Re: [RFC v2 3/4] locks: Split insert/delete block functions into flock/posix parts

[Jeff Layton]

On Wed, 4 Mar 2015 12:59:23 -0500
Jeff Layton <jlayton@xxxxxxxxxxxxxxx> wrote:

> On Wed, 4 Mar 2015 16:32:57 +0100
> Daniel Wagner <daniel.wagner@xxxxxxxxxxxx> wrote:
> 
> > On 03/04/2015 04:00 PM, Boaz Harrosh wrote:
> > > On 03/04/2015 04:20 PM, Daniel Wagner wrote:
> > >> On 03/03/2015 01:55 AM, Jeff Layton wrote:
> > >>> On Mon,  2 Mar 2015 15:25:12 +0100
> > >>> Daniel Wagner <daniel.wagner@xxxxxxxxxxxx> wrote:
> > >>>
> > > <>
> > >> I have fixed that stuff and now I am testing it. Though it seems
> > >> that there is a memory leak which can be triggered with 
> > >>
> > >> 	while true; rm -rf /tmp/a; ./lease02 /tmp/a; done
> > >>
> > >> and this happens also without any of my patches. Still trying to
> > >> figure out what's happening. Hopefully I just see a ghost.
> > >>
> > >> slabtop tells me that ftrace_event_field is constantly growing:
> > >>
> > > 
> > > check out the Kernel's leak detector it is perfect in showing you
> > > what was the exact call stack of the leaked memory.
> > 
> > Thanks for the tip. Will use it in future :)
> > 
> > I have done a quick bisect limit the search on fs/locks.c.
> > I suspect that the file_lock_context refactoring is the source of the leak.
> > bisect agrees with me
> > 
> > 
> > 8634b51f6ca298fb8b07aa4847340764903533ab is the first bad commit
> > commit 8634b51f6ca298fb8b07aa4847340764903533ab
> > Author: Jeff Layton <jlayton@xxxxxxxxxxxxxxx>
> > Date:   Fri Jan 16 15:05:55 2015 -0500
> > 
> >     locks: convert lease handling to file_lock_context
> >     
> >     Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxxxxxxx>
> >     Acked-by: Christoph Hellwig <hch@xxxxxx>
> > 
> > :040000 040000 4114db9392dc4dadb30664b71a954321e5e87bab 5b9abbaf1808a7c926c09fa2164044e0cc26fd54 M      fs
> > :040000 040000 bd569f527a195edf673c4f7d0e80bf356c7f8d1b 6362646e04dd83efc1a9e92877900797ac879e9a M      include
> > 
> 
> Thanks. I'll take a look.
> 

Huh. I'm was a bit surprised by this as I didn't really touch how the
fasync entries get handled. I added a bit of printk debugging
(primitive, I know...) and I see this:

[  458.715319] lease_modify: calling fasync_helper on ffff880035a942d0

So, the fasync_helper getting called on the fasync entry, but it's
definitely not getting freed. When I look at the object in the
debugger, it looks like call_rcu has been called on it though:

  fa_file = 0x0, 
  fa_rcu = {
    next = 0xffff8800ccd6a8a0, 
    func = 0xffffffff8122b1c0 <fasync_free_rcu>
  }

...it's almost like the rcu grace period isn't ending properly? I'll
keep poking at though to see if I can figure out what's going wrong.

-- 
Jeff Layton <jeff.layton@xxxxxxxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/