dialog: fix rtags of forking INVITE

[maximilianfridrich]

Resolves: baresip/baresip#2819

[maximilianfridrich]

A similar PR (#947) resulted in no ACKs being sent in some scenarios - see baresip/baresip#2772.

I tested this patch with the attached SIPP scenario (bob_100rel_hold.txt) and the BareSIP selftest (which checks for pending ACKs).

Still, ideally, this patch could be tested in a few different setups to ensure that ACKs are always sent correctly.

[cspiel1]

@maximilianfridrich please could you add a retest for re-INVITE which also checks that an ACK is sent/received?

[maximilianfridrich]

Apparently there's still issues with ACKs to re-INVITEs, as stated by juha-h in baresip/baresip#2819. I have looked into that in the past and could not find the cause. I will try again.

[maximilianfridrich]

I have finally found the ACK issue!

The issue was that the pl structs in dlg->route are initialized to point to a memory address within dlg->mb. In my patch I re-initialize dlg->mb and therefore the pointers in dlg->route were no longer valid. Now, after dlg->mb is reset in sip_dialog_update, dlg->route is also re-set to point into the new mbuf.

[cspiel1]

This test helps to detect missing ACKs: #1027

[sreimers]

@maximilianfridrich can you rebase against main to check if ACK tests are passed.

[maximilianfridrich]

@sreimers Done. Looks good.

[sreimers]

Thanks!