[commit message is verbose for discussion purposes - will trim it
down later. Some questions about implementation details at the end.]
Zorro Lang recently ran a new test to stress single inode extent
counts now that they are no longer limited by memory allocation.
The test was simply:
# xfs_io -f -c "falloc 0 40t" /mnt/scratch/big-file
# ~/src/xfstests-dev/punch-alternating /mnt/scratch/big-file
This test uncovered a problem where the hole punching operation
appeared to finish with no error, but apparently only created 268M
extents instead of the 10 billion it was supposed to.
Further, trying to punch out extents that should have been present
resulted in success, but no change in the extent count. It looked
like a silent failure.
While running the test and observing the behaviour in real time,
I observed the extent coutn growing at ~2M extents/minute, and saw
this after about an hour:
# xfs_io -f -c "stat" /mnt/scratch/big-file |grep next ; \
> sleep 60 ; \
> xfs_io -f -c "stat" /mnt/scratch/big-file |grep next
fsxattr.nextents = 127657993
fsxattr.nextents = 129683339
#
And a few minutes later this:
# xfs_io -f -c "stat" /mnt/scratch/big-file |grep next
fsxattr.nextents = 4177861124
#
Ah, what? Where did that 4 billion extra extents suddenly come from?
Stop the workload, unmount, mount:
# xfs_io -f -c "stat" /mnt/scratch/big-file |grep next
fsxattr.nextents = 166044375
#
And it's back at the expected number. i.e. the extent count is
correct on disk, but it's screwed up in memory. I loaded up the
extent list, and immediately:
# xfs_io -f -c "stat" /mnt/scratch/big-file |grep next
fsxattr.nextents = 4192576215
#
It's bad again. So, where does that number come from?
xfs_fill_fsxattr():
if (ip->i_df.if_flags & XFS_IFEXTENTS)
fa->fsx_nextents = xfs_iext_count(&ip->i_df);
else
fa->fsx_nextents = ip->i_d.di_nextents;
And that's the behaviour I just saw in a nutshell. The on disk count
is correct, but once the tree is loaded into memory, it goes whacky.
Clearly there's something wrong with xfs_iext_count():
inline xfs_extnum_t xfs_iext_count(struct xfs_ifork *ifp)
{
return ifp->if_bytes / sizeof(struct xfs_iext_rec);
}
Simple enough, but 134M extents is 2**27, and that's right about
where things went wrong. A struct xfs_iext_rec is 16 bytes in size,
which means 2**27 * 2**4 = 2**31 and we're right on target for an
integer overflow. And, sure enough:
struct xfs_ifork {
int if_bytes; /* bytes in if_u1 */
....
Once we get 2**27 extents in a file, we overflow if_bytes and the
in-core extent count goes wrong. And when we reach 2**28 extents,
if_bytes wraps back to zero and things really start to go wrong
there. This is where the silent failure comes from - only the first
2**28 extents can be looked up directly due to the overflow, all the
extents above this index wrap back to somewhere in the first 2**28
extents. Hence with a regular pattern, trying to punch a hole in the
range that didn't have holes mapped to a hole in the first 2**28
extents and so "succeeded" without changing anything. Hence "silent
failure"...
Fix this by converting if_bytes to a int64_t and converting all the
index variables and size calculations to use int64_t types to avoid
overflows in future. Signed integers are still used to enable easy
detection of extent count underflows. This enables scalability of
extent counts to the limits of the on-disk format - MAXEXTNUM
(2**31) extents.
Current testing is at over 500M extents and still going:
fsxattr.nextents = 517310478
Reported-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
This function doesn't use the @state parameter, so get rid of it.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
There are many, many xfs header files which are included but
unneeded (or included twice) in the xfs code, so remove them.
nb: xfs_linux.h includes about 9 headers for everyone, so those
explicit includes get removed by this. I'm not sure what the
preference is, but if we wanted explicit includes everywhere,
a followup patch could remove those xfs_*.h includes from
xfs_linux.h and move them into the files that need them.
Or it could be left as-is.
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>
The inode geometry structure isn't related to ondisk format; it's
support for the mount structure. Move it to xfs_shared.h.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
The sequence counter in the xfs_ifork structure is only updated on
COW forks. This is because the counter is currently only used to
optimize out repetitive COW fork checks at writeback time.
Tweak the extent code to update the seq counter regardless of the
fork type in preparation for using this counter on data forks as
well.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
This adds ordering of the updates and makes sure we always see the if_seq
update before the extent tree is modified.
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>
Add a simple 32-bit unsigned integer as the sequence count for
modifications to the extent list in the inode fork. This will be
used to optimize away extent list lookups in the writeback code.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Remove the verbose license text from XFS files and replace them
with SPDX tags. This does not change the license of any of the code,
merely refers to the common, up-to-date license files in LICENSES/
This change was mostly scripted. fs/xfs/Makefile and
fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected
and modified by the following command:
for f in `git grep -l "GNU General" fs/xfs/` ; do
echo $f
cat $f | awk -f hdr.awk > $f.new
mv -f $f.new $f
done
And the hdr.awk script that did the modification (including
detecting the difference between GPL-2.0 and GPL-2.0+ licenses)
is as follows:
$ cat hdr.awk
BEGIN {
hdr = 1.0
tag = "GPL-2.0"
str = ""
}
/^ \* This program is free software/ {
hdr = 2.0;
next
}
/any later version./ {
tag = "GPL-2.0+"
next
}
/^ \*\// {
if (hdr > 0.0) {
print "// SPDX-License-Identifier: " tag
print str
print $0
str=""
hdr = 0.0
next
}
print $0
next
}
/^ \* / {
if (hdr > 1.0)
next
if (hdr > 0.0) {
if (str != "")
str = str "\n"
str = str $0
next
}
print $0
next
}
/^ \*/ {
if (hdr > 0.0)
next
print $0
next
}
// {
if (hdr > 0.0) {
if (str != "")
str = str "\n"
str = str $0
next
}
print $0
}
END { }
$
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Move the tracepoint in xfs_iext_insert to after the point where we've
inserted the extent because otherwise we report stale extent data in
the ftrace output.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
found the issue by kmemleak.
unreferenced object 0xffff8800674611c0 (size 16):
xfs_iext_insert+0x82a/0xa90 [xfs]
xfs_bmap_add_extent_hole_delay+0x1e5/0x5b0 [xfs]
xfs_bmapi_reserve_delalloc+0x483/0x530 [xfs]
xfs_file_iomap_begin+0xac8/0xd40 [xfs]
iomap_apply+0xb8/0x1b0
iomap_file_buffered_write+0xac/0xe0
xfs_file_buffered_aio_write+0x198/0x420 [xfs]
xfs_file_write_iter+0x23f/0x2a0 [xfs]
__vfs_write+0x23e/0x340
vfs_write+0xe9/0x240
SyS_write+0xa1/0x120
do_syscall_64+0xda/0x260
Signed-off-by: Shu Wang <shuwang@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>
Be consistent about using uint32_t/uint8_t instead of u32/u8. This is
more so that we don't have to maintain /those/ types in xfsprogs.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
And also rename fill to nr_entries to match the rest of the code.
Reported-by: Brian Foster <bfoster@redhat.com>
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>
Reported-by: Brian Foster <bfoster@redhat.com>
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>
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>
Reported-by: Brian Foster <bfoster@redhat.com>
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>
Reported-by: Brian Foster <bfoster@redhat.com>
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>
Fix to check the correct value, and remove a duplicate handling of the
uneven record number split algorith,
Reported-by: Brian Foster <bfoster@redhat.com>
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>
We only have two places that remove 2 extents at the same time, so unroll
the loop there.
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 only have two places that insert 2 extents at the same time, so unroll
the loop there.
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>
Replace the current linear list and the indirection array for the in-core
extent list with a b+tree to avoid the need for larger memory allocations
for the indirection array when lots of extents are present. The current
extent list implementations leads to heavy pressure on the memory
allocator when modifying files with a high extent count, and can lead
to high latencies because of that.
The replacement is a b+tree with a few quirks. The leaf nodes directly
store the extent record in two u64 values. The encoding is a little bit
different from the existing in-core extent records so that the start
offset and length which are required for lookups can be retreived with
simple mask operations. The inner nodes store a 64-bit key containing
the start offset in the first half of the node, and the pointers to the
next lower level in the second half. In either case we walk the node
from the beginninig to the end and do a linear search, as that is more
efficient for the low number of cache lines touched during a search
(2 for the inner nodes, 4 for the leaf nodes) than a binary search.
We store termination markers (zero length for the leaf nodes, an
otherwise impossible high bit for the inner nodes) to terminate the key
list / records instead of storing a count to use the available cache
lines as efficiently as possible.
One quirk of the algorithm is that while we normally split a node half and
half like usual btree implementations we just spill over entries added at
the very end of the list to a new node on its own. This means we get a
100% fill grade for the common cases of bulk insertion when reading an
inode into memory, and when only sequentially appending to a file. The
downside is a slightly higher chance of splits on the first random
insertions.
Both insert and removal manually recurse into the lower levels, but
the bulk deletion of the whole tree is still implemented as a recursive
function call, although one limited by the overall depth and with very
little stack usage in every iteration.
For the first few extents we dynamically grow the list from a single
extent to the next powers of two until we have a first full leaf block
and that building the actual tree.
The code started out based on the generic lib/btree.c code from Joern
Engel based on earlier work from Peter Zijlstra, but has since been
rewritten beyond recognition.
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>
