Merge branch 'for-linus' of git://git.infradead.org/users/eparis/notify

* 'for-linus' of git://git.infradead.org/users/eparis/notify:
  inotify: Ensure we alwasy write the terminating NULL.
  inotify: fix locking around inotify watching in the idr
  inotify: do not BUG on idr entries at inotify destruction
  inotify: seperate new watch creation updating existing watches

fs/notify/inotify/inotify_fsnotify.c

@@ -105,16 +105,45 @@ static bool inotify_should_send_event(struct fsnotify_group *group, struct inode
 	return send;
 }
 
+/*
+ * This is NEVER supposed to be called.  Inotify marks should either have been
+ * removed from the idr when the watch was removed or in the
+ * fsnotify_destroy_mark_by_group() call when the inotify instance was being
+ * torn down.  This is only called if the idr is about to be freed but there
+ * are still marks in it.
+ */
 static int idr_callback(int id, void *p, void *data)
 {
-	BUG();
+	struct fsnotify_mark_entry *entry;
+	struct inotify_inode_mark_entry *ientry;
+	static bool warned = false;
+
+	if (warned)
+		return 0;
+
+	warned = false;
+	entry = p;
+	ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
+
+	WARN(1, "inotify closing but id=%d for entry=%p in group=%p still in "
+		"idr.  Probably leaking memory\n", id, p, data);
+
+	/*
+	 * I'm taking the liberty of assuming that the mark in question is a
+	 * valid address and I'm dereferencing it.  This might help to figure
+	 * out why we got here and the panic is no worse than the original
+	 * BUG() that was here.
+	 */
+	if (entry)
+		printk(KERN_WARNING "entry->group=%p inode=%p wd=%d\n",
+			entry->group, entry->inode, ientry->wd);
 	return 0;
 }
 
 static void inotify_free_group_priv(struct fsnotify_group *group)
 {
 	/* ideally the idr is empty and we won't hit the BUG in teh callback */
-	idr_for_each(&group->inotify_data.idr, idr_callback, NULL);
+	idr_for_each(&group->inotify_data.idr, idr_callback, group);
 	idr_remove_all(&group->inotify_data.idr);
 	idr_destroy(&group->inotify_data.idr);
 }

fs/notify/inotify/inotify_user.c

@@ -47,9 +47,6 @@
 
 static struct vfsmount *inotify_mnt __read_mostly;
 
-/* this just sits here and wastes global memory.  used to just pad userspace messages with zeros */
-static struct inotify_event nul_inotify_event;
-
 /* these are configurable via /proc/sys/fs/inotify/ */
 static int inotify_max_user_instances __read_mostly;
 static int inotify_max_queued_events __read_mostly;
@@ -199,8 +196,10 @@ static ssize_t copy_event_to_user(struct fsnotify_group *group,
 		inotify_free_event_priv(fsn_priv);
 	}
 
-	/* round up event->name_len so it is a multiple of event_size */
-	name_len = roundup(event->name_len, event_size);
+	/* round up event->name_len so it is a multiple of event_size
+	 * plus an extra byte for the terminating '\0'.
+	 */
+	name_len = roundup(event->name_len + 1, event_size);
 	inotify_event.len = name_len;
 
 	inotify_event.mask = inotify_mask_to_arg(event->mask);
@@ -224,8 +223,8 @@ static ssize_t copy_event_to_user(struct fsnotify_group *group,
 			return -EFAULT;
 		buf += event->name_len;
 
-		/* fill userspace with 0's from nul_inotify_event */
-		if (copy_to_user(buf, &nul_inotify_event, len_to_zero))
+		/* fill userspace with 0's */
+		if (clear_user(buf, len_to_zero))
 			return -EFAULT;
 		buf += len_to_zero;
 		event_size += name_len;
@@ -364,20 +363,53 @@ static int inotify_find_inode(const char __user *dirname, struct path *path, uns
 	return error;
 }
 
+/*
+ * Remove the mark from the idr (if present) and drop the reference
+ * on the mark because it was in the idr.
+ */
 static void inotify_remove_from_idr(struct fsnotify_group *group,
 				    struct inotify_inode_mark_entry *ientry)
 {
 	struct idr *idr;
+	struct fsnotify_mark_entry *entry;
+	struct inotify_inode_mark_entry *found_ientry;
+	int wd;
 
 	spin_lock(&group->inotify_data.idr_lock);
 	idr = &group->inotify_data.idr;
-	idr_remove(idr, ientry->wd);
-	spin_unlock(&group->inotify_data.idr_lock);
+	wd = ientry->wd;
+
+	if (wd == -1)
+		goto out;
+
+	entry = idr_find(&group->inotify_data.idr, wd);
+	if (unlikely(!entry))
+		goto out;
+
+	found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
+	if (unlikely(found_ientry != ientry)) {
+		/* We found an entry in the idr with the right wd, but it's
+		 * not the entry we were told to remove.  eparis seriously
+		 * fucked up somewhere. */
+		WARN_ON(1);
 		ientry->wd = -1;
+		goto out;
+	}
+
+	/* One ref for being in the idr, one ref held by the caller */
+	BUG_ON(atomic_read(&entry->refcnt) < 2);
+
+	idr_remove(idr, wd);
+	ientry->wd = -1;
+
+	/* removed from the idr, drop that ref */
+	fsnotify_put_mark(entry);
+out:
+	spin_unlock(&group->inotify_data.idr_lock);
 }
+
 /*
- * Send IN_IGNORED for this wd, remove this wd from the idr, and drop the
- * internal reference help on the mark because it is in the idr.
+ * Send IN_IGNORED for this wd, remove this wd from the idr.
  */
 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 				    struct fsnotify_group *group)
@@ -417,9 +449,6 @@ skip_send_ignore:
 	/* remove this entry from the idr */
 	inotify_remove_from_idr(group, ientry);
 
-	/* removed from idr, drop that reference */
-	fsnotify_put_mark(entry);
-
 	atomic_dec(&group->inotify_data.user->inotify_watches);
 }
 
@@ -431,15 +460,76 @@ static void inotify_free_mark(struct fsnotify_mark_entry *entry)
 	kmem_cache_free(inotify_inode_mark_cachep, ientry);
 }
 
-static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
+static int inotify_update_existing_watch(struct fsnotify_group *group,
+					 struct inode *inode,
+					 u32 arg)
 {
-	struct fsnotify_mark_entry *entry = NULL;
+	struct fsnotify_mark_entry *entry;
 	struct inotify_inode_mark_entry *ientry;
-	struct inotify_inode_mark_entry *tmp_ientry;
-	int ret = 0;
+	__u32 old_mask, new_mask;
+	__u32 mask;
 	int add = (arg & IN_MASK_ADD);
+	int ret;
+
+	/* don't allow invalid bits: we don't want flags set */
+	mask = inotify_arg_to_mask(arg);
+	if (unlikely(!mask))
+		return -EINVAL;
+
+	spin_lock(&inode->i_lock);
+	entry = fsnotify_find_mark_entry(group, inode);
+	spin_unlock(&inode->i_lock);
+	if (!entry)
+		return -ENOENT;
+
+	ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
+
+	spin_lock(&entry->lock);
+
+	old_mask = entry->mask;
+	if (add) {
+		entry->mask |= mask;
+		new_mask = entry->mask;
+	} else {
+		entry->mask = mask;
+		new_mask = entry->mask;
+	}
+
+	spin_unlock(&entry->lock);
+
+	if (old_mask != new_mask) {
+		/* more bits in old than in new? */
+		int dropped = (old_mask & ~new_mask);
+		/* more bits in this entry than the inode's mask? */
+		int do_inode = (new_mask & ~inode->i_fsnotify_mask);
+		/* more bits in this entry than the group? */
+		int do_group = (new_mask & ~group->mask);
+
+		/* update the inode with this new entry */
+		if (dropped || do_inode)
+			fsnotify_recalc_inode_mask(inode);
+
+		/* update the group mask with the new mask */
+		if (dropped || do_group)
+			fsnotify_recalc_group_mask(group);
+	}
+
+	/* return the wd */
+	ret = ientry->wd;
+
+	/* match the get from fsnotify_find_mark_entry() */
+	fsnotify_put_mark(entry);
+
+	return ret;
+}
+
+static int inotify_new_watch(struct fsnotify_group *group,
+			     struct inode *inode,
+			     u32 arg)
+{
+	struct inotify_inode_mark_entry *tmp_ientry;
 	__u32 mask;
-	__u32 old_mask, new_mask;
+	int ret;
 
 	/* don't allow invalid bits: we don't want flags set */
 	mask = inotify_arg_to_mask(arg);
@@ -449,17 +539,11 @@ static int inotify_update_watch(struct fsnotify_group *group, struct inode *inod
 	tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
 	if (unlikely(!tmp_ientry))
 		return -ENOMEM;
-	/* we set the mask at the end after attaching it */
+
 	fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
+	tmp_ientry->fsn_entry.mask = mask;
 	tmp_ientry->wd = -1;
 
-find_entry:
-	spin_lock(&inode->i_lock);
-	entry = fsnotify_find_mark_entry(group, inode);
-	spin_unlock(&inode->i_lock);
-	if (entry) {
-		ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
-	} else {
 	ret = -ENOSPC;
 	if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
 		goto out_err;
@@ -474,80 +558,59 @@ retry:
 				&tmp_ientry->wd);
 	spin_unlock(&group->inotify_data.idr_lock);
 	if (ret) {
+		/* idr was out of memory allocate and try again */
 		if (ret == -EAGAIN)
 			goto retry;
 		goto out_err;
 	}
 
+	/* we put the mark on the idr, take a reference */
+	fsnotify_get_mark(&tmp_ientry->fsn_entry);
+
+	/* we are on the idr, now get on the inode */
 	ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
 	if (ret) {
+		/* we failed to get on the inode, get off the idr */
 		inotify_remove_from_idr(group, tmp_ientry);
-			if (ret == -EEXIST)
-				goto find_entry;
 		goto out_err;
 	}
 
-		/* tmp_ientry has been added to the inode, so we are all set up.
-		 * now we just need to make sure tmp_ientry doesn't get freed and
-		 * we need to set up entry and ientry so the generic code can
-		 * do its thing. */
-		ientry = tmp_ientry;
-		entry = &ientry->fsn_entry;
-		tmp_ientry = NULL;
+	/* update the idr hint, who cares about races, it's just a hint */
+	group->inotify_data.last_wd = tmp_ientry->wd;
 
+	/* increment the number of watches the user has */
 	atomic_inc(&group->inotify_data.user->inotify_watches);
 
-		/* update the idr hint */
-		group->inotify_data.last_wd = ientry->wd;
-
-		/* we put the mark on the idr, take a reference */
-		fsnotify_get_mark(entry);
-	}
-
-	ret = ientry->wd;
-
-	spin_lock(&entry->lock);
-
-	old_mask = entry->mask;
-	if (add) {
-		entry->mask |= mask;
-		new_mask = entry->mask;
-	} else {
-		entry->mask = mask;
-		new_mask = entry->mask;
-	}
-
-	spin_unlock(&entry->lock);
+	/* return the watch descriptor for this new entry */
+	ret = tmp_ientry->wd;
 
-	if (old_mask != new_mask) {
-		/* more bits in old than in new? */
-		int dropped = (old_mask & ~new_mask);
-		/* more bits in this entry than the inode's mask? */
-		int do_inode = (new_mask & ~inode->i_fsnotify_mask);
-		/* more bits in this entry than the group? */
-		int do_group = (new_mask & ~group->mask);
+	/* match the ref from fsnotify_init_markentry() */
+	fsnotify_put_mark(&tmp_ientry->fsn_entry);
 
-		/* update the inode with this new entry */
-		if (dropped || do_inode)
-			fsnotify_recalc_inode_mask(inode);
+out_err:
+	if (ret < 0)
+		kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
 
-		/* update the group mask with the new mask */
-		if (dropped || do_group)
-			fsnotify_recalc_group_mask(group);
-	}
+	return ret;
+}
 
-	/* this either matches fsnotify_find_mark_entry, or init_mark_entry
-	 * depending on which path we took... */
-	fsnotify_put_mark(entry);
+static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
+{
+	int ret = 0;
 
-out_err:
-	/* could be an error, could be that we found an existing mark */
-	if (tmp_ientry) {
-		/* on the idr but didn't make it on the inode */
-		if (tmp_ientry->wd != -1)
-			inotify_remove_from_idr(group, tmp_ientry);
-		kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
-	}
+retry:
+	/* try to update and existing watch with the new arg */
+	ret = inotify_update_existing_watch(group, inode, arg);
+	/* no mark present, try to add a new one */
+	if (ret == -ENOENT)
+		ret = inotify_new_watch(group, inode, arg);
+	/*
+	 * inotify_new_watch could race with another thread which did an
+	 * inotify_new_watch between the update_existing and the add watch
+	 * here, go back and try to update an existing mark again.
+	 */
+	if (ret == -EEXIST)
+		goto retry;
 
 	return ret;
 }