Re: [PATCH 4/4] ext3: Implement delayed allocation on page_mkwrite time

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 11 May 2011 17:38:41 +0200
Jan Kara <jack@suse.cz> wrote:

> > > > considerations I've decided it's worth it and and fixed the bug...
> > > 
> > > Well.  How did you come to that decision?
> >   So my thoughts were: If a company runs a hosting or similar service and
> > some load either inadvertedly or even maliciously triggers this bug, your
> > systems can be DOSed. That's bad and you need to fix that ASAP. From my
> > experience with our SLE customers, they are willing to listen to advices
> > such as fs choice when they plan a system deployment. After that they
> > vehemently refuse any major change (and fs driver change is a major one).
> > So I'm quite certain they'd rather accept this largish ext3 change.
> > Finally, admittedly, I didn't think the patch will end up so large.
> > 
> > Looking into the patch, I could split off some cleanups and code
> > reorganizations which are 20-30% of the patch but it probably does not make
> > sense to split it more. What I think is a plus of the patch is that there
> > are only two code paths that really change - ext3_get_blocks() has two new
> > cases how it is called (to allocate only indirect blocks and to allocate
> > already reserved data block) and trucate path which has to do more work to
> > check whether indirect block can be removed.
> >   
> > > Are real users hurting from this problem?
> >   I've got a report of this from NEC
> > (http://ns3.spinics.net/lists/linux-ext4/msg20239.html) and OpenVZ people
> > were also concerned
> > (http://ns3.spinics.net/lists/linux-ext4/msg20288.html). I think there was
> > one more report of this problem but I can't find it now. So yes, there are
> > users who care.
> > 
> > >  What's the real-world case for fixing it?
> >   Sorry, I don't understand the question (or how it is different from the
> > previous one).
>   Andrew, does this change your opinion in any way?

I didn't actually have an opinion - I was checking, as usual, that the
proposed change passes the cost-vs-benefit test.

Yes, if real users are hitting the problem in real life then that
strengthens the case for fixing it.