Re: post 2.6.21 regression in F_GETLK

[Doug Chapman]

On Thu, 2007-05-10 at 16:23 -0400, J. Bruce Fields wrote:
> On Thu, May 10, 2007 at 03:38:59PM -0400, bfields wrote:
> > On Thu, May 10, 2007 at 03:30:50PM -0400, bfields wrote:
> > > On Thu, May 10, 2007 at 02:56:15PM -0400, Doug Chapman wrote:
> > > > A recent regression (introduced after 2.6.21) was caught by the LTP test
> > > > fcntl11.  It appears that F_GETLK is not properly checking for existing
> > > > F_RDLCK and allows taking out a write lock.
> 
> Hm, actually, could you double-check the test results?  Looking at your
> test case, it appears that it fails when the lock returned from the
> fcntl(.,F_GETLK,.) has an l_type != F_RDLCK.  That doesn't necessarily
> mean the F_GETLK is reporting no conflict.  I believe the bug is
> actually that it's reporting the wrong kind of conflict--so it's
> returning l_type == F_WRLCK, not F_UNLCK.


You are partly right on the test however note that it is using a start
and len that are specific to the RDLCK so that should _only_ conflict
with that lock.  I did notice that the LTP test is taking a new lock on
the entire file which should be blocked by eithe rthe RDLCK or the WRLCK
and it doesn't check both, I plan on fixing that once this is resolved.

But, much more importantly F_GETLK is returning F_UNLCK saying that
there was no conflict at all.

> Also, this affects only F_GETLK, not F_SETLK, so you're not actually
> managing to acquire a conflicting lock, right?
> 

True, this doesn't actually acquire the lock.  I have not tested to see
if trying a conflicting F_SETLK blocks as it should.  I will test that
later.  I missed lunch today so I won't get back to this until later
tonight or tomorrow....

- Doug



-
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/