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 /* Maximum number of management connection listeners. */
83 /* Settings that may be configured by the user. */
85 /* Overall mode of operation. */
86 bool discovery; /* Discover the controller automatically? */
87 bool in_band; /* Connect to controller in-band? */
89 /* Related vconns and network devices. */
90 const char *nl_name; /* Local datapath (must be "nl:" vconn). */
91 char *of_name; /* ofX network device name. */
92 const char *controller_name; /* Controller (if not discovery mode). */
93 const char *listener_names[MAX_MGMT]; /* Listen for mgmt connections. */
94 size_t n_listeners; /* Number of mgmt connection listeners. */
95 const char *monitor_name; /* Listen for traffic monitor connections. */
97 /* Failure behavior. */
98 enum fail_mode fail_mode; /* Act as learning switch if no controller? */
99 int max_idle; /* Idle time for flows in fail-open mode. */
100 int probe_interval; /* # seconds idle before sending echo request. */
101 int max_backoff; /* Max # seconds between connection attempts. */
103 /* Packet-in rate-limiting. */
104 int rate_limit; /* Tokens added to bucket per second. */
105 int burst_limit; /* Maximum number token bucket size. */
107 /* Discovery behavior. */
108 regex_t accept_controller_regex; /* Controller vconns to accept. */
109 const char *accept_controller_re; /* String version of regex. */
110 bool update_resolv_conf; /* Update /etc/resolv.conf? */
115 struct buffer *rxbuf;
116 int n_txq; /* No. of packets queued for tx on 'rconn'. */
123 #define HALF_REMOTE 1
124 struct half halves[2];
130 bool (*packet_cb)(struct relay *, int half, void *aux);
131 void (*periodic_cb)(void *aux);
132 void (*wait_cb)(void *aux);
136 static struct vlog_rate_limit vrl = VLOG_RATE_LIMIT_INIT(60, 60);
138 static void parse_options(int argc, char *argv[], struct settings *);
139 static void usage(void) NO_RETURN;
141 static struct vconn *open_passive_vconn(const char *name);
142 static struct vconn *accept_vconn(struct vconn *vconn);
144 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
146 static struct relay *relay_accept(const struct settings *, struct vconn *);
147 static void relay_run(struct relay *, const struct hook[], size_t n_hooks);
148 static void relay_wait(struct relay *);
149 static void relay_destroy(struct relay *);
151 static struct hook make_hook(bool (*packet_cb)(struct relay *, int, void *),
152 void (*periodic_cb)(void *),
153 void (*wait_cb)(void *),
156 struct switch_status;
158 static struct hook switch_status_hook_create(const struct settings *,
159 struct switch_status **);
160 static void switch_status_register_category(struct switch_status *,
161 const char *category,
162 void (*cb)(struct status_reply *,
165 static void status_reply_put(struct status_reply *, const char *, ...)
168 static void rconn_status_cb(struct status_reply *, void *rconn_);
170 static struct discovery *discovery_init(const struct settings *,
171 struct switch_status *);
172 static void discovery_question_connectivity(struct discovery *);
173 static bool discovery_run(struct discovery *, char **controller_name);
174 static void discovery_wait(struct discovery *);
176 static struct hook in_band_hook_create(const struct settings *,
177 struct switch_status *,
178 struct rconn *remote);
179 static struct hook fail_open_hook_create(const struct settings *,
180 struct switch_status *,
182 struct rconn *remote);
183 static struct hook rate_limit_hook_create(const struct settings *,
184 struct switch_status *,
186 struct rconn *remote);
189 static void modify_dhcp_request(struct dhcp_msg *, void *aux);
190 static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux);
193 main(int argc, char *argv[])
197 struct list relays = LIST_INITIALIZER(&relays);
199 struct hook hooks[8];
202 struct vconn *monitor;
204 struct vconn *listeners[MAX_MGMT];
207 struct rconn *local_rconn, *remote_rconn;
208 struct relay *controller_relay;
209 struct discovery *discovery;
210 struct switch_status *switch_status;
214 set_program_name(argv[0]);
215 register_fault_handlers();
218 parse_options(argc, argv, &s);
219 signal(SIGPIPE, SIG_IGN);
221 /* Start listening for management and monitoring connections. */
223 for (i = 0; i < s.n_listeners; i++) {
224 listeners[n_listeners++] = open_passive_vconn(s.listener_names[i]);
226 monitor = s.monitor_name ? open_passive_vconn(s.monitor_name) : NULL;
228 /* Initialize switch status hook. */
229 hooks[n_hooks++] = switch_status_hook_create(&s, &switch_status);
231 /* Start controller discovery. */
232 discovery = s.discovery ? discovery_init(&s, switch_status) : NULL;
234 /* Start listening for vlogconf requests. */
235 retval = vlog_server_listen(NULL, NULL);
237 fatal(retval, "Could not listen for vlog connections");
240 die_if_already_running();
243 VLOG_WARN("OpenFlow reference implementation version %s", VERSION);
244 VLOG_WARN("OpenFlow protocol version 0x%02x", OFP_VERSION);
246 /* Connect to datapath. */
247 local_rconn = rconn_create(0, s.max_backoff);
248 rconn_connect(local_rconn, s.nl_name);
249 switch_status_register_category(switch_status, "local",
250 rconn_status_cb, local_rconn);
252 /* Connect to controller. */
253 remote_rconn = rconn_create(s.probe_interval, s.max_backoff);
254 if (s.controller_name) {
255 retval = rconn_connect(remote_rconn, s.controller_name);
256 if (retval == EAFNOSUPPORT) {
257 fatal(0, "No support for %s vconn", s.controller_name);
260 switch_status_register_category(switch_status, "remote",
261 rconn_status_cb, remote_rconn);
263 /* Start relaying. */
264 controller_relay = relay_create(local_rconn, remote_rconn, false);
265 list_push_back(&relays, &controller_relay->node);
269 hooks[n_hooks++] = in_band_hook_create(&s, switch_status,
272 if (s.fail_mode == FAIL_OPEN) {
273 hooks[n_hooks++] = fail_open_hook_create(&s, switch_status,
274 local_rconn, remote_rconn);
277 hooks[n_hooks++] = rate_limit_hook_create(&s, switch_status,
278 local_rconn, remote_rconn);
280 assert(n_hooks <= ARRAY_SIZE(hooks));
287 LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) {
288 relay_run(r, hooks, n_hooks);
290 for (i = 0; i < n_listeners; i++) {
292 struct relay *r = relay_accept(&s, listeners[i]);
296 list_push_back(&relays, &r->node);
300 struct vconn *new = accept_vconn(monitor);
302 rconn_add_monitor(local_rconn, new);
305 for (i = 0; i < n_hooks; i++) {
306 if (hooks[i].periodic_cb) {
307 hooks[i].periodic_cb(hooks[i].aux);
311 char *controller_name;
312 if (rconn_is_connectivity_questionable(remote_rconn)) {
313 discovery_question_connectivity(discovery);
315 if (discovery_run(discovery, &controller_name)) {
316 if (controller_name) {
317 rconn_connect(remote_rconn, controller_name);
319 rconn_disconnect(remote_rconn);
324 /* Wait for something to happen. */
325 LIST_FOR_EACH (r, struct relay, node, &relays) {
328 for (i = 0; i < n_listeners; i++) {
329 vconn_accept_wait(listeners[i]);
332 vconn_accept_wait(monitor);
334 for (i = 0; i < n_hooks; i++) {
335 if (hooks[i].wait_cb) {
336 hooks[i].wait_cb(hooks[i].aux);
340 discovery_wait(discovery);
348 static struct vconn *
349 open_passive_vconn(const char *name)
354 retval = vconn_open(name, &vconn);
355 if (retval && retval != EAGAIN) {
356 fatal(retval, "opening %s", name);
358 if (!vconn_is_passive(vconn)) {
359 fatal(0, "%s is not a passive vconn", name);
364 static struct vconn *
365 accept_vconn(struct vconn *vconn)
370 retval = vconn_accept(vconn, &new);
371 if (retval && retval != EAGAIN) {
372 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
378 make_hook(bool (*packet_cb)(struct relay *, int half, void *aux),
379 void (*periodic_cb)(void *aux),
380 void (*wait_cb)(void *aux),
384 h.packet_cb = packet_cb;
385 h.periodic_cb = periodic_cb;
391 /* OpenFlow message relaying. */
393 static struct relay *
394 relay_accept(const struct settings *s, struct vconn *listen_vconn)
396 struct vconn *new_remote, *new_local;
397 char *nl_name_without_subscription;
398 struct rconn *r1, *r2;
401 new_remote = accept_vconn(listen_vconn);
406 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We
407 * only accept the former syntax in main().
409 * nl:123:0 opens a netlink connection to local datapath 123 without
410 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
412 nl_name_without_subscription = xasprintf("%s:0", s->nl_name);
413 retval = vconn_open(nl_name_without_subscription, &new_local);
415 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
416 nl_name_without_subscription, strerror(retval));
417 vconn_close(new_remote);
418 free(nl_name_without_subscription);
422 /* Create and return relay. */
423 r1 = rconn_create(0, 0);
424 rconn_connect_unreliably(r1, nl_name_without_subscription, new_local);
425 free(nl_name_without_subscription);
427 r2 = rconn_create(0, 0);
428 rconn_connect_unreliably(r2, "passive", new_remote);
430 return relay_create(r1, r2, true);
433 static struct relay *
434 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
436 struct relay *r = xcalloc(1, sizeof *r);
437 r->halves[HALF_LOCAL].rconn = local;
438 r->halves[HALF_REMOTE].rconn = remote;
439 r->is_mgmt_conn = is_mgmt_conn;
444 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
449 for (i = 0; i < 2; i++) {
450 rconn_run(r->halves[i].rconn);
453 /* Limit the number of iterations to prevent other tasks from starving. */
454 for (iteration = 0; iteration < 50; iteration++) {
455 bool progress = false;
456 for (i = 0; i < 2; i++) {
457 struct half *this = &r->halves[i];
458 struct half *peer = &r->halves[!i];
461 this->rxbuf = rconn_recv(this->rconn);
463 const struct hook *h;
464 for (h = hooks; h < &hooks[n_hooks]; h++) {
465 if (h->packet_cb(r, i, h->aux)) {
466 buffer_delete(this->rxbuf);
475 if (this->rxbuf && !this->n_txq) {
476 int retval = rconn_send(peer->rconn, this->rxbuf,
478 if (retval != EAGAIN) {
482 buffer_delete(this->rxbuf);
493 if (r->is_mgmt_conn) {
494 for (i = 0; i < 2; i++) {
495 struct half *this = &r->halves[i];
496 if (!rconn_is_alive(this->rconn)) {
505 relay_wait(struct relay *r)
509 for (i = 0; i < 2; i++) {
510 struct half *this = &r->halves[i];
512 rconn_run_wait(this->rconn);
514 rconn_recv_wait(this->rconn);
520 relay_destroy(struct relay *r)
524 list_remove(&r->node);
525 for (i = 0; i < 2; i++) {
526 struct half *this = &r->halves[i];
527 rconn_destroy(this->rconn);
528 buffer_delete(this->rxbuf);
533 /* In-band control. */
535 struct in_band_data {
536 const struct settings *s;
537 struct mac_learning *ml;
538 struct netdev *of_device;
539 struct rconn *controller;
540 uint8_t mac[ETH_ADDR_LEN];
545 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct buffer *b)
547 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
550 static const uint8_t *
551 get_controller_mac(struct in_band_data *in_band)
553 static uint32_t ip, last_nonzero_ip;
554 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
555 static time_t next_refresh = 0;
557 uint32_t last_ip = ip;
559 time_t now = time_now();
561 ip = rconn_get_ip(in_band->controller);
562 if (last_ip != ip || !next_refresh || now >= next_refresh) {
565 /* Look up MAC address. */
566 memset(mac, 0, sizeof mac);
568 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
570 VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s",
571 IP_ARGS(&ip), strerror(retval));
574 have_mac = !eth_addr_is_zero(mac);
576 /* Log changes in IP, MAC addresses. */
577 if (ip && ip != last_nonzero_ip) {
578 VLOG_DBG("controller IP address changed from "IP_FMT
579 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
580 last_nonzero_ip = ip;
582 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
583 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
585 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
586 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
589 /* Schedule next refresh.
591 * If we have an IP address but not a MAC address, then refresh
592 * quickly, since we probably will get a MAC address soon (via ARP).
593 * Otherwise, we can afford to wait a little while. */
594 next_refresh = now + (!ip || have_mac ? 10 : 1);
596 return !eth_addr_is_zero(mac) ? mac : NULL;
600 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
601 struct in_band_data *in_band)
603 const uint8_t *mac = get_controller_mac(in_band);
604 return mac && eth_addr_equals(mac, dl_addr);
608 in_band_learn_mac(struct in_band_data *in_band, const struct flow *flow)
610 uint16_t in_port = ntohs(flow->in_port);
611 if (mac_learning_learn(in_band->ml, flow->dl_src, in_port)) {
612 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
613 ETH_ADDR_ARGS(flow->dl_src), in_port);
618 in_band_packet_cb(struct relay *r, int half, void *in_band_)
620 struct in_band_data *in_band = in_band_;
621 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
622 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
623 struct ofp_packet_in *opi;
624 struct ofp_header *oh;
625 size_t pkt_ofs, pkt_len;
628 uint16_t in_port, out_port;
630 if (half != HALF_LOCAL || r->is_mgmt_conn) {
635 if (oh->type != OFPT_PACKET_IN) {
638 if (msg->size < offsetof(struct ofp_packet_in, data)) {
639 VLOG_WARN_RL(&vrl, "packet too short (%zu bytes) for packet_in",
644 /* Extract flow data from 'opi' into 'flow'. */
646 in_port = ntohs(opi->in_port);
647 pkt_ofs = offsetof(struct ofp_packet_in, data);
648 pkt_len = ntohs(opi->header.length) - pkt_ofs;
649 pkt.data = opi->data;
651 flow_extract(&pkt, in_port, &flow);
653 /* Deal with local stuff. */
654 if (in_port == OFPP_LOCAL) {
655 /* Sent by secure channel. */
656 out_port = mac_learning_lookup(in_band->ml, flow.dl_dst);
657 } else if (eth_addr_equals(flow.dl_dst, in_band->mac)) {
658 /* Sent to secure channel. */
659 out_port = OFPP_LOCAL;
660 in_band_learn_mac(in_band, &flow);
661 } else if (flow.dl_type == htons(ETH_TYPE_ARP)
662 && eth_addr_is_broadcast(flow.dl_dst)
663 && is_controller_mac(flow.dl_src, in_band)) {
664 /* ARP sent by controller. */
665 out_port = OFPP_FLOOD;
666 } else if (is_controller_mac(flow.dl_dst, in_band)
667 || is_controller_mac(flow.dl_src, in_band)) {
668 /* Traffic to or from controller. Switch it by hand. */
669 in_band_learn_mac(in_band, &flow);
670 out_port = mac_learning_lookup(in_band->ml, flow.dl_dst);
675 if (in_port == out_port) {
676 /* The input and output port match. Set up a flow to drop packets. */
677 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
678 in_band->s->max_idle, 0));
679 } else if (out_port != OFPP_FLOOD) {
680 /* The output port is known, so add a new flow. */
681 queue_tx(rc, in_band,
682 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
683 out_port, in_band->s->max_idle));
685 /* If the switch didn't buffer the packet, we need to send a copy. */
686 if (ntohl(opi->buffer_id) == UINT32_MAX) {
687 queue_tx(rc, in_band,
688 make_unbuffered_packet_out(&pkt, in_port, out_port));
691 /* We don't know that MAC. Send along the packet without setting up a
694 if (ntohl(opi->buffer_id) == UINT32_MAX) {
695 b = make_unbuffered_packet_out(&pkt, in_port, out_port);
697 b = make_buffered_packet_out(ntohl(opi->buffer_id),
700 queue_tx(rc, in_band, b);
706 in_band_status_cb(struct status_reply *sr, void *in_band_)
708 struct in_band_data *in_band = in_band_;
709 struct in_addr local_ip;
710 uint32_t controller_ip;
711 const uint8_t *controller_mac;
713 if (netdev_get_in4(in_band->of_device, &local_ip)) {
714 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
716 status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
717 ETH_ADDR_ARGS(in_band->mac));
719 controller_ip = rconn_get_ip(in_band->controller);
721 status_reply_put(sr, "controller-ip="IP_FMT,
722 IP_ARGS(&controller_ip));
724 controller_mac = get_controller_mac(in_band);
725 if (controller_mac) {
726 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
727 ETH_ADDR_ARGS(controller_mac));
732 in_band_hook_create(const struct settings *s, struct switch_status *ss,
733 struct rconn *remote)
735 struct in_band_data *in_band;
738 in_band = xcalloc(1, sizeof *in_band);
740 in_band->ml = mac_learning_create();
741 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE,
742 &in_band->of_device);
744 fatal(retval, "Could not open %s device", s->of_name);
746 memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device),
748 in_band->controller = remote;
749 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
750 return make_hook(in_band_packet_cb, NULL, NULL, in_band);
753 /* Fail open support. */
755 struct fail_open_data {
756 const struct settings *s;
757 struct rconn *local_rconn;
758 struct rconn *remote_rconn;
759 struct lswitch *lswitch;
760 int last_disconn_secs;
763 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
765 fail_open_periodic_cb(void *fail_open_)
767 struct fail_open_data *fail_open = fail_open_;
771 disconn_secs = rconn_disconnected_duration(fail_open->remote_rconn);
772 open = disconn_secs >= fail_open->s->probe_interval * 3;
773 if (open != (fail_open->lswitch != NULL)) {
775 VLOG_WARN("No longer in fail-open mode");
776 lswitch_destroy(fail_open->lswitch);
777 fail_open->lswitch = NULL;
779 VLOG_WARN("Could not connect to controller for %d seconds, "
780 "failing open", disconn_secs);
781 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
782 fail_open->s->max_idle);
783 fail_open->last_disconn_secs = disconn_secs;
785 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
786 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
787 "from controller", disconn_secs);
788 fail_open->last_disconn_secs = disconn_secs;
793 fail_open_packet_cb(struct relay *r, int half, void *fail_open_)
795 struct fail_open_data *fail_open = fail_open_;
796 if (half != HALF_LOCAL || r->is_mgmt_conn || !fail_open->lswitch) {
799 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
800 r->halves[HALF_LOCAL].rxbuf);
801 rconn_run(fail_open->local_rconn);
807 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
809 struct fail_open_data *fail_open = fail_open_;
810 const struct settings *s = fail_open->s;
811 int trigger_duration = s->probe_interval * 3;
812 int cur_duration = rconn_disconnected_duration(fail_open->remote_rconn);
814 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
815 status_reply_put(sr, "current-duration=%d", cur_duration);
816 status_reply_put(sr, "triggered=%s",
817 cur_duration >= trigger_duration ? "true" : "false");
818 status_reply_put(sr, "max-idle=%d", s->max_idle);
822 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
823 struct rconn *local_rconn, struct rconn *remote_rconn)
825 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
827 fail_open->local_rconn = local_rconn;
828 fail_open->remote_rconn = remote_rconn;
829 fail_open->lswitch = NULL;
830 switch_status_register_category(ss, "fail-open",
831 fail_open_status_cb, fail_open);
832 return make_hook(fail_open_packet_cb, fail_open_periodic_cb, NULL,
836 struct rate_limiter {
837 const struct settings *s;
838 struct rconn *remote_rconn;
840 /* One queue per physical port. */
841 struct queue queues[OFPP_MAX];
842 int n_queued; /* Sum over queues[*].n. */
843 int next_tx_port; /* Next port to check in round-robin. */
847 * It costs 1000 tokens to send a single packet_in message. A single token
848 * per message would be more straightforward, but this choice lets us avoid
849 * round-off error in refill_bucket()'s calculation of how many tokens to
850 * add to the bucket, since no division step is needed. */
851 long long int last_fill; /* Time at which we last added tokens. */
852 int tokens; /* Current number of tokens. */
854 /* Transmission queue. */
855 int n_txq; /* No. of packets waiting in rconn for tx. */
857 /* Statistics reporting. */
858 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
859 unsigned long long n_limited; /* # queued for rate limiting. */
860 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
861 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
864 /* Drop a packet from the longest queue in 'rl'. */
866 drop_packet(struct rate_limiter *rl)
868 struct queue *longest; /* Queue currently selected as longest. */
869 int n_longest; /* # of queues of same length as 'longest'. */
872 longest = &rl->queues[0];
874 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
875 if (longest->n < q->n) {
878 } else if (longest->n == q->n) {
881 /* Randomly select one of the longest queues, with a uniform
882 * distribution (Knuth algorithm 3.4.2R). */
883 if (!random_range(n_longest)) {
889 /* FIXME: do we want to pop the tail instead? */
890 buffer_delete(queue_pop_head(longest));
894 /* Remove and return the next packet to transmit (in round-robin order). */
895 static struct buffer *
896 dequeue_packet(struct rate_limiter *rl)
900 for (i = 0; i < OFPP_MAX; i++) {
901 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
902 struct queue *q = &rl->queues[port];
904 rl->next_tx_port = (port + 1) % OFPP_MAX;
906 return queue_pop_head(q);
912 /* Add tokens to the bucket based on elapsed time. */
914 refill_bucket(struct rate_limiter *rl)
916 const struct settings *s = rl->s;
917 long long int now = time_msec();
918 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
919 if (tokens >= 1000) {
921 rl->tokens = MIN(tokens, s->burst_limit * 1000);
925 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
926 * true if successful, false otherwise. (In the latter case no tokens are
929 get_token(struct rate_limiter *rl)
931 if (rl->tokens >= 1000) {
940 rate_limit_packet_cb(struct relay *r, int half, void *rl_)
942 struct rate_limiter *rl = rl_;
943 const struct settings *s = rl->s;
944 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
945 struct ofp_header *oh;
947 if (half == HALF_REMOTE) {
952 if (oh->type != OFPT_PACKET_IN) {
955 if (msg->size < offsetof(struct ofp_packet_in, data)) {
956 VLOG_WARN_RL(&vrl, "packet too short (%zu bytes) for packet_in",
961 if (!rl->n_queued && get_token(rl)) {
962 /* In the common case where we are not constrained by the rate limit,
963 * let the packet take the normal path. */
967 /* Otherwise queue it up for the periodic callback to drain out. */
968 struct ofp_packet_in *opi = msg->data;
969 int port = ntohs(opi->in_port) % OFPP_MAX;
970 if (rl->n_queued >= s->burst_limit) {
973 queue_push_tail(&rl->queues[port], buffer_clone(msg));
981 rate_limit_status_cb(struct status_reply *sr, void *rl_)
983 struct rate_limiter *rl = rl_;
985 status_reply_put(sr, "normal=%llu", rl->n_normal);
986 status_reply_put(sr, "limited=%llu", rl->n_limited);
987 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
988 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
992 rate_limit_periodic_cb(void *rl_)
994 struct rate_limiter *rl = rl_;
997 /* Drain some packets out of the bucket if possible, but limit the number
998 * of iterations to allow other code to get work done too. */
1000 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
1001 /* Use a small, arbitrary limit for the amount of queuing to do here,
1002 * because the TCP connection is responsible for buffering and there is
1003 * no point in trying to transmit faster than the TCP connection can
1005 struct buffer *b = dequeue_packet(rl);
1006 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
1013 rate_limit_wait_cb(void *rl_)
1015 struct rate_limiter *rl = rl_;
1017 if (rl->tokens >= 1000) {
1018 /* We can transmit more packets as soon as we're called again. */
1019 poll_immediate_wake();
1021 /* We have to wait for the bucket to re-fill. We could calculate
1022 * the exact amount of time here for increased smoothness. */
1023 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
1029 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
1030 struct rconn *local, struct rconn *remote)
1032 struct rate_limiter *rl;
1035 rl = xcalloc(1, sizeof *rl);
1037 rl->remote_rconn = remote;
1038 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
1039 queue_init(&rl->queues[i]);
1041 rl->last_fill = time_msec();
1042 rl->tokens = s->rate_limit * 100;
1043 switch_status_register_category(ss, "rate-limit",
1044 rate_limit_status_cb, rl);
1045 return make_hook(rate_limit_packet_cb, rate_limit_periodic_cb,
1046 rate_limit_wait_cb, rl);
1049 /* OFPST_SWITCH statistics. */
1051 struct switch_status_category {
1053 void (*cb)(struct status_reply *, void *aux);
1057 struct switch_status {
1058 const struct settings *s;
1060 struct switch_status_category categories[8];
1064 struct status_reply {
1065 struct switch_status_category *category;
1071 switch_status_packet_cb(struct relay *r, int half, void *ss_)
1073 struct switch_status *ss = ss_;
1074 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
1075 struct buffer *msg = r->halves[HALF_REMOTE].rxbuf;
1076 struct switch_status_category *c;
1077 struct ofp_stats_request *osr;
1078 struct ofp_stats_reply *reply;
1079 struct status_reply sr;
1080 struct ofp_header *oh;
1084 if (half == HALF_LOCAL) {
1089 if (oh->type != OFPT_STATS_REQUEST) {
1092 if (msg->size < sizeof(struct ofp_stats_request)) {
1093 VLOG_WARN_RL(&vrl, "packet too short (%zu bytes) for stats_request",
1099 if (osr->type != htons(OFPST_SWITCH)) {
1103 sr.request.string = (void *) (osr + 1);
1104 sr.request.length = msg->size - sizeof *osr;
1105 ds_init(&sr.output);
1106 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
1107 if (!memcmp(c->name, sr.request.string,
1108 MIN(strlen(c->name), sr.request.length))) {
1113 reply = make_openflow_xid((offsetof(struct ofp_stats_reply, body)
1114 + sr.output.length),
1115 OFPT_STATS_REPLY, osr->header.xid, &b);
1116 reply->type = htons(OFPST_SWITCH);
1118 memcpy(reply->body, sr.output.string, sr.output.length);
1119 retval = rconn_send(rc, b, NULL);
1120 if (retval && retval != EAGAIN) {
1121 VLOG_WARN("send failed (%s)", strerror(retval));
1123 ds_destroy(&sr.output);
1128 rconn_status_cb(struct status_reply *sr, void *rconn_)
1130 struct rconn *rconn = rconn_;
1131 time_t now = time_now();
1133 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
1134 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
1135 status_reply_put(sr, "backoff=%d", rconn_get_backoff(rconn));
1136 status_reply_put(sr, "is-connected=%s",
1137 rconn_is_connected(rconn) ? "true" : "false");
1138 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
1139 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
1140 status_reply_put(sr, "attempted-connections=%u",
1141 rconn_get_attempted_connections(rconn));
1142 status_reply_put(sr, "successful-connections=%u",
1143 rconn_get_successful_connections(rconn));
1144 status_reply_put(sr, "last-connection=%ld",
1145 (long int) (now - rconn_get_last_connection(rconn)));
1146 status_reply_put(sr, "time-connected=%lu",
1147 rconn_get_total_time_connected(rconn));
1148 status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(rconn));
1152 config_status_cb(struct status_reply *sr, void *s_)
1154 const struct settings *s = s_;
1157 for (i = 0; i < s->n_listeners; i++) {
1158 status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
1160 if (s->probe_interval) {
1161 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
1163 if (s->max_backoff) {
1164 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
1169 switch_status_cb(struct status_reply *sr, void *ss_)
1171 struct switch_status *ss = ss_;
1172 time_t now = time_now();
1174 status_reply_put(sr, "now=%ld", (long int) now);
1175 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
1176 status_reply_put(sr, "pid=%ld", (long int) getpid());
1180 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
1182 struct switch_status *ss = xcalloc(1, sizeof *ss);
1184 ss->booted = time_now();
1185 switch_status_register_category(ss, "config",
1186 config_status_cb, (void *) s);
1187 switch_status_register_category(ss, "switch", switch_status_cb, ss);
1189 return make_hook(switch_status_packet_cb, NULL, NULL, ss);
1193 switch_status_register_category(struct switch_status *ss,
1194 const char *category,
1195 void (*cb)(struct status_reply *,
1199 struct switch_status_category *c;
1200 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
1201 c = &ss->categories[ss->n_categories++];
1204 c->name = xstrdup(category);
1208 status_reply_put(struct status_reply *sr, const char *content, ...)
1210 size_t old_length = sr->output.length;
1214 /* Append the status reply to the output. */
1215 ds_put_format(&sr->output, "%s.", sr->category->name);
1216 va_start(args, content);
1217 ds_put_format_valist(&sr->output, content, args);
1219 if (ds_last(&sr->output) != '\n') {
1220 ds_put_char(&sr->output, '\n');
1223 /* Drop what we just added if it doesn't match the request. */
1224 added = sr->output.length - old_length;
1225 if (added < sr->request.length
1226 || memcmp(&sr->output.string[old_length],
1227 sr->request.string, sr->request.length)) {
1228 ds_truncate(&sr->output, old_length);
1233 /* Controller discovery. */
1237 const struct settings *s;
1238 struct dhclient *dhcp;
1243 discovery_status_cb(struct status_reply *sr, void *d_)
1245 struct discovery *d = d_;
1247 status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re);
1248 status_reply_put(sr, "n-changes=%d", d->n_changes);
1249 status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp));
1250 status_reply_put(sr, "state-elapsed=%u",
1251 dhclient_get_state_elapsed(d->dhcp));
1252 if (dhclient_is_bound(d->dhcp)) {
1253 uint32_t ip = dhclient_get_ip(d->dhcp);
1254 uint32_t netmask = dhclient_get_netmask(d->dhcp);
1255 uint32_t router = dhclient_get_router(d->dhcp);
1257 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
1258 uint32_t dns_server;
1262 status_reply_put(sr, "ip="IP_FMT, IP_ARGS(&ip));
1263 status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask));
1265 status_reply_put(sr, "router="IP_FMT, IP_ARGS(&router));
1268 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i, &dns_server);
1270 status_reply_put(sr, "dns%d="IP_FMT, i, IP_ARGS(&dns_server));
1273 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
1275 status_reply_put(sr, "domain=%s", domain_name);
1279 status_reply_put(sr, "lease-remaining=%u",
1280 dhclient_get_lease_remaining(d->dhcp));
1284 static struct discovery *
1285 discovery_init(const struct settings *s, struct switch_status *ss)
1287 struct netdev *netdev;
1288 struct discovery *d;
1289 struct dhclient *dhcp;
1292 /* Bring ofX network device up. */
1293 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1295 fatal(retval, "Could not open %s device", s->of_name);
1297 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
1299 fatal(retval, "Could not bring %s device up", s->of_name);
1301 netdev_close(netdev);
1303 /* Initialize DHCP client. */
1304 retval = dhclient_create(s->of_name, modify_dhcp_request,
1305 validate_dhcp_offer, (void *) s, &dhcp);
1307 fatal(retval, "Failed to initialize DHCP client");
1309 dhclient_init(dhcp, 0);
1311 d = xmalloc(sizeof *d);
1316 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
1322 discovery_question_connectivity(struct discovery *d)
1324 dhclient_force_renew(d->dhcp, 15);
1328 discovery_run(struct discovery *d, char **controller_name)
1330 dhclient_run(d->dhcp);
1331 if (!dhclient_changed(d->dhcp)) {
1335 dhclient_configure_netdev(d->dhcp);
1336 if (d->s->update_resolv_conf) {
1337 dhclient_update_resolv_conf(d->dhcp);
1340 if (dhclient_is_bound(d->dhcp)) {
1341 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
1342 DHCP_CODE_OFP_CONTROLLER_VCONN);
1343 VLOG_WARN("%s: discovered controller", *controller_name);
1346 *controller_name = NULL;
1348 VLOG_WARN("discovered controller no longer available");
1356 discovery_wait(struct discovery *d)
1358 dhclient_wait(d->dhcp);
1362 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
1364 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
1368 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
1370 const struct settings *s = s_;
1374 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
1376 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
1379 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
1381 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
1382 s->accept_controller_re);
1388 /* User interface. */
1391 parse_options(int argc, char *argv[], struct settings *s)
1394 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
1396 OPT_INACTIVITY_PROBE,
1402 static struct option long_options[] = {
1403 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
1404 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
1405 {"fail", required_argument, 0, 'F'},
1406 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
1407 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
1408 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
1409 {"listen", required_argument, 0, 'l'},
1410 {"monitor", required_argument, 0, 'm'},
1411 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
1412 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
1413 {"detach", no_argument, 0, 'D'},
1414 {"force", no_argument, 0, 'f'},
1415 {"pidfile", optional_argument, 0, 'P'},
1416 {"verbose", optional_argument, 0, 'v'},
1417 {"help", no_argument, 0, 'h'},
1418 {"version", no_argument, 0, 'V'},
1419 VCONN_SSL_LONG_OPTIONS
1422 char *short_options = long_options_to_short_options(long_options);
1423 char *accept_re = NULL;
1426 /* Set defaults that we can figure out before parsing options. */
1428 s->monitor_name = NULL;
1429 s->fail_mode = FAIL_OPEN;
1431 s->probe_interval = 15;
1432 s->max_backoff = 15;
1433 s->update_resolv_conf = true;
1439 c = getopt_long(argc, argv, short_options, long_options, NULL);
1445 case OPT_ACCEPT_VCONN:
1446 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
1449 case OPT_NO_RESOLV_CONF:
1450 s->update_resolv_conf = false;
1454 if (!strcmp(optarg, "open")) {
1455 s->fail_mode = FAIL_OPEN;
1456 } else if (!strcmp(optarg, "closed")) {
1457 s->fail_mode = FAIL_CLOSED;
1460 "-f or --fail argument must be \"open\" or \"closed\"");
1464 case OPT_INACTIVITY_PROBE:
1465 s->probe_interval = atoi(optarg);
1466 if (s->probe_interval < 5) {
1467 fatal(0, "--inactivity-probe argument must be at least 5");
1472 if (!strcmp(optarg, "permanent")) {
1473 s->max_idle = OFP_FLOW_PERMANENT;
1475 s->max_idle = atoi(optarg);
1476 if (s->max_idle < 1 || s->max_idle > 65535) {
1477 fatal(0, "--max-idle argument must be between 1 and "
1478 "65535 or the word 'permanent'");
1483 case OPT_MAX_BACKOFF:
1484 s->max_backoff = atoi(optarg);
1485 if (s->max_backoff < 1) {
1486 fatal(0, "--max-backoff argument must be at least 1");
1487 } else if (s->max_backoff > 3600) {
1488 s->max_backoff = 3600;
1492 case OPT_RATE_LIMIT:
1494 s->rate_limit = atoi(optarg);
1495 if (s->rate_limit < 1) {
1496 fatal(0, "--rate-limit argument must be at least 1");
1499 s->rate_limit = 1000;
1503 case OPT_BURST_LIMIT:
1504 s->burst_limit = atoi(optarg);
1505 if (s->burst_limit < 1) {
1506 fatal(0, "--burst-limit argument must be at least 1");
1515 set_pidfile(optarg);
1519 ignore_existing_pidfile();
1523 if (s->n_listeners >= MAX_MGMT) {
1524 fatal(0, "-l or --listen may be specified at most %d times",
1527 s->listener_names[s->n_listeners++] = optarg;
1531 if (s->monitor_name) {
1532 fatal(0, "-m or --monitor may only be specified once");
1534 s->monitor_name = optarg;
1541 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
1545 vlog_set_verbosity(optarg);
1548 VCONN_SSL_OPTION_HANDLERS
1557 free(short_options);
1561 if (argc < 1 || argc > 2) {
1562 fatal(0, "need one or two non-option arguments; use --help for usage");
1565 /* Local and remote vconns. */
1566 s->nl_name = argv[0];
1567 if (strncmp(s->nl_name, "nl:", 3)
1568 || strlen(s->nl_name) < 4
1569 || s->nl_name[strspn(s->nl_name + 3, "0123456789") + 3]) {
1570 fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"", s->nl_name);
1572 s->of_name = xasprintf("of%s", s->nl_name + 3);
1573 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
1575 /* Set accept_controller_regex. */
1577 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
1579 retval = regcomp(&s->accept_controller_regex, accept_re,
1580 REG_NOSUB | REG_EXTENDED);
1582 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
1583 char *buffer = xmalloc(length);
1584 regerror(retval, &s->accept_controller_regex, buffer, length);
1585 fatal(0, "%s: %s", accept_re, buffer);
1587 s->accept_controller_re = accept_re;
1589 /* Mode of operation. */
1590 s->discovery = s->controller_name == NULL;
1594 enum netdev_flags flags;
1595 struct netdev *netdev;
1597 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1599 fatal(retval, "Could not open %s device", s->of_name);
1602 retval = netdev_get_flags(netdev, &flags);
1604 fatal(retval, "Could not get flags for %s device", s->of_name);
1607 s->in_band = (flags & NETDEV_UP) != 0;
1608 if (s->in_band && netdev_get_in6(netdev, NULL)) {
1609 VLOG_WARN("Ignoring IPv6 address on %s device: IPv6 not supported",
1613 netdev_close(netdev);
1616 /* Rate limiting. */
1617 if (s->rate_limit) {
1618 if (s->rate_limit < 100) {
1619 VLOG_WARN("Rate limit set to unusually low value %d",
1622 if (!s->burst_limit) {
1623 s->burst_limit = s->rate_limit / 4;
1625 s->burst_limit = MAX(s->burst_limit, 1);
1626 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
1633 printf("%s: secure channel, a relay for OpenFlow messages.\n"
1634 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
1635 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
1636 "CONTROLLER is an active OpenFlow connection method; if it is\n"
1637 "omitted, then secchan performs controller discovery.\n",
1638 program_name, program_name);
1639 vconn_usage(true, true);
1640 printf("\nController discovery options:\n"
1641 " --accept-vconn=REGEX accept matching discovered controllers\n"
1642 " --no-resolv-conf do not update /etc/resolv.conf\n"
1643 "\nNetworking options:\n"
1644 " -F, --fail=open|closed when controller connection fails:\n"
1645 " closed: drop all packets\n"
1646 " open (default): act as learning switch\n"
1647 " --inactivity-probe=SECS time between inactivity probes\n"
1648 " --max-idle=SECS max idle for flows set up by secchan\n"
1649 " --max-backoff=SECS max time between controller connection\n"
1650 " attempts (default: 15 seconds)\n"
1651 " -l, --listen=METHOD allow management connections on METHOD\n"
1652 " (a passive OpenFlow connection method)\n"
1653 " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n"
1654 " (a passive OpenFlow connection method)\n"
1655 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
1656 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
1657 " --burst-limit=BURST limit on packet credit for idle time\n"
1658 "\nOther options:\n"
1659 " -D, --detach run in background as daemon\n"
1660 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
1661 " -f, --force with -P, start even if already running\n"
1662 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
1663 " -v, --verbose set maximum verbosity level\n"
1664 " -h, --help display this help message\n"
1665 " -V, --version display version information\n",