[dialog] added support for delayed dialog deletion

Once terminated, instead of deleting / destroying the in-memory dialog right away, by setting the 'delete_delay' parameter, the dialog may be kept for a while in memory, in a read-only state (no action, no changes, nothing).
This delaying may be used to help with the routing of late in-dialog request that may be received after the dialog terminated (like late BYE's due to retransmissions, cross BYE requests, auth'ed BYE request, slow ACK on re-INVITEs, etc).

This work was sponsored by 46Labs, a true OpenSIPS supporter!

modules/dialog/dialog.c

@@ -73,6 +73,7 @@ int dlg_bulk_del_no = 1; /* delete one by one */
 int seq_match_mode = SEQ_MATCH_FALLBACK;
 int options_ping_interval = 30;      /* seconds */
 int reinvite_ping_interval = 300;    /* seconds */
+int dlg_del_delay = 0;               /* in seconds, default off */
 str dlg_extra_hdrs = {NULL,0};
 int race_condition_timeout = 5; /* seconds until call termination is triggered,
 					after 200OK -> CANCEL race detection */
@@ -286,6 +287,7 @@ static param_export_t mod_params[]={
 	{ "default_timeout",       INT_PARAM, &default_timeout          },
 	{ "options_ping_interval", INT_PARAM, &options_ping_interval    },
 	{ "reinvite_ping_interval",INT_PARAM, &reinvite_ping_interval   },
+	{ "delete_delay",          INT_PARAM, &dlg_del_delay            },
 	{ "dlg_extra_hdrs",        STR_PARAM, &dlg_extra_hdrs.s         },
 	{ "dlg_match_mode",        INT_PARAM, &seq_match_mode           },
 	{ "db_url",                STR_PARAM, &db_url.s                 },
@@ -858,6 +860,11 @@ static int mod_init(void)
 		return -1;
 	}
 
+	if (init_dlg_del_timer(dlg_ondelete )!=0) {
+		LM_ERR("cannot init delete timer list\n");
+		return -1;
+	}
+
 	if (init_dlg_ping_timer()!=0) {
 		LM_ERR("cannot init ping timer\n");
 		return -1;
@@ -960,6 +967,7 @@ static void mod_destroy(void)
 	dlg_db_mode = DB_MODE_NONE;
 	destroy_dlg_table();
 	destroy_dlg_timer();
+	destroy_dlg_del_timer();
 	destroy_ping_timer();
 	destroy_dlg_callbacks( DLGCB_CREATED|DLGCB_LOADED );
 	destroy_dlg_handlers();

modules/dialog/dlg_handlers.c

@@ -1889,7 +1889,7 @@ void dlg_onroute(struct sip_msg* req, str *route_params, void *param)
 				return;
 			}
 
-			dlg = lookup_dlg( h_entry, h_id);
+			dlg = lookup_dlg( h_entry, h_id, 0);
 			if (dlg==0) {
 				LM_DBG("unable to find dialog for %.*s "
 					"with route param '%.*s'\n",
@@ -1981,6 +1981,46 @@ void dlg_onroute(struct sip_msg* req, str *route_params, void *param)
 		dlg_set_tm_dialog_ctx(dlg, t);
 
 	/* run actions for the transition */
+	if (new_state==DLG_STATE_DELETED && old_state==DLG_STATE_DELETED) {
+		/* a request after dialog termination */
+
+		/* within dialog request */
+		run_dlg_callbacks(DLGCB_REQ_WITHIN, dlg, req, dir, NULL, 0, 1);
+
+		/* update the cseq */
+		dlg_lock (d_table,d_entry);
+		if (dlg->legs[dst_leg].last_gen_cseq) {
+
+			update_val = ++(dlg->legs[dst_leg].last_gen_cseq);
+			dlg_unlock (d_table,d_entry);
+
+			if (update_msg_cseq(req,0,update_val) != 0)
+				LM_ERR("failed to update BYE msg cseq\n");
+
+			msg_cseq = &((struct cseq_body *)req->cseq->parsed)->number;
+
+			final_cseq = shm_malloc(msg_cseq->len + 1);
+			if (final_cseq == 0) {
+				LM_ERR("no more shm mem\n");
+				goto after_unlock5;
+			}
+
+			memcpy(final_cseq,msg_cseq->s,msg_cseq->len);
+			final_cseq[msg_cseq->len] = 0;
+
+			if ( d_tmb.register_tmcb( req, 0, TMCB_RESPONSE_FWDED,
+			fix_final_cseq,
+			(void*)final_cseq, free_final_cseq)<0 ) {
+				LM_ERR("failed to register TMCB (2)\n");
+			}
+		}
+		else
+			dlg_unlock (d_table,d_entry);
+
+		return;
+	}
+
+
 	if (event==DLG_EVENT_REQBYE && new_state==DLG_STATE_DELETED &&
 	old_state!=DLG_STATE_DELETED) {
 
@@ -2470,6 +2510,42 @@ void dlg_ontimeout(struct dlg_tl *tl)
 	return;
 }
 
+#define get_dlg_del_tl_payload(_tl_)  ((struct dlg_cell*)((char *)(_tl_)- \
+		(unsigned long)(&((struct dlg_cell*)0)->del_tl)))
+
+/* This timer handler is triggered when the done with "waiting" after the
+ * dialog got into the TERMINATED state (to be deleted). Take care as 
+ * the dialog may or not be ref counted here (and still in hash)
+ */
+void dlg_ondelete(struct dlg_tl *tl)
+{
+	/* just removed from del timer, it already has the marker, so it will 
+	 * not be subject to another adding */
+	/* del timer - unlocked */
+	/* dlg cell  - unlocked */
+	struct dlg_entry *d_entry;
+	struct dlg_cell *dlg;
+
+	dlg = get_dlg_del_tl_payload(tl);
+
+	d_entry = &(d_table->entries[dlg->h_entry]);
+
+	dlg_lock( d_table, d_entry);
+
+	LM_DBG("delete handler for dlg %p, ref=%d\n",dlg, dlg->ref);
+
+	if (dlg->ref<=0) {
+		/* dlg not ref'ed, so safe to destroy.*/
+		LM_DBG("destroying dlg %p\n",dlg);
+		unlink_unsafe_dlg( d_entry, dlg);
+		destroy_dlg(dlg);
+	} /* if the dialog is still ref'ed, it will be destroyed when 
+	   * the ref gets to 0 (and del timer waiting is already done) */
+
+	dlg_unlock( d_table, d_entry);
+}
+
+
 #define ROUTE_STR "Route: "
 #define CRLF "\r\n"
 #define ROUTE_LEN (sizeof(ROUTE_STR) - 1)
@@ -2847,7 +2923,7 @@ int terminate_dlg(const str *callid, unsigned int h_entry, unsigned int h_id,
 	if (callid)
 		dlg = get_dlg_by_callid(callid, 1);
 	else
-		dlg = lookup_dlg(h_entry, h_id);
+		dlg = lookup_dlg(h_entry, h_id, 1);
 
 	if(!dlg)
 		return 0;

modules/dialog/dlg_handlers.h

@@ -122,6 +122,8 @@ void dlg_onroute(struct sip_msg* req, str *rr_param, void *param);
 
 void dlg_ontimeout( struct dlg_tl *tl);
 
+void dlg_ondelete(struct dlg_tl *tl);
+
 str *dlg_get_did(struct dlg_cell *dlg);
 
 int dlg_validate_dialog( struct sip_msg* req, struct dlg_cell *dlg);

modules/dialog/dlg_hash.c

@@ -757,7 +757,8 @@ int dlg_update_routing(struct dlg_cell *dlg, unsigned int leg,
 
 
 
-struct dlg_cell* lookup_dlg( unsigned int h_entry, unsigned int h_id)
+struct dlg_cell* lookup_dlg( unsigned int h_entry, unsigned int h_id,
+															int active_only )
 {
 	struct dlg_cell *dlg;
 	struct dlg_entry *d_entry;
@@ -771,7 +772,7 @@ struct dlg_cell* lookup_dlg( unsigned int h_entry, unsigned int h_id)
 
 	for( dlg=d_entry->first ; dlg ; dlg=dlg->next ) {
 		if (dlg->h_id == h_id) {
-			if (dlg->state==DLG_STATE_DELETED) {
+			if (active_only && dlg->state==DLG_STATE_DELETED) {
 				dlg_unlock( d_table, d_entry);
 				goto not_found;
 			}
@@ -948,7 +949,7 @@ struct dlg_cell *get_dlg_by_dialog_id(str *dialog_id)
 		/* we might have a dialog did */
 		LM_DBG("ID: %*s (h_entry %u h_id %u)\n",
 				dialog_id->len, dialog_id->s, h_entry, h_id);
-		dlg = lookup_dlg(h_entry, h_id);
+		dlg = lookup_dlg(h_entry, h_id, 1);
 	}
 	if (!dlg) {
 		/* the ID is not a number, so let's consider

modules/dialog/dlg_hash.h

@@ -152,6 +152,7 @@ struct dlg_cell
 	unsigned int         initial_t_label;
 	unsigned int         replicated; /* indicates if the dialog is replicated */
 	struct dlg_tl        tl;
+	struct dlg_tl        del_tl;
 	struct dlg_ping_list *pl;
 	struct dlg_ping_list *reinvite_pl;
 	str                  terminate_reason;
@@ -327,8 +328,16 @@ void destroy_dlg(struct dlg_cell *dlg);
 			abort(); \
 		}\
 		if ((_dlg)->ref<=0) { \
-			unlink_unsafe_dlg( _d_entry, _dlg);\
-			destroy_dlg(_dlg);\
+			/* dlg good to be destried, but be sure it went first
+			 * via the delete timer */ \
+			if (dlg_del_delay==0 || \
+			insert_attempt_dlg_del_timer(&_dlg->del_tl, dlg_del_delay)==-2) {\
+				/* no delay on del or not in del timer anymore -> destroy */ \
+				LM_DBG("Destroying dialog %p\n",_dlg); \
+				unlink_unsafe_dlg( _d_entry, _dlg);\
+				destroy_dlg(_dlg);\
+			} /* else, either still in timer (-1), either
+			   * inserted now in del timer (0) -> nothing to do*/ \
 		}\
 	}while(0)
 
@@ -396,11 +405,13 @@ int dlg_update_leg_info(int leg_idx, struct dlg_cell *dlg, str* tag, str *rr,
 		str *mangled_from,str *mangled_to,str *in_sdp, str *out_sdp);
 
 int dlg_update_cseq(struct dlg_cell *dlg, unsigned int leg, str *cseq,
-						int field_type);
+		int field_type);
 
-int dlg_update_routing(struct dlg_cell *dlg, unsigned int leg,str *rr, str *contact);
+int dlg_update_routing(struct dlg_cell *dlg, unsigned int leg,str *rr,
+		str *contact);
 
-struct dlg_cell* lookup_dlg( unsigned int h_entry, unsigned int h_id);
+struct dlg_cell* lookup_dlg( unsigned int h_entry, unsigned int h_id,
+		int active_only);
 
 struct dlg_cell* get_dlg(str *callid, str *ftag, str *ttag,
 		unsigned int *dir, unsigned int *dst_leg);

modules/dialog/dlg_replication.c

@@ -611,7 +611,7 @@ int dlg_replicated_delete(bin_packet_t *packet)
 		DLG_BIN_POP(int, packet, h_id, malformed);
 
 		h_entry = dlg_hash(&call_id);
-		dlg = lookup_dlg(h_entry, h_id);
+		dlg = lookup_dlg(h_entry, h_id, 1);
 	} else {
 		dlg = get_dlg(&call_id, &from_tag, &to_tag, &dir, &dst_leg);
 	}