xfs: ensure we capture IO errors correctly
authorDave Chinner <dchinner@redhat.com>
Tue, 12 Mar 2013 12:30:34 +0000 (23:30 +1100)
committerBen Myers <bpm@sgi.com>
Mon, 18 Mar 2013 18:39:10 +0000 (13:39 -0500)
commite001873853d87674dd5b3cfa2851885023616695
tree5a88db119babf206d0822aab6e5b70cc1c975fba
parent3325beed46d8d14d873e94d89ea57ee900dec942
xfs: ensure we capture IO errors correctly

Failed buffer readahead can leave the buffer in the cache marked
with an error. Most callers that then issue a subsequent read on the
buffer do not zero the b_error field out, and so we may incorectly
detect an error during IO completion due to the stale error value
left on the buffer.

Avoid this problem by zeroing the error before IO submission. This
ensures that the only IO errors that are detected those captured
from are those captured from bio submission or completion.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>

(cherry picked from commit c163f9a1760229a95d04e37b332de7d5c1c225cd)
fs/xfs/xfs_buf.c