Bluetooth: Fix crash in SMP when unpairing
When unpairing the keys stored in hci_dev are removed. If SMP is
ongoing the SMP context will also have references to these keys, so
removing them from the hci_dev lists will make the pointers invalid.
This can result in the following type of crashes:
BUG: unable to handle kernel paging request at 6b6b6b6b
IP: [<c11f26be>] __list_del_entry+0x44/0x71
*pde = 00000000
Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
Modules linked in: hci_uart btqca btusb btintel btbcm btrtl hci_vhci rfcomm bluetooth_6lowpan bluetooth
CPU: 0 PID: 723 Comm: kworker/u5:0 Not tainted 4.3.0-rc3+ #1379
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014
Workqueue: hci0 hci_rx_work [bluetooth]
task: f19da940 ti: f1a94000 task.ti: f1a94000
EIP: 0060:[<c11f26be>] EFLAGS: 00010202 CPU: 0
EIP is at __list_del_entry+0x44/0x71
EAX: c0088d20 EBX: f30fcac0 ECX: 6b6b6b6b EDX: 6b6b6b6b
ESI: f4b60000 EDI: c0088d20 EBP: f1a95d90 ESP: f1a95d8c
DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
CR0: 8005003b CR2: 6b6b6b6b CR3: 319e5000 CR4: 00000690
Stack:
f30fcac0 f1a95db0 f82dc3e1 f1bfc000 00000000 c106524f f1bfc000 f30fd020
f1a95dc0 f1a95dd0 f82dcbdb f1a95de0 f82dcbdb 00000067 f1bfc000 f30fd020
f1a95de0 f1a95df0 f82d1126 00000067 f82d1126 00000006 f30fd020 f1bfc000
Call Trace:
[<f82dc3e1>] smp_chan_destroy+0x192/0x240 [bluetooth]
[<c106524f>] ? trace_hardirqs_on_caller+0x14e/0x169
[<f82dcbdb>] smp_teardown_cb+0x47/0x64 [bluetooth]
[<f82dcbdb>] ? smp_teardown_cb+0x47/0x64 [bluetooth]
[<f82d1126>] l2cap_chan_del+0x5d/0x14d [bluetooth]
[<f82d1126>] ? l2cap_chan_del+0x5d/0x14d [bluetooth]
[<f82d40ef>] l2cap_conn_del+0x109/0x17b [bluetooth]
[<f82d40ef>] ? l2cap_conn_del+0x109/0x17b [bluetooth]
[<f82c0205>] ? hci_event_packet+0x5b1/0x2092 [bluetooth]
[<f82d41aa>] l2cap_disconn_cfm+0x49/0x50 [bluetooth]
[<f82d41aa>] ? l2cap_disconn_cfm+0x49/0x50 [bluetooth]
[<f82c0228>] hci_event_packet+0x5d4/0x2092 [bluetooth]
[<c1332c16>] ? skb_release_data+0x6a/0x95
[<f82ce5d4>] ? hci_send_to_monitor+0xe7/0xf4 [bluetooth]
[<c1409708>] ? _raw_spin_unlock_irqrestore+0x44/0x57
[<f82b3bb0>] hci_rx_work+0xf1/0x28b [bluetooth]
[<f82b3bb0>] ? hci_rx_work+0xf1/0x28b [bluetooth]
[<c10635a0>] ? __lock_is_held+0x2e/0x44
[<c104772e>] process_one_work+0x232/0x432
[<c1071ddc>] ? rcu_read_lock_sched_held+0x50/0x5a
[<c104772e>] ? process_one_work+0x232/0x432
[<c1047d48>] worker_thread+0x1b8/0x255
[<c1047b90>] ? rescuer_thread+0x23c/0x23c
[<c104bb71>] kthread+0x91/0x96
[<c14096a7>] ? _raw_spin_unlock_irq+0x27/0x44
[<c1409d61>] ret_from_kernel_thread+0x21/0x30
[<c104bae0>] ? kthread_parkme+0x1e/0x1e
To solve the issue, introduce a new smp_cancel_pairing() API that can
be used to clean up the SMP state before touching the hci_dev lists.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
3 files changed
