Debian Codegraph @

src/modules/sca/NOTES

raw
1 Hook all transactions?
2 register callbacks for TMCB_REQUEST_IN_N & TMCB_RESPONSE_IN_N
3 if Call-Info header not found, pass
4 might be able to dispense with sca_handle_subscribe().
5 performance penalty: cost of check for Call-Info on each transaction.
6
7 OR
8
9 check for Call-Info header in script, call "sca_call_info_update"
10 must be done in onreply route, too
11 more prone to bugs because of missed cases
12 easy to misconfigure, potentially leading to low adoption
13
14
15 To consider:
16 * What to do when a subscriber reSUBSCRIBES and the dialog is different
17 from the current saved subscription? Avoiding shm_realloc seems like a
18 good idea, so perhaps just delete the current one and save the new.
19
20 Update: separate shm_alloc for dialog.id in sca_subscription_create,
21 then shm_free & shm_alloc if differing dialog found.
22
23
24 How to remove Call-Info header before relaying?
25 Use del_lump from ../../data_lump.h
26 Cf. textops module's remove_hf_f()
27
28 How to detect hold/pickup in SDP?
29 Use parse_sdp/free_sdp from ../../parser/sdp/sdp.h.
30
31 sdp_stream_cell_t *sdp_stream;
32
33 /* puts sdp_info_t pointer in msg->body */
34 rc = parse_sdp( msg );
35 if ( rc < 0 ) {
36 error;
37 }
38 /* msg->body must now be free_sdp'd */
39
40 /* for now we'll just assume a single stream per call */
41 sdp_stream = get_sdp_stream( msg, 0, 0 );
42
43 if ( sdp->is_on_hold ) {
44 ....
45 }
46
47
48
49 Call-Info meaning in context
50 If present, refers to appearance associated with the Contact URI.
51
52 e.g., a Call-Info header present in an INVITE indicates the caller
53 has appearance-index N and is using the line to place a call.
54
55 e.g., a Call-Info header present in a response indicates the callee
56 has appearance-index N
57
58
59
60 A reINVITE from another line in the SCA group has the following characteristics:
61 * RURI is the SCA line's AoR, not the callee.
62 * From and To URIs are also SCA line's AoR.
63 * SCA provides very limited way to pass callee info: only
64 in NOTIFY's appearance-uri, which is for display purposes.
65 * From-tag and Call-ID are new. There is no to-tag.
66 * A Call-Info header is present with the appearance-index the line is
67 attempting to retrieve from hold.
68
69
70 Revisiting possibility of using TMCB_ tm module hooks.
71 TMCB_E2EACK_IN is called when an ACK matching a 200OK-answered INVITE
72 is received. Could use this to send out call-info NOTIFYs when an SCA
73 line answers a call. Bug right now is that the NOTIFYs are sent out
74 before the proxy has CANCELed the INVITEs to the other members of the
75 group. Seems the handsets are dumping the line state when they get the
76 CANCELs, so the appearance goes away. On next reSUBSCRIBE/NOTIFY, the
77 handsets get the correct state again.
78
79 Using TCMB_REQ_IN is a viable option. It would allow for eliminating
80 the use of exported functions to the script. Second rev? Probably
81 more efficient than using exported functions.
82
83 Hooking TMCB_E2EACK_IN does solve the above problem. Current problem
84 is getting current URIs on both call legs, since callee in hold/pickup
85 scenario (SCA line A1 holds call to B, SCA line A2 picks up) thinks
86 To & From URI are both the SCA line's AoR. Tried sending UPDATEs from
87 the TMCB_E2EACK_IN callback, but because we're sending them before the
88 proxy relays the ACK to the callee, the callee appears to revert to
89 using the URIs in the ACK for all future packets. The callER handles
90 the UPDATE correctly, and uses the correct URIs.
91
92 Possible workarounds for the above, in order of increasing quality:
93 * Append UPDATE info to shared linked list, register short timer,
94 send all pending UPDATEs when timer alarms, clear queue.
95
96 Pros: easy to write, discrete from other code
97 Cons: This is effectively a sleep(1) workaround. Clumsy, prone
98 to error: still might cause UPDATE packet to be sent prior
99 to ACK delivery to callee, if UPDATE is appended to queue
100 just as timer fires.
101
102 * Register a TMCB_DESTROY callback in the TMCB_E2EACK_IN callback
103 to send the UPDATEs when the ACK transaction is destroyed.
104
105 Pros: fits current model fairly well, can glean all required
106 information from struct cell in callback, no need to
107 involve script.
108 Cons: also similar to a sleep(1) workaround, though less prone
109 to races. still attached to a timer (tm destroy tick),
110 meaning users may still trigger problem with rapid hold/
111 pickups.
112
113 * Register a pv in the module, set it to a known value in the
114 TMCB_E2EACK_IN callback, have the script check the value of the
115 pv, and invoke an exported sca_update function if the pv is set
116 to the known value.
117
118 Pros: fits sip-router script-based logic model well, does not
119 rely on a timer (eliminates races).
120 Cons: pv has a lot of overhead for what amounts to a flag, means
121 adding an exported function to the script.
122
123 * Register a flag, set it in TMCB_E2EACK_IN callback if UPDATEs
124 need to go to call legs, invoke sca_update from script as above.
125
126 Pros: it's just a flag.
127 Cons: setup may be complicated, given sip-router's script flag
128 handling, means adding exported function for script.
129 (see fix_flags in flags.c, e.g. registrar's
130 fix_nated_contact())
131
132
133 Dependency on To-URI is problematic for hold/pickup
134 An SCA pickup of a held call on another handset means RURI, From-URI
135 and To-URI are identical. The module currently expects those to
136 be unique. It will be possible to reconcile these, but it suggests
137 that using the sip_uri_t struct and the parse_uri functions may be
138 helpful.
139
140 No handler for t_reply (locally sent transaction-stateful replies).
141 Probably need to register callback for TMCB_LOCAL_RESPONSE_IN.
142
143 Deletion of call-info subscription when line-seize NOTIFY fails seems
144 wrongheaded.