Ensure that we always bump or drop the nlink count on the parent directory
when we do a mkdir or a rmdir(). This needs to be done by hand as we don't
have pre/post op attributes.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

If the server tells us that out layoutreturn raced with another layout
update, then we must ensure that the new layout segments are not in use
before we resend with an updated layout stateid.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

I found that injecting disconnects with v4.18-rc resulted in
random failures of the multi-threaded git regression test.

The root cause appears to be that, after a reconnect, the
RPC/RDMA transport is waking pending RPCs before the transport has
posted enough Receive buffers to receive the Replies. If a Reply
arrives before enough Receive buffers are posted, the connection
is dropped. A few connection drops happen in quick succession as
the client and server struggle to regain credit synchronization.

This regression was introduced with commit 7c8d9e7c ("xprtrdma:
Move Receive posting to Receive handler"). The client is supposed to
post a single Receive when a connection is established because
it's not supposed to send more than one RPC Call before it gets
a fresh credit grant in the first RPC Reply [RFC 8166, Section
3.3.3].

Unfortunately there appears to be a longstanding bug in the Linux
client's credit accounting mechanism. On connect, it simply dumps
all pending RPC Calls onto the new connection. It's possible it has
done this ever since the RPC/RDMA transport was added to the kernel
ten years ago.

Servers have so far been tolerant of this bad behavior. Currently no
server implementation ever changes its credit grant over reconnects,
and servers always repost enough Receives before connections are
fully established.

The Linux client implementation used to post a Receive before each
of these Calls. This has covered up the flooding send behavior.

I could try to correct this old bug so that the client sends exactly
one RPC Call and waits for a Reply. Since we are so close to the
next merge window, I'm going to instead provide a simple patch to
post enough Receives before a reconnect completes (based on the
number of credits granted to the previous connection).

The spurious disconnects will be gone, but the client will still
send multiple RPC Calls immediately after a reconnect.

Addressing the latter problem will wait for a merge window because
a) I expect it to be a large change requiring lots of testing, and
b) obviously the Linux client has interoperated successfully since
day zero while still being broken.

Fixes: 7c8d9e7c ("xprtrdma: Move Receive posting to ... ")
Cc: stable@vger.kernel.org # v4.18+
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

Remove trailing whitespace and blank line at EOF
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

After a live data migration event at the NFS server, the client may send
I/O requests to the wrong server, causing a live hang due to repeated
recovery events.  On the wire, this will appear as an I/O request failing
with NFS4ERR_BADSESSION, followed by successful CREATE_SESSION, repeatedly.
NFS4ERR_BADSSESSION is returned because the session ID being used was
issued by the other server and is not valid at the old server.

The failure is caused by async worker threads having cached the transport
(xprt) in the rpc_task structure.  After the migration recovery completes,
the task is redispatched and the task resends the request to the wrong
server based on the old value still present in tk_xprt.

The solution is to recompute the tk_xprt field of the rpc_task structure
so that the request goes to the correct server.
Signed-off-by: Bill Baker <bill.baker@oracle.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Tested-by: Helen Chao <helen.chao@oracle.com>
Fixes: fb43d172 ("SUNRPC: Use the multipath iterator to assign a ...")
Cc: stable@vger.kernel.org # v4.9+
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

If there is an error during processing of a callback message, it leads
to refrence leak on the client structure and eventually an unclean
superblock.
Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

Smatch complains that "num" can be uninitialized when kstrtoul() returns
-ERANGE.  It's true enough, but basically harmless in this case.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

kstrtoul() can return -ERANGE so Smatch complains that "num" can be
uninitialized.  We check that it's within bounds so it's not a huge
deal.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

The existing rpc_print_iostats has a few shortcomings.  First, the naming
is not consistent with other functions in the kernel that display stats.
Second, it is really displaying stats for an rpc_clnt structure as it
displays both xprt stats and per-op stats.  Third, it does not handle
rpc_clnt clones, which is important for the one in-kernel tree caller
of this function, the NFS client's nfs_show_stats function.

Fix all of the above by renaming the rpc_print_iostats to
rpc_clnt_show_stats and looping through any rpc_clnt clones via
cl_parent.

Once this interface is fixed, this addresses a problem with NFSv4.
Before this patch, the /proc/self/mountstats always showed incorrect
counts for NFSv4 lease and session related opcodes such as SEQUENCE,
RENEW, SETCLIENTID, CREATE_SESSION, etc.  These counts were always 0
even though many ops would go over the wire.  The reason for this is
there are multiple rpc_clnt structures allocated for any given NFSv4
mount, and inside nfs_show_stats() we callled into rpc_print_iostats()
which only handled one of them, nfs_server->client.  Fix these counts
by calling sunrpc's new rpc_clnt_show_stats() function, which handles
cloned rpc_clnt structs and prints the stats together.

Note that one side-effect of the above is that multiple mounts from
the same NFS server will show identical counts in the above ops due
to the fact the one rpc_clnt (representing the NFSv4 client state)
is shared across mounts.
Signed-off-by: Dave Wysochanski <dwysocha@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

Add a helper function to add the metrics in two rpc_iostats structures.
Signed-off-by: Dave Wysochanski <dwysocha@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

Refactor the output of the metrics for one RPC op into an internal
function.  No functional change.
Signed-off-by: Dave Wysochanski <dwysocha@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

This turns rpc_auth_create_args into a const as it gets passed through the
auth stack.
Signed-off-by: Sargun Dhillon <sargun@sargun.me>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

"dev->nr_children" is the number of children which were parsed
successfully in bl_parse_stripe().  It could be all of them and then, in
that case, it is equal to v->stripe.volumes_count.  Either way, the >
should be >= so that we don't go beyond the end of what we're supposed
to.

Fixes: 5c83746a ("pnfs/blocklayout: in-kernel GETDEVICEINFO XDR parsing")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: stable@vger.kernel.org # 3.17+
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

Commit 530ea421 ("nfs: Referrals should use the same proto setting
as their parent") encloses the fix with #ifdef CONFIG_SUNRPC_XPRT_RDMA.

CONFIG_SUNRPC_XPRT_RDMA is a tristate option, so it should be tested
with #if IS_ENABLED().

Fixes: 530ea421 ("nfs: Referrals should use the same proto setting as their parent")
Reported-by: Helen Chao <helen.chao@oracle.com>
Tested-by: Helen Chao <helen.chao@oracle.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Bill Baker <bill.baker@oracle.com>
Signed-off-by: Calum Mackay <calum.mackay@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

When reclaiming a delegation via CLAIM_PREVIOUS open, the server can
indicate that the delegation has been recalled since it was issued by
setting the "recalled" flag in the delegation.

Ensure that we respect the flag by initiating a delegation return when
it is set.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

Use new return type vm_fault_t for fault handler
in struct vm_operations_struct. For now, this is
just documenting that the function returns a
VM_FAULT value rather than an errno.  Once all
instances are converted, vm_fault_t will become
a distinct type.

see commit 1c8f4220 ("mm: change return type to
vm_fault_t") for reference.
Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Reviewed-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

Even though the caller of nfs_idmap_prepare_message() checks return
code in their side but it's better to add an error check for match_int()
so that we can avoid unnecessary operations when bad int arg is
detected.
Signed-off-by: Chengguang Xu <cgxu519@gmx.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

Return -ESTALE to force a lookup when the file has no more links
Signed-off-by: Lance Shelton <lance.shelton@hammerspace.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

execute_ok() will only check the mode bits if the object is not a
directory, so we don't need to revalidate the attributes in that case.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

When nfs_update_inode() sets NFS_INO_INVALID_ACCESS it is a sign that
we want to revalidate the access cache, not the inode attributes.
In fact we only want to revalidate here if we see that the mode bits
are invalid, so check for NFS_INO_INVALID_OTHER instead.
Reported-by: Olga Kornievskaia <aglo@umich.edu>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

If the writes are being rescheduled due to a pNFS error, then we really
want to immediately start a new flush. The O_DIRECT code already does
this, so we only need to worry about buffered writes.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

If the client is sending a layoutget, but the server issues a callback
to recall what it thinks may be an outstanding layout, then we may find
an uninitialised layout attached to the inode due to the layoutget.
In that case, it is appropriate to return NFS4ERR_NOMATCHING_LAYOUT
rather than NFS4ERR_DELAY, as the latter can end up deadlocking.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

Even if the results of the permissions checks failed, we should parse
the results of the layout on open call so that we can return the
layout if required.
Note that we also want to ignore the sequence counter for whether or not
a layout recall occurred. If the recall pertained to our OPEN, then the
callback will know, and will attempt to wait for us to finih processing
anyway.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

There should be no need to grab the inode lock if we're only reading
the file offset.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

If the old layout was recalled, and we returned NFS4ERR_NOMATCHINGLAYOUT
then we need to wait for all outstanding layoutget calls to complete
before we can send a new one.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

If a layout segment is carrying layoutstats or layout error information,
then we always want to return it rather than using a forgetful model.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

If a layout has been recalled, then we should fire off a layoutreturn as
soon as all the layout segments that match the recall have been retired.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

RFC5661 doesn't state directly that the client should update the layout
stateid if it returns NFS4ERR_NOMATCHING_LAYOUT in response to a recall,
however it does state that this error will "cleanly indicate completion"
on par with returning the layout. For this reason, we assume that the
client should update the layout stateid. The Linux pNFS server definitely
does expect this behaviour.

However, if the client replies NFS4ERR_DELAY, then it is stating that
the recall was not processed, so it would be very wrong to update the
layout stateid.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

If there are layout segments that are marked for return, then we need
to ensure that pnfs_mark_matching_lsegs_return() does not just
silently discard them, but it should tell the caller that there is a
layoutreturn scheduled.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

Pull USB fixes from Greg KH:
 "Here are a number of USB fixes and new device ids for 4.18-rc7.

  The largest number are a bunch of gadget driver fixes that got delayed
  in being submitted earlier due to vacation schedules, but nothing
  really huge is present in them. There are some new device ids and some
  PHY driver fixes that were connected to some USB ones. Full details
  are in the shortlog.

  All have been in linux-next for a while with no reported issues"

* tag 'usb-4.18-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb: (28 commits)
  usb: core: handle hub C_PORT_OVER_CURRENT condition
  usb: xhci: Fix memory leak in xhci_endpoint_reset()
  usb: typec: tcpm: Fix sink PDO starting index for PPS APDO selection
  usb: gadget: f_fs: Only return delayed status when len is 0
  usb: gadget: f_uac2: fix endianness of 'struct cntrl_*_lay3'
  usb: dwc2: Fix inefficient copy of unaligned buffers
  usb: dwc2: Fix DMA alignment to start at allocated boundary
  usb: dwc3: rockchip: Fix PHY documentation links.
  tools: usb: ffs-test: Fix build on big endian systems
  usb: gadget: aspeed: Workaround memory ordering issue
  usb: dwc3: gadget: remove redundant variable maxpacket
  usb: dwc2: avoid NULL dereferences
  usb/phy: fix PPC64 build errors in phy-fsl-usb.c
  usb: dwc2: host: do not delay retries for CONTROL IN transfers
  usb: gadget: u_audio: protect stream runtime fields with stream spinlock
  usb: gadget: u_audio: remove cached period bytes value
  usb: gadget: u_audio: remove caching of stream buffer parameters
  usb: gadget: u_audio: update hw_ptr in iso_complete after data copied
  usb: gadget: u_audio: fix pcm/card naming in g_audio_setup()
  usb: gadget: f_uac2: fix error handling in afunc_bind (again)
  ...

Pull staging driver fixes from Greg KH:
 "Here are three small staging driver fixes for 4.18-rc7.

  One is a revert of an earlier patch that turned out to be incorrect,
  one is a fix for the speakup drivers, and the last a fix for the
  ks7010 driver to resolve a regression.

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'staging-4.18-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  staging: speakup: fix wraparound in uaccess length check
  staging: ks7010: call 'hostif_mib_set_request_int' instead of 'hostif_mib_set_request_bool'
  Revert "staging:r8188eu: Use lib80211 to support TKIP"

Merge tag 'driver-core-4.18-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core

Pull driver core fix from Greg KH:
 "This is a single driver core fix for 4.18-rc7. It partially reverts a
  previous commit to resolve some reported issues.

  It has been in linux-next for a while now with no reported issues"

* tag 'driver-core-4.18-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
  driver core: Partially revert "driver core: correct device's shutdown order"

Pull ACPI fix from Rafael Wysocki:
 "Fix a recent ACPICA regression causing the AML parser to get confused
  and fail in some situations involving incorrect AML in an ACPI table
  (Erik Schmauss)"

* tag 'acpi-4.18-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPICA: AML Parser: ignore dispatcher error status during table load

Pull power management fix from Rafael Wysocki:
 "Fix up the recently introduced cpufreq driver for Qualcomm Kryo
  processors by adding a terminating NULL entry to its table of device
  IDs (YueHaibing)"

* tag 'pm-4.18-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  cpufreq: qcom-kryo: add NULL entry to the end of_device_id array

Pull clk fixes from Stephen Boyd:
 "One more round of updates for problems seen this -rc series. Drivers
  fixes are:

   - Amlogic Meson audio divider fix and CPU clk critical marking

   - Qualcomm multimedia GDSC marked as 'always on' to keep display
     working

   - Aspeed fixes for critical clks, resets causing clks to stay
     disabled, and an incorrect HPLL frequency calculation

   - Marvell Armada 3700 cpu clks would undervolt when switching from
     low frequencies to high frequencies because the voltage didn't
     stabilize in time so now we switch to an intermediate frequency

  Plus we have a core framework thinko that messed up the debugfs flag
  printing logic to make it not very useful"

* tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
  clk: aspeed: Support HPLL strapping on ast2400
  clk: mvebu: armada-37xx-periph: Fix switching CPU rate from 300Mhz to 1.2GHz
  clk: aspeed: Mark bclk (PCIe) and dclk (VGA) as critical
  clk/mmcc-msm8996: Make mmagic_bimc_gdsc ALWAYS_ON
  clk: aspeed: Treat a gate in reset as disabled
  clk: Really show symbolic clock flags in debugfs
  clk: qcom: gcc-msm8996: Disable halt check on UFS tx clock
  clk: meson: audio-divider is one based
  clk: meson-gxbb: set fclk_div2 as CLK_IS_CRITICAL

Merge tag 'fscache-fixes-20180725' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs

Pull fscache/cachefiles fixes from David Howells:

 - Allow cancelled operations to be queued so they can be cleaned up.

 - Fix a refcounting bug in the monitoring of reads on backend files
   whereby a race can occur between monitor objects being listed for
   work, the work processing being queued and the work processor running
   and destroying the monitor objects.

 - Fix a ref overput in object attachment, whereby a tentatively
   considered object is put in error handling without first being 'got'.

 - Fix a missing clear of the CACHEFILES_OBJECT_ACTIVE flag whereby an
   assertion occurs when we retry because it seems the object is now
   active.

 - Wait rather BUG'ing on an object collision in the depths of
   cachefiles as the active object should be being cleaned up - also
   depends on the one above.

* tag 'fscache-fixes-20180725' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs:
  cachefiles: Wait rather than BUG'ing on "Unexpected object collision"
  cachefiles: Fix missing clear of the CACHEFILES_OBJECT_ACTIVE flag
  fscache: Fix reference overput in fscache_attach_object() error handling
  cachefiles: Fix refcounting bug in backing-file read monitoring
  fscache: Allow cancelled operations to be enqueued

If we meet a conflicting object that is marked FSCACHE_OBJECT_IS_LIVE in
the active object tree, we have been emitting a BUG after logging
information about it and the new object.

Instead, we should wait for the CACHEFILES_OBJECT_ACTIVE flag to be cleared
on the old object (or return an error).  The ACTIVE flag should be cleared
after it has been removed from the active object tree.  A timeout of 60s is
used in the wait, so we shouldn't be able to get stuck there.

Fixes: 9ae326a6 ("CacheFiles: A cache that backs onto a mounted filesystem")
Signed-off-by: Kiran Kumar Modukuri <kiran.modukuri@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>

In cachefiles_mark_object_active(), the new object is marked active and
then we try to add it to the active object tree.  If a conflicting object
is already present, we want to wait for that to go away.  After the wait,
we go round again and try to re-mark the object as being active - but it's
already marked active from the first time we went through and a BUG is
issued.

Fix this by clearing the CACHEFILES_OBJECT_ACTIVE flag before we try again.

Analysis from Kiran Kumar Modukuri:

[Impact]
Oops during heavy NFS + FSCache + Cachefiles

CacheFiles: Error: Overlong wait for old active object to go away.

BUG: unable to handle kernel NULL pointer dereference at 0000000000000002

CacheFiles: Error: Object already active kernel BUG at
fs/cachefiles/namei.c:163!

[Cause]
In a heavily loaded system with big files being read and truncated, an
fscache object for a cookie is being dropped and a new object being
looked. The new object being looked for has to wait for the old object
to go away before the new object is moved to active state.

[Fix]
Clear the flag 'CACHEFILES_OBJECT_ACTIVE' for the new object when
retrying the object lookup.

[Testcase]
Have run ~100 hours of NFS stress tests and have not seen this bug recur.

[Regression Potential]
 - Limited to fscache/cachefiles.

Fixes: 9ae326a6 ("CacheFiles: A cache that backs onto a mounted filesystem")
Signed-off-by: Kiran Kumar Modukuri <kiran.modukuri@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>

When a cookie is allocated that causes fscache_object structs to be
allocated, those objects are initialised with the cookie pointer, but
aren't blessed with a ref on that cookie unless the attachment is
successfully completed in fscache_attach_object().

If attachment fails because the parent object was dying or there was a
collision, fscache_attach_object() returns without incrementing the cookie
counter - but upon failure of this function, the object is released which
then puts the cookie, whether or not a ref was taken on the cookie.

Fix this by taking a ref on the cookie when it is assigned in
fscache_object_init(), even when we're creating a root object.


Analysis from Kiran Kumar:

This bug has been seen in 4.4.0-124-generic #148-Ubuntu kernel

BugLink: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1776277

fscache cookie ref count updated incorrectly during fscache object
allocation resulting in following Oops.

kernel BUG at /build/linux-Y09MKI/linux-4.4.0/fs/fscache/internal.h:321!
kernel BUG at /build/linux-Y09MKI/linux-4.4.0/fs/fscache/cookie.c:639!

[Cause]
Two threads are trying to do operate on a cookie and two objects.

(1) One thread tries to unmount the filesystem and in process goes over a
    huge list of objects marking them dead and deleting the objects.
    cookie->usage is also decremented in following path:

      nfs_fscache_release_super_cookie
       -> __fscache_relinquish_cookie
        ->__fscache_cookie_put
        ->BUG_ON(atomic_read(&cookie->usage) <= 0);

(2) A second thread tries to lookup an object for reading data in following
    path:

    fscache_alloc_object
    1) cachefiles_alloc_object
        -> fscache_object_init
           -> assign cookie, but usage not bumped.
    2) fscache_attach_object -> fails in cant_attach_object because the
         cookie's backing object or cookie's->parent object are going away
    3) fscache_put_object
        -> cachefiles_put_object
          ->fscache_object_destroy
            ->fscache_cookie_put
               ->BUG_ON(atomic_read(&cookie->usage) <= 0);

[NOTE from dhowells] It's unclear as to the circumstances in which (2) can
take place, given that thread (1) is in nfs_kill_super(), however a
conflicting NFS mount with slightly different parameters that creates a
different superblock would do it.  A backtrace from Kiran seems to show
that this is a possibility:

    kernel BUG at/build/linux-Y09MKI/linux-4.4.0/fs/fscache/cookie.c:639!
    ...
    RIP: __fscache_cookie_put+0x3a/0x40 [fscache]
    Call Trace:
     __fscache_relinquish_cookie+0x87/0x120 [fscache]
     nfs_fscache_release_super_cookie+0x2d/0xb0 [nfs]
     nfs_kill_super+0x29/0x40 [nfs]
     deactivate_locked_super+0x48/0x80
     deactivate_super+0x5c/0x60
     cleanup_mnt+0x3f/0x90
     __cleanup_mnt+0x12/0x20
     task_work_run+0x86/0xb0
     exit_to_usermode_loop+0xc2/0xd0
     syscall_return_slowpath+0x4e/0x60
     int_ret_from_sys_call+0x25/0x9f

[Fix] Bump up the cookie usage in fscache_object_init, when it is first
being assigned a cookie atomically such that the cookie is added and bumped
up if its refcount is not zero.  Remove the assignment in
fscache_attach_object().

[Testcase]
I have run ~100 hours of NFS stress tests and not seen this bug recur.

[Regression Potential]
 - Limited to fscache/cachefiles.

Fixes: ccc4fc3d ("FS-Cache: Implement the cookie management part of the netfs API")
Signed-off-by: Kiran Kumar Modukuri <kiran.modukuri@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>

cachefiles_read_waiter() has the right to access a 'monitor' object by
virtue of being called under the waitqueue lock for one of the pages in its
purview.  However, it has no ref on that monitor object or on the
associated operation.

What it is allowed to do is to move the monitor object to the operation's
to_do list, but once it drops the work_lock, it's actually no longer
permitted to access that object.  However, it is trying to enqueue the
retrieval operation for processing - but it can only do this via a pointer
in the monitor object, something it shouldn't be doing.

If it doesn't enqueue the operation, the operation may not get processed.
If the order is flipped so that the enqueue is first, then it's possible
for the work processor to look at the to_do list before the monitor is
enqueued upon it.

Fix this by getting a ref on the operation so that we can trust that it
will still be there once we've added the monitor to the to_do list and
dropped the work_lock.  The op can then be enqueued after the lock is
dropped.

The bug can manifest in one of a couple of ways.  The first manifestation
looks like:

 FS-Cache:
 FS-Cache: Assertion failed
 FS-Cache: 6 == 5 is false
 ------------[ cut here ]------------
 kernel BUG at fs/fscache/operation.c:494!
 RIP: 0010:fscache_put_operation+0x1e3/0x1f0
 ...
 fscache_op_work_func+0x26/0x50
 process_one_work+0x131/0x290
 worker_thread+0x45/0x360
 kthread+0xf8/0x130
 ? create_worker+0x190/0x190
 ? kthread_cancel_work_sync+0x10/0x10
 ret_from_fork+0x1f/0x30

This is due to the operation being in the DEAD state (6) rather than
INITIALISED, COMPLETE or CANCELLED (5) because it's already passed through
fscache_put_operation().

The bug can also manifest like the following:

 kernel BUG at fs/fscache/operation.c:69!
 ...
    [exception RIP: fscache_enqueue_operation+246]
 ...
 #7 [ffff883fff083c10] fscache_enqueue_operation at ffffffffa0b793c6
 #8 [ffff883fff083c28] cachefiles_read_waiter at ffffffffa0b15a48
 #9 [ffff883fff083c48] __wake_up_common at ffffffff810af028

I'm not entirely certain as to which is line 69 in Lei's kernel, so I'm not
entirely clear which assertion failed.

Fixes: 9ae326a6 ("CacheFiles: A cache that backs onto a mounted filesystem")
Reported-by: Lei Xue <carmark.dlut@gmail.com>
Reported-by: Vegard Nossum <vegard.nossum@gmail.com>
Reported-by: Anthony DeRobertis <aderobertis@metrics.net>
Reported-by: NeilBrown <neilb@suse.com>
Reported-by: Daniel Axtens <dja@axtens.net>
Reported-by: Kiran Kumar Modukuri <kiran.modukuri@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Daniel Axtens <dja@axtens.net>