1 /* Copyright (c) 2008, 2009, 2010, 2011 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"
41 #include "dynamic-string.h"
48 #include "mac-learning.h"
52 #include "ofp-print.h"
54 #include "ofproto/netflow.h"
55 #include "ofproto/ofproto.h"
56 #include "ovsdb-data.h"
58 #include "poll-loop.h"
62 #include "socket-util.h"
63 #include "stream-ssl.h"
65 #include "system-stats.h"
70 #include "vswitchd/vswitch-idl.h"
71 #include "xenserver.h"
73 #include "sflow_api.h"
75 VLOG_DEFINE_THIS_MODULE(bridge);
77 COVERAGE_DEFINE(bridge_flush);
78 COVERAGE_DEFINE(bridge_process_flow);
79 COVERAGE_DEFINE(bridge_process_cfm);
80 COVERAGE_DEFINE(bridge_process_lacp);
81 COVERAGE_DEFINE(bridge_reconfigure);
82 COVERAGE_DEFINE(bridge_lacp_update);
90 struct dst builtin[32];
95 static void dst_set_init(struct dst_set *);
96 static void dst_set_add(struct dst_set *, const struct dst *);
97 static void dst_set_free(struct dst_set *);
100 /* These members are always valid. */
101 struct list port_elem; /* Element in struct port's "ifaces" list. */
102 struct port *port; /* Containing port. */
103 char *name; /* Host network device name. */
104 tag_type tag; /* Tag associated with this interface. */
105 long long delay_expires; /* Time after which 'enabled' may change. */
107 /* These members are valid only after bridge_reconfigure() causes them to
109 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
110 int dp_ifidx; /* Index within kernel datapath. */
111 struct netdev *netdev; /* Network device. */
112 bool enabled; /* May be chosen for flows? */
113 bool up; /* Is the interface up? */
114 const char *type; /* Usually same as cfg->type. */
115 const struct ovsrec_interface *cfg;
117 /* LACP information. */
118 uint16_t lacp_priority; /* LACP port priority. */
121 #define BOND_MASK 0xff
123 struct iface *iface; /* Assigned iface, or NULL if none. */
124 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
125 tag_type tag; /* Tag for bond_entry<->iface association. */
129 BM_TCP, /* Transport Layer Load Balance. */
130 BM_SLB, /* Source Load Balance. */
131 BM_AB /* Active Backup. */
134 #define MAX_MIRRORS 32
135 typedef uint32_t mirror_mask_t;
136 #define MIRROR_MASK_C(X) UINT32_C(X)
137 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
139 struct bridge *bridge;
142 struct uuid uuid; /* UUID of this "mirror" record in database. */
144 /* Selection criteria. */
145 struct shash src_ports; /* Name is port name; data is always NULL. */
146 struct shash dst_ports; /* Name is port name; data is always NULL. */
151 struct port *out_port;
155 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
157 struct bridge *bridge;
158 struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */
161 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
162 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
163 * NULL if all VLANs are trunked. */
164 const struct ovsrec_port *cfg;
167 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
168 long long int miimon_interval; /* Miimon status refresh interval. */
169 long long int miimon_next_update; /* Time of next miimon update. */
171 /* An ordinary bridge port has 1 interface.
172 * A bridge port for bonding has at least 2 interfaces. */
173 struct list ifaces; /* List of "struct iface"s. */
174 size_t n_ifaces; /* list_size(ifaces). */
177 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
178 struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */
179 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
180 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
181 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
182 long long int bond_next_fake_iface_update; /* Time of next update. */
184 /* LACP information. */
185 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
186 bool lacp_active; /* True if LACP is active */
187 bool lacp_fast; /* True if LACP is in fast mode. */
188 uint16_t lacp_priority; /* LACP system priority. */
190 /* SLB specific bonding info. */
191 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
192 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
193 long long int bond_next_rebalance; /* Next rebalancing time. */
195 /* Port mirroring info. */
196 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
197 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
198 bool is_mirror_output_port; /* Does port mirroring send frames here? */
202 struct list node; /* Node in global list of bridges. */
203 char *name; /* User-specified arbitrary name. */
204 struct mac_learning *ml; /* MAC learning table. */
205 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
206 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
207 const struct ovsrec_bridge *cfg;
209 /* OpenFlow switch processing. */
210 struct ofproto *ofproto; /* OpenFlow switch. */
212 /* Kernel datapath information. */
213 struct dpif *dpif; /* Datapath. */
214 struct hmap ifaces; /* Contains "struct iface"s. */
217 struct hmap ports; /* "struct port"s indexed by name. */
218 struct shash iface_by_name; /* "struct iface"s indexed by name. */
221 bool has_bonded_ports;
226 /* Port mirroring. */
227 struct mirror *mirrors[MAX_MIRRORS];
230 /* List of all bridges. */
231 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
233 /* OVSDB IDL used to obtain configuration. */
234 static struct ovsdb_idl *idl;
236 /* Each time this timer expires, the bridge fetches systems and interface
237 * statistics and pushes them into the database. */
238 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
239 static long long int stats_timer = LLONG_MIN;
241 /* Stores the time after which CFM statistics may be written to the database.
242 * Only updated when changes to the database require rate limiting. */
243 #define CFM_LIMIT_INTERVAL (1 * 1000) /* In milliseconds. */
244 static long long int cfm_limiter = LLONG_MIN;
246 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
247 static void bridge_destroy(struct bridge *);
248 static struct bridge *bridge_lookup(const char *name);
249 static unixctl_cb_func bridge_unixctl_dump_flows;
250 static unixctl_cb_func bridge_unixctl_reconnect;
251 static int bridge_run_one(struct bridge *);
252 static size_t bridge_get_controllers(const struct bridge *br,
253 struct ovsrec_controller ***controllersp);
254 static void bridge_reconfigure_one(struct bridge *);
255 static void bridge_reconfigure_remotes(struct bridge *,
256 const struct sockaddr_in *managers,
258 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
259 static void bridge_fetch_dp_ifaces(struct bridge *);
260 static void bridge_flush(struct bridge *);
261 static void bridge_pick_local_hw_addr(struct bridge *,
262 uint8_t ea[ETH_ADDR_LEN],
263 struct iface **hw_addr_iface);
264 static uint64_t bridge_pick_datapath_id(struct bridge *,
265 const uint8_t bridge_ea[ETH_ADDR_LEN],
266 struct iface *hw_addr_iface);
267 static uint64_t dpid_from_hash(const void *, size_t nbytes);
269 static unixctl_cb_func bridge_unixctl_fdb_show;
270 static unixctl_cb_func cfm_unixctl_show;
271 static unixctl_cb_func qos_unixctl_show;
273 static void bond_init(void);
274 static void bond_run(struct port *);
275 static void bond_wait(struct port *);
276 static void bond_rebalance_port(struct port *);
277 static void bond_send_learning_packets(struct port *);
278 static void bond_enable_slave(struct iface *iface, bool enable);
280 static void port_run(struct port *);
281 static void port_wait(struct port *);
282 static struct port *port_create(struct bridge *, const char *name);
283 static void port_reconfigure(struct port *, const struct ovsrec_port *);
284 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
285 static void port_destroy(struct port *);
286 static struct port *port_lookup(const struct bridge *, const char *name);
287 static struct iface *port_lookup_iface(const struct port *, const char *name);
288 static struct iface *port_get_an_iface(const struct port *);
289 static struct port *port_from_dp_ifidx(const struct bridge *,
291 static void port_update_bonding(struct port *);
292 static void port_update_lacp(struct port *);
294 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
295 static void mirror_destroy(struct mirror *);
296 static void mirror_reconfigure(struct bridge *);
297 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
298 static bool vlan_is_mirrored(const struct mirror *, int vlan);
300 static struct iface *iface_create(struct port *port,
301 const struct ovsrec_interface *if_cfg);
302 static void iface_destroy(struct iface *);
303 static struct iface *iface_lookup(const struct bridge *, const char *name);
304 static struct iface *iface_find(const char *name);
305 static struct iface *iface_from_dp_ifidx(const struct bridge *,
307 static void iface_set_mac(struct iface *);
308 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
309 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
310 static void iface_update_cfm(struct iface *);
311 static bool iface_refresh_cfm_stats(struct iface *iface);
312 static void iface_update_carrier(struct iface *);
313 static bool iface_get_carrier(const struct iface *);
315 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
317 static void shash_to_ovs_idl_map(struct shash *,
318 char ***keys, char ***values, size_t *n);
320 /* Hooks into ofproto processing. */
321 static struct ofhooks bridge_ofhooks;
323 /* Public functions. */
325 /* Initializes the bridge module, configuring it to obtain its configuration
326 * from an OVSDB server accessed over 'remote', which should be a string in a
327 * form acceptable to ovsdb_idl_create(). */
329 bridge_init(const char *remote)
331 /* Create connection to database. */
332 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
334 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
335 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
336 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
338 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
340 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
341 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
343 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
344 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
345 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
347 /* Register unixctl commands. */
348 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
349 unixctl_command_register("cfm/show", cfm_unixctl_show, NULL);
350 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
351 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
353 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
362 struct bridge *br, *next_br;
364 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
367 ovsdb_idl_destroy(idl);
370 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
371 * but for which the ovs-vswitchd configuration 'cfg' is required. */
373 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
375 static bool already_configured_once;
376 struct svec bridge_names;
377 struct svec dpif_names, dpif_types;
380 /* Only do this once per ovs-vswitchd run. */
381 if (already_configured_once) {
384 already_configured_once = true;
386 stats_timer = time_msec() + STATS_INTERVAL;
388 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
389 svec_init(&bridge_names);
390 for (i = 0; i < cfg->n_bridges; i++) {
391 svec_add(&bridge_names, cfg->bridges[i]->name);
393 svec_sort(&bridge_names);
395 /* Iterate over all system dpifs and delete any of them that do not appear
397 svec_init(&dpif_names);
398 svec_init(&dpif_types);
399 dp_enumerate_types(&dpif_types);
400 for (i = 0; i < dpif_types.n; i++) {
403 dp_enumerate_names(dpif_types.names[i], &dpif_names);
405 /* Delete each dpif whose name is not in 'bridge_names'. */
406 for (j = 0; j < dpif_names.n; j++) {
407 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
411 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
420 svec_destroy(&bridge_names);
421 svec_destroy(&dpif_names);
422 svec_destroy(&dpif_types);
425 /* Callback for iterate_and_prune_ifaces(). */
427 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
429 if (!iface->netdev) {
430 /* We already reported a related error, don't bother duplicating it. */
434 if (iface->dp_ifidx < 0) {
435 VLOG_ERR("%s interface not in %s, dropping",
436 iface->name, dpif_name(br->dpif));
440 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
441 iface->name, iface->dp_ifidx);
445 /* Callback for iterate_and_prune_ifaces(). */
447 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
448 void *aux OVS_UNUSED)
450 /* Set policing attributes. */
451 netdev_set_policing(iface->netdev,
452 iface->cfg->ingress_policing_rate,
453 iface->cfg->ingress_policing_burst);
455 /* Set MAC address of internal interfaces other than the local
457 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
458 iface_set_mac(iface);
464 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
465 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
466 * deletes from 'br' any ports that no longer have any interfaces. */
468 iterate_and_prune_ifaces(struct bridge *br,
469 bool (*cb)(struct bridge *, struct iface *,
473 struct port *port, *next_port;
475 HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) {
476 struct iface *iface, *next_iface;
478 LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) {
479 if (!cb(br, iface, aux)) {
480 iface_set_ofport(iface->cfg, -1);
481 iface_destroy(iface);
485 if (!port->n_ifaces) {
486 VLOG_WARN("%s port has no interfaces, dropping", port->name);
492 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
493 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
494 * responsible for freeing '*managersp' (with free()).
496 * You may be asking yourself "why does ovs-vswitchd care?", because
497 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
498 * should not be and in fact is not directly involved in that. But
499 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
500 * it has to tell in-band control where the managers are to enable that.
501 * (Thus, only managers connected in-band are collected.)
504 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
505 struct sockaddr_in **managersp, size_t *n_managersp)
507 struct sockaddr_in *managers = NULL;
508 size_t n_managers = 0;
509 struct shash targets;
512 /* Collect all of the potential targets from the "targets" columns of the
513 * rows pointed to by "manager_options", excluding any that are
515 shash_init(&targets);
516 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
517 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
519 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
520 shash_find_and_delete(&targets, m->target);
522 shash_add_once(&targets, m->target, NULL);
526 /* Now extract the targets' IP addresses. */
527 if (!shash_is_empty(&targets)) {
528 struct shash_node *node;
530 managers = xmalloc(shash_count(&targets) * sizeof *managers);
531 SHASH_FOR_EACH (node, &targets) {
532 const char *target = node->name;
533 struct sockaddr_in *sin = &managers[n_managers];
535 if ((!strncmp(target, "tcp:", 4)
536 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
537 (!strncmp(target, "ssl:", 4)
538 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
543 shash_destroy(&targets);
545 *managersp = managers;
546 *n_managersp = n_managers;
550 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
552 struct shash old_br, new_br;
553 struct shash_node *node;
554 struct bridge *br, *next;
555 struct sockaddr_in *managers;
558 int sflow_bridge_number;
560 COVERAGE_INC(bridge_reconfigure);
562 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
564 /* Collect old and new bridges. */
567 LIST_FOR_EACH (br, node, &all_bridges) {
568 shash_add(&old_br, br->name, br);
570 for (i = 0; i < ovs_cfg->n_bridges; i++) {
571 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
572 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
573 VLOG_WARN("more than one bridge named %s", br_cfg->name);
577 /* Get rid of deleted bridges and add new bridges. */
578 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
579 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
586 SHASH_FOR_EACH (node, &new_br) {
587 const char *br_name = node->name;
588 const struct ovsrec_bridge *br_cfg = node->data;
589 br = shash_find_data(&old_br, br_name);
591 /* If the bridge datapath type has changed, we need to tear it
592 * down and recreate. */
593 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
595 bridge_create(br_cfg);
598 bridge_create(br_cfg);
601 shash_destroy(&old_br);
602 shash_destroy(&new_br);
604 /* Reconfigure all bridges. */
605 LIST_FOR_EACH (br, node, &all_bridges) {
606 bridge_reconfigure_one(br);
609 /* Add and delete ports on all datapaths.
611 * The kernel will reject any attempt to add a given port to a datapath if
612 * that port already belongs to a different datapath, so we must do all
613 * port deletions before any port additions. */
614 LIST_FOR_EACH (br, node, &all_bridges) {
615 struct dpif_port_dump dump;
616 struct shash want_ifaces;
617 struct dpif_port dpif_port;
619 bridge_get_all_ifaces(br, &want_ifaces);
620 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
621 if (!shash_find(&want_ifaces, dpif_port.name)
622 && strcmp(dpif_port.name, br->name)) {
623 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
625 VLOG_WARN("failed to remove %s interface from %s: %s",
626 dpif_port.name, dpif_name(br->dpif),
631 shash_destroy(&want_ifaces);
633 LIST_FOR_EACH (br, node, &all_bridges) {
634 struct shash cur_ifaces, want_ifaces;
635 struct dpif_port_dump dump;
636 struct dpif_port dpif_port;
638 /* Get the set of interfaces currently in this datapath. */
639 shash_init(&cur_ifaces);
640 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
641 struct dpif_port *port_info = xmalloc(sizeof *port_info);
642 dpif_port_clone(port_info, &dpif_port);
643 shash_add(&cur_ifaces, dpif_port.name, port_info);
646 /* Get the set of interfaces we want on this datapath. */
647 bridge_get_all_ifaces(br, &want_ifaces);
649 hmap_clear(&br->ifaces);
650 SHASH_FOR_EACH (node, &want_ifaces) {
651 const char *if_name = node->name;
652 struct iface *iface = node->data;
653 struct dpif_port *dpif_port;
657 type = iface ? iface->type : "internal";
658 dpif_port = shash_find_data(&cur_ifaces, if_name);
660 /* If we have a port or a netdev already, and it's not the type we
661 * want, then delete the port (if any) and close the netdev (if
663 if ((dpif_port && strcmp(dpif_port->type, type))
664 || (iface && iface->netdev
665 && strcmp(type, netdev_get_type(iface->netdev)))) {
667 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
674 netdev_close(iface->netdev);
675 iface->netdev = NULL;
679 /* If the port doesn't exist or we don't have the netdev open,
680 * we need to do more work. */
681 if (!dpif_port || (iface && !iface->netdev)) {
682 struct netdev_options options;
683 struct netdev *netdev;
686 /* First open the network device. */
687 options.name = if_name;
689 options.args = &args;
690 options.ethertype = NETDEV_ETH_TYPE_NONE;
694 shash_from_ovs_idl_map(iface->cfg->key_options,
695 iface->cfg->value_options,
696 iface->cfg->n_options, &args);
698 error = netdev_open(&options, &netdev);
699 shash_destroy(&args);
702 VLOG_WARN("could not open network device %s (%s)",
703 if_name, strerror(error));
707 /* Then add the port if we haven't already. */
709 error = dpif_port_add(br->dpif, netdev, NULL);
711 netdev_close(netdev);
712 if (error == EFBIG) {
713 VLOG_ERR("ran out of valid port numbers on %s",
714 dpif_name(br->dpif));
717 VLOG_WARN("failed to add %s interface to %s: %s",
718 if_name, dpif_name(br->dpif),
725 /* Update 'iface'. */
727 iface->netdev = netdev;
728 iface->enabled = iface_get_carrier(iface);
729 iface->up = iface->enabled;
731 } else if (iface && iface->netdev) {
735 shash_from_ovs_idl_map(iface->cfg->key_options,
736 iface->cfg->value_options,
737 iface->cfg->n_options, &args);
738 netdev_set_config(iface->netdev, &args);
739 shash_destroy(&args);
742 shash_destroy(&want_ifaces);
744 SHASH_FOR_EACH (node, &cur_ifaces) {
745 struct dpif_port *port_info = node->data;
746 dpif_port_destroy(port_info);
749 shash_destroy(&cur_ifaces);
751 sflow_bridge_number = 0;
752 LIST_FOR_EACH (br, node, &all_bridges) {
755 struct iface *local_iface;
756 struct iface *hw_addr_iface;
759 bridge_fetch_dp_ifaces(br);
761 iterate_and_prune_ifaces(br, check_iface, NULL);
763 /* Pick local port hardware address, datapath ID. */
764 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
765 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
767 int error = netdev_set_etheraddr(local_iface->netdev, ea);
769 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
770 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
771 "Ethernet address: %s",
772 br->name, strerror(error));
775 memcpy(br->ea, ea, ETH_ADDR_LEN);
777 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
778 ofproto_set_datapath_id(br->ofproto, dpid);
780 dpid_string = xasprintf("%016"PRIx64, dpid);
781 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
784 /* Set NetFlow configuration on this bridge. */
785 if (br->cfg->netflow) {
786 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
787 struct netflow_options opts;
789 memset(&opts, 0, sizeof opts);
791 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
792 if (nf_cfg->engine_type) {
793 opts.engine_type = *nf_cfg->engine_type;
795 if (nf_cfg->engine_id) {
796 opts.engine_id = *nf_cfg->engine_id;
799 opts.active_timeout = nf_cfg->active_timeout;
800 if (!opts.active_timeout) {
801 opts.active_timeout = -1;
802 } else if (opts.active_timeout < 0) {
803 VLOG_WARN("bridge %s: active timeout interval set to negative "
804 "value, using default instead (%d seconds)", br->name,
805 NF_ACTIVE_TIMEOUT_DEFAULT);
806 opts.active_timeout = -1;
809 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
810 if (opts.add_id_to_iface) {
811 if (opts.engine_id > 0x7f) {
812 VLOG_WARN("bridge %s: netflow port mangling may conflict "
813 "with another vswitch, choose an engine id less "
814 "than 128", br->name);
816 if (hmap_count(&br->ports) > 508) {
817 VLOG_WARN("bridge %s: netflow port mangling will conflict "
818 "with another port when more than 508 ports are "
823 opts.collectors.n = nf_cfg->n_targets;
824 opts.collectors.names = nf_cfg->targets;
825 if (ofproto_set_netflow(br->ofproto, &opts)) {
826 VLOG_ERR("bridge %s: problem setting netflow collectors",
830 ofproto_set_netflow(br->ofproto, NULL);
833 /* Set sFlow configuration on this bridge. */
834 if (br->cfg->sflow) {
835 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
836 struct ovsrec_controller **controllers;
837 struct ofproto_sflow_options oso;
838 size_t n_controllers;
840 memset(&oso, 0, sizeof oso);
842 oso.targets.n = sflow_cfg->n_targets;
843 oso.targets.names = sflow_cfg->targets;
845 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
846 if (sflow_cfg->sampling) {
847 oso.sampling_rate = *sflow_cfg->sampling;
850 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
851 if (sflow_cfg->polling) {
852 oso.polling_interval = *sflow_cfg->polling;
855 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
856 if (sflow_cfg->header) {
857 oso.header_len = *sflow_cfg->header;
860 oso.sub_id = sflow_bridge_number++;
861 oso.agent_device = sflow_cfg->agent;
863 oso.control_ip = NULL;
864 n_controllers = bridge_get_controllers(br, &controllers);
865 for (i = 0; i < n_controllers; i++) {
866 if (controllers[i]->local_ip) {
867 oso.control_ip = controllers[i]->local_ip;
871 ofproto_set_sflow(br->ofproto, &oso);
873 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
875 ofproto_set_sflow(br->ofproto, NULL);
878 /* Update the controller and related settings. It would be more
879 * straightforward to call this from bridge_reconfigure_one(), but we
880 * can't do it there for two reasons. First, and most importantly, at
881 * that point we don't know the dp_ifidx of any interfaces that have
882 * been added to the bridge (because we haven't actually added them to
883 * the datapath). Second, at that point we haven't set the datapath ID
884 * yet; when a controller is configured, resetting the datapath ID will
885 * immediately disconnect from the controller, so it's better to set
886 * the datapath ID before the controller. */
887 bridge_reconfigure_remotes(br, managers, n_managers);
889 LIST_FOR_EACH (br, node, &all_bridges) {
892 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
896 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
897 netdev_monitor_add(port->monitor, iface->netdev);
900 port->miimon_next_update = 0;
903 port_update_lacp(port);
904 port_update_bonding(port);
906 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
907 iface_update_qos(iface, port->cfg->qos);
911 LIST_FOR_EACH (br, node, &all_bridges) {
912 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
915 LIST_FOR_EACH (br, node, &all_bridges) {
917 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
918 iface_update_cfm(iface);
924 /* ovs-vswitchd has completed initialization, so allow the process that
925 * forked us to exit successfully. */
926 daemonize_complete();
930 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
931 const struct ovsdb_idl_column *column,
934 const struct ovsdb_datum *datum;
935 union ovsdb_atom atom;
938 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
939 atom.string = (char *) key;
940 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
941 return idx == UINT_MAX ? NULL : datum->values[idx].string;
945 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
947 return get_ovsrec_key_value(&br_cfg->header_,
948 &ovsrec_bridge_col_other_config, key);
952 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
953 struct iface **hw_addr_iface)
959 *hw_addr_iface = NULL;
961 /* Did the user request a particular MAC? */
962 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
963 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
964 if (eth_addr_is_multicast(ea)) {
965 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
966 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
967 } else if (eth_addr_is_zero(ea)) {
968 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
974 /* Otherwise choose the minimum non-local MAC address among all of the
976 memset(ea, 0xff, ETH_ADDR_LEN);
977 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
978 uint8_t iface_ea[ETH_ADDR_LEN];
979 struct iface *candidate;
982 /* Mirror output ports don't participate. */
983 if (port->is_mirror_output_port) {
987 /* Choose the MAC address to represent the port. */
989 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
990 /* Find the interface with this Ethernet address (if any) so that
991 * we can provide the correct devname to the caller. */
992 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
993 uint8_t candidate_ea[ETH_ADDR_LEN];
994 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
995 && eth_addr_equals(iface_ea, candidate_ea)) {
1000 /* Choose the interface whose MAC address will represent the port.
1001 * The Linux kernel bonding code always chooses the MAC address of
1002 * the first slave added to a bond, and the Fedora networking
1003 * scripts always add slaves to a bond in alphabetical order, so
1004 * for compatibility we choose the interface with the name that is
1005 * first in alphabetical order. */
1006 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1007 if (!iface || strcmp(candidate->name, iface->name) < 0) {
1012 /* The local port doesn't count (since we're trying to choose its
1013 * MAC address anyway). */
1014 if (iface->dp_ifidx == ODPP_LOCAL) {
1019 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1021 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1022 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1023 iface->name, strerror(error));
1028 /* Compare against our current choice. */
1029 if (!eth_addr_is_multicast(iface_ea) &&
1030 !eth_addr_is_local(iface_ea) &&
1031 !eth_addr_is_reserved(iface_ea) &&
1032 !eth_addr_is_zero(iface_ea) &&
1033 eth_addr_compare_3way(iface_ea, ea) < 0)
1035 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1036 *hw_addr_iface = iface;
1039 if (eth_addr_is_multicast(ea)) {
1040 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1041 *hw_addr_iface = NULL;
1042 VLOG_WARN("bridge %s: using default bridge Ethernet "
1043 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1045 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1046 br->name, ETH_ADDR_ARGS(ea));
1050 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1051 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1052 * an interface on 'br', then that interface must be passed in as
1053 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1054 * 'hw_addr_iface' must be passed in as a null pointer. */
1056 bridge_pick_datapath_id(struct bridge *br,
1057 const uint8_t bridge_ea[ETH_ADDR_LEN],
1058 struct iface *hw_addr_iface)
1061 * The procedure for choosing a bridge MAC address will, in the most
1062 * ordinary case, also choose a unique MAC that we can use as a datapath
1063 * ID. In some special cases, though, multiple bridges will end up with
1064 * the same MAC address. This is OK for the bridges, but it will confuse
1065 * the OpenFlow controller, because each datapath needs a unique datapath
1068 * Datapath IDs must be unique. It is also very desirable that they be
1069 * stable from one run to the next, so that policy set on a datapath
1072 const char *datapath_id;
1075 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1076 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1080 if (hw_addr_iface) {
1082 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1084 * A bridge whose MAC address is taken from a VLAN network device
1085 * (that is, a network device created with vconfig(8) or similar
1086 * tool) will have the same MAC address as a bridge on the VLAN
1087 * device's physical network device.
1089 * Handle this case by hashing the physical network device MAC
1090 * along with the VLAN identifier.
1092 uint8_t buf[ETH_ADDR_LEN + 2];
1093 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1094 buf[ETH_ADDR_LEN] = vlan >> 8;
1095 buf[ETH_ADDR_LEN + 1] = vlan;
1096 return dpid_from_hash(buf, sizeof buf);
1099 * Assume that this bridge's MAC address is unique, since it
1100 * doesn't fit any of the cases we handle specially.
1105 * A purely internal bridge, that is, one that has no non-virtual
1106 * network devices on it at all, is more difficult because it has no
1107 * natural unique identifier at all.
1109 * When the host is a XenServer, we handle this case by hashing the
1110 * host's UUID with the name of the bridge. Names of bridges are
1111 * persistent across XenServer reboots, although they can be reused if
1112 * an internal network is destroyed and then a new one is later
1113 * created, so this is fairly effective.
1115 * When the host is not a XenServer, we punt by using a random MAC
1116 * address on each run.
1118 const char *host_uuid = xenserver_get_host_uuid();
1120 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1121 dpid = dpid_from_hash(combined, strlen(combined));
1127 return eth_addr_to_uint64(bridge_ea);
1131 dpid_from_hash(const void *data, size_t n)
1133 uint8_t hash[SHA1_DIGEST_SIZE];
1135 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1136 sha1_bytes(data, n, hash);
1137 eth_addr_mark_random(hash);
1138 return eth_addr_to_uint64(hash);
1142 iface_refresh_status(struct iface *iface)
1146 enum netdev_flags flags;
1155 if (!netdev_get_status(iface->netdev, &sh)) {
1157 char **keys, **values;
1159 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1160 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1165 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1168 shash_destroy_free_data(&sh);
1170 error = netdev_get_flags(iface->netdev, &flags);
1172 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1175 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1178 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1180 ovsrec_interface_set_duplex(iface->cfg,
1181 netdev_features_is_full_duplex(current)
1183 /* warning: uint64_t -> int64_t conversion */
1184 bps = netdev_features_to_bps(current);
1185 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1188 ovsrec_interface_set_duplex(iface->cfg, NULL);
1189 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1193 ovsrec_interface_set_link_state(iface->cfg,
1194 iface_get_carrier(iface) ? "up" : "down");
1196 error = netdev_get_mtu(iface->netdev, &mtu);
1197 if (!error && mtu != INT_MAX) {
1199 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1202 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1206 /* Writes 'iface''s CFM statistics to the database. Returns true if anything
1207 * changed, false otherwise. */
1209 iface_refresh_cfm_stats(struct iface *iface)
1211 const struct ovsrec_monitor *mon;
1212 const struct cfm *cfm;
1213 bool changed = false;
1216 mon = iface->cfg->monitor;
1217 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1223 for (i = 0; i < mon->n_remote_mps; i++) {
1224 const struct ovsrec_maintenance_point *mp;
1225 const struct remote_mp *rmp;
1227 mp = mon->remote_mps[i];
1228 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1230 if (mp->n_fault != 1 || mp->fault[0] != rmp->fault) {
1231 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1236 if (mon->n_fault != 1 || mon->fault[0] != cfm->fault) {
1237 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1245 iface_refresh_stats(struct iface *iface)
1251 static const struct iface_stat iface_stats[] = {
1252 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1253 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1254 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1255 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1256 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1257 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1258 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1259 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1260 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1261 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1262 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1263 { "collisions", offsetof(struct netdev_stats, collisions) },
1265 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1266 const struct iface_stat *s;
1268 char *keys[N_STATS];
1269 int64_t values[N_STATS];
1272 struct netdev_stats stats;
1274 /* Intentionally ignore return value, since errors will set 'stats' to
1275 * all-1s, and we will deal with that correctly below. */
1276 netdev_get_stats(iface->netdev, &stats);
1279 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1280 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1281 if (value != UINT64_MAX) {
1288 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1292 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1294 struct ovsdb_datum datum;
1298 get_system_stats(&stats);
1300 ovsdb_datum_from_shash(&datum, &stats);
1301 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1305 static inline const char *
1306 nx_role_to_str(enum nx_role role)
1311 case NX_ROLE_MASTER:
1316 return "*** INVALID ROLE ***";
1321 bridge_refresh_controller_status(const struct bridge *br)
1324 const struct ovsrec_controller *cfg;
1326 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1328 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1329 struct ofproto_controller_info *cinfo =
1330 shash_find_data(&info, cfg->target);
1333 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1334 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1335 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1336 (char **) cinfo->pairs.values,
1339 ovsrec_controller_set_is_connected(cfg, false);
1340 ovsrec_controller_set_role(cfg, NULL);
1341 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1345 ofproto_free_ofproto_controller_info(&info);
1351 const struct ovsrec_open_vswitch *cfg;
1353 bool datapath_destroyed;
1354 bool database_changed;
1357 /* Let each bridge do the work that it needs to do. */
1358 datapath_destroyed = false;
1359 LIST_FOR_EACH (br, node, &all_bridges) {
1360 int error = bridge_run_one(br);
1362 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1363 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1364 "forcing reconfiguration", br->name);
1365 datapath_destroyed = true;
1369 /* (Re)configure if necessary. */
1370 database_changed = ovsdb_idl_run(idl);
1371 cfg = ovsrec_open_vswitch_first(idl);
1373 /* Re-configure SSL. We do this on every trip through the main loop,
1374 * instead of just when the database changes, because the contents of the
1375 * key and certificate files can change without the database changing.
1377 * We do this before bridge_reconfigure() because that function might
1378 * initiate SSL connections and thus requires SSL to be configured. */
1379 if (cfg && cfg->ssl) {
1380 const struct ovsrec_ssl *ssl = cfg->ssl;
1382 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1383 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1386 if (database_changed || datapath_destroyed) {
1388 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1390 bridge_configure_once(cfg);
1391 bridge_reconfigure(cfg);
1393 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1394 ovsdb_idl_txn_commit(txn);
1395 ovsdb_idl_txn_destroy(txn); /* XXX */
1397 /* We still need to reconfigure to avoid dangling pointers to
1398 * now-destroyed ovsrec structures inside bridge data. */
1399 static const struct ovsrec_open_vswitch null_cfg;
1401 bridge_reconfigure(&null_cfg);
1405 /* Refresh system and interface stats if necessary. */
1406 if (time_msec() >= stats_timer) {
1408 struct ovsdb_idl_txn *txn;
1410 txn = ovsdb_idl_txn_create(idl);
1411 LIST_FOR_EACH (br, node, &all_bridges) {
1414 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1415 struct iface *iface;
1417 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1418 iface_refresh_stats(iface);
1419 iface_refresh_status(iface);
1422 bridge_refresh_controller_status(br);
1424 refresh_system_stats(cfg);
1425 ovsdb_idl_txn_commit(txn);
1426 ovsdb_idl_txn_destroy(txn); /* XXX */
1429 stats_timer = time_msec() + STATS_INTERVAL;
1432 if (time_msec() >= cfm_limiter) {
1433 struct ovsdb_idl_txn *txn;
1434 bool changed = false;
1436 txn = ovsdb_idl_txn_create(idl);
1437 LIST_FOR_EACH (br, node, &all_bridges) {
1440 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1441 struct iface *iface;
1443 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1444 changed = iface_refresh_cfm_stats(iface) || changed;
1450 cfm_limiter = time_msec() + CFM_LIMIT_INTERVAL;
1453 ovsdb_idl_txn_commit(txn);
1454 ovsdb_idl_txn_destroy(txn);
1463 LIST_FOR_EACH (br, node, &all_bridges) {
1466 ofproto_wait(br->ofproto);
1467 mac_learning_wait(br->ml);
1468 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1472 ovsdb_idl_wait(idl);
1473 poll_timer_wait_until(stats_timer);
1475 if (cfm_limiter > time_msec()) {
1476 poll_timer_wait_until(cfm_limiter);
1480 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1481 * configuration changes. */
1483 bridge_flush(struct bridge *br)
1485 COVERAGE_INC(bridge_flush);
1489 /* Bridge unixctl user interface functions. */
1491 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1492 const char *args, void *aux OVS_UNUSED)
1494 struct ds ds = DS_EMPTY_INITIALIZER;
1495 const struct bridge *br;
1496 const struct mac_entry *e;
1498 br = bridge_lookup(args);
1500 unixctl_command_reply(conn, 501, "no such bridge");
1504 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1505 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1506 struct port *port = e->port.p;
1507 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1508 port_get_an_iface(port)->dp_ifidx,
1509 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1511 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1515 /* CFM unixctl user interface functions. */
1517 cfm_unixctl_show(struct unixctl_conn *conn,
1518 const char *args, void *aux OVS_UNUSED)
1520 struct ds ds = DS_EMPTY_INITIALIZER;
1521 struct iface *iface;
1522 const struct cfm *cfm;
1524 iface = iface_find(args);
1526 unixctl_command_reply(conn, 501, "no such interface");
1530 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1533 unixctl_command_reply(conn, 501, "CFM not enabled");
1537 cfm_dump_ds(cfm, &ds);
1538 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1542 /* QoS unixctl user interface functions. */
1544 struct qos_unixctl_show_cbdata {
1546 struct iface *iface;
1550 qos_unixctl_show_cb(unsigned int queue_id,
1551 const struct shash *details,
1554 struct qos_unixctl_show_cbdata *data = aux;
1555 struct ds *ds = data->ds;
1556 struct iface *iface = data->iface;
1557 struct netdev_queue_stats stats;
1558 struct shash_node *node;
1561 ds_put_cstr(ds, "\n");
1563 ds_put_format(ds, "Queue %u:\n", queue_id);
1565 ds_put_cstr(ds, "Default:\n");
1568 SHASH_FOR_EACH (node, details) {
1569 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1572 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1574 if (stats.tx_packets != UINT64_MAX) {
1575 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1578 if (stats.tx_bytes != UINT64_MAX) {
1579 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1582 if (stats.tx_errors != UINT64_MAX) {
1583 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1586 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1587 queue_id, strerror(error));
1592 qos_unixctl_show(struct unixctl_conn *conn,
1593 const char *args, void *aux OVS_UNUSED)
1595 struct ds ds = DS_EMPTY_INITIALIZER;
1596 struct shash sh = SHASH_INITIALIZER(&sh);
1597 struct iface *iface;
1599 struct shash_node *node;
1600 struct qos_unixctl_show_cbdata data;
1603 iface = iface_find(args);
1605 unixctl_command_reply(conn, 501, "no such interface");
1609 netdev_get_qos(iface->netdev, &type, &sh);
1611 if (*type != '\0') {
1612 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1614 SHASH_FOR_EACH (node, &sh) {
1615 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1620 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1623 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1625 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1627 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1628 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1631 shash_destroy_free_data(&sh);
1635 /* Bridge reconfiguration functions. */
1636 static struct bridge *
1637 bridge_create(const struct ovsrec_bridge *br_cfg)
1642 assert(!bridge_lookup(br_cfg->name));
1643 br = xzalloc(sizeof *br);
1645 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1651 dpif_flow_flush(br->dpif);
1653 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1656 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1658 dpif_delete(br->dpif);
1659 dpif_close(br->dpif);
1664 br->name = xstrdup(br_cfg->name);
1666 br->ml = mac_learning_create();
1667 eth_addr_nicira_random(br->default_ea);
1669 hmap_init(&br->ports);
1670 hmap_init(&br->ifaces);
1671 shash_init(&br->iface_by_name);
1675 list_push_back(&all_bridges, &br->node);
1677 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1683 bridge_destroy(struct bridge *br)
1686 struct port *port, *next;
1689 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1692 list_remove(&br->node);
1693 error = dpif_delete(br->dpif);
1694 if (error && error != ENOENT) {
1695 VLOG_ERR("failed to delete %s: %s",
1696 dpif_name(br->dpif), strerror(error));
1698 dpif_close(br->dpif);
1699 ofproto_destroy(br->ofproto);
1700 mac_learning_destroy(br->ml);
1701 hmap_destroy(&br->ifaces);
1702 hmap_destroy(&br->ports);
1703 shash_destroy(&br->iface_by_name);
1709 static struct bridge *
1710 bridge_lookup(const char *name)
1714 LIST_FOR_EACH (br, node, &all_bridges) {
1715 if (!strcmp(br->name, name)) {
1722 /* Handle requests for a listing of all flows known by the OpenFlow
1723 * stack, including those normally hidden. */
1725 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1726 const char *args, void *aux OVS_UNUSED)
1731 br = bridge_lookup(args);
1733 unixctl_command_reply(conn, 501, "Unknown bridge");
1738 ofproto_get_all_flows(br->ofproto, &results);
1740 unixctl_command_reply(conn, 200, ds_cstr(&results));
1741 ds_destroy(&results);
1744 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1745 * connections and reconnect. If BRIDGE is not specified, then all bridges
1746 * drop their controller connections and reconnect. */
1748 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1749 const char *args, void *aux OVS_UNUSED)
1752 if (args[0] != '\0') {
1753 br = bridge_lookup(args);
1755 unixctl_command_reply(conn, 501, "Unknown bridge");
1758 ofproto_reconnect_controllers(br->ofproto);
1760 LIST_FOR_EACH (br, node, &all_bridges) {
1761 ofproto_reconnect_controllers(br->ofproto);
1764 unixctl_command_reply(conn, 200, NULL);
1768 bridge_run_one(struct bridge *br)
1773 error = ofproto_run1(br->ofproto);
1778 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1780 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1784 error = ofproto_run2(br->ofproto, br->flush);
1791 bridge_get_controllers(const struct bridge *br,
1792 struct ovsrec_controller ***controllersp)
1794 struct ovsrec_controller **controllers;
1795 size_t n_controllers;
1797 controllers = br->cfg->controller;
1798 n_controllers = br->cfg->n_controller;
1800 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1806 *controllersp = controllers;
1808 return n_controllers;
1812 bridge_reconfigure_one(struct bridge *br)
1814 enum ofproto_fail_mode fail_mode;
1815 struct svec snoops, old_snoops;
1816 struct port *port, *next;
1817 struct shash_node *node;
1818 struct shash new_ports;
1821 /* Collect new ports. */
1822 shash_init(&new_ports);
1823 for (i = 0; i < br->cfg->n_ports; i++) {
1824 const char *name = br->cfg->ports[i]->name;
1825 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1826 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1831 /* If we have a controller, then we need a local port. Complain if the
1832 * user didn't specify one.
1834 * XXX perhaps we should synthesize a port ourselves in this case. */
1835 if (bridge_get_controllers(br, NULL)) {
1836 char local_name[IF_NAMESIZE];
1839 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1840 local_name, sizeof local_name);
1841 if (!error && !shash_find(&new_ports, local_name)) {
1842 VLOG_WARN("bridge %s: controller specified but no local port "
1843 "(port named %s) defined",
1844 br->name, local_name);
1848 /* Get rid of deleted ports.
1849 * Get rid of deleted interfaces on ports that still exist. */
1850 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1851 const struct ovsrec_port *port_cfg;
1853 port_cfg = shash_find_data(&new_ports, port->name);
1857 port_del_ifaces(port, port_cfg);
1861 /* Create new ports.
1862 * Add new interfaces to existing ports.
1863 * Reconfigure existing ports. */
1864 SHASH_FOR_EACH (node, &new_ports) {
1865 struct port *port = port_lookup(br, node->name);
1867 port = port_create(br, node->name);
1870 port_reconfigure(port, node->data);
1871 if (!port->n_ifaces) {
1872 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1873 br->name, port->name);
1877 shash_destroy(&new_ports);
1879 /* Set the fail-mode */
1880 fail_mode = !br->cfg->fail_mode
1881 || !strcmp(br->cfg->fail_mode, "standalone")
1882 ? OFPROTO_FAIL_STANDALONE
1883 : OFPROTO_FAIL_SECURE;
1884 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1885 && !ofproto_has_primary_controller(br->ofproto)) {
1886 ofproto_flush_flows(br->ofproto);
1888 ofproto_set_fail_mode(br->ofproto, fail_mode);
1890 /* Delete all flows if we're switching from connected to standalone or vice
1891 * versa. (XXX Should we delete all flows if we are switching from one
1892 * controller to another?) */
1894 /* Configure OpenFlow controller connection snooping. */
1896 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1897 ovs_rundir(), br->name));
1898 svec_init(&old_snoops);
1899 ofproto_get_snoops(br->ofproto, &old_snoops);
1900 if (!svec_equal(&snoops, &old_snoops)) {
1901 ofproto_set_snoops(br->ofproto, &snoops);
1903 svec_destroy(&snoops);
1904 svec_destroy(&old_snoops);
1906 mirror_reconfigure(br);
1909 /* Initializes 'oc' appropriately as a management service controller for
1912 * The caller must free oc->target when it is no longer needed. */
1914 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1915 struct ofproto_controller *oc)
1917 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1918 oc->max_backoff = 0;
1919 oc->probe_interval = 60;
1920 oc->band = OFPROTO_OUT_OF_BAND;
1922 oc->burst_limit = 0;
1925 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1927 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1928 struct ofproto_controller *oc)
1930 oc->target = c->target;
1931 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1932 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1933 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1934 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1935 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1936 oc->burst_limit = (c->controller_burst_limit
1937 ? *c->controller_burst_limit : 0);
1940 /* Configures the IP stack for 'br''s local interface properly according to the
1941 * configuration in 'c'. */
1943 bridge_configure_local_iface_netdev(struct bridge *br,
1944 struct ovsrec_controller *c)
1946 struct netdev *netdev;
1947 struct in_addr mask, gateway;
1949 struct iface *local_iface;
1952 /* If there's no local interface or no IP address, give up. */
1953 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
1954 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1958 /* Bring up the local interface. */
1959 netdev = local_iface->netdev;
1960 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1962 /* Configure the IP address and netmask. */
1963 if (!c->local_netmask
1964 || !inet_aton(c->local_netmask, &mask)
1966 mask.s_addr = guess_netmask(ip.s_addr);
1968 if (!netdev_set_in4(netdev, ip, mask)) {
1969 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1970 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1973 /* Configure the default gateway. */
1974 if (c->local_gateway
1975 && inet_aton(c->local_gateway, &gateway)
1976 && gateway.s_addr) {
1977 if (!netdev_add_router(netdev, gateway)) {
1978 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1979 br->name, IP_ARGS(&gateway.s_addr));
1985 bridge_reconfigure_remotes(struct bridge *br,
1986 const struct sockaddr_in *managers,
1989 const char *disable_ib_str, *queue_id_str;
1990 bool disable_in_band = false;
1993 struct ovsrec_controller **controllers;
1994 size_t n_controllers;
1997 struct ofproto_controller *ocs;
2001 /* Check if we should disable in-band control on this bridge. */
2002 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2003 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2004 disable_in_band = true;
2007 /* Set OpenFlow queue ID for in-band control. */
2008 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2009 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2010 ofproto_set_in_band_queue(br->ofproto, queue_id);
2012 if (disable_in_band) {
2013 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2015 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2017 had_primary = ofproto_has_primary_controller(br->ofproto);
2019 n_controllers = bridge_get_controllers(br, &controllers);
2021 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2024 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2025 for (i = 0; i < n_controllers; i++) {
2026 struct ovsrec_controller *c = controllers[i];
2028 if (!strncmp(c->target, "punix:", 6)
2029 || !strncmp(c->target, "unix:", 5)) {
2030 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2032 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2033 * domain sockets and overwriting arbitrary local files. */
2034 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2035 "\"%s\" due to possibility for remote exploit",
2036 dpif_name(br->dpif), c->target);
2040 bridge_configure_local_iface_netdev(br, c);
2041 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2042 if (disable_in_band) {
2043 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2048 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2049 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2052 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2053 ofproto_flush_flows(br->ofproto);
2056 /* If there are no controllers and the bridge is in standalone
2057 * mode, set up a flow that matches every packet and directs
2058 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2059 * switch is in secure mode and we won't pass any traffic until
2060 * a controller has been defined and it tells us to do so. */
2062 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2063 union ofp_action action;
2064 struct cls_rule rule;
2066 memset(&action, 0, sizeof action);
2067 action.type = htons(OFPAT_OUTPUT);
2068 action.output.len = htons(sizeof action);
2069 action.output.port = htons(OFPP_NORMAL);
2070 cls_rule_init_catchall(&rule, 0);
2071 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2076 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2081 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2082 struct iface *iface;
2084 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2085 shash_add_once(ifaces, iface->name, iface);
2087 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2088 shash_add_once(ifaces, port->name, NULL);
2093 /* For robustness, in case the administrator moves around datapath ports behind
2094 * our back, we re-check all the datapath port numbers here.
2096 * This function will set the 'dp_ifidx' members of interfaces that have
2097 * disappeared to -1, so only call this function from a context where those
2098 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2099 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2100 * datapath, which doesn't support UINT16_MAX+1 ports. */
2102 bridge_fetch_dp_ifaces(struct bridge *br)
2104 struct dpif_port_dump dump;
2105 struct dpif_port dpif_port;
2108 /* Reset all interface numbers. */
2109 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2110 struct iface *iface;
2112 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2113 iface->dp_ifidx = -1;
2116 hmap_clear(&br->ifaces);
2118 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2119 struct iface *iface = iface_lookup(br, dpif_port.name);
2121 if (iface->dp_ifidx >= 0) {
2122 VLOG_WARN("%s reported interface %s twice",
2123 dpif_name(br->dpif), dpif_port.name);
2124 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2125 VLOG_WARN("%s reported interface %"PRIu16" twice",
2126 dpif_name(br->dpif), dpif_port.port_no);
2128 iface->dp_ifidx = dpif_port.port_no;
2129 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2130 hash_int(iface->dp_ifidx, 0));
2133 iface_set_ofport(iface->cfg,
2134 (iface->dp_ifidx >= 0
2135 ? odp_port_to_ofp_port(iface->dp_ifidx)
2141 /* Bridge packet processing functions. */
2144 bond_is_tcp_hash(const struct port *port)
2146 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2150 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2152 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2155 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2157 struct flow hash_flow;
2159 memcpy(&hash_flow, flow, sizeof hash_flow);
2160 hash_flow.vlan_tci = 0;
2162 /* The symmetric quality of this hash function is not required, but
2163 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2164 * purposes, so we use it out of convenience. */
2165 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2168 static struct bond_entry *
2169 lookup_bond_entry(const struct port *port, const struct flow *flow,
2172 assert(port->bond_mode != BM_AB);
2174 if (bond_is_tcp_hash(port)) {
2175 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2177 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2181 static struct iface *
2182 bond_choose_iface(const struct port *port)
2184 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2185 struct iface *best_down_slave;
2186 struct iface *iface;
2188 best_down_slave = NULL;
2189 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2190 if (iface->enabled) {
2192 } else if ((!best_down_slave
2193 || iface->delay_expires < best_down_slave->delay_expires)
2194 && lacp_slave_may_enable(port->lacp, iface)) {
2195 best_down_slave = iface;
2199 if (best_down_slave) {
2200 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2201 "since no other interface is up",
2202 best_down_slave->name,
2203 best_down_slave->delay_expires - time_msec());
2204 bond_enable_slave(best_down_slave, true);
2207 return best_down_slave;
2211 choose_output_iface(const struct port *port, const struct flow *flow,
2212 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2214 struct iface *iface;
2216 assert(port->n_ifaces);
2217 if (port->n_ifaces == 1) {
2218 iface = port_get_an_iface(port);
2219 } else if (port->bond_mode == BM_AB) {
2220 iface = port->active_iface;
2222 *tags |= port->no_ifaces_tag;
2226 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2227 if (!e->iface || !e->iface->enabled) {
2228 /* XXX select interface properly. The current interface selection
2229 * is only good for testing the rebalancing code. */
2230 e->iface = bond_choose_iface(port);
2232 *tags |= port->no_ifaces_tag;
2235 e->tag = tag_create_random();
2240 *dp_ifidx = iface->dp_ifidx;
2241 *tags |= iface->tag; /* Currently only used for bonding. */
2246 bond_link_status_update(struct iface *iface)
2248 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2249 struct port *port = iface->port;
2250 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2251 int updelay, downdelay;
2253 updelay = port->updelay;
2254 downdelay = port->downdelay;
2256 if (lacp_negotiated(port->lacp)) {
2261 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2262 /* Nothing to do. */
2265 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2266 iface->name, up ? "up" : "down");
2267 if (up == iface->enabled) {
2268 iface->delay_expires = LLONG_MAX;
2269 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2270 iface->name, up ? "disabled" : "enabled");
2271 } else if (up && !port->active_iface) {
2272 bond_enable_slave(iface, true);
2274 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2275 "other interface is up", iface->name, updelay);
2278 int delay = up ? updelay : downdelay;
2279 iface->delay_expires = time_msec() + delay;
2282 "interface %s: will be %s if it stays %s for %d ms",
2284 up ? "enabled" : "disabled",
2292 bond_choose_active_iface(struct port *port)
2294 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2296 port->active_iface = bond_choose_iface(port);
2297 if (port->active_iface) {
2298 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2299 port->name, port->active_iface->name);
2301 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2307 bond_enable_slave(struct iface *iface, bool enable)
2309 struct port *port = iface->port;
2310 struct bridge *br = port->bridge;
2312 /* This acts as a recursion check. If the act of disabling a slave
2313 * causes a different slave to be enabled, the flag will allow us to
2314 * skip redundant work when we reenter this function. It must be
2315 * cleared on exit to keep things safe with multiple bonds. */
2316 static bool moving_active_iface = false;
2318 iface->delay_expires = LLONG_MAX;
2319 if (enable == iface->enabled) {
2323 iface->enabled = enable;
2324 if (!iface->enabled) {
2325 VLOG_WARN("interface %s: disabled", iface->name);
2326 ofproto_revalidate(br->ofproto, iface->tag);
2327 if (iface == port->active_iface) {
2328 /* Disabling a slave can lead to another slave being immediately
2329 * enabled if there will be no active slaves but one is waiting
2330 * on an updelay. In this case we do not need to run most of the
2331 * code for the newly enabled slave since there was no period
2332 * without an active slave and it is redundant with the disabling
2334 moving_active_iface = true;
2335 bond_choose_active_iface(port);
2337 bond_send_learning_packets(port);
2339 VLOG_WARN("interface %s: enabled", iface->name);
2340 if (!port->active_iface && !moving_active_iface) {
2341 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2342 bond_choose_active_iface(port);
2343 bond_send_learning_packets(port);
2345 iface->tag = tag_create_random();
2348 moving_active_iface = false;
2351 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2352 * bond interface. */
2354 bond_update_fake_iface_stats(struct port *port)
2356 struct netdev_stats bond_stats;
2357 struct netdev *bond_dev;
2358 struct iface *iface;
2360 memset(&bond_stats, 0, sizeof bond_stats);
2362 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2363 struct netdev_stats slave_stats;
2365 if (!netdev_get_stats(iface->netdev, &slave_stats)) {
2366 /* XXX: We swap the stats here because they are swapped back when
2367 * reported by the internal device. The reason for this is
2368 * internal devices normally represent packets going into the system
2369 * but when used as fake bond device they represent packets leaving
2370 * the system. We really should do this in the internal device
2371 * itself because changing it here reverses the counts from the
2372 * perspective of the switch. However, the internal device doesn't
2373 * know what type of device it represents so we have to do it here
2375 bond_stats.tx_packets += slave_stats.rx_packets;
2376 bond_stats.tx_bytes += slave_stats.rx_bytes;
2377 bond_stats.rx_packets += slave_stats.tx_packets;
2378 bond_stats.rx_bytes += slave_stats.tx_bytes;
2382 if (!netdev_open_default(port->name, &bond_dev)) {
2383 netdev_set_stats(bond_dev, &bond_stats);
2384 netdev_close(bond_dev);
2389 bond_run(struct port *port)
2391 struct iface *iface;
2393 if (port->n_ifaces < 2) {
2397 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2398 bond_link_status_update(iface);
2401 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2402 if (time_msec() >= iface->delay_expires) {
2403 bond_enable_slave(iface, !iface->enabled);
2407 if (port->bond_fake_iface
2408 && time_msec() >= port->bond_next_fake_iface_update) {
2409 bond_update_fake_iface_stats(port);
2410 port->bond_next_fake_iface_update = time_msec() + 1000;
2415 bond_wait(struct port *port)
2417 struct iface *iface;
2419 if (port->n_ifaces < 2) {
2423 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2424 if (iface->delay_expires != LLONG_MAX) {
2425 poll_timer_wait_until(iface->delay_expires);
2429 if (port->bond_fake_iface) {
2430 poll_timer_wait_until(port->bond_next_fake_iface_update);
2435 set_dst(struct dst *dst, const struct flow *flow,
2436 const struct port *in_port, const struct port *out_port,
2439 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2440 : in_port->vlan >= 0 ? in_port->vlan
2441 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2442 : vlan_tci_to_vid(flow->vlan_tci));
2443 return choose_output_iface(out_port, flow, dst->vlan,
2444 &dst->dp_ifidx, tags);
2448 swap_dst(struct dst *p, struct dst *q)
2450 struct dst tmp = *p;
2455 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2456 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2457 * that we push to the datapath. We could in fact fully sort the array by
2458 * vlan, but in most cases there are at most two different vlan tags so that's
2459 * possibly overkill.) */
2461 partition_dsts(struct dst_set *set, int vlan)
2463 struct dst *first = set->dsts;
2464 struct dst *last = set->dsts + set->n;
2466 while (first != last) {
2468 * - All dsts < first have vlan == 'vlan'.
2469 * - All dsts >= last have vlan != 'vlan'.
2470 * - first < last. */
2471 while (first->vlan == vlan) {
2472 if (++first == last) {
2477 /* Same invariants, plus one additional:
2478 * - first->vlan != vlan.
2480 while (last[-1].vlan != vlan) {
2481 if (--last == first) {
2486 /* Same invariants, plus one additional:
2487 * - last[-1].vlan == vlan.*/
2488 swap_dst(first++, --last);
2493 mirror_mask_ffs(mirror_mask_t mask)
2495 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2500 dst_set_init(struct dst_set *set)
2502 set->dsts = set->builtin;
2504 set->allocated = ARRAY_SIZE(set->builtin);
2508 dst_set_add(struct dst_set *set, const struct dst *dst)
2510 if (set->n >= set->allocated) {
2511 size_t new_allocated;
2512 struct dst *new_dsts;
2514 new_allocated = set->allocated * 2;
2515 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2516 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2520 set->dsts = new_dsts;
2521 set->allocated = new_allocated;
2523 set->dsts[set->n++] = *dst;
2527 dst_set_free(struct dst_set *set)
2529 if (set->dsts != set->builtin) {
2535 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2538 for (i = 0; i < set->n; i++) {
2539 if (set->dsts[i].vlan == test->vlan
2540 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2548 port_trunks_vlan(const struct port *port, uint16_t vlan)
2550 return (port->vlan < 0
2551 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2555 port_includes_vlan(const struct port *port, uint16_t vlan)
2557 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2561 port_is_floodable(const struct port *port)
2563 struct iface *iface;
2565 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2566 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2574 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2575 * there is no active iface. */
2577 port_get_active_iface_tag(const struct port *port)
2579 return (port->active_iface
2580 ? port->active_iface->tag
2581 : port->no_ifaces_tag);
2584 /* Returns an arbitrary interface within 'port'.
2586 * 'port' must have at least one interface. */
2587 static struct iface *
2588 port_get_an_iface(const struct port *port)
2590 return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem);
2594 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2595 const struct port *in_port, const struct port *out_port,
2596 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2598 mirror_mask_t mirrors = in_port->src_mirrors;
2602 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2603 if (flow_vlan == 0) {
2604 flow_vlan = OFP_VLAN_NONE;
2607 if (out_port == FLOOD_PORT) {
2610 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2612 && port_is_floodable(port)
2613 && port_includes_vlan(port, vlan)
2614 && !port->is_mirror_output_port
2615 && set_dst(&dst, flow, in_port, port, tags)) {
2616 mirrors |= port->dst_mirrors;
2617 dst_set_add(set, &dst);
2620 *nf_output_iface = NF_OUT_FLOOD;
2621 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2622 dst_set_add(set, &dst);
2623 *nf_output_iface = dst.dp_ifidx;
2624 mirrors |= out_port->dst_mirrors;
2628 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2629 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2631 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2632 && !dst_is_duplicate(set, &dst)) {
2633 dst_set_add(set, &dst);
2638 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2639 if (port_includes_vlan(port, m->out_vlan)
2640 && set_dst(&dst, flow, in_port, port, tags))
2642 if (port->vlan < 0) {
2643 dst.vlan = m->out_vlan;
2645 if (dst_is_duplicate(set, &dst)) {
2649 /* Use the vlan tag on the original flow instead of
2650 * the one passed in the vlan parameter. This ensures
2651 * that we compare the vlan from before any implicit
2652 * tagging tags place. This is necessary because
2653 * dst->vlan is the final vlan, after removing implicit
2655 if (port == in_port && dst.vlan == flow_vlan) {
2656 /* Don't send out input port on same VLAN. */
2659 dst_set_add(set, &dst);
2664 mirrors &= mirrors - 1;
2667 partition_dsts(set, flow_vlan);
2670 static void OVS_UNUSED
2671 print_dsts(const struct dst_set *set)
2675 for (i = 0; i < set->n; i++) {
2676 const struct dst *dst = &set->dsts[i];
2678 printf(">p%"PRIu16, dst->dp_ifidx);
2679 if (dst->vlan != OFP_VLAN_NONE) {
2680 printf("v%"PRIu16, dst->vlan);
2686 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2687 const struct port *in_port, const struct port *out_port,
2688 tag_type *tags, struct ofpbuf *actions,
2689 uint16_t *nf_output_iface)
2696 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2699 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2700 if (cur_vlan == 0) {
2701 cur_vlan = OFP_VLAN_NONE;
2703 for (i = 0; i < set.n; i++) {
2704 const struct dst *dst = &set.dsts[i];
2705 if (dst->vlan != cur_vlan) {
2706 if (dst->vlan == OFP_VLAN_NONE) {
2707 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2710 tci = htons(dst->vlan & VLAN_VID_MASK);
2711 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2712 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2714 cur_vlan = dst->vlan;
2716 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2721 /* Returns the effective vlan of a packet, taking into account both the
2722 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2723 * the packet is untagged and -1 indicates it has an invalid header and
2724 * should be dropped. */
2725 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2726 struct port *in_port, bool have_packet)
2728 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2729 if (in_port->vlan >= 0) {
2731 /* XXX support double tagging? */
2733 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2734 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2735 "packet received on port %s configured with "
2736 "implicit VLAN %"PRIu16,
2737 br->name, vlan, in_port->name, in_port->vlan);
2741 vlan = in_port->vlan;
2743 if (!port_includes_vlan(in_port, vlan)) {
2745 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2746 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2747 "packet received on port %s not configured for "
2749 br->name, vlan, in_port->name, vlan);
2758 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2759 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2760 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2762 is_gratuitous_arp(const struct flow *flow)
2764 return (flow->dl_type == htons(ETH_TYPE_ARP)
2765 && eth_addr_is_broadcast(flow->dl_dst)
2766 && (flow->nw_proto == ARP_OP_REPLY
2767 || (flow->nw_proto == ARP_OP_REQUEST
2768 && flow->nw_src == flow->nw_dst)));
2772 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2773 struct port *in_port)
2775 struct mac_entry *mac;
2777 if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) {
2781 mac = mac_learning_insert(br->ml, flow->dl_src, vlan);
2782 if (is_gratuitous_arp(flow)) {
2783 /* We don't want to learn from gratuitous ARP packets that are
2784 * reflected back over bond slaves so we lock the learning table. */
2785 if (in_port->n_ifaces == 1) {
2786 mac_entry_set_grat_arp_lock(mac);
2787 } else if (mac_entry_is_grat_arp_locked(mac)) {
2792 if (mac_entry_is_new(mac) || mac->port.p != in_port) {
2793 /* The log messages here could actually be useful in debugging,
2794 * so keep the rate limit relatively high. */
2795 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
2796 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2797 "on port %s in VLAN %d",
2798 br->name, ETH_ADDR_ARGS(flow->dl_src),
2799 in_port->name, vlan);
2801 mac->port.p = in_port;
2802 ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac));
2806 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2807 * dropped. Returns true if they may be forwarded, false if they should be
2810 * If 'have_packet' is true, it indicates that the caller is processing a
2811 * received packet. If 'have_packet' is false, then the caller is just
2812 * revalidating an existing flow because configuration has changed. Either
2813 * way, 'have_packet' only affects logging (there is no point in logging errors
2814 * during revalidation).
2816 * Sets '*in_portp' to the input port. This will be a null pointer if
2817 * flow->in_port does not designate a known input port (in which case
2818 * is_admissible() returns false).
2820 * When returning true, sets '*vlanp' to the effective VLAN of the input
2821 * packet, as returned by flow_get_vlan().
2823 * May also add tags to '*tags', although the current implementation only does
2824 * so in one special case.
2827 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2828 tag_type *tags, int *vlanp, struct port **in_portp)
2830 struct iface *in_iface;
2831 struct port *in_port;
2834 /* Find the interface and port structure for the received packet. */
2835 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2837 /* No interface? Something fishy... */
2839 /* Odd. A few possible reasons here:
2841 * - We deleted an interface but there are still a few packets
2842 * queued up from it.
2844 * - Someone externally added an interface (e.g. with "ovs-dpctl
2845 * add-if") that we don't know about.
2847 * - Packet arrived on the local port but the local port is not
2848 * one of our bridge ports.
2850 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2852 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2853 "interface %"PRIu16, br->name, flow->in_port);
2859 *in_portp = in_port = in_iface->port;
2860 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2865 /* Drop frames for reserved multicast addresses. */
2866 if (eth_addr_is_reserved(flow->dl_dst)) {
2870 /* Drop frames on ports reserved for mirroring. */
2871 if (in_port->is_mirror_output_port) {
2873 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2874 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2875 "%s, which is reserved exclusively for mirroring",
2876 br->name, in_port->name);
2881 /* When using LACP, do not accept packets from disabled interfaces. */
2882 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2886 /* Packets received on non-LACP bonds need special attention to avoid
2888 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2889 struct mac_entry *mac;
2891 if (eth_addr_is_multicast(flow->dl_dst)) {
2892 *tags |= port_get_active_iface_tag(in_port);
2893 if (in_port->active_iface != in_iface) {
2894 /* Drop all multicast packets on inactive slaves. */
2899 /* Drop all packets for which we have learned a different input
2900 * port, because we probably sent the packet on one slave and got
2901 * it back on the other. Gratuitous ARP packets are an exception
2902 * to this rule: the host has moved to another switch. The exception
2903 * to the exception is if we locked the learning table to avoid
2904 * reflections on bond slaves. If this is the case, just drop the
2906 mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL);
2907 if (mac && mac->port.p != in_port &&
2908 (!is_gratuitous_arp(flow) || mac_entry_is_grat_arp_locked(mac))) {
2916 /* If the composed actions may be applied to any packet in the given 'flow',
2917 * returns true. Otherwise, the actions should only be applied to 'packet', or
2918 * not at all, if 'packet' was NULL. */
2920 process_flow(struct bridge *br, const struct flow *flow,
2921 const struct ofpbuf *packet, struct ofpbuf *actions,
2922 tag_type *tags, uint16_t *nf_output_iface)
2924 struct port *in_port;
2925 struct port *out_port;
2926 struct mac_entry *mac;
2929 /* Check whether we should drop packets in this flow. */
2930 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2935 /* Learn source MAC (but don't try to learn from revalidation). */
2937 update_learning_table(br, flow, vlan, in_port);
2940 /* Determine output port. */
2941 mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags);
2943 out_port = mac->port.p;
2944 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2945 /* If we are revalidating but don't have a learning entry then
2946 * eject the flow. Installing a flow that floods packets opens
2947 * up a window of time where we could learn from a packet reflected
2948 * on a bond and blackhole packets before the learning table is
2949 * updated to reflect the correct port. */
2952 out_port = FLOOD_PORT;
2955 /* Don't send packets out their input ports. */
2956 if (in_port == out_port) {
2962 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2970 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2971 struct ofpbuf *actions, tag_type *tags,
2972 uint16_t *nf_output_iface, void *br_)
2974 struct bridge *br = br_;
2976 COVERAGE_INC(bridge_process_flow);
2977 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2981 bridge_special_ofhook_cb(const struct flow *flow,
2982 const struct ofpbuf *packet, void *br_)
2984 struct iface *iface;
2985 struct bridge *br = br_;
2987 iface = iface_from_dp_ifidx(br, flow->in_port);
2989 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
2991 if (iface && iface->port->lacp && packet) {
2992 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
2995 COVERAGE_INC(bridge_process_lacp);
2996 lacp_process_pdu(iface->port->lacp, iface, pdu);
3006 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3007 const struct nlattr *actions,
3009 uint64_t n_bytes, void *br_)
3011 struct bridge *br = br_;
3012 const struct nlattr *a;
3013 struct port *in_port;
3018 /* Feed information from the active flows back into the learning table to
3019 * ensure that table is always in sync with what is actually flowing
3020 * through the datapath.
3022 * We test that 'tags' is nonzero to ensure that only flows that include an
3023 * OFPP_NORMAL action are used for learning. This works because
3024 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3025 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3026 update_learning_table(br, flow, vlan, in_port);
3029 /* Account for bond slave utilization. */
3030 if (!br->has_bonded_ports) {
3033 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3034 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3035 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3036 if (out_port && out_port->n_ifaces >= 2 &&
3037 out_port->bond_mode != BM_AB) {
3038 uint16_t vlan = (flow->vlan_tci
3039 ? vlan_tci_to_vid(flow->vlan_tci)
3041 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3042 e->tx_bytes += n_bytes;
3049 bridge_account_checkpoint_ofhook_cb(void *br_)
3051 struct bridge *br = br_;
3055 if (!br->has_bonded_ports) {
3060 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3061 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3062 && now >= port->bond_next_rebalance) {
3063 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3064 bond_rebalance_port(port);
3069 static struct ofhooks bridge_ofhooks = {
3070 bridge_normal_ofhook_cb,
3071 bridge_special_ofhook_cb,
3072 bridge_account_flow_ofhook_cb,
3073 bridge_account_checkpoint_ofhook_cb,
3076 /* Bonding functions. */
3078 /* Statistics for a single interface on a bonded port, used for load-based
3079 * bond rebalancing. */
3080 struct slave_balance {
3081 struct iface *iface; /* The interface. */
3082 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3084 /* All the "bond_entry"s that are assigned to this interface, in order of
3085 * increasing tx_bytes. */
3086 struct bond_entry **hashes;
3091 bond_mode_to_string(enum bond_mode bm) {
3092 static char *bm_slb = "balance-slb";
3093 static char *bm_ab = "active-backup";
3094 static char *bm_tcp = "balance-tcp";
3097 case BM_SLB: return bm_slb;
3098 case BM_AB: return bm_ab;
3099 case BM_TCP: return bm_tcp;
3106 /* Sorts pointers to pointers to bond_entries in ascending order by the
3107 * interface to which they are assigned, and within a single interface in
3108 * ascending order of bytes transmitted. */
3110 compare_bond_entries(const void *a_, const void *b_)
3112 const struct bond_entry *const *ap = a_;
3113 const struct bond_entry *const *bp = b_;
3114 const struct bond_entry *a = *ap;
3115 const struct bond_entry *b = *bp;
3116 if (a->iface != b->iface) {
3117 return a->iface > b->iface ? 1 : -1;
3118 } else if (a->tx_bytes != b->tx_bytes) {
3119 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3125 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3126 * *descending* order by number of bytes transmitted. */
3128 compare_slave_balance(const void *a_, const void *b_)
3130 const struct slave_balance *a = a_;
3131 const struct slave_balance *b = b_;
3132 if (a->iface->enabled != b->iface->enabled) {
3133 return a->iface->enabled ? -1 : 1;
3134 } else if (a->tx_bytes != b->tx_bytes) {
3135 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3142 swap_bals(struct slave_balance *a, struct slave_balance *b)
3144 struct slave_balance tmp = *a;
3149 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3150 * given that 'p' (and only 'p') might be in the wrong location.
3152 * This function invalidates 'p', since it might now be in a different memory
3155 resort_bals(struct slave_balance *p,
3156 struct slave_balance bals[], size_t n_bals)
3159 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3160 swap_bals(p, p - 1);
3162 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3163 swap_bals(p, p + 1);
3169 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3171 if (VLOG_IS_DBG_ENABLED()) {
3172 struct ds ds = DS_EMPTY_INITIALIZER;
3173 const struct slave_balance *b;
3175 for (b = bals; b < bals + n_bals; b++) {
3179 ds_put_char(&ds, ',');
3181 ds_put_format(&ds, " %s %"PRIu64"kB",
3182 b->iface->name, b->tx_bytes / 1024);
3184 if (!b->iface->enabled) {
3185 ds_put_cstr(&ds, " (disabled)");
3187 if (b->n_hashes > 0) {
3188 ds_put_cstr(&ds, " (");
3189 for (i = 0; i < b->n_hashes; i++) {
3190 const struct bond_entry *e = b->hashes[i];
3192 ds_put_cstr(&ds, " + ");
3194 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3195 e - port->bond_hash, e->tx_bytes / 1024);
3197 ds_put_cstr(&ds, ")");
3200 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3205 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3207 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3210 struct bond_entry *hash = from->hashes[hash_idx];
3211 struct port *port = from->iface->port;
3212 uint64_t delta = hash->tx_bytes;
3214 assert(port->bond_mode != BM_AB);
3216 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3217 "from %s to %s (now carrying %"PRIu64"kB and "
3218 "%"PRIu64"kB load, respectively)",
3219 port->name, delta / 1024, hash - port->bond_hash,
3220 from->iface->name, to->iface->name,
3221 (from->tx_bytes - delta) / 1024,
3222 (to->tx_bytes + delta) / 1024);
3224 /* Delete element from from->hashes.
3226 * We don't bother to add the element to to->hashes because not only would
3227 * it require more work, the only purpose it would be to allow that hash to
3228 * be migrated to another slave in this rebalancing run, and there is no
3229 * point in doing that. */
3230 if (hash_idx == 0) {
3233 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3234 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3238 /* Shift load away from 'from' to 'to'. */
3239 from->tx_bytes -= delta;
3240 to->tx_bytes += delta;
3242 /* Arrange for flows to be revalidated. */
3243 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3244 hash->iface = to->iface;
3245 hash->tag = tag_create_random();
3249 bond_rebalance_port(struct port *port)
3251 struct slave_balance *bals;
3253 struct bond_entry *hashes[BOND_MASK + 1];
3254 struct slave_balance *b, *from, *to;
3255 struct bond_entry *e;
3256 struct iface *iface;
3259 assert(port->bond_mode != BM_AB);
3261 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3262 * descending order of tx_bytes, so that bals[0] represents the most
3263 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3266 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3267 * array for each slave_balance structure, we sort our local array of
3268 * hashes in order by slave, so that all of the hashes for a given slave
3269 * become contiguous in memory, and then we point each 'hashes' members of
3270 * a slave_balance structure to the start of a contiguous group. */
3271 n_bals = port->n_ifaces;
3272 b = bals = xmalloc(n_bals * sizeof *bals);
3273 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3280 assert(b == &bals[n_bals]);
3281 for (i = 0; i <= BOND_MASK; i++) {
3282 hashes[i] = &port->bond_hash[i];
3284 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3285 for (i = 0; i <= BOND_MASK; i++) {
3291 for (b = bals; b < &bals[n_bals]; b++) {
3292 if (b->iface == e->iface) {
3293 b->tx_bytes += e->tx_bytes;
3295 b->hashes = &hashes[i];
3302 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3303 log_bals(bals, n_bals, port);
3305 /* Discard slaves that aren't enabled (which were sorted to the back of the
3306 * array earlier). */
3307 while (!bals[n_bals - 1].iface->enabled) {
3314 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3315 to = &bals[n_bals - 1];
3316 for (from = bals; from < to; ) {
3317 uint64_t overload = from->tx_bytes - to->tx_bytes;
3318 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3319 /* The extra load on 'from' (and all less-loaded slaves), compared
3320 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3321 * it is less than ~1Mbps. No point in rebalancing. */
3323 } else if (from->n_hashes == 1) {
3324 /* 'from' only carries a single MAC hash, so we can't shift any
3325 * load away from it, even though we want to. */
3328 /* 'from' is carrying significantly more load than 'to', and that
3329 * load is split across at least two different hashes. Pick a hash
3330 * to migrate to 'to' (the least-loaded slave), given that doing so
3331 * must decrease the ratio of the load on the two slaves by at
3334 * The sort order we use means that we prefer to shift away the
3335 * smallest hashes instead of the biggest ones. There is little
3336 * reason behind this decision; we could use the opposite sort
3337 * order to shift away big hashes ahead of small ones. */
3340 for (i = 0; i < from->n_hashes; i++) {
3341 double old_ratio, new_ratio;
3342 uint64_t delta = from->hashes[i]->tx_bytes;
3344 if (delta == 0 || from->tx_bytes - delta == 0) {
3345 /* Pointless move. */
3349 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3351 if (to->tx_bytes == 0) {
3352 /* Nothing on the new slave, move it. */
3356 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3357 new_ratio = (double)(from->tx_bytes - delta) /
3358 (to->tx_bytes + delta);
3360 if (new_ratio == 0) {
3361 /* Should already be covered but check to prevent division
3366 if (new_ratio < 1) {
3367 new_ratio = 1 / new_ratio;
3370 if (old_ratio - new_ratio > 0.1) {
3371 /* Would decrease the ratio, move it. */
3375 if (i < from->n_hashes) {
3376 bond_shift_load(from, to, i);
3378 /* If the result of the migration changed the relative order of
3379 * 'from' and 'to' swap them back to maintain invariants. */
3380 if (order_swapped) {
3381 swap_bals(from, to);
3384 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3385 * point to different slave_balance structures. It is only
3386 * valid to do these two operations in a row at all because we
3387 * know that 'from' will not move past 'to' and vice versa. */
3388 resort_bals(from, bals, n_bals);
3389 resort_bals(to, bals, n_bals);
3396 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3397 * historical data to decay to <1% in 7 rebalancing runs. */
3398 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3410 bond_send_learning_packets(struct port *port)
3412 struct bridge *br = port->bridge;
3413 struct mac_entry *e;
3414 struct ofpbuf packet;
3415 int error, n_packets, n_errors;
3417 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3421 ofpbuf_init(&packet, 128);
3422 error = n_packets = n_errors = 0;
3423 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3429 if (e->port.p == port) {
3433 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3435 flow_extract(&packet, 0, ODPP_NONE, &flow);
3437 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3443 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3449 ofpbuf_uninit(&packet);
3452 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3453 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3454 "packets, last error was: %s",
3455 port->name, n_errors, n_packets, strerror(error));
3457 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3458 port->name, n_packets);
3462 /* Bonding unixctl user interface functions. */
3465 bond_unixctl_list(struct unixctl_conn *conn,
3466 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3468 struct ds ds = DS_EMPTY_INITIALIZER;
3469 const struct bridge *br;
3471 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3473 LIST_FOR_EACH (br, node, &all_bridges) {
3476 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3477 if (port->n_ifaces > 1) {
3478 struct iface *iface;
3480 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3481 bond_mode_to_string(port->bond_mode));
3482 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3483 if (&iface->port_elem != list_front(&port->ifaces)) {
3484 ds_put_cstr(&ds, ", ");
3486 ds_put_cstr(&ds, iface->name);
3488 ds_put_char(&ds, '\n');
3492 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3496 static struct port *
3497 bond_find(const char *name)
3499 const struct bridge *br;
3501 LIST_FOR_EACH (br, node, &all_bridges) {
3504 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3505 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3514 bond_unixctl_show(struct unixctl_conn *conn,
3515 const char *args, void *aux OVS_UNUSED)
3517 struct ds ds = DS_EMPTY_INITIALIZER;
3518 const struct port *port;
3519 struct iface *iface;
3521 port = bond_find(args);
3523 unixctl_command_reply(conn, 501, "no such bond");
3527 ds_put_format(&ds, "bond_mode: %s\n",
3528 bond_mode_to_string(port->bond_mode));
3531 ds_put_format(&ds, "lacp: %s\n",
3532 port->lacp_active ? "active" : "passive");
3534 ds_put_cstr(&ds, "lacp: off\n");
3537 if (port->bond_mode != BM_AB) {
3538 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3539 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3543 ds_put_format(&ds, "bond-detect-mode: %s\n",
3544 port->monitor ? "carrier" : "miimon");
3546 if (!port->monitor) {
3547 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3548 port->miimon_interval);
3551 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3552 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3554 if (port->bond_mode != BM_AB) {
3555 ds_put_format(&ds, "next rebalance: %lld ms\n",
3556 port->bond_next_rebalance - time_msec());
3559 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3560 struct bond_entry *be;
3564 ds_put_format(&ds, "\nslave %s: %s\n",
3565 iface->name, iface->enabled ? "enabled" : "disabled");
3566 if (iface == port->active_iface) {
3567 ds_put_cstr(&ds, "\tactive slave\n");
3569 if (iface->delay_expires != LLONG_MAX) {
3570 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3571 iface->enabled ? "downdelay" : "updelay",
3572 iface->delay_expires - time_msec());
3575 if (port->bond_mode == BM_AB) {
3580 memset(&flow, 0, sizeof flow);
3581 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3582 int hash = be - port->bond_hash;
3583 struct mac_entry *me;
3585 if (be->iface != iface) {
3589 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3590 hash, be->tx_bytes / 1024);
3592 if (port->bond_mode != BM_SLB) {
3597 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3601 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3602 if (bond_hash_src(me->mac, me->vlan) == hash
3603 && me->port.p != port
3604 && choose_output_iface(port, &flow, me->vlan,
3606 && dp_ifidx == iface->dp_ifidx)
3608 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3609 ETH_ADDR_ARGS(me->mac));
3614 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3619 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3620 void *aux OVS_UNUSED)
3622 char *args = (char *) args_;
3623 char *save_ptr = NULL;
3624 char *bond_s, *hash_s, *slave_s;
3626 struct iface *iface;
3627 struct bond_entry *entry;
3630 bond_s = strtok_r(args, " ", &save_ptr);
3631 hash_s = strtok_r(NULL, " ", &save_ptr);
3632 slave_s = strtok_r(NULL, " ", &save_ptr);
3634 unixctl_command_reply(conn, 501,
3635 "usage: bond/migrate BOND HASH SLAVE");
3639 port = bond_find(bond_s);
3641 unixctl_command_reply(conn, 501, "no such bond");
3645 if (port->bond_mode != BM_SLB) {
3646 unixctl_command_reply(conn, 501, "not an SLB bond");
3650 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3651 hash = atoi(hash_s) & BOND_MASK;
3653 unixctl_command_reply(conn, 501, "bad hash");
3657 iface = port_lookup_iface(port, slave_s);
3659 unixctl_command_reply(conn, 501, "no such slave");
3663 if (!iface->enabled) {
3664 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3668 entry = &port->bond_hash[hash];
3669 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3670 entry->iface = iface;
3671 entry->tag = tag_create_random();
3672 unixctl_command_reply(conn, 200, "migrated");
3676 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3677 void *aux OVS_UNUSED)
3679 char *args = (char *) args_;
3680 char *save_ptr = NULL;
3681 char *bond_s, *slave_s;
3683 struct iface *iface;
3685 bond_s = strtok_r(args, " ", &save_ptr);
3686 slave_s = strtok_r(NULL, " ", &save_ptr);
3688 unixctl_command_reply(conn, 501,
3689 "usage: bond/set-active-slave BOND SLAVE");
3693 port = bond_find(bond_s);
3695 unixctl_command_reply(conn, 501, "no such bond");
3699 iface = port_lookup_iface(port, slave_s);
3701 unixctl_command_reply(conn, 501, "no such slave");
3705 if (!iface->enabled) {
3706 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3710 if (port->active_iface != iface) {
3711 ofproto_revalidate(port->bridge->ofproto,
3712 port_get_active_iface_tag(port));
3713 port->active_iface = iface;
3714 VLOG_INFO("port %s: active interface is now %s",
3715 port->name, iface->name);
3716 bond_send_learning_packets(port);
3717 unixctl_command_reply(conn, 200, "done");
3719 unixctl_command_reply(conn, 200, "no change");
3724 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3726 char *args = (char *) args_;
3727 char *save_ptr = NULL;
3728 char *bond_s, *slave_s;
3730 struct iface *iface;
3732 bond_s = strtok_r(args, " ", &save_ptr);
3733 slave_s = strtok_r(NULL, " ", &save_ptr);
3735 unixctl_command_reply(conn, 501,
3736 "usage: bond/enable/disable-slave BOND SLAVE");
3740 port = bond_find(bond_s);
3742 unixctl_command_reply(conn, 501, "no such bond");
3746 iface = port_lookup_iface(port, slave_s);
3748 unixctl_command_reply(conn, 501, "no such slave");
3752 bond_enable_slave(iface, enable);
3753 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3757 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3758 void *aux OVS_UNUSED)
3760 enable_slave(conn, args, true);
3764 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3765 void *aux OVS_UNUSED)
3767 enable_slave(conn, args, false);
3771 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3772 void *aux OVS_UNUSED)
3774 char *args = (char *) args_;
3775 uint8_t mac[ETH_ADDR_LEN];
3779 char *mac_s, *vlan_s;
3780 char *save_ptr = NULL;
3782 mac_s = strtok_r(args, " ", &save_ptr);
3783 vlan_s = strtok_r(NULL, " ", &save_ptr);
3786 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3787 unixctl_command_reply(conn, 501, "invalid vlan");
3791 vlan = OFP_VLAN_NONE;
3794 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3795 == ETH_ADDR_SCAN_COUNT) {
3796 hash = bond_hash_src(mac, vlan);
3798 hash_cstr = xasprintf("%u", hash);
3799 unixctl_command_reply(conn, 200, hash_cstr);
3802 unixctl_command_reply(conn, 501, "invalid mac");
3809 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3810 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3811 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3812 unixctl_command_register("bond/set-active-slave",
3813 bond_unixctl_set_active_slave, NULL);
3814 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3816 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3818 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3821 /* Port functions. */
3824 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3826 struct iface *iface = aux;
3827 uint8_t ea[ETH_ADDR_LEN];
3830 error = netdev_get_etheraddr(iface->netdev, ea);
3832 struct ofpbuf packet;
3833 struct lacp_pdu *packet_pdu;
3835 ofpbuf_init(&packet, 0);
3836 packet_pdu = compose_packet(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3837 sizeof *packet_pdu);
3838 memcpy(packet_pdu, pdu, sizeof *packet_pdu);
3839 ofproto_send_packet(iface->port->bridge->ofproto,
3840 iface->dp_ifidx, 0, &packet);
3841 ofpbuf_uninit(&packet);
3843 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3844 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3845 "(%s)", iface->name, strerror(error));
3850 port_run(struct port *port)
3852 if (port->monitor) {
3855 /* Track carrier going up and down on interfaces. */
3856 while (!netdev_monitor_poll(port->monitor, &devname)) {
3857 struct iface *iface;
3859 iface = port_lookup_iface(port, devname);
3861 iface_update_carrier(iface);
3865 } else if (time_msec() >= port->miimon_next_update) {
3866 struct iface *iface;
3868 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3869 iface_update_carrier(iface);
3871 port->miimon_next_update = time_msec() + port->miimon_interval;
3875 struct iface *iface;
3877 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3878 lacp_slave_enable(port->lacp, iface, iface->enabled);
3881 lacp_run(port->lacp, lacp_send_pdu_cb);
3888 port_wait(struct port *port)
3890 if (port->monitor) {
3891 netdev_monitor_poll_wait(port->monitor);
3893 poll_timer_wait_until(port->miimon_next_update);
3897 lacp_wait(port->lacp);
3903 static struct port *
3904 port_create(struct bridge *br, const char *name)
3908 port = xzalloc(sizeof *port);
3911 port->trunks = NULL;
3912 port->name = xstrdup(name);
3913 port->active_iface = NULL;
3914 list_init(&port->ifaces);
3916 hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0));
3918 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3925 get_port_other_config(const struct ovsrec_port *port, const char *key,
3926 const char *default_value)
3930 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3932 return value ? value : default_value;
3936 get_interface_other_config(const struct ovsrec_interface *iface,
3937 const char *key, const char *default_value)
3941 value = get_ovsrec_key_value(&iface->header_,
3942 &ovsrec_interface_col_other_config, key);
3943 return value ? value : default_value;
3947 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3949 struct iface *iface, *next;
3950 struct shash new_ifaces;
3953 /* Collect list of new interfaces. */
3954 shash_init(&new_ifaces);
3955 for (i = 0; i < cfg->n_interfaces; i++) {
3956 const char *name = cfg->interfaces[i]->name;
3957 shash_add_once(&new_ifaces, name, NULL);
3960 /* Get rid of deleted interfaces. */
3961 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
3962 if (!shash_find(&new_ifaces, iface->name)) {
3963 iface_destroy(iface);
3967 shash_destroy(&new_ifaces);
3970 /* Expires all MAC learning entries associated with 'port' and forces ofproto
3971 * to revalidate every flow. */
3973 port_flush_macs(struct port *port)
3975 struct bridge *br = port->bridge;
3976 struct mac_learning *ml = br->ml;
3977 struct mac_entry *mac, *next_mac;
3980 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
3981 if (mac->port.p == port) {
3982 mac_learning_expire(ml, mac);
3988 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3990 const char *detect_mode;
3991 struct shash new_ifaces;
3992 long long int next_rebalance, miimon_next_update, lacp_priority;
3993 bool need_flush = false;
3994 unsigned long *trunks;
4000 /* Update settings. */
4001 port->updelay = cfg->bond_updelay;
4002 if (port->updelay < 0) {
4005 port->downdelay = cfg->bond_downdelay;
4006 if (port->downdelay < 0) {
4007 port->downdelay = 0;
4009 port->bond_rebalance_interval = atoi(
4010 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4011 if (port->bond_rebalance_interval < 1000) {
4012 port->bond_rebalance_interval = 1000;
4014 next_rebalance = time_msec() + port->bond_rebalance_interval;
4015 if (port->bond_next_rebalance > next_rebalance) {
4016 port->bond_next_rebalance = next_rebalance;
4019 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4022 netdev_monitor_destroy(port->monitor);
4023 port->monitor = NULL;
4025 if (strcmp(detect_mode, "miimon")) {
4026 port->monitor = netdev_monitor_create();
4028 if (strcmp(detect_mode, "carrier")) {
4029 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4030 "defaulting to carrier", port->name, detect_mode);
4034 port->miimon_interval = atoi(
4035 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4036 if (port->miimon_interval < 100) {
4037 port->miimon_interval = 100;
4039 miimon_next_update = time_msec() + port->miimon_interval;
4040 if (port->miimon_next_update > miimon_next_update) {
4041 port->miimon_next_update = miimon_next_update;
4044 if (!port->cfg->bond_mode ||
4045 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4046 port->bond_mode = BM_SLB;
4047 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4048 port->bond_mode = BM_AB;
4049 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4050 port->bond_mode = BM_TCP;
4052 port->bond_mode = BM_SLB;
4053 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4054 port->name, port->cfg->bond_mode,
4055 bond_mode_to_string(port->bond_mode));
4058 /* Add new interfaces and update 'cfg' member of existing ones. */
4059 shash_init(&new_ifaces);
4060 for (i = 0; i < cfg->n_interfaces; i++) {
4061 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4062 struct iface *iface;
4064 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4065 VLOG_WARN("port %s: %s specified twice as port interface",
4066 port->name, if_cfg->name);
4067 iface_set_ofport(if_cfg, -1);
4071 iface = iface_lookup(port->bridge, if_cfg->name);
4073 if (iface->port != port) {
4074 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4076 port->bridge->name, if_cfg->name, iface->port->name);
4079 iface->cfg = if_cfg;
4081 iface = iface_create(port, if_cfg);
4084 /* Determine interface type. The local port always has type
4085 * "internal". Other ports take their type from the database and
4086 * default to "system" if none is specified. */
4087 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4088 : if_cfg->type[0] ? if_cfg->type
4092 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4095 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4096 iface->lacp_priority = UINT16_MAX;
4098 iface->lacp_priority = lacp_priority;
4101 shash_destroy(&new_ifaces);
4103 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4107 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4109 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4110 /* Prefer bondable links if unspecified. */
4111 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4113 port->lacp_priority = lacp_priority;
4116 if (!port->cfg->lacp) {
4117 /* XXX when LACP implementation has been sufficiently tested, enable by
4118 * default and make active on bonded ports. */
4119 lacp_destroy(port->lacp);
4121 } else if (!strcmp(port->cfg->lacp, "off")) {
4122 lacp_destroy(port->lacp);
4124 } else if (!strcmp(port->cfg->lacp, "active")) {
4126 port->lacp = lacp_create();
4128 port->lacp_active = true;
4129 } else if (!strcmp(port->cfg->lacp, "passive")) {
4131 port->lacp = lacp_create();
4133 port->lacp_active = false;
4135 VLOG_WARN("port %s: unknown LACP mode %s",
4136 port->name, port->cfg->lacp);
4137 lacp_destroy(port->lacp);
4144 if (port->n_ifaces < 2) {
4146 if (vlan >= 0 && vlan <= 4095) {
4147 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4152 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4153 * they even work as-is. But they have not been tested. */
4154 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4158 if (port->vlan != vlan) {
4163 /* Get trunked VLANs. */
4165 if (vlan < 0 && cfg->n_trunks) {
4168 trunks = bitmap_allocate(4096);
4170 for (i = 0; i < cfg->n_trunks; i++) {
4171 int trunk = cfg->trunks[i];
4173 bitmap_set1(trunks, trunk);
4179 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4180 port->name, cfg->n_trunks);
4182 if (n_errors == cfg->n_trunks) {
4183 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4185 bitmap_free(trunks);
4188 } else if (vlan >= 0 && cfg->n_trunks) {
4189 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4193 ? port->trunks != NULL
4194 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4197 bitmap_free(port->trunks);
4198 port->trunks = trunks;
4201 port_flush_macs(port);
4206 port_destroy(struct port *port)
4209 struct bridge *br = port->bridge;
4210 struct iface *iface, *next;
4213 for (i = 0; i < MAX_MIRRORS; i++) {
4214 struct mirror *m = br->mirrors[i];
4215 if (m && m->out_port == port) {
4220 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4221 iface_destroy(iface);
4224 hmap_remove(&br->ports, &port->hmap_node);
4226 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4228 port_flush_macs(port);
4230 lacp_destroy(port->lacp);
4231 netdev_monitor_destroy(port->monitor);
4232 bitmap_free(port->trunks);
4233 free(port->bond_hash);
4239 static struct port *
4240 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4242 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4243 return iface ? iface->port : NULL;
4246 static struct port *
4247 port_lookup(const struct bridge *br, const char *name)
4251 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0),
4253 if (!strcmp(port->name, name)) {
4260 static struct iface *
4261 port_lookup_iface(const struct port *port, const char *name)
4263 struct iface *iface = iface_lookup(port->bridge, name);
4264 return iface && iface->port == port ? iface : NULL;
4268 port_update_lacp(struct port *port)
4271 struct iface *iface;
4273 lacp_configure(port->lacp, port->name,
4274 port->bridge->ea, port->lacp_priority,
4275 port->lacp_active, port->lacp_fast);
4277 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
4278 lacp_slave_register(port->lacp, iface, iface->name,
4279 iface->dp_ifidx, iface->lacp_priority);
4285 port_update_bonding(struct port *port)
4287 if (port->n_ifaces < 2) {
4288 /* Not a bonded port. */
4289 free(port->bond_hash);
4290 port->bond_hash = NULL;
4291 port->bond_fake_iface = false;
4292 port->active_iface = NULL;
4293 port->no_ifaces_tag = 0;
4297 if (port->bond_mode != BM_AB && !port->bond_hash) {
4298 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4299 for (i = 0; i <= BOND_MASK; i++) {
4300 struct bond_entry *e = &port->bond_hash[i];
4304 port->bond_next_rebalance
4305 = time_msec() + port->bond_rebalance_interval;
4306 } else if (port->bond_mode == BM_AB) {
4307 free(port->bond_hash);
4308 port->bond_hash = NULL;
4311 if (!port->no_ifaces_tag) {
4312 port->no_ifaces_tag = tag_create_random();
4315 if (!port->active_iface) {
4316 bond_choose_active_iface(port);
4319 port->bond_fake_iface = port->cfg->bond_fake_iface;
4320 if (port->bond_fake_iface) {
4321 port->bond_next_fake_iface_update = time_msec();
4327 /* Interface functions. */
4329 static struct iface *
4330 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4332 struct bridge *br = port->bridge;
4333 struct iface *iface;
4334 char *name = if_cfg->name;
4336 iface = xzalloc(sizeof *iface);
4338 iface->name = xstrdup(name);
4339 iface->dp_ifidx = -1;
4340 iface->tag = tag_create_random();
4341 iface->delay_expires = LLONG_MAX;
4342 iface->netdev = NULL;
4343 iface->cfg = if_cfg;
4345 shash_add_assert(&br->iface_by_name, iface->name, iface);
4347 list_push_back(&port->ifaces, &iface->port_elem);
4350 if (port->n_ifaces > 1) {
4351 br->has_bonded_ports = true;
4354 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4362 iface_destroy(struct iface *iface)
4365 struct port *port = iface->port;
4366 struct bridge *br = port->bridge;
4367 bool del_active = port->active_iface == iface;
4369 if (port->bond_hash) {
4370 struct bond_entry *e;
4371 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4372 if (e->iface == iface) {
4378 if (iface->port->lacp) {
4379 lacp_slave_unregister(iface->port->lacp, iface);
4382 if (port->monitor && iface->netdev) {
4383 netdev_monitor_remove(port->monitor, iface->netdev);
4386 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4388 if (iface->dp_ifidx >= 0) {
4389 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4392 list_remove(&iface->port_elem);
4395 netdev_close(iface->netdev);
4398 bond_choose_active_iface(port);
4399 bond_send_learning_packets(port);
4405 bridge_flush(port->bridge);
4409 static struct iface *
4410 iface_lookup(const struct bridge *br, const char *name)
4412 return shash_find_data(&br->iface_by_name, name);
4415 static struct iface *
4416 iface_find(const char *name)
4418 const struct bridge *br;
4420 LIST_FOR_EACH (br, node, &all_bridges) {
4421 struct iface *iface = iface_lookup(br, name);
4430 static struct iface *
4431 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4433 struct iface *iface;
4435 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4436 hash_int(dp_ifidx, 0), &br->ifaces) {
4437 if (iface->dp_ifidx == dp_ifidx) {
4444 /* Set Ethernet address of 'iface', if one is specified in the configuration
4447 iface_set_mac(struct iface *iface)
4449 uint8_t ea[ETH_ADDR_LEN];
4451 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4452 if (eth_addr_is_multicast(ea)) {
4453 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4455 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4456 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4457 iface->name, iface->name);
4459 int error = netdev_set_etheraddr(iface->netdev, ea);
4461 VLOG_ERR("interface %s: setting MAC failed (%s)",
4462 iface->name, strerror(error));
4468 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4470 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4473 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4477 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4479 * The value strings in '*shash' are taken directly from values[], not copied,
4480 * so the caller should not modify or free them. */
4482 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4483 struct shash *shash)
4488 for (i = 0; i < n; i++) {
4489 shash_add(shash, keys[i], values[i]);
4493 /* Creates 'keys' and 'values' arrays from 'shash'.
4495 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4496 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4497 * are populated with with strings taken directly from 'shash' and thus have
4498 * the same ownership of the key-value pairs in shash.
4501 shash_to_ovs_idl_map(struct shash *shash,
4502 char ***keys, char ***values, size_t *n)
4506 struct shash_node *sn;
4508 count = shash_count(shash);
4510 k = xmalloc(count * sizeof *k);
4511 v = xmalloc(count * sizeof *v);
4514 SHASH_FOR_EACH(sn, shash) {
4525 struct iface_delete_queues_cbdata {
4526 struct netdev *netdev;
4527 const struct ovsdb_datum *queues;
4531 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4533 union ovsdb_atom atom;
4535 atom.integer = target;
4536 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4540 iface_delete_queues(unsigned int queue_id,
4541 const struct shash *details OVS_UNUSED, void *cbdata_)
4543 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4545 if (!queue_ids_include(cbdata->queues, queue_id)) {
4546 netdev_delete_queue(cbdata->netdev, queue_id);
4551 iface_update_carrier(struct iface *iface)
4553 bool carrier = iface_get_carrier(iface);
4554 if (carrier == iface->up) {
4558 iface->up = carrier;
4559 if (iface->port->lacp) {
4560 lacp_slave_carrier_changed(iface->port->lacp, iface);
4565 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4567 if (!qos || qos->type[0] == '\0') {
4568 netdev_set_qos(iface->netdev, NULL, NULL);
4570 struct iface_delete_queues_cbdata cbdata;
4571 struct shash details;
4574 /* Configure top-level Qos for 'iface'. */
4575 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4576 qos->n_other_config, &details);
4577 netdev_set_qos(iface->netdev, qos->type, &details);
4578 shash_destroy(&details);
4580 /* Deconfigure queues that were deleted. */
4581 cbdata.netdev = iface->netdev;
4582 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4584 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4586 /* Configure queues for 'iface'. */
4587 for (i = 0; i < qos->n_queues; i++) {
4588 const struct ovsrec_queue *queue = qos->value_queues[i];
4589 unsigned int queue_id = qos->key_queues[i];
4591 shash_from_ovs_idl_map(queue->key_other_config,
4592 queue->value_other_config,
4593 queue->n_other_config, &details);
4594 netdev_set_queue(iface->netdev, queue_id, &details);
4595 shash_destroy(&details);
4601 iface_update_cfm(struct iface *iface)
4605 uint16_t *remote_mps;
4606 struct ovsrec_monitor *mon;
4607 uint8_t maid[CCM_MAID_LEN];
4609 mon = iface->cfg->monitor;
4612 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4616 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4617 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4621 cfm.mpid = mon->mpid;
4622 cfm.interval = mon->interval ? *mon->interval : 1000;
4624 memcpy(cfm.maid, maid, sizeof cfm.maid);
4626 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4627 for(i = 0; i < mon->n_remote_mps; i++) {
4628 remote_mps[i] = mon->remote_mps[i]->mpid;
4631 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4632 &cfm, remote_mps, mon->n_remote_mps);
4636 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4637 * how 'iface''s port is configured.
4639 * Returns true if 'iface' is up, false otherwise. */
4641 iface_get_carrier(const struct iface *iface)
4643 return (iface->port->monitor
4644 ? netdev_get_carrier(iface->netdev)
4645 : netdev_get_miimon(iface->netdev));
4648 /* Port mirroring. */
4650 static struct mirror *
4651 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4655 for (i = 0; i < MAX_MIRRORS; i++) {
4656 struct mirror *m = br->mirrors[i];
4657 if (m && uuid_equals(uuid, &m->uuid)) {
4665 mirror_reconfigure(struct bridge *br)
4667 unsigned long *rspan_vlans;
4671 /* Get rid of deleted mirrors. */
4672 for (i = 0; i < MAX_MIRRORS; i++) {
4673 struct mirror *m = br->mirrors[i];
4675 const struct ovsdb_datum *mc;
4676 union ovsdb_atom atom;
4678 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4679 atom.uuid = br->mirrors[i]->uuid;
4680 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4686 /* Add new mirrors and reconfigure existing ones. */
4687 for (i = 0; i < br->cfg->n_mirrors; i++) {
4688 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4689 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4691 mirror_reconfigure_one(m, cfg);
4693 mirror_create(br, cfg);
4697 /* Update port reserved status. */
4698 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4699 port->is_mirror_output_port = false;
4701 for (i = 0; i < MAX_MIRRORS; i++) {
4702 struct mirror *m = br->mirrors[i];
4703 if (m && m->out_port) {
4704 m->out_port->is_mirror_output_port = true;
4708 /* Update flooded vlans (for RSPAN). */
4710 if (br->cfg->n_flood_vlans) {
4711 rspan_vlans = bitmap_allocate(4096);
4713 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4714 int64_t vlan = br->cfg->flood_vlans[i];
4715 if (vlan >= 0 && vlan < 4096) {
4716 bitmap_set1(rspan_vlans, vlan);
4717 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4720 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4725 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4727 mac_learning_flush(br->ml);
4732 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4737 for (i = 0; ; i++) {
4738 if (i >= MAX_MIRRORS) {
4739 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4740 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4743 if (!br->mirrors[i]) {
4748 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4750 mac_learning_flush(br->ml);
4752 br->mirrors[i] = m = xzalloc(sizeof *m);
4755 m->name = xstrdup(cfg->name);
4756 shash_init(&m->src_ports);
4757 shash_init(&m->dst_ports);
4763 mirror_reconfigure_one(m, cfg);
4767 mirror_destroy(struct mirror *m)
4770 struct bridge *br = m->bridge;
4773 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4774 port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4775 port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4778 shash_destroy(&m->src_ports);
4779 shash_destroy(&m->dst_ports);
4782 m->bridge->mirrors[m->idx] = NULL;
4787 mac_learning_flush(br->ml);
4792 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4793 struct shash *names)
4797 for (i = 0; i < n_ports; i++) {
4798 const char *name = ports[i]->name;
4799 if (port_lookup(m->bridge, name)) {
4800 shash_add_once(names, name, NULL);
4802 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4803 "port %s", m->bridge->name, m->name, name);
4809 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4815 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4817 for (i = 0; i < cfg->n_select_vlan; i++) {
4818 int64_t vlan = cfg->select_vlan[i];
4819 if (vlan < 0 || vlan > 4095) {
4820 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4821 m->bridge->name, m->name, vlan);
4823 (*vlans)[n_vlans++] = vlan;
4830 vlan_is_mirrored(const struct mirror *m, int vlan)
4834 for (i = 0; i < m->n_vlans; i++) {
4835 if (m->vlans[i] == vlan) {
4843 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4847 for (i = 0; i < m->n_vlans; i++) {
4848 if (port_trunks_vlan(p, m->vlans[i])) {
4856 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4858 struct shash src_ports, dst_ports;
4859 mirror_mask_t mirror_bit;
4860 struct port *out_port;
4867 if (strcmp(cfg->name, m->name)) {
4869 m->name = xstrdup(cfg->name);
4872 /* Get output port. */
4873 if (cfg->output_port) {
4874 out_port = port_lookup(m->bridge, cfg->output_port->name);
4876 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4877 m->bridge->name, m->name);
4883 if (cfg->output_vlan) {
4884 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4885 "output vlan; ignoring output vlan",
4886 m->bridge->name, m->name);
4888 } else if (cfg->output_vlan) {
4890 out_vlan = *cfg->output_vlan;
4892 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4893 m->bridge->name, m->name);
4898 shash_init(&src_ports);
4899 shash_init(&dst_ports);
4900 if (cfg->select_all) {
4901 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4902 shash_add_once(&src_ports, port->name, NULL);
4903 shash_add_once(&dst_ports, port->name, NULL);
4908 /* Get ports, and drop duplicates and ports that don't exist. */
4909 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4911 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4914 /* Get all the vlans, and drop duplicate and invalid vlans. */
4915 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4918 /* Update mirror data. */
4919 if (!shash_equal_keys(&m->src_ports, &src_ports)
4920 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4921 || m->n_vlans != n_vlans
4922 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4923 || m->out_port != out_port
4924 || m->out_vlan != out_vlan) {
4925 bridge_flush(m->bridge);
4926 mac_learning_flush(m->bridge->ml);
4928 shash_swap(&m->src_ports, &src_ports);
4929 shash_swap(&m->dst_ports, &dst_ports);
4932 m->n_vlans = n_vlans;
4933 m->out_port = out_port;
4934 m->out_vlan = out_vlan;
4937 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4938 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4939 if (shash_find(&m->src_ports, port->name)
4942 ? port_trunks_any_mirrored_vlan(m, port)
4943 : vlan_is_mirrored(m, port->vlan)))) {
4944 port->src_mirrors |= mirror_bit;
4946 port->src_mirrors &= ~mirror_bit;
4949 if (shash_find(&m->dst_ports, port->name)) {
4950 port->dst_mirrors |= mirror_bit;
4952 port->dst_mirrors &= ~mirror_bit;
4957 shash_destroy(&src_ports);
4958 shash_destroy(&dst_ports);