X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=d99d6a1cd4746e5c6ad4c3635827f63ac7265234;hb=8402c74b186e28c53ad51fc8813aebe64a12cd7c;hp=4cf1b86f8f4931d57a34e23788e101b83a72b2a5;hpb=2280e7223cc5d014fe60ad3be45b8e4d9d401997;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 4cf1b86f8..d99d6a1cd 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -1,4 +1,4 @@ -/* Copyright (c) 2008, 2009 Nicira Networks +/* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -17,775 +17,1000 @@ #include "bridge.h" #include #include -#include -#include #include -#include -#include -#include #include -#include -#include -#include -#include -#include #include "bitmap.h" +#include "bond.h" +#include "cfm.h" #include "coverage.h" +#include "daemon.h" #include "dirs.h" -#include "dpif.h" #include "dynamic-string.h" -#include "flow.h" #include "hash.h" +#include "hmap.h" +#include "jsonrpc.h" +#include "lacp.h" #include "list.h" -#include "mac-learning.h" #include "netdev.h" -#include "odp-util.h" #include "ofp-print.h" #include "ofpbuf.h" -#include "ofproto/netflow.h" #include "ofproto/ofproto.h" -#include "packets.h" #include "poll-loop.h" -#include "port-array.h" -#include "proc-net-compat.h" -#include "process.h" #include "sha1.h" #include "shash.h" #include "socket-util.h" -#include "svec.h" +#include "stream-ssl.h" +#include "sset.h" +#include "system-stats.h" #include "timeval.h" #include "util.h" #include "unixctl.h" -#include "vconn.h" -#include "vconn-ssl.h" #include "vswitchd/vswitch-idl.h" #include "xenserver.h" -#include "xtoxll.h" - -#define THIS_MODULE VLM_bridge #include "vlog.h" +#include "sflow_api.h" +#include "vlan-bitmap.h" -struct dst { - uint16_t vlan; - uint16_t dp_ifidx; -}; +VLOG_DEFINE_THIS_MODULE(bridge); + +COVERAGE_DEFINE(bridge_reconfigure); struct iface { /* These members are always valid. */ + struct list port_elem; /* Element in struct port's "ifaces" list. */ + struct hmap_node name_node; /* In struct bridge's "iface_by_name" hmap. */ 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.*/ - int dp_ifidx; /* Index within kernel datapath. */ + * be initialized. */ + struct hmap_node ofp_port_node; /* In struct bridge's "ifaces" hmap. */ + int ofp_port; /* OpenFlow port number, -1 if unknown. */ struct netdev *netdev; /* Network device. */ - bool enabled; /* May be chosen for flows? */ - - /* This member is only valid *during* bridge_reconfigure(). */ + const char *type; /* Usually same as cfg->type. */ const struct ovsrec_interface *cfg; }; -#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. */ -}; - -#define MAX_MIRRORS 32 -typedef uint32_t mirror_mask_t; -#define MIRROR_MASK_C(X) UINT32_C(X) -BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); struct mirror { + struct uuid uuid; /* UUID of this "mirror" record in database. */ + struct hmap_node hmap_node; /* In struct bridge's "mirrors" hmap. */ struct bridge *bridge; - size_t idx; char *name; - - /* Selection criteria. */ - struct svec src_ports; - struct svec dst_ports; - int *vlans; - size_t n_vlans; - - /* Output. */ - struct port *out_port; - int out_vlan; }; -#define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ struct port { struct bridge *bridge; - size_t port_idx; - int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ - unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */ + struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */ char *name; + const struct ovsrec_port *cfg; + /* 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; - - /* Bonding info. */ - struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ - 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? */ - - /* Port mirroring info. */ - mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ - mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ - bool is_mirror_output_port; /* Does port mirroring send frames here? */ - - /* This member is only valid *during* bridge_reconfigure(). */ - const struct ovsrec_port *cfg; + struct list ifaces; /* List of "struct iface"s. */ }; -#define DP_MAX_PORTS 255 struct bridge { - struct list node; /* Node in global list of bridges. */ + struct hmap_node node; /* In 'all_bridges'. */ char *name; /* User-specified arbitrary name. */ - struct mac_learning *ml; /* MAC learning table. */ - bool sent_config_request; /* Successfully sent config request? */ + char *type; /* Datapath type. */ + uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */ uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */ - - /* Support for remote controllers. */ - char *controller; /* NULL if there is no remote controller; - * "discover" to do controller discovery; - * otherwise a vconn name. */ + const struct ovsrec_bridge *cfg; /* OpenFlow switch processing. */ struct ofproto *ofproto; /* OpenFlow switch. */ - /* Kernel datapath information. */ - struct dpif *dpif; /* Datapath. */ - struct port_array ifaces; /* Indexed by kernel datapath port number. */ - /* Bridge ports. */ - struct port **ports; - size_t n_ports, allocated_ports; - - /* Bonding. */ - bool has_bonded_ports; - long long int bond_next_rebalance; - - /* Flow tracking. */ - bool flush; - - /* Flow statistics gathering. */ - time_t next_stats_request; + struct hmap ports; /* "struct port"s indexed by name. */ + struct hmap ifaces; /* "struct iface"s indexed by ofp_port. */ + struct hmap iface_by_name; /* "struct iface"s indexed by name. */ /* Port mirroring. */ - struct mirror *mirrors[MAX_MIRRORS]; + struct hmap mirrors; /* "struct mirror" indexed by UUID. */ - /* This member is only valid *during* bridge_reconfigure(). */ - const struct ovsrec_bridge *cfg; + /* 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. */ -static struct list all_bridges = LIST_INITIALIZER(&all_bridges); +/* All bridges, indexed by name. */ +static struct hmap all_bridges = HMAP_INITIALIZER(&all_bridges); + +/* OVSDB IDL used to obtain configuration. */ +static struct ovsdb_idl *idl; + +/* Each time this timer expires, the bridge fetches systems and interface + * statistics and pushes them into the database. */ +#define STATS_INTERVAL (5 * 1000) /* In milliseconds. */ +static long long int stats_timer = LLONG_MIN; -/* Maximum number of datapaths. */ -enum { DP_MAX = 256 }; +/* 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 char *name); +static void add_del_bridges(const struct ovsrec_open_vswitch *); +static void bridge_del_ofprotos(void); +static bool bridge_add_ofprotos(struct bridge *); +static void bridge_create(const struct ovsrec_bridge *); static void bridge_destroy(struct bridge *); static struct bridge *bridge_lookup(const char *name); static unixctl_cb_func bridge_unixctl_dump_flows; -static int bridge_run_one(struct bridge *); -static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *, - struct bridge *); -static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *, - struct bridge *); -static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces); -static void bridge_fetch_dp_ifaces(struct bridge *); -static void bridge_flush(struct bridge *); +static unixctl_cb_func bridge_unixctl_reconnect; +static size_t bridge_get_controllers(const struct bridge *br, + struct ovsrec_controller ***controllersp); +static void bridge_add_del_ports(struct bridge *); +static void bridge_add_ofproto_ports(struct bridge *); +static void bridge_del_ofproto_ports(struct bridge *); +static void bridge_refresh_ofp_port(struct bridge *); +static void bridge_configure_datapath_id(struct bridge *); +static void bridge_configure_flow_eviction_threshold(struct bridge *); +static void bridge_configure_netflow(struct bridge *); +static void bridge_configure_forward_bpdu(struct bridge *); +static void bridge_configure_sflow(struct bridge *, int *sflow_bridge_number); +static void bridge_configure_remotes(struct bridge *, + const struct sockaddr_in *managers, + size_t n_managers); static void bridge_pick_local_hw_addr(struct bridge *, uint8_t ea[ETH_ADDR_LEN], struct iface **hw_addr_iface); 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 bool bridge_has_bond_fake_iface(const struct bridge *, + const char *name); +static bool port_is_bond_fake_iface(const struct port *); -static unixctl_cb_func bridge_unixctl_fdb_show; - -static void bond_init(void); -static void bond_run(struct bridge *); -static void bond_wait(struct bridge *); -static void bond_rebalance_port(struct port *); -static void bond_send_learning_packets(struct port *); -static void bond_enable_slave(struct iface *iface, bool enable); +static unixctl_cb_func qos_unixctl_show; -static struct port *port_create(struct bridge *, const char *name); -static void port_reconfigure(struct port *, const struct ovsrec_port *); +static struct port *port_create(struct bridge *, const struct ovsrec_port *); +static void port_add_ifaces(struct port *); +static void port_del_ifaces(struct 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 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 *); - -#if 0 -static void mirror_create(struct bridge *, const char *name); +static void port_configure(struct port *); +static struct lacp_settings *port_configure_lacp(struct port *, + struct lacp_settings *); +static void port_configure_bond(struct port *, struct bond_settings *, + uint32_t *bond_stable_ids); + +static void bridge_configure_mirrors(struct bridge *); +static struct mirror *mirror_create(struct bridge *, + const struct ovsrec_mirror *); static void mirror_destroy(struct mirror *); -static void mirror_reconfigure(struct bridge *); -static void mirror_reconfigure_one(struct mirror *); -static bool vlan_is_mirrored(const struct mirror *, int vlan); -#else -static bool vlan_is_mirrored(const struct mirror *m UNUSED, int vlan UNUSED) -{ - return false; -} -#endif +static bool mirror_configure(struct mirror *, const struct ovsrec_mirror *); -static struct iface *iface_create(struct port *port, +static void iface_configure_lacp(struct iface *, struct lacp_slave_settings *); +static struct iface *iface_create(struct port *port, const struct ovsrec_interface *if_cfg); static void iface_destroy(struct iface *); static struct iface *iface_lookup(const struct bridge *, const char *name); -static struct iface *iface_from_dp_ifidx(const struct bridge *, - uint16_t dp_ifidx); -static bool iface_is_internal(const struct bridge *, const char *name); +static struct iface *iface_find(const char *name); +static struct iface *iface_from_ofp_port(const struct bridge *, + uint16_t ofp_port); static void iface_set_mac(struct iface *); - -/* Hooks into ofproto processing. */ -static struct ofhooks bridge_ofhooks; +static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport); +static void iface_configure_qos(struct iface *, const struct ovsrec_qos *); +static void iface_configure_cfm(struct iface *); +static bool iface_refresh_cfm_stats(struct iface *); +static void iface_refresh_stats(struct iface *); +static void iface_refresh_status(struct 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); /* Public functions. */ -/* Adds the name of each interface used by a bridge, including local and - * internal ports, to 'svec'. */ +/* Initializes the bridge module, configuring it to obtain its configuration + * from an OVSDB server accessed over 'remote', which should be a string in a + * form acceptable to ovsdb_idl_create(). */ void -bridge_get_ifaces(struct svec *svec) +bridge_init(const char *remote) { - struct bridge *br, *next; - size_t i, j; + /* Create connection to database. */ + idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true); + ovsdb_idl_set_lock(idl, "ovs_vswitchd"); - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + ovsdb_idl_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); - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->dp_ifidx < 0) { - VLOG_ERR("%s interface not in datapath %s, ignoring", - iface->name, dpif_name(br->dpif)); - } else { - if (iface->dp_ifidx != ODPP_LOCAL) { - svec_add(svec, iface->name); - } - } - } - } - } -} + ovsdb_idl_omit_alert(idl, &ovsrec_bridge_col_datapath_id); + ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); -void -bridge_init(const struct ovsrec_open_vswitch *cfg) -{ - struct svec bridge_names; - struct svec dpif_names; - size_t i; + ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids); + ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge); - unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + 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); - svec_init(&bridge_names); - for (i = 0; i < cfg->n_bridges; i++) { - svec_add(&bridge_names, cfg->bridges[i]->name); - } - svec_sort(&bridge_names); - - svec_init(&dpif_names); - dp_enumerate(&dpif_names); - for (i = 0; i < dpif_names.n; i++) { - const char *dpif_name = dpif_names.names[i]; - struct dpif *dpif; - int retval; - - retval = dpif_open(dpif_name, &dpif); - if (!retval) { - struct svec all_names; - size_t j; - - svec_init(&all_names); - dpif_get_all_names(dpif, &all_names); - for (j = 0; j < all_names.n; j++) { - if (svec_contains(&bridge_names, all_names.names[j])) { - goto found; - } - } - dpif_delete(dpif); - found: - svec_destroy(&all_names); - dpif_close(dpif); - } - } - svec_destroy(&dpif_names); + 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(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("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(); - bridge_reconfigure(cfg); + cfm_init(); } -#ifdef HAVE_OPENSSL -static bool -config_string_change(const char *value, char **valuep) +void +bridge_exit(void) { - if (value && (!*valuep || strcmp(value, *valuep))) { - free(*valuep); - *valuep = xstrdup(value); - return true; - } else { - return false; + struct bridge *br, *next_br; + + HMAP_FOR_EACH_SAFE (br, next_br, node, &all_bridges) { + bridge_destroy(br); } + ovsdb_idl_destroy(idl); } +/* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP + * addresses and ports into '*managersp' and '*n_managersp'. The caller is + * responsible for freeing '*managersp' (with free()). + * + * You may be asking yourself "why does ovs-vswitchd care?", because + * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd + * should not be and in fact is not directly involved in that. But + * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so + * it has to tell in-band control where the managers are to enable that. + * (Thus, only managers connected in-band are collected.) + */ static void -bridge_configure_ssl(const struct ovsrec_ssl *ssl) +collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg, + struct sockaddr_in **managersp, size_t *n_managersp) { - /* XXX SSL should be configurable on a per-bridge basis. - * XXX should be possible to de-configure SSL. */ - static char *private_key_file; - static char *certificate_file; - static char *cacert_file; - struct stat s; + struct sockaddr_in *managers = NULL; + size_t n_managers = 0; + struct sset targets; + size_t i; - if (!ssl) { - /* XXX We can't un-set SSL settings. */ - return; - } + /* 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 (config_string_change(ssl->private_key, &private_key_file)) { - vconn_ssl_set_private_key_file(private_key_file); + if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) { + sset_find_and_delete(&targets, m->target); + } else { + sset_add(&targets, m->target); + } } - if (config_string_change(ssl->certificate, &certificate_file)) { - vconn_ssl_set_certificate_file(certificate_file); - } + /* Now extract the targets' IP addresses. */ + if (!sset_is_empty(&targets)) { + const char *target; + + managers = xmalloc(sset_count(&targets) * sizeof *managers); + SSET_FOR_EACH (target, &targets) { + struct sockaddr_in *sin = &managers[n_managers]; - /* We assume that even if the filename hasn't changed, if the CA cert - * file has been removed, that we want to move back into - * boot-strapping mode. This opens a small security hole, because - * the old certificate will still be trusted until vSwitch is - * restarted. We may want to address this in vconn's SSL library. */ - if (config_string_change(ssl->ca_cert, &cacert_file) - || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) { - vconn_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert); + if ((!strncmp(target, "tcp:", 4) + && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) || + (!strncmp(target, "ssl:", 4) + && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) { + n_managers++; + } + } } + sset_destroy(&targets); + + *managersp = managers; + *n_managersp = n_managers; } -#endif -/* Attempt to create the network device 'iface_name' through the netdev - * library. */ -static int -set_up_iface(const struct ovsrec_interface *iface_cfg, bool create) +static void +bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) { - struct shash_node *node; - struct shash options; - int error; - size_t i; + struct sockaddr_in *managers; + struct bridge *br, *next; + int sflow_bridge_number; + size_t n_managers; + + COVERAGE_INC(bridge_reconfigure); - /* If a type is not explicitly declared, then assume it's an existing - * "system" device. */ - if (iface_cfg->type[0] == '\0' || !strcmp(iface_cfg->type, "system")) { - return 0; + /* Create and destroy "struct bridge"s, "struct port"s, and "struct + * iface"s according to 'ovs_cfg', with only very minimal configuration + * otherwise. + * + * This is purely an update to bridge data structures. Nothing is pushed + * down to ofproto or lower layers. */ + add_del_bridges(ovs_cfg); + HMAP_FOR_EACH (br, node, &all_bridges) { + bridge_add_del_ports(br); } - shash_init(&options); - for (i = 0; i < iface_cfg->n_options; i++) { - shash_add(&options, iface_cfg->key_options[i], - xstrdup(iface_cfg->value_options[i])); + /* Delete all datapaths and datapath ports that are no longer configured. + * + * The kernel will reject any attempt to add a given port to a datapath if + * that port already belongs to a different datapath, so we must do all + * port deletions before any port additions. A datapath always has a + * "local port" so we must delete not-configured datapaths too. */ + bridge_del_ofprotos(); + HMAP_FOR_EACH (br, node, &all_bridges) { + if (br->ofproto) { + bridge_del_ofproto_ports(br); + } } - if (create) { - error = netdev_create(iface_cfg->name, iface_cfg->type, &options); - } else { - /* xxx Check to make sure that the type hasn't changed. */ - error = netdev_reconfigure(iface_cfg->name, &options); + /* Create datapaths and datapath ports that are missing. + * + * After this is done, we have our final set of bridges, ports, and + * interfaces. Every "struct bridge" has an ofproto, every "struct port" + * has at least one iface, every "struct iface" has a valid ofp_port and + * netdev. */ + HMAP_FOR_EACH_SAFE (br, next, node, &all_bridges) { + if (!br->ofproto && !bridge_add_ofprotos(br)) { + bridge_destroy(br); + } } + HMAP_FOR_EACH (br, node, &all_bridges) { + bridge_refresh_ofp_port(br); + bridge_add_ofproto_ports(br); + } + + /* Complete the configuration. */ + sflow_bridge_number = 0; + collect_in_band_managers(ovs_cfg, &managers, &n_managers); + HMAP_FOR_EACH (br, node, &all_bridges) { + struct port *port; - SHASH_FOR_EACH (node, &options) { - free(node->data); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + port_configure(port); + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_configure_cfm(iface); + iface_configure_qos(iface, port->cfg->qos); + iface_set_mac(iface); + } + } + bridge_configure_mirrors(br); + bridge_configure_datapath_id(br); + bridge_configure_flow_eviction_threshold(br); + bridge_configure_forward_bpdu(br); + bridge_configure_remotes(br, managers, n_managers); + bridge_configure_netflow(br); + bridge_configure_sflow(br, &sflow_bridge_number); } - shash_destroy(&options); + free(managers); - return error; + /* ovs-vswitchd has completed initialization, so allow the process that + * forked us to exit successfully. */ + daemonize_complete(); } -static int -reconfigure_iface(const struct ovsrec_interface *iface_cfg) -{ - return set_up_iface(iface_cfg, false); +/* Iterate over all ofprotos and delete any of them that do not have a + * configured bridge or that are the wrong type. */ +static void +bridge_del_ofprotos(void) +{ + struct sset names; + struct sset types; + const char *type; + + sset_init(&names); + sset_init(&types); + ofproto_enumerate_types(&types); + SSET_FOR_EACH (type, &types) { + const char *name; + + ofproto_enumerate_names(type, &names); + SSET_FOR_EACH (name, &names) { + struct bridge *br = bridge_lookup(name); + if (!br || strcmp(type, br->type)) { + ofproto_delete(name, type); + } + } + } + sset_destroy(&names); + sset_destroy(&types); } - -/* iterate_and_prune_ifaces() callback function that opens the network device - * for 'iface', if it is not already open, and retrieves the interface's MAC - * address and carrier status. */ static bool -init_iface_netdev(struct bridge *br UNUSED, struct iface *iface, - void *aux UNUSED) +bridge_add_ofprotos(struct bridge *br) { - if (iface->netdev) { - return true; - } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE, - &iface->netdev)) { - netdev_get_carrier(iface->netdev, &iface->enabled); - return true; - } else { - /* If the network device can't be opened, then we're not going to try - * to do anything with this interface. */ + int error = ofproto_create(br->name, br->type, &br->ofproto); + if (error) { + VLOG_ERR("failed to create bridge %s: %s", br->name, strerror(error)); return false; } + return true; } -static bool -check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED) +static void +port_configure(struct port *port) { - if (iface->dp_ifidx >= 0) { - VLOG_DBG("%s has interface %s on port %d", - dpif_name(br->dpif), - iface->name, iface->dp_ifidx); - return true; - } else { - VLOG_ERR("%s interface not in %s, dropping", - iface->name, dpif_name(br->dpif)); - return false; + const struct ovsrec_port *cfg = port->cfg; + struct bond_settings bond_settings; + struct lacp_settings lacp_settings; + struct ofproto_bundle_settings s; + struct iface *iface; + + /* Get name. */ + s.name = port->name; + + /* Get slaves. */ + s.n_slaves = 0; + s.slaves = xmalloc(list_size(&port->ifaces) * sizeof *s.slaves); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + s.slaves[s.n_slaves++] = iface->ofp_port; } -} -static bool -set_iface_properties(struct bridge *br UNUSED, struct iface *iface, - void *aux UNUSED) -{ - /* Set policing attributes. */ - netdev_set_policing(iface->netdev, - iface->cfg->ingress_policing_rate, - iface->cfg->ingress_policing_burst); + /* Get VLAN tag. */ + s.vlan = -1; + if (cfg->tag) { + if (list_is_short(&port->ifaces)) { + if (*cfg->tag >= 0 && *cfg->tag <= 4095) { + s.vlan = *cfg->tag; + VLOG_DBG("port %s: assigning VLAN tag %d", port->name, s.vlan); + } + } else { + /* It's possible that bonded, VLAN-tagged ports make sense. Maybe + * they even work as-is. But they have not been tested. */ + VLOG_WARN("port %s: VLAN tags not supported on bonded ports", + port->name); + } + } - /* Set MAC address of internal interfaces other than the local - * interface. */ - if (iface->dp_ifidx != ODPP_LOCAL - && iface_is_internal(br, iface->name)) { - iface_set_mac(iface); + /* Get VLAN trunks. */ + s.trunks = NULL; + if (s.vlan < 0 && cfg->n_trunks) { + s.trunks = vlan_bitmap_from_array(cfg->trunks, cfg->n_trunks); + } else if (s.vlan >= 0 && cfg->n_trunks) { + VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", + port->name); } - return true; -} + /* Get LACP settings. */ + s.lacp = port_configure_lacp(port, &lacp_settings); + if (s.lacp) { + size_t i = 0; -/* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument. - * Deletes from 'br' all the interfaces for which 'cb' returns false, and then - * deletes from 'br' any ports that no longer have any interfaces. */ -static void -iterate_and_prune_ifaces(struct bridge *br, - bool (*cb)(struct bridge *, struct iface *, - void *aux), - void *aux) -{ - size_t i, j; - - for (i = 0; i < br->n_ports; ) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; ) { - struct iface *iface = port->ifaces[j]; - if (cb(br, iface, aux)) { - j++; - } else { - iface_destroy(iface); - } + s.lacp_slaves = xmalloc(s.n_slaves * sizeof *s.lacp_slaves); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_configure_lacp(iface, &s.lacp_slaves[i++]); } + } else { + s.lacp_slaves = NULL; + } - if (port->n_ifaces) { - i++; - } else { - VLOG_ERR("%s port has no interfaces, dropping", port->name); - port_destroy(port); + /* Get bond settings. */ + if (s.n_slaves > 1) { + s.bond = &bond_settings; + s.bond_stable_ids = xmalloc(s.n_slaves * sizeof *s.bond_stable_ids); + port_configure_bond(port, &bond_settings, s.bond_stable_ids); + } else { + s.bond = NULL; + s.bond_stable_ids = NULL; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + netdev_set_miimon_interval(iface->netdev, 0); } } + + /* Register. */ + ofproto_bundle_register(port->bridge->ofproto, port, &s); + + /* Clean up. */ + free(s.slaves); + free(s.trunks); + free(s.lacp_slaves); + free(s.bond_stable_ids); } -void -bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) +/* Pick local port hardware address and datapath ID for 'br'. */ +static void +bridge_configure_datapath_id(struct bridge *br) { - struct ovsdb_idl_txn *txn; - struct shash old_br, new_br; - struct shash_node *node; - struct bridge *br, *next; - size_t i; + uint8_t ea[ETH_ADDR_LEN]; + uint64_t dpid; + struct iface *local_iface; + struct iface *hw_addr_iface; + char *dpid_string; + + bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); + local_iface = iface_from_ofp_port(br, OFPP_LOCAL); + if (local_iface) { + int error = netdev_set_etheraddr(local_iface->netdev, ea); + if (error) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge " + "Ethernet address: %s", + br->name, strerror(error)); + } + } + memcpy(br->ea, ea, ETH_ADDR_LEN); - COVERAGE_INC(bridge_reconfigure); + dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface); + ofproto_set_datapath_id(br->ofproto, dpid); - txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl); + dpid_string = xasprintf("%016"PRIx64, dpid); + ovsrec_bridge_set_datapath_id(br->cfg, dpid_string); + free(dpid_string); +} - /* Collect old and new bridges. */ - shash_init(&old_br); - shash_init(&new_br); - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - shash_add(&old_br, br->name, br); - } - for (i = 0; i < ovs_cfg->n_bridges; i++) { - const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i]; - if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) { - VLOG_WARN("more than one bridge named %s", br_cfg->name); - } +/* Set NetFlow configuration on 'br'. */ +static void +bridge_configure_netflow(struct bridge *br) +{ + struct ovsrec_netflow *cfg = br->cfg->netflow; + struct netflow_options opts; + + if (!cfg) { + ofproto_set_netflow(br->ofproto, NULL); + return; } - /* Get rid of deleted bridges and add new bridges. */ - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { - struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name); - if (br_cfg) { - br->cfg = br_cfg; - } else { - bridge_destroy(br); - } + memset(&opts, 0, sizeof opts); + + /* Get default NetFlow configuration from datapath. + * Apply overrides from 'cfg'. */ + ofproto_get_netflow_ids(br->ofproto, &opts.engine_type, &opts.engine_id); + if (cfg->engine_type) { + opts.engine_type = *cfg->engine_type; } - SHASH_FOR_EACH (node, &new_br) { - const char *br_name = node->name; - const struct ovsrec_bridge *br_cfg = node->data; - if (!shash_find_data(&old_br, br_name)) { - br = bridge_create(br_name); - if (br) { - br->cfg = br_cfg; - } - } + if (cfg->engine_id) { + opts.engine_id = *cfg->engine_id; } - shash_destroy(&old_br); - shash_destroy(&new_br); -#ifdef HAVE_OPENSSL - /* Configure SSL. */ - bridge_configure_ssl(ovs_cfg->ssl); -#endif - - /* Reconfigure all bridges. */ - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - bridge_reconfigure_one(ovs_cfg, br); + /* Configure active timeout interval. */ + opts.active_timeout = cfg->active_timeout; + if (!opts.active_timeout) { + opts.active_timeout = -1; + } else if (opts.active_timeout < 0) { + VLOG_WARN("bridge %s: active timeout interval set to negative " + "value, using default instead (%d seconds)", br->name, + NF_ACTIVE_TIMEOUT_DEFAULT); + opts.active_timeout = -1; } - /* Add and delete ports on all datapaths. - * - * The kernel will reject any attempt to add a given port to a datapath if - * that port already belongs to a different datapath, so we must do all - * port deletions before any port additions. */ - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; - struct shash want_ifaces; - - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); - bridge_get_all_ifaces(br, &want_ifaces); - for (i = 0; i < n_dpif_ports; i++) { - const struct odp_port *p = &dpif_ports[i]; - if (!shash_find(&want_ifaces, p->devname) - && strcmp(p->devname, br->name)) { - int retval = dpif_port_del(br->dpif, p->port); - if (retval) { - VLOG_ERR("failed to remove %s interface from %s: %s", - p->devname, dpif_name(br->dpif), - strerror(retval)); - } - } + /* Add engine ID to interface number to disambiguate bridgs? */ + opts.add_id_to_iface = cfg->add_id_to_interface; + if (opts.add_id_to_iface) { + if (opts.engine_id > 0x7f) { + VLOG_WARN("bridge %s: NetFlow port mangling may conflict with " + "another vswitch, choose an engine id less than 128", + br->name); } - shash_destroy(&want_ifaces); - free(dpif_ports); - } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; - struct shash cur_ifaces, want_ifaces; - struct shash_node *node; - - /* Get the set of interfaces currently in this datapath. */ - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); - shash_init(&cur_ifaces); - for (i = 0; i < n_dpif_ports; i++) { - const char *name = dpif_ports[i].devname; - if (!shash_find(&cur_ifaces, name)) { - shash_add(&cur_ifaces, name, NULL); - } + 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); } - free(dpif_ports); + } - /* Get the set of interfaces we want on this datapath. */ - bridge_get_all_ifaces(br, &want_ifaces); + /* Collectors. */ + sset_init(&opts.collectors); + sset_add_array(&opts.collectors, cfg->targets, cfg->n_targets); - SHASH_FOR_EACH (node, &want_ifaces) { - const char *if_name = node->name; - struct iface *iface = node->data; + /* Configure. */ + if (ofproto_set_netflow(br->ofproto, &opts)) { + VLOG_ERR("bridge %s: problem setting netflow collectors", br->name); + } + sset_destroy(&opts.collectors); +} - if (shash_find(&cur_ifaces, if_name)) { - /* Already exists, just reconfigure it. */ - if (iface) { - reconfigure_iface(iface->cfg); - } - } else { - /* Need to add to datapath. */ - bool internal; - int error; +/* Set sFlow configuration on 'br'. */ +static void +bridge_configure_sflow(struct bridge *br, int *sflow_bridge_number) +{ + const struct ovsrec_sflow *cfg = br->cfg->sflow; + struct ovsrec_controller **controllers; + struct ofproto_sflow_options oso; + size_t n_controllers; + size_t i; - /* Add to datapath. */ - internal = iface_is_internal(br, if_name); - error = dpif_port_add(br->dpif, if_name, - internal ? ODP_PORT_INTERNAL : 0, NULL); - if (error == EFBIG) { - VLOG_ERR("ran out of valid port numbers on %s", - dpif_name(br->dpif)); - break; - } else if (error) { - VLOG_ERR("failed to add %s interface to %s: %s", - if_name, dpif_name(br->dpif), strerror(error)); - } - } - } - shash_destroy(&cur_ifaces); - shash_destroy(&want_ifaces); + if (!cfg) { + ofproto_set_sflow(br->ofproto, NULL); + return; } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - uint8_t ea[8]; - uint64_t dpid; - struct iface *local_iface; - struct iface *hw_addr_iface; - char *dpid_string; - bridge_fetch_dp_ifaces(br); - iterate_and_prune_ifaces(br, init_iface_netdev, NULL); + memset(&oso, 0, sizeof oso); - iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL); + sset_init(&oso.targets); + sset_add_array(&oso.targets, cfg->targets, cfg->n_targets); - /* Pick local port hardware address, datapath ID. */ - bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); - local_iface = bridge_get_local_iface(br); - if (local_iface) { - int error = netdev_set_etheraddr(local_iface->netdev, ea); - if (error) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge " - "Ethernet address: %s", - br->name, strerror(error)); - } - } + oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE; + if (cfg->sampling) { + oso.sampling_rate = *cfg->sampling; + } - dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface); - ofproto_set_datapath_id(br->ofproto, dpid); + oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL; + if (cfg->polling) { + oso.polling_interval = *cfg->polling; + } - dpid_string = xasprintf("%012"PRIx64, dpid); - ovsrec_bridge_set_datapath_id(br->cfg, dpid_string); - free(dpid_string); + oso.header_len = SFL_DEFAULT_HEADER_SIZE; + if (cfg->header) { + oso.header_len = *cfg->header; + } - /* Set NetFlow configuration on this bridge. */ - if (br->cfg->netflow) { - struct ovsrec_netflow *nf_cfg = br->cfg->netflow; - struct netflow_options opts; + oso.sub_id = (*sflow_bridge_number)++; + oso.agent_device = cfg->agent; - memset(&opts, 0, sizeof opts); + oso.control_ip = NULL; + n_controllers = bridge_get_controllers(br, &controllers); + for (i = 0; i < n_controllers; i++) { + if (controllers[i]->local_ip) { + oso.control_ip = controllers[i]->local_ip; + break; + } + } + ofproto_set_sflow(br->ofproto, &oso); - dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id); - if (nf_cfg->engine_type) { - opts.engine_type = nf_cfg->engine_type; - } - if (nf_cfg->engine_id) { - opts.engine_id = nf_cfg->engine_id; - } + sset_destroy(&oso.targets); +} - opts.active_timeout = nf_cfg->active_timeout; - if (!opts.active_timeout) { - opts.active_timeout = -1; - } else if (opts.active_timeout < 0) { - VLOG_WARN("bridge %s: active timeout interval set to negative " - "value, using default instead (%d seconds)", br->name, - NF_ACTIVE_TIMEOUT_DEFAULT); - opts.active_timeout = -1; - } +static bool +bridge_has_bond_fake_iface(const struct bridge *br, const char *name) +{ + const struct port *port = port_lookup(br, name); + return port && port_is_bond_fake_iface(port); +} - opts.add_id_to_iface = nf_cfg->add_id_to_interface; - if (opts.add_id_to_iface) { - if (opts.engine_id > 0x7f) { - VLOG_WARN("bridge %s: netflow port mangling may conflict " - "with another vswitch, choose an engine id less " - "than 128", br->name); - } - if (br->n_ports > 508) { - VLOG_WARN("bridge %s: netflow port mangling will conflict " - "with another port when more than 508 ports are " - "used", br->name); - } - } +static bool +port_is_bond_fake_iface(const struct port *port) +{ + return port->cfg->bond_fake_iface && !list_is_short(&port->ifaces); +} - opts.collectors.n = nf_cfg->n_targets; - opts.collectors.names = nf_cfg->targets; - if (ofproto_set_netflow(br->ofproto, &opts)) { - VLOG_ERR("bridge %s: problem setting netflow collectors", - br->name); - } - } else { - ofproto_set_netflow(br->ofproto, NULL); - } +static void +add_del_bridges(const struct ovsrec_open_vswitch *cfg) +{ + struct bridge *br, *next; + struct shash new_br; + size_t i; - /* Update the controller and related settings. It would be more - * straightforward to call this from bridge_reconfigure_one(), but we - * can't do it there for two reasons. First, and most importantly, at - * that point we don't know the dp_ifidx of any interfaces that have - * been added to the bridge (because we haven't actually added them to - * the datapath). Second, at that point we haven't set the datapath ID - * yet; when a controller is configured, resetting the datapath ID will - * immediately disconnect from the controller, so it's better to set - * the datapath ID before the controller. */ - bridge_reconfigure_controller(ovs_cfg, br); + /* Collect new bridges' names and types. */ + shash_init(&new_br); + for (i = 0; i < cfg->n_bridges; i++) { + const struct ovsrec_bridge *br_cfg = cfg->bridges[i]; + if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) { + VLOG_WARN("bridge %s specified twice", br_cfg->name); + } } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - port_update_vlan_compat(port); - port_update_bonding(port); + /* Get rid of deleted bridges or those whose types have changed. + * Update 'cfg' of bridges that still exist. */ + HMAP_FOR_EACH_SAFE (br, next, node, &all_bridges) { + br->cfg = shash_find_data(&new_br, br->name); + if (!br->cfg || strcmp(br->type, ofproto_normalize_type( + br->cfg->datapath_type))) { + bridge_destroy(br); } } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - iterate_and_prune_ifaces(br, set_iface_properties, NULL); - } - ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg); + /* Add new bridges. */ + for (i = 0; i < cfg->n_bridges; i++) { + const struct ovsrec_bridge *br_cfg = cfg->bridges[i]; + struct bridge *br = bridge_lookup(br_cfg->name); + if (!br) { + bridge_create(br_cfg); + } + } - ovsdb_idl_txn_commit(txn); - ovsdb_idl_txn_destroy(txn); /* XXX */ + shash_destroy(&new_br); } -static const char * -bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key) +/* Delete each ofproto port on 'br' that doesn't have a corresponding "struct + * iface". + * + * The kernel will reject any attempt to add a given port to a datapath if that + * port already belongs to a different datapath, so we must do all port + * deletions before any port additions. */ +static void +bridge_del_ofproto_ports(struct bridge *br) { - size_t i; + struct ofproto_port_dump dump; + struct ofproto_port ofproto_port; + + OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, br->ofproto) { + const char *name = ofproto_port.name; + struct iface *iface; + const char *type; + int error; + + /* Ignore the local port. We can't change it anyhow. */ + if (!strcmp(name, br->name)) { + continue; + } - for (i = 0; i < br_cfg->n_other_config; i++) { - if (!strcmp(br_cfg->key_other_config[i], key)) { - return br_cfg->value_other_config[i]; + /* Get the type that 'ofproto_port' should have (ordinarily the + * type of its corresponding iface) or NULL if it should be + * deleted. */ + iface = iface_lookup(br, name); + type = (iface ? iface->type + : bridge_has_bond_fake_iface(br, name) ? "internal" + : NULL); + + /* If it's the wrong type then delete the ofproto port. */ + if (type + && !strcmp(ofproto_port.type, type) + && (!iface || !iface->netdev + || !strcmp(netdev_get_type(iface->netdev), type))) { + continue; + } + error = ofproto_port_del(br->ofproto, ofproto_port.ofp_port); + if (error) { + VLOG_WARN("bridge %s: failed to remove %s interface (%s)", + br->name, name, strerror(error)); + } + if (iface) { + netdev_close(iface->netdev); + iface->netdev = NULL; } } - return NULL; } static void -bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], - struct iface **hw_addr_iface) +iface_set_ofp_port(struct iface *iface, int ofp_port) { - const char *hwaddr; - size_t i, j; - int error; + struct bridge *br = iface->port->bridge; - *hw_addr_iface = NULL; + assert(iface->ofp_port < 0 && ofp_port >= 0); + iface->ofp_port = ofp_port; + hmap_insert(&br->ifaces, &iface->ofp_port_node, hash_int(ofp_port, 0)); + iface_set_ofport(iface->cfg, ofp_port); +} - /* Did the user request a particular MAC? */ - hwaddr = bridge_get_other_config(br->cfg, "hwaddr"); - if (hwaddr && eth_addr_from_string(hwaddr, ea)) { +static void +bridge_refresh_ofp_port(struct bridge *br) +{ + struct ofproto_port_dump dump; + struct ofproto_port ofproto_port; + struct port *port; + + /* Clear all the "ofp_port"es. */ + hmap_clear(&br->ifaces); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface->ofp_port = -1; + } + } + + /* Obtain the correct "ofp_port"s from ofproto. */ + OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, br->ofproto) { + struct iface *iface = iface_lookup(br, ofproto_port.name); + if (iface) { + if (iface->ofp_port >= 0) { + VLOG_WARN("bridge %s: interface %s reported twice", + br->name, ofproto_port.name); + } else if (iface_from_ofp_port(br, ofproto_port.ofp_port)) { + VLOG_WARN("bridge %s: interface %"PRIu16" reported twice", + br->name, ofproto_port.ofp_port); + } else { + iface_set_ofp_port(iface, ofproto_port.ofp_port); + } + } + } +} + +/* Add an ofproto port for any "struct iface" that doesn't have one. + * Delete any "struct iface" for which this fails. + * Delete any "struct port" that thereby ends up with no ifaces. */ +static void +bridge_add_ofproto_ports(struct bridge *br) +{ + struct port *port, *next_port; + + HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) { + struct iface *iface, *next_iface; + struct ofproto_port ofproto_port; + + LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) { + int error; + + /* Open the netdev. */ + if (!iface->netdev) { + error = netdev_open(iface->name, iface->type, &iface->netdev); + if (error) { + VLOG_WARN("could not open network device %s (%s)", + iface->name, strerror(error)); + } + } else { + error = 0; + } + + /* Configure the netdev. */ + if (iface->netdev) { + struct shash args; + + shash_init(&args); + shash_from_ovs_idl_map(iface->cfg->key_options, + iface->cfg->value_options, + iface->cfg->n_options, &args); + error = netdev_set_config(iface->netdev, &args); + shash_destroy(&args); + + if (error) { + VLOG_WARN("could not configure network device %s (%s)", + iface->name, strerror(error)); + netdev_close(iface->netdev); + iface->netdev = NULL; + } + } + + /* Add the port, if necessary. */ + if (iface->netdev && iface->ofp_port < 0) { + uint16_t ofp_port; + int error; + + error = ofproto_port_add(br->ofproto, iface->netdev, + &ofp_port); + if (!error) { + iface_set_ofp_port(iface, ofp_port); + } else { + netdev_close(iface->netdev); + iface->netdev = NULL; + } + } + + /* Populate stats columns in new Interface rows. */ + if (iface->netdev && !iface->cfg->mtu) { + iface_refresh_stats(iface); + iface_refresh_status(iface); + } + + /* Delete the iface if we failed. */ + if (iface->netdev && iface->ofp_port >= 0) { + VLOG_DBG("bridge %s: interface %s is on port %d", + br->name, iface->name, iface->ofp_port); + } else { + if (iface->netdev) { + VLOG_ERR("bridge %s: missing %s interface, dropping", + br->name, iface->name); + } else { + /* We already reported a related error, don't bother + * duplicating it. */ + } + iface_set_ofport(iface->cfg, -1); + iface_destroy(iface); + } + } + if (list_is_empty(&port->ifaces)) { + VLOG_WARN("%s port has no interfaces, dropping", port->name); + port_destroy(port); + continue; + } + + /* Add bond fake iface if necessary. */ + if (port_is_bond_fake_iface(port)) { + if (ofproto_port_query_by_name(br->ofproto, port->name, + &ofproto_port)) { + struct netdev *netdev; + int error; + + error = netdev_open(port->name, "internal", &netdev); + if (!error) { + ofproto_port_add(br->ofproto, netdev, NULL); + netdev_close(netdev); + } else { + VLOG_WARN("could not open network device %s (%s)", + port->name, strerror(error)); + } + } else { + /* Already exists, nothing to do. */ + ofproto_port_destroy(&ofproto_port); + } + } + } +} + +static const char * +get_ovsrec_key_value(const struct ovsdb_idl_row *row, + const struct ovsdb_idl_column *column, + const char *key) +{ + const struct ovsdb_datum *datum; + union ovsdb_atom atom; + unsigned int idx; + + datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING); + atom.string = (char *) key; + idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING); + return idx == UINT_MAX ? NULL : datum->values[idx].string; +} + +static const char * +bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key) +{ + return get_ovsrec_key_value(&br_cfg->header_, + &ovsrec_bridge_col_other_config, key); +} + +/* Set Flow eviction threshold */ +static void +bridge_configure_flow_eviction_threshold(struct bridge *br) +{ + const char *threshold_str; + unsigned threshold; + + threshold_str = bridge_get_other_config(br->cfg, "flow-eviction-threshold"); + if (threshold_str) { + threshold = strtoul(threshold_str, NULL, 10); + } else { + threshold = OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT; + } + ofproto_set_flow_eviction_threshold(br->ofproto, threshold); +} + +/* Set forward BPDU option. */ +static void +bridge_configure_forward_bpdu(struct bridge *br) +{ + const char *forward_bpdu_str; + bool forward_bpdu = false; + + forward_bpdu_str = bridge_get_other_config(br->cfg, "forward-bpdu"); + if (forward_bpdu_str && !strcmp(forward_bpdu_str, "true")) { + forward_bpdu = true; + } + ofproto_set_forward_bpdu(br->ofproto, forward_bpdu); +} + +static void +bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], + struct iface **hw_addr_iface) +{ + const char *hwaddr; + struct port *port; + int error; + + *hw_addr_iface = NULL; + + /* Did the user request a particular MAC? */ + hwaddr = bridge_get_other_config(br->cfg, "hwaddr"); + if (hwaddr && eth_addr_from_string(hwaddr, ea)) { if (eth_addr_is_multicast(ea)) { VLOG_ERR("bridge %s: cannot set MAC address to multicast " "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); @@ -798,24 +1023,23 @@ 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. */ - if (port->is_mirror_output_port) { + if (ofproto_is_mirror_output_bundle(br->ofproto, port)) { continue; } /* 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)) { @@ -829,26 +1053,21 @@ 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; } } /* The local port doesn't count (since we're trying to choose its * MAC address anyway). */ - if (iface->dp_ifidx == ODPP_LOCAL) { + if (iface->ofp_port == OFPP_LOCAL) { continue; } /* Grab MAC. */ error = netdev_get_etheraddr(iface->netdev, iface_ea); if (error) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s", - iface->name, strerror(error)); continue; } } @@ -858,7 +1077,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], !eth_addr_is_local(iface_ea) && !eth_addr_is_reserved(iface_ea) && !eth_addr_is_zero(iface_ea) && - memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) + eth_addr_compare_3way(iface_ea, ea) < 0) { memcpy(ea, iface_ea, ETH_ADDR_LEN); *hw_addr_iface = iface; @@ -966,2132 +1185,944 @@ dpid_from_hash(const void *data, size_t n) return eth_addr_to_uint64(hash); } -int -bridge_run(void) +static void +iface_refresh_status(struct iface *iface) { - struct bridge *br, *next; - int retval; + struct shash sh; - retval = 0; - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { - int error = bridge_run_one(br); - if (error) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, " - "forcing reconfiguration", br->name); - if (!retval) { - retval = error; - } - } + enum netdev_flags flags; + uint32_t current; + int64_t bps; + int mtu; + int64_t mtu_64; + int error; + + if (iface_is_synthetic(iface)) { + return; } - return retval; -} -void -bridge_wait(void) -{ - struct bridge *br; + shash_init(&sh); - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - ofproto_wait(br->ofproto); - if (br->controller) { - continue; - } + if (!netdev_get_status(iface->netdev, &sh)) { + size_t n; + char **keys, **values; + + shash_to_ovs_idl_map(&sh, &keys, &values, &n); + ovsrec_interface_set_status(iface->cfg, keys, values, n); - mac_learning_wait(br->ml); - bond_wait(br); + free(keys); + free(values); + } else { + ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0); } -} -/* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s - * configuration changes. */ -static void -bridge_flush(struct bridge *br) -{ - COVERAGE_INC(bridge_flush); - br->flush = true; - mac_learning_flush(br->ml); -} + shash_destroy_free_data(&sh); -/* 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; + error = netdev_get_flags(iface->netdev, &flags); + if (!error) { + ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down"); + } + else { + ovsrec_interface_set_admin_state(iface->cfg, NULL); + } - 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; - } - } + error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL); + if (!error) { + ovsrec_interface_set_duplex(iface->cfg, + netdev_features_is_full_duplex(current) + ? "full" : "half"); + /* warning: uint64_t -> int64_t conversion */ + bps = netdev_features_to_bps(current); + ovsrec_interface_set_link_speed(iface->cfg, &bps, 1); + } + else { + ovsrec_interface_set_duplex(iface->cfg, NULL); + ovsrec_interface_set_link_speed(iface->cfg, NULL, 0); } - return NULL; -} - -/* Bridge unixctl user interface functions. */ -static void -bridge_unixctl_fdb_show(struct unixctl_conn *conn, - const char *args, void *aux UNUSED) -{ - struct ds ds = DS_EMPTY_INITIALIZER; - const struct bridge *br; - const struct mac_entry *e; + ovsrec_interface_set_link_state(iface->cfg, + iface_get_carrier(iface) ? "up" : "down"); - br = bridge_lookup(args); - if (!br) { - unixctl_command_reply(conn, 501, "no such bridge"); - return; + error = netdev_get_mtu(iface->netdev, &mtu); + if (!error && mtu != INT_MAX) { + mtu_64 = mtu; + ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1); } - - ds_put_cstr(&ds, " port VLAN MAC Age\n"); - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { - if (e->port < 0 || e->port >= br->n_ports) { - continue; - } - ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", - br->ports[e->port]->ifaces[0]->dp_ifidx, - e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); + else { + ovsrec_interface_set_mtu(iface->cfg, NULL, 0); } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); - ds_destroy(&ds); } - -/* Bridge reconfiguration functions. */ -static struct bridge * -bridge_create(const char *name) +/* Writes 'iface''s CFM statistics to the database. Returns true if anything + * changed, false otherwise. */ +static bool +iface_refresh_cfm_stats(struct iface *iface) { - struct bridge *br; - int error; + const struct ovsrec_interface *cfg = iface->cfg; + bool changed = false; + int fault; - assert(!bridge_lookup(name)); - br = xzalloc(sizeof *br); + fault = ofproto_port_get_cfm_fault(iface->port->bridge->ofproto, + iface->ofp_port); - error = dpif_create_and_open(name, &br->dpif); - if (error) { - free(br); - return NULL; + if (fault < 0) { + return false; } - dpif_flow_flush(br->dpif); - error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto); - if (error) { - VLOG_ERR("failed to create switch %s: %s", name, strerror(error)); - dpif_delete(br->dpif); - dpif_close(br->dpif); - free(br); - return NULL; + if (cfg->n_cfm_fault != 1 || cfg->cfm_fault[0] != fault) { + bool fault_bool = fault; + ovsrec_interface_set_cfm_fault(cfg, &fault_bool, 1); + changed = true; } - br->name = xstrdup(name); - br->ml = mac_learning_create(); - br->sent_config_request = false; - eth_addr_random(br->default_ea); + return changed; +} + +static bool +iface_refresh_lacp_stats(struct iface *iface) +{ + struct ofproto *ofproto = iface->port->bridge->ofproto; + int old = iface->cfg->lacp_current ? *iface->cfg->lacp_current : -1; + int new = ofproto_port_is_lacp_current(ofproto, iface->ofp_port); - port_array_init(&br->ifaces); + if (old != new) { + bool current = new; + ovsrec_interface_set_lacp_current(iface->cfg, ¤t, new >= 0); + } + return old != new; +} - br->flush = false; - br->bond_next_rebalance = time_msec() + 10000; +static void +iface_refresh_stats(struct iface *iface) +{ +#define IFACE_STATS \ + IFACE_STAT(rx_packets, "rx_packets") \ + IFACE_STAT(tx_packets, "tx_packets") \ + IFACE_STAT(rx_bytes, "rx_bytes") \ + IFACE_STAT(tx_bytes, "tx_bytes") \ + IFACE_STAT(rx_dropped, "rx_dropped") \ + IFACE_STAT(tx_dropped, "tx_dropped") \ + IFACE_STAT(rx_errors, "rx_errors") \ + IFACE_STAT(tx_errors, "tx_errors") \ + IFACE_STAT(rx_frame_errors, "rx_frame_err") \ + IFACE_STAT(rx_over_errors, "rx_over_err") \ + IFACE_STAT(rx_crc_errors, "rx_crc_err") \ + IFACE_STAT(collisions, "collisions") + +#define IFACE_STAT(MEMBER, NAME) NAME, + static char *keys[] = { IFACE_STATS }; +#undef IFACE_STAT + int64_t values[ARRAY_SIZE(keys)]; + int i; + + struct netdev_stats stats; + + if (iface_is_synthetic(iface)) { + return; + } - list_push_back(&all_bridges, &br->node); + /* 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); - VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif)); + /* Copy statistics into values[] array. */ + i = 0; +#define IFACE_STAT(MEMBER, NAME) values[i++] = stats.MEMBER; + IFACE_STATS; +#undef IFACE_STAT + assert(i == ARRAY_SIZE(keys)); - return br; + ovsrec_interface_set_statistics(iface->cfg, keys, values, ARRAY_SIZE(keys)); +#undef IFACE_STATS } -static void -bridge_destroy(struct bridge *br) +static bool +enable_system_stats(const struct ovsrec_open_vswitch *cfg) { - if (br) { - int error; + const char *enable; - while (br->n_ports > 0) { - port_destroy(br->ports[br->n_ports - 1]); - } - list_remove(&br->node); - 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); - free(br->controller); - mac_learning_destroy(br->ml); - port_array_destroy(&br->ifaces); - free(br->ports); - free(br->name); - free(br); + /* Use other-config:enable-system-stats by preference. */ + enable = get_ovsrec_key_value(&cfg->header_, + &ovsrec_open_vswitch_col_other_config, + "enable-statistics"); + if (enable) { + return !strcmp(enable, "true"); } + + /* Disable by default. */ + return false; } -static struct bridge * -bridge_lookup(const char *name) +static void +refresh_system_stats(const struct ovsrec_open_vswitch *cfg) { - struct bridge *br; + struct ovsdb_datum datum; + struct shash stats; - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - if (!strcmp(br->name, name)) { - return br; - } + shash_init(&stats); + if (enable_system_stats(cfg)) { + get_system_stats(&stats); } - return NULL; -} -bool -bridge_exists(const char *name) -{ - return bridge_lookup(name) ? true : false; + ovsdb_datum_from_shash(&datum, &stats); + ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics, + &datum); } -uint64_t -bridge_get_datapathid(const char *name) +static inline const char * +nx_role_to_str(enum nx_role role) { - struct bridge *br = bridge_lookup(name); - return br ? ofproto_get_datapath_id(br->ofproto) : 0; + switch (role) { + case NX_ROLE_OTHER: + return "other"; + case NX_ROLE_MASTER: + return "master"; + case NX_ROLE_SLAVE: + return "slave"; + default: + return "*** INVALID ROLE ***"; + } } -/* Handle requests for a listing of all flows known by the OpenFlow - * stack, including those normally hidden. */ static void -bridge_unixctl_dump_flows(struct unixctl_conn *conn, - const char *args, void *aux UNUSED) +refresh_controller_status(void) { struct bridge *br; - struct ds results; - - br = bridge_lookup(args); - if (!br) { - unixctl_command_reply(conn, 501, "Unknown bridge"); - return; - } - - ds_init(&results); - ofproto_get_all_flows(br->ofproto, &results); - - unixctl_command_reply(conn, 200, ds_cstr(&results)); - ds_destroy(&results); -} + struct shash info; + const struct ovsrec_controller *cfg; -static int -bridge_run_one(struct bridge *br) -{ - int error; + shash_init(&info); - error = ofproto_run1(br->ofproto); - if (error) { - return error; + /* Accumulate status for controllers on all bridges. */ + HMAP_FOR_EACH (br, node, &all_bridges) { + ofproto_get_ofproto_controller_info(br->ofproto, &info); } - mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); - bond_run(br); + /* Update each controller in the database with current status. */ + OVSREC_CONTROLLER_FOR_EACH(cfg, idl) { + struct ofproto_controller_info *cinfo = + shash_find_data(&info, cfg->target); - error = ofproto_run2(br->ofproto, br->flush); - br->flush = false; + if (cinfo) { + ovsrec_controller_set_is_connected(cfg, cinfo->is_connected); + ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role)); + ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys, + (char **) cinfo->pairs.values, + cinfo->pairs.n); + } else { + ovsrec_controller_set_is_connected(cfg, false); + ovsrec_controller_set_role(cfg, NULL); + ovsrec_controller_set_status(cfg, NULL, NULL, 0); + } + } - return error; + ofproto_free_ofproto_controller_info(&info); } -static const struct ovsrec_controller * -bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg, - const struct bridge *br) +void +bridge_run(void) { - const struct ovsrec_controller *controller; + const struct ovsrec_open_vswitch *cfg; - controller = (br->cfg->controller ? br->cfg->controller - : ovs_cfg->controller ? ovs_cfg->controller - : NULL); - - if (controller && !strcmp(controller->target, "none")) { - return NULL; - } - - return controller; -} + bool datapath_destroyed; + bool database_changed; + struct bridge *br; -static bool -check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_) -{ - struct svec *ifaces = ifaces_; - if (!svec_contains(ifaces, iface->name)) { - svec_add(ifaces, iface->name); - svec_sort(ifaces); - return true; - } else { - VLOG_ERR("bridge %s: %s interface is on multiple ports, " - "removing from %s", - br->name, iface->name, iface->port->name); - return false; - } -} + /* (Re)configure if necessary. */ + database_changed = ovsdb_idl_run(idl); + if (ovsdb_idl_is_lock_contended(idl)) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); + struct bridge *br, *next_br; -static void -bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, - struct bridge *br) -{ - struct shash old_ports, new_ports; - struct svec ifaces; - struct svec listeners, old_listeners; - struct svec snoops, old_snoops; - struct shash_node *node; - uint64_t mgmt_id; - size_t i; + VLOG_ERR_RL(&rl, "another ovs-vswitchd process is running, " + "disabling this process until it goes away"); - /* 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]); + HMAP_FOR_EACH_SAFE (br, next_br, node, &all_bridges) { + bridge_destroy(br); + } + return; + } else if (!ovsdb_idl_has_lock(idl)) { + return; } + cfg = ovsrec_open_vswitch_first(idl); - /* Collect new ports. */ - shash_init(&new_ports); - for (i = 0; i < br->cfg->n_ports; i++) { - const char *name = br->cfg->ports[i]->name; - if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) { - VLOG_WARN("bridge %s: %s specified twice as bridge port", - br->name, name); + /* Let each bridge do the work that it needs to do. */ + datapath_destroyed = false; + HMAP_FOR_EACH (br, node, &all_bridges) { + int error = ofproto_run(br->ofproto); + if (error) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, " + "forcing reconfiguration", br->name); + datapath_destroyed = true; } } - /* If we have a controller, then we need a local port. Complain if the - * user didn't specify one. + /* Re-configure SSL. We do this on every trip through the main loop, + * instead of just when the database changes, because the contents of the + * key and certificate files can change without the database changing. * - * XXX perhaps we should synthesize a port ourselves in this case. */ - if (bridge_get_controller(ovs_cfg, br)) { - char local_name[IF_NAMESIZE]; - int error; + * We do this before bridge_reconfigure() because that function might + * initiate SSL connections and thus requires SSL to be configured. */ + if (cfg && cfg->ssl) { + const struct ovsrec_ssl *ssl = cfg->ssl; - 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); - } + stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate); + stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert); } - dpid_from_string(ovs_cfg->management_id, &mgmt_id); - ofproto_set_mgmt_id(br->ofproto, mgmt_id); + if (database_changed || datapath_destroyed) { + if (cfg) { + struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl); - /* Get rid of deleted ports and add new ports. */ - SHASH_FOR_EACH (node, &old_ports) { - if (!shash_find(&new_ports, node->name)) { - port_destroy(node->data); - } - } - SHASH_FOR_EACH (node, &new_ports) { - struct port *port = shash_find_data(&old_ports, node->name); - if (!port) { - port = port_create(br, node->name); + bridge_reconfigure(cfg); + + ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg); + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); /* XXX */ + } else { + /* We still need to reconfigure to avoid dangling pointers to + * now-destroyed ovsrec structures inside bridge data. */ + static const struct ovsrec_open_vswitch null_cfg; + + bridge_reconfigure(&null_cfg); } - port_reconfigure(port, node->data); } - shash_destroy(&old_ports); - shash_destroy(&new_ports); - /* Check and delete duplicate interfaces. */ - svec_init(&ifaces); - iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces); - svec_destroy(&ifaces); + /* Refresh system and interface stats if necessary. */ + if (time_msec() >= stats_timer) { + if (cfg) { + struct ovsdb_idl_txn *txn; - /* 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?) */ + txn = ovsdb_idl_txn_create(idl); + HMAP_FOR_EACH (br, node, &all_bridges) { + struct port *port; -#if 0 - /* Configure OpenFlow management listeners. */ - svec_init(&listeners); - cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name); - if (!listeners.n) { - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) { - svec_clear(&listeners); - } - svec_sort_unique(&listeners); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - svec_sort_unique(&old_listeners); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_refresh_stats(iface); + iface_refresh_status(iface); + } + } + } + refresh_system_stats(cfg); + refresh_controller_status(); + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); /* XXX */ + } - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); + stats_timer = time_msec() + STATS_INTERVAL; } - svec_destroy(&listeners); - svec_destroy(&old_listeners); - /* Configure OpenFlow controller connection snooping. */ - svec_init(&snoops); - cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name); - if (!snoops.n) { - svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); - } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) { - svec_clear(&snoops); - } - svec_sort_unique(&snoops); - - svec_init(&old_snoops); - ofproto_get_snoops(br->ofproto, &old_snoops); - svec_sort_unique(&old_snoops); - - if (!svec_equal(&snoops, &old_snoops)) { - ofproto_set_snoops(br->ofproto, &snoops); - } - svec_destroy(&snoops); - svec_destroy(&old_snoops); -#else - /* Default listener. */ - svec_init(&listeners); - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); - } - svec_destroy(&listeners); - svec_destroy(&old_listeners); - - /* Default snoop. */ - svec_init(&snoops); - svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); - svec_init(&old_snoops); - ofproto_get_snoops(br->ofproto, &old_snoops); - if (!svec_equal(&snoops, &old_snoops)) { - ofproto_set_snoops(br->ofproto, &snoops); - } - svec_destroy(&snoops); - svec_destroy(&old_snoops); -#endif - -#if 0 - mirror_reconfigure(br); -#endif -} - -static void -bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg, - struct bridge *br) -{ - char *pfx = xasprintf("bridge.%s.controller", br->name); - const struct ovsrec_controller *c; + if (time_msec() >= db_limiter) { + struct ovsdb_idl_txn *txn; + bool changed = false; - c = bridge_get_controller(ovs_cfg, br); - if ((br->controller != NULL) != (c != NULL)) { - ofproto_flush_flows(br->ofproto); - } - free(br->controller); - br->controller = c ? xstrdup(c->target) : NULL; - - if (c) { - int max_backoff, probe; - int rate_limit, burst_limit; - - if (!strcmp(c->target, "discover")) { - ofproto_set_discovery(br->ofproto, true, - c->discover_accept_regex, - c->discover_update_resolv_conf); - } else { - struct iface *local_iface; - struct in_addr ip; - bool in_band; - - in_band = (!c->connection_mode - || !strcmp(c->connection_mode, "out-of-band")); - ofproto_set_discovery(br->ofproto, false, NULL, NULL); - ofproto_set_in_band(br->ofproto, in_band); - - local_iface = bridge_get_local_iface(br); - if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) { - struct netdev *netdev = local_iface->netdev; - struct in_addr ip, mask, gateway; - - if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) { - mask.s_addr = 0; - } - if (!c->local_gateway - || !inet_aton(c->local_gateway, &gateway)) { - gateway.s_addr = 0; - } + txn = ovsdb_idl_txn_create(idl); + HMAP_FOR_EACH (br, node, &all_bridges) { + struct port *port; - netdev_turn_flags_on(netdev, NETDEV_UP, true); - if (!mask.s_addr) { - mask.s_addr = guess_netmask(ip.s_addr); - } - if (!netdev_set_in4(netdev, ip, mask)) { - VLOG_INFO("bridge %s: configured IP address "IP_FMT", " - "netmask "IP_FMT, - br->name, IP_ARGS(&ip.s_addr), - IP_ARGS(&mask.s_addr)); - } + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; - if (gateway.s_addr) { - if (!netdev_add_router(netdev, gateway)) { - VLOG_INFO("bridge %s: configured gateway "IP_FMT, - br->name, IP_ARGS(&gateway.s_addr)); - } + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + changed = iface_refresh_cfm_stats(iface) || changed; + changed = iface_refresh_lacp_stats(iface) || changed; } } } - ofproto_set_failure(br->ofproto, - (!c->fail_mode - || !strcmp(c->fail_mode, "standalone") - || !strcmp(c->fail_mode, "open"))); - - probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; - ofproto_set_probe_interval(br->ofproto, probe); - - max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; - ofproto_set_max_backoff(br->ofproto, max_backoff); - - rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; - burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0; - ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit); - } else { - union ofp_action action; - flow_t flow; - - /* Set up a flow that matches every packet and directs them to - * OFPP_NORMAL (which goes to us). */ - memset(&action, 0, sizeof action); - action.type = htons(OFPAT_OUTPUT); - action.output.len = htons(sizeof action); - action.output.port = htons(OFPP_NORMAL); - memset(&flow, 0, sizeof flow); - ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0, - &action, 1, 0); - - ofproto_set_in_band(br->ofproto, false); - ofproto_set_max_backoff(br->ofproto, 1); - ofproto_set_probe_interval(br->ofproto, 5); - ofproto_set_failure(br->ofproto, false); - } - free(pfx); - - ofproto_set_controller(br->ofproto, br->controller); -} - -static void -bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) -{ - size_t i, j; - - 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]; - shash_add_once(ifaces, iface->name, iface); - } - if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { - shash_add_once(ifaces, port->name, NULL); + if (changed) { + db_limiter = time_msec() + DB_LIMIT_INTERVAL; } + + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); } } -/* For robustness, in case the administrator moves around datapath ports behind - * our back, we re-check all the datapath port numbers here. - * - * This function will set the 'dp_ifidx' members of interfaces that have - * disappeared to -1, so only call this function from a context where those - * 'struct iface's will be removed from the bridge. Otherwise, the -1 - * 'dp_ifidx'es will cause trouble later when we try to send them to the - * datapath, which doesn't support UINT16_MAX+1 ports. */ -static void -bridge_fetch_dp_ifaces(struct bridge *br) +void +bridge_wait(void) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; - size_t i, j; + ovsdb_idl_wait(idl); + if (!hmap_is_empty(&all_bridges)) { + struct bridge *br; - /* 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]; - iface->dp_ifidx = -1; + HMAP_FOR_EACH (br, node, &all_bridges) { + ofproto_wait(br->ofproto); } - } - port_array_clear(&br->ifaces); - - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); - for (i = 0; i < n_dpif_ports; i++) { - struct odp_port *p = &dpif_ports[i]; - struct iface *iface = iface_lookup(br, p->devname); - if (iface) { - if (iface->dp_ifidx >= 0) { - VLOG_WARN("%s reported interface %s twice", - dpif_name(br->dpif), p->devname); - } else if (iface_from_dp_ifidx(br, p->port)) { - VLOG_WARN("%s reported interface %"PRIu16" twice", - dpif_name(br->dpif), p->port); - } else { - port_array_set(&br->ifaces, p->port, iface); - iface->dp_ifidx = p->port; - } + poll_timer_wait_until(stats_timer); - if (iface->cfg) { - int64_t ofport = (iface->dp_ifidx >= 0 - ? odp_port_to_ofp_port(iface->dp_ifidx) - : -1); - ovsrec_interface_set_ofport(iface->cfg, &ofport, 1); - } + if (db_limiter > time_msec()) { + poll_timer_wait_until(db_limiter); } } - free(dpif_ports); } -/* Bridge packet processing functions. */ - -static int -bond_hash(const uint8_t mac[ETH_ADDR_LEN]) -{ - return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK; -} - -static struct bond_entry * -lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN]) -{ - return &port->bond_hash[bond_hash(mac)]; -} - -static int -bond_choose_iface(const struct port *port) -{ - 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) { - best_down_slave = i; - next_delay_expiration = iface->delay_expires; - } - } - - 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; -} +/* QoS unixctl user interface functions. */ -static bool -choose_output_iface(const struct port *port, const uint8_t *dl_src, - uint16_t *dp_ifidx, tag_type *tags) -{ +struct qos_unixctl_show_cbdata { + struct ds *ds; struct iface *iface; - - assert(port->n_ifaces); - if (port->n_ifaces == 1) { - iface = port->ifaces[0]; - } else { - struct bond_entry *e = lookup_bond_entry(port, dl_src); - 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, bool carrier) -{ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); - struct port *port = iface->port; - - if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { - /* Nothing to do. */ - return; - } - VLOG_INFO_RL(&rl, "interface %s: carrier %s", - iface->name, carrier ? "detected" : "dropped"); - if (carrier == iface->enabled) { - iface->delay_expires = LLONG_MAX; - VLOG_INFO_RL(&rl, "interface %s: will not be %s", - iface->name, carrier ? "disabled" : "enabled"); - } else if (carrier && port->active_iface < 0) { - bond_enable_slave(iface, true); - if (port->updelay) { - VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " - "other interface is up", iface->name, port->updelay); - } - } else { - int delay = carrier ? port->updelay : port->downdelay; - iface->delay_expires = time_msec() + delay; - if (delay) { - VLOG_INFO_RL(&rl, - "interface %s: will be %s if it stays %s for %d ms", - iface->name, - carrier ? "enabled" : "disabled", - carrier ? "up" : "down", - delay); - } - } -} +}; static void -bond_choose_active_iface(struct port *port) -{ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); +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; - 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); + ds_put_cstr(ds, "\n"); + if (queue_id) { + ds_put_format(ds, "Queue %u:\n", queue_id); } else { - VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface", - port->name); + ds_put_cstr(ds, "Default:\n"); } -} - -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; + SHASH_FOR_EACH (node, details) { + ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data); } - 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); + 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); } - iface->tag = tag_create_random(); - } - - moving_active_iface = false; - 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) { - 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 (stats.tx_bytes != UINT64_MAX) { + ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes); } - 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; - } - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->delay_expires != LLONG_MAX) { - poll_timer_wait(iface->delay_expires - time_msec()); - } + 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 bool -set_dst(struct dst *p, const flow_t *flow, - const struct port *in_port, const struct port *out_port, - tag_type *tags) -{ - p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE - : in_port->vlan >= 0 ? in_port->vlan - : ntohs(flow->dl_vlan)); - return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags); -} - static void -swap_dst(struct dst *p, struct dst *q) +qos_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) { - 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 *dsts, size_t n_dsts, int vlan) -{ - struct dst *first = dsts; - struct dst *last = dsts + n_dsts; - - 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); -} + 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; -static bool -dst_is_duplicate(const struct dst *dsts, size_t n_dsts, - const struct dst *test) -{ - size_t i; - for (i = 0; i < n_dsts; i++) { - if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) { - return true; - } + iface = iface_find(args); + if (!iface) { + unixctl_command_reply(conn, 501, "no such interface"); + return; } - return false; -} - -static bool -port_trunks_vlan(const struct port *port, uint16_t vlan) -{ - return port->vlan < 0 && bitmap_is_set(port->trunks, vlan); -} -static bool -port_includes_vlan(const struct port *port, uint16_t vlan) -{ - return vlan == port->vlan || port_trunks_vlan(port, vlan); -} + netdev_get_qos(iface->netdev, &type, &sh); -static size_t -compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, - const struct port *in_port, const struct port *out_port, - struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface) -{ - mirror_mask_t mirrors = in_port->src_mirrors; - struct dst *dst = dsts; - size_t i; + if (*type != '\0') { + ds_put_format(&ds, "QoS: %s %s\n", iface->name, type); - if (out_port == FLOOD_PORT) { - /* XXX use ODP_FLOOD if no vlans or bonding. */ - /* XXX even better, define each VLAN as a datapath port group */ - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (port != in_port && port_includes_vlan(port, vlan) - && !port->is_mirror_output_port - && set_dst(dst, flow, in_port, port, tags)) { - mirrors |= port->dst_mirrors; - dst++; - } + SHASH_FOR_EACH (node, &sh) { + ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data); } - *nf_output_iface = NF_OUT_FLOOD; - } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) { - *nf_output_iface = dst->dp_ifidx; - mirrors |= out_port->dst_mirrors; - dst++; - } - while (mirrors) { - struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1]; - if (!m->n_vlans || vlan_is_mirrored(m, vlan)) { - if (m->out_port) { - if (set_dst(dst, flow, in_port, m->out_port, tags) - && !dst_is_duplicate(dsts, dst - dsts, dst)) { - dst++; - } - } else { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (port_includes_vlan(port, m->out_vlan) - && set_dst(dst, flow, in_port, port, tags)) - { - int flow_vlan; - - if (port->vlan < 0) { - dst->vlan = m->out_vlan; - } - if (dst_is_duplicate(dsts, dst - dsts, dst)) { - continue; - } - - /* Use the vlan tag on the original flow instead of - * the one passed in the vlan parameter. This ensures - * that we compare the vlan from before any implicit - * tagging tags place. This is necessary because - * dst->vlan is the final vlan, after removing implicit - * tags. */ - flow_vlan = ntohs(flow->dl_vlan); - if (flow_vlan == 0) { - flow_vlan = OFP_VLAN_NONE; - } - if (port == in_port && dst->vlan == flow_vlan) { - /* Don't send out input port on same VLAN. */ - continue; - } - dst++; - } - } - } - } - mirrors &= mirrors - 1; - } - - partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan)); - return dst - dsts; -} - -static void UNUSED -print_dsts(const struct dst *dsts, size_t n) -{ - for (; n--; dsts++) { - printf(">p%"PRIu16, dsts->dp_ifidx); - if (dsts->vlan != OFP_VLAN_NONE) { - printf("v%"PRIu16, dsts->vlan); - } - } -} + data.ds = &ds; + data.iface = iface; + error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data); -static void -compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, - const struct port *in_port, const struct port *out_port, - tag_type *tags, struct odp_actions *actions, - uint16_t *nf_output_iface) -{ - struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)]; - size_t n_dsts; - const struct dst *p; - uint16_t cur_vlan; - - n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags, - nf_output_iface); - - cur_vlan = ntohs(flow->dl_vlan); - for (p = dsts; p < &dsts[n_dsts]; p++) { - union odp_action *a; - if (p->vlan != cur_vlan) { - if (p->vlan == OFP_VLAN_NONE) { - odp_actions_add(actions, ODPAT_STRIP_VLAN); - } else { - a = odp_actions_add(actions, ODPAT_SET_VLAN_VID); - a->vlan_vid.vlan_vid = htons(p->vlan); - } - cur_vlan = p->vlan; - } - a = odp_actions_add(actions, ODPAT_OUTPUT); - a->output.port = p->dp_ifidx; - } -} - -/* Returns the effective vlan of a packet, taking into account both the - * 802.1Q header and implicitly tagged ports. A value of 0 indicates that - * the packet is untagged and -1 indicates it has an invalid header and - * should be dropped. */ -static int flow_get_vlan(struct bridge *br, const flow_t *flow, - struct port *in_port, bool have_packet) -{ - /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet - * belongs to VLAN 0, so we should treat both cases identically. (In the - * former case, the packet has an 802.1Q header that specifies VLAN 0, - * presumably to allow a priority to be specified. In the latter case, the - * packet does not have any 802.1Q header.) */ - int vlan = ntohs(flow->dl_vlan); - if (vlan == OFP_VLAN_NONE) { - vlan = 0; - } - if (in_port->vlan >= 0) { - if (vlan) { - /* XXX support double tagging? */ - if (have_packet) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " - "packet received on port %s configured with " - "implicit VLAN %"PRIu16, - br->name, ntohs(flow->dl_vlan), - in_port->name, in_port->vlan); - } - return -1; + if (error) { + ds_put_format(&ds, "failed to dump queues: %s", strerror(error)); } - vlan = in_port->vlan; + unixctl_command_reply(conn, 200, ds_cstr(&ds)); } else { - if (!port_includes_vlan(in_port, vlan)) { - if (have_packet) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " - "packet received on port %s not configured for " - "trunking VLAN %d", - br->name, vlan, in_port->name, vlan); - } - return -1; - } + ds_put_format(&ds, "QoS not configured on %s\n", iface->name); + unixctl_command_reply(conn, 501, ds_cstr(&ds)); } - return vlan; + shash_destroy_free_data(&sh); + ds_destroy(&ds); } - + +/* Bridge reconfiguration functions. */ static void -update_learning_table(struct bridge *br, const flow_t *flow, int vlan, - struct port *in_port) +bridge_create(const struct ovsrec_bridge *br_cfg) { - tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src, - vlan, in_port->port_idx); - if (rev_tag) { - /* The log messages here could actually be useful in debugging, - * so keep the rate limit relatively high. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, - 300); - VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " - "on port %s in VLAN %d", - br->name, ETH_ADDR_ARGS(flow->dl_src), - in_port->name, vlan); - ofproto_revalidate(br->ofproto, rev_tag); - } -} - -static bool -is_bcast_arp_reply(const flow_t *flow) -{ - return (flow->dl_type == htons(ETH_TYPE_ARP) - && flow->nw_proto == ARP_OP_REPLY - && eth_addr_is_broadcast(flow->dl_dst)); -} - -/* If the composed actions may be applied to any packet in the given 'flow', - * returns true. Otherwise, the actions should only be applied to 'packet', or - * not at all, if 'packet' was NULL. */ -static bool -process_flow(struct bridge *br, const flow_t *flow, - const struct ofpbuf *packet, struct odp_actions *actions, - tag_type *tags, uint16_t *nf_output_iface) -{ - struct iface *in_iface; - struct port *in_port; - struct port *out_port = NULL; /* By default, drop the packet/flow. */ - int vlan; - int out_port_idx; - - /* Find the interface and port structure for the received packet. */ - in_iface = iface_from_dp_ifidx(br, flow->in_port); - if (!in_iface) { - /* No interface? Something fishy... */ - if (packet != NULL) { - /* Odd. A few possible reasons here: - * - * - We deleted an interface but there are still a few packets - * queued up from it. - * - * - Someone externally added an interface (e.g. with "ovs-dpctl - * add-if") that we don't know about. - * - * - Packet arrived on the local port but the local port is not - * one of our bridge ports. - */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - - VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " - "interface %"PRIu16, br->name, flow->in_port); - } - - /* Return without adding any actions, to drop packets on this flow. */ - return true; - } - in_port = in_iface->port; - vlan = flow_get_vlan(br, flow, in_port, !!packet); - if (vlan < 0) { - goto done; - } - - /* Drop frames for reserved multicast addresses. */ - if (eth_addr_is_reserved(flow->dl_dst)) { - goto done; - } - - /* Drop frames on ports reserved for mirroring. */ - if (in_port->is_mirror_output_port) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, " - "which is reserved exclusively for mirroring", - br->name, in_port->name); - goto done; - } - - /* Packets received on bonds need special attention to avoid duplicates. */ - if (in_port->n_ifaces > 1) { - int src_idx; - - if (eth_addr_is_multicast(flow->dl_dst)) { - *tags |= in_port->active_iface_tag; - if (in_port->active_iface != in_iface->port_ifidx) { - /* Drop all multicast packets on inactive slaves. */ - goto done; - } - } - - /* Drop all packets for which we have learned a different input - * port, because we probably sent the packet on one slave and got - * it back on the other. Broadcast ARP replies are an exception - * to this rule: the host has moved to another switch. */ - src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); - if (src_idx != -1 && src_idx != in_port->port_idx && - !is_bcast_arp_reply(flow)) { - goto done; - } - } + struct bridge *br; - /* MAC learning. */ - out_port = FLOOD_PORT; - /* Learn source MAC (but don't try to learn from revalidation). */ - if (packet) { - update_learning_table(br, flow, vlan, in_port); - } + assert(!bridge_lookup(br_cfg->name)); + br = xzalloc(sizeof *br); - /* Determine output port. */ - out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, - tags); - if (out_port_idx >= 0 && out_port_idx < br->n_ports) { - out_port = br->ports[out_port_idx]; - } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { - /* If we are revalidating but don't have a learning entry then - * eject the flow. Installing a flow that floods packets opens - * up a window of time where we could learn from a packet reflected - * on a bond and blackhole packets before the learning table is - * updated to reflect the correct port. */ - return false; - } + br->name = xstrdup(br_cfg->name); + br->type = xstrdup(ofproto_normalize_type(br_cfg->datapath_type)); + br->cfg = br_cfg; - /* Don't send packets out their input ports. */ - if (in_port == out_port) { - out_port = NULL; - } + /* Derive the default Ethernet address from the bridge's UUID. This should + * be unique and it will be stable between ovs-vswitchd runs. */ + memcpy(br->default_ea, &br_cfg->header_.uuid, ETH_ADDR_LEN); + eth_addr_mark_random(br->default_ea); -done: - compose_actions(br, flow, vlan, in_port, out_port, tags, actions, - nf_output_iface); + hmap_init(&br->ports); + hmap_init(&br->ifaces); + hmap_init(&br->iface_by_name); + hmap_init(&br->mirrors); - return true; + hmap_insert(&all_bridges, &br->node, hash_string(br->name, 0)); } -/* Careful: 'opp' is in host byte order and opp->port_no is an OFP port - * number. */ static void -bridge_port_changed_ofhook_cb(enum ofp_port_reason reason, - const struct ofp_phy_port *opp, - void *br_) +bridge_destroy(struct bridge *br) { - struct bridge *br = br_; - struct iface *iface; - struct port *port; - - iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no)); - if (!iface) { - return; - } - port = iface->port; + if (br) { + struct mirror *mirror, *next_mirror; + struct port *port, *next_port; - if (reason == OFPPR_DELETE) { - VLOG_WARN("bridge %s: interface %s deleted unexpectedly", - br->name, iface->name); - iface_destroy(iface); - if (!port->n_ifaces) { - VLOG_WARN("bridge %s: port %s has no interfaces, dropping", - br->name, port->name); + HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) { port_destroy(port); } - - bridge_flush(br); - } else { - if (port->n_ifaces > 1) { - bool up = !(opp->state & OFPPS_LINK_DOWN); - bond_link_status_update(iface, up); - port_update_bond_compat(port); + HMAP_FOR_EACH_SAFE (mirror, next_mirror, hmap_node, &br->mirrors) { + mirror_destroy(mirror); } + hmap_remove(&all_bridges, &br->node); + ofproto_destroy(br->ofproto); + hmap_destroy(&br->ifaces); + hmap_destroy(&br->ports); + hmap_destroy(&br->iface_by_name); + hmap_destroy(&br->mirrors); + free(br->name); + free(br->type); + free(br); } } -static bool -bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, - struct odp_actions *actions, tag_type *tags, - uint16_t *nf_output_iface, void *br_) -{ - struct bridge *br = br_; - - COVERAGE_INC(bridge_process_flow); - return process_flow(br, flow, packet, actions, tags, nf_output_iface); -} - -static void -bridge_account_flow_ofhook_cb(const flow_t *flow, - const union odp_action *actions, - size_t n_actions, unsigned long long int n_bytes, - void *br_) +static struct bridge * +bridge_lookup(const char *name) { - struct bridge *br = br_; - struct port *in_port; - const union odp_action *a; - - /* Feed information from the active flows back into the learning table - * to ensure that table is always in sync with what is actually flowing - * through the datapath. */ - in_port = port_from_dp_ifidx(br, flow->in_port); - if (in_port) { - int vlan = flow_get_vlan(br, flow, in_port, false); - if (vlan >= 0) { - update_learning_table(br, flow, vlan, in_port); - } - } - - if (!br->has_bonded_ports) { - return; - } + struct bridge *br; - for (a = actions; a < &actions[n_actions]; a++) { - if (a->type == ODPAT_OUTPUT) { - struct port *out_port = port_from_dp_ifidx(br, a->output.port); - if (out_port && out_port->n_ifaces >= 2) { - struct bond_entry *e = lookup_bond_entry(out_port, - flow->dl_src); - e->tx_bytes += n_bytes; - } + HMAP_FOR_EACH_WITH_HASH (br, node, hash_string(name, 0), &all_bridges) { + if (!strcmp(br->name, name)) { + return br; } } + return NULL; } +/* Handle requests for a listing of all flows known by the OpenFlow + * stack, including those normally hidden. */ static void -bridge_account_checkpoint_ofhook_cb(void *br_) +bridge_unixctl_dump_flows(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) { - struct bridge *br = br_; - size_t i; - - if (!br->has_bonded_ports) { - return; - } + struct bridge *br; + struct ds results; - /* The current ofproto implementation calls this callback at least once a - * second, so this timer implementation is sufficient. */ - if (time_msec() < br->bond_next_rebalance) { + br = bridge_lookup(args); + if (!br) { + unixctl_command_reply(conn, 501, "Unknown bridge"); return; } - br->bond_next_rebalance = time_msec() + 10000; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (port->n_ifaces > 1) { - bond_rebalance_port(port); - } - } -} - -static struct ofhooks bridge_ofhooks = { - bridge_port_changed_ofhook_cb, - bridge_normal_ofhook_cb, - bridge_account_flow_ofhook_cb, - bridge_account_checkpoint_ofhook_cb, -}; - -/* 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; -}; - -/* 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; - } -} + ds_init(&results); + ofproto_get_all_flows(br->ofproto, &results); -static void -swap_bals(struct slave_balance *a, struct slave_balance *b) -{ - struct slave_balance tmp = *a; - *a = *b; - *b = tmp; + unixctl_command_reply(conn, 200, ds_cstr(&results)); + ds_destroy(&results); } -/* 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. */ +/* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller + * connections and reconnect. If BRIDGE is not specified, then all bridges + * drop their controller connections and reconnect. */ static void -resort_bals(struct slave_balance *p, - struct slave_balance bals[], size_t n_bals) +bridge_unixctl_reconnect(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) { - if (n_bals > 1) { - for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) { - swap_bals(p, p - 1); + struct bridge *br; + if (args[0] != '\0') { + br = bridge_lookup(args); + if (!br) { + unixctl_command_reply(conn, 501, "Unknown bridge"); + return; } - for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) { - swap_bals(p, p + 1); + ofproto_reconnect_controllers(br->ofproto); + } else { + HMAP_FOR_EACH (br, node, &all_bridges) { + ofproto_reconnect_controllers(br->ofproto); } } + unixctl_command_reply(conn, 200, NULL); } -static void -log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port) +static size_t +bridge_get_controllers(const struct bridge *br, + struct ovsrec_controller ***controllersp) { - 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; + struct ovsrec_controller **controllers; + size_t n_controllers; - if (b > bals) { - ds_put_char(&ds, ','); - } - ds_put_format(&ds, " %s %"PRIu64"kB", - b->iface->name, b->tx_bytes / 1024); + controllers = br->cfg->controller; + n_controllers = br->cfg->n_controller; - 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, ")"); - } - } - VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds)); - ds_destroy(&ds); + if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) { + controllers = NULL; + n_controllers = 0; } -} -/* Shifts 'hash' from 'from' to 'to' within 'port'. */ -static void -bond_shift_load(struct slave_balance *from, struct slave_balance *to, - int hash_idx) -{ - struct bond_entry *hash = from->hashes[hash_idx]; - struct port *port = from->iface->port; - uint64_t delta = hash->tx_bytes; - - 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); - - /* 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++; - } else { - memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1, - (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes); + if (controllersp) { + *controllersp = controllers; } - 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(); + return n_controllers; } +/* Adds and deletes "struct port"s and "struct iface"s under 'br' to match + * those configured in 'br->cfg'. */ static void -bond_rebalance_port(struct port *port) +bridge_add_del_ports(struct bridge *br) { - struct slave_balance bals[DP_MAX_PORTS]; - size_t n_bals; - struct bond_entry *hashes[BOND_MASK + 1]; - struct slave_balance *b, *from, *to; - struct bond_entry *e; + struct port *port, *next; + struct shash_node *node; + struct shash new_ports; size_t i; - /* 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; - 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) { - return; + /* Collect new ports. */ + shash_init(&new_ports); + for (i = 0; i < br->cfg->n_ports; i++) { + const char *name = br->cfg->ports[i]->name; + if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) { + VLOG_WARN("bridge %s: %s specified twice as bridge port", + br->name, name); } } + if (bridge_get_controllers(br, NULL) + && !shash_find(&new_ports, br->name)) { + VLOG_WARN("bridge %s: no port named %s, synthesizing one", + br->name, br->name); - /* 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. */ - size_t i; - bool order_swapped; - - for (i = 0; i < from->n_hashes; i++) { - double old_ratio, new_ratio; - uint64_t delta = from->hashes[i]->tx_bytes; - - if (delta == 0 || from->tx_bytes - delta == 0) { - /* Pointless move. */ - continue; - } - - order_swapped = from->tx_bytes - delta < to->tx_bytes + delta; + br->synth_local_port.interfaces = &br->synth_local_ifacep; + br->synth_local_port.n_interfaces = 1; + br->synth_local_port.name = br->name; - 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; - } + br->synth_local_iface.name = br->name; + br->synth_local_iface.type = "internal"; - if (new_ratio < 1) { - new_ratio = 1 / new_ratio; - } + br->synth_local_ifacep = &br->synth_local_iface; - 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); - } + shash_add(&new_ports, br->name, &br->synth_local_port); + } - /* 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++; - } + /* Get rid of deleted ports. + * Get rid of deleted interfaces on ports that still exist. + * Update 'cfg' of ports that still exist. */ + HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) { + port->cfg = shash_find_data(&new_ports, port->name); + if (!port->cfg) { + port_destroy(port); + } else { + port_del_ifaces(port); } } - /* 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; + /* Create new ports. + * Add new interfaces to existing ports. */ + SHASH_FOR_EACH (node, &new_ports) { + struct port *port = port_lookup(br, node->name); + if (!port) { + struct ovsrec_port *cfg = node->data; + port = port_create(br, cfg); + } + port_add_ifaces(port); + 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(&new_ports); } +/* Initializes 'oc' appropriately as a management service controller for + * 'br'. + * + * The caller must free oc->target when it is no longer needed. */ static void -bond_send_learning_packets(struct port *port) +bridge_ofproto_controller_for_mgmt(const struct bridge *br, + struct ofproto_controller *oc) { - 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) { - return; - } - - ofpbuf_init(&packet, 128); - error = n_packets = n_errors = 0; - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { - union ofp_action actions[2], *a; - uint16_t dp_ifidx; - tag_type tags = 0; - flow_t flow; - int retval; - - if (e->port == port->port_idx - || !choose_output_iface(port, e->mac, &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++; - compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, - e->mac); - flow_extract(&packet, ODPP_NONE, &flow); - retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, - &packet); - 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); - } + oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name); + oc->max_backoff = 0; + oc->probe_interval = 60; + oc->band = OFPROTO_OUT_OF_BAND; + oc->rate_limit = 0; + oc->burst_limit = 0; } - -/* Bonding unixctl user interface functions. */ +/* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */ static void -bond_unixctl_list(struct unixctl_conn *conn, - const char *args UNUSED, void *aux UNUSED) +bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c, + struct ofproto_controller *oc) { - struct ds ds = DS_EMPTY_INITIALIZER; - const struct bridge *br; - - ds_put_cstr(&ds, "bridge\tbond\tslaves\n"); - - LIST_FOR_EACH (br, struct bridge, 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", br->name, port->name); - 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, struct bridge, 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; + oc->target = c->target; + oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; + oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; + oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band") + ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); + oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; + oc->burst_limit = (c->controller_burst_limit + ? *c->controller_burst_limit : 0); } +/* Configures the IP stack for 'br''s local interface properly according to the + * configuration in 'c'. */ static void -bond_unixctl_show(struct unixctl_conn *conn, - const char *args, void *aux UNUSED) +bridge_configure_local_iface_netdev(struct bridge *br, + struct ovsrec_controller *c) { - 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, "updelay: %d ms\n", port->updelay); - ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); - ds_put_format(&ds, "next rebalance: %lld ms\n", - port->bridge->bond_next_rebalance - time_msec()); - for (j = 0; j < port->n_ifaces; j++) { - const struct iface *iface = port->ifaces[j]; - struct bond_entry *be; - - /* 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()); - } + struct netdev *netdev; + struct in_addr mask, gateway; - /* Hashes. */ - for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { - int hash = be - port->bond_hash; - struct mac_entry *me; + struct iface *local_iface; + struct in_addr ip; - if (be->iface_idx != j) { - continue; - } - - ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", - hash, be->tx_bytes / 1024); - - /* MACs. */ - LIST_FOR_EACH (me, struct mac_entry, lru_node, - &port->bridge->ml->lrus) { - uint16_t dp_ifidx; - tag_type tags = 0; - if (bond_hash(me->mac) == hash - && me->port != port->port_idx - && choose_output_iface(port, me->mac, &dp_ifidx, &tags) - && dp_ifidx == iface->dp_ifidx) - { - ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n", - ETH_ADDR_ARGS(me->mac)); - } - } - } - } - 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 UNUSED) -{ - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *hash_s, *slave_s; - uint8_t mac[ETH_ADDR_LEN]; - 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"); + /* If there's no local interface or no IP address, give up. */ + local_iface = iface_from_ofp_port(br, OFPP_LOCAL); + if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) { return; } - port = bond_find(bond_s); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } + /* Bring up the local interface. */ + netdev = local_iface->netdev; + netdev_turn_flags_on(netdev, NETDEV_UP, true); - if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) - == ETH_ADDR_SCAN_COUNT) { - hash = bond_hash(mac); - } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) { - hash = atoi(hash_s) & BOND_MASK; - } else { - unixctl_command_reply(conn, 501, "bad hash"); - return; + /* Configure the IP address and netmask. */ + if (!c->local_netmask + || !inet_aton(c->local_netmask, &mask) + || !mask.s_addr) { + mask.s_addr = guess_netmask(ip.s_addr); } - - iface = port_lookup_iface(port, slave_s); - if (!iface) { - unixctl_command_reply(conn, 501, "no such slave"); - return; + if (!netdev_set_in4(netdev, ip, mask)) { + VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT, + br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr)); } - if (!iface->enabled) { - unixctl_command_reply(conn, 501, "cannot migrate to disabled slave"); - return; + /* Configure the default gateway. */ + if (c->local_gateway + && inet_aton(c->local_gateway, &gateway) + && gateway.s_addr) { + if (!netdev_add_router(netdev, gateway)) { + VLOG_INFO("bridge %s: configured gateway "IP_FMT, + br->name, IP_ARGS(&gateway.s_addr)); + } } - - 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 UNUSED) +bridge_configure_remotes(struct bridge *br, + const struct sockaddr_in *managers, size_t n_managers) { - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *slave_s; - struct port *port; - struct iface *iface; + const char *disable_ib_str, *queue_id_str; + bool disable_in_band = false; + int queue_id; - 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; - } + struct ovsrec_controller **controllers; + size_t n_controllers; - port = bond_find(bond_s); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } + enum ofproto_fail_mode fail_mode; - iface = port_lookup_iface(port, slave_s); - if (!iface) { - unixctl_command_reply(conn, 501, "no such slave"); - return; - } + struct ofproto_controller *ocs; + size_t n_ocs; + size_t i; - if (!iface->enabled) { - unixctl_command_reply(conn, 501, "cannot make disabled slave active"); - return; + /* Check if we should disable in-band control on this bridge. */ + disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band"); + if (disable_ib_str && !strcmp(disable_ib_str, "true")) { + disable_in_band = true; } - 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"); + /* Set OpenFlow queue ID for in-band control. */ + queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue"); + queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1; + ofproto_set_in_band_queue(br->ofproto, queue_id); + + if (disable_in_band) { + ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0); } else { - unixctl_command_reply(conn, 200, "no change"); + ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); } -} -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; + n_controllers = bridge_get_controllers(br, &controllers); - 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; - } + ocs = xmalloc((n_controllers + 1) * sizeof *ocs); + n_ocs = 0; - port = bond_find(bond_s); - if (!port) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } + bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]); + for (i = 0; i < n_controllers; i++) { + struct ovsrec_controller *c = controllers[i]; - 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"); -} + if (!strncmp(c->target, "punix:", 6) + || !strncmp(c->target, "unix:", 5)) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); -static void -bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, - void *aux UNUSED) -{ - enable_slave(conn, args, true); -} + /* Prevent remote ovsdb-server users from accessing arbitrary Unix + * domain sockets and overwriting arbitrary local files. */ + VLOG_ERR_RL(&rl, "bridge %s: not adding Unix domain socket " + "controller \"%s\" due to possibility for remote " + "exploit", br->name, c->target); + continue; + } -static void -bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, - void *aux UNUSED) -{ - enable_slave(conn, args, false); -} + bridge_configure_local_iface_netdev(br, c); + bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]); + if (disable_in_band) { + ocs[n_ocs].band = OFPROTO_OUT_OF_BAND; + } + n_ocs++; + } -static void -bond_unixctl_hash(struct unixctl_conn *conn, const char *args, - void *aux UNUSED) -{ - uint8_t mac[ETH_ADDR_LEN]; - uint8_t hash; - char *hash_cstr; + ofproto_set_controllers(br->ofproto, ocs, n_ocs); + free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */ + free(ocs); - if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) - == ETH_ADDR_SCAN_COUNT) { - hash = bond_hash(mac); + /* Set the fail-mode. */ + fail_mode = !br->cfg->fail_mode + || !strcmp(br->cfg->fail_mode, "standalone") + ? OFPROTO_FAIL_STANDALONE + : OFPROTO_FAIL_SECURE; + ofproto_set_fail_mode(br->ofproto, fail_mode); - hash_cstr = xasprintf("%u", hash); - unixctl_command_reply(conn, 200, hash_cstr); - free(hash_cstr); - } else { - unixctl_command_reply(conn, 501, "invalid mac"); - } -} + /* Configure OpenFlow controller connection snooping. */ + if (!ofproto_has_snoops(br->ofproto)) { + struct sset snoops; -static void -bond_init(void) -{ - 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); + 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); + } } /* Port functions. */ static struct port * -port_create(struct bridge *br, const char *name) +port_create(struct bridge *br, const struct ovsrec_port *cfg) { struct port *port; port = xzalloc(sizeof *port); port->bridge = br; - port->port_idx = br->n_ports; - port->vlan = -1; - port->trunks = NULL; - port->name = xstrdup(name); - port->active_iface = -1; + port->name = xstrdup(cfg->name); + port->cfg = cfg; + 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; + 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); return port; } +static const char * +get_port_other_config(const struct ovsrec_port *port, const char *key, + const char *default_value) +{ + const char *value; + + value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config, + key); + return value ? value : default_value; +} + +static const char * +get_interface_other_config(const struct ovsrec_interface *iface, + const char *key, const char *default_value) +{ + const char *value; + + value = get_ovsrec_key_value(&iface->header_, + &ovsrec_interface_col_other_config, key); + return value ? value : default_value; +} + +/* Deletes interfaces from 'port' that are no longer configured for it. */ static void -port_reconfigure(struct port *port, const struct ovsrec_port *cfg) +port_del_ifaces(struct port *port) { - struct shash old_ifaces, new_ifaces; - struct shash_node *node; - unsigned long *trunks; - int vlan; + struct iface *iface, *next; + struct sset new_ifaces; size_t i; - port->cfg = cfg; - - /* Collect old and new interfaces. */ - shash_init(&old_ifaces); - shash_init(&new_ifaces); - for (i = 0; i < port->n_ifaces; i++) { - shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]); - } - for (i = 0; i < cfg->n_interfaces; i++) { - const char *name = cfg->interfaces[i]->name; - if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) { - VLOG_WARN("port %s: %s specified twice as port interface", - port->name, name); + /* Collect list of new interfaces. */ + sset_init(&new_ifaces); + for (i = 0; i < port->cfg->n_interfaces; i++) { + const char *name = port->cfg->interfaces[i]->name; + const char *type = port->cfg->interfaces[i]->name; + if (strcmp(type, "null")) { + sset_add(&new_ifaces, name); } } - port->updelay = cfg->bond_updelay; - if (port->updelay < 0) { - port->updelay = 0; - } - port->updelay = cfg->bond_downdelay; - if (port->downdelay < 0) { - port->downdelay = 0; - } - /* Get rid of deleted interfaces and add new interfaces. */ - SHASH_FOR_EACH (node, &old_ifaces) { - if (!shash_find(&new_ifaces, node->name)) { - iface_destroy(node->data); + /* Get rid of deleted interfaces. */ + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + if (!sset_contains(&new_ifaces, iface->name)) { + iface_destroy(iface); } } - SHASH_FOR_EACH (node, &new_ifaces) { - const struct ovsrec_interface *if_cfg = node->data; - struct iface *iface; - iface = shash_find_data(&old_ifaces, if_cfg->name); - if (!iface) { - iface = iface_create(port, if_cfg); - } - iface->cfg = if_cfg; - } + sset_destroy(&new_ifaces); +} - /* Get VLAN tag. */ - vlan = -1; - if (cfg->tag) { - if (port->n_ifaces < 2) { - vlan = *cfg->tag; - if (vlan >= 0 && vlan <= 4095) { - VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan); - } else { - vlan = -1; - } - } else { - /* It's possible that bonded, VLAN-tagged ports make sense. Maybe - * they even work as-is. But they have not been tested. */ - VLOG_WARN("port %s: VLAN tags not supported on bonded ports", - port->name); +/* Adds new interfaces to 'port' and updates 'type' and 'cfg' members of + * existing ones. */ +static void +port_add_ifaces(struct port *port) +{ + struct shash new_ifaces; + struct shash_node *node; + size_t i; + + /* Collect new ifaces. */ + shash_init(&new_ifaces); + for (i = 0; i < port->cfg->n_interfaces; i++) { + const struct ovsrec_interface *cfg = port->cfg->interfaces[i]; + if (strcmp(cfg->type, "null") + && !shash_add_once(&new_ifaces, cfg->name, cfg)) { + VLOG_WARN("port %s: %s specified twice as port interface", + port->name, cfg->name); + iface_set_ofport(cfg, -1); } } - if (port->vlan != vlan) { - port->vlan = vlan; - bridge_flush(port->bridge); - } - /* Get trunked VLANs. */ - trunks = NULL; - if (vlan < 0) { - size_t n_errors; - size_t i; + /* Create new interfaces. + * Update interface types and 'cfg' members. */ + SHASH_FOR_EACH (node, &new_ifaces) { + const struct ovsrec_interface *cfg = node->data; + const char *iface_name = node->name; + struct iface *iface; - 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) { - if (n_errors) { - VLOG_ERR("port %s: no valid trunks, trunking all VLANs", - port->name); - } - bitmap_set_multiple(trunks, 0, 4096, 1); - } - } else { - if (cfg->n_trunks) { - VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", - port->name); + iface = iface_lookup(port->bridge, iface_name); + if (!iface) { + iface = iface_create(port, cfg); + } else { + iface->cfg = cfg; } - } - if (trunks == NULL - ? port->trunks != NULL - : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) { - bridge_flush(port->bridge); - } - bitmap_free(port->trunks); - port->trunks = trunks; - shash_destroy(&old_ifaces); + /* Determine interface type. The local port always has type + * "internal". Other ports take their type from the database and + * default to "system" if none is specified. */ + iface->type = (!strcmp(iface_name, port->bridge->name) ? "internal" + : cfg->type[0] ? cfg->type + : "system"); + } shash_destroy(&new_ifaces); } @@ -3100,49 +2131,32 @@ port_destroy(struct port *port) { if (port) { struct bridge *br = port->bridge; - struct port *del; - - proc_net_compat_update_vlan(port->name, NULL, 0); - proc_net_compat_update_bond(port->name, NULL); + struct iface *iface, *next; -#if 0 - for (i = 0; i < MAX_MIRRORS; i++) { - struct mirror *m = br->mirrors[i]; - if (m && m->out_port == port) { - mirror_destroy(m); - } + if (br->ofproto) { + ofproto_bundle_unregister(br->ofproto, port); } -#endif - 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); } - 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); - free(port->ifaces); - bitmap_free(port->trunks); free(port->name); free(port); - bridge_flush(br); } } -static struct port * -port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) -{ - struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx); - return iface ? iface->port : NULL; -} - static struct port * port_lookup(const struct bridge *br, const char *name) { - size_t i; + struct port *port; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0), + &br->ports) { if (!strcmp(port->name, name)) { return port; } @@ -3150,157 +2164,161 @@ port_lookup(const struct bridge *br, const char *name) return NULL; } -static struct iface * -port_lookup_iface(const struct port *port, const char *name) +static bool +enable_lacp(struct port *port, bool *activep) { - size_t j; - - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (!strcmp(iface->name, name)) { - return iface; - } + 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 { + VLOG_WARN("port %s: unknown LACP mode %s", + port->name, port->cfg->lacp); + return false; } - return NULL; } -static void -port_update_bonding(struct port *port) +static struct lacp_settings * +port_configure_lacp(struct port *port, struct lacp_settings *s) { - 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; - } + const char *lacp_time; + long long int custom_time; + int priority; + + if (!enable_lacp(port, &s->active)) { + return NULL; + } + + s->name = port->name; + memcpy(s->id, port->bridge->ea, ETH_ADDR_LEN); + + /* 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->heartbeat = !strcmp(get_port_other_config(port->cfg, + "lacp-heartbeat", + "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 { - if (!port->bond_hash) { - size_t i; - - 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_compat_is_stale = true; - port->bond_fake_iface = port->cfg->bond_fake_iface; + s->lacp_time = LACP_TIME_SLOW; } + + return s; } static void -port_update_bond_compat(struct port *port) +iface_configure_lacp(struct iface *iface, struct lacp_slave_settings *s) { - struct compat_bond_hash compat_hashes[BOND_MASK + 1]; - struct compat_bond bond; - size_t i; + int priority, portid, key; - if (port->n_ifaces < 2) { - proc_net_compat_update_bond(port->name, NULL); - return; + portid = atoi(get_interface_other_config(iface->cfg, "lacp-port-id", "0")); + priority = atoi(get_interface_other_config(iface->cfg, + "lacp-port-priority", "0")); + key = atoi(get_interface_other_config(iface->cfg, "lacp-aggregation-key", + "0")); + + if (portid <= 0 || portid > UINT16_MAX) { + portid = iface->ofp_port; } - bond.up = false; - bond.updelay = port->updelay; - bond.downdelay = port->downdelay; + if (priority <= 0 || priority > UINT16_MAX) { + priority = UINT16_MAX; + } - 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; - } - } + if (key < 0 || key > UINT16_MAX) { + key = 0; } - 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; + s->name = iface->name; + s->id = portid; + s->priority = priority; + s->key = key; +} - /* 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); +static void +port_configure_bond(struct port *port, struct bond_settings *s, + uint32_t *bond_stable_ids) +{ + const char *detect_s; + struct iface *iface; + int miimon_interval; + size_t i; + + 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 (s->balance == BM_SLB && port->bridge->cfg->n_flood_vlans) { + VLOG_WARN("port %s: SLB bonds are incompatible with flood_vlans, " + "please use another bond type or disable flood_vlans", + port->name); } - if (port->bond_fake_iface) { - struct netdev *bond_netdev; + miimon_interval = atoi(get_port_other_config(port->cfg, + "bond-miimon-interval", "0")); + if (miimon_interval <= 0) { + miimon_interval = 200; + } - if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) { - if (bond.up) { - netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); - } else { - netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); - } - netdev_close(bond_netdev); - } + detect_s = get_port_other_config(port->cfg, "bond-detect-mode", "carrier"); + if (!strcmp(detect_s, "carrier")) { + miimon_interval = 0; + } else if (strcmp(detect_s, "miimon")) { + VLOG_WARN("port %s: unsupported bond-detect-mode %s, " + "defaulting to carrier", port->name, detect_s); + miimon_interval = 0; } - proc_net_compat_update_bond(port->name, &bond); - free(bond.slaves); -} + s->up_delay = MAX(0, port->cfg->bond_updelay); + s->down_delay = MAX(0, port->cfg->bond_downdelay); + s->basis = atoi(get_port_other_config(port->cfg, "bond-hash-basis", "0")); + s->rebalance_interval = atoi( + get_port_other_config(port->cfg, "bond-rebalance-interval", "10000")); + if (s->rebalance_interval < 1000) { + s->rebalance_interval = 1000; + } -static void -port_update_vlan_compat(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; - } - } + s->fake_iface = port->cfg->bond_fake_iface; + + i = 0; + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + long long stable_id; + + stable_id = atoll(get_interface_other_config(iface->cfg, + "bond-stable-id", "0")); + if (stable_id <= 0 || stable_id >= UINT32_MAX) { + stable_id = iface->ofp_port; } + bond_stable_ids[i++] = stable_id; + + netdev_set_miimon_interval(iface->netdev, miimon_interval); } - proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan); } /* Interface functions. */ @@ -3308,40 +2326,24 @@ port_update_vlan_compat(struct port *port) static struct iface * iface_create(struct port *port, const struct ovsrec_interface *if_cfg) { + struct bridge *br = port->bridge; struct iface *iface; char *name = if_cfg->name; - int error; iface = xzalloc(sizeof *iface); iface->port = port; - iface->port_ifidx = port->n_ifaces; iface->name = xstrdup(name); - iface->dp_ifidx = -1; + iface->ofp_port = -1; iface->tag = tag_create_random(); - iface->delay_expires = LLONG_MAX; iface->netdev = NULL; + iface->cfg = if_cfg; - if (port->n_ifaces >= port->allocated_ifaces) { - port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, - sizeof *port->ifaces); - } - port->ifaces[port->n_ifaces++] = iface; - if (port->n_ifaces > 1) { - port->bridge->has_bonded_ports = true; - } + hmap_insert(&br->iface_by_name, &iface->name_node, hash_string(name, 0)); - /* Attempt to create the network interface in case it - * doesn't exist yet. */ - error = set_up_iface(if_cfg, true); - if (error) { - VLOG_WARN("could not create iface %s: %s\n", iface->name, - strerror(error)); - } + list_push_back(&port->ifaces, &iface->port_elem); VLOG_DBG("attached network device %s to port %s", iface->name, port->name); - bridge_flush(port->bridge); - return iface; } @@ -3351,86 +2353,67 @@ 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 (iface->dp_ifidx >= 0) { - port_array_set(&br->ifaces, iface->dp_ifidx, NULL); + if (br->ofproto && iface->ofp_port >= 0) { + ofproto_port_unregister(br->ofproto, iface->ofp_port); } - del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces]; - del->port_ifidx = iface->port_ifidx; + if (iface->ofp_port >= 0) { + hmap_remove(&br->ifaces, &iface->ofp_port_node); + } - netdev_close(iface->netdev); + list_remove(&iface->port_elem); + hmap_remove(&br->iface_by_name, &iface->name_node); - if (del_active) { - ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); - bond_choose_active_iface(port); - bond_send_learning_packets(port); - } + netdev_close(iface->netdev); - netdev_destroy(iface->name); free(iface->name); free(iface); - - bridge_flush(port->bridge); } } static struct iface * iface_lookup(const struct bridge *br, const char *name) { - size_t i, j; + struct iface *iface; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (!strcmp(iface->name, name)) { - return iface; - } + HMAP_FOR_EACH_WITH_HASH (iface, name_node, hash_string(name, 0), + &br->iface_by_name) { + if (!strcmp(iface->name, name)) { + return iface; } } + return NULL; } static struct iface * -iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) +iface_find(const char *name) { - return port_array_get(&br->ifaces, dp_ifidx); -} + const struct bridge *br; -/* Returns true if 'iface' is the name of an "internal" interface on bridge - * 'br', that is, an interface that is entirely simulated within the datapath. - * The local port (ODPP_LOCAL) is always an internal interface. Other local - * interfaces are created by setting "iface..internal = true". - * - * In addition, we have a kluge-y feature that creates an internal port with - * the name of a bonded port if "bonding..fake-iface = true" is set. - * This feature needs to go away in the long term. Until then, this is one - * reason why this function takes a name instead of a struct iface: the fake - * interfaces created this way do not have a struct iface. */ -static bool -iface_is_internal(const struct bridge *br, const char *if_name) -{ - /* XXX wastes time */ - struct iface *iface; - struct port *port; + HMAP_FOR_EACH (br, node, &all_bridges) { + struct iface *iface = iface_lookup(br, name); - if (!strcmp(if_name, br->name)) { - return true; + if (iface) { + return iface; + } } + return NULL; +} - iface = iface_lookup(br, if_name); - if (iface && !strcmp(iface->cfg->type, "internal")) { - return true; - } +static struct iface * +iface_from_ofp_port(const struct bridge *br, uint16_t ofp_port) +{ + struct iface *iface; - port = port_lookup(br, if_name); - if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { - return true; + HMAP_FOR_EACH_IN_BUCKET (iface, ofp_port_node, + hash_int(ofp_port, 0), &br->ifaces) { + if (iface->ofp_port == ofp_port) { + return iface; + } } - return false; + return NULL; } /* Set Ethernet address of 'iface', if one is specified in the configuration @@ -3440,13 +2423,15 @@ 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->ofp_port == OFPP_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) { @@ -3456,336 +2441,368 @@ iface_set_mac(struct iface *iface) } } } - -/* Port mirroring. */ -#if 0 +/* Sets the ofport column of 'if_cfg' to 'ofport'. */ static void -mirror_reconfigure(struct bridge *br UNUSED) -{ - struct svec old_mirrors, new_mirrors; - size_t i, n_rspan_vlans; - unsigned long *rspan_vlans; - - /* Collect old and new mirrors. */ - svec_init(&old_mirrors); - svec_init(&new_mirrors); - cfg_get_subsections(&new_mirrors, "mirror.%s", br->name); - for (i = 0; i < MAX_MIRRORS; i++) { - if (br->mirrors[i]) { - svec_add(&old_mirrors, br->mirrors[i]->name); - } - } - - /* Get rid of deleted mirrors and add new mirrors. */ - svec_sort(&old_mirrors); - assert(svec_is_unique(&old_mirrors)); - svec_sort(&new_mirrors); - assert(svec_is_unique(&new_mirrors)); - for (i = 0; i < MAX_MIRRORS; i++) { - struct mirror *m = br->mirrors[i]; - if (m && !svec_contains(&new_mirrors, m->name)) { - mirror_destroy(m); - } - } - for (i = 0; i < new_mirrors.n; i++) { - const char *name = new_mirrors.names[i]; - if (!svec_contains(&old_mirrors, name)) { - mirror_create(br, name); - } +iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport) +{ + if (if_cfg && !ovsdb_idl_row_is_synthetic(&if_cfg->header_)) { + ovsrec_interface_set_ofport(if_cfg, &ofport, 1); } - svec_destroy(&old_mirrors); - svec_destroy(&new_mirrors); +} - /* Reconfigure all mirrors. */ - for (i = 0; i < MAX_MIRRORS; i++) { - if (br->mirrors[i]) { - mirror_reconfigure_one(br->mirrors[i]); - } - } +/* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'. + * + * The value strings in '*shash' are taken directly from values[], not copied, + * so the caller should not modify or free them. */ +static void +shash_from_ovs_idl_map(char **keys, char **values, size_t n, + struct shash *shash) +{ + size_t i; - /* Update port reserved status. */ - for (i = 0; i < br->n_ports; i++) { - br->ports[i]->is_mirror_output_port = false; - } - for (i = 0; i < MAX_MIRRORS; i++) { - struct mirror *m = br->mirrors[i]; - if (m && m->out_port) { - m->out_port->is_mirror_output_port = true; - } + shash_init(shash); + for (i = 0; i < n; i++) { + shash_add(shash, keys[i], values[i]); } +} - /* Update learning disabled vlans (for RSPAN). */ - rspan_vlans = NULL; - n_rspan_vlans = cfg_count("vlan.%s.disable-learning", br->name); - if (n_rspan_vlans) { - rspan_vlans = bitmap_allocate(4096); +/* Creates 'keys' and 'values' arrays from 'shash'. + * + * Sets 'keys' and 'values' to heap allocated arrays representing the key-value + * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They + * are populated with with strings taken directly from 'shash' and thus have + * the same ownership of the key-value pairs in shash. + */ +static void +shash_to_ovs_idl_map(struct shash *shash, + char ***keys, char ***values, size_t *n) +{ + size_t i, count; + char **k, **v; + struct shash_node *sn; - for (i = 0; i < n_rspan_vlans; i++) { - int vlan = cfg_get_vlan(i, "vlan.%s.disable-learning", br->name); - if (vlan >= 0) { - bitmap_set1(rspan_vlans, vlan); - VLOG_INFO("bridge %s: disabling learning on vlan %d\n", - br->name, vlan); - } else { - VLOG_ERR("bridge %s: invalid value '%s' for learning disabled " - "VLAN", br->name, - cfg_get_string(i, "vlan.%s.disable-learning", br->name)); - } - } + count = shash_count(shash); + + k = xmalloc(count * sizeof *k); + v = xmalloc(count * sizeof *v); + + i = 0; + SHASH_FOR_EACH(sn, shash) { + k[i] = sn->name; + v[i] = sn->data; + i++; } - if (mac_learning_set_disabled_vlans(br->ml, rspan_vlans)) { - bridge_flush(br); + + *n = count; + *keys = k; + *values = v; +} + +struct iface_delete_queues_cbdata { + struct netdev *netdev; + const struct ovsdb_datum *queues; +}; + +static bool +queue_ids_include(const struct ovsdb_datum *queues, int64_t target) +{ + union ovsdb_atom atom; + + atom.integer = target; + return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX; +} + +static void +iface_delete_queues(unsigned int queue_id, + const struct shash *details OVS_UNUSED, void *cbdata_) +{ + struct iface_delete_queues_cbdata *cbdata = cbdata_; + + if (!queue_ids_include(cbdata->queues, queue_id)) { + netdev_delete_queue(cbdata->netdev, queue_id); } } static void -mirror_create(struct bridge *br, const char *name) +iface_configure_qos(struct iface *iface, const struct ovsrec_qos *qos) { - struct mirror *m; - size_t i; + if (!qos || qos->type[0] == '\0' || qos->n_queues < 1) { + netdev_set_qos(iface->netdev, NULL, NULL); + } else { + struct iface_delete_queues_cbdata cbdata; + struct shash details; + bool queue_zero; + size_t i; - for (i = 0; ; i++) { - if (i >= MAX_MIRRORS) { - VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " - "cannot create %s", br->name, MAX_MIRRORS, name); - return; + /* Configure top-level Qos for 'iface'. */ + shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config, + qos->n_other_config, &details); + netdev_set_qos(iface->netdev, qos->type, &details); + shash_destroy(&details); + + /* Deconfigure queues that were deleted. */ + cbdata.netdev = iface->netdev; + cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER, + OVSDB_TYPE_UUID); + netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata); + + /* Configure queues for 'iface'. */ + queue_zero = false; + for (i = 0; i < qos->n_queues; i++) { + const struct ovsrec_queue *queue = qos->value_queues[i]; + unsigned int queue_id = qos->key_queues[i]; + + if (queue_id == 0) { + queue_zero = true; + } + + shash_from_ovs_idl_map(queue->key_other_config, + queue->value_other_config, + queue->n_other_config, &details); + netdev_set_queue(iface->netdev, queue_id, &details); + shash_destroy(&details); } - if (!br->mirrors[i]) { - break; + if (!queue_zero) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); + VLOG_WARN_RL(&rl, "interface %s: QoS configured without a default " + "queue (queue 0). Packets not directed to a " + "correctly configured queue may be dropped.", + iface->name); } } - VLOG_INFO("created port mirror %s on bridge %s", name, br->name); - bridge_flush(br); - - br->mirrors[i] = m = xzalloc(sizeof *m); - m->bridge = br; - m->idx = i; - m->name = xstrdup(name); - svec_init(&m->src_ports); - svec_init(&m->dst_ports); - m->vlans = NULL; - m->n_vlans = 0; - m->out_vlan = -1; - m->out_port = NULL; + netdev_set_policing(iface->netdev, + iface->cfg->ingress_policing_rate, + iface->cfg->ingress_policing_burst); } static void -mirror_destroy(struct mirror *m) +iface_configure_cfm(struct iface *iface) { - if (m) { - struct bridge *br = m->bridge; - size_t i; + const struct ovsrec_interface *cfg = iface->cfg; + struct cfm_settings s; + uint16_t remote_mpid; - 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); - } + if (!cfg->n_cfm_mpid || !cfg->n_cfm_remote_mpid) { + ofproto_port_clear_cfm(iface->port->bridge->ofproto, iface->ofp_port); + return; + } - svec_destroy(&m->src_ports); - svec_destroy(&m->dst_ports); - free(m->vlans); + s.mpid = *cfg->cfm_mpid; + remote_mpid = *cfg->cfm_remote_mpid; + s.remote_mpids = &remote_mpid; + s.n_remote_mpids = 1; - m->bridge->mirrors[m->idx] = NULL; - free(m); + s.interval = atoi(get_interface_other_config(iface->cfg, "cfm_interval", + "0")); + if (s.interval <= 0) { + s.interval = 1000; + } + + ofproto_port_set_cfm(iface->port->bridge->ofproto, iface->ofp_port, &s); +} + +/* 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. */ + +static struct mirror * +mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid) +{ + struct mirror *m; - bridge_flush(br); + HMAP_FOR_EACH_IN_BUCKET (m, hmap_node, uuid_hash(uuid), &br->mirrors) { + if (uuid_equals(uuid, &m->uuid)) { + return m; + } } + return NULL; } static void -prune_ports(struct mirror *m, struct svec *ports) +bridge_configure_mirrors(struct bridge *br) { - struct svec tmp; + const struct ovsdb_datum *mc; + unsigned long *flood_vlans; + struct mirror *m, *next; size_t i; - svec_sort_unique(ports); + /* Get rid of deleted mirrors. */ + mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID); + HMAP_FOR_EACH_SAFE (m, next, hmap_node, &br->mirrors) { + union ovsdb_atom atom; - svec_init(&tmp); - for (i = 0; i < ports->n; i++) { - const char *name = ports->names[i]; - if (port_lookup(m->bridge, name)) { - svec_add(&tmp, name); - } else { - VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s", - m->bridge->name, m->name, name); + atom.uuid = m->uuid; + if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) { + mirror_destroy(m); + } + } + + /* Add new mirrors and reconfigure existing ones. */ + for (i = 0; i < br->cfg->n_mirrors; i++) { + const struct ovsrec_mirror *cfg = br->cfg->mirrors[i]; + struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid); + if (!m) { + m = mirror_create(br, cfg); + } + if (!mirror_configure(m, cfg)) { + mirror_destroy(m); } } - svec_swap(ports, &tmp); - svec_destroy(&tmp); + + /* Update flooded vlans (for RSPAN). */ + flood_vlans = vlan_bitmap_from_array(br->cfg->flood_vlans, + br->cfg->n_flood_vlans); + ofproto_set_flood_vlans(br->ofproto, flood_vlans); + bitmap_free(flood_vlans); } -static size_t -prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans) +static struct mirror * +mirror_create(struct bridge *br, const struct ovsrec_mirror *cfg) { - size_t n_vlans, i; + struct mirror *m; - /* This isn't perfect: it won't combine "0" and "00", and the textual sort - * order won't give us numeric sort order. But that's good enough for what - * we need right now. */ - svec_sort_unique(vlan_strings); + m = xzalloc(sizeof *m); + m->uuid = cfg->header_.uuid; + hmap_insert(&br->mirrors, &m->hmap_node, uuid_hash(&m->uuid)); + m->bridge = br; + m->name = xstrdup(cfg->name); - *vlans = xmalloc(sizeof *vlans * vlan_strings->n); - n_vlans = 0; - for (i = 0; i < vlan_strings->n; i++) { - const char *name = vlan_strings->names[i]; - int vlan; - if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) { - VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s", - m->bridge->name, m->name, name); - } else { - (*vlans)[n_vlans++] = vlan; + return m; +} + +static void +mirror_destroy(struct mirror *m) +{ + if (m) { + struct bridge *br = m->bridge; + + if (br->ofproto) { + ofproto_mirror_unregister(br->ofproto, m); } + + hmap_remove(&br->mirrors, &m->hmap_node); + free(m->name); + free(m); } - return n_vlans; } -static bool -vlan_is_mirrored(const struct mirror *m, int vlan) +static void +mirror_collect_ports(struct mirror *m, + struct ovsrec_port **in_ports, int n_in_ports, + void ***out_portsp, size_t *n_out_portsp) { + void **out_ports = xmalloc(n_in_ports * sizeof *out_ports); + size_t n_out_ports = 0; size_t i; - for (i = 0; i < m->n_vlans; i++) { - if (m->vlans[i] == vlan) { - return true; + for (i = 0; i < n_in_ports; i++) { + const char *name = in_ports[i]->name; + struct port *port = port_lookup(m->bridge, name); + if (port) { + out_ports[n_out_ports++] = port; + } else { + VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent " + "port %s", m->bridge->name, m->name, name); } } - return false; + *out_portsp = out_ports; + *n_out_portsp = n_out_ports; } static bool -port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) +mirror_configure(struct mirror *m, const struct ovsrec_mirror *cfg) { - size_t i; + struct ofproto_mirror_settings s; - for (i = 0; i < m->n_vlans; i++) { - if (port_trunks_vlan(p, m->vlans[i])) { - return true; - } + /* Set name. */ + if (strcmp(cfg->name, m->name)) { + free(m->name); + m->name = xstrdup(cfg->name); } - return false; -} + s.name = m->name; -static void -mirror_reconfigure_one(struct mirror *m UNUSED) -{ - char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name); - struct svec src_ports, dst_ports, ports; - struct svec vlan_strings; - mirror_mask_t mirror_bit; - const char *out_port_name; - struct port *out_port; - int out_vlan; - size_t n_vlans; - int *vlans; - size_t i; - bool mirror_all_ports; - bool any_ports_specified; - - /* Get output port. */ - out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port", - m->bridge->name, m->name); - if (out_port_name) { - out_port = port_lookup(m->bridge, out_port_name); - if (!out_port) { - VLOG_ERR("%s.output.port: bridge %s does not have a port " - "named %s", pfx, m->bridge->name, out_port_name); - mirror_destroy(m); - free(pfx); - return; + /* Get output port or VLAN. */ + if (cfg->output_port) { + s.out_bundle = port_lookup(m->bridge, cfg->output_port->name); + if (!s.out_bundle) { + VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge", + m->bridge->name, m->name); + return false; } - out_vlan = -1; + s.out_vlan = UINT16_MAX; - if (cfg_has("%s.output.vlan", pfx)) { - VLOG_ERR("%s.output.port and %s.output.vlan both specified; " - "ignoring %s.output.vlan", pfx, pfx, pfx); + if (cfg->output_vlan) { + VLOG_ERR("bridge %s: mirror %s specifies both output port and " + "output vlan; ignoring output vlan", + m->bridge->name, m->name); } - } else if (cfg_has("%s.output.vlan", pfx)) { - out_port = NULL; - out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx); + } else if (cfg->output_vlan) { + /* The database should prevent invalid VLAN values. */ + s.out_bundle = NULL; + s.out_vlan = *cfg->output_vlan; } else { - VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, " - "but exactly one is required; disabling port mirror %s", - pfx, pfx, pfx, pfx); - mirror_destroy(m); - free(pfx); - return; + VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring", + m->bridge->name, m->name); + return false; } - /* Get all the ports, and drop duplicates and ports that don't exist. */ - svec_init(&src_ports); - svec_init(&dst_ports); - svec_init(&ports); - cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx); - cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx); - cfg_get_all_keys(&ports, "%s.select.port", pfx); - any_ports_specified = src_ports.n || dst_ports.n || ports.n; - svec_append(&src_ports, &ports); - svec_append(&dst_ports, &ports); - svec_destroy(&ports); - prune_ports(m, &src_ports); - prune_ports(m, &dst_ports); - if (any_ports_specified && !src_ports.n && !dst_ports.n) { - VLOG_ERR("%s: none of the specified ports exist; " - "disabling port mirror %s", pfx, pfx); - mirror_destroy(m); - goto exit; - } - - /* Get all the vlans, and drop duplicate and invalid vlans. */ - svec_init(&vlan_strings); - cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx); - n_vlans = prune_vlans(m, &vlan_strings, &vlans); - svec_destroy(&vlan_strings); - - /* Update mirror data. */ - if (!svec_equal(&m->src_ports, &src_ports) - || !svec_equal(&m->dst_ports, &dst_ports) - || m->n_vlans != n_vlans - || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans) - || m->out_port != out_port - || m->out_vlan != out_vlan) { - bridge_flush(m->bridge); - } - svec_swap(&m->src_ports, &src_ports); - svec_swap(&m->dst_ports, &dst_ports); - free(m->vlans); - m->vlans = vlans; - m->n_vlans = n_vlans; - m->out_port = out_port; - m->out_vlan = out_vlan; - - /* If no selection criteria have been given, mirror for all ports. */ - mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans); - - /* Update ports. */ - mirror_bit = MIRROR_MASK_C(1) << m->idx; - for (i = 0; i < m->bridge->n_ports; i++) { - struct port *port = m->bridge->ports[i]; - - if (mirror_all_ports - || svec_contains(&m->src_ports, port->name) - || (m->n_vlans - && (!port->vlan - ? port_trunks_any_mirrored_vlan(m, port) - : vlan_is_mirrored(m, port->vlan)))) { - port->src_mirrors |= mirror_bit; - } else { - port->src_mirrors &= ~mirror_bit; - } + /* Get port selection. */ + if (cfg->select_all) { + size_t n_ports = hmap_count(&m->bridge->ports); + void **ports = xmalloc(n_ports * sizeof *ports); + struct port *port; + size_t i; - if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) { - port->dst_mirrors |= mirror_bit; - } else { - port->dst_mirrors &= ~mirror_bit; + i = 0; + HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) { + ports[i++] = port; } + + s.srcs = ports; + s.n_srcs = n_ports; + + s.dsts = ports; + s.n_dsts = n_ports; + } else { + /* Get ports, dropping ports that don't exist. + * The IDL ensures that there are no duplicates. */ + mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port, + &s.srcs, &s.n_srcs); + mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port, + &s.dsts, &s.n_dsts); } + /* Get VLAN selection. */ + s.src_vlans = vlan_bitmap_from_array(cfg->select_vlan, cfg->n_select_vlan); + + /* Configure. */ + ofproto_mirror_register(m->bridge->ofproto, m, &s); + /* Clean up. */ -exit: - svec_destroy(&src_ports); - svec_destroy(&dst_ports); - free(pfx); + if (s.srcs != s.dsts) { + free(s.dsts); + } + free(s.srcs); + free(s.src_vlans); + + return true; } -#endif