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>
22 #include <sys/socket.h>
24 #include "byte-order.h"
25 #include "connectivity.h"
28 #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 local_eth_src[ETH_ADDR_LEN]; /* Local eth src address. */
173 uint8_t local_eth_dst[ETH_ADDR_LEN]; /* Local eth dst address. */
175 uint8_t rmt_eth_dst[ETH_ADDR_LEN]; /* Remote eth dst address. */
177 ovs_be32 ip_src; /* IPv4 source address. */
178 ovs_be32 ip_dst; /* IPv4 destination address. */
180 uint16_t udp_src; /* UDP source port. */
182 /* All timers in milliseconds. */
183 long long int rmt_min_rx; /* bfd.RemoteMinRxInterval. */
184 long long int rmt_min_tx; /* Remote minimum TX interval. */
186 long long int cfg_min_tx; /* Configured minimum TX rate. */
187 long long int cfg_min_rx; /* Configured required minimum RX rate. */
188 long long int poll_min_tx; /* Min TX negotating in a poll sequence. */
189 long long int poll_min_rx; /* Min RX negotating in a poll sequence. */
190 long long int min_tx; /* bfd.DesiredMinTxInterval. */
191 long long int min_rx; /* bfd.RequiredMinRxInterval. */
193 long long int last_tx; /* Last TX time. */
194 long long int next_tx; /* Next TX time. */
195 long long int detect_time; /* RFC 5880 6.8.4 Detection time. */
197 bool last_forwarding; /* Last calculation of forwarding flag. */
198 int forwarding_override; /* Manual override of 'forwarding' status. */
200 atomic_bool check_tnl_key; /* Verify tunnel key of inbound packets? */
201 struct ovs_refcount ref_cnt;
203 /* When forward_if_rx is true, bfd_forwarding() will return
204 * true as long as there are incoming packets received.
205 * Note, forwarding_override still has higher priority. */
206 bool forwarding_if_rx;
207 long long int forwarding_if_rx_detect_time;
209 /* When 'bfd->forwarding_if_rx' is set, at least one bfd control packet
210 * is required to be received every 100 * bfd->cfg_min_rx. If bfd
211 * control packet is not received within this interval, even if data
212 * packets are received, the bfd->forwarding will still be false. */
213 long long int demand_rx_bfd_time;
215 /* BFD decay related variables. */
216 bool in_decay; /* True when bfd is in decay. */
217 int decay_min_rx; /* min_rx is set to decay_min_rx when */
219 int decay_rx_ctl; /* Count bfd packets received within decay */
220 /* detect interval. */
221 uint64_t decay_rx_packets; /* Packets received by 'netdev'. */
222 long long int decay_detect_time; /* Decay detection time. */
224 uint64_t flap_count; /* Counts bfd forwarding flaps. */
226 /* True when the variables returned by bfd_get_status() are changed
227 * since last check. */
231 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
232 static struct hmap all_bfds__ = HMAP_INITIALIZER(&all_bfds__);
233 static struct hmap *const all_bfds OVS_GUARDED_BY(mutex) = &all_bfds__;
235 static bool bfd_lookup_ip(const char *host_name, struct in_addr *)
237 static bool bfd_forwarding__(struct bfd *) OVS_REQUIRES(mutex);
238 static bool bfd_in_poll(const struct bfd *) OVS_REQUIRES(mutex);
239 static void bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex);
240 static const char *bfd_diag_str(enum diag) OVS_REQUIRES(mutex);
241 static const char *bfd_state_str(enum state) OVS_REQUIRES(mutex);
242 static long long int bfd_min_tx(const struct bfd *) OVS_REQUIRES(mutex);
243 static long long int bfd_tx_interval(const struct bfd *)
245 static long long int bfd_rx_interval(const struct bfd *)
247 static void bfd_set_next_tx(struct bfd *) OVS_REQUIRES(mutex);
248 static void bfd_set_state(struct bfd *, enum state, enum diag)
250 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex);
251 static void bfd_put_details(struct ds *, const struct bfd *)
253 static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQUIRES(mutex);
254 static void bfd_try_decay(struct bfd *) OVS_REQUIRES(mutex);
255 static void bfd_decay_update(struct bfd *) OVS_REQUIRES(mutex);
256 static void bfd_status_changed(struct bfd *) OVS_REQUIRES(mutex);
258 static void bfd_forwarding_if_rx_update(struct bfd *) OVS_REQUIRES(mutex);
259 static void bfd_unixctl_show(struct unixctl_conn *, int argc,
260 const char *argv[], void *aux OVS_UNUSED);
261 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
262 int argc, const char *argv[],
263 void *aux OVS_UNUSED);
264 static void log_msg(enum vlog_level, const struct msg *, const char *message,
265 const struct bfd *) OVS_REQUIRES(mutex);
267 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 20);
269 /* Returns true if the interface on which 'bfd' is running may be used to
270 * forward traffic according to the BFD session state. */
272 bfd_forwarding(struct bfd *bfd) OVS_EXCLUDED(mutex)
276 ovs_mutex_lock(&mutex);
277 ret = bfd_forwarding__(bfd);
278 ovs_mutex_unlock(&mutex);
282 /* When forwarding_if_rx is enabled, if there are packets received,
283 * updates forwarding_if_rx_detect_time. */
285 bfd_account_rx(struct bfd *bfd, const struct dpif_flow_stats *stats)
287 if (stats->n_packets && bfd->forwarding_if_rx) {
288 ovs_mutex_lock(&mutex);
289 bfd_forwarding__(bfd);
290 bfd_forwarding_if_rx_update(bfd);
291 bfd_forwarding__(bfd);
292 ovs_mutex_unlock(&mutex);
296 /* Returns and resets the 'bfd->status_changed'. */
298 bfd_check_status_change(struct bfd *bfd) OVS_EXCLUDED(mutex)
302 ovs_mutex_lock(&mutex);
303 ret = bfd->status_changed;
304 bfd->status_changed = false;
305 ovs_mutex_unlock(&mutex);
310 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
311 * intended for the OVS database. */
313 bfd_get_status(const struct bfd *bfd, struct smap *smap)
316 ovs_mutex_lock(&mutex);
317 smap_add(smap, "forwarding",
318 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
320 smap_add(smap, "state", bfd_state_str(bfd->state));
321 smap_add(smap, "diagnostic", bfd_diag_str(bfd->diag));
322 smap_add_format(smap, "flap_count", "%"PRIu64, bfd->flap_count);
324 if (bfd->state != STATE_DOWN) {
325 smap_add(smap, "remote_state", bfd_state_str(bfd->rmt_state));
326 smap_add(smap, "remote_diagnostic", bfd_diag_str(bfd->rmt_diag));
328 ovs_mutex_unlock(&mutex);
331 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
332 * according to the database configuration contained in 'cfg'. Takes ownership
333 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
334 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
335 * Also returns NULL if cfg is NULL. */
337 bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
338 struct netdev *netdev) OVS_EXCLUDED(mutex)
340 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
341 static atomic_count udp_src = ATOMIC_COUNT_INIT(0);
344 long long int min_tx, min_rx;
345 bool need_poll = false;
346 bool cfg_min_rx_changed = false;
347 bool cpath_down, forwarding_if_rx;
348 const char *hwaddr, *ip_src, *ip_dst;
349 struct in_addr in_addr;
350 uint8_t ea[ETH_ADDR_LEN];
352 if (ovsthread_once_start(&once)) {
353 unixctl_command_register("bfd/show", "[interface]", 0, 1,
354 bfd_unixctl_show, NULL);
355 unixctl_command_register("bfd/set-forwarding",
356 "[interface] normal|false|true", 1, 2,
357 bfd_unixctl_set_forwarding_override, NULL);
358 ovsthread_once_done(&once);
361 if (!cfg || !smap_get_bool(cfg, "enable", false)) {
366 ovs_mutex_lock(&mutex);
368 bfd = xzalloc(sizeof *bfd);
369 bfd->name = xstrdup(name);
370 bfd->forwarding_override = -1;
371 bfd->disc = generate_discriminator();
372 hmap_insert(all_bfds, &bfd->node, bfd->disc);
374 bfd->diag = DIAG_NONE;
377 ovs_refcount_init(&bfd->ref_cnt);
378 bfd->netdev = netdev_ref(netdev);
379 bfd->rx_packets = bfd_rx_packets(bfd);
380 bfd->in_decay = false;
383 /* RFC 5881 section 4
384 * The source port MUST be in the range 49152 through 65535. The same
385 * UDP source port number MUST be used for all BFD Control packets
386 * associated with a particular session. The source port number SHOULD
387 * be unique among all BFD sessions on the system. */
388 bfd->udp_src = (atomic_count_inc(&udp_src) % 16384) + 49152;
390 bfd_set_state(bfd, STATE_DOWN, DIAG_NONE);
392 bfd_status_changed(bfd);
395 atomic_store_relaxed(&bfd->check_tnl_key,
396 smap_get_bool(cfg, "check_tnl_key", false));
397 min_tx = smap_get_int(cfg, "min_tx", 100);
398 min_tx = MAX(min_tx, 1);
399 if (bfd->cfg_min_tx != min_tx) {
400 bfd->cfg_min_tx = min_tx;
401 if (bfd->state != STATE_UP
402 || (!bfd_in_poll(bfd) && bfd->cfg_min_tx < bfd->min_tx)) {
403 bfd->min_tx = bfd->cfg_min_tx;
408 min_rx = smap_get_int(cfg, "min_rx", 1000);
409 min_rx = MAX(min_rx, 1);
410 if (bfd->cfg_min_rx != min_rx) {
411 bfd->cfg_min_rx = min_rx;
412 if (bfd->state != STATE_UP
413 || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
414 bfd->min_rx = bfd->cfg_min_rx;
416 cfg_min_rx_changed = true;
420 decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
421 if (bfd->decay_min_rx != decay_min_rx || cfg_min_rx_changed) {
422 if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
423 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
424 bfd->name, bfd->cfg_min_rx);
425 bfd->decay_min_rx = 0;
427 bfd->decay_min_rx = decay_min_rx;
430 bfd->in_decay = false;
431 bfd_decay_update(bfd);
435 cpath_down = smap_get_bool(cfg, "cpath_down", false);
436 if (bfd->cpath_down != cpath_down) {
437 bfd->cpath_down = cpath_down;
438 bfd_set_state(bfd, bfd->state, DIAG_NONE);
442 hwaddr = smap_get(cfg, "bfd_local_src_mac");
443 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
444 memcpy(bfd->local_eth_src, ea, ETH_ADDR_LEN);
446 memset(bfd->local_eth_src, 0, ETH_ADDR_LEN);
449 hwaddr = smap_get(cfg, "bfd_local_dst_mac");
450 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
451 memcpy(bfd->local_eth_dst, ea, ETH_ADDR_LEN);
453 memset(bfd->local_eth_dst, 0, ETH_ADDR_LEN);
456 hwaddr = smap_get(cfg, "bfd_remote_dst_mac");
457 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
458 memcpy(bfd->rmt_eth_dst, ea, ETH_ADDR_LEN);
460 memset(bfd->rmt_eth_dst, 0, ETH_ADDR_LEN);
463 ip_src = smap_get(cfg, "bfd_src_ip");
464 if (ip_src && bfd_lookup_ip(ip_src, &in_addr)) {
465 memcpy(&bfd->ip_src, &in_addr, sizeof in_addr);
467 bfd->ip_src = htonl(0xA9FE0101); /* 169.254.1.1. */
470 ip_dst = smap_get(cfg, "bfd_dst_ip");
471 if (ip_dst && bfd_lookup_ip(ip_dst, &in_addr)) {
472 memcpy(&bfd->ip_dst, &in_addr, sizeof in_addr);
474 bfd->ip_dst = htonl(0xA9FE0100); /* 169.254.1.0. */
477 forwarding_if_rx = smap_get_bool(cfg, "forwarding_if_rx", false);
478 if (bfd->forwarding_if_rx != forwarding_if_rx) {
479 bfd->forwarding_if_rx = forwarding_if_rx;
480 if (bfd->state == STATE_UP && bfd->forwarding_if_rx) {
481 bfd_forwarding_if_rx_update(bfd);
483 bfd->forwarding_if_rx_detect_time = 0;
490 ovs_mutex_unlock(&mutex);
495 bfd_ref(const struct bfd *bfd_)
497 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
499 ovs_refcount_ref(&bfd->ref_cnt);
505 bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
507 if (bfd && ovs_refcount_unref_relaxed(&bfd->ref_cnt) == 1) {
508 ovs_mutex_lock(&mutex);
509 bfd_status_changed(bfd);
510 hmap_remove(all_bfds, &bfd->node);
511 netdev_close(bfd->netdev);
514 ovs_mutex_unlock(&mutex);
519 bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
521 poll_timer_wait_until(bfd_wake_time(bfd));
524 /* Returns the next wake up time. */
526 bfd_wake_time(const struct bfd *bfd) OVS_EXCLUDED(mutex)
528 long long int retval;
534 ovs_mutex_lock(&mutex);
535 if (bfd->flags & FLAG_FINAL) {
538 retval = bfd->next_tx;
539 if (bfd->state > STATE_DOWN) {
540 retval = MIN(bfd->detect_time, retval);
543 ovs_mutex_unlock(&mutex);
548 bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
553 ovs_mutex_lock(&mutex);
555 old_in_decay = bfd->in_decay;
557 if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
558 bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
560 bfd_forwarding__(bfd);
562 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
563 * configured, and decay_detect_time is reached. */
564 if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
565 && now >= bfd->decay_detect_time) {
569 if (bfd->min_tx != bfd->cfg_min_tx
570 || (bfd->min_rx != bfd->cfg_min_rx && bfd->min_rx != bfd->decay_min_rx)
571 || bfd->in_decay != old_in_decay) {
574 ovs_mutex_unlock(&mutex);
578 bfd_should_send_packet(const struct bfd *bfd) OVS_EXCLUDED(mutex)
581 ovs_mutex_lock(&mutex);
582 ret = bfd->flags & FLAG_FINAL || time_msec() >= bfd->next_tx;
583 ovs_mutex_unlock(&mutex);
588 bfd_put_packet(struct bfd *bfd, struct ofpbuf *p,
589 uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
591 long long int min_tx, min_rx;
592 struct udp_header *udp;
593 struct eth_header *eth;
594 struct ip_header *ip;
597 ovs_mutex_lock(&mutex);
599 long long int delay = time_msec() - bfd->next_tx;
600 long long int interval = bfd_tx_interval(bfd);
601 if (delay > interval * 3 / 2) {
602 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
603 " control message", bfd->name, delay, interval);
607 /* RFC 5880 Section 6.5
608 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
610 ovs_assert(!(bfd->flags & FLAG_POLL) || !(bfd->flags & FLAG_FINAL));
612 ofpbuf_reserve(p, 2); /* Properly align after the ethernet header. */
613 eth = ofpbuf_put_uninit(p, sizeof *eth);
615 eth_addr_is_zero(bfd->local_eth_src) ? eth_src
616 : bfd->local_eth_src,
619 eth_addr_is_zero(bfd->local_eth_dst) ? eth_addr_bfd
620 : bfd->local_eth_dst,
622 eth->eth_type = htons(ETH_TYPE_IP);
624 ip = ofpbuf_put_zeros(p, sizeof *ip);
625 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
626 ip->ip_tot_len = htons(sizeof *ip + sizeof *udp + sizeof *msg);
628 ip->ip_tos = IPTOS_LOWDELAY | IPTOS_THROUGHPUT;
629 ip->ip_proto = IPPROTO_UDP;
630 put_16aligned_be32(&ip->ip_src, bfd->ip_src);
631 put_16aligned_be32(&ip->ip_dst, bfd->ip_dst);
632 ip->ip_csum = csum(ip, sizeof *ip);
634 udp = ofpbuf_put_zeros(p, sizeof *udp);
635 udp->udp_src = htons(bfd->udp_src);
636 udp->udp_dst = htons(BFD_DEST_PORT);
637 udp->udp_len = htons(sizeof *udp + sizeof *msg);
639 msg = ofpbuf_put_uninit(p, sizeof *msg);
640 msg->vers_diag = (BFD_VERSION << 5) | bfd->diag;
641 msg->flags = (bfd->state & STATE_MASK) | bfd->flags;
643 msg->mult = bfd->mult;
644 msg->length = BFD_PACKET_LEN;
645 msg->my_disc = htonl(bfd->disc);
646 msg->your_disc = htonl(bfd->rmt_disc);
647 msg->min_rx_echo = htonl(0);
649 if (bfd_in_poll(bfd)) {
650 min_tx = bfd->poll_min_tx;
651 min_rx = bfd->poll_min_rx;
653 min_tx = bfd_min_tx(bfd);
654 min_rx = bfd->min_rx;
657 msg->min_tx = htonl(min_tx * 1000);
658 msg->min_rx = htonl(min_rx * 1000);
660 bfd->flags &= ~FLAG_FINAL;
662 log_msg(VLL_DBG, msg, "Sending BFD Message", bfd);
664 bfd->last_tx = time_msec();
665 bfd_set_next_tx(bfd);
666 ovs_mutex_unlock(&mutex);
670 bfd_should_process_flow(const struct bfd *bfd_, const struct flow *flow,
671 struct flow_wildcards *wc)
673 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
675 if (!eth_addr_is_zero(bfd->rmt_eth_dst)) {
676 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
678 if (memcmp(bfd->rmt_eth_dst, flow->dl_dst, ETH_ADDR_LEN)) {
683 if (flow->dl_type == htons(ETH_TYPE_IP)) {
684 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
685 if (flow->nw_proto == IPPROTO_UDP) {
686 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
687 if (flow->tp_dst == htons(BFD_DEST_PORT)) {
690 atomic_read_relaxed(&bfd->check_tnl_key, &check_tnl_key);
692 memset(&wc->masks.tunnel.tun_id, 0xff,
693 sizeof wc->masks.tunnel.tun_id);
694 return flow->tunnel.tun_id == htonll(0);
704 bfd_process_packet(struct bfd *bfd, const struct flow *flow,
705 const struct ofpbuf *p) OVS_EXCLUDED(mutex)
707 uint32_t rmt_min_rx, pkt_your_disc;
708 enum state rmt_state;
712 const uint8_t *l7 = ofpbuf_get_udp_payload(p);
715 return; /* No UDP payload. */
718 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
720 ovs_mutex_lock(&mutex);
721 /* Increments the decay rx counter. */
724 bfd_forwarding__(bfd);
726 if (flow->nw_ttl != 255) {
727 /* XXX Should drop in the kernel to prevent DOS. */
731 msg = ofpbuf_at(p, l7 - (uint8_t *)ofpbuf_data(p), BFD_PACKET_LEN);
733 VLOG_INFO_RL(&rl, "%s: Received too-short BFD control message (only "
734 "%"PRIdPTR" bytes long, at least %d required).",
735 bfd->name, (uint8_t *) ofpbuf_tail(p) - l7,
740 /* RFC 5880 Section 6.8.6
741 * If the Length field is greater than the payload of the encapsulating
742 * protocol, the packet MUST be discarded.
744 * Note that we make this check implicity. Above we use ofpbuf_at() to
745 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
746 * the encapsulating protocol. Below we require msg->length to be exactly
747 * BFD_PACKET_LEN bytes. */
749 flags = msg->flags & FLAGS_MASK;
750 rmt_state = msg->flags & STATE_MASK;
751 version = msg->vers_diag >> VERS_SHIFT;
753 log_msg(VLL_DBG, msg, "Received BFD control message", bfd);
755 if (version != BFD_VERSION) {
756 log_msg(VLL_WARN, msg, "Incorrect version", bfd);
760 /* Technically this should happen after the length check. We don't support
761 * authentication however, so it's simpler to do the check first. */
762 if (flags & FLAG_AUTH) {
763 log_msg(VLL_WARN, msg, "Authenticated control message with"
764 " authentication disabled", bfd);
768 if (msg->length != BFD_PACKET_LEN) {
769 log_msg(VLL_WARN, msg, "Unexpected length", bfd);
770 if (msg->length < BFD_PACKET_LEN) {
776 log_msg(VLL_WARN, msg, "Zero multiplier", bfd);
780 if (flags & FLAG_MULTIPOINT) {
781 log_msg(VLL_WARN, msg, "Unsupported multipoint flag", bfd);
786 log_msg(VLL_WARN, msg, "NULL my_disc", bfd);
790 pkt_your_disc = ntohl(msg->your_disc);
792 /* Technically, we should use the your discriminator field to figure
793 * out which 'struct bfd' this packet is destined towards. That way a
794 * bfd session could migrate from one interface to another
795 * transparently. This doesn't fit in with the OVS structure very
796 * well, so in this respect, we are not compliant. */
797 if (pkt_your_disc != bfd->disc) {
798 log_msg(VLL_WARN, msg, "Incorrect your_disc", bfd);
801 } else if (rmt_state > STATE_DOWN) {
802 log_msg(VLL_WARN, msg, "Null your_disc", bfd);
806 if (bfd->rmt_state != rmt_state) {
807 bfd_status_changed(bfd);
810 bfd->rmt_disc = ntohl(msg->my_disc);
811 bfd->rmt_state = rmt_state;
812 bfd->rmt_flags = flags;
813 bfd->rmt_diag = msg->vers_diag & DIAG_MASK;
815 if (flags & FLAG_FINAL && bfd_in_poll(bfd)) {
816 bfd->min_tx = bfd->poll_min_tx;
817 bfd->min_rx = bfd->poll_min_rx;
818 bfd->flags &= ~FLAG_POLL;
819 log_msg(VLL_INFO, msg, "Poll sequence terminated", bfd);
822 if (flags & FLAG_POLL) {
823 /* RFC 5880 Section 6.5
824 * When the other system receives a Poll, it immediately transmits a
825 * BFD Control packet with the Final (F) bit set, independent of any
826 * periodic BFD Control packets it may be sending
827 * (see section 6.8.7). */
828 bfd->flags &= ~FLAG_POLL;
829 bfd->flags |= FLAG_FINAL;
832 rmt_min_rx = MAX(ntohl(msg->min_rx) / 1000, 1);
833 if (bfd->rmt_min_rx != rmt_min_rx) {
834 bfd->rmt_min_rx = rmt_min_rx;
836 bfd_set_next_tx(bfd);
838 log_msg(VLL_INFO, msg, "New remote min_rx", bfd);
841 bfd->rmt_min_tx = MAX(ntohl(msg->min_tx) / 1000, 1);
842 bfd->detect_time = bfd_rx_interval(bfd) * bfd->mult + time_msec();
844 if (bfd->state == STATE_ADMIN_DOWN) {
845 VLOG_DBG_RL(&rl, "Administratively down, dropping control message.");
849 if (rmt_state == STATE_ADMIN_DOWN) {
850 if (bfd->state != STATE_DOWN) {
851 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
854 switch (bfd->state) {
856 if (rmt_state == STATE_DOWN) {
857 bfd_set_state(bfd, STATE_INIT, bfd->diag);
858 } else if (rmt_state == STATE_INIT) {
859 bfd_set_state(bfd, STATE_UP, bfd->diag);
863 if (rmt_state > STATE_DOWN) {
864 bfd_set_state(bfd, STATE_UP, bfd->diag);
868 if (rmt_state <= STATE_DOWN) {
869 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
870 log_msg(VLL_INFO, msg, "Remote signaled STATE_DOWN", bfd);
873 case STATE_ADMIN_DOWN:
878 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
880 if (bfd->forwarding_if_rx) {
881 bfd->demand_rx_bfd_time = time_msec() + 100 * bfd->cfg_min_rx;
885 bfd_forwarding__(bfd);
886 ovs_mutex_unlock(&mutex);
889 /* Must be called when the netdev owned by 'bfd' should change. */
891 bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
894 ovs_mutex_lock(&mutex);
895 if (bfd->netdev != netdev) {
896 netdev_close(bfd->netdev);
897 bfd->netdev = netdev_ref(netdev);
898 if (bfd->decay_min_rx && bfd->state == STATE_UP) {
899 bfd_decay_update(bfd);
901 if (bfd->forwarding_if_rx && bfd->state == STATE_UP) {
902 bfd_forwarding_if_rx_update(bfd);
904 bfd->rx_packets = bfd_rx_packets(bfd);
906 ovs_mutex_unlock(&mutex);
910 /* Updates the forwarding flag. If override is not configured and
911 * the forwarding flag value changes, increments the flap count.
913 * Note this function may be called multiple times in a function
914 * (e.g. bfd_account_rx) before and after the bfd state or status
915 * change. This is to capture any forwarding flag flap. */
917 bfd_forwarding__(struct bfd *bfd) OVS_REQUIRES(mutex)
919 long long int now = time_msec();
920 bool forwarding_if_rx;
921 bool last_forwarding = bfd->last_forwarding;
923 if (bfd->forwarding_override != -1) {
924 return bfd->forwarding_override == 1;
927 forwarding_if_rx = bfd->forwarding_if_rx
928 && bfd->forwarding_if_rx_detect_time > now
929 && bfd->demand_rx_bfd_time > now;
931 bfd->last_forwarding = (bfd->state == STATE_UP || forwarding_if_rx)
932 && bfd->rmt_diag != DIAG_PATH_DOWN
933 && bfd->rmt_diag != DIAG_CPATH_DOWN
934 && bfd->rmt_diag != DIAG_RCPATH_DOWN;
935 if (bfd->last_forwarding != last_forwarding) {
937 bfd_status_changed(bfd);
939 return bfd->last_forwarding;
944 bfd_lookup_ip(const char *host_name, struct in_addr *addr)
946 if (!inet_pton(AF_INET, host_name, addr)) {
947 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
954 bfd_in_poll(const struct bfd *bfd) OVS_REQUIRES(mutex)
956 return (bfd->flags & FLAG_POLL) != 0;
960 bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex)
962 if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
963 && !(bfd->flags & FLAG_FINAL)) {
964 bfd->poll_min_tx = bfd->cfg_min_tx;
965 bfd->poll_min_rx = bfd->in_decay ? bfd->decay_min_rx : bfd->cfg_min_rx;
966 bfd->flags |= FLAG_POLL;
968 VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
973 bfd_min_tx(const struct bfd *bfd) OVS_REQUIRES(mutex)
975 /* RFC 5880 Section 6.8.3
976 * When bfd.SessionState is not Up, the system MUST set
977 * bfd.DesiredMinTxInterval to a value of not less than one second
978 * (1,000,000 microseconds). This is intended to ensure that the
979 * bandwidth consumed by BFD sessions that are not Up is negligible,
980 * particularly in the case where a neighbor may not be running BFD. */
981 return (bfd->state == STATE_UP ? bfd->min_tx : MAX(bfd->min_tx, 1000));
985 bfd_tx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
987 long long int interval = bfd_min_tx(bfd);
988 return MAX(interval, bfd->rmt_min_rx);
992 bfd_rx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
994 return MAX(bfd->min_rx, bfd->rmt_min_tx);
998 bfd_set_next_tx(struct bfd *bfd) OVS_REQUIRES(mutex)
1000 long long int interval = bfd_tx_interval(bfd);
1001 interval -= interval * random_range(26) / 100;
1002 bfd->next_tx = bfd->last_tx + interval;
1006 bfd_flag_str(enum flags flags)
1008 struct ds ds = DS_EMPTY_INITIALIZER;
1009 static char flag_str[128];
1015 if (flags & FLAG_MULTIPOINT) {
1016 ds_put_cstr(&ds, "multipoint ");
1019 if (flags & FLAG_DEMAND) {
1020 ds_put_cstr(&ds, "demand ");
1023 if (flags & FLAG_AUTH) {
1024 ds_put_cstr(&ds, "auth ");
1027 if (flags & FLAG_CTL) {
1028 ds_put_cstr(&ds, "ctl ");
1031 if (flags & FLAG_FINAL) {
1032 ds_put_cstr(&ds, "final ");
1035 if (flags & FLAG_POLL) {
1036 ds_put_cstr(&ds, "poll ");
1039 /* Do not copy the trailing whitespace. */
1041 ovs_strlcpy(flag_str, ds_cstr(&ds), sizeof flag_str);
1047 bfd_state_str(enum state state)
1050 case STATE_ADMIN_DOWN: return "admin_down";
1051 case STATE_DOWN: return "down";
1052 case STATE_INIT: return "init";
1053 case STATE_UP: return "up";
1054 default: return "invalid";
1059 bfd_diag_str(enum diag diag) {
1061 case DIAG_NONE: return "No Diagnostic";
1062 case DIAG_EXPIRED: return "Control Detection Time Expired";
1063 case DIAG_ECHO_FAILED: return "Echo Function Failed";
1064 case DIAG_RMT_DOWN: return "Neighbor Signaled Session Down";
1065 case DIAG_FWD_RESET: return "Forwarding Plane Reset";
1066 case DIAG_PATH_DOWN: return "Path Down";
1067 case DIAG_CPATH_DOWN: return "Concatenated Path Down";
1068 case DIAG_ADMIN_DOWN: return "Administratively Down";
1069 case DIAG_RCPATH_DOWN: return "Reverse Concatenated Path Down";
1070 default: return "Invalid Diagnostic";
1075 log_msg(enum vlog_level level, const struct msg *p, const char *message,
1076 const struct bfd *bfd) OVS_REQUIRES(mutex)
1078 struct ds ds = DS_EMPTY_INITIALIZER;
1080 if (vlog_should_drop(THIS_MODULE, level, &rl)) {
1086 "\n\tvers:%"PRIu8" diag:\"%s\" state:%s mult:%"PRIu8
1089 "\n\tmy_disc:0x%"PRIx32" your_disc:0x%"PRIx32
1090 "\n\tmin_tx:%"PRIu32"us (%"PRIu32"ms)"
1091 "\n\tmin_rx:%"PRIu32"us (%"PRIu32"ms)"
1092 "\n\tmin_rx_echo:%"PRIu32"us (%"PRIu32"ms)",
1093 bfd->name, message, p->vers_diag >> VERS_SHIFT,
1094 bfd_diag_str(p->vers_diag & DIAG_MASK),
1095 bfd_state_str(p->flags & STATE_MASK),
1096 p->mult, p->length, bfd_flag_str(p->flags & FLAGS_MASK),
1097 ntohl(p->my_disc), ntohl(p->your_disc),
1098 ntohl(p->min_tx), ntohl(p->min_tx) / 1000,
1099 ntohl(p->min_rx), ntohl(p->min_rx) / 1000,
1100 ntohl(p->min_rx_echo), ntohl(p->min_rx_echo) / 1000);
1101 bfd_put_details(&ds, bfd);
1102 VLOG(level, "%s", ds_cstr(&ds));
1107 bfd_set_state(struct bfd *bfd, enum state state, enum diag diag)
1110 if (bfd->cpath_down) {
1111 diag = DIAG_CPATH_DOWN;
1114 if (bfd->state != state || bfd->diag != diag) {
1115 if (!VLOG_DROP_INFO(&rl)) {
1116 struct ds ds = DS_EMPTY_INITIALIZER;
1118 ds_put_format(&ds, "%s: BFD state change: %s->%s"
1119 " \"%s\"->\"%s\".\n",
1120 bfd->name, bfd_state_str(bfd->state),
1121 bfd_state_str(state), bfd_diag_str(bfd->diag),
1122 bfd_diag_str(diag));
1123 bfd_put_details(&ds, bfd);
1124 VLOG_INFO("%s", ds_cstr(&ds));
1131 if (bfd->state <= STATE_DOWN) {
1132 bfd->rmt_state = STATE_DOWN;
1133 bfd->rmt_diag = DIAG_NONE;
1134 bfd->rmt_min_rx = 1;
1137 bfd->rmt_min_tx = 0;
1138 /* Resets the min_rx if in_decay. */
1139 if (bfd->in_decay) {
1140 bfd->min_rx = bfd->cfg_min_rx;
1141 bfd->in_decay = false;
1144 /* Resets the decay when state changes to STATE_UP
1145 * and decay_min_rx is configured. */
1146 if (bfd->state == STATE_UP && bfd->decay_min_rx) {
1147 bfd_decay_update(bfd);
1150 bfd_status_changed(bfd);
1155 bfd_rx_packets(const struct bfd *bfd) OVS_REQUIRES(mutex)
1157 struct netdev_stats stats;
1159 if (!netdev_get_stats(bfd->netdev, &stats)) {
1160 return stats.rx_packets;
1166 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
1167 * the 'expect' value. */
1169 bfd_try_decay(struct bfd *bfd) OVS_REQUIRES(mutex)
1171 int64_t diff, expect;
1173 /* The 'diff' is the difference between current interface rx_packets
1174 * stats and last-time check. The 'expect' is the recorded number of
1175 * bfd control packets received within an approximately decay_min_rx
1176 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
1178 * Since the update of rx_packets stats at interface happens
1179 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1180 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1181 * more wiggle room. */
1182 diff = bfd_rx_packets(bfd) - bfd->decay_rx_packets;
1183 expect = 2 * MAX(bfd->decay_rx_ctl, 1);
1184 bfd->in_decay = diff <= expect ? true : false;
1185 bfd_decay_update(bfd);
1188 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1190 bfd_decay_update(struct bfd * bfd) OVS_REQUIRES(mutex)
1192 bfd->decay_rx_packets = bfd_rx_packets(bfd);
1193 bfd->decay_rx_ctl = 0;
1194 bfd->decay_detect_time = MAX(bfd->decay_min_rx, 2000) + time_msec();
1197 /* Records the status change and changes the global connectivity seq. */
1199 bfd_status_changed(struct bfd *bfd) OVS_REQUIRES(mutex)
1201 seq_change(connectivity_seq_get());
1202 bfd->status_changed = true;
1206 bfd_forwarding_if_rx_update(struct bfd *bfd) OVS_REQUIRES(mutex)
1208 int64_t incr = bfd_rx_interval(bfd) * bfd->mult;
1209 bfd->forwarding_if_rx_detect_time = MAX(incr, 2000) + time_msec();
1213 generate_discriminator(void)
1217 /* RFC 5880 Section 6.8.1
1218 * It SHOULD be set to a random (but still unique) value to improve
1219 * security. The value is otherwise outside the scope of this
1225 /* 'disc' is by definition random, so there's no reason to waste time
1227 disc = random_uint32();
1228 HMAP_FOR_EACH_IN_BUCKET (bfd, node, disc, all_bfds) {
1229 if (bfd->disc == disc) {
1240 bfd_find_by_name(const char *name) OVS_REQUIRES(mutex)
1244 HMAP_FOR_EACH (bfd, node, all_bfds) {
1245 if (!strcmp(bfd->name, name)) {
1253 bfd_put_details(struct ds *ds, const struct bfd *bfd) OVS_REQUIRES(mutex)
1255 ds_put_format(ds, "\tForwarding: %s\n",
1256 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
1257 ? "true" : "false");
1258 ds_put_format(ds, "\tDetect Multiplier: %d\n", bfd->mult);
1259 ds_put_format(ds, "\tConcatenated Path Down: %s\n",
1260 bfd->cpath_down ? "true" : "false");
1261 ds_put_format(ds, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd));
1262 ds_put_format(ds, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd));
1263 ds_put_format(ds, "\tDetect Time: now %+lldms\n",
1264 time_msec() - bfd->detect_time);
1265 ds_put_format(ds, "\tNext TX Time: now %+lldms\n",
1266 time_msec() - bfd->next_tx);
1267 ds_put_format(ds, "\tLast TX Time: now %+lldms\n",
1268 time_msec() - bfd->last_tx);
1270 ds_put_cstr(ds, "\n");
1272 ds_put_format(ds, "\tLocal Flags: %s\n", bfd_flag_str(bfd->flags));
1273 ds_put_format(ds, "\tLocal Session State: %s\n",
1274 bfd_state_str(bfd->state));
1275 ds_put_format(ds, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd->diag));
1276 ds_put_format(ds, "\tLocal Discriminator: 0x%"PRIx32"\n", bfd->disc);
1277 ds_put_format(ds, "\tLocal Minimum TX Interval: %lldms\n",
1279 ds_put_format(ds, "\tLocal Minimum RX Interval: %lldms\n", bfd->min_rx);
1281 ds_put_cstr(ds, "\n");
1283 ds_put_format(ds, "\tRemote Flags: %s\n", bfd_flag_str(bfd->rmt_flags));
1284 ds_put_format(ds, "\tRemote Session State: %s\n",
1285 bfd_state_str(bfd->rmt_state));
1286 ds_put_format(ds, "\tRemote Diagnostic: %s\n",
1287 bfd_diag_str(bfd->rmt_diag));
1288 ds_put_format(ds, "\tRemote Discriminator: 0x%"PRIx32"\n", bfd->rmt_disc);
1289 ds_put_format(ds, "\tRemote Minimum TX Interval: %lldms\n",
1291 ds_put_format(ds, "\tRemote Minimum RX Interval: %lldms\n",
1296 bfd_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1297 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1299 struct ds ds = DS_EMPTY_INITIALIZER;
1302 ovs_mutex_lock(&mutex);
1304 bfd = bfd_find_by_name(argv[1]);
1306 unixctl_command_reply_error(conn, "no such bfd object");
1309 bfd_put_details(&ds, bfd);
1311 HMAP_FOR_EACH (bfd, node, all_bfds) {
1312 ds_put_format(&ds, "---- %s ----\n", bfd->name);
1313 bfd_put_details(&ds, bfd);
1316 unixctl_command_reply(conn, ds_cstr(&ds));
1320 ovs_mutex_unlock(&mutex);
1325 bfd_unixctl_set_forwarding_override(struct unixctl_conn *conn, int argc,
1326 const char *argv[], void *aux OVS_UNUSED)
1329 const char *forward_str = argv[argc - 1];
1330 int forwarding_override;
1333 ovs_mutex_lock(&mutex);
1334 if (!strcasecmp("true", forward_str)) {
1335 forwarding_override = 1;
1336 } else if (!strcasecmp("false", forward_str)) {
1337 forwarding_override = 0;
1338 } else if (!strcasecmp("normal", forward_str)) {
1339 forwarding_override = -1;
1341 unixctl_command_reply_error(conn, "unknown fault string");
1346 bfd = bfd_find_by_name(argv[1]);
1348 unixctl_command_reply_error(conn, "no such BFD object");
1351 bfd->forwarding_override = forwarding_override;
1352 bfd_status_changed(bfd);
1354 HMAP_FOR_EACH (bfd, node, all_bfds) {
1355 bfd->forwarding_override = forwarding_override;
1356 bfd_status_changed(bfd);
1360 unixctl_command_reply(conn, "OK");
1363 ovs_mutex_unlock(&mutex);