On Mon, Dec 21, 2009 at 09:37:24PM -0700, Eric Blake wrote:
> According to OGAWA Hirofumi on 12/21/2009 8:05 AM:
> >> It may also be file-system dependent. On the machine where I saw the
> >> original failure:
> >>> $ uname -a
> >>> Linux fencepost 2.6.26-2-xen-amd64 #1 SMP Thu Nov 5 04:27:12 UTC 2009
> >>> x86_64 GNU/Linux
> >> $ df -T .
> >> Filesystem Type 1K-blocks Used Available Use% Mounted on
> >> /dev/sdb1 xfs 419299328 269018656 150280672 65% /srv/data
> > Thanks.
> > This is good point. This would be xfs issue or design. xfs seems to have
> > own special handling of ctime.
Yeah, it looks like the change to utimesat() back in 2.6.26 for
posix conformance made ATTR_CTIME appear outside inode truncation
and XFS wasn't updated for this change in behaviour at the VFS level.
Looks simple to fix, but I'm worried about introducing other
unintended ctime modifications - is there a test suite that checks
posix compliant atime/mtime/ctime behaviour around anywhere?