Chandan Babu reports the following livelock in xfs/708:
run fstests xfs/708 at 2024-05-04 15:35:29
XFS (loop16): EXPERIMENTAL online scrub feature in use. Use at your own risk!
XFS (loop5): Mounting V5 Filesystem e96086f0-a2f9-4424-a1d5-c75d53d823be
XFS (loop5): Ending clean mount
XFS (loop5): Quotacheck needed: Please wait.
XFS (loop5): Quotacheck: Done.
XFS (loop5): EXPERIMENTAL online scrub feature in use. Use at your own risk!
INFO: task xfs_io:143725 blocked for more than 122 seconds.
Not tainted 6.9.0-rc4+ #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
task:xfs_io state:D stack:0 pid:143725 tgid:143725 ppid:117661 flags:0x00004006
Call Trace:
<TASK>
__schedule+0x69c/0x17a0
schedule+0x74/0x1b0
io_schedule+0xc4/0x140
folio_wait_bit_common+0x254/0x650
shmem_undo_range+0x9d5/0xb40
shmem_evict_inode+0x322/0x8f0
evict+0x24e/0x560
__dentry_kill+0x17d/0x4d0
dput+0x263/0x430
__fput+0x2fc/0xaa0
task_work_run+0x132/0x210
get_signal+0x1a8/0x1910
arch_do_signal_or_restart+0x7b/0x2f0
syscall_exit_to_user_mode+0x1c2/0x200
do_syscall_64+0x72/0x170
entry_SYSCALL_64_after_hwframe+0x76/0x7e
The shmem code is trying to drop all the folios attached to a shmem
file and gets stuck on a locked folio after a bnobt repair. It looks
like the process has a signal pending, so I started looking for places
where we lock an xfile folio and then deal with a fatal signal.
I found a bug in xfarray_sort_scan via code inspection. This function
is called to set up the scanning phase of a quicksort operation, which
may involve grabbing a locked xfile folio. If we exit the function with
an error code, the caller does not call xfarray_sort_scan_done to put
the xfile folio. If _sort_scan returns an error code while si->folio is
set, we leak the reference and never unlock the folio.
Therefore, change xfarray_sort to call _scan_done on exit. This is safe
to call multiple times because it sets si->folio to NULL and ignores a
NULL si->folio. Also change _sort_scan to use an intermediate variable
so that we never pollute si->folio with an errptr.
Fixes: 232ea05277 ("xfs: enable sorting of xfile-backed arrays")
Reported-by: Chandan Babu R <chandanbabu@kernel.org>
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
We really don't want to call cond_resched every single time we go
through a loop in scrub -- there may be billions of records, and probing
into the scheduler itself has overhead. Reduce this overhead by only
calling cond_resched 10x per second; and add a counter so that we only
check jiffies once every 1000 records or so.
Surprisingly, this reduces scrub-only fstests runtime by about 2%. I
used the bmapinflate xfs_db command to produce a billion-extent file and
this stupid gadget reduced the scrub runtime by about 4%.
From a stupid microbenchmark of calling these things 1 billion times, I
estimate that cond_resched costs about 5.5ns per call; jiffes costs
about 0.3ns per read; and fatal_signal_pending costs about 0.4ns per
call.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
If the extended attributes look bad, try to sift through the rubble to
find whatever keys/values we can, stage a new attribute structure in a
temporary file and use the atomic extent swapping mechanism to commit
the results in bulk.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Convert xfarray_pagesort to handle large folios by introducing a new
xfile_get_folio routine that can return a folio of arbitrary size, and
using heapsort on the full folio. This also corrects an off-by-one bug
in the calculation of len in xfarray_pagesort that was papered over by
xfarray_want_pagesort.
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Kent Overstreet <kent.overstreet@linux.dev>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
xfiles are shmem files, not memfds.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
Now that xfile pages don't need kmapping, there is no need to cache
the kernel virtual address for them.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
XFS is generally used on 64-bit, non-highmem platforms and xfile
mappings are accessed all the time. Reduce our pain by not allowing
any highmem mappings in the xfile page cache and remove all the kmap
calls for it.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
All current and pending xfile users use the xfile_obj_load
and xfile_obj_store API, so make those the actual implementation.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
Now that we have the means to do insertion sorts of small in-memory
subsets of an xfarray, use it to improve the quicksort pivot algorithm
by reading 7 records into memory and finding the median of that. This
should prevent bad partitioning when a[lo] and a[hi] end up next to each
other in the final sort, which can happen when sorting for cntbt repair
when the free space is extremely fragmented (e.g. generic/176).
This doesn't speed up the average quicksort run by much, but it will
(hopefully) avoid the quadratic time collapse for which quicksort is
famous.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Kent Overstreet <kent.overstreet@linux.dev>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
After quicksort picks a pivot item for a particular subsort, it walks
the records in that subset from the outside in, rearranging them so that
every record less than the pivot comes before it, and every record
greater than the pivot comes after it. This scan has a lot of locality,
so we can speed it up quite a bit by grabbing the xfile backing page and
holding onto it as long as we possibly can. Doing so reduces the
runtime by another 5% on the author's computer.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Kent Overstreet <kent.overstreet@linux.dev>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
If all the records in an xfarray subset live within the same memory
page, we can short-circuit even more quicksort recursion by mapping that
page into the local CPU and using the kernel's heapsort function to sort
the subset. On the author's computer, this reduces the runtime by
another 15% on a 500,000 element array.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Kent Overstreet <kent.overstreet@linux.dev>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
In the previous patch, we created a very basic quicksort implementation
for xfile arrays. While the use of an alternate sorting algorithm to
avoid quicksort recursion on very small subsets reduces the runtime
modestly, we could do better than a load and store-heavy insertion sort,
particularly since each load and store requires a page mapping lookup in
the xfile.
For a small increase in kernel memory requirements, we could instead
bulk load the xfarray records into memory, use the kernel's existing
heapsort implementation to sort the records, and bulk store the memory
buffer back into the xfile. On the author's computer, this reduces the
runtime by about 5% on a 500,000 element array.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Kent Overstreet <kent.overstreet@linux.dev>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
The btree bulk loading code requires that records be provided in the
correct record sort order for the given btree type. In general, repair
code cannot be required to collect records in order, and it is not
feasible to insert new records in the middle of an array to maintain
sort order.
Implement a sorting algorithm so that we can sort the records just prior
to bulk loading. In principle, an xfarray could consume many gigabytes
of memory and its backing pages can be sent out to disk at any time.
This means that we cannot map the entire array into memory at once, so
we must find a way to divide the work into smaller portions (e.g. a
page) that /can/ be mapped into memory.
Quicksort seems like a reasonable fit for this purpose, since it uses a
divide and conquer strategy to keep its average runtime logarithmic.
The solution presented here is a port of the glibc implementation, which
itself is derived from the median-of-three and tail call recursion
strategies outlined by Sedgwick.
Subsequent patches will optimize the implementation further by utilizing
the kernel's heapsort on directly-mapped memory whenever possible, and
improving the quicksort pivot selection algorithm to try to avoid O(n^2)
collapses.
Note: The sorting functionality gets its own patch because the basic big
array mechanisms were plenty for a single code patch.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Kent Overstreet <kent.overstreet@linux.dev>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Create a simple 'big array' data structure for storage of fixed-size
metadata records that will be used to reconstruct a btree index. For
repair operations, the most important operations are append, iterate,
and sort.
Earlier implementations of the big array used linked lists and suffered
from severe problems -- pinning all records in kernel memory was not a
good idea and frequently lead to OOM situations; random access was very
inefficient; and record overhead for the lists was unacceptably high at
40-60%.
Therefore, the big memory array relies on the 'xfile' abstraction, which
creates a memfd file and stores the records in page cache pages. Since
the memfd is created in tmpfs, the memory pages can be pushed out to
disk if necessary and we have a built-in usage limit of 50% of physical
memory.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Kent Overstreet <kent.overstreet@linux.dev>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
