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"
49 #include "ofp-print.h"
51 #include "ofproto/netflow.h"
52 #include "ofproto/ofproto.h"
53 #include "ovsdb-data.h"
55 #include "poll-loop.h"
56 #include "proc-net-compat.h"
60 #include "socket-util.h"
61 #include "stream-ssl.h"
63 #include "system-stats.h"
68 #include "vswitchd/vswitch-idl.h"
69 #include "xenserver.h"
71 #include "sflow_api.h"
73 VLOG_DEFINE_THIS_MODULE(bridge);
75 COVERAGE_DEFINE(bridge_flush);
76 COVERAGE_DEFINE(bridge_process_flow);
77 COVERAGE_DEFINE(bridge_reconfigure);
85 /* These members are always valid. */
86 struct port *port; /* Containing port. */
87 size_t port_ifidx; /* Index within containing port. */
88 char *name; /* Host network device name. */
89 tag_type tag; /* Tag associated with this interface. */
90 long long delay_expires; /* Time after which 'enabled' may change. */
92 /* These members are valid only after bridge_reconfigure() causes them to
94 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
95 int dp_ifidx; /* Index within kernel datapath. */
96 struct netdev *netdev; /* Network device. */
97 bool enabled; /* May be chosen for flows? */
98 const char *type; /* Usually same as cfg->type. */
99 struct cfm *cfm; /* Connectivity Fault Management */
100 const struct ovsrec_interface *cfg;
103 #define BOND_MASK 0xff
105 int iface_idx; /* Index of assigned iface, or -1 if none. */
106 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
107 tag_type iface_tag; /* Tag associated with iface_idx. */
110 #define MAX_MIRRORS 32
111 typedef uint32_t mirror_mask_t;
112 #define MIRROR_MASK_C(X) UINT32_C(X)
113 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
115 struct bridge *bridge;
118 struct uuid uuid; /* UUID of this "mirror" record in database. */
120 /* Selection criteria. */
121 struct shash src_ports; /* Name is port name; data is always NULL. */
122 struct shash dst_ports; /* Name is port name; data is always NULL. */
127 struct port *out_port;
131 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
133 struct bridge *bridge;
135 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
136 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
137 * NULL if all VLANs are trunked. */
138 const struct ovsrec_port *cfg;
141 /* An ordinary bridge port has 1 interface.
142 * A bridge port for bonding has at least 2 interfaces. */
143 struct iface **ifaces;
144 size_t n_ifaces, allocated_ifaces;
147 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
148 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
149 tag_type active_iface_tag; /* Tag for bcast flows. */
150 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
151 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
152 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
153 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
154 long long int bond_next_fake_iface_update; /* Time of next update. */
155 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
156 long long int bond_next_rebalance; /* Next rebalancing time. */
157 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
159 /* Port mirroring info. */
160 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
161 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
162 bool is_mirror_output_port; /* Does port mirroring send frames here? */
165 #define DP_MAX_PORTS 255
167 struct list node; /* Node in global list of bridges. */
168 char *name; /* User-specified arbitrary name. */
169 struct mac_learning *ml; /* MAC learning table. */
170 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
171 const struct ovsrec_bridge *cfg;
173 /* OpenFlow switch processing. */
174 struct ofproto *ofproto; /* OpenFlow switch. */
176 /* Kernel datapath information. */
177 struct dpif *dpif; /* Datapath. */
178 struct hmap ifaces; /* Contains "struct iface"s. */
182 size_t n_ports, allocated_ports;
183 struct shash iface_by_name; /* "struct iface"s indexed by name. */
184 struct shash port_by_name; /* "struct port"s indexed by name. */
187 bool has_bonded_ports;
192 /* Port mirroring. */
193 struct mirror *mirrors[MAX_MIRRORS];
196 /* List of all bridges. */
197 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
199 /* OVSDB IDL used to obtain configuration. */
200 static struct ovsdb_idl *idl;
202 /* Each time this timer expires, the bridge fetches systems and interface
203 * statistics and pushes them into the database. */
204 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
205 static long long int stats_timer = LLONG_MIN;
207 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
208 static void bridge_destroy(struct bridge *);
209 static struct bridge *bridge_lookup(const char *name);
210 static unixctl_cb_func bridge_unixctl_dump_flows;
211 static unixctl_cb_func bridge_unixctl_reconnect;
212 static int bridge_run_one(struct bridge *);
213 static size_t bridge_get_controllers(const struct bridge *br,
214 struct ovsrec_controller ***controllersp);
215 static void bridge_reconfigure_one(struct bridge *);
216 static void bridge_reconfigure_remotes(struct bridge *,
217 const struct sockaddr_in *managers,
219 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
220 static void bridge_fetch_dp_ifaces(struct bridge *);
221 static void bridge_flush(struct bridge *);
222 static void bridge_pick_local_hw_addr(struct bridge *,
223 uint8_t ea[ETH_ADDR_LEN],
224 struct iface **hw_addr_iface);
225 static uint64_t bridge_pick_datapath_id(struct bridge *,
226 const uint8_t bridge_ea[ETH_ADDR_LEN],
227 struct iface *hw_addr_iface);
228 static struct iface *bridge_get_local_iface(struct bridge *);
229 static uint64_t dpid_from_hash(const void *, size_t nbytes);
231 static unixctl_cb_func bridge_unixctl_fdb_show;
233 static void bond_init(void);
234 static void bond_run(struct bridge *);
235 static void bond_wait(struct bridge *);
236 static void bond_rebalance_port(struct port *);
237 static void bond_send_learning_packets(struct port *);
238 static void bond_enable_slave(struct iface *iface, bool enable);
240 static struct port *port_create(struct bridge *, const char *name);
241 static void port_reconfigure(struct port *, const struct ovsrec_port *);
242 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
243 static void port_destroy(struct port *);
244 static struct port *port_lookup(const struct bridge *, const char *name);
245 static struct iface *port_lookup_iface(const struct port *, const char *name);
246 static struct port *port_from_dp_ifidx(const struct bridge *,
248 static void port_update_bond_compat(struct port *);
249 static void port_update_vlan_compat(struct port *);
250 static void port_update_bonding(struct port *);
252 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
253 static void mirror_destroy(struct mirror *);
254 static void mirror_reconfigure(struct bridge *);
255 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
256 static bool vlan_is_mirrored(const struct mirror *, int vlan);
258 static struct iface *iface_create(struct port *port,
259 const struct ovsrec_interface *if_cfg);
260 static void iface_destroy(struct iface *);
261 static struct iface *iface_lookup(const struct bridge *, const char *name);
262 static struct iface *iface_from_dp_ifidx(const struct bridge *,
264 static void iface_set_mac(struct iface *);
265 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
266 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
267 static void iface_update_cfm(struct iface *);
268 static void iface_refresh_cfm_stats(struct iface *iface);
269 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
271 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
274 /* Hooks into ofproto processing. */
275 static struct ofhooks bridge_ofhooks;
277 /* Public functions. */
279 /* Initializes the bridge module, configuring it to obtain its configuration
280 * from an OVSDB server accessed over 'remote', which should be a string in a
281 * form acceptable to ovsdb_idl_create(). */
283 bridge_init(const char *remote)
285 /* Create connection to database. */
286 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
288 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
289 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
290 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
292 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
294 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
295 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
297 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
298 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
299 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
301 /* Register unixctl commands. */
302 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
303 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
305 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
310 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
311 * but for which the ovs-vswitchd configuration 'cfg' is required. */
313 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
315 static bool already_configured_once;
316 struct svec bridge_names;
317 struct svec dpif_names, dpif_types;
320 /* Only do this once per ovs-vswitchd run. */
321 if (already_configured_once) {
324 already_configured_once = true;
326 stats_timer = time_msec() + STATS_INTERVAL;
328 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
329 svec_init(&bridge_names);
330 for (i = 0; i < cfg->n_bridges; i++) {
331 svec_add(&bridge_names, cfg->bridges[i]->name);
333 svec_sort(&bridge_names);
335 /* Iterate over all system dpifs and delete any of them that do not appear
337 svec_init(&dpif_names);
338 svec_init(&dpif_types);
339 dp_enumerate_types(&dpif_types);
340 for (i = 0; i < dpif_types.n; i++) {
345 dp_enumerate_names(dpif_types.names[i], &dpif_names);
347 /* For each dpif... */
348 for (j = 0; j < dpif_names.n; j++) {
349 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
351 struct svec all_names;
354 /* ...check whether any of its names is in 'bridge_names'. */
355 svec_init(&all_names);
356 dpif_get_all_names(dpif, &all_names);
357 for (k = 0; k < all_names.n; k++) {
358 if (svec_contains(&bridge_names, all_names.names[k])) {
363 /* No. Delete the dpif. */
367 svec_destroy(&all_names);
372 svec_destroy(&bridge_names);
373 svec_destroy(&dpif_names);
374 svec_destroy(&dpif_types);
377 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
378 * keys from "options" and "other_config", preferring "options" keys over
379 * "other_config" keys.
381 * The value strings in '*options' are taken directly from if_cfg, not copied,
382 * so the caller should not modify or free them. */
384 iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
388 shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options,
389 if_cfg->n_options, options);
391 for (i = 0; i < if_cfg->n_other_config; i++) {
392 char *key = if_cfg->key_other_config[i];
393 char *value = if_cfg->value_other_config[i];
395 if (!shash_find_data(options, key)) {
396 shash_add(options, key, value);
398 VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts "
399 "with \"options\" key %s", if_cfg->name, key, key);
404 /* Returns the type of network device that 'iface' should have. (This is
405 * ordinarily the same type as the interface, but the network devices for
406 * "internal" ports have type "system".) */
408 iface_get_netdev_type(const struct iface *iface)
410 return !strcmp(iface->type, "internal") ? "system" : iface->type;
413 /* Attempt to create the network device for 'iface' through the netdev
416 create_iface_netdev(struct iface *iface)
418 struct netdev_options netdev_options;
419 struct shash options;
422 memset(&netdev_options, 0, sizeof netdev_options);
423 netdev_options.name = iface->cfg->name;
424 netdev_options.type = iface_get_netdev_type(iface);
425 netdev_options.args = &options;
426 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
428 iface_get_options(iface->cfg, &options);
430 error = netdev_open(&netdev_options, &iface->netdev);
433 iface->enabled = netdev_get_carrier(iface->netdev);
436 shash_destroy(&options);
442 reconfigure_iface_netdev(struct iface *iface)
444 const char *netdev_type, *iface_type;
445 struct shash options;
448 /* Skip reconfiguration if the device has the wrong type. This shouldn't
450 iface_type = iface_get_netdev_type(iface);
451 netdev_type = netdev_get_type(iface->netdev);
452 if (iface_type && strcmp(netdev_type, iface_type)) {
453 VLOG_WARN("%s: attempting change device type from %s to %s",
454 iface->cfg->name, netdev_type, iface_type);
458 /* Reconfigure device. */
459 iface_get_options(iface->cfg, &options);
460 error = netdev_reconfigure(iface->netdev, &options);
461 shash_destroy(&options);
466 /* Callback for iterate_and_prune_ifaces(). */
468 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
470 if (!iface->netdev) {
471 /* We already reported a related error, don't bother duplicating it. */
475 if (iface->dp_ifidx < 0) {
476 VLOG_ERR("%s interface not in %s, dropping",
477 iface->name, dpif_name(br->dpif));
481 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
482 iface->name, iface->dp_ifidx);
486 /* Callback for iterate_and_prune_ifaces(). */
488 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
489 void *aux OVS_UNUSED)
491 /* Set policing attributes. */
492 netdev_set_policing(iface->netdev,
493 iface->cfg->ingress_policing_rate,
494 iface->cfg->ingress_policing_burst);
496 /* Set MAC address of internal interfaces other than the local
498 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
499 iface_set_mac(iface);
505 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
506 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
507 * deletes from 'br' any ports that no longer have any interfaces. */
509 iterate_and_prune_ifaces(struct bridge *br,
510 bool (*cb)(struct bridge *, struct iface *,
516 for (i = 0; i < br->n_ports; ) {
517 struct port *port = br->ports[i];
518 for (j = 0; j < port->n_ifaces; ) {
519 struct iface *iface = port->ifaces[j];
520 if (cb(br, iface, aux)) {
523 iface_set_ofport(iface->cfg, -1);
524 iface_destroy(iface);
528 if (port->n_ifaces) {
531 VLOG_ERR("%s port has no interfaces, dropping", port->name);
537 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
538 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
539 * responsible for freeing '*managersp' (with free()).
541 * You may be asking yourself "why does ovs-vswitchd care?", because
542 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
543 * should not be and in fact is not directly involved in that. But
544 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
545 * it has to tell in-band control where the managers are to enable that.
546 * (Thus, only managers connected in-band are collected.)
549 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
550 struct sockaddr_in **managersp, size_t *n_managersp)
552 struct sockaddr_in *managers = NULL;
553 size_t n_managers = 0;
554 struct shash targets;
557 /* Collect all of the potential targets, as the union of the "managers"
558 * column and the "targets" columns of the rows pointed to by
559 * "manager_options", excluding any that are out-of-band. */
560 shash_init(&targets);
561 for (i = 0; i < ovs_cfg->n_managers; i++) {
562 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
564 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
565 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
567 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
568 shash_find_and_delete(&targets, m->target);
570 shash_add_once(&targets, m->target, NULL);
574 /* Now extract the targets' IP addresses. */
575 if (!shash_is_empty(&targets)) {
576 struct shash_node *node;
578 managers = xmalloc(shash_count(&targets) * sizeof *managers);
579 SHASH_FOR_EACH (node, &targets) {
580 const char *target = node->name;
581 struct sockaddr_in *sin = &managers[n_managers];
583 if ((!strncmp(target, "tcp:", 4)
584 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
585 (!strncmp(target, "ssl:", 4)
586 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
591 shash_destroy(&targets);
593 *managersp = managers;
594 *n_managersp = n_managers;
598 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
600 struct shash old_br, new_br;
601 struct shash_node *node;
602 struct bridge *br, *next;
603 struct sockaddr_in *managers;
606 int sflow_bridge_number;
608 COVERAGE_INC(bridge_reconfigure);
610 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
612 /* Collect old and new bridges. */
615 LIST_FOR_EACH (br, node, &all_bridges) {
616 shash_add(&old_br, br->name, br);
618 for (i = 0; i < ovs_cfg->n_bridges; i++) {
619 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
620 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
621 VLOG_WARN("more than one bridge named %s", br_cfg->name);
625 /* Get rid of deleted bridges and add new bridges. */
626 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
627 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
634 SHASH_FOR_EACH (node, &new_br) {
635 const char *br_name = node->name;
636 const struct ovsrec_bridge *br_cfg = node->data;
637 br = shash_find_data(&old_br, br_name);
639 /* If the bridge datapath type has changed, we need to tear it
640 * down and recreate. */
641 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
643 bridge_create(br_cfg);
646 bridge_create(br_cfg);
649 shash_destroy(&old_br);
650 shash_destroy(&new_br);
652 /* Reconfigure all bridges. */
653 LIST_FOR_EACH (br, node, &all_bridges) {
654 bridge_reconfigure_one(br);
657 /* Add and delete ports on all datapaths.
659 * The kernel will reject any attempt to add a given port to a datapath if
660 * that port already belongs to a different datapath, so we must do all
661 * port deletions before any port additions. */
662 LIST_FOR_EACH (br, node, &all_bridges) {
663 struct odp_port *dpif_ports;
665 struct shash want_ifaces;
667 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
668 bridge_get_all_ifaces(br, &want_ifaces);
669 for (i = 0; i < n_dpif_ports; i++) {
670 const struct odp_port *p = &dpif_ports[i];
671 if (!shash_find(&want_ifaces, p->devname)
672 && strcmp(p->devname, br->name)) {
673 int retval = dpif_port_del(br->dpif, p->port);
675 VLOG_ERR("failed to remove %s interface from %s: %s",
676 p->devname, dpif_name(br->dpif),
681 shash_destroy(&want_ifaces);
684 LIST_FOR_EACH (br, node, &all_bridges) {
685 struct odp_port *dpif_ports;
687 struct shash cur_ifaces, want_ifaces;
689 /* Get the set of interfaces currently in this datapath. */
690 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
691 shash_init(&cur_ifaces);
692 for (i = 0; i < n_dpif_ports; i++) {
693 const char *name = dpif_ports[i].devname;
694 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
697 /* Get the set of interfaces we want on this datapath. */
698 bridge_get_all_ifaces(br, &want_ifaces);
700 hmap_clear(&br->ifaces);
701 SHASH_FOR_EACH (node, &want_ifaces) {
702 const char *if_name = node->name;
703 struct iface *iface = node->data;
704 bool internal = !iface || !strcmp(iface->type, "internal");
705 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
708 /* If we have a port or a netdev already, and it's not the type we
709 * want, then delete the port (if any) and close the netdev (if
712 ? dpif_port && !(dpif_port->flags & ODP_PORT_INTERNAL)
714 && strcmp(iface->type, netdev_get_type(iface->netdev))))
717 error = ofproto_port_del(br->ofproto, dpif_port->port);
724 netdev_close(iface->netdev);
725 iface->netdev = NULL;
729 /* If it's not an internal port, open (possibly create) the
732 if (!iface->netdev) {
733 error = create_iface_netdev(iface);
735 VLOG_WARN("could not create iface %s: %s", iface->name,
740 reconfigure_iface_netdev(iface);
744 /* If it's not part of the datapath, add it. */
746 error = dpif_port_add(br->dpif, if_name,
747 internal ? ODP_PORT_INTERNAL : 0, NULL);
748 if (error == EFBIG) {
749 VLOG_ERR("ran out of valid port numbers on %s",
750 dpif_name(br->dpif));
753 VLOG_ERR("failed to add %s interface to %s: %s",
754 if_name, dpif_name(br->dpif), strerror(error));
759 /* If it's an internal port, open the netdev. */
761 if (iface && !iface->netdev) {
762 error = create_iface_netdev(iface);
764 VLOG_WARN("could not create iface %s: %s", iface->name,
770 assert(iface->netdev != NULL);
774 shash_destroy(&cur_ifaces);
775 shash_destroy(&want_ifaces);
777 sflow_bridge_number = 0;
778 LIST_FOR_EACH (br, node, &all_bridges) {
781 struct iface *local_iface;
782 struct iface *hw_addr_iface;
785 bridge_fetch_dp_ifaces(br);
787 iterate_and_prune_ifaces(br, check_iface, NULL);
789 /* Pick local port hardware address, datapath ID. */
790 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
791 local_iface = bridge_get_local_iface(br);
793 int error = netdev_set_etheraddr(local_iface->netdev, ea);
795 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
796 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
797 "Ethernet address: %s",
798 br->name, strerror(error));
802 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
803 ofproto_set_datapath_id(br->ofproto, dpid);
805 dpid_string = xasprintf("%016"PRIx64, dpid);
806 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
809 /* Set NetFlow configuration on this bridge. */
810 if (br->cfg->netflow) {
811 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
812 struct netflow_options opts;
814 memset(&opts, 0, sizeof opts);
816 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
817 if (nf_cfg->engine_type) {
818 opts.engine_type = *nf_cfg->engine_type;
820 if (nf_cfg->engine_id) {
821 opts.engine_id = *nf_cfg->engine_id;
824 opts.active_timeout = nf_cfg->active_timeout;
825 if (!opts.active_timeout) {
826 opts.active_timeout = -1;
827 } else if (opts.active_timeout < 0) {
828 VLOG_WARN("bridge %s: active timeout interval set to negative "
829 "value, using default instead (%d seconds)", br->name,
830 NF_ACTIVE_TIMEOUT_DEFAULT);
831 opts.active_timeout = -1;
834 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
835 if (opts.add_id_to_iface) {
836 if (opts.engine_id > 0x7f) {
837 VLOG_WARN("bridge %s: netflow port mangling may conflict "
838 "with another vswitch, choose an engine id less "
839 "than 128", br->name);
841 if (br->n_ports > 508) {
842 VLOG_WARN("bridge %s: netflow port mangling will conflict "
843 "with another port when more than 508 ports are "
848 opts.collectors.n = nf_cfg->n_targets;
849 opts.collectors.names = nf_cfg->targets;
850 if (ofproto_set_netflow(br->ofproto, &opts)) {
851 VLOG_ERR("bridge %s: problem setting netflow collectors",
855 ofproto_set_netflow(br->ofproto, NULL);
858 /* Set sFlow configuration on this bridge. */
859 if (br->cfg->sflow) {
860 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
861 struct ovsrec_controller **controllers;
862 struct ofproto_sflow_options oso;
863 size_t n_controllers;
865 memset(&oso, 0, sizeof oso);
867 oso.targets.n = sflow_cfg->n_targets;
868 oso.targets.names = sflow_cfg->targets;
870 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
871 if (sflow_cfg->sampling) {
872 oso.sampling_rate = *sflow_cfg->sampling;
875 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
876 if (sflow_cfg->polling) {
877 oso.polling_interval = *sflow_cfg->polling;
880 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
881 if (sflow_cfg->header) {
882 oso.header_len = *sflow_cfg->header;
885 oso.sub_id = sflow_bridge_number++;
886 oso.agent_device = sflow_cfg->agent;
888 oso.control_ip = NULL;
889 n_controllers = bridge_get_controllers(br, &controllers);
890 for (i = 0; i < n_controllers; i++) {
891 if (controllers[i]->local_ip) {
892 oso.control_ip = controllers[i]->local_ip;
896 ofproto_set_sflow(br->ofproto, &oso);
898 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
900 ofproto_set_sflow(br->ofproto, NULL);
903 /* Update the controller and related settings. It would be more
904 * straightforward to call this from bridge_reconfigure_one(), but we
905 * can't do it there for two reasons. First, and most importantly, at
906 * that point we don't know the dp_ifidx of any interfaces that have
907 * been added to the bridge (because we haven't actually added them to
908 * the datapath). Second, at that point we haven't set the datapath ID
909 * yet; when a controller is configured, resetting the datapath ID will
910 * immediately disconnect from the controller, so it's better to set
911 * the datapath ID before the controller. */
912 bridge_reconfigure_remotes(br, managers, n_managers);
914 LIST_FOR_EACH (br, node, &all_bridges) {
915 for (i = 0; i < br->n_ports; i++) {
916 struct port *port = br->ports[i];
919 port_update_vlan_compat(port);
920 port_update_bonding(port);
922 for (j = 0; j < port->n_ifaces; j++) {
923 iface_update_qos(port->ifaces[j], port->cfg->qos);
927 LIST_FOR_EACH (br, node, &all_bridges) {
928 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
931 LIST_FOR_EACH (br, node, &all_bridges) {
933 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
934 iface_update_cfm(iface);
942 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
943 const struct ovsdb_idl_column *column,
946 const struct ovsdb_datum *datum;
947 union ovsdb_atom atom;
950 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
951 atom.string = (char *) key;
952 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
953 return idx == UINT_MAX ? NULL : datum->values[idx].string;
957 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
959 return get_ovsrec_key_value(&br_cfg->header_,
960 &ovsrec_bridge_col_other_config, key);
964 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
965 struct iface **hw_addr_iface)
971 *hw_addr_iface = NULL;
973 /* Did the user request a particular MAC? */
974 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
975 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
976 if (eth_addr_is_multicast(ea)) {
977 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
978 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
979 } else if (eth_addr_is_zero(ea)) {
980 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
986 /* Otherwise choose the minimum non-local MAC address among all of the
988 memset(ea, 0xff, sizeof ea);
989 for (i = 0; i < br->n_ports; i++) {
990 struct port *port = br->ports[i];
991 uint8_t iface_ea[ETH_ADDR_LEN];
994 /* Mirror output ports don't participate. */
995 if (port->is_mirror_output_port) {
999 /* Choose the MAC address to represent the port. */
1000 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1001 /* Find the interface with this Ethernet address (if any) so that
1002 * we can provide the correct devname to the caller. */
1004 for (j = 0; j < port->n_ifaces; j++) {
1005 struct iface *candidate = port->ifaces[j];
1006 uint8_t candidate_ea[ETH_ADDR_LEN];
1007 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1008 && eth_addr_equals(iface_ea, candidate_ea)) {
1013 /* Choose the interface whose MAC address will represent the port.
1014 * The Linux kernel bonding code always chooses the MAC address of
1015 * the first slave added to a bond, and the Fedora networking
1016 * scripts always add slaves to a bond in alphabetical order, so
1017 * for compatibility we choose the interface with the name that is
1018 * first in alphabetical order. */
1019 iface = port->ifaces[0];
1020 for (j = 1; j < port->n_ifaces; j++) {
1021 struct iface *candidate = port->ifaces[j];
1022 if (strcmp(candidate->name, iface->name) < 0) {
1027 /* The local port doesn't count (since we're trying to choose its
1028 * MAC address anyway). */
1029 if (iface->dp_ifidx == ODPP_LOCAL) {
1034 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1036 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1037 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1038 iface->name, strerror(error));
1043 /* Compare against our current choice. */
1044 if (!eth_addr_is_multicast(iface_ea) &&
1045 !eth_addr_is_local(iface_ea) &&
1046 !eth_addr_is_reserved(iface_ea) &&
1047 !eth_addr_is_zero(iface_ea) &&
1048 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
1050 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1051 *hw_addr_iface = iface;
1054 if (eth_addr_is_multicast(ea)) {
1055 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1056 *hw_addr_iface = NULL;
1057 VLOG_WARN("bridge %s: using default bridge Ethernet "
1058 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1060 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1061 br->name, ETH_ADDR_ARGS(ea));
1065 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1066 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1067 * an interface on 'br', then that interface must be passed in as
1068 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1069 * 'hw_addr_iface' must be passed in as a null pointer. */
1071 bridge_pick_datapath_id(struct bridge *br,
1072 const uint8_t bridge_ea[ETH_ADDR_LEN],
1073 struct iface *hw_addr_iface)
1076 * The procedure for choosing a bridge MAC address will, in the most
1077 * ordinary case, also choose a unique MAC that we can use as a datapath
1078 * ID. In some special cases, though, multiple bridges will end up with
1079 * the same MAC address. This is OK for the bridges, but it will confuse
1080 * the OpenFlow controller, because each datapath needs a unique datapath
1083 * Datapath IDs must be unique. It is also very desirable that they be
1084 * stable from one run to the next, so that policy set on a datapath
1087 const char *datapath_id;
1090 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1091 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1095 if (hw_addr_iface) {
1097 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1099 * A bridge whose MAC address is taken from a VLAN network device
1100 * (that is, a network device created with vconfig(8) or similar
1101 * tool) will have the same MAC address as a bridge on the VLAN
1102 * device's physical network device.
1104 * Handle this case by hashing the physical network device MAC
1105 * along with the VLAN identifier.
1107 uint8_t buf[ETH_ADDR_LEN + 2];
1108 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1109 buf[ETH_ADDR_LEN] = vlan >> 8;
1110 buf[ETH_ADDR_LEN + 1] = vlan;
1111 return dpid_from_hash(buf, sizeof buf);
1114 * Assume that this bridge's MAC address is unique, since it
1115 * doesn't fit any of the cases we handle specially.
1120 * A purely internal bridge, that is, one that has no non-virtual
1121 * network devices on it at all, is more difficult because it has no
1122 * natural unique identifier at all.
1124 * When the host is a XenServer, we handle this case by hashing the
1125 * host's UUID with the name of the bridge. Names of bridges are
1126 * persistent across XenServer reboots, although they can be reused if
1127 * an internal network is destroyed and then a new one is later
1128 * created, so this is fairly effective.
1130 * When the host is not a XenServer, we punt by using a random MAC
1131 * address on each run.
1133 const char *host_uuid = xenserver_get_host_uuid();
1135 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1136 dpid = dpid_from_hash(combined, strlen(combined));
1142 return eth_addr_to_uint64(bridge_ea);
1146 dpid_from_hash(const void *data, size_t n)
1148 uint8_t hash[SHA1_DIGEST_SIZE];
1150 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1151 sha1_bytes(data, n, hash);
1152 eth_addr_mark_random(hash);
1153 return eth_addr_to_uint64(hash);
1157 iface_refresh_cfm_stats(struct iface *iface)
1161 const struct ovsrec_monitor *mon;
1163 mon = iface->cfg->monitor;
1170 for (i = 0; i < mon->n_remote_mps; i++) {
1171 const struct ovsrec_maintenance_point *mp;
1172 const struct remote_mp *rmp;
1174 mp = mon->remote_mps[i];
1175 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1177 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1180 if (hmap_is_empty(&cfm->x_remote_mps)) {
1181 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1184 struct remote_mp *rmp;
1185 int64_t *x_remote_mps;
1187 length = hmap_count(&cfm->x_remote_mps);
1188 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1191 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1192 x_remote_mps[i++] = rmp->mpid;
1195 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1199 if (hmap_is_empty(&cfm->x_remote_maids)) {
1200 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1203 char **x_remote_maids;
1204 struct remote_maid *rmaid;
1206 length = hmap_count(&cfm->x_remote_maids);
1207 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1210 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1213 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1215 for (j = 0; j < CCM_MAID_LEN; j++) {
1216 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1221 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1223 for (i = 0; i < length; i++) {
1224 free(x_remote_maids[i]);
1226 free(x_remote_maids);
1229 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1233 iface_refresh_stats(struct iface *iface)
1239 static const struct iface_stat iface_stats[] = {
1240 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1241 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1242 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1243 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1244 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1245 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1246 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1247 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1248 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1249 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1250 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1251 { "collisions", offsetof(struct netdev_stats, collisions) },
1253 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1254 const struct iface_stat *s;
1256 char *keys[N_STATS];
1257 int64_t values[N_STATS];
1260 struct netdev_stats stats;
1262 /* Intentionally ignore return value, since errors will set 'stats' to
1263 * all-1s, and we will deal with that correctly below. */
1264 netdev_get_stats(iface->netdev, &stats);
1267 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1268 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1269 if (value != UINT64_MAX) {
1276 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1280 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1282 struct ovsdb_datum datum;
1286 get_system_stats(&stats);
1288 ovsdb_datum_from_shash(&datum, &stats);
1289 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1296 const struct ovsrec_open_vswitch *cfg;
1298 bool datapath_destroyed;
1299 bool database_changed;
1302 /* Let each bridge do the work that it needs to do. */
1303 datapath_destroyed = false;
1304 LIST_FOR_EACH (br, node, &all_bridges) {
1305 int error = bridge_run_one(br);
1307 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1308 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1309 "forcing reconfiguration", br->name);
1310 datapath_destroyed = true;
1314 /* (Re)configure if necessary. */
1315 database_changed = ovsdb_idl_run(idl);
1316 cfg = ovsrec_open_vswitch_first(idl);
1317 if (database_changed || datapath_destroyed) {
1319 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1321 bridge_configure_once(cfg);
1322 bridge_reconfigure(cfg);
1324 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1325 ovsdb_idl_txn_commit(txn);
1326 ovsdb_idl_txn_destroy(txn); /* XXX */
1328 /* We still need to reconfigure to avoid dangling pointers to
1329 * now-destroyed ovsrec structures inside bridge data. */
1330 static const struct ovsrec_open_vswitch null_cfg;
1332 bridge_reconfigure(&null_cfg);
1337 /* Re-configure SSL. We do this on every trip through the main loop,
1338 * instead of just when the database changes, because the contents of the
1339 * key and certificate files can change without the database changing. */
1340 if (cfg && cfg->ssl) {
1341 const struct ovsrec_ssl *ssl = cfg->ssl;
1343 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1344 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1348 /* Refresh system and interface stats if necessary. */
1349 if (time_msec() >= stats_timer) {
1351 struct ovsdb_idl_txn *txn;
1353 txn = ovsdb_idl_txn_create(idl);
1354 LIST_FOR_EACH (br, node, &all_bridges) {
1357 for (i = 0; i < br->n_ports; i++) {
1358 struct port *port = br->ports[i];
1361 for (j = 0; j < port->n_ifaces; j++) {
1362 struct iface *iface = port->ifaces[j];
1363 iface_refresh_stats(iface);
1364 iface_refresh_cfm_stats(iface);
1368 refresh_system_stats(cfg);
1369 ovsdb_idl_txn_commit(txn);
1370 ovsdb_idl_txn_destroy(txn); /* XXX */
1373 stats_timer = time_msec() + STATS_INTERVAL;
1381 struct iface *iface;
1383 LIST_FOR_EACH (br, node, &all_bridges) {
1384 ofproto_wait(br->ofproto);
1385 if (ofproto_has_primary_controller(br->ofproto)) {
1389 mac_learning_wait(br->ml);
1392 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1394 cfm_wait(iface->cfm);
1398 ovsdb_idl_wait(idl);
1399 poll_timer_wait_until(stats_timer);
1402 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1403 * configuration changes. */
1405 bridge_flush(struct bridge *br)
1407 COVERAGE_INC(bridge_flush);
1409 mac_learning_flush(br->ml);
1412 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1413 * such interface. */
1414 static struct iface *
1415 bridge_get_local_iface(struct bridge *br)
1419 for (i = 0; i < br->n_ports; i++) {
1420 struct port *port = br->ports[i];
1421 for (j = 0; j < port->n_ifaces; j++) {
1422 struct iface *iface = port->ifaces[j];
1423 if (iface->dp_ifidx == ODPP_LOCAL) {
1432 /* Bridge unixctl user interface functions. */
1434 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1435 const char *args, void *aux OVS_UNUSED)
1437 struct ds ds = DS_EMPTY_INITIALIZER;
1438 const struct bridge *br;
1439 const struct mac_entry *e;
1441 br = bridge_lookup(args);
1443 unixctl_command_reply(conn, 501, "no such bridge");
1447 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1448 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1449 if (e->port < 0 || e->port >= br->n_ports) {
1452 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1453 br->ports[e->port]->ifaces[0]->dp_ifidx,
1454 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1456 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1460 /* Bridge reconfiguration functions. */
1461 static struct bridge *
1462 bridge_create(const struct ovsrec_bridge *br_cfg)
1467 assert(!bridge_lookup(br_cfg->name));
1468 br = xzalloc(sizeof *br);
1470 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1476 dpif_flow_flush(br->dpif);
1478 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1481 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1483 dpif_delete(br->dpif);
1484 dpif_close(br->dpif);
1489 br->name = xstrdup(br_cfg->name);
1491 br->ml = mac_learning_create();
1492 eth_addr_nicira_random(br->default_ea);
1494 hmap_init(&br->ifaces);
1496 shash_init(&br->port_by_name);
1497 shash_init(&br->iface_by_name);
1501 list_push_back(&all_bridges, &br->node);
1503 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1509 bridge_destroy(struct bridge *br)
1514 while (br->n_ports > 0) {
1515 port_destroy(br->ports[br->n_ports - 1]);
1517 list_remove(&br->node);
1518 error = dpif_delete(br->dpif);
1519 if (error && error != ENOENT) {
1520 VLOG_ERR("failed to delete %s: %s",
1521 dpif_name(br->dpif), strerror(error));
1523 dpif_close(br->dpif);
1524 ofproto_destroy(br->ofproto);
1525 mac_learning_destroy(br->ml);
1526 hmap_destroy(&br->ifaces);
1527 shash_destroy(&br->port_by_name);
1528 shash_destroy(&br->iface_by_name);
1535 static struct bridge *
1536 bridge_lookup(const char *name)
1540 LIST_FOR_EACH (br, node, &all_bridges) {
1541 if (!strcmp(br->name, name)) {
1548 /* Handle requests for a listing of all flows known by the OpenFlow
1549 * stack, including those normally hidden. */
1551 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1552 const char *args, void *aux OVS_UNUSED)
1557 br = bridge_lookup(args);
1559 unixctl_command_reply(conn, 501, "Unknown bridge");
1564 ofproto_get_all_flows(br->ofproto, &results);
1566 unixctl_command_reply(conn, 200, ds_cstr(&results));
1567 ds_destroy(&results);
1570 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1571 * connections and reconnect. If BRIDGE is not specified, then all bridges
1572 * drop their controller connections and reconnect. */
1574 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1575 const char *args, void *aux OVS_UNUSED)
1578 if (args[0] != '\0') {
1579 br = bridge_lookup(args);
1581 unixctl_command_reply(conn, 501, "Unknown bridge");
1584 ofproto_reconnect_controllers(br->ofproto);
1586 LIST_FOR_EACH (br, node, &all_bridges) {
1587 ofproto_reconnect_controllers(br->ofproto);
1590 unixctl_command_reply(conn, 200, NULL);
1594 bridge_run_one(struct bridge *br)
1597 struct iface *iface;
1599 error = ofproto_run1(br->ofproto);
1604 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1607 error = ofproto_run2(br->ofproto, br->flush);
1610 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1611 struct ofpbuf *packet;
1617 packet = cfm_run(iface->cfm);
1619 iface_send_packet(iface, packet);
1620 ofpbuf_uninit(packet);
1629 bridge_get_controllers(const struct bridge *br,
1630 struct ovsrec_controller ***controllersp)
1632 struct ovsrec_controller **controllers;
1633 size_t n_controllers;
1635 controllers = br->cfg->controller;
1636 n_controllers = br->cfg->n_controller;
1638 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1644 *controllersp = controllers;
1646 return n_controllers;
1650 bridge_reconfigure_one(struct bridge *br)
1652 struct shash old_ports, new_ports;
1653 struct svec snoops, old_snoops;
1654 struct shash_node *node;
1655 enum ofproto_fail_mode fail_mode;
1658 /* Collect old ports. */
1659 shash_init(&old_ports);
1660 for (i = 0; i < br->n_ports; i++) {
1661 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1664 /* Collect new ports. */
1665 shash_init(&new_ports);
1666 for (i = 0; i < br->cfg->n_ports; i++) {
1667 const char *name = br->cfg->ports[i]->name;
1668 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1669 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1674 /* If we have a controller, then we need a local port. Complain if the
1675 * user didn't specify one.
1677 * XXX perhaps we should synthesize a port ourselves in this case. */
1678 if (bridge_get_controllers(br, NULL)) {
1679 char local_name[IF_NAMESIZE];
1682 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1683 local_name, sizeof local_name);
1684 if (!error && !shash_find(&new_ports, local_name)) {
1685 VLOG_WARN("bridge %s: controller specified but no local port "
1686 "(port named %s) defined",
1687 br->name, local_name);
1691 /* Get rid of deleted ports.
1692 * Get rid of deleted interfaces on ports that still exist. */
1693 SHASH_FOR_EACH (node, &old_ports) {
1694 struct port *port = node->data;
1695 const struct ovsrec_port *port_cfg;
1697 port_cfg = shash_find_data(&new_ports, node->name);
1701 port_del_ifaces(port, port_cfg);
1705 /* Create new ports.
1706 * Add new interfaces to existing ports.
1707 * Reconfigure existing ports. */
1708 SHASH_FOR_EACH (node, &new_ports) {
1709 struct port *port = shash_find_data(&old_ports, node->name);
1711 port = port_create(br, node->name);
1714 port_reconfigure(port, node->data);
1715 if (!port->n_ifaces) {
1716 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1717 br->name, port->name);
1721 shash_destroy(&old_ports);
1722 shash_destroy(&new_ports);
1724 /* Set the fail-mode */
1725 fail_mode = !br->cfg->fail_mode
1726 || !strcmp(br->cfg->fail_mode, "standalone")
1727 ? OFPROTO_FAIL_STANDALONE
1728 : OFPROTO_FAIL_SECURE;
1729 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1730 && !ofproto_has_primary_controller(br->ofproto)) {
1731 ofproto_flush_flows(br->ofproto);
1733 ofproto_set_fail_mode(br->ofproto, fail_mode);
1735 /* Delete all flows if we're switching from connected to standalone or vice
1736 * versa. (XXX Should we delete all flows if we are switching from one
1737 * controller to another?) */
1739 /* Configure OpenFlow controller connection snooping. */
1741 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1742 ovs_rundir(), br->name));
1743 svec_init(&old_snoops);
1744 ofproto_get_snoops(br->ofproto, &old_snoops);
1745 if (!svec_equal(&snoops, &old_snoops)) {
1746 ofproto_set_snoops(br->ofproto, &snoops);
1748 svec_destroy(&snoops);
1749 svec_destroy(&old_snoops);
1751 mirror_reconfigure(br);
1754 /* Initializes 'oc' appropriately as a management service controller for
1757 * The caller must free oc->target when it is no longer needed. */
1759 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1760 struct ofproto_controller *oc)
1762 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1763 oc->max_backoff = 0;
1764 oc->probe_interval = 60;
1765 oc->band = OFPROTO_OUT_OF_BAND;
1766 oc->accept_re = NULL;
1767 oc->update_resolv_conf = false;
1769 oc->burst_limit = 0;
1772 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1774 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1775 struct ofproto_controller *oc)
1777 oc->target = c->target;
1778 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1779 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1780 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1781 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1782 oc->accept_re = c->discover_accept_regex;
1783 oc->update_resolv_conf = c->discover_update_resolv_conf;
1784 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1785 oc->burst_limit = (c->controller_burst_limit
1786 ? *c->controller_burst_limit : 0);
1789 /* Configures the IP stack for 'br''s local interface properly according to the
1790 * configuration in 'c'. */
1792 bridge_configure_local_iface_netdev(struct bridge *br,
1793 struct ovsrec_controller *c)
1795 struct netdev *netdev;
1796 struct in_addr mask, gateway;
1798 struct iface *local_iface;
1801 /* Controller discovery does its own TCP/IP configuration later. */
1802 if (strcmp(c->target, "discover")) {
1806 /* If there's no local interface or no IP address, give up. */
1807 local_iface = bridge_get_local_iface(br);
1808 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1812 /* Bring up the local interface. */
1813 netdev = local_iface->netdev;
1814 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1816 /* Configure the IP address and netmask. */
1817 if (!c->local_netmask
1818 || !inet_aton(c->local_netmask, &mask)
1820 mask.s_addr = guess_netmask(ip.s_addr);
1822 if (!netdev_set_in4(netdev, ip, mask)) {
1823 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1824 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1827 /* Configure the default gateway. */
1828 if (c->local_gateway
1829 && inet_aton(c->local_gateway, &gateway)
1830 && gateway.s_addr) {
1831 if (!netdev_add_router(netdev, gateway)) {
1832 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1833 br->name, IP_ARGS(&gateway.s_addr));
1839 bridge_reconfigure_remotes(struct bridge *br,
1840 const struct sockaddr_in *managers,
1843 const char *disable_ib_str, *queue_id_str;
1844 bool disable_in_band = false;
1847 struct ovsrec_controller **controllers;
1848 size_t n_controllers;
1851 struct ofproto_controller *ocs;
1855 /* Check if we should disable in-band control on this bridge. */
1856 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1857 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1858 disable_in_band = true;
1861 /* Set OpenFlow queue ID for in-band control. */
1862 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1863 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1864 ofproto_set_in_band_queue(br->ofproto, queue_id);
1866 if (disable_in_band) {
1867 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1869 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1871 had_primary = ofproto_has_primary_controller(br->ofproto);
1873 n_controllers = bridge_get_controllers(br, &controllers);
1875 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1878 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1879 for (i = 0; i < n_controllers; i++) {
1880 struct ovsrec_controller *c = controllers[i];
1882 if (!strncmp(c->target, "punix:", 6)
1883 || !strncmp(c->target, "unix:", 5)) {
1884 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1886 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1887 * domain sockets and overwriting arbitrary local files. */
1888 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1889 "\"%s\" due to possibility for remote exploit",
1890 dpif_name(br->dpif), c->target);
1894 bridge_configure_local_iface_netdev(br, c);
1895 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1896 if (disable_in_band) {
1897 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1902 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1903 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1906 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1907 ofproto_flush_flows(br->ofproto);
1910 /* If there are no controllers and the bridge is in standalone
1911 * mode, set up a flow that matches every packet and directs
1912 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1913 * switch is in secure mode and we won't pass any traffic until
1914 * a controller has been defined and it tells us to do so. */
1916 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1917 union ofp_action action;
1918 struct cls_rule rule;
1920 memset(&action, 0, sizeof action);
1921 action.type = htons(OFPAT_OUTPUT);
1922 action.output.len = htons(sizeof action);
1923 action.output.port = htons(OFPP_NORMAL);
1924 cls_rule_init_catchall(&rule, 0);
1925 ofproto_add_flow(br->ofproto, &rule, &action, 1);
1930 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1935 for (i = 0; i < br->n_ports; i++) {
1936 struct port *port = br->ports[i];
1937 for (j = 0; j < port->n_ifaces; j++) {
1938 struct iface *iface = port->ifaces[j];
1939 shash_add_once(ifaces, iface->name, iface);
1941 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1942 shash_add_once(ifaces, port->name, NULL);
1947 /* For robustness, in case the administrator moves around datapath ports behind
1948 * our back, we re-check all the datapath port numbers here.
1950 * This function will set the 'dp_ifidx' members of interfaces that have
1951 * disappeared to -1, so only call this function from a context where those
1952 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1953 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1954 * datapath, which doesn't support UINT16_MAX+1 ports. */
1956 bridge_fetch_dp_ifaces(struct bridge *br)
1958 struct odp_port *dpif_ports;
1959 size_t n_dpif_ports;
1962 /* Reset all interface numbers. */
1963 for (i = 0; i < br->n_ports; i++) {
1964 struct port *port = br->ports[i];
1965 for (j = 0; j < port->n_ifaces; j++) {
1966 struct iface *iface = port->ifaces[j];
1967 iface->dp_ifidx = -1;
1970 hmap_clear(&br->ifaces);
1972 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1973 for (i = 0; i < n_dpif_ports; i++) {
1974 struct odp_port *p = &dpif_ports[i];
1975 struct iface *iface = iface_lookup(br, p->devname);
1977 if (iface->dp_ifidx >= 0) {
1978 VLOG_WARN("%s reported interface %s twice",
1979 dpif_name(br->dpif), p->devname);
1980 } else if (iface_from_dp_ifidx(br, p->port)) {
1981 VLOG_WARN("%s reported interface %"PRIu16" twice",
1982 dpif_name(br->dpif), p->port);
1984 iface->dp_ifidx = p->port;
1985 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1986 hash_int(iface->dp_ifidx, 0));
1989 iface_set_ofport(iface->cfg,
1990 (iface->dp_ifidx >= 0
1991 ? odp_port_to_ofp_port(iface->dp_ifidx)
1998 /* Bridge packet processing functions. */
2001 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
2003 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
2006 static struct bond_entry *
2007 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
2009 return &port->bond_hash[bond_hash(mac)];
2013 bond_choose_iface(const struct port *port)
2015 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2016 size_t i, best_down_slave = -1;
2017 long long next_delay_expiration = LLONG_MAX;
2019 for (i = 0; i < port->n_ifaces; i++) {
2020 struct iface *iface = port->ifaces[i];
2022 if (iface->enabled) {
2024 } else if (iface->delay_expires < next_delay_expiration) {
2025 best_down_slave = i;
2026 next_delay_expiration = iface->delay_expires;
2030 if (best_down_slave != -1) {
2031 struct iface *iface = port->ifaces[best_down_slave];
2033 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2034 "since no other interface is up", iface->name,
2035 iface->delay_expires - time_msec());
2036 bond_enable_slave(iface, true);
2039 return best_down_slave;
2043 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2044 uint16_t *dp_ifidx, tag_type *tags)
2046 struct iface *iface;
2048 assert(port->n_ifaces);
2049 if (port->n_ifaces == 1) {
2050 iface = port->ifaces[0];
2052 struct bond_entry *e = lookup_bond_entry(port, dl_src);
2053 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2054 || !port->ifaces[e->iface_idx]->enabled) {
2055 /* XXX select interface properly. The current interface selection
2056 * is only good for testing the rebalancing code. */
2057 e->iface_idx = bond_choose_iface(port);
2058 if (e->iface_idx < 0) {
2059 *tags |= port->no_ifaces_tag;
2062 e->iface_tag = tag_create_random();
2063 ((struct port *) port)->bond_compat_is_stale = true;
2065 *tags |= e->iface_tag;
2066 iface = port->ifaces[e->iface_idx];
2068 *dp_ifidx = iface->dp_ifidx;
2069 *tags |= iface->tag; /* Currently only used for bonding. */
2074 bond_link_status_update(struct iface *iface, bool carrier)
2076 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2077 struct port *port = iface->port;
2079 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2080 /* Nothing to do. */
2083 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
2084 iface->name, carrier ? "detected" : "dropped");
2085 if (carrier == iface->enabled) {
2086 iface->delay_expires = LLONG_MAX;
2087 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2088 iface->name, carrier ? "disabled" : "enabled");
2089 } else if (carrier && port->active_iface < 0) {
2090 bond_enable_slave(iface, true);
2091 if (port->updelay) {
2092 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2093 "other interface is up", iface->name, port->updelay);
2096 int delay = carrier ? port->updelay : port->downdelay;
2097 iface->delay_expires = time_msec() + delay;
2100 "interface %s: will be %s if it stays %s for %d ms",
2102 carrier ? "enabled" : "disabled",
2103 carrier ? "up" : "down",
2110 bond_choose_active_iface(struct port *port)
2112 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2114 port->active_iface = bond_choose_iface(port);
2115 port->active_iface_tag = tag_create_random();
2116 if (port->active_iface >= 0) {
2117 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2118 port->name, port->ifaces[port->active_iface]->name);
2120 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2126 bond_enable_slave(struct iface *iface, bool enable)
2128 struct port *port = iface->port;
2129 struct bridge *br = port->bridge;
2131 /* This acts as a recursion check. If the act of disabling a slave
2132 * causes a different slave to be enabled, the flag will allow us to
2133 * skip redundant work when we reenter this function. It must be
2134 * cleared on exit to keep things safe with multiple bonds. */
2135 static bool moving_active_iface = false;
2137 iface->delay_expires = LLONG_MAX;
2138 if (enable == iface->enabled) {
2142 iface->enabled = enable;
2143 if (!iface->enabled) {
2144 VLOG_WARN("interface %s: disabled", iface->name);
2145 ofproto_revalidate(br->ofproto, iface->tag);
2146 if (iface->port_ifidx == port->active_iface) {
2147 ofproto_revalidate(br->ofproto,
2148 port->active_iface_tag);
2150 /* Disabling a slave can lead to another slave being immediately
2151 * enabled if there will be no active slaves but one is waiting
2152 * on an updelay. In this case we do not need to run most of the
2153 * code for the newly enabled slave since there was no period
2154 * without an active slave and it is redundant with the disabling
2156 moving_active_iface = true;
2157 bond_choose_active_iface(port);
2159 bond_send_learning_packets(port);
2161 VLOG_WARN("interface %s: enabled", iface->name);
2162 if (port->active_iface < 0 && !moving_active_iface) {
2163 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2164 bond_choose_active_iface(port);
2165 bond_send_learning_packets(port);
2167 iface->tag = tag_create_random();
2170 moving_active_iface = false;
2171 port->bond_compat_is_stale = true;
2174 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2175 * bond interface. */
2177 bond_update_fake_iface_stats(struct port *port)
2179 struct netdev_stats bond_stats;
2180 struct netdev *bond_dev;
2183 memset(&bond_stats, 0, sizeof bond_stats);
2185 for (i = 0; i < port->n_ifaces; i++) {
2186 struct netdev_stats slave_stats;
2188 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2189 /* XXX: We swap the stats here because they are swapped back when
2190 * reported by the internal device. The reason for this is
2191 * internal devices normally represent packets going into the system
2192 * but when used as fake bond device they represent packets leaving
2193 * the system. We really should do this in the internal device
2194 * itself because changing it here reverses the counts from the
2195 * perspective of the switch. However, the internal device doesn't
2196 * know what type of device it represents so we have to do it here
2198 bond_stats.tx_packets += slave_stats.rx_packets;
2199 bond_stats.tx_bytes += slave_stats.rx_bytes;
2200 bond_stats.rx_packets += slave_stats.tx_packets;
2201 bond_stats.rx_bytes += slave_stats.tx_bytes;
2205 if (!netdev_open_default(port->name, &bond_dev)) {
2206 netdev_set_stats(bond_dev, &bond_stats);
2207 netdev_close(bond_dev);
2212 bond_run(struct bridge *br)
2216 for (i = 0; i < br->n_ports; i++) {
2217 struct port *port = br->ports[i];
2219 if (port->n_ifaces >= 2) {
2222 /* Track carrier going up and down on interfaces. */
2223 while (!netdev_monitor_poll(port->monitor, &devname)) {
2224 struct iface *iface;
2226 iface = port_lookup_iface(port, devname);
2228 bool carrier = netdev_get_carrier(iface->netdev);
2230 bond_link_status_update(iface, carrier);
2231 port_update_bond_compat(port);
2236 for (j = 0; j < port->n_ifaces; j++) {
2237 struct iface *iface = port->ifaces[j];
2238 if (time_msec() >= iface->delay_expires) {
2239 bond_enable_slave(iface, !iface->enabled);
2243 if (port->bond_fake_iface
2244 && time_msec() >= port->bond_next_fake_iface_update) {
2245 bond_update_fake_iface_stats(port);
2246 port->bond_next_fake_iface_update = time_msec() + 1000;
2250 if (port->bond_compat_is_stale) {
2251 port->bond_compat_is_stale = false;
2252 port_update_bond_compat(port);
2258 bond_wait(struct bridge *br)
2262 for (i = 0; i < br->n_ports; i++) {
2263 struct port *port = br->ports[i];
2264 if (port->n_ifaces < 2) {
2267 netdev_monitor_poll_wait(port->monitor);
2268 for (j = 0; j < port->n_ifaces; j++) {
2269 struct iface *iface = port->ifaces[j];
2270 if (iface->delay_expires != LLONG_MAX) {
2271 poll_timer_wait_until(iface->delay_expires);
2274 if (port->bond_fake_iface) {
2275 poll_timer_wait_until(port->bond_next_fake_iface_update);
2281 set_dst(struct dst *p, const struct flow *flow,
2282 const struct port *in_port, const struct port *out_port,
2285 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2286 : in_port->vlan >= 0 ? in_port->vlan
2287 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2288 : vlan_tci_to_vid(flow->vlan_tci));
2289 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2293 swap_dst(struct dst *p, struct dst *q)
2295 struct dst tmp = *p;
2300 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2301 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2302 * that we push to the datapath. We could in fact fully sort the array by
2303 * vlan, but in most cases there are at most two different vlan tags so that's
2304 * possibly overkill.) */
2306 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2308 struct dst *first = dsts;
2309 struct dst *last = dsts + n_dsts;
2311 while (first != last) {
2313 * - All dsts < first have vlan == 'vlan'.
2314 * - All dsts >= last have vlan != 'vlan'.
2315 * - first < last. */
2316 while (first->vlan == vlan) {
2317 if (++first == last) {
2322 /* Same invariants, plus one additional:
2323 * - first->vlan != vlan.
2325 while (last[-1].vlan != vlan) {
2326 if (--last == first) {
2331 /* Same invariants, plus one additional:
2332 * - last[-1].vlan == vlan.*/
2333 swap_dst(first++, --last);
2338 mirror_mask_ffs(mirror_mask_t mask)
2340 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2345 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2346 const struct dst *test)
2349 for (i = 0; i < n_dsts; i++) {
2350 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2358 port_trunks_vlan(const struct port *port, uint16_t vlan)
2360 return (port->vlan < 0
2361 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2365 port_includes_vlan(const struct port *port, uint16_t vlan)
2367 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2371 port_is_floodable(const struct port *port)
2375 for (i = 0; i < port->n_ifaces; i++) {
2376 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2377 port->ifaces[i]->dp_ifidx)) {
2385 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2386 const struct port *in_port, const struct port *out_port,
2387 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2389 mirror_mask_t mirrors = in_port->src_mirrors;
2391 struct dst *dst = dsts;
2394 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2395 if (flow_vlan == 0) {
2396 flow_vlan = OFP_VLAN_NONE;
2399 if (out_port == FLOOD_PORT) {
2400 /* XXX use ODP_FLOOD if no vlans or bonding. */
2401 /* XXX even better, define each VLAN as a datapath port group */
2402 for (i = 0; i < br->n_ports; i++) {
2403 struct port *port = br->ports[i];
2405 && port_is_floodable(port)
2406 && port_includes_vlan(port, vlan)
2407 && !port->is_mirror_output_port
2408 && set_dst(dst, flow, in_port, port, tags)) {
2409 mirrors |= port->dst_mirrors;
2413 *nf_output_iface = NF_OUT_FLOOD;
2414 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2415 *nf_output_iface = dst->dp_ifidx;
2416 mirrors |= out_port->dst_mirrors;
2421 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2422 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2424 if (set_dst(dst, flow, in_port, m->out_port, tags)
2425 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2429 for (i = 0; i < br->n_ports; i++) {
2430 struct port *port = br->ports[i];
2431 if (port_includes_vlan(port, m->out_vlan)
2432 && set_dst(dst, flow, in_port, port, tags))
2435 if (port->vlan < 0) {
2436 dst->vlan = m->out_vlan;
2438 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2442 /* Use the vlan tag on the original flow instead of
2443 * the one passed in the vlan parameter. This ensures
2444 * that we compare the vlan from before any implicit
2445 * tagging tags place. This is necessary because
2446 * dst->vlan is the final vlan, after removing implicit
2448 if (port == in_port && dst->vlan == flow_vlan) {
2449 /* Don't send out input port on same VLAN. */
2457 mirrors &= mirrors - 1;
2460 partition_dsts(dsts, dst - dsts, flow_vlan);
2464 static void OVS_UNUSED
2465 print_dsts(const struct dst *dsts, size_t n)
2467 for (; n--; dsts++) {
2468 printf(">p%"PRIu16, dsts->dp_ifidx);
2469 if (dsts->vlan != OFP_VLAN_NONE) {
2470 printf("v%"PRIu16, dsts->vlan);
2476 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2477 const struct port *in_port, const struct port *out_port,
2478 tag_type *tags, struct odp_actions *actions,
2479 uint16_t *nf_output_iface)
2481 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2483 const struct dst *p;
2486 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2489 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2490 if (cur_vlan == 0) {
2491 cur_vlan = OFP_VLAN_NONE;
2493 for (p = dsts; p < &dsts[n_dsts]; p++) {
2494 union odp_action *a;
2495 if (p->vlan != cur_vlan) {
2496 if (p->vlan == OFP_VLAN_NONE) {
2497 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2499 a = odp_actions_add(actions, ODPAT_SET_DL_TCI);
2500 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2501 a->dl_tci.tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2505 a = odp_actions_add(actions, ODPAT_OUTPUT);
2506 a->output.port = p->dp_ifidx;
2510 /* Returns the effective vlan of a packet, taking into account both the
2511 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2512 * the packet is untagged and -1 indicates it has an invalid header and
2513 * should be dropped. */
2514 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2515 struct port *in_port, bool have_packet)
2517 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2518 if (in_port->vlan >= 0) {
2520 /* XXX support double tagging? */
2522 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2523 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2524 "packet received on port %s configured with "
2525 "implicit VLAN %"PRIu16,
2526 br->name, vlan, in_port->name, in_port->vlan);
2530 vlan = in_port->vlan;
2532 if (!port_includes_vlan(in_port, vlan)) {
2534 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2535 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2536 "packet received on port %s not configured for "
2538 br->name, vlan, in_port->name, vlan);
2547 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2548 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2549 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2551 is_gratuitous_arp(const struct flow *flow)
2553 return (flow->dl_type == htons(ETH_TYPE_ARP)
2554 && eth_addr_is_broadcast(flow->dl_dst)
2555 && (flow->nw_proto == ARP_OP_REPLY
2556 || (flow->nw_proto == ARP_OP_REQUEST
2557 && flow->nw_src == flow->nw_dst)));
2561 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2562 struct port *in_port)
2564 enum grat_arp_lock_type lock_type;
2567 /* We don't want to learn from gratuitous ARP packets that are reflected
2568 * back over bond slaves so we lock the learning table. */
2569 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2570 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2571 GRAT_ARP_LOCK_CHECK;
2573 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2576 /* The log messages here could actually be useful in debugging,
2577 * so keep the rate limit relatively high. */
2578 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2580 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2581 "on port %s in VLAN %d",
2582 br->name, ETH_ADDR_ARGS(flow->dl_src),
2583 in_port->name, vlan);
2584 ofproto_revalidate(br->ofproto, rev_tag);
2588 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2589 * dropped. Returns true if they may be forwarded, false if they should be
2592 * If 'have_packet' is true, it indicates that the caller is processing a
2593 * received packet. If 'have_packet' is false, then the caller is just
2594 * revalidating an existing flow because configuration has changed. Either
2595 * way, 'have_packet' only affects logging (there is no point in logging errors
2596 * during revalidation).
2598 * Sets '*in_portp' to the input port. This will be a null pointer if
2599 * flow->in_port does not designate a known input port (in which case
2600 * is_admissible() returns false).
2602 * When returning true, sets '*vlanp' to the effective VLAN of the input
2603 * packet, as returned by flow_get_vlan().
2605 * May also add tags to '*tags', although the current implementation only does
2606 * so in one special case.
2609 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2610 tag_type *tags, int *vlanp, struct port **in_portp)
2612 struct iface *in_iface;
2613 struct port *in_port;
2616 /* Find the interface and port structure for the received packet. */
2617 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2619 /* No interface? Something fishy... */
2621 /* Odd. A few possible reasons here:
2623 * - We deleted an interface but there are still a few packets
2624 * queued up from it.
2626 * - Someone externally added an interface (e.g. with "ovs-dpctl
2627 * add-if") that we don't know about.
2629 * - Packet arrived on the local port but the local port is not
2630 * one of our bridge ports.
2632 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2634 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2635 "interface %"PRIu16, br->name, flow->in_port);
2641 *in_portp = in_port = in_iface->port;
2642 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2647 /* Drop frames for reserved multicast addresses. */
2648 if (eth_addr_is_reserved(flow->dl_dst)) {
2652 /* Drop frames on ports reserved for mirroring. */
2653 if (in_port->is_mirror_output_port) {
2655 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2656 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2657 "%s, which is reserved exclusively for mirroring",
2658 br->name, in_port->name);
2663 /* Packets received on bonds need special attention to avoid duplicates. */
2664 if (in_port->n_ifaces > 1) {
2666 bool is_grat_arp_locked;
2668 if (eth_addr_is_multicast(flow->dl_dst)) {
2669 *tags |= in_port->active_iface_tag;
2670 if (in_port->active_iface != in_iface->port_ifidx) {
2671 /* Drop all multicast packets on inactive slaves. */
2676 /* Drop all packets for which we have learned a different input
2677 * port, because we probably sent the packet on one slave and got
2678 * it back on the other. Gratuitous ARP packets are an exception
2679 * to this rule: the host has moved to another switch. The exception
2680 * to the exception is if we locked the learning table to avoid
2681 * reflections on bond slaves. If this is the case, just drop the
2683 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2684 &is_grat_arp_locked);
2685 if (src_idx != -1 && src_idx != in_port->port_idx &&
2686 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2694 /* If the composed actions may be applied to any packet in the given 'flow',
2695 * returns true. Otherwise, the actions should only be applied to 'packet', or
2696 * not at all, if 'packet' was NULL. */
2698 process_flow(struct bridge *br, const struct flow *flow,
2699 const struct ofpbuf *packet, struct odp_actions *actions,
2700 tag_type *tags, uint16_t *nf_output_iface)
2702 struct port *in_port;
2703 struct port *out_port;
2707 /* Check whether we should drop packets in this flow. */
2708 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2713 /* Learn source MAC (but don't try to learn from revalidation). */
2715 update_learning_table(br, flow, vlan, in_port);
2718 /* Determine output port. */
2719 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2721 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2722 out_port = br->ports[out_port_idx];
2723 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2724 /* If we are revalidating but don't have a learning entry then
2725 * eject the flow. Installing a flow that floods packets opens
2726 * up a window of time where we could learn from a packet reflected
2727 * on a bond and blackhole packets before the learning table is
2728 * updated to reflect the correct port. */
2731 out_port = FLOOD_PORT;
2734 /* Don't send packets out their input ports. */
2735 if (in_port == out_port) {
2741 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2749 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2750 struct odp_actions *actions, tag_type *tags,
2751 uint16_t *nf_output_iface, void *br_)
2753 struct iface *iface;
2754 struct bridge *br = br_;
2756 COVERAGE_INC(bridge_process_flow);
2758 iface = iface_from_dp_ifidx(br, flow->in_port);
2760 if (cfm_should_process_flow(flow)) {
2761 if (packet && iface->cfm) {
2762 cfm_process_heartbeat(iface->cfm, packet);
2767 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2771 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2772 const union odp_action *actions,
2773 size_t n_actions, unsigned long long int n_bytes,
2776 struct bridge *br = br_;
2777 const union odp_action *a;
2778 struct port *in_port;
2782 /* Feed information from the active flows back into the learning table to
2783 * ensure that table is always in sync with what is actually flowing
2784 * through the datapath.
2786 * We test that 'tags' is nonzero to ensure that only flows that include an
2787 * OFPP_NORMAL action are used for learning. This works because
2788 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2789 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2790 update_learning_table(br, flow, vlan, in_port);
2793 /* Account for bond slave utilization. */
2794 if (!br->has_bonded_ports) {
2797 for (a = actions; a < &actions[n_actions]; a++) {
2798 if (a->type == ODPAT_OUTPUT) {
2799 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2800 if (out_port && out_port->n_ifaces >= 2) {
2801 struct bond_entry *e = lookup_bond_entry(out_port,
2803 e->tx_bytes += n_bytes;
2810 bridge_account_checkpoint_ofhook_cb(void *br_)
2812 struct bridge *br = br_;
2816 if (!br->has_bonded_ports) {
2821 for (i = 0; i < br->n_ports; i++) {
2822 struct port *port = br->ports[i];
2823 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2824 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2825 bond_rebalance_port(port);
2830 static struct ofhooks bridge_ofhooks = {
2831 bridge_normal_ofhook_cb,
2832 bridge_account_flow_ofhook_cb,
2833 bridge_account_checkpoint_ofhook_cb,
2836 /* Bonding functions. */
2838 /* Statistics for a single interface on a bonded port, used for load-based
2839 * bond rebalancing. */
2840 struct slave_balance {
2841 struct iface *iface; /* The interface. */
2842 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2844 /* All the "bond_entry"s that are assigned to this interface, in order of
2845 * increasing tx_bytes. */
2846 struct bond_entry **hashes;
2850 /* Sorts pointers to pointers to bond_entries in ascending order by the
2851 * interface to which they are assigned, and within a single interface in
2852 * ascending order of bytes transmitted. */
2854 compare_bond_entries(const void *a_, const void *b_)
2856 const struct bond_entry *const *ap = a_;
2857 const struct bond_entry *const *bp = b_;
2858 const struct bond_entry *a = *ap;
2859 const struct bond_entry *b = *bp;
2860 if (a->iface_idx != b->iface_idx) {
2861 return a->iface_idx > b->iface_idx ? 1 : -1;
2862 } else if (a->tx_bytes != b->tx_bytes) {
2863 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2869 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2870 * *descending* order by number of bytes transmitted. */
2872 compare_slave_balance(const void *a_, const void *b_)
2874 const struct slave_balance *a = a_;
2875 const struct slave_balance *b = b_;
2876 if (a->iface->enabled != b->iface->enabled) {
2877 return a->iface->enabled ? -1 : 1;
2878 } else if (a->tx_bytes != b->tx_bytes) {
2879 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2886 swap_bals(struct slave_balance *a, struct slave_balance *b)
2888 struct slave_balance tmp = *a;
2893 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2894 * given that 'p' (and only 'p') might be in the wrong location.
2896 * This function invalidates 'p', since it might now be in a different memory
2899 resort_bals(struct slave_balance *p,
2900 struct slave_balance bals[], size_t n_bals)
2903 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2904 swap_bals(p, p - 1);
2906 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2907 swap_bals(p, p + 1);
2913 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2915 if (VLOG_IS_DBG_ENABLED()) {
2916 struct ds ds = DS_EMPTY_INITIALIZER;
2917 const struct slave_balance *b;
2919 for (b = bals; b < bals + n_bals; b++) {
2923 ds_put_char(&ds, ',');
2925 ds_put_format(&ds, " %s %"PRIu64"kB",
2926 b->iface->name, b->tx_bytes / 1024);
2928 if (!b->iface->enabled) {
2929 ds_put_cstr(&ds, " (disabled)");
2931 if (b->n_hashes > 0) {
2932 ds_put_cstr(&ds, " (");
2933 for (i = 0; i < b->n_hashes; i++) {
2934 const struct bond_entry *e = b->hashes[i];
2936 ds_put_cstr(&ds, " + ");
2938 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2939 e - port->bond_hash, e->tx_bytes / 1024);
2941 ds_put_cstr(&ds, ")");
2944 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2949 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2951 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2954 struct bond_entry *hash = from->hashes[hash_idx];
2955 struct port *port = from->iface->port;
2956 uint64_t delta = hash->tx_bytes;
2958 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2959 "from %s to %s (now carrying %"PRIu64"kB and "
2960 "%"PRIu64"kB load, respectively)",
2961 port->name, delta / 1024, hash - port->bond_hash,
2962 from->iface->name, to->iface->name,
2963 (from->tx_bytes - delta) / 1024,
2964 (to->tx_bytes + delta) / 1024);
2966 /* Delete element from from->hashes.
2968 * We don't bother to add the element to to->hashes because not only would
2969 * it require more work, the only purpose it would be to allow that hash to
2970 * be migrated to another slave in this rebalancing run, and there is no
2971 * point in doing that. */
2972 if (hash_idx == 0) {
2975 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2976 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2980 /* Shift load away from 'from' to 'to'. */
2981 from->tx_bytes -= delta;
2982 to->tx_bytes += delta;
2984 /* Arrange for flows to be revalidated. */
2985 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2986 hash->iface_idx = to->iface->port_ifidx;
2987 hash->iface_tag = tag_create_random();
2991 bond_rebalance_port(struct port *port)
2993 struct slave_balance bals[DP_MAX_PORTS];
2995 struct bond_entry *hashes[BOND_MASK + 1];
2996 struct slave_balance *b, *from, *to;
2997 struct bond_entry *e;
3000 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3001 * descending order of tx_bytes, so that bals[0] represents the most
3002 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3005 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3006 * array for each slave_balance structure, we sort our local array of
3007 * hashes in order by slave, so that all of the hashes for a given slave
3008 * become contiguous in memory, and then we point each 'hashes' members of
3009 * a slave_balance structure to the start of a contiguous group. */
3010 n_bals = port->n_ifaces;
3011 for (b = bals; b < &bals[n_bals]; b++) {
3012 b->iface = port->ifaces[b - bals];
3017 for (i = 0; i <= BOND_MASK; i++) {
3018 hashes[i] = &port->bond_hash[i];
3020 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3021 for (i = 0; i <= BOND_MASK; i++) {
3023 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3024 b = &bals[e->iface_idx];
3025 b->tx_bytes += e->tx_bytes;
3027 b->hashes = &hashes[i];
3032 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3033 log_bals(bals, n_bals, port);
3035 /* Discard slaves that aren't enabled (which were sorted to the back of the
3036 * array earlier). */
3037 while (!bals[n_bals - 1].iface->enabled) {
3044 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3045 to = &bals[n_bals - 1];
3046 for (from = bals; from < to; ) {
3047 uint64_t overload = from->tx_bytes - to->tx_bytes;
3048 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3049 /* The extra load on 'from' (and all less-loaded slaves), compared
3050 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3051 * it is less than ~1Mbps. No point in rebalancing. */
3053 } else if (from->n_hashes == 1) {
3054 /* 'from' only carries a single MAC hash, so we can't shift any
3055 * load away from it, even though we want to. */
3058 /* 'from' is carrying significantly more load than 'to', and that
3059 * load is split across at least two different hashes. Pick a hash
3060 * to migrate to 'to' (the least-loaded slave), given that doing so
3061 * must decrease the ratio of the load on the two slaves by at
3064 * The sort order we use means that we prefer to shift away the
3065 * smallest hashes instead of the biggest ones. There is little
3066 * reason behind this decision; we could use the opposite sort
3067 * order to shift away big hashes ahead of small ones. */
3070 for (i = 0; i < from->n_hashes; i++) {
3071 double old_ratio, new_ratio;
3072 uint64_t delta = from->hashes[i]->tx_bytes;
3074 if (delta == 0 || from->tx_bytes - delta == 0) {
3075 /* Pointless move. */
3079 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3081 if (to->tx_bytes == 0) {
3082 /* Nothing on the new slave, move it. */
3086 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3087 new_ratio = (double)(from->tx_bytes - delta) /
3088 (to->tx_bytes + delta);
3090 if (new_ratio == 0) {
3091 /* Should already be covered but check to prevent division
3096 if (new_ratio < 1) {
3097 new_ratio = 1 / new_ratio;
3100 if (old_ratio - new_ratio > 0.1) {
3101 /* Would decrease the ratio, move it. */
3105 if (i < from->n_hashes) {
3106 bond_shift_load(from, to, i);
3107 port->bond_compat_is_stale = true;
3109 /* If the result of the migration changed the relative order of
3110 * 'from' and 'to' swap them back to maintain invariants. */
3111 if (order_swapped) {
3112 swap_bals(from, to);
3115 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3116 * point to different slave_balance structures. It is only
3117 * valid to do these two operations in a row at all because we
3118 * know that 'from' will not move past 'to' and vice versa. */
3119 resort_bals(from, bals, n_bals);
3120 resort_bals(to, bals, n_bals);
3127 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3128 * historical data to decay to <1% in 7 rebalancing runs. */
3129 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3135 bond_send_learning_packets(struct port *port)
3137 struct bridge *br = port->bridge;
3138 struct mac_entry *e;
3139 struct ofpbuf packet;
3140 int error, n_packets, n_errors;
3142 if (!port->n_ifaces || port->active_iface < 0) {
3146 ofpbuf_init(&packet, 128);
3147 error = n_packets = n_errors = 0;
3148 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3149 union ofp_action actions[2], *a;
3155 if (e->port == port->port_idx
3156 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
3160 /* Compose actions. */
3161 memset(actions, 0, sizeof actions);
3164 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3165 a->vlan_vid.len = htons(sizeof *a);
3166 a->vlan_vid.vlan_vid = htons(e->vlan);
3169 a->output.type = htons(OFPAT_OUTPUT);
3170 a->output.len = htons(sizeof *a);
3171 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3176 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3178 flow_extract(&packet, 0, ODPP_NONE, &flow);
3179 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3186 ofpbuf_uninit(&packet);
3189 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3190 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3191 "packets, last error was: %s",
3192 port->name, n_errors, n_packets, strerror(error));
3194 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3195 port->name, n_packets);
3199 /* Bonding unixctl user interface functions. */
3202 bond_unixctl_list(struct unixctl_conn *conn,
3203 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3205 struct ds ds = DS_EMPTY_INITIALIZER;
3206 const struct bridge *br;
3208 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
3210 LIST_FOR_EACH (br, node, &all_bridges) {
3213 for (i = 0; i < br->n_ports; i++) {
3214 const struct port *port = br->ports[i];
3215 if (port->n_ifaces > 1) {
3218 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3219 for (j = 0; j < port->n_ifaces; j++) {
3220 const struct iface *iface = port->ifaces[j];
3222 ds_put_cstr(&ds, ", ");
3224 ds_put_cstr(&ds, iface->name);
3226 ds_put_char(&ds, '\n');
3230 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3234 static struct port *
3235 bond_find(const char *name)
3237 const struct bridge *br;
3239 LIST_FOR_EACH (br, node, &all_bridges) {
3242 for (i = 0; i < br->n_ports; i++) {
3243 struct port *port = br->ports[i];
3244 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3253 bond_unixctl_show(struct unixctl_conn *conn,
3254 const char *args, void *aux OVS_UNUSED)
3256 struct ds ds = DS_EMPTY_INITIALIZER;
3257 const struct port *port;
3260 port = bond_find(args);
3262 unixctl_command_reply(conn, 501, "no such bond");
3266 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3267 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3268 ds_put_format(&ds, "next rebalance: %lld ms\n",
3269 port->bond_next_rebalance - time_msec());
3270 for (j = 0; j < port->n_ifaces; j++) {
3271 const struct iface *iface = port->ifaces[j];
3272 struct bond_entry *be;
3275 ds_put_format(&ds, "slave %s: %s\n",
3276 iface->name, iface->enabled ? "enabled" : "disabled");
3277 if (j == port->active_iface) {
3278 ds_put_cstr(&ds, "\tactive slave\n");
3280 if (iface->delay_expires != LLONG_MAX) {
3281 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3282 iface->enabled ? "downdelay" : "updelay",
3283 iface->delay_expires - time_msec());
3287 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3288 int hash = be - port->bond_hash;
3289 struct mac_entry *me;
3291 if (be->iface_idx != j) {
3295 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3296 hash, be->tx_bytes / 1024);
3299 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3302 if (bond_hash(me->mac) == hash
3303 && me->port != port->port_idx
3304 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3305 && dp_ifidx == iface->dp_ifidx)
3307 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3308 ETH_ADDR_ARGS(me->mac));
3313 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3318 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3319 void *aux OVS_UNUSED)
3321 char *args = (char *) args_;
3322 char *save_ptr = NULL;
3323 char *bond_s, *hash_s, *slave_s;
3324 uint8_t mac[ETH_ADDR_LEN];
3326 struct iface *iface;
3327 struct bond_entry *entry;
3330 bond_s = strtok_r(args, " ", &save_ptr);
3331 hash_s = strtok_r(NULL, " ", &save_ptr);
3332 slave_s = strtok_r(NULL, " ", &save_ptr);
3334 unixctl_command_reply(conn, 501,
3335 "usage: bond/migrate BOND HASH SLAVE");
3339 port = bond_find(bond_s);
3341 unixctl_command_reply(conn, 501, "no such bond");
3345 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3346 == ETH_ADDR_SCAN_COUNT) {
3347 hash = bond_hash(mac);
3348 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3349 hash = atoi(hash_s) & BOND_MASK;
3351 unixctl_command_reply(conn, 501, "bad hash");
3355 iface = port_lookup_iface(port, slave_s);
3357 unixctl_command_reply(conn, 501, "no such slave");
3361 if (!iface->enabled) {
3362 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3366 entry = &port->bond_hash[hash];
3367 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3368 entry->iface_idx = iface->port_ifidx;
3369 entry->iface_tag = tag_create_random();
3370 port->bond_compat_is_stale = true;
3371 unixctl_command_reply(conn, 200, "migrated");
3375 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3376 void *aux OVS_UNUSED)
3378 char *args = (char *) args_;
3379 char *save_ptr = NULL;
3380 char *bond_s, *slave_s;
3382 struct iface *iface;
3384 bond_s = strtok_r(args, " ", &save_ptr);
3385 slave_s = strtok_r(NULL, " ", &save_ptr);
3387 unixctl_command_reply(conn, 501,
3388 "usage: bond/set-active-slave BOND SLAVE");
3392 port = bond_find(bond_s);
3394 unixctl_command_reply(conn, 501, "no such bond");
3398 iface = port_lookup_iface(port, slave_s);
3400 unixctl_command_reply(conn, 501, "no such slave");
3404 if (!iface->enabled) {
3405 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3409 if (port->active_iface != iface->port_ifidx) {
3410 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3411 port->active_iface = iface->port_ifidx;
3412 port->active_iface_tag = tag_create_random();
3413 VLOG_INFO("port %s: active interface is now %s",
3414 port->name, iface->name);
3415 bond_send_learning_packets(port);
3416 unixctl_command_reply(conn, 200, "done");
3418 unixctl_command_reply(conn, 200, "no change");
3423 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3425 char *args = (char *) args_;
3426 char *save_ptr = NULL;
3427 char *bond_s, *slave_s;
3429 struct iface *iface;
3431 bond_s = strtok_r(args, " ", &save_ptr);
3432 slave_s = strtok_r(NULL, " ", &save_ptr);
3434 unixctl_command_reply(conn, 501,
3435 "usage: bond/enable/disable-slave BOND SLAVE");
3439 port = bond_find(bond_s);
3441 unixctl_command_reply(conn, 501, "no such bond");
3445 iface = port_lookup_iface(port, slave_s);
3447 unixctl_command_reply(conn, 501, "no such slave");
3451 bond_enable_slave(iface, enable);
3452 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3456 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3457 void *aux OVS_UNUSED)
3459 enable_slave(conn, args, true);
3463 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3464 void *aux OVS_UNUSED)
3466 enable_slave(conn, args, false);
3470 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3471 void *aux OVS_UNUSED)
3473 uint8_t mac[ETH_ADDR_LEN];
3477 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3478 == ETH_ADDR_SCAN_COUNT) {
3479 hash = bond_hash(mac);
3481 hash_cstr = xasprintf("%u", hash);
3482 unixctl_command_reply(conn, 200, hash_cstr);
3485 unixctl_command_reply(conn, 501, "invalid mac");
3492 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3493 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3494 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3495 unixctl_command_register("bond/set-active-slave",
3496 bond_unixctl_set_active_slave, NULL);
3497 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3499 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3501 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3504 /* Port functions. */
3506 static struct port *
3507 port_create(struct bridge *br, const char *name)
3511 port = xzalloc(sizeof *port);
3513 port->port_idx = br->n_ports;
3515 port->trunks = NULL;
3516 port->name = xstrdup(name);
3517 port->active_iface = -1;
3519 if (br->n_ports >= br->allocated_ports) {
3520 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3523 br->ports[br->n_ports++] = port;
3524 shash_add_assert(&br->port_by_name, port->name, port);
3526 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3533 get_port_other_config(const struct ovsrec_port *port, const char *key,
3534 const char *default_value)
3538 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3540 return value ? value : default_value;
3544 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3546 struct shash new_ifaces;
3549 /* Collect list of new interfaces. */
3550 shash_init(&new_ifaces);
3551 for (i = 0; i < cfg->n_interfaces; i++) {
3552 const char *name = cfg->interfaces[i]->name;
3553 shash_add_once(&new_ifaces, name, NULL);
3556 /* Get rid of deleted interfaces. */
3557 for (i = 0; i < port->n_ifaces; ) {
3558 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3559 iface_destroy(port->ifaces[i]);
3565 shash_destroy(&new_ifaces);
3569 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3571 struct shash new_ifaces;
3572 long long int next_rebalance;
3573 unsigned long *trunks;
3579 /* Update settings. */
3580 port->updelay = cfg->bond_updelay;
3581 if (port->updelay < 0) {
3584 port->downdelay = cfg->bond_downdelay;
3585 if (port->downdelay < 0) {
3586 port->downdelay = 0;
3588 port->bond_rebalance_interval = atoi(
3589 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3590 if (port->bond_rebalance_interval < 1000) {
3591 port->bond_rebalance_interval = 1000;
3593 next_rebalance = time_msec() + port->bond_rebalance_interval;
3594 if (port->bond_next_rebalance > next_rebalance) {
3595 port->bond_next_rebalance = next_rebalance;
3598 /* Add new interfaces and update 'cfg' member of existing ones. */
3599 shash_init(&new_ifaces);
3600 for (i = 0; i < cfg->n_interfaces; i++) {
3601 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3602 struct iface *iface;
3604 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3605 VLOG_WARN("port %s: %s specified twice as port interface",
3606 port->name, if_cfg->name);
3607 iface_set_ofport(if_cfg, -1);
3611 iface = iface_lookup(port->bridge, if_cfg->name);
3613 if (iface->port != port) {
3614 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3616 port->bridge->name, if_cfg->name, iface->port->name);
3619 iface->cfg = if_cfg;
3621 iface = iface_create(port, if_cfg);
3624 /* Determine interface type. The local port always has type
3625 * "internal". Other ports take their type from the database and
3626 * default to "system" if none is specified. */
3627 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3628 : if_cfg->type[0] ? if_cfg->type
3631 shash_destroy(&new_ifaces);
3636 if (port->n_ifaces < 2) {
3638 if (vlan >= 0 && vlan <= 4095) {
3639 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3644 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3645 * they even work as-is. But they have not been tested. */
3646 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3650 if (port->vlan != vlan) {
3652 bridge_flush(port->bridge);
3655 /* Get trunked VLANs. */
3657 if (vlan < 0 && cfg->n_trunks) {
3660 trunks = bitmap_allocate(4096);
3662 for (i = 0; i < cfg->n_trunks; i++) {
3663 int trunk = cfg->trunks[i];
3665 bitmap_set1(trunks, trunk);
3671 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3672 port->name, cfg->n_trunks);
3674 if (n_errors == cfg->n_trunks) {
3675 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3677 bitmap_free(trunks);
3680 } else if (vlan >= 0 && cfg->n_trunks) {
3681 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3685 ? port->trunks != NULL
3686 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3687 bridge_flush(port->bridge);
3689 bitmap_free(port->trunks);
3690 port->trunks = trunks;
3694 port_destroy(struct port *port)
3697 struct bridge *br = port->bridge;
3701 proc_net_compat_update_vlan(port->name, NULL, 0);
3702 proc_net_compat_update_bond(port->name, NULL);
3704 for (i = 0; i < MAX_MIRRORS; i++) {
3705 struct mirror *m = br->mirrors[i];
3706 if (m && m->out_port == port) {
3711 while (port->n_ifaces > 0) {
3712 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3715 shash_find_and_delete_assert(&br->port_by_name, port->name);
3717 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3718 del->port_idx = port->port_idx;
3720 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3722 netdev_monitor_destroy(port->monitor);
3724 bitmap_free(port->trunks);
3731 static struct port *
3732 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3734 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3735 return iface ? iface->port : NULL;
3738 static struct port *
3739 port_lookup(const struct bridge *br, const char *name)
3741 return shash_find_data(&br->port_by_name, name);
3744 static struct iface *
3745 port_lookup_iface(const struct port *port, const char *name)
3747 struct iface *iface = iface_lookup(port->bridge, name);
3748 return iface && iface->port == port ? iface : NULL;
3752 port_update_bonding(struct port *port)
3754 if (port->monitor) {
3755 netdev_monitor_destroy(port->monitor);
3756 port->monitor = NULL;
3758 if (port->n_ifaces < 2) {
3759 /* Not a bonded port. */
3760 if (port->bond_hash) {
3761 free(port->bond_hash);
3762 port->bond_hash = NULL;
3763 port->bond_compat_is_stale = true;
3764 port->bond_fake_iface = false;
3769 if (!port->bond_hash) {
3770 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3771 for (i = 0; i <= BOND_MASK; i++) {
3772 struct bond_entry *e = &port->bond_hash[i];
3776 port->no_ifaces_tag = tag_create_random();
3777 bond_choose_active_iface(port);
3778 port->bond_next_rebalance
3779 = time_msec() + port->bond_rebalance_interval;
3781 if (port->cfg->bond_fake_iface) {
3782 port->bond_next_fake_iface_update = time_msec();
3785 port->bond_compat_is_stale = true;
3786 port->bond_fake_iface = port->cfg->bond_fake_iface;
3788 port->monitor = netdev_monitor_create();
3789 for (i = 0; i < port->n_ifaces; i++) {
3790 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
3796 port_update_bond_compat(struct port *port)
3798 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3799 struct compat_bond bond;
3802 if (port->n_ifaces < 2) {
3803 proc_net_compat_update_bond(port->name, NULL);
3808 bond.updelay = port->updelay;
3809 bond.downdelay = port->downdelay;
3812 bond.hashes = compat_hashes;
3813 if (port->bond_hash) {
3814 const struct bond_entry *e;
3815 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3816 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3817 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3818 cbh->hash = e - port->bond_hash;
3819 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3824 bond.n_slaves = port->n_ifaces;
3825 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3826 for (i = 0; i < port->n_ifaces; i++) {
3827 struct iface *iface = port->ifaces[i];
3828 struct compat_bond_slave *slave = &bond.slaves[i];
3829 slave->name = iface->name;
3831 /* We need to make the same determination as the Linux bonding
3832 * code to determine whether a slave should be consider "up".
3833 * The Linux function bond_miimon_inspect() supports four
3834 * BOND_LINK_* states:
3836 * - BOND_LINK_UP: carrier detected, updelay has passed.
3837 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3838 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3839 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3841 * The function bond_info_show_slave() only considers BOND_LINK_UP
3842 * to be "up" and anything else to be "down".
3844 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3848 netdev_get_etheraddr(iface->netdev, slave->mac);
3851 if (port->bond_fake_iface) {
3852 struct netdev *bond_netdev;
3854 if (!netdev_open_default(port->name, &bond_netdev)) {
3856 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3858 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3860 netdev_close(bond_netdev);
3864 proc_net_compat_update_bond(port->name, &bond);
3869 port_update_vlan_compat(struct port *port)
3871 struct bridge *br = port->bridge;
3872 char *vlandev_name = NULL;
3874 if (port->vlan > 0) {
3875 /* Figure out the name that the VLAN device should actually have, if it
3876 * existed. This takes some work because the VLAN device would not
3877 * have port->name in its name; rather, it would have the trunk port's
3878 * name, and 'port' would be attached to a bridge that also had the
3879 * VLAN device one of its ports. So we need to find a trunk port that
3880 * includes port->vlan.
3882 * There might be more than one candidate. This doesn't happen on
3883 * XenServer, so if it happens we just pick the first choice in
3884 * alphabetical order instead of creating multiple VLAN devices. */
3886 for (i = 0; i < br->n_ports; i++) {
3887 struct port *p = br->ports[i];
3888 if (port_trunks_vlan(p, port->vlan)
3890 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3892 uint8_t ea[ETH_ADDR_LEN];
3893 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3894 if (!eth_addr_is_multicast(ea) &&
3895 !eth_addr_is_reserved(ea) &&
3896 !eth_addr_is_zero(ea)) {
3897 vlandev_name = p->name;
3902 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3905 /* Interface functions. */
3908 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
3911 union ofp_action action;
3913 memset(&action, 0, sizeof action);
3914 action.output.type = htons(OFPAT_OUTPUT);
3915 action.output.len = htons(sizeof action);
3916 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
3918 flow_extract(packet, 0, ODPP_NONE, &flow);
3920 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
3922 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3923 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
3927 static struct iface *
3928 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3930 struct bridge *br = port->bridge;
3931 struct iface *iface;
3932 char *name = if_cfg->name;
3934 iface = xzalloc(sizeof *iface);
3936 iface->port_ifidx = port->n_ifaces;
3937 iface->name = xstrdup(name);
3938 iface->dp_ifidx = -1;
3939 iface->tag = tag_create_random();
3940 iface->delay_expires = LLONG_MAX;
3941 iface->netdev = NULL;
3942 iface->cfg = if_cfg;
3944 shash_add_assert(&br->iface_by_name, iface->name, iface);
3946 if (port->n_ifaces >= port->allocated_ifaces) {
3947 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3948 sizeof *port->ifaces);
3950 port->ifaces[port->n_ifaces++] = iface;
3951 if (port->n_ifaces > 1) {
3952 br->has_bonded_ports = true;
3955 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3963 iface_destroy(struct iface *iface)
3966 struct port *port = iface->port;
3967 struct bridge *br = port->bridge;
3968 bool del_active = port->active_iface == iface->port_ifidx;
3971 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3973 if (iface->dp_ifidx >= 0) {
3974 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
3977 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3978 del->port_ifidx = iface->port_ifidx;
3980 netdev_close(iface->netdev);
3983 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3984 bond_choose_active_iface(port);
3985 bond_send_learning_packets(port);
3988 cfm_destroy(iface->cfm);
3993 bridge_flush(port->bridge);
3997 static struct iface *
3998 iface_lookup(const struct bridge *br, const char *name)
4000 return shash_find_data(&br->iface_by_name, name);
4003 static struct iface *
4004 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4006 struct iface *iface;
4008 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4009 hash_int(dp_ifidx, 0), &br->ifaces) {
4010 if (iface->dp_ifidx == dp_ifidx) {
4017 /* Set Ethernet address of 'iface', if one is specified in the configuration
4020 iface_set_mac(struct iface *iface)
4022 uint8_t ea[ETH_ADDR_LEN];
4024 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4025 if (eth_addr_is_multicast(ea)) {
4026 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4028 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4029 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4030 iface->name, iface->name);
4032 int error = netdev_set_etheraddr(iface->netdev, ea);
4034 VLOG_ERR("interface %s: setting MAC failed (%s)",
4035 iface->name, strerror(error));
4041 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4043 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4046 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4050 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4052 * The value strings in '*shash' are taken directly from values[], not copied,
4053 * so the caller should not modify or free them. */
4055 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4056 struct shash *shash)
4061 for (i = 0; i < n; i++) {
4062 shash_add(shash, keys[i], values[i]);
4066 struct iface_delete_queues_cbdata {
4067 struct netdev *netdev;
4068 const struct ovsdb_datum *queues;
4072 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4074 union ovsdb_atom atom;
4076 atom.integer = target;
4077 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4081 iface_delete_queues(unsigned int queue_id,
4082 const struct shash *details OVS_UNUSED, void *cbdata_)
4084 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4086 if (!queue_ids_include(cbdata->queues, queue_id)) {
4087 netdev_delete_queue(cbdata->netdev, queue_id);
4092 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4094 if (!qos || qos->type[0] == '\0') {
4095 netdev_set_qos(iface->netdev, NULL, NULL);
4097 struct iface_delete_queues_cbdata cbdata;
4098 struct shash details;
4101 /* Configure top-level Qos for 'iface'. */
4102 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4103 qos->n_other_config, &details);
4104 netdev_set_qos(iface->netdev, qos->type, &details);
4105 shash_destroy(&details);
4107 /* Deconfigure queues that were deleted. */
4108 cbdata.netdev = iface->netdev;
4109 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4111 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4113 /* Configure queues for 'iface'. */
4114 for (i = 0; i < qos->n_queues; i++) {
4115 const struct ovsrec_queue *queue = qos->value_queues[i];
4116 unsigned int queue_id = qos->key_queues[i];
4118 shash_from_ovs_idl_map(queue->key_other_config,
4119 queue->value_other_config,
4120 queue->n_other_config, &details);
4121 netdev_set_queue(iface->netdev, queue_id, &details);
4122 shash_destroy(&details);
4128 iface_update_cfm(struct iface *iface)
4132 uint16_t *remote_mps;
4133 struct ovsrec_monitor *mon;
4134 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4136 mon = iface->cfg->monitor;
4142 if (netdev_get_etheraddr(iface->netdev, ea)) {
4143 VLOG_WARN("interface %s: Failed to get ethernet address. "
4144 "Skipping Monitor.", iface->name);
4148 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4149 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4154 iface->cfm = cfm_create();
4158 cfm->mpid = mon->mpid;
4159 cfm->interval = mon->interval ? *mon->interval : 1000;
4161 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4162 memcpy(cfm->maid, maid, sizeof cfm->maid);
4164 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4165 for(i = 0; i < mon->n_remote_mps; i++) {
4166 remote_mps[i] = mon->remote_mps[i]->mpid;
4168 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4171 if (!cfm_configure(iface->cfm)) {
4172 cfm_destroy(iface->cfm);
4177 /* Port mirroring. */
4179 static struct mirror *
4180 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4184 for (i = 0; i < MAX_MIRRORS; i++) {
4185 struct mirror *m = br->mirrors[i];
4186 if (m && uuid_equals(uuid, &m->uuid)) {
4194 mirror_reconfigure(struct bridge *br)
4196 unsigned long *rspan_vlans;
4199 /* Get rid of deleted mirrors. */
4200 for (i = 0; i < MAX_MIRRORS; i++) {
4201 struct mirror *m = br->mirrors[i];
4203 const struct ovsdb_datum *mc;
4204 union ovsdb_atom atom;
4206 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4207 atom.uuid = br->mirrors[i]->uuid;
4208 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4214 /* Add new mirrors and reconfigure existing ones. */
4215 for (i = 0; i < br->cfg->n_mirrors; i++) {
4216 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4217 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4219 mirror_reconfigure_one(m, cfg);
4221 mirror_create(br, cfg);
4225 /* Update port reserved status. */
4226 for (i = 0; i < br->n_ports; i++) {
4227 br->ports[i]->is_mirror_output_port = false;
4229 for (i = 0; i < MAX_MIRRORS; i++) {
4230 struct mirror *m = br->mirrors[i];
4231 if (m && m->out_port) {
4232 m->out_port->is_mirror_output_port = true;
4236 /* Update flooded vlans (for RSPAN). */
4238 if (br->cfg->n_flood_vlans) {
4239 rspan_vlans = bitmap_allocate(4096);
4241 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4242 int64_t vlan = br->cfg->flood_vlans[i];
4243 if (vlan >= 0 && vlan < 4096) {
4244 bitmap_set1(rspan_vlans, vlan);
4245 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4248 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4253 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4259 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4264 for (i = 0; ; i++) {
4265 if (i >= MAX_MIRRORS) {
4266 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4267 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4270 if (!br->mirrors[i]) {
4275 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4278 br->mirrors[i] = m = xzalloc(sizeof *m);
4281 m->name = xstrdup(cfg->name);
4282 shash_init(&m->src_ports);
4283 shash_init(&m->dst_ports);
4289 mirror_reconfigure_one(m, cfg);
4293 mirror_destroy(struct mirror *m)
4296 struct bridge *br = m->bridge;
4299 for (i = 0; i < br->n_ports; i++) {
4300 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4301 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4304 shash_destroy(&m->src_ports);
4305 shash_destroy(&m->dst_ports);
4308 m->bridge->mirrors[m->idx] = NULL;
4317 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4318 struct shash *names)
4322 for (i = 0; i < n_ports; i++) {
4323 const char *name = ports[i]->name;
4324 if (port_lookup(m->bridge, name)) {
4325 shash_add_once(names, name, NULL);
4327 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4328 "port %s", m->bridge->name, m->name, name);
4334 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4340 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4342 for (i = 0; i < cfg->n_select_vlan; i++) {
4343 int64_t vlan = cfg->select_vlan[i];
4344 if (vlan < 0 || vlan > 4095) {
4345 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4346 m->bridge->name, m->name, vlan);
4348 (*vlans)[n_vlans++] = vlan;
4355 vlan_is_mirrored(const struct mirror *m, int vlan)
4359 for (i = 0; i < m->n_vlans; i++) {
4360 if (m->vlans[i] == vlan) {
4368 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4372 for (i = 0; i < m->n_vlans; i++) {
4373 if (port_trunks_vlan(p, m->vlans[i])) {
4381 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4383 struct shash src_ports, dst_ports;
4384 mirror_mask_t mirror_bit;
4385 struct port *out_port;
4392 if (strcmp(cfg->name, m->name)) {
4394 m->name = xstrdup(cfg->name);
4397 /* Get output port. */
4398 if (cfg->output_port) {
4399 out_port = port_lookup(m->bridge, cfg->output_port->name);
4401 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4402 m->bridge->name, m->name);
4408 if (cfg->output_vlan) {
4409 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4410 "output vlan; ignoring output vlan",
4411 m->bridge->name, m->name);
4413 } else if (cfg->output_vlan) {
4415 out_vlan = *cfg->output_vlan;
4417 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4418 m->bridge->name, m->name);
4423 shash_init(&src_ports);
4424 shash_init(&dst_ports);
4425 if (cfg->select_all) {
4426 for (i = 0; i < m->bridge->n_ports; i++) {
4427 const char *name = m->bridge->ports[i]->name;
4428 shash_add_once(&src_ports, name, NULL);
4429 shash_add_once(&dst_ports, name, NULL);
4434 /* Get ports, and drop duplicates and ports that don't exist. */
4435 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4437 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4440 /* Get all the vlans, and drop duplicate and invalid vlans. */
4441 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4444 /* Update mirror data. */
4445 if (!shash_equal_keys(&m->src_ports, &src_ports)
4446 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4447 || m->n_vlans != n_vlans
4448 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4449 || m->out_port != out_port
4450 || m->out_vlan != out_vlan) {
4451 bridge_flush(m->bridge);
4453 shash_swap(&m->src_ports, &src_ports);
4454 shash_swap(&m->dst_ports, &dst_ports);
4457 m->n_vlans = n_vlans;
4458 m->out_port = out_port;
4459 m->out_vlan = out_vlan;
4462 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4463 for (i = 0; i < m->bridge->n_ports; i++) {
4464 struct port *port = m->bridge->ports[i];
4466 if (shash_find(&m->src_ports, port->name)
4469 ? port_trunks_any_mirrored_vlan(m, port)
4470 : vlan_is_mirrored(m, port->vlan)))) {
4471 port->src_mirrors |= mirror_bit;
4473 port->src_mirrors &= ~mirror_bit;
4476 if (shash_find(&m->dst_ports, port->name)) {
4477 port->dst_mirrors |= mirror_bit;
4479 port->dst_mirrors &= ~mirror_bit;
4484 shash_destroy(&src_ports);
4485 shash_destroy(&dst_ports);