Merge branch 'work.dcache' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs

Pull dcache_readdir() fixes from Al Viro:
 "The couple of patches you'd been OK with; no hlist conversion yet, and
  cursors are still in the list of children"

[ Al is referring to future work to avoid some nasty O(n**2) behavior
  with the readdir cursors when you have lots of concurrent readdirs.

  This is just a fix for a race with a trivial cleanup   - Linus ]

* 'work.dcache' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  libfs: take cursors out of list when moving past the end of directory
  Fix the locking in dcache_readdir() and friends

Author: torvalds

fs/libfs.c

@@ -89,58 +89,45 @@ int dcache_dir_close(struct inode *inode, struct file *file)
 EXPORT_SYMBOL(dcache_dir_close);
 
 /* parent is locked at least shared */
-static struct dentry *next_positive(struct dentry *parent,
-				    struct list_head *from,
-				    int count)
+/*
+ * Returns an element of siblings' list.
+ * We are looking for <count>th positive after <p>; if
+ * found, dentry is grabbed and returned to caller.
+ * If no such element exists, NULL is returned.
+ */
+static struct dentry *scan_positives(struct dentry *cursor,
+					struct list_head *p,
+					loff_t count,
+					struct dentry *last)
 {
-	unsigned *seq = &parent->d_inode->i_dir_seq, n;
-	struct dentry *res;
-	struct list_head *p;
-	bool skipped;
-	int i;
+	struct dentry *dentry = cursor->d_parent, *found = NULL;
 
-retry:
-	i = count;
-	skipped = false;
-	n = smp_load_acquire(seq) & ~1;
-	res = NULL;
-	rcu_read_lock();
-	for (p = from->next; p != &parent->d_subdirs; p = p->next) {
+	spin_lock(&dentry->d_lock);
+	while ((p = p->next) != &dentry->d_subdirs) {
 		struct dentry *d = list_entry(p, struct dentry, d_child);
-		if (!simple_positive(d)) {
-			skipped = true;
-		} else if (!--i) {
-			res = d;
-			break;
+		// we must at least skip cursors, to avoid livelocks
+		if (d->d_flags & DCACHE_DENTRY_CURSOR)
+			continue;
+		if (simple_positive(d) && !--count) {
+			spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
+			if (simple_positive(d))
+				found = dget_dlock(d);
+			spin_unlock(&d->d_lock);
+			if (likely(found))
+				break;
+			count = 1;
+		}
+		if (need_resched()) {
+			list_move(&cursor->d_child, p);
+			p = &cursor->d_child;
+			spin_unlock(&dentry->d_lock);
+			cond_resched();
+			spin_lock(&dentry->d_lock);
 		}
 	}
-	rcu_read_unlock();
-	if (skipped) {
-		smp_rmb();
-		if (unlikely(*seq != n))
-			goto retry;
-	}
-	return res;
-}
-
-static void move_cursor(struct dentry *cursor, struct list_head *after)
-{
-	struct dentry *parent = cursor->d_parent;
-	unsigned n, *seq = &parent->d_inode->i_dir_seq;
-	spin_lock(&parent->d_lock);
-	for (;;) {
-		n = *seq;
-		if (!(n & 1) && cmpxchg(seq, n, n + 1) == n)
-			break;
-		cpu_relax();
-	}
-	__list_del(cursor->d_child.prev, cursor->d_child.next);
-	if (after)
-		list_add(&cursor->d_child, after);
-	else
-		list_add_tail(&cursor->d_child, &parent->d_subdirs);
-	smp_store_release(seq, n + 2);
-	spin_unlock(&parent->d_lock);
+	spin_unlock(&dentry->d_lock);
+	dput(last);
+	return found;
 }
 
 loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
@@ -158,17 +145,25 @@ loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
 			return -EINVAL;
 	}
 	if (offset != file->f_pos) {
+		struct dentry *cursor = file->private_data;
+		struct dentry *to = NULL;
+
+		inode_lock_shared(dentry->d_inode);
+
+		if (offset > 2)
+			to = scan_positives(cursor, &dentry->d_subdirs,
+					    offset - 2, NULL);
+		spin_lock(&dentry->d_lock);
+		if (to)
+			list_move(&cursor->d_child, &to->d_child);
+		else
+			list_del_init(&cursor->d_child);
+		spin_unlock(&dentry->d_lock);
+		dput(to);
+
 		file->f_pos = offset;
-		if (file->f_pos >= 2) {
-			struct dentry *cursor = file->private_data;
-			struct dentry *to;
-			loff_t n = file->f_pos - 2;
-
-			inode_lock_shared(dentry->d_inode);
-			to = next_positive(dentry, &dentry->d_subdirs, n);
-			move_cursor(cursor, to ? &to->d_child : NULL);
-			inode_unlock_shared(dentry->d_inode);
-		}
+
+		inode_unlock_shared(dentry->d_inode);
 	}
 	return offset;
 }
@@ -190,25 +185,35 @@ int dcache_readdir(struct file *file, struct dir_context *ctx)
 {
 	struct dentry *dentry = file->f_path.dentry;
 	struct dentry *cursor = file->private_data;
-	struct list_head *p = &cursor->d_child;
-	struct dentry *next;
-	bool moved = false;
+	struct list_head *anchor = &dentry->d_subdirs;
+	struct dentry *next = NULL;
+	struct list_head *p;
 
 	if (!dir_emit_dots(file, ctx))
 		return 0;
 
 	if (ctx->pos == 2)
-		p = &dentry->d_subdirs;
-	while ((next = next_positive(dentry, p, 1)) != NULL) {
+		p = anchor;
+	else if (!list_empty(&cursor->d_child))
+		p = &cursor->d_child;
+	else
+		return 0;
+
+	while ((next = scan_positives(cursor, p, 1, next)) != NULL) {
 		if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
 			      d_inode(next)->i_ino, dt_type(d_inode(next))))
 			break;
-		moved = true;
-		p = &next->d_child;
 		ctx->pos++;
+		p = &next->d_child;
 	}
-	if (moved)
-		move_cursor(cursor, p);
+	spin_lock(&dentry->d_lock);
+	if (next)
+		list_move_tail(&cursor->d_child, &next->d_child);
+	else
+		list_del_init(&cursor->d_child);
+	spin_unlock(&dentry->d_lock);
+	dput(next);
+
 	return 0;
 }
 EXPORT_SYMBOL(dcache_readdir);