Re: ext3_free_blocks

[Theodore Tso <tytso MIT EDU>]

On Wed, Sep 19, 2001 at 09:53:41PM +0100, Stephen C. Tweedie wrote:
> Hi,
> 
> On Wed, Sep 19, 2001 at 01:30:54PM -0700, RW Hawkins wrote:
> > I am getting the following error under moderate load on one of our 
> > fileserver:
> > 
> > Sep 19 12:44:03 ara kernel: EXT3-fs warning (device sd(8,49)): 
> > ext3_free_blocks: bit already cleared for block 35949297
> > Sep 19 12:44:03 ara last message repeated 9 times
> > 
> > When this occurs NFSD goes out of control and the load pegs and the 
> > machine needs a reboot. This is kernel 2.2.19, ext3-0.0.7a with quotas 
> > enabled. Does anyone have an idea what is going on here?
> 
> Well, the immediate problem isn't ext3-specific --- the error message
> is in ext2 too.  The filesystem is trying to delete a block which is
> already free on disk.  That can happen for a huge variety of reasons:
> bad memory, disk corruption, or any sort of kernel bugs including
> filesystem faults, virtual memory faults or driver bugs.

Actually, this error message is the symptom of an old ext3 bug which
happens when you unlink a file locally while it is being written via
NFS.  (This is the one which I was tracking down while I visiting you
in Edinburgh.)

This is fixed in ext3-0.9.9 for the 2.4 kernels.  The patch should be
relatively easy to back-port to the 0.0.7 kernel for ext3 for 2.2, but
it hasn't been done yet.

						- Ted