X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=ff9ddfdf5dbf5b7c9e39cce97fdb7078bb7359aa;hb=b1da6250481f6879d0727710eea87d5f61d6a24c;hp=04898f88d9689787c3f344e692b3fc61417668dd;hpb=7cf8b2660f9813fe080a3f4fcc975099cb36417a;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 04898f88d..ff9ddfdf5 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -15,11 +15,13 @@ #include #include "bridge.h" +#include "byte-order.h" #include #include #include #include #include +#include #include #include #include @@ -30,12 +32,14 @@ #include #include #include "bitmap.h" +#include "classifier.h" #include "coverage.h" #include "dirs.h" #include "dpif.h" #include "dynamic-string.h" #include "flow.h" #include "hash.h" +#include "hmap.h" #include "jsonrpc.h" #include "list.h" #include "mac-learning.h" @@ -45,9 +49,9 @@ #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" @@ -55,17 +59,17 @@ #include "socket-util.h" #include "stream-ssl.h" #include "svec.h" +#include "system-stats.h" #include "timeval.h" #include "util.h" #include "unixctl.h" #include "vconn.h" #include "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); struct dst { uint16_t vlan; @@ -81,12 +85,12 @@ struct iface { long long delay_expires; /* Time after which 'enabled' may change. */ /* These members are valid only after bridge_reconfigure() causes them to - * be initialized.*/ + * be initialized. */ + struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */ int dp_ifidx; /* Index within kernel datapath. */ struct netdev *netdev; /* Network device. */ bool enabled; /* May be chosen for flows? */ - - /* This member is only valid *during* bridge_reconfigure(). */ + const char *type; /* Usually same as cfg->type. */ const struct ovsrec_interface *cfg; }; @@ -105,6 +109,7 @@ 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. */ @@ -124,6 +129,7 @@ struct port { int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. * NULL if all VLANs are trunked. */ + const struct ovsrec_port *cfg; char *name; /* An ordinary bridge port has 1 interface. @@ -139,17 +145,15 @@ struct port { 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 bond_next_fake_iface_update; /* Next update to fake bond stats. */ + long long int bond_next_fake_iface_update; /* Time of next update. */ int bond_rebalance_interval; /* Interval between rebalances, in ms. */ long long int bond_next_rebalance; /* Next rebalancing time. */ + struct netdev_monitor *monitor; /* Tracks carrier up/down status. */ /* Port mirroring info. */ mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ bool is_mirror_output_port; /* Does port mirroring send frames here? */ - - /* This member is only valid *during* bridge_reconfigure(). */ - const struct ovsrec_port *cfg; }; #define DP_MAX_PORTS 255 @@ -157,22 +161,15 @@ 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 default_ea[ETH_ADDR_LEN]; /* Default MAC. */ + 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; @@ -186,34 +183,31 @@ 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; 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 size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg, - const struct bridge *br, +static size_t bridge_get_controllers(const struct bridge *br, struct ovsrec_controller ***controllersp); -static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *, - struct bridge *); -static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *, - struct bridge *, +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); @@ -249,68 +243,88 @@ static void port_update_bond_compat(struct port *); static void port_update_vlan_compat(struct port *); static void port_update_bonding(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_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 shash_from_ovs_idl_map(char **keys, char **values, size_t n, + struct shash *); /* 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'. */ +/* Initializes the bridge module, configuring it to obtain its configuration + * from an OVSDB server accessed over 'remote', which should be a string in a + * form acceptable to ovsdb_idl_create(). */ void -bridge_get_ifaces(struct svec *svec) +bridge_init(const char *remote) { - struct bridge *br, *next; - size_t i, j; + /* Create connection to database. */ + idl = ovsdb_idl_create(remote, &ovsrec_idl_class); - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { - for (i = 0; i < br->n_ports; i++) { - struct port *port = br->ports[i]; + ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg); + ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics); + ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids); - for (j = 0; j < port->n_ifaces; j++) { - struct iface *iface = port->ifaces[j]; - if (iface->dp_ifidx < 0) { - VLOG_ERR("%s interface not in datapath %s, ignoring", - iface->name, dpif_name(br->dpif)); - } else { - if (iface->dp_ifidx != ODPP_LOCAL) { - svec_add(svec, iface->name); - } - } - } - } - } + ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); + + ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids); + ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge); + + ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport); + ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics); + ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids); + + /* Register unixctl commands. */ + unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, + NULL); + unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect, + NULL); + bond_init(); } -void -bridge_init(const struct ovsrec_open_vswitch *cfg) +/* 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) { + static bool already_configured_once; struct svec bridge_names; struct svec dpif_names, dpif_types; size_t i; - unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); + /* Only do this once per ovs-vswitchd run. */ + if (already_configured_once) { + return; + } + already_configured_once = true; + + stats_timer = time_msec() + STATS_INTERVAL; + /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */ svec_init(&bridge_names); for (i = 0; i < cfg->n_bridges; i++) { svec_add(&bridge_names, cfg->bridges[i]->name); } svec_sort(&bridge_names); + /* Iterate over all system dpifs and delete any of them that do not appear + * in 'cfg'. */ svec_init(&dpif_names); svec_init(&dpif_types); dp_enumerate_types(&dpif_types); @@ -321,12 +335,14 @@ bridge_init(const struct ovsrec_open_vswitch *cfg) dp_enumerate_names(dpif_types.names[i], &dpif_names); + /* For each dpif... */ 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; + /* ...check whether any of its names is in 'bridge_names'. */ svec_init(&all_names); dpif_get_all_names(dpif, &all_names); for (k = 0; k < all_names.n; k++) { @@ -334,7 +350,10 @@ bridge_init(const struct ovsrec_open_vswitch *cfg) goto found; } } + + /* No. Delete the dpif. */ dpif_delete(dpif); + found: svec_destroy(&all_names); dpif_close(dpif); @@ -344,132 +363,118 @@ bridge_init(const struct ovsrec_open_vswitch *cfg) svec_destroy(&bridge_names); svec_destroy(&dpif_names); svec_destroy(&dpif_types); - - unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, - NULL); - - bond_init(); - bridge_reconfigure(cfg); } -#ifdef HAVE_OPENSSL +/* Initializes 'options' and fills it with the options for 'if_cfg'. Merges + * keys from "options" and "other_config", preferring "options" keys over + * "other_config" keys. + * + * The value strings in '*options' are taken directly from if_cfg, not copied, + * so the caller should not modify or free them. */ static void -bridge_configure_ssl(const struct ovsrec_ssl *ssl) +iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options) { - /* 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); + size_t i; + + shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options, + if_cfg->n_options, options); + + for (i = 0; i < if_cfg->n_other_config; i++) { + char *key = if_cfg->key_other_config[i]; + char *value = if_cfg->value_other_config[i]; + + if (!shash_find_data(options, key)) { + shash_add(options, key, value); + } else { + VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts " + "with \"options\" key %s", if_cfg->name, key, key); + } } } -#endif -/* Attempt to create the network device 'iface_name' through the netdev +/* Returns the type of network device that 'iface' should have. (This is + * ordinarily the same type as the interface, but the network devices for + * "internal" ports have type "system".) */ +static const char * +iface_get_netdev_type(const struct iface *iface) +{ + return !strcmp(iface->type, "internal") ? "system" : iface->type; +} + +/* Attempt to create the network device for 'iface' through the netdev * library. */ static int -set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface, - bool create) +create_iface_netdev(struct iface *iface) { - struct shash_node *node; + struct netdev_options netdev_options; struct shash options; - int error = 0; - 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])); - } - - if (create) { - struct netdev_options netdev_options; - - 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; - } + int error; - error = netdev_open(&netdev_options, &iface->netdev); + memset(&netdev_options, 0, sizeof netdev_options); + netdev_options.name = iface->cfg->name; + netdev_options.type = iface_get_netdev_type(iface); + netdev_options.args = &options; + netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; - 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; + iface_get_options(iface->cfg, &options); - /* An "internal" config type maps to a netdev "system" type. */ - if (iface_type && !strcmp(iface_type, "internal")) { - iface_type = "system"; - } + error = netdev_open(&netdev_options, &iface->netdev); - 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; - } + if (iface->netdev) { + iface->enabled = netdev_get_carrier(iface->netdev); } - 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) +reconfigure_iface_netdev(struct iface *iface) { - return set_up_iface(iface_cfg, iface, false); -} + const char *netdev_type, *iface_type; + struct shash options; + int error; -static bool -check_iface_netdev(struct bridge *br OVS_UNUSED, 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; - } + /* Skip reconfiguration if the device has the wrong type. This shouldn't + * happen, but... */ + iface_type = iface_get_netdev_type(iface); + netdev_type = netdev_get_type(iface->netdev); + if (iface_type && strcmp(netdev_type, iface_type)) { + VLOG_WARN("%s: attempting change device type from %s to %s", + iface->cfg->name, netdev_type, iface_type); + return EINVAL; } - return true; + /* Reconfigure device. */ + iface_get_options(iface->cfg, &options); + error = netdev_reconfigure(iface->netdev, &options); + shash_destroy(&options); + + return error; } +/* Callback for iterate_and_prune_ifaces(). */ static bool -check_iface_dp_ifidx(struct bridge *br, struct iface *iface, - void *aux OVS_UNUSED) +check_iface(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->netdev) { + /* We already reported a related error, don't bother duplicating it. */ + return false; + } + + 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) @@ -481,8 +486,7 @@ 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); } @@ -507,6 +511,7 @@ iterate_and_prune_ifaces(struct bridge *br, if (cb(br, iface, aux)) { j++; } else { + iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } } @@ -529,39 +534,60 @@ iterate_and_prune_ifaces(struct bridge *br, * 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_managers(const struct ovsrec_open_vswitch *ovs_cfg, - struct sockaddr_in **managersp, size_t *n_managersp) +collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg, + struct sockaddr_in **managersp, size_t *n_managersp) { struct sockaddr_in *managers = NULL; size_t n_managers = 0; + struct shash targets; + size_t i; - if (ovs_cfg->n_managers > 0) { - size_t i; + /* Collect all of the potential targets, as the union of the "managers" + * column and the "targets" columns of the rows pointed to by + * "manager_options", excluding any that are out-of-band. */ + shash_init(&targets); + for (i = 0; i < ovs_cfg->n_managers; i++) { + shash_add_once(&targets, ovs_cfg->managers[i], NULL); + } + for (i = 0; i < ovs_cfg->n_manager_options; i++) { + struct ovsrec_manager *m = ovs_cfg->manager_options[i]; - managers = xmalloc(ovs_cfg->n_managers * sizeof *managers); - for (i = 0; i < ovs_cfg->n_managers; i++) { - const char *name = ovs_cfg->managers[i]; - struct sockaddr_in *sin = &managers[i]; + if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) { + shash_find_and_delete(&targets, m->target); + } else { + shash_add_once(&targets, m->target, NULL); + } + } - if ((!strncmp(name, "tcp:", 4) - && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) || - (!strncmp(name, "ssl:", 4) - && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) { + /* Now extract the targets' IP addresses. */ + if (!shash_is_empty(&targets)) { + struct shash_node *node; + + managers = xmalloc(shash_count(&targets) * sizeof *managers); + SHASH_FOR_EACH (node, &targets) { + const char *target = node->name; + struct sockaddr_in *sin = &managers[n_managers]; + + if ((!strncmp(target, "tcp:", 4) + && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) || + (!strncmp(target, "ssl:", 4) + && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) { n_managers++; } } } + shash_destroy(&targets); *managersp = managers; *n_managersp = n_managers; } -void +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; @@ -572,14 +598,12 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) COVERAGE_INC(bridge_reconfigure); - txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl); - - collect_managers(ovs_cfg, &managers, &n_managers); + 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++) { @@ -590,7 +614,7 @@ 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; @@ -616,14 +640,9 @@ 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. @@ -631,7 +650,7 @@ 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) { + LIST_FOR_EACH (br, node, &all_bridges) { struct odp_port *dpif_ports; size_t n_dpif_ports; struct shash want_ifaces; @@ -653,42 +672,68 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) shash_destroy(&want_ifaces); free(dpif_ports); } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { struct odp_port *dpif_ports; size_t n_dpif_ports; struct shash cur_ifaces, want_ifaces; - struct shash_node *node; /* Get the set of interfaces currently in this datapath. */ dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); shash_init(&cur_ifaces); for (i = 0; i < n_dpif_ports; i++) { const char *name = dpif_ports[i].devname; - if (!shash_find(&cur_ifaces, name)) { - shash_add(&cur_ifaces, name, NULL); - } + shash_add_once(&cur_ifaces, name, &dpif_ports[i]); } - 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; - - if (shash_find(&cur_ifaces, if_name)) { - /* Already exists, just reconfigure it. */ + bool internal = !iface || !strcmp(iface->type, "internal"); + struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name); + int error; + + /* 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 (internal + ? dpif_port && !(dpif_port->flags & ODP_PORT_INTERNAL) + : (iface->netdev + && strcmp(iface->type, netdev_get_type(iface->netdev)))) + { + if (dpif_port) { + error = ofproto_port_del(br->ofproto, dpif_port->port); + if (error) { + continue; + } + dpif_port = NULL; + } if (iface) { - reconfigure_iface(iface->cfg, iface); + netdev_close(iface->netdev); + iface->netdev = NULL; } - } else { - /* Need to add to datapath. */ - bool internal; - int error; + } - /* Add to datapath. */ - internal = iface_is_internal(br, if_name); + /* If it's not an internal port, open (possibly create) the + * netdev. */ + if (!internal) { + if (!iface->netdev) { + error = create_iface_netdev(iface); + if (error) { + VLOG_WARN("could not create iface %s: %s", iface->name, + strerror(error)); + continue; + } + } else { + reconfigure_iface_netdev(iface); + } + } + + /* If it's not part of the datapath, add it. */ + if (!dpif_port) { error = dpif_port_add(br->dpif, if_name, internal ? ODP_PORT_INTERNAL : 0, NULL); if (error == EFBIG) { @@ -698,14 +743,30 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } else if (error) { VLOG_ERR("failed to add %s interface to %s: %s", if_name, dpif_name(br->dpif), strerror(error)); + continue; } } + + /* If it's an internal port, open the netdev. */ + if (internal) { + if (iface && !iface->netdev) { + error = create_iface_netdev(iface); + if (error) { + VLOG_WARN("could not create iface %s: %s", iface->name, + strerror(error)); + continue; + } + } + } else { + assert(iface->netdev != NULL); + } } + free(dpif_ports); shash_destroy(&cur_ifaces); shash_destroy(&want_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; @@ -714,8 +775,7 @@ 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); @@ -733,7 +793,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) 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); @@ -779,7 +839,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) opts.collectors.n = nf_cfg->n_targets; opts.collectors.names = nf_cfg->targets; if (ofproto_set_netflow(br->ofproto, &opts)) { - VLOG_ERR("bridge %s: problem setting netflow collectors", + VLOG_ERR("bridge %s: problem setting netflow collectors", br->name); } } else { @@ -792,7 +852,6 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) struct ovsrec_controller **controllers; struct ofproto_sflow_options oso; size_t n_controllers; - size_t i; memset(&oso, 0, sizeof oso); @@ -818,7 +877,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) oso.agent_device = sflow_cfg->agent; oso.control_ip = NULL; - n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers); + 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; @@ -841,48 +900,48 @@ 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_remotes(ovs_cfg, br, managers, n_managers); + bridge_reconfigure_remotes(br, managers, n_managers); } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; + int j; port_update_vlan_compat(port); port_update_bonding(port); + + for (j = 0; j < port->n_ifaces; j++) { + iface_update_qos(port->ifaces[j], port->cfg->qos); + } } } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + 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); - - ovsdb_idl_txn_commit(txn); - ovsdb_idl_txn_destroy(txn); /* XXX */ - free(managers); } 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 @@ -1078,25 +1137,148 @@ dpid_from_hash(const void *data, size_t n) return eth_addr_to_uint64(hash); } -int +static void +iface_refresh_stats(struct iface *iface) +{ + struct iface_stat { + char *name; + int offset; + }; + static const struct iface_stat iface_stats[] = { + { "rx_packets", offsetof(struct netdev_stats, rx_packets) }, + { "tx_packets", offsetof(struct netdev_stats, tx_packets) }, + { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) }, + { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) }, + { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) }, + { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) }, + { "rx_errors", offsetof(struct netdev_stats, rx_errors) }, + { "tx_errors", offsetof(struct netdev_stats, tx_errors) }, + { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) }, + { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) }, + { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) }, + { "collisions", offsetof(struct netdev_stats, collisions) }, + }; + enum { N_STATS = ARRAY_SIZE(iface_stats) }; + const struct iface_stat *s; + + char *keys[N_STATS]; + int64_t values[N_STATS]; + int n; + + struct netdev_stats stats; + + /* Intentionally ignore return value, since errors will set 'stats' to + * all-1s, and we will deal with that correctly below. */ + netdev_get_stats(iface->netdev, &stats); + + n = 0; + for (s = iface_stats; s < &iface_stats[N_STATS]; s++) { + uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset); + if (value != UINT64_MAX) { + keys[n] = s->name; + values[n] = value; + n++; + } + } + + ovsrec_interface_set_statistics(iface->cfg, keys, values, n); +} + +static void +refresh_system_stats(const struct ovsrec_open_vswitch *cfg) +{ + struct ovsdb_datum datum; + struct shash stats; + + shash_init(&stats); + get_system_stats(&stats); + + ovsdb_datum_from_shash(&datum, &stats); + ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics, + &datum); +} + +void bridge_run(void) { - struct bridge *br, *next; - int retval; + const struct ovsrec_open_vswitch *cfg; + + bool datapath_destroyed; + bool database_changed; + struct bridge *br; - retval = 0; - LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { + /* 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); - if (!retval) { - retval = error; + datapath_destroyed = true; + } + } + + /* (Re)configure if necessary. */ + database_changed = ovsdb_idl_run(idl); + cfg = ovsrec_open_vswitch_first(idl); + 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); + } + } + +#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. */ + 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 + + /* Refresh system and interface stats if necessary. */ + if (time_msec() >= stats_timer) { + if (cfg) { + struct ovsdb_idl_txn *txn; + + txn = ovsdb_idl_txn_create(idl); + LIST_FOR_EACH (br, node, &all_bridges) { + size_t i; + + for (i = 0; i < br->n_ports; i++) { + struct port *port = br->ports[i]; + size_t j; + + for (j = 0; j < port->n_ifaces; j++) { + struct iface *iface = port->ifaces[j]; + iface_refresh_stats(iface); + } + } } + refresh_system_stats(cfg); + ovsdb_idl_txn_commit(txn); + ovsdb_idl_txn_destroy(txn); /* XXX */ } + + stats_timer = time_msec() + STATS_INTERVAL; } - return retval; } void @@ -1104,15 +1286,17 @@ bridge_wait(void) { struct bridge *br; - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { ofproto_wait(br->ofproto); - if (ofproto_has_controller(br->ofproto)) { + if (ofproto_has_primary_controller(br->ofproto)) { continue; } mac_learning_wait(br->ml); bond_wait(br); } + ovsdb_idl_wait(idl); + poll_timer_wait_until(stats_timer); } /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s @@ -1161,7 +1345,7 @@ bridge_unixctl_fdb_show(struct unixctl_conn *conn, } ds_put_cstr(&ds, " port VLAN MAC Age\n"); - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { if (e->port < 0 || e->port >= br->n_ports) { continue; } @@ -1205,10 +1389,9 @@ 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->ifaces); shash_init(&br->port_by_name); shash_init(&br->iface_by_name); @@ -1240,7 +1423,7 @@ bridge_destroy(struct bridge *br) dpif_close(br->dpif); ofproto_destroy(br->ofproto); mac_learning_destroy(br->ml); - port_array_destroy(&br->ifaces); + hmap_destroy(&br->ifaces); shash_destroy(&br->port_by_name); shash_destroy(&br->iface_by_name); free(br->ports); @@ -1254,7 +1437,7 @@ 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; } @@ -1262,19 +1445,6 @@ 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 @@ -1283,7 +1453,7 @@ 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"); @@ -1297,6 +1467,29 @@ 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) { @@ -1317,20 +1510,14 @@ bridge_run_one(struct bridge *br) } static size_t -bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg, - const struct bridge *br, +bridge_get_controllers(const struct bridge *br, struct ovsrec_controller ***controllersp) { struct ovsrec_controller **controllers; size_t n_controllers; - if (br->cfg->n_controller) { - controllers = br->cfg->controller; - n_controllers = br->cfg->n_controller; - } else { - controllers = ovs_cfg->controller; - n_controllers = ovs_cfg->n_controller; - } + controllers = br->cfg->controller; + n_controllers = br->cfg->n_controller; if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) { controllers = NULL; @@ -1344,82 +1531,12 @@ bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg, } static void -bridge_update_desc(struct bridge *br OVS_UNUSED) -{ -#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; - } - - 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; - } - - 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 (changed) { - ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc, - br->sw_desc, br->serial_desc, br->dp_desc); - } -#endif -} - -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 listeners, old_listeners; struct svec snoops, old_snoops; struct shash_node *node; + enum ofproto_fail_mode fail_mode; size_t i; /* Collect old ports. */ @@ -1442,7 +1559,7 @@ 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_controllers(ovs_cfg, br, NULL)) { + if (bridge_get_controllers(br, NULL)) { char local_name[IF_NAMESIZE]; int error; @@ -1488,67 +1605,23 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, shash_destroy(&old_ports); shash_destroy(&new_ports); + /* Set the fail-mode */ + fail_mode = !br->cfg->fail_mode + || !strcmp(br->cfg->fail_mode, "standalone") + ? OFPROTO_FAIL_STANDALONE + : OFPROTO_FAIL_SECURE; + if (ofproto_get_fail_mode(br->ofproto) != fail_mode + && !ofproto_has_primary_controller(br->ofproto)) { + ofproto_flush_flows(br->ofproto); + } + ofproto_set_fail_mode(br->ofproto, fail_mode); + /* Delete all flows if we're switching from connected to standalone or vice * versa. (XXX Should we delete all flows if we are switching from one * controller to another?) */ -#if 0 - /* Configure OpenFlow management listeners. */ - svec_init(&listeners); - cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name); - if (!listeners.n) { - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) { - svec_clear(&listeners); - } - svec_sort_unique(&listeners); - - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - svec_sort_unique(&old_listeners); - - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); - } - svec_destroy(&listeners); - svec_destroy(&old_listeners); - /* Configure OpenFlow controller connection snooping. */ svec_init(&snoops); - cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name); - if (!snoops.n) { - svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", - ovs_rundir, br->name)); - } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) { - svec_clear(&snoops); - } - svec_sort_unique(&snoops); - - svec_init(&old_snoops); - ofproto_get_snoops(br->ofproto, &old_snoops); - svec_sort_unique(&old_snoops); - - if (!svec_equal(&snoops, &old_snoops)) { - ofproto_set_snoops(br->ofproto, &snoops); - } - svec_destroy(&snoops); - svec_destroy(&old_snoops); -#else - /* Default listener. */ - svec_init(&listeners); - svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", - ovs_rundir, br->name)); - svec_init(&old_listeners); - ofproto_get_listeners(br->ofproto, &old_listeners); - if (!svec_equal(&listeners, &old_listeners)) { - ofproto_set_listeners(br->ofproto, &listeners); - } - svec_destroy(&listeners); - svec_destroy(&old_listeners); - - /* Default snoop. */ - svec_init(&snoops); svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", ovs_rundir, br->name)); svec_init(&old_snoops); @@ -1558,114 +1631,182 @@ bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, } svec_destroy(&snoops); svec_destroy(&old_snoops); -#endif mirror_reconfigure(br); +} - bridge_update_desc(br); +/* Initializes 'oc' appropriately as a management service controller for + * 'br'. + * + * The caller must free oc->target when it is no longer needed. */ +static void +bridge_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->accept_re = NULL; + oc->update_resolv_conf = false; + oc->rate_limit = 0; + oc->burst_limit = 0; } +/* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */ static void -bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg, - struct bridge *br, +bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c, + struct ofproto_controller *oc) +{ + oc->target = c->target; + oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; + oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; + oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band") + ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); + oc->accept_re = c->discover_accept_regex; + oc->update_resolv_conf = c->discover_update_resolv_conf; + oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; + oc->burst_limit = (c->controller_burst_limit + ? *c->controller_burst_limit : 0); +} + +/* Configures the IP stack for 'br''s local interface properly according to the + * configuration in 'c'. */ +static void +bridge_configure_local_iface_netdev(struct bridge *br, + struct ovsrec_controller *c) +{ + struct netdev *netdev; + struct in_addr mask, gateway; + + struct iface *local_iface; + struct in_addr ip; + + /* Controller discovery does its own TCP/IP configuration later. */ + if (strcmp(c->target, "discover")) { + return; + } + + /* If there's no local interface or no IP address, give up. */ + local_iface = bridge_get_local_iface(br); + if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) { + return; + } + + /* Bring up the local interface. */ + netdev = local_iface->netdev; + netdev_turn_flags_on(netdev, NETDEV_UP, true); + + /* Configure the IP address and netmask. */ + if (!c->local_netmask + || !inet_aton(c->local_netmask, &mask) + || !mask.s_addr) { + mask.s_addr = guess_netmask(ip.s_addr); + } + if (!netdev_set_in4(netdev, ip, mask)) { + VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT, + br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr)); + } + + /* Configure the default gateway. */ + if (c->local_gateway + && inet_aton(c->local_gateway, &gateway) + && gateway.s_addr) { + if (!netdev_add_router(netdev, gateway)) { + VLOG_INFO("bridge %s: configured gateway "IP_FMT, + br->name, IP_ARGS(&gateway.s_addr)); + } + } +} + +static void +bridge_reconfigure_remotes(struct bridge *br, const struct sockaddr_in *managers, size_t n_managers) { + const char *disable_ib_str, *queue_id_str; + bool disable_in_band = false; + int queue_id; + struct ovsrec_controller **controllers; size_t n_controllers; + bool had_primary; - ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); + struct ofproto_controller *ocs; + size_t n_ocs; + size_t i; - n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers); - if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) { - ofproto_flush_flows(br->ofproto); + /* Check if we should disable in-band control on this bridge. */ + disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band"); + if (disable_ib_str && !strcmp(disable_ib_str, "true")) { + disable_in_band = true; } - if (!n_controllers) { - union ofp_action action; - flow_t flow; - - /* Clear out controllers. */ - ofproto_set_controllers(br->ofproto, NULL, 0); + /* 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); - /* 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, OVSFW_ALL, 0, &action, 1, 0); + if (disable_in_band) { + ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0); } else { - struct ofproto_controller *ocs; - size_t i; - - ocs = xmalloc(n_controllers * sizeof *ocs); - for (i = 0; i < n_controllers; i++) { - struct ovsrec_controller *c = controllers[i]; - struct ofproto_controller *oc = &ocs[i]; + ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); + } + had_primary = ofproto_has_primary_controller(br->ofproto); - if (strcmp(c->target, "discover")) { - struct iface *local_iface; - struct in_addr ip; + n_controllers = bridge_get_controllers(br, &controllers); - 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; + ocs = xmalloc((n_controllers + 1) * sizeof *ocs); + n_ocs = 0; - 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; - } + bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]); + for (i = 0; i < n_controllers; i++) { + struct ovsrec_controller *c = controllers[i]; - 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)); - } + if (!strncmp(c->target, "punix:", 6) + || !strncmp(c->target, "unix:", 5)) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - 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)); - } - } - } - } + /* 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; + } - 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->fail = (!c->fail_mode - || !strcmp(c->fail_mode, "standalone") - || !strcmp(c->fail_mode, "open") - ? OFPROTO_FAIL_STANDALONE - : OFPROTO_FAIL_SECURE); - oc->band = (!c->connection_mode - || !strcmp(c->connection_mode, "in-band") - ? OFPROTO_IN_BAND - : OFPROTO_OUT_OF_BAND); - oc->accept_re = c->discover_accept_regex; - oc->update_resolv_conf = c->discover_update_resolv_conf; - oc->rate_limit = (c->controller_rate_limit - ? *c->controller_rate_limit : 0); - oc->burst_limit = (c->controller_burst_limit - ? *c->controller_burst_limit : 0); + 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; } - ofproto_set_controllers(br->ofproto, ocs, n_controllers); - free(ocs); + n_ocs++; + } + + ofproto_set_controllers(br->ofproto, ocs, n_ocs); + free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */ + free(ocs); + + if (had_primary != ofproto_has_primary_controller(br->ofproto)) { + ofproto_flush_flows(br->ofproto); + } + + /* If there are no controllers and the bridge is in standalone + * mode, set up a flow that matches every packet and directs + * them to OFPP_NORMAL (which goes to us). Otherwise, the + * switch is in secure mode and we won't pass any traffic until + * a controller has been defined and it tells us to do so. */ + if (!n_controllers + && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) { + union ofp_action action; + struct cls_rule rule; + + memset(&action, 0, sizeof action); + action.type = htons(OFPAT_OUTPUT); + action.output.len = htons(sizeof action); + action.output.port = htons(OFPP_NORMAL); + cls_rule_init_catchall(&rule, 0); + ofproto_add_flow(br->ofproto, &rule, &action, 1); } } @@ -1710,7 +1851,7 @@ bridge_fetch_dp_ifaces(struct bridge *br) iface->dp_ifidx = -1; } } - port_array_clear(&br->ifaces); + hmap_clear(&br->ifaces); dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); for (i = 0; i < n_dpif_ports; i++) { @@ -1724,16 +1865,15 @@ bridge_fetch_dp_ifaces(struct bridge *br) VLOG_WARN("%s reported interface %"PRIu16" twice", dpif_name(br->dpif), p->port); } else { - port_array_set(&br->ifaces, p->port, iface); iface->dp_ifidx = p->port; + 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); @@ -1930,10 +2070,19 @@ bond_update_fake_iface_stats(struct port *port) struct netdev_stats slave_stats; if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) { - bond_stats.rx_packets += slave_stats.rx_packets; - bond_stats.rx_bytes += slave_stats.rx_bytes; - bond_stats.tx_packets += slave_stats.tx_packets; - bond_stats.tx_bytes += slave_stats.tx_bytes; + /* 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; } } @@ -1952,6 +2101,22 @@ bond_run(struct bridge *br) struct port *port = br->ports[i]; if (port->n_ifaces >= 2) { + 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) { + bool carrier = netdev_get_carrier(iface->netdev); + + bond_link_status_update(iface, carrier); + port_update_bond_compat(port); + } + free(devname); + } + for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (time_msec() >= iface->delay_expires) { @@ -1983,6 +2148,7 @@ bond_wait(struct bridge *br) if (port->n_ifaces < 2) { continue; } + netdev_monitor_poll_wait(port->monitor); for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (iface->delay_expires != LLONG_MAX) { @@ -1996,7 +2162,7 @@ bond_wait(struct bridge *br) } static bool -set_dst(struct dst *p, const flow_t *flow, +set_dst(struct dst *p, const struct flow *flow, const struct port *in_port, const struct port *out_port, tag_type *tags) { @@ -2084,8 +2250,22 @@ port_includes_vlan(const struct port *port, uint16_t vlan) return vlan == port->vlan || port_trunks_vlan(port, vlan); } +static bool +port_is_floodable(const struct port *port) +{ + int i; + + for (i = 0; i < port->n_ifaces; i++) { + if (!ofproto_port_is_floodable(port->bridge->ofproto, + port->ifaces[i]->dp_ifidx)) { + return false; + } + } + return true; +} + static size_t -compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, +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) { @@ -2098,7 +2278,9 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, /* XXX even better, define each VLAN as a datapath port group */ for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - if (port != in_port && port_includes_vlan(port, vlan) + if (port != in_port + && port_is_floodable(port) + && port_includes_vlan(port, vlan) && !port->is_mirror_output_port && set_dst(dst, flow, in_port, port, tags)) { mirrors |= port->dst_mirrors; @@ -2173,7 +2355,7 @@ print_dsts(const struct dst *dsts, size_t n) } 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, uint16_t *nf_output_iface) @@ -2193,8 +2375,9 @@ compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, if (p->vlan == OFP_VLAN_NONE) { odp_actions_add(actions, ODPAT_STRIP_VLAN); } else { - a = odp_actions_add(actions, ODPAT_SET_VLAN_VID); - a->vlan_vid.vlan_vid = htons(p->vlan); + a = odp_actions_add(actions, ODPAT_SET_DL_TCI); + a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK); + a->dl_tci.tci |= htons(flow->dl_vlan_pcp << VLAN_PCP_SHIFT); } cur_vlan = p->vlan; } @@ -2207,7 +2390,7 @@ compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, * 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 @@ -2249,12 +2432,34 @@ static int flow_get_vlan(struct bridge *br, const flow_t *flow, return vlan; } +/* A VM broadcasts a gratuitous ARP to indicate that it has resumed after + * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to + * indicate this; newer upstream kernels use gratuitous ARP requests. */ +static bool +is_gratuitous_arp(const struct flow *flow) +{ + return (flow->dl_type == htons(ETH_TYPE_ARP) + && eth_addr_is_broadcast(flow->dl_dst) + && (flow->nw_proto == ARP_OP_REPLY + || (flow->nw_proto == ARP_OP_REQUEST + && flow->nw_src == flow->nw_dst))); +} + static void -update_learning_table(struct bridge *br, const flow_t *flow, int vlan, +update_learning_table(struct bridge *br, const struct flow *flow, int vlan, struct port *in_port) { - tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src, - vlan, in_port->port_idx); + enum grat_arp_lock_type lock_type; + tag_type rev_tag; + + /* We don't want to learn from gratuitous ARP packets that are reflected + * back over bond slaves so we lock the learning table. */ + lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE : + (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET : + GRAT_ARP_LOCK_CHECK; + + rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx, + lock_type); if (rev_tag) { /* The log messages here could actually be useful in debugging, * so keep the rate limit relatively high. */ @@ -2268,14 +2473,6 @@ update_learning_table(struct bridge *br, const flow_t *flow, int vlan, } } -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)); -} - /* 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. @@ -2297,7 +2494,7 @@ is_bcast_arp_reply(const flow_t *flow) * so in one special case. */ static bool -is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, +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; @@ -2323,7 +2520,7 @@ is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, 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); } *in_portp = NULL; @@ -2354,6 +2551,7 @@ is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, /* Packets received on bonds need special attention to avoid duplicates. */ if (in_port->n_ifaces > 1) { int src_idx; + bool is_grat_arp_locked; if (eth_addr_is_multicast(flow->dl_dst)) { *tags |= in_port->active_iface_tag; @@ -2365,11 +2563,15 @@ is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, /* Drop all packets for which we have learned a different input * port, because we probably sent the packet on one slave and got - * it back on the other. Broadcast ARP replies are an exception - * to this rule: the host has moved to another switch. */ - src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); + * it back on the other. Gratuitous ARP packets are an exception + * to this rule: the host has moved to another switch. The exception + * to the exception is if we locked the learning table to avoid + * reflections on bond slaves. If this is the case, just drop the + * packet now. */ + src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan, + &is_grat_arp_locked); if (src_idx != -1 && src_idx != in_port->port_idx && - !is_bcast_arp_reply(flow)) { + (!is_gratuitous_arp(flow) || is_grat_arp_locked)) { return false; } } @@ -2381,7 +2583,7 @@ is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, * returns true. Otherwise, the actions should only be applied to 'packet', or * not at all, if 'packet' was NULL. */ static bool -process_flow(struct bridge *br, const flow_t *flow, +process_flow(struct bridge *br, const struct flow *flow, const struct ofpbuf *packet, struct odp_actions *actions, tag_type *tags, uint16_t *nf_output_iface) { @@ -2402,7 +2604,8 @@ process_flow(struct bridge *br, const flow_t *flow, } /* Determine output port. */ - out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags); + out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags, + NULL); if (out_port_idx >= 0 && out_port_idx < br->n_ports) { out_port = br->ports[out_port_idx]; } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { @@ -2430,56 +2633,20 @@ done: 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_) -{ - 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); - } - } -} - static bool -bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, +bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, struct odp_actions *actions, tag_type *tags, uint16_t *nf_output_iface, void *br_) { struct bridge *br = br_; COVERAGE_INC(bridge_process_flow); + return process_flow(br, flow, packet, actions, tags, nf_output_iface); } static void -bridge_account_flow_ofhook_cb(const flow_t *flow, +bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, const union odp_action *actions, size_t n_actions, unsigned long long int n_bytes, void *br_) @@ -2487,20 +2654,24 @@ bridge_account_flow_ofhook_cb(const flow_t *flow, struct bridge *br = br_; const union odp_action *a; struct port *in_port; - tag_type tags = 0; + tag_type dummy = 0; 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. */ - if (is_admissible(br, flow, false, &tags, &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); @@ -2535,7 +2706,6 @@ bridge_account_checkpoint_ofhook_cb(void *br_) } static struct ofhooks bridge_ofhooks = { - bridge_port_changed_ofhook_cb, bridge_normal_ofhook_cb, bridge_account_flow_ofhook_cb, bridge_account_checkpoint_ofhook_cb, @@ -2773,7 +2943,6 @@ 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++) { @@ -2854,11 +3023,11 @@ bond_send_learning_packets(struct port *port) ofpbuf_init(&packet, 128); error = n_packets = n_errors = 0; - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { union ofp_action actions[2], *a; uint16_t dp_ifidx; tag_type tags = 0; - flow_t flow; + struct flow flow; int retval; if (e->port == port->port_idx @@ -2916,7 +3085,7 @@ bond_unixctl_list(struct unixctl_conn *conn, ds_put_cstr(&ds, "bridge\tbond\tslaves\n"); - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { size_t i; for (i = 0; i < br->n_ports; i++) { @@ -2945,7 +3114,7 @@ bond_find(const char *name) { const struct bridge *br; - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { size_t i; for (i = 0; i < br->n_ports; i++) { @@ -3005,8 +3174,7 @@ bond_unixctl_show(struct unixctl_conn *conn, hash, be->tx_bytes / 1024); /* 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 @@ -3243,10 +3411,10 @@ 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; } @@ -3291,7 +3459,7 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) if (port->updelay < 0) { port->updelay = 0; } - port->updelay = cfg->bond_downdelay; + port->downdelay = cfg->bond_downdelay; if (port->downdelay < 0) { port->downdelay = 0; } @@ -3314,6 +3482,7 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) { VLOG_WARN("port %s: %s specified twice as port interface", port->name, if_cfg->name); + iface_set_ofport(if_cfg, -1); continue; } @@ -3327,8 +3496,15 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) } iface->cfg = if_cfg; } else { - iface_create(port, if_cfg); + iface = iface_create(port, if_cfg); } + + /* Determine interface type. The local port always has type + * "internal". Other ports take their type from the database and + * default to "system" if none is specified. */ + iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal" + : if_cfg->type[0] ? if_cfg->type + : "system"); } shash_destroy(&new_ifaces); @@ -3358,7 +3534,6 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) trunks = NULL; if (vlan < 0 && cfg->n_trunks) { size_t n_errors; - size_t i; trunks = bitmap_allocate(4096); n_errors = 0; @@ -3420,6 +3595,9 @@ port_destroy(struct port *port) del = br->ports[port->port_idx] = br->ports[--br->n_ports]; del->port_idx = port->port_idx; + VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); + + netdev_monitor_destroy(port->monitor); free(port->ifaces); bitmap_free(port->trunks); free(port->name); @@ -3451,6 +3629,10 @@ port_lookup_iface(const struct port *port, const char *name) static void port_update_bonding(struct port *port) { + if (port->monitor) { + netdev_monitor_destroy(port->monitor); + port->monitor = NULL; + } if (port->n_ifaces < 2) { /* Not a bonded port. */ if (port->bond_hash) { @@ -3460,9 +3642,9 @@ port_update_bonding(struct port *port) port->bond_fake_iface = false; } } else { - if (!port->bond_hash) { - size_t i; + size_t i; + if (!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]; @@ -3480,6 +3662,11 @@ port_update_bonding(struct port *port) } port->bond_compat_is_stale = true; port->bond_fake_iface = port->cfg->bond_fake_iface; + + port->monitor = netdev_monitor_create(); + for (i = 0; i < port->n_ifaces; i++) { + netdev_monitor_add(port->monitor, port->ifaces[i]->netdev); + } } } @@ -3521,15 +3708,15 @@ port_update_bond_compat(struct port *port) /* We need to make the same determination as the Linux bonding * code to determine whether a slave should be consider "up". - * The Linux function bond_miimon_inspect() supports four + * The Linux function bond_miimon_inspect() supports four * BOND_LINK_* states: - * + * * - BOND_LINK_UP: carrier detected, updelay has passed. * - BOND_LINK_FAIL: carrier lost, downdelay in progress. * - BOND_LINK_DOWN: carrier lost, downdelay has passed. * - BOND_LINK_BACK: carrier detected, updelay in progress. * - * The function bond_info_show_slave() only considers BOND_LINK_UP + * The function bond_info_show_slave() only considers BOND_LINK_UP * to be "up" and anything else to be "down". */ slave->up = iface->enabled && iface->delay_expires == LLONG_MAX; @@ -3601,7 +3788,6 @@ 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; @@ -3615,20 +3801,6 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) 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(br, iface->name)) { - error = set_up_iface(if_cfg, iface, true); - if (error) { - VLOG_WARN("could not create iface %s: %s", iface->name, - strerror(error)); - - shash_find_and_delete_assert(&br->iface_by_name, iface->name); - free(iface->name); - free(iface); - return NULL; - } - } - if (port->n_ifaces >= port->allocated_ifaces) { port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, sizeof *port->ifaces); @@ -3657,7 +3829,7 @@ iface_destroy(struct iface *iface) 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]; @@ -3686,40 +3858,16 @@ 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) { 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 @@ -3745,53 +3893,140 @@ 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]); + } +} + +struct iface_delete_queues_cbdata { + struct netdev *netdev; + const struct ovsdb_datum *queues; +}; + +static bool +queue_ids_include(const struct ovsdb_datum *queues, int64_t target) +{ + union ovsdb_atom atom; + + atom.integer = target; + return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX; +} + +static void +iface_delete_queues(unsigned int queue_id, + const struct shash *details OVS_UNUSED, void *cbdata_) +{ + struct iface_delete_queues_cbdata *cbdata = cbdata_; + + if (!queue_ids_include(cbdata->queues, queue_id)) { + netdev_delete_queue(cbdata->netdev, queue_id); + } +} + +static void +iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos) +{ + if (!qos || qos->type[0] == '\0') { + netdev_set_qos(iface->netdev, NULL, NULL); + } else { + struct iface_delete_queues_cbdata cbdata; + struct shash details; + size_t i; + + /* Configure top-level Qos for 'iface'. */ + shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config, + qos->n_other_config, &details); + netdev_set_qos(iface->netdev, qos->type, &details); + shash_destroy(&details); + + /* Deconfigure queues that were deleted. */ + cbdata.netdev = iface->netdev; + cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER, + OVSDB_TYPE_UUID); + netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata); + + /* Configure queues for 'iface'. */ + for (i = 0; i < qos->n_queues; i++) { + const struct ovsrec_queue *queue = qos->value_queues[i]; + unsigned int queue_id = qos->key_queues[i]; + + shash_from_ovs_idl_map(queue->key_other_config, + queue->value_other_config, + queue->n_other_config, &details); + netdev_set_queue(iface->netdev, queue_id, &details); + shash_destroy(&details); + } + } +} /* 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; 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++) { @@ -3826,8 +4061,8 @@ mirror_reconfigure(struct bridge *br) } } -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; @@ -3835,21 +4070,21 @@ 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); br->mirrors[i] = m = xzalloc(sizeof *m); m->bridge = br; m->idx = i; - m->name = xstrdup(name); + m->name = xstrdup(cfg->name); shash_init(&m->src_ports); shash_init(&m->dst_ports); m->vlans = NULL; @@ -3857,7 +4092,7 @@ mirror_create(struct bridge *br, const char *name) m->out_vlan = -1; m->out_port = NULL; - return m; + mirror_reconfigure_one(m, cfg); } static void @@ -3877,6 +4112,7 @@ mirror_destroy(struct mirror *m) free(m->vlans); m->bridge->mirrors[m->idx] = NULL; + free(m->name); free(m); bridge_flush(br); @@ -3958,6 +4194,12 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) 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) { out_port = port_lookup(m->bridge, cfg->output_port->name);