ceph: do not clear I_COMPLETE from d_release
authorSage Weil <sage@newdream.net>
Mon, 28 Feb 2011 20:49:15 +0000 (12:49 -0800)
committerSage Weil <sage@newdream.net>
Thu, 3 Mar 2011 18:09:52 +0000 (10:09 -0800)
First, this was racy anyway: d_release isn't called until well after the
dentry is unhashed.  Second, this runs afoul of the recent dcache change
that clears d_parent prior to calling d_release (949854d0), causing a NULL
pointer dereference.

Signed-off-by: Sage Weil <sage@newdream.net>
fs/ceph/dir.c

index 9b4f9d9..196fd4c 100644 (file)
@@ -1029,28 +1029,8 @@ out_touch:
 static void ceph_dentry_release(struct dentry *dentry)
 {
        struct ceph_dentry_info *di = ceph_dentry(dentry);
-       struct inode *parent_inode = NULL;
-       u64 snapid = CEPH_NOSNAP;
 
-       if (!IS_ROOT(dentry)) {
-               parent_inode = dentry->d_parent->d_inode;
-               if (parent_inode)
-                       snapid = ceph_snap(parent_inode);
-       }
-       dout("dentry_release %p parent %p\n", dentry, parent_inode);
-       if (parent_inode && snapid != CEPH_SNAPDIR) {
-               struct ceph_inode_info *ci = ceph_inode(parent_inode);
-
-               spin_lock(&parent_inode->i_lock);
-               if (ci->i_shared_gen == di->lease_shared_gen ||
-                   snapid <= CEPH_MAXSNAP) {
-                       dout(" clearing %p complete (d_release)\n",
-                            parent_inode);
-                       ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
-                       ci->i_release_count++;
-               }
-               spin_unlock(&parent_inode->i_lock);
-       }
+       dout("dentry_release %p\n", dentry);
        if (di) {
                ceph_dentry_lru_del(dentry);
                if (di->lease_session)