Hi all,

I have a question about CANCEL message processing.

My call sceraio:
When an INVITE request comes I need to rewrite a domain part of the several headers(to, from, contact, SDP ip) according to the outgoing interface.
I can do it before t_relay(), but when destination user has more then one locations and they are reachable through different interfaces all forked INVITEs will have the same domain.
To avoid that problem i have tried to move rewriting part(subst from the textops module) to the branch_route.

Unfortunately that solution didn't helped me, because "To" header of the outgoing CANCEL messages is wrong (unchanged. as it was at the moment when transaction was created by the t_relay() ).
As i understood from the documentation and mailing list kamailio builds CANCEL based on outgoing INVITE.

I did a small research and found that Kamailio really takes outgoing INVITE from each branch (invite_transaction_cell -> uac[b_id].request.buffer) and builds CANCEL (build_local_reparse() from tm/t_msgbuilder.c) based on it.
But kamailio does an exception for the To header which is described below:
    cancel_branch function (from t_cancel.c) calls build_local_reparse() and fills one of the parameters with a pointer to unmodified To header.
    build_local_reparse() uses received To header to construct outgoing CANCEL.

I have changed:
    cancel = build_local_reparse(t, branch, &len, CANCEL, CANCEL_LEN, &t->to, reason);
to:
    cancel = build_local_reparse(t, branch, &len, CANCEL, CANCEL_LEN, NULL, reason);
and it works for me now.

It looks like if E2E_CANCEL_HOP_BY_HOP e2e_cancel() (t_fwd.c) will call e2e_cancel_branch() which works as i want instead of cancel_branch(), but it is just my assumption.

Why does kamailio generate CANCEL requests in such a way ? Did i miss something from the RFC3261 or kamailio documentation ?

Thanks in advance for any help !