1 /* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
4 * We are making the OpenFlow specification and associated documentation
5 * (Software) available for public use and benefit with the expectation
6 * that others will use, modify and enhance the Software and contribute
7 * those enhancements back to the community. However, since we would
8 * like to make the Software available for broadest use, with as few
9 * restrictions as possible permission is hereby granted, free of
10 * charge, to any person obtaining a copy of this Software to deal in
11 * the Software under the copyrights without restriction, including
12 * without limitation the rights to use, copy, modify, merge, publish,
13 * distribute, sublicense, and/or sell copies of the Software, and to
14 * permit persons to whom the Software is furnished to do so, subject to
15 * the following conditions:
17 * The above copyright notice and this permission notice shall be
18 * included in all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
23 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
24 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
25 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
26 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 * The name and trademarks of copyright holder(s) may NOT be used in
30 * advertising or publicity pertaining to the Software or any
31 * derivatives without specific, written prior permission.
39 #include <netinet/in.h>
49 #include "command-line.h"
53 #include "dhcp-client.h"
54 #include "dynamic-string.h"
57 #include "learning-switch.h"
59 #include "mac-learning.h"
61 #include "nicira-ext.h"
64 #include "poll-loop.h"
69 #include "vconn-ssl.h"
71 #include "vlog-socket.h"
74 #define THIS_MODULE VLM_secchan
76 /* Behavior when the connection to the controller fails. */
78 FAIL_OPEN, /* Act as learning switch. */
79 FAIL_CLOSED /* Drop all packets. */
82 /* Maximum number of management connection listeners. */
85 /* Settings that may be configured by the user. */
87 /* Overall mode of operation. */
88 bool discovery; /* Discover the controller automatically? */
89 bool in_band; /* Connect to controller in-band? */
91 /* Related vconns and network devices. */
92 const char *nl_name; /* Local datapath (must be "nl:" vconn). */
93 char *of_name; /* ofX network device name. */
94 const char *controller_name; /* Controller (if not discovery mode). */
95 const char *listener_names[MAX_MGMT]; /* Listen for mgmt connections. */
96 size_t n_listeners; /* Number of mgmt connection listeners. */
97 const char *monitor_name; /* Listen for traffic monitor connections. */
99 /* Failure behavior. */
100 enum fail_mode fail_mode; /* Act as learning switch if no controller? */
101 int max_idle; /* Idle time for flows in fail-open mode. */
102 int probe_interval; /* # seconds idle before sending echo request. */
103 int max_backoff; /* Max # seconds between connection attempts. */
105 /* Packet-in rate-limiting. */
106 int rate_limit; /* Tokens added to bucket per second. */
107 int burst_limit; /* Maximum number token bucket size. */
109 /* Discovery behavior. */
110 regex_t accept_controller_regex; /* Controller vconns to accept. */
111 const char *accept_controller_re; /* String version of regex. */
112 bool update_resolv_conf; /* Update /etc/resolv.conf? */
117 struct buffer *rxbuf;
118 int n_txq; /* No. of packets queued for tx on 'rconn'. */
125 #define HALF_REMOTE 1
126 struct half halves[2];
132 bool (*packet_cb[2])(struct relay *, void *aux);
133 void (*periodic_cb)(void *aux);
134 void (*wait_cb)(void *aux);
138 static struct vlog_rate_limit vrl = VLOG_RATE_LIMIT_INIT(60, 60);
140 static void parse_options(int argc, char *argv[], struct settings *);
141 static void usage(void) NO_RETURN;
143 static struct vconn *open_passive_vconn(const char *name);
144 static struct vconn *accept_vconn(struct vconn *vconn);
146 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
148 static struct relay *relay_accept(const struct settings *, struct vconn *);
149 static void relay_run(struct relay *, const struct hook[], size_t n_hooks);
150 static void relay_wait(struct relay *);
151 static void relay_destroy(struct relay *);
153 static struct hook make_hook(bool (*local_packet_cb)(struct relay *, void *),
154 bool (*remote_packet_cb)(struct relay *, void *),
155 void (*periodic_cb)(void *),
156 void (*wait_cb)(void *),
158 static struct ofp_packet_in *get_ofp_packet_in(struct relay *);
159 static bool get_ofp_packet_eth_header(struct relay *, struct ofp_packet_in **,
160 struct eth_header **);
161 static void get_ofp_packet_payload(struct ofp_packet_in *, struct buffer *);
163 struct switch_status;
165 static struct hook switch_status_hook_create(const struct settings *,
166 struct switch_status **);
167 static void switch_status_register_category(struct switch_status *,
168 const char *category,
169 void (*cb)(struct status_reply *,
172 static void status_reply_put(struct status_reply *, const char *, ...)
175 static void rconn_status_cb(struct status_reply *, void *rconn_);
177 static struct discovery *discovery_init(const struct settings *,
178 struct switch_status *);
179 static void discovery_question_connectivity(struct discovery *);
180 static bool discovery_run(struct discovery *, char **controller_name);
181 static void discovery_wait(struct discovery *);
183 static struct hook in_band_hook_create(const struct settings *,
184 struct switch_status *,
185 struct rconn *remote);
188 static struct hook port_watcher_create(struct rconn *local,
189 struct rconn *remote,
190 struct port_watcher **);
191 static uint32_t port_watcher_get_flags(const struct port_watcher *,
193 static void port_watcher_set_flags(struct port_watcher *,
194 int port_no, uint32_t flags, uint32_t mask);
196 static struct hook stp_hook_create(const struct settings *,
197 struct port_watcher *,
198 struct rconn *local, struct rconn *remote);
200 static struct hook fail_open_hook_create(const struct settings *,
201 struct switch_status *,
203 struct rconn *remote);
204 static struct hook rate_limit_hook_create(const struct settings *,
205 struct switch_status *,
207 struct rconn *remote);
210 static void modify_dhcp_request(struct dhcp_msg *, void *aux);
211 static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux);
214 main(int argc, char *argv[])
218 struct list relays = LIST_INITIALIZER(&relays);
220 struct hook hooks[8];
223 struct vconn *monitor;
225 struct vconn *listeners[MAX_MGMT];
228 struct rconn *local_rconn, *remote_rconn;
229 struct relay *controller_relay;
230 struct discovery *discovery;
231 struct switch_status *switch_status;
232 struct port_watcher *pw;
236 set_program_name(argv[0]);
237 register_fault_handlers();
240 parse_options(argc, argv, &s);
241 signal(SIGPIPE, SIG_IGN);
243 /* Start listening for management and monitoring connections. */
245 for (i = 0; i < s.n_listeners; i++) {
246 listeners[n_listeners++] = open_passive_vconn(s.listener_names[i]);
248 monitor = s.monitor_name ? open_passive_vconn(s.monitor_name) : NULL;
250 /* Initialize switch status hook. */
251 hooks[n_hooks++] = switch_status_hook_create(&s, &switch_status);
253 /* Start controller discovery. */
254 discovery = s.discovery ? discovery_init(&s, switch_status) : NULL;
256 /* Start listening for vlogconf requests. */
257 retval = vlog_server_listen(NULL, NULL);
259 fatal(retval, "Could not listen for vlog connections");
262 die_if_already_running();
265 VLOG_WARN("OpenFlow reference implementation version %s", VERSION);
266 VLOG_WARN("OpenFlow protocol version 0x%02x", OFP_VERSION);
268 /* Connect to datapath. */
269 local_rconn = rconn_create(0, s.max_backoff);
270 rconn_connect(local_rconn, s.nl_name);
271 switch_status_register_category(switch_status, "local",
272 rconn_status_cb, local_rconn);
274 /* Connect to controller. */
275 remote_rconn = rconn_create(s.probe_interval, s.max_backoff);
276 if (s.controller_name) {
277 retval = rconn_connect(remote_rconn, s.controller_name);
278 if (retval == EAFNOSUPPORT) {
279 fatal(0, "No support for %s vconn", s.controller_name);
282 switch_status_register_category(switch_status, "remote",
283 rconn_status_cb, remote_rconn);
285 /* Start relaying. */
286 controller_relay = relay_create(local_rconn, remote_rconn, false);
287 list_push_back(&relays, &controller_relay->node);
290 hooks[n_hooks++] = port_watcher_create(local_rconn, remote_rconn, &pw);
291 hooks[n_hooks++] = stp_hook_create(&s, pw, local_rconn, remote_rconn);
293 hooks[n_hooks++] = in_band_hook_create(&s, switch_status,
296 if (s.fail_mode == FAIL_OPEN) {
297 hooks[n_hooks++] = fail_open_hook_create(&s, switch_status,
298 local_rconn, remote_rconn);
301 hooks[n_hooks++] = rate_limit_hook_create(&s, switch_status,
302 local_rconn, remote_rconn);
304 assert(n_hooks <= ARRAY_SIZE(hooks));
311 LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) {
312 relay_run(r, hooks, n_hooks);
314 for (i = 0; i < n_listeners; i++) {
316 struct relay *r = relay_accept(&s, listeners[i]);
320 list_push_back(&relays, &r->node);
324 struct vconn *new = accept_vconn(monitor);
326 rconn_add_monitor(local_rconn, new);
329 for (i = 0; i < n_hooks; i++) {
330 if (hooks[i].periodic_cb) {
331 hooks[i].periodic_cb(hooks[i].aux);
335 char *controller_name;
336 if (rconn_is_connectivity_questionable(remote_rconn)) {
337 discovery_question_connectivity(discovery);
339 if (discovery_run(discovery, &controller_name)) {
340 if (controller_name) {
341 rconn_connect(remote_rconn, controller_name);
343 rconn_disconnect(remote_rconn);
348 /* Wait for something to happen. */
349 LIST_FOR_EACH (r, struct relay, node, &relays) {
352 for (i = 0; i < n_listeners; i++) {
353 vconn_accept_wait(listeners[i]);
356 vconn_accept_wait(monitor);
358 for (i = 0; i < n_hooks; i++) {
359 if (hooks[i].wait_cb) {
360 hooks[i].wait_cb(hooks[i].aux);
364 discovery_wait(discovery);
372 static struct vconn *
373 open_passive_vconn(const char *name)
378 retval = vconn_open(name, &vconn);
379 if (retval && retval != EAGAIN) {
380 fatal(retval, "opening %s", name);
382 if (!vconn_is_passive(vconn)) {
383 fatal(0, "%s is not a passive vconn", name);
388 static struct vconn *
389 accept_vconn(struct vconn *vconn)
394 retval = vconn_accept(vconn, &new);
395 if (retval && retval != EAGAIN) {
396 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
402 make_hook(bool (*local_packet_cb)(struct relay *, void *aux),
403 bool (*remote_packet_cb)(struct relay *, void *aux),
404 void (*periodic_cb)(void *aux),
405 void (*wait_cb)(void *aux),
409 h.packet_cb[HALF_LOCAL] = local_packet_cb;
410 h.packet_cb[HALF_REMOTE] = remote_packet_cb;
411 h.periodic_cb = periodic_cb;
417 static struct ofp_packet_in *
418 get_ofp_packet_in(struct relay *r)
420 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
421 struct ofp_header *oh = msg->data;
422 if (oh->type == OFPT_PACKET_IN) {
423 if (msg->size >= offsetof (struct ofp_packet_in, data)) {
426 VLOG_WARN("packet too short (%zu bytes) for packet_in",
434 get_ofp_packet_eth_header(struct relay *r, struct ofp_packet_in **opip,
435 struct eth_header **ethp)
437 const int min_len = offsetof(struct ofp_packet_in, data) + ETH_HEADER_LEN;
438 struct ofp_packet_in *opi = get_ofp_packet_in(r);
439 if (opi && ntohs(opi->header.length) >= min_len) {
441 *ethp = (void *) opi->data;
448 /* OpenFlow message relaying. */
450 static struct relay *
451 relay_accept(const struct settings *s, struct vconn *listen_vconn)
453 struct vconn *new_remote, *new_local;
454 char *nl_name_without_subscription;
455 struct rconn *r1, *r2;
458 new_remote = accept_vconn(listen_vconn);
463 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We
464 * only accept the former syntax in main().
466 * nl:123:0 opens a netlink connection to local datapath 123 without
467 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
469 nl_name_without_subscription = xasprintf("%s:0", s->nl_name);
470 retval = vconn_open(nl_name_without_subscription, &new_local);
472 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
473 nl_name_without_subscription, strerror(retval));
474 vconn_close(new_remote);
475 free(nl_name_without_subscription);
479 /* Create and return relay. */
480 r1 = rconn_create(0, 0);
481 rconn_connect_unreliably(r1, nl_name_without_subscription, new_local);
482 free(nl_name_without_subscription);
484 r2 = rconn_create(0, 0);
485 rconn_connect_unreliably(r2, "passive", new_remote);
487 return relay_create(r1, r2, true);
490 static struct relay *
491 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
493 struct relay *r = xcalloc(1, sizeof *r);
494 r->halves[HALF_LOCAL].rconn = local;
495 r->halves[HALF_REMOTE].rconn = remote;
496 r->is_mgmt_conn = is_mgmt_conn;
501 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
506 for (i = 0; i < 2; i++) {
507 rconn_run(r->halves[i].rconn);
510 /* Limit the number of iterations to prevent other tasks from starving. */
511 for (iteration = 0; iteration < 50; iteration++) {
512 bool progress = false;
513 for (i = 0; i < 2; i++) {
514 struct half *this = &r->halves[i];
515 struct half *peer = &r->halves[!i];
518 this->rxbuf = rconn_recv(this->rconn);
519 if (this->rxbuf && (i == HALF_REMOTE || !r->is_mgmt_conn)) {
520 const struct hook *h;
521 for (h = hooks; h < &hooks[n_hooks]; h++) {
522 if (h->packet_cb[i] && h->packet_cb[i](r, h->aux)) {
523 buffer_delete(this->rxbuf);
532 if (this->rxbuf && !this->n_txq) {
533 int retval = rconn_send(peer->rconn, this->rxbuf,
535 if (retval != EAGAIN) {
539 buffer_delete(this->rxbuf);
550 if (r->is_mgmt_conn) {
551 for (i = 0; i < 2; i++) {
552 struct half *this = &r->halves[i];
553 if (!rconn_is_alive(this->rconn)) {
562 relay_wait(struct relay *r)
566 for (i = 0; i < 2; i++) {
567 struct half *this = &r->halves[i];
569 rconn_run_wait(this->rconn);
571 rconn_recv_wait(this->rconn);
577 relay_destroy(struct relay *r)
581 list_remove(&r->node);
582 for (i = 0; i < 2; i++) {
583 struct half *this = &r->halves[i];
584 rconn_destroy(this->rconn);
585 buffer_delete(this->rxbuf);
590 /* Port status watcher. */
592 typedef void port_watcher_cb_func(uint16_t port_no,
593 const struct ofp_phy_port *old,
594 const struct ofp_phy_port *new,
597 struct port_watcher_cb {
598 port_watcher_cb_func *function;
602 struct port_watcher {
603 struct rconn *local_rconn;
604 struct rconn *remote_rconn;
605 struct ofp_phy_port ports[OFPP_MAX + 1];
606 time_t last_feature_request;
607 bool got_feature_reply;
609 struct port_watcher_cb cbs[2];
613 /* Returns the number of fields that differ from 'a' to 'b'. */
615 opp_differs(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
617 BUILD_ASSERT_DECL(sizeof *a == 36); /* Trips when we add or remove fields. */
618 return ((a->port_no != b->port_no)
619 + (memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) != 0)
620 + (memcmp(a->name, b->name, sizeof a->name) != 0)
621 + (a->flags != b->flags)
622 + (a->speed != b->speed)
623 + (a->features != b->features));
627 sanitize_opp(struct ofp_phy_port *opp)
631 for (i = 0; i < sizeof opp->name; i++) {
632 char c = opp->name[i];
633 if (c && (c < 0x20 || c > 0x7e)) {
637 opp->name[sizeof opp->name - 1] = '\0';
641 port_no_to_pw_idx(int port_no)
643 return (port_no < OFPP_MAX ? port_no
644 : port_no == OFPP_LOCAL ? OFPP_MAX
649 call_pw_callbacks(struct port_watcher *pw, int port_no,
650 const struct ofp_phy_port *old,
651 const struct ofp_phy_port *new)
653 if (opp_differs(old, new)) {
655 for (i = 0; i < pw->n_cbs; i++) {
656 pw->cbs[i].function(port_no, old, new, pw->cbs[i].aux);
662 update_phy_port(struct port_watcher *pw, struct ofp_phy_port *opp,
663 uint8_t reason, bool seen[OFPP_MAX + 1])
665 struct ofp_phy_port *pw_opp;
666 struct ofp_phy_port old;
670 port_no = ntohs(opp->port_no);
671 idx = port_no_to_pw_idx(port_no);
680 pw_opp = &pw->ports[idx];
682 if (reason == OFPPR_DELETE) {
683 memset(pw_opp, 0, sizeof *pw_opp);
684 pw_opp->port_no = htons(OFPP_NONE);
685 } else if (reason == OFPPR_MOD || reason == OFPPR_ADD) {
687 sanitize_opp(pw_opp);
689 call_pw_callbacks(pw, port_no, &old, pw_opp);
693 port_watcher_local_packet_cb(struct relay *r, void *pw_)
695 struct port_watcher *pw = pw_;
696 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
697 struct ofp_header *oh = msg->data;
699 if (oh->type == OFPT_FEATURES_REPLY
700 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
701 struct ofp_switch_features *osf = msg->data;
702 bool seen[ARRAY_SIZE(pw->ports)];
706 pw->got_feature_reply = true;
708 /* Update each port included in the message. */
709 memset(seen, 0, sizeof seen);
710 n_ports = ((msg->size - offsetof(struct ofp_switch_features, ports))
711 / sizeof *osf->ports);
712 for (i = 0; i < n_ports; i++) {
713 update_phy_port(pw, &osf->ports[i], OFPPR_MOD, seen);
716 /* Delete all the ports not included in the message. */
717 for (i = 0; i < ARRAY_SIZE(pw->ports); i++) {
719 update_phy_port(pw, &pw->ports[i], OFPPR_DELETE, NULL);
722 } else if (oh->type == OFPT_PORT_STATUS
723 && msg->size >= sizeof(struct ofp_port_status)) {
724 struct ofp_port_status *ops = msg->data;
725 update_phy_port(pw, &ops->desc, ops->reason, NULL);
731 port_watcher_periodic_cb(void *pw_)
733 struct port_watcher *pw = pw_;
735 if (!pw->got_feature_reply && time_now() >= pw->last_feature_request + 5) {
737 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
738 rconn_send_with_limit(pw->local_rconn, b, &pw->n_txq, 1);
739 pw->last_feature_request = time_now();
744 put_duplexes(struct ds *ds, const char *name, uint32_t features,
745 uint32_t hd_bit, uint32_t fd_bit)
747 if (features & (hd_bit | fd_bit)) {
748 ds_put_format(ds, " %s", name);
749 if (features & hd_bit) {
750 ds_put_cstr(ds, "(HD)");
752 if (features & fd_bit) {
753 ds_put_cstr(ds, "(FD)");
759 log_port_status(uint16_t port_no,
760 const struct ofp_phy_port *old,
761 const struct ofp_phy_port *new,
764 if (VLOG_IS_DBG_ENABLED()) {
765 bool was_enabled = old->port_no != htons(OFPP_NONE);
766 bool now_enabled = new->port_no != htons(OFPP_NONE);
767 uint32_t features = ntohl(new->features);
770 if (old->flags != new->flags && opp_differs(old, new) == 1) {
771 /* Don't care if only flags changed. */
776 ds_put_format(&ds, "\"%s\", "ETH_ADDR_FMT, new->name,
777 ETH_ADDR_ARGS(new->hw_addr));
778 if (ntohl(new->speed)) {
779 ds_put_format(&ds, ", speed %"PRIu32, ntohl(new->speed));
781 if (features & (OFPPF_10MB_HD | OFPPF_10MB_FD
782 | OFPPF_100MB_HD | OFPPF_100MB_FD
783 | OFPPF_1GB_HD | OFPPF_1GB_FD | OFPPF_10GB_FD)) {
784 ds_put_cstr(&ds, ", supports");
785 put_duplexes(&ds, "10M", features, OFPPF_10MB_HD, OFPPF_10MB_FD);
786 put_duplexes(&ds, "100M", features,
787 OFPPF_100MB_HD, OFPPF_100MB_FD);
788 put_duplexes(&ds, "1G", features, OFPPF_100MB_HD, OFPPF_100MB_FD);
789 if (features & OFPPF_10GB_FD) {
790 ds_put_cstr(&ds, " 10G");
793 if (was_enabled != now_enabled) {
795 VLOG_DBG("Port %d added: %s", port_no, ds_cstr(&ds));
797 VLOG_DBG("Port %d deleted", port_no);
800 VLOG_DBG("Port %d changed: %s", port_no, ds_cstr(&ds));
807 port_watcher_register_callback(struct port_watcher *pw,
808 port_watcher_cb_func *function,
811 assert(pw->n_cbs < ARRAY_SIZE(pw->cbs));
812 pw->cbs[pw->n_cbs].function = function;
813 pw->cbs[pw->n_cbs].aux = aux;
818 port_watcher_get_flags(const struct port_watcher *pw, int port_no)
820 int idx = port_no_to_pw_idx(port_no);
821 return idx >= 0 ? ntohl(pw->ports[idx].flags) : 0;
825 port_watcher_set_flags(struct port_watcher *pw,
826 int port_no, uint32_t flags, uint32_t mask)
828 struct ofp_phy_port old;
829 struct ofp_phy_port *p;
830 struct ofp_port_mod *opm;
831 struct ofp_port_status *ops;
835 idx = port_no_to_pw_idx(port_no);
841 if (!((ntohl(p->flags) ^ flags) & mask)) {
846 /* Update our idea of the flags. */
847 p->flags = ntohl(flags);
848 call_pw_callbacks(pw, port_no, &old, p);
850 /* Change the flags in the datapath. */
851 opm = make_openflow(sizeof *opm, OFPT_PORT_MOD, &b);
852 opm->mask = htonl(mask);
854 rconn_send(pw->local_rconn, b, NULL);
856 /* Notify the controller that the flags changed. */
857 ops = make_openflow(sizeof *ops, OFPT_PORT_STATUS, &b);
858 ops->reason = OFPPR_MOD;
860 rconn_send(pw->remote_rconn, b, NULL);
864 port_watcher_is_ready(const struct port_watcher *pw)
866 return pw->got_feature_reply;
870 port_watcher_create(struct rconn *local_rconn, struct rconn *remote_rconn,
871 struct port_watcher **pwp)
873 struct port_watcher *pw;
876 pw = *pwp = xcalloc(1, sizeof *pw);
877 pw->local_rconn = local_rconn;
878 pw->remote_rconn = remote_rconn;
879 pw->last_feature_request = TIME_MIN;
880 for (i = 0; i < OFPP_MAX; i++) {
881 pw->ports[i].port_no = htons(OFPP_NONE);
883 port_watcher_register_callback(pw, log_port_status, NULL);
884 return make_hook(port_watcher_local_packet_cb, NULL,
885 port_watcher_periodic_cb, NULL, pw);
888 /* Spanning tree protocol. */
890 /* Extra time, in seconds, at boot before going into fail-open, to give the
891 * spanning tree protocol time to figure out the network layout. */
892 #define STP_EXTRA_BOOT_TIME 30
896 struct port_watcher *pw;
897 struct rconn *local_rconn;
898 struct rconn *remote_rconn;
899 uint8_t dpid[ETH_ADDR_LEN];
900 long long int last_tick_256ths;
905 stp_local_packet_cb(struct relay *r, void *stp_)
907 struct stp_data *stp = stp_;
908 struct ofp_packet_in *opi;
909 struct eth_header *eth;
910 struct llc_header *llc;
911 struct buffer payload;
915 if (!get_ofp_packet_eth_header(r, &opi, ð)
916 || !eth_addr_equals(eth->eth_dst, stp_eth_addr)) {
920 port_no = ntohs(opi->in_port);
921 if (port_no >= STP_MAX_PORTS) {
922 /* STP only supports 255 ports. */
925 if (port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP) {
926 /* We're not doing STP on this port. */
930 if (opi->reason == OFPR_ACTION) {
931 /* The controller set up a flow for this, so we won't intercept it. */
935 get_ofp_packet_payload(opi, &payload);
936 flow_extract(&payload, port_no, &flow);
937 if (flow.dl_type != htons(OFP_DL_TYPE_NOT_ETH_TYPE)) {
938 VLOG_DBG("non-LLC frame received on STP multicast address");
941 llc = buffer_at_assert(&payload, sizeof *eth, sizeof *llc);
942 if (llc->llc_dsap != STP_LLC_DSAP) {
943 VLOG_DBG("bad DSAP 0x%02"PRIx8" received on STP multicast address",
948 /* Trim off padding on payload. */
949 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
950 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
952 if (buffer_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
953 struct stp_port *p = stp_get_port(stp->stp, port_no);
954 stp_received_bpdu(p, payload.data, payload.size);
963 return time_msec() * 256 / 1000;
967 stp_periodic_cb(void *stp_)
969 struct stp_data *stp = stp_;
970 long long int now_256ths = time_256ths();
971 long long int elapsed_256ths = now_256ths - stp->last_tick_256ths;
974 if (!port_watcher_is_ready(stp->pw)) {
975 /* Can't start STP until we know port flags, because port flags can
979 if (elapsed_256ths <= 0) {
983 stp_tick(stp->stp, MIN(INT_MAX, elapsed_256ths));
984 stp->last_tick_256ths = now_256ths;
986 while (stp_get_changed_port(stp->stp, &p)) {
987 int port_no = stp_port_no(p);
988 enum stp_state state = stp_port_get_state(p);
990 if (state != STP_DISABLED) {
991 VLOG_WARN("STP: Port %d entered %s state",
992 port_no, stp_state_name(state));
994 if (!(port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP)) {
998 flags = OFPPFL_STP_LISTEN;
1001 flags = OFPPFL_STP_LEARN;
1004 case STP_FORWARDING:
1005 flags = OFPPFL_STP_FORWARD;
1008 flags = OFPPFL_STP_BLOCK;
1011 VLOG_DBG_RL(&vrl, "STP: Port %d has bad state %x",
1013 flags = OFPPFL_STP_FORWARD;
1016 if (!stp_forward_in_state(state)) {
1017 flags |= OFPPFL_NO_FLOOD;
1019 port_watcher_set_flags(stp->pw, port_no, flags,
1020 OFPPFL_STP_MASK | OFPPFL_NO_FLOOD);
1022 /* We don't own those flags. */
1028 stp_wait_cb(void *stp_ UNUSED)
1030 poll_timer_wait(1000);
1034 send_bpdu(const void *bpdu, size_t bpdu_size, int port_no, void *stp_)
1036 struct stp_data *stp = stp_;
1037 struct eth_header *eth;
1038 struct llc_header *llc;
1039 struct buffer pkt, *opo;
1041 /* Packet skeleton. */
1042 buffer_init(&pkt, ETH_HEADER_LEN + LLC_HEADER_LEN + bpdu_size);
1043 eth = buffer_put_uninit(&pkt, sizeof *eth);
1044 llc = buffer_put_uninit(&pkt, sizeof *llc);
1045 buffer_put(&pkt, bpdu, bpdu_size);
1048 memcpy(eth->eth_dst, stp_eth_addr, ETH_ADDR_LEN);
1049 memcpy(eth->eth_src, stp->pw->ports[port_no].hw_addr, ETH_ADDR_LEN);
1050 eth->eth_type = htons(pkt.size - ETH_HEADER_LEN);
1053 llc->llc_dsap = STP_LLC_DSAP;
1054 llc->llc_ssap = STP_LLC_SSAP;
1055 llc->llc_cntl = STP_LLC_CNTL;
1057 opo = make_unbuffered_packet_out(&pkt, OFPP_NONE, port_no);
1058 buffer_uninit(&pkt);
1059 rconn_send_with_limit(stp->local_rconn, opo, &stp->n_txq, OFPP_MAX);
1063 stp_port_watcher_cb(uint16_t port_no,
1064 const struct ofp_phy_port *old,
1065 const struct ofp_phy_port *new,
1068 struct stp_data *stp = stp_;
1071 /* STP only supports a maximum of 255 ports, one less than OpenFlow. We
1072 * don't support STP on OFPP_LOCAL, either. */
1073 if (port_no >= STP_MAX_PORTS) {
1077 p = stp_get_port(stp->stp, port_no);
1078 if (new->port_no == htons(OFPP_NONE)
1079 || new->flags & htonl(OFPPFL_NO_STP)) {
1080 stp_port_disable(p);
1083 stp_port_set_speed(p, new->speed);
1088 stp_hook_create(const struct settings *s, struct port_watcher *pw,
1089 struct rconn *local, struct rconn *remote)
1091 uint8_t dpid[ETH_ADDR_LEN];
1092 struct netdev *netdev;
1093 struct stp_data *stp;
1096 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1098 fatal(retval, "Could not open %s device", s->of_name);
1100 memcpy(dpid, netdev_get_etheraddr(netdev), ETH_ADDR_LEN);
1101 netdev_close(netdev);
1103 stp = xcalloc(1, sizeof *stp);
1104 stp->stp = stp_create("stp", eth_addr_to_uint64(dpid), send_bpdu, stp);
1106 memcpy(stp->dpid, dpid, ETH_ADDR_LEN);
1107 stp->local_rconn = local;
1108 stp->remote_rconn = remote;
1109 stp->last_tick_256ths = time_256ths();
1111 port_watcher_register_callback(pw, stp_port_watcher_cb, stp);
1112 return make_hook(stp_local_packet_cb, NULL,
1113 stp_periodic_cb, stp_wait_cb, stp);
1116 /* In-band control. */
1118 struct in_band_data {
1119 const struct settings *s;
1120 struct mac_learning *ml;
1121 struct netdev *of_device;
1122 struct rconn *controller;
1123 uint8_t mac[ETH_ADDR_LEN];
1128 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct buffer *b)
1130 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
1133 static const uint8_t *
1134 get_controller_mac(struct in_band_data *in_band)
1136 static uint32_t ip, last_nonzero_ip;
1137 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
1138 static time_t next_refresh = 0;
1140 uint32_t last_ip = ip;
1142 time_t now = time_now();
1144 ip = rconn_get_ip(in_band->controller);
1145 if (last_ip != ip || !next_refresh || now >= next_refresh) {
1148 /* Look up MAC address. */
1149 memset(mac, 0, sizeof mac);
1151 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
1153 VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s",
1154 IP_ARGS(&ip), strerror(retval));
1157 have_mac = !eth_addr_is_zero(mac);
1159 /* Log changes in IP, MAC addresses. */
1160 if (ip && ip != last_nonzero_ip) {
1161 VLOG_DBG("controller IP address changed from "IP_FMT
1162 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
1163 last_nonzero_ip = ip;
1165 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
1166 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
1168 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
1169 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
1172 /* Schedule next refresh.
1174 * If we have an IP address but not a MAC address, then refresh
1175 * quickly, since we probably will get a MAC address soon (via ARP).
1176 * Otherwise, we can afford to wait a little while. */
1177 next_refresh = now + (!ip || have_mac ? 10 : 1);
1179 return !eth_addr_is_zero(mac) ? mac : NULL;
1183 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
1184 struct in_band_data *in_band)
1186 const uint8_t *mac = get_controller_mac(in_band);
1187 return mac && eth_addr_equals(mac, dl_addr);
1191 in_band_learn_mac(struct in_band_data *in_band,
1192 uint16_t in_port, const uint8_t src_mac[ETH_ADDR_LEN])
1194 if (mac_learning_learn(in_band->ml, src_mac, in_port)) {
1195 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
1196 ETH_ADDR_ARGS(src_mac), in_port);
1201 in_band_local_packet_cb(struct relay *r, void *in_band_)
1203 struct in_band_data *in_band = in_band_;
1204 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
1205 struct ofp_packet_in *opi;
1206 struct eth_header *eth;
1207 struct buffer payload;
1212 if (!get_ofp_packet_eth_header(r, &opi, ð)) {
1215 in_port = ntohs(opi->in_port);
1217 /* Deal with local stuff. */
1218 if (in_port == OFPP_LOCAL) {
1219 /* Sent by secure channel. */
1220 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1221 } else if (eth_addr_equals(eth->eth_dst, in_band->mac)) {
1222 /* Sent to secure channel. */
1223 out_port = OFPP_LOCAL;
1224 in_band_learn_mac(in_band, in_port, eth->eth_src);
1225 } else if (eth->eth_type == htons(ETH_TYPE_ARP)
1226 && eth_addr_is_broadcast(eth->eth_dst)
1227 && is_controller_mac(eth->eth_src, in_band)) {
1228 /* ARP sent by controller. */
1229 out_port = OFPP_FLOOD;
1230 } else if (is_controller_mac(eth->eth_dst, in_band)
1231 || is_controller_mac(eth->eth_src, in_band)) {
1232 /* Traffic to or from controller. Switch it by hand. */
1233 in_band_learn_mac(in_band, in_port, eth->eth_src);
1234 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1236 const uint8_t *controller_mac;
1237 controller_mac = get_controller_mac(in_band);
1238 if (eth->eth_type == htons(ETH_TYPE_ARP)
1239 && eth_addr_is_broadcast(eth->eth_dst)
1240 && is_controller_mac(eth->eth_src, in_band)) {
1241 /* ARP sent by controller. */
1242 out_port = OFPP_FLOOD;
1243 } else if (is_controller_mac(eth->eth_dst, in_band)
1244 && in_port == mac_learning_lookup(in_band->ml,
1246 /* Drop controller traffic that arrives on the controller port. */
1253 get_ofp_packet_payload(opi, &payload);
1254 flow_extract(&payload, in_port, &flow);
1255 if (in_port == out_port) {
1256 /* The input and output port match. Set up a flow to drop packets. */
1257 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
1258 in_band->s->max_idle, 0));
1259 } else if (out_port != OFPP_FLOOD) {
1260 /* The output port is known, so add a new flow. */
1261 queue_tx(rc, in_band,
1262 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
1263 out_port, in_band->s->max_idle));
1265 /* If the switch didn't buffer the packet, we need to send a copy. */
1266 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1267 queue_tx(rc, in_band,
1268 make_unbuffered_packet_out(&payload, in_port, out_port));
1271 /* We don't know that MAC. Send along the packet without setting up a
1274 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1275 b = make_unbuffered_packet_out(&payload, in_port, out_port);
1277 b = make_buffered_packet_out(ntohl(opi->buffer_id),
1280 queue_tx(rc, in_band, b);
1286 in_band_status_cb(struct status_reply *sr, void *in_band_)
1288 struct in_band_data *in_band = in_band_;
1289 struct in_addr local_ip;
1290 uint32_t controller_ip;
1291 const uint8_t *controller_mac;
1293 if (netdev_get_in4(in_band->of_device, &local_ip)) {
1294 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
1296 status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
1297 ETH_ADDR_ARGS(in_band->mac));
1299 controller_ip = rconn_get_ip(in_band->controller);
1300 if (controller_ip) {
1301 status_reply_put(sr, "controller-ip="IP_FMT,
1302 IP_ARGS(&controller_ip));
1304 controller_mac = get_controller_mac(in_band);
1305 if (controller_mac) {
1306 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
1307 ETH_ADDR_ARGS(controller_mac));
1312 get_ofp_packet_payload(struct ofp_packet_in *opi, struct buffer *payload)
1314 payload->data = opi->data;
1315 payload->size = ntohs(opi->header.length) - offsetof(struct ofp_packet_in,
1320 in_band_hook_create(const struct settings *s, struct switch_status *ss,
1321 struct rconn *remote)
1323 struct in_band_data *in_band;
1326 in_band = xcalloc(1, sizeof *in_band);
1328 in_band->ml = mac_learning_create();
1329 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE,
1330 &in_band->of_device);
1332 fatal(retval, "Could not open %s device", s->of_name);
1334 memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device),
1336 in_band->controller = remote;
1337 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
1338 return make_hook(in_band_local_packet_cb, NULL, NULL, NULL, in_band);
1341 /* Fail open support. */
1343 struct fail_open_data {
1344 const struct settings *s;
1345 struct rconn *local_rconn;
1346 struct rconn *remote_rconn;
1347 struct lswitch *lswitch;
1348 int last_disconn_secs;
1349 time_t boot_deadline;
1352 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
1354 fail_open_periodic_cb(void *fail_open_)
1356 struct fail_open_data *fail_open = fail_open_;
1360 if (time_now() < fail_open->boot_deadline) {
1363 disconn_secs = rconn_disconnected_duration(fail_open->remote_rconn);
1364 open = disconn_secs >= fail_open->s->probe_interval * 3;
1365 if (open != (fail_open->lswitch != NULL)) {
1367 VLOG_WARN("No longer in fail-open mode");
1368 lswitch_destroy(fail_open->lswitch);
1369 fail_open->lswitch = NULL;
1371 VLOG_WARN("Could not connect to controller for %d seconds, "
1372 "failing open", disconn_secs);
1373 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
1374 fail_open->s->max_idle);
1375 fail_open->last_disconn_secs = disconn_secs;
1377 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
1378 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
1379 "from controller", disconn_secs);
1380 fail_open->last_disconn_secs = disconn_secs;
1385 fail_open_local_packet_cb(struct relay *r, void *fail_open_)
1387 struct fail_open_data *fail_open = fail_open_;
1388 if (!fail_open->lswitch) {
1391 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
1392 r->halves[HALF_LOCAL].rxbuf);
1393 rconn_run(fail_open->local_rconn);
1399 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
1401 struct fail_open_data *fail_open = fail_open_;
1402 const struct settings *s = fail_open->s;
1403 int trigger_duration = s->probe_interval * 3;
1404 int cur_duration = rconn_disconnected_duration(fail_open->remote_rconn);
1406 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
1407 status_reply_put(sr, "current-duration=%d", cur_duration);
1408 status_reply_put(sr, "triggered=%s",
1409 cur_duration >= trigger_duration ? "true" : "false");
1410 status_reply_put(sr, "max-idle=%d", s->max_idle);
1414 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
1415 struct rconn *local_rconn, struct rconn *remote_rconn)
1417 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
1419 fail_open->local_rconn = local_rconn;
1420 fail_open->remote_rconn = remote_rconn;
1421 fail_open->lswitch = NULL;
1422 fail_open->boot_deadline = time_now() + s->probe_interval * 3;
1423 fail_open->boot_deadline += STP_EXTRA_BOOT_TIME;
1424 switch_status_register_category(ss, "fail-open",
1425 fail_open_status_cb, fail_open);
1426 return make_hook(fail_open_local_packet_cb, NULL,
1427 fail_open_periodic_cb, NULL, fail_open);
1430 struct rate_limiter {
1431 const struct settings *s;
1432 struct rconn *remote_rconn;
1434 /* One queue per physical port. */
1435 struct queue queues[OFPP_MAX];
1436 int n_queued; /* Sum over queues[*].n. */
1437 int next_tx_port; /* Next port to check in round-robin. */
1441 * It costs 1000 tokens to send a single packet_in message. A single token
1442 * per message would be more straightforward, but this choice lets us avoid
1443 * round-off error in refill_bucket()'s calculation of how many tokens to
1444 * add to the bucket, since no division step is needed. */
1445 long long int last_fill; /* Time at which we last added tokens. */
1446 int tokens; /* Current number of tokens. */
1448 /* Transmission queue. */
1449 int n_txq; /* No. of packets waiting in rconn for tx. */
1451 /* Statistics reporting. */
1452 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
1453 unsigned long long n_limited; /* # queued for rate limiting. */
1454 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
1455 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
1458 /* Drop a packet from the longest queue in 'rl'. */
1460 drop_packet(struct rate_limiter *rl)
1462 struct queue *longest; /* Queue currently selected as longest. */
1463 int n_longest; /* # of queues of same length as 'longest'. */
1466 longest = &rl->queues[0];
1468 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
1469 if (longest->n < q->n) {
1472 } else if (longest->n == q->n) {
1475 /* Randomly select one of the longest queues, with a uniform
1476 * distribution (Knuth algorithm 3.4.2R). */
1477 if (!random_range(n_longest)) {
1483 /* FIXME: do we want to pop the tail instead? */
1484 buffer_delete(queue_pop_head(longest));
1488 /* Remove and return the next packet to transmit (in round-robin order). */
1489 static struct buffer *
1490 dequeue_packet(struct rate_limiter *rl)
1494 for (i = 0; i < OFPP_MAX; i++) {
1495 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
1496 struct queue *q = &rl->queues[port];
1498 rl->next_tx_port = (port + 1) % OFPP_MAX;
1500 return queue_pop_head(q);
1506 /* Add tokens to the bucket based on elapsed time. */
1508 refill_bucket(struct rate_limiter *rl)
1510 const struct settings *s = rl->s;
1511 long long int now = time_msec();
1512 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
1513 if (tokens >= 1000) {
1514 rl->last_fill = now;
1515 rl->tokens = MIN(tokens, s->burst_limit * 1000);
1519 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
1520 * true if successful, false otherwise. (In the latter case no tokens are
1523 get_token(struct rate_limiter *rl)
1525 if (rl->tokens >= 1000) {
1534 rate_limit_local_packet_cb(struct relay *r, void *rl_)
1536 struct rate_limiter *rl = rl_;
1537 const struct settings *s = rl->s;
1538 struct ofp_packet_in *opi;
1540 opi = get_ofp_packet_in(r);
1545 if (!rl->n_queued && get_token(rl)) {
1546 /* In the common case where we are not constrained by the rate limit,
1547 * let the packet take the normal path. */
1551 /* Otherwise queue it up for the periodic callback to drain out. */
1552 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
1553 int port = ntohs(opi->in_port) % OFPP_MAX;
1554 if (rl->n_queued >= s->burst_limit) {
1557 queue_push_tail(&rl->queues[port], buffer_clone(msg));
1565 rate_limit_status_cb(struct status_reply *sr, void *rl_)
1567 struct rate_limiter *rl = rl_;
1569 status_reply_put(sr, "normal=%llu", rl->n_normal);
1570 status_reply_put(sr, "limited=%llu", rl->n_limited);
1571 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
1572 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
1576 rate_limit_periodic_cb(void *rl_)
1578 struct rate_limiter *rl = rl_;
1581 /* Drain some packets out of the bucket if possible, but limit the number
1582 * of iterations to allow other code to get work done too. */
1584 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
1585 /* Use a small, arbitrary limit for the amount of queuing to do here,
1586 * because the TCP connection is responsible for buffering and there is
1587 * no point in trying to transmit faster than the TCP connection can
1589 struct buffer *b = dequeue_packet(rl);
1590 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
1597 rate_limit_wait_cb(void *rl_)
1599 struct rate_limiter *rl = rl_;
1601 if (rl->tokens >= 1000) {
1602 /* We can transmit more packets as soon as we're called again. */
1603 poll_immediate_wake();
1605 /* We have to wait for the bucket to re-fill. We could calculate
1606 * the exact amount of time here for increased smoothness. */
1607 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
1613 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
1614 struct rconn *local, struct rconn *remote)
1616 struct rate_limiter *rl;
1619 rl = xcalloc(1, sizeof *rl);
1621 rl->remote_rconn = remote;
1622 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
1623 queue_init(&rl->queues[i]);
1625 rl->last_fill = time_msec();
1626 rl->tokens = s->rate_limit * 100;
1627 switch_status_register_category(ss, "rate-limit",
1628 rate_limit_status_cb, rl);
1629 return make_hook(rate_limit_local_packet_cb, NULL, rate_limit_periodic_cb,
1630 rate_limit_wait_cb, rl);
1633 /* OFPST_SWITCH statistics. */
1635 struct switch_status_category {
1637 void (*cb)(struct status_reply *, void *aux);
1641 struct switch_status {
1642 const struct settings *s;
1644 struct switch_status_category categories[8];
1648 struct status_reply {
1649 struct switch_status_category *category;
1655 switch_status_remote_packet_cb(struct relay *r, void *ss_)
1657 struct switch_status *ss = ss_;
1658 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
1659 struct buffer *msg = r->halves[HALF_REMOTE].rxbuf;
1660 struct switch_status_category *c;
1661 struct nicira_header *request;
1662 struct nicira_header *reply;
1663 struct status_reply sr;
1667 if (msg->size < sizeof(struct nicira_header)) {
1670 request = msg->data;
1671 if (request->header.type != OFPT_VENDOR
1672 || request->vendor_id != htonl(NX_VENDOR_ID)
1673 || request->subtype != htonl(NXT_STATUS_REQUEST)) {
1677 sr.request.string = (void *) (request + 1);
1678 sr.request.length = msg->size - sizeof *request;
1679 ds_init(&sr.output);
1680 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
1681 if (!memcmp(c->name, sr.request.string,
1682 MIN(strlen(c->name), sr.request.length))) {
1687 reply = make_openflow_xid(sizeof *reply + sr.output.length,
1688 OFPT_VENDOR, request->header.xid, &b);
1689 reply->vendor_id = htonl(NX_VENDOR_ID);
1690 reply->subtype = htonl(NXT_STATUS_REPLY);
1691 memcpy(reply + 1, sr.output.string, sr.output.length);
1692 retval = rconn_send(rc, b, NULL);
1693 if (retval && retval != EAGAIN) {
1694 VLOG_WARN("send failed (%s)", strerror(retval));
1696 ds_destroy(&sr.output);
1701 rconn_status_cb(struct status_reply *sr, void *rconn_)
1703 struct rconn *rconn = rconn_;
1704 time_t now = time_now();
1706 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
1707 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
1708 status_reply_put(sr, "backoff=%d", rconn_get_backoff(rconn));
1709 status_reply_put(sr, "is-connected=%s",
1710 rconn_is_connected(rconn) ? "true" : "false");
1711 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
1712 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
1713 status_reply_put(sr, "attempted-connections=%u",
1714 rconn_get_attempted_connections(rconn));
1715 status_reply_put(sr, "successful-connections=%u",
1716 rconn_get_successful_connections(rconn));
1717 status_reply_put(sr, "last-connection=%ld",
1718 (long int) (now - rconn_get_last_connection(rconn)));
1719 status_reply_put(sr, "time-connected=%lu",
1720 rconn_get_total_time_connected(rconn));
1721 status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(rconn));
1725 config_status_cb(struct status_reply *sr, void *s_)
1727 const struct settings *s = s_;
1730 for (i = 0; i < s->n_listeners; i++) {
1731 status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
1733 if (s->probe_interval) {
1734 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
1736 if (s->max_backoff) {
1737 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
1742 switch_status_cb(struct status_reply *sr, void *ss_)
1744 struct switch_status *ss = ss_;
1745 time_t now = time_now();
1747 status_reply_put(sr, "now=%ld", (long int) now);
1748 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
1749 status_reply_put(sr, "pid=%ld", (long int) getpid());
1753 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
1755 struct switch_status *ss = xcalloc(1, sizeof *ss);
1757 ss->booted = time_now();
1758 switch_status_register_category(ss, "config",
1759 config_status_cb, (void *) s);
1760 switch_status_register_category(ss, "switch", switch_status_cb, ss);
1762 return make_hook(NULL, switch_status_remote_packet_cb, NULL, NULL, ss);
1766 switch_status_register_category(struct switch_status *ss,
1767 const char *category,
1768 void (*cb)(struct status_reply *,
1772 struct switch_status_category *c;
1773 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
1774 c = &ss->categories[ss->n_categories++];
1777 c->name = xstrdup(category);
1781 status_reply_put(struct status_reply *sr, const char *content, ...)
1783 size_t old_length = sr->output.length;
1787 /* Append the status reply to the output. */
1788 ds_put_format(&sr->output, "%s.", sr->category->name);
1789 va_start(args, content);
1790 ds_put_format_valist(&sr->output, content, args);
1792 if (ds_last(&sr->output) != '\n') {
1793 ds_put_char(&sr->output, '\n');
1796 /* Drop what we just added if it doesn't match the request. */
1797 added = sr->output.length - old_length;
1798 if (added < sr->request.length
1799 || memcmp(&sr->output.string[old_length],
1800 sr->request.string, sr->request.length)) {
1801 ds_truncate(&sr->output, old_length);
1806 /* Controller discovery. */
1810 const struct settings *s;
1811 struct dhclient *dhcp;
1816 discovery_status_cb(struct status_reply *sr, void *d_)
1818 struct discovery *d = d_;
1820 status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re);
1821 status_reply_put(sr, "n-changes=%d", d->n_changes);
1822 status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp));
1823 status_reply_put(sr, "state-elapsed=%u",
1824 dhclient_get_state_elapsed(d->dhcp));
1825 if (dhclient_is_bound(d->dhcp)) {
1826 uint32_t ip = dhclient_get_ip(d->dhcp);
1827 uint32_t netmask = dhclient_get_netmask(d->dhcp);
1828 uint32_t router = dhclient_get_router(d->dhcp);
1830 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
1831 uint32_t dns_server;
1835 status_reply_put(sr, "ip="IP_FMT, IP_ARGS(&ip));
1836 status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask));
1838 status_reply_put(sr, "router="IP_FMT, IP_ARGS(&router));
1841 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i, &dns_server);
1843 status_reply_put(sr, "dns%d="IP_FMT, i, IP_ARGS(&dns_server));
1846 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
1848 status_reply_put(sr, "domain=%s", domain_name);
1852 status_reply_put(sr, "lease-remaining=%u",
1853 dhclient_get_lease_remaining(d->dhcp));
1857 static struct discovery *
1858 discovery_init(const struct settings *s, struct switch_status *ss)
1860 struct netdev *netdev;
1861 struct discovery *d;
1862 struct dhclient *dhcp;
1865 /* Bring ofX network device up. */
1866 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1868 fatal(retval, "Could not open %s device", s->of_name);
1870 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
1872 fatal(retval, "Could not bring %s device up", s->of_name);
1874 netdev_close(netdev);
1876 /* Initialize DHCP client. */
1877 retval = dhclient_create(s->of_name, modify_dhcp_request,
1878 validate_dhcp_offer, (void *) s, &dhcp);
1880 fatal(retval, "Failed to initialize DHCP client");
1882 dhclient_init(dhcp, 0);
1884 d = xmalloc(sizeof *d);
1889 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
1895 discovery_question_connectivity(struct discovery *d)
1897 dhclient_force_renew(d->dhcp, 15);
1901 discovery_run(struct discovery *d, char **controller_name)
1903 dhclient_run(d->dhcp);
1904 if (!dhclient_changed(d->dhcp)) {
1908 dhclient_configure_netdev(d->dhcp);
1909 if (d->s->update_resolv_conf) {
1910 dhclient_update_resolv_conf(d->dhcp);
1913 if (dhclient_is_bound(d->dhcp)) {
1914 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
1915 DHCP_CODE_OFP_CONTROLLER_VCONN);
1916 VLOG_WARN("%s: discovered controller", *controller_name);
1919 *controller_name = NULL;
1921 VLOG_WARN("discovered controller no longer available");
1929 discovery_wait(struct discovery *d)
1931 dhclient_wait(d->dhcp);
1935 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
1937 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
1941 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
1943 const struct settings *s = s_;
1947 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
1949 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
1952 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
1954 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
1955 s->accept_controller_re);
1961 /* User interface. */
1964 parse_options(int argc, char *argv[], struct settings *s)
1967 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
1969 OPT_INACTIVITY_PROBE,
1975 static struct option long_options[] = {
1976 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
1977 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
1978 {"fail", required_argument, 0, 'F'},
1979 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
1980 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
1981 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
1982 {"listen", required_argument, 0, 'l'},
1983 {"monitor", required_argument, 0, 'm'},
1984 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
1985 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
1986 {"detach", no_argument, 0, 'D'},
1987 {"force", no_argument, 0, 'f'},
1988 {"pidfile", optional_argument, 0, 'P'},
1989 {"verbose", optional_argument, 0, 'v'},
1990 {"help", no_argument, 0, 'h'},
1991 {"version", no_argument, 0, 'V'},
1992 VCONN_SSL_LONG_OPTIONS
1995 char *short_options = long_options_to_short_options(long_options);
1996 char *accept_re = NULL;
1999 /* Set defaults that we can figure out before parsing options. */
2001 s->monitor_name = NULL;
2002 s->fail_mode = FAIL_OPEN;
2004 s->probe_interval = 15;
2005 s->max_backoff = 15;
2006 s->update_resolv_conf = true;
2012 c = getopt_long(argc, argv, short_options, long_options, NULL);
2018 case OPT_ACCEPT_VCONN:
2019 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
2022 case OPT_NO_RESOLV_CONF:
2023 s->update_resolv_conf = false;
2027 if (!strcmp(optarg, "open")) {
2028 s->fail_mode = FAIL_OPEN;
2029 } else if (!strcmp(optarg, "closed")) {
2030 s->fail_mode = FAIL_CLOSED;
2033 "-f or --fail argument must be \"open\" or \"closed\"");
2037 case OPT_INACTIVITY_PROBE:
2038 s->probe_interval = atoi(optarg);
2039 if (s->probe_interval < 5) {
2040 fatal(0, "--inactivity-probe argument must be at least 5");
2045 if (!strcmp(optarg, "permanent")) {
2046 s->max_idle = OFP_FLOW_PERMANENT;
2048 s->max_idle = atoi(optarg);
2049 if (s->max_idle < 1 || s->max_idle > 65535) {
2050 fatal(0, "--max-idle argument must be between 1 and "
2051 "65535 or the word 'permanent'");
2056 case OPT_MAX_BACKOFF:
2057 s->max_backoff = atoi(optarg);
2058 if (s->max_backoff < 1) {
2059 fatal(0, "--max-backoff argument must be at least 1");
2060 } else if (s->max_backoff > 3600) {
2061 s->max_backoff = 3600;
2065 case OPT_RATE_LIMIT:
2067 s->rate_limit = atoi(optarg);
2068 if (s->rate_limit < 1) {
2069 fatal(0, "--rate-limit argument must be at least 1");
2072 s->rate_limit = 1000;
2076 case OPT_BURST_LIMIT:
2077 s->burst_limit = atoi(optarg);
2078 if (s->burst_limit < 1) {
2079 fatal(0, "--burst-limit argument must be at least 1");
2088 set_pidfile(optarg);
2092 ignore_existing_pidfile();
2096 if (s->n_listeners >= MAX_MGMT) {
2097 fatal(0, "-l or --listen may be specified at most %d times",
2100 s->listener_names[s->n_listeners++] = optarg;
2104 if (s->monitor_name) {
2105 fatal(0, "-m or --monitor may only be specified once");
2107 s->monitor_name = optarg;
2114 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
2118 vlog_set_verbosity(optarg);
2121 VCONN_SSL_OPTION_HANDLERS
2130 free(short_options);
2134 if (argc < 1 || argc > 2) {
2135 fatal(0, "need one or two non-option arguments; use --help for usage");
2138 /* Local and remote vconns. */
2139 s->nl_name = argv[0];
2140 if (strncmp(s->nl_name, "nl:", 3)
2141 || strlen(s->nl_name) < 4
2142 || s->nl_name[strspn(s->nl_name + 3, "0123456789") + 3]) {
2143 fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"", s->nl_name);
2145 s->of_name = xasprintf("of%s", s->nl_name + 3);
2146 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
2148 /* Set accept_controller_regex. */
2150 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
2152 retval = regcomp(&s->accept_controller_regex, accept_re,
2153 REG_NOSUB | REG_EXTENDED);
2155 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
2156 char *buffer = xmalloc(length);
2157 regerror(retval, &s->accept_controller_regex, buffer, length);
2158 fatal(0, "%s: %s", accept_re, buffer);
2160 s->accept_controller_re = accept_re;
2162 /* Mode of operation. */
2163 s->discovery = s->controller_name == NULL;
2167 enum netdev_flags flags;
2168 struct netdev *netdev;
2170 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
2172 fatal(retval, "Could not open %s device", s->of_name);
2175 retval = netdev_get_flags(netdev, &flags);
2177 fatal(retval, "Could not get flags for %s device", s->of_name);
2180 s->in_band = (flags & NETDEV_UP) != 0;
2181 if (s->in_band && netdev_get_in6(netdev, NULL)) {
2182 VLOG_WARN("Ignoring IPv6 address on %s device: IPv6 not supported",
2186 netdev_close(netdev);
2189 /* Rate limiting. */
2190 if (s->rate_limit) {
2191 if (s->rate_limit < 100) {
2192 VLOG_WARN("Rate limit set to unusually low value %d",
2195 if (!s->burst_limit) {
2196 s->burst_limit = s->rate_limit / 4;
2198 s->burst_limit = MAX(s->burst_limit, 1);
2199 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
2206 printf("%s: secure channel, a relay for OpenFlow messages.\n"
2207 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
2208 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
2209 "CONTROLLER is an active OpenFlow connection method; if it is\n"
2210 "omitted, then secchan performs controller discovery.\n",
2211 program_name, program_name);
2212 vconn_usage(true, true);
2213 printf("\nController discovery options:\n"
2214 " --accept-vconn=REGEX accept matching discovered controllers\n"
2215 " --no-resolv-conf do not update /etc/resolv.conf\n"
2216 "\nNetworking options:\n"
2217 " -F, --fail=open|closed when controller connection fails:\n"
2218 " closed: drop all packets\n"
2219 " open (default): act as learning switch\n"
2220 " --inactivity-probe=SECS time between inactivity probes\n"
2221 " --max-idle=SECS max idle for flows set up by secchan\n"
2222 " --max-backoff=SECS max time between controller connection\n"
2223 " attempts (default: 15 seconds)\n"
2224 " -l, --listen=METHOD allow management connections on METHOD\n"
2225 " (a passive OpenFlow connection method)\n"
2226 " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n"
2227 " (a passive OpenFlow connection method)\n"
2228 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
2229 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
2230 " --burst-limit=BURST limit on packet credit for idle time\n"
2231 "\nOther options:\n"
2232 " -D, --detach run in background as daemon\n"
2233 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
2234 " -f, --force with -P, start even if already running\n"
2235 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
2236 " -v, --verbose set maximum verbosity level\n"
2237 " -h, --help display this help message\n"
2238 " -V, --version display version information\n",