[PATCH v3] Bluetooth: MGMT: Fix mesh_tx Use-After-Free and leak in mesh_send()

[Hui Peng]

In mesh_send(), a struct mgmt_mesh_tx entry is allocated and added to
hdev->mesh_pending via mgmt_mesh_add(sk, hdev, send, len) before queuing
mesh_send_sync() and mesh_send_start_complete() onto hdev->req_workqueue
with mesh_tx as the callback data pointer.

Before or while the workqueue executes mesh_send_sync() and
mesh_send_start_complete(), mesh_tx can be removed from
hdev->mesh_pending and freed via mgmt_mesh_remove():
1. If an earlier MGMT_OP_MESH_SEND_CANCEL (send_cancel()) item runs on
   hdev->req_workqueue, send_cancel() calls mesh_send_complete(), which
   removes and frees mesh_tx.
2. If the management socket is closed, mgmt_cleanup() removes and frees
   all mesh_tx entries matching sk under hci_dev_lock(hdev).

When mesh_send_sync() and mesh_send_start_complete() subsequently run,
they dereference the dangling mesh_tx pointer (mesh_tx->param,
mesh_tx->handle, mesh_tx->instance), and mesh_send_start_complete()
frees mesh_tx a second time via mesh_send_complete() on error.

Furthermore, if hci_cmd_sync_queue() fails in mesh_send() (which can
only happen in the !sending branch), mesh_send() only called
mgmt_mesh_remove(mesh_tx) when if (sending) was true (an inverted
condition), leaking mesh_tx on hdev->mesh_pending.

Fix this by holding hci_dev_lock(hdev) and verifying
mesh_tx == mgmt_mesh_next(hdev, NULL) across all mesh_tx and send
dereferences in mesh_send_sync() and mesh_send_start_complete(), and
unconditionally calling mgmt_mesh_remove(mesh_tx) on
hci_cmd_sync_queue() error in mesh_send().

Kernel stack trace (Linux 7.3.0-rc3):
 ==================================================================
 BUG: KASAN: slab-use-after-free in mesh_send_sync+0x39a/0x410
 Read of size 1 at addr ffff8880132fe53b by task kworker/u9:1/143

 CPU: 1 UID: 0 PID: 143 Comm: kworker/u9:1 Not tainted 7.3.0-rc3-g5dd1818b15d9 #1 PREEMPT(lazy)
 Workqueue: hci0 hci_cmd_sync_work
 Call Trace:
  <TASK>
  dump_stack_lvl+0x70/0xa0
  print_report+0x153/0x4c6
  kasan_report+0xf1/0x120
  mesh_send_sync+0x39a/0x410
  hci_cmd_sync_work+0x14e/0x2a0
  process_one_work+0x6ff/0x1110
  worker_thread+0x4a8/0xb70
  kthread+0x307/0x3e0
  ret_from_fork+0x3ed/0x680
  ret_from_fork_asm+0x1a/0x30
  </TASK>

 Allocated by task 1:
  kasan_save_stack+0x30/0x50
  kasan_save_track+0x14/0x30
  __kasan_kmalloc+0x7f/0x90
  __kmalloc_cache_noprof+0x16a/0x380
  mgmt_mesh_add+0x5b/0x340
  mesh_send+0x22b/0x690
  hci_sock_sendmsg+0x1155/0x2040
  sock_write_iter+0x492/0x520
  vfs_write+0x671/0xd20
  ksys_write+0x1bb/0x210
  do_syscall_64+0xda/0x4b0
  entry_SYSCALL_64_after_hwframe+0x77/0x7f

 Freed by task 143:
  kasan_save_stack+0x30/0x50
  kasan_save_track+0x14/0x30
  kasan_save_free_info+0x3b/0x70
  __kasan_slab_free+0x47/0x70
  kfree+0x159/0x420
  send_cancel+0x14e/0x360
  hci_cmd_sync_work+0x14e/0x2a0
  process_one_work+0x6ff/0x1110
  worker_thread+0x4a8/0xb70
  kthread+0x307/0x3e0
  ret_from_fork+0x3ed/0x680
  ret_from_fork_asm+0x1a/0x30

 The buggy address belongs to the object at 0xffff8880132fe500
  which belongs to the cache kmalloc-96 of size 96
 The buggy address is located 59 bytes inside of
  freed 96-byte region [ffff8880132fe500, ffff8880132fe560)
 ==================================================================

Fixes: b338d91703fa ("Bluetooth: Implement support for Mesh")
Assisted-by: LLM
Signed-off-by: Hui Peng <benquike@xxxxxxxxx>
---
v3: Add a Fixes: tag. The allocator (mgmt_mesh_add() in mesh_send()),
the freeing path (mesh_send_complete() from send_cancel()), the racing
dereference (mesh_send_sync()), the unlocked hci_add_adv_instance()
call and the missing mgmt_mesh_remove() on the !sending path all came
in together with b338d91703fa ("Bluetooth: Implement support for
Mesh"), first released in v6.1. No later commit touched any of the
lines this patch changes.

This patch fixes two things that share that one origin: the
UAF/locking race, and the leak when the send is not queued. Happy to
split it into two patches if you prefer - both halves would carry the
same tag.

Backporting: the hunks apply back to v6.1, but mesh_send() picked up
extra validation in bda93eec78cd, so older stable branches may need a
trivial context fixup.

v2: Resend via git send-email with intact tab formatting and Assisted-by: LLM tag.
Reproducer (triggers KASAN slab-use-after-free in mesh_send_sync() on
Linux 7.3.0-rc3 via /dev/vhci and MGMT_OP_MESH_SEND_CANCEL +
MGMT_OP_MESH_SEND):

  /* 1. Enable Mesh experimental feature (UUID 766ef3e8-245f-05bf-8d4d-037ad763e42c)
   *    and LE on hci0 via MGMT_OP_SET_EXP_FEATURE (0x004a) and MGMT_OP_SET_LE (0x000d).
   * 2. Hold hdev->req_workqueue briefly in set_name_sync (MGMT_OP_SET_LOCAL_NAME 0x000f).
   * 3. Queue MGMT_OP_MESH_SEND_CANCEL (0x005a, handle = 0) followed immediately by
   *    MGMT_OP_MESH_SEND (0x0059) on a second MGMT control socket.
   * 4. MGMT_OP_MESH_SEND synchronously adds mesh_tx to hdev->mesh_pending (assigning
   *    handle = 1, or matching handle = 0 cancel) and queues mesh_send_sync(mesh_tx).
   *    When send_cancel() runs first on hdev->req_workqueue, it frees mesh_tx, and
   *    mesh_send_sync() immediately dereferences the freed mesh_tx. */

 net/bluetooth/mgmt.c | 25 ++++++++++++++++++-------
 1 file changed, 18 insertions(+), 7 deletions(-)

diff --git a/net/bluetooth/mgmt.c b/net/bluetooth/mgmt.c
index ac4864e56..fbf85e791 100644
--- a/net/bluetooth/mgmt.c
+++ b/net/bluetooth/mgmt.c
@@ -2306,36 +2306,53 @@ static int set_mesh(struct sock *sk, struct hci_dev *hdev, void *data, u16 len)
 static void mesh_send_start_complete(struct hci_dev *hdev, void *data, int err)
 {
 	struct mgmt_mesh_tx *mesh_tx = data;
-	struct mgmt_cp_mesh_send *send = (void *)mesh_tx->param;
+	struct mgmt_cp_mesh_send *send;
 	unsigned long mesh_send_interval;
 	u8 mgmt_err = mgmt_status(err);
 
-	/* Report any errors here, but don't report completion */
+	hci_dev_lock(hdev);
+	if (mesh_tx != mgmt_mesh_next(hdev, NULL)) {
+		hci_dev_unlock(hdev);
+		return;
+	}
 
+	/* Report any errors here, but don't report completion */
 	if (mgmt_err) {
 		hci_dev_clear_flag(hdev, HCI_MESH_SENDING);
 		/* Send Complete Error Code for handle */
 		mesh_send_complete(hdev, mesh_tx, false);
+		hci_dev_unlock(hdev);
 		return;
 	}
 
+	send = (void *)mesh_tx->param;
 	mesh_send_interval = msecs_to_jiffies((send->cnt) * 25);
 	queue_delayed_work(hdev->req_workqueue, &hdev->mesh_send_done,
 			   mesh_send_interval);
+	hci_dev_unlock(hdev);
 }
 
 static int mesh_send_sync(struct hci_dev *hdev, void *data)
 {
 	struct mgmt_mesh_tx *mesh_tx = data;
-	struct mgmt_cp_mesh_send *send = (void *)mesh_tx->param;
+	struct mgmt_cp_mesh_send *send;
 	struct adv_info *adv, *next_instance;
 	u8 instance = hdev->le_num_of_adv_sets + 1;
 	u16 timeout, duration;
 	int err = 0;
 
-	if (hdev->le_num_of_adv_sets <= hdev->adv_instance_cnt)
+	hci_dev_lock(hdev);
+	if (mesh_tx != mgmt_mesh_next(hdev, NULL)) {
+		hci_dev_unlock(hdev);
+		return MGMT_STATUS_FAILED;
+	}
+
+	if (hdev->le_num_of_adv_sets <= hdev->adv_instance_cnt) {
+		hci_dev_unlock(hdev);
 		return MGMT_STATUS_BUSY;
+	}
 
+	send = (void *)mesh_tx->param;
 	timeout = 1000;
 	duration = send->cnt * INTERVAL_TO_MS(hdev->le_adv_max_interval);
 	adv = hci_add_adv_instance(hdev, instance, 0,
@@ -2372,6 +2389,8 @@ static int mesh_send_sync(struct hci_dev *hdev, void *data)
 		instance = 0;
 	}
 
+	hci_dev_unlock(hdev);
+
 	if (instance)
 		return hci_schedule_adv_instance_sync(hdev, instance, true);
 
@@ -2534,10 +2553,8 @@ static int mesh_send(struct sock *sk, struct hci_dev *hdev, void *data, u16 len)
 		err = mgmt_cmd_status(sk, hdev->id, MGMT_OP_MESH_SEND,
 				      MGMT_STATUS_FAILED);
 
-		if (mesh_tx) {
-			if (sending)
-				mgmt_mesh_remove(mesh_tx);
-		}
+		if (mesh_tx)
+			mgmt_mesh_remove(mesh_tx);
 	} else {
 		hci_dev_set_flag(hdev, HCI_MESH_SENDING);
 
-- 
2.43.0