Hi,
Just to signal that the last change on spec_vnops.c (1.77) correct my
problem: now the system don't deadlock.
Thanks a lot.
--
Sebastien Marie
On Mon, Oct 21, 2013 at 09:59:43AM +0200, Sébastien Marie wrote:
> On Sat, Oct 19, 2013 at 05:54:22PM +0200, Sébastien Marie wrote:
> > Hi,
> >
> > I fall in a system problem using tmux: the system (OpenBSD -current
> > on i386) freeze (but no panic).
> >
>
> The freeze seems to be a dead-lock, and tmux expose it.
>
> ddb> ps
> PID PPID PGRP UID S FLAGS WAIT COMMAND
> [...]
> 13243 1 13243 0 3 0 inode tmux
> [...]
>
> Here, tmux is waiting "inode".
>
> This wait message is set here:
> ufs/ext2fs/ext2fs_vfsops.c
> 831: lockinit(&ip->i_lock, PINOD, "inode", 0, 0);
>
> ufs/ffs/ffs_vfsops.c
> 1257: lockinit(&ip->i_lock, PINOD, "inode", 0, 0);
>
> And if I let the system running, several others process fall in
> "inode" waiting (as cron, or login_passwd if I try login).
>
>
> With ddb, if I check locked vnodes, there are two on root partition.
>
> ddb> show all mounts
> flags 50000<LOCAL,ROOTFS>
> vnodecovered 0x0 syncer 0xd316aa60 data 0xd108a200
> vfsconf: ops 0xd098d7a0 name "ffs" num 1 ref 3 flags 0x1000
> statvfs cache: bsize 800 iosize 4000
> blocks 403383 free 375320 avail 355151
> files 102910 ffiles 100646 favail 100646
> f_fsidx {0x400, 0xc8a5ad54} owner 0 ctime 0x52640b1d
> syncwrites 325 asyncwrites = 340
> syncreads 8881 asyncreads = 0
> fstype "ffs" mnton "/" mntfrom "/dev/sd0a" mntspec "ab8fcda4850f14e9.a"
> locked vnodes:
> 0xd3165ea8, 0xd316a310
> [... others partitions stripped ...]
>
> ddb> show vnode 0xd3165ea8
> tag UFS(1) type VCHR(4) mount 0xd108b400 typedata 0xd0ffb100
> data 0xd3161298 usecount 2 writecount 0 holdcnt 0 numoutput 0
>
> ddb> show vnode 0xd316a310
> tag UFS(1) type VDIR(2) mount 0xd108b400 typedata 0x0
> data 0xd31851ec usecount 1 writecount 0 holdcnt 3 numoutput 0
>
>
> Does someone have any clue, about what to check or how to debug this ?
> I think I will try the option VFSLCKDEBUG in kernel, but what else ?
> --
> Sébastien Marie
