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"
65 #include "port-array.h"
70 #include "vconn-ssl.h"
72 #include "vlog-socket.h"
75 #define THIS_MODULE VLM_secchan
77 /* Behavior when the connection to the controller fails. */
79 FAIL_OPEN, /* Act as learning switch. */
80 FAIL_CLOSED /* Drop all packets. */
83 /* Maximum number of management connection listeners. */
86 /* Settings that may be configured by the user. */
88 /* Overall mode of operation. */
89 bool discovery; /* Discover the controller automatically? */
90 bool in_band; /* Connect to controller in-band? */
92 /* Related vconns and network devices. */
93 const char *dp_name; /* Local datapath. */
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_);
181 static struct discovery *discovery_init(const struct settings *,
182 struct port_watcher *,
183 struct switch_status *);
184 static void discovery_question_connectivity(struct discovery *);
185 static bool discovery_run(struct discovery *, char **controller_name);
186 static void discovery_wait(struct discovery *);
188 static struct hook in_band_hook_create(const struct settings *,
189 struct switch_status *,
190 struct port_watcher *,
191 struct rconn *remote);
193 static struct hook port_watcher_create(struct rconn *local,
194 struct rconn *remote,
195 struct port_watcher **);
196 static uint32_t port_watcher_get_config(const struct port_watcher *,
198 static const char *port_watcher_get_name(const struct port_watcher *,
199 uint16_t port_no) UNUSED;
200 static const uint8_t *port_watcher_get_hwaddr(const struct port_watcher *,
202 static void port_watcher_set_flags(struct port_watcher *, uint16_t port_no,
203 uint32_t config, uint32_t c_mask,
204 uint32_t state, uint32_t s_mask);
207 static struct hook snat_hook_create(struct port_watcher *pw);
210 static struct hook stp_hook_create(const struct settings *,
211 struct port_watcher *,
212 struct rconn *local, struct rconn *remote);
214 static struct hook fail_open_hook_create(const struct settings *,
215 struct switch_status *,
217 struct rconn *remote);
218 static struct hook rate_limit_hook_create(const struct settings *,
219 struct switch_status *,
221 struct rconn *remote);
224 static void modify_dhcp_request(struct dhcp_msg *, void *aux);
225 static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux);
228 main(int argc, char *argv[])
232 struct list relays = LIST_INITIALIZER(&relays);
234 struct hook hooks[8];
237 struct pvconn *monitor;
239 struct pvconn *listeners[MAX_MGMT];
242 struct rconn *local_rconn, *remote_rconn;
243 struct relay *controller_relay;
244 struct discovery *discovery;
245 struct switch_status *switch_status;
246 struct port_watcher *pw;
250 set_program_name(argv[0]);
251 register_fault_handlers();
254 parse_options(argc, argv, &s);
255 signal(SIGPIPE, SIG_IGN);
257 /* Start listening for management and monitoring connections. */
259 for (i = 0; i < s.n_listeners; i++) {
260 listeners[n_listeners++] = open_passive_vconn(s.listener_names[i]);
262 monitor = s.monitor_name ? open_passive_vconn(s.monitor_name) : NULL;
264 /* Initialize switch status hook. */
265 hooks[n_hooks++] = switch_status_hook_create(&s, &switch_status);
267 /* Start listening for vlogconf requests. */
268 retval = vlog_server_listen(NULL, NULL);
270 ofp_fatal(retval, "Could not listen for vlog connections");
273 die_if_already_running();
276 VLOG_WARN("OpenFlow reference implementation version %s", VERSION);
277 VLOG_WARN("OpenFlow protocol version 0x%02x", OFP_VERSION);
279 /* Connect to datapath. */
280 local_rconn = rconn_create(0, s.max_backoff);
281 rconn_connect(local_rconn, s.dp_name);
282 switch_status_register_category(switch_status, "local",
283 rconn_status_cb, local_rconn);
285 /* Connect to controller. */
286 remote_rconn = rconn_create(s.probe_interval, s.max_backoff);
287 if (s.controller_name) {
288 retval = rconn_connect(remote_rconn, s.controller_name);
289 if (retval == EAFNOSUPPORT) {
290 ofp_fatal(0, "No support for %s vconn", s.controller_name);
293 switch_status_register_category(switch_status, "remote",
294 rconn_status_cb, remote_rconn);
296 /* Start relaying. */
297 controller_relay = relay_create(local_rconn, remote_rconn, false);
298 list_push_back(&relays, &controller_relay->node);
301 hooks[n_hooks++] = port_watcher_create(local_rconn, remote_rconn, &pw);
302 discovery = s.discovery ? discovery_init(&s, pw, switch_status) : NULL;
304 hooks[n_hooks++] = snat_hook_create(pw);
307 hooks[n_hooks++] = stp_hook_create(&s, pw, local_rconn, remote_rconn);
310 hooks[n_hooks++] = in_band_hook_create(&s, switch_status, pw,
313 if (s.fail_mode == FAIL_OPEN) {
314 hooks[n_hooks++] = fail_open_hook_create(&s, switch_status,
315 local_rconn, remote_rconn);
318 hooks[n_hooks++] = rate_limit_hook_create(&s, switch_status,
319 local_rconn, remote_rconn);
321 assert(n_hooks <= ARRAY_SIZE(hooks));
328 LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) {
329 relay_run(r, hooks, n_hooks);
331 for (i = 0; i < n_listeners; i++) {
333 struct relay *r = relay_accept(&s, listeners[i]);
337 list_push_back(&relays, &r->node);
341 struct vconn *new = accept_vconn(monitor);
343 rconn_add_monitor(local_rconn, new);
346 for (i = 0; i < n_hooks; i++) {
347 if (hooks[i].periodic_cb) {
348 hooks[i].periodic_cb(hooks[i].aux);
352 char *controller_name;
353 if (rconn_is_connectivity_questionable(remote_rconn)) {
354 discovery_question_connectivity(discovery);
356 if (discovery_run(discovery, &controller_name)) {
357 if (controller_name) {
358 rconn_connect(remote_rconn, controller_name);
360 rconn_disconnect(remote_rconn);
365 /* Wait for something to happen. */
366 LIST_FOR_EACH (r, struct relay, node, &relays) {
369 for (i = 0; i < n_listeners; i++) {
370 pvconn_wait(listeners[i]);
373 pvconn_wait(monitor);
375 for (i = 0; i < n_hooks; i++) {
376 if (hooks[i].wait_cb) {
377 hooks[i].wait_cb(hooks[i].aux);
381 discovery_wait(discovery);
389 static struct pvconn *
390 open_passive_vconn(const char *name)
392 struct pvconn *pvconn;
395 retval = pvconn_open(name, &pvconn);
396 if (retval && retval != EAGAIN) {
397 ofp_fatal(retval, "opening %s", name);
402 static struct vconn *
403 accept_vconn(struct pvconn *pvconn)
408 retval = pvconn_accept(pvconn, OFP_VERSION, &new);
409 if (retval && retval != EAGAIN) {
410 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
416 make_hook(bool (*local_packet_cb)(struct relay *, void *aux),
417 bool (*remote_packet_cb)(struct relay *, void *aux),
418 void (*periodic_cb)(void *aux),
419 void (*wait_cb)(void *aux),
423 h.packet_cb[HALF_LOCAL] = local_packet_cb;
424 h.packet_cb[HALF_REMOTE] = remote_packet_cb;
425 h.periodic_cb = periodic_cb;
431 static struct ofp_packet_in *
432 get_ofp_packet_in(struct relay *r)
434 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
435 struct ofp_header *oh = msg->data;
436 if (oh->type == OFPT_PACKET_IN) {
437 if (msg->size >= offsetof (struct ofp_packet_in, data)) {
440 VLOG_WARN("packet too short (%zu bytes) for packet_in",
448 get_ofp_packet_eth_header(struct relay *r, struct ofp_packet_in **opip,
449 struct eth_header **ethp)
451 const int min_len = offsetof(struct ofp_packet_in, data) + ETH_HEADER_LEN;
452 struct ofp_packet_in *opi = get_ofp_packet_in(r);
453 if (opi && ntohs(opi->header.length) >= min_len) {
455 *ethp = (void *) opi->data;
462 /* OpenFlow message relaying. */
464 static struct relay *
465 relay_accept(const struct settings *s, struct pvconn *pvconn)
467 struct vconn *new_remote, *new_local;
468 struct rconn *r1, *r2;
473 new_remote = accept_vconn(pvconn);
478 if (sscanf(s->dp_name, "nl:%d", &nl_index) == 1) {
479 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123.
480 * nl:123:0 opens a netlink connection to local datapath 123 without
481 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
482 * messages. That's what we want here; management connections should
483 * not receive those messages, at least by default. */
484 vconn_name = xasprintf("nl:%d:0", nl_index);
486 /* We don't have a way to specify not to subscribe to those messages
487 * for other transports. (That's a defect: really this should be in
488 * the OpenFlow protocol, not the Netlink transport). */
489 VLOG_WARN_RL(&vrl, "new management connection will receive "
490 "asynchronous messages");
491 vconn_name = xstrdup(s->dp_name);
494 retval = vconn_open(vconn_name, OFP_VERSION, &new_local);
496 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
497 vconn_name, strerror(retval));
498 vconn_close(new_remote);
503 /* Create and return relay. */
504 r1 = rconn_create(0, 0);
505 rconn_connect_unreliably(r1, vconn_name, new_local);
508 r2 = rconn_create(0, 0);
509 rconn_connect_unreliably(r2, "passive", new_remote);
511 return relay_create(r1, r2, true);
514 static struct relay *
515 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
517 struct relay *r = xcalloc(1, sizeof *r);
518 r->halves[HALF_LOCAL].rconn = local;
519 r->halves[HALF_REMOTE].rconn = remote;
520 r->is_mgmt_conn = is_mgmt_conn;
525 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
530 for (i = 0; i < 2; i++) {
531 rconn_run(r->halves[i].rconn);
534 /* Limit the number of iterations to prevent other tasks from starving. */
535 for (iteration = 0; iteration < 50; iteration++) {
536 bool progress = false;
537 for (i = 0; i < 2; i++) {
538 struct half *this = &r->halves[i];
539 struct half *peer = &r->halves[!i];
542 this->rxbuf = rconn_recv(this->rconn);
543 if (this->rxbuf && (i == HALF_REMOTE || !r->is_mgmt_conn)) {
544 const struct hook *h;
545 for (h = hooks; h < &hooks[n_hooks]; h++) {
546 if (h->packet_cb[i] && h->packet_cb[i](r, h->aux)) {
547 ofpbuf_delete(this->rxbuf);
556 if (this->rxbuf && !this->n_txq) {
557 int retval = rconn_send(peer->rconn, this->rxbuf,
559 if (retval != EAGAIN) {
563 ofpbuf_delete(this->rxbuf);
574 if (r->is_mgmt_conn) {
575 for (i = 0; i < 2; i++) {
576 struct half *this = &r->halves[i];
577 if (!rconn_is_alive(this->rconn)) {
586 relay_wait(struct relay *r)
590 for (i = 0; i < 2; i++) {
591 struct half *this = &r->halves[i];
593 rconn_run_wait(this->rconn);
595 rconn_recv_wait(this->rconn);
601 relay_destroy(struct relay *r)
605 list_remove(&r->node);
606 for (i = 0; i < 2; i++) {
607 struct half *this = &r->halves[i];
608 rconn_destroy(this->rconn);
609 ofpbuf_delete(this->rxbuf);
614 /* Port status watcher. */
616 typedef void port_changed_cb_func(uint16_t port_no,
617 const struct ofp_phy_port *old,
618 const struct ofp_phy_port *new,
621 struct port_watcher_cb {
622 port_changed_cb_func *port_changed;
626 typedef void local_port_changed_cb_func(const struct ofp_phy_port *new,
629 struct port_watcher_local_cb {
630 local_port_changed_cb_func *local_port_changed;
634 struct port_watcher {
635 struct rconn *local_rconn;
636 struct rconn *remote_rconn;
637 struct port_array ports;
638 time_t last_feature_request;
639 bool got_feature_reply;
641 struct port_watcher_cb cbs[2];
643 struct port_watcher_local_cb local_cbs[2];
645 char local_port_name[OFP_MAX_PORT_NAME_LEN + 1];
648 /* Returns the number of fields that differ from 'a' to 'b'. */
650 opp_differs(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
652 BUILD_ASSERT_DECL(sizeof *a == 48); /* Trips when we add or remove fields. */
653 return ((a->port_no != b->port_no)
654 + (memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) != 0)
655 + (memcmp(a->name, b->name, sizeof a->name) != 0)
656 + (a->config != b->config)
657 + (a->state != b->state)
658 + (a->curr != b->curr)
659 + (a->advertised != b->advertised)
660 + (a->supported != b->supported)
661 + (a->peer != b->peer));
665 sanitize_opp(struct ofp_phy_port *opp)
669 for (i = 0; i < sizeof opp->name; i++) {
670 char c = opp->name[i];
671 if (c && (c < 0x20 || c > 0x7e)) {
675 opp->name[sizeof opp->name - 1] = '\0';
679 call_port_changed_callbacks(struct port_watcher *pw, int port_no,
680 const struct ofp_phy_port *old,
681 const struct ofp_phy_port *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);
691 get_port_name(const struct ofp_phy_port *port, char *name, size_t name_size)
695 memcpy(name, port->name, MIN(name_size, sizeof port->name));
696 name[name_size - 1] = '\0';
697 for (p = name; *p != '\0'; p++) {
698 if (*p < 32 || *p > 126) {
704 static struct ofp_phy_port *
705 lookup_port(const struct port_watcher *pw, uint16_t port_no)
707 return port_array_get(&pw->ports, port_no);
711 call_local_port_changed_callbacks(struct port_watcher *pw)
713 char name[OFP_MAX_PORT_NAME_LEN + 1];
714 const struct ofp_phy_port *port;
717 /* Pass the local port to the callbacks, if it exists.
718 Pass a null pointer if there is no local port. */
719 port = lookup_port(pw, OFPP_LOCAL);
721 /* Log the name of the local port. */
723 get_port_name(port, name, sizeof name);
727 if (strcmp(pw->local_port_name, name)) {
729 VLOG_WARN("Identified data path local port as \"%s\".", name);
731 VLOG_WARN("Data path has no local port.");
733 strcpy(pw->local_port_name, name);
736 /* Invoke callbacks. */
737 for (i = 0; i < pw->n_local_cbs; i++) {
738 local_port_changed_cb_func *cb = pw->local_cbs[i].local_port_changed;
739 (cb)(port, pw->local_cbs[i].aux);
744 update_phy_port(struct port_watcher *pw, struct ofp_phy_port *opp,
747 struct ofp_phy_port *old;
750 port_no = ntohs(opp->port_no);
751 old = lookup_port(pw, port_no);
753 if (reason == OFPPR_DELETE && old) {
754 call_port_changed_callbacks(pw, port_no, old, NULL);
756 port_array_set(&pw->ports, port_no, NULL);
757 } else if ((reason == OFPPR_MODIFY || reason == OFPPR_ADD)
758 && (!old || opp_differs(opp, old))) {
759 struct ofp_phy_port new = *opp;
761 call_port_changed_callbacks(pw, port_no, old, &new);
765 port_array_set(&pw->ports, port_no, xmemdup(&new, sizeof new));
771 port_watcher_local_packet_cb(struct relay *r, void *pw_)
773 struct port_watcher *pw = pw_;
774 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
775 struct ofp_header *oh = msg->data;
777 if (oh->type == OFPT_FEATURES_REPLY
778 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
779 struct ofp_switch_features *osf = msg->data;
780 bool seen[PORT_ARRAY_SIZE];
781 struct ofp_phy_port *p;
782 unsigned int port_no;
786 pw->got_feature_reply = true;
788 /* Update each port included in the message. */
789 memset(seen, false, 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);
795 seen[ntohs(opp->port_no)] = true;
798 /* Delete all the ports not included in the message. */
799 for (p = port_array_first(&pw->ports, &port_no); p;
800 p = port_array_next(&pw->ports, &port_no)) {
801 if (!seen[port_no]) {
802 update_phy_port(pw, p, OFPPR_DELETE);
806 call_local_port_changed_callbacks(pw);
807 } else if (oh->type == OFPT_PORT_STATUS
808 && msg->size >= sizeof(struct ofp_port_status)) {
809 struct ofp_port_status *ops = msg->data;
810 update_phy_port(pw, &ops->desc, ops->reason);
811 if (ops->desc.port_no == htons(OFPP_LOCAL)) {
812 call_local_port_changed_callbacks(pw);
819 port_watcher_remote_packet_cb(struct relay *r, void *pw_)
821 struct port_watcher *pw = pw_;
822 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
823 struct ofp_header *oh = msg->data;
825 if (oh->type == OFPT_PORT_MOD
826 && msg->size >= sizeof(struct ofp_port_mod)) {
827 struct ofp_port_mod *opm = msg->data;
828 uint16_t port_no = ntohs(opm->port_no);
829 struct ofp_phy_port *pw_opp = lookup_port(pw, port_no);
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);
844 port_watcher_periodic_cb(void *pw_)
846 struct port_watcher *pw = pw_;
848 if (!pw->got_feature_reply
849 && time_now() >= pw->last_feature_request + 5
850 && rconn_is_connected(pw->local_rconn)) {
852 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
853 rconn_send_with_limit(pw->local_rconn, b, &pw->n_txq, 1);
854 pw->last_feature_request = time_now();
859 port_watcher_wait_cb(void *pw_)
861 struct port_watcher *pw = pw_;
862 if (!pw->got_feature_reply && rconn_is_connected(pw->local_rconn)) {
863 if (pw->last_feature_request != TIME_MIN) {
864 poll_timer_wait(pw->last_feature_request + 5 - time_now());
866 poll_immediate_wake();
872 put_duplexes(struct ds *ds, const char *name, uint32_t features,
873 uint32_t hd_bit, uint32_t fd_bit)
875 if (features & (hd_bit | fd_bit)) {
876 ds_put_format(ds, " %s", name);
877 if (features & hd_bit) {
878 ds_put_cstr(ds, "(HD)");
880 if (features & fd_bit) {
881 ds_put_cstr(ds, "(FD)");
887 put_features(struct ds *ds, const char *name, uint32_t features)
889 if (features & (OFPPF_10MB_HD | OFPPF_10MB_FD
890 | OFPPF_100MB_HD | OFPPF_100MB_FD
891 | OFPPF_1GB_HD | OFPPF_1GB_FD | OFPPF_10GB_FD)) {
892 ds_put_cstr(ds, name);
893 put_duplexes(ds, "10M", features, OFPPF_10MB_HD, OFPPF_10MB_FD);
894 put_duplexes(ds, "100M", features,
895 OFPPF_100MB_HD, OFPPF_100MB_FD);
896 put_duplexes(ds, "1G", features, OFPPF_100MB_HD, OFPPF_100MB_FD);
897 if (features & OFPPF_10GB_FD) {
898 ds_put_cstr(ds, " 10G");
900 if (features & OFPPF_AUTONEG) {
901 ds_put_cstr(ds, " AUTO_NEG");
903 if (features & OFPPF_PAUSE) {
904 ds_put_cstr(ds, " PAUSE");
906 if (features & OFPPF_PAUSE_ASYM) {
907 ds_put_cstr(ds, " PAUSE_ASYM");
913 log_port_status(uint16_t port_no,
914 const struct ofp_phy_port *old,
915 const struct ofp_phy_port *new,
918 if (VLOG_IS_DBG_ENABLED()) {
919 if (old && new && (opp_differs(old, new)
920 == ((old->config != new->config)
921 + (old->state != new->state))))
923 /* Don't care if only state or config changed. */
926 VLOG_DBG("Port %d deleted", port_no);
929 struct ds ds = DS_EMPTY_INITIALIZER;
930 uint32_t curr = ntohl(new->curr);
931 uint32_t supported = ntohl(new->supported);
932 ds_put_format(&ds, "\"%s\", "ETH_ADDR_FMT, new->name,
933 ETH_ADDR_ARGS(new->hw_addr));
935 put_features(&ds, ", current", curr);
938 put_features(&ds, ", supports", supported);
940 VLOG_DBG("Port %d %s: %s",
941 port_no, old ? "changed" : "added", ds_cstr(&ds));
948 port_watcher_register_callback(struct port_watcher *pw,
949 port_changed_cb_func *port_changed,
952 assert(pw->n_cbs < ARRAY_SIZE(pw->cbs));
953 pw->cbs[pw->n_cbs].port_changed = port_changed;
954 pw->cbs[pw->n_cbs].aux = aux;
959 port_watcher_register_local_port_callback(struct port_watcher *pw,
960 local_port_changed_cb_func *cb,
963 assert(pw->n_local_cbs < ARRAY_SIZE(pw->local_cbs));
964 pw->local_cbs[pw->n_local_cbs].local_port_changed = cb;
965 pw->local_cbs[pw->n_local_cbs].aux = aux;
970 port_watcher_get_config(const struct port_watcher *pw, uint16_t port_no)
972 struct ofp_phy_port *p = lookup_port(pw, port_no);
973 return p ? ntohl(p->config) : 0;
977 port_watcher_get_name(const struct port_watcher *pw, uint16_t port_no)
979 struct ofp_phy_port *p = lookup_port(pw, port_no);
980 return p ? (const char *) p->name : NULL;
983 static const uint8_t *
984 port_watcher_get_hwaddr(const struct port_watcher *pw, uint16_t port_no)
986 struct ofp_phy_port *p = lookup_port(pw, port_no);
987 return p ? p->hw_addr : NULL;
991 port_watcher_set_flags(struct port_watcher *pw, uint16_t port_no,
992 uint32_t config, uint32_t c_mask,
993 uint32_t state, uint32_t s_mask)
995 struct ofp_phy_port old;
996 struct ofp_phy_port *p;
997 struct ofp_port_mod *opm;
998 struct ofp_port_status *ops;
1001 p = lookup_port(pw, port_no);
1003 VLOG_WARN_RL(&vrl, "attempting to set flags on nonexistent port %"
1008 if (!((ntohl(p->state) ^ state) & s_mask)
1009 && (!((ntohl(p->config) ^ config) & c_mask))) {
1014 /* Update our idea of the flags. */
1015 p->config = htonl((ntohl(p->config) & ~c_mask) | (config & c_mask));
1016 p->state = htonl((ntohl(p->state) & ~s_mask) | (state & s_mask));
1017 call_port_changed_callbacks(pw, port_no, &old, p);
1019 /* Change the flags in the datapath. */
1020 opm = make_openflow(sizeof *opm, OFPT_PORT_MOD, &b);
1021 opm->port_no = p->port_no;
1022 memcpy(opm->hw_addr, p->hw_addr, OFP_ETH_ALEN);
1023 opm->config = p->config;
1024 opm->mask = htonl(c_mask);
1025 opm->advertise = htonl(0);
1026 rconn_send(pw->local_rconn, b, NULL);
1028 /* Notify the controller that the flags changed. */
1029 ops = make_openflow(sizeof *ops, OFPT_PORT_STATUS, &b);
1030 ops->reason = OFPPR_MODIFY;
1032 rconn_send(pw->remote_rconn, b, NULL);
1036 port_watcher_is_ready(const struct port_watcher *pw)
1038 return pw->got_feature_reply;
1042 port_watcher_create(struct rconn *local_rconn, struct rconn *remote_rconn,
1043 struct port_watcher **pwp)
1045 struct port_watcher *pw;
1047 pw = *pwp = xcalloc(1, sizeof *pw);
1048 pw->local_rconn = local_rconn;
1049 pw->remote_rconn = remote_rconn;
1050 pw->last_feature_request = TIME_MIN;
1051 port_array_init(&pw->ports);
1052 pw->local_port_name[0] = '\0';
1053 port_watcher_register_callback(pw, log_port_status, NULL);
1054 return make_hook(port_watcher_local_packet_cb,
1055 port_watcher_remote_packet_cb,
1056 port_watcher_periodic_cb,
1057 port_watcher_wait_cb, pw);
1061 struct snat_port_conf {
1063 struct nx_snat_config config;
1067 struct port_watcher *pw;
1068 struct list port_list;
1072 /* Source-NAT configuration monitor. */
1073 #define SNAT_CMD_LEN 1024
1075 /* Commands to configure iptables. There is no programmatic interface
1076 * to iptables from the kernel, so we're stuck making command-line calls
1078 #define SNAT_FLUSH_ALL_CMD "/sbin/iptables -t nat -F"
1079 #define SNAT_FLUSH_CHAIN_CMD "/sbin/iptables -t nat -F of-snat-%s"
1081 #define SNAT_ADD_CHAIN_CMD "/sbin/iptables -t nat -N of-snat-%s"
1082 #define SNAT_CONF_CHAIN_CMD "/sbin/iptables -t nat -A POSTROUTING -o %s -j of-snat-%s"
1084 #define SNAT_ADD_IP_CMD "/sbin/iptables -t nat -A of-snat-%s -j SNAT --to %s-%s"
1085 #define SNAT_ADD_TCP_CMD "/sbin/iptables -t nat -A of-snat-%s -j SNAT -p TCP --to %s-%s:%d-%d"
1086 #define SNAT_ADD_UDP_CMD "/sbin/iptables -t nat -A of-snat-%s -j SNAT -p UDP --to %s-%s:%d-%d"
1088 #define SNAT_UNSET_CHAIN_CMD "/sbin/iptables -t nat -D POSTROUTING -o %s -j of-snat-%s"
1089 #define SNAT_DEL_CHAIN_CMD "/sbin/iptables -t nat -X of-snat-%s"
1092 snat_add_rules(const struct nx_snat_config *sc, const uint8_t *dev_name)
1094 char command[SNAT_CMD_LEN];
1095 char ip_str_start[16];
1096 char ip_str_end[16];
1099 snprintf(ip_str_start, sizeof ip_str_start, IP_FMT,
1100 IP_ARGS(&sc->ip_addr_start));
1101 snprintf(ip_str_end, sizeof ip_str_end, IP_FMT,
1102 IP_ARGS(&sc->ip_addr_end));
1104 /* We always attempt to remove existing entries, so that we know
1105 * there's a pristine state for SNAT on the interface. We just ignore
1106 * the results of these calls, since iptables will complain about
1107 * any non-existent entries. */
1109 /* Flush the chain that does the SNAT. */
1110 snprintf(command, sizeof(command), SNAT_FLUSH_CHAIN_CMD, dev_name);
1113 /* We always try to create the a new chain. */
1114 snprintf(command, sizeof(command), SNAT_ADD_CHAIN_CMD, dev_name);
1117 /* Disassociate any old SNAT chain from the POSTROUTING chain. */
1118 snprintf(command, sizeof(command), SNAT_UNSET_CHAIN_CMD, dev_name,
1122 /* Associate the new chain with the POSTROUTING hook. */
1123 snprintf(command, sizeof(command), SNAT_CONF_CHAIN_CMD, dev_name,
1125 if (system(command) != 0) {
1126 VLOG_ERR("SNAT: problem flushing chain for add");
1130 /* If configured, restrict TCP source port ranges. */
1131 if ((sc->tcp_start != 0) && (sc->tcp_end != 0)) {
1132 snprintf(command, sizeof(command), SNAT_ADD_TCP_CMD,
1133 dev_name, ip_str_start, ip_str_end,
1134 ntohs(sc->tcp_start), ntohs(sc->tcp_end));
1135 if (system(command) != 0) {
1136 VLOG_ERR("SNAT: problem adding TCP rule");
1141 /* If configured, restrict UDP source port ranges. */
1142 if ((sc->udp_start != 0) && (sc->udp_end != 0)) {
1143 snprintf(command, sizeof(command), SNAT_ADD_UDP_CMD,
1144 dev_name, ip_str_start, ip_str_end,
1145 ntohs(sc->udp_start), ntohs(sc->udp_end));
1146 if (system(command) != 0) {
1147 VLOG_ERR("SNAT: problem adding UDP rule");
1152 /* Add a rule that covers all IP traffic that would not be covered
1153 * by the prior TCP or UDP ranges. */
1154 snprintf(command, sizeof(command), SNAT_ADD_IP_CMD,
1155 dev_name, ip_str_start, ip_str_end);
1156 if (system(command) != 0) {
1157 VLOG_ERR("SNAT: problem adding base rule");
1163 snat_del_rules(const uint8_t *dev_name)
1165 char command[SNAT_CMD_LEN];
1167 /* Flush the chain that does the SNAT. */
1168 snprintf(command, sizeof(command), SNAT_FLUSH_CHAIN_CMD, dev_name);
1169 if (system(command) != 0) {
1170 VLOG_ERR("SNAT: problem flushing chain for deletion");
1174 /* Disassociate the SNAT chain from the POSTROUTING chain. */
1175 snprintf(command, sizeof(command), SNAT_UNSET_CHAIN_CMD, dev_name,
1177 if (system(command) != 0) {
1178 VLOG_ERR("SNAT: problem unsetting chain");
1182 /* Now we can finally delete our SNAT chain. */
1183 snprintf(command, sizeof(command), SNAT_DEL_CHAIN_CMD, dev_name);
1184 if (system(command) != 0) {
1185 VLOG_ERR("SNAT: problem deleting chain");
1191 snat_config(const struct nx_snat_config *sc, struct snat_data *snat)
1193 struct snat_port_conf *c, *spc=NULL;
1194 const uint8_t *netdev_name;
1196 netdev_name = (const uint8_t *) port_watcher_get_name(snat->pw,
1202 LIST_FOR_EACH(c, struct snat_port_conf, node, &snat->port_list) {
1203 if (c->config.port == sc->port) {
1209 if (sc->command == NXSC_ADD) {
1211 spc = xmalloc(sizeof(*c));
1213 VLOG_ERR("SNAT: no memory for new entry");
1216 list_push_back(&snat->port_list, &spc->node);
1218 memcpy(&spc->config, sc, sizeof(spc->config));
1219 snat_add_rules(sc, netdev_name);
1221 snat_del_rules(netdev_name);
1222 list_remove(&spc->node);
1227 snat_remote_packet_cb(struct relay *r, void *snat_)
1229 struct snat_data *snat = snat_;
1230 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
1231 struct nicira_header *request = msg->data;
1232 struct nx_act_config *nac = msg->data;
1236 if (msg->size < sizeof(struct nx_act_config)) {
1239 request = msg->data;
1240 if (request->header.type != OFPT_VENDOR
1241 || request->vendor != htonl(NX_VENDOR_ID)
1242 || request->subtype != htonl(NXT_ACT_SET_CONFIG)) {
1246 /* We're only interested in attempts to configure SNAT */
1247 if (nac->type != htons(NXAST_SNAT)) {
1251 n_configs = (msg->size - sizeof *nac) / sizeof *nac->snat;
1252 for (i=0; i<n_configs; i++) {
1253 snat_config(&nac->snat[i], snat);
1260 snat_port_changed_cb(uint16_t port_no,
1261 const struct ofp_phy_port *old,
1262 const struct ofp_phy_port *new,
1265 struct snat_data *snat = snat_;
1266 struct snat_port_conf *c;
1268 /* We're only interested in ports that went away */
1273 LIST_FOR_EACH(c, struct snat_port_conf, node, &snat->port_list) {
1274 if (c->config.port == old->port_no) {
1275 snat_del_rules(old->name);
1276 list_remove(&c->node);
1283 snat_hook_create(struct port_watcher *pw)
1286 struct snat_data *snat;
1288 ret = system(SNAT_FLUSH_ALL_CMD);
1290 VLOG_ERR("SNAT: problem flushing tables");
1293 snat = xcalloc(1, sizeof *snat);
1295 list_init(&snat->port_list);
1297 port_watcher_register_callback(pw, snat_port_changed_cb, snat);
1298 return make_hook(NULL, snat_remote_packet_cb, NULL, NULL, snat);
1300 #endif /* SUPPORT_SNAT */
1302 /* Spanning tree protocol. */
1304 /* Extra time, in seconds, at boot before going into fail-open, to give the
1305 * spanning tree protocol time to figure out the network layout. */
1306 #define STP_EXTRA_BOOT_TIME 30
1310 struct port_watcher *pw;
1311 struct rconn *local_rconn;
1312 struct rconn *remote_rconn;
1313 long long int last_tick_256ths;
1318 stp_local_packet_cb(struct relay *r, void *stp_)
1320 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
1321 struct ofp_header *oh;
1322 struct stp_data *stp = stp_;
1323 struct ofp_packet_in *opi;
1324 struct eth_header *eth;
1325 struct llc_header *llc;
1326 struct ofpbuf payload;
1331 if (oh->type == OFPT_FEATURES_REPLY
1332 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
1333 struct ofp_switch_features *osf = msg->data;
1334 osf->capabilities |= htonl(OFPC_STP);
1338 if (!get_ofp_packet_eth_header(r, &opi, ð)
1339 || !eth_addr_equals(eth->eth_dst, stp_eth_addr)) {
1343 port_no = ntohs(opi->in_port);
1344 if (port_no >= STP_MAX_PORTS) {
1345 /* STP only supports 255 ports. */
1348 if (port_watcher_get_config(stp->pw, port_no) & OFPPC_NO_STP) {
1349 /* We're not doing STP on this port. */
1353 if (opi->reason == OFPR_ACTION) {
1354 /* The controller set up a flow for this, so we won't intercept it. */
1358 get_ofp_packet_payload(opi, &payload);
1359 flow_extract(&payload, port_no, &flow);
1360 if (flow.dl_type != htons(OFP_DL_TYPE_NOT_ETH_TYPE)) {
1361 VLOG_DBG("non-LLC frame received on STP multicast address");
1364 llc = ofpbuf_at_assert(&payload, sizeof *eth, sizeof *llc);
1365 if (llc->llc_dsap != STP_LLC_DSAP) {
1366 VLOG_DBG("bad DSAP 0x%02"PRIx8" received on STP multicast address",
1371 /* Trim off padding on payload. */
1372 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1373 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1375 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1376 struct stp_port *p = stp_get_port(stp->stp, port_no);
1377 stp_received_bpdu(p, payload.data, payload.size);
1383 static long long int
1386 return time_msec() * 256 / 1000;
1390 stp_periodic_cb(void *stp_)
1392 struct stp_data *stp = stp_;
1393 long long int now_256ths = time_256ths();
1394 long long int elapsed_256ths = now_256ths - stp->last_tick_256ths;
1397 if (!port_watcher_is_ready(stp->pw)) {
1398 /* Can't start STP until we know port flags, because port flags can
1402 if (elapsed_256ths <= 0) {
1406 stp_tick(stp->stp, MIN(INT_MAX, elapsed_256ths));
1407 stp->last_tick_256ths = now_256ths;
1409 while (stp_get_changed_port(stp->stp, &p)) {
1410 int port_no = stp_port_no(p);
1411 enum stp_state s_state = stp_port_get_state(p);
1413 if (s_state != STP_DISABLED) {
1414 VLOG_WARN("STP: Port %d entered %s state",
1415 port_no, stp_state_name(s_state));
1417 if (!(port_watcher_get_config(stp->pw, port_no) & OFPPC_NO_STP)) {
1418 uint32_t p_config = 0;
1422 p_state = OFPPS_STP_LISTEN;
1425 p_state = OFPPS_STP_LEARN;
1428 case STP_FORWARDING:
1429 p_state = OFPPS_STP_FORWARD;
1432 p_state = OFPPS_STP_BLOCK;
1435 VLOG_DBG_RL(&vrl, "STP: Port %d has bad state %x",
1437 p_state = OFPPS_STP_FORWARD;
1440 if (!stp_forward_in_state(s_state)) {
1441 p_config = OFPPC_NO_FLOOD;
1443 port_watcher_set_flags(stp->pw, port_no,
1444 p_config, OFPPC_NO_FLOOD,
1445 p_state, OFPPS_STP_MASK);
1447 /* We don't own those flags. */
1453 stp_wait_cb(void *stp_ UNUSED)
1455 poll_timer_wait(1000);
1459 send_bpdu(const void *bpdu, size_t bpdu_size, int port_no, void *stp_)
1461 struct stp_data *stp = stp_;
1462 const uint8_t *port_mac;
1463 struct eth_header *eth;
1464 struct llc_header *llc;
1465 struct ofpbuf pkt, *opo;
1467 port_mac = port_watcher_get_hwaddr(stp->pw, port_no);
1469 VLOG_WARN_RL(&vrl, "cannot send BPDU on missing port %d", port_no);
1473 /* Packet skeleton. */
1474 ofpbuf_init(&pkt, ETH_HEADER_LEN + LLC_HEADER_LEN + bpdu_size);
1475 eth = ofpbuf_put_uninit(&pkt, sizeof *eth);
1476 llc = ofpbuf_put_uninit(&pkt, sizeof *llc);
1477 ofpbuf_put(&pkt, bpdu, bpdu_size);
1480 memcpy(eth->eth_dst, stp_eth_addr, ETH_ADDR_LEN);
1481 memcpy(eth->eth_src, port_mac, ETH_ADDR_LEN);
1482 eth->eth_type = htons(pkt.size - ETH_HEADER_LEN);
1485 llc->llc_dsap = STP_LLC_DSAP;
1486 llc->llc_ssap = STP_LLC_SSAP;
1487 llc->llc_cntl = STP_LLC_CNTL;
1489 opo = make_unbuffered_packet_out(&pkt, OFPP_NONE, port_no);
1490 ofpbuf_uninit(&pkt);
1491 rconn_send_with_limit(stp->local_rconn, opo, &stp->n_txq, OFPP_MAX);
1495 stp_is_port_supported(uint16_t port_no)
1497 return port_no < STP_MAX_PORTS;
1501 stp_port_changed_cb(uint16_t port_no,
1502 const struct ofp_phy_port *old,
1503 const struct ofp_phy_port *new,
1506 struct stp_data *stp = stp_;
1509 if (!stp_is_port_supported(port_no)) {
1513 p = stp_get_port(stp->stp, port_no);
1515 || new->config & htonl(OFPPC_NO_STP | OFPPC_PORT_DOWN)
1516 || new->state & htonl(OFPPS_LINK_DOWN)) {
1517 stp_port_disable(p);
1521 if (new->curr & (OFPPF_10MB_HD | OFPPF_10MB_FD)) {
1523 } else if (new->curr & (OFPPF_100MB_HD | OFPPF_100MB_FD)) {
1525 } else if (new->curr & (OFPPF_1GB_HD | OFPPF_1GB_FD)) {
1527 } else if (new->curr & OFPPF_100MB_FD) {
1530 stp_port_set_speed(p, speed);
1535 stp_local_port_changed_cb(const struct ofp_phy_port *port, void *stp_)
1537 struct stp_data *stp = stp_;
1539 stp_set_bridge_id(stp->stp, eth_addr_to_uint64(port->hw_addr));
1544 stp_hook_create(const struct settings *s, struct port_watcher *pw,
1545 struct rconn *local, struct rconn *remote)
1547 uint8_t dpid[ETH_ADDR_LEN];
1548 struct stp_data *stp;
1550 stp = xcalloc(1, sizeof *stp);
1551 eth_addr_random(dpid);
1552 stp->stp = stp_create("stp", eth_addr_to_uint64(dpid), send_bpdu, stp);
1554 stp->local_rconn = local;
1555 stp->remote_rconn = remote;
1556 stp->last_tick_256ths = time_256ths();
1558 port_watcher_register_callback(pw, stp_port_changed_cb, stp);
1559 port_watcher_register_local_port_callback(pw, stp_local_port_changed_cb,
1561 return make_hook(stp_local_packet_cb, NULL,
1562 stp_periodic_cb, stp_wait_cb, stp);
1565 /* In-band control. */
1567 struct in_band_data {
1568 const struct settings *s;
1569 struct mac_learning *ml;
1570 struct netdev *of_device;
1571 struct rconn *controller;
1576 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct ofpbuf *b)
1578 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
1581 static const uint8_t *
1582 get_controller_mac(struct in_band_data *in_band)
1584 static uint32_t ip, last_nonzero_ip;
1585 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
1586 static time_t next_refresh = 0;
1588 uint32_t last_ip = ip;
1590 time_t now = time_now();
1592 ip = rconn_get_ip(in_band->controller);
1593 if (last_ip != ip || !next_refresh || now >= next_refresh) {
1596 /* Look up MAC address. */
1597 memset(mac, 0, sizeof mac);
1598 if (ip && in_band->of_device) {
1599 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
1601 VLOG_DBG_RL(&vrl, "cannot look up controller hw address "
1602 "("IP_FMT"): %s", IP_ARGS(&ip), strerror(retval));
1605 have_mac = !eth_addr_is_zero(mac);
1607 /* Log changes in IP, MAC addresses. */
1608 if (ip && ip != last_nonzero_ip) {
1609 VLOG_DBG("controller IP address changed from "IP_FMT
1610 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
1611 last_nonzero_ip = ip;
1613 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
1614 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
1616 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
1617 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
1620 /* Schedule next refresh.
1622 * If we have an IP address but not a MAC address, then refresh
1623 * quickly, since we probably will get a MAC address soon (via ARP).
1624 * Otherwise, we can afford to wait a little while. */
1625 next_refresh = now + (!ip || have_mac ? 10 : 1);
1627 return !eth_addr_is_zero(mac) ? mac : NULL;
1631 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
1632 struct in_band_data *in_band)
1634 const uint8_t *mac = get_controller_mac(in_band);
1635 return mac && eth_addr_equals(mac, dl_addr);
1639 in_band_learn_mac(struct in_band_data *in_band,
1640 uint16_t in_port, const uint8_t src_mac[ETH_ADDR_LEN])
1642 if (mac_learning_learn(in_band->ml, src_mac, in_port)) {
1643 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
1644 ETH_ADDR_ARGS(src_mac), in_port);
1649 in_band_local_packet_cb(struct relay *r, void *in_band_)
1651 struct in_band_data *in_band = in_band_;
1652 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
1653 struct ofp_packet_in *opi;
1654 struct eth_header *eth;
1655 struct ofpbuf payload;
1660 if (!get_ofp_packet_eth_header(r, &opi, ð) || !in_band->of_device) {
1663 in_port = ntohs(opi->in_port);
1665 /* Deal with local stuff. */
1666 if (in_port == OFPP_LOCAL) {
1667 /* Sent by secure channel. */
1668 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1669 } else if (eth_addr_equals(eth->eth_dst,
1670 netdev_get_etheraddr(in_band->of_device))) {
1671 /* Sent to secure channel. */
1672 out_port = OFPP_LOCAL;
1673 in_band_learn_mac(in_band, in_port, eth->eth_src);
1674 } else if (eth->eth_type == htons(ETH_TYPE_ARP)
1675 && eth_addr_is_broadcast(eth->eth_dst)
1676 && is_controller_mac(eth->eth_src, in_band)) {
1677 /* ARP sent by controller. */
1678 out_port = OFPP_FLOOD;
1679 } else if (is_controller_mac(eth->eth_dst, in_band)
1680 || is_controller_mac(eth->eth_src, in_band)) {
1681 /* Traffic to or from controller. Switch it by hand. */
1682 in_band_learn_mac(in_band, in_port, eth->eth_src);
1683 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1685 const uint8_t *controller_mac;
1686 controller_mac = get_controller_mac(in_band);
1687 if (eth->eth_type == htons(ETH_TYPE_ARP)
1688 && eth_addr_is_broadcast(eth->eth_dst)
1689 && is_controller_mac(eth->eth_src, in_band)) {
1690 /* ARP sent by controller. */
1691 out_port = OFPP_FLOOD;
1692 } else if (is_controller_mac(eth->eth_dst, in_band)
1693 && in_port == mac_learning_lookup(in_band->ml,
1695 /* Drop controller traffic that arrives on the controller port. */
1702 get_ofp_packet_payload(opi, &payload);
1703 flow_extract(&payload, in_port, &flow);
1704 if (in_port == out_port) {
1705 /* The input and output port match. Set up a flow to drop packets. */
1706 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
1707 in_band->s->max_idle, 0));
1708 } else if (out_port != OFPP_FLOOD) {
1709 /* The output port is known, so add a new flow. */
1710 queue_tx(rc, in_band,
1711 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
1712 out_port, in_band->s->max_idle));
1714 /* If the switch didn't buffer the packet, we need to send a copy. */
1715 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1716 queue_tx(rc, in_band,
1717 make_unbuffered_packet_out(&payload, in_port, out_port));
1720 /* We don't know that MAC. Send along the packet without setting up a
1723 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1724 b = make_unbuffered_packet_out(&payload, in_port, out_port);
1726 b = make_buffered_packet_out(ntohl(opi->buffer_id),
1729 queue_tx(rc, in_band, b);
1735 in_band_status_cb(struct status_reply *sr, void *in_band_)
1737 struct in_band_data *in_band = in_band_;
1738 struct in_addr local_ip;
1739 uint32_t controller_ip;
1740 const uint8_t *controller_mac;
1742 if (in_band->of_device) {
1743 const uint8_t *mac = netdev_get_etheraddr(in_band->of_device);
1744 if (netdev_get_in4(in_band->of_device, &local_ip)) {
1745 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
1747 status_reply_put(sr, "local-mac="ETH_ADDR_FMT, ETH_ADDR_ARGS(mac));
1749 controller_ip = rconn_get_ip(in_band->controller);
1750 if (controller_ip) {
1751 status_reply_put(sr, "controller-ip="IP_FMT,
1752 IP_ARGS(&controller_ip));
1754 controller_mac = get_controller_mac(in_band);
1755 if (controller_mac) {
1756 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
1757 ETH_ADDR_ARGS(controller_mac));
1763 get_ofp_packet_payload(struct ofp_packet_in *opi, struct ofpbuf *payload)
1765 payload->data = opi->data;
1766 payload->size = ntohs(opi->header.length) - offsetof(struct ofp_packet_in,
1771 in_band_local_port_cb(const struct ofp_phy_port *port, void *in_band_)
1773 struct in_band_data *in_band = in_band_;
1775 char name[sizeof port->name + 1];
1776 get_port_name(port, name, sizeof name);
1778 if (!in_band->of_device
1779 || strcmp(netdev_get_name(in_band->of_device), name))
1782 netdev_close(in_band->of_device);
1783 error = netdev_open(name, NETDEV_ETH_TYPE_NONE,
1784 &in_band->of_device);
1786 VLOG_ERR("failed to open in-band control network device "
1787 "\"%s\": %s", name, strerror(errno));
1791 netdev_close(in_band->of_device);
1792 in_band->of_device = NULL;
1797 in_band_hook_create(const struct settings *s, struct switch_status *ss,
1798 struct port_watcher *pw, struct rconn *remote)
1800 struct in_band_data *in_band;
1802 in_band = xcalloc(1, sizeof *in_band);
1804 in_band->ml = mac_learning_create();
1805 in_band->of_device = NULL;
1806 in_band->controller = remote;
1807 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
1808 port_watcher_register_local_port_callback(pw, in_band_local_port_cb,
1810 return make_hook(in_band_local_packet_cb, NULL, NULL, NULL, in_band);
1813 /* Fail open support. */
1815 struct fail_open_data {
1816 const struct settings *s;
1817 struct rconn *local_rconn;
1818 struct rconn *remote_rconn;
1819 struct lswitch *lswitch;
1820 int last_disconn_secs;
1821 time_t boot_deadline;
1824 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
1826 fail_open_periodic_cb(void *fail_open_)
1828 struct fail_open_data *fail_open = fail_open_;
1832 if (time_now() < fail_open->boot_deadline) {
1835 disconn_secs = rconn_failure_duration(fail_open->remote_rconn);
1836 open = disconn_secs >= fail_open->s->probe_interval * 3;
1837 if (open != (fail_open->lswitch != NULL)) {
1839 VLOG_WARN("No longer in fail-open mode");
1840 lswitch_destroy(fail_open->lswitch);
1841 fail_open->lswitch = NULL;
1843 VLOG_WARN("Could not connect to controller for %d seconds, "
1844 "failing open", disconn_secs);
1845 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
1846 fail_open->s->max_idle);
1847 fail_open->last_disconn_secs = disconn_secs;
1849 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
1850 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
1851 "from controller", disconn_secs);
1852 fail_open->last_disconn_secs = disconn_secs;
1857 fail_open_local_packet_cb(struct relay *r, void *fail_open_)
1859 struct fail_open_data *fail_open = fail_open_;
1860 if (!fail_open->lswitch) {
1863 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
1864 r->halves[HALF_LOCAL].rxbuf);
1865 rconn_run(fail_open->local_rconn);
1871 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
1873 struct fail_open_data *fail_open = fail_open_;
1874 const struct settings *s = fail_open->s;
1875 int trigger_duration = s->probe_interval * 3;
1876 int cur_duration = rconn_failure_duration(fail_open->remote_rconn);
1878 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
1879 status_reply_put(sr, "current-duration=%d", cur_duration);
1880 status_reply_put(sr, "triggered=%s",
1881 cur_duration >= trigger_duration ? "true" : "false");
1882 status_reply_put(sr, "max-idle=%d", s->max_idle);
1886 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
1887 struct rconn *local_rconn, struct rconn *remote_rconn)
1889 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
1891 fail_open->local_rconn = local_rconn;
1892 fail_open->remote_rconn = remote_rconn;
1893 fail_open->lswitch = NULL;
1894 fail_open->boot_deadline = time_now() + s->probe_interval * 3;
1895 if (s->enable_stp) {
1896 fail_open->boot_deadline += STP_EXTRA_BOOT_TIME;
1898 switch_status_register_category(ss, "fail-open",
1899 fail_open_status_cb, fail_open);
1900 return make_hook(fail_open_local_packet_cb, NULL,
1901 fail_open_periodic_cb, NULL, fail_open);
1904 struct rate_limiter {
1905 const struct settings *s;
1906 struct rconn *remote_rconn;
1908 /* One queue per physical port. */
1909 struct ofp_queue queues[OFPP_MAX];
1910 int n_queued; /* Sum over queues[*].n. */
1911 int next_tx_port; /* Next port to check in round-robin. */
1915 * It costs 1000 tokens to send a single packet_in message. A single token
1916 * per message would be more straightforward, but this choice lets us avoid
1917 * round-off error in refill_bucket()'s calculation of how many tokens to
1918 * add to the bucket, since no division step is needed. */
1919 long long int last_fill; /* Time at which we last added tokens. */
1920 int tokens; /* Current number of tokens. */
1922 /* Transmission queue. */
1923 int n_txq; /* No. of packets waiting in rconn for tx. */
1925 /* Statistics reporting. */
1926 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
1927 unsigned long long n_limited; /* # queued for rate limiting. */
1928 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
1929 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
1932 /* Drop a packet from the longest queue in 'rl'. */
1934 drop_packet(struct rate_limiter *rl)
1936 struct ofp_queue *longest; /* Queue currently selected as longest. */
1937 int n_longest; /* # of queues of same length as 'longest'. */
1938 struct ofp_queue *q;
1940 longest = &rl->queues[0];
1942 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
1943 if (longest->n < q->n) {
1946 } else if (longest->n == q->n) {
1949 /* Randomly select one of the longest queues, with a uniform
1950 * distribution (Knuth algorithm 3.4.2R). */
1951 if (!random_range(n_longest)) {
1957 /* FIXME: do we want to pop the tail instead? */
1958 ofpbuf_delete(queue_pop_head(longest));
1962 /* Remove and return the next packet to transmit (in round-robin order). */
1963 static struct ofpbuf *
1964 dequeue_packet(struct rate_limiter *rl)
1968 for (i = 0; i < OFPP_MAX; i++) {
1969 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
1970 struct ofp_queue *q = &rl->queues[port];
1972 rl->next_tx_port = (port + 1) % OFPP_MAX;
1974 return queue_pop_head(q);
1980 /* Add tokens to the bucket based on elapsed time. */
1982 refill_bucket(struct rate_limiter *rl)
1984 const struct settings *s = rl->s;
1985 long long int now = time_msec();
1986 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
1987 if (tokens >= 1000) {
1988 rl->last_fill = now;
1989 rl->tokens = MIN(tokens, s->burst_limit * 1000);
1993 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
1994 * true if successful, false otherwise. (In the latter case no tokens are
1997 get_token(struct rate_limiter *rl)
1999 if (rl->tokens >= 1000) {
2008 rate_limit_local_packet_cb(struct relay *r, void *rl_)
2010 struct rate_limiter *rl = rl_;
2011 const struct settings *s = rl->s;
2012 struct ofp_packet_in *opi;
2014 opi = get_ofp_packet_in(r);
2019 if (!rl->n_queued && get_token(rl)) {
2020 /* In the common case where we are not constrained by the rate limit,
2021 * let the packet take the normal path. */
2025 /* Otherwise queue it up for the periodic callback to drain out. */
2026 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
2027 int port = ntohs(opi->in_port) % OFPP_MAX;
2028 if (rl->n_queued >= s->burst_limit) {
2031 queue_push_tail(&rl->queues[port], ofpbuf_clone(msg));
2039 rate_limit_status_cb(struct status_reply *sr, void *rl_)
2041 struct rate_limiter *rl = rl_;
2043 status_reply_put(sr, "normal=%llu", rl->n_normal);
2044 status_reply_put(sr, "limited=%llu", rl->n_limited);
2045 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
2046 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
2050 rate_limit_periodic_cb(void *rl_)
2052 struct rate_limiter *rl = rl_;
2055 /* Drain some packets out of the bucket if possible, but limit the number
2056 * of iterations to allow other code to get work done too. */
2058 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
2059 /* Use a small, arbitrary limit for the amount of queuing to do here,
2060 * because the TCP connection is responsible for buffering and there is
2061 * no point in trying to transmit faster than the TCP connection can
2063 struct ofpbuf *b = dequeue_packet(rl);
2064 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
2071 rate_limit_wait_cb(void *rl_)
2073 struct rate_limiter *rl = rl_;
2075 if (rl->tokens >= 1000) {
2076 /* We can transmit more packets as soon as we're called again. */
2077 poll_immediate_wake();
2079 /* We have to wait for the bucket to re-fill. We could calculate
2080 * the exact amount of time here for increased smoothness. */
2081 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
2087 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
2088 struct rconn *local, struct rconn *remote)
2090 struct rate_limiter *rl;
2093 rl = xcalloc(1, sizeof *rl);
2095 rl->remote_rconn = remote;
2096 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
2097 queue_init(&rl->queues[i]);
2099 rl->last_fill = time_msec();
2100 rl->tokens = s->rate_limit * 100;
2101 switch_status_register_category(ss, "rate-limit",
2102 rate_limit_status_cb, rl);
2103 return make_hook(rate_limit_local_packet_cb, NULL, rate_limit_periodic_cb,
2104 rate_limit_wait_cb, rl);
2107 /* OFPST_SWITCH statistics. */
2109 struct switch_status_category {
2111 void (*cb)(struct status_reply *, void *aux);
2115 struct switch_status {
2116 const struct settings *s;
2118 struct switch_status_category categories[8];
2122 struct status_reply {
2123 struct switch_status_category *category;
2129 switch_status_remote_packet_cb(struct relay *r, void *ss_)
2131 struct switch_status *ss = ss_;
2132 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
2133 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
2134 struct switch_status_category *c;
2135 struct nicira_header *request;
2136 struct nicira_header *reply;
2137 struct status_reply sr;
2141 if (msg->size < sizeof(struct nicira_header)) {
2144 request = msg->data;
2145 if (request->header.type != OFPT_VENDOR
2146 || request->vendor != htonl(NX_VENDOR_ID)
2147 || request->subtype != htonl(NXT_STATUS_REQUEST)) {
2151 sr.request.string = (void *) (request + 1);
2152 sr.request.length = msg->size - sizeof *request;
2153 ds_init(&sr.output);
2154 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
2155 if (!memcmp(c->name, sr.request.string,
2156 MIN(strlen(c->name), sr.request.length))) {
2161 reply = make_openflow_xid(sizeof *reply + sr.output.length,
2162 OFPT_VENDOR, request->header.xid, &b);
2163 reply->vendor = htonl(NX_VENDOR_ID);
2164 reply->subtype = htonl(NXT_STATUS_REPLY);
2165 memcpy(reply + 1, sr.output.string, sr.output.length);
2166 retval = rconn_send(rc, b, NULL);
2167 if (retval && retval != EAGAIN) {
2168 VLOG_WARN("send failed (%s)", strerror(retval));
2170 ds_destroy(&sr.output);
2175 rconn_status_cb(struct status_reply *sr, void *rconn_)
2177 struct rconn *rconn = rconn_;
2178 time_t now = time_now();
2180 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
2181 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
2182 status_reply_put(sr, "backoff=%d", rconn_get_backoff(rconn));
2183 status_reply_put(sr, "is-connected=%s",
2184 rconn_is_connected(rconn) ? "true" : "false");
2185 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
2186 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
2187 status_reply_put(sr, "attempted-connections=%u",
2188 rconn_get_attempted_connections(rconn));
2189 status_reply_put(sr, "successful-connections=%u",
2190 rconn_get_successful_connections(rconn));
2191 status_reply_put(sr, "last-connection=%ld",
2192 (long int) (now - rconn_get_last_connection(rconn)));
2193 status_reply_put(sr, "time-connected=%lu",
2194 rconn_get_total_time_connected(rconn));
2195 status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(rconn));
2199 config_status_cb(struct status_reply *sr, void *s_)
2201 const struct settings *s = s_;
2204 for (i = 0; i < s->n_listeners; i++) {
2205 status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
2207 if (s->probe_interval) {
2208 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
2210 if (s->max_backoff) {
2211 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
2216 switch_status_cb(struct status_reply *sr, void *ss_)
2218 struct switch_status *ss = ss_;
2219 time_t now = time_now();
2221 status_reply_put(sr, "now=%ld", (long int) now);
2222 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
2223 status_reply_put(sr, "pid=%ld", (long int) getpid());
2227 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
2229 struct switch_status *ss = xcalloc(1, sizeof *ss);
2231 ss->booted = time_now();
2232 switch_status_register_category(ss, "config",
2233 config_status_cb, (void *) s);
2234 switch_status_register_category(ss, "switch", switch_status_cb, ss);
2236 return make_hook(NULL, switch_status_remote_packet_cb, NULL, NULL, ss);
2240 switch_status_register_category(struct switch_status *ss,
2241 const char *category,
2242 void (*cb)(struct status_reply *,
2246 struct switch_status_category *c;
2247 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
2248 c = &ss->categories[ss->n_categories++];
2251 c->name = xstrdup(category);
2255 status_reply_put(struct status_reply *sr, const char *content, ...)
2257 size_t old_length = sr->output.length;
2261 /* Append the status reply to the output. */
2262 ds_put_format(&sr->output, "%s.", sr->category->name);
2263 va_start(args, content);
2264 ds_put_format_valist(&sr->output, content, args);
2266 if (ds_last(&sr->output) != '\n') {
2267 ds_put_char(&sr->output, '\n');
2270 /* Drop what we just added if it doesn't match the request. */
2271 added = sr->output.length - old_length;
2272 if (added < sr->request.length
2273 || memcmp(&sr->output.string[old_length],
2274 sr->request.string, sr->request.length)) {
2275 ds_truncate(&sr->output, old_length);
2280 /* Controller discovery. */
2284 const struct settings *s;
2285 struct dhclient *dhcp;
2290 discovery_status_cb(struct status_reply *sr, void *d_)
2292 struct discovery *d = d_;
2294 status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re);
2295 status_reply_put(sr, "n-changes=%d", d->n_changes);
2297 status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp));
2298 status_reply_put(sr, "state-elapsed=%u",
2299 dhclient_get_state_elapsed(d->dhcp));
2300 if (dhclient_is_bound(d->dhcp)) {
2301 uint32_t ip = dhclient_get_ip(d->dhcp);
2302 uint32_t netmask = dhclient_get_netmask(d->dhcp);
2303 uint32_t router = dhclient_get_router(d->dhcp);
2305 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
2306 uint32_t dns_server;
2310 status_reply_put(sr, "ip="IP_FMT, IP_ARGS(&ip));
2311 status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask));
2313 status_reply_put(sr, "router="IP_FMT, IP_ARGS(&router));
2316 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i,
2319 status_reply_put(sr, "dns%d="IP_FMT, i, IP_ARGS(&dns_server));
2322 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
2324 status_reply_put(sr, "domain=%s", domain_name);
2328 status_reply_put(sr, "lease-remaining=%u",
2329 dhclient_get_lease_remaining(d->dhcp));
2335 discovery_local_port_cb(const struct ofp_phy_port *port, void *d_)
2337 struct discovery *d = d_;
2339 char name[OFP_MAX_PORT_NAME_LEN + 1];
2340 struct netdev *netdev;
2343 /* Check that this was really a change. */
2344 get_port_name(port, name, sizeof name);
2345 if (d->dhcp && !strcmp(netdev_get_name(dhclient_get_netdev(d->dhcp)),
2350 /* Destroy current DHCP client. */
2351 dhclient_destroy(d->dhcp);
2354 /* Bring local network device up. */
2355 retval = netdev_open(name, NETDEV_ETH_TYPE_NONE, &netdev);
2357 VLOG_ERR("Could not open %s device, discovery disabled: %s",
2358 name, strerror(retval));
2361 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
2363 VLOG_ERR("Could not bring %s device up, discovery disabled: %s",
2364 name, strerror(retval));
2367 netdev_close(netdev);
2369 /* Initialize DHCP client. */
2370 retval = dhclient_create(name, modify_dhcp_request,
2371 validate_dhcp_offer, (void *) d->s, &d->dhcp);
2373 VLOG_ERR("Failed to initialize DHCP client, "
2374 "discovery disabled: %s", strerror(retval));
2377 dhclient_init(d->dhcp, 0);
2379 dhclient_destroy(d->dhcp);
2385 static struct discovery *
2386 discovery_init(const struct settings *s, struct port_watcher *pw,
2387 struct switch_status *ss)
2389 struct discovery *d;
2391 d = xmalloc(sizeof *d);
2396 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
2397 port_watcher_register_local_port_callback(pw, discovery_local_port_cb, d);
2403 discovery_question_connectivity(struct discovery *d)
2406 dhclient_force_renew(d->dhcp, 15);
2411 discovery_run(struct discovery *d, char **controller_name)
2414 *controller_name = NULL;
2418 dhclient_run(d->dhcp);
2419 if (!dhclient_changed(d->dhcp)) {
2423 dhclient_configure_netdev(d->dhcp);
2424 if (d->s->update_resolv_conf) {
2425 dhclient_update_resolv_conf(d->dhcp);
2428 if (dhclient_is_bound(d->dhcp)) {
2429 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
2430 DHCP_CODE_OFP_CONTROLLER_VCONN);
2431 VLOG_WARN("%s: discovered controller", *controller_name);
2434 *controller_name = NULL;
2436 VLOG_WARN("discovered controller no longer available");
2444 discovery_wait(struct discovery *d)
2447 dhclient_wait(d->dhcp);
2452 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
2454 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
2458 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
2460 const struct settings *s = s_;
2464 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
2466 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
2469 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
2471 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
2472 s->accept_controller_re);
2478 /* User interface. */
2481 parse_options(int argc, char *argv[], struct settings *s)
2484 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
2486 OPT_INACTIVITY_PROBE,
2491 OPT_BOOTSTRAP_CA_CERT,
2497 static struct option long_options[] = {
2498 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
2499 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
2500 {"fail", required_argument, 0, 'F'},
2501 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
2502 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
2503 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
2504 {"listen", required_argument, 0, 'l'},
2505 {"monitor", required_argument, 0, 'm'},
2506 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
2507 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
2508 {"stp", no_argument, 0, OPT_STP},
2509 {"no-stp", no_argument, 0, OPT_NO_STP},
2510 {"out-of-band", no_argument, 0, OPT_OUT_OF_BAND},
2511 {"in-band", no_argument, 0, OPT_IN_BAND},
2512 {"detach", no_argument, 0, 'D'},
2513 {"force", no_argument, 0, 'f'},
2514 {"pidfile", optional_argument, 0, 'P'},
2515 {"verbose", optional_argument, 0, 'v'},
2516 {"help", no_argument, 0, 'h'},
2517 {"version", no_argument, 0, 'V'},
2519 VCONN_SSL_LONG_OPTIONS
2520 {"bootstrap-ca-cert", required_argument, 0, OPT_BOOTSTRAP_CA_CERT},
2524 char *short_options = long_options_to_short_options(long_options);
2525 char *accept_re = NULL;
2528 /* Set defaults that we can figure out before parsing options. */
2530 s->monitor_name = NULL;
2531 s->fail_mode = FAIL_OPEN;
2533 s->probe_interval = 15;
2534 s->max_backoff = 15;
2535 s->update_resolv_conf = true;
2538 s->enable_stp = false;
2543 c = getopt_long(argc, argv, short_options, long_options, NULL);
2549 case OPT_ACCEPT_VCONN:
2550 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
2553 case OPT_NO_RESOLV_CONF:
2554 s->update_resolv_conf = false;
2558 if (!strcmp(optarg, "open")) {
2559 s->fail_mode = FAIL_OPEN;
2560 } else if (!strcmp(optarg, "closed")) {
2561 s->fail_mode = FAIL_CLOSED;
2563 ofp_fatal(0, "-f or --fail argument must be \"open\" "
2568 case OPT_INACTIVITY_PROBE:
2569 s->probe_interval = atoi(optarg);
2570 if (s->probe_interval < 5) {
2571 ofp_fatal(0, "--inactivity-probe argument must be at least 5");
2576 if (!strcmp(optarg, "permanent")) {
2577 s->max_idle = OFP_FLOW_PERMANENT;
2579 s->max_idle = atoi(optarg);
2580 if (s->max_idle < 1 || s->max_idle > 65535) {
2581 ofp_fatal(0, "--max-idle argument must be between 1 and "
2582 "65535 or the word 'permanent'");
2587 case OPT_MAX_BACKOFF:
2588 s->max_backoff = atoi(optarg);
2589 if (s->max_backoff < 1) {
2590 ofp_fatal(0, "--max-backoff argument must be at least 1");
2591 } else if (s->max_backoff > 3600) {
2592 s->max_backoff = 3600;
2596 case OPT_RATE_LIMIT:
2598 s->rate_limit = atoi(optarg);
2599 if (s->rate_limit < 1) {
2600 ofp_fatal(0, "--rate-limit argument must be at least 1");
2603 s->rate_limit = 1000;
2607 case OPT_BURST_LIMIT:
2608 s->burst_limit = atoi(optarg);
2609 if (s->burst_limit < 1) {
2610 ofp_fatal(0, "--burst-limit argument must be at least 1");
2615 s->enable_stp = true;
2619 s->enable_stp = false;
2622 case OPT_OUT_OF_BAND:
2635 set_pidfile(optarg);
2639 ignore_existing_pidfile();
2643 if (s->n_listeners >= MAX_MGMT) {
2645 "-l or --listen may be specified at most %d times",
2648 s->listener_names[s->n_listeners++] = optarg;
2652 if (s->monitor_name) {
2653 ofp_fatal(0, "-m or --monitor may only be specified once");
2655 s->monitor_name = optarg;
2662 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
2666 vlog_set_verbosity(optarg);
2670 VCONN_SSL_OPTION_HANDLERS
2672 case OPT_BOOTSTRAP_CA_CERT:
2673 vconn_ssl_set_ca_cert_file(optarg, true);
2684 free(short_options);
2688 if (argc < 1 || argc > 2) {
2689 ofp_fatal(0, "need one or two non-option arguments; "
2690 "use --help for usage");
2693 /* Local and remote vconns. */
2694 s->dp_name = argv[0];
2695 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
2697 /* Set accept_controller_regex. */
2699 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
2701 retval = regcomp(&s->accept_controller_regex, accept_re,
2702 REG_NOSUB | REG_EXTENDED);
2704 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
2705 char *buffer = xmalloc(length);
2706 regerror(retval, &s->accept_controller_regex, buffer, length);
2707 ofp_fatal(0, "%s: %s", accept_re, buffer);
2709 s->accept_controller_re = accept_re;
2711 /* Mode of operation. */
2712 s->discovery = s->controller_name == NULL;
2713 if (s->discovery && !s->in_band) {
2714 ofp_fatal(0, "Cannot perform discovery with out-of-band control");
2717 /* Rate limiting. */
2718 if (s->rate_limit) {
2719 if (s->rate_limit < 100) {
2720 VLOG_WARN("Rate limit set to unusually low value %d",
2723 if (!s->burst_limit) {
2724 s->burst_limit = s->rate_limit / 4;
2726 s->burst_limit = MAX(s->burst_limit, 1);
2727 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
2734 printf("%s: secure channel, a relay for OpenFlow messages.\n"
2735 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
2736 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
2737 "CONTROLLER is an active OpenFlow connection method; if it is\n"
2738 "omitted, then secchan performs controller discovery.\n",
2739 program_name, program_name);
2740 vconn_usage(true, true, true);
2741 printf("\nController discovery options:\n"
2742 " --accept-vconn=REGEX accept matching discovered controllers\n"
2743 " --no-resolv-conf do not update /etc/resolv.conf\n"
2744 "\nNetworking options:\n"
2745 " -F, --fail=open|closed when controller connection fails:\n"
2746 " closed: drop all packets\n"
2747 " open (default): act as learning switch\n"
2748 " --inactivity-probe=SECS time between inactivity probes\n"
2749 " --max-idle=SECS max idle for flows set up by secchan\n"
2750 " --max-backoff=SECS max time between controller connection\n"
2751 " attempts (default: 15 seconds)\n"
2752 " -l, --listen=METHOD allow management connections on METHOD\n"
2753 " (a passive OpenFlow connection method)\n"
2754 " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n"
2755 " (a passive OpenFlow connection method)\n"
2756 " --out-of-band controller connection is out-of-band\n"
2757 " --stp enable 802.1D Spanning Tree Protocol\n"
2758 " --no-stp disable 802.1D Spanning Tree Protocol\n"
2759 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
2760 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
2761 " --burst-limit=BURST limit on packet credit for idle time\n"
2762 "\nOther options:\n"
2763 " -D, --detach run in background as daemon\n"
2764 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
2765 " -f, --force with -P, start even if already running\n"
2766 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
2767 " -v, --verbose set maximum verbosity level\n"
2768 " -h, --help display this help message\n"
2769 " -V, --version display version information\n",