1 /* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
4 * We are making the OpenFlow specification and associated documentation
5 * (Software) available for public use and benefit with the expectation
6 * that others will use, modify and enhance the Software and contribute
7 * those enhancements back to the community. However, since we would
8 * like to make the Software available for broadest use, with as few
9 * restrictions as possible permission is hereby granted, free of
10 * charge, to any person obtaining a copy of this Software to deal in
11 * the Software under the copyrights without restriction, including
12 * without limitation the rights to use, copy, modify, merge, publish,
13 * distribute, sublicense, and/or sell copies of the Software, and to
14 * permit persons to whom the Software is furnished to do so, subject to
15 * the following conditions:
17 * The above copyright notice and this permission notice shall be
18 * included in all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
23 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
24 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
25 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
26 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 * The name and trademarks of copyright holder(s) may NOT be used in
30 * advertising or publicity pertaining to the Software or any
31 * derivatives without specific, written prior permission.
39 #include <netinet/in.h>
49 #include "command-line.h"
53 #include "dhcp-client.h"
54 #include "dynamic-string.h"
57 #include "learning-switch.h"
59 #include "mac-learning.h"
63 #include "poll-loop.h"
68 #include "vconn-ssl.h"
70 #include "vlog-socket.h"
73 #define THIS_MODULE VLM_secchan
75 /* Behavior when the connection to the controller fails. */
77 FAIL_OPEN, /* Act as learning switch. */
78 FAIL_CLOSED /* Drop all packets. */
81 /* Maximum number of management connection listeners. */
84 /* Settings that may be configured by the user. */
86 /* Overall mode of operation. */
87 bool discovery; /* Discover the controller automatically? */
88 bool in_band; /* Connect to controller in-band? */
90 /* Related vconns and network devices. */
91 const char *nl_name; /* Local datapath (must be "nl:" vconn). */
92 char *of_name; /* ofX network device name. */
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? */
116 struct buffer *rxbuf;
117 int n_txq; /* No. of packets queued for tx on 'rconn'. */
124 #define HALF_REMOTE 1
125 struct half halves[2];
131 bool (*packet_cb[2])(struct relay *, void *aux);
132 void (*periodic_cb)(void *aux);
133 void (*wait_cb)(void *aux);
137 static struct vlog_rate_limit vrl = VLOG_RATE_LIMIT_INIT(60, 60);
139 static void parse_options(int argc, char *argv[], struct settings *);
140 static void usage(void) NO_RETURN;
142 static struct vconn *open_passive_vconn(const char *name);
143 static struct vconn *accept_vconn(struct vconn *vconn);
145 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
147 static struct relay *relay_accept(const struct settings *, struct vconn *);
148 static void relay_run(struct relay *, const struct hook[], size_t n_hooks);
149 static void relay_wait(struct relay *);
150 static void relay_destroy(struct relay *);
152 static struct hook make_hook(bool (*local_packet_cb)(struct relay *, void *),
153 bool (*remote_packet_cb)(struct relay *, void *),
154 void (*periodic_cb)(void *),
155 void (*wait_cb)(void *),
157 static struct ofp_packet_in *get_ofp_packet_in(struct relay *);
158 static bool get_ofp_packet_eth_header(struct relay *, struct ofp_packet_in **,
159 struct eth_header **);
160 static void get_ofp_packet_payload(struct ofp_packet_in *, struct buffer *);
162 struct switch_status;
164 static struct hook switch_status_hook_create(const struct settings *,
165 struct switch_status **);
166 static void switch_status_register_category(struct switch_status *,
167 const char *category,
168 void (*cb)(struct status_reply *,
171 static void status_reply_put(struct status_reply *, const char *, ...)
174 static void rconn_status_cb(struct status_reply *, void *rconn_);
176 static struct discovery *discovery_init(const struct settings *,
177 struct switch_status *);
178 static void discovery_question_connectivity(struct discovery *);
179 static bool discovery_run(struct discovery *, char **controller_name);
180 static void discovery_wait(struct discovery *);
182 static struct hook in_band_hook_create(const struct settings *,
183 struct switch_status *,
184 struct rconn *remote);
187 static struct hook port_watcher_create(struct rconn *local,
188 struct rconn *remote,
189 struct port_watcher **);
190 static uint32_t port_watcher_get_flags(const struct port_watcher *,
192 static void port_watcher_set_flags(struct port_watcher *,
193 int port_no, uint32_t flags, uint32_t mask);
195 static struct hook stp_hook_create(const struct settings *,
196 struct port_watcher *,
197 struct rconn *local, struct rconn *remote);
199 static struct hook fail_open_hook_create(const struct settings *,
200 struct switch_status *,
202 struct rconn *remote);
203 static struct hook rate_limit_hook_create(const struct settings *,
204 struct switch_status *,
206 struct rconn *remote);
209 static void modify_dhcp_request(struct dhcp_msg *, void *aux);
210 static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux);
213 main(int argc, char *argv[])
217 struct list relays = LIST_INITIALIZER(&relays);
219 struct hook hooks[8];
222 struct vconn *monitor;
224 struct vconn *listeners[MAX_MGMT];
227 struct rconn *local_rconn, *remote_rconn;
228 struct relay *controller_relay;
229 struct discovery *discovery;
230 struct switch_status *switch_status;
231 struct port_watcher *pw;
235 set_program_name(argv[0]);
236 register_fault_handlers();
239 parse_options(argc, argv, &s);
240 signal(SIGPIPE, SIG_IGN);
242 /* Start listening for management and monitoring connections. */
244 for (i = 0; i < s.n_listeners; i++) {
245 listeners[n_listeners++] = open_passive_vconn(s.listener_names[i]);
247 monitor = s.monitor_name ? open_passive_vconn(s.monitor_name) : NULL;
249 /* Initialize switch status hook. */
250 hooks[n_hooks++] = switch_status_hook_create(&s, &switch_status);
252 /* Start controller discovery. */
253 discovery = s.discovery ? discovery_init(&s, switch_status) : NULL;
255 /* Start listening for vlogconf requests. */
256 retval = vlog_server_listen(NULL, NULL);
258 fatal(retval, "Could not listen for vlog connections");
261 die_if_already_running();
264 VLOG_WARN("OpenFlow reference implementation version %s", VERSION);
265 VLOG_WARN("OpenFlow protocol version 0x%02x", OFP_VERSION);
267 /* Connect to datapath. */
268 local_rconn = rconn_create(0, s.max_backoff);
269 rconn_connect(local_rconn, s.nl_name);
270 switch_status_register_category(switch_status, "local",
271 rconn_status_cb, local_rconn);
273 /* Connect to controller. */
274 remote_rconn = rconn_create(s.probe_interval, s.max_backoff);
275 if (s.controller_name) {
276 retval = rconn_connect(remote_rconn, s.controller_name);
277 if (retval == EAFNOSUPPORT) {
278 fatal(0, "No support for %s vconn", s.controller_name);
281 switch_status_register_category(switch_status, "remote",
282 rconn_status_cb, remote_rconn);
284 /* Start relaying. */
285 controller_relay = relay_create(local_rconn, remote_rconn, false);
286 list_push_back(&relays, &controller_relay->node);
289 hooks[n_hooks++] = port_watcher_create(local_rconn, remote_rconn, &pw);
290 hooks[n_hooks++] = stp_hook_create(&s, pw, local_rconn, remote_rconn);
292 hooks[n_hooks++] = in_band_hook_create(&s, switch_status,
295 if (s.fail_mode == FAIL_OPEN) {
296 hooks[n_hooks++] = fail_open_hook_create(&s, switch_status,
297 local_rconn, remote_rconn);
300 hooks[n_hooks++] = rate_limit_hook_create(&s, switch_status,
301 local_rconn, remote_rconn);
303 assert(n_hooks <= ARRAY_SIZE(hooks));
310 LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) {
311 relay_run(r, hooks, n_hooks);
313 for (i = 0; i < n_listeners; i++) {
315 struct relay *r = relay_accept(&s, listeners[i]);
319 list_push_back(&relays, &r->node);
323 struct vconn *new = accept_vconn(monitor);
325 rconn_add_monitor(local_rconn, new);
328 for (i = 0; i < n_hooks; i++) {
329 if (hooks[i].periodic_cb) {
330 hooks[i].periodic_cb(hooks[i].aux);
334 char *controller_name;
335 if (rconn_is_connectivity_questionable(remote_rconn)) {
336 discovery_question_connectivity(discovery);
338 if (discovery_run(discovery, &controller_name)) {
339 if (controller_name) {
340 rconn_connect(remote_rconn, controller_name);
342 rconn_disconnect(remote_rconn);
347 /* Wait for something to happen. */
348 LIST_FOR_EACH (r, struct relay, node, &relays) {
351 for (i = 0; i < n_listeners; i++) {
352 vconn_accept_wait(listeners[i]);
355 vconn_accept_wait(monitor);
357 for (i = 0; i < n_hooks; i++) {
358 if (hooks[i].wait_cb) {
359 hooks[i].wait_cb(hooks[i].aux);
363 discovery_wait(discovery);
371 static struct vconn *
372 open_passive_vconn(const char *name)
377 retval = vconn_open(name, &vconn);
378 if (retval && retval != EAGAIN) {
379 fatal(retval, "opening %s", name);
381 if (!vconn_is_passive(vconn)) {
382 fatal(0, "%s is not a passive vconn", name);
387 static struct vconn *
388 accept_vconn(struct vconn *vconn)
393 retval = vconn_accept(vconn, &new);
394 if (retval && retval != EAGAIN) {
395 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
401 make_hook(bool (*local_packet_cb)(struct relay *, void *aux),
402 bool (*remote_packet_cb)(struct relay *, void *aux),
403 void (*periodic_cb)(void *aux),
404 void (*wait_cb)(void *aux),
408 h.packet_cb[HALF_LOCAL] = local_packet_cb;
409 h.packet_cb[HALF_REMOTE] = remote_packet_cb;
410 h.periodic_cb = periodic_cb;
416 static struct ofp_packet_in *
417 get_ofp_packet_in(struct relay *r)
419 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
420 struct ofp_header *oh = msg->data;
421 if (oh->type == OFPT_PACKET_IN) {
422 if (msg->size >= offsetof (struct ofp_packet_in, data)) {
425 VLOG_WARN("packet too short (%zu bytes) for packet_in",
433 get_ofp_packet_eth_header(struct relay *r, struct ofp_packet_in **opip,
434 struct eth_header **ethp)
436 const int min_len = offsetof(struct ofp_packet_in, data) + ETH_HEADER_LEN;
437 struct ofp_packet_in *opi = get_ofp_packet_in(r);
438 if (opi && ntohs(opi->header.length) >= min_len) {
440 *ethp = (void *) opi->data;
447 /* OpenFlow message relaying. */
449 static struct relay *
450 relay_accept(const struct settings *s, struct vconn *listen_vconn)
452 struct vconn *new_remote, *new_local;
453 char *nl_name_without_subscription;
454 struct rconn *r1, *r2;
457 new_remote = accept_vconn(listen_vconn);
462 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We
463 * only accept the former syntax in main().
465 * nl:123:0 opens a netlink connection to local datapath 123 without
466 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
468 nl_name_without_subscription = xasprintf("%s:0", s->nl_name);
469 retval = vconn_open(nl_name_without_subscription, &new_local);
471 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
472 nl_name_without_subscription, strerror(retval));
473 vconn_close(new_remote);
474 free(nl_name_without_subscription);
478 /* Create and return relay. */
479 r1 = rconn_create(0, 0);
480 rconn_connect_unreliably(r1, nl_name_without_subscription, new_local);
481 free(nl_name_without_subscription);
483 r2 = rconn_create(0, 0);
484 rconn_connect_unreliably(r2, "passive", new_remote);
486 return relay_create(r1, r2, true);
489 static struct relay *
490 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
492 struct relay *r = xcalloc(1, sizeof *r);
493 r->halves[HALF_LOCAL].rconn = local;
494 r->halves[HALF_REMOTE].rconn = remote;
495 r->is_mgmt_conn = is_mgmt_conn;
500 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
505 for (i = 0; i < 2; i++) {
506 rconn_run(r->halves[i].rconn);
509 /* Limit the number of iterations to prevent other tasks from starving. */
510 for (iteration = 0; iteration < 50; iteration++) {
511 bool progress = false;
512 for (i = 0; i < 2; i++) {
513 struct half *this = &r->halves[i];
514 struct half *peer = &r->halves[!i];
517 this->rxbuf = rconn_recv(this->rconn);
518 if (this->rxbuf && (i == HALF_REMOTE || !r->is_mgmt_conn)) {
519 const struct hook *h;
520 for (h = hooks; h < &hooks[n_hooks]; h++) {
521 if (h->packet_cb[i] && h->packet_cb[i](r, h->aux)) {
522 buffer_delete(this->rxbuf);
531 if (this->rxbuf && !this->n_txq) {
532 int retval = rconn_send(peer->rconn, this->rxbuf,
534 if (retval != EAGAIN) {
538 buffer_delete(this->rxbuf);
549 if (r->is_mgmt_conn) {
550 for (i = 0; i < 2; i++) {
551 struct half *this = &r->halves[i];
552 if (!rconn_is_alive(this->rconn)) {
561 relay_wait(struct relay *r)
565 for (i = 0; i < 2; i++) {
566 struct half *this = &r->halves[i];
568 rconn_run_wait(this->rconn);
570 rconn_recv_wait(this->rconn);
576 relay_destroy(struct relay *r)
580 list_remove(&r->node);
581 for (i = 0; i < 2; i++) {
582 struct half *this = &r->halves[i];
583 rconn_destroy(this->rconn);
584 buffer_delete(this->rxbuf);
589 /* Port status watcher. */
591 typedef void port_watcher_cb_func(uint16_t port_no,
592 const struct ofp_phy_port *old,
593 const struct ofp_phy_port *new,
596 struct port_watcher_cb {
597 port_watcher_cb_func *function;
601 struct port_watcher {
602 struct rconn *local_rconn;
603 struct rconn *remote_rconn;
604 struct ofp_phy_port ports[OFPP_MAX + 1];
605 time_t last_feature_request;
606 bool got_feature_reply;
608 struct port_watcher_cb cbs[2];
612 /* Returns the number of fields that differ from 'a' to 'b'. */
614 opp_differs(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
616 BUILD_ASSERT_DECL(sizeof *a == 36); /* Trips when we add or remove fields. */
617 return ((a->port_no != b->port_no)
618 + (memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) != 0)
619 + (memcmp(a->name, b->name, sizeof a->name) != 0)
620 + (a->flags != b->flags)
621 + (a->speed != b->speed)
622 + (a->features != b->features));
626 sanitize_opp(struct ofp_phy_port *opp)
630 for (i = 0; i < sizeof opp->name; i++) {
631 char c = opp->name[i];
632 if (c && (c < 0x20 || c > 0x7e)) {
636 opp->name[sizeof opp->name - 1] = '\0';
640 port_no_to_pw_idx(int port_no)
642 return (port_no < OFPP_MAX ? port_no
643 : port_no == OFPP_LOCAL ? OFPP_MAX
648 call_pw_callbacks(struct port_watcher *pw, int port_no,
649 const struct ofp_phy_port *old,
650 const struct ofp_phy_port *new)
652 if (opp_differs(old, new)) {
654 for (i = 0; i < pw->n_cbs; i++) {
655 pw->cbs[i].function(port_no, old, new, pw->cbs[i].aux);
661 update_phy_port(struct port_watcher *pw, struct ofp_phy_port *opp,
662 uint8_t reason, bool seen[OFPP_MAX + 1])
664 struct ofp_phy_port *pw_opp;
665 struct ofp_phy_port old;
669 port_no = ntohs(opp->port_no);
670 idx = port_no_to_pw_idx(port_no);
679 pw_opp = &pw->ports[idx];
681 if (reason == OFPPR_DELETE) {
682 memset(pw_opp, 0, sizeof *pw_opp);
683 pw_opp->port_no = htons(OFPP_NONE);
684 } else if (reason == OFPPR_MOD || reason == OFPPR_ADD) {
686 sanitize_opp(pw_opp);
688 call_pw_callbacks(pw, port_no, &old, pw_opp);
692 port_watcher_local_packet_cb(struct relay *r, void *pw_)
694 struct port_watcher *pw = pw_;
695 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
696 struct ofp_header *oh = msg->data;
698 if (oh->type == OFPT_FEATURES_REPLY
699 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
700 struct ofp_switch_features *osf = msg->data;
701 bool seen[ARRAY_SIZE(pw->ports)];
705 pw->got_feature_reply = true;
707 /* Update each port included in the message. */
708 memset(seen, 0, sizeof seen);
709 n_ports = ((msg->size - offsetof(struct ofp_switch_features, ports))
710 / sizeof *osf->ports);
711 for (i = 0; i < n_ports; i++) {
712 update_phy_port(pw, &osf->ports[i], OFPPR_MOD, seen);
715 /* Delete all the ports not included in the message. */
716 for (i = 0; i < ARRAY_SIZE(pw->ports); i++) {
718 update_phy_port(pw, &pw->ports[i], OFPPR_DELETE, NULL);
721 } else if (oh->type == OFPT_PORT_STATUS
722 && msg->size >= sizeof(struct ofp_port_status)) {
723 struct ofp_port_status *ops = msg->data;
724 update_phy_port(pw, &ops->desc, ops->reason, NULL);
730 port_watcher_periodic_cb(void *pw_)
732 struct port_watcher *pw = pw_;
734 if (!pw->got_feature_reply && time_now() >= pw->last_feature_request + 5) {
736 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
737 rconn_send_with_limit(pw->local_rconn, b, &pw->n_txq, 1);
738 pw->last_feature_request = time_now();
743 put_duplexes(struct ds *ds, const char *name, uint32_t features,
744 uint32_t hd_bit, uint32_t fd_bit)
746 if (features & (hd_bit | fd_bit)) {
747 ds_put_format(ds, " %s", name);
748 if (features & hd_bit) {
749 ds_put_cstr(ds, "(HD)");
751 if (features & fd_bit) {
752 ds_put_cstr(ds, "(FD)");
758 log_port_status(uint16_t port_no,
759 const struct ofp_phy_port *old,
760 const struct ofp_phy_port *new,
763 if (VLOG_IS_DBG_ENABLED()) {
764 bool was_enabled = old->port_no != htons(OFPP_NONE);
765 bool now_enabled = new->port_no != htons(OFPP_NONE);
766 uint32_t features = ntohl(new->features);
769 if (old->flags != new->flags && opp_differs(old, new) == 1) {
770 /* Don't care if only flags changed. */
775 ds_put_format(&ds, "\"%s\", "ETH_ADDR_FMT, new->name,
776 ETH_ADDR_ARGS(new->hw_addr));
777 if (ntohl(new->speed)) {
778 ds_put_format(&ds, ", speed %"PRIu32, ntohl(new->speed));
780 if (features & (OFPPF_10MB_HD | OFPPF_10MB_FD
781 | OFPPF_100MB_HD | OFPPF_100MB_FD
782 | OFPPF_1GB_HD | OFPPF_1GB_FD | OFPPF_10GB_FD)) {
783 ds_put_cstr(&ds, ", supports");
784 put_duplexes(&ds, "10M", features, OFPPF_10MB_HD, OFPPF_10MB_FD);
785 put_duplexes(&ds, "100M", features,
786 OFPPF_100MB_HD, OFPPF_100MB_FD);
787 put_duplexes(&ds, "1G", features, OFPPF_100MB_HD, OFPPF_100MB_FD);
788 if (features & OFPPF_10GB_FD) {
789 ds_put_cstr(&ds, " 10G");
792 if (was_enabled != now_enabled) {
794 VLOG_DBG("Port %d added: %s", port_no, ds_cstr(&ds));
796 VLOG_DBG("Port %d deleted", port_no);
799 VLOG_DBG("Port %d changed: %s", port_no, ds_cstr(&ds));
806 port_watcher_register_callback(struct port_watcher *pw,
807 port_watcher_cb_func *function,
810 assert(pw->n_cbs < ARRAY_SIZE(pw->cbs));
811 pw->cbs[pw->n_cbs].function = function;
812 pw->cbs[pw->n_cbs].aux = aux;
817 port_watcher_get_flags(const struct port_watcher *pw, int port_no)
819 int idx = port_no_to_pw_idx(port_no);
820 return idx >= 0 ? ntohl(pw->ports[idx].flags) : 0;
824 port_watcher_set_flags(struct port_watcher *pw,
825 int port_no, uint32_t flags, uint32_t mask)
827 struct ofp_phy_port old;
828 struct ofp_phy_port *p;
829 struct ofp_port_mod *opm;
830 struct ofp_port_status *ops;
834 idx = port_no_to_pw_idx(port_no);
840 if (!((ntohl(p->flags) ^ flags) & mask)) {
845 /* Update our idea of the flags. */
846 p->flags = ntohl(flags);
847 call_pw_callbacks(pw, port_no, &old, p);
849 /* Change the flags in the datapath. */
850 opm = make_openflow(sizeof *opm, OFPT_PORT_MOD, &b);
851 opm->mask = htonl(mask);
853 rconn_send(pw->local_rconn, b, NULL);
855 /* Notify the controller that the flags changed. */
856 ops = make_openflow(sizeof *ops, OFPT_PORT_STATUS, &b);
857 ops->reason = OFPPR_MOD;
859 rconn_send(pw->remote_rconn, b, NULL);
863 port_watcher_is_ready(const struct port_watcher *pw)
865 return pw->got_feature_reply;
869 port_watcher_create(struct rconn *local_rconn, struct rconn *remote_rconn,
870 struct port_watcher **pwp)
872 struct port_watcher *pw;
875 pw = *pwp = xcalloc(1, sizeof *pw);
876 pw->local_rconn = local_rconn;
877 pw->remote_rconn = remote_rconn;
878 pw->last_feature_request = TIME_MIN;
879 for (i = 0; i < OFPP_MAX; i++) {
880 pw->ports[i].port_no = htons(OFPP_NONE);
882 port_watcher_register_callback(pw, log_port_status, NULL);
883 return make_hook(port_watcher_local_packet_cb, NULL,
884 port_watcher_periodic_cb, NULL, pw);
887 /* Spanning tree protocol. */
889 /* Extra time, in seconds, at boot before going into fail-open, to give the
890 * spanning tree protocol time to figure out the network layout. */
891 #define STP_EXTRA_BOOT_TIME 30
895 struct port_watcher *pw;
896 struct rconn *local_rconn;
897 struct rconn *remote_rconn;
898 uint8_t dpid[ETH_ADDR_LEN];
899 long long int last_tick_256ths;
904 stp_local_packet_cb(struct relay *r, void *stp_)
906 struct stp_data *stp = stp_;
907 struct ofp_packet_in *opi;
908 struct eth_header *eth;
909 struct llc_header *llc;
910 struct buffer payload;
914 if (!get_ofp_packet_eth_header(r, &opi, ð)
915 || !eth_addr_equals(eth->eth_dst, stp_eth_addr)) {
919 port_no = ntohs(opi->in_port);
920 if (port_no >= STP_MAX_PORTS) {
921 /* STP only supports 255 ports. */
924 if (port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP) {
925 /* We're not doing STP on this port. */
929 if (opi->reason == OFPR_ACTION) {
930 /* The controller set up a flow for this, so we won't intercept it. */
934 get_ofp_packet_payload(opi, &payload);
935 flow_extract(&payload, port_no, &flow);
936 if (flow.dl_type != htons(OFP_DL_TYPE_NOT_ETH_TYPE)) {
937 VLOG_DBG("non-LLC frame received on STP multicast address");
940 llc = buffer_at_assert(&payload, sizeof *eth, sizeof *llc);
941 if (llc->llc_dsap != STP_LLC_DSAP) {
942 VLOG_DBG("bad DSAP 0x%02"PRIx8" received on STP multicast address",
947 /* Trim off padding on payload. */
948 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
949 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
951 if (buffer_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
952 struct stp_port *p = stp_get_port(stp->stp, port_no);
953 stp_received_bpdu(p, payload.data, payload.size);
962 return time_msec() * 256 / 1000;
966 stp_periodic_cb(void *stp_)
968 struct stp_data *stp = stp_;
969 long long int now_256ths = time_256ths();
970 long long int elapsed_256ths = now_256ths - stp->last_tick_256ths;
973 if (!port_watcher_is_ready(stp->pw)) {
974 /* Can't start STP until we know port flags, because port flags can
978 if (elapsed_256ths <= 0) {
982 stp_tick(stp->stp, MIN(INT_MAX, elapsed_256ths));
983 stp->last_tick_256ths = now_256ths;
985 while (stp_get_changed_port(stp->stp, &p)) {
986 int port_no = stp_port_no(p);
987 enum stp_state state = stp_port_get_state(p);
989 if (state != STP_DISABLED) {
990 VLOG_WARN("STP: Port %d entered %s state",
991 port_no, stp_state_name(state));
993 if (!(port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP)) {
997 flags = OFPPFL_STP_LISTEN;
1000 flags = OFPPFL_STP_LEARN;
1003 case STP_FORWARDING:
1004 flags = OFPPFL_STP_FORWARD;
1007 flags = OFPPFL_STP_BLOCK;
1010 VLOG_DBG_RL(&vrl, "STP: Port %d has bad state %x",
1012 flags = OFPPFL_STP_FORWARD;
1015 if (!stp_forward_in_state(state)) {
1016 flags |= OFPPFL_NO_FLOOD;
1018 port_watcher_set_flags(stp->pw, port_no, flags,
1019 OFPPFL_STP_MASK | OFPPFL_NO_FLOOD);
1021 /* We don't own those flags. */
1027 stp_wait_cb(void *stp_ UNUSED)
1029 poll_timer_wait(1000);
1033 send_bpdu(const void *bpdu, size_t bpdu_size, int port_no, void *stp_)
1035 struct stp_data *stp = stp_;
1036 struct eth_header *eth;
1037 struct llc_header *llc;
1038 struct buffer pkt, *opo;
1040 /* Packet skeleton. */
1041 buffer_init(&pkt, ETH_HEADER_LEN + LLC_HEADER_LEN + bpdu_size);
1042 eth = buffer_put_uninit(&pkt, sizeof *eth);
1043 llc = buffer_put_uninit(&pkt, sizeof *llc);
1044 buffer_put(&pkt, bpdu, bpdu_size);
1047 memcpy(eth->eth_dst, stp_eth_addr, ETH_ADDR_LEN);
1048 memcpy(eth->eth_src, stp->pw->ports[port_no].hw_addr, ETH_ADDR_LEN);
1049 eth->eth_type = htons(pkt.size - ETH_HEADER_LEN);
1052 llc->llc_dsap = STP_LLC_DSAP;
1053 llc->llc_ssap = STP_LLC_SSAP;
1054 llc->llc_cntl = STP_LLC_CNTL;
1056 opo = make_unbuffered_packet_out(&pkt, OFPP_NONE, port_no);
1057 buffer_uninit(&pkt);
1058 rconn_send_with_limit(stp->local_rconn, opo, &stp->n_txq, OFPP_MAX);
1062 stp_port_watcher_cb(uint16_t port_no,
1063 const struct ofp_phy_port *old,
1064 const struct ofp_phy_port *new,
1067 struct stp_data *stp = stp_;
1070 /* STP only supports a maximum of 255 ports, one less than OpenFlow. We
1071 * don't support STP on OFPP_LOCAL, either. */
1072 if (port_no >= STP_MAX_PORTS) {
1076 p = stp_get_port(stp->stp, port_no);
1077 if (new->port_no == htons(OFPP_NONE)
1078 || new->flags & htonl(OFPPFL_NO_STP)) {
1079 stp_port_disable(p);
1082 stp_port_set_speed(p, new->speed);
1087 stp_hook_create(const struct settings *s, struct port_watcher *pw,
1088 struct rconn *local, struct rconn *remote)
1090 uint8_t dpid[ETH_ADDR_LEN];
1091 struct netdev *netdev;
1092 struct stp_data *stp;
1095 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1097 fatal(retval, "Could not open %s device", s->of_name);
1099 memcpy(dpid, netdev_get_etheraddr(netdev), ETH_ADDR_LEN);
1100 netdev_close(netdev);
1102 stp = xcalloc(1, sizeof *stp);
1103 stp->stp = stp_create("stp", eth_addr_to_uint64(dpid), send_bpdu, stp);
1105 memcpy(stp->dpid, dpid, ETH_ADDR_LEN);
1106 stp->local_rconn = local;
1107 stp->remote_rconn = remote;
1108 stp->last_tick_256ths = time_256ths();
1110 port_watcher_register_callback(pw, stp_port_watcher_cb, stp);
1111 return make_hook(stp_local_packet_cb, NULL,
1112 stp_periodic_cb, stp_wait_cb, stp);
1115 /* In-band control. */
1117 struct in_band_data {
1118 const struct settings *s;
1119 struct mac_learning *ml;
1120 struct netdev *of_device;
1121 struct rconn *controller;
1122 uint8_t mac[ETH_ADDR_LEN];
1127 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct buffer *b)
1129 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
1132 static const uint8_t *
1133 get_controller_mac(struct in_band_data *in_band)
1135 static uint32_t ip, last_nonzero_ip;
1136 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
1137 static time_t next_refresh = 0;
1139 uint32_t last_ip = ip;
1141 time_t now = time_now();
1143 ip = rconn_get_ip(in_band->controller);
1144 if (last_ip != ip || !next_refresh || now >= next_refresh) {
1147 /* Look up MAC address. */
1148 memset(mac, 0, sizeof mac);
1150 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
1152 VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s",
1153 IP_ARGS(&ip), strerror(retval));
1156 have_mac = !eth_addr_is_zero(mac);
1158 /* Log changes in IP, MAC addresses. */
1159 if (ip && ip != last_nonzero_ip) {
1160 VLOG_DBG("controller IP address changed from "IP_FMT
1161 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
1162 last_nonzero_ip = ip;
1164 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
1165 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
1167 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
1168 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
1171 /* Schedule next refresh.
1173 * If we have an IP address but not a MAC address, then refresh
1174 * quickly, since we probably will get a MAC address soon (via ARP).
1175 * Otherwise, we can afford to wait a little while. */
1176 next_refresh = now + (!ip || have_mac ? 10 : 1);
1178 return !eth_addr_is_zero(mac) ? mac : NULL;
1182 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
1183 struct in_band_data *in_band)
1185 const uint8_t *mac = get_controller_mac(in_band);
1186 return mac && eth_addr_equals(mac, dl_addr);
1190 in_band_learn_mac(struct in_band_data *in_band,
1191 uint16_t in_port, const uint8_t src_mac[ETH_ADDR_LEN])
1193 if (mac_learning_learn(in_band->ml, src_mac, in_port)) {
1194 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
1195 ETH_ADDR_ARGS(src_mac), in_port);
1200 in_band_local_packet_cb(struct relay *r, void *in_band_)
1202 struct in_band_data *in_band = in_band_;
1203 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
1204 struct ofp_packet_in *opi;
1205 struct eth_header *eth;
1206 struct buffer payload;
1211 if (!get_ofp_packet_eth_header(r, &opi, ð)) {
1214 in_port = ntohs(opi->in_port);
1216 /* Deal with local stuff. */
1217 if (in_port == OFPP_LOCAL) {
1218 /* Sent by secure channel. */
1219 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1220 } else if (eth_addr_equals(eth->eth_dst, in_band->mac)) {
1221 /* Sent to secure channel. */
1222 out_port = OFPP_LOCAL;
1223 in_band_learn_mac(in_band, in_port, eth->eth_src);
1224 } else if (eth->eth_type == htons(ETH_TYPE_ARP)
1225 && eth_addr_is_broadcast(eth->eth_dst)
1226 && is_controller_mac(eth->eth_src, in_band)) {
1227 /* ARP sent by controller. */
1228 out_port = OFPP_FLOOD;
1229 } else if (is_controller_mac(eth->eth_dst, in_band)
1230 || is_controller_mac(eth->eth_src, in_band)) {
1231 /* Traffic to or from controller. Switch it by hand. */
1232 in_band_learn_mac(in_band, in_port, eth->eth_src);
1233 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1235 const uint8_t *controller_mac;
1236 controller_mac = get_controller_mac(in_band);
1237 if (eth->eth_type == htons(ETH_TYPE_ARP)
1238 && eth_addr_is_broadcast(eth->eth_dst)
1239 && is_controller_mac(eth->eth_src, in_band)) {
1240 /* ARP sent by controller. */
1241 out_port = OFPP_FLOOD;
1242 } else if (is_controller_mac(eth->eth_dst, in_band)
1243 && in_port == mac_learning_lookup(in_band->ml,
1245 /* Drop controller traffic that arrives on the controller port. */
1252 get_ofp_packet_payload(opi, &payload);
1253 flow_extract(&payload, in_port, &flow);
1254 if (in_port == out_port) {
1255 /* The input and output port match. Set up a flow to drop packets. */
1256 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
1257 in_band->s->max_idle, 0));
1258 } else if (out_port != OFPP_FLOOD) {
1259 /* The output port is known, so add a new flow. */
1260 queue_tx(rc, in_band,
1261 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
1262 out_port, in_band->s->max_idle));
1264 /* If the switch didn't buffer the packet, we need to send a copy. */
1265 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1266 queue_tx(rc, in_band,
1267 make_unbuffered_packet_out(&payload, in_port, out_port));
1270 /* We don't know that MAC. Send along the packet without setting up a
1273 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1274 b = make_unbuffered_packet_out(&payload, in_port, out_port);
1276 b = make_buffered_packet_out(ntohl(opi->buffer_id),
1279 queue_tx(rc, in_band, b);
1285 in_band_status_cb(struct status_reply *sr, void *in_band_)
1287 struct in_band_data *in_band = in_band_;
1288 struct in_addr local_ip;
1289 uint32_t controller_ip;
1290 const uint8_t *controller_mac;
1292 if (netdev_get_in4(in_band->of_device, &local_ip)) {
1293 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
1295 status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
1296 ETH_ADDR_ARGS(in_band->mac));
1298 controller_ip = rconn_get_ip(in_band->controller);
1299 if (controller_ip) {
1300 status_reply_put(sr, "controller-ip="IP_FMT,
1301 IP_ARGS(&controller_ip));
1303 controller_mac = get_controller_mac(in_band);
1304 if (controller_mac) {
1305 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
1306 ETH_ADDR_ARGS(controller_mac));
1311 get_ofp_packet_payload(struct ofp_packet_in *opi, struct buffer *payload)
1313 payload->data = opi->data;
1314 payload->size = ntohs(opi->header.length) - offsetof(struct ofp_packet_in,
1319 in_band_hook_create(const struct settings *s, struct switch_status *ss,
1320 struct rconn *remote)
1322 struct in_band_data *in_band;
1325 in_band = xcalloc(1, sizeof *in_band);
1327 in_band->ml = mac_learning_create();
1328 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE,
1329 &in_band->of_device);
1331 fatal(retval, "Could not open %s device", s->of_name);
1333 memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device),
1335 in_band->controller = remote;
1336 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
1337 return make_hook(in_band_local_packet_cb, NULL, NULL, NULL, in_band);
1340 /* Fail open support. */
1342 struct fail_open_data {
1343 const struct settings *s;
1344 struct rconn *local_rconn;
1345 struct rconn *remote_rconn;
1346 struct lswitch *lswitch;
1347 int last_disconn_secs;
1348 time_t boot_deadline;
1351 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
1353 fail_open_periodic_cb(void *fail_open_)
1355 struct fail_open_data *fail_open = fail_open_;
1359 if (time_now() < fail_open->boot_deadline) {
1362 disconn_secs = rconn_disconnected_duration(fail_open->remote_rconn);
1363 open = disconn_secs >= fail_open->s->probe_interval * 3;
1364 if (open != (fail_open->lswitch != NULL)) {
1366 VLOG_WARN("No longer in fail-open mode");
1367 lswitch_destroy(fail_open->lswitch);
1368 fail_open->lswitch = NULL;
1370 VLOG_WARN("Could not connect to controller for %d seconds, "
1371 "failing open", disconn_secs);
1372 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
1373 fail_open->s->max_idle);
1374 fail_open->last_disconn_secs = disconn_secs;
1376 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
1377 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
1378 "from controller", disconn_secs);
1379 fail_open->last_disconn_secs = disconn_secs;
1384 fail_open_local_packet_cb(struct relay *r, void *fail_open_)
1386 struct fail_open_data *fail_open = fail_open_;
1387 if (!fail_open->lswitch) {
1390 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
1391 r->halves[HALF_LOCAL].rxbuf);
1392 rconn_run(fail_open->local_rconn);
1398 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
1400 struct fail_open_data *fail_open = fail_open_;
1401 const struct settings *s = fail_open->s;
1402 int trigger_duration = s->probe_interval * 3;
1403 int cur_duration = rconn_disconnected_duration(fail_open->remote_rconn);
1405 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
1406 status_reply_put(sr, "current-duration=%d", cur_duration);
1407 status_reply_put(sr, "triggered=%s",
1408 cur_duration >= trigger_duration ? "true" : "false");
1409 status_reply_put(sr, "max-idle=%d", s->max_idle);
1413 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
1414 struct rconn *local_rconn, struct rconn *remote_rconn)
1416 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
1418 fail_open->local_rconn = local_rconn;
1419 fail_open->remote_rconn = remote_rconn;
1420 fail_open->lswitch = NULL;
1421 fail_open->boot_deadline = time_now() + s->probe_interval * 3;
1422 fail_open->boot_deadline += STP_EXTRA_BOOT_TIME;
1423 switch_status_register_category(ss, "fail-open",
1424 fail_open_status_cb, fail_open);
1425 return make_hook(fail_open_local_packet_cb, NULL,
1426 fail_open_periodic_cb, NULL, fail_open);
1429 struct rate_limiter {
1430 const struct settings *s;
1431 struct rconn *remote_rconn;
1433 /* One queue per physical port. */
1434 struct queue queues[OFPP_MAX];
1435 int n_queued; /* Sum over queues[*].n. */
1436 int next_tx_port; /* Next port to check in round-robin. */
1440 * It costs 1000 tokens to send a single packet_in message. A single token
1441 * per message would be more straightforward, but this choice lets us avoid
1442 * round-off error in refill_bucket()'s calculation of how many tokens to
1443 * add to the bucket, since no division step is needed. */
1444 long long int last_fill; /* Time at which we last added tokens. */
1445 int tokens; /* Current number of tokens. */
1447 /* Transmission queue. */
1448 int n_txq; /* No. of packets waiting in rconn for tx. */
1450 /* Statistics reporting. */
1451 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
1452 unsigned long long n_limited; /* # queued for rate limiting. */
1453 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
1454 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
1457 /* Drop a packet from the longest queue in 'rl'. */
1459 drop_packet(struct rate_limiter *rl)
1461 struct queue *longest; /* Queue currently selected as longest. */
1462 int n_longest; /* # of queues of same length as 'longest'. */
1465 longest = &rl->queues[0];
1467 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
1468 if (longest->n < q->n) {
1471 } else if (longest->n == q->n) {
1474 /* Randomly select one of the longest queues, with a uniform
1475 * distribution (Knuth algorithm 3.4.2R). */
1476 if (!random_range(n_longest)) {
1482 /* FIXME: do we want to pop the tail instead? */
1483 buffer_delete(queue_pop_head(longest));
1487 /* Remove and return the next packet to transmit (in round-robin order). */
1488 static struct buffer *
1489 dequeue_packet(struct rate_limiter *rl)
1493 for (i = 0; i < OFPP_MAX; i++) {
1494 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
1495 struct queue *q = &rl->queues[port];
1497 rl->next_tx_port = (port + 1) % OFPP_MAX;
1499 return queue_pop_head(q);
1505 /* Add tokens to the bucket based on elapsed time. */
1507 refill_bucket(struct rate_limiter *rl)
1509 const struct settings *s = rl->s;
1510 long long int now = time_msec();
1511 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
1512 if (tokens >= 1000) {
1513 rl->last_fill = now;
1514 rl->tokens = MIN(tokens, s->burst_limit * 1000);
1518 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
1519 * true if successful, false otherwise. (In the latter case no tokens are
1522 get_token(struct rate_limiter *rl)
1524 if (rl->tokens >= 1000) {
1533 rate_limit_local_packet_cb(struct relay *r, void *rl_)
1535 struct rate_limiter *rl = rl_;
1536 const struct settings *s = rl->s;
1537 struct ofp_packet_in *opi;
1539 opi = get_ofp_packet_in(r);
1544 if (!rl->n_queued && get_token(rl)) {
1545 /* In the common case where we are not constrained by the rate limit,
1546 * let the packet take the normal path. */
1550 /* Otherwise queue it up for the periodic callback to drain out. */
1551 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
1552 int port = ntohs(opi->in_port) % OFPP_MAX;
1553 if (rl->n_queued >= s->burst_limit) {
1556 queue_push_tail(&rl->queues[port], buffer_clone(msg));
1564 rate_limit_status_cb(struct status_reply *sr, void *rl_)
1566 struct rate_limiter *rl = rl_;
1568 status_reply_put(sr, "normal=%llu", rl->n_normal);
1569 status_reply_put(sr, "limited=%llu", rl->n_limited);
1570 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
1571 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
1575 rate_limit_periodic_cb(void *rl_)
1577 struct rate_limiter *rl = rl_;
1580 /* Drain some packets out of the bucket if possible, but limit the number
1581 * of iterations to allow other code to get work done too. */
1583 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
1584 /* Use a small, arbitrary limit for the amount of queuing to do here,
1585 * because the TCP connection is responsible for buffering and there is
1586 * no point in trying to transmit faster than the TCP connection can
1588 struct buffer *b = dequeue_packet(rl);
1589 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
1596 rate_limit_wait_cb(void *rl_)
1598 struct rate_limiter *rl = rl_;
1600 if (rl->tokens >= 1000) {
1601 /* We can transmit more packets as soon as we're called again. */
1602 poll_immediate_wake();
1604 /* We have to wait for the bucket to re-fill. We could calculate
1605 * the exact amount of time here for increased smoothness. */
1606 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
1612 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
1613 struct rconn *local, struct rconn *remote)
1615 struct rate_limiter *rl;
1618 rl = xcalloc(1, sizeof *rl);
1620 rl->remote_rconn = remote;
1621 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
1622 queue_init(&rl->queues[i]);
1624 rl->last_fill = time_msec();
1625 rl->tokens = s->rate_limit * 100;
1626 switch_status_register_category(ss, "rate-limit",
1627 rate_limit_status_cb, rl);
1628 return make_hook(rate_limit_local_packet_cb, NULL, rate_limit_periodic_cb,
1629 rate_limit_wait_cb, rl);
1632 /* OFPST_SWITCH statistics. */
1634 struct switch_status_category {
1636 void (*cb)(struct status_reply *, void *aux);
1640 struct switch_status {
1641 const struct settings *s;
1643 struct switch_status_category categories[8];
1647 struct status_reply {
1648 struct switch_status_category *category;
1654 switch_status_remote_packet_cb(struct relay *r, void *ss_)
1656 struct switch_status *ss = ss_;
1657 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
1658 struct buffer *msg = r->halves[HALF_REMOTE].rxbuf;
1659 struct switch_status_category *c;
1660 struct ofp_stats_request *osr;
1661 struct ofp_stats_reply *reply;
1662 struct status_reply sr;
1663 struct ofp_header *oh;
1668 if (oh->type != OFPT_STATS_REQUEST) {
1671 if (msg->size < sizeof(struct ofp_stats_request)) {
1672 VLOG_WARN_RL(&vrl, "packet too short (%zu bytes) for stats_request",
1678 if (osr->type != htons(OFPST_SWITCH)) {
1682 sr.request.string = (void *) (osr + 1);
1683 sr.request.length = msg->size - sizeof *osr;
1684 ds_init(&sr.output);
1685 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
1686 if (!memcmp(c->name, sr.request.string,
1687 MIN(strlen(c->name), sr.request.length))) {
1692 reply = make_openflow_xid((offsetof(struct ofp_stats_reply, body)
1693 + sr.output.length),
1694 OFPT_STATS_REPLY, osr->header.xid, &b);
1695 reply->type = htons(OFPST_SWITCH);
1697 memcpy(reply->body, sr.output.string, sr.output.length);
1698 retval = rconn_send(rc, b, NULL);
1699 if (retval && retval != EAGAIN) {
1700 VLOG_WARN("send failed (%s)", strerror(retval));
1702 ds_destroy(&sr.output);
1707 rconn_status_cb(struct status_reply *sr, void *rconn_)
1709 struct rconn *rconn = rconn_;
1710 time_t now = time_now();
1712 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
1713 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
1714 status_reply_put(sr, "backoff=%d", rconn_get_backoff(rconn));
1715 status_reply_put(sr, "is-connected=%s",
1716 rconn_is_connected(rconn) ? "true" : "false");
1717 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
1718 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
1719 status_reply_put(sr, "attempted-connections=%u",
1720 rconn_get_attempted_connections(rconn));
1721 status_reply_put(sr, "successful-connections=%u",
1722 rconn_get_successful_connections(rconn));
1723 status_reply_put(sr, "last-connection=%ld",
1724 (long int) (now - rconn_get_last_connection(rconn)));
1725 status_reply_put(sr, "time-connected=%lu",
1726 rconn_get_total_time_connected(rconn));
1727 status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(rconn));
1731 config_status_cb(struct status_reply *sr, void *s_)
1733 const struct settings *s = s_;
1736 for (i = 0; i < s->n_listeners; i++) {
1737 status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
1739 if (s->probe_interval) {
1740 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
1742 if (s->max_backoff) {
1743 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
1748 switch_status_cb(struct status_reply *sr, void *ss_)
1750 struct switch_status *ss = ss_;
1751 time_t now = time_now();
1753 status_reply_put(sr, "now=%ld", (long int) now);
1754 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
1755 status_reply_put(sr, "pid=%ld", (long int) getpid());
1759 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
1761 struct switch_status *ss = xcalloc(1, sizeof *ss);
1763 ss->booted = time_now();
1764 switch_status_register_category(ss, "config",
1765 config_status_cb, (void *) s);
1766 switch_status_register_category(ss, "switch", switch_status_cb, ss);
1768 return make_hook(NULL, switch_status_remote_packet_cb, NULL, NULL, ss);
1772 switch_status_register_category(struct switch_status *ss,
1773 const char *category,
1774 void (*cb)(struct status_reply *,
1778 struct switch_status_category *c;
1779 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
1780 c = &ss->categories[ss->n_categories++];
1783 c->name = xstrdup(category);
1787 status_reply_put(struct status_reply *sr, const char *content, ...)
1789 size_t old_length = sr->output.length;
1793 /* Append the status reply to the output. */
1794 ds_put_format(&sr->output, "%s.", sr->category->name);
1795 va_start(args, content);
1796 ds_put_format_valist(&sr->output, content, args);
1798 if (ds_last(&sr->output) != '\n') {
1799 ds_put_char(&sr->output, '\n');
1802 /* Drop what we just added if it doesn't match the request. */
1803 added = sr->output.length - old_length;
1804 if (added < sr->request.length
1805 || memcmp(&sr->output.string[old_length],
1806 sr->request.string, sr->request.length)) {
1807 ds_truncate(&sr->output, old_length);
1812 /* Controller discovery. */
1816 const struct settings *s;
1817 struct dhclient *dhcp;
1822 discovery_status_cb(struct status_reply *sr, void *d_)
1824 struct discovery *d = d_;
1826 status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re);
1827 status_reply_put(sr, "n-changes=%d", d->n_changes);
1828 status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp));
1829 status_reply_put(sr, "state-elapsed=%u",
1830 dhclient_get_state_elapsed(d->dhcp));
1831 if (dhclient_is_bound(d->dhcp)) {
1832 uint32_t ip = dhclient_get_ip(d->dhcp);
1833 uint32_t netmask = dhclient_get_netmask(d->dhcp);
1834 uint32_t router = dhclient_get_router(d->dhcp);
1836 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
1837 uint32_t dns_server;
1841 status_reply_put(sr, "ip="IP_FMT, IP_ARGS(&ip));
1842 status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask));
1844 status_reply_put(sr, "router="IP_FMT, IP_ARGS(&router));
1847 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i, &dns_server);
1849 status_reply_put(sr, "dns%d="IP_FMT, i, IP_ARGS(&dns_server));
1852 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
1854 status_reply_put(sr, "domain=%s", domain_name);
1858 status_reply_put(sr, "lease-remaining=%u",
1859 dhclient_get_lease_remaining(d->dhcp));
1863 static struct discovery *
1864 discovery_init(const struct settings *s, struct switch_status *ss)
1866 struct netdev *netdev;
1867 struct discovery *d;
1868 struct dhclient *dhcp;
1871 /* Bring ofX network device up. */
1872 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1874 fatal(retval, "Could not open %s device", s->of_name);
1876 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
1878 fatal(retval, "Could not bring %s device up", s->of_name);
1880 netdev_close(netdev);
1882 /* Initialize DHCP client. */
1883 retval = dhclient_create(s->of_name, modify_dhcp_request,
1884 validate_dhcp_offer, (void *) s, &dhcp);
1886 fatal(retval, "Failed to initialize DHCP client");
1888 dhclient_init(dhcp, 0);
1890 d = xmalloc(sizeof *d);
1895 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
1901 discovery_question_connectivity(struct discovery *d)
1903 dhclient_force_renew(d->dhcp, 15);
1907 discovery_run(struct discovery *d, char **controller_name)
1909 dhclient_run(d->dhcp);
1910 if (!dhclient_changed(d->dhcp)) {
1914 dhclient_configure_netdev(d->dhcp);
1915 if (d->s->update_resolv_conf) {
1916 dhclient_update_resolv_conf(d->dhcp);
1919 if (dhclient_is_bound(d->dhcp)) {
1920 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
1921 DHCP_CODE_OFP_CONTROLLER_VCONN);
1922 VLOG_WARN("%s: discovered controller", *controller_name);
1925 *controller_name = NULL;
1927 VLOG_WARN("discovered controller no longer available");
1935 discovery_wait(struct discovery *d)
1937 dhclient_wait(d->dhcp);
1941 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
1943 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
1947 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
1949 const struct settings *s = s_;
1953 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
1955 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
1958 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
1960 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
1961 s->accept_controller_re);
1967 /* User interface. */
1970 parse_options(int argc, char *argv[], struct settings *s)
1973 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
1975 OPT_INACTIVITY_PROBE,
1981 static struct option long_options[] = {
1982 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
1983 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
1984 {"fail", required_argument, 0, 'F'},
1985 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
1986 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
1987 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
1988 {"listen", required_argument, 0, 'l'},
1989 {"monitor", required_argument, 0, 'm'},
1990 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
1991 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
1992 {"detach", no_argument, 0, 'D'},
1993 {"force", no_argument, 0, 'f'},
1994 {"pidfile", optional_argument, 0, 'P'},
1995 {"verbose", optional_argument, 0, 'v'},
1996 {"help", no_argument, 0, 'h'},
1997 {"version", no_argument, 0, 'V'},
1998 VCONN_SSL_LONG_OPTIONS
2001 char *short_options = long_options_to_short_options(long_options);
2002 char *accept_re = NULL;
2005 /* Set defaults that we can figure out before parsing options. */
2007 s->monitor_name = NULL;
2008 s->fail_mode = FAIL_OPEN;
2010 s->probe_interval = 15;
2011 s->max_backoff = 15;
2012 s->update_resolv_conf = true;
2018 c = getopt_long(argc, argv, short_options, long_options, NULL);
2024 case OPT_ACCEPT_VCONN:
2025 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
2028 case OPT_NO_RESOLV_CONF:
2029 s->update_resolv_conf = false;
2033 if (!strcmp(optarg, "open")) {
2034 s->fail_mode = FAIL_OPEN;
2035 } else if (!strcmp(optarg, "closed")) {
2036 s->fail_mode = FAIL_CLOSED;
2039 "-f or --fail argument must be \"open\" or \"closed\"");
2043 case OPT_INACTIVITY_PROBE:
2044 s->probe_interval = atoi(optarg);
2045 if (s->probe_interval < 5) {
2046 fatal(0, "--inactivity-probe argument must be at least 5");
2051 if (!strcmp(optarg, "permanent")) {
2052 s->max_idle = OFP_FLOW_PERMANENT;
2054 s->max_idle = atoi(optarg);
2055 if (s->max_idle < 1 || s->max_idle > 65535) {
2056 fatal(0, "--max-idle argument must be between 1 and "
2057 "65535 or the word 'permanent'");
2062 case OPT_MAX_BACKOFF:
2063 s->max_backoff = atoi(optarg);
2064 if (s->max_backoff < 1) {
2065 fatal(0, "--max-backoff argument must be at least 1");
2066 } else if (s->max_backoff > 3600) {
2067 s->max_backoff = 3600;
2071 case OPT_RATE_LIMIT:
2073 s->rate_limit = atoi(optarg);
2074 if (s->rate_limit < 1) {
2075 fatal(0, "--rate-limit argument must be at least 1");
2078 s->rate_limit = 1000;
2082 case OPT_BURST_LIMIT:
2083 s->burst_limit = atoi(optarg);
2084 if (s->burst_limit < 1) {
2085 fatal(0, "--burst-limit argument must be at least 1");
2094 set_pidfile(optarg);
2098 ignore_existing_pidfile();
2102 if (s->n_listeners >= MAX_MGMT) {
2103 fatal(0, "-l or --listen may be specified at most %d times",
2106 s->listener_names[s->n_listeners++] = optarg;
2110 if (s->monitor_name) {
2111 fatal(0, "-m or --monitor may only be specified once");
2113 s->monitor_name = optarg;
2120 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
2124 vlog_set_verbosity(optarg);
2127 VCONN_SSL_OPTION_HANDLERS
2136 free(short_options);
2140 if (argc < 1 || argc > 2) {
2141 fatal(0, "need one or two non-option arguments; use --help for usage");
2144 /* Local and remote vconns. */
2145 s->nl_name = argv[0];
2146 if (strncmp(s->nl_name, "nl:", 3)
2147 || strlen(s->nl_name) < 4
2148 || s->nl_name[strspn(s->nl_name + 3, "0123456789") + 3]) {
2149 fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"", s->nl_name);
2151 s->of_name = xasprintf("of%s", s->nl_name + 3);
2152 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
2154 /* Set accept_controller_regex. */
2156 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
2158 retval = regcomp(&s->accept_controller_regex, accept_re,
2159 REG_NOSUB | REG_EXTENDED);
2161 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
2162 char *buffer = xmalloc(length);
2163 regerror(retval, &s->accept_controller_regex, buffer, length);
2164 fatal(0, "%s: %s", accept_re, buffer);
2166 s->accept_controller_re = accept_re;
2168 /* Mode of operation. */
2169 s->discovery = s->controller_name == NULL;
2173 enum netdev_flags flags;
2174 struct netdev *netdev;
2176 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
2178 fatal(retval, "Could not open %s device", s->of_name);
2181 retval = netdev_get_flags(netdev, &flags);
2183 fatal(retval, "Could not get flags for %s device", s->of_name);
2186 s->in_band = (flags & NETDEV_UP) != 0;
2187 if (s->in_band && netdev_get_in6(netdev, NULL)) {
2188 VLOG_WARN("Ignoring IPv6 address on %s device: IPv6 not supported",
2192 netdev_close(netdev);
2195 /* Rate limiting. */
2196 if (s->rate_limit) {
2197 if (s->rate_limit < 100) {
2198 VLOG_WARN("Rate limit set to unusually low value %d",
2201 if (!s->burst_limit) {
2202 s->burst_limit = s->rate_limit / 4;
2204 s->burst_limit = MAX(s->burst_limit, 1);
2205 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
2212 printf("%s: secure channel, a relay for OpenFlow messages.\n"
2213 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
2214 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
2215 "CONTROLLER is an active OpenFlow connection method; if it is\n"
2216 "omitted, then secchan performs controller discovery.\n",
2217 program_name, program_name);
2218 vconn_usage(true, true);
2219 printf("\nController discovery options:\n"
2220 " --accept-vconn=REGEX accept matching discovered controllers\n"
2221 " --no-resolv-conf do not update /etc/resolv.conf\n"
2222 "\nNetworking options:\n"
2223 " -F, --fail=open|closed when controller connection fails:\n"
2224 " closed: drop all packets\n"
2225 " open (default): act as learning switch\n"
2226 " --inactivity-probe=SECS time between inactivity probes\n"
2227 " --max-idle=SECS max idle for flows set up by secchan\n"
2228 " --max-backoff=SECS max time between controller connection\n"
2229 " attempts (default: 15 seconds)\n"
2230 " -l, --listen=METHOD allow management connections on METHOD\n"
2231 " (a passive OpenFlow connection method)\n"
2232 " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n"
2233 " (a passive OpenFlow connection method)\n"
2234 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
2235 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
2236 " --burst-limit=BURST limit on packet credit for idle time\n"
2237 "\nOther options:\n"
2238 " -D, --detach run in background as daemon\n"
2239 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
2240 " -f, --force with -P, start even if already running\n"
2241 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
2242 " -v, --verbose set maximum verbosity level\n"
2243 " -h, --help display this help message\n"
2244 " -V, --version display version information\n",