Hi Daniel,

This patch needs to be applied to avoid core.

Thanks
Jijo

On Wed, Sep 19, 2012 at 10:54 AM, Jijo <realjijo@gmail.com> wrote:

Hi All,

Finally i found the issue, 

Here is one of the bad trace for SUBSCRIBE(722bytes) and NOTIFY(1282bytes) which corrupted the memory. The messages came in the order NOTIFY and SUBSCRIBE. The core is generated in a different place but I believe this could be the reason for memory corruption.

Here is the trace UDP Process 27294 processing NOTIFY and Process 27303 processing SUBSCRIBE .

 

The explanation and implementation is below

 

2012-09-19T02:06:17+01:00 [info] sipserver: [27303] INFO: <script>: CI=1-3292@10.233.20.152 -R39 - Entry R-URI=1234 FD=10.233.20.152 SI=10.233.20.152

2012-09-19T02:06:17+01:00 [info] sipserver: [27294] INFO: <script>: CI=1-3292@10.233.20.152 -R1 - Force LAN socket: tcp:10.233.20.151:5060 <null>

2012-09-19T02:06:17+01:00 [info] sipserver: [27303] INFO: <script>: CI=1-3292@10.233.20.152 -R1 - Force LAN socket: tcp:10.233.20.151:5060 <null>

2012-09-19T02:06:17+01:00 [err] sipserver: [27303] ERROR: <core> [tcp_main.c:2357]:  tcp_conn_send_put : calling wbufq_add

2012-09-19T02:06:17+01:00 [err] sipserver: [27303] ERROR: <core> [tcp_main.c:730]: ERROR: wbufq_add(722 bytes): buf:SUBSCRIBE sip:1234@10.233.20.141:5063;transport=tcp SIP/2.0

Record-Route: <sip:10.233.20.151;tran

2012-09-19T02:06:17+01:00 [err] sipserver: [27303] ERROR: <core> [tcp_main.c:747]: ERROR: wbufq_add(722 bytes): first:b00519f4 last:b00519f4

2012-09-19T02:06:17+01:00 [err] sipserver: [27303] ERROR: <core> [tcp_main.c:774]: ERROR: wbufq_add(2 last free crt_size:722): first:b00519f4 last:b00519f4

2012-09-19T02:06:17+01:00 [err] sipserver: [27294] ERROR: <core> [tcp_main.c:796]: ERROR: wbufq_insert(1282 bytes): buf:NOTIFY sip:1234@10.233.20.141:5063;transport=tcp SIP/2.0

Record-Route: <sip:10.233.20.151;transpo

2012-09-19T02:06:17+01:00 [err] sipserver: [27294] ERROR: <core> [tcp_main.c:801]: ERROR: wbufq_insert(2 last free ): first:b00519f4 last:b00519f4

2012-09-19T02:06:17+01:00 [err] sipserver: [27294] ERROR: <core> [tcp_main.c:820]: ERROR: wbufq_insert(22 last free ): first:b00519f4 last:b00519f4

2012-09-19T02:06:17+01:00 [err] sipserver: [27359] ERROR: <core> [tcp_main.c:887]: ERROR: wbufq_run(3 last free ): first:b00519f4 last:b00519f4

2012-09-19T02:06:17+01:00 [crit] sipserver: [27359] : <core> [mem/q_malloc.c:157]: BUG: qm_*: fragm. 0xb00519dc (address 0xb00519f4) end overwritten(0, 0)!

2012-09-19T02:06:18+01:00 [alert] sipserver: [27265] ALERT: <core> [main.c:755]: child process 27359 exited by a signal 11

2012-09-19T02:06:18+01:00 [alert] sipserver: [27265] ALERT: <core> [main.c:758]: core was generated

2012-09-19T02:06:18+01:00 [info] sipserver: [27265] INFO: <core> [main.c:770]: INFO: terminating due to SIGCHLD

 

Process 27294(NOTIFY) created the TCP connection structure for destination IP and just before calling wbufq_insert(), context switch happened and process 27303(SUBSCRIBE) got the cpu. Since the connection structure is already available process 27303 add the SUBSCRIBE message(722 bytes) to wbufq. Afterwards process 27294 got the CPU and invoked wbufq_insert() which basically corrupted the memory due to an overflow with the existing implementation.

 

 

inline static int _wbufq_insert(struct  tcp_connection* c, const char* data,

                                                                                                                unsigned int size)

{

                struct tcp_wbuffer_queue* q;

                struct tcp_wbuffer* wb;

               

                q=&c->wbuf_q;

                if (likely(q->first==0)) /* if empty, use wbufq_add */

                                return _wbufq_add(c, data, size);

:

:

:

:

                if ((q->first==q->last) && ((q->last->b_size-q->last_used)>=size)){

                                /* one block with enough space in it for size bytes */

 

                                memmove(q->first->buf+size, q->first->buf, size);

                                memcpy(q->first->buf, data, size);

                                q->last_used+=size;

                }

 

The above condition shall be true in this case and memmove was moving the pointer which was causing the overflow.

                        memmove(void *dest, const void *src, size_t n);

 

As per the memmove man page, the src shall be copied with size ‘n’ to a temporary buffer and then temporary buffer to dest.

 

dest is q->first->buf+size: which is basically (q->first->buf  + NOTIFY MSG SIZE). so the dst will move my 1282 bytes, so we have remaining space of only 818 bytes.

 

src is q->first->buf: which is basically copied to temp buffer with NOTIFY SIZE(1282bytes).

 

Finally we are moving the buffer from temporary buffer of size 1282 bytes to buffer which we left with 818 bytes, This basically corrupt the memory and on wbufq_run we see the memory corruption

 

2012-09-19T02:06:17+01:00 [crit] sipserver: [27359] : <core> [mem/q_malloc.c:157]: BUG: qm_*: fragm. 0xb00519dc (address 0xb00519f4) end overwritten(0, 0)!

 

I think we don’t need memove, so we can change the code as below

 

       if ((q->first==q->last) && ((q->last->b_size-q->last_used)>=size)){

                /* one block with enough space in it for size bytes */

                //memmove(q->first->buf+size, q->first->buf, size);

                memcpy(q->first->buf+q->last_used, data, size);

                q->last_used+=size;

        }


OR
we need only the else part as it always add the block to the first.

Thanks
Jijo

On Wed, Jul 18, 2012 at 12:42 PM, Daniel-Constantin Mierla <miconda@gmail.com> wrote:
Hello,

sorry, I just read the last messages in the thread and I didn't noticed it is about a crash, but thought it is about an annoying log message.

The backtrace is no longer matching the latest sources of the 3.1 branch, but I expect it is due to a double free issue, so you have to update to latest 3.1.6, as suggested by other users here. There were many fixes from 3.1.0 to 3.1.6.

Cheers,
Daniel


On 7/17/12 6:03 PM, Jijo wrote:
Hi,

This is not happening at shutdown or status check. Its aborting when the system is active.

Thanks
Jijo
On Tue, Jul 17, 2012 at 3:06 AM, Daniel-Constantin Mierla <miconda@gmail.com> wrote:
Hello,

does it keep being or it is one time and that's it?

That is printed at shut down or if pkg_status() or shm_status() is executed from some part of code or in config via cfgutils module functions.

You can get rid of them by setting memdbg and memlog to a value higher than debug global parameter.

Cheers,
Daniel


On 7/16/12 5:28 PM, Jijo wrote:
Thanks.. It is not easy to upgrade as it is happening at customer system.
Is there any change occurred for this issue.I looked at it, but didn't see anything in q_malloc.c/qm_status()


On Mon, Jul 16, 2012 at 11:12 AM, Jon Bonilla <manwe@aholab.ehu.es> wrote:
El Mon, 16 Jul 2012 10:27:42 -0400
Jijo <realjijo@gmail.com> escribió:

> Hi All,
>
> I'm observing a core intermittently at "qm_status (qm=0x786cd000) at
> mem/q_malloc.c:763" for kamailio version 3.1.0
>

I'd say that you're using a very old version. You should update your branch to
3.1.6 or upgrade to a newer branch.


_______________________________________________
sr-dev mailing list
sr-dev@lists.sip-router.org
http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-dev



_______________________________________________
sr-dev mailing list
sr-dev@lists.sip-router.org
http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-dev

-- 
Daniel-Constantin Mierla - http://www.asipto.com
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Kamailio Advanced Training, Seattle, USA, Sep 23-26, 2012 - http://asipto.com/u/katu
Kamailio Practical Workshop, Netherlands, Sep 10-12, 2012 - http://asipto.com/u/kpw


-- 
Daniel-Constantin Mierla - http://www.asipto.com
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Kamailio Advanced Training, Seattle, USA, Sep 23-26, 2012 - http://asipto.com/u/katu
Kamailio Practical Workshop, Netherlands, Sep 10-12, 2012 - http://asipto.com/u/kpw