* RE: [v4] Bluetooth: l2cap: defer conn param update to avoid conn->lock/hdev->lock inversion
2026-04-14 18:43 [PATCH v4] Bluetooth: l2cap: defer conn param update to avoid conn->lock/hdev->lock inversion Mikhail Gavrilov
@ 2026-04-14 19:18 ` bluez.test.bot
2026-04-14 20:26 ` [PATCH v4] " Pauli Virtanen
2026-04-14 20:27 ` Luiz Augusto von Dentz
2 siblings, 0 replies; 5+ messages in thread
From: bluez.test.bot @ 2026-04-14 19:18 UTC (permalink / raw)
To: linux-bluetooth, mikhail.v.gavrilov
[-- Attachment #1: Type: text/plain, Size: 7456 bytes --]
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=1081253
---Test result---
Test Summary:
CheckPatch PASS 1.14 seconds
GitLint FAIL 0.29 seconds
SubjectPrefix PASS 0.10 seconds
BuildKernel PASS 26.88 seconds
CheckAllWarning PASS 29.13 seconds
CheckSparse PASS 27.61 seconds
BuildKernel32 PASS 25.61 seconds
TestRunnerSetup PASS 579.82 seconds
TestRunner_l2cap-tester PASS 29.45 seconds
TestRunner_iso-tester PASS 44.08 seconds
TestRunner_bnep-tester PASS 6.56 seconds
TestRunner_mgmt-tester FAIL 120.96 seconds
TestRunner_rfcomm-tester PASS 9.68 seconds
TestRunner_sco-tester FAIL 14.47 seconds
TestRunner_ioctl-tester PASS 10.66 seconds
TestRunner_mesh-tester FAIL 12.21 seconds
TestRunner_smp-tester PASS 8.79 seconds
TestRunner_userchan-tester PASS 6.89 seconds
TestRunner_6lowpan-tester FAIL 8.80 seconds
IncrementalBuild FAIL 0.00 seconds
Details
##############################
Test: GitLint - FAIL
Desc: Run gitlint
Output:
[v4] Bluetooth: l2cap: defer conn param update to avoid conn->lock/hdev->lock inversion
WARNING: I3 - ignore-body-lines: gitlint will be switching from using Python regex 'match' (match beginning) to 'search' (match anywhere) semantics. Please review your ignore-body-lines.regex option accordingly. To remove this warning, set general.regex-style-search=True. More details: https://jorisroovers.github.io/gitlint/configuration/#regex-style-search
1: T1 Title exceeds max length (87>80): "[v4] Bluetooth: l2cap: defer conn param update to avoid conn->lock/hdev->lock inversion"
47: B2 Line has trailing whitespace: " "
54: B2 Line has trailing whitespace: " "
##############################
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 0.120 seconds
##############################
Test: TestRunner_sco-tester - FAIL
Desc: Run sco-tester with test-runner
Output:
WARNING: possible circular locking dependency detected
7.0.0-rc2-g40cbfa4250a5 #1 Not tainted
------------------------------------------------------
kworker/u5:2/117 is trying to acquire lock:
ffff888001abd240 (sk_lock-AF_BLUETOOTH-BTPROTO_SCO){+.+.}-{0:0}, at: sco_connect_cfm+0x358/0x8d0
but task is already holding lock:
ffff8880025f7220 (&conn->lock){+.+.}-{3:3}, at: sco_connect_cfm+0x22d/0x8d0
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (&conn->lock){+.+.}-{3:3}:
lock_acquire+0xf7/0x2c0
_raw_spin_lock+0x2a/0x40
sco_sock_connect+0x4d7/0x1280
__sys_connect+0x1a3/0x260
__x64_sys_connect+0x6e/0xb0
do_syscall_64+0xa0/0x570
entry_SYSCALL_64_after_hwframe+0x74/0x7c
-> #0 (sk_lock-AF_BLUETOOTH-BTPROTO_SCO){+.+.}-{0:0}:
check_prev_add+0xe9/0xc70
__lock_acquire+0x1457/0x1df0
lock_acquire+0xf7/0x2c0
lock_sock_nested+0x36/0xd0
sco_connect_cfm+0x358/0x8d0
hci_sync_conn_complete_evt+0x3d3/0x8e0
hci_event_packet+0x74f/0xb10
hci_rx_work+0x398/0xd00
process_scheduled_works+0xb16/0x1ac0
worker_thread+0x4ff/0xba0
kthread+0x368/0x490
ret_from_fork+0x498/0x7e0
ret_from_fork_asm+0x19/0x30
other info that might help us debug this:
...
BUG: sleeping function called from invalid context at net/core/sock.c:3782
in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 117, name: kworker/u5:2
preempt_count: 1, expected: 0
RCU nest depth: 0, expected: 0
INFO: lockdep is turned off.
CPU: 0 UID: 0 PID: 117 Comm: kworker/u5:2 Not tainted 7.0.0-rc2-g40cbfa4250a5 #1 PREEMPT(lazy)
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.13.0-1ubuntu1.1 04/01/2014
Workqueue: hci0 hci_rx_work
Call Trace:
<TASK>
dump_stack_lvl+0x49/0x60
__might_resched+0x2ea/0x500
lock_sock_nested+0x47/0xd0
? sco_connect_cfm+0x358/0x8d0
sco_connect_cfm+0x358/0x8d0
? hci_debugfs_create_conn+0x190/0x210
? __pfx_sco_connect_cfm+0x10/0x10
hci_sync_conn_complete_evt+0x3d3/0x8e0
hci_event_packet+0x74f/0xb10
? __pfx_hci_sync_conn_complete_evt+0x10/0x10
? __pfx_hci_event_packet+0x10/0x10
? mark_held_locks+0x49/0x80
? lockdep_hardirqs_on_prepare+0xd4/0x180
? _raw_spin_unlock_irqrestore+0x2c/0x50
hci_rx_work+0x398/0xd00
process_scheduled_works+0xb16/0x1ac0
? __pfx_process_scheduled_works+0x10/0x10
? lock_acquire+0xf7/0x2c0
? lock_is_held_type+0x9b/0x110
? __pfx_hci_rx_work+0x10/0x10
worker_thread+0x4ff/0xba0
? _raw_spin_unlock_irqrestore+0x2c/0x50
? __pfx_worker_thread+0x10/0x10
kthread+0x368/0x490
? _raw_spin_unlock_irq+0x23/0x40
? __pfx_kthread+0x10/0x10
ret_from_fork+0x498/0x7e0
? __pfx_ret_from_fork+0x10/0x10
? __switch_to+0x9e4/0xe50
? __switch_to_asm+0x32/0x60
...
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 2.631 seconds
Mesh - Send cancel - 2 Timed out 1.993 seconds
##############################
Test: TestRunner_6lowpan-tester - FAIL
Desc: Run 6lowpan-tester with test-runner
Output:
WARNING: possible circular locking dependency detected
7.0.0-rc2-g40cbfa4250a5 #1 Not tainted
------------------------------------------------------
kworker/0:1/11 is trying to acquire lock:
ffff88800275c940 ((wq_completion)hci0#2){+.+.}-{0:0}, at: touch_wq_lockdep_map+0x75/0x180
but task is already holding lock:
ffffffffaac4d720 (rtnl_mutex){+.+.}-{4:4}, at: lowpan_unregister_netdev+0xd/0x30
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #4 (rtnl_mutex){+.+.}-{4:4}:
lock_acquire+0xf7/0x2c0
__mutex_lock+0x16b/0x1fc0
lowpan_register_netdev+0x11/0x30
chan_ready_cb+0x836/0xd00
l2cap_recv_frame+0x6a07/0x88a0
l2cap_recv_acldata+0x790/0xdf0
hci_rx_work+0x500/0xd00
process_scheduled_works+0xb16/0x1ac0
worker_thread+0x4ff/0xba0
kthread+0x368/0x490
ret_from_fork+0x498/0x7e0
ret_from_fork_asm+0x19/0x30
-> #3 (&chan->lock#3/1){+.+.}-{4:4}:
lock_acquire+0xf7/0x2c0
__mutex_lock+0x16b/0x1fc0
l2cap_chan_connect+0x74e/0x1980
lowpan_control_write+0x523/0x660
full_proxy_write+0x10b/0x190
vfs_write+0x1c0/0xf60
ksys_write+0xf1/0x1d0
do_syscall_64+0xa0/0x570
entry_SYSCALL_64_after_hwframe+0x74/0x7c
-> #2 (&conn->lock){+.+.}-{4:4}:
...
Total: 8, Passed: 8 (100.0%), Failed: 0, Not Run: 0
##############################
Test: IncrementalBuild - FAIL
Desc: Incremental build with the patches in the series
Output:
Failed to reset to base commit
https://github.com/bluez/bluetooth-next/pull/83
---
Regards,
Linux Bluetooth
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH v4] Bluetooth: l2cap: defer conn param update to avoid conn->lock/hdev->lock inversion
2026-04-14 18:43 [PATCH v4] Bluetooth: l2cap: defer conn param update to avoid conn->lock/hdev->lock inversion Mikhail Gavrilov
2026-04-14 19:18 ` [v4] " bluez.test.bot
@ 2026-04-14 20:26 ` Pauli Virtanen
2026-04-14 20:30 ` Luiz Augusto von Dentz
2026-04-14 20:27 ` Luiz Augusto von Dentz
2 siblings, 1 reply; 5+ messages in thread
From: Pauli Virtanen @ 2026-04-14 20:26 UTC (permalink / raw)
To: Mikhail Gavrilov, luiz.dentz, marcel
Cc: pmenzel, linux-bluetooth, linux-kernel
ti, 2026-04-14 kello 23:43 +0500, Mikhail Gavrilov kirjoitti:
> When a BLE peripheral sends an L2CAP Connection Parameter Update Request
> the processing path is:
>
> process_pending_rx() [takes conn->lock]
> l2cap_le_sig_channel()
> l2cap_conn_param_update_req()
> hci_le_conn_update() [takes hdev->lock]
>
> Meanwhile other code paths take the locks in the opposite order:
>
> l2cap_chan_connect() [takes hdev->lock]
> ...
> mutex_lock(&conn->lock)
>
> l2cap_conn_ready() [hdev->lock via hci_cb_list_lock]
> ...
> mutex_lock(&conn->lock)
>
> This is a classic AB/BA deadlock which lockdep reports as a circular
> locking dependency when connecting a BLE MIDI keyboard (Carry-On FC-49).
>
> Fix this by making hci_le_conn_update() defer the HCI command through
> hci_cmd_sync_queue() so it no longer needs to take hdev->lock in the
> caller context. The sync callback uses __hci_cmd_sync_status_sk() to
> wait for the HCI_EV_LE_CONN_UPDATE_COMPLETE event, then updates the
> stored connection parameters (hci_conn_params) and notifies userspace
> (mgmt_new_conn_param) only after the controller has confirmed the update.
>
> A reference on hci_conn is held via hci_conn_get()/hci_conn_put() for
> the lifetime of the queued work to prevent use-after-free, and
> hci_conn_valid() is checked before proceeding in case the connection was
> removed while the work was pending.
>
> Fixes: f044eb0524a0 ("Bluetooth: Store latency and supervision timeout in connection params")
> Signed-off-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
> Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
> ---
>
> Changes in v4 (Luiz Augusto von Dentz, Sashiko/Gemini AI review):
> - Use hci_conn_get()/hci_conn_put() to hold a reference while work is queued
> - Use __hci_cmd_sync_status_sk() to wait for HCI_EV_LE_CONN_UPDATE_COMPLETE,
> then do params update + mgmt notification in the sync callback — removes the
> need for hci_event.c changes and hci_sent_cmd_data()
> - Use kzalloc_obj() per checkpatch recommendation
> - Fix checkpatch alignment issues
>
> Changes in v3 (Luiz Augusto von Dentz):
> - Move hci_cmd_sync_queue into hci_le_conn_update itself instead of open-coding
> the deferral in l2cap_core.c
> - Move conn_params update and mgmt_new_conn_param into
> hci_le_conn_update_complete_evt, using hci_sent_cmd_data to retrieve
> the originally requested parameters
>
> Changes in v2 (Paul Menzel, Sashiko/Gemini AI review):
> - Allocate before sending ACCEPTED response to avoid state mismatch on OOM
> - Verify connection handle and address in sync callback against reuse race
> - Expand commit message with implementation details
>
> include/net/bluetooth/hci_core.h | 2 +-
> net/bluetooth/hci_conn.c | 98 +++++++++++++++++++++++++-------
> net/bluetooth/l2cap_core.c | 12 +---
> 3 files changed, 82 insertions(+), 30 deletions(-)
>
> diff --git a/include/net/bluetooth/hci_core.h b/include/net/bluetooth/hci_core.h
> index a7bffb908c1e..aa600fbf9a53 100644
> --- a/include/net/bluetooth/hci_core.h
> +++ b/include/net/bluetooth/hci_core.h
> @@ -2495,7 +2495,7 @@ void mgmt_adv_monitor_device_lost(struct hci_dev *hdev, u16 handle,
> bdaddr_t *bdaddr, u8 addr_type);
>
> int hci_abort_conn(struct hci_conn *conn, u8 reason);
> -u8 hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> +void hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> u16 to_multiplier);
> void hci_le_start_enc(struct hci_conn *conn, __le16 ediv, __le64 rand,
> __u8 ltk[16], __u8 key_size);
> diff --git a/net/bluetooth/hci_conn.c b/net/bluetooth/hci_conn.c
> index 11d3ad8d2551..c9cc18fab1f2 100644
> --- a/net/bluetooth/hci_conn.c
> +++ b/net/bluetooth/hci_conn.c
> @@ -480,40 +480,100 @@ bool hci_setup_sync(struct hci_conn *conn, __u16 handle)
> return hci_setup_sync_conn(conn, handle);
> }
>
> -u8 hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> - u16 to_multiplier)
> +struct le_conn_update_data {
> + struct hci_conn *conn;
> + u16 min;
> + u16 max;
> + u16 latency;
> + u16 to_multiplier;
> +};
> +
> +static int le_conn_update_sync(struct hci_dev *hdev, void *data)
> {
> - struct hci_dev *hdev = conn->hdev;
> + struct le_conn_update_data *d = data;
> + struct hci_conn *conn = d->conn;
> struct hci_conn_params *params;
> struct hci_cp_le_conn_update cp;
> + u8 store_hint;
> + int err;
>
> + /* Verify connection is still alive. */
> hci_dev_lock(hdev);
> -
> - params = hci_conn_params_lookup(hdev, &conn->dst, conn->dst_type);
> - if (params) {
> - params->conn_min_interval = min;
> - params->conn_max_interval = max;
> - params->conn_latency = latency;
> - params->supervision_timeout = to_multiplier;
> + if (!hci_conn_valid(hdev, conn)) {
> + hci_dev_unlock(hdev);
> + return -ECANCELED;
> }
> -
> hci_dev_unlock(hdev);
>
> memset(&cp, 0, sizeof(cp));
> cp.handle = cpu_to_le16(conn->handle);
hci_conn_valid() should be in the same hci_dev_lock() critical section
as where the conn is dereferenced, otherwise the lock is not doing
something useful.
> - cp.conn_interval_min = cpu_to_le16(min);
> - cp.conn_interval_max = cpu_to_le16(max);
> - cp.conn_latency = cpu_to_le16(latency);
> - cp.supervision_timeout = cpu_to_le16(to_multiplier);
> + cp.conn_interval_min = cpu_to_le16(d->min);
> + cp.conn_interval_max = cpu_to_le16(d->max);
> + cp.conn_latency = cpu_to_le16(d->latency);
> + cp.supervision_timeout = cpu_to_le16(d->to_multiplier);
> cp.min_ce_len = cpu_to_le16(0x0000);
> cp.max_ce_len = cpu_to_le16(0x0000);
>
> - hci_send_cmd(hdev, HCI_OP_LE_CONN_UPDATE, sizeof(cp), &cp);
> + err = __hci_cmd_sync_status_sk(hdev, HCI_OP_LE_CONN_UPDATE,
> + sizeof(cp), &cp,
> + HCI_EV_LE_CONN_UPDATE_COMPLETE,
> + HCI_CMD_TIMEOUT, NULL);
> + if (err)
> + return err;
> +
> + /* Update stored connection parameters after the controller has
> + * confirmed the update via the LE Connection Update Complete event.
> + */
> + hci_dev_lock(hdev);
>
> - if (params)
> - return 0x01;
> + params = hci_conn_params_lookup(hdev, &conn->dst, conn->dst_type);
> + if (params) {
> + params->conn_min_interval = d->min;
> + params->conn_max_interval = d->max;
> + params->conn_latency = d->latency;
> + params->supervision_timeout = d->to_multiplier;
> + store_hint = 0x01;
> + } else {
> + store_hint = 0x00;
> + }
>
> - return 0x00;
> + hci_dev_unlock(hdev);
> +
> + mgmt_new_conn_param(hdev, &conn->dst, conn->dst_type, store_hint,
> + d->min, d->max, d->latency, d->to_multiplier);
> +
> + return 0;
> +}
> +
> +static void le_conn_update_complete(struct hci_dev *hdev, void *data, int err)
> +{
> + struct le_conn_update_data *d = data;
> +
> + hci_conn_put(d->conn);
> + kfree(d);
> +}
> +
> +void hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> + u16 to_multiplier)
> +{
> + struct le_conn_update_data *d;
> +
> + d = kzalloc_obj(*d);
> + if (!d)
> + return;
> +
> + hci_conn_get(conn);
> + d->conn = conn;
> + d->min = min;
> + d->max = max;
> + d->latency = latency;
> + d->to_multiplier = to_multiplier;
> +
> + if (hci_cmd_sync_queue(conn->hdev, le_conn_update_sync, d,
> + le_conn_update_complete) < 0) {
> + hci_conn_put(conn);
> + kfree(d);
> + }
> }
>
> void hci_le_start_enc(struct hci_conn *conn, __le16 ediv, __le64 rand,
> diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> index 95c65fece39b..aac2db1d6fbb 100644
> --- a/net/bluetooth/l2cap_core.c
> +++ b/net/bluetooth/l2cap_core.c
> @@ -4706,16 +4706,8 @@ static inline int l2cap_conn_param_update_req(struct l2cap_conn *conn,
> l2cap_send_cmd(conn, cmd->ident, L2CAP_CONN_PARAM_UPDATE_RSP,
> sizeof(rsp), &rsp);
>
> - if (!err) {
> - u8 store_hint;
> -
> - store_hint = hci_le_conn_update(hcon, min, max, latency,
> - to_multiplier);
> - mgmt_new_conn_param(hcon->hdev, &hcon->dst, hcon->dst_type,
> - store_hint, min, max, latency,
> - to_multiplier);
> -
> - }
> + if (!err)
> + hci_le_conn_update(hcon, min, max, latency, to_multiplier);
>
> return 0;
> }
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH v4] Bluetooth: l2cap: defer conn param update to avoid conn->lock/hdev->lock inversion
2026-04-14 20:26 ` [PATCH v4] " Pauli Virtanen
@ 2026-04-14 20:30 ` Luiz Augusto von Dentz
0 siblings, 0 replies; 5+ messages in thread
From: Luiz Augusto von Dentz @ 2026-04-14 20:30 UTC (permalink / raw)
To: Pauli Virtanen
Cc: Mikhail Gavrilov, marcel, pmenzel, linux-bluetooth, linux-kernel
Hi Pauli,
On Tue, Apr 14, 2026 at 4:26 PM Pauli Virtanen <pav@iki.fi> wrote:
>
> ti, 2026-04-14 kello 23:43 +0500, Mikhail Gavrilov kirjoitti:
> > When a BLE peripheral sends an L2CAP Connection Parameter Update Request
> > the processing path is:
> >
> > process_pending_rx() [takes conn->lock]
> > l2cap_le_sig_channel()
> > l2cap_conn_param_update_req()
> > hci_le_conn_update() [takes hdev->lock]
> >
> > Meanwhile other code paths take the locks in the opposite order:
> >
> > l2cap_chan_connect() [takes hdev->lock]
> > ...
> > mutex_lock(&conn->lock)
> >
> > l2cap_conn_ready() [hdev->lock via hci_cb_list_lock]
> > ...
> > mutex_lock(&conn->lock)
> >
> > This is a classic AB/BA deadlock which lockdep reports as a circular
> > locking dependency when connecting a BLE MIDI keyboard (Carry-On FC-49).
> >
> > Fix this by making hci_le_conn_update() defer the HCI command through
> > hci_cmd_sync_queue() so it no longer needs to take hdev->lock in the
> > caller context. The sync callback uses __hci_cmd_sync_status_sk() to
> > wait for the HCI_EV_LE_CONN_UPDATE_COMPLETE event, then updates the
> > stored connection parameters (hci_conn_params) and notifies userspace
> > (mgmt_new_conn_param) only after the controller has confirmed the update.
> >
> > A reference on hci_conn is held via hci_conn_get()/hci_conn_put() for
> > the lifetime of the queued work to prevent use-after-free, and
> > hci_conn_valid() is checked before proceeding in case the connection was
> > removed while the work was pending.
> >
> > Fixes: f044eb0524a0 ("Bluetooth: Store latency and supervision timeout in connection params")
> > Signed-off-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
> > Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
> > ---
> >
> > Changes in v4 (Luiz Augusto von Dentz, Sashiko/Gemini AI review):
> > - Use hci_conn_get()/hci_conn_put() to hold a reference while work is queued
> > - Use __hci_cmd_sync_status_sk() to wait for HCI_EV_LE_CONN_UPDATE_COMPLETE,
> > then do params update + mgmt notification in the sync callback — removes the
> > need for hci_event.c changes and hci_sent_cmd_data()
> > - Use kzalloc_obj() per checkpatch recommendation
> > - Fix checkpatch alignment issues
> >
> > Changes in v3 (Luiz Augusto von Dentz):
> > - Move hci_cmd_sync_queue into hci_le_conn_update itself instead of open-coding
> > the deferral in l2cap_core.c
> > - Move conn_params update and mgmt_new_conn_param into
> > hci_le_conn_update_complete_evt, using hci_sent_cmd_data to retrieve
> > the originally requested parameters
> >
> > Changes in v2 (Paul Menzel, Sashiko/Gemini AI review):
> > - Allocate before sending ACCEPTED response to avoid state mismatch on OOM
> > - Verify connection handle and address in sync callback against reuse race
> > - Expand commit message with implementation details
> >
> > include/net/bluetooth/hci_core.h | 2 +-
> > net/bluetooth/hci_conn.c | 98 +++++++++++++++++++++++++-------
> > net/bluetooth/l2cap_core.c | 12 +---
> > 3 files changed, 82 insertions(+), 30 deletions(-)
> >
> > diff --git a/include/net/bluetooth/hci_core.h b/include/net/bluetooth/hci_core.h
> > index a7bffb908c1e..aa600fbf9a53 100644
> > --- a/include/net/bluetooth/hci_core.h
> > +++ b/include/net/bluetooth/hci_core.h
> > @@ -2495,7 +2495,7 @@ void mgmt_adv_monitor_device_lost(struct hci_dev *hdev, u16 handle,
> > bdaddr_t *bdaddr, u8 addr_type);
> >
> > int hci_abort_conn(struct hci_conn *conn, u8 reason);
> > -u8 hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> > +void hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> > u16 to_multiplier);
> > void hci_le_start_enc(struct hci_conn *conn, __le16 ediv, __le64 rand,
> > __u8 ltk[16], __u8 key_size);
> > diff --git a/net/bluetooth/hci_conn.c b/net/bluetooth/hci_conn.c
> > index 11d3ad8d2551..c9cc18fab1f2 100644
> > --- a/net/bluetooth/hci_conn.c
> > +++ b/net/bluetooth/hci_conn.c
> > @@ -480,40 +480,100 @@ bool hci_setup_sync(struct hci_conn *conn, __u16 handle)
> > return hci_setup_sync_conn(conn, handle);
> > }
> >
> > -u8 hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> > - u16 to_multiplier)
> > +struct le_conn_update_data {
> > + struct hci_conn *conn;
> > + u16 min;
> > + u16 max;
> > + u16 latency;
> > + u16 to_multiplier;
> > +};
> > +
> > +static int le_conn_update_sync(struct hci_dev *hdev, void *data)
> > {
> > - struct hci_dev *hdev = conn->hdev;
> > + struct le_conn_update_data *d = data;
> > + struct hci_conn *conn = d->conn;
> > struct hci_conn_params *params;
> > struct hci_cp_le_conn_update cp;
> > + u8 store_hint;
> > + int err;
> >
> > + /* Verify connection is still alive. */
> > hci_dev_lock(hdev);
> > -
> > - params = hci_conn_params_lookup(hdev, &conn->dst, conn->dst_type);
> > - if (params) {
> > - params->conn_min_interval = min;
> > - params->conn_max_interval = max;
> > - params->conn_latency = latency;
> > - params->supervision_timeout = to_multiplier;
> > + if (!hci_conn_valid(hdev, conn)) {
> > + hci_dev_unlock(hdev);
> > + return -ECANCELED;
> > }
> > -
> > hci_dev_unlock(hdev);
> >
> > memset(&cp, 0, sizeof(cp));
> > cp.handle = cpu_to_le16(conn->handle);
>
> hci_conn_valid() should be in the same hci_dev_lock() critical section
> as where the conn is dereferenced, otherwise the lock is not doing
> something useful.
This means we should only unlock after we finish using hci_conn,
including conn->handle above.
> > - cp.conn_interval_min = cpu_to_le16(min);
> > - cp.conn_interval_max = cpu_to_le16(max);
> > - cp.conn_latency = cpu_to_le16(latency);
> > - cp.supervision_timeout = cpu_to_le16(to_multiplier);
> > + cp.conn_interval_min = cpu_to_le16(d->min);
> > + cp.conn_interval_max = cpu_to_le16(d->max);
> > + cp.conn_latency = cpu_to_le16(d->latency);
> > + cp.supervision_timeout = cpu_to_le16(d->to_multiplier);
> > cp.min_ce_len = cpu_to_le16(0x0000);
> > cp.max_ce_len = cpu_to_le16(0x0000);
> >
> > - hci_send_cmd(hdev, HCI_OP_LE_CONN_UPDATE, sizeof(cp), &cp);
> > + err = __hci_cmd_sync_status_sk(hdev, HCI_OP_LE_CONN_UPDATE,
> > + sizeof(cp), &cp,
> > + HCI_EV_LE_CONN_UPDATE_COMPLETE,
> > + HCI_CMD_TIMEOUT, NULL);
> > + if (err)
> > + return err;
> > +
> > + /* Update stored connection parameters after the controller has
> > + * confirmed the update via the LE Connection Update Complete event.
> > + */
> > + hci_dev_lock(hdev);
> >
> > - if (params)
> > - return 0x01;
> > + params = hci_conn_params_lookup(hdev, &conn->dst, conn->dst_type);
> > + if (params) {
> > + params->conn_min_interval = d->min;
> > + params->conn_max_interval = d->max;
> > + params->conn_latency = d->latency;
> > + params->supervision_timeout = d->to_multiplier;
> > + store_hint = 0x01;
> > + } else {
> > + store_hint = 0x00;
> > + }
> >
> > - return 0x00;
> > + hci_dev_unlock(hdev);
> > +
> > + mgmt_new_conn_param(hdev, &conn->dst, conn->dst_type, store_hint,
> > + d->min, d->max, d->latency, d->to_multiplier);
> > +
> > + return 0;
> > +}
> > +
> > +static void le_conn_update_complete(struct hci_dev *hdev, void *data, int err)
> > +{
> > + struct le_conn_update_data *d = data;
> > +
> > + hci_conn_put(d->conn);
> > + kfree(d);
> > +}
> > +
> > +void hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> > + u16 to_multiplier)
> > +{
> > + struct le_conn_update_data *d;
> > +
> > + d = kzalloc_obj(*d);
> > + if (!d)
> > + return;
> > +
> > + hci_conn_get(conn);
> > + d->conn = conn;
> > + d->min = min;
> > + d->max = max;
> > + d->latency = latency;
> > + d->to_multiplier = to_multiplier;
> > +
> > + if (hci_cmd_sync_queue(conn->hdev, le_conn_update_sync, d,
> > + le_conn_update_complete) < 0) {
> > + hci_conn_put(conn);
> > + kfree(d);
> > + }
> > }
> >
> > void hci_le_start_enc(struct hci_conn *conn, __le16 ediv, __le64 rand,
> > diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> > index 95c65fece39b..aac2db1d6fbb 100644
> > --- a/net/bluetooth/l2cap_core.c
> > +++ b/net/bluetooth/l2cap_core.c
> > @@ -4706,16 +4706,8 @@ static inline int l2cap_conn_param_update_req(struct l2cap_conn *conn,
> > l2cap_send_cmd(conn, cmd->ident, L2CAP_CONN_PARAM_UPDATE_RSP,
> > sizeof(rsp), &rsp);
> >
> > - if (!err) {
> > - u8 store_hint;
> > -
> > - store_hint = hci_le_conn_update(hcon, min, max, latency,
> > - to_multiplier);
> > - mgmt_new_conn_param(hcon->hdev, &hcon->dst, hcon->dst_type,
> > - store_hint, min, max, latency,
> > - to_multiplier);
> > -
> > - }
> > + if (!err)
> > + hci_le_conn_update(hcon, min, max, latency, to_multiplier);
> >
> > return 0;
> > }
--
Luiz Augusto von Dentz
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH v4] Bluetooth: l2cap: defer conn param update to avoid conn->lock/hdev->lock inversion
2026-04-14 18:43 [PATCH v4] Bluetooth: l2cap: defer conn param update to avoid conn->lock/hdev->lock inversion Mikhail Gavrilov
2026-04-14 19:18 ` [v4] " bluez.test.bot
2026-04-14 20:26 ` [PATCH v4] " Pauli Virtanen
@ 2026-04-14 20:27 ` Luiz Augusto von Dentz
2 siblings, 0 replies; 5+ messages in thread
From: Luiz Augusto von Dentz @ 2026-04-14 20:27 UTC (permalink / raw)
To: Mikhail Gavrilov; +Cc: marcel, pmenzel, linux-bluetooth, linux-kernel
Hi Mikhail,
On Tue, Apr 14, 2026 at 2:44 PM Mikhail Gavrilov
<mikhail.v.gavrilov@gmail.com> wrote:
>
> When a BLE peripheral sends an L2CAP Connection Parameter Update Request
> the processing path is:
>
> process_pending_rx() [takes conn->lock]
> l2cap_le_sig_channel()
> l2cap_conn_param_update_req()
> hci_le_conn_update() [takes hdev->lock]
>
> Meanwhile other code paths take the locks in the opposite order:
>
> l2cap_chan_connect() [takes hdev->lock]
> ...
> mutex_lock(&conn->lock)
>
> l2cap_conn_ready() [hdev->lock via hci_cb_list_lock]
> ...
> mutex_lock(&conn->lock)
>
> This is a classic AB/BA deadlock which lockdep reports as a circular
> locking dependency when connecting a BLE MIDI keyboard (Carry-On FC-49).
>
> Fix this by making hci_le_conn_update() defer the HCI command through
> hci_cmd_sync_queue() so it no longer needs to take hdev->lock in the
> caller context. The sync callback uses __hci_cmd_sync_status_sk() to
> wait for the HCI_EV_LE_CONN_UPDATE_COMPLETE event, then updates the
> stored connection parameters (hci_conn_params) and notifies userspace
> (mgmt_new_conn_param) only after the controller has confirmed the update.
>
> A reference on hci_conn is held via hci_conn_get()/hci_conn_put() for
> the lifetime of the queued work to prevent use-after-free, and
> hci_conn_valid() is checked before proceeding in case the connection was
> removed while the work was pending.
>
> Fixes: f044eb0524a0 ("Bluetooth: Store latency and supervision timeout in connection params")
> Signed-off-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
> Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
> ---
>
> Changes in v4 (Luiz Augusto von Dentz, Sashiko/Gemini AI review):
> - Use hci_conn_get()/hci_conn_put() to hold a reference while work is queued
> - Use __hci_cmd_sync_status_sk() to wait for HCI_EV_LE_CONN_UPDATE_COMPLETE,
> then do params update + mgmt notification in the sync callback — removes the
> need for hci_event.c changes and hci_sent_cmd_data()
> - Use kzalloc_obj() per checkpatch recommendation
> - Fix checkpatch alignment issues
>
> Changes in v3 (Luiz Augusto von Dentz):
> - Move hci_cmd_sync_queue into hci_le_conn_update itself instead of open-coding
> the deferral in l2cap_core.c
> - Move conn_params update and mgmt_new_conn_param into
> hci_le_conn_update_complete_evt, using hci_sent_cmd_data to retrieve
> the originally requested parameters
>
> Changes in v2 (Paul Menzel, Sashiko/Gemini AI review):
> - Allocate before sending ACCEPTED response to avoid state mismatch on OOM
> - Verify connection handle and address in sync callback against reuse race
> - Expand commit message with implementation details
>
> include/net/bluetooth/hci_core.h | 2 +-
> net/bluetooth/hci_conn.c | 98 +++++++++++++++++++++++++-------
> net/bluetooth/l2cap_core.c | 12 +---
> 3 files changed, 82 insertions(+), 30 deletions(-)
>
> diff --git a/include/net/bluetooth/hci_core.h b/include/net/bluetooth/hci_core.h
> index a7bffb908c1e..aa600fbf9a53 100644
> --- a/include/net/bluetooth/hci_core.h
> +++ b/include/net/bluetooth/hci_core.h
> @@ -2495,7 +2495,7 @@ void mgmt_adv_monitor_device_lost(struct hci_dev *hdev, u16 handle,
> bdaddr_t *bdaddr, u8 addr_type);
>
> int hci_abort_conn(struct hci_conn *conn, u8 reason);
> -u8 hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> +void hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> u16 to_multiplier);
> void hci_le_start_enc(struct hci_conn *conn, __le16 ediv, __le64 rand,
> __u8 ltk[16], __u8 key_size);
> diff --git a/net/bluetooth/hci_conn.c b/net/bluetooth/hci_conn.c
> index 11d3ad8d2551..c9cc18fab1f2 100644
> --- a/net/bluetooth/hci_conn.c
> +++ b/net/bluetooth/hci_conn.c
> @@ -480,40 +480,100 @@ bool hci_setup_sync(struct hci_conn *conn, __u16 handle)
> return hci_setup_sync_conn(conn, handle);
> }
>
> -u8 hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> - u16 to_multiplier)
> +struct le_conn_update_data {
> + struct hci_conn *conn;
> + u16 min;
> + u16 max;
> + u16 latency;
> + u16 to_multiplier;
> +};
> +
> +static int le_conn_update_sync(struct hci_dev *hdev, void *data)
> {
> - struct hci_dev *hdev = conn->hdev;
> + struct le_conn_update_data *d = data;
> + struct hci_conn *conn = d->conn;
> struct hci_conn_params *params;
> struct hci_cp_le_conn_update cp;
> + u8 store_hint;
> + int err;
>
> + /* Verify connection is still alive. */
> hci_dev_lock(hdev);
> -
> - params = hci_conn_params_lookup(hdev, &conn->dst, conn->dst_type);
> - if (params) {
> - params->conn_min_interval = min;
> - params->conn_max_interval = max;
> - params->conn_latency = latency;
> - params->supervision_timeout = to_multiplier;
> + if (!hci_conn_valid(hdev, conn)) {
> + hci_dev_unlock(hdev);
> + return -ECANCELED;
> }
> -
> hci_dev_unlock(hdev);
>
> memset(&cp, 0, sizeof(cp));
> cp.handle = cpu_to_le16(conn->handle);
> - cp.conn_interval_min = cpu_to_le16(min);
> - cp.conn_interval_max = cpu_to_le16(max);
> - cp.conn_latency = cpu_to_le16(latency);
> - cp.supervision_timeout = cpu_to_le16(to_multiplier);
> + cp.conn_interval_min = cpu_to_le16(d->min);
> + cp.conn_interval_max = cpu_to_le16(d->max);
> + cp.conn_latency = cpu_to_le16(d->latency);
> + cp.supervision_timeout = cpu_to_le16(d->to_multiplier);
> cp.min_ce_len = cpu_to_le16(0x0000);
> cp.max_ce_len = cpu_to_le16(0x0000);
>
> - hci_send_cmd(hdev, HCI_OP_LE_CONN_UPDATE, sizeof(cp), &cp);
> + err = __hci_cmd_sync_status_sk(hdev, HCI_OP_LE_CONN_UPDATE,
> + sizeof(cp), &cp,
> + HCI_EV_LE_CONN_UPDATE_COMPLETE,
> + HCI_CMD_TIMEOUT, NULL);
> + if (err)
> + return err;
> +
> + /* Update stored connection parameters after the controller has
> + * confirmed the update via the LE Connection Update Complete event.
> + */
> + hci_dev_lock(hdev);
>
> - if (params)
> - return 0x01;
> + params = hci_conn_params_lookup(hdev, &conn->dst, conn->dst_type);
> + if (params) {
> + params->conn_min_interval = d->min;
> + params->conn_max_interval = d->max;
> + params->conn_latency = d->latency;
> + params->supervision_timeout = d->to_multiplier;
> + store_hint = 0x01;
> + } else {
> + store_hint = 0x00;
> + }
>
> - return 0x00;
> + hci_dev_unlock(hdev);
> +
> + mgmt_new_conn_param(hdev, &conn->dst, conn->dst_type, store_hint,
> + d->min, d->max, d->latency, d->to_multiplier);
> +
> + return 0;
> +}
> +
> +static void le_conn_update_complete(struct hci_dev *hdev, void *data, int err)
> +{
> + struct le_conn_update_data *d = data;
> +
> + hci_conn_put(d->conn);
> + kfree(d);
> +}
> +
> +void hci_le_conn_update(struct hci_conn *conn, u16 min, u16 max, u16 latency,
> + u16 to_multiplier)
> +{
> + struct le_conn_update_data *d;
> +
> + d = kzalloc_obj(*d);
> + if (!d)
> + return;
> +
> + hci_conn_get(conn);
> + d->conn = conn;
> + d->min = min;
> + d->max = max;
> + d->latency = latency;
> + d->to_multiplier = to_multiplier;
> +
> + if (hci_cmd_sync_queue(conn->hdev, le_conn_update_sync, d,
> + le_conn_update_complete) < 0) {
> + hci_conn_put(conn);
> + kfree(d);
> + }
> }
>
> void hci_le_start_enc(struct hci_conn *conn, __le16 ediv, __le64 rand,
> diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> index 95c65fece39b..aac2db1d6fbb 100644
> --- a/net/bluetooth/l2cap_core.c
> +++ b/net/bluetooth/l2cap_core.c
> @@ -4706,16 +4706,8 @@ static inline int l2cap_conn_param_update_req(struct l2cap_conn *conn,
> l2cap_send_cmd(conn, cmd->ident, L2CAP_CONN_PARAM_UPDATE_RSP,
> sizeof(rsp), &rsp);
>
> - if (!err) {
> - u8 store_hint;
> -
> - store_hint = hci_le_conn_update(hcon, min, max, latency,
> - to_multiplier);
> - mgmt_new_conn_param(hcon->hdev, &hcon->dst, hcon->dst_type,
> - store_hint, min, max, latency,
> - to_multiplier);
> -
> - }
> + if (!err)
> + hci_le_conn_update(hcon, min, max, latency, to_multiplier);
>
> return 0;
> }
> --
> 2.53.0
https://sashiko.dev/#/patchset/20260414184415.17114-1-mikhail.v.gavrilov%40gmail.com
1. I don't think it really matters because connection intervals are
typically 60ms at most, so 6*60 = 360ms. That said, perhaps we should
use 6 * max_interval as the timeout since that is how long the
controller has to update the connection interval, well if the 6
consecutive intervals are actually a thing since I don't recall
anything like that being mentioned in the command, but perhaps it is
in the link layer. But if it times out I don't think there is a way to
cancel the change either, so I'm not sure how we'll be able to
synchronize if the controller misbehaves.
2. Well I guess that is would need to defer the response and wait the
HCI proceedure to complete to really respond properly, but that would
mean the remote could get a LL procedure to change the connection
interval before the L2CAP response, so I can see this going wrong
either way.
--
Luiz Augusto von Dentz
^ permalink raw reply [flat|nested] 5+ messages in thread