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"
61 #include "nicira-ext.h"
64 #include "poll-loop.h"
69 #include "vconn-ssl.h"
71 #include "vlog-socket.h"
74 #define THIS_MODULE VLM_secchan
76 /* Behavior when the connection to the controller fails. */
78 FAIL_OPEN, /* Act as learning switch. */
79 FAIL_CLOSED /* Drop all packets. */
82 /* Maximum number of management connection listeners. */
85 /* Settings that may be configured by the user. */
87 /* Overall mode of operation. */
88 bool discovery; /* Discover the controller automatically? */
89 bool in_band; /* Connect to controller in-band? */
91 /* Related vconns and network devices. */
92 const char *nl_name; /* Local datapath (must be "nl:" vconn). */
93 char *of_name; /* ofX network device name. */
94 const char *controller_name; /* Controller (if not discovery mode). */
95 const char *listener_names[MAX_MGMT]; /* Listen for mgmt connections. */
96 size_t n_listeners; /* Number of mgmt connection listeners. */
97 const char *monitor_name; /* Listen for traffic monitor connections. */
99 /* Failure behavior. */
100 enum fail_mode fail_mode; /* Act as learning switch if no controller? */
101 int max_idle; /* Idle time for flows in fail-open mode. */
102 int probe_interval; /* # seconds idle before sending echo request. */
103 int max_backoff; /* Max # seconds between connection attempts. */
105 /* Packet-in rate-limiting. */
106 int rate_limit; /* Tokens added to bucket per second. */
107 int burst_limit; /* Maximum number token bucket size. */
109 /* Discovery behavior. */
110 regex_t accept_controller_regex; /* Controller vconns to accept. */
111 const char *accept_controller_re; /* String version of regex. */
112 bool update_resolv_conf; /* Update /etc/resolv.conf? */
117 struct buffer *rxbuf;
118 int n_txq; /* No. of packets queued for tx on 'rconn'. */
125 #define HALF_REMOTE 1
126 struct half halves[2];
132 bool (*packet_cb[2])(struct relay *, void *aux);
133 void (*periodic_cb)(void *aux);
134 void (*wait_cb)(void *aux);
138 static struct vlog_rate_limit vrl = VLOG_RATE_LIMIT_INIT(60, 60);
140 static void parse_options(int argc, char *argv[], struct settings *);
141 static void usage(void) NO_RETURN;
143 static struct pvconn *open_passive_vconn(const char *name);
144 static struct vconn *accept_vconn(struct pvconn *pvconn);
146 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
148 static struct relay *relay_accept(const struct settings *, struct pvconn *);
149 static void relay_run(struct relay *, const struct hook[], size_t n_hooks);
150 static void relay_wait(struct relay *);
151 static void relay_destroy(struct relay *);
153 static struct hook make_hook(bool (*local_packet_cb)(struct relay *, void *),
154 bool (*remote_packet_cb)(struct relay *, void *),
155 void (*periodic_cb)(void *),
156 void (*wait_cb)(void *),
158 static struct ofp_packet_in *get_ofp_packet_in(struct relay *);
159 static bool get_ofp_packet_eth_header(struct relay *, struct ofp_packet_in **,
160 struct eth_header **);
161 static void get_ofp_packet_payload(struct ofp_packet_in *, struct buffer *);
163 struct switch_status;
165 static struct hook switch_status_hook_create(const struct settings *,
166 struct switch_status **);
167 static void switch_status_register_category(struct switch_status *,
168 const char *category,
169 void (*cb)(struct status_reply *,
172 static void status_reply_put(struct status_reply *, const char *, ...)
175 static void rconn_status_cb(struct status_reply *, void *rconn_);
177 static struct discovery *discovery_init(const struct settings *,
178 struct switch_status *);
179 static void discovery_question_connectivity(struct discovery *);
180 static bool discovery_run(struct discovery *, char **controller_name);
181 static void discovery_wait(struct discovery *);
183 static struct hook in_band_hook_create(const struct settings *,
184 struct switch_status *,
185 struct rconn *remote);
188 static struct hook port_watcher_create(struct rconn *local,
189 struct rconn *remote,
190 struct port_watcher **);
191 static uint32_t port_watcher_get_flags(const struct port_watcher *,
193 static void port_watcher_set_flags(struct port_watcher *,
194 int port_no, uint32_t flags, uint32_t mask);
196 static struct hook stp_hook_create(const struct settings *,
197 struct port_watcher *,
198 struct rconn *local, struct rconn *remote);
200 static struct hook fail_open_hook_create(const struct settings *,
201 struct switch_status *,
203 struct rconn *remote);
204 static struct hook rate_limit_hook_create(const struct settings *,
205 struct switch_status *,
207 struct rconn *remote);
210 static void modify_dhcp_request(struct dhcp_msg *, void *aux);
211 static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux);
214 main(int argc, char *argv[])
218 struct list relays = LIST_INITIALIZER(&relays);
220 struct hook hooks[8];
223 struct pvconn *monitor;
225 struct pvconn *listeners[MAX_MGMT];
228 struct rconn *local_rconn, *remote_rconn;
229 struct relay *controller_relay;
230 struct discovery *discovery;
231 struct switch_status *switch_status;
232 struct port_watcher *pw;
236 set_program_name(argv[0]);
237 register_fault_handlers();
240 parse_options(argc, argv, &s);
241 signal(SIGPIPE, SIG_IGN);
243 /* Start listening for management and monitoring connections. */
245 for (i = 0; i < s.n_listeners; i++) {
246 listeners[n_listeners++] = open_passive_vconn(s.listener_names[i]);
248 monitor = s.monitor_name ? open_passive_vconn(s.monitor_name) : NULL;
250 /* Initialize switch status hook. */
251 hooks[n_hooks++] = switch_status_hook_create(&s, &switch_status);
253 /* Start controller discovery. */
254 discovery = s.discovery ? discovery_init(&s, switch_status) : NULL;
256 /* Start listening for vlogconf requests. */
257 retval = vlog_server_listen(NULL, NULL);
259 fatal(retval, "Could not listen for vlog connections");
262 die_if_already_running();
265 VLOG_WARN("OpenFlow reference implementation version %s", VERSION);
266 VLOG_WARN("OpenFlow protocol version 0x%02x", OFP_VERSION);
268 /* Connect to datapath. */
269 local_rconn = rconn_create(0, s.max_backoff);
270 rconn_connect(local_rconn, s.nl_name);
271 switch_status_register_category(switch_status, "local",
272 rconn_status_cb, local_rconn);
274 /* Connect to controller. */
275 remote_rconn = rconn_create(s.probe_interval, s.max_backoff);
276 if (s.controller_name) {
277 retval = rconn_connect(remote_rconn, s.controller_name);
278 if (retval == EAFNOSUPPORT) {
279 fatal(0, "No support for %s vconn", s.controller_name);
282 switch_status_register_category(switch_status, "remote",
283 rconn_status_cb, remote_rconn);
285 /* Start relaying. */
286 controller_relay = relay_create(local_rconn, remote_rconn, false);
287 list_push_back(&relays, &controller_relay->node);
290 hooks[n_hooks++] = port_watcher_create(local_rconn, remote_rconn, &pw);
291 hooks[n_hooks++] = stp_hook_create(&s, pw, local_rconn, remote_rconn);
293 hooks[n_hooks++] = in_band_hook_create(&s, switch_status,
296 if (s.fail_mode == FAIL_OPEN) {
297 hooks[n_hooks++] = fail_open_hook_create(&s, switch_status,
298 local_rconn, remote_rconn);
301 hooks[n_hooks++] = rate_limit_hook_create(&s, switch_status,
302 local_rconn, remote_rconn);
304 assert(n_hooks <= ARRAY_SIZE(hooks));
311 LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) {
312 relay_run(r, hooks, n_hooks);
314 for (i = 0; i < n_listeners; i++) {
316 struct relay *r = relay_accept(&s, listeners[i]);
320 list_push_back(&relays, &r->node);
324 struct vconn *new = accept_vconn(monitor);
326 rconn_add_monitor(local_rconn, new);
329 for (i = 0; i < n_hooks; i++) {
330 if (hooks[i].periodic_cb) {
331 hooks[i].periodic_cb(hooks[i].aux);
335 char *controller_name;
336 if (rconn_is_connectivity_questionable(remote_rconn)) {
337 discovery_question_connectivity(discovery);
339 if (discovery_run(discovery, &controller_name)) {
340 if (controller_name) {
341 rconn_connect(remote_rconn, controller_name);
343 rconn_disconnect(remote_rconn);
348 /* Wait for something to happen. */
349 LIST_FOR_EACH (r, struct relay, node, &relays) {
352 for (i = 0; i < n_listeners; i++) {
353 pvconn_wait(listeners[i]);
356 pvconn_wait(monitor);
358 for (i = 0; i < n_hooks; i++) {
359 if (hooks[i].wait_cb) {
360 hooks[i].wait_cb(hooks[i].aux);
364 discovery_wait(discovery);
372 static struct pvconn *
373 open_passive_vconn(const char *name)
375 struct pvconn *pvconn;
378 retval = pvconn_open(name, &pvconn);
379 if (retval && retval != EAGAIN) {
380 fatal(retval, "opening %s", name);
385 static struct vconn *
386 accept_vconn(struct pvconn *pvconn)
391 retval = pvconn_accept(pvconn, &new);
392 if (retval && retval != EAGAIN) {
393 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
399 make_hook(bool (*local_packet_cb)(struct relay *, void *aux),
400 bool (*remote_packet_cb)(struct relay *, void *aux),
401 void (*periodic_cb)(void *aux),
402 void (*wait_cb)(void *aux),
406 h.packet_cb[HALF_LOCAL] = local_packet_cb;
407 h.packet_cb[HALF_REMOTE] = remote_packet_cb;
408 h.periodic_cb = periodic_cb;
414 static struct ofp_packet_in *
415 get_ofp_packet_in(struct relay *r)
417 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
418 struct ofp_header *oh = msg->data;
419 if (oh->type == OFPT_PACKET_IN) {
420 if (msg->size >= offsetof (struct ofp_packet_in, data)) {
423 VLOG_WARN("packet too short (%zu bytes) for packet_in",
431 get_ofp_packet_eth_header(struct relay *r, struct ofp_packet_in **opip,
432 struct eth_header **ethp)
434 const int min_len = offsetof(struct ofp_packet_in, data) + ETH_HEADER_LEN;
435 struct ofp_packet_in *opi = get_ofp_packet_in(r);
436 if (opi && ntohs(opi->header.length) >= min_len) {
438 *ethp = (void *) opi->data;
445 /* OpenFlow message relaying. */
447 static struct relay *
448 relay_accept(const struct settings *s, struct pvconn *pvconn)
450 struct vconn *new_remote, *new_local;
451 char *nl_name_without_subscription;
452 struct rconn *r1, *r2;
455 new_remote = accept_vconn(pvconn);
460 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We
461 * only accept the former syntax in main().
463 * nl:123:0 opens a netlink connection to local datapath 123 without
464 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
466 nl_name_without_subscription = xasprintf("%s:0", s->nl_name);
467 retval = vconn_open(nl_name_without_subscription, &new_local);
469 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
470 nl_name_without_subscription, strerror(retval));
471 vconn_close(new_remote);
472 free(nl_name_without_subscription);
476 /* Create and return relay. */
477 r1 = rconn_create(0, 0);
478 rconn_connect_unreliably(r1, nl_name_without_subscription, new_local);
479 free(nl_name_without_subscription);
481 r2 = rconn_create(0, 0);
482 rconn_connect_unreliably(r2, "passive", new_remote);
484 return relay_create(r1, r2, true);
487 static struct relay *
488 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
490 struct relay *r = xcalloc(1, sizeof *r);
491 r->halves[HALF_LOCAL].rconn = local;
492 r->halves[HALF_REMOTE].rconn = remote;
493 r->is_mgmt_conn = is_mgmt_conn;
498 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
503 for (i = 0; i < 2; i++) {
504 rconn_run(r->halves[i].rconn);
507 /* Limit the number of iterations to prevent other tasks from starving. */
508 for (iteration = 0; iteration < 50; iteration++) {
509 bool progress = false;
510 for (i = 0; i < 2; i++) {
511 struct half *this = &r->halves[i];
512 struct half *peer = &r->halves[!i];
515 this->rxbuf = rconn_recv(this->rconn);
516 if (this->rxbuf && (i == HALF_REMOTE || !r->is_mgmt_conn)) {
517 const struct hook *h;
518 for (h = hooks; h < &hooks[n_hooks]; h++) {
519 if (h->packet_cb[i] && h->packet_cb[i](r, h->aux)) {
520 buffer_delete(this->rxbuf);
529 if (this->rxbuf && !this->n_txq) {
530 int retval = rconn_send(peer->rconn, this->rxbuf,
532 if (retval != EAGAIN) {
536 buffer_delete(this->rxbuf);
547 if (r->is_mgmt_conn) {
548 for (i = 0; i < 2; i++) {
549 struct half *this = &r->halves[i];
550 if (!rconn_is_alive(this->rconn)) {
559 relay_wait(struct relay *r)
563 for (i = 0; i < 2; i++) {
564 struct half *this = &r->halves[i];
566 rconn_run_wait(this->rconn);
568 rconn_recv_wait(this->rconn);
574 relay_destroy(struct relay *r)
578 list_remove(&r->node);
579 for (i = 0; i < 2; i++) {
580 struct half *this = &r->halves[i];
581 rconn_destroy(this->rconn);
582 buffer_delete(this->rxbuf);
587 /* Port status watcher. */
589 typedef void port_watcher_cb_func(uint16_t port_no,
590 const struct ofp_phy_port *old,
591 const struct ofp_phy_port *new,
594 struct port_watcher_cb {
595 port_watcher_cb_func *function;
599 struct port_watcher {
600 struct rconn *local_rconn;
601 struct rconn *remote_rconn;
602 struct ofp_phy_port ports[OFPP_MAX + 1];
603 time_t last_feature_request;
604 bool got_feature_reply;
606 struct port_watcher_cb cbs[2];
610 /* Returns the number of fields that differ from 'a' to 'b'. */
612 opp_differs(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
614 BUILD_ASSERT_DECL(sizeof *a == 36); /* Trips when we add or remove fields. */
615 return ((a->port_no != b->port_no)
616 + (memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) != 0)
617 + (memcmp(a->name, b->name, sizeof a->name) != 0)
618 + (a->flags != b->flags)
619 + (a->speed != b->speed)
620 + (a->features != b->features));
624 sanitize_opp(struct ofp_phy_port *opp)
628 for (i = 0; i < sizeof opp->name; i++) {
629 char c = opp->name[i];
630 if (c && (c < 0x20 || c > 0x7e)) {
634 opp->name[sizeof opp->name - 1] = '\0';
638 port_no_to_pw_idx(int port_no)
640 return (port_no < OFPP_MAX ? port_no
641 : port_no == OFPP_LOCAL ? OFPP_MAX
646 call_pw_callbacks(struct port_watcher *pw, int port_no,
647 const struct ofp_phy_port *old,
648 const struct ofp_phy_port *new)
650 if (opp_differs(old, new)) {
652 for (i = 0; i < pw->n_cbs; i++) {
653 pw->cbs[i].function(port_no, old, new, pw->cbs[i].aux);
659 update_phy_port(struct port_watcher *pw, struct ofp_phy_port *opp,
660 uint8_t reason, bool seen[OFPP_MAX + 1])
662 struct ofp_phy_port *pw_opp;
663 struct ofp_phy_port old;
667 port_no = ntohs(opp->port_no);
668 idx = port_no_to_pw_idx(port_no);
677 pw_opp = &pw->ports[idx];
679 if (reason == OFPPR_DELETE) {
680 memset(pw_opp, 0, sizeof *pw_opp);
681 pw_opp->port_no = htons(OFPP_NONE);
682 } else if (reason == OFPPR_MOD || reason == OFPPR_ADD) {
684 sanitize_opp(pw_opp);
686 call_pw_callbacks(pw, port_no, &old, pw_opp);
690 port_watcher_local_packet_cb(struct relay *r, void *pw_)
692 struct port_watcher *pw = pw_;
693 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
694 struct ofp_header *oh = msg->data;
696 if (oh->type == OFPT_FEATURES_REPLY
697 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
698 struct ofp_switch_features *osf = msg->data;
699 bool seen[ARRAY_SIZE(pw->ports)];
703 pw->got_feature_reply = true;
705 /* Update each port included in the message. */
706 memset(seen, 0, sizeof seen);
707 n_ports = ((msg->size - offsetof(struct ofp_switch_features, ports))
708 / sizeof *osf->ports);
709 for (i = 0; i < n_ports; i++) {
710 update_phy_port(pw, &osf->ports[i], OFPPR_MOD, seen);
713 /* Delete all the ports not included in the message. */
714 for (i = 0; i < ARRAY_SIZE(pw->ports); i++) {
716 update_phy_port(pw, &pw->ports[i], OFPPR_DELETE, NULL);
719 } else if (oh->type == OFPT_PORT_STATUS
720 && msg->size >= sizeof(struct ofp_port_status)) {
721 struct ofp_port_status *ops = msg->data;
722 update_phy_port(pw, &ops->desc, ops->reason, NULL);
728 port_watcher_remote_packet_cb(struct relay *r, void *pw_)
730 struct port_watcher *pw = pw_;
731 struct buffer *msg = r->halves[HALF_REMOTE].rxbuf;
732 struct ofp_header *oh = msg->data;
734 if (oh->type == OFPT_PORT_MOD
735 && msg->size >= sizeof(struct ofp_port_mod)) {
736 struct ofp_port_mod *opm = msg->data;
737 uint16_t port_no = ntohs(opm->desc.port_no);
738 int idx = port_no_to_pw_idx(port_no);
740 struct ofp_phy_port *pw_opp = &pw->ports[idx];
741 if (pw_opp->port_no != htons(OFPP_NONE)) {
742 struct ofp_phy_port old = *pw_opp;
743 pw_opp->flags = ((pw_opp->flags & ~opm->mask)
744 | (opm->desc.flags & opm->mask));
745 call_pw_callbacks(pw, port_no, &old, pw_opp);
753 port_watcher_periodic_cb(void *pw_)
755 struct port_watcher *pw = pw_;
757 if (!pw->got_feature_reply && time_now() >= pw->last_feature_request + 5) {
759 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
760 rconn_send_with_limit(pw->local_rconn, b, &pw->n_txq, 1);
761 pw->last_feature_request = time_now();
766 put_duplexes(struct ds *ds, const char *name, uint32_t features,
767 uint32_t hd_bit, uint32_t fd_bit)
769 if (features & (hd_bit | fd_bit)) {
770 ds_put_format(ds, " %s", name);
771 if (features & hd_bit) {
772 ds_put_cstr(ds, "(HD)");
774 if (features & fd_bit) {
775 ds_put_cstr(ds, "(FD)");
781 log_port_status(uint16_t port_no,
782 const struct ofp_phy_port *old,
783 const struct ofp_phy_port *new,
786 if (VLOG_IS_DBG_ENABLED()) {
787 bool was_enabled = old->port_no != htons(OFPP_NONE);
788 bool now_enabled = new->port_no != htons(OFPP_NONE);
789 uint32_t features = ntohl(new->features);
792 if (old->flags != new->flags && opp_differs(old, new) == 1) {
793 /* Don't care if only flags changed. */
798 ds_put_format(&ds, "\"%s\", "ETH_ADDR_FMT, new->name,
799 ETH_ADDR_ARGS(new->hw_addr));
800 if (ntohl(new->speed)) {
801 ds_put_format(&ds, ", speed %"PRIu32, ntohl(new->speed));
803 if (features & (OFPPF_10MB_HD | OFPPF_10MB_FD
804 | OFPPF_100MB_HD | OFPPF_100MB_FD
805 | OFPPF_1GB_HD | OFPPF_1GB_FD | OFPPF_10GB_FD)) {
806 ds_put_cstr(&ds, ", supports");
807 put_duplexes(&ds, "10M", features, OFPPF_10MB_HD, OFPPF_10MB_FD);
808 put_duplexes(&ds, "100M", features,
809 OFPPF_100MB_HD, OFPPF_100MB_FD);
810 put_duplexes(&ds, "1G", features, OFPPF_100MB_HD, OFPPF_100MB_FD);
811 if (features & OFPPF_10GB_FD) {
812 ds_put_cstr(&ds, " 10G");
815 if (was_enabled != now_enabled) {
817 VLOG_DBG("Port %d added: %s", port_no, ds_cstr(&ds));
819 VLOG_DBG("Port %d deleted", port_no);
822 VLOG_DBG("Port %d changed: %s", port_no, ds_cstr(&ds));
829 port_watcher_register_callback(struct port_watcher *pw,
830 port_watcher_cb_func *function,
833 assert(pw->n_cbs < ARRAY_SIZE(pw->cbs));
834 pw->cbs[pw->n_cbs].function = function;
835 pw->cbs[pw->n_cbs].aux = aux;
840 port_watcher_get_flags(const struct port_watcher *pw, int port_no)
842 int idx = port_no_to_pw_idx(port_no);
843 return idx >= 0 ? ntohl(pw->ports[idx].flags) : 0;
847 port_watcher_set_flags(struct port_watcher *pw,
848 int port_no, uint32_t flags, uint32_t mask)
850 struct ofp_phy_port old;
851 struct ofp_phy_port *p;
852 struct ofp_port_mod *opm;
853 struct ofp_port_status *ops;
857 idx = port_no_to_pw_idx(port_no);
863 if (!((ntohl(p->flags) ^ flags) & mask)) {
868 /* Update our idea of the flags. */
869 p->flags = htonl((ntohl(p->flags) & ~mask) | (flags & mask));
870 call_pw_callbacks(pw, port_no, &old, p);
872 /* Change the flags in the datapath. */
873 opm = make_openflow(sizeof *opm, OFPT_PORT_MOD, &b);
874 opm->mask = htonl(mask);
876 rconn_send(pw->local_rconn, b, NULL);
878 /* Notify the controller that the flags changed. */
879 ops = make_openflow(sizeof *ops, OFPT_PORT_STATUS, &b);
880 ops->reason = OFPPR_MOD;
882 rconn_send(pw->remote_rconn, b, NULL);
886 port_watcher_is_ready(const struct port_watcher *pw)
888 return pw->got_feature_reply;
892 port_watcher_create(struct rconn *local_rconn, struct rconn *remote_rconn,
893 struct port_watcher **pwp)
895 struct port_watcher *pw;
898 pw = *pwp = xcalloc(1, sizeof *pw);
899 pw->local_rconn = local_rconn;
900 pw->remote_rconn = remote_rconn;
901 pw->last_feature_request = TIME_MIN;
902 for (i = 0; i < OFPP_MAX; i++) {
903 pw->ports[i].port_no = htons(OFPP_NONE);
905 port_watcher_register_callback(pw, log_port_status, NULL);
906 return make_hook(port_watcher_local_packet_cb,
907 port_watcher_remote_packet_cb,
908 port_watcher_periodic_cb, NULL, pw);
911 /* Spanning tree protocol. */
913 /* Extra time, in seconds, at boot before going into fail-open, to give the
914 * spanning tree protocol time to figure out the network layout. */
915 #define STP_EXTRA_BOOT_TIME 30
919 struct port_watcher *pw;
920 struct rconn *local_rconn;
921 struct rconn *remote_rconn;
922 uint8_t dpid[ETH_ADDR_LEN];
923 long long int last_tick_256ths;
928 stp_local_packet_cb(struct relay *r, void *stp_)
930 struct stp_data *stp = stp_;
931 struct ofp_packet_in *opi;
932 struct eth_header *eth;
933 struct llc_header *llc;
934 struct buffer payload;
938 if (!get_ofp_packet_eth_header(r, &opi, ð)
939 || !eth_addr_equals(eth->eth_dst, stp_eth_addr)) {
943 port_no = ntohs(opi->in_port);
944 if (port_no >= STP_MAX_PORTS) {
945 /* STP only supports 255 ports. */
948 if (port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP) {
949 /* We're not doing STP on this port. */
953 if (opi->reason == OFPR_ACTION) {
954 /* The controller set up a flow for this, so we won't intercept it. */
958 get_ofp_packet_payload(opi, &payload);
959 flow_extract(&payload, port_no, &flow);
960 if (flow.dl_type != htons(OFP_DL_TYPE_NOT_ETH_TYPE)) {
961 VLOG_DBG("non-LLC frame received on STP multicast address");
964 llc = buffer_at_assert(&payload, sizeof *eth, sizeof *llc);
965 if (llc->llc_dsap != STP_LLC_DSAP) {
966 VLOG_DBG("bad DSAP 0x%02"PRIx8" received on STP multicast address",
971 /* Trim off padding on payload. */
972 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
973 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
975 if (buffer_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
976 struct stp_port *p = stp_get_port(stp->stp, port_no);
977 stp_received_bpdu(p, payload.data, payload.size);
986 return time_msec() * 256 / 1000;
990 stp_periodic_cb(void *stp_)
992 struct stp_data *stp = stp_;
993 long long int now_256ths = time_256ths();
994 long long int elapsed_256ths = now_256ths - stp->last_tick_256ths;
997 if (!port_watcher_is_ready(stp->pw)) {
998 /* Can't start STP until we know port flags, because port flags can
1002 if (elapsed_256ths <= 0) {
1006 stp_tick(stp->stp, MIN(INT_MAX, elapsed_256ths));
1007 stp->last_tick_256ths = now_256ths;
1009 while (stp_get_changed_port(stp->stp, &p)) {
1010 int port_no = stp_port_no(p);
1011 enum stp_state state = stp_port_get_state(p);
1013 if (state != STP_DISABLED) {
1014 VLOG_WARN("STP: Port %d entered %s state",
1015 port_no, stp_state_name(state));
1017 if (!(port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP)) {
1021 flags = OFPPFL_STP_LISTEN;
1024 flags = OFPPFL_STP_LEARN;
1027 case STP_FORWARDING:
1028 flags = OFPPFL_STP_FORWARD;
1031 flags = OFPPFL_STP_BLOCK;
1034 VLOG_DBG_RL(&vrl, "STP: Port %d has bad state %x",
1036 flags = OFPPFL_STP_FORWARD;
1039 if (!stp_forward_in_state(state)) {
1040 flags |= OFPPFL_NO_FLOOD;
1042 port_watcher_set_flags(stp->pw, port_no, flags,
1043 OFPPFL_STP_MASK | OFPPFL_NO_FLOOD);
1045 /* We don't own those flags. */
1051 stp_wait_cb(void *stp_ UNUSED)
1053 poll_timer_wait(1000);
1057 send_bpdu(const void *bpdu, size_t bpdu_size, int port_no, void *stp_)
1059 struct stp_data *stp = stp_;
1060 struct eth_header *eth;
1061 struct llc_header *llc;
1062 struct buffer pkt, *opo;
1064 /* Packet skeleton. */
1065 buffer_init(&pkt, ETH_HEADER_LEN + LLC_HEADER_LEN + bpdu_size);
1066 eth = buffer_put_uninit(&pkt, sizeof *eth);
1067 llc = buffer_put_uninit(&pkt, sizeof *llc);
1068 buffer_put(&pkt, bpdu, bpdu_size);
1071 memcpy(eth->eth_dst, stp_eth_addr, ETH_ADDR_LEN);
1072 memcpy(eth->eth_src, stp->pw->ports[port_no].hw_addr, ETH_ADDR_LEN);
1073 eth->eth_type = htons(pkt.size - ETH_HEADER_LEN);
1076 llc->llc_dsap = STP_LLC_DSAP;
1077 llc->llc_ssap = STP_LLC_SSAP;
1078 llc->llc_cntl = STP_LLC_CNTL;
1080 opo = make_unbuffered_packet_out(&pkt, OFPP_NONE, port_no);
1081 buffer_uninit(&pkt);
1082 rconn_send_with_limit(stp->local_rconn, opo, &stp->n_txq, OFPP_MAX);
1086 stp_port_watcher_cb(uint16_t port_no,
1087 const struct ofp_phy_port *old,
1088 const struct ofp_phy_port *new,
1091 struct stp_data *stp = stp_;
1094 /* STP only supports a maximum of 255 ports, one less than OpenFlow. We
1095 * don't support STP on OFPP_LOCAL, either. */
1096 if (port_no >= STP_MAX_PORTS) {
1100 p = stp_get_port(stp->stp, port_no);
1101 if (new->port_no == htons(OFPP_NONE)
1102 || new->flags & htonl(OFPPFL_NO_STP)) {
1103 stp_port_disable(p);
1106 stp_port_set_speed(p, new->speed);
1111 stp_hook_create(const struct settings *s, struct port_watcher *pw,
1112 struct rconn *local, struct rconn *remote)
1114 uint8_t dpid[ETH_ADDR_LEN];
1115 struct netdev *netdev;
1116 struct stp_data *stp;
1119 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1121 fatal(retval, "Could not open %s device", s->of_name);
1123 memcpy(dpid, netdev_get_etheraddr(netdev), ETH_ADDR_LEN);
1124 netdev_close(netdev);
1126 stp = xcalloc(1, sizeof *stp);
1127 stp->stp = stp_create("stp", eth_addr_to_uint64(dpid), send_bpdu, stp);
1129 memcpy(stp->dpid, dpid, ETH_ADDR_LEN);
1130 stp->local_rconn = local;
1131 stp->remote_rconn = remote;
1132 stp->last_tick_256ths = time_256ths();
1134 port_watcher_register_callback(pw, stp_port_watcher_cb, stp);
1135 return make_hook(stp_local_packet_cb, NULL,
1136 stp_periodic_cb, stp_wait_cb, stp);
1139 /* In-band control. */
1141 struct in_band_data {
1142 const struct settings *s;
1143 struct mac_learning *ml;
1144 struct netdev *of_device;
1145 struct rconn *controller;
1146 uint8_t mac[ETH_ADDR_LEN];
1151 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct buffer *b)
1153 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
1156 static const uint8_t *
1157 get_controller_mac(struct in_band_data *in_band)
1159 static uint32_t ip, last_nonzero_ip;
1160 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
1161 static time_t next_refresh = 0;
1163 uint32_t last_ip = ip;
1165 time_t now = time_now();
1167 ip = rconn_get_ip(in_band->controller);
1168 if (last_ip != ip || !next_refresh || now >= next_refresh) {
1171 /* Look up MAC address. */
1172 memset(mac, 0, sizeof mac);
1174 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
1176 VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s",
1177 IP_ARGS(&ip), strerror(retval));
1180 have_mac = !eth_addr_is_zero(mac);
1182 /* Log changes in IP, MAC addresses. */
1183 if (ip && ip != last_nonzero_ip) {
1184 VLOG_DBG("controller IP address changed from "IP_FMT
1185 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
1186 last_nonzero_ip = ip;
1188 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
1189 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
1191 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
1192 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
1195 /* Schedule next refresh.
1197 * If we have an IP address but not a MAC address, then refresh
1198 * quickly, since we probably will get a MAC address soon (via ARP).
1199 * Otherwise, we can afford to wait a little while. */
1200 next_refresh = now + (!ip || have_mac ? 10 : 1);
1202 return !eth_addr_is_zero(mac) ? mac : NULL;
1206 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
1207 struct in_band_data *in_band)
1209 const uint8_t *mac = get_controller_mac(in_band);
1210 return mac && eth_addr_equals(mac, dl_addr);
1214 in_band_learn_mac(struct in_band_data *in_band,
1215 uint16_t in_port, const uint8_t src_mac[ETH_ADDR_LEN])
1217 if (mac_learning_learn(in_band->ml, src_mac, in_port)) {
1218 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
1219 ETH_ADDR_ARGS(src_mac), in_port);
1224 in_band_local_packet_cb(struct relay *r, void *in_band_)
1226 struct in_band_data *in_band = in_band_;
1227 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
1228 struct ofp_packet_in *opi;
1229 struct eth_header *eth;
1230 struct buffer payload;
1235 if (!get_ofp_packet_eth_header(r, &opi, ð)) {
1238 in_port = ntohs(opi->in_port);
1240 /* Deal with local stuff. */
1241 if (in_port == OFPP_LOCAL) {
1242 /* Sent by secure channel. */
1243 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1244 } else if (eth_addr_equals(eth->eth_dst, in_band->mac)) {
1245 /* Sent to secure channel. */
1246 out_port = OFPP_LOCAL;
1247 in_band_learn_mac(in_band, in_port, eth->eth_src);
1248 } else if (eth->eth_type == htons(ETH_TYPE_ARP)
1249 && eth_addr_is_broadcast(eth->eth_dst)
1250 && is_controller_mac(eth->eth_src, in_band)) {
1251 /* ARP sent by controller. */
1252 out_port = OFPP_FLOOD;
1253 } else if (is_controller_mac(eth->eth_dst, in_band)
1254 || is_controller_mac(eth->eth_src, in_band)) {
1255 /* Traffic to or from controller. Switch it by hand. */
1256 in_band_learn_mac(in_band, in_port, eth->eth_src);
1257 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1259 const uint8_t *controller_mac;
1260 controller_mac = get_controller_mac(in_band);
1261 if (eth->eth_type == htons(ETH_TYPE_ARP)
1262 && eth_addr_is_broadcast(eth->eth_dst)
1263 && is_controller_mac(eth->eth_src, in_band)) {
1264 /* ARP sent by controller. */
1265 out_port = OFPP_FLOOD;
1266 } else if (is_controller_mac(eth->eth_dst, in_band)
1267 && in_port == mac_learning_lookup(in_band->ml,
1269 /* Drop controller traffic that arrives on the controller port. */
1276 get_ofp_packet_payload(opi, &payload);
1277 flow_extract(&payload, in_port, &flow);
1278 if (in_port == out_port) {
1279 /* The input and output port match. Set up a flow to drop packets. */
1280 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
1281 in_band->s->max_idle, 0));
1282 } else if (out_port != OFPP_FLOOD) {
1283 /* The output port is known, so add a new flow. */
1284 queue_tx(rc, in_band,
1285 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
1286 out_port, in_band->s->max_idle));
1288 /* If the switch didn't buffer the packet, we need to send a copy. */
1289 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1290 queue_tx(rc, in_band,
1291 make_unbuffered_packet_out(&payload, in_port, out_port));
1294 /* We don't know that MAC. Send along the packet without setting up a
1297 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1298 b = make_unbuffered_packet_out(&payload, in_port, out_port);
1300 b = make_buffered_packet_out(ntohl(opi->buffer_id),
1303 queue_tx(rc, in_band, b);
1309 in_band_status_cb(struct status_reply *sr, void *in_band_)
1311 struct in_band_data *in_band = in_band_;
1312 struct in_addr local_ip;
1313 uint32_t controller_ip;
1314 const uint8_t *controller_mac;
1316 if (netdev_get_in4(in_band->of_device, &local_ip)) {
1317 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
1319 status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
1320 ETH_ADDR_ARGS(in_band->mac));
1322 controller_ip = rconn_get_ip(in_band->controller);
1323 if (controller_ip) {
1324 status_reply_put(sr, "controller-ip="IP_FMT,
1325 IP_ARGS(&controller_ip));
1327 controller_mac = get_controller_mac(in_band);
1328 if (controller_mac) {
1329 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
1330 ETH_ADDR_ARGS(controller_mac));
1335 get_ofp_packet_payload(struct ofp_packet_in *opi, struct buffer *payload)
1337 payload->data = opi->data;
1338 payload->size = ntohs(opi->header.length) - offsetof(struct ofp_packet_in,
1343 in_band_hook_create(const struct settings *s, struct switch_status *ss,
1344 struct rconn *remote)
1346 struct in_band_data *in_band;
1349 in_band = xcalloc(1, sizeof *in_band);
1351 in_band->ml = mac_learning_create();
1352 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE,
1353 &in_band->of_device);
1355 fatal(retval, "Could not open %s device", s->of_name);
1357 memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device),
1359 in_band->controller = remote;
1360 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
1361 return make_hook(in_band_local_packet_cb, NULL, NULL, NULL, in_band);
1364 /* Fail open support. */
1366 struct fail_open_data {
1367 const struct settings *s;
1368 struct rconn *local_rconn;
1369 struct rconn *remote_rconn;
1370 struct lswitch *lswitch;
1371 int last_disconn_secs;
1372 time_t boot_deadline;
1375 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
1377 fail_open_periodic_cb(void *fail_open_)
1379 struct fail_open_data *fail_open = fail_open_;
1383 if (time_now() < fail_open->boot_deadline) {
1386 disconn_secs = rconn_disconnected_duration(fail_open->remote_rconn);
1387 open = disconn_secs >= fail_open->s->probe_interval * 3;
1388 if (open != (fail_open->lswitch != NULL)) {
1390 VLOG_WARN("No longer in fail-open mode");
1391 lswitch_destroy(fail_open->lswitch);
1392 fail_open->lswitch = NULL;
1394 VLOG_WARN("Could not connect to controller for %d seconds, "
1395 "failing open", disconn_secs);
1396 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
1397 fail_open->s->max_idle);
1398 fail_open->last_disconn_secs = disconn_secs;
1400 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
1401 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
1402 "from controller", disconn_secs);
1403 fail_open->last_disconn_secs = disconn_secs;
1408 fail_open_local_packet_cb(struct relay *r, void *fail_open_)
1410 struct fail_open_data *fail_open = fail_open_;
1411 if (!fail_open->lswitch) {
1414 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
1415 r->halves[HALF_LOCAL].rxbuf);
1416 rconn_run(fail_open->local_rconn);
1422 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
1424 struct fail_open_data *fail_open = fail_open_;
1425 const struct settings *s = fail_open->s;
1426 int trigger_duration = s->probe_interval * 3;
1427 int cur_duration = rconn_disconnected_duration(fail_open->remote_rconn);
1429 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
1430 status_reply_put(sr, "current-duration=%d", cur_duration);
1431 status_reply_put(sr, "triggered=%s",
1432 cur_duration >= trigger_duration ? "true" : "false");
1433 status_reply_put(sr, "max-idle=%d", s->max_idle);
1437 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
1438 struct rconn *local_rconn, struct rconn *remote_rconn)
1440 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
1442 fail_open->local_rconn = local_rconn;
1443 fail_open->remote_rconn = remote_rconn;
1444 fail_open->lswitch = NULL;
1445 fail_open->boot_deadline = time_now() + s->probe_interval * 3;
1446 fail_open->boot_deadline += STP_EXTRA_BOOT_TIME;
1447 switch_status_register_category(ss, "fail-open",
1448 fail_open_status_cb, fail_open);
1449 return make_hook(fail_open_local_packet_cb, NULL,
1450 fail_open_periodic_cb, NULL, fail_open);
1453 struct rate_limiter {
1454 const struct settings *s;
1455 struct rconn *remote_rconn;
1457 /* One queue per physical port. */
1458 struct queue queues[OFPP_MAX];
1459 int n_queued; /* Sum over queues[*].n. */
1460 int next_tx_port; /* Next port to check in round-robin. */
1464 * It costs 1000 tokens to send a single packet_in message. A single token
1465 * per message would be more straightforward, but this choice lets us avoid
1466 * round-off error in refill_bucket()'s calculation of how many tokens to
1467 * add to the bucket, since no division step is needed. */
1468 long long int last_fill; /* Time at which we last added tokens. */
1469 int tokens; /* Current number of tokens. */
1471 /* Transmission queue. */
1472 int n_txq; /* No. of packets waiting in rconn for tx. */
1474 /* Statistics reporting. */
1475 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
1476 unsigned long long n_limited; /* # queued for rate limiting. */
1477 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
1478 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
1481 /* Drop a packet from the longest queue in 'rl'. */
1483 drop_packet(struct rate_limiter *rl)
1485 struct queue *longest; /* Queue currently selected as longest. */
1486 int n_longest; /* # of queues of same length as 'longest'. */
1489 longest = &rl->queues[0];
1491 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
1492 if (longest->n < q->n) {
1495 } else if (longest->n == q->n) {
1498 /* Randomly select one of the longest queues, with a uniform
1499 * distribution (Knuth algorithm 3.4.2R). */
1500 if (!random_range(n_longest)) {
1506 /* FIXME: do we want to pop the tail instead? */
1507 buffer_delete(queue_pop_head(longest));
1511 /* Remove and return the next packet to transmit (in round-robin order). */
1512 static struct buffer *
1513 dequeue_packet(struct rate_limiter *rl)
1517 for (i = 0; i < OFPP_MAX; i++) {
1518 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
1519 struct queue *q = &rl->queues[port];
1521 rl->next_tx_port = (port + 1) % OFPP_MAX;
1523 return queue_pop_head(q);
1529 /* Add tokens to the bucket based on elapsed time. */
1531 refill_bucket(struct rate_limiter *rl)
1533 const struct settings *s = rl->s;
1534 long long int now = time_msec();
1535 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
1536 if (tokens >= 1000) {
1537 rl->last_fill = now;
1538 rl->tokens = MIN(tokens, s->burst_limit * 1000);
1542 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
1543 * true if successful, false otherwise. (In the latter case no tokens are
1546 get_token(struct rate_limiter *rl)
1548 if (rl->tokens >= 1000) {
1557 rate_limit_local_packet_cb(struct relay *r, void *rl_)
1559 struct rate_limiter *rl = rl_;
1560 const struct settings *s = rl->s;
1561 struct ofp_packet_in *opi;
1563 opi = get_ofp_packet_in(r);
1568 if (!rl->n_queued && get_token(rl)) {
1569 /* In the common case where we are not constrained by the rate limit,
1570 * let the packet take the normal path. */
1574 /* Otherwise queue it up for the periodic callback to drain out. */
1575 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
1576 int port = ntohs(opi->in_port) % OFPP_MAX;
1577 if (rl->n_queued >= s->burst_limit) {
1580 queue_push_tail(&rl->queues[port], buffer_clone(msg));
1588 rate_limit_status_cb(struct status_reply *sr, void *rl_)
1590 struct rate_limiter *rl = rl_;
1592 status_reply_put(sr, "normal=%llu", rl->n_normal);
1593 status_reply_put(sr, "limited=%llu", rl->n_limited);
1594 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
1595 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
1599 rate_limit_periodic_cb(void *rl_)
1601 struct rate_limiter *rl = rl_;
1604 /* Drain some packets out of the bucket if possible, but limit the number
1605 * of iterations to allow other code to get work done too. */
1607 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
1608 /* Use a small, arbitrary limit for the amount of queuing to do here,
1609 * because the TCP connection is responsible for buffering and there is
1610 * no point in trying to transmit faster than the TCP connection can
1612 struct buffer *b = dequeue_packet(rl);
1613 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
1620 rate_limit_wait_cb(void *rl_)
1622 struct rate_limiter *rl = rl_;
1624 if (rl->tokens >= 1000) {
1625 /* We can transmit more packets as soon as we're called again. */
1626 poll_immediate_wake();
1628 /* We have to wait for the bucket to re-fill. We could calculate
1629 * the exact amount of time here for increased smoothness. */
1630 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
1636 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
1637 struct rconn *local, struct rconn *remote)
1639 struct rate_limiter *rl;
1642 rl = xcalloc(1, sizeof *rl);
1644 rl->remote_rconn = remote;
1645 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
1646 queue_init(&rl->queues[i]);
1648 rl->last_fill = time_msec();
1649 rl->tokens = s->rate_limit * 100;
1650 switch_status_register_category(ss, "rate-limit",
1651 rate_limit_status_cb, rl);
1652 return make_hook(rate_limit_local_packet_cb, NULL, rate_limit_periodic_cb,
1653 rate_limit_wait_cb, rl);
1656 /* OFPST_SWITCH statistics. */
1658 struct switch_status_category {
1660 void (*cb)(struct status_reply *, void *aux);
1664 struct switch_status {
1665 const struct settings *s;
1667 struct switch_status_category categories[8];
1671 struct status_reply {
1672 struct switch_status_category *category;
1678 switch_status_remote_packet_cb(struct relay *r, void *ss_)
1680 struct switch_status *ss = ss_;
1681 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
1682 struct buffer *msg = r->halves[HALF_REMOTE].rxbuf;
1683 struct switch_status_category *c;
1684 struct nicira_header *request;
1685 struct nicira_header *reply;
1686 struct status_reply sr;
1690 if (msg->size < sizeof(struct nicira_header)) {
1693 request = msg->data;
1694 if (request->header.type != OFPT_VENDOR
1695 || request->vendor_id != htonl(NX_VENDOR_ID)
1696 || request->subtype != htonl(NXT_STATUS_REQUEST)) {
1700 sr.request.string = (void *) (request + 1);
1701 sr.request.length = msg->size - sizeof *request;
1702 ds_init(&sr.output);
1703 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
1704 if (!memcmp(c->name, sr.request.string,
1705 MIN(strlen(c->name), sr.request.length))) {
1710 reply = make_openflow_xid(sizeof *reply + sr.output.length,
1711 OFPT_VENDOR, request->header.xid, &b);
1712 reply->vendor_id = htonl(NX_VENDOR_ID);
1713 reply->subtype = htonl(NXT_STATUS_REPLY);
1714 memcpy(reply + 1, sr.output.string, sr.output.length);
1715 retval = rconn_send(rc, b, NULL);
1716 if (retval && retval != EAGAIN) {
1717 VLOG_WARN("send failed (%s)", strerror(retval));
1719 ds_destroy(&sr.output);
1724 rconn_status_cb(struct status_reply *sr, void *rconn_)
1726 struct rconn *rconn = rconn_;
1727 time_t now = time_now();
1729 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
1730 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
1731 status_reply_put(sr, "backoff=%d", rconn_get_backoff(rconn));
1732 status_reply_put(sr, "is-connected=%s",
1733 rconn_is_connected(rconn) ? "true" : "false");
1734 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
1735 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
1736 status_reply_put(sr, "attempted-connections=%u",
1737 rconn_get_attempted_connections(rconn));
1738 status_reply_put(sr, "successful-connections=%u",
1739 rconn_get_successful_connections(rconn));
1740 status_reply_put(sr, "last-connection=%ld",
1741 (long int) (now - rconn_get_last_connection(rconn)));
1742 status_reply_put(sr, "time-connected=%lu",
1743 rconn_get_total_time_connected(rconn));
1744 status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(rconn));
1748 config_status_cb(struct status_reply *sr, void *s_)
1750 const struct settings *s = s_;
1753 for (i = 0; i < s->n_listeners; i++) {
1754 status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
1756 if (s->probe_interval) {
1757 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
1759 if (s->max_backoff) {
1760 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
1765 switch_status_cb(struct status_reply *sr, void *ss_)
1767 struct switch_status *ss = ss_;
1768 time_t now = time_now();
1770 status_reply_put(sr, "now=%ld", (long int) now);
1771 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
1772 status_reply_put(sr, "pid=%ld", (long int) getpid());
1776 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
1778 struct switch_status *ss = xcalloc(1, sizeof *ss);
1780 ss->booted = time_now();
1781 switch_status_register_category(ss, "config",
1782 config_status_cb, (void *) s);
1783 switch_status_register_category(ss, "switch", switch_status_cb, ss);
1785 return make_hook(NULL, switch_status_remote_packet_cb, NULL, NULL, ss);
1789 switch_status_register_category(struct switch_status *ss,
1790 const char *category,
1791 void (*cb)(struct status_reply *,
1795 struct switch_status_category *c;
1796 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
1797 c = &ss->categories[ss->n_categories++];
1800 c->name = xstrdup(category);
1804 status_reply_put(struct status_reply *sr, const char *content, ...)
1806 size_t old_length = sr->output.length;
1810 /* Append the status reply to the output. */
1811 ds_put_format(&sr->output, "%s.", sr->category->name);
1812 va_start(args, content);
1813 ds_put_format_valist(&sr->output, content, args);
1815 if (ds_last(&sr->output) != '\n') {
1816 ds_put_char(&sr->output, '\n');
1819 /* Drop what we just added if it doesn't match the request. */
1820 added = sr->output.length - old_length;
1821 if (added < sr->request.length
1822 || memcmp(&sr->output.string[old_length],
1823 sr->request.string, sr->request.length)) {
1824 ds_truncate(&sr->output, old_length);
1829 /* Controller discovery. */
1833 const struct settings *s;
1834 struct dhclient *dhcp;
1839 discovery_status_cb(struct status_reply *sr, void *d_)
1841 struct discovery *d = d_;
1843 status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re);
1844 status_reply_put(sr, "n-changes=%d", d->n_changes);
1845 status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp));
1846 status_reply_put(sr, "state-elapsed=%u",
1847 dhclient_get_state_elapsed(d->dhcp));
1848 if (dhclient_is_bound(d->dhcp)) {
1849 uint32_t ip = dhclient_get_ip(d->dhcp);
1850 uint32_t netmask = dhclient_get_netmask(d->dhcp);
1851 uint32_t router = dhclient_get_router(d->dhcp);
1853 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
1854 uint32_t dns_server;
1858 status_reply_put(sr, "ip="IP_FMT, IP_ARGS(&ip));
1859 status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask));
1861 status_reply_put(sr, "router="IP_FMT, IP_ARGS(&router));
1864 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i, &dns_server);
1866 status_reply_put(sr, "dns%d="IP_FMT, i, IP_ARGS(&dns_server));
1869 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
1871 status_reply_put(sr, "domain=%s", domain_name);
1875 status_reply_put(sr, "lease-remaining=%u",
1876 dhclient_get_lease_remaining(d->dhcp));
1880 static struct discovery *
1881 discovery_init(const struct settings *s, struct switch_status *ss)
1883 struct netdev *netdev;
1884 struct discovery *d;
1885 struct dhclient *dhcp;
1888 /* Bring ofX network device up. */
1889 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1891 fatal(retval, "Could not open %s device", s->of_name);
1893 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
1895 fatal(retval, "Could not bring %s device up", s->of_name);
1897 netdev_close(netdev);
1899 /* Initialize DHCP client. */
1900 retval = dhclient_create(s->of_name, modify_dhcp_request,
1901 validate_dhcp_offer, (void *) s, &dhcp);
1903 fatal(retval, "Failed to initialize DHCP client");
1905 dhclient_init(dhcp, 0);
1907 d = xmalloc(sizeof *d);
1912 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
1918 discovery_question_connectivity(struct discovery *d)
1920 dhclient_force_renew(d->dhcp, 15);
1924 discovery_run(struct discovery *d, char **controller_name)
1926 dhclient_run(d->dhcp);
1927 if (!dhclient_changed(d->dhcp)) {
1931 dhclient_configure_netdev(d->dhcp);
1932 if (d->s->update_resolv_conf) {
1933 dhclient_update_resolv_conf(d->dhcp);
1936 if (dhclient_is_bound(d->dhcp)) {
1937 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
1938 DHCP_CODE_OFP_CONTROLLER_VCONN);
1939 VLOG_WARN("%s: discovered controller", *controller_name);
1942 *controller_name = NULL;
1944 VLOG_WARN("discovered controller no longer available");
1952 discovery_wait(struct discovery *d)
1954 dhclient_wait(d->dhcp);
1958 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
1960 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
1964 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
1966 const struct settings *s = s_;
1970 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
1972 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
1975 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
1977 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
1978 s->accept_controller_re);
1984 /* User interface. */
1987 parse_options(int argc, char *argv[], struct settings *s)
1990 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
1992 OPT_INACTIVITY_PROBE,
1998 static struct option long_options[] = {
1999 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
2000 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
2001 {"fail", required_argument, 0, 'F'},
2002 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
2003 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
2004 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
2005 {"listen", required_argument, 0, 'l'},
2006 {"monitor", required_argument, 0, 'm'},
2007 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
2008 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
2009 {"detach", no_argument, 0, 'D'},
2010 {"force", no_argument, 0, 'f'},
2011 {"pidfile", optional_argument, 0, 'P'},
2012 {"verbose", optional_argument, 0, 'v'},
2013 {"help", no_argument, 0, 'h'},
2014 {"version", no_argument, 0, 'V'},
2015 VCONN_SSL_LONG_OPTIONS
2018 char *short_options = long_options_to_short_options(long_options);
2019 char *accept_re = NULL;
2022 /* Set defaults that we can figure out before parsing options. */
2024 s->monitor_name = NULL;
2025 s->fail_mode = FAIL_OPEN;
2027 s->probe_interval = 15;
2028 s->max_backoff = 15;
2029 s->update_resolv_conf = true;
2035 c = getopt_long(argc, argv, short_options, long_options, NULL);
2041 case OPT_ACCEPT_VCONN:
2042 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
2045 case OPT_NO_RESOLV_CONF:
2046 s->update_resolv_conf = false;
2050 if (!strcmp(optarg, "open")) {
2051 s->fail_mode = FAIL_OPEN;
2052 } else if (!strcmp(optarg, "closed")) {
2053 s->fail_mode = FAIL_CLOSED;
2056 "-f or --fail argument must be \"open\" or \"closed\"");
2060 case OPT_INACTIVITY_PROBE:
2061 s->probe_interval = atoi(optarg);
2062 if (s->probe_interval < 5) {
2063 fatal(0, "--inactivity-probe argument must be at least 5");
2068 if (!strcmp(optarg, "permanent")) {
2069 s->max_idle = OFP_FLOW_PERMANENT;
2071 s->max_idle = atoi(optarg);
2072 if (s->max_idle < 1 || s->max_idle > 65535) {
2073 fatal(0, "--max-idle argument must be between 1 and "
2074 "65535 or the word 'permanent'");
2079 case OPT_MAX_BACKOFF:
2080 s->max_backoff = atoi(optarg);
2081 if (s->max_backoff < 1) {
2082 fatal(0, "--max-backoff argument must be at least 1");
2083 } else if (s->max_backoff > 3600) {
2084 s->max_backoff = 3600;
2088 case OPT_RATE_LIMIT:
2090 s->rate_limit = atoi(optarg);
2091 if (s->rate_limit < 1) {
2092 fatal(0, "--rate-limit argument must be at least 1");
2095 s->rate_limit = 1000;
2099 case OPT_BURST_LIMIT:
2100 s->burst_limit = atoi(optarg);
2101 if (s->burst_limit < 1) {
2102 fatal(0, "--burst-limit argument must be at least 1");
2111 set_pidfile(optarg);
2115 ignore_existing_pidfile();
2119 if (s->n_listeners >= MAX_MGMT) {
2120 fatal(0, "-l or --listen may be specified at most %d times",
2123 s->listener_names[s->n_listeners++] = optarg;
2127 if (s->monitor_name) {
2128 fatal(0, "-m or --monitor may only be specified once");
2130 s->monitor_name = optarg;
2137 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
2141 vlog_set_verbosity(optarg);
2144 VCONN_SSL_OPTION_HANDLERS
2153 free(short_options);
2157 if (argc < 1 || argc > 2) {
2158 fatal(0, "need one or two non-option arguments; use --help for usage");
2161 /* Local and remote vconns. */
2162 s->nl_name = argv[0];
2163 if (strncmp(s->nl_name, "nl:", 3)
2164 || strlen(s->nl_name) < 4
2165 || s->nl_name[strspn(s->nl_name + 3, "0123456789") + 3]) {
2166 fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"", s->nl_name);
2168 s->of_name = xasprintf("of%s", s->nl_name + 3);
2169 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
2171 /* Set accept_controller_regex. */
2173 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
2175 retval = regcomp(&s->accept_controller_regex, accept_re,
2176 REG_NOSUB | REG_EXTENDED);
2178 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
2179 char *buffer = xmalloc(length);
2180 regerror(retval, &s->accept_controller_regex, buffer, length);
2181 fatal(0, "%s: %s", accept_re, buffer);
2183 s->accept_controller_re = accept_re;
2185 /* Mode of operation. */
2186 s->discovery = s->controller_name == NULL;
2190 enum netdev_flags flags;
2191 struct netdev *netdev;
2193 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
2195 fatal(retval, "Could not open %s device", s->of_name);
2198 retval = netdev_get_flags(netdev, &flags);
2200 fatal(retval, "Could not get flags for %s device", s->of_name);
2203 s->in_band = (flags & NETDEV_UP) != 0;
2204 if (s->in_band && netdev_get_in6(netdev, NULL)) {
2205 VLOG_WARN("Ignoring IPv6 address on %s device: IPv6 not supported",
2209 netdev_close(netdev);
2212 /* Rate limiting. */
2213 if (s->rate_limit) {
2214 if (s->rate_limit < 100) {
2215 VLOG_WARN("Rate limit set to unusually low value %d",
2218 if (!s->burst_limit) {
2219 s->burst_limit = s->rate_limit / 4;
2221 s->burst_limit = MAX(s->burst_limit, 1);
2222 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
2229 printf("%s: secure channel, a relay for OpenFlow messages.\n"
2230 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
2231 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
2232 "CONTROLLER is an active OpenFlow connection method; if it is\n"
2233 "omitted, then secchan performs controller discovery.\n",
2234 program_name, program_name);
2235 vconn_usage(true, true);
2236 printf("\nController discovery options:\n"
2237 " --accept-vconn=REGEX accept matching discovered controllers\n"
2238 " --no-resolv-conf do not update /etc/resolv.conf\n"
2239 "\nNetworking options:\n"
2240 " -F, --fail=open|closed when controller connection fails:\n"
2241 " closed: drop all packets\n"
2242 " open (default): act as learning switch\n"
2243 " --inactivity-probe=SECS time between inactivity probes\n"
2244 " --max-idle=SECS max idle for flows set up by secchan\n"
2245 " --max-backoff=SECS max time between controller connection\n"
2246 " attempts (default: 15 seconds)\n"
2247 " -l, --listen=METHOD allow management connections on METHOD\n"
2248 " (a passive OpenFlow connection method)\n"
2249 " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n"
2250 " (a passive OpenFlow connection method)\n"
2251 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
2252 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
2253 " --burst-limit=BURST limit on packet credit for idle time\n"
2254 "\nOther options:\n"
2255 " -D, --detach run in background as daemon\n"
2256 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
2257 " -f, --force with -P, start even if already running\n"
2258 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
2259 " -v, --verbose set maximum verbosity level\n"
2260 " -h, --help display this help message\n"
2261 " -V, --version display version information\n",