DragonFly BSD
DragonFly bugs List (threaded) for 2005-07
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: /dev permissions after reboot (and panic)


From: Peter Avalos <pavalos@xxxxxxxxxxxx>
Date: Sun, 17 Jul 2005 02:08:31 -0700
Mail-followup-to: bugs@crater.dragonflybsd.org

On Sun, Jul 17, 2005 at 12:47:41AM -0700, Matthew Dillon wrote:
> 
> :
> :I applied the SOCKBUF_DEBUG patch and rebooted.  After the machine came up,
> :I tried a MAKEDEV, sync, waited about a mintue, then rebooted.  I then got
> :this panic:
> :
> :panic: assertion: sb->sb_mb =3D=3D m in sbunlinkmbuf
> :mp_lock =3D 00000000; cpuid =3D 0; lapic.id =3D 00000000
> :boot() called on cpu#0
> :Uptime: 3m22s
> :
> :Matt, as you know my remote console really sucks, so I wasn't able to
> :get everything, but I did get a sucessfuly dump.  That's uploading to
> :leaf right now (*.12).
> :
> :Hopefully it's useful.
> :
> :--Peter
> 
>     Dump looks good.  The sockbuf has clearly been corrupted.  Not only
>     that, but the debug code checked that the sockbuf was valid at 
>     the beginning of the sbdrop() call and it was fine.
> 
>     This is very encouraging!
> 
>     That patch set did not entirely solve the blocking problem with
>     m_free().  There are still several places where I am calling m_free()
>     without resynchronizing the sockbuf state and I broke a record
>     delimiter in soreceive().  Here is a new patch to try.  In this patch
>     I defer all m_free() calls to the end of the routine to avoid the
>     blocking issue.  It also appears that sbdrop() was being called without
>     a critical section.  This should not have caused a problem but
>     clearly something has raced the sbdrop() call so I'm adding one for
>     good measure.
> 

Ok, that fixed the m_free() warning and subsequent panic on reboot.  It
may take a few days to figure out if it fixes my previously reported panics.

I am still getting a "busy buffer problem" on reboot, and then the /dev
permission problem.  Sometimes I don't even have time to type "cont" at
the db> prompt.  It hard-locks.  Also, attempts to get a dump by typing
'panic' have failed.

--Peter

Attachment: pgp00010.pgp
Description: PGP signature



[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]