Adding upstream version 8.4.4.upstream/8.4.4upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

1 files changed, 1189 insertions, 0 deletions

diff --git a/bfdd/dplane.c b/bfdd/dplane.c
new file mode 100644
index 0000000..c108164
--- /dev/null
+++ b/bfdd/dplane.c
@@ -0,0 +1,1189 @@
+/*
+ * BFD data plane implementation (distributed BFD).
+ *
+ * Copyright (C) 2020 Network Device Education Foundation, Inc. ("NetDEF")
+ *                    Rafael Zalamena
+ *
+ * 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; see the file COPYING; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <zebra.h>
+
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+
+#ifdef __FreeBSD__
+#include <sys/endian.h>
+#else
+#include <endian.h>
+#endif /* __FreeBSD__ */
+
+#include <errno.h>
+#include <time.h>
+
+#include "lib/hook.h"
+#include "lib/network.h"
+#include "lib/printfrr.h"
+#include "lib/stream.h"
+#include "lib/thread.h"
+
+#include "bfd.h"
+#include "bfddp_packet.h"
+
+#include "lib/openbsd-queue.h"
+
+DEFINE_MTYPE_STATIC(BFDD, BFDD_DPLANE_CTX,
+		    "Data plane client allocated memory");
+
+/** Data plane client socket buffer size. */
+#define BFD_DPLANE_CLIENT_BUF_SIZE 8192
+
+struct bfd_dplane_ctx {
+	/** Client file descriptor. */
+	int sock;
+	/** Is this a connected or accepted? */
+	bool client;
+	/** Is the socket still connecting? */
+	bool connecting;
+	/** Client/server address. */
+	union {
+		struct sockaddr sa;
+		struct sockaddr_in sin;
+		struct sockaddr_in6 sin6;
+		struct sockaddr_un sun;
+	} addr;
+	/** Address length. */
+	socklen_t addrlen;
+	/** Data plane current last used ID. */
+	uint16_t last_id;
+
+	/** Input buffer data. */
+	struct stream *inbuf;
+	/** Output buffer data. */
+	struct stream *outbuf;
+	/** Input event data. */
+	struct thread *inbufev;
+	/** Output event data. */
+	struct thread *outbufev;
+	/** Connection event. */
+	struct thread *connectev;
+
+	/** Amount of bytes read. */
+	uint64_t in_bytes;
+	/** Amount of bytes read peak. */
+	uint64_t in_bytes_peak;
+	/** Amount of bytes written. */
+	uint64_t out_bytes;
+	/** Amount of bytes written peak. */
+	uint64_t out_bytes_peak;
+	/** Amount of output buffer full events (`bfd_dplane_enqueue` failed).
+	 */
+	uint64_t out_fullev;
+
+	/** Amount of messages read (full messages). */
+	uint64_t in_msgs;
+	/** Amount of messages enqueued (maybe written). */
+	uint64_t out_msgs;
+
+	TAILQ_ENTRY(bfd_dplane_ctx) entry;
+};
+
+/**
+ * Callback type for `bfd_dplane_expect`. \see bfd_dplane_expect.
+ */
+typedef void (*bfd_dplane_expect_cb)(struct bfddp_message *msg, void *arg);
+
+static void bfd_dplane_client_connect(struct thread *t);
+static bool bfd_dplane_client_connecting(struct bfd_dplane_ctx *bdc);
+static void bfd_dplane_ctx_free(struct bfd_dplane_ctx *bdc);
+static int _bfd_dplane_add_session(struct bfd_dplane_ctx *bdc,
+				   struct bfd_session *bs);
+
+/*
+ * BFD data plane helper functions.
+ */
+static const char *bfd_dplane_messagetype2str(enum bfddp_message_type bmt)
+{
+	switch (bmt) {
+	case ECHO_REQUEST:
+		return "ECHO_REQUEST";
+	case ECHO_REPLY:
+		return "ECHO_REPLY";
+	case DP_ADD_SESSION:
+		return "DP_ADD_SESSION";
+	case DP_DELETE_SESSION:
+		return "DP_DELETE_SESSION";
+	case BFD_STATE_CHANGE:
+		return "BFD_STATE_CHANGE";
+	case DP_REQUEST_SESSION_COUNTERS:
+		return "DP_REQUEST_SESSION_COUNTERS";
+	case BFD_SESSION_COUNTERS:
+		return "BFD_SESSION_COUNTERS";
+	default:
+		return "UNKNOWN";
+	}
+}
+
+static void bfd_dplane_debug_message(const struct bfddp_message *msg)
+{
+	enum bfddp_message_type bmt;
+	char buf[256], addrs[256];
+	uint32_t flags;
+	int rv;
+
+	if (!bglobal.debug_dplane)
+		return;
+
+	bmt = ntohs(msg->header.type);
+	zlog_debug("dplane-packet: [version=%d length=%d type=%s (%d)]",
+		   msg->header.version, ntohs(msg->header.length),
+		   bfd_dplane_messagetype2str(bmt), bmt);
+
+	switch (bmt) {
+	case ECHO_REPLY:
+	case ECHO_REQUEST:
+		zlog_debug("  [dp_time=%" PRIu64 " bfdd_time=%" PRIu64 "]",
+			   be64toh(msg->data.echo.dp_time),
+			   be64toh(msg->data.echo.bfdd_time));
+		break;
+
+	case DP_ADD_SESSION:
+	case DP_DELETE_SESSION:
+		flags = ntohl(msg->data.session.flags);
+		if (flags & SESSION_IPV6)
+			snprintfrr(addrs, sizeof(addrs), "src=%pI6 dst=%pI6",
+				   &msg->data.session.src,
+				   &msg->data.session.dst);
+		else
+			snprintfrr(addrs, sizeof(addrs), "src=%pI4 dst=%pI4",
+				   (struct in_addr *)&msg->data.session.src,
+				   (struct in_addr *)&msg->data.session.dst);
+
+		buf[0] = 0;
+		if (flags & SESSION_CBIT)
+			strlcat(buf, "cpi ", sizeof(buf));
+		if (flags & SESSION_ECHO)
+			strlcat(buf, "echo ", sizeof(buf));
+		if (flags & SESSION_IPV6)
+			strlcat(buf, "ipv6 ", sizeof(buf));
+		if (flags & SESSION_DEMAND)
+			strlcat(buf, "demand ", sizeof(buf));
+		if (flags & SESSION_PASSIVE)
+			strlcat(buf, "passive ", sizeof(buf));
+		if (flags & SESSION_MULTIHOP)
+			strlcat(buf, "multihop ", sizeof(buf));
+		if (flags & SESSION_SHUTDOWN)
+			strlcat(buf, "shutdown ", sizeof(buf));
+
+		/* Remove the last space to make things prettier. */
+		rv = (int)strlen(buf);
+		if (rv > 0)
+			buf[rv - 1] = 0;
+
+		zlog_debug(
+			"  [flags=0x%08x{%s} %s ttl=%d detect_mult=%d "
+			"ifindex=%d ifname=%s]",
+			flags, buf, addrs, msg->data.session.ttl,
+			msg->data.session.detect_mult,
+			ntohl(msg->data.session.ifindex),
+			msg->data.session.ifname);
+		break;
+
+	case BFD_STATE_CHANGE:
+		buf[0] = 0;
+		flags = ntohl(msg->data.state.remote_flags);
+		if (flags & RBIT_CPI)
+			strlcat(buf, "cbit ", sizeof(buf));
+		if (flags & RBIT_DEMAND)
+			strlcat(buf, "demand ", sizeof(buf));
+		if (flags & RBIT_MP)
+			strlcat(buf, "mp ", sizeof(buf));
+
+		/* Remove the last space to make things prettier. */
+		rv = (int)strlen(buf);
+		if (rv > 0)
+			buf[rv - 1] = 0;
+
+		zlog_debug(
+			"  [lid=%u rid=%u flags=0x%02x{%s} state=%s "
+			"diagnostics=%s mult=%d tx=%u rx=%u erx=%u]",
+			ntohl(msg->data.state.lid), ntohl(msg->data.state.rid),
+			flags, buf, state_list[msg->data.state.state].str,
+			diag2str(msg->data.state.diagnostics),
+			msg->data.state.detection_multiplier,
+			ntohl(msg->data.state.desired_tx),
+			ntohl(msg->data.state.required_rx),
+			ntohl(msg->data.state.required_echo_rx));
+		break;
+
+	case DP_REQUEST_SESSION_COUNTERS:
+		zlog_debug("  [lid=%u]", ntohl(msg->data.counters_req.lid));
+		break;
+
+	case BFD_SESSION_COUNTERS:
+		zlog_debug(
+			"  [lid=%u "
+			"control{in %" PRIu64 " bytes (%" PRIu64
+			" packets), "
+			"out %" PRIu64 " bytes (%" PRIu64
+			" packets)} "
+			"echo{in %" PRIu64 " bytes (%" PRIu64
+			" packets), "
+			"out %" PRIu64 " bytes (%" PRIu64 " packets)}]",
+			ntohl(msg->data.session_counters.lid),
+			be64toh(msg->data.session_counters.control_input_bytes),
+			be64toh(msg->data.session_counters
+				.control_input_packets),
+			be64toh(msg->data.session_counters
+				.control_output_bytes),
+			be64toh(msg->data.session_counters
+				.control_output_packets),
+			be64toh(msg->data.session_counters.echo_input_bytes),
+			be64toh(msg->data.session_counters.echo_input_packets),
+			be64toh(msg->data.session_counters.echo_output_bytes),
+			be64toh(msg->data.session_counters
+				.echo_output_packets));
+		break;
+	}
+}
+
+/**
+ * Gets the next unused non zero identification.
+ *
+ * \param bdc the data plane context.
+ *
+ * \returns next usable id.
+ */
+static uint16_t bfd_dplane_next_id(struct bfd_dplane_ctx *bdc)
+{
+	bdc->last_id++;
+
+	/* Don't use reserved id `0`. */
+	if (bdc->last_id == 0)
+		bdc->last_id = 1;
+
+	return bdc->last_id;
+}
+
+static ssize_t bfd_dplane_flush(struct bfd_dplane_ctx *bdc)
+{
+	ssize_t total = 0;
+	int rv;
+
+	while (STREAM_READABLE(bdc->outbuf)) {
+		/* Flush buffer contents to socket. */
+		rv = stream_flush(bdc->outbuf, bdc->sock);
+		if (rv == -1) {
+			/* Interruption: try again. */
+			if (errno == EAGAIN || errno == EWOULDBLOCK
+			    || errno == EINTR)
+				continue;
+
+			zlog_warn("%s: socket failed: %s", __func__,
+				  strerror(errno));
+			bfd_dplane_ctx_free(bdc);
+			return 0;
+		}
+		if (rv == 0) {
+			if (bglobal.debug_dplane)
+				zlog_info("%s: connection closed", __func__);
+
+			bfd_dplane_ctx_free(bdc);
+			return 0;
+		}
+
+		/* Account total written. */
+		total += rv;
+
+		/* Account output bytes. */
+		bdc->out_bytes += (uint64_t)rv;
+
+		/* Forward pointer. */
+		stream_forward_getp(bdc->outbuf, (size_t)rv);
+	}
+
+	/* Make more space for new data. */
+	stream_pulldown(bdc->outbuf);
+
+	/* Disable write ready events. */
+	THREAD_OFF(bdc->outbufev);
+
+	return total;
+}
+
+static void bfd_dplane_write(struct thread *t)
+{
+	struct bfd_dplane_ctx *bdc = THREAD_ARG(t);
+
+	/* Handle connection stage. */
+	if (bdc->connecting && bfd_dplane_client_connecting(bdc))
+		return;
+
+	bfd_dplane_flush(bdc);
+}
+
+static void
+bfd_dplane_session_state_change(struct bfd_dplane_ctx *bdc,
+				const struct bfddp_state_change *state)
+{
+	struct bfd_session *bs;
+	uint32_t flags;
+	int old_state;
+
+	/* Look up session. */
+	bs = bfd_id_lookup(ntohl(state->lid));
+	if (bs == NULL) {
+		if (bglobal.debug_dplane)
+			zlog_debug("%s: failed to find session to update",
+				   __func__);
+		return;
+	}
+
+	flags = ntohl(state->remote_flags);
+	old_state = bs->ses_state;
+
+	/* Update session state. */
+	bs->ses_state = state->state;
+	bs->remote_diag = state->diagnostics;
+	bs->discrs.remote_discr = ntohl(state->rid);
+	bs->remote_cbit = !!(flags & RBIT_CPI);
+	bs->remote_detect_mult = state->detection_multiplier;
+	bs->remote_timers.desired_min_tx = ntohl(state->desired_tx);
+	bs->remote_timers.required_min_rx = ntohl(state->required_rx);
+	bs->remote_timers.required_min_echo = ntohl(state->required_echo_rx);
+
+	/* Notify and update counters. */
+	control_notify(bs, bs->ses_state);
+
+	/* No state change. */
+	if (old_state == bs->ses_state)
+		return;
+
+	switch (bs->ses_state) {
+	case PTM_BFD_ADM_DOWN:
+	case PTM_BFD_DOWN:
+		/* Both states mean down. */
+		if (old_state == PTM_BFD_ADM_DOWN || old_state == PTM_BFD_DOWN)
+			break;
+
+		monotime(&bs->downtime);
+		bs->stats.session_down++;
+		break;
+	case PTM_BFD_UP:
+		monotime(&bs->uptime);
+		bs->stats.session_up++;
+		break;
+	case PTM_BFD_INIT:
+		/* NOTHING */
+		break;
+
+	default:
+		zlog_warn("%s: unhandled new state %d", __func__,
+			  bs->ses_state);
+		break;
+	}
+
+	if (bglobal.debug_peer_event)
+		zlog_debug("state-change: [data plane: %s] %s -> %s",
+			   bs_to_string(bs), state_list[old_state].str,
+			   state_list[bs->ses_state].str);
+}
+
+/**
+ * Enqueue message in output buffer.
+ *
+ * \param[in,out] bdc data plane client context.
+ * \param[in] buf the message to buffer.
+ * \param[in] buflen the amount of bytes to buffer.
+ *
+ * \returns `-1` on failure (buffer full) or `0` on success.
+ */
+static int bfd_dplane_enqueue(struct bfd_dplane_ctx *bdc, const void *buf,
+			      size_t buflen)
+{
+	size_t rlen;
+
+	/* Handle not connected yet client. */
+	if (bdc->client && bdc->sock == -1)
+		return -1;
+
+	/* Not enough space. */
+	if (buflen > STREAM_WRITEABLE(bdc->outbuf)) {
+		bdc->out_fullev++;
+		return -1;
+	}
+
+	/* Show debug message if active. */
+	bfd_dplane_debug_message((struct bfddp_message *)buf);
+
+	/* Buffer the message. */
+	stream_write(bdc->outbuf, buf, buflen);
+
+	/* Account message as sent. */
+	bdc->out_msgs++;
+	/* Register peak buffered bytes. */
+	rlen = STREAM_READABLE(bdc->outbuf);
+	if (bdc->out_bytes_peak < rlen)
+		bdc->out_bytes_peak = rlen;
+
+	/* Schedule if it is not yet. */
+	if (bdc->outbufev == NULL)
+		thread_add_write(master, bfd_dplane_write, bdc, bdc->sock,
+				 &bdc->outbufev);
+
+	return 0;
+}
+
+static void bfd_dplane_echo_request_handle(struct bfd_dplane_ctx *bdc,
+					   const struct bfddp_message *bm)
+{
+	struct bfddp_message msg = {};
+	uint16_t msglen = sizeof(msg.header) + sizeof(msg.data.echo);
+	struct timeval tv;
+
+	gettimeofday(&tv, NULL);
+
+	/* Prepare header. */
+	msg.header.version = BFD_DP_VERSION;
+	msg.header.type = htons(ECHO_REPLY);
+	msg.header.length = htons(msglen);
+
+	/* Prepare payload. */
+	msg.data.echo.dp_time = bm->data.echo.dp_time;
+	msg.data.echo.bfdd_time =
+		htobe64((uint64_t)((tv.tv_sec * 1000000) + tv.tv_usec));
+
+	/* Enqueue for output. */
+	bfd_dplane_enqueue(bdc, &msg, msglen);
+}
+
+static void bfd_dplane_handle_message(struct bfddp_message *msg, void *arg)
+{
+	enum bfddp_message_type bmt;
+	struct bfd_dplane_ctx *bdc = arg;
+
+	/* Call the appropriated handler. */
+	bmt = ntohs(msg->header.type);
+	switch (bmt) {
+	case ECHO_REQUEST:
+		bfd_dplane_echo_request_handle(bdc, msg);
+		break;
+	case BFD_STATE_CHANGE:
+		bfd_dplane_session_state_change(bdc, &msg->data.state);
+		break;
+	case ECHO_REPLY:
+		/* NOTHING: we don't do anything with this information. */
+		break;
+	case DP_ADD_SESSION:
+	case DP_DELETE_SESSION:
+	case DP_REQUEST_SESSION_COUNTERS:
+		/* NOTHING: we are not supposed to receive this. */
+		break;
+	case BFD_SESSION_COUNTERS:
+		/*
+		 * NOTHING: caller of DP_REQUEST_SESSION_COUNTERS should
+		 * handle this with `bfd_dplane_expect`.
+		 */
+		break;
+
+	default:
+		zlog_debug("%s: unhandled message type %d", __func__, bmt);
+		break;
+	}
+}
+
+/**
+ * Reads the socket immediately to receive data plane answer to query.
+ *
+ * \param bdc the data plane context.
+ * \param id the message ID waiting response.
+ * \param cb the callback to call when ready.
+ * \param arg the callback argument.
+ *
+ * \return
+ * `-2` on unavailability (try again), `-1` on failure or `0` on success.
+ */
+static int bfd_dplane_expect(struct bfd_dplane_ctx *bdc, uint16_t id,
+			     bfd_dplane_expect_cb cb, void *arg)
+{
+	struct bfddp_message_header *bh;
+	size_t rlen = 0, reads = 0;
+	ssize_t rv;
+
+	/*
+	 * Don't attempt to read if buffer is full, otherwise we'll get a
+	 * bogus 'connection closed' signal (rv == 0).
+	 */
+	if (bdc->inbuf->endp == bdc->inbuf->size)
+		goto skip_read;
+
+read_again:
+	/* Attempt to read message from client. */
+	rv = stream_read_try(bdc->inbuf, bdc->sock,
+			     STREAM_WRITEABLE(bdc->inbuf));
+	if (rv == 0) {
+		if (bglobal.debug_dplane)
+			zlog_info("%s: socket closed", __func__);
+
+		bfd_dplane_ctx_free(bdc);
+		return -1;
+	}
+	if (rv == -1) {
+		zlog_warn("%s: socket failed: %s", __func__, strerror(errno));
+		bfd_dplane_ctx_free(bdc);
+		return -1;
+	}
+
+	/* We got interrupted, reschedule read. */
+	if (rv == -2)
+		return -2;
+
+	/* Account read bytes. */
+	bdc->in_bytes += (uint64_t)rv;
+	/* Register peak buffered bytes. */
+	rlen = STREAM_READABLE(bdc->inbuf);
+	if (bdc->in_bytes_peak < rlen)
+		bdc->in_bytes_peak = rlen;
+
+skip_read:
+	while (rlen > 0) {
+		bh = (struct bfddp_message_header *)stream_pnt(bdc->inbuf);
+		/* Not enough data read. */
+		if (ntohs(bh->length) > rlen)
+			goto read_again;
+
+		/* Account full message read. */
+		bdc->in_msgs++;
+
+		/* Account this message as whole read for buffer reorganize. */
+		reads++;
+
+		/* Check for bad version. */
+		if (bh->version != BFD_DP_VERSION) {
+			zlog_err("%s: bad data plane client version: %d",
+				 __func__, bh->version);
+			return -1;
+		}
+
+		/* Show debug message if active. */
+		bfd_dplane_debug_message((struct bfddp_message *)bh);
+
+		/*
+		 * Handle incoming message with callback if the ID matches,
+		 * otherwise fallback to default handler.
+		 */
+		if (id && ntohs(bh->id) == id)
+			cb((struct bfddp_message *)bh, arg);
+		else
+			bfd_dplane_handle_message((struct bfddp_message *)bh,
+						  bdc);
+
+		/* Advance current read pointer. */
+		stream_forward_getp(bdc->inbuf, ntohs(bh->length));
+
+		/* Reduce the buffer available bytes. */
+		rlen -= ntohs(bh->length);
+
+		/* Reorganize buffer to handle more bytes read. */
+		if (reads >= 3) {
+			stream_pulldown(bdc->inbuf);
+			reads = 0;
+		}
+
+		/* We found the message, return to caller. */
+		if (id && ntohs(bh->id) == id)
+			break;
+	}
+
+	return 0;
+}
+
+static void bfd_dplane_read(struct thread *t)
+{
+	struct bfd_dplane_ctx *bdc = THREAD_ARG(t);
+	int rv;
+
+	rv = bfd_dplane_expect(bdc, 0, bfd_dplane_handle_message, NULL);
+	if (rv == -1)
+		return;
+
+	stream_pulldown(bdc->inbuf);
+	thread_add_read(master, bfd_dplane_read, bdc, bdc->sock, &bdc->inbufev);
+}
+
+static void _bfd_session_register_dplane(struct hash_bucket *hb, void *arg)
+{
+	struct bfd_session *bs = hb->data;
+	struct bfd_dplane_ctx *bdc = arg;
+
+	if (bs->bdc != NULL)
+		return;
+
+	/* Disable software session. */
+	bfd_session_disable(bs);
+
+	/* Move session to data plane. */
+	_bfd_dplane_add_session(bdc, bs);
+}
+
+static struct bfd_dplane_ctx *bfd_dplane_ctx_new(int sock)
+{
+	struct bfd_dplane_ctx *bdc;
+
+	bdc = XCALLOC(MTYPE_BFDD_DPLANE_CTX, sizeof(*bdc));
+
+	bdc->sock = sock;
+	bdc->inbuf = stream_new(BFD_DPLANE_CLIENT_BUF_SIZE);
+	bdc->outbuf = stream_new(BFD_DPLANE_CLIENT_BUF_SIZE);
+
+	/* If not socket ready, skip read and session registration. */
+	if (sock == -1)
+		return bdc;
+
+	thread_add_read(master, bfd_dplane_read, bdc, sock, &bdc->inbufev);
+
+	/* Register all unattached sessions. */
+	bfd_key_iterate(_bfd_session_register_dplane, bdc);
+
+	return bdc;
+}
+
+static void _bfd_session_unregister_dplane(struct hash_bucket *hb, void *arg)
+{
+	struct bfd_session *bs = hb->data;
+	struct bfd_dplane_ctx *bdc = arg;
+
+	if (bs->bdc != bdc)
+		return;
+
+	bs->bdc = NULL;
+
+	/* Fallback to software. */
+	bfd_session_enable(bs);
+}
+
+static void bfd_dplane_ctx_free(struct bfd_dplane_ctx *bdc)
+{
+	if (bglobal.debug_dplane)
+		zlog_debug("%s: terminating data plane client %d", __func__,
+			   bdc->sock);
+
+	/* Client mode has special treatment. */
+	if (bdc->client) {
+		/* Disable connection event if any. */
+		THREAD_OFF(bdc->connectev);
+
+		/* Normal treatment on shutdown. */
+		if (bglobal.bg_shutdown)
+			goto free_resources;
+
+		/* Attempt reconnection. */
+		socket_close(&bdc->sock);
+		THREAD_OFF(bdc->inbufev);
+		THREAD_OFF(bdc->outbufev);
+		thread_add_timer(master, bfd_dplane_client_connect, bdc, 3,
+				 &bdc->connectev);
+		return;
+	}
+
+free_resources:
+	/* Remove from the list of attached data planes. */
+	TAILQ_REMOVE(&bglobal.bg_dplaneq, bdc, entry);
+
+	/* Detach all associated sessions. */
+	if (bglobal.bg_shutdown == false)
+		bfd_key_iterate(_bfd_session_unregister_dplane, bdc);
+
+	/* Free resources. */
+	socket_close(&bdc->sock);
+	stream_free(bdc->inbuf);
+	stream_free(bdc->outbuf);
+	THREAD_OFF(bdc->inbufev);
+	THREAD_OFF(bdc->outbufev);
+	XFREE(MTYPE_BFDD_DPLANE_CTX, bdc);
+}
+
+static void _bfd_dplane_session_fill(const struct bfd_session *bs,
+				     struct bfddp_message *msg)
+{
+	uint16_t msglen = sizeof(msg->header) + sizeof(msg->data.session);
+
+	/* Message header. */
+	msg->header.version = BFD_DP_VERSION;
+	msg->header.length = ntohs(msglen);
+	msg->header.type = ntohs(DP_ADD_SESSION);
+
+	/* Message payload. */
+	msg->data.session.dst = bs->key.peer;
+	msg->data.session.src = bs->key.local;
+	msg->data.session.detect_mult = bs->detect_mult;
+
+	if (bs->ifp) {
+		msg->data.session.ifindex = htonl(bs->ifp->ifindex);
+		strlcpy(msg->data.session.ifname, bs->ifp->name,
+			sizeof(msg->data.session.ifname));
+	}
+	if (bs->flags & BFD_SESS_FLAG_MH) {
+		msg->data.session.flags |= SESSION_MULTIHOP;
+		msg->data.session.ttl = bs->mh_ttl;
+	} else
+		msg->data.session.ttl = BFD_TTL_VAL;
+
+	if (bs->flags & BFD_SESS_FLAG_IPV6)
+		msg->data.session.flags |= SESSION_IPV6;
+	if (bs->flags & BFD_SESS_FLAG_ECHO)
+		msg->data.session.flags |= SESSION_ECHO;
+	if (bs->flags & BFD_SESS_FLAG_CBIT)
+		msg->data.session.flags |= SESSION_CBIT;
+	if (bs->flags & BFD_SESS_FLAG_PASSIVE)
+		msg->data.session.flags |= SESSION_PASSIVE;
+	if (bs->flags & BFD_SESS_FLAG_SHUTDOWN)
+		msg->data.session.flags |= SESSION_SHUTDOWN;
+
+	msg->data.session.flags = htonl(msg->data.session.flags);
+	msg->data.session.lid = htonl(bs->discrs.my_discr);
+	msg->data.session.min_tx = htonl(bs->timers.desired_min_tx);
+	msg->data.session.min_rx = htonl(bs->timers.required_min_rx);
+	msg->data.session.min_echo_tx = htonl(bs->timers.desired_min_echo_tx);
+	msg->data.session.min_echo_rx = htonl(bs->timers.required_min_echo_rx);
+}
+
+static int _bfd_dplane_add_session(struct bfd_dplane_ctx *bdc,
+				   struct bfd_session *bs)
+{
+	int rv;
+
+	/* Associate session. */
+	bs->bdc = bdc;
+
+	/* Reset previous state. */
+	bs->remote_diag = 0;
+	bs->local_diag = 0;
+	bs->ses_state = PTM_BFD_DOWN;
+
+	/* Enqueue message to data plane client. */
+	rv = bfd_dplane_update_session(bs);
+	if (rv != 0)
+		bs->bdc = NULL;
+
+	return rv;
+}
+
+static void _bfd_dplane_update_session_counters(struct bfddp_message *msg,
+						void *arg)
+{
+	struct bfd_session *bs = arg;
+
+	bs->stats.rx_ctrl_pkt =
+		be64toh(msg->data.session_counters.control_input_packets);
+	bs->stats.tx_ctrl_pkt =
+		be64toh(msg->data.session_counters.control_output_packets);
+	bs->stats.rx_echo_pkt =
+		be64toh(msg->data.session_counters.echo_input_packets);
+	bs->stats.tx_echo_pkt =
+		be64toh(msg->data.session_counters.echo_output_bytes);
+}
+
+/**
+ * Send message to data plane requesting the session counters.
+ *
+ * \param bs the BFD session.
+ *
+ * \returns `0` on failure or the request id.
+ */
+static uint16_t bfd_dplane_request_counters(const struct bfd_session *bs)
+{
+	struct bfddp_message msg = {};
+	size_t msglen = sizeof(msg.header) + sizeof(msg.data.counters_req);
+
+	/* Fill header information. */
+	msg.header.version = BFD_DP_VERSION;
+	msg.header.length = htons(msglen);
+	msg.header.type = htons(DP_REQUEST_SESSION_COUNTERS);
+	msg.header.id = htons(bfd_dplane_next_id(bs->bdc));
+
+	/* Session to get counters. */
+	msg.data.counters_req.lid = htonl(bs->discrs.my_discr);
+
+	/* If enqueue failed, let caller know. */
+	if (bfd_dplane_enqueue(bs->bdc, &msg, msglen) == -1)
+		return 0;
+
+	/* Flush socket. */
+	bfd_dplane_flush(bs->bdc);
+
+	return ntohs(msg.header.id);
+}
+
+/*
+ * Data plane listening socket.
+ */
+static void bfd_dplane_accept(struct thread *t)
+{
+	struct bfd_global *bg = THREAD_ARG(t);
+	struct bfd_dplane_ctx *bdc;
+	int sock;
+
+	/* Accept new connection. */
+	sock = accept(bg->bg_dplane_sock, NULL, 0);
+	if (sock == -1) {
+		zlog_warn("%s: accept failed: %s", __func__, strerror(errno));
+		goto reschedule_and_return;
+	}
+
+	/* Create and handle new connection. */
+	bdc = bfd_dplane_ctx_new(sock);
+	TAILQ_INSERT_TAIL(&bglobal.bg_dplaneq, bdc, entry);
+
+	if (bglobal.debug_dplane)
+		zlog_debug("%s: new data plane client connected", __func__);
+
+reschedule_and_return:
+	thread_add_read(master, bfd_dplane_accept, bg, bg->bg_dplane_sock,
+			&bglobal.bg_dplane_sockev);
+}
+
+/*
+ * Data plane connecting socket.
+ */
+static void _bfd_dplane_client_bootstrap(struct bfd_dplane_ctx *bdc)
+{
+	bdc->connecting = false;
+
+	/* Clean up buffers. */
+	stream_reset(bdc->inbuf);
+	stream_reset(bdc->outbuf);
+
+	/* Ask for read notifications. */
+	thread_add_read(master, bfd_dplane_read, bdc, bdc->sock, &bdc->inbufev);
+
+	/* Remove all sessions then register again to send them all. */
+	bfd_key_iterate(_bfd_session_unregister_dplane, bdc);
+	bfd_key_iterate(_bfd_session_register_dplane, bdc);
+}
+
+static bool bfd_dplane_client_connecting(struct bfd_dplane_ctx *bdc)
+{
+	int rv;
+	socklen_t rvlen = sizeof(rv);
+
+	/* Make sure `errno` is reset, then test `getsockopt` success. */
+	errno = 0;
+	if (getsockopt(bdc->sock, SOL_SOCKET, SO_ERROR, &rv, &rvlen) == -1)
+		rv = -1;
+
+	/* Connection successful. */
+	if (rv == 0) {
+		if (bglobal.debug_dplane)
+			zlog_debug("%s: connected to server: %d", __func__,
+				   bdc->sock);
+
+		_bfd_dplane_client_bootstrap(bdc);
+		return false;
+	}
+
+	switch (rv) {
+	case EINTR:
+	case EAGAIN:
+	case EALREADY:
+	case EINPROGRESS:
+		/* non error, wait more. */
+		return true;
+
+	default:
+		zlog_warn("%s: connection failed: %s", __func__,
+			  strerror(errno));
+		bfd_dplane_ctx_free(bdc);
+		return true;
+	}
+}
+
+static void bfd_dplane_client_connect(struct thread *t)
+{
+	struct bfd_dplane_ctx *bdc = THREAD_ARG(t);
+	int rv, sock;
+	socklen_t rvlen = sizeof(rv);
+
+	/* Allocate new socket. */
+	sock = socket(bdc->addr.sa.sa_family, SOCK_STREAM, 0);
+	if (sock == -1) {
+		zlog_warn("%s: failed to initialize socket: %s", __func__,
+			  strerror(errno));
+		goto reschedule_connect;
+	}
+
+	/* Set non blocking socket. */
+	set_nonblocking(sock);
+
+	/* Set 'no delay' (disables nagle algorithm) for IPv4/IPv6. */
+	rv = 1;
+	if (bdc->addr.sa.sa_family != AF_UNIX
+	    && setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, &rv, rvlen) == -1)
+		zlog_warn("%s: TCP_NODELAY: %s", __func__, strerror(errno));
+
+	/* Attempt to connect. */
+	rv = connect(sock, &bdc->addr.sa, bdc->addrlen);
+	if (rv == -1 && (errno != EINPROGRESS && errno != EAGAIN)) {
+		zlog_warn("%s: data plane connection failed: %s", __func__,
+			  strerror(errno));
+		goto reschedule_connect;
+	}
+
+	bdc->sock = sock;
+	if (rv == -1) {
+		if (bglobal.debug_dplane)
+			zlog_debug("%s: server connection in progress: %d",
+				   __func__, sock);
+
+		/* If we are not connected yet, ask for write notifications. */
+		bdc->connecting = true;
+		thread_add_write(master, bfd_dplane_write, bdc, bdc->sock,
+				 &bdc->outbufev);
+	} else {
+		if (bglobal.debug_dplane)
+			zlog_debug("%s: server connection: %d", __func__, sock);
+
+		/* Otherwise just start accepting data. */
+		_bfd_dplane_client_bootstrap(bdc);
+	}
+
+reschedule_connect:
+	THREAD_OFF(bdc->inbufev);
+	THREAD_OFF(bdc->outbufev);
+	socket_close(&sock);
+	thread_add_timer(master, bfd_dplane_client_connect, bdc, 3,
+			 &bdc->connectev);
+}
+
+static void bfd_dplane_client_init(const struct sockaddr *sa, socklen_t salen)
+{
+	struct bfd_dplane_ctx *bdc;
+
+	/* Allocate context and copy address for reconnection. */
+	bdc = bfd_dplane_ctx_new(-1);
+	if (salen <= sizeof(bdc->addr)) {
+		memcpy(&bdc->addr, sa, salen);
+		bdc->addrlen = sizeof(bdc->addr);
+	} else {
+		memcpy(&bdc->addr, sa, sizeof(bdc->addr));
+		bdc->addrlen = sizeof(bdc->addr);
+		zlog_warn("%s: server address truncated (from %d to %d)",
+			  __func__, salen, bdc->addrlen);
+	}
+
+	bdc->client = true;
+
+	thread_add_timer(master, bfd_dplane_client_connect, bdc, 0,
+			 &bdc->connectev);
+
+	/* Insert into data plane lists. */
+	TAILQ_INSERT_TAIL(&bglobal.bg_dplaneq, bdc, entry);
+}
+
+/**
+ * Termination phase of the distributed BFD infrastructure: free all allocated
+ * resources.
+ */
+static int bfd_dplane_finish_late(void)
+{
+	struct bfd_dplane_ctx *bdc;
+
+	if (bglobal.debug_dplane)
+		zlog_debug("%s: terminating distributed BFD", __func__);
+
+	/* Free all data plane client contexts. */
+	while ((bdc = TAILQ_FIRST(&bglobal.bg_dplaneq)) != NULL)
+		bfd_dplane_ctx_free(bdc);
+
+	/* Cancel accept thread and close socket. */
+	THREAD_OFF(bglobal.bg_dplane_sockev);
+	close(bglobal.bg_dplane_sock);
+
+	return 0;
+}
+
+/*
+ * Data plane exported functions.
+ */
+void bfd_dplane_init(const struct sockaddr *sa, socklen_t salen, bool client)
+{
+	int sock;
+
+	zlog_info("initializing distributed BFD");
+
+	/* Initialize queue header. */
+	TAILQ_INIT(&bglobal.bg_dplaneq);
+
+	/* Initialize listening socket. */
+	bglobal.bg_dplane_sock = -1;
+
+	/* Observe shutdown events. */
+	hook_register(frr_fini, bfd_dplane_finish_late);
+
+	/* Handle client mode. */
+	if (client) {
+		bfd_dplane_client_init(sa, salen);
+		return;
+	}
+
+	/*
+	 * Data plane socket creation:
+	 * - Set REUSEADDR option for taking over previously open socket.
+	 * - Bind to address requested (maybe IPv4, IPv6, UNIX etc...).
+	 * - Listen on that address for new connections.
+	 * - Ask to be waken up when a new connection comes.
+	 */
+	sock = socket(sa->sa_family, SOCK_STREAM, 0);
+	if (sock == -1) {
+		zlog_warn("%s: failed to initialize socket: %s", __func__,
+			  strerror(errno));
+		return;
+	}
+
+	if (sockopt_reuseaddr(sock) == -1) {
+		zlog_warn("%s: failed to set reuseaddr: %s", __func__,
+			  strerror(errno));
+		close(sock);
+		return;
+	}
+
+	/* Handle UNIX socket: delete previous socket if any. */
+	if (sa->sa_family == AF_UNIX)
+		unlink(((struct sockaddr_un *)sa)->sun_path);
+
+	if (bind(sock, sa, salen) == -1) {
+		zlog_warn("%s: failed to bind socket: %s", __func__,
+			  strerror(errno));
+		close(sock);
+		return;
+	}
+
+	if (listen(sock, SOMAXCONN) == -1) {
+		zlog_warn("%s: failed to put socket on listen: %s", __func__,
+			  strerror(errno));
+		close(sock);
+		return;
+	}
+
+	bglobal.bg_dplane_sock = sock;
+	thread_add_read(master, bfd_dplane_accept, &bglobal, sock,
+			&bglobal.bg_dplane_sockev);
+}
+
+int bfd_dplane_add_session(struct bfd_session *bs)
+{
+	struct bfd_dplane_ctx *bdc;
+
+	/* Select a data plane client to install session. */
+	TAILQ_FOREACH (bdc, &bglobal.bg_dplaneq, entry) {
+		if (_bfd_dplane_add_session(bdc, bs) == 0)
+			return 0;
+	}
+
+	return -1;
+}
+
+int bfd_dplane_update_session(const struct bfd_session *bs)
+{
+	struct bfddp_message msg = {};
+
+	if (bs->bdc == NULL)
+		return 0;
+
+	_bfd_dplane_session_fill(bs, &msg);
+
+	/* Enqueue message to data plane client. */
+	return bfd_dplane_enqueue(bs->bdc, &msg, ntohs(msg.header.length));
+}
+
+int bfd_dplane_delete_session(struct bfd_session *bs)
+{
+	struct bfddp_message msg = {};
+	int rv;
+
+	/* Not using data plane, just return success. */
+	if (bs->bdc == NULL)
+		return 0;
+
+	/* Fill most of the common fields. */
+	_bfd_dplane_session_fill(bs, &msg);
+
+	/* Change the message type. */
+	msg.header.type = ntohs(DP_DELETE_SESSION);
+
+	/* Enqueue message to data plane client. */
+	rv = bfd_dplane_enqueue(bs->bdc, &msg, ntohs(msg.header.length));
+
+	/* Remove association. */
+	bs->bdc = NULL;
+
+	return rv;
+}
+
+/*
+ * Data plane CLI.
+ */
+void bfd_dplane_show_counters(struct vty *vty)
+{
+	struct bfd_dplane_ctx *bdc;
+
+#define SHOW_COUNTER(label, counter, formatter)                                \
+	vty_out(vty, "%28s: %" formatter "\n", (label), (counter))
+
+	vty_out(vty, "%28s\n%28s\n", "Data plane", "==========");
+	TAILQ_FOREACH (bdc, &bglobal.bg_dplaneq, entry) {
+		SHOW_COUNTER("File descriptor", bdc->sock, "d");
+		SHOW_COUNTER("Input bytes", bdc->in_bytes, PRIu64);
+		SHOW_COUNTER("Input bytes peak", bdc->in_bytes_peak, PRIu64);
+		SHOW_COUNTER("Input messages", bdc->in_msgs, PRIu64);
+		SHOW_COUNTER("Input current usage", STREAM_READABLE(bdc->inbuf),
+			     "zu");
+		SHOW_COUNTER("Output bytes", bdc->out_bytes, PRIu64);
+		SHOW_COUNTER("Output bytes peak", bdc->out_bytes_peak, PRIu64);
+		SHOW_COUNTER("Output messages", bdc->out_msgs, PRIu64);
+		SHOW_COUNTER("Output full events", bdc->out_fullev, PRIu64);
+		SHOW_COUNTER("Output current usage",
+			     STREAM_READABLE(bdc->inbuf), "zu");
+		vty_out(vty, "\n");
+	}
+#undef SHOW_COUNTER
+}
+
+int bfd_dplane_update_session_counters(struct bfd_session *bs)
+{
+	uint16_t id;
+	int rv;
+
+	/* If session is not using data plane, then just return success. */
+	if (bs->bdc == NULL)
+		return 0;
+
+	/* Make the request. */
+	id = bfd_dplane_request_counters(bs);
+	if (id == 0) {
+		zlog_debug("%s: counters request failed", __func__);
+		return -1;
+	}
+
+	/* Handle interruptions. */
+	do {
+		rv = bfd_dplane_expect(bs->bdc, id,
+				       _bfd_dplane_update_session_counters, bs);
+	} while (rv == -2);
+
+	return rv;
+}