1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
40 #include "dynamic-string.h"
46 #include "mac-learning.h"
50 #include "ofp-print.h"
52 #include "ofproto/netflow.h"
53 #include "ofproto/ofproto.h"
54 #include "ovsdb-data.h"
56 #include "poll-loop.h"
57 #include "proc-net-compat.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_reconfigure);
86 struct dst builtin[32];
91 static void dst_set_init(struct dst_set *);
92 static void dst_set_add(struct dst_set *, const struct dst *);
93 static void dst_set_free(struct dst_set *);
96 /* These members are always valid. */
97 struct port *port; /* Containing port. */
98 size_t port_ifidx; /* Index within containing port. */
99 char *name; /* Host network device name. */
100 tag_type tag; /* Tag associated with this interface. */
101 long long delay_expires; /* Time after which 'enabled' may change. */
103 /* These members are valid only after bridge_reconfigure() causes them to
105 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
106 int dp_ifidx; /* Index within kernel datapath. */
107 struct netdev *netdev; /* Network device. */
108 bool enabled; /* May be chosen for flows? */
109 const char *type; /* Usually same as cfg->type. */
110 struct cfm *cfm; /* Connectivity Fault Management */
111 const struct ovsrec_interface *cfg;
114 #define BOND_MASK 0xff
116 int iface_idx; /* Index of assigned iface, or -1 if none. */
117 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
118 tag_type iface_tag; /* Tag associated with iface_idx. */
121 #define MAX_MIRRORS 32
122 typedef uint32_t mirror_mask_t;
123 #define MIRROR_MASK_C(X) UINT32_C(X)
124 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
126 struct bridge *bridge;
129 struct uuid uuid; /* UUID of this "mirror" record in database. */
131 /* Selection criteria. */
132 struct shash src_ports; /* Name is port name; data is always NULL. */
133 struct shash dst_ports; /* Name is port name; data is always NULL. */
138 struct port *out_port;
142 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
144 struct bridge *bridge;
146 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
147 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
148 * NULL if all VLANs are trunked. */
149 const struct ovsrec_port *cfg;
152 /* An ordinary bridge port has 1 interface.
153 * A bridge port for bonding has at least 2 interfaces. */
154 struct iface **ifaces;
155 size_t n_ifaces, allocated_ifaces;
158 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
159 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
160 tag_type active_iface_tag; /* Tag for bcast flows. */
161 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
162 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
163 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
164 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
165 long long int bond_next_fake_iface_update; /* Time of next update. */
166 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
167 long long int bond_next_rebalance; /* Next rebalancing time. */
168 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
170 /* Port mirroring info. */
171 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
172 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
173 bool is_mirror_output_port; /* Does port mirroring send frames here? */
177 struct list node; /* Node in global list of bridges. */
178 char *name; /* User-specified arbitrary name. */
179 struct mac_learning *ml; /* MAC learning table. */
180 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
181 const struct ovsrec_bridge *cfg;
183 /* OpenFlow switch processing. */
184 struct ofproto *ofproto; /* OpenFlow switch. */
186 /* Kernel datapath information. */
187 struct dpif *dpif; /* Datapath. */
188 struct hmap ifaces; /* Contains "struct iface"s. */
192 size_t n_ports, allocated_ports;
193 struct shash iface_by_name; /* "struct iface"s indexed by name. */
194 struct shash port_by_name; /* "struct port"s indexed by name. */
197 bool has_bonded_ports;
202 /* Port mirroring. */
203 struct mirror *mirrors[MAX_MIRRORS];
206 /* List of all bridges. */
207 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
209 /* OVSDB IDL used to obtain configuration. */
210 static struct ovsdb_idl *idl;
212 /* Each time this timer expires, the bridge fetches systems and interface
213 * statistics and pushes them into the database. */
214 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
215 static long long int stats_timer = LLONG_MIN;
217 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
218 static void bridge_destroy(struct bridge *);
219 static struct bridge *bridge_lookup(const char *name);
220 static unixctl_cb_func bridge_unixctl_dump_flows;
221 static unixctl_cb_func bridge_unixctl_reconnect;
222 static int bridge_run_one(struct bridge *);
223 static size_t bridge_get_controllers(const struct bridge *br,
224 struct ovsrec_controller ***controllersp);
225 static void bridge_reconfigure_one(struct bridge *);
226 static void bridge_reconfigure_remotes(struct bridge *,
227 const struct sockaddr_in *managers,
229 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
230 static void bridge_fetch_dp_ifaces(struct bridge *);
231 static void bridge_flush(struct bridge *);
232 static void bridge_pick_local_hw_addr(struct bridge *,
233 uint8_t ea[ETH_ADDR_LEN],
234 struct iface **hw_addr_iface);
235 static uint64_t bridge_pick_datapath_id(struct bridge *,
236 const uint8_t bridge_ea[ETH_ADDR_LEN],
237 struct iface *hw_addr_iface);
238 static struct iface *bridge_get_local_iface(struct bridge *);
239 static uint64_t dpid_from_hash(const void *, size_t nbytes);
241 static unixctl_cb_func bridge_unixctl_fdb_show;
243 static void bond_init(void);
244 static void bond_run(struct bridge *);
245 static void bond_wait(struct bridge *);
246 static void bond_rebalance_port(struct port *);
247 static void bond_send_learning_packets(struct port *);
248 static void bond_enable_slave(struct iface *iface, bool enable);
250 static struct port *port_create(struct bridge *, const char *name);
251 static void port_reconfigure(struct port *, const struct ovsrec_port *);
252 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
253 static void port_destroy(struct port *);
254 static struct port *port_lookup(const struct bridge *, const char *name);
255 static struct iface *port_lookup_iface(const struct port *, const char *name);
256 static struct port *port_from_dp_ifidx(const struct bridge *,
258 static void port_update_bond_compat(struct port *);
259 static void port_update_vlan_compat(struct port *);
260 static void port_update_bonding(struct port *);
262 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
263 static void mirror_destroy(struct mirror *);
264 static void mirror_reconfigure(struct bridge *);
265 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
266 static bool vlan_is_mirrored(const struct mirror *, int vlan);
268 static struct iface *iface_create(struct port *port,
269 const struct ovsrec_interface *if_cfg);
270 static void iface_destroy(struct iface *);
271 static struct iface *iface_lookup(const struct bridge *, const char *name);
272 static struct iface *iface_from_dp_ifidx(const struct bridge *,
274 static void iface_set_mac(struct iface *);
275 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
276 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
277 static void iface_update_cfm(struct iface *);
278 static void iface_refresh_cfm_stats(struct iface *iface);
279 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
281 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
284 /* Hooks into ofproto processing. */
285 static struct ofhooks bridge_ofhooks;
287 /* Public functions. */
289 /* Initializes the bridge module, configuring it to obtain its configuration
290 * from an OVSDB server accessed over 'remote', which should be a string in a
291 * form acceptable to ovsdb_idl_create(). */
293 bridge_init(const char *remote)
295 /* Create connection to database. */
296 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
298 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
299 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
300 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
302 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
304 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
305 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
307 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
308 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
309 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
311 /* Register unixctl commands. */
312 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
313 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
315 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
323 struct bridge *br, *next_br;
325 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
328 ovsdb_idl_destroy(idl);
331 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
332 * but for which the ovs-vswitchd configuration 'cfg' is required. */
334 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
336 static bool already_configured_once;
337 struct svec bridge_names;
338 struct svec dpif_names, dpif_types;
341 /* Only do this once per ovs-vswitchd run. */
342 if (already_configured_once) {
345 already_configured_once = true;
347 stats_timer = time_msec() + STATS_INTERVAL;
349 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
350 svec_init(&bridge_names);
351 for (i = 0; i < cfg->n_bridges; i++) {
352 svec_add(&bridge_names, cfg->bridges[i]->name);
354 svec_sort(&bridge_names);
356 /* Iterate over all system dpifs and delete any of them that do not appear
358 svec_init(&dpif_names);
359 svec_init(&dpif_types);
360 dp_enumerate_types(&dpif_types);
361 for (i = 0; i < dpif_types.n; i++) {
366 dp_enumerate_names(dpif_types.names[i], &dpif_names);
368 /* For each dpif... */
369 for (j = 0; j < dpif_names.n; j++) {
370 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
372 struct svec all_names;
375 /* ...check whether any of its names is in 'bridge_names'. */
376 svec_init(&all_names);
377 dpif_get_all_names(dpif, &all_names);
378 for (k = 0; k < all_names.n; k++) {
379 if (svec_contains(&bridge_names, all_names.names[k])) {
384 /* No. Delete the dpif. */
388 svec_destroy(&all_names);
393 svec_destroy(&bridge_names);
394 svec_destroy(&dpif_names);
395 svec_destroy(&dpif_types);
398 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
399 * keys from "options" and "other_config", preferring "options" keys over
400 * "other_config" keys. */
402 iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
406 shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options,
407 if_cfg->n_options, options);
409 for (i = 0; i < if_cfg->n_other_config; i++) {
410 char *key = if_cfg->key_other_config[i];
411 char *value = if_cfg->value_other_config[i];
413 if (!shash_find_data(options, key)) {
414 shash_add(options, key, value);
416 VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts "
417 "with \"options\" key %s", if_cfg->name, key, key);
422 /* Callback for iterate_and_prune_ifaces(). */
424 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
426 if (!iface->netdev) {
427 /* We already reported a related error, don't bother duplicating it. */
431 if (iface->dp_ifidx < 0) {
432 VLOG_ERR("%s interface not in %s, dropping",
433 iface->name, dpif_name(br->dpif));
437 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
438 iface->name, iface->dp_ifidx);
442 /* Callback for iterate_and_prune_ifaces(). */
444 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
445 void *aux OVS_UNUSED)
447 /* Set policing attributes. */
448 netdev_set_policing(iface->netdev,
449 iface->cfg->ingress_policing_rate,
450 iface->cfg->ingress_policing_burst);
452 /* Set MAC address of internal interfaces other than the local
454 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
455 iface_set_mac(iface);
461 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
462 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
463 * deletes from 'br' any ports that no longer have any interfaces. */
465 iterate_and_prune_ifaces(struct bridge *br,
466 bool (*cb)(struct bridge *, struct iface *,
472 for (i = 0; i < br->n_ports; ) {
473 struct port *port = br->ports[i];
474 for (j = 0; j < port->n_ifaces; ) {
475 struct iface *iface = port->ifaces[j];
476 if (cb(br, iface, aux)) {
479 iface_set_ofport(iface->cfg, -1);
480 iface_destroy(iface);
484 if (port->n_ifaces) {
487 VLOG_ERR("%s port has no interfaces, dropping", port->name);
493 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
494 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
495 * responsible for freeing '*managersp' (with free()).
497 * You may be asking yourself "why does ovs-vswitchd care?", because
498 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
499 * should not be and in fact is not directly involved in that. But
500 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
501 * it has to tell in-band control where the managers are to enable that.
502 * (Thus, only managers connected in-band are collected.)
505 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
506 struct sockaddr_in **managersp, size_t *n_managersp)
508 struct sockaddr_in *managers = NULL;
509 size_t n_managers = 0;
510 struct shash targets;
513 /* Collect all of the potential targets, as the union of the "managers"
514 * column and the "targets" columns of the rows pointed to by
515 * "manager_options", excluding any that are out-of-band. */
516 shash_init(&targets);
517 for (i = 0; i < ovs_cfg->n_managers; i++) {
518 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
520 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
521 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
523 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
524 shash_find_and_delete(&targets, m->target);
526 shash_add_once(&targets, m->target, NULL);
530 /* Now extract the targets' IP addresses. */
531 if (!shash_is_empty(&targets)) {
532 struct shash_node *node;
534 managers = xmalloc(shash_count(&targets) * sizeof *managers);
535 SHASH_FOR_EACH (node, &targets) {
536 const char *target = node->name;
537 struct sockaddr_in *sin = &managers[n_managers];
539 if ((!strncmp(target, "tcp:", 4)
540 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
541 (!strncmp(target, "ssl:", 4)
542 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
547 shash_destroy(&targets);
549 *managersp = managers;
550 *n_managersp = n_managers;
554 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
556 struct shash old_br, new_br;
557 struct shash_node *node;
558 struct bridge *br, *next;
559 struct sockaddr_in *managers;
562 int sflow_bridge_number;
564 COVERAGE_INC(bridge_reconfigure);
566 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
568 /* Collect old and new bridges. */
571 LIST_FOR_EACH (br, node, &all_bridges) {
572 shash_add(&old_br, br->name, br);
574 for (i = 0; i < ovs_cfg->n_bridges; i++) {
575 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
576 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
577 VLOG_WARN("more than one bridge named %s", br_cfg->name);
581 /* Get rid of deleted bridges and add new bridges. */
582 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
583 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
590 SHASH_FOR_EACH (node, &new_br) {
591 const char *br_name = node->name;
592 const struct ovsrec_bridge *br_cfg = node->data;
593 br = shash_find_data(&old_br, br_name);
595 /* If the bridge datapath type has changed, we need to tear it
596 * down and recreate. */
597 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
599 bridge_create(br_cfg);
602 bridge_create(br_cfg);
605 shash_destroy(&old_br);
606 shash_destroy(&new_br);
608 /* Reconfigure all bridges. */
609 LIST_FOR_EACH (br, node, &all_bridges) {
610 bridge_reconfigure_one(br);
613 /* Add and delete ports on all datapaths.
615 * The kernel will reject any attempt to add a given port to a datapath if
616 * that port already belongs to a different datapath, so we must do all
617 * port deletions before any port additions. */
618 LIST_FOR_EACH (br, node, &all_bridges) {
619 struct odp_port *dpif_ports;
621 struct shash want_ifaces;
623 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
624 bridge_get_all_ifaces(br, &want_ifaces);
625 for (i = 0; i < n_dpif_ports; i++) {
626 const struct odp_port *p = &dpif_ports[i];
627 if (!shash_find(&want_ifaces, p->devname)
628 && strcmp(p->devname, br->name)) {
629 int retval = dpif_port_del(br->dpif, p->port);
631 VLOG_ERR("failed to remove %s interface from %s: %s",
632 p->devname, dpif_name(br->dpif),
637 shash_destroy(&want_ifaces);
640 LIST_FOR_EACH (br, node, &all_bridges) {
641 struct odp_port *dpif_ports;
643 struct shash cur_ifaces, want_ifaces;
645 /* Get the set of interfaces currently in this datapath. */
646 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
647 shash_init(&cur_ifaces);
648 for (i = 0; i < n_dpif_ports; i++) {
649 const char *name = dpif_ports[i].devname;
650 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
653 /* Get the set of interfaces we want on this datapath. */
654 bridge_get_all_ifaces(br, &want_ifaces);
656 hmap_clear(&br->ifaces);
657 SHASH_FOR_EACH (node, &want_ifaces) {
658 const char *if_name = node->name;
659 struct iface *iface = node->data;
660 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
661 const char *type = iface ? iface->type : "internal";
664 /* If we have a port or a netdev already, and it's not the type we
665 * want, then delete the port (if any) and close the netdev (if
667 if ((dpif_port && strcmp(dpif_port->type, type))
668 || (iface && iface->netdev
669 && strcmp(type, netdev_get_type(iface->netdev)))) {
671 error = ofproto_port_del(br->ofproto, dpif_port->port);
678 netdev_close(iface->netdev);
679 iface->netdev = NULL;
683 /* If the port doesn't exist or we don't have the netdev open,
684 * we need to do more work. */
685 if (!dpif_port || (iface && !iface->netdev)) {
686 struct netdev_options options;
687 struct netdev *netdev;
690 /* First open the network device. */
691 options.name = if_name;
693 options.args = &args;
694 options.ethertype = NETDEV_ETH_TYPE_NONE;
698 iface_get_options(iface->cfg, &args);
700 error = netdev_open(&options, &netdev);
701 shash_destroy(&args);
704 VLOG_WARN("could not open network device %s (%s)",
705 if_name, strerror(error));
709 /* Then add the port if we haven't already. */
711 error = dpif_port_add(br->dpif, netdev, NULL);
713 netdev_close(netdev);
714 if (error == EFBIG) {
715 VLOG_ERR("ran out of valid port numbers on %s",
716 dpif_name(br->dpif));
719 VLOG_ERR("failed to add %s interface to %s: %s",
720 if_name, dpif_name(br->dpif),
727 /* Update 'iface'. */
729 iface->netdev = netdev;
730 iface->enabled = netdev_get_carrier(iface->netdev);
732 } else if (iface && iface->netdev) {
736 iface_get_options(iface->cfg, &args);
737 netdev_reconfigure(iface->netdev, &args);
738 shash_destroy(&args);
742 shash_destroy(&cur_ifaces);
743 shash_destroy(&want_ifaces);
745 sflow_bridge_number = 0;
746 LIST_FOR_EACH (br, node, &all_bridges) {
749 struct iface *local_iface;
750 struct iface *hw_addr_iface;
753 bridge_fetch_dp_ifaces(br);
755 iterate_and_prune_ifaces(br, check_iface, NULL);
757 /* Pick local port hardware address, datapath ID. */
758 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
759 local_iface = bridge_get_local_iface(br);
761 int error = netdev_set_etheraddr(local_iface->netdev, ea);
763 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
764 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
765 "Ethernet address: %s",
766 br->name, strerror(error));
770 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
771 ofproto_set_datapath_id(br->ofproto, dpid);
773 dpid_string = xasprintf("%016"PRIx64, dpid);
774 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
777 /* Set NetFlow configuration on this bridge. */
778 if (br->cfg->netflow) {
779 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
780 struct netflow_options opts;
782 memset(&opts, 0, sizeof opts);
784 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
785 if (nf_cfg->engine_type) {
786 opts.engine_type = *nf_cfg->engine_type;
788 if (nf_cfg->engine_id) {
789 opts.engine_id = *nf_cfg->engine_id;
792 opts.active_timeout = nf_cfg->active_timeout;
793 if (!opts.active_timeout) {
794 opts.active_timeout = -1;
795 } else if (opts.active_timeout < 0) {
796 VLOG_WARN("bridge %s: active timeout interval set to negative "
797 "value, using default instead (%d seconds)", br->name,
798 NF_ACTIVE_TIMEOUT_DEFAULT);
799 opts.active_timeout = -1;
802 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
803 if (opts.add_id_to_iface) {
804 if (opts.engine_id > 0x7f) {
805 VLOG_WARN("bridge %s: netflow port mangling may conflict "
806 "with another vswitch, choose an engine id less "
807 "than 128", br->name);
809 if (br->n_ports > 508) {
810 VLOG_WARN("bridge %s: netflow port mangling will conflict "
811 "with another port when more than 508 ports are "
816 opts.collectors.n = nf_cfg->n_targets;
817 opts.collectors.names = nf_cfg->targets;
818 if (ofproto_set_netflow(br->ofproto, &opts)) {
819 VLOG_ERR("bridge %s: problem setting netflow collectors",
823 ofproto_set_netflow(br->ofproto, NULL);
826 /* Set sFlow configuration on this bridge. */
827 if (br->cfg->sflow) {
828 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
829 struct ovsrec_controller **controllers;
830 struct ofproto_sflow_options oso;
831 size_t n_controllers;
833 memset(&oso, 0, sizeof oso);
835 oso.targets.n = sflow_cfg->n_targets;
836 oso.targets.names = sflow_cfg->targets;
838 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
839 if (sflow_cfg->sampling) {
840 oso.sampling_rate = *sflow_cfg->sampling;
843 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
844 if (sflow_cfg->polling) {
845 oso.polling_interval = *sflow_cfg->polling;
848 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
849 if (sflow_cfg->header) {
850 oso.header_len = *sflow_cfg->header;
853 oso.sub_id = sflow_bridge_number++;
854 oso.agent_device = sflow_cfg->agent;
856 oso.control_ip = NULL;
857 n_controllers = bridge_get_controllers(br, &controllers);
858 for (i = 0; i < n_controllers; i++) {
859 if (controllers[i]->local_ip) {
860 oso.control_ip = controllers[i]->local_ip;
864 ofproto_set_sflow(br->ofproto, &oso);
866 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
868 ofproto_set_sflow(br->ofproto, NULL);
871 /* Update the controller and related settings. It would be more
872 * straightforward to call this from bridge_reconfigure_one(), but we
873 * can't do it there for two reasons. First, and most importantly, at
874 * that point we don't know the dp_ifidx of any interfaces that have
875 * been added to the bridge (because we haven't actually added them to
876 * the datapath). Second, at that point we haven't set the datapath ID
877 * yet; when a controller is configured, resetting the datapath ID will
878 * immediately disconnect from the controller, so it's better to set
879 * the datapath ID before the controller. */
880 bridge_reconfigure_remotes(br, managers, n_managers);
882 LIST_FOR_EACH (br, node, &all_bridges) {
883 for (i = 0; i < br->n_ports; i++) {
884 struct port *port = br->ports[i];
887 port_update_vlan_compat(port);
888 port_update_bonding(port);
890 for (j = 0; j < port->n_ifaces; j++) {
891 iface_update_qos(port->ifaces[j], port->cfg->qos);
895 LIST_FOR_EACH (br, node, &all_bridges) {
896 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
899 LIST_FOR_EACH (br, node, &all_bridges) {
901 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
902 iface_update_cfm(iface);
910 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
911 const struct ovsdb_idl_column *column,
914 const struct ovsdb_datum *datum;
915 union ovsdb_atom atom;
918 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
919 atom.string = (char *) key;
920 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
921 return idx == UINT_MAX ? NULL : datum->values[idx].string;
925 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
927 return get_ovsrec_key_value(&br_cfg->header_,
928 &ovsrec_bridge_col_other_config, key);
932 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
933 struct iface **hw_addr_iface)
939 *hw_addr_iface = NULL;
941 /* Did the user request a particular MAC? */
942 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
943 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
944 if (eth_addr_is_multicast(ea)) {
945 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
946 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
947 } else if (eth_addr_is_zero(ea)) {
948 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
954 /* Otherwise choose the minimum non-local MAC address among all of the
956 memset(ea, 0xff, sizeof ea);
957 for (i = 0; i < br->n_ports; i++) {
958 struct port *port = br->ports[i];
959 uint8_t iface_ea[ETH_ADDR_LEN];
962 /* Mirror output ports don't participate. */
963 if (port->is_mirror_output_port) {
967 /* Choose the MAC address to represent the port. */
968 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
969 /* Find the interface with this Ethernet address (if any) so that
970 * we can provide the correct devname to the caller. */
972 for (j = 0; j < port->n_ifaces; j++) {
973 struct iface *candidate = port->ifaces[j];
974 uint8_t candidate_ea[ETH_ADDR_LEN];
975 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
976 && eth_addr_equals(iface_ea, candidate_ea)) {
981 /* Choose the interface whose MAC address will represent the port.
982 * The Linux kernel bonding code always chooses the MAC address of
983 * the first slave added to a bond, and the Fedora networking
984 * scripts always add slaves to a bond in alphabetical order, so
985 * for compatibility we choose the interface with the name that is
986 * first in alphabetical order. */
987 iface = port->ifaces[0];
988 for (j = 1; j < port->n_ifaces; j++) {
989 struct iface *candidate = port->ifaces[j];
990 if (strcmp(candidate->name, iface->name) < 0) {
995 /* The local port doesn't count (since we're trying to choose its
996 * MAC address anyway). */
997 if (iface->dp_ifidx == ODPP_LOCAL) {
1002 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1004 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1005 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1006 iface->name, strerror(error));
1011 /* Compare against our current choice. */
1012 if (!eth_addr_is_multicast(iface_ea) &&
1013 !eth_addr_is_local(iface_ea) &&
1014 !eth_addr_is_reserved(iface_ea) &&
1015 !eth_addr_is_zero(iface_ea) &&
1016 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
1018 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1019 *hw_addr_iface = iface;
1022 if (eth_addr_is_multicast(ea)) {
1023 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1024 *hw_addr_iface = NULL;
1025 VLOG_WARN("bridge %s: using default bridge Ethernet "
1026 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1028 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1029 br->name, ETH_ADDR_ARGS(ea));
1033 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1034 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1035 * an interface on 'br', then that interface must be passed in as
1036 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1037 * 'hw_addr_iface' must be passed in as a null pointer. */
1039 bridge_pick_datapath_id(struct bridge *br,
1040 const uint8_t bridge_ea[ETH_ADDR_LEN],
1041 struct iface *hw_addr_iface)
1044 * The procedure for choosing a bridge MAC address will, in the most
1045 * ordinary case, also choose a unique MAC that we can use as a datapath
1046 * ID. In some special cases, though, multiple bridges will end up with
1047 * the same MAC address. This is OK for the bridges, but it will confuse
1048 * the OpenFlow controller, because each datapath needs a unique datapath
1051 * Datapath IDs must be unique. It is also very desirable that they be
1052 * stable from one run to the next, so that policy set on a datapath
1055 const char *datapath_id;
1058 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1059 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1063 if (hw_addr_iface) {
1065 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1067 * A bridge whose MAC address is taken from a VLAN network device
1068 * (that is, a network device created with vconfig(8) or similar
1069 * tool) will have the same MAC address as a bridge on the VLAN
1070 * device's physical network device.
1072 * Handle this case by hashing the physical network device MAC
1073 * along with the VLAN identifier.
1075 uint8_t buf[ETH_ADDR_LEN + 2];
1076 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1077 buf[ETH_ADDR_LEN] = vlan >> 8;
1078 buf[ETH_ADDR_LEN + 1] = vlan;
1079 return dpid_from_hash(buf, sizeof buf);
1082 * Assume that this bridge's MAC address is unique, since it
1083 * doesn't fit any of the cases we handle specially.
1088 * A purely internal bridge, that is, one that has no non-virtual
1089 * network devices on it at all, is more difficult because it has no
1090 * natural unique identifier at all.
1092 * When the host is a XenServer, we handle this case by hashing the
1093 * host's UUID with the name of the bridge. Names of bridges are
1094 * persistent across XenServer reboots, although they can be reused if
1095 * an internal network is destroyed and then a new one is later
1096 * created, so this is fairly effective.
1098 * When the host is not a XenServer, we punt by using a random MAC
1099 * address on each run.
1101 const char *host_uuid = xenserver_get_host_uuid();
1103 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1104 dpid = dpid_from_hash(combined, strlen(combined));
1110 return eth_addr_to_uint64(bridge_ea);
1114 dpid_from_hash(const void *data, size_t n)
1116 uint8_t hash[SHA1_DIGEST_SIZE];
1118 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1119 sha1_bytes(data, n, hash);
1120 eth_addr_mark_random(hash);
1121 return eth_addr_to_uint64(hash);
1125 iface_refresh_cfm_stats(struct iface *iface)
1129 const struct ovsrec_monitor *mon;
1131 mon = iface->cfg->monitor;
1138 for (i = 0; i < mon->n_remote_mps; i++) {
1139 const struct ovsrec_maintenance_point *mp;
1140 const struct remote_mp *rmp;
1142 mp = mon->remote_mps[i];
1143 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1145 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1148 if (hmap_is_empty(&cfm->x_remote_mps)) {
1149 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1152 struct remote_mp *rmp;
1153 int64_t *x_remote_mps;
1155 length = hmap_count(&cfm->x_remote_mps);
1156 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1159 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1160 x_remote_mps[i++] = rmp->mpid;
1163 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1167 if (hmap_is_empty(&cfm->x_remote_maids)) {
1168 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1171 char **x_remote_maids;
1172 struct remote_maid *rmaid;
1174 length = hmap_count(&cfm->x_remote_maids);
1175 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1178 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1181 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1183 for (j = 0; j < CCM_MAID_LEN; j++) {
1184 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1189 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1191 for (i = 0; i < length; i++) {
1192 free(x_remote_maids[i]);
1194 free(x_remote_maids);
1197 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1201 iface_refresh_stats(struct iface *iface)
1207 static const struct iface_stat iface_stats[] = {
1208 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1209 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1210 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1211 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1212 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1213 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1214 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1215 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1216 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1217 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1218 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1219 { "collisions", offsetof(struct netdev_stats, collisions) },
1221 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1222 const struct iface_stat *s;
1224 char *keys[N_STATS];
1225 int64_t values[N_STATS];
1228 struct netdev_stats stats;
1230 /* Intentionally ignore return value, since errors will set 'stats' to
1231 * all-1s, and we will deal with that correctly below. */
1232 netdev_get_stats(iface->netdev, &stats);
1235 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1236 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1237 if (value != UINT64_MAX) {
1244 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1248 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1250 struct ovsdb_datum datum;
1254 get_system_stats(&stats);
1256 ovsdb_datum_from_shash(&datum, &stats);
1257 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1264 const struct ovsrec_open_vswitch *cfg;
1266 bool datapath_destroyed;
1267 bool database_changed;
1270 /* Let each bridge do the work that it needs to do. */
1271 datapath_destroyed = false;
1272 LIST_FOR_EACH (br, node, &all_bridges) {
1273 int error = bridge_run_one(br);
1275 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1276 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1277 "forcing reconfiguration", br->name);
1278 datapath_destroyed = true;
1282 /* (Re)configure if necessary. */
1283 database_changed = ovsdb_idl_run(idl);
1284 cfg = ovsrec_open_vswitch_first(idl);
1285 if (database_changed || datapath_destroyed) {
1287 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1289 bridge_configure_once(cfg);
1290 bridge_reconfigure(cfg);
1292 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1293 ovsdb_idl_txn_commit(txn);
1294 ovsdb_idl_txn_destroy(txn); /* XXX */
1296 /* We still need to reconfigure to avoid dangling pointers to
1297 * now-destroyed ovsrec structures inside bridge data. */
1298 static const struct ovsrec_open_vswitch null_cfg;
1300 bridge_reconfigure(&null_cfg);
1305 /* Re-configure SSL. We do this on every trip through the main loop,
1306 * instead of just when the database changes, because the contents of the
1307 * key and certificate files can change without the database changing. */
1308 if (cfg && cfg->ssl) {
1309 const struct ovsrec_ssl *ssl = cfg->ssl;
1311 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1312 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1316 /* Refresh system and interface stats if necessary. */
1317 if (time_msec() >= stats_timer) {
1319 struct ovsdb_idl_txn *txn;
1321 txn = ovsdb_idl_txn_create(idl);
1322 LIST_FOR_EACH (br, node, &all_bridges) {
1325 for (i = 0; i < br->n_ports; i++) {
1326 struct port *port = br->ports[i];
1329 for (j = 0; j < port->n_ifaces; j++) {
1330 struct iface *iface = port->ifaces[j];
1331 iface_refresh_stats(iface);
1332 iface_refresh_cfm_stats(iface);
1336 refresh_system_stats(cfg);
1337 ovsdb_idl_txn_commit(txn);
1338 ovsdb_idl_txn_destroy(txn); /* XXX */
1341 stats_timer = time_msec() + STATS_INTERVAL;
1349 struct iface *iface;
1351 LIST_FOR_EACH (br, node, &all_bridges) {
1352 ofproto_wait(br->ofproto);
1353 if (ofproto_has_primary_controller(br->ofproto)) {
1357 mac_learning_wait(br->ml);
1360 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1362 cfm_wait(iface->cfm);
1366 ovsdb_idl_wait(idl);
1367 poll_timer_wait_until(stats_timer);
1370 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1371 * configuration changes. */
1373 bridge_flush(struct bridge *br)
1375 COVERAGE_INC(bridge_flush);
1377 mac_learning_flush(br->ml);
1380 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1381 * such interface. */
1382 static struct iface *
1383 bridge_get_local_iface(struct bridge *br)
1387 for (i = 0; i < br->n_ports; i++) {
1388 struct port *port = br->ports[i];
1389 for (j = 0; j < port->n_ifaces; j++) {
1390 struct iface *iface = port->ifaces[j];
1391 if (iface->dp_ifidx == ODPP_LOCAL) {
1400 /* Bridge unixctl user interface functions. */
1402 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1403 const char *args, void *aux OVS_UNUSED)
1405 struct ds ds = DS_EMPTY_INITIALIZER;
1406 const struct bridge *br;
1407 const struct mac_entry *e;
1409 br = bridge_lookup(args);
1411 unixctl_command_reply(conn, 501, "no such bridge");
1415 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1416 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1417 if (e->port < 0 || e->port >= br->n_ports) {
1420 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1421 br->ports[e->port]->ifaces[0]->dp_ifidx,
1422 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1424 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1428 /* Bridge reconfiguration functions. */
1429 static struct bridge *
1430 bridge_create(const struct ovsrec_bridge *br_cfg)
1435 assert(!bridge_lookup(br_cfg->name));
1436 br = xzalloc(sizeof *br);
1438 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1444 dpif_flow_flush(br->dpif);
1446 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1449 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1451 dpif_delete(br->dpif);
1452 dpif_close(br->dpif);
1457 br->name = xstrdup(br_cfg->name);
1459 br->ml = mac_learning_create();
1460 eth_addr_nicira_random(br->default_ea);
1462 hmap_init(&br->ifaces);
1464 shash_init(&br->port_by_name);
1465 shash_init(&br->iface_by_name);
1469 list_push_back(&all_bridges, &br->node);
1471 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1477 bridge_destroy(struct bridge *br)
1482 while (br->n_ports > 0) {
1483 port_destroy(br->ports[br->n_ports - 1]);
1485 list_remove(&br->node);
1486 error = dpif_delete(br->dpif);
1487 if (error && error != ENOENT) {
1488 VLOG_ERR("failed to delete %s: %s",
1489 dpif_name(br->dpif), strerror(error));
1491 dpif_close(br->dpif);
1492 ofproto_destroy(br->ofproto);
1493 mac_learning_destroy(br->ml);
1494 hmap_destroy(&br->ifaces);
1495 shash_destroy(&br->port_by_name);
1496 shash_destroy(&br->iface_by_name);
1503 static struct bridge *
1504 bridge_lookup(const char *name)
1508 LIST_FOR_EACH (br, node, &all_bridges) {
1509 if (!strcmp(br->name, name)) {
1516 /* Handle requests for a listing of all flows known by the OpenFlow
1517 * stack, including those normally hidden. */
1519 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1520 const char *args, void *aux OVS_UNUSED)
1525 br = bridge_lookup(args);
1527 unixctl_command_reply(conn, 501, "Unknown bridge");
1532 ofproto_get_all_flows(br->ofproto, &results);
1534 unixctl_command_reply(conn, 200, ds_cstr(&results));
1535 ds_destroy(&results);
1538 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1539 * connections and reconnect. If BRIDGE is not specified, then all bridges
1540 * drop their controller connections and reconnect. */
1542 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1543 const char *args, void *aux OVS_UNUSED)
1546 if (args[0] != '\0') {
1547 br = bridge_lookup(args);
1549 unixctl_command_reply(conn, 501, "Unknown bridge");
1552 ofproto_reconnect_controllers(br->ofproto);
1554 LIST_FOR_EACH (br, node, &all_bridges) {
1555 ofproto_reconnect_controllers(br->ofproto);
1558 unixctl_command_reply(conn, 200, NULL);
1562 bridge_run_one(struct bridge *br)
1565 struct iface *iface;
1567 error = ofproto_run1(br->ofproto);
1572 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1575 error = ofproto_run2(br->ofproto, br->flush);
1578 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1579 struct ofpbuf *packet;
1585 packet = cfm_run(iface->cfm);
1587 iface_send_packet(iface, packet);
1588 ofpbuf_uninit(packet);
1597 bridge_get_controllers(const struct bridge *br,
1598 struct ovsrec_controller ***controllersp)
1600 struct ovsrec_controller **controllers;
1601 size_t n_controllers;
1603 controllers = br->cfg->controller;
1604 n_controllers = br->cfg->n_controller;
1606 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1612 *controllersp = controllers;
1614 return n_controllers;
1618 bridge_reconfigure_one(struct bridge *br)
1620 struct shash old_ports, new_ports;
1621 struct svec snoops, old_snoops;
1622 struct shash_node *node;
1623 enum ofproto_fail_mode fail_mode;
1626 /* Collect old ports. */
1627 shash_init(&old_ports);
1628 for (i = 0; i < br->n_ports; i++) {
1629 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1632 /* Collect new ports. */
1633 shash_init(&new_ports);
1634 for (i = 0; i < br->cfg->n_ports; i++) {
1635 const char *name = br->cfg->ports[i]->name;
1636 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1637 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1642 /* If we have a controller, then we need a local port. Complain if the
1643 * user didn't specify one.
1645 * XXX perhaps we should synthesize a port ourselves in this case. */
1646 if (bridge_get_controllers(br, NULL)) {
1647 char local_name[IF_NAMESIZE];
1650 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1651 local_name, sizeof local_name);
1652 if (!error && !shash_find(&new_ports, local_name)) {
1653 VLOG_WARN("bridge %s: controller specified but no local port "
1654 "(port named %s) defined",
1655 br->name, local_name);
1659 /* Get rid of deleted ports.
1660 * Get rid of deleted interfaces on ports that still exist. */
1661 SHASH_FOR_EACH (node, &old_ports) {
1662 struct port *port = node->data;
1663 const struct ovsrec_port *port_cfg;
1665 port_cfg = shash_find_data(&new_ports, node->name);
1669 port_del_ifaces(port, port_cfg);
1673 /* Create new ports.
1674 * Add new interfaces to existing ports.
1675 * Reconfigure existing ports. */
1676 SHASH_FOR_EACH (node, &new_ports) {
1677 struct port *port = shash_find_data(&old_ports, node->name);
1679 port = port_create(br, node->name);
1682 port_reconfigure(port, node->data);
1683 if (!port->n_ifaces) {
1684 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1685 br->name, port->name);
1689 shash_destroy(&old_ports);
1690 shash_destroy(&new_ports);
1692 /* Set the fail-mode */
1693 fail_mode = !br->cfg->fail_mode
1694 || !strcmp(br->cfg->fail_mode, "standalone")
1695 ? OFPROTO_FAIL_STANDALONE
1696 : OFPROTO_FAIL_SECURE;
1697 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1698 && !ofproto_has_primary_controller(br->ofproto)) {
1699 ofproto_flush_flows(br->ofproto);
1701 ofproto_set_fail_mode(br->ofproto, fail_mode);
1703 /* Delete all flows if we're switching from connected to standalone or vice
1704 * versa. (XXX Should we delete all flows if we are switching from one
1705 * controller to another?) */
1707 /* Configure OpenFlow controller connection snooping. */
1709 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1710 ovs_rundir(), br->name));
1711 svec_init(&old_snoops);
1712 ofproto_get_snoops(br->ofproto, &old_snoops);
1713 if (!svec_equal(&snoops, &old_snoops)) {
1714 ofproto_set_snoops(br->ofproto, &snoops);
1716 svec_destroy(&snoops);
1717 svec_destroy(&old_snoops);
1719 mirror_reconfigure(br);
1722 /* Initializes 'oc' appropriately as a management service controller for
1725 * The caller must free oc->target when it is no longer needed. */
1727 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1728 struct ofproto_controller *oc)
1730 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1731 oc->max_backoff = 0;
1732 oc->probe_interval = 60;
1733 oc->band = OFPROTO_OUT_OF_BAND;
1734 oc->accept_re = NULL;
1735 oc->update_resolv_conf = false;
1737 oc->burst_limit = 0;
1740 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1742 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1743 struct ofproto_controller *oc)
1745 oc->target = c->target;
1746 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1747 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1748 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1749 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1750 oc->accept_re = c->discover_accept_regex;
1751 oc->update_resolv_conf = c->discover_update_resolv_conf;
1752 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1753 oc->burst_limit = (c->controller_burst_limit
1754 ? *c->controller_burst_limit : 0);
1757 /* Configures the IP stack for 'br''s local interface properly according to the
1758 * configuration in 'c'. */
1760 bridge_configure_local_iface_netdev(struct bridge *br,
1761 struct ovsrec_controller *c)
1763 struct netdev *netdev;
1764 struct in_addr mask, gateway;
1766 struct iface *local_iface;
1769 /* Controller discovery does its own TCP/IP configuration later. */
1770 if (strcmp(c->target, "discover")) {
1774 /* If there's no local interface or no IP address, give up. */
1775 local_iface = bridge_get_local_iface(br);
1776 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1780 /* Bring up the local interface. */
1781 netdev = local_iface->netdev;
1782 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1784 /* Configure the IP address and netmask. */
1785 if (!c->local_netmask
1786 || !inet_aton(c->local_netmask, &mask)
1788 mask.s_addr = guess_netmask(ip.s_addr);
1790 if (!netdev_set_in4(netdev, ip, mask)) {
1791 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1792 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1795 /* Configure the default gateway. */
1796 if (c->local_gateway
1797 && inet_aton(c->local_gateway, &gateway)
1798 && gateway.s_addr) {
1799 if (!netdev_add_router(netdev, gateway)) {
1800 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1801 br->name, IP_ARGS(&gateway.s_addr));
1807 bridge_reconfigure_remotes(struct bridge *br,
1808 const struct sockaddr_in *managers,
1811 const char *disable_ib_str, *queue_id_str;
1812 bool disable_in_band = false;
1815 struct ovsrec_controller **controllers;
1816 size_t n_controllers;
1819 struct ofproto_controller *ocs;
1823 /* Check if we should disable in-band control on this bridge. */
1824 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1825 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1826 disable_in_band = true;
1829 /* Set OpenFlow queue ID for in-band control. */
1830 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1831 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1832 ofproto_set_in_band_queue(br->ofproto, queue_id);
1834 if (disable_in_band) {
1835 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1837 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1839 had_primary = ofproto_has_primary_controller(br->ofproto);
1841 n_controllers = bridge_get_controllers(br, &controllers);
1843 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1846 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1847 for (i = 0; i < n_controllers; i++) {
1848 struct ovsrec_controller *c = controllers[i];
1850 if (!strncmp(c->target, "punix:", 6)
1851 || !strncmp(c->target, "unix:", 5)) {
1852 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1854 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1855 * domain sockets and overwriting arbitrary local files. */
1856 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1857 "\"%s\" due to possibility for remote exploit",
1858 dpif_name(br->dpif), c->target);
1862 bridge_configure_local_iface_netdev(br, c);
1863 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1864 if (disable_in_band) {
1865 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1870 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1871 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1874 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1875 ofproto_flush_flows(br->ofproto);
1878 /* If there are no controllers and the bridge is in standalone
1879 * mode, set up a flow that matches every packet and directs
1880 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1881 * switch is in secure mode and we won't pass any traffic until
1882 * a controller has been defined and it tells us to do so. */
1884 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1885 union ofp_action action;
1886 struct cls_rule rule;
1888 memset(&action, 0, sizeof action);
1889 action.type = htons(OFPAT_OUTPUT);
1890 action.output.len = htons(sizeof action);
1891 action.output.port = htons(OFPP_NORMAL);
1892 cls_rule_init_catchall(&rule, 0);
1893 ofproto_add_flow(br->ofproto, &rule, &action, 1);
1898 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1903 for (i = 0; i < br->n_ports; i++) {
1904 struct port *port = br->ports[i];
1905 for (j = 0; j < port->n_ifaces; j++) {
1906 struct iface *iface = port->ifaces[j];
1907 shash_add_once(ifaces, iface->name, iface);
1909 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1910 shash_add_once(ifaces, port->name, NULL);
1915 /* For robustness, in case the administrator moves around datapath ports behind
1916 * our back, we re-check all the datapath port numbers here.
1918 * This function will set the 'dp_ifidx' members of interfaces that have
1919 * disappeared to -1, so only call this function from a context where those
1920 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1921 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1922 * datapath, which doesn't support UINT16_MAX+1 ports. */
1924 bridge_fetch_dp_ifaces(struct bridge *br)
1926 struct odp_port *dpif_ports;
1927 size_t n_dpif_ports;
1930 /* Reset all interface numbers. */
1931 for (i = 0; i < br->n_ports; i++) {
1932 struct port *port = br->ports[i];
1933 for (j = 0; j < port->n_ifaces; j++) {
1934 struct iface *iface = port->ifaces[j];
1935 iface->dp_ifidx = -1;
1938 hmap_clear(&br->ifaces);
1940 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1941 for (i = 0; i < n_dpif_ports; i++) {
1942 struct odp_port *p = &dpif_ports[i];
1943 struct iface *iface = iface_lookup(br, p->devname);
1945 if (iface->dp_ifidx >= 0) {
1946 VLOG_WARN("%s reported interface %s twice",
1947 dpif_name(br->dpif), p->devname);
1948 } else if (iface_from_dp_ifidx(br, p->port)) {
1949 VLOG_WARN("%s reported interface %"PRIu16" twice",
1950 dpif_name(br->dpif), p->port);
1952 iface->dp_ifidx = p->port;
1953 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1954 hash_int(iface->dp_ifidx, 0));
1957 iface_set_ofport(iface->cfg,
1958 (iface->dp_ifidx >= 0
1959 ? odp_port_to_ofp_port(iface->dp_ifidx)
1966 /* Bridge packet processing functions. */
1969 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
1971 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
1974 static struct bond_entry *
1975 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
1978 return &port->bond_hash[bond_hash(mac, vlan)];
1982 bond_choose_iface(const struct port *port)
1984 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1985 size_t i, best_down_slave = -1;
1986 long long next_delay_expiration = LLONG_MAX;
1988 for (i = 0; i < port->n_ifaces; i++) {
1989 struct iface *iface = port->ifaces[i];
1991 if (iface->enabled) {
1993 } else if (iface->delay_expires < next_delay_expiration) {
1994 best_down_slave = i;
1995 next_delay_expiration = iface->delay_expires;
1999 if (best_down_slave != -1) {
2000 struct iface *iface = port->ifaces[best_down_slave];
2002 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2003 "since no other interface is up", iface->name,
2004 iface->delay_expires - time_msec());
2005 bond_enable_slave(iface, true);
2008 return best_down_slave;
2012 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2013 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2015 struct iface *iface;
2017 assert(port->n_ifaces);
2018 if (port->n_ifaces == 1) {
2019 iface = port->ifaces[0];
2021 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2022 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2023 || !port->ifaces[e->iface_idx]->enabled) {
2024 /* XXX select interface properly. The current interface selection
2025 * is only good for testing the rebalancing code. */
2026 e->iface_idx = bond_choose_iface(port);
2027 if (e->iface_idx < 0) {
2028 *tags |= port->no_ifaces_tag;
2031 e->iface_tag = tag_create_random();
2032 ((struct port *) port)->bond_compat_is_stale = true;
2034 *tags |= e->iface_tag;
2035 iface = port->ifaces[e->iface_idx];
2037 *dp_ifidx = iface->dp_ifidx;
2038 *tags |= iface->tag; /* Currently only used for bonding. */
2043 bond_link_status_update(struct iface *iface, bool carrier)
2045 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2046 struct port *port = iface->port;
2048 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2049 /* Nothing to do. */
2052 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
2053 iface->name, carrier ? "detected" : "dropped");
2054 if (carrier == iface->enabled) {
2055 iface->delay_expires = LLONG_MAX;
2056 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2057 iface->name, carrier ? "disabled" : "enabled");
2058 } else if (carrier && port->active_iface < 0) {
2059 bond_enable_slave(iface, true);
2060 if (port->updelay) {
2061 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2062 "other interface is up", iface->name, port->updelay);
2065 int delay = carrier ? port->updelay : port->downdelay;
2066 iface->delay_expires = time_msec() + delay;
2069 "interface %s: will be %s if it stays %s for %d ms",
2071 carrier ? "enabled" : "disabled",
2072 carrier ? "up" : "down",
2079 bond_choose_active_iface(struct port *port)
2081 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2083 port->active_iface = bond_choose_iface(port);
2084 port->active_iface_tag = tag_create_random();
2085 if (port->active_iface >= 0) {
2086 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2087 port->name, port->ifaces[port->active_iface]->name);
2089 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2095 bond_enable_slave(struct iface *iface, bool enable)
2097 struct port *port = iface->port;
2098 struct bridge *br = port->bridge;
2100 /* This acts as a recursion check. If the act of disabling a slave
2101 * causes a different slave to be enabled, the flag will allow us to
2102 * skip redundant work when we reenter this function. It must be
2103 * cleared on exit to keep things safe with multiple bonds. */
2104 static bool moving_active_iface = false;
2106 iface->delay_expires = LLONG_MAX;
2107 if (enable == iface->enabled) {
2111 iface->enabled = enable;
2112 if (!iface->enabled) {
2113 VLOG_WARN("interface %s: disabled", iface->name);
2114 ofproto_revalidate(br->ofproto, iface->tag);
2115 if (iface->port_ifidx == port->active_iface) {
2116 ofproto_revalidate(br->ofproto,
2117 port->active_iface_tag);
2119 /* Disabling a slave can lead to another slave being immediately
2120 * enabled if there will be no active slaves but one is waiting
2121 * on an updelay. In this case we do not need to run most of the
2122 * code for the newly enabled slave since there was no period
2123 * without an active slave and it is redundant with the disabling
2125 moving_active_iface = true;
2126 bond_choose_active_iface(port);
2128 bond_send_learning_packets(port);
2130 VLOG_WARN("interface %s: enabled", iface->name);
2131 if (port->active_iface < 0 && !moving_active_iface) {
2132 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2133 bond_choose_active_iface(port);
2134 bond_send_learning_packets(port);
2136 iface->tag = tag_create_random();
2139 moving_active_iface = false;
2140 port->bond_compat_is_stale = true;
2143 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2144 * bond interface. */
2146 bond_update_fake_iface_stats(struct port *port)
2148 struct netdev_stats bond_stats;
2149 struct netdev *bond_dev;
2152 memset(&bond_stats, 0, sizeof bond_stats);
2154 for (i = 0; i < port->n_ifaces; i++) {
2155 struct netdev_stats slave_stats;
2157 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2158 /* XXX: We swap the stats here because they are swapped back when
2159 * reported by the internal device. The reason for this is
2160 * internal devices normally represent packets going into the system
2161 * but when used as fake bond device they represent packets leaving
2162 * the system. We really should do this in the internal device
2163 * itself because changing it here reverses the counts from the
2164 * perspective of the switch. However, the internal device doesn't
2165 * know what type of device it represents so we have to do it here
2167 bond_stats.tx_packets += slave_stats.rx_packets;
2168 bond_stats.tx_bytes += slave_stats.rx_bytes;
2169 bond_stats.rx_packets += slave_stats.tx_packets;
2170 bond_stats.rx_bytes += slave_stats.tx_bytes;
2174 if (!netdev_open_default(port->name, &bond_dev)) {
2175 netdev_set_stats(bond_dev, &bond_stats);
2176 netdev_close(bond_dev);
2181 bond_run(struct bridge *br)
2185 for (i = 0; i < br->n_ports; i++) {
2186 struct port *port = br->ports[i];
2188 if (port->n_ifaces >= 2) {
2191 /* Track carrier going up and down on interfaces. */
2192 while (!netdev_monitor_poll(port->monitor, &devname)) {
2193 struct iface *iface;
2195 iface = port_lookup_iface(port, devname);
2197 bool carrier = netdev_get_carrier(iface->netdev);
2199 bond_link_status_update(iface, carrier);
2200 port_update_bond_compat(port);
2205 for (j = 0; j < port->n_ifaces; j++) {
2206 struct iface *iface = port->ifaces[j];
2207 if (time_msec() >= iface->delay_expires) {
2208 bond_enable_slave(iface, !iface->enabled);
2212 if (port->bond_fake_iface
2213 && time_msec() >= port->bond_next_fake_iface_update) {
2214 bond_update_fake_iface_stats(port);
2215 port->bond_next_fake_iface_update = time_msec() + 1000;
2219 if (port->bond_compat_is_stale) {
2220 port->bond_compat_is_stale = false;
2221 port_update_bond_compat(port);
2227 bond_wait(struct bridge *br)
2231 for (i = 0; i < br->n_ports; i++) {
2232 struct port *port = br->ports[i];
2233 if (port->n_ifaces < 2) {
2236 netdev_monitor_poll_wait(port->monitor);
2237 for (j = 0; j < port->n_ifaces; j++) {
2238 struct iface *iface = port->ifaces[j];
2239 if (iface->delay_expires != LLONG_MAX) {
2240 poll_timer_wait_until(iface->delay_expires);
2243 if (port->bond_fake_iface) {
2244 poll_timer_wait_until(port->bond_next_fake_iface_update);
2250 set_dst(struct dst *dst, const struct flow *flow,
2251 const struct port *in_port, const struct port *out_port,
2254 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2255 : in_port->vlan >= 0 ? in_port->vlan
2256 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2257 : vlan_tci_to_vid(flow->vlan_tci));
2258 return choose_output_iface(out_port, flow->dl_src, dst->vlan,
2259 &dst->dp_ifidx, tags);
2263 swap_dst(struct dst *p, struct dst *q)
2265 struct dst tmp = *p;
2270 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2271 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2272 * that we push to the datapath. We could in fact fully sort the array by
2273 * vlan, but in most cases there are at most two different vlan tags so that's
2274 * possibly overkill.) */
2276 partition_dsts(struct dst_set *set, int vlan)
2278 struct dst *first = set->dsts;
2279 struct dst *last = set->dsts + set->n;
2281 while (first != last) {
2283 * - All dsts < first have vlan == 'vlan'.
2284 * - All dsts >= last have vlan != 'vlan'.
2285 * - first < last. */
2286 while (first->vlan == vlan) {
2287 if (++first == last) {
2292 /* Same invariants, plus one additional:
2293 * - first->vlan != vlan.
2295 while (last[-1].vlan != vlan) {
2296 if (--last == first) {
2301 /* Same invariants, plus one additional:
2302 * - last[-1].vlan == vlan.*/
2303 swap_dst(first++, --last);
2308 mirror_mask_ffs(mirror_mask_t mask)
2310 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2315 dst_set_init(struct dst_set *set)
2317 set->dsts = set->builtin;
2319 set->allocated = ARRAY_SIZE(set->builtin);
2323 dst_set_add(struct dst_set *set, const struct dst *dst)
2325 if (set->n >= set->allocated) {
2326 size_t new_allocated;
2327 struct dst *new_dsts;
2329 new_allocated = set->allocated * 2;
2330 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2331 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2335 set->dsts = new_dsts;
2336 set->allocated = new_allocated;
2338 set->dsts[set->n++] = *dst;
2342 dst_set_free(struct dst_set *set)
2344 if (set->dsts != set->builtin) {
2350 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2353 for (i = 0; i < set->n; i++) {
2354 if (set->dsts[i].vlan == test->vlan
2355 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2363 port_trunks_vlan(const struct port *port, uint16_t vlan)
2365 return (port->vlan < 0
2366 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2370 port_includes_vlan(const struct port *port, uint16_t vlan)
2372 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2376 port_is_floodable(const struct port *port)
2380 for (i = 0; i < port->n_ifaces; i++) {
2381 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2382 port->ifaces[i]->dp_ifidx)) {
2390 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2391 const struct port *in_port, const struct port *out_port,
2392 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2394 mirror_mask_t mirrors = in_port->src_mirrors;
2399 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2400 if (flow_vlan == 0) {
2401 flow_vlan = OFP_VLAN_NONE;
2404 if (out_port == FLOOD_PORT) {
2405 for (i = 0; i < br->n_ports; i++) {
2406 struct port *port = br->ports[i];
2408 && port_is_floodable(port)
2409 && port_includes_vlan(port, vlan)
2410 && !port->is_mirror_output_port
2411 && set_dst(&dst, flow, in_port, port, tags)) {
2412 mirrors |= port->dst_mirrors;
2413 dst_set_add(set, &dst);
2416 *nf_output_iface = NF_OUT_FLOOD;
2417 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2418 dst_set_add(set, &dst);
2419 *nf_output_iface = dst.dp_ifidx;
2420 mirrors |= out_port->dst_mirrors;
2424 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2425 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2427 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2428 && !dst_is_duplicate(set, &dst)) {
2429 dst_set_add(set, &dst);
2432 for (i = 0; i < br->n_ports; i++) {
2433 struct port *port = br->ports[i];
2434 if (port_includes_vlan(port, m->out_vlan)
2435 && set_dst(&dst, flow, in_port, port, tags))
2437 if (port->vlan < 0) {
2438 dst.vlan = m->out_vlan;
2440 if (dst_is_duplicate(set, &dst)) {
2444 /* Use the vlan tag on the original flow instead of
2445 * the one passed in the vlan parameter. This ensures
2446 * that we compare the vlan from before any implicit
2447 * tagging tags place. This is necessary because
2448 * dst->vlan is the final vlan, after removing implicit
2450 if (port == in_port && dst.vlan == flow_vlan) {
2451 /* Don't send out input port on same VLAN. */
2454 dst_set_add(set, &dst);
2459 mirrors &= mirrors - 1;
2462 partition_dsts(set, flow_vlan);
2465 static void OVS_UNUSED
2466 print_dsts(const struct dst_set *set)
2470 for (i = 0; i < set->n; i++) {
2471 const struct dst *dst = &set->dsts[i];
2473 printf(">p%"PRIu16, dst->dp_ifidx);
2474 if (dst->vlan != OFP_VLAN_NONE) {
2475 printf("v%"PRIu16, dst->vlan);
2481 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2482 const struct port *in_port, const struct port *out_port,
2483 tag_type *tags, struct ofpbuf *actions,
2484 uint16_t *nf_output_iface)
2491 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2494 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2495 if (cur_vlan == 0) {
2496 cur_vlan = OFP_VLAN_NONE;
2498 for (i = 0; i < set.n; i++) {
2499 const struct dst *dst = &set.dsts[i];
2500 if (dst->vlan != cur_vlan) {
2501 if (dst->vlan == OFP_VLAN_NONE) {
2502 nl_msg_put_flag(actions, ODPAT_STRIP_VLAN);
2505 tci = htons(dst->vlan & VLAN_VID_MASK);
2506 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2507 nl_msg_put_be16(actions, ODPAT_SET_DL_TCI, tci);
2509 cur_vlan = dst->vlan;
2511 nl_msg_put_u32(actions, ODPAT_OUTPUT, dst->dp_ifidx);
2516 /* Returns the effective vlan of a packet, taking into account both the
2517 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2518 * the packet is untagged and -1 indicates it has an invalid header and
2519 * should be dropped. */
2520 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2521 struct port *in_port, bool have_packet)
2523 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2524 if (in_port->vlan >= 0) {
2526 /* XXX support double tagging? */
2528 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2529 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2530 "packet received on port %s configured with "
2531 "implicit VLAN %"PRIu16,
2532 br->name, vlan, in_port->name, in_port->vlan);
2536 vlan = in_port->vlan;
2538 if (!port_includes_vlan(in_port, vlan)) {
2540 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2541 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2542 "packet received on port %s not configured for "
2544 br->name, vlan, in_port->name, vlan);
2553 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2554 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2555 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2557 is_gratuitous_arp(const struct flow *flow)
2559 return (flow->dl_type == htons(ETH_TYPE_ARP)
2560 && eth_addr_is_broadcast(flow->dl_dst)
2561 && (flow->nw_proto == ARP_OP_REPLY
2562 || (flow->nw_proto == ARP_OP_REQUEST
2563 && flow->nw_src == flow->nw_dst)));
2567 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2568 struct port *in_port)
2570 enum grat_arp_lock_type lock_type;
2573 /* We don't want to learn from gratuitous ARP packets that are reflected
2574 * back over bond slaves so we lock the learning table. */
2575 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2576 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2577 GRAT_ARP_LOCK_CHECK;
2579 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2582 /* The log messages here could actually be useful in debugging,
2583 * so keep the rate limit relatively high. */
2584 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2586 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2587 "on port %s in VLAN %d",
2588 br->name, ETH_ADDR_ARGS(flow->dl_src),
2589 in_port->name, vlan);
2590 ofproto_revalidate(br->ofproto, rev_tag);
2594 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2595 * dropped. Returns true if they may be forwarded, false if they should be
2598 * If 'have_packet' is true, it indicates that the caller is processing a
2599 * received packet. If 'have_packet' is false, then the caller is just
2600 * revalidating an existing flow because configuration has changed. Either
2601 * way, 'have_packet' only affects logging (there is no point in logging errors
2602 * during revalidation).
2604 * Sets '*in_portp' to the input port. This will be a null pointer if
2605 * flow->in_port does not designate a known input port (in which case
2606 * is_admissible() returns false).
2608 * When returning true, sets '*vlanp' to the effective VLAN of the input
2609 * packet, as returned by flow_get_vlan().
2611 * May also add tags to '*tags', although the current implementation only does
2612 * so in one special case.
2615 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2616 tag_type *tags, int *vlanp, struct port **in_portp)
2618 struct iface *in_iface;
2619 struct port *in_port;
2622 /* Find the interface and port structure for the received packet. */
2623 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2625 /* No interface? Something fishy... */
2627 /* Odd. A few possible reasons here:
2629 * - We deleted an interface but there are still a few packets
2630 * queued up from it.
2632 * - Someone externally added an interface (e.g. with "ovs-dpctl
2633 * add-if") that we don't know about.
2635 * - Packet arrived on the local port but the local port is not
2636 * one of our bridge ports.
2638 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2640 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2641 "interface %"PRIu16, br->name, flow->in_port);
2647 *in_portp = in_port = in_iface->port;
2648 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2653 /* Drop frames for reserved multicast addresses. */
2654 if (eth_addr_is_reserved(flow->dl_dst)) {
2658 /* Drop frames on ports reserved for mirroring. */
2659 if (in_port->is_mirror_output_port) {
2661 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2662 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2663 "%s, which is reserved exclusively for mirroring",
2664 br->name, in_port->name);
2669 /* Packets received on bonds need special attention to avoid duplicates. */
2670 if (in_port->n_ifaces > 1) {
2672 bool is_grat_arp_locked;
2674 if (eth_addr_is_multicast(flow->dl_dst)) {
2675 *tags |= in_port->active_iface_tag;
2676 if (in_port->active_iface != in_iface->port_ifidx) {
2677 /* Drop all multicast packets on inactive slaves. */
2682 /* Drop all packets for which we have learned a different input
2683 * port, because we probably sent the packet on one slave and got
2684 * it back on the other. Gratuitous ARP packets are an exception
2685 * to this rule: the host has moved to another switch. The exception
2686 * to the exception is if we locked the learning table to avoid
2687 * reflections on bond slaves. If this is the case, just drop the
2689 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2690 &is_grat_arp_locked);
2691 if (src_idx != -1 && src_idx != in_port->port_idx &&
2692 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2700 /* If the composed actions may be applied to any packet in the given 'flow',
2701 * returns true. Otherwise, the actions should only be applied to 'packet', or
2702 * not at all, if 'packet' was NULL. */
2704 process_flow(struct bridge *br, const struct flow *flow,
2705 const struct ofpbuf *packet, struct ofpbuf *actions,
2706 tag_type *tags, uint16_t *nf_output_iface)
2708 struct port *in_port;
2709 struct port *out_port;
2713 /* Check whether we should drop packets in this flow. */
2714 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2719 /* Learn source MAC (but don't try to learn from revalidation). */
2721 update_learning_table(br, flow, vlan, in_port);
2724 /* Determine output port. */
2725 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2727 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2728 out_port = br->ports[out_port_idx];
2729 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2730 /* If we are revalidating but don't have a learning entry then
2731 * eject the flow. Installing a flow that floods packets opens
2732 * up a window of time where we could learn from a packet reflected
2733 * on a bond and blackhole packets before the learning table is
2734 * updated to reflect the correct port. */
2737 out_port = FLOOD_PORT;
2740 /* Don't send packets out their input ports. */
2741 if (in_port == out_port) {
2747 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2755 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2756 struct ofpbuf *actions, tag_type *tags,
2757 uint16_t *nf_output_iface, void *br_)
2759 struct iface *iface;
2760 struct bridge *br = br_;
2762 COVERAGE_INC(bridge_process_flow);
2764 iface = iface_from_dp_ifidx(br, flow->in_port);
2766 if (cfm_should_process_flow(flow)) {
2767 if (packet && iface->cfm) {
2768 cfm_process_heartbeat(iface->cfm, packet);
2773 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2777 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2778 const struct nlattr *actions,
2780 unsigned long long int n_bytes, void *br_)
2782 struct bridge *br = br_;
2783 const struct nlattr *a;
2784 struct port *in_port;
2789 /* Feed information from the active flows back into the learning table to
2790 * ensure that table is always in sync with what is actually flowing
2791 * through the datapath.
2793 * We test that 'tags' is nonzero to ensure that only flows that include an
2794 * OFPP_NORMAL action are used for learning. This works because
2795 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2796 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2797 update_learning_table(br, flow, vlan, in_port);
2800 /* Account for bond slave utilization. */
2801 if (!br->has_bonded_ports) {
2804 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2805 if (nl_attr_type(a) == ODPAT_OUTPUT) {
2806 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2807 if (out_port && out_port->n_ifaces >= 2) {
2808 uint16_t vlan = (flow->vlan_tci
2809 ? vlan_tci_to_vid(flow->vlan_tci)
2811 struct bond_entry *e = lookup_bond_entry(out_port,
2812 flow->dl_src, vlan);
2813 e->tx_bytes += n_bytes;
2820 bridge_account_checkpoint_ofhook_cb(void *br_)
2822 struct bridge *br = br_;
2826 if (!br->has_bonded_ports) {
2831 for (i = 0; i < br->n_ports; i++) {
2832 struct port *port = br->ports[i];
2833 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2834 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2835 bond_rebalance_port(port);
2840 static struct ofhooks bridge_ofhooks = {
2841 bridge_normal_ofhook_cb,
2842 bridge_account_flow_ofhook_cb,
2843 bridge_account_checkpoint_ofhook_cb,
2846 /* Bonding functions. */
2848 /* Statistics for a single interface on a bonded port, used for load-based
2849 * bond rebalancing. */
2850 struct slave_balance {
2851 struct iface *iface; /* The interface. */
2852 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2854 /* All the "bond_entry"s that are assigned to this interface, in order of
2855 * increasing tx_bytes. */
2856 struct bond_entry **hashes;
2860 /* Sorts pointers to pointers to bond_entries in ascending order by the
2861 * interface to which they are assigned, and within a single interface in
2862 * ascending order of bytes transmitted. */
2864 compare_bond_entries(const void *a_, const void *b_)
2866 const struct bond_entry *const *ap = a_;
2867 const struct bond_entry *const *bp = b_;
2868 const struct bond_entry *a = *ap;
2869 const struct bond_entry *b = *bp;
2870 if (a->iface_idx != b->iface_idx) {
2871 return a->iface_idx > b->iface_idx ? 1 : -1;
2872 } else if (a->tx_bytes != b->tx_bytes) {
2873 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2879 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2880 * *descending* order by number of bytes transmitted. */
2882 compare_slave_balance(const void *a_, const void *b_)
2884 const struct slave_balance *a = a_;
2885 const struct slave_balance *b = b_;
2886 if (a->iface->enabled != b->iface->enabled) {
2887 return a->iface->enabled ? -1 : 1;
2888 } else if (a->tx_bytes != b->tx_bytes) {
2889 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2896 swap_bals(struct slave_balance *a, struct slave_balance *b)
2898 struct slave_balance tmp = *a;
2903 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2904 * given that 'p' (and only 'p') might be in the wrong location.
2906 * This function invalidates 'p', since it might now be in a different memory
2909 resort_bals(struct slave_balance *p,
2910 struct slave_balance bals[], size_t n_bals)
2913 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2914 swap_bals(p, p - 1);
2916 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2917 swap_bals(p, p + 1);
2923 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2925 if (VLOG_IS_DBG_ENABLED()) {
2926 struct ds ds = DS_EMPTY_INITIALIZER;
2927 const struct slave_balance *b;
2929 for (b = bals; b < bals + n_bals; b++) {
2933 ds_put_char(&ds, ',');
2935 ds_put_format(&ds, " %s %"PRIu64"kB",
2936 b->iface->name, b->tx_bytes / 1024);
2938 if (!b->iface->enabled) {
2939 ds_put_cstr(&ds, " (disabled)");
2941 if (b->n_hashes > 0) {
2942 ds_put_cstr(&ds, " (");
2943 for (i = 0; i < b->n_hashes; i++) {
2944 const struct bond_entry *e = b->hashes[i];
2946 ds_put_cstr(&ds, " + ");
2948 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2949 e - port->bond_hash, e->tx_bytes / 1024);
2951 ds_put_cstr(&ds, ")");
2954 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2959 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2961 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2964 struct bond_entry *hash = from->hashes[hash_idx];
2965 struct port *port = from->iface->port;
2966 uint64_t delta = hash->tx_bytes;
2968 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2969 "from %s to %s (now carrying %"PRIu64"kB and "
2970 "%"PRIu64"kB load, respectively)",
2971 port->name, delta / 1024, hash - port->bond_hash,
2972 from->iface->name, to->iface->name,
2973 (from->tx_bytes - delta) / 1024,
2974 (to->tx_bytes + delta) / 1024);
2976 /* Delete element from from->hashes.
2978 * We don't bother to add the element to to->hashes because not only would
2979 * it require more work, the only purpose it would be to allow that hash to
2980 * be migrated to another slave in this rebalancing run, and there is no
2981 * point in doing that. */
2982 if (hash_idx == 0) {
2985 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2986 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2990 /* Shift load away from 'from' to 'to'. */
2991 from->tx_bytes -= delta;
2992 to->tx_bytes += delta;
2994 /* Arrange for flows to be revalidated. */
2995 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2996 hash->iface_idx = to->iface->port_ifidx;
2997 hash->iface_tag = tag_create_random();
3001 bond_rebalance_port(struct port *port)
3003 struct slave_balance *bals;
3005 struct bond_entry *hashes[BOND_MASK + 1];
3006 struct slave_balance *b, *from, *to;
3007 struct bond_entry *e;
3010 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3011 * descending order of tx_bytes, so that bals[0] represents the most
3012 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3015 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3016 * array for each slave_balance structure, we sort our local array of
3017 * hashes in order by slave, so that all of the hashes for a given slave
3018 * become contiguous in memory, and then we point each 'hashes' members of
3019 * a slave_balance structure to the start of a contiguous group. */
3020 n_bals = port->n_ifaces;
3021 bals = xmalloc(n_bals * sizeof *bals);
3022 for (b = bals; b < &bals[n_bals]; b++) {
3023 b->iface = port->ifaces[b - bals];
3028 for (i = 0; i <= BOND_MASK; i++) {
3029 hashes[i] = &port->bond_hash[i];
3031 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3032 for (i = 0; i <= BOND_MASK; i++) {
3034 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3035 b = &bals[e->iface_idx];
3036 b->tx_bytes += e->tx_bytes;
3038 b->hashes = &hashes[i];
3043 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3044 log_bals(bals, n_bals, port);
3046 /* Discard slaves that aren't enabled (which were sorted to the back of the
3047 * array earlier). */
3048 while (!bals[n_bals - 1].iface->enabled) {
3055 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3056 to = &bals[n_bals - 1];
3057 for (from = bals; from < to; ) {
3058 uint64_t overload = from->tx_bytes - to->tx_bytes;
3059 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3060 /* The extra load on 'from' (and all less-loaded slaves), compared
3061 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3062 * it is less than ~1Mbps. No point in rebalancing. */
3064 } else if (from->n_hashes == 1) {
3065 /* 'from' only carries a single MAC hash, so we can't shift any
3066 * load away from it, even though we want to. */
3069 /* 'from' is carrying significantly more load than 'to', and that
3070 * load is split across at least two different hashes. Pick a hash
3071 * to migrate to 'to' (the least-loaded slave), given that doing so
3072 * must decrease the ratio of the load on the two slaves by at
3075 * The sort order we use means that we prefer to shift away the
3076 * smallest hashes instead of the biggest ones. There is little
3077 * reason behind this decision; we could use the opposite sort
3078 * order to shift away big hashes ahead of small ones. */
3081 for (i = 0; i < from->n_hashes; i++) {
3082 double old_ratio, new_ratio;
3083 uint64_t delta = from->hashes[i]->tx_bytes;
3085 if (delta == 0 || from->tx_bytes - delta == 0) {
3086 /* Pointless move. */
3090 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3092 if (to->tx_bytes == 0) {
3093 /* Nothing on the new slave, move it. */
3097 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3098 new_ratio = (double)(from->tx_bytes - delta) /
3099 (to->tx_bytes + delta);
3101 if (new_ratio == 0) {
3102 /* Should already be covered but check to prevent division
3107 if (new_ratio < 1) {
3108 new_ratio = 1 / new_ratio;
3111 if (old_ratio - new_ratio > 0.1) {
3112 /* Would decrease the ratio, move it. */
3116 if (i < from->n_hashes) {
3117 bond_shift_load(from, to, i);
3118 port->bond_compat_is_stale = true;
3120 /* If the result of the migration changed the relative order of
3121 * 'from' and 'to' swap them back to maintain invariants. */
3122 if (order_swapped) {
3123 swap_bals(from, to);
3126 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3127 * point to different slave_balance structures. It is only
3128 * valid to do these two operations in a row at all because we
3129 * know that 'from' will not move past 'to' and vice versa. */
3130 resort_bals(from, bals, n_bals);
3131 resort_bals(to, bals, n_bals);
3138 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3139 * historical data to decay to <1% in 7 rebalancing runs. */
3140 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3149 bond_send_learning_packets(struct port *port)
3151 struct bridge *br = port->bridge;
3152 struct mac_entry *e;
3153 struct ofpbuf packet;
3154 int error, n_packets, n_errors;
3156 if (!port->n_ifaces || port->active_iface < 0) {
3160 ofpbuf_init(&packet, 128);
3161 error = n_packets = n_errors = 0;
3162 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3163 union ofp_action actions[2], *a;
3169 if (e->port == port->port_idx
3170 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3174 /* Compose actions. */
3175 memset(actions, 0, sizeof actions);
3178 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3179 a->vlan_vid.len = htons(sizeof *a);
3180 a->vlan_vid.vlan_vid = htons(e->vlan);
3183 a->output.type = htons(OFPAT_OUTPUT);
3184 a->output.len = htons(sizeof *a);
3185 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3190 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3192 flow_extract(&packet, 0, ODPP_NONE, &flow);
3193 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3200 ofpbuf_uninit(&packet);
3203 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3204 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3205 "packets, last error was: %s",
3206 port->name, n_errors, n_packets, strerror(error));
3208 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3209 port->name, n_packets);
3213 /* Bonding unixctl user interface functions. */
3216 bond_unixctl_list(struct unixctl_conn *conn,
3217 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3219 struct ds ds = DS_EMPTY_INITIALIZER;
3220 const struct bridge *br;
3222 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
3224 LIST_FOR_EACH (br, node, &all_bridges) {
3227 for (i = 0; i < br->n_ports; i++) {
3228 const struct port *port = br->ports[i];
3229 if (port->n_ifaces > 1) {
3232 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3233 for (j = 0; j < port->n_ifaces; j++) {
3234 const struct iface *iface = port->ifaces[j];
3236 ds_put_cstr(&ds, ", ");
3238 ds_put_cstr(&ds, iface->name);
3240 ds_put_char(&ds, '\n');
3244 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3248 static struct port *
3249 bond_find(const char *name)
3251 const struct bridge *br;
3253 LIST_FOR_EACH (br, node, &all_bridges) {
3256 for (i = 0; i < br->n_ports; i++) {
3257 struct port *port = br->ports[i];
3258 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3267 bond_unixctl_show(struct unixctl_conn *conn,
3268 const char *args, void *aux OVS_UNUSED)
3270 struct ds ds = DS_EMPTY_INITIALIZER;
3271 const struct port *port;
3274 port = bond_find(args);
3276 unixctl_command_reply(conn, 501, "no such bond");
3280 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3281 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3282 ds_put_format(&ds, "next rebalance: %lld ms\n",
3283 port->bond_next_rebalance - time_msec());
3284 for (j = 0; j < port->n_ifaces; j++) {
3285 const struct iface *iface = port->ifaces[j];
3286 struct bond_entry *be;
3289 ds_put_format(&ds, "slave %s: %s\n",
3290 iface->name, iface->enabled ? "enabled" : "disabled");
3291 if (j == port->active_iface) {
3292 ds_put_cstr(&ds, "\tactive slave\n");
3294 if (iface->delay_expires != LLONG_MAX) {
3295 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3296 iface->enabled ? "downdelay" : "updelay",
3297 iface->delay_expires - time_msec());
3301 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3302 int hash = be - port->bond_hash;
3303 struct mac_entry *me;
3305 if (be->iface_idx != j) {
3309 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3310 hash, be->tx_bytes / 1024);
3313 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3316 if (bond_hash(me->mac, me->vlan) == hash
3317 && me->port != port->port_idx
3318 && choose_output_iface(port, me->mac, me->vlan,
3320 && dp_ifidx == iface->dp_ifidx)
3322 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3323 ETH_ADDR_ARGS(me->mac));
3328 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3333 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3334 void *aux OVS_UNUSED)
3336 char *args = (char *) args_;
3337 char *save_ptr = NULL;
3338 char *bond_s, *hash_s, *slave_s;
3340 struct iface *iface;
3341 struct bond_entry *entry;
3344 bond_s = strtok_r(args, " ", &save_ptr);
3345 hash_s = strtok_r(NULL, " ", &save_ptr);
3346 slave_s = strtok_r(NULL, " ", &save_ptr);
3348 unixctl_command_reply(conn, 501,
3349 "usage: bond/migrate BOND HASH SLAVE");
3353 port = bond_find(bond_s);
3355 unixctl_command_reply(conn, 501, "no such bond");
3359 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3360 hash = atoi(hash_s) & BOND_MASK;
3362 unixctl_command_reply(conn, 501, "bad hash");
3366 iface = port_lookup_iface(port, slave_s);
3368 unixctl_command_reply(conn, 501, "no such slave");
3372 if (!iface->enabled) {
3373 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3377 entry = &port->bond_hash[hash];
3378 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3379 entry->iface_idx = iface->port_ifidx;
3380 entry->iface_tag = tag_create_random();
3381 port->bond_compat_is_stale = true;
3382 unixctl_command_reply(conn, 200, "migrated");
3386 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3387 void *aux OVS_UNUSED)
3389 char *args = (char *) args_;
3390 char *save_ptr = NULL;
3391 char *bond_s, *slave_s;
3393 struct iface *iface;
3395 bond_s = strtok_r(args, " ", &save_ptr);
3396 slave_s = strtok_r(NULL, " ", &save_ptr);
3398 unixctl_command_reply(conn, 501,
3399 "usage: bond/set-active-slave BOND SLAVE");
3403 port = bond_find(bond_s);
3405 unixctl_command_reply(conn, 501, "no such bond");
3409 iface = port_lookup_iface(port, slave_s);
3411 unixctl_command_reply(conn, 501, "no such slave");
3415 if (!iface->enabled) {
3416 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3420 if (port->active_iface != iface->port_ifidx) {
3421 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3422 port->active_iface = iface->port_ifidx;
3423 port->active_iface_tag = tag_create_random();
3424 VLOG_INFO("port %s: active interface is now %s",
3425 port->name, iface->name);
3426 bond_send_learning_packets(port);
3427 unixctl_command_reply(conn, 200, "done");
3429 unixctl_command_reply(conn, 200, "no change");
3434 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3436 char *args = (char *) args_;
3437 char *save_ptr = NULL;
3438 char *bond_s, *slave_s;
3440 struct iface *iface;
3442 bond_s = strtok_r(args, " ", &save_ptr);
3443 slave_s = strtok_r(NULL, " ", &save_ptr);
3445 unixctl_command_reply(conn, 501,
3446 "usage: bond/enable/disable-slave BOND SLAVE");
3450 port = bond_find(bond_s);
3452 unixctl_command_reply(conn, 501, "no such bond");
3456 iface = port_lookup_iface(port, slave_s);
3458 unixctl_command_reply(conn, 501, "no such slave");
3462 bond_enable_slave(iface, enable);
3463 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3467 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3468 void *aux OVS_UNUSED)
3470 enable_slave(conn, args, true);
3474 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3475 void *aux OVS_UNUSED)
3477 enable_slave(conn, args, false);
3481 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3482 void *aux OVS_UNUSED)
3484 char *args = (char *) args_;
3485 uint8_t mac[ETH_ADDR_LEN];
3489 char *mac_s, *vlan_s;
3490 char *save_ptr = NULL;
3492 mac_s = strtok_r(args, " ", &save_ptr);
3493 vlan_s = strtok_r(NULL, " ", &save_ptr);
3496 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3497 unixctl_command_reply(conn, 501, "invalid vlan");
3501 vlan = OFP_VLAN_NONE;
3504 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3505 == ETH_ADDR_SCAN_COUNT) {
3506 hash = bond_hash(mac, vlan);
3508 hash_cstr = xasprintf("%u", hash);
3509 unixctl_command_reply(conn, 200, hash_cstr);
3512 unixctl_command_reply(conn, 501, "invalid mac");
3519 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3520 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3521 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3522 unixctl_command_register("bond/set-active-slave",
3523 bond_unixctl_set_active_slave, NULL);
3524 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3526 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3528 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3531 /* Port functions. */
3533 static struct port *
3534 port_create(struct bridge *br, const char *name)
3538 port = xzalloc(sizeof *port);
3540 port->port_idx = br->n_ports;
3542 port->trunks = NULL;
3543 port->name = xstrdup(name);
3544 port->active_iface = -1;
3546 if (br->n_ports >= br->allocated_ports) {
3547 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3550 br->ports[br->n_ports++] = port;
3551 shash_add_assert(&br->port_by_name, port->name, port);
3553 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3560 get_port_other_config(const struct ovsrec_port *port, const char *key,
3561 const char *default_value)
3565 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3567 return value ? value : default_value;
3571 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3573 struct shash new_ifaces;
3576 /* Collect list of new interfaces. */
3577 shash_init(&new_ifaces);
3578 for (i = 0; i < cfg->n_interfaces; i++) {
3579 const char *name = cfg->interfaces[i]->name;
3580 shash_add_once(&new_ifaces, name, NULL);
3583 /* Get rid of deleted interfaces. */
3584 for (i = 0; i < port->n_ifaces; ) {
3585 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3586 iface_destroy(port->ifaces[i]);
3592 shash_destroy(&new_ifaces);
3596 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3598 struct shash new_ifaces;
3599 long long int next_rebalance;
3600 unsigned long *trunks;
3606 /* Update settings. */
3607 port->updelay = cfg->bond_updelay;
3608 if (port->updelay < 0) {
3611 port->downdelay = cfg->bond_downdelay;
3612 if (port->downdelay < 0) {
3613 port->downdelay = 0;
3615 port->bond_rebalance_interval = atoi(
3616 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3617 if (port->bond_rebalance_interval < 1000) {
3618 port->bond_rebalance_interval = 1000;
3620 next_rebalance = time_msec() + port->bond_rebalance_interval;
3621 if (port->bond_next_rebalance > next_rebalance) {
3622 port->bond_next_rebalance = next_rebalance;
3625 /* Add new interfaces and update 'cfg' member of existing ones. */
3626 shash_init(&new_ifaces);
3627 for (i = 0; i < cfg->n_interfaces; i++) {
3628 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3629 struct iface *iface;
3631 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3632 VLOG_WARN("port %s: %s specified twice as port interface",
3633 port->name, if_cfg->name);
3634 iface_set_ofport(if_cfg, -1);
3638 iface = iface_lookup(port->bridge, if_cfg->name);
3640 if (iface->port != port) {
3641 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3643 port->bridge->name, if_cfg->name, iface->port->name);
3646 iface->cfg = if_cfg;
3648 iface = iface_create(port, if_cfg);
3651 /* Determine interface type. The local port always has type
3652 * "internal". Other ports take their type from the database and
3653 * default to "system" if none is specified. */
3654 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3655 : if_cfg->type[0] ? if_cfg->type
3658 shash_destroy(&new_ifaces);
3663 if (port->n_ifaces < 2) {
3665 if (vlan >= 0 && vlan <= 4095) {
3666 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3671 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3672 * they even work as-is. But they have not been tested. */
3673 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3677 if (port->vlan != vlan) {
3679 bridge_flush(port->bridge);
3682 /* Get trunked VLANs. */
3684 if (vlan < 0 && cfg->n_trunks) {
3687 trunks = bitmap_allocate(4096);
3689 for (i = 0; i < cfg->n_trunks; i++) {
3690 int trunk = cfg->trunks[i];
3692 bitmap_set1(trunks, trunk);
3698 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3699 port->name, cfg->n_trunks);
3701 if (n_errors == cfg->n_trunks) {
3702 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3704 bitmap_free(trunks);
3707 } else if (vlan >= 0 && cfg->n_trunks) {
3708 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3712 ? port->trunks != NULL
3713 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3714 bridge_flush(port->bridge);
3716 bitmap_free(port->trunks);
3717 port->trunks = trunks;
3721 port_destroy(struct port *port)
3724 struct bridge *br = port->bridge;
3728 proc_net_compat_update_vlan(port->name, NULL, 0);
3729 proc_net_compat_update_bond(port->name, NULL);
3731 for (i = 0; i < MAX_MIRRORS; i++) {
3732 struct mirror *m = br->mirrors[i];
3733 if (m && m->out_port == port) {
3738 while (port->n_ifaces > 0) {
3739 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3742 shash_find_and_delete_assert(&br->port_by_name, port->name);
3744 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3745 del->port_idx = port->port_idx;
3747 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3749 netdev_monitor_destroy(port->monitor);
3751 bitmap_free(port->trunks);
3758 static struct port *
3759 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3761 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3762 return iface ? iface->port : NULL;
3765 static struct port *
3766 port_lookup(const struct bridge *br, const char *name)
3768 return shash_find_data(&br->port_by_name, name);
3771 static struct iface *
3772 port_lookup_iface(const struct port *port, const char *name)
3774 struct iface *iface = iface_lookup(port->bridge, name);
3775 return iface && iface->port == port ? iface : NULL;
3779 port_update_bonding(struct port *port)
3781 if (port->monitor) {
3782 netdev_monitor_destroy(port->monitor);
3783 port->monitor = NULL;
3785 if (port->n_ifaces < 2) {
3786 /* Not a bonded port. */
3787 if (port->bond_hash) {
3788 free(port->bond_hash);
3789 port->bond_hash = NULL;
3790 port->bond_compat_is_stale = true;
3791 port->bond_fake_iface = false;
3796 if (!port->bond_hash) {
3797 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3798 for (i = 0; i <= BOND_MASK; i++) {
3799 struct bond_entry *e = &port->bond_hash[i];
3803 port->no_ifaces_tag = tag_create_random();
3804 bond_choose_active_iface(port);
3805 port->bond_next_rebalance
3806 = time_msec() + port->bond_rebalance_interval;
3808 if (port->cfg->bond_fake_iface) {
3809 port->bond_next_fake_iface_update = time_msec();
3812 port->bond_compat_is_stale = true;
3813 port->bond_fake_iface = port->cfg->bond_fake_iface;
3815 port->monitor = netdev_monitor_create();
3816 for (i = 0; i < port->n_ifaces; i++) {
3817 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
3823 port_update_bond_compat(struct port *port)
3825 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3826 struct compat_bond bond;
3829 if (port->n_ifaces < 2) {
3830 proc_net_compat_update_bond(port->name, NULL);
3835 bond.updelay = port->updelay;
3836 bond.downdelay = port->downdelay;
3839 bond.hashes = compat_hashes;
3840 if (port->bond_hash) {
3841 const struct bond_entry *e;
3842 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3843 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3844 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3845 cbh->hash = e - port->bond_hash;
3846 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3851 bond.n_slaves = port->n_ifaces;
3852 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3853 for (i = 0; i < port->n_ifaces; i++) {
3854 struct iface *iface = port->ifaces[i];
3855 struct compat_bond_slave *slave = &bond.slaves[i];
3856 slave->name = iface->name;
3858 /* We need to make the same determination as the Linux bonding
3859 * code to determine whether a slave should be consider "up".
3860 * The Linux function bond_miimon_inspect() supports four
3861 * BOND_LINK_* states:
3863 * - BOND_LINK_UP: carrier detected, updelay has passed.
3864 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3865 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3866 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3868 * The function bond_info_show_slave() only considers BOND_LINK_UP
3869 * to be "up" and anything else to be "down".
3871 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3875 netdev_get_etheraddr(iface->netdev, slave->mac);
3878 if (port->bond_fake_iface) {
3879 struct netdev *bond_netdev;
3881 if (!netdev_open_default(port->name, &bond_netdev)) {
3883 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3885 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3887 netdev_close(bond_netdev);
3891 proc_net_compat_update_bond(port->name, &bond);
3896 port_update_vlan_compat(struct port *port)
3898 struct bridge *br = port->bridge;
3899 char *vlandev_name = NULL;
3901 if (port->vlan > 0) {
3902 /* Figure out the name that the VLAN device should actually have, if it
3903 * existed. This takes some work because the VLAN device would not
3904 * have port->name in its name; rather, it would have the trunk port's
3905 * name, and 'port' would be attached to a bridge that also had the
3906 * VLAN device one of its ports. So we need to find a trunk port that
3907 * includes port->vlan.
3909 * There might be more than one candidate. This doesn't happen on
3910 * XenServer, so if it happens we just pick the first choice in
3911 * alphabetical order instead of creating multiple VLAN devices. */
3913 for (i = 0; i < br->n_ports; i++) {
3914 struct port *p = br->ports[i];
3915 if (port_trunks_vlan(p, port->vlan)
3917 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3919 uint8_t ea[ETH_ADDR_LEN];
3920 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3921 if (!eth_addr_is_multicast(ea) &&
3922 !eth_addr_is_reserved(ea) &&
3923 !eth_addr_is_zero(ea)) {
3924 vlandev_name = p->name;
3929 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3932 /* Interface functions. */
3935 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
3938 union ofp_action action;
3940 memset(&action, 0, sizeof action);
3941 action.output.type = htons(OFPAT_OUTPUT);
3942 action.output.len = htons(sizeof action);
3943 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
3945 flow_extract(packet, 0, ODPP_NONE, &flow);
3947 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
3949 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3950 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
3954 static struct iface *
3955 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3957 struct bridge *br = port->bridge;
3958 struct iface *iface;
3959 char *name = if_cfg->name;
3961 iface = xzalloc(sizeof *iface);
3963 iface->port_ifidx = port->n_ifaces;
3964 iface->name = xstrdup(name);
3965 iface->dp_ifidx = -1;
3966 iface->tag = tag_create_random();
3967 iface->delay_expires = LLONG_MAX;
3968 iface->netdev = NULL;
3969 iface->cfg = if_cfg;
3971 shash_add_assert(&br->iface_by_name, iface->name, iface);
3973 if (port->n_ifaces >= port->allocated_ifaces) {
3974 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3975 sizeof *port->ifaces);
3977 port->ifaces[port->n_ifaces++] = iface;
3978 if (port->n_ifaces > 1) {
3979 br->has_bonded_ports = true;
3982 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3990 iface_destroy(struct iface *iface)
3993 struct port *port = iface->port;
3994 struct bridge *br = port->bridge;
3995 bool del_active = port->active_iface == iface->port_ifidx;
3998 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4000 if (iface->dp_ifidx >= 0) {
4001 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4004 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4005 del->port_ifidx = iface->port_ifidx;
4007 netdev_close(iface->netdev);
4010 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4011 bond_choose_active_iface(port);
4012 bond_send_learning_packets(port);
4015 cfm_destroy(iface->cfm);
4020 bridge_flush(port->bridge);
4024 static struct iface *
4025 iface_lookup(const struct bridge *br, const char *name)
4027 return shash_find_data(&br->iface_by_name, name);
4030 static struct iface *
4031 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4033 struct iface *iface;
4035 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4036 hash_int(dp_ifidx, 0), &br->ifaces) {
4037 if (iface->dp_ifidx == dp_ifidx) {
4044 /* Set Ethernet address of 'iface', if one is specified in the configuration
4047 iface_set_mac(struct iface *iface)
4049 uint8_t ea[ETH_ADDR_LEN];
4051 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4052 if (eth_addr_is_multicast(ea)) {
4053 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4055 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4056 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4057 iface->name, iface->name);
4059 int error = netdev_set_etheraddr(iface->netdev, ea);
4061 VLOG_ERR("interface %s: setting MAC failed (%s)",
4062 iface->name, strerror(error));
4068 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4070 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4073 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4077 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4079 * The value strings in '*shash' are taken directly from values[], not copied,
4080 * so the caller should not modify or free them. */
4082 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4083 struct shash *shash)
4088 for (i = 0; i < n; i++) {
4089 shash_add(shash, keys[i], values[i]);
4093 struct iface_delete_queues_cbdata {
4094 struct netdev *netdev;
4095 const struct ovsdb_datum *queues;
4099 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4101 union ovsdb_atom atom;
4103 atom.integer = target;
4104 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4108 iface_delete_queues(unsigned int queue_id,
4109 const struct shash *details OVS_UNUSED, void *cbdata_)
4111 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4113 if (!queue_ids_include(cbdata->queues, queue_id)) {
4114 netdev_delete_queue(cbdata->netdev, queue_id);
4119 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4121 if (!qos || qos->type[0] == '\0') {
4122 netdev_set_qos(iface->netdev, NULL, NULL);
4124 struct iface_delete_queues_cbdata cbdata;
4125 struct shash details;
4128 /* Configure top-level Qos for 'iface'. */
4129 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4130 qos->n_other_config, &details);
4131 netdev_set_qos(iface->netdev, qos->type, &details);
4132 shash_destroy(&details);
4134 /* Deconfigure queues that were deleted. */
4135 cbdata.netdev = iface->netdev;
4136 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4138 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4140 /* Configure queues for 'iface'. */
4141 for (i = 0; i < qos->n_queues; i++) {
4142 const struct ovsrec_queue *queue = qos->value_queues[i];
4143 unsigned int queue_id = qos->key_queues[i];
4145 shash_from_ovs_idl_map(queue->key_other_config,
4146 queue->value_other_config,
4147 queue->n_other_config, &details);
4148 netdev_set_queue(iface->netdev, queue_id, &details);
4149 shash_destroy(&details);
4155 iface_update_cfm(struct iface *iface)
4159 uint16_t *remote_mps;
4160 struct ovsrec_monitor *mon;
4161 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4163 mon = iface->cfg->monitor;
4169 if (netdev_get_etheraddr(iface->netdev, ea)) {
4170 VLOG_WARN("interface %s: Failed to get ethernet address. "
4171 "Skipping Monitor.", iface->name);
4175 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4176 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4181 iface->cfm = cfm_create();
4185 cfm->mpid = mon->mpid;
4186 cfm->interval = mon->interval ? *mon->interval : 1000;
4188 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4189 memcpy(cfm->maid, maid, sizeof cfm->maid);
4191 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4192 for(i = 0; i < mon->n_remote_mps; i++) {
4193 remote_mps[i] = mon->remote_mps[i]->mpid;
4195 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4198 if (!cfm_configure(iface->cfm)) {
4199 cfm_destroy(iface->cfm);
4204 /* Port mirroring. */
4206 static struct mirror *
4207 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4211 for (i = 0; i < MAX_MIRRORS; i++) {
4212 struct mirror *m = br->mirrors[i];
4213 if (m && uuid_equals(uuid, &m->uuid)) {
4221 mirror_reconfigure(struct bridge *br)
4223 unsigned long *rspan_vlans;
4226 /* Get rid of deleted mirrors. */
4227 for (i = 0; i < MAX_MIRRORS; i++) {
4228 struct mirror *m = br->mirrors[i];
4230 const struct ovsdb_datum *mc;
4231 union ovsdb_atom atom;
4233 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4234 atom.uuid = br->mirrors[i]->uuid;
4235 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4241 /* Add new mirrors and reconfigure existing ones. */
4242 for (i = 0; i < br->cfg->n_mirrors; i++) {
4243 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4244 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4246 mirror_reconfigure_one(m, cfg);
4248 mirror_create(br, cfg);
4252 /* Update port reserved status. */
4253 for (i = 0; i < br->n_ports; i++) {
4254 br->ports[i]->is_mirror_output_port = false;
4256 for (i = 0; i < MAX_MIRRORS; i++) {
4257 struct mirror *m = br->mirrors[i];
4258 if (m && m->out_port) {
4259 m->out_port->is_mirror_output_port = true;
4263 /* Update flooded vlans (for RSPAN). */
4265 if (br->cfg->n_flood_vlans) {
4266 rspan_vlans = bitmap_allocate(4096);
4268 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4269 int64_t vlan = br->cfg->flood_vlans[i];
4270 if (vlan >= 0 && vlan < 4096) {
4271 bitmap_set1(rspan_vlans, vlan);
4272 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4275 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4280 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4286 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4291 for (i = 0; ; i++) {
4292 if (i >= MAX_MIRRORS) {
4293 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4294 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4297 if (!br->mirrors[i]) {
4302 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4305 br->mirrors[i] = m = xzalloc(sizeof *m);
4308 m->name = xstrdup(cfg->name);
4309 shash_init(&m->src_ports);
4310 shash_init(&m->dst_ports);
4316 mirror_reconfigure_one(m, cfg);
4320 mirror_destroy(struct mirror *m)
4323 struct bridge *br = m->bridge;
4326 for (i = 0; i < br->n_ports; i++) {
4327 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4328 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4331 shash_destroy(&m->src_ports);
4332 shash_destroy(&m->dst_ports);
4335 m->bridge->mirrors[m->idx] = NULL;
4344 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4345 struct shash *names)
4349 for (i = 0; i < n_ports; i++) {
4350 const char *name = ports[i]->name;
4351 if (port_lookup(m->bridge, name)) {
4352 shash_add_once(names, name, NULL);
4354 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4355 "port %s", m->bridge->name, m->name, name);
4361 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4367 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4369 for (i = 0; i < cfg->n_select_vlan; i++) {
4370 int64_t vlan = cfg->select_vlan[i];
4371 if (vlan < 0 || vlan > 4095) {
4372 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4373 m->bridge->name, m->name, vlan);
4375 (*vlans)[n_vlans++] = vlan;
4382 vlan_is_mirrored(const struct mirror *m, int vlan)
4386 for (i = 0; i < m->n_vlans; i++) {
4387 if (m->vlans[i] == vlan) {
4395 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4399 for (i = 0; i < m->n_vlans; i++) {
4400 if (port_trunks_vlan(p, m->vlans[i])) {
4408 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4410 struct shash src_ports, dst_ports;
4411 mirror_mask_t mirror_bit;
4412 struct port *out_port;
4419 if (strcmp(cfg->name, m->name)) {
4421 m->name = xstrdup(cfg->name);
4424 /* Get output port. */
4425 if (cfg->output_port) {
4426 out_port = port_lookup(m->bridge, cfg->output_port->name);
4428 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4429 m->bridge->name, m->name);
4435 if (cfg->output_vlan) {
4436 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4437 "output vlan; ignoring output vlan",
4438 m->bridge->name, m->name);
4440 } else if (cfg->output_vlan) {
4442 out_vlan = *cfg->output_vlan;
4444 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4445 m->bridge->name, m->name);
4450 shash_init(&src_ports);
4451 shash_init(&dst_ports);
4452 if (cfg->select_all) {
4453 for (i = 0; i < m->bridge->n_ports; i++) {
4454 const char *name = m->bridge->ports[i]->name;
4455 shash_add_once(&src_ports, name, NULL);
4456 shash_add_once(&dst_ports, name, NULL);
4461 /* Get ports, and drop duplicates and ports that don't exist. */
4462 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4464 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4467 /* Get all the vlans, and drop duplicate and invalid vlans. */
4468 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4471 /* Update mirror data. */
4472 if (!shash_equal_keys(&m->src_ports, &src_ports)
4473 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4474 || m->n_vlans != n_vlans
4475 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4476 || m->out_port != out_port
4477 || m->out_vlan != out_vlan) {
4478 bridge_flush(m->bridge);
4480 shash_swap(&m->src_ports, &src_ports);
4481 shash_swap(&m->dst_ports, &dst_ports);
4484 m->n_vlans = n_vlans;
4485 m->out_port = out_port;
4486 m->out_vlan = out_vlan;
4489 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4490 for (i = 0; i < m->bridge->n_ports; i++) {
4491 struct port *port = m->bridge->ports[i];
4493 if (shash_find(&m->src_ports, port->name)
4496 ? port_trunks_any_mirrored_vlan(m, port)
4497 : vlan_is_mirrored(m, port->vlan)))) {
4498 port->src_mirrors |= mirror_bit;
4500 port->src_mirrors &= ~mirror_bit;
4503 if (shash_find(&m->dst_ports, port->name)) {
4504 port->dst_mirrors |= mirror_bit;
4506 port->dst_mirrors &= ~mirror_bit;
4511 shash_destroy(&src_ports);
4512 shash_destroy(&dst_ports);