Skip to content

Latest commit

 

History

History
155 lines (110 loc) · 8.26 KB

xfs-bugs.md

File metadata and controls

155 lines (110 loc) · 8.26 KB
Raw

Table of Contents

Upstream bugs reported on kdevops for XFS

There are plenty of bugs found and reported by kdevops, you can use git grep korg for bugs reported bugzilla.kernel.org but there are also many bugs reported on other bugzillas for other distributions. This section tracks bugs perhaps not yet reported on linux-next or rc kernels.

Process for stable fixes

A table fix candidate does by no means indicate that we should immediately send the fix to stable, this is precisely why we have stable XFS maintainers to avoid fiascos to users like the recent ext4 data corruption and so someone has to go through the trouble of ensuring first:

    1. A baseline is created for the respective stable kernel
    1. Verifying that the fix does not regress that stable kernel

Please coordinate with the stable XFS maintainers:

Likewise if the issue is found not yet merged upstream a fix first needs to be vetted by the community, and ultimately the release manager will decide when / if to merge something. The release manager for XFS is Chandan Babu R chandan.babu@oracle.com.

v6.6-rc5

We have found XFS bugs and also memory management bugs. We spit this into two sections.

Defining critical bugs

Critical bugs are:

  • Crashes
  • Asserts with memory or XFS debug enabled
  • hung tasks

Critial bugs count

Since we strive to report all critical bugs and use a notation to describe their respective bugzilla.kernel.org entries we can find all critical bugs as follows:

cat workflows/fstests/expunges/6.6.0-rc5/xfs/unassigned/*.txt| egrep -i "korg|hung task"| wc -l
46

Non critical bugs count

Non critical bugs can be found withe following then:

cat workflows/fstests/expunges/6.6.0-rc5/xfs/unassigned/*.txt| grep -v korg | grep -i -v "hung task"| wc -l
410

The explanation as to what caused some of these issues can be found by using git blame on respective commit if the commit author had time and the confidence to explain the bug. Otherwise you can help to diagnose and look into the issue more carefully by inspecting the archive of results for the bug.

XXX: Add a script which let's a developer easily look for logs for a respective bug on all archives for a specific kernel.

XFS bugs

Below are the bugs reported to the community which have been found using v6.6-rc5 as a baseline.

1) korg#218224 - XFS: Assertion failed: ip->i_nblocks == 0 file: fs/xfs/xfs_inode.c, line: 2359

Luis noted how this resembles a bug found previously on s390, Chinner suggested hat perhaps the fix for that issue could be tried to see if it resolves the crash on x86_64. The fix is commit 7930d9e10370 ("xfs: recovery should not clear di_flushiter unconditionally").

git describe --contains 7930d9e10370
v6.7-rc2~10^2

This tells us that the fix was first merged into v6.7-rc2~10, and so clearly not part of v6.6-rc5 which was where the issues were found. If this fix is confirmed to resolve these issues it would be a good stable backport candidate to evaluate for integration.

2) korg#218225 - xfs assert (irec->br_blockcount & ~XFS_IEXT_LENGTH_MASK) == 0 file: fs/xfs/libxfs/xfs_iext_tree.c, line: 58

No feedback from the community yet.

3) korg#218226 - XFS: Assertion failed: bp->b_flags & XBF_DONE, file: fs/xfs/xfs_trans_buf.c, line: 241

Chinner noted that this is a known issue. Christoph posted a patch but Chinner does not believe that is a right fix and so a proper fix is still pending.

4) korg#218229 - xfs/438 hung

Chandan noted that Leah had fixed this, this is confirmed, this is fixed by commit 471de20303dd ("xfs: up(ic_sema) if flushing data device fails") merged on v6.7-rc2.

5) korg#218230 - xfs/538 hung

Chandan has root caused this issue and is working on an issue.

Memory management bugs

Note that the memory management folks do not want us to use bugzilla.kernel.org for memory management bugs. They want us to instead report issues to the mailing list directly.

1) korg#216114 - page dumped because: VM_BUG_ON_FOLIO(!folio_contains(folio, index)) and kernel BUG at mm/truncate.c:669!

Matthew acknowledged that this is a terribly rare issue, and suggested and suggested a patch to try to see if it fixes the issue.

This is confirmed to be fixed by commit fc346d0a70a1 ("mm: migrate high-order folios in swap cache correctly") merged as of v6.7-rc8.

2) korg#218227 - fsstress + compaction

Ongoing discussion here Vlastimil noted this is caused as of commit 9c5ccf2db04b ("mm: remove HUGETLB_PAGE_DTOR")).

It is not yet clear what to do or how to resolve this.