X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=44474966ead69747977ec55e387533767017c19f;hb=79efe70efcc57414f43a2d10fbacae25aa6e43ca;hp=620689fcf23d5a98ba8cb750e66162c3928622b4;hpb=6501fda909c3f70fb8df171003af32956d05fc43;p=openvswitch diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 620689fc..44474966 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -1,4 +1,4 @@ -/* Copyright (c) 2008, 2009, 2010 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 @@ -30,70 +32,101 @@ #include #include #include "bitmap.h" +#include "cfm.h" +#include "classifier.h" #include "coverage.h" +#include "daemon.h" #include "dirs.h" #include "dpif.h" #include "dynamic-string.h" #include "flow.h" #include "hash.h" +#include "hmap.h" +#include "jsonrpc.h" +#include "lacp.h" #include "list.h" #include "mac-learning.h" #include "netdev.h" +#include "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 "sset.h" #include "svec.h" +#include "system-stats.h" #include "timeval.h" #include "util.h" #include "unixctl.h" #include "vconn.h" #include "vswitchd/vswitch-idl.h" #include "xenserver.h" -#include "xtoxll.h" +#include "vlog.h" #include "sflow_api.h" -#define THIS_MODULE VLM_bridge -#include "vlog.h" +VLOG_DEFINE_THIS_MODULE(bridge); + +COVERAGE_DEFINE(bridge_flush); +COVERAGE_DEFINE(bridge_process_flow); +COVERAGE_DEFINE(bridge_process_cfm); +COVERAGE_DEFINE(bridge_process_lacp); +COVERAGE_DEFINE(bridge_reconfigure); +COVERAGE_DEFINE(bridge_lacp_update); 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 list port_elem; /* Element in struct port's "ifaces" list. */ struct port *port; /* Containing port. */ - size_t port_ifidx; /* Index within containing port. */ char *name; /* Host network device name. */ tag_type tag; /* Tag associated with this interface. */ long long delay_expires; /* Time after which 'enabled' may change. */ /* These members are valid only after bridge_reconfigure() causes them to - * be initialized.*/ + * 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. */ const struct ovsrec_interface *cfg; }; #define BOND_MASK 0xff struct bond_entry { - int iface_idx; /* Index of assigned iface, or -1 if none. */ + struct iface *iface; /* Assigned iface, or NULL if none. */ uint64_t tx_bytes; /* Count of bytes recently transmitted. */ - tag_type iface_tag; /* Tag associated with iface_idx. */ + tag_type tag; /* Tag for bond_entry<->iface association. */ +}; + +enum bond_mode { + BM_TCP, /* Transport Layer Load Balance. */ + BM_SLB, /* Source Load Balance. */ + BM_AB /* Active Backup. */ }; #define MAX_MIRRORS 32 @@ -104,10 +137,11 @@ struct mirror { struct bridge *bridge; size_t idx; char *name; + struct uuid uuid; /* UUID of this "mirror" record in database. */ /* Selection criteria. */ - struct shash src_ports; /* Name is port name; data is always NULL. */ - struct shash dst_ports; /* Name is port name; data is always NULL. */ + struct sset src_ports; /* Source port names. */ + struct sset dst_ports; /* Destination port names. */ int *vlans; size_t n_vlans; @@ -119,24 +153,37 @@ struct mirror { #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ struct port { struct bridge *bridge; - size_t port_idx; - int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ - unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */ + struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */ char *name; + int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ + unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. + * NULL if all VLANs are trunked. */ + const struct ovsrec_port *cfg; + + /* Monitoring. */ + struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */ + long long int miimon_interval; /* Miimon status refresh interval. */ + long long int miimon_next_update; /* Time of next miimon update. */ + /* An ordinary bridge port has 1 interface. * A bridge port for bonding has at least 2 interfaces. */ - struct iface **ifaces; - size_t n_ifaces, allocated_ifaces; + struct list ifaces; /* List of "struct iface"s. */ + size_t n_ifaces; /* list_size(ifaces). */ /* Bonding info. */ - struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ - int active_iface; /* Ifidx on which bcasts accepted, or -1. */ - tag_type active_iface_tag; /* Tag for bcast flows. */ + enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */ + struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */ 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? */ + long long int bond_next_fake_iface_update; /* Time of next update. */ + + /* LACP information. */ + struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */ + + /* 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. */ @@ -144,41 +191,26 @@ struct port { 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. */ - /* Description strings. */ - char *mfr_desc; /* Manufacturer. */ - char *hw_desc; /* Hardware. */ - char *sw_desc; /* Software version. */ - char *serial_desc; /* Serial number. */ - char *dp_desc; /* Datapath description. */ - /* 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 hmap ports; /* "struct port"s indexed by name. */ + struct shash iface_by_name; /* "struct iface"s indexed by name. */ /* Bonding. */ bool has_bonded_ports; @@ -186,34 +218,38 @@ struct bridge { /* 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; + +/* 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; + +/* Stores the time after which CFM statistics may be written to the database. + * Only updated when changes to the database require rate limiting. */ +#define CFM_LIMIT_INTERVAL (1 * 1000) /* In milliseconds. */ +static long long int cfm_limiter = LLONG_MIN; static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg); static void bridge_destroy(struct bridge *); static struct bridge *bridge_lookup(const char *name); static unixctl_cb_func bridge_unixctl_dump_flows; +static unixctl_cb_func bridge_unixctl_reconnect; static int bridge_run_one(struct bridge *); -static const struct ovsrec_controller *bridge_get_controller( - const struct ovsrec_open_vswitch *ovs_cfg, - const struct bridge *br); -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 *); @@ -223,250 +259,184 @@ static void bridge_pick_local_hw_addr(struct bridge *, static uint64_t bridge_pick_datapath_id(struct bridge *, const uint8_t bridge_ea[ETH_ADDR_LEN], struct iface *hw_addr_iface); -static struct iface *bridge_get_local_iface(struct bridge *); static uint64_t dpid_from_hash(const void *, size_t nbytes); static unixctl_cb_func bridge_unixctl_fdb_show; +static unixctl_cb_func cfm_unixctl_show; +static unixctl_cb_func qos_unixctl_show; static void bond_init(void); -static void bond_run(struct bridge *); -static void bond_wait(struct bridge *); +static void bond_run(struct port *); +static void bond_wait(struct port *); static void bond_rebalance_port(struct port *); static void bond_send_learning_packets(struct port *); static void bond_enable_slave(struct iface *iface, bool enable); +static void port_run(struct port *); +static void port_wait(struct port *); static struct port *port_create(struct bridge *, const char *name); static void port_reconfigure(struct port *, const struct ovsrec_port *); +static void port_del_ifaces(struct port *, const struct ovsrec_port *); static void port_destroy(struct port *); static struct port *port_lookup(const struct bridge *, const char *name); static struct iface *port_lookup_iface(const struct port *, const char *name); +static struct iface *port_get_an_iface(const struct port *); static struct port *port_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); -static void port_update_bond_compat(struct port *); -static void port_update_vlan_compat(struct port *); static void port_update_bonding(struct port *); +static void port_update_lacp(struct port *); -static struct mirror *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 *, struct ovsrec_mirror *); static bool vlan_is_mirrored(const struct mirror *, int vlan); -static struct iface *iface_create(struct port *port, +static struct iface *iface_create(struct port *port, const struct ovsrec_interface *if_cfg); static void iface_destroy(struct iface *); static struct iface *iface_lookup(const struct bridge *, const char *name); +static struct iface *iface_find(const char *name); static struct iface *iface_from_dp_ifidx(const struct bridge *, uint16_t dp_ifidx); -static 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 bool iface_refresh_cfm_stats(struct iface *iface); +static void iface_update_carrier(struct iface *); +static bool iface_get_carrier(const struct iface *); + +static void shash_from_ovs_idl_map(char **keys, char **values, size_t n, + struct shash *); +static void shash_to_ovs_idl_map(struct shash *, + char ***keys, char ***values, size_t *n); /* Hooks into ofproto processing. */ static struct ofhooks bridge_ofhooks; /* 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, dpif_types; - size_t i; + /* Create connection to database. */ + idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true); - unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); - - svec_init(&bridge_names); - for (i = 0; i < cfg->n_bridges; i++) { - svec_add(&bridge_names, cfg->bridges[i]->name); - } - svec_sort(&bridge_names); + 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); - svec_init(&dpif_types); - dp_enumerate_types(&dpif_types); - for (i = 0; i < dpif_types.n; i++) { - struct dpif *dpif; - int retval; - size_t j; + ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); - dp_enumerate_names(dpif_types.names[i], &dpif_names); + ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids); + ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge); - for (j = 0; j < dpif_names.n; j++) { - retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif); - if (!retval) { - struct svec all_names; - size_t k; - - svec_init(&all_names); - dpif_get_all_names(dpif, &all_names); - for (k = 0; k < all_names.n; k++) { - if (svec_contains(&bridge_names, all_names.names[k])) { - goto found; - } - } - dpif_delete(dpif); - found: - svec_destroy(&all_names); - dpif_close(dpif); - } - } - } - svec_destroy(&bridge_names); - svec_destroy(&dpif_names); - svec_destroy(&dpif_types); + 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("cfm/show", cfm_unixctl_show, NULL); + unixctl_command_register("qos/show", qos_unixctl_show, NULL); unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, NULL); - + unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect, + NULL); + lacp_init(); bond_init(); - bridge_reconfigure(cfg); } -#ifdef HAVE_OPENSSL -static void -bridge_configure_ssl(const struct ovsrec_ssl *ssl) +void +bridge_exit(void) { - /* XXX SSL should be configurable on a per-bridge basis. */ - if (ssl) { - stream_ssl_set_private_key_file(ssl->private_key); - stream_ssl_set_certificate_file(ssl->certificate); - stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert); + struct bridge *br, *next_br; + + 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, struct iface *iface, - 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 = 0; + static bool already_configured_once; + struct sset bridge_names; + struct sset dpif_names, dpif_types; + const char *type; size_t i; - 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])); + /* Only do this once per ovs-vswitchd run. */ + if (already_configured_once) { + return; } + already_configured_once = true; - if (create) { - struct netdev_options netdev_options; + stats_timer = time_msec() + STATS_INTERVAL; - memset(&netdev_options, 0, sizeof netdev_options); - netdev_options.name = iface_cfg->name; - if (!strcmp(iface_cfg->type, "internal")) { - /* An "internal" config type maps to a netdev "system" type. */ - netdev_options.type = "system"; - } else { - netdev_options.type = iface_cfg->type; - } - netdev_options.args = &options; - netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; - netdev_options.may_create = true; - if (iface_is_internal(iface->port->bridge, iface_cfg->name)) { - netdev_options.may_open = true; - } + /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */ + sset_init(&bridge_names); + for (i = 0; i < cfg->n_bridges; i++) { + sset_add(&bridge_names, cfg->bridges[i]->name); + } - error = netdev_open(&netdev_options, &iface->netdev); + /* Iterate over all system dpifs and delete any of them that do not appear + * in 'cfg'. */ + sset_init(&dpif_names); + sset_init(&dpif_types); + dp_enumerate_types(&dpif_types); + SSET_FOR_EACH (type, &dpif_types) { + const char *name; - if (iface->netdev) { - netdev_get_carrier(iface->netdev, &iface->enabled); - } - } else if (iface->netdev) { - const char *netdev_type = netdev_get_type(iface->netdev); - const char *iface_type = iface_cfg->type && strlen(iface_cfg->type) - ? iface_cfg->type : NULL; + dp_enumerate_names(type, &dpif_names); - /* An "internal" config type maps to a netdev "system" type. */ - if (iface_type && !strcmp(iface_type, "internal")) { - iface_type = "system"; - } + /* Delete each dpif whose name is not in 'bridge_names'. */ + SSET_FOR_EACH (name, &dpif_names) { + if (!sset_contains(&bridge_names, name)) { + struct dpif *dpif; + int retval; - if (!iface_type || !strcmp(netdev_type, iface_type)) { - error = netdev_reconfigure(iface->netdev, &options); - } else { - VLOG_WARN("%s: attempting change device type from %s to %s", - iface_cfg->name, netdev_type, iface_type); - error = EINVAL; + retval = dpif_open(name, type, &dpif); + if (!retval) { + dpif_delete(dpif); + dpif_close(dpif); + } + } } } - - SHASH_FOR_EACH (node, &options) { - free(node->data); - } - shash_destroy(&options); - - return error; -} - -static int -reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface) -{ - return set_up_iface(iface_cfg, iface, false); + sset_destroy(&bridge_names); + sset_destroy(&dpif_names); + sset_destroy(&dpif_types); } +/* Callback for iterate_and_prune_ifaces(). */ static bool -check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface, - void *aux OVS_UNUSED) +check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED) { if (!iface->netdev) { - int error = set_up_iface(iface->cfg, iface, true); - if (error) { - VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name, - strerror(error)); - return false; - } + /* We already reported a related error, don't bother duplicating it. */ + return false; } - return true; -} - -static bool -check_iface_dp_ifidx(struct bridge *br, struct iface *iface, - void *aux OVS_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 OVS_UNUSED, struct iface *iface, void *aux OVS_UNUSED) @@ -478,8 +448,7 @@ set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, /* Set MAC address of internal interfaces other than the local * interface. */ - if (iface->dp_ifidx != ODPP_LOCAL - && iface_is_internal(br, iface->name)) { + if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) { iface_set_mac(iface); } @@ -495,46 +464,100 @@ iterate_and_prune_ifaces(struct bridge *br, void *aux), void *aux) { - size_t i, j; + struct port *port, *next_port; - for (i = 0; i < br->n_ports; ) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; ) { - struct iface *iface = port->ifaces[j]; - if (cb(br, iface, aux)) { - j++; - } else { + HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) { + struct iface *iface, *next_iface; + + LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) { + if (!cb(br, iface, aux)) { + iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } } - if (port->n_ifaces) { - i++; - } else { - VLOG_ERR("%s port has no interfaces, dropping", port->name); + if (!port->n_ifaces) { + VLOG_WARN("%s port has no interfaces, dropping", port->name); port_destroy(port); } } } -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 sset targets; + size_t i; + + /* Collect all of the potential targets from the "targets" columns of the + * rows pointed to by "manager_options", excluding any that are + * out-of-band. */ + sset_init(&targets); + for (i = 0; i < ovs_cfg->n_manager_options; i++) { + struct ovsrec_manager *m = ovs_cfg->manager_options[i]; + + if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) { + sset_find_and_delete(&targets, m->target); + } else { + sset_add(&targets, m->target); + } + } + + /* Now extract the targets' IP addresses. */ + if (!sset_is_empty(&targets)) { + const char *target; + + managers = xmalloc(sset_count(&targets) * sizeof *managers); + SSET_FOR_EACH (target, &targets) { + struct sockaddr_in *sin = &managers[n_managers]; + + if ((!strncmp(target, "tcp:", 4) + && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) || + (!strncmp(target, "ssl:", 4) + && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) { + n_managers++; + } + } + } + sset_destroy(&targets); + + *managersp = managers; + *n_managersp = n_managers; +} + +static void bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) { - struct ovsdb_idl_txn *txn; 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); - txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl); + 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++) { @@ -545,7 +568,7 @@ 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; @@ -571,14 +594,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_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. @@ -586,81 +604,145 @@ 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), - strerror(retval)); + VLOG_WARN("failed to remove %s interface from %s: %s", + dpif_port.name, dpif_name(br->dpif), + strerror(retval)); } } } 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; - if (shash_find(&cur_ifaces, if_name)) { - /* Already exists, just reconfigure it. */ + /* First open the network device. */ + options.name = if_name; + options.type = type; + options.args = &args; + options.ethertype = NETDEV_ETH_TYPE_NONE; + + shash_init(&args); if (iface) { - reconfigure_iface(iface->cfg, iface); + 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; - - /* Add to datapath. */ - internal = iface_is_internal(br, if_name); - error = dpif_port_add(br->dpif, if_name, - internal ? ODP_PORT_INTERNAL : 0, NULL); - if (error == EFBIG) { - VLOG_ERR("ran out of valid port numbers on %s", - dpif_name(br->dpif)); - break; - } else if (error) { - VLOG_ERR("failed to add %s interface to %s: %s", - if_name, dpif_name(br->dpif), strerror(error)); + error = netdev_open(&options, &netdev); + shash_destroy(&args); + + if (error) { + VLOG_WARN("could not open network device %s (%s)", + if_name, strerror(error)); + continue; + } + + /* Then add the port if we haven't already. */ + if (!dpif_port) { + error = dpif_port_add(br->dpif, netdev, NULL); + if (error) { + netdev_close(netdev); + if (error == EFBIG) { + VLOG_ERR("ran out of valid port numbers on %s", + dpif_name(br->dpif)); + break; + } else { + VLOG_WARN("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 = iface_get_carrier(iface); + 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); } sflow_bridge_number = 0; - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { uint8_t ea[8]; uint64_t dpid; struct iface *local_iface; @@ -669,12 +751,11 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) bridge_fetch_dp_ifaces(br); - iterate_and_prune_ifaces(br, check_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); - local_iface = bridge_get_local_iface(br); + local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL); if (local_iface) { int error = netdev_set_etheraddr(local_iface->netdev, ea); if (error) { @@ -684,11 +765,12 @@ 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("%012"PRIx64, dpid); + dpid_string = xasprintf("%016"PRIx64, dpid); ovsrec_bridge_set_datapath_id(br->cfg, dpid_string); free(dpid_string); @@ -724,19 +806,21 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) "with another vswitch, choose an engine id less " "than 128", br->name); } - if (br->n_ports > 508) { + if (hmap_count(&br->ports) > 508) { VLOG_WARN("bridge %s: netflow port mangling will conflict " "with another port when more than 508 ports are " "used", br->name); } } - opts.collectors.n = nf_cfg->n_targets; - opts.collectors.names = nf_cfg->targets; + sset_init(&opts.collectors); + sset_add_array(&opts.collectors, + nf_cfg->targets, nf_cfg->n_targets); if (ofproto_set_netflow(br->ofproto, &opts)) { - VLOG_ERR("bridge %s: problem setting netflow collectors", + VLOG_ERR("bridge %s: problem setting netflow collectors", br->name); } + sset_destroy(&opts.collectors); } else { ofproto_set_netflow(br->ofproto, NULL); } @@ -744,13 +828,15 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) /* Set sFlow configuration on this bridge. */ if (br->cfg->sflow) { const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow; - const struct ovsrec_controller *ctrl; + 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; + sset_init(&oso.targets); + sset_add_array(&oso.targets, + sflow_cfg->targets, sflow_cfg->n_targets); oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE; if (sflow_cfg->sampling) { @@ -770,11 +856,17 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) oso.sub_id = sflow_bridge_number++; oso.agent_device = sflow_cfg->agent; - ctrl = bridge_get_controller(ovs_cfg, br); - oso.control_ip = ctrl ? ctrl->local_ip : NULL; + 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); - svec_destroy(&oso.targets); + sset_destroy(&oso.targets); } else { ofproto_set_sflow(br->ofproto, NULL); } @@ -788,46 +880,68 @@ 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) { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + LIST_FOR_EACH (br, node, &all_bridges) { + struct port *port; - port_update_vlan_compat(port); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + if (port->monitor) { + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + netdev_monitor_add(port->monitor, iface->netdev); + } + } else { + port->miimon_next_update = 0; + } + + port_update_lacp(port); port_update_bonding(port); + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_update_qos(iface, 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); } - ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg); + 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); - ovsdb_idl_txn_commit(txn); - ovsdb_idl_txn_destroy(txn); /* XXX */ + /* ovs-vswitchd has completed initialization, so allow the process that + * forked us to exit successfully. */ + daemonize_complete(); } static const char * -get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n) +get_ovsrec_key_value(const struct ovsdb_idl_row *row, + const struct ovsdb_idl_column *column, + const char *key) { - size_t i; - - for (i = 0; i < n; i++) { - if (!strcmp(keys[i], key)) { - return values[i]; - } - } - return NULL; + 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(key, - br_cfg->key_other_config, - br_cfg->value_other_config, - br_cfg->n_other_config); + return get_ovsrec_key_value(&br_cfg->header_, + &ovsrec_bridge_col_other_config, key); } static void @@ -835,7 +949,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], struct iface **hw_addr_iface) { const char *hwaddr; - size_t i, j; + struct port *port; int error; *hw_addr_iface = NULL; @@ -855,10 +969,10 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], /* Otherwise choose the minimum non-local MAC address among all of the * interfaces. */ - memset(ea, 0xff, sizeof ea); - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + memset(ea, 0xff, ETH_ADDR_LEN); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { uint8_t iface_ea[ETH_ADDR_LEN]; + struct iface *candidate; struct iface *iface; /* Mirror output ports don't participate. */ @@ -867,12 +981,11 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], } /* Choose the MAC address to represent the port. */ + iface = NULL; if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) { /* Find the interface with this Ethernet address (if any) so that * we can provide the correct devname to the caller. */ - iface = NULL; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *candidate = port->ifaces[j]; + LIST_FOR_EACH (candidate, port_elem, &port->ifaces) { uint8_t candidate_ea[ETH_ADDR_LEN]; if (!netdev_get_etheraddr(candidate->netdev, candidate_ea) && eth_addr_equals(iface_ea, candidate_ea)) { @@ -886,10 +999,8 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], * scripts always add slaves to a bond in alphabetical order, so * for compatibility we choose the interface with the name that is * first in alphabetical order. */ - iface = port->ifaces[0]; - for (j = 1; j < port->n_ifaces; j++) { - struct iface *candidate = port->ifaces[j]; - if (strcmp(candidate->name, iface->name) < 0) { + LIST_FOR_EACH (candidate, port_elem, &port->ifaces) { + if (!iface || strcmp(candidate->name, iface->name) < 0) { iface = candidate; } } @@ -915,7 +1026,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], !eth_addr_is_local(iface_ea) && !eth_addr_is_reserved(iface_ea) && !eth_addr_is_zero(iface_ea) && - memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) + eth_addr_compare_3way(iface_ea, ea) < 0) { memcpy(ea, iface_ea, ETH_ADDR_LEN); *hw_addr_iface = iface; @@ -1023,101 +1134,500 @@ 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); + } + + shash_destroy_free_data(&sh); + + 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); + } + + 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, + iface_get_carrier(iface) ? "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); } - return retval; } -void -bridge_wait(void) +/* Writes 'iface''s CFM statistics to the database. Returns true if anything + * changed, false otherwise. */ +static bool +iface_refresh_cfm_stats(struct iface *iface) { - struct bridge *br; + const struct ovsrec_monitor *mon; + const struct cfm *cfm; + bool changed = false; + size_t i; - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - ofproto_wait(br->ofproto); - if (br->controller) { - continue; + mon = iface->cfg->monitor; + cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); + + if (!cfm || !mon) { + return false; + } + + 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); + + if (mp->n_fault != 1 || mp->fault[0] != rmp->fault) { + ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1); + changed = true; } + } - mac_learning_wait(br->ml); - bond_wait(br); + if (mon->n_fault != 1 || mon->fault[0] != cfm->fault) { + ovsrec_monitor_set_fault(mon, &cfm->fault, 1); + changed = true; } + + return changed; } -/* 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_stats(struct iface *iface) { - COVERAGE_INC(bridge_flush); - br->flush = true; - mac_learning_flush(br->ml); + 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); } -/* 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) +static void +refresh_system_stats(const struct ovsrec_open_vswitch *cfg) { - size_t i, j; + struct ovsdb_datum datum; + struct shash stats; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->dp_ifidx == ODPP_LOCAL) { - return iface; - } - } - } + shash_init(&stats); + get_system_stats(&stats); - return NULL; + ovsdb_datum_from_shash(&datum, &stats); + ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics, + &datum); } - -/* Bridge unixctl user interface functions. */ + +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_unixctl_fdb_show(struct unixctl_conn *conn, - const char *args, void *aux OVS_UNUSED) +bridge_refresh_controller_status(const struct bridge *br) { - struct ds ds = DS_EMPTY_INITIALIZER; - const struct bridge *br; - const struct mac_entry *e; + struct shash info; + const struct ovsrec_controller *cfg; - br = bridge_lookup(args); - if (!br) { - unixctl_command_reply(conn, 501, "no such bridge"); - return; - } + ofproto_get_ofproto_controller_info(br->ofproto, &info); - ds_put_cstr(&ds, " port VLAN MAC Age\n"); - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { - if (e->port < 0 || e->port >= br->n_ports) { - continue; + 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); } - ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", - br->ports[e->port]->ifaces[0]->dp_ifidx, - e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); - ds_destroy(&ds); + + 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) { + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_refresh_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; + } + + if (time_msec() >= cfm_limiter) { + struct ovsdb_idl_txn *txn; + bool changed = false; + + txn = ovsdb_idl_txn_create(idl); + LIST_FOR_EACH (br, node, &all_bridges) { + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + changed = iface_refresh_cfm_stats(iface) || changed; + } + } + } + + if (changed) { + cfm_limiter = time_msec() + CFM_LIMIT_INTERVAL; + } + + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); + } +} + +void +bridge_wait(void) +{ + struct bridge *br; + + LIST_FOR_EACH (br, node, &all_bridges) { + struct port *port; + + ofproto_wait(br->ofproto); + mac_learning_wait(br->ml); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port_wait(port); + } + } + ovsdb_idl_wait(idl); + poll_timer_wait_until(stats_timer); + + if (cfm_limiter > time_msec()) { + poll_timer_wait_until(cfm_limiter); + } +} + +/* 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; +} + +/* Bridge unixctl user interface functions. */ +static void +bridge_unixctl_fdb_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + const struct bridge *br; + const struct mac_entry *e; + + br = bridge_lookup(args); + if (!br) { + unixctl_command_reply(conn, 501, "no such bridge"); + return; + } + + ds_put_cstr(&ds, " port VLAN MAC Age\n"); + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { + struct port *port = e->port.p; + ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", + port_get_an_iface(port)->dp_ifidx, + e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); + } + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + ds_destroy(&ds); +} + +/* CFM unixctl user interface functions. */ +static void +cfm_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct iface *iface; + const struct cfm *cfm; + + iface = iface_find(args); + if (!iface) { + unixctl_command_reply(conn, 501, "no such interface"); + return; + } + + cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); + + if (!cfm) { + unixctl_command_reply(conn, 501, "CFM not enabled"); + return; + } + + cfm_dump_ds(cfm, &ds); + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + ds_destroy(&ds); +} + +/* QoS unixctl user interface functions. */ + +struct qos_unixctl_show_cbdata { + struct ds *ds; + struct iface *iface; +}; + +static void +qos_unixctl_show_cb(unsigned int queue_id, + const struct shash *details, + void *aux) +{ + struct qos_unixctl_show_cbdata *data = aux; + struct ds *ds = data->ds; + struct iface *iface = data->iface; + struct netdev_queue_stats stats; + struct shash_node *node; + int error; + + ds_put_cstr(ds, "\n"); + if (queue_id) { + ds_put_format(ds, "Queue %u:\n", queue_id); + } else { + ds_put_cstr(ds, "Default:\n"); + } + + SHASH_FOR_EACH (node, details) { + ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data); + } + + error = netdev_get_queue_stats(iface->netdev, queue_id, &stats); + if (!error) { + if (stats.tx_packets != UINT64_MAX) { + ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets); + } + + if (stats.tx_bytes != UINT64_MAX) { + ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes); + } + + if (stats.tx_errors != UINT64_MAX) { + ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors); + } + } else { + ds_put_format(ds, "\tFailed to get statistics for queue %u: %s", + queue_id, strerror(error)); + } +} + +static void +qos_unixctl_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct shash sh = SHASH_INITIALIZER(&sh); + struct iface *iface; + const char *type; + struct shash_node *node; + struct qos_unixctl_show_cbdata data; + int error; + + iface = iface_find(args); + if (!iface) { + unixctl_command_reply(conn, 501, "no such interface"); + return; + } + + netdev_get_qos(iface->netdev, &type, &sh); + + if (*type != '\0') { + ds_put_format(&ds, "QoS: %s %s\n", iface->name, type); + + SHASH_FOR_EACH (node, &sh) { + ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data); + } + + data.ds = &ds; + data.iface = iface; + error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data); + + if (error) { + ds_put_format(&ds, "failed to dump queues: %s", strerror(error)); + } + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + } else { + ds_put_format(&ds, "QoS not configured on %s\n", iface->name); + unixctl_command_reply(conn, 501, ds_cstr(&ds)); + } + + shash_destroy_free_data(&sh); + ds_destroy(&ds); +} + /* Bridge reconfiguration functions. */ static struct bridge * bridge_create(const struct ovsrec_bridge *br_cfg) @@ -1134,7 +1644,6 @@ bridge_create(const struct ovsrec_bridge *br_cfg) free(br); return NULL; } - dpif_flow_flush(br->dpif); error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks, br, &br->ofproto); @@ -1150,10 +1659,11 @@ bridge_create(const struct ovsrec_bridge *br_cfg) br->name = xstrdup(br_cfg->name); br->cfg = br_cfg; br->ml = mac_learning_create(); - br->sent_config_request = false; eth_addr_nicira_random(br->default_ea); - port_array_init(&br->ifaces); + hmap_init(&br->ports); + hmap_init(&br->ifaces); + shash_init(&br->iface_by_name); br->flush = false; @@ -1168,23 +1678,24 @@ static void bridge_destroy(struct bridge *br) { if (br) { + struct port *port, *next; int error; - while (br->n_ports > 0) { - port_destroy(br->ports[br->n_ports - 1]); + HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) { + port_destroy(port); } list_remove(&br->node); + ofproto_destroy(br->ofproto); error = dpif_delete(br->dpif); if (error && error != ENOENT) { VLOG_ERR("failed to delete %s: %s", dpif_name(br->dpif), strerror(error)); } dpif_close(br->dpif); - ofproto_destroy(br->ofproto); - free(br->controller); mac_learning_destroy(br->ml); - port_array_destroy(&br->ifaces); - free(br->ports); + hmap_destroy(&br->ifaces); + hmap_destroy(&br->ports); + shash_destroy(&br->iface_by_name); free(br->name); free(br); } @@ -1195,7 +1706,7 @@ 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; } @@ -1203,19 +1714,6 @@ 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 @@ -1224,7 +1722,7 @@ bridge_unixctl_dump_flows(struct unixctl_conn *conn, { struct bridge *br; struct ds results; - + br = bridge_lookup(args); if (!br) { unixctl_command_reply(conn, 501, "Unknown bridge"); @@ -1238,9 +1736,33 @@ 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) { + struct port *port; int error; error = ofproto_run1(br->ofproto); @@ -1249,7 +1771,10 @@ bridge_run_one(struct bridge *br) } mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); - bond_run(br); + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port_run(port); + } error = ofproto_run2(br->ofproto, br->flush); br->flush = false; @@ -1257,125 +1782,36 @@ bridge_run_one(struct bridge *br) return error; } -static const struct ovsrec_controller * -bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg, - const struct bridge *br) -{ - const struct ovsrec_controller *controller; - - controller = (br->cfg->controller ? br->cfg->controller - : ovs_cfg->controller ? ovs_cfg->controller - : NULL); - - if (controller && !strcmp(controller->target, "none")) { - return NULL; - } - - return controller; -} - -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; - } -} - -static void -bridge_update_desc(struct bridge *br OVS_UNUSED) +static size_t +bridge_get_controllers(const struct bridge *br, + struct ovsrec_controller ***controllersp) { -#if 0 - bool changed = false; - const char *desc; - - desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name); - if (desc != br->mfr_desc) { - free(br->mfr_desc); - if (desc) { - br->mfr_desc = xstrdup(desc); - } else { - br->mfr_desc = xstrdup(DEFAULT_MFR_DESC); - } - changed = true; - } - - desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name); - if (desc != br->hw_desc) { - free(br->hw_desc); - if (desc) { - br->hw_desc = xstrdup(desc); - } else { - br->hw_desc = xstrdup(DEFAULT_HW_DESC); - } - changed = true; - } + struct ovsrec_controller **controllers; + size_t n_controllers; - desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name); - if (desc != br->sw_desc) { - free(br->sw_desc); - if (desc) { - br->sw_desc = xstrdup(desc); - } else { - br->sw_desc = xstrdup(DEFAULT_SW_DESC); - } - changed = true; - } - - desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name); - if (desc != br->serial_desc) { - free(br->serial_desc); - if (desc) { - br->serial_desc = xstrdup(desc); - } else { - br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); - } - changed = true; - } + controllers = br->cfg->controller; + n_controllers = br->cfg->n_controller; - desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name); - if (desc != br->dp_desc) { - free(br->dp_desc); - if (desc) { - br->dp_desc = xstrdup(desc); - } else { - br->dp_desc = xstrdup(DEFAULT_DP_DESC); - } - changed = true; + if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) { + controllers = NULL; + n_controllers = 0; } - if (changed) { - ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc, - br->sw_desc, br->serial_desc, br->dp_desc); + if (controllersp) { + *controllersp = controllers; } -#endif + 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; + enum ofproto_fail_mode fail_mode; + struct port *port, *next; struct shash_node *node; + struct shash new_ports; size_t i; - /* Collect old ports. */ - shash_init(&old_ports); - for (i = 0; i < br->n_ports; i++) { - shash_add(&old_ports, br->ports[i]->name, br->ports[i]); - } - /* Collect new ports. */ shash_init(&new_ports); for (i = 0; i < br->cfg->n_ports; i++) { @@ -1390,7 +1826,7 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, * user didn't specify one. * * XXX perhaps we should synthesize a port ourselves in this case. */ - if (bridge_get_controller(ovs_cfg, br)) { + if (bridge_get_controllers(br, NULL)) { char local_name[IF_NAMESIZE]; int error; @@ -1403,215 +1839,242 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, } } - /* Get rid of deleted ports and add new ports. */ - SHASH_FOR_EACH (node, &old_ports) { - if (!shash_find(&new_ports, node->name)) { - port_destroy(node->data); + /* Get rid of deleted ports. + * Get rid of deleted interfaces on ports that still exist. */ + HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) { + const struct ovsrec_port *port_cfg; + + port_cfg = shash_find_data(&new_ports, port->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); + struct port *port = port_lookup(br, node->name); if (!port) { port = port_create(br, node->name); } + port_reconfigure(port, node->data); + if (!port->n_ifaces) { + 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); + /* Configure OpenFlow controller connection snooping. */ + if (!ofproto_has_snoops(br->ofproto)) { + struct sset snoops; + + sset_init(&snoops); + sset_add_and_free(&snoops, xasprintf("punix:%s/%s.snoop", + ovs_rundir(), br->name)); + ofproto_set_snoops(br->ofproto, &snoops); + sset_destroy(&snoops); } - svec_sort_unique(&listeners); - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - svec_sort_unique(&old_listeners); + mirror_reconfigure(br); +} - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); - } - svec_destroy(&listeners); - svec_destroy(&old_listeners); +/* 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_ofproto_controller_for_mgmt(const struct bridge *br, + struct ofproto_controller *oc) +{ + oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name); + oc->max_backoff = 0; + oc->probe_interval = 60; + oc->band = OFPROTO_OUT_OF_BAND; + oc->rate_limit = 0; + oc->burst_limit = 0; +} - /* 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); +/* 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->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; + + /* If there's no local interface or no IP address, give up. */ + local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL); + if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) { + return; } - svec_sort_unique(&snoops); - svec_init(&old_snoops); - ofproto_get_snoops(br->ofproto, &old_snoops); - svec_sort_unique(&old_snoops); + /* Bring up the local interface. */ + netdev = local_iface->netdev; + netdev_turn_flags_on(netdev, NETDEV_UP, true); - if (!svec_equal(&snoops, &old_snoops)) { - ofproto_set_snoops(br->ofproto, &snoops); + /* 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); } - svec_destroy(&snoops); - svec_destroy(&old_snoops); -#else - /* Default listener. */ - svec_init(&listeners); - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); - } - svec_destroy(&listeners); - svec_destroy(&old_listeners); - - /* Default snoop. */ - svec_init(&snoops); - svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); - svec_init(&old_snoops); - ofproto_get_snoops(br->ofproto, &old_snoops); - if (!svec_equal(&snoops, &old_snoops)) { - ofproto_set_snoops(br->ofproto, &snoops); + 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)); } - svec_destroy(&snoops); - svec_destroy(&old_snoops); -#endif - mirror_reconfigure(br); - - bridge_update_desc(br); + /* 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_controller(const struct ovsrec_open_vswitch *ovs_cfg, - struct bridge *br) +bridge_reconfigure_remotes(struct bridge *br, + const struct sockaddr_in *managers, + size_t n_managers) { - const struct ovsrec_controller *c; + const char *disable_ib_str, *queue_id_str; + bool disable_in_band = false; + int queue_id; - c = bridge_get_controller(ovs_cfg, br); - if ((br->controller != NULL) != (c != NULL)) { - ofproto_flush_flows(br->ofproto); + struct ovsrec_controller **controllers; + size_t n_controllers; + bool had_primary; + + struct ofproto_controller *ocs; + size_t n_ocs; + size_t i; + + /* 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; } - free(br->controller); - br->controller = c ? xstrdup(c->target) : NULL; - if (c) { - int max_backoff, probe; - int 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); - 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 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; - } + 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); - 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)); - } + n_controllers = bridge_get_controllers(br, &controllers); - 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)); - } - } - } + 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; } - ofproto_set_failure(br->ofproto, - (!c->fail_mode - || !strcmp(c->fail_mode, "standalone") - || !strcmp(c->fail_mode, "open"))); + 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++; + } - probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; - ofproto_set_probe_interval(br->ofproto, probe); + ofproto_set_controllers(br->ofproto, ocs, n_ocs); + free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */ + free(ocs); - max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; - ofproto_set_max_backoff(br->ofproto, max_backoff); + if (had_primary != ofproto_has_primary_controller(br->ofproto)) { + ofproto_flush_flows(br->ofproto); + } - rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; - burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0; - ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit); - } else { + /* 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); } - - ofproto_set_controller(br->ofproto, br->controller); } static void bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) { - size_t i, j; + struct port *port; shash_init(ifaces); - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { shash_add_once(ifaces, iface->name, iface); } if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { @@ -1631,116 +2094,141 @@ 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; - size_t i, j; + struct dpif_port_dump dump; + struct dpif_port dpif_port; + struct port *port; /* Reset all interface numbers. */ - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { iface->dp_ifidx = -1; } } - 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)); } - if (iface->cfg) { - int64_t ofport = (iface->dp_ifidx >= 0 - ? odp_port_to_ofp_port(iface->dp_ifidx) - : -1); - ovsrec_interface_set_ofport(iface->cfg, &ofport, 1); - } + 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 && lacp_negotiated(port->lacp); +} + 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, vlan) & BOND_MASK; +} + +static int bond_hash_tcp(const struct flow *flow, uint16_t vlan) { - return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK; + 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 +static struct iface * bond_choose_iface(const struct port *port) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); - size_t i, best_down_slave = -1; - long long next_delay_expiration = LLONG_MAX; - - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; + struct iface *best_down_slave; + struct iface *iface; + best_down_slave = NULL; + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { if (iface->enabled) { - return i; - } else if (iface->delay_expires < next_delay_expiration) { - best_down_slave = i; - next_delay_expiration = iface->delay_expires; + return iface; + } else if ((!best_down_slave + || iface->delay_expires < best_down_slave->delay_expires) + && lacp_slave_may_enable(port->lacp, iface)) { + best_down_slave = iface; } } - if (best_down_slave != -1) { - struct iface *iface = port->ifaces[best_down_slave]; - + if (best_down_slave) { VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay " - "since no other interface is up", iface->name, - iface->delay_expires - time_msec()); - bond_enable_slave(iface, true); + "since no other interface is up", + best_down_slave->name, + best_down_slave->delay_expires - time_msec()); + bond_enable_slave(best_down_slave, true); } return best_down_slave; } static bool -choose_output_iface(const struct port *port, const uint8_t *dl_src, - uint16_t *dp_ifidx, tag_type *tags) +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]; + iface = port_get_an_iface(port); + } else if (port->bond_mode == BM_AB) { + iface = port->active_iface; + if (!iface) { + *tags |= port->no_ifaces_tag; + return false; + } } else { - struct bond_entry *e = lookup_bond_entry(port, dl_src); - if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces - || !port->ifaces[e->iface_idx]->enabled) { + struct bond_entry *e = lookup_bond_entry(port, flow, vlan); + if (!e->iface || !e->iface->enabled) { /* XXX select interface properly. The current interface selection * is only good for testing the rebalancing code. */ - e->iface_idx = bond_choose_iface(port); - if (e->iface_idx < 0) { + e->iface = bond_choose_iface(port); + if (!e->iface) { *tags |= port->no_ifaces_tag; return false; } - e->iface_tag = tag_create_random(); - ((struct port *) port)->bond_compat_is_stale = true; + e->tag = tag_create_random(); } - *tags |= e->iface_tag; - iface = port->ifaces[e->iface_idx]; + *tags |= e->tag; + iface = e->iface; } *dp_ifidx = iface->dp_ifidx; *tags |= iface->tag; /* Currently only used for bonding. */ @@ -1748,36 +2236,46 @@ 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 && lacp_slave_may_enable(port->lacp, iface); + int updelay, downdelay; + + updelay = port->updelay; + downdelay = port->downdelay; + + if (lacp_negotiated(port->lacp)) { + downdelay = 0; + updelay = 0; + } - if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { + 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) { 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); } } @@ -1789,10 +2287,9 @@ bond_choose_active_iface(struct port *port) static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); port->active_iface = bond_choose_iface(port); - port->active_iface_tag = tag_create_random(); - if (port->active_iface >= 0) { + if (port->active_iface) { VLOG_INFO_RL(&rl, "port %s: active interface is now %s", - port->name, port->ifaces[port->active_iface]->name); + port->name, port->active_iface->name); } else { VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface", port->name); @@ -1820,10 +2317,7 @@ bond_enable_slave(struct iface *iface, bool enable) if (!iface->enabled) { VLOG_WARN("interface %s: disabled", iface->name); ofproto_revalidate(br->ofproto, iface->tag); - if (iface->port_ifidx == port->active_iface) { - ofproto_revalidate(br->ofproto, - port->active_iface_tag); - + if (iface == port->active_iface) { /* Disabling a slave can lead to another slave being immediately * enabled if there will be no active slaves but one is waiting * on an updelay. In this case we do not need to run most of the @@ -1836,7 +2330,7 @@ bond_enable_slave(struct iface *iface, bool enable) bond_send_learning_packets(port); } else { VLOG_WARN("interface %s: enabled", iface->name); - if (port->active_iface < 0 && !moving_active_iface) { + if (!port->active_iface && !moving_active_iface) { ofproto_revalidate(br->ofproto, port->no_ifaces_tag); bond_choose_active_iface(port); bond_send_learning_packets(port); @@ -1845,61 +2339,102 @@ bond_enable_slave(struct iface *iface, bool enable) } moving_active_iface = false; - port->bond_compat_is_stale = true; } +/* Attempts to make the sum of the bond slaves' statistics appear on the fake + * bond interface. */ static void -bond_run(struct bridge *br) +bond_update_fake_iface_stats(struct port *port) { - size_t i, j; + struct netdev_stats bond_stats; + struct netdev *bond_dev; + struct iface *iface; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + memset(&bond_stats, 0, sizeof bond_stats); - if (port->n_ifaces >= 2) { - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (time_msec() >= iface->delay_expires) { - bond_enable_slave(iface, !iface->enabled); - } - } - } + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + struct netdev_stats slave_stats; - if (port->bond_compat_is_stale) { - port->bond_compat_is_stale = false; - port_update_bond_compat(port); + if (!netdev_get_stats(iface->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_wait(struct bridge *br) +bond_run(struct port *port) { - size_t i, j; + struct iface *iface; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - if (port->n_ifaces < 2) { - continue; + if (port->n_ifaces < 2) { + return; + } + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + bond_link_status_update(iface); + } + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + if (time_msec() >= iface->delay_expires) { + bond_enable_slave(iface, !iface->enabled); } - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->delay_expires != LLONG_MAX) { - poll_timer_wait(iface->delay_expires - time_msec()); - } + } + + if (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; + } +} + +static void +bond_wait(struct port *port) +{ + struct iface *iface; + + if (port->n_ifaces < 2) { + return; + } + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + if (iface->delay_expires != LLONG_MAX) { + poll_timer_wait_until(iface->delay_expires); } } + + if (port->bond_fake_iface) { + poll_timer_wait_until(port->bond_next_fake_iface_update); + } } static bool -set_dst(struct dst *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 @@ -1916,10 +2451,10 @@ 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: @@ -1954,13 +2489,48 @@ 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; } } @@ -1970,7 +2540,8 @@ 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 @@ -1979,54 +2550,92 @@ 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) +{ + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + if (!ofproto_port_is_floodable(port->bridge->ofproto, + iface->dp_ifidx)) { + return false; + } + } + return true; +} + +/* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if + * there is no active iface. */ +static tag_type +port_get_active_iface_tag(const struct port *port) +{ + return (port->active_iface + ? port->active_iface->tag + : port->no_ifaces_tag); +} + +/* Returns an arbitrary interface within 'port'. + * + * 'port' must have at least one interface. */ +static struct iface * +port_get_an_iface(const struct port *port) +{ + return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem); +} + +static void +compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, - struct dst 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; - size_t i; + struct dst dst; + int flow_vlan; + + 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) + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + if (port != in_port + && port_is_floodable(port) + && port_includes_vlan(port, vlan) && !port->is_mirror_output_port - && set_dst(dst, flow, in_port, port, tags)) { + && 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]; + struct port *port; + + HMAP_FOR_EACH (port, hmap_node, &br->ports) { if (port_includes_vlan(port, m->out_vlan) - && set_dst(dst, flow, in_port, port, tags)) + && 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; } @@ -2036,15 +2645,11 @@ 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); } } } @@ -2052,78 +2657,77 @@ 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 OVS_UNUSED -print_dsts(const struct dst *dsts, size_t n) +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; } @@ -2144,52 +2748,87 @@ static int flow_get_vlan(struct bridge *br, const flow_t *flow, return vlan; } -static void -update_learning_table(struct bridge *br, const flow_t *flow, int vlan, - struct port *in_port) -{ - tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src, - vlan, in_port->port_idx); - if (rev_tag) { +/* 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 struct flow *flow, int vlan, + struct port *in_port) +{ + struct mac_entry *mac; + + if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) { + return; + } + + mac = mac_learning_insert(br->ml, flow->dl_src, vlan); + if (is_gratuitous_arp(flow)) { + /* We don't want to learn from gratuitous ARP packets that are + * reflected back over bond slaves so we lock the learning table. */ + if (in_port->n_ifaces == 1) { + mac_entry_set_grat_arp_lock(mac); + } else if (mac_entry_is_grat_arp_locked(mac)) { + return; + } + } + + if (mac_entry_is_new(mac) || mac->port.p != in_port) { /* The log messages here could actually be useful in debugging, * so keep the rate limit relatively high. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, - 300); + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " "on port %s in VLAN %d", br->name, ETH_ADDR_ARGS(flow->dl_src), in_port->name, vlan); - ofproto_revalidate(br->ofproto, rev_tag); - } -} -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)); + mac->port.p = in_port; + ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac)); + } } -/* 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. */ +/* Determines whether packets in 'flow' within 'br' should be forwarded or + * dropped. Returns true if they may be forwarded, false if they should be + * dropped. + * + * If 'have_packet' is true, it indicates that the caller is processing a + * received packet. If 'have_packet' is false, then the caller is just + * revalidating an existing flow because configuration has changed. Either + * way, 'have_packet' only affects logging (there is no point in logging errors + * during revalidation). + * + * Sets '*in_portp' to the input port. This will be a null pointer if + * flow->in_port does not designate a known input port (in which case + * is_admissible() returns false). + * + * When returning true, sets '*vlanp' to the effective VLAN of the input + * packet, as returned by flow_get_vlan(). + * + * May also add tags to '*tags', although the current implementation only does + * so in one special case. + */ static bool -process_flow(struct bridge *br, const flow_t *flow, - const struct ofpbuf *packet, struct odp_actions *actions, - tag_type *tags, 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 @@ -2204,67 +2843,97 @@ 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; + } + + /* When using LACP, do not accept packets from disabled interfaces. */ + if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) { + return false; } - /* Packets received on bonds need special attention to avoid duplicates. */ - if (in_port->n_ifaces > 1) { - int src_idx; + /* Packets received on non-LACP bonds need special attention to avoid + * duplicates. */ + if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) { + struct mac_entry *mac; if (eth_addr_is_multicast(flow->dl_dst)) { - *tags |= in_port->active_iface_tag; - if (in_port->active_iface != in_iface->port_ifidx) { + *tags |= port_get_active_iface_tag(in_port); + if (in_port->active_iface != in_iface) { /* 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); - if (src_idx != -1 && src_idx != in_port->port_idx && - !is_bcast_arp_reply(flow)) { - goto done; + * 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. */ + mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL); + if (mac && mac->port.p != in_port && + (!is_gratuitous_arp(flow) || mac_entry_is_grat_arp_locked(mac))) { + return false; } } - /* 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; + struct mac_entry *mac; + int vlan; + + /* 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); - if (out_port_idx >= 0 && out_port_idx < br->n_ports) { - out_port = br->ports[out_port_idx]; + mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags); + if (mac) { + out_port = mac->port.p; } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { /* If we are revalidating but don't have a learning entry then * eject the flow. Installing a flow that floods packets opens @@ -2272,6 +2941,8 @@ 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. */ @@ -2280,91 +2951,87 @@ process_flow(struct bridge *br, const flow_t *flow, } done: - compose_actions(br, flow, vlan, in_port, out_port, tags, actions, - nf_output_iface); + if (in_port) { + compose_actions(br, flow, vlan, in_port, out_port, tags, actions, + nf_output_iface); + } return true; } -/* Careful: 'opp' is in host byte order and opp->port_no is an OFP port - * number. */ -static void -bridge_port_changed_ofhook_cb(enum ofp_port_reason reason, - const struct ofp_phy_port *opp, - void *br_) +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 bridge *br = br_; - struct iface *iface; - struct port *port; - - iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no)); - if (!iface) { - return; - } - port = iface->port; - - if (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); - } - 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); - } - } + COVERAGE_INC(bridge_process_flow); + return process_flow(br, flow, packet, actions, tags, nf_output_iface); } 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_) +bridge_special_ofhook_cb(const struct flow *flow, + const struct ofpbuf *packet, void *br_) { + struct iface *iface; struct bridge *br = br_; - COVERAGE_INC(bridge_process_flow); - return process_flow(br, flow, packet, actions, tags, nf_output_iface); + iface = iface_from_dp_ifidx(br, flow->in_port); + + if (flow->dl_type == htons(ETH_TYPE_LACP)) { + + if (iface && iface->port->lacp && packet) { + const struct lacp_pdu *pdu = parse_lacp_packet(packet); + + if (pdu) { + COVERAGE_INC(bridge_process_lacp); + lacp_process_pdu(iface->port->lacp, iface, pdu); + } + } + return false; + } + + return true; } static void -bridge_account_flow_ofhook_cb(const flow_t *flow, - const union odp_action *actions, - size_t n_actions, unsigned long long int n_bytes, - void *br_) +bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, + const struct nlattr *actions, + size_t actions_len, + uint64_t n_bytes, void *br_) { struct bridge *br = br_; + const struct nlattr *a; struct port *in_port; - const union odp_action *a; + 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. */ - 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); - } + /* Feed information from the active flows back into the learning table to + * ensure that table is always in sync with what is actually flowing + * through the datapath. + * + * We test that 'tags' is nonzero to ensure that only flows that include an + * OFPP_NORMAL action are used for learning. This works because + * bridge_normal_ofhook_cb() always sets a nonzero tag value. */ + if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) { + update_learning_table(br, flow, vlan, in_port); } + /* Account for bond slave utilization. */ if (!br->has_bonded_ports) { return; } - - for (a = actions; a < &actions[n_actions]; a++) { - if (a->type == ODPAT_OUTPUT) { - struct port *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); + 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_AB) { + 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; } } @@ -2375,17 +3042,17 @@ static void bridge_account_checkpoint_ofhook_cb(void *br_) { struct bridge *br = br_; + struct port *port; 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 && now >= port->bond_next_rebalance) { + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + if (port->n_ifaces > 1 && port->bond_mode != BM_AB + && now >= port->bond_next_rebalance) { port->bond_next_rebalance = now + port->bond_rebalance_interval; bond_rebalance_port(port); } @@ -2393,8 +3060,8 @@ bridge_account_checkpoint_ofhook_cb(void *br_) } static struct ofhooks bridge_ofhooks = { - bridge_port_changed_ofhook_cb, bridge_normal_ofhook_cb, + bridge_special_ofhook_cb, bridge_account_flow_ofhook_cb, bridge_account_checkpoint_ofhook_cb, }; @@ -2413,6 +3080,22 @@ 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. */ @@ -2423,8 +3106,8 @@ compare_bond_entries(const void *a_, const void *b_) const struct bond_entry *const *bp = b_; const struct bond_entry *a = *ap; const struct bond_entry *b = *bp; - if (a->iface_idx != b->iface_idx) { - return a->iface_idx > b->iface_idx ? 1 : -1; + if (a->iface != b->iface) { + return a->iface > b->iface ? 1 : -1; } else if (a->tx_bytes != b->tx_bytes) { return a->tx_bytes > b->tx_bytes ? 1 : -1; } else { @@ -2521,6 +3204,8 @@ 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_AB); + 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)", @@ -2548,21 +3233,24 @@ bond_shift_load(struct slave_balance *from, struct slave_balance *to, to->tx_bytes += delta; /* Arrange for flows to be revalidated. */ - ofproto_revalidate(port->bridge->ofproto, hash->iface_tag); - hash->iface_idx = to->iface->port_ifidx; - hash->iface_tag = tag_create_random(); + ofproto_revalidate(port->bridge->ofproto, hash->tag); + hash->iface = to->iface; + hash->tag = tag_create_random(); } 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; + struct iface *iface; 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 @@ -2574,25 +3262,34 @@ 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; - for (b = bals; b < &bals[n_bals]; b++) { - b->iface = port->ifaces[b - bals]; + b = bals = xmalloc(n_bals * sizeof *bals); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + b->iface = iface; b->tx_bytes = 0; b->hashes = NULL; b->n_hashes = 0; + b++; } + assert(b == &bals[n_bals]); for (i = 0; i <= BOND_MASK; i++) { hashes[i] = &port->bond_hash[i]; } qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries); for (i = 0; i <= BOND_MASK; i++) { e = hashes[i]; - if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { - b = &bals[e->iface_idx]; - b->tx_bytes += e->tx_bytes; - if (!b->hashes) { - b->hashes = &hashes[i]; + if (!e->iface) { + continue; + } + + for (b = bals; b < &bals[n_bals]; b++) { + if (b->iface == e->iface) { + b->tx_bytes += e->tx_bytes; + if (!b->hashes) { + b->hashes = &hashes[i]; + } + b->n_hashes++; + break; } - b->n_hashes++; } } qsort(bals, n_bals, sizeof *bals, compare_slave_balance); @@ -2603,7 +3300,7 @@ bond_rebalance_port(struct port *port) while (!bals[n_bals - 1].iface->enabled) { n_bals--; if (!n_bals) { - return; + goto exit; } } @@ -2631,7 +3328,6 @@ 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++) { @@ -2671,7 +3367,6 @@ bond_rebalance_port(struct port *port) } if (i < from->n_hashes) { bond_shift_load(from, to, i); - port->bond_compat_is_stale = true; /* If the result of the migration changed the relative order of * 'from' and 'to' swap them back to maintain invariants. */ @@ -2695,7 +3390,13 @@ bond_rebalance_port(struct port *port) * historical data to decay to <1% in 7 rebalancing runs. */ for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) { e->tx_bytes /= 2; + if (!e->tx_bytes) { + e->iface = NULL; + } } + +exit: + free(bals); } static void @@ -2706,45 +3407,33 @@ 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 || 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) { - union ofp_action actions[2], *a; - uint16_t dp_ifidx; + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { tag_type tags = 0; - flow_t flow; + uint16_t dp_ifidx; + struct flow flow; int retval; - if (e->port == port->port_idx - || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) { + if (e->port.p == port) { continue; } - /* Compose actions. */ - memset(actions, 0, sizeof actions); - a = actions; - if (e->vlan) { - a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID); - a->vlan_vid.len = htons(sizeof *a); - a->vlan_vid.vlan_vid = htons(e->vlan); - a++; + 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; } - a->output.type = htons(OFPAT_OUTPUT); - a->output.len = htons(sizeof *a); - a->output.port = htons(odp_port_to_ofp_port(dp_ifidx)); - a++; /* Send packet. */ n_packets++; - compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, - e->mac); - flow_extract(&packet, ODPP_NONE, &flow); - retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, - &packet); + retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet); if (retval) { error = retval; n_errors++; @@ -2772,20 +3461,19 @@ bond_unixctl_list(struct unixctl_conn *conn, 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) { - size_t i; + LIST_FOR_EACH (br, node, &all_bridges) { + struct port *port; - for (i = 0; i < br->n_ports; i++) { - const struct port *port = br->ports[i]; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { if (port->n_ifaces > 1) { - size_t j; + struct iface *iface; - ds_put_format(&ds, "%s\t%s\t", br->name, port->name); - for (j = 0; j < port->n_ifaces; j++) { - const struct iface *iface = port->ifaces[j]; - if (j) { + ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name, + bond_mode_to_string(port->bond_mode)); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + if (&iface->port_elem != list_front(&port->ifaces)) { ds_put_cstr(&ds, ", "); } ds_put_cstr(&ds, iface->name); @@ -2803,11 +3491,10 @@ bond_find(const char *name) { const struct bridge *br; - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { - size_t i; + LIST_FOR_EACH (br, node, &all_bridges) { + struct port *port; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { if (!strcmp(port->name, name) && port->n_ifaces > 1) { return port; } @@ -2822,7 +3509,7 @@ bond_unixctl_show(struct unixctl_conn *conn, { struct ds ds = DS_EMPTY_INITIALIZER; const struct port *port; - size_t j; + struct iface *iface; port = bond_find(args); if (!port) { @@ -2830,18 +3517,46 @@ 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, "lacp: %s\n", + lacp_is_active(port->lacp) ? "active" : "passive"); + } else { + ds_put_cstr(&ds, "lacp: 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->monitor ? "carrier" : "miimon"); + + if (!port->monitor) { + ds_put_format(&ds, "bond-miimon-interval: %lld\n", + port->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->bond_next_rebalance - time_msec()); - for (j = 0; j < port->n_ifaces; j++) { - const struct iface *iface = port->ifaces[j]; + + if (port->bond_mode != BM_AB) { + ds_put_format(&ds, "next rebalance: %lld ms\n", + port->bond_next_rebalance - time_msec()); + } + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { struct bond_entry *be; + struct flow flow; /* Basic info. */ - ds_put_format(&ds, "slave %s: %s\n", + ds_put_format(&ds, "\nslave %s: %s\n", iface->name, iface->enabled ? "enabled" : "disabled"); - if (j == port->active_iface) { + if (iface == port->active_iface) { ds_put_cstr(&ds, "\tactive slave\n"); } if (iface->delay_expires != LLONG_MAX) { @@ -2850,26 +3565,37 @@ bond_unixctl_show(struct unixctl_conn *conn, iface->delay_expires - time_msec()); } + if (port->bond_mode == BM_AB) { + continue; + } + /* Hashes. */ + memset(&flow, 0, sizeof flow); for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { int hash = be - port->bond_hash; struct mac_entry *me; - if (be->iface_idx != j) { + if (be->iface != iface) { continue; } 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 - && me->port != port->port_idx - && choose_output_iface(port, me->mac, &dp_ifidx, &tags) + + memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN); + if (bond_hash_src(me->mac, me->vlan) == hash + && me->port.p != port + && 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", @@ -2889,7 +3615,6 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, 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; @@ -2910,10 +3635,12 @@ 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"); @@ -2932,10 +3659,9 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, } entry = &port->bond_hash[hash]; - ofproto_revalidate(port->bridge->ofproto, entry->iface_tag); - entry->iface_idx = iface->port_ifidx; - entry->iface_tag = tag_create_random(); - port->bond_compat_is_stale = true; + ofproto_revalidate(port->bridge->ofproto, entry->tag); + entry->iface = iface; + entry->tag = tag_create_random(); unixctl_command_reply(conn, 200, "migrated"); } @@ -2974,10 +3700,10 @@ bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, return; } - if (port->active_iface != iface->port_ifidx) { - ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); - port->active_iface = iface->port_ifidx; - port->active_iface_tag = tag_create_random(); + if (port->active_iface != iface) { + ofproto_revalidate(port->bridge->ofproto, + port_get_active_iface_tag(port)); + port->active_iface = iface; VLOG_INFO("port %s: active interface is now %s", port->name, iface->name); bond_send_learning_packets(port); @@ -3035,23 +3761,39 @@ bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, } static void -bond_unixctl_hash(struct unixctl_conn *conn, const char *args, +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 @@ -3071,6 +3813,86 @@ bond_init(void) /* Port functions. */ +static void +lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu) +{ + struct iface *iface = aux; + uint8_t ea[ETH_ADDR_LEN]; + int error; + + error = netdev_get_etheraddr(iface->netdev, ea); + if (!error) { + struct ofpbuf packet; + struct lacp_pdu *packet_pdu; + + ofpbuf_init(&packet, 0); + packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, + sizeof *packet_pdu); + memcpy(packet_pdu, pdu, sizeof *packet_pdu); + ofproto_send_packet(iface->port->bridge->ofproto, + iface->dp_ifidx, 0, &packet); + ofpbuf_uninit(&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)); + } +} + +static void +port_run(struct port *port) +{ + if (port->monitor) { + char *devname; + + /* 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) { + iface_update_carrier(iface); + } + free(devname); + } + } else if (time_msec() >= port->miimon_next_update) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_update_carrier(iface); + } + port->miimon_next_update = time_msec() + port->miimon_interval; + } + + if (port->lacp) { + struct iface *iface; + + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + lacp_slave_enable(port->lacp, iface, iface->enabled); + } + + lacp_run(port->lacp, lacp_send_pdu_cb); + } + + bond_run(port); +} + +static void +port_wait(struct port *port) +{ + if (port->monitor) { + netdev_monitor_poll_wait(port->monitor); + } else { + poll_timer_wait_until(port->miimon_next_update); + } + + if (port->lacp) { + lacp_wait(port->lacp); + } + + bond_wait(port); +} + static struct port * port_create(struct bridge *br, const char *name) { @@ -3078,17 +3900,13 @@ port_create(struct bridge *br, const char *name) port = xzalloc(sizeof *port); port->bridge = br; - port->port_idx = br->n_ports; port->vlan = -1; port->trunks = NULL; port->name = xstrdup(name); - port->active_iface = -1; + port->active_iface = NULL; + list_init(&port->ifaces); - if (br->n_ports >= br->allocated_ports) { - br->ports = x2nrealloc(br->ports, &br->allocated_ports, - sizeof *br->ports); - } - br->ports[br->n_ports++] = port; + hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0)); VLOG_INFO("created port %s on bridge %s", port->name, br->name); bridge_flush(br); @@ -3100,43 +3918,84 @@ static const char * get_port_other_config(const struct ovsrec_port *port, const char *key, const char *default_value) { - const char *value = get_ovsrec_key_value(key, - port->key_other_config, - port->value_other_config, - port->n_other_config); + 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 iface *iface, *next; + struct sset new_ifaces; + size_t i; + + /* Collect list of new interfaces. */ + sset_init(&new_ifaces); + for (i = 0; i < cfg->n_interfaces; i++) { + const char *name = cfg->interfaces[i]->name; + sset_add(&new_ifaces, name); + } + + /* Get rid of deleted interfaces. */ + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + if (!sset_contains(&new_ifaces, iface->name)) { + iface_destroy(iface); + } + } + + sset_destroy(&new_ifaces); +} + +/* Expires all MAC learning entries associated with 'port' and forces ofproto + * to revalidate every flow. */ +static void +port_flush_macs(struct port *port) +{ + struct bridge *br = port->bridge; + struct mac_learning *ml = br->ml; + struct mac_entry *mac, *next_mac; + + bridge_flush(br); + LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { + if (mac->port.p == port) { + mac_learning_expire(ml, mac); + } + } +} + static void port_reconfigure(struct port *port, const struct ovsrec_port *cfg) { - struct shash old_ifaces, new_ifaces; - long long int next_rebalance; - struct shash_node *node; + const char *detect_mode; + struct sset new_ifaces; + long long int next_rebalance, miimon_next_update; + bool need_flush = false; 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]); - } - for (i = 0; i < cfg->n_interfaces; i++) { - const char *name = cfg->interfaces[i]->name; - if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) { - VLOG_WARN("port %s: %s specified twice as port interface", - port->name, name); - } - } + /* Update settings. */ port->updelay = cfg->bond_updelay; if (port->updelay < 0) { port->updelay = 0; } - port->updelay = cfg->bond_downdelay; + port->downdelay = cfg->bond_downdelay; if (port->downdelay < 0) { port->downdelay = 0; } @@ -3150,23 +4009,79 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) port->bond_next_rebalance = next_rebalance; } - /* 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); + detect_mode = get_port_other_config(cfg, "bond-detect-mode", + "carrier"); + + netdev_monitor_destroy(port->monitor); + port->monitor = NULL; + + if (strcmp(detect_mode, "miimon")) { + port->monitor = netdev_monitor_create(); + + if (strcmp(detect_mode, "carrier")) { + VLOG_WARN("port %s: unsupported bond-detect-mode %s, " + "defaulting to carrier", port->name, detect_mode); } } - SHASH_FOR_EACH (node, &new_ifaces) { - const struct ovsrec_interface *if_cfg = node->data; + + port->miimon_interval = atoi( + get_port_other_config(cfg, "bond-miimon-interval", "200")); + if (port->miimon_interval < 100) { + port->miimon_interval = 100; + } + miimon_next_update = time_msec() + port->miimon_interval; + if (port->miimon_next_update > miimon_next_update) { + port->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. */ + sset_init(&new_ifaces); + for (i = 0; i < cfg->n_interfaces; i++) { + const struct ovsrec_interface *if_cfg = cfg->interfaces[i]; struct iface *iface; - iface = shash_find_data(&old_ifaces, if_cfg->name); - if (!iface) { - iface_create(port, if_cfg); - } else { + if (!sset_add(&new_ifaces, if_cfg->name)) { + VLOG_WARN("port %s: %s specified twice as port interface", + port->name, if_cfg->name); + iface_set_ofport(if_cfg, -1); + continue; + } + + iface = iface_lookup(port->bridge, if_cfg->name); + if (iface) { + if (iface->port != port) { + VLOG_ERR("bridge %s: %s interface is on multiple ports, " + "removing from %s", + port->bridge->name, if_cfg->name, iface->port->name); + continue; + } iface->cfg = if_cfg; + } else { + iface = iface_create(port, if_cfg); } + + /* Determine interface type. The local port always has type + * "internal". Other ports take their type from the database and + * default to "system" if none is specified. */ + iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal" + : if_cfg->type[0] ? if_cfg->type + : "system"); } + sset_destroy(&new_ifaces); /* Get VLAN tag. */ vlan = -1; @@ -3187,14 +4102,13 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) } if (port->vlan != vlan) { port->vlan = vlan; - bridge_flush(port->bridge); + need_flush = true; } /* Get trunked VLANs. */ trunks = NULL; - if (vlan < 0) { + if (vlan < 0 && cfg->n_trunks) { size_t n_errors; - size_t i; trunks = bitmap_allocate(4096); n_errors = 0; @@ -3211,28 +4125,26 @@ 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 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) { - bridge_flush(port->bridge); + need_flush = true; } bitmap_free(port->trunks); port->trunks = trunks; - shash_destroy(&old_ifaces); - shash_destroy(&new_ifaces); + if (need_flush) { + port_flush_macs(port); + } } static void @@ -3240,12 +4152,9 @@ port_destroy(struct port *port) { if (port) { struct bridge *br = port->bridge; - struct port *del; + struct iface *iface, *next; int i; - proc_net_compat_update_vlan(port->name, NULL, 0); - proc_net_compat_update_bond(port->name, NULL); - for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; if (m && m->out_port == port) { @@ -3253,18 +4162,22 @@ port_destroy(struct port *port) } } - while (port->n_ifaces > 0) { - iface_destroy(port->ifaces[port->n_ifaces - 1]); + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + iface_destroy(iface); } - del = br->ports[port->port_idx] = br->ports[--br->n_ports]; - del->port_idx = port->port_idx; + hmap_remove(&br->ports, &port->hmap_node); + + VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); - free(port->ifaces); + port_flush_macs(port); + + lacp_destroy(port->lacp); + netdev_monitor_destroy(port->monitor); bitmap_free(port->trunks); + free(port->bond_hash); free(port->name); free(port); - bridge_flush(br); } } @@ -3278,10 +4191,10 @@ 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; + struct port *port; - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0), + &br->ports) { if (!strcmp(port->name, name)) { return port; } @@ -3292,156 +4205,123 @@ port_lookup(const struct bridge *br, const char *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 bool +enable_lacp(struct port *port, bool *activep) +{ + if (!port->cfg->lacp) { + /* XXX when LACP implementation has been sufficiently tested, enable by + * default and make active on bonded ports. */ + return false; + } else if (!strcmp(port->cfg->lacp, "off")) { + return false; + } else if (!strcmp(port->cfg->lacp, "active")) { + *activep = true; + return true; + } else if (!strcmp(port->cfg->lacp, "passive")) { + *activep = false; + return true; + } else { + VLOG_WARN("port %s: unknown LACP mode %s", + port->name, port->cfg->lacp); + return false; } - return NULL; } static void -port_update_bonding(struct port *port) +iface_update_lacp(struct iface *iface) { - 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; - } - } else { - if (!port->bond_hash) { - size_t i; + struct lacp_slave_settings s; + int priority; - port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); - for (i = 0; i <= BOND_MASK; i++) { - struct bond_entry *e = &port->bond_hash[i]; - e->iface_idx = -1; - e->tx_bytes = 0; - } - port->no_ifaces_tag = tag_create_random(); - bond_choose_active_iface(port); - port->bond_next_rebalance - = time_msec() + port->bond_rebalance_interval; - } - port->bond_compat_is_stale = true; - port->bond_fake_iface = port->cfg->bond_fake_iface; - } + s.name = iface->name; + s.id = iface->dp_ifidx; + priority = atoi(get_interface_other_config( + iface->cfg, "lacp-port-priority", "0")); + s.priority = (priority >= 0 && priority <= UINT16_MAX ? priority + : UINT16_MAX); + + lacp_slave_register(iface->port->lacp, iface, &s); } static void -port_update_bond_compat(struct port *port) +port_update_lacp(struct port *port) { - struct compat_bond_hash compat_hashes[BOND_MASK + 1]; - struct compat_bond bond; - size_t i; + struct lacp_settings s; + struct iface *iface; - if (port->n_ifaces < 2) { - proc_net_compat_update_bond(port->name, NULL); + if (!enable_lacp(port, &s.active)) { + lacp_destroy(port->lacp); + port->lacp = NULL; return; } - bond.up = false; - bond.updelay = port->updelay; - bond.downdelay = port->downdelay; - - bond.n_hashes = 0; - bond.hashes = compat_hashes; - if (port->bond_hash) { - const struct bond_entry *e; - for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) { - if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { - struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++]; - cbh->hash = e - port->bond_hash; - cbh->netdev_name = port->ifaces[e->iface_idx]->name; - } - } + if (!port->lacp) { + port->lacp = lacp_create(); } - bond.n_slaves = port->n_ifaces; - bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves); - for (i = 0; i < port->n_ifaces; i++) { - struct iface *iface = port->ifaces[i]; - struct compat_bond_slave *slave = &bond.slaves[i]; - slave->name = iface->name; + s.name = port->name; + memcpy(s.id, port->bridge->ea, ETH_ADDR_LEN); + s.priority = atoi(get_port_other_config(port->cfg, "lacp-system-priority", + "0")); + s.fast = !strcmp(get_port_other_config(port->cfg, "lacp-time", "slow"), + "fast"); - /* We need to make the same determination as the Linux bonding - * code to determine whether a slave should be consider "up". - * The Linux function bond_miimon_inspect() supports four - * BOND_LINK_* states: - * - * - BOND_LINK_UP: carrier detected, updelay has passed. - * - BOND_LINK_FAIL: carrier lost, downdelay in progress. - * - BOND_LINK_DOWN: carrier lost, downdelay has passed. - * - BOND_LINK_BACK: carrier detected, updelay in progress. - * - * The function bond_info_show_slave() only considers BOND_LINK_UP - * to be "up" and anything else to be "down". - */ - slave->up = iface->enabled && iface->delay_expires == LLONG_MAX; - if (slave->up) { - bond.up = true; - } - netdev_get_etheraddr(iface->netdev, slave->mac); + if (s.priority <= 0 || s.priority > UINT16_MAX) { + /* Prefer bondable links if unspecified. */ + s.priority = UINT16_MAX - (port->n_ifaces > 1); } - if (port->bond_fake_iface) { - struct netdev *bond_netdev; - - if (!netdev_open_default(port->name, &bond_netdev)) { - if (bond.up) { - netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); - } else { - netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); - } - netdev_close(bond_netdev); - } + lacp_configure(port->lacp, &s); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_update_lacp(iface); } - - proc_net_compat_update_bond(port->name, &bond); - free(bond.slaves); } static void -port_update_vlan_compat(struct port *port) +port_update_bonding(struct port *port) { - struct bridge *br = port->bridge; - char *vlandev_name = NULL; - - if (port->vlan > 0) { - /* Figure out the name that the VLAN device should actually have, if it - * existed. This takes some work because the VLAN device would not - * have port->name in its name; rather, it would have the trunk port's - * name, and 'port' would be attached to a bridge that also had the - * VLAN device one of its ports. So we need to find a trunk port that - * includes port->vlan. - * - * There might be more than one candidate. This doesn't happen on - * XenServer, so if it happens we just pick the first choice in - * alphabetical order instead of creating multiple VLAN devices. */ + if (port->n_ifaces < 2) { + /* Not a bonded port. */ + free(port->bond_hash); + port->bond_hash = NULL; + port->bond_fake_iface = false; + port->active_iface = NULL; + port->no_ifaces_tag = 0; + } else { size_t i; - for (i = 0; i < br->n_ports; i++) { - struct port *p = br->ports[i]; - if (port_trunks_vlan(p, port->vlan) - && p->n_ifaces - && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0)) - { - uint8_t ea[ETH_ADDR_LEN]; - netdev_get_etheraddr(p->ifaces[0]->netdev, ea); - if (!eth_addr_is_multicast(ea) && - !eth_addr_is_reserved(ea) && - !eth_addr_is_zero(ea)) { - vlandev_name = p->name; - } + + if (port->bond_mode != BM_AB && !port->bond_hash) { + port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); + for (i = 0; i <= BOND_MASK; i++) { + struct bond_entry *e = &port->bond_hash[i]; + e->iface = NULL; + e->tx_bytes = 0; } + port->bond_next_rebalance + = time_msec() + port->bond_rebalance_interval; + } else if (port->bond_mode == BM_AB) { + free(port->bond_hash); + port->bond_hash = NULL; + } + + if (!port->no_ifaces_tag) { + port->no_ifaces_tag = tag_create_random(); } + + if (!port->active_iface) { + bond_choose_active_iface(port); + } + + port->bond_fake_iface = port->cfg->bond_fake_iface; + if (port->bond_fake_iface) { + port->bond_next_fake_iface_update = time_msec(); + } + } - proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan); } /* Interface functions. */ @@ -3449,13 +4329,12 @@ port_update_vlan_compat(struct port *port) static struct iface * iface_create(struct port *port, const struct ovsrec_interface *if_cfg) { + struct bridge *br = port->bridge; struct iface *iface; char *name = if_cfg->name; - int error; iface = xzalloc(sizeof *iface); iface->port = port; - iface->port_ifidx = port->n_ifaces; iface->name = xstrdup(name); iface->dp_ifidx = -1; iface->tag = tag_create_random(); @@ -3463,28 +4342,18 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) iface->netdev = NULL; iface->cfg = if_cfg; - if (port->n_ifaces >= port->allocated_ifaces) { - port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, - sizeof *port->ifaces); - } - port->ifaces[port->n_ifaces++] = iface; - if (port->n_ifaces > 1) { - port->bridge->has_bonded_ports = true; - } + shash_add_assert(&br->iface_by_name, iface->name, iface); - /* Attempt to create the network interface in case it - * doesn't exist yet. */ - if (!iface_is_internal(port->bridge, iface->name)) { - error = set_up_iface(if_cfg, iface, true); - if (error) { - VLOG_WARN("could not create iface %s: %s", iface->name, - strerror(error)); - } + list_push_back(&port->ifaces, &iface->port_elem); + port->n_ifaces++; + + if (port->n_ifaces > 1) { + 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; } @@ -3495,20 +4364,37 @@ iface_destroy(struct iface *iface) if (iface) { struct port *port = iface->port; struct bridge *br = port->bridge; - bool del_active = port->active_iface == iface->port_ifidx; - struct iface *del; + bool del_active = port->active_iface == iface; + + if (port->bond_hash) { + struct bond_entry *e; + for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) { + if (e->iface == iface) { + e->iface = NULL; + } + } + } + + if (iface->port->lacp) { + lacp_slave_unregister(iface->port->lacp, iface); + } + + if (port->monitor && iface->netdev) { + 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; + list_remove(&iface->port_elem); + port->n_ifaces--; netdev_close(iface->netdev); if (del_active) { - ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); bond_choose_active_iface(port); bond_send_learning_packets(port); } @@ -3523,15 +4409,19 @@ iface_destroy(struct iface *iface) static struct iface * iface_lookup(const struct bridge *br, const char *name) { - size_t i, j; + return shash_find_data(&br->iface_by_name, name); +} - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (!strcmp(iface->name, name)) { - return iface; - } +static struct iface * +iface_find(const char *name) +{ + const struct bridge *br; + + LIST_FOR_EACH (br, node, &all_bridges) { + struct iface *iface = iface_lookup(br, name); + + if (iface) { + return iface; } } return NULL; @@ -3540,40 +4430,15 @@ iface_lookup(const struct bridge *br, const char *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 && 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 @@ -3599,57 +4464,239 @@ iface_set_mac(struct iface *iface) } } } + +/* Sets the ofport column of 'if_cfg' to 'ofport'. */ +static void +iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport) +{ + if (if_cfg) { + ovsrec_interface_set_ofport(if_cfg, &ofport, 1); + } +} + +/* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'. + * + * The value strings in '*shash' are taken directly from values[], not copied, + * so the caller should not modify or free them. */ +static void +shash_from_ovs_idl_map(char **keys, char **values, size_t n, + struct shash *shash) +{ + 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_carrier(struct iface *iface) +{ + bool carrier = iface_get_carrier(iface); + if (carrier == iface->up) { + return; + } + + iface->up = carrier; + if (iface->port->lacp) { + lacp_slave_carrier_changed(iface->port->lacp, iface); + } +} + +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 maid[CCM_MAID_LEN]; + + mon = iface->cfg->monitor; + + if (!mon) { + ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx); + return; + } + + if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) { + VLOG_WARN("interface %s: Failed to generate MAID.", iface->name); + return; + } + + cfm.mpid = mon->mpid; + cfm.interval = mon->interval ? *mon->interval : 1000; + + 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; + } + + ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx, + &cfm, remote_mps, mon->n_remote_mps); + free(remote_mps); +} + +/* Read carrier or miimon status directly from 'iface''s netdev, according to + * how 'iface''s port is configured. + * + * Returns true if 'iface' is up, false otherwise. */ +static bool +iface_get_carrier(const struct iface *iface) +{ + return (iface->port->monitor + ? netdev_get_carrier(iface->netdev) + : netdev_get_miimon(iface->netdev)); +} /* Port mirroring. */ +static struct mirror * +mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid) +{ + int i; + + for (i = 0; i < MAX_MIRRORS; i++) { + struct mirror *m = br->mirrors[i]; + if (m && uuid_equals(uuid, &m->uuid)) { + return m; + } + } + return NULL; +} + static void mirror_reconfigure(struct bridge *br) { - struct shash old_mirrors, new_mirrors; - struct shash_node *node; unsigned long *rspan_vlans; + struct port *port; int i; - /* Collect old mirrors. */ - shash_init(&old_mirrors); + /* Get rid of deleted mirrors. */ for (i = 0; i < MAX_MIRRORS; i++) { - if (br->mirrors[i]) { - shash_add(&old_mirrors, br->mirrors[i]->name, br->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); + } } } - /* Collect new mirrors. */ - shash_init(&new_mirrors); + /* Add new mirrors and reconfigure existing ones. */ for (i = 0; i < br->cfg->n_mirrors; i++) { struct ovsrec_mirror *cfg = br->cfg->mirrors[i]; - if (!shash_add_once(&new_mirrors, cfg->name, cfg)) { - VLOG_WARN("bridge %s: %s specified twice as mirror", - br->name, cfg->name); - } - } - - /* Get rid of deleted mirrors and add new mirrors. */ - SHASH_FOR_EACH (node, &old_mirrors) { - if (!shash_find(&new_mirrors, node->name)) { - mirror_destroy(node->data); - } - } - SHASH_FOR_EACH (node, &new_mirrors) { - struct mirror *mirror = shash_find_data(&old_mirrors, node->name); - if (!mirror) { - mirror = mirror_create(br, node->name); - if (!mirror) { - break; - } + struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid); + if (m) { + mirror_reconfigure_one(m, cfg); + } else { + mirror_create(br, cfg); } - mirror_reconfigure_one(mirror, node->data); } - shash_destroy(&old_mirrors); - shash_destroy(&new_mirrors); /* Update port reserved status. */ - for (i = 0; i < br->n_ports; i++) { - br->ports[i]->is_mirror_output_port = false; + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port->is_mirror_output_port = false; } for (i = 0; i < MAX_MIRRORS; i++) { struct mirror *m = br->mirrors[i]; @@ -3677,11 +4724,12 @@ mirror_reconfigure(struct bridge *br) } if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) { bridge_flush(br); + mac_learning_flush(br->ml); } } -static struct mirror * -mirror_create(struct bridge *br, const char *name) +static void +mirror_create(struct bridge *br, struct ovsrec_mirror *cfg) { struct mirror *m; size_t i; @@ -3689,29 +4737,30 @@ 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); - return NULL; + "cannot create %s", br->name, MAX_MIRRORS, cfg->name); + return; } if (!br->mirrors[i]) { break; } } - 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); + mac_learning_flush(br->ml); br->mirrors[i] = m = xzalloc(sizeof *m); m->bridge = br; m->idx = i; - m->name = xstrdup(name); - shash_init(&m->src_ports); - shash_init(&m->dst_ports); + m->name = xstrdup(cfg->name); + sset_init(&m->src_ports); + sset_init(&m->dst_ports); m->vlans = NULL; m->n_vlans = 0; m->out_vlan = -1; m->out_port = NULL; - return m; + mirror_reconfigure_one(m, cfg); } static void @@ -3719,34 +4768,36 @@ mirror_destroy(struct mirror *m) { if (m) { struct bridge *br = m->bridge; - size_t i; + struct port *port; - for (i = 0; i < br->n_ports; i++) { - br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); - br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); + HMAP_FOR_EACH (port, hmap_node, &br->ports) { + port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); + port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); } - shash_destroy(&m->src_ports); - shash_destroy(&m->dst_ports); + sset_destroy(&m->src_ports); + sset_destroy(&m->dst_ports); free(m->vlans); m->bridge->mirrors[m->idx] = NULL; + free(m->name); free(m); bridge_flush(br); + mac_learning_flush(br->ml); } } static void mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports, - struct shash *names) + struct sset *names) { size_t i; for (i = 0; i < n_ports; i++) { const char *name = ports[i]->name; if (port_lookup(m->bridge, name)) { - shash_add_once(names, name, NULL); + sset_add(names, name); } else { VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent " "port %s", m->bridge->name, m->name, name); @@ -3804,13 +4855,19 @@ port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) static void mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) { - struct shash src_ports, dst_ports; + struct sset src_ports, dst_ports; mirror_mask_t mirror_bit; struct port *out_port; + struct port *port; int out_vlan; size_t n_vlans; int *vlans; - size_t i; + + /* Set name. */ + if (strcmp(cfg->name, m->name)) { + free(m->name); + m->name = xstrdup(cfg->name); + } /* Get output port. */ if (cfg->output_port) { @@ -3838,13 +4895,12 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) return; } - shash_init(&src_ports); - shash_init(&dst_ports); + sset_init(&src_ports); + sset_init(&dst_ports); if (cfg->select_all) { - for (i = 0; i < m->bridge->n_ports; i++) { - const char *name = m->bridge->ports[i]->name; - shash_add_once(&src_ports, name, NULL); - shash_add_once(&dst_ports, name, NULL); + HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) { + sset_add(&src_ports, port->name); + sset_add(&dst_ports, port->name); } vlans = NULL; n_vlans = 0; @@ -3860,16 +4916,17 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) } /* Update mirror data. */ - if (!shash_equal_keys(&m->src_ports, &src_ports) - || !shash_equal_keys(&m->dst_ports, &dst_ports) + if (!sset_equals(&m->src_ports, &src_ports) + || !sset_equals(&m->dst_ports, &dst_ports) || m->n_vlans != n_vlans || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans) || m->out_port != out_port || m->out_vlan != out_vlan) { bridge_flush(m->bridge); + mac_learning_flush(m->bridge->ml); } - shash_swap(&m->src_ports, &src_ports); - shash_swap(&m->dst_ports, &dst_ports); + sset_swap(&m->src_ports, &src_ports); + sset_swap(&m->dst_ports, &dst_ports); free(m->vlans); m->vlans = vlans; m->n_vlans = n_vlans; @@ -3878,10 +4935,8 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) /* Update ports. */ mirror_bit = MIRROR_MASK_C(1) << m->idx; - for (i = 0; i < m->bridge->n_ports; i++) { - struct port *port = m->bridge->ports[i]; - - if (shash_find(&m->src_ports, port->name) + HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) { + if (sset_contains(&m->src_ports, port->name) || (m->n_vlans && (!port->vlan ? port_trunks_any_mirrored_vlan(m, port) @@ -3891,7 +4946,7 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) port->src_mirrors &= ~mirror_bit; } - if (shash_find(&m->dst_ports, port->name)) { + if (sset_contains(&m->dst_ports, port->name)) { port->dst_mirrors |= mirror_bit; } else { port->dst_mirrors &= ~mirror_bit; @@ -3899,6 +4954,6 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) } /* Clean up. */ - shash_destroy(&src_ports); - shash_destroy(&dst_ports); + sset_destroy(&src_ports); + sset_destroy(&dst_ports); }