fs: Use rename lock and RCU for multi-step operations

The remaining usages for dcache_lock is to allow atomic, multi-step read-side
operations over the directory tree by excluding modifications to the tree.
Also, to walk in the leaf->root direction in the tree where we don't have
a natural d_lock ordering.

This could be accomplished by taking every d_lock, but this would mean a
huge number of locks and actually gets very tricky.

Solve this instead by using the rename seqlock for multi-step read-side
operations, retry in case of a rename so we don't walk up the wrong parent.
Concurrent dentry insertions are not serialised against. Concurrent deletes
are tricky when walking up the directory: our parent might have been deleted
when dropping locks so also need to check and retry for that.

We can also use the rename lock in cases where livelock is a worry (and it
is introduced in subsequent patch).

Signed-off-by: Nick Piggin <[e-mail]>

drivers/staging/pohmelfs/path_entry.c

···
 int pohmelfs_path_length(struct pohmelfs_inode *pi)
 {
 	struct dentry *d, *root, *first;
-	int len = 1; /* Root slash */
+	int len;
+	unsigned seq;
 
-	first = d = d_find_alias(&pi->vfs_inode);
-	if (!d) {
+	first = d_find_alias(&pi->vfs_inode);
+	if (!first) {
 		dprintk("%s: ino: %llu, mode: %o.\n", __func__, pi->ino, pi->vfs_inode.i_mode);
 		return -ENOENT;
 	}
···
 	root = dget(current->fs->root.dentry);
 	spin_unlock(&current->fs->lock);
 
+rename_retry:
+	len = 1; /* Root slash */
+	d = first;
+	seq = read_seqbegin(&rename_lock);
+	rcu_read_lock();
 	spin_lock(&dcache_lock);
 
 	if (!IS_ROOT(d) && d_unhashed(d))
···
 		d = d->d_parent;
 	}
 	spin_unlock(&dcache_lock);
+	rcu_read_unlock();
+	if (read_seqretry(&rename_lock, seq))
+		goto rename_retry;
 
 	dput(root);
 	dput(first);

fs/autofs4/waitq.c

···
 {
 	struct dentry *root = sbi->sb->s_root;
 	struct dentry *tmp;
-	char *buf = *name;
+	char *buf;
 	char *p;
-	int len = 0;
+	int len;
+	unsigned seq;
 
+rename_retry:
+	buf = *name;
+	len = 0;
+	seq = read_seqbegin(&rename_lock);
+	rcu_read_lock();
 	spin_lock(&dcache_lock);
 	for (tmp = dentry ; tmp != root ; tmp = tmp->d_parent)
 		len += tmp->d_name.len + 1;
 
 	if (!len || --len > NAME_MAX) {
 		spin_unlock(&dcache_lock);
+		rcu_read_unlock();
+		if (read_seqretry(&rename_lock, seq))
+			goto rename_retry;
 		return 0;
 	}
 
···
 		strncpy(p, tmp->d_name.name, tmp->d_name.len);
 	}
 	spin_unlock(&dcache_lock);
+	rcu_read_unlock();
+	if (read_seqretry(&rename_lock, seq))
+		goto rename_retry;
 
 	return len;
 }

fs/dcache.c

···
 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
 
+EXPORT_SYMBOL(rename_lock);
 EXPORT_SYMBOL(dcache_inode_lock);
 EXPORT_SYMBOL(dcache_lock);
 
···
 	__releases(dcache_inode_lock)
 	__releases(dcache_lock)
 {
+	dentry->d_parent = NULL;
 	list_del(&dentry->d_u.d_child);
 	if (parent)
 		spin_unlock(&parent->d_lock);
···
  * Return true if the parent or its subdirectories contain
  * a mount point
  */
- 
 int have_submounts(struct dentry *parent)
 {
-	struct dentry *this_parent = parent;
+	struct dentry *this_parent;
 	struct list_head *next;
+	unsigned seq;
+
+rename_retry:
+	this_parent = parent;
+	seq = read_seqbegin(&rename_lock);
 
 	spin_lock(&dcache_lock);
 	if (d_mountpoint(parent))
···
 	 * All done at this level ... ascend and resume the search.
 	 */
 	if (this_parent != parent) {
-		next = this_parent->d_u.d_child.next;
+		struct dentry *tmp;
+		struct dentry *child;
+
+		tmp = this_parent->d_parent;
+		rcu_read_lock();
 		spin_unlock(&this_parent->d_lock);
-		this_parent = this_parent->d_parent;
+		child = this_parent;
+		this_parent = tmp;
 		spin_lock(&this_parent->d_lock);
+		/* might go back up the wrong parent if we have had a rename
+		 * or deletion */
+		if (this_parent != child->d_parent ||
+				read_seqretry(&rename_lock, seq)) {
+			spin_unlock(&this_parent->d_lock);
+			spin_unlock(&dcache_lock);
+			rcu_read_unlock();
+			goto rename_retry;
+		}
+		rcu_read_unlock();
+		next = child->d_u.d_child.next;
 		goto resume;
 	}
 	spin_unlock(&this_parent->d_lock);
 	spin_unlock(&dcache_lock);
+	if (read_seqretry(&rename_lock, seq))
+		goto rename_retry;
 	return 0; /* No mount points found in tree */
 positive:
 	spin_unlock(&dcache_lock);
+	if (read_seqretry(&rename_lock, seq))
+		goto rename_retry;
 	return 1;
 }
 EXPORT_SYMBOL(have_submounts);
···
  */
 static int select_parent(struct dentry * parent)
 {
-	struct dentry *this_parent = parent;
+	struct dentry *this_parent;
 	struct list_head *next;
+	unsigned seq;
 	int found = 0;
+
+rename_retry:
+	this_parent = parent;
+	seq = read_seqbegin(&rename_lock);
 
 	spin_lock(&dcache_lock);
 	spin_lock(&this_parent->d_lock);
···
 		struct list_head *tmp = next;
 		struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
 		next = tmp->next;
-		BUG_ON(this_parent == dentry);
 
 		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 
···
 	 */
 	if (this_parent != parent) {
 		struct dentry *tmp;
-		next = this_parent->d_u.d_child.next;
+		struct dentry *child;
+
 		tmp = this_parent->d_parent;
+		rcu_read_lock();
 		spin_unlock(&this_parent->d_lock);
-		BUG_ON(tmp == this_parent);
+		child = this_parent;
 		this_parent = tmp;
 		spin_lock(&this_parent->d_lock);
+		/* might go back up the wrong parent if we have had a rename
+		 * or deletion */
+		if (this_parent != child->d_parent ||
+				read_seqretry(&rename_lock, seq)) {
+			spin_unlock(&this_parent->d_lock);
+			spin_unlock(&dcache_lock);
+			rcu_read_unlock();
+			goto rename_retry;
+		}
+		rcu_read_unlock();
+		next = child->d_u.d_child.next;
 		goto resume;
 	}
 out:
 	spin_unlock(&this_parent->d_lock);
 	spin_unlock(&dcache_lock);
+	if (read_seqretry(&rename_lock, seq))
+		goto rename_retry;
 	return found;
 }
 
···
 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
 {
 	struct dentry * dentry = NULL;
-	unsigned long seq;
+	unsigned seq;
 
         do {
                 seq = read_seqbegin(&rename_lock);
···
  * @buffer: pointer to the end of the buffer
  * @buflen: pointer to buffer length
  *
- * Caller holds the dcache_lock.
+ * Caller holds the rename_lock.
  *
  * If path is not reachable from the supplied root, then the value of
  * root is changed (without modifying refcounts).
···
 
 	prepend(&res, &buflen, "\0", 1);
 	spin_lock(&dcache_lock);
+	write_seqlock(&rename_lock);
 	error = prepend_path(path, root, &res, &buflen);
+	write_sequnlock(&rename_lock);
 	spin_unlock(&dcache_lock);
 
 	if (error)
···
 
 	get_fs_root(current->fs, &root);
 	spin_lock(&dcache_lock);
+	write_seqlock(&rename_lock);
 	tmp = root;
 	error = path_with_deleted(path, &tmp, &res, &buflen);
 	if (error)
 		res = ERR_PTR(error);
+	write_sequnlock(&rename_lock);
 	spin_unlock(&dcache_lock);
 	path_put(&root);
 	return res;
···
 
 	get_fs_root(current->fs, &root);
 	spin_lock(&dcache_lock);
+	write_seqlock(&rename_lock);
 	tmp = root;
 	error = path_with_deleted(path, &tmp, &res, &buflen);
 	if (!error && !path_equal(&tmp, &root))
 		error = prepend_unreachable(&res, &buflen);
+	write_sequnlock(&rename_lock);
 	spin_unlock(&dcache_lock);
 	path_put(&root);
 	if (error)
···
 	char *retval;
 
 	spin_lock(&dcache_lock);
+	write_seqlock(&rename_lock);
 	retval = __dentry_path(dentry, buf, buflen);
+	write_sequnlock(&rename_lock);
 	spin_unlock(&dcache_lock);
 
 	return retval;
···
 	char *retval;
 
 	spin_lock(&dcache_lock);
+	write_seqlock(&rename_lock);
 	if (d_unlinked(dentry)) {
 		p = buf + buflen;
 		if (prepend(&p, &buflen, "//deleted", 10) != 0)
···
 		buflen++;
 	}
 	retval = __dentry_path(dentry, buf, buflen);
+	write_sequnlock(&rename_lock);
 	spin_unlock(&dcache_lock);
 	if (!IS_ERR(retval) && p)
 		*p = '/';	/* restore '/' overriden with '\0' */
···
 
 	error = -ENOENT;
 	spin_lock(&dcache_lock);
+	write_seqlock(&rename_lock);
 	if (!d_unlinked(pwd.dentry)) {
 		unsigned long len;
 		struct path tmp = root;
···
 
 		prepend(&cwd, &buflen, "\0", 1);
 		error = prepend_path(&pwd, &tmp, &cwd, &buflen);
+		write_sequnlock(&rename_lock);
 		spin_unlock(&dcache_lock);
 
 		if (error)
···
 			if (copy_to_user(buf, cwd, len))
 				error = -EFAULT;
 		}
-	} else
+	} else {
+		write_sequnlock(&rename_lock);
 		spin_unlock(&dcache_lock);
+	}
 
 out:
 	path_put(&pwd);
···
 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
 {
 	int result;
-	unsigned long seq;
+	unsigned seq;
 
 	if (new_dentry == old_dentry)
 		return 1;
 
-	/*
-	 * Need rcu_readlock to protect against the d_parent trashing
-	 * due to d_move
-	 */
-	rcu_read_lock();
 	do {
 		/* for restarting inner loop in case of seq retry */
 		seq = read_seqbegin(&rename_lock);
+		/*
+		 * Need rcu_readlock to protect against the d_parent trashing
+		 * due to d_move
+		 */
+		rcu_read_lock();
 		if (d_ancestor(old_dentry, new_dentry))
 			result = 1;
 		else
 			result = 0;
+		rcu_read_unlock();
 	} while (read_seqretry(&rename_lock, seq));
-	rcu_read_unlock();
 
 	return result;
 }
···
 
 void d_genocide(struct dentry *root)
 {
-	struct dentry *this_parent = root;
+	struct dentry *this_parent;
 	struct list_head *next;
+	unsigned seq;
 
+rename_retry:
+	this_parent = root;
+	seq = read_seqbegin(&rename_lock);
 	spin_lock(&dcache_lock);
 	spin_lock(&this_parent->d_lock);
 repeat:
···
 		struct list_head *tmp = next;
 		struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
 		next = tmp->next;
+
 		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 		if (d_unhashed(dentry) || !dentry->d_inode) {
 			spin_unlock(&dentry->d_lock);
···
 			spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
 			goto repeat;
 		}
-		dentry->d_count--;
+		if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
+			dentry->d_flags |= DCACHE_GENOCIDE;
+			dentry->d_count--;
+		}
 		spin_unlock(&dentry->d_lock);
 	}
 	if (this_parent != root) {
-		next = this_parent->d_u.d_child.next;
-		this_parent->d_count--;
+		struct dentry *tmp;
+		struct dentry *child;
+
+		tmp = this_parent->d_parent;
+		if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
+			this_parent->d_flags |= DCACHE_GENOCIDE;
+			this_parent->d_count--;
+		}
+		rcu_read_lock();
 		spin_unlock(&this_parent->d_lock);
-		this_parent = this_parent->d_parent;
+		child = this_parent;
+		this_parent = tmp;
 		spin_lock(&this_parent->d_lock);
+		/* might go back up the wrong parent if we have had a rename
+		 * or deletion */
+		if (this_parent != child->d_parent ||
+				read_seqretry(&rename_lock, seq)) {
+			spin_unlock(&this_parent->d_lock);
+			spin_unlock(&dcache_lock);
+			rcu_read_unlock();
+			goto rename_retry;
+		}
+		rcu_read_unlock();
+		next = child->d_u.d_child.next;
 		goto resume;
 	}
 	spin_unlock(&this_parent->d_lock);
 	spin_unlock(&dcache_lock);
+	if (read_seqretry(&rename_lock, seq))
+		goto rename_retry;
 }
 
 /**

fs/nfs/namespace.c

···
 	       const struct dentry *dentry,
 	       char *buffer, ssize_t buflen)
 {
-	char *end = buffer+buflen;
+	char *end;
 	int namelen;
+	unsigned seq;
 
+rename_retry:
+	end = buffer+buflen;
 	*--end = '\0';
 	buflen--;
+
+	seq = read_seqbegin(&rename_lock);
+	rcu_read_lock();
 	spin_lock(&dcache_lock);
 	while (!IS_ROOT(dentry) && dentry != droot) {
 		namelen = dentry->d_name.len;
···
 		dentry = dentry->d_parent;
 	}
 	spin_unlock(&dcache_lock);
+	rcu_read_unlock();
+	if (read_seqretry(&rename_lock, seq))
+		goto rename_retry;
 	if (*end != '/') {
 		if (--buflen < 0)
 			goto Elong;
···
 	return end;
 Elong_unlock:
 	spin_unlock(&dcache_lock);
+	rcu_read_unlock();
+	if (read_seqretry(&rename_lock, seq))
+		goto rename_retry;
 Elong:
 	return ERR_PTR(-ENAMETOOLONG);
 }

include/linux/dcache.h

···
 #define DCACHE_FSNOTIFY_PARENT_WATCHED	0x0080 /* Parent inode is watched by some fsnotify listener */
 
 #define DCACHE_CANT_MOUNT	0x0100
+#define DCACHE_GENOCIDE		0x0200
 
 extern spinlock_t dcache_inode_lock;
 extern spinlock_t dcache_lock;