Re: Oops and Panics in 2.6.21.1, 2.6.20.6 and 2.6.19.2

[Tejun Heo <htejun@gmail.com>]

Andrew Morton wrote:
>>> a number of people have hit that, on and off.
>> Yeah, I've been seeing that one.  It should have been fixed with the big
>> fat patchset.
> 
> Great - fingers crossed.
> 
>>> We were close to having a fix, I think, but then we decided that great
>>> chunks of sysfs needed rewriting and I believe that we believe that this
>>> great rewrite will fix this bug.
>> How were we gonna fix it?  If it isn't too complex, I can cook up a
>> patch for -stable series.
> 
> Do we actually understand the causes?

Yeah, I think I do.  Basically, the problem is that on-demand attach and
reclamation update sd->s_dentry but accesses to it aren't synchronized
properly.  In the big fat patchset, first I tried to fix it by removing
sd->s_dentry completely which didn't work because of shadow nodes, so
the second try was to fix the synchronization which is in -mm now.

>>  The safest approach I can think of is making
>> dentries for attributes unreclaimable but those are made reclaimable for
>> good reasons.  :-(
> 
> Yeah, that was the google workaround.  It's OK unless you happen to have
> thousands of disks on an ia32 box.

I see.  I thought there was different approach on fixing the problem.
I'll try to backport the synchronization fix but am afraid it can be too
risky for -stable.  If it seems too risky, I'll send a patch to disable
reclamation.

-- 
tejun
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/