X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=7fa819cf6e821ff8962d45e6fa669dfa2298f1d9;hb=6333182946ffd368eb7623d7408185a521058b46;hp=4a7f90bfa6b9aa3b12ac5ffcbb30a7e56f13f72b;hpb=a4e2e1f28976fd17435a7748c233286188778344;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 4a7f90bfa..7fa819cf6 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" @@ -42,6 +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" @@ -65,16 +69,29 @@ #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); + +COVERAGE_DEFINE(bridge_flush); +COVERAGE_DEFINE(bridge_process_flow); +COVERAGE_DEFINE(bridge_reconfigure); struct dst { uint16_t vlan; uint16_t dp_ifidx; }; +struct dst_set { + struct dst builtin[32]; + struct dst *dsts; + size_t n, allocated; +}; + +static void dst_set_init(struct dst_set *); +static void dst_set_add(struct dst_set *, const struct dst *); +static void dst_set_free(struct dst_set *); + struct iface { /* These members are always valid. */ struct port *port; /* Containing port. */ @@ -90,6 +107,7 @@ 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; }; @@ -100,6 +118,11 @@ struct bond_entry { tag_type iface_tag; /* Tag associated with iface_idx. */ }; +enum bond_mode { + BM_SLB, /* Source Load Balance (Default). */ + BM_AB /* Active Backup. */ +}; + #define MAX_MIRRORS 32 typedef uint32_t mirror_mask_t; #define MIRROR_MASK_C(X) UINT32_C(X) @@ -137,17 +160,23 @@ struct port { size_t n_ifaces, allocated_ifaces; /* Bonding info. */ - struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ + enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */ int active_iface; /* Ifidx on which bcasts accepted, or -1. */ tag_type active_iface_tag; /* Tag for bcast flows. */ tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */ int updelay, downdelay; /* Delay before iface goes up/down, in ms. */ bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */ bool bond_fake_iface; /* Fake a bond interface for legacy compat? */ + bool miimon; /* Use miimon instead of carrier? */ + long long int bond_miimon_interval; /* Miimon status refresh interval. */ + long long int bond_miimon_next_update; /* Time of next miimon update. */ long long int bond_next_fake_iface_update; /* Time of next update. */ + struct netdev_monitor *monitor; /* Tracks carrier up/down status. */ + + /* SLB specific bonding info. */ + struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ 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. */ @@ -155,7 +184,6 @@ struct port { bool is_mirror_output_port; /* Does port mirroring send frames here? */ }; -#define DP_MAX_PORTS 255 struct bridge { struct list node; /* Node in global list of bridges. */ char *name; /* User-specified arbitrary name. */ @@ -255,10 +283,17 @@ 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 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 *); +static void shash_to_ovs_idl_map(struct shash *, + char ***keys, char ***values, size_t *n); + /* Hooks into ofproto processing. */ static struct ofhooks bridge_ofhooks; @@ -272,10 +307,10 @@ 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); @@ -283,8 +318,8 @@ 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. */ @@ -296,6 +331,17 @@ bridge_init(const char *remote) bond_init(); } +void +bridge_exit(void) +{ + struct bridge *br, *next_br; + + LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) { + bridge_destroy(br); + } + ovsdb_idl_destroy(idl); +} + /* Performs configuration that is only necessary once at ovs-vswitchd startup, * but for which the ovs-vswitchd configuration 'cfg' is required. */ static void @@ -363,95 +409,6 @@ bridge_configure_once(const struct ovsrec_open_vswitch *cfg) svec_destroy(&dpif_types); } -/* 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. */ -static void -iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options) -{ - size_t i; - - shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options, - if_cfg->n_options, options); - - for (i = 0; i < if_cfg->n_other_config; i++) { - char *key = if_cfg->key_other_config[i]; - char *value = if_cfg->value_other_config[i]; - - if (!shash_find_data(options, key)) { - shash_add(options, key, value); - } else { - VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts " - "with \"options\" key %s", if_cfg->name, key, key); - } - } -} - -/* 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) { - netdev_get_carrier(iface->netdev, &iface->enabled); - } - - 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) @@ -509,6 +466,7 @@ iterate_and_prune_ifaces(struct bridge *br, if (cb(br, iface, aux)) { j++; } else { + iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } } @@ -531,30 +489,52 @@ 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]; + + 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); + } + } + + /* Now extract the targets' IP addresses. */ + if (!shash_is_empty(&targets)) { + struct shash_node *node; - 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]; + 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(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))) { + 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; @@ -573,7 +553,7 @@ 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); @@ -667,18 +647,16 @@ 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) { @@ -692,48 +670,66 @@ 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) { + shash_from_ovs_idl_map(iface->cfg->key_options, + iface->cfg->value_options, + iface->cfg->n_options, &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); + shash_from_ovs_idl_map(iface->cfg->key_options, + iface->cfg->value_options, + iface->cfg->n_options, &args); + netdev_reconfigure(iface->netdev, &args); + shash_destroy(&args); } } free(dpif_ports); @@ -894,6 +890,13 @@ 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); } @@ -1112,6 +1115,105 @@ dpid_from_hash(const void *data, size_t n) return eth_addr_to_uint64(hash); } +static void +iface_refresh_status(struct iface *iface) +{ + struct shash sh; + + shash_init(&sh); + + if (!netdev_get_status(iface->netdev, &sh)) { + size_t n; + char **keys, **values; + + shash_to_ovs_idl_map(&sh, &keys, &values, &n); + ovsrec_interface_set_status(iface->cfg, keys, values, n); + + free(keys); + free(values); + } else { + ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0); + } + + shash_destroy_free_data(&sh); +} + +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) { @@ -1244,6 +1346,8 @@ 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); + iface_refresh_status(iface); } } } @@ -1260,6 +1364,7 @@ void bridge_wait(void) { struct bridge *br; + struct iface *iface; LIST_FOR_EACH (br, node, &all_bridges) { ofproto_wait(br->ofproto); @@ -1269,6 +1374,12 @@ 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); @@ -1469,6 +1580,7 @@ static int bridge_run_one(struct bridge *br) { int error; + struct iface *iface; error = ofproto_run1(br->ofproto); if (error) { @@ -1481,6 +1593,21 @@ 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; } @@ -1598,7 +1725,7 @@ 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)) { @@ -1618,7 +1745,7 @@ 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; @@ -1699,6 +1826,10 @@ 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; @@ -1707,7 +1838,22 @@ 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); @@ -1732,7 +1878,11 @@ 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); @@ -1751,14 +1901,14 @@ 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); } } @@ -1822,12 +1972,10 @@ 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); @@ -1836,15 +1984,17 @@ 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)]; + assert(port->bond_mode == BM_SLB); + return &port->bond_hash[bond_hash(mac, vlan)]; } static int @@ -1879,15 +2029,21 @@ 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; assert(port->n_ifaces); if (port->n_ifaces == 1) { iface = port->ifaces[0]; - } else { - struct bond_entry *e = lookup_bond_entry(port, dl_src); + } else if (port->bond_mode == BM_AB) { + if (port->active_iface < 0) { + *tags |= port->no_ifaces_tag; + return false; + } + iface = port->ifaces[port->active_iface]; + } else if (port->bond_mode == BM_SLB){ + 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 @@ -1902,6 +2058,8 @@ choose_output_iface(const struct port *port, const uint8_t *dl_src, } *tags |= e->iface_tag; iface = port->ifaces[e->iface_idx]; + } else { + NOT_REACHED(); } *dp_ifidx = iface->dp_ifidx; *tags |= iface->tag; /* Currently only used for bonding. */ @@ -1918,8 +2076,8 @@ bond_link_status_update(struct iface *iface, bool carrier) /* Nothing to do. */ return; } - VLOG_INFO_RL(&rl, "interface %s: carrier %s", - iface->name, carrier ? "detected" : "dropped"); + VLOG_INFO_RL(&rl, "interface %s: link state %s", + iface->name, carrier ? "up" : "down"); if (carrier == iface->enabled) { iface->delay_expires = LLONG_MAX; VLOG_INFO_RL(&rl, "interface %s: will not be %s", @@ -2057,17 +2215,36 @@ bond_run(struct bridge *br) 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; - bool carrier; + if (port->monitor) { + assert(!port->miimon); + + /* 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 up = netdev_get_carrier(iface->netdev); + + bond_link_status_update(iface, up); + port_update_bond_compat(port); + } + free(devname); + } + } else { + assert(port->miimon); + + if (time_msec() >= port->bond_miimon_next_update) { + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + bool up = netdev_get_miimon(iface->netdev); - iface = port_lookup_iface(port, devname); - if (iface && !netdev_get_carrier(iface->netdev, &carrier)) { - bond_link_status_update(iface, carrier); - port_update_bond_compat(port); + bond_link_status_update(iface, up); + port_update_bond_compat(port); + } + port->bond_miimon_next_update = time_msec() + + port->bond_miimon_interval; } - free(devname); } for (j = 0; j < port->n_ifaces; j++) { @@ -2101,7 +2278,15 @@ bond_wait(struct bridge *br) if (port->n_ifaces < 2) { continue; } - netdev_monitor_poll_wait(port->monitor); + + if (port->monitor) { + netdev_monitor_poll_wait(port->monitor); + } + + if (port->miimon) { + poll_timer_wait_until(port->bond_miimon_next_update); + } + for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (iface->delay_expires != LLONG_MAX) { @@ -2115,14 +2300,16 @@ bond_wait(struct bridge *br) } static bool -set_dst(struct dst *p, const struct flow *flow, +set_dst(struct dst *dst, 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 + dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE : in_port->vlan >= 0 ? in_port->vlan - : ntohs(flow->dl_vlan)); - return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags); + : flow->vlan_tci == 0 ? OFP_VLAN_NONE + : vlan_tci_to_vid(flow->vlan_tci)); + return choose_output_iface(out_port, flow->dl_src, dst->vlan, + &dst->dp_ifidx, tags); } static void @@ -2139,10 +2326,10 @@ swap_dst(struct dst *p, struct dst *q) * vlan, but in most cases there are at most two different vlan tags so that's * possibly overkill.) */ static void -partition_dsts(struct dst *dsts, size_t n_dsts, int vlan) +partition_dsts(struct dst_set *set, int vlan) { - struct dst *first = dsts; - struct dst *last = dsts + n_dsts; + struct dst *first = set->dsts; + struct dst *last = set->dsts + set->n; while (first != last) { /* Invariants: @@ -2177,13 +2364,48 @@ mirror_mask_ffs(mirror_mask_t mask) return ffs(mask); } +static void +dst_set_init(struct dst_set *set) +{ + set->dsts = set->builtin; + set->n = 0; + set->allocated = ARRAY_SIZE(set->builtin); +} + +static void +dst_set_add(struct dst_set *set, const struct dst *dst) +{ + if (set->n >= set->allocated) { + size_t new_allocated; + struct dst *new_dsts; + + new_allocated = set->allocated * 2; + new_dsts = xmalloc(new_allocated * sizeof *new_dsts); + memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); + + dst_set_free(set); + + set->dsts = new_dsts; + set->allocated = new_allocated; + } + set->dsts[set->n++] = *dst; +} + +static void +dst_set_free(struct dst_set *set) +{ + if (set->dsts != set->builtin) { + free(set->dsts); + } +} + static bool -dst_is_duplicate(const struct dst *dsts, size_t n_dsts, - const struct dst *test) +dst_is_duplicate(const struct dst_set *set, const struct dst *test) { size_t i; - for (i = 0; i < n_dsts; i++) { - if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) { + for (i = 0; i < set->n; i++) { + if (set->dsts[i].vlan == test->vlan + && set->dsts[i].dp_ifidx == test->dp_ifidx) { return true; } } @@ -2217,56 +2439,58 @@ port_is_floodable(const struct port *port) return true; } -static size_t +static void 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) + struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface) { mirror_mask_t mirrors = in_port->src_mirrors; - struct dst *dst = dsts; + struct dst dst; + int flow_vlan; 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_is_floodable(port) && port_includes_vlan(port, vlan) && !port->is_mirror_output_port - && set_dst(dst, flow, in_port, port, tags)) { + && set_dst(&dst, flow, in_port, port, tags)) { mirrors |= port->dst_mirrors; - dst++; + dst_set_add(set, &dst); } } *nf_output_iface = NF_OUT_FLOOD; - } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) { - *nf_output_iface = dst->dp_ifidx; + } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) { + dst_set_add(set, &dst); + *nf_output_iface = dst.dp_ifidx; mirrors |= out_port->dst_mirrors; - dst++; } while (mirrors) { struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1]; if (!m->n_vlans || vlan_is_mirrored(m, vlan)) { if (m->out_port) { - if (set_dst(dst, flow, in_port, m->out_port, tags) - && !dst_is_duplicate(dsts, dst - dsts, dst)) { - dst++; + if (set_dst(&dst, flow, in_port, m->out_port, tags) + && !dst_is_duplicate(set, &dst)) { + dst_set_add(set, &dst); } } else { for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; if (port_includes_vlan(port, m->out_vlan) - && set_dst(dst, flow, in_port, port, tags)) + && set_dst(&dst, flow, in_port, port, tags)) { - int flow_vlan; - if (port->vlan < 0) { - dst->vlan = m->out_vlan; + dst.vlan = m->out_vlan; } - if (dst_is_duplicate(dsts, dst - dsts, dst)) { + if (dst_is_duplicate(set, &dst)) { continue; } @@ -2276,15 +2500,11 @@ 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) { + if (port == in_port && dst.vlan == flow_vlan) { /* Don't send out input port on same VLAN. */ continue; } - dst++; + dst_set_add(set, &dst); } } } @@ -2292,17 +2512,20 @@ 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)); - return dst - dsts; + partition_dsts(set, flow_vlan); } static void OVS_UNUSED -print_dsts(const struct dst *dsts, size_t n) +print_dsts(const struct dst_set *set) { - for (; n--; dsts++) { - printf(">p%"PRIu16, dsts->dp_ifidx); - if (dsts->vlan != OFP_VLAN_NONE) { - printf("v%"PRIu16, dsts->vlan); + size_t i; + + for (i = 0; i < set->n; i++) { + const struct dst *dst = &set->dsts[i]; + + printf(">p%"PRIu16, dst->dp_ifidx); + if (dst->vlan != OFP_VLAN_NONE) { + printf("v%"PRIu16, dst->vlan); } } } @@ -2310,33 +2533,37 @@ 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)]; - size_t n_dsts; - const struct dst *p; + struct dst_set set; uint16_t cur_vlan; + size_t i; - n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags, - nf_output_iface); + dst_set_init(&set); + compose_dsts(br, flow, vlan, in_port, out_port, &set, tags, + nf_output_iface); - cur_vlan = ntohs(flow->dl_vlan); - 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); + cur_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (cur_vlan == 0) { + cur_vlan = OFP_VLAN_NONE; + } + for (i = 0; i < set.n; i++) { + const struct dst *dst = &set.dsts[i]; + if (dst->vlan != cur_vlan) { + if (dst->vlan == OFP_VLAN_NONE) { + 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.mask = htons(VLAN_VID_MASK); + ovs_be16 tci; + tci = htons(dst->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; + cur_vlan = dst->vlan; } - a = odp_actions_add(actions, ODPAT_OUTPUT); - a->output.port = p->dp_ifidx; + nl_msg_put_u32(actions, ODPAT_OUTPUT, dst->dp_ifidx); } + dst_set_free(&set); } /* Returns the effective vlan of a packet, taking into account both the @@ -2346,25 +2573,16 @@ 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; } @@ -2537,7 +2755,7 @@ 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; @@ -2588,26 +2806,37 @@ 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; 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 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, + size_t 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 @@ -2625,12 +2854,16 @@ 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); - if (out_port && out_port->n_ifaces >= 2) { + 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 && + out_port->bond_mode == BM_SLB) { + 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; } } @@ -2651,7 +2884,8 @@ bridge_account_checkpoint_ofhook_cb(void *br_) now = time_msec(); for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) { + if (port->n_ifaces > 1 && port->bond_mode == BM_SLB + && now >= port->bond_next_rebalance) { port->bond_next_rebalance = now + port->bond_rebalance_interval; bond_rebalance_port(port); } @@ -2678,6 +2912,20 @@ struct slave_balance { size_t n_hashes; }; +static const char * +bond_mode_to_string(enum bond_mode bm) { + static char *bm_slb = "balance-slb"; + static char *bm_ab = "active-backup"; + + switch (bm) { + case BM_SLB: return bm_slb; + case BM_AB: return bm_ab; + } + + NOT_REACHED(); + return NULL; +} + /* Sorts pointers to pointers to bond_entries in ascending order by the * interface to which they are assigned, and within a single interface in * ascending order of bytes transmitted. */ @@ -2786,6 +3034,8 @@ bond_shift_load(struct slave_balance *from, struct slave_balance *to, struct port *port = from->iface->port; uint64_t delta = hash->tx_bytes; + assert(port->bond_mode == BM_SLB); + VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) " "from %s to %s (now carrying %"PRIu64"kB and " "%"PRIu64"kB load, respectively)", @@ -2821,13 +3071,15 @@ bond_shift_load(struct slave_balance *from, struct slave_balance *to, static void bond_rebalance_port(struct port *port) { - struct slave_balance bals[DP_MAX_PORTS]; + struct slave_balance *bals; size_t n_bals; struct bond_entry *hashes[BOND_MASK + 1]; struct slave_balance *b, *from, *to; struct bond_entry *e; size_t i; + assert(port->bond_mode == BM_SLB); + /* Sets up 'bals' to describe each of the port's interfaces, sorted in * descending order of tx_bytes, so that bals[0] represents the most * heavily loaded slave and bals[n_bals - 1] represents the least heavily @@ -2839,6 +3091,7 @@ bond_rebalance_port(struct port *port) * become contiguous in memory, and then we point each 'hashes' members of * a slave_balance structure to the start of a contiguous group. */ n_bals = port->n_ifaces; + bals = xmalloc(n_bals * sizeof *bals); for (b = bals; b < &bals[n_bals]; b++) { b->iface = port->ifaces[b - bals]; b->tx_bytes = 0; @@ -2868,7 +3121,7 @@ bond_rebalance_port(struct port *port) while (!bals[n_bals - 1].iface->enabled) { n_bals--; if (!n_bals) { - return; + goto exit; } } @@ -2960,6 +3213,9 @@ bond_rebalance_port(struct port *port) for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) { e->tx_bytes /= 2; } + +exit: + free(bals); } static void @@ -2984,7 +3240,7 @@ 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; } @@ -3036,7 +3292,7 @@ bond_unixctl_list(struct unixctl_conn *conn, struct ds ds = DS_EMPTY_INITIALIZER; const struct bridge *br; - ds_put_cstr(&ds, "bridge\tbond\tslaves\n"); + ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n"); LIST_FOR_EACH (br, node, &all_bridges) { size_t i; @@ -3046,7 +3302,8 @@ bond_unixctl_list(struct unixctl_conn *conn, if (port->n_ifaces > 1) { size_t j; - ds_put_format(&ds, "%s\t%s\t", br->name, port->name); + ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name, + bond_mode_to_string(port->bond_mode)); for (j = 0; j < port->n_ifaces; j++) { const struct iface *iface = port->ifaces[j]; if (j) { @@ -3094,10 +3351,18 @@ bond_unixctl_show(struct unixctl_conn *conn, return; } + ds_put_format(&ds, "bond_mode: %s\n", + bond_mode_to_string(port->bond_mode)); + ds_put_format(&ds, "bond-detect-mode: %s\n", + port->miimon ? "miimon" : "carrier"); ds_put_format(&ds, "updelay: %d ms\n", port->updelay); ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); - ds_put_format(&ds, "next rebalance: %lld ms\n", - port->bond_next_rebalance - time_msec()); + + if (port->bond_mode == BM_SLB) { + ds_put_format(&ds, "next rebalance: %lld ms\n", + port->bond_next_rebalance - time_msec()); + } + for (j = 0; j < port->n_ifaces; j++) { const struct iface *iface = port->ifaces[j]; struct bond_entry *be; @@ -3114,6 +3379,10 @@ bond_unixctl_show(struct unixctl_conn *conn, iface->delay_expires - time_msec()); } + if (port->bond_mode != BM_SLB) { + continue; + } + /* Hashes. */ for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { int hash = be - port->bond_hash; @@ -3130,9 +3399,10 @@ 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", @@ -3152,7 +3422,6 @@ 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; @@ -3173,10 +3442,12 @@ 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 (port->bond_mode != BM_SLB) { + unixctl_command_reply(conn, 501, "not an SLB bond"); + return; + } + + if (strspn(hash_s, "0123456789") == strlen(hash_s)) { hash = atoi(hash_s) & BOND_MASK; } else { unixctl_command_reply(conn, 501, "bad hash"); @@ -3298,23 +3569,39 @@ 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 @@ -3399,8 +3686,9 @@ port_del_ifaces(struct port *port, const struct ovsrec_port *cfg) static void port_reconfigure(struct port *port, const struct ovsrec_port *cfg) { + const char *detect_mode; struct shash new_ifaces; - long long int next_rebalance; + long long int next_rebalance, miimon_next_update; unsigned long *trunks; int vlan; size_t i; @@ -3426,6 +3714,41 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) port->bond_next_rebalance = next_rebalance; } + detect_mode = get_port_other_config(cfg, "bond-detect-mode", + "carrier"); + + if (!strcmp(detect_mode, "carrier")) { + port->miimon = false; + } else if (!strcmp(detect_mode, "miimon")) { + port->miimon = true; + } else { + port->miimon = false; + VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to " + "carrier", port->name, detect_mode); + } + + port->bond_miimon_interval = atoi( + get_port_other_config(cfg, "bond-miimon-interval", "200")); + if (port->bond_miimon_interval < 100) { + port->bond_miimon_interval = 100; + } + miimon_next_update = time_msec() + port->bond_miimon_interval; + if (port->bond_miimon_next_update > miimon_next_update) { + port->bond_miimon_next_update = miimon_next_update; + } + + if (!port->cfg->bond_mode || + !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) { + port->bond_mode = BM_SLB; + } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) { + port->bond_mode = BM_AB; + } else { + port->bond_mode = BM_SLB; + VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s", + port->name, port->cfg->bond_mode, + bond_mode_to_string(port->bond_mode)); + } + /* Add new interfaces and update 'cfg' member of existing ones. */ shash_init(&new_ifaces); for (i = 0; i < cfg->n_interfaces; i++) { @@ -3435,6 +3758,7 @@ 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; } @@ -3547,6 +3871,8 @@ 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); @@ -3589,12 +3915,13 @@ port_update_bonding(struct port *port) free(port->bond_hash); port->bond_hash = NULL; port->bond_compat_is_stale = true; - port->bond_fake_iface = false; } + + port->bond_fake_iface = false; } else { size_t i; - if (!port->bond_hash) { + if (port->bond_mode == BM_SLB && !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]; @@ -3609,13 +3936,18 @@ port_update_bonding(struct port *port) if (port->cfg->bond_fake_iface) { port->bond_next_fake_iface_update = time_msec(); } + } else if (port->bond_mode != BM_SLB) { + free(port->bond_hash); + port->bond_hash = NULL; } 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); + if (!port->miimon) { + port->monitor = netdev_monitor_create(); + for (i = 0; i < port->n_ifaces; i++) { + netdev_monitor_add(port->monitor, port->ifaces[i]->netdev); + } } } } @@ -3627,7 +3959,7 @@ port_update_bond_compat(struct port *port) struct compat_bond bond; size_t i; - if (port->n_ifaces < 2) { + if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) { proc_net_compat_update_bond(port->name, NULL); return; } @@ -3732,6 +4064,26 @@ 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) { @@ -3776,6 +4128,10 @@ iface_destroy(struct iface *iface) bool del_active = port->active_iface == iface->port_ifidx; struct iface *del; + if (port->monitor) { + netdev_monitor_remove(port->monitor, iface->netdev); + } + shash_find_and_delete_assert(&br->iface_by_name, iface->name); if (iface->dp_ifidx >= 0) { @@ -3793,6 +4149,8 @@ iface_destroy(struct iface *iface) bond_send_learning_packets(port); } + cfm_destroy(iface->cfm); + free(iface->name); free(iface); @@ -3844,6 +4202,15 @@ 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, @@ -3860,6 +4227,38 @@ shash_from_ovs_idl_map(char **keys, char **values, size_t n, } } +/* Creates 'keys' and 'values' arrays from 'shash'. + * + * Sets 'keys' and 'values' to heap allocated arrays representing the key-value + * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They + * are populated with with strings taken directly from 'shash' and thus have + * the same ownership of the key-value pairs in shash. + */ +static void +shash_to_ovs_idl_map(struct shash *shash, + char ***keys, char ***values, size_t *n) +{ + size_t i, count; + char **k, **v; + struct shash_node *sn; + + count = shash_count(shash); + + k = xmalloc(count * sizeof *k); + v = xmalloc(count * sizeof *v); + + i = 0; + SHASH_FOR_EACH(sn, shash) { + k[i] = sn->name; + v[i] = sn->data; + i++; + } + + *n = count; + *keys = k; + *values = v; +} + struct iface_delete_queues_cbdata { struct netdev *netdev; const struct ovsdb_datum *queues; @@ -3920,6 +4319,56 @@ 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. */