[ Upstream commit 15a77c6fe4 ]
With >=32 CPUs the userfaultfd selftest triggered a graceful but
unexpected SIGBUS because VM_FAULT_RETRY was returned by
handle_userfault() despite the UFFDIO_COPY wasn't completed.
This seems caused by rwsem waking the thread blocked in
handle_userfault() and we can't run up_read() before the wait_event
sequence is complete.
Keeping the wait_even sequence identical to the first one, would require
running userfaultfd_must_wait() again to know if the loop should be
repeated, and it would also require retaking the rwsem and revalidating
the whole vma status.
It seems simpler to wait the targeted wakeup so that if false wakeups
materialize we still wait for our specific wakeup event, unless of
course there are signals or the uffd was released.
Debug code collecting the stack trace of the wakeup showed this:
$ ./userfaultfd 100 99999
nr_pages: 25600, nr_pages_per_cpu: 800
bounces: 99998, mode: racing ver poll, userfaults: 32 35 90 232 30 138 69 82 34 30 139 40 40 31 20 19 43 13 15 28 27 38 21 43 56 22 1 17 31 8 4 2
bounces: 99997, mode: rnd ver poll, Bus error (core dumped)
save_stack_trace+0x2b/0x50
try_to_wake_up+0x2a6/0x580
wake_up_q+0x32/0x70
rwsem_wake+0xe0/0x120
call_rwsem_wake+0x1b/0x30
up_write+0x3b/0x40
vm_mmap_pgoff+0x9c/0xc0
SyS_mmap_pgoff+0x1a9/0x240
SyS_mmap+0x22/0x30
entry_SYSCALL_64_fastpath+0x1f/0xbd
0xffffffffffffffff
FAULT_FLAG_ALLOW_RETRY missing 70
CPU: 24 PID: 1054 Comm: userfaultfd Tainted: G W 4.8.0+ #30
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.3-0-ge2fc41e-prebuilt.qemu-project.org 04/01/2014
Call Trace:
dump_stack+0xb8/0x112
handle_userfault+0x572/0x650
handle_mm_fault+0x12cb/0x1520
__do_page_fault+0x175/0x500
trace_do_page_fault+0x61/0x270
do_async_page_fault+0x19/0x90
async_page_fault+0x25/0x30
This always happens when the main userfault selftest thread is running
clone() while glibc runs either mprotect or mmap (both taking mmap_sem
down_write()) to allocate the thread stack of the background threads,
while locking/userfault threads already run at full throttle and are
susceptible to false wakeups that may cause handle_userfault() to return
before than expected (which results in graceful SIGBUS at the next
attempt).
This was reproduced only with >=32 CPUs because the loop to start the
thread where clone() is too quick with fewer CPUs, while with 32 CPUs
there's already significant activity on ~32 locking and userfault
threads when the last background threads are started with clone().
This >=32 CPUs SMP race condition is likely reproducible only with the
selftest because of the much heavier userfault load it generates if
compared to real apps.
We'll have to allow "one more" VM_FAULT_RETRY for the WP support and a
patch floating around that provides it also hidden this problem but in
reality only is successfully at hiding the problem.
False wakeups could still happen again the second time
handle_userfault() is invoked, even if it's a so rare race condition
that getting false wakeups twice in a row is impossible to reproduce.
This full fix is needed for correctness, the only alternative would be
to allow VM_FAULT_RETRY to be returned infinitely. With this fix the WP
support can stick to a strict "one more" VM_FAULT_RETRY logic (no need
of returning it infinite times to avoid the SIGBUS).
Link: http://lkml.kernel.org/r/20170111005535.13832-2-aarcange@redhat.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reported-by: Shubham Kumar Sharma <shubham.kumar.sharma@oracle.com>
Tested-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Michael Rapoport <RAPOPORT@il.ibm.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <alexander.levin@verizon.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit f598f82e20 ]
Commit 8a59f5d252 ("fs/romfs: return f_fsid for statfs(2)") generates
a 64bit id from sb->s_bdev->bd_dev. This is only correct when romfs is
defined with CONFIG_ROMFS_ON_BLOCK. If romfs is only defined with
CONFIG_ROMFS_ON_MTD, sb->s_bdev is NULL, referencing sb->s_bdev->bd_dev
will triger an oops.
Richard Weinberger points out that when CONFIG_ROMFS_BACKED_BY_BOTH=y,
both CONFIG_ROMFS_ON_BLOCK and CONFIG_ROMFS_ON_MTD are defined.
Therefore when calling huge_encode_dev() to generate a 64bit id, I use
the follow order to choose parameter,
- CONFIG_ROMFS_ON_BLOCK defined
use sb->s_bdev->bd_dev
- CONFIG_ROMFS_ON_BLOCK undefined and CONFIG_ROMFS_ON_MTD defined
use sb->s_dev when,
- both CONFIG_ROMFS_ON_BLOCK and CONFIG_ROMFS_ON_MTD undefined
leave id as 0
When CONFIG_ROMFS_ON_MTD is defined and sb->s_mtd is not NULL, sb->s_dev
is set to a device ID generated by MTD_BLOCK_MAJOR and mtd index,
otherwise sb->s_dev is 0.
This is a try-best effort to generate a uniq file system ID, if all the
above conditions are not meet, f_fsid of this romfs instance will be 0.
Generally only one romfs can be built on single MTD block device, this
method is enough to identify multiple romfs instances in a computer.
Link: http://lkml.kernel.org/r/1482928596-115155-1-git-send-email-colyli@suse.de
Signed-off-by: Coly Li <colyli@suse.de>
Reported-by: Nong Li <nongli1031@gmail.com>
Tested-by: Nong Li <nongli1031@gmail.com>
Cc: Richard Weinberger <richard.weinberger@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <alexander.levin@verizon.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 6bdded59c8 ]
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:
[<ffffffffa0142883>] __fscache_wait_on_page_write+0x82/0x9b [fscache]
[<ffffffffa014336e>] __fscache_uncache_all_inode_pages+0x91/0xe1 [fscache]
[<ffffffffa01740fa>] nfs_fscache_open_file+0x59/0x9e [nfs]
[<ffffffffa01ccf41>] nfs4_file_open+0x17f/0x1b8 [nfsv4]
[<ffffffff8117350e>] do_dentry_open+0x16d/0x2b7
[<ffffffff811743ac>] vfs_open+0x5c/0x65
[<ffffffff81184185>] path_openat+0x785/0x8fb
[<ffffffff81184343>] do_filp_open+0x48/0x9e
[<ffffffff81174710>] do_sys_open+0x13b/0x1cb
[<ffffffff811747b9>] SyS_open+0x19/0x1b
[<ffffffff81001c44>] do_syscall_64+0x80/0x17a
[<ffffffff8165c2da>] return_from_SYSCALL_64+0x0/0x7a
[<ffffffffffffffff>] 0xffffffffffffffff
Reported-by: Jianhong Yin <jiyin@redhat.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Jeff Layton <jlayton@redhat.com>
Acked-by: Steve Dickson <steved@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Sasha Levin <alexander.levin@verizon.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit e26bfebdfc ]
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:
[<ffffffffa0363695>] ? fscache_object_work_func+0xa5/0x200 [fscache]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e4ac>] worker_thread+0x21c/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff816460d8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Jeremy McNicoll <jeremymc@redhat.com>
Tested-by: Frank Sorenson <sorenson@redhat.com>
Tested-by: Benjamin Coddington <bcodding@redhat.com>
Reviewed-by: Benjamin Coddington <bcodding@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Sasha Levin <alexander.levin@verizon.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 6a25478077 ]
The function glock_hash_walk walks the rhashtable by hand. This
is broken because if it catches the hash table in the middle of
a rehash, then it will miss entries.
This patch replaces the manual walk by using the rhashtable walk
interface.
Fixes: 88ffbf3e03 ("GFS2: Use resizable hash table for glocks")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <alexander.levin@verizon.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit cc2b702c52 upstream.
Variables start_idx and end_idx are supposed to hold a page index
derived from the file offsets. The int type is not the right one though,
offsets larger than 1 << 44 will get silently trimmed off the high bits.
(1 << 44 is 16TiB)
What can go wrong, if start is below the boundary and end gets trimmed:
- if there's a page after start, we'll find it (radix_tree_gang_lookup_slot)
- the final check "if (page->index <= end_idx)" will unexpectedly fail
The function will return false, ie. "there's no page in the range",
although there is at least one.
btrfs_page_exists_in_range is used to prevent races in:
* in hole punching, where we make sure there are not pages in the
truncated range, otherwise we'll wait for them to finish and redo
truncation, but we're going to replace the pages with holes anyway so
the only problem is the intermediate state
* lock_extent_direct: we want to make sure there are no pages before we
lock and start DIO, to prevent stale data reads
For practical occurence of the bug, there are several constaints. The
file must be quite large, the affected range must cross the 16TiB
boundary and the internal state of the file pages and pending operations
must match. Also, we must not have started any ordered data in the
range, otherwise we don't even reach the buggy function check.
DIO locking tries hard in several places to avoid deadlocks with
buffered IO and avoids waiting for ranges. The worst consequence seems
to be stale data read.
CC: Liu Bo <bo.li.liu@oracle.com>
Fixes: fc4adbff82 ("btrfs: Drop EXTENT_UPTODATE check in hole punching and direct locking")
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 67a7d5f561 upstream.
Currently, extent manipulation operations such as hole punch, range
zeroing, or extent shifting do not record the fact that file data has
changed and thus fdatasync(2) has a work to do. As a result if we crash
e.g. after a punch hole and fdatasync, user can still possibly see the
punched out data after journal replay. Test generic/392 fails due to
these problems.
Fix the problem by properly marking that file data has changed in these
operations.
Fixes: a4bb6b64e3
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 4f8caa60a5 upstream.
When ext4_map_blocks() is called with EXT4_GET_BLOCKS_ZERO to zero-out
allocated blocks and these blocks are actually converted from unwritten
extent the following race can happen:
CPU0 CPU1
page fault page fault
... ...
ext4_map_blocks()
ext4_ext_map_blocks()
ext4_ext_handle_unwritten_extents()
ext4_ext_convert_to_initialized()
- zero out converted extent
ext4_zeroout_es()
- inserts extent as initialized in status tree
ext4_map_blocks()
ext4_es_lookup_extent()
- finds initialized extent
write data
ext4_issue_zeroout()
- zeroes out new extent overwriting data
This problem can be reproduced by generic/340 for the fallocated case
for the last block in the file.
Fix the problem by avoiding zeroing out the area we are mapping with
ext4_map_blocks() in ext4_ext_convert_to_initialized(). It is pointless
to zero out this area in the first place as the caller asked us to
convert the area to initialized because he is just going to write data
there before the transaction finishes. To achieve this we delete the
special case of zeroing out full extent as that will be handled by the
cases below zeroing only the part of the extent that needs it. We also
instruct ext4_split_extent() that the middle of extent being split
contains data so that ext4_split_extent_at() cannot zero out full extent
in case of ENOSPC.
Fixes: 12735f8819
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 7d95eddf31 upstream.
Currently, SEEK_HOLE implementation in ext4 may both return that there's
a hole at some offset although that offset already has data and skip
some holes during a search for the next hole. The first problem is
demostrated by:
xfs_io -c "falloc 0 256k" -c "pwrite 0 56k" -c "seek -h 0" file
wrote 57344/57344 bytes at offset 0
56 KiB, 14 ops; 0.0000 sec (2.054 GiB/sec and 538461.5385 ops/sec)
Whence Result
HOLE 0
Where we can see that SEEK_HOLE wrongly returned offset 0 as containing
a hole although we have written data there. The second problem can be
demonstrated by:
xfs_io -c "falloc 0 256k" -c "pwrite 0 56k" -c "pwrite 128k 8k"
-c "seek -h 0" file
wrote 57344/57344 bytes at offset 0
56 KiB, 14 ops; 0.0000 sec (1.978 GiB/sec and 518518.5185 ops/sec)
wrote 8192/8192 bytes at offset 131072
8 KiB, 2 ops; 0.0000 sec (2 GiB/sec and 500000.0000 ops/sec)
Whence Result
HOLE 139264
Where we can see that hole at offsets 56k..128k has been ignored by the
SEEK_HOLE call.
The underlying problem is in the ext4_find_unwritten_pgoff() which is
just buggy. In some cases it fails to update returned offset when it
finds a hole (when no pages are found or when the first found page has
higher index than expected), in some cases conditions for detecting hole
are just missing (we fail to detect a situation where indices of
returned pages are not contiguous).
Fix ext4_find_unwritten_pgoff() to properly detect non-contiguous page
indices and also handle all cases where we got less pages then expected
in one place and handle it properly there.
Fixes: c8c0df241c
CC: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit b26b78cb72 upstream.
If an NFSv4 client asks us for the supattr_exclcreat, then we must
not return attributes that are unsupported by this minor version.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Fixes: 75976de655 ("NFSD: Return word2 bitmask if setting security..,")
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 9a307403d3 upstream.
if we receive a compound such that:
- the sessionid, slot, and sequence number in the SEQUENCE op
match a cached succesful reply with N ops, and
- the Nth operation of the compound is a PUTFH, PUTPUBFH,
PUTROOTFH, or RESTOREFH,
then nfsd4_sequence will return 0 and set cstate->status to
nfserr_replay_cache. The current filehandle will not be set. This will
cause us to call check_nfsd_access with first argument NULL.
To nfsd4_compound it looks like we just succesfully executed an
operation that set a filehandle, but the current filehandle is not set.
Fix this by moving the nfserr_replay_cache earlier. There was never any
reason to have it after the encode_op label, since the only case where
he hit that is when opdesc->op_func sets it.
Note that there are two ways we could hit this case:
- a client is resending a previously sent compound that ended
with one of the four PUTFH-like operations, or
- a client is sending a *new* compound that (incorrectly) shares
sessionid, slot, and sequence number with a previously sent
compound, and the length of the previously sent compound
happens to match the position of a PUTFH-like operation in the
new compound.
The second is obviously incorrect client behavior. The first is also
very strange--the only purpose of a PUTFH-like operation is to set the
current filehandle to be used by the following operation, so there's no
point in having it as the last in a compound.
So it's likely this requires a buggy or malicious client to reproduce.
Reported-by: Scott Mayhew <smayhew@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit d7fd24257a upstream.
There is an off-by-one error in loop termination conditions in
xfs_find_get_desired_pgoff() since 'end' may index a page beyond end of
desired range if 'endoff' is page aligned. It doesn't have any visible
effects but still it is good to fix it.
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit a4d768e702 upstream.
This structure copy was throwing unaligned access warnings on sparc64:
Kernel unaligned access at TPC[1043c088] xfs_btree_visit_blocks+0x88/0xe0 [xfs]
xfs_btree_copy_ptrs does a memcpy, which avoids it.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 3ecb3ac7b9 upstream.
If a malicious user corrupts the refcount btree to cause a cycle between
different levels of the tree, the next mount attempt will deadlock in
the CoW recovery routine while grabbing buffer locks. We can use the
ability to re-grab a buffer that was previous locked to a transaction to
avoid deadlocks, so do that here.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This is a partial cherry-pick of commit e89c041338
("xfs: implement the GETFSMAP ioctl"), which also adds this helper, and
a great example of why feature patches should be properly split into
their parts.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
[hch: split from the larger patch for -stable]
Signed-off-by: Christoph Hellwig <hch@lst.de>
commit 892d2a5f70 upstream.
By run fsstress long enough time enough in RHEL-7, I find an
assertion failure (harder to reproduce on linux-4.11, but problem
is still there):
XFS: Assertion failed: (iflags & BMV_IF_DELALLOC) != 0, file: fs/xfs/xfs_bmap_util.c
The assertion is in xfs_getbmap() funciton:
if (map[i].br_startblock == DELAYSTARTBLOCK &&
--> map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
ASSERT((iflags & BMV_IF_DELALLOC) != 0);
When map[i].br_startoff == XFS_B_TO_FSB(mp, XFS_ISIZE(ip)), the
startoff is just at EOF. But we only need to make sure delalloc
extents that are within EOF, not include EOF.
Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 6eadbf4c8b upstream.
When we're fulfilling a BMAPX request, jump out early if the data fork
is in local format. This prevents us from hitting a debugging check in
bmapi_read and barfing errors back to userspace. The on-disk extent
count check later isn't sufficient for IF_DELALLOC mode because da
extents are in memory and not on disk.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 0daaecacb8 upstream.
The delalloc -> real block conversion path uses an incorrect
calculation in the case where the middle part of a delalloc extent
is being converted. This is documented as a rare situation because
XFS generally attempts to maximize contiguity by converting as much
of a delalloc extent as possible.
If this situation does occur, the indlen reservation for the two new
delalloc extents left behind by the conversion of the middle range
is calculated and compared with the original reservation. If more
blocks are required, the delta is allocated from the global block
pool. This delta value can be characterized as the difference
between the new total requirement (temp + temp2) and the currently
available reservation minus those blocks that have already been
allocated (startblockval(PREV.br_startblock) - allocated).
The problem is that the current code does not account for previously
allocated blocks correctly. It subtracts the current allocation
count from the (new - old) delta rather than the old indlen
reservation. This means that more indlen blocks than have been
allocated end up stashed in the remaining extents and free space
accounting is broken as a result.
Fix up the calculation to subtract the allocated block count from
the original extent indlen and thus correctly allocate the
reservation delta based on the difference between the new total
requirement and the unused blocks from the original reservation.
Also remove a bogus assert that contradicts the fact that the new
indlen reservation can be larger than the original indlen
reservation.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit fe0be23e68 upstream.
In xfs_reflink_end_cow, we erroneously reserve only enough blocks to
handle adding 1 extent. This is problematic if we fragment free space,
have to do CoW, and then have to perform multiple bmap btree expansions.
Furthermore, the BUI recovery routine doesn't reserve /any/ blocks to
handle btree splits, so log recovery fails after our first error causes
the filesystem to go down.
Therefore, refactor the transaction block reservation macros until we
have a macro that works for our deferred (re)mapping activities, and fix
both problems by using that macro.
With 1k blocks we can hit this fairly often in g/187 if the scratch fs
is big enough.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit e20c8a517f upstream.
The quotaoff operation has a race with inode allocation that results
in a livelock. An inode allocation that occurs before the quota
status flags are updated acquires the appropriate dquots for the
inode via xfs_qm_vop_dqalloc(). It then inserts the XFS_INEW inode
into the perag radix tree, sometime later attaches the dquots to the
inode and finally clears the XFS_INEW flag. Quotaoff expects to
release the dquots from all inodes in the filesystem via
xfs_qm_dqrele_all_inodes(). This invokes the AG inode iterator,
which skips inodes in the XFS_INEW state because they are not fully
constructed. If the scan occurs after dquots have been attached to
an inode, but before XFS_INEW is cleared, the newly allocated inode
will continue to hold a reference to the applicable dquots. When
quotaoff invokes xfs_qm_dqpurge_all(), the reference count of those
dquot(s) remain elevated and the dqpurge scan spins indefinitely.
To address this problem, update the xfs_qm_dqrele_all_inodes() scan
to wait on inodes marked on the XFS_INEW state. We wait on the
inodes explicitly rather than skip and retry to avoid continuous
retry loops due to a parallel inode allocation workload. Since
quotaoff updates the quota state flags and uses a synchronous
transaction before the dqrele scan, and dquots are attached to
inodes after radix tree insertion iff quota is enabled, one INEW
waiting pass through the AG guarantees that the scan has processed
all inodes that could possibly hold dquot references.
Reported-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit ae2c4ac2dd upstream.
The AG inode iterator currently skips new inodes as such inodes are
inserted into the inode radix tree before they are fully
constructed. Certain contexts require the ability to wait on the
construction of new inodes, however. The fs-wide dquot release from
the quotaoff sequence is an example of this.
Update the AG inode iterator to support the ability to wait on
inodes flagged with XFS_INEW upon request. Create a new
xfs_inode_ag_iterator_flags() interface and support a set of
iteration flags to modify the iteration behavior. When the
XFS_AGITER_INEW_WAIT flag is set, include XFS_INEW flags in the
radix tree inode lookup and wait on them before the callback is
executed.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 756baca27f upstream.
Inodes that are inserted into the perag tree but still under
construction are flagged with the XFS_INEW bit. Most contexts either
skip such inodes when they are encountered or have the ability to
handle them.
The runtime quotaoff sequence introduces a context that must wait
for construction of such inodes to correctly ensure that all dquots
in the fs are released. In anticipation of this, support the ability
to wait on new inodes. Wake the appropriate bit when XFS_INEW is
cleared.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 20e8a06378 upstream.
The quotacheck error handling of the delwri buffer list assumes the
resident buffers are locked and doesn't clear the _XBF_DELWRI_Q flag
on the buffers that are dequeued. This can lead to assert failures
on buffer release and possibly other locking problems.
Move this code to a delwri queue cancel helper function to
encapsulate the logic required to properly release buffers from a
delwri queue. Update the helper to clear the delwri queue flag and
call it from quotacheck.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit cb52ee334a upstream.
Directory block readahead uses a complex iteration mechanism to map
between high-level directory blocks and underlying physical extents.
This mechanism attempts to traverse the higher-level dir blocks in a
manner that handles multi-fsb directory blocks and simultaneously
maintains a reference to the corresponding physical blocks.
This logic doesn't handle certain (discontiguous) physical extent
layouts correctly with multi-fsb directory blocks. For example,
consider the case of a 4k FSB filesystem with a 2 FSB (8k) directory
block size and a directory with the following extent layout:
EXT: FILE-OFFSET BLOCK-RANGE AG AG-OFFSET TOTAL
0: [0..7]: 88..95 0 (88..95) 8
1: [8..15]: 80..87 0 (80..87) 8
2: [16..39]: 168..191 0 (168..191) 24
3: [40..63]: 5242952..5242975 1 (72..95) 24
Directory block 0 spans physical extents 0 and 1, dirblk 1 lies
entirely within extent 2 and dirblk 2 spans extents 2 and 3. Because
extent 2 is larger than the directory block size, the readahead code
erroneously assumes the block is contiguous and issues a readahead
based on the physical mapping of the first fsb of the dirblk. This
results in read verifier failure and a spurious corruption or crc
failure, depending on the filesystem format.
Further, the subsequent readahead code responsible for walking
through the physical table doesn't correctly advance the physical
block reference for dirblk 2. Instead of advancing two physical
filesystem blocks, the first iteration of the loop advances 1 block
(correctly), but the subsequent iteration advances 2 more physical
blocks because the next physical extent (extent 3, above) happens to
cover more than dirblk 2. At this point, the higher-level directory
block walking is completely off the rails of the actual physical
layout of the directory for the respective mapping table.
Update the contiguous dirblock logic to consider the current offset
in the physical extent to avoid issuing directory readahead to
unrelated blocks. Also, update the mapping table advancing code to
consider the current offset within the current dirblock to avoid
advancing the mapping reference too far beyond the dirblock.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 023cc840b4 upstream.
Carlos had a case where "find" seemed to start spinning
forever and never return.
This was on a filesystem with non-default multi-fsb (8k)
directory blocks, and a fragmented directory with extents
like this:
0:[0,133646,2,0]
1:[2,195888,1,0]
2:[3,195890,1,0]
3:[4,195892,1,0]
4:[5,195894,1,0]
5:[6,195896,1,0]
6:[7,195898,1,0]
7:[8,195900,1,0]
8:[9,195902,1,0]
9:[10,195908,1,0]
10:[11,195910,1,0]
11:[12,195912,1,0]
12:[13,195914,1,0]
...
i.e. the first extent is a contiguous 2-fsb dir block, but
after that it is fragmented into 1 block extents.
At the top of the readdir path, we allocate a mapping array
which (for this filesystem geometry) can hold 10 extents; see
the assignment to map_info->map_size. During readdir, we are
therefore able to map extents 0 through 9 above into the array
for readahead purposes. If we count by 2, we see that the last
mapped index (9) is the first block of a 2-fsb directory block.
At the end of xfs_dir2_leaf_readbuf() we have 2 loops to fill
more readahead; the outer loop assumes one full dir block is
processed each loop iteration, and an inner loop that ensures
that this is so by advancing to the next extent until a full
directory block is mapped.
The problem is that this inner loop may step past the last
extent in the mapping array as it tries to reach the end of
the directory block. This will read garbage for the extent
length, and as a result the loop control variable 'j' may
become corrupted and never fail the loop conditional.
The number of valid mappings we have in our array is stored
in map->map_valid, so stop this inner loop based on that limit.
There is an ASSERT at the top of the outer loop for this
same condition, but we never made it out of the inner loop,
so the ASSERT never fired.
Huge appreciation for Carlos for debugging and isolating
the problem.
Debugged-and-analyzed-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Tested-by: Carlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: Bill O'Donnell <billodo@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 3b4683c294 upstream.
Lockdep complains about use of the iolock in inode reclaim context
because it doesn't understand that reclaim has the last reference to
the inode, and thus an iolock->reclaim->iolock deadlock is not
possible.
The iolock is technically not necessary in xfs_inactive() and was
only added to appease an assert in xfs_free_eofblocks(), which can
be called from other non-reclaim contexts. Therefore, just kill the
assert and drop the use of the iolock from reclaim context to quiet
lockdep.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 84358536dc upstream.
Apparently FIEMAP for xattrs has been broken since we switched to
the iomap backend because of an incorrect check for xattr presence.
Also fix the broken locking.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit be6324c00c upstream.
In xfs_ioc_getbmap, we should only copy the fields of struct getbmap
from userspace, or else we end up copying random stack contents into the
kernel. struct getbmap is a strict subset of getbmapx, so a partial
structure copy should work fine.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 696a562072 upstream.
The log covering background task used to be part of the xfssyncd
workqueue. That workqueue was removed as of commit 5889608df ("xfs:
syncd workqueue is no more") and the associated work item scheduled
to the xfs-log wq. The latter is used for log buffer I/O completion.
Since xfs_log_worker() can invoke a log flush, a deadlock is
possible between the xfs-log and xfs-cil workqueues. Consider the
following codepath from xfs_log_worker():
xfs_log_worker()
xfs_log_force()
_xfs_log_force()
xlog_cil_force()
xlog_cil_force_lsn()
xlog_cil_push_now()
flush_work()
The above is in xfs-log wq context and blocked waiting on the
completion of an xfs-cil work item. Concurrently, the cil push in
progress can end up blocked here:
xlog_cil_push_work()
xlog_cil_push()
xlog_write()
xlog_state_get_iclog_space()
xlog_wait(&log->l_flush_wait, ...)
The above is in xfs-cil context waiting on log buffer I/O
completion, which executes in xfs-log wq context. In this scenario
both workqueues are deadlocked waiting on eachother.
Add a new workqueue specifically for the high level log covering and
ail pushing worker, as was the case prior to commit 5889608df.
Diagnosed-by: David Jeffery <djeffery@redhat.com>
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit bf9216f922 upstream.
Fix a memory exposure problems in inumbers where we allocate an array of
structures with holes, fail to zero the holes, then blindly copy the
kernel memory contents (junk and all) into userspace.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 78420281a9 upstream.
The inline directory verifiers should be called on the inode fork data,
which means after iformat_local on the read side, and prior to
ifork_flush on the write side. This makes the fork verifier more
consistent with the way buffer verifiers work -- i.e. they will operate
on the memory buffer that the code will be reading and writing directly.
Furthermore, revise the verifier function to return -EFSCORRUPTED so
that we don't flood the logs with corruption messages and assert
notices. This has been a particular problem with xfs/348, which
triggers the XFS_WANT_CORRUPTED_RETURN assertions, which halts the
kernel when CONFIG_XFS_DEBUG=y. Disk corruption isn't supposed to do
that, at least not in a verifier.
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 630a04e79d upstream.
When we're reading or writing the data fork of an inline directory,
check the contents to make sure we're not overflowing buffers or eating
garbage data. xfs/348 corrupts an inline symlink into an inline
directory, triggering a buffer overflow bug.
v2: add more checks consistent with _dir2_sf_check and make the verifier
usable from anywhere.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 8affebe16d upstream.
xfs_find_get_desired_pgoff() is used to search for offset of hole or
data in page range [index, end] (both inclusive), and the max number
of pages to search should be at least one, if end == index.
Otherwise the only page is missed and no hole or data is found,
which is not correct.
When block size is smaller than page size, this can be demonstrated
by preallocating a file with size smaller than page size and writing
data to the last block. E.g. run this xfs_io command on a 1k block
size XFS on x86_64 host.
# xfs_io -fc "falloc 0 3k" -c "pwrite 2k 1k" \
-c "seek -d 0" /mnt/xfs/testfile
wrote 1024/1024 bytes at offset 2048
1 KiB, 1 ops; 0.0000 sec (33.675 MiB/sec and 34482.7586 ops/sec)
Whence Result
DATA EOF
Data at offset 2k was missed, and lseek(2) returned ENXIO.
This is uncovered by generic/285 subtest 07 and 08 on ppc64 host,
where pagesize is 64k. Because a recent change to generic/285
reduced the preallocated file size to smaller than 64k.
Signed-off-by: Eryu Guan <eguan@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 63db7c815b upstream.
We've had user reports of unmount hangs in xfs_wait_buftarg() that
analysis shows is due to btp->bt_io_count == -1. bt_io_count
represents the count of in-flight asynchronous buffers and thus
should always be >= 0. xfs_wait_buftarg() waits for this value to
stabilize to zero in order to ensure that all untracked (with
respect to the lru) buffers have completed I/O processing before
unmount proceeds to tear down in-core data structures.
The value of -1 implies an I/O accounting decrement race. Indeed,
the fact that xfs_buf_ioacct_dec() is called from xfs_buf_rele()
(where the buffer lock is no longer held) means that bp->b_flags can
be updated from an unsafe context. While a user-level reproducer is
currently not available, some intrusive hacks to run racing buffer
lookups/ioacct/releases from multiple threads was used to
successfully manufacture this problem.
Existing callers do not expect to acquire the buffer lock from
xfs_buf_rele(). Therefore, we can not safely update ->b_flags from
this context. It turns out that we already have separate buffer
state bits and associated serialization for dealing with buffer LRU
state in the form of ->b_state and ->b_lock. Therefore, replace the
_XBF_IN_FLIGHT flag with a ->b_state variant, update the I/O
accounting wrappers appropriately and make sure they are used with
the correct locking. This ensures that buffer in-flight state can be
modified at buffer release time without racing with modifications
from a buffer lock holder.
Fixes: 9c7504aa72 ("xfs: track and serialize in-flight async buffers against unmount")
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Tested-by: Libor Pechacek <lpechacek@suse.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 5375023ae1 upstream.
XFS SEEK_HOLE implementation could miss a hole in an unwritten extent as
can be seen by the following command:
xfs_io -c "falloc 0 256k" -c "pwrite 0 56k" -c "pwrite 128k 8k"
-c "seek -h 0" file
wrote 57344/57344 bytes at offset 0
56 KiB, 14 ops; 0.0000 sec (49.312 MiB/sec and 12623.9856 ops/sec)
wrote 8192/8192 bytes at offset 131072
8 KiB, 2 ops; 0.0000 sec (70.383 MiB/sec and 18018.0180 ops/sec)
Whence Result
HOLE 139264
Where we can see that hole at offset 56k was just ignored by SEEK_HOLE
implementation. The bug is in xfs_find_get_desired_pgoff() which does
not properly detect the case when pages are not contiguous.
Fix the problem by properly detecting when found page has larger offset
than expected.
Fixes: d126d43f63
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
