pespin submitted this change.
cp_peer: Implement local originated heartbeat procedure
Submit PFCP Hearbeat Request with configured interval,
and timeout after no PFCP Hearbeat Response received based on
configuration.
Upon timeout, the cp_peer assoc_cb() is called to notify the user that
the peer is considered not associated anymore. It will then try to
keep associating again automatically.
Default value for OSMO_PFCP_TIMER_HEARTBEAT_RESP is increased to 35s to
allow for a single Heartbeat Request packet loss.
Related: SYS#7294
Change-Id: I7efc0961e1ea39dd7f4cc6ba96be4cf5ce9a2d6c
---
M include/osmocom/pfcp/pfcp_cp_peer_private.h
M src/libosmo-pfcp/pfcp_cp_peer.c
M src/libosmo-pfcp/pfcp_endpoint.c
3 files changed, 104 insertions(+), 9 deletions(-)
diff --git a/include/osmocom/pfcp/pfcp_cp_peer_private.h b/include/osmocom/pfcp/pfcp_cp_peer_private.h
index b8b0af7..503912d 100644
--- a/include/osmocom/pfcp/pfcp_cp_peer_private.h
+++ b/include/osmocom/pfcp/pfcp_cp_peer_private.h
@@ -46,6 +46,9 @@
/* Application private data for assoc_cb, in case ep->priv does not suffice. */
void *priv;
+ struct osmo_timer_list heartbeat_tx_timer;
+ struct osmo_timer_list heartbeat_rx_timer;
+
struct osmo_use_count use_count;
struct osmo_use_count_entry use_count_buf[128];
};
diff --git a/src/libosmo-pfcp/pfcp_cp_peer.c b/src/libosmo-pfcp/pfcp_cp_peer.c
index 0c29df7..2e4fd1b 100644
--- a/src/libosmo-pfcp/pfcp_cp_peer.c
+++ b/src/libosmo-pfcp/pfcp_cp_peer.c
@@ -45,12 +45,14 @@
enum pfcp_cp_peer_fsm_event {
PFCP_CP_PEER_EV_RX_ASSOC_SETUP_RESP,
PFCP_CP_PEER_EV_RX_ASSOC_UPDATE_REQ,
+ PFCP_CP_PEER_EV_HEARTBEAT_TIMEOUT,
PFCP_CP_PEER_EV_USE_COUNT_ZERO,
};
static const struct value_string pfcp_cp_peer_fsm_event_names[] = {
OSMO_VALUE_STRING(PFCP_CP_PEER_EV_RX_ASSOC_SETUP_RESP),
OSMO_VALUE_STRING(PFCP_CP_PEER_EV_RX_ASSOC_UPDATE_REQ),
+ OSMO_VALUE_STRING(PFCP_CP_PEER_EV_HEARTBEAT_TIMEOUT),
OSMO_VALUE_STRING(PFCP_CP_PEER_EV_USE_COUNT_ZERO),
{}
};
@@ -91,6 +93,62 @@
return 0;
}
+static int on_pfcp_heartbeat_resp(struct osmo_pfcp_msg *req, struct osmo_pfcp_msg *rx_resp, const char *errmsg)
+{
+ struct osmo_fsm_inst *fi = req->ctx.peer_fi;
+ struct osmo_pfcp_cp_peer *cp_peer = fi->priv;
+
+ if (!rx_resp) {
+ OSMO_LOG_PFCP_MSG(req, LOGL_NOTICE, "Error: PFCP Heartbeat Response: %s\n",
+ errmsg ? : "no response received");
+ return 0;
+ }
+
+ OSMO_LOG_PFCP_MSG(rx_resp, LOGL_INFO, "Rx PFCP Heartbeat Response\n");
+
+ if (fi->state != PFCP_CP_PEER_ST_ASSOCIATED)
+ return 0;
+
+ unsigned int tval_rx_heartbeat_s =
+ osmo_tdef_get(cp_peer->ep->cfg.tdefs, OSMO_PFCP_TIMER_HEARTBEAT_RESP, OSMO_TDEF_S, -1);
+ osmo_timer_schedule(&cp_peer->heartbeat_rx_timer, tval_rx_heartbeat_s, 0);
+
+ return 0;
+}
+
+static void pfcp_cp_peer_tx_heartbeat_req(struct osmo_pfcp_cp_peer *cp_peer)
+{
+ struct osmo_pfcp_msg *m;
+
+ m = osmo_pfcp_cp_peer_new_req(cp_peer, OSMO_PFCP_MSGT_HEARTBEAT_REQ);
+ m->ies.heartbeat_req.recovery_time_stamp = osmo_pfcp_endpoint_get_recovery_timestamp(cp_peer->ep);
+
+ m->ctx.resp_cb = on_pfcp_heartbeat_resp;
+
+ LOG_CP_PEER(cp_peer, LOGL_INFO, "Tx PFCP Heartbeat Request\n");
+
+ if (osmo_pfcp_endpoint_tx(cp_peer->ep, m))
+ LOG_CP_PEER(cp_peer, LOGL_ERROR, "Failed to transmit PFCP Heartbeat Request to peer\n");
+}
+
+static void pfcp_cp_peer_heartbeat_tx_timer_cb(void *data)
+{
+ struct osmo_pfcp_cp_peer *cp_peer = data;
+ unsigned int tval_tx_heartbeat_s =
+ osmo_tdef_get(cp_peer->ep->cfg.tdefs, OSMO_PFCP_TIMER_HEARTBEAT_REQ, OSMO_TDEF_S, -1);
+
+ pfcp_cp_peer_tx_heartbeat_req(cp_peer);
+
+ osmo_timer_schedule(&cp_peer->heartbeat_tx_timer, tval_tx_heartbeat_s, 0);
+}
+
+static void pfcp_cp_peer_heartbeat_rx_timer_cb(void *data)
+{
+ struct osmo_pfcp_cp_peer *cp_peer = data;
+
+ osmo_fsm_inst_dispatch(cp_peer->fi, PFCP_CP_PEER_EV_HEARTBEAT_TIMEOUT, NULL);
+}
+
/* Allocate PFCP CP peer FSM and start sending PFCP Association Setup Request messages to remote_addr, using endpoint
* ep. As soon as a successful response is received, change to state PFCP_CP_PEER_ST_ASSOCIATED.
*/
@@ -117,9 +175,20 @@
osmo_use_count_make_static_entries(&cp_peer->use_count, cp_peer->use_count_buf, ARRAY_SIZE(cp_peer->use_count_buf));
osmo_fsm_inst_update_id_f_sanitize(fi, '-', osmo_sockaddr_to_str_c(OTC_SELECT, &cp_peer->remote_addr));
+
+ osmo_timer_setup(&cp_peer->heartbeat_tx_timer, pfcp_cp_peer_heartbeat_tx_timer_cb, cp_peer);
+ osmo_timer_setup(&cp_peer->heartbeat_rx_timer, pfcp_cp_peer_heartbeat_rx_timer_cb, cp_peer);
return cp_peer;
}
+
+static void pfcp_cp_peer_fsm_cleanup(struct osmo_fsm_inst *fi, enum osmo_fsm_term_cause cause)
+{
+ struct osmo_pfcp_cp_peer *cp_peer = fi->priv;
+ osmo_timer_del(&cp_peer->heartbeat_tx_timer);
+ osmo_timer_del(&cp_peer->heartbeat_rx_timer);
+}
+
void osmo_pfcp_cp_peer_free(struct osmo_pfcp_cp_peer *cp_peer)
{
if (!cp_peer)
@@ -257,8 +326,18 @@
static void pfcp_cp_peer_associated_onenter(struct osmo_fsm_inst *fi, uint32_t prev_state)
{
struct osmo_pfcp_cp_peer *cp_peer = fi->priv;
+ unsigned int tval_rx_heartbeat_s =
+ osmo_tdef_get(cp_peer->ep->cfg.tdefs, OSMO_PFCP_TIMER_HEARTBEAT_RESP, OSMO_TDEF_S, -1);
LOG_CP_PEER(cp_peer, LOGL_NOTICE, "Associated with UPF %s\n",
osmo_sockaddr_to_str_c(OTC_SELECT, &cp_peer->remote_addr));
+
+ /* Send first Heartbeat Req immediately to fetch recovery timestamp info from peer.
+ * 3GPP TS 23.007 19A: "When peer PFCP entities information is available, i.e. when the PFCP Association
+ * is still alive, the restarted PFCP entity shall send its updated Recovery Time Stamps in a Heartbeat
+ * Request message to the peer PFCP entities before initiating any PFCP session signalling"
+ */
+ osmo_timer_schedule(&cp_peer->heartbeat_tx_timer, 0, 0);
+ osmo_timer_schedule(&cp_peer->heartbeat_rx_timer, tval_rx_heartbeat_s, 0);
if (cp_peer->assoc_cb)
cp_peer->assoc_cb(cp_peer, true);
}
@@ -273,6 +352,11 @@
LOG_CP_PEER(cp_peer, LOGL_ERROR, "PFCP Association Update Request is not implemented\n");
break;
+ case PFCP_CP_PEER_EV_HEARTBEAT_TIMEOUT:
+ LOG_CP_PEER(cp_peer, LOGL_NOTICE, "Heartbeat timeout!\n");
+ osmo_pfcp_cp_peer_fsm_state_chg(PFCP_CP_PEER_ST_WAIT_ASSOC_SETUP_RESP);
+ break;
+
default:
OSMO_ASSERT(false);
}
@@ -283,6 +367,8 @@
struct osmo_pfcp_cp_peer *cp_peer = fi->priv;
LOG_CP_PEER(cp_peer, LOGL_NOTICE, "Disassociating from UPF %s\n",
osmo_sockaddr_to_str_c(OTC_SELECT, &cp_peer->remote_addr));
+ osmo_timer_del(&cp_peer->heartbeat_tx_timer);
+ osmo_timer_del(&cp_peer->heartbeat_rx_timer);
if (cp_peer->assoc_cb)
cp_peer->assoc_cb(cp_peer, false);
}
@@ -358,6 +444,7 @@
.name = "associated",
.in_event_mask = 0
| S(PFCP_CP_PEER_EV_RX_ASSOC_UPDATE_REQ)
+ | S(PFCP_CP_PEER_EV_HEARTBEAT_TIMEOUT)
,
.out_state_mask = 0
| S(PFCP_CP_PEER_ST_WAIT_ASSOC_SETUP_RESP)
@@ -394,6 +481,7 @@
.num_states = ARRAY_SIZE(pfcp_cp_peer_fsm_states),
.log_subsys = DLPFCP,
.event_names = pfcp_cp_peer_fsm_event_names,
+ .cleanup = pfcp_cp_peer_fsm_cleanup,
.timer_cb = pfcp_cp_peer_fsm_timer_cb,
.allstate_action = pfcp_cp_peer_allstate_action,
.allstate_event_mask = 0
diff --git a/src/libosmo-pfcp/pfcp_endpoint.c b/src/libosmo-pfcp/pfcp_endpoint.c
index bb7fbf3..e93e272 100644
--- a/src/libosmo-pfcp/pfcp_endpoint.c
+++ b/src/libosmo-pfcp/pfcp_endpoint.c
@@ -76,7 +76,7 @@
{ .T = OSMO_PFCP_TIMER_HEARTBEAT_REQ, .default_val = 15, .unit = OSMO_TDEF_S,
.desc = "PFCP Heartbeat Request period, how long to wait between issuing requests"
},
- { .T = OSMO_PFCP_TIMER_HEARTBEAT_RESP, .default_val = 15, .unit = OSMO_TDEF_S,
+ { .T = OSMO_PFCP_TIMER_HEARTBEAT_RESP, .default_val = 35, .unit = OSMO_TDEF_S,
.desc = "PFCP Heartbeat Response timeout, the time after which to regard a non-responding peer as disconnected"
},
{ .T = OSMO_PFCP_TIMER_GRACEFUL_REL, .default_val = 15, .unit = OSMO_TDEF_S,
@@ -128,14 +128,18 @@
return ep;
}
-static unsigned int ep_n1(struct osmo_pfcp_endpoint *ep)
+static unsigned int ep_n1(struct osmo_pfcp_endpoint *ep, const struct osmo_pfcp_msg *m)
{
+ if (m->h.message_type == OSMO_PFCP_MSGT_HEARTBEAT_REQ)
+ return 0;
return osmo_tdef_get(ep->cfg.tdefs, OSMO_PFCP_TIMER_N1, OSMO_TDEF_CUSTOM, -1);
}
-static unsigned int ep_t1(struct osmo_pfcp_endpoint *ep)
+static unsigned int ep_t1(struct osmo_pfcp_endpoint *ep, const struct osmo_pfcp_msg *m)
{
- return osmo_tdef_get(ep->cfg.tdefs, OSMO_PFCP_TIMER_T1, OSMO_TDEF_MS, -1);
+ int T = m->h.message_type == OSMO_PFCP_MSGT_HEARTBEAT_REQ ? OSMO_PFCP_TIMER_HEARTBEAT_RESP :
+ OSMO_PFCP_TIMER_T1;
+ return osmo_tdef_get(ep->cfg.tdefs, T, OSMO_TDEF_MS, -1);
}
static unsigned int ep_keep_resp(const struct osmo_pfcp_endpoint *ep, const struct osmo_pfcp_msg *m)
@@ -154,8 +158,8 @@
static bool pfcp_queue_retrans(struct osmo_pfcp_queue_entry *qe)
{
struct osmo_pfcp_endpoint *endpoint = qe->ep;
- unsigned int t1_ms = ep_t1(endpoint);
struct osmo_pfcp_msg *m = qe->m;
+ unsigned int t1_ms = ep_t1(endpoint, m);
int rc;
/* if no more attempts remaining, drop from queue */
@@ -260,12 +264,12 @@
timeout_ms = ep_keep_resp(endpoint, m);
OSMO_LOG_PFCP_MSG(m, LOGL_DEBUG, "keep sent Responses for %ums\n", timeout_ms);
} else {
- timeout_ms = ep_t1(endpoint);
- n1 = ep_n1(endpoint);
+ timeout_ms = ep_t1(endpoint, m);
+ n1 = ep_n1(endpoint, m);
OSMO_LOG_PFCP_MSG(m, LOGL_DEBUG, "retransmit unanswered Requests %u x %ums\n", n1, timeout_ms);
- /* If there are no retransmissions or no timeout, it makes no sense to add to the queue. */
- if (!n1 || !timeout_ms) {
+ /* If there are no retransmissions and no timeout, it makes no sense to add to the queue. */
+ if (!n1 && !timeout_ms) {
if (!m->is_response && m->ctx.resp_cb)
m->ctx.resp_cb(m, NULL, "PFCP timeout is zero, cannot wait for a response");
return 0;
To view, visit change 42120. To unsubscribe, or for help writing mail filters, visit settings.