On Tue, Jul 16, 2013 at 10:27:03AM +1000, Dave Chinner wrote:
> On Mon, Jul 15, 2013 at 03:58:00PM -0400, Dave Jones wrote:
> > On Sat, Jul 13, 2013 at 12:00:30PM +1000, Dave Chinner wrote:
> > > > [ 97.162665] XFS: Assertion failed: xfs_dir2_sf_lookup(args) ==
> > ENOENT, file: fs/xfs/xfs_dir2_sf.c, line: 358
> > > ....
> > >
> > > So, it's not clear what has caused this yet. Is it reproducable?
> > Just hit it again on the same machine.
> > > would be good to get a trace of lookup vs addname events from XFS,
> > > too (i.e. all the xfs_dir* and xfs_da* events) so we can see if the
> > > correct lookups were done prior to the failing addname operation...
> > <- xfs tracing noob. Got a recipe I can cut-n-paste ?
> > Or a patch if that's necessary..
> Just use your usual method of pulling tracepoints out of the kernel,
> be it with perf or (my preferred method) trace-cmd:
> # trace-cmd record -e xfs_dir\* -e xfs_da\* <command that dies>
> # trace-cmd report > trace.report.out
> The trace.report.out file is what I'd like to see. The report file
> could be large, so compressing it might be an idea...
Hmm, will that only log traces from the command that I run ?
I might need to do some kind of system-wide tracing here,
because every time this has happened so far, it's happened in
a different process. (rpcbind the first [wtf?], and chrony the 2nd)
Can trace-cmd do that ?