X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=8314c5392bd2b73c8e332bc1aab336ac551c045d;hb=f915f1a8ca180828983ef22cf2fd21b8f010b972;hp=d7f4d32893fbed64a630e42b62b243e5bf3e4fd0;hpb=e9e2856e084383b4f93b371dfba8320b51d4cd03;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index d7f4d3289..bb67e17fc 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 @@ -30,46 +32,72 @@ #include #include #include "bitmap.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 "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); + +enum lacp_status { + LACP_STATUS_CURRENT, /* Partner is up to date. */ + LACP_STATUS_EXPIRED, /* Partner is out of date. Attempt to re-sync. */ + LACP_STATUS_DEFAULTED /* Partner information is unknown. */ +}; struct dst { uint16_t vlan; uint16_t dp_ifidx; }; +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. */ struct port *port; /* Containing port. */ @@ -79,13 +107,25 @@ 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? */ - - /* This member is only valid *during* bridge_reconfigure(). */ + bool up; /* Is the interface up? */ + const char *type; /* Usually same as cfg->type. */ + struct cfm *cfm; /* Connectivity Fault Management */ const struct ovsrec_interface *cfg; + + /* LACP information. */ + enum lacp_status lacp_status; /* LACP state machine status. */ + uint16_t lacp_priority; /* LACP port priority. */ + struct lacp_info lacp_actor; /* LACP actor information. */ + struct lacp_info lacp_partner; /* LACP partner information. */ + long long int lacp_tx; /* Next LACP message transmission time. */ + long long int lacp_rx; /* Next LACP message receive time. */ + bool lacp_attached; /* Attached to its aggregator? LACP allows + this link to be chosen for flows. */ }; #define BOND_MASK 0xff @@ -95,6 +135,12 @@ struct bond_entry { tag_type iface_tag; /* Tag associated with iface_idx. */ }; +enum bond_mode { + BM_TCP, /* Transport Layer Load Balance. */ + BM_SLB, /* Source Load Balance. */ + BM_AB /* Active Backup. */ +}; + #define MAX_MIRRORS 32 typedef uint32_t mirror_mask_t; #define MIRROR_MASK_C(X) UINT32_C(X) @@ -103,10 +149,11 @@ 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; @@ -115,12 +162,19 @@ struct mirror { int out_vlan; }; +/* Flags for a port's lacp member. */ +#define LACP_ACTIVE 0x01 /* LACP is in active mode. */ +#define LACP_PASSIVE 0x02 /* LACP is in passive mode. */ +#define LACP_NEGOTIATED 0x04 /* LACP has successfully negotiated. */ + #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ struct port { struct bridge *bridge; size_t port_idx; 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. @@ -129,79 +183,90 @@ 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. */ + + /* LACP information. */ + int lacp; /* LACP status flags. 0 if LACP is off. */ + uint16_t lacp_key; /* LACP aggregation key. */ + uint16_t lacp_priority; /* LACP system priority. */ + bool lacp_need_update; /* Need to update attached interfaces? */ + + /* SLB specific bonding info. */ + struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ + int bond_rebalance_interval; /* Interval between rebalances, in ms. */ + long long int bond_next_rebalance; /* Next rebalancing time. */ /* Port mirroring info. */ mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ bool is_mirror_output_port; /* Does port mirroring send frames here? */ - - /* This member is only valid *during* bridge_reconfigure(). */ - const struct ovsrec_port *cfg; }; -#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. */ - bool sent_config_request; /* Successfully sent config request? */ + uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */ uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */ - - /* Support for remote controllers. */ - char *controller; /* NULL if there is no remote controller; - * "discover" to do controller discovery; - * otherwise a vconn name. */ + const struct ovsrec_bridge *cfg; /* OpenFlow switch processing. */ struct ofproto *ofproto; /* OpenFlow switch. */ /* Kernel datapath information. */ struct dpif *dpif; /* Datapath. */ - struct port_array ifaces; /* Indexed by kernel datapath port number. */ + 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]; - - /* This member is only valid *during* bridge_reconfigure(). */ - const struct ovsrec_bridge *cfg; }; /* 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 unixctl_cb_func bridge_unixctl_dump_flows; +static unixctl_cb_func bridge_unixctl_reconnect; static int bridge_run_one(struct bridge *); -static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *, - struct bridge *); -static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *, - struct bridge *); +static size_t bridge_get_controllers(const struct bridge *br, + struct ovsrec_controller ***controllersp); +static void bridge_reconfigure_one(struct bridge *); +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 *); @@ -216,6 +281,10 @@ static uint64_t dpid_from_hash(const void *, size_t nbytes); static unixctl_cb_func bridge_unixctl_fdb_show; +static void lacp_run(struct bridge *); +static void lacp_wait(struct bridge *); +static void lacp_process_packet(const struct ofpbuf *, struct iface *); + static void bond_init(void); static void bond_run(struct bridge *); static void bond_wait(struct bridge *); @@ -225,6 +294,7 @@ static void bond_enable_slave(struct iface *iface, bool enable); 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); @@ -233,251 +303,163 @@ static struct port *port_from_dp_ifidx(const struct bridge *, static void port_update_bond_compat(struct port *); static void port_update_vlan_compat(struct port *); static void port_update_bonding(struct port *); +static void port_update_lacp(struct port *); -#if 0 -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); -#else -static bool vlan_is_mirrored(const struct mirror *m UNUSED, int vlan UNUSED) -{ - return false; -} -#endif -static struct iface *iface_create(struct port *, const char *name); +static struct iface *iface_create(struct port *port, + const struct ovsrec_interface *if_cfg); static void iface_destroy(struct iface *); static struct iface *iface_lookup(const struct bridge *, const char *name); static struct iface *iface_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); -static bool iface_is_internal(const struct bridge *, const char *name); static void iface_set_mac(struct iface *); +static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport); +static void iface_update_qos(struct iface *, const struct ovsrec_qos *); +static void iface_update_cfm(struct iface *); +static void iface_refresh_cfm_stats(struct iface *iface); +static void iface_send_packet(struct iface *, struct ofpbuf *packet); +static uint8_t iface_get_lacp_state(const struct iface *); +static void iface_get_lacp_priority(struct iface *, struct lacp_info *); +static void iface_set_lacp_defaulted(struct iface *); +static void iface_set_lacp_expired(struct iface *); + +static 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); - } - } - } - } - } -} - +/* 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(const struct ovsrec_open_vswitch *cfg) +bridge_init(const char *remote) { - struct svec bridge_names; - struct svec dpif_names; - size_t i; - - unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + /* Create connection to database. */ + idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true); - svec_init(&bridge_names); - for (i = 0; i < cfg->n_bridges; i++) { - svec_add(&bridge_names, cfg->bridges[i]->name); - } - svec_sort(&bridge_names); + 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; + ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); - retval = dpif_open(dpif_name, &dpif); - if (!retval) { - struct svec all_names; - size_t j; + ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids); + ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge); - svec_init(&all_names); - dpif_get_all_names(dpif, &all_names); - for (j = 0; j < all_names.n; j++) { - if (svec_contains(&bridge_names, all_names.names[j])) { - goto found; - } - } - dpif_delete(dpif); - found: - svec_destroy(&all_names); - dpif_close(dpif); - } - } - svec_destroy(&dpif_names); + ovsdb_idl_omit_alert(idl, &ovsrec_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(cfg); -} - -#ifdef HAVE_OPENSSL -static bool -config_string_change(const char *value, char **valuep) -{ - if (value && (!*valuep || strcmp(value, *valuep))) { - free(*valuep); - *valuep = xstrdup(value); - return true; - } else { - return false; - } } -static void -bridge_configure_ssl(const struct ovsrec_ssl *ssl) +void +bridge_exit(void) { - /* 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; - - if (!ssl) { - /* XXX We can't un-set SSL settings. */ - return; - } - - if (config_string_change(ssl->private_key, &private_key_file)) { - vconn_ssl_set_private_key_file(private_key_file); - } - - if (config_string_change(ssl->certificate, &certificate_file)) { - vconn_ssl_set_certificate_file(certificate_file); - } + struct bridge *br, *next_br; - /* We assume that even if the filename hasn't changed, if the CA cert - * file has been removed, that we want to move back into - * boot-strapping mode. This opens a small security hole, because - * the old certificate will still be trusted until vSwitch is - * restarted. We may want to address this in vconn's SSL library. */ - if (config_string_change(ssl->ca_cert, &cacert_file) - || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) { - vconn_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert); + LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) { + bridge_destroy(br); } + ovsdb_idl_destroy(idl); } -#endif -/* Attempt to create the network device 'iface_name' through the netdev - * library. */ -static int -set_up_iface(const struct ovsrec_interface *iface_cfg, bool create) +/* 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_once(const struct ovsrec_open_vswitch *cfg) { - struct shash_node *node; - struct shash options; - int error; + static bool already_configured_once; + struct svec bridge_names; + struct svec dpif_names, dpif_types; size_t i; - /* If a type is not explicitly declared, then assume it's an existing - * "system" device. */ - if (iface_cfg->type[0] == '\0' || !strcmp(iface_cfg->type, "system")) { - return 0; + /* Only do this once per ovs-vswitchd run. */ + if (already_configured_once) { + return; } + already_configured_once = true; - shash_init(&options); - for (i = 0; i < iface_cfg->n_options; i++) { - shash_add(&options, iface_cfg->key_options[i], - xstrdup(iface_cfg->value_options[i])); - } + stats_timer = time_msec() + STATS_INTERVAL; - if (create) { - error = netdev_create(iface_cfg->name, iface_cfg->type, &options); - } else { - /* xxx Check to make sure that the type hasn't changed. */ - error = netdev_reconfigure(iface_cfg->name, &options); + /* 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); - SHASH_FOR_EACH (node, &options) { - free(node->data); + /* 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); + } + } + } } - shash_destroy(&options); - - return error; -} - -static int -create_iface(const struct ovsrec_interface *iface_cfg) -{ - return set_up_iface(iface_cfg, true); -} - -static int -reconfigure_iface(const struct ovsrec_interface *iface_cfg) -{ - return set_up_iface(iface_cfg, false); -} - -static void -destroy_iface(const char *iface_name) -{ - netdev_destroy(iface_name); + svec_destroy(&bridge_names); + svec_destroy(&dpif_names); + svec_destroy(&dpif_types); } - -/* iterate_and_prune_ifaces() callback function that opens the network device - * for 'iface', if it is not already open, and retrieves the interface's MAC - * address and carrier status. */ +/* 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_properties(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) { /* Set policing attributes. */ netdev_set_policing(iface->netdev, @@ -486,8 +468,7 @@ set_iface_properties(struct bridge *br UNUSED, struct iface *iface, /* Set MAC address of internal interfaces other than the local * interface. */ - if (iface->dp_ifidx != ODPP_LOCAL - && iface_is_internal(br, iface->name)) { + if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) { iface_set_mac(iface); } @@ -512,6 +493,7 @@ iterate_and_prune_ifaces(struct bridge *br, if (cb(br, iface, aux)) { j++; } else { + iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } } @@ -525,20 +507,85 @@ iterate_and_prune_ifaces(struct bridge *br, } } -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 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. */ shash_init(&old_br); shash_init(&new_br); - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { shash_add(&old_br, br->name, br); } for (i = 0; i < ovs_cfg->n_bridges; i++) { @@ -549,7 +596,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } /* Get rid of deleted bridges and add new bridges. */ - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { + 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; @@ -560,24 +607,24 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) SHASH_FOR_EACH (node, &new_br) { const char *br_name = node->name; const struct ovsrec_bridge *br_cfg = node->data; - if (!shash_find_data(&old_br, br_name)) { - br = bridge_create(br_name); - if (br) { - br->cfg = br_cfg; + 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); } } shash_destroy(&old_br); shash_destroy(&new_br); -#ifdef HAVE_OPENSSL - /* Configure SSL. */ - bridge_configure_ssl(ovs_cfg->ssl); -#endif - /* Reconfigure all bridges. */ - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - bridge_reconfigure_one(ovs_cfg, br); + LIST_FOR_EACH (br, node, &all_bridges) { + bridge_reconfigure_one(br); } /* Add and delete ports on all datapaths. @@ -585,99 +632,154 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) * 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; + 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 (!shash_find(&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)); } - destroy_iface(p->devname); } } shash_destroy(&want_ifaces); - free(dpif_ports); } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; + LIST_FOR_EACH (br, node, &all_bridges) { struct shash cur_ifaces, want_ifaces; - struct shash_node *node; + struct dpif_port_dump dump; + struct dpif_port dpif_port; /* Get the set of interfaces currently in this datapath. */ - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); shash_init(&cur_ifaces); - for (i = 0; i < n_dpif_ports; i++) { - const char *name = dpif_ports[i].devname; - if (!shash_find(&cur_ifaces, name)) { - shash_add(&cur_ifaces, name, NULL); - } + 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); /* Get the set of interfaces we want on this datapath. */ bridge_get_all_ifaces(br, &want_ifaces); + 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; + + 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; + } + } + + /* 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; - if (shash_find(&cur_ifaces, if_name)) { - /* Already exists, just reconfigure it. */ + shash_init(&args); if (iface) { - reconfigure_iface(iface->cfg); + shash_from_ovs_idl_map(iface->cfg->key_options, + iface->cfg->value_options, + iface->cfg->n_options, &args); } - } else { - /* Need to add to datapath. */ - bool internal; - int error; + error = netdev_open(&options, &netdev); + shash_destroy(&args); - /* Attempt to create the network interface in case it - * doesn't exist yet. */ - error = iface ? create_iface(iface->cfg) : 0; if (error) { - VLOG_WARN("could not create iface %s: %s\n", if_name, - strerror(error)); + VLOG_WARN("could not open network device %s (%s)", + if_name, strerror(error)); continue; } - /* Add to datapath. */ - internal = !iface || iface_is_internal(br, if_name); - error = dpif_port_add(br->dpif, if_name, - internal ? ODP_PORT_INTERNAL : 0, NULL); - if (error == EFBIG) { - VLOG_ERR("ran out of valid port numbers on %s", - dpif_name(br->dpif)); - break; - } else if (error) { - VLOG_ERR("failed to add %s interface to %s: %s", - if_name, dpif_name(br->dpif), strerror(error)); + /* 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); + iface->up = iface->enabled; } + } 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); } } - shash_destroy(&cur_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; + 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); @@ -691,10 +793,15 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) br->name, strerror(error)); } } + memcpy(br->ea, ea, ETH_ADDR_LEN); 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. */ if (br->cfg->netflow) { struct ovsrec_netflow *nf_cfg = br->cfg->netflow; @@ -704,10 +811,10 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id); if (nf_cfg->engine_type) { - opts.engine_type = nf_cfg->engine_type; + opts.engine_type = *nf_cfg->engine_type; } if (nf_cfg->engine_id) { - opts.engine_id = nf_cfg->engine_id; + opts.engine_id = *nf_cfg->engine_id; } opts.active_timeout = nf_cfg->active_timeout; @@ -737,13 +844,58 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) 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", + VLOG_ERR("bridge %s: problem setting netflow collectors", br->name); } } else { ofproto_set_netflow(br->ofproto, NULL); } + /* 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 * straightforward to call this from bridge_reconfigure_one(), but we * can't do it there for two reasons. First, and most importantly, at @@ -753,32 +905,71 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) * yet; when a controller is configured, resetting the datapath ID will * immediately disconnect from the controller, so it's better to set * the datapath ID before the controller. */ - bridge_reconfigure_controller(ovs_cfg, br); + 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); + port_update_lacp(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) { + 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) { + const char *hwaddr; size_t i, j; int error; *hw_addr_iface = NULL; /* Did the user request a particular MAC? */ - if (br->cfg->hwaddr && eth_addr_from_string(br->cfg->hwaddr, 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)); @@ -789,9 +980,8 @@ 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]; @@ -832,11 +1022,8 @@ 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 - || !strcmp(iface->cfg->type, "internal")) { + * MAC address anyway). */ + if (iface->dp_ifidx == ODPP_LOCAL) { continue; } @@ -852,15 +1039,16 @@ 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 " @@ -893,10 +1081,11 @@ 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; - if (br->cfg->datapath_id - && dpid_from_string(br->cfg->datapath_id, &dpid)) { + datapath_id = bridge_get_other_config(br->cfg, "datapath-id"); + if (datapath_id && dpid_from_string(datapath_id, &dpid)) { return dpid; } @@ -961,58 +1150,381 @@ dpid_from_hash(const void *data, size_t n) return eth_addr_to_uint64(hash); } -int -bridge_run(void) +static void +iface_refresh_status(struct iface *iface) { - struct bridge *br, *next; - int retval; + struct shash sh; - retval = 0; - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { - int error = bridge_run_one(br); - if (error) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, " - "forcing reconfiguration", br->name); - if (!retval) { - retval = error; - } - } + enum netdev_flags flags; + uint32_t current; + int64_t bps; + int mtu; + int64_t mtu_64; + int error; + + shash_init(&sh); + + if (!netdev_get_status(iface->netdev, &sh)) { + size_t n; + char **keys, **values; + + shash_to_ovs_idl_map(&sh, &keys, &values, &n); + ovsrec_interface_set_status(iface->cfg, keys, values, n); + + free(keys); + free(values); + } else { + ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0); } - return retval; -} -void -bridge_wait(void) -{ - struct bridge *br; + shash_destroy_free_data(&sh); - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - ofproto_wait(br->ofproto); - if (br->controller) { - continue; - } + 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); + } - mac_learning_wait(br->ml); - bond_wait(br); + 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 != INT_MAX) { + mtu_64 = mtu; + ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1); + } + else { + ovsrec_interface_set_mtu(iface->cfg, NULL, 0); } } -/* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s - * configuration changes. */ static void -bridge_flush(struct bridge *br) +iface_refresh_cfm_stats(struct iface *iface) { - COVERAGE_INC(bridge_flush); - br->flush = true; - mac_learning_flush(br->ml); -} + size_t i; + struct cfm *cfm; + const struct ovsrec_monitor *mon; -/* 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) -{ + mon = iface->cfg->monitor; + cfm = iface->cfm; + + if (!cfm || !mon) { + return; + } + + 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); + } + + if (hmap_is_empty(&cfm->x_remote_mps)) { + ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0); + } else { + size_t length; + struct remote_mp *rmp; + int64_t *x_remote_mps; + + length = hmap_count(&cfm->x_remote_mps); + x_remote_mps = xzalloc(length * sizeof *x_remote_mps); + + i = 0; + HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) { + x_remote_mps[i++] = rmp->mpid; + } + + ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length); + free(x_remote_mps); + } + + if (hmap_is_empty(&cfm->x_remote_maids)) { + ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0); + } else { + size_t length; + char **x_remote_maids; + struct remote_maid *rmaid; + + length = hmap_count(&cfm->x_remote_maids); + x_remote_maids = xzalloc(length * sizeof *x_remote_maids); + + 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); + lacp_wait(br); + 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++) { @@ -1031,7 +1543,7 @@ bridge_get_local_iface(struct bridge *br) /* Bridge unixctl user interface functions. */ static void bridge_unixctl_fdb_show(struct unixctl_conn *conn, - const char *args, void *aux UNUSED) + const char *args, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct bridge *br; @@ -1044,7 +1556,7 @@ bridge_unixctl_fdb_show(struct unixctl_conn *conn, } ds_put_cstr(&ds, " port VLAN MAC Age\n"); - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { if (e->port < 0 || e->port >= br->n_ports) { continue; } @@ -1057,41 +1569,45 @@ bridge_unixctl_fdb_show(struct unixctl_conn *conn, } /* Bridge reconfiguration functions. */ - static struct bridge * -bridge_create(const char *name) +bridge_create(const struct ovsrec_bridge *br_cfg) { struct bridge *br; int error; - assert(!bridge_lookup(name)); + assert(!bridge_lookup(br_cfg->name)); br = xzalloc(sizeof *br); - error = dpif_create_and_open(name, &br->dpif); + error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type, + &br->dpif); if (error) { free(br); return NULL; } dpif_flow_flush(br->dpif); - error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto); + error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks, + br, &br->ofproto); if (error) { - VLOG_ERR("failed to create switch %s: %s", name, strerror(error)); + 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); @@ -1117,9 +1633,10 @@ 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); @@ -1131,7 +1648,7 @@ 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; } @@ -1139,28 +1656,15 @@ 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, void *aux UNUSED) + 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"); @@ -1174,10 +1678,34 @@ bridge_unixctl_dump_flows(struct unixctl_conn *conn, 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) { @@ -1185,57 +1713,58 @@ bridge_run_one(struct bridge *br) } mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); + lacp_run(br); bond_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 struct ovsrec_controller * -bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg, - const struct bridge *br) +static size_t +bridge_get_controllers(const struct bridge *br, + struct ovsrec_controller ***controllersp) { - const struct ovsrec_controller *controller; + struct ovsrec_controller **controllers; + size_t n_controllers; - controller = (br->cfg->controller ? br->cfg->controller - : ovs_cfg->controller ? ovs_cfg->controller - : NULL); + controllers = br->cfg->controller; + n_controllers = br->cfg->n_controller; - if (controller && !strcmp(controller->target, "none")) { - return NULL; + if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) { + controllers = NULL; + n_controllers = 0; } - return controller; -} - -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(const struct ovsrec_open_vswitch *ovs_cfg, - struct bridge *br) +bridge_reconfigure_one(struct bridge *br) { struct shash old_ports, new_ports; - struct svec ifaces; - struct svec listeners, old_listeners; struct svec snoops, old_snoops; struct shash_node *node; - uint64_t mgmt_id; + enum ofproto_fail_mode fail_mode; size_t i; /* Collect old ports. */ @@ -1253,104 +1782,76 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, br->name, name); } } - if (bridge_get_controller(ovs_cfg, br)) { + + /* 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) { - shash_add_once(&new_ports, local_name, NULL); + 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); } } - dpid_from_string(ovs_cfg->management_id, &mgmt_id); - ofproto_set_mgmt_id(br->ofproto, mgmt_id); - - /* Get rid of deleted ports and add new ports. */ + /* Get rid of deleted ports. + * Get rid of deleted interfaces on ports that still exist. */ SHASH_FOR_EACH (node, &old_ports) { - if (!shash_find(&new_ports, node->name)) { - port_destroy(node->data); + struct port *port = node->data; + const struct ovsrec_port *port_cfg; + + port_cfg = shash_find_data(&new_ports, node->name); + if (!port_cfg) { + port_destroy(port); + } else { + port_del_ifaces(port, port_cfg); } } + + /* Create new ports. + * Add new interfaces to existing ports. + * Reconfigure existing ports. */ SHASH_FOR_EACH (node, &new_ports) { struct port *port = shash_find_data(&old_ports, node->name); if (!port) { port = port_create(br, node->name); } + port_reconfigure(port, node->data); + if (!port->n_ifaces) { + VLOG_WARN("bridge %s: port %s has no interfaces, dropping", + br->name, port->name); + port_destroy(port); + } } shash_destroy(&old_ports); shash_destroy(&new_ports); - /* 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?) */ -#if 0 - /* Configure OpenFlow management listeners. */ - svec_init(&listeners); - cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name); - if (!listeners.n) { - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) { - svec_clear(&listeners); - } - svec_sort_unique(&listeners); - - 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_init(&old_snoops); - ofproto_get_snoops(br->ofproto, &old_snoops); - svec_sort_unique(&old_snoops); - - if (!svec_equal(&snoops, &old_snoops)) { - ofproto_set_snoops(br->ofproto, &snoops); - } - svec_destroy(&snoops); - svec_destroy(&old_snoops); -#else - /* Default listener. */ - svec_init(&listeners); - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); - } - svec_destroy(&listeners); - svec_destroy(&old_listeners); - - /* Default snoop. */ - svec_init(&snoops); svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); + ovs_rundir(), br->name)); svec_init(&old_snoops); ofproto_get_snoops(br->ofproto, &old_snoops); if (!svec_equal(&snoops, &old_snoops)) { @@ -1358,116 +1859,183 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, } svec_destroy(&snoops); svec_destroy(&old_snoops); -#endif -#if 0 mirror_reconfigure(br); -#endif } +/* 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(const struct ovsrec_open_vswitch *ovs_cfg, - 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 struct ovsrec_controller *c; + 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; +} - c = bridge_get_controller(ovs_cfg, br); - if ((br->controller != NULL) != (c != 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 = c ? xstrdup(c->target) : NULL; - if (c) { - 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(c->target, "discover")) { - ofproto_set_discovery(br->ofproto, true, - c->discover_accept_regex, - c->discover_update_resolv_conf); - } else { - struct iface *local_iface; - struct in_addr ip; - bool in_band; - - in_band = (!c->connection_mode - || !strcmp(c->connection_mode, "out-of-band")); - ofproto_set_discovery(br->ofproto, false, NULL, NULL); - ofproto_set_in_band(br->ofproto, in_band); - - local_iface = bridge_get_local_iface(br); - if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) { - struct netdev *netdev = local_iface->netdev; - struct in_addr ip, mask, gateway; - - if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) { - mask.s_addr = 0; - } - if (!c->local_gateway - || !inet_aton(c->local_gateway, &gateway)) { - gateway.s_addr = 0; - } + /* Bring up the local interface. */ + netdev = local_iface->netdev; + netdev_turn_flags_on(netdev, NETDEV_UP, true); - 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)); } + } +} + +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; - ofproto_set_failure(br->ofproto, - (!c->fail_mode - || !strcmp(c->fail_mode, "standalone") - || !strcmp(c->fail_mode, "open"))); + struct ovsrec_controller **controllers; + size_t n_controllers; + bool had_primary; - probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; - ofproto_set_probe_interval(br->ofproto, probe); + struct ofproto_controller *ocs; + size_t n_ocs; + size_t i; - max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; - ofproto_set_max_backoff(br->ofproto, max_backoff); + /* 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; + } - rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; - burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0; - ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit); + /* 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); - ofproto_set_remote_execution(br->ofproto, NULL, NULL); /* XXX */ + 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); + + n_controllers = bridge_get_controllers(br, &controllers); + + ocs = xmalloc((n_controllers + 1) * sizeof *ocs); + n_ocs = 0; + + 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; + } + + 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); + 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 @@ -1499,8 +2067,8 @@ bridge_get_all_ifaces(const struct bridge *br, struct shash *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. */ @@ -1511,40 +2079,69 @@ 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 bool +bond_is_tcp_hash(const struct port *port) +{ + return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED; +} + static int -bond_hash(const uint8_t mac[ETH_ADDR_LEN]) +bond_hash_src(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 int bond_hash_tcp(const struct flow *flow, uint16_t vlan) +{ + struct flow hash_flow; + + memcpy(&hash_flow, flow, sizeof hash_flow); + hash_flow.vlan_tci = 0; + + /* The symmetric quality of this hash function is not required, but + * flow_hash_symmetric_l4 already exists, and is sufficient for our + * purposes, so we use it out of convenience. */ + return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK; } 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 struct flow *flow, + uint16_t vlan) { - return &port->bond_hash[bond_hash(mac)]; + assert(port->bond_mode != BM_AB); + + if (bond_is_tcp_hash(port)) { + return &port->bond_hash[bond_hash_tcp(flow, vlan)]; + } else { + return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)]; + } } static int @@ -1559,7 +2156,9 @@ bond_choose_iface(const struct port *port) if (iface->enabled) { return i; - } else if (iface->delay_expires < next_delay_expiration) { + } else if (iface->delay_expires < next_delay_expiration + && (iface->lacp_attached + || !(port->lacp & LACP_NEGOTIATED))) { best_down_slave = i; next_delay_expiration = iface->delay_expires; } @@ -1578,16 +2177,22 @@ bond_choose_iface(const struct port *port) } static bool -choose_output_iface(const struct port *port, const uint8_t *dl_src, - uint16_t *dp_ifidx, tag_type *tags) +choose_output_iface(const struct port *port, const struct flow *flow, + uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags) { struct iface *iface; assert(port->n_ifaces); if (port->n_ifaces == 1) { iface = port->ifaces[0]; + } else if (port->bond_mode == BM_AB) { + if (port->active_iface < 0) { + *tags |= port->no_ifaces_tag; + return false; + } + iface = port->ifaces[port->active_iface]; } else { - struct bond_entry *e = lookup_bond_entry(port, dl_src); + struct bond_entry *e = lookup_bond_entry(port, flow, vlan); if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces || !port->ifaces[e->iface_idx]->enabled) { /* XXX select interface properly. The current interface selection @@ -1609,36 +2214,58 @@ choose_output_iface(const struct port *port, const uint8_t *dl_src, } static void -bond_link_status_update(struct iface *iface, bool carrier) +bond_link_status_update(struct iface *iface) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); struct port *port = iface->port; + bool up = iface->up; + int updelay, downdelay; - if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { + updelay = port->updelay; + downdelay = port->downdelay; + + if (iface->port->lacp & LACP_NEGOTIATED) { + downdelay = 0; + updelay = 0; + } + + if (iface->port->lacp && up) { + /* The interface is up if it's attached to an aggregator and its + * partner is synchronized. The only exception is defaulted links. + * They are not required to have synchronized partners because they + * have no partners at all. However, they will only be attached if + * negotiations failed on all interfaces in the bond. */ + up = iface->lacp_attached + && (iface->lacp_partner.state & LACP_STATE_SYNC + || iface->lacp_status == LACP_STATUS_DEFAULTED); + } + + + if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { /* Nothing to do. */ return; } - VLOG_INFO_RL(&rl, "interface %s: carrier %s", - iface->name, carrier ? "detected" : "dropped"); - if (carrier == iface->enabled) { + VLOG_INFO_RL(&rl, "interface %s: link state %s", + iface->name, up ? "up" : "down"); + if (up == iface->enabled) { iface->delay_expires = LLONG_MAX; VLOG_INFO_RL(&rl, "interface %s: will not be %s", - iface->name, carrier ? "disabled" : "enabled"); - } else if (carrier && port->active_iface < 0) { + iface->name, up ? "disabled" : "enabled"); + } else if (up && port->active_iface < 0) { bond_enable_slave(iface, true); - if (port->updelay) { + if (updelay) { VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " - "other interface is up", iface->name, port->updelay); + "other interface is up", iface->name, updelay); } } else { - int delay = carrier ? port->updelay : port->downdelay; + int delay = up ? updelay : downdelay; iface->delay_expires = time_msec() + delay; if (delay) { VLOG_INFO_RL(&rl, "interface %s: will be %s if it stays %s for %d ms", iface->name, - carrier ? "enabled" : "disabled", - carrier ? "up" : "down", + up ? "enabled" : "disabled", + up ? "up" : "down", delay); } } @@ -1709,6 +2336,59 @@ bond_enable_slave(struct iface *iface, bool enable) port->bond_compat_is_stale = true; } +/* Attempts to make the sum of the bond slaves' statistics appear on the fake + * bond interface. */ +static void +bond_update_fake_iface_stats(struct port *port) +{ + struct netdev_stats bond_stats; + struct netdev *bond_dev; + size_t i; + + memset(&bond_stats, 0, sizeof bond_stats); + + for (i = 0; i < port->n_ifaces; i++) { + struct netdev_stats slave_stats; + + if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) { + /* XXX: We swap the stats here because they are swapped back when + * reported by the internal device. The reason for this is + * internal devices normally represent packets going into the system + * but when used as fake bond device they represent packets leaving + * the system. We really should do this in the internal device + * itself because changing it here reverses the counts from the + * perspective of the switch. However, the internal device doesn't + * know what type of device it represents so we have to do it here + * for now. */ + bond_stats.tx_packets += slave_stats.rx_packets; + bond_stats.tx_bytes += slave_stats.rx_bytes; + bond_stats.rx_packets += slave_stats.tx_packets; + bond_stats.rx_bytes += slave_stats.tx_bytes; + } + } + + if (!netdev_open_default(port->name, &bond_dev)) { + netdev_set_stats(bond_dev, &bond_stats); + netdev_close(bond_dev); + } +} + +static void +bond_link_carrier_update(struct iface *iface, bool carrier) +{ + if (carrier == iface->up) { + return; + } + + if (iface->lacp_status == LACP_STATUS_CURRENT) { + iface_set_lacp_expired(iface); + } + + iface->up = carrier; + iface->lacp_tx = 0; + iface->port->bond_compat_is_stale = true; +} + static void bond_run(struct bridge *br) { @@ -1718,12 +2398,52 @@ bond_run(struct bridge *br) struct port *port = br->ports[i]; if (port->n_ifaces >= 2) { + char *devname; + + if (port->monitor) { + assert(!port->miimon); + + /* Track carrier going up and down on interfaces. */ + while (!netdev_monitor_poll(port->monitor, &devname)) { + struct iface *iface; + + iface = port_lookup_iface(port, devname); + if (iface) { + bool up = netdev_get_carrier(iface->netdev); + bond_link_carrier_update(iface, up); + } + free(devname); + } + } else { + assert(port->miimon); + + if (time_msec() >= port->bond_miimon_next_update) { + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + bool up = netdev_get_miimon(iface->netdev); + bond_link_carrier_update(iface, up); + } + port->bond_miimon_next_update = time_msec() + + port->bond_miimon_interval; + } + } + + for (j = 0; j < port->n_ifaces; j++) { + bond_link_status_update(port->ifaces[j]); + } + for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (time_msec() >= iface->delay_expires) { bond_enable_slave(iface, !iface->enabled); } } + + if (port->bond_fake_iface + && time_msec() >= port->bond_next_fake_iface_update) { + bond_update_fake_iface_stats(port); + port->bond_next_fake_iface_update = time_msec() + 1000; + } } if (port->bond_compat_is_stale) { @@ -1743,24 +2463,38 @@ 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) { - 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, dst->vlan, + &dst->dp_ifidx, tags); } static void @@ -1777,10 +2511,10 @@ 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: @@ -1815,13 +2549,48 @@ 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; } } @@ -1831,7 +2600,8 @@ 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 @@ -1840,54 +2610,72 @@ 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, uint16_t *nf_output_iface) + 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; + 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); } } *nf_output_iface = NF_OUT_FLOOD; - } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) { - *nf_output_iface = dst->dp_ifidx; + } 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)) { - int flow_vlan; - 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; } @@ -1897,15 +2685,11 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, * tagging tags place. This is necessary because * dst->vlan is the final vlan, after removing implicit * tags. */ - flow_vlan = ntohs(flow->dl_vlan); - if (flow_vlan == 0) { - flow_vlan = OFP_VLAN_NONE; - } - if (port == in_port && dst->vlan == flow_vlan) { + if (port == in_port && dst.vlan == flow_vlan) { /* Don't send out input port on same VLAN. */ continue; } - dst++; + dst_set_add(set, &dst); } } } @@ -1913,78 +2697,77 @@ 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, - nf_output_iface); + 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 flow_t *flow, +static int flow_get_vlan(struct bridge *br, const struct flow *flow, struct port *in_port, bool have_packet) { - /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet - * belongs to VLAN 0, so we should treat both cases identically. (In the - * former case, the packet has an 802.1Q header that specifies VLAN 0, - * presumably to allow a priority to be specified. In the latter case, the - * packet does not have any 802.1Q header.) */ - int vlan = ntohs(flow->dl_vlan); - if (vlan == OFP_VLAN_NONE) { - vlan = 0; - } + 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 %"PRIu16" tagged " + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " "packet received on port %s configured with " "implicit VLAN %"PRIu16, - br->name, ntohs(flow->dl_vlan), - in_port->name, in_port->vlan); + br->name, vlan, in_port->name, in_port->vlan); } return -1; } @@ -2005,12 +2788,34 @@ static int flow_get_vlan(struct bridge *br, const flow_t *flow, 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_gratuitous_arp(const struct flow *flow) +{ + return (flow->dl_type == htons(ETH_TYPE_ARP) + && eth_addr_is_broadcast(flow->dl_dst) + && (flow->nw_proto == ARP_OP_REPLY + || (flow->nw_proto == ARP_OP_REQUEST + && flow->nw_src == flow->nw_dst))); +} + static void -update_learning_table(struct bridge *br, const flow_t *flow, int vlan, +update_learning_table(struct bridge *br, const struct flow *flow, int vlan, struct port *in_port) { - tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src, - vlan, in_port->port_idx); + 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. */ @@ -2024,33 +2829,39 @@ update_learning_table(struct bridge *br, const flow_t *flow, int vlan, } } +/* 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 -is_bcast_arp_reply(const flow_t *flow) -{ - return (flow->dl_type == htons(ETH_TYPE_ARP) - && flow->nw_proto == ARP_OP_REPLY - && eth_addr_is_broadcast(flow->dl_dst)); -} - -/* If the composed actions may be applied to any packet in the given 'flow', - * returns true. Otherwise, the actions should only be applied to 'packet', or - * not at all, if 'packet' was NULL. */ -static bool -process_flow(struct bridge *br, const flow_t *flow, - const struct ofpbuf *packet, struct odp_actions *actions, - tag_type *tags, uint16_t *nf_output_iface) +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; - int out_port_idx; /* Find the interface and port structure for the received packet. */ in_iface = iface_from_dp_ifidx(br, flow->in_port); if (!in_iface) { /* No interface? Something fishy... */ - if (packet != NULL) { + if (have_packet) { /* Odd. A few possible reasons here: * * - We deleted an interface but there are still a few packets @@ -2065,65 +2876,98 @@ 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_portp = NULL; + return false; } - in_port = in_iface->port; - vlan = flow_get_vlan(br, flow, in_port, !!packet); + *in_portp = in_port = in_iface->port; + *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet); if (vlan < 0) { - goto done; + return false; } /* Drop frames for reserved multicast addresses. */ if (eth_addr_is_reserved(flow->dl_dst)) { - goto done; + return false; } /* Drop frames on ports reserved for mirroring. */ if (in_port->is_mirror_output_port) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, " - "which is reserved exclusively for mirroring", - br->name, in_port->name); - goto done; + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " + "%s, which is reserved exclusively for mirroring", + br->name, in_port->name); + } + return false; } - /* Packets received on bonds need special attention to avoid duplicates. */ - if (in_port->n_ifaces > 1) { + /* When using LACP, do not accept packets from disabled interfaces. */ + if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) { + return false; + } + + /* Packets received on non-LACP bonds need special attention to avoid + * duplicates. */ + if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) { int src_idx; + bool is_grat_arp_locked; if (eth_addr_is_multicast(flow->dl_dst)) { *tags |= in_port->active_iface_tag; if (in_port->active_iface != in_iface->port_ifidx) { /* Drop all multicast packets on inactive slaves. */ - goto done; + return false; } } /* Drop all packets for which we have learned a different input * port, because we probably sent the packet on one slave and got - * it back on the other. Broadcast ARP replies are an exception - * to this rule: the host has moved to another switch. */ - src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); + * 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_bcast_arp_reply(flow)) { - goto done; + (!is_gratuitous_arp(flow) || is_grat_arp_locked)) { + return false; } } - /* MAC learning. */ - out_port = FLOOD_PORT; + return true; +} + +/* If the composed actions may be applied to any packet in the given 'flow', + * returns true. Otherwise, the actions should only be applied to 'packet', or + * not at all, if 'packet' was NULL. */ +static bool +process_flow(struct bridge *br, const struct flow *flow, + const struct ofpbuf *packet, struct ofpbuf *actions, + tag_type *tags, uint16_t *nf_output_iface) +{ + struct port *in_port; + struct port *out_port; + int vlan; + int out_port_idx; + + /* Check whether we should drop packets in this flow. */ + if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) { + out_port = NULL; + goto done; + } + /* Learn source MAC (but don't try to learn from revalidation). */ if (packet) { update_learning_table(br, flow, vlan, in_port); } /* Determine output port. */ - out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, - tags); + 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)) { @@ -2133,6 +2977,8 @@ process_flow(struct bridge *br, const flow_t *flow, * on a bond and blackhole packets before the learning table is * updated to reflect the correct port. */ return false; + } else { + out_port = FLOOD_PORT; } /* Don't send packets out their input ports. */ @@ -2140,129 +2986,298 @@ process_flow(struct bridge *br, const flow_t *flow, out_port = NULL; } -done: - compose_actions(br, flow, vlan, in_port, out_port, tags, actions, - nf_output_iface); +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 struct flow *flow, const struct ofpbuf *packet, + struct ofpbuf *actions, tag_type *tags, + uint16_t *nf_output_iface, void *br_) +{ + struct iface *iface; + struct bridge *br = br_; + + COVERAGE_INC(bridge_process_flow); + + iface = iface_from_dp_ifidx(br, flow->in_port); + + if (cfm_should_process_flow(flow)) { + if (packet && iface->cfm) { + cfm_process_heartbeat(iface->cfm, packet); + } + return false; + } else if (flow->dl_type == htons(ETH_TYPE_LACP)) { + if (packet) { + lacp_process_packet(packet, iface); + } + return false; + } + + return process_flow(br, flow, packet, actions, tags, nf_output_iface); +} + +static void +bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, + const struct nlattr *actions, + size_t actions_len, + unsigned long long int n_bytes, void *br_) +{ + struct bridge *br = br_; + 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; + } + 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, vlan); + e->tx_bytes += n_bytes; + } + } + } +} + +static void +bridge_account_checkpoint_ofhook_cb(void *br_) +{ + struct bridge *br = br_; + long long int now; + size_t i; + + if (!br->has_bonded_ports) { + return; + } + + now = time_msec(); + for (i = 0; i < br->n_ports; i++) { + struct port *port = br->ports[i]; + if (port->n_ifaces > 1 && port->bond_mode == BM_SLB + && now >= port->bond_next_rebalance) { + port->bond_next_rebalance = now + port->bond_rebalance_interval; + bond_rebalance_port(port); + } + } +} + +static struct ofhooks bridge_ofhooks = { + bridge_normal_ofhook_cb, + bridge_account_flow_ofhook_cb, + bridge_account_checkpoint_ofhook_cb, +}; + +/* LACP functions. */ + +static void +lacp_process_packet(const struct ofpbuf *packet, struct iface *iface) +{ + const struct lacp_pdu *pdu; + + if (!iface->port->lacp) { + return; + } + + pdu = parse_lacp_packet(packet); + if (!pdu) { + return; + } + + iface->lacp_status = LACP_STATUS_CURRENT; + iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX; - return true; + iface->lacp_actor.state = iface_get_lacp_state(iface); + if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) { + iface->lacp_tx = 0; + } + + if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) { + iface->port->lacp_need_update = true; + iface->lacp_partner = pdu->actor; + } } -/* 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_) +lacp_update_ifaces(struct port *port) { - struct bridge *br = br_; - struct iface *iface; - struct port *port; + size_t i; + struct iface *lead; + struct lacp_info lead_pri; - iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no)); - if (!iface) { + port->lacp_need_update = false; + + if (!port->lacp) { return; } - 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); + lead = NULL; + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + struct lacp_info pri; + + iface->lacp_attached = true; + ofproto_revalidate(port->bridge->ofproto, iface->tag); + + /* Don't allow loopback interfaces to send traffic or lead. */ + if (eth_addr_equals(iface->lacp_partner.sysid, + iface->lacp_actor.sysid)) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); + VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is " + "connected to its own bridge", iface->name); + iface->lacp_attached = false; + continue; } - bridge_flush(br); - } else { - if (port->n_ifaces > 1) { - bool up = !(opp->state & OFPPS_LINK_DOWN); - bond_link_status_update(iface, up); - port_update_bond_compat(port); + if (iface->lacp_status == LACP_STATUS_DEFAULTED) { + continue; + } + + iface_get_lacp_priority(iface, &pri); + + if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) { + lead = iface; + lead_pri = pri; + } + } + + if (!lead) { + port->lacp &= ~LACP_NEGOTIATED; + return; + } + + port->lacp |= LACP_NEGOTIATED; + + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + + if (iface->lacp_status == LACP_STATUS_DEFAULTED + || lead->lacp_partner.key != iface->lacp_partner.key + || !eth_addr_equals(lead->lacp_partner.sysid, + iface->lacp_partner.sysid)) { + iface->lacp_attached = false; } } } static bool -bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, - struct odp_actions *actions, tag_type *tags, - uint16_t *nf_output_iface, void *br_) +lacp_iface_may_tx(const struct iface *iface) { - struct bridge *br = br_; - - COVERAGE_INC(bridge_process_flow); - return process_flow(br, flow, packet, actions, tags, nf_output_iface); + return iface->port->lacp & LACP_ACTIVE + || iface->lacp_status != LACP_STATUS_DEFAULTED; } 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_) +lacp_run(struct bridge *br) { - struct bridge *br = br_; - struct port *in_port; - const union odp_action *a; + size_t i, j; + struct ofpbuf packet; - /* Feed information from the active flows back into the learning table - * to ensure that table is always in sync with what is actually flowing - * through the datapath. */ - in_port = port_from_dp_ifidx(br, flow->in_port); - if (in_port) { - int vlan = flow_get_vlan(br, flow, in_port, false); - if (vlan >= 0) { - update_learning_table(br, flow, vlan, in_port); + ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN); + + for (i = 0; i < br->n_ports; i++) { + struct port *port = br->ports[i]; + + if (!port->lacp) { + continue; } - } - if (!br->has_bonded_ports) { - return; - } + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; - for (a = actions; a < &actions[n_actions]; a++) { - if (a->type == ODPAT_OUTPUT) { - struct port *out_port = port_from_dp_ifidx(br, a->output.port); - if (out_port && out_port->n_ifaces >= 2) { - struct bond_entry *e = lookup_bond_entry(out_port, - flow->dl_src); - e->tx_bytes += n_bytes; + if (time_msec() > iface->lacp_rx) { + if (iface->lacp_status == LACP_STATUS_CURRENT) { + iface_set_lacp_expired(iface); + } else if (iface->lacp_status == LACP_STATUS_EXPIRED) { + iface_set_lacp_defaulted(iface); + } + } + } + + if (port->lacp_need_update) { + lacp_update_ifaces(port); + } + + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + uint8_t ea[ETH_ADDR_LEN]; + int error; + + if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) { + continue; + } + + error = netdev_get_etheraddr(iface->netdev, ea); + if (!error) { + iface->lacp_actor.state = iface_get_lacp_state(iface); + compose_lacp_packet(&packet, &iface->lacp_actor, + &iface->lacp_partner, ea); + iface_send_packet(iface, &packet); + } else { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); + VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address " + "(%s)", iface->name, strerror(error)); } + + iface->lacp_tx = time_msec() + + (iface->lacp_partner.state & LACP_STATE_TIME + ? LACP_FAST_TIME_TX + : LACP_SLOW_TIME_TX); } } + ofpbuf_uninit(&packet); } static void -bridge_account_checkpoint_ofhook_cb(void *br_) +lacp_wait(struct bridge *br) { - struct bridge *br = br_; - 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; + size_t i, j; for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - if (port->n_ifaces > 1) { - bond_rebalance_port(port); + + if (!port->lacp) { + continue; + } + + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + + if (lacp_iface_may_tx(iface)) { + poll_timer_wait_until(iface->lacp_tx); + } + + if (iface->lacp_status != LACP_STATUS_DEFAULTED) { + poll_timer_wait_until(iface->lacp_rx); + } } } } - -static struct ofhooks bridge_ofhooks = { - bridge_port_changed_ofhook_cb, - bridge_normal_ofhook_cb, - bridge_account_flow_ofhook_cb, - bridge_account_checkpoint_ofhook_cb, -}; /* Bonding functions. */ @@ -2278,6 +3293,22 @@ 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"; + static char *bm_tcp = "balance-tcp"; + + switch (bm) { + case BM_SLB: return bm_slb; + case BM_AB: return bm_ab; + case BM_TCP: return bm_tcp; + } + + NOT_REACHED(); + return NULL; +} + /* Sorts pointers to pointers to bond_entries in ascending order by the * interface to which they are assigned, and within a single interface in * ascending order of bytes transmitted. */ @@ -2386,6 +3417,8 @@ bond_shift_load(struct slave_balance *from, struct slave_balance *to, 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)", @@ -2421,13 +3454,15 @@ 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_AB); + /* Sets up 'bals' to describe each of the port's interfaces, sorted in * descending order of tx_bytes, so that bals[0] represents the most * heavily loaded slave and bals[n_bals - 1] represents the least heavily @@ -2439,6 +3474,7 @@ 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; @@ -2468,7 +3504,7 @@ bond_rebalance_port(struct port *port) while (!bals[n_bals - 1].iface->enabled) { n_bals--; if (!n_bals) { - return; + goto exit; } } @@ -2496,7 +3532,6 @@ bond_rebalance_port(struct port *port) * 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++) { @@ -2561,6 +3596,9 @@ 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 @@ -2571,21 +3609,28 @@ bond_send_learning_packets(struct port *port) struct ofpbuf packet; int error, n_packets, n_errors; - if (!port->n_ifaces || port->active_iface < 0) { + if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) { return; } ofpbuf_init(&packet, 128); error = n_packets = n_errors = 0; - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { union ofp_action actions[2], *a; 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)) { + if (e->port == port->port_idx) { + continue; + } + + compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, + e->mac); + flow_extract(&packet, 0, ODPP_NONE, &flow); + + if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) { continue; } @@ -2605,9 +3650,6 @@ bond_send_learning_packets(struct port *port) /* Send packet. */ n_packets++; - compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, - e->mac); - flow_extract(&packet, ODPP_NONE, &flow); retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, &packet); if (retval) { @@ -2632,14 +3674,14 @@ bond_send_learning_packets(struct port *port) static void bond_unixctl_list(struct unixctl_conn *conn, - const char *args UNUSED, void *aux UNUSED) + 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++) { @@ -2647,7 +3689,8 @@ bond_unixctl_list(struct unixctl_conn *conn, 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) { @@ -2668,7 +3711,7 @@ 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++) { @@ -2681,9 +3724,45 @@ bond_find(const char *name) return NULL; } +static void +ds_put_lacp_state(struct ds *ds, uint8_t state) +{ + if (state & LACP_STATE_ACT) { + ds_put_cstr(ds, "activity "); + } + + if (state & LACP_STATE_TIME) { + ds_put_cstr(ds, "timeout "); + } + + if (state & LACP_STATE_AGG) { + ds_put_cstr(ds, "aggregation "); + } + + if (state & LACP_STATE_SYNC) { + ds_put_cstr(ds, "synchronized "); + } + + if (state & LACP_STATE_COL) { + ds_put_cstr(ds, "collecting "); + } + + if (state & LACP_STATE_DIST) { + ds_put_cstr(ds, "distributing "); + } + + if (state & LACP_STATE_DEF) { + ds_put_cstr(ds, "defaulted "); + } + + if (state & LACP_STATE_EXP) { + ds_put_cstr(ds, "expired "); + } +} + static void bond_unixctl_show(struct unixctl_conn *conn, - const char *args, void *aux UNUSED) + const char *args, void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct port *port; @@ -2695,13 +3774,42 @@ bond_unixctl_show(struct unixctl_conn *conn, return; } + ds_put_format(&ds, "bond_mode: %s\n", + bond_mode_to_string(port->bond_mode)); + + if (port->lacp) { + ds_put_format(&ds, "\tlacp: %s\n", + port->lacp & LACP_ACTIVE ? "active" : "passive"); + } else { + ds_put_cstr(&ds, "\tlacp: off\n"); + } + + if (port->bond_mode != BM_AB) { + ds_put_format(&ds, "bond-hash-algorithm: %s\n", + bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb"); + } + + + ds_put_format(&ds, "bond-detect-mode: %s\n", + port->miimon ? "miimon" : "carrier"); + + if (port->miimon) { + ds_put_format(&ds, "bond-miimon-interval: %lld\n", + port->bond_miimon_interval); + } + ds_put_format(&ds, "updelay: %d ms\n", port->updelay); ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); - ds_put_format(&ds, "next rebalance: %lld ms\n", - port->bridge->bond_next_rebalance - time_msec()); + + if (port->bond_mode != BM_AB) { + ds_put_format(&ds, "next rebalance: %lld ms\n", + port->bond_next_rebalance - time_msec()); + } + for (j = 0; j < port->n_ifaces; j++) { const struct iface *iface = port->ifaces[j]; struct bond_entry *be; + struct flow flow; /* Basic info. */ ds_put_format(&ds, "slave %s: %s\n", @@ -2715,7 +3823,72 @@ bond_unixctl_show(struct unixctl_conn *conn, iface->delay_expires - time_msec()); } + if (port->lacp) { + ds_put_cstr(&ds, "\tstatus: "); + + if (iface->lacp_status == LACP_STATUS_CURRENT) { + ds_put_cstr(&ds, "current "); + } else if (iface->lacp_status == LACP_STATUS_EXPIRED) { + ds_put_cstr(&ds, "expired "); + } else { + ds_put_cstr(&ds, "defaulted "); + } + + if (iface->lacp_attached) { + ds_put_cstr(&ds, "attached "); + } + + ds_put_cstr(&ds, "\n"); + + ds_put_cstr(&ds, "\n\tactor sysid: "); + ds_put_format(&ds, ETH_ADDR_FMT, + ETH_ADDR_ARGS(iface->lacp_actor.sysid)); + ds_put_cstr(&ds, "\n"); + + ds_put_format(&ds, "\tactor sys_priority: %u\n", + ntohs(iface->lacp_actor.sys_priority)); + + ds_put_format(&ds, "\tactor portid: %u\n", + ntohs(iface->lacp_actor.portid)); + + ds_put_format(&ds, "\tactor port_priority: %u\n", + ntohs(iface->lacp_actor.port_priority)); + + ds_put_format(&ds, "\tactor key: %u\n", + ntohs(iface->lacp_actor.key)); + + ds_put_cstr(&ds, "\tactor state: "); + ds_put_lacp_state(&ds, iface_get_lacp_state(iface)); + ds_put_cstr(&ds, "\n\n"); + + ds_put_cstr(&ds, "\tpartner sysid: "); + ds_put_format(&ds, ETH_ADDR_FMT, + ETH_ADDR_ARGS(iface->lacp_partner.sysid)); + ds_put_cstr(&ds, "\n"); + + ds_put_format(&ds, "\tpartner sys_priority: %u\n", + ntohs(iface->lacp_partner.sys_priority)); + + ds_put_format(&ds, "\tpartner portid: %u\n", + ntohs(iface->lacp_partner.portid)); + + ds_put_format(&ds, "\tpartner port_priority: %u\n", + ntohs(iface->lacp_partner.port_priority)); + + ds_put_format(&ds, "\tpartner key: %u\n", + ntohs(iface->lacp_partner.key)); + + ds_put_cstr(&ds, "\tpartner state: "); + ds_put_lacp_state(&ds, iface->lacp_partner.state); + ds_put_cstr(&ds, "\n\n"); + } + + if (port->bond_mode == BM_AB) { + continue; + } + /* Hashes. */ + memset(&flow, 0, sizeof flow); for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { int hash = be - port->bond_hash; struct mac_entry *me; @@ -2727,14 +3900,20 @@ bond_unixctl_show(struct unixctl_conn *conn, ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", hash, be->tx_bytes / 1024); + if (port->bond_mode != BM_SLB) { + continue; + } + /* MACs. */ - 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 + + memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN); + if (bond_hash_src(me->mac, me->vlan) == hash && me->port != port->port_idx - && choose_output_iface(port, me->mac, &dp_ifidx, &tags) + && choose_output_iface(port, &flow, me->vlan, + &dp_ifidx, &tags) && dp_ifidx == iface->dp_ifidx) { ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n", @@ -2749,12 +3928,11 @@ bond_unixctl_show(struct unixctl_conn *conn, static void bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, - void *aux UNUSED) + 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; @@ -2775,10 +3953,12 @@ 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"); @@ -2806,7 +3986,7 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, static void bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, - void *aux UNUSED) + void *aux OVS_UNUSED) { char *args = (char *) args_; char *save_ptr = NULL; @@ -2887,36 +4067,52 @@ enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) static void bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, - void *aux UNUSED) + void *aux OVS_UNUSED) { enable_slave(conn, args, true); } static void bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, - void *aux UNUSED) + void *aux OVS_UNUSED) { enable_slave(conn, args, false); } static void -bond_unixctl_hash(struct unixctl_conn *conn, const char *args, - void *aux UNUSED) +bond_unixctl_hash(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) { - uint8_t mac[ETH_ADDR_LEN]; - uint8_t hash; - char *hash_cstr; + 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(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) - == ETH_ADDR_SCAN_COUNT) { - hash = bond_hash(mac); + if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) + == ETH_ADDR_SCAN_COUNT) { + hash = bond_hash_src(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"); - } + 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 @@ -2954,6 +4150,7 @@ port_create(struct bridge *br, const char *name) 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); @@ -2961,55 +4158,190 @@ port_create(struct bridge *br, const char *name) return port; } +static const char * +get_port_other_config(const struct ovsrec_port *port, const char *key, + const char *default_value) +{ + const char *value; + + value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config, + key); + return value ? value : default_value; +} + +static const char * +get_interface_other_config(const struct ovsrec_interface *iface, + const char *key, const char *default_value) +{ + const char *value; + + value = get_ovsrec_key_value(&iface->header_, + &ovsrec_interface_col_other_config, key); + return value ? value : default_value; +} + +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, const struct ovsrec_port *cfg) { - struct shash old_ifaces, new_ifaces; - struct shash_node *node; + const char *detect_mode; + struct shash new_ifaces; + long long int next_rebalance, miimon_next_update, lacp_priority; unsigned long *trunks; int vlan; size_t i; port->cfg = cfg; - /* Collect old and new interfaces. */ - shash_init(&old_ifaces); - shash_init(&new_ifaces); - for (i = 0; i < port->n_ifaces; i++) { - shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]); + /* Update settings. */ + port->updelay = cfg->bond_updelay; + if (port->updelay < 0) { + port->updelay = 0; + } + port->downdelay = cfg->bond_downdelay; + if (port->downdelay < 0) { + port->downdelay = 0; + } + port->bond_rebalance_interval = atoi( + get_port_other_config(cfg, "bond-rebalance-interval", "10000")); + if (port->bond_rebalance_interval < 1000) { + port->bond_rebalance_interval = 1000; + } + next_rebalance = time_msec() + port->bond_rebalance_interval; + if (port->bond_next_rebalance > next_rebalance) { + port->bond_next_rebalance = next_rebalance; + } + + detect_mode = get_port_other_config(cfg, "bond-detect-mode", + "carrier"); + + if (!strcmp(detect_mode, "carrier")) { + port->miimon = false; + } else if (!strcmp(detect_mode, "miimon")) { + port->miimon = true; + } else { + port->miimon = false; + VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to " + "carrier", port->name, detect_mode); + } + + port->bond_miimon_interval = atoi( + get_port_other_config(cfg, "bond-miimon-interval", "200")); + if (port->bond_miimon_interval < 100) { + port->bond_miimon_interval = 100; + } + miimon_next_update = time_msec() + port->bond_miimon_interval; + if (port->bond_miimon_next_update > miimon_next_update) { + port->bond_miimon_next_update = miimon_next_update; + } + + if (!port->cfg->bond_mode || + !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) { + port->bond_mode = BM_SLB; + } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) { + port->bond_mode = BM_AB; + } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) { + port->bond_mode = BM_TCP; + } else { + port->bond_mode = BM_SLB; + VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s", + port->name, port->cfg->bond_mode, + bond_mode_to_string(port->bond_mode)); } + + /* Add new interfaces and update 'cfg' member of existing ones. */ + shash_init(&new_ifaces); for (i = 0; i < cfg->n_interfaces; i++) { - const char *name = cfg->interfaces[i]->name; - if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) { + 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, name); + 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); } - } - port->updelay = cfg->bond_updelay; - if (port->updelay < 0) { - port->updelay = 0; - } - port->updelay = cfg->bond_downdelay; - if (port->downdelay < 0) { - port->downdelay = 0; - } - /* Get rid of deleted interfaces and add new interfaces. */ - SHASH_FOR_EACH (node, &old_ifaces) { - if (!shash_find(&new_ifaces, node->name)) { - iface_destroy(node->data); + /* 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"); + + lacp_priority = + atoi(get_interface_other_config(if_cfg, "lacp-port-priority", + "0")); + + if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) { + iface->lacp_priority = UINT16_MAX; + } else { + iface->lacp_priority = lacp_priority; } } - SHASH_FOR_EACH (node, &new_ifaces) { - const struct ovsrec_interface *if_cfg = node->data; - const char *if_name = node->name; - struct iface *iface; + shash_destroy(&new_ifaces); - iface = shash_find_data(&old_ifaces, if_name); - if (!iface) { - iface = iface_create(port, if_name); - } - iface->cfg = if_cfg; + lacp_priority = + atoi(get_port_other_config(cfg, "lacp-system-priority", "0")); + + if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) { + /* Prefer bondable links if unspecified. */ + port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX; + } else { + port->lacp_priority = lacp_priority; + } + + if (!port->cfg->lacp) { + /* XXX when LACP implementation has been sufficiently tested, enable by + * default and make active on bonded ports. */ + port->lacp = 0; + } else if (!strcmp(port->cfg->lacp, "off")) { + port->lacp = 0; + } else if (!strcmp(port->cfg->lacp, "active")) { + port->lacp = LACP_ACTIVE; + } else if (!strcmp(port->cfg->lacp, "passive")) { + port->lacp = LACP_PASSIVE; + } else { + VLOG_WARN("port %s: unknown LACP mode %s", + port->name, port->cfg->lacp); + port->lacp = 0; } /* Get VLAN tag. */ @@ -3036,9 +4368,8 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) /* Get trunked VLANs. */ trunks = NULL; - if (vlan < 0) { + if (vlan < 0 && cfg->n_trunks) { size_t n_errors; - size_t i; trunks = bitmap_allocate(4096); n_errors = 0; @@ -3055,17 +4386,14 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) port->name, cfg->n_trunks); } if (n_errors == cfg->n_trunks) { - if (n_errors) { - VLOG_ERR("port %s: no valid trunks, trunking all VLANs", - port->name); - } - bitmap_set_multiple(trunks, 0, 4096, 1); - } - } else { - if (cfg->n_trunks) { - VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", + 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 @@ -3074,9 +4402,6 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) } bitmap_free(port->trunks); port->trunks = trunks; - - shash_destroy(&old_ifaces); - shash_destroy(&new_ifaces); } static void @@ -3085,26 +4410,30 @@ port_destroy(struct port *port) if (port) { struct bridge *br = port->bridge; struct port *del; + int i; proc_net_compat_update_vlan(port->name, NULL, 0); proc_net_compat_update_bond(port->name, NULL); -#if 0 for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; if (m && m->out_port == port) { mirror_destroy(m); } } -#endif while (port->n_ifaces > 0) { iface_destroy(port->ifaces[port->n_ifaces - 1]); } + 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); @@ -3123,46 +4452,79 @@ 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; + struct iface *iface = iface_lookup(port->bridge, name); + return iface && iface->port == port ? iface : NULL; +} - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (!strcmp(iface->name, name)) { - return iface; +static void +port_update_lacp(struct port *port) +{ + size_t i; + bool key_changed; + + if (!port->lacp || port->n_ifaces < 1) { + return; + } + + key_changed = true; + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + + if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) { + port->lacp = 0; + return; + } + + if (iface->dp_ifidx == port->lacp_key) { + key_changed = false; } } - return NULL; + + if (key_changed) { + port->lacp_key = port->ifaces[0]->dp_ifidx; + } + + for (i = 0; i < port->n_ifaces; i++) { + struct iface *iface = port->ifaces[i]; + + iface->lacp_actor.sys_priority = htons(port->lacp_priority); + memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN); + + iface->lacp_actor.port_priority = htons(iface->lacp_priority); + iface->lacp_actor.portid = htons(iface->dp_ifidx); + iface->lacp_actor.key = htons(port->lacp_key); + + iface->lacp_tx = 0; + } + port->lacp_need_update = true; } 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) { free(port->bond_hash); port->bond_hash = NULL; port->bond_compat_is_stale = true; - port->bond_fake_iface = false; } + + port->bond_fake_iface = false; } else { - if (!port->bond_hash) { - size_t i; + size_t i; + if (port->bond_mode != BM_AB && !port->bond_hash) { port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); for (i = 0; i <= BOND_MASK; i++) { struct bond_entry *e = &port->bond_hash[i]; @@ -3171,9 +4533,25 @@ 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_AB) { + free(port->bond_hash); + port->bond_hash = NULL; } port->bond_compat_is_stale = true; port->bond_fake_iface = port->cfg->bond_fake_iface; + + if (!port->miimon) { + port->monitor = netdev_monitor_create(); + for (i = 0; i < port->n_ifaces; i++) { + netdev_monitor_add(port->monitor, port->ifaces[i]->netdev); + } + } } } @@ -3184,7 +4562,7 @@ 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; } @@ -3215,15 +4593,15 @@ port_update_bond_compat(struct port *port) /* 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 + * 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 + * 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; @@ -3236,7 +4614,7 @@ port_update_bond_compat(struct port *port) if (port->bond_fake_iface) { struct netdev *bond_netdev; - if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) { + if (!netdev_open_default(port->name, &bond_netdev)) { if (bond.up) { netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); } else { @@ -3289,10 +4667,113 @@ port_update_vlan_compat(struct port *port) /* Interface functions. */ +static void +iface_set_lacp_defaulted(struct iface *iface) +{ + memset(&iface->lacp_partner, 0xff, sizeof iface->lacp_partner); + iface->lacp_partner.state = 0; + + iface->lacp_status = LACP_STATUS_DEFAULTED; + iface->lacp_tx = 0; + iface->port->lacp_need_update = true; +} + +static void +iface_set_lacp_expired(struct iface *iface) +{ + iface->lacp_status = LACP_STATUS_EXPIRED; + iface->lacp_partner.state |= LACP_STATE_TIME; + iface->lacp_partner.state &= ~LACP_STATE_SYNC; + + iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX; + iface->lacp_tx = 0; +} + +static uint8_t +iface_get_lacp_state(const struct iface *iface) +{ + uint8_t state = 0; + + if (iface->port->lacp & LACP_ACTIVE) { + state |= LACP_STATE_ACT; + } + + if (iface->lacp_status == LACP_STATUS_DEFAULTED) { + state |= LACP_STATE_DEF; + } else if (iface->lacp_attached) { + state |= LACP_STATE_SYNC; + } + + if (iface->lacp_status == LACP_STATUS_EXPIRED) { + state |= LACP_STATE_EXP; + } + + if (iface->port->n_ifaces > 1) { + state |= LACP_STATE_AGG; + } + + if (iface->enabled) { + state |= LACP_STATE_COL | LACP_STATE_DIST; + } + + return state; +} + +/* Given 'iface', populates 'priority' with data representing its LACP link + * priority. If two priority objects populated by this function are compared + * using memcmp, the higher priority link will be less than the lower priority + * link. */ +static void +iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority) +{ + uint16_t partner_priority, actor_priority; + + /* Choose the lacp_info of the higher priority system by comparing their + * system priorities and mac addresses. */ + actor_priority = ntohs(iface->lacp_actor.sys_priority); + partner_priority = ntohs(iface->lacp_partner.sys_priority); + if (actor_priority < partner_priority) { + *priority = iface->lacp_actor; + } else if (partner_priority < actor_priority) { + *priority = iface->lacp_partner; + } else if (eth_addr_compare_3way(iface->lacp_actor.sysid, + iface->lacp_partner.sysid) < 0) { + *priority = iface->lacp_actor; + } else { + *priority = iface->lacp_partner; + } + + /* Key and state are not used in priority comparisons. */ + priority->key = 0; + priority->state = 0; +} + +static void +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 char *name) +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 = xzalloc(sizeof *iface); iface->port = port; @@ -3302,6 +4783,14 @@ 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; + iface_set_lacp_defaulted(iface); + + if (port->lacp & LACP_ACTIVE) { + iface_set_lacp_expired(iface); + } + + shash_add_assert(&br->iface_by_name, iface->name, iface); if (port->n_ifaces >= port->allocated_ifaces) { port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, @@ -3309,12 +4798,12 @@ 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; } @@ -3328,16 +4817,20 @@ 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); @@ -3345,6 +4838,11 @@ iface_destroy(struct iface *iface) bond_send_learning_packets(port); } + cfm_destroy(iface->cfm); + + free(iface->name); + free(iface); + bridge_flush(port->bridge); } } @@ -3352,57 +4850,21 @@ iface_destroy(struct iface *iface) static struct iface * iface_lookup(const struct bridge *br, const char *name) { - size_t i, j; - - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (!strcmp(iface->name, name)) { - return iface; - } - } - } - return NULL; + return shash_find_data(&br->iface_by_name, name); } static struct iface * iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) { - return port_array_get(&br->ifaces, dp_ifidx); -} - -/* Returns true if 'iface' is the name of an "internal" interface on bridge - * 'br', that is, an interface that is entirely simulated within the datapath. - * The local port (ODPP_LOCAL) is always an internal interface. Other local - * interfaces are created by setting "iface..internal = true". - * - * In addition, we have a kluge-y feature that creates an internal port with - * the name of a bonded port if "bonding..fake-iface = true" is set. - * This feature needs to go away in the long term. Until then, this is one - * reason why this function takes a name instead of a struct iface: the fake - * interfaces created this way do not have a struct iface. */ -static bool -iface_is_internal(const struct bridge *br, const char *if_name) -{ - /* XXX wastes time */ struct iface *iface; - struct port *port; - - if (!strcmp(if_name, br->name)) { - return true; - } - iface = iface_lookup(br, if_name); - if (iface && !strcmp(iface->cfg->type, "internal")) { - return true; - } - - port = port_lookup(br, if_name); - if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { - return true; + 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 false; + return NULL; } /* Set Ethernet address of 'iface', if one is specified in the configuration @@ -3428,51 +4890,220 @@ iface_set_mac(struct iface *iface) } } } - -/* Port mirroring. */ -#if 0 +/* Sets the ofport column of 'if_cfg' to 'ofport'. */ static void -mirror_reconfigure(struct bridge *br UNUSED) +iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport) { - struct svec old_mirrors, new_mirrors; - size_t i, n_rspan_vlans; - unsigned long *rspan_vlans; + if (if_cfg) { + ovsrec_interface_set_ofport(if_cfg, &ofport, 1); + } +} - /* 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); +/* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'. + * + * The value strings in '*shash' are taken directly from values[], not copied, + * so the caller should not modify or free them. */ +static void +shash_from_ovs_idl_map(char **keys, char **values, size_t n, + struct shash *shash) +{ + size_t i; + + 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); } } @@ -3487,32 +5118,30 @@ mirror_reconfigure(struct bridge *br UNUSED) } } - /* Update learning disabled vlans (for RSPAN). */ + /* Update flooded vlans (for RSPAN). */ rspan_vlans = NULL; - n_rspan_vlans = cfg_count("vlan.%s.disable-learning", br->name); - if (n_rspan_vlans) { + if (br->cfg->n_flood_vlans) { rspan_vlans = bitmap_allocate(4096); - for (i = 0; i < n_rspan_vlans; i++) { - int vlan = cfg_get_vlan(i, "vlan.%s.disable-learning", br->name); - if (vlan >= 0) { + 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 %d\n", + VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64, br->name, vlan); } else { - VLOG_ERR("bridge %s: invalid value '%s' for learning disabled " - "VLAN", br->name, - cfg_get_string(i, "vlan.%s.disable-learning", br->name)); + VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN", + br->name, vlan); } } } - if (mac_learning_set_disabled_vlans(br->ml, rspan_vlans)) { + 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; @@ -3520,7 +5149,7 @@ 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]) { @@ -3528,19 +5157,21 @@ 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 = 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 @@ -3555,11 +5186,12 @@ 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); @@ -3567,45 +5199,36 @@ 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; } @@ -3640,104 +5263,92 @@ port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) } static void -mirror_reconfigure_one(struct mirror *m UNUSED) +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) @@ -3747,7 +5358,7 @@ mirror_reconfigure_one(struct mirror *m UNUSED) 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; @@ -3755,9 +5366,6 @@ mirror_reconfigure_one(struct mirror *m UNUSED) } /* Clean up. */ -exit: - svec_destroy(&src_ports); - svec_destroy(&dst_ports); - free(pfx); + shash_destroy(&src_ports); + shash_destroy(&dst_ports); } -#endif