1 /* Copyright (c) 2013, 2014, 2015 Nicira, Inc.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License. */
18 #include <sys/types.h>
19 #include <arpa/inet.h>
20 #include <netinet/in_systm.h>
21 #include <netinet/ip.h>
22 #include <sys/socket.h>
24 #include "byte-order.h"
25 #include "connectivity.h"
27 #include "dp-packet.h"
29 #include "dynamic-string.h"
37 #include "ovs-thread.h"
38 #include "openvswitch/types.h"
40 #include "poll-loop.h"
45 #include "unaligned.h"
48 #include "openvswitch/vlog.h"
50 VLOG_DEFINE_THIS_MODULE(bfd);
54 * The goal of this module is to replace CFM with something both more flexible
55 * and standards compliant. In service of this goal, the following needs to be
59 * * Implement Demand mode.
60 * * Go through the RFC line by line and verify we comply.
61 * * Test against a hardware implementation. Preferably a popular one.
62 * * Delete BFD packets with nw_ttl != 255 in the datapath to prevent DOS
67 * - Set TOS/PCP on the outer tunnel header when encapped.
69 * - Sending BFD messages should be in its own thread/process.
71 * - Scale testing. How does it operate when there are large number of bfd
72 * sessions? Do we ever have random flaps? What's the CPU utilization?
74 * - Rely on data traffic for liveness by using BFD demand mode.
75 * If we're receiving traffic on a port, we can safely assume it's up (modulo
76 * unidrectional failures). BFD has a demand mode in which it can stay quiet
77 * unless it feels the need to check the status of the port. Using this, we
78 * can implement a strategy in which BFD only sends control messages on dark
81 * - Depending on how one interprets the spec, it appears that a BFD session
82 * can never change bfd.LocalDiag to "No Diagnostic". We should verify that
83 * this is what hardware implementations actually do. Seems like "No
84 * Diagnostic" should be set once a BFD session state goes UP. */
89 FLAG_MULTIPOINT = 1 << 0,
98 STATE_ADMIN_DOWN = 0 << 6,
105 DIAG_NONE = 0, /* No Diagnostic. */
106 DIAG_EXPIRED = 1, /* Control Detection Time Expired. */
107 DIAG_ECHO_FAILED = 2, /* Echo Function Failed. */
108 DIAG_RMT_DOWN = 3, /* Neighbor Signaled Session Down. */
109 DIAG_FWD_RESET = 4, /* Forwarding Plane Reset. */
110 DIAG_PATH_DOWN = 5, /* Path Down. */
111 DIAG_CPATH_DOWN = 6, /* Concatenated Path Down. */
112 DIAG_ADMIN_DOWN = 7, /* Administratively Down. */
113 DIAG_RCPATH_DOWN = 8 /* Reverse Concatenated Path Down. */
116 /* RFC 5880 Section 4.1
118 * 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
119 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
120 * |Vers | Diag |Sta|P|F|C|A|D|M| Detect Mult | Length |
121 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
122 * | My Discriminator |
123 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
124 * | Your Discriminator |
125 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
126 * | Desired Min TX Interval |
127 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
128 * | Required Min RX Interval |
129 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
130 * | Required Min Echo RX Interval |
131 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ */
133 uint8_t vers_diag; /* Version and diagnostic. */
134 uint8_t flags; /* 2bit State field followed by flags. */
135 uint8_t mult; /* Fault detection multiplier. */
136 uint8_t length; /* Length of this BFD message. */
137 ovs_be32 my_disc; /* My discriminator. */
138 ovs_be32 your_disc; /* Your discriminator. */
139 ovs_be32 min_tx; /* Desired minimum tx interval. */
140 ovs_be32 min_rx; /* Required minimum rx interval. */
141 ovs_be32 min_rx_echo; /* Required minimum echo rx interval. */
143 BUILD_ASSERT_DECL(BFD_PACKET_LEN == sizeof(struct msg));
145 #define DIAG_MASK 0x1f
147 #define STATE_MASK 0xC0
148 #define FLAGS_MASK 0x3f
151 struct hmap_node node; /* In 'all_bfds'. */
152 uint32_t disc; /* bfd.LocalDiscr. Key in 'all_bfds' hmap. */
154 char *name; /* Name used for logging. */
156 bool cpath_down; /* Concatenated Path Down. */
157 uint8_t mult; /* bfd.DetectMult. */
159 struct netdev *netdev;
160 uint64_t rx_packets; /* Packets received by 'netdev'. */
162 enum state state; /* bfd.SessionState. */
163 enum state rmt_state; /* bfd.RemoteSessionState. */
165 enum diag diag; /* bfd.LocalDiag. */
166 enum diag rmt_diag; /* Remote diagnostic. */
168 enum flags flags; /* Flags sent on messages. */
169 enum flags rmt_flags; /* Flags last received. */
171 uint32_t rmt_disc; /* bfd.RemoteDiscr. */
173 struct eth_addr local_eth_src; /* Local eth src address. */
174 struct eth_addr local_eth_dst; /* Local eth dst address. */
176 struct eth_addr rmt_eth_dst; /* Remote eth dst address. */
178 ovs_be32 ip_src; /* IPv4 source address. */
179 ovs_be32 ip_dst; /* IPv4 destination address. */
181 uint16_t udp_src; /* UDP source port. */
183 /* All timers in milliseconds. */
184 long long int rmt_min_rx; /* bfd.RemoteMinRxInterval. */
185 long long int rmt_min_tx; /* Remote minimum TX interval. */
187 long long int cfg_min_tx; /* Configured minimum TX rate. */
188 long long int cfg_min_rx; /* Configured required minimum RX rate. */
189 long long int poll_min_tx; /* Min TX negotating in a poll sequence. */
190 long long int poll_min_rx; /* Min RX negotating in a poll sequence. */
191 long long int min_tx; /* bfd.DesiredMinTxInterval. */
192 long long int min_rx; /* bfd.RequiredMinRxInterval. */
194 long long int last_tx; /* Last TX time. */
195 long long int next_tx; /* Next TX time. */
196 long long int detect_time; /* RFC 5880 6.8.4 Detection time. */
198 bool last_forwarding; /* Last calculation of forwarding flag. */
199 int forwarding_override; /* Manual override of 'forwarding' status. */
201 atomic_bool check_tnl_key; /* Verify tunnel key of inbound packets? */
202 struct ovs_refcount ref_cnt;
204 /* When forward_if_rx is true, bfd_forwarding() will return
205 * true as long as there are incoming packets received.
206 * Note, forwarding_override still has higher priority. */
207 bool forwarding_if_rx;
208 long long int forwarding_if_rx_detect_time;
210 /* When 'bfd->forwarding_if_rx' is set, at least one bfd control packet
211 * is required to be received every 100 * bfd->cfg_min_rx. If bfd
212 * control packet is not received within this interval, even if data
213 * packets are received, the bfd->forwarding will still be false. */
214 long long int demand_rx_bfd_time;
216 /* BFD decay related variables. */
217 bool in_decay; /* True when bfd is in decay. */
218 int decay_min_rx; /* min_rx is set to decay_min_rx when */
220 int decay_rx_ctl; /* Count bfd packets received within decay */
221 /* detect interval. */
222 uint64_t decay_rx_packets; /* Packets received by 'netdev'. */
223 long long int decay_detect_time; /* Decay detection time. */
225 uint64_t flap_count; /* Counts bfd forwarding flaps. */
227 /* True when the variables returned by bfd_get_status() are changed
228 * since last check. */
232 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
233 static struct hmap all_bfds__ = HMAP_INITIALIZER(&all_bfds__);
234 static struct hmap *const all_bfds OVS_GUARDED_BY(mutex) = &all_bfds__;
236 static bool bfd_lookup_ip(const char *host_name, struct in_addr *)
238 static bool bfd_forwarding__(struct bfd *) OVS_REQUIRES(mutex);
239 static bool bfd_in_poll(const struct bfd *) OVS_REQUIRES(mutex);
240 static void bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex);
241 static const char *bfd_diag_str(enum diag) OVS_REQUIRES(mutex);
242 static const char *bfd_state_str(enum state) OVS_REQUIRES(mutex);
243 static long long int bfd_min_tx(const struct bfd *) OVS_REQUIRES(mutex);
244 static long long int bfd_tx_interval(const struct bfd *)
246 static long long int bfd_rx_interval(const struct bfd *)
248 static void bfd_set_next_tx(struct bfd *) OVS_REQUIRES(mutex);
249 static void bfd_set_state(struct bfd *, enum state, enum diag)
251 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex);
252 static void bfd_put_details(struct ds *, const struct bfd *)
254 static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQUIRES(mutex);
255 static void bfd_try_decay(struct bfd *) OVS_REQUIRES(mutex);
256 static void bfd_decay_update(struct bfd *) OVS_REQUIRES(mutex);
257 static void bfd_status_changed(struct bfd *) OVS_REQUIRES(mutex);
259 static void bfd_forwarding_if_rx_update(struct bfd *) OVS_REQUIRES(mutex);
260 static void bfd_unixctl_show(struct unixctl_conn *, int argc,
261 const char *argv[], void *aux OVS_UNUSED);
262 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
263 int argc, const char *argv[],
264 void *aux OVS_UNUSED);
265 static void log_msg(enum vlog_level, const struct msg *, const char *message,
266 const struct bfd *) OVS_REQUIRES(mutex);
268 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 20);
270 /* Returns true if the interface on which 'bfd' is running may be used to
271 * forward traffic according to the BFD session state. */
273 bfd_forwarding(struct bfd *bfd) OVS_EXCLUDED(mutex)
277 ovs_mutex_lock(&mutex);
278 ret = bfd_forwarding__(bfd);
279 ovs_mutex_unlock(&mutex);
283 /* When forwarding_if_rx is enabled, if there are packets received,
284 * updates forwarding_if_rx_detect_time. */
286 bfd_account_rx(struct bfd *bfd, const struct dpif_flow_stats *stats)
288 if (stats->n_packets && bfd->forwarding_if_rx) {
289 ovs_mutex_lock(&mutex);
290 bfd_forwarding__(bfd);
291 bfd_forwarding_if_rx_update(bfd);
292 bfd_forwarding__(bfd);
293 ovs_mutex_unlock(&mutex);
297 /* Returns and resets the 'bfd->status_changed'. */
299 bfd_check_status_change(struct bfd *bfd) OVS_EXCLUDED(mutex)
303 ovs_mutex_lock(&mutex);
304 ret = bfd->status_changed;
305 bfd->status_changed = false;
306 ovs_mutex_unlock(&mutex);
311 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
312 * intended for the OVS database. */
314 bfd_get_status(const struct bfd *bfd, struct smap *smap)
317 ovs_mutex_lock(&mutex);
318 smap_add(smap, "forwarding",
319 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
321 smap_add(smap, "state", bfd_state_str(bfd->state));
322 smap_add(smap, "diagnostic", bfd_diag_str(bfd->diag));
323 smap_add_format(smap, "flap_count", "%"PRIu64, bfd->flap_count);
324 smap_add(smap, "remote_state", bfd_state_str(bfd->rmt_state));
325 smap_add(smap, "remote_diagnostic", bfd_diag_str(bfd->rmt_diag));
326 ovs_mutex_unlock(&mutex);
332 unixctl_command_register("bfd/show", "[interface]", 0, 1,
333 bfd_unixctl_show, NULL);
334 unixctl_command_register("bfd/set-forwarding",
335 "[interface] normal|false|true", 1, 2,
336 bfd_unixctl_set_forwarding_override, NULL);
339 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
340 * according to the database configuration contained in 'cfg'. Takes ownership
341 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
342 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
343 * Also returns NULL if cfg is NULL. */
345 bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
346 struct netdev *netdev) OVS_EXCLUDED(mutex)
348 static atomic_count udp_src = ATOMIC_COUNT_INIT(0);
351 long long int min_tx, min_rx;
352 bool need_poll = false;
353 bool cfg_min_rx_changed = false;
354 bool cpath_down, forwarding_if_rx;
355 const char *hwaddr, *ip_src, *ip_dst;
356 struct in_addr in_addr;
359 if (!cfg || !smap_get_bool(cfg, "enable", false)) {
364 ovs_mutex_lock(&mutex);
366 bfd = xzalloc(sizeof *bfd);
367 bfd->name = xstrdup(name);
368 bfd->forwarding_override = -1;
369 bfd->disc = generate_discriminator();
370 hmap_insert(all_bfds, &bfd->node, bfd->disc);
372 bfd->diag = DIAG_NONE;
375 ovs_refcount_init(&bfd->ref_cnt);
376 bfd->netdev = netdev_ref(netdev);
377 bfd->rx_packets = bfd_rx_packets(bfd);
378 bfd->in_decay = false;
381 /* RFC 5881 section 4
382 * The source port MUST be in the range 49152 through 65535. The same
383 * UDP source port number MUST be used for all BFD Control packets
384 * associated with a particular session. The source port number SHOULD
385 * be unique among all BFD sessions on the system. */
386 bfd->udp_src = (atomic_count_inc(&udp_src) % 16384) + 49152;
388 bfd_set_state(bfd, STATE_DOWN, DIAG_NONE);
390 bfd_status_changed(bfd);
393 atomic_store_relaxed(&bfd->check_tnl_key,
394 smap_get_bool(cfg, "check_tnl_key", false));
395 min_tx = smap_get_int(cfg, "min_tx", 100);
396 min_tx = MAX(min_tx, 1);
397 if (bfd->cfg_min_tx != min_tx) {
398 bfd->cfg_min_tx = min_tx;
399 if (bfd->state != STATE_UP
400 || (!bfd_in_poll(bfd) && bfd->cfg_min_tx < bfd->min_tx)) {
401 bfd->min_tx = bfd->cfg_min_tx;
406 min_rx = smap_get_int(cfg, "min_rx", 1000);
407 min_rx = MAX(min_rx, 1);
408 if (bfd->cfg_min_rx != min_rx) {
409 bfd->cfg_min_rx = min_rx;
410 if (bfd->state != STATE_UP
411 || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
412 bfd->min_rx = bfd->cfg_min_rx;
414 cfg_min_rx_changed = true;
418 decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
419 if (bfd->decay_min_rx != decay_min_rx || cfg_min_rx_changed) {
420 if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
421 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
422 bfd->name, bfd->cfg_min_rx);
423 bfd->decay_min_rx = 0;
425 bfd->decay_min_rx = decay_min_rx;
428 bfd->in_decay = false;
429 bfd_decay_update(bfd);
433 cpath_down = smap_get_bool(cfg, "cpath_down", false);
434 if (bfd->cpath_down != cpath_down) {
435 bfd->cpath_down = cpath_down;
436 bfd_set_state(bfd, bfd->state, DIAG_NONE);
440 hwaddr = smap_get(cfg, "bfd_local_src_mac");
441 if (hwaddr && eth_addr_from_string(hwaddr, &ea)) {
442 bfd->local_eth_src = ea;
444 bfd->local_eth_src = eth_addr_zero;
447 hwaddr = smap_get(cfg, "bfd_local_dst_mac");
448 if (hwaddr && eth_addr_from_string(hwaddr, &ea)) {
449 bfd->local_eth_dst = ea;
451 bfd->local_eth_dst = eth_addr_zero;
454 hwaddr = smap_get(cfg, "bfd_remote_dst_mac");
455 if (hwaddr && eth_addr_from_string(hwaddr, &ea)) {
456 bfd->rmt_eth_dst = ea;
458 bfd->rmt_eth_dst = eth_addr_zero;
461 ip_src = smap_get(cfg, "bfd_src_ip");
462 if (ip_src && bfd_lookup_ip(ip_src, &in_addr)) {
463 memcpy(&bfd->ip_src, &in_addr, sizeof in_addr);
465 bfd->ip_src = htonl(0xA9FE0101); /* 169.254.1.1. */
468 ip_dst = smap_get(cfg, "bfd_dst_ip");
469 if (ip_dst && bfd_lookup_ip(ip_dst, &in_addr)) {
470 memcpy(&bfd->ip_dst, &in_addr, sizeof in_addr);
472 bfd->ip_dst = htonl(0xA9FE0100); /* 169.254.1.0. */
475 forwarding_if_rx = smap_get_bool(cfg, "forwarding_if_rx", false);
476 if (bfd->forwarding_if_rx != forwarding_if_rx) {
477 bfd->forwarding_if_rx = forwarding_if_rx;
478 if (bfd->state == STATE_UP && bfd->forwarding_if_rx) {
479 bfd_forwarding_if_rx_update(bfd);
481 bfd->forwarding_if_rx_detect_time = 0;
488 ovs_mutex_unlock(&mutex);
493 bfd_ref(const struct bfd *bfd_)
495 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
497 ovs_refcount_ref(&bfd->ref_cnt);
503 bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
505 if (bfd && ovs_refcount_unref_relaxed(&bfd->ref_cnt) == 1) {
506 ovs_mutex_lock(&mutex);
507 bfd_status_changed(bfd);
508 hmap_remove(all_bfds, &bfd->node);
509 netdev_close(bfd->netdev);
512 ovs_mutex_unlock(&mutex);
517 bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
519 long long int wake_time = bfd_wake_time(bfd);
520 poll_timer_wait_until(wake_time);
524 /* Returns the next wake up time. */
526 bfd_wake_time(const struct bfd *bfd) OVS_EXCLUDED(mutex)
528 long long int retval;
534 ovs_mutex_lock(&mutex);
535 if (bfd->flags & FLAG_FINAL) {
538 retval = bfd->next_tx;
539 if (bfd->state > STATE_DOWN) {
540 retval = MIN(bfd->detect_time, retval);
543 ovs_mutex_unlock(&mutex);
548 bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
553 ovs_mutex_lock(&mutex);
555 old_in_decay = bfd->in_decay;
557 if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
558 bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
560 bfd_forwarding__(bfd);
562 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
563 * configured, and decay_detect_time is reached. */
564 if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
565 && now >= bfd->decay_detect_time) {
569 if (bfd->min_tx != bfd->cfg_min_tx
570 || (bfd->min_rx != bfd->cfg_min_rx && bfd->min_rx != bfd->decay_min_rx)
571 || bfd->in_decay != old_in_decay) {
574 ovs_mutex_unlock(&mutex);
578 bfd_should_send_packet(const struct bfd *bfd) OVS_EXCLUDED(mutex)
581 ovs_mutex_lock(&mutex);
582 ret = bfd->flags & FLAG_FINAL || time_msec() >= bfd->next_tx;
583 ovs_mutex_unlock(&mutex);
588 bfd_put_packet(struct bfd *bfd, struct dp_packet *p,
589 const struct eth_addr eth_src) OVS_EXCLUDED(mutex)
591 long long int min_tx, min_rx;
592 struct udp_header *udp;
593 struct eth_header *eth;
594 struct ip_header *ip;
597 ovs_mutex_lock(&mutex);
599 long long int delay = time_msec() - bfd->next_tx;
600 long long int interval = bfd_tx_interval(bfd);
601 if (delay > interval * 3 / 2) {
602 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
603 " control message", bfd->name, delay, interval);
607 /* RFC 5880 Section 6.5
608 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
610 ovs_assert(!(bfd->flags & FLAG_POLL) || !(bfd->flags & FLAG_FINAL));
612 dp_packet_reserve(p, 2); /* Properly align after the ethernet header. */
613 eth = dp_packet_put_uninit(p, sizeof *eth);
614 eth->eth_src = eth_addr_is_zero(bfd->local_eth_src)
615 ? eth_src : bfd->local_eth_src;
616 eth->eth_dst = eth_addr_is_zero(bfd->local_eth_dst)
617 ? eth_addr_bfd : bfd->local_eth_dst;
618 eth->eth_type = htons(ETH_TYPE_IP);
620 ip = dp_packet_put_zeros(p, sizeof *ip);
621 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
622 ip->ip_tot_len = htons(sizeof *ip + sizeof *udp + sizeof *msg);
624 ip->ip_tos = IPTOS_LOWDELAY | IPTOS_THROUGHPUT;
625 ip->ip_proto = IPPROTO_UDP;
626 put_16aligned_be32(&ip->ip_src, bfd->ip_src);
627 put_16aligned_be32(&ip->ip_dst, bfd->ip_dst);
628 ip->ip_csum = csum(ip, sizeof *ip);
630 udp = dp_packet_put_zeros(p, sizeof *udp);
631 udp->udp_src = htons(bfd->udp_src);
632 udp->udp_dst = htons(BFD_DEST_PORT);
633 udp->udp_len = htons(sizeof *udp + sizeof *msg);
635 msg = dp_packet_put_uninit(p, sizeof *msg);
636 msg->vers_diag = (BFD_VERSION << 5) | bfd->diag;
637 msg->flags = (bfd->state & STATE_MASK) | bfd->flags;
639 msg->mult = bfd->mult;
640 msg->length = BFD_PACKET_LEN;
641 msg->my_disc = htonl(bfd->disc);
642 msg->your_disc = htonl(bfd->rmt_disc);
643 msg->min_rx_echo = htonl(0);
645 if (bfd_in_poll(bfd)) {
646 min_tx = bfd->poll_min_tx;
647 min_rx = bfd->poll_min_rx;
649 min_tx = bfd_min_tx(bfd);
650 min_rx = bfd->min_rx;
653 msg->min_tx = htonl(min_tx * 1000);
654 msg->min_rx = htonl(min_rx * 1000);
656 bfd->flags &= ~FLAG_FINAL;
658 log_msg(VLL_DBG, msg, "Sending BFD Message", bfd);
660 bfd->last_tx = time_msec();
661 bfd_set_next_tx(bfd);
662 ovs_mutex_unlock(&mutex);
666 bfd_should_process_flow(const struct bfd *bfd_, const struct flow *flow,
667 struct flow_wildcards *wc)
669 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
671 if (!eth_addr_is_zero(bfd->rmt_eth_dst)) {
672 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
674 if (!eth_addr_equals(bfd->rmt_eth_dst, flow->dl_dst)) {
679 if (flow->dl_type == htons(ETH_TYPE_IP)) {
680 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
681 if (flow->nw_proto == IPPROTO_UDP) {
682 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
683 if (flow->tp_dst == htons(BFD_DEST_PORT)) {
686 atomic_read_relaxed(&bfd->check_tnl_key, &check_tnl_key);
688 memset(&wc->masks.tunnel.tun_id, 0xff,
689 sizeof wc->masks.tunnel.tun_id);
690 return flow->tunnel.tun_id == htonll(0);
700 bfd_process_packet(struct bfd *bfd, const struct flow *flow,
701 const struct dp_packet *p) OVS_EXCLUDED(mutex)
703 uint32_t rmt_min_rx, pkt_your_disc;
704 enum state rmt_state;
708 const uint8_t *l7 = dp_packet_get_udp_payload(p);
711 return; /* No UDP payload. */
714 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
716 ovs_mutex_lock(&mutex);
717 /* Increments the decay rx counter. */
720 bfd_forwarding__(bfd);
722 if (flow->nw_ttl != 255) {
723 /* XXX Should drop in the kernel to prevent DOS. */
727 msg = dp_packet_at(p, l7 - (uint8_t *)dp_packet_data(p), BFD_PACKET_LEN);
729 VLOG_INFO_RL(&rl, "%s: Received too-short BFD control message (only "
730 "%"PRIdPTR" bytes long, at least %d required).",
731 bfd->name, (uint8_t *) dp_packet_tail(p) - l7,
736 /* RFC 5880 Section 6.8.6
737 * If the Length field is greater than the payload of the encapsulating
738 * protocol, the packet MUST be discarded.
740 * Note that we make this check implicitly. Above we use dp_packet_at() to
741 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
742 * the encapsulating protocol. Below we require msg->length to be exactly
743 * BFD_PACKET_LEN bytes. */
745 flags = msg->flags & FLAGS_MASK;
746 rmt_state = msg->flags & STATE_MASK;
747 version = msg->vers_diag >> VERS_SHIFT;
749 log_msg(VLL_DBG, msg, "Received BFD control message", bfd);
751 if (version != BFD_VERSION) {
752 log_msg(VLL_WARN, msg, "Incorrect version", bfd);
756 /* Technically this should happen after the length check. We don't support
757 * authentication however, so it's simpler to do the check first. */
758 if (flags & FLAG_AUTH) {
759 log_msg(VLL_WARN, msg, "Authenticated control message with"
760 " authentication disabled", bfd);
764 if (msg->length != BFD_PACKET_LEN) {
765 log_msg(VLL_WARN, msg, "Unexpected length", bfd);
766 if (msg->length < BFD_PACKET_LEN) {
772 log_msg(VLL_WARN, msg, "Zero multiplier", bfd);
776 if (flags & FLAG_MULTIPOINT) {
777 log_msg(VLL_WARN, msg, "Unsupported multipoint flag", bfd);
782 log_msg(VLL_WARN, msg, "NULL my_disc", bfd);
786 pkt_your_disc = ntohl(msg->your_disc);
788 /* Technically, we should use the your discriminator field to figure
789 * out which 'struct bfd' this packet is destined towards. That way a
790 * bfd session could migrate from one interface to another
791 * transparently. This doesn't fit in with the OVS structure very
792 * well, so in this respect, we are not compliant. */
793 if (pkt_your_disc != bfd->disc) {
794 log_msg(VLL_WARN, msg, "Incorrect your_disc", bfd);
797 } else if (rmt_state > STATE_DOWN) {
798 log_msg(VLL_WARN, msg, "Null your_disc", bfd);
802 if (bfd->rmt_state != rmt_state) {
803 bfd_status_changed(bfd);
806 bfd->rmt_disc = ntohl(msg->my_disc);
807 bfd->rmt_state = rmt_state;
808 bfd->rmt_flags = flags;
809 bfd->rmt_diag = msg->vers_diag & DIAG_MASK;
811 if (flags & FLAG_FINAL && bfd_in_poll(bfd)) {
812 bfd->min_tx = bfd->poll_min_tx;
813 bfd->min_rx = bfd->poll_min_rx;
814 bfd->flags &= ~FLAG_POLL;
815 log_msg(VLL_INFO, msg, "Poll sequence terminated", bfd);
818 if (flags & FLAG_POLL) {
819 /* RFC 5880 Section 6.5
820 * When the other system receives a Poll, it immediately transmits a
821 * BFD Control packet with the Final (F) bit set, independent of any
822 * periodic BFD Control packets it may be sending
823 * (see section 6.8.7). */
824 bfd->flags &= ~FLAG_POLL;
825 bfd->flags |= FLAG_FINAL;
828 rmt_min_rx = MAX(ntohl(msg->min_rx) / 1000, 1);
829 if (bfd->rmt_min_rx != rmt_min_rx) {
830 bfd->rmt_min_rx = rmt_min_rx;
832 bfd_set_next_tx(bfd);
834 log_msg(VLL_INFO, msg, "New remote min_rx", bfd);
837 bfd->rmt_min_tx = MAX(ntohl(msg->min_tx) / 1000, 1);
838 bfd->detect_time = bfd_rx_interval(bfd) * bfd->mult + time_msec();
840 if (bfd->state == STATE_ADMIN_DOWN) {
841 VLOG_DBG_RL(&rl, "Administratively down, dropping control message.");
845 if (rmt_state == STATE_ADMIN_DOWN) {
846 if (bfd->state != STATE_DOWN) {
847 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
850 switch (bfd->state) {
852 if (rmt_state == STATE_DOWN) {
853 bfd_set_state(bfd, STATE_INIT, bfd->diag);
854 } else if (rmt_state == STATE_INIT) {
855 bfd_set_state(bfd, STATE_UP, bfd->diag);
859 if (rmt_state > STATE_DOWN) {
860 bfd_set_state(bfd, STATE_UP, bfd->diag);
864 if (rmt_state <= STATE_DOWN) {
865 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
866 log_msg(VLL_INFO, msg, "Remote signaled STATE_DOWN", bfd);
869 case STATE_ADMIN_DOWN:
874 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
876 if (bfd->forwarding_if_rx) {
877 bfd->demand_rx_bfd_time = time_msec() + 100 * bfd->cfg_min_rx;
881 bfd_forwarding__(bfd);
882 ovs_mutex_unlock(&mutex);
885 /* Must be called when the netdev owned by 'bfd' should change. */
887 bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
890 ovs_mutex_lock(&mutex);
891 if (bfd->netdev != netdev) {
892 netdev_close(bfd->netdev);
893 bfd->netdev = netdev_ref(netdev);
894 if (bfd->decay_min_rx && bfd->state == STATE_UP) {
895 bfd_decay_update(bfd);
897 if (bfd->forwarding_if_rx && bfd->state == STATE_UP) {
898 bfd_forwarding_if_rx_update(bfd);
900 bfd->rx_packets = bfd_rx_packets(bfd);
902 ovs_mutex_unlock(&mutex);
906 /* Updates the forwarding flag. If override is not configured and
907 * the forwarding flag value changes, increments the flap count.
909 * Note this function may be called multiple times in a function
910 * (e.g. bfd_account_rx) before and after the bfd state or status
911 * change. This is to capture any forwarding flag flap. */
913 bfd_forwarding__(struct bfd *bfd) OVS_REQUIRES(mutex)
915 long long int now = time_msec();
916 bool forwarding_if_rx;
917 bool last_forwarding = bfd->last_forwarding;
919 if (bfd->forwarding_override != -1) {
920 return bfd->forwarding_override == 1;
923 forwarding_if_rx = bfd->forwarding_if_rx
924 && bfd->forwarding_if_rx_detect_time > now
925 && bfd->demand_rx_bfd_time > now;
927 bfd->last_forwarding = (bfd->state == STATE_UP || forwarding_if_rx)
928 && bfd->rmt_diag != DIAG_PATH_DOWN
929 && bfd->rmt_diag != DIAG_CPATH_DOWN
930 && bfd->rmt_diag != DIAG_RCPATH_DOWN;
931 if (bfd->last_forwarding != last_forwarding) {
933 bfd_status_changed(bfd);
935 return bfd->last_forwarding;
940 bfd_lookup_ip(const char *host_name, struct in_addr *addr)
942 if (!inet_pton(AF_INET, host_name, addr)) {
943 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
950 bfd_in_poll(const struct bfd *bfd) OVS_REQUIRES(mutex)
952 return (bfd->flags & FLAG_POLL) != 0;
956 bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex)
958 if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
959 && !(bfd->flags & FLAG_FINAL)) {
960 bfd->poll_min_tx = bfd->cfg_min_tx;
961 bfd->poll_min_rx = bfd->in_decay ? bfd->decay_min_rx : bfd->cfg_min_rx;
962 bfd->flags |= FLAG_POLL;
964 VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
969 bfd_min_tx(const struct bfd *bfd) OVS_REQUIRES(mutex)
971 /* RFC 5880 Section 6.8.3
972 * When bfd.SessionState is not Up, the system MUST set
973 * bfd.DesiredMinTxInterval to a value of not less than one second
974 * (1,000,000 microseconds). This is intended to ensure that the
975 * bandwidth consumed by BFD sessions that are not Up is negligible,
976 * particularly in the case where a neighbor may not be running BFD. */
977 return (bfd->state == STATE_UP ? bfd->min_tx : MAX(bfd->min_tx, 1000));
981 bfd_tx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
983 long long int interval = bfd_min_tx(bfd);
984 return MAX(interval, bfd->rmt_min_rx);
988 bfd_rx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
990 return MAX(bfd->min_rx, bfd->rmt_min_tx);
994 bfd_set_next_tx(struct bfd *bfd) OVS_REQUIRES(mutex)
996 long long int interval = bfd_tx_interval(bfd);
997 interval -= interval * random_range(26) / 100;
998 bfd->next_tx = bfd->last_tx + interval;
1002 bfd_flag_str(enum flags flags)
1004 struct ds ds = DS_EMPTY_INITIALIZER;
1005 static char flag_str[128];
1011 if (flags & FLAG_MULTIPOINT) {
1012 ds_put_cstr(&ds, "multipoint ");
1015 if (flags & FLAG_DEMAND) {
1016 ds_put_cstr(&ds, "demand ");
1019 if (flags & FLAG_AUTH) {
1020 ds_put_cstr(&ds, "auth ");
1023 if (flags & FLAG_CTL) {
1024 ds_put_cstr(&ds, "ctl ");
1027 if (flags & FLAG_FINAL) {
1028 ds_put_cstr(&ds, "final ");
1031 if (flags & FLAG_POLL) {
1032 ds_put_cstr(&ds, "poll ");
1035 /* Do not copy the trailing whitespace. */
1037 ovs_strlcpy(flag_str, ds_cstr(&ds), sizeof flag_str);
1043 bfd_state_str(enum state state)
1046 case STATE_ADMIN_DOWN: return "admin_down";
1047 case STATE_DOWN: return "down";
1048 case STATE_INIT: return "init";
1049 case STATE_UP: return "up";
1050 default: return "invalid";
1055 bfd_diag_str(enum diag diag) {
1057 case DIAG_NONE: return "No Diagnostic";
1058 case DIAG_EXPIRED: return "Control Detection Time Expired";
1059 case DIAG_ECHO_FAILED: return "Echo Function Failed";
1060 case DIAG_RMT_DOWN: return "Neighbor Signaled Session Down";
1061 case DIAG_FWD_RESET: return "Forwarding Plane Reset";
1062 case DIAG_PATH_DOWN: return "Path Down";
1063 case DIAG_CPATH_DOWN: return "Concatenated Path Down";
1064 case DIAG_ADMIN_DOWN: return "Administratively Down";
1065 case DIAG_RCPATH_DOWN: return "Reverse Concatenated Path Down";
1066 default: return "Invalid Diagnostic";
1071 log_msg(enum vlog_level level, const struct msg *p, const char *message,
1072 const struct bfd *bfd) OVS_REQUIRES(mutex)
1074 struct ds ds = DS_EMPTY_INITIALIZER;
1076 if (vlog_should_drop(THIS_MODULE, level, &rl)) {
1082 "\n\tvers:%"PRIu8" diag:\"%s\" state:%s mult:%"PRIu8
1085 "\n\tmy_disc:0x%"PRIx32" your_disc:0x%"PRIx32
1086 "\n\tmin_tx:%"PRIu32"us (%"PRIu32"ms)"
1087 "\n\tmin_rx:%"PRIu32"us (%"PRIu32"ms)"
1088 "\n\tmin_rx_echo:%"PRIu32"us (%"PRIu32"ms)",
1089 bfd->name, message, p->vers_diag >> VERS_SHIFT,
1090 bfd_diag_str(p->vers_diag & DIAG_MASK),
1091 bfd_state_str(p->flags & STATE_MASK),
1092 p->mult, p->length, bfd_flag_str(p->flags & FLAGS_MASK),
1093 ntohl(p->my_disc), ntohl(p->your_disc),
1094 ntohl(p->min_tx), ntohl(p->min_tx) / 1000,
1095 ntohl(p->min_rx), ntohl(p->min_rx) / 1000,
1096 ntohl(p->min_rx_echo), ntohl(p->min_rx_echo) / 1000);
1097 bfd_put_details(&ds, bfd);
1098 VLOG(level, "%s", ds_cstr(&ds));
1103 bfd_set_state(struct bfd *bfd, enum state state, enum diag diag)
1106 if (bfd->cpath_down) {
1107 diag = DIAG_CPATH_DOWN;
1110 if (bfd->state != state || bfd->diag != diag) {
1111 if (!VLOG_DROP_INFO(&rl)) {
1112 struct ds ds = DS_EMPTY_INITIALIZER;
1114 ds_put_format(&ds, "%s: BFD state change: %s->%s"
1115 " \"%s\"->\"%s\".\n",
1116 bfd->name, bfd_state_str(bfd->state),
1117 bfd_state_str(state), bfd_diag_str(bfd->diag),
1118 bfd_diag_str(diag));
1119 bfd_put_details(&ds, bfd);
1120 VLOG_INFO("%s", ds_cstr(&ds));
1127 if (bfd->state <= STATE_DOWN) {
1128 bfd->rmt_state = STATE_DOWN;
1129 bfd->rmt_diag = DIAG_NONE;
1130 bfd->rmt_min_rx = 1;
1133 bfd->rmt_min_tx = 0;
1134 /* Resets the min_rx if in_decay. */
1135 if (bfd->in_decay) {
1136 bfd->min_rx = bfd->cfg_min_rx;
1137 bfd->in_decay = false;
1140 /* Resets the decay when state changes to STATE_UP
1141 * and decay_min_rx is configured. */
1142 if (bfd->state == STATE_UP && bfd->decay_min_rx) {
1143 bfd_decay_update(bfd);
1146 bfd_status_changed(bfd);
1151 bfd_rx_packets(const struct bfd *bfd) OVS_REQUIRES(mutex)
1153 struct netdev_stats stats;
1155 if (!netdev_get_stats(bfd->netdev, &stats)) {
1156 return stats.rx_packets;
1162 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
1163 * the 'expect' value. */
1165 bfd_try_decay(struct bfd *bfd) OVS_REQUIRES(mutex)
1167 int64_t diff, expect;
1169 /* The 'diff' is the difference between current interface rx_packets
1170 * stats and last-time check. The 'expect' is the recorded number of
1171 * bfd control packets received within an approximately decay_min_rx
1172 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
1174 * Since the update of rx_packets stats at interface happens
1175 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1176 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1177 * more wiggle room. */
1178 diff = bfd_rx_packets(bfd) - bfd->decay_rx_packets;
1179 expect = 2 * MAX(bfd->decay_rx_ctl, 1);
1180 bfd->in_decay = diff <= expect ? true : false;
1181 bfd_decay_update(bfd);
1184 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1186 bfd_decay_update(struct bfd * bfd) OVS_REQUIRES(mutex)
1188 bfd->decay_rx_packets = bfd_rx_packets(bfd);
1189 bfd->decay_rx_ctl = 0;
1190 bfd->decay_detect_time = MAX(bfd->decay_min_rx, 2000) + time_msec();
1193 /* Records the status change and changes the global connectivity seq. */
1195 bfd_status_changed(struct bfd *bfd) OVS_REQUIRES(mutex)
1197 seq_change(connectivity_seq_get());
1198 bfd->status_changed = true;
1202 bfd_forwarding_if_rx_update(struct bfd *bfd) OVS_REQUIRES(mutex)
1204 int64_t incr = bfd_rx_interval(bfd) * bfd->mult;
1205 bfd->forwarding_if_rx_detect_time = MAX(incr, 2000) + time_msec();
1209 generate_discriminator(void)
1213 /* RFC 5880 Section 6.8.1
1214 * It SHOULD be set to a random (but still unique) value to improve
1215 * security. The value is otherwise outside the scope of this
1221 /* 'disc' is by definition random, so there's no reason to waste time
1223 disc = random_uint32();
1224 HMAP_FOR_EACH_IN_BUCKET (bfd, node, disc, all_bfds) {
1225 if (bfd->disc == disc) {
1236 bfd_find_by_name(const char *name) OVS_REQUIRES(mutex)
1240 HMAP_FOR_EACH (bfd, node, all_bfds) {
1241 if (!strcmp(bfd->name, name)) {
1249 bfd_put_details(struct ds *ds, const struct bfd *bfd) OVS_REQUIRES(mutex)
1251 ds_put_format(ds, "\tForwarding: %s\n",
1252 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
1253 ? "true" : "false");
1254 ds_put_format(ds, "\tDetect Multiplier: %d\n", bfd->mult);
1255 ds_put_format(ds, "\tConcatenated Path Down: %s\n",
1256 bfd->cpath_down ? "true" : "false");
1257 ds_put_format(ds, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd));
1258 ds_put_format(ds, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd));
1259 ds_put_format(ds, "\tDetect Time: now %+lldms\n",
1260 time_msec() - bfd->detect_time);
1261 ds_put_format(ds, "\tNext TX Time: now %+lldms\n",
1262 time_msec() - bfd->next_tx);
1263 ds_put_format(ds, "\tLast TX Time: now %+lldms\n",
1264 time_msec() - bfd->last_tx);
1266 ds_put_cstr(ds, "\n");
1268 ds_put_format(ds, "\tLocal Flags: %s\n", bfd_flag_str(bfd->flags));
1269 ds_put_format(ds, "\tLocal Session State: %s\n",
1270 bfd_state_str(bfd->state));
1271 ds_put_format(ds, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd->diag));
1272 ds_put_format(ds, "\tLocal Discriminator: 0x%"PRIx32"\n", bfd->disc);
1273 ds_put_format(ds, "\tLocal Minimum TX Interval: %lldms\n",
1275 ds_put_format(ds, "\tLocal Minimum RX Interval: %lldms\n", bfd->min_rx);
1277 ds_put_cstr(ds, "\n");
1279 ds_put_format(ds, "\tRemote Flags: %s\n", bfd_flag_str(bfd->rmt_flags));
1280 ds_put_format(ds, "\tRemote Session State: %s\n",
1281 bfd_state_str(bfd->rmt_state));
1282 ds_put_format(ds, "\tRemote Diagnostic: %s\n",
1283 bfd_diag_str(bfd->rmt_diag));
1284 ds_put_format(ds, "\tRemote Discriminator: 0x%"PRIx32"\n", bfd->rmt_disc);
1285 ds_put_format(ds, "\tRemote Minimum TX Interval: %lldms\n",
1287 ds_put_format(ds, "\tRemote Minimum RX Interval: %lldms\n",
1292 bfd_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1293 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1295 struct ds ds = DS_EMPTY_INITIALIZER;
1298 ovs_mutex_lock(&mutex);
1300 bfd = bfd_find_by_name(argv[1]);
1302 unixctl_command_reply_error(conn, "no such bfd object");
1305 bfd_put_details(&ds, bfd);
1307 HMAP_FOR_EACH (bfd, node, all_bfds) {
1308 ds_put_format(&ds, "---- %s ----\n", bfd->name);
1309 bfd_put_details(&ds, bfd);
1312 unixctl_command_reply(conn, ds_cstr(&ds));
1316 ovs_mutex_unlock(&mutex);
1321 bfd_unixctl_set_forwarding_override(struct unixctl_conn *conn, int argc,
1322 const char *argv[], void *aux OVS_UNUSED)
1325 const char *forward_str = argv[argc - 1];
1326 int forwarding_override;
1329 ovs_mutex_lock(&mutex);
1330 if (!strcasecmp("true", forward_str)) {
1331 forwarding_override = 1;
1332 } else if (!strcasecmp("false", forward_str)) {
1333 forwarding_override = 0;
1334 } else if (!strcasecmp("normal", forward_str)) {
1335 forwarding_override = -1;
1337 unixctl_command_reply_error(conn, "unknown fault string");
1342 bfd = bfd_find_by_name(argv[1]);
1344 unixctl_command_reply_error(conn, "no such BFD object");
1347 bfd->forwarding_override = forwarding_override;
1348 bfd_status_changed(bfd);
1350 HMAP_FOR_EACH (bfd, node, all_bfds) {
1351 bfd->forwarding_override = forwarding_override;
1352 bfd_status_changed(bfd);
1356 unixctl_command_reply(conn, "OK");
1359 ovs_mutex_unlock(&mutex);