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 *dp_name; /* Local datapath. */
93 const char *controller_name; /* Controller (if not discovery mode). */
94 const char *listener_names[MAX_MGMT]; /* Listen for mgmt connections. */
95 size_t n_listeners; /* Number of mgmt connection listeners. */
96 const char *monitor_name; /* Listen for traffic monitor connections. */
98 /* Failure behavior. */
99 enum fail_mode fail_mode; /* Act as learning switch if no controller? */
100 int max_idle; /* Idle time for flows in fail-open mode. */
101 int probe_interval; /* # seconds idle before sending echo request. */
102 int max_backoff; /* Max # seconds between connection attempts. */
104 /* Packet-in rate-limiting. */
105 int rate_limit; /* Tokens added to bucket per second. */
106 int burst_limit; /* Maximum number token bucket size. */
108 /* Discovery behavior. */
109 regex_t accept_controller_regex; /* Controller vconns to accept. */
110 const char *accept_controller_re; /* String version of regex. */
111 bool update_resolv_conf; /* Update /etc/resolv.conf? */
113 /* Spanning tree protocol. */
119 struct ofpbuf *rxbuf;
120 int n_txq; /* No. of packets queued for tx on 'rconn'. */
127 #define HALF_REMOTE 1
128 struct half halves[2];
134 bool (*packet_cb[2])(struct relay *, void *aux);
135 void (*periodic_cb)(void *aux);
136 void (*wait_cb)(void *aux);
140 static struct vlog_rate_limit vrl = VLOG_RATE_LIMIT_INIT(60, 60);
142 static void parse_options(int argc, char *argv[], struct settings *);
143 static void usage(void) NO_RETURN;
145 static struct pvconn *open_passive_vconn(const char *name);
146 static struct vconn *accept_vconn(struct pvconn *pvconn);
148 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
150 static struct relay *relay_accept(const struct settings *, struct pvconn *);
151 static void relay_run(struct relay *, const struct hook[], size_t n_hooks);
152 static void relay_wait(struct relay *);
153 static void relay_destroy(struct relay *);
155 static struct hook make_hook(bool (*local_packet_cb)(struct relay *, void *),
156 bool (*remote_packet_cb)(struct relay *, void *),
157 void (*periodic_cb)(void *),
158 void (*wait_cb)(void *),
160 static struct ofp_packet_in *get_ofp_packet_in(struct relay *);
161 static bool get_ofp_packet_eth_header(struct relay *, struct ofp_packet_in **,
162 struct eth_header **);
163 static void get_ofp_packet_payload(struct ofp_packet_in *, struct ofpbuf *);
165 struct switch_status;
167 static struct hook switch_status_hook_create(const struct settings *,
168 struct switch_status **);
169 static void switch_status_register_category(struct switch_status *,
170 const char *category,
171 void (*cb)(struct status_reply *,
174 static void status_reply_put(struct status_reply *, const char *, ...)
177 static void rconn_status_cb(struct status_reply *, void *rconn_);
180 static struct discovery *discovery_init(const struct settings *,
181 struct port_watcher *,
182 struct switch_status *);
183 static void discovery_question_connectivity(struct discovery *);
184 static bool discovery_run(struct discovery *, char **controller_name);
185 static void discovery_wait(struct discovery *);
187 static struct hook in_band_hook_create(const struct settings *,
188 struct switch_status *,
189 struct port_watcher *,
190 struct rconn *remote);
192 static struct hook port_watcher_create(struct rconn *local,
193 struct rconn *remote,
194 struct port_watcher **);
195 static uint32_t port_watcher_get_config(const struct port_watcher *,
197 static void port_watcher_set_flags(struct port_watcher *, int port_no,
198 uint32_t config, uint32_t c_mask,
199 uint32_t state, uint32_t s_mask);
201 static struct hook snat_hook_create(struct port_watcher *pw);
203 static struct hook stp_hook_create(const struct settings *,
204 struct port_watcher *,
205 struct rconn *local, struct rconn *remote);
207 static struct hook fail_open_hook_create(const struct settings *,
208 struct switch_status *,
210 struct rconn *remote);
211 static struct hook rate_limit_hook_create(const struct settings *,
212 struct switch_status *,
214 struct rconn *remote);
217 static void modify_dhcp_request(struct dhcp_msg *, void *aux);
218 static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux);
221 main(int argc, char *argv[])
225 struct list relays = LIST_INITIALIZER(&relays);
227 struct hook hooks[8];
230 struct pvconn *monitor;
232 struct pvconn *listeners[MAX_MGMT];
235 struct rconn *local_rconn, *remote_rconn;
236 struct relay *controller_relay;
237 struct discovery *discovery;
238 struct switch_status *switch_status;
239 struct port_watcher *pw;
243 set_program_name(argv[0]);
244 register_fault_handlers();
247 parse_options(argc, argv, &s);
248 signal(SIGPIPE, SIG_IGN);
250 /* Start listening for management and monitoring connections. */
252 for (i = 0; i < s.n_listeners; i++) {
253 listeners[n_listeners++] = open_passive_vconn(s.listener_names[i]);
255 monitor = s.monitor_name ? open_passive_vconn(s.monitor_name) : NULL;
257 /* Initialize switch status hook. */
258 hooks[n_hooks++] = switch_status_hook_create(&s, &switch_status);
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.dp_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);
295 discovery = s.discovery ? discovery_init(&s, pw, switch_status) : NULL;
296 hooks[n_hooks++] = snat_hook_create(pw);
298 hooks[n_hooks++] = stp_hook_create(&s, pw, local_rconn, remote_rconn);
301 hooks[n_hooks++] = in_band_hook_create(&s, switch_status, pw,
304 if (s.fail_mode == FAIL_OPEN) {
305 hooks[n_hooks++] = fail_open_hook_create(&s, switch_status,
306 local_rconn, remote_rconn);
309 hooks[n_hooks++] = rate_limit_hook_create(&s, switch_status,
310 local_rconn, remote_rconn);
312 assert(n_hooks <= ARRAY_SIZE(hooks));
319 LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) {
320 relay_run(r, hooks, n_hooks);
322 for (i = 0; i < n_listeners; i++) {
324 struct relay *r = relay_accept(&s, listeners[i]);
328 list_push_back(&relays, &r->node);
332 struct vconn *new = accept_vconn(monitor);
334 rconn_add_monitor(local_rconn, new);
337 for (i = 0; i < n_hooks; i++) {
338 if (hooks[i].periodic_cb) {
339 hooks[i].periodic_cb(hooks[i].aux);
343 char *controller_name;
344 if (rconn_is_connectivity_questionable(remote_rconn)) {
345 discovery_question_connectivity(discovery);
347 if (discovery_run(discovery, &controller_name)) {
348 if (controller_name) {
349 rconn_connect(remote_rconn, controller_name);
351 rconn_disconnect(remote_rconn);
356 /* Wait for something to happen. */
357 LIST_FOR_EACH (r, struct relay, node, &relays) {
360 for (i = 0; i < n_listeners; i++) {
361 pvconn_wait(listeners[i]);
364 pvconn_wait(monitor);
366 for (i = 0; i < n_hooks; i++) {
367 if (hooks[i].wait_cb) {
368 hooks[i].wait_cb(hooks[i].aux);
372 discovery_wait(discovery);
380 static struct pvconn *
381 open_passive_vconn(const char *name)
383 struct pvconn *pvconn;
386 retval = pvconn_open(name, &pvconn);
387 if (retval && retval != EAGAIN) {
388 ofp_fatal(retval, "opening %s", name);
393 static struct vconn *
394 accept_vconn(struct pvconn *pvconn)
399 retval = pvconn_accept(pvconn, OFP_VERSION, &new);
400 if (retval && retval != EAGAIN) {
401 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
407 make_hook(bool (*local_packet_cb)(struct relay *, void *aux),
408 bool (*remote_packet_cb)(struct relay *, void *aux),
409 void (*periodic_cb)(void *aux),
410 void (*wait_cb)(void *aux),
414 h.packet_cb[HALF_LOCAL] = local_packet_cb;
415 h.packet_cb[HALF_REMOTE] = remote_packet_cb;
416 h.periodic_cb = periodic_cb;
422 static struct ofp_packet_in *
423 get_ofp_packet_in(struct relay *r)
425 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
426 struct ofp_header *oh = msg->data;
427 if (oh->type == OFPT_PACKET_IN) {
428 if (msg->size >= offsetof (struct ofp_packet_in, data)) {
431 VLOG_WARN("packet too short (%zu bytes) for packet_in",
439 get_ofp_packet_eth_header(struct relay *r, struct ofp_packet_in **opip,
440 struct eth_header **ethp)
442 const int min_len = offsetof(struct ofp_packet_in, data) + ETH_HEADER_LEN;
443 struct ofp_packet_in *opi = get_ofp_packet_in(r);
444 if (opi && ntohs(opi->header.length) >= min_len) {
446 *ethp = (void *) opi->data;
453 /* OpenFlow message relaying. */
455 static struct relay *
456 relay_accept(const struct settings *s, struct pvconn *pvconn)
458 struct vconn *new_remote, *new_local;
459 struct rconn *r1, *r2;
464 new_remote = accept_vconn(pvconn);
469 if (sscanf(s->dp_name, "nl:%d", &nl_index) == 1) {
470 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123.
471 * nl:123:0 opens a netlink connection to local datapath 123 without
472 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
473 * messages. That's what we want here; management connections should
474 * not receive those messages, at least by default. */
475 vconn_name = xasprintf("nl:%d:0", nl_index);
477 /* We don't have a way to specify not to subscribe to those messages
478 * for other transports. (That's a defect: really this should be in
479 * the OpenFlow protocol, not the Netlink transport). */
480 VLOG_WARN_RL(&vrl, "new management connection will receive "
481 "asynchronous messages");
482 vconn_name = xstrdup(s->dp_name);
485 retval = vconn_open(vconn_name, OFP_VERSION, &new_local);
487 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
488 vconn_name, strerror(retval));
489 vconn_close(new_remote);
494 /* Create and return relay. */
495 r1 = rconn_create(0, 0);
496 rconn_connect_unreliably(r1, vconn_name, new_local);
499 r2 = rconn_create(0, 0);
500 rconn_connect_unreliably(r2, "passive", new_remote);
502 return relay_create(r1, r2, true);
505 static struct relay *
506 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
508 struct relay *r = xcalloc(1, sizeof *r);
509 r->halves[HALF_LOCAL].rconn = local;
510 r->halves[HALF_REMOTE].rconn = remote;
511 r->is_mgmt_conn = is_mgmt_conn;
516 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
521 for (i = 0; i < 2; i++) {
522 rconn_run(r->halves[i].rconn);
525 /* Limit the number of iterations to prevent other tasks from starving. */
526 for (iteration = 0; iteration < 50; iteration++) {
527 bool progress = false;
528 for (i = 0; i < 2; i++) {
529 struct half *this = &r->halves[i];
530 struct half *peer = &r->halves[!i];
533 this->rxbuf = rconn_recv(this->rconn);
534 if (this->rxbuf && (i == HALF_REMOTE || !r->is_mgmt_conn)) {
535 const struct hook *h;
536 for (h = hooks; h < &hooks[n_hooks]; h++) {
537 if (h->packet_cb[i] && h->packet_cb[i](r, h->aux)) {
538 ofpbuf_delete(this->rxbuf);
547 if (this->rxbuf && !this->n_txq) {
548 int retval = rconn_send(peer->rconn, this->rxbuf,
550 if (retval != EAGAIN) {
554 ofpbuf_delete(this->rxbuf);
565 if (r->is_mgmt_conn) {
566 for (i = 0; i < 2; i++) {
567 struct half *this = &r->halves[i];
568 if (!rconn_is_alive(this->rconn)) {
577 relay_wait(struct relay *r)
581 for (i = 0; i < 2; i++) {
582 struct half *this = &r->halves[i];
584 rconn_run_wait(this->rconn);
586 rconn_recv_wait(this->rconn);
592 relay_destroy(struct relay *r)
596 list_remove(&r->node);
597 for (i = 0; i < 2; i++) {
598 struct half *this = &r->halves[i];
599 rconn_destroy(this->rconn);
600 ofpbuf_delete(this->rxbuf);
605 /* Port status watcher. */
607 typedef void port_changed_cb_func(uint16_t port_no,
608 const struct ofp_phy_port *old,
609 const struct ofp_phy_port *new,
612 struct port_watcher_cb {
613 port_changed_cb_func *port_changed;
617 typedef void local_port_changed_cb_func(const struct ofp_phy_port *new,
620 struct port_watcher_local_cb {
621 local_port_changed_cb_func *local_port_changed;
625 struct port_watcher {
626 struct rconn *local_rconn;
627 struct rconn *remote_rconn;
628 struct ofp_phy_port ports[OFPP_MAX + 1];
629 time_t last_feature_request;
630 bool got_feature_reply;
632 struct port_watcher_cb cbs[2];
634 struct port_watcher_local_cb local_cbs[2];
636 char local_port_name[OFP_MAX_PORT_NAME_LEN + 1];
639 /* Returns the number of fields that differ from 'a' to 'b'. */
641 opp_differs(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
643 BUILD_ASSERT_DECL(sizeof *a == 48); /* Trips when we add or remove fields. */
644 return ((a->port_no != b->port_no)
645 + (memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) != 0)
646 + (memcmp(a->name, b->name, sizeof a->name) != 0)
647 + (a->config != b->config)
648 + (a->state != b->state)
649 + (a->curr != b->curr)
650 + (a->advertised != b->advertised)
651 + (a->supported != b->supported)
652 + (a->peer != b->peer));
656 sanitize_opp(struct ofp_phy_port *opp)
660 for (i = 0; i < sizeof opp->name; i++) {
661 char c = opp->name[i];
662 if (c && (c < 0x20 || c > 0x7e)) {
666 opp->name[sizeof opp->name - 1] = '\0';
670 port_no_to_pw_idx(int port_no)
672 return (port_no < OFPP_MAX ? port_no
673 : port_no == OFPP_LOCAL ? OFPP_MAX
678 call_port_changed_callbacks(struct port_watcher *pw, int port_no,
679 const struct ofp_phy_port *old,
680 const struct ofp_phy_port *new)
682 if (opp_differs(old, new)) {
684 for (i = 0; i < pw->n_cbs; i++) {
685 port_changed_cb_func *port_changed = pw->cbs[i].port_changed;
686 (port_changed)(port_no, old, new, pw->cbs[i].aux);
692 get_port_name(const struct ofp_phy_port *port, char *name, size_t name_size)
696 memcpy(name, port->name, MIN(name_size, sizeof port->name));
697 name[name_size - 1] = '\0';
698 for (p = name; *p != '\0'; p++) {
699 if (*p < 32 || *p > 126) {
706 call_local_port_changed_callbacks(struct port_watcher *pw)
708 char name[OFP_MAX_PORT_NAME_LEN + 1];
709 const struct ofp_phy_port *port;
712 /* Pass the local port to the callbacks, if it exists.
713 Pass a null pointer if there is no local port. */
714 port = &pw->ports[port_no_to_pw_idx(OFPP_LOCAL)];
715 if (port->port_no != htons(OFPP_LOCAL)) {
719 /* Log the name of the local port. */
721 get_port_name(port, name, sizeof name);
725 if (strcmp(pw->local_port_name, name)) {
727 VLOG_WARN("Identified data path local port as \"%s\".", name);
729 VLOG_WARN("Data path has no local port.");
731 strcpy(pw->local_port_name, name);
734 /* Invoke callbacks. */
735 for (i = 0; i < pw->n_local_cbs; i++) {
736 local_port_changed_cb_func *cb = pw->local_cbs[i].local_port_changed;
737 (cb)(port, pw->local_cbs[i].aux);
742 update_phy_port(struct port_watcher *pw, struct ofp_phy_port *opp,
743 uint8_t reason, bool seen[OFPP_MAX + 1])
745 struct ofp_phy_port *pw_opp;
746 struct ofp_phy_port old;
750 port_no = ntohs(opp->port_no);
751 idx = port_no_to_pw_idx(port_no);
760 pw_opp = &pw->ports[idx];
762 if (reason == OFPPR_DELETE) {
763 memset(pw_opp, 0, sizeof *pw_opp);
764 pw_opp->port_no = htons(OFPP_NONE);
765 } else if (reason == OFPPR_MODIFY || reason == OFPPR_ADD) {
767 sanitize_opp(pw_opp);
769 call_port_changed_callbacks(pw, port_no, &old, pw_opp);
773 port_watcher_local_packet_cb(struct relay *r, void *pw_)
775 struct port_watcher *pw = pw_;
776 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
777 struct ofp_header *oh = msg->data;
779 if (oh->type == OFPT_FEATURES_REPLY
780 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
781 struct ofp_switch_features *osf = msg->data;
782 bool seen[ARRAY_SIZE(pw->ports)];
786 pw->got_feature_reply = true;
788 /* Update each port included in the message. */
789 memset(seen, 0, sizeof seen);
790 n_ports = ((msg->size - offsetof(struct ofp_switch_features, ports))
791 / sizeof *osf->ports);
792 for (i = 0; i < n_ports; i++) {
793 struct ofp_phy_port *opp = &osf->ports[i];
794 update_phy_port(pw, opp, OFPPR_MODIFY, seen);
797 /* Delete all the ports not included in the message. */
798 for (i = 0; i < ARRAY_SIZE(pw->ports); i++) {
800 update_phy_port(pw, &pw->ports[i], OFPPR_DELETE, NULL);
804 call_local_port_changed_callbacks(pw);
805 } else if (oh->type == OFPT_PORT_STATUS
806 && msg->size >= sizeof(struct ofp_port_status)) {
807 struct ofp_port_status *ops = msg->data;
808 update_phy_port(pw, &ops->desc, ops->reason, NULL);
809 if (ops->desc.port_no == htons(OFPP_LOCAL)) {
810 call_local_port_changed_callbacks(pw);
817 port_watcher_remote_packet_cb(struct relay *r, void *pw_)
819 struct port_watcher *pw = pw_;
820 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
821 struct ofp_header *oh = msg->data;
823 if (oh->type == OFPT_PORT_MOD
824 && msg->size >= sizeof(struct ofp_port_mod)) {
825 struct ofp_port_mod *opm = msg->data;
826 uint16_t port_no = ntohs(opm->port_no);
827 int idx = port_no_to_pw_idx(port_no);
829 struct ofp_phy_port *pw_opp = &pw->ports[idx];
830 if (pw_opp->port_no != htons(OFPP_NONE)) {
831 struct ofp_phy_port old = *pw_opp;
832 pw_opp->config = ((pw_opp->config & ~opm->mask)
833 | (opm->config & opm->mask));
834 call_port_changed_callbacks(pw, port_no, &old, pw_opp);
835 if (pw_opp->port_no == htons(OFPP_LOCAL)) {
836 call_local_port_changed_callbacks(pw);
845 port_watcher_periodic_cb(void *pw_)
847 struct port_watcher *pw = pw_;
849 if (!pw->got_feature_reply
850 && time_now() >= pw->last_feature_request + 5
851 && rconn_is_connected(pw->local_rconn)) {
853 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
854 rconn_send_with_limit(pw->local_rconn, b, &pw->n_txq, 1);
855 pw->last_feature_request = time_now();
860 port_watcher_wait_cb(void *pw_)
862 struct port_watcher *pw = pw_;
863 if (!pw->got_feature_reply && rconn_is_connected(pw->local_rconn)) {
864 if (pw->last_feature_request != TIME_MIN) {
865 poll_timer_wait(pw->last_feature_request + 5 - time_now());
867 poll_immediate_wake();
873 put_duplexes(struct ds *ds, const char *name, uint32_t features,
874 uint32_t hd_bit, uint32_t fd_bit)
876 if (features & (hd_bit | fd_bit)) {
877 ds_put_format(ds, " %s", name);
878 if (features & hd_bit) {
879 ds_put_cstr(ds, "(HD)");
881 if (features & fd_bit) {
882 ds_put_cstr(ds, "(FD)");
888 put_features(struct ds *ds, const char *name, uint32_t features)
890 if (features & (OFPPF_10MB_HD | OFPPF_10MB_FD
891 | OFPPF_100MB_HD | OFPPF_100MB_FD
892 | OFPPF_1GB_HD | OFPPF_1GB_FD | OFPPF_10GB_FD)) {
893 ds_put_cstr(ds, name);
894 put_duplexes(ds, "10M", features, OFPPF_10MB_HD, OFPPF_10MB_FD);
895 put_duplexes(ds, "100M", features,
896 OFPPF_100MB_HD, OFPPF_100MB_FD);
897 put_duplexes(ds, "1G", features, OFPPF_100MB_HD, OFPPF_100MB_FD);
898 if (features & OFPPF_10GB_FD) {
899 ds_put_cstr(ds, " 10G");
901 if (features & OFPPF_AUTONEG) {
902 ds_put_cstr(ds, " AUTO_NEG");
904 if (features & OFPPF_PAUSE) {
905 ds_put_cstr(ds, " PAUSE");
907 if (features & OFPPF_PAUSE_ASYM) {
908 ds_put_cstr(ds, " PAUSE_ASYM");
914 log_port_status(uint16_t port_no,
915 const struct ofp_phy_port *old,
916 const struct ofp_phy_port *new,
919 if (VLOG_IS_DBG_ENABLED()) {
920 bool was_enabled = old->port_no != htons(OFPP_NONE);
921 bool now_enabled = new->port_no != htons(OFPP_NONE);
922 uint32_t curr = ntohl(new->curr);
923 uint32_t supported = ntohl(new->supported);
926 if (((old->config != new->config) || (old->state != new->state))
927 && opp_differs(old, new) == 1) {
928 /* Don't care if only flags changed. */
933 ds_put_format(&ds, "\"%s\", "ETH_ADDR_FMT, new->name,
934 ETH_ADDR_ARGS(new->hw_addr));
936 put_features(&ds, ", current", curr);
939 put_features(&ds, ", supports", supported);
941 if (was_enabled != now_enabled) {
943 VLOG_DBG("Port %d added: %s", port_no, ds_cstr(&ds));
945 VLOG_DBG("Port %d deleted", port_no);
948 VLOG_DBG("Port %d changed: %s", port_no, ds_cstr(&ds));
955 port_watcher_register_callback(struct port_watcher *pw,
956 port_changed_cb_func *port_changed,
959 assert(pw->n_cbs < ARRAY_SIZE(pw->cbs));
960 pw->cbs[pw->n_cbs].port_changed = port_changed;
961 pw->cbs[pw->n_cbs].aux = aux;
966 port_watcher_register_local_port_callback(struct port_watcher *pw,
967 local_port_changed_cb_func *cb,
970 assert(pw->n_local_cbs < ARRAY_SIZE(pw->local_cbs));
971 pw->local_cbs[pw->n_local_cbs].local_port_changed = cb;
972 pw->local_cbs[pw->n_local_cbs].aux = aux;
977 port_watcher_get_config(const struct port_watcher *pw, int port_no)
979 int idx = port_no_to_pw_idx(port_no);
980 return idx >= 0 ? ntohl(pw->ports[idx].config) : 0;
984 port_watcher_set_flags(struct port_watcher *pw, int port_no,
985 uint32_t config, uint32_t c_mask,
986 uint32_t state, uint32_t s_mask)
988 struct ofp_phy_port old;
989 struct ofp_phy_port *p;
990 struct ofp_port_mod *opm;
991 struct ofp_port_status *ops;
995 idx = port_no_to_pw_idx(port_no);
1000 p = &pw->ports[idx];
1001 if (!((ntohl(p->state) ^ state) & s_mask)
1002 && (!((ntohl(p->config) ^ config) & c_mask))) {
1007 /* Update our idea of the flags. */
1008 p->config = htonl((ntohl(p->config) & ~c_mask) | (config & c_mask));
1009 p->state = htonl((ntohl(p->state) & ~s_mask) | (state & s_mask));
1010 call_port_changed_callbacks(pw, port_no, &old, p);
1012 /* Change the flags in the datapath. */
1013 opm = make_openflow(sizeof *opm, OFPT_PORT_MOD, &b);
1014 opm->port_no = p->port_no;
1015 memcpy(opm->hw_addr, p->hw_addr, OFP_ETH_ALEN);
1016 opm->config = p->config;
1017 opm->mask = htonl(c_mask);
1018 opm->advertise = htonl(0);
1019 rconn_send(pw->local_rconn, b, NULL);
1021 /* Notify the controller that the flags changed. */
1022 ops = make_openflow(sizeof *ops, OFPT_PORT_STATUS, &b);
1023 ops->reason = OFPPR_MODIFY;
1025 rconn_send(pw->remote_rconn, b, NULL);
1029 port_watcher_is_ready(const struct port_watcher *pw)
1031 return pw->got_feature_reply;
1035 port_watcher_create(struct rconn *local_rconn, struct rconn *remote_rconn,
1036 struct port_watcher **pwp)
1038 struct port_watcher *pw;
1041 pw = *pwp = xcalloc(1, sizeof *pw);
1042 pw->local_rconn = local_rconn;
1043 pw->remote_rconn = remote_rconn;
1044 pw->last_feature_request = TIME_MIN;
1045 for (i = 0; i < OFPP_MAX; i++) {
1046 pw->ports[i].port_no = htons(OFPP_NONE);
1048 pw->local_port_name[0] = '\0';
1049 port_watcher_register_callback(pw, log_port_status, NULL);
1050 return make_hook(port_watcher_local_packet_cb,
1051 port_watcher_remote_packet_cb,
1052 port_watcher_periodic_cb,
1053 port_watcher_wait_cb, pw);
1056 struct snat_port_conf {
1058 struct nx_snat_config config;
1062 struct port_watcher *pw;
1063 struct list port_list;
1067 /* Source-NAT configuration monitor. */
1068 #define SNAT_CMD_LEN 1024
1070 /* Commands to configure iptables. There is no programmatic interface
1071 * to iptables from the kernel, so we're stuck making command-line calls
1073 #define SNAT_FLUSH_ALL_CMD "/sbin/iptables -t nat -F"
1074 #define SNAT_FLUSH_CHAIN_CMD "/sbin/iptables -t nat -F of-snat-%s"
1076 #define SNAT_ADD_CHAIN_CMD "/sbin/iptables -t nat -N of-snat-%s"
1077 #define SNAT_CONF_CHAIN_CMD "/sbin/iptables -t nat -A POSTROUTING -o %s -j of-snat-%s"
1079 #define SNAT_ADD_IP_CMD "/sbin/iptables -t nat -A of-snat-%s -j SNAT --to %s-%s"
1080 #define SNAT_ADD_TCP_CMD "/sbin/iptables -t nat -A of-snat-%s -j SNAT -p TCP --to %s-%s:%d-%d"
1081 #define SNAT_ADD_UDP_CMD "/sbin/iptables -t nat -A of-snat-%s -j SNAT -p UDP --to %s-%s:%d-%d"
1083 #define SNAT_UNSET_CHAIN_CMD "/sbin/iptables -t nat -D POSTROUTING -o %s -j of-snat-%s"
1084 #define SNAT_DEL_CHAIN_CMD "/sbin/iptables -t nat -X of-snat-%s"
1087 snat_add_rules(const struct nx_snat_config *sc, const uint8_t *dev_name)
1089 char command[SNAT_CMD_LEN];
1090 char ip_str_start[16];
1091 char ip_str_end[16];
1094 snprintf(ip_str_start, sizeof ip_str_start, IP_FMT,
1095 IP_ARGS(&sc->ip_addr_start));
1096 snprintf(ip_str_end, sizeof ip_str_end, IP_FMT,
1097 IP_ARGS(&sc->ip_addr_end));
1099 /* We always attempt to remove existing entries, so that we know
1100 * there's a pristine state for SNAT on the interface. We just ignore
1101 * the results of these calls, since iptables will complain about
1102 * any non-existent entries. */
1104 /* Flush the chain that does the SNAT. */
1105 snprintf(command, sizeof(command), SNAT_FLUSH_CHAIN_CMD, dev_name);
1108 /* We always try to create the a new chain. */
1109 snprintf(command, sizeof(command), SNAT_ADD_CHAIN_CMD, dev_name);
1112 /* Disassociate any old SNAT chain from the POSTROUTING chain. */
1113 snprintf(command, sizeof(command), SNAT_UNSET_CHAIN_CMD, dev_name,
1117 /* Associate the new chain with the POSTROUTING hook. */
1118 snprintf(command, sizeof(command), SNAT_CONF_CHAIN_CMD, dev_name,
1120 if (system(command) != 0) {
1121 VLOG_ERR("SNAT: problem flushing chain for add");
1125 /* If configured, restrict TCP source port ranges. */
1126 if ((sc->tcp_start != 0) && (sc->tcp_end != 0)) {
1127 snprintf(command, sizeof(command), SNAT_ADD_TCP_CMD,
1128 dev_name, ip_str_start, ip_str_end,
1129 ntohs(sc->tcp_start), ntohs(sc->tcp_end));
1130 if (system(command) != 0) {
1131 VLOG_ERR("SNAT: problem adding TCP rule");
1136 /* If configured, restrict UDP source port ranges. */
1137 if ((sc->udp_start != 0) && (sc->udp_end != 0)) {
1138 snprintf(command, sizeof(command), SNAT_ADD_UDP_CMD,
1139 dev_name, ip_str_start, ip_str_end,
1140 ntohs(sc->udp_start), ntohs(sc->udp_end));
1141 if (system(command) != 0) {
1142 VLOG_ERR("SNAT: problem adding UDP rule");
1147 /* Add a rule that covers all IP traffic that would not be covered
1148 * by the prior TCP or UDP ranges. */
1149 snprintf(command, sizeof(command), SNAT_ADD_IP_CMD,
1150 dev_name, ip_str_start, ip_str_end);
1151 if (system(command) != 0) {
1152 VLOG_ERR("SNAT: problem adding base rule");
1158 snat_del_rules(const uint8_t *dev_name)
1160 char command[SNAT_CMD_LEN];
1162 /* Flush the chain that does the SNAT. */
1163 snprintf(command, sizeof(command), SNAT_FLUSH_CHAIN_CMD, dev_name);
1164 if (system(command) != 0) {
1165 VLOG_ERR("SNAT: problem flushing chain for deletion");
1169 /* Disassociate the SNAT chain from the POSTROUTING chain. */
1170 snprintf(command, sizeof(command), SNAT_UNSET_CHAIN_CMD, dev_name,
1172 if (system(command) != 0) {
1173 VLOG_ERR("SNAT: problem unsetting chain");
1177 /* Now we can finally delete our SNAT chain. */
1178 snprintf(command, sizeof(command), SNAT_DEL_CHAIN_CMD, dev_name);
1179 if (system(command) != 0) {
1180 VLOG_ERR("SNAT: problem deleting chain");
1186 snat_config(const struct nx_snat_config *sc, struct snat_data *snat)
1189 struct port_watcher *pw = snat->pw;
1190 struct ofp_phy_port *pw_opp;
1191 struct snat_port_conf *c, *spc=NULL;
1194 port_no = ntohs(sc->port);
1195 idx = port_no_to_pw_idx(port_no);
1200 pw_opp = &pw->ports[idx];
1201 if (htons(pw_opp->port_no) != port_no) {
1205 LIST_FOR_EACH(c, struct snat_port_conf, node, &snat->port_list) {
1206 if (c->config.port == sc->port) {
1212 if (sc->command == NXSC_ADD) {
1214 spc = xmalloc(sizeof(*c));
1216 VLOG_ERR("SNAT: no memory for new entry");
1219 list_push_back(&snat->port_list, &spc->node);
1221 memcpy(&spc->config, sc, sizeof(spc->config));
1222 snat_add_rules(sc, pw_opp->name);
1224 snat_del_rules(pw_opp->name);
1225 list_remove(&spc->node);
1230 snat_remote_packet_cb(struct relay *r, void *snat_)
1232 struct snat_data *snat = snat_;
1233 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
1234 struct nicira_header *request = msg->data;
1235 struct nx_act_config *nac = msg->data;
1239 if (msg->size < sizeof(struct nx_act_config)) {
1242 request = msg->data;
1243 if (request->header.type != OFPT_VENDOR
1244 || request->vendor != htonl(NX_VENDOR_ID)
1245 || request->subtype != htonl(NXT_ACT_SET_CONFIG)) {
1249 /* We're only interested in attempts to configure SNAT */
1250 if (nac->type != htons(NXAST_SNAT)) {
1254 n_configs = (msg->size - sizeof *nac) / sizeof *nac->snat;
1255 for (i=0; i<n_configs; i++) {
1256 snat_config(&nac->snat[i], snat);
1263 snat_port_changed_cb(uint16_t port_no,
1264 const struct ofp_phy_port *old,
1265 const struct ofp_phy_port *new,
1268 struct snat_data *snat = snat_;
1269 struct snat_port_conf *c;
1271 /* We're only interested in ports that went away */
1272 if (new->port_no != htons(OFPP_NONE)) {
1276 LIST_FOR_EACH(c, struct snat_port_conf, node, &snat->port_list) {
1277 if (c->config.port == old->port_no) {
1278 snat_del_rules(old->name);
1279 list_remove(&c->node);
1286 snat_hook_create(struct port_watcher *pw)
1289 struct snat_data *snat;
1291 ret = system(SNAT_FLUSH_ALL_CMD);
1293 VLOG_ERR("SNAT: problum flushing tables");
1296 snat = xcalloc(1, sizeof *snat);
1298 list_init(&snat->port_list);
1300 port_watcher_register_callback(pw, snat_port_changed_cb, snat);
1301 return make_hook(NULL, snat_remote_packet_cb, NULL, NULL, snat);
1304 /* Spanning tree protocol. */
1306 /* Extra time, in seconds, at boot before going into fail-open, to give the
1307 * spanning tree protocol time to figure out the network layout. */
1308 #define STP_EXTRA_BOOT_TIME 30
1312 struct port_watcher *pw;
1313 struct rconn *local_rconn;
1314 struct rconn *remote_rconn;
1315 long long int last_tick_256ths;
1320 stp_local_packet_cb(struct relay *r, void *stp_)
1322 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
1323 struct ofp_header *oh;
1324 struct stp_data *stp = stp_;
1325 struct ofp_packet_in *opi;
1326 struct eth_header *eth;
1327 struct llc_header *llc;
1328 struct ofpbuf payload;
1333 if (oh->type == OFPT_FEATURES_REPLY
1334 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
1335 struct ofp_switch_features *osf = msg->data;
1336 osf->capabilities |= htonl(OFPC_STP);
1340 if (!get_ofp_packet_eth_header(r, &opi, ð)
1341 || !eth_addr_equals(eth->eth_dst, stp_eth_addr)) {
1345 port_no = ntohs(opi->in_port);
1346 if (port_no >= STP_MAX_PORTS) {
1347 /* STP only supports 255 ports. */
1350 if (port_watcher_get_config(stp->pw, port_no) & OFPPC_NO_STP) {
1351 /* We're not doing STP on this port. */
1355 if (opi->reason == OFPR_ACTION) {
1356 /* The controller set up a flow for this, so we won't intercept it. */
1360 get_ofp_packet_payload(opi, &payload);
1361 flow_extract(&payload, port_no, &flow);
1362 if (flow.dl_type != htons(OFP_DL_TYPE_NOT_ETH_TYPE)) {
1363 VLOG_DBG("non-LLC frame received on STP multicast address");
1366 llc = ofpbuf_at_assert(&payload, sizeof *eth, sizeof *llc);
1367 if (llc->llc_dsap != STP_LLC_DSAP) {
1368 VLOG_DBG("bad DSAP 0x%02"PRIx8" received on STP multicast address",
1373 /* Trim off padding on payload. */
1374 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1375 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1377 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1378 struct stp_port *p = stp_get_port(stp->stp, port_no);
1379 stp_received_bpdu(p, payload.data, payload.size);
1385 static long long int
1388 return time_msec() * 256 / 1000;
1392 stp_periodic_cb(void *stp_)
1394 struct stp_data *stp = stp_;
1395 long long int now_256ths = time_256ths();
1396 long long int elapsed_256ths = now_256ths - stp->last_tick_256ths;
1399 if (!port_watcher_is_ready(stp->pw)) {
1400 /* Can't start STP until we know port flags, because port flags can
1404 if (elapsed_256ths <= 0) {
1408 stp_tick(stp->stp, MIN(INT_MAX, elapsed_256ths));
1409 stp->last_tick_256ths = now_256ths;
1411 while (stp_get_changed_port(stp->stp, &p)) {
1412 int port_no = stp_port_no(p);
1413 enum stp_state s_state = stp_port_get_state(p);
1415 if (s_state != STP_DISABLED) {
1416 VLOG_WARN("STP: Port %d entered %s state",
1417 port_no, stp_state_name(s_state));
1419 if (!(port_watcher_get_config(stp->pw, port_no) & OFPPC_NO_STP)) {
1420 uint32_t p_config = 0;
1424 p_state = OFPPS_STP_LISTEN;
1427 p_state = OFPPS_STP_LEARN;
1430 case STP_FORWARDING:
1431 p_state = OFPPS_STP_FORWARD;
1434 p_state = OFPPS_STP_BLOCK;
1437 VLOG_DBG_RL(&vrl, "STP: Port %d has bad state %x",
1439 p_state = OFPPS_STP_FORWARD;
1442 if (!stp_forward_in_state(s_state)) {
1443 p_config = OFPPC_NO_FLOOD;
1445 port_watcher_set_flags(stp->pw, port_no,
1446 p_config, OFPPC_NO_FLOOD,
1447 p_state, OFPPS_STP_MASK);
1449 /* We don't own those flags. */
1455 stp_wait_cb(void *stp_ UNUSED)
1457 poll_timer_wait(1000);
1461 send_bpdu(const void *bpdu, size_t bpdu_size, int port_no, void *stp_)
1463 struct stp_data *stp = stp_;
1464 struct eth_header *eth;
1465 struct llc_header *llc;
1466 struct ofpbuf pkt, *opo;
1468 /* Packet skeleton. */
1469 ofpbuf_init(&pkt, ETH_HEADER_LEN + LLC_HEADER_LEN + bpdu_size);
1470 eth = ofpbuf_put_uninit(&pkt, sizeof *eth);
1471 llc = ofpbuf_put_uninit(&pkt, sizeof *llc);
1472 ofpbuf_put(&pkt, bpdu, bpdu_size);
1475 memcpy(eth->eth_dst, stp_eth_addr, ETH_ADDR_LEN);
1476 memcpy(eth->eth_src, stp->pw->ports[port_no].hw_addr, ETH_ADDR_LEN);
1477 eth->eth_type = htons(pkt.size - ETH_HEADER_LEN);
1480 llc->llc_dsap = STP_LLC_DSAP;
1481 llc->llc_ssap = STP_LLC_SSAP;
1482 llc->llc_cntl = STP_LLC_CNTL;
1484 opo = make_unbuffered_packet_out(&pkt, OFPP_NONE, port_no);
1485 ofpbuf_uninit(&pkt);
1486 rconn_send_with_limit(stp->local_rconn, opo, &stp->n_txq, OFPP_MAX);
1490 stp_is_port_supported(uint16_t port_no)
1492 /* We should be able to support STP on all possible OpenFlow physical
1493 * ports. (But we don't support STP on OFPP_LOCAL.) */
1494 BUILD_ASSERT_DECL(STP_MAX_PORTS >= OFPP_MAX);
1495 return port_no < STP_MAX_PORTS;
1499 stp_port_changed_cb(uint16_t port_no,
1500 const struct ofp_phy_port *old,
1501 const struct ofp_phy_port *new,
1504 struct stp_data *stp = stp_;
1507 if (!stp_is_port_supported(port_no)) {
1511 p = stp_get_port(stp->stp, port_no);
1512 if (new->port_no == htons(OFPP_NONE)
1513 || new->config & htonl(OFPPC_NO_STP | OFPPC_PORT_DOWN)
1514 || new->state & htonl(OFPPS_LINK_DOWN)) {
1515 stp_port_disable(p);
1519 if (new->curr & (OFPPF_10MB_HD | OFPPF_10MB_FD)) {
1521 } else if (new->curr & (OFPPF_100MB_HD | OFPPF_100MB_FD)) {
1523 } else if (new->curr & (OFPPF_1GB_HD | OFPPF_1GB_FD)) {
1525 } else if (new->curr & OFPPF_100MB_FD) {
1528 stp_port_set_speed(p, speed);
1533 stp_local_port_changed_cb(const struct ofp_phy_port *port, void *stp_)
1535 struct stp_data *stp = stp_;
1537 stp_set_bridge_id(stp->stp, eth_addr_to_uint64(port->hw_addr));
1542 stp_hook_create(const struct settings *s, struct port_watcher *pw,
1543 struct rconn *local, struct rconn *remote)
1545 uint8_t dpid[ETH_ADDR_LEN];
1546 struct stp_data *stp;
1548 stp = xcalloc(1, sizeof *stp);
1549 eth_addr_random(dpid);
1550 stp->stp = stp_create("stp", eth_addr_to_uint64(dpid), send_bpdu, stp);
1552 stp->local_rconn = local;
1553 stp->remote_rconn = remote;
1554 stp->last_tick_256ths = time_256ths();
1556 port_watcher_register_callback(pw, stp_port_changed_cb, stp);
1557 port_watcher_register_local_port_callback(pw, stp_local_port_changed_cb,
1559 return make_hook(stp_local_packet_cb, NULL,
1560 stp_periodic_cb, stp_wait_cb, stp);
1563 /* In-band control. */
1565 struct in_band_data {
1566 const struct settings *s;
1567 struct mac_learning *ml;
1568 struct netdev *of_device;
1569 struct rconn *controller;
1574 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct ofpbuf *b)
1576 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
1579 static const uint8_t *
1580 get_controller_mac(struct in_band_data *in_band)
1582 static uint32_t ip, last_nonzero_ip;
1583 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
1584 static time_t next_refresh = 0;
1586 uint32_t last_ip = ip;
1588 time_t now = time_now();
1590 ip = rconn_get_ip(in_band->controller);
1591 if (last_ip != ip || !next_refresh || now >= next_refresh) {
1594 /* Look up MAC address. */
1595 memset(mac, 0, sizeof mac);
1596 if (ip && in_band->of_device) {
1597 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
1599 VLOG_DBG_RL(&vrl, "cannot look up controller hw address "
1600 "("IP_FMT"): %s", IP_ARGS(&ip), strerror(retval));
1603 have_mac = !eth_addr_is_zero(mac);
1605 /* Log changes in IP, MAC addresses. */
1606 if (ip && ip != last_nonzero_ip) {
1607 VLOG_DBG("controller IP address changed from "IP_FMT
1608 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
1609 last_nonzero_ip = ip;
1611 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
1612 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
1614 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
1615 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
1618 /* Schedule next refresh.
1620 * If we have an IP address but not a MAC address, then refresh
1621 * quickly, since we probably will get a MAC address soon (via ARP).
1622 * Otherwise, we can afford to wait a little while. */
1623 next_refresh = now + (!ip || have_mac ? 10 : 1);
1625 return !eth_addr_is_zero(mac) ? mac : NULL;
1629 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
1630 struct in_band_data *in_band)
1632 const uint8_t *mac = get_controller_mac(in_band);
1633 return mac && eth_addr_equals(mac, dl_addr);
1637 in_band_learn_mac(struct in_band_data *in_band,
1638 uint16_t in_port, const uint8_t src_mac[ETH_ADDR_LEN])
1640 if (mac_learning_learn(in_band->ml, src_mac, in_port)) {
1641 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
1642 ETH_ADDR_ARGS(src_mac), in_port);
1647 in_band_local_packet_cb(struct relay *r, void *in_band_)
1649 struct in_band_data *in_band = in_band_;
1650 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
1651 struct ofp_packet_in *opi;
1652 struct eth_header *eth;
1653 struct ofpbuf payload;
1658 if (!get_ofp_packet_eth_header(r, &opi, ð) || !in_band->of_device) {
1661 in_port = ntohs(opi->in_port);
1663 /* Deal with local stuff. */
1664 if (in_port == OFPP_LOCAL) {
1665 /* Sent by secure channel. */
1666 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1667 } else if (eth_addr_equals(eth->eth_dst,
1668 netdev_get_etheraddr(in_band->of_device))) {
1669 /* Sent to secure channel. */
1670 out_port = OFPP_LOCAL;
1671 in_band_learn_mac(in_band, in_port, eth->eth_src);
1672 } else if (eth->eth_type == htons(ETH_TYPE_ARP)
1673 && eth_addr_is_broadcast(eth->eth_dst)
1674 && is_controller_mac(eth->eth_src, in_band)) {
1675 /* ARP sent by controller. */
1676 out_port = OFPP_FLOOD;
1677 } else if (is_controller_mac(eth->eth_dst, in_band)
1678 || is_controller_mac(eth->eth_src, in_band)) {
1679 /* Traffic to or from controller. Switch it by hand. */
1680 in_band_learn_mac(in_band, in_port, eth->eth_src);
1681 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1683 const uint8_t *controller_mac;
1684 controller_mac = get_controller_mac(in_band);
1685 if (eth->eth_type == htons(ETH_TYPE_ARP)
1686 && eth_addr_is_broadcast(eth->eth_dst)
1687 && is_controller_mac(eth->eth_src, in_band)) {
1688 /* ARP sent by controller. */
1689 out_port = OFPP_FLOOD;
1690 } else if (is_controller_mac(eth->eth_dst, in_band)
1691 && in_port == mac_learning_lookup(in_band->ml,
1693 /* Drop controller traffic that arrives on the controller port. */
1700 get_ofp_packet_payload(opi, &payload);
1701 flow_extract(&payload, in_port, &flow);
1702 if (in_port == out_port) {
1703 /* The input and output port match. Set up a flow to drop packets. */
1704 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
1705 in_band->s->max_idle, 0));
1706 } else if (out_port != OFPP_FLOOD) {
1707 /* The output port is known, so add a new flow. */
1708 queue_tx(rc, in_band,
1709 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
1710 out_port, in_band->s->max_idle));
1712 /* If the switch didn't buffer the packet, we need to send a copy. */
1713 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1714 queue_tx(rc, in_band,
1715 make_unbuffered_packet_out(&payload, in_port, out_port));
1718 /* We don't know that MAC. Send along the packet without setting up a
1721 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1722 b = make_unbuffered_packet_out(&payload, in_port, out_port);
1724 b = make_buffered_packet_out(ntohl(opi->buffer_id),
1727 queue_tx(rc, in_band, b);
1733 in_band_status_cb(struct status_reply *sr, void *in_band_)
1735 struct in_band_data *in_band = in_band_;
1736 struct in_addr local_ip;
1737 uint32_t controller_ip;
1738 const uint8_t *controller_mac;
1740 if (in_band->of_device) {
1741 const uint8_t *mac = netdev_get_etheraddr(in_band->of_device);
1742 if (netdev_get_in4(in_band->of_device, &local_ip)) {
1743 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
1745 status_reply_put(sr, "local-mac="ETH_ADDR_FMT, ETH_ADDR_ARGS(mac));
1747 controller_ip = rconn_get_ip(in_band->controller);
1748 if (controller_ip) {
1749 status_reply_put(sr, "controller-ip="IP_FMT,
1750 IP_ARGS(&controller_ip));
1752 controller_mac = get_controller_mac(in_band);
1753 if (controller_mac) {
1754 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
1755 ETH_ADDR_ARGS(controller_mac));
1761 get_ofp_packet_payload(struct ofp_packet_in *opi, struct ofpbuf *payload)
1763 payload->data = opi->data;
1764 payload->size = ntohs(opi->header.length) - offsetof(struct ofp_packet_in,
1769 in_band_local_port_cb(const struct ofp_phy_port *port, void *in_band_)
1771 struct in_band_data *in_band = in_band_;
1773 char name[sizeof port->name + 1];
1774 get_port_name(port, name, sizeof name);
1776 if (!in_band->of_device
1777 || strcmp(netdev_get_name(in_band->of_device), name))
1780 netdev_close(in_band->of_device);
1781 error = netdev_open(name, NETDEV_ETH_TYPE_NONE,
1782 &in_band->of_device);
1784 VLOG_ERR("failed to open in-band control network device "
1785 "\"%s\": %s", name, strerror(errno));
1789 netdev_close(in_band->of_device);
1790 in_band->of_device = NULL;
1795 in_band_hook_create(const struct settings *s, struct switch_status *ss,
1796 struct port_watcher *pw, struct rconn *remote)
1798 struct in_band_data *in_band;
1800 in_band = xcalloc(1, sizeof *in_band);
1802 in_band->ml = mac_learning_create();
1803 in_band->of_device = NULL;
1804 in_band->controller = remote;
1805 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
1806 port_watcher_register_local_port_callback(pw, in_band_local_port_cb,
1808 return make_hook(in_band_local_packet_cb, NULL, NULL, NULL, in_band);
1811 /* Fail open support. */
1813 struct fail_open_data {
1814 const struct settings *s;
1815 struct rconn *local_rconn;
1816 struct rconn *remote_rconn;
1817 struct lswitch *lswitch;
1818 int last_disconn_secs;
1819 time_t boot_deadline;
1822 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
1824 fail_open_periodic_cb(void *fail_open_)
1826 struct fail_open_data *fail_open = fail_open_;
1830 if (time_now() < fail_open->boot_deadline) {
1833 disconn_secs = rconn_failure_duration(fail_open->remote_rconn);
1834 open = disconn_secs >= fail_open->s->probe_interval * 3;
1835 if (open != (fail_open->lswitch != NULL)) {
1837 VLOG_WARN("No longer in fail-open mode");
1838 lswitch_destroy(fail_open->lswitch);
1839 fail_open->lswitch = NULL;
1841 VLOG_WARN("Could not connect to controller for %d seconds, "
1842 "failing open", disconn_secs);
1843 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
1844 fail_open->s->max_idle);
1845 fail_open->last_disconn_secs = disconn_secs;
1847 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
1848 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
1849 "from controller", disconn_secs);
1850 fail_open->last_disconn_secs = disconn_secs;
1855 fail_open_local_packet_cb(struct relay *r, void *fail_open_)
1857 struct fail_open_data *fail_open = fail_open_;
1858 if (!fail_open->lswitch) {
1861 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
1862 r->halves[HALF_LOCAL].rxbuf);
1863 rconn_run(fail_open->local_rconn);
1869 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
1871 struct fail_open_data *fail_open = fail_open_;
1872 const struct settings *s = fail_open->s;
1873 int trigger_duration = s->probe_interval * 3;
1874 int cur_duration = rconn_failure_duration(fail_open->remote_rconn);
1876 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
1877 status_reply_put(sr, "current-duration=%d", cur_duration);
1878 status_reply_put(sr, "triggered=%s",
1879 cur_duration >= trigger_duration ? "true" : "false");
1880 status_reply_put(sr, "max-idle=%d", s->max_idle);
1884 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
1885 struct rconn *local_rconn, struct rconn *remote_rconn)
1887 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
1889 fail_open->local_rconn = local_rconn;
1890 fail_open->remote_rconn = remote_rconn;
1891 fail_open->lswitch = NULL;
1892 fail_open->boot_deadline = time_now() + s->probe_interval * 3;
1893 if (s->enable_stp) {
1894 fail_open->boot_deadline += STP_EXTRA_BOOT_TIME;
1896 switch_status_register_category(ss, "fail-open",
1897 fail_open_status_cb, fail_open);
1898 return make_hook(fail_open_local_packet_cb, NULL,
1899 fail_open_periodic_cb, NULL, fail_open);
1902 struct rate_limiter {
1903 const struct settings *s;
1904 struct rconn *remote_rconn;
1906 /* One queue per physical port. */
1907 struct ofp_queue queues[OFPP_MAX];
1908 int n_queued; /* Sum over queues[*].n. */
1909 int next_tx_port; /* Next port to check in round-robin. */
1913 * It costs 1000 tokens to send a single packet_in message. A single token
1914 * per message would be more straightforward, but this choice lets us avoid
1915 * round-off error in refill_bucket()'s calculation of how many tokens to
1916 * add to the bucket, since no division step is needed. */
1917 long long int last_fill; /* Time at which we last added tokens. */
1918 int tokens; /* Current number of tokens. */
1920 /* Transmission queue. */
1921 int n_txq; /* No. of packets waiting in rconn for tx. */
1923 /* Statistics reporting. */
1924 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
1925 unsigned long long n_limited; /* # queued for rate limiting. */
1926 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
1927 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
1930 /* Drop a packet from the longest queue in 'rl'. */
1932 drop_packet(struct rate_limiter *rl)
1934 struct ofp_queue *longest; /* Queue currently selected as longest. */
1935 int n_longest; /* # of queues of same length as 'longest'. */
1936 struct ofp_queue *q;
1938 longest = &rl->queues[0];
1940 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
1941 if (longest->n < q->n) {
1944 } else if (longest->n == q->n) {
1947 /* Randomly select one of the longest queues, with a uniform
1948 * distribution (Knuth algorithm 3.4.2R). */
1949 if (!random_range(n_longest)) {
1955 /* FIXME: do we want to pop the tail instead? */
1956 ofpbuf_delete(queue_pop_head(longest));
1960 /* Remove and return the next packet to transmit (in round-robin order). */
1961 static struct ofpbuf *
1962 dequeue_packet(struct rate_limiter *rl)
1966 for (i = 0; i < OFPP_MAX; i++) {
1967 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
1968 struct ofp_queue *q = &rl->queues[port];
1970 rl->next_tx_port = (port + 1) % OFPP_MAX;
1972 return queue_pop_head(q);
1978 /* Add tokens to the bucket based on elapsed time. */
1980 refill_bucket(struct rate_limiter *rl)
1982 const struct settings *s = rl->s;
1983 long long int now = time_msec();
1984 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
1985 if (tokens >= 1000) {
1986 rl->last_fill = now;
1987 rl->tokens = MIN(tokens, s->burst_limit * 1000);
1991 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
1992 * true if successful, false otherwise. (In the latter case no tokens are
1995 get_token(struct rate_limiter *rl)
1997 if (rl->tokens >= 1000) {
2006 rate_limit_local_packet_cb(struct relay *r, void *rl_)
2008 struct rate_limiter *rl = rl_;
2009 const struct settings *s = rl->s;
2010 struct ofp_packet_in *opi;
2012 opi = get_ofp_packet_in(r);
2017 if (!rl->n_queued && get_token(rl)) {
2018 /* In the common case where we are not constrained by the rate limit,
2019 * let the packet take the normal path. */
2023 /* Otherwise queue it up for the periodic callback to drain out. */
2024 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
2025 int port = ntohs(opi->in_port) % OFPP_MAX;
2026 if (rl->n_queued >= s->burst_limit) {
2029 queue_push_tail(&rl->queues[port], ofpbuf_clone(msg));
2037 rate_limit_status_cb(struct status_reply *sr, void *rl_)
2039 struct rate_limiter *rl = rl_;
2041 status_reply_put(sr, "normal=%llu", rl->n_normal);
2042 status_reply_put(sr, "limited=%llu", rl->n_limited);
2043 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
2044 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
2048 rate_limit_periodic_cb(void *rl_)
2050 struct rate_limiter *rl = rl_;
2053 /* Drain some packets out of the bucket if possible, but limit the number
2054 * of iterations to allow other code to get work done too. */
2056 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
2057 /* Use a small, arbitrary limit for the amount of queuing to do here,
2058 * because the TCP connection is responsible for buffering and there is
2059 * no point in trying to transmit faster than the TCP connection can
2061 struct ofpbuf *b = dequeue_packet(rl);
2062 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
2069 rate_limit_wait_cb(void *rl_)
2071 struct rate_limiter *rl = rl_;
2073 if (rl->tokens >= 1000) {
2074 /* We can transmit more packets as soon as we're called again. */
2075 poll_immediate_wake();
2077 /* We have to wait for the bucket to re-fill. We could calculate
2078 * the exact amount of time here for increased smoothness. */
2079 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
2085 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
2086 struct rconn *local, struct rconn *remote)
2088 struct rate_limiter *rl;
2091 rl = xcalloc(1, sizeof *rl);
2093 rl->remote_rconn = remote;
2094 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
2095 queue_init(&rl->queues[i]);
2097 rl->last_fill = time_msec();
2098 rl->tokens = s->rate_limit * 100;
2099 switch_status_register_category(ss, "rate-limit",
2100 rate_limit_status_cb, rl);
2101 return make_hook(rate_limit_local_packet_cb, NULL, rate_limit_periodic_cb,
2102 rate_limit_wait_cb, rl);
2105 /* OFPST_SWITCH statistics. */
2107 struct switch_status_category {
2109 void (*cb)(struct status_reply *, void *aux);
2113 struct switch_status {
2114 const struct settings *s;
2116 struct switch_status_category categories[8];
2120 struct status_reply {
2121 struct switch_status_category *category;
2127 switch_status_remote_packet_cb(struct relay *r, void *ss_)
2129 struct switch_status *ss = ss_;
2130 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
2131 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
2132 struct switch_status_category *c;
2133 struct nicira_header *request;
2134 struct nicira_header *reply;
2135 struct status_reply sr;
2139 if (msg->size < sizeof(struct nicira_header)) {
2142 request = msg->data;
2143 if (request->header.type != OFPT_VENDOR
2144 || request->vendor != htonl(NX_VENDOR_ID)
2145 || request->subtype != htonl(NXT_STATUS_REQUEST)) {
2149 sr.request.string = (void *) (request + 1);
2150 sr.request.length = msg->size - sizeof *request;
2151 ds_init(&sr.output);
2152 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
2153 if (!memcmp(c->name, sr.request.string,
2154 MIN(strlen(c->name), sr.request.length))) {
2159 reply = make_openflow_xid(sizeof *reply + sr.output.length,
2160 OFPT_VENDOR, request->header.xid, &b);
2161 reply->vendor = htonl(NX_VENDOR_ID);
2162 reply->subtype = htonl(NXT_STATUS_REPLY);
2163 memcpy(reply + 1, sr.output.string, sr.output.length);
2164 retval = rconn_send(rc, b, NULL);
2165 if (retval && retval != EAGAIN) {
2166 VLOG_WARN("send failed (%s)", strerror(retval));
2168 ds_destroy(&sr.output);
2173 rconn_status_cb(struct status_reply *sr, void *rconn_)
2175 struct rconn *rconn = rconn_;
2176 time_t now = time_now();
2178 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
2179 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
2180 status_reply_put(sr, "backoff=%d", rconn_get_backoff(rconn));
2181 status_reply_put(sr, "is-connected=%s",
2182 rconn_is_connected(rconn) ? "true" : "false");
2183 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
2184 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
2185 status_reply_put(sr, "attempted-connections=%u",
2186 rconn_get_attempted_connections(rconn));
2187 status_reply_put(sr, "successful-connections=%u",
2188 rconn_get_successful_connections(rconn));
2189 status_reply_put(sr, "last-connection=%ld",
2190 (long int) (now - rconn_get_last_connection(rconn)));
2191 status_reply_put(sr, "time-connected=%lu",
2192 rconn_get_total_time_connected(rconn));
2193 status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(rconn));
2197 config_status_cb(struct status_reply *sr, void *s_)
2199 const struct settings *s = s_;
2202 for (i = 0; i < s->n_listeners; i++) {
2203 status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
2205 if (s->probe_interval) {
2206 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
2208 if (s->max_backoff) {
2209 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
2214 switch_status_cb(struct status_reply *sr, void *ss_)
2216 struct switch_status *ss = ss_;
2217 time_t now = time_now();
2219 status_reply_put(sr, "now=%ld", (long int) now);
2220 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
2221 status_reply_put(sr, "pid=%ld", (long int) getpid());
2225 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
2227 struct switch_status *ss = xcalloc(1, sizeof *ss);
2229 ss->booted = time_now();
2230 switch_status_register_category(ss, "config",
2231 config_status_cb, (void *) s);
2232 switch_status_register_category(ss, "switch", switch_status_cb, ss);
2234 return make_hook(NULL, switch_status_remote_packet_cb, NULL, NULL, ss);
2238 switch_status_register_category(struct switch_status *ss,
2239 const char *category,
2240 void (*cb)(struct status_reply *,
2244 struct switch_status_category *c;
2245 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
2246 c = &ss->categories[ss->n_categories++];
2249 c->name = xstrdup(category);
2253 status_reply_put(struct status_reply *sr, const char *content, ...)
2255 size_t old_length = sr->output.length;
2259 /* Append the status reply to the output. */
2260 ds_put_format(&sr->output, "%s.", sr->category->name);
2261 va_start(args, content);
2262 ds_put_format_valist(&sr->output, content, args);
2264 if (ds_last(&sr->output) != '\n') {
2265 ds_put_char(&sr->output, '\n');
2268 /* Drop what we just added if it doesn't match the request. */
2269 added = sr->output.length - old_length;
2270 if (added < sr->request.length
2271 || memcmp(&sr->output.string[old_length],
2272 sr->request.string, sr->request.length)) {
2273 ds_truncate(&sr->output, old_length);
2278 /* Controller discovery. */
2282 const struct settings *s;
2283 struct dhclient *dhcp;
2288 discovery_status_cb(struct status_reply *sr, void *d_)
2290 struct discovery *d = d_;
2292 status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re);
2293 status_reply_put(sr, "n-changes=%d", d->n_changes);
2295 status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp));
2296 status_reply_put(sr, "state-elapsed=%u",
2297 dhclient_get_state_elapsed(d->dhcp));
2298 if (dhclient_is_bound(d->dhcp)) {
2299 uint32_t ip = dhclient_get_ip(d->dhcp);
2300 uint32_t netmask = dhclient_get_netmask(d->dhcp);
2301 uint32_t router = dhclient_get_router(d->dhcp);
2303 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
2304 uint32_t dns_server;
2308 status_reply_put(sr, "ip="IP_FMT, IP_ARGS(&ip));
2309 status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask));
2311 status_reply_put(sr, "router="IP_FMT, IP_ARGS(&router));
2314 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i,
2317 status_reply_put(sr, "dns%d="IP_FMT, i, IP_ARGS(&dns_server));
2320 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
2322 status_reply_put(sr, "domain=%s", domain_name);
2326 status_reply_put(sr, "lease-remaining=%u",
2327 dhclient_get_lease_remaining(d->dhcp));
2333 discovery_local_port_cb(const struct ofp_phy_port *port, void *d_)
2335 struct discovery *d = d_;
2337 char name[OFP_MAX_PORT_NAME_LEN + 1];
2338 struct netdev *netdev;
2341 /* Check that this was really a change. */
2342 get_port_name(port, name, sizeof name);
2343 if (d->dhcp && !strcmp(netdev_get_name(dhclient_get_netdev(d->dhcp)),
2348 /* Destroy current DHCP client. */
2349 dhclient_destroy(d->dhcp);
2352 /* Bring local network device up. */
2353 retval = netdev_open(name, NETDEV_ETH_TYPE_NONE, &netdev);
2355 VLOG_ERR("Could not open %s device, discovery disabled: %s",
2356 name, strerror(retval));
2359 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
2361 VLOG_ERR("Could not bring %s device up, discovery disabled: %s",
2362 name, strerror(retval));
2365 netdev_close(netdev);
2367 /* Initialize DHCP client. */
2368 retval = dhclient_create(name, modify_dhcp_request,
2369 validate_dhcp_offer, (void *) d->s, &d->dhcp);
2371 VLOG_ERR("Failed to initialize DHCP client, "
2372 "discovery disabled: %s", strerror(retval));
2375 dhclient_init(d->dhcp, 0);
2377 dhclient_destroy(d->dhcp);
2383 static struct discovery *
2384 discovery_init(const struct settings *s, struct port_watcher *pw,
2385 struct switch_status *ss)
2387 struct discovery *d;
2389 d = xmalloc(sizeof *d);
2394 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
2395 port_watcher_register_local_port_callback(pw, discovery_local_port_cb, d);
2401 discovery_question_connectivity(struct discovery *d)
2404 dhclient_force_renew(d->dhcp, 15);
2409 discovery_run(struct discovery *d, char **controller_name)
2412 *controller_name = NULL;
2416 dhclient_run(d->dhcp);
2417 if (!dhclient_changed(d->dhcp)) {
2421 dhclient_configure_netdev(d->dhcp);
2422 if (d->s->update_resolv_conf) {
2423 dhclient_update_resolv_conf(d->dhcp);
2426 if (dhclient_is_bound(d->dhcp)) {
2427 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
2428 DHCP_CODE_OFP_CONTROLLER_VCONN);
2429 VLOG_WARN("%s: discovered controller", *controller_name);
2432 *controller_name = NULL;
2434 VLOG_WARN("discovered controller no longer available");
2442 discovery_wait(struct discovery *d)
2445 dhclient_wait(d->dhcp);
2450 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
2452 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
2456 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
2458 const struct settings *s = s_;
2462 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
2464 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
2467 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
2469 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
2470 s->accept_controller_re);
2476 /* User interface. */
2479 parse_options(int argc, char *argv[], struct settings *s)
2482 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
2484 OPT_INACTIVITY_PROBE,
2489 OPT_BOOTSTRAP_CA_CERT,
2495 static struct option long_options[] = {
2496 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
2497 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
2498 {"fail", required_argument, 0, 'F'},
2499 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
2500 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
2501 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
2502 {"listen", required_argument, 0, 'l'},
2503 {"monitor", required_argument, 0, 'm'},
2504 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
2505 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
2506 {"stp", no_argument, 0, OPT_STP},
2507 {"no-stp", no_argument, 0, OPT_NO_STP},
2508 {"out-of-band", no_argument, 0, OPT_OUT_OF_BAND},
2509 {"in-band", no_argument, 0, OPT_IN_BAND},
2510 {"detach", no_argument, 0, 'D'},
2511 {"force", no_argument, 0, 'f'},
2512 {"pidfile", optional_argument, 0, 'P'},
2513 {"verbose", optional_argument, 0, 'v'},
2514 {"help", no_argument, 0, 'h'},
2515 {"version", no_argument, 0, 'V'},
2517 VCONN_SSL_LONG_OPTIONS
2518 {"bootstrap-ca-cert", required_argument, 0, OPT_BOOTSTRAP_CA_CERT},
2522 char *short_options = long_options_to_short_options(long_options);
2523 char *accept_re = NULL;
2526 /* Set defaults that we can figure out before parsing options. */
2528 s->monitor_name = NULL;
2529 s->fail_mode = FAIL_OPEN;
2531 s->probe_interval = 15;
2532 s->max_backoff = 15;
2533 s->update_resolv_conf = true;
2536 s->enable_stp = false;
2541 c = getopt_long(argc, argv, short_options, long_options, NULL);
2547 case OPT_ACCEPT_VCONN:
2548 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
2551 case OPT_NO_RESOLV_CONF:
2552 s->update_resolv_conf = false;
2556 if (!strcmp(optarg, "open")) {
2557 s->fail_mode = FAIL_OPEN;
2558 } else if (!strcmp(optarg, "closed")) {
2559 s->fail_mode = FAIL_CLOSED;
2561 ofp_fatal(0, "-f or --fail argument must be \"open\" "
2566 case OPT_INACTIVITY_PROBE:
2567 s->probe_interval = atoi(optarg);
2568 if (s->probe_interval < 5) {
2569 ofp_fatal(0, "--inactivity-probe argument must be at least 5");
2574 if (!strcmp(optarg, "permanent")) {
2575 s->max_idle = OFP_FLOW_PERMANENT;
2577 s->max_idle = atoi(optarg);
2578 if (s->max_idle < 1 || s->max_idle > 65535) {
2579 ofp_fatal(0, "--max-idle argument must be between 1 and "
2580 "65535 or the word 'permanent'");
2585 case OPT_MAX_BACKOFF:
2586 s->max_backoff = atoi(optarg);
2587 if (s->max_backoff < 1) {
2588 ofp_fatal(0, "--max-backoff argument must be at least 1");
2589 } else if (s->max_backoff > 3600) {
2590 s->max_backoff = 3600;
2594 case OPT_RATE_LIMIT:
2596 s->rate_limit = atoi(optarg);
2597 if (s->rate_limit < 1) {
2598 ofp_fatal(0, "--rate-limit argument must be at least 1");
2601 s->rate_limit = 1000;
2605 case OPT_BURST_LIMIT:
2606 s->burst_limit = atoi(optarg);
2607 if (s->burst_limit < 1) {
2608 ofp_fatal(0, "--burst-limit argument must be at least 1");
2613 s->enable_stp = true;
2617 s->enable_stp = false;
2620 case OPT_OUT_OF_BAND:
2633 set_pidfile(optarg);
2637 ignore_existing_pidfile();
2641 if (s->n_listeners >= MAX_MGMT) {
2643 "-l or --listen may be specified at most %d times",
2646 s->listener_names[s->n_listeners++] = optarg;
2650 if (s->monitor_name) {
2651 ofp_fatal(0, "-m or --monitor may only be specified once");
2653 s->monitor_name = optarg;
2660 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
2664 vlog_set_verbosity(optarg);
2668 VCONN_SSL_OPTION_HANDLERS
2670 case OPT_BOOTSTRAP_CA_CERT:
2671 vconn_ssl_set_ca_cert_file(optarg, true);
2682 free(short_options);
2686 if (argc < 1 || argc > 2) {
2687 ofp_fatal(0, "need one or two non-option arguments; "
2688 "use --help for usage");
2691 /* Local and remote vconns. */
2692 s->dp_name = argv[0];
2693 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
2695 /* Set accept_controller_regex. */
2697 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
2699 retval = regcomp(&s->accept_controller_regex, accept_re,
2700 REG_NOSUB | REG_EXTENDED);
2702 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
2703 char *buffer = xmalloc(length);
2704 regerror(retval, &s->accept_controller_regex, buffer, length);
2705 ofp_fatal(0, "%s: %s", accept_re, buffer);
2707 s->accept_controller_re = accept_re;
2709 /* Mode of operation. */
2710 s->discovery = s->controller_name == NULL;
2711 if (s->discovery && !s->in_band) {
2712 ofp_fatal(0, "Cannot perform discovery with out-of-band control");
2715 /* Rate limiting. */
2716 if (s->rate_limit) {
2717 if (s->rate_limit < 100) {
2718 VLOG_WARN("Rate limit set to unusually low value %d",
2721 if (!s->burst_limit) {
2722 s->burst_limit = s->rate_limit / 4;
2724 s->burst_limit = MAX(s->burst_limit, 1);
2725 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
2732 printf("%s: secure channel, a relay for OpenFlow messages.\n"
2733 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
2734 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
2735 "CONTROLLER is an active OpenFlow connection method; if it is\n"
2736 "omitted, then secchan performs controller discovery.\n",
2737 program_name, program_name);
2738 vconn_usage(true, true, true);
2739 printf("\nController discovery options:\n"
2740 " --accept-vconn=REGEX accept matching discovered controllers\n"
2741 " --no-resolv-conf do not update /etc/resolv.conf\n"
2742 "\nNetworking options:\n"
2743 " -F, --fail=open|closed when controller connection fails:\n"
2744 " closed: drop all packets\n"
2745 " open (default): act as learning switch\n"
2746 " --inactivity-probe=SECS time between inactivity probes\n"
2747 " --max-idle=SECS max idle for flows set up by secchan\n"
2748 " --max-backoff=SECS max time between controller connection\n"
2749 " attempts (default: 15 seconds)\n"
2750 " -l, --listen=METHOD allow management connections on METHOD\n"
2751 " (a passive OpenFlow connection method)\n"
2752 " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n"
2753 " (a passive OpenFlow connection method)\n"
2754 " --out-of-band controller connection is out-of-band\n"
2755 " --stp enable 802.1D Spanning Tree Protocol\n"
2756 " --no-stp disable 802.1D Spanning Tree Protocol\n"
2757 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
2758 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
2759 " --burst-limit=BURST limit on packet credit for idle time\n"
2760 "\nOther options:\n"
2761 " -D, --detach run in background as daemon\n"
2762 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
2763 " -f, --force with -P, start even if already running\n"
2764 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
2765 " -v, --verbose set maximum verbosity level\n"
2766 " -h, --help display this help message\n"
2767 " -V, --version display version information\n",