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 vconn *open_passive_vconn(const char *name);
144 static struct vconn *accept_vconn(struct vconn *vconn);
146 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
148 static struct relay *relay_accept(const struct settings *, struct vconn *);
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 vconn *monitor;
225 struct vconn *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 vconn_accept_wait(listeners[i]);
356 vconn_accept_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 vconn *
373 open_passive_vconn(const char *name)
378 retval = vconn_open(name, &vconn);
379 if (retval && retval != EAGAIN) {
380 fatal(retval, "opening %s", name);
382 if (!vconn_is_passive(vconn)) {
383 fatal(0, "%s is not a passive vconn", name);
388 static struct vconn *
389 accept_vconn(struct vconn *vconn)
394 retval = vconn_accept(vconn, &new);
395 if (retval && retval != EAGAIN) {
396 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
402 make_hook(bool (*local_packet_cb)(struct relay *, void *aux),
403 bool (*remote_packet_cb)(struct relay *, void *aux),
404 void (*periodic_cb)(void *aux),
405 void (*wait_cb)(void *aux),
409 h.packet_cb[HALF_LOCAL] = local_packet_cb;
410 h.packet_cb[HALF_REMOTE] = remote_packet_cb;
411 h.periodic_cb = periodic_cb;
417 static struct ofp_packet_in *
418 get_ofp_packet_in(struct relay *r)
420 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
421 struct ofp_header *oh = msg->data;
422 if (oh->type == OFPT_PACKET_IN) {
423 if (msg->size >= offsetof (struct ofp_packet_in, data)) {
426 VLOG_WARN("packet too short (%zu bytes) for packet_in",
434 get_ofp_packet_eth_header(struct relay *r, struct ofp_packet_in **opip,
435 struct eth_header **ethp)
437 const int min_len = offsetof(struct ofp_packet_in, data) + ETH_HEADER_LEN;
438 struct ofp_packet_in *opi = get_ofp_packet_in(r);
439 if (opi && ntohs(opi->header.length) >= min_len) {
441 *ethp = (void *) opi->data;
448 /* OpenFlow message relaying. */
450 static struct relay *
451 relay_accept(const struct settings *s, struct vconn *listen_vconn)
453 struct vconn *new_remote, *new_local;
454 char *nl_name_without_subscription;
455 struct rconn *r1, *r2;
458 new_remote = accept_vconn(listen_vconn);
463 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We
464 * only accept the former syntax in main().
466 * nl:123:0 opens a netlink connection to local datapath 123 without
467 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
469 nl_name_without_subscription = xasprintf("%s:0", s->nl_name);
470 retval = vconn_open(nl_name_without_subscription, &new_local);
472 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
473 nl_name_without_subscription, strerror(retval));
474 vconn_close(new_remote);
475 free(nl_name_without_subscription);
479 /* Create and return relay. */
480 r1 = rconn_create(0, 0);
481 rconn_connect_unreliably(r1, nl_name_without_subscription, new_local);
482 free(nl_name_without_subscription);
484 r2 = rconn_create(0, 0);
485 rconn_connect_unreliably(r2, "passive", new_remote);
487 return relay_create(r1, r2, true);
490 static struct relay *
491 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
493 struct relay *r = xcalloc(1, sizeof *r);
494 r->halves[HALF_LOCAL].rconn = local;
495 r->halves[HALF_REMOTE].rconn = remote;
496 r->is_mgmt_conn = is_mgmt_conn;
501 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
506 for (i = 0; i < 2; i++) {
507 rconn_run(r->halves[i].rconn);
510 /* Limit the number of iterations to prevent other tasks from starving. */
511 for (iteration = 0; iteration < 50; iteration++) {
512 bool progress = false;
513 for (i = 0; i < 2; i++) {
514 struct half *this = &r->halves[i];
515 struct half *peer = &r->halves[!i];
518 this->rxbuf = rconn_recv(this->rconn);
519 if (this->rxbuf && (i == HALF_REMOTE || !r->is_mgmt_conn)) {
520 const struct hook *h;
521 for (h = hooks; h < &hooks[n_hooks]; h++) {
522 if (h->packet_cb[i] && h->packet_cb[i](r, h->aux)) {
523 buffer_delete(this->rxbuf);
532 if (this->rxbuf && !this->n_txq) {
533 int retval = rconn_send(peer->rconn, this->rxbuf,
535 if (retval != EAGAIN) {
539 buffer_delete(this->rxbuf);
550 if (r->is_mgmt_conn) {
551 for (i = 0; i < 2; i++) {
552 struct half *this = &r->halves[i];
553 if (!rconn_is_alive(this->rconn)) {
562 relay_wait(struct relay *r)
566 for (i = 0; i < 2; i++) {
567 struct half *this = &r->halves[i];
569 rconn_run_wait(this->rconn);
571 rconn_recv_wait(this->rconn);
577 relay_destroy(struct relay *r)
581 list_remove(&r->node);
582 for (i = 0; i < 2; i++) {
583 struct half *this = &r->halves[i];
584 rconn_destroy(this->rconn);
585 buffer_delete(this->rxbuf);
590 /* Port status watcher. */
592 typedef void port_watcher_cb_func(uint16_t port_no,
593 const struct ofp_phy_port *old,
594 const struct ofp_phy_port *new,
597 struct port_watcher_cb {
598 port_watcher_cb_func *function;
602 struct port_watcher {
603 struct rconn *local_rconn;
604 struct rconn *remote_rconn;
605 struct ofp_phy_port ports[OFPP_MAX + 1];
606 time_t last_feature_request;
607 bool got_feature_reply;
609 struct port_watcher_cb cbs[2];
613 /* Returns the number of fields that differ from 'a' to 'b'. */
615 opp_differs(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
617 BUILD_ASSERT_DECL(sizeof *a == 36); /* Trips when we add or remove fields. */
618 return ((a->port_no != b->port_no)
619 + (memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) != 0)
620 + (memcmp(a->name, b->name, sizeof a->name) != 0)
621 + (a->flags != b->flags)
622 + (a->speed != b->speed)
623 + (a->features != b->features));
627 sanitize_opp(struct ofp_phy_port *opp)
631 for (i = 0; i < sizeof opp->name; i++) {
632 char c = opp->name[i];
633 if (c && (c < 0x20 || c > 0x7e)) {
637 opp->name[sizeof opp->name - 1] = '\0';
641 port_no_to_pw_idx(int port_no)
643 return (port_no < OFPP_MAX ? port_no
644 : port_no == OFPP_LOCAL ? OFPP_MAX
649 call_pw_callbacks(struct port_watcher *pw, int port_no,
650 const struct ofp_phy_port *old,
651 const struct ofp_phy_port *new)
653 if (opp_differs(old, new)) {
655 for (i = 0; i < pw->n_cbs; i++) {
656 pw->cbs[i].function(port_no, old, new, pw->cbs[i].aux);
662 update_phy_port(struct port_watcher *pw, struct ofp_phy_port *opp,
663 uint8_t reason, bool seen[OFPP_MAX + 1])
665 struct ofp_phy_port *pw_opp;
666 struct ofp_phy_port old;
670 port_no = ntohs(opp->port_no);
671 idx = port_no_to_pw_idx(port_no);
680 pw_opp = &pw->ports[idx];
682 if (reason == OFPPR_DELETE) {
683 memset(pw_opp, 0, sizeof *pw_opp);
684 pw_opp->port_no = htons(OFPP_NONE);
685 } else if (reason == OFPPR_MOD || reason == OFPPR_ADD) {
687 sanitize_opp(pw_opp);
689 call_pw_callbacks(pw, port_no, &old, pw_opp);
693 port_watcher_local_packet_cb(struct relay *r, void *pw_)
695 struct port_watcher *pw = pw_;
696 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
697 struct ofp_header *oh = msg->data;
699 if (oh->type == OFPT_FEATURES_REPLY
700 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
701 struct ofp_switch_features *osf = msg->data;
702 bool seen[ARRAY_SIZE(pw->ports)];
706 pw->got_feature_reply = true;
708 /* Update each port included in the message. */
709 memset(seen, 0, sizeof seen);
710 n_ports = ((msg->size - offsetof(struct ofp_switch_features, ports))
711 / sizeof *osf->ports);
712 for (i = 0; i < n_ports; i++) {
713 update_phy_port(pw, &osf->ports[i], OFPPR_MOD, seen);
716 /* Delete all the ports not included in the message. */
717 for (i = 0; i < ARRAY_SIZE(pw->ports); i++) {
719 update_phy_port(pw, &pw->ports[i], OFPPR_DELETE, NULL);
722 } else if (oh->type == OFPT_PORT_STATUS
723 && msg->size >= sizeof(struct ofp_port_status)) {
724 struct ofp_port_status *ops = msg->data;
725 update_phy_port(pw, &ops->desc, ops->reason, NULL);
731 port_watcher_remote_packet_cb(struct relay *r, void *pw_)
733 struct port_watcher *pw = pw_;
734 struct buffer *msg = r->halves[HALF_REMOTE].rxbuf;
735 struct ofp_header *oh = msg->data;
737 if (oh->type == OFPT_PORT_MOD
738 && msg->size >= sizeof(struct ofp_port_mod)) {
739 struct ofp_port_mod *opm = msg->data;
740 uint16_t port_no = ntohs(opm->desc.port_no);
741 int idx = port_no_to_pw_idx(port_no);
743 struct ofp_phy_port *pw_opp = &pw->ports[idx];
744 if (pw_opp->port_no != htons(OFPP_NONE)) {
745 struct ofp_phy_port old = *pw_opp;
746 pw_opp->flags = ((pw_opp->flags & ~opm->mask)
747 | (opm->desc.flags & opm->mask));
748 call_pw_callbacks(pw, port_no, &old, pw_opp);
756 port_watcher_periodic_cb(void *pw_)
758 struct port_watcher *pw = pw_;
760 if (!pw->got_feature_reply && time_now() >= pw->last_feature_request + 5) {
762 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
763 rconn_send_with_limit(pw->local_rconn, b, &pw->n_txq, 1);
764 pw->last_feature_request = time_now();
769 put_duplexes(struct ds *ds, const char *name, uint32_t features,
770 uint32_t hd_bit, uint32_t fd_bit)
772 if (features & (hd_bit | fd_bit)) {
773 ds_put_format(ds, " %s", name);
774 if (features & hd_bit) {
775 ds_put_cstr(ds, "(HD)");
777 if (features & fd_bit) {
778 ds_put_cstr(ds, "(FD)");
784 log_port_status(uint16_t port_no,
785 const struct ofp_phy_port *old,
786 const struct ofp_phy_port *new,
789 if (VLOG_IS_DBG_ENABLED()) {
790 bool was_enabled = old->port_no != htons(OFPP_NONE);
791 bool now_enabled = new->port_no != htons(OFPP_NONE);
792 uint32_t features = ntohl(new->features);
795 if (old->flags != new->flags && opp_differs(old, new) == 1) {
796 /* Don't care if only flags changed. */
801 ds_put_format(&ds, "\"%s\", "ETH_ADDR_FMT, new->name,
802 ETH_ADDR_ARGS(new->hw_addr));
803 if (ntohl(new->speed)) {
804 ds_put_format(&ds, ", speed %"PRIu32, ntohl(new->speed));
806 if (features & (OFPPF_10MB_HD | OFPPF_10MB_FD
807 | OFPPF_100MB_HD | OFPPF_100MB_FD
808 | OFPPF_1GB_HD | OFPPF_1GB_FD | OFPPF_10GB_FD)) {
809 ds_put_cstr(&ds, ", supports");
810 put_duplexes(&ds, "10M", features, OFPPF_10MB_HD, OFPPF_10MB_FD);
811 put_duplexes(&ds, "100M", features,
812 OFPPF_100MB_HD, OFPPF_100MB_FD);
813 put_duplexes(&ds, "1G", features, OFPPF_100MB_HD, OFPPF_100MB_FD);
814 if (features & OFPPF_10GB_FD) {
815 ds_put_cstr(&ds, " 10G");
818 if (was_enabled != now_enabled) {
820 VLOG_DBG("Port %d added: %s", port_no, ds_cstr(&ds));
822 VLOG_DBG("Port %d deleted", port_no);
825 VLOG_DBG("Port %d changed: %s", port_no, ds_cstr(&ds));
832 port_watcher_register_callback(struct port_watcher *pw,
833 port_watcher_cb_func *function,
836 assert(pw->n_cbs < ARRAY_SIZE(pw->cbs));
837 pw->cbs[pw->n_cbs].function = function;
838 pw->cbs[pw->n_cbs].aux = aux;
843 port_watcher_get_flags(const struct port_watcher *pw, int port_no)
845 int idx = port_no_to_pw_idx(port_no);
846 return idx >= 0 ? ntohl(pw->ports[idx].flags) : 0;
850 port_watcher_set_flags(struct port_watcher *pw,
851 int port_no, uint32_t flags, uint32_t mask)
853 struct ofp_phy_port old;
854 struct ofp_phy_port *p;
855 struct ofp_port_mod *opm;
856 struct ofp_port_status *ops;
860 idx = port_no_to_pw_idx(port_no);
866 if (!((ntohl(p->flags) ^ flags) & mask)) {
871 /* Update our idea of the flags. */
872 p->flags = htonl((ntohl(p->flags) & ~mask) | (flags & mask));
873 call_pw_callbacks(pw, port_no, &old, p);
875 /* Change the flags in the datapath. */
876 opm = make_openflow(sizeof *opm, OFPT_PORT_MOD, &b);
877 opm->mask = htonl(mask);
879 rconn_send(pw->local_rconn, b, NULL);
881 /* Notify the controller that the flags changed. */
882 ops = make_openflow(sizeof *ops, OFPT_PORT_STATUS, &b);
883 ops->reason = OFPPR_MOD;
885 rconn_send(pw->remote_rconn, b, NULL);
889 port_watcher_is_ready(const struct port_watcher *pw)
891 return pw->got_feature_reply;
895 port_watcher_create(struct rconn *local_rconn, struct rconn *remote_rconn,
896 struct port_watcher **pwp)
898 struct port_watcher *pw;
901 pw = *pwp = xcalloc(1, sizeof *pw);
902 pw->local_rconn = local_rconn;
903 pw->remote_rconn = remote_rconn;
904 pw->last_feature_request = TIME_MIN;
905 for (i = 0; i < OFPP_MAX; i++) {
906 pw->ports[i].port_no = htons(OFPP_NONE);
908 port_watcher_register_callback(pw, log_port_status, NULL);
909 return make_hook(port_watcher_local_packet_cb,
910 port_watcher_remote_packet_cb,
911 port_watcher_periodic_cb, NULL, pw);
914 /* Spanning tree protocol. */
916 /* Extra time, in seconds, at boot before going into fail-open, to give the
917 * spanning tree protocol time to figure out the network layout. */
918 #define STP_EXTRA_BOOT_TIME 30
922 struct port_watcher *pw;
923 struct rconn *local_rconn;
924 struct rconn *remote_rconn;
925 uint8_t dpid[ETH_ADDR_LEN];
926 long long int last_tick_256ths;
931 stp_local_packet_cb(struct relay *r, void *stp_)
933 struct stp_data *stp = stp_;
934 struct ofp_packet_in *opi;
935 struct eth_header *eth;
936 struct llc_header *llc;
937 struct buffer payload;
941 if (!get_ofp_packet_eth_header(r, &opi, ð)
942 || !eth_addr_equals(eth->eth_dst, stp_eth_addr)) {
946 port_no = ntohs(opi->in_port);
947 if (port_no >= STP_MAX_PORTS) {
948 /* STP only supports 255 ports. */
951 if (port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP) {
952 /* We're not doing STP on this port. */
956 if (opi->reason == OFPR_ACTION) {
957 /* The controller set up a flow for this, so we won't intercept it. */
961 get_ofp_packet_payload(opi, &payload);
962 flow_extract(&payload, port_no, &flow);
963 if (flow.dl_type != htons(OFP_DL_TYPE_NOT_ETH_TYPE)) {
964 VLOG_DBG("non-LLC frame received on STP multicast address");
967 llc = buffer_at_assert(&payload, sizeof *eth, sizeof *llc);
968 if (llc->llc_dsap != STP_LLC_DSAP) {
969 VLOG_DBG("bad DSAP 0x%02"PRIx8" received on STP multicast address",
974 /* Trim off padding on payload. */
975 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
976 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
978 if (buffer_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
979 struct stp_port *p = stp_get_port(stp->stp, port_no);
980 stp_received_bpdu(p, payload.data, payload.size);
989 return time_msec() * 256 / 1000;
993 stp_periodic_cb(void *stp_)
995 struct stp_data *stp = stp_;
996 long long int now_256ths = time_256ths();
997 long long int elapsed_256ths = now_256ths - stp->last_tick_256ths;
1000 if (!port_watcher_is_ready(stp->pw)) {
1001 /* Can't start STP until we know port flags, because port flags can
1005 if (elapsed_256ths <= 0) {
1009 stp_tick(stp->stp, MIN(INT_MAX, elapsed_256ths));
1010 stp->last_tick_256ths = now_256ths;
1012 while (stp_get_changed_port(stp->stp, &p)) {
1013 int port_no = stp_port_no(p);
1014 enum stp_state state = stp_port_get_state(p);
1016 if (state != STP_DISABLED) {
1017 VLOG_WARN("STP: Port %d entered %s state",
1018 port_no, stp_state_name(state));
1020 if (!(port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP)) {
1024 flags = OFPPFL_STP_LISTEN;
1027 flags = OFPPFL_STP_LEARN;
1030 case STP_FORWARDING:
1031 flags = OFPPFL_STP_FORWARD;
1034 flags = OFPPFL_STP_BLOCK;
1037 VLOG_DBG_RL(&vrl, "STP: Port %d has bad state %x",
1039 flags = OFPPFL_STP_FORWARD;
1042 if (!stp_forward_in_state(state)) {
1043 flags |= OFPPFL_NO_FLOOD;
1045 port_watcher_set_flags(stp->pw, port_no, flags,
1046 OFPPFL_STP_MASK | OFPPFL_NO_FLOOD);
1048 /* We don't own those flags. */
1054 stp_wait_cb(void *stp_ UNUSED)
1056 poll_timer_wait(1000);
1060 send_bpdu(const void *bpdu, size_t bpdu_size, int port_no, void *stp_)
1062 struct stp_data *stp = stp_;
1063 struct eth_header *eth;
1064 struct llc_header *llc;
1065 struct buffer pkt, *opo;
1067 /* Packet skeleton. */
1068 buffer_init(&pkt, ETH_HEADER_LEN + LLC_HEADER_LEN + bpdu_size);
1069 eth = buffer_put_uninit(&pkt, sizeof *eth);
1070 llc = buffer_put_uninit(&pkt, sizeof *llc);
1071 buffer_put(&pkt, bpdu, bpdu_size);
1074 memcpy(eth->eth_dst, stp_eth_addr, ETH_ADDR_LEN);
1075 memcpy(eth->eth_src, stp->pw->ports[port_no].hw_addr, ETH_ADDR_LEN);
1076 eth->eth_type = htons(pkt.size - ETH_HEADER_LEN);
1079 llc->llc_dsap = STP_LLC_DSAP;
1080 llc->llc_ssap = STP_LLC_SSAP;
1081 llc->llc_cntl = STP_LLC_CNTL;
1083 opo = make_unbuffered_packet_out(&pkt, OFPP_NONE, port_no);
1084 buffer_uninit(&pkt);
1085 rconn_send_with_limit(stp->local_rconn, opo, &stp->n_txq, OFPP_MAX);
1089 stp_port_watcher_cb(uint16_t port_no,
1090 const struct ofp_phy_port *old,
1091 const struct ofp_phy_port *new,
1094 struct stp_data *stp = stp_;
1097 /* STP only supports a maximum of 255 ports, one less than OpenFlow. We
1098 * don't support STP on OFPP_LOCAL, either. */
1099 if (port_no >= STP_MAX_PORTS) {
1103 p = stp_get_port(stp->stp, port_no);
1104 if (new->port_no == htons(OFPP_NONE)
1105 || new->flags & htonl(OFPPFL_NO_STP)) {
1106 stp_port_disable(p);
1109 stp_port_set_speed(p, new->speed);
1114 stp_hook_create(const struct settings *s, struct port_watcher *pw,
1115 struct rconn *local, struct rconn *remote)
1117 uint8_t dpid[ETH_ADDR_LEN];
1118 struct netdev *netdev;
1119 struct stp_data *stp;
1122 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1124 fatal(retval, "Could not open %s device", s->of_name);
1126 memcpy(dpid, netdev_get_etheraddr(netdev), ETH_ADDR_LEN);
1127 netdev_close(netdev);
1129 stp = xcalloc(1, sizeof *stp);
1130 stp->stp = stp_create("stp", eth_addr_to_uint64(dpid), send_bpdu, stp);
1132 memcpy(stp->dpid, dpid, ETH_ADDR_LEN);
1133 stp->local_rconn = local;
1134 stp->remote_rconn = remote;
1135 stp->last_tick_256ths = time_256ths();
1137 port_watcher_register_callback(pw, stp_port_watcher_cb, stp);
1138 return make_hook(stp_local_packet_cb, NULL,
1139 stp_periodic_cb, stp_wait_cb, stp);
1142 /* In-band control. */
1144 struct in_band_data {
1145 const struct settings *s;
1146 struct mac_learning *ml;
1147 struct netdev *of_device;
1148 struct rconn *controller;
1149 uint8_t mac[ETH_ADDR_LEN];
1154 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct buffer *b)
1156 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
1159 static const uint8_t *
1160 get_controller_mac(struct in_band_data *in_band)
1162 static uint32_t ip, last_nonzero_ip;
1163 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
1164 static time_t next_refresh = 0;
1166 uint32_t last_ip = ip;
1168 time_t now = time_now();
1170 ip = rconn_get_ip(in_band->controller);
1171 if (last_ip != ip || !next_refresh || now >= next_refresh) {
1174 /* Look up MAC address. */
1175 memset(mac, 0, sizeof mac);
1177 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
1179 VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s",
1180 IP_ARGS(&ip), strerror(retval));
1183 have_mac = !eth_addr_is_zero(mac);
1185 /* Log changes in IP, MAC addresses. */
1186 if (ip && ip != last_nonzero_ip) {
1187 VLOG_DBG("controller IP address changed from "IP_FMT
1188 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
1189 last_nonzero_ip = ip;
1191 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
1192 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
1194 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
1195 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
1198 /* Schedule next refresh.
1200 * If we have an IP address but not a MAC address, then refresh
1201 * quickly, since we probably will get a MAC address soon (via ARP).
1202 * Otherwise, we can afford to wait a little while. */
1203 next_refresh = now + (!ip || have_mac ? 10 : 1);
1205 return !eth_addr_is_zero(mac) ? mac : NULL;
1209 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
1210 struct in_band_data *in_band)
1212 const uint8_t *mac = get_controller_mac(in_band);
1213 return mac && eth_addr_equals(mac, dl_addr);
1217 in_band_learn_mac(struct in_band_data *in_band,
1218 uint16_t in_port, const uint8_t src_mac[ETH_ADDR_LEN])
1220 if (mac_learning_learn(in_band->ml, src_mac, in_port)) {
1221 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
1222 ETH_ADDR_ARGS(src_mac), in_port);
1227 in_band_local_packet_cb(struct relay *r, void *in_band_)
1229 struct in_band_data *in_band = in_band_;
1230 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
1231 struct ofp_packet_in *opi;
1232 struct eth_header *eth;
1233 struct buffer payload;
1238 if (!get_ofp_packet_eth_header(r, &opi, ð)) {
1241 in_port = ntohs(opi->in_port);
1243 /* Deal with local stuff. */
1244 if (in_port == OFPP_LOCAL) {
1245 /* Sent by secure channel. */
1246 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1247 } else if (eth_addr_equals(eth->eth_dst, in_band->mac)) {
1248 /* Sent to secure channel. */
1249 out_port = OFPP_LOCAL;
1250 in_band_learn_mac(in_band, in_port, eth->eth_src);
1251 } else if (eth->eth_type == htons(ETH_TYPE_ARP)
1252 && eth_addr_is_broadcast(eth->eth_dst)
1253 && is_controller_mac(eth->eth_src, in_band)) {
1254 /* ARP sent by controller. */
1255 out_port = OFPP_FLOOD;
1256 } else if (is_controller_mac(eth->eth_dst, in_band)
1257 || is_controller_mac(eth->eth_src, in_band)) {
1258 /* Traffic to or from controller. Switch it by hand. */
1259 in_band_learn_mac(in_band, in_port, eth->eth_src);
1260 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1262 const uint8_t *controller_mac;
1263 controller_mac = get_controller_mac(in_band);
1264 if (eth->eth_type == htons(ETH_TYPE_ARP)
1265 && eth_addr_is_broadcast(eth->eth_dst)
1266 && is_controller_mac(eth->eth_src, in_band)) {
1267 /* ARP sent by controller. */
1268 out_port = OFPP_FLOOD;
1269 } else if (is_controller_mac(eth->eth_dst, in_band)
1270 && in_port == mac_learning_lookup(in_band->ml,
1272 /* Drop controller traffic that arrives on the controller port. */
1279 get_ofp_packet_payload(opi, &payload);
1280 flow_extract(&payload, in_port, &flow);
1281 if (in_port == out_port) {
1282 /* The input and output port match. Set up a flow to drop packets. */
1283 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
1284 in_band->s->max_idle, 0));
1285 } else if (out_port != OFPP_FLOOD) {
1286 /* The output port is known, so add a new flow. */
1287 queue_tx(rc, in_band,
1288 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
1289 out_port, in_band->s->max_idle));
1291 /* If the switch didn't buffer the packet, we need to send a copy. */
1292 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1293 queue_tx(rc, in_band,
1294 make_unbuffered_packet_out(&payload, in_port, out_port));
1297 /* We don't know that MAC. Send along the packet without setting up a
1300 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1301 b = make_unbuffered_packet_out(&payload, in_port, out_port);
1303 b = make_buffered_packet_out(ntohl(opi->buffer_id),
1306 queue_tx(rc, in_band, b);
1312 in_band_status_cb(struct status_reply *sr, void *in_band_)
1314 struct in_band_data *in_band = in_band_;
1315 struct in_addr local_ip;
1316 uint32_t controller_ip;
1317 const uint8_t *controller_mac;
1319 if (netdev_get_in4(in_band->of_device, &local_ip)) {
1320 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
1322 status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
1323 ETH_ADDR_ARGS(in_band->mac));
1325 controller_ip = rconn_get_ip(in_band->controller);
1326 if (controller_ip) {
1327 status_reply_put(sr, "controller-ip="IP_FMT,
1328 IP_ARGS(&controller_ip));
1330 controller_mac = get_controller_mac(in_band);
1331 if (controller_mac) {
1332 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
1333 ETH_ADDR_ARGS(controller_mac));
1338 get_ofp_packet_payload(struct ofp_packet_in *opi, struct buffer *payload)
1340 payload->data = opi->data;
1341 payload->size = ntohs(opi->header.length) - offsetof(struct ofp_packet_in,
1346 in_band_hook_create(const struct settings *s, struct switch_status *ss,
1347 struct rconn *remote)
1349 struct in_band_data *in_band;
1352 in_band = xcalloc(1, sizeof *in_band);
1354 in_band->ml = mac_learning_create();
1355 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE,
1356 &in_band->of_device);
1358 fatal(retval, "Could not open %s device", s->of_name);
1360 memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device),
1362 in_band->controller = remote;
1363 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
1364 return make_hook(in_band_local_packet_cb, NULL, NULL, NULL, in_band);
1367 /* Fail open support. */
1369 struct fail_open_data {
1370 const struct settings *s;
1371 struct rconn *local_rconn;
1372 struct rconn *remote_rconn;
1373 struct lswitch *lswitch;
1374 int last_disconn_secs;
1375 time_t boot_deadline;
1378 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
1380 fail_open_periodic_cb(void *fail_open_)
1382 struct fail_open_data *fail_open = fail_open_;
1386 if (time_now() < fail_open->boot_deadline) {
1389 disconn_secs = rconn_disconnected_duration(fail_open->remote_rconn);
1390 open = disconn_secs >= fail_open->s->probe_interval * 3;
1391 if (open != (fail_open->lswitch != NULL)) {
1393 VLOG_WARN("No longer in fail-open mode");
1394 lswitch_destroy(fail_open->lswitch);
1395 fail_open->lswitch = NULL;
1397 VLOG_WARN("Could not connect to controller for %d seconds, "
1398 "failing open", disconn_secs);
1399 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
1400 fail_open->s->max_idle);
1401 fail_open->last_disconn_secs = disconn_secs;
1403 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
1404 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
1405 "from controller", disconn_secs);
1406 fail_open->last_disconn_secs = disconn_secs;
1411 fail_open_local_packet_cb(struct relay *r, void *fail_open_)
1413 struct fail_open_data *fail_open = fail_open_;
1414 if (!fail_open->lswitch) {
1417 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
1418 r->halves[HALF_LOCAL].rxbuf);
1419 rconn_run(fail_open->local_rconn);
1425 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
1427 struct fail_open_data *fail_open = fail_open_;
1428 const struct settings *s = fail_open->s;
1429 int trigger_duration = s->probe_interval * 3;
1430 int cur_duration = rconn_disconnected_duration(fail_open->remote_rconn);
1432 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
1433 status_reply_put(sr, "current-duration=%d", cur_duration);
1434 status_reply_put(sr, "triggered=%s",
1435 cur_duration >= trigger_duration ? "true" : "false");
1436 status_reply_put(sr, "max-idle=%d", s->max_idle);
1440 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
1441 struct rconn *local_rconn, struct rconn *remote_rconn)
1443 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
1445 fail_open->local_rconn = local_rconn;
1446 fail_open->remote_rconn = remote_rconn;
1447 fail_open->lswitch = NULL;
1448 fail_open->boot_deadline = time_now() + s->probe_interval * 3;
1449 fail_open->boot_deadline += STP_EXTRA_BOOT_TIME;
1450 switch_status_register_category(ss, "fail-open",
1451 fail_open_status_cb, fail_open);
1452 return make_hook(fail_open_local_packet_cb, NULL,
1453 fail_open_periodic_cb, NULL, fail_open);
1456 struct rate_limiter {
1457 const struct settings *s;
1458 struct rconn *remote_rconn;
1460 /* One queue per physical port. */
1461 struct queue queues[OFPP_MAX];
1462 int n_queued; /* Sum over queues[*].n. */
1463 int next_tx_port; /* Next port to check in round-robin. */
1467 * It costs 1000 tokens to send a single packet_in message. A single token
1468 * per message would be more straightforward, but this choice lets us avoid
1469 * round-off error in refill_bucket()'s calculation of how many tokens to
1470 * add to the bucket, since no division step is needed. */
1471 long long int last_fill; /* Time at which we last added tokens. */
1472 int tokens; /* Current number of tokens. */
1474 /* Transmission queue. */
1475 int n_txq; /* No. of packets waiting in rconn for tx. */
1477 /* Statistics reporting. */
1478 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
1479 unsigned long long n_limited; /* # queued for rate limiting. */
1480 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
1481 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
1484 /* Drop a packet from the longest queue in 'rl'. */
1486 drop_packet(struct rate_limiter *rl)
1488 struct queue *longest; /* Queue currently selected as longest. */
1489 int n_longest; /* # of queues of same length as 'longest'. */
1492 longest = &rl->queues[0];
1494 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
1495 if (longest->n < q->n) {
1498 } else if (longest->n == q->n) {
1501 /* Randomly select one of the longest queues, with a uniform
1502 * distribution (Knuth algorithm 3.4.2R). */
1503 if (!random_range(n_longest)) {
1509 /* FIXME: do we want to pop the tail instead? */
1510 buffer_delete(queue_pop_head(longest));
1514 /* Remove and return the next packet to transmit (in round-robin order). */
1515 static struct buffer *
1516 dequeue_packet(struct rate_limiter *rl)
1520 for (i = 0; i < OFPP_MAX; i++) {
1521 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
1522 struct queue *q = &rl->queues[port];
1524 rl->next_tx_port = (port + 1) % OFPP_MAX;
1526 return queue_pop_head(q);
1532 /* Add tokens to the bucket based on elapsed time. */
1534 refill_bucket(struct rate_limiter *rl)
1536 const struct settings *s = rl->s;
1537 long long int now = time_msec();
1538 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
1539 if (tokens >= 1000) {
1540 rl->last_fill = now;
1541 rl->tokens = MIN(tokens, s->burst_limit * 1000);
1545 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
1546 * true if successful, false otherwise. (In the latter case no tokens are
1549 get_token(struct rate_limiter *rl)
1551 if (rl->tokens >= 1000) {
1560 rate_limit_local_packet_cb(struct relay *r, void *rl_)
1562 struct rate_limiter *rl = rl_;
1563 const struct settings *s = rl->s;
1564 struct ofp_packet_in *opi;
1566 opi = get_ofp_packet_in(r);
1571 if (!rl->n_queued && get_token(rl)) {
1572 /* In the common case where we are not constrained by the rate limit,
1573 * let the packet take the normal path. */
1577 /* Otherwise queue it up for the periodic callback to drain out. */
1578 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
1579 int port = ntohs(opi->in_port) % OFPP_MAX;
1580 if (rl->n_queued >= s->burst_limit) {
1583 queue_push_tail(&rl->queues[port], buffer_clone(msg));
1591 rate_limit_status_cb(struct status_reply *sr, void *rl_)
1593 struct rate_limiter *rl = rl_;
1595 status_reply_put(sr, "normal=%llu", rl->n_normal);
1596 status_reply_put(sr, "limited=%llu", rl->n_limited);
1597 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
1598 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
1602 rate_limit_periodic_cb(void *rl_)
1604 struct rate_limiter *rl = rl_;
1607 /* Drain some packets out of the bucket if possible, but limit the number
1608 * of iterations to allow other code to get work done too. */
1610 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
1611 /* Use a small, arbitrary limit for the amount of queuing to do here,
1612 * because the TCP connection is responsible for buffering and there is
1613 * no point in trying to transmit faster than the TCP connection can
1615 struct buffer *b = dequeue_packet(rl);
1616 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
1623 rate_limit_wait_cb(void *rl_)
1625 struct rate_limiter *rl = rl_;
1627 if (rl->tokens >= 1000) {
1628 /* We can transmit more packets as soon as we're called again. */
1629 poll_immediate_wake();
1631 /* We have to wait for the bucket to re-fill. We could calculate
1632 * the exact amount of time here for increased smoothness. */
1633 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
1639 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
1640 struct rconn *local, struct rconn *remote)
1642 struct rate_limiter *rl;
1645 rl = xcalloc(1, sizeof *rl);
1647 rl->remote_rconn = remote;
1648 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
1649 queue_init(&rl->queues[i]);
1651 rl->last_fill = time_msec();
1652 rl->tokens = s->rate_limit * 100;
1653 switch_status_register_category(ss, "rate-limit",
1654 rate_limit_status_cb, rl);
1655 return make_hook(rate_limit_local_packet_cb, NULL, rate_limit_periodic_cb,
1656 rate_limit_wait_cb, rl);
1659 /* OFPST_SWITCH statistics. */
1661 struct switch_status_category {
1663 void (*cb)(struct status_reply *, void *aux);
1667 struct switch_status {
1668 const struct settings *s;
1670 struct switch_status_category categories[8];
1674 struct status_reply {
1675 struct switch_status_category *category;
1681 switch_status_remote_packet_cb(struct relay *r, void *ss_)
1683 struct switch_status *ss = ss_;
1684 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
1685 struct buffer *msg = r->halves[HALF_REMOTE].rxbuf;
1686 struct switch_status_category *c;
1687 struct nicira_header *request;
1688 struct nicira_header *reply;
1689 struct status_reply sr;
1693 if (msg->size < sizeof(struct nicira_header)) {
1696 request = msg->data;
1697 if (request->header.type != OFPT_VENDOR
1698 || request->vendor_id != htonl(NX_VENDOR_ID)
1699 || request->subtype != htonl(NXT_STATUS_REQUEST)) {
1703 sr.request.string = (void *) (request + 1);
1704 sr.request.length = msg->size - sizeof *request;
1705 ds_init(&sr.output);
1706 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
1707 if (!memcmp(c->name, sr.request.string,
1708 MIN(strlen(c->name), sr.request.length))) {
1713 reply = make_openflow_xid(sizeof *reply + sr.output.length,
1714 OFPT_VENDOR, request->header.xid, &b);
1715 reply->vendor_id = htonl(NX_VENDOR_ID);
1716 reply->subtype = htonl(NXT_STATUS_REPLY);
1717 memcpy(reply + 1, sr.output.string, sr.output.length);
1718 retval = rconn_send(rc, b, NULL);
1719 if (retval && retval != EAGAIN) {
1720 VLOG_WARN("send failed (%s)", strerror(retval));
1722 ds_destroy(&sr.output);
1727 rconn_status_cb(struct status_reply *sr, void *rconn_)
1729 struct rconn *rconn = rconn_;
1730 time_t now = time_now();
1732 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
1733 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
1734 status_reply_put(sr, "backoff=%d", rconn_get_backoff(rconn));
1735 status_reply_put(sr, "is-connected=%s",
1736 rconn_is_connected(rconn) ? "true" : "false");
1737 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
1738 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
1739 status_reply_put(sr, "attempted-connections=%u",
1740 rconn_get_attempted_connections(rconn));
1741 status_reply_put(sr, "successful-connections=%u",
1742 rconn_get_successful_connections(rconn));
1743 status_reply_put(sr, "last-connection=%ld",
1744 (long int) (now - rconn_get_last_connection(rconn)));
1745 status_reply_put(sr, "time-connected=%lu",
1746 rconn_get_total_time_connected(rconn));
1747 status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(rconn));
1751 config_status_cb(struct status_reply *sr, void *s_)
1753 const struct settings *s = s_;
1756 for (i = 0; i < s->n_listeners; i++) {
1757 status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
1759 if (s->probe_interval) {
1760 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
1762 if (s->max_backoff) {
1763 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
1768 switch_status_cb(struct status_reply *sr, void *ss_)
1770 struct switch_status *ss = ss_;
1771 time_t now = time_now();
1773 status_reply_put(sr, "now=%ld", (long int) now);
1774 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
1775 status_reply_put(sr, "pid=%ld", (long int) getpid());
1779 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
1781 struct switch_status *ss = xcalloc(1, sizeof *ss);
1783 ss->booted = time_now();
1784 switch_status_register_category(ss, "config",
1785 config_status_cb, (void *) s);
1786 switch_status_register_category(ss, "switch", switch_status_cb, ss);
1788 return make_hook(NULL, switch_status_remote_packet_cb, NULL, NULL, ss);
1792 switch_status_register_category(struct switch_status *ss,
1793 const char *category,
1794 void (*cb)(struct status_reply *,
1798 struct switch_status_category *c;
1799 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
1800 c = &ss->categories[ss->n_categories++];
1803 c->name = xstrdup(category);
1807 status_reply_put(struct status_reply *sr, const char *content, ...)
1809 size_t old_length = sr->output.length;
1813 /* Append the status reply to the output. */
1814 ds_put_format(&sr->output, "%s.", sr->category->name);
1815 va_start(args, content);
1816 ds_put_format_valist(&sr->output, content, args);
1818 if (ds_last(&sr->output) != '\n') {
1819 ds_put_char(&sr->output, '\n');
1822 /* Drop what we just added if it doesn't match the request. */
1823 added = sr->output.length - old_length;
1824 if (added < sr->request.length
1825 || memcmp(&sr->output.string[old_length],
1826 sr->request.string, sr->request.length)) {
1827 ds_truncate(&sr->output, old_length);
1832 /* Controller discovery. */
1836 const struct settings *s;
1837 struct dhclient *dhcp;
1842 discovery_status_cb(struct status_reply *sr, void *d_)
1844 struct discovery *d = d_;
1846 status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re);
1847 status_reply_put(sr, "n-changes=%d", d->n_changes);
1848 status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp));
1849 status_reply_put(sr, "state-elapsed=%u",
1850 dhclient_get_state_elapsed(d->dhcp));
1851 if (dhclient_is_bound(d->dhcp)) {
1852 uint32_t ip = dhclient_get_ip(d->dhcp);
1853 uint32_t netmask = dhclient_get_netmask(d->dhcp);
1854 uint32_t router = dhclient_get_router(d->dhcp);
1856 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
1857 uint32_t dns_server;
1861 status_reply_put(sr, "ip="IP_FMT, IP_ARGS(&ip));
1862 status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask));
1864 status_reply_put(sr, "router="IP_FMT, IP_ARGS(&router));
1867 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i, &dns_server);
1869 status_reply_put(sr, "dns%d="IP_FMT, i, IP_ARGS(&dns_server));
1872 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
1874 status_reply_put(sr, "domain=%s", domain_name);
1878 status_reply_put(sr, "lease-remaining=%u",
1879 dhclient_get_lease_remaining(d->dhcp));
1883 static struct discovery *
1884 discovery_init(const struct settings *s, struct switch_status *ss)
1886 struct netdev *netdev;
1887 struct discovery *d;
1888 struct dhclient *dhcp;
1891 /* Bring ofX network device up. */
1892 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1894 fatal(retval, "Could not open %s device", s->of_name);
1896 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
1898 fatal(retval, "Could not bring %s device up", s->of_name);
1900 netdev_close(netdev);
1902 /* Initialize DHCP client. */
1903 retval = dhclient_create(s->of_name, modify_dhcp_request,
1904 validate_dhcp_offer, (void *) s, &dhcp);
1906 fatal(retval, "Failed to initialize DHCP client");
1908 dhclient_init(dhcp, 0);
1910 d = xmalloc(sizeof *d);
1915 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
1921 discovery_question_connectivity(struct discovery *d)
1923 dhclient_force_renew(d->dhcp, 15);
1927 discovery_run(struct discovery *d, char **controller_name)
1929 dhclient_run(d->dhcp);
1930 if (!dhclient_changed(d->dhcp)) {
1934 dhclient_configure_netdev(d->dhcp);
1935 if (d->s->update_resolv_conf) {
1936 dhclient_update_resolv_conf(d->dhcp);
1939 if (dhclient_is_bound(d->dhcp)) {
1940 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
1941 DHCP_CODE_OFP_CONTROLLER_VCONN);
1942 VLOG_WARN("%s: discovered controller", *controller_name);
1945 *controller_name = NULL;
1947 VLOG_WARN("discovered controller no longer available");
1955 discovery_wait(struct discovery *d)
1957 dhclient_wait(d->dhcp);
1961 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
1963 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
1967 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
1969 const struct settings *s = s_;
1973 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
1975 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
1978 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
1980 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
1981 s->accept_controller_re);
1987 /* User interface. */
1990 parse_options(int argc, char *argv[], struct settings *s)
1993 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
1995 OPT_INACTIVITY_PROBE,
2001 static struct option long_options[] = {
2002 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
2003 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
2004 {"fail", required_argument, 0, 'F'},
2005 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
2006 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
2007 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
2008 {"listen", required_argument, 0, 'l'},
2009 {"monitor", required_argument, 0, 'm'},
2010 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
2011 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
2012 {"detach", no_argument, 0, 'D'},
2013 {"force", no_argument, 0, 'f'},
2014 {"pidfile", optional_argument, 0, 'P'},
2015 {"verbose", optional_argument, 0, 'v'},
2016 {"help", no_argument, 0, 'h'},
2017 {"version", no_argument, 0, 'V'},
2018 VCONN_SSL_LONG_OPTIONS
2021 char *short_options = long_options_to_short_options(long_options);
2022 char *accept_re = NULL;
2025 /* Set defaults that we can figure out before parsing options. */
2027 s->monitor_name = NULL;
2028 s->fail_mode = FAIL_OPEN;
2030 s->probe_interval = 15;
2031 s->max_backoff = 15;
2032 s->update_resolv_conf = true;
2038 c = getopt_long(argc, argv, short_options, long_options, NULL);
2044 case OPT_ACCEPT_VCONN:
2045 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
2048 case OPT_NO_RESOLV_CONF:
2049 s->update_resolv_conf = false;
2053 if (!strcmp(optarg, "open")) {
2054 s->fail_mode = FAIL_OPEN;
2055 } else if (!strcmp(optarg, "closed")) {
2056 s->fail_mode = FAIL_CLOSED;
2059 "-f or --fail argument must be \"open\" or \"closed\"");
2063 case OPT_INACTIVITY_PROBE:
2064 s->probe_interval = atoi(optarg);
2065 if (s->probe_interval < 5) {
2066 fatal(0, "--inactivity-probe argument must be at least 5");
2071 if (!strcmp(optarg, "permanent")) {
2072 s->max_idle = OFP_FLOW_PERMANENT;
2074 s->max_idle = atoi(optarg);
2075 if (s->max_idle < 1 || s->max_idle > 65535) {
2076 fatal(0, "--max-idle argument must be between 1 and "
2077 "65535 or the word 'permanent'");
2082 case OPT_MAX_BACKOFF:
2083 s->max_backoff = atoi(optarg);
2084 if (s->max_backoff < 1) {
2085 fatal(0, "--max-backoff argument must be at least 1");
2086 } else if (s->max_backoff > 3600) {
2087 s->max_backoff = 3600;
2091 case OPT_RATE_LIMIT:
2093 s->rate_limit = atoi(optarg);
2094 if (s->rate_limit < 1) {
2095 fatal(0, "--rate-limit argument must be at least 1");
2098 s->rate_limit = 1000;
2102 case OPT_BURST_LIMIT:
2103 s->burst_limit = atoi(optarg);
2104 if (s->burst_limit < 1) {
2105 fatal(0, "--burst-limit argument must be at least 1");
2114 set_pidfile(optarg);
2118 ignore_existing_pidfile();
2122 if (s->n_listeners >= MAX_MGMT) {
2123 fatal(0, "-l or --listen may be specified at most %d times",
2126 s->listener_names[s->n_listeners++] = optarg;
2130 if (s->monitor_name) {
2131 fatal(0, "-m or --monitor may only be specified once");
2133 s->monitor_name = optarg;
2140 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
2144 vlog_set_verbosity(optarg);
2147 VCONN_SSL_OPTION_HANDLERS
2156 free(short_options);
2160 if (argc < 1 || argc > 2) {
2161 fatal(0, "need one or two non-option arguments; use --help for usage");
2164 /* Local and remote vconns. */
2165 s->nl_name = argv[0];
2166 if (strncmp(s->nl_name, "nl:", 3)
2167 || strlen(s->nl_name) < 4
2168 || s->nl_name[strspn(s->nl_name + 3, "0123456789") + 3]) {
2169 fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"", s->nl_name);
2171 s->of_name = xasprintf("of%s", s->nl_name + 3);
2172 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
2174 /* Set accept_controller_regex. */
2176 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
2178 retval = regcomp(&s->accept_controller_regex, accept_re,
2179 REG_NOSUB | REG_EXTENDED);
2181 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
2182 char *buffer = xmalloc(length);
2183 regerror(retval, &s->accept_controller_regex, buffer, length);
2184 fatal(0, "%s: %s", accept_re, buffer);
2186 s->accept_controller_re = accept_re;
2188 /* Mode of operation. */
2189 s->discovery = s->controller_name == NULL;
2193 enum netdev_flags flags;
2194 struct netdev *netdev;
2196 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
2198 fatal(retval, "Could not open %s device", s->of_name);
2201 retval = netdev_get_flags(netdev, &flags);
2203 fatal(retval, "Could not get flags for %s device", s->of_name);
2206 s->in_band = (flags & NETDEV_UP) != 0;
2207 if (s->in_band && netdev_get_in6(netdev, NULL)) {
2208 VLOG_WARN("Ignoring IPv6 address on %s device: IPv6 not supported",
2212 netdev_close(netdev);
2215 /* Rate limiting. */
2216 if (s->rate_limit) {
2217 if (s->rate_limit < 100) {
2218 VLOG_WARN("Rate limit set to unusually low value %d",
2221 if (!s->burst_limit) {
2222 s->burst_limit = s->rate_limit / 4;
2224 s->burst_limit = MAX(s->burst_limit, 1);
2225 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
2232 printf("%s: secure channel, a relay for OpenFlow messages.\n"
2233 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
2234 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
2235 "CONTROLLER is an active OpenFlow connection method; if it is\n"
2236 "omitted, then secchan performs controller discovery.\n",
2237 program_name, program_name);
2238 vconn_usage(true, true);
2239 printf("\nController discovery options:\n"
2240 " --accept-vconn=REGEX accept matching discovered controllers\n"
2241 " --no-resolv-conf do not update /etc/resolv.conf\n"
2242 "\nNetworking options:\n"
2243 " -F, --fail=open|closed when controller connection fails:\n"
2244 " closed: drop all packets\n"
2245 " open (default): act as learning switch\n"
2246 " --inactivity-probe=SECS time between inactivity probes\n"
2247 " --max-idle=SECS max idle for flows set up by secchan\n"
2248 " --max-backoff=SECS max time between controller connection\n"
2249 " attempts (default: 15 seconds)\n"
2250 " -l, --listen=METHOD allow management connections on METHOD\n"
2251 " (a passive OpenFlow connection method)\n"
2252 " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n"
2253 " (a passive OpenFlow connection method)\n"
2254 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
2255 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
2256 " --burst-limit=BURST limit on packet credit for idle time\n"
2257 "\nOther options:\n"
2258 " -D, --detach run in background as daemon\n"
2259 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
2260 " -f, --force with -P, start even if already running\n"
2261 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
2262 " -v, --verbose set maximum verbosity level\n"
2263 " -h, --help display this help message\n"
2264 " -V, --version display version information\n",