2009-02-24 15:30:29 +00:00
|
|
|
/*
|
2018-07-24 03:51:21 +00:00
|
|
|
* Copyright (c) 2007, 2017 Oracle and/or its affiliates. All rights reserved.
|
2009-02-24 15:30:29 +00:00
|
|
|
*
|
|
|
|
* This software is available to you under a choice of one of two
|
|
|
|
* licenses. You may choose to be licensed under the terms of the GNU
|
|
|
|
* General Public License (GPL) Version 2, available from the file
|
|
|
|
* COPYING in the main directory of this source tree, or the
|
|
|
|
* OpenIB.org BSD license below:
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or
|
|
|
|
* without modification, are permitted provided that the following
|
|
|
|
* conditions are met:
|
|
|
|
*
|
|
|
|
* - Redistributions of source code must retain the above
|
|
|
|
* copyright notice, this list of conditions and the following
|
|
|
|
* disclaimer.
|
|
|
|
*
|
|
|
|
* - Redistributions in binary form must reproduce the above
|
|
|
|
* copyright notice, this list of conditions and the following
|
|
|
|
* disclaimer in the documentation and/or other materials
|
|
|
|
* provided with the distribution.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
|
|
|
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
|
|
|
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
|
|
|
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
|
|
|
|
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
|
|
|
|
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
|
|
|
|
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
|
|
* SOFTWARE.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
#include <linux/pagemap.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>
|
2009-02-24 15:30:29 +00:00
|
|
|
#include <linux/rbtree.h>
|
|
|
|
#include <linux/dma-mapping.h> /* for DMA_*_DEVICE */
|
|
|
|
|
2010-01-12 20:57:27 +00:00
|
|
|
#include "rds.h"
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* XXX
|
|
|
|
* - build with sparse
|
|
|
|
* - should we detect duplicate keys on a socket? hmm.
|
|
|
|
* - an rdma is an mlock, apply rlimit?
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* get the number of pages by looking at the page indices that the start and
|
|
|
|
* end addresses fall in.
|
|
|
|
*
|
|
|
|
* Returns 0 if the vec is invalid. It is invalid if the number of bytes
|
|
|
|
* causes the address to wrap or overflows an unsigned int. This comes
|
|
|
|
* from being stored in the 'length' member of 'struct scatterlist'.
|
|
|
|
*/
|
|
|
|
static unsigned int rds_pages_in_vec(struct rds_iovec *vec)
|
|
|
|
{
|
|
|
|
if ((vec->addr + vec->bytes <= vec->addr) ||
|
|
|
|
(vec->bytes > (u64)UINT_MAX))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
return ((vec->addr + vec->bytes + PAGE_SIZE - 1) >> PAGE_SHIFT) -
|
|
|
|
(vec->addr >> PAGE_SHIFT);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct rds_mr *rds_mr_tree_walk(struct rb_root *root, u64 key,
|
|
|
|
struct rds_mr *insert)
|
|
|
|
{
|
|
|
|
struct rb_node **p = &root->rb_node;
|
|
|
|
struct rb_node *parent = NULL;
|
|
|
|
struct rds_mr *mr;
|
|
|
|
|
|
|
|
while (*p) {
|
|
|
|
parent = *p;
|
|
|
|
mr = rb_entry(parent, struct rds_mr, r_rb_node);
|
|
|
|
|
|
|
|
if (key < mr->r_key)
|
|
|
|
p = &(*p)->rb_left;
|
|
|
|
else if (key > mr->r_key)
|
|
|
|
p = &(*p)->rb_right;
|
|
|
|
else
|
|
|
|
return mr;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (insert) {
|
|
|
|
rb_link_node(&insert->r_rb_node, parent, p);
|
|
|
|
rb_insert_color(&insert->r_rb_node, root);
|
2017-07-04 12:53:17 +00:00
|
|
|
refcount_inc(&insert->r_refcount);
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Destroy the transport-specific part of a MR.
|
|
|
|
*/
|
|
|
|
static void rds_destroy_mr(struct rds_mr *mr)
|
|
|
|
{
|
|
|
|
struct rds_sock *rs = mr->r_sock;
|
|
|
|
void *trans_private = NULL;
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
rdsdebug("RDS: destroy mr key is %x refcnt %u\n",
|
2017-07-04 12:53:17 +00:00
|
|
|
mr->r_key, refcount_read(&mr->r_refcount));
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
if (test_and_set_bit(RDS_MR_DEAD, &mr->r_state))
|
|
|
|
return;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&rs->rs_rdma_lock, flags);
|
|
|
|
if (!RB_EMPTY_NODE(&mr->r_rb_node))
|
|
|
|
rb_erase(&mr->r_rb_node, &rs->rs_rdma_keys);
|
|
|
|
trans_private = mr->r_trans_private;
|
|
|
|
mr->r_trans_private = NULL;
|
|
|
|
spin_unlock_irqrestore(&rs->rs_rdma_lock, flags);
|
|
|
|
|
|
|
|
if (trans_private)
|
|
|
|
mr->r_trans->free_mr(trans_private, mr->r_invalidate);
|
|
|
|
}
|
|
|
|
|
|
|
|
void __rds_put_mr_final(struct rds_mr *mr)
|
|
|
|
{
|
|
|
|
rds_destroy_mr(mr);
|
|
|
|
kfree(mr);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* By the time this is called we can't have any more ioctls called on
|
|
|
|
* the socket so we don't need to worry about racing with others.
|
|
|
|
*/
|
|
|
|
void rds_rdma_drop_keys(struct rds_sock *rs)
|
|
|
|
{
|
|
|
|
struct rds_mr *mr;
|
|
|
|
struct rb_node *node;
|
2010-04-01 21:09:00 +00:00
|
|
|
unsigned long flags;
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
/* Release any MRs associated with this socket */
|
2010-04-01 21:09:00 +00:00
|
|
|
spin_lock_irqsave(&rs->rs_rdma_lock, flags);
|
2009-02-24 15:30:29 +00:00
|
|
|
while ((node = rb_first(&rs->rs_rdma_keys))) {
|
2016-12-20 14:02:18 +00:00
|
|
|
mr = rb_entry(node, struct rds_mr, r_rb_node);
|
2009-02-24 15:30:29 +00:00
|
|
|
if (mr->r_trans == rs->rs_transport)
|
|
|
|
mr->r_invalidate = 0;
|
2010-04-01 21:09:00 +00:00
|
|
|
rb_erase(&mr->r_rb_node, &rs->rs_rdma_keys);
|
|
|
|
RB_CLEAR_NODE(&mr->r_rb_node);
|
|
|
|
spin_unlock_irqrestore(&rs->rs_rdma_lock, flags);
|
|
|
|
rds_destroy_mr(mr);
|
2009-02-24 15:30:29 +00:00
|
|
|
rds_mr_put(mr);
|
2010-04-01 21:09:00 +00:00
|
|
|
spin_lock_irqsave(&rs->rs_rdma_lock, flags);
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
2010-04-01 21:09:00 +00:00
|
|
|
spin_unlock_irqrestore(&rs->rs_rdma_lock, flags);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
if (rs->rs_transport && rs->rs_transport->flush_mrs)
|
|
|
|
rs->rs_transport->flush_mrs();
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Helper function to pin user pages.
|
|
|
|
*/
|
|
|
|
static int rds_pin_pages(unsigned long user_addr, unsigned int nr_pages,
|
|
|
|
struct page **pages, int write)
|
|
|
|
{
|
2020-01-15 12:43:38 +00:00
|
|
|
unsigned int gup_flags = FOLL_LONGTERM;
|
2009-02-24 15:30:29 +00:00
|
|
|
int ret;
|
|
|
|
|
2020-01-15 12:43:38 +00:00
|
|
|
if (write)
|
|
|
|
gup_flags |= FOLL_WRITE;
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2020-01-15 12:43:38 +00:00
|
|
|
ret = get_user_pages_fast(user_addr, nr_pages, gup_flags, pages);
|
2009-04-09 14:09:40 +00:00
|
|
|
if (ret >= 0 && ret < nr_pages) {
|
2009-02-24 15:30:29 +00:00
|
|
|
while (ret--)
|
|
|
|
put_page(pages[ret]);
|
|
|
|
ret = -EFAULT;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int __rds_rdma_map(struct rds_sock *rs, struct rds_get_mr_args *args,
|
2018-07-25 03:31:58 +00:00
|
|
|
u64 *cookie_ret, struct rds_mr **mr_ret,
|
|
|
|
struct rds_conn_path *cp)
|
2009-02-24 15:30:29 +00:00
|
|
|
{
|
|
|
|
struct rds_mr *mr = NULL, *found;
|
|
|
|
unsigned int nr_pages;
|
|
|
|
struct page **pages = NULL;
|
|
|
|
struct scatterlist *sg;
|
|
|
|
void *trans_private;
|
|
|
|
unsigned long flags;
|
|
|
|
rds_rdma_cookie_t cookie;
|
|
|
|
unsigned int nents;
|
|
|
|
long i;
|
|
|
|
int ret;
|
|
|
|
|
2018-07-24 03:51:21 +00:00
|
|
|
if (ipv6_addr_any(&rs->rs_bound_addr) || !rs->rs_transport) {
|
2009-02-24 15:30:29 +00:00
|
|
|
ret = -ENOTCONN; /* XXX not a great errno */
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2010-01-12 19:56:44 +00:00
|
|
|
if (!rs->rs_transport->get_mr) {
|
2009-02-24 15:30:29 +00:00
|
|
|
ret = -EOPNOTSUPP;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
nr_pages = rds_pages_in_vec(&args->vec);
|
|
|
|
if (nr_pages == 0) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-02-29 23:30:57 +00:00
|
|
|
/* Restrict the size of mr irrespective of underlying transport
|
|
|
|
* To account for unaligned mr regions, subtract one from nr_pages
|
|
|
|
*/
|
|
|
|
if ((nr_pages - 1) > (RDS_MAX_MSG_SIZE >> PAGE_SHIFT)) {
|
|
|
|
ret = -EMSGSIZE;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2009-02-24 15:30:29 +00:00
|
|
|
rdsdebug("RDS: get_mr addr %llx len %llu nr_pages %u\n",
|
|
|
|
args->vec.addr, args->vec.bytes, nr_pages);
|
|
|
|
|
|
|
|
/* XXX clamp nr_pages to limit the size of this alloc? */
|
|
|
|
pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
|
2010-01-12 19:56:44 +00:00
|
|
|
if (!pages) {
|
2009-02-24 15:30:29 +00:00
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
mr = kzalloc(sizeof(struct rds_mr), GFP_KERNEL);
|
2010-01-12 19:56:44 +00:00
|
|
|
if (!mr) {
|
2009-02-24 15:30:29 +00:00
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2017-07-04 12:53:17 +00:00
|
|
|
refcount_set(&mr->r_refcount, 1);
|
2009-02-24 15:30:29 +00:00
|
|
|
RB_CLEAR_NODE(&mr->r_rb_node);
|
|
|
|
mr->r_trans = rs->rs_transport;
|
|
|
|
mr->r_sock = rs;
|
|
|
|
|
|
|
|
if (args->flags & RDS_RDMA_USE_ONCE)
|
|
|
|
mr->r_use_once = 1;
|
|
|
|
if (args->flags & RDS_RDMA_INVALIDATE)
|
|
|
|
mr->r_invalidate = 1;
|
|
|
|
if (args->flags & RDS_RDMA_READWRITE)
|
|
|
|
mr->r_write = 1;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Pin the pages that make up the user buffer and transfer the page
|
|
|
|
* pointers to the mr's sg array. We check to see if we've mapped
|
|
|
|
* the whole region after transferring the partial page references
|
|
|
|
* to the sg array so that we can have one page ref cleanup path.
|
|
|
|
*
|
|
|
|
* For now we have no flag that tells us whether the mapping is
|
|
|
|
* r/o or r/w. We need to assume r/w, or we'll do a lot of RDMA to
|
|
|
|
* the zero page.
|
|
|
|
*/
|
2010-01-12 18:52:28 +00:00
|
|
|
ret = rds_pin_pages(args->vec.addr, nr_pages, pages, 1);
|
2009-02-24 15:30:29 +00:00
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
nents = ret;
|
|
|
|
sg = kcalloc(nents, sizeof(*sg), GFP_KERNEL);
|
2010-01-12 19:56:44 +00:00
|
|
|
if (!sg) {
|
2009-02-24 15:30:29 +00:00
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
WARN_ON(!nents);
|
|
|
|
sg_init_table(sg, nents);
|
|
|
|
|
|
|
|
/* Stick all pages into the scatterlist */
|
|
|
|
for (i = 0 ; i < nents; i++)
|
|
|
|
sg_set_page(&sg[i], pages[i], PAGE_SIZE, 0);
|
|
|
|
|
|
|
|
rdsdebug("RDS: trans_private nents is %u\n", nents);
|
|
|
|
|
|
|
|
/* Obtain a transport specific MR. If this succeeds, the
|
|
|
|
* s/g list is now owned by the MR.
|
|
|
|
* Note that dma_map() implies that pending writes are
|
|
|
|
* flushed to RAM, so no dma_sync is needed here. */
|
|
|
|
trans_private = rs->rs_transport->get_mr(sg, nents, rs,
|
2018-07-25 03:31:58 +00:00
|
|
|
&mr->r_key,
|
|
|
|
cp ? cp->cp_conn : NULL);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
if (IS_ERR(trans_private)) {
|
|
|
|
for (i = 0 ; i < nents; i++)
|
|
|
|
put_page(sg_page(&sg[i]));
|
|
|
|
kfree(sg);
|
|
|
|
ret = PTR_ERR(trans_private);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
mr->r_trans_private = trans_private;
|
|
|
|
|
|
|
|
rdsdebug("RDS: get_mr put_user key is %x cookie_addr %p\n",
|
|
|
|
mr->r_key, (void *)(unsigned long) args->cookie_addr);
|
|
|
|
|
|
|
|
/* The user may pass us an unaligned address, but we can only
|
|
|
|
* map page aligned regions. So we keep the offset, and build
|
|
|
|
* a 64bit cookie containing <R_Key, offset> and pass that
|
|
|
|
* around. */
|
|
|
|
cookie = rds_rdma_make_cookie(mr->r_key, args->vec.addr & ~PAGE_MASK);
|
|
|
|
if (cookie_ret)
|
|
|
|
*cookie_ret = cookie;
|
|
|
|
|
|
|
|
if (args->cookie_addr && put_user(cookie, (u64 __user *)(unsigned long) args->cookie_addr)) {
|
|
|
|
ret = -EFAULT;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Inserting the new MR into the rbtree bumps its
|
|
|
|
* reference count. */
|
|
|
|
spin_lock_irqsave(&rs->rs_rdma_lock, flags);
|
|
|
|
found = rds_mr_tree_walk(&rs->rs_rdma_keys, mr->r_key, mr);
|
|
|
|
spin_unlock_irqrestore(&rs->rs_rdma_lock, flags);
|
|
|
|
|
|
|
|
BUG_ON(found && found != mr);
|
|
|
|
|
|
|
|
rdsdebug("RDS: get_mr key is %x\n", mr->r_key);
|
|
|
|
if (mr_ret) {
|
2017-07-04 12:53:17 +00:00
|
|
|
refcount_inc(&mr->r_refcount);
|
2009-02-24 15:30:29 +00:00
|
|
|
*mr_ret = mr;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = 0;
|
|
|
|
out:
|
|
|
|
kfree(pages);
|
|
|
|
if (mr)
|
|
|
|
rds_mr_put(mr);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int rds_get_mr(struct rds_sock *rs, char __user *optval, int optlen)
|
|
|
|
{
|
|
|
|
struct rds_get_mr_args args;
|
|
|
|
|
|
|
|
if (optlen != sizeof(struct rds_get_mr_args))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (copy_from_user(&args, (struct rds_get_mr_args __user *)optval,
|
|
|
|
sizeof(struct rds_get_mr_args)))
|
|
|
|
return -EFAULT;
|
|
|
|
|
2018-07-25 03:31:58 +00:00
|
|
|
return __rds_rdma_map(rs, &args, NULL, NULL, NULL);
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
2009-10-30 08:54:53 +00:00
|
|
|
int rds_get_mr_for_dest(struct rds_sock *rs, char __user *optval, int optlen)
|
|
|
|
{
|
|
|
|
struct rds_get_mr_for_dest_args args;
|
|
|
|
struct rds_get_mr_args new_args;
|
|
|
|
|
|
|
|
if (optlen != sizeof(struct rds_get_mr_for_dest_args))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (copy_from_user(&args, (struct rds_get_mr_for_dest_args __user *)optval,
|
|
|
|
sizeof(struct rds_get_mr_for_dest_args)))
|
|
|
|
return -EFAULT;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Initially, just behave like get_mr().
|
|
|
|
* TODO: Implement get_mr as wrapper around this
|
|
|
|
* and deprecate it.
|
|
|
|
*/
|
|
|
|
new_args.vec = args.vec;
|
|
|
|
new_args.cookie_addr = args.cookie_addr;
|
|
|
|
new_args.flags = args.flags;
|
|
|
|
|
2018-07-25 03:31:58 +00:00
|
|
|
return __rds_rdma_map(rs, &new_args, NULL, NULL, NULL);
|
2009-10-30 08:54:53 +00:00
|
|
|
}
|
|
|
|
|
2009-02-24 15:30:29 +00:00
|
|
|
/*
|
|
|
|
* Free the MR indicated by the given R_Key
|
|
|
|
*/
|
|
|
|
int rds_free_mr(struct rds_sock *rs, char __user *optval, int optlen)
|
|
|
|
{
|
|
|
|
struct rds_free_mr_args args;
|
|
|
|
struct rds_mr *mr;
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
if (optlen != sizeof(struct rds_free_mr_args))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (copy_from_user(&args, (struct rds_free_mr_args __user *)optval,
|
|
|
|
sizeof(struct rds_free_mr_args)))
|
|
|
|
return -EFAULT;
|
|
|
|
|
|
|
|
/* Special case - a null cookie means flush all unused MRs */
|
|
|
|
if (args.cookie == 0) {
|
|
|
|
if (!rs->rs_transport || !rs->rs_transport->flush_mrs)
|
|
|
|
return -EINVAL;
|
|
|
|
rs->rs_transport->flush_mrs();
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Look up the MR given its R_key and remove it from the rbtree
|
|
|
|
* so nobody else finds it.
|
|
|
|
* This should also prevent races with rds_rdma_unuse.
|
|
|
|
*/
|
|
|
|
spin_lock_irqsave(&rs->rs_rdma_lock, flags);
|
|
|
|
mr = rds_mr_tree_walk(&rs->rs_rdma_keys, rds_rdma_cookie_key(args.cookie), NULL);
|
|
|
|
if (mr) {
|
|
|
|
rb_erase(&mr->r_rb_node, &rs->rs_rdma_keys);
|
|
|
|
RB_CLEAR_NODE(&mr->r_rb_node);
|
|
|
|
if (args.flags & RDS_RDMA_INVALIDATE)
|
|
|
|
mr->r_invalidate = 1;
|
|
|
|
}
|
|
|
|
spin_unlock_irqrestore(&rs->rs_rdma_lock, flags);
|
|
|
|
|
|
|
|
if (!mr)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* call rds_destroy_mr() ourselves so that we're sure it's done by the time
|
|
|
|
* we return. If we let rds_mr_put() do it it might not happen until
|
|
|
|
* someone else drops their ref.
|
|
|
|
*/
|
|
|
|
rds_destroy_mr(mr);
|
|
|
|
rds_mr_put(mr);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is called when we receive an extension header that
|
|
|
|
* tells us this MR was used. It allows us to implement
|
|
|
|
* use_once semantics
|
|
|
|
*/
|
|
|
|
void rds_rdma_unuse(struct rds_sock *rs, u32 r_key, int force)
|
|
|
|
{
|
|
|
|
struct rds_mr *mr;
|
|
|
|
unsigned long flags;
|
|
|
|
int zot_me = 0;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&rs->rs_rdma_lock, flags);
|
|
|
|
mr = rds_mr_tree_walk(&rs->rs_rdma_keys, r_key, NULL);
|
2010-01-12 20:37:17 +00:00
|
|
|
if (!mr) {
|
2016-07-04 02:14:10 +00:00
|
|
|
pr_debug("rds: trying to unuse MR with unknown r_key %u!\n",
|
|
|
|
r_key);
|
2010-01-12 20:37:17 +00:00
|
|
|
spin_unlock_irqrestore(&rs->rs_rdma_lock, flags);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mr->r_use_once || force) {
|
2009-02-24 15:30:29 +00:00
|
|
|
rb_erase(&mr->r_rb_node, &rs->rs_rdma_keys);
|
|
|
|
RB_CLEAR_NODE(&mr->r_rb_node);
|
|
|
|
zot_me = 1;
|
2010-01-12 20:37:17 +00:00
|
|
|
}
|
2009-02-24 15:30:29 +00:00
|
|
|
spin_unlock_irqrestore(&rs->rs_rdma_lock, flags);
|
|
|
|
|
|
|
|
/* May have to issue a dma_sync on this memory region.
|
|
|
|
* Note we could avoid this if the operation was a RDMA READ,
|
|
|
|
* but at this point we can't tell. */
|
2010-01-12 20:37:17 +00:00
|
|
|
if (mr->r_trans->sync_mr)
|
|
|
|
mr->r_trans->sync_mr(mr->r_trans_private, DMA_FROM_DEVICE);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2010-01-12 20:37:17 +00:00
|
|
|
/* If the MR was marked as invalidate, this will
|
|
|
|
* trigger an async flush. */
|
2015-08-25 19:01:59 +00:00
|
|
|
if (zot_me) {
|
2010-01-12 20:37:17 +00:00
|
|
|
rds_destroy_mr(mr);
|
2015-08-25 19:01:59 +00:00
|
|
|
rds_mr_put(mr);
|
|
|
|
}
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
2010-03-01 22:11:53 +00:00
|
|
|
void rds_rdma_free_op(struct rm_rdma_op *ro)
|
2009-02-24 15:30:29 +00:00
|
|
|
{
|
|
|
|
unsigned int i;
|
|
|
|
|
2010-03-01 22:11:53 +00:00
|
|
|
for (i = 0; i < ro->op_nents; i++) {
|
|
|
|
struct page *page = sg_page(&ro->op_sg[i]);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
/* Mark page dirty if it was possibly modified, which
|
|
|
|
* is the case for a RDMA_READ which copies from remote
|
|
|
|
* to local memory */
|
2010-03-01 22:11:53 +00:00
|
|
|
if (!ro->op_write) {
|
2015-08-22 22:45:31 +00:00
|
|
|
WARN_ON(!page->mapping && irqs_disabled());
|
2009-02-24 15:30:29 +00:00
|
|
|
set_page_dirty(page);
|
2010-03-11 13:50:06 +00:00
|
|
|
}
|
2009-02-24 15:30:29 +00:00
|
|
|
put_page(page);
|
|
|
|
}
|
|
|
|
|
2010-03-01 22:11:53 +00:00
|
|
|
kfree(ro->op_notifier);
|
|
|
|
ro->op_notifier = NULL;
|
|
|
|
ro->op_active = 0;
|
2010-01-12 22:13:15 +00:00
|
|
|
}
|
|
|
|
|
2010-01-28 00:15:48 +00:00
|
|
|
void rds_atomic_free_op(struct rm_atomic_op *ao)
|
|
|
|
{
|
|
|
|
struct page *page = sg_page(ao->op_sg);
|
|
|
|
|
|
|
|
/* Mark page dirty if it was possibly modified, which
|
|
|
|
* is the case for a RDMA_READ which copies from remote
|
|
|
|
* to local memory */
|
|
|
|
set_page_dirty(page);
|
|
|
|
put_page(page);
|
|
|
|
|
|
|
|
kfree(ao->op_notifier);
|
|
|
|
ao->op_notifier = NULL;
|
|
|
|
ao->op_active = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-01-12 22:13:15 +00:00
|
|
|
/*
|
2010-10-28 15:40:58 +00:00
|
|
|
* Count the number of pages needed to describe an incoming iovec array.
|
2010-01-12 22:13:15 +00:00
|
|
|
*/
|
2010-10-28 15:40:58 +00:00
|
|
|
static int rds_rdma_pages(struct rds_iovec iov[], int nr_iovecs)
|
|
|
|
{
|
|
|
|
int tot_pages = 0;
|
|
|
|
unsigned int nr_pages;
|
|
|
|
unsigned int i;
|
|
|
|
|
|
|
|
/* figure out the number of pages in the vector */
|
|
|
|
for (i = 0; i < nr_iovecs; i++) {
|
|
|
|
nr_pages = rds_pages_in_vec(&iov[i]);
|
|
|
|
if (nr_pages == 0)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
tot_pages += nr_pages;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* nr_pages for one entry is limited to (UINT_MAX>>PAGE_SHIFT)+1,
|
|
|
|
* so tot_pages cannot overflow without first going negative.
|
|
|
|
*/
|
|
|
|
if (tot_pages < 0)
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return tot_pages;
|
|
|
|
}
|
|
|
|
|
2018-12-16 07:01:08 +00:00
|
|
|
int rds_rdma_extra_size(struct rds_rdma_args *args,
|
|
|
|
struct rds_iov_vector *iov)
|
2010-01-12 22:13:15 +00:00
|
|
|
{
|
2018-12-16 07:01:08 +00:00
|
|
|
struct rds_iovec *vec;
|
2010-01-12 22:13:15 +00:00
|
|
|
struct rds_iovec __user *local_vec;
|
2010-10-28 15:40:58 +00:00
|
|
|
int tot_pages = 0;
|
2010-01-12 22:13:15 +00:00
|
|
|
unsigned int nr_pages;
|
|
|
|
unsigned int i;
|
|
|
|
|
|
|
|
local_vec = (struct rds_iovec __user *)(unsigned long) args->local_vec_addr;
|
|
|
|
|
2018-01-02 19:44:34 +00:00
|
|
|
if (args->nr_local == 0)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2018-12-16 07:01:08 +00:00
|
|
|
iov->iov = kcalloc(args->nr_local,
|
|
|
|
sizeof(struct rds_iovec),
|
|
|
|
GFP_KERNEL);
|
|
|
|
if (!iov->iov)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
vec = &iov->iov[0];
|
|
|
|
|
|
|
|
if (copy_from_user(vec, local_vec, args->nr_local *
|
|
|
|
sizeof(struct rds_iovec)))
|
|
|
|
return -EFAULT;
|
|
|
|
iov->len = args->nr_local;
|
|
|
|
|
2010-01-12 22:13:15 +00:00
|
|
|
/* figure out the number of pages in the vector */
|
2018-12-16 07:01:08 +00:00
|
|
|
for (i = 0; i < args->nr_local; i++, vec++) {
|
2010-01-12 22:13:15 +00:00
|
|
|
|
2018-12-16 07:01:08 +00:00
|
|
|
nr_pages = rds_pages_in_vec(vec);
|
2010-01-12 22:13:15 +00:00
|
|
|
if (nr_pages == 0)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
tot_pages += nr_pages;
|
2010-10-28 15:40:55 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* nr_pages for one entry is limited to (UINT_MAX>>PAGE_SHIFT)+1,
|
|
|
|
* so tot_pages cannot overflow without first going negative.
|
|
|
|
*/
|
2010-10-28 15:40:58 +00:00
|
|
|
if (tot_pages < 0)
|
2010-10-28 15:40:55 +00:00
|
|
|
return -EINVAL;
|
2010-01-12 22:13:15 +00:00
|
|
|
}
|
|
|
|
|
2010-10-28 15:40:58 +00:00
|
|
|
return tot_pages * sizeof(struct scatterlist);
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2010-01-28 00:07:30 +00:00
|
|
|
* The application asks for a RDMA transfer.
|
|
|
|
* Extract all arguments and set up the rdma_op
|
2009-02-24 15:30:29 +00:00
|
|
|
*/
|
2010-01-28 00:07:30 +00:00
|
|
|
int rds_cmsg_rdma_args(struct rds_sock *rs, struct rds_message *rm,
|
2018-12-16 07:01:08 +00:00
|
|
|
struct cmsghdr *cmsg,
|
|
|
|
struct rds_iov_vector *vec)
|
2009-02-24 15:30:29 +00:00
|
|
|
{
|
2010-01-28 00:07:30 +00:00
|
|
|
struct rds_rdma_args *args;
|
2010-03-01 22:11:53 +00:00
|
|
|
struct rm_rdma_op *op = &rm->rdma;
|
2010-09-18 13:42:25 +00:00
|
|
|
int nr_pages;
|
2009-02-24 15:30:29 +00:00
|
|
|
unsigned int nr_bytes;
|
|
|
|
struct page **pages = NULL;
|
2018-12-16 07:01:08 +00:00
|
|
|
struct rds_iovec *iovs;
|
2009-02-24 15:30:29 +00:00
|
|
|
unsigned int i, j;
|
2010-01-12 22:13:15 +00:00
|
|
|
int ret = 0;
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2010-01-28 00:07:30 +00:00
|
|
|
if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct rds_rdma_args))
|
2010-03-01 22:11:53 +00:00
|
|
|
|| rm->rdma.op_active)
|
2010-01-28 00:07:30 +00:00
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
args = CMSG_DATA(cmsg);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2018-07-24 03:51:21 +00:00
|
|
|
if (ipv6_addr_any(&rs->rs_bound_addr)) {
|
2009-02-24 15:30:29 +00:00
|
|
|
ret = -ENOTCONN; /* XXX not a great errno */
|
2014-10-14 19:35:08 +00:00
|
|
|
goto out_ret;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
2010-11-17 06:37:16 +00:00
|
|
|
if (args->nr_local > UIO_MAXIOV) {
|
2009-02-24 15:30:29 +00:00
|
|
|
ret = -EMSGSIZE;
|
2014-10-14 19:35:08 +00:00
|
|
|
goto out_ret;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
2018-12-16 07:01:08 +00:00
|
|
|
if (vec->len != args->nr_local) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out_ret;
|
2010-10-28 15:40:58 +00:00
|
|
|
}
|
|
|
|
|
2018-12-16 07:01:08 +00:00
|
|
|
iovs = vec->iov;
|
2010-10-28 15:40:58 +00:00
|
|
|
|
|
|
|
nr_pages = rds_rdma_pages(iovs, args->nr_local);
|
2010-10-28 15:40:56 +00:00
|
|
|
if (nr_pages < 0) {
|
|
|
|
ret = -EINVAL;
|
2018-12-16 07:01:08 +00:00
|
|
|
goto out_ret;
|
2010-10-28 15:40:56 +00:00
|
|
|
}
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2010-01-12 22:13:15 +00:00
|
|
|
pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
|
|
|
|
if (!pages) {
|
2009-02-24 15:30:29 +00:00
|
|
|
ret = -ENOMEM;
|
2018-12-16 07:01:08 +00:00
|
|
|
goto out_ret;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
2010-03-01 22:11:53 +00:00
|
|
|
op->op_write = !!(args->flags & RDS_RDMA_READWRITE);
|
|
|
|
op->op_fence = !!(args->flags & RDS_RDMA_FENCE);
|
|
|
|
op->op_notify = !!(args->flags & RDS_RDMA_NOTIFY_ME);
|
2010-03-02 00:10:40 +00:00
|
|
|
op->op_silent = !!(args->flags & RDS_RDMA_SILENT);
|
2010-03-01 22:11:53 +00:00
|
|
|
op->op_active = 1;
|
|
|
|
op->op_recverr = rs->rs_recverr;
|
2009-02-24 15:30:29 +00:00
|
|
|
WARN_ON(!nr_pages);
|
2018-12-16 07:01:09 +00:00
|
|
|
op->op_sg = rds_message_alloc_sgs(rm, nr_pages, &ret);
|
|
|
|
if (!op->op_sg)
|
2018-12-16 07:01:08 +00:00
|
|
|
goto out_pages;
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2010-03-01 22:11:53 +00:00
|
|
|
if (op->op_notify || op->op_recverr) {
|
2009-02-24 15:30:29 +00:00
|
|
|
/* We allocate an uninitialized notifier here, because
|
|
|
|
* we don't want to do that in the completion handler. We
|
|
|
|
* would have to use GFP_ATOMIC there, and don't want to deal
|
|
|
|
* with failed allocations.
|
|
|
|
*/
|
2010-03-01 22:11:53 +00:00
|
|
|
op->op_notifier = kmalloc(sizeof(struct rds_notifier), GFP_KERNEL);
|
|
|
|
if (!op->op_notifier) {
|
2009-02-24 15:30:29 +00:00
|
|
|
ret = -ENOMEM;
|
2018-12-16 07:01:08 +00:00
|
|
|
goto out_pages;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
2010-03-01 22:11:53 +00:00
|
|
|
op->op_notifier->n_user_token = args->user_token;
|
|
|
|
op->op_notifier->n_status = RDS_RDMA_SUCCESS;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* The cookie contains the R_Key of the remote memory region, and
|
|
|
|
* optionally an offset into it. This is how we implement RDMA into
|
|
|
|
* unaligned memory.
|
|
|
|
* When setting up the RDMA, we need to add that offset to the
|
|
|
|
* destination address (which is really an offset into the MR)
|
|
|
|
* FIXME: We may want to move this into ib_rdma.c
|
|
|
|
*/
|
2010-03-01 22:11:53 +00:00
|
|
|
op->op_rkey = rds_rdma_cookie_key(args->cookie);
|
|
|
|
op->op_remote_addr = args->remote_vec.addr + rds_rdma_cookie_offset(args->cookie);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
nr_bytes = 0;
|
|
|
|
|
|
|
|
rdsdebug("RDS: rdma prepare nr_local %llu rva %llx rkey %x\n",
|
|
|
|
(unsigned long long)args->nr_local,
|
|
|
|
(unsigned long long)args->remote_vec.addr,
|
2010-03-01 22:11:53 +00:00
|
|
|
op->op_rkey);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
for (i = 0; i < args->nr_local; i++) {
|
2010-10-28 15:40:58 +00:00
|
|
|
struct rds_iovec *iov = &iovs[i];
|
|
|
|
/* don't need to check, rds_rdma_pages() verified nr will be +nonzero */
|
|
|
|
unsigned int nr = rds_pages_in_vec(iov);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2010-10-28 15:40:58 +00:00
|
|
|
rs->rs_user_addr = iov->addr;
|
|
|
|
rs->rs_user_bytes = iov->bytes;
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
/* If it's a WRITE operation, we want to pin the pages for reading.
|
|
|
|
* If it's a READ operation, we need to pin the pages for writing.
|
|
|
|
*/
|
2010-10-28 15:40:58 +00:00
|
|
|
ret = rds_pin_pages(iov->addr, nr, pages, !op->op_write);
|
2009-02-24 15:30:29 +00:00
|
|
|
if (ret < 0)
|
2018-12-16 07:01:08 +00:00
|
|
|
goto out_pages;
|
2015-08-22 22:45:22 +00:00
|
|
|
else
|
|
|
|
ret = 0;
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2010-10-28 15:40:58 +00:00
|
|
|
rdsdebug("RDS: nr_bytes %u nr %u iov->bytes %llu iov->addr %llx\n",
|
|
|
|
nr_bytes, nr, iov->bytes, iov->addr);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2010-10-28 15:40:58 +00:00
|
|
|
nr_bytes += iov->bytes;
|
2009-02-24 15:30:29 +00:00
|
|
|
|
|
|
|
for (j = 0; j < nr; j++) {
|
2010-10-28 15:40:58 +00:00
|
|
|
unsigned int offset = iov->addr & ~PAGE_MASK;
|
2010-01-12 22:13:15 +00:00
|
|
|
struct scatterlist *sg;
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2010-03-01 22:11:53 +00:00
|
|
|
sg = &op->op_sg[op->op_nents + j];
|
2009-02-24 15:30:29 +00:00
|
|
|
sg_set_page(sg, pages[j],
|
2010-10-28 15:40:58 +00:00
|
|
|
min_t(unsigned int, iov->bytes, PAGE_SIZE - offset),
|
2009-02-24 15:30:29 +00:00
|
|
|
offset);
|
|
|
|
|
2010-10-28 15:40:58 +00:00
|
|
|
rdsdebug("RDS: sg->offset %x sg->len %x iov->addr %llx iov->bytes %llu\n",
|
|
|
|
sg->offset, sg->length, iov->addr, iov->bytes);
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2010-10-28 15:40:58 +00:00
|
|
|
iov->addr += sg->length;
|
|
|
|
iov->bytes -= sg->length;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
2010-03-01 22:11:53 +00:00
|
|
|
op->op_nents += nr;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
if (nr_bytes > args->remote_vec.bytes) {
|
|
|
|
rdsdebug("RDS nr_bytes %u remote_bytes %u do not match\n",
|
|
|
|
nr_bytes,
|
|
|
|
(unsigned int) args->remote_vec.bytes);
|
|
|
|
ret = -EINVAL;
|
2018-12-16 07:01:08 +00:00
|
|
|
goto out_pages;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
2010-03-01 22:11:53 +00:00
|
|
|
op->op_bytes = nr_bytes;
|
2009-02-24 15:30:29 +00:00
|
|
|
|
2018-12-16 07:01:08 +00:00
|
|
|
out_pages:
|
2009-02-24 15:30:29 +00:00
|
|
|
kfree(pages);
|
2014-10-14 19:35:08 +00:00
|
|
|
out_ret:
|
2010-01-12 22:13:15 +00:00
|
|
|
if (ret)
|
|
|
|
rds_rdma_free_op(op);
|
2010-10-28 15:40:57 +00:00
|
|
|
else
|
|
|
|
rds_stats_inc(s_send_rdma);
|
2010-01-28 00:07:30 +00:00
|
|
|
|
|
|
|
return ret;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The application wants us to pass an RDMA destination (aka MR)
|
|
|
|
* to the remote
|
|
|
|
*/
|
|
|
|
int rds_cmsg_rdma_dest(struct rds_sock *rs, struct rds_message *rm,
|
|
|
|
struct cmsghdr *cmsg)
|
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
struct rds_mr *mr;
|
|
|
|
u32 r_key;
|
|
|
|
int err = 0;
|
|
|
|
|
2009-11-30 00:55:45 +00:00
|
|
|
if (cmsg->cmsg_len < CMSG_LEN(sizeof(rds_rdma_cookie_t)) ||
|
|
|
|
rm->m_rdma_cookie != 0)
|
2009-02-24 15:30:29 +00:00
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
memcpy(&rm->m_rdma_cookie, CMSG_DATA(cmsg), sizeof(rm->m_rdma_cookie));
|
|
|
|
|
|
|
|
/* We are reusing a previously mapped MR here. Most likely, the
|
|
|
|
* application has written to the buffer, so we need to explicitly
|
|
|
|
* flush those writes to RAM. Otherwise the HCA may not see them
|
|
|
|
* when doing a DMA from that buffer.
|
|
|
|
*/
|
|
|
|
r_key = rds_rdma_cookie_key(rm->m_rdma_cookie);
|
|
|
|
|
|
|
|
spin_lock_irqsave(&rs->rs_rdma_lock, flags);
|
|
|
|
mr = rds_mr_tree_walk(&rs->rs_rdma_keys, r_key, NULL);
|
2010-01-12 19:56:44 +00:00
|
|
|
if (!mr)
|
2009-02-24 15:30:29 +00:00
|
|
|
err = -EINVAL; /* invalid r_key */
|
|
|
|
else
|
2017-07-04 12:53:17 +00:00
|
|
|
refcount_inc(&mr->r_refcount);
|
2009-02-24 15:30:29 +00:00
|
|
|
spin_unlock_irqrestore(&rs->rs_rdma_lock, flags);
|
|
|
|
|
|
|
|
if (mr) {
|
|
|
|
mr->r_trans->sync_mr(mr->r_trans_private, DMA_TO_DEVICE);
|
2010-03-01 22:11:53 +00:00
|
|
|
rm->rdma.op_rdma_mr = mr;
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The application passes us an address range it wants to enable RDMA
|
|
|
|
* to/from. We map the area, and save the <R_Key,offset> pair
|
|
|
|
* in rm->m_rdma_cookie. This causes it to be sent along to the peer
|
|
|
|
* in an extension header.
|
|
|
|
*/
|
|
|
|
int rds_cmsg_rdma_map(struct rds_sock *rs, struct rds_message *rm,
|
|
|
|
struct cmsghdr *cmsg)
|
|
|
|
{
|
2009-11-30 00:55:45 +00:00
|
|
|
if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct rds_get_mr_args)) ||
|
|
|
|
rm->m_rdma_cookie != 0)
|
2009-02-24 15:30:29 +00:00
|
|
|
return -EINVAL;
|
|
|
|
|
2018-07-25 03:31:58 +00:00
|
|
|
return __rds_rdma_map(rs, CMSG_DATA(cmsg), &rm->m_rdma_cookie,
|
|
|
|
&rm->rdma.op_rdma_mr, rm->m_conn_path);
|
2009-02-24 15:30:29 +00:00
|
|
|
}
|
2010-01-12 22:33:38 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Fill in rds_message for an atomic request.
|
|
|
|
*/
|
|
|
|
int rds_cmsg_atomic(struct rds_sock *rs, struct rds_message *rm,
|
|
|
|
struct cmsghdr *cmsg)
|
|
|
|
{
|
|
|
|
struct page *page = NULL;
|
|
|
|
struct rds_atomic_args *args;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct rds_atomic_args))
|
|
|
|
|| rm->atomic.op_active)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
args = CMSG_DATA(cmsg);
|
|
|
|
|
2010-08-25 12:51:28 +00:00
|
|
|
/* Nonmasked & masked cmsg ops converted to masked hw ops */
|
|
|
|
switch (cmsg->cmsg_type) {
|
|
|
|
case RDS_CMSG_ATOMIC_FADD:
|
|
|
|
rm->atomic.op_type = RDS_ATOMIC_TYPE_FADD;
|
|
|
|
rm->atomic.op_m_fadd.add = args->fadd.add;
|
|
|
|
rm->atomic.op_m_fadd.nocarry_mask = 0;
|
|
|
|
break;
|
|
|
|
case RDS_CMSG_MASKED_ATOMIC_FADD:
|
2010-01-12 22:33:38 +00:00
|
|
|
rm->atomic.op_type = RDS_ATOMIC_TYPE_FADD;
|
2010-08-25 12:51:28 +00:00
|
|
|
rm->atomic.op_m_fadd.add = args->m_fadd.add;
|
|
|
|
rm->atomic.op_m_fadd.nocarry_mask = args->m_fadd.nocarry_mask;
|
|
|
|
break;
|
|
|
|
case RDS_CMSG_ATOMIC_CSWP:
|
|
|
|
rm->atomic.op_type = RDS_ATOMIC_TYPE_CSWP;
|
|
|
|
rm->atomic.op_m_cswp.compare = args->cswp.compare;
|
|
|
|
rm->atomic.op_m_cswp.swap = args->cswp.swap;
|
|
|
|
rm->atomic.op_m_cswp.compare_mask = ~0;
|
|
|
|
rm->atomic.op_m_cswp.swap_mask = ~0;
|
|
|
|
break;
|
|
|
|
case RDS_CMSG_MASKED_ATOMIC_CSWP:
|
|
|
|
rm->atomic.op_type = RDS_ATOMIC_TYPE_CSWP;
|
|
|
|
rm->atomic.op_m_cswp.compare = args->m_cswp.compare;
|
|
|
|
rm->atomic.op_m_cswp.swap = args->m_cswp.swap;
|
|
|
|
rm->atomic.op_m_cswp.compare_mask = args->m_cswp.compare_mask;
|
|
|
|
rm->atomic.op_m_cswp.swap_mask = args->m_cswp.swap_mask;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
BUG(); /* should never happen */
|
2010-01-12 22:33:38 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
rm->atomic.op_notify = !!(args->flags & RDS_RDMA_NOTIFY_ME);
|
2010-03-02 00:10:40 +00:00
|
|
|
rm->atomic.op_silent = !!(args->flags & RDS_RDMA_SILENT);
|
2010-03-02 00:04:59 +00:00
|
|
|
rm->atomic.op_active = 1;
|
2010-01-12 22:33:38 +00:00
|
|
|
rm->atomic.op_recverr = rs->rs_recverr;
|
2018-12-16 07:01:09 +00:00
|
|
|
rm->atomic.op_sg = rds_message_alloc_sgs(rm, 1, &ret);
|
|
|
|
if (!rm->atomic.op_sg)
|
2010-10-28 15:40:59 +00:00
|
|
|
goto err;
|
2010-01-12 22:33:38 +00:00
|
|
|
|
|
|
|
/* verify 8 byte-aligned */
|
|
|
|
if (args->local_addr & 0x7) {
|
|
|
|
ret = -EFAULT;
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = rds_pin_pages(args->local_addr, 1, &page, 1);
|
|
|
|
if (ret != 1)
|
|
|
|
goto err;
|
|
|
|
ret = 0;
|
|
|
|
|
|
|
|
sg_set_page(rm->atomic.op_sg, page, 8, offset_in_page(args->local_addr));
|
|
|
|
|
|
|
|
if (rm->atomic.op_notify || rm->atomic.op_recverr) {
|
|
|
|
/* We allocate an uninitialized notifier here, because
|
|
|
|
* we don't want to do that in the completion handler. We
|
|
|
|
* would have to use GFP_ATOMIC there, and don't want to deal
|
|
|
|
* with failed allocations.
|
|
|
|
*/
|
|
|
|
rm->atomic.op_notifier = kmalloc(sizeof(*rm->atomic.op_notifier), GFP_KERNEL);
|
|
|
|
if (!rm->atomic.op_notifier) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
rm->atomic.op_notifier->n_user_token = args->user_token;
|
|
|
|
rm->atomic.op_notifier->n_status = RDS_RDMA_SUCCESS;
|
|
|
|
}
|
|
|
|
|
2010-01-12 18:50:48 +00:00
|
|
|
rm->atomic.op_rkey = rds_rdma_cookie_key(args->cookie);
|
2010-01-12 22:33:38 +00:00
|
|
|
rm->atomic.op_remote_addr = args->remote_addr + rds_rdma_cookie_offset(args->cookie);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
err:
|
|
|
|
if (page)
|
|
|
|
put_page(page);
|
2018-01-03 21:06:06 +00:00
|
|
|
rm->atomic.op_active = 0;
|
2010-01-12 22:33:38 +00:00
|
|
|
kfree(rm->atomic.op_notifier);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|