X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;ds=sidebyside;f=secchan%2Fsecchan.c;h=42e99da254888967124c9f74063102235e37d7e1;hb=5549180a542fa7aa17a196d55b2f5eb4cbb3159a;hp=e4f283d1ad17b822b0f5986df083c31acde3efec;hpb=903d0940fba239b2e250e9ab5f4dfdf4ecaf4bc0;p=sliver-openvswitch.git diff --git a/secchan/secchan.c b/secchan/secchan.c index e4f283d1a..42e99da25 100644 --- a/secchan/secchan.c +++ b/secchan/secchan.c @@ -40,16 +40,16 @@ #include #include #include +#include #include #include #include -#include "buffer.h" #include "command-line.h" #include "compiler.h" #include "daemon.h" -#include "dhcp.h" #include "dhcp-client.h" +#include "dhcp.h" #include "dynamic-string.h" #include "fault.h" #include "flow.h" @@ -57,6 +57,7 @@ #include "list.h" #include "mac-learning.h" #include "netdev.h" +#include "ofpbuf.h" #include "openflow.h" #include "packets.h" #include "poll-loop.h" @@ -76,6 +77,9 @@ enum fail_mode { FAIL_CLOSED /* Drop all packets. */ }; +/* Maximum number of management connection listeners. */ +#define MAX_MGMT 8 + /* Settings that may be configured by the user. */ struct settings { /* Overall mode of operation. */ @@ -86,7 +90,8 @@ struct settings { const char *nl_name; /* Local datapath (must be "nl:" vconn). */ char *of_name; /* ofX network device name. */ const char *controller_name; /* Controller (if not discovery mode). */ - const char *listen_vconn_name; /* Listens for mgmt connections. */ + const char *listener_names[MAX_MGMT]; /* Listen for mgmt connections. */ + size_t n_listeners; /* Number of mgmt connection listeners. */ /* Failure behavior. */ enum fail_mode fail_mode; /* Act as learning switch if no controller? */ @@ -94,6 +99,10 @@ struct settings { int probe_interval; /* # seconds idle before sending echo request. */ int max_backoff; /* Max # seconds between connection attempts. */ + /* Packet-in rate-limiting. */ + int rate_limit; /* Tokens added to bucket per second. */ + int burst_limit; /* Maximum number token bucket size. */ + /* Discovery behavior. */ regex_t accept_controller_regex; /* Controller vconns to accept. */ const char *accept_controller_re; /* String version of regex. */ @@ -102,7 +111,7 @@ struct settings { struct half { struct rconn *rconn; - struct buffer *rxbuf; + struct ofpbuf *rxbuf; int n_txq; /* No. of packets queued for tx on 'rconn'. */ }; @@ -119,9 +128,12 @@ struct relay { struct hook { bool (*packet_cb)(struct relay *, int half, void *aux); void (*periodic_cb)(void *aux); + void (*wait_cb)(void *aux); void *aux; }; +static struct vlog_rate_limit vrl = VLOG_RATE_LIMIT_INIT(60, 60); + static void parse_options(int argc, char *argv[], struct settings *); static void usage(void) NO_RETURN; @@ -134,17 +146,41 @@ static void relay_destroy(struct relay *); static struct hook make_hook(bool (*packet_cb)(struct relay *, int, void *), void (*periodic_cb)(void *), + void (*wait_cb)(void *), void *aux); -static struct discovery *discovery_init(const struct settings *); +struct switch_status; +struct status_reply; +static struct hook switch_status_hook_create(const struct settings *, + struct switch_status **); +static void switch_status_register_category(struct switch_status *, + const char *category, + void (*cb)(struct status_reply *, + void *aux), + void *aux); +static void status_reply_put(struct status_reply *, const char *, ...) + PRINTF_FORMAT(2, 3); + +static void rconn_status_cb(struct status_reply *, void *rconn_); + +static struct discovery *discovery_init(const struct settings *, + struct switch_status *); static void discovery_question_connectivity(struct discovery *); static bool discovery_run(struct discovery *, char **controller_name); static void discovery_wait(struct discovery *); -static struct hook in_band_hook_create(const struct settings *); +static struct hook in_band_hook_create(const struct settings *, + struct switch_status *, + struct rconn *remote); static struct hook fail_open_hook_create(const struct settings *, + struct switch_status *, struct rconn *local, struct rconn *remote); +static struct hook rate_limit_hook_create(const struct settings *, + struct switch_status *, + struct rconn *local, + struct rconn *remote); + static void modify_dhcp_request(struct dhcp_msg *, void *aux); static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux); @@ -156,13 +192,17 @@ main(int argc, char *argv[]) struct list relays = LIST_INITIALIZER(&relays); - struct hook hooks[3]; - size_t n_hooks; + struct hook hooks[8]; + size_t n_hooks = 0; + + struct vconn *listeners[MAX_MGMT]; + size_t n_listeners; struct rconn *local_rconn, *remote_rconn; - struct vconn *listen_vconn; struct relay *controller_relay; struct discovery *discovery; + struct switch_status *switch_status; + int i; int retval; set_program_name(argv[0]); @@ -170,22 +210,28 @@ main(int argc, char *argv[]) time_init(); vlog_init(); parse_options(argc, argv, &s); + signal(SIGPIPE, SIG_IGN); /* Start listening for management connections. */ - if (s.listen_vconn_name) { - retval = vconn_open(s.listen_vconn_name, &listen_vconn); + n_listeners = 0; + for (i = 0; i < s.n_listeners; i++) { + const char *name = s.listener_names[i]; + struct vconn *listener; + retval = vconn_open(name, &listener); if (retval && retval != EAGAIN) { - fatal(retval, "opening %s", s.listen_vconn_name); + fatal(retval, "opening %s", name); } - if (!vconn_is_passive(listen_vconn)) { - fatal(0, "%s is not a passive vconn", s.listen_vconn_name); + if (!vconn_is_passive(listener)) { + fatal(0, "%s is not a passive vconn", name); } - } else { - listen_vconn = NULL; + listeners[n_listeners++] = listener; } + /* Initialize switch status hook. */ + hooks[n_hooks++] = switch_status_hook_create(&s, &switch_status); + /* Start controller discovery. */ - discovery = s.discovery ? discovery_init(&s) : NULL; + discovery = s.discovery ? discovery_init(&s, switch_status) : NULL; /* Start listening for vlogconf requests. */ retval = vlog_server_listen(NULL, NULL); @@ -193,11 +239,17 @@ main(int argc, char *argv[]) fatal(retval, "Could not listen for vlog connections"); } + die_if_already_running(); daemonize(); + VLOG_WARN("OpenFlow reference implementation version %s", VERSION); + VLOG_WARN("OpenFlow protocol version 0x%02x", OFP_VERSION); + /* Connect to datapath. */ local_rconn = rconn_create(0, s.max_backoff); rconn_connect(local_rconn, s.nl_name); + switch_status_register_category(switch_status, "local", + rconn_status_cb, local_rconn); /* Connect to controller. */ remote_rconn = rconn_create(s.probe_interval, s.max_backoff); @@ -207,20 +259,26 @@ main(int argc, char *argv[]) fatal(0, "No support for %s vconn", s.controller_name); } } + switch_status_register_category(switch_status, "remote", + rconn_status_cb, remote_rconn); /* Start relaying. */ controller_relay = relay_create(local_rconn, remote_rconn, false); list_push_back(&relays, &controller_relay->node); /* Set up hooks. */ - n_hooks = 0; if (s.in_band) { - hooks[n_hooks++] = in_band_hook_create(&s); + hooks[n_hooks++] = in_band_hook_create(&s, switch_status, + remote_rconn); } if (s.fail_mode == FAIL_OPEN) { - hooks[n_hooks++] = fail_open_hook_create(&s, + hooks[n_hooks++] = fail_open_hook_create(&s, switch_status, local_rconn, remote_rconn); } + if (s.rate_limit) { + hooks[n_hooks++] = rate_limit_hook_create(&s, switch_status, + local_rconn, remote_rconn); + } assert(n_hooks <= ARRAY_SIZE(hooks)); for (;;) { @@ -231,9 +289,9 @@ main(int argc, char *argv[]) LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) { relay_run(r, hooks, n_hooks); } - if (listen_vconn) { + for (i = 0; i < n_listeners; i++) { for (;;) { - struct relay *r = relay_accept(&s, listen_vconn); + struct relay *r = relay_accept(&s, listeners[i]); if (!r) { break; } @@ -263,8 +321,13 @@ main(int argc, char *argv[]) LIST_FOR_EACH (r, struct relay, node, &relays) { relay_wait(r); } - if (listen_vconn) { - vconn_accept_wait(listen_vconn); + for (i = 0; i < n_listeners; i++) { + vconn_accept_wait(listeners[i]); + } + for (i = 0; i < n_hooks; i++) { + if (hooks[i].wait_cb) { + hooks[i].wait_cb(hooks[i].aux); + } } if (discovery) { discovery_wait(discovery); @@ -278,11 +341,13 @@ main(int argc, char *argv[]) static struct hook make_hook(bool (*packet_cb)(struct relay *, int half, void *aux), void (*periodic_cb)(void *aux), + void (*wait_cb)(void *aux), void *aux) { struct hook h; h.packet_cb = packet_cb; h.periodic_cb = periodic_cb; + h.wait_cb = wait_cb; h.aux = aux; return h; } @@ -300,7 +365,7 @@ relay_accept(const struct settings *s, struct vconn *listen_vconn) retval = vconn_accept(listen_vconn, &new_remote); if (retval) { if (retval != EAGAIN) { - VLOG_WARN("accept failed (%s)", strerror(retval)); + VLOG_WARN_RL(&vrl, "accept failed (%s)", strerror(retval)); } return NULL; } @@ -314,8 +379,8 @@ relay_accept(const struct settings *s, struct vconn *listen_vconn) nl_name_without_subscription = xasprintf("%s:0", s->nl_name); retval = vconn_open(nl_name_without_subscription, &new_local); if (retval) { - VLOG_ERR("could not connect to %s (%s)", - nl_name_without_subscription, strerror(retval)); + VLOG_ERR_RL(&vrl, "could not connect to %s (%s)", + nl_name_without_subscription, strerror(retval)); vconn_close(new_remote); free(nl_name_without_subscription); return NULL; @@ -365,7 +430,7 @@ relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks) const struct hook *h; for (h = hooks; h < &hooks[n_hooks]; h++) { if (h->packet_cb(r, i, h->aux)) { - buffer_delete(this->rxbuf); + ofpbuf_delete(this->rxbuf); this->rxbuf = NULL; progress = true; break; @@ -381,7 +446,7 @@ relay_run(struct relay *r, const struct hook hooks[], size_t n_hooks) if (!retval) { progress = true; } else { - buffer_delete(this->rxbuf); + ofpbuf_delete(this->rxbuf); } this->rxbuf = NULL; } @@ -427,7 +492,7 @@ relay_destroy(struct relay *r) for (i = 0; i < 2; i++) { struct half *this = &r->halves[i]; rconn_destroy(this->rconn); - buffer_delete(this->rxbuf); + ofpbuf_delete(this->rxbuf); } free(r); } @@ -438,18 +503,19 @@ struct in_band_data { const struct settings *s; struct mac_learning *ml; struct netdev *of_device; + struct rconn *controller; uint8_t mac[ETH_ADDR_LEN]; int n_queued; }; static void -queue_tx(struct rconn *rc, struct in_band_data *in_band, struct buffer *b) +queue_tx(struct rconn *rc, struct in_band_data *in_band, struct ofpbuf *b) { rconn_send_with_limit(rc, b, &in_band->n_queued, 10); } static const uint8_t * -get_controller_mac(struct netdev *netdev, struct rconn *controller) +get_controller_mac(struct in_band_data *in_band) { static uint32_t ip, last_nonzero_ip; static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN]; @@ -459,14 +525,14 @@ get_controller_mac(struct netdev *netdev, struct rconn *controller) time_t now = time_now(); - ip = rconn_get_ip(controller); + ip = rconn_get_ip(in_band->controller); if (last_ip != ip || !next_refresh || now >= next_refresh) { bool have_mac; /* Look up MAC address. */ memset(mac, 0, sizeof mac); if (ip) { - int retval = netdev_arp_lookup(netdev, ip, mac); + int retval = netdev_arp_lookup(in_band->of_device, ip, mac); if (retval) { VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s", IP_ARGS(&ip), strerror(retval)); @@ -498,10 +564,11 @@ get_controller_mac(struct netdev *netdev, struct rconn *controller) } static bool -is_controller_mac(const uint8_t mac[ETH_ADDR_LEN], - const uint8_t *controller_mac) +is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN], + struct in_band_data *in_band) { - return controller_mac && eth_addr_equals(mac, controller_mac); + const uint8_t *mac = get_controller_mac(in_band); + return mac && eth_addr_equals(mac, dl_addr); } static bool @@ -509,11 +576,11 @@ in_band_packet_cb(struct relay *r, int half, void *in_band_) { struct in_band_data *in_band = in_band_; struct rconn *rc = r->halves[HALF_LOCAL].rconn; - struct buffer *msg = r->halves[HALF_LOCAL].rxbuf; + struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf; struct ofp_packet_in *opi; struct ofp_header *oh; size_t pkt_ofs, pkt_len; - struct buffer pkt; + struct ofpbuf pkt; struct flow flow; uint16_t in_port, out_port; const uint8_t *controller_mac; @@ -526,8 +593,9 @@ in_band_packet_cb(struct relay *r, int half, void *in_band_) if (oh->type != OFPT_PACKET_IN) { return false; } - if (msg->size < offsetof (struct ofp_packet_in, data)) { - VLOG_WARN("packet too short (%zu bytes) for packet_in", msg->size); + 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; } @@ -541,8 +609,7 @@ in_band_packet_cb(struct relay *r, int half, void *in_band_) flow_extract(&pkt, in_port, &flow); /* Deal with local stuff. */ - controller_mac = get_controller_mac(in_band->of_device, - r->halves[HALF_REMOTE].rconn); + controller_mac = get_controller_mac(in_band); if (in_port == OFPP_LOCAL) { /* Sent by secure channel. */ out_port = mac_learning_lookup(in_band->ml, flow.dl_dst); @@ -550,26 +617,38 @@ in_band_packet_cb(struct relay *r, int half, void *in_band_) /* Sent to secure channel. */ out_port = OFPP_LOCAL; if (mac_learning_learn(in_band->ml, flow.dl_src, in_port)) { - VLOG_DBG("learned that "ETH_ADDR_FMT" is on port %"PRIu16, - ETH_ADDR_ARGS(flow.dl_src), in_port); + VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, + ETH_ADDR_ARGS(flow.dl_src), in_port); } } else if (flow.dl_type == htons(ETH_TYPE_ARP) && eth_addr_is_broadcast(flow.dl_dst) - && is_controller_mac(flow.dl_src, controller_mac)) { + && is_controller_mac(flow.dl_src, in_band)) { /* ARP sent by controller. */ out_port = OFPP_FLOOD; - } else if (is_controller_mac(flow.dl_dst, controller_mac) - && in_port == mac_learning_lookup(in_band->ml, - controller_mac)) { - /* Drop controller traffic that arrives on the controller port. */ - queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id), - in_band->s->max_idle, 0)); - return true; + } else if (is_controller_mac(flow.dl_dst, in_band)) { + if (mac_learning_learn(in_band->ml, flow.dl_src, in_port)) { + VLOG_DBG_RL(&vrl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, + ETH_ADDR_ARGS(flow.dl_src), in_port); + } + + out_port = mac_learning_lookup(in_band->ml, controller_mac); + if (in_port != out_port) { + return false; + } + + /* This is controller traffic that arrived on the controller port. + * It will get dropped below. */ + } else if (is_controller_mac(flow.dl_src, in_band)) { + out_port = mac_learning_lookup(in_band->ml, flow.dl_dst); } else { return false; } - if (out_port != OFPP_FLOOD) { + if (in_port == out_port) { + /* The input and output port match. Set up a flow to drop packets. */ + queue_tx(rc, in_band, make_add_flow(&flow, ntohl(opi->buffer_id), + in_band->s->max_idle, 0)); + } else if (out_port != OFPP_FLOOD) { /* The output port is known, so add a new flow. */ queue_tx(rc, in_band, make_add_simple_flow(&flow, ntohl(opi->buffer_id), @@ -583,7 +662,7 @@ in_band_packet_cb(struct relay *r, int half, void *in_band_) } else { /* We don't know that MAC. Send along the packet without setting up a * flow. */ - struct buffer *b; + struct ofpbuf *b; if (ntohl(opi->buffer_id) == UINT32_MAX) { b = make_unbuffered_packet_out(&pkt, in_port, out_port); } else { @@ -595,8 +674,35 @@ in_band_packet_cb(struct relay *r, int half, void *in_band_) return true; } +static void +in_band_status_cb(struct status_reply *sr, void *in_band_) +{ + 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) +in_band_hook_create(const struct settings *s, struct switch_status *ss, + struct rconn *remote) { struct in_band_data *in_band; int retval; @@ -611,8 +717,9 @@ in_band_hook_create(const struct settings *s) } memcpy(in_band->mac, netdev_get_etheraddr(in_band->of_device), ETH_ADDR_LEN); - - return make_hook(in_band_packet_cb, NULL, in_band); + 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); } /* Fail open support. */ @@ -668,17 +775,432 @@ fail_open_packet_cb(struct relay *r, int half, void *fail_open_) } } +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 rconn *local_rconn, - struct rconn *remote_rconn) +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; - return make_hook(fail_open_packet_cb, fail_open_periodic_cb, fail_open); + 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); } + +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? */ + ofpbuf_delete(queue_pop_head(longest)); + rl->n_queued--; +} + +/* Remove and return the next packet to transmit (in round-robin order). */ +static struct ofpbuf * +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 ofpbuf *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], ofpbuf_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 ofpbuf *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); +} + +/* 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 ofpbuf *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 ofpbuf *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, "backoff=%d", rconn_get_backoff(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)); + status_reply_put(sr, "state-elapsed=%u", rconn_get_state_elapsed(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); + } +} + /* Controller discovery. */ @@ -686,11 +1208,53 @@ struct discovery { const struct settings *s; struct dhclient *dhcp; - bool ever_successful; + int n_changes; }; +static void +discovery_status_cb(struct status_reply *sr, void *d_) +{ + struct discovery *d = d_; + + status_reply_put(sr, "accept-remote=%s", d->s->accept_controller_re); + status_reply_put(sr, "n-changes=%d", d->n_changes); + status_reply_put(sr, "state=%s", dhclient_get_state(d->dhcp)); + status_reply_put(sr, "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, "ip="IP_FMT, IP_ARGS(&ip)); + status_reply_put(sr, "netmask="IP_FMT, IP_ARGS(&netmask)); + if (router) { + status_reply_put(sr, "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, "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, "domain=%s", domain_name); + free(domain_name); + } + + status_reply_put(sr, "lease-remaining=%u", + dhclient_get_lease_remaining(d->dhcp)); + } +} + static struct discovery * -discovery_init(const struct settings *s) +discovery_init(const struct settings *s, struct switch_status *ss) { struct netdev *netdev; struct discovery *d; @@ -719,7 +1283,10 @@ discovery_init(const struct settings *s) d = xmalloc(sizeof *d); d->s = s; d->dhcp = dhcp; - d->ever_successful = false; + d->n_changes = 0; + + switch_status_register_category(ss, "discovery", discovery_status_cb, d); + return d; } @@ -746,11 +1313,12 @@ discovery_run(struct discovery *d, char **controller_name) *controller_name = dhcp_msg_get_string(dhclient_get_config(d->dhcp), DHCP_CODE_OFP_CONTROLLER_VCONN); VLOG_WARN("%s: discovered controller", *controller_name); - d->ever_successful = true; - } else if (controller_name) { + d->n_changes++; + } else { *controller_name = NULL; - if (d->ever_successful) { + if (d->n_changes) { VLOG_WARN("discovered controller no longer available"); + d->n_changes++; } } return true; @@ -777,13 +1345,13 @@ validate_dhcp_offer(const struct dhcp_msg *msg, void *s_) vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN); if (!vconn_name) { - VLOG_WARN("rejecting DHCP offer missing controller vconn"); + 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("rejecting controller vconn that fails to match %s", - s->accept_controller_re); + VLOG_WARN_RL(&vrl, "rejecting controller vconn that fails to match %s", + s->accept_controller_re); } free(vconn_name); return accept; @@ -799,17 +1367,22 @@ parse_options(int argc, char *argv[], struct settings *s) OPT_NO_RESOLV_CONF, OPT_INACTIVITY_PROBE, OPT_MAX_IDLE, - OPT_MAX_BACKOFF + OPT_MAX_BACKOFF, + OPT_RATE_LIMIT, + OPT_BURST_LIMIT }; static struct option long_options[] = { {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN}, {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF}, - {"fail", required_argument, 0, 'f'}, + {"fail", required_argument, 0, 'F'}, {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE}, {"max-idle", required_argument, 0, OPT_MAX_IDLE}, {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF}, {"listen", required_argument, 0, 'l'}, + {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT}, + {"burst-limit", required_argument, 0, OPT_BURST_LIMIT}, {"detach", no_argument, 0, 'D'}, + {"force", no_argument, 0, 'f'}, {"pidfile", optional_argument, 0, 'P'}, {"verbose", optional_argument, 0, 'v'}, {"help", no_argument, 0, 'h'}, @@ -822,12 +1395,14 @@ parse_options(int argc, char *argv[], struct settings *s) int retval; /* Set defaults that we can figure out before parsing options. */ - s->listen_vconn_name = NULL; + s->n_listeners = 0; s->fail_mode = FAIL_OPEN; s->max_idle = 15; s->probe_interval = 15; s->max_backoff = 15; s->update_resolv_conf = true; + s->rate_limit = 0; + s->burst_limit = 0; for (;;) { int c; @@ -845,7 +1420,7 @@ parse_options(int argc, char *argv[], struct settings *s) s->update_resolv_conf = false; break; - case 'f': + case 'F': if (!strcmp(optarg, "open")) { s->fail_mode = FAIL_OPEN; } else if (!strcmp(optarg, "closed")) { @@ -884,6 +1459,24 @@ parse_options(int argc, char *argv[], struct settings *s) } break; + case OPT_RATE_LIMIT: + if (optarg) { + s->rate_limit = atoi(optarg); + if (s->rate_limit < 1) { + fatal(0, "--rate-limit argument must be at least 1"); + } + } else { + s->rate_limit = 1000; + } + break; + + case OPT_BURST_LIMIT: + s->burst_limit = atoi(optarg); + if (s->burst_limit < 1) { + fatal(0, "--burst-limit argument must be at least 1"); + } + break; + case 'D': set_detach(); break; @@ -892,11 +1485,16 @@ parse_options(int argc, char *argv[], struct settings *s) set_pidfile(optarg); break; + case 'f': + ignore_existing_pidfile(); + break; + case 'l': - if (s->listen_vconn_name) { - fatal(0, "-l or --listen may be only specified once"); + if (s->n_listeners >= MAX_MGMT) { + fatal(0, "-l or --listen may be specified at most %d times", + MAX_MGMT); } - s->listen_vconn_name = optarg; + s->listener_names[s->n_listeners++] = optarg; break; case 'h': @@ -977,6 +1575,19 @@ parse_options(int argc, char *argv[], struct settings *s) netdev_close(netdev); } + + /* Rate limiting. */ + if (s->rate_limit) { + if (s->rate_limit < 100) { + VLOG_WARN("Rate limit set to unusually low value %d", + s->rate_limit); + } + if (!s->burst_limit) { + s->burst_limit = s->rate_limit / 4; + } + s->burst_limit = MAX(s->burst_limit, 1); + s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000); + } } static void @@ -993,7 +1604,7 @@ usage(void) " --accept-vconn=REGEX accept matching discovered controllers\n" " --no-resolv-conf do not update /etc/resolv.conf\n" "\nNetworking options:\n" - " -f, --fail=open|closed when controller connection fails:\n" + " -F, --fail=open|closed when controller connection fails:\n" " closed: drop all packets\n" " open (default): act as learning switch\n" " --inactivity-probe=SECS time between inactivity probes\n" @@ -1002,9 +1613,13 @@ usage(void) " attempts (default: 15 seconds)\n" " -l, --listen=METHOD allow management connections on METHOD\n" " (a passive OpenFlow connection method)\n" + "\nRate-limiting of \"packet-in\" messages to the controller:\n" + " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n" + " --burst-limit=BURST limit on packet credit for idle time\n" "\nOther options:\n" " -D, --detach run in background as daemon\n" " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n" + " -f, --force with -P, start even if already running\n" " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n" " -v, --verbose set maximum verbosity level\n" " -h, --help display this help message\n"