2005-04-16 22:20:36 +00:00
|
|
|
/*
|
|
|
|
* NET3 Protocol independent device support routines.
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU General Public License
|
|
|
|
* as published by the Free Software Foundation; either version
|
|
|
|
* 2 of the License, or (at your option) any later version.
|
|
|
|
*
|
|
|
|
* Derived from the non IP parts of dev.c 1.0.19
|
2005-05-05 23:16:16 +00:00
|
|
|
* Authors: Ross Biro
|
2005-04-16 22:20:36 +00:00
|
|
|
* Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
|
|
|
|
* Mark Evans, <evansmp@uhura.aston.ac.uk>
|
|
|
|
*
|
|
|
|
* Additional Authors:
|
|
|
|
* Florian la Roche <rzsfl@rz.uni-sb.de>
|
|
|
|
* Alan Cox <gw4pts@gw4pts.ampr.org>
|
|
|
|
* David Hinds <dahinds@users.sourceforge.net>
|
|
|
|
* Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
|
|
|
|
* Adam Sulmicki <adam@cfar.umd.edu>
|
|
|
|
* Pekka Riikonen <priikone@poesidon.pspt.fi>
|
|
|
|
*
|
|
|
|
* Changes:
|
|
|
|
* D.J. Barrow : Fixed bug where dev->refcnt gets set
|
|
|
|
* to 2 if register_netdev gets called
|
|
|
|
* before net_dev_init & also removed a
|
|
|
|
* few lines of code in the process.
|
|
|
|
* Alan Cox : device private ioctl copies fields back.
|
|
|
|
* Alan Cox : Transmit queue code does relevant
|
|
|
|
* stunts to keep the queue safe.
|
|
|
|
* Alan Cox : Fixed double lock.
|
|
|
|
* Alan Cox : Fixed promisc NULL pointer trap
|
|
|
|
* ???????? : Support the full private ioctl range
|
|
|
|
* Alan Cox : Moved ioctl permission check into
|
|
|
|
* drivers
|
|
|
|
* Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
|
|
|
|
* Alan Cox : 100 backlog just doesn't cut it when
|
|
|
|
* you start doing multicast video 8)
|
|
|
|
* Alan Cox : Rewrote net_bh and list manager.
|
|
|
|
* Alan Cox : Fix ETH_P_ALL echoback lengths.
|
|
|
|
* Alan Cox : Took out transmit every packet pass
|
|
|
|
* Saved a few bytes in the ioctl handler
|
|
|
|
* Alan Cox : Network driver sets packet type before
|
|
|
|
* calling netif_rx. Saves a function
|
|
|
|
* call a packet.
|
|
|
|
* Alan Cox : Hashed net_bh()
|
|
|
|
* Richard Kooijman: Timestamp fixes.
|
|
|
|
* Alan Cox : Wrong field in SIOCGIFDSTADDR
|
|
|
|
* Alan Cox : Device lock protection.
|
|
|
|
* Alan Cox : Fixed nasty side effect of device close
|
|
|
|
* changes.
|
|
|
|
* Rudi Cilibrasi : Pass the right thing to
|
|
|
|
* set_mac_address()
|
|
|
|
* Dave Miller : 32bit quantity for the device lock to
|
|
|
|
* make it work out on a Sparc.
|
|
|
|
* Bjorn Ekwall : Added KERNELD hack.
|
|
|
|
* Alan Cox : Cleaned up the backlog initialise.
|
|
|
|
* Craig Metz : SIOCGIFCONF fix if space for under
|
|
|
|
* 1 device.
|
|
|
|
* Thomas Bogendoerfer : Return ENODEV for dev_open, if there
|
|
|
|
* is no device open function.
|
|
|
|
* Andi Kleen : Fix error reporting for SIOCGIFCONF
|
|
|
|
* Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
|
|
|
|
* Cyrus Durgin : Cleaned for KMOD
|
|
|
|
* Adam Sulmicki : Bug Fix : Network Device Unload
|
|
|
|
* A network device unload needs to purge
|
|
|
|
* the backlog queue.
|
|
|
|
* Paul Rusty Russell : SIOCSIFNAME
|
|
|
|
* Pekka Riikonen : Netdev boot-time settings code
|
|
|
|
* Andrew Morton : Make unregister_netdevice wait
|
|
|
|
* indefinitely on dev->refcnt
|
|
|
|
* J Hadi Salim : - Backlog queue sampling
|
|
|
|
* - netif_rx() feedback
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <asm/uaccess.h>
|
|
|
|
#include <linux/bitops.h>
|
2006-01-11 20:17:47 +00:00
|
|
|
#include <linux/capability.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
#include <linux/cpu.h>
|
|
|
|
#include <linux/types.h>
|
|
|
|
#include <linux/kernel.h>
|
2009-11-10 07:20:34 +00:00
|
|
|
#include <linux/hash.h>
|
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
|
|
|
#include <linux/slab.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
#include <linux/sched.h>
|
2006-03-21 06:33:17 +00:00
|
|
|
#include <linux/mutex.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
#include <linux/string.h>
|
|
|
|
#include <linux/mm.h>
|
|
|
|
#include <linux/socket.h>
|
|
|
|
#include <linux/sockios.h>
|
|
|
|
#include <linux/errno.h>
|
|
|
|
#include <linux/interrupt.h>
|
|
|
|
#include <linux/if_ether.h>
|
|
|
|
#include <linux/netdevice.h>
|
|
|
|
#include <linux/etherdevice.h>
|
2008-06-19 23:15:47 +00:00
|
|
|
#include <linux/ethtool.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
#include <linux/notifier.h>
|
|
|
|
#include <linux/skbuff.h>
|
2007-09-12 10:01:34 +00:00
|
|
|
#include <net/net_namespace.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
#include <net/sock.h>
|
|
|
|
#include <linux/rtnetlink.h>
|
|
|
|
#include <linux/stat.h>
|
|
|
|
#include <net/dst.h>
|
|
|
|
#include <net/pkt_sched.h>
|
|
|
|
#include <net/checksum.h>
|
2009-11-26 06:07:08 +00:00
|
|
|
#include <net/xfrm.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
#include <linux/highmem.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/netpoll.h>
|
|
|
|
#include <linux/rcupdate.h>
|
|
|
|
#include <linux/delay.h>
|
|
|
|
#include <net/iw_handler.h>
|
|
|
|
#include <asm/current.h>
|
2005-12-03 13:39:35 +00:00
|
|
|
#include <linux/audit.h>
|
2006-06-18 04:24:58 +00:00
|
|
|
#include <linux/dmaengine.h>
|
2006-06-22 09:57:17 +00:00
|
|
|
#include <linux/err.h>
|
2006-08-15 23:34:13 +00:00
|
|
|
#include <linux/ctype.h>
|
2007-05-16 05:46:18 +00:00
|
|
|
#include <linux/if_arp.h>
|
2008-06-17 00:02:28 +00:00
|
|
|
#include <linux/if_vlan.h>
|
2008-07-15 10:47:03 +00:00
|
|
|
#include <linux/ip.h>
|
2008-09-21 05:05:50 +00:00
|
|
|
#include <net/ip.h>
|
2008-07-15 10:47:03 +00:00
|
|
|
#include <linux/ipv6.h>
|
|
|
|
#include <linux/in.h>
|
2008-07-21 16:48:06 +00:00
|
|
|
#include <linux/jhash.h>
|
|
|
|
#include <linux/random.h>
|
2009-06-15 10:02:23 +00:00
|
|
|
#include <trace/events/napi.h>
|
2010-08-23 09:45:02 +00:00
|
|
|
#include <trace/events/net.h>
|
2010-08-23 09:46:12 +00:00
|
|
|
#include <trace/events/skb.h>
|
2010-03-30 22:35:50 +00:00
|
|
|
#include <linux/pci.h>
|
2010-09-17 04:39:16 +00:00
|
|
|
#include <linux/inetdevice.h>
|
2011-01-19 11:03:53 +00:00
|
|
|
#include <linux/cpu_rmap.h>
|
2012-02-24 07:31:31 +00:00
|
|
|
#include <linux/static_key.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-10-24 04:14:45 +00:00
|
|
|
#include "net-sysfs.h"
|
|
|
|
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
/* Instead of increasing this, you should create a hash table. */
|
|
|
|
#define MAX_GRO_SKBS 8
|
|
|
|
|
2009-01-05 00:13:40 +00:00
|
|
|
/* This should be increased if a protocol with a bigger head is added. */
|
|
|
|
#define GRO_MAX_HEAD (MAX_HEADER + 128)
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static DEFINE_SPINLOCK(ptype_lock);
|
2012-11-15 08:49:10 +00:00
|
|
|
static DEFINE_SPINLOCK(offload_lock);
|
2013-02-18 19:20:33 +00:00
|
|
|
struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
|
|
|
|
struct list_head ptype_all __read_mostly; /* Taps */
|
2012-11-15 08:49:10 +00:00
|
|
|
static struct list_head offload_base __read_mostly;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/*
|
2007-05-03 22:13:45 +00:00
|
|
|
* The @dev_base_head list is protected by @dev_base_lock and the rtnl
|
2005-04-16 22:20:36 +00:00
|
|
|
* semaphore.
|
|
|
|
*
|
2009-11-04 13:43:23 +00:00
|
|
|
* Pure readers hold dev_base_lock for reading, or rcu_read_lock()
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* Writers must hold the rtnl semaphore while they loop through the
|
2007-05-03 22:13:45 +00:00
|
|
|
* dev_base_head list, and hold dev_base_lock for writing when they do the
|
2005-04-16 22:20:36 +00:00
|
|
|
* actual updates. This allows pure readers to access the list even
|
|
|
|
* while a writer is preparing to update it.
|
|
|
|
*
|
|
|
|
* To put it another way, dev_base_lock is held for writing only to
|
|
|
|
* protect against pure readers; the rtnl semaphore provides the
|
|
|
|
* protection against other writers.
|
|
|
|
*
|
|
|
|
* See, for example usages, register_netdevice() and
|
|
|
|
* unregister_netdevice(), which must be called with the rtnl
|
|
|
|
* semaphore held.
|
|
|
|
*/
|
|
|
|
DEFINE_RWLOCK(dev_base_lock);
|
|
|
|
EXPORT_SYMBOL(dev_base_lock);
|
|
|
|
|
2012-12-20 17:25:08 +00:00
|
|
|
seqcount_t devnet_rename_seq;
|
2012-11-26 05:21:08 +00:00
|
|
|
|
2011-06-21 03:11:20 +00:00
|
|
|
static inline void dev_base_seq_inc(struct net *net)
|
|
|
|
{
|
|
|
|
while (++net->dev_base_seq == 0);
|
|
|
|
}
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2012-04-15 05:58:06 +00:00
|
|
|
unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
|
|
|
|
|
2009-11-10 07:20:34 +00:00
|
|
|
return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2009-10-24 13:13:17 +00:00
|
|
|
return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2010-04-19 21:17:14 +00:00
|
|
|
static inline void rps_lock(struct softnet_data *sd)
|
2010-03-30 20:16:22 +00:00
|
|
|
{
|
|
|
|
#ifdef CONFIG_RPS
|
2010-04-19 21:17:14 +00:00
|
|
|
spin_lock(&sd->input_pkt_queue.lock);
|
2010-03-30 20:16:22 +00:00
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2010-04-19 21:17:14 +00:00
|
|
|
static inline void rps_unlock(struct softnet_data *sd)
|
2010-03-30 20:16:22 +00:00
|
|
|
{
|
|
|
|
#ifdef CONFIG_RPS
|
2010-04-19 21:17:14 +00:00
|
|
|
spin_unlock(&sd->input_pkt_queue.lock);
|
2010-03-30 20:16:22 +00:00
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2007-09-12 11:53:49 +00:00
|
|
|
/* Device list insertion */
|
2013-04-17 22:17:50 +00:00
|
|
|
static void list_netdevice(struct net_device *dev)
|
2007-09-12 11:53:49 +00:00
|
|
|
{
|
2008-03-25 12:47:49 +00:00
|
|
|
struct net *net = dev_net(dev);
|
2007-09-12 11:53:49 +00:00
|
|
|
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
write_lock_bh(&dev_base_lock);
|
2009-11-04 13:43:23 +00:00
|
|
|
list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
|
2009-10-30 07:11:27 +00:00
|
|
|
hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
|
2009-10-19 19:18:49 +00:00
|
|
|
hlist_add_head_rcu(&dev->index_hlist,
|
|
|
|
dev_index_hash(net, dev->ifindex));
|
2007-09-12 11:53:49 +00:00
|
|
|
write_unlock_bh(&dev_base_lock);
|
2011-06-21 03:11:20 +00:00
|
|
|
|
|
|
|
dev_base_seq_inc(net);
|
2007-09-12 11:53:49 +00:00
|
|
|
}
|
|
|
|
|
2009-10-19 19:18:49 +00:00
|
|
|
/* Device list removal
|
|
|
|
* caller must respect a RCU grace period before freeing/reusing dev
|
|
|
|
*/
|
2007-09-12 11:53:49 +00:00
|
|
|
static void unlist_netdevice(struct net_device *dev)
|
|
|
|
{
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
/* Unlink dev from the device chain */
|
|
|
|
write_lock_bh(&dev_base_lock);
|
2009-11-04 13:43:23 +00:00
|
|
|
list_del_rcu(&dev->dev_list);
|
2009-10-30 07:11:27 +00:00
|
|
|
hlist_del_rcu(&dev->name_hlist);
|
2009-10-19 19:18:49 +00:00
|
|
|
hlist_del_rcu(&dev->index_hlist);
|
2007-09-12 11:53:49 +00:00
|
|
|
write_unlock_bh(&dev_base_lock);
|
2011-06-21 03:11:20 +00:00
|
|
|
|
|
|
|
dev_base_seq_inc(dev_net(dev));
|
2007-09-12 11:53:49 +00:00
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/*
|
|
|
|
* Our notifier list
|
|
|
|
*/
|
|
|
|
|
2006-05-09 22:23:03 +00:00
|
|
|
static RAW_NOTIFIER_HEAD(netdev_chain);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Device drivers call our routines to queue packets here. We empty the
|
|
|
|
* queue in the local softnet handler.
|
|
|
|
*/
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
|
2010-04-17 04:17:02 +00:00
|
|
|
DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_PER_CPU_SYMBOL(softnet_data);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-07-22 21:16:42 +00:00
|
|
|
#ifdef CONFIG_LOCKDEP
|
2007-05-16 05:46:18 +00:00
|
|
|
/*
|
2008-07-09 06:13:53 +00:00
|
|
|
* register_netdevice() inits txq->_xmit_lock and sets lockdep class
|
2007-05-16 05:46:18 +00:00
|
|
|
* according to dev->type
|
|
|
|
*/
|
|
|
|
static const unsigned short netdev_lock_type[] =
|
|
|
|
{ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
|
|
|
|
ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
|
|
|
|
ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
|
|
|
|
ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
|
|
|
|
ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
|
|
|
|
ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
|
|
|
|
ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
|
|
|
|
ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
|
|
|
|
ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
|
|
|
|
ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
|
|
|
|
ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
|
|
|
|
ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
|
2012-05-10 21:14:35 +00:00
|
|
|
ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
|
|
|
|
ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
|
|
|
|
ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
|
2007-05-16 05:46:18 +00:00
|
|
|
|
2009-08-05 17:42:58 +00:00
|
|
|
static const char *const netdev_lock_name[] =
|
2007-05-16 05:46:18 +00:00
|
|
|
{"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
|
|
|
|
"_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
|
|
|
|
"_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
|
|
|
|
"_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
|
|
|
|
"_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
|
|
|
|
"_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
|
|
|
|
"_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
|
|
|
|
"_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
|
|
|
|
"_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
|
|
|
|
"_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
|
|
|
|
"_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
|
|
|
|
"_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
|
2012-05-10 21:14:35 +00:00
|
|
|
"_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
|
|
|
|
"_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
|
|
|
|
"_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
|
2007-05-16 05:46:18 +00:00
|
|
|
|
|
|
|
static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
|
2008-07-22 21:16:42 +00:00
|
|
|
static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
|
2007-05-16 05:46:18 +00:00
|
|
|
|
|
|
|
static inline unsigned short netdev_lock_pos(unsigned short dev_type)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
|
|
|
|
if (netdev_lock_type[i] == dev_type)
|
|
|
|
return i;
|
|
|
|
/* the last key is used by default */
|
|
|
|
return ARRAY_SIZE(netdev_lock_type) - 1;
|
|
|
|
}
|
|
|
|
|
2008-07-22 21:16:42 +00:00
|
|
|
static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
|
|
|
|
unsigned short dev_type)
|
2007-05-16 05:46:18 +00:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
i = netdev_lock_pos(dev_type);
|
|
|
|
lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
|
|
|
|
netdev_lock_name[i]);
|
|
|
|
}
|
2008-07-22 21:16:42 +00:00
|
|
|
|
|
|
|
static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
i = netdev_lock_pos(dev->type);
|
|
|
|
lockdep_set_class_and_name(&dev->addr_list_lock,
|
|
|
|
&netdev_addr_lock_key[i],
|
|
|
|
netdev_lock_name[i]);
|
|
|
|
}
|
2007-05-16 05:46:18 +00:00
|
|
|
#else
|
2008-07-22 21:16:42 +00:00
|
|
|
static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
|
|
|
|
unsigned short dev_type)
|
|
|
|
{
|
|
|
|
}
|
|
|
|
static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
|
2007-05-16 05:46:18 +00:00
|
|
|
{
|
|
|
|
}
|
|
|
|
#endif
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/*******************************************************************************
|
|
|
|
|
|
|
|
Protocol management and registration routines
|
|
|
|
|
|
|
|
*******************************************************************************/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Add a protocol ID to the list. Now that the input handler is
|
|
|
|
* smarter we can dispense with all the messy stuff that used to be
|
|
|
|
* here.
|
|
|
|
*
|
|
|
|
* BEWARE!!! Protocol handlers, mangling input packets,
|
|
|
|
* MUST BE last in hash buckets and checking protocol handlers
|
|
|
|
* MUST start from promiscuous ptype_all chain in net_bh.
|
|
|
|
* It is true now, do not change it.
|
|
|
|
* Explanation follows: if protocol handler, mangling packet, will
|
|
|
|
* be the first on list, it is not able to sense, that packet
|
|
|
|
* is cloned and should be copied-on-write, so that it will
|
|
|
|
* change it and subsequent readers will get broken packet.
|
|
|
|
* --ANK (980803)
|
|
|
|
*/
|
|
|
|
|
2010-09-02 03:53:46 +00:00
|
|
|
static inline struct list_head *ptype_head(const struct packet_type *pt)
|
|
|
|
{
|
|
|
|
if (pt->type == htons(ETH_P_ALL))
|
|
|
|
return &ptype_all;
|
|
|
|
else
|
|
|
|
return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* dev_add_pack - add packet handler
|
|
|
|
* @pt: packet type declaration
|
|
|
|
*
|
|
|
|
* Add a protocol handler to the networking stack. The passed &packet_type
|
|
|
|
* is linked into kernel lists and may not be freed until it has been
|
|
|
|
* removed from the kernel lists.
|
|
|
|
*
|
2007-02-09 14:24:36 +00:00
|
|
|
* This call does not sleep therefore it can not
|
2005-04-16 22:20:36 +00:00
|
|
|
* guarantee all CPU's that are in middle of receiving packets
|
|
|
|
* will see the new packet type (until the next received packet).
|
|
|
|
*/
|
|
|
|
|
|
|
|
void dev_add_pack(struct packet_type *pt)
|
|
|
|
{
|
2010-09-02 03:53:46 +00:00
|
|
|
struct list_head *head = ptype_head(pt);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-09-02 03:53:46 +00:00
|
|
|
spin_lock(&ptype_lock);
|
|
|
|
list_add_rcu(&pt->list, head);
|
|
|
|
spin_unlock(&ptype_lock);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_add_pack);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* __dev_remove_pack - remove packet handler
|
|
|
|
* @pt: packet type declaration
|
|
|
|
*
|
|
|
|
* Remove a protocol handler that was previously added to the kernel
|
|
|
|
* protocol handlers by dev_add_pack(). The passed &packet_type is removed
|
|
|
|
* from the kernel lists and can be freed or reused once this function
|
2007-02-09 14:24:36 +00:00
|
|
|
* returns.
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* The packet type might still be in use by receivers
|
|
|
|
* and must not be freed until after all the CPU's have gone
|
|
|
|
* through a quiescent state.
|
|
|
|
*/
|
|
|
|
void __dev_remove_pack(struct packet_type *pt)
|
|
|
|
{
|
2010-09-02 03:53:46 +00:00
|
|
|
struct list_head *head = ptype_head(pt);
|
2005-04-16 22:20:36 +00:00
|
|
|
struct packet_type *pt1;
|
|
|
|
|
2010-09-02 03:53:46 +00:00
|
|
|
spin_lock(&ptype_lock);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
list_for_each_entry(pt1, head, list) {
|
|
|
|
if (pt == pt1) {
|
|
|
|
list_del_rcu(&pt->list);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_warn("dev_remove_pack: %p not found\n", pt);
|
2005-04-16 22:20:36 +00:00
|
|
|
out:
|
2010-09-02 03:53:46 +00:00
|
|
|
spin_unlock(&ptype_lock);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(__dev_remove_pack);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* dev_remove_pack - remove packet handler
|
|
|
|
* @pt: packet type declaration
|
|
|
|
*
|
|
|
|
* Remove a protocol handler that was previously added to the kernel
|
|
|
|
* protocol handlers by dev_add_pack(). The passed &packet_type is removed
|
|
|
|
* from the kernel lists and can be freed or reused once this function
|
|
|
|
* returns.
|
|
|
|
*
|
|
|
|
* This call sleeps to guarantee that no CPU is looking at the packet
|
|
|
|
* type after return.
|
|
|
|
*/
|
|
|
|
void dev_remove_pack(struct packet_type *pt)
|
|
|
|
{
|
|
|
|
__dev_remove_pack(pt);
|
2007-02-09 14:24:36 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
synchronize_net();
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_remove_pack);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2012-11-15 08:49:10 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* dev_add_offload - register offload handlers
|
|
|
|
* @po: protocol offload declaration
|
|
|
|
*
|
|
|
|
* Add protocol offload handlers to the networking stack. The passed
|
|
|
|
* &proto_offload is linked into kernel lists and may not be freed until
|
|
|
|
* it has been removed from the kernel lists.
|
|
|
|
*
|
|
|
|
* This call does not sleep therefore it can not
|
|
|
|
* guarantee all CPU's that are in middle of receiving packets
|
|
|
|
* will see the new offload handlers (until the next received packet).
|
|
|
|
*/
|
|
|
|
void dev_add_offload(struct packet_offload *po)
|
|
|
|
{
|
|
|
|
struct list_head *head = &offload_base;
|
|
|
|
|
|
|
|
spin_lock(&offload_lock);
|
|
|
|
list_add_rcu(&po->list, head);
|
|
|
|
spin_unlock(&offload_lock);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_add_offload);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __dev_remove_offload - remove offload handler
|
|
|
|
* @po: packet offload declaration
|
|
|
|
*
|
|
|
|
* Remove a protocol offload handler that was previously added to the
|
|
|
|
* kernel offload handlers by dev_add_offload(). The passed &offload_type
|
|
|
|
* is removed from the kernel lists and can be freed or reused once this
|
|
|
|
* function returns.
|
|
|
|
*
|
|
|
|
* The packet type might still be in use by receivers
|
|
|
|
* and must not be freed until after all the CPU's have gone
|
|
|
|
* through a quiescent state.
|
|
|
|
*/
|
|
|
|
void __dev_remove_offload(struct packet_offload *po)
|
|
|
|
{
|
|
|
|
struct list_head *head = &offload_base;
|
|
|
|
struct packet_offload *po1;
|
|
|
|
|
2012-11-16 08:08:23 +00:00
|
|
|
spin_lock(&offload_lock);
|
2012-11-15 08:49:10 +00:00
|
|
|
|
|
|
|
list_for_each_entry(po1, head, list) {
|
|
|
|
if (po == po1) {
|
|
|
|
list_del_rcu(&po->list);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
pr_warn("dev_remove_offload: %p not found\n", po);
|
|
|
|
out:
|
2012-11-16 08:08:23 +00:00
|
|
|
spin_unlock(&offload_lock);
|
2012-11-15 08:49:10 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(__dev_remove_offload);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* dev_remove_offload - remove packet offload handler
|
|
|
|
* @po: packet offload declaration
|
|
|
|
*
|
|
|
|
* Remove a packet offload handler that was previously added to the kernel
|
|
|
|
* offload handlers by dev_add_offload(). The passed &offload_type is
|
|
|
|
* removed from the kernel lists and can be freed or reused once this
|
|
|
|
* function returns.
|
|
|
|
*
|
|
|
|
* This call sleeps to guarantee that no CPU is looking at the packet
|
|
|
|
* type after return.
|
|
|
|
*/
|
|
|
|
void dev_remove_offload(struct packet_offload *po)
|
|
|
|
{
|
|
|
|
__dev_remove_offload(po);
|
|
|
|
|
|
|
|
synchronize_net();
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_remove_offload);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/******************************************************************************
|
|
|
|
|
|
|
|
Device Boot-time Settings Routines
|
|
|
|
|
|
|
|
*******************************************************************************/
|
|
|
|
|
|
|
|
/* Boot time configuration table */
|
|
|
|
static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_boot_setup_add - add new setup entry
|
|
|
|
* @name: name of the device
|
|
|
|
* @map: configured settings for the device
|
|
|
|
*
|
|
|
|
* Adds new setup entry to the dev_boot_setup list. The function
|
|
|
|
* returns 0 on error and 1 on success. This is a generic routine to
|
|
|
|
* all netdevices.
|
|
|
|
*/
|
|
|
|
static int netdev_boot_setup_add(char *name, struct ifmap *map)
|
|
|
|
{
|
|
|
|
struct netdev_boot_setup *s;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
s = dev_boot_setup;
|
|
|
|
for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
|
|
|
|
if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
|
|
|
|
memset(s[i].name, 0, sizeof(s[i].name));
|
2008-07-02 02:57:19 +00:00
|
|
|
strlcpy(s[i].name, name, IFNAMSIZ);
|
2005-04-16 22:20:36 +00:00
|
|
|
memcpy(&s[i].map, map, sizeof(s[i].map));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_boot_setup_check - check boot time settings
|
|
|
|
* @dev: the netdevice
|
|
|
|
*
|
|
|
|
* Check boot time settings for the device.
|
|
|
|
* The found settings are set for the device to be used
|
|
|
|
* later in the device probing.
|
|
|
|
* Returns 0 if no settings found, 1 if they are.
|
|
|
|
*/
|
|
|
|
int netdev_boot_setup_check(struct net_device *dev)
|
|
|
|
{
|
|
|
|
struct netdev_boot_setup *s = dev_boot_setup;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
|
|
|
|
if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
|
2008-07-02 02:57:19 +00:00
|
|
|
!strcmp(dev->name, s[i].name)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->irq = s[i].map.irq;
|
|
|
|
dev->base_addr = s[i].map.base_addr;
|
|
|
|
dev->mem_start = s[i].map.mem_start;
|
|
|
|
dev->mem_end = s[i].map.mem_end;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(netdev_boot_setup_check);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_boot_base - get address from boot time settings
|
|
|
|
* @prefix: prefix for network device
|
|
|
|
* @unit: id for network device
|
|
|
|
*
|
|
|
|
* Check boot time settings for the base address of device.
|
|
|
|
* The found settings are set for the device to be used
|
|
|
|
* later in the device probing.
|
|
|
|
* Returns 0 if no settings found.
|
|
|
|
*/
|
|
|
|
unsigned long netdev_boot_base(const char *prefix, int unit)
|
|
|
|
{
|
|
|
|
const struct netdev_boot_setup *s = dev_boot_setup;
|
|
|
|
char name[IFNAMSIZ];
|
|
|
|
int i;
|
|
|
|
|
|
|
|
sprintf(name, "%s%d", prefix, unit);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If device already registered then return base of 1
|
|
|
|
* to indicate not to probe for this interface
|
|
|
|
*/
|
2007-09-17 18:56:21 +00:00
|
|
|
if (__dev_get_by_name(&init_net, name))
|
2005-04-16 22:20:36 +00:00
|
|
|
return 1;
|
|
|
|
|
|
|
|
for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
|
|
|
|
if (!strcmp(name, s[i].name))
|
|
|
|
return s[i].map.base_addr;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Saves at boot time configured settings for any netdevice.
|
|
|
|
*/
|
|
|
|
int __init netdev_boot_setup(char *str)
|
|
|
|
{
|
|
|
|
int ints[5];
|
|
|
|
struct ifmap map;
|
|
|
|
|
|
|
|
str = get_options(str, ARRAY_SIZE(ints), ints);
|
|
|
|
if (!str || !*str)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* Save settings */
|
|
|
|
memset(&map, 0, sizeof(map));
|
|
|
|
if (ints[0] > 0)
|
|
|
|
map.irq = ints[1];
|
|
|
|
if (ints[0] > 1)
|
|
|
|
map.base_addr = ints[2];
|
|
|
|
if (ints[0] > 2)
|
|
|
|
map.mem_start = ints[3];
|
|
|
|
if (ints[0] > 3)
|
|
|
|
map.mem_end = ints[4];
|
|
|
|
|
|
|
|
/* Add new entry to the list */
|
|
|
|
return netdev_boot_setup_add(str, &map);
|
|
|
|
}
|
|
|
|
|
|
|
|
__setup("netdev=", netdev_boot_setup);
|
|
|
|
|
|
|
|
/*******************************************************************************
|
|
|
|
|
|
|
|
Device Interface Subroutines
|
|
|
|
|
|
|
|
*******************************************************************************/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __dev_get_by_name - find a device by its name
|
2007-10-13 04:17:49 +00:00
|
|
|
* @net: the applicable net namespace
|
2005-04-16 22:20:36 +00:00
|
|
|
* @name: name to find
|
|
|
|
*
|
|
|
|
* Find an interface by name. Must be called under RTNL semaphore
|
|
|
|
* or @dev_base_lock. If the name is found a pointer to the device
|
|
|
|
* is returned. If the name is not found then %NULL is returned. The
|
|
|
|
* reference counters are not incremented so the caller must be
|
|
|
|
* careful with locks.
|
|
|
|
*/
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
struct net_device *__dev_get_by_name(struct net *net, const char *name)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2009-10-30 08:40:11 +00:00
|
|
|
struct net_device *dev;
|
|
|
|
struct hlist_head *head = dev_name_hash(net, name);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
hlist: drop the node parameter from iterators
I'm not sure why, but the hlist for each entry iterators were conceived
list_for_each_entry(pos, head, member)
The hlist ones were greedy and wanted an extra parameter:
hlist_for_each_entry(tpos, pos, head, member)
Why did they need an extra pos parameter? I'm not quite sure. Not only
they don't really need it, it also prevents the iterator from looking
exactly like the list iterator, which is unfortunate.
Besides the semantic patch, there was some manual work required:
- Fix up the actual hlist iterators in linux/list.h
- Fix up the declaration of other iterators based on the hlist ones.
- A very small amount of places were using the 'node' parameter, this
was modified to use 'obj->member' instead.
- Coccinelle didn't handle the hlist_for_each_entry_safe iterator
properly, so those had to be fixed up manually.
The semantic patch which is mostly the work of Peter Senna Tschudin is here:
@@
iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host;
type T;
expression a,c,d,e;
identifier b;
statement S;
@@
-T b;
<+... when != b
(
hlist_for_each_entry(a,
- b,
c, d) S
|
hlist_for_each_entry_continue(a,
- b,
c) S
|
hlist_for_each_entry_from(a,
- b,
c) S
|
hlist_for_each_entry_rcu(a,
- b,
c, d) S
|
hlist_for_each_entry_rcu_bh(a,
- b,
c, d) S
|
hlist_for_each_entry_continue_rcu_bh(a,
- b,
c) S
|
for_each_busy_worker(a, c,
- b,
d) S
|
ax25_uid_for_each(a,
- b,
c) S
|
ax25_for_each(a,
- b,
c) S
|
inet_bind_bucket_for_each(a,
- b,
c) S
|
sctp_for_each_hentry(a,
- b,
c) S
|
sk_for_each(a,
- b,
c) S
|
sk_for_each_rcu(a,
- b,
c) S
|
sk_for_each_from
-(a, b)
+(a)
S
+ sk_for_each_from(a) S
|
sk_for_each_safe(a,
- b,
c, d) S
|
sk_for_each_bound(a,
- b,
c) S
|
hlist_for_each_entry_safe(a,
- b,
c, d, e) S
|
hlist_for_each_entry_continue_rcu(a,
- b,
c) S
|
nr_neigh_for_each(a,
- b,
c) S
|
nr_neigh_for_each_safe(a,
- b,
c, d) S
|
nr_node_for_each(a,
- b,
c) S
|
nr_node_for_each_safe(a,
- b,
c, d) S
|
- for_each_gfn_sp(a, c, d, b) S
+ for_each_gfn_sp(a, c, d) S
|
- for_each_gfn_indirect_valid_sp(a, c, d, b) S
+ for_each_gfn_indirect_valid_sp(a, c, d) S
|
for_each_host(a,
- b,
c) S
|
for_each_host_safe(a,
- b,
c, d) S
|
for_each_mesh_entry(a,
- b,
c, d) S
)
...+>
[akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c]
[akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c]
[akpm@linux-foundation.org: checkpatch fixes]
[akpm@linux-foundation.org: fix warnings]
[akpm@linux-foudnation.org: redo intrusive kvm changes]
Tested-by: Peter Senna Tschudin <peter.senna@gmail.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 01:06:00 +00:00
|
|
|
hlist_for_each_entry(dev, head, name_hlist)
|
2005-04-16 22:20:36 +00:00
|
|
|
if (!strncmp(dev->name, name, IFNAMSIZ))
|
|
|
|
return dev;
|
2009-10-30 08:40:11 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(__dev_get_by_name);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2009-10-30 07:11:27 +00:00
|
|
|
/**
|
|
|
|
* dev_get_by_name_rcu - find a device by its name
|
|
|
|
* @net: the applicable net namespace
|
|
|
|
* @name: name to find
|
|
|
|
*
|
|
|
|
* Find an interface by name.
|
|
|
|
* If the name is found a pointer to the device is returned.
|
|
|
|
* If the name is not found then %NULL is returned.
|
|
|
|
* The reference counters are not incremented so the caller must be
|
|
|
|
* careful with locks. The caller must hold RCU lock.
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
|
|
|
|
{
|
|
|
|
struct net_device *dev;
|
|
|
|
struct hlist_head *head = dev_name_hash(net, name);
|
|
|
|
|
hlist: drop the node parameter from iterators
I'm not sure why, but the hlist for each entry iterators were conceived
list_for_each_entry(pos, head, member)
The hlist ones were greedy and wanted an extra parameter:
hlist_for_each_entry(tpos, pos, head, member)
Why did they need an extra pos parameter? I'm not quite sure. Not only
they don't really need it, it also prevents the iterator from looking
exactly like the list iterator, which is unfortunate.
Besides the semantic patch, there was some manual work required:
- Fix up the actual hlist iterators in linux/list.h
- Fix up the declaration of other iterators based on the hlist ones.
- A very small amount of places were using the 'node' parameter, this
was modified to use 'obj->member' instead.
- Coccinelle didn't handle the hlist_for_each_entry_safe iterator
properly, so those had to be fixed up manually.
The semantic patch which is mostly the work of Peter Senna Tschudin is here:
@@
iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host;
type T;
expression a,c,d,e;
identifier b;
statement S;
@@
-T b;
<+... when != b
(
hlist_for_each_entry(a,
- b,
c, d) S
|
hlist_for_each_entry_continue(a,
- b,
c) S
|
hlist_for_each_entry_from(a,
- b,
c) S
|
hlist_for_each_entry_rcu(a,
- b,
c, d) S
|
hlist_for_each_entry_rcu_bh(a,
- b,
c, d) S
|
hlist_for_each_entry_continue_rcu_bh(a,
- b,
c) S
|
for_each_busy_worker(a, c,
- b,
d) S
|
ax25_uid_for_each(a,
- b,
c) S
|
ax25_for_each(a,
- b,
c) S
|
inet_bind_bucket_for_each(a,
- b,
c) S
|
sctp_for_each_hentry(a,
- b,
c) S
|
sk_for_each(a,
- b,
c) S
|
sk_for_each_rcu(a,
- b,
c) S
|
sk_for_each_from
-(a, b)
+(a)
S
+ sk_for_each_from(a) S
|
sk_for_each_safe(a,
- b,
c, d) S
|
sk_for_each_bound(a,
- b,
c) S
|
hlist_for_each_entry_safe(a,
- b,
c, d, e) S
|
hlist_for_each_entry_continue_rcu(a,
- b,
c) S
|
nr_neigh_for_each(a,
- b,
c) S
|
nr_neigh_for_each_safe(a,
- b,
c, d) S
|
nr_node_for_each(a,
- b,
c) S
|
nr_node_for_each_safe(a,
- b,
c, d) S
|
- for_each_gfn_sp(a, c, d, b) S
+ for_each_gfn_sp(a, c, d) S
|
- for_each_gfn_indirect_valid_sp(a, c, d, b) S
+ for_each_gfn_indirect_valid_sp(a, c, d) S
|
for_each_host(a,
- b,
c) S
|
for_each_host_safe(a,
- b,
c, d) S
|
for_each_mesh_entry(a,
- b,
c, d) S
)
...+>
[akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c]
[akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c]
[akpm@linux-foundation.org: checkpatch fixes]
[akpm@linux-foundation.org: fix warnings]
[akpm@linux-foudnation.org: redo intrusive kvm changes]
Tested-by: Peter Senna Tschudin <peter.senna@gmail.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 01:06:00 +00:00
|
|
|
hlist_for_each_entry_rcu(dev, head, name_hlist)
|
2009-10-30 07:11:27 +00:00
|
|
|
if (!strncmp(dev->name, name, IFNAMSIZ))
|
|
|
|
return dev;
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_get_by_name_rcu);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* dev_get_by_name - find a device by its name
|
2007-10-13 04:17:49 +00:00
|
|
|
* @net: the applicable net namespace
|
2005-04-16 22:20:36 +00:00
|
|
|
* @name: name to find
|
|
|
|
*
|
|
|
|
* Find an interface by name. This can be called from any
|
|
|
|
* context and does its own locking. The returned handle has
|
|
|
|
* the usage count incremented and the caller must use dev_put() to
|
|
|
|
* release it when it is no longer needed. %NULL is returned if no
|
|
|
|
* matching device is found.
|
|
|
|
*/
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
struct net_device *dev_get_by_name(struct net *net, const char *name)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev;
|
|
|
|
|
2009-10-30 07:11:27 +00:00
|
|
|
rcu_read_lock();
|
|
|
|
dev = dev_get_by_name_rcu(net, name);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (dev)
|
|
|
|
dev_hold(dev);
|
2009-10-30 07:11:27 +00:00
|
|
|
rcu_read_unlock();
|
2005-04-16 22:20:36 +00:00
|
|
|
return dev;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_get_by_name);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* __dev_get_by_index - find a device by its ifindex
|
2007-10-13 04:17:49 +00:00
|
|
|
* @net: the applicable net namespace
|
2005-04-16 22:20:36 +00:00
|
|
|
* @ifindex: index of device
|
|
|
|
*
|
|
|
|
* Search for an interface by index. Returns %NULL if the device
|
|
|
|
* is not found or a pointer to the device. The device has not
|
|
|
|
* had its reference counter increased so the caller must be careful
|
|
|
|
* about locking. The caller must hold either the RTNL semaphore
|
|
|
|
* or @dev_base_lock.
|
|
|
|
*/
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
struct net_device *__dev_get_by_index(struct net *net, int ifindex)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2009-10-30 08:40:11 +00:00
|
|
|
struct net_device *dev;
|
|
|
|
struct hlist_head *head = dev_index_hash(net, ifindex);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
hlist: drop the node parameter from iterators
I'm not sure why, but the hlist for each entry iterators were conceived
list_for_each_entry(pos, head, member)
The hlist ones were greedy and wanted an extra parameter:
hlist_for_each_entry(tpos, pos, head, member)
Why did they need an extra pos parameter? I'm not quite sure. Not only
they don't really need it, it also prevents the iterator from looking
exactly like the list iterator, which is unfortunate.
Besides the semantic patch, there was some manual work required:
- Fix up the actual hlist iterators in linux/list.h
- Fix up the declaration of other iterators based on the hlist ones.
- A very small amount of places were using the 'node' parameter, this
was modified to use 'obj->member' instead.
- Coccinelle didn't handle the hlist_for_each_entry_safe iterator
properly, so those had to be fixed up manually.
The semantic patch which is mostly the work of Peter Senna Tschudin is here:
@@
iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host;
type T;
expression a,c,d,e;
identifier b;
statement S;
@@
-T b;
<+... when != b
(
hlist_for_each_entry(a,
- b,
c, d) S
|
hlist_for_each_entry_continue(a,
- b,
c) S
|
hlist_for_each_entry_from(a,
- b,
c) S
|
hlist_for_each_entry_rcu(a,
- b,
c, d) S
|
hlist_for_each_entry_rcu_bh(a,
- b,
c, d) S
|
hlist_for_each_entry_continue_rcu_bh(a,
- b,
c) S
|
for_each_busy_worker(a, c,
- b,
d) S
|
ax25_uid_for_each(a,
- b,
c) S
|
ax25_for_each(a,
- b,
c) S
|
inet_bind_bucket_for_each(a,
- b,
c) S
|
sctp_for_each_hentry(a,
- b,
c) S
|
sk_for_each(a,
- b,
c) S
|
sk_for_each_rcu(a,
- b,
c) S
|
sk_for_each_from
-(a, b)
+(a)
S
+ sk_for_each_from(a) S
|
sk_for_each_safe(a,
- b,
c, d) S
|
sk_for_each_bound(a,
- b,
c) S
|
hlist_for_each_entry_safe(a,
- b,
c, d, e) S
|
hlist_for_each_entry_continue_rcu(a,
- b,
c) S
|
nr_neigh_for_each(a,
- b,
c) S
|
nr_neigh_for_each_safe(a,
- b,
c, d) S
|
nr_node_for_each(a,
- b,
c) S
|
nr_node_for_each_safe(a,
- b,
c, d) S
|
- for_each_gfn_sp(a, c, d, b) S
+ for_each_gfn_sp(a, c, d) S
|
- for_each_gfn_indirect_valid_sp(a, c, d, b) S
+ for_each_gfn_indirect_valid_sp(a, c, d) S
|
for_each_host(a,
- b,
c) S
|
for_each_host_safe(a,
- b,
c, d) S
|
for_each_mesh_entry(a,
- b,
c, d) S
)
...+>
[akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c]
[akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c]
[akpm@linux-foundation.org: checkpatch fixes]
[akpm@linux-foundation.org: fix warnings]
[akpm@linux-foudnation.org: redo intrusive kvm changes]
Tested-by: Peter Senna Tschudin <peter.senna@gmail.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 01:06:00 +00:00
|
|
|
hlist_for_each_entry(dev, head, index_hlist)
|
2005-04-16 22:20:36 +00:00
|
|
|
if (dev->ifindex == ifindex)
|
|
|
|
return dev;
|
2009-10-30 08:40:11 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(__dev_get_by_index);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2009-10-19 19:18:49 +00:00
|
|
|
/**
|
|
|
|
* dev_get_by_index_rcu - find a device by its ifindex
|
|
|
|
* @net: the applicable net namespace
|
|
|
|
* @ifindex: index of device
|
|
|
|
*
|
|
|
|
* Search for an interface by index. Returns %NULL if the device
|
|
|
|
* is not found or a pointer to the device. The device has not
|
|
|
|
* had its reference counter increased so the caller must be careful
|
|
|
|
* about locking. The caller must hold RCU lock.
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
|
|
|
|
{
|
|
|
|
struct net_device *dev;
|
|
|
|
struct hlist_head *head = dev_index_hash(net, ifindex);
|
|
|
|
|
hlist: drop the node parameter from iterators
I'm not sure why, but the hlist for each entry iterators were conceived
list_for_each_entry(pos, head, member)
The hlist ones were greedy and wanted an extra parameter:
hlist_for_each_entry(tpos, pos, head, member)
Why did they need an extra pos parameter? I'm not quite sure. Not only
they don't really need it, it also prevents the iterator from looking
exactly like the list iterator, which is unfortunate.
Besides the semantic patch, there was some manual work required:
- Fix up the actual hlist iterators in linux/list.h
- Fix up the declaration of other iterators based on the hlist ones.
- A very small amount of places were using the 'node' parameter, this
was modified to use 'obj->member' instead.
- Coccinelle didn't handle the hlist_for_each_entry_safe iterator
properly, so those had to be fixed up manually.
The semantic patch which is mostly the work of Peter Senna Tschudin is here:
@@
iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host;
type T;
expression a,c,d,e;
identifier b;
statement S;
@@
-T b;
<+... when != b
(
hlist_for_each_entry(a,
- b,
c, d) S
|
hlist_for_each_entry_continue(a,
- b,
c) S
|
hlist_for_each_entry_from(a,
- b,
c) S
|
hlist_for_each_entry_rcu(a,
- b,
c, d) S
|
hlist_for_each_entry_rcu_bh(a,
- b,
c, d) S
|
hlist_for_each_entry_continue_rcu_bh(a,
- b,
c) S
|
for_each_busy_worker(a, c,
- b,
d) S
|
ax25_uid_for_each(a,
- b,
c) S
|
ax25_for_each(a,
- b,
c) S
|
inet_bind_bucket_for_each(a,
- b,
c) S
|
sctp_for_each_hentry(a,
- b,
c) S
|
sk_for_each(a,
- b,
c) S
|
sk_for_each_rcu(a,
- b,
c) S
|
sk_for_each_from
-(a, b)
+(a)
S
+ sk_for_each_from(a) S
|
sk_for_each_safe(a,
- b,
c, d) S
|
sk_for_each_bound(a,
- b,
c) S
|
hlist_for_each_entry_safe(a,
- b,
c, d, e) S
|
hlist_for_each_entry_continue_rcu(a,
- b,
c) S
|
nr_neigh_for_each(a,
- b,
c) S
|
nr_neigh_for_each_safe(a,
- b,
c, d) S
|
nr_node_for_each(a,
- b,
c) S
|
nr_node_for_each_safe(a,
- b,
c, d) S
|
- for_each_gfn_sp(a, c, d, b) S
+ for_each_gfn_sp(a, c, d) S
|
- for_each_gfn_indirect_valid_sp(a, c, d, b) S
+ for_each_gfn_indirect_valid_sp(a, c, d) S
|
for_each_host(a,
- b,
c) S
|
for_each_host_safe(a,
- b,
c, d) S
|
for_each_mesh_entry(a,
- b,
c, d) S
)
...+>
[akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c]
[akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c]
[akpm@linux-foundation.org: checkpatch fixes]
[akpm@linux-foundation.org: fix warnings]
[akpm@linux-foudnation.org: redo intrusive kvm changes]
Tested-by: Peter Senna Tschudin <peter.senna@gmail.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 01:06:00 +00:00
|
|
|
hlist_for_each_entry_rcu(dev, head, index_hlist)
|
2009-10-19 19:18:49 +00:00
|
|
|
if (dev->ifindex == ifindex)
|
|
|
|
return dev;
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_get_by_index_rcu);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* dev_get_by_index - find a device by its ifindex
|
2007-10-13 04:17:49 +00:00
|
|
|
* @net: the applicable net namespace
|
2005-04-16 22:20:36 +00:00
|
|
|
* @ifindex: index of device
|
|
|
|
*
|
|
|
|
* Search for an interface by index. Returns NULL if the device
|
|
|
|
* is not found or a pointer to the device. The device returned has
|
|
|
|
* had a reference added and the pointer is safe until the user calls
|
|
|
|
* dev_put to indicate they have finished with it.
|
|
|
|
*/
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
struct net_device *dev_get_by_index(struct net *net, int ifindex)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev;
|
|
|
|
|
2009-10-19 19:18:49 +00:00
|
|
|
rcu_read_lock();
|
|
|
|
dev = dev_get_by_index_rcu(net, ifindex);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (dev)
|
|
|
|
dev_hold(dev);
|
2009-10-19 19:18:49 +00:00
|
|
|
rcu_read_unlock();
|
2005-04-16 22:20:36 +00:00
|
|
|
return dev;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_get_by_index);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
2010-12-05 01:23:53 +00:00
|
|
|
* dev_getbyhwaddr_rcu - find a device by its hardware address
|
2007-10-13 04:17:49 +00:00
|
|
|
* @net: the applicable net namespace
|
2005-04-16 22:20:36 +00:00
|
|
|
* @type: media type of device
|
|
|
|
* @ha: hardware address
|
|
|
|
*
|
|
|
|
* Search for an interface by MAC address. Returns NULL if the device
|
2011-01-24 21:16:16 +00:00
|
|
|
* is not found or a pointer to the device.
|
|
|
|
* The caller must hold RCU or RTNL.
|
2010-12-05 01:23:53 +00:00
|
|
|
* The returned device has not had its ref count increased
|
2005-04-16 22:20:36 +00:00
|
|
|
* and the caller must therefore be careful about locking
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
2010-12-05 01:23:53 +00:00
|
|
|
struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
|
|
|
|
const char *ha)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev;
|
|
|
|
|
2010-12-05 01:23:53 +00:00
|
|
|
for_each_netdev_rcu(net, dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
if (dev->type == type &&
|
|
|
|
!memcmp(dev->dev_addr, ha, dev->addr_len))
|
2007-05-03 22:13:45 +00:00
|
|
|
return dev;
|
|
|
|
|
|
|
|
return NULL;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2010-12-05 01:23:53 +00:00
|
|
|
EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
|
2005-09-22 07:44:55 +00:00
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev;
|
|
|
|
|
2007-05-03 10:28:13 +00:00
|
|
|
ASSERT_RTNL();
|
2007-09-17 18:56:21 +00:00
|
|
|
for_each_netdev(net, dev)
|
2007-05-03 10:28:13 +00:00
|
|
|
if (dev->type == type)
|
2007-05-03 22:13:45 +00:00
|
|
|
return dev;
|
|
|
|
|
|
|
|
return NULL;
|
2007-05-03 10:28:13 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(__dev_getfirstbyhwtype);
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
|
2007-05-03 10:28:13 +00:00
|
|
|
{
|
2010-03-18 11:27:25 +00:00
|
|
|
struct net_device *dev, *ret = NULL;
|
2007-05-03 10:28:13 +00:00
|
|
|
|
2010-03-18 11:27:25 +00:00
|
|
|
rcu_read_lock();
|
|
|
|
for_each_netdev_rcu(net, dev)
|
|
|
|
if (dev->type == type) {
|
|
|
|
dev_hold(dev);
|
|
|
|
ret = dev;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
return ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_getfirstbyhwtype);
|
|
|
|
|
|
|
|
/**
|
2010-06-07 11:42:13 +00:00
|
|
|
* dev_get_by_flags_rcu - find any device with given flags
|
2007-10-13 04:17:49 +00:00
|
|
|
* @net: the applicable net namespace
|
2005-04-16 22:20:36 +00:00
|
|
|
* @if_flags: IFF_* values
|
|
|
|
* @mask: bitmask of bits in if_flags to check
|
|
|
|
*
|
|
|
|
* Search for any interface with the given flags. Returns NULL if a device
|
2010-06-07 11:42:13 +00:00
|
|
|
* is not found or a pointer to the device. Must be called inside
|
|
|
|
* rcu_read_lock(), and result refcount is unchanged.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
|
2010-06-07 11:42:13 +00:00
|
|
|
struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
|
2009-09-03 08:29:39 +00:00
|
|
|
unsigned short mask)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-05-03 22:13:45 +00:00
|
|
|
struct net_device *dev, *ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-05-03 22:13:45 +00:00
|
|
|
ret = NULL;
|
2009-11-04 13:43:23 +00:00
|
|
|
for_each_netdev_rcu(net, dev) {
|
2005-04-16 22:20:36 +00:00
|
|
|
if (((dev->flags ^ if_flags) & mask) == 0) {
|
2007-05-03 22:13:45 +00:00
|
|
|
ret = dev;
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2007-05-03 22:13:45 +00:00
|
|
|
return ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2010-06-07 11:42:13 +00:00
|
|
|
EXPORT_SYMBOL(dev_get_by_flags_rcu);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* dev_valid_name - check if name is okay for network device
|
|
|
|
* @name: name string
|
|
|
|
*
|
|
|
|
* Network device names need to be valid file names to
|
2006-08-15 23:34:13 +00:00
|
|
|
* to allow sysfs to work. We also disallow any kind of
|
|
|
|
* whitespace.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
2012-03-06 21:12:15 +00:00
|
|
|
bool dev_valid_name(const char *name)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2006-08-15 23:34:13 +00:00
|
|
|
if (*name == '\0')
|
2012-03-06 21:12:15 +00:00
|
|
|
return false;
|
2006-08-30 00:06:13 +00:00
|
|
|
if (strlen(name) >= IFNAMSIZ)
|
2012-03-06 21:12:15 +00:00
|
|
|
return false;
|
2006-08-15 23:34:13 +00:00
|
|
|
if (!strcmp(name, ".") || !strcmp(name, ".."))
|
2012-03-06 21:12:15 +00:00
|
|
|
return false;
|
2006-08-15 23:34:13 +00:00
|
|
|
|
|
|
|
while (*name) {
|
|
|
|
if (*name == '/' || isspace(*name))
|
2012-03-06 21:12:15 +00:00
|
|
|
return false;
|
2006-08-15 23:34:13 +00:00
|
|
|
name++;
|
|
|
|
}
|
2012-03-06 21:12:15 +00:00
|
|
|
return true;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_valid_name);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
2007-09-12 11:48:45 +00:00
|
|
|
* __dev_alloc_name - allocate a name for a device
|
|
|
|
* @net: network namespace to allocate the device name in
|
2005-04-16 22:20:36 +00:00
|
|
|
* @name: name format string
|
2007-09-12 11:48:45 +00:00
|
|
|
* @buf: scratch buffer and result name string
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* Passed a format string - eg "lt%d" it will try and find a suitable
|
2006-05-26 20:25:24 +00:00
|
|
|
* id. It scans list of devices to build up a free map, then chooses
|
|
|
|
* the first empty slot. The caller must hold the dev_base or rtnl lock
|
|
|
|
* while allocating the name and adding the device in order to avoid
|
|
|
|
* duplicates.
|
|
|
|
* Limited to bits_per_byte * page size devices (ie 32K on most platforms).
|
|
|
|
* Returns the number of the unit assigned or a negative errno code.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
|
2007-09-12 11:48:45 +00:00
|
|
|
static int __dev_alloc_name(struct net *net, const char *name, char *buf)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
int i = 0;
|
|
|
|
const char *p;
|
|
|
|
const int max_netdevices = 8*PAGE_SIZE;
|
2007-10-09 08:59:42 +00:00
|
|
|
unsigned long *inuse;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct net_device *d;
|
|
|
|
|
|
|
|
p = strnchr(name, IFNAMSIZ-1, '%');
|
|
|
|
if (p) {
|
|
|
|
/*
|
|
|
|
* Verify the string as this thing may have come from
|
|
|
|
* the user. There must be either one "%d" and no other "%"
|
|
|
|
* characters.
|
|
|
|
*/
|
|
|
|
if (p[1] != 'd' || strchr(p + 2, '%'))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
/* Use one page as a bit array of possible slots */
|
2007-10-09 08:59:42 +00:00
|
|
|
inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (!inuse)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
for_each_netdev(net, d) {
|
2005-04-16 22:20:36 +00:00
|
|
|
if (!sscanf(d->name, name, &i))
|
|
|
|
continue;
|
|
|
|
if (i < 0 || i >= max_netdevices)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/* avoid cases where sscanf is not exact inverse of printf */
|
2007-09-12 11:48:45 +00:00
|
|
|
snprintf(buf, IFNAMSIZ, name, i);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (!strncmp(buf, d->name, IFNAMSIZ))
|
|
|
|
set_bit(i, inuse);
|
|
|
|
}
|
|
|
|
|
|
|
|
i = find_first_zero_bit(inuse, max_netdevices);
|
|
|
|
free_page((unsigned long) inuse);
|
|
|
|
}
|
|
|
|
|
2009-11-18 02:36:59 +00:00
|
|
|
if (buf != name)
|
|
|
|
snprintf(buf, IFNAMSIZ, name, i);
|
2007-09-12 11:48:45 +00:00
|
|
|
if (!__dev_get_by_name(net, buf))
|
2005-04-16 22:20:36 +00:00
|
|
|
return i;
|
|
|
|
|
|
|
|
/* It is possible to run out of possible slots
|
|
|
|
* when the name is long and there isn't enough space left
|
|
|
|
* for the digits, or if all bits are used.
|
|
|
|
*/
|
|
|
|
return -ENFILE;
|
|
|
|
}
|
|
|
|
|
2007-09-12 11:48:45 +00:00
|
|
|
/**
|
|
|
|
* dev_alloc_name - allocate a name for a device
|
|
|
|
* @dev: device
|
|
|
|
* @name: name format string
|
|
|
|
*
|
|
|
|
* Passed a format string - eg "lt%d" it will try and find a suitable
|
|
|
|
* id. It scans list of devices to build up a free map, then chooses
|
|
|
|
* the first empty slot. The caller must hold the dev_base or rtnl lock
|
|
|
|
* while allocating the name and adding the device in order to avoid
|
|
|
|
* duplicates.
|
|
|
|
* Limited to bits_per_byte * page size devices (ie 32K on most platforms).
|
|
|
|
* Returns the number of the unit assigned or a negative errno code.
|
|
|
|
*/
|
|
|
|
|
|
|
|
int dev_alloc_name(struct net_device *dev, const char *name)
|
|
|
|
{
|
|
|
|
char buf[IFNAMSIZ];
|
|
|
|
struct net *net;
|
|
|
|
int ret;
|
|
|
|
|
2008-03-25 12:47:49 +00:00
|
|
|
BUG_ON(!dev_net(dev));
|
|
|
|
net = dev_net(dev);
|
2007-09-12 11:48:45 +00:00
|
|
|
ret = __dev_alloc_name(net, name, buf);
|
|
|
|
if (ret >= 0)
|
|
|
|
strlcpy(dev->name, buf, IFNAMSIZ);
|
|
|
|
return ret;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_alloc_name);
|
2007-09-12 11:48:45 +00:00
|
|
|
|
2012-09-13 20:58:27 +00:00
|
|
|
static int dev_alloc_name_ns(struct net *net,
|
|
|
|
struct net_device *dev,
|
|
|
|
const char *name)
|
2009-11-18 02:36:59 +00:00
|
|
|
{
|
2012-09-13 20:58:27 +00:00
|
|
|
char buf[IFNAMSIZ];
|
|
|
|
int ret;
|
2010-05-19 10:12:19 +00:00
|
|
|
|
2012-09-13 20:58:27 +00:00
|
|
|
ret = __dev_alloc_name(net, name, buf);
|
|
|
|
if (ret >= 0)
|
|
|
|
strlcpy(dev->name, buf, IFNAMSIZ);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int dev_get_valid_name(struct net *net,
|
|
|
|
struct net_device *dev,
|
|
|
|
const char *name)
|
|
|
|
{
|
|
|
|
BUG_ON(!net);
|
2010-05-19 10:12:19 +00:00
|
|
|
|
2009-11-18 02:36:59 +00:00
|
|
|
if (!dev_valid_name(name))
|
|
|
|
return -EINVAL;
|
|
|
|
|
2011-04-30 01:21:32 +00:00
|
|
|
if (strchr(name, '%'))
|
2012-09-13 20:58:27 +00:00
|
|
|
return dev_alloc_name_ns(net, dev, name);
|
2009-11-18 02:36:59 +00:00
|
|
|
else if (__dev_get_by_name(net, name))
|
|
|
|
return -EEXIST;
|
2010-05-19 10:12:19 +00:00
|
|
|
else if (dev->name != name)
|
|
|
|
strlcpy(dev->name, name, IFNAMSIZ);
|
2009-11-18 02:36:59 +00:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* dev_change_name - change name of a device
|
|
|
|
* @dev: device
|
|
|
|
* @newname: name (or format string) must be at least IFNAMSIZ
|
|
|
|
*
|
|
|
|
* Change name of a device, can pass format strings "eth%d".
|
|
|
|
* for wildcarding.
|
|
|
|
*/
|
2008-09-30 09:22:14 +00:00
|
|
|
int dev_change_name(struct net_device *dev, const char *newname)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-07-31 00:03:38 +00:00
|
|
|
char oldname[IFNAMSIZ];
|
2005-04-16 22:20:36 +00:00
|
|
|
int err = 0;
|
2007-07-31 00:03:38 +00:00
|
|
|
int ret;
|
2007-09-17 18:56:21 +00:00
|
|
|
struct net *net;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
ASSERT_RTNL();
|
2008-03-25 12:47:49 +00:00
|
|
|
BUG_ON(!dev_net(dev));
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-03-25 12:47:49 +00:00
|
|
|
net = dev_net(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (dev->flags & IFF_UP)
|
|
|
|
return -EBUSY;
|
|
|
|
|
2012-12-20 17:25:08 +00:00
|
|
|
write_seqcount_begin(&devnet_rename_seq);
|
2012-11-26 05:21:08 +00:00
|
|
|
|
|
|
|
if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
|
2012-12-20 17:25:08 +00:00
|
|
|
write_seqcount_end(&devnet_rename_seq);
|
2007-10-26 10:53:42 +00:00
|
|
|
return 0;
|
2012-11-26 05:21:08 +00:00
|
|
|
}
|
2007-10-26 10:53:42 +00:00
|
|
|
|
2007-07-31 00:03:38 +00:00
|
|
|
memcpy(oldname, dev->name, IFNAMSIZ);
|
|
|
|
|
2012-09-13 20:58:27 +00:00
|
|
|
err = dev_get_valid_name(net, dev, newname);
|
2012-11-26 05:21:08 +00:00
|
|
|
if (err < 0) {
|
2012-12-20 17:25:08 +00:00
|
|
|
write_seqcount_end(&devnet_rename_seq);
|
2009-11-18 02:36:59 +00:00
|
|
|
return err;
|
2012-11-26 05:21:08 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-07-31 00:03:38 +00:00
|
|
|
rollback:
|
2010-05-05 00:36:49 +00:00
|
|
|
ret = device_rename(&dev->dev, dev->name);
|
|
|
|
if (ret) {
|
|
|
|
memcpy(dev->name, oldname, IFNAMSIZ);
|
2012-12-20 17:25:08 +00:00
|
|
|
write_seqcount_end(&devnet_rename_seq);
|
2010-05-05 00:36:49 +00:00
|
|
|
return ret;
|
2008-05-15 05:33:38 +00:00
|
|
|
}
|
2007-07-30 23:35:46 +00:00
|
|
|
|
2012-12-20 17:25:08 +00:00
|
|
|
write_seqcount_end(&devnet_rename_seq);
|
2012-11-26 05:21:08 +00:00
|
|
|
|
2007-07-30 23:35:46 +00:00
|
|
|
write_lock_bh(&dev_base_lock);
|
2011-05-17 17:56:59 +00:00
|
|
|
hlist_del_rcu(&dev->name_hlist);
|
2009-10-30 07:11:27 +00:00
|
|
|
write_unlock_bh(&dev_base_lock);
|
|
|
|
|
|
|
|
synchronize_rcu();
|
|
|
|
|
|
|
|
write_lock_bh(&dev_base_lock);
|
|
|
|
hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
|
2007-07-30 23:35:46 +00:00
|
|
|
write_unlock_bh(&dev_base_lock);
|
|
|
|
|
2007-09-16 22:42:43 +00:00
|
|
|
ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
|
2007-07-31 00:03:38 +00:00
|
|
|
ret = notifier_to_errno(ret);
|
|
|
|
|
|
|
|
if (ret) {
|
2009-11-15 23:30:24 +00:00
|
|
|
/* err >= 0 after dev_alloc_name() or stores the first errno */
|
|
|
|
if (err >= 0) {
|
2007-07-31 00:03:38 +00:00
|
|
|
err = ret;
|
2012-12-20 17:25:08 +00:00
|
|
|
write_seqcount_begin(&devnet_rename_seq);
|
2007-07-31 00:03:38 +00:00
|
|
|
memcpy(dev->name, oldname, IFNAMSIZ);
|
|
|
|
goto rollback;
|
2009-11-15 23:30:24 +00:00
|
|
|
} else {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_err("%s: name change rollback failed: %d\n",
|
2009-11-15 23:30:24 +00:00
|
|
|
dev->name, ret);
|
2007-07-31 00:03:38 +00:00
|
|
|
}
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2008-09-23 04:28:11 +00:00
|
|
|
/**
|
|
|
|
* dev_set_alias - change ifalias of a device
|
|
|
|
* @dev: device
|
|
|
|
* @alias: name up to IFALIASZ
|
2008-09-30 09:23:58 +00:00
|
|
|
* @len: limit of bytes to copy from info
|
2008-09-23 04:28:11 +00:00
|
|
|
*
|
|
|
|
* Set ifalias for a device,
|
|
|
|
*/
|
|
|
|
int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
|
|
|
|
{
|
2012-08-08 00:33:25 +00:00
|
|
|
char *new_ifalias;
|
|
|
|
|
2008-09-23 04:28:11 +00:00
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
if (len >= IFALIASZ)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2008-09-24 04:23:19 +00:00
|
|
|
if (!len) {
|
2012-11-20 00:57:04 +00:00
|
|
|
kfree(dev->ifalias);
|
|
|
|
dev->ifalias = NULL;
|
2008-09-24 04:23:19 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2012-08-08 00:33:25 +00:00
|
|
|
new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
|
|
|
|
if (!new_ifalias)
|
2008-09-23 04:28:11 +00:00
|
|
|
return -ENOMEM;
|
2012-08-08 00:33:25 +00:00
|
|
|
dev->ifalias = new_ifalias;
|
2008-09-23 04:28:11 +00:00
|
|
|
|
|
|
|
strlcpy(dev->ifalias, alias, len+1);
|
|
|
|
return len;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-05-29 21:13:47 +00:00
|
|
|
/**
|
2006-05-26 20:25:24 +00:00
|
|
|
* netdev_features_change - device changes features
|
2005-05-29 21:13:47 +00:00
|
|
|
* @dev: device to cause notification
|
|
|
|
*
|
|
|
|
* Called to indicate a device has changed features.
|
|
|
|
*/
|
|
|
|
void netdev_features_change(struct net_device *dev)
|
|
|
|
{
|
2007-09-16 22:42:43 +00:00
|
|
|
call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
|
2005-05-29 21:13:47 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_features_change);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* netdev_state_change - device changes state
|
|
|
|
* @dev: device to cause notification
|
|
|
|
*
|
|
|
|
* Called to indicate a device has changed state. This function calls
|
|
|
|
* the notifier chains for netdev_chain and sends a NEWLINK message
|
|
|
|
* to the routing socket.
|
|
|
|
*/
|
|
|
|
void netdev_state_change(struct net_device *dev)
|
|
|
|
{
|
|
|
|
if (dev->flags & IFF_UP) {
|
2007-09-16 22:42:43 +00:00
|
|
|
call_netdevice_notifiers(NETDEV_CHANGE, dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
|
|
|
|
}
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(netdev_state_change);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2012-08-09 22:14:56 +00:00
|
|
|
/**
|
|
|
|
* netdev_notify_peers - notify network peers about existence of @dev
|
|
|
|
* @dev: network device
|
|
|
|
*
|
|
|
|
* Generate traffic such that interested network peers are aware of
|
|
|
|
* @dev, such as by generating a gratuitous ARP. This may be used when
|
|
|
|
* a device wants to inform the rest of the network about some sort of
|
|
|
|
* reconfiguration such as a failover event or virtual machine
|
|
|
|
* migration.
|
|
|
|
*/
|
|
|
|
void netdev_notify_peers(struct net_device *dev)
|
2008-06-14 01:12:00 +00:00
|
|
|
{
|
2012-08-09 22:14:56 +00:00
|
|
|
rtnl_lock();
|
|
|
|
call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
|
|
|
|
rtnl_unlock();
|
2008-06-14 01:12:00 +00:00
|
|
|
}
|
2012-08-09 22:14:56 +00:00
|
|
|
EXPORT_SYMBOL(netdev_notify_peers);
|
2008-06-14 01:12:00 +00:00
|
|
|
|
2010-02-26 06:34:53 +00:00
|
|
|
static int __dev_open(struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-20 05:32:24 +00:00
|
|
|
const struct net_device_ops *ops = dev->netdev_ops;
|
2009-05-29 23:39:53 +00:00
|
|
|
int ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-05-08 09:53:17 +00:00
|
|
|
ASSERT_RTNL();
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
if (!netif_device_present(dev))
|
|
|
|
return -ENODEV;
|
|
|
|
|
netpoll: protect napi_poll and poll_controller during dev_[open|close]
Ivan Vercera was recently backporting commit
9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that,
while this patch protects the tg3 driver from having its ndo_poll_controller
routine called during device initalization, it does nothing for the driver
during shutdown. I.e. it would be entirely possible to have the
ndo_poll_controller method (or subsequently the ndo_poll) routine called for a
driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or
ndo_open routine could be called. Given that the two latter routines tend to
initizlize and free many data structures that the former two rely on, the result
can easily be data corruption or various other crashes. Furthermore, it seems
that this is potentially a problem with all net drivers that support netpoll,
and so this should ideally be fixed in a common path.
As Ben H Pointed out to me, we can't preform dev_open/dev_close in atomic
context, so I've come up with this solution. We can use a mutex to sleep in
open/close paths and just do a mutex_trylock in the napi poll path and abandon
the poll attempt if we're locked, as we'll just retry the poll on the next send
anyway.
I've tested this here by flooding netconsole with messages on a system whos nic
driver I modfied to periodically return NETDEV_TX_BUSY, so that the netpoll tx
workqueue would be forced to send frames and poll the device. While this was
going on I rapidly ifdown/up'ed the interface and watched for any problems.
I've not found any.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Ivan Vecera <ivecera@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Francois Romieu <romieu@fr.zoreil.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-02-05 08:05:43 +00:00
|
|
|
/* Block netpoll from trying to do any rx path servicing.
|
|
|
|
* If we don't do this there is a chance ndo_poll_controller
|
|
|
|
* or ndo_poll may be running while we open the device
|
|
|
|
*/
|
2013-05-27 19:53:31 +00:00
|
|
|
netpoll_rx_disable(dev);
|
netpoll: protect napi_poll and poll_controller during dev_[open|close]
Ivan Vercera was recently backporting commit
9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that,
while this patch protects the tg3 driver from having its ndo_poll_controller
routine called during device initalization, it does nothing for the driver
during shutdown. I.e. it would be entirely possible to have the
ndo_poll_controller method (or subsequently the ndo_poll) routine called for a
driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or
ndo_open routine could be called. Given that the two latter routines tend to
initizlize and free many data structures that the former two rely on, the result
can easily be data corruption or various other crashes. Furthermore, it seems
that this is potentially a problem with all net drivers that support netpoll,
and so this should ideally be fixed in a common path.
As Ben H Pointed out to me, we can't preform dev_open/dev_close in atomic
context, so I've come up with this solution. We can use a mutex to sleep in
open/close paths and just do a mutex_trylock in the napi poll path and abandon
the poll attempt if we're locked, as we'll just retry the poll on the next send
anyway.
I've tested this here by flooding netconsole with messages on a system whos nic
driver I modfied to periodically return NETDEV_TX_BUSY, so that the netpoll tx
workqueue would be forced to send frames and poll the device. While this was
going on I rapidly ifdown/up'ed the interface and watched for any problems.
I've not found any.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Ivan Vecera <ivecera@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Francois Romieu <romieu@fr.zoreil.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-02-05 08:05:43 +00:00
|
|
|
|
2009-05-29 23:39:53 +00:00
|
|
|
ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
|
|
|
|
ret = notifier_to_errno(ret);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
set_bit(__LINK_STATE_START, &dev->state);
|
2007-10-24 03:19:37 +00:00
|
|
|
|
2008-11-20 05:32:24 +00:00
|
|
|
if (ops->ndo_validate_addr)
|
|
|
|
ret = ops->ndo_validate_addr(dev);
|
2007-10-24 03:19:37 +00:00
|
|
|
|
2008-11-20 05:32:24 +00:00
|
|
|
if (!ret && ops->ndo_open)
|
|
|
|
ret = ops->ndo_open(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
netpoll: protect napi_poll and poll_controller during dev_[open|close]
Ivan Vercera was recently backporting commit
9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that,
while this patch protects the tg3 driver from having its ndo_poll_controller
routine called during device initalization, it does nothing for the driver
during shutdown. I.e. it would be entirely possible to have the
ndo_poll_controller method (or subsequently the ndo_poll) routine called for a
driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or
ndo_open routine could be called. Given that the two latter routines tend to
initizlize and free many data structures that the former two rely on, the result
can easily be data corruption or various other crashes. Furthermore, it seems
that this is potentially a problem with all net drivers that support netpoll,
and so this should ideally be fixed in a common path.
As Ben H Pointed out to me, we can't preform dev_open/dev_close in atomic
context, so I've come up with this solution. We can use a mutex to sleep in
open/close paths and just do a mutex_trylock in the napi poll path and abandon
the poll attempt if we're locked, as we'll just retry the poll on the next send
anyway.
I've tested this here by flooding netconsole with messages on a system whos nic
driver I modfied to periodically return NETDEV_TX_BUSY, so that the netpoll tx
workqueue would be forced to send frames and poll the device. While this was
going on I rapidly ifdown/up'ed the interface and watched for any problems.
I've not found any.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Ivan Vecera <ivecera@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Francois Romieu <romieu@fr.zoreil.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-02-05 08:05:43 +00:00
|
|
|
netpoll_rx_enable(dev);
|
|
|
|
|
2007-10-24 03:19:37 +00:00
|
|
|
if (ret)
|
|
|
|
clear_bit(__LINK_STATE_START, &dev->state);
|
|
|
|
else {
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->flags |= IFF_UP;
|
2009-02-07 06:06:43 +00:00
|
|
|
net_dmaengine_get();
|
2007-06-27 08:28:10 +00:00
|
|
|
dev_set_rx_mode(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
dev_activate(dev);
|
2012-07-05 01:23:25 +00:00
|
|
|
add_device_randomness(dev->dev_addr, dev->addr_len);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2007-10-24 03:19:37 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2010-02-26 06:34:53 +00:00
|
|
|
* dev_open - prepare an interface for use.
|
|
|
|
* @dev: device to open
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
2010-02-26 06:34:53 +00:00
|
|
|
* Takes a device from down to up state. The device's private open
|
|
|
|
* function is invoked and then the multicast lists are loaded. Finally
|
|
|
|
* the device is moved into the up state and a %NETDEV_UP message is
|
|
|
|
* sent to the netdev notifier chain.
|
|
|
|
*
|
|
|
|
* Calling this function on an active interface is a nop. On a failure
|
|
|
|
* a negative errno code is returned.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
2010-02-26 06:34:53 +00:00
|
|
|
int dev_open(struct net_device *dev)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (dev->flags & IFF_UP)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
ret = __dev_open(dev);
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
|
|
|
|
call_netdevice_notifiers(NETDEV_UP, dev);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_open);
|
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
static int __dev_close_many(struct list_head *head)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2010-12-13 12:44:07 +00:00
|
|
|
struct net_device *dev;
|
2008-05-08 09:53:17 +00:00
|
|
|
|
2010-02-26 06:34:53 +00:00
|
|
|
ASSERT_RTNL();
|
2007-09-12 12:33:25 +00:00
|
|
|
might_sleep();
|
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
list_for_each_entry(dev, head, unreg_list) {
|
|
|
|
call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
clear_bit(__LINK_STATE_START, &dev->state);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
/* Synchronize to scheduled poll. We cannot touch poll list, it
|
|
|
|
* can be even on different cpu. So just clear netif_running().
|
|
|
|
*
|
|
|
|
* dev->stop() will invoke napi_disable() on all of it's
|
|
|
|
* napi_struct instances on this device.
|
|
|
|
*/
|
|
|
|
smp_mb__after_clear_bit(); /* Commit netif_running(). */
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
dev_deactivate_many(head);
|
2008-02-13 07:10:11 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
list_for_each_entry(dev, head, unreg_list) {
|
|
|
|
const struct net_device_ops *ops = dev->netdev_ops;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
/*
|
|
|
|
* Call the device specific close. This cannot fail.
|
|
|
|
* Only if device is UP
|
|
|
|
*
|
|
|
|
* We allow it to be called even after a DETACH hot-plug
|
|
|
|
* event.
|
|
|
|
*/
|
|
|
|
if (ops->ndo_stop)
|
|
|
|
ops->ndo_stop(dev);
|
|
|
|
|
|
|
|
dev->flags &= ~IFF_UP;
|
|
|
|
net_dmaengine_put();
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int __dev_close(struct net_device *dev)
|
|
|
|
{
|
2011-02-17 22:54:38 +00:00
|
|
|
int retval;
|
2010-12-13 12:44:07 +00:00
|
|
|
LIST_HEAD(single);
|
|
|
|
|
netpoll: protect napi_poll and poll_controller during dev_[open|close]
Ivan Vercera was recently backporting commit
9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that,
while this patch protects the tg3 driver from having its ndo_poll_controller
routine called during device initalization, it does nothing for the driver
during shutdown. I.e. it would be entirely possible to have the
ndo_poll_controller method (or subsequently the ndo_poll) routine called for a
driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or
ndo_open routine could be called. Given that the two latter routines tend to
initizlize and free many data structures that the former two rely on, the result
can easily be data corruption or various other crashes. Furthermore, it seems
that this is potentially a problem with all net drivers that support netpoll,
and so this should ideally be fixed in a common path.
As Ben H Pointed out to me, we can't preform dev_open/dev_close in atomic
context, so I've come up with this solution. We can use a mutex to sleep in
open/close paths and just do a mutex_trylock in the napi poll path and abandon
the poll attempt if we're locked, as we'll just retry the poll on the next send
anyway.
I've tested this here by flooding netconsole with messages on a system whos nic
driver I modfied to periodically return NETDEV_TX_BUSY, so that the netpoll tx
workqueue would be forced to send frames and poll the device. While this was
going on I rapidly ifdown/up'ed the interface and watched for any problems.
I've not found any.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Ivan Vecera <ivecera@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Francois Romieu <romieu@fr.zoreil.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-02-05 08:05:43 +00:00
|
|
|
/* Temporarily disable netpoll until the interface is down */
|
2013-05-27 19:53:31 +00:00
|
|
|
netpoll_rx_disable(dev);
|
netpoll: protect napi_poll and poll_controller during dev_[open|close]
Ivan Vercera was recently backporting commit
9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that,
while this patch protects the tg3 driver from having its ndo_poll_controller
routine called during device initalization, it does nothing for the driver
during shutdown. I.e. it would be entirely possible to have the
ndo_poll_controller method (or subsequently the ndo_poll) routine called for a
driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or
ndo_open routine could be called. Given that the two latter routines tend to
initizlize and free many data structures that the former two rely on, the result
can easily be data corruption or various other crashes. Furthermore, it seems
that this is potentially a problem with all net drivers that support netpoll,
and so this should ideally be fixed in a common path.
As Ben H Pointed out to me, we can't preform dev_open/dev_close in atomic
context, so I've come up with this solution. We can use a mutex to sleep in
open/close paths and just do a mutex_trylock in the napi poll path and abandon
the poll attempt if we're locked, as we'll just retry the poll on the next send
anyway.
I've tested this here by flooding netconsole with messages on a system whos nic
driver I modfied to periodically return NETDEV_TX_BUSY, so that the netpoll tx
workqueue would be forced to send frames and poll the device. While this was
going on I rapidly ifdown/up'ed the interface and watched for any problems.
I've not found any.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Ivan Vecera <ivecera@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Francois Romieu <romieu@fr.zoreil.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-02-05 08:05:43 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
list_add(&dev->unreg_list, &single);
|
2011-02-17 22:54:38 +00:00
|
|
|
retval = __dev_close_many(&single);
|
|
|
|
list_del(&single);
|
netpoll: protect napi_poll and poll_controller during dev_[open|close]
Ivan Vercera was recently backporting commit
9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that,
while this patch protects the tg3 driver from having its ndo_poll_controller
routine called during device initalization, it does nothing for the driver
during shutdown. I.e. it would be entirely possible to have the
ndo_poll_controller method (or subsequently the ndo_poll) routine called for a
driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or
ndo_open routine could be called. Given that the two latter routines tend to
initizlize and free many data structures that the former two rely on, the result
can easily be data corruption or various other crashes. Furthermore, it seems
that this is potentially a problem with all net drivers that support netpoll,
and so this should ideally be fixed in a common path.
As Ben H Pointed out to me, we can't preform dev_open/dev_close in atomic
context, so I've come up with this solution. We can use a mutex to sleep in
open/close paths and just do a mutex_trylock in the napi poll path and abandon
the poll attempt if we're locked, as we'll just retry the poll on the next send
anyway.
I've tested this here by flooding netconsole with messages on a system whos nic
driver I modfied to periodically return NETDEV_TX_BUSY, so that the netpoll tx
workqueue would be forced to send frames and poll the device. While this was
going on I rapidly ifdown/up'ed the interface and watched for any problems.
I've not found any.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Ivan Vecera <ivecera@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Francois Romieu <romieu@fr.zoreil.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-02-05 08:05:43 +00:00
|
|
|
|
|
|
|
netpoll_rx_enable(dev);
|
2011-02-17 22:54:38 +00:00
|
|
|
return retval;
|
2010-12-13 12:44:07 +00:00
|
|
|
}
|
|
|
|
|
2011-01-19 21:23:22 +00:00
|
|
|
static int dev_close_many(struct list_head *head)
|
2010-12-13 12:44:07 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev, *tmp;
|
|
|
|
LIST_HEAD(tmp_list);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
list_for_each_entry_safe(dev, tmp, head, unreg_list)
|
|
|
|
if (!(dev->flags & IFF_UP))
|
|
|
|
list_move(&dev->unreg_list, &tmp_list);
|
|
|
|
|
|
|
|
__dev_close_many(head);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
list_for_each_entry(dev, head, unreg_list) {
|
|
|
|
rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
|
|
|
|
call_netdevice_notifiers(NETDEV_DOWN, dev);
|
|
|
|
}
|
2010-02-26 06:34:53 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
/* rollback_registered_many needs the complete original list */
|
|
|
|
list_splice(&tmp_list, head);
|
2010-02-26 06:34:53 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* dev_close - shutdown an interface.
|
|
|
|
* @dev: device to shutdown
|
|
|
|
*
|
|
|
|
* This function moves an active device into down state. A
|
|
|
|
* %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
|
|
|
|
* is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
|
|
|
|
* chain.
|
|
|
|
*/
|
|
|
|
int dev_close(struct net_device *dev)
|
|
|
|
{
|
2011-05-10 19:26:06 +00:00
|
|
|
if (dev->flags & IFF_UP) {
|
|
|
|
LIST_HEAD(single);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
netpoll: protect napi_poll and poll_controller during dev_[open|close]
Ivan Vercera was recently backporting commit
9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that,
while this patch protects the tg3 driver from having its ndo_poll_controller
routine called during device initalization, it does nothing for the driver
during shutdown. I.e. it would be entirely possible to have the
ndo_poll_controller method (or subsequently the ndo_poll) routine called for a
driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or
ndo_open routine could be called. Given that the two latter routines tend to
initizlize and free many data structures that the former two rely on, the result
can easily be data corruption or various other crashes. Furthermore, it seems
that this is potentially a problem with all net drivers that support netpoll,
and so this should ideally be fixed in a common path.
As Ben H Pointed out to me, we can't preform dev_open/dev_close in atomic
context, so I've come up with this solution. We can use a mutex to sleep in
open/close paths and just do a mutex_trylock in the napi poll path and abandon
the poll attempt if we're locked, as we'll just retry the poll on the next send
anyway.
I've tested this here by flooding netconsole with messages on a system whos nic
driver I modfied to periodically return NETDEV_TX_BUSY, so that the netpoll tx
workqueue would be forced to send frames and poll the device. While this was
going on I rapidly ifdown/up'ed the interface and watched for any problems.
I've not found any.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Ivan Vecera <ivecera@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Francois Romieu <romieu@fr.zoreil.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-02-05 08:05:43 +00:00
|
|
|
/* Block netpoll rx while the interface is going down */
|
2013-05-27 19:53:31 +00:00
|
|
|
netpoll_rx_disable(dev);
|
netpoll: protect napi_poll and poll_controller during dev_[open|close]
Ivan Vercera was recently backporting commit
9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that,
while this patch protects the tg3 driver from having its ndo_poll_controller
routine called during device initalization, it does nothing for the driver
during shutdown. I.e. it would be entirely possible to have the
ndo_poll_controller method (or subsequently the ndo_poll) routine called for a
driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or
ndo_open routine could be called. Given that the two latter routines tend to
initizlize and free many data structures that the former two rely on, the result
can easily be data corruption or various other crashes. Furthermore, it seems
that this is potentially a problem with all net drivers that support netpoll,
and so this should ideally be fixed in a common path.
As Ben H Pointed out to me, we can't preform dev_open/dev_close in atomic
context, so I've come up with this solution. We can use a mutex to sleep in
open/close paths and just do a mutex_trylock in the napi poll path and abandon
the poll attempt if we're locked, as we'll just retry the poll on the next send
anyway.
I've tested this here by flooding netconsole with messages on a system whos nic
driver I modfied to periodically return NETDEV_TX_BUSY, so that the netpoll tx
workqueue would be forced to send frames and poll the device. While this was
going on I rapidly ifdown/up'ed the interface and watched for any problems.
I've not found any.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Ivan Vecera <ivecera@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Francois Romieu <romieu@fr.zoreil.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-02-05 08:05:43 +00:00
|
|
|
|
2011-05-10 19:26:06 +00:00
|
|
|
list_add(&dev->unreg_list, &single);
|
|
|
|
dev_close_many(&single);
|
|
|
|
list_del(&single);
|
netpoll: protect napi_poll and poll_controller during dev_[open|close]
Ivan Vercera was recently backporting commit
9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that,
while this patch protects the tg3 driver from having its ndo_poll_controller
routine called during device initalization, it does nothing for the driver
during shutdown. I.e. it would be entirely possible to have the
ndo_poll_controller method (or subsequently the ndo_poll) routine called for a
driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or
ndo_open routine could be called. Given that the two latter routines tend to
initizlize and free many data structures that the former two rely on, the result
can easily be data corruption or various other crashes. Furthermore, it seems
that this is potentially a problem with all net drivers that support netpoll,
and so this should ideally be fixed in a common path.
As Ben H Pointed out to me, we can't preform dev_open/dev_close in atomic
context, so I've come up with this solution. We can use a mutex to sleep in
open/close paths and just do a mutex_trylock in the napi poll path and abandon
the poll attempt if we're locked, as we'll just retry the poll on the next send
anyway.
I've tested this here by flooding netconsole with messages on a system whos nic
driver I modfied to periodically return NETDEV_TX_BUSY, so that the netpoll tx
workqueue would be forced to send frames and poll the device. While this was
going on I rapidly ifdown/up'ed the interface and watched for any problems.
I've not found any.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Ivan Vecera <ivecera@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Francois Romieu <romieu@fr.zoreil.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-02-05 08:05:43 +00:00
|
|
|
|
|
|
|
netpoll_rx_enable(dev);
|
2011-05-10 19:26:06 +00:00
|
|
|
}
|
2013-05-27 19:53:31 +00:00
|
|
|
return 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_close);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
|
2008-06-19 23:15:47 +00:00
|
|
|
/**
|
|
|
|
* dev_disable_lro - disable Large Receive Offload on a device
|
|
|
|
* @dev: device
|
|
|
|
*
|
|
|
|
* Disable Large Receive Offload (LRO) on a net device. Must be
|
|
|
|
* called under RTNL. This is needed if received packets may be
|
|
|
|
* forwarded to another interface.
|
|
|
|
*/
|
|
|
|
void dev_disable_lro(struct net_device *dev)
|
|
|
|
{
|
2011-05-24 08:31:09 +00:00
|
|
|
/*
|
|
|
|
* If we're trying to disable lro on a vlan device
|
|
|
|
* use the underlying physical device instead
|
|
|
|
*/
|
|
|
|
if (is_vlan_dev(dev))
|
|
|
|
dev = vlan_dev_real_dev(dev);
|
|
|
|
|
2011-11-15 15:29:55 +00:00
|
|
|
dev->wanted_features &= ~NETIF_F_LRO;
|
|
|
|
netdev_update_features(dev);
|
2011-03-18 16:56:34 +00:00
|
|
|
|
2011-04-21 12:42:15 +00:00
|
|
|
if (unlikely(dev->features & NETIF_F_LRO))
|
|
|
|
netdev_WARN(dev, "failed to disable LRO!\n");
|
2008-06-19 23:15:47 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_disable_lro);
|
|
|
|
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
static int dev_boot_phase = 1;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* register_netdevice_notifier - register a network notifier block
|
|
|
|
* @nb: notifier
|
|
|
|
*
|
|
|
|
* Register a notifier to be called when network device events occur.
|
|
|
|
* The notifier passed is linked into the kernel structures and must
|
|
|
|
* not be reused until it has been unregistered. A negative errno code
|
|
|
|
* is returned on a failure.
|
|
|
|
*
|
|
|
|
* When registered all registration and up events are replayed
|
2007-02-09 14:24:36 +00:00
|
|
|
* to the new notifier to allow device to have a race free
|
2005-04-16 22:20:36 +00:00
|
|
|
* view of the network device list.
|
|
|
|
*/
|
|
|
|
|
|
|
|
int register_netdevice_notifier(struct notifier_block *nb)
|
|
|
|
{
|
|
|
|
struct net_device *dev;
|
2007-07-31 00:03:38 +00:00
|
|
|
struct net_device *last;
|
2007-09-17 18:56:21 +00:00
|
|
|
struct net *net;
|
2005-04-16 22:20:36 +00:00
|
|
|
int err;
|
|
|
|
|
|
|
|
rtnl_lock();
|
2006-05-09 22:23:03 +00:00
|
|
|
err = raw_notifier_chain_register(&netdev_chain, nb);
|
2007-07-31 00:03:38 +00:00
|
|
|
if (err)
|
|
|
|
goto unlock;
|
2007-09-17 18:56:21 +00:00
|
|
|
if (dev_boot_phase)
|
|
|
|
goto unlock;
|
|
|
|
for_each_net(net) {
|
|
|
|
for_each_netdev(net, dev) {
|
|
|
|
err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
|
|
|
|
err = notifier_to_errno(err);
|
|
|
|
if (err)
|
|
|
|
goto rollback;
|
|
|
|
|
|
|
|
if (!(dev->flags & IFF_UP))
|
|
|
|
continue;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
nb->notifier_call(nb, NETDEV_UP, dev);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2007-07-31 00:03:38 +00:00
|
|
|
|
|
|
|
unlock:
|
2005-04-16 22:20:36 +00:00
|
|
|
rtnl_unlock();
|
|
|
|
return err;
|
2007-07-31 00:03:38 +00:00
|
|
|
|
|
|
|
rollback:
|
|
|
|
last = dev;
|
2007-09-17 18:56:21 +00:00
|
|
|
for_each_net(net) {
|
|
|
|
for_each_netdev(net, dev) {
|
|
|
|
if (dev == last)
|
2011-12-01 04:43:07 +00:00
|
|
|
goto outroll;
|
2007-07-31 00:03:38 +00:00
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
if (dev->flags & IFF_UP) {
|
|
|
|
nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
|
|
|
|
nb->notifier_call(nb, NETDEV_DOWN, dev);
|
|
|
|
}
|
|
|
|
nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
|
2007-07-31 00:03:38 +00:00
|
|
|
}
|
|
|
|
}
|
2007-11-14 23:53:16 +00:00
|
|
|
|
2011-12-01 04:43:07 +00:00
|
|
|
outroll:
|
2007-11-14 23:53:16 +00:00
|
|
|
raw_notifier_chain_unregister(&netdev_chain, nb);
|
2007-07-31 00:03:38 +00:00
|
|
|
goto unlock;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(register_netdevice_notifier);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* unregister_netdevice_notifier - unregister a network notifier block
|
|
|
|
* @nb: notifier
|
|
|
|
*
|
|
|
|
* Unregister a notifier previously registered by
|
|
|
|
* register_netdevice_notifier(). The notifier is unlinked into the
|
|
|
|
* kernel structures and may then be reused. A negative errno code
|
|
|
|
* is returned on a failure.
|
2012-04-06 15:33:35 +00:00
|
|
|
*
|
|
|
|
* After unregistering unregister and down device events are synthesized
|
|
|
|
* for all devices on the device list to the removed notifier to remove
|
|
|
|
* the need for special case cleanup code.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
int unregister_netdevice_notifier(struct notifier_block *nb)
|
|
|
|
{
|
2012-04-06 15:33:35 +00:00
|
|
|
struct net_device *dev;
|
|
|
|
struct net *net;
|
2006-03-25 09:24:25 +00:00
|
|
|
int err;
|
|
|
|
|
|
|
|
rtnl_lock();
|
2006-05-09 22:23:03 +00:00
|
|
|
err = raw_notifier_chain_unregister(&netdev_chain, nb);
|
2012-04-06 15:33:35 +00:00
|
|
|
if (err)
|
|
|
|
goto unlock;
|
|
|
|
|
|
|
|
for_each_net(net) {
|
|
|
|
for_each_netdev(net, dev) {
|
|
|
|
if (dev->flags & IFF_UP) {
|
|
|
|
nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
|
|
|
|
nb->notifier_call(nb, NETDEV_DOWN, dev);
|
|
|
|
}
|
|
|
|
nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
unlock:
|
2006-03-25 09:24:25 +00:00
|
|
|
rtnl_unlock();
|
|
|
|
return err;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(unregister_netdevice_notifier);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* call_netdevice_notifiers - call all network notifier blocks
|
|
|
|
* @val: value passed unmodified to notifier function
|
2007-10-13 04:17:49 +00:00
|
|
|
* @dev: net_device pointer passed unmodified to notifier function
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* Call all network notifier blocks. Parameters and return value
|
2006-05-09 22:23:03 +00:00
|
|
|
* are as for raw_notifier_call_chain().
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
|
2007-09-16 22:33:32 +00:00
|
|
|
int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2010-04-20 08:45:37 +00:00
|
|
|
ASSERT_RTNL();
|
2007-09-16 22:33:32 +00:00
|
|
|
return raw_notifier_call_chain(&netdev_chain, val, dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2011-03-24 13:24:01 +00:00
|
|
|
EXPORT_SYMBOL(call_netdevice_notifiers);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2012-02-24 07:31:31 +00:00
|
|
|
static struct static_key netstamp_needed __read_mostly;
|
2011-11-28 11:16:50 +00:00
|
|
|
#ifdef HAVE_JUMP_LABEL
|
2012-02-24 07:31:31 +00:00
|
|
|
/* We are not allowed to call static_key_slow_dec() from irq context
|
2011-11-28 11:16:50 +00:00
|
|
|
* If net_disable_timestamp() is called from irq context, defer the
|
2012-02-24 07:31:31 +00:00
|
|
|
* static_key_slow_dec() calls.
|
2011-11-28 11:16:50 +00:00
|
|
|
*/
|
|
|
|
static atomic_t netstamp_needed_deferred;
|
|
|
|
#endif
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
void net_enable_timestamp(void)
|
|
|
|
{
|
2011-11-28 11:16:50 +00:00
|
|
|
#ifdef HAVE_JUMP_LABEL
|
|
|
|
int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
|
|
|
|
|
|
|
|
if (deferred) {
|
|
|
|
while (--deferred)
|
2012-02-24 07:31:31 +00:00
|
|
|
static_key_slow_dec(&netstamp_needed);
|
2011-11-28 11:16:50 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
#endif
|
2012-02-24 07:31:31 +00:00
|
|
|
static_key_slow_inc(&netstamp_needed);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(net_enable_timestamp);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
void net_disable_timestamp(void)
|
|
|
|
{
|
2011-11-28 11:16:50 +00:00
|
|
|
#ifdef HAVE_JUMP_LABEL
|
|
|
|
if (in_interrupt()) {
|
|
|
|
atomic_inc(&netstamp_needed_deferred);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
#endif
|
2012-02-24 07:31:31 +00:00
|
|
|
static_key_slow_dec(&netstamp_needed);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(net_disable_timestamp);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
net: Consistent skb timestamping
With RPS inclusion, skb timestamping is not consistent in RX path.
If netif_receive_skb() is used, its deferred after RPS dispatch.
If netif_rx() is used, its done before RPS dispatch.
This can give strange tcpdump timestamps results.
I think timestamping should be done as soon as possible in the receive
path, to get meaningful values (ie timestamps taken at the time packet
was delivered by NIC driver to our stack), even if NAPI already can
defer timestamping a bit (RPS can help to reduce the gap)
Tom Herbert prefer to sample timestamps after RPS dispatch. In case
sampling is expensive (HPET/acpi_pm on x86), this makes sense.
Let admins switch from one mode to another, using a new
sysctl, /proc/sys/net/core/netdev_tstamp_prequeue
Its default value (1), means timestamps are taken as soon as possible,
before backlog queueing, giving accurate timestamps.
Setting a 0 value permits to sample timestamps when processing backlog,
after RPS dispatch, to lower the load of the pre-RPS cpu.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-16 06:57:10 +00:00
|
|
|
static inline void net_timestamp_set(struct sk_buff *skb)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2011-11-15 04:12:55 +00:00
|
|
|
skb->tstamp.tv64 = 0;
|
2012-02-24 07:31:31 +00:00
|
|
|
if (static_key_false(&netstamp_needed))
|
2005-08-15 00:24:31 +00:00
|
|
|
__net_timestamp(skb);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2011-11-15 04:12:55 +00:00
|
|
|
#define net_timestamp_check(COND, SKB) \
|
2012-02-24 07:31:31 +00:00
|
|
|
if (static_key_false(&netstamp_needed)) { \
|
2011-11-15 04:12:55 +00:00
|
|
|
if ((COND) && !(SKB)->tstamp.tv64) \
|
|
|
|
__net_timestamp(SKB); \
|
|
|
|
} \
|
net: Consistent skb timestamping
With RPS inclusion, skb timestamping is not consistent in RX path.
If netif_receive_skb() is used, its deferred after RPS dispatch.
If netif_rx() is used, its done before RPS dispatch.
This can give strange tcpdump timestamps results.
I think timestamping should be done as soon as possible in the receive
path, to get meaningful values (ie timestamps taken at the time packet
was delivered by NIC driver to our stack), even if NAPI already can
defer timestamping a bit (RPS can help to reduce the gap)
Tom Herbert prefer to sample timestamps after RPS dispatch. In case
sampling is expensive (HPET/acpi_pm on x86), this makes sense.
Let admins switch from one mode to another, using a new
sysctl, /proc/sys/net/core/netdev_tstamp_prequeue
Its default value (1), means timestamps are taken as soon as possible,
before backlog queueing, giving accurate timestamps.
Setting a 0 value permits to sample timestamps when processing backlog,
after RPS dispatch, to lower the load of the pre-RPS cpu.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-16 06:57:10 +00:00
|
|
|
|
2011-03-30 09:42:17 +00:00
|
|
|
static inline bool is_skb_forwardable(struct net_device *dev,
|
|
|
|
struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
unsigned int len;
|
|
|
|
|
|
|
|
if (!(dev->flags & IFF_UP))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
|
|
|
|
if (skb->len <= len)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
/* if TSO is enabled, we don't care about the length as the packet
|
|
|
|
* could be forwarded without being segmented before
|
|
|
|
*/
|
|
|
|
if (skb_is_gso(skb))
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2009-11-26 06:07:08 +00:00
|
|
|
/**
|
|
|
|
* dev_forward_skb - loopback an skb to another netif
|
|
|
|
*
|
|
|
|
* @dev: destination network device
|
|
|
|
* @skb: buffer to forward
|
|
|
|
*
|
|
|
|
* return values:
|
|
|
|
* NET_RX_SUCCESS (no congestion)
|
2010-05-06 07:53:53 +00:00
|
|
|
* NET_RX_DROP (packet was dropped, but freed)
|
2009-11-26 06:07:08 +00:00
|
|
|
*
|
|
|
|
* dev_forward_skb can be used for injecting an skb from the
|
|
|
|
* start_xmit function of one device into the receive queue
|
|
|
|
* of another device.
|
|
|
|
*
|
|
|
|
* The receiving device may be in another namespace, so
|
|
|
|
* we have to clear all information in the skb that could
|
|
|
|
* impact namespace isolation.
|
|
|
|
*/
|
|
|
|
int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
|
|
|
|
{
|
2011-08-31 08:03:29 +00:00
|
|
|
if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
|
|
|
|
if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
|
|
|
|
atomic_long_inc(&dev->rx_dropped);
|
|
|
|
kfree_skb(skb);
|
|
|
|
return NET_RX_DROP;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-11-26 06:07:08 +00:00
|
|
|
skb_orphan(skb);
|
|
|
|
|
2011-03-30 09:42:17 +00:00
|
|
|
if (unlikely(!is_skb_forwardable(dev, skb))) {
|
2010-09-30 21:06:55 +00:00
|
|
|
atomic_long_inc(&dev->rx_dropped);
|
2010-05-06 07:53:53 +00:00
|
|
|
kfree_skb(skb);
|
2009-11-26 06:07:08 +00:00
|
|
|
return NET_RX_DROP;
|
2010-05-06 07:53:53 +00:00
|
|
|
}
|
2012-03-27 15:55:44 +00:00
|
|
|
skb->skb_iif = 0;
|
2012-05-11 03:03:34 +00:00
|
|
|
skb_dst_drop(skb);
|
2009-11-26 06:07:08 +00:00
|
|
|
skb->tstamp.tv64 = 0;
|
|
|
|
skb->pkt_type = PACKET_HOST;
|
|
|
|
skb->protocol = eth_type_trans(skb, dev);
|
2012-05-11 03:03:34 +00:00
|
|
|
skb->mark = 0;
|
|
|
|
secpath_reset(skb);
|
|
|
|
nf_reset(skb);
|
2013-04-05 18:42:05 +00:00
|
|
|
nf_reset_trace(skb);
|
2009-11-26 06:07:08 +00:00
|
|
|
return netif_rx(skb);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(dev_forward_skb);
|
|
|
|
|
2010-12-15 19:57:25 +00:00
|
|
|
static inline int deliver_skb(struct sk_buff *skb,
|
|
|
|
struct packet_type *pt_prev,
|
|
|
|
struct net_device *orig_dev)
|
|
|
|
{
|
2012-07-20 09:23:17 +00:00
|
|
|
if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
|
|
|
|
return -ENOMEM;
|
2010-12-15 19:57:25 +00:00
|
|
|
atomic_inc(&skb->users);
|
|
|
|
return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
|
|
|
|
}
|
|
|
|
|
2012-08-16 22:02:58 +00:00
|
|
|
static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
|
|
|
|
{
|
2012-11-06 02:10:10 +00:00
|
|
|
if (!ptype->af_packet_priv || !skb->sk)
|
2012-08-16 22:02:58 +00:00
|
|
|
return false;
|
|
|
|
|
|
|
|
if (ptype->id_match)
|
|
|
|
return ptype->id_match(ptype, skb->sk);
|
|
|
|
else if ((struct sock *)ptype->af_packet_priv == skb->sk)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/*
|
|
|
|
* Support routine. Sends outgoing frames to any network
|
|
|
|
* taps currently in use.
|
|
|
|
*/
|
|
|
|
|
2006-06-22 09:57:17 +00:00
|
|
|
static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct packet_type *ptype;
|
2010-12-15 19:57:25 +00:00
|
|
|
struct sk_buff *skb2 = NULL;
|
|
|
|
struct packet_type *pt_prev = NULL;
|
2005-08-15 00:24:31 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
rcu_read_lock();
|
|
|
|
list_for_each_entry_rcu(ptype, &ptype_all, list) {
|
|
|
|
/* Never send packets back to the socket
|
|
|
|
* they originated from - MvS (miquels@drinkel.ow.org)
|
|
|
|
*/
|
|
|
|
if ((ptype->dev == dev || !ptype->dev) &&
|
2012-08-16 22:02:58 +00:00
|
|
|
(!skb_loop_sk(ptype, skb))) {
|
2010-12-15 19:57:25 +00:00
|
|
|
if (pt_prev) {
|
|
|
|
deliver_skb(skb2, pt_prev, skb->dev);
|
|
|
|
pt_prev = ptype;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
skb2 = skb_clone(skb, GFP_ATOMIC);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (!skb2)
|
|
|
|
break;
|
|
|
|
|
2010-12-20 21:22:51 +00:00
|
|
|
net_timestamp_set(skb2);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* skb->nh should be correctly
|
|
|
|
set by sender, so that the second statement is
|
|
|
|
just protection against buggy protocols.
|
|
|
|
*/
|
2007-03-19 22:30:44 +00:00
|
|
|
skb_reset_mac_header(skb2);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-04-11 03:50:43 +00:00
|
|
|
if (skb_network_header(skb2) < skb2->data ||
|
2007-04-20 03:29:13 +00:00
|
|
|
skb2->network_header > skb2->tail) {
|
2012-05-13 21:56:26 +00:00
|
|
|
net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
|
|
|
|
ntohs(skb2->protocol),
|
|
|
|
dev->name);
|
2007-04-11 03:45:18 +00:00
|
|
|
skb_reset_network_header(skb2);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2007-04-11 04:21:55 +00:00
|
|
|
skb2->transport_header = skb2->network_header;
|
2005-04-16 22:20:36 +00:00
|
|
|
skb2->pkt_type = PACKET_OUTGOING;
|
2010-12-15 19:57:25 +00:00
|
|
|
pt_prev = ptype;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
2010-12-15 19:57:25 +00:00
|
|
|
if (pt_prev)
|
|
|
|
pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
|
|
|
|
2012-07-10 10:55:09 +00:00
|
|
|
/**
|
|
|
|
* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
|
2011-01-17 08:06:04 +00:00
|
|
|
* @dev: Network device
|
|
|
|
* @txq: number of queues available
|
|
|
|
*
|
|
|
|
* If real_num_tx_queues is changed the tc mappings may no longer be
|
|
|
|
* valid. To resolve this verify the tc mapping remains valid and if
|
|
|
|
* not NULL the mapping. With no priorities mapping to this
|
|
|
|
* offset/count pair it will no longer be used. In the worst case TC0
|
|
|
|
* is invalid nothing can be done so disable priority mappings. If is
|
|
|
|
* expected that drivers will fix this mapping if they can before
|
|
|
|
* calling netif_set_real_num_tx_queues.
|
|
|
|
*/
|
2011-01-20 19:18:08 +00:00
|
|
|
static void netif_setup_tc(struct net_device *dev, unsigned int txq)
|
2011-01-17 08:06:04 +00:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
|
|
|
|
|
|
|
|
/* If TC0 is invalidated disable TC mapping */
|
|
|
|
if (tc->offset + tc->count > txq) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
|
2011-01-17 08:06:04 +00:00
|
|
|
dev->num_tc = 0;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Invalidated prio to tc mappings set to TC0 */
|
|
|
|
for (i = 1; i < TC_BITMASK + 1; i++) {
|
|
|
|
int q = netdev_get_prio_tc_map(dev, i);
|
|
|
|
|
|
|
|
tc = &dev->tc_to_txq[q];
|
|
|
|
if (tc->offset + tc->count > txq) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
|
|
|
|
i, q);
|
2011-01-17 08:06:04 +00:00
|
|
|
netdev_set_prio_tc_map(dev, i, 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-01-10 08:57:02 +00:00
|
|
|
#ifdef CONFIG_XPS
|
|
|
|
static DEFINE_MUTEX(xps_map_mutex);
|
|
|
|
#define xmap_dereference(P) \
|
|
|
|
rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
|
|
|
|
|
2013-01-10 08:57:17 +00:00
|
|
|
static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
|
|
|
|
int cpu, u16 index)
|
2013-01-10 08:57:02 +00:00
|
|
|
{
|
2013-01-10 08:57:17 +00:00
|
|
|
struct xps_map *map = NULL;
|
|
|
|
int pos;
|
2013-01-10 08:57:02 +00:00
|
|
|
|
2013-01-10 08:57:17 +00:00
|
|
|
if (dev_maps)
|
|
|
|
map = xmap_dereference(dev_maps->cpu_map[cpu]);
|
2013-01-10 08:57:02 +00:00
|
|
|
|
2013-01-10 08:57:17 +00:00
|
|
|
for (pos = 0; map && pos < map->len; pos++) {
|
|
|
|
if (map->queues[pos] == index) {
|
2013-01-10 08:57:02 +00:00
|
|
|
if (map->len > 1) {
|
|
|
|
map->queues[pos] = map->queues[--map->len];
|
|
|
|
} else {
|
2013-01-10 08:57:17 +00:00
|
|
|
RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
|
2013-01-10 08:57:02 +00:00
|
|
|
kfree_rcu(map, rcu);
|
|
|
|
map = NULL;
|
|
|
|
}
|
2013-01-10 08:57:17 +00:00
|
|
|
break;
|
2013-01-10 08:57:02 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-01-10 08:57:17 +00:00
|
|
|
return map;
|
|
|
|
}
|
|
|
|
|
2013-01-10 08:57:46 +00:00
|
|
|
static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
|
2013-01-10 08:57:17 +00:00
|
|
|
{
|
|
|
|
struct xps_dev_maps *dev_maps;
|
2013-01-10 08:57:46 +00:00
|
|
|
int cpu, i;
|
2013-01-10 08:57:17 +00:00
|
|
|
bool active = false;
|
|
|
|
|
|
|
|
mutex_lock(&xps_map_mutex);
|
|
|
|
dev_maps = xmap_dereference(dev->xps_maps);
|
|
|
|
|
|
|
|
if (!dev_maps)
|
|
|
|
goto out_no_maps;
|
|
|
|
|
|
|
|
for_each_possible_cpu(cpu) {
|
2013-01-10 08:57:46 +00:00
|
|
|
for (i = index; i < dev->num_tx_queues; i++) {
|
|
|
|
if (!remove_xps_queue(dev_maps, cpu, i))
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (i == dev->num_tx_queues)
|
2013-01-10 08:57:17 +00:00
|
|
|
active = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!active) {
|
2013-01-10 08:57:02 +00:00
|
|
|
RCU_INIT_POINTER(dev->xps_maps, NULL);
|
|
|
|
kfree_rcu(dev_maps, rcu);
|
|
|
|
}
|
|
|
|
|
2013-01-10 08:57:46 +00:00
|
|
|
for (i = index; i < dev->num_tx_queues; i++)
|
|
|
|
netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
|
|
|
|
NUMA_NO_NODE);
|
|
|
|
|
2013-01-10 08:57:02 +00:00
|
|
|
out_no_maps:
|
|
|
|
mutex_unlock(&xps_map_mutex);
|
|
|
|
}
|
|
|
|
|
2013-01-10 08:57:35 +00:00
|
|
|
static struct xps_map *expand_xps_map(struct xps_map *map,
|
|
|
|
int cpu, u16 index)
|
|
|
|
{
|
|
|
|
struct xps_map *new_map;
|
|
|
|
int alloc_len = XPS_MIN_MAP_ALLOC;
|
|
|
|
int i, pos;
|
|
|
|
|
|
|
|
for (pos = 0; map && pos < map->len; pos++) {
|
|
|
|
if (map->queues[pos] != index)
|
|
|
|
continue;
|
|
|
|
return map;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Need to add queue to this CPU's existing map */
|
|
|
|
if (map) {
|
|
|
|
if (pos < map->alloc_len)
|
|
|
|
return map;
|
|
|
|
|
|
|
|
alloc_len = map->alloc_len * 2;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Need to allocate new map to store queue on this CPU's map */
|
|
|
|
new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
|
|
|
|
cpu_to_node(cpu));
|
|
|
|
if (!new_map)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
for (i = 0; i < pos; i++)
|
|
|
|
new_map->queues[i] = map->queues[i];
|
|
|
|
new_map->alloc_len = alloc_len;
|
|
|
|
new_map->len = pos;
|
|
|
|
|
|
|
|
return new_map;
|
|
|
|
}
|
|
|
|
|
2013-01-10 08:57:02 +00:00
|
|
|
int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
|
|
|
|
{
|
2013-01-10 08:57:35 +00:00
|
|
|
struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
|
2013-01-10 08:57:02 +00:00
|
|
|
struct xps_map *map, *new_map;
|
|
|
|
int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
|
2013-01-10 08:57:35 +00:00
|
|
|
int cpu, numa_node_id = -2;
|
|
|
|
bool active = false;
|
2013-01-10 08:57:02 +00:00
|
|
|
|
|
|
|
mutex_lock(&xps_map_mutex);
|
|
|
|
|
|
|
|
dev_maps = xmap_dereference(dev->xps_maps);
|
|
|
|
|
2013-01-10 08:57:35 +00:00
|
|
|
/* allocate memory for queue storage */
|
|
|
|
for_each_online_cpu(cpu) {
|
|
|
|
if (!cpumask_test_cpu(cpu, mask))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (!new_dev_maps)
|
|
|
|
new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
|
2013-02-22 06:38:44 +00:00
|
|
|
if (!new_dev_maps) {
|
|
|
|
mutex_unlock(&xps_map_mutex);
|
2013-01-10 08:57:35 +00:00
|
|
|
return -ENOMEM;
|
2013-02-22 06:38:44 +00:00
|
|
|
}
|
2013-01-10 08:57:35 +00:00
|
|
|
|
|
|
|
map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
|
|
|
|
NULL;
|
|
|
|
|
|
|
|
map = expand_xps_map(map, cpu, index);
|
|
|
|
if (!map)
|
|
|
|
goto error;
|
|
|
|
|
|
|
|
RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!new_dev_maps)
|
|
|
|
goto out_no_new_maps;
|
|
|
|
|
2013-01-10 08:57:02 +00:00
|
|
|
for_each_possible_cpu(cpu) {
|
2013-01-10 08:57:35 +00:00
|
|
|
if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
|
|
|
|
/* add queue to CPU maps */
|
|
|
|
int pos = 0;
|
|
|
|
|
|
|
|
map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
|
|
|
|
while ((pos < map->len) && (map->queues[pos] != index))
|
|
|
|
pos++;
|
|
|
|
|
|
|
|
if (pos == map->len)
|
|
|
|
map->queues[map->len++] = index;
|
2013-01-10 08:57:02 +00:00
|
|
|
#ifdef CONFIG_NUMA
|
|
|
|
if (numa_node_id == -2)
|
|
|
|
numa_node_id = cpu_to_node(cpu);
|
|
|
|
else if (numa_node_id != cpu_to_node(cpu))
|
|
|
|
numa_node_id = -1;
|
|
|
|
#endif
|
2013-01-10 08:57:35 +00:00
|
|
|
} else if (dev_maps) {
|
|
|
|
/* fill in the new device map from the old device map */
|
|
|
|
map = xmap_dereference(dev_maps->cpu_map[cpu]);
|
|
|
|
RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
|
2013-01-10 08:57:02 +00:00
|
|
|
}
|
2013-01-10 08:57:35 +00:00
|
|
|
|
2013-01-10 08:57:02 +00:00
|
|
|
}
|
|
|
|
|
2013-01-10 08:57:35 +00:00
|
|
|
rcu_assign_pointer(dev->xps_maps, new_dev_maps);
|
|
|
|
|
2013-01-10 08:57:02 +00:00
|
|
|
/* Cleanup old maps */
|
2013-01-10 08:57:35 +00:00
|
|
|
if (dev_maps) {
|
|
|
|
for_each_possible_cpu(cpu) {
|
|
|
|
new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
|
|
|
|
map = xmap_dereference(dev_maps->cpu_map[cpu]);
|
|
|
|
if (map && map != new_map)
|
|
|
|
kfree_rcu(map, rcu);
|
|
|
|
}
|
2013-01-10 08:57:02 +00:00
|
|
|
|
2013-01-10 08:57:35 +00:00
|
|
|
kfree_rcu(dev_maps, rcu);
|
2013-01-10 08:57:02 +00:00
|
|
|
}
|
|
|
|
|
2013-01-10 08:57:35 +00:00
|
|
|
dev_maps = new_dev_maps;
|
|
|
|
active = true;
|
2013-01-10 08:57:02 +00:00
|
|
|
|
2013-01-10 08:57:35 +00:00
|
|
|
out_no_new_maps:
|
|
|
|
/* update Tx queue numa node */
|
2013-01-10 08:57:02 +00:00
|
|
|
netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
|
|
|
|
(numa_node_id >= 0) ? numa_node_id :
|
|
|
|
NUMA_NO_NODE);
|
|
|
|
|
2013-01-10 08:57:35 +00:00
|
|
|
if (!dev_maps)
|
|
|
|
goto out_no_maps;
|
|
|
|
|
|
|
|
/* removes queue from unused CPUs */
|
|
|
|
for_each_possible_cpu(cpu) {
|
|
|
|
if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (remove_xps_queue(dev_maps, cpu, index))
|
|
|
|
active = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* free map if not active */
|
|
|
|
if (!active) {
|
|
|
|
RCU_INIT_POINTER(dev->xps_maps, NULL);
|
|
|
|
kfree_rcu(dev_maps, rcu);
|
|
|
|
}
|
|
|
|
|
|
|
|
out_no_maps:
|
2013-01-10 08:57:02 +00:00
|
|
|
mutex_unlock(&xps_map_mutex);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
error:
|
2013-01-10 08:57:35 +00:00
|
|
|
/* remove any maps that we added */
|
|
|
|
for_each_possible_cpu(cpu) {
|
|
|
|
new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
|
|
|
|
map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
|
|
|
|
NULL;
|
|
|
|
if (new_map && new_map != map)
|
|
|
|
kfree(new_map);
|
|
|
|
}
|
|
|
|
|
2013-01-10 08:57:02 +00:00
|
|
|
mutex_unlock(&xps_map_mutex);
|
|
|
|
|
|
|
|
kfree(new_dev_maps);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_set_xps_queue);
|
|
|
|
|
|
|
|
#endif
|
2010-07-01 13:21:57 +00:00
|
|
|
/*
|
|
|
|
* Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
|
|
|
|
* greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
|
|
|
|
*/
|
2010-10-18 18:04:39 +00:00
|
|
|
int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
|
2010-07-01 13:21:57 +00:00
|
|
|
{
|
2010-11-21 13:17:27 +00:00
|
|
|
int rc;
|
|
|
|
|
2010-10-18 18:04:39 +00:00
|
|
|
if (txq < 1 || txq > dev->num_tx_queues)
|
|
|
|
return -EINVAL;
|
2010-07-01 13:21:57 +00:00
|
|
|
|
2011-02-15 19:39:21 +00:00
|
|
|
if (dev->reg_state == NETREG_REGISTERED ||
|
|
|
|
dev->reg_state == NETREG_UNREGISTERING) {
|
2010-10-18 18:04:39 +00:00
|
|
|
ASSERT_RTNL();
|
|
|
|
|
2010-11-21 13:17:27 +00:00
|
|
|
rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
|
|
|
|
txq);
|
2010-11-26 08:36:09 +00:00
|
|
|
if (rc)
|
|
|
|
return rc;
|
|
|
|
|
2011-01-17 08:06:04 +00:00
|
|
|
if (dev->num_tc)
|
|
|
|
netif_setup_tc(dev, txq);
|
|
|
|
|
2013-01-10 08:57:46 +00:00
|
|
|
if (txq < dev->real_num_tx_queues) {
|
2010-10-18 18:04:39 +00:00
|
|
|
qdisc_reset_all_tx_gt(dev, txq);
|
2013-01-10 08:57:46 +00:00
|
|
|
#ifdef CONFIG_XPS
|
|
|
|
netif_reset_xps_queues_gt(dev, txq);
|
|
|
|
#endif
|
|
|
|
}
|
2010-07-01 13:21:57 +00:00
|
|
|
}
|
2010-10-18 18:04:39 +00:00
|
|
|
|
|
|
|
dev->real_num_tx_queues = txq;
|
|
|
|
return 0;
|
2010-07-01 13:21:57 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_set_real_num_tx_queues);
|
2006-03-29 23:57:29 +00:00
|
|
|
|
2010-09-27 08:24:33 +00:00
|
|
|
#ifdef CONFIG_RPS
|
|
|
|
/**
|
|
|
|
* netif_set_real_num_rx_queues - set actual number of RX queues used
|
|
|
|
* @dev: Network device
|
|
|
|
* @rxq: Actual number of RX queues
|
|
|
|
*
|
|
|
|
* This must be called either with the rtnl_lock held or before
|
|
|
|
* registration of the net device. Returns 0 on success, or a
|
2010-10-08 17:33:39 +00:00
|
|
|
* negative error code. If called before registration, it always
|
|
|
|
* succeeds.
|
2010-09-27 08:24:33 +00:00
|
|
|
*/
|
|
|
|
int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
|
|
|
|
{
|
|
|
|
int rc;
|
|
|
|
|
2010-10-18 18:00:16 +00:00
|
|
|
if (rxq < 1 || rxq > dev->num_rx_queues)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2010-09-27 08:24:33 +00:00
|
|
|
if (dev->reg_state == NETREG_REGISTERED) {
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
|
|
|
|
rxq);
|
|
|
|
if (rc)
|
|
|
|
return rc;
|
|
|
|
}
|
|
|
|
|
|
|
|
dev->real_num_rx_queues = rxq;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_set_real_num_rx_queues);
|
|
|
|
#endif
|
|
|
|
|
2012-07-10 10:55:09 +00:00
|
|
|
/**
|
|
|
|
* netif_get_num_default_rss_queues - default number of RSS queues
|
2012-07-01 03:18:50 +00:00
|
|
|
*
|
|
|
|
* This routine should set an upper limit on the number of RSS queues
|
|
|
|
* used by default by multiqueue devices.
|
|
|
|
*/
|
2012-07-10 10:54:38 +00:00
|
|
|
int netif_get_num_default_rss_queues(void)
|
2012-07-01 03:18:50 +00:00
|
|
|
{
|
|
|
|
return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_get_num_default_rss_queues);
|
|
|
|
|
2008-08-18 04:54:43 +00:00
|
|
|
static inline void __netif_reschedule(struct Qdisc *q)
|
2006-03-29 23:57:29 +00:00
|
|
|
{
|
2008-08-18 04:54:43 +00:00
|
|
|
struct softnet_data *sd;
|
|
|
|
unsigned long flags;
|
2006-03-29 23:57:29 +00:00
|
|
|
|
2008-08-18 04:54:43 +00:00
|
|
|
local_irq_save(flags);
|
|
|
|
sd = &__get_cpu_var(softnet_data);
|
2010-04-26 23:06:24 +00:00
|
|
|
q->next_sched = NULL;
|
|
|
|
*sd->output_queue_tailp = q;
|
|
|
|
sd->output_queue_tailp = &q->next_sched;
|
2008-08-18 04:54:43 +00:00
|
|
|
raise_softirq_irqoff(NET_TX_SOFTIRQ);
|
|
|
|
local_irq_restore(flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
void __netif_schedule(struct Qdisc *q)
|
|
|
|
{
|
|
|
|
if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
|
|
|
|
__netif_reschedule(q);
|
2006-03-29 23:57:29 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(__netif_schedule);
|
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
void dev_kfree_skb_irq(struct sk_buff *skb)
|
2006-03-29 23:57:29 +00:00
|
|
|
{
|
2010-08-03 07:24:04 +00:00
|
|
|
if (atomic_dec_and_test(&skb->users)) {
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
struct softnet_data *sd;
|
|
|
|
unsigned long flags;
|
2006-03-29 23:57:29 +00:00
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
local_irq_save(flags);
|
|
|
|
sd = &__get_cpu_var(softnet_data);
|
|
|
|
skb->next = sd->completion_queue;
|
|
|
|
sd->completion_queue = skb;
|
|
|
|
raise_softirq_irqoff(NET_TX_SOFTIRQ);
|
|
|
|
local_irq_restore(flags);
|
|
|
|
}
|
2006-03-29 23:57:29 +00:00
|
|
|
}
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
EXPORT_SYMBOL(dev_kfree_skb_irq);
|
2006-03-29 23:57:29 +00:00
|
|
|
|
|
|
|
void dev_kfree_skb_any(struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
if (in_irq() || irqs_disabled())
|
|
|
|
dev_kfree_skb_irq(skb);
|
|
|
|
else
|
|
|
|
dev_kfree_skb(skb);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_kfree_skb_any);
|
|
|
|
|
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
/**
|
|
|
|
* netif_device_detach - mark device as removed
|
|
|
|
* @dev: network device
|
|
|
|
*
|
|
|
|
* Mark device as removed from system and therefore no longer available.
|
|
|
|
*/
|
2006-03-29 23:57:29 +00:00
|
|
|
void netif_device_detach(struct net_device *dev)
|
|
|
|
{
|
|
|
|
if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
|
|
|
|
netif_running(dev)) {
|
2009-04-08 13:15:22 +00:00
|
|
|
netif_tx_stop_all_queues(dev);
|
2006-03-29 23:57:29 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_device_detach);
|
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
/**
|
|
|
|
* netif_device_attach - mark device as attached
|
|
|
|
* @dev: network device
|
|
|
|
*
|
|
|
|
* Mark device as attached from system and restart if needed.
|
|
|
|
*/
|
2006-03-29 23:57:29 +00:00
|
|
|
void netif_device_attach(struct net_device *dev)
|
|
|
|
{
|
|
|
|
if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
|
|
|
|
netif_running(dev)) {
|
2009-04-08 13:15:22 +00:00
|
|
|
netif_tx_wake_all_queues(dev);
|
2007-02-09 14:24:36 +00:00
|
|
|
__netdev_watchdog_up(dev);
|
2006-03-29 23:57:29 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_device_attach);
|
|
|
|
|
2012-01-17 07:57:56 +00:00
|
|
|
static void skb_warn_bad_offload(const struct sk_buff *skb)
|
|
|
|
{
|
2012-01-17 10:00:40 +00:00
|
|
|
static const netdev_features_t null_features = 0;
|
2012-01-17 07:57:56 +00:00
|
|
|
struct net_device *dev = skb->dev;
|
|
|
|
const char *driver = "";
|
|
|
|
|
2013-04-19 10:45:52 +00:00
|
|
|
if (!net_ratelimit())
|
|
|
|
return;
|
|
|
|
|
2012-01-17 07:57:56 +00:00
|
|
|
if (dev && dev->dev.parent)
|
|
|
|
driver = dev_driver_string(dev->dev.parent);
|
|
|
|
|
|
|
|
WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
|
|
|
|
"gso_type=%d ip_summed=%d\n",
|
2012-01-17 10:00:40 +00:00
|
|
|
driver, dev ? &dev->features : &null_features,
|
|
|
|
skb->sk ? &skb->sk->sk_route_caps : &null_features,
|
2012-01-17 07:57:56 +00:00
|
|
|
skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
|
|
|
|
skb_shinfo(skb)->gso_type, skb->ip_summed);
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/*
|
|
|
|
* Invalidate hardware checksum when packet is to be mangled, and
|
|
|
|
* complete checksum manually on outgoing path.
|
|
|
|
*/
|
2006-08-29 23:44:56 +00:00
|
|
|
int skb_checksum_help(struct sk_buff *skb)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2006-11-15 05:24:49 +00:00
|
|
|
__wsum csum;
|
2007-04-09 18:59:07 +00:00
|
|
|
int ret = 0, offset;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-08-29 23:44:56 +00:00
|
|
|
if (skb->ip_summed == CHECKSUM_COMPLETE)
|
2006-07-08 20:34:56 +00:00
|
|
|
goto out_set_summed;
|
|
|
|
|
|
|
|
if (unlikely(skb_shinfo(skb)->gso_size)) {
|
2012-01-17 07:57:56 +00:00
|
|
|
skb_warn_bad_offload(skb);
|
|
|
|
return -EINVAL;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2013-01-25 20:34:37 +00:00
|
|
|
/* Before computing a checksum, we should make sure no frag could
|
|
|
|
* be modified by an external entity : checksum could be wrong.
|
|
|
|
*/
|
|
|
|
if (skb_has_shared_frag(skb)) {
|
|
|
|
ret = __skb_linearize(skb);
|
|
|
|
if (ret)
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2010-12-14 15:24:08 +00:00
|
|
|
offset = skb_checksum_start_offset(skb);
|
2007-10-15 08:47:15 +00:00
|
|
|
BUG_ON(offset >= skb_headlen(skb));
|
|
|
|
csum = skb_checksum(skb, offset, skb->len - offset, 0);
|
|
|
|
|
|
|
|
offset += skb->csum_offset;
|
|
|
|
BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
|
|
|
|
|
|
|
|
if (skb_cloned(skb) &&
|
|
|
|
!skb_clone_writable(skb, offset + sizeof(__sum16))) {
|
2005-04-16 22:20:36 +00:00
|
|
|
ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
|
|
|
|
if (ret)
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2007-10-15 08:47:15 +00:00
|
|
|
*(__sum16 *)(skb->data + offset) = csum_fold(csum);
|
2006-07-08 20:34:56 +00:00
|
|
|
out_set_summed:
|
2005-04-16 22:20:36 +00:00
|
|
|
skb->ip_summed = CHECKSUM_NONE;
|
2007-02-09 14:24:36 +00:00
|
|
|
out:
|
2005-04-16 22:20:36 +00:00
|
|
|
return ret;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(skb_checksum_help);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2013-03-07 09:28:01 +00:00
|
|
|
__be16 skb_network_protocol(struct sk_buff *skb)
|
2006-06-22 09:57:17 +00:00
|
|
|
{
|
2006-11-15 04:48:11 +00:00
|
|
|
__be16 type = skb->protocol;
|
2013-03-11 20:30:44 +00:00
|
|
|
int vlan_depth = ETH_HLEN;
|
2006-06-22 09:57:17 +00:00
|
|
|
|
2013-05-07 20:41:07 +00:00
|
|
|
/* Tunnel gso handlers can set protocol to ethernet. */
|
|
|
|
if (type == htons(ETH_P_TEB)) {
|
|
|
|
struct ethhdr *eth;
|
|
|
|
|
|
|
|
if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
eth = (struct ethhdr *)skb_mac_header(skb);
|
|
|
|
type = eth->h_proto;
|
|
|
|
}
|
|
|
|
|
net: vlan: add 802.1ad support
Add support for 802.1ad VLAN devices. This mainly consists of checking for
ETH_P_8021AD in addition to ETH_P_8021Q in a couple of places and check
offloading capabilities based on the used protocol.
Configuration is done using "ip link":
# ip link add link eth0 eth0.1000 \
type vlan proto 802.1ad id 1000
# ip link add link eth0.1000 eth0.1000.1000 \
type vlan proto 802.1q id 1000
52:54:00:12:34:56 > 92:b1:54:28:e4:8c, ethertype 802.1Q (0x8100), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto ICMP (1), length 84)
20.1.0.2 > 20.1.0.1: ICMP echo request, id 3003, seq 8, length 64
92:b1:54:28:e4:8c > 52:54:00:12:34:56, ethertype 802.1Q-QinQ (0x88a8), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 47944, offset 0, flags [none], proto ICMP (1), length 84)
20.1.0.1 > 20.1.0.2: ICMP echo reply, id 3003, seq 8, length 64
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-19 02:04:31 +00:00
|
|
|
while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
|
2010-10-29 12:14:54 +00:00
|
|
|
struct vlan_hdr *vh;
|
2010-10-20 13:56:04 +00:00
|
|
|
|
2010-10-29 12:14:54 +00:00
|
|
|
if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
|
2013-03-07 09:28:01 +00:00
|
|
|
return 0;
|
2010-10-20 13:56:04 +00:00
|
|
|
|
2010-10-29 12:14:54 +00:00
|
|
|
vh = (struct vlan_hdr *)(skb->data + vlan_depth);
|
|
|
|
type = vh->h_vlan_encapsulated_proto;
|
|
|
|
vlan_depth += VLAN_HLEN;
|
2010-10-20 13:56:04 +00:00
|
|
|
}
|
|
|
|
|
2013-03-07 09:28:01 +00:00
|
|
|
return type;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* skb_mac_gso_segment - mac layer segmentation handler.
|
|
|
|
* @skb: buffer to segment
|
|
|
|
* @features: features for the output path (see dev->features)
|
|
|
|
*/
|
|
|
|
struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
|
|
|
|
netdev_features_t features)
|
|
|
|
{
|
|
|
|
struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
|
|
|
|
struct packet_offload *ptype;
|
|
|
|
__be16 type = skb_network_protocol(skb);
|
|
|
|
|
|
|
|
if (unlikely(!type))
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
|
2006-06-22 09:57:17 +00:00
|
|
|
__skb_pull(skb, skb->mac_len);
|
|
|
|
|
|
|
|
rcu_read_lock();
|
2012-11-15 08:49:11 +00:00
|
|
|
list_for_each_entry_rcu(ptype, &offload_base, list) {
|
2012-11-15 08:49:23 +00:00
|
|
|
if (ptype->type == type && ptype->callbacks.gso_segment) {
|
2006-08-29 23:44:56 +00:00
|
|
|
if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
|
2013-02-14 09:44:55 +00:00
|
|
|
int err;
|
|
|
|
|
2012-11-15 08:49:23 +00:00
|
|
|
err = ptype->callbacks.gso_send_check(skb);
|
2006-07-08 20:34:56 +00:00
|
|
|
segs = ERR_PTR(err);
|
|
|
|
if (err || skb_gso_ok(skb, features))
|
|
|
|
break;
|
2007-04-11 03:50:43 +00:00
|
|
|
__skb_push(skb, (skb->data -
|
|
|
|
skb_network_header(skb)));
|
2006-07-08 20:34:56 +00:00
|
|
|
}
|
2012-11-15 08:49:23 +00:00
|
|
|
segs = ptype->callbacks.gso_segment(skb, features);
|
2006-06-22 09:57:17 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
2007-03-19 22:33:04 +00:00
|
|
|
__skb_push(skb, skb->data - skb_mac_header(skb));
|
2006-06-27 20:22:38 +00:00
|
|
|
|
2006-06-22 09:57:17 +00:00
|
|
|
return segs;
|
|
|
|
}
|
2013-02-14 09:44:55 +00:00
|
|
|
EXPORT_SYMBOL(skb_mac_gso_segment);
|
|
|
|
|
|
|
|
|
|
|
|
/* openvswitch calls this on rx path, so we need a different check.
|
|
|
|
*/
|
|
|
|
static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
|
|
|
|
{
|
|
|
|
if (tx_path)
|
|
|
|
return skb->ip_summed != CHECKSUM_PARTIAL;
|
|
|
|
else
|
|
|
|
return skb->ip_summed == CHECKSUM_NONE;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __skb_gso_segment - Perform segmentation on skb.
|
|
|
|
* @skb: buffer to segment
|
|
|
|
* @features: features for the output path (see dev->features)
|
|
|
|
* @tx_path: whether it is called in TX path
|
|
|
|
*
|
|
|
|
* This function segments the given skb and returns a list of segments.
|
|
|
|
*
|
|
|
|
* It may return NULL if the skb requires no segmentation. This is
|
|
|
|
* only possible when GSO is used for verifying header integrity.
|
|
|
|
*/
|
|
|
|
struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
|
|
|
|
netdev_features_t features, bool tx_path)
|
|
|
|
{
|
|
|
|
if (unlikely(skb_needs_check(skb, tx_path))) {
|
|
|
|
int err;
|
|
|
|
|
|
|
|
skb_warn_bad_offload(skb);
|
|
|
|
|
|
|
|
if (skb_header_cloned(skb) &&
|
|
|
|
(err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
|
|
|
|
return ERR_PTR(err);
|
|
|
|
}
|
|
|
|
|
2013-02-14 14:02:41 +00:00
|
|
|
SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
|
2013-02-14 09:44:55 +00:00
|
|
|
skb_reset_mac_header(skb);
|
|
|
|
skb_reset_mac_len(skb);
|
|
|
|
|
|
|
|
return skb_mac_gso_segment(skb, features);
|
|
|
|
}
|
2013-02-05 16:36:38 +00:00
|
|
|
EXPORT_SYMBOL(__skb_gso_segment);
|
2006-06-22 09:57:17 +00:00
|
|
|
|
2005-11-10 21:01:24 +00:00
|
|
|
/* Take action when hardware reception checksum errors are detected. */
|
|
|
|
#ifdef CONFIG_BUG
|
|
|
|
void netdev_rx_csum_fault(struct net_device *dev)
|
|
|
|
{
|
|
|
|
if (net_ratelimit()) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
|
2005-11-10 21:01:24 +00:00
|
|
|
dump_stack();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_rx_csum_fault);
|
|
|
|
#endif
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Actually, we should eliminate this check as soon as we know, that:
|
|
|
|
* 1. IOMMU is present and allows to map all the memory.
|
|
|
|
* 2. No high memory really exists on this machine.
|
|
|
|
*/
|
|
|
|
|
2010-04-02 20:34:49 +00:00
|
|
|
static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2006-06-27 20:33:10 +00:00
|
|
|
#ifdef CONFIG_HIGHMEM
|
2005-04-16 22:20:36 +00:00
|
|
|
int i;
|
2010-03-30 22:35:50 +00:00
|
|
|
if (!(dev->features & NETIF_F_HIGHDMA)) {
|
2011-08-22 23:44:58 +00:00
|
|
|
for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
|
|
|
|
skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
|
|
|
|
if (PageHighMem(skb_frag_page(frag)))
|
2010-03-30 22:35:50 +00:00
|
|
|
return 1;
|
2011-08-22 23:44:58 +00:00
|
|
|
}
|
2010-03-30 22:35:50 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-03-30 22:35:50 +00:00
|
|
|
if (PCI_DMA_BUS_IS_PHYS) {
|
|
|
|
struct device *pdev = dev->dev.parent;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-04-02 20:34:49 +00:00
|
|
|
if (!pdev)
|
|
|
|
return 0;
|
2010-03-30 22:35:50 +00:00
|
|
|
for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
|
2011-08-22 23:44:58 +00:00
|
|
|
skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
|
|
|
|
dma_addr_t addr = page_to_phys(skb_frag_page(frag));
|
2010-03-30 22:35:50 +00:00
|
|
|
if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
}
|
2006-06-27 20:33:10 +00:00
|
|
|
#endif
|
2005-04-16 22:20:36 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-06-22 09:57:17 +00:00
|
|
|
struct dev_gso_cb {
|
|
|
|
void (*destructor)(struct sk_buff *skb);
|
|
|
|
};
|
|
|
|
|
|
|
|
#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
|
|
|
|
|
|
|
|
static void dev_gso_skb_destructor(struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
struct dev_gso_cb *cb;
|
|
|
|
|
|
|
|
do {
|
|
|
|
struct sk_buff *nskb = skb->next;
|
|
|
|
|
|
|
|
skb->next = nskb->next;
|
|
|
|
nskb->next = NULL;
|
|
|
|
kfree_skb(nskb);
|
|
|
|
} while (skb->next);
|
|
|
|
|
|
|
|
cb = DEV_GSO_CB(skb);
|
|
|
|
if (cb->destructor)
|
|
|
|
cb->destructor(skb);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* dev_gso_segment - Perform emulated hardware segmentation on skb.
|
|
|
|
* @skb: buffer to segment
|
2011-01-09 06:23:33 +00:00
|
|
|
* @features: device features as applicable to this skb
|
2006-06-22 09:57:17 +00:00
|
|
|
*
|
|
|
|
* This function segments the given skb and stores the list of segments
|
|
|
|
* in skb->next.
|
|
|
|
*/
|
2011-11-15 15:29:55 +00:00
|
|
|
static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
|
2006-06-22 09:57:17 +00:00
|
|
|
{
|
|
|
|
struct sk_buff *segs;
|
2006-06-27 20:22:38 +00:00
|
|
|
|
|
|
|
segs = skb_gso_segment(skb, features);
|
|
|
|
|
|
|
|
/* Verifying header integrity only. */
|
|
|
|
if (!segs)
|
|
|
|
return 0;
|
2006-06-22 09:57:17 +00:00
|
|
|
|
2008-04-29 08:03:09 +00:00
|
|
|
if (IS_ERR(segs))
|
2006-06-22 09:57:17 +00:00
|
|
|
return PTR_ERR(segs);
|
|
|
|
|
|
|
|
skb->next = segs;
|
|
|
|
DEV_GSO_CB(skb)->destructor = skb->destructor;
|
|
|
|
skb->destructor = dev_gso_skb_destructor;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-15 15:29:55 +00:00
|
|
|
static netdev_features_t harmonize_features(struct sk_buff *skb,
|
|
|
|
__be16 protocol, netdev_features_t features)
|
2011-01-09 06:23:31 +00:00
|
|
|
{
|
2012-09-19 15:49:00 +00:00
|
|
|
if (skb->ip_summed != CHECKSUM_NONE &&
|
|
|
|
!can_checksum_protocol(features, protocol)) {
|
2011-01-09 06:23:31 +00:00
|
|
|
features &= ~NETIF_F_ALL_CSUM;
|
|
|
|
} else if (illegal_highdma(skb->dev, skb)) {
|
|
|
|
features &= ~NETIF_F_SG;
|
|
|
|
}
|
|
|
|
|
|
|
|
return features;
|
|
|
|
}
|
|
|
|
|
2011-11-15 15:29:55 +00:00
|
|
|
netdev_features_t netif_skb_features(struct sk_buff *skb)
|
2010-10-29 12:14:55 +00:00
|
|
|
{
|
|
|
|
__be16 protocol = skb->protocol;
|
2011-11-15 15:29:55 +00:00
|
|
|
netdev_features_t features = skb->dev->features;
|
2010-10-29 12:14:55 +00:00
|
|
|
|
2012-07-30 15:57:00 +00:00
|
|
|
if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
|
|
|
|
features &= ~NETIF_F_GSO_MASK;
|
|
|
|
|
net: vlan: add 802.1ad support
Add support for 802.1ad VLAN devices. This mainly consists of checking for
ETH_P_8021AD in addition to ETH_P_8021Q in a couple of places and check
offloading capabilities based on the used protocol.
Configuration is done using "ip link":
# ip link add link eth0 eth0.1000 \
type vlan proto 802.1ad id 1000
# ip link add link eth0.1000 eth0.1000.1000 \
type vlan proto 802.1q id 1000
52:54:00:12:34:56 > 92:b1:54:28:e4:8c, ethertype 802.1Q (0x8100), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto ICMP (1), length 84)
20.1.0.2 > 20.1.0.1: ICMP echo request, id 3003, seq 8, length 64
92:b1:54:28:e4:8c > 52:54:00:12:34:56, ethertype 802.1Q-QinQ (0x88a8), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 47944, offset 0, flags [none], proto ICMP (1), length 84)
20.1.0.1 > 20.1.0.2: ICMP echo reply, id 3003, seq 8, length 64
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-19 02:04:31 +00:00
|
|
|
if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
|
2010-10-29 12:14:55 +00:00
|
|
|
struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
|
|
|
|
protocol = veh->h_vlan_encapsulated_proto;
|
2011-01-09 06:23:31 +00:00
|
|
|
} else if (!vlan_tx_tag_present(skb)) {
|
|
|
|
return harmonize_features(skb, protocol, features);
|
|
|
|
}
|
2010-10-29 12:14:55 +00:00
|
|
|
|
net: vlan: add 802.1ad support
Add support for 802.1ad VLAN devices. This mainly consists of checking for
ETH_P_8021AD in addition to ETH_P_8021Q in a couple of places and check
offloading capabilities based on the used protocol.
Configuration is done using "ip link":
# ip link add link eth0 eth0.1000 \
type vlan proto 802.1ad id 1000
# ip link add link eth0.1000 eth0.1000.1000 \
type vlan proto 802.1q id 1000
52:54:00:12:34:56 > 92:b1:54:28:e4:8c, ethertype 802.1Q (0x8100), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto ICMP (1), length 84)
20.1.0.2 > 20.1.0.1: ICMP echo request, id 3003, seq 8, length 64
92:b1:54:28:e4:8c > 52:54:00:12:34:56, ethertype 802.1Q-QinQ (0x88a8), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 47944, offset 0, flags [none], proto ICMP (1), length 84)
20.1.0.1 > 20.1.0.2: ICMP echo reply, id 3003, seq 8, length 64
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-19 02:04:31 +00:00
|
|
|
features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
|
|
|
|
NETIF_F_HW_VLAN_STAG_TX);
|
2011-01-09 06:23:31 +00:00
|
|
|
|
net: vlan: add 802.1ad support
Add support for 802.1ad VLAN devices. This mainly consists of checking for
ETH_P_8021AD in addition to ETH_P_8021Q in a couple of places and check
offloading capabilities based on the used protocol.
Configuration is done using "ip link":
# ip link add link eth0 eth0.1000 \
type vlan proto 802.1ad id 1000
# ip link add link eth0.1000 eth0.1000.1000 \
type vlan proto 802.1q id 1000
52:54:00:12:34:56 > 92:b1:54:28:e4:8c, ethertype 802.1Q (0x8100), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto ICMP (1), length 84)
20.1.0.2 > 20.1.0.1: ICMP echo request, id 3003, seq 8, length 64
92:b1:54:28:e4:8c > 52:54:00:12:34:56, ethertype 802.1Q-QinQ (0x88a8), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 47944, offset 0, flags [none], proto ICMP (1), length 84)
20.1.0.1 > 20.1.0.2: ICMP echo reply, id 3003, seq 8, length 64
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-19 02:04:31 +00:00
|
|
|
if (protocol != htons(ETH_P_8021Q) && protocol != htons(ETH_P_8021AD)) {
|
2011-01-09 06:23:31 +00:00
|
|
|
return harmonize_features(skb, protocol, features);
|
|
|
|
} else {
|
|
|
|
features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
|
net: vlan: add 802.1ad support
Add support for 802.1ad VLAN devices. This mainly consists of checking for
ETH_P_8021AD in addition to ETH_P_8021Q in a couple of places and check
offloading capabilities based on the used protocol.
Configuration is done using "ip link":
# ip link add link eth0 eth0.1000 \
type vlan proto 802.1ad id 1000
# ip link add link eth0.1000 eth0.1000.1000 \
type vlan proto 802.1q id 1000
52:54:00:12:34:56 > 92:b1:54:28:e4:8c, ethertype 802.1Q (0x8100), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto ICMP (1), length 84)
20.1.0.2 > 20.1.0.1: ICMP echo request, id 3003, seq 8, length 64
92:b1:54:28:e4:8c > 52:54:00:12:34:56, ethertype 802.1Q-QinQ (0x88a8), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 47944, offset 0, flags [none], proto ICMP (1), length 84)
20.1.0.1 > 20.1.0.2: ICMP echo reply, id 3003, seq 8, length 64
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-19 02:04:31 +00:00
|
|
|
NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
|
|
|
|
NETIF_F_HW_VLAN_STAG_TX;
|
2011-01-09 06:23:31 +00:00
|
|
|
return harmonize_features(skb, protocol, features);
|
|
|
|
}
|
2010-10-29 12:14:55 +00:00
|
|
|
}
|
2011-01-09 06:23:31 +00:00
|
|
|
EXPORT_SYMBOL(netif_skb_features);
|
2010-10-29 12:14:55 +00:00
|
|
|
|
2010-06-16 14:18:12 +00:00
|
|
|
/*
|
|
|
|
* Returns true if either:
|
|
|
|
* 1. skb has frag_list and the device doesn't support FRAGLIST, or
|
2012-08-27 23:39:24 +00:00
|
|
|
* 2. skb is fragmented and the device does not support SG.
|
2010-06-16 14:18:12 +00:00
|
|
|
*/
|
|
|
|
static inline int skb_needs_linearize(struct sk_buff *skb,
|
2013-05-01 22:36:49 +00:00
|
|
|
netdev_features_t features)
|
2010-06-16 14:18:12 +00:00
|
|
|
{
|
2011-01-09 06:23:34 +00:00
|
|
|
return skb_is_nonlinear(skb) &&
|
|
|
|
((skb_has_frag_list(skb) &&
|
|
|
|
!(features & NETIF_F_FRAGLIST)) ||
|
2010-10-29 12:14:53 +00:00
|
|
|
(skb_shinfo(skb)->nr_frags &&
|
2011-01-09 06:23:34 +00:00
|
|
|
!(features & NETIF_F_SG)));
|
2010-06-16 14:18:12 +00:00
|
|
|
}
|
|
|
|
|
2008-07-17 08:56:23 +00:00
|
|
|
int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
|
|
|
|
struct netdev_queue *txq)
|
2006-06-22 09:57:17 +00:00
|
|
|
{
|
2008-11-21 04:14:53 +00:00
|
|
|
const struct net_device_ops *ops = dev->netdev_ops;
|
2009-11-10 06:14:14 +00:00
|
|
|
int rc = NETDEV_TX_OK;
|
2011-05-30 21:48:34 +00:00
|
|
|
unsigned int skb_len;
|
2008-11-21 04:14:53 +00:00
|
|
|
|
2006-06-22 09:57:17 +00:00
|
|
|
if (likely(!skb->next)) {
|
2011-11-15 15:29:55 +00:00
|
|
|
netdev_features_t features;
|
2011-01-09 06:23:32 +00:00
|
|
|
|
net: release dst entry in dev_hard_start_xmit()
One point of contention in high network loads is the dst_release() performed
when a transmited skb is freed. This is because NIC tx completion calls
dev_kree_skb() long after original call to dev_queue_xmit(skb).
CPU cache is cold and the atomic op in dst_release() stalls. On SMP, this is
quite visible if one CPU is 100% handling softirqs for a network device,
since dst_clone() is done by other cpus, involving cache line ping pongs.
It seems right place to release dst is in dev_hard_start_xmit(), for most
devices but ones that are virtual, and some exceptions.
David Miller suggested to define a new device flag, set in alloc_netdev_mq()
(so that most devices set it at init time), and carefuly unset in devices
which dont want a NULL skb->dst in their ndo_start_xmit().
List of devices that must clear this flag is :
- loopback device, because it calls netif_rx() and quoting Patrick :
"ip_route_input() doesn't accept loopback addresses, so loopback packets
already need to have a dst_entry attached."
- appletalk/ipddp.c : needs skb->dst in its xmit function
- And all devices that call again dev_queue_xmit() from their xmit function
(as some classifiers need skb->dst) : bonding, vlan, macvlan, eql, ifb, hdlc_fr
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-05-19 05:19:19 +00:00
|
|
|
/*
|
2011-03-31 01:57:33 +00:00
|
|
|
* If device doesn't need skb->dst, release it right now while
|
net: release dst entry in dev_hard_start_xmit()
One point of contention in high network loads is the dst_release() performed
when a transmited skb is freed. This is because NIC tx completion calls
dev_kree_skb() long after original call to dev_queue_xmit(skb).
CPU cache is cold and the atomic op in dst_release() stalls. On SMP, this is
quite visible if one CPU is 100% handling softirqs for a network device,
since dst_clone() is done by other cpus, involving cache line ping pongs.
It seems right place to release dst is in dev_hard_start_xmit(), for most
devices but ones that are virtual, and some exceptions.
David Miller suggested to define a new device flag, set in alloc_netdev_mq()
(so that most devices set it at init time), and carefuly unset in devices
which dont want a NULL skb->dst in their ndo_start_xmit().
List of devices that must clear this flag is :
- loopback device, because it calls netif_rx() and quoting Patrick :
"ip_route_input() doesn't accept loopback addresses, so loopback packets
already need to have a dst_entry attached."
- appletalk/ipddp.c : needs skb->dst in its xmit function
- And all devices that call again dev_queue_xmit() from their xmit function
(as some classifiers need skb->dst) : bonding, vlan, macvlan, eql, ifb, hdlc_fr
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-05-19 05:19:19 +00:00
|
|
|
* its hot in this cpu cache
|
|
|
|
*/
|
2009-06-02 05:19:30 +00:00
|
|
|
if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
|
|
|
|
skb_dst_drop(skb);
|
|
|
|
|
2011-01-09 06:23:32 +00:00
|
|
|
features = netif_skb_features(skb);
|
|
|
|
|
2010-10-20 13:56:04 +00:00
|
|
|
if (vlan_tx_tag_present(skb) &&
|
2013-04-19 02:04:30 +00:00
|
|
|
!vlan_hw_offload_capable(features, skb->vlan_proto)) {
|
|
|
|
skb = __vlan_put_tag(skb, skb->vlan_proto,
|
|
|
|
vlan_tx_tag_get(skb));
|
2010-10-20 13:56:04 +00:00
|
|
|
if (unlikely(!skb))
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
skb->vlan_tci = 0;
|
|
|
|
}
|
|
|
|
|
2012-12-07 14:14:15 +00:00
|
|
|
/* If encapsulation offload request, verify we are testing
|
|
|
|
* hardware encapsulation features instead of standard
|
|
|
|
* features for the netdev
|
|
|
|
*/
|
|
|
|
if (skb->encapsulation)
|
|
|
|
features &= dev->hw_enc_features;
|
|
|
|
|
2011-01-09 06:23:32 +00:00
|
|
|
if (netif_needs_gso(skb, features)) {
|
2011-01-09 06:23:33 +00:00
|
|
|
if (unlikely(dev_gso_segment(skb, features)))
|
2010-04-22 08:02:07 +00:00
|
|
|
goto out_kfree_skb;
|
|
|
|
if (skb->next)
|
|
|
|
goto gso;
|
2010-06-16 14:18:12 +00:00
|
|
|
} else {
|
2011-01-09 06:23:34 +00:00
|
|
|
if (skb_needs_linearize(skb, features) &&
|
2010-06-16 14:18:12 +00:00
|
|
|
__skb_linearize(skb))
|
|
|
|
goto out_kfree_skb;
|
|
|
|
|
|
|
|
/* If packet is not checksummed and device does not
|
|
|
|
* support checksumming for this protocol, complete
|
|
|
|
* checksumming here.
|
|
|
|
*/
|
|
|
|
if (skb->ip_summed == CHECKSUM_PARTIAL) {
|
2012-12-07 14:14:15 +00:00
|
|
|
if (skb->encapsulation)
|
|
|
|
skb_set_inner_transport_header(skb,
|
|
|
|
skb_checksum_start_offset(skb));
|
|
|
|
else
|
|
|
|
skb_set_transport_header(skb,
|
|
|
|
skb_checksum_start_offset(skb));
|
2011-01-09 06:23:35 +00:00
|
|
|
if (!(features & NETIF_F_ALL_CSUM) &&
|
2010-06-16 14:18:12 +00:00
|
|
|
skb_checksum_help(skb))
|
|
|
|
goto out_kfree_skb;
|
|
|
|
}
|
2010-04-22 08:02:07 +00:00
|
|
|
}
|
|
|
|
|
2012-09-18 20:44:49 +00:00
|
|
|
if (!list_empty(&ptype_all))
|
|
|
|
dev_queue_xmit_nit(skb, dev);
|
|
|
|
|
2011-05-30 21:48:34 +00:00
|
|
|
skb_len = skb->len;
|
2009-02-12 05:03:37 +00:00
|
|
|
rc = ops->ndo_start_xmit(skb, dev);
|
2011-05-30 21:48:34 +00:00
|
|
|
trace_net_dev_xmit(skb, rc, dev, skb_len);
|
2009-07-06 02:23:38 +00:00
|
|
|
if (rc == NETDEV_TX_OK)
|
2009-05-26 05:58:01 +00:00
|
|
|
txq_trans_update(txq);
|
2009-02-12 05:03:37 +00:00
|
|
|
return rc;
|
2006-06-22 09:57:17 +00:00
|
|
|
}
|
|
|
|
|
2006-06-27 20:22:38 +00:00
|
|
|
gso:
|
2006-06-22 09:57:17 +00:00
|
|
|
do {
|
|
|
|
struct sk_buff *nskb = skb->next;
|
|
|
|
|
|
|
|
skb->next = nskb->next;
|
|
|
|
nskb->next = NULL;
|
2009-12-09 20:59:58 +00:00
|
|
|
|
2012-09-18 20:44:49 +00:00
|
|
|
if (!list_empty(&ptype_all))
|
|
|
|
dev_queue_xmit_nit(nskb, dev);
|
|
|
|
|
2011-05-30 21:48:34 +00:00
|
|
|
skb_len = nskb->len;
|
2008-11-21 04:14:53 +00:00
|
|
|
rc = ops->ndo_start_xmit(nskb, dev);
|
2011-05-30 21:48:34 +00:00
|
|
|
trace_net_dev_xmit(nskb, rc, dev, skb_len);
|
2009-07-06 02:23:38 +00:00
|
|
|
if (unlikely(rc != NETDEV_TX_OK)) {
|
2009-11-10 06:14:14 +00:00
|
|
|
if (rc & ~NETDEV_TX_MASK)
|
|
|
|
goto out_kfree_gso_skb;
|
2006-06-26 06:57:04 +00:00
|
|
|
nskb->next = skb->next;
|
2006-06-22 09:57:17 +00:00
|
|
|
skb->next = nskb;
|
|
|
|
return rc;
|
|
|
|
}
|
2009-05-26 05:58:01 +00:00
|
|
|
txq_trans_update(txq);
|
2011-11-28 16:32:44 +00:00
|
|
|
if (unlikely(netif_xmit_stopped(txq) && skb->next))
|
2006-06-26 06:57:04 +00:00
|
|
|
return NETDEV_TX_BUSY;
|
2006-06-22 09:57:17 +00:00
|
|
|
} while (skb->next);
|
2007-02-09 14:24:36 +00:00
|
|
|
|
2009-11-10 06:14:14 +00:00
|
|
|
out_kfree_gso_skb:
|
2013-04-29 13:02:42 +00:00
|
|
|
if (likely(skb->next == NULL)) {
|
2009-11-10 06:14:14 +00:00
|
|
|
skb->destructor = DEV_GSO_CB(skb)->destructor;
|
2013-04-29 13:02:42 +00:00
|
|
|
consume_skb(skb);
|
|
|
|
return rc;
|
|
|
|
}
|
2006-06-22 09:57:17 +00:00
|
|
|
out_kfree_skb:
|
|
|
|
kfree_skb(skb);
|
2010-10-20 13:56:04 +00:00
|
|
|
out:
|
2009-11-10 06:14:14 +00:00
|
|
|
return rc;
|
2006-06-22 09:57:17 +00:00
|
|
|
}
|
|
|
|
|
2013-01-10 12:36:42 +00:00
|
|
|
static void qdisc_pkt_len_init(struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
const struct skb_shared_info *shinfo = skb_shinfo(skb);
|
|
|
|
|
|
|
|
qdisc_skb_cb(skb)->pkt_len = skb->len;
|
|
|
|
|
|
|
|
/* To get more precise estimation of bytes sent on wire,
|
|
|
|
* we add to pkt_len the headers size of all segments
|
|
|
|
*/
|
|
|
|
if (shinfo->gso_size) {
|
2013-01-16 05:14:21 +00:00
|
|
|
unsigned int hdr_len;
|
2013-03-25 20:19:59 +00:00
|
|
|
u16 gso_segs = shinfo->gso_segs;
|
2013-01-10 12:36:42 +00:00
|
|
|
|
2013-01-16 05:14:21 +00:00
|
|
|
/* mac layer + network layer */
|
|
|
|
hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
|
|
|
|
|
|
|
|
/* + transport layer */
|
2013-01-10 12:36:42 +00:00
|
|
|
if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
|
|
|
|
hdr_len += tcp_hdrlen(skb);
|
|
|
|
else
|
|
|
|
hdr_len += sizeof(struct udphdr);
|
2013-03-25 20:19:59 +00:00
|
|
|
|
|
|
|
if (shinfo->gso_type & SKB_GSO_DODGY)
|
|
|
|
gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
|
|
|
|
shinfo->gso_size);
|
|
|
|
|
|
|
|
qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
|
2013-01-10 12:36:42 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-08-06 01:44:21 +00:00
|
|
|
static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
|
|
|
|
struct net_device *dev,
|
|
|
|
struct netdev_queue *txq)
|
|
|
|
{
|
|
|
|
spinlock_t *root_lock = qdisc_lock(q);
|
2011-01-20 03:48:19 +00:00
|
|
|
bool contended;
|
2009-08-06 01:44:21 +00:00
|
|
|
int rc;
|
|
|
|
|
2013-01-10 12:36:42 +00:00
|
|
|
qdisc_pkt_len_init(skb);
|
2011-01-20 03:48:19 +00:00
|
|
|
qdisc_calculate_pkt_len(skb, q);
|
2010-06-02 12:09:29 +00:00
|
|
|
/*
|
|
|
|
* Heuristic to force contended enqueues to serialize on a
|
|
|
|
* separate lock before trying to get qdisc main lock.
|
|
|
|
* This permits __QDISC_STATE_RUNNING owner to get the lock more often
|
|
|
|
* and dequeue packets faster.
|
|
|
|
*/
|
2011-01-20 03:48:19 +00:00
|
|
|
contended = qdisc_is_running(q);
|
2010-06-02 12:09:29 +00:00
|
|
|
if (unlikely(contended))
|
|
|
|
spin_lock(&q->busylock);
|
|
|
|
|
2009-08-06 01:44:21 +00:00
|
|
|
spin_lock(root_lock);
|
|
|
|
if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
|
|
|
|
kfree_skb(skb);
|
|
|
|
rc = NET_XMIT_DROP;
|
|
|
|
} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
|
2010-06-02 10:23:51 +00:00
|
|
|
qdisc_run_begin(q)) {
|
2009-08-06 01:44:21 +00:00
|
|
|
/*
|
|
|
|
* This is a work-conserving queue; there are no old skbs
|
|
|
|
* waiting to be sent out; and the qdisc is not running -
|
|
|
|
* xmit the skb directly.
|
|
|
|
*/
|
2010-05-11 23:19:48 +00:00
|
|
|
if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
|
|
|
|
skb_dst_force(skb);
|
2011-01-09 08:30:54 +00:00
|
|
|
|
|
|
|
qdisc_bstats_update(q, skb);
|
|
|
|
|
2010-06-02 12:09:29 +00:00
|
|
|
if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
|
|
|
|
if (unlikely(contended)) {
|
|
|
|
spin_unlock(&q->busylock);
|
|
|
|
contended = false;
|
|
|
|
}
|
2009-08-06 01:44:21 +00:00
|
|
|
__qdisc_run(q);
|
2010-06-02 12:09:29 +00:00
|
|
|
} else
|
2010-06-02 10:23:51 +00:00
|
|
|
qdisc_run_end(q);
|
2009-08-06 01:44:21 +00:00
|
|
|
|
|
|
|
rc = NET_XMIT_SUCCESS;
|
|
|
|
} else {
|
2010-05-11 23:19:48 +00:00
|
|
|
skb_dst_force(skb);
|
2011-01-20 03:48:19 +00:00
|
|
|
rc = q->enqueue(skb, q) & NET_XMIT_MASK;
|
2010-06-02 12:09:29 +00:00
|
|
|
if (qdisc_run_begin(q)) {
|
|
|
|
if (unlikely(contended)) {
|
|
|
|
spin_unlock(&q->busylock);
|
|
|
|
contended = false;
|
|
|
|
}
|
|
|
|
__qdisc_run(q);
|
|
|
|
}
|
2009-08-06 01:44:21 +00:00
|
|
|
}
|
|
|
|
spin_unlock(root_lock);
|
2010-06-02 12:09:29 +00:00
|
|
|
if (unlikely(contended))
|
|
|
|
spin_unlock(&q->busylock);
|
2009-08-06 01:44:21 +00:00
|
|
|
return rc;
|
|
|
|
}
|
|
|
|
|
2011-11-22 05:10:51 +00:00
|
|
|
#if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
|
|
|
|
static void skb_update_prio(struct sk_buff *skb)
|
|
|
|
{
|
2011-11-25 07:44:54 +00:00
|
|
|
struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
|
2011-11-22 05:10:51 +00:00
|
|
|
|
2012-07-08 21:45:10 +00:00
|
|
|
if (!skb->priority && skb->sk && map) {
|
|
|
|
unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
|
|
|
|
|
|
|
|
if (prioidx < map->priomap_len)
|
|
|
|
skb->priority = map->priomap[prioidx];
|
|
|
|
}
|
2011-11-22 05:10:51 +00:00
|
|
|
}
|
|
|
|
#else
|
|
|
|
#define skb_update_prio(skb)
|
|
|
|
#endif
|
|
|
|
|
2010-09-29 20:23:09 +00:00
|
|
|
static DEFINE_PER_CPU(int, xmit_recursion);
|
2010-10-25 19:51:55 +00:00
|
|
|
#define RECURSION_LIMIT 10
|
2010-09-29 20:23:09 +00:00
|
|
|
|
2012-06-12 10:16:35 +00:00
|
|
|
/**
|
|
|
|
* dev_loopback_xmit - loop back @skb
|
|
|
|
* @skb: buffer to transmit
|
|
|
|
*/
|
|
|
|
int dev_loopback_xmit(struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
skb_reset_mac_header(skb);
|
|
|
|
__skb_pull(skb, skb_network_offset(skb));
|
|
|
|
skb->pkt_type = PACKET_LOOPBACK;
|
|
|
|
skb->ip_summed = CHECKSUM_UNNECESSARY;
|
|
|
|
WARN_ON(!skb_dst(skb));
|
|
|
|
skb_dst_force(skb);
|
|
|
|
netif_rx_ni(skb);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_loopback_xmit);
|
|
|
|
|
2008-07-22 21:09:06 +00:00
|
|
|
/**
|
|
|
|
* dev_queue_xmit - transmit a buffer
|
|
|
|
* @skb: buffer to transmit
|
|
|
|
*
|
|
|
|
* Queue a buffer for transmission to a network device. The caller must
|
|
|
|
* have set the device and priority and built the buffer before calling
|
|
|
|
* this function. The function can be called from an interrupt.
|
|
|
|
*
|
|
|
|
* A negative errno code is returned on a failure. A success does not
|
|
|
|
* guarantee the frame will be transmitted as it may be dropped due
|
|
|
|
* to congestion or traffic shaping.
|
|
|
|
*
|
|
|
|
* -----------------------------------------------------------------------------------
|
|
|
|
* I notice this method can also return errors from the queue disciplines,
|
|
|
|
* including NET_XMIT_DROP, which is a positive value. So, errors can also
|
|
|
|
* be positive.
|
|
|
|
*
|
|
|
|
* Regardless of the return value, the skb is consumed, so it is currently
|
|
|
|
* difficult to retry a send to this method. (You can bump the ref count
|
|
|
|
* before sending to hold a reference for retry if you are careful.)
|
|
|
|
*
|
|
|
|
* When calling this method, interrupts MUST be enabled. This is because
|
|
|
|
* the BH enable code must have IRQs enabled so that it will not deadlock.
|
|
|
|
* --BLG
|
|
|
|
*/
|
2005-04-16 22:20:36 +00:00
|
|
|
int dev_queue_xmit(struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
struct net_device *dev = skb->dev;
|
2008-07-09 00:18:23 +00:00
|
|
|
struct netdev_queue *txq;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct Qdisc *q;
|
|
|
|
int rc = -ENOMEM;
|
|
|
|
|
2013-02-05 20:22:20 +00:00
|
|
|
skb_reset_mac_header(skb);
|
|
|
|
|
2007-02-09 14:24:36 +00:00
|
|
|
/* Disable soft irqs for various locks below. Also
|
|
|
|
* stops preemption for RCU.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
2007-02-09 14:24:36 +00:00
|
|
|
rcu_read_lock_bh();
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2011-11-22 05:10:51 +00:00
|
|
|
skb_update_prio(skb);
|
|
|
|
|
2012-09-17 20:16:31 +00:00
|
|
|
txq = netdev_pick_tx(dev, skb);
|
2010-02-23 01:04:49 +00:00
|
|
|
q = rcu_dereference_bh(txq->qdisc);
|
2008-07-16 09:15:04 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
#ifdef CONFIG_NET_CLS_ACT
|
2009-09-03 08:29:39 +00:00
|
|
|
skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
|
2005-04-16 22:20:36 +00:00
|
|
|
#endif
|
2010-08-23 09:45:02 +00:00
|
|
|
trace_net_dev_queue(skb);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (q->enqueue) {
|
2009-08-06 01:44:21 +00:00
|
|
|
rc = __dev_xmit_skb(skb, q, dev, txq);
|
2008-07-16 09:15:04 +00:00
|
|
|
goto out;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* The device has no queue. Common case for software devices:
|
|
|
|
loopback, all the sorts of tunnels...
|
|
|
|
|
2006-06-09 19:20:56 +00:00
|
|
|
Really, it is unlikely that netif_tx_lock protection is necessary
|
|
|
|
here. (f.e. loopback and IP tunnels are clean ignoring statistics
|
2005-04-16 22:20:36 +00:00
|
|
|
counters.)
|
|
|
|
However, it is possible, that they rely on protection
|
|
|
|
made by us here.
|
|
|
|
|
|
|
|
Check this and shot the lock. It is not prone from deadlocks.
|
|
|
|
Either shot noqueue qdisc, it is even simpler 8)
|
|
|
|
*/
|
|
|
|
if (dev->flags & IFF_UP) {
|
|
|
|
int cpu = smp_processor_id(); /* ok because BHs are off */
|
|
|
|
|
2008-07-09 06:13:53 +00:00
|
|
|
if (txq->xmit_lock_owner != cpu) {
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-09-29 20:23:09 +00:00
|
|
|
if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
|
|
|
|
goto recursion_alert;
|
|
|
|
|
2008-07-09 06:13:53 +00:00
|
|
|
HARD_TX_LOCK(dev, txq, cpu);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2011-11-28 16:32:44 +00:00
|
|
|
if (!netif_xmit_stopped(txq)) {
|
2010-09-29 20:23:09 +00:00
|
|
|
__this_cpu_inc(xmit_recursion);
|
2009-11-10 06:14:14 +00:00
|
|
|
rc = dev_hard_start_xmit(skb, dev, txq);
|
2010-09-29 20:23:09 +00:00
|
|
|
__this_cpu_dec(xmit_recursion);
|
2009-11-10 06:14:14 +00:00
|
|
|
if (dev_xmit_complete(rc)) {
|
2008-07-09 06:13:53 +00:00
|
|
|
HARD_TX_UNLOCK(dev, txq);
|
2005-04-16 22:20:36 +00:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
2008-07-09 06:13:53 +00:00
|
|
|
HARD_TX_UNLOCK(dev, txq);
|
2012-05-13 21:56:26 +00:00
|
|
|
net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
|
|
|
|
dev->name);
|
2005-04-16 22:20:36 +00:00
|
|
|
} else {
|
|
|
|
/* Recursion is detected! It is possible,
|
2010-09-29 20:23:09 +00:00
|
|
|
* unfortunately
|
|
|
|
*/
|
|
|
|
recursion_alert:
|
2012-05-13 21:56:26 +00:00
|
|
|
net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
|
|
|
|
dev->name);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
rc = -ENETDOWN;
|
2006-06-22 09:28:18 +00:00
|
|
|
rcu_read_unlock_bh();
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
kfree_skb(skb);
|
|
|
|
return rc;
|
|
|
|
out:
|
2006-06-22 09:28:18 +00:00
|
|
|
rcu_read_unlock_bh();
|
2005-04-16 22:20:36 +00:00
|
|
|
return rc;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_queue_xmit);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
|
|
|
|
/*=======================================================================
|
|
|
|
Receiver routines
|
|
|
|
=======================================================================*/
|
|
|
|
|
2007-03-12 21:33:50 +00:00
|
|
|
int netdev_max_backlog __read_mostly = 1000;
|
2012-09-27 19:29:05 +00:00
|
|
|
EXPORT_SYMBOL(netdev_max_backlog);
|
|
|
|
|
net: Consistent skb timestamping
With RPS inclusion, skb timestamping is not consistent in RX path.
If netif_receive_skb() is used, its deferred after RPS dispatch.
If netif_rx() is used, its done before RPS dispatch.
This can give strange tcpdump timestamps results.
I think timestamping should be done as soon as possible in the receive
path, to get meaningful values (ie timestamps taken at the time packet
was delivered by NIC driver to our stack), even if NAPI already can
defer timestamping a bit (RPS can help to reduce the gap)
Tom Herbert prefer to sample timestamps after RPS dispatch. In case
sampling is expensive (HPET/acpi_pm on x86), this makes sense.
Let admins switch from one mode to another, using a new
sysctl, /proc/sys/net/core/netdev_tstamp_prequeue
Its default value (1), means timestamps are taken as soon as possible,
before backlog queueing, giving accurate timestamps.
Setting a 0 value permits to sample timestamps when processing backlog,
after RPS dispatch, to lower the load of the pre-RPS cpu.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-16 06:57:10 +00:00
|
|
|
int netdev_tstamp_prequeue __read_mostly = 1;
|
2007-03-12 21:33:50 +00:00
|
|
|
int netdev_budget __read_mostly = 300;
|
|
|
|
int weight_p __read_mostly = 64; /* old backlog weight */
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-05-07 05:07:48 +00:00
|
|
|
/* Called with irq disabled */
|
|
|
|
static inline void ____napi_schedule(struct softnet_data *sd,
|
|
|
|
struct napi_struct *napi)
|
|
|
|
{
|
|
|
|
list_add_tail(&napi->poll_list, &sd->poll_list);
|
|
|
|
__raise_softirq_irqoff(NET_RX_SOFTIRQ);
|
|
|
|
}
|
|
|
|
|
2010-08-04 06:15:52 +00:00
|
|
|
#ifdef CONFIG_RPS
|
|
|
|
|
|
|
|
/* One global table that all flow-based protocols share. */
|
2010-10-25 03:02:02 +00:00
|
|
|
struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
|
2010-08-04 06:15:52 +00:00
|
|
|
EXPORT_SYMBOL(rps_sock_flow_table);
|
|
|
|
|
2012-02-24 07:31:31 +00:00
|
|
|
struct static_key rps_needed __read_mostly;
|
2011-11-17 03:13:26 +00:00
|
|
|
|
2011-01-19 11:03:53 +00:00
|
|
|
static struct rps_dev_flow *
|
|
|
|
set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
|
|
|
|
struct rps_dev_flow *rflow, u16 next_cpu)
|
|
|
|
{
|
2011-10-03 04:42:46 +00:00
|
|
|
if (next_cpu != RPS_NO_CPU) {
|
2011-01-19 11:03:53 +00:00
|
|
|
#ifdef CONFIG_RFS_ACCEL
|
|
|
|
struct netdev_rx_queue *rxqueue;
|
|
|
|
struct rps_dev_flow_table *flow_table;
|
|
|
|
struct rps_dev_flow *old_rflow;
|
|
|
|
u32 flow_id;
|
|
|
|
u16 rxq_index;
|
|
|
|
int rc;
|
|
|
|
|
|
|
|
/* Should we steer this flow to a different hardware queue? */
|
2011-02-15 20:32:04 +00:00
|
|
|
if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
|
|
|
|
!(dev->features & NETIF_F_NTUPLE))
|
2011-01-19 11:03:53 +00:00
|
|
|
goto out;
|
|
|
|
rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
|
|
|
|
if (rxq_index == skb_get_rx_queue(skb))
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
rxqueue = dev->_rx + rxq_index;
|
|
|
|
flow_table = rcu_dereference(rxqueue->rps_flow_table);
|
|
|
|
if (!flow_table)
|
|
|
|
goto out;
|
|
|
|
flow_id = skb->rxhash & flow_table->mask;
|
|
|
|
rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
|
|
|
|
rxq_index, flow_id);
|
|
|
|
if (rc < 0)
|
|
|
|
goto out;
|
|
|
|
old_rflow = rflow;
|
|
|
|
rflow = &flow_table->flows[flow_id];
|
|
|
|
rflow->filter = rc;
|
|
|
|
if (old_rflow->filter == rflow->filter)
|
|
|
|
old_rflow->filter = RPS_NO_FILTER;
|
|
|
|
out:
|
|
|
|
#endif
|
|
|
|
rflow->last_qtail =
|
2011-10-03 04:42:46 +00:00
|
|
|
per_cpu(softnet_data, next_cpu).input_queue_head;
|
2011-01-19 11:03:53 +00:00
|
|
|
}
|
|
|
|
|
2011-10-03 04:42:46 +00:00
|
|
|
rflow->cpu = next_cpu;
|
2011-01-19 11:03:53 +00:00
|
|
|
return rflow;
|
|
|
|
}
|
|
|
|
|
2010-08-04 06:15:52 +00:00
|
|
|
/*
|
|
|
|
* get_rps_cpu is called from netif_receive_skb and returns the target
|
|
|
|
* CPU from the RPS map of the receiving queue for a given skb.
|
|
|
|
* rcu_read_lock must be held on entry.
|
|
|
|
*/
|
|
|
|
static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
|
|
|
|
struct rps_dev_flow **rflowp)
|
|
|
|
{
|
|
|
|
struct netdev_rx_queue *rxqueue;
|
2010-10-25 03:02:02 +00:00
|
|
|
struct rps_map *map;
|
2010-08-04 06:15:52 +00:00
|
|
|
struct rps_dev_flow_table *flow_table;
|
|
|
|
struct rps_sock_flow_table *sock_flow_table;
|
|
|
|
int cpu = -1;
|
|
|
|
u16 tcpu;
|
|
|
|
|
|
|
|
if (skb_rx_queue_recorded(skb)) {
|
|
|
|
u16 index = skb_get_rx_queue(skb);
|
2010-09-27 08:24:33 +00:00
|
|
|
if (unlikely(index >= dev->real_num_rx_queues)) {
|
|
|
|
WARN_ONCE(dev->real_num_rx_queues > 1,
|
|
|
|
"%s received packet on queue %u, but number "
|
|
|
|
"of RX queues is %u\n",
|
|
|
|
dev->name, index, dev->real_num_rx_queues);
|
2010-08-04 06:15:52 +00:00
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
rxqueue = dev->_rx + index;
|
|
|
|
} else
|
|
|
|
rxqueue = dev->_rx;
|
|
|
|
|
2010-10-25 03:02:02 +00:00
|
|
|
map = rcu_dereference(rxqueue->rps_map);
|
|
|
|
if (map) {
|
2011-02-01 00:23:42 +00:00
|
|
|
if (map->len == 1 &&
|
2011-08-11 19:30:52 +00:00
|
|
|
!rcu_access_pointer(rxqueue->rps_flow_table)) {
|
2010-09-03 23:12:37 +00:00
|
|
|
tcpu = map->cpus[0];
|
|
|
|
if (cpu_online(tcpu))
|
|
|
|
cpu = tcpu;
|
|
|
|
goto done;
|
|
|
|
}
|
2011-08-11 19:30:52 +00:00
|
|
|
} else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
|
2010-08-04 06:15:52 +00:00
|
|
|
goto done;
|
2010-09-03 23:12:37 +00:00
|
|
|
}
|
2010-08-04 06:15:52 +00:00
|
|
|
|
2010-08-17 19:00:56 +00:00
|
|
|
skb_reset_network_header(skb);
|
2010-08-04 06:15:52 +00:00
|
|
|
if (!skb_get_rxhash(skb))
|
|
|
|
goto done;
|
|
|
|
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
flow_table = rcu_dereference(rxqueue->rps_flow_table);
|
|
|
|
sock_flow_table = rcu_dereference(rps_sock_flow_table);
|
|
|
|
if (flow_table && sock_flow_table) {
|
|
|
|
u16 next_cpu;
|
|
|
|
struct rps_dev_flow *rflow;
|
|
|
|
|
|
|
|
rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
|
|
|
|
tcpu = rflow->cpu;
|
|
|
|
|
|
|
|
next_cpu = sock_flow_table->ents[skb->rxhash &
|
|
|
|
sock_flow_table->mask];
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the desired CPU (where last recvmsg was done) is
|
|
|
|
* different from current CPU (one in the rx-queue flow
|
|
|
|
* table entry), switch if one of the following holds:
|
|
|
|
* - Current CPU is unset (equal to RPS_NO_CPU).
|
|
|
|
* - Current CPU is offline.
|
|
|
|
* - The current CPU's queue tail has advanced beyond the
|
|
|
|
* last packet that was enqueued using this table entry.
|
|
|
|
* This guarantees that all previous packets for the flow
|
|
|
|
* have been dequeued, thus preserving in order delivery.
|
|
|
|
*/
|
|
|
|
if (unlikely(tcpu != next_cpu) &&
|
|
|
|
(tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
|
|
|
|
((int)(per_cpu(softnet_data, tcpu).input_queue_head -
|
2012-11-16 09:04:15 +00:00
|
|
|
rflow->last_qtail)) >= 0)) {
|
|
|
|
tcpu = next_cpu;
|
2011-01-19 11:03:53 +00:00
|
|
|
rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
|
2012-11-16 09:04:15 +00:00
|
|
|
}
|
2011-01-19 11:03:53 +00:00
|
|
|
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
|
|
|
|
*rflowp = rflow;
|
|
|
|
cpu = tcpu;
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-03-16 08:03:29 +00:00
|
|
|
if (map) {
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
|
2010-03-16 08:03:29 +00:00
|
|
|
|
|
|
|
if (cpu_online(tcpu)) {
|
|
|
|
cpu = tcpu;
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
done:
|
|
|
|
return cpu;
|
|
|
|
}
|
|
|
|
|
2011-01-19 11:03:53 +00:00
|
|
|
#ifdef CONFIG_RFS_ACCEL
|
|
|
|
|
|
|
|
/**
|
|
|
|
* rps_may_expire_flow - check whether an RFS hardware filter may be removed
|
|
|
|
* @dev: Device on which the filter was set
|
|
|
|
* @rxq_index: RX queue index
|
|
|
|
* @flow_id: Flow ID passed to ndo_rx_flow_steer()
|
|
|
|
* @filter_id: Filter ID returned by ndo_rx_flow_steer()
|
|
|
|
*
|
|
|
|
* Drivers that implement ndo_rx_flow_steer() should periodically call
|
|
|
|
* this function for each installed filter and remove the filters for
|
|
|
|
* which it returns %true.
|
|
|
|
*/
|
|
|
|
bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
|
|
|
|
u32 flow_id, u16 filter_id)
|
|
|
|
{
|
|
|
|
struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
|
|
|
|
struct rps_dev_flow_table *flow_table;
|
|
|
|
struct rps_dev_flow *rflow;
|
|
|
|
bool expire = true;
|
|
|
|
int cpu;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
flow_table = rcu_dereference(rxqueue->rps_flow_table);
|
|
|
|
if (flow_table && flow_id <= flow_table->mask) {
|
|
|
|
rflow = &flow_table->flows[flow_id];
|
|
|
|
cpu = ACCESS_ONCE(rflow->cpu);
|
|
|
|
if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
|
|
|
|
((int)(per_cpu(softnet_data, cpu).input_queue_head -
|
|
|
|
rflow->last_qtail) <
|
|
|
|
(int)(10 * flow_table->mask)))
|
|
|
|
expire = false;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
return expire;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(rps_may_expire_flow);
|
|
|
|
|
|
|
|
#endif /* CONFIG_RFS_ACCEL */
|
|
|
|
|
2010-03-16 08:03:29 +00:00
|
|
|
/* Called from hardirq (IPI) context */
|
2010-04-19 21:17:14 +00:00
|
|
|
static void rps_trigger_softirq(void *data)
|
2010-03-16 08:03:29 +00:00
|
|
|
{
|
2010-04-19 21:17:14 +00:00
|
|
|
struct softnet_data *sd = data;
|
|
|
|
|
2010-05-07 05:07:48 +00:00
|
|
|
____napi_schedule(sd, &sd->backlog);
|
2010-05-02 05:42:16 +00:00
|
|
|
sd->received_rps++;
|
2010-03-16 08:03:29 +00:00
|
|
|
}
|
2010-04-19 21:17:14 +00:00
|
|
|
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
#endif /* CONFIG_RPS */
|
2010-03-16 08:03:29 +00:00
|
|
|
|
2010-04-19 21:17:14 +00:00
|
|
|
/*
|
|
|
|
* Check if this softnet_data structure is another cpu one
|
|
|
|
* If yes, queue it to our IPI list and return 1
|
|
|
|
* If no, return 0
|
|
|
|
*/
|
|
|
|
static int rps_ipi_queued(struct softnet_data *sd)
|
|
|
|
{
|
|
|
|
#ifdef CONFIG_RPS
|
|
|
|
struct softnet_data *mysd = &__get_cpu_var(softnet_data);
|
|
|
|
|
|
|
|
if (sd != mysd) {
|
|
|
|
sd->rps_ipi_next = mysd->rps_ipi_list;
|
|
|
|
mysd->rps_ipi_list = sd;
|
|
|
|
|
|
|
|
__raise_softirq_irqoff(NET_RX_SOFTIRQ);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
#endif /* CONFIG_RPS */
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2013-05-20 04:02:32 +00:00
|
|
|
#ifdef CONFIG_NET_FLOW_LIMIT
|
|
|
|
int netdev_flow_limit_table_len __read_mostly = (1 << 12);
|
|
|
|
#endif
|
|
|
|
|
|
|
|
static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
|
|
|
|
{
|
|
|
|
#ifdef CONFIG_NET_FLOW_LIMIT
|
|
|
|
struct sd_flow_limit *fl;
|
|
|
|
struct softnet_data *sd;
|
|
|
|
unsigned int old_flow, new_flow;
|
|
|
|
|
|
|
|
if (qlen < (netdev_max_backlog >> 1))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
sd = &__get_cpu_var(softnet_data);
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
fl = rcu_dereference(sd->flow_limit);
|
|
|
|
if (fl) {
|
|
|
|
new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
|
|
|
|
old_flow = fl->history[fl->history_head];
|
|
|
|
fl->history[fl->history_head] = new_flow;
|
|
|
|
|
|
|
|
fl->history_head++;
|
|
|
|
fl->history_head &= FLOW_LIMIT_HISTORY - 1;
|
|
|
|
|
|
|
|
if (likely(fl->buckets[old_flow]))
|
|
|
|
fl->buckets[old_flow]--;
|
|
|
|
|
|
|
|
if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
|
|
|
|
fl->count++;
|
|
|
|
rcu_read_unlock();
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
#endif
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2010-03-16 08:03:29 +00:00
|
|
|
/*
|
|
|
|
* enqueue_to_backlog is called to queue an skb to a per CPU backlog
|
|
|
|
* queue (may be a remote CPU queue).
|
|
|
|
*/
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
|
|
|
|
unsigned int *qtail)
|
2010-03-16 08:03:29 +00:00
|
|
|
{
|
2010-04-19 21:17:14 +00:00
|
|
|
struct softnet_data *sd;
|
2010-03-16 08:03:29 +00:00
|
|
|
unsigned long flags;
|
2013-05-20 04:02:32 +00:00
|
|
|
unsigned int qlen;
|
2010-03-16 08:03:29 +00:00
|
|
|
|
2010-04-19 21:17:14 +00:00
|
|
|
sd = &per_cpu(softnet_data, cpu);
|
2010-03-16 08:03:29 +00:00
|
|
|
|
|
|
|
local_irq_save(flags);
|
|
|
|
|
2010-04-19 21:17:14 +00:00
|
|
|
rps_lock(sd);
|
2013-05-20 04:02:32 +00:00
|
|
|
qlen = skb_queue_len(&sd->input_pkt_queue);
|
|
|
|
if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
|
2010-04-27 22:07:33 +00:00
|
|
|
if (skb_queue_len(&sd->input_pkt_queue)) {
|
2010-03-16 08:03:29 +00:00
|
|
|
enqueue:
|
2010-04-19 21:17:14 +00:00
|
|
|
__skb_queue_tail(&sd->input_pkt_queue, skb);
|
2010-05-20 18:37:59 +00:00
|
|
|
input_queue_tail_incr_save(sd, qtail);
|
2010-04-19 21:17:14 +00:00
|
|
|
rps_unlock(sd);
|
2010-03-30 20:16:22 +00:00
|
|
|
local_irq_restore(flags);
|
2010-03-16 08:03:29 +00:00
|
|
|
return NET_RX_SUCCESS;
|
|
|
|
}
|
|
|
|
|
2010-05-06 23:51:21 +00:00
|
|
|
/* Schedule NAPI for backlog device
|
|
|
|
* We can use non atomic operation since we own the queue lock
|
|
|
|
*/
|
|
|
|
if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
|
2010-04-19 21:17:14 +00:00
|
|
|
if (!rps_ipi_queued(sd))
|
2010-05-07 05:07:48 +00:00
|
|
|
____napi_schedule(sd, &sd->backlog);
|
2010-03-16 08:03:29 +00:00
|
|
|
}
|
|
|
|
goto enqueue;
|
|
|
|
}
|
|
|
|
|
2010-05-02 05:42:16 +00:00
|
|
|
sd->dropped++;
|
2010-04-19 21:17:14 +00:00
|
|
|
rps_unlock(sd);
|
2010-03-16 08:03:29 +00:00
|
|
|
|
|
|
|
local_irq_restore(flags);
|
|
|
|
|
2010-09-30 21:06:55 +00:00
|
|
|
atomic_long_inc(&skb->dev->rx_dropped);
|
2010-03-16 08:03:29 +00:00
|
|
|
kfree_skb(skb);
|
|
|
|
return NET_RX_DROP;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* netif_rx - post buffer to the network code
|
|
|
|
* @skb: buffer to post
|
|
|
|
*
|
|
|
|
* This function receives a packet from a device driver and queues it for
|
|
|
|
* the upper (protocol) levels to process. It always succeeds. The buffer
|
|
|
|
* may be dropped during processing for congestion control or by the
|
|
|
|
* protocol layers.
|
|
|
|
*
|
|
|
|
* return values:
|
|
|
|
* NET_RX_SUCCESS (no congestion)
|
|
|
|
* NET_RX_DROP (packet was dropped)
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
int netif_rx(struct sk_buff *skb)
|
|
|
|
{
|
2010-04-15 07:14:07 +00:00
|
|
|
int ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* if netpoll wants it, pretend we never saw it */
|
|
|
|
if (netpoll_rx(skb))
|
|
|
|
return NET_RX_DROP;
|
|
|
|
|
2011-11-15 04:12:55 +00:00
|
|
|
net_timestamp_check(netdev_tstamp_prequeue, skb);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-08-23 09:45:02 +00:00
|
|
|
trace_netif_rx(skb);
|
2010-03-24 19:13:54 +00:00
|
|
|
#ifdef CONFIG_RPS
|
2012-02-24 07:31:31 +00:00
|
|
|
if (static_key_false(&rps_needed)) {
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
struct rps_dev_flow voidflow, *rflow = &voidflow;
|
2010-04-15 07:14:07 +00:00
|
|
|
int cpu;
|
|
|
|
|
2010-08-08 03:35:43 +00:00
|
|
|
preempt_disable();
|
2010-04-15 07:14:07 +00:00
|
|
|
rcu_read_lock();
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
|
|
|
|
cpu = get_rps_cpu(skb->dev, skb, &rflow);
|
2010-04-15 07:14:07 +00:00
|
|
|
if (cpu < 0)
|
|
|
|
cpu = smp_processor_id();
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
|
|
|
|
ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
|
|
|
|
|
2010-04-15 07:14:07 +00:00
|
|
|
rcu_read_unlock();
|
2010-08-08 03:35:43 +00:00
|
|
|
preempt_enable();
|
2011-11-17 03:13:26 +00:00
|
|
|
} else
|
|
|
|
#endif
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
{
|
|
|
|
unsigned int qtail;
|
|
|
|
ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
|
|
|
|
put_cpu();
|
|
|
|
}
|
2010-04-15 07:14:07 +00:00
|
|
|
return ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(netif_rx);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
int netif_rx_ni(struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
|
|
|
|
preempt_disable();
|
|
|
|
err = netif_rx(skb);
|
|
|
|
if (local_softirq_pending())
|
|
|
|
do_softirq();
|
|
|
|
preempt_enable();
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_rx_ni);
|
|
|
|
|
|
|
|
static void net_tx_action(struct softirq_action *h)
|
|
|
|
{
|
|
|
|
struct softnet_data *sd = &__get_cpu_var(softnet_data);
|
|
|
|
|
|
|
|
if (sd->completion_queue) {
|
|
|
|
struct sk_buff *clist;
|
|
|
|
|
|
|
|
local_irq_disable();
|
|
|
|
clist = sd->completion_queue;
|
|
|
|
sd->completion_queue = NULL;
|
|
|
|
local_irq_enable();
|
|
|
|
|
|
|
|
while (clist) {
|
|
|
|
struct sk_buff *skb = clist;
|
|
|
|
clist = clist->next;
|
|
|
|
|
2008-07-26 04:43:18 +00:00
|
|
|
WARN_ON(atomic_read(&skb->users));
|
2010-08-23 09:46:12 +00:00
|
|
|
trace_kfree_skb(skb, net_tx_action);
|
2005-04-16 22:20:36 +00:00
|
|
|
__kfree_skb(skb);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sd->output_queue) {
|
2008-07-16 09:15:04 +00:00
|
|
|
struct Qdisc *head;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
local_irq_disable();
|
|
|
|
head = sd->output_queue;
|
|
|
|
sd->output_queue = NULL;
|
2010-04-26 23:06:24 +00:00
|
|
|
sd->output_queue_tailp = &sd->output_queue;
|
2005-04-16 22:20:36 +00:00
|
|
|
local_irq_enable();
|
|
|
|
|
|
|
|
while (head) {
|
2008-07-16 09:15:04 +00:00
|
|
|
struct Qdisc *q = head;
|
|
|
|
spinlock_t *root_lock;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
head = head->next_sched;
|
|
|
|
|
2008-08-03 03:02:43 +00:00
|
|
|
root_lock = qdisc_lock(q);
|
2008-07-16 09:15:04 +00:00
|
|
|
if (spin_trylock(root_lock)) {
|
2008-08-18 04:54:43 +00:00
|
|
|
smp_mb__before_clear_bit();
|
|
|
|
clear_bit(__QDISC_STATE_SCHED,
|
|
|
|
&q->state);
|
2008-07-16 09:15:04 +00:00
|
|
|
qdisc_run(q);
|
|
|
|
spin_unlock(root_lock);
|
2005-04-16 22:20:36 +00:00
|
|
|
} else {
|
2008-08-19 11:00:36 +00:00
|
|
|
if (!test_bit(__QDISC_STATE_DEACTIVATED,
|
2008-09-08 01:41:21 +00:00
|
|
|
&q->state)) {
|
2008-08-19 11:00:36 +00:00
|
|
|
__netif_reschedule(q);
|
2008-09-08 01:41:21 +00:00
|
|
|
} else {
|
|
|
|
smp_mb__before_clear_bit();
|
|
|
|
clear_bit(__QDISC_STATE_SCHED,
|
|
|
|
&q->state);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-06-01 21:52:08 +00:00
|
|
|
#if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
|
|
|
|
(defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
|
2009-06-05 05:35:28 +00:00
|
|
|
/* This hook is defined here for ATM LANE */
|
|
|
|
int (*br_fdb_test_addr_hook)(struct net_device *dev,
|
|
|
|
unsigned char *addr) __read_mostly;
|
2009-09-11 18:50:08 +00:00
|
|
|
EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
|
2009-06-05 05:35:28 +00:00
|
|
|
#endif
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
#ifdef CONFIG_NET_CLS_ACT
|
|
|
|
/* TODO: Maybe we should just force sch_ingress to be compiled in
|
|
|
|
* when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
|
|
|
|
* a compare and 2 stores extra right now if we dont have it on
|
|
|
|
* but have CONFIG_NET_CLS_ACT
|
2011-03-31 01:57:33 +00:00
|
|
|
* NOTE: This doesn't stop any functionality; if you dont have
|
|
|
|
* the ingress scheduler, you just can't add policies on ingress.
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
*/
|
2010-10-02 06:11:55 +00:00
|
|
|
static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev = skb->dev;
|
2007-10-14 07:38:47 +00:00
|
|
|
u32 ttl = G_TC_RTTL(skb->tc_verd);
|
2008-07-09 00:33:13 +00:00
|
|
|
int result = TC_ACT_OK;
|
|
|
|
struct Qdisc *q;
|
2007-02-09 14:24:36 +00:00
|
|
|
|
2010-08-01 07:33:23 +00:00
|
|
|
if (unlikely(MAX_RED_LOOP < ttl++)) {
|
2012-05-13 21:56:26 +00:00
|
|
|
net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
|
|
|
|
skb->skb_iif, dev->ifindex);
|
2007-10-14 07:38:47 +00:00
|
|
|
return TC_ACT_SHOT;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-10-14 07:38:47 +00:00
|
|
|
skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
|
|
|
|
skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-07-17 07:53:03 +00:00
|
|
|
q = rxq->qdisc;
|
2008-07-30 09:37:46 +00:00
|
|
|
if (q != &noop_qdisc) {
|
2008-07-17 07:53:03 +00:00
|
|
|
spin_lock(qdisc_lock(q));
|
2008-08-18 04:51:03 +00:00
|
|
|
if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
|
|
|
|
result = qdisc_enqueue_root(skb, q);
|
2008-07-17 07:53:03 +00:00
|
|
|
spin_unlock(qdisc_lock(q));
|
|
|
|
}
|
2007-10-14 07:38:47 +00:00
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
2005-08-10 02:36:29 +00:00
|
|
|
|
2007-10-14 07:38:47 +00:00
|
|
|
static inline struct sk_buff *handle_ing(struct sk_buff *skb,
|
|
|
|
struct packet_type **pt_prev,
|
|
|
|
int *ret, struct net_device *orig_dev)
|
|
|
|
{
|
2010-10-02 06:11:55 +00:00
|
|
|
struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
|
|
|
|
|
|
|
|
if (!rxq || rxq->qdisc == &noop_qdisc)
|
2007-10-14 07:38:47 +00:00
|
|
|
goto out;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-10-14 07:38:47 +00:00
|
|
|
if (*pt_prev) {
|
|
|
|
*ret = deliver_skb(skb, *pt_prev, orig_dev);
|
|
|
|
*pt_prev = NULL;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2010-10-02 06:11:55 +00:00
|
|
|
switch (ing_filter(skb, rxq)) {
|
2007-10-14 07:38:47 +00:00
|
|
|
case TC_ACT_SHOT:
|
|
|
|
case TC_ACT_STOLEN:
|
|
|
|
kfree_skb(skb);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
skb->tc_verd = 0;
|
|
|
|
return skb;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2010-06-01 21:52:08 +00:00
|
|
|
/**
|
|
|
|
* netdev_rx_handler_register - register receive handler
|
|
|
|
* @dev: device to register a handler for
|
|
|
|
* @rx_handler: receive handler to register
|
2010-06-10 03:34:59 +00:00
|
|
|
* @rx_handler_data: data pointer that is used by rx handler
|
2010-06-01 21:52:08 +00:00
|
|
|
*
|
|
|
|
* Register a receive hander for a device. This handler will then be
|
|
|
|
* called from __netif_receive_skb. A negative errno code is returned
|
|
|
|
* on a failure.
|
|
|
|
*
|
|
|
|
* The caller must hold the rtnl_mutex.
|
2011-03-12 03:14:39 +00:00
|
|
|
*
|
|
|
|
* For a general description of rx_handler, see enum rx_handler_result.
|
2010-06-01 21:52:08 +00:00
|
|
|
*/
|
|
|
|
int netdev_rx_handler_register(struct net_device *dev,
|
2010-06-10 03:34:59 +00:00
|
|
|
rx_handler_func_t *rx_handler,
|
|
|
|
void *rx_handler_data)
|
2010-06-01 21:52:08 +00:00
|
|
|
{
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
if (dev->rx_handler)
|
|
|
|
return -EBUSY;
|
|
|
|
|
net: add a synchronize_net() in netdev_rx_handler_unregister()
commit 35d48903e97819 (bonding: fix rx_handler locking) added a race
in bonding driver, reported by Steven Rostedt who did a very good
diagnosis :
<quoting Steven>
I'm currently debugging a crash in an old 3.0-rt kernel that one of our
customers is seeing. The bug happens with a stress test that loads and
unloads the bonding module in a loop (I don't know all the details as
I'm not the one that is directly interacting with the customer). But the
bug looks to be something that may still be present and possibly present
in mainline too. It will just be much harder to trigger it in mainline.
In -rt, interrupts are threads, and can schedule in and out just like
any other thread. Note, mainline now supports interrupt threads so this
may be easily reproducible in mainline as well. I don't have the ability
to tell the customer to try mainline or other kernels, so my hands are
somewhat tied to what I can do.
But according to a core dump, I tracked down that the eth irq thread
crashed in bond_handle_frame() here:
slave = bond_slave_get_rcu(skb->dev);
bond = slave->bond; <--- BUG
the slave returned was NULL and accessing slave->bond caused a NULL
pointer dereference.
Looking at the code that unregisters the handler:
void netdev_rx_handler_unregister(struct net_device *dev)
{
ASSERT_RTNL();
RCU_INIT_POINTER(dev->rx_handler, NULL);
RCU_INIT_POINTER(dev->rx_handler_data, NULL);
}
Which is basically:
dev->rx_handler = NULL;
dev->rx_handler_data = NULL;
And looking at __netif_receive_skb() we have:
rx_handler = rcu_dereference(skb->dev->rx_handler);
if (rx_handler) {
if (pt_prev) {
ret = deliver_skb(skb, pt_prev, orig_dev);
pt_prev = NULL;
}
switch (rx_handler(&skb)) {
My question to all of you is, what stops this interrupt from happening
while the bonding module is unloading? What happens if the interrupt
triggers and we have this:
CPU0 CPU1
---- ----
rx_handler = skb->dev->rx_handler
netdev_rx_handler_unregister() {
dev->rx_handler = NULL;
dev->rx_handler_data = NULL;
rx_handler()
bond_handle_frame() {
slave = skb->dev->rx_handler;
bond = slave->bond; <-- NULL pointer dereference!!!
What protection am I missing in the bond release handler that would
prevent the above from happening?
</quoting Steven>
We can fix bug this in two ways. First is adding a test in
bond_handle_frame() and others to check if rx_handler_data is NULL.
A second way is adding a synchronize_net() in
netdev_rx_handler_unregister() to make sure that a rcu protected reader
has the guarantee to see a non NULL rx_handler_data.
The second way is better as it avoids an extra test in fast path.
Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jiri Pirko <jpirko@redhat.com>
Cc: Paul E. McKenney <paulmck@us.ibm.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-29 03:01:22 +00:00
|
|
|
/* Note: rx_handler_data must be set before rx_handler */
|
2010-06-10 03:34:59 +00:00
|
|
|
rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
|
2010-06-01 21:52:08 +00:00
|
|
|
rcu_assign_pointer(dev->rx_handler, rx_handler);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_rx_handler_unregister - unregister receive handler
|
|
|
|
* @dev: device to unregister a handler from
|
|
|
|
*
|
2013-03-18 02:59:52 +00:00
|
|
|
* Unregister a receive handler from a device.
|
2010-06-01 21:52:08 +00:00
|
|
|
*
|
|
|
|
* The caller must hold the rtnl_mutex.
|
|
|
|
*/
|
|
|
|
void netdev_rx_handler_unregister(struct net_device *dev)
|
|
|
|
{
|
|
|
|
|
|
|
|
ASSERT_RTNL();
|
2011-08-01 16:19:00 +00:00
|
|
|
RCU_INIT_POINTER(dev->rx_handler, NULL);
|
net: add a synchronize_net() in netdev_rx_handler_unregister()
commit 35d48903e97819 (bonding: fix rx_handler locking) added a race
in bonding driver, reported by Steven Rostedt who did a very good
diagnosis :
<quoting Steven>
I'm currently debugging a crash in an old 3.0-rt kernel that one of our
customers is seeing. The bug happens with a stress test that loads and
unloads the bonding module in a loop (I don't know all the details as
I'm not the one that is directly interacting with the customer). But the
bug looks to be something that may still be present and possibly present
in mainline too. It will just be much harder to trigger it in mainline.
In -rt, interrupts are threads, and can schedule in and out just like
any other thread. Note, mainline now supports interrupt threads so this
may be easily reproducible in mainline as well. I don't have the ability
to tell the customer to try mainline or other kernels, so my hands are
somewhat tied to what I can do.
But according to a core dump, I tracked down that the eth irq thread
crashed in bond_handle_frame() here:
slave = bond_slave_get_rcu(skb->dev);
bond = slave->bond; <--- BUG
the slave returned was NULL and accessing slave->bond caused a NULL
pointer dereference.
Looking at the code that unregisters the handler:
void netdev_rx_handler_unregister(struct net_device *dev)
{
ASSERT_RTNL();
RCU_INIT_POINTER(dev->rx_handler, NULL);
RCU_INIT_POINTER(dev->rx_handler_data, NULL);
}
Which is basically:
dev->rx_handler = NULL;
dev->rx_handler_data = NULL;
And looking at __netif_receive_skb() we have:
rx_handler = rcu_dereference(skb->dev->rx_handler);
if (rx_handler) {
if (pt_prev) {
ret = deliver_skb(skb, pt_prev, orig_dev);
pt_prev = NULL;
}
switch (rx_handler(&skb)) {
My question to all of you is, what stops this interrupt from happening
while the bonding module is unloading? What happens if the interrupt
triggers and we have this:
CPU0 CPU1
---- ----
rx_handler = skb->dev->rx_handler
netdev_rx_handler_unregister() {
dev->rx_handler = NULL;
dev->rx_handler_data = NULL;
rx_handler()
bond_handle_frame() {
slave = skb->dev->rx_handler;
bond = slave->bond; <-- NULL pointer dereference!!!
What protection am I missing in the bond release handler that would
prevent the above from happening?
</quoting Steven>
We can fix bug this in two ways. First is adding a test in
bond_handle_frame() and others to check if rx_handler_data is NULL.
A second way is adding a synchronize_net() in
netdev_rx_handler_unregister() to make sure that a rcu protected reader
has the guarantee to see a non NULL rx_handler_data.
The second way is better as it avoids an extra test in fast path.
Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jiri Pirko <jpirko@redhat.com>
Cc: Paul E. McKenney <paulmck@us.ibm.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-29 03:01:22 +00:00
|
|
|
/* a reader seeing a non NULL rx_handler in a rcu_read_lock()
|
|
|
|
* section has a guarantee to see a non NULL rx_handler_data
|
|
|
|
* as well.
|
|
|
|
*/
|
|
|
|
synchronize_net();
|
2011-08-01 16:19:00 +00:00
|
|
|
RCU_INIT_POINTER(dev->rx_handler_data, NULL);
|
2010-06-01 21:52:08 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
|
|
|
|
|
2012-07-31 23:44:26 +00:00
|
|
|
/*
|
|
|
|
* Limit the use of PFMEMALLOC reserves to those protocols that implement
|
|
|
|
* the special handling of PFMEMALLOC skbs.
|
|
|
|
*/
|
|
|
|
static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
switch (skb->protocol) {
|
|
|
|
case __constant_htons(ETH_P_ARP):
|
|
|
|
case __constant_htons(ETH_P_IP):
|
|
|
|
case __constant_htons(ETH_P_IPV6):
|
|
|
|
case __constant_htons(ETH_P_8021Q):
|
net: vlan: add 802.1ad support
Add support for 802.1ad VLAN devices. This mainly consists of checking for
ETH_P_8021AD in addition to ETH_P_8021Q in a couple of places and check
offloading capabilities based on the used protocol.
Configuration is done using "ip link":
# ip link add link eth0 eth0.1000 \
type vlan proto 802.1ad id 1000
# ip link add link eth0.1000 eth0.1000.1000 \
type vlan proto 802.1q id 1000
52:54:00:12:34:56 > 92:b1:54:28:e4:8c, ethertype 802.1Q (0x8100), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto ICMP (1), length 84)
20.1.0.2 > 20.1.0.1: ICMP echo request, id 3003, seq 8, length 64
92:b1:54:28:e4:8c > 52:54:00:12:34:56, ethertype 802.1Q-QinQ (0x88a8), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 47944, offset 0, flags [none], proto ICMP (1), length 84)
20.1.0.1 > 20.1.0.2: ICMP echo reply, id 3003, seq 8, length 64
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-19 02:04:31 +00:00
|
|
|
case __constant_htons(ETH_P_8021AD):
|
2012-07-31 23:44:26 +00:00
|
|
|
return true;
|
|
|
|
default:
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-02-14 20:57:38 +00:00
|
|
|
static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct packet_type *ptype, *pt_prev;
|
2010-06-01 21:52:08 +00:00
|
|
|
rx_handler_func_t *rx_handler;
|
2005-08-10 02:34:12 +00:00
|
|
|
struct net_device *orig_dev;
|
2011-02-28 18:48:59 +00:00
|
|
|
struct net_device *null_or_dev;
|
2011-03-12 03:14:39 +00:00
|
|
|
bool deliver_exact = false;
|
2005-04-16 22:20:36 +00:00
|
|
|
int ret = NET_RX_DROP;
|
2006-11-15 04:48:11 +00:00
|
|
|
__be16 type;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2011-11-15 04:12:55 +00:00
|
|
|
net_timestamp_check(!netdev_tstamp_prequeue, skb);
|
2009-09-30 23:42:42 +00:00
|
|
|
|
2010-08-23 09:45:02 +00:00
|
|
|
trace_netif_receive_skb(skb);
|
2008-11-04 22:49:57 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* if we've gotten here through NAPI, check netpoll */
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
if (netpoll_receive_skb(skb))
|
2012-07-31 23:44:26 +00:00
|
|
|
goto out;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-07-03 01:22:00 +00:00
|
|
|
orig_dev = skb->dev;
|
2006-02-22 00:36:44 +00:00
|
|
|
|
2007-04-11 03:45:18 +00:00
|
|
|
skb_reset_network_header(skb);
|
2013-01-07 09:28:21 +00:00
|
|
|
if (!skb_transport_header_was_set(skb))
|
|
|
|
skb_reset_transport_header(skb);
|
2011-06-10 06:56:58 +00:00
|
|
|
skb_reset_mac_len(skb);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
pt_prev = NULL;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
|
2011-02-28 18:48:59 +00:00
|
|
|
another_round:
|
2012-07-23 23:27:54 +00:00
|
|
|
skb->skb_iif = skb->dev->ifindex;
|
2011-02-28 18:48:59 +00:00
|
|
|
|
|
|
|
__this_cpu_inc(softnet_data.processed);
|
|
|
|
|
net: vlan: add 802.1ad support
Add support for 802.1ad VLAN devices. This mainly consists of checking for
ETH_P_8021AD in addition to ETH_P_8021Q in a couple of places and check
offloading capabilities based on the used protocol.
Configuration is done using "ip link":
# ip link add link eth0 eth0.1000 \
type vlan proto 802.1ad id 1000
# ip link add link eth0.1000 eth0.1000.1000 \
type vlan proto 802.1q id 1000
52:54:00:12:34:56 > 92:b1:54:28:e4:8c, ethertype 802.1Q (0x8100), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto ICMP (1), length 84)
20.1.0.2 > 20.1.0.1: ICMP echo request, id 3003, seq 8, length 64
92:b1:54:28:e4:8c > 52:54:00:12:34:56, ethertype 802.1Q-QinQ (0x88a8), length 106: vlan 1000, p 0, ethertype 802.1Q, vlan 1000, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 47944, offset 0, flags [none], proto ICMP (1), length 84)
20.1.0.1 > 20.1.0.2: ICMP echo reply, id 3003, seq 8, length 64
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-19 02:04:31 +00:00
|
|
|
if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
|
|
|
|
skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
|
2011-04-07 19:48:33 +00:00
|
|
|
skb = vlan_untag(skb);
|
|
|
|
if (unlikely(!skb))
|
2012-07-31 23:44:26 +00:00
|
|
|
goto unlock;
|
2011-04-07 19:48:33 +00:00
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
#ifdef CONFIG_NET_CLS_ACT
|
|
|
|
if (skb->tc_verd & TC_NCLS) {
|
|
|
|
skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
|
|
|
|
goto ncls;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2013-02-14 20:57:38 +00:00
|
|
|
if (pfmemalloc)
|
2012-07-31 23:44:26 +00:00
|
|
|
goto skip_taps;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
list_for_each_entry_rcu(ptype, &ptype_all, list) {
|
2011-02-28 18:48:59 +00:00
|
|
|
if (!ptype->dev || ptype->dev == skb->dev) {
|
2007-02-09 14:24:36 +00:00
|
|
|
if (pt_prev)
|
2005-08-10 02:34:12 +00:00
|
|
|
ret = deliver_skb(skb, pt_prev, orig_dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
pt_prev = ptype;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-07-31 23:44:26 +00:00
|
|
|
skip_taps:
|
2005-04-16 22:20:36 +00:00
|
|
|
#ifdef CONFIG_NET_CLS_ACT
|
2007-10-14 07:38:47 +00:00
|
|
|
skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
|
|
|
|
if (!skb)
|
2012-07-31 23:44:26 +00:00
|
|
|
goto unlock;
|
2005-04-16 22:20:36 +00:00
|
|
|
ncls:
|
|
|
|
#endif
|
|
|
|
|
2013-02-14 20:57:38 +00:00
|
|
|
if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
|
2012-07-31 23:44:26 +00:00
|
|
|
goto drop;
|
|
|
|
|
2011-10-10 09:16:41 +00:00
|
|
|
if (vlan_tx_tag_present(skb)) {
|
|
|
|
if (pt_prev) {
|
|
|
|
ret = deliver_skb(skb, pt_prev, orig_dev);
|
|
|
|
pt_prev = NULL;
|
|
|
|
}
|
vlan: don't deliver frames for unknown vlans to protocols
6a32e4f9dd9219261f8856f817e6655114cfec2f made the vlan code skip marking
vlan-tagged frames for not locally configured vlans as PACKET_OTHERHOST if
there was an rx_handler, as the rx_handler could cause the frame to be received
on a different (virtual) vlan-capable interface where that vlan might be
configured.
As rx_handlers do not necessarily return RX_HANDLER_ANOTHER, this could cause
frames for unknown vlans to be delivered to the protocol stack as if they had
been received untagged.
For example, if an ipv6 router advertisement that's tagged for a locally not
configured vlan is received on an interface with macvlan interfaces attached,
macvlan's rx_handler returns RX_HANDLER_PASS after delivering the frame to the
macvlan interfaces, which caused it to be passed to the protocol stack, leading
to ipv6 addresses for the announced prefix being configured even though those
are completely unusable on the underlying interface.
The fix moves marking as PACKET_OTHERHOST after the rx_handler so the
rx_handler, if there is one, sees the frame unchanged, but afterwards,
before the frame is delivered to the protocol stack, it gets marked whether
there is an rx_handler or not.
Signed-off-by: Florian Zumbiehl <florz@florz.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-10-07 15:51:58 +00:00
|
|
|
if (vlan_do_receive(&skb))
|
2011-10-10 09:16:41 +00:00
|
|
|
goto another_round;
|
|
|
|
else if (unlikely(!skb))
|
2012-07-31 23:44:26 +00:00
|
|
|
goto unlock;
|
2011-10-10 09:16:41 +00:00
|
|
|
}
|
|
|
|
|
vlan: don't deliver frames for unknown vlans to protocols
6a32e4f9dd9219261f8856f817e6655114cfec2f made the vlan code skip marking
vlan-tagged frames for not locally configured vlans as PACKET_OTHERHOST if
there was an rx_handler, as the rx_handler could cause the frame to be received
on a different (virtual) vlan-capable interface where that vlan might be
configured.
As rx_handlers do not necessarily return RX_HANDLER_ANOTHER, this could cause
frames for unknown vlans to be delivered to the protocol stack as if they had
been received untagged.
For example, if an ipv6 router advertisement that's tagged for a locally not
configured vlan is received on an interface with macvlan interfaces attached,
macvlan's rx_handler returns RX_HANDLER_PASS after delivering the frame to the
macvlan interfaces, which caused it to be passed to the protocol stack, leading
to ipv6 addresses for the announced prefix being configured even though those
are completely unusable on the underlying interface.
The fix moves marking as PACKET_OTHERHOST after the rx_handler so the
rx_handler, if there is one, sees the frame unchanged, but afterwards,
before the frame is delivered to the protocol stack, it gets marked whether
there is an rx_handler or not.
Signed-off-by: Florian Zumbiehl <florz@florz.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-10-07 15:51:58 +00:00
|
|
|
rx_handler = rcu_dereference(skb->dev->rx_handler);
|
2010-06-01 21:52:08 +00:00
|
|
|
if (rx_handler) {
|
|
|
|
if (pt_prev) {
|
|
|
|
ret = deliver_skb(skb, pt_prev, orig_dev);
|
|
|
|
pt_prev = NULL;
|
|
|
|
}
|
2011-03-12 03:14:39 +00:00
|
|
|
switch (rx_handler(&skb)) {
|
|
|
|
case RX_HANDLER_CONSUMED:
|
2013-03-08 07:03:38 +00:00
|
|
|
ret = NET_RX_SUCCESS;
|
2012-07-31 23:44:26 +00:00
|
|
|
goto unlock;
|
2011-03-12 03:14:39 +00:00
|
|
|
case RX_HANDLER_ANOTHER:
|
2011-02-28 18:48:59 +00:00
|
|
|
goto another_round;
|
2011-03-12 03:14:39 +00:00
|
|
|
case RX_HANDLER_EXACT:
|
|
|
|
deliver_exact = true;
|
|
|
|
case RX_HANDLER_PASS:
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
BUG();
|
|
|
|
}
|
2010-06-01 21:52:08 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
vlan: don't deliver frames for unknown vlans to protocols
6a32e4f9dd9219261f8856f817e6655114cfec2f made the vlan code skip marking
vlan-tagged frames for not locally configured vlans as PACKET_OTHERHOST if
there was an rx_handler, as the rx_handler could cause the frame to be received
on a different (virtual) vlan-capable interface where that vlan might be
configured.
As rx_handlers do not necessarily return RX_HANDLER_ANOTHER, this could cause
frames for unknown vlans to be delivered to the protocol stack as if they had
been received untagged.
For example, if an ipv6 router advertisement that's tagged for a locally not
configured vlan is received on an interface with macvlan interfaces attached,
macvlan's rx_handler returns RX_HANDLER_PASS after delivering the frame to the
macvlan interfaces, which caused it to be passed to the protocol stack, leading
to ipv6 addresses for the announced prefix being configured even though those
are completely unusable on the underlying interface.
The fix moves marking as PACKET_OTHERHOST after the rx_handler so the
rx_handler, if there is one, sees the frame unchanged, but afterwards,
before the frame is delivered to the protocol stack, it gets marked whether
there is an rx_handler or not.
Signed-off-by: Florian Zumbiehl <florz@florz.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-10-07 15:51:58 +00:00
|
|
|
if (vlan_tx_nonzero_tag_present(skb))
|
|
|
|
skb->pkt_type = PACKET_OTHERHOST;
|
|
|
|
|
2011-02-28 18:48:59 +00:00
|
|
|
/* deliver only exact match when indicated */
|
2011-03-12 03:14:39 +00:00
|
|
|
null_or_dev = deliver_exact ? skb->dev : NULL;
|
bonding: allow arp_ip_targets on separate vlans to use arp validation
This allows a bond device to specify an arp_ip_target as a host that is
not on the same vlan as the base bond device and still use arp
validation. A configuration like this, now works:
BONDING_OPTS="mode=active-backup arp_interval=1000 arp_ip_target=10.0.100.1 arp_validate=3"
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
2: eth1: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 qlen 1000
link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff
3: eth0: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 qlen 1000
link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff
8: bond0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP> mtu 1500 qdisc noqueue
link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff
inet6 fe80::213:21ff:febe:33e9/64 scope link
valid_lft forever preferred_lft forever
9: bond0.100@bond0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP> mtu 1500 qdisc noqueue
link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff
inet 10.0.100.2/24 brd 10.0.100.255 scope global bond0.100
inet6 fe80::213:21ff:febe:33e9/64 scope link
valid_lft forever preferred_lft forever
Ethernet Channel Bonding Driver: v3.6.0 (September 26, 2009)
Bonding Mode: fault-tolerance (active-backup)
Primary Slave: None
Currently Active Slave: eth1
MII Status: up
MII Polling Interval (ms): 0
Up Delay (ms): 0
Down Delay (ms): 0
ARP Polling Interval (ms): 1000
ARP IP target/s (n.n.n.n form): 10.0.100.1
Slave Interface: eth1
MII Status: up
Link Failure Count: 1
Permanent HW addr: 00:40:05:30:ff:30
Slave Interface: eth0
MII Status: up
Link Failure Count: 0
Permanent HW addr: 00:13:21:be:33:e9
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-12-14 10:48:58 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
type = skb->protocol;
|
2007-11-26 12:12:58 +00:00
|
|
|
list_for_each_entry_rcu(ptype,
|
|
|
|
&ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
|
2011-02-28 18:48:59 +00:00
|
|
|
if (ptype->type == type &&
|
2011-02-28 20:26:31 +00:00
|
|
|
(ptype->dev == null_or_dev || ptype->dev == skb->dev ||
|
|
|
|
ptype->dev == orig_dev)) {
|
2007-02-09 14:24:36 +00:00
|
|
|
if (pt_prev)
|
2005-08-10 02:34:12 +00:00
|
|
|
ret = deliver_skb(skb, pt_prev, orig_dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
pt_prev = ptype;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (pt_prev) {
|
2012-07-20 09:23:17 +00:00
|
|
|
if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
|
2012-09-15 22:44:16 +00:00
|
|
|
goto drop;
|
2012-07-20 09:23:17 +00:00
|
|
|
else
|
|
|
|
ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
} else {
|
2012-07-31 23:44:26 +00:00
|
|
|
drop:
|
2010-09-30 21:06:55 +00:00
|
|
|
atomic_long_inc(&skb->dev->rx_dropped);
|
2005-04-16 22:20:36 +00:00
|
|
|
kfree_skb(skb);
|
|
|
|
/* Jamal, now you will not able to escape explaining
|
|
|
|
* me how you were going to use this. :-)
|
|
|
|
*/
|
|
|
|
ret = NET_RX_DROP;
|
|
|
|
}
|
|
|
|
|
2012-07-31 23:44:26 +00:00
|
|
|
unlock:
|
2005-04-16 22:20:36 +00:00
|
|
|
rcu_read_unlock();
|
2012-07-31 23:44:26 +00:00
|
|
|
out:
|
2013-02-14 20:57:38 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int __netif_receive_skb(struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
|
|
|
|
unsigned long pflags = current->flags;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* PFMEMALLOC skbs are special, they should
|
|
|
|
* - be delivered to SOCK_MEMALLOC sockets only
|
|
|
|
* - stay away from userspace
|
|
|
|
* - have bounded memory usage
|
|
|
|
*
|
|
|
|
* Use PF_MEMALLOC as this saves us from propagating the allocation
|
|
|
|
* context down to all allocation sites.
|
|
|
|
*/
|
|
|
|
current->flags |= PF_MEMALLOC;
|
|
|
|
ret = __netif_receive_skb_core(skb, true);
|
|
|
|
tsk_restore_flags(current, pflags, PF_MEMALLOC);
|
|
|
|
} else
|
|
|
|
ret = __netif_receive_skb_core(skb, false);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return ret;
|
|
|
|
}
|
2010-03-16 08:03:29 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* netif_receive_skb - process receive buffer from network
|
|
|
|
* @skb: buffer to process
|
|
|
|
*
|
|
|
|
* netif_receive_skb() is the main receive data processing function.
|
|
|
|
* It always succeeds. The buffer may be dropped during processing
|
|
|
|
* for congestion control or by the protocol layers.
|
|
|
|
*
|
|
|
|
* This function may only be called from softirq context and interrupts
|
|
|
|
* should be enabled.
|
|
|
|
*
|
|
|
|
* Return values (usually ignored):
|
|
|
|
* NET_RX_SUCCESS: no congestion
|
|
|
|
* NET_RX_DROP: packet was dropped
|
|
|
|
*/
|
|
|
|
int netif_receive_skb(struct sk_buff *skb)
|
|
|
|
{
|
2011-11-15 04:12:55 +00:00
|
|
|
net_timestamp_check(netdev_tstamp_prequeue, skb);
|
net: Consistent skb timestamping
With RPS inclusion, skb timestamping is not consistent in RX path.
If netif_receive_skb() is used, its deferred after RPS dispatch.
If netif_rx() is used, its done before RPS dispatch.
This can give strange tcpdump timestamps results.
I think timestamping should be done as soon as possible in the receive
path, to get meaningful values (ie timestamps taken at the time packet
was delivered by NIC driver to our stack), even if NAPI already can
defer timestamping a bit (RPS can help to reduce the gap)
Tom Herbert prefer to sample timestamps after RPS dispatch. In case
sampling is expensive (HPET/acpi_pm on x86), this makes sense.
Let admins switch from one mode to another, using a new
sysctl, /proc/sys/net/core/netdev_tstamp_prequeue
Its default value (1), means timestamps are taken as soon as possible,
before backlog queueing, giving accurate timestamps.
Setting a 0 value permits to sample timestamps when processing backlog,
after RPS dispatch, to lower the load of the pre-RPS cpu.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-16 06:57:10 +00:00
|
|
|
|
2010-07-17 08:49:36 +00:00
|
|
|
if (skb_defer_rx_timestamp(skb))
|
|
|
|
return NET_RX_SUCCESS;
|
|
|
|
|
2010-03-24 19:13:54 +00:00
|
|
|
#ifdef CONFIG_RPS
|
2012-02-24 07:31:31 +00:00
|
|
|
if (static_key_false(&rps_needed)) {
|
net: Consistent skb timestamping
With RPS inclusion, skb timestamping is not consistent in RX path.
If netif_receive_skb() is used, its deferred after RPS dispatch.
If netif_rx() is used, its done before RPS dispatch.
This can give strange tcpdump timestamps results.
I think timestamping should be done as soon as possible in the receive
path, to get meaningful values (ie timestamps taken at the time packet
was delivered by NIC driver to our stack), even if NAPI already can
defer timestamping a bit (RPS can help to reduce the gap)
Tom Herbert prefer to sample timestamps after RPS dispatch. In case
sampling is expensive (HPET/acpi_pm on x86), this makes sense.
Let admins switch from one mode to another, using a new
sysctl, /proc/sys/net/core/netdev_tstamp_prequeue
Its default value (1), means timestamps are taken as soon as possible,
before backlog queueing, giving accurate timestamps.
Setting a 0 value permits to sample timestamps when processing backlog,
after RPS dispatch, to lower the load of the pre-RPS cpu.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-16 06:57:10 +00:00
|
|
|
struct rps_dev_flow voidflow, *rflow = &voidflow;
|
|
|
|
int cpu, ret;
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
|
net: Consistent skb timestamping
With RPS inclusion, skb timestamping is not consistent in RX path.
If netif_receive_skb() is used, its deferred after RPS dispatch.
If netif_rx() is used, its done before RPS dispatch.
This can give strange tcpdump timestamps results.
I think timestamping should be done as soon as possible in the receive
path, to get meaningful values (ie timestamps taken at the time packet
was delivered by NIC driver to our stack), even if NAPI already can
defer timestamping a bit (RPS can help to reduce the gap)
Tom Herbert prefer to sample timestamps after RPS dispatch. In case
sampling is expensive (HPET/acpi_pm on x86), this makes sense.
Let admins switch from one mode to another, using a new
sysctl, /proc/sys/net/core/netdev_tstamp_prequeue
Its default value (1), means timestamps are taken as soon as possible,
before backlog queueing, giving accurate timestamps.
Setting a 0 value permits to sample timestamps when processing backlog,
after RPS dispatch, to lower the load of the pre-RPS cpu.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-16 06:57:10 +00:00
|
|
|
rcu_read_lock();
|
|
|
|
|
|
|
|
cpu = get_rps_cpu(skb->dev, skb, &rflow);
|
2010-03-16 08:03:29 +00:00
|
|
|
|
net: Consistent skb timestamping
With RPS inclusion, skb timestamping is not consistent in RX path.
If netif_receive_skb() is used, its deferred after RPS dispatch.
If netif_rx() is used, its done before RPS dispatch.
This can give strange tcpdump timestamps results.
I think timestamping should be done as soon as possible in the receive
path, to get meaningful values (ie timestamps taken at the time packet
was delivered by NIC driver to our stack), even if NAPI already can
defer timestamping a bit (RPS can help to reduce the gap)
Tom Herbert prefer to sample timestamps after RPS dispatch. In case
sampling is expensive (HPET/acpi_pm on x86), this makes sense.
Let admins switch from one mode to another, using a new
sysctl, /proc/sys/net/core/netdev_tstamp_prequeue
Its default value (1), means timestamps are taken as soon as possible,
before backlog queueing, giving accurate timestamps.
Setting a 0 value permits to sample timestamps when processing backlog,
after RPS dispatch, to lower the load of the pre-RPS cpu.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-16 06:57:10 +00:00
|
|
|
if (cpu >= 0) {
|
|
|
|
ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
|
|
|
|
rcu_read_unlock();
|
2011-11-17 03:13:26 +00:00
|
|
|
return ret;
|
net: Consistent skb timestamping
With RPS inclusion, skb timestamping is not consistent in RX path.
If netif_receive_skb() is used, its deferred after RPS dispatch.
If netif_rx() is used, its done before RPS dispatch.
This can give strange tcpdump timestamps results.
I think timestamping should be done as soon as possible in the receive
path, to get meaningful values (ie timestamps taken at the time packet
was delivered by NIC driver to our stack), even if NAPI already can
defer timestamping a bit (RPS can help to reduce the gap)
Tom Herbert prefer to sample timestamps after RPS dispatch. In case
sampling is expensive (HPET/acpi_pm on x86), this makes sense.
Let admins switch from one mode to another, using a new
sysctl, /proc/sys/net/core/netdev_tstamp_prequeue
Its default value (1), means timestamps are taken as soon as possible,
before backlog queueing, giving accurate timestamps.
Setting a 0 value permits to sample timestamps when processing backlog,
after RPS dispatch, to lower the load of the pre-RPS cpu.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-05-16 06:57:10 +00:00
|
|
|
}
|
2011-11-17 03:13:26 +00:00
|
|
|
rcu_read_unlock();
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
}
|
2010-03-19 00:45:44 +00:00
|
|
|
#endif
|
2011-11-17 03:13:26 +00:00
|
|
|
return __netif_receive_skb(skb);
|
2010-03-16 08:03:29 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(netif_receive_skb);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-04-19 05:07:33 +00:00
|
|
|
/* Network device is going away, flush any packets still pending
|
|
|
|
* Called with irqs disabled.
|
|
|
|
*/
|
2010-03-30 20:16:22 +00:00
|
|
|
static void flush_backlog(void *arg)
|
2008-08-04 04:29:57 +00:00
|
|
|
{
|
2010-03-30 20:16:22 +00:00
|
|
|
struct net_device *dev = arg;
|
2010-04-19 21:17:14 +00:00
|
|
|
struct softnet_data *sd = &__get_cpu_var(softnet_data);
|
2008-08-04 04:29:57 +00:00
|
|
|
struct sk_buff *skb, *tmp;
|
|
|
|
|
2010-04-19 21:17:14 +00:00
|
|
|
rps_lock(sd);
|
2010-04-27 22:07:33 +00:00
|
|
|
skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
|
2008-08-04 04:29:57 +00:00
|
|
|
if (skb->dev == dev) {
|
2010-04-19 21:17:14 +00:00
|
|
|
__skb_unlink(skb, &sd->input_pkt_queue);
|
2008-08-04 04:29:57 +00:00
|
|
|
kfree_skb(skb);
|
2010-05-20 18:37:59 +00:00
|
|
|
input_queue_head_incr(sd);
|
2008-08-04 04:29:57 +00:00
|
|
|
}
|
2010-04-27 22:07:33 +00:00
|
|
|
}
|
2010-04-19 21:17:14 +00:00
|
|
|
rps_unlock(sd);
|
2010-04-27 22:07:33 +00:00
|
|
|
|
|
|
|
skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
|
|
|
|
if (skb->dev == dev) {
|
|
|
|
__skb_unlink(skb, &sd->process_queue);
|
|
|
|
kfree_skb(skb);
|
2010-05-20 18:37:59 +00:00
|
|
|
input_queue_head_incr(sd);
|
2010-04-27 22:07:33 +00:00
|
|
|
}
|
|
|
|
}
|
2008-08-04 04:29:57 +00:00
|
|
|
}
|
|
|
|
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
static int napi_gro_complete(struct sk_buff *skb)
|
|
|
|
{
|
2012-11-15 08:49:11 +00:00
|
|
|
struct packet_offload *ptype;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
__be16 type = skb->protocol;
|
2012-11-15 08:49:11 +00:00
|
|
|
struct list_head *head = &offload_base;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
int err = -ENOENT;
|
|
|
|
|
2012-12-06 13:54:59 +00:00
|
|
|
BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
|
|
|
|
|
2009-04-14 22:11:06 +00:00
|
|
|
if (NAPI_GRO_CB(skb)->count == 1) {
|
|
|
|
skb_shinfo(skb)->gso_size = 0;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
goto out;
|
2009-04-14 22:11:06 +00:00
|
|
|
}
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
list_for_each_entry_rcu(ptype, head, list) {
|
2012-11-15 08:49:23 +00:00
|
|
|
if (ptype->type != type || !ptype->callbacks.gro_complete)
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
continue;
|
|
|
|
|
2012-11-15 08:49:23 +00:00
|
|
|
err = ptype->callbacks.gro_complete(skb);
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
if (err) {
|
|
|
|
WARN_ON(&ptype->list == head);
|
|
|
|
kfree_skb(skb);
|
|
|
|
return NET_RX_SUCCESS;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
return netif_receive_skb(skb);
|
|
|
|
}
|
|
|
|
|
2012-10-06 08:08:49 +00:00
|
|
|
/* napi->gro_list contains packets ordered by age.
|
|
|
|
* youngest packets at the head of it.
|
|
|
|
* Complete skbs in reverse order to reduce latencies.
|
|
|
|
*/
|
|
|
|
void napi_gro_flush(struct napi_struct *napi, bool flush_old)
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
{
|
2012-10-06 08:08:49 +00:00
|
|
|
struct sk_buff *skb, *prev = NULL;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
|
2012-10-06 08:08:49 +00:00
|
|
|
/* scan list and build reverse chain */
|
|
|
|
for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
|
|
|
|
skb->prev = prev;
|
|
|
|
prev = skb;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (skb = prev; skb; skb = prev) {
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
skb->next = NULL;
|
2012-10-06 08:08:49 +00:00
|
|
|
|
|
|
|
if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
|
|
|
|
return;
|
|
|
|
|
|
|
|
prev = skb->prev;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
napi_gro_complete(skb);
|
2012-10-06 08:08:49 +00:00
|
|
|
napi->gro_count--;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
napi->gro_list = NULL;
|
|
|
|
}
|
2010-08-31 18:25:32 +00:00
|
|
|
EXPORT_SYMBOL(napi_gro_flush);
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
|
2012-12-10 13:28:16 +00:00
|
|
|
static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
struct sk_buff *p;
|
|
|
|
unsigned int maclen = skb->dev->hard_header_len;
|
|
|
|
|
|
|
|
for (p = napi->gro_list; p; p = p->next) {
|
|
|
|
unsigned long diffs;
|
|
|
|
|
|
|
|
diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
|
|
|
|
diffs |= p->vlan_tci ^ skb->vlan_tci;
|
|
|
|
if (maclen == ETH_HLEN)
|
|
|
|
diffs |= compare_ether_header(skb_mac_header(p),
|
|
|
|
skb_gro_mac_header(skb));
|
|
|
|
else if (!diffs)
|
|
|
|
diffs = memcmp(skb_mac_header(p),
|
|
|
|
skb_gro_mac_header(skb),
|
|
|
|
maclen);
|
|
|
|
NAPI_GRO_CB(p)->same_flow = !diffs;
|
|
|
|
NAPI_GRO_CB(p)->flush = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-11-28 21:55:25 +00:00
|
|
|
static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
{
|
|
|
|
struct sk_buff **pp = NULL;
|
2012-11-15 08:49:11 +00:00
|
|
|
struct packet_offload *ptype;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
__be16 type = skb->protocol;
|
2012-11-15 08:49:11 +00:00
|
|
|
struct list_head *head = &offload_base;
|
2008-12-26 22:57:42 +00:00
|
|
|
int same_flow;
|
2009-10-29 07:17:09 +00:00
|
|
|
enum gro_result ret;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
|
2010-08-05 01:19:11 +00:00
|
|
|
if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
goto normal;
|
|
|
|
|
2010-08-23 07:13:46 +00:00
|
|
|
if (skb_is_gso(skb) || skb_has_frag_list(skb))
|
2009-01-14 22:36:12 +00:00
|
|
|
goto normal;
|
|
|
|
|
2012-12-10 13:28:16 +00:00
|
|
|
gro_list_prepare(napi, skb);
|
|
|
|
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
rcu_read_lock();
|
|
|
|
list_for_each_entry_rcu(ptype, head, list) {
|
2012-11-15 08:49:23 +00:00
|
|
|
if (ptype->type != type || !ptype->callbacks.gro_receive)
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
continue;
|
|
|
|
|
2009-01-29 14:19:50 +00:00
|
|
|
skb_set_network_header(skb, skb_gro_offset(skb));
|
2013-02-14 17:31:48 +00:00
|
|
|
skb_reset_mac_len(skb);
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
NAPI_GRO_CB(skb)->same_flow = 0;
|
|
|
|
NAPI_GRO_CB(skb)->flush = 0;
|
2009-01-05 00:13:40 +00:00
|
|
|
NAPI_GRO_CB(skb)->free = 0;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
|
2012-11-15 08:49:23 +00:00
|
|
|
pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
if (&ptype->list == head)
|
|
|
|
goto normal;
|
|
|
|
|
2008-12-26 22:57:42 +00:00
|
|
|
same_flow = NAPI_GRO_CB(skb)->same_flow;
|
2009-01-29 14:19:48 +00:00
|
|
|
ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
|
2008-12-26 22:57:42 +00:00
|
|
|
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
if (pp) {
|
|
|
|
struct sk_buff *nskb = *pp;
|
|
|
|
|
|
|
|
*pp = nskb->next;
|
|
|
|
nskb->next = NULL;
|
|
|
|
napi_gro_complete(nskb);
|
2009-02-08 18:00:36 +00:00
|
|
|
napi->gro_count--;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
}
|
|
|
|
|
2008-12-26 22:57:42 +00:00
|
|
|
if (same_flow)
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
goto ok;
|
|
|
|
|
2009-02-08 18:00:36 +00:00
|
|
|
if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
goto normal;
|
|
|
|
|
2009-02-08 18:00:36 +00:00
|
|
|
napi->gro_count++;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
NAPI_GRO_CB(skb)->count = 1;
|
2012-10-06 08:08:49 +00:00
|
|
|
NAPI_GRO_CB(skb)->age = jiffies;
|
2009-01-29 14:19:50 +00:00
|
|
|
skb_shinfo(skb)->gso_size = skb_gro_len(skb);
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
skb->next = napi->gro_list;
|
|
|
|
napi->gro_list = skb;
|
2009-01-29 14:19:48 +00:00
|
|
|
ret = GRO_HELD;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
|
2009-02-01 09:24:55 +00:00
|
|
|
pull:
|
2009-05-26 18:50:31 +00:00
|
|
|
if (skb_headlen(skb) < skb_gro_offset(skb)) {
|
|
|
|
int grow = skb_gro_offset(skb) - skb_headlen(skb);
|
|
|
|
|
|
|
|
BUG_ON(skb->end - skb->tail < grow);
|
|
|
|
|
|
|
|
memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
|
|
|
|
|
|
|
|
skb->tail += grow;
|
|
|
|
skb->data_len -= grow;
|
|
|
|
|
|
|
|
skb_shinfo(skb)->frags[0].page_offset += grow;
|
2011-10-18 21:00:24 +00:00
|
|
|
skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
|
2009-05-26 18:50:31 +00:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
|
2011-08-22 23:44:58 +00:00
|
|
|
skb_frag_unref(skb, 0);
|
2009-05-26 18:50:31 +00:00
|
|
|
memmove(skb_shinfo(skb)->frags,
|
|
|
|
skb_shinfo(skb)->frags + 1,
|
2010-08-11 02:02:10 +00:00
|
|
|
--skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
|
2009-05-26 18:50:31 +00:00
|
|
|
}
|
2009-02-01 09:24:55 +00:00
|
|
|
}
|
|
|
|
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
ok:
|
2009-01-29 14:19:48 +00:00
|
|
|
return ret;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
|
|
|
|
normal:
|
2009-02-01 09:24:55 +00:00
|
|
|
ret = GRO_NORMAL;
|
|
|
|
goto pull;
|
2009-01-05 00:13:40 +00:00
|
|
|
}
|
2009-01-06 18:49:34 +00:00
|
|
|
|
2009-01-05 00:13:40 +00:00
|
|
|
|
2012-11-28 21:55:25 +00:00
|
|
|
static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
|
2009-01-05 00:13:40 +00:00
|
|
|
{
|
2009-01-29 14:19:48 +00:00
|
|
|
switch (ret) {
|
|
|
|
case GRO_NORMAL:
|
2009-10-30 04:36:53 +00:00
|
|
|
if (netif_receive_skb(skb))
|
|
|
|
ret = GRO_DROP;
|
|
|
|
break;
|
2009-01-05 00:13:40 +00:00
|
|
|
|
2009-01-29 14:19:48 +00:00
|
|
|
case GRO_DROP:
|
2009-01-05 00:13:40 +00:00
|
|
|
kfree_skb(skb);
|
|
|
|
break;
|
2009-10-29 07:17:09 +00:00
|
|
|
|
2012-04-19 07:07:40 +00:00
|
|
|
case GRO_MERGED_FREE:
|
2012-04-30 08:10:34 +00:00
|
|
|
if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
|
|
|
|
kmem_cache_free(skbuff_head_cache, skb);
|
|
|
|
else
|
|
|
|
__kfree_skb(skb);
|
2012-04-19 07:07:40 +00:00
|
|
|
break;
|
|
|
|
|
2009-10-29 07:17:09 +00:00
|
|
|
case GRO_HELD:
|
|
|
|
case GRO_MERGED:
|
|
|
|
break;
|
2009-01-05 00:13:40 +00:00
|
|
|
}
|
|
|
|
|
2009-10-30 04:36:53 +00:00
|
|
|
return ret;
|
2009-01-29 14:19:48 +00:00
|
|
|
}
|
|
|
|
|
2012-10-06 22:28:06 +00:00
|
|
|
static void skb_gro_reset_offset(struct sk_buff *skb)
|
2009-05-26 18:50:21 +00:00
|
|
|
{
|
2012-10-06 22:28:06 +00:00
|
|
|
const struct skb_shared_info *pinfo = skb_shinfo(skb);
|
|
|
|
const skb_frag_t *frag0 = &pinfo->frags[0];
|
|
|
|
|
2009-05-26 18:50:21 +00:00
|
|
|
NAPI_GRO_CB(skb)->data_offset = 0;
|
|
|
|
NAPI_GRO_CB(skb)->frag0 = NULL;
|
2009-05-26 18:50:27 +00:00
|
|
|
NAPI_GRO_CB(skb)->frag0_len = 0;
|
2009-05-26 18:50:21 +00:00
|
|
|
|
2009-05-26 18:50:23 +00:00
|
|
|
if (skb->mac_header == skb->tail &&
|
2012-10-06 22:28:06 +00:00
|
|
|
pinfo->nr_frags &&
|
|
|
|
!PageHighMem(skb_frag_page(frag0))) {
|
|
|
|
NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
|
|
|
|
NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
|
2009-05-26 18:50:27 +00:00
|
|
|
}
|
2009-05-26 18:50:21 +00:00
|
|
|
}
|
|
|
|
|
2009-10-30 04:36:53 +00:00
|
|
|
gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
|
2009-01-29 14:19:48 +00:00
|
|
|
{
|
2009-01-29 14:19:50 +00:00
|
|
|
skb_gro_reset_offset(skb);
|
|
|
|
|
2012-12-10 13:28:16 +00:00
|
|
|
return napi_skb_finish(dev_gro_receive(napi, skb), skb);
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(napi_gro_receive);
|
|
|
|
|
2010-10-19 07:12:10 +00:00
|
|
|
static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
|
2009-01-06 18:49:34 +00:00
|
|
|
{
|
|
|
|
__skb_pull(skb, skb_headlen(skb));
|
2012-03-21 06:58:03 +00:00
|
|
|
/* restore the reserve we had after netdev_alloc_skb_ip_align() */
|
|
|
|
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
|
2010-10-20 13:56:06 +00:00
|
|
|
skb->vlan_tci = 0;
|
2011-01-30 04:44:54 +00:00
|
|
|
skb->dev = napi->dev;
|
2011-02-02 22:53:25 +00:00
|
|
|
skb->skb_iif = 0;
|
2009-01-06 18:49:34 +00:00
|
|
|
|
|
|
|
napi->skb = skb;
|
|
|
|
}
|
|
|
|
|
2009-04-16 09:02:07 +00:00
|
|
|
struct sk_buff *napi_get_frags(struct napi_struct *napi)
|
2009-01-05 00:13:40 +00:00
|
|
|
{
|
|
|
|
struct sk_buff *skb = napi->skb;
|
|
|
|
|
|
|
|
if (!skb) {
|
2009-10-13 05:34:20 +00:00
|
|
|
skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
|
|
|
|
if (skb)
|
|
|
|
napi->skb = skb;
|
2009-01-29 14:19:52 +00:00
|
|
|
}
|
2009-01-06 18:49:34 +00:00
|
|
|
return skb;
|
|
|
|
}
|
2009-04-16 09:02:07 +00:00
|
|
|
EXPORT_SYMBOL(napi_get_frags);
|
2009-01-06 18:49:34 +00:00
|
|
|
|
2012-11-28 21:55:25 +00:00
|
|
|
static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
|
2009-10-30 04:36:53 +00:00
|
|
|
gro_result_t ret)
|
2009-01-06 18:49:34 +00:00
|
|
|
{
|
2009-01-29 14:19:48 +00:00
|
|
|
switch (ret) {
|
|
|
|
case GRO_NORMAL:
|
2009-01-29 14:19:50 +00:00
|
|
|
case GRO_HELD:
|
2010-02-16 20:25:43 +00:00
|
|
|
skb->protocol = eth_type_trans(skb, skb->dev);
|
2009-01-29 14:19:50 +00:00
|
|
|
|
2009-10-30 04:36:53 +00:00
|
|
|
if (ret == GRO_HELD)
|
|
|
|
skb_gro_pull(skb, -ETH_HLEN);
|
|
|
|
else if (netif_receive_skb(skb))
|
|
|
|
ret = GRO_DROP;
|
2009-01-29 14:19:50 +00:00
|
|
|
break;
|
2009-01-05 00:13:40 +00:00
|
|
|
|
2009-01-29 14:19:48 +00:00
|
|
|
case GRO_DROP:
|
|
|
|
case GRO_MERGED_FREE:
|
|
|
|
napi_reuse_skb(napi, skb);
|
|
|
|
break;
|
2009-10-29 07:17:09 +00:00
|
|
|
|
|
|
|
case GRO_MERGED:
|
|
|
|
break;
|
2009-01-29 14:19:48 +00:00
|
|
|
}
|
2009-01-05 00:13:40 +00:00
|
|
|
|
2009-10-30 04:36:53 +00:00
|
|
|
return ret;
|
2009-01-05 00:13:40 +00:00
|
|
|
}
|
2009-01-29 14:19:48 +00:00
|
|
|
|
2012-05-18 20:49:06 +00:00
|
|
|
static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
|
2009-04-16 09:02:07 +00:00
|
|
|
{
|
|
|
|
struct sk_buff *skb = napi->skb;
|
|
|
|
struct ethhdr *eth;
|
2009-05-26 18:50:28 +00:00
|
|
|
unsigned int hlen;
|
|
|
|
unsigned int off;
|
2009-04-16 09:02:07 +00:00
|
|
|
|
|
|
|
napi->skb = NULL;
|
|
|
|
|
|
|
|
skb_reset_mac_header(skb);
|
|
|
|
skb_gro_reset_offset(skb);
|
|
|
|
|
2009-05-26 18:50:28 +00:00
|
|
|
off = skb_gro_offset(skb);
|
|
|
|
hlen = off + sizeof(*eth);
|
|
|
|
eth = skb_gro_header_fast(skb, off);
|
|
|
|
if (skb_gro_header_hard(skb, hlen)) {
|
|
|
|
eth = skb_gro_header_slow(skb, hlen, off);
|
|
|
|
if (unlikely(!eth)) {
|
|
|
|
napi_reuse_skb(napi, skb);
|
|
|
|
skb = NULL;
|
|
|
|
goto out;
|
|
|
|
}
|
2009-04-16 09:02:07 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
skb_gro_pull(skb, sizeof(*eth));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This works because the only protocols we care about don't require
|
|
|
|
* special handling. We'll fix it up properly at the end.
|
|
|
|
*/
|
|
|
|
skb->protocol = eth->h_proto;
|
|
|
|
|
|
|
|
out:
|
|
|
|
return skb;
|
|
|
|
}
|
|
|
|
|
2009-10-30 04:36:53 +00:00
|
|
|
gro_result_t napi_gro_frags(struct napi_struct *napi)
|
2009-01-29 14:19:48 +00:00
|
|
|
{
|
2009-04-16 09:02:07 +00:00
|
|
|
struct sk_buff *skb = napi_frags_skb(napi);
|
2009-01-29 14:19:48 +00:00
|
|
|
|
|
|
|
if (!skb)
|
2009-10-30 04:36:53 +00:00
|
|
|
return GRO_DROP;
|
2009-01-29 14:19:48 +00:00
|
|
|
|
2012-12-10 13:28:16 +00:00
|
|
|
return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
|
2009-01-29 14:19:48 +00:00
|
|
|
}
|
2009-01-05 00:13:40 +00:00
|
|
|
EXPORT_SYMBOL(napi_gro_frags);
|
|
|
|
|
2010-04-22 07:22:45 +00:00
|
|
|
/*
|
|
|
|
* net_rps_action sends any pending IPI's for rps.
|
|
|
|
* Note: called with local irq disabled, but exits with local irq enabled.
|
|
|
|
*/
|
|
|
|
static void net_rps_action_and_irq_enable(struct softnet_data *sd)
|
|
|
|
{
|
|
|
|
#ifdef CONFIG_RPS
|
|
|
|
struct softnet_data *remsd = sd->rps_ipi_list;
|
|
|
|
|
|
|
|
if (remsd) {
|
|
|
|
sd->rps_ipi_list = NULL;
|
|
|
|
|
|
|
|
local_irq_enable();
|
|
|
|
|
|
|
|
/* Send pending IPI's to kick RPS processing on remote cpus. */
|
|
|
|
while (remsd) {
|
|
|
|
struct softnet_data *next = remsd->rps_ipi_next;
|
|
|
|
|
|
|
|
if (cpu_online(remsd->cpu))
|
|
|
|
__smp_call_function_single(remsd->cpu,
|
|
|
|
&remsd->csd, 0);
|
|
|
|
remsd = next;
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
#endif
|
|
|
|
local_irq_enable();
|
|
|
|
}
|
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
static int process_backlog(struct napi_struct *napi, int quota)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
int work = 0;
|
2010-05-07 05:07:48 +00:00
|
|
|
struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-04-22 07:22:45 +00:00
|
|
|
#ifdef CONFIG_RPS
|
|
|
|
/* Check if we have pending ipi, its better to send them now,
|
|
|
|
* not waiting net_rx_action() end.
|
|
|
|
*/
|
|
|
|
if (sd->rps_ipi_list) {
|
|
|
|
local_irq_disable();
|
|
|
|
net_rps_action_and_irq_enable(sd);
|
|
|
|
}
|
|
|
|
#endif
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
napi->weight = weight_p;
|
2010-04-27 22:07:33 +00:00
|
|
|
local_irq_disable();
|
|
|
|
while (work < quota) {
|
2005-04-16 22:20:36 +00:00
|
|
|
struct sk_buff *skb;
|
2010-04-27 22:07:33 +00:00
|
|
|
unsigned int qlen;
|
|
|
|
|
|
|
|
while ((skb = __skb_dequeue(&sd->process_queue))) {
|
|
|
|
local_irq_enable();
|
|
|
|
__netif_receive_skb(skb);
|
|
|
|
local_irq_disable();
|
2010-05-20 18:37:59 +00:00
|
|
|
input_queue_head_incr(sd);
|
|
|
|
if (++work >= quota) {
|
|
|
|
local_irq_enable();
|
|
|
|
return work;
|
|
|
|
}
|
2010-04-27 22:07:33 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-04-19 21:17:14 +00:00
|
|
|
rps_lock(sd);
|
2010-04-27 22:07:33 +00:00
|
|
|
qlen = skb_queue_len(&sd->input_pkt_queue);
|
2010-05-20 18:37:59 +00:00
|
|
|
if (qlen)
|
2010-04-27 22:07:33 +00:00
|
|
|
skb_queue_splice_tail_init(&sd->input_pkt_queue,
|
|
|
|
&sd->process_queue);
|
2010-05-20 18:37:59 +00:00
|
|
|
|
2010-04-27 22:07:33 +00:00
|
|
|
if (qlen < quota - work) {
|
2010-05-07 05:07:48 +00:00
|
|
|
/*
|
|
|
|
* Inline a custom version of __napi_complete().
|
|
|
|
* only current cpu owns and manipulates this napi,
|
|
|
|
* and NAPI_STATE_SCHED is the only possible flag set on backlog.
|
|
|
|
* we can use a plain write instead of clear_bit(),
|
|
|
|
* and we dont need an smp_mb() memory barrier.
|
|
|
|
*/
|
|
|
|
list_del(&napi->poll_list);
|
|
|
|
napi->state = 0;
|
|
|
|
|
2010-04-27 22:07:33 +00:00
|
|
|
quota = work + qlen;
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
}
|
2010-04-19 21:17:14 +00:00
|
|
|
rps_unlock(sd);
|
2010-04-27 22:07:33 +00:00
|
|
|
}
|
|
|
|
local_irq_enable();
|
2005-04-16 22:20:36 +00:00
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
return work;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
/**
|
|
|
|
* __napi_schedule - schedule for receive
|
2007-10-13 04:17:49 +00:00
|
|
|
* @n: entry to schedule
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
*
|
|
|
|
* The entry's receive function will be scheduled to run
|
|
|
|
*/
|
2008-02-13 23:03:16 +00:00
|
|
|
void __napi_schedule(struct napi_struct *n)
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
{
|
|
|
|
unsigned long flags;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
local_irq_save(flags);
|
2010-05-07 05:07:48 +00:00
|
|
|
____napi_schedule(&__get_cpu_var(softnet_data), n);
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
local_irq_restore(flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
EXPORT_SYMBOL(__napi_schedule);
|
|
|
|
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
void __napi_complete(struct napi_struct *n)
|
|
|
|
{
|
|
|
|
BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
|
|
|
|
BUG_ON(n->gro_list);
|
|
|
|
|
|
|
|
list_del(&n->poll_list);
|
|
|
|
smp_mb__before_clear_bit();
|
|
|
|
clear_bit(NAPI_STATE_SCHED, &n->state);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(__napi_complete);
|
|
|
|
|
|
|
|
void napi_complete(struct napi_struct *n)
|
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* don't let napi dequeue from the cpu poll list
|
|
|
|
* just in case its running on a different cpu
|
|
|
|
*/
|
|
|
|
if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
|
|
|
|
return;
|
|
|
|
|
2012-10-06 08:08:49 +00:00
|
|
|
napi_gro_flush(n, false);
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
local_irq_save(flags);
|
|
|
|
__napi_complete(n);
|
|
|
|
local_irq_restore(flags);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(napi_complete);
|
|
|
|
|
|
|
|
void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
|
|
|
|
int (*poll)(struct napi_struct *, int), int weight)
|
|
|
|
{
|
|
|
|
INIT_LIST_HEAD(&napi->poll_list);
|
2009-02-08 18:00:36 +00:00
|
|
|
napi->gro_count = 0;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
napi->gro_list = NULL;
|
2009-01-05 00:13:40 +00:00
|
|
|
napi->skb = NULL;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
napi->poll = poll;
|
2013-03-05 15:57:22 +00:00
|
|
|
if (weight > NAPI_POLL_WEIGHT)
|
|
|
|
pr_err_once("netif_napi_add() called with weight %d on device %s\n",
|
|
|
|
weight, dev->name);
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
napi->weight = weight;
|
|
|
|
list_add(&napi->dev_list, &dev->napi_list);
|
|
|
|
napi->dev = dev;
|
2009-01-05 00:13:40 +00:00
|
|
|
#ifdef CONFIG_NETPOLL
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
spin_lock_init(&napi->poll_lock);
|
|
|
|
napi->poll_owner = -1;
|
|
|
|
#endif
|
|
|
|
set_bit(NAPI_STATE_SCHED, &napi->state);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_napi_add);
|
|
|
|
|
|
|
|
void netif_napi_del(struct napi_struct *napi)
|
|
|
|
{
|
|
|
|
struct sk_buff *skb, *next;
|
|
|
|
|
2008-12-26 09:35:35 +00:00
|
|
|
list_del_init(&napi->dev_list);
|
2009-04-16 09:02:07 +00:00
|
|
|
napi_free_frags(napi);
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
|
|
|
|
for (skb = napi->gro_list; skb; skb = next) {
|
|
|
|
next = skb->next;
|
|
|
|
skb->next = NULL;
|
|
|
|
kfree_skb(skb);
|
|
|
|
}
|
|
|
|
|
|
|
|
napi->gro_list = NULL;
|
2009-02-08 18:00:36 +00:00
|
|
|
napi->gro_count = 0;
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_napi_del);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static void net_rx_action(struct softirq_action *h)
|
|
|
|
{
|
2010-04-22 07:22:45 +00:00
|
|
|
struct softnet_data *sd = &__get_cpu_var(softnet_data);
|
2008-11-04 01:14:38 +00:00
|
|
|
unsigned long time_limit = jiffies + 2;
|
2005-06-24 03:14:40 +00:00
|
|
|
int budget = netdev_budget;
|
2005-08-12 02:27:43 +00:00
|
|
|
void *have;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
local_irq_disable();
|
|
|
|
|
2010-04-22 07:22:45 +00:00
|
|
|
while (!list_empty(&sd->poll_list)) {
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
struct napi_struct *n;
|
|
|
|
int work, weight;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
/* If softirq window is exhuasted then punt.
|
2008-11-04 01:14:38 +00:00
|
|
|
* Allow this to run for 2 jiffies since which will allow
|
|
|
|
* an average latency of 1.5/HZ.
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
*/
|
2013-03-05 07:15:13 +00:00
|
|
|
if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
|
2005-04-16 22:20:36 +00:00
|
|
|
goto softnet_break;
|
|
|
|
|
|
|
|
local_irq_enable();
|
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
/* Even though interrupts have been re-enabled, this
|
|
|
|
* access is safe because interrupts can only add new
|
|
|
|
* entries to the tail of this list, and only ->poll()
|
|
|
|
* calls can remove this head entry from the list.
|
|
|
|
*/
|
2010-04-22 07:22:45 +00:00
|
|
|
n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
have = netpoll_poll_lock(n);
|
|
|
|
|
|
|
|
weight = n->weight;
|
|
|
|
|
2007-10-30 04:28:47 +00:00
|
|
|
/* This NAPI_STATE_SCHED test is for avoiding a race
|
|
|
|
* with netpoll's poll_napi(). Only the entity which
|
|
|
|
* obtains the lock and sees NAPI_STATE_SCHED set will
|
|
|
|
* actually make the ->poll() call. Therefore we avoid
|
2011-03-31 01:57:33 +00:00
|
|
|
* accidentally calling ->poll() when NAPI is not scheduled.
|
2007-10-30 04:28:47 +00:00
|
|
|
*/
|
|
|
|
work = 0;
|
2009-05-21 07:36:08 +00:00
|
|
|
if (test_bit(NAPI_STATE_SCHED, &n->state)) {
|
2007-10-30 04:28:47 +00:00
|
|
|
work = n->poll(n, weight);
|
2009-05-21 07:36:08 +00:00
|
|
|
trace_napi_poll(n);
|
|
|
|
}
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
|
|
|
|
WARN_ON_ONCE(work > weight);
|
|
|
|
|
|
|
|
budget -= work;
|
|
|
|
|
|
|
|
local_irq_disable();
|
|
|
|
|
|
|
|
/* Drivers must not modify the NAPI state if they
|
|
|
|
* consume the entire weight. In such cases this code
|
|
|
|
* still "owns" the NAPI instance and therefore can
|
|
|
|
* move the instance around on the list at-will.
|
|
|
|
*/
|
2008-01-08 05:00:40 +00:00
|
|
|
if (unlikely(work == weight)) {
|
2009-06-27 02:27:04 +00:00
|
|
|
if (unlikely(napi_disable_pending(n))) {
|
|
|
|
local_irq_enable();
|
|
|
|
napi_complete(n);
|
|
|
|
local_irq_disable();
|
2012-10-06 08:08:49 +00:00
|
|
|
} else {
|
|
|
|
if (n->gro_list) {
|
|
|
|
/* flush too old packets
|
|
|
|
* If HZ < 1000, flush all packets.
|
|
|
|
*/
|
|
|
|
local_irq_enable();
|
|
|
|
napi_gro_flush(n, HZ >= 1000);
|
|
|
|
local_irq_disable();
|
|
|
|
}
|
2010-04-22 07:22:45 +00:00
|
|
|
list_move_tail(&n->poll_list, &sd->poll_list);
|
2012-10-06 08:08:49 +00:00
|
|
|
}
|
2008-01-08 05:00:40 +00:00
|
|
|
}
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
|
|
|
|
netpoll_poll_unlock(have);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
out:
|
2010-04-22 07:22:45 +00:00
|
|
|
net_rps_action_and_irq_enable(sd);
|
2010-03-16 08:03:29 +00:00
|
|
|
|
2006-06-18 04:24:58 +00:00
|
|
|
#ifdef CONFIG_NET_DMA
|
|
|
|
/*
|
|
|
|
* There may not be any more sk_buffs coming right now, so push
|
|
|
|
* any pending DMA copies to hardware
|
|
|
|
*/
|
2009-01-06 18:38:14 +00:00
|
|
|
dma_issue_pending_all();
|
2006-06-18 04:24:58 +00:00
|
|
|
#endif
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return;
|
|
|
|
|
|
|
|
softnet_break:
|
2010-05-02 05:42:16 +00:00
|
|
|
sd->time_squeeze++;
|
2005-04-16 22:20:36 +00:00
|
|
|
__raise_softirq_irqoff(NET_RX_SOFTIRQ);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2013-01-03 22:48:49 +00:00
|
|
|
struct netdev_upper {
|
|
|
|
struct net_device *dev;
|
|
|
|
bool master;
|
|
|
|
struct list_head list;
|
|
|
|
struct rcu_head rcu;
|
|
|
|
struct list_head search_list;
|
|
|
|
};
|
|
|
|
|
|
|
|
static void __append_search_uppers(struct list_head *search_list,
|
|
|
|
struct net_device *dev)
|
|
|
|
{
|
|
|
|
struct netdev_upper *upper;
|
|
|
|
|
|
|
|
list_for_each_entry(upper, &dev->upper_dev_list, list) {
|
|
|
|
/* check if this upper is not already in search list */
|
|
|
|
if (list_empty(&upper->search_list))
|
|
|
|
list_add_tail(&upper->search_list, search_list);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool __netdev_search_upper_dev(struct net_device *dev,
|
|
|
|
struct net_device *upper_dev)
|
|
|
|
{
|
|
|
|
LIST_HEAD(search_list);
|
|
|
|
struct netdev_upper *upper;
|
|
|
|
struct netdev_upper *tmp;
|
|
|
|
bool ret = false;
|
|
|
|
|
|
|
|
__append_search_uppers(&search_list, dev);
|
|
|
|
list_for_each_entry(upper, &search_list, search_list) {
|
|
|
|
if (upper->dev == upper_dev) {
|
|
|
|
ret = true;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
__append_search_uppers(&search_list, upper->dev);
|
|
|
|
}
|
|
|
|
list_for_each_entry_safe(upper, tmp, &search_list, search_list)
|
|
|
|
INIT_LIST_HEAD(&upper->search_list);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct netdev_upper *__netdev_find_upper(struct net_device *dev,
|
|
|
|
struct net_device *upper_dev)
|
|
|
|
{
|
|
|
|
struct netdev_upper *upper;
|
|
|
|
|
|
|
|
list_for_each_entry(upper, &dev->upper_dev_list, list) {
|
|
|
|
if (upper->dev == upper_dev)
|
|
|
|
return upper;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_has_upper_dev - Check if device is linked to an upper device
|
|
|
|
* @dev: device
|
|
|
|
* @upper_dev: upper device to check
|
|
|
|
*
|
|
|
|
* Find out if a device is linked to specified upper device and return true
|
|
|
|
* in case it is. Note that this checks only immediate upper device,
|
|
|
|
* not through a complete stack of devices. The caller must hold the RTNL lock.
|
|
|
|
*/
|
|
|
|
bool netdev_has_upper_dev(struct net_device *dev,
|
|
|
|
struct net_device *upper_dev)
|
|
|
|
{
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
return __netdev_find_upper(dev, upper_dev);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_has_upper_dev);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_has_any_upper_dev - Check if device is linked to some device
|
|
|
|
* @dev: device
|
|
|
|
*
|
|
|
|
* Find out if a device is linked to an upper device and return true in case
|
|
|
|
* it is. The caller must hold the RTNL lock.
|
|
|
|
*/
|
|
|
|
bool netdev_has_any_upper_dev(struct net_device *dev)
|
|
|
|
{
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
return !list_empty(&dev->upper_dev_list);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_has_any_upper_dev);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_master_upper_dev_get - Get master upper device
|
|
|
|
* @dev: device
|
|
|
|
*
|
|
|
|
* Find a master upper device and return pointer to it or NULL in case
|
|
|
|
* it's not there. The caller must hold the RTNL lock.
|
|
|
|
*/
|
|
|
|
struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
|
|
|
|
{
|
|
|
|
struct netdev_upper *upper;
|
|
|
|
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
if (list_empty(&dev->upper_dev_list))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
upper = list_first_entry(&dev->upper_dev_list,
|
|
|
|
struct netdev_upper, list);
|
|
|
|
if (likely(upper->master))
|
|
|
|
return upper->dev;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_master_upper_dev_get);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_master_upper_dev_get_rcu - Get master upper device
|
|
|
|
* @dev: device
|
|
|
|
*
|
|
|
|
* Find a master upper device and return pointer to it or NULL in case
|
|
|
|
* it's not there. The caller must hold the RCU read lock.
|
|
|
|
*/
|
|
|
|
struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
|
|
|
|
{
|
|
|
|
struct netdev_upper *upper;
|
|
|
|
|
|
|
|
upper = list_first_or_null_rcu(&dev->upper_dev_list,
|
|
|
|
struct netdev_upper, list);
|
|
|
|
if (upper && likely(upper->master))
|
|
|
|
return upper->dev;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
|
|
|
|
|
|
|
|
static int __netdev_upper_dev_link(struct net_device *dev,
|
|
|
|
struct net_device *upper_dev, bool master)
|
|
|
|
{
|
|
|
|
struct netdev_upper *upper;
|
|
|
|
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
if (dev == upper_dev)
|
|
|
|
return -EBUSY;
|
|
|
|
|
|
|
|
/* To prevent loops, check if dev is not upper device to upper_dev. */
|
|
|
|
if (__netdev_search_upper_dev(upper_dev, dev))
|
|
|
|
return -EBUSY;
|
|
|
|
|
|
|
|
if (__netdev_find_upper(dev, upper_dev))
|
|
|
|
return -EEXIST;
|
|
|
|
|
|
|
|
if (master && netdev_master_upper_dev_get(dev))
|
|
|
|
return -EBUSY;
|
|
|
|
|
|
|
|
upper = kmalloc(sizeof(*upper), GFP_KERNEL);
|
|
|
|
if (!upper)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
upper->dev = upper_dev;
|
|
|
|
upper->master = master;
|
|
|
|
INIT_LIST_HEAD(&upper->search_list);
|
|
|
|
|
|
|
|
/* Ensure that master upper link is always the first item in list. */
|
|
|
|
if (master)
|
|
|
|
list_add_rcu(&upper->list, &dev->upper_dev_list);
|
|
|
|
else
|
|
|
|
list_add_tail_rcu(&upper->list, &dev->upper_dev_list);
|
|
|
|
dev_hold(upper_dev);
|
2013-05-25 04:12:10 +00:00
|
|
|
call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
|
2013-01-03 22:48:49 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_upper_dev_link - Add a link to the upper device
|
|
|
|
* @dev: device
|
|
|
|
* @upper_dev: new upper device
|
|
|
|
*
|
|
|
|
* Adds a link to device which is upper to this one. The caller must hold
|
|
|
|
* the RTNL lock. On a failure a negative errno code is returned.
|
|
|
|
* On success the reference counts are adjusted and the function
|
|
|
|
* returns zero.
|
|
|
|
*/
|
|
|
|
int netdev_upper_dev_link(struct net_device *dev,
|
|
|
|
struct net_device *upper_dev)
|
|
|
|
{
|
|
|
|
return __netdev_upper_dev_link(dev, upper_dev, false);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_upper_dev_link);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_master_upper_dev_link - Add a master link to the upper device
|
|
|
|
* @dev: device
|
|
|
|
* @upper_dev: new upper device
|
|
|
|
*
|
|
|
|
* Adds a link to device which is upper to this one. In this case, only
|
|
|
|
* one master upper device can be linked, although other non-master devices
|
|
|
|
* might be linked as well. The caller must hold the RTNL lock.
|
|
|
|
* On a failure a negative errno code is returned. On success the reference
|
|
|
|
* counts are adjusted and the function returns zero.
|
|
|
|
*/
|
|
|
|
int netdev_master_upper_dev_link(struct net_device *dev,
|
|
|
|
struct net_device *upper_dev)
|
|
|
|
{
|
|
|
|
return __netdev_upper_dev_link(dev, upper_dev, true);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_master_upper_dev_link);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* netdev_upper_dev_unlink - Removes a link to upper device
|
|
|
|
* @dev: device
|
|
|
|
* @upper_dev: new upper device
|
|
|
|
*
|
|
|
|
* Removes a link to device which is upper to this one. The caller must hold
|
|
|
|
* the RTNL lock.
|
|
|
|
*/
|
|
|
|
void netdev_upper_dev_unlink(struct net_device *dev,
|
|
|
|
struct net_device *upper_dev)
|
|
|
|
{
|
|
|
|
struct netdev_upper *upper;
|
|
|
|
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
upper = __netdev_find_upper(dev, upper_dev);
|
|
|
|
if (!upper)
|
|
|
|
return;
|
|
|
|
list_del_rcu(&upper->list);
|
|
|
|
dev_put(upper_dev);
|
|
|
|
kfree_rcu(upper, rcu);
|
2013-05-25 04:12:10 +00:00
|
|
|
call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
|
2013-01-03 22:48:49 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_upper_dev_unlink);
|
|
|
|
|
2008-10-07 22:26:48 +00:00
|
|
|
static void dev_change_rx_flags(struct net_device *dev, int flags)
|
|
|
|
{
|
2008-11-20 05:32:24 +00:00
|
|
|
const struct net_device_ops *ops = dev->netdev_ops;
|
|
|
|
|
|
|
|
if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
|
|
|
|
ops->ndo_change_rx_flags(dev, flags);
|
2008-10-07 22:26:48 +00:00
|
|
|
}
|
|
|
|
|
2008-06-18 08:48:28 +00:00
|
|
|
static int __dev_set_promiscuity(struct net_device *dev, int inc)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2011-11-30 21:42:26 +00:00
|
|
|
unsigned int old_flags = dev->flags;
|
2012-05-23 23:01:57 +00:00
|
|
|
kuid_t uid;
|
|
|
|
kgid_t gid;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-07-15 01:51:31 +00:00
|
|
|
ASSERT_RTNL();
|
|
|
|
|
2008-06-18 08:48:28 +00:00
|
|
|
dev->flags |= IFF_PROMISC;
|
|
|
|
dev->promiscuity += inc;
|
|
|
|
if (dev->promiscuity == 0) {
|
|
|
|
/*
|
|
|
|
* Avoid overflow.
|
|
|
|
* If inc causes overflow, untouch promisc and return error.
|
|
|
|
*/
|
|
|
|
if (inc < 0)
|
|
|
|
dev->flags &= ~IFF_PROMISC;
|
|
|
|
else {
|
|
|
|
dev->promiscuity -= inc;
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
|
|
|
|
dev->name);
|
2008-06-18 08:48:28 +00:00
|
|
|
return -EOVERFLOW;
|
|
|
|
}
|
|
|
|
}
|
2005-07-05 22:11:06 +00:00
|
|
|
if (dev->flags != old_flags) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_info("device %s %s promiscuous mode\n",
|
|
|
|
dev->name,
|
|
|
|
dev->flags & IFF_PROMISC ? "entered" : "left");
|
2008-11-13 23:39:10 +00:00
|
|
|
if (audit_enabled) {
|
|
|
|
current_uid_gid(&uid, &gid);
|
2008-01-24 03:57:45 +00:00
|
|
|
audit_log(current->audit_context, GFP_ATOMIC,
|
|
|
|
AUDIT_ANOM_PROMISCUOUS,
|
|
|
|
"dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
|
|
|
|
dev->name, (dev->flags & IFF_PROMISC),
|
|
|
|
(old_flags & IFF_PROMISC),
|
2012-09-11 05:39:43 +00:00
|
|
|
from_kuid(&init_user_ns, audit_get_loginuid(current)),
|
2012-05-23 23:01:57 +00:00
|
|
|
from_kuid(&init_user_ns, uid),
|
|
|
|
from_kgid(&init_user_ns, gid),
|
2008-01-24 03:57:45 +00:00
|
|
|
audit_get_sessionid(current));
|
2008-11-13 23:39:10 +00:00
|
|
|
}
|
2007-07-15 01:51:31 +00:00
|
|
|
|
2008-10-07 22:26:48 +00:00
|
|
|
dev_change_rx_flags(dev, IFF_PROMISC);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-06-18 08:48:28 +00:00
|
|
|
return 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2007-06-27 08:28:10 +00:00
|
|
|
/**
|
|
|
|
* dev_set_promiscuity - update promiscuity count on a device
|
|
|
|
* @dev: device
|
|
|
|
* @inc: modifier
|
|
|
|
*
|
|
|
|
* Add or remove promiscuity from a device. While the count in the device
|
|
|
|
* remains above zero the interface remains promiscuous. Once it hits zero
|
|
|
|
* the device reverts back to normal filtering operation. A negative inc
|
|
|
|
* value is used to drop promiscuity on the device.
|
2008-06-18 08:48:28 +00:00
|
|
|
* Return 0 if successful or a negative errno code on error.
|
2007-06-27 08:28:10 +00:00
|
|
|
*/
|
2008-06-18 08:48:28 +00:00
|
|
|
int dev_set_promiscuity(struct net_device *dev, int inc)
|
2007-06-27 08:28:10 +00:00
|
|
|
{
|
2011-11-30 21:42:26 +00:00
|
|
|
unsigned int old_flags = dev->flags;
|
2008-06-18 08:48:28 +00:00
|
|
|
int err;
|
2007-06-27 08:28:10 +00:00
|
|
|
|
2008-06-18 08:48:28 +00:00
|
|
|
err = __dev_set_promiscuity(dev, inc);
|
2008-07-06 22:49:08 +00:00
|
|
|
if (err < 0)
|
2008-06-18 08:48:28 +00:00
|
|
|
return err;
|
2007-06-27 08:28:10 +00:00
|
|
|
if (dev->flags != old_flags)
|
|
|
|
dev_set_rx_mode(dev);
|
2008-06-18 08:48:28 +00:00
|
|
|
return err;
|
2007-06-27 08:28:10 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_set_promiscuity);
|
2007-06-27 08:28:10 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* dev_set_allmulti - update allmulti count on a device
|
|
|
|
* @dev: device
|
|
|
|
* @inc: modifier
|
|
|
|
*
|
|
|
|
* Add or remove reception of all multicast frames to a device. While the
|
|
|
|
* count in the device remains above zero the interface remains listening
|
|
|
|
* to all interfaces. Once it hits zero the device reverts back to normal
|
|
|
|
* filtering operation. A negative @inc value is used to drop the counter
|
|
|
|
* when releasing a resource needing all multicasts.
|
2008-06-18 08:48:28 +00:00
|
|
|
* Return 0 if successful or a negative errno code on error.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
|
2008-06-18 08:48:28 +00:00
|
|
|
int dev_set_allmulti(struct net_device *dev, int inc)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2011-11-30 21:42:26 +00:00
|
|
|
unsigned int old_flags = dev->flags;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-07-15 01:51:31 +00:00
|
|
|
ASSERT_RTNL();
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->flags |= IFF_ALLMULTI;
|
2008-06-18 08:48:28 +00:00
|
|
|
dev->allmulti += inc;
|
|
|
|
if (dev->allmulti == 0) {
|
|
|
|
/*
|
|
|
|
* Avoid overflow.
|
|
|
|
* If inc causes overflow, untouch allmulti and return error.
|
|
|
|
*/
|
|
|
|
if (inc < 0)
|
|
|
|
dev->flags &= ~IFF_ALLMULTI;
|
|
|
|
else {
|
|
|
|
dev->allmulti -= inc;
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
|
|
|
|
dev->name);
|
2008-06-18 08:48:28 +00:00
|
|
|
return -EOVERFLOW;
|
|
|
|
}
|
|
|
|
}
|
2007-07-15 01:51:31 +00:00
|
|
|
if (dev->flags ^ old_flags) {
|
2008-10-07 22:26:48 +00:00
|
|
|
dev_change_rx_flags(dev, IFF_ALLMULTI);
|
2007-06-27 08:28:10 +00:00
|
|
|
dev_set_rx_mode(dev);
|
2007-07-15 01:51:31 +00:00
|
|
|
}
|
2008-06-18 08:48:28 +00:00
|
|
|
return 0;
|
2007-06-27 08:28:10 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_set_allmulti);
|
2007-06-27 08:28:10 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Upload unicast and multicast address lists to device and
|
|
|
|
* configure RX filtering. When the device doesn't support unicast
|
2007-12-20 22:02:06 +00:00
|
|
|
* filtering it is put in promiscuous mode while unicast addresses
|
2007-06-27 08:28:10 +00:00
|
|
|
* are present.
|
|
|
|
*/
|
|
|
|
void __dev_set_rx_mode(struct net_device *dev)
|
|
|
|
{
|
2008-11-20 05:32:24 +00:00
|
|
|
const struct net_device_ops *ops = dev->netdev_ops;
|
|
|
|
|
2007-06-27 08:28:10 +00:00
|
|
|
/* dev_open will call this function so the list will stay sane. */
|
|
|
|
if (!(dev->flags&IFF_UP))
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (!netif_device_present(dev))
|
2007-07-19 01:43:23 +00:00
|
|
|
return;
|
2007-06-27 08:28:10 +00:00
|
|
|
|
2011-08-16 06:29:00 +00:00
|
|
|
if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
|
2007-06-27 08:28:10 +00:00
|
|
|
/* Unicast addresses changes may only happen under the rtnl,
|
|
|
|
* therefore calling __dev_set_promiscuity here is safe.
|
|
|
|
*/
|
2010-01-25 21:36:10 +00:00
|
|
|
if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
|
2007-06-27 08:28:10 +00:00
|
|
|
__dev_set_promiscuity(dev, 1);
|
2011-07-25 23:17:35 +00:00
|
|
|
dev->uc_promisc = true;
|
2010-01-25 21:36:10 +00:00
|
|
|
} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
|
2007-06-27 08:28:10 +00:00
|
|
|
__dev_set_promiscuity(dev, -1);
|
2011-07-25 23:17:35 +00:00
|
|
|
dev->uc_promisc = false;
|
2007-06-27 08:28:10 +00:00
|
|
|
}
|
|
|
|
}
|
2011-08-16 06:29:00 +00:00
|
|
|
|
|
|
|
if (ops->ndo_set_rx_mode)
|
|
|
|
ops->ndo_set_rx_mode(dev);
|
2007-06-27 08:28:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void dev_set_rx_mode(struct net_device *dev)
|
|
|
|
{
|
2008-07-15 07:15:08 +00:00
|
|
|
netif_addr_lock_bh(dev);
|
2007-06-27 08:28:10 +00:00
|
|
|
__dev_set_rx_mode(dev);
|
2008-07-15 07:15:08 +00:00
|
|
|
netif_addr_unlock_bh(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2008-09-30 09:23:58 +00:00
|
|
|
/**
|
|
|
|
* dev_get_flags - get flags reported to userspace
|
|
|
|
* @dev: device
|
|
|
|
*
|
|
|
|
* Get the combination of flag bits exported through APIs to userspace.
|
|
|
|
*/
|
2012-04-15 05:58:06 +00:00
|
|
|
unsigned int dev_get_flags(const struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2012-04-15 05:58:06 +00:00
|
|
|
unsigned int flags;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
flags = (dev->flags & ~(IFF_PROMISC |
|
|
|
|
IFF_ALLMULTI |
|
2006-03-21 01:09:11 +00:00
|
|
|
IFF_RUNNING |
|
|
|
|
IFF_LOWER_UP |
|
|
|
|
IFF_DORMANT)) |
|
2005-04-16 22:20:36 +00:00
|
|
|
(dev->gflags & (IFF_PROMISC |
|
|
|
|
IFF_ALLMULTI));
|
|
|
|
|
2006-03-21 01:09:11 +00:00
|
|
|
if (netif_running(dev)) {
|
|
|
|
if (netif_oper_up(dev))
|
|
|
|
flags |= IFF_RUNNING;
|
|
|
|
if (netif_carrier_ok(dev))
|
|
|
|
flags |= IFF_LOWER_UP;
|
|
|
|
if (netif_dormant(dev))
|
|
|
|
flags |= IFF_DORMANT;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
return flags;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_get_flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-02-26 06:34:53 +00:00
|
|
|
int __dev_change_flags(struct net_device *dev, unsigned int flags)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2011-11-30 21:42:26 +00:00
|
|
|
unsigned int old_flags = dev->flags;
|
2010-02-26 06:34:53 +00:00
|
|
|
int ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-07-15 01:51:31 +00:00
|
|
|
ASSERT_RTNL();
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/*
|
|
|
|
* Set the flags on our device.
|
|
|
|
*/
|
|
|
|
|
|
|
|
dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
|
|
|
|
IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
|
|
|
|
IFF_AUTOMEDIA)) |
|
|
|
|
(dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
|
|
|
|
IFF_ALLMULTI));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Load in the correct multicast list now the flags have changed.
|
|
|
|
*/
|
|
|
|
|
2008-10-07 22:26:48 +00:00
|
|
|
if ((old_flags ^ flags) & IFF_MULTICAST)
|
|
|
|
dev_change_rx_flags(dev, IFF_MULTICAST);
|
2007-07-15 01:51:31 +00:00
|
|
|
|
2007-06-27 08:28:10 +00:00
|
|
|
dev_set_rx_mode(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Have we downed the interface. We handle IFF_UP ourselves
|
|
|
|
* according to user attempts to set it, rather than blindly
|
|
|
|
* setting it.
|
|
|
|
*/
|
|
|
|
|
|
|
|
ret = 0;
|
|
|
|
if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
|
2010-02-26 06:34:53 +00:00
|
|
|
ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if (!ret)
|
2007-06-27 08:28:10 +00:00
|
|
|
dev_set_rx_mode(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
if ((flags ^ dev->gflags) & IFF_PROMISC) {
|
2009-09-03 08:29:39 +00:00
|
|
|
int inc = (flags & IFF_PROMISC) ? 1 : -1;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->gflags ^= IFF_PROMISC;
|
|
|
|
dev_set_promiscuity(dev, inc);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
|
|
|
|
is important. Some (broken) drivers set IFF_PROMISC, when
|
|
|
|
IFF_ALLMULTI is requested not asking us and not reporting.
|
|
|
|
*/
|
|
|
|
if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
|
2009-09-03 08:29:39 +00:00
|
|
|
int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->gflags ^= IFF_ALLMULTI;
|
|
|
|
dev_set_allmulti(dev, inc);
|
|
|
|
}
|
|
|
|
|
2010-02-26 06:34:53 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
|
|
|
|
{
|
|
|
|
unsigned int changes = dev->flags ^ old_flags;
|
|
|
|
|
|
|
|
if (changes & IFF_UP) {
|
|
|
|
if (dev->flags & IFF_UP)
|
|
|
|
call_netdevice_notifiers(NETDEV_UP, dev);
|
|
|
|
else
|
|
|
|
call_netdevice_notifiers(NETDEV_DOWN, dev);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (dev->flags & IFF_UP &&
|
|
|
|
(changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
|
|
|
|
call_netdevice_notifiers(NETDEV_CHANGE, dev);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* dev_change_flags - change device settings
|
|
|
|
* @dev: device
|
|
|
|
* @flags: device state flags
|
|
|
|
*
|
|
|
|
* Change settings on device based state flags. The flags are
|
|
|
|
* in the userspace exported format.
|
|
|
|
*/
|
2011-11-30 21:42:26 +00:00
|
|
|
int dev_change_flags(struct net_device *dev, unsigned int flags)
|
2010-02-26 06:34:53 +00:00
|
|
|
{
|
2011-11-30 21:42:26 +00:00
|
|
|
int ret;
|
|
|
|
unsigned int changes, old_flags = dev->flags;
|
2010-02-26 06:34:53 +00:00
|
|
|
|
|
|
|
ret = __dev_change_flags(dev, flags);
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
changes = old_flags ^ dev->flags;
|
2007-06-05 23:03:03 +00:00
|
|
|
if (changes)
|
|
|
|
rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-02-26 06:34:53 +00:00
|
|
|
__dev_notify_flags(dev, old_flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
return ret;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_change_flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-09-30 09:23:58 +00:00
|
|
|
/**
|
|
|
|
* dev_set_mtu - Change maximum transfer unit
|
|
|
|
* @dev: device
|
|
|
|
* @new_mtu: new transfer unit
|
|
|
|
*
|
|
|
|
* Change the maximum transfer size of the network device.
|
|
|
|
*/
|
2005-04-16 22:20:36 +00:00
|
|
|
int dev_set_mtu(struct net_device *dev, int new_mtu)
|
|
|
|
{
|
2008-11-20 05:32:24 +00:00
|
|
|
const struct net_device_ops *ops = dev->netdev_ops;
|
2005-04-16 22:20:36 +00:00
|
|
|
int err;
|
|
|
|
|
|
|
|
if (new_mtu == dev->mtu)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* MTU must be positive. */
|
|
|
|
if (new_mtu < 0)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (!netif_device_present(dev))
|
|
|
|
return -ENODEV;
|
|
|
|
|
|
|
|
err = 0;
|
2008-11-20 05:32:24 +00:00
|
|
|
if (ops->ndo_change_mtu)
|
|
|
|
err = ops->ndo_change_mtu(dev, new_mtu);
|
2005-04-16 22:20:36 +00:00
|
|
|
else
|
|
|
|
dev->mtu = new_mtu;
|
2008-11-20 05:32:24 +00:00
|
|
|
|
2012-12-03 01:16:32 +00:00
|
|
|
if (!err)
|
2007-09-16 22:42:43 +00:00
|
|
|
call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
return err;
|
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_set_mtu);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2011-01-13 23:38:30 +00:00
|
|
|
/**
|
|
|
|
* dev_set_group - Change group this device belongs to
|
|
|
|
* @dev: device
|
|
|
|
* @new_group: group this device should belong to
|
|
|
|
*/
|
|
|
|
void dev_set_group(struct net_device *dev, int new_group)
|
|
|
|
{
|
|
|
|
dev->group = new_group;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_set_group);
|
|
|
|
|
2008-09-30 09:23:58 +00:00
|
|
|
/**
|
|
|
|
* dev_set_mac_address - Change Media Access Control Address
|
|
|
|
* @dev: device
|
|
|
|
* @sa: new address
|
|
|
|
*
|
|
|
|
* Change the hardware (MAC) address of the device
|
|
|
|
*/
|
2005-04-16 22:20:36 +00:00
|
|
|
int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
|
|
|
|
{
|
2008-11-20 05:32:24 +00:00
|
|
|
const struct net_device_ops *ops = dev->netdev_ops;
|
2005-04-16 22:20:36 +00:00
|
|
|
int err;
|
|
|
|
|
2008-11-20 05:32:24 +00:00
|
|
|
if (!ops->ndo_set_mac_address)
|
2005-04-16 22:20:36 +00:00
|
|
|
return -EOPNOTSUPP;
|
|
|
|
if (sa->sa_family != dev->type)
|
|
|
|
return -EINVAL;
|
|
|
|
if (!netif_device_present(dev))
|
|
|
|
return -ENODEV;
|
2008-11-20 05:32:24 +00:00
|
|
|
err = ops->ndo_set_mac_address(dev, sa);
|
2013-01-01 03:30:14 +00:00
|
|
|
if (err)
|
|
|
|
return err;
|
2013-01-01 03:30:16 +00:00
|
|
|
dev->addr_assign_type = NET_ADDR_SET;
|
2013-01-01 03:30:14 +00:00
|
|
|
call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
|
2012-07-05 01:23:25 +00:00
|
|
|
add_device_randomness(dev->dev_addr, dev->addr_len);
|
2013-01-01 03:30:14 +00:00
|
|
|
return 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(dev_set_mac_address);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2012-12-27 23:49:37 +00:00
|
|
|
/**
|
|
|
|
* dev_change_carrier - Change device carrier
|
|
|
|
* @dev: device
|
2013-03-04 12:32:43 +00:00
|
|
|
* @new_carrier: new value
|
2012-12-27 23:49:37 +00:00
|
|
|
*
|
|
|
|
* Change device carrier
|
|
|
|
*/
|
|
|
|
int dev_change_carrier(struct net_device *dev, bool new_carrier)
|
|
|
|
{
|
|
|
|
const struct net_device_ops *ops = dev->netdev_ops;
|
|
|
|
|
|
|
|
if (!ops->ndo_change_carrier)
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
if (!netif_device_present(dev))
|
|
|
|
return -ENODEV;
|
|
|
|
return ops->ndo_change_carrier(dev, new_carrier);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dev_change_carrier);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* dev_new_index - allocate an ifindex
|
2007-10-13 04:17:49 +00:00
|
|
|
* @net: the applicable net namespace
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* Returns a suitable unique value for a new device interface
|
|
|
|
* number. The caller must hold the rtnl semaphore or the
|
|
|
|
* dev_base_lock to be sure it remains unique.
|
|
|
|
*/
|
2007-09-17 18:56:21 +00:00
|
|
|
static int dev_new_index(struct net *net)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2012-08-08 21:53:19 +00:00
|
|
|
int ifindex = net->ifindex;
|
2005-04-16 22:20:36 +00:00
|
|
|
for (;;) {
|
|
|
|
if (++ifindex <= 0)
|
|
|
|
ifindex = 1;
|
2007-09-17 18:56:21 +00:00
|
|
|
if (!__dev_get_by_index(net, ifindex))
|
2012-08-08 21:53:19 +00:00
|
|
|
return net->ifindex = ifindex;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Delayed registration/unregisteration */
|
2007-12-07 08:49:17 +00:00
|
|
|
static LIST_HEAD(net_todo_list);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-03-09 04:46:03 +00:00
|
|
|
static void net_set_todo(struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
list_add_tail(&dev->todo_list, &net_todo_list);
|
|
|
|
}
|
|
|
|
|
2009-10-27 07:04:19 +00:00
|
|
|
static void rollback_registered_many(struct list_head *head)
|
2007-10-30 22:38:18 +00:00
|
|
|
{
|
net: Handle NETREG_UNINITIALIZED devices correctly
Fix two problems:
1. If unregister_netdevice_many() is called with both registered
and unregistered devices, rollback_registered_many() bails out
when it reaches the first unregistered device. The processing
of the prior registered devices is unfinished, and the
remaining devices are skipped, and possible registered netdev's
are leaked/unregistered.
2. System hangs or panics depending on how the devices are passed,
since when netdev_run_todo() runs, some devices were not fully
processed.
Tested by passing intermingled unregistered and registered vlan
devices to unregister_netdevice_many() as follows:
1. dev, fake_dev1, fake_dev2: hangs in run_todo
("unregister_netdevice: waiting for eth1.100 to become
free. Usage count = 1")
2. fake_dev1, dev, fake_dev2: failure during de-registration
and next registration, followed by a vlan driver Oops
during subsequent registration.
Confirmed that the patch fixes both cases.
Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-12-08 22:26:02 +00:00
|
|
|
struct net_device *dev, *tmp;
|
2009-10-27 07:04:19 +00:00
|
|
|
|
2007-10-30 22:38:18 +00:00
|
|
|
BUG_ON(dev_boot_phase);
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
net: Handle NETREG_UNINITIALIZED devices correctly
Fix two problems:
1. If unregister_netdevice_many() is called with both registered
and unregistered devices, rollback_registered_many() bails out
when it reaches the first unregistered device. The processing
of the prior registered devices is unfinished, and the
remaining devices are skipped, and possible registered netdev's
are leaked/unregistered.
2. System hangs or panics depending on how the devices are passed,
since when netdev_run_todo() runs, some devices were not fully
processed.
Tested by passing intermingled unregistered and registered vlan
devices to unregister_netdevice_many() as follows:
1. dev, fake_dev1, fake_dev2: hangs in run_todo
("unregister_netdevice: waiting for eth1.100 to become
free. Usage count = 1")
2. fake_dev1, dev, fake_dev2: failure during de-registration
and next registration, followed by a vlan driver Oops
during subsequent registration.
Confirmed that the patch fixes both cases.
Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-12-08 22:26:02 +00:00
|
|
|
list_for_each_entry_safe(dev, tmp, head, unreg_list) {
|
2009-10-27 07:04:19 +00:00
|
|
|
/* Some devices call without registering
|
net: Handle NETREG_UNINITIALIZED devices correctly
Fix two problems:
1. If unregister_netdevice_many() is called with both registered
and unregistered devices, rollback_registered_many() bails out
when it reaches the first unregistered device. The processing
of the prior registered devices is unfinished, and the
remaining devices are skipped, and possible registered netdev's
are leaked/unregistered.
2. System hangs or panics depending on how the devices are passed,
since when netdev_run_todo() runs, some devices were not fully
processed.
Tested by passing intermingled unregistered and registered vlan
devices to unregister_netdevice_many() as follows:
1. dev, fake_dev1, fake_dev2: hangs in run_todo
("unregister_netdevice: waiting for eth1.100 to become
free. Usage count = 1")
2. fake_dev1, dev, fake_dev2: failure during de-registration
and next registration, followed by a vlan driver Oops
during subsequent registration.
Confirmed that the patch fixes both cases.
Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-12-08 22:26:02 +00:00
|
|
|
* for initialization unwind. Remove those
|
|
|
|
* devices and proceed with the remaining.
|
2009-10-27 07:04:19 +00:00
|
|
|
*/
|
|
|
|
if (dev->reg_state == NETREG_UNINITIALIZED) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_debug("unregister_netdevice: device %s/%p never was registered\n",
|
|
|
|
dev->name, dev);
|
2007-10-30 22:38:18 +00:00
|
|
|
|
2009-10-27 07:04:19 +00:00
|
|
|
WARN_ON(1);
|
net: Handle NETREG_UNINITIALIZED devices correctly
Fix two problems:
1. If unregister_netdevice_many() is called with both registered
and unregistered devices, rollback_registered_many() bails out
when it reaches the first unregistered device. The processing
of the prior registered devices is unfinished, and the
remaining devices are skipped, and possible registered netdev's
are leaked/unregistered.
2. System hangs or panics depending on how the devices are passed,
since when netdev_run_todo() runs, some devices were not fully
processed.
Tested by passing intermingled unregistered and registered vlan
devices to unregister_netdevice_many() as follows:
1. dev, fake_dev1, fake_dev2: hangs in run_todo
("unregister_netdevice: waiting for eth1.100 to become
free. Usage count = 1")
2. fake_dev1, dev, fake_dev2: failure during de-registration
and next registration, followed by a vlan driver Oops
during subsequent registration.
Confirmed that the patch fixes both cases.
Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-12-08 22:26:02 +00:00
|
|
|
list_del(&dev->unreg_list);
|
|
|
|
continue;
|
2009-10-27 07:04:19 +00:00
|
|
|
}
|
2011-05-19 12:24:16 +00:00
|
|
|
dev->dismantle = true;
|
2009-10-27 07:04:19 +00:00
|
|
|
BUG_ON(dev->reg_state != NETREG_REGISTERED);
|
2010-12-13 12:44:07 +00:00
|
|
|
}
|
2007-10-30 22:38:18 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
/* If device is running, close it first. */
|
|
|
|
dev_close_many(head);
|
2007-10-30 22:38:18 +00:00
|
|
|
|
2010-12-13 12:44:07 +00:00
|
|
|
list_for_each_entry(dev, head, unreg_list) {
|
2009-10-27 07:04:19 +00:00
|
|
|
/* And unlink it from device chain. */
|
|
|
|
unlist_netdevice(dev);
|
2007-10-30 22:38:18 +00:00
|
|
|
|
2009-10-27 07:04:19 +00:00
|
|
|
dev->reg_state = NETREG_UNREGISTERING;
|
|
|
|
}
|
2007-10-30 22:38:18 +00:00
|
|
|
|
|
|
|
synchronize_net();
|
|
|
|
|
2009-10-27 07:04:19 +00:00
|
|
|
list_for_each_entry(dev, head, unreg_list) {
|
|
|
|
/* Shutdown queueing discipline. */
|
|
|
|
dev_shutdown(dev);
|
2007-10-30 22:38:18 +00:00
|
|
|
|
|
|
|
|
2009-10-27 07:04:19 +00:00
|
|
|
/* Notify protocols, that we are about to destroy
|
|
|
|
this device. They should clean all the things.
|
|
|
|
*/
|
|
|
|
call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
|
2007-10-30 22:38:18 +00:00
|
|
|
|
2010-02-26 06:34:51 +00:00
|
|
|
if (!dev->rtnl_link_ops ||
|
|
|
|
dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
|
|
|
|
rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
|
|
|
|
|
2009-10-27 07:04:19 +00:00
|
|
|
/*
|
|
|
|
* Flush the unicast and multicast chains
|
|
|
|
*/
|
2010-04-01 21:22:09 +00:00
|
|
|
dev_uc_flush(dev);
|
2010-04-01 21:22:57 +00:00
|
|
|
dev_mc_flush(dev);
|
2007-10-30 22:38:18 +00:00
|
|
|
|
2009-10-27 07:04:19 +00:00
|
|
|
if (dev->netdev_ops->ndo_uninit)
|
|
|
|
dev->netdev_ops->ndo_uninit(dev);
|
2007-10-30 22:38:18 +00:00
|
|
|
|
2013-01-03 22:48:49 +00:00
|
|
|
/* Notifier chain MUST detach us all upper devices. */
|
|
|
|
WARN_ON(netdev_has_any_upper_dev(dev));
|
2007-10-30 22:38:18 +00:00
|
|
|
|
2009-10-27 07:04:19 +00:00
|
|
|
/* Remove entries from kobject tree */
|
|
|
|
netdev_unregister_kobject(dev);
|
2013-01-10 08:57:46 +00:00
|
|
|
#ifdef CONFIG_XPS
|
|
|
|
/* Remove XPS queueing entries */
|
|
|
|
netif_reset_xps_queues_gt(dev, 0);
|
|
|
|
#endif
|
2009-10-27 07:04:19 +00:00
|
|
|
}
|
2007-10-30 22:38:18 +00:00
|
|
|
|
2011-10-13 22:25:23 +00:00
|
|
|
synchronize_net();
|
2009-11-16 13:49:35 +00:00
|
|
|
|
2009-11-29 15:45:58 +00:00
|
|
|
list_for_each_entry(dev, head, unreg_list)
|
2009-10-27 07:04:19 +00:00
|
|
|
dev_put(dev);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rollback_registered(struct net_device *dev)
|
|
|
|
{
|
|
|
|
LIST_HEAD(single);
|
|
|
|
|
|
|
|
list_add(&dev->unreg_list, &single);
|
|
|
|
rollback_registered_many(&single);
|
2011-02-17 22:59:19 +00:00
|
|
|
list_del(&single);
|
2007-10-30 22:38:18 +00:00
|
|
|
}
|
|
|
|
|
2011-11-15 15:29:55 +00:00
|
|
|
static netdev_features_t netdev_fix_features(struct net_device *dev,
|
|
|
|
netdev_features_t features)
|
2008-10-23 08:11:29 +00:00
|
|
|
{
|
2011-01-22 12:14:12 +00:00
|
|
|
/* Fix illegal checksum combinations */
|
|
|
|
if ((features & NETIF_F_HW_CSUM) &&
|
|
|
|
(features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
|
2011-05-16 19:14:21 +00:00
|
|
|
netdev_warn(dev, "mixed HW and IP checksum settings.\n");
|
2011-01-22 12:14:12 +00:00
|
|
|
features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
|
|
|
|
}
|
|
|
|
|
2008-10-23 08:11:29 +00:00
|
|
|
/* TSO requires that SG is present as well. */
|
2011-04-12 14:38:37 +00:00
|
|
|
if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
|
2011-05-16 19:14:21 +00:00
|
|
|
netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
|
2011-04-12 14:38:37 +00:00
|
|
|
features &= ~NETIF_F_ALL_TSO;
|
2008-10-23 08:11:29 +00:00
|
|
|
}
|
|
|
|
|
2013-03-07 09:28:01 +00:00
|
|
|
if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
|
|
|
|
!(features & NETIF_F_IP_CSUM)) {
|
|
|
|
netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
|
|
|
|
features &= ~NETIF_F_TSO;
|
|
|
|
features &= ~NETIF_F_TSO_ECN;
|
|
|
|
}
|
|
|
|
|
|
|
|
if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
|
|
|
|
!(features & NETIF_F_IPV6_CSUM)) {
|
|
|
|
netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
|
|
|
|
features &= ~NETIF_F_TSO6;
|
|
|
|
}
|
|
|
|
|
2011-04-12 14:47:15 +00:00
|
|
|
/* TSO ECN requires that TSO is present as well. */
|
|
|
|
if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
|
|
|
|
features &= ~NETIF_F_TSO_ECN;
|
|
|
|
|
2011-02-15 16:59:16 +00:00
|
|
|
/* Software GSO depends on SG. */
|
|
|
|
if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
|
2011-05-16 19:14:21 +00:00
|
|
|
netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
|
2011-02-15 16:59:16 +00:00
|
|
|
features &= ~NETIF_F_GSO;
|
|
|
|
}
|
|
|
|
|
2011-01-24 23:45:15 +00:00
|
|
|
/* UFO needs SG and checksumming */
|
2008-10-23 08:11:29 +00:00
|
|
|
if (features & NETIF_F_UFO) {
|
2010-11-30 06:38:00 +00:00
|
|
|
/* maybe split UFO into V4 and V6? */
|
|
|
|
if (!((features & NETIF_F_GEN_CSUM) ||
|
|
|
|
(features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
|
|
|
|
== (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
|
2011-05-16 19:14:21 +00:00
|
|
|
netdev_dbg(dev,
|
2011-01-24 23:45:15 +00:00
|
|
|
"Dropping NETIF_F_UFO since no checksum offload features.\n");
|
2008-10-23 08:11:29 +00:00
|
|
|
features &= ~NETIF_F_UFO;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!(features & NETIF_F_SG)) {
|
2011-05-16 19:14:21 +00:00
|
|
|
netdev_dbg(dev,
|
2011-01-24 23:45:15 +00:00
|
|
|
"Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
|
2008-10-23 08:11:29 +00:00
|
|
|
features &= ~NETIF_F_UFO;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return features;
|
|
|
|
}
|
|
|
|
|
2011-04-03 05:48:47 +00:00
|
|
|
int __netdev_update_features(struct net_device *dev)
|
2011-02-15 16:59:17 +00:00
|
|
|
{
|
2011-11-15 15:29:55 +00:00
|
|
|
netdev_features_t features;
|
2011-02-15 16:59:17 +00:00
|
|
|
int err = 0;
|
|
|
|
|
2011-04-12 09:56:38 +00:00
|
|
|
ASSERT_RTNL();
|
|
|
|
|
2011-02-15 16:59:17 +00:00
|
|
|
features = netdev_get_wanted_features(dev);
|
|
|
|
|
|
|
|
if (dev->netdev_ops->ndo_fix_features)
|
|
|
|
features = dev->netdev_ops->ndo_fix_features(dev, features);
|
|
|
|
|
|
|
|
/* driver might be less strict about feature dependencies */
|
|
|
|
features = netdev_fix_features(dev, features);
|
|
|
|
|
|
|
|
if (dev->features == features)
|
2011-04-03 05:48:47 +00:00
|
|
|
return 0;
|
2011-02-15 16:59:17 +00:00
|
|
|
|
2011-11-15 15:29:55 +00:00
|
|
|
netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
|
|
|
|
&dev->features, &features);
|
2011-02-15 16:59:17 +00:00
|
|
|
|
|
|
|
if (dev->netdev_ops->ndo_set_features)
|
|
|
|
err = dev->netdev_ops->ndo_set_features(dev, features);
|
|
|
|
|
2011-04-03 05:48:47 +00:00
|
|
|
if (unlikely(err < 0)) {
|
2011-02-15 16:59:17 +00:00
|
|
|
netdev_err(dev,
|
2011-11-15 15:29:55 +00:00
|
|
|
"set_features() failed (%d); wanted %pNF, left %pNF\n",
|
|
|
|
err, &features, &dev->features);
|
2011-04-03 05:48:47 +00:00
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!err)
|
|
|
|
dev->features = features;
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2011-05-07 03:22:17 +00:00
|
|
|
/**
|
|
|
|
* netdev_update_features - recalculate device features
|
|
|
|
* @dev: the device to check
|
|
|
|
*
|
|
|
|
* Recalculate dev->features set and send notifications if it
|
|
|
|
* has changed. Should be called after driver or hardware dependent
|
|
|
|
* conditions might have changed that influence the features.
|
|
|
|
*/
|
2011-04-03 05:48:47 +00:00
|
|
|
void netdev_update_features(struct net_device *dev)
|
|
|
|
{
|
|
|
|
if (__netdev_update_features(dev))
|
|
|
|
netdev_features_change(dev);
|
2011-02-15 16:59:17 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_update_features);
|
|
|
|
|
2011-05-07 03:22:17 +00:00
|
|
|
/**
|
|
|
|
* netdev_change_features - recalculate device features
|
|
|
|
* @dev: the device to check
|
|
|
|
*
|
|
|
|
* Recalculate dev->features set and send notifications even
|
|
|
|
* if they have not changed. Should be called instead of
|
|
|
|
* netdev_update_features() if also dev->vlan_features might
|
|
|
|
* have changed to allow the changes to be propagated to stacked
|
|
|
|
* VLAN devices.
|
|
|
|
*/
|
|
|
|
void netdev_change_features(struct net_device *dev)
|
|
|
|
{
|
|
|
|
__netdev_update_features(dev);
|
|
|
|
netdev_features_change(dev);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_change_features);
|
|
|
|
|
2009-12-03 23:59:22 +00:00
|
|
|
/**
|
|
|
|
* netif_stacked_transfer_operstate - transfer operstate
|
|
|
|
* @rootdev: the root or lower level device to transfer state from
|
|
|
|
* @dev: the device to transfer operstate to
|
|
|
|
*
|
|
|
|
* Transfer operational state from root to device. This is normally
|
|
|
|
* called when a stacking relationship exists between the root
|
|
|
|
* device and the device(a leaf device).
|
|
|
|
*/
|
|
|
|
void netif_stacked_transfer_operstate(const struct net_device *rootdev,
|
|
|
|
struct net_device *dev)
|
|
|
|
{
|
|
|
|
if (rootdev->operstate == IF_OPER_DORMANT)
|
|
|
|
netif_dormant_on(dev);
|
|
|
|
else
|
|
|
|
netif_dormant_off(dev);
|
|
|
|
|
|
|
|
if (netif_carrier_ok(rootdev)) {
|
|
|
|
if (!netif_carrier_ok(dev))
|
|
|
|
netif_carrier_on(dev);
|
|
|
|
} else {
|
|
|
|
if (netif_carrier_ok(dev))
|
|
|
|
netif_carrier_off(dev);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netif_stacked_transfer_operstate);
|
|
|
|
|
2010-11-26 08:36:09 +00:00
|
|
|
#ifdef CONFIG_RPS
|
2010-09-23 17:26:35 +00:00
|
|
|
static int netif_alloc_rx_queues(struct net_device *dev)
|
|
|
|
{
|
|
|
|
unsigned int i, count = dev->num_rx_queues;
|
2010-10-18 18:00:16 +00:00
|
|
|
struct netdev_rx_queue *rx;
|
2010-09-23 17:26:35 +00:00
|
|
|
|
2010-10-18 18:00:16 +00:00
|
|
|
BUG_ON(count < 1);
|
2010-09-23 17:26:35 +00:00
|
|
|
|
2010-10-18 18:00:16 +00:00
|
|
|
rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
|
2013-02-04 16:48:16 +00:00
|
|
|
if (!rx)
|
2010-10-18 18:00:16 +00:00
|
|
|
return -ENOMEM;
|
2013-02-04 16:48:16 +00:00
|
|
|
|
2010-10-18 18:00:16 +00:00
|
|
|
dev->_rx = rx;
|
|
|
|
|
|
|
|
for (i = 0; i < count; i++)
|
2010-11-09 10:47:38 +00:00
|
|
|
rx[i].dev = dev;
|
2010-09-23 17:26:35 +00:00
|
|
|
return 0;
|
|
|
|
}
|
2010-11-26 08:36:09 +00:00
|
|
|
#endif
|
2010-09-23 17:26:35 +00:00
|
|
|
|
2010-12-04 02:31:41 +00:00
|
|
|
static void netdev_init_one_queue(struct net_device *dev,
|
|
|
|
struct netdev_queue *queue, void *_unused)
|
|
|
|
{
|
|
|
|
/* Initialize queue lock */
|
|
|
|
spin_lock_init(&queue->_xmit_lock);
|
|
|
|
netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
|
|
|
|
queue->xmit_lock_owner = -1;
|
2010-12-14 03:09:15 +00:00
|
|
|
netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
|
2010-12-04 02:31:41 +00:00
|
|
|
queue->dev = dev;
|
2011-11-28 16:33:09 +00:00
|
|
|
#ifdef CONFIG_BQL
|
|
|
|
dql_init(&queue->dql, HZ);
|
|
|
|
#endif
|
2010-12-04 02:31:41 +00:00
|
|
|
}
|
|
|
|
|
2010-10-18 18:04:39 +00:00
|
|
|
static int netif_alloc_netdev_queues(struct net_device *dev)
|
|
|
|
{
|
|
|
|
unsigned int count = dev->num_tx_queues;
|
|
|
|
struct netdev_queue *tx;
|
|
|
|
|
|
|
|
BUG_ON(count < 1);
|
|
|
|
|
|
|
|
tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
|
2013-02-04 16:48:16 +00:00
|
|
|
if (!tx)
|
2010-10-18 18:04:39 +00:00
|
|
|
return -ENOMEM;
|
2013-02-04 16:48:16 +00:00
|
|
|
|
2010-10-18 18:04:39 +00:00
|
|
|
dev->_tx = tx;
|
2010-11-21 13:17:27 +00:00
|
|
|
|
2010-10-18 18:04:39 +00:00
|
|
|
netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
|
|
|
|
spin_lock_init(&dev->tx_global_lock);
|
2010-12-04 02:31:41 +00:00
|
|
|
|
|
|
|
return 0;
|
2010-10-18 18:04:39 +00:00
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* register_netdevice - register a network device
|
|
|
|
* @dev: device to register
|
|
|
|
*
|
|
|
|
* Take a completed network device structure and add it to the kernel
|
|
|
|
* interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
|
|
|
|
* chain. 0 is returned on success. A negative errno code is returned
|
|
|
|
* on a failure to set up the device, or if the name is a duplicate.
|
|
|
|
*
|
|
|
|
* Callers must hold the rtnl semaphore. You may want
|
|
|
|
* register_netdev() instead of this.
|
|
|
|
*
|
|
|
|
* BUGS:
|
|
|
|
* The locking appears insufficient to guarantee two parallel registers
|
|
|
|
* will not get the same name.
|
|
|
|
*/
|
|
|
|
|
|
|
|
int register_netdevice(struct net_device *dev)
|
|
|
|
{
|
|
|
|
int ret;
|
2008-11-20 05:32:24 +00:00
|
|
|
struct net *net = dev_net(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
BUG_ON(dev_boot_phase);
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
2006-05-10 20:21:17 +00:00
|
|
|
might_sleep();
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* When net_device's are persistent, this will be fatal. */
|
|
|
|
BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
|
2008-11-20 05:32:24 +00:00
|
|
|
BUG_ON(!net);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-07-15 07:08:33 +00:00
|
|
|
spin_lock_init(&dev->addr_list_lock);
|
2008-07-22 21:16:42 +00:00
|
|
|
netdev_set_addr_lockdep_class(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
dev->iflink = -1;
|
|
|
|
|
2012-09-13 20:58:27 +00:00
|
|
|
ret = dev_get_valid_name(net, dev, dev->name);
|
2011-05-12 15:46:56 +00:00
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Init, if this function is available */
|
2008-11-20 05:32:24 +00:00
|
|
|
if (dev->netdev_ops->ndo_init) {
|
|
|
|
ret = dev->netdev_ops->ndo_init(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (ret) {
|
|
|
|
if (ret > 0)
|
|
|
|
ret = -EIO;
|
2006-11-14 00:02:22 +00:00
|
|
|
goto out;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
2007-02-09 14:24:36 +00:00
|
|
|
|
2013-04-19 02:04:27 +00:00
|
|
|
if (((dev->hw_features | dev->features) &
|
|
|
|
NETIF_F_HW_VLAN_CTAG_FILTER) &&
|
2013-01-29 15:14:16 +00:00
|
|
|
(!dev->netdev_ops->ndo_vlan_rx_add_vid ||
|
|
|
|
!dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
|
|
|
|
netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto err_uninit;
|
|
|
|
}
|
|
|
|
|
2012-08-08 21:52:46 +00:00
|
|
|
ret = -EBUSY;
|
|
|
|
if (!dev->ifindex)
|
|
|
|
dev->ifindex = dev_new_index(net);
|
|
|
|
else if (__dev_get_by_index(net, dev->ifindex))
|
|
|
|
goto err_uninit;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
if (dev->iflink == -1)
|
|
|
|
dev->iflink = dev->ifindex;
|
|
|
|
|
2011-02-15 16:59:17 +00:00
|
|
|
/* Transfer changeable features to wanted_features and enable
|
|
|
|
* software offloads (GSO and GRO).
|
|
|
|
*/
|
|
|
|
dev->hw_features |= NETIF_F_SOFT_FEATURES;
|
2011-02-22 16:52:28 +00:00
|
|
|
dev->features |= NETIF_F_SOFT_FEATURES;
|
|
|
|
dev->wanted_features = dev->features & dev->hw_features;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2011-04-05 05:30:30 +00:00
|
|
|
/* Turn on no cache copy if HW is doing checksum */
|
2011-11-15 15:29:55 +00:00
|
|
|
if (!(dev->flags & IFF_LOOPBACK)) {
|
|
|
|
dev->hw_features |= NETIF_F_NOCACHE_COPY;
|
|
|
|
if (dev->features & NETIF_F_ALL_CSUM) {
|
|
|
|
dev->wanted_features |= NETIF_F_NOCACHE_COPY;
|
|
|
|
dev->features |= NETIF_F_NOCACHE_COPY;
|
|
|
|
}
|
2011-04-05 05:30:30 +00:00
|
|
|
}
|
|
|
|
|
2011-07-14 21:41:11 +00:00
|
|
|
/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
|
2010-09-15 09:24:24 +00:00
|
|
|
*/
|
2011-07-14 21:41:11 +00:00
|
|
|
dev->vlan_features |= NETIF_F_HIGHDMA;
|
2010-09-15 09:24:24 +00:00
|
|
|
|
2013-03-07 09:28:08 +00:00
|
|
|
/* Make NETIF_F_SG inheritable to tunnel devices.
|
|
|
|
*/
|
|
|
|
dev->hw_enc_features |= NETIF_F_SG;
|
|
|
|
|
2013-05-23 21:02:52 +00:00
|
|
|
/* Make NETIF_F_SG inheritable to MPLS.
|
|
|
|
*/
|
|
|
|
dev->mpls_features |= NETIF_F_SG;
|
|
|
|
|
2009-10-02 05:15:27 +00:00
|
|
|
ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
|
|
|
|
ret = notifier_to_errno(ret);
|
|
|
|
if (ret)
|
|
|
|
goto err_uninit;
|
|
|
|
|
2007-09-27 05:02:53 +00:00
|
|
|
ret = netdev_register_kobject(dev);
|
2006-05-10 20:21:17 +00:00
|
|
|
if (ret)
|
2007-07-30 23:29:40 +00:00
|
|
|
goto err_uninit;
|
2006-05-10 20:21:17 +00:00
|
|
|
dev->reg_state = NETREG_REGISTERED;
|
|
|
|
|
2011-04-03 05:48:47 +00:00
|
|
|
__netdev_update_features(dev);
|
2011-02-22 16:52:28 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/*
|
|
|
|
* Default initial state at registry is that the
|
|
|
|
* device is present.
|
|
|
|
*/
|
|
|
|
|
|
|
|
set_bit(__LINK_STATE_PRESENT, &dev->state);
|
|
|
|
|
net: Set device operstate at registration time
The operstate of a device is initially IF_OPER_UNKNOWN and is updated
asynchronously by linkwatch after each change of carrier state
reported by the driver. The default carrier state of a net device is
on, and this will never be changed on drivers that do not support
carrier detection, thus the operstate remains IF_OPER_UNKNOWN.
For devices that do support carrier detection, the driver must set the
carrier state to off initially, then poll the hardware state when the
device is opened. However, we must not activate linkwatch for a
unregistered device, and commit b473001 ('net: Do not fire linkwatch
events until the device is registered.') ensured that we don't. But
this means that the operstate for many devices that support carrier
detection remains IF_OPER_UNKNOWN when it should be IF_OPER_DOWN.
The same issue exists with the dormant state.
The proper initialisation sequence, avoiding a race with opening of
the device, is:
rtnl_lock();
rc = register_netdevice(dev);
if (rc)
goto out_unlock;
netif_carrier_off(dev); /* or netif_dormant_on(dev) */
rtnl_unlock();
but it seems silly that this should have to be repeated in so many
drivers. Further, the operstate seen immediately after opening the
device may still be IF_OPER_UNKNOWN due to the asynchronous nature of
linkwatch.
Commit 22604c8 ('net: Fix for initial link state in 2.6.28') attempted
to fix this by setting the operstate synchronously, but it was
reverted as it could lead to deadlock.
This initialises the operstate synchronously at registration time
only.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-20 21:16:51 +00:00
|
|
|
linkwatch_init_dev(dev);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
dev_init_scheduler(dev);
|
|
|
|
dev_hold(dev);
|
2007-09-12 11:53:49 +00:00
|
|
|
list_netdevice(dev);
|
2012-07-05 01:23:25 +00:00
|
|
|
add_device_randomness(dev->dev_addr, dev->addr_len);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2013-01-08 01:38:25 +00:00
|
|
|
/* If the device has permanent device address, driver should
|
|
|
|
* set dev_addr and also addr_assign_type should be set to
|
|
|
|
* NET_ADDR_PERM (default value).
|
|
|
|
*/
|
|
|
|
if (dev->addr_assign_type == NET_ADDR_PERM)
|
|
|
|
memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Notify protocols, that a new device appeared. */
|
2007-09-16 22:42:43 +00:00
|
|
|
ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
|
2007-07-31 00:03:38 +00:00
|
|
|
ret = notifier_to_errno(ret);
|
2007-10-30 22:38:18 +00:00
|
|
|
if (ret) {
|
|
|
|
rollback_registered(dev);
|
|
|
|
dev->reg_state = NETREG_UNREGISTERED;
|
|
|
|
}
|
2009-12-12 22:11:15 +00:00
|
|
|
/*
|
|
|
|
* Prevent userspace races by waiting until the network
|
|
|
|
* device is fully setup before sending notifications.
|
|
|
|
*/
|
2010-02-26 06:34:51 +00:00
|
|
|
if (!dev->rtnl_link_ops ||
|
|
|
|
dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
|
|
|
|
rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
out:
|
|
|
|
return ret;
|
2007-07-30 23:29:40 +00:00
|
|
|
|
|
|
|
err_uninit:
|
2008-11-20 05:32:24 +00:00
|
|
|
if (dev->netdev_ops->ndo_uninit)
|
|
|
|
dev->netdev_ops->ndo_uninit(dev);
|
2007-07-30 23:29:40 +00:00
|
|
|
goto out;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(register_netdevice);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2009-01-15 05:05:05 +00:00
|
|
|
/**
|
|
|
|
* init_dummy_netdev - init a dummy network device for NAPI
|
|
|
|
* @dev: device to init
|
|
|
|
*
|
|
|
|
* This takes a network device structure and initialize the minimum
|
|
|
|
* amount of fields so it can be used to schedule NAPI polls without
|
|
|
|
* registering a full blown interface. This is to be used by drivers
|
|
|
|
* that need to tie several hardware interfaces to a single NAPI
|
|
|
|
* poll scheduler due to HW limitations.
|
|
|
|
*/
|
|
|
|
int init_dummy_netdev(struct net_device *dev)
|
|
|
|
{
|
|
|
|
/* Clear everything. Note we don't initialize spinlocks
|
|
|
|
* are they aren't supposed to be taken by any of the
|
|
|
|
* NAPI code and this dummy netdev is supposed to be
|
|
|
|
* only ever used for NAPI polls
|
|
|
|
*/
|
|
|
|
memset(dev, 0, sizeof(struct net_device));
|
|
|
|
|
|
|
|
/* make sure we BUG if trying to hit standard
|
|
|
|
* register/unregister code path
|
|
|
|
*/
|
|
|
|
dev->reg_state = NETREG_DUMMY;
|
|
|
|
|
|
|
|
/* NAPI wants this */
|
|
|
|
INIT_LIST_HEAD(&dev->napi_list);
|
|
|
|
|
|
|
|
/* a dummy interface is started by default */
|
|
|
|
set_bit(__LINK_STATE_PRESENT, &dev->state);
|
|
|
|
set_bit(__LINK_STATE_START, &dev->state);
|
|
|
|
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
/* Note : We dont allocate pcpu_refcnt for dummy devices,
|
|
|
|
* because users of this 'device' dont need to change
|
|
|
|
* its refcount.
|
|
|
|
*/
|
|
|
|
|
2009-01-15 05:05:05 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(init_dummy_netdev);
|
|
|
|
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* register_netdev - register a network device
|
|
|
|
* @dev: device to register
|
|
|
|
*
|
|
|
|
* Take a completed network device structure and add it to the kernel
|
|
|
|
* interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
|
|
|
|
* chain. 0 is returned on success. A negative errno code is returned
|
|
|
|
* on a failure to set up the device, or if the name is a duplicate.
|
|
|
|
*
|
2007-04-21 05:14:10 +00:00
|
|
|
* This is a wrapper around register_netdevice that takes the rtnl semaphore
|
2005-04-16 22:20:36 +00:00
|
|
|
* and expands the device name if you passed a format string to
|
|
|
|
* alloc_netdev.
|
|
|
|
*/
|
|
|
|
int register_netdev(struct net_device *dev)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
|
|
|
|
rtnl_lock();
|
|
|
|
err = register_netdevice(dev);
|
|
|
|
rtnl_unlock();
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(register_netdev);
|
|
|
|
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
int netdev_refcnt_read(const struct net_device *dev)
|
|
|
|
{
|
|
|
|
int i, refcnt = 0;
|
|
|
|
|
|
|
|
for_each_possible_cpu(i)
|
|
|
|
refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
|
|
|
|
return refcnt;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_refcnt_read);
|
|
|
|
|
2012-07-10 10:55:09 +00:00
|
|
|
/**
|
2005-04-16 22:20:36 +00:00
|
|
|
* netdev_wait_allrefs - wait until all references are gone.
|
2012-08-18 14:36:44 +00:00
|
|
|
* @dev: target net_device
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* This is called when unregistering network devices.
|
|
|
|
*
|
|
|
|
* Any protocol or device that holds a reference should register
|
|
|
|
* for netdevice notification, and cleanup and put back the
|
|
|
|
* reference if they receive an UNREGISTER event.
|
|
|
|
* We can get stuck here if buggy protocols don't correctly
|
2007-02-09 14:24:36 +00:00
|
|
|
* call dev_put.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
static void netdev_wait_allrefs(struct net_device *dev)
|
|
|
|
{
|
|
|
|
unsigned long rebroadcast_time, warning_time;
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
int refcnt;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
linkwatch: linkwatch_forget_dev() to speedup device dismantle
Herbert Xu a écrit :
> On Tue, Nov 17, 2009 at 04:26:04AM -0800, David Miller wrote:
>> Really, the link watch stuff is just due for a redesign. I don't
>> think a simple hack is going to cut it this time, sorry Eric :-)
>
> I have no objections against any redesigns, but since the only
> caller of linkwatch_forget_dev runs in process context with the
> RTNL, it could also legally emit those events.
Thanks guys, here an updated version then, before linkwatch surgery ?
In this version, I force the event to be sent synchronously.
[PATCH net-next-2.6] linkwatch: linkwatch_forget_dev() to speedup device dismantle
time ip link del eth3.103 ; time ip link del eth3.104 ; time ip link del eth3.105
real 0m0.266s
user 0m0.000s
sys 0m0.001s
real 0m0.770s
user 0m0.000s
sys 0m0.000s
real 0m1.022s
user 0m0.000s
sys 0m0.000s
One problem of current schem in vlan dismantle phase is the
holding of device done by following chain :
vlan_dev_stop() ->
netif_carrier_off(dev) ->
linkwatch_fire_event(dev) ->
dev_hold() ...
And __linkwatch_run_queue() runs up to one second later...
A generic fix to this problem is to add a linkwatch_forget_dev() method
to unlink the device from the list of watched devices.
dev->link_watch_next becomes dev->link_watch_list (and use a bit more memory),
to be able to unlink device in O(1).
After patch :
time ip link del eth3.103 ; time ip link del eth3.104 ; time ip link del eth3.105
real 0m0.024s
user 0m0.000s
sys 0m0.000s
real 0m0.032s
user 0m0.000s
sys 0m0.001s
real 0m0.033s
user 0m0.000s
sys 0m0.000s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-11-17 05:59:21 +00:00
|
|
|
linkwatch_forget_dev(dev);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
rebroadcast_time = warning_time = jiffies;
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
refcnt = netdev_refcnt_read(dev);
|
|
|
|
|
|
|
|
while (refcnt != 0) {
|
2005-04-16 22:20:36 +00:00
|
|
|
if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
|
2006-03-21 06:23:58 +00:00
|
|
|
rtnl_lock();
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Rebroadcast unregister notification */
|
2007-09-16 22:42:43 +00:00
|
|
|
call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2012-08-22 21:50:59 +00:00
|
|
|
__rtnl_unlock();
|
2012-08-22 17:19:46 +00:00
|
|
|
rcu_barrier();
|
2012-08-22 21:50:59 +00:00
|
|
|
rtnl_lock();
|
|
|
|
|
2012-08-22 17:19:46 +00:00
|
|
|
call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
|
|
|
|
&dev->state)) {
|
|
|
|
/* We must not have linkwatch events
|
|
|
|
* pending on unregister. If this
|
|
|
|
* happens, we simply run the queue
|
|
|
|
* unscheduled, resulting in a noop
|
|
|
|
* for this device.
|
|
|
|
*/
|
|
|
|
linkwatch_run_queue();
|
|
|
|
}
|
|
|
|
|
2006-03-21 06:23:58 +00:00
|
|
|
__rtnl_unlock();
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
rebroadcast_time = jiffies;
|
|
|
|
}
|
|
|
|
|
|
|
|
msleep(250);
|
|
|
|
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
refcnt = netdev_refcnt_read(dev);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
if (time_after(jiffies, warning_time + 10 * HZ)) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
|
|
|
|
dev->name, refcnt);
|
2005-04-16 22:20:36 +00:00
|
|
|
warning_time = jiffies;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* The sequence is:
|
|
|
|
*
|
|
|
|
* rtnl_lock();
|
|
|
|
* ...
|
|
|
|
* register_netdevice(x1);
|
|
|
|
* register_netdevice(x2);
|
|
|
|
* ...
|
|
|
|
* unregister_netdevice(y1);
|
|
|
|
* unregister_netdevice(y2);
|
|
|
|
* ...
|
|
|
|
* rtnl_unlock();
|
|
|
|
* free_netdev(y1);
|
|
|
|
* free_netdev(y2);
|
|
|
|
*
|
2008-10-07 22:50:03 +00:00
|
|
|
* We are invoked by rtnl_unlock().
|
2005-04-16 22:20:36 +00:00
|
|
|
* This allows us to deal with problems:
|
2006-05-10 20:21:17 +00:00
|
|
|
* 1) We can delete sysfs objects which invoke hotplug
|
2005-04-16 22:20:36 +00:00
|
|
|
* without deadlocking with linkwatch via keventd.
|
|
|
|
* 2) Since we run with the RTNL semaphore not held, we can sleep
|
|
|
|
* safely in order to wait for the netdev refcnt to drop to zero.
|
2008-10-07 22:50:03 +00:00
|
|
|
*
|
|
|
|
* We must not return until all unregister events added during
|
|
|
|
* the interval the lock was held have been completed.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
void netdev_run_todo(void)
|
|
|
|
{
|
2006-06-23 09:05:55 +00:00
|
|
|
struct list_head list;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Snapshot list, allow later requests */
|
2006-06-23 09:05:55 +00:00
|
|
|
list_replace_init(&net_todo_list, &list);
|
2008-10-07 22:50:03 +00:00
|
|
|
|
|
|
|
__rtnl_unlock();
|
2006-06-23 09:05:55 +00:00
|
|
|
|
2012-08-22 17:19:46 +00:00
|
|
|
|
|
|
|
/* Wait for rcu callbacks to finish before next phase */
|
2011-10-13 22:25:23 +00:00
|
|
|
if (!list_empty(&list))
|
|
|
|
rcu_barrier();
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
while (!list_empty(&list)) {
|
|
|
|
struct net_device *dev
|
2010-02-24 14:01:38 +00:00
|
|
|
= list_first_entry(&list, struct net_device, todo_list);
|
2005-04-16 22:20:36 +00:00
|
|
|
list_del(&dev->todo_list);
|
|
|
|
|
2012-08-22 21:50:59 +00:00
|
|
|
rtnl_lock();
|
2012-08-22 17:19:46 +00:00
|
|
|
call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
|
2012-08-22 21:50:59 +00:00
|
|
|
__rtnl_unlock();
|
2012-08-22 17:19:46 +00:00
|
|
|
|
2006-05-10 20:21:17 +00:00
|
|
|
if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_err("network todo '%s' but state %d\n",
|
2006-05-10 20:21:17 +00:00
|
|
|
dev->name, dev->reg_state);
|
|
|
|
dump_stack();
|
|
|
|
continue;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-05-10 20:21:17 +00:00
|
|
|
dev->reg_state = NETREG_UNREGISTERED;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-03-30 20:16:22 +00:00
|
|
|
on_each_cpu(flush_backlog, dev, 1);
|
2008-08-04 04:29:57 +00:00
|
|
|
|
2006-05-10 20:21:17 +00:00
|
|
|
netdev_wait_allrefs(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-05-10 20:21:17 +00:00
|
|
|
/* paranoia */
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
BUG_ON(netdev_refcnt_read(dev));
|
2011-08-11 19:30:52 +00:00
|
|
|
WARN_ON(rcu_access_pointer(dev->ip_ptr));
|
|
|
|
WARN_ON(rcu_access_pointer(dev->ip6_ptr));
|
2008-07-26 04:43:18 +00:00
|
|
|
WARN_ON(dev->dn_ptr);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-05-10 20:21:17 +00:00
|
|
|
if (dev->destructor)
|
|
|
|
dev->destructor(dev);
|
2007-05-19 22:39:25 +00:00
|
|
|
|
|
|
|
/* Free network device */
|
|
|
|
kobject_put(&dev->dev.kobj);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-07-09 09:11:52 +00:00
|
|
|
/* Convert net_device_stats to rtnl_link_stats64. They have the same
|
|
|
|
* fields in the same order, with only the type differing.
|
|
|
|
*/
|
2012-03-05 04:50:09 +00:00
|
|
|
void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
|
|
|
|
const struct net_device_stats *netdev_stats)
|
2010-07-09 09:11:52 +00:00
|
|
|
{
|
|
|
|
#if BITS_PER_LONG == 64
|
2012-03-05 04:50:09 +00:00
|
|
|
BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
|
|
|
|
memcpy(stats64, netdev_stats, sizeof(*stats64));
|
2010-07-09 09:11:52 +00:00
|
|
|
#else
|
|
|
|
size_t i, n = sizeof(*stats64) / sizeof(u64);
|
|
|
|
const unsigned long *src = (const unsigned long *)netdev_stats;
|
|
|
|
u64 *dst = (u64 *)stats64;
|
|
|
|
|
|
|
|
BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
|
|
|
|
sizeof(*stats64) / sizeof(u64));
|
|
|
|
for (i = 0; i < n; i++)
|
|
|
|
dst[i] = src[i];
|
|
|
|
#endif
|
|
|
|
}
|
2012-03-05 04:50:09 +00:00
|
|
|
EXPORT_SYMBOL(netdev_stats_to_stats64);
|
2010-07-09 09:11:52 +00:00
|
|
|
|
2008-11-20 05:40:23 +00:00
|
|
|
/**
|
|
|
|
* dev_get_stats - get network device statistics
|
|
|
|
* @dev: device to get statistics from
|
2010-07-07 21:58:56 +00:00
|
|
|
* @storage: place to store stats
|
2008-11-20 05:40:23 +00:00
|
|
|
*
|
2010-07-09 09:12:41 +00:00
|
|
|
* Get network statistics from device. Return @storage.
|
|
|
|
* The device driver may provide its own method by setting
|
|
|
|
* dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
|
|
|
|
* otherwise the internal statistics structure is used.
|
2008-11-20 05:40:23 +00:00
|
|
|
*/
|
2010-07-09 09:12:41 +00:00
|
|
|
struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
|
|
|
|
struct rtnl_link_stats64 *storage)
|
2009-05-18 00:34:33 +00:00
|
|
|
{
|
2008-11-20 05:40:23 +00:00
|
|
|
const struct net_device_ops *ops = dev->netdev_ops;
|
|
|
|
|
2010-07-07 21:58:56 +00:00
|
|
|
if (ops->ndo_get_stats64) {
|
|
|
|
memset(storage, 0, sizeof(*storage));
|
2010-09-30 21:06:55 +00:00
|
|
|
ops->ndo_get_stats64(dev, storage);
|
|
|
|
} else if (ops->ndo_get_stats) {
|
2010-07-09 09:11:52 +00:00
|
|
|
netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
|
2010-09-30 21:06:55 +00:00
|
|
|
} else {
|
|
|
|
netdev_stats_to_stats64(storage, &dev->stats);
|
2010-07-07 21:58:56 +00:00
|
|
|
}
|
2010-09-30 21:06:55 +00:00
|
|
|
storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
|
2010-07-07 21:58:56 +00:00
|
|
|
return storage;
|
2007-03-28 21:29:08 +00:00
|
|
|
}
|
2008-11-20 05:40:23 +00:00
|
|
|
EXPORT_SYMBOL(dev_get_stats);
|
2007-03-28 21:29:08 +00:00
|
|
|
|
2010-10-02 06:11:55 +00:00
|
|
|
struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
|
2008-07-09 00:18:23 +00:00
|
|
|
{
|
2010-10-02 06:11:55 +00:00
|
|
|
struct netdev_queue *queue = dev_ingress_queue(dev);
|
2008-07-09 00:18:23 +00:00
|
|
|
|
2010-10-02 06:11:55 +00:00
|
|
|
#ifdef CONFIG_NET_CLS_ACT
|
|
|
|
if (queue)
|
|
|
|
return queue;
|
|
|
|
queue = kzalloc(sizeof(*queue), GFP_KERNEL);
|
|
|
|
if (!queue)
|
|
|
|
return NULL;
|
|
|
|
netdev_init_one_queue(dev, queue, NULL);
|
|
|
|
queue->qdisc = &noop_qdisc;
|
|
|
|
queue->qdisc_sleeping = &noop_qdisc;
|
|
|
|
rcu_assign_pointer(dev->ingress_queue, queue);
|
|
|
|
#endif
|
|
|
|
return queue;
|
2008-07-08 23:55:56 +00:00
|
|
|
}
|
|
|
|
|
2012-09-16 09:17:26 +00:00
|
|
|
static const struct ethtool_ops default_ethtool_ops;
|
|
|
|
|
2013-01-10 23:19:10 +00:00
|
|
|
void netdev_set_default_ethtool_ops(struct net_device *dev,
|
|
|
|
const struct ethtool_ops *ops)
|
|
|
|
{
|
|
|
|
if (dev->ethtool_ops == &default_ethtool_ops)
|
|
|
|
dev->ethtool_ops = ops;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
2011-01-09 19:36:31 +00:00
|
|
|
* alloc_netdev_mqs - allocate network device
|
2005-04-16 22:20:36 +00:00
|
|
|
* @sizeof_priv: size of private data to allocate space for
|
|
|
|
* @name: device name format string
|
|
|
|
* @setup: callback to initialize device
|
2011-01-09 19:36:31 +00:00
|
|
|
* @txqs: the number of TX subqueues to allocate
|
|
|
|
* @rxqs: the number of RX subqueues to allocate
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* Allocates a struct net_device with private data area for driver use
|
2007-07-06 20:36:20 +00:00
|
|
|
* and performs basic initialization. Also allocates subquue structs
|
2011-01-09 19:36:31 +00:00
|
|
|
* for each queue on the device.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
2011-01-09 19:36:31 +00:00
|
|
|
struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
|
|
|
|
void (*setup)(struct net_device *),
|
|
|
|
unsigned int txqs, unsigned int rxqs)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev;
|
2008-07-21 20:28:44 +00:00
|
|
|
size_t alloc_size;
|
2009-05-27 04:42:37 +00:00
|
|
|
struct net_device *p;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-08-30 00:06:13 +00:00
|
|
|
BUG_ON(strlen(name) >= sizeof(dev->name));
|
|
|
|
|
2011-01-09 19:36:31 +00:00
|
|
|
if (txqs < 1) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
|
2010-10-18 17:55:58 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2011-01-09 19:36:31 +00:00
|
|
|
#ifdef CONFIG_RPS
|
|
|
|
if (rxqs < 1) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
|
2011-01-09 19:36:31 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2008-07-17 08:56:23 +00:00
|
|
|
alloc_size = sizeof(struct net_device);
|
2008-04-18 22:43:32 +00:00
|
|
|
if (sizeof_priv) {
|
|
|
|
/* ensure 32-byte alignment of private area */
|
2009-05-27 04:42:37 +00:00
|
|
|
alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
|
2008-04-18 22:43:32 +00:00
|
|
|
alloc_size += sizeof_priv;
|
|
|
|
}
|
|
|
|
/* ensure 32-byte alignment of whole construct */
|
2009-05-27 04:42:37 +00:00
|
|
|
alloc_size += NETDEV_ALIGN - 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-04-07 05:38:28 +00:00
|
|
|
p = kzalloc(alloc_size, GFP_KERNEL);
|
2013-02-04 16:48:16 +00:00
|
|
|
if (!p)
|
2005-04-16 22:20:36 +00:00
|
|
|
return NULL;
|
|
|
|
|
2009-05-27 04:42:37 +00:00
|
|
|
dev = PTR_ALIGN(p, NETDEV_ALIGN);
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->padded = (char *)dev - (char *)p;
|
2009-05-08 13:30:17 +00:00
|
|
|
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
dev->pcpu_refcnt = alloc_percpu(int);
|
|
|
|
if (!dev->pcpu_refcnt)
|
2010-10-18 18:04:39 +00:00
|
|
|
goto free_p;
|
2009-05-08 13:30:17 +00:00
|
|
|
|
|
|
|
if (dev_addr_init(dev))
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
goto free_pcpu;
|
2009-05-08 13:30:17 +00:00
|
|
|
|
2010-04-01 21:22:57 +00:00
|
|
|
dev_mc_init(dev);
|
2010-04-01 21:22:09 +00:00
|
|
|
dev_uc_init(dev);
|
2009-05-22 23:22:17 +00:00
|
|
|
|
2008-03-25 12:47:49 +00:00
|
|
|
dev_net_set(dev, &init_net);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2011-02-08 23:02:50 +00:00
|
|
|
dev->gso_max_size = GSO_MAX_SIZE;
|
2012-07-30 15:57:00 +00:00
|
|
|
dev->gso_max_segs = GSO_MAX_SEGS;
|
2011-02-08 23:02:50 +00:00
|
|
|
|
|
|
|
INIT_LIST_HEAD(&dev->napi_list);
|
|
|
|
INIT_LIST_HEAD(&dev->unreg_list);
|
|
|
|
INIT_LIST_HEAD(&dev->link_watch_list);
|
2013-01-03 22:48:49 +00:00
|
|
|
INIT_LIST_HEAD(&dev->upper_dev_list);
|
2011-02-08 23:02:50 +00:00
|
|
|
dev->priv_flags = IFF_XMIT_DST_RELEASE;
|
|
|
|
setup(dev);
|
|
|
|
|
2011-01-09 19:36:31 +00:00
|
|
|
dev->num_tx_queues = txqs;
|
|
|
|
dev->real_num_tx_queues = txqs;
|
2010-11-09 10:47:30 +00:00
|
|
|
if (netif_alloc_netdev_queues(dev))
|
2011-02-08 23:02:50 +00:00
|
|
|
goto free_all;
|
2008-07-17 07:34:19 +00:00
|
|
|
|
2010-03-24 19:13:54 +00:00
|
|
|
#ifdef CONFIG_RPS
|
2011-01-09 19:36:31 +00:00
|
|
|
dev->num_rx_queues = rxqs;
|
|
|
|
dev->real_num_rx_queues = rxqs;
|
2010-11-09 10:47:38 +00:00
|
|
|
if (netif_alloc_rx_queues(dev))
|
2011-02-08 23:02:50 +00:00
|
|
|
goto free_all;
|
2010-03-24 19:13:54 +00:00
|
|
|
#endif
|
2010-03-16 08:03:29 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
strcpy(dev->name, name);
|
2011-01-13 23:38:30 +00:00
|
|
|
dev->group = INIT_NETDEV_GROUP;
|
2012-09-16 09:17:26 +00:00
|
|
|
if (!dev->ethtool_ops)
|
|
|
|
dev->ethtool_ops = &default_ethtool_ops;
|
2005-04-16 22:20:36 +00:00
|
|
|
return dev;
|
2009-05-08 13:30:17 +00:00
|
|
|
|
2011-02-08 23:02:50 +00:00
|
|
|
free_all:
|
|
|
|
free_netdev(dev);
|
|
|
|
return NULL;
|
|
|
|
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
free_pcpu:
|
|
|
|
free_percpu(dev->pcpu_refcnt);
|
2010-11-09 10:47:30 +00:00
|
|
|
kfree(dev->_tx);
|
2010-11-09 10:47:38 +00:00
|
|
|
#ifdef CONFIG_RPS
|
|
|
|
kfree(dev->_rx);
|
|
|
|
#endif
|
|
|
|
|
2009-05-08 13:30:17 +00:00
|
|
|
free_p:
|
|
|
|
kfree(p);
|
|
|
|
return NULL;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2011-01-09 19:36:31 +00:00
|
|
|
EXPORT_SYMBOL(alloc_netdev_mqs);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* free_netdev - free network device
|
|
|
|
* @dev: device
|
|
|
|
*
|
2007-02-09 14:24:36 +00:00
|
|
|
* This function does the last stage of destroying an allocated device
|
|
|
|
* interface. The reference to the device object is released.
|
2005-04-16 22:20:36 +00:00
|
|
|
* If this is the last reference then it will be freed.
|
|
|
|
*/
|
|
|
|
void free_netdev(struct net_device *dev)
|
|
|
|
{
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
struct napi_struct *p, *n;
|
|
|
|
|
2008-04-16 09:02:18 +00:00
|
|
|
release_net(dev_net(dev));
|
|
|
|
|
2008-07-17 07:34:19 +00:00
|
|
|
kfree(dev->_tx);
|
2010-11-09 10:47:38 +00:00
|
|
|
#ifdef CONFIG_RPS
|
|
|
|
kfree(dev->_rx);
|
|
|
|
#endif
|
2008-07-17 07:34:19 +00:00
|
|
|
|
2011-08-11 19:30:52 +00:00
|
|
|
kfree(rcu_dereference_protected(dev->ingress_queue, 1));
|
2010-10-02 06:11:55 +00:00
|
|
|
|
2009-05-05 02:48:28 +00:00
|
|
|
/* Flush device addresses */
|
|
|
|
dev_addr_flush(dev);
|
|
|
|
|
net: Add Generic Receive Offload infrastructure
This patch adds the top-level GRO (Generic Receive Offload) infrastructure.
This is pretty similar to LRO except that this is protocol-independent.
Instead of holding packets in an lro_mgr structure, they're now held in
napi_struct.
For drivers that intend to use this, they can set the NETIF_F_GRO bit and
call napi_gro_receive instead of netif_receive_skb or just call netif_rx.
The latter will call napi_receive_skb automatically. When napi_gro_receive
is used, the driver must either call napi_complete/napi_rx_complete, or
call napi_gro_flush in softirq context if the driver uses the primitives
__napi_complete/__napi_rx_complete.
Protocols will set the gro_receive and gro_complete function pointers in
order to participate in this scheme.
In addition to the packet, gro_receive will get a list of currently held
packets. Each packet in the list has a same_flow field which is non-zero
if it is a potential match for the new packet. For each packet that may
match, they also have a flush field which is non-zero if the held packet
must not be merged with the new packet.
Once gro_receive has determined that the new skb matches a held packet,
the held packet may be processed immediately if the new skb cannot be
merged with it. In this case gro_receive should return the pointer to
the existing skb in gro_list. Otherwise the new skb should be merged into
the existing packet and NULL should be returned, unless the new skb makes
it impossible for any further merges to be made (e.g., FIN packet) where
the merged skb should be returned.
Whenever the skb is merged into an existing entry, the gro_receive
function should set NAPI_GRO_CB(skb)->same_flow. Note that if an skb
merely matches an existing entry but can't be merged with it, then
this shouldn't be set.
If gro_receive finds it pointless to hold the new skb for future merging,
it should set NAPI_GRO_CB(skb)->flush.
Held packets will be flushed by napi_gro_flush which is called by
napi_complete and napi_rx_complete.
Currently held packets are stored in a singly liked list just like LRO.
The list is limited to a maximum of 8 entries. In future, this may be
expanded to use a hash table to allow more flows to be held for merging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-12-16 07:38:52 +00:00
|
|
|
list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
|
|
|
|
netif_napi_del(p);
|
|
|
|
|
net: percpu net_device refcount
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-10-11 10:22:12 +00:00
|
|
|
free_percpu(dev->pcpu_refcnt);
|
|
|
|
dev->pcpu_refcnt = NULL;
|
|
|
|
|
2006-05-26 20:25:24 +00:00
|
|
|
/* Compatibility with error handling in drivers */
|
2005-04-16 22:20:36 +00:00
|
|
|
if (dev->reg_state == NETREG_UNINITIALIZED) {
|
|
|
|
kfree((char *)dev - dev->padded);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
|
|
|
|
dev->reg_state = NETREG_RELEASED;
|
|
|
|
|
2002-04-09 19:14:34 +00:00
|
|
|
/* will free via device release */
|
|
|
|
put_device(&dev->dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(free_netdev);
|
2007-02-09 14:24:36 +00:00
|
|
|
|
2008-09-30 09:23:58 +00:00
|
|
|
/**
|
|
|
|
* synchronize_net - Synchronize with packet receive processing
|
|
|
|
*
|
|
|
|
* Wait for packets currently being received to be done.
|
|
|
|
* Does not block later packets from starting.
|
|
|
|
*/
|
2007-02-09 14:24:36 +00:00
|
|
|
void synchronize_net(void)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
might_sleep();
|
2011-05-23 23:07:32 +00:00
|
|
|
if (rtnl_is_locked())
|
|
|
|
synchronize_rcu_expedited();
|
|
|
|
else
|
|
|
|
synchronize_rcu();
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-09-03 08:29:39 +00:00
|
|
|
EXPORT_SYMBOL(synchronize_net);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/**
|
2009-10-27 07:03:04 +00:00
|
|
|
* unregister_netdevice_queue - remove device from the kernel
|
2005-04-16 22:20:36 +00:00
|
|
|
* @dev: device
|
2009-10-27 07:03:04 +00:00
|
|
|
* @head: list
|
2009-11-23 04:43:13 +00:00
|
|
|
*
|
2005-04-16 22:20:36 +00:00
|
|
|
* This function shuts down a device interface and removes it
|
2007-12-11 10:28:03 +00:00
|
|
|
* from the kernel tables.
|
2009-10-27 07:03:04 +00:00
|
|
|
* If head not NULL, device is queued to be unregistered later.
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* Callers must hold the rtnl semaphore. You may want
|
|
|
|
* unregister_netdev() instead of this.
|
|
|
|
*/
|
|
|
|
|
2009-10-27 07:03:04 +00:00
|
|
|
void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-12-13 03:21:56 +00:00
|
|
|
ASSERT_RTNL();
|
|
|
|
|
2009-10-27 07:03:04 +00:00
|
|
|
if (head) {
|
2009-10-30 14:51:13 +00:00
|
|
|
list_move_tail(&dev->unreg_list, head);
|
2009-10-27 07:03:04 +00:00
|
|
|
} else {
|
|
|
|
rollback_registered(dev);
|
|
|
|
/* Finish processing unregister after unlock */
|
|
|
|
net_set_todo(dev);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2009-10-27 07:03:04 +00:00
|
|
|
EXPORT_SYMBOL(unregister_netdevice_queue);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2009-10-27 07:04:19 +00:00
|
|
|
/**
|
|
|
|
* unregister_netdevice_many - unregister many devices
|
|
|
|
* @head: list of devices
|
|
|
|
*/
|
|
|
|
void unregister_netdevice_many(struct list_head *head)
|
|
|
|
{
|
|
|
|
struct net_device *dev;
|
|
|
|
|
|
|
|
if (!list_empty(head)) {
|
|
|
|
rollback_registered_many(head);
|
|
|
|
list_for_each_entry(dev, head, unreg_list)
|
|
|
|
net_set_todo(dev);
|
|
|
|
}
|
|
|
|
}
|
2009-10-27 07:06:49 +00:00
|
|
|
EXPORT_SYMBOL(unregister_netdevice_many);
|
2009-10-27 07:04:19 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* unregister_netdev - remove device from the kernel
|
|
|
|
* @dev: device
|
|
|
|
*
|
|
|
|
* This function shuts down a device interface and removes it
|
2007-12-11 10:28:03 +00:00
|
|
|
* from the kernel tables.
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* This is just a wrapper for unregister_netdevice that takes
|
|
|
|
* the rtnl semaphore. In general you want to use this and not
|
|
|
|
* unregister_netdevice.
|
|
|
|
*/
|
|
|
|
void unregister_netdev(struct net_device *dev)
|
|
|
|
{
|
|
|
|
rtnl_lock();
|
|
|
|
unregister_netdevice(dev);
|
|
|
|
rtnl_unlock();
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(unregister_netdev);
|
|
|
|
|
2007-09-12 11:53:49 +00:00
|
|
|
/**
|
|
|
|
* dev_change_net_namespace - move device to different nethost namespace
|
|
|
|
* @dev: device
|
|
|
|
* @net: network namespace
|
|
|
|
* @pat: If not NULL name pattern to try if the current device name
|
|
|
|
* is already taken in the destination network namespace.
|
|
|
|
*
|
|
|
|
* This function shuts down a device interface and moves it
|
|
|
|
* to a new network namespace. On success 0 is returned, on
|
|
|
|
* a failure a netagive errno code is returned.
|
|
|
|
*
|
|
|
|
* Callers must hold the rtnl semaphore.
|
|
|
|
*/
|
|
|
|
|
|
|
|
int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
/* Don't allow namespace local devices to be moved. */
|
|
|
|
err = -EINVAL;
|
|
|
|
if (dev->features & NETIF_F_NETNS_LOCAL)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* Ensure the device has been registrered */
|
|
|
|
if (dev->reg_state != NETREG_REGISTERED)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* Get out if there is nothing todo */
|
|
|
|
err = 0;
|
2008-03-25 18:57:35 +00:00
|
|
|
if (net_eq(dev_net(dev), net))
|
2007-09-12 11:53:49 +00:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* Pick the destination device name, and ensure
|
|
|
|
* we can use it in the destination network namespace.
|
|
|
|
*/
|
|
|
|
err = -EEXIST;
|
2009-11-18 02:36:59 +00:00
|
|
|
if (__dev_get_by_name(net, dev->name)) {
|
2007-09-12 11:53:49 +00:00
|
|
|
/* We get here if we can't use the current device name */
|
|
|
|
if (!pat)
|
|
|
|
goto out;
|
2012-09-13 20:58:27 +00:00
|
|
|
if (dev_get_valid_name(net, dev, pat) < 0)
|
2007-09-12 11:53:49 +00:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* And now a mini version of register_netdevice unregister_netdevice.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* If device is running close it first. */
|
2007-10-10 09:49:09 +00:00
|
|
|
dev_close(dev);
|
2007-09-12 11:53:49 +00:00
|
|
|
|
|
|
|
/* And unlink it from device chain */
|
|
|
|
err = -ENODEV;
|
|
|
|
unlist_netdevice(dev);
|
|
|
|
|
|
|
|
synchronize_net();
|
|
|
|
|
|
|
|
/* Shutdown queueing discipline. */
|
|
|
|
dev_shutdown(dev);
|
|
|
|
|
|
|
|
/* Notify protocols, that we are about to destroy
|
|
|
|
this device. They should clean all the things.
|
2010-09-17 03:22:19 +00:00
|
|
|
|
|
|
|
Note that dev->reg_state stays at NETREG_REGISTERED.
|
|
|
|
This is wanted because this way 8021q and macvlan know
|
|
|
|
the device is just moving and can keep their slaves up.
|
2007-09-12 11:53:49 +00:00
|
|
|
*/
|
|
|
|
call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
|
net: dev: fix the incorrect hold of net namespace's lo device
When moving a net device from one net namespace to another
net namespace,dev_change_net_namespace calls NETDEV_DOWN
event,so the original net namespace's dst entries which
beloned to this net device will be put into dst_garbage
list.
then dev_change_net_namespace will set this net device's
net to the new net namespace.
If we unregister this net device's driver, this will trigger
the NETDEV_UNREGISTER_FINAL event, dst_ifdown will be called,
and get this net device's dst entries from dst_garbage list,
put these entries' dev to the new net namespace's lo device.
It's not what we want,actually we need these dst entries hold
the original net namespace's lo device,this incorrect device
holding will trigger emg message like below.
unregister_netdevice: waiting for lo to become free. Usage count = 1
so we should call NETDEV_UNREGISTER_FINAL event in
dev_change_net_namespace too,in order to make sure dst entries
already in the dst_garbage list, we need rcu_barrier before we
call NETDEV_UNREGISTER_FINAL event.
With help form Eric Dumazet.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-23 15:36:55 +00:00
|
|
|
rcu_barrier();
|
|
|
|
call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
|
2011-10-21 06:24:20 +00:00
|
|
|
rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
|
2007-09-12 11:53:49 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Flush the unicast and multicast chains
|
|
|
|
*/
|
2010-04-01 21:22:09 +00:00
|
|
|
dev_uc_flush(dev);
|
2010-04-01 21:22:57 +00:00
|
|
|
dev_mc_flush(dev);
|
2007-09-12 11:53:49 +00:00
|
|
|
|
net: dev_change_net_namespace: send a KOBJ_REMOVED/KOBJ_ADD
When a new nic is created in namespace ns1, the kernel sends a KOBJ_ADD uevent
to ns1. When the nic is moved to ns2, we only send a KOBJ_MOVE to ns2, and
nothing to ns1.
This patch changes that behavior so that when moving a nic from ns1 to ns2, we
send a KOBJ_REMOVED to ns1 and KOBJ_ADD to ns2. (The KOBJ_MOVE is still
sent to ns2).
The effects of this can be seen when starting and stopping containers in
an upstart based host. Lxc will create a pair of veth nics, the kernel
sends KOBJ_ADD, and upstart starts network-instance jobs for each. When
one nic is moved to the container, because no KOBJ_REMOVED event is
received, the network-instance job for that veth never goes away. This
was reported at https://bugs.launchpad.net/ubuntu/+source/lxc/+bug/1065589
With this patch the networ-instance jobs properly go away.
The other oddness solved here is that if a nic is passed into a running
upstart-based container, without this patch no network-instance job is
started in the container. But when the container creates a new nic
itself (ip link add new type veth) then network-interface jobs are
created. With this patch, behavior comes in line with a regular host.
v2: also send KOBJ_ADD to new netns. There will then be a
_MOVE event from the device_rename() call, but that should
be innocuous.
Signed-off-by: Serge Hallyn <serge.hallyn@canonical.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Daniel Lezcano <daniel.lezcano@free.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-12-03 16:17:12 +00:00
|
|
|
/* Send a netdev-removed uevent to the old namespace */
|
|
|
|
kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
|
|
|
|
|
2007-09-12 11:53:49 +00:00
|
|
|
/* Actually switch the network namespace */
|
2008-03-25 12:47:49 +00:00
|
|
|
dev_net_set(dev, net);
|
2007-09-12 11:53:49 +00:00
|
|
|
|
|
|
|
/* If there is an ifindex conflict assign a new one */
|
|
|
|
if (__dev_get_by_index(net, dev->ifindex)) {
|
|
|
|
int iflink = (dev->iflink == dev->ifindex);
|
|
|
|
dev->ifindex = dev_new_index(net);
|
|
|
|
if (iflink)
|
|
|
|
dev->iflink = dev->ifindex;
|
|
|
|
}
|
|
|
|
|
net: dev_change_net_namespace: send a KOBJ_REMOVED/KOBJ_ADD
When a new nic is created in namespace ns1, the kernel sends a KOBJ_ADD uevent
to ns1. When the nic is moved to ns2, we only send a KOBJ_MOVE to ns2, and
nothing to ns1.
This patch changes that behavior so that when moving a nic from ns1 to ns2, we
send a KOBJ_REMOVED to ns1 and KOBJ_ADD to ns2. (The KOBJ_MOVE is still
sent to ns2).
The effects of this can be seen when starting and stopping containers in
an upstart based host. Lxc will create a pair of veth nics, the kernel
sends KOBJ_ADD, and upstart starts network-instance jobs for each. When
one nic is moved to the container, because no KOBJ_REMOVED event is
received, the network-instance job for that veth never goes away. This
was reported at https://bugs.launchpad.net/ubuntu/+source/lxc/+bug/1065589
With this patch the networ-instance jobs properly go away.
The other oddness solved here is that if a nic is passed into a running
upstart-based container, without this patch no network-instance job is
started in the container. But when the container creates a new nic
itself (ip link add new type veth) then network-interface jobs are
created. With this patch, behavior comes in line with a regular host.
v2: also send KOBJ_ADD to new netns. There will then be a
_MOVE event from the device_rename() call, but that should
be innocuous.
Signed-off-by: Serge Hallyn <serge.hallyn@canonical.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Daniel Lezcano <daniel.lezcano@free.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-12-03 16:17:12 +00:00
|
|
|
/* Send a netdev-add uevent to the new namespace */
|
|
|
|
kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
|
|
|
|
|
2007-09-27 05:02:53 +00:00
|
|
|
/* Fixup kobjects */
|
2010-05-05 00:36:49 +00:00
|
|
|
err = device_rename(&dev->dev, dev->name);
|
2007-09-27 05:02:53 +00:00
|
|
|
WARN_ON(err);
|
2007-09-12 11:53:49 +00:00
|
|
|
|
|
|
|
/* Add the device back in the hashes */
|
|
|
|
list_netdevice(dev);
|
|
|
|
|
|
|
|
/* Notify protocols, that a new device appeared. */
|
|
|
|
call_netdevice_notifiers(NETDEV_REGISTER, dev);
|
|
|
|
|
2009-12-12 22:11:15 +00:00
|
|
|
/*
|
|
|
|
* Prevent userspace races by waiting until the network
|
|
|
|
* device is fully setup before sending notifications.
|
|
|
|
*/
|
|
|
|
rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
|
|
|
|
|
2007-09-12 11:53:49 +00:00
|
|
|
synchronize_net();
|
|
|
|
err = 0;
|
|
|
|
out:
|
|
|
|
return err;
|
|
|
|
}
|
2009-07-13 22:33:35 +00:00
|
|
|
EXPORT_SYMBOL_GPL(dev_change_net_namespace);
|
2007-09-12 11:53:49 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static int dev_cpu_callback(struct notifier_block *nfb,
|
|
|
|
unsigned long action,
|
|
|
|
void *ocpu)
|
|
|
|
{
|
|
|
|
struct sk_buff **list_skb;
|
|
|
|
struct sk_buff *skb;
|
|
|
|
unsigned int cpu, oldcpu = (unsigned long)ocpu;
|
|
|
|
struct softnet_data *sd, *oldsd;
|
|
|
|
|
2007-05-09 09:35:10 +00:00
|
|
|
if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
|
2005-04-16 22:20:36 +00:00
|
|
|
return NOTIFY_OK;
|
|
|
|
|
|
|
|
local_irq_disable();
|
|
|
|
cpu = smp_processor_id();
|
|
|
|
sd = &per_cpu(softnet_data, cpu);
|
|
|
|
oldsd = &per_cpu(softnet_data, oldcpu);
|
|
|
|
|
|
|
|
/* Find end of our completion_queue. */
|
|
|
|
list_skb = &sd->completion_queue;
|
|
|
|
while (*list_skb)
|
|
|
|
list_skb = &(*list_skb)->next;
|
|
|
|
/* Append completion queue from offline CPU. */
|
|
|
|
*list_skb = oldsd->completion_queue;
|
|
|
|
oldsd->completion_queue = NULL;
|
|
|
|
|
|
|
|
/* Append output queue from offline CPU. */
|
2010-04-26 23:06:24 +00:00
|
|
|
if (oldsd->output_queue) {
|
|
|
|
*sd->output_queue_tailp = oldsd->output_queue;
|
|
|
|
sd->output_queue_tailp = oldsd->output_queue_tailp;
|
|
|
|
oldsd->output_queue = NULL;
|
|
|
|
oldsd->output_queue_tailp = &oldsd->output_queue;
|
|
|
|
}
|
2011-06-06 20:50:03 +00:00
|
|
|
/* Append NAPI poll list from offline CPU. */
|
|
|
|
if (!list_empty(&oldsd->poll_list)) {
|
|
|
|
list_splice_init(&oldsd->poll_list, &sd->poll_list);
|
|
|
|
raise_softirq_irqoff(NET_RX_SOFTIRQ);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
raise_softirq_irqoff(NET_TX_SOFTIRQ);
|
|
|
|
local_irq_enable();
|
|
|
|
|
|
|
|
/* Process offline CPU's input_pkt_queue */
|
2010-05-20 18:37:59 +00:00
|
|
|
while ((skb = __skb_dequeue(&oldsd->process_queue))) {
|
2005-04-16 22:20:36 +00:00
|
|
|
netif_rx(skb);
|
2010-05-20 18:37:59 +00:00
|
|
|
input_queue_head_incr(oldsd);
|
rfs: Receive Flow Steering
This patch implements receive flow steering (RFS). RFS steers
received packets for layer 3 and 4 processing to the CPU where
the application for the corresponding flow is running. RFS is an
extension of Receive Packet Steering (RPS).
The basic idea of RFS is that when an application calls recvmsg
(or sendmsg) the application's running CPU is stored in a hash
table that is indexed by the connection's rxhash which is stored in
the socket structure. The rxhash is passed in skb's received on
the connection from netif_receive_skb. For each received packet,
the associated rxhash is used to look up the CPU in the hash table,
if a valid CPU is set then the packet is steered to that CPU using
the RPS mechanisms.
The convolution of the simple approach is that it would potentially
allow OOO packets. If threads are thrashing around CPUs or multiple
threads are trying to read from the same sockets, a quickly changing
CPU value in the hash table could cause rampant OOO packets--
we consider this a non-starter.
To avoid OOO packets, this solution implements two types of hash
tables: rps_sock_flow_table and rps_dev_flow_table.
rps_sock_table is a global hash table. Each entry is just a CPU
number and it is populated in recvmsg and sendmsg as described above.
This table contains the "desired" CPUs for flows.
rps_dev_flow_table is specific to each device queue. Each entry
contains a CPU and a tail queue counter. The CPU is the "current"
CPU for a matching flow. The tail queue counter holds the value
of a tail queue counter for the associated CPU's backlog queue at
the time of last enqueue for a flow matching the entry.
Each backlog queue has a queue head counter which is incremented
on dequeue, and so a queue tail counter is computed as queue head
count + queue length. When a packet is enqueued on a backlog queue,
the current value of the queue tail counter is saved in the hash
entry of the rps_dev_flow_table.
And now the trick: when selecting the CPU for RPS (get_rps_cpu)
the rps_sock_flow table and the rps_dev_flow table for the RX queue
are consulted. When the desired CPU for the flow (found in the
rps_sock_flow table) does not match the current CPU (found in the
rps_dev_flow table), the current CPU is changed to the desired CPU
if one of the following is true:
- The current CPU is unset (equal to RPS_NO_CPU)
- Current CPU is offline
- The current CPU's queue head counter >= queue tail counter in the
rps_dev_flow table. This checks if the queue tail has advanced
beyond the last packet that was enqueued using this table entry.
This guarantees that all packets queued using this entry have been
dequeued, thus preserving in order delivery.
Making each queue have its own rps_dev_flow table has two advantages:
1) the tail queue counters will be written on each receive, so
keeping the table local to interrupting CPU s good for locality. 2)
this allows lockless access to the table-- the CPU number and queue
tail counter need to be accessed together under mutual exclusion
from netif_receive_skb, we assume that this is only called from
device napi_poll which is non-reentrant.
This patch implements RFS for TCP and connected UDP sockets.
It should be usable for other flow oriented protocols.
There are two configuration parameters for RFS. The
"rps_flow_entries" kernel init parameter sets the number of
entries in the rps_sock_flow_table, the per rxqueue sysfs entry
"rps_flow_cnt" contains the number of entries in the rps_dev_flow
table for the rxqueue. Both are rounded to power of two.
The obvious benefit of RFS (over just RPS) is that it achieves
CPU locality between the receive processing for a flow and the
applications processing; this can result in increased performance
(higher pps, lower latency).
The benefits of RFS are dependent on cache hierarchy, application
load, and other factors. On simple benchmarks, we don't necessarily
see improvement and sometimes see degradation. However, for more
complex benchmarks and for applications where cache pressure is
much higher this technique seems to perform very well.
Below are some benchmark results which show the potential benfit of
this patch. The netperf test has 500 instances of netperf TCP_RR
test with 1 byte req. and resp. The RPC test is an request/response
test similar in structure to netperf RR test ith 100 threads on
each host, but does more work in userspace that netperf.
e1000e on 8 core Intel
No RFS or RPS 104K tps at 30% CPU
No RFS (best RPS config): 290K tps at 63% CPU
RFS 303K tps at 61% CPU
RPC test tps CPU% 50/90/99% usec latency Latency StdDev
No RFS/RPS 103K 48% 757/900/3185 4472.35
RPS only: 174K 73% 415/993/2468 491.66
RFS 223K 73% 379/651/1382 315.61
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-04-16 23:01:27 +00:00
|
|
|
}
|
2010-05-20 18:37:59 +00:00
|
|
|
while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
|
2010-04-27 22:07:33 +00:00
|
|
|
netif_rx(skb);
|
2010-05-20 18:37:59 +00:00
|
|
|
input_queue_head_incr(oldsd);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
return NOTIFY_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2007-08-10 22:47:58 +00:00
|
|
|
/**
|
2008-10-23 08:11:29 +00:00
|
|
|
* netdev_increment_features - increment feature set by one
|
|
|
|
* @all: current feature set
|
|
|
|
* @one: new feature set
|
|
|
|
* @mask: mask feature set
|
2007-08-10 22:47:58 +00:00
|
|
|
*
|
|
|
|
* Computes a new feature set after adding a device with feature set
|
2008-10-23 08:11:29 +00:00
|
|
|
* @one to the master device with current feature set @all. Will not
|
|
|
|
* enable anything that is off in @mask. Returns the new feature set.
|
2007-08-10 22:47:58 +00:00
|
|
|
*/
|
2011-11-15 15:29:55 +00:00
|
|
|
netdev_features_t netdev_increment_features(netdev_features_t all,
|
|
|
|
netdev_features_t one, netdev_features_t mask)
|
2008-10-23 08:11:29 +00:00
|
|
|
{
|
2011-04-22 06:31:16 +00:00
|
|
|
if (mask & NETIF_F_GEN_CSUM)
|
|
|
|
mask |= NETIF_F_ALL_CSUM;
|
|
|
|
mask |= NETIF_F_VLAN_CHALLENGED;
|
2007-08-10 22:47:58 +00:00
|
|
|
|
2011-04-22 06:31:16 +00:00
|
|
|
all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
|
|
|
|
all &= one | ~NETIF_F_ALL_FOR_ALL;
|
2011-04-05 05:30:30 +00:00
|
|
|
|
2011-04-22 06:31:16 +00:00
|
|
|
/* If one device supports hw checksumming, set for all. */
|
|
|
|
if (all & NETIF_F_GEN_CSUM)
|
|
|
|
all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
|
2007-08-10 22:47:58 +00:00
|
|
|
|
|
|
|
return all;
|
|
|
|
}
|
2008-10-23 08:11:29 +00:00
|
|
|
EXPORT_SYMBOL(netdev_increment_features);
|
2007-08-10 22:47:58 +00:00
|
|
|
|
2007-09-16 22:40:33 +00:00
|
|
|
static struct hlist_head *netdev_create_hash(void)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
struct hlist_head *hash;
|
|
|
|
|
|
|
|
hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
|
|
|
|
if (hash != NULL)
|
|
|
|
for (i = 0; i < NETDEV_HASHENTRIES; i++)
|
|
|
|
INIT_HLIST_HEAD(&hash[i]);
|
|
|
|
|
|
|
|
return hash;
|
|
|
|
}
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
/* Initialize per network namespace state */
|
2007-10-09 03:38:39 +00:00
|
|
|
static int __net_init netdev_init(struct net *net)
|
2007-09-17 18:56:21 +00:00
|
|
|
{
|
2012-07-18 09:06:07 +00:00
|
|
|
if (net != &init_net)
|
|
|
|
INIT_LIST_HEAD(&net->dev_base_head);
|
2007-09-17 18:56:21 +00:00
|
|
|
|
2007-09-16 22:40:33 +00:00
|
|
|
net->dev_name_head = netdev_create_hash();
|
|
|
|
if (net->dev_name_head == NULL)
|
|
|
|
goto err_name;
|
2007-09-17 18:56:21 +00:00
|
|
|
|
2007-09-16 22:40:33 +00:00
|
|
|
net->dev_index_head = netdev_create_hash();
|
|
|
|
if (net->dev_index_head == NULL)
|
|
|
|
goto err_idx;
|
2007-09-17 18:56:21 +00:00
|
|
|
|
|
|
|
return 0;
|
2007-09-16 22:40:33 +00:00
|
|
|
|
|
|
|
err_idx:
|
|
|
|
kfree(net->dev_name_head);
|
|
|
|
err_name:
|
|
|
|
return -ENOMEM;
|
2007-09-17 18:56:21 +00:00
|
|
|
}
|
|
|
|
|
2008-09-30 09:23:58 +00:00
|
|
|
/**
|
|
|
|
* netdev_drivername - network driver for the device
|
|
|
|
* @dev: network device
|
|
|
|
*
|
|
|
|
* Determine network driver for device.
|
|
|
|
*/
|
2011-06-06 23:41:33 +00:00
|
|
|
const char *netdev_drivername(const struct net_device *dev)
|
2008-07-21 20:31:48 +00:00
|
|
|
{
|
2008-09-30 09:22:14 +00:00
|
|
|
const struct device_driver *driver;
|
|
|
|
const struct device *parent;
|
2011-06-06 23:41:33 +00:00
|
|
|
const char *empty = "";
|
2008-07-21 20:31:48 +00:00
|
|
|
|
|
|
|
parent = dev->dev.parent;
|
|
|
|
if (!parent)
|
2011-06-06 23:41:33 +00:00
|
|
|
return empty;
|
2008-07-21 20:31:48 +00:00
|
|
|
|
|
|
|
driver = parent->driver;
|
|
|
|
if (driver && driver->name)
|
2011-06-06 23:41:33 +00:00
|
|
|
return driver->name;
|
|
|
|
return empty;
|
2008-07-21 20:31:48 +00:00
|
|
|
}
|
|
|
|
|
2012-09-13 03:12:19 +00:00
|
|
|
static int __netdev_printk(const char *level, const struct net_device *dev,
|
2010-06-27 01:02:35 +00:00
|
|
|
struct va_format *vaf)
|
|
|
|
{
|
|
|
|
int r;
|
|
|
|
|
2012-09-13 03:12:19 +00:00
|
|
|
if (dev && dev->dev.parent) {
|
2012-09-13 03:14:11 +00:00
|
|
|
r = dev_printk_emit(level[1] - '0',
|
|
|
|
dev->dev.parent,
|
|
|
|
"%s %s %s: %pV",
|
|
|
|
dev_driver_string(dev->dev.parent),
|
|
|
|
dev_name(dev->dev.parent),
|
|
|
|
netdev_name(dev), vaf);
|
2012-09-13 03:12:19 +00:00
|
|
|
} else if (dev) {
|
2010-06-27 01:02:35 +00:00
|
|
|
r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
|
2012-09-13 03:12:19 +00:00
|
|
|
} else {
|
2010-06-27 01:02:35 +00:00
|
|
|
r = printk("%s(NULL net_device): %pV", level, vaf);
|
2012-09-13 03:12:19 +00:00
|
|
|
}
|
2010-06-27 01:02:35 +00:00
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
int netdev_printk(const char *level, const struct net_device *dev,
|
|
|
|
const char *format, ...)
|
|
|
|
{
|
|
|
|
struct va_format vaf;
|
|
|
|
va_list args;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
va_start(args, format);
|
|
|
|
|
|
|
|
vaf.fmt = format;
|
|
|
|
vaf.va = &args;
|
|
|
|
|
|
|
|
r = __netdev_printk(level, dev, &vaf);
|
2012-09-13 03:12:19 +00:00
|
|
|
|
2010-06-27 01:02:35 +00:00
|
|
|
va_end(args);
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(netdev_printk);
|
|
|
|
|
|
|
|
#define define_netdev_printk_level(func, level) \
|
|
|
|
int func(const struct net_device *dev, const char *fmt, ...) \
|
|
|
|
{ \
|
|
|
|
int r; \
|
|
|
|
struct va_format vaf; \
|
|
|
|
va_list args; \
|
|
|
|
\
|
|
|
|
va_start(args, fmt); \
|
|
|
|
\
|
|
|
|
vaf.fmt = fmt; \
|
|
|
|
vaf.va = &args; \
|
|
|
|
\
|
|
|
|
r = __netdev_printk(level, dev, &vaf); \
|
2012-09-13 03:12:19 +00:00
|
|
|
\
|
2010-06-27 01:02:35 +00:00
|
|
|
va_end(args); \
|
|
|
|
\
|
|
|
|
return r; \
|
|
|
|
} \
|
|
|
|
EXPORT_SYMBOL(func);
|
|
|
|
|
|
|
|
define_netdev_printk_level(netdev_emerg, KERN_EMERG);
|
|
|
|
define_netdev_printk_level(netdev_alert, KERN_ALERT);
|
|
|
|
define_netdev_printk_level(netdev_crit, KERN_CRIT);
|
|
|
|
define_netdev_printk_level(netdev_err, KERN_ERR);
|
|
|
|
define_netdev_printk_level(netdev_warn, KERN_WARNING);
|
|
|
|
define_netdev_printk_level(netdev_notice, KERN_NOTICE);
|
|
|
|
define_netdev_printk_level(netdev_info, KERN_INFO);
|
|
|
|
|
2007-10-09 03:38:39 +00:00
|
|
|
static void __net_exit netdev_exit(struct net *net)
|
2007-09-17 18:56:21 +00:00
|
|
|
{
|
|
|
|
kfree(net->dev_name_head);
|
|
|
|
kfree(net->dev_index_head);
|
|
|
|
}
|
|
|
|
|
2007-11-13 11:23:50 +00:00
|
|
|
static struct pernet_operations __net_initdata netdev_net_ops = {
|
2007-09-17 18:56:21 +00:00
|
|
|
.init = netdev_init,
|
|
|
|
.exit = netdev_exit,
|
|
|
|
};
|
|
|
|
|
2007-10-09 03:38:39 +00:00
|
|
|
static void __net_exit default_device_exit(struct net *net)
|
2007-09-12 11:53:49 +00:00
|
|
|
{
|
2009-11-29 22:25:30 +00:00
|
|
|
struct net_device *dev, *aux;
|
2007-09-12 11:53:49 +00:00
|
|
|
/*
|
2009-11-29 22:25:30 +00:00
|
|
|
* Push all migratable network devices back to the
|
2007-09-12 11:53:49 +00:00
|
|
|
* initial network namespace
|
|
|
|
*/
|
|
|
|
rtnl_lock();
|
2009-11-29 22:25:30 +00:00
|
|
|
for_each_netdev_safe(net, dev, aux) {
|
2007-09-12 11:53:49 +00:00
|
|
|
int err;
|
netns: Fix arbitrary net_device-s corruptions on net_ns stop.
When a net namespace is destroyed, some devices (those, not killed
on ns stop explicitly) are moved back to init_net.
The problem, is that this net_ns change has one point of failure -
the __dev_alloc_name() may be called if a name collision occurs (and
this is easy to trigger). This allocator performs a likely-to-fail
GFP_ATOMIC allocation to find a suitable number. Other possible
conditions that may cause error (for device being ns local or not
registered) are always false in this case.
So, when this call fails, the device is unregistered. But this is
*not* the right thing to do, since after this the device may be
released (and kfree-ed) improperly. E. g. bridges require more
actions (sysfs update, timer disarming, etc.), some other devices
want to remove their private areas from lists, etc.
I. e. arbitrary use-after-free cases may occur.
The proposed fix is the following: since the only reason for the
dev_change_net_namespace to fail is the name generation, we may
give it a unique fall-back name w/o %d-s in it - the dev<ifindex>
one, since ifindexes are still unique.
So make this change, raise the failure-case printk loglevel to
EMERG and replace the unregister_netdevice call with BUG().
[ Use snprintf() -DaveM ]
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-05-08 08:24:25 +00:00
|
|
|
char fb_name[IFNAMSIZ];
|
2007-09-12 11:53:49 +00:00
|
|
|
|
|
|
|
/* Ignore unmoveable devices (i.e. loopback) */
|
|
|
|
if (dev->features & NETIF_F_NETNS_LOCAL)
|
|
|
|
continue;
|
|
|
|
|
2009-11-29 22:25:30 +00:00
|
|
|
/* Leave virtual devices for the generic cleanup */
|
|
|
|
if (dev->rtnl_link_ops)
|
|
|
|
continue;
|
2008-11-05 23:59:38 +00:00
|
|
|
|
2011-03-31 01:57:33 +00:00
|
|
|
/* Push remaining network devices to init_net */
|
netns: Fix arbitrary net_device-s corruptions on net_ns stop.
When a net namespace is destroyed, some devices (those, not killed
on ns stop explicitly) are moved back to init_net.
The problem, is that this net_ns change has one point of failure -
the __dev_alloc_name() may be called if a name collision occurs (and
this is easy to trigger). This allocator performs a likely-to-fail
GFP_ATOMIC allocation to find a suitable number. Other possible
conditions that may cause error (for device being ns local or not
registered) are always false in this case.
So, when this call fails, the device is unregistered. But this is
*not* the right thing to do, since after this the device may be
released (and kfree-ed) improperly. E. g. bridges require more
actions (sysfs update, timer disarming, etc.), some other devices
want to remove their private areas from lists, etc.
I. e. arbitrary use-after-free cases may occur.
The proposed fix is the following: since the only reason for the
dev_change_net_namespace to fail is the name generation, we may
give it a unique fall-back name w/o %d-s in it - the dev<ifindex>
one, since ifindexes are still unique.
So make this change, raise the failure-case printk loglevel to
EMERG and replace the unregister_netdevice call with BUG().
[ Use snprintf() -DaveM ]
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-05-08 08:24:25 +00:00
|
|
|
snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
|
|
|
|
err = dev_change_net_namespace(dev, &init_net, fb_name);
|
2007-09-12 11:53:49 +00:00
|
|
|
if (err) {
|
2012-02-01 10:54:43 +00:00
|
|
|
pr_emerg("%s: failed to move %s to init_net: %d\n",
|
|
|
|
__func__, dev->name, err);
|
netns: Fix arbitrary net_device-s corruptions on net_ns stop.
When a net namespace is destroyed, some devices (those, not killed
on ns stop explicitly) are moved back to init_net.
The problem, is that this net_ns change has one point of failure -
the __dev_alloc_name() may be called if a name collision occurs (and
this is easy to trigger). This allocator performs a likely-to-fail
GFP_ATOMIC allocation to find a suitable number. Other possible
conditions that may cause error (for device being ns local or not
registered) are always false in this case.
So, when this call fails, the device is unregistered. But this is
*not* the right thing to do, since after this the device may be
released (and kfree-ed) improperly. E. g. bridges require more
actions (sysfs update, timer disarming, etc.), some other devices
want to remove their private areas from lists, etc.
I. e. arbitrary use-after-free cases may occur.
The proposed fix is the following: since the only reason for the
dev_change_net_namespace to fail is the name generation, we may
give it a unique fall-back name w/o %d-s in it - the dev<ifindex>
one, since ifindexes are still unique.
So make this change, raise the failure-case printk loglevel to
EMERG and replace the unregister_netdevice call with BUG().
[ Use snprintf() -DaveM ]
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-05-08 08:24:25 +00:00
|
|
|
BUG();
|
2007-09-12 11:53:49 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
rtnl_unlock();
|
|
|
|
}
|
|
|
|
|
2009-12-03 02:29:04 +00:00
|
|
|
static void __net_exit default_device_exit_batch(struct list_head *net_list)
|
|
|
|
{
|
|
|
|
/* At exit all network devices most be removed from a network
|
tree-wide: fix comment/printk typos
"gadget", "through", "command", "maintain", "maintain", "controller", "address",
"between", "initiali[zs]e", "instead", "function", "select", "already",
"equal", "access", "management", "hierarchy", "registration", "interest",
"relative", "memory", "offset", "already",
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
2010-11-01 19:38:34 +00:00
|
|
|
* namespace. Do this in the reverse order of registration.
|
2009-12-03 02:29:04 +00:00
|
|
|
* Do this across as many network namespaces as possible to
|
|
|
|
* improve batching efficiency.
|
|
|
|
*/
|
|
|
|
struct net_device *dev;
|
|
|
|
struct net *net;
|
|
|
|
LIST_HEAD(dev_kill_list);
|
|
|
|
|
|
|
|
rtnl_lock();
|
|
|
|
list_for_each_entry(net, net_list, exit_list) {
|
|
|
|
for_each_netdev_reverse(net, dev) {
|
|
|
|
if (dev->rtnl_link_ops)
|
|
|
|
dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
|
|
|
|
else
|
|
|
|
unregister_netdevice_queue(dev, &dev_kill_list);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
unregister_netdevice_many(&dev_kill_list);
|
2011-02-17 22:59:19 +00:00
|
|
|
list_del(&dev_kill_list);
|
2009-12-03 02:29:04 +00:00
|
|
|
rtnl_unlock();
|
|
|
|
}
|
|
|
|
|
2007-11-13 11:23:50 +00:00
|
|
|
static struct pernet_operations __net_initdata default_device_ops = {
|
2007-09-12 11:53:49 +00:00
|
|
|
.exit = default_device_exit,
|
2009-12-03 02:29:04 +00:00
|
|
|
.exit_batch = default_device_exit_batch,
|
2007-09-12 11:53:49 +00:00
|
|
|
};
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/*
|
|
|
|
* Initialize the DEV module. At boot time this walks the device list and
|
|
|
|
* unhooks any devices that fail to initialise (normally hardware not
|
|
|
|
* present) and leaves us with a valid list of present and active devices.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is called single threaded during boot, so no need
|
|
|
|
* to take the rtnl semaphore.
|
|
|
|
*/
|
|
|
|
static int __init net_dev_init(void)
|
|
|
|
{
|
|
|
|
int i, rc = -ENOMEM;
|
|
|
|
|
|
|
|
BUG_ON(!dev_boot_phase);
|
|
|
|
|
|
|
|
if (dev_proc_init())
|
|
|
|
goto out;
|
|
|
|
|
2007-09-27 05:02:53 +00:00
|
|
|
if (netdev_kobject_init())
|
2005-04-16 22:20:36 +00:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
INIT_LIST_HEAD(&ptype_all);
|
2007-11-26 12:12:58 +00:00
|
|
|
for (i = 0; i < PTYPE_HASH_SIZE; i++)
|
2005-04-16 22:20:36 +00:00
|
|
|
INIT_LIST_HEAD(&ptype_base[i]);
|
|
|
|
|
2012-11-15 08:49:10 +00:00
|
|
|
INIT_LIST_HEAD(&offload_base);
|
|
|
|
|
2007-09-17 18:56:21 +00:00
|
|
|
if (register_pernet_subsys(&netdev_net_ops))
|
|
|
|
goto out;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Initialise the packet receive queues.
|
|
|
|
*/
|
|
|
|
|
2006-04-11 05:52:50 +00:00
|
|
|
for_each_possible_cpu(i) {
|
2010-04-19 21:17:14 +00:00
|
|
|
struct softnet_data *sd = &per_cpu(softnet_data, i);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2010-05-02 05:42:16 +00:00
|
|
|
memset(sd, 0, sizeof(*sd));
|
2010-04-19 21:17:14 +00:00
|
|
|
skb_queue_head_init(&sd->input_pkt_queue);
|
2010-04-27 22:07:33 +00:00
|
|
|
skb_queue_head_init(&sd->process_queue);
|
2010-04-19 21:17:14 +00:00
|
|
|
sd->completion_queue = NULL;
|
|
|
|
INIT_LIST_HEAD(&sd->poll_list);
|
2010-04-26 23:06:24 +00:00
|
|
|
sd->output_queue = NULL;
|
|
|
|
sd->output_queue_tailp = &sd->output_queue;
|
2010-03-24 19:13:54 +00:00
|
|
|
#ifdef CONFIG_RPS
|
2010-04-19 21:17:14 +00:00
|
|
|
sd->csd.func = rps_trigger_softirq;
|
|
|
|
sd->csd.info = sd;
|
|
|
|
sd->csd.flags = 0;
|
|
|
|
sd->cpu = i;
|
2010-03-19 00:45:44 +00:00
|
|
|
#endif
|
2010-03-16 08:03:29 +00:00
|
|
|
|
2010-04-19 21:17:14 +00:00
|
|
|
sd->backlog.poll = process_backlog;
|
|
|
|
sd->backlog.weight = weight_p;
|
|
|
|
sd->backlog.gro_list = NULL;
|
|
|
|
sd->backlog.gro_count = 0;
|
2013-05-20 04:02:32 +00:00
|
|
|
|
|
|
|
#ifdef CONFIG_NET_FLOW_LIMIT
|
|
|
|
sd->flow_limit = NULL;
|
|
|
|
#endif
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
dev_boot_phase = 0;
|
|
|
|
|
2008-11-08 06:54:20 +00:00
|
|
|
/* The loopback device is special if any other network devices
|
|
|
|
* is present in a network namespace the loopback device must
|
|
|
|
* be present. Since we now dynamically allocate and free the
|
|
|
|
* loopback device ensure this invariant is maintained by
|
|
|
|
* keeping the loopback device as the first device on the
|
|
|
|
* list of network devices. Ensuring the loopback devices
|
|
|
|
* is the first device that appears and the last network device
|
|
|
|
* that disappears.
|
|
|
|
*/
|
|
|
|
if (register_pernet_device(&loopback_net_ops))
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
if (register_pernet_device(&default_device_ops))
|
|
|
|
goto out;
|
|
|
|
|
Remove argument from open_softirq which is always NULL
As git-grep shows, open_softirq() is always called with the last argument
being NULL
block/blk-core.c: open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
kernel/hrtimer.c: open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);
kernel/rcuclassic.c: open_softirq(RCU_SOFTIRQ, rcu_process_callbacks, NULL);
kernel/rcupreempt.c: open_softirq(RCU_SOFTIRQ, rcu_process_callbacks, NULL);
kernel/sched.c: open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL);
kernel/softirq.c: open_softirq(TASKLET_SOFTIRQ, tasklet_action, NULL);
kernel/softirq.c: open_softirq(HI_SOFTIRQ, tasklet_hi_action, NULL);
kernel/timer.c: open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
net/core/dev.c: open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
net/core/dev.c: open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
This observation has already been made by Matthew Wilcox in June 2002
(http://www.cs.helsinki.fi/linux/linux-kernel/2002-25/0687.html)
"I notice that none of the current softirq routines use the data element
passed to them."
and the situation hasn't changed since them. So it appears we can safely
remove that extra argument to save 128 (54) bytes of kernel data (text).
Signed-off-by: Carlos R. Mafra <crmafra@ift.unesp.br>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-15 14:15:37 +00:00
|
|
|
open_softirq(NET_TX_SOFTIRQ, net_tx_action);
|
|
|
|
open_softirq(NET_RX_SOFTIRQ, net_rx_action);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
hotcpu_notifier(dev_cpu_callback, 0);
|
|
|
|
dst_init();
|
|
|
|
rc = 0;
|
|
|
|
out:
|
|
|
|
return rc;
|
|
|
|
}
|
|
|
|
|
|
|
|
subsys_initcall(net_dev_init);
|