VFS: Make clone_mnt()/copy_tree()/collect_mounts() return errors
authorDavid Howells <dhowells@redhat.com>
Mon, 25 Jun 2012 11:55:18 +0000 (12:55 +0100)
committerAl Viro <viro@zeniv.linux.org.uk>
Sat, 14 Jul 2012 12:37:27 +0000 (16:37 +0400)
copy_tree() can theoretically fail in a case other than ENOMEM, but always
returns NULL which is interpreted by callers as -ENOMEM.  Change it to return
an explicit error.

Also change clone_mnt() for consistency and because union mounts will add new
error cases.

Thanks to Andreas Gruenbacher <agruen@suse.de> for a bug fix.
[AV: folded braino fix by Dan Carpenter]

Original-author: Valerie Aurora <vaurora@redhat.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Cc: Valerie Aurora <valerie.aurora@gmail.com>
Cc: Andreas Gruenbacher <agruen@suse.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/namespace.c
fs/pnode.c
kernel/audit_tree.c

index 8f412abcb67f6eb9c90a80a14ceb539a8b54c4f9..be1b07a774f14bd7ed0833a63c8977507d08baf9 100644 (file)
@@ -708,56 +708,60 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
                                        int flag)
 {
        struct super_block *sb = old->mnt.mnt_sb;
-       struct mount *mnt = alloc_vfsmnt(old->mnt_devname);
+       struct mount *mnt;
+       int err;
 
-       if (mnt) {
-               if (flag & (CL_SLAVE | CL_PRIVATE))
-                       mnt->mnt_group_id = 0; /* not a peer of original */
-               else
-                       mnt->mnt_group_id = old->mnt_group_id;
-
-               if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
-                       int err = mnt_alloc_group_id(mnt);
-                       if (err)
-                               goto out_free;
-               }
+       mnt = alloc_vfsmnt(old->mnt_devname);
+       if (!mnt)
+               return ERR_PTR(-ENOMEM);
 
-               mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
-               atomic_inc(&sb->s_active);
-               mnt->mnt.mnt_sb = sb;
-               mnt->mnt.mnt_root = dget(root);
-               mnt->mnt_mountpoint = mnt->mnt.mnt_root;
-               mnt->mnt_parent = mnt;
-               br_write_lock(&vfsmount_lock);
-               list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
-               br_write_unlock(&vfsmount_lock);
+       if (flag & (CL_SLAVE | CL_PRIVATE))
+               mnt->mnt_group_id = 0; /* not a peer of original */
+       else
+               mnt->mnt_group_id = old->mnt_group_id;
 
-               if (flag & CL_SLAVE) {
-                       list_add(&mnt->mnt_slave, &old->mnt_slave_list);
-                       mnt->mnt_master = old;
-                       CLEAR_MNT_SHARED(mnt);
-               } else if (!(flag & CL_PRIVATE)) {
-                       if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
-                               list_add(&mnt->mnt_share, &old->mnt_share);
-                       if (IS_MNT_SLAVE(old))
-                               list_add(&mnt->mnt_slave, &old->mnt_slave);
-                       mnt->mnt_master = old->mnt_master;
-               }
-               if (flag & CL_MAKE_SHARED)
-                       set_mnt_shared(mnt);
-
-               /* stick the duplicate mount on the same expiry list
-                * as the original if that was on one */
-               if (flag & CL_EXPIRE) {
-                       if (!list_empty(&old->mnt_expire))
-                               list_add(&mnt->mnt_expire, &old->mnt_expire);
-               }
+       if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
+               err = mnt_alloc_group_id(mnt);
+               if (err)
+                       goto out_free;
+       }
+
+       mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
+       atomic_inc(&sb->s_active);
+       mnt->mnt.mnt_sb = sb;
+       mnt->mnt.mnt_root = dget(root);
+       mnt->mnt_mountpoint = mnt->mnt.mnt_root;
+       mnt->mnt_parent = mnt;
+       br_write_lock(&vfsmount_lock);
+       list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
+       br_write_unlock(&vfsmount_lock);
+
+       if (flag & CL_SLAVE) {
+               list_add(&mnt->mnt_slave, &old->mnt_slave_list);
+               mnt->mnt_master = old;
+               CLEAR_MNT_SHARED(mnt);
+       } else if (!(flag & CL_PRIVATE)) {
+               if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
+                       list_add(&mnt->mnt_share, &old->mnt_share);
+               if (IS_MNT_SLAVE(old))
+                       list_add(&mnt->mnt_slave, &old->mnt_slave);
+               mnt->mnt_master = old->mnt_master;
+       }
+       if (flag & CL_MAKE_SHARED)
+               set_mnt_shared(mnt);
+
+       /* stick the duplicate mount on the same expiry list
+        * as the original if that was on one */
+       if (flag & CL_EXPIRE) {
+               if (!list_empty(&old->mnt_expire))
+                       list_add(&mnt->mnt_expire, &old->mnt_expire);
        }
+
        return mnt;
 
  out_free:
        free_vfsmnt(mnt);
-       return NULL;
+       return ERR_PTR(err);
 }
 
 static inline void mntfree(struct mount *mnt)
@@ -1242,11 +1246,12 @@ struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
        struct path path;
 
        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
-               return NULL;
+               return ERR_PTR(-EINVAL);
 
        res = q = clone_mnt(mnt, dentry, flag);
-       if (!q)
-               goto Enomem;
+       if (IS_ERR(q))
+               return q;
+
        q->mnt_mountpoint = mnt->mnt_mountpoint;
 
        p = mnt;
@@ -1268,8 +1273,8 @@ struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
                        path.mnt = &q->mnt;
                        path.dentry = p->mnt_mountpoint;
                        q = clone_mnt(p, p->mnt.mnt_root, flag);
-                       if (!q)
-                               goto Enomem;
+                       if (IS_ERR(q))
+                               goto out;
                        br_write_lock(&vfsmount_lock);
                        list_add_tail(&q->mnt_list, &res->mnt_list);
                        attach_mnt(q, &path);
@@ -1277,7 +1282,7 @@ struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
                }
        }
        return res;
-Enomem:
+out:
        if (res) {
                LIST_HEAD(umount_list);
                br_write_lock(&vfsmount_lock);
@@ -1285,9 +1290,11 @@ Enomem:
                br_write_unlock(&vfsmount_lock);
                release_mounts(&umount_list);
        }
-       return NULL;
+       return q;
 }
 
+/* Caller should check returned pointer for errors */
+
 struct vfsmount *collect_mounts(struct path *path)
 {
        struct mount *tree;
@@ -1295,7 +1302,9 @@ struct vfsmount *collect_mounts(struct path *path)
        tree = copy_tree(real_mount(path->mnt), path->dentry,
                         CL_COPY_ALL | CL_PRIVATE);
        up_write(&namespace_sem);
-       return tree ? &tree->mnt : NULL;
+       if (IS_ERR(tree))
+               return NULL;
+       return &tree->mnt;
 }
 
 void drop_collected_mounts(struct vfsmount *mnt)
@@ -1590,14 +1599,15 @@ static int do_loopback(struct path *path, char *old_name,
        if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
                goto out2;
 
-       err = -ENOMEM;
        if (recurse)
                mnt = copy_tree(old, old_path.dentry, 0);
        else
                mnt = clone_mnt(old, old_path.dentry, 0);
 
-       if (!mnt)
-               goto out2;
+       if (IS_ERR(mnt)) {
+               err = PTR_ERR(mnt);
+               goto out;
+       }
 
        err = graft_tree(mnt, path);
        if (err) {
@@ -2211,10 +2221,10 @@ static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
        down_write(&namespace_sem);
        /* First pass: copy the tree topology */
        new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
-       if (!new) {
+       if (IS_ERR(new)) {
                up_write(&namespace_sem);
                kfree(new_ns);
-               return ERR_PTR(-ENOMEM);
+               return ERR_CAST(new);
        }
        new_ns->root = new;
        br_write_lock(&vfsmount_lock);
index bed378db075813350362c39f423d1b4335240bfa..3e000a51ac0d09556d184d26422a91d0bc4c0ff9 100644 (file)
@@ -237,8 +237,9 @@ int propagate_mnt(struct mount *dest_mnt, struct dentry *dest_dentry,
 
                source =  get_source(m, prev_dest_mnt, prev_src_mnt, &type);
 
-               if (!(child = copy_tree(source, source->mnt.mnt_root, type))) {
-                       ret = -ENOMEM;
+               child = copy_tree(source, source->mnt.mnt_root, type);
+               if (IS_ERR(child)) {
+                       ret = PTR_ERR(child);
                        list_splice(tree_list, tmp_list.prev);
                        goto out;
                }
index 5bf0790497e7779cf1001f6cb30c214de812ccd5..3a5ca582ba1ebe2373803bfbf1887357d05b523e 100644 (file)
@@ -595,7 +595,7 @@ void audit_trim_trees(void)
 
                root_mnt = collect_mounts(&path);
                path_put(&path);
-               if (!root_mnt)
+               if (IS_ERR(root_mnt))
                        goto skip_it;
 
                spin_lock(&hash_lock);
@@ -669,8 +669,8 @@ int audit_add_tree_rule(struct audit_krule *rule)
                goto Err;
        mnt = collect_mounts(&path);
        path_put(&path);
-       if (!mnt) {
-               err = -ENOMEM;
+       if (IS_ERR(mnt)) {
+               err = PTR_ERR(mnt);
                goto Err;
        }
 
@@ -719,8 +719,8 @@ int audit_tag_tree(char *old, char *new)
                return err;
        tagged = collect_mounts(&path2);
        path_put(&path2);
-       if (!tagged)
-               return -ENOMEM;
+       if (IS_ERR(tagged))
+               return PTR_ERR(tagged);
 
        err = kern_path(old, 0, &path1);
        if (err) {