summaryrefslogtreecommitdiffstats
path: root/fs/namespace.c
diff options
context:
space:
mode:
authorAl Viro <viro@zeniv.linux.org.uk>2011-01-17 01:47:59 -0500
committerAl Viro <viro@zeniv.linux.org.uk>2011-01-17 01:47:59 -0500
commitb1e75df45a3d8a490b8648e44632debc5eea04b1 (patch)
tree4ee3aaa516675ca9715d673c4399f1b517d4a690 /fs/namespace.c
parent15f9a3f3e199647fe0cac19302c5033cf031372d (diff)
downloadlinux-b1e75df45a3d8a490b8648e44632debc5eea04b1.tar.gz
linux-b1e75df45a3d8a490b8648e44632debc5eea04b1.tar.bz2
linux-b1e75df45a3d8a490b8648e44632debc5eea04b1.zip
tidy up around finish_automount()
do_add_mount() and mnt_clear_expiry() are not needed outside of namespace.c anymore, now that namei has finish_automount() to use. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Diffstat (limited to 'fs/namespace.c')
-rw-r--r--fs/namespace.c46
1 files changed, 18 insertions, 28 deletions
diff --git a/fs/namespace.c b/fs/namespace.c
index 31aefc8e5fa6..7b0b95371696 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1872,6 +1872,8 @@ out:
return err;
}
+static int do_add_mount(struct vfsmount *, struct path *, int);
+
/*
* create a new mount for userspace and request it to be added into the
* namespace's tree
@@ -1909,25 +1911,31 @@ int finish_automount(struct vfsmount *m, struct path *path)
if (m->mnt_sb == path->mnt->mnt_sb &&
m->mnt_root == path->dentry) {
- mnt_clear_expiry(m);
- mntput(m);
- mntput(m);
- return -ELOOP;
+ err = -ELOOP;
+ goto fail;
}
err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
- if (err) {
- mnt_clear_expiry(m);
- mntput(m);
- mntput(m);
+ if (!err)
+ return 0;
+fail:
+ /* remove m from any expiration list it may be on */
+ if (!list_empty(&m->mnt_expire)) {
+ down_write(&namespace_sem);
+ br_write_lock(vfsmount_lock);
+ list_del_init(&m->mnt_expire);
+ br_write_unlock(vfsmount_lock);
+ up_write(&namespace_sem);
}
+ mntput(m);
+ mntput(m);
return err;
}
/*
* add a mount into a namespace's mount tree
*/
-int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
+static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
{
int err;
@@ -1954,11 +1962,7 @@ int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
goto unlock;
newmnt->mnt_flags = mnt_flags;
- if ((err = graft_tree(newmnt, path)))
- goto unlock;
-
- up_write(&namespace_sem);
- return 0;
+ err = graft_tree(newmnt, path);
unlock:
up_write(&namespace_sem);
@@ -1983,20 +1987,6 @@ void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
EXPORT_SYMBOL(mnt_set_expiry);
/*
- * Remove a vfsmount from any expiration list it may be on
- */
-void mnt_clear_expiry(struct vfsmount *mnt)
-{
- if (!list_empty(&mnt->mnt_expire)) {
- down_write(&namespace_sem);
- br_write_lock(vfsmount_lock);
- list_del_init(&mnt->mnt_expire);
- br_write_unlock(vfsmount_lock);
- up_write(&namespace_sem);
- }
-}
-
-/*
* process a list of expirable mountpoints with the intent of discarding any
* mountpoints that aren't in use and haven't been touched since last we came
* here