[PATCH] shared mounts handling: umount

An unmount of a mount creates a umount event on the parent. If the parent is a shared mount, it gets propagated to all mounts in the peer group. Signed-off-by: Ram Pai <linuxram@us.ibm.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] shared mounts handling: umount
An unmount of a mount creates a umount event on the parent. If the parent is a shared mount, it gets propagated to all mounts in the peer group. Signed-off-by: Ram Pai <linuxram@us.ibm.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Ram Pai · Linus Torvalds
1 parent 2144440327
Showing 5 changed files with 128 additions and 20 deletions Side-by-side Diff
fs/namespace.c
fs/pnode.c
fs/pnode.h
include/linux/dcache.h
include/linux/fs.h
@@ -86,31 +86,44 @@
 }
  
 /*
- * Now, lookup_mnt increments the ref count before returning
- * the vfsmount struct.
+ * find the first or last mount at @dentry on vfsmount @mnt depending on
+ * @dir. If @dir is set return the first mount else return the last mount.
  */
-struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
+struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
+			      int dir)
 {
 	struct list_head *head = mount_hashtable + hash(mnt, dentry);
 	struct list_head *tmp = head;
 	struct vfsmount *p, *found = NULL;
  
-	spin_lock(&vfsmount_lock);
 	for (;;) {
-		tmp = tmp->next;
+		tmp = dir ? tmp->next : tmp->prev;
 		p = NULL;
 		if (tmp == head)
 			break;
 		p = list_entry(tmp, struct vfsmount, mnt_hash);
 		if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
-			found = mntget(p);
+			found = p;
 			break;
 		}
 	}
-	spin_unlock(&vfsmount_lock);
 	return found;
 }
  
+/*
+ * lookup_mnt increments the ref count before returning
+ * the vfsmount struct.
+ */
+struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
+{
+	struct vfsmount *child_mnt;
+	spin_lock(&vfsmount_lock);
+	if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
+		mntget(child_mnt);
+	spin_unlock(&vfsmount_lock);
+	return child_mnt;
+}
+
 static inline int check_mnt(struct vfsmount *mnt)
 {
 	return mnt->mnt_namespace == current->namespace;
@@ -404,9 +417,12 @@
  */
 int may_umount(struct vfsmount *mnt)
 {
-	if (atomic_read(&mnt->mnt_count) > 2)
-		return -EBUSY;
-	return 0;
+	int ret = 0;
+	spin_lock(&vfsmount_lock);
+	if (propagate_mount_busy(mnt, 2))
+		ret = -EBUSY;
+	spin_unlock(&vfsmount_lock);
+	return ret;
 }
  
 EXPORT_SYMBOL(may_umount);
@@ -433,7 +449,7 @@
 	}
 }
  
-void umount_tree(struct vfsmount *mnt, struct list_head *kill)
+void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
 {
 	struct vfsmount *p;
  
@@ -442,6 +458,9 @@
 		list_add(&p->mnt_hash, kill);
 	}
  
+	if (propagate)
+		propagate_umount(kill);
+
 	list_for_each_entry(p, kill, mnt_hash) {
 		list_del_init(&p->mnt_expire);
 		list_del_init(&p->mnt_list);
@@ -450,6 +469,7 @@
 		list_del_init(&p->mnt_child);
 		if (p->mnt_parent != p)
 			mnt->mnt_mountpoint->d_mounted--;
+		change_mnt_propagation(p, MS_PRIVATE);
 	}
 }
  
  
@@ -526,9 +546,9 @@
 	event++;
  
 	retval = -EBUSY;
-	if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
+	if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
 		if (!list_empty(&mnt->mnt_list))
-			umount_tree(mnt, &umount_list);
+			umount_tree(mnt, 1, &umount_list);
 		retval = 0;
 	}
 	spin_unlock(&vfsmount_lock);
@@ -651,7 +671,7 @@
 	if (res) {
 		LIST_HEAD(umount_list);
 		spin_lock(&vfsmount_lock);
-		umount_tree(res, &umount_list);
+		umount_tree(res, 0, &umount_list);
 		spin_unlock(&vfsmount_lock);
 		release_mounts(&umount_list);
 	}
@@ -827,7 +847,7 @@
 	if (err) {
 		LIST_HEAD(umount_list);
 		spin_lock(&vfsmount_lock);
-		umount_tree(mnt, &umount_list);
+		umount_tree(mnt, 0, &umount_list);
 		spin_unlock(&vfsmount_lock);
 		release_mounts(&umount_list);
 	}
  
@@ -1023,12 +1043,12 @@
 	 * Check that it is still dead: the count should now be 2 - as
 	 * contributed by the vfsmount parent and the mntget above
 	 */
-	if (atomic_read(&mnt->mnt_count) == 2) {
+	if (!propagate_mount_busy(mnt, 2)) {
 		/* delete from the namespace */
 		touch_namespace(mnt->mnt_namespace);
 		list_del_init(&mnt->mnt_list);
 		mnt->mnt_namespace = NULL;
-		umount_tree(mnt, umounts);
+		umount_tree(mnt, 1, umounts);
 		spin_unlock(&vfsmount_lock);
 	} else {
 		/*
@@ -1647,7 +1667,7 @@
 	spin_unlock(&vfsmount_lock);
 	down_write(&namespace_sem);
 	spin_lock(&vfsmount_lock);
-	umount_tree(root, &umount_list);
+	umount_tree(root, 0, &umount_list);
 	spin_unlock(&vfsmount_lock);
 	up_write(&namespace_sem);
 	release_mounts(&umount_list);
@@ -99,10 +99,95 @@
 	while (!list_empty(&tmp_list)) {
 		child = list_entry(tmp_list.next, struct vfsmount, mnt_hash);
 		list_del_init(&child->mnt_hash);
-		umount_tree(child, &umount_list);
+		umount_tree(child, 0, &umount_list);
 	}
 	spin_unlock(&vfsmount_lock);
 	release_mounts(&umount_list);
 	return ret;
+}
+
+/*
+ * return true if the refcount is greater than count
+ */
+static inline int do_refcount_check(struct vfsmount *mnt, int count)
+{
+	int mycount = atomic_read(&mnt->mnt_count);
+	return (mycount > count);
+}
+
+/*
+ * check if the mount 'mnt' can be unmounted successfully.
+ * @mnt: the mount to be checked for unmount
+ * NOTE: unmounting 'mnt' would naturally propagate to all
+ * other mounts its parent propagates to.
+ * Check if any of these mounts that **do not have submounts**
+ * have more references than 'refcnt'. If so return busy.
+ */
+int propagate_mount_busy(struct vfsmount *mnt, int refcnt)
+{
+	struct vfsmount *m, *child;
+	struct vfsmount *parent = mnt->mnt_parent;
+	int ret = 0;
+
+	if (mnt == parent)
+		return do_refcount_check(mnt, refcnt);
+
+	/*
+	 * quickly check if the current mount can be unmounted.
+	 * If not, we don't have to go checking for all other
+	 * mounts
+	 */
+	if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
+		return 1;
+
+	for (m = propagation_next(parent, parent); m;
+	     		m = propagation_next(m, parent)) {
+		child = __lookup_mnt(m, mnt->mnt_mountpoint, 0);
+		if (child && list_empty(&child->mnt_mounts) &&
+		    (ret = do_refcount_check(child, 1)))
+			break;
+	}
+	return ret;
+}
+
+/*
+ * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
+ * parent propagates to.
+ */
+static void __propagate_umount(struct vfsmount *mnt)
+{
+	struct vfsmount *parent = mnt->mnt_parent;
+	struct vfsmount *m;
+
+	BUG_ON(parent == mnt);
+
+	for (m = propagation_next(parent, parent); m;
+			m = propagation_next(m, parent)) {
+
+		struct vfsmount *child = __lookup_mnt(m,
+					mnt->mnt_mountpoint, 0);
+		/*
+		 * umount the child only if the child has no
+		 * other children
+		 */
+		if (child && list_empty(&child->mnt_mounts)) {
+			list_del(&child->mnt_hash);
+			list_add_tail(&child->mnt_hash, &mnt->mnt_hash);
+		}
+	}
+}
+
+/*
+ * collect all mounts that receive propagation from the mount in @list,
+ * and return these additional mounts in the same list.
+ * @list: the list of mounts to be unmounted.
+ */
+int propagate_umount(struct list_head *list)
+{
+	struct vfsmount *mnt;
+
+	list_for_each_entry(mnt, list, mnt_hash)
+		__propagate_umount(mnt);
+	return 0;
 }
@@ -29,5 +29,7 @@
 void change_mnt_propagation(struct vfsmount *, int);
 int propagate_mnt(struct vfsmount *, struct dentry *, struct vfsmount *,
 		struct list_head *);
+int propagate_umount(struct list_head *);
+int propagate_mount_busy(struct vfsmount *, int);
 #endif /* _LINUX_PNODE_H */
@@ -329,6 +329,7 @@
 }
  
 extern struct vfsmount *lookup_mnt(struct vfsmount *, struct dentry *);
+extern struct vfsmount *__lookup_mnt(struct vfsmount *, struct dentry *, int);
 extern struct dentry *lookup_create(struct nameidata *nd, int is_dir);
  
 extern int sysctl_vfs_cache_pressure;
@@ -1251,7 +1251,7 @@
 extern struct vfsmount *kern_mount(struct file_system_type *);
 extern int may_umount_tree(struct vfsmount *);
 extern int may_umount(struct vfsmount *);
-extern void umount_tree(struct vfsmount *, struct list_head *);
+extern void umount_tree(struct vfsmount *, int, struct list_head *);
 extern void release_mounts(struct list_head *);
 extern long do_mount(char *, char *, char *, unsigned long, void *);
 extern struct vfsmount *copy_tree(struct vfsmount *, struct dentry *, int);
...	...	@@ -86,31 +86,44 @@
86	86	}
87	87
88	88	/*
89		- * Now, lookup_mnt increments the ref count before returning
90		- * the vfsmount struct.
	89	+ * find the first or last mount at @dentry on vfsmount @mnt depending on
	90	+ * @dir. If @dir is set return the first mount else return the last mount.
91	91	*/
92		-struct vfsmount lookup_mnt(struct vfsmount mnt, struct dentry *dentry)
	92	+struct vfsmount __lookup_mnt(struct vfsmount mnt, struct dentry *dentry,
	93	+ int dir)
93	94	{
94	95	struct list_head *head = mount_hashtable + hash(mnt, dentry);
95	96	struct list_head *tmp = head;
96	97	struct vfsmount p, found = NULL;
97	98
98		- spin_lock(&vfsmount_lock);
99	99	for (;;) {
100		- tmp = tmp->next;
	100	+ tmp = dir ? tmp->next : tmp->prev;
101	101	p = NULL;
102	102	if (tmp == head)
103	103	break;
104	104	p = list_entry(tmp, struct vfsmount, mnt_hash);
105	105	if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
106		- found = mntget(p);
	106	+ found = p;
107	107	break;
108	108	}
109	109	}
110		- spin_unlock(&vfsmount_lock);
111	110	return found;
112	111	}
113	112
	113	+/*
	114	+ * lookup_mnt increments the ref count before returning
	115	+ * the vfsmount struct.
	116	+ */
	117	+struct vfsmount lookup_mnt(struct vfsmount mnt, struct dentry *dentry)
	118	+{
	119	+ struct vfsmount *child_mnt;
	120	+ spin_lock(&vfsmount_lock);
	121	+ if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
	122	+ mntget(child_mnt);
	123	+ spin_unlock(&vfsmount_lock);
	124	+ return child_mnt;
	125	+}
	126	+
114	127	static inline int check_mnt(struct vfsmount *mnt)
115	128	{
116	129	return mnt->mnt_namespace == current->namespace;
...	...	@@ -404,9 +417,12 @@
404	417	*/
405	418	int may_umount(struct vfsmount *mnt)
406	419	{
407		- if (atomic_read(&mnt->mnt_count) > 2)
408		- return -EBUSY;
409		- return 0;
	420	+ int ret = 0;
	421	+ spin_lock(&vfsmount_lock);
	422	+ if (propagate_mount_busy(mnt, 2))
	423	+ ret = -EBUSY;
	424	+ spin_unlock(&vfsmount_lock);
	425	+ return ret;
410	426	}
411	427
412	428	EXPORT_SYMBOL(may_umount);
...	...	@@ -433,7 +449,7 @@
433	449	}
434	450	}
435	451
436		-void umount_tree(struct vfsmount mnt, struct list_head kill)
	452	+void umount_tree(struct vfsmount mnt, int propagate, struct list_head kill)
437	453	{
438	454	struct vfsmount *p;
439	455
...	...	@@ -442,6 +458,9 @@
442	458	list_add(&p->mnt_hash, kill);
443	459	}
444	460
	461	+ if (propagate)
	462	+ propagate_umount(kill);
	463	+
445	464	list_for_each_entry(p, kill, mnt_hash) {
446	465	list_del_init(&p->mnt_expire);
447	466	list_del_init(&p->mnt_list);
...	...	@@ -450,6 +469,7 @@
450	469	list_del_init(&p->mnt_child);
451	470	if (p->mnt_parent != p)
452	471	mnt->mnt_mountpoint->d_mounted--;
	472	+ change_mnt_propagation(p, MS_PRIVATE);
453	473	}
454	474	}
455	475
456	476
...	...	@@ -526,9 +546,9 @@
526	546	event++;
527	547
528	548	retval = -EBUSY;
529		- if (atomic_read(&mnt->mnt_count) == 2 \|\| flags & MNT_DETACH) {
	549	+ if (flags & MNT_DETACH \|\| !propagate_mount_busy(mnt, 2)) {
530	550	if (!list_empty(&mnt->mnt_list))
531		- umount_tree(mnt, &umount_list);
	551	+ umount_tree(mnt, 1, &umount_list);
532	552	retval = 0;
533	553	}
534	554	spin_unlock(&vfsmount_lock);
...	...	@@ -651,7 +671,7 @@
651	671	if (res) {
652	672	LIST_HEAD(umount_list);
653	673	spin_lock(&vfsmount_lock);
654		- umount_tree(res, &umount_list);
	674	+ umount_tree(res, 0, &umount_list);
655	675	spin_unlock(&vfsmount_lock);
656	676	release_mounts(&umount_list);
657	677	}
...	...	@@ -827,7 +847,7 @@
827	847	if (err) {
828	848	LIST_HEAD(umount_list);
829	849	spin_lock(&vfsmount_lock);
830		- umount_tree(mnt, &umount_list);
	850	+ umount_tree(mnt, 0, &umount_list);
831	851	spin_unlock(&vfsmount_lock);
832	852	release_mounts(&umount_list);
833	853	}
834	854
...	...	@@ -1023,12 +1043,12 @@
1023	1043	* Check that it is still dead: the count should now be 2 - as
1024	1044	* contributed by the vfsmount parent and the mntget above
1025	1045	*/
1026		- if (atomic_read(&mnt->mnt_count) == 2) {
	1046	+ if (!propagate_mount_busy(mnt, 2)) {
1027	1047	/* delete from the namespace */
1028	1048	touch_namespace(mnt->mnt_namespace);
1029	1049	list_del_init(&mnt->mnt_list);
1030	1050	mnt->mnt_namespace = NULL;
1031		- umount_tree(mnt, umounts);
	1051	+ umount_tree(mnt, 1, umounts);
1032	1052	spin_unlock(&vfsmount_lock);
1033	1053	} else {
1034	1054	/*
...	...	@@ -1647,7 +1667,7 @@
1647	1667	spin_unlock(&vfsmount_lock);
1648	1668	down_write(&namespace_sem);
1649	1669	spin_lock(&vfsmount_lock);
1650		- umount_tree(root, &umount_list);
	1670	+ umount_tree(root, 0, &umount_list);
1651	1671	spin_unlock(&vfsmount_lock);
1652	1672	up_write(&namespace_sem);
1653	1673	release_mounts(&umount_list);
...	...	@@ -99,10 +99,95 @@
99	99	while (!list_empty(&tmp_list)) {
100	100	child = list_entry(tmp_list.next, struct vfsmount, mnt_hash);
101	101	list_del_init(&child->mnt_hash);
102		- umount_tree(child, &umount_list);
	102	+ umount_tree(child, 0, &umount_list);
103	103	}
104	104	spin_unlock(&vfsmount_lock);
105	105	release_mounts(&umount_list);
106	106	return ret;
	107	+}
	108	+
	109	+/*
	110	+ * return true if the refcount is greater than count
	111	+ */
	112	+static inline int do_refcount_check(struct vfsmount *mnt, int count)
	113	+{
	114	+ int mycount = atomic_read(&mnt->mnt_count);
	115	+ return (mycount > count);
	116	+}
	117	+
	118	+/*
	119	+ * check if the mount 'mnt' can be unmounted successfully.
	120	+ * @mnt: the mount to be checked for unmount
	121	+ * NOTE: unmounting 'mnt' would naturally propagate to all
	122	+ * other mounts its parent propagates to.
	123	+ * Check if any of these mounts that do not have submounts
	124	+ * have more references than 'refcnt'. If so return busy.
	125	+ */
	126	+int propagate_mount_busy(struct vfsmount *mnt, int refcnt)
	127	+{
	128	+ struct vfsmount m, child;
	129	+ struct vfsmount *parent = mnt->mnt_parent;
	130	+ int ret = 0;
	131	+
	132	+ if (mnt == parent)
	133	+ return do_refcount_check(mnt, refcnt);
	134	+
	135	+ /*
	136	+ * quickly check if the current mount can be unmounted.
	137	+ * If not, we don't have to go checking for all other
	138	+ * mounts
	139	+ */
	140	+ if (!list_empty(&mnt->mnt_mounts) \|\| do_refcount_check(mnt, refcnt))
	141	+ return 1;
	142	+
	143	+ for (m = propagation_next(parent, parent); m;
	144	+ m = propagation_next(m, parent)) {
	145	+ child = __lookup_mnt(m, mnt->mnt_mountpoint, 0);
	146	+ if (child && list_empty(&child->mnt_mounts) &&
	147	+ (ret = do_refcount_check(child, 1)))
	148	+ break;
	149	+ }
	150	+ return ret;
	151	+}
	152	+
	153	+/*
	154	+ * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
	155	+ * parent propagates to.
	156	+ */
	157	+static void __propagate_umount(struct vfsmount *mnt)
	158	+{
	159	+ struct vfsmount *parent = mnt->mnt_parent;
	160	+ struct vfsmount *m;
	161	+
	162	+ BUG_ON(parent == mnt);
	163	+
	164	+ for (m = propagation_next(parent, parent); m;
	165	+ m = propagation_next(m, parent)) {
	166	+
	167	+ struct vfsmount *child = __lookup_mnt(m,
	168	+ mnt->mnt_mountpoint, 0);
	169	+ /*
	170	+ * umount the child only if the child has no
	171	+ * other children
	172	+ */
	173	+ if (child && list_empty(&child->mnt_mounts)) {
	174	+ list_del(&child->mnt_hash);
	175	+ list_add_tail(&child->mnt_hash, &mnt->mnt_hash);
	176	+ }
	177	+ }
	178	+}
	179	+
	180	+/*
	181	+ * collect all mounts that receive propagation from the mount in @list,
	182	+ * and return these additional mounts in the same list.
	183	+ * @list: the list of mounts to be unmounted.
	184	+ */
	185	+int propagate_umount(struct list_head *list)
	186	+{
	187	+ struct vfsmount *mnt;
	188	+
	189	+ list_for_each_entry(mnt, list, mnt_hash)
	190	+ __propagate_umount(mnt);
	191	+ return 0;
107	192	}
...	...	@@ -29,5 +29,7 @@
29	29	void change_mnt_propagation(struct vfsmount *, int);
30	30	int propagate_mnt(struct vfsmount , struct dentry , struct vfsmount *,
31	31	struct list_head *);
	32	+int propagate_umount(struct list_head *);
	33	+int propagate_mount_busy(struct vfsmount *, int);
32	34	#endif /* _LINUX_PNODE_H */
...	...	@@ -329,6 +329,7 @@
329	329	}
330	330
331	331	extern struct vfsmount lookup_mnt(struct vfsmount , struct dentry *);
	332	+extern struct vfsmount __lookup_mnt(struct vfsmount , struct dentry *, int);
332	333	extern struct dentry lookup_create(struct nameidata nd, int is_dir);
333	334
334	335	extern int sysctl_vfs_cache_pressure;
...	...	@@ -1251,7 +1251,7 @@
1251	1251	extern struct vfsmount kern_mount(struct file_system_type );
1252	1252	extern int may_umount_tree(struct vfsmount *);
1253	1253	extern int may_umount(struct vfsmount *);
1254		-extern void umount_tree(struct vfsmount , struct list_head );
	1254	+extern void umount_tree(struct vfsmount , int, struct list_head );
1255	1255	extern void release_mounts(struct list_head *);
1256	1256	extern long do_mount(char , char , char , unsigned long, void );
1257	1257	extern struct vfsmount copy_tree(struct vfsmount , struct dentry *, int);