summaryrefslogtreecommitdiff
path: root/fs/configfs/dir.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2020-11-14 03:07:53 +0300
committerLinus Torvalds <torvalds@linux-foundation.org>2020-11-14 03:07:53 +0300
commitf01c30de86f1047e9bae1b1b1417b0ce8dcd15b1 (patch)
treebc1c21aae8cd50d1206bdee027df866e9d78ba24 /fs/configfs/dir.c
parentd9315f5634c94500b91039895f40051a7ac79e28 (diff)
parent9b8523423b23ee3dfd88e32f5b7207be56a4e782 (diff)
downloadlinux-f01c30de86f1047e9bae1b1b1417b0ce8dcd15b1.tar.xz
Merge tag 'vfs-5.10-fixes-2' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
Pull fs freeze fix and cleanups from Darrick Wong: "A single vfs fix for 5.10, along with two subsequent cleanups. A very long time ago, a hack was added to the vfs fs freeze protection code to work around lockdep complaints about XFS, which would try to run a transaction (which requires intwrite protection) to finalize an xfs freeze (by which time the vfs had already taken intwrite). Fast forward a few years, and XFS fixed the recursive intwrite problem on its own, and the hack became unnecessary. Fast forward almost a decade, and latent bugs in the code converting this hack from freeze flags to freeze locks combine with lockdep bugs to make this reproduce frequently enough to notice page faults racing with freeze. Since the hack is unnecessary and causes thread race errors, just get rid of it completely. Making this kind of vfs change midway through a cycle makes me nervous, but a large enough number of the usual VFS/ext4/XFS/btrfs suspects have said this looks good and solves a real problem vector. And once that removal is done, __sb_start_write is now simple enough that it becomes possible to refactor the function into smaller, simpler static inline helpers in linux/fs.h. The cleanup is straightforward. Summary: - Finally remove the "convert to trylock" weirdness in the fs freezer code. It was necessary 10 years ago to deal with nested transactions in XFS, but we've long since removed that; and now this is causing subtle race conditions when lockdep goes offline and sb_start_* aren't prepared to retry a trylock failure. - Minor cleanups of the sb_start_* fs freeze helpers" * tag 'vfs-5.10-fixes-2' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux: vfs: move __sb_{start,end}_write* to fs.h vfs: separate __sb_start_write into blocking and non-blocking helpers vfs: remove lockdep bogosity in __sb_start_write
Diffstat (limited to 'fs/configfs/dir.c')
0 files changed, 0 insertions, 0 deletions