[RFC] cachefiles: vfs_mkdir() might succeed leaving dentry negative unhashed

Commit Message

Al Viro May 11, 2018, 9:16 p.m. UTC

[same story as with nfsd; recovery is even simpler here]

That can (and does, on some filesystems) happen - ->mkdir() (and thus
vfs_mkdir()) can legitimately leave its argument negative and just
unhash it, counting upon the lookup to pick the object we'd created
next time we try to look at that name.

Some vfs_mkdir() callers forget about that possibility...

Signed-off-by: Al Viro <[e-mail]>
---

Patch

diff --git a/fs/cachefiles/namei.c b/fs/cachefiles/namei.c
index 0daa1e3fe0df..ab0bbe93b398 100644
--- a/fs/cachefiles/namei.c
+++ b/fs/cachefiles/namei.c
@@ -572,6 +572,11 @@  int cachefiles_walk_to_object(struct cachefiles_object *parent,
 			if (ret < 0)
 				goto create_error;
 
+			if (unlikely(d_unhashed(next))) {
+				dput(next);
+				inode_unlock(d_inode(dir));
+				goto lookup_again;
+			}
 			ASSERT(d_backing_inode(next));
 
 			_debug("mkdir -> %p{%p{ino=%lu}}",
@@ -764,6 +769,7 @@  struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
 	/* search the current directory for the element name */
 	inode_lock(d_inode(dir));
 
+retry:
 	start = jiffies;
 	subdir = lookup_one_len(dirname, dir, strlen(dirname));
 	cachefiles_hist(cachefiles_lookup_histogram, start);
@@ -793,6 +799,10 @@  struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
 		if (ret < 0)
 			goto mkdir_error;
 
+		if (unlikely(d_unhashed(subdir))) {
+			dput(subdir);
+			goto retry;
+		}
 		ASSERT(d_backing_inode(subdir));
 
 		_debug("mkdir -> %p{%p{ino=%lu}}",