* derivatives without specific, written prior permission.
*/
-#include <assert.h>
+#include <config.h>
+
#include <errno.h>
#include <getopt.h>
-#include <inttypes.h>
-#include <netinet/in.h>
-#include <poll.h>
+#include <limits.h>
+#include <signal.h>
#include <stdlib.h>
#include <string.h>
-#include <unistd.h>
-#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 "learning-switch.h"
+#include "ofpbuf.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 lswitch *lswitch;
+ struct rconn *rconn;
};
-/* -H, --hub: Use dumb hub instead of learning switch? */
-static bool hub = false;
+/* Learn the ports on which MAC addresses appear? */
+static bool learn_macs = true;
-/* -n, --noflow: Pass traffic, but don't setup flows in switch */
-static bool noflow = false;
+/* 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_hub(struct switch_ *, struct ofp_packet_in *);
-static void process_noflow(struct switch_ *, struct ofp_packet_in *);
-
-static void switch_init(void);
-static void process_switch(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 pvconn *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);
-
- if (!hub && !noflow) {
- switch_init();
- }
+ signal(SIGPIPE, SIG_IGN);
if (argc - optind < 1) {
- fatal(0, "at least one vconn argument required; use --help for usage");
+ ofp_fatal(0, "at least one vconn argument required; "
+ "use --help for usage");
}
retval = vlog_server_listen(NULL, NULL);
if (retval) {
- fatal(retval, "Could not listen for vlog connections");
+ ofp_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, OFP_VERSION, &vconn);
+ if (!retval) {
if (n_switches >= MAX_SWITCHES) {
- fatal(0, "max %d switch connections", n_switches);
+ ofp_fatal(0, "max %d switch connections", n_switches);
}
- switches[n_switches++] = this;
+ new_switch(&switches[n_switches++], vconn, name);
+ continue;
+ } else if (retval == EAFNOSUPPORT) {
+ struct pvconn *pvconn;
+ retval = pvconn_open(name, &pvconn);
+ if (!retval) {
+ if (n_listeners >= MAX_LISTENERS) {
+ ofp_fatal(0, "max %d passive connections", n_listeners);
+ }
+ listeners[n_listeners++] = pvconn;
+ }
+ }
+ if (retval) {
+ VLOG_ERR("%s: connect: %s", name, strerror(retval));
}
}
- if (n_switches == 0) {
- fatal(0, "could not connect to any switches");
+ if (n_switches == 0 && n_listeners == 0) {
+ ofp_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 = pvconn_accept(listeners[i], OFP_VERSION, &new_vconn);
+ if (!retval || retval == EAGAIN) {
+ if (!retval) {
+ new_switch(&switches[n_switches++], new_vconn, "tcp");
+ }
+ i++;
+ } else {
+ pvconn_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;
- }
- printf("accept!\n");
- 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) {
}
/* 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++) {
+ pvconn_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);
- }
- return this;
-}
-
static void
-close_switch(struct switch_ *this)
+new_switch(struct switch_ *sw, struct vconn *vconn, const char *name)
{
- if (this) {
- printf("dropped!\n");
- free(this->name);
- vconn_close(this->vconn);
- queue_destroy(&this->txq);
- free(this);
- }
+ sw->rconn = rconn_new_from_vconn(name, vconn);
+ sw->lswitch = lswitch_create(sw->rconn, learn_macs,
+ setup_flows ? max_idle : -1);
}
-static void
-send_features_request(struct switch_ *this)
-{
- 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_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);
-
- /* 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);
-
- this->last_features_request = now;
- }
-}
-
-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 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;
- }
-
- 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 if (noflow) {
- process_noflow(sw, opi);
- } else if (hub) {
- process_hub(sw, opi);
- } else {
- process_switch(sw, opi);
- }
- } else {
- ofp_print(stdout, msg->data, msg->size, 2);
- }
-}
-
-static void
-process_hub(struct switch_ *sw, struct ofp_packet_in *opi)
-{
- 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, ntohs(opi->in_port), &flow);
-
- /* Add new flow. */
- queue_tx(sw, make_add_simple_flow(&flow, ntohl(opi->buffer_id),
- OFPP_FLOOD));
-
- /* 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, ntohs(flow.in_port),
- OFPP_FLOOD));
- }
-}
-
-static void
-process_noflow(struct switch_ *sw, struct ofp_packet_in *opi)
-{
- /* If the switch didn't buffer the packet, we need to send a copy. */
- if (ntohl(opi->buffer_id) == UINT32_MAX) {
- size_t pkt_ofs, pkt_len;
- struct buffer pkt;
-
- /* 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;
-
- queue_tx(sw, make_unbuffered_packet_out(&pkt, ntohs(opi->in_port),
- OFPP_FLOOD));
- } else {
- queue_tx(sw, make_buffered_packet_out(ntohl(opi->buffer_id),
- ntohs(opi->in_port), OFPP_FLOOD));
- }
-}
-
-
-#define MAC_HASH_BITS 10
-#define MAC_HASH_MASK (MAC_HASH_SIZE - 1)
-#define MAC_HASH_SIZE (1u << MAC_HASH_BITS)
-
-#define MAC_MAX 1024
-
-struct mac_source {
- struct list hash_list;
- struct list lru_list;
- uint64_t datapath_id;
- uint8_t mac[ETH_ADDR_LEN];
- uint16_t port;
-};
-
-static struct list mac_table[MAC_HASH_SIZE];
-static struct list lrus;
-static size_t mac_count;
-
-static void
-switch_init(void)
-{
- int i;
-
- list_init(&lrus);
- for (i = 0; i < MAC_HASH_SIZE; i++) {
- list_init(&mac_table[i]);
- }
-}
-
-static struct list *
-mac_table_bucket(uint64_t datapath_id, const uint8_t mac[ETH_ADDR_LEN])
-{
- uint32_t hash;
- hash = hash_fnv(&datapath_id, sizeof datapath_id, HASH_FNV_BASIS);
- hash = hash_fnv(mac, ETH_ADDR_LEN, hash);
- return &mac_table[hash & MAC_HASH_BITS];
-}
-
-static void
-process_switch(struct switch_ *sw, struct ofp_packet_in *opi)
+static int
+do_switching(struct switch_ *sw)
{
- size_t pkt_ofs, pkt_len;
- struct buffer pkt;
- struct flow flow;
-
- uint16_t out_port;
-
- /* 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, ntohs(opi->in_port), &flow);
-
- /* Learn the source. */
- if (!eth_addr_is_multicast(flow.dl_src)) {
- struct mac_source *src;
- struct list *bucket;
- bool found;
-
- bucket = mac_table_bucket(sw->datapath_id, flow.dl_src);
- found = false;
- LIST_FOR_EACH (src, struct mac_source, hash_list, bucket) {
- if (src->datapath_id == sw->datapath_id
- && eth_addr_equals(src->mac, flow.dl_src)) {
- found = true;
- break;
- }
- }
-
- if (!found) {
- /* Learn a new address. */
-
- if (mac_count >= MAC_MAX) {
- /* Drop the least recently used mac source. */
- struct mac_source *lru;
- lru = CONTAINER_OF(lrus.next, struct mac_source, lru_list);
- list_remove(&lru->hash_list);
- list_remove(&lru->lru_list);
- free(lru);
- } else {
- mac_count++;
- }
+ unsigned int packets_sent;
+ struct ofpbuf *msg;
- /* Create new mac source */
- src = xmalloc(sizeof *src);
- src->datapath_id = sw->datapath_id;
- memcpy(src->mac, flow.dl_src, ETH_ADDR_LEN);
- src->port = -1;
- list_push_front(bucket, &src->hash_list);
- list_push_back(&lrus, &src->lru_list);
- } else {
- /* Make 'src' most-recently-used. */
- list_remove(&src->lru_list);
- list_push_back(&lrus, &src->lru_list);
- }
+ packets_sent = rconn_packets_sent(sw->rconn);
- if (ntohs(flow.in_port) != src->port) {
- src->port = ntohs(flow.in_port);
- VLOG_DBG("learned that "ETH_ADDR_FMT" is on datapath %"
- PRIx64" port %d",
- ETH_ADDR_ARGS(src->mac), ntohll(src->datapath_id),
- src->port);
- }
- } else {
- VLOG_DBG("multicast packet source "ETH_ADDR_FMT,
- ETH_ADDR_ARGS(flow.dl_src));
+ msg = rconn_recv(sw->rconn);
+ if (msg) {
+ lswitch_process_packet(sw->lswitch, sw->rconn, msg);
+ ofpbuf_delete(msg);
}
+ rconn_run(sw->rconn);
- /* Figure out the destination. */
- out_port = OFPP_FLOOD;
- if (!eth_addr_is_multicast(flow.dl_dst)) {
- struct mac_source *dst;
- struct list *bucket;
-
- bucket = mac_table_bucket(sw->datapath_id, flow.dl_dst);
- LIST_FOR_EACH (dst, struct mac_source, hash_list, bucket) {
- if (dst->datapath_id == sw->datapath_id
- && eth_addr_equals(dst->mac, flow.dl_dst)) {
- out_port = dst->port;
- break;
- }
- }
- }
-
- if (out_port != OFPP_FLOOD) {
- /* The output port is known, 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, ntohs(flow.in_port),
- out_port));
- }
- } else {
- /* We don't know that MAC. Flood the packet. */
- struct buffer *b;
- if (ntohl(opi->buffer_id) == UINT32_MAX) {
- b = make_unbuffered_packet_out(&pkt, ntohs(flow.in_port), out_port);
- } else {
- b = make_buffered_packet_out(ntohl(opi->buffer_id),
- ntohs(flow.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,
+ OPT_PEER_CA_CERT
+ };
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'},
+#ifdef HAVE_OPENSSL
VCONN_SSL_LONG_OPTIONS
+ {"peer-ca-cert", required_argument, 0, OPT_PEER_CA_CERT},
+#endif
{0, 0, 0, 0},
};
char *short_options = long_options_to_short_options(long_options);
}
switch (c) {
+ case 'D':
+ set_detach();
+ break;
+
+ case 'P':
+ set_pidfile(optarg);
+ break;
+
+ case 'f':
+ ignore_existing_pidfile();
+ break;
+
case 'H':
- hub = true;
+ learn_macs = false;
break;
case 'n':
- noflow = true;
+ 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) {
+ ofp_fatal(0, "--max-idle argument must be between 1 and "
+ "65535 or the word 'permanent'");
+ }
+ }
break;
case 'h':
vlog_set_verbosity(optarg);
break;
+#ifdef HAVE_OPENSSL
VCONN_SSL_OPTION_HANDLERS
+ case OPT_PEER_CA_CERT:
+ vconn_ssl_set_peer_ca_cert_file(optarg);
+ break;
+#endif
+
case '?':
exit(EXIT_FAILURE);
"usage: %s [OPTIONS] METHOD\n"
"where METHOD is any OpenFlow connection method.\n",
program_name, program_name);
- vconn_usage(true, true);
+ vconn_usage(true, true, false);
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);
}