+ struct in_band_data *in_band = in_band_;
+ struct in_addr local_ip;
+ uint32_t controller_ip;
+ const uint8_t *controller_mac;
+
+ if (netdev_get_in4(in_band->of_device, &local_ip)) {
+ status_reply_put(sr, "local-ip="IP_FMT, IP_ARGS(&local_ip.s_addr));
+ }
+ status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
+ ETH_ADDR_ARGS(in_band->mac));
+
+ controller_ip = rconn_get_ip(in_band->controller);
+ if (controller_ip) {
+ status_reply_put(sr, "controller-ip="IP_FMT,
+ IP_ARGS(&controller_ip));
+ }
+ controller_mac = get_controller_mac(in_band);
+ if (controller_mac) {
+ status_reply_put(sr, "controller-mac="ETH_ADDR_FMT,
+ ETH_ADDR_ARGS(controller_mac));
+ }
+}
+
+static struct hook
+in_band_hook_create(const struct settings *s, struct switch_status *ss,
+ struct rconn *remote)
+{
+ struct in_band_data *in_band;
+ int retval;
+
+ in_band = xcalloc(1, sizeof *in_band);
+ in_band->s = s;
+ in_band->ml = mac_learning_create();
+ retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE,
+ &in_band->of_device);
+ if (retval) {
+ fatal(retval, "Could not open %s device", s->of_name);
+ }
+ memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device),
+ ETH_ADDR_LEN);
+ in_band->controller = remote;
+ switch_status_register_category(ss, "in-band", in_band_status_cb, in_band);
+ return make_hook(in_band_packet_cb, NULL, NULL, in_band);
+}
+\f
+/* Fail open support. */
+
+struct fail_open_data {
+ const struct settings *s;
+ struct rconn *local_rconn;
+ struct rconn *remote_rconn;
+ struct lswitch *lswitch;
+ int last_disconn_secs;
+};
+
+/* Causes 'r' to enter or leave fail-open mode, if appropriate. */
+static void
+fail_open_periodic_cb(void *fail_open_)
+{
+ struct fail_open_data *fail_open = fail_open_;
+ int disconn_secs;
+ bool open;
+
+ disconn_secs = rconn_disconnected_duration(fail_open->remote_rconn);
+ open = disconn_secs >= fail_open->s->probe_interval * 3;
+ if (open != (fail_open->lswitch != NULL)) {
+ if (!open) {
+ VLOG_WARN("No longer in fail-open mode");
+ lswitch_destroy(fail_open->lswitch);
+ fail_open->lswitch = NULL;
+ } else {
+ VLOG_WARN("Could not connect to controller for %d seconds, "
+ "failing open", disconn_secs);
+ fail_open->lswitch = lswitch_create(fail_open->local_rconn, true,
+ fail_open->s->max_idle);
+ fail_open->last_disconn_secs = disconn_secs;
+ }
+ } else if (open && disconn_secs > fail_open->last_disconn_secs + 60) {
+ VLOG_WARN("Still in fail-open mode after %d seconds disconnected "
+ "from controller", disconn_secs);
+ fail_open->last_disconn_secs = disconn_secs;
+ }
+}
+
+static bool
+fail_open_packet_cb(struct relay *r, int half, void *fail_open_)
+{
+ struct fail_open_data *fail_open = fail_open_;
+ if (half != HALF_LOCAL || r->is_mgmt_conn || !fail_open->lswitch) {
+ return false;
+ } else {
+ lswitch_process_packet(fail_open->lswitch, fail_open->local_rconn,
+ r->halves[HALF_LOCAL].rxbuf);
+ rconn_run(fail_open->local_rconn);
+ return true;
+ }
+}
+
+static void
+fail_open_status_cb(struct status_reply *sr, void *fail_open_)
+{
+ struct fail_open_data *fail_open = fail_open_;
+ const struct settings *s = fail_open->s;
+ int trigger_duration = s->probe_interval * 3;
+ int cur_duration = rconn_disconnected_duration(fail_open->remote_rconn);
+
+ status_reply_put(sr, "trigger-duration=%d", trigger_duration);
+ status_reply_put(sr, "current-duration=%d", cur_duration);
+ status_reply_put(sr, "triggered=%s",
+ cur_duration >= trigger_duration ? "true" : "false");
+ status_reply_put(sr, "max-idle=%d", s->max_idle);
+}
+
+static struct hook
+fail_open_hook_create(const struct settings *s, struct switch_status *ss,
+ struct rconn *local_rconn, struct rconn *remote_rconn)
+{
+ struct fail_open_data *fail_open = xmalloc(sizeof *fail_open);
+ fail_open->s = s;
+ fail_open->local_rconn = local_rconn;
+ fail_open->remote_rconn = remote_rconn;
+ fail_open->lswitch = NULL;
+ switch_status_register_category(ss, "fail-open",
+ fail_open_status_cb, fail_open);
+ return make_hook(fail_open_packet_cb, fail_open_periodic_cb, NULL,
+ fail_open);
+}
+\f
+struct rate_limiter {
+ const struct settings *s;
+ struct rconn *remote_rconn;
+
+ /* One queue per physical port. */
+ struct queue queues[OFPP_MAX];
+ int n_queued; /* Sum over queues[*].n. */
+ int next_tx_port; /* Next port to check in round-robin. */
+
+ /* Token bucket.
+ *
+ * It costs 1000 tokens to send a single packet_in message. A single token
+ * per message would be more straightforward, but this choice lets us avoid
+ * round-off error in refill_bucket()'s calculation of how many tokens to
+ * add to the bucket, since no division step is needed. */
+ long long int last_fill; /* Time at which we last added tokens. */
+ int tokens; /* Current number of tokens. */
+
+ /* Transmission queue. */
+ int n_txq; /* No. of packets waiting in rconn for tx. */
+
+ /* Statistics reporting. */
+ unsigned long long n_normal; /* # txed w/o rate limit queuing. */
+ unsigned long long n_limited; /* # queued for rate limiting. */
+ unsigned long long n_queue_dropped; /* # dropped due to queue overflow. */
+ unsigned long long n_tx_dropped; /* # dropped due to tx overflow. */
+};
+
+/* Drop a packet from the longest queue in 'rl'. */
+static void
+drop_packet(struct rate_limiter *rl)
+{
+ struct queue *longest; /* Queue currently selected as longest. */
+ int n_longest; /* # of queues of same length as 'longest'. */
+ struct queue *q;
+
+ longest = &rl->queues[0];
+ n_longest = 1;
+ for (q = &rl->queues[0]; q < &rl->queues[OFPP_MAX]; q++) {
+ if (longest->n < q->n) {
+ longest = q;
+ n_longest = 1;
+ } else if (longest->n == q->n) {
+ n_longest++;
+
+ /* Randomly select one of the longest queues, with a uniform
+ * distribution (Knuth algorithm 3.4.2R). */
+ if (!random_range(n_longest)) {
+ longest = q;
+ }
+ }
+ }
+
+ /* FIXME: do we want to pop the tail instead? */
+ buffer_delete(queue_pop_head(longest));
+ rl->n_queued--;
+}
+
+/* Remove and return the next packet to transmit (in round-robin order). */
+static struct buffer *
+dequeue_packet(struct rate_limiter *rl)
+{
+ unsigned int i;
+
+ for (i = 0; i < OFPP_MAX; i++) {
+ unsigned int port = (rl->next_tx_port + i) % OFPP_MAX;
+ struct queue *q = &rl->queues[port];
+ if (q->n) {
+ rl->next_tx_port = (port + 1) % OFPP_MAX;
+ rl->n_queued--;
+ return queue_pop_head(q);
+ }
+ }
+ NOT_REACHED();
+}
+
+/* Add tokens to the bucket based on elapsed time. */
+static void
+refill_bucket(struct rate_limiter *rl)
+{
+ const struct settings *s = rl->s;
+ long long int now = time_msec();
+ long long int tokens = (now - rl->last_fill) * s->rate_limit + rl->tokens;
+ if (tokens >= 1000) {
+ rl->last_fill = now;
+ rl->tokens = MIN(tokens, s->burst_limit * 1000);
+ }
+}
+
+/* Attempts to remove enough tokens from 'rl' to transmit a packet. Returns
+ * true if successful, false otherwise. (In the latter case no tokens are
+ * removed.) */
+static bool
+get_token(struct rate_limiter *rl)
+{
+ if (rl->tokens >= 1000) {
+ rl->tokens -= 1000;
+ return true;
+ } else {
+ return false;
+ }
+}
+
+static bool
+rate_limit_packet_cb(struct relay *r, int half, void *rl_)
+{
+ struct rate_limiter *rl = rl_;
+ const struct settings *s = rl->s;
+ struct buffer *msg = r->halves[HALF_LOCAL].rxbuf;
+ struct ofp_header *oh;
+
+ if (half == HALF_REMOTE) {
+ return false;
+ }
+
+ oh = msg->data;
+ if (oh->type != OFPT_PACKET_IN) {
+ return false;
+ }
+ if (msg->size < offsetof(struct ofp_packet_in, data)) {
+ VLOG_WARN_RL(&vrl, "packet too short (%zu bytes) for packet_in",
+ msg->size);
+ return false;
+ }
+
+ if (!rl->n_queued && get_token(rl)) {
+ /* In the common case where we are not constrained by the rate limit,
+ * let the packet take the normal path. */
+ rl->n_normal++;
+ return false;
+ } else {
+ /* Otherwise queue it up for the periodic callback to drain out. */
+ struct ofp_packet_in *opi = msg->data;
+ int port = ntohs(opi->in_port) % OFPP_MAX;
+ if (rl->n_queued >= s->burst_limit) {
+ drop_packet(rl);
+ }
+ queue_push_tail(&rl->queues[port], buffer_clone(msg));
+ rl->n_queued++;
+ rl->n_limited++;
+ return true;
+ }
+}
+
+static void
+rate_limit_status_cb(struct status_reply *sr, void *rl_)
+{
+ struct rate_limiter *rl = rl_;
+
+ status_reply_put(sr, "normal=%llu", rl->n_normal);
+ status_reply_put(sr, "limited=%llu", rl->n_limited);
+ status_reply_put(sr, "queue-dropped=%llu", rl->n_queue_dropped);
+ status_reply_put(sr, "tx-dropped=%llu", rl->n_tx_dropped);
+}
+
+static void
+rate_limit_periodic_cb(void *rl_)
+{
+ struct rate_limiter *rl = rl_;
+ int i;
+
+ /* Drain some packets out of the bucket if possible, but limit the number
+ * of iterations to allow other code to get work done too. */
+ refill_bucket(rl);
+ for (i = 0; rl->n_queued && get_token(rl) && i < 50; i++) {
+ /* Use a small, arbitrary limit for the amount of queuing to do here,
+ * because the TCP connection is responsible for buffering and there is
+ * no point in trying to transmit faster than the TCP connection can
+ * handle. */
+ struct buffer *b = dequeue_packet(rl);
+ if (rconn_send_with_limit(rl->remote_rconn, b, &rl->n_txq, 10)) {
+ rl->n_tx_dropped++;
+ }
+ }
+}
+
+static void
+rate_limit_wait_cb(void *rl_)
+{
+ struct rate_limiter *rl = rl_;
+ if (rl->n_queued) {
+ if (rl->tokens >= 1000) {
+ /* We can transmit more packets as soon as we're called again. */
+ poll_immediate_wake();
+ } else {
+ /* We have to wait for the bucket to re-fill. We could calculate
+ * the exact amount of time here for increased smoothness. */
+ poll_timer_wait(TIME_UPDATE_INTERVAL / 2);
+ }
+ }
+}
+
+static struct hook
+rate_limit_hook_create(const struct settings *s, struct switch_status *ss,
+ struct rconn *local, struct rconn *remote)
+{
+ struct rate_limiter *rl;
+ size_t i;
+
+ rl = xcalloc(1, sizeof *rl);
+ rl->s = s;
+ rl->remote_rconn = remote;
+ for (i = 0; i < ARRAY_SIZE(rl->queues); i++) {
+ queue_init(&rl->queues[i]);
+ }
+ rl->last_fill = time_msec();
+ rl->tokens = s->rate_limit * 100;
+ switch_status_register_category(ss, "rate-limit",
+ rate_limit_status_cb, rl);
+ return make_hook(rate_limit_packet_cb, rate_limit_periodic_cb,
+ rate_limit_wait_cb, rl);
+}
+\f
+/* OFPST_SWITCH statistics. */
+
+struct switch_status_category {
+ char *name;
+ void (*cb)(struct status_reply *, void *aux);
+ void *aux;
+};
+
+struct switch_status {
+ const struct settings *s;
+ time_t booted;
+ struct switch_status_category categories[8];
+ int n_categories;
+};
+
+struct status_reply {
+ struct switch_status_category *category;
+ struct ds request;
+ struct ds output;
+};
+
+static bool
+switch_status_packet_cb(struct relay *r, int half, void *ss_)
+{
+ struct switch_status *ss = ss_;
+ struct rconn *rc = r->halves[HALF_REMOTE].rconn;
+ struct buffer *msg = r->halves[HALF_REMOTE].rxbuf;
+ struct switch_status_category *c;
+ struct ofp_stats_request *osr;
+ struct ofp_stats_reply *reply;
+ struct status_reply sr;
+ struct ofp_header *oh;
+ struct buffer *b;
+ int retval;
+
+ if (half == HALF_LOCAL) {
+ return false;
+ }
+
+ oh = msg->data;
+ if (oh->type != OFPT_STATS_REQUEST) {
+ return false;
+ }
+ if (msg->size < sizeof(struct ofp_stats_request)) {
+ VLOG_WARN_RL(&vrl, "packet too short (%zu bytes) for stats_request",
+ msg->size);
+ return false;
+ }
+
+ osr = msg->data;
+ if (osr->type != htons(OFPST_SWITCH)) {
+ return false;
+ }
+
+ sr.request.string = (void *) (osr + 1);
+ sr.request.length = msg->size - sizeof *osr;
+ ds_init(&sr.output);
+ for (c = ss->categories; c < &ss->categories[ss->n_categories]; c++) {
+ if (!memcmp(c->name, sr.request.string,
+ MIN(strlen(c->name), sr.request.length))) {
+ sr.category = c;
+ c->cb(&sr, c->aux);
+ }
+ }
+ reply = make_openflow_xid((offsetof(struct ofp_stats_reply, body)
+ + sr.output.length),
+ OFPT_STATS_REPLY, osr->header.xid, &b);
+ reply->type = htons(OFPST_SWITCH);
+ reply->flags = 0;
+ memcpy(reply->body, sr.output.string, sr.output.length);
+ retval = rconn_send(rc, b, NULL);
+ if (retval && retval != EAGAIN) {
+ VLOG_WARN("send failed (%s)", strerror(retval));
+ }
+ ds_destroy(&sr.output);
+ return true;
+}
+
+static void
+rconn_status_cb(struct status_reply *sr, void *rconn_)
+{
+ struct rconn *rconn = rconn_;
+ time_t now = time_now();
+
+ status_reply_put(sr, "name=%s", rconn_get_name(rconn));
+ status_reply_put(sr, "state=%s", rconn_get_state(rconn));
+ status_reply_put(sr, "is-connected=%s",
+ rconn_is_connected(rconn) ? "true" : "false");
+ status_reply_put(sr, "sent-msgs=%u", rconn_packets_sent(rconn));
+ status_reply_put(sr, "received-msgs=%u", rconn_packets_received(rconn));
+ status_reply_put(sr, "attempted-connections=%u",
+ rconn_get_attempted_connections(rconn));
+ status_reply_put(sr, "successful-connections=%u",
+ rconn_get_successful_connections(rconn));
+ status_reply_put(sr, "last-connection=%ld",
+ (long int) (now - rconn_get_last_connection(rconn)));
+ status_reply_put(sr, "time-connected=%lu",
+ rconn_get_total_time_connected(rconn));
+}
+
+static void
+config_status_cb(struct status_reply *sr, void *s_)
+{
+ const struct settings *s = s_;
+ size_t i;
+
+ for (i = 0; i < s->n_listeners; i++) {
+ status_reply_put(sr, "management%zu=%s", i, s->listener_names[i]);
+ }
+ if (s->probe_interval) {
+ status_reply_put(sr, "probe-interval=%d", s->probe_interval);
+ }
+ if (s->max_backoff) {
+ status_reply_put(sr, "max-backoff=%d", s->max_backoff);
+ }
+}
+
+static void
+switch_status_cb(struct status_reply *sr, void *ss_)
+{
+ struct switch_status *ss = ss_;
+ time_t now = time_now();
+
+ status_reply_put(sr, "now=%ld", (long int) now);
+ status_reply_put(sr, "uptime=%ld", (long int) (now - ss->booted));
+ status_reply_put(sr, "pid=%ld", (long int) getpid());
+}
+
+static struct hook
+switch_status_hook_create(const struct settings *s, struct switch_status **ssp)
+{
+ struct switch_status *ss = xcalloc(1, sizeof *ss);
+ ss->s = s;
+ ss->booted = time_now();
+ switch_status_register_category(ss, "config",
+ config_status_cb, (void *) s);
+ switch_status_register_category(ss, "switch", switch_status_cb, ss);
+ *ssp = ss;
+ return make_hook(switch_status_packet_cb, NULL, NULL, ss);
+}
+
+static void
+switch_status_register_category(struct switch_status *ss,
+ const char *category,
+ void (*cb)(struct status_reply *,
+ void *aux),
+ void *aux)
+{
+ struct switch_status_category *c;
+ assert(ss->n_categories < ARRAY_SIZE(ss->categories));
+ c = &ss->categories[ss->n_categories++];
+ c->cb = cb;
+ c->aux = aux;
+ c->name = xstrdup(category);
+}
+
+static void
+status_reply_put(struct status_reply *sr, const char *content, ...)
+{
+ size_t old_length = sr->output.length;
+ size_t added;
+ va_list args;
+
+ /* Append the status reply to the output. */
+ ds_put_format(&sr->output, "%s.", sr->category->name);
+ va_start(args, content);
+ ds_put_format_valist(&sr->output, content, args);
+ va_end(args);
+ if (ds_last(&sr->output) != '\n') {
+ ds_put_char(&sr->output, '\n');
+ }
+
+ /* Drop what we just added if it doesn't match the request. */
+ added = sr->output.length - old_length;
+ if (added < sr->request.length
+ || memcmp(&sr->output.string[old_length],
+ sr->request.string, sr->request.length)) {
+ ds_truncate(&sr->output, old_length);
+ }
+}
+
+\f
+/* Controller discovery. */
+
+struct discovery
+{
+ const struct settings *s;
+ struct dhclient *dhcp;
+ int n_changes;
+};
+
+static void
+discovery_status_cb(struct status_reply *sr, void *d_)
+{
+ struct discovery *d = d_;
+
+ status_reply_put(sr, "discovery.accept-remote=%s",
+ d->s->accept_controller_re);
+ status_reply_put(sr, "discovery.n-changes=%d", d->n_changes);
+ status_reply_put(sr, "discovery.state=%s", dhclient_get_state(d->dhcp));
+ status_reply_put(sr, "discovery.state-elapsed=%u",
+ dhclient_get_state_elapsed(d->dhcp));
+ if (dhclient_is_bound(d->dhcp)) {
+ uint32_t ip = dhclient_get_ip(d->dhcp);
+ uint32_t netmask = dhclient_get_netmask(d->dhcp);
+ uint32_t router = dhclient_get_router(d->dhcp);
+
+ const struct dhcp_msg *cfg = dhclient_get_config(d->dhcp);
+ uint32_t dns_server;
+ char *domain_name;
+ int i;
+
+ status_reply_put(sr, "discovery.ip="IP_FMT, IP_ARGS(&ip));
+ status_reply_put(sr, "discovery.netmask="IP_FMT, IP_ARGS(&netmask));
+ if (router) {
+ status_reply_put(sr, "discovery.router="IP_FMT, IP_ARGS(&router));
+ }
+
+ for (i = 0; dhcp_msg_get_ip(cfg, DHCP_CODE_DNS_SERVER, i, &dns_server);
+ i++) {
+ status_reply_put(sr, "discovery.dns%d="IP_FMT,
+ i, IP_ARGS(&dns_server));
+ }
+
+ domain_name = dhcp_msg_get_string(cfg, DHCP_CODE_DOMAIN_NAME);
+ if (domain_name) {
+ status_reply_put(sr, "discovery.domain=%s", domain_name);
+ free(domain_name);
+ }
+
+ status_reply_put(sr, "discovery.lease-remaining=%u",
+ dhclient_get_lease_remaining(d->dhcp));
+ }
+}
+
+static struct discovery *
+discovery_init(const struct settings *s, struct switch_status *ss)
+{
+ struct netdev *netdev;
+ struct discovery *d;
+ struct dhclient *dhcp;
+ int retval;
+
+ /* Bring ofX network device up. */
+ retval = netdev_open(s->of_name, NETDEV_ETH_TYPE_NONE, &netdev);
+ if (retval) {
+ fatal(retval, "Could not open %s device", s->of_name);
+ }
+ retval = netdev_turn_flags_on(netdev, NETDEV_UP, true);
+ if (retval) {
+ fatal(retval, "Could not bring %s device up", s->of_name);
+ }
+ netdev_close(netdev);
+
+ /* Initialize DHCP client. */
+ retval = dhclient_create(s->of_name, modify_dhcp_request,
+ validate_dhcp_offer, (void *) s, &dhcp);
+ if (retval) {
+ fatal(retval, "Failed to initialize DHCP client");
+ }
+ dhclient_init(dhcp, 0);
+
+ d = xmalloc(sizeof *d);
+ d->s = s;
+ d->dhcp = dhcp;
+ d->n_changes = 0;
+
+ switch_status_register_category(ss, "discovery", discovery_status_cb, d);
+
+ return d;
+}
+
+static void
+discovery_question_connectivity(struct discovery *d)
+{
+ dhclient_force_renew(d->dhcp, 15);
+}
+
+static bool
+discovery_run(struct discovery *d, char **controller_name)
+{
+ dhclient_run(d->dhcp);
+ if (!dhclient_changed(d->dhcp)) {
+ return false;
+ }
+
+ dhclient_configure_netdev(d->dhcp);
+ if (d->s->update_resolv_conf) {
+ dhclient_update_resolv_conf(d->dhcp);
+ }
+
+ if (dhclient_is_bound(d->dhcp)) {
+ *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp),
+ DHCP_CODE_OFP_CONTROLLER_VCONN);
+ VLOG_WARN("%s: discovered controller", *controller_name);
+ d->n_changes++;
+ } else {
+ *controller_name = NULL;
+ if (d->n_changes) {
+ VLOG_WARN("discovered controller no longer available");
+ d->n_changes++;
+ }
+ }
+ return true;
+}
+
+static void
+discovery_wait(struct discovery *d)
+{
+ dhclient_wait(d->dhcp);
+}
+
+static void
+modify_dhcp_request(struct dhcp_msg *msg, void *aux)
+{
+ dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow");
+}
+
+static bool
+validate_dhcp_offer(const struct dhcp_msg *msg, void *s_)
+{
+ const struct settings *s = s_;
+ char *vconn_name;
+ bool accept;
+
+ vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN);
+ if (!vconn_name) {
+ VLOG_WARN_RL(&vrl, "rejecting DHCP offer missing controller vconn");
+ return false;
+ }
+ accept = !regexec(&s->accept_controller_regex, vconn_name, 0, NULL, 0);
+ if (!accept) {
+ VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s",
+ s->accept_controller_re);
+ }
+ free(vconn_name);
+ return accept;
+}
+\f
+/* User interface. */
+
+static void
+parse_options(int argc, char *argv[], struct settings *s)
+{
+ enum {
+ OPT_ACCEPT_VCONN = UCHAR_MAX + 1,
+ OPT_NO_RESOLV_CONF,
+ OPT_INACTIVITY_PROBE,
+ OPT_MAX_IDLE,
+ OPT_MAX_BACKOFF,
+ OPT_RATE_LIMIT,
+ OPT_BURST_LIMIT
+ };