Hi Linus,
Could you pull these changes please? There are two, affecting the
functions that iterates over the pagecache unmarking or unlocking pages
after an op is complete:
(1) xas_for_each() loops must call xas_retry() first thing and immediately
do a "continue" in the case that the extracted value is a special
value that indicates that the walk raced with a modification. Fix the
unlock and unmark loops to do this.
(2) The maths in the unlock loop is dodgy as it could, theoretically, at
some point in the future end up with a starting file pointer that is
in the middle of a folio. This will cause a subtraction to go
negative - but the number is unsigned. Fix the maths to use absolute
file positions instead of relative page indices.
Thanks,
David
Link:
https://lore.kernel.org/r/166749229733.107206.17482609105741691452.st...@warthog.procyon.org.uk/
# v1
Link:
https://lore.kernel.org/r/166757987929.950645.12595273010425381286.st...@warthog.procyon.org.uk/
# v2
---
The following changes since commit f0c4d9fc9cc9462659728d168387191387e903cc:
Linux 6.1-rc4 (2022-11-06 15:07:11 -0800)
are available in the Git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git
tags/netfs-fixes-20221115
for you to fetch changes up to 5e51c627c5acbcf82bb552e17533a79d2a6a2600:
netfs: Fix dodgy maths (2022-11-15 16:56:07 +0000)
----------------------------------------------------------------
netfslib fixes
----------------------------------------------------------------
David Howells (2):
netfs: Fix missing xas_retry() calls in xarray iteration
netfs: Fix dodgy maths
fs/netfs/buffered_read.c | 20 +++++++++++++-------
fs/netfs/io.c | 3 +++
2 files changed, 16 insertions(+), 7 deletions(-)
--
Linux-cachefs mailing list
Linux-cachefs@redhat.com
https://listman.redhat.com/mailman/listinfo/linux-cachefs
