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"
24 #include "connectivity.h"
27 #include "dynamic-string.h"
36 #include "ovs-thread.h"
37 #include "openvswitch/types.h"
39 #include "poll-loop.h"
44 #include "unaligned.h"
49 VLOG_DEFINE_THIS_MODULE(bfd);
53 * The goal of this module is to replace CFM with something both more flexible
54 * and standards compliant. In service of this goal, the following needs to be
58 * * Implement Demand mode.
59 * * Go through the RFC line by line and verify we comply.
60 * * Test against a hardware implementation. Preferably a popular one.
61 * * Delete BFD packets with nw_ttl != 255 in the datapath to prevent DOS
66 * - Set TOS/PCP on the outer tunnel header when encapped.
68 * - Sending BFD messages should be in its own thread/process.
70 * - Scale testing. How does it operate when there are large number of bfd
71 * sessions? Do we ever have random flaps? What's the CPU utilization?
73 * - Rely on data traffic for liveness by using BFD demand mode.
74 * If we're receiving traffic on a port, we can safely assume it's up (modulo
75 * unidrectional failures). BFD has a demand mode in which it can stay quiet
76 * unless it feels the need to check the status of the port. Using this, we
77 * can implement a strategy in which BFD only sends control messages on dark
80 * - Depending on how one interprets the spec, it appears that a BFD session
81 * can never change bfd.LocalDiag to "No Diagnostic". We should verify that
82 * this is what hardware implementations actually do. Seems like "No
83 * Diagnostic" should be set once a BFD session state goes UP. */
88 FLAG_MULTIPOINT = 1 << 0,
97 STATE_ADMIN_DOWN = 0 << 6,
104 DIAG_NONE = 0, /* No Diagnostic. */
105 DIAG_EXPIRED = 1, /* Control Detection Time Expired. */
106 DIAG_ECHO_FAILED = 2, /* Echo Function Failed. */
107 DIAG_RMT_DOWN = 3, /* Neighbor Signaled Session Down. */
108 DIAG_FWD_RESET = 4, /* Forwarding Plane Reset. */
109 DIAG_PATH_DOWN = 5, /* Path Down. */
110 DIAG_CPATH_DOWN = 6, /* Concatenated Path Down. */
111 DIAG_ADMIN_DOWN = 7, /* Administratively Down. */
112 DIAG_RCPATH_DOWN = 8 /* Reverse Concatenated Path Down. */
115 /* RFC 5880 Section 4.1
117 * 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
118 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
119 * |Vers | Diag |Sta|P|F|C|A|D|M| Detect Mult | Length |
120 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
121 * | My Discriminator |
122 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
123 * | Your Discriminator |
124 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
125 * | Desired Min TX Interval |
126 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
127 * | Required Min RX Interval |
128 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
129 * | Required Min Echo RX Interval |
130 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ */
132 uint8_t vers_diag; /* Version and diagnostic. */
133 uint8_t flags; /* 2bit State field followed by flags. */
134 uint8_t mult; /* Fault detection multiplier. */
135 uint8_t length; /* Length of this BFD message. */
136 ovs_be32 my_disc; /* My discriminator. */
137 ovs_be32 your_disc; /* Your discriminator. */
138 ovs_be32 min_tx; /* Desired minimum tx interval. */
139 ovs_be32 min_rx; /* Required minimum rx interval. */
140 ovs_be32 min_rx_echo; /* Required minimum echo rx interval. */
142 BUILD_ASSERT_DECL(BFD_PACKET_LEN == sizeof(struct msg));
144 #define DIAG_MASK 0x1f
146 #define STATE_MASK 0xC0
147 #define FLAGS_MASK 0x3f
150 struct hmap_node node; /* In 'all_bfds'. */
151 uint32_t disc; /* bfd.LocalDiscr. Key in 'all_bfds' hmap. */
153 char *name; /* Name used for logging. */
155 bool cpath_down; /* Concatenated Path Down. */
156 uint8_t mult; /* bfd.DetectMult. */
158 struct netdev *netdev;
159 uint64_t rx_packets; /* Packets received by 'netdev'. */
161 enum state state; /* bfd.SessionState. */
162 enum state rmt_state; /* bfd.RemoteSessionState. */
164 enum diag diag; /* bfd.LocalDiag. */
165 enum diag rmt_diag; /* Remote diagnostic. */
167 enum flags flags; /* Flags sent on messages. */
168 enum flags rmt_flags; /* Flags last received. */
170 uint32_t rmt_disc; /* bfd.RemoteDiscr. */
172 uint8_t eth_dst[ETH_ADDR_LEN];/* Ethernet destination address. */
173 bool eth_dst_set; /* 'eth_dst' set through database. */
175 uint16_t udp_src; /* UDP source port. */
177 /* All timers in milliseconds. */
178 long long int rmt_min_rx; /* bfd.RemoteMinRxInterval. */
179 long long int rmt_min_tx; /* Remote minimum TX interval. */
181 long long int cfg_min_tx; /* Configured minimum TX rate. */
182 long long int cfg_min_rx; /* Configured required minimum RX rate. */
183 long long int poll_min_tx; /* Min TX negotating in a poll sequence. */
184 long long int poll_min_rx; /* Min RX negotating in a poll sequence. */
185 long long int min_tx; /* bfd.DesiredMinTxInterval. */
186 long long int min_rx; /* bfd.RequiredMinRxInterval. */
188 long long int last_tx; /* Last TX time. */
189 long long int next_tx; /* Next TX time. */
190 long long int detect_time; /* RFC 5880 6.8.4 Detection time. */
192 bool last_forwarding; /* Last calculation of forwarding flag. */
193 int forwarding_override; /* Manual override of 'forwarding' status. */
195 atomic_bool check_tnl_key; /* Verify tunnel key of inbound packets? */
196 struct ovs_refcount ref_cnt;
198 /* When forward_if_rx is true, bfd_forwarding() will return
199 * true as long as there are incoming packets received.
200 * Note, forwarding_override still has higher priority. */
201 bool forwarding_if_rx;
202 long long int forwarding_if_rx_detect_time;
204 /* BFD decay related variables. */
205 bool in_decay; /* True when bfd is in decay. */
206 int decay_min_rx; /* min_rx is set to decay_min_rx when */
208 int decay_rx_ctl; /* Count bfd packets received within decay */
209 /* detect interval. */
210 uint64_t decay_rx_packets; /* Packets received by 'netdev'. */
211 long long int decay_detect_time; /* Decay detection time. */
213 uint64_t flap_count; /* Counts bfd forwarding flaps. */
216 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
217 static struct hmap all_bfds__ = HMAP_INITIALIZER(&all_bfds__);
218 static struct hmap *const all_bfds OVS_GUARDED_BY(mutex) = &all_bfds__;
220 static bool bfd_forwarding__(struct bfd *) OVS_REQUIRES(mutex);
221 static bool bfd_in_poll(const struct bfd *) OVS_REQUIRES(mutex);
222 static void bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex);
223 static const char *bfd_diag_str(enum diag) OVS_REQUIRES(mutex);
224 static const char *bfd_state_str(enum state) OVS_REQUIRES(mutex);
225 static long long int bfd_min_tx(const struct bfd *) OVS_REQUIRES(mutex);
226 static long long int bfd_tx_interval(const struct bfd *)
228 static long long int bfd_rx_interval(const struct bfd *)
230 static void bfd_set_next_tx(struct bfd *) OVS_REQUIRES(mutex);
231 static void bfd_set_state(struct bfd *, enum state, enum diag)
233 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex);
234 static void bfd_put_details(struct ds *, const struct bfd *)
236 static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQUIRES(mutex);
237 static void bfd_try_decay(struct bfd *) OVS_REQUIRES(mutex);
238 static void bfd_decay_update(struct bfd *) OVS_REQUIRES(mutex);
240 static void bfd_forwarding_if_rx_update(struct bfd *) OVS_REQUIRES(mutex);
241 static void bfd_unixctl_show(struct unixctl_conn *, int argc,
242 const char *argv[], void *aux OVS_UNUSED);
243 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
244 int argc, const char *argv[],
245 void *aux OVS_UNUSED);
246 static void log_msg(enum vlog_level, const struct msg *, const char *message,
247 const struct bfd *) OVS_REQUIRES(mutex);
249 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 20);
251 /* Returns true if the interface on which 'bfd' is running may be used to
252 * forward traffic according to the BFD session state. */
254 bfd_forwarding(struct bfd *bfd) OVS_EXCLUDED(mutex)
258 ovs_mutex_lock(&mutex);
259 ret = bfd_forwarding__(bfd);
260 ovs_mutex_unlock(&mutex);
264 /* When forwarding_if_rx is enabled, if there are packets received,
265 * updates forwarding_if_rx_detect_time. */
267 bfd_account_rx(struct bfd *bfd, const struct dpif_flow_stats *stats)
269 if (stats->n_packets && bfd->forwarding_if_rx) {
270 ovs_mutex_lock(&mutex);
271 bfd_forwarding__(bfd);
272 bfd_forwarding_if_rx_update(bfd);
273 bfd_forwarding__(bfd);
274 ovs_mutex_unlock(&mutex);
278 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
279 * intended for the OVS database. */
281 bfd_get_status(const struct bfd *bfd, struct smap *smap)
284 ovs_mutex_lock(&mutex);
285 smap_add(smap, "forwarding",
286 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
288 smap_add(smap, "state", bfd_state_str(bfd->state));
289 smap_add(smap, "diagnostic", bfd_diag_str(bfd->diag));
290 smap_add_format(smap, "flap_count", "%"PRIu64, bfd->flap_count);
292 if (bfd->state != STATE_DOWN) {
293 smap_add(smap, "remote_state", bfd_state_str(bfd->rmt_state));
294 smap_add(smap, "remote_diagnostic", bfd_diag_str(bfd->rmt_diag));
296 ovs_mutex_unlock(&mutex);
299 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
300 * according to the database configuration contained in 'cfg'. Takes ownership
301 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
302 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
303 * Also returns NULL if cfg is NULL. */
305 bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
306 struct netdev *netdev) OVS_EXCLUDED(mutex)
308 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
309 static atomic_uint16_t udp_src = ATOMIC_VAR_INIT(0);
312 long long int min_tx, min_rx;
313 bool need_poll = false;
314 bool cfg_min_rx_changed = false;
315 bool cpath_down, forwarding_if_rx;
317 uint8_t ea[ETH_ADDR_LEN];
319 if (ovsthread_once_start(&once)) {
320 unixctl_command_register("bfd/show", "[interface]", 0, 1,
321 bfd_unixctl_show, NULL);
322 unixctl_command_register("bfd/set-forwarding",
323 "[interface] normal|false|true", 1, 2,
324 bfd_unixctl_set_forwarding_override, NULL);
325 ovsthread_once_done(&once);
328 if (!cfg || !smap_get_bool(cfg, "enable", false)) {
333 ovs_mutex_lock(&mutex);
335 bfd = xzalloc(sizeof *bfd);
336 bfd->name = xstrdup(name);
337 bfd->forwarding_override = -1;
338 bfd->disc = generate_discriminator();
339 hmap_insert(all_bfds, &bfd->node, bfd->disc);
341 bfd->diag = DIAG_NONE;
344 ovs_refcount_init(&bfd->ref_cnt);
345 bfd->netdev = netdev_ref(netdev);
346 bfd->rx_packets = bfd_rx_packets(bfd);
347 bfd->in_decay = false;
350 /* RFC 5881 section 4
351 * The source port MUST be in the range 49152 through 65535. The same
352 * UDP source port number MUST be used for all BFD Control packets
353 * associated with a particular session. The source port number SHOULD
354 * be unique among all BFD sessions on the system. */
355 atomic_add(&udp_src, 1, &bfd->udp_src);
356 bfd->udp_src = (bfd->udp_src % 16384) + 49152;
358 bfd_set_state(bfd, STATE_DOWN, DIAG_NONE);
360 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
363 atomic_store(&bfd->check_tnl_key,
364 smap_get_bool(cfg, "check_tnl_key", false));
365 min_tx = smap_get_int(cfg, "min_tx", 100);
366 min_tx = MAX(min_tx, 100);
367 if (bfd->cfg_min_tx != min_tx) {
368 bfd->cfg_min_tx = min_tx;
369 if (bfd->state != STATE_UP
370 || (!bfd_in_poll(bfd) && bfd->cfg_min_tx < bfd->min_tx)) {
371 bfd->min_tx = bfd->cfg_min_tx;
376 min_rx = smap_get_int(cfg, "min_rx", 1000);
377 min_rx = MAX(min_rx, 100);
378 if (bfd->cfg_min_rx != min_rx) {
379 bfd->cfg_min_rx = min_rx;
380 if (bfd->state != STATE_UP
381 || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
382 bfd->min_rx = bfd->cfg_min_rx;
384 cfg_min_rx_changed = true;
388 decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
389 if (bfd->decay_min_rx != decay_min_rx || cfg_min_rx_changed) {
390 if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
391 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
392 bfd->name, bfd->cfg_min_rx);
393 bfd->decay_min_rx = 0;
395 bfd->decay_min_rx = decay_min_rx;
398 bfd->in_decay = false;
399 bfd_decay_update(bfd);
403 cpath_down = smap_get_bool(cfg, "cpath_down", false);
404 if (bfd->cpath_down != cpath_down) {
405 bfd->cpath_down = cpath_down;
406 if (bfd->diag == DIAG_NONE || bfd->diag == DIAG_CPATH_DOWN) {
407 bfd_set_state(bfd, bfd->state, DIAG_NONE);
412 hwaddr = smap_get(cfg, "bfd_dst_mac");
413 if (hwaddr && eth_addr_from_string(hwaddr, ea) && !eth_addr_is_zero(ea)) {
414 memcpy(bfd->eth_dst, ea, ETH_ADDR_LEN);
415 bfd->eth_dst_set = true;
416 } else if (bfd->eth_dst_set) {
417 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
418 bfd->eth_dst_set = false;
421 forwarding_if_rx = smap_get_bool(cfg, "forwarding_if_rx", false);
422 if (bfd->forwarding_if_rx != forwarding_if_rx) {
423 bfd->forwarding_if_rx = forwarding_if_rx;
424 if (bfd->state == STATE_UP && bfd->forwarding_if_rx) {
425 bfd_forwarding_if_rx_update(bfd);
427 bfd->forwarding_if_rx_detect_time = 0;
434 ovs_mutex_unlock(&mutex);
439 bfd_ref(const struct bfd *bfd_)
441 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
443 ovs_refcount_ref(&bfd->ref_cnt);
449 bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
451 if (bfd && ovs_refcount_unref(&bfd->ref_cnt) == 1) {
452 ovs_mutex_lock(&mutex);
453 hmap_remove(all_bfds, &bfd->node);
454 netdev_close(bfd->netdev);
455 ovs_refcount_destroy(&bfd->ref_cnt);
458 ovs_mutex_unlock(&mutex);
463 bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
465 poll_timer_wait_until(bfd_wake_time(bfd));
468 /* Returns the next wake up time. */
470 bfd_wake_time(const struct bfd *bfd) OVS_EXCLUDED(mutex)
472 long long int retval;
478 ovs_mutex_lock(&mutex);
479 if (bfd->flags & FLAG_FINAL) {
482 retval = bfd->next_tx;
483 if (bfd->state > STATE_DOWN) {
484 retval = MIN(bfd->detect_time, retval);
487 ovs_mutex_unlock(&mutex);
492 bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
497 ovs_mutex_lock(&mutex);
499 old_in_decay = bfd->in_decay;
501 if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
502 bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
504 bfd_forwarding__(bfd);
506 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
507 * configured, and decay_detect_time is reached. */
508 if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
509 && now >= bfd->decay_detect_time) {
513 if (bfd->min_tx != bfd->cfg_min_tx
514 || (bfd->min_rx != bfd->cfg_min_rx && bfd->min_rx != bfd->decay_min_rx)
515 || bfd->in_decay != old_in_decay) {
518 ovs_mutex_unlock(&mutex);
522 bfd_should_send_packet(const struct bfd *bfd) OVS_EXCLUDED(mutex)
525 ovs_mutex_lock(&mutex);
526 ret = bfd->flags & FLAG_FINAL || time_msec() >= bfd->next_tx;
527 ovs_mutex_unlock(&mutex);
532 bfd_put_packet(struct bfd *bfd, struct ofpbuf *p,
533 uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
535 long long int min_tx, min_rx;
536 struct udp_header *udp;
537 struct eth_header *eth;
538 struct ip_header *ip;
541 ovs_mutex_lock(&mutex);
543 long long int delay = time_msec() - bfd->next_tx;
544 long long int interval = bfd_tx_interval(bfd);
545 if (delay > interval * 3 / 2) {
546 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
547 " control message", bfd->name, delay, interval);
551 /* RFC 5880 Section 6.5
552 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
554 ovs_assert(!(bfd->flags & FLAG_POLL) || !(bfd->flags & FLAG_FINAL));
556 ofpbuf_reserve(p, 2); /* Properly align after the ethernet header. */
557 eth = ofpbuf_put_uninit(p, sizeof *eth);
558 memcpy(eth->eth_src, eth_src, ETH_ADDR_LEN);
559 memcpy(eth->eth_dst, bfd->eth_dst, ETH_ADDR_LEN);
560 eth->eth_type = htons(ETH_TYPE_IP);
562 ip = ofpbuf_put_zeros(p, sizeof *ip);
563 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
564 ip->ip_tot_len = htons(sizeof *ip + sizeof *udp + sizeof *msg);
566 ip->ip_tos = IPTOS_LOWDELAY | IPTOS_THROUGHPUT;
567 ip->ip_proto = IPPROTO_UDP;
568 /* Use link local addresses: */
569 put_16aligned_be32(&ip->ip_src, htonl(0xA9FE0100)); /* 169.254.1.0. */
570 put_16aligned_be32(&ip->ip_dst, htonl(0xA9FE0101)); /* 169.254.1.1. */
571 ip->ip_csum = csum(ip, sizeof *ip);
573 udp = ofpbuf_put_zeros(p, sizeof *udp);
574 udp->udp_src = htons(bfd->udp_src);
575 udp->udp_dst = htons(BFD_DEST_PORT);
576 udp->udp_len = htons(sizeof *udp + sizeof *msg);
578 msg = ofpbuf_put_uninit(p, sizeof *msg);
579 msg->vers_diag = (BFD_VERSION << 5) | bfd->diag;
580 msg->flags = (bfd->state & STATE_MASK) | bfd->flags;
582 msg->mult = bfd->mult;
583 msg->length = BFD_PACKET_LEN;
584 msg->my_disc = htonl(bfd->disc);
585 msg->your_disc = htonl(bfd->rmt_disc);
586 msg->min_rx_echo = htonl(0);
588 if (bfd_in_poll(bfd)) {
589 min_tx = bfd->poll_min_tx;
590 min_rx = bfd->poll_min_rx;
592 min_tx = bfd_min_tx(bfd);
593 min_rx = bfd->min_rx;
596 msg->min_tx = htonl(min_tx * 1000);
597 msg->min_rx = htonl(min_rx * 1000);
599 bfd->flags &= ~FLAG_FINAL;
601 log_msg(VLL_DBG, msg, "Sending BFD Message", bfd);
603 bfd->last_tx = time_msec();
604 bfd_set_next_tx(bfd);
605 ovs_mutex_unlock(&mutex);
609 bfd_should_process_flow(const struct bfd *bfd_, const struct flow *flow,
610 struct flow_wildcards *wc)
612 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
615 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
616 if (bfd->eth_dst_set && memcmp(bfd->eth_dst, flow->dl_dst, ETH_ADDR_LEN)) {
620 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
621 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
623 atomic_read(&bfd->check_tnl_key, &check_tnl_key);
625 memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id);
627 return (flow->dl_type == htons(ETH_TYPE_IP)
628 && flow->nw_proto == IPPROTO_UDP
629 && flow->tp_dst == htons(BFD_DEST_PORT)
630 && (!check_tnl_key || flow->tunnel.tun_id == htonll(0)));
634 bfd_process_packet(struct bfd *bfd, const struct flow *flow,
635 const struct ofpbuf *p) OVS_EXCLUDED(mutex)
637 uint32_t rmt_min_rx, pkt_your_disc;
638 enum state rmt_state;
643 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
645 ovs_mutex_lock(&mutex);
646 /* Increments the decay rx counter. */
649 bfd_forwarding__(bfd);
651 if (flow->nw_ttl != 255) {
652 /* XXX Should drop in the kernel to prevent DOS. */
656 msg = ofpbuf_at(p, (uint8_t *)p->l7 - (uint8_t *)p->data, BFD_PACKET_LEN);
658 VLOG_INFO_RL(&rl, "%s: Received too-short BFD control message (only "
659 "%"PRIdPTR" bytes long, at least %d required).",
660 bfd->name, (uint8_t *) ofpbuf_tail(p) - (uint8_t *) p->l7,
665 /* RFC 5880 Section 6.8.6
666 * If the Length field is greater than the payload of the encapsulating
667 * protocol, the packet MUST be discarded.
669 * Note that we make this check implicity. Above we use ofpbuf_at() to
670 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
671 * the encapsulating protocol. Below we require msg->length to be exactly
672 * BFD_PACKET_LEN bytes. */
674 flags = msg->flags & FLAGS_MASK;
675 rmt_state = msg->flags & STATE_MASK;
676 version = msg->vers_diag >> VERS_SHIFT;
678 log_msg(VLL_DBG, msg, "Received BFD control message", bfd);
680 if (version != BFD_VERSION) {
681 log_msg(VLL_WARN, msg, "Incorrect version", bfd);
685 /* Technically this should happen after the length check. We don't support
686 * authentication however, so it's simpler to do the check first. */
687 if (flags & FLAG_AUTH) {
688 log_msg(VLL_WARN, msg, "Authenticated control message with"
689 " authentication disabled", bfd);
693 if (msg->length != BFD_PACKET_LEN) {
694 log_msg(VLL_WARN, msg, "Unexpected length", bfd);
695 if (msg->length < BFD_PACKET_LEN) {
701 log_msg(VLL_WARN, msg, "Zero multiplier", bfd);
705 if (flags & FLAG_MULTIPOINT) {
706 log_msg(VLL_WARN, msg, "Unsupported multipoint flag", bfd);
711 log_msg(VLL_WARN, msg, "NULL my_disc", bfd);
715 pkt_your_disc = ntohl(msg->your_disc);
717 /* Technically, we should use the your discriminator field to figure
718 * out which 'struct bfd' this packet is destined towards. That way a
719 * bfd session could migrate from one interface to another
720 * transparently. This doesn't fit in with the OVS structure very
721 * well, so in this respect, we are not compliant. */
722 if (pkt_your_disc != bfd->disc) {
723 log_msg(VLL_WARN, msg, "Incorrect your_disc", bfd);
726 } else if (rmt_state > STATE_DOWN) {
727 log_msg(VLL_WARN, msg, "Null your_disc", bfd);
731 if (bfd->rmt_state != rmt_state) {
732 seq_change(connectivity_seq_get());
735 bfd->rmt_disc = ntohl(msg->my_disc);
736 bfd->rmt_state = rmt_state;
737 bfd->rmt_flags = flags;
738 bfd->rmt_diag = msg->vers_diag & DIAG_MASK;
740 if (flags & FLAG_FINAL && bfd_in_poll(bfd)) {
741 bfd->min_tx = bfd->poll_min_tx;
742 bfd->min_rx = bfd->poll_min_rx;
743 bfd->flags &= ~FLAG_POLL;
744 log_msg(VLL_INFO, msg, "Poll sequence terminated", bfd);
747 if (flags & FLAG_POLL) {
748 /* RFC 5880 Section 6.5
749 * When the other system receives a Poll, it immediately transmits a
750 * BFD Control packet with the Final (F) bit set, independent of any
751 * periodic BFD Control packets it may be sending
752 * (see section 6.8.7). */
753 bfd->flags &= ~FLAG_POLL;
754 bfd->flags |= FLAG_FINAL;
757 rmt_min_rx = MAX(ntohl(msg->min_rx) / 1000, 1);
758 if (bfd->rmt_min_rx != rmt_min_rx) {
759 bfd->rmt_min_rx = rmt_min_rx;
761 bfd_set_next_tx(bfd);
763 log_msg(VLL_INFO, msg, "New remote min_rx", bfd);
766 bfd->rmt_min_tx = MAX(ntohl(msg->min_tx) / 1000, 1);
767 bfd->detect_time = bfd_rx_interval(bfd) * bfd->mult + time_msec();
769 if (bfd->state == STATE_ADMIN_DOWN) {
770 VLOG_DBG_RL(&rl, "Administratively down, dropping control message.");
774 if (rmt_state == STATE_ADMIN_DOWN) {
775 if (bfd->state != STATE_DOWN) {
776 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
779 switch (bfd->state) {
781 if (rmt_state == STATE_DOWN) {
782 bfd_set_state(bfd, STATE_INIT, bfd->diag);
783 } else if (rmt_state == STATE_INIT) {
784 bfd_set_state(bfd, STATE_UP, bfd->diag);
788 if (rmt_state > STATE_DOWN) {
789 bfd_set_state(bfd, STATE_UP, bfd->diag);
793 if (rmt_state <= STATE_DOWN) {
794 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
795 log_msg(VLL_INFO, msg, "Remote signaled STATE_DOWN", bfd);
798 case STATE_ADMIN_DOWN:
803 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
806 bfd_forwarding__(bfd);
807 ovs_mutex_unlock(&mutex);
810 /* Must be called when the netdev owned by 'bfd' should change. */
812 bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
815 ovs_mutex_lock(&mutex);
816 if (bfd->netdev != netdev) {
817 netdev_close(bfd->netdev);
818 bfd->netdev = netdev_ref(netdev);
819 if (bfd->decay_min_rx && bfd->state == STATE_UP) {
820 bfd_decay_update(bfd);
822 if (bfd->forwarding_if_rx && bfd->state == STATE_UP) {
823 bfd_forwarding_if_rx_update(bfd);
825 bfd->rx_packets = bfd_rx_packets(bfd);
827 ovs_mutex_unlock(&mutex);
831 /* Updates the forwarding flag. If override is not configured and
832 * the forwarding flag value changes, increments the flap count.
834 * Note this function may be called multiple times in a function
835 * (e.g. bfd_account_rx) before and after the bfd state or status
836 * change. This is to capture any forwarding flag flap. */
838 bfd_forwarding__(struct bfd *bfd) OVS_REQUIRES(mutex)
841 bool last_forwarding = bfd->last_forwarding;
843 if (bfd->forwarding_override != -1) {
844 return bfd->forwarding_override == 1;
847 time = bfd->forwarding_if_rx_detect_time;
848 bfd->last_forwarding = (bfd->state == STATE_UP
849 || (bfd->forwarding_if_rx && time > time_msec()))
850 && bfd->rmt_diag != DIAG_PATH_DOWN
851 && bfd->rmt_diag != DIAG_CPATH_DOWN
852 && bfd->rmt_diag != DIAG_RCPATH_DOWN;
853 if (bfd->last_forwarding != last_forwarding) {
855 seq_change(connectivity_seq_get());
857 return bfd->last_forwarding;
862 bfd_in_poll(const struct bfd *bfd) OVS_REQUIRES(mutex)
864 return (bfd->flags & FLAG_POLL) != 0;
868 bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex)
870 if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
871 && !(bfd->flags & FLAG_FINAL)) {
872 bfd->poll_min_tx = bfd->cfg_min_tx;
873 bfd->poll_min_rx = bfd->in_decay ? bfd->decay_min_rx : bfd->cfg_min_rx;
874 bfd->flags |= FLAG_POLL;
876 VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
881 bfd_min_tx(const struct bfd *bfd) OVS_REQUIRES(mutex)
883 /* RFC 5880 Section 6.8.3
884 * When bfd.SessionState is not Up, the system MUST set
885 * bfd.DesiredMinTxInterval to a value of not less than one second
886 * (1,000,000 microseconds). This is intended to ensure that the
887 * bandwidth consumed by BFD sessions that are not Up is negligible,
888 * particularly in the case where a neighbor may not be running BFD. */
889 return (bfd->state == STATE_UP ? bfd->min_tx : MAX(bfd->min_tx, 1000));
893 bfd_tx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
895 long long int interval = bfd_min_tx(bfd);
896 return MAX(interval, bfd->rmt_min_rx);
900 bfd_rx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
902 return MAX(bfd->min_rx, bfd->rmt_min_tx);
906 bfd_set_next_tx(struct bfd *bfd) OVS_REQUIRES(mutex)
908 long long int interval = bfd_tx_interval(bfd);
909 interval -= interval * random_range(26) / 100;
910 bfd->next_tx = bfd->last_tx + interval;
914 bfd_flag_str(enum flags flags)
916 struct ds ds = DS_EMPTY_INITIALIZER;
917 static char flag_str[128];
923 if (flags & FLAG_MULTIPOINT) {
924 ds_put_cstr(&ds, "multipoint ");
927 if (flags & FLAG_DEMAND) {
928 ds_put_cstr(&ds, "demand ");
931 if (flags & FLAG_AUTH) {
932 ds_put_cstr(&ds, "auth ");
935 if (flags & FLAG_CTL) {
936 ds_put_cstr(&ds, "ctl ");
939 if (flags & FLAG_FINAL) {
940 ds_put_cstr(&ds, "final ");
943 if (flags & FLAG_POLL) {
944 ds_put_cstr(&ds, "poll ");
947 /* Do not copy the trailing whitespace. */
949 ovs_strlcpy(flag_str, ds_cstr(&ds), sizeof flag_str);
955 bfd_state_str(enum state state)
958 case STATE_ADMIN_DOWN: return "admin_down";
959 case STATE_DOWN: return "down";
960 case STATE_INIT: return "init";
961 case STATE_UP: return "up";
962 default: return "invalid";
967 bfd_diag_str(enum diag diag) {
969 case DIAG_NONE: return "No Diagnostic";
970 case DIAG_EXPIRED: return "Control Detection Time Expired";
971 case DIAG_ECHO_FAILED: return "Echo Function Failed";
972 case DIAG_RMT_DOWN: return "Neighbor Signaled Session Down";
973 case DIAG_FWD_RESET: return "Forwarding Plane Reset";
974 case DIAG_PATH_DOWN: return "Path Down";
975 case DIAG_CPATH_DOWN: return "Concatenated Path Down";
976 case DIAG_ADMIN_DOWN: return "Administratively Down";
977 case DIAG_RCPATH_DOWN: return "Reverse Concatenated Path Down";
978 default: return "Invalid Diagnostic";
983 log_msg(enum vlog_level level, const struct msg *p, const char *message,
984 const struct bfd *bfd) OVS_REQUIRES(mutex)
986 struct ds ds = DS_EMPTY_INITIALIZER;
988 if (vlog_should_drop(THIS_MODULE, level, &rl)) {
994 "\n\tvers:%"PRIu8" diag:\"%s\" state:%s mult:%"PRIu8
997 "\n\tmy_disc:0x%"PRIx32" your_disc:0x%"PRIx32
998 "\n\tmin_tx:%"PRIu32"us (%"PRIu32"ms)"
999 "\n\tmin_rx:%"PRIu32"us (%"PRIu32"ms)"
1000 "\n\tmin_rx_echo:%"PRIu32"us (%"PRIu32"ms)",
1001 bfd->name, message, p->vers_diag >> VERS_SHIFT,
1002 bfd_diag_str(p->vers_diag & DIAG_MASK),
1003 bfd_state_str(p->flags & STATE_MASK),
1004 p->mult, p->length, bfd_flag_str(p->flags & FLAGS_MASK),
1005 ntohl(p->my_disc), ntohl(p->your_disc),
1006 ntohl(p->min_tx), ntohl(p->min_tx) / 1000,
1007 ntohl(p->min_rx), ntohl(p->min_rx) / 1000,
1008 ntohl(p->min_rx_echo), ntohl(p->min_rx_echo) / 1000);
1009 bfd_put_details(&ds, bfd);
1010 VLOG(level, "%s", ds_cstr(&ds));
1015 bfd_set_state(struct bfd *bfd, enum state state, enum diag diag)
1018 if (diag == DIAG_NONE && bfd->cpath_down) {
1019 diag = DIAG_CPATH_DOWN;
1022 if (bfd->state != state || bfd->diag != diag) {
1023 if (!VLOG_DROP_INFO(&rl)) {
1024 struct ds ds = DS_EMPTY_INITIALIZER;
1026 ds_put_format(&ds, "%s: BFD state change: %s->%s"
1027 " \"%s\"->\"%s\".\n",
1028 bfd->name, bfd_state_str(bfd->state),
1029 bfd_state_str(state), bfd_diag_str(bfd->diag),
1030 bfd_diag_str(diag));
1031 bfd_put_details(&ds, bfd);
1032 VLOG_INFO("%s", ds_cstr(&ds));
1039 if (bfd->state <= STATE_DOWN) {
1040 bfd->rmt_state = STATE_DOWN;
1041 bfd->rmt_diag = DIAG_NONE;
1042 bfd->rmt_min_rx = 1;
1045 bfd->rmt_min_tx = 0;
1046 /* Resets the min_rx if in_decay. */
1047 if (bfd->in_decay) {
1048 bfd->min_rx = bfd->cfg_min_rx;
1049 bfd->in_decay = false;
1052 /* Resets the decay when state changes to STATE_UP
1053 * and decay_min_rx is configured. */
1054 if (bfd->state == STATE_UP && bfd->decay_min_rx) {
1055 bfd_decay_update(bfd);
1058 seq_change(connectivity_seq_get());
1063 bfd_rx_packets(const struct bfd *bfd) OVS_REQUIRES(mutex)
1065 struct netdev_stats stats;
1067 if (!netdev_get_stats(bfd->netdev, &stats)) {
1068 return stats.rx_packets;
1074 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
1075 * the 'expect' value. */
1077 bfd_try_decay(struct bfd *bfd) OVS_REQUIRES(mutex)
1079 int64_t diff, expect;
1081 /* The 'diff' is the difference between current interface rx_packets
1082 * stats and last-time check. The 'expect' is the recorded number of
1083 * bfd control packets received within an approximately decay_min_rx
1084 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
1086 * Since the update of rx_packets stats at interface happens
1087 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1088 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1089 * more wiggle room. */
1090 diff = bfd_rx_packets(bfd) - bfd->decay_rx_packets;
1091 expect = 2 * MAX(bfd->decay_rx_ctl, 1);
1092 bfd->in_decay = diff <= expect ? true : false;
1093 bfd_decay_update(bfd);
1096 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1098 bfd_decay_update(struct bfd * bfd) OVS_REQUIRES(mutex)
1100 bfd->decay_rx_packets = bfd_rx_packets(bfd);
1101 bfd->decay_rx_ctl = 0;
1102 bfd->decay_detect_time = MAX(bfd->decay_min_rx, 2000) + time_msec();
1106 bfd_forwarding_if_rx_update(struct bfd *bfd) OVS_REQUIRES(mutex)
1108 int64_t incr = bfd_rx_interval(bfd) * bfd->mult;
1109 bfd->forwarding_if_rx_detect_time = MAX(incr, 2000) + time_msec();
1113 generate_discriminator(void)
1117 /* RFC 5880 Section 6.8.1
1118 * It SHOULD be set to a random (but still unique) value to improve
1119 * security. The value is otherwise outside the scope of this
1125 /* 'disc' is by definition random, so there's no reason to waste time
1127 disc = random_uint32();
1128 HMAP_FOR_EACH_IN_BUCKET (bfd, node, disc, all_bfds) {
1129 if (bfd->disc == disc) {
1140 bfd_find_by_name(const char *name) OVS_REQUIRES(mutex)
1144 HMAP_FOR_EACH (bfd, node, all_bfds) {
1145 if (!strcmp(bfd->name, name)) {
1153 bfd_put_details(struct ds *ds, const struct bfd *bfd) OVS_REQUIRES(mutex)
1155 ds_put_format(ds, "\tForwarding: %s\n",
1156 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
1157 ? "true" : "false");
1158 ds_put_format(ds, "\tDetect Multiplier: %d\n", bfd->mult);
1159 ds_put_format(ds, "\tConcatenated Path Down: %s\n",
1160 bfd->cpath_down ? "true" : "false");
1161 ds_put_format(ds, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd));
1162 ds_put_format(ds, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd));
1163 ds_put_format(ds, "\tDetect Time: now %+lldms\n",
1164 time_msec() - bfd->detect_time);
1165 ds_put_format(ds, "\tNext TX Time: now %+lldms\n",
1166 time_msec() - bfd->next_tx);
1167 ds_put_format(ds, "\tLast TX Time: now %+lldms\n",
1168 time_msec() - bfd->last_tx);
1170 ds_put_cstr(ds, "\n");
1172 ds_put_format(ds, "\tLocal Flags: %s\n", bfd_flag_str(bfd->flags));
1173 ds_put_format(ds, "\tLocal Session State: %s\n",
1174 bfd_state_str(bfd->state));
1175 ds_put_format(ds, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd->diag));
1176 ds_put_format(ds, "\tLocal Discriminator: 0x%"PRIx32"\n", bfd->disc);
1177 ds_put_format(ds, "\tLocal Minimum TX Interval: %lldms\n",
1179 ds_put_format(ds, "\tLocal Minimum RX Interval: %lldms\n", bfd->min_rx);
1181 ds_put_cstr(ds, "\n");
1183 ds_put_format(ds, "\tRemote Flags: %s\n", bfd_flag_str(bfd->rmt_flags));
1184 ds_put_format(ds, "\tRemote Session State: %s\n",
1185 bfd_state_str(bfd->rmt_state));
1186 ds_put_format(ds, "\tRemote Diagnostic: %s\n",
1187 bfd_diag_str(bfd->rmt_diag));
1188 ds_put_format(ds, "\tRemote Discriminator: 0x%"PRIx32"\n", bfd->rmt_disc);
1189 ds_put_format(ds, "\tRemote Minimum TX Interval: %lldms\n",
1191 ds_put_format(ds, "\tRemote Minimum RX Interval: %lldms\n",
1196 bfd_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1197 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1199 struct ds ds = DS_EMPTY_INITIALIZER;
1202 ovs_mutex_lock(&mutex);
1204 bfd = bfd_find_by_name(argv[1]);
1206 unixctl_command_reply_error(conn, "no such bfd object");
1209 bfd_put_details(&ds, bfd);
1211 HMAP_FOR_EACH (bfd, node, all_bfds) {
1212 ds_put_format(&ds, "---- %s ----\n", bfd->name);
1213 bfd_put_details(&ds, bfd);
1216 unixctl_command_reply(conn, ds_cstr(&ds));
1220 ovs_mutex_unlock(&mutex);
1225 bfd_unixctl_set_forwarding_override(struct unixctl_conn *conn, int argc,
1226 const char *argv[], void *aux OVS_UNUSED)
1229 const char *forward_str = argv[argc - 1];
1230 int forwarding_override;
1233 ovs_mutex_lock(&mutex);
1234 if (!strcasecmp("true", forward_str)) {
1235 forwarding_override = 1;
1236 } else if (!strcasecmp("false", forward_str)) {
1237 forwarding_override = 0;
1238 } else if (!strcasecmp("normal", forward_str)) {
1239 forwarding_override = -1;
1241 unixctl_command_reply_error(conn, "unknown fault string");
1246 bfd = bfd_find_by_name(argv[1]);
1248 unixctl_command_reply_error(conn, "no such BFD object");
1251 bfd->forwarding_override = forwarding_override;
1253 HMAP_FOR_EACH (bfd, node, all_bfds) {
1254 bfd->forwarding_override = forwarding_override;
1258 unixctl_command_reply(conn, "OK");
1261 ovs_mutex_unlock(&mutex);