We do not need to lookup a hashed negative directory since we have
already revalidated it before and have found it to be fine.
This also prevents a crash in cifs_lookup() when it attempts to rehash
the already hashed negative lookup dentry.
The patch has been tested using the reproducer at
https://bugzilla.redhat.com/show_bug.cgi?id=867344#c28
Cc: <redacted> # 3.6.x
Reported-by: Vit Zahradka <redacted>
Signed-off-by: Sachin Prabhu <redacted>
---
fs/cifs/dir.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/fs/cifs/dir.c b/fs/cifs/dir.c
index 7c0a812..e26d0a6 100644
--- a/fs/cifs/dir.c
+++ b/fs/cifs/dir.c
@@ -398,6 +398,12 @@ cifs_atomic_open(struct inode *inode, struct dentry *direntry,
* in network traffic in the other paths.
*/
if (!(oflags & O_CREAT)) {
+ /* Check for hashed negative dentry. We have already revalidated
+ * the dentry and it is fine. No need to perform another lookup.
+ */
+ if (!d_unhashed(direntry))
+ return -ENOENT;
+
struct dentry *res = cifs_lookup(inode, direntry, 0);
if (IS_ERR(res))
return PTR_ERR(res);--
1.7.11.7