X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=0a6501a3675a5a5a741b7735cc31a7dd76164281;hb=83db796889bf98f8699fbd7305ffe920a0b527a1;hp=13e897fd5586061987b9e38b1417f9de1651c976;hpb=ef73f86cd4f7ec0cd2d3f9526167a0d0f6be4f63;p=openvswitch diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 13e897fd..0a6501a3 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. @@ -32,8 +32,10 @@ #include #include #include "bitmap.h" +#include "cfm.h" #include "classifier.h" #include "coverage.h" +#include "daemon.h" #include "dirs.h" #include "dpif.h" #include "dynamic-string.h" @@ -41,9 +43,11 @@ #include "hash.h" #include "hmap.h" #include "jsonrpc.h" +#include "lacp.h" #include "list.h" #include "mac-learning.h" #include "netdev.h" +#include "netlink.h" #include "odp-util.h" #include "ofp-print.h" #include "ofpbuf.h" @@ -52,7 +56,6 @@ #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" @@ -71,15 +74,32 @@ 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; 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 list port_elem; /* Element in struct port's "ifaces" list. */ struct port *port; /* Containing port. */ - size_t port_ifidx; /* Index within containing port. */ char *name; /* Host network device name. */ tag_type tag; /* Tag associated with this interface. */ long long delay_expires; /* Time after which 'enabled' may change. */ @@ -90,15 +110,25 @@ 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. */ const struct ovsrec_interface *cfg; + + /* LACP information. */ + uint16_t lacp_priority; /* LACP port priority. */ }; #define BOND_MASK 0xff struct bond_entry { - int iface_idx; /* Index of assigned iface, or -1 if none. */ + struct iface *iface; /* Assigned iface, or NULL if none. */ uint64_t tx_bytes; /* Count of bytes recently transmitted. */ - tag_type iface_tag; /* Tag associated with iface_idx. */ + tag_type tag; /* Tag for bond_entry<->iface association. */ +}; + +enum bond_mode { + BM_TCP, /* Transport Layer Load Balance. */ + BM_SLB, /* Source Load Balance. */ + BM_AB /* Active Backup. */ }; #define MAX_MIRRORS 32 @@ -132,23 +162,34 @@ struct port { const struct ovsrec_port *cfg; char *name; + /* Monitoring. */ + struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */ + long long int miimon_interval; /* Miimon status refresh interval. */ + long long int miimon_next_update; /* Time of next miimon update. */ + /* An ordinary bridge port has 1 interface. * A bridge port for bonding has at least 2 interfaces. */ - struct iface **ifaces; - size_t n_ifaces, allocated_ifaces; + struct list ifaces; /* List of "struct iface"s. */ + size_t n_ifaces; /* list_size(ifaces). */ /* Bonding info. */ - struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ - int active_iface; /* Ifidx on which bcasts accepted, or -1. */ - tag_type active_iface_tag; /* Tag for bcast flows. */ + enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */ + struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */ 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? */ long long int bond_next_fake_iface_update; /* Time of next update. */ + + /* LACP information. */ + struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */ + bool lacp_active; /* True if LACP is active */ + bool lacp_fast; /* True if LACP is in fast mode. */ + uint16_t lacp_priority; /* LACP system priority. */ + + /* 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. */ @@ -156,11 +197,11 @@ 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; @@ -219,29 +260,31 @@ static void bridge_pick_local_hw_addr(struct bridge *, static uint64_t bridge_pick_datapath_id(struct bridge *, const uint8_t bridge_ea[ETH_ADDR_LEN], struct iface *hw_addr_iface); -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 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 *); static void port_destroy(struct port *); 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 iface *port_get_an_iface(const struct port *); 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 *); @@ -253,14 +296,21 @@ 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 *); static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport); static void iface_update_qos(struct iface *, const struct ovsrec_qos *); +static void iface_update_cfm(struct iface *); +static void iface_refresh_cfm_stats(struct iface *iface); +static void iface_update_carrier(struct iface *); +static bool iface_get_carrier(const 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; @@ -291,13 +341,26 @@ 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, NULL); + lacp_init(); 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 @@ -329,34 +392,22 @@ 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); } } } @@ -365,95 +416,6 @@ bridge_configure_once(const struct ovsrec_open_vswitch *cfg) svec_destroy(&dpif_types); } -/* Initializes 'options' and fills it with the options for 'if_cfg'. Merges - * keys from "options" and "other_config", preferring "options" keys over - * "other_config" keys. - * - * The value strings in '*options' are taken directly from if_cfg, not copied, - * so the caller should not modify or free them. */ -static void -iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options) -{ - size_t i; - - shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options, - if_cfg->n_options, options); - - for (i = 0; i < if_cfg->n_other_config; i++) { - char *key = if_cfg->key_other_config[i]; - char *value = if_cfg->value_other_config[i]; - - if (!shash_find_data(options, key)) { - shash_add(options, key, value); - } else { - VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts " - "with \"options\" key %s", if_cfg->name, key, key); - } - } -} - -/* Returns the type of network device that 'iface' should have. (This is - * ordinarily the same type as the interface, but the network devices for - * "internal" ports have type "system".) */ -static const char * -iface_get_netdev_type(const struct iface *iface) -{ - return !strcmp(iface->type, "internal") ? "system" : iface->type; -} - -/* Attempt to create the network device for 'iface' through the netdev - * library. */ -static int -create_iface_netdev(struct iface *iface) -{ - struct netdev_options netdev_options; - struct shash options; - int error; - - memset(&netdev_options, 0, sizeof netdev_options); - netdev_options.name = iface->cfg->name; - netdev_options.type = iface_get_netdev_type(iface); - netdev_options.args = &options; - netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; - - iface_get_options(iface->cfg, &options); - - error = netdev_open(&netdev_options, &iface->netdev); - - if (iface->netdev) { - iface->enabled = netdev_get_carrier(iface->netdev); - } - - shash_destroy(&options); - - return error; -} - -static int -reconfigure_iface_netdev(struct iface *iface) -{ - const char *netdev_type, *iface_type; - struct shash options; - int error; - - /* Skip reconfiguration if the device has the wrong type. This shouldn't - * happen, but... */ - iface_type = iface_get_netdev_type(iface); - netdev_type = netdev_get_type(iface->netdev); - if (iface_type && strcmp(netdev_type, iface_type)) { - VLOG_WARN("%s: attempting change device type from %s to %s", - iface->cfg->name, netdev_type, iface_type); - return EINVAL; - } - - /* Reconfigure device. */ - iface_get_options(iface->cfg, &options); - error = netdev_reconfigure(iface->netdev, &options); - shash_destroy(&options); - - return error; -} - /* Callback for iterate_and_prune_ifaces(). */ static bool check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED) @@ -502,15 +464,14 @@ iterate_and_prune_ifaces(struct bridge *br, void *aux), void *aux) { - size_t i, j; + size_t i; for (i = 0; i < br->n_ports; ) { struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; ) { - struct iface *iface = port->ifaces[j]; - if (cb(br, iface, aux)) { - j++; - } else { + struct iface *iface, *next; + + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + if (!cb(br, iface, aux)) { iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } @@ -519,7 +480,7 @@ 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); } } @@ -545,13 +506,10 @@ 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]; @@ -651,38 +609,35 @@ 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), - strerror(retval)); + VLOG_WARN("failed to remove %s interface from %s: %s", + 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. */ @@ -692,20 +647,21 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) SHASH_FOR_EACH (node, &want_ifaces) { const char *if_name = node->name; struct iface *iface = node->data; - bool internal = !iface || !strcmp(iface->type, "internal"); - struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name); + 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). */ - if (internal - ? dpif_port && !(dpif_port->flags & ODP_PORT_INTERNAL) - : (iface->netdev - && strcmp(iface->type, netdev_get_type(iface->netdev)))) - { + if ((dpif_port && strcmp(dpif_port->type, type)) + || (iface && iface->netdev + && strcmp(type, netdev_get_type(iface->netdev)))) { if (dpif_port) { - error = ofproto_port_del(br->ofproto, dpif_port->port); + error = ofproto_port_del(br->ofproto, dpif_port->port_no); if (error) { continue; } @@ -717,53 +673,77 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } } - /* If it's not an internal port, open (possibly create) the - * netdev. */ - if (!internal) { - if (!iface->netdev) { - error = create_iface_netdev(iface); - if (error) { - VLOG_WARN("could not create iface %s: %s", iface->name, - strerror(error)); - continue; - } - } else { - reconfigure_iface_netdev(iface); + /* If the port doesn't exist or we don't have the netdev open, + * we need to do more work. */ + if (!dpif_port || (iface && !iface->netdev)) { + struct netdev_options options; + struct netdev *netdev; + struct shash args; + + /* First open the network device. */ + options.name = if_name; + options.type = type; + options.args = &args; + options.ethertype = NETDEV_ETH_TYPE_NONE; + + shash_init(&args); + if (iface) { + shash_from_ovs_idl_map(iface->cfg->key_options, + iface->cfg->value_options, + iface->cfg->n_options, &args); } - } + error = netdev_open(&options, &netdev); + shash_destroy(&args); - /* If it's not part of the datapath, add it. */ - if (!dpif_port) { - error = dpif_port_add(br->dpif, if_name, - internal ? ODP_PORT_INTERNAL : 0, NULL); - if (error == EFBIG) { - VLOG_ERR("ran out of valid port numbers on %s", - dpif_name(br->dpif)); - break; - } else if (error) { - VLOG_ERR("failed to add %s interface to %s: %s", - if_name, dpif_name(br->dpif), strerror(error)); + if (error) { + VLOG_WARN("could not open network device %s (%s)", + if_name, strerror(error)); continue; } - } - /* If it's an internal port, open the netdev. */ - if (internal) { - if (iface && !iface->netdev) { - error = create_iface_netdev(iface); + /* Then add the port if we haven't already. */ + if (!dpif_port) { + error = dpif_port_add(br->dpif, netdev, NULL); if (error) { - VLOG_WARN("could not create iface %s: %s", iface->name, - strerror(error)); - continue; + netdev_close(netdev); + if (error == EFBIG) { + VLOG_ERR("ran out of valid port numbers on %s", + dpif_name(br->dpif)); + break; + } else { + VLOG_WARN("failed to add %s interface to %s: %s", + if_name, dpif_name(br->dpif), + strerror(error)); + continue; + } } } - } else { - assert(iface->netdev != NULL); + + /* Update 'iface'. */ + if (iface) { + iface->netdev = netdev; + iface->enabled = iface_get_carrier(iface); + iface->up = iface->enabled; + } + } else if (iface && iface->netdev) { + struct shash args; + + shash_init(&args); + shash_from_ovs_idl_map(iface->cfg->key_options, + iface->cfg->value_options, + iface->cfg->n_options, &args); + netdev_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) { @@ -779,7 +759,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) /* Pick local port hardware address, datapath ID. */ bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); - local_iface = bridge_get_local_iface(br); + local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL); if (local_iface) { int error = netdev_set_etheraddr(local_iface->netdev, ea); if (error) { @@ -789,6 +769,7 @@ 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); @@ -905,13 +886,21 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) LIST_FOR_EACH (br, node, &all_bridges) { for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - int j; + struct iface *iface; + + if (port->monitor) { + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + netdev_monitor_add(port->monitor, iface->netdev); + } + } else { + port->miimon_next_update = 0; + } - port_update_vlan_compat(port); + port_update_lacp(port); port_update_bonding(port); - for (j = 0; j < port->n_ifaces; j++) { - iface_update_qos(port->ifaces[j], port->cfg->qos); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_update_qos(iface, port->cfg->qos); } } } @@ -919,7 +908,18 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) iterate_and_prune_ifaces(br, set_iface_properties, NULL); } + LIST_FOR_EACH (br, node, &all_bridges) { + struct iface *iface; + HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { + iface_update_cfm(iface); + } + } + free(managers); + + /* ovs-vswitchd has completed initialization, so allow the process that + * forked us to exit successfully. */ + daemonize_complete(); } static const char * @@ -949,7 +949,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], struct iface **hw_addr_iface) { const char *hwaddr; - size_t i, j; + size_t i; int error; *hw_addr_iface = NULL; @@ -969,10 +969,11 @@ 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]; + struct iface *candidate; struct iface *iface; /* Mirror output ports don't participate. */ @@ -981,12 +982,11 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], } /* Choose the MAC address to represent the port. */ + iface = NULL; if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) { /* Find the interface with this Ethernet address (if any) so that * we can provide the correct devname to the caller. */ - iface = NULL; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *candidate = port->ifaces[j]; + LIST_FOR_EACH (candidate, port_elem, &port->ifaces) { uint8_t candidate_ea[ETH_ADDR_LEN]; if (!netdev_get_etheraddr(candidate->netdev, candidate_ea) && eth_addr_equals(iface_ea, candidate_ea)) { @@ -1000,10 +1000,8 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], * scripts always add slaves to a bond in alphabetical order, so * for compatibility we choose the interface with the name that is * first in alphabetical order. */ - iface = port->ifaces[0]; - for (j = 1; j < port->n_ifaces; j++) { - struct iface *candidate = port->ifaces[j]; - if (strcmp(candidate->name, iface->name) < 0) { + LIST_FOR_EACH (candidate, port_elem, &port->ifaces) { + if (!iface || strcmp(candidate->name, iface->name) < 0) { iface = candidate; } } @@ -1029,7 +1027,7 @@ 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; @@ -1137,6 +1135,147 @@ 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, + iface_get_carrier(iface) ? "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) +{ + const struct ovsrec_monitor *mon; + const struct cfm *cfm; + size_t i; + + mon = iface->cfg->monitor; + cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); + + if (!cfm || !mon) { + return; + } + + for (i = 0; i < mon->n_remote_mps; i++) { + const struct ovsrec_maintenance_point *mp; + const struct remote_mp *rmp; + + mp = mon->remote_mps[i]; + rmp = cfm_get_remote_mp(cfm, mp->mpid); + + ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1); + } + + if (hmap_is_empty(&cfm->x_remote_mps)) { + ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0); + } else { + size_t length; + struct remote_mp *rmp; + int64_t *x_remote_mps; + + length = hmap_count(&cfm->x_remote_mps); + x_remote_mps = xzalloc(length * sizeof *x_remote_mps); + + i = 0; + HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) { + x_remote_mps[i++] = rmp->mpid; + } + + ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length); + free(x_remote_mps); + } + + if (hmap_is_empty(&cfm->x_remote_maids)) { + ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0); + } else { + size_t length; + char **x_remote_maids; + struct remote_maid *rmaid; + + length = hmap_count(&cfm->x_remote_maids); + x_remote_maids = xzalloc(length * sizeof *x_remote_maids); + + i = 0; + HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) { + size_t j; + + x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1); + + for (j = 0; j < CCM_MAID_LEN; j++) { + snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx", + rmaid->maid[j]); + } + i++; + } + ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length); + + for (i = 0; i < length; i++) { + free(x_remote_maids[i]); + } + free(x_remote_maids); + } + + ovsrec_monitor_set_fault(mon, &cfm->fault, 1); +} + static void iface_refresh_stats(struct iface *iface) { @@ -1198,6 +1337,49 @@ 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) { @@ -1222,6 +1404,20 @@ 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); @@ -1241,18 +1437,6 @@ 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) { @@ -1264,13 +1448,15 @@ bridge_run(void) for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - size_t j; + struct iface *iface; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { 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); @@ -1287,13 +1473,18 @@ bridge_wait(void) struct bridge *br; 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); + + for (i = 0; i < br->n_ports; i++) { + port_wait(br->ports[i]); + } } ovsdb_idl_wait(idl); poll_timer_wait_until(stats_timer); @@ -1308,26 +1499,6 @@ bridge_flush(struct bridge *br) br->flush = true; mac_learning_flush(br->ml); } - -/* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no - * such interface. */ -static struct iface * -bridge_get_local_iface(struct bridge *br) -{ - size_t i, j; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->dp_ifidx == ODPP_LOCAL) { - return iface; - } - } - } - - return NULL; -} /* Bridge unixctl user interface functions. */ static void @@ -1350,55 +1521,148 @@ bridge_unixctl_fdb_show(struct unixctl_conn *conn, continue; } ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", - br->ports[e->port]->ifaces[0]->dp_ifidx, + port_get_an_iface(br->ports[e->port])->dp_ifidx, e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); } unixctl_command_reply(conn, 200, ds_cstr(&ds)); ds_destroy(&ds); } -/* Bridge reconfiguration functions. */ -static struct bridge * -bridge_create(const struct ovsrec_bridge *br_cfg) +/* 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 bridge *br; + 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; - assert(!bridge_lookup(br_cfg->name)); - br = xzalloc(sizeof *br); - - error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type, - &br->dpif); - if (error) { - free(br); - return NULL; + ds_put_cstr(ds, "\n"); + if (queue_id) { + ds_put_format(ds, "Queue %u:\n", queue_id); + } else { + ds_put_cstr(ds, "Default:\n"); } - dpif_flow_flush(br->dpif); - error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks, - br, &br->ofproto); - if (error) { - VLOG_ERR("failed to create switch %s: %s", br_cfg->name, - strerror(error)); - dpif_delete(br->dpif); - dpif_close(br->dpif); - free(br); - return NULL; + SHASH_FOR_EACH (node, details) { + ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data); } - br->name = xstrdup(br_cfg->name); - br->cfg = br_cfg; - br->ml = mac_learning_create(); - eth_addr_nicira_random(br->default_ea); - - hmap_init(&br->ifaces); - - shash_init(&br->port_by_name); - shash_init(&br->iface_by_name); + 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); + } - br->flush = false; + if (stats.tx_bytes != UINT64_MAX) { + ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes); + } - list_push_back(&all_bridges, &br->node); + 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) +{ + struct bridge *br; + int error; + + assert(!bridge_lookup(br_cfg->name)); + br = xzalloc(sizeof *br); + + error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type, + &br->dpif); + if (error) { + free(br); + return NULL; + } + dpif_flow_flush(br->dpif); + + error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks, + br, &br->ofproto); + if (error) { + VLOG_ERR("failed to create switch %s: %s", br_cfg->name, + strerror(error)); + dpif_delete(br->dpif); + dpif_close(br->dpif); + free(br); + return NULL; + } + + br->name = xstrdup(br_cfg->name); + br->cfg = br_cfg; + br->ml = mac_learning_create(); + eth_addr_nicira_random(br->default_ea); + + hmap_init(&br->ifaces); + + shash_init(&br->port_by_name); + shash_init(&br->iface_by_name); + + br->flush = false; + + list_push_back(&all_bridges, &br->node); VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif)); @@ -1493,6 +1757,7 @@ bridge_unixctl_reconnect(struct unixctl_conn *conn, static int bridge_run_one(struct bridge *br) { + size_t i; int error; error = ofproto_run1(br->ofproto); @@ -1501,7 +1766,10 @@ 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; @@ -1623,7 +1891,7 @@ bridge_reconfigure_one(struct bridge *br) /* Configure OpenFlow controller connection snooping. */ svec_init(&snoops); svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); + ovs_rundir(), br->name)); svec_init(&old_snoops); ofproto_get_snoops(br->ofproto, &old_snoops); if (!svec_equal(&snoops, &old_snoops)) { @@ -1643,12 +1911,10 @@ static void bridge_ofproto_controller_for_mgmt(const struct bridge *br, struct ofproto_controller *oc) { - oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name); + oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name); oc->max_backoff = 0; oc->probe_interval = 60; oc->band = OFPROTO_OUT_OF_BAND; - oc->accept_re = NULL; - oc->update_resolv_conf = false; oc->rate_limit = 0; oc->burst_limit = 0; } @@ -1663,8 +1929,6 @@ 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); @@ -1682,13 +1946,8 @@ 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); + local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL); if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) { return; } @@ -1813,13 +2072,14 @@ bridge_reconfigure_remotes(struct bridge *br, static void bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) { - size_t i, j; + size_t i; shash_init(ifaces); for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { shash_add_once(ifaces, iface->name, iface); } if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { @@ -1839,33 +2099,32 @@ 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; - size_t i, j; + struct dpif_port_dump dump; + struct dpif_port dpif_port; + size_t i; /* Reset all interface numbers. */ for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { iface->dp_ifidx = -1; } } 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)); } @@ -1876,78 +2135,106 @@ 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 && lacp_negotiated(port->lacp); +} + static int -bond_hash(const uint8_t mac[ETH_ADDR_LEN]) +bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan) { - return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK; + return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK; +} + +static 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)]; + 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 +static struct iface * bond_choose_iface(const struct port *port) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); - size_t i, best_down_slave = -1; - long long next_delay_expiration = LLONG_MAX; - - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; + struct iface *best_down_slave; + struct iface *iface; + best_down_slave = NULL; + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { if (iface->enabled) { - return i; - } else if (iface->delay_expires < next_delay_expiration) { - best_down_slave = i; - next_delay_expiration = iface->delay_expires; + return iface; + } else if ((!best_down_slave + || iface->delay_expires < best_down_slave->delay_expires) + && lacp_slave_may_enable(port->lacp, iface)) { + best_down_slave = iface; } } - if (best_down_slave != -1) { - struct iface *iface = port->ifaces[best_down_slave]; - + if (best_down_slave) { VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay " - "since no other interface is up", iface->name, - iface->delay_expires - time_msec()); - bond_enable_slave(iface, true); + "since no other interface is up", + best_down_slave->name, + best_down_slave->delay_expires - time_msec()); + bond_enable_slave(best_down_slave, true); } return best_down_slave; } static bool -choose_output_iface(const struct port *port, const uint8_t *dl_src, - uint16_t *dp_ifidx, tag_type *tags) +choose_output_iface(const struct port *port, const struct flow *flow, + uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags) { struct iface *iface; assert(port->n_ifaces); if (port->n_ifaces == 1) { - iface = port->ifaces[0]; + iface = port_get_an_iface(port); + } else if (port->bond_mode == BM_AB) { + iface = port->active_iface; + if (!iface) { + *tags |= port->no_ifaces_tag; + return false; + } } else { - struct bond_entry *e = lookup_bond_entry(port, dl_src); - if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces - || !port->ifaces[e->iface_idx]->enabled) { + struct bond_entry *e = lookup_bond_entry(port, flow, vlan); + if (!e->iface || !e->iface->enabled) { /* XXX select interface properly. The current interface selection * is only good for testing the rebalancing code. */ - e->iface_idx = bond_choose_iface(port); - if (e->iface_idx < 0) { + e->iface = bond_choose_iface(port); + if (!e->iface) { *tags |= port->no_ifaces_tag; return false; } - e->iface_tag = tag_create_random(); - ((struct port *) port)->bond_compat_is_stale = true; + e->tag = tag_create_random(); } - *tags |= e->iface_tag; - iface = port->ifaces[e->iface_idx]; + *tags |= e->tag; + iface = e->iface; } *dp_ifidx = iface->dp_ifidx; *tags |= iface->tag; /* Currently only used for bonding. */ @@ -1955,36 +2242,46 @@ 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 && lacp_slave_may_enable(port->lacp, iface); + int updelay, downdelay; + + updelay = port->updelay; + downdelay = port->downdelay; - if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { + if (lacp_negotiated(port->lacp)) { + downdelay = 0; + updelay = 0; + } + + 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) { 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); } } @@ -1996,10 +2293,9 @@ bond_choose_active_iface(struct port *port) static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); port->active_iface = bond_choose_iface(port); - port->active_iface_tag = tag_create_random(); - if (port->active_iface >= 0) { + if (port->active_iface) { VLOG_INFO_RL(&rl, "port %s: active interface is now %s", - port->name, port->ifaces[port->active_iface]->name); + port->name, port->active_iface->name); } else { VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface", port->name); @@ -2027,10 +2323,7 @@ bond_enable_slave(struct iface *iface, bool enable) if (!iface->enabled) { VLOG_WARN("interface %s: disabled", iface->name); ofproto_revalidate(br->ofproto, iface->tag); - if (iface->port_ifidx == port->active_iface) { - ofproto_revalidate(br->ofproto, - port->active_iface_tag); - + if (iface == port->active_iface) { /* Disabling a slave can lead to another slave being immediately * enabled if there will be no active slaves but one is waiting * on an updelay. In this case we do not need to run most of the @@ -2043,7 +2336,7 @@ bond_enable_slave(struct iface *iface, bool enable) bond_send_learning_packets(port); } else { VLOG_WARN("interface %s: enabled", iface->name); - if (port->active_iface < 0 && !moving_active_iface) { + if (!port->active_iface && !moving_active_iface) { ofproto_revalidate(br->ofproto, port->no_ifaces_tag); bond_choose_active_iface(port); bond_send_learning_packets(port); @@ -2052,7 +2345,6 @@ 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 @@ -2062,14 +2354,14 @@ bond_update_fake_iface_stats(struct port *port) { struct netdev_stats bond_stats; struct netdev *bond_dev; - size_t i; + struct iface *iface; memset(&bond_stats, 0, sizeof bond_stats); - for (i = 0; i < port->n_ifaces; i++) { + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { struct netdev_stats slave_stats; - if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) { + if (!netdev_get_stats(iface->netdev, &slave_stats)) { /* XXX: We swap the stats here because they are swapped back when * reported by the internal device. The reason for this is * internal devices normally represent packets going into the system @@ -2093,83 +2385,62 @@ bond_update_fake_iface_stats(struct port *port) } static void -bond_run(struct bridge *br) +bond_run(struct port *port) { - size_t i, j; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - - if (port->n_ifaces >= 2) { - char *devname; - - /* Track carrier going up and down on interfaces. */ - while (!netdev_monitor_poll(port->monitor, &devname)) { - struct iface *iface; - - iface = port_lookup_iface(port, devname); - if (iface) { - bool carrier = netdev_get_carrier(iface->netdev); + struct iface *iface; - bond_link_status_update(iface, carrier); - port_update_bond_compat(port); - } - free(devname); - } + if (port->n_ifaces < 2) { + return; + } - 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); - } - } + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + bond_link_status_update(iface); + } - 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; - } + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + 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; + struct iface *iface; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (port->n_ifaces < 2) { - continue; - } - netdev_monitor_poll_wait(port->monitor); - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->delay_expires != LLONG_MAX) { - poll_timer_wait_until(iface->delay_expires); - } - } - if (port->bond_fake_iface) { - poll_timer_wait_until(port->bond_next_fake_iface_update); + if (port->n_ifaces < 2) { + return; + } + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + 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 -set_dst(struct dst *p, const struct flow *flow, +set_dst(struct dst *dst, const struct flow *flow, const struct port *in_port, const struct port *out_port, tag_type *tags) { - p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE + dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE : in_port->vlan >= 0 ? in_port->vlan - : ntohs(flow->dl_vlan)); - return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags); + : flow->vlan_tci == 0 ? OFP_VLAN_NONE + : vlan_tci_to_vid(flow->vlan_tci)); + return choose_output_iface(out_port, flow, dst->vlan, + &dst->dp_ifidx, tags); } static void @@ -2186,10 +2457,10 @@ 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: @@ -2224,13 +2495,48 @@ 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; } } @@ -2253,67 +2559,88 @@ port_includes_vlan(const struct port *port, uint16_t vlan) static bool port_is_floodable(const struct port *port) { - int i; + struct iface *iface; - for (i = 0; i < port->n_ifaces; i++) { + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { if (!ofproto_port_is_floodable(port->bridge->ofproto, - port->ifaces[i]->dp_ifidx)) { + iface->dp_ifidx)) { return false; } } return true; } -static size_t +/* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if + * there is no active iface. */ +static tag_type +port_get_active_iface_tag(const struct port *port) +{ + return (port->active_iface + ? port->active_iface->tag + : port->no_ifaces_tag); +} + +/* Returns an arbitrary interface within 'port'. + * + * 'port' must have at least one interface. */ +static struct iface * +port_get_an_iface(const struct port *port) +{ + return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem); +} + +static void compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, - struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface) + struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface) { mirror_mask_t mirrors = in_port->src_mirrors; - struct dst *dst = dsts; + struct dst dst; + int flow_vlan; size_t i; + flow_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (flow_vlan == 0) { + flow_vlan = OFP_VLAN_NONE; + } + if (out_port == FLOOD_PORT) { - /* XXX use ODP_FLOOD if no vlans or bonding. */ - /* XXX even better, define each VLAN as a datapath port group */ for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; if (port != in_port && port_is_floodable(port) && port_includes_vlan(port, vlan) && !port->is_mirror_output_port - && set_dst(dst, flow, in_port, port, tags)) { + && set_dst(&dst, flow, in_port, port, tags)) { mirrors |= port->dst_mirrors; - dst++; + dst_set_add(set, &dst); } } *nf_output_iface = NF_OUT_FLOOD; - } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) { - *nf_output_iface = dst->dp_ifidx; + } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) { + dst_set_add(set, &dst); + *nf_output_iface = dst.dp_ifidx; mirrors |= out_port->dst_mirrors; - dst++; } while (mirrors) { struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1]; if (!m->n_vlans || vlan_is_mirrored(m, vlan)) { if (m->out_port) { - if (set_dst(dst, flow, in_port, m->out_port, tags) - && !dst_is_duplicate(dsts, dst - dsts, dst)) { - dst++; + if (set_dst(&dst, flow, in_port, m->out_port, tags) + && !dst_is_duplicate(set, &dst)) { + dst_set_add(set, &dst); } } else { for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; if (port_includes_vlan(port, m->out_vlan) - && set_dst(dst, flow, in_port, port, tags)) + && set_dst(&dst, flow, in_port, port, tags)) { - int flow_vlan; - if (port->vlan < 0) { - dst->vlan = m->out_vlan; + dst.vlan = m->out_vlan; } - if (dst_is_duplicate(dsts, dst - dsts, dst)) { + if (dst_is_duplicate(set, &dst)) { continue; } @@ -2323,15 +2650,11 @@ compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, * tagging tags place. This is necessary because * dst->vlan is the final vlan, after removing implicit * tags. */ - flow_vlan = ntohs(flow->dl_vlan); - if (flow_vlan == 0) { - flow_vlan = OFP_VLAN_NONE; - } - if (port == in_port && dst->vlan == flow_vlan) { + if (port == in_port && dst.vlan == flow_vlan) { /* Don't send out input port on same VLAN. */ continue; } - dst++; + dst_set_add(set, &dst); } } } @@ -2339,17 +2662,20 @@ compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, mirrors &= mirrors - 1; } - partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan)); - return dst - dsts; + partition_dsts(set, flow_vlan); } static void OVS_UNUSED -print_dsts(const struct dst *dsts, size_t n) +print_dsts(const struct dst_set *set) { - for (; n--; dsts++) { - printf(">p%"PRIu16, dsts->dp_ifidx); - if (dsts->vlan != OFP_VLAN_NONE) { - printf("v%"PRIu16, dsts->vlan); + size_t i; + + for (i = 0; i < set->n; i++) { + const struct dst *dst = &set->dsts[i]; + + printf(">p%"PRIu16, dst->dp_ifidx); + if (dst->vlan != OFP_VLAN_NONE) { + printf("v%"PRIu16, dst->vlan); } } } @@ -2357,33 +2683,37 @@ print_dsts(const struct dst *dsts, size_t n) static void compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, - tag_type *tags, struct odp_actions *actions, + tag_type *tags, struct ofpbuf *actions, uint16_t *nf_output_iface) { - struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)]; - size_t n_dsts; - const struct dst *p; + struct dst_set set; uint16_t cur_vlan; + size_t i; - n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags, - nf_output_iface); + dst_set_init(&set); + compose_dsts(br, flow, vlan, in_port, out_port, &set, tags, + nf_output_iface); - cur_vlan = ntohs(flow->dl_vlan); - for (p = dsts; p < &dsts[n_dsts]; p++) { - union odp_action *a; - if (p->vlan != cur_vlan) { - if (p->vlan == OFP_VLAN_NONE) { - odp_actions_add(actions, ODPAT_STRIP_VLAN); + cur_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (cur_vlan == 0) { + cur_vlan = OFP_VLAN_NONE; + } + for (i = 0; i < set.n; i++) { + const struct dst *dst = &set.dsts[i]; + if (dst->vlan != cur_vlan) { + if (dst->vlan == OFP_VLAN_NONE) { + nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN); } else { - a = odp_actions_add(actions, ODPAT_SET_DL_TCI); - a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK); - a->dl_tci.tci |= htons(flow->dl_vlan_pcp << VLAN_PCP_SHIFT); + ovs_be16 tci; + tci = htons(dst->vlan & VLAN_VID_MASK); + tci |= flow->vlan_tci & htons(VLAN_PCP_MASK); + nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci); } - cur_vlan = p->vlan; + cur_vlan = dst->vlan; } - a = odp_actions_add(actions, ODPAT_OUTPUT); - a->output.port = p->dp_ifidx; + nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx); } + dst_set_free(&set); } /* Returns the effective vlan of a packet, taking into account both the @@ -2393,25 +2723,16 @@ compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan, static int flow_get_vlan(struct bridge *br, const struct flow *flow, struct port *in_port, bool have_packet) { - /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet - * belongs to VLAN 0, so we should treat both cases identically. (In the - * former case, the packet has an 802.1Q header that specifies VLAN 0, - * presumably to allow a priority to be specified. In the latter case, the - * packet does not have any 802.1Q header.) */ - int vlan = ntohs(flow->dl_vlan); - if (vlan == OFP_VLAN_NONE) { - vlan = 0; - } + int vlan = vlan_tci_to_vid(flow->vlan_tci); if (in_port->vlan >= 0) { if (vlan) { /* XXX support double tagging? */ if (have_packet) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " "packet received on port %s configured with " "implicit VLAN %"PRIu16, - br->name, ntohs(flow->dl_vlan), - in_port->name, in_port->vlan); + br->name, vlan, in_port->name, in_port->vlan); } return -1; } @@ -2548,14 +2869,20 @@ 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 (lacp_negotiated(in_port->lacp) && !in_iface->enabled) { + return false; + } + + /* Packets received on non-LACP bonds need special attention to avoid + * duplicates. */ + if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) { int src_idx; bool is_grat_arp_locked; if (eth_addr_is_multicast(flow->dl_dst)) { - *tags |= in_port->active_iface_tag; - if (in_port->active_iface != in_iface->port_ifidx) { + *tags |= port_get_active_iface_tag(in_port); + if (in_port->active_iface != in_iface) { /* Drop all multicast packets on inactive slaves. */ return false; } @@ -2584,7 +2911,7 @@ is_admissible(struct bridge *br, const struct flow *flow, bool have_packet, * not at all, if 'packet' was NULL. */ static bool process_flow(struct bridge *br, const struct flow *flow, - const struct ofpbuf *packet, struct odp_actions *actions, + const struct ofpbuf *packet, struct ofpbuf *actions, tag_type *tags, uint16_t *nf_output_iface) { struct port *in_port; @@ -2635,26 +2962,51 @@ done: static bool bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, - struct odp_actions *actions, tag_type *tags, + struct ofpbuf *actions, tag_type *tags, uint16_t *nf_output_iface, void *br_) { struct 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 (flow->dl_type == htons(ETH_TYPE_LACP)) { + + if (iface && iface->port->lacp && packet) { + const struct lacp_pdu *pdu = parse_lacp_packet(packet); + + if (pdu) { + COVERAGE_INC(bridge_process_lacp); + lacp_process_pdu(iface->port->lacp, iface, pdu); + } + } + return false; + } + + return true; +} + static void bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, - const union odp_action *actions, - size_t n_actions, unsigned long long int n_bytes, - void *br_) + const struct nlattr *actions, + size_t actions_len, + uint64_t n_bytes, void *br_) { struct bridge *br = br_; - const union odp_action *a; + const struct nlattr *a; struct port *in_port; tag_type dummy = 0; + unsigned int left; int vlan; /* Feed information from the active flows back into the learning table to @@ -2672,12 +3024,15 @@ bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, if (!br->has_bonded_ports) { return; } - for (a = actions; a < &actions[n_actions]; a++) { - if (a->type == ODPAT_OUTPUT) { - struct port *out_port = port_from_dp_ifidx(br, a->output.port); - if (out_port && out_port->n_ifaces >= 2) { - struct bond_entry *e = lookup_bond_entry(out_port, - flow->dl_src); + NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) { + 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_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, vlan); e->tx_bytes += n_bytes; } } @@ -2698,7 +3053,8 @@ 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_AB + && now >= port->bond_next_rebalance) { port->bond_next_rebalance = now + port->bond_rebalance_interval; bond_rebalance_port(port); } @@ -2707,6 +3063,7 @@ 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, }; @@ -2725,6 +3082,22 @@ 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. */ @@ -2735,8 +3108,8 @@ compare_bond_entries(const void *a_, const void *b_) const struct bond_entry *const *bp = b_; const struct bond_entry *a = *ap; const struct bond_entry *b = *bp; - if (a->iface_idx != b->iface_idx) { - return a->iface_idx > b->iface_idx ? 1 : -1; + if (a->iface != b->iface) { + return a->iface > b->iface ? 1 : -1; } else if (a->tx_bytes != b->tx_bytes) { return a->tx_bytes > b->tx_bytes ? 1 : -1; } else { @@ -2833,6 +3206,8 @@ 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_AB); + 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)", @@ -2860,21 +3235,24 @@ bond_shift_load(struct slave_balance *from, struct slave_balance *to, to->tx_bytes += delta; /* Arrange for flows to be revalidated. */ - ofproto_revalidate(port->bridge->ofproto, hash->iface_tag); - hash->iface_idx = to->iface->port_ifidx; - hash->iface_tag = tag_create_random(); + ofproto_revalidate(port->bridge->ofproto, hash->tag); + hash->iface = to->iface; + hash->tag = tag_create_random(); } 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; + struct iface *iface; 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 @@ -2886,25 +3264,34 @@ 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; - for (b = bals; b < &bals[n_bals]; b++) { - b->iface = port->ifaces[b - bals]; + b = bals = xmalloc(n_bals * sizeof *bals); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + b->iface = iface; b->tx_bytes = 0; b->hashes = NULL; b->n_hashes = 0; + b++; } + assert(b == &bals[n_bals]); for (i = 0; i <= BOND_MASK; i++) { hashes[i] = &port->bond_hash[i]; } qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries); for (i = 0; i <= BOND_MASK; i++) { e = hashes[i]; - if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { - b = &bals[e->iface_idx]; - b->tx_bytes += e->tx_bytes; - if (!b->hashes) { - b->hashes = &hashes[i]; + if (!e->iface) { + continue; + } + + for (b = bals; b < &bals[n_bals]; b++) { + if (b->iface == e->iface) { + b->tx_bytes += e->tx_bytes; + if (!b->hashes) { + b->hashes = &hashes[i]; + } + b->n_hashes++; + break; } - b->n_hashes++; } } qsort(bals, n_bals, sizeof *bals, compare_slave_balance); @@ -2915,7 +3302,7 @@ bond_rebalance_port(struct port *port) while (!bals[n_bals - 1].iface->enabled) { n_bals--; if (!n_bals) { - return; + goto exit; } } @@ -2982,7 +3369,6 @@ 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. */ @@ -3006,7 +3392,13 @@ bond_rebalance_port(struct port *port) * historical data to decay to <1% in 7 rebalancing runs. */ for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) { e->tx_bytes /= 2; + if (!e->tx_bytes) { + e->iface = NULL; + } } + +exit: + free(bals); } static void @@ -3017,45 +3409,33 @@ 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 || bond_is_tcp_hash(port)) { return; } ofpbuf_init(&packet, 128); error = n_packets = n_errors = 0; LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { - union ofp_action actions[2], *a; - uint16_t dp_ifidx; tag_type tags = 0; + uint16_t dp_ifidx; struct flow flow; int retval; - if (e->port == port->port_idx - || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) { + if (e->port == port->port_idx) { continue; } - /* Compose actions. */ - memset(actions, 0, sizeof actions); - a = actions; - if (e->vlan) { - a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID); - a->vlan_vid.len = htons(sizeof *a); - a->vlan_vid.vlan_vid = htons(e->vlan); - a++; + 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; } - a->output.type = htons(OFPAT_OUTPUT); - a->output.len = htons(sizeof *a); - a->output.port = htons(odp_port_to_ofp_port(dp_ifidx)); - a++; /* 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); + retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet); if (retval) { error = retval; n_errors++; @@ -3083,7 +3463,7 @@ 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; @@ -3091,12 +3471,12 @@ bond_unixctl_list(struct unixctl_conn *conn, for (i = 0; i < br->n_ports; i++) { const struct port *port = br->ports[i]; if (port->n_ifaces > 1) { - size_t j; + struct iface *iface; - ds_put_format(&ds, "%s\t%s\t", br->name, port->name); - for (j = 0; j < port->n_ifaces; j++) { - const struct iface *iface = port->ifaces[j]; - if (j) { + ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name, + bond_mode_to_string(port->bond_mode)); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + if (&iface->port_elem != list_front(&port->ifaces)) { ds_put_cstr(&ds, ", "); } ds_put_cstr(&ds, iface->name); @@ -3133,7 +3513,7 @@ bond_unixctl_show(struct unixctl_conn *conn, { struct ds ds = DS_EMPTY_INITIALIZER; const struct port *port; - size_t j; + struct iface *iface; port = bond_find(args); if (!port) { @@ -3141,18 +3521,46 @@ 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, "lacp: %s\n", + port->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->monitor ? "carrier" : "miimon"); + + if (!port->monitor) { + ds_put_format(&ds, "bond-miimon-interval: %lld\n", + port->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()); - for (j = 0; j < port->n_ifaces; j++) { - const struct iface *iface = port->ifaces[j]; + + if (port->bond_mode != BM_AB) { + ds_put_format(&ds, "next rebalance: %lld ms\n", + port->bond_next_rebalance - time_msec()); + } + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { 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) { + if (iface == port->active_iface) { ds_put_cstr(&ds, "\tactive slave\n"); } if (iface->delay_expires != LLONG_MAX) { @@ -3161,25 +3569,37 @@ bond_unixctl_show(struct unixctl_conn *conn, iface->delay_expires - time_msec()); } + 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; - if (be->iface_idx != j) { + if (be->iface != iface) { continue; } 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) == 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, &dp_ifidx, &tags) + && choose_output_iface(port, &flow, me->vlan, + &dp_ifidx, &tags) && dp_ifidx == iface->dp_ifidx) { ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n", @@ -3199,7 +3619,6 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, char *args = (char *) args_; char *save_ptr = NULL; char *bond_s, *hash_s, *slave_s; - uint8_t mac[ETH_ADDR_LEN]; struct port *port; struct iface *iface; struct bond_entry *entry; @@ -3220,10 +3639,12 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, return; } - if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) - == ETH_ADDR_SCAN_COUNT) { - hash = bond_hash(mac); - } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) { + if (port->bond_mode != BM_SLB) { + unixctl_command_reply(conn, 501, "not an SLB bond"); + return; + } + + if (strspn(hash_s, "0123456789") == strlen(hash_s)) { hash = atoi(hash_s) & BOND_MASK; } else { unixctl_command_reply(conn, 501, "bad hash"); @@ -3242,10 +3663,9 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, } entry = &port->bond_hash[hash]; - 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; + ofproto_revalidate(port->bridge->ofproto, entry->tag); + entry->iface = iface; + entry->tag = tag_create_random(); unixctl_command_reply(conn, 200, "migrated"); } @@ -3284,10 +3704,10 @@ bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, return; } - if (port->active_iface != iface->port_ifidx) { - ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); - port->active_iface = iface->port_ifidx; - port->active_iface_tag = tag_create_random(); + if (port->active_iface != iface) { + ofproto_revalidate(port->bridge->ofproto, + port_get_active_iface_tag(port)); + port->active_iface = iface; VLOG_INFO("port %s: active interface is now %s", port->name, iface->name); bond_send_learning_packets(port); @@ -3345,23 +3765,39 @@ bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, } static void -bond_unixctl_hash(struct unixctl_conn *conn, const char *args, +bond_unixctl_hash(struct unixctl_conn *conn, const char *args_, void *aux OVS_UNUSED) { - uint8_t mac[ETH_ADDR_LEN]; - uint8_t hash; - char *hash_cstr; + char *args = (char *) args_; + uint8_t mac[ETH_ADDR_LEN]; + uint8_t hash; + char *hash_cstr; + unsigned int vlan; + char *mac_s, *vlan_s; + char *save_ptr = NULL; - if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) - == ETH_ADDR_SCAN_COUNT) { - hash = bond_hash(mac); + mac_s = strtok_r(args, " ", &save_ptr); + vlan_s = strtok_r(NULL, " ", &save_ptr); + + if (vlan_s) { + if (sscanf(vlan_s, "%u", &vlan) != 1) { + unixctl_command_reply(conn, 501, "invalid vlan"); + return; + } + } else { + vlan = OFP_VLAN_NONE; + } + + if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) + == ETH_ADDR_SCAN_COUNT) { + hash = bond_hash_src(mac, vlan); - hash_cstr = xasprintf("%u", hash); - unixctl_command_reply(conn, 200, hash_cstr); - free(hash_cstr); - } else { - unixctl_command_reply(conn, 501, "invalid mac"); - } + hash_cstr = xasprintf("%u", hash); + unixctl_command_reply(conn, 200, hash_cstr); + free(hash_cstr); + } else { + unixctl_command_reply(conn, 501, "invalid mac"); + } } static void @@ -3381,6 +3817,83 @@ bond_init(void) /* Port functions. */ +static void +lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu) +{ + struct iface *iface = aux; + uint8_t ea[ETH_ADDR_LEN]; + int error; + + error = netdev_get_etheraddr(iface->netdev, ea); + if (!error) { + struct ofpbuf packet; + + ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN); + compose_lacp_packet(&packet, ea, pdu); + ofproto_send_packet(iface->port->bridge->ofproto, + iface->dp_ifidx, 0, &packet); + ofpbuf_uninit(&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)); + } +} + +static void +port_run(struct port *port) +{ + if (port->monitor) { + char *devname; + + /* Track carrier going up and down on interfaces. */ + while (!netdev_monitor_poll(port->monitor, &devname)) { + struct iface *iface; + + iface = port_lookup_iface(port, devname); + if (iface) { + iface_update_carrier(iface); + } + free(devname); + } + } else if (time_msec() >= port->miimon_next_update) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_update_carrier(iface); + } + port->miimon_next_update = time_msec() + port->miimon_interval; + } + + if (port->lacp) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + lacp_slave_enable(port->lacp, iface, iface->enabled); + } + + lacp_run(port->lacp, lacp_send_pdu_cb); + } + + bond_run(port); +} + +static void +port_wait(struct port *port) +{ + if (port->monitor) { + netdev_monitor_poll_wait(port->monitor); + } else { + poll_timer_wait_until(port->miimon_next_update); + } + + if (port->lacp) { + lacp_wait(port->lacp); + } + + bond_wait(port); +} + static struct port * port_create(struct bridge *br, const char *name) { @@ -3392,7 +3905,8 @@ port_create(struct bridge *br, const char *name) port->vlan = -1; port->trunks = NULL; port->name = xstrdup(name); - port->active_iface = -1; + port->active_iface = NULL; + list_init(&port->ifaces); if (br->n_ports >= br->allocated_ports) { br->ports = x2nrealloc(br->ports, &br->allocated_ports, @@ -3418,9 +3932,21 @@ 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) { + struct iface *iface, *next; struct shash new_ifaces; size_t i; @@ -3432,11 +3958,9 @@ port_del_ifaces(struct port *port, const struct ovsrec_port *cfg) } /* Get rid of deleted interfaces. */ - for (i = 0; i < port->n_ifaces; ) { - if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) { - iface_destroy(port->ifaces[i]); - } else { - i++; + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + if (!shash_find(&new_ifaces, iface->name)) { + iface_destroy(iface); } } @@ -3446,8 +3970,9 @@ 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; @@ -3473,6 +3998,45 @@ 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"); + + netdev_monitor_destroy(port->monitor); + port->monitor = NULL; + + if (strcmp(detect_mode, "miimon")) { + port->monitor = netdev_monitor_create(); + + if (strcmp(detect_mode, "carrier")) { + VLOG_WARN("port %s: unsupported bond-detect-mode %s, " + "defaulting to carrier", port->name, detect_mode); + } + } + + port->miimon_interval = atoi( + get_port_other_config(cfg, "bond-miimon-interval", "200")); + if (port->miimon_interval < 100) { + port->miimon_interval = 100; + } + miimon_next_update = time_msec() + port->miimon_interval; + if (port->miimon_next_update > miimon_next_update) { + port->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++) { @@ -3505,9 +4069,57 @@ 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); + port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"), + "fast"); + + 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. */ + lacp_destroy(port->lacp); + port->lacp = NULL; + } else if (!strcmp(port->cfg->lacp, "off")) { + lacp_destroy(port->lacp); + port->lacp = NULL; + } else if (!strcmp(port->cfg->lacp, "active")) { + if (!port->lacp) { + port->lacp = lacp_create(); + } + port->lacp_active = true; + } else if (!strcmp(port->cfg->lacp, "passive")) { + if (!port->lacp) { + port->lacp = lacp_create(); + } + port->lacp_active = false; + } else { + VLOG_WARN("port %s: unknown LACP mode %s", + port->name, port->cfg->lacp); + lacp_destroy(port->lacp); + port->lacp = NULL; + } + /* Get VLAN tag. */ vlan = -1; if (cfg->tag) { @@ -3573,12 +4185,10 @@ port_destroy(struct port *port) { if (port) { struct bridge *br = port->bridge; + struct iface *iface, *next; 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) { @@ -3586,8 +4196,8 @@ port_destroy(struct port *port) } } - while (port->n_ifaces > 0) { - iface_destroy(port->ifaces[port->n_ifaces - 1]); + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + iface_destroy(iface); } shash_find_and_delete_assert(&br->port_by_name, port->name); @@ -3598,7 +4208,6 @@ port_destroy(struct port *port) VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); netdev_monitor_destroy(port->monitor); - free(port->ifaces); bitmap_free(port->trunks); free(port->name); free(port); @@ -3627,157 +4236,63 @@ port_lookup_iface(const struct port *port, const char *name) } static void -port_update_bonding(struct port *port) +port_update_lacp(struct port *port) { - if (port->monitor) { - netdev_monitor_destroy(port->monitor); - port->monitor = NULL; + if (port->lacp) { + struct iface *iface; + + lacp_configure(port->lacp, port->name, + port->bridge->ea, port->lacp_priority, + port->lacp_active, port->lacp_fast); + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + lacp_slave_register(port->lacp, iface, iface->name, + iface->dp_ifidx, iface->lacp_priority); + } } +} + +static void +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; - port->bond_fake_iface = false; - } + free(port->bond_hash); + port->bond_hash = NULL; + port->bond_fake_iface = false; + port->active_iface = NULL; + port->no_ifaces_tag = 0; } 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]; - e->iface_idx = -1; + e->iface = NULL; 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(); - } - } - 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); + } else if (port->bond_mode == BM_AB) { + free(port->bond_hash); + port->bond_hash = NULL; } - } -} - -static void -port_update_bond_compat(struct port *port) -{ - struct compat_bond_hash compat_hashes[BOND_MASK + 1]; - struct compat_bond bond; - size_t i; - if (port->n_ifaces < 2) { - proc_net_compat_update_bond(port->name, NULL); - return; - } - - bond.up = false; - bond.updelay = port->updelay; - bond.downdelay = port->downdelay; - - 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; - } + if (!port->no_ifaces_tag) { + port->no_ifaces_tag = tag_create_random(); } - } - - 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; - /* 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; + if (!port->active_iface) { + bond_choose_active_iface(port); } - netdev_get_etheraddr(iface->netdev, slave->mac); - } - if (port->bond_fake_iface) { - struct netdev *bond_netdev; - - 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); + port->bond_fake_iface = port->cfg->bond_fake_iface; + if (port->bond_fake_iface) { + port->bond_next_fake_iface_update = time_msec(); } - } - proc_net_compat_update_bond(port->name, &bond); - free(bond.slaves); -} - -static void -port_update_vlan_compat(struct port *port) -{ - 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; - } - } - } } - proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan); } /* Interface functions. */ @@ -3791,7 +4306,6 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) iface = xzalloc(sizeof *iface); iface->port = port; - iface->port_ifidx = port->n_ifaces; iface->name = xstrdup(name); iface->dp_ifidx = -1; iface->tag = tag_create_random(); @@ -3801,11 +4315,9 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) shash_add_assert(&br->iface_by_name, iface->name, iface); - if (port->n_ifaces >= port->allocated_ifaces) { - port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, - sizeof *port->ifaces); - } - port->ifaces[port->n_ifaces++] = iface; + list_push_back(&port->ifaces, &iface->port_elem); + port->n_ifaces++; + if (port->n_ifaces > 1) { br->has_bonded_ports = true; } @@ -3823,8 +4335,24 @@ iface_destroy(struct iface *iface) if (iface) { struct port *port = iface->port; struct bridge *br = port->bridge; - bool del_active = port->active_iface == iface->port_ifidx; - struct iface *del; + bool del_active = port->active_iface == iface; + + if (port->bond_hash) { + struct bond_entry *e; + for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) { + if (e->iface == iface) { + e->iface = NULL; + } + } + } + + if (iface->port->lacp) { + lacp_slave_unregister(iface->port->lacp, iface); + } + + if (port->monitor && iface->netdev) { + netdev_monitor_remove(port->monitor, iface->netdev); + } shash_find_and_delete_assert(&br->iface_by_name, iface->name); @@ -3832,13 +4360,12 @@ iface_destroy(struct iface *iface) hmap_remove(&br->ifaces, &iface->dp_ifidx_node); } - del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces]; - del->port_ifidx = iface->port_ifidx; + list_remove(&iface->port_elem); + port->n_ifaces--; netdev_close(iface->netdev); if (del_active) { - ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); bond_choose_active_iface(port); bond_send_learning_packets(port); } @@ -3856,6 +4383,21 @@ 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) { @@ -3919,6 +4461,38 @@ 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; @@ -3944,6 +4518,20 @@ iface_delete_queues(unsigned int queue_id, } } +static void +iface_update_carrier(struct iface *iface) +{ + bool carrier = iface_get_carrier(iface); + if (carrier == iface->up) { + return; + } + + iface->up = carrier; + if (iface->port->lacp) { + lacp_slave_carrier_changed(iface->port->lacp, iface); + } +} + static void iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos) { @@ -3979,6 +4567,61 @@ iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos) } } } + +static void +iface_update_cfm(struct iface *iface) +{ + size_t i; + struct cfm cfm; + uint16_t *remote_mps; + struct ovsrec_monitor *mon; + uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN]; + + mon = iface->cfg->monitor; + + if (!mon) { + ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); + return; + } + + if (netdev_get_etheraddr(iface->netdev, ea)) { + VLOG_WARN("interface %s: Failed to get ethernet address. " + "Skipping Monitor.", iface->name); + return; + } + + if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) { + VLOG_WARN("interface %s: Failed to generate MAID.", iface->name); + return; + } + + cfm.mpid = mon->mpid; + cfm.interval = mon->interval ? *mon->interval : 1000; + + memcpy(cfm.eth_src, ea, sizeof cfm.eth_src); + memcpy(cfm.maid, maid, sizeof cfm.maid); + + remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps); + for(i = 0; i < mon->n_remote_mps; i++) { + remote_mps[i] = mon->remote_mps[i]->mpid; + } + + ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx, + &cfm, remote_mps, mon->n_remote_mps); + free(remote_mps); +} + +/* Read carrier or miimon status directly from 'iface''s netdev, according to + * how 'iface''s port is configured. + * + * Returns true if 'iface' is up, false otherwise. */ +static bool +iface_get_carrier(const struct iface *iface) +{ + return (iface->port->monitor + ? netdev_get_carrier(iface->netdev) + : netdev_get_miimon(iface->netdev)); +} /* Port mirroring. */