Re: kernel panic happens when disconnecting Bluetooth headset

On Tue, Sep 22, 2009 at 01:18:26PM -0700, Nick Pelly wrote:

On Mon, Sep 21, 2009 at 6:29 PM, Nick Pelly [off-list ref] wrote:

quoted

On Mon, Sep 21, 2009 at 5:52 PM, Nick Pelly [off-list ref] wrote:

quoted

On Mon, Sep 14, 2009 at 2:10 AM, Lan Zhu [off-list ref] wrote:

quoted

Hi Marcel,

2009/9/12 Marcel Holtmann [off-list ref]:

quoted

Hi Zhu,

quoted

quoted

quoted

We met a issue that kernel panic happens when disconnecting some kinds
of Bluetooth headset, then we did some analysis and made some changes
on kernel code which have avoided the panic happening. Would you
please help to check if our analysis and fix is correct?

=============
Issue description
=============
On Android platform(kernel 2.6.29), disconnecting Bluetooth headset
may cause kernel panic on certain conditions.

(Pre-condition is android paired with headset.)
Initiate the connection from android, disconnect it from android, result is OK.
Initiate the connection from android, disconnect it from headset, result is OK.
Initiate the connection from headset, disconnect it from headset, result is OK.
Initiate the connection from headset, disconnect it from android, for
Motorola H12 headset, result is OK.
Initiate the connection from headset, disconnect it from android, for
Motorola H620/560 headset, result is kernel panic.

=============
Kernel panic point
=============
kernel panic at __list_del() in the function rfcomm_session_del() ,
panic reason is "Unable to handle kernel paging request at virtual
address 00200200"

=============
Kernel log analysis
=============
rfcomm_session_del() is still called after the session entry is
removed from the list. Then __list_del() will cause kernel panic
because of the incorrect pointer. This situation occurs when calling
rfcomm_recv_ua() when the socket state is BT_CLOSED . So we need to
find out why the socket state become BT_CLOSED before we calling
rfcomm_recv_ua().

# [  171.677429] rfcomm_sock_sendmsg: sock ce9a0960, sk cc5c4c00
[  171.683532] rfcomm_dlc_send: dlc cd3fe920 mtu 255 len 14
[  171.689422] rfcomm_process_rx: session cc751be0 state 1 qlen 0
[  171.695709] rfcomm_process_rx: @@@ @@@ sk_state = 1
[  171.701110] rfcomm_process_dlcs: session cc751be0 state 1
[  171.706939] rfcomm_process_tx: dlc cd3fe920 state 1 cfc 40
rx_credits 33 tx_credits 31
[  171.715515] rfcomm_send_frame: session cc751be0 len 18
[  171.721130] rfcomm_session_put: in rfcomm_session_put, s->refcnt = 3
[  174.127807] rfcomm_sock_sendmsg: sock ce9a0960, sk cc5c4c00
[  174.134490] rfcomm_dlc_send: dlc cd3fe920 mtu 255 len 6
[  174.141540] rfcomm_process_rx: session cc751be0 state 1 qlen 0
[  174.148498] rfcomm_process_rx: @@@ @@@ sk_state = 1
[  174.154968] rfcomm_process_dlcs: session cc751be0 state 1
[  174.161437] rfcomm_process_tx: dlc cd3fe920 state 1 cfc 40
rx_credits 33 tx_credits 30
[  174.171173] rfcomm_send_frame: session cc751be0 len 10
[  174.177642] rfcomm_session_put: in rfcomm_session_put, s->refcnt = 3
[  174.205932] rfcomm_sock_release: sock ce9a0960, sk cc5c4c00
[  174.212707] rfcomm_sock_shutdown: sock ce9a0960, sk cc5c4c00
[  174.220031] __rfcomm_sock_close: sk cc5c4c00 state 1 socket ce9a0960
[  174.227508] __rfcomm_dlc_close: dlc cd3fe920 state 1 dlci 20 err 0
session cc751be0
[  174.236877] rfcomm_send_disc: cc751be0 dlci 20
[  174.242706] rfcomm_send_frame: session cc751be0 len 4
[  174.248962] rfcomm_dlc_set_timer: dlc cd3fe920 state 8 timeout 2560
[  174.256835] rfcomm_sock_kill: sk cc5c4c00 state 1 refcnt 2
[  174.263336] rfcomm_sock_destruct: sk cc5c4c00 dlc cd3fe920
[  174.399444] rfcomm_l2data_ready: ccf70400 bytes 4
[  174.404724] rfcomm_process_rx: session cc751be0 state 1 qlen 1
[  174.411010] rfcomm_process_rx: @@@ @@@ sk_state = 1
[  174.416412] rfcomm_recv_ua: session cc751be0 state 1 dlci 20
[  174.422515] __rfcomm_dlc_close: dlc cd3fe920 state 9 dlci 20 err 0
session cc751be0
[  174.430816] rfcomm_dlc_clear_timer: dlc cd3fe920 state 9
[  174.436553] rfcomm_dlc_unlink: dlc cd3fe920 refcnt 1 session cc751be0
[  174.443572] rfcomm_dlc_free: cd3fe920
[  174.447570] rfcomm_session_put: in rfcomm_session_put, s->refcnt = 3
[  174.454528] rfcomm_send_disc: cc751be0 dlci 0
[  174.459259] rfcomm_send_frame: session cc751be0 len 4
[  174.464904] rfcomm_process_dlcs: session cc751be0 state 8
[  174.470703] rfcomm_session_put: in rfcomm_session_put, s->refcnt = 2
[  174.898284] rfcomm_l2data_ready: ccf70400 bytes 4
[  174.903442] rfcomm_l2state_change: ccf70400 state 9
[  174.908874] rfcomm_process_rx: session cc751be0 state 8 qlen 1
[  174.915130] rfcomm_process_rx: @@@ @@@ sk_state = 9
[  174.920562] rfcomm_recv_ua: session cc751be0 state 8 dlci 0
[  174.926574] rfcomm_session_put: in rfcomm_session_put, s->refcnt = 2
[  174.933532] rfcomm_process_rx: @@@ @@@ sk_state == BT_CLOSED , s->initiator=0
[  174.941253] rfcomm_session_put: in rfcomm_session_put, s->refcnt = 1
[  174.948211] rfcomm_session_del: session cc751be0 state 8
[  174.953918] @@@@ in rfcomm_session_del()
[  174.958312] @@@@ s->list = cc751be0
[  174.962097] @@@@ s->list.next = ccbfe9a0
[  174.966369] @@@@ s->list.prev = c047d524
[  174.970733] @@@@ list is valid, call list_del()
[  174.975646] @@@@ after list_del()
[  174.979278] @@@@ s->list = cc751be0
[  174.983184] @@@@ s->list.next = 00100100
[  174.987457] @@@@ s->list.prev = 00200200
[  174.991729] rfcomm_session_close: session cc751be0 state 8 err 104
[  174.998504] rfcomm_session_put: in rfcomm_session_put, s->refcnt = 1
[  175.005310] rfcomm_session_del: session cc751be0 state 9
[  175.011169] @@@@ in rfcomm_session_del()
[  175.015441] @@@@ s->list = cc751be0
[  175.019409] @@@@ s->list.next = 00100100
[  175.023651] @@@@ s->list.prev = 00200200
[  175.027923] @@@@ list is valid, call list_del()
[  175.032958] Unable to handle kernel paging request at virtual
address 00200200
[  175.040679] pgd = c0004000
[  175.043792] [00200200] *pgd=00000000
[  175.047821] Internal error: Oops: 817 [#1]
[  175.052246] Modules linked in:
[  175.055725] CPU: 0    Not tainted  (2.6.29-omap1-dirty #34)
[  175.061859] PC is at rfcomm_session_del+0x6c/0x108
[  175.067047] LR is at release_console_sem+0x190/0x1a0
[  175.072509] pc : [<c033ded8>]    lr : [<c0066308>]    psr: 60000013
[  175.072509] sp : cc1abf38  ip : cc1abe68  fp : cc1abf4c
[  175.084960] r10: cc751c04  r9 : c036d2fc  r8 : cc751be0
[  175.090545] r7 : 00000068  r6 : cc751c04  r5 : 00000009  r4 : cc751be0
[  175.097656] r3 : 00100100  r2 : 00100100  r1 : 00200200  r0 : c0422876

=============
HCI log analysis
=============
Compare the hcidump log of the correct case with the one of the panic
case, we found there is only one difference in the message sequence.
In the panic case, headset send L2CAP Disconn_Req immediately after
sending rfcomm UA frame to android. We think this is the reason that
cause the socket state become BT_CLOSED.

Please compare these two log, pay attention to the message direction
of the last Disconn_Req.


Log of correct case:
----------------------------


009-09-10 15:27:28.963519 < ACL data: handle 1 flags 0x02 dlen 22
   L2CAP(d): cid 0x0047 len 18 [psm 3]
     RFCOMM(d): UIH: cr 0 dlci 20 pf 0 ilen 14 fcs 0xeb
     0000: 0d 0a 2b 43 49 45 56 3a  20 37 2c 33 0d 0a        ..+CIEV: 7,3..
009-09-10 15:27:28.967272 > HCI Event: Number of Completed Packets (0x13) plen

   handle 1 packets 1
009-09-10 15:27:29.243945 < ACL data: handle 1 flags 0x02 dlen 8
   L2CAP(d): cid 0x0047 len 4 [psm 3]
     RFCOMM(s): DISC: cr 0 dlci 20 pf 1 ilen 0 fcs 0x7d
009-09-10 15:27:29.247363 > HCI Event: Number of Completed Packets (0x13) plen

   handle 1 packets 1
009-09-10 15:27:29.274890 > ACL data: handle 1 flags 0x02 dlen 8
   L2CAP(d): cid 0x0040 len 4 [psm 3]
     RFCOMM(s): UA: cr 0 dlci 20 pf 1 ilen 0 fcs 0x57
009-09-10 15:27:29.296343 < ACL data: handle 1 flags 0x02 dlen 8
   L2CAP(d): cid 0x0047 len 4 [psm 3]
     RFCOMM(s): DISC: cr 0 dlci 0 pf 1 ilen 0 fcs 0x9c
009-09-10 15:27:29.298480 > HCI Event: Number of Completed Packets (0x13) plen

   handle 1 packets 1
009-09-10 15:27:29.319873 > ACL data: handle 1 flags 0x02 dlen 8
   L2CAP(d): cid 0x0040 len 4 [psm 3]
     RFCOMM(s): UA: cr 0 dlci 0 pf 1 ilen 0 fcs 0xb6
009-09-10 15:27:29.320727 < ACL data: handle 1 flags 0x02 dlen 12
   L2CAP(s): Disconn req: dcid 0x0047 scid 0x0040
009-09-10 15:27:29.323474 > HCI Event: Number of Completed Packets (0x13) plen

   handle 1 packets 1
009-09-10 15:27:29.337237 > ACL data: handle 1 flags 0x02 dlen 12
   L2CAP(s): Disconn rsp: dcid 0x0047 scid 0x0040




log of panic case:
------------------------



2009-09-10 13:34:24.020208 < ACL data: handle 1 flags 0x02 dlen 22
    L2CAP(d): cid 0x0041 len 18 [psm 3]
      RFCOMM(d): UIH: cr 0 dlci 20 pf 0 ilen 14 fcs 0xeb
      0000: 0d 0a 2b 43 49 45 56 3a  20 37 2c 33 0d 0a        ..+CIEV: 7,3..
2009-09-10 13:34:24.281256 > HCI Event: Number of Completed Packets (0x13) plen
5
    handle 1 packets 1
2009-09-10 13:34:24.083580 < ACL data: handle 1 flags 0x02 dlen 8
    L2CAP(d): cid 0x0041 len 4 [psm 3]
      RFCOMM(s): DISC: cr 0 dlci 20 pf 1 ilen 0 fcs 0x7d
2009-09-10 13:34:24.529442 > HCI Event: Number of Completed Packets (0x13) plen
5
    handle 1 packets 1
2009-09-10 13:34:24.531914 > ACL data: handle 1 flags 0x02 dlen 8
    L2CAP(d): cid 0x0040 len 4 [psm 3]
      RFCOMM(s): UA: cr 0 dlci 20 pf 1 ilen 0 fcs 0x57
2009-09-10 13:34:24.533135 < ACL data: handle 1 flags 0x02 dlen 8
    L2CAP(d): cid 0x0041 len 4 [psm 3]
      RFCOMM(s): DISC: cr 0 dlci 0 pf 1 ilen 0 fcs 0x9c
2009-09-10 13:34:25.028649 > HCI Event: Number of Completed Packets (0x13) plen
5
    handle 1 packets 1
2009-09-10 13:34:25.032128 > ACL data: handle 1 flags 0x02 dlen 8
    L2CAP(d): cid 0x0040 len 4 [psm 3]
      RFCOMM(s): UA: cr 0 dlci 0 pf 1 ilen 0 fcs 0xb6
2009-09-10 13:34:25.032341 > ACL data: handle 1 flags 0x02 dlen 12
    L2CAP(s): Disconn req: dcid 0x0040 scid 0x0041
2009-09-10 13:34:25.032646 < ACL data: handle 1 flags 0x02 dlen 12
    L2CAP(s): Disconn rsp: dcid 0x0040 scid 0x0041

=============
Analysis Result
=============
For some kinds of Bluetooth headset such as Motorola H560/H620 which
are based on BCM2044S, they will send L2CAP Disconn_Req command right
after sending rfcomm UA frame. This L2CAP Disconn_Req will cause the
rfcomm socket state become BT_CLOSED before completely handling UA
frame, thus it will cause kernel panic. I think we can ignore the
received rfcomm frames if socket state is BT_CLOSED, because it
doesn't make sense in the BT_CLOSED state.


=============
Changed Code
=============
We changed the code in the function rfcomm_process_rx() in
net/bluetooth/rfcomm/core.c, check the socket state first before
handling the received framew. If the socket state is BT_CLOSED, we
don't handle any rfcomm frames but just close the session.

The change is like below

+       if (sk->sk_state != BT_CLOSED) {
        /* Get data directly from socket receive queue without copying it. */
        while ((skb = skb_dequeue(&sk->sk_receive_queue))) {
                skb_orphan(skb);
                rfcomm_recv_frame(s, skb);
        }
-
-       if (sk->sk_state == BT_CLOSED) {
+       } else {
                if (!s->initiator)
                        rfcomm_session_put(s);

                rfcomm_session_close(s, sk->sk_err);
         }

so I do see the issue here, but I don't agree with the fix since it
changes behavior that might cause other issues. So in case the frame
processing leads to sk->sk_state == BT_CLOSED we are not closing the
connection anymore if we make it depend on a state before the frame
processing. And nothing guarantees that rfcomm_process_rx gets scheduled
again.

diff --git a/net/bluetooth/rfcomm/core.c b/net/bluetooth/rfcomm/core.c
index 94b3388..606143b 100644
--- a/net/bluetooth/rfcomm/core.c
+++ b/net/bluetooth/rfcomm/core.c

@@ -1798,12 +1798,16 @@ static inline void rfcomm_process_rx(struct rfcomm_session *s)

       BT_DBG("session %p state %ld qlen %d", s, s->state, skb_queue_len(&sk->sk_receive_queue));

+       rfcomm_session_hold(s);
+
       /* Get data directly from socket receive queue without copying it. */
       while ((skb = skb_dequeue(&sk->sk_receive_queue))) {
               skb_orphan(skb);
               rfcomm_recv_frame(s, skb);
       }

+       rfcomm_session_put(s);
+
       if (sk->sk_state == BT_CLOSED) {
               if (!s->initiator)
                       rfcomm_session_put(s);

What does the above patch do for you? Since if I read it correctly, then
the rfcomm_recv_ua causes the rfcomm_session_put to trigger the closing
of the session. And then in this case it is delayed until after all
frames are processed.

I've tried your patch but unfortunately kernel panic still happened.

From the log I noticed that if rfcomm_l2state_change is called before
rfcomm_process_rx, kernel panic will happen definitely.

Below lines are in the correct log,

[  139.323852] rfcomm_l2data_ready: ccf70000 bytes 4
[  139.346252] rfcomm_process_rx: session ccb94ce0 state 8 qlen 1
...
[  139.457519] rfcomm_l2state_change: ccf70000 state 9
(disconnect ok)

In the above case, when process_rx, the code in the condition "if
(sk->sk_state == BT_CLOSED)" will never run.

Below lines are in the panic log,

[  174.898284] rfcomm_l2data_ready: ccf70400 bytes 4
[  174.903442] rfcomm_l2state_change: ccf70400 state 9
[  174.908874] rfcomm_process_rx: session cc751be0 state 8 qlen 1
...
( then panic)

In the above case, sk_state is changed to 9 (BT_CLOSED) firstly,  then
process_rx, so the code in the condition "if (sk->sk_state ==
BT_CLOSED) " will be run, it will call session_put twice. I think this
is the root cause of panic.

I know why it happens, that is not the problem. My point is not to break
current scheduling assumptions.

So if you move the rfcomm_session_put() now at the end of the function,
then it should be fine, right?

Regards

Marcel

You are right. I moved the rfcomm_session_put() at the end of
rfcomm_process_tx() then kernel panic doesn't happen any longer.

The changed code is like below,

@@ -1796,6 +1796,8 @@ static inline void rfcomm_process_rx(struct rfcomm_session

       BT_DBG("session %p state %ld qlen %d", s, s->state, skb_queue_len(&sk->s

+       rfcomm_session_hold(s);
+
       /* Get data directly from socket receive queue without copying it. */
       while ((skb = skb_dequeue(&sk->sk_receive_queue))) {
               skb_orphan(skb);

@@ -1808,6 +1810,8 @@ static inline void rfcomm_process_rx(struct rfcomm_session

               rfcomm_session_close(s, sk->sk_err);
       }
+
+       rfcomm_session_put(s);
 }

 static inline void rfcomm_accept_connection(struct rfcomm_session *s)

Please submit this change to bluez release.

Unfortunately, with this change I get a panic disconnecting from
Motorola H270 in the case that the headset initiated RFCOMM and we
disconnect RFCOMM.

Here is the hcidump:

2009-09-21 17:22:37.384811 < ACL data: handle 1 flags 0x02 dlen 22
   L2CAP(d): cid 0x0041 len 18 [psm 3]
     RFCOMM(d): UIH: cr 0 dlci 20 pf 0 ilen 14 fcs 0xeb
     0000: 0d 0a 2b 43 49 45 56 3a  20 37 2c 33 0d 0a        ..+CIEV: 7,3..
2009-09-21 17:22:37.502273 > HCI Event: Number of Completed Packets
(0x13) plen 5
   handle 1 packets 1
2009-09-21 17:22:37.788895 < ACL data: handle 1 flags 0x02 dlen 8
   L2CAP(d): cid 0x0041 len 4 [psm 3]
     RFCOMM(s): DISC: cr 0 dlci 20 pf 1 ilen 0 fcs 0x7d
2009-09-21 17:22:37.906204 > HCI Event: Number of Completed Packets
(0x13) plen 5
   handle 1 packets 1
2009-09-21 17:22:37.933090 > ACL data: handle 1 flags 0x02 dlen 8
   L2CAP(d): cid 0x0040 len 4 [psm 3]
     RFCOMM(s): UA: cr 0 dlci 20 pf 1 ilen 0 fcs 0x57
2009-09-21 17:22:38.636764 < ACL data: handle 1 flags 0x02 dlen 8
   L2CAP(d): cid 0x0041 len 4 [psm 3]
     RFCOMM(s): DISC: cr 0 dlci 0 pf 1 ilen 0 fcs 0x9c
2009-09-21 17:22:38.744125 > HCI Event: Number of Completed Packets
(0x13) plen 5
   handle 1 packets 1
2009-09-21 17:22:38.763687 > ACL data: handle 1 flags 0x02 dlen 8
   L2CAP(d): cid 0x0040 len 4 [psm 3]
     RFCOMM(s): UA: cr 0 dlci 0 pf 1 ilen 0 fcs 0xb6
2009-09-21 17:22:38.783554 > ACL data: handle 1 flags 0x02 dlen 12
   L2CAP(s): Disconn req: dcid 0x0040 scid 0x0041
2009-09-21 17:22:39.029526 < ACL data: handle 1 flags 0x02 dlen 12
   L2CAP(s): Disconn rsp: dcid 0x0040 scid 0x0041
2009-09-21 17:22:39.136581 > HCI Event: Number of Completed Packets
(0x13) plen 5
   handle 1 packets 1
2009-09-21 17:22:41.337203 > HCI Event: Disconn Complete (0x05) plen 4
   status 0x00 handle 1 reason 0x13
   Reason: Remote User Terminated Connection

And the panic:

<7>[ 3161.665557] rfcomm:rfcomm_session_del: session c9c06ad0 state 9
<7>[ 3161.671905] l2cap:l2cap_sock_release: sock cea04360, sk c97f02f8
<7>[ 3161.678497] l2cap:l2cap_sock_shutdown: sock cea04360, sk c97f02f8
<7>[ 3161.685028] l2cap:l2cap_sock_kill: sk c97f02f8 state 9
<7>[ 3161.695587] l2cap:l2cap_sock_destruct: sk c97f02f8
<4>[ 3161.700805] npelly 1911 rfcomm_process_sessions session c9c06ad0
refcnt 1802201963
<7>[ 3161.709014] rfcomm:rfcomm_process_dlcs: session c9c06ad0 state 1802201963
<7>[ 3161.716308] rfcomm:rfcomm_process_dlcs: session c9c06ad0 dlc 6b6b6b6b
<1>[ 3161.726776] Unable to handle kernel paging request at virtual
address 6b6b6b6b
<1>[ 3161.734619] pgd = c0004000
<1>[ 3161.737609] [6b6b6b6b] *pgd=00000000
<4>[ 3161.741638] Internal error: Oops: 5 [#1] PREEMPT
<4>[ 3161.746734] Modules linked in:
<4>[ 3161.750213] CPU: 0    Not tainted
(2.6.29-omap1-07358-g9a3fd55-dirty #206)
<4>[ 3161.757629] PC is at rfcomm_process_dlcs+0x108/0x590
<4>[ 3161.762969] LR is at preempt_schedule+0x44/0x54
<4>[ 3161.767852] pc : [<c03911f4>]    lr : [<c03a27c4>]    psr: 60000113
<4>[ 3161.767883] sp : ccdf9e80  ip : ccdf9dd8  fp : ccdf9edc
<4>[ 3161.780273] r10: 00000000  r9 : c9c06af4  r8 : c9c06ad0
<4>[ 3161.786010] r7 : 00000000  r6 : c9c06ad0  r5 : c4c68680  r4 : 6b6b6b6b
<4>[ 3161.792968] r3 : c9c06ae0  r2 : ccdf8000  r1 : c61a8940  r0 : 0000004c
<4>[ 3161.800079] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM
Segment kernel
<4>[ 3161.807983] Control: 10c5387d  Table: 86db8019  DAC: 00000017
<4>[ 3161.814147]
<4>[ 3161.814147] PC: 0xc0391174:
[...]
<4>[ 3162.973175] Backtrace:
<4>[ 3162.976013] [<c03910ec>] (rfcomm_process_dlcs+0x0/0x590) from
[<c03930b0>] (rfcomm_process_sessions+0x1a34/0x1a9c)
<4>[ 3162.987579] [<c039167c>] (rfcomm_process_sessions+0x0/0x1a9c)
from [<c03932ec>] (rfcomm_run+0x1d4/0x2ac)
<4>[ 3162.998199] [<c0393118>] (rfcomm_run+0x0/0x2ac) from
[<c008e7d8>] (kthread+0x5c/0x94)
<4>[ 3163.013763] [<c008e77c>] (kthread+0x0/0x94) from [<c007c998>]
(do_exit+0x0/0x714)


Seems like this fix avoids the panic due to calling
rfcomm_session_close() on a deleted session, but does not always
address the unbalanced rfcomm_session_put() which may be the root
cause.

Lan Zhu suspected this in the original post, and his original fix does
in fact fix this panic as well as the originally reported panic,
because it avoids the unbalanced rfcomm_session_put().

Marcel I know you are concerned about the original fix changing
scheduling assumptions, are you able to comment on this further?

Are there any other suggestions for patches for this issue? I have
spent the best part of the day trying to figure this one out, but the
recounting in the rfcomm core is quite subtle and I think it really
needs someone familiar with the code to have a quick look and come up
with the safest patch. I can run tests.

In the mean time, I am doing some testing of Lan Zhu's original fix
and if there are no better suggestions we will run with that one for
Android.

Nick


Some more analysis:

With the RFCOMM connection in idle there are 2 references on s->refcnt

However three references are removed during disconnect with the H270
- rfcomm_process_sessions() -> __rfcomm_dlc_close() -> rfcomm_dlc_unlink()
- rfcomm_process_sessions() -> rfcomm_process_rx() -> rfcomm_recv_ua()
with dlci = 0 and s->state = BT_DISCONN
- rfcomm_process_sessions() -> rfcomm_process_rx() with sk_state =
BT_CLOSED and s->initiator = 0

in that order.

On another headset, for example the Moto H350, we only see the first
two references removed during disconnect.

- rfcomm_process_sessions() -> __rfcomm_dlc_close() -> rfcomm_dlc_unlink()
- rfcomm_process_sessions() -> rfcomm_process_rx() -> rfcomm_recv_ua()
with dlci = 0 and s->state = BT_DISCONN

How about this. We still call rfcomm_process_rx(), but avoid the
rfcomm_session_put() due to RFCOMM UA when the socket state is
BT_CLOSED.

It is less invasive, so might address Marcel's concerns with regard to
scheduling changes.

diff --git a/net/bluetooth/rfcomm/core.c b/net/bluetooth/rfcomm/core.c
index c109a3a..333c6e9 100644
--- a/net/bluetooth/rfcomm/core.c
+++ b/net/bluetooth/rfcomm/core.c

@@ -1105,6 +1105,8 @@ static int rfcomm_recv_ua(struct rfcomm_session

*s, u8 dlci)
               }
       } else {
               /* Control channel */
+               struct socket *sock = s->sock;
+               struct sock *sk = sock->sk;
               switch (s->state) {
               case BT_CONNECT:
                       s->state = BT_CONNECTED;

@@ -1112,7 +1114,8 @@ static int rfcomm_recv_ua(struct rfcomm_session

*s, u8 dlci)
                       break;

               case BT_DISCONN:
-                       rfcomm_session_put(s);
+                       if (sk->sk_state != BT_CLOSED)
+                               rfcomm_session_put(s);
                       break;
               }
       }

I made a minor style improvement and added commit message. Patch available from

http://android.git.kernel.org/?p=kernel/common.git;a=commit;h=1048e007842da2d6440679e1ca80f45438a6369d

What about following fix? (building test only)

--- linux-2.6.orig/net/bluetooth/rfcomm/core.c	2009-09-19 15:25:23.000000000 +0800
+++ linux-2.6/net/bluetooth/rfcomm/core.c	2009-09-23 15:15:30.000000000 +0800

@@ -1927,16 +1927,15 @@ static inline void rfcomm_process_sessio
 
 		rfcomm_session_hold(s);
 
-		switch (s->state) {
-		case BT_BOUND:
-			rfcomm_check_connection(s);
-			break;
-
-		default:
-			rfcomm_process_rx(s);
-			break;
+		rfcomm_check_connection(s);
+
+		if (s->state == BT_CLOSED) {
+			rfcomm_session_put(s);
+			continue;
 		}
 
+		rfcomm_process_rx(s);
+
 		rfcomm_process_dlcs(s);
 
 		rfcomm_session_put(s);