netdev-dpdk: fix mbuf leaks
[cascardo/ovs.git] / lib / learning-switch.c
index 426cb37..19a90db 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -25,6 +25,7 @@
 
 #include "byte-order.h"
 #include "classifier.h"
+#include "dp-packet.h"
 #include "flow.h"
 #include "hmap.h"
 #include "mac-learning.h"
@@ -41,8 +42,8 @@
 #include "shash.h"
 #include "simap.h"
 #include "timeval.h"
-#include "vconn.h"
-#include "vlog.h"
+#include "openvswitch/vconn.h"
+#include "openvswitch/vlog.h"
 
 VLOG_DEFINE_THIS_MODULE(learning_switch);
 
@@ -89,6 +90,7 @@ struct lswitch {
      * to set up the flow table. */
     const struct ofputil_flow_mod *default_flows;
     size_t n_default_flows;
+    enum ofputil_protocol usable_protocols;
 };
 
 /* The log messages here could actually be useful in debugging, so keep the
@@ -104,6 +106,13 @@ static enum ofperr process_switch_features(struct lswitch *,
 static void process_packet_in(struct lswitch *, const struct ofp_header *);
 static void process_echo_request(struct lswitch *, const struct ofp_header *);
 
+static ofp_port_t get_mac_entry_ofp_port(const struct mac_learning *ml,
+                                         const struct mac_entry *)
+    OVS_REQ_RDLOCK(ml->rwlock);
+static void set_mac_entry_ofp_port(struct mac_learning *ml,
+                                   struct mac_entry *, ofp_port_t)
+    OVS_REQ_WRLOCK(ml->rwlock);
+
 /* Creates and returns a new learning switch whose configuration is given by
  * 'cfg'.
  *
@@ -161,6 +170,7 @@ lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
 
     sw->default_flows = cfg->default_flows;
     sw->n_default_flows = cfg->n_default_flows;
+    sw->usable_protocols = cfg->usable_protocols;
 
     sw->queued = rconn_packet_counter_create();
 
@@ -171,12 +181,44 @@ static void
 lswitch_handshake(struct lswitch *sw)
 {
     enum ofputil_protocol protocol;
+    enum ofp_version version;
 
     send_features_request(sw);
 
-    protocol = ofputil_protocol_from_ofp_version(rconn_get_version(sw->rconn));
+    version = rconn_get_version(sw->rconn);
+    protocol = ofputil_protocol_from_ofp_version(version);
+    if (version >= OFP13_VERSION) {
+        /* OpenFlow 1.3 and later by default drop packets that miss in the flow
+         * table.  Set up a flow to send packets to the controller by
+         * default. */
+        struct ofpact_output output;
+        struct ofpbuf *msg;
+        int error;
+
+        ofpact_init_OUTPUT(&output);
+        output.port = OFPP_CONTROLLER;
+        output.max_len = OFP_DEFAULT_MISS_SEND_LEN;
+
+        struct ofputil_flow_mod fm = {
+            .match = MATCH_CATCHALL_INITIALIZER,
+            .priority = 0,
+            .table_id = 0,
+            .command = OFPFC_ADD,
+            .buffer_id = UINT32_MAX,
+            .out_port = OFPP_NONE,
+            .out_group = OFPG_ANY,
+            .ofpacts = &output.ofpact,
+            .ofpacts_len = sizeof output,
+        };
+
+        msg = ofputil_encode_flow_mod(&fm, protocol);
+        error = rconn_send(sw->rconn, msg, NULL);
+        if (error) {
+            VLOG_INFO_RL(&rl, "%s: failed to add default flow (%s)",
+                         rconn_get_name(sw->rconn), ovs_strerror(error));
+        }
+    }
     if (sw->default_flows) {
-        enum ofputil_protocol usable_protocols;
         struct ofpbuf *msg = NULL;
         int error = 0;
         size_t i;
@@ -188,10 +230,8 @@ lswitch_handshake(struct lswitch *sw)
          * This could be improved by actually negotiating a mutually acceptable
          * flow format with the switch, but that would require an asynchronous
          * state machine.  This version ought to work fine in practice. */
-        usable_protocols = ofputil_flow_mod_usable_protocols(
-            sw->default_flows, sw->n_default_flows);
-        if (!(protocol & usable_protocols)) {
-            enum ofputil_protocol want = rightmost_1bit(usable_protocols);
+        if (!(protocol & sw->usable_protocols)) {
+            enum ofputil_protocol want = rightmost_1bit(sw->usable_protocols);
             while (!error) {
                 msg = ofputil_encode_set_protocol(protocol, want, &protocol);
                 if (!msg) {
@@ -200,7 +240,7 @@ lswitch_handshake(struct lswitch *sw)
                 error = rconn_send(sw->rconn, msg, NULL);
             }
         }
-        if (protocol & usable_protocols) {
+        if (protocol & sw->usable_protocols) {
             for (i = 0; !error && i < sw->n_default_flows; i++) {
                 msg = ofputil_encode_flow_mod(&sw->default_flows[i], protocol);
                 error = rconn_send(sw->rconn, msg, NULL);
@@ -208,7 +248,7 @@ lswitch_handshake(struct lswitch *sw)
 
             if (error) {
                 VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)",
-                             rconn_get_name(sw->rconn), strerror(error));
+                             rconn_get_name(sw->rconn), ovs_strerror(error));
             }
         } else {
             VLOG_INFO_RL(&rl, "%s: failed to set usable protocol",
@@ -237,7 +277,7 @@ lswitch_destroy(struct lswitch *sw)
             free(node);
         }
         shash_destroy(&sw->queue_names);
-        mac_learning_destroy(sw->ml);
+        mac_learning_unref(sw->ml);
         rconn_packet_counter_destroy(sw->queued);
         free(sw);
     }
@@ -251,7 +291,9 @@ lswitch_run(struct lswitch *sw)
     int i;
 
     if (sw->ml) {
-        mac_learning_run(sw->ml, NULL);
+        ovs_rwlock_wrlock(&sw->ml->rwlock);
+        mac_learning_run(sw->ml);
+        ovs_rwlock_unlock(&sw->ml->rwlock);
     }
 
     rconn_run(sw->rconn);
@@ -283,7 +325,9 @@ void
 lswitch_wait(struct lswitch *sw)
 {
     if (sw->ml) {
+        ovs_rwlock_rdlock(&sw->ml->rwlock);
         mac_learning_wait(sw->ml);
+        ovs_rwlock_unlock(&sw->ml->rwlock);
     }
     rconn_run_wait(sw->rconn);
     rconn_recv_wait(sw->rconn);
@@ -310,12 +354,9 @@ lswitch_process_packet(struct lswitch *sw, const struct ofpbuf *msg)
         return;
     }
 
-    switch (type) {
-    case OFPTYPE_ECHO_REQUEST:
+    if (type == OFPTYPE_ECHO_REQUEST) {
         process_echo_request(sw, msg->data);
-        break;
-
-    case OFPTYPE_FEATURES_REPLY:
+    } else if (type == OFPTYPE_FEATURES_REPLY) {
         if (sw->state == S_FEATURES_REPLY) {
             if (!process_switch_features(sw, msg->data)) {
                 sw->state = S_SWITCHING;
@@ -323,82 +364,15 @@ lswitch_process_packet(struct lswitch *sw, const struct ofpbuf *msg)
                 rconn_disconnect(sw->rconn);
             }
         }
-        break;
-
-    case OFPTYPE_PACKET_IN:
+    } else if (type == OFPTYPE_PACKET_IN) {
         process_packet_in(sw, msg->data);
-        break;
-
-    case OFPTYPE_FLOW_REMOVED:
+    } else if (type == OFPTYPE_FLOW_REMOVED) {
         /* Nothing to do. */
-        break;
-
-    case OFPTYPE_HELLO:
-    case OFPTYPE_ERROR:
-    case OFPTYPE_ECHO_REPLY:
-    case OFPTYPE_FEATURES_REQUEST:
-    case OFPTYPE_GET_CONFIG_REQUEST:
-    case OFPTYPE_GET_CONFIG_REPLY:
-    case OFPTYPE_SET_CONFIG:
-    case OFPTYPE_PORT_STATUS:
-    case OFPTYPE_PACKET_OUT:
-    case OFPTYPE_FLOW_MOD:
-    case OFPTYPE_PORT_MOD:
-    case OFPTYPE_BARRIER_REQUEST:
-    case OFPTYPE_BARRIER_REPLY:
-    case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
-    case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
-    case OFPTYPE_DESC_STATS_REQUEST:
-    case OFPTYPE_DESC_STATS_REPLY:
-    case OFPTYPE_FLOW_STATS_REQUEST:
-    case OFPTYPE_FLOW_STATS_REPLY:
-    case OFPTYPE_AGGREGATE_STATS_REQUEST:
-    case OFPTYPE_AGGREGATE_STATS_REPLY:
-    case OFPTYPE_TABLE_STATS_REQUEST:
-    case OFPTYPE_TABLE_STATS_REPLY:
-    case OFPTYPE_PORT_STATS_REQUEST:
-    case OFPTYPE_PORT_STATS_REPLY:
-    case OFPTYPE_QUEUE_STATS_REQUEST:
-    case OFPTYPE_QUEUE_STATS_REPLY:
-    case OFPTYPE_PORT_DESC_STATS_REQUEST:
-    case OFPTYPE_PORT_DESC_STATS_REPLY:
-    case OFPTYPE_ROLE_REQUEST:
-    case OFPTYPE_ROLE_REPLY:
-    case OFPTYPE_SET_FLOW_FORMAT:
-    case OFPTYPE_FLOW_MOD_TABLE_ID:
-    case OFPTYPE_SET_PACKET_IN_FORMAT:
-    case OFPTYPE_FLOW_AGE:
-    case OFPTYPE_SET_CONTROLLER_ID:
-    case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
-    case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
-    case OFPTYPE_FLOW_MONITOR_CANCEL:
-    case OFPTYPE_FLOW_MONITOR_PAUSED:
-    case OFPTYPE_FLOW_MONITOR_RESUMED:
-    case OFPTYPE_GET_ASYNC_REQUEST:
-    case OFPTYPE_GET_ASYNC_REPLY:
-    case OFPTYPE_SET_ASYNC_CONFIG:
-    case OFPTYPE_METER_MOD:
-    case OFPTYPE_GROUP_REQUEST:
-    case OFPTYPE_GROUP_REPLY:
-    case OFPTYPE_GROUP_DESC_REQUEST:
-    case OFPTYPE_GROUP_DESC_REPLY:
-    case OFPTYPE_GROUP_FEATURES_REQUEST:
-    case OFPTYPE_GROUP_FEATURES_REPLY:
-    case OFPTYPE_METER_REQUEST:
-    case OFPTYPE_METER_REPLY:
-    case OFPTYPE_METER_CONFIG_REQUEST:
-    case OFPTYPE_METER_CONFIG_REPLY:
-    case OFPTYPE_METER_FEATURES_REQUEST:
-    case OFPTYPE_METER_FEATURES_REPLY:
-    case OFPTYPE_TABLE_FEATURES_REQUEST:
-    case OFPTYPE_TABLE_FEATURES_REPLY:
-    default:
-        if (VLOG_IS_DBG_ENABLED()) {
-            char *s = ofp_to_string(msg->data, msg->size, 2);
-            VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
-                        sw->datapath_id, s);
-            free(s);
-        }
+    } else if (VLOG_IS_DBG_ENABLED()) {
+        char *s = ofp_to_string(msg->data, msg->size, 2);
+        VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
+                    sw->datapath_id, s);
+        free(s);
     }
 }
 \f
@@ -406,7 +380,6 @@ static void
 send_features_request(struct lswitch *sw)
 {
     struct ofpbuf *b;
-    struct ofp_switch_config *osc;
     int ofp_version = rconn_get_version(sw->rconn);
 
     ovs_assert(ofp_version > 0 && ofp_version < 0xff);
@@ -416,10 +389,10 @@ send_features_request(struct lswitch *sw)
     queue_tx(sw, b);
 
     /* Send OFPT_SET_CONFIG. */
-    b = ofpraw_alloc(OFPRAW_OFPT_SET_CONFIG, ofp_version, sizeof *osc);
-    osc = ofpbuf_put_zeros(b, sizeof *osc);
-    osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
-    queue_tx(sw, b);
+    struct ofputil_switch_config config = {
+        .miss_send_len = OFP_DEFAULT_MISS_SEND_LEN
+    };
+    queue_tx(sw, ofputil_encode_set_config(&config, ofp_version));
 }
 
 static void
@@ -433,7 +406,7 @@ queue_tx(struct lswitch *sw, struct ofpbuf *b)
         } else {
             VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
                          sw->datapath_id, rconn_get_name(sw->rconn),
-                         strerror(retval));
+                         ovs_strerror(retval));
         }
     }
 }
@@ -443,10 +416,9 @@ process_switch_features(struct lswitch *sw, struct ofp_header *oh)
 {
     struct ofputil_switch_features features;
     struct ofputil_phy_port port;
-    enum ofperr error;
-    struct ofpbuf b;
 
-    error = ofputil_decode_switch_features(oh, &features, &b);
+    struct ofpbuf b = ofpbuf_const_initializer(oh, ntohs(oh->length));
+    enum ofperr error = ofputil_pull_switch_features(&b, &features);
     if (error) {
         VLOG_ERR("received invalid switch feature reply (%s)",
                  ofperr_to_string(error));
@@ -472,17 +444,22 @@ lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
     ofp_port_t out_port;
 
     /* Learn the source MAC. */
-    if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
-        struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, 0);
-        if (mac_entry_is_new(mac)
-            || mac->port.ofp_port != flow->in_port.ofp_port) {
-            VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
-                        "port %"PRIu16, sw->datapath_id,
-                        ETH_ADDR_ARGS(flow->dl_src), flow->in_port.ofp_port);
-
-            mac->port.ofp_port = flow->in_port.ofp_port;
-            mac_learning_changed(sw->ml, mac);
+    if (sw->ml) {
+        ovs_rwlock_wrlock(&sw->ml->rwlock);
+        if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
+            struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src,
+                                                        0);
+            if (get_mac_entry_ofp_port(sw->ml, mac)
+                != flow->in_port.ofp_port) {
+                VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
+                            "port %"PRIu16, sw->datapath_id,
+                            ETH_ADDR_ARGS(flow->dl_src),
+                            flow->in_port.ofp_port);
+
+                set_mac_entry_ofp_port(sw->ml, mac, flow->in_port.ofp_port);
+            }
         }
+        ovs_rwlock_unlock(&sw->ml->rwlock);
     }
 
     /* Drop frames for reserved multicast addresses. */
@@ -494,14 +471,17 @@ lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
     if (sw->ml) {
         struct mac_entry *mac;
 
-        mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
+        ovs_rwlock_rdlock(&sw->ml->rwlock);
+        mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0);
         if (mac) {
-            out_port = mac->port.ofp_port;
+            out_port = get_mac_entry_ofp_port(sw->ml, mac);
             if (out_port == flow->in_port.ofp_port) {
                 /* Don't send a packet back out its input port. */
+                ovs_rwlock_unlock(&sw->ml->rwlock);
                 return OFPP_NONE;
             }
         }
+        ovs_rwlock_unlock(&sw->ml->rwlock);
     }
 
     /* Check if we need to use "NORMAL" action. */
@@ -531,6 +511,7 @@ static void
 process_packet_in(struct lswitch *sw, const struct ofp_header *oh)
 {
     struct ofputil_packet_in pi;
+    uint32_t buffer_id;
     uint32_t queue_id;
     ofp_port_t out_port;
 
@@ -540,11 +521,10 @@ process_packet_in(struct lswitch *sw, const struct ofp_header *oh)
     struct ofputil_packet_out po;
     enum ofperr error;
 
-    struct ofpbuf pkt;
+    struct dp_packet pkt;
     struct flow flow;
-    union flow_in_port in_port_;
 
-    error = ofputil_decode_packet_in(&pi, oh);
+    error = ofputil_decode_packet_in(oh, true, &pi, NULL, &buffer_id, NULL);
     if (error) {
         VLOG_WARN_RL(&rl, "failed to decode packet-in: %s",
                      ofperr_to_string(error));
@@ -558,17 +538,17 @@ process_packet_in(struct lswitch *sw, const struct ofp_header *oh)
         return;
     }
 
-    /* Extract flow data from 'opi' into 'flow'. */
-    ofpbuf_use_const(&pkt, pi.packet, pi.packet_len);
-    in_port_.ofp_port = pi.fmd.in_port;
-    flow_extract(&pkt, 0, 0, NULL, &in_port_, &flow);
-    flow.tunnel.tun_id = pi.fmd.tun_id;
+    /* Extract flow data from 'pi' into 'flow'. */
+    dp_packet_use_const(&pkt, pi.packet, pi.packet_len);
+    flow_extract(&pkt, &flow);
+    flow.in_port.ofp_port = pi.flow_metadata.flow.in_port.ofp_port;
+    flow.tunnel.tun_id = pi.flow_metadata.flow.tunnel.tun_id;
 
     /* Choose output port. */
     out_port = lswitch_choose_destination(sw, &flow);
 
     /* Make actions. */
-    queue_id = get_queue_id(sw, pi.fmd.in_port);
+    queue_id = get_queue_id(sw, pi.flow_metadata.flow.in_port.ofp_port);
     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
     if (out_port == OFPP_NONE) {
         /* No actions. */
@@ -580,51 +560,49 @@ process_packet_in(struct lswitch *sw, const struct ofp_header *oh)
         enqueue->port = out_port;
         enqueue->queue = queue_id;
     }
-    ofpact_pad(&ofpacts);
 
     /* Prepare packet_out in case we need one. */
-    po.buffer_id = pi.buffer_id;
-    if (po.buffer_id == UINT32_MAX) {
-        po.packet = pkt.data;
-        po.packet_len = pkt.size;
+    po.buffer_id = buffer_id;
+    if (buffer_id == UINT32_MAX) {
+        po.packet = dp_packet_data(&pkt);
+        po.packet_len = dp_packet_size(&pkt);
     } else {
         po.packet = NULL;
         po.packet_len = 0;
     }
-    po.in_port = pi.fmd.in_port;
+    po.in_port = pi.flow_metadata.flow.in_port.ofp_port;
     po.ofpacts = ofpacts.data;
     po.ofpacts_len = ofpacts.size;
 
     /* Send the packet, and possibly the whole flow, to the output port. */
     if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
-        struct ofputil_flow_mod fm;
-        struct ofpbuf *buffer;
-
         /* The output port is known, or we always flood everything, so add a
          * new flow. */
-        memset(&fm, 0, sizeof fm);
+        struct ofputil_flow_mod fm = {
+            .priority = 1, /* Must be > 0 because of table-miss flow entry. */
+            .table_id = 0xff,
+            .command = OFPFC_ADD,
+            .idle_timeout = sw->max_idle,
+            .buffer_id = buffer_id,
+            .out_port = OFPP_NONE,
+            .ofpacts = ofpacts.data,
+            .ofpacts_len = ofpacts.size,
+        };
         match_init(&fm.match, &flow, &sw->wc);
         ofputil_normalize_match_quiet(&fm.match);
-        fm.priority = 0;
-        fm.table_id = 0xff;
-        fm.command = OFPFC_ADD;
-        fm.idle_timeout = sw->max_idle;
-        fm.buffer_id = pi.buffer_id;
-        fm.out_port = OFPP_NONE;
-        fm.ofpacts = ofpacts.data;
-        fm.ofpacts_len = ofpacts.size;
-        buffer = ofputil_encode_flow_mod(&fm, sw->protocol);
+
+        struct ofpbuf *buffer = ofputil_encode_flow_mod(&fm, sw->protocol);
 
         queue_tx(sw, buffer);
 
         /* If the switch didn't buffer the packet, we need to send a copy. */
-        if (pi.buffer_id == UINT32_MAX && out_port != OFPP_NONE) {
+        if (buffer_id == UINT32_MAX && out_port != OFPP_NONE) {
             queue_tx(sw, ofputil_encode_packet_out(&po, sw->protocol));
         }
     } else {
         /* We don't know that MAC, or we don't set up flows.  Send along the
          * packet without setting up a flow. */
-        if (pi.buffer_id != UINT32_MAX || out_port != OFPP_NONE) {
+        if (buffer_id != UINT32_MAX || out_port != OFPP_NONE) {
             queue_tx(sw, ofputil_encode_packet_out(&po, sw->protocol));
         }
     }
@@ -635,3 +613,20 @@ process_echo_request(struct lswitch *sw, const struct ofp_header *rq)
 {
     queue_tx(sw, make_echo_reply(rq));
 }
+
+static ofp_port_t
+get_mac_entry_ofp_port(const struct mac_learning *ml,
+                       const struct mac_entry *e)
+    OVS_REQ_RDLOCK(ml->rwlock)
+{
+    void *port = mac_entry_get_port(ml, e);
+    return (OVS_FORCE ofp_port_t) (uintptr_t) port;
+}
+
+static void
+set_mac_entry_ofp_port(struct mac_learning *ml,
+                       struct mac_entry *e, ofp_port_t ofp_port)
+    OVS_REQ_WRLOCK(ml->rwlock)
+{
+    mac_entry_set_port(ml, e, (void *) (OVS_FORCE uintptr_t) ofp_port);
+}