As part of the head/tail discovery process, log recovery locates the
head block and then reverse seeks to find the start of the last active
record in the log. This is non-trivial as the record itself could have
wrapped around the end of the physical log. Log recovery torn write
detection potentially needs to walk further behind the last record in
the log, as multiple log I/Os can be in-flight at one time during a
crash event.
Therefore, refactor the reverse log record header search mechanism into
a new helper that supports the ability to seek past an arbitrary number
of log records (or until the tail is hit). Update the head/tail search
mechanism to call the new helper, but otherwise there is no change in
log recovery behavior.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Log recovery torn write detection uses CRC verification over a range of
the active log to identify torn writes. Since the generic log recovery
pass code implements a superset of the functionality required for CRC
verification, it can be easily modified to support a CRC verification
only pass.
Create a new CRC pass type and update the log record processing helper
to skip everything beyond CRC verification when in this mode. This pass
will be invoked in subsequent patches to implement torn write detection.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Each log recovery pass walks from the tail block to the head block and
processes records appropriately based on the associated log pass type.
There are various failure conditions that can occur through this
sequence, such as I/O errors, CRC errors, etc. Log torn write detection
will perform CRC verification near the head of the log to detect torn
writes and trim torn records from the log appropriately.
As it is, xlog_do_recovery_pass() only returns an error code in the
event of CRC failure, which isn't enough information to trim the head of
the log. Update xlog_do_recovery_pass() to optionally return the start
block of the associated record when an error occurs. This patch contains
no functional changes.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Log record CRC verification currently occurs during active log recovery,
immediately before a log record is unpacked. Therefore, the CRC
calculation code is buried within the data unpack function. CRC
verification pass support only needs to go so far as check the CRC, but
this is not easily allowed as the code is currently organized.
Since we now have a new log record processing helper, pull the record
CRC verification code out from the unpack helper and open-code it at the
top of the new process helper. This facilitates the ability to modify
how records are processed based on the type of the current pass. This
patch contains no functional changes.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
xlog_do_recovery_pass() duplicates a couple function calls related to
processing log records because the function must handle wrapping around
the end of the log if the head is behind the tail. This is implemented
as separate loops. CRC verification pass support will modify how records
are processed in both of these loops.
Rather than continue to duplicate code, factor the calls that process a
log record into a new helper and call that helper from both loops. This
patch contains no functional changes.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
XFS log records have separate fields for the record size and the iclog
size used to write the record. mkfs.xfs zeroes the log and writes an
unmount record to generate a clean log for the subsequent mount. The
userspace record logging code has a bug where the iclog size (h_size)
field of the log record is hardcoded to 32k, even if a log stripe unit
is specified. The log record length is correctly extended to the stripe
unit. Since the kernel log recovery code uses the h_size field to
determine the log buffer size, this means that the kernel can attempt to
read/process records larger than the buffer size and overrun the buffer.
This has historically not been a problem because the kernel doesn't
actually run through log recovery in the clean unmount case. Instead,
the kernel detects that a single unmount record exists between the head
and tail and pushes the tail forward such that the log is viewed as
clean (head == tail). Once CRC verification is enabled, however, all
records at the head of the log are verified for CRC errors and thus we
are susceptible to overrun problems if the iclog field is not correct.
While the core problem must be fixed in userspace, this is historical
behavior that must be detected in the kernel to avoid severe side
effects such as memory corruption and crashes. Update the log buffer
size calculation code to detect this condition, warn the user and resize
the log buffer based on the log stripe unit. Return a corruption error
in cases where this does not look like a clean filesystem (i.e., the log
record header indicates more than one operation).
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The xattr_handler operations are currently all passed a file system
specific flags value which the operations can use to disambiguate between
different handlers; some file systems use that to distinguish the xattr
namespace, for example. In some oprations, it would be useful to also have
access to the handler prefix. To allow that, pass a pointer to the handler
to operations instead of the flags value alone.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
This update contains:
o per-mount operational statistics in sysfs
o fixes for concurrent aio append write submission
o various logging fixes
o detection of zeroed logs and invalid log sequence numbers on v5 filesystems
o memory allocation failure message improvements
o a bunch of xattr/ACL fixes
o fdatasync optimisation
o miscellaneous other fixes and cleanups
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)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=zu+5
-----END PGP SIGNATURE-----
Merge tag 'xfs-for-linus-4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs
Pull xfs updates from Dave Chinner:
"There is nothing really major here - the only significant addition is
the per-mount operation statistics infrastructure. Otherwises there's
various ACL, xattr, DAX, AIO and logging fixes, and a smattering of
small cleanups and fixes elsewhere.
Summary:
- per-mount operational statistics in sysfs
- fixes for concurrent aio append write submission
- various logging fixes
- detection of zeroed logs and invalid log sequence numbers on v5 filesystems
- memory allocation failure message improvements
- a bunch of xattr/ACL fixes
- fdatasync optimisation
- miscellaneous other fixes and cleanups"
* tag 'xfs-for-linus-4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs: (39 commits)
xfs: give all workqueues rescuer threads
xfs: fix log recovery op header validation assert
xfs: Fix error path in xfs_get_acl
xfs: optimise away log forces on timestamp updates for fdatasync
xfs: don't leak uuid table on rmmod
xfs: invalidate cached acl if set via ioctl
xfs: Plug memory leak in xfs_attrmulti_attr_set
xfs: Validate the length of on-disk ACLs
xfs: invalidate cached acl if set directly via xattr
xfs: xfs_filemap_pmd_fault treats read faults as write faults
xfs: add ->pfn_mkwrite support for DAX
xfs: DAX does not use IO completion callbacks
xfs: Don't use unwritten extents for DAX
xfs: introduce BMAPI_ZERO for allocating zeroed extents
xfs: fix inode size update overflow in xfs_map_direct()
xfs: clear PF_NOFREEZE for xfsaild kthread
xfs: fix an error code in xfs_fs_fill_super()
xfs: stats are no longer dependent on CONFIG_PROC_FS
xfs: simplify /proc teardown & error handling
xfs: per-filesystem stats counter implementation
...
The function currently called "__block_page_mkwrite()" used to be called
"block_page_mkwrite()" until a wrapper for this function was added by:
commit 24da4fab5a ("vfs: Create __block_page_mkwrite() helper passing
error values back")
This wrapper, the current "block_page_mkwrite()", is currently unused.
__block_page_mkwrite() is used directly by ext4, nilfs2 and xfs.
Remove the unused wrapper, rename __block_page_mkwrite() back to
block_page_mkwrite() and update the comment above block_page_mkwrite().
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.com>
Cc: Jan Kara <jack@suse.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
We're consistently hitting deadlocks here with XFS on recent kernels.
After some digging through the crash files, it looks like everyone in
the system is waiting for XFS to reclaim memory.
Something like this:
PID: 2733434 TASK: ffff8808cd242800 CPU: 19 COMMAND: "java"
#0 [ffff880019c53588] __schedule at ffffffff818c4df2
#1 [ffff880019c535d8] schedule at ffffffff818c5517
#2 [ffff880019c535f8] _xfs_log_force_lsn at ffffffff81316348
#3 [ffff880019c53688] xfs_log_force_lsn at ffffffff813164fb
#4 [ffff880019c536b8] xfs_iunpin_wait at ffffffff8130835e
#5 [ffff880019c53728] xfs_reclaim_inode at ffffffff812fd453
#6 [ffff880019c53778] xfs_reclaim_inodes_ag at ffffffff812fd8c7
#7 [ffff880019c53928] xfs_reclaim_inodes_nr at ffffffff812fe433
#8 [ffff880019c53958] xfs_fs_free_cached_objects at ffffffff8130d3b9
#9 [ffff880019c53968] super_cache_scan at ffffffff811a6f73
#10 [ffff880019c539c8] shrink_slab at ffffffff811460e6
#11 [ffff880019c53aa8] shrink_zone at ffffffff8114a53f
#12 [ffff880019c53b48] do_try_to_free_pages at ffffffff8114a8ba
#13 [ffff880019c53be8] try_to_free_pages at ffffffff8114ad5a
#14 [ffff880019c53c78] __alloc_pages_nodemask at ffffffff8113e1b8
#15 [ffff880019c53d88] alloc_kmem_pages_node at ffffffff8113e671
#16 [ffff880019c53dd8] copy_process at ffffffff8104f781
#17 [ffff880019c53ec8] do_fork at ffffffff8105129c
#18 [ffff880019c53f38] sys_clone at ffffffff810515b6
#19 [ffff880019c53f48] stub_clone at ffffffff818c8e4d
xfs_log_force_lsn is waiting for logs to get cleaned, which is waiting
for IO, which is waiting for workers to complete the IO which is waiting
for worker threads that don't exist yet:
PID: 2752451 TASK: ffff880bd6bdda00 CPU: 37 COMMAND: "kworker/37:1"
#0 [ffff8808d20abbb0] __schedule at ffffffff818c4df2
#1 [ffff8808d20abc00] schedule at ffffffff818c5517
#2 [ffff8808d20abc20] schedule_timeout at ffffffff818c7c6c
#3 [ffff8808d20abcc0] wait_for_completion_killable at ffffffff818c6495
#4 [ffff8808d20abd30] kthread_create_on_node at ffffffff8106ec82
#5 [ffff8808d20abdf0] create_worker at ffffffff8106752f
#6 [ffff8808d20abe40] worker_thread at ffffffff810699be
#7 [ffff8808d20abec0] kthread at ffffffff8106ef59
#8 [ffff8808d20abf50] ret_from_fork at ffffffff818c8ac8
I think we should be using WQ_MEM_RECLAIM to make sure this thread
pool makes progress when we're not able to allocate new workers.
[dchinner: make all workqueues WQ_MEM_RECLAIM]
Signed-off-by: Chris Mason <clm@fb.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Commit 89cebc84 ("xfs: validate transaction header length on log
recovery") added additional validation of the on-disk op header length
to protect from buffer overflow during log recovery. It accounts for the
fact that the transaction header can be split across multiple op
headers. It added an assert for when this occurs that verifies the
length of the second part of a split transaction header is less than a
full transaction header. In other words, it expects that the first op
header of a split transaction header includes at least some portion of
the transaction header.
This expectation is not always valid as a zero-length op header can
exist for the first op header of a split transaction header (see
xlog_recover_add_to_trans() for details). This means that the second op
header can have a valid, full length transaction header and thus the
full header is copied in xlog_recover_add_to_cont_trans(). Fix the
assert in xlog_recover_add_to_cont_trans() to handle this case correctly
and require that the op header length is less than or equal to a full
transaction header.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Error codes from xfs_attr_get other than -ENOATTR were not properly
reported. Fix that.
In addition, the declaration of struct xfs_inode in xfs_acl.h isn't needed.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
__GFP_WAIT has been used to identify atomic context in callers that hold
spinlocks or are in interrupts. They are expected to be high priority and
have access one of two watermarks lower than "min" which can be referred
to as the "atomic reserve". __GFP_HIGH users get access to the first
lower watermark and can be called the "high priority reserve".
Over time, callers had a requirement to not block when fallback options
were available. Some have abused __GFP_WAIT leading to a situation where
an optimisitic allocation with a fallback option can access atomic
reserves.
This patch uses __GFP_ATOMIC to identify callers that are truely atomic,
cannot sleep and have no alternative. High priority users continue to use
__GFP_HIGH. __GFP_DIRECT_RECLAIM identifies callers that can sleep and
are willing to enter direct reclaim. __GFP_KSWAPD_RECLAIM to identify
callers that want to wake kswapd for background reclaim. __GFP_WAIT is
redefined as a caller that is willing to enter direct reclaim and wake
kswapd for background reclaim.
This patch then converts a number of sites
o __GFP_ATOMIC is used by callers that are high priority and have memory
pools for those requests. GFP_ATOMIC uses this flag.
o Callers that have a limited mempool to guarantee forward progress clear
__GFP_DIRECT_RECLAIM but keep __GFP_KSWAPD_RECLAIM. bio allocations fall
into this category where kswapd will still be woken but atomic reserves
are not used as there is a one-entry mempool to guarantee progress.
o Callers that are checking if they are non-blocking should use the
helper gfpflags_allow_blocking() where possible. This is because
checking for __GFP_WAIT as was done historically now can trigger false
positives. Some exceptions like dm-crypt.c exist where the code intent
is clearer if __GFP_DIRECT_RECLAIM is used instead of the helper due to
flag manipulations.
o Callers that built their own GFP flags instead of starting with GFP_KERNEL
and friends now also need to specify __GFP_KSWAPD_RECLAIM.
The first key hazard to watch out for is callers that removed __GFP_WAIT
and was depending on access to atomic reserves for inconspicuous reasons.
In some cases it may be appropriate for them to use __GFP_HIGH.
The second key hazard is callers that assembled their own combination of
GFP flags instead of starting with something like GFP_KERNEL. They may
now wish to specify __GFP_KSWAPD_RECLAIM. It's almost certainly harmless
if it's missed in most cases as other activity will wake kswapd.
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
xfs: timestamp updates cause excessive fdatasync log traffic
Sage Weil reported that a ceph test workload was writing to the
log on every fdatasync during an overwrite workload. Event tracing
showed that the only metadata modification being made was the
timestamp updates during the write(2) syscall, but fdatasync(2)
is supposed to ignore them. The key observation was that the
transactions in the log all looked like this:
INODE: #regs: 4 ino: 0x8b flags: 0x45 dsize: 32
And contained a flags field of 0x45 or 0x85, and had data and
attribute forks following the inode core. This means that the
timestamp updates were triggering dirty relogging of previously
logged parts of the inode that hadn't yet been flushed back to
disk.
There are two parts to this problem. The first is that XFS relogs
dirty regions in subsequent transactions, so it carries around the
fields that have been dirtied since the last time the inode was
written back to disk, not since the last time the inode was forced
into the log.
The second part is that on v5 filesystems, the inode change count
update during inode dirtying also sets the XFS_ILOG_CORE flag, so
on v5 filesystems this makes a timestamp update dirty the entire
inode.
As a result when fdatasync is run, it looks at the dirty fields in
the inode, and sees more than just the timestamp flag, even though
the only metadata change since the last fdatasync was just the
timestamps. Hence we force the log on every subsequent fdatasync
even though it is not needed.
To fix this, add a new field to the inode log item that tracks
changes since the last time fsync/fdatasync forced the log to flush
the changes to the journal. This flag is updated when we dirty the
inode, but we do it before updating the change count so it does not
carry the "core dirty" flag from timestamp updates. The fields are
zeroed when the inode is marked clean (due to writeback/freeing) or
when an fsync/datasync forces the log. Hence if we only dirty the
timestamps on the inode between fsync/fdatasync calls, the fdatasync
will not trigger another log force.
Over 100 runs of the test program:
Ext4 baseline:
runtime: 1.63s +/- 0.24s
avg lat: 1.59ms +/- 0.24ms
iops: ~2000
XFS, vanilla kernel:
runtime: 2.45s +/- 0.18s
avg lat: 2.39ms +/- 0.18ms
log forces: ~400/s
iops: ~1000
XFS, patched kernel:
runtime: 1.49s +/- 0.26s
avg lat: 1.46ms +/- 0.25ms
log forces: ~30/s
iops: ~1500
Reported-by: Sage Weil <sage@redhat.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Don't leak the UUID table when the module is unloaded.
(Found with kmemleak.)
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Setting or removing the "SGI_ACL_[FILE|DEFAULT]" attributes via the
XFS_IOC_ATTRMULTI_BY_HANDLE ioctl completely bypasses the POSIX ACL
infrastructure, like setting the "trusted.SGI_ACL_[FILE|DEFAULT]" xattrs
did until commit 6caa1056. Similar to that commit, invalidate cached
acls when setting/removing them via the ioctl as well.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
When setting attributes via XFS_IOC_ATTRMULTI_BY_HANDLE, the user-space
buffer is copied into a new kernel-space buffer via memdup_user; that
buffer then isn't freed.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
In xfs_acl_from_disk, instead of trusting that xfs_acl.acl_cnt is correct,
make sure that the length of the attributes is correct as well. Also, turn
the aclp parameter into a const pointer.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
ACLs are stored as extended attributes of the inode to which they apply.
XFS converts the standard "system.posix_acl_[access|default]" attribute
names used to control ACLs to "trusted.SGI_ACL_[FILE|DEFAULT]" as stored
on-disk. These xattrs are directly exposed in on-disk format via
getxattr/setxattr, without any ACL aware code in the path to perform
validation, etc. This is partly historical and supports backup/restore
applications such as xfsdump to back up and restore the binary blob that
represents ACLs as-is.
Andreas reports that the ACLs observed via the getfacl interface is not
consistent when ACLs are set directly via the setxattr path. This occurs
because the ACLs are cached in-core against the inode and the xattr path
has no knowledge that the operation relates to ACLs.
Update the xattr set codepath to trap writes of the special XFS ACL
attributes and invalidate the associated cached ACL when this occurs.
This ensures that the correct ACLs are used on a subsequent operation
through the actual ACL interface.
Note that this does not update or add support for setting the ACL xattrs
directly beyond the restore use case that requires a correctly formatted
binary blob and to restore a consistent i_mode at the same time. It is
still possible for a root user to set an invalid or inconsistent (with
i_mode) ACL blob on-disk and potentially cause corruption.
[ With fixes from Andreas Gruenbacher. ]
Reported-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The code initially committed didn't have the same checks for write
faults as the dax_pmd_fault code and hence treats all faults as
write faults. We can get read faults through this path because they
is no pmd_mkwrite path for write faults similar to the normal page
fault path. Hence we need to ensure that we only do c/mtime updates
on write faults, and freeze protection is unnecessary for read
faults.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
->pfn_mkwrite support is needed so that when a page with allocated
backing store takes a write fault we can check that the fault has
not raced with a truncate and is pointing to a region beyond the
current end of file.
This also allows us to update the timestamp on the inode, too, which
fixes a generic/080 failure.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
For DAX, we are now doing block zeroing during allocation. This
means we no longer need a special DAX fault IO completion callback
to do unwritten extent conversion. Because mmap never extends the
file size (it SEGVs the process) we don't need a callback to update
the file size, either. Hence we can remove the completion callbacks
from the __dax_fault and __dax_mkwrite calls.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
DAX has a page fault serialisation problem with block allocation.
Because it allows concurrent page faults and does not have a page
lock to serialise faults to the same page, it can get two concurrent
faults to the page that race.
When two read faults race, this isn't a huge problem as the data
underlying the page is not changing and so "detect and drop" works
just fine. The issues are to do with write faults.
When two write faults occur, we serialise block allocation in
get_blocks() so only one faul will allocate the extent. It will,
however, be marked as an unwritten extent, and that is where the
problem lies - the DAX fault code cannot differentiate between a
block that was just allocated and a block that was preallocated and
needs zeroing. The result is that both write faults end up zeroing
the block and attempting to convert it back to written.
The problem is that the first fault can zero and convert before the
second fault starts zeroing, resulting in the zeroing for the second
fault overwriting the data that the first fault wrote with zeros.
The second fault then attempts to convert the unwritten extent,
which is then a no-op because it's already written. Data loss occurs
as a result of this race.
Because there is no sane locking construct in the page fault code
that we can use for serialisation across the page faults, we need to
ensure block allocation and zeroing occurs atomically in the
filesystem. This means we can still take concurrent page faults and
the only time they will serialise is in the filesystem
mapping/allocation callback. The page fault code will always see
written, initialised extents, so we will be able to remove the
unwritten extent handling from the DAX code when all filesystems are
converted.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
To enable DAX to do atomic allocation of zeroed extents, we need to
drive the block zeroing deep into the allocator. Because
xfs_bmapi_write() can return merged extents on allocation that were
only partially allocated (i.e. requested range spans allocated and
hole regions, allocation into the hole was contiguous), we cannot
zero the extent returned from xfs_bmapi_write() as that can
overwrite existing data with zeros.
Hence we have to drive the extent zeroing into the allocation code,
prior to where we merge the extents into the BMBT and return the
resultant map. This means we need to propagate this need down to
the xfs_alloc_vextent() and issue the block zeroing at this point.
While this functionality is being introduced for DAX, there is no
reason why it is specific to DAX - we can per-zero blocks during the
allocation transaction on any type of device. It's just slow (and
usually slower than unwritten allocation and conversion) on
traditional block devices so doesn't tend to get used. We can,
however, hook hardware zeroing optimisations via sb_issue_zeroout()
to this operation, so it may be useful in future and hence the
"allocate zeroed blocks" API needs to be implementation neutral.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Both direct IO and DAX pass an offset and count into get_blocks that
will overflow a s64 variable when an IO goes into the last supported
block in a file (i.e. at offset 2^63 - 1FSB bytes). This can be seen
from the tracing:
xfs_get_blocks_alloc: [...] offset 0x7ffffffffffff000 count 4096
xfs_gbmap_direct: [...] offset 0x7ffffffffffff000 count 4096
xfs_gbmap_direct_none:[...] offset 0x7ffffffffffff000 count 4096
0x7ffffffffffff000 + 4096 = 0x8000000000000000, and hence that
overflows the s64 offset and we fail to detect the need for a
filesize update and an ioend is not allocated.
This is *mostly* avoided for direct IO because such extending IOs
occur with full block allocation, and so the "IS_UNWRITTEN()" check
still evaluates as true and we get an ioend that way. However, doing
single sector extending IOs to this last block will expose the fact
that file size updates will not occur after the first allocating
direct IO as the overflow will then be exposed.
There is one further complexity: the DAX page fault path also
exposes the same issue in block allocation. However, page faults
cannot extend the file size, so in this case we want to allocate the
block but do not want to allocate an ioend to enable file size
update at IO completion. Hence we now need to distinguish between
the direct IO patch allocation and dax fault path allocation to
avoid leaking ioend structures.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Since xfsaild has been converted to kthread in 0030807c, it calls
try_to_freeze() during every AIL push iteration. It however doesn't set
itself as freezable, and therefore this try_to_freeze() will never do
anything.
Before (hopefully eventually) kthread freezing gets converted to fileystem
freezing, we'd rather mark xfsaild freezable (as it can generate I/O
during suspend).
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
If alloc_percpu() fails, we accidentally return PTR_ERR(NULL), which
means success, but we intended to return -ENOMEM.
Fixes: 225e463558 ('xfs: per-filesystem stats in sysfs')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Bill O'Donnell <billodo@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
So we need to fix the makefile to understand this, otherwise build
errors with CONFIG_PROC_FS=n occur.
Reported-and-tested-by: Jim Davis <jim.epost@gmail.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
remove_proc_subtree() was added in 3.9, and can be
used to simplify our procfile creation error handling
and cleanup, removing the nested gotos. It simply
removes fs/xfs and everything created under it.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
This patch modifies the stats counting macros and the callers
to those macros to properly increment, decrement, and add-to
the xfs stats counts. The counts for global and per-fs stats
are correctly advanced, and cleared by writing a "1" to the
corresponding clear file.
global counts: /sys/fs/xfs/stats/stats
per-fs counts: /sys/fs/xfs/sda*/stats/stats
global clear: /sys/fs/xfs/stats/stats_clear
per-fs clear: /sys/fs/xfs/sda*/stats/stats_clear
[dchinner: cleaned up macro variables, removed CONFIG_FS_PROC around
stats structures and macros. ]
Signed-off-by: Bill O'Donnell <billodo@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
This patch implements per-filesystem stats objects in sysfs. It
depends on the application of the previous patch series that
develops the infrastructure to support both xfs global stats and
xfs per-fs stats in sysfs.
Stats objects are instantiated when an xfs filesystem is mounted
and deleted on unmount. With this patch, the stats directory is
created and populated with the familiar stats and stats_clear files.
Example:
/sys/fs/xfs/sda9/stats/stats
/sys/fs/xfs/sda9/stats/stats_clear
With this patch, the individual counts within the new per-fs
stats file(s) remain at zero. Functions that use the the macros
to increment, decrement, and add-to the per-fs stats counts will
be covered in a separate new patch to follow this one. Note that
the counts within the global stats file (/sys/fs/xfs/stats/stats)
advance normally and can be cleared as it was prior to this patch.
[dchinner: move setup/teardown to xfs_fs_{fill|put}_super() so
it is down before/after any path that uses the per-mount stats. ]
Signed-off-by: Bill O'Donnell <billodo@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
In an effort to get more useful out of "possible memory
allocation deadlock" messages, print the size of the
requested allocation, and dump the stack if the xfs error
level is tuned high.
The stack dump is implemented in define_xfs_printk_level()
for error levels >= LOGLEVEL_ERR, partly because it
seems generically useful, and also because kmem.c has
no knowledge of xfs error level tunables or other such bits,
it's very kmem-specific.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The gcc undefined behavior sanitizer caught this; surely
any sane memcpy implementation will no-op if size == 0,
but behavior with a *src of NULL is technically undefined
(declared nonnull), so avoid it here.
We are actually in this situation frequently via
xlog_commit_record(), because:
struct xfs_log_iovec reg = {
.i_addr = NULL,
.i_len = 0,
.i_type = XLOG_REG_TYPE_COMMIT,
};
Reported-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The total field from struct xfs_alloc_arg is a bit of an unknown
commodity. It is documented as the total block requirement for the
transaction and is used in this manner from most call sites by virtue of
passing the total block reservation of the transaction associated with
an allocation. Several xfs_bmapi_write() callers pass hardcoded values
of 0 or 1 for the total block requirement, which is a historical oddity
without any clear reasoning.
The xfs_iomap_write_direct() caller, for example, passes 0 for the total
block requirement. This has been determined to cause problems in the
form of ABBA deadlocks of AGF buffers due to incorrect AG selection in
the block allocator. Specifically, the xfs_alloc_space_available()
function incorrectly selects an AG that doesn't actually have sufficient
space for the allocation. This occurs because the args.total field is 0
and thus the remaining free space check on the AG doesn't actually
consider the size of the allocation request. This locks the AGF buffer,
the allocation attempt proceeds and ultimately fails (in
xfs_alloc_fix_minleft()), and xfs_alloc_vexent() moves on to the next
AG. In turn, this can lead to incorrect AG locking order (if the
allocator wraps around, attempting to lock AG 0 after acquiring AG N)
and thus deadlock if racing with another operation. This problem has
been reproduced via generic/299 on smallish (1GB) ramdisk test devices.
To avoid this problem, replace the undocumented hardcoded total
parameters from the iomap and utility callers to pass the block
reservation used for the associated transaction. This is consistent with
other xfs_bmapi_write() callers throughout XFS. The assumption is that
the total field allows the selection of an AG that can handle the entire
operation rather than simply the allocation/range being requested (e.g.,
resulting btree splits, etc.). This addresses the aforementioned
generic/299 hang by ensuring AG selection only occurs when the
allocation can be satisfied by the AG.
Reported-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Currently, we depends on Linux XATTR value for on disk
definition. Which causes trouble on other platforms and
maybe also if this value was to change.
Fix it by creating a custom definition independent from
those in Linux (although with the same values), so it is OK
with the be16 fields used for holding these attributes.
This patch reflects a change in xfsprogs.
Signed-off-by: Jan Tulak <jtulak@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Remove a hard dependency of Linux XATTR_LIST_MAX value by using
a prefixed version. This patch reflects the same change in xfsprogs.
Signed-off-by: Jan Tulak <jtulak@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Just fix two typos in code comments.
Signed-off-by: Geliang Tang <geliangtang@163.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Add a tracepoint in xfs_zero_eof() to facilitate tracking and debugging
EOF zeroing events. This has proven useful in the context of other
direct I/O tracepoints to ensure EOF zeroing occurs within appropriate
file ranges.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
XFS supports and typically allows concurrent asynchronous direct I/O
submission to a single file. One exception to the rule is that file
extending dio writes that start beyond the current EOF (e.g.,
potentially create a hole at EOF) require exclusive I/O access to the
file. This is because such writes must zero any pre-existing blocks
beyond EOF that are exposed by virtue of now residing within EOF as a
result of the write about to be submitted.
Before EOF zeroing can occur, the current file i_size must be stabilized
to avoid data corruption. In this scenario, XFS upgrades the iolock to
exclude any further I/O submission, waits on in-flight I/O to complete
to ensure i_size is up to date (i_size is updated on dio write
completion) and restarts the various checks against the state of the
file. The problem is that this protection sequence is triggered only
when the iolock is currently held shared. While this is true for async
dio in most cases, the caller may upgrade the lock in advance based on
arbitrary circumstances with respect to EOF zeroing. For example, the
iolock is always acquired exclusively if the start offset is not block
aligned. This means that even though the iolock is already held
exclusive for such I/Os, pending I/O is not drained and thus EOF zeroing
can occur based on an unstable i_size.
This problem has been reproduced as guest data corruption in virtual
machines with file-backed qcow2 virtual disks hosted on an XFS
filesystem. The virtual disks must be configured with aio=native mode
and the must not be truncated out to the maximum file size (as some virt
managers will do).
Update xfs_file_aio_write_checks() to unconditionally drain in-flight
dio before EOF zeroing can occur. Rather than trigger the wait based on
iolock state, use a new flag and upgrade the iolock when necessary. Note
that this results in a full restart of the inode checks even when the
iolock was already held exclusive when technically it is only required
to recheck i_size. This should be a rare enough occurrence that it is
preferable to keep the code simple rather than create an alternate
restart jump target.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Since the onset of v5 superblocks, the LSN of the last modification has
been included in a variety of on-disk data structures. This LSN is used
to provide log recovery ordering guarantees (e.g., to ensure an older
log recovery item is not replayed over a newer target data structure).
While this works correctly from the point a filesystem is formatted and
mounted, userspace tools have some problematic behaviors that defeat
this mechanism. For example, xfs_repair historically zeroes out the log
unconditionally (regardless of whether corruption is detected). If this
occurs, the LSN of the filesystem is reset and the log is now in a
problematic state with respect to on-disk metadata structures that might
have a larger LSN. Until either the log catches up to the highest
previously used metadata LSN or each affected data structure is modified
and written out without incident (which resets the metadata LSN), log
recovery is susceptible to filesystem corruption.
This problem is ultimately addressed and repaired in the associated
userspace tools. The kernel is still responsible to detect the problem
and notify the user that something is wrong. Check the superblock LSN at
mount time and fail the mount if it is invalid. From that point on,
trigger verifier failure on any metadata I/O where an invalid LSN is
detected. This results in a filesystem shutdown and guarantees that we
do not log metadata changes with invalid LSNs on disk. Since this is a
known issue with a known recovery path, present a warning to instruct
the user how to recover.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
This patch adds comm name and pid to warning messages printed by
kmem_alloc(), kmem_zone_alloc() and xfs_buf_allocate_memory().
This will help telling which memory allocations (e.g. kernel worker
threads, OOM victim tasks, neither) are stalling because these functions
are passing __GFP_NOWARN which suppresses not only backtrace but comm name
and pid.
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
A local format symlink inode is converted to extent format when an
extended attribute is set on an inode as part of the attribute fork
creation. This means a block is allocated, the local symlink target name
is copied to the block and the block is logged. Currently,
xfs_bmap_local_to_extents() handles logging the remote block data based
on the size of the data fork prior to the conversion. This is not
correct on v5 superblock filesystems, which add an additional header to
remote symlink blocks that is nonexistent in local format inodes.
As a result, the full length of the remote symlink block content is not
logged. This can lead to corruption should a crash occur and log
recovery replay this transaction.
Since a callout is already used to initialize the new remote symlink
block, update the local-to-extents conversion mechanism to make the
callout also responsible for logging the block. It is already required
to set the log buffer type and format the block appropriately based on
the superblock version. This ensures the remote symlink is always logged
correctly. Note that xfs_bmap_local_to_extents() is only called for
symlinks so there are no other callouts that require modification.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
