xfs: fix periodic log flushing
authorDave Chinner <dchinner@redhat.com>
Fri, 25 Jan 2013 18:45:07 +0000 (12:45 -0600)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 4 Feb 2013 00:27:07 +0000 (18:27 -0600)
commit8121aecbe073a880758ab5464a305361dbcb7134
tree75a52c90ec3314136e621611489110853884525d
parentd8525659e782fa14ce8baa2b197d03dac3a7cc90
xfs: fix periodic log flushing

[Please take this patch for -stable in kernels 3.5-3.7.  It doesn't have an
equivalent upstream commit because the code was removed before the bug was
discovered.  See f661f1e0bf50 and 7e18530bef6a.]

There is a logic inversion in xfssyncd_worker() which means that the
log is not periodically forced or idled correctly. This means that
metadata changes aggregated in memory do not get flushed in a timely
manner, and hence if filesystem is not cleanly unmounted those
changes can be lost. This loss can manifest itself even hours after
the changes were made if the filesystem is left to idle without a
sync() occurring between the last modification and the
crash/shutdown occuring.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/xfs/xfs_sync.c