Re: ext4+quota+nfs issue

[Mingming <cmm@us.ibm.com>]

On Tue, 2009-09-15 at 00:22 +0530, Aneesh Kumar K.V wrote:
> On Mon, Sep 14, 2009 at 07:50:56PM +0200, Jan Kara wrote:
> >   I've found some time to look into this and I can see a few problems in
> > the code. Firstly, what may cause your problems:
> >   vfs_dq_claim_blocks() is called in ext4_mb_mark_diskspace_used(). But
> > as far as I can understand the code, ext4_mb_normalize_request() can
> > increase the amount of space we really allocate and thus we try to
> > allocate more blocks than we have actually reserved in quota. Aneesh, is
> > that right?
> 
> ext4_mb_mark_diskspace_used use ac->ac_b_ex.fe_len which is NOT the normalized
> request len. it is min(allocated_len, original_len). So i guess that code
> should be safe
> 

That's right. ac->ac_b_ex.fe_len is the number of blocks that account
for the original request. It is used for update the per-fs free blocks
counter and the per-fs dirty(delayed) blocks counter too. Quota counters
are updated based on this value.

> 
> >   Secondly, ext4_da_reserve_space() seems to have a bug that it can reserve
> > quota blocks multiple times if ext4_claim_free_blocks() fail and we retry
> > the allocation. We should release the quota reservation before restarting.
> > Actually, when we find out we cannot reserve quota space, we could force
> > some delayed allocated writes to disk (thus possibly release some quota
> > in case we have overestimated the amount of blocks needed). But that's
> > a different issue.
> 
> That would imply the file system was full. But the dumpe2fs ouput list
> large number of free blocks. But yes the code should have released the
> quota reservation before trying block reservation again.
> 
> 

I'll send out a patch shortly.

> >   Thirdly, ext4_indirect_calc_metadata_amount() is wrong for sparse files.
> > The worst case is 3 metadata blocks per data block if we make the file
> > sufficiently sparse and there's no easy way around that...
> > 
> 

If this is a real concern, I am all for fix it.  Just we have consider
the worse case before, at that time it seems a little overprotect that
we have always account for the worse case. In regularly cases, we pretty
much always reserve more metadata(indirect) blocks than needed...

> -aneesh
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html