X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=bb67e17fc91d041800e6d97d358e63bc9b89365a;hb=f915f1a8ca180828983ef22cf2fd21b8f010b972;hp=6691746f14aea565b41061265043ddd5d08921e3;hpb=cf22f8cba328c0fb47a7e6811563f773fc232904;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 6691746f1..bb67e17fc 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -1,4 +1,4 @@ -/* Copyright (c) 2008, 2009, 2010 Nicira Networks +/* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -77,11 +77,27 @@ COVERAGE_DEFINE(bridge_flush); COVERAGE_DEFINE(bridge_process_flow); COVERAGE_DEFINE(bridge_reconfigure); +enum lacp_status { + LACP_STATUS_CURRENT, /* Partner is up to date. */ + LACP_STATUS_EXPIRED, /* Partner is out of date. Attempt to re-sync. */ + LACP_STATUS_DEFAULTED /* Partner information is unknown. */ +}; + 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. */ @@ -96,9 +112,20 @@ struct iface { int dp_ifidx; /* Index within kernel datapath. */ struct netdev *netdev; /* Network device. */ bool enabled; /* May be chosen for flows? */ + bool up; /* Is the interface up? */ 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 state machine 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. */ + bool lacp_attached; /* Attached to its aggregator? LACP allows + this link to be chosen for flows. */ }; #define BOND_MASK 0xff @@ -108,6 +135,12 @@ struct bond_entry { tag_type iface_tag; /* Tag associated with iface_idx. */ }; +enum bond_mode { + BM_TCP, /* Transport Layer Load Balance. */ + BM_SLB, /* Source Load Balance. */ + BM_AB /* Active Backup. */ +}; + #define MAX_MIRRORS 32 typedef uint32_t mirror_mask_t; #define MIRROR_MASK_C(X) UINT32_C(X) @@ -129,6 +162,11 @@ 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; @@ -145,17 +183,29 @@ 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. */ + + /* 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. */ 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. */ @@ -163,11 +213,11 @@ 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. */ struct mac_learning *ml; /* MAC learning table. */ + uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */ uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */ const struct ovsrec_bridge *cfg; @@ -231,6 +281,10 @@ static uint64_t dpid_from_hash(const void *, size_t nbytes); static unixctl_cb_func bridge_unixctl_fdb_show; +static void lacp_run(struct bridge *); +static void lacp_wait(struct bridge *); +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 *); @@ -249,6 +303,7 @@ static struct port *port_from_dp_ifidx(const struct bridge *, 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 *); @@ -268,9 +323,16 @@ 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 *); +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; @@ -308,6 +370,17 @@ 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 @@ -339,34 +412,22 @@ bridge_configure_once(const struct ovsrec_open_vswitch *cfg) svec_init(&dpif_types); dp_enumerate_types(&dpif_types); for (i = 0; i < dpif_types.n; i++) { - struct dpif *dpif; - int retval; size_t j; dp_enumerate_names(dpif_types.names[i], &dpif_names); - /* For each dpif... */ + /* Delete each dpif whose name is not in 'bridge_names'. */ for (j = 0; j < dpif_names.n; j++) { - retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif); - if (!retval) { - struct svec all_names; - size_t k; - - /* ...check whether any of its names is in 'bridge_names'. */ - svec_init(&all_names); - dpif_get_all_names(dpif, &all_names); - for (k = 0; k < all_names.n; k++) { - if (svec_contains(&bridge_names, all_names.names[k])) { - goto found; - } + if (!svec_contains(&bridge_names, dpif_names.names[j])) { + struct dpif *dpif; + int retval; + + retval = dpif_open(dpif_names.names[j], dpif_types.names[i], + &dpif); + if (!retval) { + dpif_delete(dpif); + dpif_close(dpif); } - - /* No. Delete the dpif. */ - dpif_delete(dpif); - - found: - svec_destroy(&all_names); - dpif_close(dpif); } } } @@ -375,30 +436,6 @@ 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. */ -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); - } - } -} - /* Callback for iterate_and_prune_ifaces(). */ static bool check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED) @@ -596,38 +633,35 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) * that port already belongs to a different datapath, so we must do all * port deletions before any port additions. */ LIST_FOR_EACH (br, node, &all_bridges) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; + struct dpif_port_dump dump; struct shash want_ifaces; + struct dpif_port dpif_port; - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); bridge_get_all_ifaces(br, &want_ifaces); - for (i = 0; i < n_dpif_ports; i++) { - const struct odp_port *p = &dpif_ports[i]; - if (!shash_find(&want_ifaces, p->devname) - && strcmp(p->devname, br->name)) { - int retval = dpif_port_del(br->dpif, p->port); + DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) { + if (!shash_find(&want_ifaces, dpif_port.name) + && 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", - p->devname, dpif_name(br->dpif), + dpif_port.name, dpif_name(br->dpif), strerror(retval)); } } } shash_destroy(&want_ifaces); - free(dpif_ports); } LIST_FOR_EACH (br, node, &all_bridges) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; struct shash cur_ifaces, want_ifaces; + struct dpif_port_dump dump; + struct dpif_port dpif_port; /* Get the set of interfaces currently in this datapath. */ - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); shash_init(&cur_ifaces); - for (i = 0; i < n_dpif_ports; i++) { - const char *name = dpif_ports[i].devname; - shash_add_once(&cur_ifaces, name, &dpif_ports[i]); + DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) { + struct dpif_port *port_info = xmalloc(sizeof *port_info); + dpif_port_clone(port_info, &dpif_port); + shash_add(&cur_ifaces, dpif_port.name, port_info); } /* Get the set of interfaces we want on this datapath. */ @@ -637,10 +671,13 @@ 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; - struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name); - const char *type = iface ? iface->type : "internal"; + struct dpif_port *dpif_port; + const char *type; int error; + type = iface ? iface->type : "internal"; + dpif_port = shash_find_data(&cur_ifaces, if_name); + /* 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). */ @@ -648,7 +685,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) || (iface && iface->netdev && strcmp(type, netdev_get_type(iface->netdev)))) { if (dpif_port) { - error = ofproto_port_del(br->ofproto, dpif_port->port); + error = ofproto_port_del(br->ofproto, dpif_port->port_no); if (error) { continue; } @@ -675,7 +712,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) shash_init(&args); if (iface) { - iface_get_options(iface->cfg, &args); + 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); @@ -708,19 +747,27 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) if (iface) { iface->netdev = netdev; iface->enabled = netdev_get_carrier(iface->netdev); + iface->up = iface->enabled; } } else if (iface && iface->netdev) { struct shash args; shash_init(&args); - iface_get_options(iface->cfg, &args); - netdev_reconfigure(iface->netdev, &args); + shash_from_ovs_idl_map(iface->cfg->key_options, + iface->cfg->value_options, + iface->cfg->n_options, &args); + netdev_set_config(iface->netdev, &args); shash_destroy(&args); } } - free(dpif_ports); - shash_destroy(&cur_ifaces); shash_destroy(&want_ifaces); + + SHASH_FOR_EACH (node, &cur_ifaces) { + struct dpif_port *port_info = node->data; + dpif_port_destroy(port_info); + free(port_info); + } + shash_destroy(&cur_ifaces); } sflow_bridge_number = 0; LIST_FOR_EACH (br, node, &all_bridges) { @@ -746,6 +793,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) br->name, strerror(error)); } } + memcpy(br->ea, ea, ETH_ADDR_LEN); dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface); ofproto_set_datapath_id(br->ofproto, dpid); @@ -866,6 +914,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) 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); @@ -993,7 +1042,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], !eth_addr_is_local(iface_ea) && !eth_addr_is_reserved(iface_ea) && !eth_addr_is_zero(iface_ea) && - memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) + eth_addr_compare_3way(iface_ea, ea) < 0) { memcpy(ea, iface_ea, ETH_ADDR_LEN); *hw_addr_iface = iface; @@ -1101,6 +1150,72 @@ 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; + + enum netdev_flags flags; + uint32_t current; + int64_t bps; + int mtu; + int64_t mtu_64; + int error; + + 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); + + error = netdev_get_flags(iface->netdev, &flags); + if (!error) { + ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down"); + } + else { + ovsrec_interface_set_admin_state(iface->cfg, NULL); + } + + error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL); + if (!error) { + ovsrec_interface_set_duplex(iface->cfg, + netdev_features_is_full_duplex(current) + ? "full" : "half"); + /* warning: uint64_t -> int64_t conversion */ + bps = netdev_features_to_bps(current); + ovsrec_interface_set_link_speed(iface->cfg, &bps, 1); + } + else { + ovsrec_interface_set_duplex(iface->cfg, NULL); + ovsrec_interface_set_link_speed(iface->cfg, NULL, 0); + } + + + ovsrec_interface_set_link_state(iface->cfg, + netdev_get_carrier(iface->netdev) + ? "up" : "down"); + + error = netdev_get_mtu(iface->netdev, &mtu); + if (!error && mtu != INT_MAX) { + mtu_64 = mtu; + ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1); + } + else { + ovsrec_interface_set_mtu(iface->cfg, NULL, 0); + } +} + static void iface_refresh_cfm_stats(struct iface *iface) { @@ -1238,6 +1353,49 @@ refresh_system_stats(const struct ovsrec_open_vswitch *cfg) &datum); } +static inline const char * +nx_role_to_str(enum nx_role role) +{ + switch (role) { + case NX_ROLE_OTHER: + return "other"; + case NX_ROLE_MASTER: + return "master"; + case NX_ROLE_SLAVE: + return "slave"; + default: + return "*** INVALID ROLE ***"; + } +} + +static void +bridge_refresh_controller_status(const struct bridge *br) +{ + struct shash info; + const struct ovsrec_controller *cfg; + + ofproto_get_ofproto_controller_info(br->ofproto, &info); + + OVSREC_CONTROLLER_FOR_EACH(cfg, idl) { + struct ofproto_controller_info *cinfo = + shash_find_data(&info, cfg->target); + + if (cinfo) { + ovsrec_controller_set_is_connected(cfg, cinfo->is_connected); + ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role)); + ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys, + (char **) cinfo->pairs.values, + cinfo->pairs.n); + } else { + ovsrec_controller_set_is_connected(cfg, false); + ovsrec_controller_set_role(cfg, NULL); + ovsrec_controller_set_status(cfg, NULL, NULL, 0); + } + } + + ofproto_free_ofproto_controller_info(&info); +} + void bridge_run(void) { @@ -1262,6 +1420,20 @@ bridge_run(void) /* (Re)configure if necessary. */ database_changed = ovsdb_idl_run(idl); cfg = ovsrec_open_vswitch_first(idl); +#ifdef HAVE_OPENSSL + /* Re-configure SSL. We do this on every trip through the main loop, + * instead of just when the database changes, because the contents of the + * key and certificate files can change without the database changing. + * + * We do this before bridge_reconfigure() because that function might + * initiate SSL connections and thus requires SSL to be configured. */ + if (cfg && cfg->ssl) { + const struct ovsrec_ssl *ssl = cfg->ssl; + + stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate); + stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert); + } +#endif if (database_changed || datapath_destroyed) { if (cfg) { struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl); @@ -1281,18 +1453,6 @@ bridge_run(void) } } -#ifdef HAVE_OPENSSL - /* Re-configure SSL. We do this on every trip through the main loop, - * instead of just when the database changes, because the contents of the - * key and certificate files can change without the database changing. */ - if (cfg && cfg->ssl) { - const struct ovsrec_ssl *ssl = cfg->ssl; - - stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate); - stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert); - } -#endif - /* Refresh system and interface stats if necessary. */ if (time_msec() >= stats_timer) { if (cfg) { @@ -1310,8 +1470,10 @@ bridge_run(void) struct iface *iface = port->ifaces[j]; iface_refresh_stats(iface); iface_refresh_cfm_stats(iface); + iface_refresh_status(iface); } } + bridge_refresh_controller_status(br); } refresh_system_stats(cfg); ovsdb_idl_txn_commit(txn); @@ -1335,6 +1497,7 @@ bridge_wait(void) } mac_learning_wait(br->ml); + lacp_wait(br); bond_wait(br); HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { @@ -1550,6 +1713,7 @@ bridge_run_one(struct bridge *br) } mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); + lacp_run(br); bond_run(br); error = ofproto_run2(br->ofproto, br->flush); @@ -1903,8 +2067,8 @@ bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) static void bridge_fetch_dp_ifaces(struct bridge *br) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; + struct dpif_port_dump dump; + struct dpif_port dpif_port; size_t i, j; /* Reset all interface numbers. */ @@ -1917,19 +2081,17 @@ bridge_fetch_dp_ifaces(struct bridge *br) } hmap_clear(&br->ifaces); - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); - for (i = 0; i < n_dpif_ports; i++) { - struct odp_port *p = &dpif_ports[i]; - struct iface *iface = iface_lookup(br, p->devname); + DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) { + struct iface *iface = iface_lookup(br, dpif_port.name); if (iface) { if (iface->dp_ifidx >= 0) { VLOG_WARN("%s reported interface %s twice", - dpif_name(br->dpif), p->devname); - } else if (iface_from_dp_ifidx(br, p->port)) { + dpif_name(br->dpif), dpif_port.name); + } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) { VLOG_WARN("%s reported interface %"PRIu16" twice", - dpif_name(br->dpif), p->port); + dpif_name(br->dpif), dpif_port.port_no); } else { - iface->dp_ifidx = p->port; + iface->dp_ifidx = dpif_port.port_no; hmap_insert(&br->ifaces, &iface->dp_ifidx_node, hash_int(iface->dp_ifidx, 0)); } @@ -1940,22 +2102,46 @@ bridge_fetch_dp_ifaces(struct bridge *br) : -1)); } } - free(dpif_ports); } /* 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) { - 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 @@ -1970,7 +2156,9 @@ 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_attached + || !(port->lacp & LACP_NEGOTIATED))) { best_down_slave = i; next_delay_expiration = iface->delay_expires; } @@ -1989,7 +2177,7 @@ 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; @@ -1997,8 +2185,14 @@ choose_output_iface(const struct port *port, const uint8_t *dl_src, assert(port->n_ifaces); if (port->n_ifaces == 1) { iface = port->ifaces[0]; + } 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 { - struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan); + 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 @@ -2020,36 +2214,58 @@ choose_output_iface(const struct port *port, const uint8_t *dl_src, } static void -bond_link_status_update(struct iface *iface, bool carrier) +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 up = iface->up; + int updelay, downdelay; + + updelay = port->updelay; + downdelay = port->downdelay; - if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { + 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_attached + && (iface->lacp_partner.state & LACP_STATE_SYNC + || iface->lacp_status == LACP_STATUS_DEFAULTED); + } + + + if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { /* Nothing to do. */ return; } - VLOG_INFO_RL(&rl, "interface %s: carrier %s", - iface->name, carrier ? "detected" : "dropped"); - if (carrier == iface->enabled) { + VLOG_INFO_RL(&rl, "interface %s: link state %s", + 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); } } @@ -2157,6 +2373,22 @@ bond_update_fake_iface_stats(struct port *port) } } +static void +bond_link_carrier_update(struct iface *iface, bool carrier) +{ + if (carrier == iface->up) { + return; + } + + if (iface->lacp_status == LACP_STATUS_CURRENT) { + iface_set_lacp_expired(iface); + } + + iface->up = carrier; + iface->lacp_tx = 0; + iface->port->bond_compat_is_stale = true; +} + static void bond_run(struct bridge *br) { @@ -2168,18 +2400,36 @@ 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; + if (port->monitor) { + assert(!port->miimon); - iface = port_lookup_iface(port, devname); - if (iface) { - bool carrier = netdev_get_carrier(iface->netdev); + /* Track carrier going up and down on interfaces. */ + while (!netdev_monitor_poll(port->monitor, &devname)) { + struct iface *iface; - bond_link_status_update(iface, carrier); - port_update_bond_compat(port); + 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); + + 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); + bond_link_carrier_update(iface, up); + } + port->bond_miimon_next_update = time_msec() + + port->bond_miimon_interval; } - free(devname); + } + + for (j = 0; j < port->n_ifaces; j++) { + bond_link_status_update(port->ifaces[j]); } for (j = 0; j < port->n_ifaces; j++) { @@ -2213,7 +2463,15 @@ 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) { @@ -2227,16 +2485,16 @@ 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 : flow->vlan_tci == 0 ? OFP_VLAN_NONE : vlan_tci_to_vid(flow->vlan_tci)); - return choose_output_iface(out_port, flow->dl_src, p->vlan, - &p->dp_ifidx, tags); + return choose_output_iface(out_port, flow, dst->vlan, + &dst->dp_ifidx, tags); } static void @@ -2253,10 +2511,10 @@ 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: @@ -2291,13 +2549,48 @@ 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; } } @@ -2331,14 +2624,14 @@ 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; int flow_vlan; - struct dst *dst = dsts; size_t i; flow_vlan = vlan_tci_to_vid(flow->vlan_tci); @@ -2347,45 +2640,42 @@ compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, } 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)) { - 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; } @@ -2395,11 +2685,11 @@ 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. */ - 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); } } } @@ -2407,17 +2697,20 @@ compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, mirrors &= mirrors - 1; } - partition_dsts(dsts, dst - dsts, flow_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); } } } @@ -2428,32 +2721,34 @@ compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan, 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 = vlan_tci_to_vid(flow->vlan_tci); if (cur_vlan == 0) { cur_vlan = OFP_VLAN_NONE; } - for (p = dsts; p < &dsts[n_dsts]; p++) { - if (p->vlan != cur_vlan) { - if (p->vlan == OFP_VLAN_NONE) { - nl_msg_put_flag(actions, ODPAT_STRIP_VLAN); + 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, ODP_ACTION_ATTR_STRIP_VLAN); } else { ovs_be16 tci; - tci = htons(p->vlan & VLAN_VID_MASK); + 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); + nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci); } - cur_vlan = p->vlan; + cur_vlan = dst->vlan; } - nl_msg_put_u32(actions, ODPAT_OUTPUT, p->dp_ifidx); + nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx); } + dst_set_free(&set); } /* Returns the effective vlan of a packet, taking into account both the @@ -2609,8 +2904,14 @@ 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; @@ -2711,6 +3012,11 @@ bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, cfm_process_heartbeat(iface->cfm, packet); } return false; + } else if (flow->dl_type == htons(ETH_TYPE_LACP)) { + if (packet) { + lacp_process_packet(packet, iface); + } + return false; } return process_flow(br, flow, packet, actions, tags, nf_output_iface); @@ -2745,14 +3051,14 @@ bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, return; } NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) { - if (nl_attr_type(a) == ODPAT_OUTPUT) { + 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) { + 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, vlan); + struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan); e->tx_bytes += n_bytes; } } @@ -2773,7 +3079,8 @@ 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); } @@ -2786,6 +3093,192 @@ static struct ofhooks bridge_ofhooks = { 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_STATUS_CURRENT; + 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; + + port->lacp_need_update = false; + + if (!port->lacp) { + return; + } + + lead = NULL; + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + struct lacp_info pri; + + iface->lacp_attached = true; + 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)) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); + VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is " + "connected to its own bridge", iface->name); + iface->lacp_attached = false; + continue; + } + + if (iface->lacp_status == LACP_STATUS_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_STATUS_DEFAULTED + || lead->lacp_partner.key != iface->lacp_partner.key + || !eth_addr_equals(lead->lacp_partner.sysid, + iface->lacp_partner.sysid)) { + iface->lacp_attached = false; + } + } +} + +static bool +lacp_iface_may_tx(const struct iface *iface) +{ + return iface->port->lacp & LACP_ACTIVE + || iface->lacp_status != LACP_STATUS_DEFAULTED; +} + +static void +lacp_run(struct bridge *br) +{ + size_t i, j; + struct ofpbuf packet; + + ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN); + + for (i = 0; i < br->n_ports; i++) { + struct port *port = br->ports[i]; + + if (!port->lacp) { + continue; + } + + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + + if (time_msec() > iface->lacp_rx) { + if (iface->lacp_status == LACP_STATUS_CURRENT) { + iface_set_lacp_expired(iface); + } else if (iface->lacp_status == LACP_STATUS_EXPIRED) { + iface_set_lacp_defaulted(iface); + } + } + } + + if (port->lacp_need_update) { + lacp_update_ifaces(port); + } + + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + 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 bridge *br) +{ + size_t i, j; + + for (i = 0; i < br->n_ports; i++) { + struct port *port = br->ports[i]; + + if (!port->lacp) { + continue; + } + + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + + if (lacp_iface_may_tx(iface)) { + poll_timer_wait_until(iface->lacp_tx); + } + + if (iface->lacp_status != LACP_STATUS_DEFAULTED) { + poll_timer_wait_until(iface->lacp_rx); + } + } + } +} + /* Bonding functions. */ /* Statistics for a single interface on a bonded port, used for load-based @@ -2800,6 +3293,22 @@ 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"; + 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(); + 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. */ @@ -2908,6 +3417,8 @@ 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)", @@ -2943,13 +3454,15 @@ 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_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 * heavily loaded slave and bals[n_bals - 1] represents the least heavily @@ -2961,6 +3474,7 @@ 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; @@ -2990,7 +3504,7 @@ bond_rebalance_port(struct port *port) while (!bals[n_bals - 1].iface->enabled) { n_bals--; if (!n_bals) { - return; + goto exit; } } @@ -3082,6 +3596,9 @@ 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 @@ -3092,7 +3609,7 @@ 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; } @@ -3105,8 +3622,15 @@ 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; } @@ -3126,9 +3650,6 @@ 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) { @@ -3158,7 +3679,7 @@ 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; @@ -3168,7 +3689,8 @@ 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) { @@ -3202,6 +3724,42 @@ 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) @@ -3216,13 +3774,42 @@ bond_unixctl_show(struct unixctl_conn *conn, return; } + ds_put_format(&ds, "bond_mode: %s\n", + bond_mode_to_string(port->bond_mode)); + + if (port->lacp) { + ds_put_format(&ds, "\tlacp: %s\n", + port->lacp & LACP_ACTIVE ? "active" : "passive"); + } else { + ds_put_cstr(&ds, "\tlacp: 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"); + + if (port->miimon) { + ds_put_format(&ds, "bond-miimon-interval: %lld\n", + port->bond_miimon_interval); + } + 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_AB) { + 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; + struct flow flow; /* Basic info. */ ds_put_format(&ds, "slave %s: %s\n", @@ -3236,7 +3823,72 @@ bond_unixctl_show(struct unixctl_conn *conn, iface->delay_expires - time_msec()); } + if (port->lacp) { + ds_put_cstr(&ds, "\tstatus: "); + + if (iface->lacp_status == LACP_STATUS_CURRENT) { + ds_put_cstr(&ds, "current "); + } else if (iface->lacp_status == LACP_STATUS_EXPIRED) { + ds_put_cstr(&ds, "expired "); + } else { + ds_put_cstr(&ds, "defaulted "); + } + + if (iface->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\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; @@ -3248,13 +3900,19 @@ 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) { @@ -3295,6 +3953,11 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, return; } + 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 { @@ -3442,7 +4105,7 @@ 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); @@ -3506,6 +4169,17 @@ 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) { @@ -3534,8 +4208,9 @@ 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, lacp_priority; unsigned long *trunks; int vlan; size_t i; @@ -3561,6 +4236,43 @@ 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 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", + 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++) { @@ -3593,9 +4305,45 @@ 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) { @@ -3714,6 +4462,49 @@ 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) { + 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) { @@ -3727,12 +4518,13 @@ 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_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]; @@ -3747,13 +4539,18 @@ 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_AB) { + 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); + } } } } @@ -3765,7 +4562,7 @@ 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; } @@ -3870,6 +4667,87 @@ port_update_vlan_compat(struct port *port) /* Interface functions. */ +static void +iface_set_lacp_defaulted(struct iface *iface) +{ + memset(&iface->lacp_partner, 0xff, sizeof iface->lacp_partner); + iface->lacp_partner.state = 0; + + iface->lacp_status = LACP_STATUS_DEFAULTED; + iface->lacp_tx = 0; + iface->port->lacp_need_update = true; +} + +static void +iface_set_lacp_expired(struct iface *iface) +{ + iface->lacp_status = LACP_STATUS_EXPIRED; + iface->lacp_partner.state |= LACP_STATE_TIME; + iface->lacp_partner.state &= ~LACP_STATE_SYNC; + + 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; + } + + if (iface->lacp_status == LACP_STATUS_DEFAULTED) { + state |= LACP_STATE_DEF; + } else if (iface->lacp_attached) { + state |= LACP_STATE_SYNC; + } + + if (iface->lacp_status == LACP_STATUS_EXPIRED) { + state |= LACP_STATE_EXP; + } + + if (iface->port->n_ifaces > 1) { + state |= LACP_STATE_AGG; + } + + 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 +iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority) +{ + 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; + } + + /* Key and state are not used in priority comparisons. */ + priority->key = 0; + priority->state = 0; +} + static void iface_send_packet(struct iface *iface, struct ofpbuf *packet) { @@ -3906,6 +4784,11 @@ 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); @@ -3934,6 +4817,10 @@ 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) { @@ -4029,6 +4916,38 @@ 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;