X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=9916414db990431c6c2970ce3cb600304c021f54;hb=b3907fbc6c74ddad7507d0f7abb1f5a2528cd2be;hp=5085703a362a2f2f8eb9a9b493459742ab994297;hpb=94db54073ea372406d4d82a009301984b097bc6b;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 5085703a3..9916414db 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -32,6 +32,8 @@ #include #include #include "bitmap.h" +#include "cfm.h" +#include "classifier.h" #include "coverage.h" #include "dirs.h" #include "dpif.h" @@ -70,6 +72,10 @@ 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; @@ -90,6 +96,7 @@ struct iface { 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; }; @@ -257,6 +264,9 @@ static struct iface *iface_from_dp_ifidx(const struct bridge *, 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 *); @@ -273,10 +283,10 @@ 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); @@ -284,8 +294,8 @@ 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. */ @@ -366,10 +376,7 @@ 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) { @@ -391,68 +398,6 @@ 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) @@ -691,18 +636,16 @@ 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) { @@ -716,48 +659,62 @@ 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); @@ -918,6 +875,13 @@ 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); } @@ -1136,6 +1100,82 @@ 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) { @@ -1268,6 +1308,7 @@ 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); } } } @@ -1284,6 +1325,7 @@ void bridge_wait(void) { struct bridge *br; + struct iface *iface; LIST_FOR_EACH (br, node, &all_bridges) { ofproto_wait(br->ofproto); @@ -1293,6 +1335,12 @@ 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); @@ -1493,6 +1541,7 @@ static int bridge_run_one(struct bridge *br) { int error; + struct iface *iface; error = ofproto_run1(br->ofproto); if (error) { @@ -1505,6 +1554,21 @@ 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; } @@ -1622,7 +1686,7 @@ 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)) { @@ -1642,7 +1706,7 @@ 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; @@ -1723,23 +1787,29 @@ 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; - const char *disable_ib_str; - bool disable_in_band = false; struct ofproto_controller *ocs; size_t n_ocs; size_t i; - /* 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 { @@ -1792,14 +1862,14 @@ bridge_reconfigure_remotes(struct bridge *br, if (!n_controllers && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) { union ofp_action action; - struct flow 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); } } @@ -2161,7 +2231,8 @@ set_dst(struct dst *p, const struct flow *flow, { 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); } @@ -2263,9 +2334,15 @@ compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, 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 */ @@ -2301,7 +2378,6 @@ compose_dsts(const struct bridge *br, const struct flow *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; @@ -2316,10 +2392,6 @@ compose_dsts(const struct bridge *br, const struct flow *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; @@ -2332,7 +2404,7 @@ compose_dsts(const struct bridge *br, const struct flow *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; } @@ -2361,7 +2433,10 @@ compose_actions(struct bridge *br, const struct flow *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) { @@ -2370,7 +2445,7 @@ compose_actions(struct bridge *br, const struct flow *flow, uint16_t 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 |= htons(flow->dl_vlan_pcp << VLAN_PCP_SHIFT); + a->dl_tci.tci |= flow->vlan_tci & htons(VLAN_PCP_MASK); } cur_vlan = p->vlan; } @@ -2386,25 +2461,16 @@ compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan, 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; } @@ -2631,10 +2697,20 @@ 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); } @@ -3775,6 +3851,26 @@ 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) { @@ -3836,6 +3932,8 @@ iface_destroy(struct iface *iface) bond_send_learning_packets(port); } + cfm_destroy(iface->cfm); + free(iface->name); free(iface); @@ -3972,6 +4070,56 @@ 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. */