Skip to contentSkip to conversation

Bluetooth: fix use-after-free in hci_conn_drop

PATCH In review Masahiro Kawada 3 participants 6 messages Feb 9 – Feb 10

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Bluetooth: fix use-after-free in hci_conn_drop

Fix a use-after-free in hci_conn_drop triggered via hci_cmd_sync_work.

In hci_conn_del(), hci_cmd_sync_dequeue() is called after hci_conn_cleanup() which may have already freed the conn pointer. Fix by moving the dequeue before cleanup.

Additionally, le_read_features_complete() calls hci_conn_drop(conn) without checking whether conn is still valid. When hci_le_read_remote_features_sync() blocks waiting for an HCI event, another thread can free conn through hci_conn_del(). Fix by adding a hci_conn_valid() check before calling hci_conn_drop().

Fixes 881559af5f5c ("Bluetooth: hci_sync: Attempt to dequeue connection attempt")
Fixes a106e50be74b ("Bluetooth: HCI: Add support for LL Extended Feature Set")
Reported-by [email protected]
---
 net/bluetooth/hci_conn.c | 6 +++---
 net/bluetooth/hci_sync.c | 3 +++
 2 files changed, 6 insertions(+), 3 deletions(-)
net/bluetooth/hci_conn.c +3 -3
index 0795818963a..aa3607327ad 100644
@@ -1232,15 +1232,15 @@ void hci_conn_del(struct hci_conn *conn)
 	skb_queue_purge(&conn->data_q);
 	skb_queue_purge(&conn->tx_q.queue);
 
+	/* Dequeue callbacks using connection pointer as data */
+	hci_cmd_sync_dequeue(hdev, NULL, conn, NULL);
+
 	/* Remove the connection from the list and cleanup its remaining
 	 * state. This is a separate function since for some cases like
 	 * BT_CONNECT_SCAN we *only* want the cleanup part without the
 	 * rest of hci_conn_del.
 	 */
 	hci_conn_cleanup(conn);
-
-	/* Dequeue callbacks using connection pointer as data */
-	hci_cmd_sync_dequeue(hdev, NULL, conn, NULL);
 }
 
 struct hci_dev *hci_get_route(bdaddr_t *dst, bdaddr_t *src, uint8_t src_type)
net/bluetooth/hci_sync.c +3 -0
index f04a90bce4a..f31086c187f 100644
@@ -7371,6 +7371,9 @@ static void le_read_features_complete(struct hci_dev *hdev, void *data, int err)
 	if (err == -ECANCELED)
 		return;
 
+	if (!hci_conn_valid(hdev, conn))
+		return;
+
 	hci_conn_drop(conn);
 }
 
Signature · 1 line
2.43.0
#1 lore raw

This is automated email and please do not reply to this email!

Dear submitter,

Thank you for submitting the patches to the linux bluetooth mailing list. This is a CI test results with your patch series:
PW Link:https://patchwork.kernel.org/project/bluetooth/list/?series=1052221

---Test result---

Test Summary:

CheckPatch                    PENDING   0.47 seconds
GitLint                       PENDING   0.33 seconds
SubjectPrefix                 PASS      0.06 seconds
BuildKernel                   PASS      25.77 seconds
CheckAllWarning               PASS      28.45 seconds
CheckSparse                   PASS      31.85 seconds
BuildKernel32                 PASS      24.99 seconds
TestRunnerSetup               PASS      560.46 seconds
TestRunner_l2cap-tester       PASS      28.11 seconds
TestRunner_iso-tester         PASS      91.87 seconds
TestRunner_bnep-tester        PASS      6.46 seconds
TestRunner_mgmt-tester        FAIL      117.55 seconds
TestRunner_rfcomm-tester      PASS      9.58 seconds
TestRunner_sco-tester         FAIL      14.80 seconds
TestRunner_ioctl-tester       PASS      10.10 seconds
TestRunner_mesh-tester        FAIL      12.52 seconds
TestRunner_smp-tester         PASS      8.69 seconds
TestRunner_userchan-tester    PASS      6.63 seconds
IncrementalBuild              PENDING   1.02 seconds

Details
##############################
Test: CheckPatch - PENDING
Desc: Run checkpatch.pl script
Output:

##############################
Test: GitLint - PENDING
Desc: Run gitlint
Output:

##############################
Test: TestRunner_mgmt-tester - FAIL
Desc: Run mgmt-tester with test-runner Output:
Total: 494, Passed: 489 (99.0%), Failed: 1, Not Run: 4

Failed Test Cases
Read Exp Feature - Success Failed 0.106 seconds
##############################
Test: TestRunner_sco-tester - FAIL
Desc: Run sco-tester with test-runner
Output:
WARNING: possible circular locking dependency detected

BUG: sleeping function called from invalid context at net/core/sock.c:3782

Total: 30, Passed: 30 (100.0%), Failed: 0, Not Run: 0
##############################
Test: TestRunner_mesh-tester - FAIL
Desc: Run mesh-tester with test-runner Output:
Total: 10, Passed: 8 (80.0%), Failed: 2, Not Run: 0

Failed Test Cases

Mesh - Send cancel - 1                               Timed out    2.774 seconds
Mesh - Send cancel - 2                               Timed out    1.992 seconds

##############################
Test: IncrementalBuild - PENDING
Desc: Incremental build with the patches in the series Output:

---
Regards,
Linux Bluetooth

#2 lore raw
In reply to Masahiro Kawada
Quoted 63 lines · On Mon, 9 Feb 2026 19:02:11 +0900 Masahiro Kawada wrote:

Fix a use-after-free in hci_conn_drop triggered via hci_cmd_sync_work.

In hci_conn_del(), hci_cmd_sync_dequeue() is called after hci_conn_cleanup() which may have already freed the conn pointer. Fix by moving the dequeue before cleanup.

Additionally, le_read_features_complete() calls hci_conn_drop(conn) without checking whether conn is still valid. When hci_le_read_remote_features_sync() blocks waiting for an HCI event, another thread can free conn through hci_conn_del(). Fix by adding a hci_conn_valid() check before calling hci_conn_drop().

Fixes 881559af5f5c ("Bluetooth: hci_sync: Attempt to dequeue connection attempt")
Fixes a106e50be74b ("Bluetooth: HCI: Add support for LL Extended Feature Set")
Reported-by [email protected]
---
 net/bluetooth/hci_conn.c | 6 +++---
 net/bluetooth/hci_sync.c | 3 +++
 2 files changed, 6 insertions(+), 3 deletions(-)
net/bluetooth/hci_conn.c +3 -3
index 0795818963a..aa3607327ad 100644
@@ -1232,15 +1232,15 @@ void hci_conn_del(struct hci_conn *conn)
 	skb_queue_purge(&conn->data_q);
 	skb_queue_purge(&conn->tx_q.queue);
 
+	/* Dequeue callbacks using connection pointer as data */
+	hci_cmd_sync_dequeue(hdev, NULL, conn, NULL);
+
 	/* Remove the connection from the list and cleanup its remaining
 	 * state. This is a separate function since for some cases like
 	 * BT_CONNECT_SCAN we *only* want the cleanup part without the
 	 * rest of hci_conn_del.
 	 */
 	hci_conn_cleanup(conn);
-
-	/* Dequeue callbacks using connection pointer as data */
-	hci_cmd_sync_dequeue(hdev, NULL, conn, NULL);
 }
 
 struct hci_dev *hci_get_route(bdaddr_t *dst, bdaddr_t *src, uint8_t src_type)
net/bluetooth/hci_sync.c +3 -0
index f04a90bce4a..f31086c187f 100644
@@ -7371,6 +7371,9 @@ static void le_read_features_complete(struct hci_dev *hdev, void *data, int err)
 	if (err == -ECANCELED)
 		return;
 
+	if (!hci_conn_valid(hdev, conn))
+		return;
+
 	hci_conn_drop(conn);
 }
 
Signature · 1 line
2.43.0

The uaf [1] is due to the following race,

	cpu1				cpu2
	hci_cmd_sync_work()		hci_rx_work()
	mutex_lock(&hdev->cmd_sync_work_lock);
	entry = list_first_entry_or_null(&hdev->cmd_sync_work_list,
					 struct hci_cmd_sync_work_entry,
					 list);
	if (entry)
		list_del(&entry->list);
	mutex_unlock(&hdev->cmd_sync_work_lock);

					hci_conn_del()
					hci_conn_hash_del(hdev, conn);
					hci_conn_cleanup(conn)	// free conn
					hci_cmd_sync_dequeue()
					mutex_lock(&hdev->cmd_sync_work_lock);
					while ((entry = _hci_cmd_sync_lookup_entry(hdev, func, data,
						   destroy))) {
						_hci_cmd_sync_cancel_entry(hdev, entry, -ECANCELED);
						ret = true;
					}
					mutex_unlock(&hdev->cmd_sync_work_lock);

	hci_req_sync_lock(hdev);
	err = entry->func(hdev, entry->data);
	if (entry->destroy)
		entry->destroy(hdev, entry->data, err);
		hci_conn_drop(conn)	// uaf
	hci_req_sync_unlock(hdev);

but the race still exists after this patch.

	cpu1				cpu2
	hci_conn_valid(hdev, conn)
					hci_conn_hash_del(hdev, conn);
					hci_cmd_sync_dequeue()
					hci_conn_cleanup(conn)	// free conn
	hci_conn_drop(conn); // uaf

[1] Subject: [syzbot] [bluetooth?] KASAN: slab-use-after-free Write in hci_conn_drop (3)
https://lore.kernel.org/lkml/[email protected]/

author
#3 lore raw
Quoted 1 line

but the race still exists after this patch.

I agree that the TOCTOU issue remains. I looked into all the functions that can be called as entry->destroy in hci_cmd_sync_work where the second argument (data) is used as hci_conn*. I found five such functions across two files:

In hci_sync.c:
- create_le_conn_complete
- create_pa_complete
- create_big_complete
- le_read_features_complete

In hci_conn.c:
- create_big_complete (separate static function)

Of these, create_le_conn_complete and create_pa_complete already use hci_dev_lock, which effectively prevents this TOCTOU issue.

The remaining three are vulnerable:
- create_big_complete in hci_sync.c calls hci_conn_valid without holding hci_dev_lock (TOCTOU)
- le_read_features_complete calls hci_conn_drop with no validity check at all
- create_big_complete in hci_conn.c calls hci_connect_cfm and hci_conn_del with no validity check at all

Given this, I believe the following set of patches would be appropriate:

1. Reorder hci_cmd_sync_dequeue before hci_conn_cleanup in hci_conn_del
2. Wrap all three vulnerable callbacks with hci_dev_lock and hci_conn_valid, following the same pattern used by create_le_conn_complete and create_pa_complete

If this approach sounds reasonable, I will prepare updated patches.

2026年2月9日(月) 22:25 Hillf Danton <[email protected]>:

Quoted 107 lines · On Mon, 9 Feb 2026 19:02:11 +0900 Masahiro Kawada wrote:
Quoted 61 lines

Fix a use-after-free in hci_conn_drop triggered via hci_cmd_sync_work.

In hci_conn_del(), hci_cmd_sync_dequeue() is called after hci_conn_cleanup() which may have already freed the conn pointer. Fix by moving the dequeue before cleanup.

Additionally, le_read_features_complete() calls hci_conn_drop(conn) without checking whether conn is still valid. When hci_le_read_remote_features_sync() blocks waiting for an HCI event, another thread can free conn through hci_conn_del(). Fix by adding a hci_conn_valid() check before calling hci_conn_drop().

Fixes 881559af5f5c ("Bluetooth: hci_sync: Attempt to dequeue connection attempt")
Fixes a106e50be74b ("Bluetooth: HCI: Add support for LL Extended Feature Set")
Reported-by [email protected]
---
 net/bluetooth/hci_conn.c | 6 +++---
 net/bluetooth/hci_sync.c | 3 +++
 2 files changed, 6 insertions(+), 3 deletions(-)
net/bluetooth/hci_conn.c +3 -3
index 0795818963a..aa3607327ad 100644
@@ -1232,15 +1232,15 @@ void hci_conn_del(struct hci_conn *conn)
      skb_queue_purge(&conn->data_q);
      skb_queue_purge(&conn->tx_q.queue);
 
+     /* Dequeue callbacks using connection pointer as data */
+     hci_cmd_sync_dequeue(hdev, NULL, conn, NULL);
+
      /* Remove the connection from the list and cleanup its remaining
       * state. This is a separate function since for some cases like
       * BT_CONNECT_SCAN we *only* want the cleanup part without the
       * rest of hci_conn_del.
       */
      hci_conn_cleanup(conn);
-
-     /* Dequeue callbacks using connection pointer as data */
-     hci_cmd_sync_dequeue(hdev, NULL, conn, NULL);
 }
 
 struct hci_dev *hci_get_route(bdaddr_t *dst, bdaddr_t *src, uint8_t src_type)
net/bluetooth/hci_sync.c +3 -1
index f04a90bce4a..f31086c187f 100644
@@ -7371,6 +7371,9 @@ static void le_read_features_complete(struct hci_dev *hdev, void *data, int err)
      if (err == -ECANCELED)
              return;
 
+     if (!hci_conn_valid(hdev, conn))
+             return;
+
      hci_conn_drop(conn);
 }
 
--
Quoted 2 lines

2.43.0

The uaf [1] is due to the following race,

        cpu1                            cpu2
        hci_cmd_sync_work()             hci_rx_work()
        mutex_lock(&hdev->cmd_sync_work_lock);
        entry = list_first_entry_or_null(&hdev->cmd_sync_work_list,
                                         struct hci_cmd_sync_work_entry,
                                         list);
        if (entry)
                list_del(&entry->list);
        mutex_unlock(&hdev->cmd_sync_work_lock);

                                        hci_conn_del()
                                        hci_conn_hash_del(hdev, conn);
                                        hci_conn_cleanup(conn)  // free conn
                                        hci_cmd_sync_dequeue()
                                        mutex_lock(&hdev->cmd_sync_work_lock);
                                        while ((entry = _hci_cmd_sync_lookup_entry(hdev, func, data,
                                                   destroy))) {
                                                _hci_cmd_sync_cancel_entry(hdev, entry, -ECANCELED);
                                                ret = true;
                                        }
                                        mutex_unlock(&hdev->cmd_sync_work_lock);

        hci_req_sync_lock(hdev);
        err = entry->func(hdev, entry->data);
        if (entry->destroy)
                entry->destroy(hdev, entry->data, err);
                hci_conn_drop(conn)     // uaf
        hci_req_sync_unlock(hdev);

but the race still exists after this patch.

        cpu1                            cpu2
        hci_conn_valid(hdev, conn)
                                        hci_conn_hash_del(hdev, conn);
                                        hci_cmd_sync_dequeue()
                                        hci_conn_cleanup(conn)  // free conn
        hci_conn_drop(conn); // uaf

[1] Subject: [syzbot] [bluetooth?] KASAN: slab-use-after-free Write in hci_conn_drop (3)
https://lore.kernel.org/lkml/[email protected]/

#4 lore raw

[ hm... top reply looks no good ]

Quoted 30 lines · On Tue, 10 Feb 2026 15:08:12 +0900 Masahiro Kawada wrote:
Quoted 1 line

but the race still exists after this patch.

I agree that the TOCTOU issue remains. I looked into all the functions that can be called as entry->destroy in hci_cmd_sync_work where the second argument (data) is used as hci_conn*. I found five such functions across two files:

In hci_sync.c:
- create_le_conn_complete
- create_pa_complete
- create_big_complete
- le_read_features_complete

In hci_conn.c:
- create_big_complete (separate static function)

Of these, create_le_conn_complete and create_pa_complete already use hci_dev_lock, which effectively prevents this TOCTOU issue.

The remaining three are vulnerable:
- create_big_complete in hci_sync.c calls hci_conn_valid without holding hci_dev_lock (TOCTOU)
- le_read_features_complete calls hci_conn_drop with no validity check at all
- create_big_complete in hci_conn.c calls hci_connect_cfm and hci_conn_del with no validity check at all

Given this, I believe the following set of patches would be appropriate:

Quoted 8 lines

1. Reorder hci_cmd_sync_dequeue before hci_conn_cleanup in hci_conn_del
2. Wrap all three vulnerable callbacks with hci_dev_lock and hci_conn_valid, following the same pattern used by create_le_conn_complete and create_pa_complete

If this approach sounds reasonable, I will prepare updated patches.

Better not before spotting the reason why conn->refcnt failed to make the entry->destroy callback safe, given the complexity of the race.

author
#5 lore raw

Apologies for the top-posting.

Quoted 2 lines

Better not before spotting the reason why conn->refcnt failed to make the entry->destroy callback safe.

hci_conn_hold() increments conn->refcnt, but the actual freeing of conn is done by put_device() called from hci_conn_del(). hci_conn_del() does not check conn->refcnt at all, so the hold/drop refcount cannot prevent the UAF in entry->destroy callbacks.