[Iaxclient-devel] Possible Solution To IAX2 Client Inbound Call Bug

[Steven S. <ss...@so...>]

Ok, I can't do the dance of joy yet.  Somebody else here has to try this
and agree that the solution works across the board.  BUT, so far I can't
get it to fail (which is a first for me).  So.... a minor hooray is in
order.

The fix:

In iaxclient_lib.c find the function "iaxc_handle_regreply()"

Towards the bottom (line 847 on mine, but it's pretty hacked up) you
will see the following:

    // XXX I think the session is no longer valid.. at least, that's
    // what miniphone does, and re-using the session doesn't seem to
    // work!
    cur->session = NULL;

To make the fix:

    // XXX I think the session is no longer valid.. at least, that's
    // what miniphone does, and re-using the session doesn't seem to
    // work!
    iax_destroy(cur->session);
    cur->session = NULL;

That's it.  Just explicitly destroy the session with the
iax_destroy(cur->session); call.  That seems to do it.  The root of the
error seemed to be this:

When the system went to re-register, it created a session that was then
included in the pool of available sessions on the heap.  Steve's
original version simply cut the session loose from the registration - it
never explicitly killed it.  So the session hung around with all of its
state information intact.  Somehow, and I am still trying to fathom the
specific reason, the session was then re-used for a subsequent incoming
call.  Because its state information, including the InSeq and OutSeq
values were still intact from the registration interchange, the
iax_header_to_event function saw the InSeq number as not matching the
arriving event message and chucked the whole thing.  Sneaky.

Please test this hypothesis and see if it holds.  So far I haven't
missed a call, and I don't see any leakage of memory, handles, threads,
etc.  Those of you working on the Linux version of this especially give
it a shot -- I don't have any working Linux client code.

Sorry I don't have a formal diff to present.  Hopefully Steve K can make
the change on the master copy in CVS.

Good luck.

Steve