X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=931f987c49624a7141cbe4c31c0c91bb3e2242a7;hb=1e82e503c5358f8dce9eb2105448f0ec894d57bc;hp=7ce1f25fcec131c460b4ab22801a01dec91cdc4b;hpb=3a55ef14838f9552fa824b90599aa2a6493bb04a;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 7ce1f25fc..931f987c4 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -1,4 +1,4 @@ -/* Copyright (c) 2008, 2009 Nicira Networks +/* Copyright (c) 2008, 2009, 2010 Nicira Networks * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -20,6 +20,7 @@ #include #include #include +#include #include #include #include @@ -30,35 +31,39 @@ #include #include #include "bitmap.h" -#include "cfg.h" #include "coverage.h" #include "dirs.h" #include "dpif.h" #include "dynamic-string.h" #include "flow.h" #include "hash.h" +#include "jsonrpc.h" #include "list.h" #include "mac-learning.h" #include "netdev.h" #include "odp-util.h" #include "ofp-print.h" #include "ofpbuf.h" +#include "ofproto/netflow.h" +#include "ofproto/ofproto.h" #include "packets.h" #include "poll-loop.h" #include "port-array.h" #include "proc-net-compat.h" #include "process.h" -#include "secchan/ofproto.h" +#include "sha1.h" +#include "shash.h" #include "socket-util.h" -#include "stp.h" +#include "stream-ssl.h" #include "svec.h" #include "timeval.h" #include "util.h" #include "unixctl.h" #include "vconn.h" -#include "vconn-ssl.h" +#include "vswitchd/vswitch-idl.h" #include "xenserver.h" #include "xtoxll.h" +#include "sflow_api.h" #define THIS_MODULE VLM_bridge #include "vlog.h" @@ -68,20 +73,22 @@ struct dst { uint16_t dp_ifidx; }; -extern uint64_t mgmt_id; - struct iface { + /* These members are always valid. */ struct port *port; /* Containing port. */ size_t port_ifidx; /* Index within containing port. */ - char *name; /* Host network device name. */ - int dp_ifidx; /* Index within kernel datapath. */ - - uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */ - tag_type tag; /* Tag associated with this interface. */ - bool enabled; /* May be chosen for flows? */ long long delay_expires; /* Time after which 'enabled' may change. */ + + /* These members are valid only after bridge_reconfigure() causes them to + * 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; }; #define BOND_MASK 0xff @@ -101,8 +108,8 @@ 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; @@ -116,7 +123,8 @@ 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. */ char *name; /* An ordinary bridge port has 1 interface. @@ -131,44 +139,50 @@ struct port { tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */ int updelay, downdelay; /* Delay before iface goes up/down, in ms. */ bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */ + bool bond_fake_iface; /* Fake a bond interface for legacy compat? */ + long 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? */ - /* Spanning tree info. */ - enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */ - tag_type stp_state_tag; /* Tag for STP state change. */ + /* This member is only valid *during* bridge_reconfigure(). */ + const struct ovsrec_port *cfg; }; #define DP_MAX_PORTS 255 struct bridge { struct list node; /* Node in global list of bridges. */ char *name; /* User-specified arbitrary name. */ - struct mac_learning *ml; /* MAC learning table, or null not to learn. */ + 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. */ - /* 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; /* Kernel datapath. */ + struct dpif *dpif; /* Datapath. */ struct port_array ifaces; /* Indexed by kernel datapath port number. */ /* 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; @@ -179,9 +193,8 @@ struct bridge { /* Port mirroring. */ struct mirror *mirrors[MAX_MIRRORS]; - /* Spanning tree. */ - struct stp *stp; - long long int stp_last_tick; + /* This member is only valid *during* bridge_reconfigure(). */ + const struct ovsrec_bridge *cfg; }; /* List of all bridges. */ @@ -190,34 +203,44 @@ static struct list all_bridges = LIST_INITIALIZER(&all_bridges); /* Maximum number of datapaths. */ enum { DP_MAX = 256 }; -static struct bridge *bridge_create(const char *name); +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 void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *); +static unixctl_cb_func bridge_unixctl_dump_flows; static int bridge_run_one(struct bridge *); -static void bridge_reconfigure_one(struct bridge *); -static void bridge_reconfigure_controller(struct bridge *); -static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces); +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_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 *); static void bridge_pick_local_hw_addr(struct bridge *, uint8_t ea[ETH_ADDR_LEN], - const char **devname); + struct iface **hw_addr_iface); static uint64_t bridge_pick_datapath_id(struct bridge *, const uint8_t bridge_ea[ETH_ADDR_LEN], - const char *devname); + struct iface *hw_addr_iface); +static struct iface *bridge_get_local_iface(struct bridge *); static uint64_t dpid_from_hash(const void *, size_t nbytes); -static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args); +static unixctl_cb_func bridge_unixctl_fdb_show; static void bond_init(void); static void bond_run(struct bridge *); static void bond_wait(struct bridge *); static void bond_rebalance_port(struct port *); static void bond_send_learning_packets(struct port *); +static void bond_enable_slave(struct iface *iface, bool enable); -static void port_create(struct bridge *, const char *name); -static void port_reconfigure(struct port *); +static struct port *port_create(struct bridge *, const char *name); +static void port_reconfigure(struct port *, const struct ovsrec_port *); +static void port_del_ifaces(struct port *, const struct ovsrec_port *); static void port_destroy(struct port *); static struct port *port_lookup(const struct bridge *, const char *name); static struct iface *port_lookup_iface(const struct port *, const char *name); @@ -225,19 +248,16 @@ static struct port *port_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); static void port_update_bond_compat(struct port *); static void port_update_vlan_compat(struct port *); +static void port_update_bonding(struct port *); -static void 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); -static void brstp_reconfigure(struct bridge *); -static void brstp_adjust_timers(struct bridge *); -static void brstp_run(struct bridge *); -static void brstp_wait(struct bridge *); - -static void iface_create(struct port *, const char *name); +static struct iface *iface_create(struct port *port, + const struct ovsrec_interface *if_cfg); static void iface_destroy(struct iface *); static struct iface *iface_lookup(const struct bridge *, const char *name); static struct iface *iface_from_dp_ifidx(const struct bridge *, @@ -265,8 +285,8 @@ bridge_get_ifaces(struct svec *svec) 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 dp%u, ignoring", - iface->name, dpif_id(&br->dpif)); + 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); @@ -277,144 +297,330 @@ bridge_get_ifaces(struct svec *svec) } } -/* The caller must already have called cfg_read(). */ void -bridge_init(void) +bridge_init(const struct ovsrec_open_vswitch *cfg) { - int retval; - int i; + struct svec bridge_names; + struct svec dpif_names, dpif_types; + size_t i; - bond_init(); + unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + + svec_init(&bridge_names); + for (i = 0; i < cfg->n_bridges; i++) { + svec_add(&bridge_names, cfg->bridges[i]->name); + } + svec_sort(&bridge_names); - unixctl_command_register("fdb/show", bridge_unixctl_fdb_show); + svec_init(&dpif_names); + 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; - for (i = 0; i < DP_MAX; i++) { - struct dpif dpif; - char devname[16]; + dp_enumerate_names(dpif_types.names[i], &dpif_names); - sprintf(devname, "dp%d", i); - retval = dpif_open(devname, &dpif); - if (!retval) { - char dpif_name[IF_NAMESIZE]; - if (dpif_get_name(&dpif, dpif_name, sizeof dpif_name) - || !cfg_has("bridge.%s.port", dpif_name)) { - dpif_delete(&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; + + 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; + } + } + dpif_delete(dpif); + found: + svec_destroy(&all_names); + dpif_close(dpif); } - dpif_close(&dpif); - } else if (retval != ENODEV) { - VLOG_ERR("failed to delete datapath dp%d: %s", - i, strerror(retval)); } } + svec_destroy(&bridge_names); + svec_destroy(&dpif_names); + svec_destroy(&dpif_types); - unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows); + unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, + NULL); - bridge_reconfigure(); + 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. */ +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; + + shash_init(&options); + for (i = 0; i < iface_cfg->n_options; i++) { + shash_add(&options, iface_cfg->key_options[i], + xstrdup(iface_cfg->value_options[i])); + } + + if (create) { + 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"; + } + + 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_FOR_EACH (node, &options) { + free(node->data); + } + shash_destroy(&options); + + return error; +} + +static int +reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface) +{ + return set_up_iface(iface_cfg, iface, false); +} + +static bool +check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface, + void *aux OVS_UNUSED) +{ + 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 -config_string_change(const char *key, char **valuep) +check_iface_dp_ifidx(struct bridge *br, struct iface *iface, + void *aux OVS_UNUSED) { - const char *value = cfg_get_string(0, "%s", key); - if (value && (!*valuep || strcmp(value, *valuep))) { - free(*valuep); - *valuep = xstrdup(value); + if (iface->dp_ifidx >= 0) { + VLOG_DBG("%s has interface %s on port %d", + dpif_name(br->dpif), + iface->name, iface->dp_ifidx); return true; } else { + VLOG_ERR("%s interface not in %s, dropping", + iface->name, dpif_name(br->dpif)); return false; } } -static void -bridge_configure_ssl(void) +static bool +set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, + void *aux OVS_UNUSED) { - /* 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; + /* Set policing attributes. */ + netdev_set_policing(iface->netdev, + iface->cfg->ingress_policing_rate, + iface->cfg->ingress_policing_burst); - if (config_string_change("ssl.private-key", &private_key_file)) { - vconn_ssl_set_private_key_file(private_key_file); + /* Set MAC address of internal interfaces other than the local + * interface. */ + if (iface->dp_ifidx != ODPP_LOCAL + && iface_is_internal(br, iface->name)) { + iface_set_mac(iface); } - if (config_string_change("ssl.certificate", &certificate_file)) { - vconn_ssl_set_certificate_file(certificate_file); + return true; +} + +/* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument. + * Deletes from 'br' all the interfaces for which 'cb' returns false, and then + * deletes from 'br' any ports that no longer have any interfaces. */ +static void +iterate_and_prune_ifaces(struct bridge *br, + bool (*cb)(struct bridge *, struct iface *, + void *aux), + void *aux) +{ + size_t i, j; + + for (i = 0; i < br->n_ports; ) { + struct port *port = br->ports[i]; + for (j = 0; j < port->n_ifaces; ) { + struct iface *iface = port->ifaces[j]; + if (cb(br, iface, aux)) { + j++; + } else { + iface_destroy(iface); + } + } + + if (port->n_ifaces) { + i++; + } else { + VLOG_ERR("%s port has no interfaces, dropping", port->name); + port_destroy(port); + } } +} + +/* 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]; - /* 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) - || (stat(cacert_file, &s) && errno == ENOENT)) { - vconn_ssl_set_ca_cert_file(cacert_file, - cfg_get_bool(0, "ssl.bootstrap-ca-cert")); + 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; } -#endif void -bridge_reconfigure(void) +bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) { - struct svec old_br, new_br, raw_new_br; + struct ovsdb_idl_txn *txn; + struct shash old_br, new_br; + struct shash_node *node; struct bridge *br, *next; - size_t i, j; + struct sockaddr_in *managers; + size_t n_managers; + size_t i; + int sflow_bridge_number; COVERAGE_INC(bridge_reconfigure); - /* Collect old bridges. */ - svec_init(&old_br); + 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); + shash_init(&new_br); LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - svec_add(&old_br, br->name); - } - - /* Collect new bridges. */ - svec_init(&raw_new_br); - cfg_get_subsections(&raw_new_br, "bridge"); - svec_init(&new_br); - for (i = 0; i < raw_new_br.n; i++) { - const char *name = raw_new_br.names[i]; - if ((!strncmp(name, "dp", 2) && isdigit(name[2])) || - (!strncmp(name, "nl:", 3) && isdigit(name[3]))) { - VLOG_ERR("%s is not a valid bridge name (bridges may not be " - "named \"dp\" or \"nl:\" followed by a digit)", name); - } else { - svec_add(&new_br, name); + shash_add(&old_br, br->name, br); + } + for (i = 0; i < ovs_cfg->n_bridges; i++) { + const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i]; + if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) { + VLOG_WARN("more than one bridge named %s", br_cfg->name); } } - svec_destroy(&raw_new_br); /* Get rid of deleted bridges and add new bridges. */ - svec_sort(&old_br); - svec_sort(&new_br); - assert(svec_is_unique(&old_br)); - assert(svec_is_unique(&new_br)); LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { - if (!svec_contains(&new_br, br->name)) { + struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name); + if (br_cfg) { + br->cfg = br_cfg; + } else { bridge_destroy(br); } } - for (i = 0; i < new_br.n; i++) { - const char *name = new_br.names[i]; - if (!svec_contains(&old_br, name)) { - bridge_create(name); + SHASH_FOR_EACH (node, &new_br) { + const char *br_name = node->name; + const struct ovsrec_bridge *br_cfg = node->data; + 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); } } - svec_destroy(&old_br); - svec_destroy(&new_br); + shash_destroy(&old_br); + shash_destroy(&new_br); #ifdef HAVE_OPENSSL /* Configure SSL. */ - bridge_configure_ssl(); + bridge_configure_ssl(ovs_cfg->ssl); #endif /* Reconfigure all bridges. */ LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - bridge_reconfigure_one(br); + bridge_reconfigure_one(ovs_cfg, br); } /* Add and delete ports on all datapaths. @@ -425,111 +631,94 @@ bridge_reconfigure(void) LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { struct odp_port *dpif_ports; size_t n_dpif_ports; - struct svec want_ifaces; + struct shash want_ifaces; - dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports); + dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); bridge_get_all_ifaces(br, &want_ifaces); for (i = 0; i < n_dpif_ports; i++) { const struct odp_port *p = &dpif_ports[i]; - if (!svec_contains(&want_ifaces, p->devname) + if (!shash_find(&want_ifaces, p->devname) && strcmp(p->devname, br->name)) { - int retval = dpif_port_del(&br->dpif, p->port); + int retval = dpif_port_del(br->dpif, p->port); if (retval) { - VLOG_ERR("failed to remove %s interface from dp%u: %s", - p->devname, dpif_id(&br->dpif), strerror(retval)); + VLOG_ERR("failed to remove %s interface from %s: %s", + p->devname, dpif_name(br->dpif), + strerror(retval)); } } } - svec_destroy(&want_ifaces); + shash_destroy(&want_ifaces); free(dpif_ports); } LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { struct odp_port *dpif_ports; size_t n_dpif_ports; - struct svec cur_ifaces, want_ifaces, add_ifaces; - int next_port_no; + struct shash cur_ifaces, want_ifaces; + struct shash_node *node; - dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports); - svec_init(&cur_ifaces); + /* Get the set of interfaces currently in this datapath. */ + dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); + shash_init(&cur_ifaces); for (i = 0; i < n_dpif_ports; i++) { - svec_add(&cur_ifaces, dpif_ports[i].devname); + const char *name = dpif_ports[i].devname; + if (!shash_find(&cur_ifaces, name)) { + shash_add(&cur_ifaces, name, NULL); + } } free(dpif_ports); - svec_sort_unique(&cur_ifaces); + + /* Get the set of interfaces we want on this datapath. */ bridge_get_all_ifaces(br, &want_ifaces); - svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL); - next_port_no = 1; - for (i = 0; i < add_ifaces.n; i++) { - const char *if_name = add_ifaces.names[i]; - for (;;) { + SHASH_FOR_EACH (node, &want_ifaces) { + const char *if_name = node->name; + struct iface *iface = node->data; + + if (shash_find(&cur_ifaces, if_name)) { + /* Already exists, just reconfigure it. */ + if (iface) { + reconfigure_iface(iface->cfg, iface); + } + } else { + /* Need to add to datapath. */ bool internal; int error; /* Add to datapath. */ internal = iface_is_internal(br, if_name); - error = dpif_port_add(&br->dpif, if_name, next_port_no++, - internal ? ODP_PORT_INTERNAL : 0); - if (error != EEXIST) { - if (next_port_no >= 256) { - VLOG_ERR("ran out of valid port numbers on dp%u", - dpif_id(&br->dpif)); - goto out; - } - if (error) { - VLOG_ERR("failed to add %s interface to dp%u: %s", - if_name, dpif_id(&br->dpif), strerror(error)); - } + error = dpif_port_add(br->dpif, if_name, + internal ? ODP_PORT_INTERNAL : 0, NULL); + if (error == EFBIG) { + VLOG_ERR("ran out of valid port numbers on %s", + dpif_name(br->dpif)); break; + } else if (error) { + VLOG_ERR("failed to add %s interface to %s: %s", + if_name, dpif_name(br->dpif), strerror(error)); } } } - out: - svec_destroy(&cur_ifaces); - svec_destroy(&want_ifaces); - svec_destroy(&add_ifaces); + 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; - struct iface *local_iface = NULL; - const char *devname; - uint8_t engine_type = br->dpif.minor; - uint8_t engine_id = br->dpif.minor; - bool add_id_to_iface = false; - struct svec nf_hosts; + struct iface *local_iface; + struct iface *hw_addr_iface; + char *dpid_string; bridge_fetch_dp_ifaces(br); - for (i = 0; i < br->n_ports; ) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; ) { - struct iface *iface = port->ifaces[j]; - if (iface->dp_ifidx < 0) { - VLOG_ERR("%s interface not in dp%u, dropping", - iface->name, dpif_id(&br->dpif)); - iface_destroy(iface); - } else { - if (iface->dp_ifidx == ODPP_LOCAL) { - local_iface = iface; - } - VLOG_DBG("dp%u has interface %s on port %d", - dpif_id(&br->dpif), iface->name, iface->dp_ifidx); - j++; - } - } - if (!port->n_ifaces) { - VLOG_ERR("%s port has no interfaces, dropping", port->name); - port_destroy(port); - continue; - } - i++; - } + 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. */ - bridge_pick_local_hw_addr(br, ea, &devname); + bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); + local_iface = bridge_get_local_iface(br); if (local_iface) { - int error = netdev_nodev_set_etheraddr(local_iface->name, ea); + int error = netdev_set_etheraddr(local_iface->netdev, ea); if (error) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge " @@ -538,39 +727,107 @@ bridge_reconfigure(void) } } - dpid = bridge_pick_datapath_id(br, ea, devname); + dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface); ofproto_set_datapath_id(br->ofproto, dpid); + dpid_string = xasprintf("%012"PRIx64, dpid); + ovsrec_bridge_set_datapath_id(br->cfg, dpid_string); + free(dpid_string); + /* Set NetFlow configuration on this bridge. */ - if (cfg_has("netflow.%s.engine-type", br->name)) { - engine_type = cfg_get_int(0, "netflow.%s.engine-type", - br->name); - } - if (cfg_has("netflow.%s.engine-id", br->name)) { - engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name); - } - if (cfg_has("netflow.%s.add-id-to-iface", br->name)) { - add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface", - br->name); - } - if (add_id_to_iface && engine_id > 0x7f) { - VLOG_WARN("bridge %s: netflow port mangling may conflict with " - "another vswitch, choose an engine id less than 128", - br->name); - } - if (add_id_to_iface && br->n_ports > 0x1ff) { - VLOG_WARN("bridge %s: netflow port mangling will conflict with " - "another port when 512 or more ports are used", - br->name); + if (br->cfg->netflow) { + struct ovsrec_netflow *nf_cfg = br->cfg->netflow; + struct netflow_options opts; + + memset(&opts, 0, sizeof opts); + + dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id); + if (nf_cfg->engine_type) { + opts.engine_type = *nf_cfg->engine_type; + } + if (nf_cfg->engine_id) { + opts.engine_id = *nf_cfg->engine_id; + } + + opts.active_timeout = nf_cfg->active_timeout; + if (!opts.active_timeout) { + opts.active_timeout = -1; + } else if (opts.active_timeout < 0) { + VLOG_WARN("bridge %s: active timeout interval set to negative " + "value, using default instead (%d seconds)", br->name, + NF_ACTIVE_TIMEOUT_DEFAULT); + opts.active_timeout = -1; + } + + opts.add_id_to_iface = nf_cfg->add_id_to_interface; + if (opts.add_id_to_iface) { + if (opts.engine_id > 0x7f) { + VLOG_WARN("bridge %s: netflow port mangling may conflict " + "with another vswitch, choose an engine id less " + "than 128", br->name); + } + if (br->n_ports > 508) { + VLOG_WARN("bridge %s: netflow port mangling will conflict " + "with another port when more than 508 ports are " + "used", br->name); + } + } + + opts.collectors.n = nf_cfg->n_targets; + opts.collectors.names = nf_cfg->targets; + if (ofproto_set_netflow(br->ofproto, &opts)) { + VLOG_ERR("bridge %s: problem setting netflow collectors", + br->name); + } + } else { + ofproto_set_netflow(br->ofproto, NULL); } - svec_init(&nf_hosts); - cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name); - if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type, - engine_id, add_id_to_iface)) { - VLOG_ERR("bridge %s: problem setting netflow collectors", - br->name); + + /* 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); } - svec_destroy(&nf_hosts); /* Update the controller and related settings. It would be more * straightforward to call this from bridge_reconfigure_one(), but we @@ -581,42 +838,63 @@ bridge_reconfigure(void) * 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(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]; port_update_vlan_compat(port); - - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->dp_ifidx != ODPP_LOCAL - && iface_is_internal(br, iface->name)) { - iface_set_mac(iface); - } - } + port_update_bonding(port); } } LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - brstp_reconfigure(br); + 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) +{ + size_t 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], - const char **devname) + struct iface **hw_addr_iface) { - uint64_t requested_ea; + const char *hwaddr; size_t i, j; int error; - *devname = NULL; + *hw_addr_iface = NULL; /* Did the user request a particular MAC? */ - requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name); - if (requested_ea) { - eth_addr_from_uint64(requested_ea, ea); + hwaddr = bridge_get_other_config(br->cfg, "hwaddr"); + if (hwaddr && eth_addr_from_string(hwaddr, ea)) { if (eth_addr_is_multicast(ea)) { VLOG_ERR("bridge %s: cannot set MAC address to multicast " "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); @@ -627,14 +905,12 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], } } - /* Otherwise choose the minimum MAC address among all of the interfaces. - * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the - * MAC of the physical interface in such an environment.) */ + /* Otherwise choose the minimum non-local MAC address among all of the + * interfaces. */ memset(ea, 0xff, sizeof ea); for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; uint8_t iface_ea[ETH_ADDR_LEN]; - uint64_t iface_ea_u64; struct iface *iface; /* Mirror output ports don't participate. */ @@ -643,18 +919,14 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], } /* Choose the MAC address to represent the port. */ - iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name); - if (iface_ea_u64) { - /* User specified explicitly. */ - eth_addr_from_uint64(iface_ea_u64, iface_ea); - + if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) { /* Find the interface with this Ethernet address (if any) so that * we can provide the correct devname to the caller. */ iface = NULL; for (j = 0; j < port->n_ifaces; j++) { struct iface *candidate = port->ifaces[j]; uint8_t candidate_ea[ETH_ADDR_LEN]; - if (!netdev_nodev_get_etheraddr(candidate->name, candidate_ea) + if (!netdev_get_etheraddr(candidate->netdev, candidate_ea) && eth_addr_equals(iface_ea, candidate_ea)) { iface = candidate; } @@ -675,16 +947,13 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], } /* The local port doesn't count (since we're trying to choose its - * MAC address anyway). Other internal ports don't count because - * we really want a physical MAC if we can get it, and internal - * ports typically have randomly generated MACs. */ - if (iface->dp_ifidx == ODPP_LOCAL - || cfg_get_bool(0, "iface.%s.internal", iface->name)) { + * MAC address anyway). */ + if (iface->dp_ifidx == ODPP_LOCAL) { continue; } /* Grab MAC. */ - error = netdev_nodev_get_etheraddr(iface->name, iface_ea); + error = netdev_get_etheraddr(iface->netdev, iface_ea); if (error) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s", @@ -695,17 +964,18 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], /* Compare against our current choice. */ if (!eth_addr_is_multicast(iface_ea) && + !eth_addr_is_local(iface_ea) && !eth_addr_is_reserved(iface_ea) && !eth_addr_is_zero(iface_ea) && memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) { memcpy(ea, iface_ea, ETH_ADDR_LEN); - *devname = iface ? iface->name : NULL; + *hw_addr_iface = iface; } } - if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) { + if (eth_addr_is_multicast(ea)) { memcpy(ea, br->default_ea, ETH_ADDR_LEN); - *devname = NULL; + *hw_addr_iface = NULL; VLOG_WARN("bridge %s: using default bridge Ethernet " "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); } else { @@ -716,13 +986,13 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], /* Choose and returns the datapath ID for bridge 'br' given that the bridge * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of - * a network device, then that network device's name must be passed in as - * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be - * passed in as a null pointer. */ + * an interface on 'br', then that interface must be passed in as + * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then + * 'hw_addr_iface' must be passed in as a null pointer. */ static uint64_t bridge_pick_datapath_id(struct bridge *br, const uint8_t bridge_ea[ETH_ADDR_LEN], - const char *devname) + struct iface *hw_addr_iface) { /* * The procedure for choosing a bridge MAC address will, in the most @@ -736,16 +1006,17 @@ bridge_pick_datapath_id(struct bridge *br, * stable from one run to the next, so that policy set on a datapath * "sticks". */ + const char *datapath_id; uint64_t dpid; - dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name); - if (dpid) { + datapath_id = bridge_get_other_config(br->cfg, "datapath-id"); + if (datapath_id && dpid_from_string(datapath_id, &dpid)) { return dpid; } - if (devname) { + if (hw_addr_iface) { int vlan; - if (!netdev_get_vlan_vid(devname, &vlan)) { + if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) { /* * A bridge whose MAC address is taken from a VLAN network device * (that is, a network device created with vconfig(8) or similar @@ -832,15 +1103,12 @@ 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; } - if (br->ml) { - mac_learning_wait(br->ml); - } + mac_learning_wait(br->ml); bond_wait(br); - brstp_wait(br); } } @@ -851,17 +1119,37 @@ bridge_flush(struct bridge *br) { COVERAGE_INC(bridge_flush); br->flush = true; - if (br->ml) { - mac_learning_flush(br->ml); + mac_learning_flush(br->ml); +} + +/* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no + * such interface. */ +static struct iface * +bridge_get_local_iface(struct bridge *br) +{ + size_t i, j; + + for (i = 0; i < br->n_ports; i++) { + struct port *port = br->ports[i]; + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + if (iface->dp_ifidx == ODPP_LOCAL) { + return iface; + } + } } + + return NULL; } /* Bridge unixctl user interface functions. */ static void -bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args) +bridge_unixctl_fdb_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct bridge *br; + const struct mac_entry *e; br = bridge_lookup(args); if (!br) { @@ -870,70 +1158,63 @@ bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args) } ds_put_cstr(&ds, " port VLAN MAC Age\n"); - if (br->ml) { - const struct mac_entry *e; - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { - if (e->port < 0 || e->port >= br->n_ports) { - continue; - } - ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", - br->ports[e->port]->ifaces[0]->dp_ifidx, - e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); + LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { + if (e->port < 0 || e->port >= br->n_ports) { + continue; } + ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", + br->ports[e->port]->ifaces[0]->dp_ifidx, + e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); } unixctl_command_reply(conn, 200, ds_cstr(&ds)); ds_destroy(&ds); } /* Bridge reconfiguration functions. */ - static struct bridge * -bridge_create(const char *name) +bridge_create(const struct ovsrec_bridge *br_cfg) { struct bridge *br; int error; - assert(!bridge_lookup(name)); - br = xcalloc(1, sizeof *br); + assert(!bridge_lookup(br_cfg->name)); + br = xzalloc(sizeof *br); - error = dpif_create(name, &br->dpif); - if (error == EEXIST) { - error = dpif_open(name, &br->dpif); - if (error) { - VLOG_ERR("datapath %s already exists but cannot be opened: %s", - name, strerror(error)); - free(br); - return NULL; - } - dpif_flow_flush(&br->dpif); - } else if (error) { - VLOG_ERR("failed to create datapath %s: %s", name, strerror(error)); + 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)); - dpif_delete(&br->dpif); - dpif_close(&br->dpif); + 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); - VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif)); + VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif)); return br; } @@ -948,16 +1229,17 @@ bridge_destroy(struct bridge *br) port_destroy(br->ports[br->n_ports - 1]); } list_remove(&br->node); - error = dpif_delete(&br->dpif); + error = dpif_delete(br->dpif); if (error && error != ENOENT) { - VLOG_ERR("failed to delete dp%u: %s", - dpif_id(&br->dpif), strerror(error)); + VLOG_ERR("failed to delete %s: %s", + dpif_name(br->dpif), strerror(error)); } - dpif_close(&br->dpif); + 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); @@ -993,7 +1275,8 @@ bridge_get_datapathid(const char *name) /* 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) +bridge_unixctl_dump_flows(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) { struct bridge *br; struct ds results; @@ -1021,11 +1304,8 @@ bridge_run_one(struct bridge *br) return error; } - if (br->ml) { - mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); - } + mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); bond_run(br); - brstp_run(br); error = ofproto_run2(br->ofproto, br->flush); br->flush = false; @@ -1033,103 +1313,183 @@ bridge_run_one(struct bridge *br) return error; } -static const char * -bridge_get_controller(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 char *controller; + 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; + } - controller = cfg_get_string(0, "bridge.%s.controller", br->name); - if (!controller) { - controller = cfg_get_string(0, "mgmt.controller"); + if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) { + controllers = NULL; + n_controllers = 0; } - return controller && controller[0] ? controller : NULL; + + if (controllersp) { + *controllersp = controllers; + } + return n_controllers; } static void -bridge_reconfigure_one(struct bridge *br) +bridge_update_desc(struct bridge *br OVS_UNUSED) { - struct svec old_ports, new_ports, ifaces; - struct svec listeners, old_listeners; - struct svec snoops, old_snoops; - size_t i, j; - - /* Collect old ports. */ - svec_init(&old_ports); - for (i = 0; i < br->n_ports; i++) { - svec_add(&old_ports, br->ports[i]->name); +#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; } - svec_sort(&old_ports); - assert(svec_is_unique(&old_ports)); - /* Collect new ports. */ - svec_init(&new_ports); - cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name); - svec_sort(&new_ports); - if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) { - svec_add(&new_ports, br->name); - svec_sort(&new_ports); - } - if (!svec_is_unique(&new_ports)) { - VLOG_WARN("bridge %s: %s specified twice as bridge port", - br->name, svec_get_duplicate(&new_ports)); - svec_unique(&new_ports); + 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; } - ofproto_set_mgmt_id(br->ofproto, mgmt_id); + 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; + } - /* Get rid of deleted ports and add new ports. */ - for (i = 0; i < br->n_ports; ) { - struct port *port = br->ports[i]; - if (!svec_contains(&new_ports, port->name)) { - port_destroy(port); + 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 { - i++; + br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); } + changed = true; } - for (i = 0; i < new_ports.n; i++) { - const char *name = new_ports.names[i]; - if (!svec_contains(&old_ports, name)) { - port_create(br, name); + + 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; } - svec_destroy(&old_ports); - svec_destroy(&new_ports); - /* Reconfigure all ports. */ + 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) +{ + struct shash old_ports, new_ports; + struct svec listeners, old_listeners; + struct svec snoops, old_snoops; + struct shash_node *node; + size_t i; + + /* Collect old ports. */ + shash_init(&old_ports); for (i = 0; i < br->n_ports; i++) { - port_reconfigure(br->ports[i]); + shash_add(&old_ports, br->ports[i]->name, br->ports[i]); } - /* Check and delete duplicate interfaces. */ - svec_init(&ifaces); - for (i = 0; i < br->n_ports; ) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; ) { - struct iface *iface = port->ifaces[j]; - if (svec_contains(&ifaces, iface->name)) { - VLOG_ERR("bridge %s: %s interface is on multiple ports, " - "removing from %s", - br->name, iface->name, port->name); - iface_destroy(iface); - } else { - svec_add(&ifaces, iface->name); - svec_sort(&ifaces); - j++; - } + /* Collect new ports. */ + shash_init(&new_ports); + for (i = 0; i < br->cfg->n_ports; i++) { + const char *name = br->cfg->ports[i]->name; + if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) { + VLOG_WARN("bridge %s: %s specified twice as bridge port", + br->name, name); } - if (!port->n_ifaces) { - VLOG_ERR("%s port has no interfaces, dropping", port->name); + } + + /* If we have a controller, then we need a local port. Complain if the + * user didn't specify one. + * + * XXX perhaps we should synthesize a port ourselves in this case. */ + if (bridge_get_controllers(ovs_cfg, br, NULL)) { + char local_name[IF_NAMESIZE]; + int error; + + error = dpif_port_get_name(br->dpif, ODPP_LOCAL, + local_name, sizeof local_name); + if (!error && !shash_find(&new_ports, local_name)) { + VLOG_WARN("bridge %s: controller specified but no local port " + "(port named %s) defined", + br->name, local_name); + } + } + + /* Get rid of deleted ports. + * Get rid of deleted interfaces on ports that still exist. */ + SHASH_FOR_EACH (node, &old_ports) { + struct port *port = node->data; + const struct ovsrec_port *port_cfg; + + port_cfg = shash_find_data(&new_ports, node->name); + if (!port_cfg) { port_destroy(port); } else { - i++; + port_del_ifaces(port, port_cfg); } } - svec_destroy(&ifaces); + + /* 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); /* 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); @@ -1171,56 +1531,95 @@ bridge_reconfigure_one(struct bridge *br) } 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); + ofproto_get_snoops(br->ofproto, &old_snoops); + if (!svec_equal(&snoops, &old_snoops)) { + ofproto_set_snoops(br->ofproto, &snoops); + } + svec_destroy(&snoops); + svec_destroy(&old_snoops); +#endif mirror_reconfigure(br); + + bridge_update_desc(br); } static void -bridge_reconfigure_controller(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 char *controller; + struct ovsrec_controller **controllers; + size_t n_controllers; - controller = bridge_get_controller(br); - if ((br->controller != NULL) != (controller != 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 = controller ? xstrdup(controller) : NULL; - if (controller) { - const char *fail_mode; - int max_backoff, probe; - int rate_limit, burst_limit; + if (!n_controllers) { + union ofp_action action; + flow_t flow; - if (!strcmp(controller, "discover")) { - bool update_resolv_conf = true; + /* Clear out controllers. */ + ofproto_set_controllers(br->ofproto, NULL, 0); - if (cfg_has("%s.update-resolv.conf", pfx)) { - update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf", - pfx); - } - ofproto_set_discovery(br->ofproto, true, - cfg_get_string(0, "%s.accept-regex", pfx), - update_resolv_conf); - } else { - struct netdev *netdev; - bool in_band; - int error; - - in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED, - "%s.in-band", pfx) - || cfg_get_bool(0, "%s.in-band", pfx)); - ofproto_set_discovery(br->ofproto, false, NULL, NULL); - ofproto_set_in_band(br->ofproto, in_band); - - error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev); - if (!error) { - if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) { - struct in_addr ip, mask, gateway; - ip.s_addr = cfg_get_ip(0, "%s.ip", pfx); - mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx); - gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx); + /* 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) { @@ -1234,115 +1633,55 @@ bridge_reconfigure_controller(struct bridge *br) } if (gateway.s_addr) { - if (!netdev_add_router(gateway)) { + if (!netdev_add_router(netdev, gateway)) { VLOG_INFO("bridge %s: configured gateway "IP_FMT, br->name, IP_ARGS(&gateway.s_addr)); } } } - netdev_close(netdev); - } - } - - fail_mode = cfg_get_string(0, "%s.fail-mode", pfx); - if (!fail_mode) { - fail_mode = cfg_get_string(0, "mgmt.fail-mode"); - } - ofproto_set_failure(br->ofproto, - (!fail_mode - || !strcmp(fail_mode, "standalone") - || !strcmp(fail_mode, "open"))); - - probe = cfg_get_int(0, "%s.inactivity-probe", pfx); - if (probe < 5) { - probe = cfg_get_int(0, "mgmt.inactivity-probe"); - if (probe < 5) { - probe = 5; - } - } - ofproto_set_probe_interval(br->ofproto, probe); - - max_backoff = cfg_get_int(0, "%s.max-backoff", pfx); - if (!max_backoff) { - max_backoff = cfg_get_int(0, "mgmt.max-backoff"); - if (!max_backoff) { - max_backoff = 8; } - } - ofproto_set_max_backoff(br->ofproto, max_backoff); - - rate_limit = cfg_get_int(0, "%s.rate-limit", pfx); - if (!rate_limit) { - rate_limit = cfg_get_int(0, "mgmt.rate-limit"); - } - burst_limit = cfg_get_int(0, "%s.burst-limit", pfx); - if (!burst_limit) { - burst_limit = cfg_get_int(0, "mgmt.burst-limit"); - } - ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit); - - ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx)); - if (cfg_has("%s.commands.acl", pfx)) { - struct svec command_acls; - char *command_acl; - - svec_init(&command_acls); - cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx); - command_acl = svec_join(&command_acls, ",", ""); - - ofproto_set_remote_execution(br->ofproto, command_acl, - cfg_get_string(0, "%s.commands.dir", - pfx)); - - svec_destroy(&command_acls); - free(command_acl); - } else { - ofproto_set_remote_execution(br->ofproto, NULL, NULL); - } - } else { - union ofp_action action; - flow_t flow; - - /* 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, OFPFW_ALL, 0, - &action, 1, 0); - - 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); - ofproto_set_stp(br->ofproto, false); + 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); } - free(pfx); - - ofproto_set_controller(br->ofproto, br->controller); } static void -bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces) +bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) { size_t i, j; - svec_init(ifaces); + shash_init(ifaces); for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; - svec_add(ifaces, iface->name); + shash_add_once(ifaces, iface->name, iface); } - if (port->n_ifaces > 1 - && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) { - svec_add(ifaces, port->name); + if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { + shash_add_once(ifaces, port->name, NULL); } } - svec_sort_unique(ifaces); } /* For robustness, in case the administrator moves around datapath ports behind @@ -1370,21 +1709,28 @@ bridge_fetch_dp_ifaces(struct bridge *br) } port_array_clear(&br->ifaces); - dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports); + dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); for (i = 0; i < n_dpif_ports; i++) { struct odp_port *p = &dpif_ports[i]; struct iface *iface = iface_lookup(br, p->devname); if (iface) { if (iface->dp_ifidx >= 0) { - VLOG_WARN("dp%u reported interface %s twice", - dpif_id(&br->dpif), p->devname); + VLOG_WARN("%s reported interface %s twice", + dpif_name(br->dpif), p->devname); } else if (iface_from_dp_ifidx(br, p->port)) { - VLOG_WARN("dp%u reported interface %"PRIu16" twice", - dpif_id(&br->dpif), p->port); + VLOG_WARN("%s reported interface %"PRIu16" twice", + dpif_name(br->dpif), p->port); } else { port_array_set(&br->ifaces, p->port, iface); iface->dp_ifidx = p->port; } + + if (iface->cfg) { + int64_t ofport = (iface->dp_ifidx >= 0 + ? odp_port_to_ofp_port(iface->dp_ifidx) + : -1); + ovsrec_interface_set_ofport(iface->cfg, &ofport, 1); + } } } free(dpif_ports); @@ -1407,13 +1753,31 @@ lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN]) static int bond_choose_iface(const struct port *port) { - size_t i; + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); + size_t i, best_down_slave = -1; + long long next_delay_expiration = LLONG_MAX; + for (i = 0; i < port->n_ifaces; i++) { - if (port->ifaces[i]->enabled) { + struct iface *iface = port->ifaces[i]; + + if (iface->enabled) { return i; + } else if (iface->delay_expires < next_delay_expiration) { + best_down_slave = i; + next_delay_expiration = iface->delay_expires; } } - return -1; + + if (best_down_slave != -1) { + struct iface *iface = port->ifaces[best_down_slave]; + + VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay " + "since no other interface is up", iface->name, + iface->delay_expires - time_msec()); + bond_enable_slave(iface, true); + } + + return best_down_slave; } static bool @@ -1463,10 +1827,12 @@ bond_link_status_update(struct iface *iface, bool carrier) iface->delay_expires = LLONG_MAX; VLOG_INFO_RL(&rl, "interface %s: will not be %s", iface->name, carrier ? "disabled" : "enabled"); - } else if (carrier && port->updelay && port->active_iface < 0) { - iface->delay_expires = time_msec(); - VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " - "other interface is up", iface->name, port->updelay); + } else if (carrier && port->active_iface < 0) { + bond_enable_slave(iface, true); + if (port->updelay) { + VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " + "other interface is up", iface->name, port->updelay); + } } else { int delay = carrier ? port->updelay : port->downdelay; iface->delay_expires = time_msec() + delay; @@ -1503,6 +1869,12 @@ bond_enable_slave(struct iface *iface, bool enable) struct port *port = iface->port; struct bridge *br = port->bridge; + /* This acts as a recursion check. If the act of disabling a slave + * causes a different slave to be enabled, the flag will allow us to + * skip redundant work when we reenter this function. It must be + * cleared on exit to keep things safe with multiple bonds. */ + static bool moving_active_iface = false; + iface->delay_expires = LLONG_MAX; if (enable == iface->enabled) { return; @@ -1515,19 +1887,57 @@ bond_enable_slave(struct iface *iface, bool enable) if (iface->port_ifidx == port->active_iface) { ofproto_revalidate(br->ofproto, port->active_iface_tag); + + /* Disabling a slave can lead to another slave being immediately + * enabled if there will be no active slaves but one is waiting + * on an updelay. In this case we do not need to run most of the + * code for the newly enabled slave since there was no period + * without an active slave and it is redundant with the disabling + * path. */ + moving_active_iface = true; bond_choose_active_iface(port); } bond_send_learning_packets(port); } else { VLOG_WARN("interface %s: enabled", iface->name); - if (port->active_iface < 0) { + if (port->active_iface < 0 && !moving_active_iface) { ofproto_revalidate(br->ofproto, port->no_ifaces_tag); bond_choose_active_iface(port); bond_send_learning_packets(port); } iface->tag = tag_create_random(); } - port_update_bond_compat(port); + + moving_active_iface = false; + 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)) { + 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; + } + } + + if (!netdev_open_default(port->name, &bond_dev)) { + netdev_set_stats(bond_dev, &bond_stats); + netdev_close(bond_dev); + } } static void @@ -1538,20 +1948,25 @@ bond_run(struct bridge *br) for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; + if (port->n_ifaces >= 2) { + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + if (time_msec() >= iface->delay_expires) { + bond_enable_slave(iface, !iface->enabled); + } + } + + if (port->bond_fake_iface + && time_msec() >= port->bond_next_fake_iface_update) { + bond_update_fake_iface_stats(port); + port->bond_next_fake_iface_update = time_msec() + 1000; + } + } + if (port->bond_compat_is_stale) { port->bond_compat_is_stale = false; port_update_bond_compat(port); } - - if (port->n_ifaces < 2) { - continue; - } - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (time_msec() >= iface->delay_expires) { - bond_enable_slave(iface, !iface->enabled); - } - } } } @@ -1568,9 +1983,12 @@ 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); + } } } @@ -1579,18 +1997,6 @@ set_dst(struct dst *p, const flow_t *flow, const struct port *in_port, const struct port *out_port, tag_type *tags) { - /* STP handling. - * - * XXX This uses too many tags: any broadcast flow will get one tag per - * destination port, and thus a broadcast on a switch of any size is likely - * to have all tag bits set. We should figure out a way to be smarter. - * - * This is OK when STP is disabled, because stp_state_tag is 0 then. */ - *tags |= out_port->stp_state_tag; - if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) { - return false; - } - p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE : in_port->vlan >= 0 ? in_port->vlan : ntohs(flow->dl_vlan)); @@ -1665,7 +2071,8 @@ 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 @@ -1677,13 +2084,12 @@ port_includes_vlan(const struct port *port, uint16_t vlan) static size_t compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, - struct dst dsts[], tag_type *tags) + struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface) { mirror_mask_t mirrors = in_port->src_mirrors; struct dst *dst = dsts; size_t i; - *tags |= in_port->stp_state_tag; if (out_port == FLOOD_PORT) { /* XXX use ODP_FLOOD if no vlans or bonding. */ /* XXX even better, define each VLAN as a datapath port group */ @@ -1696,7 +2102,9 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, dst++; } } + *nf_output_iface = NF_OUT_FLOOD; } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) { + *nf_output_iface = dst->dp_ifidx; mirrors |= out_port->dst_mirrors; dst++; } @@ -1750,7 +2158,7 @@ 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++) { @@ -1764,14 +2172,16 @@ print_dsts(const struct dst *dsts, size_t n) static void compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, - tag_type *tags, struct odp_actions *actions) + tag_type *tags, struct odp_actions *actions, + uint16_t *nf_output_iface) { struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)]; size_t n_dsts; const struct dst *p; uint16_t cur_vlan; - n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags); + n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags, + nf_output_iface); cur_vlan = ntohs(flow->dl_vlan); for (p = dsts; p < &dsts[n_dsts]; p++) { @@ -1790,34 +2200,117 @@ compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, } } +/* Returns the effective vlan of a packet, taking into account both the + * 802.1Q header and implicitly tagged ports. A value of 0 indicates that + * the packet is untagged and -1 indicates it has an invalid header and + * should be dropped. */ +static int flow_get_vlan(struct bridge *br, const flow_t *flow, + struct port *in_port, bool have_packet) +{ + /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet + * belongs to VLAN 0, so we should treat both cases identically. (In the + * former case, the packet has an 802.1Q header that specifies VLAN 0, + * presumably to allow a priority to be specified. In the latter case, the + * packet does not have any 802.1Q header.) */ + int vlan = ntohs(flow->dl_vlan); + if (vlan == OFP_VLAN_NONE) { + vlan = 0; + } + if (in_port->vlan >= 0) { + if (vlan) { + /* XXX support double tagging? */ + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " + "packet received on port %s configured with " + "implicit VLAN %"PRIu16, + br->name, ntohs(flow->dl_vlan), + in_port->name, in_port->vlan); + } + return -1; + } + vlan = in_port->vlan; + } else { + if (!port_includes_vlan(in_port, vlan)) { + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " + "packet received on port %s not configured for " + "trunking VLAN %d", + br->name, vlan, in_port->name, vlan); + } + return -1; + } + } + + return vlan; +} + +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); + if (rev_tag) { + /* The log messages here could actually be useful in debugging, + * so keep the rate limit relatively high. */ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, + 300); + VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " + "on port %s in VLAN %d", + br->name, ETH_ADDR_ARGS(flow->dl_src), + in_port->name, vlan); + ofproto_revalidate(br->ofproto, rev_tag); + } +} + +/* 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_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet) +is_gratuitous_arp(const flow_t *flow) { - struct arp_eth_header *arp = (struct arp_eth_header *) packet->data; return (flow->dl_type == htons(ETH_TYPE_ARP) && eth_addr_is_broadcast(flow->dl_dst) - && packet->size >= sizeof(struct arp_eth_header) - && arp->ar_op == ARP_OP_REQUEST); + && (flow->nw_proto == ARP_OP_REPLY + || (flow->nw_proto == ARP_OP_REQUEST + && flow->nw_src == flow->nw_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. */ +/* 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 -process_flow(struct bridge *br, const flow_t *flow, - const struct ofpbuf *packet, struct odp_actions *actions, - tag_type *tags) +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; /* 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 @@ -1835,66 +2328,29 @@ 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_port = in_iface->port; - - /* Figure out what VLAN this packet belongs to. - * - * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet - * belongs to VLAN 0, so we should treat both cases identically. (In the - * former case, the packet has an 802.1Q header that specifies VLAN 0, - * presumably to allow a priority to be specified. In the latter case, the - * packet does not have any 802.1Q header.) */ - vlan = ntohs(flow->dl_vlan); - if (vlan == OFP_VLAN_NONE) { - vlan = 0; - } - if (in_port->vlan >= 0) { - if (vlan) { - /* XXX support double tagging? */ - if (packet != NULL) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " - "packet received on port %s configured with " - "implicit VLAN %"PRIu16, - br->name, ntohs(flow->dl_vlan), - in_port->name, in_port->vlan); - } - goto done; - } - vlan = in_port->vlan; - } else { - if (!port_includes_vlan(in_port, vlan)) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " - "packet received on port %s not configured for " - "trunking VLAN %d", - br->name, vlan, in_port->name, vlan); - goto done; - } + *in_portp = NULL; + return false; } - - /* Drop frames for ports that STP wants entirely killed (both for - * forwarding and for learning). Later, after we do learning, we'll drop - * the frames that STP wants to do learning but not forwarding on. */ - if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) { - goto done; + *in_portp = in_port = in_iface->port; + *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet); + if (vlan < 0) { + 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. */ @@ -1905,69 +2361,75 @@ process_flow(struct bridge *br, const flow_t *flow, *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 + * it back on the other. Gratuitous ARP packets are an exception * to this rule: the host has moved to another switch. */ src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); if (src_idx != -1 && src_idx != in_port->port_idx && - (!packet || !is_bcast_arp_reply(flow, packet))) { - goto done; - } - } - - /* MAC learning. */ - out_port = FLOOD_PORT; - if (br->ml) { - int out_port_idx; - - /* Learn source MAC (but don't try to learn from revalidation). */ - if (packet) { - tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src, - vlan, in_port->port_idx); - if (rev_tag) { - /* The log messages here could actually be useful in debugging, - * so keep the rate limit relatively high. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, - 300); - VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " - "on port %s in VLAN %d", - br->name, ETH_ADDR_ARGS(flow->dl_src), - in_port->name, vlan); - ofproto_revalidate(br->ofproto, rev_tag); - } + !is_gratuitous_arp(flow)) { + return false; } + } - /* Determine output port. */ - out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, - tags); - if (out_port_idx >= 0 && out_port_idx < br->n_ports) { - out_port = br->ports[out_port_idx]; - } + 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); + if (out_port_idx >= 0 && out_port_idx < br->n_ports) { + out_port = br->ports[out_port_idx]; + } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { + /* If we are revalidating but don't have a learning entry then + * eject the flow. Installing a flow that floods packets opens + * up a window of time where we could learn from a packet reflected + * on a bond and blackhole packets before the learning table is + * updated to reflect the correct port. */ + return false; + } else { + out_port = FLOOD_PORT; } - /* Don't send packets out their input ports. Don't forward frames that STP - * wants us to discard. */ - if (in_port == out_port || in_port->stp_state == STP_LEARNING) { + /* Don't send packets out their input ports. */ + if (in_port == out_port) { out_port = NULL; } done: - compose_actions(br, flow, vlan, in_port, out_port, tags, actions); + if (in_port) { + compose_actions(br, flow, vlan, in_port, out_port, tags, actions, + nf_output_iface); + } - /* - * We send out only a single packet, instead of setting up a flow, if the - * packet is an ARP directed to broadcast that arrived on a bonded - * interface. In such a situation ARP requests and replies must be handled - * differently, but OpenFlow unfortunately can't distinguish them. - */ - return (in_port->n_ifaces < 2 - || flow->dl_type != htons(ETH_TYPE_ARP) - || !eth_addr_is_broadcast(flow->dl_dst)); + return true; } /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port @@ -1999,7 +2461,6 @@ bridge_port_changed_ofhook_cb(enum ofp_port_reason reason, bridge_flush(br); } else { - memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN); if (port->n_ifaces > 1) { bool up = !(opp->state & OFPPS_LINK_DOWN); bond_link_status_update(iface, up); @@ -2010,20 +2471,13 @@ bridge_port_changed_ofhook_cb(enum ofp_port_reason reason, static bool bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, - struct odp_actions *actions, tag_type *tags, void *br_) + struct odp_actions *actions, tag_type *tags, + uint16_t *nf_output_iface, void *br_) { struct bridge *br = br_; -#if 0 - if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE) - && eth_addr_equals(flow->dl_dst, stp_eth_addr)) { - brstp_receive(br, flow, payload); - return true; - } -#endif - COVERAGE_INC(bridge_process_flow); - return process_flow(br, flow, packet, actions, tags); + return process_flow(br, flow, packet, actions, tags, nf_output_iface); } static void @@ -2034,6 +2488,16 @@ 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; + 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)) { + update_learning_table(br, flow, vlan, in_port); + } if (!br->has_bonded_ports) { return; @@ -2041,9 +2505,10 @@ bridge_account_flow_ofhook_cb(const flow_t *flow, for (a = actions; a < &actions[n_actions]; a++) { if (a->type == ODPAT_OUTPUT) { - struct port *port = port_from_dp_ifidx(br, a->output.port); - if (port && port->n_ifaces >= 2) { - struct bond_entry *e = lookup_bond_entry(port, flow->dl_src); + struct port *out_port = port_from_dp_ifidx(br, a->output.port); + if (out_port && out_port->n_ifaces >= 2) { + struct bond_entry *e = lookup_bond_entry(out_port, + flow->dl_src); e->tx_bytes += n_bytes; } } @@ -2054,22 +2519,18 @@ static void bridge_account_checkpoint_ofhook_cb(void *br_) { struct bridge *br = br_; - size_t i; - - if (!br->has_bonded_ports) { - return; - } + long long int now; + size_t i; - /* 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) { + if (!br->has_bonded_ports) { 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); } } @@ -2198,8 +2659,9 @@ log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port) /* Shifts 'hash' from 'from' to 'to' within 'port'. */ static void bond_shift_load(struct slave_balance *from, struct slave_balance *to, - struct bond_entry *hash) + int hash_idx) { + struct bond_entry *hash = from->hashes[hash_idx]; struct port *port = from->iface->port; uint64_t delta = hash->tx_bytes; @@ -2217,12 +2679,11 @@ bond_shift_load(struct slave_balance *from, struct slave_balance *to, * it require more work, the only purpose it would be to allow that hash to * be migrated to another slave in this rebalancing run, and there is no * point in doing that. */ - if (from->hashes[0] == hash) { + if (hash_idx == 0) { from->hashes++; } else { - int i = hash - from->hashes[0]; - memmove(from->hashes + i, from->hashes + i + 1, - (from->n_hashes - (i + 1)) * sizeof *from->hashes); + memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1, + (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes); } from->n_hashes--; @@ -2307,22 +2768,60 @@ bond_rebalance_port(struct port *port) /* 'from' is carrying significantly more load than 'to', and that * load is split across at least two different hashes. Pick a hash * to migrate to 'to' (the least-loaded slave), given that doing so - * must not cause 'to''s load to exceed 'from''s load. + * must decrease the ratio of the load on the two slaves by at + * least 0.1. * * The sort order we use means that we prefer to shift away the * smallest hashes instead of the biggest ones. There is little * reason behind this decision; we could use the opposite sort * order to shift away big hashes ahead of small ones. */ size_t i; + bool order_swapped; for (i = 0; i < from->n_hashes; i++) { + double old_ratio, new_ratio; uint64_t delta = from->hashes[i]->tx_bytes; - if (to->tx_bytes + delta < from->tx_bytes - delta) { + + if (delta == 0 || from->tx_bytes - delta == 0) { + /* Pointless move. */ + continue; + } + + order_swapped = from->tx_bytes - delta < to->tx_bytes + delta; + + if (to->tx_bytes == 0) { + /* Nothing on the new slave, move it. */ + break; + } + + old_ratio = (double)from->tx_bytes / to->tx_bytes; + new_ratio = (double)(from->tx_bytes - delta) / + (to->tx_bytes + delta); + + if (new_ratio == 0) { + /* Should already be covered but check to prevent division + * by zero. */ + continue; + } + + if (new_ratio < 1) { + new_ratio = 1 / new_ratio; + } + + if (old_ratio - new_ratio > 0.1) { + /* Would decrease the ratio, move it. */ break; } } if (i < from->n_hashes) { - bond_shift_load(from, to, from->hashes[i]); + bond_shift_load(from, to, i); + port->bond_compat_is_stale = true; + + /* If the result of the migration changed the relative order of + * 'from' and 'to' swap them back to maintain invariants. */ + if (order_swapped) { + swap_bals(from, to); + } /* Re-sort 'bals'. Note that this may make 'from' and 'to' * point to different slave_balance structures. It is only @@ -2333,7 +2832,6 @@ bond_rebalance_port(struct port *port) } else { from++; } - port->bond_compat_is_stale = true; } } @@ -2352,7 +2850,7 @@ bond_send_learning_packets(struct port *port) struct ofpbuf packet; int error, n_packets, n_errors; - if (!port->n_ifaces || port->active_iface < 0 || !br->ml) { + if (!port->n_ifaces || port->active_iface < 0) { return; } @@ -2388,7 +2886,7 @@ 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) { @@ -2412,7 +2910,8 @@ bond_send_learning_packets(struct port *port) /* Bonding unixctl user interface functions. */ static void -bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED) +bond_unixctl_list(struct unixctl_conn *conn, + const char *args OVS_UNUSED, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct bridge *br; @@ -2462,7 +2961,8 @@ bond_find(const char *name) } static void -bond_unixctl_show(struct unixctl_conn *conn, const char *args) +bond_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct port *port; @@ -2477,7 +2977,7 @@ bond_unixctl_show(struct unixctl_conn *conn, const char *args) 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; @@ -2503,14 +3003,10 @@ bond_unixctl_show(struct unixctl_conn *conn, const char *args) continue; } - ds_put_format(&ds, "\thash %d: %lld kB load\n", + ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", hash, be->tx_bytes / 1024); /* MACs. */ - if (!port->bridge->ml) { - break; - } - LIST_FOR_EACH (me, struct mac_entry, lru_node, &port->bridge->ml->lrus) { uint16_t dp_ifidx; @@ -2531,7 +3027,8 @@ bond_unixctl_show(struct unixctl_conn *conn, const char *args) } static void -bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_) +bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) { char *args = (char *) args_; char *save_ptr = NULL; @@ -2587,7 +3084,8 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_) } static void -bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_) +bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) { char *args = (char *) args_; char *save_ptr = NULL; @@ -2667,118 +3165,184 @@ enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) } static void -bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args) +bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, + void *aux OVS_UNUSED) { enable_slave(conn, args, true); } static void -bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args) +bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, + void *aux OVS_UNUSED) { enable_slave(conn, args, false); } +static void +bond_unixctl_hash(struct unixctl_conn *conn, const char *args, + void *aux OVS_UNUSED) +{ + uint8_t mac[ETH_ADDR_LEN]; + uint8_t hash; + char *hash_cstr; + + if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) + == ETH_ADDR_SCAN_COUNT) { + hash = bond_hash(mac); + + hash_cstr = xasprintf("%u", hash); + unixctl_command_reply(conn, 200, hash_cstr); + free(hash_cstr); + } else { + unixctl_command_reply(conn, 501, "invalid mac"); + } +} + static void bond_init(void) { - unixctl_command_register("bond/list", bond_unixctl_list); - unixctl_command_register("bond/show", bond_unixctl_show); - unixctl_command_register("bond/migrate", bond_unixctl_migrate); + unixctl_command_register("bond/list", bond_unixctl_list, NULL); + unixctl_command_register("bond/show", bond_unixctl_show, NULL); + unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL); unixctl_command_register("bond/set-active-slave", - bond_unixctl_set_active_slave); - unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave); - unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave); + bond_unixctl_set_active_slave, NULL); + unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave, + NULL); + unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave, + NULL); + unixctl_command_register("bond/hash", bond_unixctl_hash, NULL); } /* Port functions. */ -static void +static struct port * port_create(struct bridge *br, const char *name) { struct port *port; - port = xcalloc(1, sizeof *port); + port = xzalloc(sizeof *port); port->bridge = br; port->port_idx = br->n_ports; port->vlan = -1; port->trunks = NULL; port->name = xstrdup(name); port->active_iface = -1; - port->stp_state = STP_DISABLED; - port->stp_state_tag = 0; if (br->n_ports >= br->allocated_ports) { br->ports = x2nrealloc(br->ports, &br->allocated_ports, sizeof *br->ports); } br->ports[br->n_ports++] = port; + shash_add_assert(&br->port_by_name, port->name, port); VLOG_INFO("created port %s on bridge %s", port->name, br->name); bridge_flush(br); + + 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_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) +port_reconfigure(struct port *port, const struct ovsrec_port *cfg) { - bool bonded = cfg_has_section("bonding.%s", port->name); - struct svec old_ifaces, new_ifaces; + struct shash new_ifaces; + long long int next_rebalance; unsigned long *trunks; int vlan; size_t i; - /* Collect old and new interfaces. */ - svec_init(&old_ifaces); - svec_init(&new_ifaces); - for (i = 0; i < port->n_ifaces; i++) { - svec_add(&old_ifaces, port->ifaces[i]->name); + port->cfg = cfg; + + /* Update settings. */ + port->updelay = cfg->bond_updelay; + if (port->updelay < 0) { + port->updelay = 0; + } + port->updelay = cfg->bond_downdelay; + if (port->downdelay < 0) { + port->downdelay = 0; + } + port->bond_rebalance_interval = atoi( + get_port_other_config(cfg, "bond-rebalance-interval", "10000")); + if (port->bond_rebalance_interval < 1000) { + port->bond_rebalance_interval = 1000; + } + next_rebalance = time_msec() + port->bond_rebalance_interval; + if (port->bond_next_rebalance > next_rebalance) { + port->bond_next_rebalance = next_rebalance; } - svec_sort(&old_ifaces); - if (bonded) { - cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name); - if (!new_ifaces.n) { - VLOG_ERR("port %s: no interfaces specified for bonded port", - port->name); - } else if (new_ifaces.n == 1) { - VLOG_WARN("port %s: only 1 interface specified for bonded port", - port->name); - } - port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name); - if (port->updelay < 0) { - port->updelay = 0; - } - port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name); - if (port->downdelay < 0) { - port->downdelay = 0; + /* 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; + + 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; } - } else { - svec_init(&new_ifaces); - svec_add(&new_ifaces, port->name); - } - /* Get rid of deleted interfaces and add new interfaces. */ - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - if (!svec_contains(&new_ifaces, iface->name)) { - iface_destroy(iface); + 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 { - i++; - } - } - for (i = 0; i < new_ifaces.n; i++) { - const char *name = new_ifaces.names[i]; - if (!svec_contains(&old_ifaces, name)) { - iface_create(port, name); + iface_create(port, if_cfg); } } + shash_destroy(&new_ifaces); /* Get VLAN tag. */ vlan = -1; - if (cfg_has("vlan.%s.tag", port->name)) { - if (!bonded) { - vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name); + if (cfg->tag) { + if (port->n_ifaces < 2) { + vlan = *cfg->tag; if (vlan >= 0 && vlan <= 4095) { VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan); + } else { + vlan = -1; } } else { /* It's possible that bonded, VLAN-tagged ports make sense. Maybe @@ -2794,15 +3358,14 @@ port_reconfigure(struct port *port) /* Get trunked VLANs. */ trunks = NULL; - if (vlan < 0) { - size_t n_trunks, n_errors; + if (vlan < 0 && cfg->n_trunks) { + size_t n_errors; size_t i; trunks = bitmap_allocate(4096); - n_trunks = cfg_count("vlan.%s.trunks", port->name); n_errors = 0; - for (i = 0; i < n_trunks; i++) { - int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name); + for (i = 0; i < cfg->n_trunks; i++) { + int trunk = cfg->trunks[i]; if (trunk >= 0) { bitmap_set1(trunks, trunk); } else { @@ -2811,20 +3374,17 @@ port_reconfigure(struct port *port) } if (n_errors) { VLOG_ERR("port %s: invalid values for %zu trunk VLANs", - port->name, n_trunks); - } - if (n_errors == n_trunks) { - if (n_errors) { - VLOG_ERR("port %s: no valid trunks, trunking all VLANs", - port->name); - } - bitmap_set_multiple(trunks, 0, 4096, 1); + port->name, cfg->n_trunks); } - } else { - if (cfg_has("vlan.%s.trunks", port->name)) { - VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan", - port->name, port->name); + if (n_errors == cfg->n_trunks) { + 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 @@ -2833,9 +3393,6 @@ port_reconfigure(struct port *port) } bitmap_free(port->trunks); port->trunks = trunks; - - svec_destroy(&old_ifaces); - svec_destroy(&new_ifaces); } static void @@ -2844,7 +3401,7 @@ port_destroy(struct port *port) if (port) { struct bridge *br = port->bridge; struct port *del; - size_t i; + int i; proc_net_compat_update_vlan(port->name, NULL, 0); proc_net_compat_update_bond(port->name, NULL); @@ -2860,6 +3417,8 @@ 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; @@ -2881,29 +3440,14 @@ 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 @@ -2915,6 +3459,7 @@ port_update_bonding(struct port *port) free(port->bond_hash); port->bond_hash = NULL; port->bond_compat_is_stale = true; + port->bond_fake_iface = false; } } else { if (!port->bond_hash) { @@ -2928,8 +3473,15 @@ 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; } } @@ -2986,13 +3538,13 @@ port_update_bond_compat(struct port *port) if (slave->up) { bond.up = true; } - memcpy(slave->mac, iface->mac, ETH_ADDR_LEN); + netdev_get_etheraddr(iface->netdev, slave->mac); } - if (cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) { + 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 { @@ -3030,7 +3582,8 @@ port_update_vlan_compat(struct port *port) && p->n_ifaces && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0)) { - const uint8_t *ea = p->ifaces[0]->mac; + uint8_t ea[ETH_ADDR_LEN]; + netdev_get_etheraddr(p->ifaces[0]->netdev, ea); if (!eth_addr_is_multicast(ea) && !eth_addr_is_reserved(ea) && !eth_addr_is_zero(ea)) { @@ -3044,29 +3597,38 @@ port_update_vlan_compat(struct port *port) /* Interface functions. */ -static void -iface_create(struct port *port, const char *name) +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; - iface = xcalloc(1, sizeof *iface); + iface = xzalloc(sizeof *iface); iface->port = port; iface->port_ifidx = port->n_ifaces; iface->name = xstrdup(name); iface->dp_ifidx = -1; iface->tag = tag_create_random(); iface->delay_expires = LLONG_MAX; + iface->netdev = NULL; + iface->cfg = if_cfg; - if (!cfg_get_bool(0, "iface.%s.internal", iface->name)) { - netdev_nodev_get_etheraddr(name, iface->mac); - netdev_nodev_get_carrier(name, &iface->enabled); - } else { - /* Internal interfaces are created later by the call to dpif_port_add() - * in bridge_reconfigure(). Until then, we can't obtain any - * information about them. (There's no real value in doing so, anyway, - * because the 'mac' and 'enabled' values are only used for interfaces - * that are bond slaves, and it doesn't normally make sense to bond an - * internal interface.) */ + 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) { @@ -3075,13 +3637,14 @@ iface_create(struct port *port, const char *name) } port->ifaces[port->n_ifaces++] = iface; if (port->n_ifaces > 1) { - port->bridge->has_bonded_ports = true; + br->has_bonded_ports = true; } VLOG_DBG("attached network device %s to port %s", iface->name, port->name); - port_update_bonding(port); - bridge_flush(port->bridge); + bridge_flush(br); + + return iface; } static void @@ -3093,6 +3656,8 @@ 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); } @@ -3100,8 +3665,7 @@ iface_destroy(struct iface *iface) del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces]; del->port_ifidx = iface->port_ifidx; - free(iface->name); - free(iface); + netdev_close(iface->netdev); if (del_active) { ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); @@ -3109,7 +3673,9 @@ iface_destroy(struct iface *iface) bond_send_learning_packets(port); } - port_update_bonding(port); + free(iface->name); + free(iface); + bridge_flush(port->bridge); } } @@ -3117,18 +3683,7 @@ 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 * @@ -3148,20 +3703,24 @@ iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) * reason why this function takes a name instead of a struct iface: the fake * interfaces created this way do not have a struct iface. */ static bool -iface_is_internal(const struct bridge *br, const char *iface) +iface_is_internal(const struct bridge *br, const char *if_name) { - if (!strcmp(iface, br->name) - || cfg_get_bool(0, "iface.%s.internal", iface)) { + struct iface *iface; + struct port *port; + + if (!strcmp(if_name, br->name)) { return true; } - if (cfg_get_bool(0, "bonding.%s.fake-iface", iface)) { - struct port *port = port_lookup(br, iface); - if (port && port->n_ifaces > 1) { - return true; - } + iface = iface_lookup(br, if_name); + if (iface && !strcmp(iface->cfg->type, "internal")) { + return true; } + port = port_lookup(br, if_name); + if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) { + return true; + } return false; } @@ -3170,11 +3729,9 @@ iface_is_internal(const struct bridge *br, const char *iface) static void iface_set_mac(struct iface *iface) { - uint64_t mac = cfg_get_mac(0, "iface.%s.mac", iface->name); - if (mac) { - static uint8_t ea[ETH_ADDR_LEN]; + uint8_t ea[ETH_ADDR_LEN]; - eth_addr_from_uint64(mac, ea); + if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) { if (eth_addr_is_multicast(ea)) { VLOG_ERR("interface %s: cannot set MAC to multicast address", iface->name); @@ -3182,7 +3739,7 @@ iface_set_mac(struct iface *iface) VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead", iface->name, iface->name); } else { - int error = netdev_nodev_set_etheraddr(iface->name, ea); + int error = netdev_set_etheraddr(iface->netdev, ea); if (error) { VLOG_ERR("interface %s: setting MAC failed (%s)", iface->name, strerror(error)); @@ -3196,45 +3753,47 @@ iface_set_mac(struct iface *iface) static void mirror_reconfigure(struct bridge *br) { - struct svec old_mirrors, new_mirrors; - size_t i; + 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++) { @@ -3246,9 +3805,30 @@ mirror_reconfigure(struct bridge *br) m->out_port->is_mirror_output_port = true; } } + + /* Update flooded vlans (for RSPAN). */ + rspan_vlans = NULL; + if (br->cfg->n_flood_vlans) { + rspan_vlans = bitmap_allocate(4096); + + 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 %"PRId64, + br->name, vlan); + } else { + VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN", + br->name, vlan); + } + } + } + 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; @@ -3258,7 +3838,7 @@ 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; @@ -3268,16 +3848,18 @@ mirror_create(struct bridge *br, const char *name) VLOG_INFO("created port mirror %s on bridge %s", name, br->name); bridge_flush(br); - br->mirrors[i] = m = xcalloc(1, sizeof *m); + br->mirrors[i] = m = xzalloc(sizeof *m); 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 @@ -3292,8 +3874,8 @@ 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; @@ -3304,45 +3886,36 @@ 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; } @@ -3377,104 +3950,86 @@ port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) } static void -mirror_reconfigure_one(struct mirror *m) +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) @@ -3484,7 +4039,7 @@ mirror_reconfigure_one(struct mirror *m) 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; @@ -3492,215 +4047,6 @@ mirror_reconfigure_one(struct mirror *m) } /* Clean up. */ -exit: - svec_destroy(&src_ports); - svec_destroy(&dst_ports); - free(pfx); -} - -/* Spanning tree protocol. */ - -static void brstp_update_port_state(struct port *); - -static void -brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_) -{ - struct bridge *br = br_; - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - struct iface *iface = iface_from_dp_ifidx(br, port_no); - if (!iface) { - VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", - br->name, port_no); - } else if (eth_addr_is_zero(iface->mac)) { - VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC", - br->name, port_no); - } else { - union ofp_action action; - struct eth_header *eth = pkt->l2; - flow_t flow; - - memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN); - - memset(&action, 0, sizeof action); - action.type = htons(OFPAT_OUTPUT); - action.output.len = htons(sizeof action); - action.output.port = htons(port_no); - - flow_extract(pkt, ODPP_NONE, &flow); - ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt); - } - ofpbuf_delete(pkt); -} - -static void -brstp_reconfigure(struct bridge *br) -{ - size_t i; - - if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) { - if (br->stp) { - stp_destroy(br->stp); - br->stp = NULL; - - bridge_flush(br); - } - } else { - uint64_t bridge_address, bridge_id; - int bridge_priority; - - bridge_address = cfg_get_mac(0, "stp.%s.address", br->name); - if (!bridge_address) { - if (br->stp) { - bridge_address = (stp_get_bridge_id(br->stp) - & ((UINT64_C(1) << 48) - 1)); - } else { - uint8_t mac[ETH_ADDR_LEN]; - eth_addr_random(mac); - bridge_address = eth_addr_to_uint64(mac); - } - } - - if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority", - br->name)) { - bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name); - } else { - bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY; - } - - bridge_id = bridge_address | ((uint64_t) bridge_priority << 48); - if (!br->stp) { - br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br); - br->stp_last_tick = time_msec(); - bridge_flush(br); - } else { - if (bridge_id != stp_get_bridge_id(br->stp)) { - stp_set_bridge_id(br->stp, bridge_id); - bridge_flush(br); - } - } - - for (i = 0; i < br->n_ports; i++) { - struct port *p = br->ports[i]; - int dp_ifidx; - struct stp_port *sp; - int path_cost, priority; - bool enable; - - if (!p->n_ifaces) { - continue; - } - dp_ifidx = p->ifaces[0]->dp_ifidx; - if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) { - continue; - } - - sp = stp_get_port(br->stp, dp_ifidx); - enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED, - "stp.%s.port.%s.enabled", - br->name, p->name) - || cfg_get_bool(0, "stp.%s.port.%s.enabled", - br->name, p->name)); - if (p->is_mirror_output_port) { - enable = false; - } - if (enable != (stp_port_get_state(sp) != STP_DISABLED)) { - bridge_flush(br); /* Might not be necessary. */ - if (enable) { - stp_port_enable(sp); - } else { - stp_port_disable(sp); - } - } - - path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost", - br->name, p->name); - stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */); - - priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED, - "stp.%s.port.%s.priority", - br->name, p->name) - ? cfg_get_int(0, "stp.%s.port.%s.priority", - br->name, p->name) - : STP_DEFAULT_PORT_PRIORITY); - stp_port_set_priority(sp, priority); - } - - brstp_adjust_timers(br); - } - for (i = 0; i < br->n_ports; i++) { - brstp_update_port_state(br->ports[i]); - } -} - -static void -brstp_update_port_state(struct port *p) -{ - struct bridge *br = p->bridge; - enum stp_state state; - - /* Figure out new state. */ - state = STP_DISABLED; - if (br->stp && p->n_ifaces > 0) { - int dp_ifidx = p->ifaces[0]->dp_ifidx; - if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) { - state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx)); - } - } - - /* Update state. */ - if (p->stp_state != state) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10); - VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s", - p->name, stp_state_name(p->stp_state), - stp_state_name(state)); - if (p->stp_state == STP_DISABLED) { - bridge_flush(br); - } else { - ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag); - } - p->stp_state = state; - p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0 - : tag_create_random()); - } -} - -static void -brstp_adjust_timers(struct bridge *br) -{ - int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name); - int max_age = cfg_get_int(0, "stp.%s.max-age", br->name); - int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name); - - stp_set_hello_time(br->stp, hello_time ? hello_time : 2000); - stp_set_max_age(br->stp, max_age ? max_age : 20000); - stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000); -} - -static void -brstp_run(struct bridge *br) -{ - if (br->stp) { - long long int now = time_msec(); - long long int elapsed = now - br->stp_last_tick; - struct stp_port *sp; - - if (elapsed > 0) { - stp_tick(br->stp, MIN(INT_MAX, elapsed)); - br->stp_last_tick = now; - } - while (stp_get_changed_port(br->stp, &sp)) { - struct port *p = port_from_dp_ifidx(br, stp_port_no(sp)); - if (p) { - brstp_update_port_state(p); - } - } - } -} - -static void -brstp_wait(struct bridge *br) -{ - if (br->stp) { - poll_timer_wait(1000); - } + shash_destroy(&src_ports); + shash_destroy(&dst_ports); }