1 /* Copyright (c) 2013 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>
23 #include "byte-order.h"
26 #include "dynamic-string.h"
35 #include "ovs-thread.h"
36 #include "openvswitch/types.h"
38 #include "poll-loop.h"
46 VLOG_DEFINE_THIS_MODULE(bfd);
50 * The goal of this module is to replace CFM with something both more flexible
51 * and standards compliant. In service of this goal, the following needs to be
55 * * Implement Demand mode.
56 * * Go through the RFC line by line and verify we comply.
57 * * Test against a hardware implementation. Preferably a popular one.
58 * * Delete BFD packets with nw_ttl != 255 in the datapath to prevent DOS
63 * - Set TOS/PCP on the outer tunnel header when encapped.
65 * - Sending BFD messages should be in its own thread/process.
67 * - Scale testing. How does it operate when there are large number of bfd
68 * sessions? Do we ever have random flaps? What's the CPU utilization?
70 * - Rely on data traffic for liveness by using BFD demand mode.
71 * If we're receiving traffic on a port, we can safely assume it's up (modulo
72 * unidrectional failures). BFD has a demand mode in which it can stay quiet
73 * unless it feels the need to check the status of the port. Using this, we
74 * can implement a strategy in which BFD only sends control messages on dark
77 * - Depending on how one interprets the spec, it appears that a BFD session
78 * can never change bfd.LocalDiag to "No Diagnostic". We should verify that
79 * this is what hardware implementations actually do. Seems like "No
80 * Diagnostic" should be set once a BFD session state goes UP. */
85 FLAG_MULTIPOINT = 1 << 0,
94 STATE_ADMIN_DOWN = 0 << 6,
101 DIAG_NONE = 0, /* No Diagnostic. */
102 DIAG_EXPIRED = 1, /* Control Detection Time Expired. */
103 DIAG_ECHO_FAILED = 2, /* Echo Function Failed. */
104 DIAG_RMT_DOWN = 3, /* Neighbor Signaled Session Down. */
105 DIAG_FWD_RESET = 4, /* Forwarding Plane Reset. */
106 DIAG_PATH_DOWN = 5, /* Path Down. */
107 DIAG_CPATH_DOWN = 6, /* Concatenated Path Down. */
108 DIAG_ADMIN_DOWN = 7, /* Administratively Down. */
109 DIAG_RCPATH_DOWN = 8 /* Reverse Concatenated Path Down. */
112 /* RFC 5880 Section 4.1
114 * 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
115 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
116 * |Vers | Diag |Sta|P|F|C|A|D|M| Detect Mult | Length |
117 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
118 * | My Discriminator |
119 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
120 * | Your Discriminator |
121 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
122 * | Desired Min TX Interval |
123 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
124 * | Required Min RX Interval |
125 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
126 * | Required Min Echo RX Interval |
127 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ */
129 uint8_t vers_diag; /* Version and diagnostic. */
130 uint8_t flags; /* 2bit State field followed by flags. */
131 uint8_t mult; /* Fault detection multiplier. */
132 uint8_t length; /* Length of this BFD message. */
133 ovs_be32 my_disc; /* My discriminator. */
134 ovs_be32 your_disc; /* Your discriminator. */
135 ovs_be32 min_tx; /* Desired minimum tx interval. */
136 ovs_be32 min_rx; /* Required minimum rx interval. */
137 ovs_be32 min_rx_echo; /* Required minimum echo rx interval. */
139 BUILD_ASSERT_DECL(BFD_PACKET_LEN == sizeof(struct msg));
141 #define DIAG_MASK 0x1f
143 #define STATE_MASK 0xC0
144 #define FLAGS_MASK 0x3f
147 struct hmap_node node; /* In 'all_bfds'. */
148 uint32_t disc; /* bfd.LocalDiscr. Key in 'all_bfds' hmap. */
150 char *name; /* Name used for logging. */
152 bool cpath_down; /* Concatenated Path Down. */
153 uint8_t mult; /* bfd.DetectMult. */
155 struct netdev *netdev;
156 uint64_t rx_packets; /* Packets received by 'netdev'. */
158 enum state state; /* bfd.SessionState. */
159 enum state rmt_state; /* bfd.RemoteSessionState. */
161 enum diag diag; /* bfd.LocalDiag. */
162 enum diag rmt_diag; /* Remote diagnostic. */
164 enum flags flags; /* Flags sent on messages. */
165 enum flags rmt_flags; /* Flags last received. */
167 uint32_t rmt_disc; /* bfd.RemoteDiscr. */
169 uint8_t eth_dst[ETH_ADDR_LEN];/* Ethernet destination address. */
170 bool eth_dst_set; /* 'eth_dst' set through database. */
172 uint16_t udp_src; /* UDP source port. */
174 /* All timers in milliseconds. */
175 long long int rmt_min_rx; /* bfd.RemoteMinRxInterval. */
176 long long int rmt_min_tx; /* Remote minimum TX interval. */
178 long long int cfg_min_tx; /* Configured minimum TX rate. */
179 long long int cfg_min_rx; /* Configured required minimum RX rate. */
180 long long int poll_min_tx; /* Min TX negotating in a poll sequence. */
181 long long int poll_min_rx; /* Min RX negotating in a poll sequence. */
182 long long int min_tx; /* bfd.DesiredMinTxInterval. */
183 long long int min_rx; /* bfd.RequiredMinRxInterval. */
185 long long int last_tx; /* Last TX time. */
186 long long int next_tx; /* Next TX time. */
187 long long int detect_time; /* RFC 5880 6.8.4 Detection time. */
189 int forwarding_override; /* Manual override of 'forwarding' status. */
191 atomic_bool check_tnl_key; /* Verify tunnel key of inbound packets? */
194 /* When forward_if_rx is true, bfd_forwarding() will return
195 * true as long as there are incoming packets received.
196 * Note, forwarding_override still has higher priority. */
197 bool forwarding_if_rx;
198 long long int forwarding_if_rx_detect_time;
200 /* BFD decay related variables. */
201 bool in_decay; /* True when bfd is in decay. */
202 int decay_min_rx; /* min_rx is set to decay_min_rx when */
204 int decay_rx_ctl; /* Count bfd packets received within decay */
205 /* detect interval. */
206 uint64_t decay_rx_packets; /* Packets received by 'netdev'. */
207 long long int decay_detect_time; /* Decay detection time. */
210 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
211 static struct hmap all_bfds__ = HMAP_INITIALIZER(&all_bfds__);
212 static struct hmap *const all_bfds OVS_GUARDED_BY(mutex) = &all_bfds__;
214 static bool bfd_forwarding__(const struct bfd *) OVS_REQUIRES(mutex);
215 static bool bfd_in_poll(const struct bfd *) OVS_REQUIRES(mutex);
216 static void bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex);
217 static const char *bfd_diag_str(enum diag) OVS_REQUIRES(mutex);
218 static const char *bfd_state_str(enum state) OVS_REQUIRES(mutex);
219 static long long int bfd_min_tx(const struct bfd *) OVS_REQUIRES(mutex);
220 static long long int bfd_tx_interval(const struct bfd *)
222 static long long int bfd_rx_interval(const struct bfd *)
224 static void bfd_set_next_tx(struct bfd *) OVS_REQUIRES(mutex);
225 static void bfd_set_state(struct bfd *, enum state, enum diag)
227 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex);
228 static void bfd_put_details(struct ds *, const struct bfd *)
230 static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQUIRES(mutex);
231 static void bfd_try_decay(struct bfd *) OVS_REQUIRES(mutex);
232 static void bfd_decay_update(struct bfd *) OVS_REQUIRES(mutex);
233 static void bfd_check_rx(struct bfd *) OVS_REQUIRES(mutex);
234 static void bfd_forwarding_if_rx_update(struct bfd *) OVS_REQUIRES(mutex);
235 static void bfd_unixctl_show(struct unixctl_conn *, int argc,
236 const char *argv[], void *aux OVS_UNUSED);
237 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
238 int argc, const char *argv[],
239 void *aux OVS_UNUSED);
240 static void log_msg(enum vlog_level, const struct msg *, const char *message,
241 const struct bfd *) OVS_REQUIRES(mutex);
243 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 20);
245 /* Returns true if the interface on which 'bfd' is running may be used to
246 * forward traffic according to the BFD session state. */
248 bfd_forwarding(const struct bfd *bfd) OVS_EXCLUDED(mutex)
252 ovs_mutex_lock(&mutex);
253 ret = bfd_forwarding__(bfd);
254 ovs_mutex_unlock(&mutex);
258 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
259 * intended for the OVS database. */
261 bfd_get_status(const struct bfd *bfd, struct smap *smap)
264 ovs_mutex_lock(&mutex);
265 smap_add(smap, "forwarding", bfd_forwarding__(bfd)? "true" : "false");
266 smap_add(smap, "state", bfd_state_str(bfd->state));
267 smap_add(smap, "diagnostic", bfd_diag_str(bfd->diag));
269 if (bfd->state != STATE_DOWN) {
270 smap_add(smap, "remote_state", bfd_state_str(bfd->rmt_state));
271 smap_add(smap, "remote_diagnostic", bfd_diag_str(bfd->rmt_diag));
273 ovs_mutex_unlock(&mutex);
276 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
277 * according to the database configuration contained in 'cfg'. Takes ownership
278 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
279 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
280 * Also returns NULL if cfg is NULL. */
282 bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
283 struct netdev *netdev) OVS_EXCLUDED(mutex)
285 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
286 static atomic_uint16_t udp_src = ATOMIC_VAR_INIT(0);
289 long long int min_tx, min_rx;
290 bool need_poll = false;
291 bool cfg_min_rx_changed = false;
292 bool cpath_down, forwarding_if_rx;
294 uint8_t ea[ETH_ADDR_LEN];
296 if (ovsthread_once_start(&once)) {
297 unixctl_command_register("bfd/show", "[interface]", 0, 1,
298 bfd_unixctl_show, NULL);
299 unixctl_command_register("bfd/set-forwarding",
300 "[interface] normal|false|true", 1, 2,
301 bfd_unixctl_set_forwarding_override, NULL);
302 ovsthread_once_done(&once);
305 if (!cfg || !smap_get_bool(cfg, "enable", false)) {
310 ovs_mutex_lock(&mutex);
312 bfd = xzalloc(sizeof *bfd);
313 bfd->name = xstrdup(name);
314 bfd->forwarding_override = -1;
315 bfd->disc = generate_discriminator();
316 hmap_insert(all_bfds, &bfd->node, bfd->disc);
318 bfd->diag = DIAG_NONE;
321 atomic_init(&bfd->ref_cnt, 1);
322 bfd->netdev = netdev_ref(netdev);
323 bfd->rx_packets = bfd_rx_packets(bfd);
324 bfd->in_decay = false;
326 /* RFC 5881 section 4
327 * The source port MUST be in the range 49152 through 65535. The same
328 * UDP source port number MUST be used for all BFD Control packets
329 * associated with a particular session. The source port number SHOULD
330 * be unique among all BFD sessions on the system. */
331 atomic_add(&udp_src, 1, &bfd->udp_src);
332 bfd->udp_src = (bfd->udp_src % 16384) + 49152;
334 bfd_set_state(bfd, STATE_DOWN, DIAG_NONE);
336 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
339 atomic_store(&bfd->check_tnl_key,
340 smap_get_bool(cfg, "check_tnl_key", false));
341 min_tx = smap_get_int(cfg, "min_tx", 100);
342 min_tx = MAX(min_tx, 100);
343 if (bfd->cfg_min_tx != min_tx) {
344 bfd->cfg_min_tx = min_tx;
345 if (bfd->state != STATE_UP
346 || (!bfd_in_poll(bfd) && bfd->cfg_min_tx < bfd->min_tx)) {
347 bfd->min_tx = bfd->cfg_min_tx;
352 min_rx = smap_get_int(cfg, "min_rx", 1000);
353 min_rx = MAX(min_rx, 100);
354 if (bfd->cfg_min_rx != min_rx) {
355 bfd->cfg_min_rx = min_rx;
356 if (bfd->state != STATE_UP
357 || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
358 bfd->min_rx = bfd->cfg_min_rx;
360 cfg_min_rx_changed = true;
364 decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
365 if (bfd->decay_min_rx != decay_min_rx || cfg_min_rx_changed) {
366 if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
367 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
368 bfd->name, bfd->cfg_min_rx);
369 bfd->decay_min_rx = 0;
371 bfd->decay_min_rx = decay_min_rx;
374 bfd->in_decay = false;
375 bfd_decay_update(bfd);
379 cpath_down = smap_get_bool(cfg, "cpath_down", false);
380 if (bfd->cpath_down != cpath_down) {
381 bfd->cpath_down = cpath_down;
382 if (bfd->diag == DIAG_NONE || bfd->diag == DIAG_CPATH_DOWN) {
383 bfd_set_state(bfd, bfd->state, DIAG_NONE);
388 hwaddr = smap_get(cfg, "bfd_dst_mac");
389 if (hwaddr && eth_addr_from_string(hwaddr, ea) && !eth_addr_is_zero(ea)) {
390 memcpy(bfd->eth_dst, ea, ETH_ADDR_LEN);
391 bfd->eth_dst_set = true;
392 } else if (bfd->eth_dst_set) {
393 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
394 bfd->eth_dst_set = false;
397 forwarding_if_rx = smap_get_bool(cfg, "forwarding_if_rx", false);
398 if (bfd->forwarding_if_rx != forwarding_if_rx) {
399 bfd->forwarding_if_rx = forwarding_if_rx;
400 if (bfd->state == STATE_UP && bfd->forwarding_if_rx) {
401 bfd_forwarding_if_rx_update(bfd);
403 bfd->forwarding_if_rx_detect_time = 0;
410 ovs_mutex_unlock(&mutex);
415 bfd_ref(const struct bfd *bfd_)
417 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
420 atomic_add(&bfd->ref_cnt, 1, &orig);
421 ovs_assert(orig > 0);
427 bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
432 atomic_sub(&bfd->ref_cnt, 1, &orig);
433 ovs_assert(orig > 0);
435 ovs_mutex_lock(&mutex);
436 hmap_remove(all_bfds, &bfd->node);
437 netdev_close(bfd->netdev);
440 ovs_mutex_unlock(&mutex);
446 bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
448 ovs_mutex_lock(&mutex);
449 if (bfd->flags & FLAG_FINAL) {
450 poll_immediate_wake();
453 poll_timer_wait_until(bfd->next_tx);
454 if (bfd->state > STATE_DOWN) {
455 poll_timer_wait_until(bfd->detect_time);
457 ovs_mutex_unlock(&mutex);
461 bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
466 ovs_mutex_lock(&mutex);
468 old_in_decay = bfd->in_decay;
470 if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
471 bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
474 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
475 * configured, and decay_detect_time is reached. */
476 if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
477 && now >= bfd->decay_detect_time) {
481 /* Always checks the reception of any packet. */
484 if (bfd->min_tx != bfd->cfg_min_tx
485 || (bfd->min_rx != bfd->cfg_min_rx && bfd->min_rx != bfd->decay_min_rx)
486 || bfd->in_decay != old_in_decay) {
489 ovs_mutex_unlock(&mutex);
493 bfd_should_send_packet(const struct bfd *bfd) OVS_EXCLUDED(mutex)
496 ovs_mutex_lock(&mutex);
497 ret = bfd->flags & FLAG_FINAL || time_msec() >= bfd->next_tx;
498 ovs_mutex_unlock(&mutex);
503 bfd_put_packet(struct bfd *bfd, struct ofpbuf *p,
504 uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
506 long long int min_tx, min_rx;
507 struct udp_header *udp;
508 struct eth_header *eth;
509 struct ip_header *ip;
512 ovs_mutex_lock(&mutex);
514 long long int delay = time_msec() - bfd->next_tx;
515 long long int interval = bfd_tx_interval(bfd);
516 if (delay > interval * 3 / 2) {
517 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
518 " control message", bfd->name, delay, interval);
522 /* RFC 5880 Section 6.5
523 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
525 ovs_assert(!(bfd->flags & FLAG_POLL) || !(bfd->flags & FLAG_FINAL));
527 ofpbuf_reserve(p, 2); /* Properly align after the ethernet header. */
528 eth = ofpbuf_put_uninit(p, sizeof *eth);
529 memcpy(eth->eth_src, eth_src, ETH_ADDR_LEN);
530 memcpy(eth->eth_dst, bfd->eth_dst, ETH_ADDR_LEN);
531 eth->eth_type = htons(ETH_TYPE_IP);
533 ip = ofpbuf_put_zeros(p, sizeof *ip);
534 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
535 ip->ip_tot_len = htons(sizeof *ip + sizeof *udp + sizeof *msg);
537 ip->ip_tos = IPTOS_LOWDELAY | IPTOS_THROUGHPUT;
538 ip->ip_proto = IPPROTO_UDP;
539 ip->ip_src = htonl(0xA9FE0100); /* 169.254.1.0 Link Local. */
540 ip->ip_dst = htonl(0xA9FE0101); /* 169.254.1.1 Link Local. */
541 ip->ip_csum = csum(ip, sizeof *ip);
543 udp = ofpbuf_put_zeros(p, sizeof *udp);
544 udp->udp_src = htons(bfd->udp_src);
545 udp->udp_dst = htons(BFD_DEST_PORT);
546 udp->udp_len = htons(sizeof *udp + sizeof *msg);
548 msg = ofpbuf_put_uninit(p, sizeof *msg);
549 msg->vers_diag = (BFD_VERSION << 5) | bfd->diag;
550 msg->flags = (bfd->state & STATE_MASK) | bfd->flags;
552 msg->mult = bfd->mult;
553 msg->length = BFD_PACKET_LEN;
554 msg->my_disc = htonl(bfd->disc);
555 msg->your_disc = htonl(bfd->rmt_disc);
556 msg->min_rx_echo = htonl(0);
558 if (bfd_in_poll(bfd)) {
559 min_tx = bfd->poll_min_tx;
560 min_rx = bfd->poll_min_rx;
562 min_tx = bfd_min_tx(bfd);
563 min_rx = bfd->min_rx;
566 msg->min_tx = htonl(min_tx * 1000);
567 msg->min_rx = htonl(min_rx * 1000);
569 bfd->flags &= ~FLAG_FINAL;
571 log_msg(VLL_DBG, msg, "Sending BFD Message", bfd);
573 bfd->last_tx = time_msec();
574 bfd_set_next_tx(bfd);
575 ovs_mutex_unlock(&mutex);
579 bfd_should_process_flow(const struct bfd *bfd_, const struct flow *flow,
580 struct flow_wildcards *wc)
582 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
585 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
586 if (bfd->eth_dst_set && memcmp(bfd->eth_dst, flow->dl_dst, ETH_ADDR_LEN)) {
590 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
591 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
593 atomic_read(&bfd->check_tnl_key, &check_tnl_key);
595 memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id);
597 return (flow->dl_type == htons(ETH_TYPE_IP)
598 && flow->nw_proto == IPPROTO_UDP
599 && flow->tp_dst == htons(BFD_DEST_PORT)
600 && (!check_tnl_key || flow->tunnel.tun_id == htonll(0)));
604 bfd_process_packet(struct bfd *bfd, const struct flow *flow,
605 const struct ofpbuf *p) OVS_EXCLUDED(mutex)
607 uint32_t rmt_min_rx, pkt_your_disc;
608 enum state rmt_state;
613 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
615 ovs_mutex_lock(&mutex);
616 /* Increments the decay rx counter. */
619 if (flow->nw_ttl != 255) {
620 /* XXX Should drop in the kernel to prevent DOS. */
624 msg = ofpbuf_at(p, (uint8_t *)p->l7 - (uint8_t *)p->data, BFD_PACKET_LEN);
626 VLOG_INFO_RL(&rl, "%s: Received unparseable BFD control message.",
631 /* RFC 5880 Section 6.8.6
632 * If the Length field is greater than the payload of the encapsulating
633 * protocol, the packet MUST be discarded.
635 * Note that we make this check implicity. Above we use ofpbuf_at() to
636 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
637 * the encapsulating protocol. Below we require msg->length to be exactly
638 * BFD_PACKET_LEN bytes. */
640 flags = msg->flags & FLAGS_MASK;
641 rmt_state = msg->flags & STATE_MASK;
642 version = msg->vers_diag >> VERS_SHIFT;
644 log_msg(VLL_DBG, msg, "Received BFD control message", bfd);
646 if (version != BFD_VERSION) {
647 log_msg(VLL_WARN, msg, "Incorrect version", bfd);
651 /* Technically this should happen after the length check. We don't support
652 * authentication however, so it's simpler to do the check first. */
653 if (flags & FLAG_AUTH) {
654 log_msg(VLL_WARN, msg, "Authenticated control message with"
655 " authentication disabled", bfd);
659 if (msg->length != BFD_PACKET_LEN) {
660 log_msg(VLL_WARN, msg, "Unexpected length", bfd);
661 if (msg->length < BFD_PACKET_LEN) {
667 log_msg(VLL_WARN, msg, "Zero multiplier", bfd);
671 if (flags & FLAG_MULTIPOINT) {
672 log_msg(VLL_WARN, msg, "Unsupported multipoint flag", bfd);
677 log_msg(VLL_WARN, msg, "NULL my_disc", bfd);
681 pkt_your_disc = ntohl(msg->your_disc);
683 /* Technically, we should use the your discriminator field to figure
684 * out which 'struct bfd' this packet is destined towards. That way a
685 * bfd session could migrate from one interface to another
686 * transparently. This doesn't fit in with the OVS structure very
687 * well, so in this respect, we are not compliant. */
688 if (pkt_your_disc != bfd->disc) {
689 log_msg(VLL_WARN, msg, "Incorrect your_disc", bfd);
692 } else if (rmt_state > STATE_DOWN) {
693 log_msg(VLL_WARN, msg, "Null your_disc", bfd);
697 bfd->rmt_disc = ntohl(msg->my_disc);
698 bfd->rmt_state = rmt_state;
699 bfd->rmt_flags = flags;
700 bfd->rmt_diag = msg->vers_diag & DIAG_MASK;
702 if (flags & FLAG_FINAL && bfd_in_poll(bfd)) {
703 bfd->min_tx = bfd->poll_min_tx;
704 bfd->min_rx = bfd->poll_min_rx;
705 bfd->flags &= ~FLAG_POLL;
706 log_msg(VLL_INFO, msg, "Poll sequence terminated", bfd);
709 if (flags & FLAG_POLL) {
710 /* RFC 5880 Section 6.5
711 * When the other system receives a Poll, it immediately transmits a
712 * BFD Control packet with the Final (F) bit set, independent of any
713 * periodic BFD Control packets it may be sending
714 * (see section 6.8.7). */
715 bfd->flags &= ~FLAG_POLL;
716 bfd->flags |= FLAG_FINAL;
719 rmt_min_rx = MAX(ntohl(msg->min_rx) / 1000, 1);
720 if (bfd->rmt_min_rx != rmt_min_rx) {
721 bfd->rmt_min_rx = rmt_min_rx;
722 bfd_set_next_tx(bfd);
723 log_msg(VLL_INFO, msg, "New remote min_rx", bfd);
726 bfd->rmt_min_tx = MAX(ntohl(msg->min_tx) / 1000, 1);
727 bfd->detect_time = bfd_rx_interval(bfd) * bfd->mult + time_msec();
729 if (bfd->state == STATE_ADMIN_DOWN) {
730 VLOG_DBG_RL(&rl, "Administratively down, dropping control message.");
734 if (rmt_state == STATE_ADMIN_DOWN) {
735 if (bfd->state != STATE_DOWN) {
736 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
739 switch (bfd->state) {
741 if (rmt_state == STATE_DOWN) {
742 bfd_set_state(bfd, STATE_INIT, bfd->diag);
743 } else if (rmt_state == STATE_INIT) {
744 bfd_set_state(bfd, STATE_UP, bfd->diag);
748 if (rmt_state > STATE_DOWN) {
749 bfd_set_state(bfd, STATE_UP, bfd->diag);
753 if (rmt_state <= STATE_DOWN) {
754 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
755 log_msg(VLL_INFO, msg, "Remote signaled STATE_DOWN", bfd);
758 case STATE_ADMIN_DOWN:
763 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
766 ovs_mutex_unlock(&mutex);
769 /* Must be called when the netdev owned by 'bfd' should change. */
771 bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
774 ovs_mutex_lock(&mutex);
775 if (bfd->netdev != netdev) {
776 netdev_close(bfd->netdev);
777 bfd->netdev = netdev_ref(netdev);
778 if (bfd->decay_min_rx && bfd->state == STATE_UP) {
779 bfd_decay_update(bfd);
781 if (bfd->forwarding_if_rx && bfd->state == STATE_UP) {
782 bfd_forwarding_if_rx_update(bfd);
784 bfd->rx_packets = bfd_rx_packets(bfd);
786 ovs_mutex_unlock(&mutex);
791 bfd_forwarding__(const struct bfd *bfd) OVS_REQUIRES(mutex)
795 if (bfd->forwarding_override != -1) {
796 return bfd->forwarding_override == 1;
799 time = bfd->forwarding_if_rx_detect_time;
800 return (bfd->state == STATE_UP
801 || (bfd->forwarding_if_rx && time > time_msec()))
802 && bfd->rmt_diag != DIAG_PATH_DOWN
803 && bfd->rmt_diag != DIAG_CPATH_DOWN
804 && bfd->rmt_diag != DIAG_RCPATH_DOWN;
809 bfd_in_poll(const struct bfd *bfd) OVS_REQUIRES(mutex)
811 return (bfd->flags & FLAG_POLL) != 0;
815 bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex)
817 if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
818 && !(bfd->flags & FLAG_FINAL)) {
819 bfd->poll_min_tx = bfd->cfg_min_tx;
820 bfd->poll_min_rx = bfd->in_decay ? bfd->decay_min_rx : bfd->cfg_min_rx;
821 bfd->flags |= FLAG_POLL;
823 VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
828 bfd_min_tx(const struct bfd *bfd) OVS_REQUIRES(mutex)
830 /* RFC 5880 Section 6.8.3
831 * When bfd.SessionState is not Up, the system MUST set
832 * bfd.DesiredMinTxInterval to a value of not less than one second
833 * (1,000,000 microseconds). This is intended to ensure that the
834 * bandwidth consumed by BFD sessions that are not Up is negligible,
835 * particularly in the case where a neighbor may not be running BFD. */
836 return (bfd->state == STATE_UP ? bfd->min_tx : MAX(bfd->min_tx, 1000));
840 bfd_tx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
842 long long int interval = bfd_min_tx(bfd);
843 return MAX(interval, bfd->rmt_min_rx);
847 bfd_rx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
849 return MAX(bfd->min_rx, bfd->rmt_min_tx);
853 bfd_set_next_tx(struct bfd *bfd) OVS_REQUIRES(mutex)
855 long long int interval = bfd_tx_interval(bfd);
856 interval -= interval * random_range(26) / 100;
857 bfd->next_tx = bfd->last_tx + interval;
861 bfd_flag_str(enum flags flags)
863 struct ds ds = DS_EMPTY_INITIALIZER;
864 static char flag_str[128];
870 if (flags & FLAG_MULTIPOINT) {
871 ds_put_cstr(&ds, "multipoint ");
874 if (flags & FLAG_DEMAND) {
875 ds_put_cstr(&ds, "demand ");
878 if (flags & FLAG_AUTH) {
879 ds_put_cstr(&ds, "auth ");
882 if (flags & FLAG_CTL) {
883 ds_put_cstr(&ds, "ctl ");
886 if (flags & FLAG_FINAL) {
887 ds_put_cstr(&ds, "final ");
890 if (flags & FLAG_POLL) {
891 ds_put_cstr(&ds, "poll ");
894 /* Do not copy the trailing whitespace. */
896 ovs_strlcpy(flag_str, ds_cstr(&ds), sizeof flag_str);
902 bfd_state_str(enum state state)
905 case STATE_ADMIN_DOWN: return "admin_down";
906 case STATE_DOWN: return "down";
907 case STATE_INIT: return "init";
908 case STATE_UP: return "up";
909 default: return "invalid";
914 bfd_diag_str(enum diag diag) {
916 case DIAG_NONE: return "No Diagnostic";
917 case DIAG_EXPIRED: return "Control Detection Time Expired";
918 case DIAG_ECHO_FAILED: return "Echo Function Failed";
919 case DIAG_RMT_DOWN: return "Neighbor Signaled Session Down";
920 case DIAG_FWD_RESET: return "Forwarding Plane Reset";
921 case DIAG_PATH_DOWN: return "Path Down";
922 case DIAG_CPATH_DOWN: return "Concatenated Path Down";
923 case DIAG_ADMIN_DOWN: return "Administratively Down";
924 case DIAG_RCPATH_DOWN: return "Reverse Concatenated Path Down";
925 default: return "Invalid Diagnostic";
930 log_msg(enum vlog_level level, const struct msg *p, const char *message,
931 const struct bfd *bfd) OVS_REQUIRES(mutex)
933 struct ds ds = DS_EMPTY_INITIALIZER;
935 if (vlog_should_drop(THIS_MODULE, level, &rl)) {
941 "\n\tvers:%"PRIu8" diag:\"%s\" state:%s mult:%"PRIu8
944 "\n\tmy_disc:0x%"PRIx32" your_disc:0x%"PRIx32
945 "\n\tmin_tx:%"PRIu32"us (%"PRIu32"ms)"
946 "\n\tmin_rx:%"PRIu32"us (%"PRIu32"ms)"
947 "\n\tmin_rx_echo:%"PRIu32"us (%"PRIu32"ms)",
948 bfd->name, message, p->vers_diag >> VERS_SHIFT,
949 bfd_diag_str(p->vers_diag & DIAG_MASK),
950 bfd_state_str(p->flags & STATE_MASK),
951 p->mult, p->length, bfd_flag_str(p->flags & FLAGS_MASK),
952 ntohl(p->my_disc), ntohl(p->your_disc),
953 ntohl(p->min_tx), ntohl(p->min_tx) / 1000,
954 ntohl(p->min_rx), ntohl(p->min_rx) / 1000,
955 ntohl(p->min_rx_echo), ntohl(p->min_rx_echo) / 1000);
956 bfd_put_details(&ds, bfd);
957 VLOG(level, "%s", ds_cstr(&ds));
962 bfd_set_state(struct bfd *bfd, enum state state, enum diag diag)
965 if (diag == DIAG_NONE && bfd->cpath_down) {
966 diag = DIAG_CPATH_DOWN;
969 if (bfd->state != state || bfd->diag != diag) {
970 if (!VLOG_DROP_INFO(&rl)) {
971 struct ds ds = DS_EMPTY_INITIALIZER;
973 ds_put_format(&ds, "%s: BFD state change: %s->%s"
974 " \"%s\"->\"%s\".\n",
975 bfd->name, bfd_state_str(bfd->state),
976 bfd_state_str(state), bfd_diag_str(bfd->diag),
978 bfd_put_details(&ds, bfd);
979 VLOG_INFO("%s", ds_cstr(&ds));
986 if (bfd->state <= STATE_DOWN) {
987 bfd->rmt_state = STATE_DOWN;
988 bfd->rmt_diag = DIAG_NONE;
993 /* Resets the min_rx if in_decay. */
995 bfd->min_rx = bfd->cfg_min_rx;
996 bfd->in_decay = false;
999 /* Resets the decay when state changes to STATE_UP
1000 * and decay_min_rx is configured. */
1001 if (bfd->state == STATE_UP && bfd->decay_min_rx) {
1002 bfd_decay_update(bfd);
1008 bfd_rx_packets(const struct bfd *bfd) OVS_REQUIRES(mutex)
1010 struct netdev_stats stats;
1012 if (!netdev_get_stats(bfd->netdev, &stats)) {
1013 return stats.rx_packets;
1019 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
1020 * the 'expect' value. */
1022 bfd_try_decay(struct bfd *bfd) OVS_REQUIRES(mutex)
1024 int64_t diff, expect;
1026 /* The 'diff' is the difference between current interface rx_packets
1027 * stats and last-time check. The 'expect' is the recorded number of
1028 * bfd control packets received within an approximately decay_min_rx
1029 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
1031 * Since the update of rx_packets stats at interface happens
1032 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1033 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1034 * more wiggle room. */
1035 diff = bfd_rx_packets(bfd) - bfd->decay_rx_packets;
1036 expect = 2 * MAX(bfd->decay_rx_ctl, 1);
1037 bfd->in_decay = diff <= expect ? true : false;
1038 bfd_decay_update(bfd);
1041 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1043 bfd_decay_update(struct bfd * bfd) OVS_REQUIRES(mutex)
1045 bfd->decay_rx_packets = bfd_rx_packets(bfd);
1046 bfd->decay_rx_ctl = 0;
1047 bfd->decay_detect_time = MAX(bfd->decay_min_rx, 2000) + time_msec();
1050 /* Checks if there are packets received during the time since last call.
1051 * If forwarding_if_rx is enabled and packets are received, updates the
1052 * forwarding_if_rx_detect_time. */
1054 bfd_check_rx(struct bfd *bfd) OVS_REQUIRES(mutex)
1056 uint64_t rx_packets = bfd_rx_packets(bfd);
1059 diff = rx_packets - bfd->rx_packets;
1060 bfd->rx_packets = rx_packets;
1062 VLOG_INFO_RL(&rl, "rx_packets count is smaller than last time.");
1064 if (bfd->forwarding_if_rx && diff > 0) {
1065 bfd_forwarding_if_rx_update(bfd);
1069 /* Updates the forwarding_if_rx_detect_time. */
1071 bfd_forwarding_if_rx_update(struct bfd *bfd) OVS_REQUIRES(mutex)
1073 int64_t incr = bfd_rx_interval(bfd) * bfd->mult;
1074 bfd->forwarding_if_rx_detect_time = MAX(incr, 2000) + time_msec();
1078 generate_discriminator(void)
1082 /* RFC 5880 Section 6.8.1
1083 * It SHOULD be set to a random (but still unique) value to improve
1084 * security. The value is otherwise outside the scope of this
1090 /* 'disc' is by definition random, so there's no reason to waste time
1092 disc = random_uint32();
1093 HMAP_FOR_EACH_IN_BUCKET (bfd, node, disc, all_bfds) {
1094 if (bfd->disc == disc) {
1105 bfd_find_by_name(const char *name) OVS_REQUIRES(mutex)
1109 HMAP_FOR_EACH (bfd, node, all_bfds) {
1110 if (!strcmp(bfd->name, name)) {
1118 bfd_put_details(struct ds *ds, const struct bfd *bfd) OVS_REQUIRES(mutex)
1120 ds_put_format(ds, "\tForwarding: %s\n",
1121 bfd_forwarding__(bfd) ? "true" : "false");
1122 ds_put_format(ds, "\tDetect Multiplier: %d\n", bfd->mult);
1123 ds_put_format(ds, "\tConcatenated Path Down: %s\n",
1124 bfd->cpath_down ? "true" : "false");
1125 ds_put_format(ds, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd));
1126 ds_put_format(ds, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd));
1127 ds_put_format(ds, "\tDetect Time: now %+lldms\n",
1128 time_msec() - bfd->detect_time);
1129 ds_put_format(ds, "\tNext TX Time: now %+lldms\n",
1130 time_msec() - bfd->next_tx);
1131 ds_put_format(ds, "\tLast TX Time: now %+lldms\n",
1132 time_msec() - bfd->last_tx);
1134 ds_put_cstr(ds, "\n");
1136 ds_put_format(ds, "\tLocal Flags: %s\n", bfd_flag_str(bfd->flags));
1137 ds_put_format(ds, "\tLocal Session State: %s\n",
1138 bfd_state_str(bfd->state));
1139 ds_put_format(ds, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd->diag));
1140 ds_put_format(ds, "\tLocal Discriminator: 0x%"PRIx32"\n", bfd->disc);
1141 ds_put_format(ds, "\tLocal Minimum TX Interval: %lldms\n",
1143 ds_put_format(ds, "\tLocal Minimum RX Interval: %lldms\n", bfd->min_rx);
1145 ds_put_cstr(ds, "\n");
1147 ds_put_format(ds, "\tRemote Flags: %s\n", bfd_flag_str(bfd->rmt_flags));
1148 ds_put_format(ds, "\tRemote Session State: %s\n",
1149 bfd_state_str(bfd->rmt_state));
1150 ds_put_format(ds, "\tRemote Diagnostic: %s\n",
1151 bfd_diag_str(bfd->rmt_diag));
1152 ds_put_format(ds, "\tRemote Discriminator: 0x%"PRIx32"\n", bfd->rmt_disc);
1153 ds_put_format(ds, "\tRemote Minimum TX Interval: %lldms\n",
1155 ds_put_format(ds, "\tRemote Minimum RX Interval: %lldms\n",
1160 bfd_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1161 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1163 struct ds ds = DS_EMPTY_INITIALIZER;
1166 ovs_mutex_lock(&mutex);
1168 bfd = bfd_find_by_name(argv[1]);
1170 unixctl_command_reply_error(conn, "no such bfd object");
1173 bfd_put_details(&ds, bfd);
1175 HMAP_FOR_EACH (bfd, node, all_bfds) {
1176 ds_put_format(&ds, "---- %s ----\n", bfd->name);
1177 bfd_put_details(&ds, bfd);
1180 unixctl_command_reply(conn, ds_cstr(&ds));
1184 ovs_mutex_unlock(&mutex);
1189 bfd_unixctl_set_forwarding_override(struct unixctl_conn *conn, int argc,
1190 const char *argv[], void *aux OVS_UNUSED)
1193 const char *forward_str = argv[argc - 1];
1194 int forwarding_override;
1197 ovs_mutex_lock(&mutex);
1198 if (!strcasecmp("true", forward_str)) {
1199 forwarding_override = 1;
1200 } else if (!strcasecmp("false", forward_str)) {
1201 forwarding_override = 0;
1202 } else if (!strcasecmp("normal", forward_str)) {
1203 forwarding_override = -1;
1205 unixctl_command_reply_error(conn, "unknown fault string");
1210 bfd = bfd_find_by_name(argv[1]);
1212 unixctl_command_reply_error(conn, "no such BFD object");
1215 bfd->forwarding_override = forwarding_override;
1217 HMAP_FOR_EACH (bfd, node, all_bfds) {
1218 bfd->forwarding_override = forwarding_override;
1222 unixctl_command_reply(conn, "OK");
1225 ovs_mutex_unlock(&mutex);