X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=controller%2Fcontroller.c;h=2d79afaa4fd58bb4c06c16bd73fc5ea1119c2572;hb=e9ab3dbf6d648c98f62248907baeea65cfc0505a;hp=3a29043df729df269917ae950d1194f6338dfd4e;hpb=8f4da69c169f4a69d8563ebf8dbe8934748fb969;p=sliver-openvswitch.git diff --git a/controller/controller.c b/controller/controller.c index 3a29043df..2d79afaa4 100644 --- a/controller/controller.c +++ b/controller/controller.c @@ -31,51 +31,39 @@ * derivatives without specific, written prior permission. */ -#include +#include + #include #include -#include -#include -#include +#include +#include #include #include -#include #include "buffer.h" #include "command-line.h" #include "compiler.h" +#include "daemon.h" #include "fault.h" -#include "flow.h" -#include "hash.h" -#include "list.h" -#include "ofp-print.h" -#include "mac-learning.h" +#include "learning-switch.h" #include "openflow.h" -#include "packets.h" #include "poll-loop.h" -#include "queue.h" -#include "time.h" +#include "rconn.h" +#include "timeval.h" #include "util.h" #include "vconn-ssl.h" #include "vconn.h" #include "vlog-socket.h" -#include "xtoxll.h" #include "vlog.h" #define THIS_MODULE VLM_controller #define MAX_SWITCHES 16 -#define MAX_TXQ 128 +#define MAX_LISTENERS 16 struct switch_ { - char *name; - struct vconn *vconn; - - uint64_t datapath_id; - time_t last_features_request; - - struct queue txq; - struct mac_learning *ml; + struct lswitch *lswitch; + struct rconn *rconn; }; /* Learn the ports on which MAC addresses appear? */ @@ -84,35 +72,29 @@ static bool learn_macs = true; /* Set up flows? (If not, every packet is processed at the controller.) */ static bool setup_flows = true; +/* --max-idle: Maximum idle time, in seconds, before flows expire. */ +static int max_idle = 60; + +static int do_switching(struct switch_ *); +static void new_switch(struct switch_ *, struct vconn *, const char *name); static void parse_options(int argc, char *argv[]); static void usage(void) NO_RETURN; -static struct switch_ *connect_switch(const char *name); -static struct switch_ *new_switch(const char *name, struct vconn *); -static void close_switch(struct switch_ *); - -static void queue_tx(struct switch_ *, struct buffer *); - -static void send_features_request(struct switch_ *); - -static int do_switch_recv(struct switch_ *this); -static int do_switch_send(struct switch_ *this); - -static void process_packet(struct switch_ *, struct buffer *); -static void process_packet_in(struct switch_ *, struct ofp_packet_in *); - int main(int argc, char *argv[]) { - struct switch_ *switches[MAX_SWITCHES]; - int n_switches; + struct switch_ switches[MAX_SWITCHES]; + struct vconn *listeners[MAX_LISTENERS]; + int n_switches, n_listeners; int retval; int i; set_program_name(argv[0]); register_fault_handlers(); + time_init(); vlog_init(); parse_options(argc, argv); + signal(SIGPIPE, SIG_IGN); if (argc - optind < 1) { fatal(0, "at least one vconn argument required; use --help for usage"); @@ -123,58 +105,74 @@ main(int argc, char *argv[]) fatal(retval, "Could not listen for vlog connections"); } - n_switches = 0; - for (i = 0; i < argc - optind; i++) { - struct switch_ *this = connect_switch(argv[optind + i]); - if (this) { + n_switches = n_listeners = 0; + for (i = optind; i < argc; i++) { + const char *name = argv[i]; + struct vconn *vconn; + int retval; + + retval = vconn_open(name, &vconn); + if (retval) { + VLOG_ERR("%s: connect: %s", name, strerror(retval)); + continue; + } + + if (vconn_is_passive(vconn)) { + if (n_listeners >= MAX_LISTENERS) { + fatal(0, "max %d passive connections", n_listeners); + } + listeners[n_listeners++] = vconn; + } else { if (n_switches >= MAX_SWITCHES) { fatal(0, "max %d switch connections", n_switches); } - switches[n_switches++] = this; + new_switch(&switches[n_switches++], vconn, name); } } - if (n_switches == 0) { - fatal(0, "could not connect to any switches"); + if (n_switches == 0 && n_listeners == 0) { + fatal(0, "no active or passive switch connections"); } - while (n_switches > 0) { - /* Do some work. Limit the number of iterations so that callbacks - * registered with the poll loop don't starve. */ + die_if_already_running(); + daemonize(); + + while (n_switches > 0 || n_listeners > 0) { int iteration; int i; + + /* Accept connections on listening vconns. */ + for (i = 0; i < n_listeners && n_switches < MAX_SWITCHES; ) { + struct vconn *new_vconn; + int retval; + + retval = vconn_accept(listeners[i], &new_vconn); + if (!retval || retval == EAGAIN) { + if (!retval) { + new_switch(&switches[n_switches++], new_vconn, "tcp"); + } + i++; + } else { + vconn_close(listeners[i]); + listeners[i] = listeners[--n_listeners]; + } + } + + /* Do some switching work. Limit the number of iterations so that + * callbacks registered with the poll loop don't starve. */ for (iteration = 0; iteration < 50; iteration++) { bool progress = false; for (i = 0; i < n_switches; ) { - struct switch_ *this = switches[i]; - int retval; - - if (vconn_is_passive(this->vconn)) { - retval = 0; - while (n_switches < MAX_SWITCHES) { - struct vconn *new_vconn; - retval = vconn_accept(this->vconn, &new_vconn); - if (retval) { - break; - } - switches[n_switches++] = new_switch("tcp", new_vconn); + struct switch_ *this = &switches[i]; + int retval = do_switching(this); + if (!retval || retval == EAGAIN) { + if (!retval) { + progress = true; } + i++; } else { - retval = do_switch_recv(this); - if (!retval || retval == EAGAIN) { - do { - retval = do_switch_send(this); - if (!retval) { - progress = true; - } - } while (!retval); - } - } - - if (retval && retval != EAGAIN) { - close_switch(this); + lswitch_destroy(this->lswitch); + rconn_destroy(this->rconn); switches[i] = switches[--n_switches]; - } else { - i++; } } if (!progress) { @@ -183,234 +181,61 @@ main(int argc, char *argv[]) } /* Wait for something to happen. */ - for (i = 0; i < n_switches; i++) { - struct switch_ *this = switches[i]; - if (vconn_is_passive(this->vconn)) { - if (n_switches < MAX_SWITCHES) { - vconn_accept_wait(this->vconn); - } - } else { - vconn_recv_wait(this->vconn); - if (this->txq.n) { - vconn_send_wait(this->vconn); - } + if (n_switches < MAX_SWITCHES) { + for (i = 0; i < n_listeners; i++) { + vconn_accept_wait(listeners[i]); } } + for (i = 0; i < n_switches; i++) { + struct switch_ *sw = &switches[i]; + rconn_run_wait(sw->rconn); + rconn_recv_wait(sw->rconn); + } poll_block(); } return 0; } -static int -do_switch_recv(struct switch_ *this) -{ - struct buffer *msg; - int retval; - - retval = vconn_recv(this->vconn, &msg); - if (!retval) { - process_packet(this, msg); - buffer_delete(msg); - } - return retval; -} - -static int -do_switch_send(struct switch_ *this) -{ - int retval = 0; - if (this->txq.n) { - struct buffer *next = this->txq.head->next; - retval = vconn_send(this->vconn, this->txq.head); - if (retval) { - return retval; - } - queue_advance_head(&this->txq, next); - return 0; - } - return EAGAIN; -} - -struct switch_ * -connect_switch(const char *name) -{ - struct vconn *vconn; - int retval; - - retval = vconn_open(name, &vconn); - if (retval) { - VLOG_ERR("%s: connect: %s", name, strerror(retval)); - return NULL; - } - - return new_switch(name, vconn); -} - -static struct switch_ * -new_switch(const char *name, struct vconn *vconn) -{ - struct switch_ *this = xmalloc(sizeof *this); - memset(this, 0, sizeof *this); - this->name = xstrdup(name); - this->vconn = vconn; - queue_init(&this->txq); - this->last_features_request = 0; - if (!vconn_is_passive(vconn)) { - send_features_request(this); - } - if (learn_macs) { - this->ml = mac_learning_create(); - } - return this; -} - -static void -close_switch(struct switch_ *this) -{ - if (this) { - free(this->name); - vconn_close(this->vconn); - queue_destroy(&this->txq); - mac_learning_destroy(this->ml); - free(this); - } -} - static void -send_features_request(struct switch_ *this) +new_switch(struct switch_ *sw, struct vconn *vconn, const char *name) { - time_t now = time(0); - if (now >= this->last_features_request + 1) { - struct buffer *b; - struct ofp_header *ofr; - struct ofp_switch_config *osc; - - /* Send OFPT_FEATURES_REQUEST. */ - b = buffer_new(0); - ofr = buffer_put_uninit(b, sizeof *ofr); - memset(ofr, 0, sizeof *ofr); - ofr->type = OFPT_FEATURES_REQUEST; - ofr->version = OFP_VERSION; - ofr->length = htons(sizeof *ofr); - queue_tx(this, b); - - /* Send OFPT_SET_CONFIG. */ - b = buffer_new(0); - osc = buffer_put_uninit(b, sizeof *osc); - memset(osc, 0, sizeof *osc); - osc->header.type = OFPT_SET_CONFIG; - osc->header.version = OFP_VERSION; - osc->header.length = htons(sizeof *osc); - osc->flags = htons(OFPC_SEND_FLOW_EXP); - osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN); - queue_tx(this, b); - - this->last_features_request = now; - } + sw->rconn = rconn_new_from_vconn(name, vconn); + sw->lswitch = lswitch_create(sw->rconn, learn_macs, + setup_flows ? max_idle : -1); } -static void -queue_tx(struct switch_ *this, struct buffer *b) -{ - queue_push_tail(&this->txq, b); -} - -static void -process_packet(struct switch_ *sw, struct buffer *msg) +static int +do_switching(struct switch_ *sw) { - static const size_t min_size[UINT8_MAX + 1] = { - [0 ... UINT8_MAX] = sizeof (struct ofp_header), - [OFPT_FEATURES_REPLY] = sizeof (struct ofp_switch_features), - [OFPT_PACKET_IN] = offsetof (struct ofp_packet_in, data), - }; - struct ofp_header *oh; - - oh = msg->data; - if (msg->size < min_size[oh->type]) { - VLOG_WARN("%s: too short (%zu bytes) for type %"PRIu8" (min %zu)", - sw->name, msg->size, oh->type, min_size[oh->type]); - return; - } + unsigned int packets_sent; + struct buffer *msg; - if (oh->type == OFPT_FEATURES_REPLY) { - struct ofp_switch_features *osf = msg->data; - sw->datapath_id = osf->datapath_id; - } else if (sw->datapath_id == 0) { - send_features_request(sw); - } else if (oh->type == OFPT_PACKET_IN) { - struct ofp_packet_in *opi = msg->data; - if (sw->txq.n >= MAX_TXQ) { - /* FIXME: ratelimit. */ - VLOG_WARN("%s: tx queue overflow", sw->name); - } else { - process_packet_in(sw, opi); - } - } else { - if (VLOG_IS_DBG_ENABLED()) { - char *p = ofp_to_string(msg->data, msg->size, 2); - VLOG_DBG("OpenFlow packet ignored: %s", p); - free(p); - } - } -} + packets_sent = rconn_packets_sent(sw->rconn); -static void -process_packet_in(struct switch_ *sw, struct ofp_packet_in *opi) -{ - uint16_t in_port = ntohs(opi->in_port); - uint16_t out_port = OFPP_FLOOD; - - size_t pkt_ofs, pkt_len; - struct buffer pkt; - struct flow flow; - - /* Extract flow data from 'opi' into 'flow'. */ - pkt_ofs = offsetof(struct ofp_packet_in, data); - pkt_len = ntohs(opi->header.length) - pkt_ofs; - pkt.data = opi->data; - pkt.size = pkt_len; - flow_extract(&pkt, in_port, &flow); - - if (learn_macs) { - if (mac_learning_learn(sw->ml, flow.dl_src, in_port)) { - VLOG_DBG("learned that "ETH_ADDR_FMT" is on datapath %" - PRIx64" port %"PRIu16, ETH_ADDR_ARGS(flow.dl_src), - ntohll(sw->datapath_id), in_port); - } - out_port = mac_learning_lookup(sw->ml, flow.dl_dst); + msg = rconn_recv(sw->rconn); + if (msg) { + lswitch_process_packet(sw->lswitch, sw->rconn, msg); + buffer_delete(msg); } + rconn_run(sw->rconn); - if (setup_flows && (!learn_macs || out_port != OFPP_FLOOD)) { - /* The output port is known, or we always flood everything, so add a - * new flow. */ - queue_tx(sw, make_add_simple_flow(&flow, ntohl(opi->buffer_id), - out_port)); - - /* If the switch didn't buffer the packet, we need to send a copy. */ - if (ntohl(opi->buffer_id) == UINT32_MAX) { - queue_tx(sw, make_unbuffered_packet_out(&pkt, in_port, out_port)); - } - } else { - /* We don't know that MAC, or we don't set up flows. Send along the - * packet without setting up a flow. */ - struct buffer *b; - if (ntohl(opi->buffer_id) == UINT32_MAX) { - b = make_unbuffered_packet_out(&pkt, in_port, out_port); - } else { - b = make_buffered_packet_out(ntohl(opi->buffer_id), - in_port, out_port); - } - queue_tx(sw, b); - } + return (!rconn_is_alive(sw->rconn) ? EOF + : rconn_packets_sent(sw->rconn) != packets_sent ? 0 + : EAGAIN); } static void parse_options(int argc, char *argv[]) { + enum { OPT_MAX_IDLE = UCHAR_MAX + 1 }; static struct option long_options[] = { + {"detach", no_argument, 0, 'D'}, + {"pidfile", optional_argument, 0, 'P'}, + {"force", no_argument, 0, 'f'}, {"hub", no_argument, 0, 'H'}, {"noflow", no_argument, 0, 'n'}, + {"max-idle", required_argument, 0, OPT_MAX_IDLE}, {"verbose", optional_argument, 0, 'v'}, {"help", no_argument, 0, 'h'}, {"version", no_argument, 0, 'V'}, @@ -429,6 +254,18 @@ parse_options(int argc, char *argv[]) } switch (c) { + case 'D': + set_detach(); + break; + + case 'P': + set_pidfile(optarg); + break; + + case 'f': + ignore_existing_pidfile(); + break; + case 'H': learn_macs = false; break; @@ -437,6 +274,18 @@ parse_options(int argc, char *argv[]) setup_flows = false; break; + case OPT_MAX_IDLE: + if (!strcmp(optarg, "permanent")) { + max_idle = OFP_FLOW_PERMANENT; + } else { + max_idle = atoi(optarg); + if (max_idle < 1 || max_idle > 65535) { + fatal(0, "--max-idle argument must be between 1 and " + "65535 or the word 'permanent'"); + } + } + break; + case 'h': usage(); @@ -469,10 +318,16 @@ usage(void) program_name, program_name); vconn_usage(true, true); printf("\nOther options:\n" + " -D, --detach run in background as daemon\n" + " -P, --pidfile[=FILE] create pidfile (default: %s/controller.pid)\n" + " -f, --force with -P, start even if already running\n" " -H, --hub act as hub instead of learning switch\n" " -n, --noflow pass traffic, but don't add flows\n" + " --max-idle=SECS max idle time for new flows\n" + " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n" " -v, --verbose set maximum verbosity level\n" " -h, --help display this help message\n" - " -V, --version display version information\n"); + " -V, --version display version information\n", + RUNDIR); exit(EXIT_SUCCESS); }