Re: Uninitialized extent races

[Jan Kara <jack@suse.cz>]

On Tue 01-01-13 00:31:46, Zheng Liu wrote:
> On Mon, Dec 31, 2012 at 09:32:21AM +0100, Jan Kara wrote:
> > On Mon 24-12-12 19:17:45, Zheng Liu wrote:
> > > On Fri, Dec 21, 2012 at 01:02:43PM -0500, Theodore Ts'o wrote:
> > > > On Fri, Dec 21, 2012 at 05:19:29PM +0100, Jan Kara wrote:
> > > > >   No, I'm speaking about merging currently uninitialized extents. I.e.
> > > > > suppose someone does the following on a filesystem with dioread_nolock so
> > > > > that writeback happens via unwritten extents:
> > > > >   fd = open("file", O_RDWR);
> > > > >   pwrite(fd, buf, 4096, 0);
> > > > > 					flusher thread starts writing
> > > > > 					we create uninitialized extent for
> > > > > 					  range 0-4096
> > > > >   fallocate(fd, 0, 4096, 4096);
> > > > >     - we merge extents and now have just 1 uninitialized extent for range
> > > > >       0-8192
> > > > > 					ext4_convert_unwritten_extents() now
> > > > > 					  has to split the extent to finish
> > > > > 					  the IO.
> > > > 
> > > > Ah, I see.  Disabling the the merging that might take place as a
> > > > result of the fallocate.  Yes, I agree that's a completely sane thing
> > > > to do.
> > > > 
> > > > The alternate approach would be to add a flag in the extent status
> > > > tree indicating that an unwritten conversion is pending, but that
> > > > would add more complexity.
> > > 
> > > Sorry for delay reply.  Indeed we could add a flag in extent status tree
> > > to indicate an pending unwritten extent, and I believe that it can bring
> > > us some benefits.  But I wonder whether this case often happens.  Do we
> > > have some real workloads?
> >   It doesn't happen often but it *can* happen. Thus you have to implement
> > a code which handles the case. I don't think bit in extent status tree is
> > really necessary. Just disabling merging of uninitialized extents is
> > simple. If we see there are some real workloads which have problems with
> > it, we can resort to a more complex solution using extent tree...
> 
> Thanks for your explanation.  I don't know whether or not you have
> generated a patch for this problem.  I am willing to make it in a proper
> time.  If you have begun to generate it, please let me know. :-)
  Disabling the merging is trivial and I have a patch for that. Just making
all other changes so that Christoph's DIO patches can work is non-trivial.
I already have several smaller fixes and cleanups to make things easier but
writeback path still has locking issues - I have a solution in mind but
whether it will be needed or not depends on what I asked in the other email
- whether extent status tree can really be used or not...

								Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR