Bluetooth: When HCI work queue is drained, only queue chained work

The HCI command, event, and data packet processing workqueue is drained
to avoid deadlock in commit
76727c02 ("Bluetooth: Call drain_workqueue() before resetting state").

There is another delayed work, which will queue command to this drained
workqueue. Which results in the following error report:

Bluetooth: hci2: command 0x040f tx timeout
WARNING: CPU: 1 PID: 18374 at kernel/workqueue.c:1438 __queue_work+0xdad/0x1140
Workqueue: events hci_cmd_timeout
RIP: 0010:__queue_work+0xdad/0x1140
RSP: 0000:ffffc90002cffc60 EFLAGS: 00010093
RAX: 0000000000000000 RBX: ffff8880b9d3ec00 RCX: 0000000000000000
RDX: ffff888024ba0000 RSI: ffffffff814e048d RDI: ffff8880b9d3ec08
RBP: 0000000000000008 R08: 0000000000000000 R09: 00000000b9d39700
R10: ffffffff814f73c6 R11: 0000000000000000 R12: ffff88807cce4c60
R13: 0000000000000000 R14: ffff8880796d8800 R15: ffff8880796d8800
FS:  0000000000000000(0000) GS:ffff8880b9d00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000000c0174b4000 CR3: 000000007cae9000 CR4: 00000000003506e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 <TASK>
 ? queue_work_on+0xcb/0x110
 ? lockdep_hardirqs_off+0x90/0xd0
 queue_work_on+0xee/0x110
 process_one_work+0x996/0x1610
 ? pwq_dec_nr_in_flight+0x2a0/0x2a0
 ? rwlock_bug.part.0+0x90/0x90
 ? _raw_spin_lock_irq+0x41/0x50
 worker_thread+0x665/0x1080
 ? process_one_work+0x1610/0x1610
 kthread+0x2e9/0x3a0
 ? kthread_complete_and_exit+0x40/0x40
 ret_from_fork+0x1f/0x30
 </TASK>

To fix this, we can add a new HCI_DRAIN_WQ flag, and don't queue the
timeout workqueue while command workqueue is draining.

Fixes: 76727c02 ("Bluetooth: Call drain_workqueue() before resetting state")
Reported-by: syzbot+63bed493aebbf6872647@syzkaller.appspotmail.com
Signed-off-by: Schspa Shi <schspa@gmail.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>

include/net/bluetooth/hci.h

@@ -361,6 +361,7 @@ enum {
 	HCI_QUALITY_REPORT,
 	HCI_OFFLOAD_CODECS_ENABLED,
 	HCI_LE_SIMULTANEOUS_ROLES,
+	HCI_CMD_DRAIN_WORKQUEUE,
 
 	__HCI_NUM_FLAGS,
 };

net/bluetooth/hci_core.c

@@ -594,6 +594,11 @@ static int hci_dev_do_reset(struct hci_dev *hdev)
 	skb_queue_purge(&hdev->rx_q);
 	skb_queue_purge(&hdev->cmd_q);
 
+	/* Cancel these to avoid queueing non-chained pending work */
+	hci_dev_set_flag(hdev, HCI_CMD_DRAIN_WORKQUEUE);
+	cancel_delayed_work(&hdev->cmd_timer);
+	cancel_delayed_work(&hdev->ncmd_timer);
+
 	/* Avoid potential lockdep warnings from the *_flush() calls by
 	 * ensuring the workqueue is empty up front.
 	 */
@@ -607,6 +612,8 @@ static int hci_dev_do_reset(struct hci_dev *hdev)
 	if (hdev->flush)
 		hdev->flush(hdev);
 
+	hci_dev_clear_flag(hdev, HCI_CMD_DRAIN_WORKQUEUE);
+
 	atomic_set(&hdev->cmd_cnt, 1);
 	hdev->acl_cnt = 0; hdev->sco_cnt = 0; hdev->le_cnt = 0;
 
@@ -3864,7 +3871,8 @@ static void hci_cmd_work(struct work_struct *work)
 			if (res < 0)
 				__hci_cmd_sync_cancel(hdev, -res);
 
-			if (test_bit(HCI_RESET, &hdev->flags))
+			if (test_bit(HCI_RESET, &hdev->flags) ||
+			    hci_dev_test_flag(hdev, HCI_CMD_DRAIN_WORKQUEUE))
 				cancel_delayed_work(&hdev->cmd_timer);
 			else
 				schedule_delayed_work(&hdev->cmd_timer,

net/bluetooth/hci_event.c

@@ -3768,8 +3768,9 @@ static inline void handle_cmd_cnt_and_timer(struct hci_dev *hdev, u8 ncmd)
 			cancel_delayed_work(&hdev->ncmd_timer);
 			atomic_set(&hdev->cmd_cnt, 1);
 		} else {
-			schedule_delayed_work(&hdev->ncmd_timer,
-					      HCI_NCMD_TIMEOUT);
+			if (!hci_dev_test_flag(hdev, HCI_CMD_DRAIN_WORKQUEUE))
+				schedule_delayed_work(&hdev->ncmd_timer,
+						      HCI_NCMD_TIMEOUT);
 		}
 	}
 }