Skip to contentSkip to conversation

inotify: invalid mask should return a error number but not set it

PATCH Superseded Zhao Hongjiang 2 participants 3 messages Apr 18, 2013 – Apr 25, 2013

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 inotify: invalid mask should return a error number but not set it

When we run the crackerjack testsuit, inotify_add_watch test is stalled cause the invalid mask 0, the task is waiting for the event but it never come. This should return -EINVAL and it do is before the commit 676a0675cf9200 ("inotify: remove broken mask checks causing unmount to be EINVAL"). The commit remove the invalid mask check simply, but the invalid mask check is needed indeed.

Check the mask wether in the ALL_INOTIFY_BITS before the inotify_arg_to_mask call, if is not, just return -EINVAL.

Because IN_UNMOUNT is in ALL_INOTIFY_BITS, so this change will not trigger the problem that above commit fixed.

---
 fs/notify/inotify/inotify_user.c |    6 ++++++
 1 files changed, 6 insertions(+), 0 deletions(-)
fs/notify/inotify/inotify_user.c +6 -0
index e0f7c12..b024bc1 100644
@@ -573,6 +573,9 @@ static int inotify_update_existing_watch(struct fsnotify_group *group,
 	int ret;
 
 	/* don't allow invalid bits: we don't want flags set */
+	if (unlikely(!(arg & ALL_INOTIFY_BITS)))
+		return -EINVAL;
+
 	mask = inotify_arg_to_mask(arg);
 
 	fsn_mark = fsnotify_find_inode_mark(group, inode);
@@ -624,6 +627,9 @@ static int inotify_new_watch(struct fsnotify_group *group,
 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 
 	/* don't allow invalid bits: we don't want flags set */
+	if (unlikely(!(arg & ALL_INOTIFY_BITS)))
+		return -EINVAL;
+
 	mask = inotify_arg_to_mask(arg);
 
 	tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);

-- 1.7.1

#1 lore raw
Quoted 8 lines · Zhao Hongjiang

When we run the crackerjack testsuit, inotify_add_watch test is stalled cause the invalid mask 0, the task is waiting for the event but it never come. This should return -EINVAL and it do is before the commit 676a0675cf9200 ("inotify: remove broken mask checks causing unmount to be EINVAL"). The commit remove the invalid mask check simply, but the invalid mask check is needed indeed.

Check the mask wether in the ALL_INOTIFY_BITS before the inotify_arg_to_mask call, if is not, just return -EINVAL.

If you want to filter out zero as EINVAL, then it should be done right at the top of the syscall entry and not duplicated twice here in the guts of the code, I think. I can send a suggested patch if that description is not 100% self evident.

Paul.
--

Quoted 36 lines

Because IN_UNMOUNT is in ALL_INOTIFY_BITS, so this change will not trigger the problem that above commit fixed.

---
 fs/notify/inotify/inotify_user.c |    6 ++++++
 1 files changed, 6 insertions(+), 0 deletions(-)
fs/notify/inotify/inotify_user.c +6 -1
index e0f7c12..b024bc1 100644
@@ -573,6 +573,9 @@ static int inotify_update_existing_watch(struct fsnotify_group *group,
 	int ret;
 
 	/* don't allow invalid bits: we don't want flags set */
+	if (unlikely(!(arg & ALL_INOTIFY_BITS)))
+		return -EINVAL;
+
 	mask = inotify_arg_to_mask(arg);
 
 	fsn_mark = fsnotify_find_inode_mark(group, inode);
@@ -624,6 +627,9 @@ static int inotify_new_watch(struct fsnotify_group *group,
 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 
 	/* don't allow invalid bits: we don't want flags set */
+	if (unlikely(!(arg & ALL_INOTIFY_BITS)))
+		return -EINVAL;
+
 	mask = inotify_arg_to_mask(arg);
 
 	tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
-- 1.7.1
#2 lore raw
Quoted 7 lines · Paul Gortmaker
Quoted 6 lines · Zhao Hongjiang

When we run the crackerjack testsuit, inotify_add_watch test is stalled cause the invalid mask 0, the task is waiting for the event but it never come. This should return -EINVAL and it do is before the commit 676a0675cf9200 ("inotify: remove broken mask checks causing unmount to be EINVAL"). The commit remove the invalid mask check simply, but the invalid mask check is needed indeed.

Quoted 8 lines
Quoted 2 lines

Check the mask wether in the ALL_INOTIFY_BITS before the inotify_arg_to_mask call, if is not, just return -EINVAL.

If you want to filter out zero as EINVAL, then it should be done right at the top of the syscall entry and not duplicated twice here in the guts of the code, I think. I can send a suggested patch if that description is not 100% self evident.

Not only filter out zero but also other invalid mask like 0x00001000 which are not in ALL_INOTIFY_BITS. i just leave the check at the place before, but move to the top of the syscall entry is more reasonable. I'll send the v2 shortly. Thanks.

Zhao Hongjiang.

Quoted 42 lines

Paul.
--

Quoted 26 lines

Because IN_UNMOUNT is in ALL_INOTIFY_BITS, so this change will not trigger the problem that above commit fixed.

---
 fs/notify/inotify/inotify_user.c |    6 ++++++
 1 files changed, 6 insertions(+), 0 deletions(-)
fs/notify/inotify/inotify_user.c +3 -0
index e0f7c12..b024bc1 100644
@@ -573,6 +573,9 @@ static int inotify_update_existing_watch(struct fsnotify_group *group,
 	int ret;
 
 	/* don't allow invalid bits: we don't want flags set */
+	if (unlikely(!(arg & ALL_INOTIFY_BITS)))
+		return -EINVAL;
+
 	mask = inotify_arg_to_mask(arg);
 
 	fsn_mark = fsnotify_find_inode_mark(group, inode);

fs/notify/inotify/inotify_user.c static int inotify_new_watch(struct fsnotify_group *group,

Quoted 10 lines
fs/notify/inotify/inotify_user.c +3 -1
@@ -624,6 +627,9 @@ static int inotify_new_watch(struct fsnotify_group *group,
 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 
 	/* don't allow invalid bits: we don't want flags set */
+	if (unlikely(!(arg & ALL_INOTIFY_BITS)))
+		return -EINVAL;
+
 	mask = inotify_arg_to_mask(arg);
 
 	tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
-- 1.7.1

.