X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=7fcfd58eddf5bf57061a9e753363c172e678c1f3;hb=130f6e5faa6190ca00000ebcb627e2d4ca3d4b37;hp=7081512e9273ccef5010471f21a561a87e221290;hpb=8398cf7efc3979cd6c8152915a584c13873ee322;p=openvswitch diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 7081512e..3637ec65 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. @@ -15,11 +15,13 @@ #include #include "bridge.h" +#include "byte-order.h" #include #include #include #include #include +#include #include #include #include @@ -30,45 +32,65 @@ #include #include #include "bitmap.h" -#include "cfg.h" +#include "cfm.h" +#include "classifier.h" #include "coverage.h" #include "dirs.h" #include "dpif.h" #include "dynamic-string.h" #include "flow.h" #include "hash.h" +#include "hmap.h" +#include "jsonrpc.h" #include "list.h" #include "mac-learning.h" #include "netdev.h" +#include "netlink.h" #include "odp-util.h" #include "ofp-print.h" #include "ofpbuf.h" +#include "ofproto/netflow.h" #include "ofproto/ofproto.h" +#include "ovsdb-data.h" #include "packets.h" #include "poll-loop.h" -#include "port-array.h" #include "proc-net-compat.h" #include "process.h" +#include "sha1.h" +#include "shash.h" #include "socket-util.h" -#include "stp.h" +#include "stream-ssl.h" #include "svec.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" + +VLOG_DEFINE_THIS_MODULE(bridge); + +COVERAGE_DEFINE(bridge_flush); +COVERAGE_DEFINE(bridge_process_flow); +COVERAGE_DEFINE(bridge_reconfigure); struct dst { uint16_t vlan; uint16_t dp_ifidx; }; -extern uint64_t mgmt_id; +struct dst_set { + struct dst builtin[32]; + struct dst *dsts; + size_t n, allocated; +}; + +static void dst_set_init(struct dst_set *); +static void dst_set_add(struct dst_set *, const struct dst *); +static void dst_set_free(struct dst_set *); struct iface { /* These members are always valid. */ @@ -79,10 +101,14 @@ struct iface { long long delay_expires; /* Time after which 'enabled' may change. */ /* These members are valid only after bridge_reconfigure() causes them to - * be initialized.*/ + * be initialized. */ + struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */ int dp_ifidx; /* Index within kernel datapath. */ struct netdev *netdev; /* Network device. */ bool enabled; /* May be chosen for flows? */ + const char *type; /* Usually same as cfg->type. */ + struct cfm *cfm; /* Connectivity Fault Management */ + const struct ovsrec_interface *cfg; }; #define BOND_MASK 0xff @@ -92,6 +118,11 @@ struct bond_entry { tag_type iface_tag; /* Tag associated with iface_idx. */ }; +enum bond_mode { + BM_SLB, /* Source Load Balance (Default). */ + BM_AB /* Active Backup. */ +}; + #define MAX_MIRRORS 32 typedef uint32_t mirror_mask_t; #define MIRROR_MASK_C(X) UINT32_C(X) @@ -100,10 +131,11 @@ struct mirror { struct bridge *bridge; size_t idx; char *name; + struct uuid uuid; /* UUID of this "mirror" record in database. */ /* Selection criteria. */ - struct svec src_ports; - struct svec dst_ports; + struct shash src_ports; /* Name is port name; data is always NULL. */ + struct shash dst_ports; /* Name is port name; data is always NULL. */ int *vlans; size_t n_vlans; @@ -117,7 +149,9 @@ struct port { struct bridge *bridge; size_t port_idx; int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ - unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */ + unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. + * NULL if all VLANs are trunked. */ + const struct ovsrec_port *cfg; char *name; /* An ordinary bridge port has 1 interface. @@ -126,79 +160,84 @@ struct port { size_t n_ifaces, allocated_ifaces; /* Bonding info. */ - struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ + enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */ int active_iface; /* Ifidx on which bcasts accepted, or -1. */ tag_type active_iface_tag; /* Tag for bcast flows. */ tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */ int updelay, downdelay; /* Delay before iface goes up/down, in ms. */ bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */ + bool bond_fake_iface; /* Fake a bond interface for legacy compat? */ + bool miimon; /* Use miimon instead of carrier? */ + long long int bond_miimon_interval; /* Miimon status refresh interval. */ + long long int bond_miimon_next_update; /* Time of next miimon update. */ + long long int bond_next_fake_iface_update; /* Time of next update. */ + struct netdev_monitor *monitor; /* Tracks carrier up/down status. */ + + /* SLB specific bonding info. */ + struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ + int bond_rebalance_interval; /* Interval between rebalances, in ms. */ + long long int bond_next_rebalance; /* Next rebalancing time. */ /* Port mirroring info. */ mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ bool is_mirror_output_port; /* Does port mirroring send frames here? */ - - /* Spanning tree info. */ - enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */ - tag_type stp_state_tag; /* Tag for STP state change. */ }; -#define DP_MAX_PORTS 255 struct bridge { struct list node; /* Node in global list of bridges. */ char *name; /* User-specified arbitrary name. */ - struct mac_learning *ml; /* MAC learning table, or null not to learn. */ - bool sent_config_request; /* Successfully sent config request? */ + struct mac_learning *ml; /* MAC learning table. */ 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. */ + struct hmap ifaces; /* Contains "struct iface"s. */ /* Bridge ports. */ struct port **ports; size_t n_ports, allocated_ports; + struct shash iface_by_name; /* "struct iface"s indexed by name. */ + struct shash port_by_name; /* "struct port"s indexed by name. */ /* Bonding. */ bool has_bonded_ports; - long long int bond_next_rebalance; /* Flow tracking. */ bool flush; - /* Flow statistics gathering. */ - time_t next_stats_request; - /* Port mirroring. */ struct mirror *mirrors[MAX_MIRRORS]; - - /* Spanning tree. */ - struct stp *stp; - long long int stp_last_tick; }; /* List of all bridges. */ static struct list all_bridges = LIST_INITIALIZER(&all_bridges); -/* Maximum number of datapaths. */ -enum { DP_MAX = 256 }; +/* OVSDB IDL used to obtain configuration. */ +static struct ovsdb_idl *idl; -static struct bridge *bridge_create(const char *name); +/* Each time this timer expires, the bridge fetches 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; + +static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg); static void bridge_destroy(struct bridge *); static struct bridge *bridge_lookup(const char *name); -static void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *); +static unixctl_cb_func bridge_unixctl_dump_flows; +static unixctl_cb_func bridge_unixctl_reconnect; static int bridge_run_one(struct bridge *); +static size_t bridge_get_controllers(const struct bridge *br, + struct ovsrec_controller ***controllersp); static void bridge_reconfigure_one(struct bridge *); -static void bridge_reconfigure_controller(struct bridge *); -static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces); +static void bridge_reconfigure_remotes(struct bridge *, + const struct sockaddr_in *managers, + size_t n_managers); +static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces); static void bridge_fetch_dp_ifaces(struct bridge *); static void bridge_flush(struct bridge *); static void bridge_pick_local_hw_addr(struct bridge *, @@ -210,16 +249,18 @@ static uint64_t bridge_pick_datapath_id(struct bridge *, static struct iface *bridge_get_local_iface(struct bridge *); static uint64_t dpid_from_hash(const void *, size_t nbytes); -static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args); +static unixctl_cb_func bridge_unixctl_fdb_show; static void bond_init(void); static void bond_run(struct bridge *); static void bond_wait(struct bridge *); static void bond_rebalance_port(struct port *); static void bond_send_learning_packets(struct port *); +static void bond_enable_slave(struct iface *iface, bool enable); -static void port_create(struct bridge *, const char *name); -static void port_reconfigure(struct port *); +static struct port *port_create(struct bridge *, const char *name); +static void port_reconfigure(struct port *, const struct ovsrec_port *); +static void port_del_ifaces(struct port *, const struct ovsrec_port *); static void port_destroy(struct port *); static struct port *port_lookup(const struct bridge *, const char *name); static struct iface *port_lookup_iface(const struct port *, const char *name); @@ -229,183 +270,169 @@ static void port_update_bond_compat(struct port *); static void port_update_vlan_compat(struct port *); static void port_update_bonding(struct port *); -static void mirror_create(struct bridge *, const char *name); +static void mirror_create(struct bridge *, struct ovsrec_mirror *); static void mirror_destroy(struct mirror *); static void mirror_reconfigure(struct bridge *); -static void mirror_reconfigure_one(struct mirror *); +static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *); static bool vlan_is_mirrored(const struct mirror *, int vlan); -static void brstp_reconfigure(struct bridge *); -static void brstp_adjust_timers(struct bridge *); -static void brstp_run(struct bridge *); -static void brstp_wait(struct bridge *); - -static void iface_create(struct port *, const char *name); +static struct iface *iface_create(struct port *port, + const struct ovsrec_interface *if_cfg); static void iface_destroy(struct iface *); static struct iface *iface_lookup(const struct bridge *, const char *name); static struct iface *iface_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); +static void iface_set_mac(struct iface *); +static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport); +static void iface_update_qos(struct iface *, const struct ovsrec_qos *); +static void iface_update_cfm(struct iface *); +static void iface_refresh_cfm_stats(struct iface *iface); +static void iface_send_packet(struct iface *, struct ofpbuf *packet); + +static void shash_from_ovs_idl_map(char **keys, char **values, size_t n, + struct shash *); +static void shash_to_ovs_idl_map(struct shash *, + char ***keys, char ***values, size_t *n); + /* Hooks into ofproto processing. */ static struct ofhooks bridge_ofhooks; /* Public functions. */ -/* Adds the name of each interface used by a bridge, including local and - * internal ports, to 'svec'. */ -void -bridge_get_ifaces(struct svec *svec) -{ - struct bridge *br, *next; - size_t i, j; - - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - - 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); - } - } - } - } - } -} - -/* The caller must already have called cfg_read(). */ +/* 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_init(void) +bridge_init(const char *remote) { - struct svec dpif_names; - size_t i; + /* Create connection to database. */ + idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true); - unixctl_command_register("fdb/show", bridge_unixctl_fdb_show); + 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); - 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; + ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); - svec_init(&all_names); - dpif_get_all_names(dpif, &all_names); - for (j = 0; j < all_names.n; j++) { - if (cfg_has("bridge.%s.port", all_names.names[j])) { - goto found; - } - } - dpif_delete(dpif); - found: - svec_destroy(&all_names); - dpif_close(dpif); - } - } + ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids); + ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge); - unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport); + ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics); + ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids); + /* Register unixctl commands. */ + unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, + NULL); + unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect, + NULL); bond_init(); - bridge_reconfigure(); } -#ifdef HAVE_OPENSSL -static bool -config_string_change(const char *key, char **valuep) +void +bridge_exit(void) { - const char *value = cfg_get_string(0, "%s", key); - if (value && (!*valuep || strcmp(value, *valuep))) { - free(*valuep); - *valuep = xstrdup(value); - return true; - } else { - return false; + struct bridge *br, *next_br; + + LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) { + bridge_destroy(br); } + ovsdb_idl_destroy(idl); } +/* Performs configuration that is only necessary once at ovs-vswitchd startup, + * but for which the ovs-vswitchd configuration 'cfg' is required. */ static void -bridge_configure_ssl(void) +bridge_configure_once(const struct ovsrec_open_vswitch *cfg) { - /* 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; + static bool already_configured_once; + struct svec bridge_names; + struct svec dpif_names, dpif_types; + size_t i; - if (config_string_change("ssl.private-key", &private_key_file)) { - vconn_ssl_set_private_key_file(private_key_file); + /* Only do this once per ovs-vswitchd run. */ + if (already_configured_once) { + return; } + already_configured_once = true; - if (config_string_change("ssl.certificate", &certificate_file)) { - vconn_ssl_set_certificate_file(certificate_file); + stats_timer = time_msec() + STATS_INTERVAL; + + /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */ + svec_init(&bridge_names); + for (i = 0; i < cfg->n_bridges; i++) { + svec_add(&bridge_names, cfg->bridges[i]->name); } + svec_sort(&bridge_names); - /* 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, - cfg_get_bool(0, "ssl.bootstrap-ca-cert")); + /* Iterate over all system dpifs and delete any of them that do not appear + * in 'cfg'. */ + svec_init(&dpif_names); + svec_init(&dpif_types); + dp_enumerate_types(&dpif_types); + for (i = 0; i < dpif_types.n; i++) { + size_t j; + + dp_enumerate_names(dpif_types.names[i], &dpif_names); + + /* Delete each dpif whose name is not in 'bridge_names'. */ + for (j = 0; j < dpif_names.n; j++) { + if (!svec_contains(&bridge_names, dpif_names.names[j])) { + struct dpif *dpif; + int retval; + + retval = dpif_open(dpif_names.names[j], dpif_types.names[i], + &dpif); + if (!retval) { + dpif_delete(dpif); + dpif_close(dpif); + } + } + } } + svec_destroy(&bridge_names); + svec_destroy(&dpif_names); + svec_destroy(&dpif_types); } -#endif -/* 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. */ +/* Callback for iterate_and_prune_ifaces(). */ static bool -init_iface_netdev(struct bridge *br UNUSED, struct iface *iface, - void *aux UNUSED) -{ - if (iface->netdev) { - return true; - } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE, - &iface->netdev)) { - netdev_get_carrier(iface->netdev, &iface->enabled); - return true; - } else { - /* If the network device can't be opened, then we're not going to try - * to do anything with this interface. */ +check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED) +{ + if (!iface->netdev) { + /* We already reported a related error, don't bother duplicating it. */ return false; } -} -static bool -check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED) -{ - 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 { + if (iface->dp_ifidx < 0) { VLOG_ERR("%s interface not in %s, dropping", iface->name, dpif_name(br->dpif)); return false; } + + VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif), + iface->name, iface->dp_ifidx); + return true; } +/* Callback for iterate_and_prune_ifaces(). */ static bool -set_iface_policing(struct bridge *br UNUSED, struct iface *iface, - void *aux UNUSED) +set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, + void *aux OVS_UNUSED) { - int rate = cfg_get_int(0, "port.%s.ingress.policing-rate", iface->name); - int burst = cfg_get_int(0, "port.%s.ingress.policing-burst", iface->name); - netdev_set_policing(iface->netdev, rate, burst); + /* Set policing attributes. */ + netdev_set_policing(iface->netdev, + iface->cfg->ingress_policing_rate, + iface->cfg->ingress_policing_burst); + + /* Set MAC address of internal interfaces other than the local + * interface. */ + if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) { + iface_set_mac(iface); + } + return true; } @@ -427,6 +454,7 @@ iterate_and_prune_ifaces(struct bridge *br, if (cb(br, iface, aux)) { j++; } else { + iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } } @@ -440,49 +468,123 @@ iterate_and_prune_ifaces(struct bridge *br, } } -void -bridge_reconfigure(void) +/* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP + * addresses and ports into '*managersp' and '*n_managersp'. The caller is + * responsible for freeing '*managersp' (with free()). + * + * You may be asking yourself "why does ovs-vswitchd care?", because + * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd + * should not be and in fact is not directly involved in that. But + * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so + * it has to tell in-band control where the managers are to enable that. + * (Thus, only managers connected in-band are collected.) + */ +static void +collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg, + struct sockaddr_in **managersp, size_t *n_managersp) +{ + struct sockaddr_in *managers = NULL; + size_t n_managers = 0; + struct shash targets; + size_t i; + + /* Collect all of the potential targets, as the union of the "managers" + * column and the "targets" columns of the rows pointed to by + * "manager_options", excluding any that are out-of-band. */ + shash_init(&targets); + for (i = 0; i < ovs_cfg->n_managers; i++) { + shash_add_once(&targets, ovs_cfg->managers[i], NULL); + } + for (i = 0; i < ovs_cfg->n_manager_options; i++) { + struct ovsrec_manager *m = ovs_cfg->manager_options[i]; + + if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) { + shash_find_and_delete(&targets, m->target); + } else { + shash_add_once(&targets, m->target, NULL); + } + } + + /* Now extract the targets' IP addresses. */ + if (!shash_is_empty(&targets)) { + struct shash_node *node; + + managers = xmalloc(shash_count(&targets) * sizeof *managers); + SHASH_FOR_EACH (node, &targets) { + const char *target = node->name; + struct sockaddr_in *sin = &managers[n_managers]; + + 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++; + } + } + } + shash_destroy(&targets); + + *managersp = managers; + *n_managersp = n_managers; +} + +static void +bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) { - struct svec old_br, new_br; + struct shash old_br, new_br; + struct shash_node *node; struct bridge *br, *next; + struct sockaddr_in *managers; + size_t n_managers; size_t i; + int sflow_bridge_number; COVERAGE_INC(bridge_reconfigure); + collect_in_band_managers(ovs_cfg, &managers, &n_managers); + /* Collect old and new bridges. */ - svec_init(&old_br); - svec_init(&new_br); - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - svec_add(&old_br, br->name); + shash_init(&old_br); + shash_init(&new_br); + LIST_FOR_EACH (br, 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); + } } - cfg_get_subsections(&new_br, "bridge"); /* Get rid of deleted bridges and add new bridges. */ - svec_sort(&old_br); - svec_sort(&new_br); - assert(svec_is_unique(&old_br)); - assert(svec_is_unique(&new_br)); - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { - if (!svec_contains(&new_br, br->name)) { + LIST_FOR_EACH_SAFE (br, next, 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); } } - for (i = 0; i < new_br.n; i++) { - const char *name = new_br.names[i]; - if (!svec_contains(&old_br, name)) { - bridge_create(name); + SHASH_FOR_EACH (node, &new_br) { + const char *br_name = node->name; + const struct ovsrec_bridge *br_cfg = node->data; + br = shash_find_data(&old_br, br_name); + if (br) { + /* If the bridge datapath type has changed, we need to tear it + * down and recreate. */ + if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) { + bridge_destroy(br); + bridge_create(br_cfg); + } + } else { + bridge_create(br_cfg); } } - svec_destroy(&old_br); - svec_destroy(&new_br); - -#ifdef HAVE_OPENSSL - /* Configure SSL. */ - bridge_configure_ssl(); -#endif + shash_destroy(&old_br); + shash_destroy(&new_br); /* Reconfigure all bridges. */ - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { bridge_reconfigure_one(br); } @@ -491,88 +593,153 @@ bridge_reconfigure(void) * 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 svec want_ifaces; + LIST_FOR_EACH (br, node, &all_bridges) { + struct dpif_port_dump dump; + struct shash want_ifaces; + struct dpif_port dpif_port; - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); bridge_get_all_ifaces(br, &want_ifaces); - for (i = 0; i < n_dpif_ports; i++) { - const struct odp_port *p = &dpif_ports[i]; - if (!svec_contains(&want_ifaces, p->devname) - && strcmp(p->devname, br->name)) { - int retval = dpif_port_del(br->dpif, p->port); + DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) { + if (!shash_find(&want_ifaces, dpif_port.name) + && strcmp(dpif_port.name, br->name)) { + int retval = dpif_port_del(br->dpif, dpif_port.port_no); if (retval) { VLOG_ERR("failed to remove %s interface from %s: %s", - p->devname, dpif_name(br->dpif), + dpif_port.name, dpif_name(br->dpif), strerror(retval)); } } } - svec_destroy(&want_ifaces); - free(dpif_ports); + shash_destroy(&want_ifaces); } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; - struct svec cur_ifaces, want_ifaces, add_ifaces; + LIST_FOR_EACH (br, node, &all_bridges) { + struct shash cur_ifaces, want_ifaces; + struct dpif_port_dump dump; + struct dpif_port dpif_port; - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); - svec_init(&cur_ifaces); - for (i = 0; i < n_dpif_ports; i++) { - svec_add(&cur_ifaces, dpif_ports[i].devname); + /* Get the set of interfaces currently in this datapath. */ + shash_init(&cur_ifaces); + DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) { + struct dpif_port *port_info = xmalloc(sizeof *port_info); + dpif_port_clone(port_info, &dpif_port); + shash_add(&cur_ifaces, dpif_port.name, port_info); } - free(dpif_ports); - svec_sort_unique(&cur_ifaces); + + /* Get the set of interfaces we want on this datapath. */ bridge_get_all_ifaces(br, &want_ifaces); - svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL); - for (i = 0; i < add_ifaces.n; i++) { - const char *if_name = add_ifaces.names[i]; - bool internal; + hmap_clear(&br->ifaces); + SHASH_FOR_EACH (node, &want_ifaces) { + const char *if_name = node->name; + struct iface *iface = node->data; + struct dpif_port *dpif_port; + const char *type; int error; - /* It's an internal interface if it's marked that way, or if - * it's a bonded interface for which we're faking up a network - * device. */ - internal = cfg_get_bool(0, "iface.%s.internal", if_name); - if (cfg_get_bool(0, "bonding.%s.fake-iface", if_name)) { - struct port *port = port_lookup(br, if_name); - if (port && port->n_ifaces > 1) { - internal = true; + type = iface ? iface->type : "internal"; + dpif_port = shash_find_data(&cur_ifaces, if_name); + + /* If we have a port or a netdev already, and it's not the type we + * want, then delete the port (if any) and close the netdev (if + * any). */ + if ((dpif_port && strcmp(dpif_port->type, type)) + || (iface && iface->netdev + && strcmp(type, netdev_get_type(iface->netdev)))) { + if (dpif_port) { + error = ofproto_port_del(br->ofproto, dpif_port->port_no); + if (error) { + continue; + } + dpif_port = NULL; + } + if (iface) { + netdev_close(iface->netdev); + iface->netdev = NULL; } } - /* Add to datapath. */ - error = dpif_port_add(br->dpif, if_name, - internal ? ODP_PORT_INTERNAL : 0, NULL); - if (error == EFBIG) { - VLOG_ERR("ran out of valid port numbers on %s", - dpif_name(br->dpif)); - break; - } else if (error) { - VLOG_ERR("failed to add %s interface to %s: %s", - if_name, dpif_name(br->dpif), strerror(error)); + /* If the port doesn't exist or we don't have the netdev open, + * we need to do more work. */ + if (!dpif_port || (iface && !iface->netdev)) { + struct netdev_options options; + struct netdev *netdev; + struct shash args; + + /* First open the network device. */ + options.name = if_name; + options.type = type; + options.args = &args; + options.ethertype = NETDEV_ETH_TYPE_NONE; + + shash_init(&args); + if (iface) { + shash_from_ovs_idl_map(iface->cfg->key_options, + iface->cfg->value_options, + iface->cfg->n_options, &args); + } + error = netdev_open(&options, &netdev); + shash_destroy(&args); + + if (error) { + VLOG_WARN("could not open network device %s (%s)", + if_name, strerror(error)); + continue; + } + + /* Then add the port if we haven't already. */ + if (!dpif_port) { + error = dpif_port_add(br->dpif, netdev, NULL); + if (error) { + netdev_close(netdev); + if (error == EFBIG) { + VLOG_ERR("ran out of valid port numbers on %s", + dpif_name(br->dpif)); + break; + } else { + VLOG_ERR("failed to add %s interface to %s: %s", + if_name, dpif_name(br->dpif), + strerror(error)); + continue; + } + } + } + + /* Update 'iface'. */ + if (iface) { + iface->netdev = netdev; + iface->enabled = netdev_get_carrier(iface->netdev); + } + } else if (iface && 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); + netdev_set_config(iface->netdev, &args); + shash_destroy(&args); } } - svec_destroy(&cur_ifaces); - svec_destroy(&want_ifaces); - svec_destroy(&add_ifaces); + shash_destroy(&want_ifaces); + + SHASH_FOR_EACH (node, &cur_ifaces) { + struct dpif_port *port_info = node->data; + dpif_port_destroy(port_info); + free(port_info); + } + shash_destroy(&cur_ifaces); } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + sflow_bridge_number = 0; + LIST_FOR_EACH (br, node, &all_bridges) { uint8_t ea[8]; uint64_t dpid; struct iface *local_iface; struct iface *hw_addr_iface; - uint8_t engine_type, engine_id; - bool add_id_to_iface = false; - struct svec nf_hosts; + char *dpid_string; bridge_fetch_dp_ifaces(br); - iterate_and_prune_ifaces(br, init_iface_netdev, NULL); - iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL); + iterate_and_prune_ifaces(br, check_iface, NULL); /* Pick local port hardware address, datapath ID. */ bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); @@ -590,35 +757,102 @@ bridge_reconfigure(void) dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface); ofproto_set_datapath_id(br->ofproto, dpid); + dpid_string = xasprintf("%016"PRIx64, dpid); + ovsrec_bridge_set_datapath_id(br->cfg, dpid_string); + free(dpid_string); + /* Set NetFlow configuration on this bridge. */ - dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id); - if (cfg_has("netflow.%s.engine-type", br->name)) { - engine_type = cfg_get_int(0, "netflow.%s.engine-type", - br->name); - } - if (cfg_has("netflow.%s.engine-id", br->name)) { - engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name); - } - if (cfg_has("netflow.%s.add-id-to-iface", br->name)) { - add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface", - br->name); - } - if (add_id_to_iface && engine_id > 0x7f) { - VLOG_WARN("bridge %s: netflow port mangling may conflict with " - "another vswitch, choose an engine id less than 128", - br->name); - } - if (add_id_to_iface && br->n_ports > 0x1ff) { - VLOG_WARN("bridge %s: netflow port mangling will conflict with " - "another port when 512 or more ports are used", - br->name); + if (br->cfg->netflow) { + struct ovsrec_netflow *nf_cfg = br->cfg->netflow; + struct netflow_options opts; + + memset(&opts, 0, sizeof opts); + + dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id); + if (nf_cfg->engine_type) { + opts.engine_type = *nf_cfg->engine_type; + } + if (nf_cfg->engine_id) { + opts.engine_id = *nf_cfg->engine_id; + } + + opts.active_timeout = nf_cfg->active_timeout; + if (!opts.active_timeout) { + opts.active_timeout = -1; + } else if (opts.active_timeout < 0) { + VLOG_WARN("bridge %s: active timeout interval set to negative " + "value, using default instead (%d seconds)", br->name, + NF_ACTIVE_TIMEOUT_DEFAULT); + opts.active_timeout = -1; + } + + opts.add_id_to_iface = nf_cfg->add_id_to_interface; + if (opts.add_id_to_iface) { + if (opts.engine_id > 0x7f) { + VLOG_WARN("bridge %s: netflow port mangling may conflict " + "with another vswitch, choose an engine id less " + "than 128", br->name); + } + if (br->n_ports > 508) { + VLOG_WARN("bridge %s: netflow port mangling will conflict " + "with another port when more than 508 ports are " + "used", br->name); + } + } + + opts.collectors.n = nf_cfg->n_targets; + opts.collectors.names = nf_cfg->targets; + if (ofproto_set_netflow(br->ofproto, &opts)) { + VLOG_ERR("bridge %s: problem setting netflow collectors", + br->name); + } + } else { + ofproto_set_netflow(br->ofproto, NULL); } - svec_init(&nf_hosts); - cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name); - if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type, - engine_id, add_id_to_iface)) { - VLOG_ERR("bridge %s: problem setting netflow collectors", - br->name); + + /* Set sFlow configuration on this bridge. */ + if (br->cfg->sflow) { + const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow; + struct ovsrec_controller **controllers; + struct ofproto_sflow_options oso; + size_t n_controllers; + + memset(&oso, 0, sizeof oso); + + oso.targets.n = sflow_cfg->n_targets; + oso.targets.names = sflow_cfg->targets; + + oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE; + if (sflow_cfg->sampling) { + oso.sampling_rate = *sflow_cfg->sampling; + } + + oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL; + if (sflow_cfg->polling) { + oso.polling_interval = *sflow_cfg->polling; + } + + oso.header_len = SFL_DEFAULT_HEADER_SIZE; + if (sflow_cfg->header) { + oso.header_len = *sflow_cfg->header; + } + + oso.sub_id = sflow_bridge_number++; + oso.agent_device = sflow_cfg->agent; + + oso.control_ip = NULL; + n_controllers = bridge_get_controllers(br, &controllers); + for (i = 0; i < n_controllers; i++) { + if (controllers[i]->local_ip) { + oso.control_ip = controllers[i]->local_ip; + break; + } + } + ofproto_set_sflow(br->ofproto, &oso); + + /* Do not destroy oso.targets because it is owned by sflow_cfg. */ + } else { + ofproto_set_sflow(br->ofproto, NULL); } /* Update the controller and related settings. It would be more @@ -630,35 +864,70 @@ bridge_reconfigure(void) * yet; when a controller is configured, resetting the datapath ID will * immediately disconnect from the controller, so it's better to set * the datapath ID before the controller. */ - bridge_reconfigure_controller(br); + bridge_reconfigure_remotes(br, managers, n_managers); } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; + int j; + port_update_vlan_compat(port); port_update_bonding(port); + + for (j = 0; j < port->n_ifaces; j++) { + iface_update_qos(port->ifaces[j], port->cfg->qos); + } } } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - brstp_reconfigure(br); - iterate_and_prune_ifaces(br, set_iface_policing, NULL); + LIST_FOR_EACH (br, node, &all_bridges) { + iterate_and_prune_ifaces(br, set_iface_properties, NULL); + } + + LIST_FOR_EACH (br, node, &all_bridges) { + struct iface *iface; + HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { + iface_update_cfm(iface); + } } + + free(managers); +} + +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); } static void bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], struct iface **hw_addr_iface) { - uint64_t requested_ea; + const char *hwaddr; size_t i, j; int error; *hw_addr_iface = NULL; /* Did the user request a particular MAC? */ - requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name); - if (requested_ea) { - eth_addr_from_uint64(requested_ea, ea); + hwaddr = bridge_get_other_config(br->cfg, "hwaddr"); + if (hwaddr && eth_addr_from_string(hwaddr, ea)) { if (eth_addr_is_multicast(ea)) { VLOG_ERR("bridge %s: cannot set MAC address to multicast " "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); @@ -669,14 +938,12 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], } } - /* Otherwise choose the minimum MAC address among all of the interfaces. - * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the - * MAC of the physical interface in such an environment.) */ + /* Otherwise choose the minimum non-local MAC address among all of the + * interfaces. */ memset(ea, 0xff, sizeof ea); for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; uint8_t iface_ea[ETH_ADDR_LEN]; - uint64_t iface_ea_u64; struct iface *iface; /* Mirror output ports don't participate. */ @@ -685,11 +952,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], } /* Choose the MAC address to represent the port. */ - iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name); - if (iface_ea_u64) { - /* User specified explicitly. */ - eth_addr_from_uint64(iface_ea_u64, iface_ea); - + if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) { /* Find the interface with this Ethernet address (if any) so that * we can provide the correct devname to the caller. */ iface = NULL; @@ -717,11 +980,8 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], } /* The local port doesn't count (since we're trying to choose its - * MAC address anyway). Other internal ports don't count because - * we really want a physical MAC if we can get it, and internal - * ports typically have randomly generated MACs. */ - if (iface->dp_ifidx == ODPP_LOCAL - || cfg_get_bool(0, "iface.%s.internal", iface->name)) { + * MAC address anyway). */ + if (iface->dp_ifidx == ODPP_LOCAL) { continue; } @@ -737,15 +997,16 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], /* Compare against our current choice. */ if (!eth_addr_is_multicast(iface_ea) && + !eth_addr_is_local(iface_ea) && !eth_addr_is_reserved(iface_ea) && !eth_addr_is_zero(iface_ea) && - memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) + eth_addr_compare_3way(iface_ea, ea) < 0) { memcpy(ea, iface_ea, ETH_ADDR_LEN); *hw_addr_iface = iface; } } - if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) { + if (eth_addr_is_multicast(ea)) { memcpy(ea, br->default_ea, ETH_ADDR_LEN); *hw_addr_iface = NULL; VLOG_WARN("bridge %s: using default bridge Ethernet " @@ -778,10 +1039,11 @@ bridge_pick_datapath_id(struct bridge *br, * stable from one run to the next, so that policy set on a datapath * "sticks". */ + const char *datapath_id; uint64_t dpid; - dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name); - if (dpid) { + datapath_id = bridge_get_other_config(br->cfg, "datapath-id"); + if (datapath_id && dpid_from_string(datapath_id, &dpid)) { return dpid; } @@ -846,152 +1108,463 @@ 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; - } - } - } - return retval; -} + enum netdev_flags flags; + uint32_t current; + int64_t bps; + int mtu; + int64_t mtu_64; + int error; -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; - if (br->ml) { - mac_learning_wait(br->ml); - } - bond_wait(br); - brstp_wait(br); - } -} + shash_to_ovs_idl_map(&sh, &keys, &values, &n); + ovsrec_interface_set_status(iface->cfg, keys, values, n); -/* 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; - if (br->ml) { - mac_learning_flush(br->ml); + free(keys); + free(values); + } else { + ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0); } -} -/* 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; + shash_destroy_free_data(&sh); - 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_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); } - return NULL; + 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); + } + + + ovsrec_interface_set_link_state(iface->cfg, + netdev_get_carrier(iface->netdev) + ? "up" : "down"); + + error = netdev_get_mtu(iface->netdev, &mtu); + if (!error) { + mtu_64 = mtu; + ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1); + } + else { + ovsrec_interface_set_mtu(iface->cfg, NULL, 0); + } } - -/* Bridge unixctl user interface functions. */ + static void -bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args) +iface_refresh_cfm_stats(struct iface *iface) { - struct ds ds = DS_EMPTY_INITIALIZER; - const struct bridge *br; + size_t i; + struct cfm *cfm; + const struct ovsrec_monitor *mon; - br = bridge_lookup(args); - if (!br) { - unixctl_command_reply(conn, 501, "no such bridge"); + mon = iface->cfg->monitor; + cfm = iface->cfm; + + if (!cfm || !mon) { return; } - ds_put_cstr(&ds, " port VLAN MAC Age\n"); - if (br->ml) { - const struct mac_entry *e; - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { - if (e->port < 0 || e->port >= br->n_ports) { - continue; - } - ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", - br->ports[e->port]->ifaces[0]->dp_ifidx, - e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); - } + for (i = 0; i < mon->n_remote_mps; i++) { + const struct ovsrec_maintenance_point *mp; + const struct remote_mp *rmp; + + mp = mon->remote_mps[i]; + rmp = cfm_get_remote_mp(cfm, mp->mpid); + + ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1); } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); - ds_destroy(&ds); -} - -/* Bridge reconfiguration functions. */ -static struct bridge * -bridge_create(const char *name) -{ - struct bridge *br; - int error; + if (hmap_is_empty(&cfm->x_remote_mps)) { + ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0); + } else { + size_t length; + struct remote_mp *rmp; + int64_t *x_remote_mps; - assert(!bridge_lookup(name)); - br = xcalloc(1, sizeof *br); + length = hmap_count(&cfm->x_remote_mps); + x_remote_mps = xzalloc(length * sizeof *x_remote_mps); - error = dpif_create(name, &br->dpif); - if (error == EEXIST || error == EBUSY) { - error = dpif_open(name, &br->dpif); - if (error) { - VLOG_ERR("datapath %s already exists but cannot be opened: %s", - name, strerror(error)); - free(br); - return NULL; - } - dpif_flow_flush(br->dpif); - } else if (error) { - VLOG_ERR("failed to create datapath %s: %s", name, strerror(error)); - free(br); - return NULL; + i = 0; + HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) { + x_remote_mps[i++] = rmp->mpid; + } + + ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length); + free(x_remote_mps); } - error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto); - if (error) { - VLOG_ERR("failed to create switch %s: %s", name, strerror(error)); + if (hmap_is_empty(&cfm->x_remote_maids)) { + ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0); + } else { + size_t length; + char **x_remote_maids; + struct remote_maid *rmaid; + + length = hmap_count(&cfm->x_remote_maids); + x_remote_maids = xzalloc(length * sizeof *x_remote_maids); + + i = 0; + HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) { + size_t j; + + x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1); + + for (j = 0; j < CCM_MAID_LEN; j++) { + snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx", + rmaid->maid[j]); + } + i++; + } + ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length); + + for (i = 0; i < length; i++) { + free(x_remote_maids[i]); + } + free(x_remote_maids); + } + + ovsrec_monitor_set_fault(mon, &cfm->fault, 1); +} + +static void +iface_refresh_stats(struct iface *iface) +{ + struct iface_stat { + char *name; + int offset; + }; + static const struct iface_stat iface_stats[] = { + { "rx_packets", offsetof(struct netdev_stats, rx_packets) }, + { "tx_packets", offsetof(struct netdev_stats, tx_packets) }, + { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) }, + { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) }, + { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) }, + { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) }, + { "rx_errors", offsetof(struct netdev_stats, rx_errors) }, + { "tx_errors", offsetof(struct netdev_stats, tx_errors) }, + { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) }, + { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) }, + { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) }, + { "collisions", offsetof(struct netdev_stats, collisions) }, + }; + enum { N_STATS = ARRAY_SIZE(iface_stats) }; + const struct iface_stat *s; + + char *keys[N_STATS]; + int64_t values[N_STATS]; + int n; + + struct netdev_stats stats; + + /* Intentionally ignore return value, since errors will set 'stats' to + * all-1s, and we will deal with that correctly below. */ + netdev_get_stats(iface->netdev, &stats); + + n = 0; + for (s = iface_stats; s < &iface_stats[N_STATS]; s++) { + uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset); + if (value != UINT64_MAX) { + keys[n] = s->name; + values[n] = value; + n++; + } + } + + ovsrec_interface_set_statistics(iface->cfg, keys, values, n); +} + +static void +refresh_system_stats(const struct ovsrec_open_vswitch *cfg) +{ + struct ovsdb_datum datum; + struct shash stats; + + shash_init(&stats); + get_system_stats(&stats); + + ovsdb_datum_from_shash(&datum, &stats); + ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics, + &datum); +} + +static inline const char * +nx_role_to_str(enum nx_role role) +{ + switch (role) { + case NX_ROLE_OTHER: + return "other"; + case NX_ROLE_MASTER: + return "master"; + case NX_ROLE_SLAVE: + return "slave"; + default: + return "*** INVALID ROLE ***"; + } +} + +static void +bridge_refresh_controller_status(const struct bridge *br) +{ + struct shash info; + const struct ovsrec_controller *cfg; + + ofproto_get_ofproto_controller_info(br->ofproto, &info); + + OVSREC_CONTROLLER_FOR_EACH(cfg, idl) { + struct ofproto_controller_info *cinfo = + shash_find_data(&info, cfg->target); + + 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); + } + } + + ofproto_free_ofproto_controller_info(&info); +} + +void +bridge_run(void) +{ + const struct ovsrec_open_vswitch *cfg; + + bool datapath_destroyed; + bool database_changed; + struct bridge *br; + + /* Let each bridge do the work that it needs to do. */ + datapath_destroyed = false; + LIST_FOR_EACH (br, 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); + datapath_destroyed = true; + } + } + + /* (Re)configure if necessary. */ + database_changed = ovsdb_idl_run(idl); + cfg = ovsrec_open_vswitch_first(idl); +#ifdef HAVE_OPENSSL + /* 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. + * + * 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; + + stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate); + stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert); + } +#endif + if (database_changed || datapath_destroyed) { + if (cfg) { + struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl); + + bridge_configure_once(cfg); + bridge_reconfigure(cfg); + + ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg); + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); /* XXX */ + } else { + /* We still need to reconfigure to avoid dangling pointers to + * now-destroyed ovsrec structures inside bridge data. */ + static const struct ovsrec_open_vswitch null_cfg; + + bridge_reconfigure(&null_cfg); + } + } + + /* Refresh system and interface stats if necessary. */ + if (time_msec() >= stats_timer) { + if (cfg) { + struct ovsdb_idl_txn *txn; + + txn = ovsdb_idl_txn_create(idl); + LIST_FOR_EACH (br, node, &all_bridges) { + size_t i; + + for (i = 0; i < br->n_ports; i++) { + struct port *port = br->ports[i]; + size_t j; + + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + iface_refresh_stats(iface); + iface_refresh_cfm_stats(iface); + iface_refresh_status(iface); + } + } + bridge_refresh_controller_status(br); + } + refresh_system_stats(cfg); + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); /* XXX */ + } + + stats_timer = time_msec() + STATS_INTERVAL; + } +} + +void +bridge_wait(void) +{ + struct bridge *br; + struct iface *iface; + + LIST_FOR_EACH (br, node, &all_bridges) { + ofproto_wait(br->ofproto); + if (ofproto_has_primary_controller(br->ofproto)) { + continue; + } + + mac_learning_wait(br->ml); + bond_wait(br); + + HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { + if (iface->cfm) { + cfm_wait(iface->cfm); + } + } + } + ovsdb_idl_wait(idl); + poll_timer_wait_until(stats_timer); +} + +/* 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); +} + +/* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no + * such interface. */ +static struct iface * +bridge_get_local_iface(struct bridge *br) +{ + size_t i, j; + + for (i = 0; i < br->n_ports; i++) { + struct port *port = br->ports[i]; + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + if (iface->dp_ifidx == ODPP_LOCAL) { + return iface; + } + } + } + + return NULL; +} + +/* Bridge unixctl user interface functions. */ +static void +bridge_unixctl_fdb_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + const struct bridge *br; + const struct mac_entry *e; + + br = bridge_lookup(args); + if (!br) { + unixctl_command_reply(conn, 501, "no such bridge"); + return; + } + + ds_put_cstr(&ds, " port VLAN MAC Age\n"); + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { + if (e->port < 0 || e->port >= br->n_ports) { + continue; + } + ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", + br->ports[e->port]->ifaces[0]->dp_ifidx, + e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); + } + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + ds_destroy(&ds); +} + +/* Bridge reconfiguration functions. */ +static struct bridge * +bridge_create(const struct ovsrec_bridge *br_cfg) +{ + struct bridge *br; + int error; + + assert(!bridge_lookup(br_cfg->name)); + br = xzalloc(sizeof *br); + + error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type, + &br->dpif); + if (error) { + free(br); + return NULL; + } + dpif_flow_flush(br->dpif); + + error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks, + br, &br->ofproto); + if (error) { + VLOG_ERR("failed to create switch %s: %s", br_cfg->name, + strerror(error)); dpif_delete(br->dpif); dpif_close(br->dpif); free(br); return NULL; } - br->name = xstrdup(name); + br->name = xstrdup(br_cfg->name); + br->cfg = br_cfg; br->ml = mac_learning_create(); - br->sent_config_request = false; - eth_addr_random(br->default_ea); + eth_addr_nicira_random(br->default_ea); - port_array_init(&br->ifaces); + hmap_init(&br->ifaces); + + shash_init(&br->port_by_name); + shash_init(&br->iface_by_name); br->flush = false; - br->bond_next_rebalance = time_msec() + 10000; list_push_back(&all_bridges, &br->node); @@ -1017,9 +1590,10 @@ bridge_destroy(struct bridge *br) } dpif_close(br->dpif); ofproto_destroy(br->ofproto); - free(br->controller); mac_learning_destroy(br->ml); - port_array_destroy(&br->ifaces); + hmap_destroy(&br->ifaces); + shash_destroy(&br->port_by_name); + shash_destroy(&br->iface_by_name); free(br->ports); free(br->name); free(br); @@ -1031,7 +1605,7 @@ bridge_lookup(const char *name) { struct bridge *br; - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { if (!strcmp(br->name, name)) { return br; } @@ -1039,27 +1613,15 @@ bridge_lookup(const char *name) return NULL; } -bool -bridge_exists(const char *name) -{ - return bridge_lookup(name) ? true : false; -} - -uint64_t -bridge_get_datapathid(const char *name) -{ - struct bridge *br = bridge_lookup(name); - return br ? ofproto_get_datapath_id(br->ofproto) : 0; -} - /* Handle requests for a listing of all flows known by the OpenFlow * stack, including those normally hidden. */ static void -bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args) +bridge_unixctl_dump_flows(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) { struct bridge *br; struct ds results; - + br = bridge_lookup(args); if (!br) { unixctl_command_reply(conn, 501, "Unknown bridge"); @@ -1073,163 +1635,181 @@ bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args) ds_destroy(&results); } +/* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller + * connections and reconnect. If BRIDGE is not specified, then all bridges + * drop their controller connections and reconnect. */ +static void +bridge_unixctl_reconnect(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct bridge *br; + if (args[0] != '\0') { + br = bridge_lookup(args); + if (!br) { + unixctl_command_reply(conn, 501, "Unknown bridge"); + return; + } + ofproto_reconnect_controllers(br->ofproto); + } else { + LIST_FOR_EACH (br, node, &all_bridges) { + ofproto_reconnect_controllers(br->ofproto); + } + } + unixctl_command_reply(conn, 200, NULL); +} + static int bridge_run_one(struct bridge *br) { int error; + struct iface *iface; error = ofproto_run1(br->ofproto); if (error) { return error; } - if (br->ml) { - mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); - } + mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); bond_run(br); - brstp_run(br); error = ofproto_run2(br->ofproto, br->flush); br->flush = false; + HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { + struct ofpbuf *packet; + + if (!iface->cfm) { + continue; + } + + packet = cfm_run(iface->cfm); + if (packet) { + iface_send_packet(iface, packet); + ofpbuf_uninit(packet); + free(packet); + } + } + return error; } -static const char * -bridge_get_controller(const struct bridge *br) +static size_t +bridge_get_controllers(const struct bridge *br, + struct ovsrec_controller ***controllersp) { - const char *controller; + struct ovsrec_controller **controllers; + size_t n_controllers; + + controllers = br->cfg->controller; + n_controllers = br->cfg->n_controller; - controller = cfg_get_string(0, "bridge.%s.controller", br->name); - if (!controller) { - controller = cfg_get_string(0, "mgmt.controller"); + if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) { + controllers = NULL; + n_controllers = 0; } - return controller && controller[0] ? controller : NULL; -} -static bool -check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_) -{ - struct svec *ifaces = ifaces_; - if (!svec_contains(ifaces, iface->name)) { - svec_add(ifaces, iface->name); - svec_sort(ifaces); - return true; - } else { - VLOG_ERR("bridge %s: %s interface is on multiple ports, " - "removing from %s", - br->name, iface->name, iface->port->name); - return false; + if (controllersp) { + *controllersp = controllers; } + return n_controllers; } static void bridge_reconfigure_one(struct bridge *br) { - struct svec old_ports, new_ports, ifaces; - struct svec listeners, old_listeners; + struct shash old_ports, new_ports; struct svec snoops, old_snoops; + struct shash_node *node; + enum ofproto_fail_mode fail_mode; size_t i; /* Collect old ports. */ - svec_init(&old_ports); + shash_init(&old_ports); for (i = 0; i < br->n_ports; i++) { - svec_add(&old_ports, br->ports[i]->name); + shash_add(&old_ports, br->ports[i]->name, br->ports[i]); } - svec_sort(&old_ports); - assert(svec_is_unique(&old_ports)); /* Collect new ports. */ - svec_init(&new_ports); - cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name); - svec_sort(&new_ports); - if (bridge_get_controller(br)) { + shash_init(&new_ports); + for (i = 0; i < br->cfg->n_ports; i++) { + const char *name = br->cfg->ports[i]->name; + if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) { + VLOG_WARN("bridge %s: %s specified twice as bridge port", + br->name, name); + } + } + + /* If we have a controller, then we need a local port. Complain if the + * user didn't specify one. + * + * XXX perhaps we should synthesize a port ourselves in this case. */ + if (bridge_get_controllers(br, NULL)) { char local_name[IF_NAMESIZE]; int error; error = dpif_port_get_name(br->dpif, ODPP_LOCAL, local_name, sizeof local_name); - if (!error && !svec_contains(&new_ports, local_name)) { - svec_add(&new_ports, local_name); - svec_sort(&new_ports); + if (!error && !shash_find(&new_ports, local_name)) { + VLOG_WARN("bridge %s: controller specified but no local port " + "(port named %s) defined", + br->name, local_name); } } - if (!svec_is_unique(&new_ports)) { - VLOG_WARN("bridge %s: %s specified twice as bridge port", - br->name, svec_get_duplicate(&new_ports)); - svec_unique(&new_ports); - } - ofproto_set_mgmt_id(br->ofproto, mgmt_id); + /* Get rid of deleted ports. + * Get rid of deleted interfaces on ports that still exist. */ + SHASH_FOR_EACH (node, &old_ports) { + struct port *port = node->data; + const struct ovsrec_port *port_cfg; - /* Get rid of deleted ports and add new ports. */ - for (i = 0; i < br->n_ports; ) { - struct port *port = br->ports[i]; - if (!svec_contains(&new_ports, port->name)) { + port_cfg = shash_find_data(&new_ports, node->name); + if (!port_cfg) { port_destroy(port); } else { - i++; + port_del_ifaces(port, port_cfg); } } - for (i = 0; i < new_ports.n; i++) { - const char *name = new_ports.names[i]; - if (!svec_contains(&old_ports, name)) { - port_create(br, name); + + /* Create new ports. + * Add new interfaces to existing ports. + * Reconfigure existing ports. */ + SHASH_FOR_EACH (node, &new_ports) { + struct port *port = shash_find_data(&old_ports, node->name); + if (!port) { + port = port_create(br, node->name); } - } - svec_destroy(&old_ports); - svec_destroy(&new_ports); - /* Reconfigure all ports. */ - for (i = 0; i < br->n_ports; i++) { - port_reconfigure(br->ports[i]); + port_reconfigure(port, node->data); + if (!port->n_ifaces) { + VLOG_WARN("bridge %s: port %s has no interfaces, dropping", + br->name, port->name); + port_destroy(port); + } } + shash_destroy(&old_ports); + shash_destroy(&new_ports); - /* Check and delete duplicate interfaces. */ - svec_init(&ifaces); - iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces); - svec_destroy(&ifaces); + /* Set the fail-mode */ + fail_mode = !br->cfg->fail_mode + || !strcmp(br->cfg->fail_mode, "standalone") + ? OFPROTO_FAIL_STANDALONE + : OFPROTO_FAIL_SECURE; + if (ofproto_get_fail_mode(br->ofproto) != fail_mode + && !ofproto_has_primary_controller(br->ofproto)) { + ofproto_flush_flows(br->ofproto); + } + ofproto_set_fail_mode(br->ofproto, fail_mode); /* Delete all flows if we're switching from connected to standalone or vice * versa. (XXX Should we delete all flows if we are switching from one * controller to another?) */ - /* Configure OpenFlow management listeners. */ - svec_init(&listeners); - cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name); - if (!listeners.n) { - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) { - svec_clear(&listeners); - } - svec_sort_unique(&listeners); - - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - svec_sort_unique(&old_listeners); - - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); - } - svec_destroy(&listeners); - svec_destroy(&old_listeners); - /* Configure OpenFlow controller connection snooping. */ svec_init(&snoops); - cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name); - if (!snoops.n) { - svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); - } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) { - svec_clear(&snoops); - } - svec_sort_unique(&snoops); - + svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", + ovs_rundir(), br->name)); 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); } @@ -1239,172 +1819,197 @@ bridge_reconfigure_one(struct bridge *br) mirror_reconfigure(br); } +/* Initializes 'oc' appropriately as a management service controller for + * 'br'. + * + * The caller must free oc->target when it is no longer needed. */ static void -bridge_reconfigure_controller(struct bridge *br) +bridge_ofproto_controller_for_mgmt(const struct bridge *br, + struct ofproto_controller *oc) { - char *pfx = xasprintf("bridge.%s.controller", br->name); - const char *controller; + 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->accept_re = NULL; + oc->update_resolv_conf = false; + oc->rate_limit = 0; + oc->burst_limit = 0; +} - controller = bridge_get_controller(br); - if ((br->controller != NULL) != (controller != NULL)) { - ofproto_flush_flows(br->ofproto); +/* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */ +static void +bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c, + struct ofproto_controller *oc) +{ + 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->accept_re = c->discover_accept_regex; + oc->update_resolv_conf = c->discover_update_resolv_conf; + oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; + oc->burst_limit = (c->controller_burst_limit + ? *c->controller_burst_limit : 0); +} + +/* Configures the IP stack for 'br''s local interface properly according to the + * configuration in 'c'. */ +static void +bridge_configure_local_iface_netdev(struct bridge *br, + struct ovsrec_controller *c) +{ + struct netdev *netdev; + struct in_addr mask, gateway; + + struct iface *local_iface; + struct in_addr ip; + + /* Controller discovery does its own TCP/IP configuration later. */ + if (strcmp(c->target, "discover")) { + return; } - free(br->controller); - br->controller = controller ? xstrdup(controller) : NULL; - if (controller) { - const char *fail_mode; - int max_backoff, probe; - int rate_limit, burst_limit; + /* If there's no local interface or no IP address, give up. */ + local_iface = bridge_get_local_iface(br); + if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) { + return; + } - if (!strcmp(controller, "discover")) { - bool update_resolv_conf = true; + /* Bring up the local interface. */ + netdev = local_iface->netdev; + netdev_turn_flags_on(netdev, NETDEV_UP, true); - if (cfg_has("%s.update-resolv.conf", pfx)) { - update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf", - pfx); - } - ofproto_set_discovery(br->ofproto, true, - cfg_get_string(0, "%s.accept-regex", pfx), - update_resolv_conf); - } else { - struct iface *local_iface; - bool in_band; - - in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED, - "%s.in-band", pfx) - || cfg_get_bool(0, "%s.in-band", pfx)); - ofproto_set_discovery(br->ofproto, false, NULL, NULL); - ofproto_set_in_band(br->ofproto, in_band); - - local_iface = bridge_get_local_iface(br); - if (local_iface - && cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) { - struct netdev *netdev = local_iface->netdev; - struct in_addr ip, mask, gateway; - ip.s_addr = cfg_get_ip(0, "%s.ip", pfx); - mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx); - gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx); - - 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)); - } + /* 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); + } + if (!netdev_set_in4(netdev, ip, mask)) { + VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT, + br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr)); + } - if (gateway.s_addr) { - if (!netdev_add_router(netdev, gateway)) { - VLOG_INFO("bridge %s: configured gateway "IP_FMT, - br->name, IP_ARGS(&gateway.s_addr)); - } - } - } + /* 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)); } + } +} - fail_mode = cfg_get_string(0, "%s.fail-mode", pfx); - if (!fail_mode) { - fail_mode = cfg_get_string(0, "mgmt.fail-mode"); - } - ofproto_set_failure(br->ofproto, - (!fail_mode - || !strcmp(fail_mode, "standalone") - || !strcmp(fail_mode, "open"))); +static void +bridge_reconfigure_remotes(struct bridge *br, + const struct sockaddr_in *managers, + size_t n_managers) +{ + const char *disable_ib_str, *queue_id_str; + bool disable_in_band = false; + int queue_id; - probe = cfg_get_int(0, "%s.inactivity-probe", pfx); - if (probe < 5) { - probe = cfg_get_int(0, "mgmt.inactivity-probe"); - if (probe < 5) { - probe = 5; - } - } - ofproto_set_probe_interval(br->ofproto, probe); + struct ovsrec_controller **controllers; + size_t n_controllers; + bool had_primary; - max_backoff = cfg_get_int(0, "%s.max-backoff", pfx); - if (!max_backoff) { - max_backoff = cfg_get_int(0, "mgmt.max-backoff"); - if (!max_backoff) { - max_backoff = 8; - } - } - ofproto_set_max_backoff(br->ofproto, max_backoff); + struct ofproto_controller *ocs; + size_t n_ocs; + size_t i; - rate_limit = cfg_get_int(0, "%s.rate-limit", pfx); - if (!rate_limit) { - rate_limit = cfg_get_int(0, "mgmt.rate-limit"); - } - burst_limit = cfg_get_int(0, "%s.burst-limit", pfx); - if (!burst_limit) { - burst_limit = cfg_get_int(0, "mgmt.burst-limit"); - } - ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit); + /* 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; + } - ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx)); + /* 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 (cfg_has("%s.commands.acl", pfx)) { - struct svec command_acls; - char *command_acl; + if (disable_in_band) { + ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0); + } else { + ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); + } + had_primary = ofproto_has_primary_controller(br->ofproto); - svec_init(&command_acls); - cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx); - command_acl = svec_join(&command_acls, ",", ""); + n_controllers = bridge_get_controllers(br, &controllers); - ofproto_set_remote_execution(br->ofproto, command_acl, - cfg_get_string(0, "%s.commands.dir", - pfx)); + ocs = xmalloc((n_controllers + 1) * sizeof *ocs); + n_ocs = 0; - svec_destroy(&command_acls); - free(command_acl); - } else { - ofproto_set_remote_execution(br->ofproto, NULL, NULL); + bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]); + for (i = 0; i < n_controllers; i++) { + struct ovsrec_controller *c = controllers[i]; + + if (!strncmp(c->target, "punix:", 6) + || !strncmp(c->target, "unix:", 5)) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + + /* Prevent remote ovsdb-server users from accessing arbitrary Unix + * domain sockets and overwriting arbitrary local files. */ + VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller " + "\"%s\" due to possibility for remote exploit", + dpif_name(br->dpif), c->target); + continue; } - } else { + + 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++; + } + + ofproto_set_controllers(br->ofproto, ocs, n_ocs); + free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */ + free(ocs); + + if (had_primary != ofproto_has_primary_controller(br->ofproto)) { + ofproto_flush_flows(br->ofproto); + } + + /* If there are no controllers and the bridge is in standalone + * mode, set up a flow that matches every packet and directs + * them to OFPP_NORMAL (which goes to us). Otherwise, the + * switch is in secure mode and we won't pass any traffic until + * a controller has been defined and it tells us to do so. */ + if (!n_controllers + && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) { union ofp_action action; - flow_t flow; + struct cls_rule rule; - /* Set up a flow that matches every packet and directs them to - * OFPP_NORMAL (which goes to us). */ memset(&action, 0, sizeof action); action.type = htons(OFPAT_OUTPUT); action.output.len = htons(sizeof action); action.output.port = htons(OFPP_NORMAL); - memset(&flow, 0, sizeof flow); - ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0, - &action, 1, 0); - - ofproto_set_in_band(br->ofproto, false); - ofproto_set_max_backoff(br->ofproto, 1); - ofproto_set_probe_interval(br->ofproto, 5); - ofproto_set_failure(br->ofproto, false); - ofproto_set_stp(br->ofproto, false); + cls_rule_init_catchall(&rule, 0); + ofproto_add_flow(br->ofproto, &rule, &action, 1); } - free(pfx); - - ofproto_set_controller(br->ofproto, br->controller); } static void -bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces) +bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) { size_t i, j; - svec_init(ifaces); + shash_init(ifaces); for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; - svec_add(ifaces, iface->name); + shash_add_once(ifaces, iface->name, iface); } - if (port->n_ifaces > 1 - && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) { - svec_add(ifaces, port->name); + if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { + shash_add_once(ifaces, port->name, NULL); } } - svec_sort_unique(ifaces); } /* For robustness, in case the administrator moves around datapath ports behind @@ -1418,8 +2023,8 @@ bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces) static void bridge_fetch_dp_ifaces(struct bridge *br) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; + struct dpif_port_dump dump; + struct dpif_port dpif_port; size_t i, j; /* Reset all interface numbers. */ @@ -1430,65 +2035,94 @@ bridge_fetch_dp_ifaces(struct bridge *br) iface->dp_ifidx = -1; } } - port_array_clear(&br->ifaces); + hmap_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); + DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) { + struct iface *iface = iface_lookup(br, dpif_port.name); 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)) { + dpif_name(br->dpif), dpif_port.name); + } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) { VLOG_WARN("%s reported interface %"PRIu16" twice", - dpif_name(br->dpif), p->port); + dpif_name(br->dpif), dpif_port.port_no); } else { - port_array_set(&br->ifaces, p->port, iface); - iface->dp_ifidx = p->port; + iface->dp_ifidx = dpif_port.port_no; + hmap_insert(&br->ifaces, &iface->dp_ifidx_node, + hash_int(iface->dp_ifidx, 0)); } + + iface_set_ofport(iface->cfg, + (iface->dp_ifidx >= 0 + ? odp_port_to_ofp_port(iface->dp_ifidx) + : -1)); } } - free(dpif_ports); } /* Bridge packet processing functions. */ static int -bond_hash(const uint8_t mac[ETH_ADDR_LEN]) +bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan) { - return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK; + return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK; } static struct bond_entry * -lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN]) +lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN], + uint16_t vlan) { - return &port->bond_hash[bond_hash(mac)]; + assert(port->bond_mode == BM_SLB); + return &port->bond_hash[bond_hash(mac, vlan)]; } static int bond_choose_iface(const struct port *port) { - size_t i; + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); + size_t i, best_down_slave = -1; + long long next_delay_expiration = LLONG_MAX; + for (i = 0; i < port->n_ifaces; i++) { - if (port->ifaces[i]->enabled) { + struct iface *iface = port->ifaces[i]; + + if (iface->enabled) { return i; + } else if (iface->delay_expires < next_delay_expiration) { + best_down_slave = i; + next_delay_expiration = iface->delay_expires; } } - return -1; + + if (best_down_slave != -1) { + struct iface *iface = port->ifaces[best_down_slave]; + + VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay " + "since no other interface is up", iface->name, + iface->delay_expires - time_msec()); + bond_enable_slave(iface, true); + } + + return best_down_slave; } static bool choose_output_iface(const struct port *port, const uint8_t *dl_src, - uint16_t *dp_ifidx, tag_type *tags) + uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags) { struct iface *iface; assert(port->n_ifaces); if (port->n_ifaces == 1) { iface = port->ifaces[0]; - } else { - struct bond_entry *e = lookup_bond_entry(port, dl_src); + } else if (port->bond_mode == BM_AB) { + if (port->active_iface < 0) { + *tags |= port->no_ifaces_tag; + return false; + } + iface = port->ifaces[port->active_iface]; + } else if (port->bond_mode == BM_SLB){ + struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan); if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces || !port->ifaces[e->iface_idx]->enabled) { /* XXX select interface properly. The current interface selection @@ -1503,6 +2137,8 @@ choose_output_iface(const struct port *port, const uint8_t *dl_src, } *tags |= e->iface_tag; iface = port->ifaces[e->iface_idx]; + } else { + NOT_REACHED(); } *dp_ifidx = iface->dp_ifidx; *tags |= iface->tag; /* Currently only used for bonding. */ @@ -1519,16 +2155,18 @@ bond_link_status_update(struct iface *iface, bool carrier) /* Nothing to do. */ return; } - VLOG_INFO_RL(&rl, "interface %s: carrier %s", - iface->name, carrier ? "detected" : "dropped"); + VLOG_INFO_RL(&rl, "interface %s: link state %s", + iface->name, carrier ? "up" : "down"); 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->updelay && port->active_iface < 0) { - iface->delay_expires = time_msec(); - VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " - "other interface is up", iface->name, port->updelay); + } else if (carrier && port->active_iface < 0) { + bond_enable_slave(iface, true); + if (port->updelay) { + VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " + "other interface is up", iface->name, port->updelay); + } } else { int delay = carrier ? port->updelay : port->downdelay; iface->delay_expires = time_msec() + delay; @@ -1565,6 +2203,12 @@ bond_enable_slave(struct iface *iface, bool enable) struct port *port = iface->port; struct bridge *br = port->bridge; + /* This acts as a recursion check. If the act of disabling a slave + * causes a different slave to be enabled, the flag will allow us to + * skip redundant work when we reenter this function. It must be + * cleared on exit to keep things safe with multiple bonds. */ + static bool moving_active_iface = false; + iface->delay_expires = LLONG_MAX; if (enable == iface->enabled) { return; @@ -1577,18 +2221,66 @@ bond_enable_slave(struct iface *iface, bool enable) if (iface->port_ifidx == port->active_iface) { ofproto_revalidate(br->ofproto, port->active_iface_tag); + + /* Disabling a slave can lead to another slave being immediately + * enabled if there will be no active slaves but one is waiting + * on an updelay. In this case we do not need to run most of the + * code for the newly enabled slave since there was no period + * without an active slave and it is redundant with the disabling + * path. */ + moving_active_iface = true; bond_choose_active_iface(port); } bond_send_learning_packets(port); } else { VLOG_WARN("interface %s: enabled", iface->name); - if (port->active_iface < 0) { + if (port->active_iface < 0 && !moving_active_iface) { ofproto_revalidate(br->ofproto, port->no_ifaces_tag); bond_choose_active_iface(port); bond_send_learning_packets(port); } iface->tag = tag_create_random(); } + + moving_active_iface = false; + port->bond_compat_is_stale = true; +} + +/* Attempts to make the sum of the bond slaves' statistics appear on the fake + * bond interface. */ +static void +bond_update_fake_iface_stats(struct port *port) +{ + struct netdev_stats bond_stats; + struct netdev *bond_dev; + size_t i; + + memset(&bond_stats, 0, sizeof bond_stats); + + for (i = 0; i < port->n_ifaces; i++) { + struct netdev_stats slave_stats; + + if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) { + /* XXX: We swap the stats here because they are swapped back when + * reported by the internal device. The reason for this is + * internal devices normally represent packets going into the system + * but when used as fake bond device they represent packets leaving + * the system. We really should do this in the internal device + * itself because changing it here reverses the counts from the + * perspective of the switch. However, the internal device doesn't + * know what type of device it represents so we have to do it here + * for now. */ + bond_stats.tx_packets += slave_stats.rx_packets; + bond_stats.tx_bytes += slave_stats.rx_bytes; + bond_stats.rx_packets += slave_stats.tx_packets; + bond_stats.rx_bytes += slave_stats.tx_bytes; + } + } + + if (!netdev_open_default(port->name, &bond_dev)) { + netdev_set_stats(bond_dev, &bond_stats); + netdev_close(bond_dev); + } } static void @@ -1599,20 +2291,59 @@ bond_run(struct bridge *br) for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; + if (port->n_ifaces >= 2) { + char *devname; + + if (port->monitor) { + assert(!port->miimon); + + /* Track carrier going up and down on interfaces. */ + while (!netdev_monitor_poll(port->monitor, &devname)) { + struct iface *iface; + + iface = port_lookup_iface(port, devname); + if (iface) { + bool up = netdev_get_carrier(iface->netdev); + + bond_link_status_update(iface, up); + port_update_bond_compat(port); + } + free(devname); + } + } else { + assert(port->miimon); + + if (time_msec() >= port->bond_miimon_next_update) { + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + bool up = netdev_get_miimon(iface->netdev); + + bond_link_status_update(iface, up); + port_update_bond_compat(port); + } + port->bond_miimon_next_update = time_msec() + + port->bond_miimon_interval; + } + } + + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + if (time_msec() >= iface->delay_expires) { + bond_enable_slave(iface, !iface->enabled); + } + } + + if (port->bond_fake_iface + && time_msec() >= port->bond_next_fake_iface_update) { + bond_update_fake_iface_stats(port); + port->bond_next_fake_iface_update = time_msec() + 1000; + } + } + if (port->bond_compat_is_stale) { port->bond_compat_is_stale = false; port_update_bond_compat(port); } - - if (port->n_ifaces < 2) { - continue; - } - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (time_msec() >= iface->delay_expires) { - bond_enable_slave(iface, !iface->enabled); - } - } } } @@ -1626,36 +2357,38 @@ bond_wait(struct bridge *br) if (port->n_ifaces < 2) { continue; } + + if (port->monitor) { + netdev_monitor_poll_wait(port->monitor); + } + + if (port->miimon) { + poll_timer_wait_until(port->bond_miimon_next_update); + } + for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (iface->delay_expires != LLONG_MAX) { - poll_timer_wait(iface->delay_expires - time_msec()); + poll_timer_wait_until(iface->delay_expires); } } + if (port->bond_fake_iface) { + poll_timer_wait_until(port->bond_next_fake_iface_update); + } } } static bool -set_dst(struct dst *p, const flow_t *flow, +set_dst(struct dst *dst, const struct flow *flow, const struct port *in_port, const struct port *out_port, tag_type *tags) { - /* STP handling. - * - * XXX This uses too many tags: any broadcast flow will get one tag per - * destination port, and thus a broadcast on a switch of any size is likely - * to have all tag bits set. We should figure out a way to be smarter. - * - * This is OK when STP is disabled, because stp_state_tag is 0 then. */ - *tags |= out_port->stp_state_tag; - if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) { - return false; - } - - p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE + dst->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); + : flow->vlan_tci == 0 ? OFP_VLAN_NONE + : vlan_tci_to_vid(flow->vlan_tci)); + return choose_output_iface(out_port, flow->dl_src, dst->vlan, + &dst->dp_ifidx, tags); } static void @@ -1672,10 +2405,10 @@ swap_dst(struct dst *p, struct dst *q) * 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) +partition_dsts(struct dst_set *set, int vlan) { - struct dst *first = dsts; - struct dst *last = dsts + n_dsts; + struct dst *first = set->dsts; + struct dst *last = set->dsts + set->n; while (first != last) { /* Invariants: @@ -1710,13 +2443,48 @@ mirror_mask_ffs(mirror_mask_t mask) return ffs(mask); } +static void +dst_set_init(struct dst_set *set) +{ + set->dsts = set->builtin; + set->n = 0; + set->allocated = ARRAY_SIZE(set->builtin); +} + +static void +dst_set_add(struct dst_set *set, const struct dst *dst) +{ + if (set->n >= set->allocated) { + size_t new_allocated; + struct dst *new_dsts; + + new_allocated = set->allocated * 2; + new_dsts = xmalloc(new_allocated * sizeof *new_dsts); + memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); + + dst_set_free(set); + + set->dsts = new_dsts; + set->allocated = new_allocated; + } + set->dsts[set->n++] = *dst; +} + +static void +dst_set_free(struct dst_set *set) +{ + if (set->dsts != set->builtin) { + free(set->dsts); + } +} + static bool -dst_is_duplicate(const struct dst *dsts, size_t n_dsts, - const struct dst *test) +dst_is_duplicate(const struct dst_set *set, 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) { + for (i = 0; i < set->n; i++) { + if (set->dsts[i].vlan == test->vlan + && set->dsts[i].dp_ifidx == test->dp_ifidx) { return true; } } @@ -1726,7 +2494,8 @@ dst_is_duplicate(const struct dst *dsts, size_t n_dsts, static bool port_trunks_vlan(const struct port *port, uint16_t vlan) { - return port->vlan < 0 && bitmap_is_set(port->trunks, vlan); + return (port->vlan < 0 + && (!port->trunks || bitmap_is_set(port->trunks, vlan))); } static bool @@ -1735,59 +2504,86 @@ port_includes_vlan(const struct port *port, uint16_t vlan) return vlan == port->vlan || port_trunks_vlan(port, vlan); } -static size_t -compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, +static bool +port_is_floodable(const struct port *port) +{ + int i; + + for (i = 0; i < port->n_ifaces; i++) { + if (!ofproto_port_is_floodable(port->bridge->ofproto, + port->ifaces[i]->dp_ifidx)) { + return false; + } + } + return true; +} + +static void +compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, - struct dst dsts[], tag_type *tags) + struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface) { mirror_mask_t mirrors = in_port->src_mirrors; - struct dst *dst = dsts; + struct dst dst; + int flow_vlan; size_t i; - *tags |= in_port->stp_state_tag; + flow_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (flow_vlan == 0) { + flow_vlan = OFP_VLAN_NONE; + } + 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) + if (port != in_port + && port_is_floodable(port) + && port_includes_vlan(port, vlan) && !port->is_mirror_output_port - && set_dst(dst, flow, in_port, port, tags)) { + && set_dst(&dst, flow, in_port, port, tags)) { mirrors |= port->dst_mirrors; - dst++; + dst_set_add(set, &dst); } } - } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) { + *nf_output_iface = NF_OUT_FLOOD; + } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) { + dst_set_add(set, &dst); + *nf_output_iface = dst.dp_ifidx; mirrors |= out_port->dst_mirrors; - 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++; + if (set_dst(&dst, flow, in_port, m->out_port, tags) + && !dst_is_duplicate(set, &dst)) { + dst_set_add(set, &dst); } } else { for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; if (port_includes_vlan(port, m->out_vlan) - && set_dst(dst, flow, in_port, port, tags)) + && set_dst(&dst, flow, in_port, port, tags)) { if (port->vlan < 0) { - dst->vlan = m->out_vlan; + dst.vlan = m->out_vlan; } - if (dst_is_duplicate(dsts, dst - dsts, dst)) { + if (dst_is_duplicate(set, &dst)) { continue; } - if (dst->dp_ifidx == flow->in_port - && dst->vlan == vlan) { + + /* 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. */ + if (port == in_port && dst.vlan == flow_vlan) { /* Don't send out input port on same VLAN. */ continue; } - dst++; + dst_set_add(set, &dst); } } } @@ -1795,78 +2591,171 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, mirrors &= mirrors - 1; } - partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan)); - return dst - dsts; + partition_dsts(set, flow_vlan); } -static void UNUSED -print_dsts(const struct dst *dsts, size_t n) +static void OVS_UNUSED +print_dsts(const struct dst_set *set) { - for (; n--; dsts++) { - printf(">p%"PRIu16, dsts->dp_ifidx); - if (dsts->vlan != OFP_VLAN_NONE) { - printf("v%"PRIu16, dsts->vlan); + size_t i; + + for (i = 0; i < set->n; i++) { + const struct dst *dst = &set->dsts[i]; + + printf(">p%"PRIu16, dst->dp_ifidx); + if (dst->vlan != OFP_VLAN_NONE) { + printf("v%"PRIu16, dst->vlan); } } } static void -compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, +compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, - tag_type *tags, struct odp_actions *actions) + tag_type *tags, struct ofpbuf *actions, + uint16_t *nf_output_iface) { - struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)]; - size_t n_dsts; - const struct dst *p; + struct dst_set set; uint16_t cur_vlan; + size_t i; - n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags); + dst_set_init(&set); + compose_dsts(br, flow, vlan, in_port, out_port, &set, 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); + cur_vlan = vlan_tci_to_vid(flow->vlan_tci); + if (cur_vlan == 0) { + cur_vlan = OFP_VLAN_NONE; + } + for (i = 0; i < set.n; i++) { + const struct dst *dst = &set.dsts[i]; + if (dst->vlan != cur_vlan) { + if (dst->vlan == OFP_VLAN_NONE) { + nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN); } else { - a = odp_actions_add(actions, ODPAT_SET_VLAN_VID); - a->vlan_vid.vlan_vid = htons(p->vlan); + ovs_be16 tci; + tci = htons(dst->vlan & VLAN_VID_MASK); + tci |= flow->vlan_tci & htons(VLAN_PCP_MASK); + nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci); } - cur_vlan = p->vlan; + cur_vlan = dst->vlan; } - a = odp_actions_add(actions, ODPAT_OUTPUT); - a->output.port = p->dp_ifidx; + nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx); } + dst_set_free(&set); +} + +/* 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 struct flow *flow, + struct port *in_port, bool have_packet) +{ + int vlan = vlan_tci_to_vid(flow->vlan_tci); + if (in_port->vlan >= 0) { + if (vlan) { + /* XXX support double tagging? */ + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " + "packet received on port %s configured with " + "implicit VLAN %"PRIu16, + br->name, vlan, in_port->name, in_port->vlan); + } + return -1; + } + vlan = in_port->vlan; + } else { + if (!port_includes_vlan(in_port, vlan)) { + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " + "packet received on port %s not configured for " + "trunking VLAN %d", + br->name, vlan, in_port->name, vlan); + } + return -1; + } + } + + return vlan; } +/* A VM broadcasts a gratuitous ARP to indicate that it has resumed after + * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to + * indicate this; newer upstream kernels use gratuitous ARP requests. */ static bool -is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet) +is_gratuitous_arp(const struct flow *flow) { - struct arp_eth_header *arp = (struct arp_eth_header *) packet->data; return (flow->dl_type == htons(ETH_TYPE_ARP) && eth_addr_is_broadcast(flow->dl_dst) - && packet->size >= sizeof(struct arp_eth_header) - && arp->ar_op == ARP_OP_REQUEST); + && (flow->nw_proto == ARP_OP_REPLY + || (flow->nw_proto == ARP_OP_REQUEST + && flow->nw_src == flow->nw_dst))); } -/* If the composed actions may be applied to any packet in the given 'flow', - * returns true. Otherwise, the actions should only be applied to 'packet', or - * not at all, if 'packet' was NULL. */ +static void +update_learning_table(struct bridge *br, const struct flow *flow, int vlan, + struct port *in_port) +{ + enum grat_arp_lock_type lock_type; + tag_type rev_tag; + + /* We don't want to learn from gratuitous ARP packets that are reflected + * back over bond slaves so we lock the learning table. */ + lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE : + (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET : + GRAT_ARP_LOCK_CHECK; + + rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx, + lock_type); + if (rev_tag) { + /* 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); + } +} + +/* Determines whether packets in 'flow' within 'br' should be forwarded or + * dropped. Returns true if they may be forwarded, false if they should be + * dropped. + * + * If 'have_packet' is true, it indicates that the caller is processing a + * received packet. If 'have_packet' is false, then the caller is just + * revalidating an existing flow because configuration has changed. Either + * way, 'have_packet' only affects logging (there is no point in logging errors + * during revalidation). + * + * Sets '*in_portp' to the input port. This will be a null pointer if + * flow->in_port does not designate a known input port (in which case + * is_admissible() returns false). + * + * When returning true, sets '*vlanp' to the effective VLAN of the input + * packet, as returned by flow_get_vlan(). + * + * May also add tags to '*tags', although the current implementation only does + * so in one special case. + */ static bool -process_flow(struct bridge *br, const flow_t *flow, - const struct ofpbuf *packet, struct odp_actions *actions, - tag_type *tags) +is_admissible(struct bridge *br, const struct flow *flow, bool have_packet, + tag_type *tags, int *vlanp, struct port **in_portp) { struct iface *in_iface; struct port *in_port; - struct port *out_port = NULL; /* By default, drop the packet/flow. */ int vlan; /* Find the interface and port structure for the received packet. */ in_iface = iface_from_dp_ifidx(br, flow->in_port); if (!in_iface) { /* No interface? Something fishy... */ - if (packet != NULL) { + if (have_packet) { /* Odd. A few possible reasons here: * * - We deleted an interface but there are still a few packets @@ -1881,223 +2770,179 @@ process_flow(struct bridge *br, const flow_t *flow, 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); + "interface %"PRIu16, br->name, flow->in_port); } - /* Return without adding any actions, to drop packets on this flow. */ - return true; - } - in_port = in_iface->port; - - /* Figure out what VLAN this packet belongs to. - * - * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet - * belongs to VLAN 0, so we should treat both cases identically. (In the - * former case, the packet has an 802.1Q header that specifies VLAN 0, - * presumably to allow a priority to be specified. In the latter case, the - * packet does not have any 802.1Q header.) */ - vlan = ntohs(flow->dl_vlan); - if (vlan == OFP_VLAN_NONE) { - vlan = 0; - } - if (in_port->vlan >= 0) { - if (vlan) { - /* XXX support double tagging? */ - if (packet != NULL) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " - "packet received on port %s configured with " - "implicit VLAN %"PRIu16, - br->name, ntohs(flow->dl_vlan), - in_port->name, in_port->vlan); - } - goto done; - } - vlan = in_port->vlan; - } else { - if (!port_includes_vlan(in_port, vlan)) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " - "packet received on port %s not configured for " - "trunking VLAN %d", - br->name, vlan, in_port->name, vlan); - goto done; - } + *in_portp = NULL; + return false; } - - /* Drop frames for ports that STP wants entirely killed (both for - * forwarding and for learning). Later, after we do learning, we'll drop - * the frames that STP wants to do learning but not forwarding on. */ - if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) { - goto done; + *in_portp = in_port = in_iface->port; + *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet); + if (vlan < 0) { + return false; } /* Drop frames for reserved multicast addresses. */ if (eth_addr_is_reserved(flow->dl_dst)) { - goto done; + return false; } /* Drop frames on ports reserved for mirroring. */ if (in_port->is_mirror_output_port) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, " - "which is reserved exclusively for mirroring", - br->name, in_port->name); - goto done; - } - - /* Multicast (and broadcast) packets on bonds need special attention, to - * avoid receiving duplicates. */ - if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) { - *tags |= in_port->active_iface_tag; - if (in_port->active_iface != in_iface->port_ifidx) { - /* Drop all multicast packets on inactive slaves. */ - goto done; - } else { - /* Drop all multicast packets for which we have learned a different - * input port, because we probably sent the packet on one slaves - * and got it back on the active slave. Broadcast ARP replies are - * an exception to this rule: the host has moved to another - * switch. */ - int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); - if (src_idx != -1 && src_idx != in_port->port_idx) { - if (packet) { - if (!is_bcast_arp_reply(flow, packet)) { - goto done; - } - } else { - /* No way to know whether it's an ARP reply, because the - * flow entry doesn't include enough information and we - * don't have a packet. Punt. */ - return false; - } - } + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " + "%s, which is reserved exclusively for mirroring", + br->name, in_port->name); } + return false; } - /* MAC learning. */ - out_port = FLOOD_PORT; - if (br->ml) { - int out_port_idx; + /* Packets received on bonds need special attention to avoid duplicates. */ + if (in_port->n_ifaces > 1) { + int src_idx; + bool is_grat_arp_locked; - /* Learn source MAC (but don't try to learn from revalidation). */ - if (packet) { - tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src, - vlan, in_port->port_idx); - if (rev_tag) { - /* The log messages here could actually be useful in debugging, - * so keep the rate limit relatively high. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, - 300); - VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " - "on port %s in VLAN %d", - br->name, ETH_ADDR_ARGS(flow->dl_src), - in_port->name, vlan); - ofproto_revalidate(br->ofproto, rev_tag); + 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. */ + return false; } } - /* Determine output port. */ - out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, - tags); - if (out_port_idx >= 0 && out_port_idx < br->n_ports) { - out_port = br->ports[out_port_idx]; + /* Drop all packets for which we have learned a different input + * port, because we probably sent the packet on one slave and got + * it back on the other. Gratuitous ARP packets are an exception + * to this rule: the host has moved to another switch. The exception + * to the exception is if we locked the learning table to avoid + * reflections on bond slaves. If this is the case, just drop the + * packet now. */ + src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan, + &is_grat_arp_locked); + if (src_idx != -1 && src_idx != in_port->port_idx && + (!is_gratuitous_arp(flow) || is_grat_arp_locked)) { + return false; } } - /* Don't send packets out their input ports. Don't forward frames that STP - * wants us to discard. */ - if (in_port == out_port || in_port->stp_state == STP_LEARNING) { - out_port = NULL; - } - -done: - compose_actions(br, flow, vlan, in_port, out_port, tags, actions); - - /* - * We send out only a single packet, instead of setting up a flow, if the - * packet is an ARP directed to broadcast that arrived on a bonded - * interface. In such a situation ARP requests and replies must be handled - * differently, but OpenFlow unfortunately can't distinguish them. - */ - return (in_port->n_ifaces < 2 - || flow->dl_type != htons(ETH_TYPE_ARP) - || !eth_addr_is_broadcast(flow->dl_dst)); + return true; } -/* Careful: 'opp' is in host byte order and opp->port_no is an OFP port - * number. */ -static void -bridge_port_changed_ofhook_cb(enum ofp_port_reason reason, - const struct ofp_phy_port *opp, - void *br_) +/* 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 struct flow *flow, + const struct ofpbuf *packet, struct ofpbuf *actions, + tag_type *tags, uint16_t *nf_output_iface) { - struct bridge *br = br_; - struct iface *iface; - struct port *port; + struct port *in_port; + struct port *out_port; + int vlan; + int out_port_idx; - iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no)); - if (!iface) { - return; + /* Check whether we should drop packets in this flow. */ + if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) { + out_port = NULL; + goto done; } - port = iface->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); - port_destroy(port); - } + /* Learn source MAC (but don't try to learn from revalidation). */ + if (packet) { + update_learning_table(br, flow, vlan, in_port); + } - bridge_flush(br); + /* Determine output port. */ + out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags, + NULL); + if (out_port_idx >= 0 && out_port_idx < br->n_ports) { + out_port = br->ports[out_port_idx]; + } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { + /* If we are revalidating but don't have a learning entry then + * eject the flow. Installing a flow that floods packets opens + * up a window of time where we could learn from a packet reflected + * on a bond and blackhole packets before the learning table is + * updated to reflect the correct port. */ + return false; } else { - if (port->n_ifaces > 1) { - bool up = !(opp->state & OFPPS_LINK_DOWN); - bond_link_status_update(iface, up); - port_update_bond_compat(port); - } + out_port = FLOOD_PORT; + } + + /* Don't send packets out their input ports. */ + if (in_port == out_port) { + out_port = NULL; + } + +done: + if (in_port) { + compose_actions(br, flow, vlan, in_port, out_port, tags, actions, + nf_output_iface); } + + return true; } static bool -bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, - struct odp_actions *actions, tag_type *tags, void *br_) +bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, + struct ofpbuf *actions, tag_type *tags, + uint16_t *nf_output_iface, void *br_) { + struct iface *iface; struct bridge *br = br_; -#if 0 - if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE) - && eth_addr_equals(flow->dl_dst, stp_eth_addr)) { - brstp_receive(br, flow, payload); - return true; + COVERAGE_INC(bridge_process_flow); + + iface = iface_from_dp_ifidx(br, flow->in_port); + + if (cfm_should_process_flow(flow)) { + if (packet && iface->cfm) { + cfm_process_heartbeat(iface->cfm, packet); + } + return false; } -#endif - COVERAGE_INC(bridge_process_flow); - return process_flow(br, flow, packet, actions, tags); + return process_flow(br, flow, packet, actions, tags, nf_output_iface); } static void -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_) +bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, + const struct nlattr *actions, + size_t actions_len, + unsigned long long int n_bytes, void *br_) { struct bridge *br = br_; - const union odp_action *a; + const struct nlattr *a; + struct port *in_port; + tag_type dummy = 0; + unsigned int left; + int vlan; + + /* Feed information from the active flows back into the learning table to + * ensure that table is always in sync with what is actually flowing + * through the datapath. + * + * We test that 'tags' is nonzero to ensure that only flows that include an + * OFPP_NORMAL action are used for learning. This works because + * bridge_normal_ofhook_cb() always sets a nonzero tag value. */ + if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) { + update_learning_table(br, flow, vlan, in_port); + } + /* Account for bond slave utilization. */ if (!br->has_bonded_ports) { return; } - - for (a = actions; a < &actions[n_actions]; a++) { - if (a->type == ODPAT_OUTPUT) { - struct port *port = port_from_dp_ifidx(br, a->output.port); - if (port && port->n_ifaces >= 2) { - struct bond_entry *e = lookup_bond_entry(port, flow->dl_src); + NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) { + if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) { + struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a)); + if (out_port && out_port->n_ifaces >= 2 && + out_port->bond_mode == BM_SLB) { + uint16_t vlan = (flow->vlan_tci + ? vlan_tci_to_vid(flow->vlan_tci) + : OFP_VLAN_NONE); + struct bond_entry *e = lookup_bond_entry(out_port, + flow->dl_src, vlan); e->tx_bytes += n_bytes; } } @@ -2108,29 +2953,25 @@ static void bridge_account_checkpoint_ofhook_cb(void *br_) { struct bridge *br = br_; + long long int now; size_t i; if (!br->has_bonded_ports) { return; } - /* The current ofproto implementation calls this callback at least once a - * second, so this timer implementation is sufficient. */ - if (time_msec() < br->bond_next_rebalance) { - return; - } - br->bond_next_rebalance = time_msec() + 10000; - + now = time_msec(); for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - if (port->n_ifaces > 1) { + if (port->n_ifaces > 1 && port->bond_mode == BM_SLB + && now >= port->bond_next_rebalance) { + port->bond_next_rebalance = now + port->bond_rebalance_interval; bond_rebalance_port(port); } } } static struct ofhooks bridge_ofhooks = { - bridge_port_changed_ofhook_cb, bridge_normal_ofhook_cb, bridge_account_flow_ofhook_cb, bridge_account_checkpoint_ofhook_cb, @@ -2150,6 +2991,20 @@ struct slave_balance { size_t n_hashes; }; +static const char * +bond_mode_to_string(enum bond_mode bm) { + static char *bm_slb = "balance-slb"; + static char *bm_ab = "active-backup"; + + switch (bm) { + case BM_SLB: return bm_slb; + case BM_AB: return bm_ab; + } + + NOT_REACHED(); + return NULL; +} + /* Sorts pointers to pointers to bond_entries in ascending order by the * interface to which they are assigned, and within a single interface in * ascending order of bytes transmitted. */ @@ -2252,11 +3107,14 @@ log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port) /* Shifts 'hash' from 'from' to 'to' within 'port'. */ static void bond_shift_load(struct slave_balance *from, struct slave_balance *to, - struct bond_entry *hash) + int hash_idx) { + struct bond_entry *hash = from->hashes[hash_idx]; struct port *port = from->iface->port; uint64_t delta = hash->tx_bytes; + assert(port->bond_mode == BM_SLB); + VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) " "from %s to %s (now carrying %"PRIu64"kB and " "%"PRIu64"kB load, respectively)", @@ -2271,12 +3129,11 @@ bond_shift_load(struct slave_balance *from, struct slave_balance *to, * it require more work, the only purpose it would be to allow that hash to * be migrated to another slave in this rebalancing run, and there is no * point in doing that. */ - if (from->hashes[0] == hash) { + if (hash_idx == 0) { from->hashes++; } else { - int i = hash - from->hashes[0]; - memmove(from->hashes + i, from->hashes + i + 1, - (from->n_hashes - (i + 1)) * sizeof *from->hashes); + memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1, + (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes); } from->n_hashes--; @@ -2293,13 +3150,15 @@ bond_shift_load(struct slave_balance *from, struct slave_balance *to, static void bond_rebalance_port(struct port *port) { - struct slave_balance bals[DP_MAX_PORTS]; + struct slave_balance *bals; size_t n_bals; struct bond_entry *hashes[BOND_MASK + 1]; struct slave_balance *b, *from, *to; struct bond_entry *e; size_t i; + assert(port->bond_mode == BM_SLB); + /* 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 @@ -2311,6 +3170,7 @@ bond_rebalance_port(struct port *port) * become contiguous in memory, and then we point each 'hashes' members of * a slave_balance structure to the start of a contiguous group. */ n_bals = port->n_ifaces; + bals = xmalloc(n_bals * sizeof *bals); for (b = bals; b < &bals[n_bals]; b++) { b->iface = port->ifaces[b - bals]; b->tx_bytes = 0; @@ -2340,7 +3200,7 @@ bond_rebalance_port(struct port *port) while (!bals[n_bals - 1].iface->enabled) { n_bals--; if (!n_bals) { - return; + goto exit; } } @@ -2361,22 +3221,59 @@ bond_rebalance_port(struct port *port) /* 'from' is carrying significantly more load than 'to', and that * load is split across at least two different hashes. Pick a hash * to migrate to 'to' (the least-loaded slave), given that doing so - * must not cause 'to''s load to exceed 'from''s load. + * must decrease the ratio of the load on the two slaves by at + * least 0.1. * * The sort order we use means that we prefer to shift away the * smallest hashes instead of the biggest ones. There is little * reason behind this decision; we could use the opposite sort * order to shift away big hashes ahead of small ones. */ - size_t i; + bool order_swapped; for (i = 0; i < from->n_hashes; i++) { + double old_ratio, new_ratio; uint64_t delta = from->hashes[i]->tx_bytes; - if (to->tx_bytes + delta < from->tx_bytes - delta) { + + if (delta == 0 || from->tx_bytes - delta == 0) { + /* Pointless move. */ + continue; + } + + order_swapped = from->tx_bytes - delta < to->tx_bytes + delta; + + if (to->tx_bytes == 0) { + /* Nothing on the new slave, move it. */ + break; + } + + old_ratio = (double)from->tx_bytes / to->tx_bytes; + new_ratio = (double)(from->tx_bytes - delta) / + (to->tx_bytes + delta); + + if (new_ratio == 0) { + /* Should already be covered but check to prevent division + * by zero. */ + continue; + } + + if (new_ratio < 1) { + new_ratio = 1 / new_ratio; + } + + if (old_ratio - new_ratio > 0.1) { + /* Would decrease the ratio, move it. */ break; } } if (i < from->n_hashes) { - bond_shift_load(from, to, from->hashes[i]); + bond_shift_load(from, to, i); + port->bond_compat_is_stale = true; + + /* If the result of the migration changed the relative order of + * 'from' and 'to' swap them back to maintain invariants. */ + if (order_swapped) { + swap_bals(from, to); + } /* Re-sort 'bals'. Note that this may make 'from' and 'to' * point to different slave_balance structures. It is only @@ -2387,7 +3284,6 @@ bond_rebalance_port(struct port *port) } else { from++; } - port->bond_compat_is_stale = true; } } @@ -2396,6 +3292,9 @@ bond_rebalance_port(struct port *port) for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) { e->tx_bytes /= 2; } + +exit: + free(bals); } static void @@ -2406,44 +3305,24 @@ bond_send_learning_packets(struct port *port) struct ofpbuf packet; int error, n_packets, n_errors; - if (!port->n_ifaces || port->active_iface < 0 || !br->ml) { + if (!port->n_ifaces || port->active_iface < 0) { return; } ofpbuf_init(&packet, 128); error = n_packets = n_errors = 0; - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { - static const char s[] = "Open vSwitch Bond Failover"; + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { union ofp_action actions[2], *a; - struct eth_header *eth; - struct llc_snap_header *llc_snap; uint16_t dp_ifidx; tag_type tags = 0; - flow_t flow; + struct flow flow; int retval; if (e->port == port->port_idx - || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) { + || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) { continue; } - /* Compose packet to send. */ - ofpbuf_clear(&packet); - eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN); - llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN); - ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */ - ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN); - - memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN); - memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN); - eth->eth_type = htons(packet.size - ETH_HEADER_LEN); - - llc_snap->llc.llc_dsap = LLC_DSAP_SNAP; - llc_snap->llc.llc_ssap = LLC_SSAP_SNAP; - llc_snap->llc.llc_cntl = LLC_CNTL_SNAP; - memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3); - llc_snap->snap.snap_type = htons(0xf177); /* Random number. */ - /* Compose actions. */ memset(actions, 0, sizeof actions); a = actions; @@ -2460,7 +3339,9 @@ bond_send_learning_packets(struct port *port) /* Send packet. */ n_packets++; - flow_extract(&packet, ODPP_NONE, &flow); + compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, + e->mac); + flow_extract(&packet, 0, ODPP_NONE, &flow); retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, &packet); if (retval) { @@ -2484,14 +3365,15 @@ bond_send_learning_packets(struct port *port) /* Bonding unixctl user interface functions. */ static void -bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED) +bond_unixctl_list(struct unixctl_conn *conn, + const char *args OVS_UNUSED, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct bridge *br; - ds_put_cstr(&ds, "bridge\tbond\tslaves\n"); + ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n"); - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { size_t i; for (i = 0; i < br->n_ports; i++) { @@ -2499,7 +3381,8 @@ bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED) if (port->n_ifaces > 1) { size_t j; - ds_put_format(&ds, "%s\t%s\t", br->name, port->name); + ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name, + bond_mode_to_string(port->bond_mode)); for (j = 0; j < port->n_ifaces; j++) { const struct iface *iface = port->ifaces[j]; if (j) { @@ -2520,7 +3403,7 @@ bond_find(const char *name) { const struct bridge *br; - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { size_t i; for (i = 0; i < br->n_ports; i++) { @@ -2534,7 +3417,8 @@ bond_find(const char *name) } static void -bond_unixctl_show(struct unixctl_conn *conn, const char *args) +bond_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct port *port; @@ -2546,10 +3430,24 @@ bond_unixctl_show(struct unixctl_conn *conn, const char *args) return; } + ds_put_format(&ds, "bond_mode: %s\n", + bond_mode_to_string(port->bond_mode)); + ds_put_format(&ds, "bond-detect-mode: %s\n", + port->miimon ? "miimon" : "carrier"); + + if (port->miimon) { + ds_put_format(&ds, "bond-miimon-interval: %lld\n", + port->bond_miimon_interval); + } + ds_put_format(&ds, "updelay: %d ms\n", port->updelay); ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); - ds_put_format(&ds, "next rebalance: %lld ms\n", - port->bridge->bond_next_rebalance - time_msec()); + + if (port->bond_mode == BM_SLB) { + ds_put_format(&ds, "next rebalance: %lld ms\n", + port->bond_next_rebalance - time_msec()); + } + for (j = 0; j < port->n_ifaces; j++) { const struct iface *iface = port->ifaces[j]; struct bond_entry *be; @@ -2566,6 +3464,10 @@ bond_unixctl_show(struct unixctl_conn *conn, const char *args) iface->delay_expires - time_msec()); } + if (port->bond_mode != BM_SLB) { + continue; + } + /* Hashes. */ for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { int hash = be - port->bond_hash; @@ -2575,21 +3477,17 @@ bond_unixctl_show(struct unixctl_conn *conn, const char *args) continue; } - ds_put_format(&ds, "\thash %d: %lld kB load\n", + ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", hash, be->tx_bytes / 1024); /* MACs. */ - if (!port->bridge->ml) { - break; - } - - LIST_FOR_EACH (me, struct mac_entry, lru_node, - &port->bridge->ml->lrus) { + LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) { uint16_t dp_ifidx; tag_type tags = 0; - if (bond_hash(me->mac) == hash + if (bond_hash(me->mac, me->vlan) == hash && me->port != port->port_idx - && choose_output_iface(port, me->mac, &dp_ifidx, &tags) + && choose_output_iface(port, me->mac, me->vlan, + &dp_ifidx, &tags) && dp_ifidx == iface->dp_ifidx) { ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n", @@ -2603,12 +3501,12 @@ bond_unixctl_show(struct unixctl_conn *conn, const char *args) } static void -bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_) +bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) { char *args = (char *) args_; char *save_ptr = NULL; char *bond_s, *hash_s, *slave_s; - uint8_t mac[ETH_ADDR_LEN]; struct port *port; struct iface *iface; struct bond_entry *entry; @@ -2629,10 +3527,12 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_) return; } - 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)) { + if (port->bond_mode != BM_SLB) { + unixctl_command_reply(conn, 501, "not an SLB bond"); + return; + } + + if (strspn(hash_s, "0123456789") == strlen(hash_s)) { hash = atoi(hash_s) & BOND_MASK; } else { unixctl_command_reply(conn, 501, "bad hash"); @@ -2659,7 +3559,8 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_) } static void -bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_) +bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) { char *args = (char *) args_; char *save_ptr = NULL; @@ -2739,118 +3640,244 @@ enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) } static void -bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args) +bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, + void *aux OVS_UNUSED) { enable_slave(conn, args, true); } static void -bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args) +bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, + void *aux OVS_UNUSED) { enable_slave(conn, args, false); } +static void +bond_unixctl_hash(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) +{ + char *args = (char *) args_; + uint8_t mac[ETH_ADDR_LEN]; + uint8_t hash; + char *hash_cstr; + unsigned int vlan; + char *mac_s, *vlan_s; + char *save_ptr = NULL; + + mac_s = strtok_r(args, " ", &save_ptr); + vlan_s = strtok_r(NULL, " ", &save_ptr); + + if (vlan_s) { + if (sscanf(vlan_s, "%u", &vlan) != 1) { + unixctl_command_reply(conn, 501, "invalid vlan"); + return; + } + } else { + vlan = OFP_VLAN_NONE; + } + + if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) + == ETH_ADDR_SCAN_COUNT) { + hash = bond_hash(mac, vlan); + + hash_cstr = xasprintf("%u", hash); + unixctl_command_reply(conn, 200, hash_cstr); + free(hash_cstr); + } else { + unixctl_command_reply(conn, 501, "invalid mac"); + } +} + static void bond_init(void) { - unixctl_command_register("bond/list", bond_unixctl_list); - unixctl_command_register("bond/show", bond_unixctl_show); - unixctl_command_register("bond/migrate", bond_unixctl_migrate); + unixctl_command_register("bond/list", bond_unixctl_list, NULL); + unixctl_command_register("bond/show", bond_unixctl_show, NULL); + unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL); unixctl_command_register("bond/set-active-slave", - bond_unixctl_set_active_slave); - unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave); - unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave); + bond_unixctl_set_active_slave, NULL); + unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave, + NULL); + unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave, + NULL); + unixctl_command_register("bond/hash", bond_unixctl_hash, NULL); } /* Port functions. */ -static void +static struct port * port_create(struct bridge *br, const char *name) { struct port *port; - port = xcalloc(1, sizeof *port); + port = xzalloc(sizeof *port); port->bridge = br; port->port_idx = br->n_ports; port->vlan = -1; port->trunks = NULL; port->name = xstrdup(name); port->active_iface = -1; - port->stp_state = STP_DISABLED; - port->stp_state_tag = 0; if (br->n_ports >= br->allocated_ports) { br->ports = x2nrealloc(br->ports, &br->allocated_ports, sizeof *br->ports); } br->ports[br->n_ports++] = port; + shash_add_assert(&br->port_by_name, port->name, port); VLOG_INFO("created port %s on bridge %s", port->name, br->name); bridge_flush(br); + + return port; +} + +static const char * +get_port_other_config(const struct ovsrec_port *port, const char *key, + const char *default_value) +{ + const char *value; + + value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config, + key); + return value ? value : default_value; +} + +static void +port_del_ifaces(struct port *port, const struct ovsrec_port *cfg) +{ + struct shash new_ifaces; + size_t i; + + /* Collect list of new interfaces. */ + shash_init(&new_ifaces); + for (i = 0; i < cfg->n_interfaces; i++) { + const char *name = cfg->interfaces[i]->name; + shash_add_once(&new_ifaces, name, NULL); + } + + /* Get rid of deleted interfaces. */ + for (i = 0; i < port->n_ifaces; ) { + if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) { + iface_destroy(port->ifaces[i]); + } else { + i++; + } + } + + shash_destroy(&new_ifaces); } static void -port_reconfigure(struct port *port) +port_reconfigure(struct port *port, const struct ovsrec_port *cfg) { - bool bonded = cfg_has_section("bonding.%s", port->name); - struct svec old_ifaces, new_ifaces; + const char *detect_mode; + struct shash new_ifaces; + long long int next_rebalance, miimon_next_update; unsigned long *trunks; int vlan; size_t i; - /* Collect old and new interfaces. */ - svec_init(&old_ifaces); - svec_init(&new_ifaces); - for (i = 0; i < port->n_ifaces; i++) { - svec_add(&old_ifaces, port->ifaces[i]->name); + port->cfg = cfg; + + /* Update settings. */ + port->updelay = cfg->bond_updelay; + if (port->updelay < 0) { + port->updelay = 0; + } + port->downdelay = cfg->bond_downdelay; + if (port->downdelay < 0) { + port->downdelay = 0; + } + port->bond_rebalance_interval = atoi( + get_port_other_config(cfg, "bond-rebalance-interval", "10000")); + if (port->bond_rebalance_interval < 1000) { + port->bond_rebalance_interval = 1000; + } + next_rebalance = time_msec() + port->bond_rebalance_interval; + if (port->bond_next_rebalance > next_rebalance) { + port->bond_next_rebalance = next_rebalance; } - svec_sort(&old_ifaces); - if (bonded) { - cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name); - if (!new_ifaces.n) { - VLOG_ERR("port %s: no interfaces specified for bonded port", - port->name); - } else if (new_ifaces.n == 1) { - VLOG_WARN("port %s: only 1 interface specified for bonded port", - port->name); - } - port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name); - if (port->updelay < 0) { - port->updelay = 0; - } - port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name); - if (port->downdelay < 0) { - port->downdelay = 0; - } + detect_mode = get_port_other_config(cfg, "bond-detect-mode", + "carrier"); + + if (!strcmp(detect_mode, "carrier")) { + port->miimon = false; + } else if (!strcmp(detect_mode, "miimon")) { + port->miimon = true; } else { - svec_init(&new_ifaces); - svec_add(&new_ifaces, port->name); + port->miimon = false; + VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to " + "carrier", port->name, detect_mode); } - /* Get rid of deleted interfaces and add new interfaces. */ - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - if (!svec_contains(&new_ifaces, iface->name)) { - iface_destroy(iface); - } else { - i++; - } + port->bond_miimon_interval = atoi( + get_port_other_config(cfg, "bond-miimon-interval", "200")); + if (port->bond_miimon_interval < 100) { + port->bond_miimon_interval = 100; + } + miimon_next_update = time_msec() + port->bond_miimon_interval; + if (port->bond_miimon_next_update > miimon_next_update) { + port->bond_miimon_next_update = miimon_next_update; + } + + if (!port->cfg->bond_mode || + !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) { + port->bond_mode = BM_SLB; + } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) { + port->bond_mode = BM_AB; + } else { + port->bond_mode = BM_SLB; + VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s", + port->name, port->cfg->bond_mode, + bond_mode_to_string(port->bond_mode)); } - for (i = 0; i < new_ifaces.n; i++) { - const char *name = new_ifaces.names[i]; - if (!svec_contains(&old_ifaces, name)) { - iface_create(port, name); + + /* Add new interfaces and update 'cfg' member of existing ones. */ + shash_init(&new_ifaces); + for (i = 0; i < cfg->n_interfaces; i++) { + const struct ovsrec_interface *if_cfg = cfg->interfaces[i]; + struct iface *iface; + + if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) { + VLOG_WARN("port %s: %s specified twice as port interface", + port->name, if_cfg->name); + iface_set_ofport(if_cfg, -1); + continue; + } + + iface = iface_lookup(port->bridge, if_cfg->name); + if (iface) { + if (iface->port != port) { + VLOG_ERR("bridge %s: %s interface is on multiple ports, " + "removing from %s", + port->bridge->name, if_cfg->name, iface->port->name); + continue; + } + iface->cfg = if_cfg; + } else { + iface = iface_create(port, if_cfg); } + + /* 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(if_cfg->name, port->bridge->name) ? "internal" + : if_cfg->type[0] ? if_cfg->type + : "system"); } + shash_destroy(&new_ifaces); /* Get VLAN tag. */ vlan = -1; - if (cfg_has("vlan.%s.tag", port->name)) { - if (!bonded) { - vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name); + if (cfg->tag) { + if (port->n_ifaces < 2) { + vlan = *cfg->tag; if (vlan >= 0 && vlan <= 4095) { VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan); + } else { + vlan = -1; } } else { /* It's possible that bonded, VLAN-tagged ports make sense. Maybe @@ -2866,15 +3893,13 @@ port_reconfigure(struct port *port) /* Get trunked VLANs. */ trunks = NULL; - if (vlan < 0) { - size_t n_trunks, n_errors; - size_t i; + if (vlan < 0 && cfg->n_trunks) { + size_t n_errors; trunks = bitmap_allocate(4096); - n_trunks = cfg_count("vlan.%s.trunks", port->name); n_errors = 0; - for (i = 0; i < n_trunks; i++) { - int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name); + for (i = 0; i < cfg->n_trunks; i++) { + int trunk = cfg->trunks[i]; if (trunk >= 0) { bitmap_set1(trunks, trunk); } else { @@ -2883,20 +3908,17 @@ port_reconfigure(struct port *port) } if (n_errors) { VLOG_ERR("port %s: invalid values for %zu trunk VLANs", - port->name, n_trunks); - } - if (n_errors == n_trunks) { - if (n_errors) { - VLOG_ERR("port %s: no valid trunks, trunking all VLANs", - port->name); - } - bitmap_set_multiple(trunks, 0, 4096, 1); + port->name, cfg->n_trunks); } - } else { - if (cfg_has("vlan.%s.trunks", port->name)) { - VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan", - port->name, port->name); + if (n_errors == cfg->n_trunks) { + VLOG_ERR("port %s: no valid trunks, trunking all VLANs", + port->name); + bitmap_free(trunks); + trunks = NULL; } + } else if (vlan >= 0 && cfg->n_trunks) { + VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", + port->name); } if (trunks == NULL ? port->trunks != NULL @@ -2905,9 +3927,6 @@ port_reconfigure(struct port *port) } bitmap_free(port->trunks); port->trunks = trunks; - - svec_destroy(&old_ifaces); - svec_destroy(&new_ifaces); } static void @@ -2916,7 +3935,7 @@ port_destroy(struct port *port) if (port) { struct bridge *br = port->bridge; struct port *del; - size_t i; + int i; proc_net_compat_update_vlan(port->name, NULL, 0); proc_net_compat_update_bond(port->name, NULL); @@ -2932,9 +3951,14 @@ port_destroy(struct port *port) iface_destroy(port->ifaces[port->n_ifaces - 1]); } + shash_find_and_delete_assert(&br->port_by_name, port->name); + del = br->ports[port->port_idx] = br->ports[--br->n_ports]; del->port_idx = port->port_idx; + VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); + + netdev_monitor_destroy(port->monitor); free(port->ifaces); bitmap_free(port->trunks); free(port->name); @@ -2953,34 +3977,23 @@ port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) static struct port * port_lookup(const struct bridge *br, const char *name) { - size_t i; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (!strcmp(port->name, name)) { - return port; - } - } - return NULL; + return shash_find_data(&br->port_by_name, name); } static struct iface * port_lookup_iface(const struct port *port, const char *name) { - size_t j; - - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (!strcmp(iface->name, name)) { - return iface; - } - } - return NULL; + struct iface *iface = iface_lookup(port->bridge, name); + return iface && iface->port == port ? iface : NULL; } static void port_update_bonding(struct port *port) { + if (port->monitor) { + netdev_monitor_destroy(port->monitor); + port->monitor = NULL; + } if (port->n_ifaces < 2) { /* Not a bonded port. */ if (port->bond_hash) { @@ -2988,10 +4001,12 @@ port_update_bonding(struct port *port) port->bond_hash = NULL; port->bond_compat_is_stale = true; } + + port->bond_fake_iface = false; } else { - if (!port->bond_hash) { - size_t i; + size_t i; + if (port->bond_mode == BM_SLB && !port->bond_hash) { port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); for (i = 0; i <= BOND_MASK; i++) { struct bond_entry *e = &port->bond_hash[i]; @@ -3000,8 +4015,25 @@ port_update_bonding(struct port *port) } port->no_ifaces_tag = tag_create_random(); bond_choose_active_iface(port); + port->bond_next_rebalance + = time_msec() + port->bond_rebalance_interval; + + if (port->cfg->bond_fake_iface) { + port->bond_next_fake_iface_update = time_msec(); + } + } else if (port->bond_mode != BM_SLB) { + free(port->bond_hash); + port->bond_hash = NULL; } port->bond_compat_is_stale = true; + port->bond_fake_iface = port->cfg->bond_fake_iface; + + if (!port->miimon) { + port->monitor = netdev_monitor_create(); + for (i = 0; i < port->n_ifaces; i++) { + netdev_monitor_add(port->monitor, port->ifaces[i]->netdev); + } + } } } @@ -3012,7 +4044,7 @@ port_update_bond_compat(struct port *port) struct compat_bond bond; size_t i; - if (port->n_ifaces < 2) { + if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) { proc_net_compat_update_bond(port->name, NULL); return; } @@ -3040,14 +4072,40 @@ port_update_bond_compat(struct port *port) struct iface *iface = port->ifaces[i]; struct compat_bond_slave *slave = &bond.slaves[i]; slave->name = iface->name; - slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) || - (!iface->enabled && iface->delay_expires != LLONG_MAX)); + + /* We need to make the same determination as the Linux bonding + * code to determine whether a slave should be consider "up". + * The Linux function bond_miimon_inspect() supports four + * BOND_LINK_* states: + * + * - BOND_LINK_UP: carrier detected, updelay has passed. + * - BOND_LINK_FAIL: carrier lost, downdelay in progress. + * - BOND_LINK_DOWN: carrier lost, downdelay has passed. + * - BOND_LINK_BACK: carrier detected, updelay in progress. + * + * The function bond_info_show_slave() only considers BOND_LINK_UP + * to be "up" and anything else to be "down". + */ + slave->up = iface->enabled && iface->delay_expires == LLONG_MAX; if (slave->up) { bond.up = true; } netdev_get_etheraddr(iface->netdev, slave->mac); } + if (port->bond_fake_iface) { + struct netdev *bond_netdev; + + if (!netdev_open_default(port->name, &bond_netdev)) { + if (bond.up) { + netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); + } else { + netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); + } + netdev_close(bond_netdev); + } + } + proc_net_compat_update_bond(port->name, &bond); free(bond.slaves); } @@ -3092,11 +4150,33 @@ port_update_vlan_compat(struct port *port) /* Interface functions. */ static void -iface_create(struct port *port, const char *name) +iface_send_packet(struct iface *iface, struct ofpbuf *packet) +{ + struct flow flow; + union ofp_action action; + + memset(&action, 0, sizeof action); + action.output.type = htons(OFPAT_OUTPUT); + action.output.len = htons(sizeof action); + action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx)); + + flow_extract(packet, 0, ODPP_NONE, &flow); + + if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1, + packet)) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name); + } +} + +static struct iface * +iface_create(struct port *port, const struct ovsrec_interface *if_cfg) { + struct bridge *br = port->bridge; struct iface *iface; + char *name = if_cfg->name; - iface = xcalloc(1, sizeof *iface); + iface = xzalloc(sizeof *iface); iface->port = port; iface->port_ifidx = port->n_ifaces; iface->name = xstrdup(name); @@ -3104,6 +4184,9 @@ iface_create(struct port *port, const char *name) iface->tag = tag_create_random(); iface->delay_expires = LLONG_MAX; iface->netdev = NULL; + iface->cfg = if_cfg; + + shash_add_assert(&br->iface_by_name, iface->name, iface); if (port->n_ifaces >= port->allocated_ifaces) { port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, @@ -3111,12 +4194,14 @@ iface_create(struct port *port, const char *name) } port->ifaces[port->n_ifaces++] = iface; if (port->n_ifaces > 1) { - port->bridge->has_bonded_ports = true; + br->has_bonded_ports = true; } VLOG_DBG("attached network device %s to port %s", iface->name, port->name); - bridge_flush(port->bridge); + bridge_flush(br); + + return iface; } static void @@ -3128,16 +4213,20 @@ iface_destroy(struct iface *iface) bool del_active = port->active_iface == iface->port_ifidx; struct iface *del; + if (port->monitor) { + netdev_monitor_remove(port->monitor, iface->netdev); + } + + shash_find_and_delete_assert(&br->iface_by_name, iface->name); + if (iface->dp_ifidx >= 0) { - port_array_set(&br->ifaces, iface->dp_ifidx, NULL); + hmap_remove(&br->ifaces, &iface->dp_ifidx_node); } del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces]; del->port_ifidx = iface->port_ifidx; netdev_close(iface->netdev); - free(iface->name); - free(iface); if (del_active) { ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); @@ -3145,6 +4234,11 @@ iface_destroy(struct iface *iface) bond_send_learning_packets(port); } + cfm_destroy(iface->cfm); + + free(iface->name); + free(iface); + bridge_flush(port->bridge); } } @@ -3152,68 +4246,260 @@ iface_destroy(struct iface *iface) static struct iface * iface_lookup(const struct bridge *br, const char *name) { - size_t i, j; + return shash_find_data(&br->iface_by_name, name); +} - 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; - } +static struct iface * +iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) +{ + struct iface *iface; + + HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node, + hash_int(dp_ifidx, 0), &br->ifaces) { + if (iface->dp_ifidx == dp_ifidx) { + return iface; } } return NULL; } -static struct iface * -iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) +/* Set Ethernet address of 'iface', if one is specified in the configuration + * file. */ +static void +iface_set_mac(struct iface *iface) { - return port_array_get(&br->ifaces, dp_ifidx); + uint8_t ea[ETH_ADDR_LEN]; + + if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) { + if (eth_addr_is_multicast(ea)) { + VLOG_ERR("interface %s: cannot set MAC to multicast address", + iface->name); + } else if (iface->dp_ifidx == ODPP_LOCAL) { + VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead", + iface->name, iface->name); + } else { + int error = netdev_set_etheraddr(iface->netdev, ea); + if (error) { + VLOG_ERR("interface %s: setting MAC failed (%s)", + iface->name, strerror(error)); + } + } + } } - -/* Port mirroring. */ +/* Sets the ofport column of 'if_cfg' to 'ofport'. */ static void -mirror_reconfigure(struct bridge *br) +iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport) +{ + if (if_cfg) { + ovsrec_interface_set_ofport(if_cfg, &ofport, 1); + } +} + +/* 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) { - struct svec old_mirrors, new_mirrors; size_t i; - /* 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); + shash_init(shash); + for (i = 0; i < n; i++) { + shash_add(shash, keys[i], values[i]); + } +} + +/* 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; + + 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++; + } + + *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 +iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos) +{ + if (!qos || qos->type[0] == '\0') { + netdev_set_qos(iface->netdev, NULL, NULL); + } else { + struct iface_delete_queues_cbdata cbdata; + struct shash details; + size_t i; + + /* Configure top-level Qos for 'iface'. */ + shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config, + qos->n_other_config, &details); + netdev_set_qos(iface->netdev, qos->type, &details); + shash_destroy(&details); + + /* Deconfigure queues that were deleted. */ + cbdata.netdev = iface->netdev; + cbdata.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'. */ + for (i = 0; i < qos->n_queues; i++) { + const struct ovsrec_queue *queue = qos->value_queues[i]; + unsigned int queue_id = qos->key_queues[i]; + + shash_from_ovs_idl_map(queue->key_other_config, + queue->value_other_config, + queue->n_other_config, &details); + netdev_set_queue(iface->netdev, queue_id, &details); + shash_destroy(&details); } } +} + +static void +iface_update_cfm(struct iface *iface) +{ + size_t i; + struct cfm *cfm; + uint16_t *remote_mps; + struct ovsrec_monitor *mon; + uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN]; + + mon = iface->cfg->monitor; + + if (!mon) { + return; + } + + if (netdev_get_etheraddr(iface->netdev, ea)) { + VLOG_WARN("interface %s: Failed to get ethernet address. " + "Skipping Monitor.", iface->name); + return; + } + + if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) { + VLOG_WARN("interface %s: Failed to generate MAID.", iface->name); + return; + } + + if (!iface->cfm) { + iface->cfm = cfm_create(); + } + + cfm = iface->cfm; + cfm->mpid = mon->mpid; + cfm->interval = mon->interval ? *mon->interval : 1000; + + memcpy(cfm->eth_src, ea, sizeof cfm->eth_src); + memcpy(cfm->maid, maid, sizeof cfm->maid); + + remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps); + for(i = 0; i < mon->n_remote_mps; i++) { + remote_mps[i] = mon->remote_mps[i]->mpid; + } + cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps); + free(remote_mps); + + if (!cfm_configure(iface->cfm)) { + cfm_destroy(iface->cfm); + iface->cfm = NULL; + } +} + +/* Port mirroring. */ + +static struct mirror * +mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid) +{ + int i; - /* Get rid of deleted mirrors and add new mirrors. */ - svec_sort(&old_mirrors); - assert(svec_is_unique(&old_mirrors)); - svec_sort(&new_mirrors); - assert(svec_is_unique(&new_mirrors)); for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; - if (m && !svec_contains(&new_mirrors, m->name)) { - mirror_destroy(m); + if (m && uuid_equals(uuid, &m->uuid)) { + return 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); + return NULL; +} + +static void +mirror_reconfigure(struct bridge *br) +{ + unsigned long *rspan_vlans; + int i; + + /* Get rid of deleted mirrors. */ + for (i = 0; i < MAX_MIRRORS; i++) { + struct mirror *m = br->mirrors[i]; + if (m) { + const struct ovsdb_datum *mc; + union ovsdb_atom atom; + + mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID); + atom.uuid = br->mirrors[i]->uuid; + if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) { + mirror_destroy(m); + } } } - 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]); + /* Add new mirrors and reconfigure existing ones. */ + for (i = 0; i < br->cfg->n_mirrors; i++) { + struct ovsrec_mirror *cfg = br->cfg->mirrors[i]; + struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid); + if (m) { + mirror_reconfigure_one(m, cfg); + } else { + mirror_create(br, cfg); } } @@ -3227,10 +4513,31 @@ mirror_reconfigure(struct bridge *br) m->out_port->is_mirror_output_port = true; } } + + /* Update flooded vlans (for RSPAN). */ + rspan_vlans = NULL; + if (br->cfg->n_flood_vlans) { + rspan_vlans = bitmap_allocate(4096); + + for (i = 0; i < br->cfg->n_flood_vlans; i++) { + int64_t vlan = br->cfg->flood_vlans[i]; + if (vlan >= 0 && vlan < 4096) { + bitmap_set1(rspan_vlans, vlan); + VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64, + br->name, vlan); + } else { + VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN", + br->name, vlan); + } + } + } + if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) { + bridge_flush(br); + } } static void -mirror_create(struct bridge *br, const char *name) +mirror_create(struct bridge *br, struct ovsrec_mirror *cfg) { struct mirror *m; size_t i; @@ -3238,7 +4545,7 @@ mirror_create(struct bridge *br, const char *name) 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); + "cannot create %s", br->name, MAX_MIRRORS, cfg->name); return; } if (!br->mirrors[i]) { @@ -3246,19 +4553,21 @@ mirror_create(struct bridge *br, const char *name) } } - VLOG_INFO("created port mirror %s on bridge %s", name, br->name); + VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name); bridge_flush(br); - br->mirrors[i] = m = xcalloc(1, sizeof *m); + br->mirrors[i] = m = xzalloc(sizeof *m); m->bridge = br; m->idx = i; - m->name = xstrdup(name); - svec_init(&m->src_ports); - svec_init(&m->dst_ports); + m->name = xstrdup(cfg->name); + shash_init(&m->src_ports); + shash_init(&m->dst_ports); m->vlans = NULL; m->n_vlans = 0; m->out_vlan = -1; m->out_port = NULL; + + mirror_reconfigure_one(m, cfg); } static void @@ -3273,11 +4582,12 @@ mirror_destroy(struct mirror *m) br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); } - svec_destroy(&m->src_ports); - svec_destroy(&m->dst_ports); + shash_destroy(&m->src_ports); + shash_destroy(&m->dst_ports); free(m->vlans); m->bridge->mirrors[m->idx] = NULL; + free(m->name); free(m); bridge_flush(br); @@ -3285,45 +4595,36 @@ mirror_destroy(struct mirror *m) } static void -prune_ports(struct mirror *m, struct svec *ports) +mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports, + struct shash *names) { - struct svec tmp; size_t i; - svec_sort_unique(ports); - - svec_init(&tmp); - for (i = 0; i < ports->n; i++) { - const char *name = ports->names[i]; + for (i = 0; i < n_ports; i++) { + const char *name = ports[i]->name; if (port_lookup(m->bridge, name)) { - svec_add(&tmp, name); + shash_add_once(names, name, NULL); } else { - VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s", - m->bridge->name, m->name, name); + VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent " + "port %s", m->bridge->name, m->name, name); } } - svec_swap(ports, &tmp); - svec_destroy(&tmp); } static size_t -prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans) +mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg, + int **vlans) { - size_t n_vlans, i; - - /* This isn't perfect: it won't combine "0" and "00", and the textual sort - * order won't give us numeric sort order. But that's good enough for what - * we need right now. */ - svec_sort_unique(vlan_strings); + size_t n_vlans; + size_t i; - *vlans = xmalloc(sizeof *vlans * vlan_strings->n); + *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan); n_vlans = 0; - for (i = 0; i < vlan_strings->n; i++) { - const char *name = vlan_strings->names[i]; - int vlan; - if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) { - VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s", - m->bridge->name, m->name, name); + for (i = 0; i < cfg->n_select_vlan; i++) { + int64_t vlan = cfg->select_vlan[i]; + if (vlan < 0 || vlan > 4095) { + VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64, + m->bridge->name, m->name, vlan); } else { (*vlans)[n_vlans++] = vlan; } @@ -3358,104 +4659,92 @@ port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) } static void -mirror_reconfigure_one(struct mirror *m) +mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) { - char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name); - struct svec src_ports, dst_ports, ports; - struct svec vlan_strings; + struct shash src_ports, dst_ports; mirror_mask_t mirror_bit; - const char *out_port_name; struct port *out_port; int out_vlan; size_t n_vlans; int *vlans; size_t i; - bool mirror_all_ports; - bool any_ports_specified; + + /* Set name. */ + if (strcmp(cfg->name, m->name)) { + free(m->name); + m->name = xstrdup(cfg->name); + } /* Get output port. */ - out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port", - m->bridge->name, m->name); - if (out_port_name) { - out_port = port_lookup(m->bridge, out_port_name); + if (cfg->output_port) { + out_port = port_lookup(m->bridge, cfg->output_port->name); if (!out_port) { - VLOG_ERR("%s.output.port: bridge %s does not have a port " - "named %s", pfx, m->bridge->name, out_port_name); + VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge", + m->bridge->name, m->name); mirror_destroy(m); - free(pfx); return; } out_vlan = -1; - if (cfg_has("%s.output.vlan", pfx)) { - VLOG_ERR("%s.output.port and %s.output.vlan both specified; " - "ignoring %s.output.vlan", pfx, pfx, pfx); + if (cfg->output_vlan) { + VLOG_ERR("bridge %s: mirror %s specifies both output port and " + "output vlan; ignoring output vlan", + m->bridge->name, m->name); } - } else if (cfg_has("%s.output.vlan", pfx)) { + } else if (cfg->output_vlan) { out_port = NULL; - out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx); + out_vlan = *cfg->output_vlan; } else { - VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, " - "but exactly one is required; disabling port mirror %s", - pfx, pfx, pfx, pfx); + VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring", + m->bridge->name, m->name); mirror_destroy(m); - free(pfx); return; } - /* Get all the ports, and drop duplicates and ports that don't exist. */ - svec_init(&src_ports); - svec_init(&dst_ports); - svec_init(&ports); - cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx); - cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx); - cfg_get_all_keys(&ports, "%s.select.port", pfx); - any_ports_specified = src_ports.n || dst_ports.n || ports.n; - svec_append(&src_ports, &ports); - svec_append(&dst_ports, &ports); - svec_destroy(&ports); - prune_ports(m, &src_ports); - prune_ports(m, &dst_ports); - if (any_ports_specified && !src_ports.n && !dst_ports.n) { - VLOG_ERR("%s: none of the specified ports exist; " - "disabling port mirror %s", pfx, pfx); - mirror_destroy(m); - goto exit; - } + shash_init(&src_ports); + shash_init(&dst_ports); + if (cfg->select_all) { + for (i = 0; i < m->bridge->n_ports; i++) { + const char *name = m->bridge->ports[i]->name; + shash_add_once(&src_ports, name, NULL); + shash_add_once(&dst_ports, name, NULL); + } + vlans = NULL; + n_vlans = 0; + } else { + /* Get ports, and drop duplicates and ports that don't exist. */ + mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port, + &src_ports); + mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port, + &dst_ports); - /* Get all the vlans, and drop duplicate and invalid vlans. */ - svec_init(&vlan_strings); - cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx); - n_vlans = prune_vlans(m, &vlan_strings, &vlans); - svec_destroy(&vlan_strings); + /* Get all the vlans, and drop duplicate and invalid vlans. */ + n_vlans = mirror_collect_vlans(m, cfg, &vlans); + } /* Update mirror data. */ - if (!svec_equal(&m->src_ports, &src_ports) - || !svec_equal(&m->dst_ports, &dst_ports) + if (!shash_equal_keys(&m->src_ports, &src_ports) + || !shash_equal_keys(&m->dst_ports, &dst_ports) || m->n_vlans != n_vlans || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans) || m->out_port != out_port || m->out_vlan != out_vlan) { bridge_flush(m->bridge); } - svec_swap(&m->src_ports, &src_ports); - svec_swap(&m->dst_ports, &dst_ports); + shash_swap(&m->src_ports, &src_ports); + shash_swap(&m->dst_ports, &dst_ports); free(m->vlans); m->vlans = vlans; m->n_vlans = n_vlans; m->out_port = out_port; m->out_vlan = out_vlan; - /* If no selection criteria have been given, mirror for all ports. */ - mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans); - /* Update ports. */ mirror_bit = MIRROR_MASK_C(1) << m->idx; for (i = 0; i < m->bridge->n_ports; i++) { struct port *port = m->bridge->ports[i]; - if (mirror_all_ports - || svec_contains(&m->src_ports, port->name) + if (shash_find(&m->src_ports, port->name) || (m->n_vlans && (!port->vlan ? port_trunks_any_mirrored_vlan(m, port) @@ -3465,7 +4754,7 @@ mirror_reconfigure_one(struct mirror *m) port->src_mirrors &= ~mirror_bit; } - if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) { + if (shash_find(&m->dst_ports, port->name)) { port->dst_mirrors |= mirror_bit; } else { port->dst_mirrors &= ~mirror_bit; @@ -3473,217 +4762,6 @@ mirror_reconfigure_one(struct mirror *m) } /* Clean up. */ -exit: - svec_destroy(&src_ports); - svec_destroy(&dst_ports); - free(pfx); -} - -/* Spanning tree protocol. */ - -static void brstp_update_port_state(struct port *); - -static void -brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_) -{ - struct bridge *br = br_; - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - struct iface *iface = iface_from_dp_ifidx(br, port_no); - if (!iface) { - VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", - br->name, port_no); - } else { - struct eth_header *eth = pkt->l2; - - netdev_get_etheraddr(iface->netdev, eth->eth_src); - if (eth_addr_is_zero(eth->eth_src)) { - VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " - "with unknown MAC", br->name, port_no); - } else { - union ofp_action action; - flow_t flow; - - memset(&action, 0, sizeof action); - action.type = htons(OFPAT_OUTPUT); - action.output.len = htons(sizeof action); - action.output.port = htons(port_no); - - flow_extract(pkt, ODPP_NONE, &flow); - ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt); - } - } - ofpbuf_delete(pkt); -} - -static void -brstp_reconfigure(struct bridge *br) -{ - size_t i; - - if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) { - if (br->stp) { - stp_destroy(br->stp); - br->stp = NULL; - - bridge_flush(br); - } - } else { - uint64_t bridge_address, bridge_id; - int bridge_priority; - - bridge_address = cfg_get_mac(0, "stp.%s.address", br->name); - if (!bridge_address) { - if (br->stp) { - bridge_address = (stp_get_bridge_id(br->stp) - & ((UINT64_C(1) << 48) - 1)); - } else { - uint8_t mac[ETH_ADDR_LEN]; - eth_addr_random(mac); - bridge_address = eth_addr_to_uint64(mac); - } - } - - if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority", - br->name)) { - bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name); - } else { - bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY; - } - - bridge_id = bridge_address | ((uint64_t) bridge_priority << 48); - if (!br->stp) { - br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br); - br->stp_last_tick = time_msec(); - bridge_flush(br); - } else { - if (bridge_id != stp_get_bridge_id(br->stp)) { - stp_set_bridge_id(br->stp, bridge_id); - bridge_flush(br); - } - } - - for (i = 0; i < br->n_ports; i++) { - struct port *p = br->ports[i]; - int dp_ifidx; - struct stp_port *sp; - int path_cost, priority; - bool enable; - - if (!p->n_ifaces) { - continue; - } - dp_ifidx = p->ifaces[0]->dp_ifidx; - if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) { - continue; - } - - sp = stp_get_port(br->stp, dp_ifidx); - enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED, - "stp.%s.port.%s.enabled", - br->name, p->name) - || cfg_get_bool(0, "stp.%s.port.%s.enabled", - br->name, p->name)); - if (p->is_mirror_output_port) { - enable = false; - } - if (enable != (stp_port_get_state(sp) != STP_DISABLED)) { - bridge_flush(br); /* Might not be necessary. */ - if (enable) { - stp_port_enable(sp); - } else { - stp_port_disable(sp); - } - } - - path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost", - br->name, p->name); - stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */); - - priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED, - "stp.%s.port.%s.priority", - br->name, p->name) - ? cfg_get_int(0, "stp.%s.port.%s.priority", - br->name, p->name) - : STP_DEFAULT_PORT_PRIORITY); - stp_port_set_priority(sp, priority); - } - - brstp_adjust_timers(br); - } - for (i = 0; i < br->n_ports; i++) { - brstp_update_port_state(br->ports[i]); - } -} - -static void -brstp_update_port_state(struct port *p) -{ - struct bridge *br = p->bridge; - enum stp_state state; - - /* Figure out new state. */ - state = STP_DISABLED; - if (br->stp && p->n_ifaces > 0) { - int dp_ifidx = p->ifaces[0]->dp_ifidx; - if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) { - state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx)); - } - } - - /* Update state. */ - if (p->stp_state != state) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10); - VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s", - p->name, stp_state_name(p->stp_state), - stp_state_name(state)); - if (p->stp_state == STP_DISABLED) { - bridge_flush(br); - } else { - ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag); - } - p->stp_state = state; - p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0 - : tag_create_random()); - } -} - -static void -brstp_adjust_timers(struct bridge *br) -{ - int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name); - int max_age = cfg_get_int(0, "stp.%s.max-age", br->name); - int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name); - - stp_set_hello_time(br->stp, hello_time ? hello_time : 2000); - stp_set_max_age(br->stp, max_age ? max_age : 20000); - stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000); -} - -static void -brstp_run(struct bridge *br) -{ - if (br->stp) { - long long int now = time_msec(); - long long int elapsed = now - br->stp_last_tick; - struct stp_port *sp; - - if (elapsed > 0) { - stp_tick(br->stp, MIN(INT_MAX, elapsed)); - br->stp_last_tick = now; - } - while (stp_get_changed_port(br->stp, &sp)) { - struct port *p = port_from_dp_ifidx(br, stp_port_no(sp)); - if (p) { - brstp_update_port_state(p); - } - } - } -} - -static void -brstp_wait(struct bridge *br) -{ - if (br->stp) { - poll_timer_wait(1000); - } + shash_destroy(&src_ports); + shash_destroy(&dst_ports); }