X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=598b0016d35d9c7e0820fffb3dcd2bc042b9c8ba;hb=d295e8e97acae13552a5b220d3fbcff8201064a2;hp=31e6cb073452f1efca9408b85836d4aa308f6123;hpb=cd11000ba2ce68a6eaeabed7efcb061b39ceaa44;p=openvswitch diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 31e6cb07..598b0016 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -20,6 +20,7 @@ #include #include #include +#include #include #include #include @@ -45,6 +46,7 @@ #include "ofpbuf.h" #include "ofproto/netflow.h" #include "ofproto/ofproto.h" +#include "ovsdb-data.h" #include "packets.h" #include "poll-loop.h" #include "port-array.h" @@ -61,11 +63,11 @@ #include "vconn.h" #include "vswitchd/vswitch-idl.h" #include "xenserver.h" +#include "vlog.h" #include "xtoxll.h" #include "sflow_api.h" -#define THIS_MODULE VLM_bridge -#include "vlog.h" +VLOG_DEFINE_THIS_MODULE(bridge) struct dst { uint16_t vlan; @@ -81,12 +83,10 @@ struct iface { long long delay_expires; /* Time after which 'enabled' may change. */ /* These members are valid only after bridge_reconfigure() causes them to - * be initialized.*/ + * be initialized. */ int dp_ifidx; /* Index within kernel datapath. */ struct netdev *netdev; /* Network device. */ bool enabled; /* May be chosen for flows? */ - - /* This member is only valid *during* bridge_reconfigure(). */ const struct ovsrec_interface *cfg; }; @@ -105,6 +105,7 @@ struct mirror { struct bridge *bridge; size_t idx; char *name; + struct uuid uuid; /* UUID of this "mirror" record in database. */ /* Selection criteria. */ struct shash src_ports; /* Name is port name; data is always NULL. */ @@ -122,7 +123,9 @@ struct port { struct bridge *bridge; size_t port_idx; int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ - unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */ + 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. @@ -138,7 +141,7 @@ struct port { 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 bond_next_fake_iface_update; /* Next update to fake bond stats. */ + 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. */ @@ -146,9 +149,6 @@ struct port { mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ bool is_mirror_output_port; /* Does port mirroring send frames here? */ - - /* This member is only valid *during* bridge_reconfigure(). */ - const struct ovsrec_port *cfg; }; #define DP_MAX_PORTS 255 @@ -156,19 +156,12 @@ struct bridge { struct list node; /* Node in global list of bridges. */ char *name; /* User-specified arbitrary name. */ struct mac_learning *ml; /* MAC learning table. */ - bool sent_config_request; /* Successfully sent config request? */ uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */ + const struct ovsrec_bridge *cfg; /* OpenFlow switch processing. */ struct ofproto *ofproto; /* OpenFlow switch. */ - /* Description strings. */ - char *mfr_desc; /* Manufacturer. */ - char *hw_desc; /* Hardware. */ - char *sw_desc; /* Software version. */ - char *serial_desc; /* Serial number. */ - char *dp_desc; /* Datapath description. */ - /* Kernel datapath information. */ struct dpif *dpif; /* Datapath. */ struct port_array ifaces; /* Indexed by kernel datapath port number. */ @@ -176,6 +169,8 @@ struct bridge { /* Bridge ports. */ struct port **ports; size_t n_ports, allocated_ports; + 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; @@ -183,36 +178,33 @@ struct bridge { /* Flow tracking. */ bool flush; - /* Flow statistics gathering. */ - time_t next_stats_request; - /* Port mirroring. */ struct mirror *mirrors[MAX_MIRRORS]; - - /* This member is only valid *during* bridge_reconfigure(). */ - const struct ovsrec_bridge *cfg; }; /* List of all bridges. */ static struct list all_bridges = LIST_INITIALIZER(&all_bridges); -/* Maximum number of datapaths. */ -enum { DP_MAX = 256 }; +/* OVSDB IDL used to obtain configuration. */ +static struct ovsdb_idl *idl; + +/* Each time this timer expires, the bridge fetches statistics for every + * interface and pushes them into the database. */ +#define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */ +static long long int iface_stats_timer = 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); static unixctl_cb_func bridge_unixctl_dump_flows; +static unixctl_cb_func bridge_unixctl_reconnect; static int bridge_run_one(struct bridge *); -static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg, - const struct bridge *br, +static size_t bridge_get_controllers(const struct bridge *br, struct ovsrec_controller ***controllersp); -static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *, - struct bridge *); -static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *, - struct bridge *, - const struct sockaddr_in *managers, - size_t n_managers); +static void bridge_reconfigure_one(struct bridge *); +static void bridge_reconfigure_remotes(struct bridge *, + const struct sockaddr_in *managers, + size_t n_managers); static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces); static void bridge_fetch_dp_ifaces(struct bridge *); static void bridge_flush(struct bridge *); @@ -236,6 +228,7 @@ static void bond_enable_slave(struct iface *iface, bool enable); 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); @@ -245,13 +238,13 @@ static void port_update_bond_compat(struct port *); static void port_update_vlan_compat(struct port *); static void port_update_bonding(struct port *); -static struct mirror *mirror_create(struct bridge *, const char *name); +static void mirror_create(struct bridge *, struct ovsrec_mirror *); static void mirror_destroy(struct mirror *); static void mirror_reconfigure(struct bridge *); static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *); static bool vlan_is_mirrored(const struct mirror *, int vlan); -static struct iface *iface_create(struct port *port, +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); @@ -259,54 +252,71 @@ static struct iface *iface_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); static bool iface_is_internal(const struct bridge *, const char *name); static void iface_set_mac(struct iface *); +static void iface_update_qos(struct iface *, const struct ovsrec_qos *); /* Hooks into ofproto processing. */ static struct ofhooks bridge_ofhooks; /* Public functions. */ -/* Adds the name of each interface used by a bridge, including local and - * internal ports, to 'svec'. */ +/* Initializes the bridge module, configuring it to obtain its configuration + * from an OVSDB server accessed over 'remote', which should be a string in a + * form acceptable to ovsdb_idl_create(). */ void -bridge_get_ifaces(struct svec *svec) +bridge_init(const char *remote) { - struct bridge *br, *next; - size_t i, j; + /* Create connection to database. */ + idl = ovsdb_idl_create(remote, &ovsrec_idl_class); - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + 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(idl, &ovsrec_open_vswitch_col_external_ids); - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->dp_ifidx < 0) { - VLOG_ERR("%s interface not in datapath %s, ignoring", - iface->name, dpif_name(br->dpif)); - } else { - if (iface->dp_ifidx != ODPP_LOCAL) { - svec_add(svec, iface->name); - } - } - } - } - } + ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); + + 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(idl, &ovsrec_interface_col_external_ids); + + /* Register unixctl commands. */ + unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, + NULL); + unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect, + NULL); + bond_init(); } -void -bridge_init(const struct ovsrec_open_vswitch *cfg) +/* 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; size_t i; - unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + /* Only do this once per ovs-vswitchd run. */ + if (already_configured_once) { + return; + } + already_configured_once = true; + + iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL; + /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */ svec_init(&bridge_names); for (i = 0; i < cfg->n_bridges; i++) { svec_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); dp_enumerate_types(&dpif_types); @@ -317,12 +327,14 @@ bridge_init(const struct ovsrec_open_vswitch *cfg) 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++) { @@ -330,7 +342,10 @@ bridge_init(const struct ovsrec_open_vswitch *cfg) goto found; } } + + /* No. Delete the dpif. */ dpif_delete(dpif); + found: svec_destroy(&all_names); dpif_close(dpif); @@ -340,26 +355,7 @@ bridge_init(const struct ovsrec_open_vswitch *cfg) svec_destroy(&bridge_names); svec_destroy(&dpif_names); svec_destroy(&dpif_types); - - unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, - NULL); - - bond_init(); - bridge_reconfigure(cfg); -} - -#ifdef HAVE_OPENSSL -static void -bridge_configure_ssl(const struct ovsrec_ssl *ssl) -{ - /* XXX SSL should be configurable on a per-bridge basis. */ - if (ssl) { - stream_ssl_set_private_key_file(ssl->private_key); - stream_ssl_set_certificate_file(ssl->certificate); - stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert); - } } -#endif /* Attempt to create the network device 'iface_name' through the netdev * library. */ @@ -367,7 +363,6 @@ static int set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface, bool create) { - struct shash_node *node; struct shash options; int error = 0; size_t i; @@ -391,10 +386,6 @@ set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface, } netdev_options.args = &options; netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; - netdev_options.may_create = true; - if (iface_is_internal(iface->port->bridge, iface_cfg->name)) { - netdev_options.may_open = true; - } error = netdev_open(&netdev_options, &iface->netdev); @@ -419,11 +410,7 @@ set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface, error = EINVAL; } } - - SHASH_FOR_EACH (node, &options) { - free(node->data); - } - shash_destroy(&options); + shash_destroy_free_data(&options); return error; } @@ -554,10 +541,9 @@ collect_managers(const struct ovsrec_open_vswitch *ovs_cfg, *n_managersp = n_managers; } -void +static void bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) { - struct ovsdb_idl_txn *txn; struct shash old_br, new_br; struct shash_node *node; struct bridge *br, *next; @@ -568,8 +554,6 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) COVERAGE_INC(bridge_reconfigure); - txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl); - collect_managers(ovs_cfg, &managers, &n_managers); /* Collect old and new bridges. */ @@ -612,14 +596,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) shash_destroy(&old_br); shash_destroy(&new_br); -#ifdef HAVE_OPENSSL - /* Configure SSL. */ - bridge_configure_ssl(ovs_cfg->ssl); -#endif - /* Reconfigure all bridges. */ LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - bridge_reconfigure_one(ovs_cfg, br); + bridge_reconfigure_one(br); } /* Add and delete ports on all datapaths. @@ -660,9 +639,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) shash_init(&cur_ifaces); for (i = 0; i < n_dpif_ports; i++) { const char *name = dpif_ports[i].devname; - if (!shash_find(&cur_ifaces, name)) { - shash_add(&cur_ifaces, name, NULL); - } + shash_add_once(&cur_ifaces, name, NULL); } free(dpif_ports); @@ -729,7 +706,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface); ofproto_set_datapath_id(br->ofproto, dpid); - dpid_string = xasprintf("%012"PRIx64, dpid); + dpid_string = xasprintf("%016"PRIx64, dpid); ovsrec_bridge_set_datapath_id(br->cfg, dpid_string); free(dpid_string); @@ -775,7 +752,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) opts.collectors.n = nf_cfg->n_targets; opts.collectors.names = nf_cfg->targets; if (ofproto_set_netflow(br->ofproto, &opts)) { - VLOG_ERR("bridge %s: problem setting netflow collectors", + VLOG_ERR("bridge %s: problem setting netflow collectors", br->name); } } else { @@ -814,7 +791,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) oso.agent_device = sflow_cfg->agent; oso.control_ip = NULL; - n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers); + n_controllers = bridge_get_controllers(br, &controllers); for (i = 0; i < n_controllers; i++) { if (controllers[i]->local_ip) { oso.control_ip = controllers[i]->local_ip; @@ -823,7 +800,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } ofproto_set_sflow(br->ofproto, &oso); - svec_destroy(&oso.targets); + /* Do not destroy oso.targets because it is owned by sflow_cfg. */ } else { ofproto_set_sflow(br->ofproto, NULL); } @@ -837,48 +814,48 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) * yet; when a controller is configured, resetting the datapath ID will * immediately disconnect from the controller, so it's better to set * the datapath ID before the controller. */ - bridge_reconfigure_controller(ovs_cfg, br, managers, n_managers); + bridge_reconfigure_remotes(br, managers, n_managers); } LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; + int j; port_update_vlan_compat(port); port_update_bonding(port); + + for (j = 0; j < port->n_ifaces; j++) { + iface_update_qos(port->ifaces[j], port->cfg->qos); + } } } LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { iterate_and_prune_ifaces(br, set_iface_properties, NULL); } - ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg); - - ovsdb_idl_txn_commit(txn); - ovsdb_idl_txn_destroy(txn); /* XXX */ - free(managers); } static const char * -get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n) +get_ovsrec_key_value(const struct ovsdb_idl_row *row, + const struct ovsdb_idl_column *column, + const char *key) { - size_t i; - - for (i = 0; i < n; i++) { - if (!strcmp(keys[i], key)) { - return values[i]; - } - } - return NULL; + const struct ovsdb_datum *datum; + union ovsdb_atom atom; + unsigned int idx; + + datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING); + atom.string = (char *) key; + idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING); + return idx == UINT_MAX ? NULL : datum->values[idx].string; } static const char * bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key) { - return get_ovsrec_key_value(key, - br_cfg->key_other_config, - br_cfg->value_other_config, - br_cfg->n_other_config); + return get_ovsrec_key_value(&br_cfg->header_, + &ovsrec_bridge_col_other_config, key); } static void @@ -1074,25 +1051,131 @@ dpid_from_hash(const void *data, size_t n) return eth_addr_to_uint64(hash); } -int +static void +iface_refresh_stats(struct iface *iface) +{ + struct iface_stat { + char *name; + int offset; + }; + static const struct iface_stat iface_stats[] = { + { "rx_packets", offsetof(struct netdev_stats, rx_packets) }, + { "tx_packets", offsetof(struct netdev_stats, tx_packets) }, + { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) }, + { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) }, + { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) }, + { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) }, + { "rx_errors", offsetof(struct netdev_stats, rx_errors) }, + { "tx_errors", offsetof(struct netdev_stats, tx_errors) }, + { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) }, + { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) }, + { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) }, + { "collisions", offsetof(struct netdev_stats, collisions) }, + }; + enum { N_STATS = ARRAY_SIZE(iface_stats) }; + const struct iface_stat *s; + + char *keys[N_STATS]; + int64_t values[N_STATS]; + int n; + + struct netdev_stats stats; + + /* Intentionally ignore return value, since errors will set 'stats' to + * all-1s, and we will deal with that correctly below. */ + netdev_get_stats(iface->netdev, &stats); + + n = 0; + for (s = iface_stats; s < &iface_stats[N_STATS]; s++) { + uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset); + if (value != UINT64_MAX) { + keys[n] = s->name; + values[n] = value; + n++; + } + } + + ovsrec_interface_set_statistics(iface->cfg, keys, values, n); +} + +void bridge_run(void) { - struct bridge *br, *next; - int retval; + const struct ovsrec_open_vswitch *cfg; - retval = 0; - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { + bool datapath_destroyed; + bool database_changed; + struct bridge *br; + + /* Let each bridge do the work that it needs to do. */ + datapath_destroyed = false; + LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { int error = bridge_run_one(br); if (error) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, " "forcing reconfiguration", br->name); - if (!retval) { - retval = error; + datapath_destroyed = true; + } + } + + /* (Re)configure if necessary. */ + database_changed = ovsdb_idl_run(idl); + cfg = ovsrec_open_vswitch_first(idl); + if (database_changed || datapath_destroyed) { + if (cfg) { + struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl); + + bridge_configure_once(cfg); + bridge_reconfigure(cfg); + + ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg); + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); /* XXX */ + } else { + /* We still need to reconfigure to avoid dangling pointers to + * now-destroyed ovsrec structures inside bridge data. */ + static const struct ovsrec_open_vswitch null_cfg; + + bridge_reconfigure(&null_cfg); + } + } + +#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 interface stats if necessary. */ + if (time_msec() >= iface_stats_timer) { + struct ovsdb_idl_txn *txn; + + txn = ovsdb_idl_txn_create(idl); + LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + size_t i; + + for (i = 0; i < br->n_ports; i++) { + struct port *port = br->ports[i]; + size_t j; + + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + iface_refresh_stats(iface); + } } } + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); /* XXX */ + + iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL; } - return retval; } void @@ -1102,13 +1185,15 @@ bridge_wait(void) LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { ofproto_wait(br->ofproto); - if (ofproto_has_controller(br->ofproto)) { + if (ofproto_has_primary_controller(br->ofproto)) { continue; } mac_learning_wait(br->ml); bond_wait(br); } + ovsdb_idl_wait(idl); + poll_timer_wait_until(iface_stats_timer); } /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s @@ -1201,11 +1286,13 @@ bridge_create(const struct ovsrec_bridge *br_cfg) br->name = xstrdup(br_cfg->name); br->cfg = br_cfg; br->ml = mac_learning_create(); - br->sent_config_request = false; eth_addr_nicira_random(br->default_ea); port_array_init(&br->ifaces); + shash_init(&br->port_by_name); + shash_init(&br->iface_by_name); + br->flush = false; list_push_back(&all_bridges, &br->node); @@ -1234,6 +1321,8 @@ bridge_destroy(struct bridge *br) ofproto_destroy(br->ofproto); mac_learning_destroy(br->ml); port_array_destroy(&br->ifaces); + shash_destroy(&br->port_by_name); + shash_destroy(&br->iface_by_name); free(br->ports); free(br->name); free(br); @@ -1253,19 +1342,6 @@ bridge_lookup(const char *name) return NULL; } -bool -bridge_exists(const char *name) -{ - return bridge_lookup(name) ? true : false; -} - -uint64_t -bridge_get_datapathid(const char *name) -{ - struct bridge *br = bridge_lookup(name); - return br ? ofproto_get_datapath_id(br->ofproto) : 0; -} - /* Handle requests for a listing of all flows known by the OpenFlow * stack, including those normally hidden. */ static void @@ -1274,7 +1350,7 @@ bridge_unixctl_dump_flows(struct unixctl_conn *conn, { struct bridge *br; struct ds results; - + br = bridge_lookup(args); if (!br) { unixctl_command_reply(conn, 501, "Unknown bridge"); @@ -1288,6 +1364,29 @@ bridge_unixctl_dump_flows(struct unixctl_conn *conn, ds_destroy(&results); } +/* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller + * connections and reconnect. If BRIDGE is not specified, then all bridges + * drop their controller connections and reconnect. */ +static void +bridge_unixctl_reconnect(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct bridge *br; + if (args[0] != '\0') { + br = bridge_lookup(args); + if (!br) { + unixctl_command_reply(conn, 501, "Unknown bridge"); + return; + } + ofproto_reconnect_controllers(br->ofproto); + } else { + LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + ofproto_reconnect_controllers(br->ofproto); + } + } + unixctl_command_reply(conn, 200, NULL); +} + static int bridge_run_one(struct bridge *br) { @@ -1308,20 +1407,14 @@ bridge_run_one(struct bridge *br) } static size_t -bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg, - const struct bridge *br, +bridge_get_controllers(const struct bridge *br, struct ovsrec_controller ***controllersp) { struct ovsrec_controller **controllers; size_t n_controllers; - if (br->cfg->n_controller) { - controllers = br->cfg->controller; - n_controllers = br->cfg->n_controller; - } else { - controllers = ovs_cfg->controller; - n_controllers = ovs_cfg->n_controller; - } + controllers = br->cfg->controller; + n_controllers = br->cfg->n_controller; if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) { controllers = NULL; @@ -1334,100 +1427,13 @@ bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg, return n_controllers; } -static bool -check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_) -{ - struct svec *ifaces = ifaces_; - if (!svec_contains(ifaces, iface->name)) { - svec_add(ifaces, iface->name); - svec_sort(ifaces); - return true; - } else { - VLOG_ERR("bridge %s: %s interface is on multiple ports, " - "removing from %s", - br->name, iface->name, iface->port->name); - return false; - } -} - static void -bridge_update_desc(struct bridge *br OVS_UNUSED) -{ -#if 0 - bool changed = false; - const char *desc; - - desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name); - if (desc != br->mfr_desc) { - free(br->mfr_desc); - if (desc) { - br->mfr_desc = xstrdup(desc); - } else { - br->mfr_desc = xstrdup(DEFAULT_MFR_DESC); - } - changed = true; - } - - desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name); - if (desc != br->hw_desc) { - free(br->hw_desc); - if (desc) { - br->hw_desc = xstrdup(desc); - } else { - br->hw_desc = xstrdup(DEFAULT_HW_DESC); - } - changed = true; - } - - desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name); - if (desc != br->sw_desc) { - free(br->sw_desc); - if (desc) { - br->sw_desc = xstrdup(desc); - } else { - br->sw_desc = xstrdup(DEFAULT_SW_DESC); - } - changed = true; - } - - desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name); - if (desc != br->serial_desc) { - free(br->serial_desc); - if (desc) { - br->serial_desc = xstrdup(desc); - } else { - br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); - } - changed = true; - } - - desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name); - if (desc != br->dp_desc) { - free(br->dp_desc); - if (desc) { - br->dp_desc = xstrdup(desc); - } else { - br->dp_desc = xstrdup(DEFAULT_DP_DESC); - } - changed = true; - } - - if (changed) { - ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc, - br->sw_desc, br->serial_desc, br->dp_desc); - } -#endif -} - -static void -bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, - struct bridge *br) +bridge_reconfigure_one(struct bridge *br) { struct shash old_ports, new_ports; - struct svec ifaces; - struct svec listeners, old_listeners; struct svec snoops, old_snoops; struct shash_node *node; + enum ofproto_fail_mode fail_mode; size_t i; /* Collect old ports. */ @@ -1450,7 +1456,7 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, * user didn't specify one. * * XXX perhaps we should synthesize a port ourselves in this case. */ - if (bridge_get_controllers(ovs_cfg, br, NULL)) { + if (bridge_get_controllers(br, NULL)) { char local_name[IF_NAMESIZE]; int error; @@ -1463,88 +1469,56 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, } } - /* Get rid of deleted ports and add new ports. */ + /* Get rid of deleted ports. + * Get rid of deleted interfaces on ports that still exist. */ SHASH_FOR_EACH (node, &old_ports) { - if (!shash_find(&new_ports, node->name)) { - port_destroy(node->data); + struct port *port = node->data; + const struct ovsrec_port *port_cfg; + + port_cfg = shash_find_data(&new_ports, node->name); + if (!port_cfg) { + port_destroy(port); + } else { + port_del_ifaces(port, port_cfg); } } + + /* Create new ports. + * 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); if (!port) { port = port_create(br, node->name); } + port_reconfigure(port, node->data); + if (!port->n_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); - /* Check and delete duplicate interfaces. */ - svec_init(&ifaces); - iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces); - svec_destroy(&ifaces); + /* Set the fail-mode */ + fail_mode = !br->cfg->fail_mode + || !strcmp(br->cfg->fail_mode, "standalone") + ? OFPROTO_FAIL_STANDALONE + : OFPROTO_FAIL_SECURE; + if (ofproto_get_fail_mode(br->ofproto) != fail_mode + && !ofproto_has_primary_controller(br->ofproto)) { + ofproto_flush_flows(br->ofproto); + } + ofproto_set_fail_mode(br->ofproto, fail_mode); /* Delete all flows if we're switching from connected to standalone or vice * versa. (XXX Should we delete all flows if we are switching from one * controller to another?) */ -#if 0 - /* Configure OpenFlow management listeners. */ - svec_init(&listeners); - cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name); - if (!listeners.n) { - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) { - svec_clear(&listeners); - } - svec_sort_unique(&listeners); - - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - svec_sort_unique(&old_listeners); - - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); - } - svec_destroy(&listeners); - svec_destroy(&old_listeners); - /* Configure OpenFlow controller connection snooping. */ svec_init(&snoops); - cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name); - if (!snoops.n) { - svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); - } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) { - svec_clear(&snoops); - } - svec_sort_unique(&snoops); - - svec_init(&old_snoops); - ofproto_get_snoops(br->ofproto, &old_snoops); - svec_sort_unique(&old_snoops); - - if (!svec_equal(&snoops, &old_snoops)) { - ofproto_set_snoops(br->ofproto, &snoops); - } - svec_destroy(&snoops); - svec_destroy(&old_snoops); -#else - /* Default listener. */ - svec_init(&listeners); - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); - } - svec_destroy(&listeners); - svec_destroy(&old_listeners); - - /* Default snoop. */ - svec_init(&snoops); svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", ovs_rundir, br->name)); svec_init(&old_snoops); @@ -1554,114 +1528,159 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, } svec_destroy(&snoops); svec_destroy(&old_snoops); -#endif mirror_reconfigure(br); +} + +/* Initializes 'oc' appropriately as a management service controller for + * 'br'. + * + * The caller must free oc->target when it is no longer needed. */ +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->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; +} + +/* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */ +static void +bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c, + struct ofproto_controller *oc) +{ + oc->target = c->target; + oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; + 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); +} + +/* Configures the IP stack for 'br''s local interface properly according to the + * configuration in 'c'. */ +static void +bridge_configure_local_iface_netdev(struct bridge *br, + struct ovsrec_controller *c) +{ + struct netdev *netdev; + struct in_addr mask, gateway; + + struct iface *local_iface; + struct in_addr ip; + + /* Controller discovery does its own TCP/IP configuration later. */ + if (strcmp(c->target, "discover")) { + return; + } + + /* If there's no local interface or no IP address, give up. */ + local_iface = bridge_get_local_iface(br); + if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) { + return; + } + + /* Bring up the local interface. */ + netdev = local_iface->netdev; + netdev_turn_flags_on(netdev, NETDEV_UP, true); + + /* Configure the IP address and netmask. */ + if (!c->local_netmask + || !inet_aton(c->local_netmask, &mask) + || !mask.s_addr) { + mask.s_addr = guess_netmask(ip.s_addr); + } + if (!netdev_set_in4(netdev, ip, mask)) { + VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT, + br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr)); + } - bridge_update_desc(br); + /* Configure the default gateway. */ + if (c->local_gateway + && inet_aton(c->local_gateway, &gateway) + && gateway.s_addr) { + if (!netdev_add_router(netdev, gateway)) { + VLOG_INFO("bridge %s: configured gateway "IP_FMT, + br->name, IP_ARGS(&gateway.s_addr)); + } + } } static void -bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg, - struct bridge *br, - const struct sockaddr_in *managers, - size_t n_managers) +bridge_reconfigure_remotes(struct bridge *br, + const struct sockaddr_in *managers, + size_t n_managers) { struct ovsrec_controller **controllers; size_t n_controllers; + bool had_primary; + + struct ofproto_controller *ocs; + size_t n_ocs; + size_t i; ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); + had_primary = ofproto_has_primary_controller(br->ofproto); + + n_controllers = bridge_get_controllers(br, &controllers); + + ocs = xmalloc((n_controllers + 1) * sizeof *ocs); + n_ocs = 0; - n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers); - if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) { + bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]); + for (i = 0; i < n_controllers; i++) { + struct ovsrec_controller *c = controllers[i]; + + if (!strncmp(c->target, "punix:", 6) + || !strncmp(c->target, "unix:", 5)) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + + /* Prevent remote ovsdb-server users from accessing arbitrary Unix + * domain sockets and overwriting arbitrary local files. */ + VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller " + "\"%s\" due to possibility for remote exploit", + dpif_name(br->dpif), c->target); + continue; + } + + bridge_configure_local_iface_netdev(br, c); + bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]); + } + + ofproto_set_controllers(br->ofproto, ocs, n_ocs); + free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */ + free(ocs); + + if (had_primary != ofproto_has_primary_controller(br->ofproto)) { ofproto_flush_flows(br->ofproto); } - if (!n_controllers) { + /* If there are no controllers and the bridge is in standalone + * mode, set up a flow that matches every packet and directs + * them to OFPP_NORMAL (which goes to us). Otherwise, the + * switch is in secure mode and we won't pass any traffic until + * a controller has been defined and it tells us to do so. */ + if (!n_controllers + && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) { union ofp_action action; flow_t flow; - /* Clear out controllers. */ - ofproto_set_controllers(br->ofproto, NULL, 0); - - /* Set up a flow that matches every packet and directs them to - * OFPP_NORMAL (which goes to us). */ 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); - } else { - struct ofproto_controller *ocs; - size_t i; - - ocs = xmalloc(n_controllers * sizeof *ocs); - for (i = 0; i < n_controllers; i++) { - struct ovsrec_controller *c = controllers[i]; - struct ofproto_controller *oc = &ocs[i]; - - if (strcmp(c->target, "discover")) { - struct iface *local_iface; - struct in_addr ip; - - local_iface = bridge_get_local_iface(br); - if (local_iface && c->local_ip - && inet_aton(c->local_ip, &ip)) { - struct netdev *netdev = local_iface->netdev; - struct in_addr mask, gateway; - - if (!c->local_netmask - || !inet_aton(c->local_netmask, &mask)) { - mask.s_addr = 0; - } - if (!c->local_gateway - || !inet_aton(c->local_gateway, &gateway)) { - gateway.s_addr = 0; - } - - netdev_turn_flags_on(netdev, NETDEV_UP, true); - if (!mask.s_addr) { - mask.s_addr = guess_netmask(ip.s_addr); - } - if (!netdev_set_in4(netdev, ip, mask)) { - VLOG_INFO("bridge %s: configured IP address "IP_FMT", " - "netmask "IP_FMT, - br->name, IP_ARGS(&ip.s_addr), - IP_ARGS(&mask.s_addr)); - } - - if (gateway.s_addr) { - if (!netdev_add_router(netdev, gateway)) { - VLOG_INFO("bridge %s: configured gateway "IP_FMT, - br->name, IP_ARGS(&gateway.s_addr)); - } - } - } - } - - oc->target = c->target; - oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; - oc->probe_interval = (c->inactivity_probe - ? *c->inactivity_probe / 1000 : 5); - oc->fail = (!c->fail_mode - || !strcmp(c->fail_mode, "standalone") - || !strcmp(c->fail_mode, "open") - ? OFPROTO_FAIL_STANDALONE - : OFPROTO_FAIL_SECURE); - 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); - } - ofproto_set_controllers(br->ofproto, ocs, n_controllers); - free(ocs); } } @@ -1926,10 +1945,19 @@ bond_update_fake_iface_stats(struct port *port) struct netdev_stats slave_stats; if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) { - bond_stats.rx_packets += slave_stats.rx_packets; - bond_stats.rx_bytes += slave_stats.rx_bytes; - bond_stats.tx_packets += slave_stats.tx_packets; - bond_stats.tx_bytes += slave_stats.tx_bytes; + /* 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; } } @@ -1982,11 +2010,11 @@ bond_wait(struct bridge *br) for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (iface->delay_expires != LLONG_MAX) { - poll_timer_wait(iface->delay_expires - time_msec()); + poll_timer_wait_until(iface->delay_expires); } } if (port->bond_fake_iface) { - poll_timer_wait(port->bond_next_fake_iface_update - time_msec()); + poll_timer_wait_until(port->bond_next_fake_iface_update); } } } @@ -2070,7 +2098,8 @@ dst_is_duplicate(const struct dst *dsts, size_t n_dsts, static bool port_trunks_vlan(const struct port *port, uint16_t vlan) { - return port->vlan < 0 && bitmap_is_set(port->trunks, vlan); + return (port->vlan < 0 + && (!port->trunks || bitmap_is_set(port->trunks, vlan))); } static bool @@ -2244,12 +2273,34 @@ static int flow_get_vlan(struct bridge *br, const flow_t *flow, return vlan; } +/* A VM broadcasts a gratuitous ARP to indicate that it has resumed after + * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to + * indicate this; newer upstream kernels use gratuitous ARP requests. */ +static bool +is_gratuitous_arp(const flow_t *flow) +{ + return (flow->dl_type == htons(ETH_TYPE_ARP) + && eth_addr_is_broadcast(flow->dl_dst) + && (flow->nw_proto == ARP_OP_REPLY + || (flow->nw_proto == ARP_OP_REQUEST + && flow->nw_src == flow->nw_dst))); +} + static void update_learning_table(struct bridge *br, const flow_t *flow, int vlan, struct port *in_port) { - tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src, - vlan, in_port->port_idx); + 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) { /* The log messages here could actually be useful in debugging, * so keep the rate limit relatively high. */ @@ -2263,14 +2314,6 @@ update_learning_table(struct bridge *br, const flow_t *flow, int vlan, } } -static bool -is_bcast_arp_reply(const flow_t *flow) -{ - return (flow->dl_type == htons(ETH_TYPE_ARP) - && flow->nw_proto == ARP_OP_REPLY - && eth_addr_is_broadcast(flow->dl_dst)); -} - /* Determines whether packets in 'flow' within 'br' should be forwarded or * dropped. Returns true if they may be forwarded, false if they should be * dropped. @@ -2318,7 +2361,7 @@ is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " - "interface %"PRIu16, br->name, flow->in_port); + "interface %"PRIu16, br->name, flow->in_port); } *in_portp = NULL; @@ -2349,6 +2392,7 @@ is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, /* 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 (eth_addr_is_multicast(flow->dl_dst)) { *tags |= in_port->active_iface_tag; @@ -2360,11 +2404,15 @@ is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, /* 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. Broadcast ARP replies are an exception - * to this rule: the host has moved to another switch. */ - src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); + * 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_bcast_arp_reply(flow)) { + (!is_gratuitous_arp(flow) || is_grat_arp_locked)) { return false; } } @@ -2397,7 +2445,8 @@ process_flow(struct bridge *br, const flow_t *flow, } /* Determine output port. */ - out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags); + 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)) { @@ -2470,11 +2519,12 @@ bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, 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 flow_t *flow, +bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags, const union odp_action *actions, size_t n_actions, unsigned long long int n_bytes, void *br_) @@ -2482,20 +2532,24 @@ bridge_account_flow_ofhook_cb(const flow_t *flow, struct bridge *br = br_; const union odp_action *a; struct port *in_port; - tag_type tags = 0; + 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. */ - if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) { + /* 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); @@ -3226,6 +3280,7 @@ port_create(struct bridge *br, const char *name) sizeof *br->ports); } br->ports[br->n_ports++] = port; + shash_add_assert(&br->port_by_name, port->name, port); VLOG_INFO("created port %s on bridge %s", port->name, br->name); bridge_flush(br); @@ -3237,43 +3292,55 @@ static const char * get_port_other_config(const struct ovsrec_port *port, const char *key, const char *default_value) { - const char *value = get_ovsrec_key_value(key, - port->key_other_config, - port->value_other_config, - port->n_other_config); + const char *value; + + value = get_ovsrec_key_value(&port->header_, &ovsrec_port_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; + size_t i; + + /* Collect list of new interfaces. */ + shash_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); + } + + /* 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++; + } + } + + shash_destroy(&new_ifaces); +} + static void port_reconfigure(struct port *port, const struct ovsrec_port *cfg) { - struct shash old_ifaces, new_ifaces; + struct shash new_ifaces; long long int next_rebalance; - struct shash_node *node; unsigned long *trunks; int vlan; size_t i; port->cfg = cfg; - /* Collect old and new interfaces. */ - shash_init(&old_ifaces); - shash_init(&new_ifaces); - for (i = 0; i < port->n_ifaces; i++) { - shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]); - } - for (i = 0; i < cfg->n_interfaces; i++) { - const char *name = cfg->interfaces[i]->name; - if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) { - VLOG_WARN("port %s: %s specified twice as port interface", - port->name, name); - } - } + /* Update settings. */ port->updelay = cfg->bond_updelay; if (port->updelay < 0) { port->updelay = 0; } - port->updelay = cfg->bond_downdelay; + port->downdelay = cfg->bond_downdelay; if (port->downdelay < 0) { port->downdelay = 0; } @@ -3287,23 +3354,32 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) port->bond_next_rebalance = next_rebalance; } - /* Get rid of deleted interfaces and add new interfaces. */ - SHASH_FOR_EACH (node, &old_ifaces) { - if (!shash_find(&new_ifaces, node->name)) { - iface_destroy(node->data); - } - } - SHASH_FOR_EACH (node, &new_ifaces) { - const struct ovsrec_interface *if_cfg = node->data; + /* Add new interfaces and update 'cfg' member of existing ones. */ + shash_init(&new_ifaces); + for (i = 0; i < cfg->n_interfaces; i++) { + const struct ovsrec_interface *if_cfg = cfg->interfaces[i]; struct iface *iface; - iface = shash_find_data(&old_ifaces, if_cfg->name); - if (!iface) { - iface_create(port, if_cfg); - } else { + if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) { + VLOG_WARN("port %s: %s specified twice as port interface", + port->name, if_cfg->name); + continue; + } + + iface = iface_lookup(port->bridge, if_cfg->name); + if (iface) { + if (iface->port != port) { + VLOG_ERR("bridge %s: %s interface is on multiple ports, " + "removing from %s", + port->bridge->name, if_cfg->name, iface->port->name); + continue; + } iface->cfg = if_cfg; + } else { + iface_create(port, if_cfg); } } + shash_destroy(&new_ifaces); /* Get VLAN tag. */ vlan = -1; @@ -3329,7 +3405,7 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) /* Get trunked VLANs. */ trunks = NULL; - if (vlan < 0) { + if (vlan < 0 && cfg->n_trunks) { size_t n_errors; size_t i; @@ -3348,17 +3424,14 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) port->name, cfg->n_trunks); } if (n_errors == cfg->n_trunks) { - if (n_errors) { - VLOG_ERR("port %s: no valid trunks, trunking all VLANs", - port->name); - } - bitmap_set_multiple(trunks, 0, 4096, 1); - } - } else { - if (cfg->n_trunks) { - VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", + VLOG_ERR("port %s: no valid trunks, trunking all VLANs", port->name); + bitmap_free(trunks); + trunks = NULL; } + } else if (vlan >= 0 && cfg->n_trunks) { + VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", + port->name); } if (trunks == NULL ? port->trunks != NULL @@ -3367,9 +3440,6 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) } bitmap_free(port->trunks); port->trunks = trunks; - - shash_destroy(&old_ifaces); - shash_destroy(&new_ifaces); } static void @@ -3394,6 +3464,8 @@ port_destroy(struct port *port) iface_destroy(port->ifaces[port->n_ifaces - 1]); } + 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; @@ -3415,29 +3487,14 @@ port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) static struct port * port_lookup(const struct bridge *br, const char *name) { - size_t i; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (!strcmp(port->name, name)) { - return port; - } - } - return NULL; + return shash_find_data(&br->port_by_name, name); } static struct iface * port_lookup_iface(const struct port *port, const char *name) { - size_t j; - - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (!strcmp(iface->name, name)) { - return iface; - } - } - return NULL; + struct iface *iface = iface_lookup(port->bridge, name); + return iface && iface->port == port ? iface : NULL; } static void @@ -3513,15 +3570,15 @@ port_update_bond_compat(struct port *port) /* 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 + * 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 + * 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; @@ -3590,6 +3647,7 @@ port_update_vlan_compat(struct port *port) static struct iface * iface_create(struct port *port, const struct ovsrec_interface *if_cfg) { + struct bridge *br = port->bridge; struct iface *iface; char *name = if_cfg->name; int error; @@ -3604,28 +3662,34 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) iface->netdev = NULL; iface->cfg = if_cfg; + shash_add_assert(&br->iface_by_name, iface->name, iface); + + /* Attempt to create the network interface in case it doesn't exist yet. */ + if (!iface_is_internal(br, iface->name)) { + error = set_up_iface(if_cfg, iface, true); + if (error) { + VLOG_WARN("could not create iface %s: %s", iface->name, + strerror(error)); + + shash_find_and_delete_assert(&br->iface_by_name, iface->name); + free(iface->name); + free(iface); + return NULL; + } + } + 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) { - port->bridge->has_bonded_ports = true; - } - - /* Attempt to create the network interface in case it - * doesn't exist yet. */ - if (!iface_is_internal(port->bridge, iface->name)) { - error = set_up_iface(if_cfg, iface, true); - if (error) { - VLOG_WARN("could not create iface %s: %s", iface->name, - strerror(error)); - } + br->has_bonded_ports = true; } VLOG_DBG("attached network device %s to port %s", iface->name, port->name); - bridge_flush(port->bridge); + bridge_flush(br); return iface; } @@ -3639,6 +3703,8 @@ iface_destroy(struct iface *iface) bool del_active = port->active_iface == iface->port_ifidx; struct iface *del; + shash_find_and_delete_assert(&br->iface_by_name, iface->name); + if (iface->dp_ifidx >= 0) { port_array_set(&br->ifaces, iface->dp_ifidx, NULL); } @@ -3664,18 +3730,7 @@ iface_destroy(struct iface *iface) static struct iface * iface_lookup(const struct bridge *br, const char *name) { - 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 (!strcmp(iface->name, name)) { - return iface; - } - } - } - return NULL; + return shash_find_data(&br->iface_by_name, name); } static struct iface * @@ -3697,7 +3752,6 @@ iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) static bool iface_is_internal(const struct bridge *br, const char *if_name) { - /* XXX wastes time */ struct iface *iface; struct port *port; @@ -3740,53 +3794,127 @@ iface_set_mac(struct iface *iface) } } } + +static void +shash_from_ovs_idl_map(char **keys, char **values, size_t n, + struct shash *shash) +{ + size_t i; + + shash_init(shash); + for (i = 0; i < n; i++) { + shash_add(shash, keys[i], values[i]); + } +} + +struct iface_delete_queues_cbdata { + struct netdev *netdev; + const struct ovsdb_datum *queues; +}; + +static bool +queue_ids_include(const struct ovsdb_datum *queues, int64_t target) +{ + union ovsdb_atom atom; + + atom.integer = target; + return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX; +} + +static void +iface_delete_queues(unsigned int queue_id, + const struct shash *details OVS_UNUSED, void *cbdata_) +{ + struct iface_delete_queues_cbdata *cbdata = cbdata_; + + if (!queue_ids_include(cbdata->queues, queue_id)) { + netdev_delete_queue(cbdata->netdev, queue_id); + } +} + +static void +iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos) +{ + if (!qos || qos->type[0] == '\0') { + netdev_set_qos(iface->netdev, NULL, NULL); + } else { + struct iface_delete_queues_cbdata cbdata; + struct shash details; + size_t i; + + /* Configure top-level Qos for 'iface'. */ + shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config, + qos->n_other_config, &details); + netdev_set_qos(iface->netdev, qos->type, &details); + shash_destroy(&details); + + /* Deconfigure queues that were deleted. */ + cbdata.netdev = iface->netdev; + cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER, + OVSDB_TYPE_UUID); + netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata); + + /* Configure queues for 'iface'. */ + for (i = 0; i < qos->n_queues; i++) { + const struct ovsrec_queue *queue = qos->value_queues[i]; + unsigned int queue_id = qos->key_queues[i]; + + shash_from_ovs_idl_map(queue->key_other_config, + queue->value_other_config, + queue->n_other_config, &details); + netdev_set_queue(iface->netdev, queue_id, &details); + shash_destroy(&details); + } + } +} /* Port mirroring. */ +static struct mirror * +mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid) +{ + int i; + + for (i = 0; i < MAX_MIRRORS; i++) { + struct mirror *m = br->mirrors[i]; + if (m && uuid_equals(uuid, &m->uuid)) { + return m; + } + } + return NULL; +} + static void mirror_reconfigure(struct bridge *br) { - struct shash old_mirrors, new_mirrors; - struct shash_node *node; unsigned long *rspan_vlans; int i; - /* Collect old mirrors. */ - shash_init(&old_mirrors); + /* Get rid of deleted mirrors. */ for (i = 0; i < MAX_MIRRORS; i++) { - if (br->mirrors[i]) { - shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]); + struct mirror *m = br->mirrors[i]; + if (m) { + const struct ovsdb_datum *mc; + union ovsdb_atom atom; + + mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID); + atom.uuid = br->mirrors[i]->uuid; + if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) { + mirror_destroy(m); + } } } - /* Collect new mirrors. */ - shash_init(&new_mirrors); + /* Add new mirrors and reconfigure existing ones. */ for (i = 0; i < br->cfg->n_mirrors; i++) { struct ovsrec_mirror *cfg = br->cfg->mirrors[i]; - if (!shash_add_once(&new_mirrors, cfg->name, cfg)) { - VLOG_WARN("bridge %s: %s specified twice as mirror", - br->name, cfg->name); - } - } - - /* Get rid of deleted mirrors and add new mirrors. */ - SHASH_FOR_EACH (node, &old_mirrors) { - if (!shash_find(&new_mirrors, node->name)) { - mirror_destroy(node->data); - } - } - SHASH_FOR_EACH (node, &new_mirrors) { - struct mirror *mirror = shash_find_data(&old_mirrors, node->name); - if (!mirror) { - mirror = mirror_create(br, node->name); - if (!mirror) { - break; - } + struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid); + if (m) { + mirror_reconfigure_one(m, cfg); + } else { + mirror_create(br, cfg); } - mirror_reconfigure_one(mirror, node->data); } - shash_destroy(&old_mirrors); - shash_destroy(&new_mirrors); /* Update port reserved status. */ for (i = 0; i < br->n_ports; i++) { @@ -3821,8 +3949,8 @@ mirror_reconfigure(struct bridge *br) } } -static struct mirror * -mirror_create(struct bridge *br, const char *name) +static void +mirror_create(struct bridge *br, struct ovsrec_mirror *cfg) { struct mirror *m; size_t i; @@ -3830,21 +3958,21 @@ mirror_create(struct bridge *br, const char *name) for (i = 0; ; i++) { if (i >= MAX_MIRRORS) { VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " - "cannot create %s", br->name, MAX_MIRRORS, name); - return NULL; + "cannot create %s", br->name, MAX_MIRRORS, cfg->name); + return; } if (!br->mirrors[i]) { break; } } - VLOG_INFO("created port mirror %s on bridge %s", name, br->name); + VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name); bridge_flush(br); br->mirrors[i] = m = xzalloc(sizeof *m); m->bridge = br; m->idx = i; - m->name = xstrdup(name); + m->name = xstrdup(cfg->name); shash_init(&m->src_ports); shash_init(&m->dst_ports); m->vlans = NULL; @@ -3852,7 +3980,7 @@ mirror_create(struct bridge *br, const char *name) m->out_vlan = -1; m->out_port = NULL; - return m; + mirror_reconfigure_one(m, cfg); } static void @@ -3872,6 +4000,7 @@ mirror_destroy(struct mirror *m) free(m->vlans); m->bridge->mirrors[m->idx] = NULL; + free(m->name); free(m); bridge_flush(br); @@ -3953,6 +4082,12 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) int *vlans; size_t i; + /* Set name. */ + if (strcmp(cfg->name, m->name)) { + free(m->name); + m->name = xstrdup(cfg->name); + } + /* Get output port. */ if (cfg->output_port) { out_port = port_lookup(m->bridge, cfg->output_port->name);