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"
42 #include "unaligned.h"
47 VLOG_DEFINE_THIS_MODULE(bfd);
51 * The goal of this module is to replace CFM with something both more flexible
52 * and standards compliant. In service of this goal, the following needs to be
56 * * Implement Demand mode.
57 * * Go through the RFC line by line and verify we comply.
58 * * Test against a hardware implementation. Preferably a popular one.
59 * * Delete BFD packets with nw_ttl != 255 in the datapath to prevent DOS
64 * - Set TOS/PCP on the outer tunnel header when encapped.
66 * - Sending BFD messages should be in its own thread/process.
68 * - Scale testing. How does it operate when there are large number of bfd
69 * sessions? Do we ever have random flaps? What's the CPU utilization?
71 * - Rely on data traffic for liveness by using BFD demand mode.
72 * If we're receiving traffic on a port, we can safely assume it's up (modulo
73 * unidrectional failures). BFD has a demand mode in which it can stay quiet
74 * unless it feels the need to check the status of the port. Using this, we
75 * can implement a strategy in which BFD only sends control messages on dark
78 * - Depending on how one interprets the spec, it appears that a BFD session
79 * can never change bfd.LocalDiag to "No Diagnostic". We should verify that
80 * this is what hardware implementations actually do. Seems like "No
81 * Diagnostic" should be set once a BFD session state goes UP. */
86 FLAG_MULTIPOINT = 1 << 0,
95 STATE_ADMIN_DOWN = 0 << 6,
102 DIAG_NONE = 0, /* No Diagnostic. */
103 DIAG_EXPIRED = 1, /* Control Detection Time Expired. */
104 DIAG_ECHO_FAILED = 2, /* Echo Function Failed. */
105 DIAG_RMT_DOWN = 3, /* Neighbor Signaled Session Down. */
106 DIAG_FWD_RESET = 4, /* Forwarding Plane Reset. */
107 DIAG_PATH_DOWN = 5, /* Path Down. */
108 DIAG_CPATH_DOWN = 6, /* Concatenated Path Down. */
109 DIAG_ADMIN_DOWN = 7, /* Administratively Down. */
110 DIAG_RCPATH_DOWN = 8 /* Reverse Concatenated Path Down. */
113 /* RFC 5880 Section 4.1
115 * 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
116 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
117 * |Vers | Diag |Sta|P|F|C|A|D|M| Detect Mult | Length |
118 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
119 * | My Discriminator |
120 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
121 * | Your Discriminator |
122 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
123 * | Desired Min TX Interval |
124 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
125 * | Required Min RX Interval |
126 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
127 * | Required Min Echo RX Interval |
128 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ */
130 uint8_t vers_diag; /* Version and diagnostic. */
131 uint8_t flags; /* 2bit State field followed by flags. */
132 uint8_t mult; /* Fault detection multiplier. */
133 uint8_t length; /* Length of this BFD message. */
134 ovs_be32 my_disc; /* My discriminator. */
135 ovs_be32 your_disc; /* Your discriminator. */
136 ovs_be32 min_tx; /* Desired minimum tx interval. */
137 ovs_be32 min_rx; /* Required minimum rx interval. */
138 ovs_be32 min_rx_echo; /* Required minimum echo rx interval. */
140 BUILD_ASSERT_DECL(BFD_PACKET_LEN == sizeof(struct msg));
142 #define DIAG_MASK 0x1f
144 #define STATE_MASK 0xC0
145 #define FLAGS_MASK 0x3f
148 struct hmap_node node; /* In 'all_bfds'. */
149 uint32_t disc; /* bfd.LocalDiscr. Key in 'all_bfds' hmap. */
151 char *name; /* Name used for logging. */
153 bool cpath_down; /* Concatenated Path Down. */
154 uint8_t mult; /* bfd.DetectMult. */
156 struct netdev *netdev;
157 uint64_t rx_packets; /* Packets received by 'netdev'. */
159 enum state state; /* bfd.SessionState. */
160 enum state rmt_state; /* bfd.RemoteSessionState. */
162 enum diag diag; /* bfd.LocalDiag. */
163 enum diag rmt_diag; /* Remote diagnostic. */
165 enum flags flags; /* Flags sent on messages. */
166 enum flags rmt_flags; /* Flags last received. */
168 uint32_t rmt_disc; /* bfd.RemoteDiscr. */
170 uint8_t eth_dst[ETH_ADDR_LEN];/* Ethernet destination address. */
171 bool eth_dst_set; /* 'eth_dst' set through database. */
173 uint16_t udp_src; /* UDP source port. */
175 /* All timers in milliseconds. */
176 long long int rmt_min_rx; /* bfd.RemoteMinRxInterval. */
177 long long int rmt_min_tx; /* Remote minimum TX interval. */
179 long long int cfg_min_tx; /* Configured minimum TX rate. */
180 long long int cfg_min_rx; /* Configured required minimum RX rate. */
181 long long int poll_min_tx; /* Min TX negotating in a poll sequence. */
182 long long int poll_min_rx; /* Min RX negotating in a poll sequence. */
183 long long int min_tx; /* bfd.DesiredMinTxInterval. */
184 long long int min_rx; /* bfd.RequiredMinRxInterval. */
186 long long int last_tx; /* Last TX time. */
187 long long int next_tx; /* Next TX time. */
188 long long int detect_time; /* RFC 5880 6.8.4 Detection time. */
190 bool last_forwarding; /* Last calculation of forwarding flag. */
191 int forwarding_override; /* Manual override of 'forwarding' status. */
193 atomic_bool check_tnl_key; /* Verify tunnel key of inbound packets? */
196 /* When forward_if_rx is true, bfd_forwarding() will return
197 * true as long as there are incoming packets received.
198 * Note, forwarding_override still has higher priority. */
199 bool forwarding_if_rx;
200 long long int forwarding_if_rx_detect_time;
202 /* BFD decay related variables. */
203 bool in_decay; /* True when bfd is in decay. */
204 int decay_min_rx; /* min_rx is set to decay_min_rx when */
206 int decay_rx_ctl; /* Count bfd packets received within decay */
207 /* detect interval. */
208 uint64_t decay_rx_packets; /* Packets received by 'netdev'. */
209 long long int decay_detect_time; /* Decay detection time. */
211 uint64_t flap_count; /* Counts bfd forwarding flaps. */
214 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
215 static struct hmap all_bfds__ = HMAP_INITIALIZER(&all_bfds__);
216 static struct hmap *const all_bfds OVS_GUARDED_BY(mutex) = &all_bfds__;
218 static bool bfd_forwarding__(struct bfd *) OVS_REQUIRES(mutex);
219 static bool bfd_in_poll(const struct bfd *) OVS_REQUIRES(mutex);
220 static void bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex);
221 static const char *bfd_diag_str(enum diag) OVS_REQUIRES(mutex);
222 static const char *bfd_state_str(enum state) OVS_REQUIRES(mutex);
223 static long long int bfd_min_tx(const struct bfd *) OVS_REQUIRES(mutex);
224 static long long int bfd_tx_interval(const struct bfd *)
226 static long long int bfd_rx_interval(const struct bfd *)
228 static void bfd_set_next_tx(struct bfd *) OVS_REQUIRES(mutex);
229 static void bfd_set_state(struct bfd *, enum state, enum diag)
231 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex);
232 static void bfd_put_details(struct ds *, const struct bfd *)
234 static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQUIRES(mutex);
235 static void bfd_try_decay(struct bfd *) OVS_REQUIRES(mutex);
236 static void bfd_decay_update(struct bfd *) OVS_REQUIRES(mutex);
238 static void bfd_forwarding_if_rx_update(struct bfd *) OVS_REQUIRES(mutex);
239 static void bfd_unixctl_show(struct unixctl_conn *, int argc,
240 const char *argv[], void *aux OVS_UNUSED);
241 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
242 int argc, const char *argv[],
243 void *aux OVS_UNUSED);
244 static void log_msg(enum vlog_level, const struct msg *, const char *message,
245 const struct bfd *) OVS_REQUIRES(mutex);
247 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 20);
249 /* Returns true if the interface on which 'bfd' is running may be used to
250 * forward traffic according to the BFD session state. */
252 bfd_forwarding(struct bfd *bfd) OVS_EXCLUDED(mutex)
256 ovs_mutex_lock(&mutex);
257 ret = bfd_forwarding__(bfd);
258 ovs_mutex_unlock(&mutex);
262 /* When forwarding_if_rx is enabled, if there are packets received,
263 * updates forwarding_if_rx_detect_time. */
265 bfd_account_rx(struct bfd *bfd, const struct dpif_flow_stats *stats)
267 if (stats->n_packets && bfd->forwarding_if_rx) {
268 ovs_mutex_lock(&mutex);
269 bfd_forwarding__(bfd);
270 bfd_forwarding_if_rx_update(bfd);
271 bfd_forwarding__(bfd);
272 ovs_mutex_unlock(&mutex);
276 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
277 * intended for the OVS database. */
279 bfd_get_status(const struct bfd *bfd, struct smap *smap)
282 ovs_mutex_lock(&mutex);
283 smap_add(smap, "forwarding",
284 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
286 smap_add(smap, "state", bfd_state_str(bfd->state));
287 smap_add(smap, "diagnostic", bfd_diag_str(bfd->diag));
288 smap_add_format(smap, "flap_count", "%"PRIu64, bfd->flap_count);
290 if (bfd->state != STATE_DOWN) {
291 smap_add(smap, "remote_state", bfd_state_str(bfd->rmt_state));
292 smap_add(smap, "remote_diagnostic", bfd_diag_str(bfd->rmt_diag));
294 ovs_mutex_unlock(&mutex);
297 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
298 * according to the database configuration contained in 'cfg'. Takes ownership
299 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
300 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
301 * Also returns NULL if cfg is NULL. */
303 bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
304 struct netdev *netdev) OVS_EXCLUDED(mutex)
306 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
307 static atomic_uint16_t udp_src = ATOMIC_VAR_INIT(0);
310 long long int min_tx, min_rx;
311 bool need_poll = false;
312 bool cfg_min_rx_changed = false;
313 bool cpath_down, forwarding_if_rx;
315 uint8_t ea[ETH_ADDR_LEN];
317 if (ovsthread_once_start(&once)) {
318 unixctl_command_register("bfd/show", "[interface]", 0, 1,
319 bfd_unixctl_show, NULL);
320 unixctl_command_register("bfd/set-forwarding",
321 "[interface] normal|false|true", 1, 2,
322 bfd_unixctl_set_forwarding_override, NULL);
323 ovsthread_once_done(&once);
326 if (!cfg || !smap_get_bool(cfg, "enable", false)) {
331 ovs_mutex_lock(&mutex);
333 bfd = xzalloc(sizeof *bfd);
334 bfd->name = xstrdup(name);
335 bfd->forwarding_override = -1;
336 bfd->disc = generate_discriminator();
337 hmap_insert(all_bfds, &bfd->node, bfd->disc);
339 bfd->diag = DIAG_NONE;
342 atomic_init(&bfd->ref_cnt, 1);
343 bfd->netdev = netdev_ref(netdev);
344 bfd->rx_packets = bfd_rx_packets(bfd);
345 bfd->in_decay = false;
348 /* RFC 5881 section 4
349 * The source port MUST be in the range 49152 through 65535. The same
350 * UDP source port number MUST be used for all BFD Control packets
351 * associated with a particular session. The source port number SHOULD
352 * be unique among all BFD sessions on the system. */
353 atomic_add(&udp_src, 1, &bfd->udp_src);
354 bfd->udp_src = (bfd->udp_src % 16384) + 49152;
356 bfd_set_state(bfd, STATE_DOWN, DIAG_NONE);
358 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
361 atomic_store(&bfd->check_tnl_key,
362 smap_get_bool(cfg, "check_tnl_key", false));
363 min_tx = smap_get_int(cfg, "min_tx", 100);
364 min_tx = MAX(min_tx, 100);
365 if (bfd->cfg_min_tx != min_tx) {
366 bfd->cfg_min_tx = min_tx;
367 if (bfd->state != STATE_UP
368 || (!bfd_in_poll(bfd) && bfd->cfg_min_tx < bfd->min_tx)) {
369 bfd->min_tx = bfd->cfg_min_tx;
374 min_rx = smap_get_int(cfg, "min_rx", 1000);
375 min_rx = MAX(min_rx, 100);
376 if (bfd->cfg_min_rx != min_rx) {
377 bfd->cfg_min_rx = min_rx;
378 if (bfd->state != STATE_UP
379 || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
380 bfd->min_rx = bfd->cfg_min_rx;
382 cfg_min_rx_changed = true;
386 decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
387 if (bfd->decay_min_rx != decay_min_rx || cfg_min_rx_changed) {
388 if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
389 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
390 bfd->name, bfd->cfg_min_rx);
391 bfd->decay_min_rx = 0;
393 bfd->decay_min_rx = decay_min_rx;
396 bfd->in_decay = false;
397 bfd_decay_update(bfd);
401 cpath_down = smap_get_bool(cfg, "cpath_down", false);
402 if (bfd->cpath_down != cpath_down) {
403 bfd->cpath_down = cpath_down;
404 if (bfd->diag == DIAG_NONE || bfd->diag == DIAG_CPATH_DOWN) {
405 bfd_set_state(bfd, bfd->state, DIAG_NONE);
410 hwaddr = smap_get(cfg, "bfd_dst_mac");
411 if (hwaddr && eth_addr_from_string(hwaddr, ea) && !eth_addr_is_zero(ea)) {
412 memcpy(bfd->eth_dst, ea, ETH_ADDR_LEN);
413 bfd->eth_dst_set = true;
414 } else if (bfd->eth_dst_set) {
415 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
416 bfd->eth_dst_set = false;
419 forwarding_if_rx = smap_get_bool(cfg, "forwarding_if_rx", false);
420 if (bfd->forwarding_if_rx != forwarding_if_rx) {
421 bfd->forwarding_if_rx = forwarding_if_rx;
422 if (bfd->state == STATE_UP && bfd->forwarding_if_rx) {
423 bfd_forwarding_if_rx_update(bfd);
425 bfd->forwarding_if_rx_detect_time = 0;
432 ovs_mutex_unlock(&mutex);
437 bfd_ref(const struct bfd *bfd_)
439 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
442 atomic_add(&bfd->ref_cnt, 1, &orig);
443 ovs_assert(orig > 0);
449 bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
454 atomic_sub(&bfd->ref_cnt, 1, &orig);
455 ovs_assert(orig > 0);
457 ovs_mutex_lock(&mutex);
458 hmap_remove(all_bfds, &bfd->node);
459 netdev_close(bfd->netdev);
462 ovs_mutex_unlock(&mutex);
468 bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
470 poll_timer_wait_until(bfd_wake_time(bfd));
473 /* Returns the next wake up time. */
475 bfd_wake_time(const struct bfd *bfd) OVS_EXCLUDED(mutex)
477 long long int retval;
483 ovs_mutex_lock(&mutex);
484 if (bfd->flags & FLAG_FINAL) {
487 retval = bfd->next_tx;
488 if (bfd->state > STATE_DOWN) {
489 retval = MIN(bfd->detect_time, retval);
492 ovs_mutex_unlock(&mutex);
497 bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
502 ovs_mutex_lock(&mutex);
504 old_in_decay = bfd->in_decay;
506 if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
507 bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
508 bfd_forwarding__(bfd);
511 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
512 * configured, and decay_detect_time is reached. */
513 if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
514 && now >= bfd->decay_detect_time) {
518 if (bfd->min_tx != bfd->cfg_min_tx
519 || (bfd->min_rx != bfd->cfg_min_rx && bfd->min_rx != bfd->decay_min_rx)
520 || bfd->in_decay != old_in_decay) {
523 ovs_mutex_unlock(&mutex);
527 bfd_should_send_packet(const struct bfd *bfd) OVS_EXCLUDED(mutex)
530 ovs_mutex_lock(&mutex);
531 ret = bfd->flags & FLAG_FINAL || time_msec() >= bfd->next_tx;
532 ovs_mutex_unlock(&mutex);
537 bfd_put_packet(struct bfd *bfd, struct ofpbuf *p,
538 uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
540 long long int min_tx, min_rx;
541 struct udp_header *udp;
542 struct eth_header *eth;
543 struct ip_header *ip;
546 ovs_mutex_lock(&mutex);
548 long long int delay = time_msec() - bfd->next_tx;
549 long long int interval = bfd_tx_interval(bfd);
550 if (delay > interval * 3 / 2) {
551 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
552 " control message", bfd->name, delay, interval);
556 /* RFC 5880 Section 6.5
557 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
559 ovs_assert(!(bfd->flags & FLAG_POLL) || !(bfd->flags & FLAG_FINAL));
561 ofpbuf_reserve(p, 2); /* Properly align after the ethernet header. */
562 eth = ofpbuf_put_uninit(p, sizeof *eth);
563 memcpy(eth->eth_src, eth_src, ETH_ADDR_LEN);
564 memcpy(eth->eth_dst, bfd->eth_dst, ETH_ADDR_LEN);
565 eth->eth_type = htons(ETH_TYPE_IP);
567 ip = ofpbuf_put_zeros(p, sizeof *ip);
568 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
569 ip->ip_tot_len = htons(sizeof *ip + sizeof *udp + sizeof *msg);
571 ip->ip_tos = IPTOS_LOWDELAY | IPTOS_THROUGHPUT;
572 ip->ip_proto = IPPROTO_UDP;
573 /* Use link local addresses: */
574 put_16aligned_be32(&ip->ip_src, htonl(0xA9FE0100)); /* 169.254.1.0. */
575 put_16aligned_be32(&ip->ip_dst, htonl(0xA9FE0101)); /* 169.254.1.1. */
576 ip->ip_csum = csum(ip, sizeof *ip);
578 udp = ofpbuf_put_zeros(p, sizeof *udp);
579 udp->udp_src = htons(bfd->udp_src);
580 udp->udp_dst = htons(BFD_DEST_PORT);
581 udp->udp_len = htons(sizeof *udp + sizeof *msg);
583 msg = ofpbuf_put_uninit(p, sizeof *msg);
584 msg->vers_diag = (BFD_VERSION << 5) | bfd->diag;
585 msg->flags = (bfd->state & STATE_MASK) | bfd->flags;
587 msg->mult = bfd->mult;
588 msg->length = BFD_PACKET_LEN;
589 msg->my_disc = htonl(bfd->disc);
590 msg->your_disc = htonl(bfd->rmt_disc);
591 msg->min_rx_echo = htonl(0);
593 if (bfd_in_poll(bfd)) {
594 min_tx = bfd->poll_min_tx;
595 min_rx = bfd->poll_min_rx;
597 min_tx = bfd_min_tx(bfd);
598 min_rx = bfd->min_rx;
601 msg->min_tx = htonl(min_tx * 1000);
602 msg->min_rx = htonl(min_rx * 1000);
604 bfd->flags &= ~FLAG_FINAL;
606 log_msg(VLL_DBG, msg, "Sending BFD Message", bfd);
608 bfd->last_tx = time_msec();
609 bfd_set_next_tx(bfd);
610 ovs_mutex_unlock(&mutex);
614 bfd_should_process_flow(const struct bfd *bfd_, const struct flow *flow,
615 struct flow_wildcards *wc)
617 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
620 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
621 if (bfd->eth_dst_set && memcmp(bfd->eth_dst, flow->dl_dst, ETH_ADDR_LEN)) {
625 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
626 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
628 atomic_read(&bfd->check_tnl_key, &check_tnl_key);
630 memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id);
632 return (flow->dl_type == htons(ETH_TYPE_IP)
633 && flow->nw_proto == IPPROTO_UDP
634 && flow->tp_dst == htons(BFD_DEST_PORT)
635 && (!check_tnl_key || flow->tunnel.tun_id == htonll(0)));
639 bfd_process_packet(struct bfd *bfd, const struct flow *flow,
640 const struct ofpbuf *p) OVS_EXCLUDED(mutex)
642 uint32_t rmt_min_rx, pkt_your_disc;
643 enum state rmt_state;
648 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
650 ovs_mutex_lock(&mutex);
651 /* Increments the decay rx counter. */
654 bfd_forwarding__(bfd);
656 if (flow->nw_ttl != 255) {
657 /* XXX Should drop in the kernel to prevent DOS. */
661 msg = ofpbuf_at(p, (uint8_t *)p->l7 - (uint8_t *)p->data, BFD_PACKET_LEN);
663 VLOG_INFO_RL(&rl, "%s: Received too-short BFD control message (only "
664 "%"PRIdPTR" bytes long, at least %d required).",
665 bfd->name, (uint8_t *) ofpbuf_tail(p) - (uint8_t *) p->l7,
670 /* RFC 5880 Section 6.8.6
671 * If the Length field is greater than the payload of the encapsulating
672 * protocol, the packet MUST be discarded.
674 * Note that we make this check implicity. Above we use ofpbuf_at() to
675 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
676 * the encapsulating protocol. Below we require msg->length to be exactly
677 * BFD_PACKET_LEN bytes. */
679 flags = msg->flags & FLAGS_MASK;
680 rmt_state = msg->flags & STATE_MASK;
681 version = msg->vers_diag >> VERS_SHIFT;
683 log_msg(VLL_DBG, msg, "Received BFD control message", bfd);
685 if (version != BFD_VERSION) {
686 log_msg(VLL_WARN, msg, "Incorrect version", bfd);
690 /* Technically this should happen after the length check. We don't support
691 * authentication however, so it's simpler to do the check first. */
692 if (flags & FLAG_AUTH) {
693 log_msg(VLL_WARN, msg, "Authenticated control message with"
694 " authentication disabled", bfd);
698 if (msg->length != BFD_PACKET_LEN) {
699 log_msg(VLL_WARN, msg, "Unexpected length", bfd);
700 if (msg->length < BFD_PACKET_LEN) {
706 log_msg(VLL_WARN, msg, "Zero multiplier", bfd);
710 if (flags & FLAG_MULTIPOINT) {
711 log_msg(VLL_WARN, msg, "Unsupported multipoint flag", bfd);
716 log_msg(VLL_WARN, msg, "NULL my_disc", bfd);
720 pkt_your_disc = ntohl(msg->your_disc);
722 /* Technically, we should use the your discriminator field to figure
723 * out which 'struct bfd' this packet is destined towards. That way a
724 * bfd session could migrate from one interface to another
725 * transparently. This doesn't fit in with the OVS structure very
726 * well, so in this respect, we are not compliant. */
727 if (pkt_your_disc != bfd->disc) {
728 log_msg(VLL_WARN, msg, "Incorrect your_disc", bfd);
731 } else if (rmt_state > STATE_DOWN) {
732 log_msg(VLL_WARN, msg, "Null your_disc", bfd);
736 bfd->rmt_disc = ntohl(msg->my_disc);
737 bfd->rmt_state = rmt_state;
738 bfd->rmt_flags = flags;
739 bfd->rmt_diag = msg->vers_diag & DIAG_MASK;
741 if (flags & FLAG_FINAL && bfd_in_poll(bfd)) {
742 bfd->min_tx = bfd->poll_min_tx;
743 bfd->min_rx = bfd->poll_min_rx;
744 bfd->flags &= ~FLAG_POLL;
745 log_msg(VLL_INFO, msg, "Poll sequence terminated", bfd);
748 if (flags & FLAG_POLL) {
749 /* RFC 5880 Section 6.5
750 * When the other system receives a Poll, it immediately transmits a
751 * BFD Control packet with the Final (F) bit set, independent of any
752 * periodic BFD Control packets it may be sending
753 * (see section 6.8.7). */
754 bfd->flags &= ~FLAG_POLL;
755 bfd->flags |= FLAG_FINAL;
758 rmt_min_rx = MAX(ntohl(msg->min_rx) / 1000, 1);
759 if (bfd->rmt_min_rx != rmt_min_rx) {
760 bfd->rmt_min_rx = rmt_min_rx;
761 bfd_set_next_tx(bfd);
762 log_msg(VLL_INFO, msg, "New remote min_rx", bfd);
765 bfd->rmt_min_tx = MAX(ntohl(msg->min_tx) / 1000, 1);
766 bfd->detect_time = bfd_rx_interval(bfd) * bfd->mult + time_msec();
768 if (bfd->state == STATE_ADMIN_DOWN) {
769 VLOG_DBG_RL(&rl, "Administratively down, dropping control message.");
773 if (rmt_state == STATE_ADMIN_DOWN) {
774 if (bfd->state != STATE_DOWN) {
775 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
778 switch (bfd->state) {
780 if (rmt_state == STATE_DOWN) {
781 bfd_set_state(bfd, STATE_INIT, bfd->diag);
782 } else if (rmt_state == STATE_INIT) {
783 bfd_set_state(bfd, STATE_UP, bfd->diag);
787 if (rmt_state > STATE_DOWN) {
788 bfd_set_state(bfd, STATE_UP, bfd->diag);
792 if (rmt_state <= STATE_DOWN) {
793 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
794 log_msg(VLL_INFO, msg, "Remote signaled STATE_DOWN", bfd);
797 case STATE_ADMIN_DOWN:
802 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
805 bfd_forwarding__(bfd);
806 ovs_mutex_unlock(&mutex);
809 /* Must be called when the netdev owned by 'bfd' should change. */
811 bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
814 ovs_mutex_lock(&mutex);
815 if (bfd->netdev != netdev) {
816 netdev_close(bfd->netdev);
817 bfd->netdev = netdev_ref(netdev);
818 if (bfd->decay_min_rx && bfd->state == STATE_UP) {
819 bfd_decay_update(bfd);
821 if (bfd->forwarding_if_rx && bfd->state == STATE_UP) {
822 bfd_forwarding_if_rx_update(bfd);
824 bfd->rx_packets = bfd_rx_packets(bfd);
826 ovs_mutex_unlock(&mutex);
830 /* Updates the forwarding flag. If override is not configured and
831 * the forwarding flag value changes, increments the flap count.
833 * Note this function may be called multiple times in a function
834 * (e.g. bfd_account_rx) before and after the bfd state or status
835 * change. This is to capture any forwarding flag flap. */
837 bfd_forwarding__(struct bfd *bfd) OVS_REQUIRES(mutex)
840 bool last_forwarding = bfd->last_forwarding;
842 if (bfd->forwarding_override != -1) {
843 return bfd->forwarding_override == 1;
846 time = bfd->forwarding_if_rx_detect_time;
847 bfd->last_forwarding = (bfd->state == STATE_UP
848 || (bfd->forwarding_if_rx && time > time_msec()))
849 && bfd->rmt_diag != DIAG_PATH_DOWN
850 && bfd->rmt_diag != DIAG_CPATH_DOWN
851 && bfd->rmt_diag != DIAG_RCPATH_DOWN;
852 if (bfd->last_forwarding != last_forwarding) {
855 return bfd->last_forwarding;
860 bfd_in_poll(const struct bfd *bfd) OVS_REQUIRES(mutex)
862 return (bfd->flags & FLAG_POLL) != 0;
866 bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex)
868 if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
869 && !(bfd->flags & FLAG_FINAL)) {
870 bfd->poll_min_tx = bfd->cfg_min_tx;
871 bfd->poll_min_rx = bfd->in_decay ? bfd->decay_min_rx : bfd->cfg_min_rx;
872 bfd->flags |= FLAG_POLL;
874 VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
879 bfd_min_tx(const struct bfd *bfd) OVS_REQUIRES(mutex)
881 /* RFC 5880 Section 6.8.3
882 * When bfd.SessionState is not Up, the system MUST set
883 * bfd.DesiredMinTxInterval to a value of not less than one second
884 * (1,000,000 microseconds). This is intended to ensure that the
885 * bandwidth consumed by BFD sessions that are not Up is negligible,
886 * particularly in the case where a neighbor may not be running BFD. */
887 return (bfd->state == STATE_UP ? bfd->min_tx : MAX(bfd->min_tx, 1000));
891 bfd_tx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
893 long long int interval = bfd_min_tx(bfd);
894 return MAX(interval, bfd->rmt_min_rx);
898 bfd_rx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
900 return MAX(bfd->min_rx, bfd->rmt_min_tx);
904 bfd_set_next_tx(struct bfd *bfd) OVS_REQUIRES(mutex)
906 long long int interval = bfd_tx_interval(bfd);
907 interval -= interval * random_range(26) / 100;
908 bfd->next_tx = bfd->last_tx + interval;
912 bfd_flag_str(enum flags flags)
914 struct ds ds = DS_EMPTY_INITIALIZER;
915 static char flag_str[128];
921 if (flags & FLAG_MULTIPOINT) {
922 ds_put_cstr(&ds, "multipoint ");
925 if (flags & FLAG_DEMAND) {
926 ds_put_cstr(&ds, "demand ");
929 if (flags & FLAG_AUTH) {
930 ds_put_cstr(&ds, "auth ");
933 if (flags & FLAG_CTL) {
934 ds_put_cstr(&ds, "ctl ");
937 if (flags & FLAG_FINAL) {
938 ds_put_cstr(&ds, "final ");
941 if (flags & FLAG_POLL) {
942 ds_put_cstr(&ds, "poll ");
945 /* Do not copy the trailing whitespace. */
947 ovs_strlcpy(flag_str, ds_cstr(&ds), sizeof flag_str);
953 bfd_state_str(enum state state)
956 case STATE_ADMIN_DOWN: return "admin_down";
957 case STATE_DOWN: return "down";
958 case STATE_INIT: return "init";
959 case STATE_UP: return "up";
960 default: return "invalid";
965 bfd_diag_str(enum diag diag) {
967 case DIAG_NONE: return "No Diagnostic";
968 case DIAG_EXPIRED: return "Control Detection Time Expired";
969 case DIAG_ECHO_FAILED: return "Echo Function Failed";
970 case DIAG_RMT_DOWN: return "Neighbor Signaled Session Down";
971 case DIAG_FWD_RESET: return "Forwarding Plane Reset";
972 case DIAG_PATH_DOWN: return "Path Down";
973 case DIAG_CPATH_DOWN: return "Concatenated Path Down";
974 case DIAG_ADMIN_DOWN: return "Administratively Down";
975 case DIAG_RCPATH_DOWN: return "Reverse Concatenated Path Down";
976 default: return "Invalid Diagnostic";
981 log_msg(enum vlog_level level, const struct msg *p, const char *message,
982 const struct bfd *bfd) OVS_REQUIRES(mutex)
984 struct ds ds = DS_EMPTY_INITIALIZER;
986 if (vlog_should_drop(THIS_MODULE, level, &rl)) {
992 "\n\tvers:%"PRIu8" diag:\"%s\" state:%s mult:%"PRIu8
995 "\n\tmy_disc:0x%"PRIx32" your_disc:0x%"PRIx32
996 "\n\tmin_tx:%"PRIu32"us (%"PRIu32"ms)"
997 "\n\tmin_rx:%"PRIu32"us (%"PRIu32"ms)"
998 "\n\tmin_rx_echo:%"PRIu32"us (%"PRIu32"ms)",
999 bfd->name, message, p->vers_diag >> VERS_SHIFT,
1000 bfd_diag_str(p->vers_diag & DIAG_MASK),
1001 bfd_state_str(p->flags & STATE_MASK),
1002 p->mult, p->length, bfd_flag_str(p->flags & FLAGS_MASK),
1003 ntohl(p->my_disc), ntohl(p->your_disc),
1004 ntohl(p->min_tx), ntohl(p->min_tx) / 1000,
1005 ntohl(p->min_rx), ntohl(p->min_rx) / 1000,
1006 ntohl(p->min_rx_echo), ntohl(p->min_rx_echo) / 1000);
1007 bfd_put_details(&ds, bfd);
1008 VLOG(level, "%s", ds_cstr(&ds));
1013 bfd_set_state(struct bfd *bfd, enum state state, enum diag diag)
1016 if (diag == DIAG_NONE && bfd->cpath_down) {
1017 diag = DIAG_CPATH_DOWN;
1020 if (bfd->state != state || bfd->diag != diag) {
1021 if (!VLOG_DROP_INFO(&rl)) {
1022 struct ds ds = DS_EMPTY_INITIALIZER;
1024 ds_put_format(&ds, "%s: BFD state change: %s->%s"
1025 " \"%s\"->\"%s\".\n",
1026 bfd->name, bfd_state_str(bfd->state),
1027 bfd_state_str(state), bfd_diag_str(bfd->diag),
1028 bfd_diag_str(diag));
1029 bfd_put_details(&ds, bfd);
1030 VLOG_INFO("%s", ds_cstr(&ds));
1037 if (bfd->state <= STATE_DOWN) {
1038 bfd->rmt_state = STATE_DOWN;
1039 bfd->rmt_diag = DIAG_NONE;
1040 bfd->rmt_min_rx = 1;
1043 bfd->rmt_min_tx = 0;
1044 /* Resets the min_rx if in_decay. */
1045 if (bfd->in_decay) {
1046 bfd->min_rx = bfd->cfg_min_rx;
1047 bfd->in_decay = false;
1050 /* Resets the decay when state changes to STATE_UP
1051 * and decay_min_rx is configured. */
1052 if (bfd->state == STATE_UP && bfd->decay_min_rx) {
1053 bfd_decay_update(bfd);
1059 bfd_rx_packets(const struct bfd *bfd) OVS_REQUIRES(mutex)
1061 struct netdev_stats stats;
1063 if (!netdev_get_stats(bfd->netdev, &stats)) {
1064 return stats.rx_packets;
1070 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
1071 * the 'expect' value. */
1073 bfd_try_decay(struct bfd *bfd) OVS_REQUIRES(mutex)
1075 int64_t diff, expect;
1077 /* The 'diff' is the difference between current interface rx_packets
1078 * stats and last-time check. The 'expect' is the recorded number of
1079 * bfd control packets received within an approximately decay_min_rx
1080 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
1082 * Since the update of rx_packets stats at interface happens
1083 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1084 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1085 * more wiggle room. */
1086 diff = bfd_rx_packets(bfd) - bfd->decay_rx_packets;
1087 expect = 2 * MAX(bfd->decay_rx_ctl, 1);
1088 bfd->in_decay = diff <= expect ? true : false;
1089 bfd_decay_update(bfd);
1092 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1094 bfd_decay_update(struct bfd * bfd) OVS_REQUIRES(mutex)
1096 bfd->decay_rx_packets = bfd_rx_packets(bfd);
1097 bfd->decay_rx_ctl = 0;
1098 bfd->decay_detect_time = MAX(bfd->decay_min_rx, 2000) + time_msec();
1102 bfd_forwarding_if_rx_update(struct bfd *bfd) OVS_REQUIRES(mutex)
1104 int64_t incr = bfd_rx_interval(bfd) * bfd->mult;
1105 bfd->forwarding_if_rx_detect_time = MAX(incr, 2000) + time_msec();
1109 generate_discriminator(void)
1113 /* RFC 5880 Section 6.8.1
1114 * It SHOULD be set to a random (but still unique) value to improve
1115 * security. The value is otherwise outside the scope of this
1121 /* 'disc' is by definition random, so there's no reason to waste time
1123 disc = random_uint32();
1124 HMAP_FOR_EACH_IN_BUCKET (bfd, node, disc, all_bfds) {
1125 if (bfd->disc == disc) {
1136 bfd_find_by_name(const char *name) OVS_REQUIRES(mutex)
1140 HMAP_FOR_EACH (bfd, node, all_bfds) {
1141 if (!strcmp(bfd->name, name)) {
1149 bfd_put_details(struct ds *ds, const struct bfd *bfd) OVS_REQUIRES(mutex)
1151 ds_put_format(ds, "\tForwarding: %s\n",
1152 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
1153 ? "true" : "false");
1154 ds_put_format(ds, "\tDetect Multiplier: %d\n", bfd->mult);
1155 ds_put_format(ds, "\tConcatenated Path Down: %s\n",
1156 bfd->cpath_down ? "true" : "false");
1157 ds_put_format(ds, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd));
1158 ds_put_format(ds, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd));
1159 ds_put_format(ds, "\tDetect Time: now %+lldms\n",
1160 time_msec() - bfd->detect_time);
1161 ds_put_format(ds, "\tNext TX Time: now %+lldms\n",
1162 time_msec() - bfd->next_tx);
1163 ds_put_format(ds, "\tLast TX Time: now %+lldms\n",
1164 time_msec() - bfd->last_tx);
1166 ds_put_cstr(ds, "\n");
1168 ds_put_format(ds, "\tLocal Flags: %s\n", bfd_flag_str(bfd->flags));
1169 ds_put_format(ds, "\tLocal Session State: %s\n",
1170 bfd_state_str(bfd->state));
1171 ds_put_format(ds, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd->diag));
1172 ds_put_format(ds, "\tLocal Discriminator: 0x%"PRIx32"\n", bfd->disc);
1173 ds_put_format(ds, "\tLocal Minimum TX Interval: %lldms\n",
1175 ds_put_format(ds, "\tLocal Minimum RX Interval: %lldms\n", bfd->min_rx);
1177 ds_put_cstr(ds, "\n");
1179 ds_put_format(ds, "\tRemote Flags: %s\n", bfd_flag_str(bfd->rmt_flags));
1180 ds_put_format(ds, "\tRemote Session State: %s\n",
1181 bfd_state_str(bfd->rmt_state));
1182 ds_put_format(ds, "\tRemote Diagnostic: %s\n",
1183 bfd_diag_str(bfd->rmt_diag));
1184 ds_put_format(ds, "\tRemote Discriminator: 0x%"PRIx32"\n", bfd->rmt_disc);
1185 ds_put_format(ds, "\tRemote Minimum TX Interval: %lldms\n",
1187 ds_put_format(ds, "\tRemote Minimum RX Interval: %lldms\n",
1192 bfd_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1193 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1195 struct ds ds = DS_EMPTY_INITIALIZER;
1198 ovs_mutex_lock(&mutex);
1200 bfd = bfd_find_by_name(argv[1]);
1202 unixctl_command_reply_error(conn, "no such bfd object");
1205 bfd_put_details(&ds, bfd);
1207 HMAP_FOR_EACH (bfd, node, all_bfds) {
1208 ds_put_format(&ds, "---- %s ----\n", bfd->name);
1209 bfd_put_details(&ds, bfd);
1212 unixctl_command_reply(conn, ds_cstr(&ds));
1216 ovs_mutex_unlock(&mutex);
1221 bfd_unixctl_set_forwarding_override(struct unixctl_conn *conn, int argc,
1222 const char *argv[], void *aux OVS_UNUSED)
1225 const char *forward_str = argv[argc - 1];
1226 int forwarding_override;
1229 ovs_mutex_lock(&mutex);
1230 if (!strcasecmp("true", forward_str)) {
1231 forwarding_override = 1;
1232 } else if (!strcasecmp("false", forward_str)) {
1233 forwarding_override = 0;
1234 } else if (!strcasecmp("normal", forward_str)) {
1235 forwarding_override = -1;
1237 unixctl_command_reply_error(conn, "unknown fault string");
1242 bfd = bfd_find_by_name(argv[1]);
1244 unixctl_command_reply_error(conn, "no such BFD object");
1247 bfd->forwarding_override = forwarding_override;
1249 HMAP_FOR_EACH (bfd, node, all_bfds) {
1250 bfd->forwarding_override = forwarding_override;
1254 unixctl_command_reply(conn, "OK");
1257 ovs_mutex_unlock(&mutex);