[PATCH] Bluetooth: Prevent double register of suspend

5 views
Skip to first unread message

Abhishek Pandit-Subedi

unread,
Sep 27, 2022, 1:37:07 AMSep 27
to linux-b...@vger.kernel.org, Abhishek Pandit-Subedi, syzbot, David S. Miller, Eric Dumazet, Jakub Kicinski, Johan Hedberg, Luiz Augusto von Dentz, Marcel Holtmann, Paolo Abeni, linux-...@vger.kernel.org, net...@vger.kernel.org
From: Abhishek Pandit-Subedi <abhishe...@chromium.org>

Suspend notifier should only be registered and unregistered once per
hdev. Since hci_sock and hci_register_dev run in different work queues
(sock vs driver), add hci_dev_lock to avoid double registering.

Reported-by: syzbot <syzk...@googlegroups.com>
Signed-off-by: Abhishek Pandit-Subedi <abhishe...@chromium.org>
---
This is fixing a syzbot reported warning. Tested in the following ways:
* Normal start-up of driver with bluez.
* Start/stop loop using HCI_USER_CHANNEL (sock path).
* USB reset triggering hci_dev_unregister (driver path).

------------[ cut here ]------------
double register detected
WARNING: CPU: 0 PID: 2657 at kernel/notifier.c:27
notifier_chain_register kernel/notifier.c:27 [inline]
WARNING: CPU: 0 PID: 2657 at kernel/notifier.c:27
notifier_chain_register+0x5c/0x124 kernel/notifier.c:22
Modules linked in:
CPU: 0 PID: 2657 Comm: syz-executor212 Not tainted
5.10.136-syzkaller-19376-g6f46a5fe0124 #0
8f0771607702f5ef7184d2ee33bd0acd70219fc4
Hardware name: Google Google Compute Engine/Google Compute Engine,
BIOS Google 07/22/2022
RIP: 0010:notifier_chain_register kernel/notifier.c:27 [inline]
RIP: 0010:notifier_chain_register+0x5c/0x124 kernel/notifier.c:22
Code: 6a 41 00 4c 8b 23 4d 85 e4 0f 84 88 00 00 00 e8 c2 1e 19 00 49
39 ec 75 18 e8 b8 1e 19 00 48 c7 c7 80 6d ca 84 e8 2c 68 48 03 <0f> 0b
e9 af 00 00 00 e8 a0 1e 19 00 48 8d 7d 10 48 89 f8 48 c1 e8
RSP: 0018:ffffc900009d7da8 EFLAGS: 00010286
RAX: 0000000000000000 RBX: ffff8881076fd1d8 RCX: 0000000000000000
RDX: 0000001810895100 RSI: ffff888110895100 RDI: fffff5200013afa7
RBP: ffff88811a4191d0 R08: ffffffff813b8ca1 R09: 0000000080000000
R10: 0000000000000000 R11: 0000000000000005 R12: ffff88811a4191d0
R13: dffffc0000000000 R14: 0000000000000000 R15: 0000000000000000
FS: 00005555571f5300(0000) GS:ffff8881f6c00000(0000)
knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000078e3857f3075 CR3: 000000010d668000 CR4: 00000000003506f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
blocking_notifier_chain_register+0x8c/0xa6 kernel/notifier.c:254
hci_register_suspend_notifier net/bluetooth/hci_core.c:2733
[inline]
hci_register_suspend_notifier+0x6b/0x7c
net/bluetooth/hci_core.c:2727
hci_sock_release+0x270/0x3cf net/bluetooth/hci_sock.c:889
__sock_release+0xcd/0x1de net/socket.c:597
sock_close+0x18/0x1c net/socket.c:1267
__fput+0x418/0x729 fs/file_table.c:281
task_work_run+0x12b/0x15b kernel/task_work.c:151
tracehook_notify_resume include/linux/tracehook.h:188 [inline]
exit_to_user_mode_loop kernel/entry/common.c:165 [inline]
exit_to_user_mode_prepare+0x8f/0x130 kernel/entry/common.c:192
syscall_exit_to_user_mode+0x172/0x1b2 kernel/entry/common.c:268
entry_SYSCALL_64_after_hwframe+0x61/0xc6
RIP: 0033:0x78e38575e1db
Code: 0f 05 48 3d 00 f0 ff ff 77 45 c3 0f 1f 40 00 48 83 ec 18 89
7c 24 0c e8 63 fc ff ff 8b 7c 24 0c 41 89 c0 b8 03 00 00 00 0f 05
<48> 3d 00 f0 ff ff 77 35 44 89 c7 89 44 24 0c e8 a1 fc ff ff 8b 44
RSP: 002b:00007ffffc20a0b0 EFLAGS: 00000293 ORIG_RAX:
0000000000000003
RAX: 0000000000000000 RBX: 0000000000000006 RCX: 000078e38575e1db
RDX: ffffffffffffffb8 RSI: 0000000020000000 RDI: 0000000000000005
RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000150
R10: 0000000000000000 R11: 0000000000000293 R12: 000000000000e155
R13: 00007ffffc20a140 R14: 00007ffffc20a130 R15: 00007ffffc20a0e8

include/net/bluetooth/hci.h | 1 +
net/bluetooth/hci_core.c | 16 ++++++++++++++--
2 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/include/net/bluetooth/hci.h b/include/net/bluetooth/hci.h
index e004ba04a9ae..36304c217151 100644
--- a/include/net/bluetooth/hci.h
+++ b/include/net/bluetooth/hci.h
@@ -353,6 +353,7 @@ enum {
HCI_OFFLOAD_CODECS_ENABLED,
HCI_LE_SIMULTANEOUS_ROLES,
HCI_CMD_DRAIN_WORKQUEUE,
+ HCI_SUSPEND_REGISTERED,

HCI_MESH_EXPERIMENTAL,
HCI_MESH,
diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index 66c7cdba0d32..5a32d17c69b8 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -2760,10 +2760,18 @@ int hci_register_suspend_notifier(struct hci_dev *hdev)
{
int ret = 0;

- if (!test_bit(HCI_QUIRK_NO_SUSPEND_NOTIFIER, &hdev->quirks)) {
+ hci_dev_lock(hdev);
+ if (!test_bit(HCI_QUIRK_NO_SUSPEND_NOTIFIER, &hdev->quirks) &&
+ !hci_dev_test_and_set_flag(hdev, HCI_SUSPEND_REGISTERED)) {
+ memset(&hdev->suspend_notifier, 0,
+ sizeof(hdev->suspend_notifier));
hdev->suspend_notifier.notifier_call = hci_suspend_notifier;
ret = register_pm_notifier(&hdev->suspend_notifier);
+
+ if (ret)
+ hci_dev_clear_flag(hdev, HCI_SUSPEND_REGISTERED);
}
+ hci_dev_unlock(hdev);

return ret;
}
@@ -2772,8 +2780,12 @@ int hci_unregister_suspend_notifier(struct hci_dev *hdev)
{
int ret = 0;

- if (!test_bit(HCI_QUIRK_NO_SUSPEND_NOTIFIER, &hdev->quirks))
+ hci_dev_lock(hdev);
+ if (!test_bit(HCI_QUIRK_NO_SUSPEND_NOTIFIER, &hdev->quirks) &&
+ hci_dev_test_and_clear_flag(hdev, HCI_SUSPEND_REGISTERED)) {
ret = unregister_pm_notifier(&hdev->suspend_notifier);
+ }
+ hci_dev_unlock(hdev);

return ret;
}
--
2.37.3.998.g577e59143f-goog

Luiz Augusto von Dentz

unread,
Sep 27, 2022, 1:37:07 AMSep 27
to Abhishek Pandit-Subedi, linux-b...@vger.kernel.org, Abhishek Pandit-Subedi, syzbot, David S. Miller, Eric Dumazet, Jakub Kicinski, Johan Hedberg, Marcel Holtmann, Paolo Abeni, linux-...@vger.kernel.org, net...@vger.kernel.org
Hi Abhishek,
Perhaps it would have been better to stop calling these on hci_sock.c
and just make the notifier callback check
hci_dev_test_and_set_flag(hdev, HCI_USER_CHANNEL) return NOTIFY_DONE.

> --
> 2.37.3.998.g577e59143f-goog
>


--
Luiz Augusto von Dentz

Abhishek Pandit-Subedi

unread,
Sep 27, 2022, 1:37:15 AMSep 27
to Luiz Augusto von Dentz, Abhishek Pandit-Subedi, linux-b...@vger.kernel.org, syzbot, David S. Miller, Eric Dumazet, Jakub Kicinski, Johan Hedberg, Marcel Holtmann, Paolo Abeni, linux-...@vger.kernel.org, net...@vger.kernel.org
Actually, yeah that's a much simpler change... Got a little
tunnel-visioned trying to fix the error :)
I'll send up a v2 tomorrow after testing.

Abhishek Pandit-Subedi

unread,
Sep 27, 2022, 12:58:19 PMSep 27
to linux-b...@vger.kernel.org, Abhishek Pandit-Subedi, syzbot, David S. Miller, Eric Dumazet, Jakub Kicinski, Johan Hedberg, Luiz Augusto von Dentz, Marcel Holtmann, Paolo Abeni, linux-...@vger.kernel.org, net...@vger.kernel.org
From: Abhishek Pandit-Subedi <abhishe...@chromium.org>

Suspend notifier should only be registered and unregistered once per
hdev. Simplify this by only registering during driver registration and
simply exiting early when HCI_USER_CHANNEL is set.

Reported-by: syzbot <syzk...@googlegroups.com>
Fixes: 359ee4f834f5 (Bluetooth: Unregister suspend with userchannel)
Changes in v2:
- Removed suspend registration from hci_sock.
- Exit hci_suspend_notifier early if user channel.

net/bluetooth/hci_core.c | 4 ++++
net/bluetooth/hci_sock.c | 3 ---
2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index 66c7cdba0d32..86ce2dd1c7fb 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -2406,6 +2406,10 @@ static int hci_suspend_notifier(struct notifier_block *nb, unsigned long action,
container_of(nb, struct hci_dev, suspend_notifier);
int ret = 0;

+ /* Userspace has full control of this device. Do nothing. */
+ if (hci_dev_test_flag(hdev, HCI_USER_CHANNEL))
+ return NOTIFY_DONE;
+
if (action == PM_SUSPEND_PREPARE)
ret = hci_suspend_dev(hdev);
else if (action == PM_POST_SUSPEND)
diff --git a/net/bluetooth/hci_sock.c b/net/bluetooth/hci_sock.c
index b2a33a05c93e..06581223238c 100644
--- a/net/bluetooth/hci_sock.c
+++ b/net/bluetooth/hci_sock.c
@@ -887,7 +887,6 @@ static int hci_sock_release(struct socket *sock)
*/
hci_dev_do_close(hdev);
hci_dev_clear_flag(hdev, HCI_USER_CHANNEL);
- hci_register_suspend_notifier(hdev);
mgmt_index_added(hdev);
}

@@ -1216,7 +1215,6 @@ static int hci_sock_bind(struct socket *sock, struct sockaddr *addr,
}

mgmt_index_removed(hdev);
- hci_unregister_suspend_notifier(hdev);

err = hci_dev_open(hdev->id);
if (err) {
@@ -1231,7 +1229,6 @@ static int hci_sock_bind(struct socket *sock, struct sockaddr *addr,
err = 0;
} else {
hci_dev_clear_flag(hdev, HCI_USER_CHANNEL);
- hci_register_suspend_notifier(hdev);
mgmt_index_added(hdev);
hci_dev_put(hdev);
goto done;
--
2.37.3.998.g577e59143f-goog

Luiz Augusto von Dentz

unread,
Sep 28, 2022, 3:32:06 PMSep 28
to Abhishek Pandit-Subedi, linux-b...@vger.kernel.org, Abhishek Pandit-Subedi, syzbot, David S. Miller, Eric Dumazet, Jakub Kicinski, Johan Hedberg, Marcel Holtmann, Paolo Abeni, linux-...@vger.kernel.org, net...@vger.kernel.org
Hi Abhishek,
Looks like our CI got stuck for some reason, do you mind sending a v3
just to confirm nothing breaks with these changes?

Abhishek Pandit-Subedi

unread,
Sep 28, 2022, 3:59:36 PMSep 28
to Luiz Augusto von Dentz, Abhishek Pandit-Subedi, linux-b...@vger.kernel.org, syzbot, David S. Miller, Eric Dumazet, Jakub Kicinski, Johan Hedberg, Marcel Holtmann, Paolo Abeni, linux-...@vger.kernel.org, net...@vger.kernel.org
Sure, resending with a v3.

Abhishek Pandit-Subedi

unread,
Sep 28, 2022, 4:00:36 PMSep 28
to linux-b...@vger.kernel.org, Abhishek Pandit-Subedi, syzbot, David S. Miller, Eric Dumazet, Jakub Kicinski, Johan Hedberg, Luiz Augusto von Dentz, Marcel Holtmann, Paolo Abeni, linux-...@vger.kernel.org, net...@vger.kernel.org
Changes in v3:
- No changes.
2.38.0.rc1.362.ged0d419d3c-goog

patchwork-b...@kernel.org

unread,
Sep 29, 2022, 3:45:38 AMSep 29
to Abhishek Pandit-Subedi, linux-b...@vger.kernel.org, abhishe...@chromium.org, syzk...@googlegroups.com, da...@davemloft.net, edum...@google.com, ku...@kernel.org, johan....@gmail.com, luiz....@gmail.com, mar...@holtmann.org, pab...@redhat.com, linux-...@vger.kernel.org, net...@vger.kernel.org
Hello:

This patch was applied to bluetooth/bluetooth-next.git (master)
by Luiz Augusto von Dentz <luiz.vo...@intel.com>:

On Wed, 28 Sep 2022 13:00:30 -0700 you wrote:
> From: Abhishek Pandit-Subedi <abhishe...@chromium.org>
>
> Suspend notifier should only be registered and unregistered once per
> hdev. Simplify this by only registering during driver registration and
> simply exiting early when HCI_USER_CHANNEL is set.
>
> Reported-by: syzbot <syzk...@googlegroups.com>
> Fixes: 359ee4f834f5 (Bluetooth: Unregister suspend with userchannel)
> Signed-off-by: Abhishek Pandit-Subedi <abhishe...@chromium.org>
>
> [...]

Here is the summary with links:
- [v3] Bluetooth: Prevent double register of suspend
https://git.kernel.org/bluetooth/bluetooth-next/c/4b8af331bb4d

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html


Reply all
Reply to author
Forward
0 new messages