X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=9fb257954227585a84552c7e1a82ce8d767ed3eb;hb=195c8086244e33ec42fd9fc8354eaedfd849bbba;hp=b2b997dc3c6d0ff2568de46755b9e4b50e8e7690;hpb=c835c3bf87d8ebf71cb13810a6ef2c167e012180;p=openvswitch diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index b2b997dc..9fb25795 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -35,6 +35,7 @@ #include "cfm.h" #include "classifier.h" #include "coverage.h" +#include "daemon.h" #include "dirs.h" #include "dpif.h" #include "dynamic-string.h" @@ -54,7 +55,6 @@ #include "ovsdb-data.h" #include "packets.h" #include "poll-loop.h" -#include "proc-net-compat.h" #include "process.h" #include "sha1.h" #include "shash.h" @@ -75,7 +75,10 @@ VLOG_DEFINE_THIS_MODULE(bridge); COVERAGE_DEFINE(bridge_flush); COVERAGE_DEFINE(bridge_process_flow); +COVERAGE_DEFINE(bridge_process_cfm); +COVERAGE_DEFINE(bridge_process_lacp); COVERAGE_DEFINE(bridge_reconfigure); +COVERAGE_DEFINE(bridge_lacp_update); struct dst { uint16_t vlan; @@ -92,6 +95,13 @@ 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 *); +enum lacp_status { + LACP_CURRENT = 0x01, /* Current State. */ + LACP_EXPIRED = 0x02, /* Expired State. */ + LACP_DEFAULTED = 0x04, /* Partner is defaulted. */ + LACP_ATTACHED = 0x08, /* Attached. Interface may be choosen for flows. */ +}; + struct iface { /* These members are always valid. */ struct port *port; /* Containing port. */ @@ -110,6 +120,14 @@ struct iface { const char *type; /* Usually same as cfg->type. */ struct cfm *cfm; /* Connectivity Fault Management */ const struct ovsrec_interface *cfg; + + /* LACP information. */ + enum lacp_status lacp_status; /* LACP status. */ + uint16_t lacp_priority; /* LACP port priority. */ + struct lacp_info lacp_actor; /* LACP actor information. */ + struct lacp_info lacp_partner; /* LACP partner information. */ + long long int lacp_tx; /* Next LACP message transmission time. */ + long long int lacp_rx; /* Next LACP message receive time. */ }; #define BOND_MASK 0xff @@ -120,7 +138,8 @@ struct bond_entry { }; enum bond_mode { - BM_SLB, /* Source Load Balance (Default). */ + BM_TCP, /* Transport Layer Load Balance. */ + BM_SLB, /* Source Load Balance. */ BM_AB /* Active Backup. */ }; @@ -145,6 +164,11 @@ struct mirror { int out_vlan; }; +/* Flags for a port's lacp member. */ +#define LACP_ACTIVE 0x01 /* LACP is in active mode. */ +#define LACP_PASSIVE 0x02 /* LACP is in passive mode. */ +#define LACP_NEGOTIATED 0x04 /* LACP has successfully negotiated. */ + #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ struct port { struct bridge *bridge; @@ -166,7 +190,6 @@ struct port { 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. */ @@ -174,6 +197,12 @@ struct port { long long int bond_next_fake_iface_update; /* Time of next update. */ struct netdev_monitor *monitor; /* Tracks carrier up/down status. */ + /* LACP information. */ + int lacp; /* LACP status flags. 0 if LACP is off. */ + uint16_t lacp_key; /* LACP aggregation key. */ + uint16_t lacp_priority; /* LACP system priority. */ + bool lacp_need_update; /* Need to update attached interfaces? */ + /* 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. */ @@ -252,14 +281,21 @@ static struct iface *bridge_get_local_iface(struct bridge *); static uint64_t dpid_from_hash(const void *, size_t nbytes); static unixctl_cb_func bridge_unixctl_fdb_show; +static unixctl_cb_func qos_unixctl_show; + +static void lacp_run(struct port *); +static void lacp_wait(struct port *); +static void lacp_process_packet(const struct ofpbuf *, struct iface *); static void bond_init(void); -static void bond_run(struct bridge *); -static void bond_wait(struct bridge *); +static void bond_run(struct port *); +static void bond_wait(struct port *); static void bond_rebalance_port(struct port *); static void bond_send_learning_packets(struct port *); static void bond_enable_slave(struct iface *iface, bool enable); +static void port_run(struct port *); +static void port_wait(struct port *); static struct port *port_create(struct bridge *, const char *name); static void port_reconfigure(struct port *, const struct ovsrec_port *); static void port_del_ifaces(struct port *, const struct ovsrec_port *); @@ -268,9 +304,8 @@ static struct port *port_lookup(const struct bridge *, const char *name); static struct iface *port_lookup_iface(const struct port *, const char *name); static struct port *port_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); -static void port_update_bond_compat(struct port *); -static void port_update_vlan_compat(struct port *); static void port_update_bonding(struct port *); +static void port_update_lacp(struct port *); static void mirror_create(struct bridge *, struct ovsrec_mirror *); static void mirror_destroy(struct mirror *); @@ -282,6 +317,7 @@ static struct iface *iface_create(struct port *port, const struct ovsrec_interface *if_cfg); static void iface_destroy(struct iface *); static struct iface *iface_lookup(const struct bridge *, const char *name); +static struct iface *iface_find(const char *name); static struct iface *iface_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); static void iface_set_mac(struct iface *); @@ -290,6 +326,10 @@ 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 uint8_t iface_get_lacp_state(const struct iface *); +static void iface_get_lacp_priority(struct iface *, struct lacp_info *); +static void iface_set_lacp_defaulted(struct iface *); +static void iface_set_lacp_expired(struct iface *); static void shash_from_ovs_idl_map(char **keys, char **values, size_t n, struct shash *); @@ -326,6 +366,7 @@ bridge_init(const char *remote) /* Register unixctl commands. */ unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + unixctl_command_register("qos/show", qos_unixctl_show, NULL); unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, NULL); unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect, @@ -464,7 +505,7 @@ iterate_and_prune_ifaces(struct bridge *br, if (port->n_ifaces) { i++; } else { - VLOG_ERR("%s port has no interfaces, dropping", port->name); + VLOG_WARN("%s port has no interfaces, dropping", port->name); port_destroy(port); } } @@ -490,13 +531,10 @@ collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg, struct shash targets; 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. */ + /* Collect all of the potential targets from 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]; @@ -606,9 +644,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) && strcmp(dpif_port.name, br->name)) { int retval = dpif_port_del(br->dpif, dpif_port.port_no); if (retval) { - VLOG_ERR("failed to remove %s interface from %s: %s", - dpif_port.name, dpif_name(br->dpif), - strerror(retval)); + VLOG_WARN("failed to remove %s interface from %s: %s", + dpif_port.name, dpif_name(br->dpif), + strerror(retval)); } } } @@ -698,9 +736,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) dpif_name(br->dpif)); break; } else { - VLOG_ERR("failed to add %s interface to %s: %s", - if_name, dpif_name(br->dpif), - strerror(error)); + VLOG_WARN("failed to add %s interface to %s: %s", + if_name, dpif_name(br->dpif), + strerror(error)); continue; } } @@ -875,8 +913,8 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) struct port *port = br->ports[i]; int j; - port_update_vlan_compat(port); port_update_bonding(port); + port_update_lacp(port); for (j = 0; j < port->n_ifaces; j++) { iface_update_qos(port->ifaces[j], port->cfg->qos); @@ -895,6 +933,10 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } free(managers); + + /* ovs-vswitchd has completed initialization, so allow the process that + * forked us to exit successfully. */ + daemonize_complete(); } static const char * @@ -944,7 +986,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], /* Otherwise choose the minimum non-local MAC address among all of the * interfaces. */ - memset(ea, 0xff, sizeof ea); + memset(ea, 0xff, ETH_ADDR_LEN); for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; uint8_t iface_ea[ETH_ADDR_LEN]; @@ -1169,7 +1211,7 @@ iface_refresh_status(struct iface *iface) ? "up" : "down"); error = netdev_get_mtu(iface->netdev, &mtu); - if (!error) { + if (!error && mtu != INT_MAX) { mtu_64 = mtu; ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1); } @@ -1450,21 +1492,19 @@ void bridge_wait(void) { struct bridge *br; - struct iface *iface; LIST_FOR_EACH (br, node, &all_bridges) { + size_t i; + ofproto_wait(br->ofproto); if (ofproto_has_primary_controller(br->ofproto)) { continue; } mac_learning_wait(br->ml); - bond_wait(br); - HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { - if (iface->cfm) { - cfm_wait(iface->cfm); - } + for (i = 0; i < br->n_ports; i++) { + port_wait(br->ports[i]); } } ovsdb_idl_wait(idl); @@ -1529,6 +1569,99 @@ bridge_unixctl_fdb_show(struct unixctl_conn *conn, ds_destroy(&ds); } +/* QoS unixctl user interface functions. */ + +struct qos_unixctl_show_cbdata { + struct ds *ds; + struct iface *iface; +}; + +static void +qos_unixctl_show_cb(unsigned int queue_id, + const struct shash *details, + void *aux) +{ + struct qos_unixctl_show_cbdata *data = aux; + struct ds *ds = data->ds; + struct iface *iface = data->iface; + struct netdev_queue_stats stats; + struct shash_node *node; + int error; + + ds_put_cstr(ds, "\n"); + if (queue_id) { + ds_put_format(ds, "Queue %u:\n", queue_id); + } else { + ds_put_cstr(ds, "Default:\n"); + } + + SHASH_FOR_EACH (node, details) { + ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data); + } + + error = netdev_get_queue_stats(iface->netdev, queue_id, &stats); + if (!error) { + if (stats.tx_packets != UINT64_MAX) { + ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets); + } + + if (stats.tx_bytes != UINT64_MAX) { + ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes); + } + + if (stats.tx_errors != UINT64_MAX) { + ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors); + } + } else { + ds_put_format(ds, "\tFailed to get statistics for queue %u: %s", + queue_id, strerror(error)); + } +} + +static void +qos_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct shash sh = SHASH_INITIALIZER(&sh); + struct iface *iface; + const char *type; + struct shash_node *node; + struct qos_unixctl_show_cbdata data; + int error; + + iface = iface_find(args); + if (!iface) { + unixctl_command_reply(conn, 501, "no such interface"); + return; + } + + netdev_get_qos(iface->netdev, &type, &sh); + + if (*type != '\0') { + ds_put_format(&ds, "QoS: %s %s\n", iface->name, type); + + SHASH_FOR_EACH (node, &sh) { + ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data); + } + + data.ds = &ds; + data.iface = iface; + error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data); + + if (error) { + ds_put_format(&ds, "failed to dump queues: %s", strerror(error)); + } + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + } else { + ds_put_format(&ds, "QoS not configured on %s\n", iface->name); + unixctl_command_reply(conn, 501, ds_cstr(&ds)); + } + + shash_destroy_free_data(&sh); + ds_destroy(&ds); +} + /* Bridge reconfiguration functions. */ static struct bridge * bridge_create(const struct ovsrec_bridge *br_cfg) @@ -1665,8 +1798,8 @@ bridge_unixctl_reconnect(struct unixctl_conn *conn, static int bridge_run_one(struct bridge *br) { + size_t i; int error; - struct iface *iface; error = ofproto_run1(br->ofproto); if (error) { @@ -1674,26 +1807,14 @@ bridge_run_one(struct bridge *br) } mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); - bond_run(br); + + for (i = 0; i < br->n_ports; i++) { + port_run(br->ports[i]); + } 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; } @@ -1835,8 +1956,6 @@ bridge_ofproto_controller_for_mgmt(const struct bridge *br, oc->max_backoff = 0; oc->probe_interval = 60; oc->band = OFPROTO_OUT_OF_BAND; - oc->accept_re = NULL; - oc->update_resolv_conf = false; oc->rate_limit = 0; oc->burst_limit = 0; } @@ -1851,8 +1970,6 @@ bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c, oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band") ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); - oc->accept_re = c->discover_accept_regex; - oc->update_resolv_conf = c->discover_update_resolv_conf; oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; oc->burst_limit = (c->controller_burst_limit ? *c->controller_burst_limit : 0); @@ -1870,11 +1987,6 @@ bridge_configure_local_iface_netdev(struct bridge *br, struct iface *local_iface; struct in_addr ip; - /* Controller discovery does its own TCP/IP configuration later. */ - if (strcmp(c->target, "discover")) { - return; - } - /* If there's no local interface or no IP address, give up. */ local_iface = bridge_get_local_iface(br); if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) { @@ -2066,18 +2178,42 @@ bridge_fetch_dp_ifaces(struct bridge *br) /* Bridge packet processing functions. */ +static bool +bond_is_tcp_hash(const struct port *port) +{ + return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED; +} + static int -bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan) +bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan) { return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK; } +static int bond_hash_tcp(const struct flow *flow, uint16_t vlan) +{ + struct flow hash_flow; + + memcpy(&hash_flow, flow, sizeof hash_flow); + hash_flow.vlan_tci = 0; + + /* The symmetric quality of this hash function is not required, but + * flow_hash_symmetric_l4 already exists, and is sufficient for our + * purposes, so we use it out of convenience. */ + return flow_hash_symmetric_l4(&hash_flow, 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 struct flow *flow, uint16_t vlan) { - assert(port->bond_mode == BM_SLB); - return &port->bond_hash[bond_hash(mac, vlan)]; + assert(port->bond_mode != BM_AB); + + if (bond_is_tcp_hash(port)) { + return &port->bond_hash[bond_hash_tcp(flow, vlan)]; + } else { + return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)]; + } } static int @@ -2092,7 +2228,9 @@ bond_choose_iface(const struct port *port) if (iface->enabled) { return i; - } else if (iface->delay_expires < next_delay_expiration) { + } else if (iface->delay_expires < next_delay_expiration + && (iface->lacp_status & LACP_ATTACHED + || !(port->lacp & LACP_NEGOTIATED))) { best_down_slave = i; next_delay_expiration = iface->delay_expires; } @@ -2111,7 +2249,7 @@ bond_choose_iface(const struct port *port) } static bool -choose_output_iface(const struct port *port, const uint8_t *dl_src, +choose_output_iface(const struct port *port, const struct flow *flow, uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags) { struct iface *iface; @@ -2125,8 +2263,8 @@ choose_output_iface(const struct port *port, const uint8_t *dl_src, 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); + } else { + struct bond_entry *e = lookup_bond_entry(port, flow, 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 @@ -2137,12 +2275,9 @@ choose_output_iface(const struct port *port, const uint8_t *dl_src, return false; } e->iface_tag = tag_create_random(); - ((struct port *) port)->bond_compat_is_stale = true; } *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. */ @@ -2154,33 +2289,54 @@ bond_link_status_update(struct iface *iface) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); struct port *port = iface->port; - bool carrier = iface->up; + bool up = iface->up; + int updelay, downdelay; - if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { + updelay = port->updelay; + downdelay = port->downdelay; + + if (iface->port->lacp & LACP_NEGOTIATED) { + downdelay = 0; + updelay = 0; + } + + if (iface->port->lacp && up) { + /* The interface is up if it's attached to an aggregator and its + * partner is synchronized. The only exception is defaulted links. + * They are not required to have synchronized partners because they + * have no partners at all. However, they will only be attached if + * negotiations failed on all interfaces in the bond. */ + up = iface->lacp_status & LACP_ATTACHED + && (iface->lacp_partner.state & LACP_STATE_SYNC + || iface->lacp_status & LACP_DEFAULTED); + } + + + if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { /* Nothing to do. */ return; } VLOG_INFO_RL(&rl, "interface %s: link state %s", - iface->name, carrier ? "up" : "down"); - if (carrier == iface->enabled) { + iface->name, up ? "up" : "down"); + if (up == iface->enabled) { iface->delay_expires = LLONG_MAX; VLOG_INFO_RL(&rl, "interface %s: will not be %s", - iface->name, carrier ? "disabled" : "enabled"); - } else if (carrier && port->active_iface < 0) { + iface->name, up ? "disabled" : "enabled"); + } else if (up && port->active_iface < 0) { bond_enable_slave(iface, true); - if (port->updelay) { + if (updelay) { VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " - "other interface is up", iface->name, port->updelay); + "other interface is up", iface->name, updelay); } } else { - int delay = carrier ? port->updelay : port->downdelay; + int delay = up ? updelay : downdelay; iface->delay_expires = time_msec() + delay; if (delay) { VLOG_INFO_RL(&rl, "interface %s: will be %s if it stays %s for %d ms", iface->name, - carrier ? "enabled" : "disabled", - carrier ? "up" : "down", + up ? "enabled" : "disabled", + up ? "up" : "down", delay); } } @@ -2248,7 +2404,6 @@ bond_enable_slave(struct iface *iface, bool enable) } moving_active_iface = false; - port->bond_compat_is_stale = true; } /* Attempts to make the sum of the bond slaves' statistics appear on the fake @@ -2289,106 +2444,103 @@ bond_update_fake_iface_stats(struct port *port) } static void -bond_run(struct bridge *br) +bond_link_carrier_update(struct iface *iface, bool carrier) { - size_t i, j; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + if (carrier == iface->up) { + return; + } - if (port->n_ifaces >= 2) { - char *devname; + if (iface->lacp_status & LACP_CURRENT) { + iface_set_lacp_expired(iface); + } - if (port->monitor) { - assert(!port->miimon); + iface->up = carrier; + iface->lacp_tx = 0; +} - /* Track carrier going up and down on interfaces. */ - while (!netdev_monitor_poll(port->monitor, &devname)) { - struct iface *iface; +static void +bond_run(struct port *port) +{ + size_t i; + char *devname; - iface = port_lookup_iface(port, devname); - if (iface) { - bool up = netdev_get_carrier(iface->netdev); + if (port->n_ifaces < 2) { + return; + } - if (up != iface->up) { - port->bond_compat_is_stale = true; - } - iface->up = up; - } - free(devname); - } - } else { - assert(port->miimon); + if (port->monitor) { + 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); + /* Track carrier going up and down on interfaces. */ + while (!netdev_monitor_poll(port->monitor, &devname)) { + struct iface *iface; - if (up != iface->up) { - port->bond_compat_is_stale = true; - } - iface->up = up; - } - port->bond_miimon_next_update = time_msec() + - port->bond_miimon_interval; - } + iface = port_lookup_iface(port, devname); + if (iface) { + bool up = netdev_get_carrier(iface->netdev); + bond_link_carrier_update(iface, up); } + free(devname); + } + } else { + assert(port->miimon); - for (j = 0; j < port->n_ifaces; j++) { - bond_link_status_update(port->ifaces[j]); + if (time_msec() >= port->bond_miimon_next_update) { + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + bool up = netdev_get_miimon(iface->netdev); + bond_link_carrier_update(iface, up); } + port->bond_miimon_next_update = time_msec() + + port->bond_miimon_interval; + } + } - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (time_msec() >= iface->delay_expires) { - bond_enable_slave(iface, !iface->enabled); - } - } + for (i = 0; i < port->n_ifaces; i++) { + bond_link_status_update(port->ifaces[i]); + } - if (port->bond_fake_iface - && time_msec() >= port->bond_next_fake_iface_update) { - bond_update_fake_iface_stats(port); - port->bond_next_fake_iface_update = time_msec() + 1000; - } + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + if (time_msec() >= iface->delay_expires) { + bond_enable_slave(iface, !iface->enabled); } + } - if (port->bond_compat_is_stale) { - port->bond_compat_is_stale = false; - port_update_bond_compat(port); - } + if (port->bond_fake_iface + && time_msec() >= port->bond_next_fake_iface_update) { + bond_update_fake_iface_stats(port); + port->bond_next_fake_iface_update = time_msec() + 1000; } } static void -bond_wait(struct bridge *br) +bond_wait(struct port *port) { - size_t i, j; + size_t i; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (port->n_ifaces < 2) { - continue; - } + if (port->n_ifaces < 2) { + return; + } - if (port->monitor) { - 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); - } + 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) { - poll_timer_wait_until(iface->delay_expires); - } - } - if (port->bond_fake_iface) { - poll_timer_wait_until(port->bond_next_fake_iface_update); + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + if (iface->delay_expires != LLONG_MAX) { + poll_timer_wait_until(iface->delay_expires); } } + + if (port->bond_fake_iface) { + poll_timer_wait_until(port->bond_next_fake_iface_update); + } } static bool @@ -2400,7 +2552,7 @@ set_dst(struct dst *dst, const struct flow *flow, : in_port->vlan >= 0 ? in_port->vlan : flow->vlan_tci == 0 ? OFP_VLAN_NONE : vlan_tci_to_vid(flow->vlan_tci)); - return choose_output_iface(out_port, flow->dl_src, dst->vlan, + return choose_output_iface(out_port, flow, dst->vlan, &dst->dp_ifidx, tags); } @@ -2811,8 +2963,14 @@ is_admissible(struct bridge *br, const struct flow *flow, bool have_packet, return false; } - /* Packets received on bonds need special attention to avoid duplicates. */ - if (in_port->n_ifaces > 1) { + /* When using LACP, do not accept packets from disabled interfaces. */ + if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) { + return false; + } + + /* Packets received on non-LACP bonds need special attention to avoid + * duplicates. */ + if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) { int src_idx; bool is_grat_arp_locked; @@ -2901,28 +3059,45 @@ bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, 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); + return process_flow(br, flow, packet, actions, tags, nf_output_iface); +} + +static bool +bridge_special_ofhook_cb(const struct flow *flow, + const struct ofpbuf *packet, void *br_) +{ + struct iface *iface; + struct bridge *br = br_; iface = iface_from_dp_ifidx(br, flow->in_port); if (cfm_should_process_flow(flow)) { - if (packet && iface->cfm) { + + if (iface && packet && iface->cfm) { + COVERAGE_INC(bridge_process_cfm); cfm_process_heartbeat(iface->cfm, packet); } return false; + } else if (flow->dl_type == htons(ETH_TYPE_LACP)) { + + if (iface && packet) { + COVERAGE_INC(bridge_process_lacp); + lacp_process_packet(packet, iface); + } + return false; } - return process_flow(br, flow, packet, actions, tags, nf_output_iface); + return true; } static void bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, const struct nlattr *actions, size_t actions_len, - unsigned long long int n_bytes, void *br_) + uint64_t n_bytes, void *br_) { struct bridge *br = br_; const struct nlattr *a; @@ -2950,12 +3125,11 @@ bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, if (nl_attr_type(a) == ODP_ACTION_ATTR_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) { + out_port->bond_mode != BM_AB) { 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, vlan); + struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan); e->tx_bytes += n_bytes; } } @@ -2976,7 +3150,7 @@ 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 && port->bond_mode == BM_SLB + if (port->n_ifaces > 1 && port->bond_mode != BM_AB && now >= port->bond_next_rebalance) { port->bond_next_rebalance = now + port->bond_rebalance_interval; bond_rebalance_port(port); @@ -2986,10 +3160,193 @@ bridge_account_checkpoint_ofhook_cb(void *br_) static struct ofhooks bridge_ofhooks = { bridge_normal_ofhook_cb, + bridge_special_ofhook_cb, bridge_account_flow_ofhook_cb, bridge_account_checkpoint_ofhook_cb, }; +/* LACP functions. */ + +static void +lacp_process_packet(const struct ofpbuf *packet, struct iface *iface) +{ + const struct lacp_pdu *pdu; + + if (!iface->port->lacp) { + return; + } + + pdu = parse_lacp_packet(packet); + if (!pdu) { + return; + } + + iface->lacp_status |= LACP_CURRENT; + iface->lacp_status &= ~(LACP_EXPIRED | LACP_DEFAULTED); + iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX; + + iface->lacp_actor.state = iface_get_lacp_state(iface); + if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) { + iface->lacp_tx = 0; + } + + if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) { + iface->port->lacp_need_update = true; + iface->lacp_partner = pdu->actor; + } +} + +static void +lacp_update_ifaces(struct port *port) +{ + size_t i; + struct iface *lead; + struct lacp_info lead_pri; + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); + + port->lacp_need_update = false; + COVERAGE_INC(bridge_lacp_update); + + if (!port->lacp) { + return; + } + + VLOG_DBG_RL(&rl, "port %s: re-evaluating LACP link status", port->name); + + lead = NULL; + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + struct lacp_info pri; + + iface->lacp_status |= LACP_ATTACHED; + ofproto_revalidate(port->bridge->ofproto, iface->tag); + + /* Don't allow loopback interfaces to send traffic or lead. */ + if (eth_addr_equals(iface->lacp_partner.sysid, + iface->lacp_actor.sysid)) { + VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is " + "connected to its own bridge", iface->name); + iface->lacp_status &= ~LACP_ATTACHED; + continue; + } + + if (iface->lacp_status & LACP_DEFAULTED) { + continue; + } + + iface_get_lacp_priority(iface, &pri); + + if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) { + lead = iface; + lead_pri = pri; + } + } + + if (!lead) { + port->lacp &= ~LACP_NEGOTIATED; + return; + } + + port->lacp |= LACP_NEGOTIATED; + + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + + if (iface->lacp_status & LACP_DEFAULTED + || lead->lacp_partner.key != iface->lacp_partner.key + || !eth_addr_equals(lead->lacp_partner.sysid, + iface->lacp_partner.sysid)) { + iface->lacp_status &= ~LACP_ATTACHED; + } + } +} + +static bool +lacp_iface_may_tx(const struct iface *iface) +{ + return iface->port->lacp & LACP_ACTIVE + || iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED); +} + +static void +lacp_run(struct port *port) +{ + size_t i; + struct ofpbuf packet; + + if (!port->lacp) { + return; + } + + ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN); + + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + + if (time_msec() > iface->lacp_rx) { + if (iface->lacp_status & LACP_CURRENT) { + iface_set_lacp_expired(iface); + } else if (iface->lacp_status & LACP_EXPIRED) { + iface_set_lacp_defaulted(iface); + } + } + } + + if (port->lacp_need_update) { + lacp_update_ifaces(port); + } + + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + uint8_t ea[ETH_ADDR_LEN]; + int error; + + if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) { + continue; + } + + error = netdev_get_etheraddr(iface->netdev, ea); + if (!error) { + iface->lacp_actor.state = iface_get_lacp_state(iface); + compose_lacp_packet(&packet, &iface->lacp_actor, + &iface->lacp_partner, ea); + iface_send_packet(iface, &packet); + } else { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); + VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address " + "(%s)", iface->name, strerror(error)); + } + + iface->lacp_tx = time_msec() + + (iface->lacp_partner.state & LACP_STATE_TIME + ? LACP_FAST_TIME_TX + : LACP_SLOW_TIME_TX); + } + ofpbuf_uninit(&packet); +} + +static void +lacp_wait(struct port *port) +{ + size_t i; + + if (!port->lacp) { + return; + } + + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + + if (lacp_iface_may_tx(iface)) { + poll_timer_wait_until(iface->lacp_tx); + } + + if (iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED)) { + poll_timer_wait_until(iface->lacp_rx); + } + } +} + /* Bonding functions. */ /* Statistics for a single interface on a bonded port, used for load-based @@ -3008,10 +3365,12 @@ static const char * bond_mode_to_string(enum bond_mode bm) { static char *bm_slb = "balance-slb"; static char *bm_ab = "active-backup"; + static char *bm_tcp = "balance-tcp"; switch (bm) { case BM_SLB: return bm_slb; case BM_AB: return bm_ab; + case BM_TCP: return bm_tcp; } NOT_REACHED(); @@ -3126,7 +3485,7 @@ 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); + assert(port->bond_mode != BM_AB); VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) " "from %s to %s (now carrying %"PRIu64"kB and " @@ -3170,7 +3529,7 @@ bond_rebalance_port(struct port *port) struct bond_entry *e; size_t i; - assert(port->bond_mode == BM_SLB); + assert(port->bond_mode != BM_AB); /* 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 @@ -3280,7 +3639,6 @@ bond_rebalance_port(struct port *port) } if (i < from->n_hashes) { bond_shift_load(from, to, i); - port->bond_compat_is_stale = true; /* If the result of the migration changed the relative order of * 'from' and 'to' swap them back to maintain invariants. */ @@ -3318,7 +3676,7 @@ bond_send_learning_packets(struct port *port) struct ofpbuf packet; int error, n_packets, n_errors; - if (!port->n_ifaces || port->active_iface < 0) { + if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) { return; } @@ -3331,8 +3689,15 @@ bond_send_learning_packets(struct port *port) struct flow flow; int retval; - if (e->port == port->port_idx - || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) { + if (e->port == port->port_idx) { + continue; + } + + compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, + e->mac); + flow_extract(&packet, 0, ODPP_NONE, &flow); + + if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) { continue; } @@ -3352,9 +3717,6 @@ bond_send_learning_packets(struct port *port) /* Send packet. */ n_packets++; - compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, - e->mac); - flow_extract(&packet, 0, ODPP_NONE, &flow); retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, &packet); if (retval) { @@ -3429,6 +3791,42 @@ bond_find(const char *name) return NULL; } +static void +ds_put_lacp_state(struct ds *ds, uint8_t state) +{ + if (state & LACP_STATE_ACT) { + ds_put_cstr(ds, "activity "); + } + + if (state & LACP_STATE_TIME) { + ds_put_cstr(ds, "timeout "); + } + + if (state & LACP_STATE_AGG) { + ds_put_cstr(ds, "aggregation "); + } + + if (state & LACP_STATE_SYNC) { + ds_put_cstr(ds, "synchronized "); + } + + if (state & LACP_STATE_COL) { + ds_put_cstr(ds, "collecting "); + } + + if (state & LACP_STATE_DIST) { + ds_put_cstr(ds, "distributing "); + } + + if (state & LACP_STATE_DEF) { + ds_put_cstr(ds, "defaulted "); + } + + if (state & LACP_STATE_EXP) { + ds_put_cstr(ds, "expired "); + } +} + static void bond_unixctl_show(struct unixctl_conn *conn, const char *args, void *aux OVS_UNUSED) @@ -3445,6 +3843,20 @@ bond_unixctl_show(struct unixctl_conn *conn, ds_put_format(&ds, "bond_mode: %s\n", bond_mode_to_string(port->bond_mode)); + + if (port->lacp) { + ds_put_format(&ds, "lacp: %s\n", + port->lacp & LACP_ACTIVE ? "active" : "passive"); + } else { + ds_put_cstr(&ds, "lacp: off\n"); + } + + if (port->bond_mode != BM_AB) { + ds_put_format(&ds, "bond-hash-algorithm: %s\n", + bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb"); + } + + ds_put_format(&ds, "bond-detect-mode: %s\n", port->miimon ? "miimon" : "carrier"); @@ -3456,7 +3868,7 @@ bond_unixctl_show(struct unixctl_conn *conn, ds_put_format(&ds, "updelay: %d ms\n", port->updelay); ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); - if (port->bond_mode == BM_SLB) { + if (port->bond_mode != BM_AB) { ds_put_format(&ds, "next rebalance: %lld ms\n", port->bond_next_rebalance - time_msec()); } @@ -3464,9 +3876,10 @@ bond_unixctl_show(struct unixctl_conn *conn, for (j = 0; j < port->n_ifaces; j++) { const struct iface *iface = port->ifaces[j]; struct bond_entry *be; + struct flow flow; /* Basic info. */ - ds_put_format(&ds, "slave %s: %s\n", + ds_put_format(&ds, "\nslave %s: %s\n", iface->name, iface->enabled ? "enabled" : "disabled"); if (j == port->active_iface) { ds_put_cstr(&ds, "\tactive slave\n"); @@ -3477,11 +3890,76 @@ bond_unixctl_show(struct unixctl_conn *conn, iface->delay_expires - time_msec()); } - if (port->bond_mode != BM_SLB) { + if (port->lacp) { + ds_put_cstr(&ds, "\tstatus: "); + + if (iface->lacp_status & LACP_CURRENT) { + ds_put_cstr(&ds, "current "); + } + + if (iface->lacp_status & LACP_EXPIRED) { + ds_put_cstr(&ds, "expired "); + } + + if (iface->lacp_status & LACP_DEFAULTED) { + ds_put_cstr(&ds, "defaulted "); + } + + if (iface->lacp_status & LACP_ATTACHED) { + ds_put_cstr(&ds, "attached "); + } + + ds_put_cstr(&ds, "\n"); + + ds_put_cstr(&ds, "\n\tactor sysid: "); + ds_put_format(&ds, ETH_ADDR_FMT, + ETH_ADDR_ARGS(iface->lacp_actor.sysid)); + ds_put_cstr(&ds, "\n"); + + ds_put_format(&ds, "\tactor sys_priority: %u\n", + ntohs(iface->lacp_actor.sys_priority)); + + ds_put_format(&ds, "\tactor portid: %u\n", + ntohs(iface->lacp_actor.portid)); + + ds_put_format(&ds, "\tactor port_priority: %u\n", + ntohs(iface->lacp_actor.port_priority)); + + ds_put_format(&ds, "\tactor key: %u\n", + ntohs(iface->lacp_actor.key)); + + ds_put_cstr(&ds, "\tactor state: "); + ds_put_lacp_state(&ds, iface_get_lacp_state(iface)); + ds_put_cstr(&ds, "\n\n"); + + ds_put_cstr(&ds, "\tpartner sysid: "); + ds_put_format(&ds, ETH_ADDR_FMT, + ETH_ADDR_ARGS(iface->lacp_partner.sysid)); + ds_put_cstr(&ds, "\n"); + + ds_put_format(&ds, "\tpartner sys_priority: %u\n", + ntohs(iface->lacp_partner.sys_priority)); + + ds_put_format(&ds, "\tpartner portid: %u\n", + ntohs(iface->lacp_partner.portid)); + + ds_put_format(&ds, "\tpartner port_priority: %u\n", + ntohs(iface->lacp_partner.port_priority)); + + ds_put_format(&ds, "\tpartner key: %u\n", + ntohs(iface->lacp_partner.key)); + + ds_put_cstr(&ds, "\tpartner state: "); + ds_put_lacp_state(&ds, iface->lacp_partner.state); + ds_put_cstr(&ds, "\n"); + } + + if (port->bond_mode == BM_AB) { continue; } /* Hashes. */ + memset(&flow, 0, sizeof flow); for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { int hash = be - port->bond_hash; struct mac_entry *me; @@ -3493,13 +3971,19 @@ bond_unixctl_show(struct unixctl_conn *conn, ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", hash, be->tx_bytes / 1024); + if (port->bond_mode != BM_SLB) { + continue; + } + /* MACs. */ LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) { uint16_t dp_ifidx; tag_type tags = 0; - if (bond_hash(me->mac, me->vlan) == hash + + memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN); + if (bond_hash_src(me->mac, me->vlan) == hash && me->port != port->port_idx - && choose_output_iface(port, me->mac, me->vlan, + && choose_output_iface(port, &flow, me->vlan, &dp_ifidx, &tags) && dp_ifidx == iface->dp_ifidx) { @@ -3567,7 +4051,6 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, ofproto_revalidate(port->bridge->ofproto, entry->iface_tag); entry->iface_idx = iface->port_ifidx; entry->iface_tag = tag_create_random(); - port->bond_compat_is_stale = true; unixctl_command_reply(conn, 200, "migrated"); } @@ -3692,7 +4175,7 @@ bond_unixctl_hash(struct unixctl_conn *conn, const char *args_, if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) == ETH_ADDR_SCAN_COUNT) { - hash = bond_hash(mac, vlan); + hash = bond_hash_src(mac, vlan); hash_cstr = xasprintf("%u", hash); unixctl_command_reply(conn, 200, hash_cstr); @@ -3719,6 +4202,44 @@ bond_init(void) /* Port functions. */ +static void +port_run(struct port *port) +{ + size_t i; + + lacp_run(port); + bond_run(port); + + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + + if (iface->cfm) { + struct ofpbuf *packet = cfm_run(iface->cfm); + if (packet) { + iface_send_packet(iface, packet); + ofpbuf_uninit(packet); + free(packet); + } + } + } +} + +static void +port_wait(struct port *port) +{ + size_t i; + + lacp_wait(port); + bond_wait(port); + + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + if (iface->cfm) { + cfm_wait(iface->cfm); + } + } +} + static struct port * port_create(struct bridge *br, const char *name) { @@ -3756,6 +4277,17 @@ get_port_other_config(const struct ovsrec_port *port, const char *key, return value ? value : default_value; } +static const char * +get_interface_other_config(const struct ovsrec_interface *iface, + const char *key, const char *default_value) +{ + const char *value; + + value = get_ovsrec_key_value(&iface->header_, + &ovsrec_interface_col_other_config, key); + return value ? value : default_value; +} + static void port_del_ifaces(struct port *port, const struct ovsrec_port *cfg) { @@ -3786,7 +4318,7 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) { const char *detect_mode; struct shash new_ifaces; - long long int next_rebalance, miimon_next_update; + long long int next_rebalance, miimon_next_update, lacp_priority; unsigned long *trunks; int vlan; size_t i; @@ -3840,6 +4372,8 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) port->bond_mode = BM_SLB; } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) { port->bond_mode = BM_AB; + } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) { + port->bond_mode = BM_TCP; } else { port->bond_mode = BM_SLB; VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s", @@ -3879,9 +4413,45 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal" : if_cfg->type[0] ? if_cfg->type : "system"); + + lacp_priority = + atoi(get_interface_other_config(if_cfg, "lacp-port-priority", + "0")); + + if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) { + iface->lacp_priority = UINT16_MAX; + } else { + iface->lacp_priority = lacp_priority; + } } shash_destroy(&new_ifaces); + lacp_priority = + atoi(get_port_other_config(cfg, "lacp-system-priority", "0")); + + if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) { + /* Prefer bondable links if unspecified. */ + port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX; + } else { + port->lacp_priority = lacp_priority; + } + + if (!port->cfg->lacp) { + /* XXX when LACP implementation has been sufficiently tested, enable by + * default and make active on bonded ports. */ + port->lacp = 0; + } else if (!strcmp(port->cfg->lacp, "off")) { + port->lacp = 0; + } else if (!strcmp(port->cfg->lacp, "active")) { + port->lacp = LACP_ACTIVE; + } else if (!strcmp(port->cfg->lacp, "passive")) { + port->lacp = LACP_PASSIVE; + } else { + VLOG_WARN("port %s: unknown LACP mode %s", + port->name, port->cfg->lacp); + port->lacp = 0; + } + /* Get VLAN tag. */ vlan = -1; if (cfg->tag) { @@ -3950,9 +4520,6 @@ port_destroy(struct port *port) struct port *del; int i; - proc_net_compat_update_vlan(port->name, NULL, 0); - proc_net_compat_update_bond(port->name, NULL); - for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; if (m && m->out_port == port) { @@ -4000,6 +4567,52 @@ port_lookup_iface(const struct port *port, const char *name) return iface && iface->port == port ? iface : NULL; } +static void +port_update_lacp(struct port *port) +{ + size_t i; + bool key_changed; + + if (!port->lacp || port->n_ifaces < 1) { + for (i = 0; i < port->n_ifaces; i++) { + iface_set_lacp_defaulted(port->ifaces[i]); + } + return; + } + + key_changed = true; + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + + if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) { + port->lacp = 0; + return; + } + + if (iface->dp_ifidx == port->lacp_key) { + key_changed = false; + } + } + + if (key_changed) { + port->lacp_key = port->ifaces[0]->dp_ifidx; + } + + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + + iface->lacp_actor.sys_priority = htons(port->lacp_priority); + memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN); + + iface->lacp_actor.port_priority = htons(iface->lacp_priority); + iface->lacp_actor.portid = htons(iface->dp_ifidx); + iface->lacp_actor.key = htons(port->lacp_key); + + iface->lacp_tx = 0; + } + port->lacp_need_update = true; +} + static void port_update_bonding(struct port *port) { @@ -4009,37 +4622,40 @@ port_update_bonding(struct port *port) } if (port->n_ifaces < 2) { /* Not a bonded port. */ - if (port->bond_hash) { - free(port->bond_hash); - port->bond_hash = NULL; - port->bond_compat_is_stale = true; - } - + free(port->bond_hash); + port->bond_hash = NULL; port->bond_fake_iface = false; + port->active_iface = -1; + port->no_ifaces_tag = 0; } else { size_t i; - if (port->bond_mode == BM_SLB && !port->bond_hash) { + if (port->bond_mode != BM_AB && !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]; e->iface_idx = -1; e->tx_bytes = 0; } - port->no_ifaces_tag = tag_create_random(); - bond_choose_active_iface(port); port->bond_next_rebalance = time_msec() + port->bond_rebalance_interval; - - if (port->cfg->bond_fake_iface) { - port->bond_next_fake_iface_update = time_msec(); - } - } else if (port->bond_mode != BM_SLB) { + } else if (port->bond_mode == BM_AB) { free(port->bond_hash); port->bond_hash = NULL; } - port->bond_compat_is_stale = true; + + if (!port->no_ifaces_tag) { + port->no_ifaces_tag = tag_create_random(); + } + + if (port->active_iface < 0) { + bond_choose_active_iface(port); + } + port->bond_fake_iface = port->cfg->bond_fake_iface; + if (port->bond_fake_iface) { + port->bond_next_fake_iface_update = time_msec(); + } if (!port->miimon) { port->monitor = netdev_monitor_create(); @@ -4049,118 +4665,92 @@ port_update_bonding(struct port *port) } } } + +/* Interface functions. */ static void -port_update_bond_compat(struct port *port) +iface_set_lacp_defaulted(struct iface *iface) { - struct compat_bond_hash compat_hashes[BOND_MASK + 1]; - struct compat_bond bond; - size_t i; + memset(&iface->lacp_partner, 0, sizeof iface->lacp_partner); - if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) { - proc_net_compat_update_bond(port->name, NULL); - return; - } + iface->lacp_status |= LACP_DEFAULTED; + iface->lacp_status &= ~(LACP_CURRENT | LACP_EXPIRED); + iface->lacp_tx = 0; + iface->port->lacp_need_update = true; +} - bond.up = false; - bond.updelay = port->updelay; - bond.downdelay = port->downdelay; +static void +iface_set_lacp_expired(struct iface *iface) +{ + iface->lacp_status &= ~LACP_CURRENT; + iface->lacp_status |= LACP_EXPIRED; + iface->lacp_partner.state |= LACP_STATE_TIME; + iface->lacp_partner.state &= ~LACP_STATE_SYNC; - bond.n_hashes = 0; - bond.hashes = compat_hashes; - if (port->bond_hash) { - const struct bond_entry *e; - for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) { - if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { - struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++]; - cbh->hash = e - port->bond_hash; - cbh->netdev_name = port->ifaces[e->iface_idx]->name; - } - } + iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX; + iface->lacp_tx = 0; +} + +static uint8_t +iface_get_lacp_state(const struct iface *iface) +{ + uint8_t state = 0; + + if (iface->port->lacp & LACP_ACTIVE) { + state |= LACP_STATE_ACT; } - bond.n_slaves = port->n_ifaces; - bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves); - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - struct compat_bond_slave *slave = &bond.slaves[i]; - slave->name = iface->name; + if (iface->lacp_status & LACP_ATTACHED) { + state |= LACP_STATE_SYNC; + } - /* We need to make the same determination as the Linux bonding - * code to determine whether a slave should be consider "up". - * The Linux function bond_miimon_inspect() supports four - * BOND_LINK_* states: - * - * - BOND_LINK_UP: carrier detected, updelay has passed. - * - BOND_LINK_FAIL: carrier lost, downdelay in progress. - * - BOND_LINK_DOWN: carrier lost, downdelay has passed. - * - BOND_LINK_BACK: carrier detected, updelay in progress. - * - * The function bond_info_show_slave() only considers BOND_LINK_UP - * to be "up" and anything else to be "down". - */ - slave->up = iface->enabled && iface->delay_expires == LLONG_MAX; - if (slave->up) { - bond.up = true; - } - netdev_get_etheraddr(iface->netdev, slave->mac); + if (iface->lacp_status & LACP_DEFAULTED) { + state |= LACP_STATE_DEF; } - if (port->bond_fake_iface) { - struct netdev *bond_netdev; + if (iface->lacp_status & LACP_EXPIRED) { + state |= LACP_STATE_EXP; + } - if (!netdev_open_default(port->name, &bond_netdev)) { - if (bond.up) { - netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); - } else { - netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); - } - netdev_close(bond_netdev); - } + if (iface->port->n_ifaces > 1) { + state |= LACP_STATE_AGG; } - proc_net_compat_update_bond(port->name, &bond); - free(bond.slaves); + if (iface->enabled) { + state |= LACP_STATE_COL | LACP_STATE_DIST; + } + + return state; } +/* Given 'iface', populates 'priority' with data representing its LACP link + * priority. If two priority objects populated by this function are compared + * using memcmp, the higher priority link will be less than the lower priority + * link. */ static void -port_update_vlan_compat(struct port *port) +iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority) { - struct bridge *br = port->bridge; - char *vlandev_name = NULL; - - if (port->vlan > 0) { - /* Figure out the name that the VLAN device should actually have, if it - * existed. This takes some work because the VLAN device would not - * have port->name in its name; rather, it would have the trunk port's - * name, and 'port' would be attached to a bridge that also had the - * VLAN device one of its ports. So we need to find a trunk port that - * includes port->vlan. - * - * There might be more than one candidate. This doesn't happen on - * XenServer, so if it happens we just pick the first choice in - * alphabetical order instead of creating multiple VLAN devices. */ - size_t i; - for (i = 0; i < br->n_ports; i++) { - struct port *p = br->ports[i]; - if (port_trunks_vlan(p, port->vlan) - && p->n_ifaces - && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0)) - { - uint8_t ea[ETH_ADDR_LEN]; - netdev_get_etheraddr(p->ifaces[0]->netdev, ea); - if (!eth_addr_is_multicast(ea) && - !eth_addr_is_reserved(ea) && - !eth_addr_is_zero(ea)) { - vlandev_name = p->name; - } - } - } + uint16_t partner_priority, actor_priority; + + /* Choose the lacp_info of the higher priority system by comparing their + * system priorities and mac addresses. */ + actor_priority = ntohs(iface->lacp_actor.sys_priority); + partner_priority = ntohs(iface->lacp_partner.sys_priority); + if (actor_priority < partner_priority) { + *priority = iface->lacp_actor; + } else if (partner_priority < actor_priority) { + *priority = iface->lacp_partner; + } else if (eth_addr_compare_3way(iface->lacp_actor.sysid, + iface->lacp_partner.sysid) < 0) { + *priority = iface->lacp_actor; + } else { + *priority = iface->lacp_partner; } - proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan); + + /* Key and state are not used in priority comparisons. */ + priority->key = 0; + priority->state = 0; } - -/* Interface functions. */ static void iface_send_packet(struct iface *iface, struct ofpbuf *packet) @@ -4198,6 +4788,11 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) iface->delay_expires = LLONG_MAX; iface->netdev = NULL; iface->cfg = if_cfg; + iface_set_lacp_defaulted(iface); + + if (port->lacp & LACP_ACTIVE) { + iface_set_lacp_expired(iface); + } shash_add_assert(&br->iface_by_name, iface->name, iface); @@ -4262,6 +4857,21 @@ iface_lookup(const struct bridge *br, const char *name) return shash_find_data(&br->iface_by_name, name); } +static struct iface * +iface_find(const char *name) +{ + const struct bridge *br; + + LIST_FOR_EACH (br, node, &all_bridges) { + struct iface *iface = iface_lookup(br, name); + + if (iface) { + return iface; + } + } + return NULL; +} + static struct iface * iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) { @@ -4430,6 +5040,8 @@ iface_update_cfm(struct iface *iface) mon = iface->cfg->monitor; if (!mon) { + cfm_destroy(iface->cfm); + iface->cfm = NULL; return; }