Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 31 May 2011 17:50:46 +0200
From:      Edgar =?iso-8859-1?B?RnXf?= <ef@math.uni-bonn.de>
To:        freebsd-fs@freebsd.org
Subject:   softdep-related panic (allocdirect_merge)
Message-ID:  <20110531155046.GB9327@gumme.math.uni-bonn.de>

next in thread | raw e-mail | index | archive | help
I hope this is the right place to ask.

I'm experiencing softdep-related panics on NetBSD and am trying to find out whether they may be due to a FreeBSD fix not having been ported over.


On a NetBSD-4.0.1/amd NFS server which has happily been runnig soft updates for years without problems, suddenly, presumably because of having updated the ~150 clients, I was struck with two nearly identical panics within hours.
The panics were

allocdirect_merge: ob 0 != nb 244139088 || lbn 12 >= 12 || osize 0 != nsize 16384

with only the nb value different between the two.
The second time, I successfully took a dump. However, I can't possibly make that public because it contains user data. Of course, I can look up whatever may help in analysing the problem.

I'm not expecting much help from the NetBSD side because they are phasing out softdeps in favour of WAPBL. It looks like older unresolved softdep-related PRs have been closed in that course.

I've found two related FreeBSD PRs, namely kern/126089 and kern/23258. Unfortunally, the former has been closed due to feedback timeout and with the later, only the locking-against-myself error is mentioned in the comments.

The system in question is UP.
The filesystem in question was nowhere close to being full when the panic happened. However, it may be that someone is exceeding his quota limit.
There were no snapshots active or taken on the filesystem at the time of the panic.

I've browsed through the CVS history of ffs_softdep.c, but haven't found any commit that looks obviously relevant to my problem. I'm aware that the root cause needn't to be in that code.

Is anyone aware of a fix in the FreeBSD code that has fixed a similar problem? I could then check whether that fix has been ported to NetBSD.

Thank you


P.S. For the time being, I've mounted the relevant filesystem non-softdep.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20110531155046.GB9327>