On 4/10/12 2:12 AM, Dave Chinner wrote: > On Thu, Apr 05, 2012 at 12:28:35PM -0700, Eric Sandeen wrote: ... >> kmem_free(pag); >> + pag = NULL; >> } > > As it is, I don't quite understand what setting a local variable to > NULL is supposed to acheive here - the perag structure has been > removed from the radix tree and zeroing a local variable doesn't do > anything globally visible... er, right. Ok. > What is more likely is that the log mount failure path is not > flushing the buftarg correctly before tearing down the perag > structures. Indeed, a log mount failure (and a perag data > intialisation failure) will simply tear down the perag structures > without waiting for cached buffers to be flushed in xfs_mountfs(), > so that is going to the root cause of this problem. i.e. we need > xfs_flush_buftarg() calls in the mountfs error handling path.... Ok, makes sense. -Eric > Cheers, > > Dave. _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
