Skip to contentSkip to conversation

thermal/debugfs: Fix a memory leak on removal and locking

PATCH In review Rafael J. Wysocki 3 participants 12 messages Apr 25, 2024 – Apr 26, 2024 Thermal

lore
Share & tools
lore
#0 lore raw

cover thermal/debugfs: Fix a memory leak on removal and locking

Hi Everyone,

This series fixes a memory leak on thermal zone removal caused by the thermal debug code and two locking issues that may cause the kernel to crash due to race conditions.

Please review urgently.

The series applies on top of the linux-next branch in linux-pm.git.

It is also present in the thermal-core-next branch in that tree.

Thanks!

#1 lore raw

PATCH 1/3 thermal/debugfs: Free all thermal zone debug memory on zone removal

From: Rafael J. Wysocki <[email protected]>

Because thermal_debug_tz_remove() does not free all memory allocated for thermal zone diagnostics, some of that memory becomes unreachable after freeing the thermal zone's struct thermal_debugfs object.

Address this by making thermal_debug_tz_remove() free all of the memory in question.

Fixes 7ef01f228c9f ("thermal/debugfs: Add thermal debugfs information for mitigation episodes")
Cc 6.8+ <[email protected]> # 6.8+
---
 drivers/thermal/thermal_debugfs.c |   13 +++++++++++++
 1 file changed, 13 insertions(+)
drivers/thermal/thermal_debugfs.c +13 -0
Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================
@@ -832,15 +832,28 @@ void thermal_debug_tz_add(struct thermal
 void thermal_debug_tz_remove(struct thermal_zone_device *tz)
 {
 	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct tz_episode *tze, *tmp;
+	struct tz_debugfs *tz_dbg;
+	int *trips_crossed;
 
 	if (!thermal_dbg)
 		return;
 
+	tz_dbg = &thermal_dbg->tz_dbg;
+
 	mutex_lock(&thermal_dbg->lock);
 
+	trips_crossed = tz_dbg->trips_crossed;
+
+	list_for_each_entry_safe(tze, tmp, &tz_dbg->tz_episodes, node) {
+		list_del(&tze->node);
+		kfree(tze);
+	}
+
 	tz->debugfs = NULL;
 
 	mutex_unlock(&thermal_dbg->lock);
 
 	thermal_debugfs_remove_id(thermal_dbg);
+	kfree(trips_crossed);
 }
reviewer Reviewed-by on v1 1/3
#2 lore raw

Re: [PATCH v1 1/3] thermal/debugfs: Free all thermal zone debug memory on zone removal

Quoted 43 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Because thermal_debug_tz_remove() does not free all memory allocated for thermal zone diagnostics, some of that memory becomes unreachable after freeing the thermal zone's struct thermal_debugfs object.

Address this by making thermal_debug_tz_remove() free all of the memory in question.

Fixes 7ef01f228c9f ("thermal/debugfs: Add thermal debugfs information for mitigation episodes")
Cc 6.8+ <[email protected]> # 6.8+
---
  drivers/thermal/thermal_debugfs.c |   13 +++++++++++++
  1 file changed, 13 insertions(+)

Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================

drivers/thermal/thermal_debugfs.c +9 -0
@@ -832,15 +832,28 @@ void thermal_debug_tz_add(struct thermal
  void thermal_debug_tz_remove(struct thermal_zone_device *tz)
  {
  	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct tz_episode *tze, *tmp;
+	struct tz_debugfs *tz_dbg;
+	int *trips_crossed;
 
  	if (!thermal_dbg)
  		return;
 
+	tz_dbg = &thermal_dbg->tz_dbg;
+
  	mutex_lock(&thermal_dbg->lock);
 
+	trips_crossed = tz_dbg->trips_crossed;
+
+	list_for_each_entry_safe(tze, tmp, &tz_dbg->tz_episodes, node) {
+		list_del(&tze->node);

drivers/thermal/thermal_debugfs.c void thermal_debug_tz_add(struct thermal · line 859

Quoted 10 lines
drivers/thermal/thermal_debugfs.c +4 -0
+		kfree(tze);
+	}
+
  	tz->debugfs = NULL;
 
  	mutex_unlock(&thermal_dbg->lock);
 
  	thermal_debugfs_remove_id(thermal_dbg);
+	kfree(trips_crossed);
  }
#3 lore raw

PATCH 2/3 thermal/debugfs: Fix thermal zone locking

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

With the current thermal zone locking arrangement in the debugfs code, user space can open the "mitigations" file for a thermal zone before the zone's debugfs pointer is set which will result in a NULL pointer dereference in tze_seq_start().

While this could be addressed by putting the creation of the "mitigations" file under thermal_dbg->lock, there is still a problem with thermal_debug_tz_remove() that is not called under the thermal zone lock and can run in parallel with the other functions accessing the thermal zone's struct thermal_debugfs object. Then, it may clear tz->debugfs after one of those functions has checked it and the struct thermal_debugfs object may be freed prematurely.

To address both problems described above at once, use the observation that thermal_debug_tz_trip_up(), thermal_debug_tz_trip_down(), and thermal_debug_update_trip_stats() all run under the thermal zone lock and because they all acquire thermal_dbg->lock for the thermal zone's struct thermal_debugfs object, they must wait on that lock if it is held while they are running and their callers (holding the thermal zone lock) must wait along with them. This means that tze_seq_start() may as well acquire tz->lock instead of thermal_dbg->lock and check the struct thermal_debugfs object pointer retrieved from the thermal zone against NULL under it.

Then, tz->lock can also be acquired by thermal_debug_tz_add() and thermal_debug_tz_remove() to eliminate the race conditions at hand.

Rearrange the code in question accordingly and remove the thermal_dbg->lock locking, which is now redundant, from it.

Fixes 7ef01f228c9f ("thermal/debugfs: Add thermal debugfs information for mitigation episodes")
Cc 6.8+ <[email protected]> # 6.8+
---
 drivers/thermal/thermal_debugfs.c |   60 +++++++++++++++++---------------------
 1 file changed, 28 insertions(+), 32 deletions(-)
drivers/thermal/thermal_debugfs.c +28 -32
Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================
@@ -551,8 +551,6 @@ void thermal_debug_tz_trip_up(struct the
 	if (!thermal_dbg)
 		return;
 
-	mutex_lock(&thermal_dbg->lock);
-
 	tz_dbg = &thermal_dbg->tz_dbg;
 
 	/*
@@ -591,7 +589,7 @@ void thermal_debug_tz_trip_up(struct the
 	if (!tz_dbg->nr_trips) {
 		tze = thermal_debugfs_tz_event_alloc(tz, now);
 		if (!tze)
-			goto unlock;
+			return;
 
 		list_add(&tze->node, &tz_dbg->tz_episodes);
 	}
@@ -613,9 +611,6 @@ void thermal_debug_tz_trip_up(struct the
 
 	tze = list_first_entry(&tz_dbg->tz_episodes, struct tz_episode, node);
 	tze->trip_stats[trip_id].timestamp = now;
-
-unlock:
-	mutex_unlock(&thermal_dbg->lock);
 }
 
 void thermal_debug_tz_trip_down(struct thermal_zone_device *tz,
@@ -631,8 +626,6 @@ void thermal_debug_tz_trip_down(struct t
 	if (!thermal_dbg)
 		return;
 
-	mutex_lock(&thermal_dbg->lock);
-
 	tz_dbg = &thermal_dbg->tz_dbg;
 
 	/*
@@ -643,7 +636,7 @@ void thermal_debug_tz_trip_down(struct t
 	 * no mitigation mechanism yet at boot time.
 	 */
 	if (!tz_dbg->nr_trips)
-		goto out;
+		return;
 
 	for (i = tz_dbg->nr_trips - 1; i >= 0; i--) {
 		if (tz_dbg->trips_crossed[i] == trip_id)
@@ -651,7 +644,7 @@ void thermal_debug_tz_trip_down(struct t
 	}
 
 	if (i < 0)
-		goto out;
+		return;
 
 	tz_dbg->nr_trips--;
 
@@ -671,9 +664,6 @@ void thermal_debug_tz_trip_down(struct t
 	 */
 	if (!tz_dbg->nr_trips)
 		tze->duration = ktime_sub(now, tze->timestamp);
-
-out:
-	mutex_unlock(&thermal_dbg->lock);
 }
 
 void thermal_debug_update_trip_stats(struct thermal_zone_device *tz)
@@ -686,12 +676,10 @@ void thermal_debug_update_trip_stats(str
 	if (!thermal_dbg)
 		return;
 
-	mutex_lock(&thermal_dbg->lock);
-
 	tz_dbg = &thermal_dbg->tz_dbg;
 
 	if (!tz_dbg->nr_trips)
-		goto out;
+		return;
 
 	tze = list_first_entry(&tz_dbg->tz_episodes, struct tz_episode, node);
 
@@ -704,19 +692,22 @@ void thermal_debug_update_trip_stats(str
 		trip_stats->avg += (tz->temperature - trip_stats->avg) /
 					++trip_stats->count;
 	}
-out:
-	mutex_unlock(&thermal_dbg->lock);
 }
 
 static void *tze_seq_start(struct seq_file *s, loff_t *pos)
 {
 	struct thermal_zone_device *tz = s->private;
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
-	struct tz_debugfs *tz_dbg = &thermal_dbg->tz_dbg;
+	struct thermal_debugfs *thermal_dbg;
 
-	mutex_lock(&thermal_dbg->lock);
+	mutex_lock(&tz->lock);
 
-	return seq_list_start(&tz_dbg->tz_episodes, *pos);
+	thermal_dbg = tz->debugfs;
+	if (!thermal_dbg) {
+		mutex_unlock(&tz->lock);
+		return NULL;
+	}
+
+	return seq_list_start(&thermal_dbg->tz_dbg.tz_episodes, *pos);
 }
 
 static void *tze_seq_next(struct seq_file *s, void *v, loff_t *pos)
@@ -731,9 +722,8 @@ static void *tze_seq_next(struct seq_fil
 static void tze_seq_stop(struct seq_file *s, void *v)
 {
 	struct thermal_zone_device *tz = s->private;
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
 
-	mutex_unlock(&thermal_dbg->lock);
+	mutex_unlock(&tz->lock);
 }
 
 static int tze_seq_show(struct seq_file *s, void *v)
@@ -826,23 +816,33 @@ void thermal_debug_tz_add(struct thermal
 
 	debugfs_create_file("mitigations", 0400, thermal_dbg->d_top, tz, &tze_fops);
 
+	mutex_lock(&tz->lock);
+
 	tz->debugfs = thermal_dbg;
+
+	mutex_unlock(&tz->lock);
 }
 
 void thermal_debug_tz_remove(struct thermal_zone_device *tz)
 {
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct thermal_debugfs *thermal_dbg;
 	struct tz_episode *tze, *tmp;
 	struct tz_debugfs *tz_dbg;
 	int *trips_crossed;
 
-	if (!thermal_dbg)
+	mutex_lock(&tz->lock);
+
+	thermal_dbg = tz->debugfs;
+	if (!thermal_dbg) {
+		mutex_unlock(&tz->lock);
 		return;
+	}
 
-	tz_dbg = &thermal_dbg->tz_dbg;
+	tz->debugfs = NULL;
 
-	mutex_lock(&thermal_dbg->lock);
+	mutex_unlock(&tz->lock);
 
+	tz_dbg = &thermal_dbg->tz_dbg;
 	trips_crossed = tz_dbg->trips_crossed;
 
 	list_for_each_entry_safe(tze, tmp, &tz_dbg->tz_episodes, node) {
@@ -850,10 +850,6 @@ void thermal_debug_tz_remove(struct ther
 		kfree(tze);
 	}
 
-	tz->debugfs = NULL;
-
-	mutex_unlock(&thermal_dbg->lock);
-
 	thermal_debugfs_remove_id(thermal_dbg);
 	kfree(trips_crossed);
 }
#4 lore raw

PATCH 2/3 thermal/debugfs: Fix two locking issues with thermal zone debug

From: Rafael J. Wysocki <[email protected]>

With the current thermal zone locking arrangement in the debugfs code, user space can open the "mitigations" file for a thermal zone before the zone's debugfs pointer is set which will result in a NULL pointer dereference in tze_seq_start().

Moreover, thermal_debug_tz_remove() is not called under the thermal zone lock, so can run in parallel with the other functions accessing the thermal zone's struct thermal_debugfs object. Then, it may clear tz->debugfs after one of those functions has checked it and the struct thermal_debugfs object may be freed prematurely.

To address the first problem, pass a pointer to the thermal zone's struct thermal_debugfs object to debugfs_create_file() in thermal_debug_tz_add() and make tze_seq_start(), tze_seq_next(), tze_seq_stop(), and tze_seq_show() retrieve it from s->private instead of a pointer to the thermal zone object. This will ensure that tz_debugfs will be valid across the "mitigations" file accesses until thermal_debugfs_remove_id() called by thermal_debug_tz_remove() removes that file.

To address the second problem, use tz->lock in thermal_debug_tz_remove() around the tz->debugfs value check (in case the same thermal zone is removed at the same time in two differet threads) and its reset to NULL.

Fixes 7ef01f228c9f ("thermal/debugfs: Add thermal debugfs information for mitigation episodes")
Cc 6.8+ <[email protected]> # 6.8+

---

This is an alternative fix for the issues addressed by

https://lore.kernel.org/linux-pm/1888579.tdWV9SEqCh@kreacher/

and I slightly prefer it, because it is less intrusive and makes the thermal zone debug code more consistent with the analogous code for cdevs.

Accordingly, I've replace the above with this patch in the thermal-core-next branch in linux-pm.git.

---
 drivers/thermal/thermal_debugfs.c |   34 ++++++++++++++++++++++------------
 1 file changed, 22 insertions(+), 12 deletions(-)
drivers/thermal/thermal_debugfs.c +22 -12
Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================
@@ -139,11 +139,13 @@ struct tz_episode {
  * we keep track of the current position in the history array.
  *
  * @tz_episodes: a list of thermal mitigation episodes
+ * @tz: thermal zone this object belongs to
  * @trips_crossed: an array of trip points crossed by id
  * @nr_trips: the number of trip points currently being crossed
  */
 struct tz_debugfs {
 	struct list_head tz_episodes;
+	struct thermal_zone_device *tz;
 	int *trips_crossed;
 	int nr_trips;
 };
@@ -710,8 +712,7 @@ out:
 
 static void *tze_seq_start(struct seq_file *s, loff_t *pos)
 {
-	struct thermal_zone_device *tz = s->private;
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct thermal_debugfs *thermal_dbg = s->private;
 	struct tz_debugfs *tz_dbg = &thermal_dbg->tz_dbg;
 
 	mutex_lock(&thermal_dbg->lock);
@@ -721,8 +722,7 @@ static void *tze_seq_start(struct seq_fi
 
 static void *tze_seq_next(struct seq_file *s, void *v, loff_t *pos)
 {
-	struct thermal_zone_device *tz = s->private;
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct thermal_debugfs *thermal_dbg = s->private;
 	struct tz_debugfs *tz_dbg = &thermal_dbg->tz_dbg;
 
 	return seq_list_next(v, &tz_dbg->tz_episodes, pos);
@@ -730,15 +730,15 @@ static void *tze_seq_next(struct seq_fil
 
 static void tze_seq_stop(struct seq_file *s, void *v)
 {
-	struct thermal_zone_device *tz = s->private;
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct thermal_debugfs *thermal_dbg = s->private;
 
 	mutex_unlock(&thermal_dbg->lock);
 }
 
 static int tze_seq_show(struct seq_file *s, void *v)
 {
-	struct thermal_zone_device *tz = s->private;
+	struct thermal_debugfs *thermal_dbg = s->private;
+	struct thermal_zone_device *tz = thermal_dbg->tz_dbg.tz;
 	struct thermal_trip_desc *td;
 	struct tz_episode *tze;
 	const char *type;
@@ -816,6 +816,8 @@ void thermal_debug_tz_add(struct thermal
 
 	tz_dbg = &thermal_dbg->tz_dbg;
 
+	tz_dbg->tz = tz;
+
 	tz_dbg->trips_crossed = kzalloc(sizeof(int) * tz->num_trips, GFP_KERNEL);
 	if (!tz_dbg->trips_crossed) {
 		thermal_debugfs_remove_id(thermal_dbg);
@@ -824,20 +826,30 @@ void thermal_debug_tz_add(struct thermal
 
 	INIT_LIST_HEAD(&tz_dbg->tz_episodes);
 
-	debugfs_create_file("mitigations", 0400, thermal_dbg->d_top, tz, &tze_fops);
+	debugfs_create_file("mitigations", 0400, thermal_dbg->d_top,
+			    thermal_dbg, &tze_fops);
 
 	tz->debugfs = thermal_dbg;
 }
 
 void thermal_debug_tz_remove(struct thermal_zone_device *tz)
 {
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct thermal_debugfs *thermal_dbg;
 	struct tz_episode *tze, *tmp;
 	struct tz_debugfs *tz_dbg;
 	int *trips_crossed;
 
-	if (!thermal_dbg)
+	mutex_lock(&tz->lock);
+
+	thermal_dbg = tz->debugfs;
+	if (!thermal_dbg) {
+		mutex_unlock(&tz->lock);
 		return;
+	}
+
+	tz->debugfs = NULL;
+
+	mutex_unlock(&tz->lock);
 
 	tz_dbg = &thermal_dbg->tz_dbg;
 
@@ -850,8 +862,6 @@ void thermal_debug_tz_remove(struct ther
 		kfree(tze);
 	}
 
-	tz->debugfs = NULL;
-
 	mutex_unlock(&thermal_dbg->lock);
 
 	thermal_debugfs_remove_id(thermal_dbg);
reviewer Reviewed-by on v1 2/3
#5 lore raw

Re: [Alternative][PATCH v1 2/3] thermal/debugfs: Fix two locking issues with thermal zone debug

Quoted 22 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

With the current thermal zone locking arrangement in the debugfs code, user space can open the "mitigations" file for a thermal zone before the zone's debugfs pointer is set which will result in a NULL pointer dereference in tze_seq_start().

Moreover, thermal_debug_tz_remove() is not called under the thermal zone lock, so can run in parallel with the other functions accessing the thermal zone's struct thermal_debugfs object. Then, it may clear tz->debugfs after one of those functions has checked it and the struct thermal_debugfs object may be freed prematurely.

To address the first problem, pass a pointer to the thermal zone's struct thermal_debugfs object to debugfs_create_file() in thermal_debug_tz_add() and make tze_seq_start(), tze_seq_next(), tze_seq_stop(), and tze_seq_show() retrieve it from s->private instead of a pointer to the thermal zone object. This will ensure that tz_debugfs will be valid across the "mitigations" file accesses until thermal_debugfs_remove_id() called by thermal_debug_tz_remove() removes that file.

Quoted 3 lines

To address the second problem, use tz->lock in thermal_debug_tz_remove() around the tz->debugfs value check (in case the same thermal zone is removed at the same time in two differet threads) and its reset to NULL.

s/differet/different/

Quoted 6 lines
Fixes 7ef01f228c9f ("thermal/debugfs: Add thermal debugfs information for mitigation episodes")
Cc 6.8+ <[email protected]> # 6.8+

---

Quoted 7 lines

This is an alternative fix for the issues addressed by

https://lore.kernel.org/linux-pm/1888579.tdWV9SEqCh@kreacher/

and I slightly prefer it, because it is less intrusive and makes the thermal zone debug code more consistent with the analogous code for cdevs.

I also prefer this one.

Quoted 112 lines

Accordingly, I've replace the above with this patch in the thermal-core-next branch in linux-pm.git.

---
  drivers/thermal/thermal_debugfs.c |   34 ++++++++++++++++++++++------------
  1 file changed, 22 insertions(+), 12 deletions(-)

Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================

drivers/thermal/thermal_debugfs.c +22 -10
@@ -139,11 +139,13 @@ struct tz_episode {
   * we keep track of the current position in the history array.
   *
   * @tz_episodes: a list of thermal mitigation episodes
+ * @tz: thermal zone this object belongs to
   * @trips_crossed: an array of trip points crossed by id
   * @nr_trips: the number of trip points currently being crossed
   */
  struct tz_debugfs {
  	struct list_head tz_episodes;
+	struct thermal_zone_device *tz;
  	int *trips_crossed;
  	int nr_trips;
  };
@@ -710,8 +712,7 @@ out:
 
  static void *tze_seq_start(struct seq_file *s, loff_t *pos)
  {
-	struct thermal_zone_device *tz = s->private;
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct thermal_debugfs *thermal_dbg = s->private;
  	struct tz_debugfs *tz_dbg = &thermal_dbg->tz_dbg;
 
  	mutex_lock(&thermal_dbg->lock);
@@ -721,8 +722,7 @@ static void *tze_seq_start(struct seq_fi
 
  static void *tze_seq_next(struct seq_file *s, void *v, loff_t *pos)
  {
-	struct thermal_zone_device *tz = s->private;
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct thermal_debugfs *thermal_dbg = s->private;
  	struct tz_debugfs *tz_dbg = &thermal_dbg->tz_dbg;
 
  	return seq_list_next(v, &tz_dbg->tz_episodes, pos);
@@ -730,15 +730,15 @@ static void *tze_seq_next(struct seq_fil
 
  static void tze_seq_stop(struct seq_file *s, void *v)
  {
-	struct thermal_zone_device *tz = s->private;
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct thermal_debugfs *thermal_dbg = s->private;
 
  	mutex_unlock(&thermal_dbg->lock);
  }
 
  static int tze_seq_show(struct seq_file *s, void *v)
  {
-	struct thermal_zone_device *tz = s->private;
+	struct thermal_debugfs *thermal_dbg = s->private;
+	struct thermal_zone_device *tz = thermal_dbg->tz_dbg.tz;
  	struct thermal_trip_desc *td;
  	struct tz_episode *tze;
  	const char *type;
@@ -816,6 +816,8 @@ void thermal_debug_tz_add(struct thermal
 
  	tz_dbg = &thermal_dbg->tz_dbg;
 
+	tz_dbg->tz = tz;
+
  	tz_dbg->trips_crossed = kzalloc(sizeof(int) * tz->num_trips, GFP_KERNEL);
  	if (!tz_dbg->trips_crossed) {
  		thermal_debugfs_remove_id(thermal_dbg);
@@ -824,20 +826,30 @@ void thermal_debug_tz_add(struct thermal
 
  	INIT_LIST_HEAD(&tz_dbg->tz_episodes);
 
-	debugfs_create_file("mitigations", 0400, thermal_dbg->d_top, tz, &tze_fops);
+	debugfs_create_file("mitigations", 0400, thermal_dbg->d_top,
+			    thermal_dbg, &tze_fops);
 
  	tz->debugfs = thermal_dbg;
  }
 
  void thermal_debug_tz_remove(struct thermal_zone_device *tz)
  {
-	struct thermal_debugfs *thermal_dbg = tz->debugfs;
+	struct thermal_debugfs *thermal_dbg;
  	struct tz_episode *tze, *tmp;
  	struct tz_debugfs *tz_dbg;
  	int *trips_crossed;
 
-	if (!thermal_dbg)
+	mutex_lock(&tz->lock);
+
+	thermal_dbg = tz->debugfs;
+	if (!thermal_dbg) {
+		mutex_unlock(&tz->lock);
  		return;
+	}
+
+	tz->debugfs = NULL;
+
+	mutex_unlock(&tz->lock);
 
  	tz_dbg = &thermal_dbg->tz_dbg;
 

drivers/thermal/thermal_debugfs.c void thermal_debug_tz_remove(struct ther · line 867

Quoted 8 lines
drivers/thermal/thermal_debugfs.c +0 -2
@@ -850,8 +862,6 @@ void thermal_debug_tz_remove(struct ther
  		kfree(tze);
  	}
 
-	tz->debugfs = NULL;
-
  	mutex_unlock(&thermal_dbg->lock);
 
  	thermal_debugfs_remove_id(thermal_dbg);

LGTM, with that minor spelling fixed:

#6 lore raw

PATCH 3/3 thermal/debugfs: Prevent use-after-free from occurring after cdev removal

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Since thermal_debug_cdev_remove() does not run under cdev->lock, it can run in parallel with thermal_debug_cdev_state_update() and it may free the struct thermal_debugfs object used by the latter after it has been checked against NULL.

If that happens, thermal_debug_cdev_state_update() will access memory that has been freed already causing the kernel to crash.

Address this by using cdev->lock in thermal_debug_cdev_remove() around the cdev->debugfs value check (in case the same cdev is removed at the same time in two differet threads) and its reset to NULL.

Fixes 755113d76786 ("thermal/debugfs: Add thermal cooling device debugfs information")
Cc 6.8+ <[email protected]> # 6.8+
---
 drivers/thermal/thermal_debugfs.c |   10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/thermal/thermal_debugfs.c +8 -2
Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================
@@ -503,15 +503,21 @@ void thermal_debug_cdev_add(struct therm
  */
 void thermal_debug_cdev_remove(struct thermal_cooling_device *cdev)
 {
-	struct thermal_debugfs *thermal_dbg = cdev->debugfs;
+	struct thermal_debugfs *thermal_dbg;
 
+	mutex_lock(&cdev->lock);
+
+	thermal_dbg = cdev->debugfs;
 	if (!thermal_dbg)
 		return;
 
+	cdev->debugfs = NULL;
+
+	mutex_unlock(&cdev->lock);
+
 	mutex_lock(&thermal_dbg->lock);
 
 	thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
-	cdev->debugfs = NULL;
 
 	mutex_unlock(&thermal_dbg->lock);
 
reviewer
#7 lore raw

Re: [PATCH v1 3/3] thermal/debugfs: Prevent use-after-free from occurring after cdev removal

Quoted 7 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Since thermal_debug_cdev_remove() does not run under cdev->lock, it can run in parallel with thermal_debug_cdev_state_update() and it may free the struct thermal_debugfs object used by the latter after it has been checked against NULL.

Quoted 6 lines

If that happens, thermal_debug_cdev_state_update() will access memory that has been freed already causing the kernel to crash.

Address this by using cdev->lock in thermal_debug_cdev_remove() around the cdev->debugfs value check (in case the same cdev is removed at the same time in two differet threads) and its reset to NULL.

s/differet/different/

Quoted 13 lines
Fixes 755113d76786 ("thermal/debugfs: Add thermal cooling device debugfs information")
Cc 6.8+ <[email protected]> # 6.8+
---
  drivers/thermal/thermal_debugfs.c |   10 ++++++++--
  1 file changed, 8 insertions(+), 2 deletions(-)

Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================

drivers/thermal/thermal_debugfs.c void thermal_debug_cdev_add(struct therm · line 511

Quoted 10 lines
drivers/thermal/thermal_debugfs.c +4 -1
@@ -503,15 +503,21 @@ void thermal_debug_cdev_add(struct therm
   */
  void thermal_debug_cdev_remove(struct thermal_cooling_device *cdev)
  {
-	struct thermal_debugfs *thermal_dbg = cdev->debugfs;
+	struct thermal_debugfs *thermal_dbg;
 
+	mutex_lock(&cdev->lock);
+
+	thermal_dbg = cdev->debugfs;
  	if (!thermal_dbg)

mutex_unlock(&cdev->lock) missing here

Quoted 17 lines
drivers/thermal/thermal_debugfs.c +4 -1
  		return;
 
+	cdev->debugfs = NULL;
+
+	mutex_unlock(&cdev->lock);
+
  	mutex_lock(&thermal_dbg->lock);
 
  	thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
-	cdev->debugfs = NULL;
 
  	mutex_unlock(&thermal_dbg->lock);
 
#8 lore raw

Re: [PATCH v1 3/3] thermal/debugfs: Prevent use-after-free from occurring after cdev removal

Quoted 39 lines · Lukasz Luba
Quoted 7 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Since thermal_debug_cdev_remove() does not run under cdev->lock, it can run in parallel with thermal_debug_cdev_state_update() and it may free the struct thermal_debugfs object used by the latter after it has been checked against NULL.

Quoted 6 lines

If that happens, thermal_debug_cdev_state_update() will access memory that has been freed already causing the kernel to crash.

Address this by using cdev->lock in thermal_debug_cdev_remove() around the cdev->debugfs value check (in case the same cdev is removed at the same time in two differet threads) and its reset to NULL.

s/differet/different/

Quoted 19 lines
Fixes 755113d76786 ("thermal/debugfs: Add thermal cooling device debugfs information")
Cc 6.8+ <[email protected]> # 6.8+
---
  drivers/thermal/thermal_debugfs.c |   10 ++++++++--
  1 file changed, 8 insertions(+), 2 deletions(-)

Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================

drivers/thermal/thermal_debugfs.c +1 -1
@@ -503,15 +503,21 @@ void thermal_debug_cdev_add(struct therm
   */
  void thermal_debug_cdev_remove(struct thermal_cooling_device *cdev)
  {
-     struct thermal_debugfs *thermal_dbg = cdev->debugfs;
+     struct thermal_debugfs *thermal_dbg;
 
Quoted 6 lines

drivers/thermal/thermal_debugfs.c void thermal_debug_cdev_add(struct therm · line 511

Quoted 4 lines
drivers/thermal/thermal_debugfs.c +3 -0
+     mutex_lock(&cdev->lock);
+
+     thermal_dbg = cdev->debugfs;
      if (!thermal_dbg)

mutex_unlock(&cdev->lock) missing here

Good catch, thanks!

Ho-hum, I'm not sure why I haven't added it here ...

I'll send a v2 of this patch shortly.

Quoted 18 lines
Quoted 17 lines
drivers/thermal/thermal_debugfs.c +4 -1
              return;
 
+     cdev->debugfs = NULL;
+
+     mutex_unlock(&cdev->lock);
+
      mutex_lock(&thermal_dbg->lock);
 
      thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
-     cdev->debugfs = NULL;
 
      mutex_unlock(&thermal_dbg->lock);
 
#9 lore raw

PATCH v2 3/3 thermal/debugfs: Prevent use-after-free from occurring after cdev removal

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Since thermal_debug_cdev_remove() does not run under cdev->lock, it can run in parallel with thermal_debug_cdev_state_update() and it may free the struct thermal_debugfs object used by the latter after it has been checked against NULL.

If that happens, thermal_debug_cdev_state_update() will access memory that has been freed already causing the kernel to crash.

Address this by using cdev->lock in thermal_debug_cdev_remove() around the cdev->debugfs value check (in case the same cdev is removed at the same time in two different threads) and its reset to NULL.

Fixes 755113d76786 ("thermal/debugfs: Add thermal cooling device debugfs information")
Cc 6.8+ <[email protected]> # 6.8+

---

v1 -> v2: Add missing mutex_unlock() (Lukasz).

---
 drivers/thermal/thermal_debugfs.c |   14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)
drivers/thermal/thermal_debugfs.c +11 -3
Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================
@@ -505,15 +505,23 @@ void thermal_debug_cdev_add(struct therm
  */
 void thermal_debug_cdev_remove(struct thermal_cooling_device *cdev)
 {
-	struct thermal_debugfs *thermal_dbg = cdev->debugfs;
+	struct thermal_debugfs *thermal_dbg;
 
-	if (!thermal_dbg)
+	mutex_lock(&cdev->lock);
+
+	thermal_dbg = cdev->debugfs;
+	if (!thermal_dbg) {
+		mutex_unlock(&cdev->lock);
 		return;
+	}
+
+	cdev->debugfs = NULL;
+
+	mutex_unlock(&cdev->lock);
 
 	mutex_lock(&thermal_dbg->lock);
 
 	thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
-	cdev->debugfs = NULL;
 
 	mutex_unlock(&thermal_dbg->lock);
 
reviewer Reviewed-by on v2 3/3
#10 lore raw

Re: [PATCH v2 3/3] thermal/debugfs: Prevent use-after-free from occurring after cdev removal

Quoted 50 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Since thermal_debug_cdev_remove() does not run under cdev->lock, it can run in parallel with thermal_debug_cdev_state_update() and it may free the struct thermal_debugfs object used by the latter after it has been checked against NULL.

If that happens, thermal_debug_cdev_state_update() will access memory that has been freed already causing the kernel to crash.

Address this by using cdev->lock in thermal_debug_cdev_remove() around the cdev->debugfs value check (in case the same cdev is removed at the same time in two different threads) and its reset to NULL.

Fixes 755113d76786 ("thermal/debugfs: Add thermal cooling device debugfs information")
Cc 6.8+ <[email protected]> # 6.8+

---

v1 -> v2: Add missing mutex_unlock() (Lukasz).

---
  drivers/thermal/thermal_debugfs.c |   14 +++++++++++---
  1 file changed, 11 insertions(+), 3 deletions(-)

Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================

drivers/thermal/thermal_debugfs.c +9 -2
@@ -505,15 +505,23 @@ void thermal_debug_cdev_add(struct therm
   */
  void thermal_debug_cdev_remove(struct thermal_cooling_device *cdev)
  {
-	struct thermal_debugfs *thermal_dbg = cdev->debugfs;
+	struct thermal_debugfs *thermal_dbg;
 
-	if (!thermal_dbg)
+	mutex_lock(&cdev->lock);
+
+	thermal_dbg = cdev->debugfs;
+	if (!thermal_dbg) {
+		mutex_unlock(&cdev->lock);
  		return;
+	}
+
+	cdev->debugfs = NULL;

drivers/thermal/thermal_debugfs.c void thermal_debug_cdev_add(struct therm · line 527

Quoted 10 lines
drivers/thermal/thermal_debugfs.c +2 -1
+
+	mutex_unlock(&cdev->lock);
 
  	mutex_lock(&thermal_dbg->lock);
 
  	thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
-	cdev->debugfs = NULL;
 
  	mutex_unlock(&thermal_dbg->lock);
 

It looks good now

#11 lore raw

Re: [PATCH v2 3/3] thermal/debugfs: Prevent use-after-free from occurring after cdev removal

Quoted 67 lines · Lukasz Luba
Quoted 50 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Since thermal_debug_cdev_remove() does not run under cdev->lock, it can run in parallel with thermal_debug_cdev_state_update() and it may free the struct thermal_debugfs object used by the latter after it has been checked against NULL.

If that happens, thermal_debug_cdev_state_update() will access memory that has been freed already causing the kernel to crash.

Address this by using cdev->lock in thermal_debug_cdev_remove() around the cdev->debugfs value check (in case the same cdev is removed at the same time in two different threads) and its reset to NULL.

Fixes 755113d76786 ("thermal/debugfs: Add thermal cooling device debugfs information")
Cc 6.8+ <[email protected]> # 6.8+

---

v1 -> v2: Add missing mutex_unlock() (Lukasz).

---
  drivers/thermal/thermal_debugfs.c |   14 +++++++++++---
  1 file changed, 11 insertions(+), 3 deletions(-)

Index: linux-pm/drivers/thermal/thermal_debugfs.c
===================================================================

drivers/thermal/thermal_debugfs.c +9 -2
@@ -505,15 +505,23 @@ void thermal_debug_cdev_add(struct therm
   */
  void thermal_debug_cdev_remove(struct thermal_cooling_device *cdev)
  {
-     struct thermal_debugfs *thermal_dbg = cdev->debugfs;
+     struct thermal_debugfs *thermal_dbg;
 
-     if (!thermal_dbg)
+     mutex_lock(&cdev->lock);
+
+     thermal_dbg = cdev->debugfs;
+     if (!thermal_dbg) {
+             mutex_unlock(&cdev->lock);
              return;
+     }
+
+     cdev->debugfs = NULL;

drivers/thermal/thermal_debugfs.c void thermal_debug_cdev_add(struct therm · line 527

Quoted 10 lines
drivers/thermal/thermal_debugfs.c +2 -1
+
+     mutex_unlock(&cdev->lock);
 
      mutex_lock(&thermal_dbg->lock);
 
      thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
-     cdev->debugfs = NULL;
 
      mutex_unlock(&thermal_dbg->lock);
 

It looks good now

Quoted 1 line

Thanks!