1 /* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
4 * We are making the OpenFlow specification and associated documentation
5 * (Software) available for public use and benefit with the expectation
6 * that others will use, modify and enhance the Software and contribute
7 * those enhancements back to the community. However, since we would
8 * like to make the Software available for broadest use, with as few
9 * restrictions as possible permission is hereby granted, free of
10 * charge, to any person obtaining a copy of this Software to deal in
11 * the Software under the copyrights without restriction, including
12 * without limitation the rights to use, copy, modify, merge, publish,
13 * distribute, sublicense, and/or sell copies of the Software, and to
14 * permit persons to whom the Software is furnished to do so, subject to
15 * the following conditions:
17 * The above copyright notice and this permission notice shall be
18 * included in all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
23 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
24 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
25 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
26 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 * The name and trademarks of copyright holder(s) may NOT be used in
30 * advertising or publicity pertaining to the Software or any
31 * derivatives without specific, written prior permission.
39 #include <netinet/in.h>
49 #include "command-line.h"
53 #include "dhcp-client.h"
54 #include "dynamic-string.h"
57 #include "learning-switch.h"
59 #include "mac-learning.h"
63 #include "poll-loop.h"
67 #include "vconn-ssl.h"
69 #include "vlog-socket.h"
72 #define THIS_MODULE VLM_secchan
74 /* Behavior when the connection to the controller fails. */
76 FAIL_OPEN, /* Act as learning switch. */
77 FAIL_CLOSED /* Drop all packets. */
80 /* Settings that may be configured by the user. */
82 /* Overall mode of operation. */
83 bool discovery; /* Discover the controller automatically? */
84 bool in_band; /* Connect to controller in-band? */
86 /* Related vconns and network devices. */
87 const char *nl_name; /* Local datapath (must be "nl:" vconn). */
88 char *of_name; /* ofX network device name. */
89 const char *controller_name; /* Controller (if not discovery mode). */
90 const char *listen_vconn_name; /* Listens for mgmt connections. */
92 /* Failure behavior. */
93 enum fail_mode fail_mode; /* Act as learning switch if no controller? */
94 int max_idle; /* Idle time for flows in fail-open mode. */
95 int probe_interval; /* # seconds idle before sending echo request. */
96 int max_backoff; /* Max # seconds between connection attempts. */
98 /* Packet-in rate-limiting. */
99 int rate_limit; /* Tokens added to bucket per second. */
100 int burst_limit; /* Maximum number token bucket size. */
102 /* Discovery behavior. */
103 regex_t accept_controller_regex; /* Controller vconns to accept. */
104 const char *accept_controller_re; /* String version of regex. */
105 bool update_resolv_conf; /* Update /etc/resolv.conf? */
110 struct buffer *rxbuf;
111 int n_txq; /* No. of packets queued for tx on 'rconn'. */
118 #define HALF_REMOTE 1
119 struct half halves[2];
125 bool (*packet_cb)(struct relay *, int half, void *aux);
126 void (*periodic_cb)(void *aux);
127 void (*wait_cb)(void *aux);
131 static struct vlog_rate_limit vrl = VLOG_RATE_LIMIT_INIT(60, 60);
133 static void parse_options(int argc, char *argv[], struct settings *);
134 static void usage(void) NO_RETURN;
136 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
138 static struct relay *relay_accept(const struct settings *, struct vconn *);
139 static void relay_run(struct relay *, const struct hook[], size_t n_hooks);
140 static void relay_wait(struct relay *);
141 static void relay_destroy(struct relay *);
143 static struct hook make_hook(bool (*packet_cb)(struct relay *, int, void *),
144 void (*periodic_cb)(void *),
145 void (*wait_cb)(void *),
148 struct switch_status;
150 static struct hook switch_status_hook_create(const struct settings *,
151 struct switch_status **);
152 static void switch_status_register_category(struct switch_status *,
153 const char *category,
154 void (*cb)(struct status_reply *,
157 static void status_reply_put(struct status_reply *, const char *, ...)
160 static void rconn_status_cb(struct status_reply *, void *rconn_);
162 static struct discovery *discovery_init(const struct settings *,
163 struct switch_status *);
164 static void discovery_question_connectivity(struct discovery *);
165 static bool discovery_run(struct discovery *, char **controller_name);
166 static void discovery_wait(struct discovery *);
168 static struct hook in_band_hook_create(const struct settings *,
169 struct switch_status *,
170 struct rconn *remote);
171 static struct hook fail_open_hook_create(const struct settings *,
172 struct switch_status *,
174 struct rconn *remote);
175 static struct hook rate_limit_hook_create(const struct settings *,
176 struct switch_status *,
178 struct rconn *remote);
181 static void modify_dhcp_request(struct dhcp_msg *, void *aux);
182 static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux);
185 main(int argc, char *argv[])
189 struct list relays = LIST_INITIALIZER(&relays);
191 struct hook hooks[8];
194 struct rconn *local_rconn, *remote_rconn;
195 struct vconn *listen_vconn;
196 struct relay *controller_relay;
197 struct discovery *discovery;
198 struct switch_status *switch_status;
201 set_program_name(argv[0]);
202 register_fault_handlers();
205 parse_options(argc, argv, &s);
206 signal(SIGPIPE, SIG_IGN);
208 /* Start listening for management connections. */
209 if (s.listen_vconn_name) {
210 retval = vconn_open(s.listen_vconn_name, &listen_vconn);
211 if (retval && retval != EAGAIN) {
212 fatal(retval, "opening %s", s.listen_vconn_name);
214 if (!vconn_is_passive(listen_vconn)) {
215 fatal(0, "%s is not a passive vconn", s.listen_vconn_name);
221 /* Initialize switch status hook. */
222 hooks[n_hooks++] = switch_status_hook_create(&s, &switch_status);
224 /* Start controller discovery. */
225 discovery = s.discovery ? discovery_init(&s, switch_status) : NULL;
227 /* Start listening for vlogconf requests. */
228 retval = vlog_server_listen(NULL, NULL);
230 fatal(retval, "Could not listen for vlog connections");
235 VLOG_WARN("OpenFlow reference implementation version %s", VERSION);
236 VLOG_WARN("OpenFlow protocol version 0x%02x", OFP_VERSION);
238 /* Connect to datapath. */
239 local_rconn = rconn_create(0, s.max_backoff);
240 rconn_connect(local_rconn, s.nl_name);
241 switch_status_register_category(switch_status, "local",
242 rconn_status_cb, local_rconn);
244 /* Connect to controller. */
245 remote_rconn = rconn_create(s.probe_interval, s.max_backoff);
246 if (s.controller_name) {
247 retval = rconn_connect(remote_rconn, s.controller_name);
248 if (retval == EAFNOSUPPORT) {
249 fatal(0, "No support for %s vconn", s.controller_name);
252 switch_status_register_category(switch_status, "remote",
253 rconn_status_cb, remote_rconn);
255 /* Start relaying. */
256 controller_relay = relay_create(local_rconn, remote_rconn, false);
257 list_push_back(&relays, &controller_relay->node);
261 hooks[n_hooks++] = in_band_hook_create(&s, switch_status,
264 if (s.fail_mode == FAIL_OPEN) {
265 hooks[n_hooks++] = fail_open_hook_create(&s, switch_status,
266 local_rconn, remote_rconn);
269 hooks[n_hooks++] = rate_limit_hook_create(&s, switch_status,
270 local_rconn, remote_rconn);
272 assert(n_hooks <= ARRAY_SIZE(hooks));
279 LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) {
280 relay_run(r, hooks, n_hooks);
284 struct relay *r = relay_accept(&s, listen_vconn);
288 list_push_back(&relays, &r->node);
291 for (i = 0; i < n_hooks; i++) {
292 if (hooks[i].periodic_cb) {
293 hooks[i].periodic_cb(hooks[i].aux);
297 char *controller_name;
298 if (rconn_is_connectivity_questionable(remote_rconn)) {
299 discovery_question_connectivity(discovery);
301 if (discovery_run(discovery, &controller_name)) {
302 if (controller_name) {
303 rconn_connect(remote_rconn, controller_name);
305 rconn_disconnect(remote_rconn);
310 /* Wait for something to happen. */
311 LIST_FOR_EACH (r, struct relay, node, &relays) {
315 vconn_accept_wait(listen_vconn);
317 for (i = 0; i < n_hooks; i++) {
318 if (hooks[i].wait_cb) {
319 hooks[i].wait_cb(hooks[i].aux);
323 discovery_wait(discovery);
332 make_hook(bool (*packet_cb)(struct relay *, int half, void *aux),
333 void (*periodic_cb)(void *aux),
334 void (*wait_cb)(void *aux),
338 h.packet_cb = packet_cb;
339 h.periodic_cb = periodic_cb;
345 /* OpenFlow message relaying. */
347 static struct relay *
348 relay_accept(const struct settings *s, struct vconn *listen_vconn)
350 struct vconn *new_remote, *new_local;
351 char *nl_name_without_subscription;
352 struct rconn *r1, *r2;
355 retval = vconn_accept(listen_vconn, &new_remote);
357 if (retval != EAGAIN) {
358 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
363 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We
364 * only accept the former syntax in main().
366 * nl:123:0 opens a netlink connection to local datapath 123 without
367 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
369 nl_name_without_subscription = xasprintf("%s:0", s->nl_name);
370 retval = vconn_open(nl_name_without_subscription, &new_local);
372 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
373 nl_name_without_subscription, strerror(retval));
374 vconn_close(new_remote);
375 free(nl_name_without_subscription);
379 /* Create and return relay. */
380 r1 = rconn_create(0, 0);
381 rconn_connect_unreliably(r1, nl_name_without_subscription, new_local);
382 free(nl_name_without_subscription);
384 r2 = rconn_create(0, 0);
385 rconn_connect_unreliably(r2, "passive", new_remote);
387 return relay_create(r1, r2, true);
390 static struct relay *
391 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
393 struct relay *r = xcalloc(1, sizeof *r);
394 r->halves[HALF_LOCAL].rconn = local;
395 r->halves[HALF_REMOTE].rconn = remote;
396 r->is_mgmt_conn = is_mgmt_conn;
401 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
406 for (i = 0; i < 2; i++) {
407 rconn_run(r->halves[i].rconn);
410 /* Limit the number of iterations to prevent other tasks from starving. */
411 for (iteration = 0; iteration < 50; iteration++) {
412 bool progress = false;
413 for (i = 0; i < 2; i++) {
414 struct half *this = &r->halves[i];
415 struct half *peer = &r->halves[!i];
418 this->rxbuf = rconn_recv(this->rconn);
420 const struct hook *h;
421 for (h = hooks; h < &hooks[n_hooks]; h++) {
422 if (h->packet_cb(r, i, h->aux)) {
423 buffer_delete(this->rxbuf);
432 if (this->rxbuf && !this->n_txq) {
433 int retval = rconn_send(peer->rconn, this->rxbuf,
435 if (retval != EAGAIN) {
439 buffer_delete(this->rxbuf);
450 if (r->is_mgmt_conn) {
451 for (i = 0; i < 2; i++) {
452 struct half *this = &r->halves[i];
453 if (!rconn_is_alive(this->rconn)) {
462 relay_wait(struct relay *r)
466 for (i = 0; i < 2; i++) {
467 struct half *this = &r->halves[i];
469 rconn_run_wait(this->rconn);
471 rconn_recv_wait(this->rconn);
477 relay_destroy(struct relay *r)
481 list_remove(&r->node);
482 for (i = 0; i < 2; i++) {
483 struct half *this = &r->halves[i];
484 rconn_destroy(this->rconn);
485 buffer_delete(this->rxbuf);
490 /* In-band control. */
492 struct in_band_data {
493 const struct settings *s;
494 struct mac_learning *ml;
495 struct netdev *of_device;
496 struct rconn *controller;
497 uint8_t mac[ETH_ADDR_LEN];
502 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct buffer *b)
504 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
507 static const uint8_t *
508 get_controller_mac(struct in_band_data *in_band)
510 static uint32_t ip, last_nonzero_ip;
511 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
512 static time_t next_refresh = 0;
514 uint32_t last_ip = ip;
516 time_t now = time_now();
518 ip = rconn_get_ip(in_band->controller);
519 if (last_ip != ip || !next_refresh || now >= next_refresh) {
522 /* Look up MAC address. */
523 memset(mac, 0, sizeof mac);
525 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
527 VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s",
528 IP_ARGS(&ip), strerror(retval));
531 have_mac = !eth_addr_is_zero(mac);
533 /* Log changes in IP, MAC addresses. */
534 if (ip && ip != last_nonzero_ip) {
535 VLOG_DBG("controller IP address changed from "IP_FMT
536 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
537 last_nonzero_ip = ip;
539 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
540 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
542 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
543 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
546 /* Schedule next refresh.
548 * If we have an IP address but not a MAC address, then refresh
549 * quickly, since we probably will get a MAC address soon (via ARP).
550 * Otherwise, we can afford to wait a little while. */
551 next_refresh = now + (!ip || have_mac ? 10 : 1);
553 return !eth_addr_is_zero(mac) ? mac : NULL;
557 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
558 struct in_band_data *in_band)
560 const uint8_t *mac = get_controller_mac(in_band);
561 return mac && eth_addr_equals(mac, dl_addr);
565 in_band_packet_cb(struct relay *r, int half, void *in_band_)
567 struct in_band_data *in_band = in_band_;
568 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
569 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
570 struct ofp_packet_in *opi;
571 struct ofp_header *oh;
572 size_t pkt_ofs, pkt_len;
575 uint16_t in_port, out_port;
576 const uint8_t *controller_mac;
578 if (half != HALF_LOCAL || r->is_mgmt_conn) {
583 if (oh->type != OFPT_PACKET_IN) {
586 if (msg->size < offsetof(struct ofp_packet_in, data)) {
587 VLOG_WARN_RL(&vrl, "packet too short (%zu bytes) for packet_in",
592 /* Extract flow data from 'opi' into 'flow'. */
594 in_port = ntohs(opi->in_port);
595 pkt_ofs = offsetof(struct ofp_packet_in, data);
596 pkt_len = ntohs(opi->header.length) - pkt_ofs;
597 pkt.data = opi->data;
599 flow_extract(&pkt, in_port, &flow);
601 /* Deal with local stuff. */
602 controller_mac = get_controller_mac(in_band);
603 if (in_port == OFPP_LOCAL) {
604 /* Sent by secure channel. */
605 out_port = mac_learning_lookup(in_band->ml, flow.dl_dst);
606 } else if (eth_addr_equals(flow.dl_dst, in_band->mac)) {
607 /* Sent to secure channel. */
608 out_port = OFPP_LOCAL;
609 if (mac_learning_learn(in_band->ml, flow.dl_src, in_port)) {
610 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
611 ETH_ADDR_ARGS(flow.dl_src), in_port);
613 } else if (flow.dl_type == htons(ETH_TYPE_ARP)
614 && eth_addr_is_broadcast(flow.dl_dst)
615 && is_controller_mac(flow.dl_src, in_band)) {
616 /* ARP sent by controller. */
617 out_port = OFPP_FLOOD;
618 } else if (is_controller_mac(flow.dl_dst, in_band)
619 && in_port == mac_learning_lookup(in_band->ml,
621 /* Drop controller traffic that arrives on the controller port. */
622 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
623 in_band->s->max_idle, 0));
629 if (out_port != OFPP_FLOOD) {
630 /* The output port is known, so add a new flow. */
631 queue_tx(rc, in_band,
632 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
633 out_port, in_band->s->max_idle));
635 /* If the switch didn't buffer the packet, we need to send a copy. */
636 if (ntohl(opi->buffer_id) == UINT32_MAX) {
637 queue_tx(rc, in_band,
638 make_unbuffered_packet_out(&pkt, in_port, out_port));
641 /* We don't know that MAC. Send along the packet without setting up a
644 if (ntohl(opi->buffer_id) == UINT32_MAX) {
645 b = make_unbuffered_packet_out(&pkt, in_port, out_port);
647 b = make_buffered_packet_out(ntohl(opi->buffer_id),
650 queue_tx(rc, in_band, b);
656 in_band_status_cb(struct status_reply *sr, void *in_band_)
658 struct in_band_data *in_band = in_band_;
659 struct in_addr local_ip;
660 uint32_t controller_ip;
661 const uint8_t *controller_mac;
663 if (netdev_get_in4(in_band->of_device, &local_ip)) {
664 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
666 status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
667 ETH_ADDR_ARGS(in_band->mac));
669 controller_ip = rconn_get_ip(in_band->controller);
671 status_reply_put(sr, "controller-ip="IP_FMT,
672 IP_ARGS(&controller_ip));
674 controller_mac = get_controller_mac(in_band);
675 if (controller_mac) {
676 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
677 ETH_ADDR_ARGS(controller_mac));
682 in_band_hook_create(const struct settings *s, struct switch_status *ss,
683 struct rconn *remote)
685 struct in_band_data *in_band;
688 in_band = xcalloc(1, sizeof *in_band);
690 in_band->ml = mac_learning_create();
691 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE,
692 &in_band->of_device);
694 fatal(retval, "Could not open %s device", s->of_name);
696 memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device),
698 in_band->controller = remote;
699 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
700 return make_hook(in_band_packet_cb, NULL, NULL, in_band);
703 /* Fail open support. */
705 struct fail_open_data {
706 const struct settings *s;
707 struct rconn *local_rconn;
708 struct rconn *remote_rconn;
709 struct lswitch *lswitch;
710 int last_disconn_secs;
713 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
715 fail_open_periodic_cb(void *fail_open_)
717 struct fail_open_data *fail_open = fail_open_;
721 disconn_secs = rconn_disconnected_duration(fail_open->remote_rconn);
722 open = disconn_secs >= fail_open->s->probe_interval * 3;
723 if (open != (fail_open->lswitch != NULL)) {
725 VLOG_WARN("No longer in fail-open mode");
726 lswitch_destroy(fail_open->lswitch);
727 fail_open->lswitch = NULL;
729 VLOG_WARN("Could not connect to controller for %d seconds, "
730 "failing open", disconn_secs);
731 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
732 fail_open->s->max_idle);
733 fail_open->last_disconn_secs = disconn_secs;
735 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
736 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
737 "from controller", disconn_secs);
738 fail_open->last_disconn_secs = disconn_secs;
743 fail_open_packet_cb(struct relay *r, int half, void *fail_open_)
745 struct fail_open_data *fail_open = fail_open_;
746 if (half != HALF_LOCAL || r->is_mgmt_conn || !fail_open->lswitch) {
749 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
750 r->halves[HALF_LOCAL].rxbuf);
751 rconn_run(fail_open->local_rconn);
757 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
759 struct fail_open_data *fail_open = fail_open_;
760 const struct settings *s = fail_open->s;
761 int trigger_duration = s->probe_interval * 3;
762 int cur_duration = rconn_disconnected_duration(fail_open->remote_rconn);
764 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
765 status_reply_put(sr, "current-duration=%d", cur_duration);
766 status_reply_put(sr, "triggered=%s",
767 cur_duration >= trigger_duration ? "true" : "false");
768 status_reply_put(sr, "max-idle=%d", s->max_idle);
772 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
773 struct rconn *local_rconn, struct rconn *remote_rconn)
775 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
777 fail_open->local_rconn = local_rconn;
778 fail_open->remote_rconn = remote_rconn;
779 fail_open->lswitch = NULL;
780 switch_status_register_category(ss, "fail-open",
781 fail_open_status_cb, fail_open);
782 return make_hook(fail_open_packet_cb, fail_open_periodic_cb, NULL,
786 struct rate_limiter {
787 const struct settings *s;
788 struct rconn *remote_rconn;
790 /* One queue per physical port. */
791 struct queue queues[OFPP_MAX];
792 int n_queued; /* Sum over queues[*].n. */
793 int next_tx_port; /* Next port to check in round-robin. */
797 * It costs 1000 tokens to send a single packet_in message. A single token
798 * per message would be more straightforward, but this choice lets us avoid
799 * round-off error in refill_bucket()'s calculation of how many tokens to
800 * add to the bucket, since no division step is needed. */
801 long long int last_fill; /* Time at which we last added tokens. */
802 int tokens; /* Current number of tokens. */
804 /* Transmission queue. */
805 int n_txq; /* No. of packets waiting in rconn for tx. */
807 /* Statistics reporting. */
808 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
809 unsigned long long n_limited; /* # queued for rate limiting. */
810 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
811 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
814 /* Drop a packet from the longest queue in 'rl'. */
816 drop_packet(struct rate_limiter *rl)
818 struct queue *longest; /* Queue currently selected as longest. */
819 int n_longest; /* # of queues of same length as 'longest'. */
822 longest = &rl->queues[0];
824 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
825 if (longest->n < q->n) {
828 } else if (longest->n == q->n) {
831 /* Randomly select one of the longest queues, with a uniform
832 * distribution (Knuth algorithm 3.4.2R). */
833 if (!random_range(n_longest)) {
839 /* FIXME: do we want to pop the tail instead? */
840 buffer_delete(queue_pop_head(longest));
844 /* Remove and return the next packet to transmit (in round-robin order). */
845 static struct buffer *
846 dequeue_packet(struct rate_limiter *rl)
850 for (i = 0; i < OFPP_MAX; i++) {
851 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
852 struct queue *q = &rl->queues[port];
854 rl->next_tx_port = (port + 1) % OFPP_MAX;
856 return queue_pop_head(q);
862 /* Add tokens to the bucket based on elapsed time. */
864 refill_bucket(struct rate_limiter *rl)
866 const struct settings *s = rl->s;
867 long long int now = time_msec();
868 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
869 if (tokens >= 1000) {
871 rl->tokens = MIN(tokens, s->burst_limit * 1000);
875 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
876 * true if successful, false otherwise. (In the latter case no tokens are
879 get_token(struct rate_limiter *rl)
881 if (rl->tokens >= 1000) {
890 rate_limit_packet_cb(struct relay *r, int half, void *rl_)
892 struct rate_limiter *rl = rl_;
893 const struct settings *s = rl->s;
894 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
895 struct ofp_header *oh;
897 if (half == HALF_REMOTE) {
902 if (oh->type != OFPT_PACKET_IN) {
905 if (msg->size < offsetof(struct ofp_packet_in, data)) {
906 VLOG_WARN_RL(&vrl, "packet too short (%zu bytes) for packet_in",
911 if (!rl->n_queued && get_token(rl)) {
912 /* In the common case where we are not constrained by the rate limit,
913 * let the packet take the normal path. */
917 /* Otherwise queue it up for the periodic callback to drain out. */
918 struct ofp_packet_in *opi = msg->data;
919 int port = ntohs(opi->in_port) % OFPP_MAX;
920 if (rl->n_queued >= s->burst_limit) {
923 queue_push_tail(&rl->queues[port], buffer_clone(msg));
931 rate_limit_status_cb(struct status_reply *sr, void *rl_)
933 struct rate_limiter *rl = rl_;
935 status_reply_put(sr, "normal=%llu", rl->n_normal);
936 status_reply_put(sr, "limited=%llu", rl->n_limited);
937 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
938 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
942 rate_limit_periodic_cb(void *rl_)
944 struct rate_limiter *rl = rl_;
947 /* Drain some packets out of the bucket if possible, but limit the number
948 * of iterations to allow other code to get work done too. */
950 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
951 /* Use a small, arbitrary limit for the amount of queuing to do here,
952 * because the TCP connection is responsible for buffering and there is
953 * no point in trying to transmit faster than the TCP connection can
955 struct buffer *b = dequeue_packet(rl);
956 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
963 rate_limit_wait_cb(void *rl_)
965 struct rate_limiter *rl = rl_;
967 if (rl->tokens >= 1000) {
968 /* We can transmit more packets as soon as we're called again. */
969 poll_immediate_wake();
971 /* We have to wait for the bucket to re-fill. We could calculate
972 * the exact amount of time here for increased smoothness. */
973 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
979 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
980 struct rconn *local, struct rconn *remote)
982 struct rate_limiter *rl;
985 rl = xcalloc(1, sizeof *rl);
987 rl->remote_rconn = remote;
988 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
989 queue_init(&rl->queues[i]);
991 rl->last_fill = time_msec();
992 rl->tokens = s->rate_limit * 100;
993 switch_status_register_category(ss, "rate-limit",
994 rate_limit_status_cb, rl);
995 return make_hook(rate_limit_packet_cb, rate_limit_periodic_cb,
996 rate_limit_wait_cb, rl);
999 /* OFPST_SWITCH statistics. */
1001 struct switch_status_category {
1003 void (*cb)(struct status_reply *, void *aux);
1007 struct switch_status {
1008 const struct settings *s;
1010 struct switch_status_category categories[8];
1014 struct status_reply {
1015 struct switch_status_category *category;
1021 switch_status_packet_cb(struct relay *r, int half, void *ss_)
1023 struct switch_status *ss = ss_;
1024 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
1025 struct buffer *msg = r->halves[HALF_REMOTE].rxbuf;
1026 struct switch_status_category *c;
1027 struct ofp_stats_request *osr;
1028 struct ofp_stats_reply *reply;
1029 struct status_reply sr;
1030 struct ofp_header *oh;
1034 if (half == HALF_LOCAL) {
1039 if (oh->type != OFPT_STATS_REQUEST) {
1042 if (msg->size < sizeof(struct ofp_stats_request)) {
1043 VLOG_WARN_RL(&vrl, "packet too short (%zu bytes) for stats_request",
1049 if (osr->type != htons(OFPST_SWITCH)) {
1053 sr.request.string = (void *) (osr + 1);
1054 sr.request.length = msg->size - sizeof *osr;
1055 ds_init(&sr.output);
1056 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
1057 if (!memcmp(c->name, sr.request.string,
1058 MIN(strlen(c->name), sr.request.length))) {
1063 reply = make_openflow_xid((offsetof(struct ofp_stats_reply, body)
1064 + sr.output.length),
1065 OFPT_STATS_REPLY, osr->header.xid, &b);
1066 reply->type = htons(OFPST_SWITCH);
1068 memcpy(reply->body, sr.output.string, sr.output.length);
1069 retval = rconn_send(rc, b, NULL);
1070 if (retval && retval != EAGAIN) {
1071 VLOG_WARN("send failed (%s)", strerror(retval));
1073 ds_destroy(&sr.output);
1078 rconn_status_cb(struct status_reply *sr, void *rconn_)
1080 struct rconn *rconn = rconn_;
1081 time_t now = time_now();
1083 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
1084 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
1085 status_reply_put(sr, "is-connected=%s",
1086 rconn_is_connected(rconn) ? "true" : "false");
1087 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
1088 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
1089 status_reply_put(sr, "attempted-connections=%u",
1090 rconn_get_attempted_connections(rconn));
1091 status_reply_put(sr, "successful-connections=%u",
1092 rconn_get_successful_connections(rconn));
1093 status_reply_put(sr, "last-connection=%ld",
1094 (long int) (now - rconn_get_last_connection(rconn)));
1095 status_reply_put(sr, "time-connected=%lu",
1096 rconn_get_total_time_connected(rconn));
1100 config_status_cb(struct status_reply *sr, void *s_)
1102 const struct settings *s = s_;
1104 if (s->listen_vconn_name) {
1105 status_reply_put(sr, "management=%s", s->listen_vconn_name);
1107 if (s->probe_interval) {
1108 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
1110 if (s->max_backoff) {
1111 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
1116 switch_status_cb(struct status_reply *sr, void *ss_)
1118 struct switch_status *ss = ss_;
1119 time_t now = time_now();
1121 status_reply_put(sr, "now=%ld", (long int) now);
1122 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
1123 status_reply_put(sr, "pid=%ld", (long int) getpid());
1127 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
1129 struct switch_status *ss = xcalloc(1, sizeof *ss);
1131 ss->booted = time_now();
1132 switch_status_register_category(ss, "config",
1133 config_status_cb, (void *) s);
1134 switch_status_register_category(ss, "switch", switch_status_cb, ss);
1136 return make_hook(switch_status_packet_cb, NULL, NULL, ss);
1140 switch_status_register_category(struct switch_status *ss,
1141 const char *category,
1142 void (*cb)(struct status_reply *,
1146 struct switch_status_category *c;
1147 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
1148 c = &ss->categories[ss->n_categories++];
1151 c->name = xstrdup(category);
1155 status_reply_put(struct status_reply *sr, const char *content, ...)
1157 size_t old_length = sr->output.length;
1161 /* Append the status reply to the output. */
1162 ds_put_format(&sr->output, "%s.", sr->category->name);
1163 va_start(args, content);
1164 ds_put_format_valist(&sr->output, content, args);
1166 if (ds_last(&sr->output) != '\n') {
1167 ds_put_char(&sr->output, '\n');
1170 /* Drop what we just added if it doesn't match the request. */
1171 added = sr->output.length - old_length;
1172 if (added < sr->request.length
1173 || memcmp(&sr->output.string[old_length],
1174 sr->request.string, sr->request.length)) {
1175 ds_truncate(&sr->output, old_length);
1180 /* Controller discovery. */
1184 const struct settings *s;
1185 struct dhclient *dhcp;
1190 discovery_status_cb(struct status_reply *sr, void *d_)
1192 struct discovery *d = d_;
1194 status_reply_put(sr, "discovery.accept-remote=%s",
1195 d->s->accept_controller_re);
1196 status_reply_put(sr, "discovery.n-changes=%d", d->n_changes);
1197 status_reply_put(sr, "discovery.state=%s", dhclient_get_state(d->dhcp));
1198 status_reply_put(sr, "discovery.state-elapsed=%u",
1199 dhclient_get_state_elapsed(d->dhcp));
1200 if (dhclient_is_bound(d->dhcp)) {
1201 uint32_t ip = dhclient_get_ip(d->dhcp);
1202 uint32_t netmask = dhclient_get_netmask(d->dhcp);
1203 uint32_t router = dhclient_get_router(d->dhcp);
1205 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
1206 uint32_t dns_server;
1210 status_reply_put(sr, "discovery.ip="IP_FMT, IP_ARGS(&ip));
1211 status_reply_put(sr, "discovery.netmask="IP_FMT, IP_ARGS(&netmask));
1213 status_reply_put(sr, "discovery.router="IP_FMT, IP_ARGS(&router));
1216 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i, &dns_server);
1218 status_reply_put(sr, "discovery.dns%d="IP_FMT,
1219 i, IP_ARGS(&dns_server));
1222 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
1224 status_reply_put(sr, "discovery.domain=%s", domain_name);
1228 status_reply_put(sr, "discovery.lease-remaining=%u",
1229 dhclient_get_lease_remaining(d->dhcp));
1233 static struct discovery *
1234 discovery_init(const struct settings *s, struct switch_status *ss)
1236 struct netdev *netdev;
1237 struct discovery *d;
1238 struct dhclient *dhcp;
1241 /* Bring ofX network device up. */
1242 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1244 fatal(retval, "Could not open %s device", s->of_name);
1246 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
1248 fatal(retval, "Could not bring %s device up", s->of_name);
1250 netdev_close(netdev);
1252 /* Initialize DHCP client. */
1253 retval = dhclient_create(s->of_name, modify_dhcp_request,
1254 validate_dhcp_offer, (void *) s, &dhcp);
1256 fatal(retval, "Failed to initialize DHCP client");
1258 dhclient_init(dhcp, 0);
1260 d = xmalloc(sizeof *d);
1265 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
1271 discovery_question_connectivity(struct discovery *d)
1273 dhclient_force_renew(d->dhcp, 15);
1277 discovery_run(struct discovery *d, char **controller_name)
1279 dhclient_run(d->dhcp);
1280 if (!dhclient_changed(d->dhcp)) {
1284 dhclient_configure_netdev(d->dhcp);
1285 if (d->s->update_resolv_conf) {
1286 dhclient_update_resolv_conf(d->dhcp);
1289 if (dhclient_is_bound(d->dhcp)) {
1290 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
1291 DHCP_CODE_OFP_CONTROLLER_VCONN);
1292 VLOG_WARN("%s: discovered controller", *controller_name);
1295 *controller_name = NULL;
1297 VLOG_WARN("discovered controller no longer available");
1305 discovery_wait(struct discovery *d)
1307 dhclient_wait(d->dhcp);
1311 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
1313 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
1317 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
1319 const struct settings *s = s_;
1323 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
1325 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
1328 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
1330 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
1331 s->accept_controller_re);
1337 /* User interface. */
1340 parse_options(int argc, char *argv[], struct settings *s)
1343 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
1345 OPT_INACTIVITY_PROBE,
1351 static struct option long_options[] = {
1352 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
1353 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
1354 {"fail", required_argument, 0, 'f'},
1355 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
1356 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
1357 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
1358 {"listen", required_argument, 0, 'l'},
1359 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
1360 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
1361 {"detach", no_argument, 0, 'D'},
1362 {"pidfile", optional_argument, 0, 'P'},
1363 {"verbose", optional_argument, 0, 'v'},
1364 {"help", no_argument, 0, 'h'},
1365 {"version", no_argument, 0, 'V'},
1366 VCONN_SSL_LONG_OPTIONS
1369 char *short_options = long_options_to_short_options(long_options);
1370 char *accept_re = NULL;
1373 /* Set defaults that we can figure out before parsing options. */
1374 s->listen_vconn_name = NULL;
1375 s->fail_mode = FAIL_OPEN;
1377 s->probe_interval = 15;
1378 s->max_backoff = 15;
1379 s->update_resolv_conf = true;
1385 c = getopt_long(argc, argv, short_options, long_options, NULL);
1391 case OPT_ACCEPT_VCONN:
1392 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
1395 case OPT_NO_RESOLV_CONF:
1396 s->update_resolv_conf = false;
1400 if (!strcmp(optarg, "open")) {
1401 s->fail_mode = FAIL_OPEN;
1402 } else if (!strcmp(optarg, "closed")) {
1403 s->fail_mode = FAIL_CLOSED;
1406 "-f or --fail argument must be \"open\" or \"closed\"");
1410 case OPT_INACTIVITY_PROBE:
1411 s->probe_interval = atoi(optarg);
1412 if (s->probe_interval < 5) {
1413 fatal(0, "--inactivity-probe argument must be at least 5");
1418 if (!strcmp(optarg, "permanent")) {
1419 s->max_idle = OFP_FLOW_PERMANENT;
1421 s->max_idle = atoi(optarg);
1422 if (s->max_idle < 1 || s->max_idle > 65535) {
1423 fatal(0, "--max-idle argument must be between 1 and "
1424 "65535 or the word 'permanent'");
1429 case OPT_MAX_BACKOFF:
1430 s->max_backoff = atoi(optarg);
1431 if (s->max_backoff < 1) {
1432 fatal(0, "--max-backoff argument must be at least 1");
1433 } else if (s->max_backoff > 3600) {
1434 s->max_backoff = 3600;
1438 case OPT_RATE_LIMIT:
1440 s->rate_limit = atoi(optarg);
1441 if (s->rate_limit < 1) {
1442 fatal(0, "--rate-limit argument must be at least 1");
1445 s->rate_limit = 1000;
1449 case OPT_BURST_LIMIT:
1450 s->burst_limit = atoi(optarg);
1451 if (s->burst_limit < 1) {
1452 fatal(0, "--burst-limit argument must be at least 1");
1461 set_pidfile(optarg);
1465 if (s->listen_vconn_name) {
1466 fatal(0, "-l or --listen may be only specified once");
1468 s->listen_vconn_name = optarg;
1475 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
1479 vlog_set_verbosity(optarg);
1482 VCONN_SSL_OPTION_HANDLERS
1491 free(short_options);
1495 if (argc < 1 || argc > 2) {
1496 fatal(0, "need one or two non-option arguments; use --help for usage");
1499 /* Local and remote vconns. */
1500 s->nl_name = argv[0];
1501 if (strncmp(s->nl_name, "nl:", 3)
1502 || strlen(s->nl_name) < 4
1503 || s->nl_name[strspn(s->nl_name + 3, "0123456789") + 3]) {
1504 fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"", s->nl_name);
1506 s->of_name = xasprintf("of%s", s->nl_name + 3);
1507 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
1509 /* Set accept_controller_regex. */
1511 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
1513 retval = regcomp(&s->accept_controller_regex, accept_re,
1514 REG_NOSUB | REG_EXTENDED);
1516 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
1517 char *buffer = xmalloc(length);
1518 regerror(retval, &s->accept_controller_regex, buffer, length);
1519 fatal(0, "%s: %s", accept_re, buffer);
1521 s->accept_controller_re = accept_re;
1523 /* Mode of operation. */
1524 s->discovery = s->controller_name == NULL;
1528 enum netdev_flags flags;
1529 struct netdev *netdev;
1531 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1533 fatal(retval, "Could not open %s device", s->of_name);
1536 retval = netdev_get_flags(netdev, &flags);
1538 fatal(retval, "Could not get flags for %s device", s->of_name);
1541 s->in_band = (flags & NETDEV_UP) != 0;
1542 if (s->in_band && netdev_get_in6(netdev, NULL)) {
1543 VLOG_WARN("Ignoring IPv6 address on %s device: IPv6 not supported",
1547 netdev_close(netdev);
1550 /* Rate limiting. */
1551 if (s->rate_limit) {
1552 if (s->rate_limit < 100) {
1553 VLOG_WARN("Rate limit set to unusually low value %d",
1556 if (!s->burst_limit) {
1557 s->burst_limit = s->rate_limit / 4;
1559 s->burst_limit = MAX(s->burst_limit, 1);
1560 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
1567 printf("%s: secure channel, a relay for OpenFlow messages.\n"
1568 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
1569 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
1570 "CONTROLLER is an active OpenFlow connection method; if it is\n"
1571 "omitted, then secchan performs controller discovery.\n",
1572 program_name, program_name);
1573 vconn_usage(true, true);
1574 printf("\nController discovery options:\n"
1575 " --accept-vconn=REGEX accept matching discovered controllers\n"
1576 " --no-resolv-conf do not update /etc/resolv.conf\n"
1577 "\nNetworking options:\n"
1578 " -f, --fail=open|closed when controller connection fails:\n"
1579 " closed: drop all packets\n"
1580 " open (default): act as learning switch\n"
1581 " --inactivity-probe=SECS time between inactivity probes\n"
1582 " --max-idle=SECS max idle for flows set up by secchan\n"
1583 " --max-backoff=SECS max time between controller connection\n"
1584 " attempts (default: 15 seconds)\n"
1585 " -l, --listen=METHOD allow management connections on METHOD\n"
1586 " (a passive OpenFlow connection method)\n"
1587 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
1588 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
1589 " --burst-limit=BURST limit on packet credit for idle time\n"
1590 "\nOther options:\n"
1591 " -D, --detach run in background as daemon\n"
1592 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
1593 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
1594 " -v, --verbose set maximum verbosity level\n"
1595 " -h, --help display this help message\n"
1596 " -V, --version display version information\n",