X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=07daacdb6ac7ce3471da5a491c11896ae3424599;hb=00c08589876b7c1cd8f57e5ebb3e66bb164c5a3d;hp=aee84d8a6e3870564fc0de3e13b62e706fa91423;hpb=99707a7afeea9877334825a18b027662bbabc270;p=openvswitch diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index aee84d8a..07daacdb 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. @@ -15,6 +15,7 @@ #include #include "bridge.h" +#include "byte-order.h" #include #include #include @@ -31,7 +32,11 @@ #include #include #include "bitmap.h" +#include "bond.h" +#include "cfm.h" +#include "classifier.h" #include "coverage.h" +#include "daemon.h" #include "dirs.h" #include "dpif.h" #include "dynamic-string.h" @@ -39,9 +44,11 @@ #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" @@ -50,12 +57,12 @@ #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" #include "socket-util.h" #include "stream-ssl.h" +#include "sset.h" #include "svec.h" #include "system-stats.h" #include "timeval.h" @@ -65,41 +72,45 @@ #include "vswitchd/vswitch-idl.h" #include "xenserver.h" #include "vlog.h" -#include "xtoxll.h" #include "sflow_api.h" -VLOG_DEFINE_THIS_MODULE(bridge) +VLOG_DEFINE_THIS_MODULE(bridge); + +COVERAGE_DEFINE(bridge_flush); +COVERAGE_DEFINE(bridge_process_flow); +COVERAGE_DEFINE(bridge_reconfigure); struct dst { + struct iface *iface; 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. */ /* These members are valid only after bridge_reconfigure() causes them to * be initialized. */ struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */ int dp_ifidx; /* Index within kernel datapath. */ struct netdev *netdev; /* Network device. */ - bool enabled; /* May be chosen for flows? */ const char *type; /* Usually same as cfg->type. */ const struct ovsrec_interface *cfg; }; -#define BOND_MASK 0xff -struct bond_entry { - int iface_idx; /* Index of assigned iface, or -1 if none. */ - uint64_t tx_bytes; /* Count of bytes recently transmitted. */ - tag_type iface_tag; /* Tag associated with iface_idx. */ -}; - #define MAX_MIRRORS 32 typedef uint32_t mirror_mask_t; #define MIRROR_MASK_C(X) UINT32_C(X) @@ -111,8 +122,8 @@ struct mirror { struct uuid uuid; /* UUID of this "mirror" record in database. */ /* Selection criteria. */ - struct shash src_ports; /* Name is port name; data is always NULL. */ - struct shash dst_ports; /* Name is port name; data is always NULL. */ + struct sset src_ports; /* Source port names. */ + struct sset dst_ports; /* Destination port names. */ int *vlans; size_t n_vlans; @@ -124,30 +135,20 @@ struct mirror { #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ struct port { struct bridge *bridge; - size_t port_idx; + struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */ + char *name; + int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. * NULL if all VLANs are trunked. */ const struct ovsrec_port *cfg; - char *name; /* 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. */ /* 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. */ - 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. */ - 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. */ + struct bond *bond; /* Port mirroring info. */ mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ @@ -155,11 +156,11 @@ 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; @@ -168,13 +169,11 @@ struct bridge { /* Kernel datapath information. */ struct dpif *dpif; /* Datapath. */ - struct hmap ifaces; /* Contains "struct iface"s. */ + struct hmap ifaces; /* "struct iface"s indexed by dp_ifidx. */ /* Bridge ports. */ - struct port **ports; - size_t n_ports, allocated_ports; + struct hmap ports; /* "struct port"s indexed by name. */ struct shash iface_by_name; /* "struct iface"s indexed by name. */ - struct shash port_by_name; /* "struct port"s indexed by name. */ /* Bonding. */ bool has_bonded_ports; @@ -197,6 +196,11 @@ static struct ovsdb_idl *idl; #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */ static long long int stats_timer = LLONG_MIN; +/* Stores the time after which CFM statistics may be written to the database. + * Only updated when changes to the database require rate limiting. */ +#define CFM_LIMIT_INTERVAL (1 * 1000) /* In milliseconds. */ +static long long int cfm_limiter = LLONG_MIN; + static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg); static void bridge_destroy(struct bridge *); static struct bridge *bridge_lookup(const char *name); @@ -218,29 +222,24 @@ 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 cfm_unixctl_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_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_reconfigure_bond(struct port *); +static void port_send_learning_packets(struct port *); static void mirror_create(struct bridge *, struct ovsrec_mirror *); static void mirror_destroy(struct mirror *); @@ -252,13 +251,20 @@ 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 bool iface_refresh_cfm_stats(struct iface *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; @@ -272,10 +278,10 @@ void bridge_init(const char *remote) { /* Create connection to database. */ - idl = ovsdb_idl_create(remote, &ovsrec_idl_class); + idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true); - ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg); - ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics); + ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg); + ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics); ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids); ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); @@ -283,12 +289,14 @@ bridge_init(const char *remote) ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids); ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge); - ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport); - ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics); ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids); /* Register unixctl commands. */ unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + unixctl_command_register("cfm/show", cfm_unixctl_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, @@ -296,14 +304,26 @@ bridge_init(const char *remote) bond_init(); } +void +bridge_exit(void) +{ + struct bridge *br, *next_br; + + LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) { + bridge_destroy(br); + } + ovsdb_idl_destroy(idl); +} + /* Performs configuration that is only necessary once at ovs-vswitchd startup, * but for which the ovs-vswitchd configuration 'cfg' is required. */ static void bridge_configure_once(const struct ovsrec_open_vswitch *cfg) { static bool already_configured_once; - struct svec bridge_names; - struct svec dpif_names, dpif_types; + struct sset bridge_names; + struct sset dpif_names, dpif_types; + const char *type; size_t i; /* Only do this once per ovs-vswitchd run. */ @@ -315,141 +335,38 @@ bridge_configure_once(const struct ovsrec_open_vswitch *cfg) stats_timer = time_msec() + STATS_INTERVAL; /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */ - svec_init(&bridge_names); + sset_init(&bridge_names); for (i = 0; i < cfg->n_bridges; i++) { - svec_add(&bridge_names, cfg->bridges[i]->name); + sset_add(&bridge_names, cfg->bridges[i]->name); } - svec_sort(&bridge_names); /* Iterate over all system dpifs and delete any of them that do not appear * in 'cfg'. */ - svec_init(&dpif_names); - svec_init(&dpif_types); + sset_init(&dpif_names); + sset_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... */ - 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; - } - } - - /* No. Delete the dpif. */ - dpif_delete(dpif); - - found: - svec_destroy(&all_names); - dpif_close(dpif); - } - } - } - svec_destroy(&bridge_names); - svec_destroy(&dpif_names); - 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; + SSET_FOR_EACH (type, &dpif_types) { + const char *name; - shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options, - if_cfg->n_options, options); + dp_enumerate_names(type, &dpif_names); - 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]; + /* Delete each dpif whose name is not in 'bridge_names'. */ + SSET_FOR_EACH (name, &dpif_names) { + if (!sset_contains(&bridge_names, name)) { + struct dpif *dpif; + int retval; - 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); + retval = dpif_open(name, type, &dpif); + if (!retval) { + dpif_delete(dpif); + dpif_close(dpif); + } + } } } -} - -/* Returns the type of network device that 'iface' should have. (This is - * ordinarily the same type as the interface, but the network devices for - * "internal" ports have type "system".) */ -static const char * -iface_get_netdev_type(const struct iface *iface) -{ - return !strcmp(iface->type, "internal") ? "system" : iface->type; -} - -/* Attempt to create the network device for 'iface' through the netdev - * library. */ -static int -create_iface_netdev(struct iface *iface) -{ - struct netdev_options netdev_options; - struct shash options; - int error; - - memset(&netdev_options, 0, sizeof netdev_options); - netdev_options.name = iface->cfg->name; - netdev_options.type = iface_get_netdev_type(iface); - netdev_options.args = &options; - netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; - - iface_get_options(iface->cfg, &options); - - error = netdev_open(&netdev_options, &iface->netdev); - - if (iface->netdev) { - netdev_get_carrier(iface->netdev, &iface->enabled); - } - - shash_destroy(&options); - - return error; -} - -static int -reconfigure_iface_netdev(struct iface *iface) -{ - const char *netdev_type, *iface_type; - struct shash options; - int error; - - /* Skip reconfiguration if the device has the wrong type. This shouldn't - * happen, but... */ - iface_type = iface_get_netdev_type(iface); - netdev_type = netdev_get_type(iface->netdev); - if (iface_type && strcmp(netdev_type, iface_type)) { - VLOG_WARN("%s: attempting change device type from %s to %s", - iface->cfg->name, netdev_type, iface_type); - return EINVAL; - } - - /* Reconfigure device. */ - iface_get_options(iface->cfg, &options); - error = netdev_reconfigure(iface->netdev, &options); - shash_destroy(&options); - - return error; + sset_destroy(&bridge_names); + sset_destroy(&dpif_names); + sset_destroy(&dpif_types); } /* Callback for iterate_and_prune_ifaces(). */ @@ -500,23 +417,20 @@ iterate_and_prune_ifaces(struct bridge *br, void *aux), void *aux) { - size_t i, j; - - 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 port *port, *next_port; + + HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) { + struct iface *iface, *next_iface; + + LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) { + if (!cb(br, iface, aux)) { + iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } } - if (port->n_ifaces) { - i++; - } else { - VLOG_ERR("%s port has no interfaces, dropping", port->name); + if (list_is_empty(&port->ifaces)) { + VLOG_WARN("%s port has no interfaces, dropping", port->name); port_destroy(port); } } @@ -531,30 +445,48 @@ iterate_and_prune_ifaces(struct bridge *br, * should not be and in fact is not directly involved in that. But * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so * it has to tell in-band control where the managers are to enable that. + * (Thus, only managers connected in-band are collected.) */ static void -collect_managers(const struct ovsrec_open_vswitch *ovs_cfg, - struct sockaddr_in **managersp, size_t *n_managersp) +collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg, + struct sockaddr_in **managersp, size_t *n_managersp) { struct sockaddr_in *managers = NULL; size_t n_managers = 0; + struct sset targets; + size_t i; - if (ovs_cfg->n_managers > 0) { - size_t i; + /* 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. */ + sset_init(&targets); + for (i = 0; i < ovs_cfg->n_manager_options; i++) { + struct ovsrec_manager *m = ovs_cfg->manager_options[i]; + + if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) { + sset_find_and_delete(&targets, m->target); + } else { + sset_add(&targets, m->target); + } + } + + /* Now extract the targets' IP addresses. */ + if (!sset_is_empty(&targets)) { + const char *target; - managers = xmalloc(ovs_cfg->n_managers * sizeof *managers); - for (i = 0; i < ovs_cfg->n_managers; i++) { - const char *name = ovs_cfg->managers[i]; - struct sockaddr_in *sin = &managers[i]; + managers = xmalloc(sset_count(&targets) * sizeof *managers); + SSET_FOR_EACH (target, &targets) { + struct sockaddr_in *sin = &managers[n_managers]; - if ((!strncmp(name, "tcp:", 4) - && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) || - (!strncmp(name, "ssl:", 4) - && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) { + if ((!strncmp(target, "tcp:", 4) + && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) || + (!strncmp(target, "ssl:", 4) + && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) { n_managers++; } } } + sset_destroy(&targets); *managersp = managers; *n_managersp = n_managers; @@ -573,7 +505,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) COVERAGE_INC(bridge_reconfigure); - collect_managers(ovs_cfg, &managers, &n_managers); + collect_in_band_managers(ovs_cfg, &managers, &n_managers); /* Collect old and new bridges. */ shash_init(&old_br); @@ -626,38 +558,35 @@ 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. */ @@ -667,20 +596,21 @@ 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; } @@ -692,57 +622,79 @@ 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; + } + } 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) { - uint8_t ea[8]; + uint8_t ea[ETH_ADDR_LEN]; uint64_t dpid; struct iface *local_iface; struct iface *hw_addr_iface; @@ -750,11 +702,15 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) bridge_fetch_dp_ifaces(br); + /* Delete interfaces that cannot be opened. + * + * From this point forward we are guaranteed that every "struct iface" + * has nonnull 'netdev' and correct 'dp_ifidx'. */ iterate_and_prune_ifaces(br, check_iface, NULL); /* 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) { @@ -764,6 +720,7 @@ 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); @@ -804,19 +761,21 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) "with another vswitch, choose an engine id less " "than 128", br->name); } - if (br->n_ports > 508) { + if (hmap_count(&br->ports) > 508) { VLOG_WARN("bridge %s: netflow port mangling will conflict " "with another port when more than 508 ports are " "used", br->name); } } - opts.collectors.n = nf_cfg->n_targets; - opts.collectors.names = nf_cfg->targets; + sset_init(&opts.collectors); + sset_add_array(&opts.collectors, + nf_cfg->targets, nf_cfg->n_targets); if (ofproto_set_netflow(br->ofproto, &opts)) { VLOG_ERR("bridge %s: problem setting netflow collectors", br->name); } + sset_destroy(&opts.collectors); } else { ofproto_set_netflow(br->ofproto, NULL); } @@ -830,8 +789,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) memset(&oso, 0, sizeof oso); - oso.targets.n = sflow_cfg->n_targets; - oso.targets.names = sflow_cfg->targets; + sset_init(&oso.targets); + sset_add_array(&oso.targets, + sflow_cfg->targets, sflow_cfg->n_targets); oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE; if (sflow_cfg->sampling) { @@ -861,7 +821,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } ofproto_set_sflow(br->ofproto, &oso); - /* Do not destroy oso.targets because it is owned by sflow_cfg. */ + sset_destroy(&oso.targets); } else { ofproto_set_sflow(br->ofproto, NULL); } @@ -878,15 +838,16 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) bridge_reconfigure_remotes(br, managers, n_managers); } LIST_FOR_EACH (br, node, &all_bridges) { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - int j; + struct port *port; + + br->has_bonded_ports = false; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; - port_update_vlan_compat(port); - port_update_bonding(port); + port_reconfigure_bond(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); } } } @@ -894,7 +855,18 @@ 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 * @@ -924,7 +896,7 @@ 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; + struct port *port; int error; *hw_addr_iface = NULL; @@ -944,10 +916,10 @@ 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); - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + memset(ea, 0xff, ETH_ADDR_LEN); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { uint8_t iface_ea[ETH_ADDR_LEN]; + struct iface *candidate; struct iface *iface; /* Mirror output ports don't participate. */ @@ -956,12 +928,11 @@ 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)) { @@ -975,10 +946,8 @@ 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; } } @@ -1004,7 +973,7 @@ 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; @@ -1112,6 +1081,109 @@ 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); + } +} + +/* Writes 'iface''s CFM statistics to the database. Returns true if anything + * changed, false otherwise. */ +static bool +iface_refresh_cfm_stats(struct iface *iface) +{ + const struct ovsrec_monitor *mon; + const struct cfm *cfm; + bool changed = false; + size_t i; + + mon = iface->cfg->monitor; + cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); + + if (!cfm || !mon) { + return false; + } + + 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); + + if (mp->n_fault != 1 || mp->fault[0] != rmp->fault) { + ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1); + changed = true; + } + } + + if (mon->n_fault != 1 || mon->fault[0] != cfm->fault) { + ovsrec_monitor_set_fault(mon, &cfm->fault, 1); + changed = true; + } + + return changed; +} + static void iface_refresh_stats(struct iface *iface) { @@ -1173,6 +1245,49 @@ 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) { @@ -1197,6 +1312,20 @@ 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); @@ -1216,18 +1345,6 @@ 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) { @@ -1235,17 +1352,17 @@ bridge_run(void) txn = ovsdb_idl_txn_create(idl); LIST_FOR_EACH (br, node, &all_bridges) { - size_t i; + struct port *port; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - size_t j; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + 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_status(iface); } } + bridge_refresh_controller_status(br); } refresh_system_stats(cfg); ovsdb_idl_txn_commit(txn); @@ -1254,6 +1371,31 @@ bridge_run(void) stats_timer = time_msec() + STATS_INTERVAL; } + + if (time_msec() >= cfm_limiter) { + struct ovsdb_idl_txn *txn; + bool changed = false; + + txn = ovsdb_idl_txn_create(idl); + LIST_FOR_EACH (br, node, &all_bridges) { + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + changed = iface_refresh_cfm_stats(iface) || changed; + } + } + } + + if (changed) { + cfm_limiter = time_msec() + CFM_LIMIT_INTERVAL; + } + + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); + } } void @@ -1262,16 +1404,20 @@ bridge_wait(void) struct bridge *br; LIST_FOR_EACH (br, node, &all_bridges) { - ofproto_wait(br->ofproto); - if (ofproto_has_primary_controller(br->ofproto)) { - continue; - } + struct port *port; + ofproto_wait(br->ofproto); mac_learning_wait(br->ml); - bond_wait(br); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port_wait(port); + } } ovsdb_idl_wait(idl); poll_timer_wait_until(stats_timer); + + if (cfm_limiter > time_msec()) { + poll_timer_wait_until(cfm_limiter); + } } /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s @@ -1281,27 +1427,6 @@ bridge_flush(struct bridge *br) { COVERAGE_INC(bridge_flush); 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. */ @@ -1321,55 +1446,171 @@ bridge_unixctl_fdb_show(struct unixctl_conn *conn, ds_put_cstr(&ds, " port VLAN MAC Age\n"); LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { - if (e->port < 0 || e->port >= br->n_ports) { - continue; - } + struct port *port = e->port.p; ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", - br->ports[e->port]->ifaces[0]->dp_ifidx, + port_get_an_iface(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) +/* CFM unixctl user interface functions. */ +static void +cfm_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) { - struct bridge *br; - int error; - - assert(!bridge_lookup(br_cfg->name)); - br = xzalloc(sizeof *br); + struct ds ds = DS_EMPTY_INITIALIZER; + struct iface *iface; + const struct cfm *cfm; - error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type, - &br->dpif); - if (error) { - free(br); - return NULL; + iface = iface_find(args); + if (!iface) { + unixctl_command_reply(conn, 501, "no such interface"); + return; } - 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; + cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); + + if (!cfm) { + unixctl_command_reply(conn, 501, "CFM not enabled"); + return; } - br->name = xstrdup(br_cfg->name); - br->cfg = br_cfg; - br->ml = mac_learning_create(); - eth_addr_nicira_random(br->default_ea); + cfm_dump_ds(cfm, &ds); + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + ds_destroy(&ds); +} + +/* QoS unixctl user interface functions. */ - hmap_init(&br->ifaces); +struct qos_unixctl_show_cbdata { + struct ds *ds; + struct iface *iface; +}; - shash_init(&br->port_by_name); - shash_init(&br->iface_by_name); +static void +qos_unixctl_show_cb(unsigned int queue_id, + const struct shash *details, + void *aux) +{ + struct qos_unixctl_show_cbdata *data = aux; + struct ds *ds = data->ds; + struct iface *iface = data->iface; + struct netdev_queue_stats stats; + struct shash_node *node; + int error; + + ds_put_cstr(ds, "\n"); + if (queue_id) { + ds_put_format(ds, "Queue %u:\n", queue_id); + } else { + ds_put_cstr(ds, "Default:\n"); + } + + SHASH_FOR_EACH (node, details) { + ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data); + } + + error = netdev_get_queue_stats(iface->netdev, queue_id, &stats); + if (!error) { + if (stats.tx_packets != UINT64_MAX) { + ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets); + } + + if (stats.tx_bytes != UINT64_MAX) { + ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes); + } + + if (stats.tx_errors != UINT64_MAX) { + ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors); + } + } else { + ds_put_format(ds, "\tFailed to get statistics for queue %u: %s", + queue_id, strerror(error)); + } +} + +static void +qos_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct shash sh = SHASH_INITIALIZER(&sh); + struct iface *iface; + const char *type; + struct shash_node *node; + struct qos_unixctl_show_cbdata data; + int error; + + iface = iface_find(args); + if (!iface) { + unixctl_command_reply(conn, 501, "no such interface"); + return; + } + + netdev_get_qos(iface->netdev, &type, &sh); + + if (*type != '\0') { + ds_put_format(&ds, "QoS: %s %s\n", iface->name, type); + + SHASH_FOR_EACH (node, &sh) { + ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data); + } + + data.ds = &ds; + data.iface = iface; + error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data); + + if (error) { + ds_put_format(&ds, "failed to dump queues: %s", strerror(error)); + } + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + } else { + ds_put_format(&ds, "QoS not configured on %s\n", iface->name); + unixctl_command_reply(conn, 501, ds_cstr(&ds)); + } + + shash_destroy_free_data(&sh); + ds_destroy(&ds); +} + +/* Bridge reconfiguration functions. */ +static struct bridge * +bridge_create(const struct ovsrec_bridge *br_cfg) +{ + 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; + } + + 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->ports); + hmap_init(&br->ifaces); + shash_init(&br->iface_by_name); br->flush = false; @@ -1384,24 +1625,24 @@ static void bridge_destroy(struct bridge *br) { if (br) { + struct port *port, *next; int error; - while (br->n_ports > 0) { - port_destroy(br->ports[br->n_ports - 1]); + HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) { + port_destroy(port); } list_remove(&br->node); + ofproto_destroy(br->ofproto); error = dpif_delete(br->dpif); if (error && error != ENOENT) { VLOG_ERR("failed to delete %s: %s", dpif_name(br->dpif), strerror(error)); } dpif_close(br->dpif); - ofproto_destroy(br->ofproto); mac_learning_destroy(br->ml); hmap_destroy(&br->ifaces); - shash_destroy(&br->port_by_name); + hmap_destroy(&br->ports); shash_destroy(&br->iface_by_name); - free(br->ports); free(br->name); free(br); } @@ -1468,6 +1709,7 @@ bridge_unixctl_reconnect(struct unixctl_conn *conn, static int bridge_run_one(struct bridge *br) { + struct port *port; int error; error = ofproto_run1(br->ofproto); @@ -1476,7 +1718,10 @@ bridge_run_one(struct bridge *br) } mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); - bond_run(br); + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port_run(port); + } error = ofproto_run2(br->ofproto, br->flush); br->flush = false; @@ -1508,18 +1753,12 @@ bridge_get_controllers(const struct bridge *br, static void bridge_reconfigure_one(struct bridge *br) { - struct shash old_ports, new_ports; - struct svec snoops, old_snoops; - struct shash_node *node; enum ofproto_fail_mode fail_mode; + struct port *port, *next; + struct shash_node *node; + struct shash new_ports; size_t i; - /* Collect old ports. */ - shash_init(&old_ports); - for (i = 0; i < br->n_ports; i++) { - shash_add(&old_ports, br->ports[i]->name, br->ports[i]); - } - /* Collect new ports. */ shash_init(&new_ports); for (i = 0; i < br->cfg->n_ports; i++) { @@ -1549,11 +1788,10 @@ bridge_reconfigure_one(struct bridge *br) /* Get rid of deleted ports. * Get rid of deleted interfaces on ports that still exist. */ - SHASH_FOR_EACH (node, &old_ports) { - struct port *port = node->data; + HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) { const struct ovsrec_port *port_cfg; - port_cfg = shash_find_data(&new_ports, node->name); + port_cfg = shash_find_data(&new_ports, port->name); if (!port_cfg) { port_destroy(port); } else { @@ -1565,19 +1803,18 @@ bridge_reconfigure_one(struct bridge *br) * Add new interfaces to existing ports. * Reconfigure existing ports. */ SHASH_FOR_EACH (node, &new_ports) { - struct port *port = shash_find_data(&old_ports, node->name); + struct port *port = port_lookup(br, node->name); if (!port) { port = port_create(br, node->name); } port_reconfigure(port, node->data); - if (!port->n_ifaces) { + if (list_is_empty(&port->ifaces)) { VLOG_WARN("bridge %s: port %s has no interfaces, dropping", br->name, port->name); port_destroy(port); } } - shash_destroy(&old_ports); shash_destroy(&new_ports); /* Set the fail-mode */ @@ -1596,16 +1833,15 @@ bridge_reconfigure_one(struct bridge *br) * controller to another?) */ /* Configure OpenFlow controller connection snooping. */ - svec_init(&snoops); - svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); - svec_init(&old_snoops); - ofproto_get_snoops(br->ofproto, &old_snoops); - if (!svec_equal(&snoops, &old_snoops)) { + if (!ofproto_has_snoops(br->ofproto)) { + struct sset snoops; + + sset_init(&snoops); + sset_add_and_free(&snoops, xasprintf("punix:%s/%s.snoop", + ovs_rundir(), br->name)); ofproto_set_snoops(br->ofproto, &snoops); + sset_destroy(&snoops); } - svec_destroy(&snoops); - svec_destroy(&old_snoops); mirror_reconfigure(br); } @@ -1618,12 +1854,10 @@ 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; } @@ -1638,8 +1872,6 @@ 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); @@ -1657,13 +1889,8 @@ 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; } @@ -1699,23 +1926,29 @@ bridge_reconfigure_remotes(struct bridge *br, const struct sockaddr_in *managers, size_t n_managers) { + const char *disable_ib_str, *queue_id_str; + bool disable_in_band = false; + int queue_id; + struct ovsrec_controller **controllers; size_t n_controllers; bool had_primary; - const char *disable_ib_str; - bool disable_in_band = false; struct ofproto_controller *ocs; size_t n_ocs; size_t i; - /* Check if we should disable in-band control on this bridge. */ disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band"); if (disable_ib_str && !strcmp(disable_ib_str, "true")) { disable_in_band = true; } + /* Set OpenFlow queue ID for in-band control. */ + queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue"); + queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1; + ofproto_set_in_band_queue(br->ofproto, queue_id); + if (disable_in_band) { ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0); } else { @@ -1768,30 +2001,30 @@ bridge_reconfigure_remotes(struct bridge *br, if (!n_controllers && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) { union ofp_action action; - struct flow flow; + struct cls_rule rule; memset(&action, 0, sizeof action); action.type = htons(OFPAT_OUTPUT); action.output.len = htons(sizeof action); action.output.port = htons(OFPP_NORMAL); - memset(&flow, 0, sizeof flow); - ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0); + cls_rule_init_catchall(&rule, 0); + ofproto_add_flow(br->ofproto, &rule, &action, 1); } } static void bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) { - size_t i, j; + struct port *port; 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]; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + 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) { + if (!list_is_short(&port->ifaces) && port->cfg->bond_fake_iface) { shash_add_once(ifaces, port->name, NULL); } } @@ -1808,399 +2041,112 @@ 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; + struct port *port; /* 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]; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + 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)); } - if (iface->cfg) { - int64_t ofport = (iface->dp_ifidx >= 0 - ? odp_port_to_ofp_port(iface->dp_ifidx) - : -1); - ovsrec_interface_set_ofport(iface->cfg, &ofport, 1); - } + iface_set_ofport(iface->cfg, + (iface->dp_ifidx >= 0 + ? odp_port_to_ofp_port(iface->dp_ifidx) + : -1)); } } - free(dpif_ports); } /* Bridge packet processing functions. */ -static int -bond_hash(const uint8_t mac[ETH_ADDR_LEN]) -{ - return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK; -} - -static struct bond_entry * -lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN]) -{ - return &port->bond_hash[bond_hash(mac)]; -} - -static int -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]; - - if (iface->enabled) { - return i; - } else if (iface->delay_expires < next_delay_expiration) { - best_down_slave = i; - next_delay_expiration = iface->delay_expires; - } - } - - if (best_down_slave != -1) { - struct iface *iface = port->ifaces[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); - } - - 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) -{ - struct iface *iface; - - assert(port->n_ifaces); - if (port->n_ifaces == 1) { - iface = port->ifaces[0]; - } else { - struct bond_entry *e = lookup_bond_entry(port, dl_src); - if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces - || !port->ifaces[e->iface_idx]->enabled) { - /* XXX select interface properly. The current interface selection - * is only good for testing the rebalancing code. */ - e->iface_idx = bond_choose_iface(port); - if (e->iface_idx < 0) { - *tags |= port->no_ifaces_tag; - return false; - } - e->iface_tag = tag_create_random(); - ((struct port *) port)->bond_compat_is_stale = true; - } - *tags |= e->iface_tag; - iface = port->ifaces[e->iface_idx]; - } - *dp_ifidx = iface->dp_ifidx; - *tags |= iface->tag; /* Currently only used for bonding. */ - return true; -} - -static void -bond_link_status_update(struct iface *iface, bool carrier) +set_dst(struct dst *dst, const struct flow *flow, + const struct port *in_port, const struct port *out_port, + tag_type *tags) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); - struct port *port = iface->port; - - if ((carrier == 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) { - 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) { - bond_enable_slave(iface, true); - if (port->updelay) { - VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " - "other interface is up", iface->name, port->updelay); - } - } else { - int delay = carrier ? port->updelay : port->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", - delay); - } - } -} + dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE + : in_port->vlan >= 0 ? in_port->vlan + : flow->vlan_tci == 0 ? OFP_VLAN_NONE + : vlan_tci_to_vid(flow->vlan_tci)); -static void -bond_choose_active_iface(struct port *port) -{ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); + dst->iface = (!out_port->bond + ? port_get_an_iface(out_port) + : bond_choose_output_slave(out_port->bond, flow, + dst->vlan, tags)); - port->active_iface = bond_choose_iface(port); - port->active_iface_tag = tag_create_random(); - if (port->active_iface >= 0) { - VLOG_INFO_RL(&rl, "port %s: active interface is now %s", - port->name, port->ifaces[port->active_iface]->name); - } else { - VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface", - port->name); - } + return dst->iface != NULL; } -static void -bond_enable_slave(struct iface *iface, bool enable) +static int +mirror_mask_ffs(mirror_mask_t mask) { - struct port *port = iface->port; - struct bridge *br = port->bridge; - - /* This acts as a recursion check. If the act of disabling a slave - * causes a different slave to be enabled, the flag will allow us to - * skip redundant work when we reenter this function. It must be - * cleared on exit to keep things safe with multiple bonds. */ - static bool moving_active_iface = false; - - iface->delay_expires = LLONG_MAX; - if (enable == iface->enabled) { - return; - } - - iface->enabled = 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); - - /* 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 - * code for the newly enabled slave since there was no period - * without an active slave and it is redundant with the disabling - * path. */ - moving_active_iface = true; - bond_choose_active_iface(port); - } - bond_send_learning_packets(port); - } else { - VLOG_WARN("interface %s: enabled", iface->name); - if (port->active_iface < 0 && !moving_active_iface) { - ofproto_revalidate(br->ofproto, port->no_ifaces_tag); - bond_choose_active_iface(port); - bond_send_learning_packets(port); - } - iface->tag = tag_create_random(); - } - - moving_active_iface = false; - port->bond_compat_is_stale = true; + BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); + return ffs(mask); } -/* Attempts to make the sum of the bond slaves' statistics appear on the fake - * bond interface. */ static void -bond_update_fake_iface_stats(struct port *port) +dst_set_init(struct dst_set *set) { - struct netdev_stats bond_stats; - struct netdev *bond_dev; - size_t i; - - memset(&bond_stats, 0, sizeof bond_stats); - - for (i = 0; i < port->n_ifaces; i++) { - struct netdev_stats slave_stats; - - if (!netdev_get_stats(port->ifaces[i]->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 - * but when used as fake bond device they represent packets leaving - * the system. We really should do this in the internal device - * itself because changing it here reverses the counts from the - * perspective of the switch. However, the internal device doesn't - * know what type of device it represents so we have to do it here - * for now. */ - bond_stats.tx_packets += slave_stats.rx_packets; - bond_stats.tx_bytes += slave_stats.rx_bytes; - bond_stats.rx_packets += slave_stats.tx_packets; - bond_stats.rx_bytes += slave_stats.tx_bytes; - } - } - - if (!netdev_open_default(port->name, &bond_dev)) { - netdev_set_stats(bond_dev, &bond_stats); - netdev_close(bond_dev); - } + set->dsts = set->builtin; + set->n = 0; + set->allocated = ARRAY_SIZE(set->builtin); } static void -bond_run(struct bridge *br) +dst_set_add(struct dst_set *set, const struct dst *dst) { - 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; - bool carrier; - - iface = port_lookup_iface(port, devname); - if (iface && !netdev_get_carrier(iface->netdev, &carrier)) { - bond_link_status_update(iface, carrier); - port_update_bond_compat(port); - } - free(devname); - } + if (set->n >= set->allocated) { + size_t new_allocated; + struct dst *new_dsts; - 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); - } - } - - 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; - } - } - - if (port->bond_compat_is_stale) { - port->bond_compat_is_stale = false; - port_update_bond_compat(port); - } - } -} + new_allocated = set->allocated * 2; + new_dsts = xmalloc(new_allocated * sizeof *new_dsts); + memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); -static void -bond_wait(struct bridge *br) -{ - size_t i, j; + dst_set_free(set); - 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); - } + set->dsts = new_dsts; + set->allocated = new_allocated; } + set->dsts[set->n++] = *dst; } -static bool -set_dst(struct dst *p, 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 - : in_port->vlan >= 0 ? in_port->vlan - : ntohs(flow->dl_vlan)); - return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags); -} - -static void -swap_dst(struct dst *p, struct dst *q) -{ - struct dst tmp = *p; - *p = *q; - *q = tmp; -} - -/* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in - * 'dsts'. (This may help performance by reducing the number of VLAN changes - * that we push to the datapath. We could in fact fully sort the array by - * 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) +dst_set_free(struct dst_set *set) { - struct dst *first = dsts; - struct dst *last = dsts + n_dsts; - - while (first != last) { - /* Invariants: - * - All dsts < first have vlan == 'vlan'. - * - All dsts >= last have vlan != 'vlan'. - * - first < last. */ - while (first->vlan == vlan) { - if (++first == last) { - return; - } - } - - /* Same invariants, plus one additional: - * - first->vlan != vlan. - */ - while (last[-1].vlan != vlan) { - if (--last == first) { - return; - } - } - - /* Same invariants, plus one additional: - * - last[-1].vlan == vlan.*/ - swap_dst(first++, --last); + if (set->dsts != set->builtin) { + free(set->dsts); } } -static int -mirror_mask_ffs(mirror_mask_t mask) -{ - BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); - return ffs(mask); -} - 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].iface == test->iface) { return true; } } @@ -2223,67 +2169,94 @@ 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 an arbitrary interface within 'port'. */ +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; - size_t i; + struct dst dst; 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]; + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { 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)) { - mirrors |= port->dst_mirrors; - dst++; + && set_dst(&dst, flow, in_port, port, tags)) { + 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; - mirrors |= out_port->dst_mirrors; - dst++; + } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) { + dst_set_add(set, &dst); + *nf_output_iface = dst.iface->dp_ifidx; + } +} + +static void +compose_mirror_dsts(const struct bridge *br, const struct flow *flow, + uint16_t vlan, const struct port *in_port, + struct dst_set *set, tag_type *tags) +{ + mirror_mask_t mirrors; + int flow_vlan; + size_t i; + + mirrors = in_port->src_mirrors; + for (i = 0; i < set->n; i++) { + mirrors |= set->dsts[i].iface->port->dst_mirrors; + } + + if (!mirrors) { + return; + } + + flow_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (flow_vlan == 0) { + flow_vlan = OFP_VLAN_NONE; } while (mirrors) { struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1]; if (!m->n_vlans || vlan_is_mirrored(m, vlan)) { + struct dst dst; + 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]; + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { 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; } @@ -2293,95 +2266,85 @@ 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); } } } } mirrors &= mirrors - 1; } - - partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan)); - return dst - dsts; -} - -static void OVS_UNUSED -print_dsts(const struct dst *dsts, size_t n) -{ - for (; n--; dsts++) { - printf(">p%"PRIu16, dsts->dp_ifidx); - if (dsts->vlan != OFP_VLAN_NONE) { - printf("v%"PRIu16, dsts->vlan); - } - } } 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; - uint16_t cur_vlan; - - n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, 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); - } 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); - } - cur_vlan = p->vlan; + uint16_t initial_vlan, cur_vlan; + const struct dst *dst; + struct dst_set set; + + dst_set_init(&set); + compose_dsts(br, flow, vlan, in_port, out_port, &set, tags, + nf_output_iface); + compose_mirror_dsts(br, flow, vlan, in_port, &set, tags); + + /* Output all the packets we can without having to change the VLAN. */ + initial_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (initial_vlan == 0) { + initial_vlan = OFP_VLAN_NONE; + } + for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { + if (dst->vlan != initial_vlan) { + continue; } - a = odp_actions_add(actions, ODPAT_OUTPUT); - a->output.port = p->dp_ifidx; + nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->iface->dp_ifidx); } -} -/* Returns the effective vlan of a packet, taking into account both the + /* Then output the rest. */ + cur_vlan = initial_vlan; + for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { + if (dst->vlan == initial_vlan) { + continue; + } + if (dst->vlan != cur_vlan) { + if (dst->vlan == OFP_VLAN_NONE) { + nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN); + } else { + ovs_be16 tci; + tci = htons(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 = dst->vlan; + } + nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->iface->dp_ifidx); + } + + dst_set_free(&set); +} + +/* Returns the effective vlan of a packet, taking into account both the * 802.1Q header and implicitly tagged ports. A value of 0 indicates that * the packet is untagged and -1 indicates it has an invalid header and * should be dropped. */ 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; } @@ -2419,27 +2382,34 @@ static void update_learning_table(struct bridge *br, const struct flow *flow, int vlan, struct port *in_port) { - enum grat_arp_lock_type lock_type; - tag_type rev_tag; - - /* We don't want to learn from gratuitous ARP packets that are reflected - * back over bond slaves so we lock the learning table. */ - lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE : - (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET : - GRAT_ARP_LOCK_CHECK; - - rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx, - lock_type); - if (rev_tag) { + struct mac_entry *mac; + + if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) { + return; + } + + mac = mac_learning_insert(br->ml, flow->dl_src, vlan); + if (is_gratuitous_arp(flow)) { + /* We don't want to learn from gratuitous ARP packets that are + * reflected back over bond slaves so we lock the learning table. */ + if (!in_port->bond) { + mac_entry_set_grat_arp_lock(mac); + } else if (mac_entry_is_grat_arp_locked(mac)) { + return; + } + } + + if (mac_entry_is_new(mac) || mac->port.p != in_port) { /* The log messages here could actually be useful in debugging, * so keep the rate limit relatively high. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, - 300); + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " "on port %s in VLAN %d", br->name, ETH_ADDR_ARGS(flow->dl_src), in_port->name, vlan); - ofproto_revalidate(br->ofproto, rev_tag); + + mac->port.p = in_port; + ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac)); } } @@ -2518,31 +2488,25 @@ 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) { - int src_idx; - bool is_grat_arp_locked; + if (in_port->bond) { + struct mac_entry *mac; - if (eth_addr_is_multicast(flow->dl_dst)) { - *tags |= in_port->active_iface_tag; - if (in_port->active_iface != in_iface->port_ifidx) { - /* Drop all multicast packets on inactive slaves. */ - return false; - } - } + switch (bond_check_admissibility(in_port->bond, in_iface, + flow->dl_dst, tags)) { + case BV_ACCEPT: + break; + + case BV_DROP: + return false; - /* Drop all packets for which we have learned a different input - * port, because we probably sent the packet on one slave and got - * it back on the other. Gratuitous ARP packets are an exception - * to this rule: the host has moved to another switch. The exception - * to the exception is if we locked the learning table to avoid - * reflections on bond slaves. If this is the case, just drop the - * packet now. */ - src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan, - &is_grat_arp_locked); - if (src_idx != -1 && src_idx != in_port->port_idx && - (!is_gratuitous_arp(flow) || is_grat_arp_locked)) { + case BV_DROP_IF_MOVED: + mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL); + if (mac && mac->port.p != in_port && + (!is_gratuitous_arp(flow) + || mac_entry_is_grat_arp_locked(mac))) { return false; + } + break; } } @@ -2554,13 +2518,13 @@ 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; struct port *out_port; + struct mac_entry *mac; int vlan; - int out_port_idx; /* Check whether we should drop packets in this flow. */ if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) { @@ -2568,788 +2532,153 @@ process_flow(struct bridge *br, const struct flow *flow, goto done; } - /* Learn source MAC (but don't try to learn from revalidation). */ - if (packet) { - update_learning_table(br, flow, vlan, in_port); - } - - /* Determine output port. */ - out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags, - NULL); - if (out_port_idx >= 0 && out_port_idx < br->n_ports) { - out_port = br->ports[out_port_idx]; - } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { - /* If we are revalidating but don't have a learning entry then - * eject the flow. Installing a flow that floods packets opens - * up a window of time where we could learn from a packet reflected - * on a bond and blackhole packets before the learning table is - * updated to reflect the correct port. */ - return false; - } else { - out_port = FLOOD_PORT; - } - - /* Don't send packets out their input ports. */ - if (in_port == out_port) { - out_port = NULL; - } - -done: - if (in_port) { - compose_actions(br, flow, vlan, in_port, out_port, tags, actions, - nf_output_iface); - } - - return true; -} - -static bool -bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, - struct odp_actions *actions, tag_type *tags, - uint16_t *nf_output_iface, void *br_) -{ - struct bridge *br = br_; - - COVERAGE_INC(bridge_process_flow); - - return process_flow(br, flow, packet, actions, tags, nf_output_iface); -} - -static void -bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, - const union odp_action *actions, - size_t n_actions, unsigned long long int n_bytes, - void *br_) -{ - struct bridge *br = br_; - const union odp_action *a; - struct port *in_port; - tag_type dummy = 0; - int vlan; - - /* Feed information from the active flows back into the learning table to - * ensure that table is always in sync with what is actually flowing - * through the datapath. - * - * We test that 'tags' is nonzero to ensure that only flows that include an - * OFPP_NORMAL action are used for learning. This works because - * bridge_normal_ofhook_cb() always sets a nonzero tag value. */ - if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) { - update_learning_table(br, flow, vlan, in_port); - } - - /* Account for bond slave utilization. */ - 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); - e->tx_bytes += n_bytes; - } - } - } -} - -static void -bridge_account_checkpoint_ofhook_cb(void *br_) -{ - struct bridge *br = br_; - long long int now; - size_t i; - - if (!br->has_bonded_ports) { - return; - } - - 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) { - port->bond_next_rebalance = now + port->bond_rebalance_interval; - bond_rebalance_port(port); - } - } -} - -static struct ofhooks bridge_ofhooks = { - bridge_normal_ofhook_cb, - bridge_account_flow_ofhook_cb, - bridge_account_checkpoint_ofhook_cb, -}; - -/* Bonding functions. */ - -/* Statistics for a single interface on a bonded port, used for load-based - * bond rebalancing. */ -struct slave_balance { - struct iface *iface; /* The interface. */ - uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */ - - /* All the "bond_entry"s that are assigned to this interface, in order of - * increasing tx_bytes. */ - struct bond_entry **hashes; - size_t n_hashes; -}; - -/* 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. */ -static int -compare_bond_entries(const void *a_, const void *b_) -{ - const struct bond_entry *const *ap = a_; - 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; - } else if (a->tx_bytes != b->tx_bytes) { - return a->tx_bytes > b->tx_bytes ? 1 : -1; - } else { - return 0; - } -} - -/* Sorts slave_balances so that enabled ports come first, and otherwise in - * *descending* order by number of bytes transmitted. */ -static int -compare_slave_balance(const void *a_, const void *b_) -{ - const struct slave_balance *a = a_; - const struct slave_balance *b = b_; - if (a->iface->enabled != b->iface->enabled) { - return a->iface->enabled ? -1 : 1; - } else if (a->tx_bytes != b->tx_bytes) { - return a->tx_bytes > b->tx_bytes ? -1 : 1; - } else { - return 0; - } -} - -static void -swap_bals(struct slave_balance *a, struct slave_balance *b) -{ - struct slave_balance tmp = *a; - *a = *b; - *b = tmp; -} - -/* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order - * given that 'p' (and only 'p') might be in the wrong location. - * - * This function invalidates 'p', since it might now be in a different memory - * location. */ -static void -resort_bals(struct slave_balance *p, - struct slave_balance bals[], size_t n_bals) -{ - if (n_bals > 1) { - for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) { - swap_bals(p, p - 1); - } - for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) { - swap_bals(p, p + 1); - } - } -} - -static void -log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port) -{ - if (VLOG_IS_DBG_ENABLED()) { - struct ds ds = DS_EMPTY_INITIALIZER; - const struct slave_balance *b; - - for (b = bals; b < bals + n_bals; b++) { - size_t i; - - if (b > bals) { - ds_put_char(&ds, ','); - } - ds_put_format(&ds, " %s %"PRIu64"kB", - b->iface->name, b->tx_bytes / 1024); - - if (!b->iface->enabled) { - ds_put_cstr(&ds, " (disabled)"); - } - if (b->n_hashes > 0) { - ds_put_cstr(&ds, " ("); - for (i = 0; i < b->n_hashes; i++) { - const struct bond_entry *e = b->hashes[i]; - if (i > 0) { - ds_put_cstr(&ds, " + "); - } - ds_put_format(&ds, "h%td: %"PRIu64"kB", - e - port->bond_hash, e->tx_bytes / 1024); - } - ds_put_cstr(&ds, ")"); - } - } - VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds)); - ds_destroy(&ds); - } -} - -/* Shifts 'hash' from 'from' to 'to' within 'port'. */ -static void -bond_shift_load(struct slave_balance *from, struct slave_balance *to, - int hash_idx) -{ - struct bond_entry *hash = from->hashes[hash_idx]; - struct port *port = from->iface->port; - uint64_t delta = hash->tx_bytes; - - 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)", - port->name, delta / 1024, hash - port->bond_hash, - from->iface->name, to->iface->name, - (from->tx_bytes - delta) / 1024, - (to->tx_bytes + delta) / 1024); - - /* Delete element from from->hashes. - * - * We don't bother to add the element to to->hashes because not only would - * it require more work, the only purpose it would be to allow that hash to - * be migrated to another slave in this rebalancing run, and there is no - * point in doing that. */ - if (hash_idx == 0) { - from->hashes++; - } else { - memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1, - (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes); - } - from->n_hashes--; - - /* Shift load away from 'from' to 'to'. */ - from->tx_bytes -= delta; - 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(); -} - -static void -bond_rebalance_port(struct port *port) -{ - struct slave_balance bals[DP_MAX_PORTS]; - size_t n_bals; - struct bond_entry *hashes[BOND_MASK + 1]; - struct slave_balance *b, *from, *to; - struct bond_entry *e; - size_t i; - - /* 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 - * loaded slave. - * - * The code is a bit tricky: to avoid dynamically allocating a 'hashes' - * array for each slave_balance structure, we sort our local array of - * hashes in order by slave, so that all of the hashes for a given slave - * 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->tx_bytes = 0; - b->hashes = NULL; - b->n_hashes = 0; - } - 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]; - } - b->n_hashes++; - } - } - qsort(bals, n_bals, sizeof *bals, compare_slave_balance); - log_bals(bals, n_bals, port); - - /* Discard slaves that aren't enabled (which were sorted to the back of the - * array earlier). */ - while (!bals[n_bals - 1].iface->enabled) { - n_bals--; - if (!n_bals) { - return; - } - } - - /* Shift load from the most-loaded slaves to the least-loaded slaves. */ - to = &bals[n_bals - 1]; - for (from = bals; from < to; ) { - uint64_t overload = from->tx_bytes - to->tx_bytes; - if (overload < to->tx_bytes >> 5 || overload < 100000) { - /* The extra load on 'from' (and all less-loaded slaves), compared - * to that of 'to' (the least-loaded slave), is less than ~3%, or - * it is less than ~1Mbps. No point in rebalancing. */ - break; - } else if (from->n_hashes == 1) { - /* 'from' only carries a single MAC hash, so we can't shift any - * load away from it, even though we want to. */ - from++; - } else { - /* 'from' is carrying significantly more load than 'to', and that - * load is split across at least two different hashes. Pick a hash - * to migrate to 'to' (the least-loaded slave), given that doing so - * must decrease the ratio of the load on the two slaves by at - * least 0.1. - * - * The sort order we use means that we prefer to shift away the - * smallest hashes instead of the biggest ones. There is little - * reason behind this decision; we could use the opposite sort - * order to shift away big hashes ahead of small ones. */ - bool order_swapped; - - for (i = 0; i < from->n_hashes; i++) { - double old_ratio, new_ratio; - uint64_t delta = from->hashes[i]->tx_bytes; - - if (delta == 0 || from->tx_bytes - delta == 0) { - /* Pointless move. */ - continue; - } - - order_swapped = from->tx_bytes - delta < to->tx_bytes + delta; - - if (to->tx_bytes == 0) { - /* Nothing on the new slave, move it. */ - break; - } - - old_ratio = (double)from->tx_bytes / to->tx_bytes; - new_ratio = (double)(from->tx_bytes - delta) / - (to->tx_bytes + delta); - - if (new_ratio == 0) { - /* Should already be covered but check to prevent division - * by zero. */ - continue; - } - - if (new_ratio < 1) { - new_ratio = 1 / new_ratio; - } - - if (old_ratio - new_ratio > 0.1) { - /* Would decrease the ratio, move it. */ - break; - } - } - 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. */ - if (order_swapped) { - swap_bals(from, to); - } - - /* Re-sort 'bals'. Note that this may make 'from' and 'to' - * point to different slave_balance structures. It is only - * valid to do these two operations in a row at all because we - * know that 'from' will not move past 'to' and vice versa. */ - resort_bals(from, bals, n_bals); - resort_bals(to, bals, n_bals); - } else { - from++; - } - } - } - - /* Implement exponentially weighted moving average. A weight of 1/2 causes - * 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; - } -} - -static void -bond_send_learning_packets(struct port *port) -{ - struct bridge *br = port->bridge; - struct mac_entry *e; - struct ofpbuf packet; - int error, n_packets, n_errors; - - if (!port->n_ifaces || port->active_iface < 0) { - 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; - struct flow flow; - int retval; - - if (e->port == port->port_idx - || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) { - 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++; - } - 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); - if (retval) { - error = retval; - n_errors++; - } - } - ofpbuf_uninit(&packet); - - if (n_errors) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " - "packets, last error was: %s", - port->name, n_errors, n_packets, strerror(error)); - } else { - VLOG_DBG("bond %s: sent %d gratuitous learning packets", - port->name, n_packets); - } -} - -/* Bonding unixctl user interface functions. */ - -static void -bond_unixctl_list(struct unixctl_conn *conn, - const char *args OVS_UNUSED, void *aux OVS_UNUSED) -{ - struct ds ds = DS_EMPTY_INITIALIZER; - const struct bridge *br; - - ds_put_cstr(&ds, "bridge\tbond\tslaves\n"); - - LIST_FOR_EACH (br, node, &all_bridges) { - size_t i; - - for (i = 0; i < br->n_ports; i++) { - const struct port *port = br->ports[i]; - if (port->n_ifaces > 1) { - size_t j; - - 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_cstr(&ds, ", "); - } - ds_put_cstr(&ds, iface->name); - } - ds_put_char(&ds, '\n'); - } - } - } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); - ds_destroy(&ds); -} - -static struct port * -bond_find(const char *name) -{ - const struct bridge *br; - - LIST_FOR_EACH (br, node, &all_bridges) { - size_t i; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (!strcmp(port->name, name) && port->n_ifaces > 1) { - return port; - } - } - } - return NULL; -} - -static void -bond_unixctl_show(struct unixctl_conn *conn, - const char *args, void *aux OVS_UNUSED) -{ - struct ds ds = DS_EMPTY_INITIALIZER; - const struct port *port; - size_t j; - - port = bond_find(args); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } - - 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]; - struct bond_entry *be; - - /* Basic info. */ - ds_put_format(&ds, "slave %s: %s\n", - iface->name, iface->enabled ? "enabled" : "disabled"); - if (j == port->active_iface) { - ds_put_cstr(&ds, "\tactive slave\n"); - } - if (iface->delay_expires != LLONG_MAX) { - ds_put_format(&ds, "\t%s expires in %lld ms\n", - iface->enabled ? "downdelay" : "updelay", - iface->delay_expires - time_msec()); - } - - /* Hashes. */ - 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) { - continue; - } - - ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", - hash, be->tx_bytes / 1024); - - /* 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 - && me->port != port->port_idx - && choose_output_iface(port, me->mac, &dp_ifidx, &tags) - && dp_ifidx == iface->dp_ifidx) - { - ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n", - ETH_ADDR_ARGS(me->mac)); - } - } - } - } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); - ds_destroy(&ds); -} - -static void -bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, - void *aux OVS_UNUSED) -{ - 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; - int hash; - - bond_s = strtok_r(args, " ", &save_ptr); - hash_s = strtok_r(NULL, " ", &save_ptr); - slave_s = strtok_r(NULL, " ", &save_ptr); - if (!slave_s) { - unixctl_command_reply(conn, 501, - "usage: bond/migrate BOND HASH SLAVE"); - return; - } - - port = bond_find(bond_s); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; + /* Learn source MAC (but don't try to learn from revalidation). */ + if (packet) { + update_learning_table(br, flow, vlan, in_port); } - 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)) { - hash = atoi(hash_s) & BOND_MASK; + /* Determine output port. */ + mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags); + if (mac) { + out_port = mac->port.p; + } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { + /* If we are revalidating but don't have a learning entry then + * eject the flow. Installing a flow that floods packets opens + * up a window of time where we could learn from a packet reflected + * on a bond and blackhole packets before the learning table is + * updated to reflect the correct port. */ + return false; } else { - unixctl_command_reply(conn, 501, "bad hash"); - return; + out_port = FLOOD_PORT; } - iface = port_lookup_iface(port, slave_s); - if (!iface) { - unixctl_command_reply(conn, 501, "no such slave"); - return; + /* Don't send packets out their input ports. */ + if (in_port == out_port) { + out_port = NULL; } - if (!iface->enabled) { - unixctl_command_reply(conn, 501, "cannot migrate to disabled slave"); - return; +done: + if (in_port) { + compose_actions(br, flow, vlan, in_port, out_port, tags, actions, + nf_output_iface); } - 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; - unixctl_command_reply(conn, 200, "migrated"); + return true; } -static void -bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, - void *aux OVS_UNUSED) +static bool +bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, + struct ofpbuf *actions, tag_type *tags, + uint16_t *nf_output_iface, void *br_) { - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *slave_s; - struct port *port; - struct iface *iface; + struct bridge *br = br_; - bond_s = strtok_r(args, " ", &save_ptr); - slave_s = strtok_r(NULL, " ", &save_ptr); - if (!slave_s) { - unixctl_command_reply(conn, 501, - "usage: bond/set-active-slave BOND SLAVE"); - return; - } + COVERAGE_INC(bridge_process_flow); + return process_flow(br, flow, packet, actions, tags, nf_output_iface); +} - port = bond_find(bond_s); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } +static bool +bridge_special_ofhook_cb(const struct flow *flow, + const struct ofpbuf *packet, void *br_) +{ + struct iface *iface; + struct bridge *br = br_; - iface = port_lookup_iface(port, slave_s); - if (!iface) { - unixctl_command_reply(conn, 501, "no such slave"); - return; - } + iface = iface_from_dp_ifidx(br, flow->in_port); - if (!iface->enabled) { - unixctl_command_reply(conn, 501, "cannot make disabled slave active"); - return; + if (flow->dl_type == htons(ETH_TYPE_LACP)) { + if (iface && iface->port->bond && packet) { + bond_process_lacp(iface->port->bond, iface, packet); + } + return false; } - 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(); - VLOG_INFO("port %s: active interface is now %s", - port->name, iface->name); - bond_send_learning_packets(port); - unixctl_command_reply(conn, 200, "done"); - } else { - unixctl_command_reply(conn, 200, "no change"); - } + return true; } static void -enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) +bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, + const struct nlattr *actions, + size_t actions_len, + uint64_t n_bytes, void *br_) { - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *slave_s; - struct port *port; - struct iface *iface; + struct bridge *br = br_; + const struct nlattr *a; + struct port *in_port; + tag_type dummy = 0; + unsigned int left; + int vlan; - bond_s = strtok_r(args, " ", &save_ptr); - slave_s = strtok_r(NULL, " ", &save_ptr); - if (!slave_s) { - unixctl_command_reply(conn, 501, - "usage: bond/enable/disable-slave BOND SLAVE"); - return; + /* Feed information from the active flows back into the learning table to + * ensure that table is always in sync with what is actually flowing + * through the datapath. + * + * We test that 'tags' is nonzero to ensure that only flows that include an + * OFPP_NORMAL action are used for learning. This works because + * bridge_normal_ofhook_cb() always sets a nonzero tag value. */ + if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) { + update_learning_table(br, flow, vlan, in_port); } - port = bond_find(bond_s); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); + /* Account for bond slave utilization. */ + if (!br->has_bonded_ports) { return; } - - iface = port_lookup_iface(port, slave_s); - if (!iface) { - unixctl_command_reply(conn, 501, "no such slave"); - return; + 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->bond) { + uint16_t vlan = (flow->vlan_tci + ? vlan_tci_to_vid(flow->vlan_tci) + : OFP_VLAN_NONE); + bond_account(out_port->bond, flow, vlan, n_bytes); + } + } } - - bond_enable_slave(iface, enable); - unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled"); } static void -bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, - void *aux OVS_UNUSED) +bridge_account_checkpoint_ofhook_cb(void *br_) { - enable_slave(conn, args, true); -} + struct bridge *br = br_; + struct port *port; -static void -bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, - void *aux OVS_UNUSED) -{ - enable_slave(conn, args, false); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + if (port->bond) { + bond_rebalance(port->bond, + ofproto_get_revalidate_set(br->ofproto)); + } + } } +static struct ofhooks bridge_ofhooks = { + bridge_normal_ofhook_cb, + bridge_special_ofhook_cb, + bridge_account_flow_ofhook_cb, + bridge_account_checkpoint_ofhook_cb, +}; + +/* Port functions. */ + static void -bond_unixctl_hash(struct unixctl_conn *conn, const char *args, - void *aux OVS_UNUSED) +port_run(struct port *port) { - uint8_t mac[ETH_ADDR_LEN]; - uint8_t hash; - char *hash_cstr; - - if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) - == ETH_ADDR_SCAN_COUNT) { - hash = bond_hash(mac); - - hash_cstr = xasprintf("%u", hash); - unixctl_command_reply(conn, 200, hash_cstr); - free(hash_cstr); - } else { - unixctl_command_reply(conn, 501, "invalid mac"); - } + if (port->bond) { + bond_run(port->bond, + ofproto_get_revalidate_set(port->bridge->ofproto)); + if (bond_should_send_learning_packets(port->bond)) { + port_send_learning_packets(port); + } + } } static void -bond_init(void) +port_wait(struct port *port) { - unixctl_command_register("bond/list", bond_unixctl_list, NULL); - unixctl_command_register("bond/show", bond_unixctl_show, NULL); - unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL); - unixctl_command_register("bond/set-active-slave", - bond_unixctl_set_active_slave, NULL); - unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave, - NULL); - unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave, - NULL); - unixctl_command_register("bond/hash", bond_unixctl_hash, NULL); + if (port->bond) { + bond_wait(port->bond); + } } - -/* Port functions. */ static struct port * port_create(struct bridge *br, const char *name) @@ -3358,18 +2687,12 @@ port_create(struct bridge *br, const char *name) port = xzalloc(sizeof *port); port->bridge = br; - port->port_idx = br->n_ports; port->vlan = -1; port->trunks = NULL; port->name = xstrdup(name); - port->active_iface = -1; + list_init(&port->ifaces); - if (br->n_ports >= br->allocated_ports) { - br->ports = x2nrealloc(br->ports, &br->allocated_ports, - sizeof *br->ports); - } - br->ports[br->n_ports++] = port; - shash_add_assert(&br->port_by_name, port->name, port); + hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0)); VLOG_INFO("created port %s on bridge %s", port->name, br->name); bridge_flush(br); @@ -3388,70 +2711,80 @@ 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 shash new_ifaces; + struct iface *iface, *next; + struct sset new_ifaces; size_t i; /* Collect list of new interfaces. */ - shash_init(&new_ifaces); + sset_init(&new_ifaces); for (i = 0; i < cfg->n_interfaces; i++) { const char *name = cfg->interfaces[i]->name; - shash_add_once(&new_ifaces, name, NULL); + sset_add(&new_ifaces, name); } /* 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 (!sset_contains(&new_ifaces, iface->name)) { + iface_destroy(iface); } } - shash_destroy(&new_ifaces); + sset_destroy(&new_ifaces); +} + +/* Expires all MAC learning entries associated with 'port' and forces ofproto + * to revalidate every flow. */ +static void +port_flush_macs(struct port *port) +{ + struct bridge *br = port->bridge; + struct mac_learning *ml = br->ml; + struct mac_entry *mac, *next_mac; + + bridge_flush(br); + LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { + if (mac->port.p == port) { + mac_learning_expire(ml, mac); + } + } } static void port_reconfigure(struct port *port, const struct ovsrec_port *cfg) { - struct shash new_ifaces; - long long int next_rebalance; + struct sset new_ifaces; + bool need_flush = false; unsigned long *trunks; int vlan; size_t i; port->cfg = cfg; - /* Update settings. */ - port->updelay = cfg->bond_updelay; - if (port->updelay < 0) { - port->updelay = 0; - } - port->downdelay = cfg->bond_downdelay; - if (port->downdelay < 0) { - port->downdelay = 0; - } - port->bond_rebalance_interval = atoi( - get_port_other_config(cfg, "bond-rebalance-interval", "10000")); - if (port->bond_rebalance_interval < 1000) { - port->bond_rebalance_interval = 1000; - } - next_rebalance = time_msec() + port->bond_rebalance_interval; - if (port->bond_next_rebalance > next_rebalance) { - port->bond_next_rebalance = next_rebalance; - } /* Add new interfaces and update 'cfg' member of existing ones. */ - shash_init(&new_ifaces); + sset_init(&new_ifaces); for (i = 0; i < cfg->n_interfaces; i++) { const struct ovsrec_interface *if_cfg = cfg->interfaces[i]; struct iface *iface; - if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) { + if (!sset_add(&new_ifaces, if_cfg->name)) { VLOG_WARN("port %s: %s specified twice as port interface", port->name, if_cfg->name); + iface_set_ofport(if_cfg, -1); continue; } @@ -3475,12 +2808,12 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) : if_cfg->type[0] ? if_cfg->type : "system"); } - shash_destroy(&new_ifaces); + sset_destroy(&new_ifaces); /* Get VLAN tag. */ vlan = -1; if (cfg->tag) { - if (port->n_ifaces < 2) { + if (list_is_short(&port->ifaces)) { vlan = *cfg->tag; if (vlan >= 0 && vlan <= 4095) { VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan); @@ -3496,7 +2829,7 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) } if (port->vlan != vlan) { port->vlan = vlan; - bridge_flush(port->bridge); + need_flush = true; } /* Get trunked VLANs. */ @@ -3531,10 +2864,14 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) if (trunks == NULL ? port->trunks != NULL : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) { - bridge_flush(port->bridge); + need_flush = true; } bitmap_free(port->trunks); port->trunks = trunks; + + if (need_flush) { + port_flush_macs(port); + } } static void @@ -3542,12 +2879,9 @@ port_destroy(struct port *port) { if (port) { struct bridge *br = port->bridge; - struct port *del; + struct iface *iface, *next; 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) { @@ -3555,23 +2889,19 @@ 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); - - del = br->ports[port->port_idx] = br->ports[--br->n_ports]; - del->port_idx = port->port_idx; + hmap_remove(&br->ports, &port->hmap_node); VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); - netdev_monitor_destroy(port->monitor); - free(port->ifaces); + port_flush_macs(port); + bitmap_free(port->trunks); free(port->name); free(port); - bridge_flush(br); } } @@ -3585,168 +2915,165 @@ port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) static struct port * port_lookup(const struct bridge *br, const char *name) { - return shash_find_data(&br->port_by_name, name); + struct port *port; + + HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0), + &br->ports) { + if (!strcmp(port->name, name)) { + return port; + } + } + return NULL; } -static struct iface * -port_lookup_iface(const struct port *port, const char *name) +static bool +enable_lacp(struct port *port, bool *activep) { - struct iface *iface = iface_lookup(port->bridge, name); - return iface && iface->port == port ? iface : NULL; + if (!port->cfg->lacp) { + /* XXX when LACP implementation has been sufficiently tested, enable by + * default and make active on bonded ports. */ + return false; + } else if (!strcmp(port->cfg->lacp, "off")) { + return false; + } else if (!strcmp(port->cfg->lacp, "active")) { + *activep = true; + return true; + } else if (!strcmp(port->cfg->lacp, "passive")) { + *activep = false; + return true; + } else { + VLOG_WARN("port %s: unknown LACP mode %s", + port->name, port->cfg->lacp); + return false; + } } -static void -port_update_bonding(struct port *port) +static struct lacp_settings * +port_reconfigure_bond_lacp(struct port *port, struct lacp_settings *s) { - if (port->monitor) { - netdev_monitor_destroy(port->monitor); - port->monitor = NULL; + if (!enable_lacp(port, &s->active)) { + return NULL; } - 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; - } - } else { - size_t i; - - if (!port->bond_hash) { - port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); - for (i = 0; i <= BOND_MASK; i++) { - struct bond_entry *e = &port->bond_hash[i]; - e->iface_idx = -1; - e->tx_bytes = 0; - } - port->no_ifaces_tag = tag_create_random(); - bond_choose_active_iface(port); - port->bond_next_rebalance - = time_msec() + port->bond_rebalance_interval; - if (port->cfg->bond_fake_iface) { - port->bond_next_fake_iface_update = time_msec(); - } - } - port->bond_compat_is_stale = true; - port->bond_fake_iface = port->cfg->bond_fake_iface; + s->name = port->name; + memcpy(s->id, port->bridge->ea, ETH_ADDR_LEN); + s->priority = atoi(get_port_other_config(port->cfg, "lacp-system-priority", + "0")); + s->fast = !strcmp(get_port_other_config(port->cfg, "lacp-time", "slow"), + "fast"); - port->monitor = netdev_monitor_create(); - for (i = 0; i < port->n_ifaces; i++) { - netdev_monitor_add(port->monitor, port->ifaces[i]->netdev); - } + if (s->priority <= 0 || s->priority > UINT16_MAX) { + /* Prefer bondable links if unspecified. */ + s->priority = UINT16_MAX - !list_is_short(&port->ifaces); } + return s; } static void -port_update_bond_compat(struct port *port) +iface_reconfigure_bond(struct iface *iface) { - struct compat_bond_hash compat_hashes[BOND_MASK + 1]; - struct compat_bond bond; - size_t i; + struct lacp_slave_settings s; + int priority; + + s.name = iface->name; + s.id = iface->dp_ifidx; + priority = atoi(get_interface_other_config( + iface->cfg, "lacp-port-priority", "0")); + s.priority = (priority >= 0 && priority <= UINT16_MAX + ? priority : UINT16_MAX); + bond_slave_register(iface->port->bond, iface, iface->netdev, &s); +} + +static void +port_reconfigure_bond(struct port *port) +{ + struct lacp_settings lacp_settings; + struct bond_settings s; + const char *detect_s; + struct iface *iface; - if (port->n_ifaces < 2) { - proc_net_compat_update_bond(port->name, NULL); + if (list_is_short(&port->ifaces)) { + /* Not a bonded port. */ + bond_destroy(port->bond); + port->bond = NULL; return; } - bond.up = false; - bond.updelay = port->updelay; - bond.downdelay = port->downdelay; + port->bridge->has_bonded_ports = true; - 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; - } - } + s.name = port->name; + s.balance = BM_SLB; + if (port->cfg->bond_mode + && !bond_mode_from_string(&s.balance, port->cfg->bond_mode)) { + VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s", + port->name, port->cfg->bond_mode, + bond_mode_to_string(s.balance)); } - 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; + s.detect = BLSM_CARRIER; + detect_s = get_port_other_config(port->cfg, "bond-detect-mode", NULL); + if (detect_s && !bond_detect_mode_from_string(&s.detect, detect_s)) { + VLOG_WARN("port %s: unsupported bond-detect-mode %s, " + "defaulting to %s", + port->name, detect_s, bond_detect_mode_to_string(s.detect)); + } - /* We need to make the same determination as the Linux bonding - * code to determine whether a slave should be consider "up". - * The Linux function bond_miimon_inspect() supports four - * BOND_LINK_* states: - * - * - BOND_LINK_UP: carrier detected, updelay has passed. - * - BOND_LINK_FAIL: carrier lost, downdelay in progress. - * - BOND_LINK_DOWN: carrier lost, downdelay has passed. - * - BOND_LINK_BACK: carrier detected, updelay in progress. - * - * The function bond_info_show_slave() only considers BOND_LINK_UP - * to be "up" and anything else to be "down". - */ - slave->up = iface->enabled && iface->delay_expires == LLONG_MAX; - if (slave->up) { - bond.up = true; - } - netdev_get_etheraddr(iface->netdev, slave->mac); + s.miimon_interval = atoi( + get_port_other_config(port->cfg, "bond-miimon-interval", "200")); + if (s.miimon_interval < 100) { + s.miimon_interval = 100; + } + + s.up_delay = MAX(0, port->cfg->bond_updelay); + s.down_delay = MAX(0, port->cfg->bond_downdelay); + s.rebalance_interval = atoi( + get_port_other_config(port->cfg, "bond-rebalance-interval", "10000")); + if (s.rebalance_interval < 1000) { + s.rebalance_interval = 1000; } - if (port->bond_fake_iface) { - struct netdev *bond_netdev; + s.fake_iface = port->cfg->bond_fake_iface; + s.lacp = port_reconfigure_bond_lacp(port, &lacp_settings); - if (!netdev_open_default(port->name, &bond_netdev)) { - if (bond.up) { - netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); - } else { - netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); - } - netdev_close(bond_netdev); - } + if (!port->bond) { + port->bond = bond_create(&s); + } else { + bond_reconfigure(port->bond, &s); } - proc_net_compat_update_bond(port->name, &bond); - free(bond.slaves); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_reconfigure_bond(iface); + } } static void -port_update_vlan_compat(struct port *port) +port_send_learning_packets(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; - } + int error, n_packets, n_errors; + struct mac_entry *e; + + error = n_packets = n_errors = 0; + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { + if (e->port.p != port) { + int ret = bond_send_learning_packet(port->bond, e->mac, e->vlan); + if (ret) { + error = ret; + n_errors++; } + n_packets++; } } - proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan); + + if (n_errors) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " + "packets, last error was: %s", + port->name, n_errors, n_packets, strerror(error)); + } else { + VLOG_DBG("bond %s: sent %d gratuitous learning packets", + port->name, n_packets); + } } /* Interface functions. */ @@ -3760,24 +3087,15 @@ 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(); - iface->delay_expires = LLONG_MAX; iface->netdev = NULL; iface->cfg = 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; - if (port->n_ifaces > 1) { - br->has_bonded_ports = true; - } + list_push_back(&port->ifaces, &iface->port_elem); VLOG_DBG("attached network device %s to port %s", iface->name, port->name); @@ -3792,8 +3110,10 @@ 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; + + if (port->bond) { + bond_slave_unregister(port->bond, iface); + } shash_find_and_delete_assert(&br->iface_by_name, iface->name); @@ -3801,17 +3121,10 @@ 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); 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); - } - free(iface->name); free(iface); @@ -3825,6 +3138,21 @@ 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) { @@ -3863,6 +3191,15 @@ iface_set_mac(struct iface *iface) } } +/* Sets the ofport column of 'if_cfg' to 'ofport'. */ +static void +iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport) +{ + if (if_cfg) { + ovsrec_interface_set_ofport(if_cfg, &ofport, 1); + } +} + /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'. * * The value strings in '*shash' are taken directly from values[], not copied, @@ -3879,6 +3216,38 @@ 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; @@ -3939,6 +3308,53 @@ 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 maid[CCM_MAID_LEN]; + + mon = iface->cfg->monitor; + + if (!mon) { + ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); + 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.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) +{ + /* XXX */ + return netdev_get_carrier(iface->netdev); +} /* Port mirroring. */ @@ -3960,6 +3376,7 @@ static void mirror_reconfigure(struct bridge *br) { unsigned long *rspan_vlans; + struct port *port; int i; /* Get rid of deleted mirrors. */ @@ -3989,8 +3406,8 @@ mirror_reconfigure(struct bridge *br) } /* Update port reserved status. */ - for (i = 0; i < br->n_ports; i++) { - br->ports[i]->is_mirror_output_port = false; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port->is_mirror_output_port = false; } for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; @@ -4018,6 +3435,7 @@ mirror_reconfigure(struct bridge *br) } if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) { bridge_flush(br); + mac_learning_flush(br->ml); } } @@ -4040,13 +3458,14 @@ mirror_create(struct bridge *br, struct ovsrec_mirror *cfg) VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name); bridge_flush(br); + mac_learning_flush(br->ml); br->mirrors[i] = m = xzalloc(sizeof *m); m->bridge = br; m->idx = i; m->name = xstrdup(cfg->name); - shash_init(&m->src_ports); - shash_init(&m->dst_ports); + sset_init(&m->src_ports); + sset_init(&m->dst_ports); m->vlans = NULL; m->n_vlans = 0; m->out_vlan = -1; @@ -4060,15 +3479,15 @@ mirror_destroy(struct mirror *m) { if (m) { struct bridge *br = m->bridge; - size_t i; + struct port *port; - for (i = 0; i < br->n_ports; i++) { - br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); - br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); + port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); } - shash_destroy(&m->src_ports); - shash_destroy(&m->dst_ports); + sset_destroy(&m->src_ports); + sset_destroy(&m->dst_ports); free(m->vlans); m->bridge->mirrors[m->idx] = NULL; @@ -4076,19 +3495,20 @@ mirror_destroy(struct mirror *m) free(m); bridge_flush(br); + mac_learning_flush(br->ml); } } static void mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports, - struct shash *names) + struct sset *names) { size_t i; for (i = 0; i < n_ports; i++) { const char *name = ports[i]->name; if (port_lookup(m->bridge, name)) { - shash_add_once(names, name, NULL); + sset_add(names, name); } else { VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent " "port %s", m->bridge->name, m->name, name); @@ -4146,13 +3566,13 @@ port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) static void mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) { - struct shash src_ports, dst_ports; + struct sset src_ports, dst_ports; mirror_mask_t mirror_bit; struct port *out_port; + struct port *port; int out_vlan; size_t n_vlans; int *vlans; - size_t i; /* Set name. */ if (strcmp(cfg->name, m->name)) { @@ -4186,13 +3606,12 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) return; } - shash_init(&src_ports); - shash_init(&dst_ports); + sset_init(&src_ports); + sset_init(&dst_ports); if (cfg->select_all) { - for (i = 0; i < m->bridge->n_ports; i++) { - const char *name = m->bridge->ports[i]->name; - shash_add_once(&src_ports, name, NULL); - shash_add_once(&dst_ports, name, NULL); + HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) { + sset_add(&src_ports, port->name); + sset_add(&dst_ports, port->name); } vlans = NULL; n_vlans = 0; @@ -4208,16 +3627,17 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) } /* Update mirror data. */ - if (!shash_equal_keys(&m->src_ports, &src_ports) - || !shash_equal_keys(&m->dst_ports, &dst_ports) + if (!sset_equals(&m->src_ports, &src_ports) + || !sset_equals(&m->dst_ports, &dst_ports) || m->n_vlans != n_vlans || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans) || m->out_port != out_port || m->out_vlan != out_vlan) { bridge_flush(m->bridge); + mac_learning_flush(m->bridge->ml); } - shash_swap(&m->src_ports, &src_ports); - shash_swap(&m->dst_ports, &dst_ports); + sset_swap(&m->src_ports, &src_ports); + sset_swap(&m->dst_ports, &dst_ports); free(m->vlans); m->vlans = vlans; m->n_vlans = n_vlans; @@ -4226,10 +3646,8 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) /* Update ports. */ mirror_bit = MIRROR_MASK_C(1) << m->idx; - for (i = 0; i < m->bridge->n_ports; i++) { - struct port *port = m->bridge->ports[i]; - - if (shash_find(&m->src_ports, port->name) + HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) { + if (sset_contains(&m->src_ports, port->name) || (m->n_vlans && (!port->vlan ? port_trunks_any_mirrored_vlan(m, port) @@ -4239,7 +3657,7 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) port->src_mirrors &= ~mirror_bit; } - if (shash_find(&m->dst_ports, port->name)) { + if (sset_contains(&m->dst_ports, port->name)) { port->dst_mirrors |= mirror_bit; } else { port->dst_mirrors &= ~mirror_bit; @@ -4247,6 +3665,6 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) } /* Clean up. */ - shash_destroy(&src_ports); - shash_destroy(&dst_ports); + sset_destroy(&src_ports); + sset_destroy(&dst_ports); }