Hi all, Here's a couple patches to add the log recovery delay instrumentation required for reproducing the log recovery buf race problem uncovered by Alex: http://oss.sgi.com/archives/xfs/2014-08/msg00155.html It looked easier to dump this tunable in /proc, but it seems like we want to move away from adding more things there. This series defines a generic sys/fs/xfs/debug location to serve a similar purpose and exports the log recovery delay tunable therein. The original problem is now easily reproduced with a 10s or so log recovery delay and the xfstests test I posted the other day: http://oss.sgi.com/archives/xfs/2014-08/msg00261.html One thing that comes to mind as I write this is whether it might be a good idea to only export this debug subdirectory for DEBUG enabled kernels. Thoughts? Any other thoughts, reviews or flames are appreciated as well. Brian Brian Foster (2): xfs: add debug sysfs attribute set xfs: export log_recovery_delay to delay mount time log recovery fs/xfs/xfs_globals.c | 4 +++ fs/xfs/xfs_log_recover.c | 12 ++++++++ fs/xfs/xfs_super.c | 15 ++++++++-- fs/xfs/xfs_sysctl.h | 5 ++++ fs/xfs/xfs_sysfs.c | 71 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/xfs/xfs_sysfs.h | 1 + 6 files changed, 106 insertions(+), 2 deletions(-) -- 1.8.3.1 _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
