On Sun, Aug 23, 2009 at 05:33:35PM -0700, Steve Langasek wrote: > I agree that these are equally as important as *passdb.tdb* to back up. But > "copy to /var/backups" is not an appropriate backup strategy - these files > are already under /var/lib and should therefore already be part of any > system backup policy. We should not be backing them up piecemeal in the > samba package.
Ah, OK. I was under the impression that /var/backups was meant for "critical odds-and-ends that should be backed up by default even on systems managed by extremely lax sysadmins". A bit of googling for "debian policy /var/backups" shows similar confusion by others. > > One other point: many well-meaning sysadmins may be doing rsync backups of > > their entire systems, including /var/lib/samba, thinking that they are safe > > in > > case of any catastrophe. But they might not be aware that the modification > > date on the *.tdb files rarely gets changed, even if the contents do, > > because > > Samba uses mmap() to access them. > > Er, use of mmap() should not affect the filesystem semantics of mtime. I > think you're mistaken here. You are right, mmap() *should* update mtime according to "man mmap". But I do not believe it ever since I got bitten by a bug like http://his.luky.org/ML/linux-kernel.2005/msg28040.html or http://his.luky.org/ML/linux-kernel.2005/msg28040.html a number of years ago. I have been cautious of how I backup mmapped data ever since. FYI, I just did a test with smbpasswd on my Lenny system with the 2.6.26-2-vserver-686 kernel and the /var/lib/samba/passdb.tdb file changed mtime as it should. Good. -- Erik Rossen OpenPGP key: 2935D0B9 ros...@rossen.ch On fonce la tête dans http://www.rtfm-sarl.ch le guidon et on RTFM. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
