[linux-cifs-client] [PATCH] cifs: Detect errors in cifs_writepages when called from pdflush

[Peter Schwenke]

The return code isn't checked in the case of cifs_writepages being
invoked as a result of pdflush. This results in data integrity
problems when errors such as a network break occur when
CIFSSMBWrite2 is called.  The resulting EAGAIN isn't acted
on.

Signed-off-by: Peter Schwenke <peter at bluetoad.com.au>
---
 fs/cifs/file.c |   11 +++++++++++
 1 files changed, 11 insertions(+), 0 deletions(-)

diff --git a/fs/cifs/file.c b/fs/cifs/file.c
index 042b122..8cc1094 100644
--- a/fs/cifs/file.c
+++ b/fs/cifs/file.c
@@ -1374,6 +1374,17 @@ retry:
 						set_bit(AS_ENOSPC, &mapping->flags);
 					else
 						set_bit(AS_EIO, &mapping->flags);
+					/*
+					 * The return code isn't checked in the
+					 * case of cifs_writepages being
+					 * invoked as a result of pdflush.
+					 * See generic_sync_sb_inodes()
+					 */
+					if (current_is_pdflush())
+						if (rc == -ENOSPC)
+							CIFS_I(mapping->host)->write_behind_rc =  rc;
+						else
+							CIFS_I(mapping->host)->write_behind_rc =  -EIO;
 				} else {
 					cifs_stats_bytes_written(cifs_sb->tcon,
 								 bytes_written);
-- 
1.5.4.3