much milder d_walk() race

d_walk() relies upon the tree not getting rearranged under it without
rename_lock being touched.  And we do grab rename_lock around the
places that change the tree topology.  Unfortunately, branch reordering
is just as bad from d_walk() POV and we have two places that do it
without touching rename_lock - one in handling of cursors (for ramfs-style
directories) and another in autofs.  autofs one is a separate story; this
commit deals with the cursors.
	* mark cursor dentries explicitly at allocation time
	* make __dentry_kill() leave ->d_child.next pointing to the next
non-cursor sibling, making sure that it won't be moved around unnoticed
before the parent is relocked on ascend-to-parent path in d_walk().
	* make d_walk() skip cursors explicitly; strictly speaking it's
not necessary (all callbacks we pass to d_walk() are no-ops on cursors),
but it makes analysis easier.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

1 files changed, 52 insertions, 6 deletions

diff --git a/fs/dcache.c b/fs/dcache.c
index 817c243..b7eddfd 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -507,6 +507,44 @@ void d_drop(struct dentry *dentry)
 }
 EXPORT_SYMBOL(d_drop);
 
+static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
+{
+	struct dentry *next;
+	/*
+	 * Inform d_walk() and shrink_dentry_list() that we are no longer
+	 * attached to the dentry tree
+	 */
+	dentry->d_flags |= DCACHE_DENTRY_KILLED;
+	if (unlikely(list_empty(&dentry->d_child)))
+		return;
+	__list_del_entry(&dentry->d_child);
+	/*
+	 * Cursors can move around the list of children.  While we'd been
+	 * a normal list member, it didn't matter - ->d_child.next would've
+	 * been updated.  However, from now on it won't be and for the
+	 * things like d_walk() it might end up with a nasty surprise.
+	 * Normally d_walk() doesn't care about cursors moving around -
+	 * ->d_lock on parent prevents that and since a cursor has no children
+	 * of its own, we get through it without ever unlocking the parent.
+	 * There is one exception, though - if we ascend from a child that
+	 * gets killed as soon as we unlock it, the next sibling is found
+	 * using the value left in its ->d_child.next.  And if _that_
+	 * pointed to a cursor, and cursor got moved (e.g. by lseek())
+	 * before d_walk() regains parent->d_lock, we'll end up skipping
+	 * everything the cursor had been moved past.
+	 *
+	 * Solution: make sure that the pointer left behind in ->d_child.next
+	 * points to something that won't be moving around.  I.e. skip the
+	 * cursors.
+	 */
+	while (dentry->d_child.next != &parent->d_subdirs) {
+		next = list_entry(dentry->d_child.next, struct dentry, d_child);
+		if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
+			break;
+		dentry->d_child.next = next->d_child.next;
+	}
+}
+
 static void __dentry_kill(struct dentry *dentry)
 {
 	struct dentry *parent = NULL;
@@ -532,12 +570,7 @@ static void __dentry_kill(struct dentry *dentry)
 	}
 	/* if it was on the hash then remove it */
 	__d_drop(dentry);
-	__list_del_entry(&dentry->d_child);
-	/*
-	 * Inform d_walk() that we are no longer attached to the
-	 * dentry tree
-	 */
-	dentry->d_flags |= DCACHE_DENTRY_KILLED;
+	dentry_unlist(dentry, parent);
 	if (parent)
 		spin_unlock(&parent->d_lock);
 	dentry_iput(dentry);
@@ -1203,6 +1236,9 @@ resume:
 		struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
 		next = tmp->next;
 
+		if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
+			continue;
+
 		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 
 		ret = enter(data, dentry);
@@ -1651,6 +1687,16 @@ struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
 }
 EXPORT_SYMBOL(d_alloc);
 
+struct dentry *d_alloc_cursor(struct dentry * parent)
+{
+	struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
+	if (dentry) {
+		dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
+		dentry->d_parent = dget(parent);
+	}
+	return dentry;
+}
+
 /**
  * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
  * @sb: the superblock