vfs: fix renameat to retry on ESTALE errors
Jeff Layton [Tue, 11 Dec 2012 17:10:10 +0000 (12:10 -0500)]
...as always, rename is the messiest of the bunch. We have to track
whether to retry or not via a separate flag since the error handling
is already quite complex.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

fs/namei.c

index 8a262c2..43a97ee 100644 (file)
@@ -3840,15 +3840,17 @@ SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
        struct nameidata oldnd, newnd;
        struct filename *from;
        struct filename *to;
+       unsigned int lookup_flags = 0;
+       bool should_retry = false;
        int error;
-
-       from = user_path_parent(olddfd, oldname, &oldnd, 0);
+retry:
+       from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
        if (IS_ERR(from)) {
                error = PTR_ERR(from);
                goto exit;
        }
 
-       to = user_path_parent(newdfd, newname, &newnd, 0);
+       to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
        if (IS_ERR(to)) {
                error = PTR_ERR(to);
                goto exit1;
@@ -3920,11 +3922,18 @@ exit3:
        unlock_rename(new_dir, old_dir);
        mnt_drop_write(oldnd.path.mnt);
 exit2:
+       if (retry_estale(error, lookup_flags))
+               should_retry = true;
        path_put(&newnd.path);
        putname(to);
 exit1:
        path_put(&oldnd.path);
        putname(from);
+       if (should_retry) {
+               should_retry = false;
+               lookup_flags |= LOOKUP_REVAL;
+               goto retry;
+       }
 exit:
        return error;
 }