X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=e3e5a4c717a25ef9ed57e7dd5724267c6360f3e3;hb=0fb7b9156ac7d319a51890630857698ad1877335;hp=54830dcd9d6a8b2b128dd72166df7ecb748fc1c1;hpb=9f5073d8a8c6c21c4e72660b8478a0e9d78764be;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 54830dcd9..e3e5a4c71 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -32,9 +32,11 @@ #include #include #include "bitmap.h" +#include "bond.h" #include "cfm.h" #include "classifier.h" #include "coverage.h" +#include "daemon.h" #include "dirs.h" #include "dpif.h" #include "dynamic-string.h" @@ -42,6 +44,7 @@ #include "hash.h" #include "hmap.h" #include "jsonrpc.h" +#include "lacp.h" #include "list.h" #include "mac-learning.h" #include "netdev.h" @@ -54,12 +57,12 @@ #include "ovsdb-data.h" #include "packets.h" #include "poll-loop.h" -#include "proc-net-compat.h" #include "process.h" #include "sha1.h" #include "shash.h" #include "socket-util.h" #include "stream-ssl.h" +#include "sset.h" #include "svec.h" #include "system-stats.h" #include "timeval.h" @@ -70,6 +73,7 @@ #include "xenserver.h" #include "vlog.h" #include "sflow_api.h" +#include "vlan-bitmap.h" VLOG_DEFINE_THIS_MODULE(bridge); @@ -77,15 +81,9 @@ COVERAGE_DEFINE(bridge_flush); COVERAGE_DEFINE(bridge_process_flow); COVERAGE_DEFINE(bridge_reconfigure); -enum lacp_status { - LACP_STATUS_CURRENT, /* Partner is up to date. */ - LACP_STATUS_EXPIRED, /* Partner is out of date. Attempt to re-sync. */ - LACP_STATUS_DEFAULTED /* Partner information is unknown. */ -}; - struct dst { + struct iface *iface; uint16_t vlan; - uint16_t dp_ifidx; }; struct dst_set { @@ -100,45 +98,18 @@ static void dst_set_free(struct dst_set *); struct iface { /* These members are always valid. */ + struct list port_elem; /* Element in struct port's "ifaces" list. */ struct port *port; /* Containing port. */ - size_t port_ifidx; /* Index within containing port. */ char *name; /* Host network device name. */ tag_type tag; /* Tag associated with this interface. */ - long long delay_expires; /* Time after which 'enabled' may change. */ /* These members are valid only after bridge_reconfigure() causes them to * be initialized. */ struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */ int dp_ifidx; /* Index within kernel datapath. */ struct netdev *netdev; /* Network device. */ - bool enabled; /* May be chosen for flows? */ - bool up; /* Is the interface up? */ const char *type; /* Usually same as cfg->type. */ - struct cfm *cfm; /* Connectivity Fault Management */ const struct ovsrec_interface *cfg; - - /* LACP information. */ - enum lacp_status lacp_status; /* LACP state machine status. */ - uint16_t lacp_priority; /* LACP port priority. */ - struct lacp_info lacp_actor; /* LACP actor information. */ - struct lacp_info lacp_partner; /* LACP partner information. */ - long long int lacp_tx; /* Next LACP message transmission time. */ - long long int lacp_rx; /* Next LACP message receive time. */ - bool lacp_attached; /* Attached to its aggregator? LACP allows - this link to be chosen for flows. */ -}; - -#define BOND_MASK 0xff -struct bond_entry { - int iface_idx; /* Index of assigned iface, or -1 if none. */ - uint64_t tx_bytes; /* Count of bytes recently transmitted. */ - tag_type iface_tag; /* Tag associated with iface_idx. */ -}; - -enum bond_mode { - BM_TCP, /* Transport Layer Load Balance. */ - BM_SLB, /* Source Load Balance. */ - BM_AB /* Active Backup. */ }; #define MAX_MIRRORS 32 @@ -152,8 +123,8 @@ struct mirror { struct uuid uuid; /* UUID of this "mirror" record in database. */ /* Selection criteria. */ - struct shash src_ports; /* Name is port name; data is always NULL. */ - struct shash dst_ports; /* Name is port name; data is always NULL. */ + struct sset src_ports; /* Source port names. */ + struct sset dst_ports; /* Destination port names. */ int *vlans; size_t n_vlans; @@ -162,50 +133,25 @@ struct mirror { int out_vlan; }; -/* Flags for a port's lacp member. */ -#define LACP_ACTIVE 0x01 /* LACP is in active mode. */ -#define LACP_PASSIVE 0x02 /* LACP is in passive mode. */ -#define LACP_NEGOTIATED 0x04 /* LACP has successfully negotiated. */ - #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ struct port { struct bridge *bridge; - size_t port_idx; + struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */ + char *name; + int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. * NULL if all VLANs are trunked. */ const struct ovsrec_port *cfg; - char *name; /* An ordinary bridge port has 1 interface. * A bridge port for bonding has at least 2 interfaces. */ - struct iface **ifaces; - size_t n_ifaces, allocated_ifaces; + struct list ifaces; /* List of "struct iface"s. */ + + struct lacp *lacp; /* NULL if LACP is not enabled. */ /* Bonding info. */ - enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */ - int active_iface; /* Ifidx on which bcasts accepted, or -1. */ - tag_type active_iface_tag; /* Tag for bcast flows. */ - tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */ - int updelay, downdelay; /* Delay before iface goes up/down, in ms. */ - bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */ - bool bond_fake_iface; /* Fake a bond interface for legacy compat? */ - bool miimon; /* Use miimon instead of carrier? */ - long long int bond_miimon_interval; /* Miimon status refresh interval. */ - long long int bond_miimon_next_update; /* Time of next miimon update. */ - long long int bond_next_fake_iface_update; /* Time of next update. */ - struct netdev_monitor *monitor; /* Tracks carrier up/down status. */ - - /* LACP information. */ - int lacp; /* LACP status flags. 0 if LACP is off. */ - uint16_t lacp_key; /* LACP aggregation key. */ - uint16_t lacp_priority; /* LACP system priority. */ - bool lacp_need_update; /* Need to update attached interfaces? */ - - /* SLB specific bonding info. */ - struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ - int bond_rebalance_interval; /* Interval between rebalances, in ms. */ - long long int bond_next_rebalance; /* Next rebalancing time. */ + struct bond *bond; /* Port mirroring info. */ mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ @@ -226,13 +172,11 @@ struct bridge { /* Kernel datapath information. */ struct dpif *dpif; /* Datapath. */ - struct hmap ifaces; /* Contains "struct iface"s. */ + struct hmap ifaces; /* "struct iface"s indexed by dp_ifidx. */ /* Bridge ports. */ - struct port **ports; - size_t n_ports, allocated_ports; + struct hmap ports; /* "struct port"s indexed by name. */ struct shash iface_by_name; /* "struct iface"s indexed by name. */ - struct shash port_by_name; /* "struct port"s indexed by name. */ /* Bonding. */ bool has_bonded_ports; @@ -242,6 +186,11 @@ struct bridge { /* Port mirroring. */ struct mirror *mirrors[MAX_MIRRORS]; + + /* Synthetic local port if necessary. */ + struct ovsrec_port synth_local_port; + struct ovsrec_interface synth_local_iface; + struct ovsrec_interface *synth_local_ifacep; }; /* List of all bridges. */ @@ -255,6 +204,12 @@ static struct ovsdb_idl *idl; #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */ static long long int stats_timer = LLONG_MIN; +/* Stores the time after which rate limited statistics may be written to the + * database. Only updated when changes to the database require rate limiting. + */ +#define DB_LIMIT_INTERVAL (1 * 1000) /* In milliseconds. */ +static long long int db_limiter = LLONG_MIN; + static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg); static void bridge_destroy(struct bridge *); static struct bridge *bridge_lookup(const char *name); @@ -276,34 +231,25 @@ static void bridge_pick_local_hw_addr(struct bridge *, static uint64_t bridge_pick_datapath_id(struct bridge *, const uint8_t bridge_ea[ETH_ADDR_LEN], struct iface *hw_addr_iface); -static struct iface *bridge_get_local_iface(struct bridge *); static uint64_t dpid_from_hash(const void *, size_t nbytes); static unixctl_cb_func bridge_unixctl_fdb_show; +static unixctl_cb_func cfm_unixctl_show; +static unixctl_cb_func qos_unixctl_show; -static void lacp_run(struct bridge *); -static void lacp_wait(struct bridge *); -static void lacp_process_packet(const struct ofpbuf *, struct iface *); - -static void bond_init(void); -static void bond_run(struct bridge *); -static void bond_wait(struct bridge *); -static void bond_rebalance_port(struct port *); -static void bond_send_learning_packets(struct port *); -static void bond_enable_slave(struct iface *iface, bool enable); - +static void port_run(struct port *); +static void port_wait(struct port *); static struct port *port_create(struct bridge *, const char *name); static void port_reconfigure(struct port *, const struct ovsrec_port *); static void port_del_ifaces(struct port *, const struct ovsrec_port *); static void port_destroy(struct port *); static struct port *port_lookup(const struct bridge *, const char *name); -static struct iface *port_lookup_iface(const struct port *, const char *name); +static struct iface *port_get_an_iface(const struct port *); static struct port *port_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); -static void port_update_bond_compat(struct port *); -static void port_update_vlan_compat(struct port *); -static void port_update_bonding(struct port *); -static void port_update_lacp(struct port *); +static void port_reconfigure_lacp(struct port *); +static void port_reconfigure_bond(struct port *); +static void port_send_learning_packets(struct port *); static void mirror_create(struct bridge *, struct ovsrec_mirror *); static void mirror_destroy(struct mirror *); @@ -315,25 +261,22 @@ static struct iface *iface_create(struct port *port, const struct ovsrec_interface *if_cfg); static void iface_destroy(struct iface *); static struct iface *iface_lookup(const struct bridge *, const char *name); +static struct iface *iface_find(const char *name); static struct iface *iface_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); static void iface_set_mac(struct iface *); static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport); static void iface_update_qos(struct iface *, const struct ovsrec_qos *); static void iface_update_cfm(struct iface *); -static void iface_refresh_cfm_stats(struct iface *iface); -static void iface_send_packet(struct iface *, struct ofpbuf *packet); -static uint8_t iface_get_lacp_state(const struct iface *); -static void iface_get_lacp_priority(struct iface *, struct lacp_info *); -static void iface_set_lacp_defaulted(struct iface *); -static void iface_set_lacp_expired(struct iface *); +static bool iface_refresh_cfm_stats(struct iface *iface); +static bool iface_get_carrier(const struct iface *); +static bool iface_is_synthetic(const struct iface *); static void shash_from_ovs_idl_map(char **keys, char **values, size_t n, struct shash *); static void shash_to_ovs_idl_map(struct shash *, char ***keys, char ***values, size_t *n); - /* Hooks into ofproto processing. */ static struct ofhooks bridge_ofhooks; @@ -351,22 +294,63 @@ bridge_init(const char *remote) ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg); ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics); ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids); + ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_ovs_version); + ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_db_version); + ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_system_type); + ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_system_version); + ovsdb_idl_omit_alert(idl, &ovsrec_bridge_col_datapath_id); ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids); ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_admin_state); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_duplex); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_link_speed); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_link_state); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_mtu); ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport); ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_status); ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids); + ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_is_connected); + ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_role); + ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_status); + ovsdb_idl_omit(idl, &ovsrec_controller_col_external_ids); + + ovsdb_idl_omit_alert(idl, &ovsrec_maintenance_point_col_fault); + + ovsdb_idl_omit_alert(idl, &ovsrec_monitor_col_fault); + + ovsdb_idl_omit(idl, &ovsrec_qos_col_external_ids); + + ovsdb_idl_omit(idl, &ovsrec_queue_col_external_ids); + + ovsdb_idl_omit(idl, &ovsrec_mirror_col_external_ids); + + ovsdb_idl_omit(idl, &ovsrec_netflow_col_external_ids); + + ovsdb_idl_omit(idl, &ovsrec_sflow_col_external_ids); + + ovsdb_idl_omit(idl, &ovsrec_manager_col_external_ids); + ovsdb_idl_omit(idl, &ovsrec_manager_col_inactivity_probe); + ovsdb_idl_omit(idl, &ovsrec_manager_col_is_connected); + ovsdb_idl_omit(idl, &ovsrec_manager_col_max_backoff); + ovsdb_idl_omit(idl, &ovsrec_manager_col_status); + + ovsdb_idl_omit(idl, &ovsrec_ssl_col_external_ids); + /* Register unixctl commands. */ unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + unixctl_command_register("cfm/show", cfm_unixctl_show, NULL); + unixctl_command_register("qos/show", qos_unixctl_show, NULL); unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, NULL); unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect, NULL); + lacp_init(); bond_init(); } @@ -387,8 +371,9 @@ static void bridge_configure_once(const struct ovsrec_open_vswitch *cfg) { static bool already_configured_once; - struct svec bridge_names; - struct svec dpif_names, dpif_types; + struct sset bridge_names; + struct sset dpif_names, dpif_types; + const char *type; size_t i; /* Only do this once per ovs-vswitchd run. */ @@ -400,30 +385,28 @@ bridge_configure_once(const struct ovsrec_open_vswitch *cfg) stats_timer = time_msec() + STATS_INTERVAL; /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */ - svec_init(&bridge_names); + sset_init(&bridge_names); for (i = 0; i < cfg->n_bridges; i++) { - svec_add(&bridge_names, cfg->bridges[i]->name); + sset_add(&bridge_names, cfg->bridges[i]->name); } - svec_sort(&bridge_names); /* Iterate over all system dpifs and delete any of them that do not appear * in 'cfg'. */ - svec_init(&dpif_names); - svec_init(&dpif_types); + sset_init(&dpif_names); + sset_init(&dpif_types); dp_enumerate_types(&dpif_types); - for (i = 0; i < dpif_types.n; i++) { - size_t j; + SSET_FOR_EACH (type, &dpif_types) { + const char *name; - dp_enumerate_names(dpif_types.names[i], &dpif_names); + dp_enumerate_names(type, &dpif_names); /* Delete each dpif whose name is not in 'bridge_names'. */ - for (j = 0; j < dpif_names.n; j++) { - if (!svec_contains(&bridge_names, dpif_names.names[j])) { + SSET_FOR_EACH (name, &dpif_names) { + if (!sset_contains(&bridge_names, name)) { struct dpif *dpif; int retval; - retval = dpif_open(dpif_names.names[j], dpif_types.names[i], - &dpif); + retval = dpif_open(name, type, &dpif); if (!retval) { dpif_delete(dpif); dpif_close(dpif); @@ -431,9 +414,9 @@ bridge_configure_once(const struct ovsrec_open_vswitch *cfg) } } } - svec_destroy(&bridge_names); - svec_destroy(&dpif_names); - svec_destroy(&dpif_types); + sset_destroy(&bridge_names); + sset_destroy(&dpif_names); + sset_destroy(&dpif_types); } /* Callback for iterate_and_prune_ifaces(). */ @@ -468,9 +451,7 @@ set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, /* Set MAC address of internal interfaces other than the local * interface. */ - if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) { - iface_set_mac(iface); - } + iface_set_mac(iface); return true; } @@ -484,24 +465,20 @@ iterate_and_prune_ifaces(struct bridge *br, void *aux), void *aux) { - size_t i, j; - - for (i = 0; i < br->n_ports; ) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; ) { - struct iface *iface = port->ifaces[j]; - if (cb(br, iface, aux)) { - j++; - } else { + struct port *port, *next_port; + + HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) { + struct iface *iface, *next_iface; + + LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) { + if (!cb(br, iface, aux)) { iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } } - if (port->n_ifaces) { - i++; - } else { - VLOG_ERR("%s port has no interfaces, dropping", port->name); + if (list_is_empty(&port->ifaces)) { + VLOG_WARN("%s port has no interfaces, dropping", port->name); port_destroy(port); } } @@ -524,33 +501,29 @@ collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg, { struct sockaddr_in *managers = NULL; size_t n_managers = 0; - struct shash targets; + struct sset targets; size_t i; - /* Collect all of the potential targets, as the union of the "managers" - * column and the "targets" columns of the rows pointed to by - * "manager_options", excluding any that are out-of-band. */ - shash_init(&targets); - for (i = 0; i < ovs_cfg->n_managers; i++) { - shash_add_once(&targets, ovs_cfg->managers[i], NULL); - } + /* Collect all of the potential targets from the "targets" columns of the + * rows pointed to by "manager_options", excluding any that are + * out-of-band. */ + sset_init(&targets); for (i = 0; i < ovs_cfg->n_manager_options; i++) { struct ovsrec_manager *m = ovs_cfg->manager_options[i]; if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) { - shash_find_and_delete(&targets, m->target); + sset_find_and_delete(&targets, m->target); } else { - shash_add_once(&targets, m->target, NULL); + sset_add(&targets, m->target); } } /* Now extract the targets' IP addresses. */ - if (!shash_is_empty(&targets)) { - struct shash_node *node; + if (!sset_is_empty(&targets)) { + const char *target; - managers = xmalloc(shash_count(&targets) * sizeof *managers); - SHASH_FOR_EACH (node, &targets) { - const char *target = node->name; + managers = xmalloc(sset_count(&targets) * sizeof *managers); + SSET_FOR_EACH (target, &targets) { struct sockaddr_in *sin = &managers[n_managers]; if ((!strncmp(target, "tcp:", 4) @@ -561,7 +534,7 @@ collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg, } } } - shash_destroy(&targets); + sset_destroy(&targets); *managersp = managers; *n_managersp = n_managers; @@ -643,9 +616,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) && strcmp(dpif_port.name, br->name)) { int retval = dpif_port_del(br->dpif, dpif_port.port_no); if (retval) { - VLOG_ERR("failed to remove %s interface from %s: %s", - dpif_port.name, dpif_name(br->dpif), - strerror(retval)); + VLOG_WARN("failed to remove %s interface from %s: %s", + dpif_port.name, dpif_name(br->dpif), + strerror(retval)); } } } @@ -692,6 +665,13 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) dpif_port = NULL; } if (iface) { + if (iface->port->bond) { + /* The bond has a pointer to the netdev, so remove it + * from the bond before closing the netdev. The slave + * will get added back to the bond later, after a new + * netdev is available. */ + bond_slave_unregister(iface->port->bond, iface); + } netdev_close(iface->netdev); iface->netdev = NULL; } @@ -735,9 +715,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) dpif_name(br->dpif)); break; } else { - VLOG_ERR("failed to add %s interface to %s: %s", - if_name, dpif_name(br->dpif), - strerror(error)); + VLOG_WARN("failed to add %s interface to %s: %s", + if_name, dpif_name(br->dpif), + strerror(error)); continue; } } @@ -746,8 +726,6 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) /* Update 'iface'. */ if (iface) { iface->netdev = netdev; - iface->enabled = netdev_get_carrier(iface->netdev); - iface->up = iface->enabled; } } else if (iface && iface->netdev) { struct shash args; @@ -771,7 +749,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } sflow_bridge_number = 0; LIST_FOR_EACH (br, node, &all_bridges) { - uint8_t ea[8]; + uint8_t ea[ETH_ADDR_LEN]; uint64_t dpid; struct iface *local_iface; struct iface *hw_addr_iface; @@ -779,11 +757,15 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) bridge_fetch_dp_ifaces(br); + /* Delete interfaces that cannot be opened. + * + * From this point forward we are guaranteed that every "struct iface" + * has nonnull 'netdev' and correct 'dp_ifidx'. */ iterate_and_prune_ifaces(br, check_iface, NULL); /* Pick local port hardware address, datapath ID. */ bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); - local_iface = bridge_get_local_iface(br); + local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL); if (local_iface) { int error = netdev_set_etheraddr(local_iface->netdev, ea); if (error) { @@ -834,19 +816,21 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) "with another vswitch, choose an engine id less " "than 128", br->name); } - if (br->n_ports > 508) { + if (hmap_count(&br->ports) > 508) { VLOG_WARN("bridge %s: netflow port mangling will conflict " "with another port when more than 508 ports are " "used", br->name); } } - opts.collectors.n = nf_cfg->n_targets; - opts.collectors.names = nf_cfg->targets; + sset_init(&opts.collectors); + sset_add_array(&opts.collectors, + nf_cfg->targets, nf_cfg->n_targets); if (ofproto_set_netflow(br->ofproto, &opts)) { VLOG_ERR("bridge %s: problem setting netflow collectors", br->name); } + sset_destroy(&opts.collectors); } else { ofproto_set_netflow(br->ofproto, NULL); } @@ -860,8 +844,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) memset(&oso, 0, sizeof oso); - oso.targets.n = sflow_cfg->n_targets; - oso.targets.names = sflow_cfg->targets; + sset_init(&oso.targets); + sset_add_array(&oso.targets, + sflow_cfg->targets, sflow_cfg->n_targets); oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE; if (sflow_cfg->sampling) { @@ -891,7 +876,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } ofproto_set_sflow(br->ofproto, &oso); - /* Do not destroy oso.targets because it is owned by sflow_cfg. */ + sset_destroy(&oso.targets); } else { ofproto_set_sflow(br->ofproto, NULL); } @@ -908,16 +893,17 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) bridge_reconfigure_remotes(br, managers, n_managers); } LIST_FOR_EACH (br, node, &all_bridges) { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - int j; + struct port *port; - port_update_vlan_compat(port); - port_update_bonding(port); - port_update_lacp(port); + br->has_bonded_ports = false; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; - for (j = 0; j < port->n_ifaces; j++) { - iface_update_qos(port->ifaces[j], port->cfg->qos); + port_reconfigure_lacp(port); + port_reconfigure_bond(port); + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_update_qos(iface, port->cfg->qos); } } } @@ -925,14 +911,23 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) iterate_and_prune_ifaces(br, set_iface_properties, NULL); } + /* Some reconfiguration operations require the bridge to have been run at + * least once. */ LIST_FOR_EACH (br, node, &all_bridges) { struct iface *iface; + + bridge_run_one(br); + HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { iface_update_cfm(iface); } } free(managers); + + /* ovs-vswitchd has completed initialization, so allow the process that + * forked us to exit successfully. */ + daemonize_complete(); } static const char * @@ -962,7 +957,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], struct iface **hw_addr_iface) { const char *hwaddr; - size_t i, j; + struct port *port; int error; *hw_addr_iface = NULL; @@ -982,10 +977,10 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], /* Otherwise choose the minimum non-local MAC address among all of the * interfaces. */ - memset(ea, 0xff, sizeof ea); - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + memset(ea, 0xff, ETH_ADDR_LEN); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { uint8_t iface_ea[ETH_ADDR_LEN]; + struct iface *candidate; struct iface *iface; /* Mirror output ports don't participate. */ @@ -994,12 +989,11 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], } /* Choose the MAC address to represent the port. */ + iface = NULL; if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) { /* Find the interface with this Ethernet address (if any) so that * we can provide the correct devname to the caller. */ - iface = NULL; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *candidate = port->ifaces[j]; + LIST_FOR_EACH (candidate, port_elem, &port->ifaces) { uint8_t candidate_ea[ETH_ADDR_LEN]; if (!netdev_get_etheraddr(candidate->netdev, candidate_ea) && eth_addr_equals(iface_ea, candidate_ea)) { @@ -1013,10 +1007,8 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], * scripts always add slaves to a bond in alphabetical order, so * for compatibility we choose the interface with the name that is * first in alphabetical order. */ - iface = port->ifaces[0]; - for (j = 1; j < port->n_ifaces; j++) { - struct iface *candidate = port->ifaces[j]; - if (strcmp(candidate->name, iface->name) < 0) { + LIST_FOR_EACH (candidate, port_elem, &port->ifaces) { + if (!iface || strcmp(candidate->name, iface->name) < 0) { iface = candidate; } } @@ -1162,6 +1154,10 @@ iface_refresh_status(struct iface *iface) int64_t mtu_64; int error; + if (iface_is_synthetic(iface)) { + return; + } + shash_init(&sh); if (!netdev_get_status(iface->netdev, &sh)) { @@ -1203,11 +1199,10 @@ iface_refresh_status(struct iface *iface) ovsrec_interface_set_link_state(iface->cfg, - netdev_get_carrier(iface->netdev) - ? "up" : "down"); + iface_get_carrier(iface) ? "up" : "down"); error = netdev_get_mtu(iface->netdev, &mtu); - if (!error) { + if (!error && mtu != INT_MAX) { mtu_64 = mtu; ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1); } @@ -1216,18 +1211,21 @@ iface_refresh_status(struct iface *iface) } } -static void +/* Writes 'iface''s CFM statistics to the database. Returns true if anything + * changed, false otherwise. */ +static bool iface_refresh_cfm_stats(struct iface *iface) { - size_t i; - struct cfm *cfm; const struct ovsrec_monitor *mon; + const struct cfm *cfm; + bool changed = false; + size_t i; mon = iface->cfg->monitor; - cfm = iface->cfm; + cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); if (!cfm || !mon) { - return; + return false; } for (i = 0; i < mon->n_remote_mps; i++) { @@ -1237,59 +1235,39 @@ iface_refresh_cfm_stats(struct iface *iface) mp = mon->remote_mps[i]; rmp = cfm_get_remote_mp(cfm, mp->mpid); - ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1); - } - - if (hmap_is_empty(&cfm->x_remote_mps)) { - ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0); - } else { - size_t length; - struct remote_mp *rmp; - int64_t *x_remote_mps; - - length = hmap_count(&cfm->x_remote_mps); - x_remote_mps = xzalloc(length * sizeof *x_remote_mps); - - i = 0; - HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) { - x_remote_mps[i++] = rmp->mpid; + if (mp->n_fault != 1 || mp->fault[0] != rmp->fault) { + ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1); + changed = true; } - - ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length); - free(x_remote_mps); } - if (hmap_is_empty(&cfm->x_remote_maids)) { - ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0); - } else { - size_t length; - char **x_remote_maids; - struct remote_maid *rmaid; - - length = hmap_count(&cfm->x_remote_maids); - x_remote_maids = xzalloc(length * sizeof *x_remote_maids); + if (mon->n_fault != 1 || mon->fault[0] != cfm->fault) { + ovsrec_monitor_set_fault(mon, &cfm->fault, 1); + changed = true; + } - i = 0; - HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) { - size_t j; + return changed; +} - x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1); +static bool +iface_refresh_lacp_stats(struct iface *iface) +{ + bool *db_current = iface->cfg->lacp_current; + bool changed = false; - for (j = 0; j < CCM_MAID_LEN; j++) { - snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx", - rmaid->maid[j]); - } - i++; - } - ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length); + if (iface->port->lacp) { + bool current = lacp_slave_is_current(iface->port->lacp, iface); - for (i = 0; i < length; i++) { - free(x_remote_maids[i]); + if (!db_current || *db_current != current) { + changed = true; + ovsrec_interface_set_lacp_current(iface->cfg, ¤t, 1); } - free(x_remote_maids); + } else if (db_current) { + changed = true; + ovsrec_interface_set_lacp_current(iface->cfg, NULL, 0); } - ovsrec_monitor_set_fault(mon, &cfm->fault, 1); + return changed; } static void @@ -1322,6 +1300,10 @@ iface_refresh_stats(struct iface *iface) struct netdev_stats stats; + if (iface_is_synthetic(iface)) { + return; + } + /* Intentionally ignore return value, since errors will set 'stats' to * all-1s, and we will deal with that correctly below. */ netdev_get_stats(iface->netdev, &stats); @@ -1460,16 +1442,13 @@ bridge_run(void) txn = ovsdb_idl_txn_create(idl); LIST_FOR_EACH (br, node, &all_bridges) { - size_t i; + struct port *port; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - size_t j; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { iface_refresh_stats(iface); - iface_refresh_cfm_stats(iface); iface_refresh_status(iface); } } @@ -1482,32 +1461,54 @@ bridge_run(void) stats_timer = time_msec() + STATS_INTERVAL; } + + if (time_msec() >= db_limiter) { + struct ovsdb_idl_txn *txn; + bool changed = false; + + txn = ovsdb_idl_txn_create(idl); + LIST_FOR_EACH (br, node, &all_bridges) { + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + changed = iface_refresh_cfm_stats(iface) || changed; + changed = iface_refresh_lacp_stats(iface) || changed; + } + } + } + + if (changed) { + db_limiter = time_msec() + DB_LIMIT_INTERVAL; + } + + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); + } } void bridge_wait(void) { struct bridge *br; - struct iface *iface; LIST_FOR_EACH (br, node, &all_bridges) { - ofproto_wait(br->ofproto); - if (ofproto_has_primary_controller(br->ofproto)) { - continue; - } + struct port *port; + ofproto_wait(br->ofproto); mac_learning_wait(br->ml); - lacp_wait(br); - bond_wait(br); - - HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { - if (iface->cfm) { - cfm_wait(iface->cfm); - } + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port_wait(port); } } ovsdb_idl_wait(idl); poll_timer_wait_until(stats_timer); + + if (db_limiter > time_msec()) { + poll_timer_wait_until(db_limiter); + } } /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s @@ -1517,27 +1518,6 @@ bridge_flush(struct bridge *br) { COVERAGE_INC(bridge_flush); br->flush = true; - mac_learning_flush(br->ml); -} - -/* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no - * such interface. */ -static struct iface * -bridge_get_local_iface(struct bridge *br) -{ - size_t i, j; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->dp_ifidx == ODPP_LOCAL) { - return iface; - } - } - } - - return NULL; } /* Bridge unixctl user interface functions. */ @@ -1557,17 +1537,135 @@ bridge_unixctl_fdb_show(struct unixctl_conn *conn, ds_put_cstr(&ds, " port VLAN MAC Age\n"); LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { - if (e->port < 0 || e->port >= br->n_ports) { - continue; - } + struct port *port = e->port.p; ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", - br->ports[e->port]->ifaces[0]->dp_ifidx, + port_get_an_iface(port)->dp_ifidx, e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); } unixctl_command_reply(conn, 200, ds_cstr(&ds)); ds_destroy(&ds); } +/* CFM unixctl user interface functions. */ +static void +cfm_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct iface *iface; + const struct cfm *cfm; + + iface = iface_find(args); + if (!iface) { + unixctl_command_reply(conn, 501, "no such interface"); + return; + } + + cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); + + if (!cfm) { + unixctl_command_reply(conn, 501, "CFM not enabled"); + return; + } + + cfm_dump_ds(cfm, &ds); + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + ds_destroy(&ds); +} + +/* QoS unixctl user interface functions. */ + +struct qos_unixctl_show_cbdata { + struct ds *ds; + struct iface *iface; +}; + +static void +qos_unixctl_show_cb(unsigned int queue_id, + const struct shash *details, + void *aux) +{ + struct qos_unixctl_show_cbdata *data = aux; + struct ds *ds = data->ds; + struct iface *iface = data->iface; + struct netdev_queue_stats stats; + struct shash_node *node; + int error; + + ds_put_cstr(ds, "\n"); + if (queue_id) { + ds_put_format(ds, "Queue %u:\n", queue_id); + } else { + ds_put_cstr(ds, "Default:\n"); + } + + SHASH_FOR_EACH (node, details) { + ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data); + } + + error = netdev_get_queue_stats(iface->netdev, queue_id, &stats); + if (!error) { + if (stats.tx_packets != UINT64_MAX) { + ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets); + } + + if (stats.tx_bytes != UINT64_MAX) { + ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes); + } + + if (stats.tx_errors != UINT64_MAX) { + ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors); + } + } else { + ds_put_format(ds, "\tFailed to get statistics for queue %u: %s", + queue_id, strerror(error)); + } +} + +static void +qos_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct shash sh = SHASH_INITIALIZER(&sh); + struct iface *iface; + const char *type; + struct shash_node *node; + struct qos_unixctl_show_cbdata data; + int error; + + iface = iface_find(args); + if (!iface) { + unixctl_command_reply(conn, 501, "no such interface"); + return; + } + + netdev_get_qos(iface->netdev, &type, &sh); + + if (*type != '\0') { + ds_put_format(&ds, "QoS: %s %s\n", iface->name, type); + + SHASH_FOR_EACH (node, &sh) { + ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data); + } + + data.ds = &ds; + data.iface = iface; + error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data); + + if (error) { + ds_put_format(&ds, "failed to dump queues: %s", strerror(error)); + } + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + } else { + ds_put_format(&ds, "QoS not configured on %s\n", iface->name); + unixctl_command_reply(conn, 501, ds_cstr(&ds)); + } + + shash_destroy_free_data(&sh); + ds_destroy(&ds); +} + /* Bridge reconfiguration functions. */ static struct bridge * bridge_create(const struct ovsrec_bridge *br_cfg) @@ -1584,7 +1682,6 @@ bridge_create(const struct ovsrec_bridge *br_cfg) free(br); return NULL; } - dpif_flow_flush(br->dpif); error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks, br, &br->ofproto); @@ -1602,9 +1699,8 @@ bridge_create(const struct ovsrec_bridge *br_cfg) br->ml = mac_learning_create(); eth_addr_nicira_random(br->default_ea); + hmap_init(&br->ports); hmap_init(&br->ifaces); - - shash_init(&br->port_by_name); shash_init(&br->iface_by_name); br->flush = false; @@ -1620,24 +1716,29 @@ static void bridge_destroy(struct bridge *br) { if (br) { + struct port *port, *next; int error; + int i; - while (br->n_ports > 0) { - port_destroy(br->ports[br->n_ports - 1]); + HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) { + port_destroy(port); + } + for (i = 0; i < MAX_MIRRORS; i++) { + mirror_destroy(br->mirrors[i]); } list_remove(&br->node); + ofproto_destroy(br->ofproto); error = dpif_delete(br->dpif); if (error && error != ENOENT) { VLOG_ERR("failed to delete %s: %s", dpif_name(br->dpif), strerror(error)); } dpif_close(br->dpif); - ofproto_destroy(br->ofproto); mac_learning_destroy(br->ml); hmap_destroy(&br->ifaces); - shash_destroy(&br->port_by_name); + hmap_destroy(&br->ports); shash_destroy(&br->iface_by_name); - free(br->ports); + free(br->synth_local_iface.type); free(br->name); free(br); } @@ -1704,8 +1805,8 @@ bridge_unixctl_reconnect(struct unixctl_conn *conn, static int bridge_run_one(struct bridge *br) { + struct port *port; int error; - struct iface *iface; error = ofproto_run1(br->ofproto); if (error) { @@ -1713,27 +1814,14 @@ bridge_run_one(struct bridge *br) } mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); - lacp_run(br); - bond_run(br); + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port_run(port); + } error = ofproto_run2(br->ofproto, br->flush); br->flush = false; - HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { - struct ofpbuf *packet; - - if (!iface->cfm) { - continue; - } - - packet = cfm_run(iface->cfm); - if (packet) { - iface_send_packet(iface, packet); - ofpbuf_uninit(packet); - free(packet); - } - } - return error; } @@ -1761,18 +1849,12 @@ bridge_get_controllers(const struct bridge *br, static void bridge_reconfigure_one(struct bridge *br) { - struct shash old_ports, new_ports; - struct svec snoops, old_snoops; - struct shash_node *node; enum ofproto_fail_mode fail_mode; + struct port *port, *next; + struct shash_node *node; + struct shash new_ports; size_t i; - /* Collect old ports. */ - shash_init(&old_ports); - for (i = 0; i < br->n_ports; i++) { - shash_add(&old_ports, br->ports[i]->name, br->ports[i]); - } - /* Collect new ports. */ shash_init(&new_ports); for (i = 0; i < br->cfg->n_ports; i++) { @@ -1782,31 +1864,36 @@ bridge_reconfigure_one(struct bridge *br) br->name, name); } } + if (!shash_find(&new_ports, br->name)) { + struct dpif_port dpif_port; + char *type; - /* 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(br, NULL)) { - char local_name[IF_NAMESIZE]; - int error; + VLOG_WARN("bridge %s: no port named %s, synthesizing one", + br->name, br->name); - 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); - } + dpif_port_query_by_number(br->dpif, ODPP_LOCAL, &dpif_port); + type = xstrdup(dpif_port.type ? dpif_port.type : "internal"); + dpif_port_destroy(&dpif_port); + + br->synth_local_port.interfaces = &br->synth_local_ifacep; + br->synth_local_port.n_interfaces = 1; + br->synth_local_port.name = br->name; + + br->synth_local_iface.name = br->name; + free(br->synth_local_iface.type); + br->synth_local_iface.type = type; + + br->synth_local_ifacep = &br->synth_local_iface; + + shash_add(&new_ports, br->name, &br->synth_local_port); } /* Get rid of deleted ports. * Get rid of deleted interfaces on ports that still exist. */ - SHASH_FOR_EACH (node, &old_ports) { - struct port *port = node->data; + HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) { const struct ovsrec_port *port_cfg; - port_cfg = shash_find_data(&new_ports, node->name); + port_cfg = shash_find_data(&new_ports, port->name); if (!port_cfg) { port_destroy(port); } else { @@ -1818,19 +1905,18 @@ bridge_reconfigure_one(struct bridge *br) * Add new interfaces to existing ports. * Reconfigure existing ports. */ SHASH_FOR_EACH (node, &new_ports) { - struct port *port = shash_find_data(&old_ports, node->name); + struct port *port = port_lookup(br, node->name); if (!port) { port = port_create(br, node->name); } port_reconfigure(port, node->data); - if (!port->n_ifaces) { + if (list_is_empty(&port->ifaces)) { VLOG_WARN("bridge %s: port %s has no interfaces, dropping", br->name, port->name); port_destroy(port); } } - shash_destroy(&old_ports); shash_destroy(&new_ports); /* Set the fail-mode */ @@ -1838,27 +1924,18 @@ bridge_reconfigure_one(struct bridge *br) || !strcmp(br->cfg->fail_mode, "standalone") ? OFPROTO_FAIL_STANDALONE : OFPROTO_FAIL_SECURE; - if (ofproto_get_fail_mode(br->ofproto) != fail_mode - && !ofproto_has_primary_controller(br->ofproto)) { - ofproto_flush_flows(br->ofproto); - } ofproto_set_fail_mode(br->ofproto, fail_mode); - /* Delete all flows if we're switching from connected to standalone or vice - * versa. (XXX Should we delete all flows if we are switching from one - * controller to another?) */ - /* Configure OpenFlow controller connection snooping. */ - svec_init(&snoops); - svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir(), br->name)); - svec_init(&old_snoops); - ofproto_get_snoops(br->ofproto, &old_snoops); - if (!svec_equal(&snoops, &old_snoops)) { + if (!ofproto_has_snoops(br->ofproto)) { + struct sset snoops; + + sset_init(&snoops); + sset_add_and_free(&snoops, xasprintf("punix:%s/%s.snoop", + ovs_rundir(), br->name)); ofproto_set_snoops(br->ofproto, &snoops); + sset_destroy(&snoops); } - svec_destroy(&snoops); - svec_destroy(&old_snoops); mirror_reconfigure(br); } @@ -1875,8 +1952,6 @@ bridge_ofproto_controller_for_mgmt(const struct bridge *br, oc->max_backoff = 0; oc->probe_interval = 60; oc->band = OFPROTO_OUT_OF_BAND; - oc->accept_re = NULL; - oc->update_resolv_conf = false; oc->rate_limit = 0; oc->burst_limit = 0; } @@ -1891,8 +1966,6 @@ bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c, oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band") ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); - oc->accept_re = c->discover_accept_regex; - oc->update_resolv_conf = c->discover_update_resolv_conf; oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; oc->burst_limit = (c->controller_burst_limit ? *c->controller_burst_limit : 0); @@ -1910,13 +1983,8 @@ bridge_configure_local_iface_netdev(struct bridge *br, struct iface *local_iface; struct in_addr ip; - /* Controller discovery does its own TCP/IP configuration later. */ - if (strcmp(c->target, "discover")) { - return; - } - /* If there's no local interface or no IP address, give up. */ - local_iface = bridge_get_local_iface(br); + local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL); if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) { return; } @@ -1958,7 +2026,6 @@ bridge_reconfigure_remotes(struct bridge *br, struct ovsrec_controller **controllers; size_t n_controllers; - bool had_primary; struct ofproto_controller *ocs; size_t n_ocs; @@ -1980,7 +2047,6 @@ bridge_reconfigure_remotes(struct bridge *br, } else { ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); } - had_primary = ofproto_has_primary_controller(br->ofproto); n_controllers = bridge_get_controllers(br, &controllers); @@ -2014,43 +2080,21 @@ bridge_reconfigure_remotes(struct bridge *br, ofproto_set_controllers(br->ofproto, ocs, n_ocs); free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */ free(ocs); - - if (had_primary != ofproto_has_primary_controller(br->ofproto)) { - ofproto_flush_flows(br->ofproto); - } - - /* If there are no controllers and the bridge is in standalone - * mode, set up a flow that matches every packet and directs - * them to OFPP_NORMAL (which goes to us). Otherwise, the - * switch is in secure mode and we won't pass any traffic until - * a controller has been defined and it tells us to do so. */ - if (!n_controllers - && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) { - union ofp_action action; - struct cls_rule rule; - - memset(&action, 0, sizeof action); - action.type = htons(OFPAT_OUTPUT); - action.output.len = htons(sizeof action); - action.output.port = htons(OFPP_NORMAL); - cls_rule_init_catchall(&rule, 0); - ofproto_add_flow(br->ofproto, &rule, &action, 1); - } } static void bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) { - size_t i, j; + struct port *port; shash_init(ifaces); - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { shash_add_once(ifaces, iface->name, iface); } - if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { + if (!list_is_short(&port->ifaces) && port->cfg->bond_fake_iface) { shash_add_once(ifaces, port->name, NULL); } } @@ -2069,13 +2113,13 @@ bridge_fetch_dp_ifaces(struct bridge *br) { struct dpif_port_dump dump; struct dpif_port dpif_port; - size_t i, j; + struct port *port; /* Reset all interface numbers. */ - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { iface->dp_ifidx = -1; } } @@ -2107,482 +2151,64 @@ bridge_fetch_dp_ifaces(struct bridge *br) /* Bridge packet processing functions. */ static bool -bond_is_tcp_hash(const struct port *port) +set_dst(struct dst *dst, const struct flow *flow, + const struct port *in_port, const struct port *out_port, + tag_type *tags) { - return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED; + dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE + : in_port->vlan >= 0 ? in_port->vlan + : flow->vlan_tci == 0 ? OFP_VLAN_NONE + : vlan_tci_to_vid(flow->vlan_tci)); + + dst->iface = (!out_port->bond + ? port_get_an_iface(out_port) + : bond_choose_output_slave(out_port->bond, flow, + dst->vlan, tags)); + + return dst->iface != NULL; } static int -bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan) +mirror_mask_ffs(mirror_mask_t mask) { - return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK; + BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); + return ffs(mask); } -static int bond_hash_tcp(const struct flow *flow, uint16_t vlan) +static void +dst_set_init(struct dst_set *set) { - struct flow hash_flow; - - memcpy(&hash_flow, flow, sizeof hash_flow); - hash_flow.vlan_tci = 0; - - /* The symmetric quality of this hash function is not required, but - * flow_hash_symmetric_l4 already exists, and is sufficient for our - * purposes, so we use it out of convenience. */ - return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK; + set->dsts = set->builtin; + set->n = 0; + set->allocated = ARRAY_SIZE(set->builtin); } -static struct bond_entry * -lookup_bond_entry(const struct port *port, const struct flow *flow, - uint16_t vlan) +static void +dst_set_add(struct dst_set *set, const struct dst *dst) { - assert(port->bond_mode != BM_AB); + if (set->n >= set->allocated) { + size_t new_allocated; + struct dst *new_dsts; - if (bond_is_tcp_hash(port)) { - return &port->bond_hash[bond_hash_tcp(flow, vlan)]; - } else { - return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)]; + new_allocated = set->allocated * 2; + new_dsts = xmalloc(new_allocated * sizeof *new_dsts); + memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); + + dst_set_free(set); + + set->dsts = new_dsts; + set->allocated = new_allocated; } + set->dsts[set->n++] = *dst; } -static int -bond_choose_iface(const struct port *port) +static void +dst_set_free(struct dst_set *set) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); - size_t i, best_down_slave = -1; - long long next_delay_expiration = LLONG_MAX; - - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - - if (iface->enabled) { - return i; - } else if (iface->delay_expires < next_delay_expiration - && (iface->lacp_attached - || !(port->lacp & LACP_NEGOTIATED))) { - best_down_slave = i; - next_delay_expiration = iface->delay_expires; - } + if (set->dsts != set->builtin) { + free(set->dsts); } - - if (best_down_slave != -1) { - struct iface *iface = port->ifaces[best_down_slave]; - - VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay " - "since no other interface is up", iface->name, - iface->delay_expires - time_msec()); - bond_enable_slave(iface, true); - } - - return best_down_slave; -} - -static bool -choose_output_iface(const struct port *port, const struct flow *flow, - uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags) -{ - struct iface *iface; - - assert(port->n_ifaces); - if (port->n_ifaces == 1) { - iface = port->ifaces[0]; - } else if (port->bond_mode == BM_AB) { - if (port->active_iface < 0) { - *tags |= port->no_ifaces_tag; - return false; - } - iface = port->ifaces[port->active_iface]; - } else { - struct bond_entry *e = lookup_bond_entry(port, flow, vlan); - if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces - || !port->ifaces[e->iface_idx]->enabled) { - /* XXX select interface properly. The current interface selection - * is only good for testing the rebalancing code. */ - e->iface_idx = bond_choose_iface(port); - if (e->iface_idx < 0) { - *tags |= port->no_ifaces_tag; - return false; - } - e->iface_tag = tag_create_random(); - ((struct port *) port)->bond_compat_is_stale = true; - } - *tags |= e->iface_tag; - iface = port->ifaces[e->iface_idx]; - } - *dp_ifidx = iface->dp_ifidx; - *tags |= iface->tag; /* Currently only used for bonding. */ - return true; -} - -static void -bond_link_status_update(struct iface *iface) -{ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); - struct port *port = iface->port; - bool up = iface->up; - int updelay, downdelay; - - updelay = port->updelay; - downdelay = port->downdelay; - - if (iface->port->lacp & LACP_NEGOTIATED) { - downdelay = 0; - updelay = 0; - } - - if (iface->port->lacp && up) { - /* The interface is up if it's attached to an aggregator and its - * partner is synchronized. The only exception is defaulted links. - * They are not required to have synchronized partners because they - * have no partners at all. However, they will only be attached if - * negotiations failed on all interfaces in the bond. */ - up = iface->lacp_attached - && (iface->lacp_partner.state & LACP_STATE_SYNC - || iface->lacp_status == LACP_STATUS_DEFAULTED); - } - - - if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { - /* Nothing to do. */ - return; - } - VLOG_INFO_RL(&rl, "interface %s: link state %s", - iface->name, up ? "up" : "down"); - if (up == iface->enabled) { - iface->delay_expires = LLONG_MAX; - VLOG_INFO_RL(&rl, "interface %s: will not be %s", - iface->name, up ? "disabled" : "enabled"); - } else if (up && port->active_iface < 0) { - bond_enable_slave(iface, true); - if (updelay) { - VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " - "other interface is up", iface->name, updelay); - } - } else { - int delay = up ? updelay : downdelay; - iface->delay_expires = time_msec() + delay; - if (delay) { - VLOG_INFO_RL(&rl, - "interface %s: will be %s if it stays %s for %d ms", - iface->name, - up ? "enabled" : "disabled", - up ? "up" : "down", - delay); - } - } -} - -static void -bond_choose_active_iface(struct port *port) -{ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); - - port->active_iface = bond_choose_iface(port); - port->active_iface_tag = tag_create_random(); - if (port->active_iface >= 0) { - VLOG_INFO_RL(&rl, "port %s: active interface is now %s", - port->name, port->ifaces[port->active_iface]->name); - } else { - VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface", - port->name); - } -} - -static void -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; - } - - iface->enabled = enable; - if (!iface->enabled) { - VLOG_WARN("interface %s: disabled", iface->name); - ofproto_revalidate(br->ofproto, iface->tag); - if (iface->port_ifidx == port->active_iface) { - ofproto_revalidate(br->ofproto, - port->active_iface_tag); - - /* Disabling a slave can lead to another slave being immediately - * enabled if there will be no active slaves but one is waiting - * on an updelay. In this case we do not need to run most of the - * code for the newly enabled slave since there was no period - * without an active slave and it is redundant with the disabling - * path. */ - moving_active_iface = true; - bond_choose_active_iface(port); - } - bond_send_learning_packets(port); - } else { - VLOG_WARN("interface %s: enabled", iface->name); - if (port->active_iface < 0 && !moving_active_iface) { - ofproto_revalidate(br->ofproto, port->no_ifaces_tag); - bond_choose_active_iface(port); - bond_send_learning_packets(port); - } - iface->tag = tag_create_random(); - } - - moving_active_iface = false; - port->bond_compat_is_stale = true; -} - -/* Attempts to make the sum of the bond slaves' statistics appear on the fake - * bond interface. */ -static void -bond_update_fake_iface_stats(struct port *port) -{ - struct netdev_stats bond_stats; - struct netdev *bond_dev; - size_t i; - - memset(&bond_stats, 0, sizeof bond_stats); - - for (i = 0; i < port->n_ifaces; i++) { - struct netdev_stats slave_stats; - - if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) { - /* XXX: We swap the stats here because they are swapped back when - * reported by the internal device. The reason for this is - * internal devices normally represent packets going into the system - * but when used as fake bond device they represent packets leaving - * the system. We really should do this in the internal device - * itself because changing it here reverses the counts from the - * perspective of the switch. However, the internal device doesn't - * know what type of device it represents so we have to do it here - * for now. */ - bond_stats.tx_packets += slave_stats.rx_packets; - bond_stats.tx_bytes += slave_stats.rx_bytes; - bond_stats.rx_packets += slave_stats.tx_packets; - bond_stats.rx_bytes += slave_stats.tx_bytes; - } - } - - if (!netdev_open_default(port->name, &bond_dev)) { - netdev_set_stats(bond_dev, &bond_stats); - netdev_close(bond_dev); - } -} - -static void -bond_link_carrier_update(struct iface *iface, bool carrier) -{ - if (carrier == iface->up) { - return; - } - - if (iface->lacp_status == LACP_STATUS_CURRENT) { - iface_set_lacp_expired(iface); - } - - iface->up = carrier; - iface->lacp_tx = 0; - iface->port->bond_compat_is_stale = true; -} - -static void -bond_run(struct bridge *br) -{ - size_t i, j; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - - if (port->n_ifaces >= 2) { - char *devname; - - if (port->monitor) { - assert(!port->miimon); - - /* Track carrier going up and down on interfaces. */ - while (!netdev_monitor_poll(port->monitor, &devname)) { - struct iface *iface; - - iface = port_lookup_iface(port, devname); - if (iface) { - bool up = netdev_get_carrier(iface->netdev); - bond_link_carrier_update(iface, up); - } - free(devname); - } - } else { - assert(port->miimon); - - if (time_msec() >= port->bond_miimon_next_update) { - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - bool up = netdev_get_miimon(iface->netdev); - bond_link_carrier_update(iface, up); - } - port->bond_miimon_next_update = time_msec() + - port->bond_miimon_interval; - } - } - - for (j = 0; j < port->n_ifaces; j++) { - bond_link_status_update(port->ifaces[j]); - } - - 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); - } - } -} - -static void -bond_wait(struct bridge *br) -{ - size_t i, j; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (port->n_ifaces < 2) { - continue; - } - - if (port->monitor) { - netdev_monitor_poll_wait(port->monitor); - } - - if (port->miimon) { - poll_timer_wait_until(port->bond_miimon_next_update); - } - - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->delay_expires != LLONG_MAX) { - poll_timer_wait_until(iface->delay_expires); - } - } - if (port->bond_fake_iface) { - poll_timer_wait_until(port->bond_next_fake_iface_update); - } - } -} - -static bool -set_dst(struct dst *dst, const struct flow *flow, - const struct port *in_port, const struct port *out_port, - tag_type *tags) -{ - dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE - : in_port->vlan >= 0 ? in_port->vlan - : flow->vlan_tci == 0 ? OFP_VLAN_NONE - : vlan_tci_to_vid(flow->vlan_tci)); - return choose_output_iface(out_port, flow, dst->vlan, - &dst->dp_ifidx, tags); -} - -static void -swap_dst(struct dst *p, struct dst *q) -{ - struct dst tmp = *p; - *p = *q; - *q = tmp; -} - -/* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in - * 'dsts'. (This may help performance by reducing the number of VLAN changes - * that we push to the datapath. We could in fact fully sort the array by - * vlan, but in most cases there are at most two different vlan tags so that's - * possibly overkill.) */ -static void -partition_dsts(struct dst_set *set, int vlan) -{ - struct dst *first = set->dsts; - struct dst *last = set->dsts + set->n; - - while (first != last) { - /* Invariants: - * - All dsts < first have vlan == 'vlan'. - * - All dsts >= last have vlan != 'vlan'. - * - first < last. */ - while (first->vlan == vlan) { - if (++first == last) { - return; - } - } - - /* Same invariants, plus one additional: - * - first->vlan != vlan. - */ - while (last[-1].vlan != vlan) { - if (--last == first) { - return; - } - } - - /* Same invariants, plus one additional: - * - last[-1].vlan == vlan.*/ - swap_dst(first++, --last); - } -} - -static int -mirror_mask_ffs(mirror_mask_t mask) -{ - BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); - return ffs(mask); -} - -static void -dst_set_init(struct dst_set *set) -{ - set->dsts = set->builtin; - set->n = 0; - set->allocated = ARRAY_SIZE(set->builtin); -} - -static void -dst_set_add(struct dst_set *set, const struct dst *dst) -{ - if (set->n >= set->allocated) { - size_t new_allocated; - struct dst *new_dsts; - - new_allocated = set->allocated * 2; - new_dsts = xmalloc(new_allocated * sizeof *new_dsts); - memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); - - dst_set_free(set); - - set->dsts = new_dsts; - set->allocated = new_allocated; - } - set->dsts[set->n++] = *dst; -} - -static void -dst_set_free(struct dst_set *set) -{ - if (set->dsts != set->builtin) { - free(set->dsts); - } -} +} static bool dst_is_duplicate(const struct dst_set *set, const struct dst *test) @@ -2590,7 +2216,7 @@ dst_is_duplicate(const struct dst_set *set, const struct dst *test) size_t i; for (i = 0; i < set->n; i++) { if (set->dsts[i].vlan == test->vlan - && set->dsts[i].dp_ifidx == test->dp_ifidx) { + && set->dsts[i].iface == test->iface) { return true; } } @@ -2600,8 +2226,7 @@ dst_is_duplicate(const struct dst_set *set, const struct dst *test) static bool port_trunks_vlan(const struct port *port, uint16_t vlan) { - return (port->vlan < 0 - && (!port->trunks || bitmap_is_set(port->trunks, vlan))); + return (port->vlan < 0 || vlan_bitmap_contains(port->trunks, vlan)); } static bool @@ -2613,62 +2238,87 @@ port_includes_vlan(const struct port *port, uint16_t vlan) static bool port_is_floodable(const struct port *port) { - int i; + struct iface *iface; - for (i = 0; i < port->n_ifaces; i++) { + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { if (!ofproto_port_is_floodable(port->bridge->ofproto, - port->ifaces[i]->dp_ifidx)) { + iface->dp_ifidx)) { return false; } } return true; } +/* Returns an arbitrary interface within 'port'. */ +static struct iface * +port_get_an_iface(const struct port *port) +{ + return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem); +} + static void compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface) { - mirror_mask_t mirrors = in_port->src_mirrors; struct dst dst; - int flow_vlan; - size_t i; - - flow_vlan = vlan_tci_to_vid(flow->vlan_tci); - if (flow_vlan == 0) { - flow_vlan = OFP_VLAN_NONE; - } if (out_port == FLOOD_PORT) { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { if (port != in_port && port_is_floodable(port) && port_includes_vlan(port, vlan) && !port->is_mirror_output_port && set_dst(&dst, flow, in_port, port, tags)) { - mirrors |= port->dst_mirrors; dst_set_add(set, &dst); } } *nf_output_iface = NF_OUT_FLOOD; } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) { dst_set_add(set, &dst); - *nf_output_iface = dst.dp_ifidx; - mirrors |= out_port->dst_mirrors; + *nf_output_iface = dst.iface->dp_ifidx; + } +} + +static void +compose_mirror_dsts(const struct bridge *br, const struct flow *flow, + uint16_t vlan, const struct port *in_port, + struct dst_set *set, tag_type *tags) +{ + mirror_mask_t mirrors; + int flow_vlan; + size_t i; + + mirrors = in_port->src_mirrors; + for (i = 0; i < set->n; i++) { + mirrors |= set->dsts[i].iface->port->dst_mirrors; + } + + if (!mirrors) { + return; + } + + flow_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (flow_vlan == 0) { + flow_vlan = OFP_VLAN_NONE; } while (mirrors) { struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1]; if (!m->n_vlans || vlan_is_mirrored(m, vlan)) { + struct dst dst; + if (m->out_port) { if (set_dst(&dst, flow, in_port, m->out_port, tags) && !dst_is_duplicate(set, &dst)) { dst_set_add(set, &dst); } } else { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { if (port_includes_vlan(port, m->out_vlan) && set_dst(&dst, flow, in_port, port, tags)) { @@ -2696,23 +2346,6 @@ compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, } mirrors &= mirrors - 1; } - - partition_dsts(set, flow_vlan); -} - -static void OVS_UNUSED -print_dsts(const struct dst_set *set) -{ - size_t i; - - for (i = 0; i < set->n; i++) { - const struct dst *dst = &set->dsts[i]; - - printf(">p%"PRIu16, dst->dp_ifidx); - if (dst->vlan != OFP_VLAN_NONE) { - printf("v%"PRIu16, dst->vlan); - } - } } static void @@ -2721,20 +2354,33 @@ compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan, tag_type *tags, struct ofpbuf *actions, uint16_t *nf_output_iface) { + uint16_t initial_vlan, cur_vlan; + const struct dst *dst; struct dst_set set; - uint16_t cur_vlan; - size_t i; dst_set_init(&set); compose_dsts(br, flow, vlan, in_port, out_port, &set, tags, nf_output_iface); + compose_mirror_dsts(br, flow, vlan, in_port, &set, tags); - cur_vlan = vlan_tci_to_vid(flow->vlan_tci); - if (cur_vlan == 0) { - cur_vlan = OFP_VLAN_NONE; + /* Output all the packets we can without having to change the VLAN. */ + initial_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (initial_vlan == 0) { + initial_vlan = OFP_VLAN_NONE; } - for (i = 0; i < set.n; i++) { - const struct dst *dst = &set.dsts[i]; + for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { + if (dst->vlan != initial_vlan) { + continue; + } + nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->iface->dp_ifidx); + } + + /* Then output the rest. */ + cur_vlan = initial_vlan; + for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { + if (dst->vlan == initial_vlan) { + continue; + } if (dst->vlan != cur_vlan) { if (dst->vlan == OFP_VLAN_NONE) { nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN); @@ -2746,8 +2392,9 @@ compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan, } cur_vlan = dst->vlan; } - nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx); + nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->iface->dp_ifidx); } + dst_set_free(&set); } @@ -2761,7 +2408,6 @@ static int flow_get_vlan(struct bridge *br, const struct flow *flow, int vlan = vlan_tci_to_vid(flow->vlan_tci); if (in_port->vlan >= 0) { if (vlan) { - /* XXX support double tagging? */ if (have_packet) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " @@ -2805,27 +2451,34 @@ static void update_learning_table(struct bridge *br, const struct flow *flow, int vlan, struct port *in_port) { - enum grat_arp_lock_type lock_type; - tag_type rev_tag; - - /* We don't want to learn from gratuitous ARP packets that are reflected - * back over bond slaves so we lock the learning table. */ - lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE : - (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET : - GRAT_ARP_LOCK_CHECK; - - rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx, - lock_type); - if (rev_tag) { + struct mac_entry *mac; + + if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) { + return; + } + + mac = mac_learning_insert(br->ml, flow->dl_src, vlan); + if (is_gratuitous_arp(flow)) { + /* We don't want to learn from gratuitous ARP packets that are + * reflected back over bond slaves so we lock the learning table. */ + if (!in_port->bond) { + mac_entry_set_grat_arp_lock(mac); + } else if (mac_entry_is_grat_arp_locked(mac)) { + return; + } + } + + if (mac_entry_is_new(mac) || mac->port.p != in_port) { /* The log messages here could actually be useful in debugging, * so keep the rate limit relatively high. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, - 300); + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " "on port %s in VLAN %d", br->name, ETH_ADDR_ARGS(flow->dl_src), in_port->name, vlan); - ofproto_revalidate(br->ofproto, rev_tag); + + mac->port.p = in_port; + ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac)); } } @@ -2904,39 +2557,27 @@ is_admissible(struct bridge *br, const struct flow *flow, bool have_packet, return false; } - /* When using LACP, do not accept packets from disabled interfaces. */ - if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) { - return false; - } + if (in_port->bond) { + struct mac_entry *mac; + + switch (bond_check_admissibility(in_port->bond, in_iface, + flow->dl_dst, tags)) { + case BV_ACCEPT: + break; - /* Packets received on non-LACP bonds need special attention to avoid - * duplicates. */ - if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) { - int src_idx; - bool is_grat_arp_locked; + case BV_DROP: + return false; - 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. */ + case BV_DROP_IF_MOVED: + mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL); + if (mac && mac->port.p != in_port && + (!is_gratuitous_arp(flow) + || mac_entry_is_grat_arp_locked(mac))) { return false; } + break; } - - /* Drop all packets for which we have learned a different input - * port, because we probably sent the packet on one slave and got - * it back on the other. Gratuitous ARP packets are an exception - * to this rule: the host has moved to another switch. The exception - * to the exception is if we locked the learning table to avoid - * reflections on bond slaves. If this is the case, just drop the - * packet now. */ - src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan, - &is_grat_arp_locked); - if (src_idx != -1 && src_idx != in_port->port_idx && - (!is_gratuitous_arp(flow) || is_grat_arp_locked)) { - return false; - } - } + } return true; } @@ -2951,8 +2592,8 @@ process_flow(struct bridge *br, const struct flow *flow, { struct port *in_port; struct port *out_port; + struct mac_entry *mac; int vlan; - int out_port_idx; /* Check whether we should drop packets in this flow. */ if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) { @@ -2966,10 +2607,9 @@ process_flow(struct bridge *br, const struct flow *flow, } /* Determine output port. */ - out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags, - NULL); - if (out_port_idx >= 0 && out_port_idx < br->n_ports) { - out_port = br->ports[out_port_idx]; + mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags); + if (mac) { + out_port = mac->port.p; } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { /* If we are revalidating but don't have a learning entry then * eject the flow. Installing a flow that floods packets opens @@ -3000,33 +2640,39 @@ bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, struct ofpbuf *actions, tag_type *tags, uint16_t *nf_output_iface, void *br_) { - struct iface *iface; struct bridge *br = br_; COVERAGE_INC(bridge_process_flow); + return process_flow(br, flow, packet, actions, tags, nf_output_iface); +} + +static bool +bridge_special_ofhook_cb(const struct flow *flow, + const struct ofpbuf *packet, void *br_) +{ + struct iface *iface; + struct bridge *br = br_; iface = iface_from_dp_ifidx(br, flow->in_port); - if (cfm_should_process_flow(flow)) { - if (packet && iface->cfm) { - cfm_process_heartbeat(iface->cfm, packet); - } - return false; - } else if (flow->dl_type == htons(ETH_TYPE_LACP)) { - if (packet) { - lacp_process_packet(packet, iface); + if (flow->dl_type == htons(ETH_TYPE_LACP)) { + if (iface && iface->port->lacp && packet) { + const struct lacp_pdu *pdu = parse_lacp_packet(packet); + if (pdu) { + lacp_process_pdu(iface->port->lacp, iface, pdu); + } } return false; } - return process_flow(br, flow, packet, actions, tags, nf_output_iface); + return true; } static void bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, const struct nlattr *actions, size_t actions_len, - unsigned long long int n_bytes, void *br_) + uint64_t n_bytes, void *br_) { struct bridge *br = br_; const struct nlattr *a; @@ -3053,13 +2699,11 @@ bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) { if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) { struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a)); - if (out_port && out_port->n_ifaces >= 2 && - out_port->bond_mode == BM_SLB) { + if (out_port && out_port->bond) { uint16_t vlan = (flow->vlan_tci ? vlan_tci_to_vid(flow->vlan_tci) : OFP_VLAN_NONE); - struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan); - e->tx_bytes += n_bytes; + bond_account(out_port->bond, flow, vlan, n_bytes); } } } @@ -3069,1068 +2713,117 @@ static void bridge_account_checkpoint_ofhook_cb(void *br_) { struct bridge *br = br_; - long long int now; - size_t i; - - if (!br->has_bonded_ports) { - return; - } - - now = time_msec(); - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (port->n_ifaces > 1 && port->bond_mode == BM_SLB - && now >= port->bond_next_rebalance) { - port->bond_next_rebalance = now + port->bond_rebalance_interval; - bond_rebalance_port(port); - } - } -} - -static struct ofhooks bridge_ofhooks = { - bridge_normal_ofhook_cb, - bridge_account_flow_ofhook_cb, - bridge_account_checkpoint_ofhook_cb, -}; - -/* LACP functions. */ - -static void -lacp_process_packet(const struct ofpbuf *packet, struct iface *iface) -{ - const struct lacp_pdu *pdu; - - if (!iface->port->lacp) { - return; - } - - pdu = parse_lacp_packet(packet); - if (!pdu) { - return; - } - - iface->lacp_status = LACP_STATUS_CURRENT; - iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX; - - iface->lacp_actor.state = iface_get_lacp_state(iface); - if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) { - iface->lacp_tx = 0; - } - - if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) { - iface->port->lacp_need_update = true; - iface->lacp_partner = pdu->actor; - } -} - -static void -lacp_update_ifaces(struct port *port) -{ - size_t i; - struct iface *lead; - struct lacp_info lead_pri; - - port->lacp_need_update = false; - - if (!port->lacp) { - return; - } - - lead = NULL; - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - struct lacp_info pri; - - iface->lacp_attached = true; - ofproto_revalidate(port->bridge->ofproto, iface->tag); - - /* Don't allow loopback interfaces to send traffic or lead. */ - if (eth_addr_equals(iface->lacp_partner.sysid, - iface->lacp_actor.sysid)) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); - VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is " - "connected to its own bridge", iface->name); - iface->lacp_attached = false; - continue; - } - - if (iface->lacp_status == LACP_STATUS_DEFAULTED) { - continue; - } - - iface_get_lacp_priority(iface, &pri); - - if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) { - lead = iface; - lead_pri = pri; - } - } - - if (!lead) { - port->lacp &= ~LACP_NEGOTIATED; - return; - } - - port->lacp |= LACP_NEGOTIATED; - - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - - if (iface->lacp_status == LACP_STATUS_DEFAULTED - || lead->lacp_partner.key != iface->lacp_partner.key - || !eth_addr_equals(lead->lacp_partner.sysid, - iface->lacp_partner.sysid)) { - iface->lacp_attached = false; - } - } -} - -static bool -lacp_iface_may_tx(const struct iface *iface) -{ - return iface->port->lacp & LACP_ACTIVE - || iface->lacp_status != LACP_STATUS_DEFAULTED; -} - -static void -lacp_run(struct bridge *br) -{ - size_t i, j; - struct ofpbuf packet; - - ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN); - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - - if (!port->lacp) { - continue; - } - - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - - if (time_msec() > iface->lacp_rx) { - if (iface->lacp_status == LACP_STATUS_CURRENT) { - iface_set_lacp_expired(iface); - } else if (iface->lacp_status == LACP_STATUS_EXPIRED) { - iface_set_lacp_defaulted(iface); - } - } - } - - if (port->lacp_need_update) { - lacp_update_ifaces(port); - } - - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - uint8_t ea[ETH_ADDR_LEN]; - int error; - - if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) { - continue; - } - - error = netdev_get_etheraddr(iface->netdev, ea); - if (!error) { - iface->lacp_actor.state = iface_get_lacp_state(iface); - compose_lacp_packet(&packet, &iface->lacp_actor, - &iface->lacp_partner, ea); - iface_send_packet(iface, &packet); - } else { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); - VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address " - "(%s)", iface->name, strerror(error)); - } - - iface->lacp_tx = time_msec() + - (iface->lacp_partner.state & LACP_STATE_TIME - ? LACP_FAST_TIME_TX - : LACP_SLOW_TIME_TX); - } - } - ofpbuf_uninit(&packet); -} - -static void -lacp_wait(struct bridge *br) -{ - size_t i, j; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - - if (!port->lacp) { - continue; - } - - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - - if (lacp_iface_may_tx(iface)) { - poll_timer_wait_until(iface->lacp_tx); - } - - if (iface->lacp_status != LACP_STATUS_DEFAULTED) { - poll_timer_wait_until(iface->lacp_rx); - } - } - } -} - -/* Bonding functions. */ - -/* Statistics for a single interface on a bonded port, used for load-based - * bond rebalancing. */ -struct slave_balance { - struct iface *iface; /* The interface. */ - uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */ - - /* All the "bond_entry"s that are assigned to this interface, in order of - * increasing tx_bytes. */ - struct bond_entry **hashes; - size_t n_hashes; -}; - -static const char * -bond_mode_to_string(enum bond_mode bm) { - static char *bm_slb = "balance-slb"; - static char *bm_ab = "active-backup"; - static char *bm_tcp = "balance-tcp"; - - switch (bm) { - case BM_SLB: return bm_slb; - case BM_AB: return bm_ab; - case BM_TCP: return bm_tcp; - } - - NOT_REACHED(); - return NULL; -} - -/* Sorts pointers to pointers to bond_entries in ascending order by the - * interface to which they are assigned, and within a single interface in - * ascending order of bytes transmitted. */ -static int -compare_bond_entries(const void *a_, const void *b_) -{ - const struct bond_entry *const *ap = a_; - const struct bond_entry *const *bp = b_; - const struct bond_entry *a = *ap; - const struct bond_entry *b = *bp; - if (a->iface_idx != b->iface_idx) { - return a->iface_idx > b->iface_idx ? 1 : -1; - } else if (a->tx_bytes != b->tx_bytes) { - return a->tx_bytes > b->tx_bytes ? 1 : -1; - } else { - return 0; - } -} - -/* Sorts slave_balances so that enabled ports come first, and otherwise in - * *descending* order by number of bytes transmitted. */ -static int -compare_slave_balance(const void *a_, const void *b_) -{ - const struct slave_balance *a = a_; - const struct slave_balance *b = b_; - if (a->iface->enabled != b->iface->enabled) { - return a->iface->enabled ? -1 : 1; - } else if (a->tx_bytes != b->tx_bytes) { - return a->tx_bytes > b->tx_bytes ? -1 : 1; - } else { - return 0; - } -} - -static void -swap_bals(struct slave_balance *a, struct slave_balance *b) -{ - struct slave_balance tmp = *a; - *a = *b; - *b = tmp; -} - -/* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order - * given that 'p' (and only 'p') might be in the wrong location. - * - * This function invalidates 'p', since it might now be in a different memory - * location. */ -static void -resort_bals(struct slave_balance *p, - struct slave_balance bals[], size_t n_bals) -{ - if (n_bals > 1) { - for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) { - swap_bals(p, p - 1); - } - for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) { - swap_bals(p, p + 1); - } - } -} - -static void -log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port) -{ - if (VLOG_IS_DBG_ENABLED()) { - struct ds ds = DS_EMPTY_INITIALIZER; - const struct slave_balance *b; - - for (b = bals; b < bals + n_bals; b++) { - size_t i; - - if (b > bals) { - ds_put_char(&ds, ','); - } - ds_put_format(&ds, " %s %"PRIu64"kB", - b->iface->name, b->tx_bytes / 1024); + struct port *port; - if (!b->iface->enabled) { - ds_put_cstr(&ds, " (disabled)"); - } - if (b->n_hashes > 0) { - ds_put_cstr(&ds, " ("); - for (i = 0; i < b->n_hashes; i++) { - const struct bond_entry *e = b->hashes[i]; - if (i > 0) { - ds_put_cstr(&ds, " + "); - } - ds_put_format(&ds, "h%td: %"PRIu64"kB", - e - port->bond_hash, e->tx_bytes / 1024); - } - ds_put_cstr(&ds, ")"); - } + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + if (port->bond) { + bond_rebalance(port->bond, + ofproto_get_revalidate_set(br->ofproto)); } - VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds)); - ds_destroy(&ds); } } -/* Shifts 'hash' from 'from' to 'to' within 'port'. */ -static void -bond_shift_load(struct slave_balance *from, struct slave_balance *to, - int hash_idx) +static uint16_t +bridge_autopath_ofhook_cb(const struct flow *flow, uint32_t ofp_port, + tag_type *tags, void *br_) { - struct bond_entry *hash = from->hashes[hash_idx]; - struct port *port = from->iface->port; - uint64_t delta = hash->tx_bytes; - - assert(port->bond_mode == BM_SLB); - - VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) " - "from %s to %s (now carrying %"PRIu64"kB and " - "%"PRIu64"kB load, respectively)", - port->name, delta / 1024, hash - port->bond_hash, - from->iface->name, to->iface->name, - (from->tx_bytes - delta) / 1024, - (to->tx_bytes + delta) / 1024); + struct bridge *br = br_; + uint16_t odp_port = ofp_port_to_odp_port(ofp_port); + struct port *port = port_from_dp_ifidx(br, odp_port); + uint16_t ret; - /* Delete element from from->hashes. - * - * We don't bother to add the element to to->hashes because not only would - * it require more work, the only purpose it would be to allow that hash to - * be migrated to another slave in this rebalancing run, and there is no - * point in doing that. */ - if (hash_idx == 0) { - from->hashes++; + if (!port) { + ret = ODPP_NONE; + } else if (list_is_short(&port->ifaces)) { + ret = odp_port; } else { - memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1, - (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes); - } - from->n_hashes--; - - /* Shift load away from 'from' to 'to'. */ - from->tx_bytes -= delta; - to->tx_bytes += delta; - - /* Arrange for flows to be revalidated. */ - ofproto_revalidate(port->bridge->ofproto, hash->iface_tag); - hash->iface_idx = to->iface->port_ifidx; - hash->iface_tag = tag_create_random(); -} - -static void -bond_rebalance_port(struct port *port) -{ - struct slave_balance *bals; - size_t n_bals; - struct bond_entry *hashes[BOND_MASK + 1]; - struct slave_balance *b, *from, *to; - struct bond_entry *e; - size_t i; - - assert(port->bond_mode != BM_AB); - - /* Sets up 'bals' to describe each of the port's interfaces, sorted in - * descending order of tx_bytes, so that bals[0] represents the most - * heavily loaded slave and bals[n_bals - 1] represents the least heavily - * loaded slave. - * - * The code is a bit tricky: to avoid dynamically allocating a 'hashes' - * array for each slave_balance structure, we sort our local array of - * hashes in order by slave, so that all of the hashes for a given slave - * become contiguous in memory, and then we point each 'hashes' members of - * a slave_balance structure to the start of a contiguous group. */ - n_bals = port->n_ifaces; - bals = xmalloc(n_bals * sizeof *bals); - for (b = bals; b < &bals[n_bals]; b++) { - b->iface = port->ifaces[b - bals]; - b->tx_bytes = 0; - b->hashes = NULL; - b->n_hashes = 0; - } - for (i = 0; i <= BOND_MASK; i++) { - hashes[i] = &port->bond_hash[i]; - } - qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries); - for (i = 0; i <= BOND_MASK; i++) { - e = hashes[i]; - if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { - b = &bals[e->iface_idx]; - b->tx_bytes += e->tx_bytes; - if (!b->hashes) { - b->hashes = &hashes[i]; - } - b->n_hashes++; - } - } - qsort(bals, n_bals, sizeof *bals, compare_slave_balance); - log_bals(bals, n_bals, port); - - /* Discard slaves that aren't enabled (which were sorted to the back of the - * array earlier). */ - while (!bals[n_bals - 1].iface->enabled) { - n_bals--; - if (!n_bals) { - goto exit; - } - } - - /* Shift load from the most-loaded slaves to the least-loaded slaves. */ - to = &bals[n_bals - 1]; - for (from = bals; from < to; ) { - uint64_t overload = from->tx_bytes - to->tx_bytes; - if (overload < to->tx_bytes >> 5 || overload < 100000) { - /* The extra load on 'from' (and all less-loaded slaves), compared - * to that of 'to' (the least-loaded slave), is less than ~3%, or - * it is less than ~1Mbps. No point in rebalancing. */ - break; - } else if (from->n_hashes == 1) { - /* 'from' only carries a single MAC hash, so we can't shift any - * load away from it, even though we want to. */ - from++; - } else { - /* 'from' is carrying significantly more load than 'to', and that - * load is split across at least two different hashes. Pick a hash - * to migrate to 'to' (the least-loaded slave), given that doing so - * must decrease the ratio of the load on the two slaves by at - * least 0.1. - * - * The sort order we use means that we prefer to shift away the - * smallest hashes instead of the biggest ones. There is little - * reason behind this decision; we could use the opposite sort - * order to shift away big hashes ahead of small ones. */ - bool order_swapped; - - for (i = 0; i < from->n_hashes; i++) { - double old_ratio, new_ratio; - uint64_t delta = from->hashes[i]->tx_bytes; - - if (delta == 0 || from->tx_bytes - delta == 0) { - /* Pointless move. */ - continue; - } - - order_swapped = from->tx_bytes - delta < to->tx_bytes + delta; - - if (to->tx_bytes == 0) { - /* Nothing on the new slave, move it. */ - break; - } - - old_ratio = (double)from->tx_bytes / to->tx_bytes; - new_ratio = (double)(from->tx_bytes - delta) / - (to->tx_bytes + delta); - - if (new_ratio == 0) { - /* Should already be covered but check to prevent division - * by zero. */ - continue; - } - - if (new_ratio < 1) { - new_ratio = 1 / new_ratio; - } - - if (old_ratio - new_ratio > 0.1) { - /* Would decrease the ratio, move it. */ - break; - } - } - if (i < from->n_hashes) { - bond_shift_load(from, to, i); - port->bond_compat_is_stale = true; - - /* If the result of the migration changed the relative order of - * 'from' and 'to' swap them back to maintain invariants. */ - if (order_swapped) { - swap_bals(from, to); - } - - /* Re-sort 'bals'. Note that this may make 'from' and 'to' - * point to different slave_balance structures. It is only - * valid to do these two operations in a row at all because we - * know that 'from' will not move past 'to' and vice versa. */ - resort_bals(from, bals, n_bals); - resort_bals(to, bals, n_bals); - } else { - from++; - } - } - } + struct iface *iface; - /* Implement exponentially weighted moving average. A weight of 1/2 causes - * historical data to decay to <1% in 7 rebalancing runs. */ - for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) { - e->tx_bytes /= 2; + /* Autopath does not support VLAN hashing. */ + iface = bond_choose_output_slave(port->bond, flow, + OFP_VLAN_NONE, tags); + ret = iface ? iface->dp_ifidx : ODPP_NONE; } -exit: - free(bals); + return odp_port_to_ofp_port(ret); } -static void -bond_send_learning_packets(struct port *port) -{ - struct bridge *br = port->bridge; - struct mac_entry *e; - struct ofpbuf packet; - int error, n_packets, n_errors; - - if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) { - return; - } - - ofpbuf_init(&packet, 128); - error = n_packets = n_errors = 0; - LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { - union ofp_action actions[2], *a; - uint16_t dp_ifidx; - tag_type tags = 0; - struct flow flow; - int retval; - - if (e->port == port->port_idx) { - continue; - } - - compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, - e->mac); - flow_extract(&packet, 0, ODPP_NONE, &flow); - - if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) { - continue; - } - - /* Compose actions. */ - memset(actions, 0, sizeof actions); - a = actions; - if (e->vlan) { - a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID); - a->vlan_vid.len = htons(sizeof *a); - a->vlan_vid.vlan_vid = htons(e->vlan); - a++; - } - a->output.type = htons(OFPAT_OUTPUT); - a->output.len = htons(sizeof *a); - a->output.port = htons(odp_port_to_ofp_port(dp_ifidx)); - a++; - - /* Send packet. */ - n_packets++; - retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, - &packet); - if (retval) { - error = retval; - n_errors++; - } - } - ofpbuf_uninit(&packet); - - if (n_errors) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " - "packets, last error was: %s", - port->name, n_errors, n_packets, strerror(error)); - } else { - VLOG_DBG("bond %s: sent %d gratuitous learning packets", - port->name, n_packets); - } -} +static struct ofhooks bridge_ofhooks = { + bridge_normal_ofhook_cb, + bridge_special_ofhook_cb, + bridge_account_flow_ofhook_cb, + bridge_account_checkpoint_ofhook_cb, + bridge_autopath_ofhook_cb, +}; -/* Bonding unixctl user interface functions. */ - -static void -bond_unixctl_list(struct unixctl_conn *conn, - const char *args OVS_UNUSED, void *aux OVS_UNUSED) -{ - struct ds ds = DS_EMPTY_INITIALIZER; - const struct bridge *br; - - ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n"); - - LIST_FOR_EACH (br, node, &all_bridges) { - size_t i; - - for (i = 0; i < br->n_ports; i++) { - const struct port *port = br->ports[i]; - if (port->n_ifaces > 1) { - size_t j; - - ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name, - bond_mode_to_string(port->bond_mode)); - for (j = 0; j < port->n_ifaces; j++) { - const struct iface *iface = port->ifaces[j]; - if (j) { - ds_put_cstr(&ds, ", "); - } - ds_put_cstr(&ds, iface->name); - } - ds_put_char(&ds, '\n'); - } - } - } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); - ds_destroy(&ds); -} - -static struct port * -bond_find(const char *name) -{ - const struct bridge *br; - - LIST_FOR_EACH (br, node, &all_bridges) { - size_t i; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (!strcmp(port->name, name) && port->n_ifaces > 1) { - return port; - } - } - } - return NULL; -} - -static void -ds_put_lacp_state(struct ds *ds, uint8_t state) -{ - if (state & LACP_STATE_ACT) { - ds_put_cstr(ds, "activity "); - } - - if (state & LACP_STATE_TIME) { - ds_put_cstr(ds, "timeout "); - } - - if (state & LACP_STATE_AGG) { - ds_put_cstr(ds, "aggregation "); - } - - if (state & LACP_STATE_SYNC) { - ds_put_cstr(ds, "synchronized "); - } - - if (state & LACP_STATE_COL) { - ds_put_cstr(ds, "collecting "); - } - - if (state & LACP_STATE_DIST) { - ds_put_cstr(ds, "distributing "); - } - - if (state & LACP_STATE_DEF) { - ds_put_cstr(ds, "defaulted "); - } - - if (state & LACP_STATE_EXP) { - ds_put_cstr(ds, "expired "); - } -} +/* Port functions. */ static void -bond_unixctl_show(struct unixctl_conn *conn, - const char *args, void *aux OVS_UNUSED) +lacp_send_pdu_cb(void *iface_, const struct lacp_pdu *pdu) { - struct ds ds = DS_EMPTY_INITIALIZER; - const struct port *port; - size_t j; - - port = bond_find(args); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } - - ds_put_format(&ds, "bond_mode: %s\n", - bond_mode_to_string(port->bond_mode)); - - if (port->lacp) { - ds_put_format(&ds, "\tlacp: %s\n", - port->lacp & LACP_ACTIVE ? "active" : "passive"); - } else { - ds_put_cstr(&ds, "\tlacp: off\n"); - } - - if (port->bond_mode != BM_AB) { - ds_put_format(&ds, "bond-hash-algorithm: %s\n", - bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb"); - } - - - ds_put_format(&ds, "bond-detect-mode: %s\n", - port->miimon ? "miimon" : "carrier"); - - if (port->miimon) { - ds_put_format(&ds, "bond-miimon-interval: %lld\n", - port->bond_miimon_interval); - } - - ds_put_format(&ds, "updelay: %d ms\n", port->updelay); - ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); - - if (port->bond_mode != BM_AB) { - ds_put_format(&ds, "next rebalance: %lld ms\n", - port->bond_next_rebalance - time_msec()); - } - - for (j = 0; j < port->n_ifaces; j++) { - const struct iface *iface = port->ifaces[j]; - struct bond_entry *be; - struct flow flow; - - /* Basic info. */ - ds_put_format(&ds, "slave %s: %s\n", - iface->name, iface->enabled ? "enabled" : "disabled"); - if (j == port->active_iface) { - ds_put_cstr(&ds, "\tactive slave\n"); - } - if (iface->delay_expires != LLONG_MAX) { - ds_put_format(&ds, "\t%s expires in %lld ms\n", - iface->enabled ? "downdelay" : "updelay", - iface->delay_expires - time_msec()); - } - - if (port->lacp) { - ds_put_cstr(&ds, "\tstatus: "); - - if (iface->lacp_status == LACP_STATUS_CURRENT) { - ds_put_cstr(&ds, "current "); - } else if (iface->lacp_status == LACP_STATUS_EXPIRED) { - ds_put_cstr(&ds, "expired "); - } else { - ds_put_cstr(&ds, "defaulted "); - } - - if (iface->lacp_attached) { - ds_put_cstr(&ds, "attached "); - } - - ds_put_cstr(&ds, "\n"); - - ds_put_cstr(&ds, "\n\tactor sysid: "); - ds_put_format(&ds, ETH_ADDR_FMT, - ETH_ADDR_ARGS(iface->lacp_actor.sysid)); - ds_put_cstr(&ds, "\n"); - - ds_put_format(&ds, "\tactor sys_priority: %u\n", - ntohs(iface->lacp_actor.sys_priority)); - - ds_put_format(&ds, "\tactor portid: %u\n", - ntohs(iface->lacp_actor.portid)); - - ds_put_format(&ds, "\tactor port_priority: %u\n", - ntohs(iface->lacp_actor.port_priority)); - - ds_put_format(&ds, "\tactor key: %u\n", - ntohs(iface->lacp_actor.key)); - - ds_put_cstr(&ds, "\tactor state: "); - ds_put_lacp_state(&ds, iface_get_lacp_state(iface)); - ds_put_cstr(&ds, "\n\n"); - - ds_put_cstr(&ds, "\tpartner sysid: "); - ds_put_format(&ds, ETH_ADDR_FMT, - ETH_ADDR_ARGS(iface->lacp_partner.sysid)); - ds_put_cstr(&ds, "\n"); - - ds_put_format(&ds, "\tpartner sys_priority: %u\n", - ntohs(iface->lacp_partner.sys_priority)); - - ds_put_format(&ds, "\tpartner portid: %u\n", - ntohs(iface->lacp_partner.portid)); - - ds_put_format(&ds, "\tpartner port_priority: %u\n", - ntohs(iface->lacp_partner.port_priority)); - - ds_put_format(&ds, "\tpartner key: %u\n", - ntohs(iface->lacp_partner.key)); - - ds_put_cstr(&ds, "\tpartner state: "); - ds_put_lacp_state(&ds, iface->lacp_partner.state); - ds_put_cstr(&ds, "\n\n"); - } - - if (port->bond_mode == BM_AB) { - continue; - } - - /* Hashes. */ - memset(&flow, 0, sizeof flow); - for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { - int hash = be - port->bond_hash; - struct mac_entry *me; - - if (be->iface_idx != j) { - continue; - } - - ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", - hash, be->tx_bytes / 1024); - - if (port->bond_mode != BM_SLB) { - continue; - } + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); + struct iface *iface = iface_; + uint8_t ea[ETH_ADDR_LEN]; + int error; - /* MACs. */ - LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) { - uint16_t dp_ifidx; - tag_type tags = 0; - - memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN); - if (bond_hash_src(me->mac, me->vlan) == hash - && me->port != port->port_idx - && choose_output_iface(port, &flow, me->vlan, - &dp_ifidx, &tags) - && dp_ifidx == iface->dp_ifidx) - { - ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n", - ETH_ADDR_ARGS(me->mac)); - } - } + error = netdev_get_etheraddr(iface->netdev, ea); + if (!error) { + struct lacp_pdu *packet_pdu; + struct ofpbuf packet; + + ofpbuf_init(&packet, 0); + packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, + sizeof *packet_pdu); + *packet_pdu = *pdu; + error = netdev_send(iface->netdev, &packet); + if (error) { + VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed " + "(%s)", iface->port->name, iface->name, + strerror(error)); } - } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); - ds_destroy(&ds); -} - -static void -bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, - void *aux OVS_UNUSED) -{ - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *hash_s, *slave_s; - struct port *port; - struct iface *iface; - struct bond_entry *entry; - int hash; - - bond_s = strtok_r(args, " ", &save_ptr); - hash_s = strtok_r(NULL, " ", &save_ptr); - slave_s = strtok_r(NULL, " ", &save_ptr); - if (!slave_s) { - unixctl_command_reply(conn, 501, - "usage: bond/migrate BOND HASH SLAVE"); - return; - } - - port = bond_find(bond_s); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } - - if (port->bond_mode != BM_SLB) { - unixctl_command_reply(conn, 501, "not an SLB bond"); - return; - } - - if (strspn(hash_s, "0123456789") == strlen(hash_s)) { - hash = atoi(hash_s) & BOND_MASK; + ofpbuf_uninit(&packet); } else { - unixctl_command_reply(conn, 501, "bad hash"); - return; + VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " + "%s (%s)", iface->port->name, iface->name, + strerror(error)); } - - iface = port_lookup_iface(port, slave_s); - if (!iface) { - unixctl_command_reply(conn, 501, "no such slave"); - return; - } - - if (!iface->enabled) { - unixctl_command_reply(conn, 501, "cannot migrate to disabled slave"); - return; - } - - entry = &port->bond_hash[hash]; - ofproto_revalidate(port->bridge->ofproto, entry->iface_tag); - entry->iface_idx = iface->port_ifidx; - entry->iface_tag = tag_create_random(); - port->bond_compat_is_stale = true; - unixctl_command_reply(conn, 200, "migrated"); -} - -static void -bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, - void *aux OVS_UNUSED) -{ - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *slave_s; - struct port *port; - struct iface *iface; - - bond_s = strtok_r(args, " ", &save_ptr); - slave_s = strtok_r(NULL, " ", &save_ptr); - if (!slave_s) { - unixctl_command_reply(conn, 501, - "usage: bond/set-active-slave BOND SLAVE"); - return; - } - - port = bond_find(bond_s); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } - - iface = port_lookup_iface(port, slave_s); - if (!iface) { - unixctl_command_reply(conn, 501, "no such slave"); - return; - } - - if (!iface->enabled) { - unixctl_command_reply(conn, 501, "cannot make disabled slave active"); - return; - } - - if (port->active_iface != iface->port_ifidx) { - ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); - port->active_iface = iface->port_ifidx; - port->active_iface_tag = tag_create_random(); - VLOG_INFO("port %s: active interface is now %s", - port->name, iface->name); - bond_send_learning_packets(port); - unixctl_command_reply(conn, 200, "done"); - } else { - unixctl_command_reply(conn, 200, "no change"); - } -} - -static void -enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) -{ - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *slave_s; - struct port *port; - struct iface *iface; - - bond_s = strtok_r(args, " ", &save_ptr); - slave_s = strtok_r(NULL, " ", &save_ptr); - if (!slave_s) { - unixctl_command_reply(conn, 501, - "usage: bond/enable/disable-slave BOND SLAVE"); - return; - } - - port = bond_find(bond_s); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } - - iface = port_lookup_iface(port, slave_s); - if (!iface) { - unixctl_command_reply(conn, 501, "no such slave"); - return; - } - - bond_enable_slave(iface, enable); - unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled"); -} - -static void -bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, - void *aux OVS_UNUSED) -{ - enable_slave(conn, args, true); -} - -static void -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) +port_run(struct port *port) { - char *args = (char *) args_; - uint8_t mac[ETH_ADDR_LEN]; - uint8_t hash; - char *hash_cstr; - unsigned int vlan; - char *mac_s, *vlan_s; - char *save_ptr = NULL; - - mac_s = strtok_r(args, " ", &save_ptr); - vlan_s = strtok_r(NULL, " ", &save_ptr); - - if (vlan_s) { - if (sscanf(vlan_s, "%u", &vlan) != 1) { - unixctl_command_reply(conn, 501, "invalid vlan"); - return; - } - } else { - vlan = OFP_VLAN_NONE; + if (port->lacp) { + lacp_run(port->lacp, lacp_send_pdu_cb); } - if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) - == ETH_ADDR_SCAN_COUNT) { - hash = bond_hash_src(mac, vlan); + if (port->bond) { + struct iface *iface; - hash_cstr = xasprintf("%u", hash); - unixctl_command_reply(conn, 200, hash_cstr); - free(hash_cstr); - } else { - unixctl_command_reply(conn, 501, "invalid mac"); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + bool may_enable = lacp_slave_may_enable(port->lacp, iface); + bond_slave_set_lacp_may_enable(port->bond, iface, may_enable); + } + + bond_run(port->bond, + ofproto_get_revalidate_set(port->bridge->ofproto), + lacp_negotiated(port->lacp)); + if (bond_should_send_learning_packets(port->bond)) { + port_send_learning_packets(port); + } } } static void -bond_init(void) +port_wait(struct port *port) { - unixctl_command_register("bond/list", bond_unixctl_list, NULL); - unixctl_command_register("bond/show", bond_unixctl_show, NULL); - unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL); - unixctl_command_register("bond/set-active-slave", - bond_unixctl_set_active_slave, NULL); - unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave, - NULL); - unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave, - NULL); - unixctl_command_register("bond/hash", bond_unixctl_hash, NULL); + if (port->lacp) { + lacp_wait(port->lacp); + } + + if (port->bond) { + bond_wait(port->bond); + } } - -/* Port functions. */ static struct port * port_create(struct bridge *br, const char *name) @@ -4139,18 +2832,12 @@ port_create(struct bridge *br, const char *name) port = xzalloc(sizeof *port); port->bridge = br; - port->port_idx = br->n_ports; port->vlan = -1; port->trunks = NULL; port->name = xstrdup(name); - port->active_iface = -1; + list_init(&port->ifaces); - if (br->n_ports >= br->allocated_ports) { - br->ports = x2nrealloc(br->ports, &br->allocated_ports, - sizeof *br->ports); - } - br->ports[br->n_ports++] = port; - shash_add_assert(&br->port_by_name, port->name, port); + hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0)); VLOG_INFO("created port %s on bridge %s", port->name, br->name); bridge_flush(br); @@ -4183,103 +2870,63 @@ get_interface_other_config(const struct ovsrec_interface *iface, static void port_del_ifaces(struct port *port, const struct ovsrec_port *cfg) { - struct shash new_ifaces; + struct iface *iface, *next; + struct sset new_ifaces; size_t i; /* Collect list of new interfaces. */ - shash_init(&new_ifaces); + sset_init(&new_ifaces); for (i = 0; i < cfg->n_interfaces; i++) { const char *name = cfg->interfaces[i]->name; - shash_add_once(&new_ifaces, name, NULL); + sset_add(&new_ifaces, name); } /* Get rid of deleted interfaces. */ - for (i = 0; i < port->n_ifaces; ) { - if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) { - iface_destroy(port->ifaces[i]); - } else { - i++; + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + if (!sset_contains(&new_ifaces, iface->name)) { + iface_destroy(iface); } } - shash_destroy(&new_ifaces); + sset_destroy(&new_ifaces); +} + +/* Expires all MAC learning entries associated with 'port' and forces ofproto + * to revalidate every flow. */ +static void +port_flush_macs(struct port *port) +{ + struct bridge *br = port->bridge; + struct mac_learning *ml = br->ml; + struct mac_entry *mac, *next_mac; + + bridge_flush(br); + LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { + if (mac->port.p == port) { + mac_learning_expire(ml, mac); + } + } } static void port_reconfigure(struct port *port, const struct ovsrec_port *cfg) { - const char *detect_mode; - struct shash new_ifaces; - long long int next_rebalance, miimon_next_update, lacp_priority; + struct sset new_ifaces; + bool need_flush = false; unsigned long *trunks; int vlan; size_t i; port->cfg = cfg; - /* Update settings. */ - port->updelay = cfg->bond_updelay; - if (port->updelay < 0) { - port->updelay = 0; - } - port->downdelay = cfg->bond_downdelay; - if (port->downdelay < 0) { - port->downdelay = 0; - } - port->bond_rebalance_interval = atoi( - get_port_other_config(cfg, "bond-rebalance-interval", "10000")); - if (port->bond_rebalance_interval < 1000) { - port->bond_rebalance_interval = 1000; - } - next_rebalance = time_msec() + port->bond_rebalance_interval; - if (port->bond_next_rebalance > next_rebalance) { - port->bond_next_rebalance = next_rebalance; - } - - detect_mode = get_port_other_config(cfg, "bond-detect-mode", - "carrier"); - - if (!strcmp(detect_mode, "carrier")) { - port->miimon = false; - } else if (!strcmp(detect_mode, "miimon")) { - port->miimon = true; - } else { - port->miimon = false; - VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to " - "carrier", port->name, detect_mode); - } - - port->bond_miimon_interval = atoi( - get_port_other_config(cfg, "bond-miimon-interval", "200")); - if (port->bond_miimon_interval < 100) { - port->bond_miimon_interval = 100; - } - miimon_next_update = time_msec() + port->bond_miimon_interval; - if (port->bond_miimon_next_update > miimon_next_update) { - port->bond_miimon_next_update = miimon_next_update; - } - - if (!port->cfg->bond_mode || - !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) { - port->bond_mode = BM_SLB; - } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) { - port->bond_mode = BM_AB; - } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) { - port->bond_mode = BM_TCP; - } else { - port->bond_mode = BM_SLB; - VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s", - port->name, port->cfg->bond_mode, - bond_mode_to_string(port->bond_mode)); - } /* Add new interfaces and update 'cfg' member of existing ones. */ - shash_init(&new_ifaces); + sset_init(&new_ifaces); for (i = 0; i < cfg->n_interfaces; i++) { const struct ovsrec_interface *if_cfg = cfg->interfaces[i]; struct iface *iface; - if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) { + if (!sset_add(&new_ifaces, if_cfg->name)) { VLOG_WARN("port %s: %s specified twice as port interface", port->name, if_cfg->name); iface_set_ofport(if_cfg, -1); @@ -4305,49 +2952,13 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal" : if_cfg->type[0] ? if_cfg->type : "system"); - - lacp_priority = - atoi(get_interface_other_config(if_cfg, "lacp-port-priority", - "0")); - - if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) { - iface->lacp_priority = UINT16_MAX; - } else { - iface->lacp_priority = lacp_priority; - } - } - shash_destroy(&new_ifaces); - - lacp_priority = - atoi(get_port_other_config(cfg, "lacp-system-priority", "0")); - - if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) { - /* Prefer bondable links if unspecified. */ - port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX; - } else { - port->lacp_priority = lacp_priority; - } - - if (!port->cfg->lacp) { - /* XXX when LACP implementation has been sufficiently tested, enable by - * default and make active on bonded ports. */ - port->lacp = 0; - } else if (!strcmp(port->cfg->lacp, "off")) { - port->lacp = 0; - } else if (!strcmp(port->cfg->lacp, "active")) { - port->lacp = LACP_ACTIVE; - } else if (!strcmp(port->cfg->lacp, "passive")) { - port->lacp = LACP_PASSIVE; - } else { - VLOG_WARN("port %s: unknown LACP mode %s", - port->name, port->cfg->lacp); - port->lacp = 0; } + sset_destroy(&new_ifaces); /* Get VLAN tag. */ vlan = -1; if (cfg->tag) { - if (port->n_ifaces < 2) { + if (list_is_short(&port->ifaces)) { vlan = *cfg->tag; if (vlan >= 0 && vlan <= 4095) { VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan); @@ -4363,45 +2974,30 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) } if (port->vlan != vlan) { port->vlan = vlan; - bridge_flush(port->bridge); + need_flush = true; } /* Get trunked VLANs. */ trunks = NULL; if (vlan < 0 && cfg->n_trunks) { - size_t n_errors; - - trunks = bitmap_allocate(4096); - n_errors = 0; - for (i = 0; i < cfg->n_trunks; i++) { - int trunk = cfg->trunks[i]; - if (trunk >= 0) { - bitmap_set1(trunks, trunk); - } else { - n_errors++; - } - } - if (n_errors) { - VLOG_ERR("port %s: invalid values for %zu trunk VLANs", - port->name, cfg->n_trunks); - } - if (n_errors == cfg->n_trunks) { + trunks = vlan_bitmap_from_array(cfg->trunks, cfg->n_trunks); + if (!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 - : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) { - bridge_flush(port->bridge); + if (!vlan_bitmap_equal(trunks, port->trunks)) { + need_flush = true; } bitmap_free(port->trunks); port->trunks = trunks; + + if (need_flush) { + port_flush_macs(port); + } } static void @@ -4409,12 +3005,9 @@ port_destroy(struct port *port) { if (port) { struct bridge *br = port->bridge; - struct port *del; + struct iface *iface, *next; int i; - proc_net_compat_update_vlan(port->name, NULL, 0); - proc_net_compat_update_bond(port->name, NULL); - for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; if (m && m->out_port == port) { @@ -4422,23 +3015,21 @@ port_destroy(struct port *port) } } - while (port->n_ifaces > 0) { - iface_destroy(port->ifaces[port->n_ifaces - 1]); + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + iface_destroy(iface); } - shash_find_and_delete_assert(&br->port_by_name, port->name); - - del = br->ports[port->port_idx] = br->ports[--br->n_ports]; - del->port_idx = port->port_idx; + hmap_remove(&br->ports, &port->hmap_node); VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); - netdev_monitor_destroy(port->monitor); - free(port->ifaces); + bond_destroy(port->bond); + lacp_destroy(port->lacp); + port_flush_macs(port); + bitmap_free(port->trunks); free(port->name); free(port); - bridge_flush(br); } } @@ -4452,321 +3043,221 @@ port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) static struct port * port_lookup(const struct bridge *br, const char *name) { - return shash_find_data(&br->port_by_name, name); -} - -static struct iface * -port_lookup_iface(const struct port *port, const char *name) -{ - struct iface *iface = iface_lookup(port->bridge, name); - return iface && iface->port == port ? iface : NULL; -} - -static void -port_update_lacp(struct port *port) -{ - size_t i; - bool key_changed; - - if (!port->lacp || port->n_ifaces < 1) { - return; - } - - key_changed = true; - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - - if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) { - port->lacp = 0; - return; - } + struct port *port; - if (iface->dp_ifidx == port->lacp_key) { - key_changed = false; + HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0), + &br->ports) { + if (!strcmp(port->name, name)) { + return port; } } - - if (key_changed) { - port->lacp_key = port->ifaces[0]->dp_ifidx; - } - - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - - iface->lacp_actor.sys_priority = htons(port->lacp_priority); - memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN); - - iface->lacp_actor.port_priority = htons(iface->lacp_priority); - iface->lacp_actor.portid = htons(iface->dp_ifidx); - iface->lacp_actor.key = htons(port->lacp_key); - - iface->lacp_tx = 0; - } - port->lacp_need_update = true; + return NULL; } -static void -port_update_bonding(struct port *port) +static bool +enable_lacp(struct port *port, bool *activep) { - if (port->monitor) { - netdev_monitor_destroy(port->monitor); - port->monitor = NULL; - } - if (port->n_ifaces < 2) { - /* Not a bonded port. */ - if (port->bond_hash) { - free(port->bond_hash); - port->bond_hash = NULL; - port->bond_compat_is_stale = true; - } - - port->bond_fake_iface = false; + if (!port->cfg->lacp) { + /* XXX when LACP implementation has been sufficiently tested, enable by + * default and make active on bonded ports. */ + return false; + } else if (!strcmp(port->cfg->lacp, "off")) { + return false; + } else if (!strcmp(port->cfg->lacp, "active")) { + *activep = true; + return true; + } else if (!strcmp(port->cfg->lacp, "passive")) { + *activep = false; + return true; } else { - size_t i; - - if (port->bond_mode != BM_AB && !port->bond_hash) { - port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); - for (i = 0; i <= BOND_MASK; i++) { - struct bond_entry *e = &port->bond_hash[i]; - e->iface_idx = -1; - e->tx_bytes = 0; - } - port->no_ifaces_tag = tag_create_random(); - bond_choose_active_iface(port); - port->bond_next_rebalance - = time_msec() + port->bond_rebalance_interval; - - if (port->cfg->bond_fake_iface) { - port->bond_next_fake_iface_update = time_msec(); - } - } else if (port->bond_mode == BM_AB) { - free(port->bond_hash); - port->bond_hash = NULL; - } - port->bond_compat_is_stale = true; - port->bond_fake_iface = port->cfg->bond_fake_iface; - - if (!port->miimon) { - port->monitor = netdev_monitor_create(); - for (i = 0; i < port->n_ifaces; i++) { - netdev_monitor_add(port->monitor, port->ifaces[i]->netdev); - } - } + VLOG_WARN("port %s: unknown LACP mode %s", + port->name, port->cfg->lacp); + return false; } } static void -port_update_bond_compat(struct port *port) +iface_reconfigure_lacp(struct iface *iface) { - struct compat_bond_hash compat_hashes[BOND_MASK + 1]; - struct compat_bond bond; - size_t i; + struct lacp_slave_settings s; + int priority, portid; - if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) { - proc_net_compat_update_bond(port->name, NULL); - return; - } - - bond.up = false; - bond.updelay = port->updelay; - bond.downdelay = port->downdelay; - - bond.n_hashes = 0; - bond.hashes = compat_hashes; - if (port->bond_hash) { - const struct bond_entry *e; - for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) { - if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { - struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++]; - cbh->hash = e - port->bond_hash; - cbh->netdev_name = port->ifaces[e->iface_idx]->name; - } - } - } - - bond.n_slaves = port->n_ifaces; - bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves); - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - struct compat_bond_slave *slave = &bond.slaves[i]; - slave->name = iface->name; + portid = atoi(get_interface_other_config(iface->cfg, "lacp-port-id", "0")); + priority = atoi(get_interface_other_config(iface->cfg, + "lacp-port-priority", "0")); - /* 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 (portid <= 0 || portid > UINT16_MAX) { + portid = iface->dp_ifidx; } - if (port->bond_fake_iface) { - struct netdev *bond_netdev; - - if (!netdev_open_default(port->name, &bond_netdev)) { - if (bond.up) { - netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); - } else { - netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); - } - netdev_close(bond_netdev); - } + if (priority <= 0 || priority > UINT16_MAX) { + priority = UINT16_MAX; } - proc_net_compat_update_bond(port->name, &bond); - free(bond.slaves); + s.name = iface->name; + s.id = portid; + s.priority = priority; + lacp_slave_register(iface->port->lacp, iface, &s); } static void -port_update_vlan_compat(struct port *port) +port_reconfigure_lacp(struct port *port) { - struct bridge *br = port->bridge; - char *vlandev_name = NULL; - - if (port->vlan > 0) { - /* Figure out the name that the VLAN device should actually have, if it - * existed. This takes some work because the VLAN device would not - * have port->name in its name; rather, it would have the trunk port's - * name, and 'port' would be attached to a bridge that also had the - * VLAN device one of its ports. So we need to find a trunk port that - * includes port->vlan. - * - * There might be more than one candidate. This doesn't happen on - * XenServer, so if it happens we just pick the first choice in - * alphabetical order instead of creating multiple VLAN devices. */ - size_t i; - for (i = 0; i < br->n_ports; i++) { - struct port *p = br->ports[i]; - if (port_trunks_vlan(p, port->vlan) - && p->n_ifaces - && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0)) - { - uint8_t ea[ETH_ADDR_LEN]; - netdev_get_etheraddr(p->ifaces[0]->netdev, ea); - if (!eth_addr_is_multicast(ea) && - !eth_addr_is_reserved(ea) && - !eth_addr_is_zero(ea)) { - vlandev_name = p->name; - } - } - } + static struct lacp_settings s; + struct iface *iface; + uint8_t sysid[ETH_ADDR_LEN]; + const char *sysid_str; + const char *lacp_time; + long long int custom_time; + int priority; + + if (!enable_lacp(port, &s.active)) { + lacp_destroy(port->lacp); + port->lacp = NULL; + return; } - proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan); -} - -/* Interface functions. */ -static void -iface_set_lacp_defaulted(struct iface *iface) -{ - memset(&iface->lacp_partner, 0xff, sizeof iface->lacp_partner); - iface->lacp_partner.state = 0; + sysid_str = get_port_other_config(port->cfg, "lacp-system-id", NULL); + if (sysid_str && eth_addr_from_string(sysid_str, sysid)) { + memcpy(s.id, sysid, ETH_ADDR_LEN); + } else { + memcpy(s.id, port->bridge->ea, ETH_ADDR_LEN); + } + + s.name = port->name; + + /* Prefer bondable links if unspecified. */ + priority = atoi(get_port_other_config(port->cfg, "lacp-system-priority", + "0")); + s.priority = (priority > 0 && priority <= UINT16_MAX + ? priority + : UINT16_MAX - !list_is_short(&port->ifaces)); + + s.strict = !strcmp(get_port_other_config(port->cfg, "lacp-strict", + "false"), + "true"); + + lacp_time = get_port_other_config(port->cfg, "lacp-time", "slow"); + custom_time = atoi(lacp_time); + if (!strcmp(lacp_time, "fast")) { + s.lacp_time = LACP_TIME_FAST; + } else if (!strcmp(lacp_time, "slow")) { + s.lacp_time = LACP_TIME_SLOW; + } else if (custom_time > 0) { + s.lacp_time = LACP_TIME_CUSTOM; + s.custom_time = custom_time; + } else { + s.lacp_time = LACP_TIME_SLOW; + } - iface->lacp_status = LACP_STATUS_DEFAULTED; - iface->lacp_tx = 0; - iface->port->lacp_need_update = true; -} + if (!port->lacp) { + port->lacp = lacp_create(); + } -static void -iface_set_lacp_expired(struct iface *iface) -{ - iface->lacp_status = LACP_STATUS_EXPIRED; - iface->lacp_partner.state |= LACP_STATE_TIME; - iface->lacp_partner.state &= ~LACP_STATE_SYNC; + lacp_configure(port->lacp, &s); - iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX; - iface->lacp_tx = 0; + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_reconfigure_lacp(iface); + } } -static uint8_t -iface_get_lacp_state(const struct iface *iface) +static void +port_reconfigure_bond(struct port *port) { - uint8_t state = 0; + struct bond_settings s; + const char *detect_s; + struct iface *iface; - if (iface->port->lacp & LACP_ACTIVE) { - state |= LACP_STATE_ACT; + if (list_is_short(&port->ifaces)) { + /* Not a bonded port. */ + bond_destroy(port->bond); + port->bond = NULL; + return; } - if (iface->lacp_status == LACP_STATUS_DEFAULTED) { - state |= LACP_STATE_DEF; - } else if (iface->lacp_attached) { - state |= LACP_STATE_SYNC; + port->bridge->has_bonded_ports = true; + + s.name = port->name; + s.balance = BM_SLB; + if (port->cfg->bond_mode + && !bond_mode_from_string(&s.balance, port->cfg->bond_mode)) { + VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s", + port->name, port->cfg->bond_mode, + bond_mode_to_string(s.balance)); } - if (iface->lacp_status == LACP_STATUS_EXPIRED) { - state |= LACP_STATE_EXP; + s.detect = BLSM_CARRIER; + detect_s = get_port_other_config(port->cfg, "bond-detect-mode", NULL); + if (detect_s && !bond_detect_mode_from_string(&s.detect, detect_s)) { + VLOG_WARN("port %s: unsupported bond-detect-mode %s, " + "defaulting to %s", + port->name, detect_s, bond_detect_mode_to_string(s.detect)); } - if (iface->port->n_ifaces > 1) { - state |= LACP_STATE_AGG; + s.miimon_interval = atoi( + get_port_other_config(port->cfg, "bond-miimon-interval", "200")); + if (s.miimon_interval < 100) { + s.miimon_interval = 100; } - if (iface->enabled) { - state |= LACP_STATE_COL | LACP_STATE_DIST; + s.up_delay = MAX(0, port->cfg->bond_updelay); + s.down_delay = MAX(0, port->cfg->bond_downdelay); + s.rebalance_interval = atoi( + get_port_other_config(port->cfg, "bond-rebalance-interval", "10000")); + if (s.rebalance_interval < 1000) { + s.rebalance_interval = 1000; } - return state; -} + s.fake_iface = port->cfg->bond_fake_iface; -/* Given 'iface', populates 'priority' with data representing its LACP link - * priority. If two priority objects populated by this function are compared - * using memcmp, the higher priority link will be less than the lower priority - * link. */ -static void -iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority) -{ - uint16_t partner_priority, actor_priority; - - /* Choose the lacp_info of the higher priority system by comparing their - * system priorities and mac addresses. */ - actor_priority = ntohs(iface->lacp_actor.sys_priority); - partner_priority = ntohs(iface->lacp_partner.sys_priority); - if (actor_priority < partner_priority) { - *priority = iface->lacp_actor; - } else if (partner_priority < actor_priority) { - *priority = iface->lacp_partner; - } else if (eth_addr_compare_3way(iface->lacp_actor.sysid, - iface->lacp_partner.sysid) < 0) { - *priority = iface->lacp_actor; + if (!port->bond) { + port->bond = bond_create(&s); } else { - *priority = iface->lacp_partner; + if (bond_reconfigure(port->bond, &s)) { + bridge_flush(port->bridge); + } } - /* Key and state are not used in priority comparisons. */ - priority->key = 0; - priority->state = 0; + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + uint16_t stable_id = (port->lacp + ? lacp_slave_get_port_id(port->lacp, iface) + : iface->dp_ifidx); + bond_slave_register(iface->port->bond, iface, stable_id, + iface->netdev); + } } static void -iface_send_packet(struct iface *iface, struct ofpbuf *packet) +port_send_learning_packets(struct port *port) { - struct flow flow; - union ofp_action action; - - memset(&action, 0, sizeof action); - action.output.type = htons(OFPAT_OUTPUT); - action.output.len = htons(sizeof action); - action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx)); + struct bridge *br = port->bridge; + int error, n_packets, n_errors; + struct mac_entry *e; - flow_extract(packet, 0, ODPP_NONE, &flow); + error = n_packets = n_errors = 0; + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { + if (e->port.p != port) { + int ret = bond_send_learning_packet(port->bond, e->mac, e->vlan); + if (ret) { + error = ret; + n_errors++; + } + n_packets++; + } + } - if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1, - packet)) { + if (n_errors) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name); + VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " + "packets, last error was: %s", + port->name, n_errors, n_packets, strerror(error)); + } else { + VLOG_DBG("bond %s: sent %d gratuitous learning packets", + port->name, n_packets); } } + +/* Interface functions. */ static struct iface * iface_create(struct port *port, const struct ovsrec_interface *if_cfg) @@ -4777,29 +3268,15 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) iface = xzalloc(sizeof *iface); iface->port = port; - iface->port_ifidx = port->n_ifaces; iface->name = xstrdup(name); iface->dp_ifidx = -1; iface->tag = tag_create_random(); - iface->delay_expires = LLONG_MAX; iface->netdev = NULL; iface->cfg = if_cfg; - iface_set_lacp_defaulted(iface); - - if (port->lacp & LACP_ACTIVE) { - iface_set_lacp_expired(iface); - } shash_add_assert(&br->iface_by_name, iface->name, iface); - if (port->n_ifaces >= port->allocated_ifaces) { - port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, - sizeof *port->ifaces); - } - port->ifaces[port->n_ifaces++] = iface; - if (port->n_ifaces > 1) { - br->has_bonded_ports = true; - } + list_push_back(&port->ifaces, &iface->port_elem); VLOG_DBG("attached network device %s to port %s", iface->name, port->name); @@ -4814,11 +3291,13 @@ iface_destroy(struct iface *iface) if (iface) { struct port *port = iface->port; struct bridge *br = port->bridge; - bool del_active = port->active_iface == iface->port_ifidx; - struct iface *del; - if (port->monitor) { - netdev_monitor_remove(port->monitor, iface->netdev); + if (port->bond) { + bond_slave_unregister(port->bond, iface); + } + + if (port->lacp) { + lacp_slave_unregister(port->lacp, iface); } shash_find_and_delete_assert(&br->iface_by_name, iface->name); @@ -4827,19 +3306,10 @@ iface_destroy(struct iface *iface) hmap_remove(&br->ifaces, &iface->dp_ifidx_node); } - del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces]; - del->port_ifidx = iface->port_ifidx; + list_remove(&iface->port_elem); netdev_close(iface->netdev); - if (del_active) { - ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); - bond_choose_active_iface(port); - bond_send_learning_packets(port); - } - - cfm_destroy(iface->cfm); - free(iface->name); free(iface); @@ -4853,6 +3323,21 @@ iface_lookup(const struct bridge *br, const char *name) return shash_find_data(&br->iface_by_name, name); } +static struct iface * +iface_find(const char *name) +{ + const struct bridge *br; + + LIST_FOR_EACH (br, node, &all_bridges) { + struct iface *iface = iface_lookup(br, name); + + if (iface) { + return iface; + } + } + return NULL; +} + static struct iface * iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) { @@ -4874,13 +3359,15 @@ 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)) { + if (!strcmp(iface->type, "internal") + && iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) { + if (iface->dp_ifidx == ODPP_LOCAL) { + VLOG_ERR("interface %s: ignoring mac in Interface record " + "(use Bridge record to set local port's mac)", + iface->name); + } else 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) { @@ -4895,7 +3382,7 @@ iface_set_mac(struct iface *iface) static void iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport) { - if (if_cfg) { + if (if_cfg && !ovsdb_idl_row_is_synthetic(&if_cfg->header_)) { ovsrec_interface_set_ofport(if_cfg, &ofport, 1); } } @@ -5013,20 +3500,15 @@ static void iface_update_cfm(struct iface *iface) { size_t i; - struct cfm *cfm; + struct cfm cfm; uint16_t *remote_mps; struct ovsrec_monitor *mon; - uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN]; + uint8_t maid[CCM_MAID_LEN]; mon = iface->cfg->monitor; if (!mon) { - return; - } - - if (netdev_get_etheraddr(iface->netdev, ea)) { - VLOG_WARN("interface %s: Failed to get ethernet address. " - "Skipping Monitor.", iface->name); + ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); return; } @@ -5035,28 +3517,38 @@ iface_update_cfm(struct iface *iface) return; } - if (!iface->cfm) { - iface->cfm = cfm_create(); - } - - cfm = iface->cfm; - cfm->mpid = mon->mpid; - cfm->interval = mon->interval ? *mon->interval : 1000; + cfm.mpid = mon->mpid; + cfm.interval = mon->interval ? *mon->interval : 1000; - memcpy(cfm->eth_src, ea, sizeof cfm->eth_src); - memcpy(cfm->maid, maid, sizeof cfm->maid); + memcpy(cfm.maid, maid, sizeof cfm.maid); remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps); for(i = 0; i < mon->n_remote_mps; i++) { remote_mps[i] = mon->remote_mps[i]->mpid; } - cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps); + + ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx, + &cfm, remote_mps, mon->n_remote_mps); free(remote_mps); +} - if (!cfm_configure(iface->cfm)) { - cfm_destroy(iface->cfm); - iface->cfm = NULL; - } +/* Read carrier or miimon status directly from 'iface''s netdev, according to + * how 'iface''s port is configured. + * + * Returns true if 'iface' is up, false otherwise. */ +static bool +iface_get_carrier(const struct iface *iface) +{ + /* XXX */ + return netdev_get_carrier(iface->netdev); +} + +/* Returns true if 'iface' is synthetic, that is, if we constructed it locally + * instead of obtaining it from the database. */ +static bool +iface_is_synthetic(const struct iface *iface) +{ + return ovsdb_idl_row_is_synthetic(&iface->cfg->header_); } /* Port mirroring. */ @@ -5079,6 +3571,7 @@ static void mirror_reconfigure(struct bridge *br) { unsigned long *rspan_vlans; + struct port *port; int i; /* Get rid of deleted mirrors. */ @@ -5108,8 +3601,8 @@ mirror_reconfigure(struct bridge *br) } /* Update port reserved status. */ - for (i = 0; i < br->n_ports; i++) { - br->ports[i]->is_mirror_output_port = false; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port->is_mirror_output_port = false; } for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; @@ -5121,22 +3614,12 @@ mirror_reconfigure(struct bridge *br) /* 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); - } - } + rspan_vlans = vlan_bitmap_from_array(br->cfg->flood_vlans, + br->cfg->n_flood_vlans); } if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) { bridge_flush(br); + mac_learning_flush(br->ml); } } @@ -5159,13 +3642,15 @@ mirror_create(struct bridge *br, struct ovsrec_mirror *cfg) VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name); bridge_flush(br); + mac_learning_flush(br->ml); br->mirrors[i] = m = xzalloc(sizeof *m); + m->uuid = cfg->header_.uuid; m->bridge = br; m->idx = i; m->name = xstrdup(cfg->name); - shash_init(&m->src_ports); - shash_init(&m->dst_ports); + sset_init(&m->src_ports); + sset_init(&m->dst_ports); m->vlans = NULL; m->n_vlans = 0; m->out_vlan = -1; @@ -5179,15 +3664,15 @@ mirror_destroy(struct mirror *m) { if (m) { struct bridge *br = m->bridge; - size_t i; + struct port *port; - for (i = 0; i < br->n_ports; i++) { - br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); - br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); + port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); } - shash_destroy(&m->src_ports); - shash_destroy(&m->dst_ports); + sset_destroy(&m->src_ports); + sset_destroy(&m->dst_ports); free(m->vlans); m->bridge->mirrors[m->idx] = NULL; @@ -5195,19 +3680,20 @@ mirror_destroy(struct mirror *m) free(m); bridge_flush(br); + mac_learning_flush(br->ml); } } static void mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports, - struct shash *names) + struct sset *names) { size_t i; for (i = 0; i < n_ports; i++) { const char *name = ports[i]->name; if (port_lookup(m->bridge, name)) { - shash_add_once(names, name, NULL); + sset_add(names, name); } else { VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent " "port %s", m->bridge->name, m->name, name); @@ -5249,29 +3735,16 @@ vlan_is_mirrored(const struct mirror *m, int vlan) return false; } -static bool -port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) -{ - size_t i; - - for (i = 0; i < m->n_vlans; i++) { - if (port_trunks_vlan(p, m->vlans[i])) { - return true; - } - } - return false; -} - static void mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) { - struct shash src_ports, dst_ports; + struct sset src_ports, dst_ports; mirror_mask_t mirror_bit; struct port *out_port; + struct port *port; int out_vlan; size_t n_vlans; int *vlans; - size_t i; /* Set name. */ if (strcmp(cfg->name, m->name)) { @@ -5305,13 +3778,12 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) return; } - shash_init(&src_ports); - shash_init(&dst_ports); + sset_init(&src_ports); + sset_init(&dst_ports); if (cfg->select_all) { - for (i = 0; i < m->bridge->n_ports; i++) { - const char *name = m->bridge->ports[i]->name; - shash_add_once(&src_ports, name, NULL); - shash_add_once(&dst_ports, name, NULL); + HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) { + sset_add(&src_ports, port->name); + sset_add(&dst_ports, port->name); } vlans = NULL; n_vlans = 0; @@ -5327,16 +3799,17 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) } /* Update mirror data. */ - if (!shash_equal_keys(&m->src_ports, &src_ports) - || !shash_equal_keys(&m->dst_ports, &dst_ports) + if (!sset_equals(&m->src_ports, &src_ports) + || !sset_equals(&m->dst_ports, &dst_ports) || m->n_vlans != n_vlans || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans) || m->out_port != out_port || m->out_vlan != out_vlan) { bridge_flush(m->bridge); + mac_learning_flush(m->bridge->ml); } - shash_swap(&m->src_ports, &src_ports); - shash_swap(&m->dst_ports, &dst_ports); + sset_swap(&m->src_ports, &src_ports); + sset_swap(&m->dst_ports, &dst_ports); free(m->vlans); m->vlans = vlans; m->n_vlans = n_vlans; @@ -5345,20 +3818,14 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) /* Update ports. */ mirror_bit = MIRROR_MASK_C(1) << m->idx; - for (i = 0; i < m->bridge->n_ports; i++) { - struct port *port = m->bridge->ports[i]; - - if (shash_find(&m->src_ports, port->name) - || (m->n_vlans - && (!port->vlan - ? port_trunks_any_mirrored_vlan(m, port) - : vlan_is_mirrored(m, port->vlan)))) { + HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) { + if (sset_contains(&m->src_ports, port->name)) { port->src_mirrors |= mirror_bit; } else { port->src_mirrors &= ~mirror_bit; } - if (shash_find(&m->dst_ports, port->name)) { + if (sset_contains(&m->dst_ports, port->name)) { port->dst_mirrors |= mirror_bit; } else { port->dst_mirrors &= ~mirror_bit; @@ -5366,6 +3833,6 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) } /* Clean up. */ - shash_destroy(&src_ports); - shash_destroy(&dst_ports); + sset_destroy(&src_ports); + sset_destroy(&dst_ports); }