1 /* Copyright (c) 2013 Nicira, Inc.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License. */
18 #include <sys/types.h>
19 #include <arpa/inet.h>
20 #include <netinet/in_systm.h>
21 #include <netinet/ip.h>
23 #include "byte-order.h"
26 #include "dynamic-string.h"
35 #include "ovs-thread.h"
36 #include "openvswitch/types.h"
38 #include "poll-loop.h"
42 #include "unaligned.h"
47 VLOG_DEFINE_THIS_MODULE(bfd
);
51 * The goal of this module is to replace CFM with something both more flexible
52 * and standards compliant. In service of this goal, the following needs to be
56 * * Implement Demand mode.
57 * * Go through the RFC line by line and verify we comply.
58 * * Test against a hardware implementation. Preferably a popular one.
59 * * Delete BFD packets with nw_ttl != 255 in the datapath to prevent DOS
64 * - Set TOS/PCP on the outer tunnel header when encapped.
66 * - Sending BFD messages should be in its own thread/process.
68 * - Scale testing. How does it operate when there are large number of bfd
69 * sessions? Do we ever have random flaps? What's the CPU utilization?
71 * - Rely on data traffic for liveness by using BFD demand mode.
72 * If we're receiving traffic on a port, we can safely assume it's up (modulo
73 * unidrectional failures). BFD has a demand mode in which it can stay quiet
74 * unless it feels the need to check the status of the port. Using this, we
75 * can implement a strategy in which BFD only sends control messages on dark
78 * - Depending on how one interprets the spec, it appears that a BFD session
79 * can never change bfd.LocalDiag to "No Diagnostic". We should verify that
80 * this is what hardware implementations actually do. Seems like "No
81 * Diagnostic" should be set once a BFD session state goes UP. */
86 FLAG_MULTIPOINT
= 1 << 0,
95 STATE_ADMIN_DOWN
= 0 << 6,
102 DIAG_NONE
= 0, /* No Diagnostic. */
103 DIAG_EXPIRED
= 1, /* Control Detection Time Expired. */
104 DIAG_ECHO_FAILED
= 2, /* Echo Function Failed. */
105 DIAG_RMT_DOWN
= 3, /* Neighbor Signaled Session Down. */
106 DIAG_FWD_RESET
= 4, /* Forwarding Plane Reset. */
107 DIAG_PATH_DOWN
= 5, /* Path Down. */
108 DIAG_CPATH_DOWN
= 6, /* Concatenated Path Down. */
109 DIAG_ADMIN_DOWN
= 7, /* Administratively Down. */
110 DIAG_RCPATH_DOWN
= 8 /* Reverse Concatenated Path Down. */
113 /* RFC 5880 Section 4.1
115 * 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
116 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
117 * |Vers | Diag |Sta|P|F|C|A|D|M| Detect Mult | Length |
118 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
119 * | My Discriminator |
120 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
121 * | Your Discriminator |
122 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
123 * | Desired Min TX Interval |
124 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
125 * | Required Min RX Interval |
126 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
127 * | Required Min Echo RX Interval |
128 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ */
130 uint8_t vers_diag
; /* Version and diagnostic. */
131 uint8_t flags
; /* 2bit State field followed by flags. */
132 uint8_t mult
; /* Fault detection multiplier. */
133 uint8_t length
; /* Length of this BFD message. */
134 ovs_be32 my_disc
; /* My discriminator. */
135 ovs_be32 your_disc
; /* Your discriminator. */
136 ovs_be32 min_tx
; /* Desired minimum tx interval. */
137 ovs_be32 min_rx
; /* Required minimum rx interval. */
138 ovs_be32 min_rx_echo
; /* Required minimum echo rx interval. */
140 BUILD_ASSERT_DECL(BFD_PACKET_LEN
== sizeof(struct msg
));
142 #define DIAG_MASK 0x1f
144 #define STATE_MASK 0xC0
145 #define FLAGS_MASK 0x3f
148 struct hmap_node node
; /* In 'all_bfds'. */
149 uint32_t disc
; /* bfd.LocalDiscr. Key in 'all_bfds' hmap. */
151 char *name
; /* Name used for logging. */
153 bool cpath_down
; /* Concatenated Path Down. */
154 uint8_t mult
; /* bfd.DetectMult. */
156 struct netdev
*netdev
;
157 uint64_t rx_packets
; /* Packets received by 'netdev'. */
159 enum state state
; /* bfd.SessionState. */
160 enum state rmt_state
; /* bfd.RemoteSessionState. */
162 enum diag diag
; /* bfd.LocalDiag. */
163 enum diag rmt_diag
; /* Remote diagnostic. */
165 enum flags flags
; /* Flags sent on messages. */
166 enum flags rmt_flags
; /* Flags last received. */
168 uint32_t rmt_disc
; /* bfd.RemoteDiscr. */
170 uint8_t eth_dst
[ETH_ADDR_LEN
];/* Ethernet destination address. */
171 bool eth_dst_set
; /* 'eth_dst' set through database. */
173 uint16_t udp_src
; /* UDP source port. */
175 /* All timers in milliseconds. */
176 long long int rmt_min_rx
; /* bfd.RemoteMinRxInterval. */
177 long long int rmt_min_tx
; /* Remote minimum TX interval. */
179 long long int cfg_min_tx
; /* Configured minimum TX rate. */
180 long long int cfg_min_rx
; /* Configured required minimum RX rate. */
181 long long int poll_min_tx
; /* Min TX negotating in a poll sequence. */
182 long long int poll_min_rx
; /* Min RX negotating in a poll sequence. */
183 long long int min_tx
; /* bfd.DesiredMinTxInterval. */
184 long long int min_rx
; /* bfd.RequiredMinRxInterval. */
186 long long int last_tx
; /* Last TX time. */
187 long long int next_tx
; /* Next TX time. */
188 long long int detect_time
; /* RFC 5880 6.8.4 Detection time. */
190 bool forwarding
; /* Interface capability of packet I/O. */
191 int forwarding_override
; /* Manual override of 'forwarding' status. */
193 atomic_bool check_tnl_key
; /* Verify tunnel key of inbound packets? */
196 /* When forward_if_rx is true, bfd_forwarding() will return
197 * true as long as there are incoming packets received.
198 * Note, forwarding_override still has higher priority. */
199 bool forwarding_if_rx
;
200 long long int forwarding_if_rx_detect_time
;
202 /* BFD decay related variables. */
203 bool in_decay
; /* True when bfd is in decay. */
204 int decay_min_rx
; /* min_rx is set to decay_min_rx when */
206 int decay_rx_ctl
; /* Count bfd packets received within decay */
207 /* detect interval. */
208 uint64_t decay_rx_packets
; /* Packets received by 'netdev'. */
209 long long int decay_detect_time
; /* Decay detection time. */
211 uint64_t flap_count
; /* Counts bfd forwarding flaps. */
214 static struct ovs_mutex mutex
= OVS_MUTEX_INITIALIZER
;
215 static struct hmap all_bfds__
= HMAP_INITIALIZER(&all_bfds__
);
216 static struct hmap
*const all_bfds
OVS_GUARDED_BY(mutex
) = &all_bfds__
;
218 static bool bfd_forwarding__(struct bfd
*) OVS_REQUIRES(mutex
);
219 static bool bfd_in_poll(const struct bfd
*) OVS_REQUIRES(mutex
);
220 static void bfd_poll(struct bfd
*bfd
) OVS_REQUIRES(mutex
);
221 static const char *bfd_diag_str(enum diag
) OVS_REQUIRES(mutex
);
222 static const char *bfd_state_str(enum state
) OVS_REQUIRES(mutex
);
223 static long long int bfd_min_tx(const struct bfd
*) OVS_REQUIRES(mutex
);
224 static long long int bfd_tx_interval(const struct bfd
*)
226 static long long int bfd_rx_interval(const struct bfd
*)
228 static void bfd_set_next_tx(struct bfd
*) OVS_REQUIRES(mutex
);
229 static void bfd_set_state(struct bfd
*, enum state
, enum diag
)
231 static uint32_t generate_discriminator(void) OVS_REQUIRES(mutex
);
232 static void bfd_put_details(struct ds
*, const struct bfd
*)
234 static uint64_t bfd_rx_packets(const struct bfd
*) OVS_REQUIRES(mutex
);
235 static void bfd_try_decay(struct bfd
*) OVS_REQUIRES(mutex
);
236 static void bfd_decay_update(struct bfd
*) OVS_REQUIRES(mutex
);
237 static void bfd_check_rx(struct bfd
*) OVS_REQUIRES(mutex
);
238 static void bfd_forwarding_if_rx_update(struct bfd
*) OVS_REQUIRES(mutex
);
239 static void bfd_unixctl_show(struct unixctl_conn
*, int argc
,
240 const char *argv
[], void *aux OVS_UNUSED
);
241 static void bfd_unixctl_set_forwarding_override(struct unixctl_conn
*,
242 int argc
, const char *argv
[],
243 void *aux OVS_UNUSED
);
244 static void log_msg(enum vlog_level
, const struct msg
*, const char *message
,
245 const struct bfd
*) OVS_REQUIRES(mutex
);
247 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(20, 20);
249 /* Returns true if the interface on which 'bfd' is running may be used to
250 * forward traffic according to the BFD session state. */
252 bfd_forwarding(struct bfd
*bfd
) OVS_EXCLUDED(mutex
)
256 ovs_mutex_lock(&mutex
);
257 ret
= bfd_forwarding__(bfd
);
258 ovs_mutex_unlock(&mutex
);
262 /* Returns a 'smap' of key value pairs representing the status of 'bfd'
263 * intended for the OVS database. */
265 bfd_get_status(const struct bfd
*bfd
, struct smap
*smap
)
268 ovs_mutex_lock(&mutex
);
269 smap_add(smap
, "forwarding", bfd
->forwarding
? "true" : "false");
270 smap_add(smap
, "state", bfd_state_str(bfd
->state
));
271 smap_add(smap
, "diagnostic", bfd_diag_str(bfd
->diag
));
272 smap_add_format(smap
, "flap_count", "%"PRIu64
, bfd
->flap_count
);
274 if (bfd
->state
!= STATE_DOWN
) {
275 smap_add(smap
, "remote_state", bfd_state_str(bfd
->rmt_state
));
276 smap_add(smap
, "remote_diagnostic", bfd_diag_str(bfd
->rmt_diag
));
278 ovs_mutex_unlock(&mutex
);
281 /* Initializes, destroys, or reconfigures the BFD session 'bfd' (named 'name'),
282 * according to the database configuration contained in 'cfg'. Takes ownership
283 * of 'bfd', which may be NULL. Returns a BFD object which may be used as a
284 * handle for the session, or NULL if BFD is not enabled according to 'cfg'.
285 * Also returns NULL if cfg is NULL. */
287 bfd_configure(struct bfd
*bfd
, const char *name
, const struct smap
*cfg
,
288 struct netdev
*netdev
) OVS_EXCLUDED(mutex
)
290 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
291 static atomic_uint16_t udp_src
= ATOMIC_VAR_INIT(0);
294 long long int min_tx
, min_rx
;
295 bool need_poll
= false;
296 bool cfg_min_rx_changed
= false;
297 bool cpath_down
, forwarding_if_rx
;
299 uint8_t ea
[ETH_ADDR_LEN
];
301 if (ovsthread_once_start(&once
)) {
302 unixctl_command_register("bfd/show", "[interface]", 0, 1,
303 bfd_unixctl_show
, NULL
);
304 unixctl_command_register("bfd/set-forwarding",
305 "[interface] normal|false|true", 1, 2,
306 bfd_unixctl_set_forwarding_override
, NULL
);
307 ovsthread_once_done(&once
);
310 if (!cfg
|| !smap_get_bool(cfg
, "enable", false)) {
315 ovs_mutex_lock(&mutex
);
317 bfd
= xzalloc(sizeof *bfd
);
318 bfd
->name
= xstrdup(name
);
319 bfd
->forwarding
= false;
320 bfd
->forwarding_override
= -1;
321 bfd
->disc
= generate_discriminator();
322 hmap_insert(all_bfds
, &bfd
->node
, bfd
->disc
);
324 bfd
->diag
= DIAG_NONE
;
327 atomic_init(&bfd
->ref_cnt
, 1);
328 bfd
->netdev
= netdev_ref(netdev
);
329 bfd
->rx_packets
= bfd_rx_packets(bfd
);
330 bfd
->in_decay
= false;
333 /* RFC 5881 section 4
334 * The source port MUST be in the range 49152 through 65535. The same
335 * UDP source port number MUST be used for all BFD Control packets
336 * associated with a particular session. The source port number SHOULD
337 * be unique among all BFD sessions on the system. */
338 atomic_add(&udp_src
, 1, &bfd
->udp_src
);
339 bfd
->udp_src
= (bfd
->udp_src
% 16384) + 49152;
341 bfd_set_state(bfd
, STATE_DOWN
, DIAG_NONE
);
343 memcpy(bfd
->eth_dst
, eth_addr_bfd
, ETH_ADDR_LEN
);
346 atomic_store(&bfd
->check_tnl_key
,
347 smap_get_bool(cfg
, "check_tnl_key", false));
348 min_tx
= smap_get_int(cfg
, "min_tx", 100);
349 min_tx
= MAX(min_tx
, 100);
350 if (bfd
->cfg_min_tx
!= min_tx
) {
351 bfd
->cfg_min_tx
= min_tx
;
352 if (bfd
->state
!= STATE_UP
353 || (!bfd_in_poll(bfd
) && bfd
->cfg_min_tx
< bfd
->min_tx
)) {
354 bfd
->min_tx
= bfd
->cfg_min_tx
;
359 min_rx
= smap_get_int(cfg
, "min_rx", 1000);
360 min_rx
= MAX(min_rx
, 100);
361 if (bfd
->cfg_min_rx
!= min_rx
) {
362 bfd
->cfg_min_rx
= min_rx
;
363 if (bfd
->state
!= STATE_UP
364 || (!bfd_in_poll(bfd
) && bfd
->cfg_min_rx
> bfd
->min_rx
)) {
365 bfd
->min_rx
= bfd
->cfg_min_rx
;
367 cfg_min_rx_changed
= true;
371 decay_min_rx
= smap_get_int(cfg
, "decay_min_rx", 0);
372 if (bfd
->decay_min_rx
!= decay_min_rx
|| cfg_min_rx_changed
) {
373 if (decay_min_rx
> 0 && decay_min_rx
< bfd
->cfg_min_rx
) {
374 VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
375 bfd
->name
, bfd
->cfg_min_rx
);
376 bfd
->decay_min_rx
= 0;
378 bfd
->decay_min_rx
= decay_min_rx
;
381 bfd
->in_decay
= false;
382 bfd_decay_update(bfd
);
386 cpath_down
= smap_get_bool(cfg
, "cpath_down", false);
387 if (bfd
->cpath_down
!= cpath_down
) {
388 bfd
->cpath_down
= cpath_down
;
389 if (bfd
->diag
== DIAG_NONE
|| bfd
->diag
== DIAG_CPATH_DOWN
) {
390 bfd_set_state(bfd
, bfd
->state
, DIAG_NONE
);
395 hwaddr
= smap_get(cfg
, "bfd_dst_mac");
396 if (hwaddr
&& eth_addr_from_string(hwaddr
, ea
) && !eth_addr_is_zero(ea
)) {
397 memcpy(bfd
->eth_dst
, ea
, ETH_ADDR_LEN
);
398 bfd
->eth_dst_set
= true;
399 } else if (bfd
->eth_dst_set
) {
400 memcpy(bfd
->eth_dst
, eth_addr_bfd
, ETH_ADDR_LEN
);
401 bfd
->eth_dst_set
= false;
404 forwarding_if_rx
= smap_get_bool(cfg
, "forwarding_if_rx", false);
405 if (bfd
->forwarding_if_rx
!= forwarding_if_rx
) {
406 bfd
->forwarding_if_rx
= forwarding_if_rx
;
407 if (bfd
->state
== STATE_UP
&& bfd
->forwarding_if_rx
) {
408 bfd_forwarding_if_rx_update(bfd
);
410 bfd
->forwarding_if_rx_detect_time
= 0;
417 ovs_mutex_unlock(&mutex
);
422 bfd_ref(const struct bfd
*bfd_
)
424 struct bfd
*bfd
= CONST_CAST(struct bfd
*, bfd_
);
427 atomic_add(&bfd
->ref_cnt
, 1, &orig
);
428 ovs_assert(orig
> 0);
434 bfd_unref(struct bfd
*bfd
) OVS_EXCLUDED(mutex
)
439 atomic_sub(&bfd
->ref_cnt
, 1, &orig
);
440 ovs_assert(orig
> 0);
442 ovs_mutex_lock(&mutex
);
443 hmap_remove(all_bfds
, &bfd
->node
);
444 netdev_close(bfd
->netdev
);
447 ovs_mutex_unlock(&mutex
);
453 bfd_wait(const struct bfd
*bfd
) OVS_EXCLUDED(mutex
)
455 poll_timer_wait_until(bfd_wake_time(bfd
));
458 /* Returns the next wake up time. */
460 bfd_wake_time(const struct bfd
*bfd
) OVS_EXCLUDED(mutex
)
462 long long int retval
;
468 ovs_mutex_lock(&mutex
);
469 if (bfd
->flags
& FLAG_FINAL
) {
472 retval
= bfd
->next_tx
;
473 if (bfd
->state
> STATE_DOWN
) {
474 retval
= MIN(bfd
->detect_time
, retval
);
477 ovs_mutex_unlock(&mutex
);
482 bfd_run(struct bfd
*bfd
) OVS_EXCLUDED(mutex
)
487 ovs_mutex_lock(&mutex
);
489 old_in_decay
= bfd
->in_decay
;
491 if (bfd
->state
> STATE_DOWN
&& now
>= bfd
->detect_time
) {
492 bfd_set_state(bfd
, STATE_DOWN
, DIAG_EXPIRED
);
495 /* Decay may only happen when state is STATE_UP, bfd->decay_min_rx is
496 * configured, and decay_detect_time is reached. */
497 if (bfd
->state
== STATE_UP
&& bfd
->decay_min_rx
> 0
498 && now
>= bfd
->decay_detect_time
) {
502 /* Always checks the reception of any packet. */
505 if (bfd
->min_tx
!= bfd
->cfg_min_tx
506 || (bfd
->min_rx
!= bfd
->cfg_min_rx
&& bfd
->min_rx
!= bfd
->decay_min_rx
)
507 || bfd
->in_decay
!= old_in_decay
) {
510 ovs_mutex_unlock(&mutex
);
514 bfd_should_send_packet(const struct bfd
*bfd
) OVS_EXCLUDED(mutex
)
517 ovs_mutex_lock(&mutex
);
518 ret
= bfd
->flags
& FLAG_FINAL
|| time_msec() >= bfd
->next_tx
;
519 ovs_mutex_unlock(&mutex
);
524 bfd_put_packet(struct bfd
*bfd
, struct ofpbuf
*p
,
525 uint8_t eth_src
[ETH_ADDR_LEN
]) OVS_EXCLUDED(mutex
)
527 long long int min_tx
, min_rx
;
528 struct udp_header
*udp
;
529 struct eth_header
*eth
;
530 struct ip_header
*ip
;
533 ovs_mutex_lock(&mutex
);
535 long long int delay
= time_msec() - bfd
->next_tx
;
536 long long int interval
= bfd_tx_interval(bfd
);
537 if (delay
> interval
* 3 / 2) {
538 VLOG_INFO("%s: long delay of %lldms (expected %lldms) sending BFD"
539 " control message", bfd
->name
, delay
, interval
);
543 /* RFC 5880 Section 6.5
544 * A BFD Control packet MUST NOT have both the Poll (P) and Final (F) bits
546 ovs_assert(!(bfd
->flags
& FLAG_POLL
) || !(bfd
->flags
& FLAG_FINAL
));
548 ofpbuf_reserve(p
, 2); /* Properly align after the ethernet header. */
549 eth
= ofpbuf_put_uninit(p
, sizeof *eth
);
550 memcpy(eth
->eth_src
, eth_src
, ETH_ADDR_LEN
);
551 memcpy(eth
->eth_dst
, bfd
->eth_dst
, ETH_ADDR_LEN
);
552 eth
->eth_type
= htons(ETH_TYPE_IP
);
554 ip
= ofpbuf_put_zeros(p
, sizeof *ip
);
555 ip
->ip_ihl_ver
= IP_IHL_VER(5, 4);
556 ip
->ip_tot_len
= htons(sizeof *ip
+ sizeof *udp
+ sizeof *msg
);
558 ip
->ip_tos
= IPTOS_LOWDELAY
| IPTOS_THROUGHPUT
;
559 ip
->ip_proto
= IPPROTO_UDP
;
560 /* Use link local addresses: */
561 put_16aligned_be32(&ip
->ip_src
, htonl(0xA9FE0100)); /* 169.254.1.0. */
562 put_16aligned_be32(&ip
->ip_dst
, htonl(0xA9FE0101)); /* 169.254.1.1. */
563 ip
->ip_csum
= csum(ip
, sizeof *ip
);
565 udp
= ofpbuf_put_zeros(p
, sizeof *udp
);
566 udp
->udp_src
= htons(bfd
->udp_src
);
567 udp
->udp_dst
= htons(BFD_DEST_PORT
);
568 udp
->udp_len
= htons(sizeof *udp
+ sizeof *msg
);
570 msg
= ofpbuf_put_uninit(p
, sizeof *msg
);
571 msg
->vers_diag
= (BFD_VERSION
<< 5) | bfd
->diag
;
572 msg
->flags
= (bfd
->state
& STATE_MASK
) | bfd
->flags
;
574 msg
->mult
= bfd
->mult
;
575 msg
->length
= BFD_PACKET_LEN
;
576 msg
->my_disc
= htonl(bfd
->disc
);
577 msg
->your_disc
= htonl(bfd
->rmt_disc
);
578 msg
->min_rx_echo
= htonl(0);
580 if (bfd_in_poll(bfd
)) {
581 min_tx
= bfd
->poll_min_tx
;
582 min_rx
= bfd
->poll_min_rx
;
584 min_tx
= bfd_min_tx(bfd
);
585 min_rx
= bfd
->min_rx
;
588 msg
->min_tx
= htonl(min_tx
* 1000);
589 msg
->min_rx
= htonl(min_rx
* 1000);
591 bfd
->flags
&= ~FLAG_FINAL
;
593 log_msg(VLL_DBG
, msg
, "Sending BFD Message", bfd
);
595 bfd
->last_tx
= time_msec();
596 bfd_set_next_tx(bfd
);
597 ovs_mutex_unlock(&mutex
);
601 bfd_should_process_flow(const struct bfd
*bfd_
, const struct flow
*flow
,
602 struct flow_wildcards
*wc
)
604 struct bfd
*bfd
= CONST_CAST(struct bfd
*, bfd_
);
607 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
608 if (bfd
->eth_dst_set
&& memcmp(bfd
->eth_dst
, flow
->dl_dst
, ETH_ADDR_LEN
)) {
612 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
613 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
615 atomic_read(&bfd
->check_tnl_key
, &check_tnl_key
);
617 memset(&wc
->masks
.tunnel
.tun_id
, 0xff, sizeof wc
->masks
.tunnel
.tun_id
);
619 return (flow
->dl_type
== htons(ETH_TYPE_IP
)
620 && flow
->nw_proto
== IPPROTO_UDP
621 && flow
->tp_dst
== htons(BFD_DEST_PORT
)
622 && (!check_tnl_key
|| flow
->tunnel
.tun_id
== htonll(0)));
626 bfd_process_packet(struct bfd
*bfd
, const struct flow
*flow
,
627 const struct ofpbuf
*p
) OVS_EXCLUDED(mutex
)
629 uint32_t rmt_min_rx
, pkt_your_disc
;
630 enum state rmt_state
;
635 /* This function is designed to follow section RFC 5880 6.8.6 closely. */
637 ovs_mutex_lock(&mutex
);
638 /* Increments the decay rx counter. */
641 if (flow
->nw_ttl
!= 255) {
642 /* XXX Should drop in the kernel to prevent DOS. */
646 msg
= ofpbuf_at(p
, (uint8_t *)p
->l7
- (uint8_t *)p
->data
, BFD_PACKET_LEN
);
648 VLOG_INFO_RL(&rl
, "%s: Received too-short BFD control message (only "
649 "%"PRIdPTR
" bytes long, at least %d required).",
650 bfd
->name
, (uint8_t *) ofpbuf_tail(p
) - (uint8_t *) p
->l7
,
655 /* RFC 5880 Section 6.8.6
656 * If the Length field is greater than the payload of the encapsulating
657 * protocol, the packet MUST be discarded.
659 * Note that we make this check implicity. Above we use ofpbuf_at() to
660 * ensure that there are at least BFD_PACKET_LEN bytes in the payload of
661 * the encapsulating protocol. Below we require msg->length to be exactly
662 * BFD_PACKET_LEN bytes. */
664 flags
= msg
->flags
& FLAGS_MASK
;
665 rmt_state
= msg
->flags
& STATE_MASK
;
666 version
= msg
->vers_diag
>> VERS_SHIFT
;
668 log_msg(VLL_DBG
, msg
, "Received BFD control message", bfd
);
670 if (version
!= BFD_VERSION
) {
671 log_msg(VLL_WARN
, msg
, "Incorrect version", bfd
);
675 /* Technically this should happen after the length check. We don't support
676 * authentication however, so it's simpler to do the check first. */
677 if (flags
& FLAG_AUTH
) {
678 log_msg(VLL_WARN
, msg
, "Authenticated control message with"
679 " authentication disabled", bfd
);
683 if (msg
->length
!= BFD_PACKET_LEN
) {
684 log_msg(VLL_WARN
, msg
, "Unexpected length", bfd
);
685 if (msg
->length
< BFD_PACKET_LEN
) {
691 log_msg(VLL_WARN
, msg
, "Zero multiplier", bfd
);
695 if (flags
& FLAG_MULTIPOINT
) {
696 log_msg(VLL_WARN
, msg
, "Unsupported multipoint flag", bfd
);
701 log_msg(VLL_WARN
, msg
, "NULL my_disc", bfd
);
705 pkt_your_disc
= ntohl(msg
->your_disc
);
707 /* Technically, we should use the your discriminator field to figure
708 * out which 'struct bfd' this packet is destined towards. That way a
709 * bfd session could migrate from one interface to another
710 * transparently. This doesn't fit in with the OVS structure very
711 * well, so in this respect, we are not compliant. */
712 if (pkt_your_disc
!= bfd
->disc
) {
713 log_msg(VLL_WARN
, msg
, "Incorrect your_disc", bfd
);
716 } else if (rmt_state
> STATE_DOWN
) {
717 log_msg(VLL_WARN
, msg
, "Null your_disc", bfd
);
721 bfd
->rmt_disc
= ntohl(msg
->my_disc
);
722 bfd
->rmt_state
= rmt_state
;
723 bfd
->rmt_flags
= flags
;
724 bfd
->rmt_diag
= msg
->vers_diag
& DIAG_MASK
;
726 if (flags
& FLAG_FINAL
&& bfd_in_poll(bfd
)) {
727 bfd
->min_tx
= bfd
->poll_min_tx
;
728 bfd
->min_rx
= bfd
->poll_min_rx
;
729 bfd
->flags
&= ~FLAG_POLL
;
730 log_msg(VLL_INFO
, msg
, "Poll sequence terminated", bfd
);
733 if (flags
& FLAG_POLL
) {
734 /* RFC 5880 Section 6.5
735 * When the other system receives a Poll, it immediately transmits a
736 * BFD Control packet with the Final (F) bit set, independent of any
737 * periodic BFD Control packets it may be sending
738 * (see section 6.8.7). */
739 bfd
->flags
&= ~FLAG_POLL
;
740 bfd
->flags
|= FLAG_FINAL
;
743 rmt_min_rx
= MAX(ntohl(msg
->min_rx
) / 1000, 1);
744 if (bfd
->rmt_min_rx
!= rmt_min_rx
) {
745 bfd
->rmt_min_rx
= rmt_min_rx
;
746 bfd_set_next_tx(bfd
);
747 log_msg(VLL_INFO
, msg
, "New remote min_rx", bfd
);
750 bfd
->rmt_min_tx
= MAX(ntohl(msg
->min_tx
) / 1000, 1);
751 bfd
->detect_time
= bfd_rx_interval(bfd
) * bfd
->mult
+ time_msec();
753 if (bfd
->state
== STATE_ADMIN_DOWN
) {
754 VLOG_DBG_RL(&rl
, "Administratively down, dropping control message.");
758 if (rmt_state
== STATE_ADMIN_DOWN
) {
759 if (bfd
->state
!= STATE_DOWN
) {
760 bfd_set_state(bfd
, STATE_DOWN
, DIAG_RMT_DOWN
);
763 switch (bfd
->state
) {
765 if (rmt_state
== STATE_DOWN
) {
766 bfd_set_state(bfd
, STATE_INIT
, bfd
->diag
);
767 } else if (rmt_state
== STATE_INIT
) {
768 bfd_set_state(bfd
, STATE_UP
, bfd
->diag
);
772 if (rmt_state
> STATE_DOWN
) {
773 bfd_set_state(bfd
, STATE_UP
, bfd
->diag
);
777 if (rmt_state
<= STATE_DOWN
) {
778 bfd_set_state(bfd
, STATE_DOWN
, DIAG_RMT_DOWN
);
779 log_msg(VLL_INFO
, msg
, "Remote signaled STATE_DOWN", bfd
);
782 case STATE_ADMIN_DOWN
:
787 /* XXX: RFC 5880 Section 6.8.6 Demand mode related calculations here. */
790 ovs_mutex_unlock(&mutex
);
793 /* Must be called when the netdev owned by 'bfd' should change. */
795 bfd_set_netdev(struct bfd
*bfd
, const struct netdev
*netdev
)
798 ovs_mutex_lock(&mutex
);
799 if (bfd
->netdev
!= netdev
) {
800 netdev_close(bfd
->netdev
);
801 bfd
->netdev
= netdev_ref(netdev
);
802 if (bfd
->decay_min_rx
&& bfd
->state
== STATE_UP
) {
803 bfd_decay_update(bfd
);
805 if (bfd
->forwarding_if_rx
&& bfd
->state
== STATE_UP
) {
806 bfd_forwarding_if_rx_update(bfd
);
808 bfd
->rx_packets
= bfd_rx_packets(bfd
);
810 ovs_mutex_unlock(&mutex
);
814 /* Updates the forwarding flag. If override is not configured and
815 * the forwarding flag value changes, increments the flap count. */
817 bfd_forwarding__(struct bfd
*bfd
) OVS_REQUIRES(mutex
)
820 bool forwarding_old
= bfd
->forwarding
;
822 if (bfd
->forwarding_override
!= -1) {
823 return bfd
->forwarding_override
== 1;
826 time
= bfd
->forwarding_if_rx_detect_time
;
827 bfd
->forwarding
= (bfd
->state
== STATE_UP
828 || (bfd
->forwarding_if_rx
&& time
> time_msec()))
829 && bfd
->rmt_diag
!= DIAG_PATH_DOWN
830 && bfd
->rmt_diag
!= DIAG_CPATH_DOWN
831 && bfd
->rmt_diag
!= DIAG_RCPATH_DOWN
;
832 if (bfd
->forwarding
!= forwarding_old
) {
835 return bfd
->forwarding
;
840 bfd_in_poll(const struct bfd
*bfd
) OVS_REQUIRES(mutex
)
842 return (bfd
->flags
& FLAG_POLL
) != 0;
846 bfd_poll(struct bfd
*bfd
) OVS_REQUIRES(mutex
)
848 if (bfd
->state
> STATE_DOWN
&& !bfd_in_poll(bfd
)
849 && !(bfd
->flags
& FLAG_FINAL
)) {
850 bfd
->poll_min_tx
= bfd
->cfg_min_tx
;
851 bfd
->poll_min_rx
= bfd
->in_decay
? bfd
->decay_min_rx
: bfd
->cfg_min_rx
;
852 bfd
->flags
|= FLAG_POLL
;
854 VLOG_INFO_RL(&rl
, "%s: Initiating poll sequence", bfd
->name
);
859 bfd_min_tx(const struct bfd
*bfd
) OVS_REQUIRES(mutex
)
861 /* RFC 5880 Section 6.8.3
862 * When bfd.SessionState is not Up, the system MUST set
863 * bfd.DesiredMinTxInterval to a value of not less than one second
864 * (1,000,000 microseconds). This is intended to ensure that the
865 * bandwidth consumed by BFD sessions that are not Up is negligible,
866 * particularly in the case where a neighbor may not be running BFD. */
867 return (bfd
->state
== STATE_UP
? bfd
->min_tx
: MAX(bfd
->min_tx
, 1000));
871 bfd_tx_interval(const struct bfd
*bfd
) OVS_REQUIRES(mutex
)
873 long long int interval
= bfd_min_tx(bfd
);
874 return MAX(interval
, bfd
->rmt_min_rx
);
878 bfd_rx_interval(const struct bfd
*bfd
) OVS_REQUIRES(mutex
)
880 return MAX(bfd
->min_rx
, bfd
->rmt_min_tx
);
884 bfd_set_next_tx(struct bfd
*bfd
) OVS_REQUIRES(mutex
)
886 long long int interval
= bfd_tx_interval(bfd
);
887 interval
-= interval
* random_range(26) / 100;
888 bfd
->next_tx
= bfd
->last_tx
+ interval
;
892 bfd_flag_str(enum flags flags
)
894 struct ds ds
= DS_EMPTY_INITIALIZER
;
895 static char flag_str
[128];
901 if (flags
& FLAG_MULTIPOINT
) {
902 ds_put_cstr(&ds
, "multipoint ");
905 if (flags
& FLAG_DEMAND
) {
906 ds_put_cstr(&ds
, "demand ");
909 if (flags
& FLAG_AUTH
) {
910 ds_put_cstr(&ds
, "auth ");
913 if (flags
& FLAG_CTL
) {
914 ds_put_cstr(&ds
, "ctl ");
917 if (flags
& FLAG_FINAL
) {
918 ds_put_cstr(&ds
, "final ");
921 if (flags
& FLAG_POLL
) {
922 ds_put_cstr(&ds
, "poll ");
925 /* Do not copy the trailing whitespace. */
927 ovs_strlcpy(flag_str
, ds_cstr(&ds
), sizeof flag_str
);
933 bfd_state_str(enum state state
)
936 case STATE_ADMIN_DOWN
: return "admin_down";
937 case STATE_DOWN
: return "down";
938 case STATE_INIT
: return "init";
939 case STATE_UP
: return "up";
940 default: return "invalid";
945 bfd_diag_str(enum diag diag
) {
947 case DIAG_NONE
: return "No Diagnostic";
948 case DIAG_EXPIRED
: return "Control Detection Time Expired";
949 case DIAG_ECHO_FAILED
: return "Echo Function Failed";
950 case DIAG_RMT_DOWN
: return "Neighbor Signaled Session Down";
951 case DIAG_FWD_RESET
: return "Forwarding Plane Reset";
952 case DIAG_PATH_DOWN
: return "Path Down";
953 case DIAG_CPATH_DOWN
: return "Concatenated Path Down";
954 case DIAG_ADMIN_DOWN
: return "Administratively Down";
955 case DIAG_RCPATH_DOWN
: return "Reverse Concatenated Path Down";
956 default: return "Invalid Diagnostic";
961 log_msg(enum vlog_level level
, const struct msg
*p
, const char *message
,
962 const struct bfd
*bfd
) OVS_REQUIRES(mutex
)
964 struct ds ds
= DS_EMPTY_INITIALIZER
;
966 if (vlog_should_drop(THIS_MODULE
, level
, &rl
)) {
972 "\n\tvers:%"PRIu8
" diag:\"%s\" state:%s mult:%"PRIu8
975 "\n\tmy_disc:0x%"PRIx32
" your_disc:0x%"PRIx32
976 "\n\tmin_tx:%"PRIu32
"us (%"PRIu32
"ms)"
977 "\n\tmin_rx:%"PRIu32
"us (%"PRIu32
"ms)"
978 "\n\tmin_rx_echo:%"PRIu32
"us (%"PRIu32
"ms)",
979 bfd
->name
, message
, p
->vers_diag
>> VERS_SHIFT
,
980 bfd_diag_str(p
->vers_diag
& DIAG_MASK
),
981 bfd_state_str(p
->flags
& STATE_MASK
),
982 p
->mult
, p
->length
, bfd_flag_str(p
->flags
& FLAGS_MASK
),
983 ntohl(p
->my_disc
), ntohl(p
->your_disc
),
984 ntohl(p
->min_tx
), ntohl(p
->min_tx
) / 1000,
985 ntohl(p
->min_rx
), ntohl(p
->min_rx
) / 1000,
986 ntohl(p
->min_rx_echo
), ntohl(p
->min_rx_echo
) / 1000);
987 bfd_put_details(&ds
, bfd
);
988 VLOG(level
, "%s", ds_cstr(&ds
));
993 bfd_set_state(struct bfd
*bfd
, enum state state
, enum diag diag
)
996 if (diag
== DIAG_NONE
&& bfd
->cpath_down
) {
997 diag
= DIAG_CPATH_DOWN
;
1000 if (bfd
->state
!= state
|| bfd
->diag
!= diag
) {
1001 if (!VLOG_DROP_INFO(&rl
)) {
1002 struct ds ds
= DS_EMPTY_INITIALIZER
;
1004 ds_put_format(&ds
, "%s: BFD state change: %s->%s"
1005 " \"%s\"->\"%s\".\n",
1006 bfd
->name
, bfd_state_str(bfd
->state
),
1007 bfd_state_str(state
), bfd_diag_str(bfd
->diag
),
1008 bfd_diag_str(diag
));
1009 bfd_put_details(&ds
, bfd
);
1010 VLOG_INFO("%s", ds_cstr(&ds
));
1017 if (bfd
->state
<= STATE_DOWN
) {
1018 bfd
->rmt_state
= STATE_DOWN
;
1019 bfd
->rmt_diag
= DIAG_NONE
;
1020 bfd
->rmt_min_rx
= 1;
1023 bfd
->rmt_min_tx
= 0;
1024 /* Resets the min_rx if in_decay. */
1025 if (bfd
->in_decay
) {
1026 bfd
->min_rx
= bfd
->cfg_min_rx
;
1027 bfd
->in_decay
= false;
1030 /* Resets the decay when state changes to STATE_UP
1031 * and decay_min_rx is configured. */
1032 if (bfd
->state
== STATE_UP
&& bfd
->decay_min_rx
) {
1033 bfd_decay_update(bfd
);
1037 /* Updates the forwarding flag. */
1038 bfd_forwarding__(bfd
);
1042 bfd_rx_packets(const struct bfd
*bfd
) OVS_REQUIRES(mutex
)
1044 struct netdev_stats stats
;
1046 if (!netdev_get_stats(bfd
->netdev
, &stats
)) {
1047 return stats
.rx_packets
;
1053 /* Decays the bfd->min_rx to bfd->decay_min_rx when 'diff' is less than
1054 * the 'expect' value. */
1056 bfd_try_decay(struct bfd
*bfd
) OVS_REQUIRES(mutex
)
1058 int64_t diff
, expect
;
1060 /* The 'diff' is the difference between current interface rx_packets
1061 * stats and last-time check. The 'expect' is the recorded number of
1062 * bfd control packets received within an approximately decay_min_rx
1063 * (2000 ms if decay_min_rx is less than 2000 ms) interval.
1065 * Since the update of rx_packets stats at interface happens
1066 * asynchronously to the bfd_rx_packets() function, the 'diff' value
1067 * can be jittered. Thusly, we double the decay_rx_ctl to provide
1068 * more wiggle room. */
1069 diff
= bfd_rx_packets(bfd
) - bfd
->decay_rx_packets
;
1070 expect
= 2 * MAX(bfd
->decay_rx_ctl
, 1);
1071 bfd
->in_decay
= diff
<= expect
? true : false;
1072 bfd_decay_update(bfd
);
1075 /* Updates the rx_packets, decay_rx_ctl and decay_detect_time. */
1077 bfd_decay_update(struct bfd
* bfd
) OVS_REQUIRES(mutex
)
1079 bfd
->decay_rx_packets
= bfd_rx_packets(bfd
);
1080 bfd
->decay_rx_ctl
= 0;
1081 bfd
->decay_detect_time
= MAX(bfd
->decay_min_rx
, 2000) + time_msec();
1084 /* Checks if there are packets received during the time since last call.
1085 * If forwarding_if_rx is enabled and packets are received, updates the
1086 * forwarding_if_rx_detect_time. */
1088 bfd_check_rx(struct bfd
*bfd
) OVS_REQUIRES(mutex
)
1090 uint64_t rx_packets
= bfd_rx_packets(bfd
);
1093 diff
= rx_packets
- bfd
->rx_packets
;
1094 bfd
->rx_packets
= rx_packets
;
1096 VLOG_INFO_RL(&rl
, "rx_packets count is smaller than last time.");
1098 if (bfd
->forwarding_if_rx
&& diff
> 0) {
1099 bfd_forwarding_if_rx_update(bfd
);
1103 /* Updates the forwarding_if_rx_detect_time and the forwarding flag. */
1105 bfd_forwarding_if_rx_update(struct bfd
*bfd
) OVS_REQUIRES(mutex
)
1107 int64_t incr
= bfd_rx_interval(bfd
) * bfd
->mult
;
1108 bfd
->forwarding_if_rx_detect_time
= MAX(incr
, 2000) + time_msec();
1109 bfd_forwarding__(bfd
);
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", bfd
->forwarding
? "true" : "false");
1156 ds_put_format(ds
, "\tDetect Multiplier: %d\n", bfd
->mult
);
1157 ds_put_format(ds
, "\tConcatenated Path Down: %s\n",
1158 bfd
->cpath_down
? "true" : "false");
1159 ds_put_format(ds
, "\tTX Interval: Approx %lldms\n", bfd_tx_interval(bfd
));
1160 ds_put_format(ds
, "\tRX Interval: Approx %lldms\n", bfd_rx_interval(bfd
));
1161 ds_put_format(ds
, "\tDetect Time: now %+lldms\n",
1162 time_msec() - bfd
->detect_time
);
1163 ds_put_format(ds
, "\tNext TX Time: now %+lldms\n",
1164 time_msec() - bfd
->next_tx
);
1165 ds_put_format(ds
, "\tLast TX Time: now %+lldms\n",
1166 time_msec() - bfd
->last_tx
);
1168 ds_put_cstr(ds
, "\n");
1170 ds_put_format(ds
, "\tLocal Flags: %s\n", bfd_flag_str(bfd
->flags
));
1171 ds_put_format(ds
, "\tLocal Session State: %s\n",
1172 bfd_state_str(bfd
->state
));
1173 ds_put_format(ds
, "\tLocal Diagnostic: %s\n", bfd_diag_str(bfd
->diag
));
1174 ds_put_format(ds
, "\tLocal Discriminator: 0x%"PRIx32
"\n", bfd
->disc
);
1175 ds_put_format(ds
, "\tLocal Minimum TX Interval: %lldms\n",
1177 ds_put_format(ds
, "\tLocal Minimum RX Interval: %lldms\n", bfd
->min_rx
);
1179 ds_put_cstr(ds
, "\n");
1181 ds_put_format(ds
, "\tRemote Flags: %s\n", bfd_flag_str(bfd
->rmt_flags
));
1182 ds_put_format(ds
, "\tRemote Session State: %s\n",
1183 bfd_state_str(bfd
->rmt_state
));
1184 ds_put_format(ds
, "\tRemote Diagnostic: %s\n",
1185 bfd_diag_str(bfd
->rmt_diag
));
1186 ds_put_format(ds
, "\tRemote Discriminator: 0x%"PRIx32
"\n", bfd
->rmt_disc
);
1187 ds_put_format(ds
, "\tRemote Minimum TX Interval: %lldms\n",
1189 ds_put_format(ds
, "\tRemote Minimum RX Interval: %lldms\n",
1194 bfd_unixctl_show(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
1195 void *aux OVS_UNUSED
) OVS_EXCLUDED(mutex
)
1197 struct ds ds
= DS_EMPTY_INITIALIZER
;
1200 ovs_mutex_lock(&mutex
);
1202 bfd
= bfd_find_by_name(argv
[1]);
1204 unixctl_command_reply_error(conn
, "no such bfd object");
1207 bfd_put_details(&ds
, bfd
);
1209 HMAP_FOR_EACH (bfd
, node
, all_bfds
) {
1210 ds_put_format(&ds
, "---- %s ----\n", bfd
->name
);
1211 bfd_put_details(&ds
, bfd
);
1214 unixctl_command_reply(conn
, ds_cstr(&ds
));
1218 ovs_mutex_unlock(&mutex
);
1223 bfd_unixctl_set_forwarding_override(struct unixctl_conn
*conn
, int argc
,
1224 const char *argv
[], void *aux OVS_UNUSED
)
1227 const char *forward_str
= argv
[argc
- 1];
1228 int forwarding_override
;
1231 ovs_mutex_lock(&mutex
);
1232 if (!strcasecmp("true", forward_str
)) {
1233 forwarding_override
= 1;
1234 } else if (!strcasecmp("false", forward_str
)) {
1235 forwarding_override
= 0;
1236 } else if (!strcasecmp("normal", forward_str
)) {
1237 forwarding_override
= -1;
1239 unixctl_command_reply_error(conn
, "unknown fault string");
1244 bfd
= bfd_find_by_name(argv
[1]);
1246 unixctl_command_reply_error(conn
, "no such BFD object");
1249 bfd
->forwarding_override
= forwarding_override
;
1251 HMAP_FOR_EACH (bfd
, node
, all_bfds
) {
1252 bfd
->forwarding_override
= forwarding_override
;
1256 unixctl_command_reply(conn
, "OK");
1259 ovs_mutex_unlock(&mutex
);