Skip to contentSkip to conversation

notify: unused event private race

PATCH New Eric Paris 1 participant 3 messages Aug 17, 2009

Subsystems (2)
lore
Share & tools
lore

Eric Paris

author
#0 lore raw

PATCH 1/3 notify: unused event private race

inotify decides if private data it passed to get added to an event was used by checking list_empty(). But it's possible that the event may have been dequeued and the private event removed so it would look empty. The fix is to use the return code from fsnotify_add_notify_event rather than looking at the list.

---

 fs/notify/inotify/inotify_fsnotify.c |   13 +++++++------
 fs/notify/inotify/inotify_user.c     |    7 +++----
 fs/notify/notification.c             |    7 +++----
 3 files changed, 13 insertions(+), 14 deletions(-)
fs/notify/inotify/inotify_fsnotify.c +7 -6
index 47cd258..5dcbafe 100644
@@ -62,13 +62,14 @@ static int inotify_handle_event(struct fsnotify_group *group, struct fsnotify_ev
 	event_priv->wd = wd;
 
 	ret = fsnotify_add_notify_event(group, event, fsn_event_priv);
-	/* EEXIST is not an error */
-	if (ret == -EEXIST)
-		ret = 0;
-
-	/* did event_priv get attached? */
-	if (list_empty(&fsn_event_priv->event_list))
+	if (ret) {
 		inotify_free_event_priv(fsn_event_priv);
+		/* EEXIST says we tail matched, EOVERFLOW isn't something
+		 * to report up the stack. */
+		if ((ret == -EEXIST) ||
+		    (ret == -EOVERFLOW))
+			ret = 0;
+	}
 
 	/*
 	 * If we hold the entry until after the event is on the queue
fs/notify/inotify/inotify_user.c +3 -4
index c17d5a3..54cbded 100644
@@ -386,6 +386,7 @@ void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 	struct fsnotify_event *ignored_event;
 	struct inotify_event_private_data *event_priv;
 	struct fsnotify_event_private_data *fsn_event_priv;
+	int ret;
 
 	ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
 					      FSNOTIFY_EVENT_NONE, NULL, 0,
@@ -404,10 +405,8 @@ void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 	fsn_event_priv->group = group;
 	event_priv->wd = ientry->wd;
 
-	fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
-
-	/* did the private data get added? */
-	if (list_empty(&fsn_event_priv->event_list))
+	ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
+	if (ret)
 		inotify_free_event_priv(fsn_event_priv);
 
 skip_send_ignore:
fs/notify/notification.c +3 -4
index 5213685..74b3cf3 100644
@@ -171,9 +171,7 @@ int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_even
 	struct list_head *list = &group->notification_list;
 	struct fsnotify_event_holder *last_holder;
 	struct fsnotify_event *last_event;
-
-	/* easy to tell if priv was attached to the event */
-	INIT_LIST_HEAD(&priv->event_list);
+	int ret = 0;
 
 	/*
 	 * There is one fsnotify_event_holder embedded inside each fsnotify_event.
@@ -194,6 +192,7 @@ alloc_holder:
 
 	if (group->q_len >= group->max_events) {
 		event = &q_overflow_event;
+		ret = -EOVERFLOW;
 		/* sorry, no private data on the overflow event */
 		priv = NULL;
 	}
@@ -235,7 +234,7 @@ alloc_holder:
 	mutex_unlock(&group->notification_mutex);
 
 	wake_up(&group->notification_waitq);
-	return 0;
+	return ret;
 }
 
 /*
author
#1 lore raw

PATCH 2/3 inotify: tail drop inotify q_overflow events

In f44aebcc the tail drop logic of events with no file backing (q_overflow and in_ignored) was reversed so IN_IGNORED events would never be tail dropped. This now means that Q_OVERFLOW events are NOT tail dropped. The fix is to not tail drop IN_IGNORED, but to tail drop Q_OVERFLOW.

---

 fs/notify/notification.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)
fs/notify/notification.c +4 -0
index 74b3cf3..3816d57 100644
@@ -153,6 +153,10 @@ static bool event_compare(struct fsnotify_event *old, struct fsnotify_event *new
 				return true;
 			break;
 		case (FSNOTIFY_EVENT_NONE):
+			if (old->mask & FS_Q_OVERFLOW)
+				return true;
+			else if (old->mask & FS_IN_IGNORED)
+				return false;
 			return false;
 		};
 	}
author
#2 lore raw

PATCH 3/3 inotify: start watch descriptor count at 1

In reply to Eric Paris

The inotify_add_watch man page specifies that inotify_add_watch() will return a non-negative integer. However, historically the inotify watches started at 1, not at 0. Turns out that the inotifywait program provided by the inotify-tools package doesn't properly handle a 0 watch descriptor. In 7e790dd5 we changed from starting at 1 to starting at 0. This patch starts at 1, just like in previous kernels, but also just like in previous kernels it's possible for it to wrap back to 0. This preserves the kernel functionality exactly like it was before the patch (neither method broke the spec)

---

 fs/notify/inotify/inotify_user.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
fs/notify/inotify/inotify_user.c +1 -1
index 54cbded..ca0f36b 100644
@@ -567,7 +567,7 @@ static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsign
 
 	spin_lock_init(&group->inotify_data.idr_lock);
 	idr_init(&group->inotify_data.idr);
-	group->inotify_data.last_wd = 0;
+	group->inotify_data.last_wd = 1;
 	group->inotify_data.user = user;
 	group->inotify_data.fa = NULL;