[Iaxclient-devel] Re: iaxcomm iaxclient registry problem

[S <s...@ka...>]

Hi Steve,

I will send a FAX to digium. First I have to find a FAX machine ;-))
(Yes there are people out which have no fax-machine (this machines are 
old fashioned stuff) If I have send the fax I let you know.

All RISC machines need pointers word align, but some one have a special 
handling to remedy the problem. For example the DEC ALPHA chip get a 
trap and realign the pointer (in the same way I have done). This means 
for every pointer wich is not align the CPU has to do some extra work.
Also I think some new compiler have options to handle this problem. 
(They add some extra code into you program).

------------------------------------------------------------

I found a other problem in iax.c function iax_header_to_event() in
  case IAX_COMMAND_TXREQ:

In the new version there is a new assignment to session->transfer .
If I let the line in I get instantly a crash. Do you know why this line 
is added ?

case IAX_COMMAND_TXREQ:
#if 0
                 session->transfer = *e->ies.apparent_addr;
#endif
                 session->transfer.sin_family = AF_INET;
                 session->transfercallno = e->ies.callno;
                 session->transferring = TRANSFER_BEGIN;

-----------------------------------------------------------------

Do you have any comments about the registry problem and my 
IAX_COMMAND_POKE fix ?

It fix the registry problem but after a time the client is unregisterd 
and registed again. So there must be a other problem to.


Regards

   Stephan

PS : See comments in the mail below.

Steve Kann wrote:
 >
 > Stephan,
 >
 >     Can you (or have you already) faxed in a disclaimer to digium for
 > your libiax2 changes?
 >
 >     If you have, than I'd gladly take this change and send it up to 
Digium.
 >
 >     As far as your alignment patch, I guess my questions are:
 >
 > 1) First, the patches are definitely not needed for all RISC platforms;
 > the code works just fine on PPC, for example.
Maybe the PPC has some special handling for not align pointer.

 >
 > 2) Why haven't others who are using iaxclient on zaurus running into
 > this same problem?
Maybe they use some special compiler (or compiler options)

 >
 > That patch seems like a lot of clutter to add to the code if it isn't
 > really necessary..
You are right it looks out like clutter, but the problem is the design 
mistake in the IAX protocol. I hope in a future version of IAX they take 
  care of word align. On 64Bit CPU's the problem getting bigger....
Most porting problems are align bugs. A good candiate to find a align 
bug is the Sparc CPU (32bit)  and Super Sparc III CPU (64bit), Or the 
Alpha CPU (where the align warning are not disabled).
I have no idea how sensitively the AMD Opteron CPU is.

 >
 > -SteveK
 >
 > S wrote:
 >
 > > Hi Steve,
 > >
 > > I found the registry bug in the libiax2/src/iax2.c,
 > > In function iax_header_to_event(...) is no POKE command handle, so I
 > > add this line and it work (only POKE case section) :
 > >
 > >             case IAX_COMMAND_PING:
 > >                 /* Just immediately reply */
 > >                 e->etype = IAX_EVENT_PING;
 > >                 e->ts = ts;
 > >                 e = schedule_delivery(e, ts);
 > >                 break;
 > >             case IAX_COMMAND_POKE: /* SKA fix registry problem */
 > >                 send_command_immediate( session,
 > >                                         AST_FRAME_IAX,
 > >                                         IAX_COMMAND_PONG,
 > >                                         ts, NULL, 0, fh->iseqno);
 > >                 e = NULL;
 > >                 break;
 > >             case IAX_COMMAND_PONG:
 > >
 > > Please add the POKE case section to libiax2/src/iax2.c line 2119
 > >
 > > iax2 show peers
 > > skaAtHome  172.16.16.160   (D)  255.255.255.255  4569      OK (22 ms)
 > >
 > > I suspect there is still a other bug, because afer a while ( and 4
 > > times refreshing registration ) I get a UNREACHABLE and a while later
 > > I get a REACHABLE
 > >
 > > Regards
 > >
 > >   Stephan
 > >
 > > PS : This is the next stop to iax phone on a ARM RISC Zaurs PDA.
 > > PS : There are still some align patches for iax2-parser.c out...
 > >
 > >
 > >
 > > S wrote:
 > >
 > >> Hi IAX Guru's,
 > >>
 > >> I have the problem to register a iaxclient (iaxcomm or my own Zaurus
 > >> client) to asterisk.
 > >> Here is my observation on asterisk with iax2 debug on where a
 > >> iaxclient try to register :
 > >>
 > >> Rx --> REGREQ
 > >> TX <-- REGAUTH
 > >> Rx --> REGREQ
 > >> TX <-- POKE
 > >> TX <-- REGACK
 > >> RX --> ACK
 > >> RX --> ACK
 > >> TX <--POKE
 > >>
 > >> iax2 show peers
 > >> Name/Username    Host                 Mask             Port  Status
 > >> muggel/muggel    (Unspecified)   (D)  255.255.255.255  0    UNKNOWN
 > >> skaAtHome        172.16.16.160   (D)  255.255.255.255  4569 
UNREACHABLE
 > >> ska/ska          (Unspecified)   (D)  255.255.255.255  0    UNKNOWN
 > >>
 > >>
 > >> And here is the iax2 debug while one asterisk register on a second 
one.
 > >> RX --> REGREQ
 > >> TX <-- REGAUTH
 > >> RX --> REGREQ
 > >> TX <-- REGACK
 > >> RX --> ACK
 > >> TX <-- POKE
 > >> RX --> PONG
 > >> Peer 'XXX' is now REACHABLE!
 > >> TX <-- ACK
 > >>
 > >> The different between both traces is the POKE PONG sequence.
 > >> In the libiax2/src/iax.c file is no code for the POKE request ?
 > >>
 > >> Any comment's ?
 > >> What is going wrong ?
 > >>
 > >> Is there anyone able to register a iaxclient to asterisk ?
 > >> ( iax2 show peers   --> Status OK )
 > >>
 > >> Regards
 > >>
 > >>     Stephan
 > >>
 > >>
 > >

// strings attachment.doc | more