This series addresses an issue where XFS might not actually flush the
inode size if we wrote out the inode data during umount, which was
recently reported on #xfs on irc. (If the reported is active on the
list please chime in so that I can add your reported-by tag!).
The first patch is the guts of the fix, and the second fixes various
issues with ->write_inode that I found during the audit. To me patch
1 is a clear Linux 3.1 candidate, and I would tend to nominate patch 2
as well.
I think we also have some issue with the VFS writeback code not handling
inode redirtying from the I/O completion handler nicely, but I'll send
another patch for that.
|