MSG_ZEROCOPY ensures that pinned user pages do not exceed the limit.
If no limit is set, skip this accounting as otherwise expensive
atomic_long operations are called for no reason.
This accounting is already skipped for privileged (CAP_IPC_LOCK)
users. Rely on the same mechanism: if no mmp->user is set,
mm_unaccount_pinned_pages does not decrement either.
Tested by running tools/testing/selftests/net/msg_zerocopy.sh with
an unprivileged user for the TXMODE binary:
ip netns exec "${NS1}" sudo -u "{$USER}" "${BIN}" "-${IP}" ...
Signed-off-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230214155740.3448763-1-willemdebruijn.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
skbuff_head_cache is misnamed (perhaps for historical reasons?)
because it does not hold heads. Head is the buffer which skb->data
points to, and also where shinfo lives. struct sk_buff is a metadata
structure, not the head.
Eric recently added skb_small_head_cache (which allocates actual
head buffers), let that serve as an excuse to finally clean this up :)
Leave the user-space visible name intact, it could possibly be uAPI.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Recent removal of ksize() in alloc_skb() increased
performance because we no longer read
the associated struct page.
We have an equivalent cost at kfree_skb() time.
kfree(skb->head) has to access a struct page,
often cold in cpu caches to get the owning
struct kmem_cache.
Considering that many allocations are small (at least for TCP ones)
we can have our own kmem_cache to avoid the cache line miss.
This also saves memory because these small heads
are no longer padded to 1024 bytes.
CONFIG_SLUB=y
$ grep skbuff_small_head /proc/slabinfo
skbuff_small_head 2907 2907 640 51 8 : tunables 0 0 0 : slabdata 57 57 0
CONFIG_SLAB=y
$ grep skbuff_small_head /proc/slabinfo
skbuff_small_head 607 624 640 6 1 : tunables 54 27 8 : slabdata 104 104 5
Notes:
- After Kees Cook patches and this one, we might
be able to revert commit
dbae2b0628 ("net: skb: introduce and use a single page frag cache")
because GRO_MAX_HEAD is also small.
- This patch is a NOP for CONFIG_SLOB=y builds.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
All kmalloc_reserve() callers have to make the same computation,
we can factorize them, to prepare following patch in the series.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This is a cleanup patch, to prepare following change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We have many places using this expression:
SKB_DATA_ALIGN(sizeof(struct skb_shared_info))
Use of SKB_HEAD_ALIGN() will allow to clean them.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
First user of skb_poison_list is in kfree_skb_list_reason, to catch bugs
earlier like introduced in commit eedade12f4 ("net: kfree_skb_list use
kmem_cache_free_bulk"). For completeness mentioned bug have been fixed in
commit f72ff8b81e ("net: fix kfree_skb_list use of skb_mark_not_on_list").
In case of a bug like mentioned commit we would have seen OOPS with:
general protection fault, probably for non-canonical address 0xdead000000000870
And content of one the registers e.g. R13: dead000000000800
In this case skb->len is at offset 112 bytes (0x70) why fault happens at
0x800+0x70 = 0x870
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This include was added for skb_find_text() but all we need there
is a forward declaration of struct ts_config.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
A bug was introduced by commit eedade12f4 ("net: kfree_skb_list use
kmem_cache_free_bulk"). It unconditionally unlinked the SKB list via
invoking skb_mark_not_on_list().
In this patch we choose to remove the skb_mark_not_on_list() call as it
isn't necessary. It would be possible and correct to call
skb_mark_not_on_list() only when __kfree_skb_reason() returns true,
meaning the SKB is ready to be free'ed, as it calls/check skb_unref().
This fix is needed as kfree_skb_list() is also invoked on skb_shared_info
frag_list (skb_drop_fraglist() calling kfree_skb_list()). A frag_list can
have SKBs with elevated refcnt due to cloning via skb_clone_fraglist(),
which takes a reference on all SKBs in the list. This implies the
invariant that all SKBs in the list must have the same refcnt, when using
kfree_skb_list().
Reported-by: syzbot+c8a2e66e37eee553c4fd@syzkaller.appspotmail.com
Reported-and-tested-by: syzbot+c8a2e66e37eee553c4fd@syzkaller.appspotmail.com
Fixes: eedade12f4 ("net: kfree_skb_list use kmem_cache_free_bulk")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/167421088417.1125894.9761158218878962159.stgit@firesoul
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The kfree_skb_list function walks SKB (via skb->next) and frees them
individually to the SLUB/SLAB allocator (kmem_cache). It is more
efficient to bulk free them via the kmem_cache_free_bulk API.
This patches create a stack local array with SKBs to bulk free while
walking the list. Bulk array size is limited to 16 SKBs to trade off
stack usage and efficiency. The SLUB kmem_cache "skbuff_head_cache"
uses objsize 256 bytes usually in an order-1 page 8192 bytes that is
32 objects per slab (can vary on archs and due to SLUB sharing). Thus,
for SLUB the optimal bulk free case is 32 objects belonging to same
slab, but runtime this isn't likely to occur.
The expected gain from using kmem_cache bulk alloc and free API
have been assessed via a microbencmark kernel module[1].
The module 'slab_bulk_test01' results at bulk 16 element:
kmem-in-loop Per elem: 109 cycles(tsc) 30.532 ns (step:16)
kmem-bulk Per elem: 64 cycles(tsc) 17.905 ns (step:16)
More detailed description of benchmarks avail in [2].
[1] https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm
[2] https://github.com/xdp-project/xdp-project/blob/master/areas/mem/kfree_skb_list01.org
V2: rename function to kfree_skb_add_bulk.
Reviewed-by: Saeed Mahameed <saeed@kernel.org>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The SKB drop reason uses __builtin_return_address(0) to give the call
"location" to trace_kfree_skb() tracepoint skb:kfree_skb.
To keep this stable for compilers kfree_skb_reason() is annotated with
__fix_address (noinline __noclone) as fixed in commit c205cc7534
("net: skb: prevent the split of kfree_skb_reason() by gcc").
The function kfree_skb_list_reason() invoke kfree_skb_reason(), which
cause the __builtin_return_address(0) "location" to report the
unexpected address of kfree_skb_list_reason.
Example output from 'perf script':
kpktgend_0 1337 [000] 81.002597: skb:kfree_skb: skbaddr=0xffff888144824700 protocol=2048 location=kfree_skb_list_reason+0x1e reason: QDISC_DROP
Patch creates an __always_inline __kfree_skb_reason() helper call that
is called from both kfree_skb_list() and kfree_skb_list_reason().
Suggestions for solutions that shares code better are welcome.
As preparation for next patch move __kfree_skb() invocation out of
this helper function.
Reviewed-by: Saeed Mahameed <saeed@kernel.org>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Since commit ce098da149 ("skbuff: Introduce slab_build_skb()")
drivers trying to build skb around slab-backed buffers should
go via slab_build_skb() rather than passing frag_size = 0 to
the main build_skb().
Remove the copy'n'pasted comments about 0 meaning slab.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extending the tail can have some unexpected side effects if a program uses
a helper like BPF_FUNC_skb_pull_data to read partial content beyond the
head skb headlen when all the skbs in the gso frag_list are linear with no
head_frag -
kernel BUG at net/core/skbuff.c:4219!
pc : skb_segment+0xcf4/0xd2c
lr : skb_segment+0x63c/0xd2c
Call trace:
skb_segment+0xcf4/0xd2c
__udp_gso_segment+0xa4/0x544
udp4_ufo_fragment+0x184/0x1c0
inet_gso_segment+0x16c/0x3a4
skb_mac_gso_segment+0xd4/0x1b0
__skb_gso_segment+0xcc/0x12c
udp_rcv_segment+0x54/0x16c
udp_queue_rcv_skb+0x78/0x144
udp_unicast_rcv_skb+0x8c/0xa4
__udp4_lib_rcv+0x490/0x68c
udp_rcv+0x20/0x30
ip_protocol_deliver_rcu+0x1b0/0x33c
ip_local_deliver+0xd8/0x1f0
ip_rcv+0x98/0x1a4
deliver_ptype_list_skb+0x98/0x1ec
__netif_receive_skb_core+0x978/0xc60
Fix this by marking these skbs as GSO_DODGY so segmentation can handle
the tail updates accordingly.
Fixes: 3dcbdb134f ("net: gso: Fix skb_segment splat when splitting gso_size mangled skb having linear-headed frag_list")
Signed-off-by: Sean Tranchetti <quic_stranche@quicinc.com>
Signed-off-by: Subash Abhinov Kasiviswanathan <quic_subashab@quicinc.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/1671084718-24796-1-git-send-email-quic_subashab@quicinc.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
syzkaller reported:
BUG: KASAN: slab-out-of-bounds in __build_skb_around+0x235/0x340 net/core/skbuff.c:294
Write of size 32 at addr ffff88802aa172c0 by task syz-executor413/5295
For bpf_prog_test_run_skb(), which uses a kmalloc()ed buffer passed to
build_skb().
When build_skb() is passed a frag_size of 0, it means the buffer came
from kmalloc. In these cases, ksize() is used to find its actual size,
but since the allocation may not have been made to that size, actually
perform the krealloc() call so that all the associated buffer size
checking will be correctly notified (and use the "new" pointer so that
compiler hinting works correctly). Split this logic out into a new
interface, slab_build_skb(), but leave the original 0 checking for now
to catch any stragglers.
Reported-by: syzbot+fda18eaa8c12534ccb3b@syzkaller.appspotmail.com
Link: https://groups.google.com/g/syzkaller-bugs/c/UnIKxTtU5-0/m/-wbXinkgAQAJ
Fixes: 38931d8989 ("mm: Make ksize() a reporting-only function")
Cc: Pavel Begunkov <asml.silence@gmail.com>
Cc: pepsipu <soopthegoop@gmail.com>
Cc: syzbot+fda18eaa8c12534ccb3b@syzkaller.appspotmail.com
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: kasan-dev <kasan-dev@googlegroups.com>
Cc: Andrii Nakryiko <andrii@kernel.org>
Cc: ast@kernel.org
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Hao Luo <haoluo@google.com>
Cc: Jesper Dangaard Brouer <hawk@kernel.org>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: jolsa@kernel.org
Cc: KP Singh <kpsingh@kernel.org>
Cc: martin.lau@linux.dev
Cc: Stanislav Fomichev <sdf@google.com>
Cc: song@kernel.org
Cc: Yonghong Song <yhs@fb.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20221208060256.give.994-kees@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Adopt atomic_long_try_cmpxchg() in mm_account_pinned_pages()
as it is slightly more efficient.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
-----BEGIN PGP SIGNATURE-----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=q04p
-----END PGP SIGNATURE-----
Merge tag 'rxrpc-next-20221108' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs
rxrpc changes
David Howells says:
====================
rxrpc: Increasing SACK size and moving away from softirq, part 1
AF_RXRPC has some issues that need addressing:
(1) The SACK table has a maximum capacity of 255, but for modern networks
that isn't sufficient. This is hard to increase in the upstream code
because of the way the application thread is coupled to the softirq
and retransmission side through a ring buffer. Adjustments to the rx
protocol allows a capacity of up to 8192, and having a ring
sufficiently large to accommodate that would use an excessive amount
of memory as this is per-call.
(2) Processing ACKs in softirq mode causes the ACKs get conflated, with
only the most recent being considered. Whilst this has the upside
that the retransmission algorithm only needs to deal with the most
recent ACK, it causes DATA transmission for a call to be very bursty
because DATA packets cannot be transmitted in softirq mode. Rather
transmission must be delegated to either the application thread or a
workqueue, so there tend to be sudden bursts of traffic for any
particular call due to scheduling delays.
(3) All crypto in a single call is done in series; however, each DATA
packet is individually encrypted so encryption and decryption of large
calls could be parallelised if spare CPU resources are available.
This is the first of a number of sets of patches that try and address them.
The overall aims of these changes include:
(1) To get rid of the TxRx ring and instead pass the packets round in
queues (eg. sk_buff_head). On the Tx side, each ACK packet comes with
a SACK table that can be parsed as-is, so there's no particular need
to maintain our own; we just have to refer to the ACK.
On the Rx side, we do need to maintain a SACK table with one bit per
entry - but only if packets go missing - and we don't want to have to
perform a complex transformation to get the information into an ACK
packet.
(2) To try and move almost all processing of received packets out of the
softirq handler and into a high-priority kernel I/O thread. Only the
transferral of packets would be left there. I would still use the
encap_rcv hook to receive packets as there's a noticeable performance
drop from letting the UDP socket put the packets into its own queue
and then getting them out of there.
(3) To make the I/O thread also do all the transmission. The app thread
would be responsible for packaging the data into packets and then
buffering them for the I/O thread to transmit. This would make it
easier for the app thread to run ahead of the I/O thread, and would
mean the I/O thread is less likely to have to wait around for a new
packet to come available for transmission.
(4) To logically partition the socket/UAPI/KAPI side of things from the
I/O side of things. The local endpoint, connection, peer and call
objects would belong to the I/O side. The socket side would not then
touch the private internals of calls and suchlike and would not change
their states. It would only look at the send queue, receive queue and
a way to pass a message to cause an abort.
(5) To remove as much locking, synchronisation, barriering and atomic ops
as possible from the I/O side. Exclusion would be achieved by
limiting modification of state to the I/O thread only. Locks would
still need to be used in communication with the UDP socket and the
AF_RXRPC socket API.
(6) To provide crypto offload kernel threads that, when there's slack in
the system, can see packets that need crypting and provide
parallelisation in dealing with them.
(7) To remove the use of system timers. Since each timer would then send
a poke to the I/O thread, which would then deal with it when it had
the opportunity, there seems no point in using system timers if,
instead, a list of timeouts can be sensibly consulted. An I/O thread
only then needs to schedule with a timeout when it is idle.
(8) To use zero-copy sendmsg to send packets. This would make use of the
I/O thread being the sole transmitter on the socket to manage the
dead-reckoning sequencing of the completion notifications. There is a
problem with zero-copy, though: the UDP socket doesn't handle running
out of option memory very gracefully.
With regard to this first patchset, the changes made include:
(1) Some fixes, including a fallback for proc_create_net_single_write(),
setting ack.bufferSize to 0 in ACK packets and a fix for rxrpc
congestion management, which shouldn't be saving the cwnd value
between calls.
(2) Improvements in rxrpc tracepoints, including splitting the timer
tracepoint into a set-timer and a timer-expired trace.
(3) Addition of a new proc file to display some stats.
(4) Some code cleanups, including removing some unused bits and
unnecessary header inclusions.
(5) A change to the recently added UDP encap_err_rcv hook so that it has
the same signature as {ip,ipv6}_icmp_error(), and then just have rxrpc
point its UDP socket's hook directly at those.
(6) Definition of a new struct, rxrpc_txbuf, that is used to hold
transmissible packets of DATA and ACK type in a single 2KiB block
rather than using an sk_buff. This allows the buffer to be on a
number of queues simultaneously more easily, and also guarantees that
the entire block is in a single unit for zerocopy purposes and that
the data payload is aligned for in-place crypto purposes.
(7) ACK txbufs are allocated at proposal and queued for later transmission
rather than being stored in a single place in the rxrpc_call struct,
which means only a single ACK can be pending transmission at a time.
The queue is then drained at various points. This allows the ACK
generation code to be simplified.
(8) The Rx ring buffer is removed. When a jumbo packet is received (which
comprises a number of ordinary DATA packets glued together), it used
to be pointed to by the ring multiple times, with an annotation in a
side ring indicating which subpacket was in that slot - but this is no
longer possible. Instead, the packet is cloned once for each
subpacket, barring the last, and the range of data is set in the skb
private area. This makes it easier for the subpackets in a jumbo
packet to be decrypted in parallel.
(9) The Tx ring buffer is removed. The side annotation ring that held the
SACK information is also removed. Instead, in the event of packet
loss, the SACK data attached an ACK packet is parsed.
(10) Allocate an skcipher request when needed in the rxkad security class
rather than caching one in the rxrpc_call struct. This deals with a
race between externally-driven call disconnection getting rid of the
skcipher request and sendmsg/recvmsg trying to use it because they
haven't seen the completion yet. This is also needed to support
parallelisation as the skcipher request cannot be used by two or more
threads simultaneously.
(11) Call udp_sendmsg() and udpv6_sendmsg() directly rather than going
through kernel_sendmsg() so that we can provide our own iterator
(zerocopy explicitly doesn't work with a KVEC iterator). This also
lets us avoid the overhead of the security hook.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Improve the tracking of which packets need to be transmitted by saving the
last ACK packet that we receive that has a populated soft-ACK table rather
than marking packets. Then we can step through the soft-ACK table and look
at the packets we've transmitted beyond that to determine which packets we
might want to retransmit.
We also look at the highest serial number that has been acked to try and
guess which packets we've transmitted the peer is likely to have seen. If
necessary, we send a ping to retrieve that number.
One downside that might be a problem is that we can't then compare the
previous acked/unacked state so easily in rxrpc_input_soft_acks() - which
is a potential problem for the slow-start algorithm.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Since commit 3dcbdb134f ("net: gso: Fix skb_segment splat when
splitting gso_size mangled skb having linear-headed frag_list"), it is
allowed to change gso_size of a GRO packet. However, that commit assumes
that "checking the first list_skb member suffices; i.e if either of the
list_skb members have non head_frag head, then the first one has too".
It turns out this assumption does not hold. We've seen BUG_ON being hit
in skb_segment when skbs on the frag_list had differing head_frag with
the vmxnet3 driver. This happens because __netdev_alloc_skb and
__napi_alloc_skb can return a skb that is page backed or kmalloced
depending on the requested size. As the result, the last small skb in
the GRO packet can be kmalloced.
There are three different locations where this can be fixed:
(1) We could check head_frag in GRO and not allow GROing skbs with
different head_frag. However, that would lead to performance
regression on normal forward paths with unmodified gso_size, where
!head_frag in the last packet is not a problem.
(2) Set a flag in bpf_skb_net_grow and bpf_skb_net_shrink indicating
that NETIF_F_SG is undesirable. That would need to eat a bit in
sk_buff. Furthermore, that flag can be unset when all skbs on the
frag_list are page backed. To retain good performance,
bpf_skb_net_grow/shrink would have to walk the frag_list.
(3) Walk the frag_list in skb_segment when determining whether
NETIF_F_SG should be cleared. This of course slows things down.
This patch implements (3). To limit the performance impact in
skb_segment, the list is walked only for skbs with SKB_GSO_DODGY set
that have gso_size changed. Normal paths thus will not hit it.
We could check only the last skb but since we need to walk the whole
list anyway, let's stay on the safe side.
Fixes: 3dcbdb134f ("net: gso: Fix skb_segment splat when splitting gso_size mangled skb having linear-headed frag_list")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/e04426a6a91baf4d1081e1b478c82b5de25fdf21.1667407944.git.jbenc@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When an skb with a frag list is consumed, we currently
pretend all skbs in the frag list were dropped.
In order to fix this, add a @reason argument to skb_release_data()
and skb_release_all().
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This will allow to simply use in the future:
kfree_skb_reason(skb, reason);
Instead of repeating sequences like:
if (dropped)
kfree_skb_reason(skb, reason);
else
consume_skb(skb);
For instance, following patch in the series is adding
@reason to skb_release_data() and skb_release_all(),
so that we can propagate a meaningful @reason whenever
consume_skb()/kfree_skb() have to take care of a potential frag_list.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
drivers/net/can/usb/kvaser_usb/kvaser_usb_leaf.c
2871edb32f ("can: kvaser_usb: Fix possible completions during init_completion")
abb8670938 ("can: kvaser_usb_leaf: Ignore stale bus-off after start")
8d21f5927a ("can: kvaser_usb_leaf: Fix improved state not being reported")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Instead of discovering the kmalloc bucket size _after_ allocation, round
up proactively so the allocation is explicitly made for the full size,
allowing the compiler to correctly reason about the resulting size of
the buffer through the existing __alloc_size() hint.
This will allow for kernels built with CONFIG_UBSAN_BOUNDS or the
coming dynamic bounds checking under CONFIG_FORTIFY_SOURCE to gain
back the __alloc_size() hints that were temporarily reverted in commit
93dd04ab0b ("slab: remove __alloc_size attribute from __kmalloc_track_caller")
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: netdev@vger.kernel.org
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: David Rientjes <rientjes@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Link: https://patchwork.kernel.org/project/netdevbpf/patch/20221021234713.you.031-kees@kernel.org/
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20221025223811.up.360-kees@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
skb_pp_recycle() is only used by skb_free_head() in
skbuff.c, so move it to skbuff.c.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
After commit 3226b158e6 ("net: avoid 32 x truesize under-estimation
for tiny skbs") we are observing 10-20% regressions in performance
tests with small packets. The perf trace points to high pressure on
the slab allocator.
This change tries to improve the allocation schema for small packets
using an idea originally suggested by Eric: a new per CPU page frag is
introduced and used in __napi_alloc_skb to cope with small allocation
requests.
To ensure that the above does not lead to excessive truesize
underestimation, the frag size for small allocation is inflated to 1K
and all the above is restricted to build with 4K page size.
Note that we need to update accordingly the run-time check introduced
with commit fd9ea57f4e ("net: add napi_get_frags_check() helper").
Alex suggested a smart page refcount schema to reduce the number
of atomic operations and deal properly with pfmemalloc pages.
Under small packet UDP flood, I measure a 15% peak tput increases.
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Suggested-by: Alexander H Duyck <alexanderduyck@fb.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/6b6f65957c59f86a353fc09a5127e83a32ab5999.1664350652.git.pabeni@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We can benefit from a smaller struct ubuf_info, so leave only mandatory
fields and let users to decide how they want to extend it. Convert
MSG_ZEROCOPY to struct ubuf_info_msgzc and remove duplicated fields.
This reduces the size from 48 bytes to just 16.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
drivers/net/ethernet/freescale/fec.h
7d650df99d ("net: fec: add pm_qos support on imx6q platform")
40c79ce13b ("net: fec: add stop mode support for imx8 platform")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
As Eric reported, the 'reason' field is not presented when trace the
kfree_skb event by perf:
$ perf record -e skb:kfree_skb -a sleep 10
$ perf script
ip_defrag 14605 [021] 221.614303: skb:kfree_skb:
skbaddr=0xffff9d2851242700 protocol=34525 location=0xffffffffa39346b1
reason:
The cause seems to be passing kernel address directly to TP_printk(),
which is not right. As the enum 'skb_drop_reason' is not exported to
user space through TRACE_DEFINE_ENUM(), perf can't get the drop reason
string from the 'reason' field, which is a number.
Therefore, we introduce the macro DEFINE_DROP_REASON(), which is used
to define the trace enum by TRACE_DEFINE_ENUM(). With the help of
DEFINE_DROP_REASON(), now we can remove the auto-generate that we
introduced in the commit ec43908dd5
("net: skb: use auto-generation to convert skb drop reason to string"),
and define the string array 'drop_reasons'.
Hmmmm...now we come back to the situation that have to maintain drop
reasons in both enum skb_drop_reason and DEFINE_DROP_REASON. But they
are both in dropreason.h, which makes it easier.
After this commit, now the format of kfree_skb is like this:
$ cat /tracing/events/skb/kfree_skb/format
name: kfree_skb
ID: 1524
format:
field:unsigned short common_type; offset:0; size:2; signed:0;
field:unsigned char common_flags; offset:2; size:1; signed:0;
field:unsigned char common_preempt_count; offset:3; size:1; signed:0;
field:int common_pid; offset:4; size:4; signed:1;
field:void * skbaddr; offset:8; size:8; signed:0;
field:void * location; offset:16; size:8; signed:0;
field:unsigned short protocol; offset:24; size:2; signed:0;
field:enum skb_drop_reason reason; offset:28; size:4; signed:0;
print fmt: "skbaddr=%p protocol=%u location=%p reason: %s", REC->skbaddr, REC->protocol, REC->location, __print_symbolic(REC->reason, { 1, "NOT_SPECIFIED" }, { 2, "NO_SOCKET" } ......
Fixes: ec43908dd5 ("net: skb: use auto-generation to convert skb drop reason to string")
Link: https://lore.kernel.org/netdev/CANn89i+bx0ybvE55iMYf5GJM48WwV1HNpdm9Q6t-HaEstqpCSA@mail.gmail.com/
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While reading sysctl_tstamp_allow_data, it can be changed
concurrently. Thus, we need to add READ_ONCE() to its reader.
Fixes: b245be1f4d ("net-timestamp: no-payload only sysctl")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
skb_copy_bits() could fail, which requires a check on the return
value.
Signed-off-by: Li Zhong <floridsleeves@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sometimes, gcc will optimize the function by spliting it to two or
more functions. In this case, kfree_skb_reason() is splited to
kfree_skb_reason and kfree_skb_reason.part.0. However, the
function/tracepoint trace_kfree_skb() in it needs the return address
of kfree_skb_reason().
This split makes the call chains becomes:
kfree_skb_reason() -> kfree_skb_reason.part.0 -> trace_kfree_skb()
which makes the return address that passed to trace_kfree_skb() be
kfree_skb().
Therefore, introduce '__fix_address', which is the combination of
'__noclone' and 'noinline', and apply it to kfree_skb_reason() to
prevent to from being splited or made inline.
(Is it better to simply apply '__noclone oninline' to kfree_skb_reason?
I'm thinking maybe other functions have the same problems)
Meanwhile, wrap 'skb_unref()' with 'unlikely()', as the compiler thinks
it is likely return true and splits kfree_skb_reason().
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pavel Begunkov says:
====================
io_uring zerocopy send
The patchset implements io_uring zerocopy send. It works with both registered
and normal buffers, mixing is allowed but not recommended. Apart from usual
request completions, just as with MSG_ZEROCOPY, io_uring separately notifies
the userspace when buffers are freed and can be reused (see API design below),
which is delivered into io_uring's Completion Queue. Those "buffer-free"
notifications are not necessarily per request, but the userspace has control
over it and should explicitly attaching a number of requests to a single
notification. The series also adds some internal optimisations when used with
registered buffers like removing page referencing.
From the kernel networking perspective there are two main changes. The first
one is passing ubuf_info into the network layer from io_uring (inside of an
in kernel struct msghdr). This allows extra optimisations, e.g. ubuf_info
caching on the io_uring side, but also helps to avoid cross-referencing
and synchronisation problems. The second part is an optional optimisation
removing page referencing for requests with registered buffers.
Benchmarking UDP with an optimised version of the selftest (see [1]), which
sends a bunch of requests, waits for completions and repeats. "+ flush" column
posts one additional "buffer-free" notification per request, and just "zc"
doesn't post buffer notifications at all.
NIC (requests / second):
IO size | non-zc | zc | zc + flush
4000 | 495134 | 606420 (+22%) | 558971 (+12%)
1500 | 551808 | 577116 (+4.5%) | 565803 (+2.5%)
1000 | 584677 | 592088 (+1.2%) | 560885 (-4%)
600 | 596292 | 598550 (+0.4%) | 555366 (-6.7%)
dummy (requests / second):
IO size | non-zc | zc | zc + flush
8000 | 1299916 | 2396600 (+84%) | 2224219 (+71%)
4000 | 1869230 | 2344146 (+25%) | 2170069 (+16%)
1200 | 2071617 | 2361960 (+14%) | 2203052 (+6%)
600 | 2106794 | 2381527 (+13%) | 2195295 (+4%)
Previously it also brought a massive performance speedup compared to the
msg_zerocopy tool (see [3]), which is probably not super interesting. There
is also an additional bunch of refcounting optimisations that was omitted from
the series for simplicity and as they don't change the picture drastically,
they will be sent as follow up, as well as flushing optimisations closing the
performance gap b/w two last columns.
For TCP on localhost (with hacks enabling localhost zerocopy) and including
additional overhead for receive:
IO size | non-zc | zc
1200 | 4174 | 4148
4096 | 7597 | 11228
Using a real NIC 1200 bytes, zc is worse than non-zc ~5-10%, maybe the
omitted optimisations will somewhat help, should look better for 4000,
but couldn't test properly because of setup problems.
Links:
liburing (benchmark + tests):
[1] https://github.com/isilence/liburing/tree/zc_v4
kernel repo:
[2] https://github.com/isilence/linux/tree/zc_v4
RFC v1:
[3] https://lore.kernel.org/io-uring/cover.1638282789.git.asml.silence@gmail.com/
RFC v2:
https://lore.kernel.org/io-uring/cover.1640029579.git.asml.silence@gmail.com/
Net patches based:
git@github.com:isilence/linux.git zc_v4-net-base
or
https://github.com/isilence/linux/tree/zc_v4-net-base
API design overview:
The series introduces an io_uring concept of notifactors. From the userspace
perspective it's an entity to which it can bind one or more requests and then
requesting to flush it. Flushing a notifier makes it impossible to attach new
requests to it, and instructs the notifier to post a completion once all
requests attached to it are completed and the kernel doesn't need the buffers
anymore.
Notifications are stored in notification slots, which should be registered as
an array in io_uring. Each slot stores only one notifier at any particular
moment. Flushing removes it from the slot and the slot automatically replaces
it with a new notifier. All operations with notifiers are done by specifying
an index of a slot it's currently in.
When registering a notification the userspace specifies a u64 tag for each
slot, which will be copied in notification completion entries as
cqe::user_data. cqe::res is 0 and cqe::flags is equal to wrap around u32
sequence number counting notifiers of a slot.
====================
Link: https://lore.kernel.org/r/cover.1657643355.git.asml.silence@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Some users like io_uring can do page pinning more efficiently, so we
want a way to delegate referencing to other subsystems. For that add
a new flag called SKBFL_MANAGED_FRAG_REFS. When set, skb doesn't hold
page references and upper layers are responsivle to managing page
lifetime.
It's allowed to convert skbs from managed to normal by calling
skb_zcopy_downgrade_managed(). The function will take all needed
page references and clear the flag. It's needed, for instance,
to avoid mixing managed modes.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add support for custom iov_iter handling to msghdr. The idea is that
in-kernel subsystems want control over how an SG is split.
Signed-off-by: David Ahern <dsahern@kernel.org>
[pavel: move callback into msghdr]
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We don't want to list every single ubuf_info callback in
skb_orphan_frags(), add a flag controlling the behaviour.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We should not append MSG_ZEROCOPY requests to skbuff with non
MSG_ZEROCOPY ubuf_info, they might be not compatible.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
TCP allocates 'fast clones' skbs for packets in tx queues.
Currently, __alloc_skb() initializes the companion fclone
field to SKB_FCLONE_CLONE, and leaves other fields untouched.
It makes sense to defer this init much later in skb_clone(),
because all fclone fields are copied and hot in cpu caches
at that time.
This removes one cache line miss in __alloc_skb(), cost seen
on an host with 256 cpus all competing on memory accesses.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the len fields manipulation in the skbs to a helper function.
There is a comment specifically requesting this and there are several
other areas in the code displaying the same pattern which can be
refactored.
This improves code readability.
Signed-off-by: Richard Gobert <richardbgobert@gmail.com>
Link: https://lore.kernel.org/r/20220622160853.GA6478@debian
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
NAPI cache skb_count is being checked twice without condition. Change to
checking the second time only if the first check is run.
Signed-off-by: Sieng Piaw Liew <liew.s.piaw@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 6454eca81e ("net: Use lockdep_assert_in_softirq()
in napi_consume_skb()") added a check in napi_consume_skb()
which is a bit weak.
napi_consume_skb() and __napi_alloc_skb() should only
be used from BH context, not from hard irq or nmi context,
otherwise we could have races.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Remove this check from fast path unless CONFIG_DEBUG_NET=y
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It is annoying to add new skb drop reasons to 'enum skb_drop_reason'
and TRACE_SKB_DROP_REASON in trace/event/skb.h, and it's easy to forget
to add the new reasons we added to TRACE_SKB_DROP_REASON.
TRACE_SKB_DROP_REASON is used to convert drop reason of type number
to string. For now, the string we passed to user space is exactly the
same as the name in 'enum skb_drop_reason' with a 'SKB_DROP_REASON_'
prefix. Therefore, we can use 'auto-generation' to generate these
drop reasons to string at build time.
The new source 'dropreason_str.c' will be auto generated during build
time, which contains the string array
'const char * const drop_reasons[]'.
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When user sets skb_defer_max to 1 the kick threshold is 0
(half of 1). If we increment queue length before the check
the kick will never happen, and the skb may get stranded.
This is likely harmless but can be avoided by moving the
increment after the check. This way skb_defer_max == 1
will always kick. Still a silly config to have, but
somehow that feels more correct.
While at it drop a comment which seems to be outdated
or confusing, and wrap the defer_count write with
a WRITE_ONCE() since it's read on the fast path
that avoids taking the lock.
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20220518185522.2038683-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
commit 68822bdf76 ("net: generalize skb freeing
deferral to per-cpu lists") added another per-cpu
cache of skbs. It was expected to be small,
and an IPI was forced whenever the list reached 128
skbs.
We might need to be able to control more precisely
queue capacity and added latency.
An IPI is generated whenever queue reaches half capacity.
Default value of the new limit is 64.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
