summaryrefslogtreecommitdiffstats
path: root/fs/namei.c
diff options
context:
space:
mode:
authorSage Weil <sage@inktank.com>2012-08-15 13:30:12 -0700
committerMiklos Szeredi <mszeredi@suse.cz>2012-08-16 19:29:09 +0200
commit62b2ce964bb901f00a480104bd35a2e1f8d2cf58 (patch)
treef8014f9c014c84fdabcacfbde91b839ad32de385 /fs/namei.c
parentaf109bca94a8a223c4632a4ff769b3419fe7ed8c (diff)
vfs: fix propagation of atomic_open create error on negative dentry
If ->atomic_open() returns -ENOENT, we take care to return the create error (e.g., EACCES), if any. Do the same when ->atomic_open() returns 1 and provides a negative dentry. This fixes a regression where an unprivileged open O_CREAT fails with ENOENT instead of EACCES, introduced with the new atomic_open code. It is tested by the open/08.t test in the pjd posix test suite, and was observed on top of fuse (backed by ceph-fuse). Signed-off-by: Sage Weil <sage@inktank.com> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Diffstat (limited to 'fs/namei.c')
-rw-r--r--fs/namei.c4
1 files changed, 4 insertions, 0 deletions
diff --git a/fs/namei.c b/fs/namei.c
index 26c28ec4f4a..db76b866a09 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -2489,6 +2489,10 @@ static int atomic_open(struct nameidata *nd, struct dentry *dentry,
dput(dentry);
dentry = file->f_path.dentry;
}
+ if (create_error && dentry->d_inode == NULL) {
+ error = create_error;
+ goto out;
+ }
goto looked_up;
}