add pfcp_endpoint

Related: SYS#5599
Change-Id: Ic8d42e201b63064a71b40ca45a5a40e29941e8ac

5 files changed, 585 insertions, 0 deletions

diff --git a/include/osmocom/pfcp/Makefile.am b/include/osmocom/pfcp/Makefile.am
index 7bd5ea3..5d2db19 100644
--- a/include/osmocom/pfcp/Makefile.am
+++ b/include/osmocom/pfcp/Makefile.am
@@ -1,4 +1,5 @@
 pfcp_HEADERS = \
+	pfcp_endpoint.h \
 	pfcp_ies_custom.h \
 	pfcp_ies_auto.h \
 	pfcp_msg.h \
diff --git a/include/osmocom/pfcp/pfcp_endpoint.h b/include/osmocom/pfcp/pfcp_endpoint.h
new file mode 100644
index 0000000..acc878e
--- /dev/null
+++ b/include/osmocom/pfcp/pfcp_endpoint.h
@@ -0,0 +1,107 @@
+/*
+ * (C) 2021-2022 by sysmocom - s.f.m.c. GmbH <info@sysmocom.de>
+ * All Rights Reserved.
+ *
+ * Author: Neels Janosch Hofmeyr <nhofmeyr@sysmocom.de>
+ *
+ * SPDX-License-Identifier: GPL-2.0+
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#pragma once
+
+#include <osmocom/core/socket.h>
+#include <osmocom/core/select.h>
+#include <osmocom/core/tdef.h>
+
+#include <osmocom/pfcp/pfcp_msg.h>
+
+struct osmo_pfcp_endpoint;
+struct osmo_fsm_inst;
+
+#define OSMO_PFCP_TIMER_HEARTBEAT_REQ -19
+#define OSMO_PFCP_TIMER_HEARTBEAT_RESP -20
+#define OSMO_PFCP_TIMER_GRACEFUL_REL -21
+#define OSMO_PFCP_TIMER_T1 -22
+#define OSMO_PFCP_TIMER_N1 -23
+#define OSMO_PFCP_TIMER_KEEP_RESP -24
+#define OSMO_PFCP_TIMER_ASSOC_RETRY -26
+
+extern struct osmo_tdef osmo_pfcp_tdefs[];
+
+/* Ownership of m remains with the caller / m will be deallocated by the caller.
+ * \param ep  The PFCP endpoint that received and decoded the message.
+ * \param m  The message that was received.
+ * \param req  If m is a PFCP Response to an earlier Request, req is that request message. Otherwise req is NULL.
+ */
+typedef void (*osmo_pfcp_endpoint_cb)(struct osmo_pfcp_endpoint *ep, struct osmo_pfcp_msg *m,
+				      struct osmo_pfcp_msg *req);
+
+/* Send/receive PFCP messages to/from remote PFCP endpoints. */
+struct osmo_pfcp_endpoint {
+	struct {
+		/* Local address */
+		struct osmo_sockaddr local_addr;
+		/* Local PFCP Node ID, as sent in outgoing messages' Node ID IE */
+		struct osmo_pfcp_ie_node_id local_node_id;
+
+		/* Timer definitions to use, if any. See t1_ms, keep_resp_ms. Use osmo_pfcp_tdefs by default. It is
+		 * convenient to add osmo_pfcp_tdefs as one of your program's osmo_tdef_group entries and call
+		 * osmo_tdef_vty_init() to expose PFCP timers on the VTY. */
+		const struct osmo_tdef *tdefs;
+	} cfg;
+
+	/* PFCP socket */
+	struct osmo_fd pfcp_fd;
+
+	/* The time at which this endpoint last restarted, as seconds since unix epoch. */
+	uint32_t recovery_time_stamp;
+
+	/* State for determining the next sequence number for transmitting a request message */
+	uint32_t seq_nr_state;
+
+	/* This function is called just after decoding and before handling the message.
+	 * This function may set ctx.peer_fi and ctx.session_fi, used for logging context during message decoding.
+	 * The caller may also use these fi pointers to reduce lookup iterations in rx_msg().
+	 */
+	osmo_pfcp_endpoint_cb set_msg_ctx;
+
+	/* Callback to receive single incoming PFCP messages from a remote peer, already decoded. */
+	osmo_pfcp_endpoint_cb rx_msg;
+
+	/* application-private data */
+	void *priv;
+
+	/* All transmitted PFCP Request messages, list of osmo_pfcp_queue_entry.
+	 * For a transmitted Request message, wait for a matching Response from a remote peer; if none arrives,
+	 * retransmit (see n1 and t1_ms). */
+	struct llist_head sent_requests;
+	/* All transmitted PFCP Response messages, list of osmo_pfcp_queue_entry.
+	 * For a transmitted Response message, keep it in the queue for a fixed amount of time. If the peer retransmits
+	 * the original Request, do not dispatch the Request, but respond with the queued message directly. */
+	struct llist_head sent_responses;
+};
+
+struct osmo_pfcp_endpoint *osmo_pfcp_endpoint_create(void *ctx, void *priv);
+int osmo_pfcp_endpoint_bind(struct osmo_pfcp_endpoint *ep);
+void osmo_pfcp_endpoint_close(struct osmo_pfcp_endpoint *ep);
+void osmo_pfcp_endpoint_free(struct osmo_pfcp_endpoint **ep);
+
+int osmo_pfcp_endpoint_tx(struct osmo_pfcp_endpoint *ep, struct osmo_pfcp_msg *m);
+int osmo_pfcp_endpoint_tx_data(struct osmo_pfcp_endpoint *ep, struct osmo_pfcp_msg *m);
+int osmo_pfcp_endpoint_tx_heartbeat_req(struct osmo_pfcp_endpoint *ep, const struct osmo_sockaddr *remote_addr);
+
+void osmo_pfcp_endpoint_invalidate_ctx(struct osmo_pfcp_endpoint *ep, struct osmo_fsm_inst *deleted_fi);
diff --git a/include/osmocom/pfcp/pfcp_msg.h b/include/osmocom/pfcp/pfcp_msg.h
index 949800e..2833408 100644
--- a/include/osmocom/pfcp/pfcp_msg.h
+++ b/include/osmocom/pfcp/pfcp_msg.h
@@ -147,6 +147,9 @@ struct osmo_pfcp_msg {
 		osmo_pfcp_resp_cb resp_cb;
 		void *priv;
 	} ctx;
+
+	/* When a message gets encoded, the encoded packet is cached here for possible retransmissions. */
+	struct msgb *encoded;
 };
 
 /* Given a &osmo_pfcp_msg->ies pointer, return the &osmo_pfcp_msg.
diff --git a/src/libosmo-pfcp/Makefile.am b/src/libosmo-pfcp/Makefile.am
index c8accb9..db81648 100644
--- a/src/libosmo-pfcp/Makefile.am
+++ b/src/libosmo-pfcp/Makefile.am
@@ -24,6 +24,7 @@ noinst_LIBRARIES = \
 	$(NULL)
 
 libosmo_pfcp_a_SOURCES = \
+	pfcp_endpoint.c \
 	pfcp_ies_custom.c \
 	pfcp_msg.c \
 	pfcp_strs.c \
diff --git a/src/libosmo-pfcp/pfcp_endpoint.c b/src/libosmo-pfcp/pfcp_endpoint.c
new file mode 100644
index 0000000..83b7c1a
--- /dev/null
+++ b/src/libosmo-pfcp/pfcp_endpoint.c
@@ -0,0 +1,473 @@
+/*
+ * (C) 2021-2022 by sysmocom - s.f.m.c. GmbH <info@sysmocom.de>
+ * All Rights Reserved.
+ *
+ * Author: Neels Janosch Hofmeyr <nhofmeyr@sysmocom.de>
+ *
+ * SPDX-License-Identifier: GPL-2.0+
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include <errno.h>
+#include <unistd.h>
+#include <time.h>
+
+#include <osmocom/core/talloc.h>
+#include <osmocom/core/timer.h>
+#include <osmocom/core/tdef.h>
+
+#include <osmocom/pfcp/pfcp_endpoint.h>
+#include <osmocom/pfcp/pfcp_msg.h>
+
+/*! Entry of pfcp_endpoint message queue of PFCP messages, for re-transsions. */
+struct osmo_pfcp_queue_entry {
+	/* entry in per-peer list of messages waiting for a response */
+	struct llist_head entry;
+	/* back-pointer */
+	struct osmo_pfcp_endpoint *ep;
+	/* message we have transmitted */
+	struct osmo_pfcp_msg *m;
+	/* T1 timer: How long to wait for response before retransmitting */
+	struct osmo_timer_list t1;
+	/* N1: number of pending re-transmissions */
+	unsigned int n1_remaining;
+};
+
+/* Find a matching osmo_pfcp_queue_entry for given rx_hdr.
+ * A returned osmo_pfcp_queue_entry is guaranteed to be a Response if rx is a Request, and vice versa. */
+static struct osmo_pfcp_queue_entry *
+osmo_pfcp_queue_find(struct llist_head *queue, const struct osmo_pfcp_msg *rx)
+{
+	struct osmo_pfcp_queue_entry *qe;
+	/* It's important to match only a Request to a Response and vice versa, because the remote peer makes its own
+	 * sequence_nr. There could be a collision of sequence_nr. But as long as all Requests look for a Response and
+	 * vice versa, the sequence_nr scopes don't overlap. */
+	llist_for_each_entry(qe, queue, entry) {
+		if (qe->m->is_response != rx->is_response
+		    && qe->m->h.sequence_nr == rx->h.sequence_nr)
+			return qe;
+	}
+	return NULL;
+}
+
+/* clean up and deallocate the given osmo_pfcp_queue_entry */
+static void osmo_pfcp_queue_del(struct osmo_pfcp_queue_entry *qe)
+{
+	/* see also the talloc destructor: osmo_pfcp_queue_destructor() */
+	talloc_free(qe);
+}
+
+static int osmo_pfcp_queue_destructor(struct osmo_pfcp_queue_entry *qe)
+{
+	osmo_timer_del(&qe->t1);
+	llist_del(&qe->entry);
+	return 0;
+}
+
+struct osmo_tdef osmo_pfcp_tdefs[] = {
+	{ .T = OSMO_PFCP_MSGT_HEARTBEAT_REQ, .default_val = 15, .unit = OSMO_TDEF_S,
+	  .desc = "PFCP Heartbeat Request period, how long to wait between issuing requests"
+	},
+	{ .T = OSMO_PFCP_MSGT_HEARTBEAT_RESP, .default_val = 15, .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,
+	  .desc = "PFCP peer graceful shutdown timeout, how long to keep the peer's state after a peer requested"
+		  " graceful shutdown"
+	},
+	{ .T = OSMO_PFCP_TIMER_T1, .default_val = 3000, .unit = OSMO_TDEF_MS,
+	  .desc = "PFCP request timeout, how long after a missing response to retransmit a PFCP request"
+	},
+	{ .T = OSMO_PFCP_TIMER_N1, .default_val = 3, .unit = OSMO_TDEF_CUSTOM,
+	  .desc = "Number of PFCP request retransmission attempts"
+	},
+	{ .T = OSMO_PFCP_TIMER_KEEP_RESP, .default_val = 10000, .unit = OSMO_TDEF_MS,
+	  .desc = "PFCP response timeout, how long to keep a response, in case its same request is retransmitted by the peer"
+	},
+	{ .T = OSMO_PFCP_TIMER_ASSOC_RETRY, .default_val = 15, .unit = OSMO_TDEF_S,
+	  .desc = "Idle time between attempts of PFCP Association Setup (CPF)"
+	},
+	{}
+};
+
+struct osmo_pfcp_endpoint *osmo_pfcp_endpoint_create(void *ctx, void *priv)
+{
+	struct osmo_pfcp_endpoint *ep = talloc_zero(ctx, struct osmo_pfcp_endpoint);
+	uint32_t unix_time;
+	if (!ep)
+		return NULL;
+
+	INIT_LLIST_HEAD(&ep->sent_requests);
+	INIT_LLIST_HEAD(&ep->sent_responses);
+
+	ep->cfg.tdefs = osmo_pfcp_tdefs;
+	ep->priv = priv;
+	ep->pfcp_fd.fd = -1;
+
+	/* time() returns seconds since 1970 (UNIX epoch), but the recovery_time_stamp is coded in the NTP format, which is
+	 * seconds since 1900, the NTP era 0. 2208988800L is the offset between UNIX epoch and NTP era 0.
+	 * TODO: what happens when we enter NTP era 1? Is it sufficient to integer-wrap? */
+	unix_time = time(NULL);
+	ep->recovery_time_stamp = unix_time + 2208988800L;
+	LOGP(DLPFCP, LOGL_NOTICE, "PFCP endpoint: recovery timestamp = 0x%08x (%u seconds since UNIX epoch,"
+	     " which is %u seconds since NTP era 0; IETF RFC 5905)\n",
+	     ep->recovery_time_stamp, unix_time, ep->recovery_time_stamp);
+
+	return ep;
+}
+
+static unsigned int ep_n1(struct osmo_pfcp_endpoint *ep)
+{
+	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)
+{
+	return osmo_tdef_get(ep->cfg.tdefs, OSMO_PFCP_TIMER_T1, OSMO_TDEF_MS, -1);
+}
+
+static unsigned int ep_keep_resp(struct osmo_pfcp_endpoint *ep)
+{
+	return osmo_tdef_get(ep->cfg.tdefs, OSMO_PFCP_TIMER_KEEP_RESP, OSMO_TDEF_MS, -1);
+}
+
+static int osmo_pfcp_endpoint_tx_data_no_logging(struct osmo_pfcp_endpoint *ep, struct osmo_pfcp_msg *m);
+
+/* Return true to keep the message in the queue, false for dropping from the queue. */
+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;
+	int rc;
+
+	/* re-transmit */
+	if (qe->n1_remaining)
+		qe->n1_remaining--;
+	OSMO_LOG_PFCP_MSG(m, LOGL_INFO, "re-sending (%u attempts remaining)\n", qe->n1_remaining);
+
+	rc = osmo_pfcp_endpoint_tx_data_no_logging(endpoint, m);
+	/* If encoding failed, it cannot ever succeed. Drop the queue entry. */
+	if (rc)
+		return false;
+	/* if no more attempts remaining, drop from queue */
+	if (!qe->n1_remaining)
+		return false;
+	/* re-schedule timer, keep in queue */
+	osmo_timer_schedule(&qe->t1, t1_ms/1000, t1_ms%1000);
+	return true;
+}
+
+/* T1 for a given queue entry has expired */
+static void pfcp_queue_timer_cb(void *data)
+{
+	struct osmo_pfcp_queue_entry *qe = data;
+	bool keep;
+
+	if (qe->m->is_response) {
+		/* The response has waited in the queue for any retransmissions of its initiating request. Now that time
+		 * has passed and the response can be dropped from the queue. */
+		keep = false;
+	} else {
+		/* The request is still here, which means it has not received a response from the remote side.
+		 * Retransmit the request. */
+		keep = pfcp_queue_retrans(qe);
+	}
+
+	if (keep)
+		return;
+	/* Drop the queue entry. No more retransmissions. */
+	if (!qe->m->is_response && qe->m->ctx.resp_cb)
+		qe->m->ctx.resp_cb(qe->m, NULL, "PFCP retransmissions elapsed, no response received");
+	osmo_pfcp_queue_del(qe);
+}
+
+/* Directly encode and transmit the message, without storing in the retrans_queue. */
+static int osmo_pfcp_endpoint_tx_data_no_logging(struct osmo_pfcp_endpoint *ep, struct osmo_pfcp_msg *m)
+{
+	int rc;
+
+	if (!m->encoded) {
+		/* Allocate msgb as child of the message m, so that when m gets deallocated at the end of
+		 * retransmission queueing, the msgb gets deallocated with it. */
+		m->encoded = msgb_alloc_c(m, OSMO_PFCP_MSGB_ALLOC_SIZE, "PFCP-tx");
+		OSMO_ASSERT(m->encoded);
+		rc = osmo_pfcp_msg_encode(m->encoded, m);
+		if (rc) {
+			msgb_free(m->encoded);
+			m->encoded = NULL;
+			return rc;
+		}
+	}
+
+	rc = sendto(ep->pfcp_fd.fd, msgb_data(m->encoded), msgb_length(m->encoded), 0,
+		    (struct sockaddr *)&m->remote_addr, sizeof(m->remote_addr));
+	if (rc != msgb_length(m->encoded)) {
+		OSMO_LOG_PFCP_MSG(m, LOGL_ERROR, "sendto() failed: rc = %d != length %u\n",
+				  rc, msgb_length(m->encoded));
+		return -EIO;
+	}
+	return 0;
+}
+
+int osmo_pfcp_endpoint_tx_data(struct osmo_pfcp_endpoint *ep, struct osmo_pfcp_msg *m)
+{
+	OSMO_LOG_PFCP_MSG(m, LOGL_INFO, "sending\n");
+	return osmo_pfcp_endpoint_tx_data_no_logging(ep, m);
+}
+
+int osmo_pfcp_endpoint_tx_heartbeat_req(struct osmo_pfcp_endpoint *ep, const struct osmo_sockaddr *remote_addr)
+{
+	struct osmo_pfcp_msg *tx = osmo_pfcp_msg_alloc_tx(OTC_SELECT, remote_addr, NULL, NULL,
+							  OSMO_PFCP_MSGT_HEARTBEAT_REQ);
+	tx->ies.heartbeat_req.recovery_time_stamp = ep->recovery_time_stamp;
+	tx->h.sequence_nr = osmo_pfcp_next_seq_nr(&ep->seq_nr_state);
+	return osmo_pfcp_endpoint_tx_data(ep, tx);
+}
+
+/* add a given msgb to the queue of per-peer messages waiting for a response */
+static int osmo_pfcp_endpoint_retrans_queue_add(struct osmo_pfcp_endpoint *endpoint, struct osmo_pfcp_msg *m)
+{
+	struct osmo_pfcp_queue_entry *qe;
+	unsigned int n1 = ep_n1(endpoint);
+	unsigned int t1_ms = ep_t1(endpoint);
+	unsigned int keep_resp_ms = ep_keep_resp(endpoint);
+	unsigned int timeout = m->is_response ? keep_resp_ms : t1_ms;
+
+	LOGP(DLPFCP, LOGL_DEBUG, "retransmit unanswered Requests %u x %ums; keep sent Responses for %ums\n",
+	     n1, t1_ms, keep_resp_ms);
+	/* If there are no retransmissions or no timeout, it makes no sense to add to the queue. */
+	if (!n1 || !t1_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;
+	}
+
+	qe = talloc(endpoint, struct osmo_pfcp_queue_entry);
+	OSMO_ASSERT(qe);
+	*qe = (struct osmo_pfcp_queue_entry){
+		.ep = endpoint,
+		.m = m,
+		.n1_remaining = m->is_response ? 0 : n1,
+	};
+	talloc_steal(qe, m);
+
+	/* Slight optimization: Add sent requests to the start of the list: we will usually receive a response shortly
+	 * after sending a request, removing that entry from the queue quickly.
+	 * Add sent responses to the end of the list: they will rarely be retransmitted at all. */
+	if (m->is_response)
+		llist_add_tail(&qe->entry, &endpoint->sent_responses);
+	else
+		llist_add_tail(&qe->entry, &endpoint->sent_requests);
+	talloc_set_destructor(qe, osmo_pfcp_queue_destructor);
+
+	osmo_timer_setup(&qe->t1, pfcp_queue_timer_cb, qe);
+	osmo_timer_schedule(&qe->t1, timeout/1000, timeout%1000);
+	return 0;
+}
+
+/* Transmit a PFCP message.
+ * Store the message in the local message queue for possible retransmissions.
+ * On success, return zero, and pass ownership of m to ep. ep deallocates m when all retransmissions are done / a reply
+ * has been received.
+ * On error, return nonzero, and immediately deallocate m. */
+int osmo_pfcp_endpoint_tx(struct osmo_pfcp_endpoint *ep, struct osmo_pfcp_msg *m)
+{
+	struct osmo_pfcp_ie_node_id *node_id;
+	int rc;
+	if (!m->is_response)
+		m->h.sequence_nr = osmo_pfcp_next_seq_nr(&ep->seq_nr_state);
+	node_id = osmo_pfcp_msg_node_id(m);
+	if (node_id)
+		*node_id = ep->cfg.local_node_id;
+
+	rc = osmo_pfcp_endpoint_tx_data(ep, m);
+	if (rc) {
+		if (!m->is_response && m->ctx.resp_cb)
+			m->ctx.resp_cb(m, NULL, "Failed to transmit request");
+		osmo_pfcp_msg_free(m);
+		return rc;
+	}
+	osmo_pfcp_endpoint_retrans_queue_add(ep, m);
+	return 0;
+}
+
+static void osmo_pfcp_endpoint_handle_rx(struct osmo_pfcp_endpoint *ep, struct osmo_pfcp_msg *m)
+{
+	bool dispatch_rx = true;
+	struct osmo_pfcp_queue_entry *prev_msg;
+	struct osmo_pfcp_msg *req;
+
+	if (m->h.message_type == OSMO_PFCP_MSGT_HEARTBEAT_REQ) {
+		/* Directly answer with a Heartbeat Response. */
+		struct osmo_pfcp_msg *resp = osmo_pfcp_msg_alloc_tx(OTC_SELECT, NULL, NULL, m, OSMO_PFCP_MSGT_HEARTBEAT_RESP);
+		resp->ies.heartbeat_resp.recovery_time_stamp = ep->recovery_time_stamp;
+		osmo_pfcp_endpoint_tx_data(ep, resp);
+		/* Still also dispatch the Rx event to the peer. */
+	}
+
+	/* If this is receiving a response, search for matching sent request that is now completed.
+	 * If this is receiving a request, search for a matching sent response that can be retransmitted.
+	 * A match is found by sequence_nr. */
+	prev_msg = osmo_pfcp_queue_find(m->is_response ? &ep->sent_requests : &ep->sent_responses, m);
+
+	if (prev_msg && !m->is_response) {
+		/* m is a request, and we have already sent a response to this same request earlier. Retransmit the same
+		 * response, and don't dispatch the msg rx. Keep our response queued in case the request is
+		 * retransmitted yet another time. */
+
+		/* Populate message context to point at peer and session, if applicable.
+		 * With that context applied, log message rx. */
+		if (ep->set_msg_ctx)
+			ep->set_msg_ctx(ep, m, NULL);
+		OSMO_LOG_PFCP_MSG(m, LOGL_INFO, "received retransmission of earlier request\n");
+
+		/* Also log on the earlier PFCP msg that it is resent */
+		OSMO_LOG_PFCP_MSG(prev_msg->m, LOGL_INFO, "re-sending cached response\n");
+		osmo_pfcp_endpoint_tx_data_no_logging(ep, prev_msg->m);
+		return;
+	}
+
+	req = NULL;
+	if (prev_msg && m->is_response) {
+		/* m is a response to the earlier request prev_msg->m. The request is now ACKed and can be dropped from
+		 * the retransmission queue: see 'if (req)' below. */
+		req = prev_msg->m;
+	}
+
+	/* Populate message context to point at peer and session, if applicable.
+	 * With that context applied, log message rx. */
+	if (ep->set_msg_ctx)
+		ep->set_msg_ctx(ep, m, req);
+	OSMO_LOG_PFCP_MSG(m, LOGL_INFO, "received\n");
+
+	if (req && req->ctx.resp_cb) {
+		int rc = req->ctx.resp_cb(req, m, NULL);
+		/* Only dispatch the response to rx_msg() when resp_cb() asks for it with rc == 1 (or when there is no
+		 * resp_cb()). */
+		if (rc != 1) {
+			dispatch_rx = false;
+			OSMO_LOG_PFCP_MSG(m, LOGL_DEBUG,
+					  "response handled by m->resp_cb(), not dispatching to rx_msg()\n");
+		}
+	}
+
+	if (dispatch_rx)
+		ep->rx_msg(ep, m, req);
+	if (req)
+		osmo_pfcp_queue_del(prev_msg);
+}
+
+/* call-back for PFCP socket file descriptor */
+static int osmo_pfcp_fd_cb(struct osmo_fd *ofd, unsigned int what)
+{
+	int rc;
+	struct osmo_pfcp_endpoint *ep = ofd->data;
+
+	if (what & OSMO_FD_READ) {
+		struct osmo_sockaddr remote;
+		socklen_t remote_len = sizeof(remote);
+		struct msgb *msg = msgb_alloc_c(OTC_SELECT, OSMO_PFCP_MSGB_ALLOC_SIZE, "PFCP-rx");
+		if (!msg)
+			return -ENOMEM;
+
+		msg->l3h = msg->tail;
+		rc = recvfrom(ofd->fd, msg->tail, msgb_tailroom(msg), 0, (struct sockaddr *)&remote, &remote_len);
+		if (rc <= 0)
+			return -EIO;
+		msgb_put(msg, rc);
+
+		OSMO_ASSERT(ep->rx_msg);
+
+		/* This may be a bundle of PFCP messages. Parse and receive each message received, by shifting l4h
+		 * through the message bundle. */
+		msg->l4h = msg->l3h;
+		while (msgb_l4len(msg)) {
+			struct osmo_gtlv_load tlv;
+			struct osmo_pfcp_msg *m = osmo_pfcp_msg_alloc_rx(OTC_SELECT, &remote);
+			m->encoded = msg;
+
+			rc = osmo_pfcp_msg_decode_header(&tlv, m, msg);
+			if (rc < 0)
+				break;
+			msg->l4h += rc;
+
+			rc = osmo_pfcp_msg_decode_tlv(m, &tlv);
+			/* If errors occurred, they have already been logged on DLPFCP. */
+			if (rc == 0)
+				osmo_pfcp_endpoint_handle_rx(ep, m);
+			osmo_pfcp_msg_free(m);
+		}
+		msgb_free(msg);
+	}
+	return 0;
+}
+
+/*! bind a PFCP endpoint to its configured address (ep->cfg.local_addr).
+ * \return 0 on success, negative on error. */
+int osmo_pfcp_endpoint_bind(struct osmo_pfcp_endpoint *ep)
+{
+	int rc;
+	/* close the existing socket, if any */
+	osmo_pfcp_endpoint_close(ep);
+
+	if (!ep->rx_msg) {
+		LOGP(DLPFCP, LOGL_ERROR, "missing rx_msg cb at osmo_pfcp_endpoint\n");
+		return -EINVAL;
+	}
+
+	/* create the new socket, binding to configured local address */
+	ep->pfcp_fd.cb = osmo_pfcp_fd_cb;
+	ep->pfcp_fd.data = ep;
+	rc = osmo_sock_init_osa_ofd(&ep->pfcp_fd, SOCK_DGRAM, IPPROTO_UDP, &ep->cfg.local_addr, NULL, OSMO_SOCK_F_BIND);
+	if (rc < 0)
+		return rc;
+	return 0;
+}
+
+void osmo_pfcp_endpoint_close(struct osmo_pfcp_endpoint *ep)
+{
+	struct osmo_pfcp_queue_entry *qe;
+	while ((qe = llist_first_entry_or_null(&ep->sent_requests, struct osmo_pfcp_queue_entry, entry)))
+		osmo_pfcp_queue_del(qe);
+	while ((qe = llist_first_entry_or_null(&ep->sent_responses, struct osmo_pfcp_queue_entry, entry)))
+		osmo_pfcp_queue_del(qe);
+
+	if (ep->pfcp_fd.fd != -1) {
+		osmo_fd_unregister(&ep->pfcp_fd);
+		close(ep->pfcp_fd.fd);
+		ep->pfcp_fd.fd = -1;
+	}
+}
+
+void osmo_pfcp_endpoint_free(struct osmo_pfcp_endpoint **ep)
+{
+	if (!*ep)
+		return;
+	osmo_pfcp_endpoint_close(*ep);
+	talloc_free(*ep);
+	*ep = NULL;
+}
+
+/* Call osmo_pfcp_msg_invalidate_ctx(deleted_fi) on all queued osmo_pfcp_msg instances in the retrans_queue. */
+void osmo_pfcp_endpoint_invalidate_ctx(struct osmo_pfcp_endpoint *ep, struct osmo_fsm_inst *deleted_fi)
+{
+	struct osmo_pfcp_queue_entry *qe;
+	llist_for_each_entry(qe, &ep->sent_requests, entry)
+		osmo_pfcp_msg_invalidate_ctx(qe->m, deleted_fi);
+	llist_for_each_entry(qe, &ep->sent_responses, entry)
+		osmo_pfcp_msg_invalidate_ctx(qe->m, deleted_fi);
+}