[Iaxclient-devel] Aha: The next IAX2 bug (PPC only, I guess)..

[Steve K. <st...@st...>]

Mark,

    Here's a good one, that took a bit to figure out also.  This problem
only occurs with MacOSX, although it just depends on which way the stack
grows..

in iax_send, you have the following code:
        unsigned char buffer[4096];             /* Buffer -- must
preceed fr2 */
        struct iax_frame fr2;
        struct iax_frame *fr;

and later, you do this:
        if (now) {
                fr = &fr2;
        } else
                fr = iax_frame_new(DIRECTION_OUTGRESS, f->datalen);
        if (!fr) {
                IAXERROR "Out of memory\n");
                return -1;
        }

        fprintf(stderr, "now=%d, fr=%x, &fr2=%x\n", now, fr, &fr2);
        fprintf(stderr, "f->datalen=%d\n", f->datalen);

        /* Copy our prospective frame into our immediate or
retransmitted wrapper */
        iax_frame_wrap(fr, f);

        fprintf(stderr, "now=%d, fr=%x, &fr2=%x\n", now, fr, &fr2);


(fprintfs are mine) which if (now=1) and f has data, copies some items
from f into fr2, and then copies data to from f to fr.  This data ends
up past the end of fr2, which you account for, on machines with one
stack direction, with the buffer on the stack.

So, I think it would be better to just do this instead:
        double buffer[1024];             /* Buffer -- does double
guarantee us proper alignment? */
        struct iax_frame *fr;

[...]

       if (now) {
                fr = (struct iax_frame *)&buffer;
        } else
                fr = iax_frame_new(DIRECTION_OUTGRESS, f->datalen);
        if (!fr) {
                IAXERROR "Out of memory\n");
                return -1;
        }

        fprintf(stderr, "now=%d, fr=%x\n", now, fr);
        fprintf(stderr, "f->datalen=%d\n", f->datalen);

        /* Copy our prospective frame into our immediate or
retransmitted wrapper */
        iax_frame_wrap(fr, f);

        fprintf(stderr, "now=%d, fr=%x\n", now, fr);


Does this seem like it will be more portable?

It works for me on MacOSX/PPC, Linux/x86, Win2K/x86.

Attached also please find my current diff between my local libiax2 and
digium's CVS.  It is still mostly portability fixes, plus the two
bugfixes (this one, and the buf[len-1] = 0 issue).

Next on the list: using IAX2, for some reason the DNID isn't being
honored.  This is kinda strange, because using "iax2 debug" on the
console of a * server, it shows the DNID properly, but I still end up in
the "s" extension for the context, instead of the requested extension..


-SteveK




-- 
  Steve Kann - Chief Engineer - 520 8th Ave #2300 NY 10018 -  (212) 533-1775
                HorizonLive.com - collaborate . interact . learn
   "The box said 'Requires Windows 95, NT, or better,' so I installed Linux."