Commit 8f9d02f4 authored by Kuba Pawlak's avatar Kuba Pawlak Committed by Marcel Holtmann

Bluetooth: Remove SCO fragments on connection close

SCO packet reassembler may have a fragment of SCO packet, from
previous connection, cached and not removed when SCO connection
is ended. Packets from new SCO connection are then going to be
attached to that fragment, creating an invalid SCO packets.

Controllers like Intel's WilkinsPeak are always fragmenting
SCO packet into 3 parts (#1, #2, #3). Packet #1 contains
SCO header and audio data, others just audio data. if there is
a fragment cached from previous connection, i.e. #1, first
SCO packet from new connection is going to be attached to it
creating packet consisting of fragments #1-#1-#2. This will
be forwarded to upper layers. After that, fragment #3 is going
to be used as a starting point for another SCO packet.
It does not contain a SCO header, but the code expects it,
casts a SCO header structure on it, and reads whatever audio
data happens to be there as SCO packet length and handle.
From that point on, we are assembling random data into SCO
packets. Usually it recovers quickly as initial audio data
contains mostly zeros (muted stream), but setups of over
4 seconds were observed.
Issue manifests itself by printing on the console:
Bluetooth: hci0 SCO packet for unknown connection handle 48
Bluetooth: hci0 SCO packet for unknown connection handle 2560
Bluetooth: hci0 SCO packet for unknown connection handle 12288
It may also show random handles if audio data was non-zeroed.
Hcidump shows SCO packets with random length and handles.

Few messages with handle 0 at connection creation are OK
for some controllers (like WilkinsPeak), as there are SCO packets
with zeroed handle at the beginning (possible controller bug).
Few of such messages at connection end, with a handle looking
sane (around 256, 512, 768 ...) is also OK, as these are last
SCO packets that were assembled and sent up, before connection
was ended, but were not handled in time.

This issue may still manifest itself on WilkinsPeak as it sometimes,
at SCO connection creation, does not send third fragment of first
SCO packet (#1-#2-#1-#2-#3...). This is a firmware bug and this
patch does not address it.
Signed-off-by: default avatarKuba Pawlak <kubax.t.pawlak@intel.com>
Signed-off-by: default avatarMarcel Holtmann <marcel@holtmann.org>
parent f7552473
...@@ -1277,6 +1277,20 @@ static void btusb_work(struct work_struct *work) ...@@ -1277,6 +1277,20 @@ static void btusb_work(struct work_struct *work)
clear_bit(BTUSB_ISOC_RUNNING, &data->flags); clear_bit(BTUSB_ISOC_RUNNING, &data->flags);
usb_kill_anchored_urbs(&data->isoc_anchor); usb_kill_anchored_urbs(&data->isoc_anchor);
/* When isochronous alternate setting needs to be
* changed, because SCO connection has been added
* or removed, a packet fragment may be left in the
* reassembling state. This could lead to wrongly
* assembled fragments.
*
* Clear outstanding fragment when selecting a new
* alternate setting.
*/
spin_lock(&data->rxlock);
kfree_skb(data->sco_skb);
data->sco_skb = NULL;
spin_unlock(&data->rxlock);
if (__set_isoc_interface(hdev, new_alts) < 0) if (__set_isoc_interface(hdev, new_alts) < 0)
return; return;
} }
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment