Bluetooth: MGMT: Fix mesh_tx Use-After-Free and leak in mesh_send()
Subsystems (2)
A message that is not in the archive (sent off-list or not yet received): [email protected]
PATCH v3 1/1 Bluetooth: MGMT: Fix mesh_tx Use-After-Free and leak in mesh_send()
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) ==================================================================
---
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(-)
net/bluetooth/mgmt.c +25 -8
@@ -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 );
Signature · 1 line
2.43.0
This is an automated email and please do not reply to this email.
Dear Submitter,
Thank you for submitting the patches to the linux bluetooth mailing list.
While preparing the CI tests, the patches you submitted couldn't be applied to the current HEAD of the repository.
----- Output -----
error: patch failed: net/bluetooth/mgmt.c:2306
error: net/bluetooth/mgmt.c: patch does not apply
hint: Use 'git am --show-current-patch=diff' to see the failed patch
hint: When you have resolved this problem, run "git am --continue".
hint: If you prefer to skip this patch, run "git am --skip" instead.
hint: To restore the original branch and stop patching, run "git am --abort".
hint: Disable this message with "git config set advice.mergeConflict false"
Please resolve the issue and submit the patches again.
---
Regards,
Linux Bluetooth
Please drop this patch, and sorry for the noise.
Two things are wrong with it.
First, the mechanical problem the CI already reported: I based it on Linus' tree rather than bluetooth-next, so it does not apply.
Second, and more to the point, the use-after-free half of it is no longer needed. 71af682ba469 ("Bluetooth: mgmt: Dequeue pending mesh_send_sync entries on cancel") fixed that on 2026-09-15, and fixed it better than this patch did. It dequeues the pending mesh_send_sync entry at the point where mesh_tx is freed, which addresses the problem at its source. What I had done instead was take hci_dev_lock() across mesh_send_sync() and re-validate mesh_tx against mgmt_mesh_next(), which only narrows the window for the two callers I happened to look at and would not help any future path that frees a queued mesh_tx. I should have checked bluetooth-next before sending.
What does survive is the other half: mesh_tx is leaked when hci_cmd_sync_queue() fails in mesh_send(), because the cleanup is guarded by "if (sending)" while that error can only occur when sending is false. That is still present in bluetooth-next today. I have sent it separately as a one-hunk patch against the right tree:
[PATCH] Bluetooth: MGMT: fix mesh_tx leak on hci_cmd_sync_queue() failure
Message-ID: <[email protected]>
Hui