Jeff Layton [Fri, 10 Jul 2015 19:58:42 +0000 (15:58 -0400)]
nfs: plug memory leak when ->prepare_layoutcommit fails
"data" is currently leaked when the prepare_layoutcommit operation
returns an error. Put the cred before taking the spinlock in that
case, take the lock and then goto out_unlock which will drop the
lock and then free "data".
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
sunrpc: translate -EAGAIN to -ENOBUFS when socket is writable.
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.
J. Bruce Fields [Thu, 23 Jul 2015 15:08:43 +0000 (11:08 -0400)]
NFSv4.2: handle NFS-specific llseek errors
Handle NFS-specific llseek errors instead of letting them leak out to
userspace.
Reported-by: Benjamin Coddington <bcodding@redhat.com> Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
NFS: Don't clear desc->pg_moreio in nfs_do_recoalesce()
Recoalescing does not affect whether or not we've already sent off
I/O, and doing so means that we end up sending a bunch of synchronous
for cases where we actually need to be using unstable writes.
If the function exits early, then we must put those requests that were
not processed back onto the &mirror->pg_list so they can be cleaned up
by nfs_pgio_error().
Fixes: a7d42ddb30997 ("nfs: add mirroring support to pgio layer") Cc: stable@vger.kernel.org # v4.0+ Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
NFS: nfs_mark_for_revalidate should always set NFS_INO_REVAL_PAGECACHE
I'm not aware of any existing bugs around this, but the expectation is
that nfs_mark_for_revalidate() should always force a revalidation of
the cached metadata.
Setting the change attribute has been mandatory for all NFS versions, since
commit 3a1556e8662c ("NFSv2/v3: Simulate the change attribute"). We should
therefore not have anything be conditional on it being set/unset.
NFS: Don't revalidate the mapping if both size and change attr are up to date
If we've ensured that the size and the change attribute are both correct,
then there is no point in marking those attributes as needing revalidation
again. Only do so if we know the size is incorrect and was not updated.
pNFS writes don't return attributes, however that doesn't mean that we
should ignore the fact that they may be extending the file. This patch
ensures that if a write is seen to extend the file, then we always set
an attribute barrier, and update the cached file size.
If a process reopens the file before we can send off the CLOSE/DELEGRETURN,
then pnfs_roc_drain() may end up waiting for a new set of layout segments
that are marked as return-on-close, but haven't yet been returned.
Fix this by only waiting for those layout segments that were invalidated in
pnfs_roc().
pNFS: Fix races between return-on-close and layoutreturn.
If one or more of the layout segments reports an error during I/O, then
we may have to send a layoutreturn to report the error back to the NFS
metadata server.
This patch ensures that the return-on-close code can detect the
outstanding layoutreturn, and not preempt it.
The server has encountered an unrecoverable fault with the
backchannel (e.g., it has lost track of the sequence ID for a slot
in the backchannel). The client MUST stop sending more requests
on the session's fore channel, wait for all outstanding requests
to complete on the fore and back channel, and then destroy the
session.
NFSv4.1: Handle SEQ4_STATUS_RECALLABLE_STATE_REVOKED status bit correctly
Try to handle this for now by invalidating all outstanding layouts for this
server and then testing all the open+lock+delegation stateids.
At some later stage, we may want to optimise by separating out the testing of
delegation stateids only, and adding testing of layout stateids.
NFSv4.1: Handle SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED status bit correctly.
If the server tells us that only some state has been revoked, then we
need to run the full TEST_STATEID dog and pony show in order to discover
which locks and delegations are still OK. Currently we blow away all
state, which means that we lose all locks!
SUNRPC: Don't confuse ENOBUFS with a write_space issue
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>
SUNRPC: Don't reencode message if transmission failed with ENOBUFS
If we're running out of buffer memory when transmitting data, then
we want to just delay for a moment, and then continue transmitting
the remainder of the message.
The problem is almost exactly the same as the one fixed by feaff8e5b2cf.
We must take a reference to the struct file when running the LOCKU
compound to prevent the final fput from running until the operation is
complete.
Reported-by: Jean Spector <jean@primarydata.com> Signed-off-by: Jeff Layton <jeff.layton@primarydata.com> Cc: stable@vger.kernel.org Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
NeilBrown [Mon, 29 Jun 2015 04:28:54 +0000 (14:28 +1000)]
NFSv4: When returning a delegation, don't reclaim an incompatible open mode.
It is possible to have an active open with one mode, and a delegation
for the same file with a different mode.
In particular, a WR_ONLY open and an RD_ONLY delegation.
This happens if a WR_ONLY open is followed by a RD_ONLY open which
provides a delegation, but is then close.
When returning the delegation, we currently try to claim opens for
every open type (n_rdwr, n_rdonly, n_wronly). As there is no harm
in claiming an open for a mode that we already have, this is often
simplest.
However if the delegation only provides a subset of the modes that we
currently have open, this will produce an error from the server.
So when claiming open modes prior to returning a delegation, skip the
open request if the mode is not covered by the delegation - the open_stateid
must already cover that mode, so there is nothing to do.
Trond Myklebust [Fri, 26 Jun 2015 18:51:32 +0000 (14:51 -0400)]
pNFS/flexfiles: Turn off layoutcommit for servers that don't need it
This patch ensures that we record the value of 'ffl_flags' from
the layout, and then checks for the presence of the
FF_FLAGS_NO_LAYOUTCOMMIT flag before deciding whether or not to
call pnfs_set_layoutcommit().
The effect is that servers now can decide whether or not they want
the client to call layoutcommit before returning a writeable layout.
Trond Myklebust [Fri, 26 Jun 2015 18:01:59 +0000 (14:01 -0400)]
Merge branch 'layoutstats'
* layoutstats:
pnfs/flexfiles: protect ktime manipulation with mirror lock
nfs: provide pnfs_report_layoutstat when NFS42 is disabled
pnfs/flexfiles: report layoutstat regularly
nfs42: serialize LAYOUTSTATS calls of the same file
pnfs/flexfiles: encode LAYOUTSTATS flexfiles specific data
pnfs/flexfiles: add ff_layout_prepare_layoutstats
pNFS/flexfiles: track when layout is first used
pNFS/flexfiles: add layoutstats tracking
pNFS/flexfiles: Remove unused struct members user_name, group_name
pnfs: add pnfs_report_layoutstat helper function
pNFS: fill in nfs42_layoutstat_ops
NFSv.2/pnfs Add a LAYOUTSTATS rpc function
Peng Tao [Thu, 25 Jun 2015 10:19:32 +0000 (18:19 +0800)]
nfs: provide pnfs_report_layoutstat when NFS42 is disabled
kbuild test robot reported:
fs/built-in.o: In function `pnfs_report_layoutstat':
>> (.text+0x151a1c): undefined reference to `nfs42_proc_layoutstats_generic'
Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Peng Tao <tao.peng@primarydata.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Commit 9597c13b forbade opens with O_APPEND|O_DIRECT for NFSv4:
nfs: verify open flags before allowing an atomic open
Currently, you can open a NFSv4 file with O_APPEND|O_DIRECT, but cannot
fcntl(F_SETFL,...) with those flags. This flag combination is explicitly
forbidden on NFSv3 opens, and it seems like it should also be on NFSv4.
However, you can still open a file with O_DIRECT|O_APPEND if there exists a
cached dentry for the file because nfs4_file_open() is used instead of
nfs_atomic_open() and the check is bypassed. Add the check in
nfs4_file_open() as well.
Signed-off-by: Benjamin Coddington <bcodding@redhat.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Jeff Layton [Wed, 24 Jun 2015 16:10:24 +0000 (12:10 -0400)]
nfs: always update creds in mirror, even when we have an already connected ds
A ds can be associated with more than one mirror, but we currently skip
setting a mirror's credentials if we find that it's already set up with
a connected client.
The upshot is that we can end up sending DS writes with MDS credentials
instead of properly setting them up. Fix nfs4_ff_layout_prepare_ds to
always verify that the mirror's credentials are set up, even when we
have a DS that's already connected.
Reported-by: Tom Haynes <thomas.haynes@primarydata.com> Signed-off-by: Jeff Layton <jeff.layton@primarydata.com> Cc: stable@vger.kernel.org # 4.0+ Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Jeff Layton [Wed, 24 Jun 2015 16:10:23 +0000 (12:10 -0400)]
nfs: fix potential credential leak in ff_layout_update_mirror_cred
If we have two tasks racing to update a mirror's credentials, then they
can end up leaking one (or more) sets of credentials. The first task
will set mirror->cred and then the second task will just overwrite it.
Use a cmpxchg to ensure that the creds are only set once. If we get to
the point where we would set mirror->cred and find that they're already
set, then we just release the creds that were just found.
Peng Tao [Tue, 23 Jun 2015 11:52:01 +0000 (19:52 +0800)]
pnfs/flexfiles: add ff_layout_prepare_layoutstats
It fills in the generic part of LAYOUTSTATS call. One thing to note
is that we don't really track if IO is continuous or not. So just fake
to use the completed bytes for it.
Still missing flexfiles specific part, which will be included in the next patch.
Reviewed-by: Jeff Layton <jeff.layton@primarydata.com> Signed-off-by: Peng Tao <tao.peng@primarydata.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Trond Myklebust [Mon, 22 Jun 2015 13:55:08 +0000 (09:55 -0400)]
Merge branch 'bugfixes'
* bugfixes:
NFS: Ensure we set NFS_CONTEXT_RESEND_WRITES when requeuing writes
pNFS: Fix a memory leak when attempted pnfs fails
NFS: Ensure that we update the sequence id under the slot table lock
nfs: Initialize cb_sequenceres information before validate_seqid()
nfs: Only update callback sequnce id when CB_SEQUENCE success
NFSv4: nfs4_handle_delegation_recall_error should ignore EAGAIN
Trond Myklebust [Sat, 20 Jun 2015 19:31:54 +0000 (15:31 -0400)]
SUNRPC: Set the TCP user timeout option on client sockets
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.
Trond Myklebust [Fri, 19 Jun 2015 20:17:57 +0000 (16:17 -0400)]
SUNRPC: Ensure we release the TCP socket once it has been closed
This fixes a regression introduced by commit caf4ccd4e88cf2 ("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>
Trond Myklebust [Fri, 19 Jun 2015 17:04:13 +0000 (13:04 -0400)]
SUNRPC: Handle connection issues correctly on the back channel
If the back channel is disconnected, we can and should just fail the
transmission. The expectation is that the NFSv4.1 server will always
retransmit any outstanding callbacks once the connection is
re-established.
Trond Myklebust [Wed, 17 Jun 2015 23:56:22 +0000 (19:56 -0400)]
NFS: Ensure we set NFS_CONTEXT_RESEND_WRITES when requeuing writes
If a write attempt fails, and the write is queued up for resending to
the server, as opposed to being dropped, then we need to set the
appropriate flag so that nfs_file_fsync() does the right thing.
Trond Myklebust [Wed, 17 Jun 2015 23:41:51 +0000 (19:41 -0400)]
pNFS: Fix a memory leak when attempted pnfs fails
pnfs_do_write() expects the call to pnfs_write_through_mds() to free the
pgio header and to release the layout segment before exiting. The problem
is that nfs_pgio_data_destroy() doesn't actually do this; it only frees
the memory allocated by nfs_generic_pgio().
Ditto for pnfs_do_read()...
Fix in both cases is to add a call to hdr->release(hdr).
Trond Myklebust [Tue, 16 Jun 2015 15:36:47 +0000 (11:36 -0400)]
Merge tag 'nfs-rdma-for-4.2' of git://git.linux-nfs.org/projects/anna/nfs-rdma
NFS: NFSoRDMA Client Changes
These patches continue to build up for improving the rsize and wsize that the
NFS client uses when talking over RDMA. In addition, these patches also add
in scalability enhancements and other bugfixes.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
* tag 'nfs-rdma-for-4.2' of git://git.linux-nfs.org/projects/anna/nfs-rdma: (142 commits)
xprtrdma: Reduce per-transport MR allocation
xprtrdma: Stack relief in fmr_op_map()
xprtrdma: Split rb_lock
xprtrdma: Remove rpcrdma_ia::ri_memreg_strategy
xprtrdma: Remove ->ro_reset
xprtrdma: Remove unused LOCAL_INV recovery logic
xprtrdma: Acquire MRs in rpcrdma_register_external()
xprtrdma: Introduce an FRMR recovery workqueue
xprtrdma: Acquire FMRs in rpcrdma_fmr_register_external()
xprtrdma: Introduce helpers for allocating MWs
xprtrdma: Use ib_device pointer safely
xprtrdma: Remove rr_func
xprtrdma: Replace rpcrdma_rep::rr_buffer with rr_rxprt
xprtrdma: Warn when there are orphaned IB objects
...
Client can receives stateid-type error (eg., BAD_STATEID) on SETATTR when
delegation stateid was used. When no open state exists, in case of application
calling truncate() on the file, client has no state to recover and fails with
EIO.
Instead, upon such error, return the bad delegation and then resend the
SETATTR with a zero stateid.
Signed-off: Olga Kornievskaia <kolga@netapp.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Jeff Layton [Tue, 9 Jun 2015 23:44:00 +0000 (19:44 -0400)]
nfs: make nfs4_init_uniform_client_string use a dynamically allocated buffer
Change the uniform client string generator to dynamically allocate the
NFSv4 client name string buffer. With this patch, we can eliminate the
buffers that are embedded within the "args" structs and simply use the
name string that is hanging off the client.
This uniform string case is a little simpler than the nonuniform since
we don't need to deal with RCU, but we do have two different cases,
depending on whether there is a uniquifier or not.
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Jeff Layton [Tue, 9 Jun 2015 23:43:59 +0000 (19:43 -0400)]
nfs: make nfs4_init_nonuniform_client_string use a dynamically allocated buffer
The way the *_client_string functions work is a little goofy. They build
the string in an on-stack buffer and then use kstrdup to copy it. This
is not only stack-heavy but artificially limits the size of the client
name string. Change it so that we determine the length of the string,
allocate it and then scnprintf into it.
Since the contents of the nonuniform string depend on rcu-managed data
structures, it's possible that they'll change between when we allocate
the string and when we go to fill it. If that happens, free the string,
recalculate the length and try again. If it the mismatch isn't resolved
on the second try then just give up and return -EINVAL.
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Jeff Layton [Tue, 9 Jun 2015 23:43:58 +0000 (19:43 -0400)]
nfs: update maxsz values for SETCLIENTID and EXCHANGE_ID
The spec allows for up to NFS4_OPAQUE_LIMIT (1k). While we'll almost
certainly never use that much, these ops are generally the only ones
in the compound so we might as well allow for them to be that large.
Also, the existing code didn't add in a word for the opaque length
field for either name string. Fix that while we're in there.
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Jeff Layton [Tue, 9 Jun 2015 23:43:57 +0000 (19:43 -0400)]
nfs: convert setclientid and exchange_id encoders to use clp->cl_owner_id
...instead of buffers that are part of their arg structs. We already
hold a reference to the client, so we might as well use the allocated
buffer. In the event that we can't allocate the clp->cl_owner_id, then
just return -ENOMEM.
Note too that we switch from a GFP_KERNEL allocation here to GFP_NOFS.
It's possible we could end up trying to do a SETCLIENTID or EXCHANGE_ID
in order to reclaim some memory, and the GFP_KERNEL allocations in the
existing code could cause recursion back into NFS reclaim.
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Neil Brown [Mon, 15 Jun 2015 05:55:30 +0000 (15:55 +1000)]
SUNRPC: never enqueue a ->rq_cong request on ->sending
If the sending queue has a task without ->rq_cong set at the front,
and then a number of tasks with ->rq_cong set such that they use
the entire congestion window, then the queue deadlocks. The first
entry cannot be processed until later entries complete.
This scenario has been seen with a client using UDP to access a server,
and the network connection breaking for a period of time - it doesn't
recover.
It never really makes sense for an ->rq_cong request to be on the ->sending
queue, but it can happen when a request is being retried, and finds
the transport if locked (XPRT_LOCKED). In this case we simple call
__xprt_put_cong() and the deadlock goes away.
Chuck Lever [Tue, 26 May 2015 15:53:13 +0000 (11:53 -0400)]
xprtrdma: Split rb_lock
/proc/lock_stat showed contention between rpcrdma_buffer_get/put
and the MR allocation functions during I/O intensive workloads.
Now that MRs are no longer allocated in rpcrdma_buffer_get(),
there's no reason the rb_mws list has to be managed using the
same lock as the send/receive buffers. Split that lock. The
new lock does not need to disable interrupts because buffer
get/put is never called in an interrupt context.
struct rpcrdma_buffer is re-arranged to ensure rb_mwlock and rb_mws
are always in a different cacheline than rb_lock and the buffer
pointers.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Reviewed-by: Sagi Grimberg <sagig@mellanox.com> Tested-By: Devesh Sharma <devesh.sharma@avagotech.com> Reviewed-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Chuck Lever [Tue, 26 May 2015 15:52:54 +0000 (11:52 -0400)]
xprtrdma: Remove ->ro_reset
An RPC can exit at any time. When it does so, xprt_rdma_free() is
called, and it calls ->op_unmap().
If ->ro_reset() is running due to a transport disconnect, the two
methods can race while processing the same rpcrdma_mw. The results
are unpredictable.
Because of this, in previous patches I've altered ->ro_map() to
handle MR reset. ->ro_reset() is no longer needed and can be
removed.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Reviewed-by: Sagi Grimberg <sagig@mellanox.com> Reviewed-by: Devesh Sharma <devesh.sharma@avagotech.com> Tested-By: Devesh Sharma <devesh.sharma@avagotech.com> Reviewed-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Chuck Lever [Tue, 26 May 2015 15:52:35 +0000 (11:52 -0400)]
xprtrdma: Acquire MRs in rpcrdma_register_external()
Acquiring 64 MRs in rpcrdma_buffer_get() while holding the buffer
pool lock is expensive, and unnecessary because most modern adapters
can transfer 100s of KBs of payload using just a single MR.
Instead, acquire MRs one-at-a-time as chunks are registered, and
return them to rb_mws immediately during deregistration.
Note: commit 539431a437d2 ("xprtrdma: Don't invalidate FRMRs if
registration fails") is reverted: There is now a valid case where
registration can fail (with -ENOMEM) but the QP is still in RTS.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Tested-By: Devesh Sharma <devesh.sharma@avagotech.com> Reviewed-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Chuck Lever [Tue, 26 May 2015 15:52:25 +0000 (11:52 -0400)]
xprtrdma: Introduce an FRMR recovery workqueue
After a transport disconnect, FRMRs can be left in an undetermined
state. In particular, the MR's rkey is no good.
Currently, FRMRs are fixed up by the transport connect worker, but
that can race with ->ro_unmap if an RPC happens to exit while the
transport connect worker is running.
A better way of dealing with broken FRMRs is to detect them before
they are re-used by ->ro_map. Such FRMRs are either already invalid
or are owned by the sending RPC, and thus no race with ->ro_unmap
is possible.
Introduce a mechanism for handing broken FRMRs to a workqueue to be
reset in a context that is appropriate for allocating resources
(ie. an ib_alloc_fast_reg_mr() API call).
This mechanism is not yet used, but will be in subsequent patches.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Tested-By: Devesh Sharma <devesh.sharma@avagotech.com> Reviewed-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Chuck Lever [Tue, 26 May 2015 15:52:16 +0000 (11:52 -0400)]
xprtrdma: Acquire FMRs in rpcrdma_fmr_register_external()
Acquiring 64 FMRs in rpcrdma_buffer_get() while holding the buffer
pool lock is expensive, and unnecessary because FMR mode can
transfer up to a 1MB payload using just a single ib_fmr.
Instead, acquire ib_fmrs one-at-a-time as chunks are registered, and
return them to rb_mws immediately during deregistration.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Tested-By: Devesh Sharma <devesh.sharma@avagotech.com> Reviewed-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Chuck Lever [Tue, 26 May 2015 15:51:56 +0000 (11:51 -0400)]
xprtrdma: Use ib_device pointer safely
The connect worker can replace ri_id, but prevents ri_id->device
from changing during the lifetime of a transport instance. The old
ID is kept around until a new ID is created and the ->device is
confirmed to be the same.
Cache a copy of ri_id->device in rpcrdma_ia and in rpcrdma_rep.
The cached copy can be used safely in code that does not serialize
with the connect worker.
Other code can use it to save an extra address generation (one
pointer dereference instead of two).
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Tested-By: Devesh Sharma <devesh.sharma@avagotech.com> Reviewed-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Chuck Lever [Tue, 26 May 2015 15:51:27 +0000 (11:51 -0400)]
xprtrdma: Warn when there are orphaned IB objects
WARN during transport destruction if ib_dealloc_pd() fails. This is
a sign that xprtrdma orphaned one or more RDMA API objects at some
point, which can pin lower layer kernel modules and cause shutdown
to hang.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Reviewed-by: Sagi Grimberg <sagig@mellanox.com> Reviewed-by: Devesh Sharma <devesh.sharma@avagotech.com> Tested-By: Devesh Sharma <devesh.sharma@avagotech.com> Reviewed-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Trond Myklebust [Fri, 12 Jun 2015 01:13:58 +0000 (21:13 -0400)]
NFS: Ensure that we update the sequence id under the slot table lock
Fix a callback slot table regression.
Fixes: e937ee714b2d ("nfs: Only update callback sequnce id when CB_SEQUENCE success") Cc: Kinglong Mee <kinglongmee@gmail.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Jeff Layton [Thu, 4 Jun 2015 22:40:13 +0000 (18:40 -0400)]
nfs: deny backchannel RPCs with an incorrect authflavor instead of dropping them
A drop should really only be done when the frame is malformed or we have
reason to think that there is some sort of DoS going on. When we get an
RPC with bad auth, we should send back an error instead.
Cc: Andy Adamson <William.Adamson@netapp.com> Signed-off-by: Jeff Layton <jeff.layton@primarydata.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Chuck Lever [Thu, 11 Jun 2015 17:47:10 +0000 (13:47 -0400)]
SUNRPC: Address kbuild warning in net/sunrpc/debugfs.c
Cross-compile test on ARCH=mn10300:
In file included from include/linux/list.h:8:0,
from include/linux/wait.h:6,
from include/linux/fs.h:6,
from include/linux/debugfs.h:18,
from net/sunrpc/debugfs.c:7:
net/sunrpc/debugfs.c: In function 'fault_disconnect_write':
include/linux/kernel.h:723:17: warning: comparison of distinct pointer
types lacks a cast
(void) (&_min1 == &_min2); \
^
>> net/sunrpc/debugfs.c:307:8: note: in expansion of macro 'min'
len = min(len, sizeof(buffer) - 1);
Fixes: ('SUNRPC: Transport fault injection') Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Kinglong Mee [Tue, 2 Jun 2015 10:58:46 +0000 (18:58 +0800)]
nfs: Only update callback sequnce id when CB_SEQUENCE success
When testing pnfs layout, nfsd got error NFS4ERR_SEQ_MISORDERED.
It is caused by nfs return NFS4ERR_DELAY before validate_seqid(),
don't update the sequnce id, but nfsd updates the sequnce id !!!
According to RFC5661 20.9.3,
" If CB_SEQUENCE returns an error, then the state of the slot
(sequence ID, cached reply) MUST NOT change. "
Signed-off-by: Kinglong Mee <kinglongmee@gmail.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Vaishali Thakkar [Wed, 10 Jun 2015 14:45:29 +0000 (20:15 +0530)]
NFS: Convert use of __constant_htonl to htonl
In little endian cases, the macro htonl unfolds to __swab32 which
provides special case for constants. In big endian cases,
__constant_htonl and htonl expand directly to the same expression.
So, replace __constant_htonl with htonl with the goal of getting
rid of the definition of __constant_htonl completely.
The semantic patch that performs this transformation is as follows:
Chuck Lever [Mon, 11 May 2015 18:02:25 +0000 (14:02 -0400)]
SUNRPC: Transport fault injection
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
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>
Anna Schumaker [Wed, 10 Jun 2015 20:54:28 +0000 (16:54 -0400)]
NFS: Remove unused nfs_rw_ops->rw_release() function
This was only ever set to nfs_writeback_release_common(), a function
which is completely empty. Let's just drop this function pointer and
simplify the code a bit.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Jeff Layton [Wed, 3 Jun 2015 20:14:29 +0000 (16:14 -0400)]
sunrpc: turn swapper_enable/disable functions into rpc_xprt_ops
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>