David Chinner wrote: >>> Also, the iolock can be held across I/O so it's possible you've lost an >>> I/O. Any I/O errors in the syslog? >> No, no I/O errors at all. See the kern.log above, I could even do dd(1) >> from the md1 (dm-crypt on raid1), no errors either. > > Oh, dm-crypt. Well, I'd definitely start looking there. XFS has a > history of exposing dm-crypt bugs, and these hangs appear to be I/O > congestion/scheduling related and not XFS. Also, we haven't changed > anything related to plug/unplug of block devices in XFS recently, so > that also points to some other change as well... Yes, there is bug in dm-crypt... Please try if the patch here helps: http://lkml.org/lkml/2008/3/14/71 Thanks, Milan -- mbroz@xxxxxxxxxx -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
