Stream protocols such as TCP can often build up a backlog of data to be
read due to ordering. Combine this with the fact that some workloads such
as NFS read()-intensive workloads need to receive a lot of data per RPC
call, and it turns out that receiving the data from inside a softirq
context can cause starvation.
The following patch moves the TCP data receive into a workqueue context.
We still end up calling tcp_read_sock(), but we do so from a process
context, meaning that softirqs are enabled for most of the time.
With this patch, I see a doubling of read bandwidth when running a
multi-threaded iozone workload between a virtual client and server setup.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Move the TCP data receive loop out of xs_tcp_data_ready(). Doing so
will allow us to move the data receive out of the softirq context in
a set of followup patches.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Under all conditions, it should be quite sufficient just to mark
the socket as disconnected. It will then be closed by the
transport shutdown or reconnect code.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Commit 718ba5b873, moved the responsibility for unlocking the socket to
xs_tcp_setup_socket, meaning that the socket will be unlocked before we
know that it has finished trying to connect. The following patch is based on
an initial patch by Russell King to ensure that we delay clearing the
XPRT_CONNECTING flag until we either know that we failed to initiate
a connection attempt, or the connection attempt itself failed.
Fixes: 718ba5b873 ("SUNRPC: Add helpers to prevent socket create from racing")
Reported-by: Russell King <linux@arm.linux.org.uk>
Reported-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Benjamin Coddington <bcodding@redhat.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
When we're destroying the socket transport, we need to ensure that
we cancel any existing delayed connection attempts, and order them
w.r.t. the call to xs_close().
Reported-by:"Suzuki K. Poulose" <suzuki.poulose@arm.com>
Acked-by: Jeff Layton <jlayton@poochiereds.net>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Add a shutdown() call before we release the socket in order to ensure the
reset is sent before we try to reconnect.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Follow up to commit c4a7ca7749 ("SUNRPC: Allow waiting on memory
allocation"). Allows the RPC socket code to do non-IO blocking.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
It is rather pointless to test the value of transport->inet after
calling xs_reset_transport(), since it will always be zero, and
so we will never see any exponential back off behaviour.
Also don't force early connections for SOFTCONN tasks. If the server
disconnects us, we should respect the exponential backoff.
Cc: stable@vger.kernel.org # 4.0+
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Highlights include:
Stable patches:
- Fix a situation where the client uses the wrong (zero) stateid.
- Fix a memory leak in nfs_do_recoalesce
Bugfixes:
- Plug a memory leak when ->prepare_layoutcommit fails
- Fix an Oops in the NFSv4 open code
- Fix a backchannel deadlock
- Fix a livelock in sunrpc when sendmsg fails due to low memory availability
- Don't revalidate the mapping if both size and change attr are up to date
- Ensure we don't miss a file extension when doing pNFS
- Several fixes to handle NFSv4.1 sequence operation status bits correctly
- Several pNFS layout return bugfixes
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=wlUD
-----END PGP SIGNATURE-----
Merge tag 'nfs-for-4.2-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Pull NFS client bugfixes from Trond Myklebust:
"Highlights include:
Stable patches:
- Fix a situation where the client uses the wrong (zero) stateid.
- Fix a memory leak in nfs_do_recoalesce
Bugfixes:
- Plug a memory leak when ->prepare_layoutcommit fails
- Fix an Oops in the NFSv4 open code
- Fix a backchannel deadlock
- Fix a livelock in sunrpc when sendmsg fails due to low memory
availability
- Don't revalidate the mapping if both size and change attr are up to
date
- Ensure we don't miss a file extension when doing pNFS
- Several fixes to handle NFSv4.1 sequence operation status bits
correctly
- Several pNFS layout return bugfixes"
* tag 'nfs-for-4.2-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (28 commits)
nfs: Fix an oops caused by using other thread's stack space in ASYNC mode
nfs: plug memory leak when ->prepare_layoutcommit fails
SUNRPC: Report TCP errors to the caller
sunrpc: translate -EAGAIN to -ENOBUFS when socket is writable.
NFSv4.2: handle NFS-specific llseek errors
NFS: Don't clear desc->pg_moreio in nfs_do_recoalesce()
NFS: Fix a memory leak in nfs_do_recoalesce
NFS: nfs_mark_for_revalidate should always set NFS_INO_REVAL_PAGECACHE
NFS: Remove the "NFS_CAP_CHANGE_ATTR" capability
NFS: Set NFS_INO_REVAL_PAGECACHE if the change attribute is uninitialised
NFS: Don't revalidate the mapping if both size and change attr are up to date
NFSv4/pnfs: Ensure we don't miss a file extension
NFSv4: We must set NFS_OPEN_STATE flag in nfs_resync_open_stateid_locked
SUNRPC: xprt_complete_bc_request must also decrement the free slot count
SUNRPC: Fix a backchannel deadlock
pNFS: Don't throw out valid layout segments
pNFS: pnfs_roc_drain() fix a race with open
pNFS: Fix races between return-on-close and layoutreturn.
pNFS: pnfs_roc_drain should return 'true' when sleeping
pNFS: Layoutreturn must invalidate all existing layout segments.
...
The networking layer does not reliably report the distinction between
a non-block write failing because:
1/ the queue is too full already and
2/ a memory allocation attempt failed.
The distinction is important because in the first case it is
appropriate to retry as soon as the socket reports that it is
writable, and in the second case a small delay is required as the
socket will most likely report as writable but kmalloc could still
fail.
sk_stream_wait_memory() exhibits this distinction nicely, setting
'vm_wait' if a small wait is needed. However in the non-blocking case
it always returns -EAGAIN no matter the cause of the failure. This
-EAGAIN call get all the way to sunrpc.
The sunrpc layer expects EAGAIN to indicate the first cause, and
ENOBUFS to indicate the second. Various documentation suggests that
this is not unreasonable, but does not guarantee the desired error
codes.
The result of getting -EAGAIN when -ENOBUFS is expected is that the
send is tried again in a tight loop and soft lockups are reported.
so: add tests after calls to xs_sendpages() to translate -EAGAIN into
-ENOBUFS if the socket is writable. This cannot happen inside
xs_sendpages() as the test for "is socket writable" is different
between TCP and UDP.
With this change, the tight loop retrying xs_sendpages() becomes a
loop which only retries every 250ms, and so will not trigger a
soft-lockup warning.
It is possible that the write did fail because the queue was too full
and by the time xs_sendpages() completed, the queue was writable
again. In this case an extra 250ms delay is inserted that isn't
really needed. This circumstance suggests a degree of congestion so a
delay is not necessarily a bad thing, and it can only cause a single
250ms delay, not a series of them.
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
ENOBUFS means that memory allocations are failing due to an actual
low memory situation. It should not be confused with being out of
socket buffer space.
Handle the problem by just punting to the delay in call_status.
Reported-by: Neil Brown <neilb@suse.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Highlights include:
Stable patches:
- Fix a crash in the NFSv4 file locking code.
- Fix an fsync() regression, where we were failing to retry I/O in some
circumstances.
- Fix an infinite loop in NFSv4.0 OPEN stateid recovery
- Fix a memory leak when an attempted pnfs fails.
- Fix a memory leak in the backchannel code
- Large hostnames were not supported correctly in NFSv4.1
- Fix a pNFS/flexfiles bug that was impeding error reporting on I/O.
- Fix a couple of credential issues in pNFS/flexfiles
Bugfixes + cleanups:
- Open flag sanity checks in the NFSv4 atomic open codepath
- More NFSv4 delegation related bugfixes
- Various NFSv4.1 backchannel bugfixes and cleanups
- Fix the NFS swap socket code
- Various cleanups of the NFSv4 SETCLIENTID and EXCHANGE_ID code
- Fix a UDP transport deadlock issue
Features:
- More RDMA client transport improvements
- NFSv4.2 LAYOUTSTATS functionality for pnfs flexfiles.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=AREr
-----END PGP SIGNATURE-----
Merge tag 'nfs-for-4.2-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Pull NFS client updates from Trond Myklebust:
"Highlights include:
Stable patches:
- Fix a crash in the NFSv4 file locking code.
- Fix an fsync() regression, where we were failing to retry I/O in
some circumstances.
- Fix an infinite loop in NFSv4.0 OPEN stateid recovery
- Fix a memory leak when an attempted pnfs fails.
- Fix a memory leak in the backchannel code
- Large hostnames were not supported correctly in NFSv4.1
- Fix a pNFS/flexfiles bug that was impeding error reporting on I/O.
- Fix a couple of credential issues in pNFS/flexfiles
Bugfixes + cleanups:
- Open flag sanity checks in the NFSv4 atomic open codepath
- More NFSv4 delegation related bugfixes
- Various NFSv4.1 backchannel bugfixes and cleanups
- Fix the NFS swap socket code
- Various cleanups of the NFSv4 SETCLIENTID and EXCHANGE_ID code
- Fix a UDP transport deadlock issue
Features:
- More RDMA client transport improvements
- NFSv4.2 LAYOUTSTATS functionality for pnfs flexfiles"
* tag 'nfs-for-4.2-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (87 commits)
nfs: Remove invalid tk_pid from debug message
nfs: Remove invalid NFS_ATTR_FATTR_V4_REFERRAL checking in nfs4_get_rootfh
nfs: Drop bad comment in nfs41_walk_client_list()
nfs: Remove unneeded micro checking of CONFIG_PROC_FS
nfs: Don't setting FILE_CREATED flags always
nfs: Use remove_proc_subtree() instead remove_proc_entry()
nfs: Remove unused argument in nfs_server_set_fsinfo()
nfs: Fix a memory leak when meeting an unsupported state protect
nfs: take extra reference to fl->fl_file when running a LOCKU operation
NFSv4: When returning a delegation, don't reclaim an incompatible open mode.
NFSv4.2: LAYOUTSTATS is optional to implement
NFSv4.2: Fix up a decoding error in layoutstats
pNFS/flexfiles: Fix the reset of struct pgio_header when resending
pNFS/flexfiles: Turn off layoutcommit for servers that don't need it
pnfs/flexfiles: protect ktime manipulation with mirror lock
nfs: provide pnfs_report_layoutstat when NFS42 is disabled
nfs: verify open flags before allowing open
nfs: always update creds in mirror, even when we have an already connected ds
nfs: fix potential credential leak in ff_layout_update_mirror_cred
pnfs/flexfiles: report layoutstat regularly
...
Use the TCP_USER_TIMEOUT socket option to advertise to the server
how long we will keep the connection open if there is unacknowledged
data. See RFC5482.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
This fixes a regression introduced by commit caf4ccd4e8 ("SUNRPC:
Make xs_tcp_close() do a socket shutdown rather than a sock_release").
Prior to that commit, the autoclose feature would ensure that an
idle connection would result in the socket being both disconnected and
released, whereas now only gets disconnected.
While the current behaviour is harmless, it does leave the port bound
until either RPC traffic resumes or the RPC client is shut down.
Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
It has been exceptionally useful to exercise the logic that handles
local immediate errors and RDMA connection loss. To enable
developers to test this regularly and repeatably, add logic to
simulate connection loss every so often.
Fault injection is disabled by default. It is enabled with
$ sudo echo xxx > /sys/kernel/debug/sunrpc/inject_fault/disconnect
where "xxx" is a large positive number of transport method calls
before a disconnect. A value of several thousand is usually a good
number that allows reasonable forward progress while still causing a
lot of connection drops.
These hooks are disabled when SUNRPC_DEBUG is turned off.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
RDMA xprts don't have a sock_xprt, but an rdma_xprt, so the
xs_swapper_enable/disable functions will likely oops when fed an RDMA
xprt. Turn these functions into rpc_xprt_ops so that that doesn't
occur. For now the RDMA versions are no-ops that just return -EINVAL
on an attempt to swapon.
Cc: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
It's possible that we could race with a call to xs_reset_transport, in
which case the xprt->inet pointer could be zeroed out while we're
accessing it. Lock the xprt before we try to set memalloc on it.
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
We currently increment the memalloc_socks counter if we have a xprt that
is associated with a swapfile. That socket can be replaced however
during a reconnect event, and the memalloc_socks counter is never
decremented if that occurs.
When tearing down a xprt socket, check to see if the xprt is set up for
swapping and sk_clear_memalloc before releasing the socket if so.
Acked-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Split xs_swapper into enable/disable functions and eliminate the
"enable" flag.
Currently, it's racy if you have multiple swapon/swapoff operations
running in parallel over the same xprt. Also fix it so that we only
set it to a memalloc socket on a 0->1 transition and only clear it
on a 1->0 transition.
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Jerome reported seeing a warning pop when working with a swapfile on
NFS. The nfs_swap_activate can end up calling sk_set_memalloc while
holding the rcu_read_lock and that function can sleep.
To fix that, we need to take a reference to the xprt while holding the
rcu_read_lock, set the socket up for swapping and then drop that
reference. But, xprt_put is not exported and having NFS deal with the
underlying xprt is a bit of layering violation anyway.
Fix this by adding a set of activate/deactivate functions that take a
rpc_clnt pointer instead of an rpc_xprt, and have nfs_swap_activate and
nfs_swap_deactivate call those.
Also, add a per-rpc_clnt atomic counter to keep track of the number of
active swapfiles associated with it. When the counter does a 0->1
transition, we enable swapping on the xprt, when we do a 1->0 transition
we disable swapping on it.
This also allows us to be a bit more selective with the RPC_TASK_SWAPPER
flag. If non-swapper and swapper clnts are sharing a xprt, then we only
need to flag the tasks from the swapper clnt with that flag.
Acked-by: Mel Gorman <mgorman@suse.de>
Reported-by: Jerome Marchand <jmarchan@redhat.com>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Several functions have outdated arguments listed in the doc comments.
Drop documentation for arguments that no longer exist.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Most code already uses consts for the struct kernel_param_ops,
sweep the kernel for the last offending stragglers. Other than
include/linux/moduleparam.h and kernel/params.c all other changes
were generated with the following Coccinelle SmPL patch. Merge
conflicts between trees can be handled with Coccinelle.
In the future git could get Coccinelle merge support to deal with
patch --> fail --> grammar --> Coccinelle --> new patch conflicts
automatically for us on patches where the grammar is available and
the patch is of high confidence. Consider this a feature request.
Test compiled on x86_64 against:
* allnoconfig
* allmodconfig
* allyesconfig
@ const_found @
identifier ops;
@@
const struct kernel_param_ops ops = {
};
@ const_not_found depends on !const_found @
identifier ops;
@@
-struct kernel_param_ops ops = {
+const struct kernel_param_ops ops = {
};
Generated-by: Coccinelle SmPL
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Junio C Hamano <gitster@pobox.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: cocci@systeme.lip6.fr
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
xs_tcp_close() is now just a call to xs_tcp_shutdown(), so remove it,
and replace the entry in xs_tcp_ops.
Suggested-by: Anna Schumaker <anna.schumaker@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Now that the linger code is gone, the xs_tcp_fin_timeout variable has
no real function. Keep it for now, since it is part of the /proc
interface, but only define it if that /proc interface is enabled.
Suggested-by: Anna Schumaker <Anna.Schumaker@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
If the connection reset is due to an active call on our side, then
the state change is sometimes not reported. Catch those instances
using xs_error_report() instead.
Also remove the xs_tcp_shutdown() call in xs_tcp_send_request() as
the change in behaviour makes it redundant.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Use of socket shutdown() means that we monitor the shutdown process
through the xs_tcp_state_change() callback, so it is preferable to
a full close in all cases unless we're destroying the transport.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
The previous behaviour left the connection half-open in order to try
to scrape the last replies from the socket. Now that we have more reliable
reconnection, change the behaviour to close down the socket faster.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Now that we no longer use the partial shutdown code when closing the
socket, we no longer need to worry about the TCP linger2 state.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Instead we rely on SO_REUSEPORT to provide the reconnection semantics
that we need for NFSv2/v3.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
It is not safe to call xs_reset_transport() from inside xs_udp_setup_socket()
or xs_tcp_setup_socket(), since they do not own the correct locks. Instead,
do it in xs_connect().
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
The socket lock is currently held by the task that is requesting the
connection be established. While that is efficient in the case where
the connection happens quickly, it is racy in the case where it doesn't.
What we really want is for the connect helper to be able to block access
to the socket while it is being set up.
This patch does so by arranging to transfer the socket lock from the
task that is requesting the connect attempt, and then releasing that
lock once everything is done.
This scheme also gives us automatic protection against collisions with
the RPC close code, so we can kill the cancel_delayed_work_sync()
call in xs_close().
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Now that we can reuse bound ports after a close, we never really want to
clear the transport's source port after it has been set. Doing so really
messes up the NFSv3 DRC on the server.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Now that we're setting SO_REUSEPORT, we still need to handle the
case where a connect() is attempted, but the old socket is still
lingering.
Essentially, all we want to do here is handle the error by waiting
a few seconds and then retrying.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
When using TCP, we need the ability to reuse port numbers after
a disconnection, so that the NFSv3 server knows that we're the same
client. Currently we use a hack to work around the TCP socket's
TIME_WAIT: we send an RST instead of closing, which doesn't
always work...
The SO_REUSEPORT option added in Linux 3.9 allows us to bind multiple
TCP connections to the same source address+port combination, and thus
to use ordinary TCP close() instead of the current hack.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
It's always set to whatever CONFIG_SUNRPC_DEBUG is, so just use that.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Add tracepoints inside the main loop on xs_tcp_data_recv that allow
us to keep an eye on what's happening during each phase of it.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
...so we can keep track of when calls are sent and replies received.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Now that nfs_release_page() doesn't block indefinitely, other deadlock
avoidance mechanisms aren't needed.
- it doesn't hurt for kswapd to block occasionally. If it doesn't
want to block it would clear __GFP_WAIT. The current_is_kswapd()
was only added to avoid deadlocks and we have a new approach for
that.
- memory allocation in the SUNRPC layer can very rarely try to
->releasepage() a page it is trying to handle. The deadlock
is removed as nfs_release_page() doesn't block indefinitely.
So we don't need to set PF_FSTRANS for sunrpc network operations any
more.
Signed-off-by: NeilBrown <neilb@suse.de>
Acked-by: Jeff Layton <jlayton@primarydata.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
If an iptables drop rule is added for an nfs server, the client can end up in
a softlockup. Because of the way that xs_sendpages() is structured, the -EPERM
is ignored since the prior bits of the packet may have been successfully queued
and thus xs_sendpages() returns a non-zero value. Then, xs_udp_send_request()
thinks that because some bits were queued it should return -EAGAIN. We then try
the request again and again, resulting in cpu spinning. Reproducer:
1) open a file on the nfs server '/nfs/foo' (mounted using udp)
2) iptables -A OUTPUT -d <nfs server ip> -j DROP
3) write to /nfs/foo
4) close /nfs/foo
5) iptables -D OUTPUT -d <nfs server ip> -j DROP
The softlockup occurs in step 4 above.
The previous patch, allows xs_sendpages() to return both a sent count and
any error values that may have occurred. Thus, if we get an -EPERM, return
that to the higher level code.
With this patch in place we can successfully abort the above sequence and
avoid the softlockup.
I also tried the above test case on an nfs mount on tcp and although the system
does not softlockup, I still ended up with the 'hung_task' firing after 120
seconds, due to the i/o being stuck. The tcp case appears a bit harder to fix,
since -EPERM appears to get ignored much lower down in the stack and does not
propogate up to xs_sendpages(). This case is not quite as insidious as the
softlockup and it is not addressed here.
Reported-by: Yigong Lou <ylou@akamai.com>
Signed-off-by: Jason Baron <jbaron@akamai.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
If an error is returned after the first bits of a packet have already been
successfully queued, xs_sendpages() will return a positive 'int' value
indicating success. Callers seem to treat this as -EAGAIN.
However, there are cases where its not a question of waiting for the write
queue to drain. For example, when there is an iptables rule dropping packets
to the destination, the lower level code can return -EPERM only after parts
of the packet have been successfully queued. In this case, we can end up
continuously retrying resulting in a kernel softlockup.
This patch is intended to make no changes in behavior but is in preparation for
subsequent patches that can make decisions based on both on the number of bytes
sent by xs_sendpages() and any errors that may have be returned.
Signed-off-by: Jason Baron <jbaron@akamai.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
When aborting a connection to preserve source ports, don't wake the task in
xs_error_report. This allows tasks with RPC_TASK_SOFTCONN to succeed if the
connection needs to be re-established since it preserves the task's status
instead of setting it to the status of the aborting kernel_connect().
This may also avoid a potential conflict on the socket's lock.
Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
Cc: stable@vger.kernel.org # 3.14+
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
When attempting to establish a local ephemeral endpoint for a TCP or UDP
socket, do not explicitly call bind, instead let it happen implicilty when the
socket is first used.
The main motivating factor for this change is when TCP runs out of unique
ephemeral ports (i.e. cannot find any ephemeral ports which are not a part of
*any* TCP connection). In this situation if you explicitly call bind, then the
call will fail with EADDRINUSE. However, if you allow the allocation of an
ephemeral port to happen implicitly as part of connect (or other functions),
then ephemeral ports can be reused, so long as the combination of (local_ip,
local_port, remote_ip, remote_port) is unique for TCP sockets on the system.
This doesn't matter for UDP sockets, but it seemed easiest to treat TCP and UDP
sockets the same.
This can allow mount.nfs(8) to continue to function successfully, even in the
face of misbehaving applications which are creating a large number of TCP
connections.
Signed-off-by: Chris Perl <chris.perl@gmail.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
