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.
37 #include <netinet/in.h>
45 #include "command-line.h"
50 #include "learning-switch.h"
52 #include "mac-learning.h"
56 #include "poll-loop.h"
59 #include "vconn-ssl.h"
61 #include "vlog-socket.h"
64 #define THIS_MODULE VLM_secchan
66 #include "ofp-print.h"
68 static const char *listen_vconn_name;
75 /* Behavior when the connection to the controller fails. */
77 FAIL_OPEN, /* Act as learning switch. */
78 FAIL_CLOSED /* Drop all packets. */
86 struct half halves[2];
89 struct lswitch *lswitch;
92 static struct list relays = LIST_INITIALIZER(&relays);
94 /* Enable the local port? */
95 static int local_port;
97 /* MAC address of local port. */
98 static uint8_t local_mac[ETH_ADDR_LEN];
100 /* MAC learning table for local port. */
101 static struct mac_learning *local_ml;
103 /* -f, --fail: Behavior when the connection to the controller fails. */
104 static enum fail_mode fail_mode = FAIL_OPEN;
106 /* The OpenFlow virtual network device ofX. */
107 static struct netdev *of_device;
109 /* --inactivity-probe: Number of seconds without receiving a message from the
110 controller before sending an inactivity probe. */
111 static int probe_interval = 15;
113 /* --max-idle: Idle time to assign to flows created by learning switch when in
115 static int max_idle = 15;
117 /* --max-backoff: Maximum interval between controller connection attempts, in
119 static int max_backoff = 15;
121 static void parse_options(int argc, char *argv[]);
122 static void usage(void) NO_RETURN;
124 static void new_management_connection(const char *nl_name, struct vconn *new_remote);
125 static struct relay *relay_create(struct rconn *local, struct rconn *remote,
127 static void relay_run(struct relay *);
128 static void relay_wait(struct relay *);
129 static void relay_destroy(struct relay *);
131 static bool local_hook(struct relay *r);
132 static bool failing_open(struct relay *r);
133 static bool fail_open_hook(struct relay *r);
136 main(int argc, char *argv[])
138 struct rconn *local_rconn, *remote_rconn;
139 struct vconn *listen_vconn;
140 struct relay *controller_relay;
145 set_program_name(argv[0]);
146 register_fault_handlers();
148 parse_options(argc, argv);
150 if (argc - optind != 2) {
152 "need exactly two non-option arguments; use --help for usage");
154 nl_name = argv[optind];
155 if (strncmp(nl_name, "nl:", 3)
156 || strlen(nl_name) < 4
157 || nl_name[strspn(nl_name + 3, "0123456789") + 3]) {
158 fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"", nl_name);
161 if (listen_vconn_name) {
162 retval = vconn_open(listen_vconn_name, &listen_vconn);
163 if (retval && retval != EAGAIN) {
164 fatal(retval, "opening %s", listen_vconn_name);
166 if (!vconn_is_passive(listen_vconn)) {
167 fatal(0, "%s is not a passive vconn", listen_vconn_name);
173 snprintf(of_name, sizeof of_name, "of%s", nl_name + 3);
174 retval = netdev_open(of_name, NETDEV_ETH_TYPE_NONE, &of_device);
176 enum netdev_flags flags;
177 retval = netdev_get_flags(of_device, &flags);
179 if (flags & NETDEV_UP) {
183 memcpy(local_mac, netdev_get_etheraddr(of_device),
185 if (netdev_get_in6(of_device, &in6)) {
186 VLOG_WARN("Ignoring IPv6 address on %s device: "
187 "IPv6 not supported", of_name);
189 local_ml = mac_learning_create();
192 error(retval, "Could not get flags for %s device", of_name);
195 error(retval, "Could not open %s device", of_name);
198 retval = vlog_server_listen(NULL, NULL);
200 fatal(retval, "Could not listen for vlog connections");
205 local_rconn = rconn_create(1, 0, max_backoff);
206 retval = rconn_connect(local_rconn, nl_name);
207 if (retval == EAFNOSUPPORT) {
208 fatal(0, "No support for %s vconn", nl_name);
211 remote_rconn = rconn_create(1, probe_interval, max_backoff);
212 retval = rconn_connect(remote_rconn, argv[optind + 1]);
213 if (retval == EAFNOSUPPORT) {
214 fatal(0, "No support for %s vconn", argv[optind + 1]);
216 controller_relay = relay_create(local_rconn, remote_rconn, false);
221 LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) {
226 struct vconn *new_remote;
227 retval = vconn_accept(listen_vconn, &new_remote);
229 if (retval != EAGAIN) {
230 VLOG_WARN("accept failed (%s)", strerror(retval));
234 new_management_connection(nl_name, new_remote);
237 failing_open(controller_relay);
239 /* Wait for something to happen. */
240 LIST_FOR_EACH (r, struct relay, node, &relays) {
244 vconn_accept_wait(listen_vconn);
253 new_management_connection(const char *nl_name, struct vconn *new_remote)
255 char *nl_name_without_subscription;
256 struct vconn *new_local;
257 struct rconn *r1, *r2;
260 /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We
261 * only accept the former syntax in main().
263 * nl:123:0 opens a netlink connection to local datapath 123 without
264 * obtaining a subscription for ofp_packet_in or ofp_flow_expired
266 nl_name_without_subscription = xasprintf("%s:0", nl_name);
267 retval = vconn_open(nl_name_without_subscription, &new_local);
269 VLOG_ERR("could not connect to %s (%s)",
270 nl_name_without_subscription, strerror(retval));
271 vconn_close(new_remote);
272 free(nl_name_without_subscription);
276 /* Add it to the relay list. */
277 r1 = rconn_new_from_vconn(nl_name_without_subscription, 1, new_local);
278 r2 = rconn_new_from_vconn("passive", 1, new_remote);
279 relay_create(r1, r2, true);
281 free(nl_name_without_subscription);
284 static struct relay *
285 relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn)
290 r = xmalloc(sizeof *r);
291 r->halves[HALF_LOCAL].rconn = local;
292 r->halves[HALF_REMOTE].rconn = remote;
293 for (i = 0; i < 2; i++) {
294 r->halves[i].rxbuf = NULL;
296 r->is_mgmt_conn = is_mgmt_conn;
298 list_push_back(&relays, &r->node);
303 relay_run(struct relay *r)
308 for (i = 0; i < 2; i++) {
309 rconn_run(r->halves[i].rconn);
312 /* Limit the number of iterations to prevent other tasks from starving. */
313 for (iteration = 0; iteration < 50; iteration++) {
314 bool progress = false;
315 for (i = 0; i < 2; i++) {
316 struct half *this = &r->halves[i];
317 struct half *peer = &r->halves[!i];
320 this->rxbuf = rconn_recv(this->rconn);
321 if (this->rxbuf && !r->is_mgmt_conn && i == HALF_LOCAL
322 && (local_hook(r) || fail_open_hook(r))) {
323 buffer_delete(this->rxbuf);
329 int retval = rconn_send(peer->rconn, this->rxbuf);
330 if (retval != EAGAIN) {
334 buffer_delete(this->rxbuf);
345 for (i = 0; i < 2; i++) {
346 struct half *this = &r->halves[i];
347 if (!rconn_is_alive(this->rconn)) {
355 relay_wait(struct relay *r)
359 for (i = 0; i < 2; i++) {
360 struct half *this = &r->halves[i];
362 rconn_run_wait(this->rconn);
364 rconn_recv_wait(this->rconn);
370 relay_destroy(struct relay *r)
374 list_remove(&r->node);
375 for (i = 0; i < 2; i++) {
376 struct half *this = &r->halves[i];
377 rconn_destroy(this->rconn);
378 buffer_delete(this->rxbuf);
384 queue_tx(struct rconn *rc, struct buffer *b)
386 if (rconn_force_send(rc, b)) {
392 is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN],
393 struct rconn *controller)
395 static uint32_t ip, last_nonzero_ip;
396 static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN];
397 static time_t next_refresh = 0;
399 uint32_t last_ip = ip;
401 time_t now = time(0);
403 ip = rconn_get_ip(controller);
404 if (last_ip != ip || !next_refresh || now >= next_refresh) {
407 /* Look up MAC address. */
408 memset(mac, 0, sizeof mac);
410 int retval = netdev_arp_lookup(of_device, ip, mac);
412 VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s",
413 IP_ARGS(&ip), strerror(retval));
416 have_mac = !eth_addr_is_zero(mac);
418 /* Log changes in IP, MAC addresses. */
419 if (ip && ip != last_nonzero_ip) {
420 VLOG_DBG("controller IP address changed from "IP_FMT
421 " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip));
422 last_nonzero_ip = ip;
424 if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) {
425 VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to "
427 ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac));
428 memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN);
431 /* Schedule next refresh.
433 * If we have an IP address but not a MAC address, then refresh
434 * quickly, since we probably will get a MAC address soon (via ARP).
435 * Otherwise, we can afford to wait a little while. */
436 next_refresh = now + (!ip || have_mac ? 10 : 1);
438 return !eth_addr_is_zero(mac) && eth_addr_equals(mac, dl_addr);
442 local_hook(struct relay *r)
444 struct rconn *rc = r->halves[HALF_LOCAL].rconn;
445 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
446 struct ofp_packet_in *opi;
447 struct ofp_header *oh;
448 size_t pkt_ofs, pkt_len;
451 uint16_t in_port, out_port;
458 if (oh->type != OFPT_PACKET_IN) {
461 if (msg->size < offsetof (struct ofp_packet_in, data)) {
462 VLOG_WARN("packet too short (%zu bytes) for packet_in", msg->size);
466 /* Extract flow data from 'opi' into 'flow'. */
468 in_port = ntohs(opi->in_port);
469 pkt_ofs = offsetof(struct ofp_packet_in, data);
470 pkt_len = ntohs(opi->header.length) - pkt_ofs;
471 pkt.data = opi->data;
473 flow_extract(&pkt, in_port, &flow);
475 /* Deal with local stuff. */
476 if (in_port == OFPP_LOCAL) {
477 out_port = mac_learning_lookup(local_ml, flow.dl_dst);
478 } else if (eth_addr_equals(flow.dl_dst, local_mac)) {
479 out_port = OFPP_LOCAL;
480 if (mac_learning_learn(local_ml, flow.dl_src, in_port)) {
481 VLOG_DBG("learned that "ETH_ADDR_FMT" is on port %"PRIu16,
482 ETH_ADDR_ARGS(flow.dl_src), in_port);
484 } else if (flow.dl_type == htons(ETH_TYPE_ARP)
485 && eth_addr_is_broadcast(flow.dl_dst)
486 && is_controller_mac(flow.dl_src,
487 r->halves[HALF_REMOTE].rconn)) {
488 out_port = OFPP_FLOOD;
493 if (out_port != OFPP_FLOOD) {
494 /* The output port is known, so add a new flow. */
495 queue_tx(rc, make_add_simple_flow(&flow, ntohl(opi->buffer_id),
496 out_port, max_idle));
498 /* If the switch didn't buffer the packet, we need to send a copy. */
499 if (ntohl(opi->buffer_id) == UINT32_MAX) {
500 queue_tx(rc, make_unbuffered_packet_out(&pkt, in_port, out_port));
503 /* We don't know that MAC. Send along the packet without setting up a
506 if (ntohl(opi->buffer_id) == UINT32_MAX) {
507 b = make_unbuffered_packet_out(&pkt, in_port, out_port);
509 b = make_buffered_packet_out(ntohl(opi->buffer_id),
517 /* Causess 'r' to enter or leave fail-open mode, if appropriate. Returns true
518 * if 'r' is in fail-open fail, false otherwise. */
520 failing_open(struct relay *r)
522 struct rconn *local = r->halves[HALF_LOCAL].rconn;
523 struct rconn *remote = r->halves[HALF_REMOTE].rconn;
524 int disconnected_duration;
526 if (fail_mode == FAIL_CLOSED) {
527 /* We fail closed, so there's never anything to do. */
531 disconnected_duration = rconn_disconnected_duration(remote);
532 if (disconnected_duration < probe_interval * 3) {
533 /* It's not time to fail open yet. */
534 if (r->lswitch && rconn_is_connected(remote)) {
535 /* We're connected, so drop the learning switch. */
536 VLOG_WARN("No longer in fail-open mode");
537 lswitch_destroy(r->lswitch);
544 VLOG_WARN("Could not connect to controller for %d seconds, "
545 "failing open", disconnected_duration);
546 r->lswitch = lswitch_create(local, true, max_idle);
552 fail_open_hook(struct relay *r)
554 if (!failing_open(r)) {
557 struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
558 struct rconn *local = r->halves[HALF_LOCAL].rconn;
559 lswitch_process_packet(r->lswitch, local, msg);
566 parse_options(int argc, char *argv[])
569 OPT_INACTIVITY_PROBE = UCHAR_MAX + 1,
573 static struct option long_options[] = {
574 {"fail", required_argument, 0, 'f'},
575 {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE},
576 {"max-idle", required_argument, 0, OPT_MAX_IDLE},
577 {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF},
578 {"listen", required_argument, 0, 'l'},
579 {"detach", no_argument, 0, 'D'},
580 {"pidfile", optional_argument, 0, 'P'},
581 {"verbose", optional_argument, 0, 'v'},
582 {"help", no_argument, 0, 'h'},
583 {"version", no_argument, 0, 'V'},
584 VCONN_SSL_LONG_OPTIONS
587 char *short_options = long_options_to_short_options(long_options);
592 c = getopt_long(argc, argv, short_options, long_options, NULL);
599 if (!strcmp(optarg, "open")) {
600 fail_mode = FAIL_OPEN;
601 } else if (!strcmp(optarg, "closed")) {
602 fail_mode = FAIL_CLOSED;
605 "-f or --fail argument must be \"open\" or \"closed\"");
609 case OPT_INACTIVITY_PROBE:
610 probe_interval = atoi(optarg);
611 if (probe_interval < 5) {
612 fatal(0, "--inactivity-probe argument must be at least 5");
617 if (!strcmp(optarg, "permanent")) {
618 max_idle = OFP_FLOW_PERMANENT;
620 max_idle = atoi(optarg);
621 if (max_idle < 1 || max_idle > 65535) {
622 fatal(0, "--max-idle argument must be between 1 and "
623 "65535 or the word 'permanent'");
628 case OPT_MAX_BACKOFF:
629 max_backoff = atoi(optarg);
630 if (max_backoff < 1) {
631 fatal(0, "--max-backoff argument must be at least 1");
632 } else if (max_backoff > 3600) {
642 set_pidfile(optarg ? optarg : "secchan.pid");
646 if (listen_vconn_name) {
647 fatal(0, "-l or --listen may be only specified once");
649 listen_vconn_name = optarg;
656 printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]);
660 vlog_set_verbosity(optarg);
663 VCONN_SSL_OPTION_HANDLERS
678 printf("%s: secure channel, a relay for OpenFlow messages.\n"
679 "usage: %s [OPTIONS] nl:DP_IDX CONTROLLER\n"
680 "where nl:DP_IDX is a datapath that has been added with dpctl\n"
681 "and CONTROLLER is an active OpenFlow connection method.\n",
682 program_name, program_name);
683 vconn_usage(true, true);
684 printf("\nNetworking options:\n"
685 " -f, --fail=open|closed when controller connection fails:\n"
686 " closed: drop all packets\n"
687 " open (default): act as learning switch\n"
688 " --inactivity-probe=SECS time between inactivity probes\n"
689 " --max-idle=SECS max idle for flows set up by secchan\n"
690 " --max-backoff=SECS max time between controller connection\n"
691 " attempts (default: 15 seconds)\n"
692 " -l, --listen=METHOD allow management connections on METHOD\n"
693 " (a passive OpenFlow connection method)\n"
695 " -D, --detach run in background as daemon\n"
696 " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n"
697 " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n"
698 " -v, --verbose set maximum verbosity level\n"
699 " -h, --help display this help message\n"
700 " -V, --version display version information\n",