Re: Segmentation fault on registered h323 calls ( solved using h323plus )
Jan Willamowius <
jan@...>
2010-08-05 13:16:12 GMT
It would be great if these changes could be included into the standard
Yate release!
Reagrds,
Jan
--
Jan Willamowius, jan@..., http://www.gnugk.org/
Francisco Olarte (M) wrote:
> We are running yate for interconnects, mainly as h233->sip converter.
>
> We repeatedely got signal 11 when working registered to a gatekeeper. It
> looked like some race condition in openh323 code, as backtraces showed both
> in openh323 transport threads and in yate thread2.
>
> This ocurred both when calls were passed to sip and when calls where routed
> to a conference ( one for each call ), or parked ( we did this to rule out
> the possibility of sip being the cause). It looked like some kind of race
> condition as it happened earlier both when call load was raised , when cpu
> was slower and when call duration was lowered.
>
> When we worked without registering the h323 endpoint to a gatekeeper every
> test went fine.
>
> This was tested with yate 2.0 / oh323 1.18 + pw 1.10, yate 2.2 / oh323 1.18
> + pw 1.10 and oh323 1.19 + pw 1.11 and yate 3.0 / oh323 1.19 + pw 1.11
>
> As we need to be registered to the gatekeeper to work properly ( we use a
> home-brew central gatekeeper wich does the routing ), and although I've read
> in the list that h323chan will need to be rewritten to work with h323plus I
> decided to give a try.
>
> After a surprisingly short hack session I've managed to build yate 3 with
> h323plus. Not only this, it's working registered to a gatekeeper, and it
> doesn't dump ( or at least not in our tests, the openh323 1.19 version core
> dumped under a 240 call load in less than a minute, this doesn't. Also it
> dumped under a 120 call 30 to 300 seconds duration call load in minutes, we
> have it running for a couple of hours and it hasn't got a single hiccup (
> after being 15 minutes in 240 10-50 seconds calls in the same instance ). We
> are going to leave it running but it seems pretty stable.
>
> So I'm thinking our problem is in openh323. I'm not knowledgeable enough in
> configure / yate sources to make a full patch, but I'd like to share what I
> did in case someone wants to consider adding h323 plus support to yate.
>
>
> This is what I did.
>
> 1.- Downloaded and installed ptlib 2.8.2 and h323plus-20100525 ( from cvs ).
>
>
> This versions were chosen because we use gnugk too and I have just compiled
> them when building a static binary for gnugk2.3.2 which I was planning to
> use as a h323-h323 gateway to isolate yate from the central gatekeeper.
>
> I unpacked them in ~/src/ptlib-2.8.2 and ~/src/h323plus-20100525 and
> configured them like this:
>
> for pwlib:
>
> ~/src/ptlib-2.8.2 $ ./configure --prefix=$HOME
>
> for h323plus:
>
> ~/src/h323plus-20100525 $
> ./configure --prefix=$HOME \
> --disable-video \
> --disable-h460 \
> --disable-h46018 \
> --disable-h46023 \
> --disable-h460p \
> --disable-t38 \
> --disable-t120 \
> --disable-file \
> --enable-gnugk
>
> ( h323plus directed me to run aclocal <at> plugins dir, I did it and it worked
> fine )
>
> This configuration was choosen for the GNUGK, I wanted to use the same libs
> in yate.
>
> Then I ran make opt in both of them ( gone well ) and exported variables to
> point at them:
>
> export PTLIBDIR=~/src/ptlib-2.8.2
> export PWLIBDIR=$PTLIBDIR
> export OPENH323DIR=~/src/h323plus-20100525
> export H323PLUSDIR=$OPENH323DIR
>
>
> 2.- Hacked a little to make yate configure script find everything. What I
> did was:
>
> 2.a.- Hack lib directory in ptlib, which has a different name:
>
> ~/src/ptlib-2.8.2 $ mkdir lib
> ~/src/ptlib-2.8.2 $ cd lib
> ~/src/ptlib-2.8.2/lib $ cp ../lib_linux_x86_64/libpt.so libpt_r.so
>
> 2.b.- Hack lib directory in h323plus, same reason:
>
> ~/src/h323plus-20100525/lib $ ln -s libh323_linux_x86_64_.so
> libh323_linux_x86_64_r.so
>
> 2.c - run yate3 configure script. I used the same previously used with
> openh323 1.19 ( from which I copied the dir structure in the previous links
> ).
>
> ./configure \
> --prefix=$HOME \
> --enable-inline \
> --disable-dahdi \
> --disable-zaptel \
> --disable-wpcard \
> --disable-tdmcard \
> --disable-wanpipe \
> --disable-ilbc \
> --enable-rtti \
> --without-libpq \
> --without-mysql \
> --without-wphwec \
> --without-libgsm \
> --without-libspeex \
> --without-amrnb \
> --without-spandsp \
> --with-pwlib=$PWLIBDIR \
> --with-openh323=$OPENH323DIR \
> --with-openssl \
> --without-libqt4
>
> This ran ok, and properly found ptlib/h323plus:
> ....
> checking for Pwlib in /home/gnugk/src/ptlib-2.8.2... sources RTTI: none
> checking for OpenH323 in /home/gnugk/src/h323plus-20100525... sources 1.22.0
> ....
>
>
> 3.- The I started the make session, which I'll describe:
>
> 3.-a first make gave a lot of error relating to unknwon BOOL type in
> h323chan. Looking at the docs I saw BOOL type was now called PBoolean, so I
> added
>
> typedef PBoolean BOOL;
>
> at the top, between
>
> #include <string.h>
>
> and
>
> using namespace TelEngine;
>
>
> 3.b Second make, this time I got a single error:
>
> g++ -Wall -I.. -I.. -O2 -fno-check-new -frtti -fno-exceptions -fPIC
> -DHAVE_GCC_FORMAT_CHECK -finline -Winline -export-dynamic -shared
> -Wl,--unresolved-symbols=ignore-in-shared-libs
> -Wl,--retain-symbols-file,/dev/null -o h323chan.yate -DPHAS_TEMPLATES
> -D_REENTRANT -DP_HAS_SEMAPHORES -I/home/gnugk/src/ptlib-2.8.2/include
> -I/home/gnugk/src/ptlib-2.8.2/include/ptlib/unix/ptlib
> -I/home/gnugk/src/ptlib-2.8.2/include/ptlib/unix
> -I/home/gnugk/src/ptlib-2.8.2/include/ptlib
> -I/home/gnugk/src/h323plus-20100525/include h323chan.cpp
> -L/home/gnugk/src/h323plus-20100525/lib -lh323_linux_x86_64_r
> -L/home/gnugk/src/ptlib-2.8.2/lib -lpt_r -L.. -lyate
> h323chan.cpp: In function 'void<unnamed>::ListRegisteredCaps(int)':
> h323chan.cpp:641: error: conversion from
> '__gnu_cxx::__normal_iterator<std::basic_string<char,
> std::char_traits<char>, std::allocator<char> >*,
> std::vector<std::basic_string<char, std::char_traits<char>,
> std::allocator<char> >, std::allocator<std::basic_string<char,
> std::char_traits<char>, std::allocator<char> > > > >' to non-scalar type
> '__gnu_cxx::__normal_iterator<const PString*, std::vector<PString,
> std::allocator<PString> > >' requested
> h323chan.cpp:641: error: no match for 'operator!=' in 'find !=
> list.std::vector<_Tp, _Alloc>::end [with _Tp = std::basic_string<char,
> std::char_traits<char>, std::allocator<char> >, _Alloc =
> std::allocator<std::basic_string<char, std::char_traits<char>,
> std::allocator<char> > >]()'
> /home/gnugk/src/ptlib-2.8.2/include/ptlib/pluginmgr.h: At global scope:
> /home/gnugk/src/ptlib-2.8.2/include/ptlib/pluginmgr.h:167: warning:
> 'PFactoryLoader::PluginLoaderStartup_loader' defined but not used
> make[1]: *** [h323chan.yate] Error 1
>
> The guilty party was the ListRegisteredCaps function:
>
>
> static void ListRegisteredCaps(int level)
> {
> PFactory<H323Capability>::KeyList_T list =
> PFactory<H323Capability>::GetKeyList();
> for (std::vector<PString>::const_iterator find = list.begin(); find !=
> list.end(); ++find)
> Debug(level,"Registed capability: '%s'",(const char*)*find);
> }
>
> Which has two problems, first the for should be written using list declared
> type, not assuming it:
>
> for (PFactory<H323Capability>::KeyList_T::const_iterator find =
> list.begin(); find != list.end(); ++find)
>
> And second, once fixed this basic_string is not convertible to const char *,
> so :
>
> Debug(level,"Registed capability: '%s'",find->c_str());
>
> Which left the function as:
>
> static void ListRegisteredCaps(int level)
> {
> PFactory<H323Capability>::KeyList_T list =
> PFactory<H323Capability>::GetKeyList();
> #if 0
> for (std::vector<PString>::const_iterator find = list.begin(); find !=
> list.end(); ++find)
> Debug(level,"Registed capability: '%s'",(const char*)*find);
> #else
> for (PFactory<H323Capability>::KeyList_T::const_iterator find =
> list.begin(); find != list.end(); ++find)
> Debug(level,"Registed capability: '%s'",find->c_str());
> #endif
> }
>
> Once this was fixed compilation went ok, but then I got link errors.
>
> 3.c.- Link error due to 64 bit fixing.
>
> One of the link errors was an undefined reference to
> H323GatekeeperServer::MonitorMain(PThread, int).
>
> Dumping the libh323_.a I noticed it had a
> H323GatekeeperServer::MonitorMain(PThread, long) inside.
>
> The declaration of the method in h323plus gkserver.cxx was
> H323GatekeeperServer::MonitorMain(PThread, INT).
>
> And I found this definition inside ptlibs header files:
>
> /home/gnugk/src/ptlib-2.8.2/include/ptlib/unix/ptlib/contain.h
> 81-#ifdef P_64BIT
> 82:typedef long INT;
> 83:typedef unsigned long UINT;
> 84-#else
> 85:typedef int INT;
> 86:typedef unsigned int UINT;
> 87-#endif
>
> So I went into the modules/Makefile and changed:
>
>
> h323chan.yate: LOCALFLAGS = -DPHAS_TEMPLATES -D_REENTRANT -DP_HAS_SEMAPHORES
> -I/home/gnugk/src/ptlib-2.8.2/include
> -I/home/gnugk/src/ptlib-2.8.2/include/ptlib/unix/ptlib
> -I/home/gnugk/src/ptlib-2.8.2/include/ptlib/unix
> -I/home/gnugk/src/ptlib-2.8.2/include/ptlib
> -I/home/gnugk/src/h323plus-20100525/include
>
> to
>
> h323chan.yate: LOCALFLAGS = -DP_64BIT -DPHAS_TEMPLATES -D_REENTRANT
> -DP_HAS_SEMAPHORES -I/home/gnugk/src/ptlib-2.8.2/include
> -I/home/gnugk/src/ptlib-2.8.2/include/ptlib/unix/ptlib
> -I/home/gnugk/src/ptlib-2.8.2/include/ptlib/unix
> -I/home/gnugk/src/ptlib-2.8.2/include/ptlib
> -I/home/gnugk/src/h323plus-20100525/include
>
>
> This led to make ok, make install ok. Then I just did
>
> export LD_LIBRARY_PATH=~/lib:$OPENH323DIR/lib:$PWLIBDIR/lib_linux_x86_64_o
>
> and ~/bin/yate ran ok.
>
> After that we copied our test configs and everything is running well,
>
>
> If someone wants to consider adding support for this configuration to yate I
> can provide more info.
>
>
>
> ------------------
>
> Francisco Olarte Sanz.
>
>
>
> ---DIFF for h323chan.cpp------------
>
> gnugk <at> gkaps ~ $ diff -u ~yate/src/yate/modules/h323chan.cpp
> yate3-ptx/modules/h323chan.cpp
> --- /home/yate/src/yate/modules/h323chan.cpp 2010-03-12
> 16:44:09.000000000 +0100
> +++ yate3-ptx/modules/h323chan.cpp 2010-08-04 17:14:40.961827685 +0200
> <at> <at> -82,6 +82,7 <at> <at>
> #include <yateversn.h>
>
> #include <string.h>
> +typedef PBoolean BOOL;
>
> using namespace TelEngine;
> namespace { // anonymous
> <at> <at> -637,8 +638,13 <at> <at>
> static void ListRegisteredCaps(int level)
> {
> PFactory<H323Capability>::KeyList_T list =
> PFactory<H323Capability>::GetKeyList();
> +#if 0
> for (std::vector<PString>::const_iterator find = list.begin(); find !=
> list.end(); ++find)
> Debug(level,"Registed capability: '%s'",(const char*)*find);
> +#else
> + for (PFactory<H323Capability>::KeyList_T::const_iterator find =
> list.begin(); find != list.end(); ++find)
> + Debug(level,"Registed capability: '%s'",find->c_str());
> +#endif
> }
> #else
> // This class is used just to find out if a capability is registered
RSS Feed