datapath: Replace "struct odp_action" by Netlink attributes.
[openvswitch] / vswitchd / bridge.c
index 0103eb99bd89a4b9d1b20dfb95511e73902bd2e6..8ade873faf5133c7043374c7fb1dbef7ddf60b78 100644 (file)
@@ -45,6 +45,7 @@
 #include "list.h"
 #include "mac-learning.h"
 #include "netdev.h"
+#include "netlink.h"
 #include "odp-util.h"
 #include "ofp-print.h"
 #include "ofpbuf.h"
 
 VLOG_DEFINE_THIS_MODULE(bridge);
 
+COVERAGE_DEFINE(bridge_flush);
+COVERAGE_DEFINE(bridge_process_flow);
+COVERAGE_DEFINE(bridge_reconfigure);
+
 struct dst {
     uint16_t vlan;
     uint16_t dp_ifidx;
@@ -372,10 +377,7 @@ bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
 
 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
  * keys from "options" and "other_config", preferring "options" keys over
- * "other_config" keys.
- *
- * The value strings in '*options' are taken directly from if_cfg, not copied,
- * so the caller should not modify or free them. */
+ * "other_config" keys. */
 static void
 iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
 {
@@ -397,68 +399,6 @@ iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
     }
 }
 
-/* Returns the type of network device that 'iface' should have.  (This is
- * ordinarily the same type as the interface, but the network devices for
- * "internal" ports have type "system".) */
-static const char *
-iface_get_netdev_type(const struct iface *iface)
-{
-    return !strcmp(iface->type, "internal") ? "system" : iface->type;
-}
-
-/* Attempt to create the network device for 'iface' through the netdev
- * library. */
-static int
-create_iface_netdev(struct iface *iface)
-{
-    struct netdev_options netdev_options;
-    struct shash options;
-    int error;
-
-    memset(&netdev_options, 0, sizeof netdev_options);
-    netdev_options.name = iface->cfg->name;
-    netdev_options.type = iface_get_netdev_type(iface);
-    netdev_options.args = &options;
-    netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
-
-    iface_get_options(iface->cfg, &options);
-
-    error = netdev_open(&netdev_options, &iface->netdev);
-
-    if (iface->netdev) {
-        iface->enabled = netdev_get_carrier(iface->netdev);
-    }
-
-    shash_destroy(&options);
-
-    return error;
-}
-
-static int
-reconfigure_iface_netdev(struct iface *iface)
-{
-    const char *netdev_type, *iface_type;
-    struct shash options;
-    int error;
-
-    /* Skip reconfiguration if the device has the wrong type. This shouldn't
-     * happen, but... */
-    iface_type = iface_get_netdev_type(iface);
-    netdev_type = netdev_get_type(iface->netdev);
-    if (iface_type && strcmp(netdev_type, iface_type)) {
-        VLOG_WARN("%s: attempting change device type from %s to %s",
-                  iface->cfg->name, netdev_type, iface_type);
-        return EINVAL;
-    }
-
-    /* Reconfigure device. */
-    iface_get_options(iface->cfg, &options);
-    error = netdev_reconfigure(iface->netdev, &options);
-    shash_destroy(&options);
-
-    return error;
-}
-
 /* Callback for iterate_and_prune_ifaces(). */
 static bool
 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
@@ -697,18 +637,16 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
         SHASH_FOR_EACH (node, &want_ifaces) {
             const char *if_name = node->name;
             struct iface *iface = node->data;
-            bool internal = !iface || !strcmp(iface->type, "internal");
             struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
+            const char *type = iface ? iface->type : "internal";
             int error;
 
             /* If we have a port or a netdev already, and it's not the type we
              * want, then delete the port (if any) and close the netdev (if
              * any). */
-            if (internal
-                ? dpif_port && !(dpif_port->flags & ODP_PORT_INTERNAL)
-                : (iface->netdev
-                   && strcmp(iface->type, netdev_get_type(iface->netdev))))
-            {
+            if ((dpif_port && strcmp(dpif_port->type, type))
+                || (iface && iface->netdev
+                    && strcmp(type, netdev_get_type(iface->netdev)))) {
                 if (dpif_port) {
                     error = ofproto_port_del(br->ofproto, dpif_port->port);
                     if (error) {
@@ -722,48 +660,62 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
                 }
             }
 
-            /* If it's not an internal port, open (possibly create) the
-             * netdev. */
-            if (!internal) {
-                if (!iface->netdev) {
-                    error = create_iface_netdev(iface);
-                    if (error) {
-                        VLOG_WARN("could not create iface %s: %s", iface->name,
-                                  strerror(error));
-                        continue;
-                    }
-                } else {
-                    reconfigure_iface_netdev(iface);
+            /* If the port doesn't exist or we don't have the netdev open,
+             * we need to do more work. */
+            if (!dpif_port || (iface && !iface->netdev)) {
+                struct netdev_options options;
+                struct netdev *netdev;
+                struct shash args;
+
+                /* First open the network device. */
+                options.name = if_name;
+                options.type = type;
+                options.args = &args;
+                options.ethertype = NETDEV_ETH_TYPE_NONE;
+
+                shash_init(&args);
+                if (iface) {
+                    iface_get_options(iface->cfg, &args);
                 }
-            }
+                error = netdev_open(&options, &netdev);
+                shash_destroy(&args);
 
-            /* If it's not part of the datapath, add it. */
-            if (!dpif_port) {
-                error = dpif_port_add(br->dpif, if_name,
-                                      internal ? ODP_PORT_INTERNAL : 0, NULL);
-                if (error == EFBIG) {
-                    VLOG_ERR("ran out of valid port numbers on %s",
-                             dpif_name(br->dpif));
-                    break;
-                } else if (error) {
-                    VLOG_ERR("failed to add %s interface to %s: %s",
-                             if_name, dpif_name(br->dpif), strerror(error));
+                if (error) {
+                    VLOG_WARN("could not open network device %s (%s)",
+                              if_name, strerror(error));
                     continue;
                 }
-            }
 
-            /* If it's an internal port, open the netdev. */
-            if (internal) {
-                if (iface && !iface->netdev) {
-                    error = create_iface_netdev(iface);
+                /* Then add the port if we haven't already. */
+                if (!dpif_port) {
+                    error = dpif_port_add(br->dpif, netdev, NULL);
                     if (error) {
-                        VLOG_WARN("could not create iface %s: %s", iface->name,
-                                  strerror(error));
-                        continue;
+                        netdev_close(netdev);
+                        if (error == EFBIG) {
+                            VLOG_ERR("ran out of valid port numbers on %s",
+                                     dpif_name(br->dpif));
+                            break;
+                        } else {
+                            VLOG_ERR("failed to add %s interface to %s: %s",
+                                     if_name, dpif_name(br->dpif),
+                                     strerror(error));
+                            continue;
+                        }
                     }
                 }
-            } else {
-                assert(iface->netdev != NULL);
+
+                /* Update 'iface'. */
+                if (iface) {
+                    iface->netdev = netdev;
+                    iface->enabled = netdev_get_carrier(iface->netdev);
+                }
+            } else if (iface && iface->netdev) {
+                struct shash args;
+
+                shash_init(&args);
+                iface_get_options(iface->cfg, &args);
+                netdev_reconfigure(iface->netdev, &args);
+                shash_destroy(&args);
             }
         }
         free(dpif_ports);
@@ -1994,15 +1946,16 @@ bridge_fetch_dp_ifaces(struct bridge *br)
 /* Bridge packet processing functions. */
 
 static int
-bond_hash(const uint8_t mac[ETH_ADDR_LEN])
+bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
 {
-    return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
+    return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
 }
 
 static struct bond_entry *
-lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
+lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
+                  uint16_t vlan)
 {
-    return &port->bond_hash[bond_hash(mac)];
+    return &port->bond_hash[bond_hash(mac, vlan)];
 }
 
 static int
@@ -2037,7 +1990,7 @@ bond_choose_iface(const struct port *port)
 
 static bool
 choose_output_iface(const struct port *port, const uint8_t *dl_src,
-                    uint16_t *dp_ifidx, tag_type *tags)
+                    uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
 {
     struct iface *iface;
 
@@ -2045,7 +1998,7 @@ choose_output_iface(const struct port *port, const uint8_t *dl_src,
     if (port->n_ifaces == 1) {
         iface = port->ifaces[0];
     } else {
-        struct bond_entry *e = lookup_bond_entry(port, dl_src);
+        struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
         if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
             || !port->ifaces[e->iface_idx]->enabled) {
             /* XXX select interface properly.  The current interface selection
@@ -2282,7 +2235,8 @@ set_dst(struct dst *p, const struct flow *flow,
               : in_port->vlan >= 0 ? in_port->vlan
               : flow->vlan_tci == 0 ? OFP_VLAN_NONE
               : vlan_tci_to_vid(flow->vlan_tci));
-    return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
+    return choose_output_iface(out_port, flow->dl_src, p->vlan,
+                               &p->dp_ifidx, tags);
 }
 
 static void
@@ -2471,7 +2425,7 @@ print_dsts(const struct dst *dsts, size_t n)
 static void
 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
                 const struct port *in_port, const struct port *out_port,
-                tag_type *tags, struct odp_actions *actions,
+                tag_type *tags, struct ofpbuf *actions,
                 uint16_t *nf_output_iface)
 {
     struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
@@ -2487,19 +2441,18 @@ compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
         cur_vlan = OFP_VLAN_NONE;
     }
     for (p = dsts; p < &dsts[n_dsts]; p++) {
-        union odp_action *a;
         if (p->vlan != cur_vlan) {
             if (p->vlan == OFP_VLAN_NONE) {
-                odp_actions_add(actions, ODPAT_STRIP_VLAN);
+                nl_msg_put_flag(actions, ODPAT_STRIP_VLAN);
             } else {
-                a = odp_actions_add(actions, ODPAT_SET_DL_TCI);
-                a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
-                a->dl_tci.tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
+                ovs_be16 tci;
+                tci = htons(p->vlan & VLAN_VID_MASK);
+                tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
+                nl_msg_put_be16(actions, ODPAT_SET_DL_TCI, tci);
             }
             cur_vlan = p->vlan;
         }
-        a = odp_actions_add(actions, ODPAT_OUTPUT);
-        a->output.port = p->dp_ifidx;
+        nl_msg_put_u32(actions, ODPAT_OUTPUT, p->dp_ifidx);
     }
 }
 
@@ -2692,7 +2645,7 @@ is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
  * not at all, if 'packet' was NULL. */
 static bool
 process_flow(struct bridge *br, const struct flow *flow,
-             const struct ofpbuf *packet, struct odp_actions *actions,
+             const struct ofpbuf *packet, struct ofpbuf *actions,
              tag_type *tags, uint16_t *nf_output_iface)
 {
     struct port *in_port;
@@ -2743,7 +2696,7 @@ done:
 
 static bool
 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
-                        struct odp_actions *actions, tag_type *tags,
+                        struct ofpbuf *actions, tag_type *tags,
                         uint16_t *nf_output_iface, void *br_)
 {
     struct iface *iface;
@@ -2765,14 +2718,15 @@ bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
 
 static void
 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
-                              const union odp_action *actions,
-                              size_t n_actions, unsigned long long int n_bytes,
-                              void *br_)
+                              const struct nlattr *actions,
+                              unsigned int actions_len,
+                              unsigned long long int n_bytes, void *br_)
 {
     struct bridge *br = br_;
-    const union odp_action *a;
+    const struct nlattr *a;
     struct port *in_port;
     tag_type dummy = 0;
+    unsigned int left;
     int vlan;
 
     /* Feed information from the active flows back into the learning table to
@@ -2790,12 +2744,15 @@ bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
     if (!br->has_bonded_ports) {
         return;
     }
-    for (a = actions; a < &actions[n_actions]; a++) {
-        if (a->type == ODPAT_OUTPUT) {
-            struct port *out_port = port_from_dp_ifidx(br, a->output.port);
+    NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
+        if (nl_attr_type(a) == ODPAT_OUTPUT) {
+            struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
             if (out_port && out_port->n_ifaces >= 2) {
+                uint16_t vlan = (flow->vlan_tci
+                                 ? vlan_tci_to_vid(flow->vlan_tci)
+                                 : OFP_VLAN_NONE);
                 struct bond_entry *e = lookup_bond_entry(out_port,
-                                                         flow->dl_src);
+                                                         flow->dl_src, vlan);
                 e->tx_bytes += n_bytes;
             }
         }
@@ -3149,7 +3106,7 @@ bond_send_learning_packets(struct port *port)
         int retval;
 
         if (e->port == port->port_idx
-            || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
+            || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
             continue;
         }
 
@@ -3295,9 +3252,10 @@ bond_unixctl_show(struct unixctl_conn *conn,
             LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
                 uint16_t dp_ifidx;
                 tag_type tags = 0;
-                if (bond_hash(me->mac) == hash
+                if (bond_hash(me->mac, me->vlan) == hash
                     && me->port != port->port_idx
-                    && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
+                    && choose_output_iface(port, me->mac, me->vlan,
+                                           &dp_ifidx, &tags)
                     && dp_ifidx == iface->dp_ifidx)
                 {
                     ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
@@ -3317,7 +3275,6 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
     char *args = (char *) args_;
     char *save_ptr = NULL;
     char *bond_s, *hash_s, *slave_s;
-    uint8_t mac[ETH_ADDR_LEN];
     struct port *port;
     struct iface *iface;
     struct bond_entry *entry;
@@ -3338,10 +3295,7 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
         return;
     }
 
-    if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
-        == ETH_ADDR_SCAN_COUNT) {
-        hash = bond_hash(mac);
-    } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
+    if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
         hash = atoi(hash_s) & BOND_MASK;
     } else {
         unixctl_command_reply(conn, 501, "bad hash");
@@ -3463,23 +3417,39 @@ bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
 }
 
 static void
-bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
+bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
                   void *aux OVS_UNUSED)
 {
-       uint8_t mac[ETH_ADDR_LEN];
-       uint8_t hash;
-       char *hash_cstr;
-
-       if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
-           == ETH_ADDR_SCAN_COUNT) {
-               hash = bond_hash(mac);
-
-               hash_cstr = xasprintf("%u", hash);
-               unixctl_command_reply(conn, 200, hash_cstr);
-               free(hash_cstr);
-       } else {
-               unixctl_command_reply(conn, 501, "invalid mac");
-       }
+    char *args = (char *) args_;
+    uint8_t mac[ETH_ADDR_LEN];
+    uint8_t hash;
+    char *hash_cstr;
+    unsigned int vlan;
+    char *mac_s, *vlan_s;
+    char *save_ptr = NULL;
+
+    mac_s  = strtok_r(args, " ", &save_ptr);
+    vlan_s = strtok_r(NULL, " ", &save_ptr);
+
+    if (vlan_s) {
+        if (sscanf(vlan_s, "%u", &vlan) != 1) {
+            unixctl_command_reply(conn, 501, "invalid vlan");
+            return;
+        }
+    } else {
+        vlan = OFP_VLAN_NONE;
+    }
+
+    if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
+        == ETH_ADDR_SCAN_COUNT) {
+        hash = bond_hash(mac, vlan);
+
+        hash_cstr = xasprintf("%u", hash);
+        unixctl_command_reply(conn, 200, hash_cstr);
+        free(hash_cstr);
+    } else {
+        unixctl_command_reply(conn, 501, "invalid mac");
+    }
 }
 
 static void