Bogdan,
I tried to reproduce this with some simple instructions, but it didn’t
reproduce. There must be some dependency on other functions or
configurations we are using. It would be too hard to try to figure out
exactly what that is, so I will have to capture the data for you. It
will just take a while for me to revive an older configuration that
allows me to enable the memory debugger.
Ben Newlin
*From: *Bogdan-Andrei Iancu <[email protected]>
*Date: *Tuesday, August 2, 2016 at 3:47 AM
*To: *"Newlin, Ben" <[email protected]>, OpenSIPS users mailling
list <[email protected]>
*Subject: *Re: [OpenSIPS-Users] OpenSIPS fix_route_dialog crashes
Ben,
To make it easier, please send me the instructions on how to reproduce
the crash.
Thanks and Regards,
Bogdan-Andrei Iancu
OpenSIPS Founder and Developer
http://www.opensips-solutions.com
On 01.08.2016 20:17, Newlin, Ben wrote:
Bogdan,
I am not familiar with gdb, so I double check what you’ve
assessed. If there are some other steps with gdb you would like me
to perform, just let me know what to do.
Is there a way to compile the memory debugger without using the
interactive `make menuconfig` command? Our build system is
completely automated, so it impossible for me to do it this way.
Can I pass the options as build parameters or alter the makefile
in some way?
I can provide a SIPp scenario which should reproduce the issue on
any basic script that uses Dialog with topology hiding, if that
would be easier.
Ben Newlin
*From: *Bogdan-Andrei Iancu <[email protected]>
<mailto:[email protected]>
*Date: *Monday, August 1, 2016 at 10:57 AM
*To: *OpenSIPS users mailling list <[email protected]>
<mailto:[email protected]>, "Newlin, Ben"
<[email protected]> <mailto:[email protected]>
*Subject: *Re: [OpenSIPS-Users] OpenSIPS fix_route_dialog crashes
Hi Ben,
According to the BT, the crash is in a pkg_malloc() call:
route = pkg_malloc(size);
Please double check this with gdb info.
If so, this indicate a memory corruption and we have 2 options here:
- you compile with memory debugger (see my previous emails)
- provide step-by-step indications on how to reproduce this crash.
Thanks and Regards,
Bogdan-Andrei Iancu
OpenSIPS Founder and Developer
http://www.opensips-solutions.com
On 29.07.2016 15:54, Newlin, Ben wrote:
This is 1.11.6, running on CentOS 7.
Ben Newlin
*From: *<[email protected]>
<mailto:[email protected]> on behalf of
Bogdan-Andrei Iancu <[email protected]>
<mailto:[email protected]>
*Reply-To: *OpenSIPS users mailling list
<[email protected]> <mailto:[email protected]>
*Date: *Friday, July 29, 2016 at 8:50 AM
*To: *"Newlin, Ben" <[email protected]>
<mailto:[email protected]>, OpenSIPS users mailling list
<[email protected]> <mailto:[email protected]>
*Subject: *Re: [OpenSIPS-Users] OpenSIPS fix_route_dialog crashes
Ben,
What OpenSIPS version is this (the crashing one) ? 1.11 or 2.1 ?
Regards,
Bogdan-Andrei Iancu
OpenSIPS Founder and Developer
http://www.opensips-solutions.com
On 27.07.2016 19:02, Newlin, Ben wrote:
I have identified that these crashes are occurring when
the far end system is not returning the Record-Route
headers in the 200 OK response. The headers are present in
the 180 response, but not the 200 OK. I have reproduced
the scenario using SIPp and captured a SIP trace:
http://pastebin.com/ckKk3EhY <http://pastebin.com/ckKk3EhY>
The crash occurs on receipt of the ACK request and attempt
to match the dialog.
I also captured a BT for this scenario as well, in case
anything specific in the trace made the issue easier to
find: http://pastebin.com/cM3FhPiw
I am working with the other system to try to fix their
behavior.
Ideally the Record-Route headers from previous replies
could be used in this case to allow the call to succeed,
but I don’t know if that is possible.
Thanks,
Ben Newlin
*From: *"Newlin, Ben" <[email protected]>
<mailto:[email protected]>
*Date: *Wednesday, July 27, 2016 at 9:44 AM
*To: *Bogdan-Andrei Iancu <[email protected]>
<mailto:[email protected]>, OpenSIPS users mailling list
<[email protected]> <mailto:[email protected]>
*Subject: *Re: [OpenSIPS-Users] OpenSIPS fix_route_dialog
crashes
Bogdan,
This is a different scenario than the other you responded
to. As I said, we have two types of servers that work
together. One is a load-balancer and runs as a proxy. It
uses double Record-Route because it sends messages between
public and private networks. Then we have our other
servers using TH which receive those requests. We are not
using TH and RR on the same server (although I would like to).
If validate_dialog() and fix_route_dialog() (and possibly
loose_route()) should not be called when using TH, I
believe the documentation should reference that. It states
that match_dialog() must be used with TH, but does not
indicate that the other functions should not be used or
that the functionality won’t work. There is also no
documentation of the incompatibility between RR and TH.
Either way, I ran a test where I removed all calls to
loose_route(), validate_dialog(), and fix_route_dialog()
from my script. The crash still occurred and the BT still
pointed to fix_route_dialog() function. So it must be
getting called from within Dialog module somewhere. That
BT is here: http://pastebin.com/wu2X2Hxh
I collected this BT with loose_route() being called from
my script, but not validate_dialog() or
fix_route_dialog(): http://pastebin.com/6V7yPaHF
This BT was collected with all three functions being
called from my script: http://pastebin.com/fZYYdndn
Ben Newlin
*From: *Bogdan-Andrei Iancu <[email protected]>
<mailto:[email protected]>
*Date: *Wednesday, July 27, 2016 at 3:57 AM
*To: *OpenSIPS users mailling list
<[email protected]>
<mailto:[email protected]>, "Newlin, Ben"
<[email protected]> <mailto:[email protected]>
*Subject: *Re: [OpenSIPS-Users] OpenSIPS fix_route_dialog
crashes
Hi Ben,
First, if you use TH, makes no sense to do Record-Routing
- there are 2 SIP concepts that overlaps. You either act
as an end-point (by doing TH), either as a proxy (doing RR).
If doing TH, makes no sense to use validate + fix as these
functions check and repair the routing information in the
request (like Route and Contact headers). if you do TH,
this routing info is actually hidden and added by
OpenSIPS, so there is nothing to fix and repair.
Nevertheless, this should not crash or corrupt OpenSIPS.
HAve you managed to get a corefile ?
Also if you suspect memory corruption, you can compile-in
the memory debugger - see
http://www.opensips.org/Documentation/TroubleShooting-OutOfMem
.
Regards,
Bogdan-Andrei Iancu
OpenSIPS Founder and Developer
http://www.opensips-solutions.com
On 26.07.2016 23:20, Newlin, Ben wrote:
I have had 3 OpenSIPS server crashes in the last week.
All were due to segmentation faults. I was not able to
capture core dumps; I am configuring that now to catch
the next crash.
My logs leading up to the crash are full of errors
from fix_route_dialog() complaining about invalid URIs
for sequential requests:
Jul 26 19:34:02 [220] ERROR:dialog:fix_route_dialog:
Failed to parse SIP uri
Jul 26 19:34:02 [220] ERROR:core:parse_uri: bad uri,
state 0 parsed: <ip:1> (4) /
<ip:10.18.8.18:5060;ftag=gK0448f137;lr;r2=on>> (44)
Jul 26 19:11:19 [218] ERROR:dialog:fix_route_dialog:
Failed to parse SIP uri
Jul 26 19:11:19 [218] ERROR:core:parse_uri: bad uri,
state 0 parsed: <b0i2> (4) /
<b0i2yjor;transport=udp<sip:10.18.8.17:5060;ftag=7207ce89;lr;r2=on>
(65)
Jul 26 17:43:19 [220] ERROR:dialog:fix_route_dialog:
Failed to parse SIP uri
Jul 26 17:43:19 [220] ERROR:core:parse_uri: bad uri,
state 0 parsed: <ervi> (4) /
<ervice_id6fdbc70f-2438-4726-807c-0d081df4d87> (44)
Many times the “URI” displayed in the error message is
actually internal OpenSIPS variables, as in the last
error above. When they are from the SIP message, I
have verified that the messages themselves are
correctly formatted. This leads me to believe there is
memory corruption occurring.
This all started when I updated my load-balancer
servers to use Record-Routing, specifically the
“double_rr” mechanism for when multiple interfaces
exist. The Record-Routing is occurring on different
servers which have not crashed. Only the servers
receiving the Record-Routed messages are experiencing
the errors.
Here is a piece of the code processing sequential
requests. I am using the topology_hiding()
functionality of the Dialog module. Are
validate_dialog() and fix_route_dialog() still valid
in a topology_hiding scenario?
if (t_check_trans())
setflag(SEQ_REQUEST);
if (has_totag())
{
loose_route();
if (match_dialog())
{
if (!validate_dialog())
fix_route_dialog();
if (is_method("BYE"))
setflag(ACC_FLAG);
setflag(SEQ_REQUEST);
}
else if (!isflagset(SEQ_REQUEST))
{
if (!is_method("ACK")) {
route(rlog, LV_ERROR, "check_sequential", "Sequential
request not matched");
route(reply_error, "481", "Call Does Not Exist");
}
return(EXIT);
}
}
I will attempt to get core dumps of future crashes.
Thanks,
Ben Newlin
_______________________________________________
Users mailing list
[email protected] <mailto:[email protected]>
http://lists.opensips.org/cgi-bin/mailman/listinfo/users
_______________________________________________
Users mailing list
[email protected] <mailto:[email protected]>
http://lists.opensips.org/cgi-bin/mailman/listinfo/users
