linux.git/fs/fscache/object.c, branch v4.13

fscache: Fix dead object requeue

2017-01-31T18:23:09+00:00

Under some circumstances, an fscache object can become queued such that it
fscache_object_work_func() can be called once the object is in the
OBJECT_DEAD state.  This results in the kernel oopsing when it tries to
invoke the handler for the state (which is hard coded to 0x2).

The way this comes about is something like the following:

 (1) The object dispatcher is processing a work state for an object.  This
     is done in workqueue context.

 (2) An out-of-band event comes in that isn't masked, causing the object to
     be queued, say EV_KILL.

 (3) The object dispatcher finishes processing the current work state on
     that object and then sees there's another event to process, so,
     without returning to the workqueue core, it processes that event too.
     It then follows the chain of events that initiates until we reach
     OBJECT_DEAD without going through a wait state (such as
     WAIT_FOR_CLEARANCE).

     At this point, object->events may be 0, object->event_mask will be 0
     and oob_event_mask will be 0.

 (4) The object dispatcher returns to the workqueue processor, and in due
     course, this sees that the object's work item is still queued and
     invokes it again.

 (5) The current state is a work state (OBJECT_DEAD), so the dispatcher
     jumps to it - resulting in an OOPS.

When I'm seeing this, the work state in (1) appears to have been either
LOOK_UP_OBJECT or CREATE_OBJECT (object->oob_table is
fscache_osm_lookup_oob).

The window for (2) is very small:

 (A) object->event_mask is cleared whilst the event dispatch process is
     underway - though there's no memory barrier to force this to the top
     of the function.

     The window, therefore is from the time the object was selected by the
     workqueue processor and made requeueable to the time the mask was
     cleared.

 (B) fscache_raise_event() will only queue the object if it manages to set
     the event bit and the corresponding event_mask bit was set.

     The enqueuement is then deferred slightly whilst we get a ref on the
     object and get the per-CPU variable for workqueue congestion.  This
     slight deferral slightly increases the probability by allowing extra
     time for the workqueue to make the item requeueable.

Handle this by giving the dead state a processor function and checking the
for the dead state address rather than seeing if the processor function is
address 0x2.  The dead state processor function can then set a flag to
indicate that it's occurred and give a warning if it occurs more than once
per object.

If this race occurs, an oops similar to the following is seen (note the RIP
value):

BUG: unable to handle kernel NULL pointer dereference at 0000000000000002
IP: [<0000000000000002>] 0x1
PGD 0
Oops: 0010 [#1] SMP
Modules linked in: ...
CPU: 17 PID: 16077 Comm: kworker/u48:9 Not tainted 3.10.0-327.18.2.el7.x86_64 #1
Hardware name: HP ProLiant DL380 Gen9/ProLiant DL380 Gen9, BIOS P89 12/27/2015
Workqueue: fscache_object fscache_object_work_func [fscache]
task: ffff880302b63980 ti: ffff880717544000 task.ti: ffff880717544000
RIP: 0010:[<0000000000000002>]  [<0000000000000002>] 0x1
RSP: 0018:ffff880717547df8  EFLAGS: 00010202
RAX: ffffffffa0368640 RBX: ffff880edf7a4480 RCX: dead000000200200
RDX: 0000000000000002 RSI: 00000000ffffffff RDI: ffff880edf7a4480
RBP: ffff880717547e18 R08: 0000000000000000 R09: dfc40a25cb3a4510
R10: dfc40a25cb3a4510 R11: 0000000000000400 R12: 0000000000000000
R13: ffff880edf7a4510 R14: ffff8817f6153400 R15: 0000000000000600
FS:  0000000000000000(0000) GS:ffff88181f420000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000002 CR3: 000000000194a000 CR4: 00000000001407e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Stack:
 ffffffffa0363695 ffff880edf7a4510 ffff88093f16f900 ffff8817faa4ec00
 ffff880717547e60 ffffffff8109d5db 00000000faa4ec18 0000000000000000
 ffff8817faa4ec18 ffff88093f16f930 ffff880302b63980 ffff88093f16f900
Call Trace:
 [] ? fscache_object_work_func+0xa5/0x200 [fscache]
 [] process_one_work+0x17b/0x470
 [] worker_thread+0x21c/0x400
 [] ? rescuer_thread+0x400/0x400
 [] kthread+0xcf/0xe0
 [] ? kthread_create_on_node+0x140/0x140
 [] ret_from_fork+0x58/0x90
 [] ? kthread_create_on_node+0x140/0x140

Signed-off-by: David Howells 
Acked-by: Jeremy McNicoll 
Tested-by: Frank Sorenson 
Tested-by: Benjamin Coddington 
Reviewed-by: Benjamin Coddington 
Signed-off-by: Al Viro

fscache: Clear outstanding writes when disabling a cookie

2017-01-31T18:23:09+00:00

fscache_disable_cookie() needs to clear the outstanding writes on the
cookie it's disabling because they cannot be completed after.

Without this, fscache_nfs_open_file() gets stuck because it disables the
cookie when the file is opened for writing but can't uncache the pages till
afterwards - otherwise there's a race between the open routine and anyone
who already has it open R/O and is still reading from it.

Looking in /proc/pid/stack of the offending process shows:

[] __fscache_wait_on_page_write+0x82/0x9b [fscache]
[] __fscache_uncache_all_inode_pages+0x91/0xe1 [fscache]
[] nfs_fscache_open_file+0x59/0x9e [nfs]
[] nfs4_file_open+0x17f/0x1b8 [nfsv4]
[] do_dentry_open+0x16d/0x2b7
[] vfs_open+0x5c/0x65
[] path_openat+0x785/0x8fb
[] do_filp_open+0x48/0x9e
[] do_sys_open+0x13b/0x1cb
[] SyS_open+0x19/0x1b
[] do_syscall_64+0x80/0x17a
[] return_from_SYSCALL_64+0x0/0x7a
[] 0xffffffffffffffff

Reported-by: Jianhong Yin 
Signed-off-by: David Howells 
Acked-by: Jeff Layton 
Acked-by: Steve Dickson 
Signed-off-by: Al Viro

FS-Cache: The operation cancellation method needs calling in more places

2015-04-02T13:28:53+00:00

Any time an incomplete operation is cancelled, the operation cancellation
function needs to be called to clean up.  This is currently being passed
directly to some of the functions that might want to call it, but not all.

Instead, pass the cancellation method pointer to the fscache_operation_init()
and have that cache it in the operation struct.  Further, plug in a dummy
cancellation handler if the caller declines to set one as this allows us to
call the function unconditionally (the extra overhead isn't worth bothering
about as we don't expect to be calling this typically).

The cancellation method must thence be called everywhere the CANCELLED state
is set.  Note that we call it *before* setting the CANCELLED state such that
the method can use the old state value to guide its operation.

fscache_do_cancel_retrieval() needs moving higher up in the sources so that
the init function can use it now.

Without this, the following oops may be seen:

	FS-Cache: Assertion failed
	FS-Cache: 3 == 0 is false
	------------[ cut here ]------------
	kernel BUG at ../fs/fscache/page.c:261!
	...
	RIP: 0010:[]  fscache_release_retrieval_op+0x77/0x100
	 [] fscache_put_operation+0x114/0x2da
	 [] __fscache_read_or_alloc_pages+0x358/0x3b3
	 [] __nfs_readpages_from_fscache+0x59/0xbf [nfs]
	 [] nfs_readpages+0x10c/0x185 [nfs]
	 [] ? alloc_pages_current+0x119/0x13e
	 [] ? __page_cache_alloc+0xfb/0x10a
	 [] __do_page_cache_readahead+0x188/0x22c
	 [] ondemand_readahead+0x29e/0x2af
	 [] page_cache_sync_readahead+0x38/0x3a
	 [] generic_file_read_iter+0x1a2/0x55a
	 [] ? nfs_revalidate_mapping+0xd6/0x288 [nfs]
	 [] nfs_file_read+0x49/0x70 [nfs]
	 [] new_sync_read+0x78/0x9c
	 [] __vfs_read+0x13/0x38
	 [] vfs_read+0x95/0x121
	 [] SyS_read+0x4c/0x8a
	 [] system_call_fastpath+0x12/0x17

The assertion is showing that the remaining number of pages (n_pages) is not 0
when the operation is being released.

Signed-off-by: David Howells 
Reviewed-by: Steve Dickson 
Acked-by: Jeff Layton

FS-Cache: Synchronise object death state change vs operation submission

2015-04-02T13:28:53+00:00

When an object is being marked as no longer live, do this under the object
spinlock to prevent a race with operation submission targeted on that object.

The problem occurs due to the following pair of intertwined sequences when the
cache tries to create an object that would take it over the hard available
space limit:

 NETFS INTERFACE
 ===============
 (A) The netfs calls fscache_acquire_cookie().  object creation is deferred to
     the object state machine and the netfs is allowed to continue.

	OBJECT STATE MACHINE KTHREAD
	============================
	(1) The object is looked up on disk by fscache_look_up_object()
	    calling cachefiles_walk_to_object().  The latter finds that the
	    object is not yet represented on disk and calls
	    fscache_object_lookup_negative().

	(2) fscache_object_lookup_negative() sets FSCACHE_COOKIE_NO_DATA_YET
	    and clears FSCACHE_COOKIE_LOOKING_UP, thus allowing the netfs to
	    start queuing read operations.

 (B) The netfs calls fscache_read_or_alloc_pages().  This calls
     fscache_wait_for_deferred_lookup() which sees FSCACHE_COOKIE_LOOKING_UP
     become clear, allowing the read to begin.

 (C) A read operation is set up and passed to fscache_submit_op() to deal
     with.

	(3) cachefiles_walk_to_object() calls cachefiles_has_space(), which
	    fails (or one of the file operations to create stuff fails).
	    cachefiles returns an error to fscache.

	(4) fscache_look_up_object() transits to the LOOKUP_FAILURE state,

	(5) fscache_lookup_failure() sets FSCACHE_OBJECT_LOOKED_UP and
	    FSCACHE_COOKIE_UNAVAILABLE and clears FSCACHE_COOKIE_LOOKING_UP
	    then transits to the KILL_OBJECT state.

	(6) fscache_kill_object() clears FSCACHE_OBJECT_IS_LIVE in an attempt
	    to reject any further requests from the netfs.

	(7) object->n_ops is examined and found to be 0.
	    fscache_kill_object() transits to the DROP_OBJECT state.

 (D) fscache_submit_op() locks the object spinlock, sees if it can dispatch
     the op immediately by calling fscache_object_is_active() - which fails
     since FSCACHE_OBJECT_IS_AVAILABLE has not yet been set.

 (E) fscache_submit_op() then tests FSCACHE_OBJECT_LOOKED_UP - which is set.
     It then queues the object and increments object->n_ops.

	(8) fscache_drop_object() releases the object and eventually
	    fscache_put_object() calls cachefiles_put_object() which suffers
	    an assertion failure here:

		ASSERTCMP(object->fscache.n_ops, ==, 0);

Locking the object spinlock in step (6) around the clearance of
FSCACHE_OBJECT_IS_LIVE ensures that the the decision trees in
fscache_submit_op() and fscache_submit_exclusive_op() don't see the IS_LIVE
flag being cleared mid-decision: either the op is queued before step (7) - in
which case fscache_kill_object() will see n_ops>0 and will deal with the op -
or the op will be rejected.

This, combined with rejecting op submission if the target object is dying, fix
the problem.

The problem shows up as the following oops:

CacheFiles: Assertion failed
CacheFiles: 1 == 0 is false
------------[ cut here ]------------
kernel BUG at ../fs/cachefiles/interface.c:339!
...
RIP: 0010:[]  [] cachefiles_put_object+0x2a4/0x301 [cachefiles]
...
Call Trace:
 [] fscache_put_object+0x18/0x21 [fscache]
 [] fscache_object_work_func+0x3ba/0x3c9 [fscache]
 [] process_one_work+0x226/0x441
 [] worker_thread+0x273/0x36b
 [] ? rescuer_thread+0x2e1/0x2e1
 [] kthread+0x10e/0x116
 [] ? kthread_create_on_node+0x1bb/0x1bb
 [] ret_from_fork+0x7c/0xb0
 [] ? kthread_create_on_node+0x1bb/0x1bb

Signed-off-by: David Howells 
Reviewed-by: Steve Dickson 
Acked-by: Jeff Layton

FS-Cache: Handle a new operation submitted against a killed object

2015-04-02T13:28:53+00:00

Reject new operations that are being submitted against an object if that
object has failed its lookup or creation states or has been killed by the
cache backend for some other reason, such as having been culled.

Signed-off-by: David Howells 
Reviewed-by: Steve Dickson 
Acked-by: Jeff Layton

FS-Cache: Count culled objects and objects rejected due to lack of space

2015-02-24T10:05:27+00:00

Count the number of objects that get culled by the cache backend and the
number of objects that the cache backend declines to instantiate due to lack
of space in the cache.

These numbers are made available through /proc/fs/fscache/stats

Signed-off-by: David Howells 
Reviewed-by: Steve Dickson 
Acked-by: Jeff Layton

FS-Cache: Reduce cookie ref count if submit fails.

2014-08-27T14:29:34+00:00

I've been seeing issues with disposing cookies under vma pressure. The symptom
is that the refcount gets out of sync. In this case we fail to decrement the
refcount if submit fails. I found this while auditing the error in and around
cookie operations.

Signed-off-by: Milosz Tanski 
Signed-off-by: David Howells

FS-Cache: Handle removal of unadded object to the fscache_object_list rb tree

2014-02-17T21:47:35+00:00

When FS-Cache allocates an object, the following sequence of events can
occur:

 -->fscache_alloc_object()
    -->cachefiles_alloc_object() [via cache->ops->alloc_object]
    <--[returns new object]
    -->fscache_attach_object()
    <--[failed]
    -->cachefiles_put_object() [via cache->ops->put_object]
       -->fscache_object_destroy()
          -->fscache_objlist_remove()
             -->rb_erase() to remove the object from fscache_object_list.

resulting in a crash in the rbtree code.

The problem is that the object is only added to fscache_object_list on
the success path of fscache_attach_object() where it calls
fscache_objlist_add().

So if fscache_attach_object() fails, the object won't have been added to
the objlist rbtree.  We do, however, unconditionally try to remove the
object from the tree.

Thanks to NeilBrown for finding this and suggesting this solution.

Reported-by: NeilBrown 
Signed-off-by: David Howells 
Tested-by: (a customer of) NeilBrown 
Signed-off-by: Linus Torvalds

Merge branch 'for-3.13/core' of git://git.kernel.dk/linux-block

2013-11-14T03:08:14+00:00

Pull block IO core updates from Jens Axboe:
 "This is the pull request for the core changes in the block layer for
  3.13.  It contains:

   - The new blk-mq request interface.

     This is a new and more scalable queueing model that marries the
     best part of the request based interface we currently have (which
     is fully featured, but scales poorly) and the bio based "interface"
     which the new drivers for high IOPS devices end up using because
     it's much faster than the request based one.

     The bio interface has no block layer support, since it taps into
     the stack much earlier.  This means that drivers end up having to
     implement a lot of functionality on their own, like tagging,
     timeout handling, requeue, etc.  The blk-mq interface provides all
     these.  Some drivers even provide a switch to select bio or rq and
     has code to handle both, since things like merging only works in
     the rq model and hence is faster for some workloads.  This is a
     huge mess.  Conversion of these drivers nets us a substantial code
     reduction.  Initial results on converting SCSI to this model even
     shows an 8x improvement on single queue devices.  So while the
     model was intended to work on the newer multiqueue devices, it has
     substantial improvements for "classic" hardware as well.  This code
     has gone through extensive testing and development, it's now ready
     to go.  A pull request is coming to convert virtio-blk to this
     model will be will be coming as well, with more drivers scheduled
     for 3.14 conversion.

   - Two blktrace fixes from Jan and Chen Gang.

   - A plug merge fix from Alireza Haghdoost.

   - Conversion of __get_cpu_var() from Christoph Lameter.

   - Fix for sector_div() with 64-bit divider from Geert Uytterhoeven.

   - A fix for a race between request completion and the timeout
     handling from Jeff Moyer.  This is what caused the merge conflict
     with blk-mq/core, in case you are looking at that.

   - A dm stacking fix from Mike Snitzer.

   - A code consolidation fix and duplicated code removal from Kent
     Overstreet.

   - A handful of block bug fixes from Mikulas Patocka, fixing a loop
     crash and memory corruption on blk cg.

   - Elevator switch bug fix from Tomoki Sekiyama.

  A heads-up that I had to rebase this branch.  Initially the immutable
  bio_vecs had been queued up for inclusion, but a week later, it became
  clear that it wasn't fully cooked yet.  So the decision was made to
  pull this out and postpone it until 3.14.  It was a straight forward
  rebase, just pruning out the immutable series and the later fixes of
  problems with it.  The rest of the patches applied directly and no
  further changes were made"

* 'for-3.13/core' of git://git.kernel.dk/linux-block: (31 commits)
  block: replace IS_ERR and PTR_ERR with PTR_ERR_OR_ZERO
  block: replace IS_ERR and PTR_ERR with PTR_ERR_OR_ZERO
  block: Do not call sector_div() with a 64-bit divisor
  kernel: trace: blktrace: remove redundent memcpy() in compat_blk_trace_setup()
  block: Consolidate duplicated bio_trim() implementations
  block: Use rw_copy_check_uvector()
  block: Enable sysfs nomerge control for I/O requests in the plug list
  block: properly stack underlying max_segment_size to DM device
  elevator: acquire q->sysfs_lock in elevator_change()
  elevator: Fix a race in elevator switching and md device initialization
  block: Replace __get_cpu_var uses
  bdi: test bdi_init failure
  block: fix a probe argument to blk_register_region
  loop: fix crash if blk_alloc_queue fails
  blk-core: Fix memory corruption if blkcg_init_queue fails
  block: fix race between request completion and timeout handling
  blktrace: Send BLK_TN_PROCESS events to all running traces
  blk-mq: don't disallow request merges for req->special being set
  blk-mq: mq plug list breakage
  blk-mq: fix for flush deadlock
  ...

block: Replace __get_cpu_var uses

2013-11-08T15:59:58+00:00

__get_cpu_var() is used for multiple purposes in the kernel source. One of
them is address calculation via the form &__get_cpu_var(x).  This calculates
the address for the instance of the percpu variable of the current processor
based on an offset.

Other use cases are for storing and retrieving data from the current
processors percpu area.  __get_cpu_var() can be used as an lvalue when
writing data or on the right side of an assignment.

__get_cpu_var() is defined as :

#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))

__get_cpu_var() always only does an address determination. However, store
and retrieve operations could use a segment prefix (or global register on
other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a
percpu area and use optimized assembly code to read and write per cpu
variables.

This patch converts __get_cpu_var into either an explicit address
calculation using this_cpu_ptr() or into a use of this_cpu operations that
use the offset.  Thereby address calculations are avoided and less registers
are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so
the macro is removed too.

The patch set includes passes over all arches as well. Once these operations
are used throughout then specialized macros can be defined in non -x86
arches as well in order to optimize per cpu access by f.e.  using a global
register that may be set to the per cpu base.

Transformations done to __get_cpu_var()

1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);

2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);

3. Retrieve the content of the current processors instance of a per cpu
variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);

4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));

5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);

6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)

Signed-off-by: Christoph Lameter 
Signed-off-by: Jens Axboe