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 /* BFD decay related variables. */
195 bool in_decay; /* True when bfd is in decay. */
196 int decay_min_rx; /* min_rx is set to decay_min_rx when */
198 int decay_rx_ctl; /* Count bfd packets received within decay */
199 /* detect interval. */
200 long long int decay_detect_time; /* Decay detection time. */
203 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
204 static struct hmap all_bfds__ = HMAP_INITIALIZER(&all_bfds__);
205 static struct hmap *const all_bfds OVS_GUARDED_BY(mutex) = &all_bfds__;
207 static bool bfd_forwarding__(const struct bfd *) OVS_REQUIRES(mutex);
208 static bool bfd_in_poll(const struct bfd *) OVS_REQUIRES(mutex);
209 static void bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex);
210 static const char *bfd_diag_str(enum diag) OVS_REQUIRES(mutex);
211 static const char *bfd_state_str(enum state) OVS_REQUIRES(mutex);
212 static long long int bfd_min_tx(const struct bfd *) OVS_REQUIRES(mutex);
213 static long long int bfd_tx_interval(const struct bfd *)
215 static long long int bfd_rx_interval(const struct bfd *)
217 static void bfd_set_next_tx(struct bfd *) OVS_REQUIRES(mutex);
218 static void bfd_set_state(struct bfd *, enum state, enum diag)
220 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex);
221 static void bfd_put_details(struct ds *, const struct bfd *)
223 static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQUIRES(mutex);
224 static void bfd_try_decay(struct bfd *) OVS_REQUIRES(mutex);
225 static void bfd_decay_update(struct bfd *) OVS_REQUIRES(mutex);
226 static void bfd_unixctl_show(struct unixctl_conn *, int argc,
227 const char *argv[], void *aux OVS_UNUSED);
228 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
229 int argc, const char *argv[],
230 void *aux OVS_UNUSED);
231 static void log_msg(enum vlog_level, const struct msg *, const char *message,
232 const struct bfd *) OVS_REQUIRES(mutex);
234 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 20);
236 /* Returns true if the interface on which 'bfd' is running may be used to
237 * forward traffic according to the BFD session state. */
239 bfd_forwarding(const struct bfd *bfd) OVS_EXCLUDED(mutex)
243 ovs_mutex_lock(&mutex);
244 ret = bfd_forwarding__(bfd);
245 ovs_mutex_unlock(&mutex);
249 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
250 * intended for the OVS database. */
252 bfd_get_status(const struct bfd *bfd, struct smap *smap)
255 ovs_mutex_lock(&mutex);
256 smap_add(smap, "forwarding", bfd_forwarding__(bfd)? "true" : "false");
257 smap_add(smap, "state", bfd_state_str(bfd->state));
258 smap_add(smap, "diagnostic", bfd_diag_str(bfd->diag));
260 if (bfd->state != STATE_DOWN) {
261 smap_add(smap, "remote_state", bfd_state_str(bfd->rmt_state));
262 smap_add(smap, "remote_diagnostic", bfd_diag_str(bfd->rmt_diag));
264 ovs_mutex_unlock(&mutex);
267 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
268 * according to the database configuration contained in 'cfg'. Takes ownership
269 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
270 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
271 * Also returns NULL if cfg is NULL. */
273 bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
274 struct netdev *netdev) OVS_EXCLUDED(mutex)
276 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
277 static atomic_uint16_t udp_src = ATOMIC_VAR_INIT(0);
280 long long int min_tx, min_rx;
281 bool need_poll = false;
282 bool cfg_min_rx_changed = false;
285 uint8_t ea[ETH_ADDR_LEN];
287 if (ovsthread_once_start(&once)) {
288 unixctl_command_register("bfd/show", "[interface]", 0, 1,
289 bfd_unixctl_show, NULL);
290 unixctl_command_register("bfd/set-forwarding",
291 "[interface] normal|false|true", 1, 2,
292 bfd_unixctl_set_forwarding_override, NULL);
293 ovsthread_once_done(&once);
296 if (!cfg || !smap_get_bool(cfg, "enable", false)) {
301 ovs_mutex_lock(&mutex);
303 bfd = xzalloc(sizeof *bfd);
304 bfd->name = xstrdup(name);
305 bfd->forwarding_override = -1;
306 bfd->disc = generate_discriminator();
307 hmap_insert(all_bfds, &bfd->node, bfd->disc);
309 bfd->diag = DIAG_NONE;
312 atomic_init(&bfd->ref_cnt, 1);
313 bfd->netdev = netdev_ref(netdev);
314 bfd->in_decay = false;
316 /* RFC 5881 section 4
317 * The source port MUST be in the range 49152 through 65535. The same
318 * UDP source port number MUST be used for all BFD Control packets
319 * associated with a particular session. The source port number SHOULD
320 * be unique among all BFD sessions on the system. */
321 atomic_add(&udp_src, 1, &bfd->udp_src);
322 bfd->udp_src = (bfd->udp_src % 16384) + 49152;
324 bfd_set_state(bfd, STATE_DOWN, DIAG_NONE);
326 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
329 atomic_store(&bfd->check_tnl_key,
330 smap_get_bool(cfg, "check_tnl_key", false));
331 min_tx = smap_get_int(cfg, "min_tx", 100);
332 min_tx = MAX(min_tx, 100);
333 if (bfd->cfg_min_tx != min_tx) {
334 bfd->cfg_min_tx = min_tx;
335 if (bfd->state != STATE_UP
336 || (!bfd_in_poll(bfd) && bfd->cfg_min_tx < bfd->min_tx)) {
337 bfd->min_tx = bfd->cfg_min_tx;
342 min_rx = smap_get_int(cfg, "min_rx", 1000);
343 min_rx = MAX(min_rx, 100);
344 if (bfd->cfg_min_rx != min_rx) {
345 bfd->cfg_min_rx = min_rx;
346 if (bfd->state != STATE_UP
347 || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
348 bfd->min_rx = bfd->cfg_min_rx;
350 cfg_min_rx_changed = true;
354 decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
355 if (bfd->decay_min_rx != decay_min_rx || cfg_min_rx_changed) {
356 if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
357 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
358 bfd->name, bfd->cfg_min_rx);
359 bfd->decay_min_rx = 0;
361 bfd->decay_min_rx = decay_min_rx;
364 bfd->in_decay = false;
365 bfd_decay_update(bfd);
369 cpath_down = smap_get_bool(cfg, "cpath_down", false);
370 if (bfd->cpath_down != cpath_down) {
371 bfd->cpath_down = cpath_down;
372 if (bfd->diag == DIAG_NONE || bfd->diag == DIAG_CPATH_DOWN) {
373 bfd_set_state(bfd, bfd->state, DIAG_NONE);
378 hwaddr = smap_get(cfg, "bfd_dst_mac");
379 if (hwaddr && eth_addr_from_string(hwaddr, ea) && !eth_addr_is_zero(ea)) {
380 memcpy(bfd->eth_dst, ea, ETH_ADDR_LEN);
381 bfd->eth_dst_set = true;
382 } else if (bfd->eth_dst_set) {
383 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
384 bfd->eth_dst_set = false;
390 ovs_mutex_unlock(&mutex);
395 bfd_ref(const struct bfd *bfd_)
397 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
400 atomic_add(&bfd->ref_cnt, 1, &orig);
401 ovs_assert(orig > 0);
407 bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
412 atomic_sub(&bfd->ref_cnt, 1, &orig);
413 ovs_assert(orig > 0);
415 ovs_mutex_lock(&mutex);
416 hmap_remove(all_bfds, &bfd->node);
417 netdev_close(bfd->netdev);
420 ovs_mutex_unlock(&mutex);
426 bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
428 ovs_mutex_lock(&mutex);
429 if (bfd->flags & FLAG_FINAL) {
430 poll_immediate_wake();
433 poll_timer_wait_until(bfd->next_tx);
434 if (bfd->state > STATE_DOWN) {
435 poll_timer_wait_until(bfd->detect_time);
437 ovs_mutex_unlock(&mutex);
441 bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
446 ovs_mutex_lock(&mutex);
448 old_in_decay = bfd->in_decay;
450 if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
451 bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
454 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
455 * configured, and decay_detect_time is reached. */
456 if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
457 && now >= bfd->decay_detect_time) {
461 if (bfd->min_tx != bfd->cfg_min_tx
462 || (bfd->min_rx != bfd->cfg_min_rx && bfd->min_rx != bfd->decay_min_rx)
463 || bfd->in_decay != old_in_decay) {
466 ovs_mutex_unlock(&mutex);
470 bfd_should_send_packet(const struct bfd *bfd) OVS_EXCLUDED(mutex)
473 ovs_mutex_lock(&mutex);
474 ret = bfd->flags & FLAG_FINAL || time_msec() >= bfd->next_tx;
475 ovs_mutex_unlock(&mutex);
480 bfd_put_packet(struct bfd *bfd, struct ofpbuf *p,
481 uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
483 long long int min_tx, min_rx;
484 struct udp_header *udp;
485 struct eth_header *eth;
486 struct ip_header *ip;
489 ovs_mutex_lock(&mutex);
491 long long int delay = time_msec() - bfd->next_tx;
492 long long int interval = bfd_tx_interval(bfd);
493 if (delay > interval * 3 / 2) {
494 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
495 " control message", bfd->name, delay, interval);
499 /* RFC 5880 Section 6.5
500 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
502 ovs_assert(!(bfd->flags & FLAG_POLL) || !(bfd->flags & FLAG_FINAL));
504 ofpbuf_reserve(p, 2); /* Properly align after the ethernet header. */
505 eth = ofpbuf_put_uninit(p, sizeof *eth);
506 memcpy(eth->eth_src, eth_src, ETH_ADDR_LEN);
507 memcpy(eth->eth_dst, bfd->eth_dst, ETH_ADDR_LEN);
508 eth->eth_type = htons(ETH_TYPE_IP);
510 ip = ofpbuf_put_zeros(p, sizeof *ip);
511 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
512 ip->ip_tot_len = htons(sizeof *ip + sizeof *udp + sizeof *msg);
514 ip->ip_tos = IPTOS_LOWDELAY | IPTOS_THROUGHPUT;
515 ip->ip_proto = IPPROTO_UDP;
516 ip->ip_src = htonl(0xA9FE0100); /* 169.254.1.0 Link Local. */
517 ip->ip_dst = htonl(0xA9FE0101); /* 169.254.1.1 Link Local. */
518 ip->ip_csum = csum(ip, sizeof *ip);
520 udp = ofpbuf_put_zeros(p, sizeof *udp);
521 udp->udp_src = htons(bfd->udp_src);
522 udp->udp_dst = htons(BFD_DEST_PORT);
523 udp->udp_len = htons(sizeof *udp + sizeof *msg);
525 msg = ofpbuf_put_uninit(p, sizeof *msg);
526 msg->vers_diag = (BFD_VERSION << 5) | bfd->diag;
527 msg->flags = (bfd->state & STATE_MASK) | bfd->flags;
529 msg->mult = bfd->mult;
530 msg->length = BFD_PACKET_LEN;
531 msg->my_disc = htonl(bfd->disc);
532 msg->your_disc = htonl(bfd->rmt_disc);
533 msg->min_rx_echo = htonl(0);
535 if (bfd_in_poll(bfd)) {
536 min_tx = bfd->poll_min_tx;
537 min_rx = bfd->poll_min_rx;
539 min_tx = bfd_min_tx(bfd);
540 min_rx = bfd->min_rx;
543 msg->min_tx = htonl(min_tx * 1000);
544 msg->min_rx = htonl(min_rx * 1000);
546 bfd->flags &= ~FLAG_FINAL;
548 log_msg(VLL_DBG, msg, "Sending BFD Message", bfd);
550 bfd->last_tx = time_msec();
551 bfd_set_next_tx(bfd);
552 ovs_mutex_unlock(&mutex);
556 bfd_should_process_flow(const struct bfd *bfd, const struct flow *flow,
557 struct flow_wildcards *wc)
560 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
561 if (bfd->eth_dst_set && memcmp(bfd->eth_dst, flow->dl_dst, ETH_ADDR_LEN)) {
565 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
566 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
568 atomic_read(&bfd->check_tnl_key, &check_tnl_key);
570 memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id);
572 return (flow->dl_type == htons(ETH_TYPE_IP)
573 && flow->nw_proto == IPPROTO_UDP
574 && flow->tp_dst == htons(BFD_DEST_PORT)
575 && (!check_tnl_key || flow->tunnel.tun_id == htonll(0)));
579 bfd_process_packet(struct bfd *bfd, const struct flow *flow,
580 const struct ofpbuf *p) OVS_EXCLUDED(mutex)
582 uint32_t rmt_min_rx, pkt_your_disc;
583 enum state rmt_state;
588 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
590 ovs_mutex_lock(&mutex);
591 /* Increments the decay rx counter. */
594 if (flow->nw_ttl != 255) {
595 /* XXX Should drop in the kernel to prevent DOS. */
599 msg = ofpbuf_at(p, (uint8_t *)p->l7 - (uint8_t *)p->data, BFD_PACKET_LEN);
601 VLOG_INFO_RL(&rl, "%s: Received unparseable BFD control message.",
606 /* RFC 5880 Section 6.8.6
607 * If the Length field is greater than the payload of the encapsulating
608 * protocol, the packet MUST be discarded.
610 * Note that we make this check implicity. Above we use ofpbuf_at() to
611 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
612 * the encapsulating protocol. Below we require msg->length to be exactly
613 * BFD_PACKET_LEN bytes. */
615 flags = msg->flags & FLAGS_MASK;
616 rmt_state = msg->flags & STATE_MASK;
617 version = msg->vers_diag >> VERS_SHIFT;
619 log_msg(VLL_DBG, msg, "Received BFD control message", bfd);
621 if (version != BFD_VERSION) {
622 log_msg(VLL_WARN, msg, "Incorrect version", bfd);
626 /* Technically this should happen after the length check. We don't support
627 * authentication however, so it's simpler to do the check first. */
628 if (flags & FLAG_AUTH) {
629 log_msg(VLL_WARN, msg, "Authenticated control message with"
630 " authentication disabled", bfd);
634 if (msg->length != BFD_PACKET_LEN) {
635 log_msg(VLL_WARN, msg, "Unexpected length", bfd);
636 if (msg->length < BFD_PACKET_LEN) {
642 log_msg(VLL_WARN, msg, "Zero multiplier", bfd);
646 if (flags & FLAG_MULTIPOINT) {
647 log_msg(VLL_WARN, msg, "Unsupported multipoint flag", bfd);
652 log_msg(VLL_WARN, msg, "NULL my_disc", bfd);
656 pkt_your_disc = ntohl(msg->your_disc);
658 /* Technically, we should use the your discriminator field to figure
659 * out which 'struct bfd' this packet is destined towards. That way a
660 * bfd session could migrate from one interface to another
661 * transparently. This doesn't fit in with the OVS structure very
662 * well, so in this respect, we are not compliant. */
663 if (pkt_your_disc != bfd->disc) {
664 log_msg(VLL_WARN, msg, "Incorrect your_disc", bfd);
667 } else if (rmt_state > STATE_DOWN) {
668 log_msg(VLL_WARN, msg, "Null your_disc", bfd);
672 bfd->rmt_disc = ntohl(msg->my_disc);
673 bfd->rmt_state = rmt_state;
674 bfd->rmt_flags = flags;
675 bfd->rmt_diag = msg->vers_diag & DIAG_MASK;
677 if (flags & FLAG_FINAL && bfd_in_poll(bfd)) {
678 bfd->min_tx = bfd->poll_min_tx;
679 bfd->min_rx = bfd->poll_min_rx;
680 bfd->flags &= ~FLAG_POLL;
681 log_msg(VLL_INFO, msg, "Poll sequence terminated", bfd);
684 if (flags & FLAG_POLL) {
685 /* RFC 5880 Section 6.5
686 * When the other system receives a Poll, it immediately transmits a
687 * BFD Control packet with the Final (F) bit set, independent of any
688 * periodic BFD Control packets it may be sending
689 * (see section 6.8.7). */
690 bfd->flags &= ~FLAG_POLL;
691 bfd->flags |= FLAG_FINAL;
694 rmt_min_rx = MAX(ntohl(msg->min_rx) / 1000, 1);
695 if (bfd->rmt_min_rx != rmt_min_rx) {
696 bfd->rmt_min_rx = rmt_min_rx;
697 bfd_set_next_tx(bfd);
698 log_msg(VLL_INFO, msg, "New remote min_rx", bfd);
701 bfd->rmt_min_tx = MAX(ntohl(msg->min_tx) / 1000, 1);
702 bfd->detect_time = bfd_rx_interval(bfd) * bfd->mult + time_msec();
704 if (bfd->state == STATE_ADMIN_DOWN) {
705 VLOG_DBG_RL(&rl, "Administratively down, dropping control message.");
709 if (rmt_state == STATE_ADMIN_DOWN) {
710 if (bfd->state != STATE_DOWN) {
711 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
714 switch (bfd->state) {
716 if (rmt_state == STATE_DOWN) {
717 bfd_set_state(bfd, STATE_INIT, bfd->diag);
718 } else if (rmt_state == STATE_INIT) {
719 bfd_set_state(bfd, STATE_UP, bfd->diag);
723 if (rmt_state > STATE_DOWN) {
724 bfd_set_state(bfd, STATE_UP, bfd->diag);
728 if (rmt_state <= STATE_DOWN) {
729 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
730 log_msg(VLL_INFO, msg, "Remote signaled STATE_DOWN", bfd);
733 case STATE_ADMIN_DOWN:
738 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
741 ovs_mutex_unlock(&mutex);
744 /* Must be called when the netdev owned by 'bfd' should change. */
746 bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
749 ovs_mutex_lock(&mutex);
750 if (bfd->netdev != netdev) {
751 netdev_close(bfd->netdev);
752 bfd->netdev = netdev_ref(netdev);
753 if (bfd->decay_min_rx) {
754 bfd_decay_update(bfd);
757 ovs_mutex_unlock(&mutex);
762 bfd_forwarding__(const struct bfd *bfd) OVS_REQUIRES(mutex)
764 if (bfd->forwarding_override != -1) {
765 return bfd->forwarding_override == 1;
768 return bfd->state == STATE_UP
769 && bfd->rmt_diag != DIAG_PATH_DOWN
770 && bfd->rmt_diag != DIAG_CPATH_DOWN
771 && bfd->rmt_diag != DIAG_RCPATH_DOWN;
776 bfd_in_poll(const struct bfd *bfd) OVS_REQUIRES(mutex)
778 return (bfd->flags & FLAG_POLL) != 0;
782 bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex)
784 if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
785 && !(bfd->flags & FLAG_FINAL)) {
786 bfd->poll_min_tx = bfd->cfg_min_tx;
787 bfd->poll_min_rx = bfd->in_decay ? bfd->decay_min_rx : bfd->cfg_min_rx;
788 bfd->flags |= FLAG_POLL;
790 VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
795 bfd_min_tx(const struct bfd *bfd) OVS_REQUIRES(mutex)
797 /* RFC 5880 Section 6.8.3
798 * When bfd.SessionState is not Up, the system MUST set
799 * bfd.DesiredMinTxInterval to a value of not less than one second
800 * (1,000,000 microseconds). This is intended to ensure that the
801 * bandwidth consumed by BFD sessions that are not Up is negligible,
802 * particularly in the case where a neighbor may not be running BFD. */
803 return (bfd->state == STATE_UP ? bfd->min_tx : MAX(bfd->min_tx, 1000));
807 bfd_tx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
809 long long int interval = bfd_min_tx(bfd);
810 return MAX(interval, bfd->rmt_min_rx);
814 bfd_rx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
816 return MAX(bfd->min_rx, bfd->rmt_min_tx);
820 bfd_set_next_tx(struct bfd *bfd) OVS_REQUIRES(mutex)
822 long long int interval = bfd_tx_interval(bfd);
823 interval -= interval * random_range(26) / 100;
824 bfd->next_tx = bfd->last_tx + interval;
828 bfd_flag_str(enum flags flags)
830 struct ds ds = DS_EMPTY_INITIALIZER;
831 static char flag_str[128];
837 if (flags & FLAG_MULTIPOINT) {
838 ds_put_cstr(&ds, "multipoint ");
841 if (flags & FLAG_DEMAND) {
842 ds_put_cstr(&ds, "demand ");
845 if (flags & FLAG_AUTH) {
846 ds_put_cstr(&ds, "auth ");
849 if (flags & FLAG_CTL) {
850 ds_put_cstr(&ds, "ctl ");
853 if (flags & FLAG_FINAL) {
854 ds_put_cstr(&ds, "final ");
857 if (flags & FLAG_POLL) {
858 ds_put_cstr(&ds, "poll ");
861 /* Do not copy the trailing whitespace. */
863 ovs_strlcpy(flag_str, ds_cstr(&ds), sizeof flag_str);
869 bfd_state_str(enum state state)
872 case STATE_ADMIN_DOWN: return "admin_down";
873 case STATE_DOWN: return "down";
874 case STATE_INIT: return "init";
875 case STATE_UP: return "up";
876 default: return "invalid";
881 bfd_diag_str(enum diag diag) {
883 case DIAG_NONE: return "No Diagnostic";
884 case DIAG_EXPIRED: return "Control Detection Time Expired";
885 case DIAG_ECHO_FAILED: return "Echo Function Failed";
886 case DIAG_RMT_DOWN: return "Neighbor Signaled Session Down";
887 case DIAG_FWD_RESET: return "Forwarding Plane Reset";
888 case DIAG_PATH_DOWN: return "Path Down";
889 case DIAG_CPATH_DOWN: return "Concatenated Path Down";
890 case DIAG_ADMIN_DOWN: return "Administratively Down";
891 case DIAG_RCPATH_DOWN: return "Reverse Concatenated Path Down";
892 default: return "Invalid Diagnostic";
897 log_msg(enum vlog_level level, const struct msg *p, const char *message,
898 const struct bfd *bfd) OVS_REQUIRES(mutex)
900 struct ds ds = DS_EMPTY_INITIALIZER;
902 if (vlog_should_drop(THIS_MODULE, level, &rl)) {
908 "\n\tvers:%"PRIu8" diag:\"%s\" state:%s mult:%"PRIu8
911 "\n\tmy_disc:0x%"PRIx32" your_disc:0x%"PRIx32
912 "\n\tmin_tx:%"PRIu32"us (%"PRIu32"ms)"
913 "\n\tmin_rx:%"PRIu32"us (%"PRIu32"ms)"
914 "\n\tmin_rx_echo:%"PRIu32"us (%"PRIu32"ms)",
915 bfd->name, message, p->vers_diag >> VERS_SHIFT,
916 bfd_diag_str(p->vers_diag & DIAG_MASK),
917 bfd_state_str(p->flags & STATE_MASK),
918 p->mult, p->length, bfd_flag_str(p->flags & FLAGS_MASK),
919 ntohl(p->my_disc), ntohl(p->your_disc),
920 ntohl(p->min_tx), ntohl(p->min_tx) / 1000,
921 ntohl(p->min_rx), ntohl(p->min_rx) / 1000,
922 ntohl(p->min_rx_echo), ntohl(p->min_rx_echo) / 1000);
923 bfd_put_details(&ds, bfd);
924 VLOG(level, "%s", ds_cstr(&ds));
929 bfd_set_state(struct bfd *bfd, enum state state, enum diag diag)
932 if (diag == DIAG_NONE && bfd->cpath_down) {
933 diag = DIAG_CPATH_DOWN;
936 if (bfd->state != state || bfd->diag != diag) {
937 if (!VLOG_DROP_INFO(&rl)) {
938 struct ds ds = DS_EMPTY_INITIALIZER;
940 ds_put_format(&ds, "%s: BFD state change: %s->%s"
941 " \"%s\"->\"%s\".\n",
942 bfd->name, bfd_state_str(bfd->state),
943 bfd_state_str(state), bfd_diag_str(bfd->diag),
945 bfd_put_details(&ds, bfd);
946 VLOG_INFO("%s", ds_cstr(&ds));
953 if (bfd->state <= STATE_DOWN) {
954 bfd->rmt_state = STATE_DOWN;
955 bfd->rmt_diag = DIAG_NONE;
960 /* Resets the min_rx if in_decay. */
962 bfd->min_rx = bfd->cfg_min_rx;
963 bfd->in_decay = false;
966 /* Resets the decay when state changes to STATE_UP
967 * and decay_min_rx is configured. */
968 if (bfd->state == STATE_UP && bfd->decay_min_rx) {
969 bfd_decay_update(bfd);
975 bfd_rx_packets(const struct bfd *bfd) OVS_REQUIRES(mutex)
977 struct netdev_stats stats;
979 if (!netdev_get_stats(bfd->netdev, &stats)) {
980 return stats.rx_packets;
986 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
987 * the 'expect' value. */
989 bfd_try_decay(struct bfd *bfd) OVS_REQUIRES(mutex)
991 int64_t diff, expect;
993 /* The 'diff' is the difference between current interface rx_packets
994 * stats and last-time check. The 'expect' is the recorded number of
995 * bfd control packets received within an approximately decay_min_rx
996 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
998 * Since the update of rx_packets stats at interface happens
999 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1000 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1001 * more wiggle room. */
1002 diff = bfd_rx_packets(bfd) - bfd->rx_packets;
1003 expect = 2 * MAX(bfd->decay_rx_ctl, 1);
1004 bfd->in_decay = diff <= expect ? true : false;
1005 bfd_decay_update(bfd);
1008 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1010 bfd_decay_update(struct bfd * bfd) OVS_REQUIRES(mutex)
1012 bfd->rx_packets = bfd_rx_packets(bfd);
1013 bfd->decay_rx_ctl = 0;
1014 bfd->decay_detect_time = MAX(bfd->decay_min_rx, 2000) + time_msec();
1018 generate_discriminator(void)
1022 /* RFC 5880 Section 6.8.1
1023 * It SHOULD be set to a random (but still unique) value to improve
1024 * security. The value is otherwise outside the scope of this
1030 /* 'disc' is by definition random, so there's no reason to waste time
1032 disc = random_uint32();
1033 HMAP_FOR_EACH_IN_BUCKET (bfd, node, disc, all_bfds) {
1034 if (bfd->disc == disc) {
1045 bfd_find_by_name(const char *name) OVS_REQUIRES(mutex)
1049 HMAP_FOR_EACH (bfd, node, all_bfds) {
1050 if (!strcmp(bfd->name, name)) {
1058 bfd_put_details(struct ds *ds, const struct bfd *bfd) OVS_REQUIRES(mutex)
1060 ds_put_format(ds, "\tForwarding: %s\n",
1061 bfd_forwarding__(bfd) ? "true" : "false");
1062 ds_put_format(ds, "\tDetect Multiplier: %d\n", bfd->mult);
1063 ds_put_format(ds, "\tConcatenated Path Down: %s\n",
1064 bfd->cpath_down ? "true" : "false");
1065 ds_put_format(ds, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd));
1066 ds_put_format(ds, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd));
1067 ds_put_format(ds, "\tDetect Time: now %+lldms\n",
1068 time_msec() - bfd->detect_time);
1069 ds_put_format(ds, "\tNext TX Time: now %+lldms\n",
1070 time_msec() - bfd->next_tx);
1071 ds_put_format(ds, "\tLast TX Time: now %+lldms\n",
1072 time_msec() - bfd->last_tx);
1074 ds_put_cstr(ds, "\n");
1076 ds_put_format(ds, "\tLocal Flags: %s\n", bfd_flag_str(bfd->flags));
1077 ds_put_format(ds, "\tLocal Session State: %s\n",
1078 bfd_state_str(bfd->state));
1079 ds_put_format(ds, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd->diag));
1080 ds_put_format(ds, "\tLocal Discriminator: 0x%"PRIx32"\n", bfd->disc);
1081 ds_put_format(ds, "\tLocal Minimum TX Interval: %lldms\n",
1083 ds_put_format(ds, "\tLocal Minimum RX Interval: %lldms\n", bfd->min_rx);
1085 ds_put_cstr(ds, "\n");
1087 ds_put_format(ds, "\tRemote Flags: %s\n", bfd_flag_str(bfd->rmt_flags));
1088 ds_put_format(ds, "\tRemote Session State: %s\n",
1089 bfd_state_str(bfd->rmt_state));
1090 ds_put_format(ds, "\tRemote Diagnostic: %s\n",
1091 bfd_diag_str(bfd->rmt_diag));
1092 ds_put_format(ds, "\tRemote Discriminator: 0x%"PRIx32"\n", bfd->rmt_disc);
1093 ds_put_format(ds, "\tRemote Minimum TX Interval: %lldms\n",
1095 ds_put_format(ds, "\tRemote Minimum RX Interval: %lldms\n",
1100 bfd_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1101 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1103 struct ds ds = DS_EMPTY_INITIALIZER;
1106 ovs_mutex_lock(&mutex);
1108 bfd = bfd_find_by_name(argv[1]);
1110 unixctl_command_reply_error(conn, "no such bfd object");
1113 bfd_put_details(&ds, bfd);
1115 HMAP_FOR_EACH (bfd, node, all_bfds) {
1116 ds_put_format(&ds, "---- %s ----\n", bfd->name);
1117 bfd_put_details(&ds, bfd);
1120 unixctl_command_reply(conn, ds_cstr(&ds));
1124 ovs_mutex_unlock(&mutex);
1129 bfd_unixctl_set_forwarding_override(struct unixctl_conn *conn, int argc,
1130 const char *argv[], void *aux OVS_UNUSED)
1133 const char *forward_str = argv[argc - 1];
1134 int forwarding_override;
1137 ovs_mutex_lock(&mutex);
1138 if (!strcasecmp("true", forward_str)) {
1139 forwarding_override = 1;
1140 } else if (!strcasecmp("false", forward_str)) {
1141 forwarding_override = 0;
1142 } else if (!strcasecmp("normal", forward_str)) {
1143 forwarding_override = -1;
1145 unixctl_command_reply_error(conn, "unknown fault string");
1150 bfd = bfd_find_by_name(argv[1]);
1152 unixctl_command_reply_error(conn, "no such BFD object");
1155 bfd->forwarding_override = forwarding_override;
1157 HMAP_FOR_EACH (bfd, node, all_bfds) {
1158 bfd->forwarding_override = forwarding_override;
1162 unixctl_command_reply(conn, "OK");
1165 ovs_mutex_unlock(&mutex);