1 /* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
4 * We are making the OpenFlow specification and associated documentation
5 * (Software) available for public use and benefit with the expectation
6 * that others will use, modify and enhance the Software and contribute
7 * those enhancements back to the community. However, since we would
8 * like to make the Software available for broadest use, with as few
9 * restrictions as possible permission is hereby granted, free of
10 * charge, to any person obtaining a copy of this Software to deal in
11 * the Software under the copyrights without restriction, including
12 * without limitation the rights to use, copy, modify, merge, publish,
13 * distribute, sublicense, and/or sell copies of the Software, and to
14 * permit persons to whom the Software is furnished to do so, subject to
15 * the following conditions:
17 * The above copyright notice and this permission notice shall be
18 * included in all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
23 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
24 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
25 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
26 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 * The name and trademarks of copyright holder(s) may NOT be used in
30 * advertising or publicity pertaining to the Software or any
31 * derivatives without specific, written prior permission.
39 #include <netinet/in.h>
48 #include "command-line.h"
51 #include "dhcp-client.h"
53 #include "dynamic-string.h"
56 #include "learning-switch.h"
58 #include "mac-learning.h"
60 #include "nicira-ext.h"
64 #include "poll-loop.h"
69 #include "vconn-ssl.h"
71 #include "vlog-socket.h"
74 #define THIS_MODULE VLM_secchan
76 /* Behavior when the connection to the controller fails. */
78 FAIL_OPEN, /* Act as learning switch. */
79 FAIL_CLOSED /* Drop all packets. */
82 /* Maximum number of management connection listeners. */
85 /* Settings that may be configured by the user. */
87 /* Overall mode of operation. */
88 bool discovery; /* Discover the controller automatically? */
89 bool in_band; /* Connect to controller in-band? */
91 /* Related vconns and network devices. */
92 const char *nl_name; /* Local datapath (must be "nl:" vconn). */
93 char *of_name; /* ofX network device name. */
94 const char *controller_name; /* Controller (if not discovery mode). */
95 const char *listener_names[MAX_MGMT]; /* Listen for mgmt connections. */
96 size_t n_listeners; /* Number of mgmt connection listeners. */
97 const char *monitor_name; /* Listen for traffic monitor connections. */
99 /* Failure behavior. */
100 enum fail_mode fail_mode; /* Act as learning switch if no controller? */
101 int max_idle; /* Idle time for flows in fail-open mode. */
102 int probe_interval; /* # seconds idle before sending echo request. */
103 int max_backoff; /* Max # seconds between connection attempts. */
105 /* Packet-in rate-limiting. */
106 int rate_limit; /* Tokens added to bucket per second. */
107 int burst_limit; /* Maximum number token bucket size. */
109 /* Discovery behavior. */
110 regex_t accept_controller_regex; /* Controller vconns to accept. */
111 const char *accept_controller_re; /* String version of regex. */
112 bool update_resolv_conf; /* Update /etc/resolv.conf? */
117 struct ofpbuf *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 pvconn *open_passive_vconn(const char *name);
144 static struct vconn *accept_vconn(struct pvconn *pvconn);
146 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
148 static struct relay *relay_accept(const struct settings *, struct pvconn *);
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 ofpbuf *);
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 pvconn *monitor;
225 struct pvconn *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 ofp_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 ofp_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 pvconn_wait(listeners[i]);
356 pvconn_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 pvconn *
373 open_passive_vconn(const char *name)
375 struct pvconn *pvconn;
378 retval = pvconn_open(name, &pvconn);
379 if (retval && retval != EAGAIN) {
380 ofp_fatal(retval, "opening %s", name);
385 static struct vconn *
386 accept_vconn(struct pvconn *pvconn)
391 retval = pvconn_accept(pvconn, OFP_VERSION, &new);
392 if (retval && retval != EAGAIN) {
393 VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval));
399 make_hook(bool (*local_packet_cb)(struct relay *, void *aux),
400 bool (*remote_packet_cb)(struct relay *, void *aux),
401 void (*periodic_cb)(void *aux),
402 void (*wait_cb)(void *aux),
406 h.packet_cb[HALF_LOCAL] = local_packet_cb;
407 h.packet_cb[HALF_REMOTE] = remote_packet_cb;
408 h.periodic_cb = periodic_cb;
414 static struct ofp_packet_in *
415 get_ofp_packet_in(struct relay *r)
417 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
418 struct ofp_header *oh = msg->data;
419 if (oh->type == OFPT_PACKET_IN) {
420 if (msg->size >= offsetof (struct ofp_packet_in, data)) {
423 VLOG_WARN("packet too short (%zu bytes) for packet_in",
431 get_ofp_packet_eth_header(struct relay *r, struct ofp_packet_in **opip,
432 struct eth_header **ethp)
434 const int min_len = offsetof(struct ofp_packet_in, data) + ETH_HEADER_LEN;
435 struct ofp_packet_in *opi = get_ofp_packet_in(r);
436 if (opi && ntohs(opi->header.length) >= min_len) {
438 *ethp = (void *) opi->data;
445 /* OpenFlow message relaying. */
447 static struct relay *
448 relay_accept(const struct settings *s, struct pvconn *pvconn)
450 struct vconn *new_remote, *new_local;
451 char *nl_name_without_subscription;
452 struct rconn *r1, *r2;
455 new_remote = accept_vconn(pvconn);
460 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We
461 * only accept the former syntax in main().
463 * nl:123:0 opens a netlink connection to local datapath 123 without
464 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
466 nl_name_without_subscription = xasprintf("%s:0", s->nl_name);
467 retval = vconn_open(nl_name_without_subscription, OFP_VERSION, &new_local);
469 VLOG_ERR_RL(&vrl, "could not connect to %s (%s)",
470 nl_name_without_subscription, strerror(retval));
471 vconn_close(new_remote);
472 free(nl_name_without_subscription);
476 /* Create and return relay. */
477 r1 = rconn_create(0, 0);
478 rconn_connect_unreliably(r1, nl_name_without_subscription, new_local);
479 free(nl_name_without_subscription);
481 r2 = rconn_create(0, 0);
482 rconn_connect_unreliably(r2, "passive", new_remote);
484 return relay_create(r1, r2, true);
487 static struct relay *
488 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
490 struct relay *r = xcalloc(1, sizeof *r);
491 r->halves[HALF_LOCAL].rconn = local;
492 r->halves[HALF_REMOTE].rconn = remote;
493 r->is_mgmt_conn = is_mgmt_conn;
498 relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks)
503 for (i = 0; i < 2; i++) {
504 rconn_run(r->halves[i].rconn);
507 /* Limit the number of iterations to prevent other tasks from starving. */
508 for (iteration = 0; iteration < 50; iteration++) {
509 bool progress = false;
510 for (i = 0; i < 2; i++) {
511 struct half *this = &r->halves[i];
512 struct half *peer = &r->halves[!i];
515 this->rxbuf = rconn_recv(this->rconn);
516 if (this->rxbuf && (i == HALF_REMOTE || !r->is_mgmt_conn)) {
517 const struct hook *h;
518 for (h = hooks; h < &hooks[n_hooks]; h++) {
519 if (h->packet_cb[i] && h->packet_cb[i](r, h->aux)) {
520 ofpbuf_delete(this->rxbuf);
529 if (this->rxbuf && !this->n_txq) {
530 int retval = rconn_send(peer->rconn, this->rxbuf,
532 if (retval != EAGAIN) {
536 ofpbuf_delete(this->rxbuf);
547 if (r->is_mgmt_conn) {
548 for (i = 0; i < 2; i++) {
549 struct half *this = &r->halves[i];
550 if (!rconn_is_alive(this->rconn)) {
559 relay_wait(struct relay *r)
563 for (i = 0; i < 2; i++) {
564 struct half *this = &r->halves[i];
566 rconn_run_wait(this->rconn);
568 rconn_recv_wait(this->rconn);
574 relay_destroy(struct relay *r)
578 list_remove(&r->node);
579 for (i = 0; i < 2; i++) {
580 struct half *this = &r->halves[i];
581 rconn_destroy(this->rconn);
582 ofpbuf_delete(this->rxbuf);
587 /* Port status watcher. */
589 typedef void edit_port_cb_func(struct ofp_phy_port *port, void *aux);
590 typedef void port_changed_cb_func(uint16_t port_no,
591 const struct ofp_phy_port *old,
592 const struct ofp_phy_port *new,
595 struct port_watcher_cb {
596 edit_port_cb_func *edit_port;
597 port_changed_cb_func *port_changed;
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_port_changed_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 port_changed_cb_func *port_changed = pw->cbs[i].port_changed;
657 (port_changed)(port_no, old, new, pw->cbs[i].aux);
664 call_edit_port_callbacks(struct port_watcher *pw, struct ofp_phy_port *p)
667 for (i = 0; i < pw->n_cbs; i++) {
668 edit_port_cb_func *edit_port = pw->cbs[i].edit_port;
670 (edit_port)(p, pw->cbs[i].aux);
676 update_phy_port(struct port_watcher *pw, struct ofp_phy_port *opp,
677 uint8_t reason, bool seen[OFPP_MAX + 1])
679 struct ofp_phy_port *pw_opp;
680 struct ofp_phy_port old;
684 port_no = ntohs(opp->port_no);
685 idx = port_no_to_pw_idx(port_no);
694 pw_opp = &pw->ports[idx];
696 if (reason == OFPPR_DELETE) {
697 memset(pw_opp, 0, sizeof *pw_opp);
698 pw_opp->port_no = htons(OFPP_NONE);
699 } else if (reason == OFPPR_MOD || reason == OFPPR_ADD) {
701 sanitize_opp(pw_opp);
703 call_port_changed_callbacks(pw, port_no, &old, pw_opp);
707 port_watcher_local_packet_cb(struct relay *r, void *pw_)
709 struct port_watcher *pw = pw_;
710 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
711 struct ofp_header *oh = msg->data;
713 if (oh->type == OFPT_FEATURES_REPLY
714 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
715 struct ofp_switch_features *osf = msg->data;
716 bool seen[ARRAY_SIZE(pw->ports)];
720 pw->got_feature_reply = true;
722 /* Update each port included in the message. */
723 memset(seen, 0, sizeof seen);
724 n_ports = ((msg->size - offsetof(struct ofp_switch_features, ports))
725 / sizeof *osf->ports);
726 for (i = 0; i < n_ports; i++) {
727 struct ofp_phy_port *opp = &osf->ports[i];
728 call_edit_port_callbacks(pw, opp);
729 update_phy_port(pw, opp, OFPPR_MOD, seen);
732 /* Delete all the ports not included in the message. */
733 for (i = 0; i < ARRAY_SIZE(pw->ports); i++) {
735 update_phy_port(pw, &pw->ports[i], OFPPR_DELETE, NULL);
738 } else if (oh->type == OFPT_PORT_STATUS
739 && msg->size >= sizeof(struct ofp_port_status)) {
740 struct ofp_port_status *ops = msg->data;
741 call_edit_port_callbacks(pw, &ops->desc);
742 update_phy_port(pw, &ops->desc, ops->reason, NULL);
748 port_watcher_remote_packet_cb(struct relay *r, void *pw_)
750 struct port_watcher *pw = pw_;
751 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
752 struct ofp_header *oh = msg->data;
754 if (oh->type == OFPT_PORT_MOD
755 && msg->size >= sizeof(struct ofp_port_mod)) {
756 struct ofp_port_mod *opm = msg->data;
757 uint16_t port_no = ntohs(opm->desc.port_no);
758 int idx = port_no_to_pw_idx(port_no);
760 struct ofp_phy_port *pw_opp = &pw->ports[idx];
761 if (pw_opp->port_no != htons(OFPP_NONE)) {
762 struct ofp_phy_port old = *pw_opp;
763 pw_opp->flags = ((pw_opp->flags & ~opm->mask)
764 | (opm->desc.flags & opm->mask));
765 call_port_changed_callbacks(pw, port_no, &old, pw_opp);
773 port_watcher_periodic_cb(void *pw_)
775 struct port_watcher *pw = pw_;
777 if (!pw->got_feature_reply && time_now() >= pw->last_feature_request + 5) {
779 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
780 rconn_send_with_limit(pw->local_rconn, b, &pw->n_txq, 1);
781 pw->last_feature_request = time_now();
786 put_duplexes(struct ds *ds, const char *name, uint32_t features,
787 uint32_t hd_bit, uint32_t fd_bit)
789 if (features & (hd_bit | fd_bit)) {
790 ds_put_format(ds, " %s", name);
791 if (features & hd_bit) {
792 ds_put_cstr(ds, "(HD)");
794 if (features & fd_bit) {
795 ds_put_cstr(ds, "(FD)");
801 log_port_status(uint16_t port_no,
802 const struct ofp_phy_port *old,
803 const struct ofp_phy_port *new,
806 if (VLOG_IS_DBG_ENABLED()) {
807 bool was_enabled = old->port_no != htons(OFPP_NONE);
808 bool now_enabled = new->port_no != htons(OFPP_NONE);
809 uint32_t features = ntohl(new->features);
812 if (old->flags != new->flags && opp_differs(old, new) == 1) {
813 /* Don't care if only flags changed. */
818 ds_put_format(&ds, "\"%s\", "ETH_ADDR_FMT, new->name,
819 ETH_ADDR_ARGS(new->hw_addr));
820 if (ntohl(new->speed)) {
821 ds_put_format(&ds, ", speed %"PRIu32, ntohl(new->speed));
823 if (features & (OFPPF_10MB_HD | OFPPF_10MB_FD
824 | OFPPF_100MB_HD | OFPPF_100MB_FD
825 | OFPPF_1GB_HD | OFPPF_1GB_FD | OFPPF_10GB_FD)) {
826 ds_put_cstr(&ds, ", supports");
827 put_duplexes(&ds, "10M", features, OFPPF_10MB_HD, OFPPF_10MB_FD);
828 put_duplexes(&ds, "100M", features,
829 OFPPF_100MB_HD, OFPPF_100MB_FD);
830 put_duplexes(&ds, "1G", features, OFPPF_100MB_HD, OFPPF_100MB_FD);
831 if (features & OFPPF_10GB_FD) {
832 ds_put_cstr(&ds, " 10G");
835 if (was_enabled != now_enabled) {
837 VLOG_DBG("Port %d added: %s", port_no, ds_cstr(&ds));
839 VLOG_DBG("Port %d deleted", port_no);
842 VLOG_DBG("Port %d changed: %s", port_no, ds_cstr(&ds));
849 port_watcher_register_callback(struct port_watcher *pw,
850 edit_port_cb_func *edit_port,
851 port_changed_cb_func *port_changed,
854 assert(pw->n_cbs < ARRAY_SIZE(pw->cbs));
855 pw->cbs[pw->n_cbs].edit_port = edit_port;
856 pw->cbs[pw->n_cbs].port_changed = port_changed;
857 pw->cbs[pw->n_cbs].aux = aux;
862 port_watcher_get_flags(const struct port_watcher *pw, int port_no)
864 int idx = port_no_to_pw_idx(port_no);
865 return idx >= 0 ? ntohl(pw->ports[idx].flags) : 0;
869 port_watcher_set_flags(struct port_watcher *pw,
870 int port_no, uint32_t flags, uint32_t mask)
872 struct ofp_phy_port old;
873 struct ofp_phy_port *p;
874 struct ofp_port_mod *opm;
875 struct ofp_port_status *ops;
879 idx = port_no_to_pw_idx(port_no);
885 if (!((ntohl(p->flags) ^ flags) & mask)) {
890 /* Update our idea of the flags. */
891 p->flags = htonl((ntohl(p->flags) & ~mask) | (flags & mask));
892 call_port_changed_callbacks(pw, port_no, &old, p);
894 /* Change the flags in the datapath. */
895 opm = make_openflow(sizeof *opm, OFPT_PORT_MOD, &b);
896 opm->mask = htonl(mask);
898 rconn_send(pw->local_rconn, b, NULL);
900 /* Notify the controller that the flags changed. */
901 ops = make_openflow(sizeof *ops, OFPT_PORT_STATUS, &b);
902 ops->reason = OFPPR_MOD;
904 rconn_send(pw->remote_rconn, b, NULL);
908 port_watcher_is_ready(const struct port_watcher *pw)
910 return pw->got_feature_reply;
914 port_watcher_create(struct rconn *local_rconn, struct rconn *remote_rconn,
915 struct port_watcher **pwp)
917 struct port_watcher *pw;
920 pw = *pwp = xcalloc(1, sizeof *pw);
921 pw->local_rconn = local_rconn;
922 pw->remote_rconn = remote_rconn;
923 pw->last_feature_request = TIME_MIN;
924 for (i = 0; i < OFPP_MAX; i++) {
925 pw->ports[i].port_no = htons(OFPP_NONE);
927 port_watcher_register_callback(pw, NULL, log_port_status, NULL);
928 return make_hook(port_watcher_local_packet_cb,
929 port_watcher_remote_packet_cb,
930 port_watcher_periodic_cb, NULL, pw);
933 /* Spanning tree protocol. */
935 /* Extra time, in seconds, at boot before going into fail-open, to give the
936 * spanning tree protocol time to figure out the network layout. */
937 #define STP_EXTRA_BOOT_TIME 30
941 struct port_watcher *pw;
942 struct rconn *local_rconn;
943 struct rconn *remote_rconn;
944 uint8_t dpid[ETH_ADDR_LEN];
945 long long int last_tick_256ths;
950 stp_local_packet_cb(struct relay *r, void *stp_)
952 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
953 struct ofp_header *oh;
954 struct stp_data *stp = stp_;
955 struct ofp_packet_in *opi;
956 struct eth_header *eth;
957 struct llc_header *llc;
958 struct ofpbuf payload;
963 if (oh->type == OFPT_FEATURES_REPLY
964 && msg->size >= offsetof(struct ofp_switch_features, ports)) {
965 struct ofp_switch_features *osf = msg->data;
966 osf->capabilities |= htonl(OFPC_STP);
970 if (!get_ofp_packet_eth_header(r, &opi, ð)
971 || !eth_addr_equals(eth->eth_dst, stp_eth_addr)) {
975 port_no = ntohs(opi->in_port);
976 if (port_no >= STP_MAX_PORTS) {
977 /* STP only supports 255 ports. */
980 if (port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP) {
981 /* We're not doing STP on this port. */
985 if (opi->reason == OFPR_ACTION) {
986 /* The controller set up a flow for this, so we won't intercept it. */
990 get_ofp_packet_payload(opi, &payload);
991 flow_extract(&payload, port_no, &flow);
992 if (flow.dl_type != htons(OFP_DL_TYPE_NOT_ETH_TYPE)) {
993 VLOG_DBG("non-LLC frame received on STP multicast address");
996 llc = ofpbuf_at_assert(&payload, sizeof *eth, sizeof *llc);
997 if (llc->llc_dsap != STP_LLC_DSAP) {
998 VLOG_DBG("bad DSAP 0x%02"PRIx8" received on STP multicast address",
1003 /* Trim off padding on payload. */
1004 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1005 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1007 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1008 struct stp_port *p = stp_get_port(stp->stp, port_no);
1009 stp_received_bpdu(p, payload.data, payload.size);
1015 static long long int
1018 return time_msec() * 256 / 1000;
1022 stp_periodic_cb(void *stp_)
1024 struct stp_data *stp = stp_;
1025 long long int now_256ths = time_256ths();
1026 long long int elapsed_256ths = now_256ths - stp->last_tick_256ths;
1029 if (!port_watcher_is_ready(stp->pw)) {
1030 /* Can't start STP until we know port flags, because port flags can
1034 if (elapsed_256ths <= 0) {
1038 stp_tick(stp->stp, MIN(INT_MAX, elapsed_256ths));
1039 stp->last_tick_256ths = now_256ths;
1041 while (stp_get_changed_port(stp->stp, &p)) {
1042 int port_no = stp_port_no(p);
1043 enum stp_state state = stp_port_get_state(p);
1045 if (state != STP_DISABLED) {
1046 VLOG_WARN("STP: Port %d entered %s state",
1047 port_no, stp_state_name(state));
1049 if (!(port_watcher_get_flags(stp->pw, port_no) & OFPPFL_NO_STP)) {
1053 flags = OFPPFL_STP_LISTEN;
1056 flags = OFPPFL_STP_LEARN;
1059 case STP_FORWARDING:
1060 flags = OFPPFL_STP_FORWARD;
1063 flags = OFPPFL_STP_BLOCK;
1066 VLOG_DBG_RL(&vrl, "STP: Port %d has bad state %x",
1068 flags = OFPPFL_STP_FORWARD;
1071 if (!stp_forward_in_state(state)) {
1072 flags |= OFPPFL_NO_FLOOD;
1074 port_watcher_set_flags(stp->pw, port_no, flags,
1075 OFPPFL_STP_MASK | OFPPFL_NO_FLOOD);
1077 /* We don't own those flags. */
1083 stp_wait_cb(void *stp_ UNUSED)
1085 poll_timer_wait(1000);
1089 send_bpdu(const void *bpdu, size_t bpdu_size, int port_no, void *stp_)
1091 struct stp_data *stp = stp_;
1092 struct eth_header *eth;
1093 struct llc_header *llc;
1094 struct ofpbuf pkt, *opo;
1096 /* Packet skeleton. */
1097 ofpbuf_init(&pkt, ETH_HEADER_LEN + LLC_HEADER_LEN + bpdu_size);
1098 eth = ofpbuf_put_uninit(&pkt, sizeof *eth);
1099 llc = ofpbuf_put_uninit(&pkt, sizeof *llc);
1100 ofpbuf_put(&pkt, bpdu, bpdu_size);
1103 memcpy(eth->eth_dst, stp_eth_addr, ETH_ADDR_LEN);
1104 memcpy(eth->eth_src, stp->pw->ports[port_no].hw_addr, ETH_ADDR_LEN);
1105 eth->eth_type = htons(pkt.size - ETH_HEADER_LEN);
1108 llc->llc_dsap = STP_LLC_DSAP;
1109 llc->llc_ssap = STP_LLC_SSAP;
1110 llc->llc_cntl = STP_LLC_CNTL;
1112 opo = make_unbuffered_packet_out(&pkt, OFPP_NONE, port_no);
1113 ofpbuf_uninit(&pkt);
1114 rconn_send_with_limit(stp->local_rconn, opo, &stp->n_txq, OFPP_MAX);
1118 stp_is_port_supported(uint16_t port_no)
1120 /* STP only supports a maximum of 255 ports, one less than OpenFlow. We
1121 * don't support STP on OFPP_LOCAL, either. */
1122 return port_no < STP_MAX_PORTS;
1126 stp_edit_port_cb(struct ofp_phy_port *p, void *stp_ UNUSED)
1128 uint16_t port_no = ntohs(p->port_no);
1129 if (stp_is_port_supported(port_no)) {
1130 p->features |= htonl(OFPPF_STP);
1135 stp_port_changed_cb(uint16_t port_no,
1136 const struct ofp_phy_port *old,
1137 const struct ofp_phy_port *new,
1140 struct stp_data *stp = stp_;
1143 if (!stp_is_port_supported(port_no)) {
1147 p = stp_get_port(stp->stp, port_no);
1148 if (new->port_no == htons(OFPP_NONE)
1149 || new->flags & htonl(OFPPFL_NO_STP)) {
1150 stp_port_disable(p);
1153 stp_port_set_speed(p, new->speed);
1158 stp_hook_create(const struct settings *s, struct port_watcher *pw,
1159 struct rconn *local, struct rconn *remote)
1161 uint8_t dpid[ETH_ADDR_LEN];
1162 struct netdev *netdev;
1163 struct stp_data *stp;
1166 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1168 ofp_fatal(retval, "Could not open %s device", s->of_name);
1170 memcpy(dpid, netdev_get_etheraddr(netdev), ETH_ADDR_LEN);
1171 netdev_close(netdev);
1173 stp = xcalloc(1, sizeof *stp);
1174 stp->stp = stp_create("stp", eth_addr_to_uint64(dpid), send_bpdu, stp);
1176 memcpy(stp->dpid, dpid, ETH_ADDR_LEN);
1177 stp->local_rconn = local;
1178 stp->remote_rconn = remote;
1179 stp->last_tick_256ths = time_256ths();
1181 port_watcher_register_callback(pw, stp_edit_port_cb,
1182 stp_port_changed_cb, stp);
1183 return make_hook(stp_local_packet_cb, NULL,
1184 stp_periodic_cb, stp_wait_cb, stp);
1187 /* In-band control. */
1189 struct in_band_data {
1190 const struct settings *s;
1191 struct mac_learning *ml;
1192 struct netdev *of_device;
1193 struct rconn *controller;
1194 uint8_t mac[ETH_ADDR_LEN];
1199 queue_tx(struct rconn *rc, struct in_band_data *in_band, struct ofpbuf *b)
1201 rconn_send_with_limit(rc, b, &in_band->n_queued, 10);
1204 static const uint8_t *
1205 get_controller_mac(struct in_band_data *in_band)
1207 static uint32_t ip, last_nonzero_ip;
1208 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
1209 static time_t next_refresh = 0;
1211 uint32_t last_ip = ip;
1213 time_t now = time_now();
1215 ip = rconn_get_ip(in_band->controller);
1216 if (last_ip != ip || !next_refresh || now >= next_refresh) {
1219 /* Look up MAC address. */
1220 memset(mac, 0, sizeof mac);
1222 int retval = netdev_arp_lookup(in_band->of_device, ip, mac);
1224 VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s",
1225 IP_ARGS(&ip), strerror(retval));
1228 have_mac = !eth_addr_is_zero(mac);
1230 /* Log changes in IP, MAC addresses. */
1231 if (ip && ip != last_nonzero_ip) {
1232 VLOG_DBG("controller IP address changed from "IP_FMT
1233 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
1234 last_nonzero_ip = ip;
1236 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
1237 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
1239 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
1240 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
1243 /* Schedule next refresh.
1245 * If we have an IP address but not a MAC address, then refresh
1246 * quickly, since we probably will get a MAC address soon (via ARP).
1247 * Otherwise, we can afford to wait a little while. */
1248 next_refresh = now + (!ip || have_mac ? 10 : 1);
1250 return !eth_addr_is_zero(mac) ? mac : NULL;
1254 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
1255 struct in_band_data *in_band)
1257 const uint8_t *mac = get_controller_mac(in_band);
1258 return mac && eth_addr_equals(mac, dl_addr);
1262 in_band_learn_mac(struct in_band_data *in_band,
1263 uint16_t in_port, const uint8_t src_mac[ETH_ADDR_LEN])
1265 if (mac_learning_learn(in_band->ml, src_mac, in_port)) {
1266 VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
1267 ETH_ADDR_ARGS(src_mac), in_port);
1272 in_band_local_packet_cb(struct relay *r, void *in_band_)
1274 struct in_band_data *in_band = in_band_;
1275 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
1276 struct ofp_packet_in *opi;
1277 struct eth_header *eth;
1278 struct ofpbuf payload;
1283 if (!get_ofp_packet_eth_header(r, &opi, ð)) {
1286 in_port = ntohs(opi->in_port);
1288 /* Deal with local stuff. */
1289 if (in_port == OFPP_LOCAL) {
1290 /* Sent by secure channel. */
1291 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1292 } else if (eth_addr_equals(eth->eth_dst, in_band->mac)) {
1293 /* Sent to secure channel. */
1294 out_port = OFPP_LOCAL;
1295 in_band_learn_mac(in_band, in_port, eth->eth_src);
1296 } else if (eth->eth_type == htons(ETH_TYPE_ARP)
1297 && eth_addr_is_broadcast(eth->eth_dst)
1298 && is_controller_mac(eth->eth_src, in_band)) {
1299 /* ARP sent by controller. */
1300 out_port = OFPP_FLOOD;
1301 } else if (is_controller_mac(eth->eth_dst, in_band)
1302 || is_controller_mac(eth->eth_src, in_band)) {
1303 /* Traffic to or from controller. Switch it by hand. */
1304 in_band_learn_mac(in_band, in_port, eth->eth_src);
1305 out_port = mac_learning_lookup(in_band->ml, eth->eth_dst);
1307 const uint8_t *controller_mac;
1308 controller_mac = get_controller_mac(in_band);
1309 if (eth->eth_type == htons(ETH_TYPE_ARP)
1310 && eth_addr_is_broadcast(eth->eth_dst)
1311 && is_controller_mac(eth->eth_src, in_band)) {
1312 /* ARP sent by controller. */
1313 out_port = OFPP_FLOOD;
1314 } else if (is_controller_mac(eth->eth_dst, in_band)
1315 && in_port == mac_learning_lookup(in_band->ml,
1317 /* Drop controller traffic that arrives on the controller port. */
1324 get_ofp_packet_payload(opi, &payload);
1325 flow_extract(&payload, in_port, &flow);
1326 if (in_port == out_port) {
1327 /* The input and output port match. Set up a flow to drop packets. */
1328 queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id),
1329 in_band->s->max_idle, 0));
1330 } else if (out_port != OFPP_FLOOD) {
1331 /* The output port is known, so add a new flow. */
1332 queue_tx(rc, in_band,
1333 make_add_simple_flow(&flow, ntohl(opi->buffer_id),
1334 out_port, in_band->s->max_idle));
1336 /* If the switch didn't buffer the packet, we need to send a copy. */
1337 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1338 queue_tx(rc, in_band,
1339 make_unbuffered_packet_out(&payload, in_port, out_port));
1342 /* We don't know that MAC. Send along the packet without setting up a
1345 if (ntohl(opi->buffer_id) == UINT32_MAX) {
1346 b = make_unbuffered_packet_out(&payload, in_port, out_port);
1348 b = make_buffered_packet_out(ntohl(opi->buffer_id),
1351 queue_tx(rc, in_band, b);
1357 in_band_status_cb(struct status_reply *sr, void *in_band_)
1359 struct in_band_data *in_band = in_band_;
1360 struct in_addr local_ip;
1361 uint32_t controller_ip;
1362 const uint8_t *controller_mac;
1364 if (netdev_get_in4(in_band->of_device, &local_ip)) {
1365 status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
1367 status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
1368 ETH_ADDR_ARGS(in_band->mac));
1370 controller_ip = rconn_get_ip(in_band->controller);
1371 if (controller_ip) {
1372 status_reply_put(sr, "controller-ip="IP_FMT,
1373 IP_ARGS(&controller_ip));
1375 controller_mac = get_controller_mac(in_band);
1376 if (controller_mac) {
1377 status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
1378 ETH_ADDR_ARGS(controller_mac));
1383 get_ofp_packet_payload(struct ofp_packet_in *opi, struct ofpbuf *payload)
1385 payload->data = opi->data;
1386 payload->size = ntohs(opi->header.length) - offsetof(struct ofp_packet_in,
1391 in_band_hook_create(const struct settings *s, struct switch_status *ss,
1392 struct rconn *remote)
1394 struct in_band_data *in_band;
1397 in_band = xcalloc(1, sizeof *in_band);
1399 in_band->ml = mac_learning_create();
1400 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE,
1401 &in_band->of_device);
1403 ofp_fatal(retval, "Could not open %s device", s->of_name);
1405 memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device),
1407 in_band->controller = remote;
1408 switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
1409 return make_hook(in_band_local_packet_cb, NULL, NULL, NULL, in_band);
1412 /* Fail open support. */
1414 struct fail_open_data {
1415 const struct settings *s;
1416 struct rconn *local_rconn;
1417 struct rconn *remote_rconn;
1418 struct lswitch *lswitch;
1419 int last_disconn_secs;
1420 time_t boot_deadline;
1423 /* Causes 'r' to enter or leave fail-open mode, if appropriate. */
1425 fail_open_periodic_cb(void *fail_open_)
1427 struct fail_open_data *fail_open = fail_open_;
1431 if (time_now() < fail_open->boot_deadline) {
1434 disconn_secs = rconn_disconnected_duration(fail_open->remote_rconn);
1435 open = disconn_secs >= fail_open->s->probe_interval * 3;
1436 if (open != (fail_open->lswitch != NULL)) {
1438 VLOG_WARN("No longer in fail-open mode");
1439 lswitch_destroy(fail_open->lswitch);
1440 fail_open->lswitch = NULL;
1442 VLOG_WARN("Could not connect to controller for %d seconds, "
1443 "failing open", disconn_secs);
1444 fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
1445 fail_open->s->max_idle);
1446 fail_open->last_disconn_secs = disconn_secs;
1448 } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
1449 VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
1450 "from controller", disconn_secs);
1451 fail_open->last_disconn_secs = disconn_secs;
1456 fail_open_local_packet_cb(struct relay *r, void *fail_open_)
1458 struct fail_open_data *fail_open = fail_open_;
1459 if (!fail_open->lswitch) {
1462 lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
1463 r->halves[HALF_LOCAL].rxbuf);
1464 rconn_run(fail_open->local_rconn);
1470 fail_open_status_cb(struct status_reply *sr, void *fail_open_)
1472 struct fail_open_data *fail_open = fail_open_;
1473 const struct settings *s = fail_open->s;
1474 int trigger_duration = s->probe_interval * 3;
1475 int cur_duration = rconn_disconnected_duration(fail_open->remote_rconn);
1477 status_reply_put(sr, "trigger-duration=%d", trigger_duration);
1478 status_reply_put(sr, "current-duration=%d", cur_duration);
1479 status_reply_put(sr, "triggered=%s",
1480 cur_duration >= trigger_duration ? "true" : "false");
1481 status_reply_put(sr, "max-idle=%d", s->max_idle);
1485 fail_open_hook_create(const struct settings *s, struct switch_status *ss,
1486 struct rconn *local_rconn, struct rconn *remote_rconn)
1488 struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
1490 fail_open->local_rconn = local_rconn;
1491 fail_open->remote_rconn = remote_rconn;
1492 fail_open->lswitch = NULL;
1493 fail_open->boot_deadline = time_now() + s->probe_interval * 3;
1494 fail_open->boot_deadline += STP_EXTRA_BOOT_TIME;
1495 switch_status_register_category(ss, "fail-open",
1496 fail_open_status_cb, fail_open);
1497 return make_hook(fail_open_local_packet_cb, NULL,
1498 fail_open_periodic_cb, NULL, fail_open);
1501 struct rate_limiter {
1502 const struct settings *s;
1503 struct rconn *remote_rconn;
1505 /* One queue per physical port. */
1506 struct ofp_queue queues[OFPP_MAX];
1507 int n_queued; /* Sum over queues[*].n. */
1508 int next_tx_port; /* Next port to check in round-robin. */
1512 * It costs 1000 tokens to send a single packet_in message. A single token
1513 * per message would be more straightforward, but this choice lets us avoid
1514 * round-off error in refill_bucket()'s calculation of how many tokens to
1515 * add to the bucket, since no division step is needed. */
1516 long long int last_fill; /* Time at which we last added tokens. */
1517 int tokens; /* Current number of tokens. */
1519 /* Transmission queue. */
1520 int n_txq; /* No. of packets waiting in rconn for tx. */
1522 /* Statistics reporting. */
1523 unsigned long long n_normal; /* # txed w/o rate limit queuing. */
1524 unsigned long long n_limited; /* # queued for rate limiting. */
1525 unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
1526 unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
1529 /* Drop a packet from the longest queue in 'rl'. */
1531 drop_packet(struct rate_limiter *rl)
1533 struct ofp_queue *longest; /* Queue currently selected as longest. */
1534 int n_longest; /* # of queues of same length as 'longest'. */
1535 struct ofp_queue *q;
1537 longest = &rl->queues[0];
1539 for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
1540 if (longest->n < q->n) {
1543 } else if (longest->n == q->n) {
1546 /* Randomly select one of the longest queues, with a uniform
1547 * distribution (Knuth algorithm 3.4.2R). */
1548 if (!random_range(n_longest)) {
1554 /* FIXME: do we want to pop the tail instead? */
1555 ofpbuf_delete(queue_pop_head(longest));
1559 /* Remove and return the next packet to transmit (in round-robin order). */
1560 static struct ofpbuf *
1561 dequeue_packet(struct rate_limiter *rl)
1565 for (i = 0; i < OFPP_MAX; i++) {
1566 unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
1567 struct ofp_queue *q = &rl->queues[port];
1569 rl->next_tx_port = (port + 1) % OFPP_MAX;
1571 return queue_pop_head(q);
1577 /* Add tokens to the bucket based on elapsed time. */
1579 refill_bucket(struct rate_limiter *rl)
1581 const struct settings *s = rl->s;
1582 long long int now = time_msec();
1583 long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
1584 if (tokens >= 1000) {
1585 rl->last_fill = now;
1586 rl->tokens = MIN(tokens, s->burst_limit * 1000);
1590 /* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
1591 * true if successful, false otherwise. (In the latter case no tokens are
1594 get_token(struct rate_limiter *rl)
1596 if (rl->tokens >= 1000) {
1605 rate_limit_local_packet_cb(struct relay *r, void *rl_)
1607 struct rate_limiter *rl = rl_;
1608 const struct settings *s = rl->s;
1609 struct ofp_packet_in *opi;
1611 opi = get_ofp_packet_in(r);
1616 if (!rl->n_queued && get_token(rl)) {
1617 /* In the common case where we are not constrained by the rate limit,
1618 * let the packet take the normal path. */
1622 /* Otherwise queue it up for the periodic callback to drain out. */
1623 struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf;
1624 int port = ntohs(opi->in_port) % OFPP_MAX;
1625 if (rl->n_queued >= s->burst_limit) {
1628 queue_push_tail(&rl->queues[port], ofpbuf_clone(msg));
1636 rate_limit_status_cb(struct status_reply *sr, void *rl_)
1638 struct rate_limiter *rl = rl_;
1640 status_reply_put(sr, "normal=%llu", rl->n_normal);
1641 status_reply_put(sr, "limited=%llu", rl->n_limited);
1642 status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
1643 status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
1647 rate_limit_periodic_cb(void *rl_)
1649 struct rate_limiter *rl = rl_;
1652 /* Drain some packets out of the bucket if possible, but limit the number
1653 * of iterations to allow other code to get work done too. */
1655 for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
1656 /* Use a small, arbitrary limit for the amount of queuing to do here,
1657 * because the TCP connection is responsible for buffering and there is
1658 * no point in trying to transmit faster than the TCP connection can
1660 struct ofpbuf *b = dequeue_packet(rl);
1661 if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
1668 rate_limit_wait_cb(void *rl_)
1670 struct rate_limiter *rl = rl_;
1672 if (rl->tokens >= 1000) {
1673 /* We can transmit more packets as soon as we're called again. */
1674 poll_immediate_wake();
1676 /* We have to wait for the bucket to re-fill. We could calculate
1677 * the exact amount of time here for increased smoothness. */
1678 poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
1684 rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
1685 struct rconn *local, struct rconn *remote)
1687 struct rate_limiter *rl;
1690 rl = xcalloc(1, sizeof *rl);
1692 rl->remote_rconn = remote;
1693 for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
1694 queue_init(&rl->queues[i]);
1696 rl->last_fill = time_msec();
1697 rl->tokens = s->rate_limit * 100;
1698 switch_status_register_category(ss, "rate-limit",
1699 rate_limit_status_cb, rl);
1700 return make_hook(rate_limit_local_packet_cb, NULL, rate_limit_periodic_cb,
1701 rate_limit_wait_cb, rl);
1704 /* OFPST_SWITCH statistics. */
1706 struct switch_status_category {
1708 void (*cb)(struct status_reply *, void *aux);
1712 struct switch_status {
1713 const struct settings *s;
1715 struct switch_status_category categories[8];
1719 struct status_reply {
1720 struct switch_status_category *category;
1726 switch_status_remote_packet_cb(struct relay *r, void *ss_)
1728 struct switch_status *ss = ss_;
1729 struct rconn *rc = r->halves[HALF_REMOTE].rconn;
1730 struct ofpbuf *msg = r->halves[HALF_REMOTE].rxbuf;
1731 struct switch_status_category *c;
1732 struct nicira_header *request;
1733 struct nicira_header *reply;
1734 struct status_reply sr;
1738 if (msg->size < sizeof(struct nicira_header)) {
1741 request = msg->data;
1742 if (request->header.type != OFPT_VENDOR
1743 || request->vendor_id != htonl(NX_VENDOR_ID)
1744 || request->subtype != htonl(NXT_STATUS_REQUEST)) {
1748 sr.request.string = (void *) (request + 1);
1749 sr.request.length = msg->size - sizeof *request;
1750 ds_init(&sr.output);
1751 for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
1752 if (!memcmp(c->name, sr.request.string,
1753 MIN(strlen(c->name), sr.request.length))) {
1758 reply = make_openflow_xid(sizeof *reply + sr.output.length,
1759 OFPT_VENDOR, request->header.xid, &b);
1760 reply->vendor_id = htonl(NX_VENDOR_ID);
1761 reply->subtype = htonl(NXT_STATUS_REPLY);
1762 memcpy(reply + 1, sr.output.string, sr.output.length);
1763 retval = rconn_send(rc, b, NULL);
1764 if (retval && retval != EAGAIN) {
1765 VLOG_WARN("send failed (%s)", strerror(retval));
1767 ds_destroy(&sr.output);
1772 rconn_status_cb(struct status_reply *sr, void *rconn_)
1774 struct rconn *rconn = rconn_;
1775 time_t now = time_now();
1777 status_reply_put(sr, "name=%s", rconn_get_name(rconn));
1778 status_reply_put(sr, "state=%s", rconn_get_state(rconn));
1779 status_reply_put(sr, "backoff=%d", rconn_get_backoff(rconn));
1780 status_reply_put(sr, "is-connected=%s",
1781 rconn_is_connected(rconn) ? "true" : "false");
1782 status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
1783 status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
1784 status_reply_put(sr, "attempted-connections=%u",
1785 rconn_get_attempted_connections(rconn));
1786 status_reply_put(sr, "successful-connections=%u",
1787 rconn_get_successful_connections(rconn));
1788 status_reply_put(sr, "last-connection=%ld",
1789 (long int) (now - rconn_get_last_connection(rconn)));
1790 status_reply_put(sr, "time-connected=%lu",
1791 rconn_get_total_time_connected(rconn));
1792 status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(rconn));
1796 config_status_cb(struct status_reply *sr, void *s_)
1798 const struct settings *s = s_;
1801 for (i = 0; i < s->n_listeners; i++) {
1802 status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
1804 if (s->probe_interval) {
1805 status_reply_put(sr, "probe-interval=%d", s->probe_interval);
1807 if (s->max_backoff) {
1808 status_reply_put(sr, "max-backoff=%d", s->max_backoff);
1813 switch_status_cb(struct status_reply *sr, void *ss_)
1815 struct switch_status *ss = ss_;
1816 time_t now = time_now();
1818 status_reply_put(sr, "now=%ld", (long int) now);
1819 status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
1820 status_reply_put(sr, "pid=%ld", (long int) getpid());
1824 switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
1826 struct switch_status *ss = xcalloc(1, sizeof *ss);
1828 ss->booted = time_now();
1829 switch_status_register_category(ss, "config",
1830 config_status_cb, (void *) s);
1831 switch_status_register_category(ss, "switch", switch_status_cb, ss);
1833 return make_hook(NULL, switch_status_remote_packet_cb, NULL, NULL, ss);
1837 switch_status_register_category(struct switch_status *ss,
1838 const char *category,
1839 void (*cb)(struct status_reply *,
1843 struct switch_status_category *c;
1844 assert(ss->n_categories < ARRAY_SIZE(ss->categories));
1845 c = &ss->categories[ss->n_categories++];
1848 c->name = xstrdup(category);
1852 status_reply_put(struct status_reply *sr, const char *content, ...)
1854 size_t old_length = sr->output.length;
1858 /* Append the status reply to the output. */
1859 ds_put_format(&sr->output, "%s.", sr->category->name);
1860 va_start(args, content);
1861 ds_put_format_valist(&sr->output, content, args);
1863 if (ds_last(&sr->output) != '\n') {
1864 ds_put_char(&sr->output, '\n');
1867 /* Drop what we just added if it doesn't match the request. */
1868 added = sr->output.length - old_length;
1869 if (added < sr->request.length
1870 || memcmp(&sr->output.string[old_length],
1871 sr->request.string, sr->request.length)) {
1872 ds_truncate(&sr->output, old_length);
1877 /* Controller discovery. */
1881 const struct settings *s;
1882 struct dhclient *dhcp;
1887 discovery_status_cb(struct status_reply *sr, void *d_)
1889 struct discovery *d = d_;
1891 status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re);
1892 status_reply_put(sr, "n-changes=%d", d->n_changes);
1893 status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp));
1894 status_reply_put(sr, "state-elapsed=%u",
1895 dhclient_get_state_elapsed(d->dhcp));
1896 if (dhclient_is_bound(d->dhcp)) {
1897 uint32_t ip = dhclient_get_ip(d->dhcp);
1898 uint32_t netmask = dhclient_get_netmask(d->dhcp);
1899 uint32_t router = dhclient_get_router(d->dhcp);
1901 const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
1902 uint32_t dns_server;
1906 status_reply_put(sr, "ip="IP_FMT, IP_ARGS(&ip));
1907 status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask));
1909 status_reply_put(sr, "router="IP_FMT, IP_ARGS(&router));
1912 for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i, &dns_server);
1914 status_reply_put(sr, "dns%d="IP_FMT, i, IP_ARGS(&dns_server));
1917 domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
1919 status_reply_put(sr, "domain=%s", domain_name);
1923 status_reply_put(sr, "lease-remaining=%u",
1924 dhclient_get_lease_remaining(d->dhcp));
1928 static struct discovery *
1929 discovery_init(const struct settings *s, struct switch_status *ss)
1931 struct netdev *netdev;
1932 struct discovery *d;
1933 struct dhclient *dhcp;
1936 /* Bring ofX network device up. */
1937 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
1939 ofp_fatal(retval, "Could not open %s device", s->of_name);
1941 retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
1943 ofp_fatal(retval, "Could not bring %s device up", s->of_name);
1945 netdev_close(netdev);
1947 /* Initialize DHCP client. */
1948 retval = dhclient_create(s->of_name, modify_dhcp_request,
1949 validate_dhcp_offer, (void *) s, &dhcp);
1951 ofp_fatal(retval, "Failed to initialize DHCP client");
1953 dhclient_init(dhcp, 0);
1955 d = xmalloc(sizeof *d);
1960 switch_status_register_category(ss, "discovery", discovery_status_cb, d);
1966 discovery_question_connectivity(struct discovery *d)
1968 dhclient_force_renew(d->dhcp, 15);
1972 discovery_run(struct discovery *d, char **controller_name)
1974 dhclient_run(d->dhcp);
1975 if (!dhclient_changed(d->dhcp)) {
1979 dhclient_configure_netdev(d->dhcp);
1980 if (d->s->update_resolv_conf) {
1981 dhclient_update_resolv_conf(d->dhcp);
1984 if (dhclient_is_bound(d->dhcp)) {
1985 *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
1986 DHCP_CODE_OFP_CONTROLLER_VCONN);
1987 VLOG_WARN("%s: discovered controller", *controller_name);
1990 *controller_name = NULL;
1992 VLOG_WARN("discovered controller no longer available");
2000 discovery_wait(struct discovery *d)
2002 dhclient_wait(d->dhcp);
2006 modify_dhcp_request(struct dhcp_msg *msg, void *aux)
2008 dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
2012 validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
2014 const struct settings *s = s_;
2018 vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
2020 VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
2023 accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
2025 VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
2026 s->accept_controller_re);
2032 /* User interface. */
2035 parse_options(int argc, char *argv[], struct settings *s)
2038 OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
2040 OPT_INACTIVITY_PROBE,
2045 OPT_BOOTSTRAP_CA_CERT
2047 static struct option long_options[] = {
2048 {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN},
2049 {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF},
2050 {"fail", required_argument, 0, 'F'},
2051 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
2052 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
2053 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
2054 {"listen", required_argument, 0, 'l'},
2055 {"monitor", required_argument, 0, 'm'},
2056 {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT},
2057 {"burst-limit", required_argument, 0, OPT_BURST_LIMIT},
2058 {"detach", no_argument, 0, 'D'},
2059 {"force", no_argument, 0, 'f'},
2060 {"pidfile", optional_argument, 0, 'P'},
2061 {"verbose", optional_argument, 0, 'v'},
2062 {"help", no_argument, 0, 'h'},
2063 {"version", no_argument, 0, 'V'},
2065 VCONN_SSL_LONG_OPTIONS
2066 {"bootstrap-ca-cert", required_argument, 0, OPT_BOOTSTRAP_CA_CERT},
2070 char *short_options = long_options_to_short_options(long_options);
2071 char *accept_re = NULL;
2074 /* Set defaults that we can figure out before parsing options. */
2076 s->monitor_name = NULL;
2077 s->fail_mode = FAIL_OPEN;
2079 s->probe_interval = 15;
2080 s->max_backoff = 15;
2081 s->update_resolv_conf = true;
2087 c = getopt_long(argc, argv, short_options, long_options, NULL);
2093 case OPT_ACCEPT_VCONN:
2094 accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg);
2097 case OPT_NO_RESOLV_CONF:
2098 s->update_resolv_conf = false;
2102 if (!strcmp(optarg, "open")) {
2103 s->fail_mode = FAIL_OPEN;
2104 } else if (!strcmp(optarg, "closed")) {
2105 s->fail_mode = FAIL_CLOSED;
2107 ofp_fatal(0, "-f or --fail argument must be \"open\" "
2112 case OPT_INACTIVITY_PROBE:
2113 s->probe_interval = atoi(optarg);
2114 if (s->probe_interval < 5) {
2115 ofp_fatal(0, "--inactivity-probe argument must be at least 5");
2120 if (!strcmp(optarg, "permanent")) {
2121 s->max_idle = OFP_FLOW_PERMANENT;
2123 s->max_idle = atoi(optarg);
2124 if (s->max_idle < 1 || s->max_idle > 65535) {
2125 ofp_fatal(0, "--max-idle argument must be between 1 and "
2126 "65535 or the word 'permanent'");
2131 case OPT_MAX_BACKOFF:
2132 s->max_backoff = atoi(optarg);
2133 if (s->max_backoff < 1) {
2134 ofp_fatal(0, "--max-backoff argument must be at least 1");
2135 } else if (s->max_backoff > 3600) {
2136 s->max_backoff = 3600;
2140 case OPT_RATE_LIMIT:
2142 s->rate_limit = atoi(optarg);
2143 if (s->rate_limit < 1) {
2144 ofp_fatal(0, "--rate-limit argument must be at least 1");
2147 s->rate_limit = 1000;
2151 case OPT_BURST_LIMIT:
2152 s->burst_limit = atoi(optarg);
2153 if (s->burst_limit < 1) {
2154 ofp_fatal(0, "--burst-limit argument must be at least 1");
2163 set_pidfile(optarg);
2167 ignore_existing_pidfile();
2171 if (s->n_listeners >= MAX_MGMT) {
2173 "-l or --listen may be specified at most %d times",
2176 s->listener_names[s->n_listeners++] = optarg;
2180 if (s->monitor_name) {
2181 ofp_fatal(0, "-m or --monitor may only be specified once");
2183 s->monitor_name = optarg;
2190 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
2194 vlog_set_verbosity(optarg);
2198 VCONN_SSL_OPTION_HANDLERS
2200 case OPT_BOOTSTRAP_CA_CERT:
2201 vconn_ssl_set_ca_cert_file(optarg, true);
2212 free(short_options);
2216 if (argc < 1 || argc > 2) {
2217 ofp_fatal(0, "need one or two non-option arguments; "
2218 "use --help for usage");
2221 /* Local and remote vconns. */
2222 s->nl_name = argv[0];
2223 if (strncmp(s->nl_name, "nl:", 3)
2224 || strlen(s->nl_name) < 4
2225 || s->nl_name[strspn(s->nl_name + 3, "0123456789") + 3]) {
2226 ofp_fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"",
2229 s->of_name = xasprintf("of%s", s->nl_name + 3);
2230 s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL;
2232 /* Set accept_controller_regex. */
2234 accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*";
2236 retval = regcomp(&s->accept_controller_regex, accept_re,
2237 REG_NOSUB | REG_EXTENDED);
2239 size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0);
2240 char *buffer = xmalloc(length);
2241 regerror(retval, &s->accept_controller_regex, buffer, length);
2242 ofp_fatal(0, "%s: %s", accept_re, buffer);
2244 s->accept_controller_re = accept_re;
2246 /* Mode of operation. */
2247 s->discovery = s->controller_name == NULL;
2251 enum netdev_flags flags;
2252 struct netdev *netdev;
2254 retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
2256 ofp_fatal(retval, "Could not open %s device", s->of_name);
2259 retval = netdev_get_flags(netdev, &flags);
2261 ofp_fatal(retval, "Could not get flags for %s device", s->of_name);
2264 s->in_band = (flags & NETDEV_UP) != 0;
2265 if (s->in_band && netdev_get_in6(netdev, NULL)) {
2266 VLOG_WARN("Ignoring IPv6 address on %s device: IPv6 not supported",
2270 netdev_close(netdev);
2273 /* Rate limiting. */
2274 if (s->rate_limit) {
2275 if (s->rate_limit < 100) {
2276 VLOG_WARN("Rate limit set to unusually low value %d",
2279 if (!s->burst_limit) {
2280 s->burst_limit = s->rate_limit / 4;
2282 s->burst_limit = MAX(s->burst_limit, 1);
2283 s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000);
2290 printf("%s: secure channel, a relay for OpenFlow messages.\n"
2291 "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n"
2292 "where nl:DP_IDX is a datapath that has been added with dpctl.\n"
2293 "CONTROLLER is an active OpenFlow connection method; if it is\n"
2294 "omitted, then secchan performs controller discovery.\n",
2295 program_name, program_name);
2296 vconn_usage(true, true, true);
2297 printf("\nController discovery options:\n"
2298 " --accept-vconn=REGEX accept matching discovered controllers\n"
2299 " --no-resolv-conf do not update /etc/resolv.conf\n"
2300 "\nNetworking options:\n"
2301 " -F, --fail=open|closed when controller connection fails:\n"
2302 " closed: drop all packets\n"
2303 " open (default): act as learning switch\n"
2304 " --inactivity-probe=SECS time between inactivity probes\n"
2305 " --max-idle=SECS max idle for flows set up by secchan\n"
2306 " --max-backoff=SECS max time between controller connection\n"
2307 " attempts (default: 15 seconds)\n"
2308 " -l, --listen=METHOD allow management connections on METHOD\n"
2309 " (a passive OpenFlow connection method)\n"
2310 " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n"
2311 " (a passive OpenFlow connection method)\n"
2312 "\nRate-limiting of \"packet-in\" messages to the controller:\n"
2313 " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n"
2314 " --burst-limit=BURST limit on packet credit for idle time\n"
2315 "\nOther options:\n"
2316 " -D, --detach run in background as daemon\n"
2317 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
2318 " -f, --force with -P, start even if already running\n"
2319 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
2320 " -v, --verbose set maximum verbosity level\n"
2321 " -h, --help display this help message\n"
2322 " -V, --version display version information\n",