Send the features request before the set config to be OpenFlow-compliant.
[sliver-openvswitch.git] / controller / controller.c
index 43c0b3f..8b69686 100644 (file)
@@ -1,22 +1,34 @@
-/* Copyright (C) 2007 Board of Trustees, Leland Stanford Jr. University.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+/* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
+ * Junior University
+ * 
+ * We are making the OpenFlow specification and associated documentation
+ * (Software) available for public use and benefit with the expectation
+ * that others will use, modify and enhance the Software and contribute
+ * those enhancements back to the community. However, since we would
+ * like to make the Software available for broadest use, with as few
+ * restrictions as possible permission is hereby granted, free of
+ * charge, to any person obtaining a copy of this Software to deal in
+ * the Software under the copyrights without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT.  IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ * 
+ * The name and trademarks of copyright holder(s) may NOT be used in
+ * advertising or publicity pertaining to the Software or any
+ * derivatives without specific, written prior permission.
  */
 
 #include <assert.h>
 #include "flow.h"
 #include "hash.h"
 #include "list.h"
-#include "mac.h"
 #include "ofp-print.h"
 #include "openflow.h"
+#include "packets.h"
+#include "poll-loop.h"
+#include "queue.h"
 #include "time.h"
 #include "util.h"
 #include "vconn-ssl.h"
 struct switch_ {
     char *name;
     struct vconn *vconn;
-    struct pollfd *pollfd;
 
     uint64_t datapath_id;
-    time_t last_control_hello;
+    time_t last_features_request;
 
-    int n_txq;
-    struct buffer *txq, *tx_tail;
+    struct queue txq;
 };
 
 /* -H, --hub: Use dumb hub instead of learning switch? */
@@ -79,7 +91,7 @@ static void close_switch(struct switch_ *);
 
 static void queue_tx(struct switch_ *, struct buffer *);
 
-static void send_control_hello(struct switch_ *);
+static void send_features_request(struct switch_ *);
 
 static int do_switch_recv(struct switch_ *this);
 static int do_switch_send(struct switch_ *this);
@@ -95,8 +107,6 @@ int
 main(int argc, char *argv[])
 {
     struct switch_ *switches[MAX_SWITCHES];
-    struct pollfd pollfds[MAX_SWITCHES + 1];
-    struct vlog_server *vlog_server;
     int n_switches;
     int retval;
     int i;
@@ -114,7 +124,7 @@ main(int argc, char *argv[])
         fatal(0, "at least one vconn argument required; use --help for usage");
     }
 
-    retval = vlog_server_listen(NULL, &vlog_server);
+    retval = vlog_server_listen(NULL, NULL);
     if (retval) {
         fatal(retval, "Could not listen for vlog connections");
     }
@@ -132,99 +142,67 @@ main(int argc, char *argv[])
     if (n_switches == 0) {
         fatal(0, "could not connect to any switches");
     }
-    
-    while (n_switches > 0) {
-        size_t n_ready;
-        int retval;
-
-        /* Wait until there's something to do. */
-        n_ready = 0;
-        for (i = 0; i < n_switches; i++) {
-            struct switch_ *this = switches[i];
-            int want;
-
-            if (vconn_is_passive(this->vconn)) {
-                want = n_switches < MAX_SWITCHES ? WANT_ACCEPT : 0;
-            } else {
-                want = WANT_RECV;
-                if (this->n_txq) {
-                    want |= WANT_SEND;
-                }
-            }
-
-            this->pollfd = &pollfds[i];
-            this->pollfd->fd = -1;
-            this->pollfd->events = 0;
-            n_ready += vconn_prepoll(this->vconn, want, this->pollfd);
-        }
-        if (vlog_server) {
-            pollfds[n_switches].fd = vlog_server_get_fd(vlog_server);
-            pollfds[n_switches].events = POLLIN;
-        }
-        do {
-            retval = poll(pollfds, n_switches + (vlog_server != NULL),
-                          n_ready ? 0 : -1);
-        } while (retval < 0 && errno == EINTR);
-        if (retval < 0 || (retval == 0 && !n_ready)) {
-            fatal(retval < 0 ? errno : 0, "poll");
-        }
-
-        /* Let each connection deal with any pending operations. */
-        for (i = 0; i < n_switches; i++) {
-            struct switch_ *this = switches[i];
-            vconn_postpoll(this->vconn, &this->pollfd->revents);
-            if (this->pollfd->revents & POLLERR) {
-                this->pollfd->revents |= POLLIN | POLLOUT;
-            }
-        }
-        if (vlog_server && pollfds[n_switches].revents) {
-            vlog_server_poll(vlog_server);
-        }
 
-        for (i = 0; i < n_switches; ) {
-            struct switch_ *this = switches[i];
+    while (n_switches > 0) {
+        /* Do some work.  Limit the number of iterations so that callbacks
+         * registered with the poll loop don't starve. */
+        int iteration;
+        int i;
+        for (iteration = 0; iteration < 50; iteration++) {
+            bool progress = false;
+            for (i = 0; i < n_switches; ) {
+                struct switch_ *this = switches[i];
+                int retval;
 
-            if (this->pollfd) {
-                retval = 0;
                 if (vconn_is_passive(this->vconn)) {
-                    if (this->pollfd->revents & POLLIN) {
+                    retval = 0;
+                    while (n_switches < MAX_SWITCHES) {
                         struct vconn *new_vconn;
-                        while (n_switches < MAX_SWITCHES 
-                               && (retval = vconn_accept(this->vconn,
-                                                         &new_vconn)) == 0) {
-                            switches[n_switches++] = new_switch("tcp",
-                                                                new_vconn);
+                        retval = vconn_accept(this->vconn, &new_vconn);
+                        if (retval) {
+                            break;
                         }
+                        switches[n_switches++] = new_switch("tcp", new_vconn);
                     }
                 } else {
-                    bool may_read = this->pollfd->revents & POLLIN;
-                    bool may_write = this->pollfd->revents & POLLOUT;
-                    if (may_read) {
-                        retval = do_switch_recv(this);
-                        if (!retval || retval == EAGAIN) {
-                            retval = 0;
-
-                            /* Enable writing to avoid round trip through poll
-                             * in common case. */
-                            may_write = true;
-                        }
-                    }
-                    while ((!retval || retval == EAGAIN) && may_write) {
-                        retval = do_switch_send(this);
-                        may_write = !retval;
+                    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);
                     switches[i] = switches[--n_switches];
-                    continue;
+                } else {
+                    i++;
+                }
+            }
+            if (!progress) {
+                break;
+            }
+        }
+
+        /* 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 {
-                /* New switch that hasn't been polled yet. */
+                vconn_recv_wait(this->vconn);
+                if (this->txq.n) {
+                    vconn_send_wait(this->vconn);
+                }
             }
-            i++;
         }
+        poll_block();
     }
 
     return 0;
@@ -248,19 +226,13 @@ static int
 do_switch_send(struct switch_ *this) 
 {
     int retval = 0;
-    if (this->n_txq) {
-        struct buffer *next = this->txq->next;
-
-        retval = vconn_send(this->vconn, this->txq);
+    if (this->txq.n) {
+        struct buffer *next = this->txq.head->next;
+        retval = vconn_send(this->vconn, this->txq.head);
         if (retval) {
             return retval;
         }
-
-        this->txq = next;
-        if (this->txq == NULL) {
-            this->tx_tail = NULL;
-        }
-        this->n_txq--;
+        queue_advance_head(&this->txq, next);
         return 0;
     }
     return EAGAIN;
@@ -288,13 +260,10 @@ new_switch(const char *name, struct vconn *vconn)
     memset(this, 0, sizeof *this);
     this->name = xstrdup(name);
     this->vconn = vconn;
-    this->pollfd = NULL;
-    this->n_txq = 0;
-    this->txq = NULL;
-    this->tx_tail = NULL;
-    this->last_control_hello = 0;
+    queue_init(&this->txq);
+    this->last_features_request = 0;
     if (!vconn_is_passive(vconn)) {
-        send_control_hello(this);
+        send_features_request(this);
     }
     return this;
 }
@@ -303,93 +272,59 @@ static void
 close_switch(struct switch_ *this) 
 {
     if (this) {
-        struct buffer *cur, *next;
-
         free(this->name);
         vconn_close(this->vconn);
-        for (cur = this->txq; cur != NULL; cur = next) {
-            next = cur->next;
-            buffer_delete(cur);
-        }
+        queue_destroy(&this->txq);
         free(this);
     }
 }
 
 static void
-send_control_hello(struct switch_ *this)
+send_features_request(struct switch_ *this)
 {
     time_t now = time(0);
-    if (now >= this->last_control_hello + 1) {
+    if (now >= this->last_features_request + 1) {
         struct buffer *b;
-        struct ofp_control_hello *och;
+        struct ofp_header *ofr;
+        struct ofp_switch_config *osc;
 
+        /* Send OFPT_FEATURES_REQUEST. */
         b = buffer_new(0);
-        och = buffer_put_uninit(b, sizeof *och);
-        memset(och, 0, sizeof *och);
-        och->header.version = OFP_VERSION;
-        och->header.length = htons(sizeof *och);
-
-        och->version = htonl(OFP_VERSION);
-        och->flags = htons(OFP_CHELLO_SEND_FLOW_EXP);
-        och->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
+        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_control_hello = now;
-    }
-}
+        /* 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);
 
-static void
-check_txq(struct switch_ *this UNUSED)
-{
-#if 0
-    struct buffer *iter;
-    size_t n;
-
-    assert(this->n_txq == 0
-           ? this->txq == NULL && this->tx_tail == NULL
-           : this->txq != NULL && this->tx_tail != NULL);
-
-    n = 0;
-    for (iter = this->txq; iter != NULL; iter = iter->next) {
-        n++;
-        assert((iter->next != NULL) == (iter != this->tx_tail));
+        this->last_features_request = now;
     }
-    assert(n == this->n_txq);
-#endif
 }
 
 static void
 queue_tx(struct switch_ *this, struct buffer *b) 
 {
-    check_txq(this);
-
-    b->next = NULL;
-    if (this->n_txq++) {
-        this->tx_tail->next = b;
-    } else {
-        this->txq = b;
-    }
-    this->tx_tail = b;
-
-    check_txq(this);
+    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] = SIZE_MAX,
-        [OFPT_CONTROL_HELLO] = sizeof (struct ofp_control_hello),
-        [OFPT_DATA_HELLO] = sizeof (struct ofp_data_hello),
+        [0 ... UINT8_MAX] = sizeof (struct ofp_header),
+        [OFPT_FEATURES_REPLY] = sizeof (struct ofp_switch_features),
         [OFPT_PACKET_IN] = offsetof (struct ofp_packet_in, data),
-        [OFPT_PACKET_OUT] = sizeof (struct ofp_packet_out),
-        [OFPT_FLOW_MOD] = sizeof (struct ofp_flow_mod),
-        [OFPT_FLOW_EXPIRED] = sizeof (struct ofp_flow_expired),
-        [OFPT_TABLE] = sizeof (struct ofp_table),
-        [OFPT_PORT_MOD] = sizeof (struct ofp_port_mod),
-        [OFPT_PORT_STATUS] = sizeof (struct ofp_port_status),
-        [OFPT_FLOW_STAT_REQUEST] = sizeof (struct ofp_flow_stat_request),
-        [OFPT_FLOW_STAT_REPLY] = sizeof (struct ofp_flow_stat_reply),
     };
     struct ofp_header *oh;
 
@@ -400,28 +335,26 @@ process_packet(struct switch_ *sw, struct buffer *msg)
         return;
     }
 
-    if (oh->type == OFPT_DATA_HELLO) {
-        struct ofp_data_hello *odh = msg->data;
-        sw->datapath_id = odh->datapath_id;
+    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_control_hello(sw);
-        return;
-    }
-
-    if (oh->type == OFPT_PACKET_IN) {
-        if (sw->n_txq >= MAX_TXQ) {
+        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, msg->data);
+            process_noflow(sw, opi);
         } else if (hub) {
-            process_hub(sw, msg->data);
+            process_hub(sw, opi);
         } else {
-            process_switch(sw, msg->data);
+            process_switch(sw, opi);
         }
-        return;
+    } else {
+        ofp_print(stdout, msg->data, msg->size, 2); 
     }
-
-    ofp_print(stdout, msg->data, msg->size, 2);
 }
 
 static void
@@ -527,7 +460,7 @@ process_switch(struct switch_ *sw, struct ofp_packet_in *opi)
     flow_extract(&pkt, ntohs(opi->in_port), &flow);
 
     /* Learn the source. */
-    if (!mac_is_multicast(flow.dl_src)) {
+    if (!eth_addr_is_multicast(flow.dl_src)) {
         struct mac_source *src;
         struct list *bucket;
         bool found;
@@ -536,7 +469,7 @@ process_switch(struct switch_ *sw, struct ofp_packet_in *opi)
         found = false;
         LIST_FOR_EACH (src, struct mac_source, hash_list, bucket) {
             if (src->datapath_id == sw->datapath_id
-                && mac_equals(src->mac, flow.dl_src)) {
+                && eth_addr_equals(src->mac, flow.dl_src)) {
                 found = true;
                 break;
             }
@@ -571,24 +504,26 @@ process_switch(struct switch_ *sw, struct ofp_packet_in *opi)
 
         if (ntohs(flow.in_port) != src->port) {
             src->port = ntohs(flow.in_port);
-            VLOG_DBG("learned that "MAC_FMT" is on datapath %"PRIx64" port %d",
-                     MAC_ARGS(src->mac), ntohll(src->datapath_id),
+            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 "MAC_FMT, MAC_ARGS(flow.dl_src));
+        VLOG_DBG("multicast packet source "ETH_ADDR_FMT,
+                 ETH_ADDR_ARGS(flow.dl_src));
     }
 
     /* Figure out the destination. */
     out_port = OFPP_FLOOD;
-    if (!mac_is_multicast(flow.dl_dst)) {
+    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
-                && mac_equals(dst->mac, flow.dl_dst)) {
+                && eth_addr_equals(dst->mac, flow.dl_dst)) {
                 out_port = dst->port;
                 break;
             }
@@ -627,11 +562,7 @@ parse_options(int argc, char *argv[])
         {"verbose",     optional_argument, 0, 'v'},
         {"help",        no_argument, 0, 'h'},
         {"version",     no_argument, 0, 'V'},
-#ifdef HAVE_OPENSSL
-        {"private-key", required_argument, 0, 'p'},
-        {"certificate", required_argument, 0, 'c'},
-        {"ca-cert",     required_argument, 0, 'C'},
-#endif
+        VCONN_SSL_LONG_OPTIONS
         {0, 0, 0, 0},
     };
     char *short_options = long_options_to_short_options(long_options);
@@ -665,19 +596,7 @@ parse_options(int argc, char *argv[])
             vlog_set_verbosity(optarg);
             break;
 
-#ifdef HAVE_OPENSSL
-        case 'p':
-            vconn_ssl_set_private_key_file(optarg);
-            break;
-
-        case 'c':
-            vconn_ssl_set_certificate_file(optarg);
-            break;
-
-        case 'C':
-            vconn_ssl_set_ca_cert_file(optarg);
-            break;
-#endif
+        VCONN_SSL_OPTION_HANDLERS
 
         case '?':
             exit(EXIT_FAILURE);
@@ -693,21 +612,10 @@ static void
 usage(void)
 {
     printf("%s: OpenFlow controller\n"
-           "usage: %s [OPTIONS] VCONN\n"
-           "where VCONN is one of the following:\n"
-           "  ptcp:[PORT]             listen to TCP PORT (default: %d)\n",
-           program_name, program_name, OFP_TCP_PORT);
-#ifdef HAVE_NETLINK
-    printf("  nl:DP_IDX               via netlink to local datapath DP_IDX\n");
-#endif
-#ifdef HAVE_OPENSSL
-    printf("  pssl:[PORT]             listen for SSL on PORT (default: %d)\n"
-           "\nPKI configuration (required to use SSL):\n"
-           "  -p, --private-key=FILE  file with private key\n"
-           "  -c, --certificate=FILE  file with certificate for private key\n"
-           "  -C, --ca-cert=FILE      file with peer CA certificate\n",
-           OFP_SSL_PORT);
-#endif
+           "usage: %s [OPTIONS] METHOD\n"
+           "where METHOD is any OpenFlow connection method.\n",
+           program_name, program_name);
+    vconn_usage(true, true);
     printf("\nOther options:\n"
            "  -H, --hub               act as hub instead of learning switch\n"
            "  -n, --noflow            pass traffic, but don't add flows\n"