X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=4cf1b86f8f4931d57a34e23788e101b83a72b2a5;hb=9467fe624698a9b72f81bb1ddfce33a5fc85281c;hp=25df8d48e443007c45bc45c0b160b0b8233e67b8;hpb=3c71830aef987c86338d2d0223d4e6913aea25a4;p=openvswitch diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 25df8d48..4cf1b86f 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -30,7 +30,6 @@ #include #include #include "bitmap.h" -#include "cfg.h" #include "coverage.h" #include "dirs.h" #include "dpif.h" @@ -43,20 +42,23 @@ #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 "sha1.h" +#include "shash.h" #include "socket-util.h" -#include "stp.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" @@ -68,8 +70,6 @@ struct dst { uint16_t dp_ifidx; }; -extern uint64_t mgmt_id; - struct iface { /* These members are always valid. */ struct port *port; /* Containing port. */ @@ -83,6 +83,9 @@ struct iface { 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 @@ -132,22 +135,22 @@ 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? */ /* 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. */ @@ -180,9 +183,8 @@ 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. */ @@ -194,11 +196,13 @@ enum { DP_MAX = 256 }; static struct bridge *bridge_create(const char *name); 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 void bridge_reconfigure_one(const struct ovsrec_open_vswitch *, + struct bridge *); +static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *, + struct bridge *); +static void bridge_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 *, @@ -210,16 +214,17 @@ static uint64_t bridge_pick_datapath_id(struct bridge *, 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_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); @@ -229,22 +234,27 @@ static void port_update_bond_compat(struct port *); static void port_update_vlan_compat(struct port *); static void port_update_bonding(struct port *); +#if 0 static void mirror_create(struct bridge *, const char *name); static void mirror_destroy(struct mirror *); static void mirror_reconfigure(struct bridge *); static void mirror_reconfigure_one(struct mirror *); static bool vlan_is_mirrored(const struct mirror *, int vlan); +#else +static bool vlan_is_mirrored(const struct mirror *m UNUSED, int vlan UNUSED) +{ + return false; +} +#endif -static 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 *, uint16_t dp_ifidx); +static bool iface_is_internal(const struct bridge *, const char *name); +static void iface_set_mac(struct iface *); /* Hooks into ofproto processing. */ static struct ofhooks bridge_ofhooks; @@ -278,14 +288,20 @@ 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) { + struct svec bridge_names; struct svec dpif_names; size_t i; - unixctl_command_register("fdb/show", bridge_unixctl_fdb_show); + 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); svec_init(&dpif_names); dp_enumerate(&dpif_names); @@ -302,7 +318,7 @@ bridge_init(void) svec_init(&all_names); dpif_get_all_names(dpif, &all_names); for (j = 0; j < all_names.n; j++) { - if (cfg_has("bridge.%s.port", all_names.names[j])) { + if (svec_contains(&bridge_names, all_names.names[j])) { goto found; } } @@ -312,18 +328,19 @@ bridge_init(void) dpif_close(dpif); } } + svec_destroy(&dpif_names); - unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows); + unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, + NULL); bond_init(); - bridge_reconfigure(); + bridge_reconfigure(cfg); } #ifdef HAVE_OPENSSL static bool -config_string_change(const char *key, char **valuep) +config_string_change(const char *value, char **valuep) { - const char *value = cfg_get_string(0, "%s", key); if (value && (!*valuep || strcmp(value, *valuep))) { free(*valuep); *valuep = xstrdup(value); @@ -334,7 +351,7 @@ config_string_change(const char *key, char **valuep) } static void -bridge_configure_ssl(void) +bridge_configure_ssl(const struct ovsrec_ssl *ssl) { /* XXX SSL should be configurable on a per-bridge basis. * XXX should be possible to de-configure SSL. */ @@ -343,11 +360,16 @@ bridge_configure_ssl(void) static char *cacert_file; struct stat s; - if (config_string_change("ssl.private-key", &private_key_file)) { + if (!ssl) { + /* XXX We can't un-set SSL settings. */ + return; + } + + if (config_string_change(ssl->private_key, &private_key_file)) { vconn_ssl_set_private_key_file(private_key_file); } - if (config_string_change("ssl.certificate", &certificate_file)) { + if (config_string_change(ssl->certificate, &certificate_file)) { vconn_ssl_set_certificate_file(certificate_file); } @@ -356,14 +378,57 @@ bridge_configure_ssl(void) * 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) + if (config_string_change(ssl->ca_cert, &cacert_file) || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) { - vconn_ssl_set_ca_cert_file(cacert_file, - cfg_get_bool(0, "ssl.bootstrap-ca-cert")); + vconn_ssl_set_ca_cert_file(cacert_file, 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, bool create) +{ + struct shash_node *node; + struct shash options; + int error; + size_t i; + + /* If a type is not explicitly declared, then assume it's an existing + * "system" device. */ + if (iface_cfg->type[0] == '\0' || !strcmp(iface_cfg->type, "system")) { + return 0; + } + + shash_init(&options); + for (i = 0; i < iface_cfg->n_options; i++) { + shash_add(&options, iface_cfg->key_options[i], + xstrdup(iface_cfg->value_options[i])); + } + + if (create) { + error = netdev_create(iface_cfg->name, iface_cfg->type, &options); + } else { + /* xxx Check to make sure that the type hasn't changed. */ + error = netdev_reconfigure(iface_cfg->name, &options); + } + + SHASH_FOR_EACH (node, &options) { + free(node->data); + } + shash_destroy(&options); + + return error; +} + +static int +reconfigure_iface(const struct ovsrec_interface *iface_cfg) +{ + return set_up_iface(iface_cfg, false); +} + + /* iterate_and_prune_ifaces() callback function that opens the network device * for 'iface', if it is not already open, and retrieves the interface's MAC * address and carrier status. */ @@ -400,12 +465,21 @@ check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED) } static bool -set_iface_policing(struct bridge *br UNUSED, struct iface *iface, +set_iface_properties(struct bridge *br UNUSED, struct iface *iface, void *aux UNUSED) { - int rate = cfg_get_int(0, "port.%s.ingress.policing-rate", iface->name); - int burst = cfg_get_int(0, "port.%s.ingress.policing-burst", iface->name); - netdev_set_policing(iface->netdev, rate, burst); + /* Set policing attributes. */ + netdev_set_policing(iface->netdev, + iface->cfg->ingress_policing_rate, + iface->cfg->ingress_policing_burst); + + /* 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); + } + return true; } @@ -441,49 +515,61 @@ iterate_and_prune_ifaces(struct bridge *br, } void -bridge_reconfigure(void) +bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) { - struct svec old_br, new_br; + struct ovsdb_idl_txn *txn; + struct shash old_br, new_br; + struct shash_node *node; struct bridge *br, *next; size_t i; COVERAGE_INC(bridge_reconfigure); + txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl); + /* Collect old and new bridges. */ - svec_init(&old_br); - svec_init(&new_br); + shash_init(&old_br); + shash_init(&new_br); LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - svec_add(&old_br, 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); + } } - cfg_get_subsections(&new_br, "bridge"); /* 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; + if (!shash_find_data(&old_br, br_name)) { + br = bridge_create(br_name); + if (br) { + br->cfg = 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. @@ -494,13 +580,13 @@ 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); 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); if (retval) { @@ -510,64 +596,66 @@ bridge_reconfigure(void) } } } - 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; + struct shash cur_ifaces, want_ifaces; + struct shash_node *node; + /* Get the set of interfaces currently in this datapath. */ dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); - svec_init(&cur_ifaces); + 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); - - for (i = 0; i < add_ifaces.n; i++) { - const char *if_name = add_ifaces.names[i]; - bool internal; - int error; - - /* It's an internal interface if it's marked that way, or if - * it's a bonded interface for which we're faking up a network - * device. */ - internal = cfg_get_bool(0, "iface.%s.internal", if_name); - if (cfg_get_bool(0, "bonding.%s.fake-iface", if_name)) { - struct port *port = port_lookup(br, if_name); - if (port && port->n_ifaces > 1) { - internal = true; - } - } - /* Add to datapath. */ - 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)); + 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); + } + } 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, + 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)); + } } } - svec_destroy(&cur_ifaces); - svec_destroy(&want_ifaces); - svec_destroy(&add_ifaces); + shash_destroy(&cur_ifaces); + shash_destroy(&want_ifaces); } LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { uint8_t ea[8]; uint64_t dpid; struct iface *local_iface; struct iface *hw_addr_iface; - uint8_t engine_type, engine_id; - bool add_id_to_iface = false; - struct svec nf_hosts; + char *dpid_string; bridge_fetch_dp_ifaces(br); iterate_and_prune_ifaces(br, init_iface_netdev, NULL); @@ -590,35 +678,57 @@ bridge_reconfigure(void) 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. */ - dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id); - 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); - } - 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); + 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); } /* Update the controller and related settings. It would be more @@ -630,35 +740,52 @@ 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_controller(ovs_cfg, br); } 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); port_update_bonding(port); } } LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - brstp_reconfigure(br); - iterate_and_prune_ifaces(br, set_iface_policing, NULL); + 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 */ +} + +static const char * +bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key) +{ + size_t i; + + for (i = 0; i < br_cfg->n_other_config; i++) { + if (!strcmp(br_cfg->key_other_config[i], key)) { + return br_cfg->value_other_config[i]; + } } + return NULL; } static void bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], struct iface **hw_addr_iface) { - uint64_t requested_ea; + const char *hwaddr; size_t i, j; int error; *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)); @@ -669,14 +796,12 @@ 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. */ @@ -685,11 +810,7 @@ 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; @@ -717,11 +838,8 @@ 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; } @@ -737,14 +855,16 @@ 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); *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); *hw_addr_iface = NULL; VLOG_WARN("bridge %s: using default bridge Ethernet " @@ -777,10 +897,11 @@ 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; } @@ -877,11 +998,8 @@ bridge_wait(void) continue; } - if (br->ml) { - mac_learning_wait(br->ml); - } + mac_learning_wait(br->ml); bond_wait(br); - brstp_wait(br); } } @@ -892,9 +1010,7 @@ 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 @@ -919,10 +1035,12 @@ bridge_get_local_iface(struct bridge *br) /* Bridge unixctl user interface functions. */ static void -bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args) +bridge_unixctl_fdb_show(struct unixctl_conn *conn, + const char *args, void *aux UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct bridge *br; + const struct mac_entry *e; br = bridge_lookup(args); if (!br) { @@ -931,16 +1049,13 @@ 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); @@ -955,23 +1070,14 @@ bridge_create(const char *name) int error; assert(!bridge_lookup(name)); - br = xcalloc(1, sizeof *br); + br = xzalloc(sizeof *br); - error = dpif_create(name, &br->dpif); - if (error == EEXIST || error == EBUSY) { - 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(name, &br->dpif); + if (error) { free(br); return NULL; } + dpif_flow_flush(br->dpif); error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto); if (error) { @@ -1054,7 +1160,8 @@ 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 UNUSED) { struct bridge *br; struct ds results; @@ -1082,11 +1189,8 @@ 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; @@ -1094,16 +1198,21 @@ bridge_run_one(struct bridge *br) return error; } -static const char * -bridge_get_controller(const struct bridge *br) +static const struct ovsrec_controller * +bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg, + const struct bridge *br) { - const char *controller; + const struct ovsrec_controller *controller; - controller = cfg_get_string(0, "bridge.%s.controller", br->name); - if (!controller) { - controller = cfg_get_string(0, "mgmt.controller"); + controller = (br->cfg->controller ? br->cfg->controller + : ovs_cfg->controller ? ovs_cfg->controller + : NULL); + + if (controller && !strcmp(controller->target, "none")) { + return NULL; } - return controller && controller[0] ? controller : NULL; + + return controller; } static bool @@ -1123,66 +1232,68 @@ check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_) } static void -bridge_reconfigure_one(struct bridge *br) +bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, + struct bridge *br) { - struct svec old_ports, new_ports, ifaces; + struct shash old_ports, new_ports; + struct svec ifaces; struct svec listeners, old_listeners; struct svec snoops, old_snoops; + struct shash_node *node; + uint64_t mgmt_id; size_t i; /* Collect old ports. */ - svec_init(&old_ports); + shash_init(&old_ports); for (i = 0; i < br->n_ports; i++) { - svec_add(&old_ports, br->ports[i]->name); + shash_add(&old_ports, br->ports[i]->name, br->ports[i]); } - 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)) { + 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 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_controller(ovs_cfg, br)) { char local_name[IF_NAMESIZE]; int error; error = dpif_port_get_name(br->dpif, ODPP_LOCAL, local_name, sizeof local_name); - if (!error && !svec_contains(&new_ports, local_name)) { - svec_add(&new_ports, local_name); - svec_sort(&new_ports); + 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); } } - 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); - } + dpid_from_string(ovs_cfg->management_id, &mgmt_id); ofproto_set_mgmt_id(br->ofproto, mgmt_id); /* Get rid of deleted ports and add new ports. */ - for (i = 0; i < br->n_ports; ) { - struct port *port = br->ports[i]; - if (!svec_contains(&new_ports, port->name)) { - port_destroy(port); - } else { - i++; + SHASH_FOR_EACH (node, &old_ports) { + if (!shash_find(&new_ports, node->name)) { + port_destroy(node->data); } } - 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); + 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); } - svec_destroy(&old_ports); - svec_destroy(&new_ports); - - /* Reconfigure all ports. */ - for (i = 0; i < br->n_ports; i++) { - port_reconfigure(br->ports[i]); - } + shash_destroy(&old_ports); + shash_destroy(&new_ports); /* Check and delete duplicate interfaces. */ svec_init(&ifaces); @@ -1193,6 +1304,7 @@ bridge_reconfigure_one(struct bridge *br) * 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); @@ -1234,56 +1346,81 @@ 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 +#if 0 mirror_reconfigure(br); +#endif } static void -bridge_reconfigure_controller(struct bridge *br) +bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg, + struct bridge *br) { char *pfx = xasprintf("bridge.%s.controller", br->name); - const char *controller; + const struct ovsrec_controller *c; - controller = bridge_get_controller(br); - if ((br->controller != NULL) != (controller != NULL)) { + c = bridge_get_controller(ovs_cfg, br); + if ((br->controller != NULL) != (c != NULL)) { ofproto_flush_flows(br->ofproto); } free(br->controller); - br->controller = controller ? xstrdup(controller) : NULL; + br->controller = c ? xstrdup(c->target) : NULL; - if (controller) { - const char *fail_mode; + if (c) { int max_backoff, probe; int rate_limit, burst_limit; - if (!strcmp(controller, "discover")) { - bool update_resolv_conf = true; - - if (cfg_has("%s.update-resolv.conf", pfx)) { - update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf", - pfx); - } + if (!strcmp(c->target, "discover")) { ofproto_set_discovery(br->ofproto, true, - cfg_get_string(0, "%s.accept-regex", pfx), - update_resolv_conf); + c->discover_accept_regex, + c->discover_update_resolv_conf); } else { struct iface *local_iface; + struct in_addr ip; bool in_band; - in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED, - "%s.in-band", pfx) - || cfg_get_bool(0, "%s.in-band", pfx)); + in_band = (!c->connection_mode + || !strcmp(c->connection_mode, "out-of-band")); ofproto_set_discovery(br->ofproto, false, NULL, NULL); ofproto_set_in_band(br->ofproto, in_band); local_iface = bridge_get_local_iface(br); - if (local_iface - && cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) { + if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) { struct netdev *netdev = local_iface->netdev; 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); + + 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) { @@ -1305,62 +1442,20 @@ bridge_reconfigure_controller(struct bridge *br) } } - 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; - } - } + (!c->fail_mode + || !strcmp(c->fail_mode, "standalone") + || !strcmp(c->fail_mode, "open"))); + + probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; ofproto_set_probe_interval(br->ofproto, probe); - max_backoff = 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; - } - } + max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 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"); - } + rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; + burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0; ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit); - - 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; @@ -1379,7 +1474,6 @@ bridge_reconfigure_controller(struct bridge *br) 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); } free(pfx); @@ -1387,23 +1481,21 @@ bridge_reconfigure_controller(struct bridge *br) } 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 @@ -1446,6 +1538,13 @@ bridge_fetch_dp_ifaces(struct bridge *br) 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); @@ -1468,13 +1567,31 @@ 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 @@ -1524,10 +1641,12 @@ 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; @@ -1564,6 +1683,12 @@ 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; @@ -1576,18 +1701,29 @@ 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(); } + + moving_active_iface = false; + port->bond_compat_is_stale = true; } static void @@ -1598,20 +1734,19 @@ 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_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); - } - } } } @@ -1639,18 +1774,6 @@ 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)); @@ -1737,13 +1860,12 @@ 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 */ @@ -1756,7 +1878,9 @@ 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++; } @@ -1773,14 +1897,28 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; if (port_includes_vlan(port, m->out_vlan) - && set_dst(dst, flow, in_port, port, tags) - && !dst_is_duplicate(dsts, dst - dsts, dst)) + && set_dst(dst, flow, in_port, port, tags)) { + int flow_vlan; + if (port->vlan < 0) { dst->vlan = m->out_vlan; } - if (dst->dp_ifidx == flow->in_port - && dst->vlan == vlan) { + if (dst_is_duplicate(dsts, dst - dsts, dst)) { + continue; + } + + /* Use the vlan tag on the original flow instead of + * the one passed in the vlan parameter. This ensures + * that we compare the vlan from before any implicit + * tagging tags place. This is necessary because + * dst->vlan is the final vlan, after removing implicit + * tags. */ + flow_vlan = ntohs(flow->dl_vlan); + if (flow_vlan == 0) { + flow_vlan = OFP_VLAN_NONE; + } + if (port == in_port && dst->vlan == flow_vlan) { /* Don't send out input port on same VLAN. */ continue; } @@ -1810,14 +1948,16 @@ 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++) { @@ -1836,14 +1976,77 @@ 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); + } +} + static bool -is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet) +is_bcast_arp_reply(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 + && eth_addr_is_broadcast(flow->dl_dst)); } /* If the composed actions may be applied to any packet in the given 'flow', @@ -1852,12 +2055,13 @@ is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet) static bool process_flow(struct bridge *br, const flow_t *flow, const struct ofpbuf *packet, struct odp_actions *actions, - tag_type *tags) + tag_type *tags, uint16_t *nf_output_iface) { struct iface *in_iface; struct port *in_port; struct port *out_port = NULL; /* By default, drop the packet/flow. */ int vlan; + int out_port_idx; /* Find the interface and port structure for the received packet. */ in_iface = iface_from_dp_ifidx(br, flow->in_port); @@ -1885,47 +2089,8 @@ process_flow(struct bridge *br, const flow_t *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; - } - } - - /* 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)) { + vlan = flow_get_vlan(br, flow, in_port, !!packet); + if (vlan < 0) { goto done; } @@ -1943,83 +2108,60 @@ process_flow(struct bridge *br, const flow_t *flow, goto done; } - /* Multicast (and broadcast) packets on bonds need special attention, to - * avoid receiving duplicates. */ - if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) { - *tags |= in_port->active_iface_tag; - if (in_port->active_iface != in_iface->port_ifidx) { - /* Drop all multicast packets on inactive slaves. */ - goto done; - } else { - /* Drop all multicast packets for which we have learned a different - * input port, because we probably sent the packet on one slaves - * and got it back on the active slave. Broadcast ARP replies are - * an exception to this rule: the host has moved to another - * switch. */ - int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); - if (src_idx != -1 && src_idx != in_port->port_idx) { - if (packet) { - if (!is_bcast_arp_reply(flow, packet)) { - goto done; - } - } else { - /* No way to know whether it's an ARP reply, because the - * flow entry doesn't include enough information and we - * don't have a packet. Punt. */ - return false; - } + /* Packets received on bonds need special attention to avoid duplicates. */ + if (in_port->n_ifaces > 1) { + int src_idx; + + if (eth_addr_is_multicast(flow->dl_dst)) { + *tags |= in_port->active_iface_tag; + if (in_port->active_iface != in_iface->port_ifidx) { + /* Drop all multicast packets on inactive slaves. */ + goto done; } } + + /* Drop all packets for which we have learned a different input + * port, because we probably sent the packet on one slave and got + * it back on the other. Broadcast ARP replies are an exception + * to this rule: the host has moved to another switch. */ + src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); + if (src_idx != -1 && src_idx != in_port->port_idx && + !is_bcast_arp_reply(flow)) { + 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); - } - } - - /* 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]; - } + /* 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; } - /* 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); + 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 @@ -2061,20 +2203,13 @@ 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 @@ -2084,17 +2219,30 @@ bridge_account_flow_ofhook_cb(const flow_t *flow, void *br_) { struct bridge *br = br_; + struct port *in_port; const union odp_action *a; + /* Feed information from the active flows back into the learning table + * to ensure that table is always in sync with what is actually flowing + * through the datapath. */ + in_port = port_from_dp_ifidx(br, flow->in_port); + if (in_port) { + int vlan = flow_get_vlan(br, flow, in_port, false); + if (vlan >= 0) { + update_learning_table(br, flow, vlan, in_port); + } + } + if (!br->has_bonded_ports) { return; } 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; } } @@ -2249,8 +2397,9 @@ 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; @@ -2268,12 +2417,11 @@ 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--; @@ -2358,22 +2506,60 @@ 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 @@ -2384,7 +2570,6 @@ bond_rebalance_port(struct port *port) } else { from++; } - port->bond_compat_is_stale = true; } } @@ -2403,17 +2588,14 @@ 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; } ofpbuf_init(&packet, 128); error = n_packets = n_errors = 0; LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { - static const char s[] = "Open vSwitch Bond Failover"; union ofp_action actions[2], *a; - struct eth_header *eth; - struct llc_snap_header *llc_snap; uint16_t dp_ifidx; tag_type tags = 0; flow_t flow; @@ -2424,23 +2606,6 @@ bond_send_learning_packets(struct port *port) continue; } - /* Compose packet to send. */ - ofpbuf_clear(&packet); - eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN); - llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN); - ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */ - ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN); - - memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN); - memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN); - eth->eth_type = htons(packet.size - ETH_HEADER_LEN); - - llc_snap->llc.llc_dsap = LLC_DSAP_SNAP; - llc_snap->llc.llc_ssap = LLC_SSAP_SNAP; - llc_snap->llc.llc_cntl = LLC_CNTL_SNAP; - memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3); - llc_snap->snap.snap_type = htons(0xf177); /* Random number. */ - /* Compose actions. */ memset(actions, 0, sizeof actions); a = actions; @@ -2457,6 +2622,8 @@ bond_send_learning_packets(struct port *port) /* Send packet. */ n_packets++; + compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, + e->mac); flow_extract(&packet, ODPP_NONE, &flow); retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, &packet); @@ -2481,7 +2648,8 @@ 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 UNUSED, void *aux UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct bridge *br; @@ -2531,7 +2699,8 @@ 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 UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct port *port; @@ -2572,14 +2741,10 @@ 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; @@ -2600,7 +2765,8 @@ 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 UNUSED) { char *args = (char *) args_; char *save_ptr = NULL; @@ -2656,7 +2822,8 @@ 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 UNUSED) { char *args = (char *) args_; char *save_ptr = NULL; @@ -2736,45 +2903,68 @@ 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 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 UNUSED) { enable_slave(conn, args, false); } +static void +bond_unixctl_hash(struct unixctl_conn *conn, const char *args, + void *aux 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, @@ -2784,70 +2974,69 @@ port_create(struct bridge *br, const char *name) VLOG_INFO("created port %s on bridge %s", port->name, br->name); bridge_flush(br); + + return port; } 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 old_ifaces, new_ifaces; + struct shash_node *node; unsigned long *trunks; int vlan; size_t i; + port->cfg = cfg; + /* Collect old and new interfaces. */ - svec_init(&old_ifaces); - svec_init(&new_ifaces); + shash_init(&old_ifaces); + shash_init(&new_ifaces); for (i = 0; i < port->n_ifaces; i++) { - svec_add(&old_ifaces, port->ifaces[i]->name); + shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]); } - 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; + for (i = 0; i < cfg->n_interfaces; i++) { + const char *name = cfg->interfaces[i]->name; + if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) { + VLOG_WARN("port %s: %s specified twice as port interface", + port->name, name); } - } else { - svec_init(&new_ifaces); - svec_add(&new_ifaces, port->name); + } + port->updelay = cfg->bond_updelay; + if (port->updelay < 0) { + port->updelay = 0; + } + port->updelay = cfg->bond_downdelay; + if (port->downdelay < 0) { + port->downdelay = 0; } /* 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); - } else { - i++; + SHASH_FOR_EACH (node, &old_ifaces) { + if (!shash_find(&new_ifaces, node->name)) { + iface_destroy(node->data); } } - 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); + SHASH_FOR_EACH (node, &new_ifaces) { + const struct ovsrec_interface *if_cfg = node->data; + struct iface *iface; + + iface = shash_find_data(&old_ifaces, if_cfg->name); + if (!iface) { + iface = iface_create(port, if_cfg); } + iface->cfg = if_cfg; } /* 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 @@ -2864,14 +3053,13 @@ port_reconfigure(struct port *port) /* Get trunked VLANs. */ trunks = NULL; if (vlan < 0) { - size_t n_trunks, n_errors; + 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 { @@ -2880,9 +3068,9 @@ port_reconfigure(struct port *port) } if (n_errors) { VLOG_ERR("port %s: invalid values for %zu trunk VLANs", - port->name, n_trunks); + port->name, cfg->n_trunks); } - if (n_errors == n_trunks) { + if (n_errors == cfg->n_trunks) { if (n_errors) { VLOG_ERR("port %s: no valid trunks, trunking all VLANs", port->name); @@ -2890,9 +3078,9 @@ port_reconfigure(struct port *port) bitmap_set_multiple(trunks, 0, 4096, 1); } } 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 (cfg->n_trunks) { + VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", + port->name); } } if (trunks == NULL @@ -2903,8 +3091,8 @@ port_reconfigure(struct port *port) bitmap_free(port->trunks); port->trunks = trunks; - svec_destroy(&old_ifaces); - svec_destroy(&new_ifaces); + shash_destroy(&old_ifaces); + shash_destroy(&new_ifaces); } static void @@ -2913,17 +3101,18 @@ port_destroy(struct port *port) if (port) { struct bridge *br = port->bridge; struct port *del; - size_t i; proc_net_compat_update_vlan(port->name, NULL, 0); proc_net_compat_update_bond(port->name, NULL); +#if 0 for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; if (m && m->out_port == port) { mirror_destroy(m); } } +#endif while (port->n_ifaces > 0) { iface_destroy(port->ifaces[port->n_ifaces - 1]); @@ -2984,6 +3173,7 @@ 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) { @@ -2999,6 +3189,7 @@ port_update_bonding(struct port *port) bond_choose_active_iface(port); } port->bond_compat_is_stale = true; + port->bond_fake_iface = port->cfg->bond_fake_iface; } } @@ -3037,14 +3228,40 @@ port_update_bond_compat(struct port *port) struct iface *iface = port->ifaces[i]; struct compat_bond_slave *slave = &bond.slaves[i]; slave->name = iface->name; - slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) || - (!iface->enabled && iface->delay_expires != LLONG_MAX)); + + /* We need to make the same determination as the Linux bonding + * code to determine whether a slave should be consider "up". + * The Linux function bond_miimon_inspect() supports four + * BOND_LINK_* states: + * + * - BOND_LINK_UP: carrier detected, updelay has passed. + * - BOND_LINK_FAIL: carrier lost, downdelay in progress. + * - BOND_LINK_DOWN: carrier lost, downdelay has passed. + * - BOND_LINK_BACK: carrier detected, updelay in progress. + * + * The function bond_info_show_slave() only considers BOND_LINK_UP + * to be "up" and anything else to be "down". + */ + slave->up = iface->enabled && iface->delay_expires == LLONG_MAX; if (slave->up) { bond.up = true; } netdev_get_etheraddr(iface->netdev, slave->mac); } + if (port->bond_fake_iface) { + struct netdev *bond_netdev; + + if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) { + if (bond.up) { + netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); + } else { + netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); + } + netdev_close(bond_netdev); + } + } + proc_net_compat_update_bond(port->name, &bond); free(bond.slaves); } @@ -3088,12 +3305,14 @@ 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 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); @@ -3111,9 +3330,19 @@ iface_create(struct port *port, const char *name) port->bridge->has_bonded_ports = true; } + /* Attempt to create the network interface in case it + * doesn't exist yet. */ + error = set_up_iface(if_cfg, true); + if (error) { + VLOG_WARN("could not create iface %s: %s\n", iface->name, + strerror(error)); + } + VLOG_DBG("attached network device %s to port %s", iface->name, port->name); bridge_flush(port->bridge); + + return iface; } static void @@ -3133,8 +3362,6 @@ iface_destroy(struct iface *iface) del->port_ifidx = iface->port_ifidx; netdev_close(iface->netdev); - free(iface->name); - free(iface); if (del_active) { ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); @@ -3142,6 +3369,10 @@ iface_destroy(struct iface *iface) bond_send_learning_packets(port); } + netdev_destroy(iface->name); + free(iface->name); + free(iface); + bridge_flush(port->bridge); } } @@ -3168,14 +3399,73 @@ iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) { return port_array_get(&br->ifaces, dp_ifidx); } + +/* Returns true if 'iface' is the name of an "internal" interface on bridge + * 'br', that is, an interface that is entirely simulated within the datapath. + * The local port (ODPP_LOCAL) is always an internal interface. Other local + * interfaces are created by setting "iface..internal = true". + * + * In addition, we have a kluge-y feature that creates an internal port with + * the name of a bonded port if "bonding..fake-iface = true" is set. + * This feature needs to go away in the long term. Until then, this is one + * 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 *if_name) +{ + /* XXX wastes time */ + struct iface *iface; + struct port *port; + + if (!strcmp(if_name, br->name)) { + 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->n_ifaces > 1 && port->cfg->bond_fake_iface) { + return true; + } + return false; +} + +/* Set Ethernet address of 'iface', if one is specified in the configuration + * file. */ +static void +iface_set_mac(struct iface *iface) +{ + uint8_t ea[ETH_ADDR_LEN]; + + 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); + } else if (iface->dp_ifidx == ODPP_LOCAL) { + VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead", + iface->name, iface->name); + } else { + int error = netdev_set_etheraddr(iface->netdev, ea); + if (error) { + VLOG_ERR("interface %s: setting MAC failed (%s)", + iface->name, strerror(error)); + } + } + } +} /* Port mirroring. */ +#if 0 static void -mirror_reconfigure(struct bridge *br) +mirror_reconfigure(struct bridge *br UNUSED) { struct svec old_mirrors, new_mirrors; - size_t i; + size_t i, n_rspan_vlans; + unsigned long *rspan_vlans; /* Collect old and new mirrors. */ svec_init(&old_mirrors); @@ -3224,6 +3514,29 @@ mirror_reconfigure(struct bridge *br) m->out_port->is_mirror_output_port = true; } } + + /* Update learning disabled vlans (for RSPAN). */ + rspan_vlans = NULL; + n_rspan_vlans = cfg_count("vlan.%s.disable-learning", br->name); + if (n_rspan_vlans) { + rspan_vlans = bitmap_allocate(4096); + + for (i = 0; i < n_rspan_vlans; i++) { + int vlan = cfg_get_vlan(i, "vlan.%s.disable-learning", br->name); + if (vlan >= 0) { + bitmap_set1(rspan_vlans, vlan); + VLOG_INFO("bridge %s: disabling learning on vlan %d\n", + br->name, vlan); + } else { + VLOG_ERR("bridge %s: invalid value '%s' for learning disabled " + "VLAN", br->name, + cfg_get_string(i, "vlan.%s.disable-learning", br->name)); + } + } + } + if (mac_learning_set_disabled_vlans(br->ml, rspan_vlans)) { + bridge_flush(br); + } } static void @@ -3246,7 +3559,7 @@ 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); @@ -3355,7 +3668,7 @@ 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 UNUSED) { char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name); struct svec src_ports, dst_ports, ports; @@ -3368,6 +3681,7 @@ mirror_reconfigure_one(struct mirror *m) 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", @@ -3406,11 +3720,18 @@ mirror_reconfigure_one(struct mirror *m) 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; + } /* Get all the vlans, and drop duplicate and invalid vlans. */ svec_init(&vlan_strings); @@ -3462,216 +3783,9 @@ 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 { - struct eth_header *eth = pkt->l2; - - netdev_get_etheraddr(iface->netdev, eth->eth_src); - if (eth_addr_is_zero(eth->eth_src)) { - VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " - "with unknown MAC", br->name, port_no); - } else { - union ofp_action action; - flow_t flow; - - 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); - } -} +#endif