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 ovs_be32 ip_src; /* IPv4 source address. */
176 ovs_be32 ip_dst; /* IPv4 destination address. */
178 uint16_t udp_src; /* UDP source port. */
180 /* All timers in milliseconds. */
181 long long int rmt_min_rx; /* bfd.RemoteMinRxInterval. */
182 long long int rmt_min_tx; /* Remote minimum TX interval. */
184 long long int cfg_min_tx; /* Configured minimum TX rate. */
185 long long int cfg_min_rx; /* Configured required minimum RX rate. */
186 long long int poll_min_tx; /* Min TX negotating in a poll sequence. */
187 long long int poll_min_rx; /* Min RX negotating in a poll sequence. */
188 long long int min_tx; /* bfd.DesiredMinTxInterval. */
189 long long int min_rx; /* bfd.RequiredMinRxInterval. */
191 long long int last_tx; /* Last TX time. */
192 long long int next_tx; /* Next TX time. */
193 long long int detect_time; /* RFC 5880 6.8.4 Detection time. */
195 bool last_forwarding; /* Last calculation of forwarding flag. */
196 int forwarding_override; /* Manual override of 'forwarding' status. */
198 atomic_bool check_tnl_key; /* Verify tunnel key of inbound packets? */
199 struct ovs_refcount ref_cnt;
201 /* When forward_if_rx is true, bfd_forwarding() will return
202 * true as long as there are incoming packets received.
203 * Note, forwarding_override still has higher priority. */
204 bool forwarding_if_rx;
205 long long int forwarding_if_rx_detect_time;
207 /* BFD decay related variables. */
208 bool in_decay; /* True when bfd is in decay. */
209 int decay_min_rx; /* min_rx is set to decay_min_rx when */
211 int decay_rx_ctl; /* Count bfd packets received within decay */
212 /* detect interval. */
213 uint64_t decay_rx_packets; /* Packets received by 'netdev'. */
214 long long int decay_detect_time; /* Decay detection time. */
216 uint64_t flap_count; /* Counts bfd forwarding flaps. */
219 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
220 static struct hmap all_bfds__ = HMAP_INITIALIZER(&all_bfds__);
221 static struct hmap *const all_bfds OVS_GUARDED_BY(mutex) = &all_bfds__;
223 static bool bfd_lookup_ip(const char *host_name, struct in_addr *)
225 static bool bfd_forwarding__(struct bfd *) OVS_REQUIRES(mutex);
226 static bool bfd_in_poll(const struct bfd *) OVS_REQUIRES(mutex);
227 static void bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex);
228 static const char *bfd_diag_str(enum diag) OVS_REQUIRES(mutex);
229 static const char *bfd_state_str(enum state) OVS_REQUIRES(mutex);
230 static long long int bfd_min_tx(const struct bfd *) OVS_REQUIRES(mutex);
231 static long long int bfd_tx_interval(const struct bfd *)
233 static long long int bfd_rx_interval(const struct bfd *)
235 static void bfd_set_next_tx(struct bfd *) OVS_REQUIRES(mutex);
236 static void bfd_set_state(struct bfd *, enum state, enum diag)
238 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex);
239 static void bfd_put_details(struct ds *, const struct bfd *)
241 static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQUIRES(mutex);
242 static void bfd_try_decay(struct bfd *) OVS_REQUIRES(mutex);
243 static void bfd_decay_update(struct bfd *) OVS_REQUIRES(mutex);
245 static void bfd_forwarding_if_rx_update(struct bfd *) OVS_REQUIRES(mutex);
246 static void bfd_unixctl_show(struct unixctl_conn *, int argc,
247 const char *argv[], void *aux OVS_UNUSED);
248 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
249 int argc, const char *argv[],
250 void *aux OVS_UNUSED);
251 static void log_msg(enum vlog_level, const struct msg *, const char *message,
252 const struct bfd *) OVS_REQUIRES(mutex);
254 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 20);
256 /* Returns true if the interface on which 'bfd' is running may be used to
257 * forward traffic according to the BFD session state. */
259 bfd_forwarding(struct bfd *bfd) OVS_EXCLUDED(mutex)
263 ovs_mutex_lock(&mutex);
264 ret = bfd_forwarding__(bfd);
265 ovs_mutex_unlock(&mutex);
269 /* When forwarding_if_rx is enabled, if there are packets received,
270 * updates forwarding_if_rx_detect_time. */
272 bfd_account_rx(struct bfd *bfd, const struct dpif_flow_stats *stats)
274 if (stats->n_packets && bfd->forwarding_if_rx) {
275 ovs_mutex_lock(&mutex);
276 bfd_forwarding__(bfd);
277 bfd_forwarding_if_rx_update(bfd);
278 bfd_forwarding__(bfd);
279 ovs_mutex_unlock(&mutex);
283 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
284 * intended for the OVS database. */
286 bfd_get_status(const struct bfd *bfd, struct smap *smap)
289 ovs_mutex_lock(&mutex);
290 smap_add(smap, "forwarding",
291 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
293 smap_add(smap, "state", bfd_state_str(bfd->state));
294 smap_add(smap, "diagnostic", bfd_diag_str(bfd->diag));
295 smap_add_format(smap, "flap_count", "%"PRIu64, bfd->flap_count);
297 if (bfd->state != STATE_DOWN) {
298 smap_add(smap, "remote_state", bfd_state_str(bfd->rmt_state));
299 smap_add(smap, "remote_diagnostic", bfd_diag_str(bfd->rmt_diag));
301 ovs_mutex_unlock(&mutex);
304 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
305 * according to the database configuration contained in 'cfg'. Takes ownership
306 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
307 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
308 * Also returns NULL if cfg is NULL. */
310 bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
311 struct netdev *netdev) OVS_EXCLUDED(mutex)
313 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
314 static atomic_uint16_t udp_src = ATOMIC_VAR_INIT(0);
317 long long int min_tx, min_rx;
318 bool need_poll = false;
319 bool cfg_min_rx_changed = false;
320 bool cpath_down, forwarding_if_rx;
321 const char *hwaddr, *ip_src, *ip_dst;
322 struct in_addr in_addr;
323 uint8_t ea[ETH_ADDR_LEN];
325 if (ovsthread_once_start(&once)) {
326 unixctl_command_register("bfd/show", "[interface]", 0, 1,
327 bfd_unixctl_show, NULL);
328 unixctl_command_register("bfd/set-forwarding",
329 "[interface] normal|false|true", 1, 2,
330 bfd_unixctl_set_forwarding_override, NULL);
331 ovsthread_once_done(&once);
334 if (!cfg || !smap_get_bool(cfg, "enable", false)) {
339 ovs_mutex_lock(&mutex);
341 bfd = xzalloc(sizeof *bfd);
342 bfd->name = xstrdup(name);
343 bfd->forwarding_override = -1;
344 bfd->disc = generate_discriminator();
345 hmap_insert(all_bfds, &bfd->node, bfd->disc);
347 bfd->diag = DIAG_NONE;
350 ovs_refcount_init(&bfd->ref_cnt);
351 bfd->netdev = netdev_ref(netdev);
352 bfd->rx_packets = bfd_rx_packets(bfd);
353 bfd->in_decay = false;
356 /* RFC 5881 section 4
357 * The source port MUST be in the range 49152 through 65535. The same
358 * UDP source port number MUST be used for all BFD Control packets
359 * associated with a particular session. The source port number SHOULD
360 * be unique among all BFD sessions on the system. */
361 atomic_add(&udp_src, 1, &bfd->udp_src);
362 bfd->udp_src = (bfd->udp_src % 16384) + 49152;
364 bfd_set_state(bfd, STATE_DOWN, DIAG_NONE);
366 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
369 atomic_store(&bfd->check_tnl_key,
370 smap_get_bool(cfg, "check_tnl_key", false));
371 min_tx = smap_get_int(cfg, "min_tx", 100);
372 min_tx = MAX(min_tx, 100);
373 if (bfd->cfg_min_tx != min_tx) {
374 bfd->cfg_min_tx = min_tx;
375 if (bfd->state != STATE_UP
376 || (!bfd_in_poll(bfd) && bfd->cfg_min_tx < bfd->min_tx)) {
377 bfd->min_tx = bfd->cfg_min_tx;
382 min_rx = smap_get_int(cfg, "min_rx", 1000);
383 min_rx = MAX(min_rx, 100);
384 if (bfd->cfg_min_rx != min_rx) {
385 bfd->cfg_min_rx = min_rx;
386 if (bfd->state != STATE_UP
387 || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
388 bfd->min_rx = bfd->cfg_min_rx;
390 cfg_min_rx_changed = true;
394 decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
395 if (bfd->decay_min_rx != decay_min_rx || cfg_min_rx_changed) {
396 if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
397 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
398 bfd->name, bfd->cfg_min_rx);
399 bfd->decay_min_rx = 0;
401 bfd->decay_min_rx = decay_min_rx;
404 bfd->in_decay = false;
405 bfd_decay_update(bfd);
409 cpath_down = smap_get_bool(cfg, "cpath_down", false);
410 if (bfd->cpath_down != cpath_down) {
411 bfd->cpath_down = cpath_down;
412 bfd_set_state(bfd, bfd->state, DIAG_NONE);
416 hwaddr = smap_get(cfg, "bfd_dst_mac");
417 if (hwaddr && eth_addr_from_string(hwaddr, ea) && !eth_addr_is_zero(ea)) {
418 memcpy(bfd->eth_dst, ea, ETH_ADDR_LEN);
419 bfd->eth_dst_set = true;
420 } else if (bfd->eth_dst_set) {
421 memcpy(bfd->eth_dst, eth_addr_bfd, ETH_ADDR_LEN);
422 bfd->eth_dst_set = false;
425 ip_src = smap_get(cfg, "bfd_src_ip");
426 if (ip_src && bfd_lookup_ip(ip_src, &in_addr)) {
427 memcpy(&bfd->ip_src, &in_addr, sizeof in_addr);
429 bfd->ip_src = htonl(0xA9FE0100); /* 169.254.1.0. */
432 ip_dst = smap_get(cfg, "bfd_dst_ip");
433 if (ip_dst && bfd_lookup_ip(ip_dst, &in_addr)) {
434 memcpy(&bfd->ip_dst, &in_addr, sizeof in_addr);
436 bfd->ip_dst = htonl(0xA9FE0101); /* 169.254.1.1. */
439 forwarding_if_rx = smap_get_bool(cfg, "forwarding_if_rx", false);
440 if (bfd->forwarding_if_rx != forwarding_if_rx) {
441 bfd->forwarding_if_rx = forwarding_if_rx;
442 if (bfd->state == STATE_UP && bfd->forwarding_if_rx) {
443 bfd_forwarding_if_rx_update(bfd);
445 bfd->forwarding_if_rx_detect_time = 0;
452 ovs_mutex_unlock(&mutex);
457 bfd_ref(const struct bfd *bfd_)
459 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
461 ovs_refcount_ref(&bfd->ref_cnt);
467 bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
469 if (bfd && ovs_refcount_unref(&bfd->ref_cnt) == 1) {
470 ovs_mutex_lock(&mutex);
471 hmap_remove(all_bfds, &bfd->node);
472 netdev_close(bfd->netdev);
473 ovs_refcount_destroy(&bfd->ref_cnt);
476 ovs_mutex_unlock(&mutex);
481 bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
483 poll_timer_wait_until(bfd_wake_time(bfd));
486 /* Returns the next wake up time. */
488 bfd_wake_time(const struct bfd *bfd) OVS_EXCLUDED(mutex)
490 long long int retval;
496 ovs_mutex_lock(&mutex);
497 if (bfd->flags & FLAG_FINAL) {
500 retval = bfd->next_tx;
501 if (bfd->state > STATE_DOWN) {
502 retval = MIN(bfd->detect_time, retval);
505 ovs_mutex_unlock(&mutex);
510 bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
515 ovs_mutex_lock(&mutex);
517 old_in_decay = bfd->in_decay;
519 if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
520 bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
522 bfd_forwarding__(bfd);
524 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
525 * configured, and decay_detect_time is reached. */
526 if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
527 && now >= bfd->decay_detect_time) {
531 if (bfd->min_tx != bfd->cfg_min_tx
532 || (bfd->min_rx != bfd->cfg_min_rx && bfd->min_rx != bfd->decay_min_rx)
533 || bfd->in_decay != old_in_decay) {
536 ovs_mutex_unlock(&mutex);
540 bfd_should_send_packet(const struct bfd *bfd) OVS_EXCLUDED(mutex)
543 ovs_mutex_lock(&mutex);
544 ret = bfd->flags & FLAG_FINAL || time_msec() >= bfd->next_tx;
545 ovs_mutex_unlock(&mutex);
550 bfd_put_packet(struct bfd *bfd, struct ofpbuf *p,
551 uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex)
553 long long int min_tx, min_rx;
554 struct udp_header *udp;
555 struct eth_header *eth;
556 struct ip_header *ip;
559 ovs_mutex_lock(&mutex);
561 long long int delay = time_msec() - bfd->next_tx;
562 long long int interval = bfd_tx_interval(bfd);
563 if (delay > interval * 3 / 2) {
564 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
565 " control message", bfd->name, delay, interval);
569 /* RFC 5880 Section 6.5
570 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
572 ovs_assert(!(bfd->flags & FLAG_POLL) || !(bfd->flags & FLAG_FINAL));
574 ofpbuf_reserve(p, 2); /* Properly align after the ethernet header. */
575 eth = ofpbuf_put_uninit(p, sizeof *eth);
576 memcpy(eth->eth_src, eth_src, ETH_ADDR_LEN);
577 memcpy(eth->eth_dst, bfd->eth_dst, ETH_ADDR_LEN);
578 eth->eth_type = htons(ETH_TYPE_IP);
580 ip = ofpbuf_put_zeros(p, sizeof *ip);
581 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
582 ip->ip_tot_len = htons(sizeof *ip + sizeof *udp + sizeof *msg);
584 ip->ip_tos = IPTOS_LOWDELAY | IPTOS_THROUGHPUT;
585 ip->ip_proto = IPPROTO_UDP;
586 put_16aligned_be32(&ip->ip_src, bfd->ip_src);
587 put_16aligned_be32(&ip->ip_dst, bfd->ip_dst);
588 ip->ip_csum = csum(ip, sizeof *ip);
590 udp = ofpbuf_put_zeros(p, sizeof *udp);
591 udp->udp_src = htons(bfd->udp_src);
592 udp->udp_dst = htons(BFD_DEST_PORT);
593 udp->udp_len = htons(sizeof *udp + sizeof *msg);
595 msg = ofpbuf_put_uninit(p, sizeof *msg);
596 msg->vers_diag = (BFD_VERSION << 5) | bfd->diag;
597 msg->flags = (bfd->state & STATE_MASK) | bfd->flags;
599 msg->mult = bfd->mult;
600 msg->length = BFD_PACKET_LEN;
601 msg->my_disc = htonl(bfd->disc);
602 msg->your_disc = htonl(bfd->rmt_disc);
603 msg->min_rx_echo = htonl(0);
605 if (bfd_in_poll(bfd)) {
606 min_tx = bfd->poll_min_tx;
607 min_rx = bfd->poll_min_rx;
609 min_tx = bfd_min_tx(bfd);
610 min_rx = bfd->min_rx;
613 msg->min_tx = htonl(min_tx * 1000);
614 msg->min_rx = htonl(min_rx * 1000);
616 bfd->flags &= ~FLAG_FINAL;
618 log_msg(VLL_DBG, msg, "Sending BFD Message", bfd);
620 bfd->last_tx = time_msec();
621 bfd_set_next_tx(bfd);
622 ovs_mutex_unlock(&mutex);
626 bfd_should_process_flow(const struct bfd *bfd_, const struct flow *flow,
627 struct flow_wildcards *wc)
629 struct bfd *bfd = CONST_CAST(struct bfd *, bfd_);
632 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
633 if (bfd->eth_dst_set && memcmp(bfd->eth_dst, flow->dl_dst, ETH_ADDR_LEN)) {
637 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
638 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
640 atomic_read(&bfd->check_tnl_key, &check_tnl_key);
642 memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id);
644 return (flow->dl_type == htons(ETH_TYPE_IP)
645 && flow->nw_proto == IPPROTO_UDP
646 && flow->tp_dst == htons(BFD_DEST_PORT)
647 && (!check_tnl_key || flow->tunnel.tun_id == htonll(0)));
651 bfd_process_packet(struct bfd *bfd, const struct flow *flow,
652 const struct ofpbuf *p) OVS_EXCLUDED(mutex)
654 uint32_t rmt_min_rx, pkt_your_disc;
655 enum state rmt_state;
660 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
662 ovs_mutex_lock(&mutex);
663 /* Increments the decay rx counter. */
666 bfd_forwarding__(bfd);
668 if (flow->nw_ttl != 255) {
669 /* XXX Should drop in the kernel to prevent DOS. */
673 msg = ofpbuf_at(p, (uint8_t *)p->l7 - (uint8_t *)p->data, BFD_PACKET_LEN);
675 VLOG_INFO_RL(&rl, "%s: Received too-short BFD control message (only "
676 "%"PRIdPTR" bytes long, at least %d required).",
677 bfd->name, (uint8_t *) ofpbuf_tail(p) - (uint8_t *) p->l7,
682 /* RFC 5880 Section 6.8.6
683 * If the Length field is greater than the payload of the encapsulating
684 * protocol, the packet MUST be discarded.
686 * Note that we make this check implicity. Above we use ofpbuf_at() to
687 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
688 * the encapsulating protocol. Below we require msg->length to be exactly
689 * BFD_PACKET_LEN bytes. */
691 flags = msg->flags & FLAGS_MASK;
692 rmt_state = msg->flags & STATE_MASK;
693 version = msg->vers_diag >> VERS_SHIFT;
695 log_msg(VLL_DBG, msg, "Received BFD control message", bfd);
697 if (version != BFD_VERSION) {
698 log_msg(VLL_WARN, msg, "Incorrect version", bfd);
702 /* Technically this should happen after the length check. We don't support
703 * authentication however, so it's simpler to do the check first. */
704 if (flags & FLAG_AUTH) {
705 log_msg(VLL_WARN, msg, "Authenticated control message with"
706 " authentication disabled", bfd);
710 if (msg->length != BFD_PACKET_LEN) {
711 log_msg(VLL_WARN, msg, "Unexpected length", bfd);
712 if (msg->length < BFD_PACKET_LEN) {
718 log_msg(VLL_WARN, msg, "Zero multiplier", bfd);
722 if (flags & FLAG_MULTIPOINT) {
723 log_msg(VLL_WARN, msg, "Unsupported multipoint flag", bfd);
728 log_msg(VLL_WARN, msg, "NULL my_disc", bfd);
732 pkt_your_disc = ntohl(msg->your_disc);
734 /* Technically, we should use the your discriminator field to figure
735 * out which 'struct bfd' this packet is destined towards. That way a
736 * bfd session could migrate from one interface to another
737 * transparently. This doesn't fit in with the OVS structure very
738 * well, so in this respect, we are not compliant. */
739 if (pkt_your_disc != bfd->disc) {
740 log_msg(VLL_WARN, msg, "Incorrect your_disc", bfd);
743 } else if (rmt_state > STATE_DOWN) {
744 log_msg(VLL_WARN, msg, "Null your_disc", bfd);
748 if (bfd->rmt_state != rmt_state) {
749 seq_change(connectivity_seq_get());
752 bfd->rmt_disc = ntohl(msg->my_disc);
753 bfd->rmt_state = rmt_state;
754 bfd->rmt_flags = flags;
755 bfd->rmt_diag = msg->vers_diag & DIAG_MASK;
757 if (flags & FLAG_FINAL && bfd_in_poll(bfd)) {
758 bfd->min_tx = bfd->poll_min_tx;
759 bfd->min_rx = bfd->poll_min_rx;
760 bfd->flags &= ~FLAG_POLL;
761 log_msg(VLL_INFO, msg, "Poll sequence terminated", bfd);
764 if (flags & FLAG_POLL) {
765 /* RFC 5880 Section 6.5
766 * When the other system receives a Poll, it immediately transmits a
767 * BFD Control packet with the Final (F) bit set, independent of any
768 * periodic BFD Control packets it may be sending
769 * (see section 6.8.7). */
770 bfd->flags &= ~FLAG_POLL;
771 bfd->flags |= FLAG_FINAL;
774 rmt_min_rx = MAX(ntohl(msg->min_rx) / 1000, 1);
775 if (bfd->rmt_min_rx != rmt_min_rx) {
776 bfd->rmt_min_rx = rmt_min_rx;
778 bfd_set_next_tx(bfd);
780 log_msg(VLL_INFO, msg, "New remote min_rx", bfd);
783 bfd->rmt_min_tx = MAX(ntohl(msg->min_tx) / 1000, 1);
784 bfd->detect_time = bfd_rx_interval(bfd) * bfd->mult + time_msec();
786 if (bfd->state == STATE_ADMIN_DOWN) {
787 VLOG_DBG_RL(&rl, "Administratively down, dropping control message.");
791 if (rmt_state == STATE_ADMIN_DOWN) {
792 if (bfd->state != STATE_DOWN) {
793 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
796 switch (bfd->state) {
798 if (rmt_state == STATE_DOWN) {
799 bfd_set_state(bfd, STATE_INIT, bfd->diag);
800 } else if (rmt_state == STATE_INIT) {
801 bfd_set_state(bfd, STATE_UP, bfd->diag);
805 if (rmt_state > STATE_DOWN) {
806 bfd_set_state(bfd, STATE_UP, bfd->diag);
810 if (rmt_state <= STATE_DOWN) {
811 bfd_set_state(bfd, STATE_DOWN, DIAG_RMT_DOWN);
812 log_msg(VLL_INFO, msg, "Remote signaled STATE_DOWN", bfd);
815 case STATE_ADMIN_DOWN:
820 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
823 bfd_forwarding__(bfd);
824 ovs_mutex_unlock(&mutex);
827 /* Must be called when the netdev owned by 'bfd' should change. */
829 bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
832 ovs_mutex_lock(&mutex);
833 if (bfd->netdev != netdev) {
834 netdev_close(bfd->netdev);
835 bfd->netdev = netdev_ref(netdev);
836 if (bfd->decay_min_rx && bfd->state == STATE_UP) {
837 bfd_decay_update(bfd);
839 if (bfd->forwarding_if_rx && bfd->state == STATE_UP) {
840 bfd_forwarding_if_rx_update(bfd);
842 bfd->rx_packets = bfd_rx_packets(bfd);
844 ovs_mutex_unlock(&mutex);
848 /* Updates the forwarding flag. If override is not configured and
849 * the forwarding flag value changes, increments the flap count.
851 * Note this function may be called multiple times in a function
852 * (e.g. bfd_account_rx) before and after the bfd state or status
853 * change. This is to capture any forwarding flag flap. */
855 bfd_forwarding__(struct bfd *bfd) OVS_REQUIRES(mutex)
858 bool last_forwarding = bfd->last_forwarding;
860 if (bfd->forwarding_override != -1) {
861 return bfd->forwarding_override == 1;
864 time = bfd->forwarding_if_rx_detect_time;
865 bfd->last_forwarding = (bfd->state == STATE_UP
866 || (bfd->forwarding_if_rx && time > time_msec()))
867 && bfd->rmt_diag != DIAG_PATH_DOWN
868 && bfd->rmt_diag != DIAG_CPATH_DOWN
869 && bfd->rmt_diag != DIAG_RCPATH_DOWN;
870 if (bfd->last_forwarding != last_forwarding) {
872 seq_change(connectivity_seq_get());
874 return bfd->last_forwarding;
879 bfd_lookup_ip(const char *host_name, struct in_addr *addr)
881 if (!inet_aton(host_name, addr)) {
882 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
889 bfd_in_poll(const struct bfd *bfd) OVS_REQUIRES(mutex)
891 return (bfd->flags & FLAG_POLL) != 0;
895 bfd_poll(struct bfd *bfd) OVS_REQUIRES(mutex)
897 if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
898 && !(bfd->flags & FLAG_FINAL)) {
899 bfd->poll_min_tx = bfd->cfg_min_tx;
900 bfd->poll_min_rx = bfd->in_decay ? bfd->decay_min_rx : bfd->cfg_min_rx;
901 bfd->flags |= FLAG_POLL;
903 VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
908 bfd_min_tx(const struct bfd *bfd) OVS_REQUIRES(mutex)
910 /* RFC 5880 Section 6.8.3
911 * When bfd.SessionState is not Up, the system MUST set
912 * bfd.DesiredMinTxInterval to a value of not less than one second
913 * (1,000,000 microseconds). This is intended to ensure that the
914 * bandwidth consumed by BFD sessions that are not Up is negligible,
915 * particularly in the case where a neighbor may not be running BFD. */
916 return (bfd->state == STATE_UP ? bfd->min_tx : MAX(bfd->min_tx, 1000));
920 bfd_tx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
922 long long int interval = bfd_min_tx(bfd);
923 return MAX(interval, bfd->rmt_min_rx);
927 bfd_rx_interval(const struct bfd *bfd) OVS_REQUIRES(mutex)
929 return MAX(bfd->min_rx, bfd->rmt_min_tx);
933 bfd_set_next_tx(struct bfd *bfd) OVS_REQUIRES(mutex)
935 long long int interval = bfd_tx_interval(bfd);
936 interval -= interval * random_range(26) / 100;
937 bfd->next_tx = bfd->last_tx + interval;
941 bfd_flag_str(enum flags flags)
943 struct ds ds = DS_EMPTY_INITIALIZER;
944 static char flag_str[128];
950 if (flags & FLAG_MULTIPOINT) {
951 ds_put_cstr(&ds, "multipoint ");
954 if (flags & FLAG_DEMAND) {
955 ds_put_cstr(&ds, "demand ");
958 if (flags & FLAG_AUTH) {
959 ds_put_cstr(&ds, "auth ");
962 if (flags & FLAG_CTL) {
963 ds_put_cstr(&ds, "ctl ");
966 if (flags & FLAG_FINAL) {
967 ds_put_cstr(&ds, "final ");
970 if (flags & FLAG_POLL) {
971 ds_put_cstr(&ds, "poll ");
974 /* Do not copy the trailing whitespace. */
976 ovs_strlcpy(flag_str, ds_cstr(&ds), sizeof flag_str);
982 bfd_state_str(enum state state)
985 case STATE_ADMIN_DOWN: return "admin_down";
986 case STATE_DOWN: return "down";
987 case STATE_INIT: return "init";
988 case STATE_UP: return "up";
989 default: return "invalid";
994 bfd_diag_str(enum diag diag) {
996 case DIAG_NONE: return "No Diagnostic";
997 case DIAG_EXPIRED: return "Control Detection Time Expired";
998 case DIAG_ECHO_FAILED: return "Echo Function Failed";
999 case DIAG_RMT_DOWN: return "Neighbor Signaled Session Down";
1000 case DIAG_FWD_RESET: return "Forwarding Plane Reset";
1001 case DIAG_PATH_DOWN: return "Path Down";
1002 case DIAG_CPATH_DOWN: return "Concatenated Path Down";
1003 case DIAG_ADMIN_DOWN: return "Administratively Down";
1004 case DIAG_RCPATH_DOWN: return "Reverse Concatenated Path Down";
1005 default: return "Invalid Diagnostic";
1010 log_msg(enum vlog_level level, const struct msg *p, const char *message,
1011 const struct bfd *bfd) OVS_REQUIRES(mutex)
1013 struct ds ds = DS_EMPTY_INITIALIZER;
1015 if (vlog_should_drop(THIS_MODULE, level, &rl)) {
1021 "\n\tvers:%"PRIu8" diag:\"%s\" state:%s mult:%"PRIu8
1024 "\n\tmy_disc:0x%"PRIx32" your_disc:0x%"PRIx32
1025 "\n\tmin_tx:%"PRIu32"us (%"PRIu32"ms)"
1026 "\n\tmin_rx:%"PRIu32"us (%"PRIu32"ms)"
1027 "\n\tmin_rx_echo:%"PRIu32"us (%"PRIu32"ms)",
1028 bfd->name, message, p->vers_diag >> VERS_SHIFT,
1029 bfd_diag_str(p->vers_diag & DIAG_MASK),
1030 bfd_state_str(p->flags & STATE_MASK),
1031 p->mult, p->length, bfd_flag_str(p->flags & FLAGS_MASK),
1032 ntohl(p->my_disc), ntohl(p->your_disc),
1033 ntohl(p->min_tx), ntohl(p->min_tx) / 1000,
1034 ntohl(p->min_rx), ntohl(p->min_rx) / 1000,
1035 ntohl(p->min_rx_echo), ntohl(p->min_rx_echo) / 1000);
1036 bfd_put_details(&ds, bfd);
1037 VLOG(level, "%s", ds_cstr(&ds));
1042 bfd_set_state(struct bfd *bfd, enum state state, enum diag diag)
1045 if (bfd->cpath_down) {
1046 diag = DIAG_CPATH_DOWN;
1049 if (bfd->state != state || bfd->diag != diag) {
1050 if (!VLOG_DROP_INFO(&rl)) {
1051 struct ds ds = DS_EMPTY_INITIALIZER;
1053 ds_put_format(&ds, "%s: BFD state change: %s->%s"
1054 " \"%s\"->\"%s\".\n",
1055 bfd->name, bfd_state_str(bfd->state),
1056 bfd_state_str(state), bfd_diag_str(bfd->diag),
1057 bfd_diag_str(diag));
1058 bfd_put_details(&ds, bfd);
1059 VLOG_INFO("%s", ds_cstr(&ds));
1066 if (bfd->state <= STATE_DOWN) {
1067 bfd->rmt_state = STATE_DOWN;
1068 bfd->rmt_diag = DIAG_NONE;
1069 bfd->rmt_min_rx = 1;
1072 bfd->rmt_min_tx = 0;
1073 /* Resets the min_rx if in_decay. */
1074 if (bfd->in_decay) {
1075 bfd->min_rx = bfd->cfg_min_rx;
1076 bfd->in_decay = false;
1079 /* Resets the decay when state changes to STATE_UP
1080 * and decay_min_rx is configured. */
1081 if (bfd->state == STATE_UP && bfd->decay_min_rx) {
1082 bfd_decay_update(bfd);
1085 seq_change(connectivity_seq_get());
1090 bfd_rx_packets(const struct bfd *bfd) OVS_REQUIRES(mutex)
1092 struct netdev_stats stats;
1094 if (!netdev_get_stats(bfd->netdev, &stats)) {
1095 return stats.rx_packets;
1101 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
1102 * the 'expect' value. */
1104 bfd_try_decay(struct bfd *bfd) OVS_REQUIRES(mutex)
1106 int64_t diff, expect;
1108 /* The 'diff' is the difference between current interface rx_packets
1109 * stats and last-time check. The 'expect' is the recorded number of
1110 * bfd control packets received within an approximately decay_min_rx
1111 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
1113 * Since the update of rx_packets stats at interface happens
1114 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1115 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1116 * more wiggle room. */
1117 diff = bfd_rx_packets(bfd) - bfd->decay_rx_packets;
1118 expect = 2 * MAX(bfd->decay_rx_ctl, 1);
1119 bfd->in_decay = diff <= expect ? true : false;
1120 bfd_decay_update(bfd);
1123 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1125 bfd_decay_update(struct bfd * bfd) OVS_REQUIRES(mutex)
1127 bfd->decay_rx_packets = bfd_rx_packets(bfd);
1128 bfd->decay_rx_ctl = 0;
1129 bfd->decay_detect_time = MAX(bfd->decay_min_rx, 2000) + time_msec();
1133 bfd_forwarding_if_rx_update(struct bfd *bfd) OVS_REQUIRES(mutex)
1135 int64_t incr = bfd_rx_interval(bfd) * bfd->mult;
1136 bfd->forwarding_if_rx_detect_time = MAX(incr, 2000) + time_msec();
1140 generate_discriminator(void)
1144 /* RFC 5880 Section 6.8.1
1145 * It SHOULD be set to a random (but still unique) value to improve
1146 * security. The value is otherwise outside the scope of this
1152 /* 'disc' is by definition random, so there's no reason to waste time
1154 disc = random_uint32();
1155 HMAP_FOR_EACH_IN_BUCKET (bfd, node, disc, all_bfds) {
1156 if (bfd->disc == disc) {
1167 bfd_find_by_name(const char *name) OVS_REQUIRES(mutex)
1171 HMAP_FOR_EACH (bfd, node, all_bfds) {
1172 if (!strcmp(bfd->name, name)) {
1180 bfd_put_details(struct ds *ds, const struct bfd *bfd) OVS_REQUIRES(mutex)
1182 ds_put_format(ds, "\tForwarding: %s\n",
1183 bfd_forwarding__(CONST_CAST(struct bfd *, bfd))
1184 ? "true" : "false");
1185 ds_put_format(ds, "\tDetect Multiplier: %d\n", bfd->mult);
1186 ds_put_format(ds, "\tConcatenated Path Down: %s\n",
1187 bfd->cpath_down ? "true" : "false");
1188 ds_put_format(ds, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd));
1189 ds_put_format(ds, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd));
1190 ds_put_format(ds, "\tDetect Time: now %+lldms\n",
1191 time_msec() - bfd->detect_time);
1192 ds_put_format(ds, "\tNext TX Time: now %+lldms\n",
1193 time_msec() - bfd->next_tx);
1194 ds_put_format(ds, "\tLast TX Time: now %+lldms\n",
1195 time_msec() - bfd->last_tx);
1197 ds_put_cstr(ds, "\n");
1199 ds_put_format(ds, "\tLocal Flags: %s\n", bfd_flag_str(bfd->flags));
1200 ds_put_format(ds, "\tLocal Session State: %s\n",
1201 bfd_state_str(bfd->state));
1202 ds_put_format(ds, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd->diag));
1203 ds_put_format(ds, "\tLocal Discriminator: 0x%"PRIx32"\n", bfd->disc);
1204 ds_put_format(ds, "\tLocal Minimum TX Interval: %lldms\n",
1206 ds_put_format(ds, "\tLocal Minimum RX Interval: %lldms\n", bfd->min_rx);
1208 ds_put_cstr(ds, "\n");
1210 ds_put_format(ds, "\tRemote Flags: %s\n", bfd_flag_str(bfd->rmt_flags));
1211 ds_put_format(ds, "\tRemote Session State: %s\n",
1212 bfd_state_str(bfd->rmt_state));
1213 ds_put_format(ds, "\tRemote Diagnostic: %s\n",
1214 bfd_diag_str(bfd->rmt_diag));
1215 ds_put_format(ds, "\tRemote Discriminator: 0x%"PRIx32"\n", bfd->rmt_disc);
1216 ds_put_format(ds, "\tRemote Minimum TX Interval: %lldms\n",
1218 ds_put_format(ds, "\tRemote Minimum RX Interval: %lldms\n",
1223 bfd_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[],
1224 void *aux OVS_UNUSED) OVS_EXCLUDED(mutex)
1226 struct ds ds = DS_EMPTY_INITIALIZER;
1229 ovs_mutex_lock(&mutex);
1231 bfd = bfd_find_by_name(argv[1]);
1233 unixctl_command_reply_error(conn, "no such bfd object");
1236 bfd_put_details(&ds, bfd);
1238 HMAP_FOR_EACH (bfd, node, all_bfds) {
1239 ds_put_format(&ds, "---- %s ----\n", bfd->name);
1240 bfd_put_details(&ds, bfd);
1243 unixctl_command_reply(conn, ds_cstr(&ds));
1247 ovs_mutex_unlock(&mutex);
1252 bfd_unixctl_set_forwarding_override(struct unixctl_conn *conn, int argc,
1253 const char *argv[], void *aux OVS_UNUSED)
1256 const char *forward_str = argv[argc - 1];
1257 int forwarding_override;
1260 ovs_mutex_lock(&mutex);
1261 if (!strcasecmp("true", forward_str)) {
1262 forwarding_override = 1;
1263 } else if (!strcasecmp("false", forward_str)) {
1264 forwarding_override = 0;
1265 } else if (!strcasecmp("normal", forward_str)) {
1266 forwarding_override = -1;
1268 unixctl_command_reply_error(conn, "unknown fault string");
1273 bfd = bfd_find_by_name(argv[1]);
1275 unixctl_command_reply_error(conn, "no such BFD object");
1278 bfd->forwarding_override = forwarding_override;
1280 HMAP_FOR_EACH (bfd, node, all_bfds) {
1281 bfd->forwarding_override = forwarding_override;
1285 unixctl_command_reply(conn, "OK");
1288 ovs_mutex_unlock(&mutex);