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>
48 #include "command-line.h"
51 #include "dhcp-client.h"
53 #include "dynamic-string.h"
56 #include "learning-switch.h"
58 #include "mac-learning.h"
60 #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? */
114 /* Spanning tree protocol. */
120 struct ofpbuf *rxbuf;
121 int n_txq; /* No. of packets queued for tx on 'rconn'. */
128 #define HALF_REMOTE 1
129 struct half halves[2];
135 bool (*packet_cb[2])(struct relay *, void *aux);
136 void (*periodic_cb)(void *aux);
137 void (*wait_cb)(void *aux);
141 static struct vlog_rate_limit vrl = VLOG_RATE_LIMIT_INIT(60, 60);
143 static void parse_options(int argc, char *argv[], struct settings *);
144 static void usage(void) NO_RETURN;
146 static struct pvconn *open_passive_vconn(const char *name);
147 static struct vconn *accept_vconn(struct pvconn *pvconn);
149 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
151 static struct relay *relay_accept(const struct settings *, struct pvconn *);
152 static void relay_run(struct relay *, const struct hook[], size_t n_hooks);
153 static void relay_wait(struct relay *);
154 static void relay_destroy(struct relay *);
156 static struct hook make_hook(bool (*local_packet_cb)(struct relay *, void *),
157 bool (*remote_packet_cb)(struct relay *, void *),
158 void (*periodic_cb)(void *),
159 void (*wait_cb)(void *),
161 static struct ofp_packet_in *get_ofp_packet_in(struct relay *);
162 static bool get_ofp_packet_eth_header(struct relay *, struct ofp_packet_in **,
163 struct eth_header **);
164 static void get_ofp_packet_payload(struct ofp_packet_in *, struct ofpbuf *);
166 struct switch_status;
168 static struct hook switch_status_hook_create(const struct settings *,
169 struct switch_status **);
170 static void switch_status_register_category(struct switch_status *,
171 const char *category,
172 void (*cb)(struct status_reply *,
175 static void status_reply_put(struct status_reply *, const char *, ...)
178 static void rconn_status_cb(struct status_reply *, void *rconn_);
180 static struct discovery *discovery_init(const struct settings *,
181 struct switch_status *);
182 static void discovery_question_connectivity(struct discovery *);
183 static bool discovery_run(struct discovery *, char **controller_name);
184 static void discovery_wait(struct discovery *);
186 static struct hook in_band_hook_create(const struct settings *,
187 struct switch_status *,
188 struct rconn *remote);
191 static struct hook port_watcher_create(struct rconn *local,
192 struct rconn *remote,
193 struct port_watcher **);
194 static uint32_t port_watcher_get_config(const struct port_watcher *,
196 static void port_watcher_set_flags(struct port_watcher *, int port_no,
197 uint32_t config, uint32_t c_mask,
198 uint32_t state, uint32_t s_mask);
200 static struct hook stp_hook_create(const struct settings *,
201 struct port_watcher *,
202 struct rconn *local, struct rconn *remote);
204 static struct hook fail_open_hook_create(const struct settings *,
205 struct switch_status *,
207 struct rconn *remote);
208 static struct hook rate_limit_hook_create(const struct settings *,
209 struct switch_status *,
211 struct rconn *remote);
214 static void modify_dhcp_request(struct dhcp_msg *, void *aux);
215 static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux);
218 main(int argc, char *argv[])
222 struct list relays = LIST_INITIALIZER(&relays);
224 struct hook hooks[8];
227 struct pvconn *monitor;
229 struct pvconn *listeners[MAX_MGMT];
232 struct rconn *local_rconn, *remote_rconn;
233 struct relay *controller_relay;
234 struct discovery *discovery;
235 struct switch_status *switch_status;
236 struct port_watcher *pw;
240 set_program_name(argv[0]);
241 register_fault_handlers();
244 parse_options(argc, argv, &s);
245 signal(SIGPIPE, SIG_IGN);
247 /* Start listening for management and monitoring connections. */
249 for (i = 0; i < s.n_listeners; i++) {
250 listeners[n_listeners++] = open_passive_vconn(s.listener_names[i]);
252 monitor = s.monitor_name ? open_passive_vconn(s.monitor_name) : NULL;
254 /* Initialize switch status hook. */
255 hooks[n_hooks++] = switch_status_hook_create(&s, &switch_status);
257 /* Start controller discovery. */
258 discovery = s.discovery ? discovery_init(&s, switch_status) : NULL;
260 /* Start listening for vlogconf requests. */
261 retval = vlog_server_listen(NULL, NULL);
263 ofp_fatal(retval, "Could not listen for vlog connections");
266 die_if_already_running();
269 VLOG_WARN("OpenFlow reference implementation version %s", VERSION);
270 VLOG_WARN("OpenFlow protocol version 0x%02x", OFP_VERSION);
272 /* Connect to datapath. */
273 local_rconn = rconn_create(0, s.max_backoff);
274 rconn_connect(local_rconn, s.nl_name);
275 switch_status_register_category(switch_status, "local",
276 rconn_status_cb, local_rconn);
278 /* Connect to controller. */
279 remote_rconn = rconn_create(s.probe_interval, s.max_backoff);
280 if (s.controller_name) {
281 retval = rconn_connect(remote_rconn, s.controller_name);
282 if (retval == EAFNOSUPPORT) {
283 ofp_fatal(0, "No support for %s vconn", s.controller_name);
286 switch_status_register_category(switch_status, "remote",
287 rconn_status_cb, remote_rconn);
289 /* Start relaying. */
290 controller_relay = relay_create(local_rconn, remote_rconn, false);
291 list_push_back(&relays, &controller_relay->node);
294 hooks[n_hooks++] = port_watcher_create(local_rconn, remote_rconn, &pw);
296 hooks[n_hooks++] = stp_hook_create(&s, pw, local_rconn, remote_rconn);
299 hooks[n_hooks++] = in_band_hook_create(&s, switch_status,
302 if (s.fail_mode == FAIL_OPEN) {
303 hooks[n_hooks++] = fail_open_hook_create(&s, switch_status,
304 local_rconn, remote_rconn);
307 hooks[n_hooks++] = rate_limit_hook_create(&s, switch_status,
308 local_rconn, remote_rconn);
310 assert(n_hooks <= ARRAY_SIZE(hooks));
317 LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) {
318 relay_run(r, hooks, n_hooks);
320 for (i = 0; i < n_listeners; i++) {
322 struct relay *r = relay_accept(&s, listeners[i]);
326 list_push_back(&relays, &r->node);
330 struct vconn *new = accept_vconn(monitor);
332 rconn_add_monitor(local_rconn, new);
335 for (i = 0; i < n_hooks; i++) {
336 if (hooks[i].periodic_cb) {
337 hooks[i].periodic_cb(hooks[i].aux);
341 char *controller_name;
342 if (rconn_is_connectivity_questionable(remote_rconn)) {
343 discovery_question_connectivity(discovery);
345 if (discovery_run(discovery, &controller_name)) {
346 if (controller_name) {
347 rconn_connect(remote_rconn, controller_name);
349 rconn_disconnect(remote_rconn);
354 /* Wait for something to happen. */
355 LIST_FOR_EACH (r, struct relay, node, &relays) {
358 for (i = 0; i < n_listeners; i++) {
359 pvconn_wait(listeners[i]);
362 pvconn_wait(monitor);
364 for (i = 0; i < n_hooks; i++) {
365 if (hooks[i].wait_cb) {
366 hooks[i].wait_cb(hooks[i].aux);
370 discovery_wait(discovery);
378 static struct pvconn *
379 open_passive_vconn(const char *name)
381 struct pvconn *pvconn;
384 retval = pvconn_open(name, &pvconn);
385 if (retval && retval != EAGAIN) {
386 ofp_fatal(retval, "opening %s", name);
391 static struct vconn *
392 accept_vconn(struct pvconn *pvconn)
397 retval = pvconn_accept(pvconn, OFP_VERSION, &new);
398 if (retval && retval != EAGAIN) {
399 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
405 make_hook(bool (*local_packet_cb)(struct relay *, void *aux),
406 bool (*remote_packet_cb)(struct relay *, void *aux),
407 void (*periodic_cb)(void *aux),
408 void (*wait_cb)(void *aux),
412 h.packet_cb[HALF_LOCAL] = local_packet_cb;
413 h.packet_cb[HALF_REMOTE] = remote_packet_cb;
414 h.periodic_cb = periodic_cb;
420 static struct ofp_packet_in *
421 get_ofp_packet_in(struct relay *r)
423 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
424 struct ofp_header *oh = msg->data;
425 if (oh->type == OFPT_PACKET_IN) {
426 if (msg->size >= offsetof (struct ofp_packet_in, data)) {
429 VLOG_WARN("packet too short (%zu bytes) for packet_in",
437 get_ofp_packet_eth_header(struct relay *r, struct ofp_packet_in **opip,
438 struct eth_header **ethp)
440 const int min_len = offsetof(struct ofp_packet_in, data) + ETH_HEADER_LEN;
441 struct ofp_packet_in *opi = get_ofp_packet_in(r);
442 if (opi && ntohs(opi->header.length) >= min_len) {
444 *ethp = (void *) opi->data;
451 /* OpenFlow message relaying. */
453 static struct relay *
454 relay_accept(const struct settings *s, struct pvconn *pvconn)
456 struct vconn *new_remote, *new_local;
457 char *nl_name_without_subscription;
458 struct rconn *r1, *r2;
461 new_remote = accept_vconn(pvconn);
466 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We
467 * only accept the former syntax in main().
469 * nl:123:0 opens a netlink connection to local datapath 123 without
470 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
472 nl_name_without_subscription = xasprintf("%s:0", s->nl_name);
473 retval = vconn_open(nl_name_without_subscription, OFP_VERSION, &new_local);
475 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
476 nl_name_without_subscription, strerror(retval));
477 vconn_close(new_remote);
478 free(nl_name_without_subscription);
482 /* Create and return relay. */
483 r1 = rconn_create(0, 0);
484 rconn_connect_unreliably(r1, nl_name_without_subscription, new_local);
485 free(nl_name_without_subscription);
487 r2 = rconn_create(0, 0);
488 rconn_connect_unreliably(r2, "passive", new_remote);
490 return relay_create(r1, r2, true);
493 static struct relay *
494 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
496 struct relay *r = xcalloc(1, sizeof *r);
497 r->halves[HALF_LOCAL].rconn = local;
498 r->halves[HALF_REMOTE].rconn = remote;
499 r->is_mgmt_conn = is_mgmt_conn;
504 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
509 for (i = 0; i < 2; i++) {
510 rconn_run(r->halves[i].rconn);
513 /* Limit the number of iterations to prevent other tasks from starving. */
514 for (iteration = 0; iteration < 50; iteration++) {
515 bool progress = false;
516 for (i = 0; i < 2; i++) {
517 struct half *this = &r->halves[i];
518 struct half *peer = &r->halves[!i];
521 this->rxbuf = rconn_recv(this->rconn);
522 if (this->rxbuf && (i == HALF_REMOTE || !r->is_mgmt_conn)) {
523 const struct hook *h;
524 for (h = hooks; h < &hooks[n_hooks]; h++) {
525 if (h->packet_cb[i] && h->packet_cb[i](r, h->aux)) {
526 ofpbuf_delete(this->rxbuf);
535 if (this->rxbuf && !this->n_txq) {
536 int retval = rconn_send(peer->rconn, this->rxbuf,
538 if (retval != EAGAIN) {
542 ofpbuf_delete(this->rxbuf);
553 if (r->is_mgmt_conn) {
554 for (i = 0; i < 2; i++) {
555 struct half *this = &r->halves[i];
556 if (!rconn_is_alive(this->rconn)) {
565 relay_wait(struct relay *r)
569 for (i = 0; i < 2; i++) {
570 struct half *this = &r->halves[i];
572 rconn_run_wait(this->rconn);
574 rconn_recv_wait(this->rconn);
580 relay_destroy(struct relay *r)
584 list_remove(&r->node);
585 for (i = 0; i < 2; i++) {
586 struct half *this = &r->halves[i];
587 rconn_destroy(this->rconn);
588 ofpbuf_delete(this->rxbuf);
593 /* Port status watcher. */
595 typedef void port_changed_cb_func(uint16_t port_no,
596 const struct ofp_phy_port *old,
597 const struct ofp_phy_port *new,
600 struct port_watcher_cb {
601 port_changed_cb_func *port_changed;
605 struct port_watcher {
606 struct rconn *local_rconn;
607 struct rconn *remote_rconn;
608 struct ofp_phy_port ports[OFPP_MAX + 1];
609 time_t last_feature_request;
610 bool got_feature_reply;
612 struct port_watcher_cb cbs[2];
616 /* Returns the number of fields that differ from 'a' to 'b'. */
618 opp_differs(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
620 BUILD_ASSERT_DECL(sizeof *a == 48); /* Trips when we add or remove fields. */
621 return ((a->port_no != b->port_no)
622 + (memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) != 0)
623 + (memcmp(a->name, b->name, sizeof a->name) != 0)
624 + (a->config != b->config)
625 + (a->state != b->state)
626 + (a->curr != b->curr)
627 + (a->advertised != b->advertised)
628 + (a->supported != b->supported)
629 + (a->peer != b->peer));
633 sanitize_opp(struct ofp_phy_port *opp)
637 for (i = 0; i < sizeof opp->name; i++) {
638 char c = opp->name[i];
639 if (c && (c < 0x20 || c > 0x7e)) {
643 opp->name[sizeof opp->name - 1] = '\0';
647 port_no_to_pw_idx(int port_no)
649 return (port_no < OFPP_MAX ? port_no
650 : port_no == OFPP_LOCAL ? OFPP_MAX
655 call_port_changed_callbacks(struct port_watcher *pw, int port_no,
656 const struct ofp_phy_port *old,
657 const struct ofp_phy_port *new)
659 if (opp_differs(old, new)) {
661 for (i = 0; i < pw->n_cbs; i++) {
662 port_changed_cb_func *port_changed = pw->cbs[i].port_changed;
664 (port_changed)(port_no, old, new, pw->cbs[i].aux);
671 update_phy_port(struct port_watcher *pw, struct ofp_phy_port *opp,
672 uint8_t reason, bool seen[OFPP_MAX + 1])
674 struct ofp_phy_port *pw_opp;
675 struct ofp_phy_port old;
679 port_no = ntohs(opp->port_no);
680 idx = port_no_to_pw_idx(port_no);
689 pw_opp = &pw->ports[idx];
691 if (reason == OFPPR_DELETE) {
692 memset(pw_opp, 0, sizeof *pw_opp);
693 pw_opp->port_no = htons(OFPP_NONE);
694 } else if (reason == OFPPR_MODIFY || reason == OFPPR_ADD) {
696 sanitize_opp(pw_opp);
698 call_port_changed_callbacks(pw, port_no, &old, pw_opp);
702 port_watcher_local_packet_cb(struct relay *r, void *pw_)
704 struct port_watcher *pw = pw_;
705 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
706 struct ofp_header *oh = msg->data;
708 if (oh->type == OFPT_FEATURES_REPLY
709 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
710 struct ofp_switch_features *osf = msg->data;
711 bool seen[ARRAY_SIZE(pw->ports)];
715 pw->got_feature_reply = true;
717 /* Update each port included in the message. */
718 memset(seen, 0, sizeof seen);
719 n_ports = ((msg->size - offsetof(struct ofp_switch_features, ports))
720 / sizeof *osf->ports);
721 for (i = 0; i < n_ports; i++) {
722 struct ofp_phy_port *opp = &osf->ports[i];
723 update_phy_port(pw, opp, OFPPR_MODIFY, seen);
726 /* Delete all the ports not included in the message. */
727 for (i = 0; i < ARRAY_SIZE(pw->ports); i++) {
729 update_phy_port(pw, &pw->ports[i], OFPPR_DELETE, NULL);
732 } else if (oh->type == OFPT_PORT_STATUS
733 && msg->size >= sizeof(struct ofp_port_status)) {
734 struct ofp_port_status *ops = msg->data;
735 update_phy_port(pw, &ops->desc, ops->reason, NULL);
741 port_watcher_remote_packet_cb(struct relay *r, void *pw_)
743 struct port_watcher *pw = pw_;
744 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
745 struct ofp_header *oh = msg->data;
747 if (oh->type == OFPT_PORT_MOD
748 && msg->size >= sizeof(struct ofp_port_mod)) {
749 struct ofp_port_mod *opm = msg->data;
750 uint16_t port_no = ntohs(opm->port_no);
751 int idx = port_no_to_pw_idx(port_no);
753 struct ofp_phy_port *pw_opp = &pw->ports[idx];
754 if (pw_opp->port_no != htons(OFPP_NONE)) {
755 struct ofp_phy_port old = *pw_opp;
756 pw_opp->config = ((pw_opp->config & ~opm->mask)
757 | (opm->config & opm->mask));
758 call_port_changed_callbacks(pw, port_no, &old, pw_opp);
766 port_watcher_periodic_cb(void *pw_)
768 struct port_watcher *pw = pw_;
770 if (!pw->got_feature_reply && time_now() >= pw->last_feature_request + 5) {
772 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
773 rconn_send_with_limit(pw->local_rconn, b, &pw->n_txq, 1);
774 pw->last_feature_request = time_now();
779 put_duplexes(struct ds *ds, const char *name, uint32_t features,
780 uint32_t hd_bit, uint32_t fd_bit)
782 if (features & (hd_bit | fd_bit)) {
783 ds_put_format(ds, " %s", name);
784 if (features & hd_bit) {
785 ds_put_cstr(ds, "(HD)");
787 if (features & fd_bit) {
788 ds_put_cstr(ds, "(FD)");
794 put_features(struct ds *ds, const char *name, uint32_t features) {
795 if (features & (OFPPF_10MB_HD | OFPPF_10MB_FD
796 | OFPPF_100MB_HD | OFPPF_100MB_FD
797 | OFPPF_1GB_HD | OFPPF_1GB_FD | OFPPF_10GB_FD)) {
798 ds_put_cstr(ds, name);
799 put_duplexes(ds, "10M", features, OFPPF_10MB_HD, OFPPF_10MB_FD);
800 put_duplexes(ds, "100M", features,
801 OFPPF_100MB_HD, OFPPF_100MB_FD);
802 put_duplexes(ds, "1G", features, OFPPF_100MB_HD, OFPPF_100MB_FD);
803 if (features & OFPPF_10GB_FD) {
804 ds_put_cstr(ds, " 10G");
806 if (features & OFPPF_AUTONEG) {
807 ds_put_cstr(ds, " AUTO_NEG");
809 if (features & OFPPF_PAUSE) {
810 ds_put_cstr(ds, " PAUSE");
812 if (features & OFPPF_PAUSE_ASYM) {
813 ds_put_cstr(ds, " PAUSE_ASYM");
819 log_port_status(uint16_t port_no,
820 const struct ofp_phy_port *old,
821 const struct ofp_phy_port *new,
824 if (VLOG_IS_DBG_ENABLED()) {
825 bool was_enabled = old->port_no != htons(OFPP_NONE);
826 bool now_enabled = new->port_no != htons(OFPP_NONE);
827 uint32_t curr = ntohl(new->curr);
828 uint32_t supported = ntohl(new->supported);
831 if (((old->config != new->config) || (old->state != new->state))
832 && opp_differs(old, new) == 1) {
833 /* Don't care if only flags changed. */
838 ds_put_format(&ds, "\"%s\", "ETH_ADDR_FMT, new->name,
839 ETH_ADDR_ARGS(new->hw_addr));
841 put_features(&ds, ", current", curr);
844 put_features(&ds, ", supports", supported);
846 if (was_enabled != now_enabled) {
848 VLOG_DBG("Port %d added: %s", port_no, ds_cstr(&ds));
850 VLOG_DBG("Port %d deleted", port_no);
853 VLOG_DBG("Port %d changed: %s", port_no, ds_cstr(&ds));
860 port_watcher_register_callback(struct port_watcher *pw,
861 port_changed_cb_func *port_changed,
864 assert(pw->n_cbs < ARRAY_SIZE(pw->cbs));
865 pw->cbs[pw->n_cbs].port_changed = port_changed;
866 pw->cbs[pw->n_cbs].aux = aux;
871 port_watcher_get_config(const struct port_watcher *pw, int port_no)
873 int idx = port_no_to_pw_idx(port_no);
874 return idx >= 0 ? ntohl(pw->ports[idx].config) : 0;
878 port_watcher_set_flags(struct port_watcher *pw, int port_no,
879 uint32_t config, uint32_t c_mask,
880 uint32_t state, uint32_t s_mask)
882 struct ofp_phy_port old;
883 struct ofp_phy_port *p;
884 struct ofp_port_mod *opm;
885 struct ofp_port_status *ops;
889 idx = port_no_to_pw_idx(port_no);
895 if (!((ntohl(p->state) ^ state) & s_mask)
896 && (!((ntohl(p->config) ^ config) & c_mask))) {
901 /* Update our idea of the flags. */
902 p->config = htonl((ntohl(p->config) & ~c_mask) | (config & c_mask));
903 p->state = htonl((ntohl(p->state) & ~s_mask) | (state & s_mask));
904 call_port_changed_callbacks(pw, port_no, &old, p);
906 /* Change the flags in the datapath. */
907 opm = make_openflow(sizeof *opm, OFPT_PORT_MOD, &b);
908 opm->port_no = p->port_no;
909 memcpy(opm->hw_addr, p->hw_addr, OFP_ETH_ALEN);
910 opm->config = p->config;
911 opm->mask = htonl(c_mask);
912 opm->advertise = htonl(0);
913 rconn_send(pw->local_rconn, b, NULL);
915 /* Notify the controller that the flags changed. */
916 ops = make_openflow(sizeof *ops, OFPT_PORT_STATUS, &b);
917 ops->reason = OFPPR_MODIFY;
919 rconn_send(pw->remote_rconn, b, NULL);
923 port_watcher_is_ready(const struct port_watcher *pw)
925 return pw->got_feature_reply;
929 port_watcher_create(struct rconn *local_rconn, struct rconn *remote_rconn,
930 struct port_watcher **pwp)
932 struct port_watcher *pw;
935 pw = *pwp = xcalloc(1, sizeof *pw);
936 pw->local_rconn = local_rconn;
937 pw->remote_rconn = remote_rconn;
938 pw->last_feature_request = TIME_MIN;
939 for (i = 0; i < OFPP_MAX; i++) {
940 pw->ports[i].port_no = htons(OFPP_NONE);
942 port_watcher_register_callback(pw, log_port_status, NULL);
943 return make_hook(port_watcher_local_packet_cb,
944 port_watcher_remote_packet_cb,
945 port_watcher_periodic_cb, NULL, pw);
948 /* Spanning tree protocol. */
950 /* Extra time, in seconds, at boot before going into fail-open, to give the
951 * spanning tree protocol time to figure out the network layout. */
952 #define STP_EXTRA_BOOT_TIME 30
956 struct port_watcher *pw;
957 struct rconn *local_rconn;
958 struct rconn *remote_rconn;
959 uint8_t dpid[ETH_ADDR_LEN];
960 long long int last_tick_256ths;
965 stp_local_packet_cb(struct relay *r, void *stp_)
967 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
968 struct ofp_header *oh;
969 struct stp_data *stp = stp_;
970 struct ofp_packet_in *opi;
971 struct eth_header *eth;
972 struct llc_header *llc;
973 struct ofpbuf payload;
978 if (oh->type == OFPT_FEATURES_REPLY
979 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
980 struct ofp_switch_features *osf = msg->data;
981 osf->capabilities |= htonl(OFPC_STP);
985 if (!get_ofp_packet_eth_header(r, &opi, ð)
986 || !eth_addr_equals(eth->eth_dst, stp_eth_addr)) {
990 port_no = ntohs(opi->in_port);
991 if (port_no >= STP_MAX_PORTS) {
992 /* STP only supports 255 ports. */
995 if (port_watcher_get_config(stp->pw, port_no) & OFPPC_NO_STP) {
996 /* We're not doing STP on this port. */
1000 if (opi->reason == OFPR_ACTION) {
1001 /* The controller set up a flow for this, so we won't intercept it. */
1005 get_ofp_packet_payload(opi, &payload);
1006 flow_extract(&payload, port_no, &flow);
1007 if (flow.dl_type != htons(OFP_DL_TYPE_NOT_ETH_TYPE)) {
1008 VLOG_DBG("non-LLC frame received on STP multicast address");
1011 llc = ofpbuf_at_assert(&payload, sizeof *eth, sizeof *llc);
1012 if (llc->llc_dsap != STP_LLC_DSAP) {
1013 VLOG_DBG("bad DSAP 0x%02"PRIx8" received on STP multicast address",
1018 /* Trim off padding on payload. */
1019 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1020 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1022 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1023 struct stp_port *p = stp_get_port(stp->stp, port_no);
1024 stp_received_bpdu(p, payload.data, payload.size);
1030 static long long int
1033 return time_msec() * 256 / 1000;
1037 stp_periodic_cb(void *stp_)
1039 struct stp_data *stp = stp_;
1040 long long int now_256ths = time_256ths();
1041 long long int elapsed_256ths = now_256ths - stp->last_tick_256ths;
1044 if (!port_watcher_is_ready(stp->pw)) {
1045 /* Can't start STP until we know port flags, because port flags can
1049 if (elapsed_256ths <= 0) {
1053 stp_tick(stp->stp, MIN(INT_MAX, elapsed_256ths));
1054 stp->last_tick_256ths = now_256ths;
1056 while (stp_get_changed_port(stp->stp, &p)) {
1057 int port_no = stp_port_no(p);
1058 enum stp_state s_state = stp_port_get_state(p);
1060 if (s_state != STP_DISABLED) {
1061 VLOG_WARN("STP: Port %d entered %s state",
1062 port_no, stp_state_name(s_state));
1064 if (!(port_watcher_get_config(stp->pw, port_no) & OFPPC_NO_STP)) {
1065 uint32_t p_config = 0;
1069 p_state = OFPPS_STP_LISTEN;
1072 p_state = OFPPS_STP_LEARN;
1075 case STP_FORWARDING:
1076 p_state = OFPPS_STP_FORWARD;
1079 p_state = OFPPS_STP_BLOCK;
1082 VLOG_DBG_RL(&vrl, "STP: Port %d has bad state %x",
1084 p_state = OFPPS_STP_FORWARD;
1087 if (!stp_forward_in_state(s_state)) {
1088 p_config = OFPPC_NO_FLOOD;
1090 port_watcher_set_flags(stp->pw, port_no,
1091 p_config, OFPPC_NO_FLOOD,
1092 p_state, OFPPS_STP_MASK);
1094 /* We don't own those flags. */
1100 stp_wait_cb(void *stp_ UNUSED)
1102 poll_timer_wait(1000);
1106 send_bpdu(const void *bpdu, size_t bpdu_size, int port_no, void *stp_)
1108 struct stp_data *stp = stp_;
1109 struct eth_header *eth;
1110 struct llc_header *llc;
1111 struct ofpbuf pkt, *opo;
1113 /* Packet skeleton. */
1114 ofpbuf_init(&pkt, ETH_HEADER_LEN + LLC_HEADER_LEN + bpdu_size);
1115 eth = ofpbuf_put_uninit(&pkt, sizeof *eth);
1116 llc = ofpbuf_put_uninit(&pkt, sizeof *llc);
1117 ofpbuf_put(&pkt, bpdu, bpdu_size);
1120 memcpy(eth->eth_dst, stp_eth_addr, ETH_ADDR_LEN);
1121 memcpy(eth->eth_src, stp->pw->ports[port_no].hw_addr, ETH_ADDR_LEN);
1122 eth->eth_type = htons(pkt.size - ETH_HEADER_LEN);
1125 llc->llc_dsap = STP_LLC_DSAP;
1126 llc->llc_ssap = STP_LLC_SSAP;
1127 llc->llc_cntl = STP_LLC_CNTL;
1129 opo = make_unbuffered_packet_out(&pkt, OFPP_NONE, port_no);
1130 ofpbuf_uninit(&pkt);
1131 rconn_send_with_limit(stp->local_rconn, opo, &stp->n_txq, OFPP_MAX);
1135 stp_is_port_supported(uint16_t port_no)
1137 /* We should be able to support STP on all possible OpenFlow physical
1138 * ports. (But we don't support STP on OFPP_LOCAL.) */
1139 BUILD_ASSERT_DECL(STP_MAX_PORTS >= OFPP_MAX);
1140 return port_no < STP_MAX_PORTS;
1144 stp_port_changed_cb(uint16_t port_no,
1145 const struct ofp_phy_port *old,
1146 const struct ofp_phy_port *new,
1149 struct stp_data *stp = stp_;
1152 if (!stp_is_port_supported(port_no)) {
1156 p = stp_get_port(stp->stp, port_no);
1157 if (new->port_no == htons(OFPP_NONE)
1158 || new->config & htonl(OFPPC_NO_STP | OFPPC_PORT_DOWN)
1159 || new->state & htonl(OFPPS_LINK_DOWN)) {
1160 stp_port_disable(p);
1164 if (new->curr & (OFPPF_10MB_HD | OFPPF_10MB_FD)) {
1166 } else if (new->curr & (OFPPF_100MB_HD | OFPPF_100MB_FD)) {
1168 } else if (new->curr & (OFPPF_1GB_HD | OFPPF_1GB_FD)) {
1170 } else if (new->curr & OFPPF_100MB_FD) {
1173 stp_port_set_speed(p, speed);
1178 stp_hook_create(const struct settings *s, struct port_watcher *pw,
1179 struct rconn *local, struct rconn *remote)
1181 uint8_t dpid[ETH_ADDR_LEN];
1182 struct netdev *netdev;
1183 struct stp_data *stp;
1186 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1188 ofp_fatal(retval, "Could not open %s device", s->of_name);
1190 memcpy(dpid, netdev_get_etheraddr(netdev), ETH_ADDR_LEN);
1191 netdev_close(netdev);
1193 stp = xcalloc(1, sizeof *stp);
1194 stp->stp = stp_create("stp", eth_addr_to_uint64(dpid), send_bpdu, stp);
1196 memcpy(stp->dpid, dpid, ETH_ADDR_LEN);
1197 stp->local_rconn = local;
1198 stp->remote_rconn = remote;
1199 stp->last_tick_256ths = time_256ths();
1201 port_watcher_register_callback(pw, stp_port_changed_cb, stp);
1202 return make_hook(stp_local_packet_cb, NULL,
1203 stp_periodic_cb, stp_wait_cb, stp);
1206 /* In-band control. */
1208 struct in_band_data {
1209 const struct settings *s;
1210 struct mac_learning *ml;
1211 struct netdev *of_device;
1212 struct rconn *controller;
1213 uint8_t mac[ETH_ADDR_LEN];
1218 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct ofpbuf *b)
1220 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
1223 static const uint8_t *
1224 get_controller_mac(struct in_band_data *in_band)
1226 static uint32_t ip, last_nonzero_ip;
1227 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
1228 static time_t next_refresh = 0;
1230 uint32_t last_ip = ip;
1232 time_t now = time_now();
1234 ip = rconn_get_ip(in_band->controller);
1235 if (last_ip != ip || !next_refresh || now >= next_refresh) {
1238 /* Look up MAC address. */
1239 memset(mac, 0, sizeof mac);
1241 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
1243 VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s",
1244 IP_ARGS(&ip), strerror(retval));
1247 have_mac = !eth_addr_is_zero(mac);
1249 /* Log changes in IP, MAC addresses. */
1250 if (ip && ip != last_nonzero_ip) {
1251 VLOG_DBG("controller IP address changed from "IP_FMT
1252 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
1253 last_nonzero_ip = ip;
1255 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
1256 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
1258 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
1259 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
1262 /* Schedule next refresh.
1264 * If we have an IP address but not a MAC address, then refresh
1265 * quickly, since we probably will get a MAC address soon (via ARP).
1266 * Otherwise, we can afford to wait a little while. */
1267 next_refresh = now + (!ip || have_mac ? 10 : 1);
1269 return !eth_addr_is_zero(mac) ? mac : NULL;
1273 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
1274 struct in_band_data *in_band)
1276 const uint8_t *mac = get_controller_mac(in_band);
1277 return mac && eth_addr_equals(mac, dl_addr);
1281 in_band_learn_mac(struct in_band_data *in_band,
1282 uint16_t in_port, const uint8_t src_mac[ETH_ADDR_LEN])
1284 if (mac_learning_learn(in_band->ml, src_mac, in_port)) {
1285 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
1286 ETH_ADDR_ARGS(src_mac), in_port);
1291 in_band_local_packet_cb(struct relay *r, void *in_band_)
1293 struct in_band_data *in_band = in_band_;
1294 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
1295 struct ofp_packet_in *opi;
1296 struct eth_header *eth;
1297 struct ofpbuf payload;
1302 if (!get_ofp_packet_eth_header(r, &opi, ð)) {
1305 in_port = ntohs(opi->in_port);
1307 /* Deal with local stuff. */
1308 if (in_port == OFPP_LOCAL) {
1309 /* Sent by secure channel. */
1310 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1311 } else if (eth_addr_equals(eth->eth_dst, in_band->mac)) {
1312 /* Sent to secure channel. */
1313 out_port = OFPP_LOCAL;
1314 in_band_learn_mac(in_band, in_port, eth->eth_src);
1315 } else if (eth->eth_type == htons(ETH_TYPE_ARP)
1316 && eth_addr_is_broadcast(eth->eth_dst)
1317 && is_controller_mac(eth->eth_src, in_band)) {
1318 /* ARP sent by controller. */
1319 out_port = OFPP_FLOOD;
1320 } else if (is_controller_mac(eth->eth_dst, in_band)
1321 || is_controller_mac(eth->eth_src, in_band)) {
1322 /* Traffic to or from controller. Switch it by hand. */
1323 in_band_learn_mac(in_band, in_port, eth->eth_src);
1324 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1326 const uint8_t *controller_mac;
1327 controller_mac = get_controller_mac(in_band);
1328 if (eth->eth_type == htons(ETH_TYPE_ARP)
1329 && eth_addr_is_broadcast(eth->eth_dst)
1330 && is_controller_mac(eth->eth_src, in_band)) {
1331 /* ARP sent by controller. */
1332 out_port = OFPP_FLOOD;
1333 } else if (is_controller_mac(eth->eth_dst, in_band)
1334 && in_port == mac_learning_lookup(in_band->ml,
1336 /* Drop controller traffic that arrives on the controller port. */
1343 get_ofp_packet_payload(opi, &payload);
1344 flow_extract(&payload, in_port, &flow);
1345 if (in_port == out_port) {
1346 /* The input and output port match. Set up a flow to drop packets. */
1347 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
1348 in_band->s->max_idle, 0));
1349 } else if (out_port != OFPP_FLOOD) {
1350 /* The output port is known, so add a new flow. */
1351 queue_tx(rc, in_band,
1352 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
1353 out_port, in_band->s->max_idle));
1355 /* If the switch didn't buffer the packet, we need to send a copy. */
1356 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1357 queue_tx(rc, in_band,
1358 make_unbuffered_packet_out(&payload, in_port, out_port));
1361 /* We don't know that MAC. Send along the packet without setting up a
1364 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1365 b = make_unbuffered_packet_out(&payload, in_port, out_port);
1367 b = make_buffered_packet_out(ntohl(opi->buffer_id),
1370 queue_tx(rc, in_band, b);
1376 in_band_status_cb(struct status_reply *sr, void *in_band_)
1378 struct in_band_data *in_band = in_band_;
1379 struct in_addr local_ip;
1380 uint32_t controller_ip;
1381 const uint8_t *controller_mac;
1383 if (netdev_get_in4(in_band->of_device, &local_ip)) {
1384 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
1386 status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
1387 ETH_ADDR_ARGS(in_band->mac));
1389 controller_ip = rconn_get_ip(in_band->controller);
1390 if (controller_ip) {
1391 status_reply_put(sr, "controller-ip="IP_FMT,
1392 IP_ARGS(&controller_ip));
1394 controller_mac = get_controller_mac(in_band);
1395 if (controller_mac) {
1396 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
1397 ETH_ADDR_ARGS(controller_mac));
1402 get_ofp_packet_payload(struct ofp_packet_in *opi, struct ofpbuf *payload)
1404 payload->data = opi->data;
1405 payload->size = ntohs(opi->header.length) - offsetof(struct ofp_packet_in,
1410 in_band_hook_create(const struct settings *s, struct switch_status *ss,
1411 struct rconn *remote)
1413 struct in_band_data *in_band;
1416 in_band = xcalloc(1, sizeof *in_band);
1418 in_band->ml = mac_learning_create();
1419 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE,
1420 &in_band->of_device);
1422 ofp_fatal(retval, "Could not open %s device", s->of_name);
1424 memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device),
1426 in_band->controller = remote;
1427 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
1428 return make_hook(in_band_local_packet_cb, NULL, NULL, NULL, in_band);
1431 /* Fail open support. */
1433 struct fail_open_data {
1434 const struct settings *s;
1435 struct rconn *local_rconn;
1436 struct rconn *remote_rconn;
1437 struct lswitch *lswitch;
1438 int last_disconn_secs;
1439 time_t boot_deadline;
1442 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
1444 fail_open_periodic_cb(void *fail_open_)
1446 struct fail_open_data *fail_open = fail_open_;
1450 if (time_now() < fail_open->boot_deadline) {
1453 disconn_secs = rconn_disconnected_duration(fail_open->remote_rconn);
1454 open = disconn_secs >= fail_open->s->probe_interval * 3;
1455 if (open != (fail_open->lswitch != NULL)) {
1457 VLOG_WARN("No longer in fail-open mode");
1458 lswitch_destroy(fail_open->lswitch);
1459 fail_open->lswitch = NULL;
1461 VLOG_WARN("Could not connect to controller for %d seconds, "
1462 "failing open", disconn_secs);
1463 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
1464 fail_open->s->max_idle);
1465 fail_open->last_disconn_secs = disconn_secs;
1467 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
1468 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
1469 "from controller", disconn_secs);
1470 fail_open->last_disconn_secs = disconn_secs;
1475 fail_open_local_packet_cb(struct relay *r, void *fail_open_)
1477 struct fail_open_data *fail_open = fail_open_;
1478 if (!fail_open->lswitch) {
1481 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
1482 r->halves[HALF_LOCAL].rxbuf);
1483 rconn_run(fail_open->local_rconn);
1489 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
1491 struct fail_open_data *fail_open = fail_open_;
1492 const struct settings *s = fail_open->s;
1493 int trigger_duration = s->probe_interval * 3;
1494 int cur_duration = rconn_disconnected_duration(fail_open->remote_rconn);
1496 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
1497 status_reply_put(sr, "current-duration=%d", cur_duration);
1498 status_reply_put(sr, "triggered=%s",
1499 cur_duration >= trigger_duration ? "true" : "false");
1500 status_reply_put(sr, "max-idle=%d", s->max_idle);
1504 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
1505 struct rconn *local_rconn, struct rconn *remote_rconn)
1507 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
1509 fail_open->local_rconn = local_rconn;
1510 fail_open->remote_rconn = remote_rconn;
1511 fail_open->lswitch = NULL;
1512 fail_open->boot_deadline = time_now() + s->probe_interval * 3;
1513 if (s->enable_stp) {
1514 fail_open->boot_deadline += STP_EXTRA_BOOT_TIME;
1516 switch_status_register_category(ss, "fail-open",
1517 fail_open_status_cb, fail_open);
1518 return make_hook(fail_open_local_packet_cb, NULL,
1519 fail_open_periodic_cb, NULL, fail_open);
1522 struct rate_limiter {
1523 const struct settings *s;
1524 struct rconn *remote_rconn;
1526 /* One queue per physical port. */
1527 struct ofp_queue queues[OFPP_MAX];
1528 int n_queued; /* Sum over queues[*].n. */
1529 int next_tx_port; /* Next port to check in round-robin. */
1533 * It costs 1000 tokens to send a single packet_in message. A single token
1534 * per message would be more straightforward, but this choice lets us avoid
1535 * round-off error in refill_bucket()'s calculation of how many tokens to
1536 * add to the bucket, since no division step is needed. */
1537 long long int last_fill; /* Time at which we last added tokens. */
1538 int tokens; /* Current number of tokens. */
1540 /* Transmission queue. */
1541 int n_txq; /* No. of packets waiting in rconn for tx. */
1543 /* Statistics reporting. */
1544 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
1545 unsigned long long n_limited; /* # queued for rate limiting. */
1546 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
1547 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
1550 /* Drop a packet from the longest queue in 'rl'. */
1552 drop_packet(struct rate_limiter *rl)
1554 struct ofp_queue *longest; /* Queue currently selected as longest. */
1555 int n_longest; /* # of queues of same length as 'longest'. */
1556 struct ofp_queue *q;
1558 longest = &rl->queues[0];
1560 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
1561 if (longest->n < q->n) {
1564 } else if (longest->n == q->n) {
1567 /* Randomly select one of the longest queues, with a uniform
1568 * distribution (Knuth algorithm 3.4.2R). */
1569 if (!random_range(n_longest)) {
1575 /* FIXME: do we want to pop the tail instead? */
1576 ofpbuf_delete(queue_pop_head(longest));
1580 /* Remove and return the next packet to transmit (in round-robin order). */
1581 static struct ofpbuf *
1582 dequeue_packet(struct rate_limiter *rl)
1586 for (i = 0; i < OFPP_MAX; i++) {
1587 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
1588 struct ofp_queue *q = &rl->queues[port];
1590 rl->next_tx_port = (port + 1) % OFPP_MAX;
1592 return queue_pop_head(q);
1598 /* Add tokens to the bucket based on elapsed time. */
1600 refill_bucket(struct rate_limiter *rl)
1602 const struct settings *s = rl->s;
1603 long long int now = time_msec();
1604 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
1605 if (tokens >= 1000) {
1606 rl->last_fill = now;
1607 rl->tokens = MIN(tokens, s->burst_limit * 1000);
1611 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
1612 * true if successful, false otherwise. (In the latter case no tokens are
1615 get_token(struct rate_limiter *rl)
1617 if (rl->tokens >= 1000) {
1626 rate_limit_local_packet_cb(struct relay *r, void *rl_)
1628 struct rate_limiter *rl = rl_;
1629 const struct settings *s = rl->s;
1630 struct ofp_packet_in *opi;
1632 opi = get_ofp_packet_in(r);
1637 if (!rl->n_queued && get_token(rl)) {
1638 /* In the common case where we are not constrained by the rate limit,
1639 * let the packet take the normal path. */
1643 /* Otherwise queue it up for the periodic callback to drain out. */
1644 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
1645 int port = ntohs(opi->in_port) % OFPP_MAX;
1646 if (rl->n_queued >= s->burst_limit) {
1649 queue_push_tail(&rl->queues[port], ofpbuf_clone(msg));
1657 rate_limit_status_cb(struct status_reply *sr, void *rl_)
1659 struct rate_limiter *rl = rl_;
1661 status_reply_put(sr, "normal=%llu", rl->n_normal);
1662 status_reply_put(sr, "limited=%llu", rl->n_limited);
1663 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
1664 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
1668 rate_limit_periodic_cb(void *rl_)
1670 struct rate_limiter *rl = rl_;
1673 /* Drain some packets out of the bucket if possible, but limit the number
1674 * of iterations to allow other code to get work done too. */
1676 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
1677 /* Use a small, arbitrary limit for the amount of queuing to do here,
1678 * because the TCP connection is responsible for buffering and there is
1679 * no point in trying to transmit faster than the TCP connection can
1681 struct ofpbuf *b = dequeue_packet(rl);
1682 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
1689 rate_limit_wait_cb(void *rl_)
1691 struct rate_limiter *rl = rl_;
1693 if (rl->tokens >= 1000) {
1694 /* We can transmit more packets as soon as we're called again. */
1695 poll_immediate_wake();
1697 /* We have to wait for the bucket to re-fill. We could calculate
1698 * the exact amount of time here for increased smoothness. */
1699 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
1705 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
1706 struct rconn *local, struct rconn *remote)
1708 struct rate_limiter *rl;
1711 rl = xcalloc(1, sizeof *rl);
1713 rl->remote_rconn = remote;
1714 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
1715 queue_init(&rl->queues[i]);
1717 rl->last_fill = time_msec();
1718 rl->tokens = s->rate_limit * 100;
1719 switch_status_register_category(ss, "rate-limit",
1720 rate_limit_status_cb, rl);
1721 return make_hook(rate_limit_local_packet_cb, NULL, rate_limit_periodic_cb,
1722 rate_limit_wait_cb, rl);
1725 /* OFPST_SWITCH statistics. */
1727 struct switch_status_category {
1729 void (*cb)(struct status_reply *, void *aux);
1733 struct switch_status {
1734 const struct settings *s;
1736 struct switch_status_category categories[8];
1740 struct status_reply {
1741 struct switch_status_category *category;
1747 switch_status_remote_packet_cb(struct relay *r, void *ss_)
1749 struct switch_status *ss = ss_;
1750 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
1751 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
1752 struct switch_status_category *c;
1753 struct nicira_header *request;
1754 struct nicira_header *reply;
1755 struct status_reply sr;
1759 if (msg->size < sizeof(struct nicira_header)) {
1762 request = msg->data;
1763 if (request->header.type != OFPT_VENDOR
1764 || request->vendor_id != htonl(NX_VENDOR_ID)
1765 || request->subtype != htonl(NXT_STATUS_REQUEST)) {
1769 sr.request.string = (void *) (request + 1);
1770 sr.request.length = msg->size - sizeof *request;
1771 ds_init(&sr.output);
1772 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
1773 if (!memcmp(c->name, sr.request.string,
1774 MIN(strlen(c->name), sr.request.length))) {
1779 reply = make_openflow_xid(sizeof *reply + sr.output.length,
1780 OFPT_VENDOR, request->header.xid, &b);
1781 reply->vendor_id = htonl(NX_VENDOR_ID);
1782 reply->subtype = htonl(NXT_STATUS_REPLY);
1783 memcpy(reply + 1, sr.output.string, sr.output.length);
1784 retval = rconn_send(rc, b, NULL);
1785 if (retval && retval != EAGAIN) {
1786 VLOG_WARN("send failed (%s)", strerror(retval));
1788 ds_destroy(&sr.output);
1793 rconn_status_cb(struct status_reply *sr, void *rconn_)
1795 struct rconn *rconn = rconn_;
1796 time_t now = time_now();
1798 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
1799 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
1800 status_reply_put(sr, "backoff=%d", rconn_get_backoff(rconn));
1801 status_reply_put(sr, "is-connected=%s",
1802 rconn_is_connected(rconn) ? "true" : "false");
1803 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
1804 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
1805 status_reply_put(sr, "attempted-connections=%u",
1806 rconn_get_attempted_connections(rconn));
1807 status_reply_put(sr, "successful-connections=%u",
1808 rconn_get_successful_connections(rconn));
1809 status_reply_put(sr, "last-connection=%ld",
1810 (long int) (now - rconn_get_last_connection(rconn)));
1811 status_reply_put(sr, "time-connected=%lu",
1812 rconn_get_total_time_connected(rconn));
1813 status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(rconn));
1817 config_status_cb(struct status_reply *sr, void *s_)
1819 const struct settings *s = s_;
1822 for (i = 0; i < s->n_listeners; i++) {
1823 status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
1825 if (s->probe_interval) {
1826 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
1828 if (s->max_backoff) {
1829 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
1834 switch_status_cb(struct status_reply *sr, void *ss_)
1836 struct switch_status *ss = ss_;
1837 time_t now = time_now();
1839 status_reply_put(sr, "now=%ld", (long int) now);
1840 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
1841 status_reply_put(sr, "pid=%ld", (long int) getpid());
1845 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
1847 struct switch_status *ss = xcalloc(1, sizeof *ss);
1849 ss->booted = time_now();
1850 switch_status_register_category(ss, "config",
1851 config_status_cb, (void *) s);
1852 switch_status_register_category(ss, "switch", switch_status_cb, ss);
1854 return make_hook(NULL, switch_status_remote_packet_cb, NULL, NULL, ss);
1858 switch_status_register_category(struct switch_status *ss,
1859 const char *category,
1860 void (*cb)(struct status_reply *,
1864 struct switch_status_category *c;
1865 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
1866 c = &ss->categories[ss->n_categories++];
1869 c->name = xstrdup(category);
1873 status_reply_put(struct status_reply *sr, const char *content, ...)
1875 size_t old_length = sr->output.length;
1879 /* Append the status reply to the output. */
1880 ds_put_format(&sr->output, "%s.", sr->category->name);
1881 va_start(args, content);
1882 ds_put_format_valist(&sr->output, content, args);
1884 if (ds_last(&sr->output) != '\n') {
1885 ds_put_char(&sr->output, '\n');
1888 /* Drop what we just added if it doesn't match the request. */
1889 added = sr->output.length - old_length;
1890 if (added < sr->request.length
1891 || memcmp(&sr->output.string[old_length],
1892 sr->request.string, sr->request.length)) {
1893 ds_truncate(&sr->output, old_length);
1898 /* Controller discovery. */
1902 const struct settings *s;
1903 struct dhclient *dhcp;
1908 discovery_status_cb(struct status_reply *sr, void *d_)
1910 struct discovery *d = d_;
1912 status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re);
1913 status_reply_put(sr, "n-changes=%d", d->n_changes);
1914 status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp));
1915 status_reply_put(sr, "state-elapsed=%u",
1916 dhclient_get_state_elapsed(d->dhcp));
1917 if (dhclient_is_bound(d->dhcp)) {
1918 uint32_t ip = dhclient_get_ip(d->dhcp);
1919 uint32_t netmask = dhclient_get_netmask(d->dhcp);
1920 uint32_t router = dhclient_get_router(d->dhcp);
1922 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
1923 uint32_t dns_server;
1927 status_reply_put(sr, "ip="IP_FMT, IP_ARGS(&ip));
1928 status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask));
1930 status_reply_put(sr, "router="IP_FMT, IP_ARGS(&router));
1933 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i, &dns_server);
1935 status_reply_put(sr, "dns%d="IP_FMT, i, IP_ARGS(&dns_server));
1938 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
1940 status_reply_put(sr, "domain=%s", domain_name);
1944 status_reply_put(sr, "lease-remaining=%u",
1945 dhclient_get_lease_remaining(d->dhcp));
1949 static struct discovery *
1950 discovery_init(const struct settings *s, struct switch_status *ss)
1952 struct netdev *netdev;
1953 struct discovery *d;
1954 struct dhclient *dhcp;
1957 /* Bring ofX network device up. */
1958 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1960 ofp_fatal(retval, "Could not open %s device", s->of_name);
1962 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
1964 ofp_fatal(retval, "Could not bring %s device up", s->of_name);
1966 netdev_close(netdev);
1968 /* Initialize DHCP client. */
1969 retval = dhclient_create(s->of_name, modify_dhcp_request,
1970 validate_dhcp_offer, (void *) s, &dhcp);
1972 ofp_fatal(retval, "Failed to initialize DHCP client");
1974 dhclient_init(dhcp, 0);
1976 d = xmalloc(sizeof *d);
1981 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
1987 discovery_question_connectivity(struct discovery *d)
1989 dhclient_force_renew(d->dhcp, 15);
1993 discovery_run(struct discovery *d, char **controller_name)
1995 dhclient_run(d->dhcp);
1996 if (!dhclient_changed(d->dhcp)) {
2000 dhclient_configure_netdev(d->dhcp);
2001 if (d->s->update_resolv_conf) {
2002 dhclient_update_resolv_conf(d->dhcp);
2005 if (dhclient_is_bound(d->dhcp)) {
2006 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
2007 DHCP_CODE_OFP_CONTROLLER_VCONN);
2008 VLOG_WARN("%s: discovered controller", *controller_name);
2011 *controller_name = NULL;
2013 VLOG_WARN("discovered controller no longer available");
2021 discovery_wait(struct discovery *d)
2023 dhclient_wait(d->dhcp);
2027 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
2029 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
2033 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
2035 const struct settings *s = s_;
2039 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
2041 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
2044 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
2046 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
2047 s->accept_controller_re);
2053 /* User interface. */
2056 parse_options(int argc, char *argv[], struct settings *s)
2059 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
2061 OPT_INACTIVITY_PROBE,
2066 OPT_BOOTSTRAP_CA_CERT,
2070 static struct option long_options[] = {
2071 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
2072 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
2073 {"fail", required_argument, 0, 'F'},
2074 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
2075 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
2076 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
2077 {"listen", required_argument, 0, 'l'},
2078 {"monitor", required_argument, 0, 'm'},
2079 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
2080 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
2081 {"stp", no_argument, 0, OPT_STP},
2082 {"no-stp", no_argument, 0, OPT_NO_STP},
2083 {"detach", no_argument, 0, 'D'},
2084 {"force", no_argument, 0, 'f'},
2085 {"pidfile", optional_argument, 0, 'P'},
2086 {"verbose", optional_argument, 0, 'v'},
2087 {"help", no_argument, 0, 'h'},
2088 {"version", no_argument, 0, 'V'},
2090 VCONN_SSL_LONG_OPTIONS
2091 {"bootstrap-ca-cert", required_argument, 0, OPT_BOOTSTRAP_CA_CERT},
2095 char *short_options = long_options_to_short_options(long_options);
2096 char *accept_re = NULL;
2099 /* Set defaults that we can figure out before parsing options. */
2101 s->monitor_name = NULL;
2102 s->fail_mode = FAIL_OPEN;
2104 s->probe_interval = 15;
2105 s->max_backoff = 15;
2106 s->update_resolv_conf = true;
2109 s->enable_stp = false;
2113 c = getopt_long(argc, argv, short_options, long_options, NULL);
2119 case OPT_ACCEPT_VCONN:
2120 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
2123 case OPT_NO_RESOLV_CONF:
2124 s->update_resolv_conf = false;
2128 if (!strcmp(optarg, "open")) {
2129 s->fail_mode = FAIL_OPEN;
2130 } else if (!strcmp(optarg, "closed")) {
2131 s->fail_mode = FAIL_CLOSED;
2133 ofp_fatal(0, "-f or --fail argument must be \"open\" "
2138 case OPT_INACTIVITY_PROBE:
2139 s->probe_interval = atoi(optarg);
2140 if (s->probe_interval < 5) {
2141 ofp_fatal(0, "--inactivity-probe argument must be at least 5");
2146 if (!strcmp(optarg, "permanent")) {
2147 s->max_idle = OFP_FLOW_PERMANENT;
2149 s->max_idle = atoi(optarg);
2150 if (s->max_idle < 1 || s->max_idle > 65535) {
2151 ofp_fatal(0, "--max-idle argument must be between 1 and "
2152 "65535 or the word 'permanent'");
2157 case OPT_MAX_BACKOFF:
2158 s->max_backoff = atoi(optarg);
2159 if (s->max_backoff < 1) {
2160 ofp_fatal(0, "--max-backoff argument must be at least 1");
2161 } else if (s->max_backoff > 3600) {
2162 s->max_backoff = 3600;
2166 case OPT_RATE_LIMIT:
2168 s->rate_limit = atoi(optarg);
2169 if (s->rate_limit < 1) {
2170 ofp_fatal(0, "--rate-limit argument must be at least 1");
2173 s->rate_limit = 1000;
2177 case OPT_BURST_LIMIT:
2178 s->burst_limit = atoi(optarg);
2179 if (s->burst_limit < 1) {
2180 ofp_fatal(0, "--burst-limit argument must be at least 1");
2185 s->enable_stp = true;
2189 s->enable_stp = false;
2197 set_pidfile(optarg);
2201 ignore_existing_pidfile();
2205 if (s->n_listeners >= MAX_MGMT) {
2207 "-l or --listen may be specified at most %d times",
2210 s->listener_names[s->n_listeners++] = optarg;
2214 if (s->monitor_name) {
2215 ofp_fatal(0, "-m or --monitor may only be specified once");
2217 s->monitor_name = optarg;
2224 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
2228 vlog_set_verbosity(optarg);
2232 VCONN_SSL_OPTION_HANDLERS
2234 case OPT_BOOTSTRAP_CA_CERT:
2235 vconn_ssl_set_ca_cert_file(optarg, true);
2246 free(short_options);
2250 if (argc < 1 || argc > 2) {
2251 ofp_fatal(0, "need one or two non-option arguments; "
2252 "use --help for usage");
2255 /* Local and remote vconns. */
2256 s->nl_name = argv[0];
2257 if (strncmp(s->nl_name, "nl:", 3)
2258 || strlen(s->nl_name) < 4
2259 || s->nl_name[strspn(s->nl_name + 3, "0123456789") + 3]) {
2260 ofp_fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"",
2263 s->of_name = xasprintf("of%s", s->nl_name + 3);
2264 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
2266 /* Set accept_controller_regex. */
2268 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
2270 retval = regcomp(&s->accept_controller_regex, accept_re,
2271 REG_NOSUB | REG_EXTENDED);
2273 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
2274 char *buffer = xmalloc(length);
2275 regerror(retval, &s->accept_controller_regex, buffer, length);
2276 ofp_fatal(0, "%s: %s", accept_re, buffer);
2278 s->accept_controller_re = accept_re;
2280 /* Mode of operation. */
2281 s->discovery = s->controller_name == NULL;
2285 enum netdev_flags flags;
2286 struct netdev *netdev;
2288 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
2290 ofp_fatal(retval, "Could not open %s device", s->of_name);
2293 retval = netdev_get_flags(netdev, &flags);
2295 ofp_fatal(retval, "Could not get flags for %s device", s->of_name);
2298 s->in_band = (flags & NETDEV_UP) != 0;
2299 if (s->in_band && netdev_get_in6(netdev, NULL)) {
2300 VLOG_WARN("Ignoring IPv6 address on %s device: IPv6 not supported",
2304 netdev_close(netdev);
2307 /* Rate limiting. */
2308 if (s->rate_limit) {
2309 if (s->rate_limit < 100) {
2310 VLOG_WARN("Rate limit set to unusually low value %d",
2313 if (!s->burst_limit) {
2314 s->burst_limit = s->rate_limit / 4;
2316 s->burst_limit = MAX(s->burst_limit, 1);
2317 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
2324 printf("%s: secure channel, a relay for OpenFlow messages.\n"
2325 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
2326 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
2327 "CONTROLLER is an active OpenFlow connection method; if it is\n"
2328 "omitted, then secchan performs controller discovery.\n",
2329 program_name, program_name);
2330 vconn_usage(true, true, true);
2331 printf("\nController discovery options:\n"
2332 " --accept-vconn=REGEX accept matching discovered controllers\n"
2333 " --no-resolv-conf do not update /etc/resolv.conf\n"
2334 "\nNetworking options:\n"
2335 " -F, --fail=open|closed when controller connection fails:\n"
2336 " closed: drop all packets\n"
2337 " open (default): act as learning switch\n"
2338 " --inactivity-probe=SECS time between inactivity probes\n"
2339 " --max-idle=SECS max idle for flows set up by secchan\n"
2340 " --max-backoff=SECS max time between controller connection\n"
2341 " attempts (default: 15 seconds)\n"
2342 " -l, --listen=METHOD allow management connections on METHOD\n"
2343 " (a passive OpenFlow connection method)\n"
2344 " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n"
2345 " (a passive OpenFlow connection method)\n"
2346 " --stp enable 802.1D Spanning Tree Protocol\n"
2347 " --no-stp disable 802.1D Spanning Tree Protocol\n"
2348 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
2349 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
2350 " --burst-limit=BURST limit on packet credit for idle time\n"
2351 "\nOther options:\n"
2352 " -D, --detach run in background as daemon\n"
2353 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
2354 " -f, --force with -P, start even if already running\n"
2355 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
2356 " -v, --verbose set maximum verbosity level\n"
2357 " -h, --help display this help message\n"
2358 " -V, --version display version information\n",