Re: 2.6.0-t3: vfs/ext3 do_lookup bug?!

[Andrew Morton <akpm@osdl.org>]

Martin Zwickel <martin.zwickel@technotrend.de> wrote:
>
> On Wed, 20 Aug 2003 23:41:19 -0700
> Andrew Morton <akpm@osdl.org> bubbled:
> 
> > Martin Zwickel <martin.zwickel@technotrend.de> wrote:
> > >
> > > cutted-dmesg.txt  text/plain (15496 bytes)
> > 
> > Try `dmesg -s 1000000'.   The silly thing has too small a buffer.
> 
> too late.. :(
> rebooted and fscked.
> on reboot, my console did hang up while unmounting fs's and i got tons of
> strange errors about something on my fs(where the processes got stuck). can't
> remeber the outputs, was too much and too fast.
> only a sysrq-b helped.
> 
> on another fs i got some "Deleted inode ###### has zero dtime.  Fix<y>?".

Sigh.  Well the filesystem obviously shat itself, so the fsck errors aren't
that surprising.

My guess would be that something oopsed while holding a directory semaphore
and you missed the oops.  Maybe you were in X at the time?

If it happens again, please remember that dmesg needs the `-s 1000000'
option to prevent it from truncating output.

> ps.: 2.6.0-t3 scheduler performance is not that good...

It's pretty bad.  I'm running 2.6.0-test3-mm1 here which has Ingo/Con
goodies and it is significantly improved.

Once that code is working sufficiently well for everyone and there is
consensus that the general direction is correct and the possible
regressions with mixed database workloads are sorted out, we'll fix it up. 
So don't panic yet.