X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=b7f068bf0cd181407bc7cf74b8cca5b52c774bab;hb=efdd908807544de7bee8644fafc1400e6a32d936;hp=0a8a9e09b49ff59cb6b9dac15784f9207a5ba6d0;hpb=2c30e5d117bd45afaf0a107509b2c92b9a797380;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 0a8a9e09b..b7f068bf0 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -20,6 +20,7 @@ #include #include #include +#include #include #include #include @@ -36,6 +37,7 @@ #include "dynamic-string.h" #include "flow.h" #include "hash.h" +#include "jsonrpc.h" #include "list.h" #include "mac-learning.h" #include "netdev.h" @@ -61,6 +63,7 @@ #include "vswitchd/vswitch-idl.h" #include "xenserver.h" #include "xtoxll.h" +#include "sflow_api.h" #define THIS_MODULE VLM_bridge #include "vlog.h" @@ -79,12 +82,10 @@ 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; }; @@ -105,8 +106,8 @@ struct mirror { char *name; /* Selection criteria. */ - struct svec src_ports; - struct svec dst_ports; + struct shash src_ports; /* Name is port name; data is always NULL. */ + struct shash dst_ports; /* Name is port name; data is always NULL. */ int *vlans; size_t n_vlans; @@ -120,7 +121,9 @@ 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. @@ -136,14 +139,14 @@ 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. */ + int bond_rebalance_interval; /* Interval between rebalances, in ms. */ + long long int bond_next_rebalance; /* Next rebalancing time. */ /* Port mirroring info. */ 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 @@ -151,13 +154,8 @@ 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. */ - - /* Support for remote controllers. */ - char *controller; /* NULL if there is no remote controller; - * "discover" to do controller discovery; - * otherwise a vconn name. */ + const struct ovsrec_bridge *cfg; /* OpenFlow switch processing. */ struct ofproto *ofproto; /* OpenFlow switch. */ @@ -169,39 +167,45 @@ 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; - long long int bond_next_rebalance; /* 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; -static struct bridge *bridge_create(const char *name); +/* 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, + 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 *); +static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *, + 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 *); @@ -225,6 +229,7 @@ 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); @@ -234,18 +239,11 @@ static void port_update_bond_compat(struct port *); static void port_update_vlan_compat(struct port *); static void port_update_bonding(struct port *); -#if 0 -static void mirror_create(struct bridge *, const char *name); +static struct mirror *mirror_create(struct bridge *, const char *name); static void mirror_destroy(struct mirror *); static void mirror_reconfigure(struct bridge *); -static void mirror_reconfigure_one(struct mirror *); +static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *); static bool vlan_is_mirrored(const struct mirror *, int vlan); -#else -static bool vlan_is_mirrored(const struct mirror *m UNUSED, int vlan UNUSED) -{ - return false; -} -#endif static struct iface *iface_create(struct port *port, const struct ovsrec_interface *if_cfg); @@ -255,132 +253,107 @@ 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; - - 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]; + /* Create connection to database. */ + idl = ovsdb_idl_create(remote, &ovsrec_idl_class); - 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); - } - } - } - } - } + /* 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; + 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); - dp_enumerate(&dpif_names); - for (i = 0; i < dpif_names.n; i++) { - const char *dpif_name = dpif_names.names[i]; + svec_init(&dpif_types); + dp_enumerate_types(&dpif_types); + for (i = 0; i < dpif_types.n; i++) { struct dpif *dpif; int retval; + size_t j; - retval = dpif_open(dpif_name, &dpif); - if (!retval) { - struct svec all_names; - size_t j; + dp_enumerate_names(dpif_types.names[i], &dpif_names); - svec_init(&all_names); - dpif_get_all_names(dpif, &all_names); - for (j = 0; j < all_names.n; j++) { - if (svec_contains(&bridge_names, all_names.names[j])) { - goto found; + /* 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); } - dpif_delete(dpif); - found: - svec_destroy(&all_names); - dpif_close(dpif); } } + svec_destroy(&bridge_names); svec_destroy(&dpif_names); - - unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, - NULL); - - bond_init(); - bridge_reconfigure(cfg); + svec_destroy(&dpif_types); } #ifdef HAVE_OPENSSL -static bool -config_string_change(const char *value, char **valuep) -{ - if (value && (!*valuep || strcmp(value, *valuep))) { - free(*valuep); - *valuep = xstrdup(value); - return true; - } else { - return false; - } -} - static void bridge_configure_ssl(const struct ovsrec_ssl *ssl) { - /* XXX SSL should be configurable on a per-bridge basis. - * XXX should be possible to de-configure SSL. */ - static char *private_key_file; - static char *certificate_file; - static char *cacert_file; - struct stat s; - - if (!ssl) { - /* XXX We can't un-set SSL settings. */ - return; - } - - if (config_string_change(ssl->private_key, &private_key_file)) { - stream_ssl_set_private_key_file(private_key_file); - } - - if (config_string_change(ssl->certificate, &certificate_file)) { - stream_ssl_set_certificate_file(certificate_file); - } - - /* We assume that even if the filename hasn't changed, if the CA cert - * file has been removed, that we want to move back into - * boot-strapping mode. This opens a small security hole, because - * the old certificate will still be trusted until vSwitch is - * restarted. We may want to address this in vconn's SSL library. */ - if (config_string_change(ssl->ca_cert, &cacert_file) - || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) { - stream_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert); + /* 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 @@ -388,19 +361,13 @@ bridge_configure_ssl(const struct ovsrec_ssl *ssl) /* Attempt to create the network device 'iface_name' through the netdev * library. */ static int -set_up_iface(const struct ovsrec_interface *iface_cfg, bool create) +set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface, + bool create) { - struct shash_node *node; struct shash options; - int error; + int error = 0; size_t i; - /* If a type is not explicitly declared, then assume it's an existing - * "system" device. */ - if (iface_cfg->type[0] == '\0' || !strcmp(iface_cfg->type, "system")) { - return 0; - } - shash_init(&options); for (i = 0; i < iface_cfg->n_options; i++) { shash_add(&options, iface_cfg->key_options[i], @@ -408,49 +375,72 @@ set_up_iface(const struct ovsrec_interface *iface_cfg, bool create) } if (create) { - error = netdev_create(iface_cfg->name, iface_cfg->type, &options); - } else { - /* xxx Check to make sure that the type hasn't changed. */ - error = netdev_reconfigure(iface_cfg->name, &options); - } + struct netdev_options netdev_options; + + memset(&netdev_options, 0, sizeof netdev_options); + netdev_options.name = iface_cfg->name; + if (!strcmp(iface_cfg->type, "internal")) { + /* An "internal" config type maps to a netdev "system" type. */ + netdev_options.type = "system"; + } else { + netdev_options.type = iface_cfg->type; + } + netdev_options.args = &options; + netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; + + error = netdev_open(&netdev_options, &iface->netdev); + + if (iface->netdev) { + netdev_get_carrier(iface->netdev, &iface->enabled); + } + } else if (iface->netdev) { + const char *netdev_type = netdev_get_type(iface->netdev); + const char *iface_type = iface_cfg->type && strlen(iface_cfg->type) + ? iface_cfg->type : NULL; + + /* An "internal" config type maps to a netdev "system" type. */ + if (iface_type && !strcmp(iface_type, "internal")) { + iface_type = "system"; + } - SHASH_FOR_EACH (node, &options) { - free(node->data); + if (!iface_type || !strcmp(netdev_type, iface_type)) { + error = netdev_reconfigure(iface->netdev, &options); + } else { + VLOG_WARN("%s: attempting change device type from %s to %s", + iface_cfg->name, netdev_type, iface_type); + error = EINVAL; + } } - shash_destroy(&options); + shash_destroy_free_data(&options); return error; } static int -reconfigure_iface(const struct ovsrec_interface *iface_cfg) +reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface) { - return set_up_iface(iface_cfg, false); + return set_up_iface(iface_cfg, iface, false); } - -/* iterate_and_prune_ifaces() callback function that opens the network device - * for 'iface', if it is not already open, and retrieves the interface's MAC - * address and carrier status. */ static bool -init_iface_netdev(struct bridge *br UNUSED, struct iface *iface, - void *aux UNUSED) +check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface, + void *aux OVS_UNUSED) { - if (iface->netdev) { - return true; - } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE, - &iface->netdev)) { - netdev_get_carrier(iface->netdev, &iface->enabled); - return true; - } else { - /* If the network device can't be opened, then we're not going to try - * to do anything with this interface. */ - return false; + if (!iface->netdev) { + int error = set_up_iface(iface->cfg, iface, true); + if (error) { + VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name, + strerror(error)); + return false; + } } + + return true; } static bool -check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED) +check_iface_dp_ifidx(struct bridge *br, struct iface *iface, + void *aux OVS_UNUSED) { if (iface->dp_ifidx >= 0) { VLOG_DBG("%s has interface %s on port %d", @@ -465,8 +455,8 @@ check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED) } static bool -set_iface_properties(struct bridge *br UNUSED, struct iface *iface, - void *aux UNUSED) +set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, + void *aux OVS_UNUSED) { /* Set policing attributes. */ netdev_set_policing(iface->netdev, @@ -514,18 +504,58 @@ iterate_and_prune_ifaces(struct bridge *br, } } -void +/* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP + * addresses and ports into '*managersp' and '*n_managersp'. The caller is + * responsible for freeing '*managersp' (with free()). + * + * You may be asking yourself "why does ovs-vswitchd care?", because + * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd + * 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. + */ +static void +collect_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; + + if (ovs_cfg->n_managers > 0) { + size_t i; + + 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]; + + 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))) { + n_managers++; + } + } + } + + *managersp = managers; + *n_managersp = n_managers; +} + +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; + struct sockaddr_in *managers; + size_t n_managers; size_t i; + int sflow_bridge_number; 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. */ shash_init(&old_br); @@ -552,11 +582,16 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) SHASH_FOR_EACH (node, &new_br) { const char *br_name = node->name; const struct ovsrec_bridge *br_cfg = node->data; - if (!shash_find_data(&old_br, br_name)) { - br = bridge_create(br_name); - if (br) { - br->cfg = br_cfg; + br = shash_find_data(&old_br, br_name); + if (br) { + /* If the bridge datapath type has changed, we need to tear it + * down and recreate. */ + if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) { + bridge_destroy(br); + bridge_create(br_cfg); } + } else { + bridge_create(br_cfg); } } shash_destroy(&old_br); @@ -610,9 +645,7 @@ 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); @@ -626,7 +659,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) if (shash_find(&cur_ifaces, if_name)) { /* Already exists, just reconfigure it. */ if (iface) { - reconfigure_iface(iface->cfg); + reconfigure_iface(iface->cfg, iface); } } else { /* Need to add to datapath. */ @@ -650,6 +683,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) shash_destroy(&cur_ifaces); shash_destroy(&want_ifaces); } + sflow_bridge_number = 0; LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { uint8_t ea[8]; uint64_t dpid; @@ -658,8 +692,8 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) char *dpid_string; bridge_fetch_dp_ifaces(br); - iterate_and_prune_ifaces(br, init_iface_netdev, NULL); + iterate_and_prune_ifaces(br, check_iface_netdev, NULL); iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL); /* Pick local port hardware address, datapath ID. */ @@ -678,7 +712,7 @@ 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); @@ -691,10 +725,10 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id); if (nf_cfg->engine_type) { - opts.engine_type = nf_cfg->engine_type; + opts.engine_type = *nf_cfg->engine_type; } if (nf_cfg->engine_id) { - opts.engine_id = nf_cfg->engine_id; + opts.engine_id = *nf_cfg->engine_id; } opts.active_timeout = nf_cfg->active_timeout; @@ -731,6 +765,52 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) ofproto_set_netflow(br->ofproto, NULL); } + /* Set sFlow configuration on this bridge. */ + if (br->cfg->sflow) { + const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow; + struct ovsrec_controller **controllers; + struct ofproto_sflow_options oso; + size_t n_controllers; + size_t i; + + memset(&oso, 0, sizeof oso); + + oso.targets.n = sflow_cfg->n_targets; + oso.targets.names = sflow_cfg->targets; + + oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE; + if (sflow_cfg->sampling) { + oso.sampling_rate = *sflow_cfg->sampling; + } + + oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL; + if (sflow_cfg->polling) { + oso.polling_interval = *sflow_cfg->polling; + } + + oso.header_len = SFL_DEFAULT_HEADER_SIZE; + if (sflow_cfg->header) { + oso.header_len = *sflow_cfg->header; + } + + oso.sub_id = sflow_bridge_number++; + oso.agent_device = sflow_cfg->agent; + + oso.control_ip = NULL; + n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers); + for (i = 0; i < n_controllers; i++) { + if (controllers[i]->local_ip) { + oso.control_ip = controllers[i]->local_ip; + break; + } + } + ofproto_set_sflow(br->ofproto, &oso); + + /* Do not destroy oso.targets because it is owned by sflow_cfg. */ + } else { + ofproto_set_sflow(br->ofproto, NULL); + } + /* Update the controller and related settings. It would be more * straightforward to call this from bridge_reconfigure_one(), but we * can't do it there for two reasons. First, and most importantly, at @@ -740,39 +820,50 @@ 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); + bridge_reconfigure_remotes(ovs_cfg, 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 * -bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key) +get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n) { size_t i; - for (i = 0; i < br_cfg->n_other_config; i++) { - if (!strcmp(br_cfg->key_other_config[i], key)) { - return br_cfg->value_other_config[i]; + for (i = 0; i < n; i++) { + if (!strcmp(keys[i], key)) { + return values[i]; } } return NULL; } +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); +} + static void bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], struct iface **hw_addr_iface) @@ -966,25 +1057,115 @@ 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; + bool datapath_destroyed; + struct bridge *br; - retval = 0; - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { + /* 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. */ + if (ovsdb_idl_run(idl) || datapath_destroyed) { + const struct ovsrec_open_vswitch *cfg = ovsrec_open_vswitch_first(idl); + 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); + } + } + + /* 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 @@ -994,13 +1175,15 @@ bridge_wait(void) LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { ofproto_wait(br->ofproto); - if (br->controller) { + if (ofproto_has_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 @@ -1036,7 +1219,7 @@ bridge_get_local_iface(struct bridge *br) /* Bridge unixctl user interface functions. */ static void bridge_unixctl_fdb_show(struct unixctl_conn *conn, - const char *args, void *aux UNUSED) + const char *args, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct bridge *br; @@ -1062,41 +1245,45 @@ bridge_unixctl_fdb_show(struct unixctl_conn *conn, } /* Bridge reconfiguration functions. */ - static struct bridge * -bridge_create(const char *name) +bridge_create(const struct ovsrec_bridge *br_cfg) { struct bridge *br; int error; - assert(!bridge_lookup(name)); + assert(!bridge_lookup(br_cfg->name)); br = xzalloc(sizeof *br); - error = dpif_create_and_open(name, &br->dpif); + error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type, + &br->dpif); if (error) { free(br); return NULL; } dpif_flow_flush(br->dpif); - error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto); + 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", name, strerror(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(name); + br->name = xstrdup(br_cfg->name); + br->cfg = br_cfg; br->ml = mac_learning_create(); - br->sent_config_request = false; - eth_addr_random(br->default_ea); + 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; - br->bond_next_rebalance = time_msec() + 10000; list_push_back(&all_bridges, &br->node); @@ -1122,9 +1309,10 @@ bridge_destroy(struct bridge *br) } dpif_close(br->dpif); ofproto_destroy(br->ofproto); - free(br->controller); 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); @@ -1144,24 +1332,11 @@ 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 bridge_unixctl_dump_flows(struct unixctl_conn *conn, - const char *args, void *aux UNUSED) + const char *args, void *aux OVS_UNUSED) { struct bridge *br; struct ds results; @@ -1179,6 +1354,29 @@ 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) { @@ -1198,37 +1396,31 @@ bridge_run_one(struct bridge *br) return error; } -static const struct ovsrec_controller * -bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg, - const struct bridge *br) +static size_t +bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg, + const struct bridge *br, + struct ovsrec_controller ***controllersp) { - const struct ovsrec_controller *controller; + struct ovsrec_controller **controllers; + size_t n_controllers; - controller = (br->cfg->controller ? br->cfg->controller - : ovs_cfg->controller ? ovs_cfg->controller - : NULL); - - if (controller && !strcmp(controller->target, "none")) { - return NULL; + 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; } - return controller; -} + if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) { + controllers = NULL; + n_controllers = 0; + } -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; + if (controllersp) { + *controllersp = controllers; } + return n_controllers; } static void @@ -1236,11 +1428,9 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, 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; - uint64_t mgmt_id; size_t i; /* Collect old ports. */ @@ -1263,7 +1453,7 @@ 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_controller(ovs_cfg, br)) { + if (bridge_get_controllers(ovs_cfg, br, NULL)) { char local_name[IF_NAMESIZE]; int error; @@ -1276,78 +1466,44 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, } } - dpid_from_string(ovs_cfg->management_id, &mgmt_id); - ofproto_set_mgmt_id(br->ofproto, mgmt_id); - - /* 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); - /* 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. */ + /* Configure OpenFlow management listener. */ svec_init(&listeners); svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name)); @@ -1359,7 +1515,7 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, svec_destroy(&listeners); svec_destroy(&old_listeners); - /* Default snoop. */ + /* Configure OpenFlow controller connection snooping. */ svec_init(&snoops); svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", ovs_rundir, br->name)); @@ -1370,96 +1526,33 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, } svec_destroy(&snoops); svec_destroy(&old_snoops); -#endif -#if 0 mirror_reconfigure(br); -#endif } static void -bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg, - struct bridge *br) +bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg, + struct bridge *br, + const struct sockaddr_in *managers, + size_t n_managers) { - char *pfx = xasprintf("bridge.%s.controller", br->name); - const struct ovsrec_controller *c; + struct ovsrec_controller **controllers; + size_t n_controllers; - c = bridge_get_controller(ovs_cfg, br); - if ((br->controller != NULL) != (c != NULL)) { + ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); + + n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers); + if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) { ofproto_flush_flows(br->ofproto); } - free(br->controller); - br->controller = c ? xstrdup(c->target) : NULL; - - if (c) { - int max_backoff, probe; - int rate_limit, burst_limit; - if (!strcmp(c->target, "discover")) { - ofproto_set_discovery(br->ofproto, true, - c->discover_accept_regex, - c->discover_update_resolv_conf); - } else { - struct iface *local_iface; - struct in_addr ip; - bool in_band; - - in_band = (!c->connection_mode - || !strcmp(c->connection_mode, "out-of-band")); - ofproto_set_discovery(br->ofproto, false, NULL, NULL); - ofproto_set_in_band(br->ofproto, in_band); - - 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 ip, 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)); - } - } - } - } - - ofproto_set_failure(br->ofproto, - (!c->fail_mode - || !strcmp(c->fail_mode, "standalone") - || !strcmp(c->fail_mode, "open"))); - - probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; - ofproto_set_probe_interval(br->ofproto, probe); - - max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; - ofproto_set_max_backoff(br->ofproto, max_backoff); - - rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; - burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0; - ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit); - } else { + if (!n_controllers) { 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); @@ -1467,17 +1560,78 @@ bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg, action.output.len = htons(sizeof action); action.output.port = htons(OFPP_NORMAL); memset(&flow, 0, sizeof flow); - ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0, - &action, 1, 0); + ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0); + } else { + struct ofproto_controller *ocs; + size_t i; - ofproto_set_in_band(br->ofproto, false); - ofproto_set_max_backoff(br->ofproto, 1); - ofproto_set_probe_interval(br->ofproto, 5); - ofproto_set_failure(br->ofproto, false); - } - free(pfx); + 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)); + } + } + } + } - ofproto_set_controller(br->ofproto, br->controller); + 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); + } } static void @@ -1726,6 +1880,43 @@ bond_enable_slave(struct iface *iface, bool enable) port->bond_compat_is_stale = true; } +/* 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) +{ + 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); + } +} + static void bond_run(struct bridge *br) { @@ -1741,6 +1932,12 @@ bond_run(struct bridge *br) 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) { @@ -1763,9 +1960,12 @@ 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_until(port->bond_next_fake_iface_update); + } } } @@ -1848,7 +2048,8 @@ 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 @@ -1934,7 +2135,7 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, return dst - dsts; } -static void UNUSED +static void OVS_UNUSED print_dsts(const struct dst *dsts, size_t n) { for (; n--; dsts++) { @@ -2022,12 +2223,34 @@ 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. */ @@ -2041,33 +2264,39 @@ update_learning_table(struct bridge *br, const flow_t *flow, int vlan, } } +/* 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. + * + * If 'have_packet' is true, it indicates that the caller is processing a + * received packet. If 'have_packet' is false, then the caller is just + * revalidating an existing flow because configuration has changed. Either + * way, 'have_packet' only affects logging (there is no point in logging errors + * during revalidation). + * + * Sets '*in_portp' to the input port. This will be a null pointer if + * flow->in_port does not designate a known input port (in which case + * is_admissible() returns false). + * + * When returning true, sets '*vlanp' to the effective VLAN of the input + * packet, as returned by flow_get_vlan(). + * + * May also add tags to '*tags', although the current implementation only does + * so in one special case. + */ 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)); -} - -/* If the composed actions may be applied to any packet in the given 'flow', - * returns true. Otherwise, the actions should only be applied to 'packet', or - * not at all, if 'packet' was NULL. */ -static bool -process_flow(struct bridge *br, const flow_t *flow, - const struct ofpbuf *packet, struct odp_actions *actions, - tag_type *tags, uint16_t *nf_output_iface) +is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, + tag_type *tags, int *vlanp, struct port **in_portp) { struct iface *in_iface; struct port *in_port; - struct port *out_port = NULL; /* By default, drop the packet/flow. */ int vlan; - int out_port_idx; /* Find the interface and port structure for the received packet. */ in_iface = iface_from_dp_ifidx(br, flow->in_port); if (!in_iface) { /* No interface? Something fishy... */ - if (packet != NULL) { + if (have_packet) { /* Odd. A few possible reasons here: * * - We deleted an interface but there are still a few packets @@ -2085,62 +2314,89 @@ process_flow(struct bridge *br, const flow_t *flow, "interface %"PRIu16, br->name, flow->in_port); } - /* Return without adding any actions, to drop packets on this flow. */ - return true; + *in_portp = NULL; + return false; } - in_port = in_iface->port; - vlan = flow_get_vlan(br, flow, in_port, !!packet); + *in_portp = in_port = in_iface->port; + *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet); if (vlan < 0) { - goto done; + return false; } /* Drop frames for reserved multicast addresses. */ if (eth_addr_is_reserved(flow->dl_dst)) { - goto done; + return false; } /* Drop frames on ports reserved for mirroring. */ if (in_port->is_mirror_output_port) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, " - "which is reserved exclusively for mirroring", - br->name, in_port->name); - goto done; + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " + "%s, which is reserved exclusively for mirroring", + br->name, in_port->name); + } + 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 (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. */ - goto done; + 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. 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)) { - goto done; + (!is_gratuitous_arp(flow) || is_grat_arp_locked)) { + return false; } } - /* MAC learning. */ - out_port = FLOOD_PORT; + return true; +} + +/* If the composed actions may be applied to any packet in the given 'flow', + * returns true. Otherwise, the actions should only be applied to 'packet', or + * not at all, if 'packet' was NULL. */ +static bool +process_flow(struct bridge *br, const flow_t *flow, + const struct ofpbuf *packet, struct odp_actions *actions, + tag_type *tags, uint16_t *nf_output_iface) +{ + struct port *in_port; + struct port *out_port; + 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)) { + out_port = NULL; + 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); + 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)) { @@ -2150,6 +2406,8 @@ process_flow(struct bridge *br, const flow_t *flow, * 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. */ @@ -2158,8 +2416,10 @@ process_flow(struct bridge *br, const flow_t *flow, } done: - compose_actions(br, flow, vlan, in_port, out_port, tags, actions, - nf_output_iface); + if (in_port) { + compose_actions(br, flow, vlan, in_port, out_port, tags, actions, + nf_output_iface); + } return true; } @@ -2219,18 +2479,16 @@ bridge_account_flow_ofhook_cb(const flow_t *flow, void *br_) { struct bridge *br = br_; - struct port *in_port; const union odp_action *a; + struct port *in_port; + tag_type tags = 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. */ - in_port = port_from_dp_ifidx(br, flow->in_port); - if (in_port) { - int vlan = flow_get_vlan(br, flow, in_port, false); - if (vlan >= 0) { - update_learning_table(br, flow, vlan, in_port); - } + if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) { + update_learning_table(br, flow, vlan, in_port); } if (!br->has_bonded_ports) { @@ -2253,22 +2511,18 @@ 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; } - /* The current ofproto implementation calls this callback at least once a - * second, so this timer implementation is sufficient. */ - if (time_msec() < br->bond_next_rebalance) { - return; - } - br->bond_next_rebalance = time_msec() + 10000; - + now = time_msec(); for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - if (port->n_ifaces > 1) { + if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) { + port->bond_next_rebalance = now + port->bond_rebalance_interval; bond_rebalance_port(port); } } @@ -2624,7 +2878,7 @@ bond_send_learning_packets(struct port *port) n_packets++; compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, e->mac); - flow_extract(&packet, ODPP_NONE, &flow); + flow_extract(&packet, 0, ODPP_NONE, &flow); retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, &packet); if (retval) { @@ -2649,7 +2903,7 @@ bond_send_learning_packets(struct port *port) static void bond_unixctl_list(struct unixctl_conn *conn, - const char *args UNUSED, void *aux UNUSED) + const char *args OVS_UNUSED, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct bridge *br; @@ -2700,7 +2954,7 @@ bond_find(const char *name) static void bond_unixctl_show(struct unixctl_conn *conn, - const char *args, void *aux UNUSED) + const char *args, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct port *port; @@ -2715,7 +2969,7 @@ bond_unixctl_show(struct unixctl_conn *conn, ds_put_format(&ds, "updelay: %d ms\n", port->updelay); ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); ds_put_format(&ds, "next rebalance: %lld ms\n", - port->bridge->bond_next_rebalance - time_msec()); + 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; @@ -2766,7 +3020,7 @@ bond_unixctl_show(struct unixctl_conn *conn, static void bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, - void *aux UNUSED) + void *aux OVS_UNUSED) { char *args = (char *) args_; char *save_ptr = NULL; @@ -2823,7 +3077,7 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, static void bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, - void *aux UNUSED) + void *aux OVS_UNUSED) { char *args = (char *) args_; char *save_ptr = NULL; @@ -2904,21 +3158,21 @@ enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) static void bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, - void *aux UNUSED) + void *aux OVS_UNUSED) { enable_slave(conn, args, true); } static void bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, - void *aux UNUSED) + void *aux OVS_UNUSED) { enable_slave(conn, args, false); } static void bond_unixctl_hash(struct unixctl_conn *conn, const char *args, - void *aux UNUSED) + void *aux OVS_UNUSED) { uint8_t mac[ETH_ADDR_LEN]; uint8_t hash; @@ -2971,6 +3225,7 @@ 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); @@ -2978,30 +3233,54 @@ port_create(struct bridge *br, const char *name) return port; } +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); + return value ? value : default_value; +} + static void -port_reconfigure(struct port *port, const struct ovsrec_port *cfg) +port_del_ifaces(struct port *port, const struct ovsrec_port *cfg) { - struct shash old_ifaces, new_ifaces; - struct shash_node *node; - unsigned long *trunks; - int vlan; + struct shash new_ifaces; size_t i; - port->cfg = cfg; - - /* Collect old and new interfaces. */ - shash_init(&old_ifaces); + /* Collect list of new interfaces. */ 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); + 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 new_ifaces; + long long int next_rebalance; + 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; @@ -3010,23 +3289,42 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) if (port->downdelay < 0) { port->downdelay = 0; } - - /* 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); - } + 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; } - SHASH_FOR_EACH (node, &new_ifaces) { - const struct ovsrec_interface *if_cfg = node->data; + 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); + 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 = iface_create(port, if_cfg); + 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); } - iface->cfg = if_cfg; } + shash_destroy(&new_ifaces); /* Get VLAN tag. */ vlan = -1; @@ -3052,7 +3350,7 @@ 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; @@ -3071,17 +3369,14 @@ 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 @@ -3090,9 +3385,6 @@ 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 @@ -3101,23 +3393,24 @@ port_destroy(struct port *port) if (port) { struct bridge *br = port->bridge; struct port *del; + int i; proc_net_compat_update_vlan(port->name, NULL, 0); proc_net_compat_update_bond(port->name, NULL); -#if 0 for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; if (m && m->out_port == port) { mirror_destroy(m); } } -#endif while (port->n_ifaces > 0) { 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; @@ -3139,29 +3432,14 @@ 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 @@ -3187,6 +3465,12 @@ port_update_bonding(struct port *port) } 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; @@ -3252,7 +3536,7 @@ port_update_bond_compat(struct port *port) if (port->bond_fake_iface) { struct netdev *bond_netdev; - if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) { + if (!netdev_open_default(port->name, &bond_netdev)) { if (bond.up) { netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); } else { @@ -3308,6 +3592,7 @@ 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; @@ -3320,6 +3605,23 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) 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); + + /* 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, @@ -3327,20 +3629,12 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) } 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. */ - error = set_up_iface(if_cfg, true); - if (error) { - VLOG_WARN("could not create iface %s: %s\n", 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; } @@ -3354,6 +3648,8 @@ 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); } @@ -3369,7 +3665,6 @@ iface_destroy(struct iface *iface) bond_send_learning_packets(port); } - netdev_destroy(iface->name); free(iface->name); free(iface); @@ -3380,18 +3675,7 @@ 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 * @@ -3413,7 +3697,6 @@ 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; @@ -3456,53 +3739,137 @@ 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 int64_t *queue_ids; + size_t n_queue_ids; +}; + +static bool +queue_ids_include(const int64_t *ids, size_t n, int64_t target) +{ + size_t low = 0; + size_t high = n; + + while (low < high) { + size_t mid = low + (high - low) / 2; + if (target > ids[mid]) { + high = mid; + } else if (target < ids[mid]) { + low = mid + 1; + } else { + return true; + } + } + return false; +} + +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->queue_ids, cbdata->n_queue_ids, 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.queue_ids = qos->key_queues; + cbdata.n_queue_ids = qos->n_queues; + 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. */ -#if 0 static void -mirror_reconfigure(struct bridge *br UNUSED) +mirror_reconfigure(struct bridge *br) { - struct svec old_mirrors, new_mirrors; - size_t i, n_rspan_vlans; + struct shash old_mirrors, new_mirrors; + struct shash_node *node; unsigned long *rspan_vlans; + int i; - /* Collect old and new mirrors. */ - svec_init(&old_mirrors); - svec_init(&new_mirrors); - cfg_get_subsections(&new_mirrors, "mirror.%s", br->name); + /* Collect old mirrors. */ + shash_init(&old_mirrors); for (i = 0; i < MAX_MIRRORS; i++) { if (br->mirrors[i]) { - svec_add(&old_mirrors, br->mirrors[i]->name); + shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]); } } - /* Get rid of deleted mirrors and add new mirrors. */ - svec_sort(&old_mirrors); - assert(svec_is_unique(&old_mirrors)); - svec_sort(&new_mirrors); - assert(svec_is_unique(&new_mirrors)); - for (i = 0; i < MAX_MIRRORS; i++) { - struct mirror *m = br->mirrors[i]; - if (m && !svec_contains(&new_mirrors, m->name)) { - mirror_destroy(m); + /* Collect new mirrors. */ + shash_init(&new_mirrors); + 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); } } - for (i = 0; i < new_mirrors.n; i++) { - const char *name = new_mirrors.names[i]; - if (!svec_contains(&old_mirrors, name)) { - mirror_create(br, 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); } } - svec_destroy(&old_mirrors); - svec_destroy(&new_mirrors); - - /* Reconfigure all mirrors. */ - for (i = 0; i < MAX_MIRRORS; i++) { - if (br->mirrors[i]) { - mirror_reconfigure_one(br->mirrors[i]); + 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; + } } + 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++) { @@ -3515,31 +3882,29 @@ mirror_reconfigure(struct bridge *br UNUSED) } } - /* Update learning disabled vlans (for RSPAN). */ + /* Update flooded vlans (for RSPAN). */ rspan_vlans = NULL; - n_rspan_vlans = cfg_count("vlan.%s.disable-learning", br->name); - if (n_rspan_vlans) { + if (br->cfg->n_flood_vlans) { rspan_vlans = bitmap_allocate(4096); - for (i = 0; i < n_rspan_vlans; i++) { - int vlan = cfg_get_vlan(i, "vlan.%s.disable-learning", br->name); - if (vlan >= 0) { + for (i = 0; i < br->cfg->n_flood_vlans; i++) { + int64_t vlan = br->cfg->flood_vlans[i]; + if (vlan >= 0 && vlan < 4096) { bitmap_set1(rspan_vlans, vlan); - VLOG_INFO("bridge %s: disabling learning on vlan %d\n", + VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64, br->name, vlan); } else { - VLOG_ERR("bridge %s: invalid value '%s' for learning disabled " - "VLAN", br->name, - cfg_get_string(i, "vlan.%s.disable-learning", br->name)); + VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN", + br->name, vlan); } } } - if (mac_learning_set_disabled_vlans(br->ml, rspan_vlans)) { + if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) { bridge_flush(br); } } -static void +static struct mirror * mirror_create(struct bridge *br, const char *name) { struct mirror *m; @@ -3549,7 +3914,7 @@ mirror_create(struct bridge *br, const char *name) if (i >= MAX_MIRRORS) { VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " "cannot create %s", br->name, MAX_MIRRORS, name); - return; + return NULL; } if (!br->mirrors[i]) { break; @@ -3563,12 +3928,14 @@ mirror_create(struct bridge *br, const char *name) m->bridge = br; m->idx = i; m->name = xstrdup(name); - svec_init(&m->src_ports); - svec_init(&m->dst_ports); + shash_init(&m->src_ports); + shash_init(&m->dst_ports); m->vlans = NULL; m->n_vlans = 0; m->out_vlan = -1; m->out_port = NULL; + + return m; } static void @@ -3583,8 +3950,8 @@ mirror_destroy(struct mirror *m) br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); } - svec_destroy(&m->src_ports); - svec_destroy(&m->dst_ports); + shash_destroy(&m->src_ports); + shash_destroy(&m->dst_ports); free(m->vlans); m->bridge->mirrors[m->idx] = NULL; @@ -3595,45 +3962,36 @@ mirror_destroy(struct mirror *m) } static void -prune_ports(struct mirror *m, struct svec *ports) +mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports, + struct shash *names) { - struct svec tmp; size_t i; - svec_sort_unique(ports); - - svec_init(&tmp); - for (i = 0; i < ports->n; i++) { - const char *name = ports->names[i]; + for (i = 0; i < n_ports; i++) { + const char *name = ports[i]->name; if (port_lookup(m->bridge, name)) { - svec_add(&tmp, name); + shash_add_once(names, name, NULL); } else { - VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s", - m->bridge->name, m->name, name); + VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent " + "port %s", m->bridge->name, m->name, name); } } - svec_swap(ports, &tmp); - svec_destroy(&tmp); } static size_t -prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans) +mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg, + int **vlans) { - size_t n_vlans, i; - - /* This isn't perfect: it won't combine "0" and "00", and the textual sort - * order won't give us numeric sort order. But that's good enough for what - * we need right now. */ - svec_sort_unique(vlan_strings); + size_t n_vlans; + size_t i; - *vlans = xmalloc(sizeof *vlans * vlan_strings->n); + *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan); n_vlans = 0; - for (i = 0; i < vlan_strings->n; i++) { - const char *name = vlan_strings->names[i]; - int vlan; - if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) { - VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s", - m->bridge->name, m->name, name); + for (i = 0; i < cfg->n_select_vlan; i++) { + int64_t vlan = cfg->select_vlan[i]; + if (vlan < 0 || vlan > 4095) { + VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64, + m->bridge->name, m->name, vlan); } else { (*vlans)[n_vlans++] = vlan; } @@ -3668,104 +4026,86 @@ port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) } static void -mirror_reconfigure_one(struct mirror *m UNUSED) +mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) { - char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name); - struct svec src_ports, dst_ports, ports; - struct svec vlan_strings; + struct shash src_ports, dst_ports; mirror_mask_t mirror_bit; - const char *out_port_name; struct port *out_port; int out_vlan; size_t n_vlans; int *vlans; size_t i; - bool mirror_all_ports; - bool any_ports_specified; /* Get output port. */ - out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port", - m->bridge->name, m->name); - if (out_port_name) { - out_port = port_lookup(m->bridge, out_port_name); + if (cfg->output_port) { + out_port = port_lookup(m->bridge, cfg->output_port->name); if (!out_port) { - VLOG_ERR("%s.output.port: bridge %s does not have a port " - "named %s", pfx, m->bridge->name, out_port_name); + VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge", + m->bridge->name, m->name); mirror_destroy(m); - free(pfx); return; } out_vlan = -1; - if (cfg_has("%s.output.vlan", pfx)) { - VLOG_ERR("%s.output.port and %s.output.vlan both specified; " - "ignoring %s.output.vlan", pfx, pfx, pfx); + if (cfg->output_vlan) { + VLOG_ERR("bridge %s: mirror %s specifies both output port and " + "output vlan; ignoring output vlan", + m->bridge->name, m->name); } - } else if (cfg_has("%s.output.vlan", pfx)) { + } else if (cfg->output_vlan) { out_port = NULL; - out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx); + out_vlan = *cfg->output_vlan; } else { - VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, " - "but exactly one is required; disabling port mirror %s", - pfx, pfx, pfx, pfx); + VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring", + m->bridge->name, m->name); mirror_destroy(m); - free(pfx); return; } - /* Get all the ports, and drop duplicates and ports that don't exist. */ - svec_init(&src_ports); - svec_init(&dst_ports); - svec_init(&ports); - cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx); - cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx); - cfg_get_all_keys(&ports, "%s.select.port", pfx); - any_ports_specified = src_ports.n || dst_ports.n || ports.n; - svec_append(&src_ports, &ports); - svec_append(&dst_ports, &ports); - svec_destroy(&ports); - prune_ports(m, &src_ports); - prune_ports(m, &dst_ports); - if (any_ports_specified && !src_ports.n && !dst_ports.n) { - VLOG_ERR("%s: none of the specified ports exist; " - "disabling port mirror %s", pfx, pfx); - mirror_destroy(m); - goto exit; - } + shash_init(&src_ports); + shash_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); + } + vlans = NULL; + n_vlans = 0; + } else { + /* Get ports, and drop duplicates and ports that don't exist. */ + mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port, + &src_ports); + mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port, + &dst_ports); - /* Get all the vlans, and drop duplicate and invalid vlans. */ - svec_init(&vlan_strings); - cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx); - n_vlans = prune_vlans(m, &vlan_strings, &vlans); - svec_destroy(&vlan_strings); + /* Get all the vlans, and drop duplicate and invalid vlans. */ + n_vlans = mirror_collect_vlans(m, cfg, &vlans); + } /* Update mirror data. */ - if (!svec_equal(&m->src_ports, &src_ports) - || !svec_equal(&m->dst_ports, &dst_ports) + if (!shash_equal_keys(&m->src_ports, &src_ports) + || !shash_equal_keys(&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); } - svec_swap(&m->src_ports, &src_ports); - svec_swap(&m->dst_ports, &dst_ports); + shash_swap(&m->src_ports, &src_ports); + shash_swap(&m->dst_ports, &dst_ports); free(m->vlans); m->vlans = vlans; m->n_vlans = n_vlans; m->out_port = out_port; m->out_vlan = out_vlan; - /* If no selection criteria have been given, mirror for all ports. */ - mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans); - /* 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 (mirror_all_ports - || svec_contains(&m->src_ports, port->name) + if (shash_find(&m->src_ports, port->name) || (m->n_vlans && (!port->vlan ? port_trunks_any_mirrored_vlan(m, port) @@ -3775,7 +4115,7 @@ mirror_reconfigure_one(struct mirror *m UNUSED) port->src_mirrors &= ~mirror_bit; } - if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) { + if (shash_find(&m->dst_ports, port->name)) { port->dst_mirrors |= mirror_bit; } else { port->dst_mirrors &= ~mirror_bit; @@ -3783,9 +4123,6 @@ mirror_reconfigure_one(struct mirror *m UNUSED) } /* Clean up. */ -exit: - svec_destroy(&src_ports); - svec_destroy(&dst_ports); - free(pfx); + shash_destroy(&src_ports); + shash_destroy(&dst_ports); } -#endif