A debug mode write failure mechanism was introduced to XFS in commit
801cc4e17a ("xfs: debug mode forced buffered write failure") to
facilitate targeted testing of delalloc indirect reservation management
from userspace. This code was subsequently rendered ineffective by the
move to iomap based buffered writes in commit 68a9f5e700 ("xfs:
implement iomap based buffered write path"). This likely went unnoticed
because the associated userspace code had not made it into xfstests.
Resurrect this mechanism to facilitate effective indlen reservation
testing from xfstests. The move to iomap based buffered writes relocated
the hook this mechanism needs to return write failure from XFS to
generic code. The failure trigger must remain in XFS. Given that
limitation, convert this from a write failure mechanism to one that
simply drops writes without returning failure to userspace. Rename all
"fail_writes" references to "drop_writes" to illustrate the point. This
is more hacky than preferred, but still triggers the XFS error handling
behavior required to drive the indlen tests. This is only available in
DEBUG mode and for testing purposes only.
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>
The buffered write failure handling code in
xfs_file_iomap_end_delalloc() has a couple minor problems. First, if
written == 0, start_fsb is not rounded down and it fails to kill off a
delalloc block if the start offset is block unaligned. This results in a
lingering delalloc block and broken delalloc block accounting detected
at unmount time. Fix this by rounding down start_fsb in the unlikely
event that written == 0.
Second, it is possible for a failed overwrite of a delalloc extent to
leave dirty pagecache around over a hole in the file. This is because is
possible to hit ->iomap_end() on write failure before the iomap code has
attempted to allocate pagecache, and thus has no need to clean it up. If
the targeted delalloc extent was successfully written by a previous
write, however, then it does still have dirty pages when ->iomap_end()
punches out the underlying blocks. This ultimately results in writeback
over a hole. To fix this problem, unconditionally punch out the
pagecache from XFS before the associated delalloc range.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Instead of preallocating all the required COW blocks in the high-level
write code do it inside the iomap code, like we do for all other I/O.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Factor a helper to calculate the extent-size aligned block out of the
iomap code, so that it can be reused by the upcoming reflink dio code.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
We currently fall back from direct to buffered writes if we detect a
remaining shared extent in the iomap_begin callback. But by the time
iomap_begin is called for the potentially unaligned end block we might
have already written most of the data to disk, which we'd now write
again using buffered I/O. To avoid this reject all writes to reflinked
files before starting I/O so that we are guaranteed to only write the
data once.
The alternative would be to unshare the unaligned start and/or end block
before doing the I/O. I think that's doable, and will actually be
required to support reflinks on DAX file system. But it will take a
little more time and I'd rather get rid of the double write ASAP.
Signed-off-by: Christoph Hellwig <hch@lst.de>
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>
Christoph Hellwig pointed out that there's a potentially nasty race when
performing simultaneous nearby directio cow writes:
"Thread 1 writes a range from B to c
" B --------- C
p
"a little later thread 2 writes from A to B
" A --------- B
p
[editor's note: the 'p' denote cowextsize boundaries, which I added to
make this more clear]
"but the code preallocates beyond B into the range where thread
"1 has just written, but ->end_io hasn't been called yet.
"But once ->end_io is called thread 2 has already allocated
"up to the extent size hint into the write range of thread 1,
"so the end_io handler will splice the unintialized blocks from
"that preallocation back into the file right after B."
We can avoid this race by ensuring that thread 1 cannot accidentally
remap the blocks that thread 2 allocated (as part of speculative
preallocation) as part of t2's write preparation in t1's end_io handler.
The way we make this happen is by taking advantage of the unwritten
extent flag as an intermediate step.
Recall that when we begin the process of writing data to shared blocks,
we create a delayed allocation extent in the CoW fork:
D: --RRRRRRSSSRRRRRRRR---
C: ------DDDDDDD---------
When a thread prepares to CoW some dirty data out to disk, it will now
convert the delalloc reservation into an /unwritten/ allocated extent in
the cow fork. The da conversion code tries to opportunistically
allocate as much of a (speculatively prealloc'd) extent as possible, so
we may end up allocating a larger extent than we're actually writing
out:
D: --RRRRRRSSSRRRRRRRR---
U: ------UUUUUUU---------
Next, we convert only the part of the extent that we're actively
planning to write to normal (i.e. not unwritten) status:
D: --RRRRRRSSSRRRRRRRR---
U: ------UURRUUU---------
If the write succeeds, the end_cow function will now scan the relevant
range of the CoW fork for real extents and remap only the real extents
into the data fork:
D: --RRRRRRRRSRRRRRRRR---
U: ------UU--UUU---------
This ensures that we never obliterate valid data fork extents with
unwritten blocks from the CoW fork.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Due to the way how xfs_iomap_write_allocate tries to convert the whole
found extents from delalloc to real space we can run into a race
condition with multiple threads doing writes to this same extent.
For the non-COW case that is harmless as the only thing that can happen
is that we call xfs_bmapi_write on an extent that has already been
converted to a real allocation. For COW writes where we move the extent
from the COW to the data fork after I/O completion the race is, however,
not quite as harmless. In the worst case we are now calling
xfs_bmapi_write on a region that contains hole in the COW work, which
will trip up an assert in debug builds or lead to file system corruption
in non-debug builds. This seems to be reproducible with workloads of
small O_DSYNC write, although so far I've not managed to come up with
a with an isolated reproducer.
The fix for the issue is relatively simple: tell xfs_bmapi_write
that we are only asked to convert delayed allocations and skip holes
in that case.
Signed-off-by: Christoph Hellwig <hch@lst.de>
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>
Straight switch over to using iomap for direct I/O - we already have the
non-COW dio path in write_begin for DAX and files with extent size hints,
so nothing to add there. The COW path is ported over from the old
get_blocks version and a bit of a mess, but I have some work in progress
to make it look more like the buffered I/O COW path.
This gets rid of xfs_get_blocks_direct and the last caller of
xfs_get_blocks with the create flag set, so all that code can be removed.
Last but not least I've removed a comment in xfs_filemap_fault that
refers to xfs_get_blocks entirely instead of updating it - while the
reference is correct, the whole DAX fault path looks different than
the non-DAX one, so it seems rather pointless.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Jens Axboe <axboe@fb.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
xfs_file_iomap_begin_delay() implements post-eof speculative
preallocation by extending the block count of the requested delayed
allocation. Now that xfs_bmapi_reserve_delalloc() has been updated to
handle prealloc blocks separately and tag the inode, update
xfs_file_iomap_begin_delay() to use the new parameter and rely on the
former to tag the inode.
Note that this patch does not change behavior.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Speculative preallocation is currently processed entirely by the callers
of xfs_bmapi_reserve_delalloc(). The caller determines how much
preallocation to include, adjusts the extent length and passes down the
resulting request.
While this works fine for post-eof speculative preallocation, it is not
as reliable for COW fork preallocation. COW fork preallocation is
implemented via the cowextszhint, which aligns the start offset as well
as the length of the extent. Further, it is difficult for the caller to
accurately identify when preallocation occurs because the returned
extent could have been merged with neighboring extents in the fork.
To simplify this situation and facilitate further COW fork preallocation
enhancements, update xfs_bmapi_reserve_delalloc() to take a separate
preallocation parameter to incorporate into the allocation request. The
preallocation blocks value is tacked onto the end of the request and
adjusted to accommodate neighboring extents and extent size limits.
Since xfs_bmapi_reserve_delalloc() now knows precisely how much
preallocation was included in the allocation, it can also tag the inodes
appropriately to support preallocation reclaim.
Note that xfs_bmapi_reserve_delalloc() callers are not yet updated to
use the preallocation mechanism. This patch should not change behavior
outside of correctly tagging reflink inodes when start offset
preallocation occurs (which the caller does not handle correctly).
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
And only lookup the previous extent inside xfs_iomap_prealloc_size
if we actually need it.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
We can easily lookup the previous extent for the cases where we need it,
which saves the callers from looking it up for us later in the series.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Instead of reserving space as the first thing in write_begin move it past
reading the extent in the data fork. That way we only have to read from
the data fork once and can reuse that information for trimming the extent
to the shared/unshared boundary. Additionally this allows to easily
limit the actual write size to said boundary, and avoid a roundtrip on the
ilock.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
There is no need to trim an extent into a shared or non-shared one, or
report any flags for plain old reads.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Create a per-inode extent size allocator hint for copy-on-write. This
hint is separate from the existing extent size hint so that CoW can
take advantage of the fragmentation-reducing properties of extent size
hints without disabling delalloc for regular writes.
The extent size hint that's fed to the allocator during a copy on
write operation is the greater of the cowextsize and regular extsize
hint.
During reflink, if we're sharing the entire source file to the entire
destination file and the destination file doesn't already have a
cowextsize hint, propagate the source file's cowextsize hint to the
destination file.
Furthermore, zero the bulkstat buffer prior to setting the fields
so that we don't copy kernel memory contents into userspace.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Report shared extents through the iomap interface so that FIEMAP flags
shared blocks accurately. Have xfs_vm_bmap return zero for reflinked
files because the bmap-based swap code requires static block mappings,
which is incompatible with copy on write.
NOTE: Existing userspace bmap users such as lilo will have the same
problem with reflink files.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Modify xfs_bmap_add_extent_delay_real() so that we can convert delayed
allocation extents in the CoW fork to real allocations, and wire this
up all the way back to xfs_iomap_write_allocate(). In a subsequent
patch, we'll modify the writepage handler to call this.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Wire up iomap_begin to detect shared extents and create delayed allocation
extents in the CoW fork:
1) Check if we already have an extent in the COW fork for the area.
If so nothing to do, we can move along.
2) Look up block number for the current extent, and if there is none
it's not shared move along.
3) Unshare the current extent as far as we are going to write into it.
For this we avoid an additional COW fork lookup and use the
information we set aside in step 1) above.
4) Goto 1) unless we've covered the whole range.
Last but not least, this updates the xfs_reflink_reserve_cow_range calling
convention to pass a byte offset and length, as that is what both callers
expect anyway. This patch has been refactored considerably as part of the
iomap transition.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Allow the creation of delayed allocation extents in the CoW fork. In
a subsequent patch we'll wire up iomap_begin to actually do this via
reflink helper functions.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Another users of buffer_heads bytes the dust.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
We always just read the extent first, and will later lock exlusively
after first dropping the lock in case we actually allocate blocks.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Currently xfs_iomap_write_delay does up to lookups in the inode
extent tree, which is rather costly especially with the new iomap
based write path and small write sizes.
But it turns out that the low-level xfs_bmap_search_extents gives us
all the information we need in the regular delalloc buffered write
path:
- it will return us an extent covering the block we are looking up
if it exists. In that case we can simply return that extent to
the caller and are done
- it will tell us if we are beyoned the last current allocated
block with an eof return parameter. In that case we can create a
delalloc reservation and use the also returned information about
the last extent in the file as the hint to size our delalloc
reservation.
- it can tell us that we are writing into a hole, but that there is
an extent beyoned this hole. In this case we can create a
delalloc reservation that covers the requested size (possible
capped to the next existing allocation).
All that can be done in one single routine instead of bouncing up
and down a few layers. This reduced the CPU overhead of the block
mapping routines and also simplified the code a lot.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
And drop the pointless mp argument to xfs_iomap_eof_align_last_fsb,
while we're at it.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
We'll need it earlier in the file soon, so the unchanged function to
the top of xfs_iomap.c
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Use a special read-only iomap_ops implementation to support fiemap on
the attr fork.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
We'll never get nimap == 0 for a successful return from xfs_bmapi_read,
so don't try to handle it.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The space reservations was without an explaination in commit
"Add error reporting calls in error paths that return EFSCORRUPTED"
back in 2003. There is no reason to reserve disk blocks in the
transaction when allocating blocks for delalloc space as we already
reserved the space when creating the delalloc extent.
With this fix we stop running out of the reserved pool in
generic/229, which has happened for long time with small blocksize
file systems, and has increased in severity with the new buffered
write path.
[ dchinner: we still need to pass the block reservation into
xfs_bmapi_write() to ensure we don't deadlock during AG selection.
See commit dbd5c8c ("xfs: pass total block res. as total
xfs_bmapi_write() parameter") for more details on why this is
necessary. ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Mechanical change of flist/free_list to dfops, since they're now
deferred ops, not just a freeing list.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Drop the compatibility shims that we were using to integrate the new
deferred operation mechanism into the existing code. No new code.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Restructure everything that used xfs_bmap_free to use xfs_defer_ops
instead. For now we'll just remove the old symbols and play some
cpp magic to make it work; in the next patch we'll actually rename
everything.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Convert XFS to use the new iomap based multipage write path. This involves
implementing the ->iomap_begin and ->iomap_end methods, and switching the
buffered file write, page_mkwrite and xfs_iozero paths to the new iomap
helpers.
With this change __xfs_get_blocks will never be used for buffered writes,
and the code handling them can be removed.
Based on earlier code from Dave Chinner.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
And ensure it works for RT subvolume files an set the block device,
both of which will be needed to be able to use the function in the
buffered write path.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Merge xfs_trans_reserve and xfs_trans_alloc into a single function call
that returns a transaction with all the required log and block reservations,
and which allows passing transaction flags directly to avoid the cumbersome
_xfs_trans_alloc interface.
While we're at it we also get rid of the transaction type argument that has
been superflous since we stopped supporting the non-CIL logging mode. The
guts of it will be removed in another patch.
[dchinner: fixed transaction leak in error path in xfs_setattr_nonsize]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Calls to xfs_bmap_finish() and xfs_trans_ijoin(), and the
associated comments were replicated several times across
the attribute code, all dealing with what to do if the
transaction was or wasn't committed.
And in that replicated code, an ASSERT() test of an
uninitialized variable occurs in several locations:
error = xfs_attr_thing(&args);
if (!error) {
error = xfs_bmap_finish(&args.trans, args.flist,
&committed);
}
if (error) {
ASSERT(committed);
If the first xfs_attr_thing() failed, we'd skip the xfs_bmap_finish,
never set "committed", and then test it in the ASSERT.
Fix this up by moving the committed state internal to xfs_bmap_finish,
and add a new inode argument. If an inode is passed in, it is passed
through to __xfs_trans_roll() and joined to the transaction there if
the transaction was committed.
xfs_qm_dqalloc() was a little unique in that it called bjoin rather
than ijoin, but as Dave points out we can detect the committed state
but checking whether (*tpp != tp).
Addresses-Coverity-Id: 102360
Addresses-Coverity-Id: 102361
Addresses-Coverity-Id: 102363
Addresses-Coverity-Id: 102364
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Commit 1ca1915 ("xfs: Don't use unwritten extents for DAX") enabled
the DAX allocation call to dip into the reserve pool in case it was
converting unwritten extents rather than allocating blocks. This was
a direct copy of the unwritten extent conversion code, but had an
unintended side effect of allowing normal data block allocation to
use the reserve pool. Hence normal block allocation could deplete
the reserve pool and prevent unwritten extent conversion at ENOSPC,
hence violating fallocate guarantees on preallocated space.
Fix it by checking whether the incoming map from __xfs_get_blocks()
spans an unwritten extent and only use the reserve pool if the
allocation covers an unwritten extent.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Tested-by: Ross Zwisler <ross.zwisler@linux.intel.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>
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>
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>
The iomap codepath (via get_blocks()) acquires and release the inode
lock in the case of a direct write that requires block allocation. This
is because xfs_iomap_write_direct() allocates a transaction, which means
the ilock must be dropped and reacquired after the transaction is
allocated and reserved.
xfs_iomap_write_direct() invokes xfs_iomap_eof_align_last_fsb() before
the transaction is created and thus before the ilock is reacquired. This
can lead to calls to xfs_iread_extents() and reads of the in-core extent
list without any synchronization (via xfs_bmap_eof() and
xfs_bmap_last_extent()). xfs_iread_extents() assert fails if the ilock
is not held, but this is not currently seen in practice as the current
callers had already invoked xfs_bmapi_read().
What has been seen in practice are reports of crashes down in the
xfs_bmap_eof() codepath on direct writes due to seemingly bogus pointer
references from xfs_iext_get_ext(). While an explicit reproducer is not
currently available to confirm the cause of the problem, crash analysis
and code inspection from David Jeffrey had identified the insufficient
locking.
xfs_iomap_eof_align_last_fsb() is called from other contexts with the
inode lock already held, so we cannot acquire it therein.
__xfs_get_blocks() acquires and drops the ilock with variable flags to
cover the event that the extent list must be read in. The common case is
that __xfs_get_blocks() acquires the shared ilock. To provide locking
around the last extent alignment call without adding more lock cycles to
the dio path, update xfs_iomap_write_direct() to expect the shared ilock
held on entry and do the extent alignment under its protection. Demote
the lock, if necessary, from __xfs_get_blocks() and push the
xfs_qm_dqattach() call outside of the shared lock critical section.
Also, add an assert to document that the extent list is always expected
to be present in this path. Otherwise, we risk a call to
xfs_iread_extents() while under the shared ilock. This is safe as all
current callers have executed an xfs_bmapi_read() call under the current
iolock context.
Reported-by: David Jeffery <djeffery@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 flags argument to xfs_trans_commit is not useful for most callers, as
a commit of a transaction without a permanent log reservation must pass
0 here, and all callers for a transaction with a permanent log reservation
except for xfs_trans_roll must pass XFS_TRANS_RELEASE_LOG_RES. So remove
the flags argument from the public xfs_trans_commit interfaces, and
introduce low-level __xfs_trans_commit variant just for xfs_trans_roll
that regrants a log reservation instead of releasing it.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
xfs_trans_cancel takes two flags arguments: XFS_TRANS_RELEASE_LOG_RES and
XFS_TRANS_ABORT. Both of them are a direct product of the transaction
state, and can be deducted:
- any dirty transaction needs XFS_TRANS_ABORT to be properly canceled,
and XFS_TRANS_ABORT is a noop for a transaction that is not dirty.
- any transaction with a permanent log reservation needs
XFS_TRANS_RELEASE_LOG_RES to be properly canceled, and passing
XFS_TRANS_RELEASE_LOG_RES for a transaction without a permanent
log reservation is invalid.
So just remove the flags argument and do the right thing.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Now that the in-core superblock infrastructure has been replaced with
generic per-cpu counters, we don't need it anymore. Nuke it from
orbit so we are sure that it won't haunt us again...
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
XFS has hand-rolled per-cpu counters for the superblock since before
there was any generic implementation. The free block counter is
special in that it is used for ENOSPC detection outside transaction
contexts for for delayed allocation. This means that the counter
needs to be accurate at zero. The current per-cpu counter code jumps
through lots of hoops to ensure we never run past zero, but we don't
need to make all those jumps with the generic counter
implementation.
The generic counter implementation allows us to pass a "batch"
threshold at which the addition/subtraction to the counter value
will be folded back into global value under lock. We can use this
feature to reduce the batch size as we approach 0 in a very similar
manner to the existing counters and their rebalance algorithm. If we
use a batch size of 1 as we approach 0, then every addition and
subtraction will be done against the global value and hence allow
accurate detection of zero threshold crossing.
Hence we can replace the handrolled, accurate-at-zero counters with
generic percpu counters.
Note: this removes just enough of the icsb infrastructure to compile
without warnings. The rest will go in subsequent commits.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
