Skip to contentSkip to conversation

fs/namespace.c: WARN if mnt_count has become negative

PATCH In review Eric Biggers 1 participant 3 messages Nov 1, 2020 – Dec 2, 2020 VFS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fs/namespace.c: WARN if mnt_count has become negative

From: Eric Biggers <[email protected]>

Missing calls to mntget() (or equivalently, too many calls to mntput()) are hard to detect because mntput() delays freeing mounts using task_work_add(), then again using call_rcu(). As a result, mnt_count can often be decremented to -1 without getting a KASAN use-after-free report. Such cases are still bugs though, and they point to real use-after-frees being possible.

For an example of this, see the bug fixed by commit 1b0b9cc8d379 ("vfs: fsmount: add missing mntget()"), discussed at
https://lkml.kernel.org/linux-fsdevel/[email protected]/T/#u.
This bug *should* have been trivial to find. But actually, it wasn't found until syzkaller happened to use fchdir() to manipulate the reference count just right for the bug to be noticeable.

Address this by making mntput_no_expire() issue a WARN if mnt_count has become negative.

---
 fs/namespace.c | 9 ++++++---
 fs/pnode.h     | 2 +-
 2 files changed, 7 insertions(+), 4 deletions(-)
fs/namespace.c +6 -3
index cebaa3e817940..93006abe7946a 100644
@@ -156,10 +156,10 @@ static inline void mnt_add_count(struct mount *mnt, int n)
 /*
  * vfsmount lock must be held for write
  */
-unsigned int mnt_get_count(struct mount *mnt)
+int mnt_get_count(struct mount *mnt)
 {
 #ifdef CONFIG_SMP
-	unsigned int count = 0;
+	int count = 0;
 	int cpu;
 
 	for_each_possible_cpu(cpu) {
@@ -1139,6 +1139,7 @@ static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
 static void mntput_no_expire(struct mount *mnt)
 {
 	LIST_HEAD(list);
+	int count;
 
 	rcu_read_lock();
 	if (likely(READ_ONCE(mnt->mnt_ns))) {
@@ -1162,7 +1163,9 @@ static void mntput_no_expire(struct mount *mnt)
 	 */
 	smp_mb();
 	mnt_add_count(mnt, -1);
-	if (mnt_get_count(mnt)) {
+	count = mnt_get_count(mnt);
+	if (count != 0) {
+		WARN_ON(count < 0);
 		rcu_read_unlock();
 		unlock_mount_hash();
 		return;
fs/pnode.h +1 -1
index 49a058c73e4c7..26f74e092bd98 100644
@@ -44,7 +44,7 @@ int propagate_mount_busy(struct mount *, int);
 void propagate_mount_unlock(struct mount *);
 void mnt_release_group_id(struct mount *);
 int get_dominating_id(struct mount *mnt, const struct path *root);
-unsigned int mnt_get_count(struct mount *mnt);
+int mnt_get_count(struct mount *mnt);
 void mnt_set_mountpoint(struct mount *, struct mountpoint *,
 			struct mount *);
 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp,

base-commit: c2dc4c073fb71b50904493657a7622b481b346e3

Signature · 1 line
2.29.1
#1 lore raw
Quoted 68 lines · Eric Biggers

From: Eric Biggers <[email protected]>

Missing calls to mntget() (or equivalently, too many calls to mntput()) are hard to detect because mntput() delays freeing mounts using task_work_add(), then again using call_rcu(). As a result, mnt_count can often be decremented to -1 without getting a KASAN use-after-free report. Such cases are still bugs though, and they point to real use-after-frees being possible.

For an example of this, see the bug fixed by commit 1b0b9cc8d379 ("vfs: fsmount: add missing mntget()"), discussed at
https://lkml.kernel.org/linux-fsdevel/[email protected]/T/#u.
This bug *should* have been trivial to find. But actually, it wasn't found until syzkaller happened to use fchdir() to manipulate the reference count just right for the bug to be noticeable.

Address this by making mntput_no_expire() issue a WARN if mnt_count has become negative.

---
 fs/namespace.c | 9 ++++++---
 fs/pnode.h     | 2 +-
 2 files changed, 7 insertions(+), 4 deletions(-)
fs/namespace.c +6 -3
index cebaa3e817940..93006abe7946a 100644
@@ -156,10 +156,10 @@ static inline void mnt_add_count(struct mount *mnt, int n)
 /*
  * vfsmount lock must be held for write
  */
-unsigned int mnt_get_count(struct mount *mnt)
+int mnt_get_count(struct mount *mnt)
 {
 #ifdef CONFIG_SMP
-	unsigned int count = 0;
+	int count = 0;
 	int cpu;
 
 	for_each_possible_cpu(cpu) {
@@ -1139,6 +1139,7 @@ static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
 static void mntput_no_expire(struct mount *mnt)
 {
 	LIST_HEAD(list);
+	int count;
 
 	rcu_read_lock();
 	if (likely(READ_ONCE(mnt->mnt_ns))) {
@@ -1162,7 +1163,9 @@ static void mntput_no_expire(struct mount *mnt)
 	 */
 	smp_mb();
 	mnt_add_count(mnt, -1);
-	if (mnt_get_count(mnt)) {
+	count = mnt_get_count(mnt);
+	if (count != 0) {
+		WARN_ON(count < 0);
 		rcu_read_unlock();
 		unlock_mount_hash();
 		return;
fs/pnode.h +0 -0
index 49a058c73e4c7..26f74e092bd98 100644

fs/pnode.h int propagate_mount_busy(struct mount *, int); · line 50

Quoted 8 lines
fs/pnode.h +1 -1
@@ -44,7 +44,7 @@ int propagate_mount_busy(struct mount *, int);
 void propagate_mount_unlock(struct mount *);
 void mnt_release_group_id(struct mount *);
 int get_dominating_id(struct mount *mnt, const struct path *root);
-unsigned int mnt_get_count(struct mount *mnt);
+int mnt_get_count(struct mount *mnt);
 void mnt_set_mountpoint(struct mount *, struct mountpoint *,
 			struct mount *);
 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp,

Ping?

#2 lore raw
Quoted 69 lines · Eric Biggers
Quoted 68 lines · Eric Biggers

From: Eric Biggers <[email protected]>

Missing calls to mntget() (or equivalently, too many calls to mntput()) are hard to detect because mntput() delays freeing mounts using task_work_add(), then again using call_rcu(). As a result, mnt_count can often be decremented to -1 without getting a KASAN use-after-free report. Such cases are still bugs though, and they point to real use-after-frees being possible.

For an example of this, see the bug fixed by commit 1b0b9cc8d379 ("vfs: fsmount: add missing mntget()"), discussed at
https://lkml.kernel.org/linux-fsdevel/[email protected]/T/#u.
This bug *should* have been trivial to find. But actually, it wasn't found until syzkaller happened to use fchdir() to manipulate the reference count just right for the bug to be noticeable.

Address this by making mntput_no_expire() issue a WARN if mnt_count has become negative.

---
 fs/namespace.c | 9 ++++++---
 fs/pnode.h     | 2 +-
 2 files changed, 7 insertions(+), 4 deletions(-)
fs/namespace.c +6 -3
index cebaa3e817940..93006abe7946a 100644
@@ -156,10 +156,10 @@ static inline void mnt_add_count(struct mount *mnt, int n)
 /*
  * vfsmount lock must be held for write
  */
-unsigned int mnt_get_count(struct mount *mnt)
+int mnt_get_count(struct mount *mnt)
 {
 #ifdef CONFIG_SMP
-	unsigned int count = 0;
+	int count = 0;
 	int cpu;
 
 	for_each_possible_cpu(cpu) {
@@ -1139,6 +1139,7 @@ static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
 static void mntput_no_expire(struct mount *mnt)
 {
 	LIST_HEAD(list);
+	int count;
 
 	rcu_read_lock();
 	if (likely(READ_ONCE(mnt->mnt_ns))) {
@@ -1162,7 +1163,9 @@ static void mntput_no_expire(struct mount *mnt)
 	 */
 	smp_mb();
 	mnt_add_count(mnt, -1);
-	if (mnt_get_count(mnt)) {
+	count = mnt_get_count(mnt);
+	if (count != 0) {
+		WARN_ON(count < 0);
 		rcu_read_unlock();
 		unlock_mount_hash();
 		return;
fs/pnode.h +0 -0
index 49a058c73e4c7..26f74e092bd98 100644
@@ -44,7 +44,7 @@ int propagate_mount_busy(struct mount *, int);
 void propagate_mount_unlock(struct mount *);
Quoted 8 lines

fs/pnode.h int propagate_mount_busy(struct mount *, int); · line 50

Quoted 8 lines
fs/pnode.h +1 -1
 void mnt_release_group_id(struct mount *);
 int get_dominating_id(struct mount *mnt, const struct path *root);
-unsigned int mnt_get_count(struct mount *mnt);
+int mnt_get_count(struct mount *mnt);
 void mnt_set_mountpoint(struct mount *, struct mountpoint *,
 			struct mount *);
 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp,

Ping.