Bug#613925: linux-image-3.2.0-4-amd64: reiserfs: deadlock affecting writes, getdents
reassign 613925 src:linux
thanks
> This bug seems to still exist in the current kernel version for Wheezy.
> But it is very rare; today it happened after 18 days' uptime, and once
> before after 30+ days uptime. This is a desktop machine with a single
> SSD SATA drive with usually only light I/O workload.
>
> Today I noticed all my iceweasel and xfce4-terminal windows had frozen.
>
> >From a virtual terminal, as root, "ls /tmp" froze (in the getdents
> syscall) but could be killed. My other reiserfs partitions on the same
> block device worked fine.
>
> My 'master' xfce4-terminal process was in a blocked state and could not
> be killed. This, and some unreaped xfce4-terminal child processes that
> I'd tried to kill already, had some open, deleted inodes on /tmp
>
> I thought I was going to have to reboot, but just out of curiosity I
> killed iceweasel (via the 'non-responding window' feature of xfce4
> window manager). Immediately my system recovered with /tmp
> readable/writable again.
>
> I think this suggests it was not a hardware issue, but some deadlock
> between processes doing I/O on reiserfs.
>
> Last time this happened to me it was the /var partition rather than /tmp
>
> This problem didn't ever occur in many months of using the Squeeze
> 2.6.32 kernel.
>
> [I would miss reiserfs very much. It has provided me with 100%
> durability for many years and performed well. Even when I once trashed
> the underlying mdraid on another system, reiserfsck rescued my data].
Hi Steven,
hi Bastien,
Jeff Mahoney has posted patches in the upstream bug in May:
https://bugzilla.kernel.org/show_bug.cgi?id=29162
Are you in a position to test them and provide him with feedback?
Cheers,
Moritz
Reply to: