RCU'd vfsmounts

* RCU-delayed freeing of vfsmounts
* vfsmount_lock replaced with a seqlock (mount_lock)
* sequence number from mount_lock is stored in nameidata->m_seq and
used when we exit RCU mode
* new vfsmount flag - MNT_SYNC_UMOUNT. Set by umount_tree() when its
caller knows that vfsmount will have no surviving references.
* synchronize_rcu() done between unlocking namespace_sem in namespace_unlock()
and doing pending mntput().
* new helper: legitimize_mnt(mnt, seq). Checks the mount_lock sequence
number against seq, then grabs reference to mnt. Then it rechecks mount_lock
again to close the race and either returns success or drops the reference it
has acquired. The subtle point is that in case of MNT_SYNC_UMOUNT we can
simply decrement the refcount and sod off - aforementioned synchronize_rcu()
makes sure that final mntput() won't come until we leave RCU mode. We need
that, since we don't want to end up with some lazy pathwalk racing with
umount() and stealing the final mntput() from it - caller of umount() may
expect it to return only once the fs is shut down and we don't want to break
that. In other cases (i.e. with MNT_SYNC_UMOUNT absent) we have to do
full-blown mntput() in case of mount_lock sequence number mismatch happening
just as we'd grabbed the reference, but in those cases we won't be stealing
the final mntput() from anything that would care.
* mntput_no_expire() doesn't lock anything on the fast path now. Incidentally,
SMP and UP cases are handled the same way - no ifdefs there.
* normal pathname resolution does *not* do any writes to mount_lock. It does,
of course, bump the refcounts of vfsmount and dentry in the very end, but that's
it.

Signed-off-by: Al Viro <[e-mail]>

fs/dcache.c

···
 	struct vfsmount *vfsmnt = path->mnt;
 	struct mount *mnt = real_mount(vfsmnt);
 	int error = 0;
-	unsigned seq = 0;
+	unsigned seq, m_seq = 0;
 	char *bptr;
 	int blen;
 
-	br_read_lock(&vfsmount_lock);
 	rcu_read_lock();
+restart_mnt:
+	read_seqbegin_or_lock(&mount_lock, &m_seq);
+	seq = 0;
 restart:
 	bptr = *buffer;
 	blen = *buflen;
+	error = 0;
 	read_seqbegin_or_lock(&rename_lock, &seq);
 	while (dentry != root->dentry || vfsmnt != root->mnt) {
 		struct dentry * parent;
 
 		if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
+			struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
 			/* Global root? */
-			if (mnt_has_parent(mnt)) {
-				dentry = mnt->mnt_mountpoint;
-				mnt = mnt->mnt_parent;
+			if (mnt != parent) {
+				dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
+				mnt = parent;
 				vfsmnt = &mnt->mnt;
 				continue;
 			}
···
 		goto restart;
 	}
 	done_seqretry(&rename_lock, seq);
-	br_read_unlock(&vfsmount_lock);
+	if (need_seqretry(&mount_lock, m_seq)) {
+		m_seq = 1;
+		goto restart_mnt;
+	}
+	done_seqretry(&mount_lock, m_seq);
 
 	if (error >= 0 && bptr == *buffer) {
 		if (--blen < 0)

fs/mount.h

···
 #include <linux/mount.h>
 #include <linux/seq_file.h>
 #include <linux/poll.h>
-#include <linux/lglock.h>
 
 struct mnt_namespace {
 	atomic_t		count;
···
 	struct mount *mnt_parent;
 	struct dentry *mnt_mountpoint;
 	struct vfsmount mnt;
+	struct rcu_head mnt_rcu;
 #ifdef CONFIG_SMP
 	struct mnt_pcp __percpu *mnt_pcp;
 #else
···
 extern struct mount *__lookup_mnt(struct vfsmount *, struct dentry *);
 extern struct mount *__lookup_mnt_last(struct vfsmount *, struct dentry *);
 
+extern bool legitimize_mnt(struct vfsmount *, unsigned);
+
 static inline void get_mnt_ns(struct mnt_namespace *ns)
 {
 	atomic_inc(&ns->count);
 }
 
-extern struct lglock vfsmount_lock;
+extern seqlock_t mount_lock;
 
 static inline void lock_mount_hash(void)
 {
-	br_write_lock(&vfsmount_lock);
+	write_seqlock(&mount_lock);
 }
 
 static inline void unlock_mount_hash(void)
 {
-	br_write_unlock(&vfsmount_lock);
+	write_sequnlock(&mount_lock);
 }
 
 struct proc_mounts {

fs/namei.c

···
 
 static inline void lock_rcu_walk(void)
 {
-	br_read_lock(&vfsmount_lock);
 	rcu_read_lock();
 }
 
 static inline void unlock_rcu_walk(void)
 {
 	rcu_read_unlock();
-	br_read_unlock(&vfsmount_lock);
 }
 
 /**
···
 	BUG_ON(!(nd->flags & LOOKUP_RCU));
 
 	/*
-	 * Get a reference to the parent first: we're
-	 * going to make "path_put(nd->path)" valid in
-	 * non-RCU context for "terminate_walk()".
-	 *
-	 * If this doesn't work, return immediately with
-	 * RCU walking still active (and then we will do
-	 * the RCU walk cleanup in terminate_walk()).
+	 * After legitimizing the bastards, terminate_walk()
+	 * will do the right thing for non-RCU mode, and all our
+	 * subsequent exit cases should rcu_read_unlock()
+	 * before returning.  Do vfsmount first; if dentry
+	 * can't be legitimized, just set nd->path.dentry to NULL
+	 * and rely on dput(NULL) being a no-op.
 	 */
-	if (!lockref_get_not_dead(&parent->d_lockref))
+	if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
 		return -ECHILD;
-
-	/*
-	 * After the mntget(), we terminate_walk() will do
-	 * the right thing for non-RCU mode, and all our
-	 * subsequent exit cases should unlock_rcu_walk()
-	 * before returning.
-	 */
-	mntget(nd->path.mnt);
 	nd->flags &= ~LOOKUP_RCU;
+
+	if (!lockref_get_not_dead(&parent->d_lockref)) {
+		nd->path.dentry = NULL;	
+		unlock_rcu_walk();
+		return -ECHILD;
+	}
 
 	/*
 	 * For a negative lookup, the lookup sequence point is the parents
···
 		if (!(nd->flags & LOOKUP_ROOT))
 			nd->root.mnt = NULL;
 
+		if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
+			unlock_rcu_walk();
+			return -ECHILD;
+		}
 		if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
 			unlock_rcu_walk();
+			mntput(nd->path.mnt);
 			return -ECHILD;
 		}
 		if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
 			unlock_rcu_walk();
 			dput(dentry);
+			mntput(nd->path.mnt);
 			return -ECHILD;
 		}
-		mntget(nd->path.mnt);
 		unlock_rcu_walk();
 	}
 
···
 	struct mount *parent;
 	struct dentry *mountpoint;
 
-	br_read_lock(&vfsmount_lock);
+	read_seqlock_excl(&mount_lock);
 	parent = mnt->mnt_parent;
 	if (parent == mnt) {
-		br_read_unlock(&vfsmount_lock);
+		read_sequnlock_excl(&mount_lock);
 		return 0;
 	}
 	mntget(&parent->mnt);
 	mountpoint = dget(mnt->mnt_mountpoint);
-	br_read_unlock(&vfsmount_lock);
+	read_sequnlock_excl(&mount_lock);
 	dput(path->dentry);
 	path->dentry = mountpoint;
 	mntput(path->mnt);
···
 
 			/* Something is mounted on this dentry in another
 			 * namespace and/or whatever was mounted there in this
-			 * namespace got unmounted before we managed to get the
-			 * vfsmount_lock */
+			 * namespace got unmounted before lookup_mnt() could
+			 * get it */
 		}
 
 		/* Handle an automount point */
···
 		if (flags & LOOKUP_RCU) {
 			lock_rcu_walk();
 			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
+			nd->m_seq = read_seqbegin(&mount_lock);
 		} else {
 			path_get(&nd->path);
 		}
···
 
 	nd->root.mnt = NULL;
 
+	nd->m_seq = read_seqbegin(&mount_lock);
 	if (*name=='/') {
 		if (flags & LOOKUP_RCU) {
 			lock_rcu_walk();

fs/namespace.c

···
  * It should be taken for write in all cases where the vfsmount
  * tree or hash is modified or when a vfsmount structure is modified.
  */
-DEFINE_BRLOCK(vfsmount_lock);
+__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
 
 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
 {
···
 	kmem_cache_free(mnt_cache, mnt);
 }
 
+/* call under rcu_read_lock */
+bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
+{
+	struct mount *mnt;
+	if (read_seqretry(&mount_lock, seq))
+		return false;
+	if (bastard == NULL)
+		return true;
+	mnt = real_mount(bastard);
+	mnt_add_count(mnt, 1);
+	if (likely(!read_seqretry(&mount_lock, seq)))
+		return true;
+	if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
+		mnt_add_count(mnt, -1);
+		return false;
+	}
+	rcu_read_unlock();
+	mntput(bastard);
+	rcu_read_lock();
+	return false;
+}
+
 /*
  * find the first mount at @dentry on vfsmount @mnt.
- * vfsmount_lock must be held for read or write.
+ * call under rcu_read_lock()
  */
 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
 {
 	struct list_head *head = mount_hashtable + hash(mnt, dentry);
 	struct mount *p;
 
-	list_for_each_entry(p, head, mnt_hash)
+	list_for_each_entry_rcu(p, head, mnt_hash)
 		if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
 			return p;
 	return NULL;
···
 
 /*
  * find the last mount at @dentry on vfsmount @mnt.
- * vfsmount_lock must be held for read or write.
+ * mount_lock must be held.
  */
 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
 {
···
 struct vfsmount *lookup_mnt(struct path *path)
 {
 	struct mount *child_mnt;
+	struct vfsmount *m;
+	unsigned seq;
 
-	br_read_lock(&vfsmount_lock);
-	child_mnt = __lookup_mnt(path->mnt, path->dentry);
-	if (child_mnt) {
-		mnt_add_count(child_mnt, 1);
-		br_read_unlock(&vfsmount_lock);
-		return &child_mnt->mnt;
-	} else {
-		br_read_unlock(&vfsmount_lock);
-		return NULL;
-	}
+	rcu_read_lock();
+	do {
+		seq = read_seqbegin(&mount_lock);
+		child_mnt = __lookup_mnt(path->mnt, path->dentry);
+		m = child_mnt ? &child_mnt->mnt : NULL;
+	} while (!legitimize_mnt(m, seq));
+	rcu_read_unlock();
+	return m;
 }
 
 static struct mountpoint *new_mountpoint(struct dentry *dentry)
···
 	return ERR_PTR(err);
 }
 
+static void delayed_free(struct rcu_head *head)
+{
+	struct mount *mnt = container_of(head, struct mount, mnt_rcu);
+	kfree(mnt->mnt_devname);
+#ifdef CONFIG_SMP
+	free_percpu(mnt->mnt_pcp);
+#endif
+	kmem_cache_free(mnt_cache, mnt);
+}
+
 static void mntput_no_expire(struct mount *mnt)
 {
 put_again:
-#ifdef CONFIG_SMP
-	br_read_lock(&vfsmount_lock);
-	if (likely(mnt->mnt_ns)) {
-		/* shouldn't be the last one */
-		mnt_add_count(mnt, -1);
-		br_read_unlock(&vfsmount_lock);
+	rcu_read_lock();
+	mnt_add_count(mnt, -1);
+	if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
+		rcu_read_unlock();
 		return;
 	}
-	br_read_unlock(&vfsmount_lock);
-
 	lock_mount_hash();
-	mnt_add_count(mnt, -1);
 	if (mnt_get_count(mnt)) {
+		rcu_read_unlock();
 		unlock_mount_hash();
 		return;
 	}
-#else
-	mnt_add_count(mnt, -1);
-	if (likely(mnt_get_count(mnt)))
-		return;
-	lock_mount_hash();
-#endif
 	if (unlikely(mnt->mnt_pinned)) {
 		mnt_add_count(mnt, mnt->mnt_pinned + 1);
 		mnt->mnt_pinned = 0;
+		rcu_read_unlock();
 		unlock_mount_hash();
 		acct_auto_close_mnt(&mnt->mnt);
 		goto put_again;
 	}
+	if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
+		rcu_read_unlock();
+		unlock_mount_hash();
+		return;
+	}
+	mnt->mnt.mnt_flags |= MNT_DOOMED;
+	rcu_read_unlock();
 
 	list_del(&mnt->mnt_instance);
 	unlock_mount_hash();
···
 	fsnotify_vfsmount_delete(&mnt->mnt);
 	dput(mnt->mnt.mnt_root);
 	deactivate_super(mnt->mnt.mnt_sb);
-	free_vfsmnt(mnt);
+	mnt_free_id(mnt);
+	call_rcu(&mnt->mnt_rcu, delayed_free);
 }
 
 void mntput(struct vfsmount *mnt)
···
 	list_splice_init(&unmounted, &head);
 	up_write(&namespace_sem);
 
+	synchronize_rcu();
+
 	while (!list_empty(&head)) {
 		mnt = list_first_entry(&head, struct mount, mnt_hash);
 		list_del_init(&mnt->mnt_hash);
···
 }
 
 /*
- * vfsmount lock must be held for write
+ * mount_lock must be held
  * namespace_sem must be held for write
+ * how = 0 => just this tree, don't propagate
+ * how = 1 => propagate; we know that nobody else has reference to any victims
+ * how = 2 => lazy umount
  */
-void umount_tree(struct mount *mnt, int propagate)
+void umount_tree(struct mount *mnt, int how)
 {
 	LIST_HEAD(tmp_list);
 	struct mount *p;
···
 	for (p = mnt; p; p = next_mnt(p, mnt))
 		list_move(&p->mnt_hash, &tmp_list);
 
-	if (propagate)
+	if (how)
 		propagate_umount(&tmp_list);
 
 	list_for_each_entry(p, &tmp_list, mnt_hash) {
···
 		list_del_init(&p->mnt_list);
 		__touch_mnt_namespace(p->mnt_ns);
 		p->mnt_ns = NULL;
+		if (how < 2)
+			p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
 		list_del_init(&p->mnt_child);
 		if (mnt_has_parent(p)) {
 			put_mountpoint(p->mnt_mp);
···
 	lock_mount_hash();
 	event++;
 
-	if (!(flags & MNT_DETACH))
-		shrink_submounts(mnt);
-
-	retval = -EBUSY;
-	if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
+	if (flags & MNT_DETACH) {
 		if (!list_empty(&mnt->mnt_list))
-			umount_tree(mnt, 1);
+			umount_tree(mnt, 2);
 		retval = 0;
+	} else {
+		shrink_submounts(mnt);
+		retval = -EBUSY;
+		if (!propagate_mount_busy(mnt, 2)) {
+			if (!list_empty(&mnt->mnt_list))
+				umount_tree(mnt, 1);
+			retval = 0;
+		}
 	}
 	unlock_mount_hash();
 	namespace_unlock();
···
 	struct mount *parent;
 	int err;
 
-	mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
+	mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL | MNT_DOOMED | MNT_SYNC_UMOUNT);
 
 	mp = lock_mount(path);
 	if (IS_ERR(mp))
···
  * process a list of expirable mountpoints with the intent of discarding any
  * submounts of a specific parent mountpoint
  *
- * vfsmount_lock must be held for write
+ * mount_lock must be held for write
  */
 static void shrink_submounts(struct mount *mnt)
 {
···
 /*
  * Return true if path is reachable from root
  *
- * namespace_sem or vfsmount_lock is held
+ * namespace_sem or mount_lock is held
  */
 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
 			 const struct path *root)
···
 int path_is_under(struct path *path1, struct path *path2)
 {
 	int res;
-	br_read_lock(&vfsmount_lock);
+	read_seqlock_excl(&mount_lock);
 	res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
-	br_read_unlock(&vfsmount_lock);
+	read_sequnlock_excl(&mount_lock);
 	return res;
 }
 EXPORT_SYMBOL(path_is_under);
···
 	for (u = 0; u < HASH_SIZE; u++)
 		INIT_LIST_HEAD(&mountpoint_hashtable[u]);
 
-	br_lock_init(&vfsmount_lock);
-
 	err = sysfs_init();
 	if (err)
 		printk(KERN_WARNING "%s: sysfs_init error: %d\n",
···
 {
 	/* release long term mount so mount point can be released */
 	if (!IS_ERR_OR_NULL(mnt)) {
-		lock_mount_hash();
 		real_mount(mnt)->mnt_ns = NULL;
-		unlock_mount_hash();
+		synchronize_rcu();	/* yecchhh... */
 		mntput(mnt);
 	}
 }

include/linux/mount.h

···
 
 #define MNT_LOCK_READONLY	0x400000
 #define MNT_LOCKED		0x800000
+#define MNT_DOOMED		0x1000000
+#define MNT_SYNC_UMOUNT		0x2000000
 
 struct vfsmount {
 	struct dentry *mnt_root;	/* root of the mounted tree */

include/linux/namei.h

···
 	struct path	root;
 	struct inode	*inode; /* path.dentry.d_inode */
 	unsigned int	flags;
-	unsigned	seq;
+	unsigned	seq, m_seq;
 	int		last_type;
 	unsigned	depth;
 	char *saved_names[MAX_NESTED_LINKS + 1];