On Thu, Oct 25, 2012 at 09:01:44PM +0100, Stuart Henderson wrote:
> On 2012/10/25 20:17, Han Boetes wrote:
> all processes wanting disk i/o stuck with waitchan inode - I've seen
> that before as have a few others I think - I haven't hit it myself on
> -current for months now though.

I know this bug, too.  When I analysed it on my 5.2 box, I found a
ksh that did try to exit, did hold the vnode log and was waiting
for ttyout.  The potential reader of the pseudo tty was a sshd
waiting for the vnode lock.  Could be something simmilar here.

This tmux tries to exit:

> >    PID   PPID   PGRP    UID  S       FLAGS  WAIT          COMMAND
> >  23658      1  23658   1000  3      0x2080  ttyout        tmux

In this state it holds a inode lock to its /dev/ttyp trying to write
some final output.  The reading end of the pseudo tty might be one
of those sshd waiting for the inode lock.  This would be the dead
lock:

> >  18819  15214  18819      0  3    0x200000  inode         sshd
> >   9596  15214   9596      0  3    0x200000  inode         sshd
> >  26335  15214  26335      0  3    0x200000  inode         sshd
> >  11569  15214  11569      0  3    0x200000  inode         sshd
> >  13416  15214  13416      0  3    0x200000  inode         sshd
> >  19251  22785  22785   1000  3           0  inode         sshd
> >   8787  15214   8787      0  3           0  inode         sshd

This is just a wild guess what might went wrong here, based on
something similar I have seen before.

bluhm

Reply via email to