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 /* When 'bfd->forwarding_if_rx' is set, at least one bfd control packet
207 * is required to be received every 100 * bfd->cfg_min_rx. If bfd
208 * control packet is not received within this interval, even if data
209 * packets are received, the bfd->forwarding will still be false. */
210 long long int demand_rx_bfd_time;
212 /* BFD decay related variables. */
213 bool in_decay; /* True when bfd is in decay. */
214 int decay_min_rx; /* min_rx is set to decay_min_rx when */
216 int decay_rx_ctl; /* Count bfd packets received within decay */
217 /* detect interval. */
218 uint64_t decay_rx_packets; /* Packets received by 'netdev'. */
219 long long int decay_detect_time; /* Decay detection time. */
221 uint64_t flap_count; /* Counts bfd forwarding flaps. */
223 /* True when the variables returned by bfd_get_status() are changed
224 * since last check. */
228 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
229 static struct hmap all_bfds__ = HMAP_INITIALIZER(&all_bfds__);
230 static struct hmap *const all_bfds OVS_GUARDED_BY(mutex) = &all_bfds__;
232 static bool bfd_lookup_ip(const char *host_name, struct in_addr *)
234 static bool bfd_forwarding__(struct bfd *) OVS_REQUIRES(mutex);
235 static bool bfd_in_poll(const struct bfd *) OVS_REQUIRES(mutex);
236 static void bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex);
237 static const char *bfd_diag_str(enum diag) OVS_REQUIRES(mutex);
238 static const char *bfd_state_str(enum state) OVS_REQUIRES(mutex);
239 static long long int bfd_min_tx(const struct bfd *) OVS_REQUIRES(mutex);
240 static long long int bfd_tx_interval(const struct bfd *)
242 static long long int bfd_rx_interval(const struct bfd *)
244 static void bfd_set_next_tx(struct bfd *) OVS_REQUIRES(mutex);
245 static void bfd_set_state(struct bfd *, enum state, enum diag)
247 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex);
248 static void bfd_put_details(struct ds *, const struct bfd *)
250 static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQUIRES(mutex);
251 static void bfd_try_decay(struct bfd *) OVS_REQUIRES(mutex);
252 static void bfd_decay_update(struct bfd *) OVS_REQUIRES(mutex);
253 static void bfd_status_changed(struct bfd *) OVS_REQUIRES(mutex);
255 static void bfd_forwarding_if_rx_update(struct bfd *) OVS_REQUIRES(mutex);
256 static void bfd_unixctl_show(struct unixctl_conn *, int argc,
257 const char *argv[], void *aux OVS_UNUSED);
258 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
259 int argc, const char *argv[],
260 void *aux OVS_UNUSED);
261 static void log_msg(enum vlog_level, const struct msg *, const char *message,
262 const struct bfd *) OVS_REQUIRES(mutex);
264 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 20);
266 /* Returns true if the interface on which 'bfd' is running may be used to
267 * forward traffic according to the BFD session state. */
269 bfd_forwarding(struct bfd *bfd) OVS_EXCLUDED(mutex)
273 ovs_mutex_lock(&mutex);
274 ret = bfd_forwarding__(bfd);
275 ovs_mutex_unlock(&mutex);
279 /* When forwarding_if_rx is enabled, if there are packets received,
280 * updates forwarding_if_rx_detect_time. */
282 bfd_account_rx(struct bfd *bfd, const struct dpif_flow_stats *stats)
284 if (stats->n_packets && bfd->forwarding_if_rx) {
285 ovs_mutex_lock(&mutex);
286 bfd_forwarding__(bfd);
287 bfd_forwarding_if_rx_update(bfd);
288 bfd_forwarding__(bfd);
289 ovs_mutex_unlock(&mutex);
293 /* Returns and resets the 'bfd->status_changed'. */
295 bfd_check_status_change(struct bfd *bfd) OVS_EXCLUDED(mutex)
299 ovs_mutex_lock(&mutex);
300 ret = bfd->status_changed;
301 bfd->status_changed = false;
302 ovs_mutex_unlock(&mutex);
307 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
308 * intended for the OVS database. */
310 bfd_get_status(const struct bfd *bfd, struct smap *smap)
313 ovs_mutex_lock(&mutex);
314 smap_add(smap, "forwarding",
315 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
317 smap_add(smap, "state", bfd_state_str(bfd->state));
318 smap_add(smap, "diagnostic", bfd_diag_str(bfd->diag));
319 smap_add_format(smap, "flap_count", "%"PRIu64, bfd->flap_count);
321 if (bfd->state != STATE_DOWN) {
322 smap_add(smap, "remote_state", bfd_state_str(bfd->rmt_state));
323 smap_add(smap, "remote_diagnostic", bfd_diag_str(bfd->rmt_diag));
325 ovs_mutex_unlock(&mutex);
328 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
329 * according to the database configuration contained in 'cfg'. Takes ownership
330 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
331 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
332 * Also returns NULL if cfg is NULL. */
334 bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
335 struct netdev *netdev) OVS_EXCLUDED(mutex)
337 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
338 static atomic_uint16_t udp_src = ATOMIC_VAR_INIT(0);
341 long long int min_tx, min_rx;
342 bool need_poll = false;
343 bool cfg_min_rx_changed = false;
344 bool cpath_down, forwarding_if_rx;
345 const char *hwaddr, *ip_src, *ip_dst;
346 struct in_addr in_addr;
347 uint8_t ea[ETH_ADDR_LEN];
349 if (ovsthread_once_start(&once)) {
350 unixctl_command_register("bfd/show", "[interface]", 0, 1,
351 bfd_unixctl_show, NULL);
352 unixctl_command_register("bfd/set-forwarding",
353 "[interface] normal|false|true", 1, 2,
354 bfd_unixctl_set_forwarding_override, NULL);
355 ovsthread_once_done(&once);
358 if (!cfg || !smap_get_bool(cfg, "enable", false)) {
363 ovs_mutex_lock(&mutex);
365 bfd = xzalloc(sizeof *bfd);
366 bfd->name = xstrdup(name);
367 bfd->forwarding_override = -1;
368 bfd->disc = generate_discriminator();
369 hmap_insert(all_bfds, &bfd->node, bfd->disc);
371 bfd->diag = DIAG_NONE;
374 ovs_refcount_init(&bfd->ref_cnt);
375 bfd->netdev = netdev_ref(netdev);
376 bfd->rx_packets = bfd_rx_packets(bfd);
377 bfd->in_decay = false;
380 /* RFC 5881 section 4
381 * The source port MUST be in the range 49152 through 65535. The same
382 * UDP source port number MUST be used for all BFD Control packets
383 * associated with a particular session. The source port number SHOULD
384 * be unique among all BFD sessions on the system. */
385 atomic_add(&udp_src, 1, &bfd->udp_src);
386 bfd->udp_src = (bfd->udp_src % 16384) + 49152;
388 bfd_set_state(bfd, STATE_DOWN, DIAG_NONE);
390 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
393 atomic_store(&bfd->check_tnl_key,
394 smap_get_bool(cfg, "check_tnl_key", false));
395 min_tx = smap_get_int(cfg, "min_tx", 100);
396 min_tx = MAX(min_tx, 100);
397 if (bfd->cfg_min_tx != min_tx) {
398 bfd->cfg_min_tx = min_tx;
399 if (bfd->state != STATE_UP
400 || (!bfd_in_poll(bfd) && bfd->cfg_min_tx < bfd->min_tx)) {
401 bfd->min_tx = bfd->cfg_min_tx;
406 min_rx = smap_get_int(cfg, "min_rx", 1000);
407 min_rx = MAX(min_rx, 100);
408 if (bfd->cfg_min_rx != min_rx) {
409 bfd->cfg_min_rx = min_rx;
410 if (bfd->state != STATE_UP
411 || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
412 bfd->min_rx = bfd->cfg_min_rx;
414 cfg_min_rx_changed = true;
418 decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
419 if (bfd->decay_min_rx != decay_min_rx || cfg_min_rx_changed) {
420 if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
421 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
422 bfd->name, bfd->cfg_min_rx);
423 bfd->decay_min_rx = 0;
425 bfd->decay_min_rx = decay_min_rx;
428 bfd->in_decay = false;
429 bfd_decay_update(bfd);
433 cpath_down = smap_get_bool(cfg, "cpath_down", false);
434 if (bfd->cpath_down != cpath_down) {
435 bfd->cpath_down = cpath_down;
436 bfd_set_state(bfd, bfd->state, DIAG_NONE);
440 hwaddr = smap_get(cfg, "bfd_dst_mac");
441 if (hwaddr && eth_addr_from_string(hwaddr, ea) && !eth_addr_is_zero(ea)) {
442 memcpy(bfd->eth_dst, ea, ETH_ADDR_LEN);
443 bfd->eth_dst_set = true;
444 } else if (bfd->eth_dst_set) {
445 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
446 bfd->eth_dst_set = false;
449 ip_src = smap_get(cfg, "bfd_src_ip");
450 if (ip_src && bfd_lookup_ip(ip_src, &in_addr)) {
451 memcpy(&bfd->ip_src, &in_addr, sizeof in_addr);
453 bfd->ip_src = htonl(0xA9FE0100); /* 169.254.1.0. */
456 ip_dst = smap_get(cfg, "bfd_dst_ip");
457 if (ip_dst && bfd_lookup_ip(ip_dst, &in_addr)) {
458 memcpy(&bfd->ip_dst, &in_addr, sizeof in_addr);
460 bfd->ip_dst = htonl(0xA9FE0101); /* 169.254.1.1. */
463 forwarding_if_rx = smap_get_bool(cfg, "forwarding_if_rx", false);
464 if (bfd->forwarding_if_rx != forwarding_if_rx) {
465 bfd->forwarding_if_rx = forwarding_if_rx;
466 if (bfd->state == STATE_UP && bfd->forwarding_if_rx) {
467 bfd_forwarding_if_rx_update(bfd);
469 bfd->forwarding_if_rx_detect_time = 0;
476 ovs_mutex_unlock(&mutex);
481 bfd_ref(const struct bfd *bfd_)
483 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
485 ovs_refcount_ref(&bfd->ref_cnt);
491 bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
493 if (bfd && ovs_refcount_unref(&bfd->ref_cnt) == 1) {
494 ovs_mutex_lock(&mutex);
495 hmap_remove(all_bfds, &bfd->node);
496 netdev_close(bfd->netdev);
499 ovs_mutex_unlock(&mutex);
504 bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
506 poll_timer_wait_until(bfd_wake_time(bfd));
509 /* Returns the next wake up time. */
511 bfd_wake_time(const struct bfd *bfd) OVS_EXCLUDED(mutex)
513 long long int retval;
519 ovs_mutex_lock(&mutex);
520 if (bfd->flags & FLAG_FINAL) {
523 retval = bfd->next_tx;
524 if (bfd->state > STATE_DOWN) {
525 retval = MIN(bfd->detect_time, retval);
528 ovs_mutex_unlock(&mutex);
533 bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
538 ovs_mutex_lock(&mutex);
540 old_in_decay = bfd->in_decay;
542 if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
543 bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
545 bfd_forwarding__(bfd);
547 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
548 * configured, and decay_detect_time is reached. */
549 if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
550 && now >= bfd->decay_detect_time) {
554 if (bfd->min_tx != bfd->cfg_min_tx
555 || (bfd->min_rx != bfd->cfg_min_rx && bfd->min_rx != bfd->decay_min_rx)
556 || bfd->in_decay != old_in_decay) {
559 ovs_mutex_unlock(&mutex);
563 bfd_should_send_packet(const struct bfd *bfd) OVS_EXCLUDED(mutex)
566 ovs_mutex_lock(&mutex);
567 ret = bfd->flags & FLAG_FINAL || time_msec() >= bfd->next_tx;
568 ovs_mutex_unlock(&mutex);
573 bfd_put_packet(struct bfd *bfd, struct ofpbuf *p,
574 uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
576 long long int min_tx, min_rx;
577 struct udp_header *udp;
578 struct eth_header *eth;
579 struct ip_header *ip;
582 ovs_mutex_lock(&mutex);
584 long long int delay = time_msec() - bfd->next_tx;
585 long long int interval = bfd_tx_interval(bfd);
586 if (delay > interval * 3 / 2) {
587 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
588 " control message", bfd->name, delay, interval);
592 /* RFC 5880 Section 6.5
593 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
595 ovs_assert(!(bfd->flags & FLAG_POLL) || !(bfd->flags & FLAG_FINAL));
597 ofpbuf_reserve(p, 2); /* Properly align after the ethernet header. */
598 eth = ofpbuf_put_uninit(p, sizeof *eth);
599 memcpy(eth->eth_src, eth_src, ETH_ADDR_LEN);
600 memcpy(eth->eth_dst, bfd->eth_dst, ETH_ADDR_LEN);
601 eth->eth_type = htons(ETH_TYPE_IP);
603 ip = ofpbuf_put_zeros(p, sizeof *ip);
604 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
605 ip->ip_tot_len = htons(sizeof *ip + sizeof *udp + sizeof *msg);
607 ip->ip_tos = IPTOS_LOWDELAY | IPTOS_THROUGHPUT;
608 ip->ip_proto = IPPROTO_UDP;
609 put_16aligned_be32(&ip->ip_src, bfd->ip_src);
610 put_16aligned_be32(&ip->ip_dst, bfd->ip_dst);
611 ip->ip_csum = csum(ip, sizeof *ip);
613 udp = ofpbuf_put_zeros(p, sizeof *udp);
614 udp->udp_src = htons(bfd->udp_src);
615 udp->udp_dst = htons(BFD_DEST_PORT);
616 udp->udp_len = htons(sizeof *udp + sizeof *msg);
618 msg = ofpbuf_put_uninit(p, sizeof *msg);
619 msg->vers_diag = (BFD_VERSION << 5) | bfd->diag;
620 msg->flags = (bfd->state & STATE_MASK) | bfd->flags;
622 msg->mult = bfd->mult;
623 msg->length = BFD_PACKET_LEN;
624 msg->my_disc = htonl(bfd->disc);
625 msg->your_disc = htonl(bfd->rmt_disc);
626 msg->min_rx_echo = htonl(0);
628 if (bfd_in_poll(bfd)) {
629 min_tx = bfd->poll_min_tx;
630 min_rx = bfd->poll_min_rx;
632 min_tx = bfd_min_tx(bfd);
633 min_rx = bfd->min_rx;
636 msg->min_tx = htonl(min_tx * 1000);
637 msg->min_rx = htonl(min_rx * 1000);
639 bfd->flags &= ~FLAG_FINAL;
641 log_msg(VLL_DBG, msg, "Sending BFD Message", bfd);
643 bfd->last_tx = time_msec();
644 bfd_set_next_tx(bfd);
645 ovs_mutex_unlock(&mutex);
649 bfd_should_process_flow(const struct bfd *bfd_, const struct flow *flow,
650 struct flow_wildcards *wc)
652 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
655 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
656 if (bfd->eth_dst_set && memcmp(bfd->eth_dst, flow->dl_dst, ETH_ADDR_LEN)) {
660 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
661 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
663 atomic_read(&bfd->check_tnl_key, &check_tnl_key);
665 memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id);
667 return (flow->dl_type == htons(ETH_TYPE_IP)
668 && flow->nw_proto == IPPROTO_UDP
669 && flow->tp_dst == htons(BFD_DEST_PORT)
670 && (!check_tnl_key || flow->tunnel.tun_id == htonll(0)));
674 bfd_process_packet(struct bfd *bfd, const struct flow *flow,
675 const struct ofpbuf *p) OVS_EXCLUDED(mutex)
677 uint32_t rmt_min_rx, pkt_your_disc;
678 enum state rmt_state;
682 const uint8_t *l7 = ofpbuf_get_udp_payload(p);
685 return; /* No UDP payload. */
688 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
690 ovs_mutex_lock(&mutex);
691 /* Increments the decay rx counter. */
694 bfd_forwarding__(bfd);
696 if (flow->nw_ttl != 255) {
697 /* XXX Should drop in the kernel to prevent DOS. */
701 msg = ofpbuf_at(p, l7 - (uint8_t *)ofpbuf_data(p), BFD_PACKET_LEN);
703 VLOG_INFO_RL(&rl, "%s: Received too-short BFD control message (only "
704 "%"PRIdPTR" bytes long, at least %d required).",
705 bfd->name, (uint8_t *) ofpbuf_tail(p) - l7,
710 /* RFC 5880 Section 6.8.6
711 * If the Length field is greater than the payload of the encapsulating
712 * protocol, the packet MUST be discarded.
714 * Note that we make this check implicity. Above we use ofpbuf_at() to
715 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
716 * the encapsulating protocol. Below we require msg->length to be exactly
717 * BFD_PACKET_LEN bytes. */
719 flags = msg->flags & FLAGS_MASK;
720 rmt_state = msg->flags & STATE_MASK;
721 version = msg->vers_diag >> VERS_SHIFT;
723 log_msg(VLL_DBG, msg, "Received BFD control message", bfd);
725 if (version != BFD_VERSION) {
726 log_msg(VLL_WARN, msg, "Incorrect version", bfd);
730 /* Technically this should happen after the length check. We don't support
731 * authentication however, so it's simpler to do the check first. */
732 if (flags & FLAG_AUTH) {
733 log_msg(VLL_WARN, msg, "Authenticated control message with"
734 " authentication disabled", bfd);
738 if (msg->length != BFD_PACKET_LEN) {
739 log_msg(VLL_WARN, msg, "Unexpected length", bfd);
740 if (msg->length < BFD_PACKET_LEN) {
746 log_msg(VLL_WARN, msg, "Zero multiplier", bfd);
750 if (flags & FLAG_MULTIPOINT) {
751 log_msg(VLL_WARN, msg, "Unsupported multipoint flag", bfd);
756 log_msg(VLL_WARN, msg, "NULL my_disc", bfd);
760 pkt_your_disc = ntohl(msg->your_disc);
762 /* Technically, we should use the your discriminator field to figure
763 * out which 'struct bfd' this packet is destined towards. That way a
764 * bfd session could migrate from one interface to another
765 * transparently. This doesn't fit in with the OVS structure very
766 * well, so in this respect, we are not compliant. */
767 if (pkt_your_disc != bfd->disc) {
768 log_msg(VLL_WARN, msg, "Incorrect your_disc", bfd);
771 } else if (rmt_state > STATE_DOWN) {
772 log_msg(VLL_WARN, msg, "Null your_disc", bfd);
776 if (bfd->rmt_state != rmt_state) {
777 bfd_status_changed(bfd);
780 bfd->rmt_disc = ntohl(msg->my_disc);
781 bfd->rmt_state = rmt_state;
782 bfd->rmt_flags = flags;
783 bfd->rmt_diag = msg->vers_diag & DIAG_MASK;
785 if (flags & FLAG_FINAL && bfd_in_poll(bfd)) {
786 bfd->min_tx = bfd->poll_min_tx;
787 bfd->min_rx = bfd->poll_min_rx;
788 bfd->flags &= ~FLAG_POLL;
789 log_msg(VLL_INFO, msg, "Poll sequence terminated", bfd);
792 if (flags & FLAG_POLL) {
793 /* RFC 5880 Section 6.5
794 * When the other system receives a Poll, it immediately transmits a
795 * BFD Control packet with the Final (F) bit set, independent of any
796 * periodic BFD Control packets it may be sending
797 * (see section 6.8.7). */
798 bfd->flags &= ~FLAG_POLL;
799 bfd->flags |= FLAG_FINAL;
802 rmt_min_rx = MAX(ntohl(msg->min_rx) / 1000, 1);
803 if (bfd->rmt_min_rx != rmt_min_rx) {
804 bfd->rmt_min_rx = rmt_min_rx;
806 bfd_set_next_tx(bfd);
808 log_msg(VLL_INFO, msg, "New remote min_rx", bfd);
811 bfd->rmt_min_tx = MAX(ntohl(msg->min_tx) / 1000, 1);
812 bfd->detect_time = bfd_rx_interval(bfd) * bfd->mult + time_msec();
814 if (bfd->state == STATE_ADMIN_DOWN) {
815 VLOG_DBG_RL(&rl, "Administratively down, dropping control message.");
819 if (rmt_state == STATE_ADMIN_DOWN) {
820 if (bfd->state != STATE_DOWN) {
821 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
824 switch (bfd->state) {
826 if (rmt_state == STATE_DOWN) {
827 bfd_set_state(bfd, STATE_INIT, bfd->diag);
828 } else if (rmt_state == STATE_INIT) {
829 bfd_set_state(bfd, STATE_UP, bfd->diag);
833 if (rmt_state > STATE_DOWN) {
834 bfd_set_state(bfd, STATE_UP, bfd->diag);
838 if (rmt_state <= STATE_DOWN) {
839 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
840 log_msg(VLL_INFO, msg, "Remote signaled STATE_DOWN", bfd);
843 case STATE_ADMIN_DOWN:
848 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
850 if (bfd->forwarding_if_rx) {
851 bfd->demand_rx_bfd_time = time_msec() + 100 * bfd->cfg_min_rx;
855 bfd_forwarding__(bfd);
856 ovs_mutex_unlock(&mutex);
859 /* Must be called when the netdev owned by 'bfd' should change. */
861 bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
864 ovs_mutex_lock(&mutex);
865 if (bfd->netdev != netdev) {
866 netdev_close(bfd->netdev);
867 bfd->netdev = netdev_ref(netdev);
868 if (bfd->decay_min_rx && bfd->state == STATE_UP) {
869 bfd_decay_update(bfd);
871 if (bfd->forwarding_if_rx && bfd->state == STATE_UP) {
872 bfd_forwarding_if_rx_update(bfd);
874 bfd->rx_packets = bfd_rx_packets(bfd);
876 ovs_mutex_unlock(&mutex);
880 /* Updates the forwarding flag. If override is not configured and
881 * the forwarding flag value changes, increments the flap count.
883 * Note this function may be called multiple times in a function
884 * (e.g. bfd_account_rx) before and after the bfd state or status
885 * change. This is to capture any forwarding flag flap. */
887 bfd_forwarding__(struct bfd *bfd) OVS_REQUIRES(mutex)
889 long long int now = time_msec();
890 bool forwarding_if_rx;
891 bool last_forwarding = bfd->last_forwarding;
893 if (bfd->forwarding_override != -1) {
894 return bfd->forwarding_override == 1;
897 forwarding_if_rx = bfd->forwarding_if_rx
898 && bfd->forwarding_if_rx_detect_time > now
899 && bfd->demand_rx_bfd_time > now;
901 bfd->last_forwarding = (bfd->state == STATE_UP || forwarding_if_rx)
902 && bfd->rmt_diag != DIAG_PATH_DOWN
903 && bfd->rmt_diag != DIAG_CPATH_DOWN
904 && bfd->rmt_diag != DIAG_RCPATH_DOWN;
905 if (bfd->last_forwarding != last_forwarding) {
907 bfd_status_changed(bfd);
909 return bfd->last_forwarding;
914 bfd_lookup_ip(const char *host_name, struct in_addr *addr)
916 if (!inet_pton(AF_INET, host_name, addr)) {
917 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
924 bfd_in_poll(const struct bfd *bfd) OVS_REQUIRES(mutex)
926 return (bfd->flags & FLAG_POLL) != 0;
930 bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex)
932 if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
933 && !(bfd->flags & FLAG_FINAL)) {
934 bfd->poll_min_tx = bfd->cfg_min_tx;
935 bfd->poll_min_rx = bfd->in_decay ? bfd->decay_min_rx : bfd->cfg_min_rx;
936 bfd->flags |= FLAG_POLL;
938 VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
943 bfd_min_tx(const struct bfd *bfd) OVS_REQUIRES(mutex)
945 /* RFC 5880 Section 6.8.3
946 * When bfd.SessionState is not Up, the system MUST set
947 * bfd.DesiredMinTxInterval to a value of not less than one second
948 * (1,000,000 microseconds). This is intended to ensure that the
949 * bandwidth consumed by BFD sessions that are not Up is negligible,
950 * particularly in the case where a neighbor may not be running BFD. */
951 return (bfd->state == STATE_UP ? bfd->min_tx : MAX(bfd->min_tx, 1000));
955 bfd_tx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
957 long long int interval = bfd_min_tx(bfd);
958 return MAX(interval, bfd->rmt_min_rx);
962 bfd_rx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
964 return MAX(bfd->min_rx, bfd->rmt_min_tx);
968 bfd_set_next_tx(struct bfd *bfd) OVS_REQUIRES(mutex)
970 long long int interval = bfd_tx_interval(bfd);
971 interval -= interval * random_range(26) / 100;
972 bfd->next_tx = bfd->last_tx + interval;
976 bfd_flag_str(enum flags flags)
978 struct ds ds = DS_EMPTY_INITIALIZER;
979 static char flag_str[128];
985 if (flags & FLAG_MULTIPOINT) {
986 ds_put_cstr(&ds, "multipoint ");
989 if (flags & FLAG_DEMAND) {
990 ds_put_cstr(&ds, "demand ");
993 if (flags & FLAG_AUTH) {
994 ds_put_cstr(&ds, "auth ");
997 if (flags & FLAG_CTL) {
998 ds_put_cstr(&ds, "ctl ");
1001 if (flags & FLAG_FINAL) {
1002 ds_put_cstr(&ds, "final ");
1005 if (flags & FLAG_POLL) {
1006 ds_put_cstr(&ds, "poll ");
1009 /* Do not copy the trailing whitespace. */
1011 ovs_strlcpy(flag_str, ds_cstr(&ds), sizeof flag_str);
1017 bfd_state_str(enum state state)
1020 case STATE_ADMIN_DOWN: return "admin_down";
1021 case STATE_DOWN: return "down";
1022 case STATE_INIT: return "init";
1023 case STATE_UP: return "up";
1024 default: return "invalid";
1029 bfd_diag_str(enum diag diag) {
1031 case DIAG_NONE: return "No Diagnostic";
1032 case DIAG_EXPIRED: return "Control Detection Time Expired";
1033 case DIAG_ECHO_FAILED: return "Echo Function Failed";
1034 case DIAG_RMT_DOWN: return "Neighbor Signaled Session Down";
1035 case DIAG_FWD_RESET: return "Forwarding Plane Reset";
1036 case DIAG_PATH_DOWN: return "Path Down";
1037 case DIAG_CPATH_DOWN: return "Concatenated Path Down";
1038 case DIAG_ADMIN_DOWN: return "Administratively Down";
1039 case DIAG_RCPATH_DOWN: return "Reverse Concatenated Path Down";
1040 default: return "Invalid Diagnostic";
1045 log_msg(enum vlog_level level, const struct msg *p, const char *message,
1046 const struct bfd *bfd) OVS_REQUIRES(mutex)
1048 struct ds ds = DS_EMPTY_INITIALIZER;
1050 if (vlog_should_drop(THIS_MODULE, level, &rl)) {
1056 "\n\tvers:%"PRIu8" diag:\"%s\" state:%s mult:%"PRIu8
1059 "\n\tmy_disc:0x%"PRIx32" your_disc:0x%"PRIx32
1060 "\n\tmin_tx:%"PRIu32"us (%"PRIu32"ms)"
1061 "\n\tmin_rx:%"PRIu32"us (%"PRIu32"ms)"
1062 "\n\tmin_rx_echo:%"PRIu32"us (%"PRIu32"ms)",
1063 bfd->name, message, p->vers_diag >> VERS_SHIFT,
1064 bfd_diag_str(p->vers_diag & DIAG_MASK),
1065 bfd_state_str(p->flags & STATE_MASK),
1066 p->mult, p->length, bfd_flag_str(p->flags & FLAGS_MASK),
1067 ntohl(p->my_disc), ntohl(p->your_disc),
1068 ntohl(p->min_tx), ntohl(p->min_tx) / 1000,
1069 ntohl(p->min_rx), ntohl(p->min_rx) / 1000,
1070 ntohl(p->min_rx_echo), ntohl(p->min_rx_echo) / 1000);
1071 bfd_put_details(&ds, bfd);
1072 VLOG(level, "%s", ds_cstr(&ds));
1077 bfd_set_state(struct bfd *bfd, enum state state, enum diag diag)
1080 if (bfd->cpath_down) {
1081 diag = DIAG_CPATH_DOWN;
1084 if (bfd->state != state || bfd->diag != diag) {
1085 if (!VLOG_DROP_INFO(&rl)) {
1086 struct ds ds = DS_EMPTY_INITIALIZER;
1088 ds_put_format(&ds, "%s: BFD state change: %s->%s"
1089 " \"%s\"->\"%s\".\n",
1090 bfd->name, bfd_state_str(bfd->state),
1091 bfd_state_str(state), bfd_diag_str(bfd->diag),
1092 bfd_diag_str(diag));
1093 bfd_put_details(&ds, bfd);
1094 VLOG_INFO("%s", ds_cstr(&ds));
1101 if (bfd->state <= STATE_DOWN) {
1102 bfd->rmt_state = STATE_DOWN;
1103 bfd->rmt_diag = DIAG_NONE;
1104 bfd->rmt_min_rx = 1;
1107 bfd->rmt_min_tx = 0;
1108 /* Resets the min_rx if in_decay. */
1109 if (bfd->in_decay) {
1110 bfd->min_rx = bfd->cfg_min_rx;
1111 bfd->in_decay = false;
1114 /* Resets the decay when state changes to STATE_UP
1115 * and decay_min_rx is configured. */
1116 if (bfd->state == STATE_UP && bfd->decay_min_rx) {
1117 bfd_decay_update(bfd);
1120 bfd_status_changed(bfd);
1125 bfd_rx_packets(const struct bfd *bfd) OVS_REQUIRES(mutex)
1127 struct netdev_stats stats;
1129 if (!netdev_get_stats(bfd->netdev, &stats)) {
1130 return stats.rx_packets;
1136 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
1137 * the 'expect' value. */
1139 bfd_try_decay(struct bfd *bfd) OVS_REQUIRES(mutex)
1141 int64_t diff, expect;
1143 /* The 'diff' is the difference between current interface rx_packets
1144 * stats and last-time check. The 'expect' is the recorded number of
1145 * bfd control packets received within an approximately decay_min_rx
1146 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
1148 * Since the update of rx_packets stats at interface happens
1149 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1150 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1151 * more wiggle room. */
1152 diff = bfd_rx_packets(bfd) - bfd->decay_rx_packets;
1153 expect = 2 * MAX(bfd->decay_rx_ctl, 1);
1154 bfd->in_decay = diff <= expect ? true : false;
1155 bfd_decay_update(bfd);
1158 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1160 bfd_decay_update(struct bfd * bfd) OVS_REQUIRES(mutex)
1162 bfd->decay_rx_packets = bfd_rx_packets(bfd);
1163 bfd->decay_rx_ctl = 0;
1164 bfd->decay_detect_time = MAX(bfd->decay_min_rx, 2000) + time_msec();
1167 /* Records the status change and changes the global connectivity seq. */
1169 bfd_status_changed(struct bfd *bfd) OVS_REQUIRES(mutex)
1171 seq_change(connectivity_seq_get());
1172 bfd->status_changed = true;
1176 bfd_forwarding_if_rx_update(struct bfd *bfd) OVS_REQUIRES(mutex)
1178 int64_t incr = bfd_rx_interval(bfd) * bfd->mult;
1179 bfd->forwarding_if_rx_detect_time = MAX(incr, 2000) + time_msec();
1183 generate_discriminator(void)
1187 /* RFC 5880 Section 6.8.1
1188 * It SHOULD be set to a random (but still unique) value to improve
1189 * security. The value is otherwise outside the scope of this
1195 /* 'disc' is by definition random, so there's no reason to waste time
1197 disc = random_uint32();
1198 HMAP_FOR_EACH_IN_BUCKET (bfd, node, disc, all_bfds) {
1199 if (bfd->disc == disc) {
1210 bfd_find_by_name(const char *name) OVS_REQUIRES(mutex)
1214 HMAP_FOR_EACH (bfd, node, all_bfds) {
1215 if (!strcmp(bfd->name, name)) {
1223 bfd_put_details(struct ds *ds, const struct bfd *bfd) OVS_REQUIRES(mutex)
1225 ds_put_format(ds, "\tForwarding: %s\n",
1226 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
1227 ? "true" : "false");
1228 ds_put_format(ds, "\tDetect Multiplier: %d\n", bfd->mult);
1229 ds_put_format(ds, "\tConcatenated Path Down: %s\n",
1230 bfd->cpath_down ? "true" : "false");
1231 ds_put_format(ds, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd));
1232 ds_put_format(ds, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd));
1233 ds_put_format(ds, "\tDetect Time: now %+lldms\n",
1234 time_msec() - bfd->detect_time);
1235 ds_put_format(ds, "\tNext TX Time: now %+lldms\n",
1236 time_msec() - bfd->next_tx);
1237 ds_put_format(ds, "\tLast TX Time: now %+lldms\n",
1238 time_msec() - bfd->last_tx);
1240 ds_put_cstr(ds, "\n");
1242 ds_put_format(ds, "\tLocal Flags: %s\n", bfd_flag_str(bfd->flags));
1243 ds_put_format(ds, "\tLocal Session State: %s\n",
1244 bfd_state_str(bfd->state));
1245 ds_put_format(ds, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd->diag));
1246 ds_put_format(ds, "\tLocal Discriminator: 0x%"PRIx32"\n", bfd->disc);
1247 ds_put_format(ds, "\tLocal Minimum TX Interval: %lldms\n",
1249 ds_put_format(ds, "\tLocal Minimum RX Interval: %lldms\n", bfd->min_rx);
1251 ds_put_cstr(ds, "\n");
1253 ds_put_format(ds, "\tRemote Flags: %s\n", bfd_flag_str(bfd->rmt_flags));
1254 ds_put_format(ds, "\tRemote Session State: %s\n",
1255 bfd_state_str(bfd->rmt_state));
1256 ds_put_format(ds, "\tRemote Diagnostic: %s\n",
1257 bfd_diag_str(bfd->rmt_diag));
1258 ds_put_format(ds, "\tRemote Discriminator: 0x%"PRIx32"\n", bfd->rmt_disc);
1259 ds_put_format(ds, "\tRemote Minimum TX Interval: %lldms\n",
1261 ds_put_format(ds, "\tRemote Minimum RX Interval: %lldms\n",
1266 bfd_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1267 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1269 struct ds ds = DS_EMPTY_INITIALIZER;
1272 ovs_mutex_lock(&mutex);
1274 bfd = bfd_find_by_name(argv[1]);
1276 unixctl_command_reply_error(conn, "no such bfd object");
1279 bfd_put_details(&ds, bfd);
1281 HMAP_FOR_EACH (bfd, node, all_bfds) {
1282 ds_put_format(&ds, "---- %s ----\n", bfd->name);
1283 bfd_put_details(&ds, bfd);
1286 unixctl_command_reply(conn, ds_cstr(&ds));
1290 ovs_mutex_unlock(&mutex);
1295 bfd_unixctl_set_forwarding_override(struct unixctl_conn *conn, int argc,
1296 const char *argv[], void *aux OVS_UNUSED)
1299 const char *forward_str = argv[argc - 1];
1300 int forwarding_override;
1303 ovs_mutex_lock(&mutex);
1304 if (!strcasecmp("true", forward_str)) {
1305 forwarding_override = 1;
1306 } else if (!strcasecmp("false", forward_str)) {
1307 forwarding_override = 0;
1308 } else if (!strcasecmp("normal", forward_str)) {
1309 forwarding_override = -1;
1311 unixctl_command_reply_error(conn, "unknown fault string");
1316 bfd = bfd_find_by_name(argv[1]);
1318 unixctl_command_reply_error(conn, "no such BFD object");
1321 bfd->forwarding_override = forwarding_override;
1322 bfd_status_changed(bfd);
1324 HMAP_FOR_EACH (bfd, node, all_bfds) {
1325 bfd->forwarding_override = forwarding_override;
1326 bfd_status_changed(bfd);
1330 unixctl_command_reply(conn, "OK");
1333 ovs_mutex_unlock(&mutex);