Use the standard gfp mask instead of using GFP_NOWAIT. The latter causes
issues when under memory pressure.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
If the call to rpc_alloc_task() fails, then ensure that the calldata is
released, and that rpc_run_task() and rpc_run_bc_task() bail out early.
Reported-by: NeilBrown <neilb@suse.de>
Fixes: 910ad38697 ("NFS: Fix memory allocation in rpc_alloc_task()")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
As for rpc_malloc(), we first try allocating from the slab, then fall
back to a non-waiting allocation from the mempool.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
When in a low memory situation, we do want rpciod to kick off direct
reclaim in the case where that helps, however we don't want it looping
forever in mempool_alloc().
So first try allocating from the slab using GFP_KERNEL | __GFP_NORETRY,
and then fall back to a GFP_NOWAIT allocation from the mempool.
Ditto for rpc_alloc_task()
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
rpc tasks can be marked as RPC_TASK_SWAPPER. This causes GFP_MEMALLOC
to be used for some allocations. This is needed in some cases, but not
in all where it is currently provided, and in some where it isn't
provided.
Currently *all* tasks associated with a rpc_client on which swap is
enabled get the flag and hence some GFP_MEMALLOC support.
GFP_MEMALLOC is provided for ->buf_alloc() but only swap-writes need it.
However xdr_alloc_bvec does not get GFP_MEMALLOC - though it often does
need it.
xdr_alloc_bvec is called while the XPRT_LOCK is held. If this blocks,
then it blocks all other queued tasks. So this allocation needs
GFP_MEMALLOC for *all* requests, not just writes, when the xprt is used
for any swap writes.
Similarly, if the transport is not connected, that will block all
requests including swap writes, so memory allocations should get
GFP_MEMALLOC if swap writes are possible.
So with this patch:
1/ we ONLY set RPC_TASK_SWAPPER for swap writes.
2/ __rpc_execute() sets PF_MEMALLOC while handling any task
with RPC_TASK_SWAPPER set, or when handling any task that
holds the XPRT_LOCKED lock on an xprt used for swap.
This removes the need for the RPC_IS_SWAPPER() test
in ->buf_alloc handlers.
3/ xprt_prepare_transmit() sets PF_MEMALLOC after locking
any task to a swapper xprt. __rpc_execute() will clear it.
3/ PF_MEMALLOC is set for all the connect workers.
Reviewed-by: Chuck Lever <chuck.lever@oracle.com> (for xprtrdma parts)
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Currently, tasks marked as "swapper" tasks get put to the front of
non-priority rpc_queues, and are sorted earlier than non-swapper tasks on
the transport's ->xmit_queue.
This is pointless as currently *all* tasks for a mount that has swap
enabled on *any* file are marked as "swapper" tasks. So the net result
is that the non-priority rpc_queues are reverse-ordered (LIFO).
This scheduling boost is not necessary to avoid deadlocks, and hurts
fairness, so remove it. If there were a need to expedite some requests,
the tk_priority mechanism is a more appropriate tool.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
When memory is short, new worker threads cannot be created and we depend
on the minimum one rpciod thread to be able to handle everything.
So it must not block waiting for memory.
mempools are particularly a problem as memory can only be released back
to the mempool by an async rpc task running. If all available
workqueue threads are waiting on the mempool, no thread is available to
return anything.
rpc_malloc() can block, and this might cause deadlocks.
So check RPC_IS_ASYNC(), rather than RPC_IS_SWAPPER() to determine if
blocking is acceptable.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
The sections which should not re-enter the filesystem are already
protected with memalloc_nofs_save/restore calls, so it is better to use
GFP_KERNEL in these calls to allow better performance for synchronous
RPC calls.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
The current range of RPC task PIDs is 0..65535. That's not adequate
for distinguishing tasks across multiple rpc_clnts running high
throughput workloads.
To help relieve this situation and to reduce the bottleneck of
having a single atomic for assigning all RPC task PIDs, assign task
PIDs per rpc_clnt.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
When find a task from wait queue to wake up, a non-privileged task may
be found out, rather than the privileged. This maybe lead a deadlock
same as commit dfe1fe75e0 ("NFSv4: Fix deadlock between nfs4_evict_inode()
and nfs4_opendata_get_inode()"):
Privileged delegreturn task is queued to privileged list because all
the slots are assigned. If there has no enough slot to wake up the
non-privileged batch tasks(session less than 8 slot), then the privileged
delegreturn task maybe lost waked up because the found out task can't
get slot since the session is on draining.
So we should treate the privileged task as the emergency task, and
execute it as for as we can.
Reported-by: Hulk Robot <hulkci@huawei.com>
Fixes: 5fcdfacc01 ("NFSv4: Return delegations synchronously in evict_inode")
Cc: stable@vger.kernel.org
Signed-off-by: Zhang Xiaoxu <zhangxiaoxu5@huawei.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
The 'queue->nr' will wraparound from 0 to 255 when only current
priority queue has tasks. This maybe lead a deadlock same as commit
dfe1fe75e0 ("NFSv4: Fix deadlock between nfs4_evict_inode()
and nfs4_opendata_get_inode()"):
Privileged delegreturn task is queued to privileged list because all
the slots are assigned. When non-privileged task complete and release
the slot, a non-privileged maybe picked out. It maybe allocate slot
failed when the session on draining.
If the 'queue->nr' has wraparound to 255, and no enough slot to
service it, then the privileged delegreturn will lost to wake up.
So we should avoid the wraparound on 'queue->nr'.
Reported-by: Hulk Robot <hulkci@huawei.com>
Fixes: 5fcdfacc01 ("NFSv4: Return delegations synchronously in evict_inode")
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
Signed-off-by: Zhang Xiaoxu <zhangxiaoxu5@huawei.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
We could recurse into NFS doing memory reclaim while sending a sync task,
which might result in a deadlock. Set memalloc_nofs_save for sync task
execution.
Fixes: a1231fda7e ("SUNRPC: Set memalloc_nofs_save() on all rpciod/xprtiod jobs")
Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Currently, we wake up the tasks by priority queue ordering, which means
that we ignore the batching that is supposed to help with QoS issues.
Fixes: c049f8ea9a ("SUNRPC: Remove the bh-safe lock requirement on the rpc_wait_queue->lock")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Remove redundant call sites or call sites that are already covered
by tracepoints.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Remove several redundant dprintk call sites, and replace a couple of
potentially useful ones with tracepoints.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
These instruments don't appear to add any substantial value.
We already have this at the termination of each RPC:
iozone-2617 [002] 975.713126: rpc_stats_latency: task:418@5 xid=0x260eab5d nfsv3 LOOKUP backlog=15 rtt=32 execute=58
iozone-2617 [002] 975.713127: xprt_release_cong: task:418@5 snd_task:4294967295 cong=256 cwnd=16384
iozone-2617 [002] 975.713127: xprt_put_cong: task:418@5 snd_task:4294967295 cong=0 cwnd=16384
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Introduce a tracepoint in call_allocate that reports the exact
sizes in the RPC buffer allocation request and the status of the
result. This helps catch problems with XDR buffer provisioning,
and replaces transport-specific debugging instrumentation.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Move the test for whether a task is already queued to prevent
corruption of the timer list in __rpc_sleep_on_priority_timeout().
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Add a flag to signal to the RPC layer that the credential is already
pinned for the duration of the RPC call.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
RPC tasks on the backchannel never invoke xprt_complete_rqst(), so
there is no way to report their tk_status at completion. Also, any
RPC task that exits via rpc_exit_task() before it is replied to will
also disappear without a trace.
Introduce a trace point that is symmetrical with rpc_task_begin that
captures the termination status of each RPC task.
Sample trace output for callback requests initiated on the server:
kworker/u8:12-448 [003] 127.025240: rpc_task_end: task:50@3 flags=ASYNC|DYNAMIC|SOFT|SOFTCONN|SENT runstate=RUNNING|ACTIVE status=0 action=rpc_exit_task
kworker/u8:12-448 [002] 127.567310: rpc_task_end: task:51@3 flags=ASYNC|DYNAMIC|SOFT|SOFTCONN|SENT runstate=RUNNING|ACTIVE status=0 action=rpc_exit_task
kworker/u8:12-448 [001] 130.506817: rpc_task_end: task:52@3 flags=ASYNC|DYNAMIC|SOFT|SOFTCONN|SENT runstate=RUNNING|ACTIVE status=0 action=rpc_exit_task
Odd, though, that I never see trace_rpc_task_complete, either in the
forward or backchannel. Should it be removed?
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Jon Hunter: "I have been tracking down another suspend/NFS related
issue where again I am seeing random delays exiting suspend. The delays
can be up to a couple minutes in the worst case and this is causing a
suspend test we have to fail."
Change the use of a deferrable work to a standard delayed one.
Reported-by: Jon Hunter <jonathanh@nvidia.com>
Tested-by: Jon Hunter <jonathanh@nvidia.com>
Fixes: 7e0a0e38fc ("SUNRPC: Replace the queue timer with a delayed work function")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Ensure that we set task->tk_rpc_status for all RPC level errors so that
the caller can distinguish between those and server reply status errors.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up: commit c544577dad ("SUNRPC: Clean up transport write
space handling") appears to have removed the last caller of
rpc_wake_up_queued_task_on_wq().
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Ensure that we do the required accounting for the round robin queue
when the caller to rpc_init_task() has passed in a transport to be
used.
Reported-by: Olga Kornievskaia <aglo@umich.edu>
Reported-by: Neil Brown <neilb@suse.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
NFSoRDMA client updates for 5.3
New features:
- Add a way to place MRs back on the free list
- Reduce context switching
- Add new trace events
Bugfixes and cleanups:
- Fix a BUG when tracing is enabled with NFSv4.1
- Fix a use-after-free in rpcrdma_post_recvs
- Replace use of xdr_stream_pos in rpcrdma_marshal_req
- Fix occasional transport deadlock
- Fix show_nfs_errors macros, other tracing improvements
- Remove RPCRDMA_REQ_F_PENDING and fr_state
- Various simplifications and refactors
Adapt and apply changes that were made to the TCP socket connect
code. See the following commits for details on the purpose of
these changes:
Commit 7196dbb02e ("SUNRPC: Allow changing of the TCP timeout parameters on the fly")
Commit 3851f1cdb2 ("SUNRPC: Limit the reconnect backoff timer to the max RPC message timeout")
Commit 02910177ae ("SUNRPC: Fix reconnection timeouts")
Some common transport code is moved to xprt.c to satisfy the code
duplication police.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
For diagnostic purposes, it would be useful to have an rpc_iostats
metric of RPCs completing with tk_status < 0. Unfortunately,
tk_status is reset inside the rpc_call_done functions for each
operation, and the call to tally the per-op metrics comes after
rpc_call_done. Refactor the call to rpc_count_iostat earlier in
rpc_exit_task so we can count these RPCs completing in error.
Signed-off-by: Dave Wysochanski <dwysocha@redhat.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
The queue timer function, which walks the RPC queue in order to locate
candidates for waking up is one of the current constraints against
removing the bh-safe queue spin locks. Replace it with a delayed
work queue, so that we can do the actual rpc task wake ups from an
ordinary process context.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Jon Hunter reports:
"I have been noticing intermittent failures with a system suspend test on
some of our machines that have a NFS mounted root file-system. Bisecting
this issue points to your commit 431235818b ("SUNRPC: Declare RPC
timers as TIMER_DEFERRABLE") and reverting this on top of v5.2-rc3 does
appear to resolve the problem.
The cause of the suspend failure appears to be a long delay observed
sometimes when resuming from suspend, and this is causing our test to
timeout."
This reverts commit 431235818b.
Reported-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Add SPDX license identifiers to all files which:
- Have no license information of any form
- Have EXPORT_.*_SYMBOL_GPL inside which was used in the
initial scan/conversion to ignore the file
These files fall under the project license, GPL v2 only. The resulting SPDX
license identifier is:
GPL-2.0-only
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Don't wake idle CPUs only for the purpose of servicing an RPC
queue timeout.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Simplify the setting of queue timeouts by using the timer_reduce()
function.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Add a helper to ensure that debugfs and friends print out the
correct current task timeout value.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Clean up the RPC task sleep interfaces by replacing the task->tk_timeout
'hidden parameter' to rpc_sleep_on() with a new function that takes an
absolute timeout.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
None of the callers set the 'action' argument, so let's just remove it.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
rpc_sleep_on() does not need to set the task->tk_callback under the
queue lock, so move that out.
Also refactor the check for whether the task is active.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The RPC_TASK_KILLED flag should really not be set from another context
because it can clobber data in the struct task when task->tk_flags is
changed non-atomically.
Let's therefore swap out RPC_TASK_KILLED with an atomic flag, and add
a function to set that flag and safely wake up the task.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The RPC task wakeup calls all check for RPC_IS_QUEUED() before taking any
locks. In addition, rpc_exit() already calls rpc_wake_up_queued_task().
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Convert the remaining gfp_flags arguments in sunrpc to standard reclaiming
allocations, now that we set memalloc_nofs_save() as appropriate.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Set memalloc_nofs_save() on all the rpciod/xprtiod jobs so that we
ensure memory allocations for asynchronous rpc calls don't ever end
up recursing back to the NFS layer for memory reclaim.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
SUNRPC has two sorts of credentials, both of which appear as
"struct rpc_cred".
There are "generic credentials" which are supplied by clients
such as NFS and passed in 'struct rpc_message' to indicate
which user should be used to authorize the request, and there
are low-level credentials such as AUTH_NULL, AUTH_UNIX, AUTH_GSS
which describe the credential to be sent over the wires.
This patch replaces all the generic credentials by 'struct cred'
pointers - the credential structure used throughout Linux.
For machine credentials, there is a special 'struct cred *' pointer
which is statically allocated and recognized where needed as
having a special meaning. A look-up of a low-level cred will
map this to a machine credential.
Signed-off-by: NeilBrown <neilb@suse.com>
Acked-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The credential passed in rpc_message.rpc_cred is always a
generic credential except in one instance.
When gss_destroying_context() calls rpc_call_null(), it passes
a specific credential that it needs to destroy.
In this case the RPC acts *on* the credential rather than
being authorized by it.
This special case deserves explicit support and providing that will
mean that rpc_message.rpc_cred is *always* generic, allowing
some optimizations.
So add "tk_op_cred" to rpc_task and "rpc_op_cred" to the setup data.
Use this to pass the cred down from rpc_call_null(), and have
rpcauth_bindcred() notice it and bind it in place.
Credit to kernel test robot <fengguang.wu@intel.com> for finding
a bug in earlier version of this patch.
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Fix up the priority queue to not batch by owner, but by queue, so that
we allow '1 << priority' elements to be dequeued before switching to
the next priority queue.
The owner field is still used to wake up requests in round robin order
by owner to avoid single processes hogging the RPC layer by loading the
queues.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Trond Myklebust