1 /* Copyright (c) 2013, 2014 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"
24 #include "connectivity.h"
27 #include "dynamic-string.h"
35 #include "ovs-thread.h"
36 #include "openvswitch/types.h"
38 #include "poll-loop.h"
43 #include "unaligned.h"
48 VLOG_DEFINE_THIS_MODULE(bfd);
52 * The goal of this module is to replace CFM with something both more flexible
53 * and standards compliant. In service of this goal, the following needs to be
57 * * Implement Demand mode.
58 * * Go through the RFC line by line and verify we comply.
59 * * Test against a hardware implementation. Preferably a popular one.
60 * * Delete BFD packets with nw_ttl != 255 in the datapath to prevent DOS
65 * - Set TOS/PCP on the outer tunnel header when encapped.
67 * - Sending BFD messages should be in its own thread/process.
69 * - Scale testing. How does it operate when there are large number of bfd
70 * sessions? Do we ever have random flaps? What's the CPU utilization?
72 * - Rely on data traffic for liveness by using BFD demand mode.
73 * If we're receiving traffic on a port, we can safely assume it's up (modulo
74 * unidrectional failures). BFD has a demand mode in which it can stay quiet
75 * unless it feels the need to check the status of the port. Using this, we
76 * can implement a strategy in which BFD only sends control messages on dark
79 * - Depending on how one interprets the spec, it appears that a BFD session
80 * can never change bfd.LocalDiag to "No Diagnostic". We should verify that
81 * this is what hardware implementations actually do. Seems like "No
82 * Diagnostic" should be set once a BFD session state goes UP. */
87 FLAG_MULTIPOINT = 1 << 0,
96 STATE_ADMIN_DOWN = 0 << 6,
103 DIAG_NONE = 0, /* No Diagnostic. */
104 DIAG_EXPIRED = 1, /* Control Detection Time Expired. */
105 DIAG_ECHO_FAILED = 2, /* Echo Function Failed. */
106 DIAG_RMT_DOWN = 3, /* Neighbor Signaled Session Down. */
107 DIAG_FWD_RESET = 4, /* Forwarding Plane Reset. */
108 DIAG_PATH_DOWN = 5, /* Path Down. */
109 DIAG_CPATH_DOWN = 6, /* Concatenated Path Down. */
110 DIAG_ADMIN_DOWN = 7, /* Administratively Down. */
111 DIAG_RCPATH_DOWN = 8 /* Reverse Concatenated Path Down. */
114 /* RFC 5880 Section 4.1
116 * 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
117 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
118 * |Vers | Diag |Sta|P|F|C|A|D|M| Detect Mult | Length |
119 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
120 * | My Discriminator |
121 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
122 * | Your Discriminator |
123 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
124 * | Desired Min TX Interval |
125 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
126 * | Required Min RX Interval |
127 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
128 * | Required Min Echo RX Interval |
129 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ */
131 uint8_t vers_diag; /* Version and diagnostic. */
132 uint8_t flags; /* 2bit State field followed by flags. */
133 uint8_t mult; /* Fault detection multiplier. */
134 uint8_t length; /* Length of this BFD message. */
135 ovs_be32 my_disc; /* My discriminator. */
136 ovs_be32 your_disc; /* Your discriminator. */
137 ovs_be32 min_tx; /* Desired minimum tx interval. */
138 ovs_be32 min_rx; /* Required minimum rx interval. */
139 ovs_be32 min_rx_echo; /* Required minimum echo rx interval. */
141 BUILD_ASSERT_DECL(BFD_PACKET_LEN == sizeof(struct msg));
143 #define DIAG_MASK 0x1f
145 #define STATE_MASK 0xC0
146 #define FLAGS_MASK 0x3f
149 struct hmap_node node; /* In 'all_bfds'. */
150 uint32_t disc; /* bfd.LocalDiscr. Key in 'all_bfds' hmap. */
152 char *name; /* Name used for logging. */
154 bool cpath_down; /* Concatenated Path Down. */
155 uint8_t mult; /* bfd.DetectMult. */
157 struct netdev *netdev;
158 uint64_t rx_packets; /* Packets received by 'netdev'. */
160 enum state state; /* bfd.SessionState. */
161 enum state rmt_state; /* bfd.RemoteSessionState. */
163 enum diag diag; /* bfd.LocalDiag. */
164 enum diag rmt_diag; /* Remote diagnostic. */
166 enum flags flags; /* Flags sent on messages. */
167 enum flags rmt_flags; /* Flags last received. */
169 uint32_t rmt_disc; /* bfd.RemoteDiscr. */
171 uint8_t eth_dst[ETH_ADDR_LEN];/* Ethernet destination address. */
172 bool eth_dst_set; /* 'eth_dst' set through database. */
174 ovs_be32 ip_src; /* IPv4 source address. */
175 ovs_be32 ip_dst; /* IPv4 destination address. */
177 uint16_t udp_src; /* UDP source port. */
179 /* All timers in milliseconds. */
180 long long int rmt_min_rx; /* bfd.RemoteMinRxInterval. */
181 long long int rmt_min_tx; /* Remote minimum TX interval. */
183 long long int cfg_min_tx; /* Configured minimum TX rate. */
184 long long int cfg_min_rx; /* Configured required minimum RX rate. */
185 long long int poll_min_tx; /* Min TX negotating in a poll sequence. */
186 long long int poll_min_rx; /* Min RX negotating in a poll sequence. */
187 long long int min_tx; /* bfd.DesiredMinTxInterval. */
188 long long int min_rx; /* bfd.RequiredMinRxInterval. */
190 long long int last_tx; /* Last TX time. */
191 long long int next_tx; /* Next TX time. */
192 long long int detect_time; /* RFC 5880 6.8.4 Detection time. */
194 bool last_forwarding; /* Last calculation of forwarding flag. */
195 int forwarding_override; /* Manual override of 'forwarding' status. */
197 atomic_bool check_tnl_key; /* Verify tunnel key of inbound packets? */
198 struct ovs_refcount ref_cnt;
200 /* When forward_if_rx is true, bfd_forwarding() will return
201 * true as long as there are incoming packets received.
202 * Note, forwarding_override still has higher priority. */
203 bool forwarding_if_rx;
204 long long int forwarding_if_rx_detect_time;
206 /* BFD decay related variables. */
207 bool in_decay; /* True when bfd is in decay. */
208 int decay_min_rx; /* min_rx is set to decay_min_rx when */
210 int decay_rx_ctl; /* Count bfd packets received within decay */
211 /* detect interval. */
212 uint64_t decay_rx_packets; /* Packets received by 'netdev'. */
213 long long int decay_detect_time; /* Decay detection time. */
215 uint64_t flap_count; /* Counts bfd forwarding flaps. */
217 /* True when the variables returned by bfd_get_status() are changed
218 * since last check. */
222 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
223 static struct hmap all_bfds__ = HMAP_INITIALIZER(&all_bfds__);
224 static struct hmap *const all_bfds OVS_GUARDED_BY(mutex) = &all_bfds__;
226 static bool bfd_lookup_ip(const char *host_name, struct in_addr *)
228 static bool bfd_forwarding__(struct bfd *) OVS_REQUIRES(mutex);
229 static bool bfd_in_poll(const struct bfd *) OVS_REQUIRES(mutex);
230 static void bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex);
231 static const char *bfd_diag_str(enum diag) OVS_REQUIRES(mutex);
232 static const char *bfd_state_str(enum state) OVS_REQUIRES(mutex);
233 static long long int bfd_min_tx(const struct bfd *) OVS_REQUIRES(mutex);
234 static long long int bfd_tx_interval(const struct bfd *)
236 static long long int bfd_rx_interval(const struct bfd *)
238 static void bfd_set_next_tx(struct bfd *) OVS_REQUIRES(mutex);
239 static void bfd_set_state(struct bfd *, enum state, enum diag)
241 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex);
242 static void bfd_put_details(struct ds *, const struct bfd *)
244 static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQUIRES(mutex);
245 static void bfd_try_decay(struct bfd *) OVS_REQUIRES(mutex);
246 static void bfd_decay_update(struct bfd *) OVS_REQUIRES(mutex);
247 static void bfd_status_changed(struct bfd *) OVS_REQUIRES(mutex);
249 static void bfd_forwarding_if_rx_update(struct bfd *) OVS_REQUIRES(mutex);
250 static void bfd_unixctl_show(struct unixctl_conn *, int argc,
251 const char *argv[], void *aux OVS_UNUSED);
252 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
253 int argc, const char *argv[],
254 void *aux OVS_UNUSED);
255 static void log_msg(enum vlog_level, const struct msg *, const char *message,
256 const struct bfd *) OVS_REQUIRES(mutex);
258 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 20);
260 /* Returns true if the interface on which 'bfd' is running may be used to
261 * forward traffic according to the BFD session state. */
263 bfd_forwarding(struct bfd *bfd) OVS_EXCLUDED(mutex)
267 ovs_mutex_lock(&mutex);
268 ret = bfd_forwarding__(bfd);
269 ovs_mutex_unlock(&mutex);
273 /* When forwarding_if_rx is enabled, if there are packets received,
274 * updates forwarding_if_rx_detect_time. */
276 bfd_account_rx(struct bfd *bfd, const struct dpif_flow_stats *stats)
278 if (stats->n_packets && bfd->forwarding_if_rx) {
279 ovs_mutex_lock(&mutex);
280 bfd_forwarding__(bfd);
281 bfd_forwarding_if_rx_update(bfd);
282 bfd_forwarding__(bfd);
283 ovs_mutex_unlock(&mutex);
287 /* Returns and resets the 'bfd->status_changed'. */
289 bfd_check_status_change(struct bfd *bfd) OVS_EXCLUDED(mutex)
293 ovs_mutex_lock(&mutex);
294 ret = bfd->status_changed;
295 bfd->status_changed = false;
296 ovs_mutex_unlock(&mutex);
301 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
302 * intended for the OVS database. */
304 bfd_get_status(const struct bfd *bfd, struct smap *smap)
307 ovs_mutex_lock(&mutex);
308 smap_add(smap, "forwarding",
309 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
311 smap_add(smap, "state", bfd_state_str(bfd->state));
312 smap_add(smap, "diagnostic", bfd_diag_str(bfd->diag));
313 smap_add_format(smap, "flap_count", "%"PRIu64, bfd->flap_count);
315 if (bfd->state != STATE_DOWN) {
316 smap_add(smap, "remote_state", bfd_state_str(bfd->rmt_state));
317 smap_add(smap, "remote_diagnostic", bfd_diag_str(bfd->rmt_diag));
319 ovs_mutex_unlock(&mutex);
322 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
323 * according to the database configuration contained in 'cfg'. Takes ownership
324 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
325 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
326 * Also returns NULL if cfg is NULL. */
328 bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
329 struct netdev *netdev) OVS_EXCLUDED(mutex)
331 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
332 static atomic_uint16_t udp_src = ATOMIC_VAR_INIT(0);
335 long long int min_tx, min_rx;
336 bool need_poll = false;
337 bool cfg_min_rx_changed = false;
338 bool cpath_down, forwarding_if_rx;
339 const char *hwaddr, *ip_src, *ip_dst;
340 struct in_addr in_addr;
341 uint8_t ea[ETH_ADDR_LEN];
343 if (ovsthread_once_start(&once)) {
344 unixctl_command_register("bfd/show", "[interface]", 0, 1,
345 bfd_unixctl_show, NULL);
346 unixctl_command_register("bfd/set-forwarding",
347 "[interface] normal|false|true", 1, 2,
348 bfd_unixctl_set_forwarding_override, NULL);
349 ovsthread_once_done(&once);
352 if (!cfg || !smap_get_bool(cfg, "enable", false)) {
357 ovs_mutex_lock(&mutex);
359 bfd = xzalloc(sizeof *bfd);
360 bfd->name = xstrdup(name);
361 bfd->forwarding_override = -1;
362 bfd->disc = generate_discriminator();
363 hmap_insert(all_bfds, &bfd->node, bfd->disc);
365 bfd->diag = DIAG_NONE;
368 ovs_refcount_init(&bfd->ref_cnt);
369 bfd->netdev = netdev_ref(netdev);
370 bfd->rx_packets = bfd_rx_packets(bfd);
371 bfd->in_decay = false;
374 /* RFC 5881 section 4
375 * The source port MUST be in the range 49152 through 65535. The same
376 * UDP source port number MUST be used for all BFD Control packets
377 * associated with a particular session. The source port number SHOULD
378 * be unique among all BFD sessions on the system. */
379 atomic_add(&udp_src, 1, &bfd->udp_src);
380 bfd->udp_src = (bfd->udp_src % 16384) + 49152;
382 bfd_set_state(bfd, STATE_DOWN, DIAG_NONE);
384 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
387 atomic_store(&bfd->check_tnl_key,
388 smap_get_bool(cfg, "check_tnl_key", false));
389 min_tx = smap_get_int(cfg, "min_tx", 100);
390 min_tx = MAX(min_tx, 100);
391 if (bfd->cfg_min_tx != min_tx) {
392 bfd->cfg_min_tx = min_tx;
393 if (bfd->state != STATE_UP
394 || (!bfd_in_poll(bfd) && bfd->cfg_min_tx < bfd->min_tx)) {
395 bfd->min_tx = bfd->cfg_min_tx;
400 min_rx = smap_get_int(cfg, "min_rx", 1000);
401 min_rx = MAX(min_rx, 100);
402 if (bfd->cfg_min_rx != min_rx) {
403 bfd->cfg_min_rx = min_rx;
404 if (bfd->state != STATE_UP
405 || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
406 bfd->min_rx = bfd->cfg_min_rx;
408 cfg_min_rx_changed = true;
412 decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
413 if (bfd->decay_min_rx != decay_min_rx || cfg_min_rx_changed) {
414 if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
415 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
416 bfd->name, bfd->cfg_min_rx);
417 bfd->decay_min_rx = 0;
419 bfd->decay_min_rx = decay_min_rx;
422 bfd->in_decay = false;
423 bfd_decay_update(bfd);
427 cpath_down = smap_get_bool(cfg, "cpath_down", false);
428 if (bfd->cpath_down != cpath_down) {
429 bfd->cpath_down = cpath_down;
430 bfd_set_state(bfd, bfd->state, DIAG_NONE);
434 hwaddr = smap_get(cfg, "bfd_dst_mac");
435 if (hwaddr && eth_addr_from_string(hwaddr, ea) && !eth_addr_is_zero(ea)) {
436 memcpy(bfd->eth_dst, ea, ETH_ADDR_LEN);
437 bfd->eth_dst_set = true;
438 } else if (bfd->eth_dst_set) {
439 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
440 bfd->eth_dst_set = false;
443 ip_src = smap_get(cfg, "bfd_src_ip");
444 if (ip_src && bfd_lookup_ip(ip_src, &in_addr)) {
445 memcpy(&bfd->ip_src, &in_addr, sizeof in_addr);
447 bfd->ip_src = htonl(0xA9FE0100); /* 169.254.1.0. */
450 ip_dst = smap_get(cfg, "bfd_dst_ip");
451 if (ip_dst && bfd_lookup_ip(ip_dst, &in_addr)) {
452 memcpy(&bfd->ip_dst, &in_addr, sizeof in_addr);
454 bfd->ip_dst = htonl(0xA9FE0101); /* 169.254.1.1. */
457 forwarding_if_rx = smap_get_bool(cfg, "forwarding_if_rx", false);
458 if (bfd->forwarding_if_rx != forwarding_if_rx) {
459 bfd->forwarding_if_rx = forwarding_if_rx;
460 if (bfd->state == STATE_UP && bfd->forwarding_if_rx) {
461 bfd_forwarding_if_rx_update(bfd);
463 bfd->forwarding_if_rx_detect_time = 0;
470 ovs_mutex_unlock(&mutex);
475 bfd_ref(const struct bfd *bfd_)
477 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
479 ovs_refcount_ref(&bfd->ref_cnt);
485 bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
487 if (bfd && ovs_refcount_unref(&bfd->ref_cnt) == 1) {
488 ovs_mutex_lock(&mutex);
489 hmap_remove(all_bfds, &bfd->node);
490 netdev_close(bfd->netdev);
493 ovs_mutex_unlock(&mutex);
498 bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
500 poll_timer_wait_until(bfd_wake_time(bfd));
503 /* Returns the next wake up time. */
505 bfd_wake_time(const struct bfd *bfd) OVS_EXCLUDED(mutex)
507 long long int retval;
513 ovs_mutex_lock(&mutex);
514 if (bfd->flags & FLAG_FINAL) {
517 retval = bfd->next_tx;
518 if (bfd->state > STATE_DOWN) {
519 retval = MIN(bfd->detect_time, retval);
522 ovs_mutex_unlock(&mutex);
527 bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
532 ovs_mutex_lock(&mutex);
534 old_in_decay = bfd->in_decay;
536 if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
537 bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
539 bfd_forwarding__(bfd);
541 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
542 * configured, and decay_detect_time is reached. */
543 if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
544 && now >= bfd->decay_detect_time) {
548 if (bfd->min_tx != bfd->cfg_min_tx
549 || (bfd->min_rx != bfd->cfg_min_rx && bfd->min_rx != bfd->decay_min_rx)
550 || bfd->in_decay != old_in_decay) {
553 ovs_mutex_unlock(&mutex);
557 bfd_should_send_packet(const struct bfd *bfd) OVS_EXCLUDED(mutex)
560 ovs_mutex_lock(&mutex);
561 ret = bfd->flags & FLAG_FINAL || time_msec() >= bfd->next_tx;
562 ovs_mutex_unlock(&mutex);
567 bfd_put_packet(struct bfd *bfd, struct ofpbuf *p,
568 uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
570 long long int min_tx, min_rx;
571 struct udp_header *udp;
572 struct eth_header *eth;
573 struct ip_header *ip;
576 ovs_mutex_lock(&mutex);
578 long long int delay = time_msec() - bfd->next_tx;
579 long long int interval = bfd_tx_interval(bfd);
580 if (delay > interval * 3 / 2) {
581 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
582 " control message", bfd->name, delay, interval);
586 /* RFC 5880 Section 6.5
587 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
589 ovs_assert(!(bfd->flags & FLAG_POLL) || !(bfd->flags & FLAG_FINAL));
591 ofpbuf_reserve(p, 2); /* Properly align after the ethernet header. */
592 eth = ofpbuf_put_uninit(p, sizeof *eth);
593 memcpy(eth->eth_src, eth_src, ETH_ADDR_LEN);
594 memcpy(eth->eth_dst, bfd->eth_dst, ETH_ADDR_LEN);
595 eth->eth_type = htons(ETH_TYPE_IP);
597 ip = ofpbuf_put_zeros(p, sizeof *ip);
598 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
599 ip->ip_tot_len = htons(sizeof *ip + sizeof *udp + sizeof *msg);
601 ip->ip_tos = IPTOS_LOWDELAY | IPTOS_THROUGHPUT;
602 ip->ip_proto = IPPROTO_UDP;
603 put_16aligned_be32(&ip->ip_src, bfd->ip_src);
604 put_16aligned_be32(&ip->ip_dst, bfd->ip_dst);
605 ip->ip_csum = csum(ip, sizeof *ip);
607 udp = ofpbuf_put_zeros(p, sizeof *udp);
608 udp->udp_src = htons(bfd->udp_src);
609 udp->udp_dst = htons(BFD_DEST_PORT);
610 udp->udp_len = htons(sizeof *udp + sizeof *msg);
612 msg = ofpbuf_put_uninit(p, sizeof *msg);
613 msg->vers_diag = (BFD_VERSION << 5) | bfd->diag;
614 msg->flags = (bfd->state & STATE_MASK) | bfd->flags;
616 msg->mult = bfd->mult;
617 msg->length = BFD_PACKET_LEN;
618 msg->my_disc = htonl(bfd->disc);
619 msg->your_disc = htonl(bfd->rmt_disc);
620 msg->min_rx_echo = htonl(0);
622 if (bfd_in_poll(bfd)) {
623 min_tx = bfd->poll_min_tx;
624 min_rx = bfd->poll_min_rx;
626 min_tx = bfd_min_tx(bfd);
627 min_rx = bfd->min_rx;
630 msg->min_tx = htonl(min_tx * 1000);
631 msg->min_rx = htonl(min_rx * 1000);
633 bfd->flags &= ~FLAG_FINAL;
635 log_msg(VLL_DBG, msg, "Sending BFD Message", bfd);
637 bfd->last_tx = time_msec();
638 bfd_set_next_tx(bfd);
639 ovs_mutex_unlock(&mutex);
643 bfd_should_process_flow(const struct bfd *bfd_, const struct flow *flow,
644 struct flow_wildcards *wc)
646 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
649 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
650 if (bfd->eth_dst_set && memcmp(bfd->eth_dst, flow->dl_dst, ETH_ADDR_LEN)) {
654 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
655 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
657 atomic_read(&bfd->check_tnl_key, &check_tnl_key);
659 memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id);
661 return (flow->dl_type == htons(ETH_TYPE_IP)
662 && flow->nw_proto == IPPROTO_UDP
663 && flow->tp_dst == htons(BFD_DEST_PORT)
664 && (!check_tnl_key || flow->tunnel.tun_id == htonll(0)));
668 bfd_process_packet(struct bfd *bfd, const struct flow *flow,
669 const struct ofpbuf *p) OVS_EXCLUDED(mutex)
671 uint32_t rmt_min_rx, pkt_your_disc;
672 enum state rmt_state;
676 const uint8_t *l7 = ofpbuf_get_udp_payload(p);
679 return; /* No UDP payload. */
682 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
684 ovs_mutex_lock(&mutex);
685 /* Increments the decay rx counter. */
688 bfd_forwarding__(bfd);
690 if (flow->nw_ttl != 255) {
691 /* XXX Should drop in the kernel to prevent DOS. */
695 msg = ofpbuf_at(p, l7 - (uint8_t *)ofpbuf_data(p), BFD_PACKET_LEN);
697 VLOG_INFO_RL(&rl, "%s: Received too-short BFD control message (only "
698 "%"PRIdPTR" bytes long, at least %d required).",
699 bfd->name, (uint8_t *) ofpbuf_tail(p) - l7,
704 /* RFC 5880 Section 6.8.6
705 * If the Length field is greater than the payload of the encapsulating
706 * protocol, the packet MUST be discarded.
708 * Note that we make this check implicity. Above we use ofpbuf_at() to
709 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
710 * the encapsulating protocol. Below we require msg->length to be exactly
711 * BFD_PACKET_LEN bytes. */
713 flags = msg->flags & FLAGS_MASK;
714 rmt_state = msg->flags & STATE_MASK;
715 version = msg->vers_diag >> VERS_SHIFT;
717 log_msg(VLL_DBG, msg, "Received BFD control message", bfd);
719 if (version != BFD_VERSION) {
720 log_msg(VLL_WARN, msg, "Incorrect version", bfd);
724 /* Technically this should happen after the length check. We don't support
725 * authentication however, so it's simpler to do the check first. */
726 if (flags & FLAG_AUTH) {
727 log_msg(VLL_WARN, msg, "Authenticated control message with"
728 " authentication disabled", bfd);
732 if (msg->length != BFD_PACKET_LEN) {
733 log_msg(VLL_WARN, msg, "Unexpected length", bfd);
734 if (msg->length < BFD_PACKET_LEN) {
740 log_msg(VLL_WARN, msg, "Zero multiplier", bfd);
744 if (flags & FLAG_MULTIPOINT) {
745 log_msg(VLL_WARN, msg, "Unsupported multipoint flag", bfd);
750 log_msg(VLL_WARN, msg, "NULL my_disc", bfd);
754 pkt_your_disc = ntohl(msg->your_disc);
756 /* Technically, we should use the your discriminator field to figure
757 * out which 'struct bfd' this packet is destined towards. That way a
758 * bfd session could migrate from one interface to another
759 * transparently. This doesn't fit in with the OVS structure very
760 * well, so in this respect, we are not compliant. */
761 if (pkt_your_disc != bfd->disc) {
762 log_msg(VLL_WARN, msg, "Incorrect your_disc", bfd);
765 } else if (rmt_state > STATE_DOWN) {
766 log_msg(VLL_WARN, msg, "Null your_disc", bfd);
770 if (bfd->rmt_state != rmt_state) {
771 bfd_status_changed(bfd);
774 bfd->rmt_disc = ntohl(msg->my_disc);
775 bfd->rmt_state = rmt_state;
776 bfd->rmt_flags = flags;
777 bfd->rmt_diag = msg->vers_diag & DIAG_MASK;
779 if (flags & FLAG_FINAL && bfd_in_poll(bfd)) {
780 bfd->min_tx = bfd->poll_min_tx;
781 bfd->min_rx = bfd->poll_min_rx;
782 bfd->flags &= ~FLAG_POLL;
783 log_msg(VLL_INFO, msg, "Poll sequence terminated", bfd);
786 if (flags & FLAG_POLL) {
787 /* RFC 5880 Section 6.5
788 * When the other system receives a Poll, it immediately transmits a
789 * BFD Control packet with the Final (F) bit set, independent of any
790 * periodic BFD Control packets it may be sending
791 * (see section 6.8.7). */
792 bfd->flags &= ~FLAG_POLL;
793 bfd->flags |= FLAG_FINAL;
796 rmt_min_rx = MAX(ntohl(msg->min_rx) / 1000, 1);
797 if (bfd->rmt_min_rx != rmt_min_rx) {
798 bfd->rmt_min_rx = rmt_min_rx;
800 bfd_set_next_tx(bfd);
802 log_msg(VLL_INFO, msg, "New remote min_rx", bfd);
805 bfd->rmt_min_tx = MAX(ntohl(msg->min_tx) / 1000, 1);
806 bfd->detect_time = bfd_rx_interval(bfd) * bfd->mult + time_msec();
808 if (bfd->state == STATE_ADMIN_DOWN) {
809 VLOG_DBG_RL(&rl, "Administratively down, dropping control message.");
813 if (rmt_state == STATE_ADMIN_DOWN) {
814 if (bfd->state != STATE_DOWN) {
815 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
818 switch (bfd->state) {
820 if (rmt_state == STATE_DOWN) {
821 bfd_set_state(bfd, STATE_INIT, bfd->diag);
822 } else if (rmt_state == STATE_INIT) {
823 bfd_set_state(bfd, STATE_UP, bfd->diag);
827 if (rmt_state > STATE_DOWN) {
828 bfd_set_state(bfd, STATE_UP, bfd->diag);
832 if (rmt_state <= STATE_DOWN) {
833 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
834 log_msg(VLL_INFO, msg, "Remote signaled STATE_DOWN", bfd);
837 case STATE_ADMIN_DOWN:
842 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
845 bfd_forwarding__(bfd);
846 ovs_mutex_unlock(&mutex);
849 /* Must be called when the netdev owned by 'bfd' should change. */
851 bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
854 ovs_mutex_lock(&mutex);
855 if (bfd->netdev != netdev) {
856 netdev_close(bfd->netdev);
857 bfd->netdev = netdev_ref(netdev);
858 if (bfd->decay_min_rx && bfd->state == STATE_UP) {
859 bfd_decay_update(bfd);
861 if (bfd->forwarding_if_rx && bfd->state == STATE_UP) {
862 bfd_forwarding_if_rx_update(bfd);
864 bfd->rx_packets = bfd_rx_packets(bfd);
866 ovs_mutex_unlock(&mutex);
870 /* Updates the forwarding flag. If override is not configured and
871 * the forwarding flag value changes, increments the flap count.
873 * Note this function may be called multiple times in a function
874 * (e.g. bfd_account_rx) before and after the bfd state or status
875 * change. This is to capture any forwarding flag flap. */
877 bfd_forwarding__(struct bfd *bfd) OVS_REQUIRES(mutex)
880 bool last_forwarding = bfd->last_forwarding;
882 if (bfd->forwarding_override != -1) {
883 return bfd->forwarding_override == 1;
886 time = bfd->forwarding_if_rx_detect_time;
887 bfd->last_forwarding = (bfd->state == STATE_UP
888 || (bfd->forwarding_if_rx && time > time_msec()))
889 && bfd->rmt_diag != DIAG_PATH_DOWN
890 && bfd->rmt_diag != DIAG_CPATH_DOWN
891 && bfd->rmt_diag != DIAG_RCPATH_DOWN;
892 if (bfd->last_forwarding != last_forwarding) {
894 bfd_status_changed(bfd);
896 return bfd->last_forwarding;
901 bfd_lookup_ip(const char *host_name, struct in_addr *addr)
903 if (!inet_pton(AF_INET, host_name, addr)) {
904 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
911 bfd_in_poll(const struct bfd *bfd) OVS_REQUIRES(mutex)
913 return (bfd->flags & FLAG_POLL) != 0;
917 bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex)
919 if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
920 && !(bfd->flags & FLAG_FINAL)) {
921 bfd->poll_min_tx = bfd->cfg_min_tx;
922 bfd->poll_min_rx = bfd->in_decay ? bfd->decay_min_rx : bfd->cfg_min_rx;
923 bfd->flags |= FLAG_POLL;
925 VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
930 bfd_min_tx(const struct bfd *bfd) OVS_REQUIRES(mutex)
932 /* RFC 5880 Section 6.8.3
933 * When bfd.SessionState is not Up, the system MUST set
934 * bfd.DesiredMinTxInterval to a value of not less than one second
935 * (1,000,000 microseconds). This is intended to ensure that the
936 * bandwidth consumed by BFD sessions that are not Up is negligible,
937 * particularly in the case where a neighbor may not be running BFD. */
938 return (bfd->state == STATE_UP ? bfd->min_tx : MAX(bfd->min_tx, 1000));
942 bfd_tx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
944 long long int interval = bfd_min_tx(bfd);
945 return MAX(interval, bfd->rmt_min_rx);
949 bfd_rx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
951 return MAX(bfd->min_rx, bfd->rmt_min_tx);
955 bfd_set_next_tx(struct bfd *bfd) OVS_REQUIRES(mutex)
957 long long int interval = bfd_tx_interval(bfd);
958 interval -= interval * random_range(26) / 100;
959 bfd->next_tx = bfd->last_tx + interval;
963 bfd_flag_str(enum flags flags)
965 struct ds ds = DS_EMPTY_INITIALIZER;
966 static char flag_str[128];
972 if (flags & FLAG_MULTIPOINT) {
973 ds_put_cstr(&ds, "multipoint ");
976 if (flags & FLAG_DEMAND) {
977 ds_put_cstr(&ds, "demand ");
980 if (flags & FLAG_AUTH) {
981 ds_put_cstr(&ds, "auth ");
984 if (flags & FLAG_CTL) {
985 ds_put_cstr(&ds, "ctl ");
988 if (flags & FLAG_FINAL) {
989 ds_put_cstr(&ds, "final ");
992 if (flags & FLAG_POLL) {
993 ds_put_cstr(&ds, "poll ");
996 /* Do not copy the trailing whitespace. */
998 ovs_strlcpy(flag_str, ds_cstr(&ds), sizeof flag_str);
1004 bfd_state_str(enum state state)
1007 case STATE_ADMIN_DOWN: return "admin_down";
1008 case STATE_DOWN: return "down";
1009 case STATE_INIT: return "init";
1010 case STATE_UP: return "up";
1011 default: return "invalid";
1016 bfd_diag_str(enum diag diag) {
1018 case DIAG_NONE: return "No Diagnostic";
1019 case DIAG_EXPIRED: return "Control Detection Time Expired";
1020 case DIAG_ECHO_FAILED: return "Echo Function Failed";
1021 case DIAG_RMT_DOWN: return "Neighbor Signaled Session Down";
1022 case DIAG_FWD_RESET: return "Forwarding Plane Reset";
1023 case DIAG_PATH_DOWN: return "Path Down";
1024 case DIAG_CPATH_DOWN: return "Concatenated Path Down";
1025 case DIAG_ADMIN_DOWN: return "Administratively Down";
1026 case DIAG_RCPATH_DOWN: return "Reverse Concatenated Path Down";
1027 default: return "Invalid Diagnostic";
1032 log_msg(enum vlog_level level, const struct msg *p, const char *message,
1033 const struct bfd *bfd) OVS_REQUIRES(mutex)
1035 struct ds ds = DS_EMPTY_INITIALIZER;
1037 if (vlog_should_drop(THIS_MODULE, level, &rl)) {
1043 "\n\tvers:%"PRIu8" diag:\"%s\" state:%s mult:%"PRIu8
1046 "\n\tmy_disc:0x%"PRIx32" your_disc:0x%"PRIx32
1047 "\n\tmin_tx:%"PRIu32"us (%"PRIu32"ms)"
1048 "\n\tmin_rx:%"PRIu32"us (%"PRIu32"ms)"
1049 "\n\tmin_rx_echo:%"PRIu32"us (%"PRIu32"ms)",
1050 bfd->name, message, p->vers_diag >> VERS_SHIFT,
1051 bfd_diag_str(p->vers_diag & DIAG_MASK),
1052 bfd_state_str(p->flags & STATE_MASK),
1053 p->mult, p->length, bfd_flag_str(p->flags & FLAGS_MASK),
1054 ntohl(p->my_disc), ntohl(p->your_disc),
1055 ntohl(p->min_tx), ntohl(p->min_tx) / 1000,
1056 ntohl(p->min_rx), ntohl(p->min_rx) / 1000,
1057 ntohl(p->min_rx_echo), ntohl(p->min_rx_echo) / 1000);
1058 bfd_put_details(&ds, bfd);
1059 VLOG(level, "%s", ds_cstr(&ds));
1064 bfd_set_state(struct bfd *bfd, enum state state, enum diag diag)
1067 if (bfd->cpath_down) {
1068 diag = DIAG_CPATH_DOWN;
1071 if (bfd->state != state || bfd->diag != diag) {
1072 if (!VLOG_DROP_INFO(&rl)) {
1073 struct ds ds = DS_EMPTY_INITIALIZER;
1075 ds_put_format(&ds, "%s: BFD state change: %s->%s"
1076 " \"%s\"->\"%s\".\n",
1077 bfd->name, bfd_state_str(bfd->state),
1078 bfd_state_str(state), bfd_diag_str(bfd->diag),
1079 bfd_diag_str(diag));
1080 bfd_put_details(&ds, bfd);
1081 VLOG_INFO("%s", ds_cstr(&ds));
1088 if (bfd->state <= STATE_DOWN) {
1089 bfd->rmt_state = STATE_DOWN;
1090 bfd->rmt_diag = DIAG_NONE;
1091 bfd->rmt_min_rx = 1;
1094 bfd->rmt_min_tx = 0;
1095 /* Resets the min_rx if in_decay. */
1096 if (bfd->in_decay) {
1097 bfd->min_rx = bfd->cfg_min_rx;
1098 bfd->in_decay = false;
1101 /* Resets the decay when state changes to STATE_UP
1102 * and decay_min_rx is configured. */
1103 if (bfd->state == STATE_UP && bfd->decay_min_rx) {
1104 bfd_decay_update(bfd);
1107 bfd_status_changed(bfd);
1112 bfd_rx_packets(const struct bfd *bfd) OVS_REQUIRES(mutex)
1114 struct netdev_stats stats;
1116 if (!netdev_get_stats(bfd->netdev, &stats)) {
1117 return stats.rx_packets;
1123 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
1124 * the 'expect' value. */
1126 bfd_try_decay(struct bfd *bfd) OVS_REQUIRES(mutex)
1128 int64_t diff, expect;
1130 /* The 'diff' is the difference between current interface rx_packets
1131 * stats and last-time check. The 'expect' is the recorded number of
1132 * bfd control packets received within an approximately decay_min_rx
1133 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
1135 * Since the update of rx_packets stats at interface happens
1136 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1137 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1138 * more wiggle room. */
1139 diff = bfd_rx_packets(bfd) - bfd->decay_rx_packets;
1140 expect = 2 * MAX(bfd->decay_rx_ctl, 1);
1141 bfd->in_decay = diff <= expect ? true : false;
1142 bfd_decay_update(bfd);
1145 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1147 bfd_decay_update(struct bfd * bfd) OVS_REQUIRES(mutex)
1149 bfd->decay_rx_packets = bfd_rx_packets(bfd);
1150 bfd->decay_rx_ctl = 0;
1151 bfd->decay_detect_time = MAX(bfd->decay_min_rx, 2000) + time_msec();
1154 /* Records the status change and changes the global connectivity seq. */
1156 bfd_status_changed(struct bfd *bfd) OVS_REQUIRES(mutex)
1158 seq_change(connectivity_seq_get());
1159 bfd->status_changed = true;
1163 bfd_forwarding_if_rx_update(struct bfd *bfd) OVS_REQUIRES(mutex)
1165 int64_t incr = bfd_rx_interval(bfd) * bfd->mult;
1166 bfd->forwarding_if_rx_detect_time = MAX(incr, 2000) + time_msec();
1170 generate_discriminator(void)
1174 /* RFC 5880 Section 6.8.1
1175 * It SHOULD be set to a random (but still unique) value to improve
1176 * security. The value is otherwise outside the scope of this
1182 /* 'disc' is by definition random, so there's no reason to waste time
1184 disc = random_uint32();
1185 HMAP_FOR_EACH_IN_BUCKET (bfd, node, disc, all_bfds) {
1186 if (bfd->disc == disc) {
1197 bfd_find_by_name(const char *name) OVS_REQUIRES(mutex)
1201 HMAP_FOR_EACH (bfd, node, all_bfds) {
1202 if (!strcmp(bfd->name, name)) {
1210 bfd_put_details(struct ds *ds, const struct bfd *bfd) OVS_REQUIRES(mutex)
1212 ds_put_format(ds, "\tForwarding: %s\n",
1213 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
1214 ? "true" : "false");
1215 ds_put_format(ds, "\tDetect Multiplier: %d\n", bfd->mult);
1216 ds_put_format(ds, "\tConcatenated Path Down: %s\n",
1217 bfd->cpath_down ? "true" : "false");
1218 ds_put_format(ds, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd));
1219 ds_put_format(ds, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd));
1220 ds_put_format(ds, "\tDetect Time: now %+lldms\n",
1221 time_msec() - bfd->detect_time);
1222 ds_put_format(ds, "\tNext TX Time: now %+lldms\n",
1223 time_msec() - bfd->next_tx);
1224 ds_put_format(ds, "\tLast TX Time: now %+lldms\n",
1225 time_msec() - bfd->last_tx);
1227 ds_put_cstr(ds, "\n");
1229 ds_put_format(ds, "\tLocal Flags: %s\n", bfd_flag_str(bfd->flags));
1230 ds_put_format(ds, "\tLocal Session State: %s\n",
1231 bfd_state_str(bfd->state));
1232 ds_put_format(ds, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd->diag));
1233 ds_put_format(ds, "\tLocal Discriminator: 0x%"PRIx32"\n", bfd->disc);
1234 ds_put_format(ds, "\tLocal Minimum TX Interval: %lldms\n",
1236 ds_put_format(ds, "\tLocal Minimum RX Interval: %lldms\n", bfd->min_rx);
1238 ds_put_cstr(ds, "\n");
1240 ds_put_format(ds, "\tRemote Flags: %s\n", bfd_flag_str(bfd->rmt_flags));
1241 ds_put_format(ds, "\tRemote Session State: %s\n",
1242 bfd_state_str(bfd->rmt_state));
1243 ds_put_format(ds, "\tRemote Diagnostic: %s\n",
1244 bfd_diag_str(bfd->rmt_diag));
1245 ds_put_format(ds, "\tRemote Discriminator: 0x%"PRIx32"\n", bfd->rmt_disc);
1246 ds_put_format(ds, "\tRemote Minimum TX Interval: %lldms\n",
1248 ds_put_format(ds, "\tRemote Minimum RX Interval: %lldms\n",
1253 bfd_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1254 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1256 struct ds ds = DS_EMPTY_INITIALIZER;
1259 ovs_mutex_lock(&mutex);
1261 bfd = bfd_find_by_name(argv[1]);
1263 unixctl_command_reply_error(conn, "no such bfd object");
1266 bfd_put_details(&ds, bfd);
1268 HMAP_FOR_EACH (bfd, node, all_bfds) {
1269 ds_put_format(&ds, "---- %s ----\n", bfd->name);
1270 bfd_put_details(&ds, bfd);
1273 unixctl_command_reply(conn, ds_cstr(&ds));
1277 ovs_mutex_unlock(&mutex);
1282 bfd_unixctl_set_forwarding_override(struct unixctl_conn *conn, int argc,
1283 const char *argv[], void *aux OVS_UNUSED)
1286 const char *forward_str = argv[argc - 1];
1287 int forwarding_override;
1290 ovs_mutex_lock(&mutex);
1291 if (!strcasecmp("true", forward_str)) {
1292 forwarding_override = 1;
1293 } else if (!strcasecmp("false", forward_str)) {
1294 forwarding_override = 0;
1295 } else if (!strcasecmp("normal", forward_str)) {
1296 forwarding_override = -1;
1298 unixctl_command_reply_error(conn, "unknown fault string");
1303 bfd = bfd_find_by_name(argv[1]);
1305 unixctl_command_reply_error(conn, "no such BFD object");
1308 bfd->forwarding_override = forwarding_override;
1309 bfd_status_changed(bfd);
1311 HMAP_FOR_EACH (bfd, node, all_bfds) {
1312 bfd->forwarding_override = forwarding_override;
1313 bfd_status_changed(bfd);
1317 unixctl_command_reply(conn, "OK");
1320 ovs_mutex_unlock(&mutex);