Factor learning switch out of controller into library.
[sliver-openvswitch.git] / controller / controller.c
index 3a29043..0cc4258 100644 (file)
  * derivatives without specific, written prior permission.
  */
 
-#include <assert.h>
 #include <errno.h>
 #include <getopt.h>
-#include <inttypes.h>
-#include <netinet/in.h>
-#include <poll.h>
 #include <stdlib.h>
 #include <string.h>
-#include <unistd.h>
 
 #include "buffer.h"
 #include "command-line.h"
 #include "compiler.h"
 #include "fault.h"
-#include "flow.h"
-#include "hash.h"
-#include "list.h"
-#include "ofp-print.h"
-#include "mac-learning.h"
-#include "openflow.h"
-#include "packets.h"
+#include "learning-switch.h"
 #include "poll-loop.h"
-#include "queue.h"
-#include "time.h"
+#include "rconn.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? */
@@ -84,28 +65,17 @@ static bool learn_macs = true;
 /* Set up flows?  (If not, every packet is processed at the controller.) */
 static bool setup_flows = true;
 
+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;
 
@@ -123,58 +93,71 @@ 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. */
+    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) {
@@ -183,226 +166,47 @@ 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) 
+new_switch(struct switch_ *sw, struct vconn *vconn, const char *name)
 {
-    if (this) {
-        free(this->name);
-        vconn_close(this->vconn);
-        queue_destroy(&this->txq);
-        mac_learning_destroy(this->ml);
-        free(this);
-    }
+    sw->rconn = rconn_new_from_vconn(name, 128, vconn);
+    sw->lswitch = lswitch_create(sw->rconn, learn_macs, setup_flows);
 }
 
-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_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;
-    }
-}
-
-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