X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=0103eb99bd89a4b9d1b20dfb95511e73902bd2e6;hb=f48667811c04d848dc345aa0bbda29d889b5f960;hp=5210da161ac6d136f419f6a4779d33f7a5285907;hpb=43776b8fa5df28adfe08e7bdc8bffdac252b1e01;p=openvswitch diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 5210da16..0103eb99 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -15,6 +15,7 @@ #include #include "bridge.h" +#include "byte-order.h" #include #include #include @@ -31,6 +32,8 @@ #include #include #include "bitmap.h" +#include "cfm.h" +#include "classifier.h" #include "coverage.h" #include "dirs.h" #include "dpif.h" @@ -65,10 +68,9 @@ #include "vswitchd/vswitch-idl.h" #include "xenserver.h" #include "vlog.h" -#include "xtoxll.h" #include "sflow_api.h" -VLOG_DEFINE_THIS_MODULE(bridge) +VLOG_DEFINE_THIS_MODULE(bridge); struct dst { uint16_t vlan; @@ -89,6 +91,8 @@ struct iface { int dp_ifidx; /* Index within kernel datapath. */ struct netdev *netdev; /* Network device. */ bool enabled; /* May be chosen for flows? */ + const char *type; /* Usually same as cfg->type. */ + struct cfm *cfm; /* Connectivity Fault Management */ const struct ovsrec_interface *cfg; }; @@ -146,6 +150,7 @@ struct port { long long int bond_next_fake_iface_update; /* Time of next update. */ int bond_rebalance_interval; /* Interval between rebalances, in ms. */ long long int bond_next_rebalance; /* Next rebalancing time. */ + struct netdev_monitor *monitor; /* Tracks carrier up/down status. */ /* Port mirroring info. */ mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ @@ -252,9 +257,12 @@ static void iface_destroy(struct iface *); static struct iface *iface_lookup(const struct bridge *, const char *name); static struct iface *iface_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); -static bool iface_is_internal(const struct bridge *, const char *name); static void iface_set_mac(struct iface *); +static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport); static void iface_update_qos(struct iface *, const struct ovsrec_qos *); +static void iface_update_cfm(struct iface *); +static void iface_refresh_cfm_stats(struct iface *iface); +static void iface_send_packet(struct iface *, struct ofpbuf *packet); static void shash_from_ovs_idl_map(char **keys, char **values, size_t n, struct shash *); @@ -271,10 +279,10 @@ void bridge_init(const char *remote) { /* Create connection to database. */ - idl = ovsdb_idl_create(remote, &ovsrec_idl_class); + idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true); - ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg); - ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics); + ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg); + ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics); ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids); ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); @@ -282,8 +290,8 @@ bridge_init(const char *remote) ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids); ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge); - ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport); - ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics); ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids); /* Register unixctl commands. */ @@ -389,6 +397,15 @@ 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 @@ -400,12 +417,7 @@ create_iface_netdev(struct iface *iface) memset(&netdev_options, 0, sizeof netdev_options); netdev_options.name = iface->cfg->name; - if (!strcmp(iface->cfg->type, "internal")) { - /* An "internal" config type maps to a netdev "system" type. */ - netdev_options.type = "system"; - } else { - netdev_options.type = iface->cfg->type; - } + netdev_options.type = iface_get_netdev_type(iface); netdev_options.args = &options; netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; @@ -414,7 +426,7 @@ create_iface_netdev(struct iface *iface) error = netdev_open(&netdev_options, &iface->netdev); if (iface->netdev) { - netdev_get_carrier(iface->netdev, &iface->enabled); + iface->enabled = netdev_get_carrier(iface->netdev); } shash_destroy(&options); @@ -431,9 +443,7 @@ reconfigure_iface_netdev(struct iface *iface) /* Skip reconfiguration if the device has the wrong type. This shouldn't * happen, but... */ - iface_type = (!iface->cfg->type[0] ? NULL - : !strcmp(iface->cfg->type, "internal") ? "system" - : iface->cfg->type); + 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", @@ -449,38 +459,27 @@ reconfigure_iface_netdev(struct iface *iface) return error; } +/* Callback for iterate_and_prune_ifaces(). */ static bool -check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface, - void *aux OVS_UNUSED) +check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED) { if (!iface->netdev) { - int error = create_iface_netdev(iface); - if (error) { - VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name, - strerror(error)); - return false; - } + /* We already reported a related error, don't bother duplicating it. */ + return false; } - return true; -} - -static bool -check_iface_dp_ifidx(struct bridge *br, struct iface *iface, - void *aux OVS_UNUSED) -{ - if (iface->dp_ifidx >= 0) { - VLOG_DBG("%s has interface %s on port %d", - dpif_name(br->dpif), - iface->name, iface->dp_ifidx); - return true; - } else { + if (iface->dp_ifidx < 0) { VLOG_ERR("%s interface not in %s, dropping", iface->name, dpif_name(br->dpif)); return false; } + + VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif), + iface->name, iface->dp_ifidx); + return true; } +/* Callback for iterate_and_prune_ifaces(). */ static bool set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, void *aux OVS_UNUSED) @@ -492,8 +491,7 @@ set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, /* Set MAC address of internal interfaces other than the local * interface. */ - if (iface->dp_ifidx != ODPP_LOCAL - && iface_is_internal(br, iface->name)) { + if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) { iface_set_mac(iface); } @@ -518,6 +516,7 @@ iterate_and_prune_ifaces(struct bridge *br, if (cb(br, iface, aux)) { j++; } else { + iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } } @@ -540,30 +539,52 @@ iterate_and_prune_ifaces(struct bridge *br, * should not be and in fact is not directly involved in that. But * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so * it has to tell in-band control where the managers are to enable that. + * (Thus, only managers connected in-band are collected.) */ static void -collect_managers(const struct ovsrec_open_vswitch *ovs_cfg, - struct sockaddr_in **managersp, size_t *n_managersp) +collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg, + struct sockaddr_in **managersp, size_t *n_managersp) { struct sockaddr_in *managers = NULL; size_t n_managers = 0; + struct shash targets; + size_t i; - if (ovs_cfg->n_managers > 0) { - size_t i; + /* Collect all of the potential targets, as the union of the "managers" + * column and the "targets" columns of the rows pointed to by + * "manager_options", excluding any that are out-of-band. */ + shash_init(&targets); + for (i = 0; i < ovs_cfg->n_managers; i++) { + shash_add_once(&targets, ovs_cfg->managers[i], NULL); + } + for (i = 0; i < ovs_cfg->n_manager_options; i++) { + struct ovsrec_manager *m = ovs_cfg->manager_options[i]; - managers = xmalloc(ovs_cfg->n_managers * sizeof *managers); - for (i = 0; i < ovs_cfg->n_managers; i++) { - const char *name = ovs_cfg->managers[i]; - struct sockaddr_in *sin = &managers[i]; + if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) { + shash_find_and_delete(&targets, m->target); + } else { + shash_add_once(&targets, m->target, NULL); + } + } - if ((!strncmp(name, "tcp:", 4) - && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) || - (!strncmp(name, "ssl:", 4) - && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) { + /* Now extract the targets' IP addresses. */ + if (!shash_is_empty(&targets)) { + struct shash_node *node; + + managers = xmalloc(shash_count(&targets) * sizeof *managers); + SHASH_FOR_EACH (node, &targets) { + const char *target = node->name; + struct sockaddr_in *sin = &managers[n_managers]; + + if ((!strncmp(target, "tcp:", 4) + && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) || + (!strncmp(target, "ssl:", 4) + && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) { n_managers++; } } } + shash_destroy(&targets); *managersp = managers; *n_managersp = n_managers; @@ -582,7 +603,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) COVERAGE_INC(bridge_reconfigure); - collect_managers(ovs_cfg, &managers, &n_managers); + collect_in_band_managers(ovs_cfg, &managers, &n_managers); /* Collect old and new bridges. */ shash_init(&old_br); @@ -666,30 +687,58 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) shash_init(&cur_ifaces); for (i = 0; i < n_dpif_ports; i++) { const char *name = dpif_ports[i].devname; - shash_add_once(&cur_ifaces, name, NULL); + shash_add_once(&cur_ifaces, name, &dpif_ports[i]); } - free(dpif_ports); /* Get the set of interfaces we want on this datapath. */ bridge_get_all_ifaces(br, &want_ifaces); + hmap_clear(&br->ifaces); 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); + 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) { + error = ofproto_port_del(br->ofproto, dpif_port->port); + if (error) { + continue; + } + dpif_port = NULL; + } + if (iface) { + netdev_close(iface->netdev); + iface->netdev = NULL; + } + } - if (shash_find(&cur_ifaces, if_name)) { - /* Already exists on the datapath. If we have it open, - * reconfigure it; otherwise we'll open it later. */ - if (iface && iface->netdev) { + /* 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); } - } else { - /* Need to add to datapath. */ - bool internal; - int error; + } - /* Add to datapath. */ - internal = iface_is_internal(br, if_name); + /* 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) { @@ -699,9 +748,25 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } else if (error) { VLOG_ERR("failed to add %s interface to %s: %s", if_name, dpif_name(br->dpif), strerror(error)); + continue; } } + + /* If it's an internal port, open the netdev. */ + if (internal) { + if (iface && !iface->netdev) { + error = create_iface_netdev(iface); + if (error) { + VLOG_WARN("could not create iface %s: %s", iface->name, + strerror(error)); + continue; + } + } + } else { + assert(iface->netdev != NULL); + } } + free(dpif_ports); shash_destroy(&cur_ifaces); shash_destroy(&want_ifaces); } @@ -715,8 +780,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) bridge_fetch_dp_ifaces(br); - iterate_and_prune_ifaces(br, check_iface_netdev, NULL); - iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL); + iterate_and_prune_ifaces(br, check_iface, NULL); /* Pick local port hardware address, datapath ID. */ bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); @@ -860,6 +924,13 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) iterate_and_prune_ifaces(br, set_iface_properties, NULL); } + LIST_FOR_EACH (br, node, &all_bridges) { + struct iface *iface; + HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { + iface_update_cfm(iface); + } + } + free(managers); } @@ -1078,6 +1149,82 @@ dpid_from_hash(const void *data, size_t n) return eth_addr_to_uint64(hash); } +static void +iface_refresh_cfm_stats(struct iface *iface) +{ + size_t i; + struct cfm *cfm; + const struct ovsrec_monitor *mon; + + mon = iface->cfg->monitor; + cfm = iface->cfm; + + if (!cfm || !mon) { + return; + } + + for (i = 0; i < mon->n_remote_mps; i++) { + const struct ovsrec_maintenance_point *mp; + const struct remote_mp *rmp; + + mp = mon->remote_mps[i]; + rmp = cfm_get_remote_mp(cfm, mp->mpid); + + ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1); + } + + if (hmap_is_empty(&cfm->x_remote_mps)) { + ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0); + } else { + size_t length; + struct remote_mp *rmp; + int64_t *x_remote_mps; + + length = hmap_count(&cfm->x_remote_mps); + x_remote_mps = xzalloc(length * sizeof *x_remote_mps); + + i = 0; + HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) { + x_remote_mps[i++] = rmp->mpid; + } + + ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length); + free(x_remote_mps); + } + + if (hmap_is_empty(&cfm->x_remote_maids)) { + ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0); + } else { + size_t length; + char **x_remote_maids; + struct remote_maid *rmaid; + + length = hmap_count(&cfm->x_remote_maids); + x_remote_maids = xzalloc(length * sizeof *x_remote_maids); + + i = 0; + HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) { + size_t j; + + x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1); + + for (j = 0; j < CCM_MAID_LEN; j++) { + snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx", + rmaid->maid[j]); + } + i++; + } + ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length); + + for (i = 0; i < length; i++) { + free(x_remote_maids[i]); + } + free(x_remote_maids); + } + + ovsrec_monitor_set_fault(mon, &cfm->fault, 1); +} + static void iface_refresh_stats(struct iface *iface) { @@ -1210,6 +1357,7 @@ bridge_run(void) for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; iface_refresh_stats(iface); + iface_refresh_cfm_stats(iface); } } } @@ -1226,6 +1374,7 @@ void bridge_wait(void) { struct bridge *br; + struct iface *iface; LIST_FOR_EACH (br, node, &all_bridges) { ofproto_wait(br->ofproto); @@ -1235,6 +1384,12 @@ bridge_wait(void) mac_learning_wait(br->ml); bond_wait(br); + + HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { + if (iface->cfm) { + cfm_wait(iface->cfm); + } + } } ovsdb_idl_wait(idl); poll_timer_wait_until(stats_timer); @@ -1435,6 +1590,7 @@ static int bridge_run_one(struct bridge *br) { int error; + struct iface *iface; error = ofproto_run1(br->ofproto); if (error) { @@ -1447,6 +1603,21 @@ bridge_run_one(struct bridge *br) error = ofproto_run2(br->ofproto, br->flush); br->flush = false; + HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { + struct ofpbuf *packet; + + if (!iface->cfm) { + continue; + } + + packet = cfm_run(iface->cfm); + if (packet) { + iface_send_packet(iface, packet); + ofpbuf_uninit(packet); + free(packet); + } + } + return error; } @@ -1564,7 +1735,7 @@ bridge_reconfigure_one(struct bridge *br) /* Configure OpenFlow controller connection snooping. */ svec_init(&snoops); svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); + ovs_rundir(), br->name)); svec_init(&old_snoops); ofproto_get_snoops(br->ofproto, &old_snoops); if (!svec_equal(&snoops, &old_snoops)) { @@ -1584,7 +1755,7 @@ static void bridge_ofproto_controller_for_mgmt(const struct bridge *br, struct ofproto_controller *oc) { - oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name); + oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name); oc->max_backoff = 0; oc->probe_interval = 60; oc->band = OFPROTO_OUT_OF_BAND; @@ -1665,6 +1836,10 @@ bridge_reconfigure_remotes(struct bridge *br, const struct sockaddr_in *managers, size_t n_managers) { + const char *disable_ib_str, *queue_id_str; + bool disable_in_band = false; + int queue_id; + struct ovsrec_controller **controllers; size_t n_controllers; bool had_primary; @@ -1673,7 +1848,22 @@ bridge_reconfigure_remotes(struct bridge *br, size_t n_ocs; size_t i; - ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); + /* Check if we should disable in-band control on this bridge. */ + disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band"); + if (disable_ib_str && !strcmp(disable_ib_str, "true")) { + disable_in_band = true; + } + + /* Set OpenFlow queue ID for in-band control. */ + queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue"); + queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1; + ofproto_set_in_band_queue(br->ofproto, queue_id); + + if (disable_in_band) { + ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0); + } else { + ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); + } had_primary = ofproto_has_primary_controller(br->ofproto); n_controllers = bridge_get_controllers(br, &controllers); @@ -1698,7 +1888,11 @@ bridge_reconfigure_remotes(struct bridge *br, } bridge_configure_local_iface_netdev(br, c); - bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]); + bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]); + if (disable_in_band) { + ocs[n_ocs].band = OFPROTO_OUT_OF_BAND; + } + n_ocs++; } ofproto_set_controllers(br->ofproto, ocs, n_ocs); @@ -1717,14 +1911,14 @@ bridge_reconfigure_remotes(struct bridge *br, if (!n_controllers && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) { union ofp_action action; - flow_t flow; + struct cls_rule rule; memset(&action, 0, sizeof action); action.type = htons(OFPAT_OUTPUT); action.output.len = htons(sizeof action); action.output.port = htons(OFPP_NORMAL); - memset(&flow, 0, sizeof flow); - ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0); + cls_rule_init_catchall(&rule, 0); + ofproto_add_flow(br->ofproto, &rule, &action, 1); } } @@ -1788,12 +1982,10 @@ bridge_fetch_dp_ifaces(struct bridge *br) hash_int(iface->dp_ifidx, 0)); } - if (iface->cfg) { - int64_t ofport = (iface->dp_ifidx >= 0 - ? odp_port_to_ofp_port(iface->dp_ifidx) - : -1); - ovsrec_interface_set_ofport(iface->cfg, &ofport, 1); - } + iface_set_ofport(iface->cfg, + (iface->dp_ifidx >= 0 + ? odp_port_to_ofp_port(iface->dp_ifidx) + : -1)); } } free(dpif_ports); @@ -2021,6 +2213,22 @@ bond_run(struct bridge *br) struct port *port = br->ports[i]; if (port->n_ifaces >= 2) { + char *devname; + + /* Track carrier going up and down on interfaces. */ + while (!netdev_monitor_poll(port->monitor, &devname)) { + struct iface *iface; + + iface = port_lookup_iface(port, devname); + if (iface) { + bool carrier = netdev_get_carrier(iface->netdev); + + bond_link_status_update(iface, carrier); + port_update_bond_compat(port); + } + free(devname); + } + for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (time_msec() >= iface->delay_expires) { @@ -2052,6 +2260,7 @@ bond_wait(struct bridge *br) if (port->n_ifaces < 2) { continue; } + netdev_monitor_poll_wait(port->monitor); for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (iface->delay_expires != LLONG_MAX) { @@ -2065,13 +2274,14 @@ bond_wait(struct bridge *br) } static bool -set_dst(struct dst *p, const flow_t *flow, +set_dst(struct dst *p, const struct flow *flow, const struct port *in_port, const struct port *out_port, tag_type *tags) { p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE : in_port->vlan >= 0 ? in_port->vlan - : ntohs(flow->dl_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); } @@ -2153,21 +2363,43 @@ port_includes_vlan(const struct port *port, uint16_t vlan) return vlan == port->vlan || port_trunks_vlan(port, vlan); } +static bool +port_is_floodable(const struct port *port) +{ + int i; + + for (i = 0; i < port->n_ifaces; i++) { + if (!ofproto_port_is_floodable(port->bridge->ofproto, + port->ifaces[i]->dp_ifidx)) { + return false; + } + } + return true; +} + static size_t -compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, +compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface) { mirror_mask_t mirrors = in_port->src_mirrors; + int flow_vlan; struct dst *dst = dsts; size_t i; + flow_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (flow_vlan == 0) { + flow_vlan = OFP_VLAN_NONE; + } + if (out_port == FLOOD_PORT) { /* XXX use ODP_FLOOD if no vlans or bonding. */ /* XXX even better, define each VLAN as a datapath port group */ for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - if (port != in_port && port_includes_vlan(port, vlan) + if (port != in_port + && port_is_floodable(port) + && port_includes_vlan(port, vlan) && !port->is_mirror_output_port && set_dst(dst, flow, in_port, port, tags)) { mirrors |= port->dst_mirrors; @@ -2195,7 +2427,6 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, if (port_includes_vlan(port, m->out_vlan) && set_dst(dst, flow, in_port, port, tags)) { - int flow_vlan; if (port->vlan < 0) { dst->vlan = m->out_vlan; @@ -2210,10 +2441,6 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, * tagging tags place. This is necessary because * dst->vlan is the final vlan, after removing implicit * tags. */ - flow_vlan = ntohs(flow->dl_vlan); - if (flow_vlan == 0) { - flow_vlan = OFP_VLAN_NONE; - } if (port == in_port && dst->vlan == flow_vlan) { /* Don't send out input port on same VLAN. */ continue; @@ -2226,7 +2453,7 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, mirrors &= mirrors - 1; } - partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan)); + partition_dsts(dsts, dst - dsts, flow_vlan); return dst - dsts; } @@ -2242,7 +2469,7 @@ print_dsts(const struct dst *dsts, size_t n) } static void -compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, +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, uint16_t *nf_output_iface) @@ -2255,15 +2482,19 @@ compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags, nf_output_iface); - cur_vlan = ntohs(flow->dl_vlan); + cur_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (cur_vlan == 0) { + 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); } else { - a = odp_actions_add(actions, ODPAT_SET_VLAN_VID); - a->vlan_vid.vlan_vid = htons(p->vlan); + 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); } cur_vlan = p->vlan; } @@ -2276,28 +2507,19 @@ compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, * 802.1Q header and implicitly tagged ports. A value of 0 indicates that * the packet is untagged and -1 indicates it has an invalid header and * should be dropped. */ -static int flow_get_vlan(struct bridge *br, const flow_t *flow, +static int flow_get_vlan(struct bridge *br, const struct flow *flow, struct port *in_port, bool have_packet) { - /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet - * belongs to VLAN 0, so we should treat both cases identically. (In the - * former case, the packet has an 802.1Q header that specifies VLAN 0, - * presumably to allow a priority to be specified. In the latter case, the - * packet does not have any 802.1Q header.) */ - int vlan = ntohs(flow->dl_vlan); - if (vlan == OFP_VLAN_NONE) { - vlan = 0; - } + int vlan = vlan_tci_to_vid(flow->vlan_tci); if (in_port->vlan >= 0) { if (vlan) { /* XXX support double tagging? */ if (have_packet) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " "packet received on port %s configured with " "implicit VLAN %"PRIu16, - br->name, ntohs(flow->dl_vlan), - in_port->name, in_port->vlan); + br->name, vlan, in_port->name, in_port->vlan); } return -1; } @@ -2322,7 +2544,7 @@ static int flow_get_vlan(struct bridge *br, const flow_t *flow, * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to * indicate this; newer upstream kernels use gratuitous ARP requests. */ static bool -is_gratuitous_arp(const flow_t *flow) +is_gratuitous_arp(const struct flow *flow) { return (flow->dl_type == htons(ETH_TYPE_ARP) && eth_addr_is_broadcast(flow->dl_dst) @@ -2332,7 +2554,7 @@ is_gratuitous_arp(const flow_t *flow) } static void -update_learning_table(struct bridge *br, const flow_t *flow, int vlan, +update_learning_table(struct bridge *br, const struct flow *flow, int vlan, struct port *in_port) { enum grat_arp_lock_type lock_type; @@ -2380,7 +2602,7 @@ update_learning_table(struct bridge *br, const flow_t *flow, int vlan, * so in one special case. */ static bool -is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, +is_admissible(struct bridge *br, const struct flow *flow, bool have_packet, tag_type *tags, int *vlanp, struct port **in_portp) { struct iface *in_iface; @@ -2469,7 +2691,7 @@ is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, * returns true. Otherwise, the actions should only be applied to 'packet', or * not at all, if 'packet' was NULL. */ static bool -process_flow(struct bridge *br, const flow_t *flow, +process_flow(struct bridge *br, const struct flow *flow, const struct ofpbuf *packet, struct odp_actions *actions, tag_type *tags, uint16_t *nf_output_iface) { @@ -2519,57 +2741,30 @@ done: return true; } -/* Careful: 'opp' is in host byte order and opp->port_no is an OFP port - * number. */ -static void -bridge_port_changed_ofhook_cb(enum ofp_port_reason reason, - const struct ofp_phy_port *opp, - void *br_) -{ - struct bridge *br = br_; - struct iface *iface; - struct port *port; - - iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no)); - if (!iface) { - return; - } - port = iface->port; - - if (reason == OFPPR_DELETE) { - VLOG_WARN("bridge %s: interface %s deleted unexpectedly", - br->name, iface->name); - iface_destroy(iface); - if (!port->n_ifaces) { - VLOG_WARN("bridge %s: port %s has no interfaces, dropping", - br->name, port->name); - port_destroy(port); - } - - bridge_flush(br); - } else { - if (port->n_ifaces > 1) { - bool up = !(opp->state & OFPPS_LINK_DOWN); - bond_link_status_update(iface, up); - port_update_bond_compat(port); - } - } -} - static bool -bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, +bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, struct odp_actions *actions, tag_type *tags, uint16_t *nf_output_iface, void *br_) { + struct iface *iface; struct bridge *br = br_; COVERAGE_INC(bridge_process_flow); + iface = iface_from_dp_ifidx(br, flow->in_port); + + if (cfm_should_process_flow(flow)) { + if (packet && iface->cfm) { + cfm_process_heartbeat(iface->cfm, packet); + } + return false; + } + return process_flow(br, flow, packet, actions, tags, nf_output_iface); } static void -bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags, +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_) @@ -2629,7 +2824,6 @@ bridge_account_checkpoint_ofhook_cb(void *br_) } static struct ofhooks bridge_ofhooks = { - bridge_port_changed_ofhook_cb, bridge_normal_ofhook_cb, bridge_account_flow_ofhook_cb, bridge_account_checkpoint_ofhook_cb, @@ -2951,7 +3145,7 @@ bond_send_learning_packets(struct port *port) union ofp_action actions[2], *a; uint16_t dp_ifidx; tag_type tags = 0; - flow_t flow; + struct flow flow; int retval; if (e->port == port->port_idx @@ -3406,6 +3600,7 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) { VLOG_WARN("port %s: %s specified twice as port interface", port->name, if_cfg->name); + iface_set_ofport(if_cfg, -1); continue; } @@ -3419,8 +3614,15 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) } iface->cfg = if_cfg; } else { - iface_create(port, if_cfg); + iface = iface_create(port, if_cfg); } + + /* Determine interface type. The local port always has type + * "internal". Other ports take their type from the database and + * default to "system" if none is specified. */ + iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal" + : if_cfg->type[0] ? if_cfg->type + : "system"); } shash_destroy(&new_ifaces); @@ -3511,6 +3713,9 @@ port_destroy(struct port *port) del = br->ports[port->port_idx] = br->ports[--br->n_ports]; del->port_idx = port->port_idx; + VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); + + netdev_monitor_destroy(port->monitor); free(port->ifaces); bitmap_free(port->trunks); free(port->name); @@ -3542,6 +3747,10 @@ port_lookup_iface(const struct port *port, const char *name) static void port_update_bonding(struct port *port) { + if (port->monitor) { + netdev_monitor_destroy(port->monitor); + port->monitor = NULL; + } if (port->n_ifaces < 2) { /* Not a bonded port. */ if (port->bond_hash) { @@ -3551,9 +3760,9 @@ port_update_bonding(struct port *port) port->bond_fake_iface = false; } } else { - if (!port->bond_hash) { - size_t i; + size_t i; + if (!port->bond_hash) { port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); for (i = 0; i <= BOND_MASK; i++) { struct bond_entry *e = &port->bond_hash[i]; @@ -3571,6 +3780,11 @@ port_update_bonding(struct port *port) } port->bond_compat_is_stale = true; port->bond_fake_iface = port->cfg->bond_fake_iface; + + port->monitor = netdev_monitor_create(); + for (i = 0; i < port->n_ifaces; i++) { + netdev_monitor_add(port->monitor, port->ifaces[i]->netdev); + } } } @@ -3686,13 +3900,32 @@ port_update_vlan_compat(struct port *port) /* Interface functions. */ +static void +iface_send_packet(struct iface *iface, struct ofpbuf *packet) +{ + struct flow flow; + union ofp_action action; + + memset(&action, 0, sizeof action); + action.output.type = htons(OFPAT_OUTPUT); + action.output.len = htons(sizeof action); + action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx)); + + flow_extract(packet, 0, ODPP_NONE, &flow); + + if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1, + packet)) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name); + } +} + static struct iface * iface_create(struct port *port, const struct ovsrec_interface *if_cfg) { struct bridge *br = port->bridge; struct iface *iface; char *name = if_cfg->name; - int error; iface = xzalloc(sizeof *iface); iface->port = port; @@ -3706,20 +3939,6 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) shash_add_assert(&br->iface_by_name, iface->name, iface); - /* Attempt to create the network interface in case it doesn't exist yet. */ - if (!iface_is_internal(br, iface->name)) { - error = create_iface_netdev(iface); - if (error) { - VLOG_WARN("could not create iface %s: %s", iface->name, - strerror(error)); - - shash_find_and_delete_assert(&br->iface_by_name, iface->name); - free(iface->name); - free(iface); - return NULL; - } - } - if (port->n_ifaces >= port->allocated_ifaces) { port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, sizeof *port->ifaces); @@ -3762,6 +3981,8 @@ iface_destroy(struct iface *iface) bond_send_learning_packets(port); } + cfm_destroy(iface->cfm); + free(iface->name); free(iface); @@ -3789,38 +4010,6 @@ iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) return NULL; } -/* Returns true if 'iface' is the name of an "internal" interface on bridge - * 'br', that is, an interface that is entirely simulated within the datapath. - * The local port (ODPP_LOCAL) is always an internal interface. Other local - * interfaces are created by setting "iface..internal = true". - * - * In addition, we have a kluge-y feature that creates an internal port with - * the name of a bonded port if "bonding..fake-iface = true" is set. - * This feature needs to go away in the long term. Until then, this is one - * reason why this function takes a name instead of a struct iface: the fake - * interfaces created this way do not have a struct iface. */ -static bool -iface_is_internal(const struct bridge *br, const char *if_name) -{ - struct iface *iface; - struct port *port; - - if (!strcmp(if_name, br->name)) { - return true; - } - - iface = iface_lookup(br, if_name); - if (iface && !strcmp(iface->cfg->type, "internal")) { - return true; - } - - port = port_lookup(br, if_name); - if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) { - return true; - } - return false; -} - /* Set Ethernet address of 'iface', if one is specified in the configuration * file. */ static void @@ -3845,6 +4034,15 @@ iface_set_mac(struct iface *iface) } } +/* Sets the ofport column of 'if_cfg' to 'ofport'. */ +static void +iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport) +{ + if (if_cfg) { + ovsrec_interface_set_ofport(if_cfg, &ofport, 1); + } +} + /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'. * * The value strings in '*shash' are taken directly from values[], not copied, @@ -3921,6 +4119,56 @@ iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos) } } } + +static void +iface_update_cfm(struct iface *iface) +{ + size_t i; + struct cfm *cfm; + uint16_t *remote_mps; + struct ovsrec_monitor *mon; + uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN]; + + mon = iface->cfg->monitor; + + if (!mon) { + return; + } + + if (netdev_get_etheraddr(iface->netdev, ea)) { + VLOG_WARN("interface %s: Failed to get ethernet address. " + "Skipping Monitor.", iface->name); + return; + } + + if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) { + VLOG_WARN("interface %s: Failed to generate MAID.", iface->name); + return; + } + + if (!iface->cfm) { + iface->cfm = cfm_create(); + } + + cfm = iface->cfm; + cfm->mpid = mon->mpid; + cfm->interval = mon->interval ? *mon->interval : 1000; + + memcpy(cfm->eth_src, ea, sizeof cfm->eth_src); + memcpy(cfm->maid, maid, sizeof cfm->maid); + + remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps); + for(i = 0; i < mon->n_remote_mps; i++) { + remote_mps[i] = mon->remote_mps[i]->mpid; + } + cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps); + free(remote_mps); + + if (!cfm_configure(iface->cfm)) { + cfm_destroy(iface->cfm); + iface->cfm = NULL; + } +} /* Port mirroring. */