Bug 219458 - bluetooth connection fails after L2CAP: Fix uaf in l2cap_connect patch
Summary: bluetooth connection fails after L2CAP: Fix uaf in l2cap_connect patch
Status: RESOLVED CODE_FIX
Alias: None
Product: Drivers
Classification: Unclassified
Component: Bluetooth (show other bugs)
Hardware: All Linux
: P3 normal
Assignee: linux-bluetooth@vger.kernel.org
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2024-11-02 22:26 UTC by Giuseppe Ghibò
Modified: 2024-11-18 15:53 UTC (History)
2 users (show)

See Also:
Kernel Version:
Subsystem:
Regression: No
Bisected commit-id:


Attachments
btmon output (connection working) (9.61 KB, text/plain)
2024-11-07 20:46 UTC, Giuseppe Ghibò
Details
btmon output (connection not working) (5.45 KB, text/plain)
2024-11-07 20:47 UTC, Giuseppe Ghibò
Details
Bluetooth: hci_core: Fix calling mgmt_device_connected (1.16 KB, patch)
2024-11-08 16:31 UTC, Luiz Von Dentz
Details | Diff

Description Giuseppe Ghibò 2024-11-02 22:26:51 UTC
Hi guys.

After a recent kernel update I get problems in bluetooth connectivity. Mostly some bluetooth audio devices (like speakers or headphones) are getting very difficult to connect, even if already paired. And when the connection is finally established they do not seem to do it in the correct way, to the extent that the devices are not later correctly attached to pulseaudio or pipewire, and not listed there. I thought the problem could be in bluez's bluetoothd, so I tried several version of bluez, up to 5.78, including the current git version, but mostly the behaviour hadn't changed.

I noticed this behaviour in 6.6.59 (including up to the current stable-queue for 6.6.59 as of today, which will be merged into next 6.6.60). Ditto for kernel series 6.11.x, in particular 6.11.5 and 6.11.6 show the same behaviour.

After a long series of attempts, I found the culprit can be traced back to these single patch with subject "Bluetooth: L2CAP: Fix uaf in l2cap_connect":

https://git.kernel.org/pub/scm/linux/kernel/git/stable/stable-queue.git/tree/releases/6.6.55/bluetooth-l2cap-fix-uaf-in-l2cap_connect.patch

and

https://git.kernel.org/pub/scm/linux/kernel/git/stable/stable-queue.git/tree/releases/6.11.3/bluetooth-l2cap-fix-uaf-in-l2cap_connect.patch

which were included starting from kernels 6.6.55 and 6.11.3.

By reverting this patch, the connection with the device gets back working again like a breeze, either at connecting and disconnecting several times in a row, while before this reverting the connection was pretty clumsy.

Apparently this patch was for a security fix, but probably it's still incomplete, as there is this side effect.

Difficult to track in logs, as logsa are pretty verbose anyway, as even in a situation where a connection working there are a lot of warnings; mostly in a broken connection with the patch included, logs are like:

bluetoothd[10741]: Failed to set mode: Failed (0x03)
bluetoothd[10741]: No matching connection for device
bluetoothd[10741]: No matching connection for device

while in a working connection, with the patch reverted, logs are like:

bluetoothd[10863]: Failed to set mode: Failed (0x03)
...
bluetoothd[10863]: No matching connection for device
...
bluetoothd[10863]: /org/bluez/hci0/dev_<mac_address>/sep2/fd2: fd(42) ready
Comment 1 Luiz Von Dentz 2024-11-05 14:59:51 UTC
(In reply to Giuseppe Ghibò from comment #0)
> Difficult to track in logs, as logsa are pretty verbose anyway, as even in a
> situation where a connection working there are a lot of warnings; mostly in
> a broken connection with the patch included, logs are like:
> 
> bluetoothd[10741]: Failed to set mode: Failed (0x03)
> bluetoothd[10741]: No matching connection for device
> bluetoothd[10741]: No matching connection for device
> 
> while in a working connection, with the patch reverted, logs are like:
> 
> bluetoothd[10863]: Failed to set mode: Failed (0x03)
> ...
> bluetoothd[10863]: No matching connection for device
> ...
> bluetoothd[10863]: /org/bluez/hci0/dev_<mac_address>/sep2/fd2: fd(42) ready

Could you please attach the btmon trace when this happens?
Comment 2 Giuseppe Ghibò 2024-11-07 20:46:31 UTC
Created attachment 307181 [details]
btmon output (connection working)
Comment 3 Giuseppe Ghibò 2024-11-07 20:47:25 UTC
Created attachment 307182 [details]
btmon output (connection not working)
Comment 4 Giuseppe Ghibò 2024-11-07 20:58:13 UTC
btmon output added. As you can see in the working case the behaviour diverges from packet #16 (in the first case, working) and after packet #94 in the "not working" case, both calls ACL Data RX with L2CAP: Information Request:, but in the working case it's followed by HCI Event: Read Remote... and other commands, while in the not working case it's immediately followed by "MGMT Event: Device Connected", like if something was truncated.

[working]
> ACL Data RX: Handle 256 flags 0x02 dlen 10              #16 [hci0] 413.700565
      L2CAP: Information Request (0x0a) ident 1 len 2
        Type: Extended features supported (0x0002)
> HCI Event: Read Remote Extended Featu.. (0x23) plen 13  #17 [hci0] 413.700588


[not working]
> ACL Data RX: Handle 256 flags 0x02 dlen 10               #94 [hci0] 20.205711
      L2CAP: Information Request (0x0a) ident 7 len 2
        Type: Extended features supported (0x0002)
@ MGMT Event: Device Connected (0x000b) plen 13       {0x0001} [hci0] 20.205764
        BR/EDR Address: AA:BB:CC:DD:EE:FF (Company)
Comment 5 Luiz Von Dentz 2024-11-08 16:31:18 UTC
Created attachment 307186 [details]
Bluetooth: hci_core: Fix calling mgmt_device_connected

Please try with the following patch.
Comment 6 Luiz Von Dentz 2024-11-11 14:56:15 UTC
(In reply to Giuseppe Ghibò from comment #4)
> btmon output added. As you can see in the working case the behaviour
> diverges from packet #16 (in the first case, working) and after packet #94
> in the "not working" case, both calls ACL Data RX with L2CAP: Information
> Request:, but in the working case it's followed by HCI Event: Read Remote...
> and other commands, while in the not working case it's immediately followed
> by "MGMT Event: Device Connected", like if something was truncated.
> 
> [working]
> > ACL Data RX: Handle 256 flags 0x02 dlen 10              #16 [hci0]
> 413.700565
>       L2CAP: Information Request (0x0a) ident 1 len 2
>         Type: Extended features supported (0x0002)
> > HCI Event: Read Remote Extended Featu.. (0x23) plen 13  #17 [hci0]
> 413.700588
> 
> 
> [not working]
> > ACL Data RX: Handle 256 flags 0x02 dlen 10               #94 [hci0]
> 20.205711
>       L2CAP: Information Request (0x0a) ident 7 len 2
>         Type: Extended features supported (0x0002)
> @ MGMT Event: Device Connected (0x000b) plen 13       {0x0001} [hci0]
> 20.205764
>         BR/EDR Address: AA:BB:CC:DD:EE:FF (Company)

Any chance to check with the patch above?
Comment 7 Pablo Yanez 2024-11-12 00:53:13 UTC
Hi Luiz

I can check the patch. I ran on the same problem (when Updating from 6.6.52 to 6.6.58 on Gentoo) and I submitted a bug report there. They asked me to do a git bisect of linux-vanilla and my bisect also reported that patch.

I'll try applying the patch and let you know if this solves the problem for me.
Comment 8 Giuseppe Ghibò 2024-11-12 01:18:41 UTC
(In reply to Luiz Von Dentz from comment #6)
> (In reply to Giuseppe Ghibò from comment #4)
> > btmon output added. As you can see in the working case the behaviour
> > diverges from packet #16 (in the first case, working) and after packet #94
> > in the "not working" case, both calls ACL Data RX with L2CAP: Information
> > Request:, but in the working case it's followed by HCI Event: Read
> Remote...
> > and other commands, while in the not working case it's immediately followed
> > by "MGMT Event: Device Connected", like if something was truncated.
> > 
> > [working]
> > > ACL Data RX: Handle 256 flags 0x02 dlen 10              #16 [hci0]
> > 413.700565
> >       L2CAP: Information Request (0x0a) ident 1 len 2
> >         Type: Extended features supported (0x0002)
> > > HCI Event: Read Remote Extended Featu.. (0x23) plen 13  #17 [hci0]
> > 413.700588
> > 
> > 
> > [not working]
> > > ACL Data RX: Handle 256 flags 0x02 dlen 10               #94 [hci0]
> > 20.205711
> >       L2CAP: Information Request (0x0a) ident 7 len 2
> >         Type: Extended features supported (0x0002)
> > @ MGMT Event: Device Connected (0x000b) plen 13       {0x0001} [hci0]
> > 20.205764
> >         BR/EDR Address: AA:BB:CC:DD:EE:FF (Company)
> 
> Any chance to check with the patch above?

Yes, just finished applying the patch (and not reverting the previous) to current 6.6.60+stable-queue as of today, and it didn't show the previous problem anymore.
Comment 9 Pablo Yanez 2024-11-12 02:10:28 UTC
Hi Luiz

Like Guiseppe, on 6.6.58 I didn't revert the original commit and applied the patch (thus removing only the if clause and the mgmt_device_connected call) and it seems that the problem is gone.

I was able to connect and disconnect multiple times in succession, I was able to remove my USB-Bluetooth device and insert it again and was able to connect again, I tried that several times and it always worked. Now it seems fine.
Comment 10 Luiz Von Dentz 2024-11-12 18:01:26 UTC
Patch has been applied to bluetooth-next, pull-request in its way:

https://patchwork.kernel.org/project/bluetooth/patch/20241112175326.930800-1-luiz.dentz@gmail.com/
Comment 11 Giuseppe Ghibò 2024-11-18 15:53:16 UTC
I've seen this patch is now included in 6.6.621, 6.11.9, 6.12.0. Thanks. Closing.
Comment 12 Giuseppe Ghibò 2024-11-18 15:53:41 UTC
Of course 6.6.62 not 6.6.621.

Note You need to log in before you can comment on or make changes to this bug.