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 port *port; /* Containing port. */
102 size_t port_ifidx; /* Index within 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 int iface_idx; /* Index of assigned iface, or -1 if none. */
124 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
125 tag_type iface_tag; /* Tag associated with iface_idx. */
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;
159 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
160 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
161 * NULL if all VLANs are trunked. */
162 const struct ovsrec_port *cfg;
166 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
167 long long int miimon_interval; /* Miimon status refresh interval. */
168 long long int miimon_next_update; /* Time of next miimon update. */
170 /* An ordinary bridge port has 1 interface.
171 * A bridge port for bonding has at least 2 interfaces. */
172 struct iface **ifaces;
173 size_t n_ifaces, allocated_ifaces;
176 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
177 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
178 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
179 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
180 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
181 long long int bond_next_fake_iface_update; /* Time of next update. */
183 /* LACP information. */
184 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
185 bool lacp_active; /* True if LACP is active */
186 bool lacp_fast; /* True if LACP is in fast mode. */
187 uint16_t lacp_priority; /* LACP system priority. */
189 /* SLB specific bonding info. */
190 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
191 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
192 long long int bond_next_rebalance; /* Next rebalancing time. */
194 /* Port mirroring info. */
195 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
196 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
197 bool is_mirror_output_port; /* Does port mirroring send frames here? */
201 struct list node; /* Node in global list of bridges. */
202 char *name; /* User-specified arbitrary name. */
203 struct mac_learning *ml; /* MAC learning table. */
204 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
205 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
206 const struct ovsrec_bridge *cfg;
208 /* OpenFlow switch processing. */
209 struct ofproto *ofproto; /* OpenFlow switch. */
211 /* Kernel datapath information. */
212 struct dpif *dpif; /* Datapath. */
213 struct hmap ifaces; /* Contains "struct iface"s. */
217 size_t n_ports, allocated_ports;
218 struct shash iface_by_name; /* "struct iface"s indexed by name. */
219 struct shash port_by_name; /* "struct port"s indexed by name. */
222 bool has_bonded_ports;
227 /* Port mirroring. */
228 struct mirror *mirrors[MAX_MIRRORS];
231 /* List of all bridges. */
232 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
234 /* OVSDB IDL used to obtain configuration. */
235 static struct ovsdb_idl *idl;
237 /* Each time this timer expires, the bridge fetches systems and interface
238 * statistics and pushes them into the database. */
239 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
240 static long long int stats_timer = LLONG_MIN;
242 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
243 static void bridge_destroy(struct bridge *);
244 static struct bridge *bridge_lookup(const char *name);
245 static unixctl_cb_func bridge_unixctl_dump_flows;
246 static unixctl_cb_func bridge_unixctl_reconnect;
247 static int bridge_run_one(struct bridge *);
248 static size_t bridge_get_controllers(const struct bridge *br,
249 struct ovsrec_controller ***controllersp);
250 static void bridge_reconfigure_one(struct bridge *);
251 static void bridge_reconfigure_remotes(struct bridge *,
252 const struct sockaddr_in *managers,
254 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
255 static void bridge_fetch_dp_ifaces(struct bridge *);
256 static void bridge_flush(struct bridge *);
257 static void bridge_pick_local_hw_addr(struct bridge *,
258 uint8_t ea[ETH_ADDR_LEN],
259 struct iface **hw_addr_iface);
260 static uint64_t bridge_pick_datapath_id(struct bridge *,
261 const uint8_t bridge_ea[ETH_ADDR_LEN],
262 struct iface *hw_addr_iface);
263 static struct iface *bridge_get_local_iface(struct bridge *);
264 static uint64_t dpid_from_hash(const void *, size_t nbytes);
266 static unixctl_cb_func bridge_unixctl_fdb_show;
267 static unixctl_cb_func qos_unixctl_show;
269 static void bond_init(void);
270 static void bond_run(struct port *);
271 static void bond_wait(struct port *);
272 static void bond_rebalance_port(struct port *);
273 static void bond_send_learning_packets(struct port *);
274 static void bond_enable_slave(struct iface *iface, bool enable);
276 static void port_run(struct port *);
277 static void port_wait(struct port *);
278 static struct port *port_create(struct bridge *, const char *name);
279 static void port_reconfigure(struct port *, const struct ovsrec_port *);
280 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
281 static void port_destroy(struct port *);
282 static struct port *port_lookup(const struct bridge *, const char *name);
283 static struct iface *port_lookup_iface(const struct port *, const char *name);
284 static struct port *port_from_dp_ifidx(const struct bridge *,
286 static void port_update_bonding(struct port *);
287 static void port_update_lacp(struct port *);
289 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
290 static void mirror_destroy(struct mirror *);
291 static void mirror_reconfigure(struct bridge *);
292 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
293 static bool vlan_is_mirrored(const struct mirror *, int vlan);
295 static struct iface *iface_create(struct port *port,
296 const struct ovsrec_interface *if_cfg);
297 static void iface_destroy(struct iface *);
298 static struct iface *iface_lookup(const struct bridge *, const char *name);
299 static struct iface *iface_find(const char *name);
300 static struct iface *iface_from_dp_ifidx(const struct bridge *,
302 static void iface_set_mac(struct iface *);
303 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
304 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
305 static void iface_update_cfm(struct iface *);
306 static void iface_refresh_cfm_stats(struct iface *iface);
307 static void iface_update_carrier(struct iface *);
308 static bool iface_get_carrier(const struct iface *);
310 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
312 static void shash_to_ovs_idl_map(struct shash *,
313 char ***keys, char ***values, size_t *n);
315 /* Hooks into ofproto processing. */
316 static struct ofhooks bridge_ofhooks;
318 /* Public functions. */
320 /* Initializes the bridge module, configuring it to obtain its configuration
321 * from an OVSDB server accessed over 'remote', which should be a string in a
322 * form acceptable to ovsdb_idl_create(). */
324 bridge_init(const char *remote)
326 /* Create connection to database. */
327 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
329 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
330 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
331 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
333 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
335 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
336 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
338 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
339 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
340 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
342 /* Register unixctl commands. */
343 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
344 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
345 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
347 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
356 struct bridge *br, *next_br;
358 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
361 ovsdb_idl_destroy(idl);
364 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
365 * but for which the ovs-vswitchd configuration 'cfg' is required. */
367 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
369 static bool already_configured_once;
370 struct svec bridge_names;
371 struct svec dpif_names, dpif_types;
374 /* Only do this once per ovs-vswitchd run. */
375 if (already_configured_once) {
378 already_configured_once = true;
380 stats_timer = time_msec() + STATS_INTERVAL;
382 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
383 svec_init(&bridge_names);
384 for (i = 0; i < cfg->n_bridges; i++) {
385 svec_add(&bridge_names, cfg->bridges[i]->name);
387 svec_sort(&bridge_names);
389 /* Iterate over all system dpifs and delete any of them that do not appear
391 svec_init(&dpif_names);
392 svec_init(&dpif_types);
393 dp_enumerate_types(&dpif_types);
394 for (i = 0; i < dpif_types.n; i++) {
397 dp_enumerate_names(dpif_types.names[i], &dpif_names);
399 /* Delete each dpif whose name is not in 'bridge_names'. */
400 for (j = 0; j < dpif_names.n; j++) {
401 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
405 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
414 svec_destroy(&bridge_names);
415 svec_destroy(&dpif_names);
416 svec_destroy(&dpif_types);
419 /* Callback for iterate_and_prune_ifaces(). */
421 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
423 if (!iface->netdev) {
424 /* We already reported a related error, don't bother duplicating it. */
428 if (iface->dp_ifidx < 0) {
429 VLOG_ERR("%s interface not in %s, dropping",
430 iface->name, dpif_name(br->dpif));
434 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
435 iface->name, iface->dp_ifidx);
439 /* Callback for iterate_and_prune_ifaces(). */
441 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
442 void *aux OVS_UNUSED)
444 /* Set policing attributes. */
445 netdev_set_policing(iface->netdev,
446 iface->cfg->ingress_policing_rate,
447 iface->cfg->ingress_policing_burst);
449 /* Set MAC address of internal interfaces other than the local
451 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
452 iface_set_mac(iface);
458 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
459 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
460 * deletes from 'br' any ports that no longer have any interfaces. */
462 iterate_and_prune_ifaces(struct bridge *br,
463 bool (*cb)(struct bridge *, struct iface *,
469 for (i = 0; i < br->n_ports; ) {
470 struct port *port = br->ports[i];
471 for (j = 0; j < port->n_ifaces; ) {
472 struct iface *iface = port->ifaces[j];
473 if (cb(br, iface, aux)) {
476 iface_set_ofport(iface->cfg, -1);
477 iface_destroy(iface);
481 if (port->n_ifaces) {
484 VLOG_WARN("%s port has no interfaces, dropping", port->name);
490 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
491 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
492 * responsible for freeing '*managersp' (with free()).
494 * You may be asking yourself "why does ovs-vswitchd care?", because
495 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
496 * should not be and in fact is not directly involved in that. But
497 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
498 * it has to tell in-band control where the managers are to enable that.
499 * (Thus, only managers connected in-band are collected.)
502 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
503 struct sockaddr_in **managersp, size_t *n_managersp)
505 struct sockaddr_in *managers = NULL;
506 size_t n_managers = 0;
507 struct shash targets;
510 /* Collect all of the potential targets from the "targets" columns of the
511 * rows pointed to by "manager_options", excluding any that are
513 shash_init(&targets);
514 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
515 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
517 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
518 shash_find_and_delete(&targets, m->target);
520 shash_add_once(&targets, m->target, NULL);
524 /* Now extract the targets' IP addresses. */
525 if (!shash_is_empty(&targets)) {
526 struct shash_node *node;
528 managers = xmalloc(shash_count(&targets) * sizeof *managers);
529 SHASH_FOR_EACH (node, &targets) {
530 const char *target = node->name;
531 struct sockaddr_in *sin = &managers[n_managers];
533 if ((!strncmp(target, "tcp:", 4)
534 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
535 (!strncmp(target, "ssl:", 4)
536 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
541 shash_destroy(&targets);
543 *managersp = managers;
544 *n_managersp = n_managers;
548 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
550 struct shash old_br, new_br;
551 struct shash_node *node;
552 struct bridge *br, *next;
553 struct sockaddr_in *managers;
556 int sflow_bridge_number;
558 COVERAGE_INC(bridge_reconfigure);
560 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
562 /* Collect old and new bridges. */
565 LIST_FOR_EACH (br, node, &all_bridges) {
566 shash_add(&old_br, br->name, br);
568 for (i = 0; i < ovs_cfg->n_bridges; i++) {
569 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
570 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
571 VLOG_WARN("more than one bridge named %s", br_cfg->name);
575 /* Get rid of deleted bridges and add new bridges. */
576 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
577 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
584 SHASH_FOR_EACH (node, &new_br) {
585 const char *br_name = node->name;
586 const struct ovsrec_bridge *br_cfg = node->data;
587 br = shash_find_data(&old_br, br_name);
589 /* If the bridge datapath type has changed, we need to tear it
590 * down and recreate. */
591 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
593 bridge_create(br_cfg);
596 bridge_create(br_cfg);
599 shash_destroy(&old_br);
600 shash_destroy(&new_br);
602 /* Reconfigure all bridges. */
603 LIST_FOR_EACH (br, node, &all_bridges) {
604 bridge_reconfigure_one(br);
607 /* Add and delete ports on all datapaths.
609 * The kernel will reject any attempt to add a given port to a datapath if
610 * that port already belongs to a different datapath, so we must do all
611 * port deletions before any port additions. */
612 LIST_FOR_EACH (br, node, &all_bridges) {
613 struct dpif_port_dump dump;
614 struct shash want_ifaces;
615 struct dpif_port dpif_port;
617 bridge_get_all_ifaces(br, &want_ifaces);
618 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
619 if (!shash_find(&want_ifaces, dpif_port.name)
620 && strcmp(dpif_port.name, br->name)) {
621 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
623 VLOG_WARN("failed to remove %s interface from %s: %s",
624 dpif_port.name, dpif_name(br->dpif),
629 shash_destroy(&want_ifaces);
631 LIST_FOR_EACH (br, node, &all_bridges) {
632 struct shash cur_ifaces, want_ifaces;
633 struct dpif_port_dump dump;
634 struct dpif_port dpif_port;
636 /* Get the set of interfaces currently in this datapath. */
637 shash_init(&cur_ifaces);
638 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
639 struct dpif_port *port_info = xmalloc(sizeof *port_info);
640 dpif_port_clone(port_info, &dpif_port);
641 shash_add(&cur_ifaces, dpif_port.name, port_info);
644 /* Get the set of interfaces we want on this datapath. */
645 bridge_get_all_ifaces(br, &want_ifaces);
647 hmap_clear(&br->ifaces);
648 SHASH_FOR_EACH (node, &want_ifaces) {
649 const char *if_name = node->name;
650 struct iface *iface = node->data;
651 struct dpif_port *dpif_port;
655 type = iface ? iface->type : "internal";
656 dpif_port = shash_find_data(&cur_ifaces, if_name);
658 /* If we have a port or a netdev already, and it's not the type we
659 * want, then delete the port (if any) and close the netdev (if
661 if ((dpif_port && strcmp(dpif_port->type, type))
662 || (iface && iface->netdev
663 && strcmp(type, netdev_get_type(iface->netdev)))) {
665 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
672 netdev_close(iface->netdev);
673 iface->netdev = NULL;
677 /* If the port doesn't exist or we don't have the netdev open,
678 * we need to do more work. */
679 if (!dpif_port || (iface && !iface->netdev)) {
680 struct netdev_options options;
681 struct netdev *netdev;
684 /* First open the network device. */
685 options.name = if_name;
687 options.args = &args;
688 options.ethertype = NETDEV_ETH_TYPE_NONE;
692 shash_from_ovs_idl_map(iface->cfg->key_options,
693 iface->cfg->value_options,
694 iface->cfg->n_options, &args);
696 error = netdev_open(&options, &netdev);
697 shash_destroy(&args);
700 VLOG_WARN("could not open network device %s (%s)",
701 if_name, strerror(error));
705 /* Then add the port if we haven't already. */
707 error = dpif_port_add(br->dpif, netdev, NULL);
709 netdev_close(netdev);
710 if (error == EFBIG) {
711 VLOG_ERR("ran out of valid port numbers on %s",
712 dpif_name(br->dpif));
715 VLOG_WARN("failed to add %s interface to %s: %s",
716 if_name, dpif_name(br->dpif),
723 /* Update 'iface'. */
725 iface->netdev = netdev;
726 iface->enabled = iface_get_carrier(iface);
727 iface->up = iface->enabled;
729 } else if (iface && iface->netdev) {
733 shash_from_ovs_idl_map(iface->cfg->key_options,
734 iface->cfg->value_options,
735 iface->cfg->n_options, &args);
736 netdev_set_config(iface->netdev, &args);
737 shash_destroy(&args);
740 shash_destroy(&want_ifaces);
742 SHASH_FOR_EACH (node, &cur_ifaces) {
743 struct dpif_port *port_info = node->data;
744 dpif_port_destroy(port_info);
747 shash_destroy(&cur_ifaces);
749 sflow_bridge_number = 0;
750 LIST_FOR_EACH (br, node, &all_bridges) {
753 struct iface *local_iface;
754 struct iface *hw_addr_iface;
757 bridge_fetch_dp_ifaces(br);
759 iterate_and_prune_ifaces(br, check_iface, NULL);
761 /* Pick local port hardware address, datapath ID. */
762 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
763 local_iface = bridge_get_local_iface(br);
765 int error = netdev_set_etheraddr(local_iface->netdev, ea);
767 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
768 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
769 "Ethernet address: %s",
770 br->name, strerror(error));
773 memcpy(br->ea, ea, ETH_ADDR_LEN);
775 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
776 ofproto_set_datapath_id(br->ofproto, dpid);
778 dpid_string = xasprintf("%016"PRIx64, dpid);
779 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
782 /* Set NetFlow configuration on this bridge. */
783 if (br->cfg->netflow) {
784 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
785 struct netflow_options opts;
787 memset(&opts, 0, sizeof opts);
789 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
790 if (nf_cfg->engine_type) {
791 opts.engine_type = *nf_cfg->engine_type;
793 if (nf_cfg->engine_id) {
794 opts.engine_id = *nf_cfg->engine_id;
797 opts.active_timeout = nf_cfg->active_timeout;
798 if (!opts.active_timeout) {
799 opts.active_timeout = -1;
800 } else if (opts.active_timeout < 0) {
801 VLOG_WARN("bridge %s: active timeout interval set to negative "
802 "value, using default instead (%d seconds)", br->name,
803 NF_ACTIVE_TIMEOUT_DEFAULT);
804 opts.active_timeout = -1;
807 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
808 if (opts.add_id_to_iface) {
809 if (opts.engine_id > 0x7f) {
810 VLOG_WARN("bridge %s: netflow port mangling may conflict "
811 "with another vswitch, choose an engine id less "
812 "than 128", br->name);
814 if (br->n_ports > 508) {
815 VLOG_WARN("bridge %s: netflow port mangling will conflict "
816 "with another port when more than 508 ports are "
821 opts.collectors.n = nf_cfg->n_targets;
822 opts.collectors.names = nf_cfg->targets;
823 if (ofproto_set_netflow(br->ofproto, &opts)) {
824 VLOG_ERR("bridge %s: problem setting netflow collectors",
828 ofproto_set_netflow(br->ofproto, NULL);
831 /* Set sFlow configuration on this bridge. */
832 if (br->cfg->sflow) {
833 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
834 struct ovsrec_controller **controllers;
835 struct ofproto_sflow_options oso;
836 size_t n_controllers;
838 memset(&oso, 0, sizeof oso);
840 oso.targets.n = sflow_cfg->n_targets;
841 oso.targets.names = sflow_cfg->targets;
843 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
844 if (sflow_cfg->sampling) {
845 oso.sampling_rate = *sflow_cfg->sampling;
848 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
849 if (sflow_cfg->polling) {
850 oso.polling_interval = *sflow_cfg->polling;
853 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
854 if (sflow_cfg->header) {
855 oso.header_len = *sflow_cfg->header;
858 oso.sub_id = sflow_bridge_number++;
859 oso.agent_device = sflow_cfg->agent;
861 oso.control_ip = NULL;
862 n_controllers = bridge_get_controllers(br, &controllers);
863 for (i = 0; i < n_controllers; i++) {
864 if (controllers[i]->local_ip) {
865 oso.control_ip = controllers[i]->local_ip;
869 ofproto_set_sflow(br->ofproto, &oso);
871 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
873 ofproto_set_sflow(br->ofproto, NULL);
876 /* Update the controller and related settings. It would be more
877 * straightforward to call this from bridge_reconfigure_one(), but we
878 * can't do it there for two reasons. First, and most importantly, at
879 * that point we don't know the dp_ifidx of any interfaces that have
880 * been added to the bridge (because we haven't actually added them to
881 * the datapath). Second, at that point we haven't set the datapath ID
882 * yet; when a controller is configured, resetting the datapath ID will
883 * immediately disconnect from the controller, so it's better to set
884 * the datapath ID before the controller. */
885 bridge_reconfigure_remotes(br, managers, n_managers);
887 LIST_FOR_EACH (br, node, &all_bridges) {
888 for (i = 0; i < br->n_ports; i++) {
889 struct port *port = br->ports[i];
893 for (j = 0; j < port->n_ifaces; j++) {
894 netdev_monitor_add(port->monitor, port->ifaces[j]->netdev);
897 port->miimon_next_update = 0;
900 port_update_lacp(port);
901 port_update_bonding(port);
903 for (j = 0; j < port->n_ifaces; j++) {
904 iface_update_qos(port->ifaces[j], port->cfg->qos);
908 LIST_FOR_EACH (br, node, &all_bridges) {
909 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
912 LIST_FOR_EACH (br, node, &all_bridges) {
914 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
915 iface_update_cfm(iface);
921 /* ovs-vswitchd has completed initialization, so allow the process that
922 * forked us to exit successfully. */
923 daemonize_complete();
927 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
928 const struct ovsdb_idl_column *column,
931 const struct ovsdb_datum *datum;
932 union ovsdb_atom atom;
935 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
936 atom.string = (char *) key;
937 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
938 return idx == UINT_MAX ? NULL : datum->values[idx].string;
942 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
944 return get_ovsrec_key_value(&br_cfg->header_,
945 &ovsrec_bridge_col_other_config, key);
949 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
950 struct iface **hw_addr_iface)
956 *hw_addr_iface = NULL;
958 /* Did the user request a particular MAC? */
959 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
960 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
961 if (eth_addr_is_multicast(ea)) {
962 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
963 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
964 } else if (eth_addr_is_zero(ea)) {
965 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
971 /* Otherwise choose the minimum non-local MAC address among all of the
973 memset(ea, 0xff, ETH_ADDR_LEN);
974 for (i = 0; i < br->n_ports; i++) {
975 struct port *port = br->ports[i];
976 uint8_t iface_ea[ETH_ADDR_LEN];
979 /* Mirror output ports don't participate. */
980 if (port->is_mirror_output_port) {
984 /* Choose the MAC address to represent the port. */
985 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
986 /* Find the interface with this Ethernet address (if any) so that
987 * we can provide the correct devname to the caller. */
989 for (j = 0; j < port->n_ifaces; j++) {
990 struct iface *candidate = port->ifaces[j];
991 uint8_t candidate_ea[ETH_ADDR_LEN];
992 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
993 && eth_addr_equals(iface_ea, candidate_ea)) {
998 /* Choose the interface whose MAC address will represent the port.
999 * The Linux kernel bonding code always chooses the MAC address of
1000 * the first slave added to a bond, and the Fedora networking
1001 * scripts always add slaves to a bond in alphabetical order, so
1002 * for compatibility we choose the interface with the name that is
1003 * first in alphabetical order. */
1004 iface = port->ifaces[0];
1005 for (j = 1; j < port->n_ifaces; j++) {
1006 struct iface *candidate = port->ifaces[j];
1007 if (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);
1207 iface_refresh_cfm_stats(struct iface *iface)
1209 const struct ovsrec_monitor *mon;
1210 const struct cfm *cfm;
1213 mon = iface->cfg->monitor;
1214 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1220 for (i = 0; i < mon->n_remote_mps; i++) {
1221 const struct ovsrec_maintenance_point *mp;
1222 const struct remote_mp *rmp;
1224 mp = mon->remote_mps[i];
1225 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1227 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1230 if (hmap_is_empty(&cfm->x_remote_mps)) {
1231 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1234 struct remote_mp *rmp;
1235 int64_t *x_remote_mps;
1237 length = hmap_count(&cfm->x_remote_mps);
1238 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1241 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1242 x_remote_mps[i++] = rmp->mpid;
1245 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1249 if (hmap_is_empty(&cfm->x_remote_maids)) {
1250 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1253 char **x_remote_maids;
1254 struct remote_maid *rmaid;
1256 length = hmap_count(&cfm->x_remote_maids);
1257 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1260 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1263 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1265 for (j = 0; j < CCM_MAID_LEN; j++) {
1266 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1271 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1273 for (i = 0; i < length; i++) {
1274 free(x_remote_maids[i]);
1276 free(x_remote_maids);
1279 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1283 iface_refresh_stats(struct iface *iface)
1289 static const struct iface_stat iface_stats[] = {
1290 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1291 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1292 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1293 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1294 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1295 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1296 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1297 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1298 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1299 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1300 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1301 { "collisions", offsetof(struct netdev_stats, collisions) },
1303 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1304 const struct iface_stat *s;
1306 char *keys[N_STATS];
1307 int64_t values[N_STATS];
1310 struct netdev_stats stats;
1312 /* Intentionally ignore return value, since errors will set 'stats' to
1313 * all-1s, and we will deal with that correctly below. */
1314 netdev_get_stats(iface->netdev, &stats);
1317 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1318 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1319 if (value != UINT64_MAX) {
1326 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1330 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1332 struct ovsdb_datum datum;
1336 get_system_stats(&stats);
1338 ovsdb_datum_from_shash(&datum, &stats);
1339 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1343 static inline const char *
1344 nx_role_to_str(enum nx_role role)
1349 case NX_ROLE_MASTER:
1354 return "*** INVALID ROLE ***";
1359 bridge_refresh_controller_status(const struct bridge *br)
1362 const struct ovsrec_controller *cfg;
1364 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1366 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1367 struct ofproto_controller_info *cinfo =
1368 shash_find_data(&info, cfg->target);
1371 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1372 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1373 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1374 (char **) cinfo->pairs.values,
1377 ovsrec_controller_set_is_connected(cfg, false);
1378 ovsrec_controller_set_role(cfg, NULL);
1379 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1383 ofproto_free_ofproto_controller_info(&info);
1389 const struct ovsrec_open_vswitch *cfg;
1391 bool datapath_destroyed;
1392 bool database_changed;
1395 /* Let each bridge do the work that it needs to do. */
1396 datapath_destroyed = false;
1397 LIST_FOR_EACH (br, node, &all_bridges) {
1398 int error = bridge_run_one(br);
1400 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1401 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1402 "forcing reconfiguration", br->name);
1403 datapath_destroyed = true;
1407 /* (Re)configure if necessary. */
1408 database_changed = ovsdb_idl_run(idl);
1409 cfg = ovsrec_open_vswitch_first(idl);
1411 /* Re-configure SSL. We do this on every trip through the main loop,
1412 * instead of just when the database changes, because the contents of the
1413 * key and certificate files can change without the database changing.
1415 * We do this before bridge_reconfigure() because that function might
1416 * initiate SSL connections and thus requires SSL to be configured. */
1417 if (cfg && cfg->ssl) {
1418 const struct ovsrec_ssl *ssl = cfg->ssl;
1420 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1421 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1424 if (database_changed || datapath_destroyed) {
1426 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1428 bridge_configure_once(cfg);
1429 bridge_reconfigure(cfg);
1431 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1432 ovsdb_idl_txn_commit(txn);
1433 ovsdb_idl_txn_destroy(txn); /* XXX */
1435 /* We still need to reconfigure to avoid dangling pointers to
1436 * now-destroyed ovsrec structures inside bridge data. */
1437 static const struct ovsrec_open_vswitch null_cfg;
1439 bridge_reconfigure(&null_cfg);
1443 /* Refresh system and interface stats if necessary. */
1444 if (time_msec() >= stats_timer) {
1446 struct ovsdb_idl_txn *txn;
1448 txn = ovsdb_idl_txn_create(idl);
1449 LIST_FOR_EACH (br, node, &all_bridges) {
1452 for (i = 0; i < br->n_ports; i++) {
1453 struct port *port = br->ports[i];
1456 for (j = 0; j < port->n_ifaces; j++) {
1457 struct iface *iface = port->ifaces[j];
1458 iface_refresh_stats(iface);
1459 iface_refresh_cfm_stats(iface);
1460 iface_refresh_status(iface);
1463 bridge_refresh_controller_status(br);
1465 refresh_system_stats(cfg);
1466 ovsdb_idl_txn_commit(txn);
1467 ovsdb_idl_txn_destroy(txn); /* XXX */
1470 stats_timer = time_msec() + STATS_INTERVAL;
1479 LIST_FOR_EACH (br, node, &all_bridges) {
1482 ofproto_wait(br->ofproto);
1483 if (ofproto_has_primary_controller(br->ofproto)) {
1487 mac_learning_wait(br->ml);
1489 for (i = 0; i < br->n_ports; i++) {
1490 port_wait(br->ports[i]);
1493 ovsdb_idl_wait(idl);
1494 poll_timer_wait_until(stats_timer);
1497 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1498 * configuration changes. */
1500 bridge_flush(struct bridge *br)
1502 COVERAGE_INC(bridge_flush);
1504 mac_learning_flush(br->ml);
1507 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1508 * such interface. */
1509 static struct iface *
1510 bridge_get_local_iface(struct bridge *br)
1514 for (i = 0; i < br->n_ports; i++) {
1515 struct port *port = br->ports[i];
1516 for (j = 0; j < port->n_ifaces; j++) {
1517 struct iface *iface = port->ifaces[j];
1518 if (iface->dp_ifidx == ODPP_LOCAL) {
1527 /* Bridge unixctl user interface functions. */
1529 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1530 const char *args, void *aux OVS_UNUSED)
1532 struct ds ds = DS_EMPTY_INITIALIZER;
1533 const struct bridge *br;
1534 const struct mac_entry *e;
1536 br = bridge_lookup(args);
1538 unixctl_command_reply(conn, 501, "no such bridge");
1542 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1543 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1544 if (e->port < 0 || e->port >= br->n_ports) {
1547 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1548 br->ports[e->port]->ifaces[0]->dp_ifidx,
1549 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1551 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1555 /* QoS unixctl user interface functions. */
1557 struct qos_unixctl_show_cbdata {
1559 struct iface *iface;
1563 qos_unixctl_show_cb(unsigned int queue_id,
1564 const struct shash *details,
1567 struct qos_unixctl_show_cbdata *data = aux;
1568 struct ds *ds = data->ds;
1569 struct iface *iface = data->iface;
1570 struct netdev_queue_stats stats;
1571 struct shash_node *node;
1574 ds_put_cstr(ds, "\n");
1576 ds_put_format(ds, "Queue %u:\n", queue_id);
1578 ds_put_cstr(ds, "Default:\n");
1581 SHASH_FOR_EACH (node, details) {
1582 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1585 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1587 if (stats.tx_packets != UINT64_MAX) {
1588 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1591 if (stats.tx_bytes != UINT64_MAX) {
1592 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1595 if (stats.tx_errors != UINT64_MAX) {
1596 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1599 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1600 queue_id, strerror(error));
1605 qos_unixctl_show(struct unixctl_conn *conn,
1606 const char *args, void *aux OVS_UNUSED)
1608 struct ds ds = DS_EMPTY_INITIALIZER;
1609 struct shash sh = SHASH_INITIALIZER(&sh);
1610 struct iface *iface;
1612 struct shash_node *node;
1613 struct qos_unixctl_show_cbdata data;
1616 iface = iface_find(args);
1618 unixctl_command_reply(conn, 501, "no such interface");
1622 netdev_get_qos(iface->netdev, &type, &sh);
1624 if (*type != '\0') {
1625 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1627 SHASH_FOR_EACH (node, &sh) {
1628 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1633 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1636 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1638 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1640 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1641 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1644 shash_destroy_free_data(&sh);
1648 /* Bridge reconfiguration functions. */
1649 static struct bridge *
1650 bridge_create(const struct ovsrec_bridge *br_cfg)
1655 assert(!bridge_lookup(br_cfg->name));
1656 br = xzalloc(sizeof *br);
1658 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1664 dpif_flow_flush(br->dpif);
1666 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1669 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1671 dpif_delete(br->dpif);
1672 dpif_close(br->dpif);
1677 br->name = xstrdup(br_cfg->name);
1679 br->ml = mac_learning_create();
1680 eth_addr_nicira_random(br->default_ea);
1682 hmap_init(&br->ifaces);
1684 shash_init(&br->port_by_name);
1685 shash_init(&br->iface_by_name);
1689 list_push_back(&all_bridges, &br->node);
1691 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1697 bridge_destroy(struct bridge *br)
1702 while (br->n_ports > 0) {
1703 port_destroy(br->ports[br->n_ports - 1]);
1705 list_remove(&br->node);
1706 error = dpif_delete(br->dpif);
1707 if (error && error != ENOENT) {
1708 VLOG_ERR("failed to delete %s: %s",
1709 dpif_name(br->dpif), strerror(error));
1711 dpif_close(br->dpif);
1712 ofproto_destroy(br->ofproto);
1713 mac_learning_destroy(br->ml);
1714 hmap_destroy(&br->ifaces);
1715 shash_destroy(&br->port_by_name);
1716 shash_destroy(&br->iface_by_name);
1723 static struct bridge *
1724 bridge_lookup(const char *name)
1728 LIST_FOR_EACH (br, node, &all_bridges) {
1729 if (!strcmp(br->name, name)) {
1736 /* Handle requests for a listing of all flows known by the OpenFlow
1737 * stack, including those normally hidden. */
1739 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1740 const char *args, void *aux OVS_UNUSED)
1745 br = bridge_lookup(args);
1747 unixctl_command_reply(conn, 501, "Unknown bridge");
1752 ofproto_get_all_flows(br->ofproto, &results);
1754 unixctl_command_reply(conn, 200, ds_cstr(&results));
1755 ds_destroy(&results);
1758 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1759 * connections and reconnect. If BRIDGE is not specified, then all bridges
1760 * drop their controller connections and reconnect. */
1762 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1763 const char *args, void *aux OVS_UNUSED)
1766 if (args[0] != '\0') {
1767 br = bridge_lookup(args);
1769 unixctl_command_reply(conn, 501, "Unknown bridge");
1772 ofproto_reconnect_controllers(br->ofproto);
1774 LIST_FOR_EACH (br, node, &all_bridges) {
1775 ofproto_reconnect_controllers(br->ofproto);
1778 unixctl_command_reply(conn, 200, NULL);
1782 bridge_run_one(struct bridge *br)
1787 error = ofproto_run1(br->ofproto);
1792 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1794 for (i = 0; i < br->n_ports; i++) {
1795 port_run(br->ports[i]);
1798 error = ofproto_run2(br->ofproto, br->flush);
1805 bridge_get_controllers(const struct bridge *br,
1806 struct ovsrec_controller ***controllersp)
1808 struct ovsrec_controller **controllers;
1809 size_t n_controllers;
1811 controllers = br->cfg->controller;
1812 n_controllers = br->cfg->n_controller;
1814 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1820 *controllersp = controllers;
1822 return n_controllers;
1826 bridge_reconfigure_one(struct bridge *br)
1828 struct shash old_ports, new_ports;
1829 struct svec snoops, old_snoops;
1830 struct shash_node *node;
1831 enum ofproto_fail_mode fail_mode;
1834 /* Collect old ports. */
1835 shash_init(&old_ports);
1836 for (i = 0; i < br->n_ports; i++) {
1837 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1840 /* Collect new ports. */
1841 shash_init(&new_ports);
1842 for (i = 0; i < br->cfg->n_ports; i++) {
1843 const char *name = br->cfg->ports[i]->name;
1844 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1845 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1850 /* If we have a controller, then we need a local port. Complain if the
1851 * user didn't specify one.
1853 * XXX perhaps we should synthesize a port ourselves in this case. */
1854 if (bridge_get_controllers(br, NULL)) {
1855 char local_name[IF_NAMESIZE];
1858 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1859 local_name, sizeof local_name);
1860 if (!error && !shash_find(&new_ports, local_name)) {
1861 VLOG_WARN("bridge %s: controller specified but no local port "
1862 "(port named %s) defined",
1863 br->name, local_name);
1867 /* Get rid of deleted ports.
1868 * Get rid of deleted interfaces on ports that still exist. */
1869 SHASH_FOR_EACH (node, &old_ports) {
1870 struct port *port = node->data;
1871 const struct ovsrec_port *port_cfg;
1873 port_cfg = shash_find_data(&new_ports, node->name);
1877 port_del_ifaces(port, port_cfg);
1881 /* Create new ports.
1882 * Add new interfaces to existing ports.
1883 * Reconfigure existing ports. */
1884 SHASH_FOR_EACH (node, &new_ports) {
1885 struct port *port = shash_find_data(&old_ports, node->name);
1887 port = port_create(br, node->name);
1890 port_reconfigure(port, node->data);
1891 if (!port->n_ifaces) {
1892 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1893 br->name, port->name);
1897 shash_destroy(&old_ports);
1898 shash_destroy(&new_ports);
1900 /* Set the fail-mode */
1901 fail_mode = !br->cfg->fail_mode
1902 || !strcmp(br->cfg->fail_mode, "standalone")
1903 ? OFPROTO_FAIL_STANDALONE
1904 : OFPROTO_FAIL_SECURE;
1905 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1906 && !ofproto_has_primary_controller(br->ofproto)) {
1907 ofproto_flush_flows(br->ofproto);
1909 ofproto_set_fail_mode(br->ofproto, fail_mode);
1911 /* Delete all flows if we're switching from connected to standalone or vice
1912 * versa. (XXX Should we delete all flows if we are switching from one
1913 * controller to another?) */
1915 /* Configure OpenFlow controller connection snooping. */
1917 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1918 ovs_rundir(), br->name));
1919 svec_init(&old_snoops);
1920 ofproto_get_snoops(br->ofproto, &old_snoops);
1921 if (!svec_equal(&snoops, &old_snoops)) {
1922 ofproto_set_snoops(br->ofproto, &snoops);
1924 svec_destroy(&snoops);
1925 svec_destroy(&old_snoops);
1927 mirror_reconfigure(br);
1930 /* Initializes 'oc' appropriately as a management service controller for
1933 * The caller must free oc->target when it is no longer needed. */
1935 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1936 struct ofproto_controller *oc)
1938 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1939 oc->max_backoff = 0;
1940 oc->probe_interval = 60;
1941 oc->band = OFPROTO_OUT_OF_BAND;
1943 oc->burst_limit = 0;
1946 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1948 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1949 struct ofproto_controller *oc)
1951 oc->target = c->target;
1952 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1953 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1954 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1955 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1956 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1957 oc->burst_limit = (c->controller_burst_limit
1958 ? *c->controller_burst_limit : 0);
1961 /* Configures the IP stack for 'br''s local interface properly according to the
1962 * configuration in 'c'. */
1964 bridge_configure_local_iface_netdev(struct bridge *br,
1965 struct ovsrec_controller *c)
1967 struct netdev *netdev;
1968 struct in_addr mask, gateway;
1970 struct iface *local_iface;
1973 /* If there's no local interface or no IP address, give up. */
1974 local_iface = bridge_get_local_iface(br);
1975 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1979 /* Bring up the local interface. */
1980 netdev = local_iface->netdev;
1981 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1983 /* Configure the IP address and netmask. */
1984 if (!c->local_netmask
1985 || !inet_aton(c->local_netmask, &mask)
1987 mask.s_addr = guess_netmask(ip.s_addr);
1989 if (!netdev_set_in4(netdev, ip, mask)) {
1990 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1991 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1994 /* Configure the default gateway. */
1995 if (c->local_gateway
1996 && inet_aton(c->local_gateway, &gateway)
1997 && gateway.s_addr) {
1998 if (!netdev_add_router(netdev, gateway)) {
1999 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2000 br->name, IP_ARGS(&gateway.s_addr));
2006 bridge_reconfigure_remotes(struct bridge *br,
2007 const struct sockaddr_in *managers,
2010 const char *disable_ib_str, *queue_id_str;
2011 bool disable_in_band = false;
2014 struct ovsrec_controller **controllers;
2015 size_t n_controllers;
2018 struct ofproto_controller *ocs;
2022 /* Check if we should disable in-band control on this bridge. */
2023 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2024 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2025 disable_in_band = true;
2028 /* Set OpenFlow queue ID for in-band control. */
2029 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2030 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2031 ofproto_set_in_band_queue(br->ofproto, queue_id);
2033 if (disable_in_band) {
2034 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2036 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2038 had_primary = ofproto_has_primary_controller(br->ofproto);
2040 n_controllers = bridge_get_controllers(br, &controllers);
2042 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2045 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2046 for (i = 0; i < n_controllers; i++) {
2047 struct ovsrec_controller *c = controllers[i];
2049 if (!strncmp(c->target, "punix:", 6)
2050 || !strncmp(c->target, "unix:", 5)) {
2051 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2053 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2054 * domain sockets and overwriting arbitrary local files. */
2055 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2056 "\"%s\" due to possibility for remote exploit",
2057 dpif_name(br->dpif), c->target);
2061 bridge_configure_local_iface_netdev(br, c);
2062 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2063 if (disable_in_band) {
2064 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2069 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2070 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2073 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2074 ofproto_flush_flows(br->ofproto);
2077 /* If there are no controllers and the bridge is in standalone
2078 * mode, set up a flow that matches every packet and directs
2079 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2080 * switch is in secure mode and we won't pass any traffic until
2081 * a controller has been defined and it tells us to do so. */
2083 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2084 union ofp_action action;
2085 struct cls_rule rule;
2087 memset(&action, 0, sizeof action);
2088 action.type = htons(OFPAT_OUTPUT);
2089 action.output.len = htons(sizeof action);
2090 action.output.port = htons(OFPP_NORMAL);
2091 cls_rule_init_catchall(&rule, 0);
2092 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2097 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2102 for (i = 0; i < br->n_ports; i++) {
2103 struct port *port = br->ports[i];
2104 for (j = 0; j < port->n_ifaces; j++) {
2105 struct iface *iface = port->ifaces[j];
2106 shash_add_once(ifaces, iface->name, iface);
2108 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2109 shash_add_once(ifaces, port->name, NULL);
2114 /* For robustness, in case the administrator moves around datapath ports behind
2115 * our back, we re-check all the datapath port numbers here.
2117 * This function will set the 'dp_ifidx' members of interfaces that have
2118 * disappeared to -1, so only call this function from a context where those
2119 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2120 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2121 * datapath, which doesn't support UINT16_MAX+1 ports. */
2123 bridge_fetch_dp_ifaces(struct bridge *br)
2125 struct dpif_port_dump dump;
2126 struct dpif_port dpif_port;
2129 /* Reset all interface numbers. */
2130 for (i = 0; i < br->n_ports; i++) {
2131 struct port *port = br->ports[i];
2132 for (j = 0; j < port->n_ifaces; j++) {
2133 struct iface *iface = port->ifaces[j];
2134 iface->dp_ifidx = -1;
2137 hmap_clear(&br->ifaces);
2139 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2140 struct iface *iface = iface_lookup(br, dpif_port.name);
2142 if (iface->dp_ifidx >= 0) {
2143 VLOG_WARN("%s reported interface %s twice",
2144 dpif_name(br->dpif), dpif_port.name);
2145 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2146 VLOG_WARN("%s reported interface %"PRIu16" twice",
2147 dpif_name(br->dpif), dpif_port.port_no);
2149 iface->dp_ifidx = dpif_port.port_no;
2150 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2151 hash_int(iface->dp_ifidx, 0));
2154 iface_set_ofport(iface->cfg,
2155 (iface->dp_ifidx >= 0
2156 ? odp_port_to_ofp_port(iface->dp_ifidx)
2162 /* Bridge packet processing functions. */
2165 bond_is_tcp_hash(const struct port *port)
2167 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2171 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2173 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2176 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2178 struct flow hash_flow;
2180 memcpy(&hash_flow, flow, sizeof hash_flow);
2181 hash_flow.vlan_tci = 0;
2183 /* The symmetric quality of this hash function is not required, but
2184 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2185 * purposes, so we use it out of convenience. */
2186 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2189 static struct bond_entry *
2190 lookup_bond_entry(const struct port *port, const struct flow *flow,
2193 assert(port->bond_mode != BM_AB);
2195 if (bond_is_tcp_hash(port)) {
2196 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2198 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2203 bond_choose_iface(const struct port *port)
2205 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2206 size_t i, best_down_slave = -1;
2207 long long next_delay_expiration = LLONG_MAX;
2209 for (i = 0; i < port->n_ifaces; i++) {
2210 struct iface *iface = port->ifaces[i];
2212 if (iface->enabled) {
2214 } else if (iface->delay_expires < next_delay_expiration
2215 && lacp_slave_may_enable(port->lacp, iface)) {
2216 best_down_slave = i;
2217 next_delay_expiration = iface->delay_expires;
2221 if (best_down_slave != -1) {
2222 struct iface *iface = port->ifaces[best_down_slave];
2224 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2225 "since no other interface is up", iface->name,
2226 iface->delay_expires - time_msec());
2227 bond_enable_slave(iface, true);
2230 return best_down_slave;
2234 choose_output_iface(const struct port *port, const struct flow *flow,
2235 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2237 struct iface *iface;
2239 assert(port->n_ifaces);
2240 if (port->n_ifaces == 1) {
2241 iface = port->ifaces[0];
2242 } else if (port->bond_mode == BM_AB) {
2243 if (port->active_iface < 0) {
2244 *tags |= port->no_ifaces_tag;
2247 iface = port->ifaces[port->active_iface];
2249 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2250 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2251 || !port->ifaces[e->iface_idx]->enabled) {
2252 /* XXX select interface properly. The current interface selection
2253 * is only good for testing the rebalancing code. */
2254 e->iface_idx = bond_choose_iface(port);
2255 if (e->iface_idx < 0) {
2256 *tags |= port->no_ifaces_tag;
2259 e->iface_tag = tag_create_random();
2261 *tags |= e->iface_tag;
2262 iface = port->ifaces[e->iface_idx];
2264 *dp_ifidx = iface->dp_ifidx;
2265 *tags |= iface->tag; /* Currently only used for bonding. */
2270 bond_link_status_update(struct iface *iface)
2272 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2273 struct port *port = iface->port;
2274 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2275 int updelay, downdelay;
2277 updelay = port->updelay;
2278 downdelay = port->downdelay;
2280 if (lacp_negotiated(port->lacp)) {
2285 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2286 /* Nothing to do. */
2289 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2290 iface->name, up ? "up" : "down");
2291 if (up == iface->enabled) {
2292 iface->delay_expires = LLONG_MAX;
2293 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2294 iface->name, up ? "disabled" : "enabled");
2295 } else if (up && port->active_iface < 0) {
2296 bond_enable_slave(iface, true);
2298 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2299 "other interface is up", iface->name, updelay);
2302 int delay = up ? updelay : downdelay;
2303 iface->delay_expires = time_msec() + delay;
2306 "interface %s: will be %s if it stays %s for %d ms",
2308 up ? "enabled" : "disabled",
2316 bond_choose_active_iface(struct port *port)
2318 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2320 port->active_iface = bond_choose_iface(port);
2321 if (port->active_iface >= 0) {
2322 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2323 port->name, port->ifaces[port->active_iface]->name);
2325 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2331 bond_enable_slave(struct iface *iface, bool enable)
2333 struct port *port = iface->port;
2334 struct bridge *br = port->bridge;
2336 /* This acts as a recursion check. If the act of disabling a slave
2337 * causes a different slave to be enabled, the flag will allow us to
2338 * skip redundant work when we reenter this function. It must be
2339 * cleared on exit to keep things safe with multiple bonds. */
2340 static bool moving_active_iface = false;
2342 iface->delay_expires = LLONG_MAX;
2343 if (enable == iface->enabled) {
2347 iface->enabled = enable;
2348 if (!iface->enabled) {
2349 VLOG_WARN("interface %s: disabled", iface->name);
2350 ofproto_revalidate(br->ofproto, iface->tag);
2351 if (iface->port_ifidx == port->active_iface) {
2352 /* Disabling a slave can lead to another slave being immediately
2353 * enabled if there will be no active slaves but one is waiting
2354 * on an updelay. In this case we do not need to run most of the
2355 * code for the newly enabled slave since there was no period
2356 * without an active slave and it is redundant with the disabling
2358 moving_active_iface = true;
2359 bond_choose_active_iface(port);
2361 bond_send_learning_packets(port);
2363 VLOG_WARN("interface %s: enabled", iface->name);
2364 if (port->active_iface < 0 && !moving_active_iface) {
2365 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2366 bond_choose_active_iface(port);
2367 bond_send_learning_packets(port);
2369 iface->tag = tag_create_random();
2372 moving_active_iface = false;
2375 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2376 * bond interface. */
2378 bond_update_fake_iface_stats(struct port *port)
2380 struct netdev_stats bond_stats;
2381 struct netdev *bond_dev;
2384 memset(&bond_stats, 0, sizeof bond_stats);
2386 for (i = 0; i < port->n_ifaces; i++) {
2387 struct netdev_stats slave_stats;
2389 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2390 /* XXX: We swap the stats here because they are swapped back when
2391 * reported by the internal device. The reason for this is
2392 * internal devices normally represent packets going into the system
2393 * but when used as fake bond device they represent packets leaving
2394 * the system. We really should do this in the internal device
2395 * itself because changing it here reverses the counts from the
2396 * perspective of the switch. However, the internal device doesn't
2397 * know what type of device it represents so we have to do it here
2399 bond_stats.tx_packets += slave_stats.rx_packets;
2400 bond_stats.tx_bytes += slave_stats.rx_bytes;
2401 bond_stats.rx_packets += slave_stats.tx_packets;
2402 bond_stats.rx_bytes += slave_stats.tx_bytes;
2406 if (!netdev_open_default(port->name, &bond_dev)) {
2407 netdev_set_stats(bond_dev, &bond_stats);
2408 netdev_close(bond_dev);
2413 bond_run(struct port *port)
2417 if (port->n_ifaces < 2) {
2421 for (i = 0; i < port->n_ifaces; i++) {
2422 bond_link_status_update(port->ifaces[i]);
2425 for (i = 0; i < port->n_ifaces; i++) {
2426 struct iface *iface = port->ifaces[i];
2427 if (time_msec() >= iface->delay_expires) {
2428 bond_enable_slave(iface, !iface->enabled);
2432 if (port->bond_fake_iface
2433 && time_msec() >= port->bond_next_fake_iface_update) {
2434 bond_update_fake_iface_stats(port);
2435 port->bond_next_fake_iface_update = time_msec() + 1000;
2440 bond_wait(struct port *port)
2444 if (port->n_ifaces < 2) {
2448 for (i = 0; i < port->n_ifaces; i++) {
2449 struct iface *iface = port->ifaces[i];
2450 if (iface->delay_expires != LLONG_MAX) {
2451 poll_timer_wait_until(iface->delay_expires);
2455 if (port->bond_fake_iface) {
2456 poll_timer_wait_until(port->bond_next_fake_iface_update);
2461 set_dst(struct dst *dst, const struct flow *flow,
2462 const struct port *in_port, const struct port *out_port,
2465 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2466 : in_port->vlan >= 0 ? in_port->vlan
2467 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2468 : vlan_tci_to_vid(flow->vlan_tci));
2469 return choose_output_iface(out_port, flow, dst->vlan,
2470 &dst->dp_ifidx, tags);
2474 swap_dst(struct dst *p, struct dst *q)
2476 struct dst tmp = *p;
2481 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2482 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2483 * that we push to the datapath. We could in fact fully sort the array by
2484 * vlan, but in most cases there are at most two different vlan tags so that's
2485 * possibly overkill.) */
2487 partition_dsts(struct dst_set *set, int vlan)
2489 struct dst *first = set->dsts;
2490 struct dst *last = set->dsts + set->n;
2492 while (first != last) {
2494 * - All dsts < first have vlan == 'vlan'.
2495 * - All dsts >= last have vlan != 'vlan'.
2496 * - first < last. */
2497 while (first->vlan == vlan) {
2498 if (++first == last) {
2503 /* Same invariants, plus one additional:
2504 * - first->vlan != vlan.
2506 while (last[-1].vlan != vlan) {
2507 if (--last == first) {
2512 /* Same invariants, plus one additional:
2513 * - last[-1].vlan == vlan.*/
2514 swap_dst(first++, --last);
2519 mirror_mask_ffs(mirror_mask_t mask)
2521 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2526 dst_set_init(struct dst_set *set)
2528 set->dsts = set->builtin;
2530 set->allocated = ARRAY_SIZE(set->builtin);
2534 dst_set_add(struct dst_set *set, const struct dst *dst)
2536 if (set->n >= set->allocated) {
2537 size_t new_allocated;
2538 struct dst *new_dsts;
2540 new_allocated = set->allocated * 2;
2541 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2542 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2546 set->dsts = new_dsts;
2547 set->allocated = new_allocated;
2549 set->dsts[set->n++] = *dst;
2553 dst_set_free(struct dst_set *set)
2555 if (set->dsts != set->builtin) {
2561 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2564 for (i = 0; i < set->n; i++) {
2565 if (set->dsts[i].vlan == test->vlan
2566 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2574 port_trunks_vlan(const struct port *port, uint16_t vlan)
2576 return (port->vlan < 0
2577 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2581 port_includes_vlan(const struct port *port, uint16_t vlan)
2583 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2587 port_is_floodable(const struct port *port)
2591 for (i = 0; i < port->n_ifaces; i++) {
2592 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2593 port->ifaces[i]->dp_ifidx)) {
2600 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2601 * there is no active iface. */
2603 port_get_active_iface_tag(const struct port *port)
2605 return (port->active_iface >= 0
2606 ? port->ifaces[port->active_iface]->tag
2607 : port->no_ifaces_tag);
2611 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2612 const struct port *in_port, const struct port *out_port,
2613 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2615 mirror_mask_t mirrors = in_port->src_mirrors;
2620 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2621 if (flow_vlan == 0) {
2622 flow_vlan = OFP_VLAN_NONE;
2625 if (out_port == FLOOD_PORT) {
2626 for (i = 0; i < br->n_ports; i++) {
2627 struct port *port = br->ports[i];
2629 && port_is_floodable(port)
2630 && port_includes_vlan(port, vlan)
2631 && !port->is_mirror_output_port
2632 && set_dst(&dst, flow, in_port, port, tags)) {
2633 mirrors |= port->dst_mirrors;
2634 dst_set_add(set, &dst);
2637 *nf_output_iface = NF_OUT_FLOOD;
2638 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2639 dst_set_add(set, &dst);
2640 *nf_output_iface = dst.dp_ifidx;
2641 mirrors |= out_port->dst_mirrors;
2645 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2646 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2648 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2649 && !dst_is_duplicate(set, &dst)) {
2650 dst_set_add(set, &dst);
2653 for (i = 0; i < br->n_ports; i++) {
2654 struct port *port = br->ports[i];
2655 if (port_includes_vlan(port, m->out_vlan)
2656 && set_dst(&dst, flow, in_port, port, tags))
2658 if (port->vlan < 0) {
2659 dst.vlan = m->out_vlan;
2661 if (dst_is_duplicate(set, &dst)) {
2665 /* Use the vlan tag on the original flow instead of
2666 * the one passed in the vlan parameter. This ensures
2667 * that we compare the vlan from before any implicit
2668 * tagging tags place. This is necessary because
2669 * dst->vlan is the final vlan, after removing implicit
2671 if (port == in_port && dst.vlan == flow_vlan) {
2672 /* Don't send out input port on same VLAN. */
2675 dst_set_add(set, &dst);
2680 mirrors &= mirrors - 1;
2683 partition_dsts(set, flow_vlan);
2686 static void OVS_UNUSED
2687 print_dsts(const struct dst_set *set)
2691 for (i = 0; i < set->n; i++) {
2692 const struct dst *dst = &set->dsts[i];
2694 printf(">p%"PRIu16, dst->dp_ifidx);
2695 if (dst->vlan != OFP_VLAN_NONE) {
2696 printf("v%"PRIu16, dst->vlan);
2702 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2703 const struct port *in_port, const struct port *out_port,
2704 tag_type *tags, struct ofpbuf *actions,
2705 uint16_t *nf_output_iface)
2712 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2715 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2716 if (cur_vlan == 0) {
2717 cur_vlan = OFP_VLAN_NONE;
2719 for (i = 0; i < set.n; i++) {
2720 const struct dst *dst = &set.dsts[i];
2721 if (dst->vlan != cur_vlan) {
2722 if (dst->vlan == OFP_VLAN_NONE) {
2723 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2726 tci = htons(dst->vlan & VLAN_VID_MASK);
2727 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2728 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2730 cur_vlan = dst->vlan;
2732 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2737 /* Returns the effective vlan of a packet, taking into account both the
2738 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2739 * the packet is untagged and -1 indicates it has an invalid header and
2740 * should be dropped. */
2741 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2742 struct port *in_port, bool have_packet)
2744 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2745 if (in_port->vlan >= 0) {
2747 /* XXX support double tagging? */
2749 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2750 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2751 "packet received on port %s configured with "
2752 "implicit VLAN %"PRIu16,
2753 br->name, vlan, in_port->name, in_port->vlan);
2757 vlan = in_port->vlan;
2759 if (!port_includes_vlan(in_port, vlan)) {
2761 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2762 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2763 "packet received on port %s not configured for "
2765 br->name, vlan, in_port->name, vlan);
2774 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2775 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2776 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2778 is_gratuitous_arp(const struct flow *flow)
2780 return (flow->dl_type == htons(ETH_TYPE_ARP)
2781 && eth_addr_is_broadcast(flow->dl_dst)
2782 && (flow->nw_proto == ARP_OP_REPLY
2783 || (flow->nw_proto == ARP_OP_REQUEST
2784 && flow->nw_src == flow->nw_dst)));
2788 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2789 struct port *in_port)
2791 enum grat_arp_lock_type lock_type;
2794 /* We don't want to learn from gratuitous ARP packets that are reflected
2795 * back over bond slaves so we lock the learning table. */
2796 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2797 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2798 GRAT_ARP_LOCK_CHECK;
2800 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2803 /* The log messages here could actually be useful in debugging,
2804 * so keep the rate limit relatively high. */
2805 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2807 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2808 "on port %s in VLAN %d",
2809 br->name, ETH_ADDR_ARGS(flow->dl_src),
2810 in_port->name, vlan);
2811 ofproto_revalidate(br->ofproto, rev_tag);
2815 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2816 * dropped. Returns true if they may be forwarded, false if they should be
2819 * If 'have_packet' is true, it indicates that the caller is processing a
2820 * received packet. If 'have_packet' is false, then the caller is just
2821 * revalidating an existing flow because configuration has changed. Either
2822 * way, 'have_packet' only affects logging (there is no point in logging errors
2823 * during revalidation).
2825 * Sets '*in_portp' to the input port. This will be a null pointer if
2826 * flow->in_port does not designate a known input port (in which case
2827 * is_admissible() returns false).
2829 * When returning true, sets '*vlanp' to the effective VLAN of the input
2830 * packet, as returned by flow_get_vlan().
2832 * May also add tags to '*tags', although the current implementation only does
2833 * so in one special case.
2836 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2837 tag_type *tags, int *vlanp, struct port **in_portp)
2839 struct iface *in_iface;
2840 struct port *in_port;
2843 /* Find the interface and port structure for the received packet. */
2844 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2846 /* No interface? Something fishy... */
2848 /* Odd. A few possible reasons here:
2850 * - We deleted an interface but there are still a few packets
2851 * queued up from it.
2853 * - Someone externally added an interface (e.g. with "ovs-dpctl
2854 * add-if") that we don't know about.
2856 * - Packet arrived on the local port but the local port is not
2857 * one of our bridge ports.
2859 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2861 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2862 "interface %"PRIu16, br->name, flow->in_port);
2868 *in_portp = in_port = in_iface->port;
2869 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2874 /* Drop frames for reserved multicast addresses. */
2875 if (eth_addr_is_reserved(flow->dl_dst)) {
2879 /* Drop frames on ports reserved for mirroring. */
2880 if (in_port->is_mirror_output_port) {
2882 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2883 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2884 "%s, which is reserved exclusively for mirroring",
2885 br->name, in_port->name);
2890 /* When using LACP, do not accept packets from disabled interfaces. */
2891 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2895 /* Packets received on non-LACP bonds need special attention to avoid
2897 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2899 bool is_grat_arp_locked;
2901 if (eth_addr_is_multicast(flow->dl_dst)) {
2902 *tags |= port_get_active_iface_tag(in_port);
2903 if (in_port->active_iface != in_iface->port_ifidx) {
2904 /* Drop all multicast packets on inactive slaves. */
2909 /* Drop all packets for which we have learned a different input
2910 * port, because we probably sent the packet on one slave and got
2911 * it back on the other. Gratuitous ARP packets are an exception
2912 * to this rule: the host has moved to another switch. The exception
2913 * to the exception is if we locked the learning table to avoid
2914 * reflections on bond slaves. If this is the case, just drop the
2916 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2917 &is_grat_arp_locked);
2918 if (src_idx != -1 && src_idx != in_port->port_idx &&
2919 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2927 /* If the composed actions may be applied to any packet in the given 'flow',
2928 * returns true. Otherwise, the actions should only be applied to 'packet', or
2929 * not at all, if 'packet' was NULL. */
2931 process_flow(struct bridge *br, const struct flow *flow,
2932 const struct ofpbuf *packet, struct ofpbuf *actions,
2933 tag_type *tags, uint16_t *nf_output_iface)
2935 struct port *in_port;
2936 struct port *out_port;
2940 /* Check whether we should drop packets in this flow. */
2941 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2946 /* Learn source MAC (but don't try to learn from revalidation). */
2948 update_learning_table(br, flow, vlan, in_port);
2951 /* Determine output port. */
2952 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2954 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2955 out_port = br->ports[out_port_idx];
2956 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2957 /* If we are revalidating but don't have a learning entry then
2958 * eject the flow. Installing a flow that floods packets opens
2959 * up a window of time where we could learn from a packet reflected
2960 * on a bond and blackhole packets before the learning table is
2961 * updated to reflect the correct port. */
2964 out_port = FLOOD_PORT;
2967 /* Don't send packets out their input ports. */
2968 if (in_port == out_port) {
2974 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2982 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2983 struct ofpbuf *actions, tag_type *tags,
2984 uint16_t *nf_output_iface, void *br_)
2986 struct bridge *br = br_;
2988 COVERAGE_INC(bridge_process_flow);
2989 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2993 bridge_special_ofhook_cb(const struct flow *flow,
2994 const struct ofpbuf *packet, void *br_)
2996 struct iface *iface;
2997 struct bridge *br = br_;
2999 iface = iface_from_dp_ifidx(br, flow->in_port);
3001 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3003 if (iface && iface->port->lacp && packet) {
3004 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
3007 COVERAGE_INC(bridge_process_lacp);
3008 lacp_process_pdu(iface->port->lacp, iface, pdu);
3018 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3019 const struct nlattr *actions,
3021 uint64_t n_bytes, void *br_)
3023 struct bridge *br = br_;
3024 const struct nlattr *a;
3025 struct port *in_port;
3030 /* Feed information from the active flows back into the learning table to
3031 * ensure that table is always in sync with what is actually flowing
3032 * through the datapath.
3034 * We test that 'tags' is nonzero to ensure that only flows that include an
3035 * OFPP_NORMAL action are used for learning. This works because
3036 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3037 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3038 update_learning_table(br, flow, vlan, in_port);
3041 /* Account for bond slave utilization. */
3042 if (!br->has_bonded_ports) {
3045 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3046 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3047 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3048 if (out_port && out_port->n_ifaces >= 2 &&
3049 out_port->bond_mode != BM_AB) {
3050 uint16_t vlan = (flow->vlan_tci
3051 ? vlan_tci_to_vid(flow->vlan_tci)
3053 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3054 e->tx_bytes += n_bytes;
3061 bridge_account_checkpoint_ofhook_cb(void *br_)
3063 struct bridge *br = br_;
3067 if (!br->has_bonded_ports) {
3072 for (i = 0; i < br->n_ports; i++) {
3073 struct port *port = br->ports[i];
3074 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3075 && now >= port->bond_next_rebalance) {
3076 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3077 bond_rebalance_port(port);
3082 static struct ofhooks bridge_ofhooks = {
3083 bridge_normal_ofhook_cb,
3084 bridge_special_ofhook_cb,
3085 bridge_account_flow_ofhook_cb,
3086 bridge_account_checkpoint_ofhook_cb,
3089 /* Bonding functions. */
3091 /* Statistics for a single interface on a bonded port, used for load-based
3092 * bond rebalancing. */
3093 struct slave_balance {
3094 struct iface *iface; /* The interface. */
3095 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3097 /* All the "bond_entry"s that are assigned to this interface, in order of
3098 * increasing tx_bytes. */
3099 struct bond_entry **hashes;
3104 bond_mode_to_string(enum bond_mode bm) {
3105 static char *bm_slb = "balance-slb";
3106 static char *bm_ab = "active-backup";
3107 static char *bm_tcp = "balance-tcp";
3110 case BM_SLB: return bm_slb;
3111 case BM_AB: return bm_ab;
3112 case BM_TCP: return bm_tcp;
3119 /* Sorts pointers to pointers to bond_entries in ascending order by the
3120 * interface to which they are assigned, and within a single interface in
3121 * ascending order of bytes transmitted. */
3123 compare_bond_entries(const void *a_, const void *b_)
3125 const struct bond_entry *const *ap = a_;
3126 const struct bond_entry *const *bp = b_;
3127 const struct bond_entry *a = *ap;
3128 const struct bond_entry *b = *bp;
3129 if (a->iface_idx != b->iface_idx) {
3130 return a->iface_idx > b->iface_idx ? 1 : -1;
3131 } else if (a->tx_bytes != b->tx_bytes) {
3132 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3138 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3139 * *descending* order by number of bytes transmitted. */
3141 compare_slave_balance(const void *a_, const void *b_)
3143 const struct slave_balance *a = a_;
3144 const struct slave_balance *b = b_;
3145 if (a->iface->enabled != b->iface->enabled) {
3146 return a->iface->enabled ? -1 : 1;
3147 } else if (a->tx_bytes != b->tx_bytes) {
3148 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3155 swap_bals(struct slave_balance *a, struct slave_balance *b)
3157 struct slave_balance tmp = *a;
3162 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3163 * given that 'p' (and only 'p') might be in the wrong location.
3165 * This function invalidates 'p', since it might now be in a different memory
3168 resort_bals(struct slave_balance *p,
3169 struct slave_balance bals[], size_t n_bals)
3172 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3173 swap_bals(p, p - 1);
3175 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3176 swap_bals(p, p + 1);
3182 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3184 if (VLOG_IS_DBG_ENABLED()) {
3185 struct ds ds = DS_EMPTY_INITIALIZER;
3186 const struct slave_balance *b;
3188 for (b = bals; b < bals + n_bals; b++) {
3192 ds_put_char(&ds, ',');
3194 ds_put_format(&ds, " %s %"PRIu64"kB",
3195 b->iface->name, b->tx_bytes / 1024);
3197 if (!b->iface->enabled) {
3198 ds_put_cstr(&ds, " (disabled)");
3200 if (b->n_hashes > 0) {
3201 ds_put_cstr(&ds, " (");
3202 for (i = 0; i < b->n_hashes; i++) {
3203 const struct bond_entry *e = b->hashes[i];
3205 ds_put_cstr(&ds, " + ");
3207 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3208 e - port->bond_hash, e->tx_bytes / 1024);
3210 ds_put_cstr(&ds, ")");
3213 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3218 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3220 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3223 struct bond_entry *hash = from->hashes[hash_idx];
3224 struct port *port = from->iface->port;
3225 uint64_t delta = hash->tx_bytes;
3227 assert(port->bond_mode != BM_AB);
3229 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3230 "from %s to %s (now carrying %"PRIu64"kB and "
3231 "%"PRIu64"kB load, respectively)",
3232 port->name, delta / 1024, hash - port->bond_hash,
3233 from->iface->name, to->iface->name,
3234 (from->tx_bytes - delta) / 1024,
3235 (to->tx_bytes + delta) / 1024);
3237 /* Delete element from from->hashes.
3239 * We don't bother to add the element to to->hashes because not only would
3240 * it require more work, the only purpose it would be to allow that hash to
3241 * be migrated to another slave in this rebalancing run, and there is no
3242 * point in doing that. */
3243 if (hash_idx == 0) {
3246 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3247 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3251 /* Shift load away from 'from' to 'to'. */
3252 from->tx_bytes -= delta;
3253 to->tx_bytes += delta;
3255 /* Arrange for flows to be revalidated. */
3256 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3257 hash->iface_idx = to->iface->port_ifidx;
3258 hash->iface_tag = tag_create_random();
3262 bond_rebalance_port(struct port *port)
3264 struct slave_balance *bals;
3266 struct bond_entry *hashes[BOND_MASK + 1];
3267 struct slave_balance *b, *from, *to;
3268 struct bond_entry *e;
3271 assert(port->bond_mode != BM_AB);
3273 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3274 * descending order of tx_bytes, so that bals[0] represents the most
3275 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3278 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3279 * array for each slave_balance structure, we sort our local array of
3280 * hashes in order by slave, so that all of the hashes for a given slave
3281 * become contiguous in memory, and then we point each 'hashes' members of
3282 * a slave_balance structure to the start of a contiguous group. */
3283 n_bals = port->n_ifaces;
3284 bals = xmalloc(n_bals * sizeof *bals);
3285 for (b = bals; b < &bals[n_bals]; b++) {
3286 b->iface = port->ifaces[b - bals];
3291 for (i = 0; i <= BOND_MASK; i++) {
3292 hashes[i] = &port->bond_hash[i];
3294 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3295 for (i = 0; i <= BOND_MASK; i++) {
3297 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3298 b = &bals[e->iface_idx];
3299 b->tx_bytes += e->tx_bytes;
3301 b->hashes = &hashes[i];
3306 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3307 log_bals(bals, n_bals, port);
3309 /* Discard slaves that aren't enabled (which were sorted to the back of the
3310 * array earlier). */
3311 while (!bals[n_bals - 1].iface->enabled) {
3318 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3319 to = &bals[n_bals - 1];
3320 for (from = bals; from < to; ) {
3321 uint64_t overload = from->tx_bytes - to->tx_bytes;
3322 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3323 /* The extra load on 'from' (and all less-loaded slaves), compared
3324 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3325 * it is less than ~1Mbps. No point in rebalancing. */
3327 } else if (from->n_hashes == 1) {
3328 /* 'from' only carries a single MAC hash, so we can't shift any
3329 * load away from it, even though we want to. */
3332 /* 'from' is carrying significantly more load than 'to', and that
3333 * load is split across at least two different hashes. Pick a hash
3334 * to migrate to 'to' (the least-loaded slave), given that doing so
3335 * must decrease the ratio of the load on the two slaves by at
3338 * The sort order we use means that we prefer to shift away the
3339 * smallest hashes instead of the biggest ones. There is little
3340 * reason behind this decision; we could use the opposite sort
3341 * order to shift away big hashes ahead of small ones. */
3344 for (i = 0; i < from->n_hashes; i++) {
3345 double old_ratio, new_ratio;
3346 uint64_t delta = from->hashes[i]->tx_bytes;
3348 if (delta == 0 || from->tx_bytes - delta == 0) {
3349 /* Pointless move. */
3353 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3355 if (to->tx_bytes == 0) {
3356 /* Nothing on the new slave, move it. */
3360 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3361 new_ratio = (double)(from->tx_bytes - delta) /
3362 (to->tx_bytes + delta);
3364 if (new_ratio == 0) {
3365 /* Should already be covered but check to prevent division
3370 if (new_ratio < 1) {
3371 new_ratio = 1 / new_ratio;
3374 if (old_ratio - new_ratio > 0.1) {
3375 /* Would decrease the ratio, move it. */
3379 if (i < from->n_hashes) {
3380 bond_shift_load(from, to, i);
3382 /* If the result of the migration changed the relative order of
3383 * 'from' and 'to' swap them back to maintain invariants. */
3384 if (order_swapped) {
3385 swap_bals(from, to);
3388 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3389 * point to different slave_balance structures. It is only
3390 * valid to do these two operations in a row at all because we
3391 * know that 'from' will not move past 'to' and vice versa. */
3392 resort_bals(from, bals, n_bals);
3393 resort_bals(to, bals, n_bals);
3400 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3401 * historical data to decay to <1% in 7 rebalancing runs. */
3402 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3411 bond_send_learning_packets(struct port *port)
3413 struct bridge *br = port->bridge;
3414 struct mac_entry *e;
3415 struct ofpbuf packet;
3416 int error, n_packets, n_errors;
3418 if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
3422 ofpbuf_init(&packet, 128);
3423 error = n_packets = n_errors = 0;
3424 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3430 if (e->port == port->port_idx) {
3434 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3436 flow_extract(&packet, 0, ODPP_NONE, &flow);
3438 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3444 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3450 ofpbuf_uninit(&packet);
3453 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3454 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3455 "packets, last error was: %s",
3456 port->name, n_errors, n_packets, strerror(error));
3458 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3459 port->name, n_packets);
3463 /* Bonding unixctl user interface functions. */
3466 bond_unixctl_list(struct unixctl_conn *conn,
3467 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3469 struct ds ds = DS_EMPTY_INITIALIZER;
3470 const struct bridge *br;
3472 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3474 LIST_FOR_EACH (br, node, &all_bridges) {
3477 for (i = 0; i < br->n_ports; i++) {
3478 const struct port *port = br->ports[i];
3479 if (port->n_ifaces > 1) {
3482 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3483 bond_mode_to_string(port->bond_mode));
3484 for (j = 0; j < port->n_ifaces; j++) {
3485 const struct iface *iface = port->ifaces[j];
3487 ds_put_cstr(&ds, ", ");
3489 ds_put_cstr(&ds, iface->name);
3491 ds_put_char(&ds, '\n');
3495 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3499 static struct port *
3500 bond_find(const char *name)
3502 const struct bridge *br;
3504 LIST_FOR_EACH (br, node, &all_bridges) {
3507 for (i = 0; i < br->n_ports; i++) {
3508 struct port *port = br->ports[i];
3509 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3518 bond_unixctl_show(struct unixctl_conn *conn,
3519 const char *args, void *aux OVS_UNUSED)
3521 struct ds ds = DS_EMPTY_INITIALIZER;
3522 const struct port *port;
3525 port = bond_find(args);
3527 unixctl_command_reply(conn, 501, "no such bond");
3531 ds_put_format(&ds, "bond_mode: %s\n",
3532 bond_mode_to_string(port->bond_mode));
3535 ds_put_format(&ds, "lacp: %s\n",
3536 port->lacp_active ? "active" : "passive");
3538 ds_put_cstr(&ds, "lacp: off\n");
3541 if (port->bond_mode != BM_AB) {
3542 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3543 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3547 ds_put_format(&ds, "bond-detect-mode: %s\n",
3548 port->monitor ? "carrier" : "miimon");
3550 if (!port->monitor) {
3551 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3552 port->miimon_interval);
3555 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3556 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3558 if (port->bond_mode != BM_AB) {
3559 ds_put_format(&ds, "next rebalance: %lld ms\n",
3560 port->bond_next_rebalance - time_msec());
3563 for (j = 0; j < port->n_ifaces; j++) {
3564 const struct iface *iface = port->ifaces[j];
3565 struct bond_entry *be;
3569 ds_put_format(&ds, "\nslave %s: %s\n",
3570 iface->name, iface->enabled ? "enabled" : "disabled");
3571 if (j == port->active_iface) {
3572 ds_put_cstr(&ds, "\tactive slave\n");
3574 if (iface->delay_expires != LLONG_MAX) {
3575 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3576 iface->enabled ? "downdelay" : "updelay",
3577 iface->delay_expires - time_msec());
3580 if (port->bond_mode == BM_AB) {
3585 memset(&flow, 0, sizeof flow);
3586 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3587 int hash = be - port->bond_hash;
3588 struct mac_entry *me;
3590 if (be->iface_idx != j) {
3594 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3595 hash, be->tx_bytes / 1024);
3597 if (port->bond_mode != BM_SLB) {
3602 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3606 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3607 if (bond_hash_src(me->mac, me->vlan) == hash
3608 && me->port != port->port_idx
3609 && choose_output_iface(port, &flow, me->vlan,
3611 && dp_ifidx == iface->dp_ifidx)
3613 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3614 ETH_ADDR_ARGS(me->mac));
3619 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3624 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3625 void *aux OVS_UNUSED)
3627 char *args = (char *) args_;
3628 char *save_ptr = NULL;
3629 char *bond_s, *hash_s, *slave_s;
3631 struct iface *iface;
3632 struct bond_entry *entry;
3635 bond_s = strtok_r(args, " ", &save_ptr);
3636 hash_s = strtok_r(NULL, " ", &save_ptr);
3637 slave_s = strtok_r(NULL, " ", &save_ptr);
3639 unixctl_command_reply(conn, 501,
3640 "usage: bond/migrate BOND HASH SLAVE");
3644 port = bond_find(bond_s);
3646 unixctl_command_reply(conn, 501, "no such bond");
3650 if (port->bond_mode != BM_SLB) {
3651 unixctl_command_reply(conn, 501, "not an SLB bond");
3655 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3656 hash = atoi(hash_s) & BOND_MASK;
3658 unixctl_command_reply(conn, 501, "bad hash");
3662 iface = port_lookup_iface(port, slave_s);
3664 unixctl_command_reply(conn, 501, "no such slave");
3668 if (!iface->enabled) {
3669 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3673 entry = &port->bond_hash[hash];
3674 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3675 entry->iface_idx = iface->port_ifidx;
3676 entry->iface_tag = tag_create_random();
3677 unixctl_command_reply(conn, 200, "migrated");
3681 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3682 void *aux OVS_UNUSED)
3684 char *args = (char *) args_;
3685 char *save_ptr = NULL;
3686 char *bond_s, *slave_s;
3688 struct iface *iface;
3690 bond_s = strtok_r(args, " ", &save_ptr);
3691 slave_s = strtok_r(NULL, " ", &save_ptr);
3693 unixctl_command_reply(conn, 501,
3694 "usage: bond/set-active-slave BOND SLAVE");
3698 port = bond_find(bond_s);
3700 unixctl_command_reply(conn, 501, "no such bond");
3704 iface = port_lookup_iface(port, slave_s);
3706 unixctl_command_reply(conn, 501, "no such slave");
3710 if (!iface->enabled) {
3711 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3715 if (port->active_iface != iface->port_ifidx) {
3716 ofproto_revalidate(port->bridge->ofproto,
3717 port_get_active_iface_tag(port));
3718 port->active_iface = iface->port_ifidx;
3719 VLOG_INFO("port %s: active interface is now %s",
3720 port->name, iface->name);
3721 bond_send_learning_packets(port);
3722 unixctl_command_reply(conn, 200, "done");
3724 unixctl_command_reply(conn, 200, "no change");
3729 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3731 char *args = (char *) args_;
3732 char *save_ptr = NULL;
3733 char *bond_s, *slave_s;
3735 struct iface *iface;
3737 bond_s = strtok_r(args, " ", &save_ptr);
3738 slave_s = strtok_r(NULL, " ", &save_ptr);
3740 unixctl_command_reply(conn, 501,
3741 "usage: bond/enable/disable-slave BOND SLAVE");
3745 port = bond_find(bond_s);
3747 unixctl_command_reply(conn, 501, "no such bond");
3751 iface = port_lookup_iface(port, slave_s);
3753 unixctl_command_reply(conn, 501, "no such slave");
3757 bond_enable_slave(iface, enable);
3758 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3762 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3763 void *aux OVS_UNUSED)
3765 enable_slave(conn, args, true);
3769 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3770 void *aux OVS_UNUSED)
3772 enable_slave(conn, args, false);
3776 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3777 void *aux OVS_UNUSED)
3779 char *args = (char *) args_;
3780 uint8_t mac[ETH_ADDR_LEN];
3784 char *mac_s, *vlan_s;
3785 char *save_ptr = NULL;
3787 mac_s = strtok_r(args, " ", &save_ptr);
3788 vlan_s = strtok_r(NULL, " ", &save_ptr);
3791 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3792 unixctl_command_reply(conn, 501, "invalid vlan");
3796 vlan = OFP_VLAN_NONE;
3799 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3800 == ETH_ADDR_SCAN_COUNT) {
3801 hash = bond_hash_src(mac, vlan);
3803 hash_cstr = xasprintf("%u", hash);
3804 unixctl_command_reply(conn, 200, hash_cstr);
3807 unixctl_command_reply(conn, 501, "invalid mac");
3814 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3815 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3816 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3817 unixctl_command_register("bond/set-active-slave",
3818 bond_unixctl_set_active_slave, NULL);
3819 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3821 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3823 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3826 /* Port functions. */
3829 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3831 struct iface *iface = aux;
3832 uint8_t ea[ETH_ADDR_LEN];
3835 error = netdev_get_etheraddr(iface->netdev, ea);
3837 struct ofpbuf packet;
3839 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3840 compose_lacp_packet(&packet, ea, pdu);
3841 ofproto_send_packet(iface->port->bridge->ofproto,
3842 iface->dp_ifidx, 0, &packet);
3843 ofpbuf_uninit(&packet);
3845 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3846 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3847 "(%s)", iface->name, strerror(error));
3852 port_run(struct port *port)
3854 if (port->monitor) {
3857 /* Track carrier going up and down on interfaces. */
3858 while (!netdev_monitor_poll(port->monitor, &devname)) {
3859 struct iface *iface;
3861 iface = port_lookup_iface(port, devname);
3863 iface_update_carrier(iface);
3867 } else if (time_msec() >= port->miimon_next_update) {
3870 for (i = 0; i < port->n_ifaces; i++) {
3871 struct iface *iface = port->ifaces[i];
3872 iface_update_carrier(iface);
3874 port->miimon_next_update = time_msec() + port->miimon_interval;
3880 for (i = 0; i < port->n_ifaces; i++) {
3881 struct iface *iface = port->ifaces[i];
3882 lacp_slave_enable(port->lacp, iface, iface->enabled);
3885 lacp_run(port->lacp, lacp_send_pdu_cb);
3892 port_wait(struct port *port)
3894 if (port->monitor) {
3895 netdev_monitor_poll_wait(port->monitor);
3897 poll_timer_wait_until(port->miimon_next_update);
3901 lacp_wait(port->lacp);
3907 static struct port *
3908 port_create(struct bridge *br, const char *name)
3912 port = xzalloc(sizeof *port);
3914 port->port_idx = br->n_ports;
3916 port->trunks = NULL;
3917 port->name = xstrdup(name);
3918 port->active_iface = -1;
3920 if (br->n_ports >= br->allocated_ports) {
3921 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3924 br->ports[br->n_ports++] = port;
3925 shash_add_assert(&br->port_by_name, port->name, port);
3927 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3934 get_port_other_config(const struct ovsrec_port *port, const char *key,
3935 const char *default_value)
3939 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3941 return value ? value : default_value;
3945 get_interface_other_config(const struct ovsrec_interface *iface,
3946 const char *key, const char *default_value)
3950 value = get_ovsrec_key_value(&iface->header_,
3951 &ovsrec_interface_col_other_config, key);
3952 return value ? value : default_value;
3956 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3958 struct shash new_ifaces;
3961 /* Collect list of new interfaces. */
3962 shash_init(&new_ifaces);
3963 for (i = 0; i < cfg->n_interfaces; i++) {
3964 const char *name = cfg->interfaces[i]->name;
3965 shash_add_once(&new_ifaces, name, NULL);
3968 /* Get rid of deleted interfaces. */
3969 for (i = 0; i < port->n_ifaces; ) {
3970 struct iface *iface = port->ifaces[i];
3971 if (!shash_find(&new_ifaces, iface->name)) {
3972 iface_destroy(iface);
3978 shash_destroy(&new_ifaces);
3982 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3984 const char *detect_mode;
3985 struct shash new_ifaces;
3986 long long int next_rebalance, miimon_next_update, lacp_priority;
3987 unsigned long *trunks;
3993 /* Update settings. */
3994 port->updelay = cfg->bond_updelay;
3995 if (port->updelay < 0) {
3998 port->downdelay = cfg->bond_downdelay;
3999 if (port->downdelay < 0) {
4000 port->downdelay = 0;
4002 port->bond_rebalance_interval = atoi(
4003 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4004 if (port->bond_rebalance_interval < 1000) {
4005 port->bond_rebalance_interval = 1000;
4007 next_rebalance = time_msec() + port->bond_rebalance_interval;
4008 if (port->bond_next_rebalance > next_rebalance) {
4009 port->bond_next_rebalance = next_rebalance;
4012 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4015 netdev_monitor_destroy(port->monitor);
4016 port->monitor = NULL;
4018 if (strcmp(detect_mode, "miimon")) {
4019 port->monitor = netdev_monitor_create();
4021 if (strcmp(detect_mode, "carrier")) {
4022 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4023 "defaulting to carrier", port->name, detect_mode);
4027 port->miimon_interval = atoi(
4028 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4029 if (port->miimon_interval < 100) {
4030 port->miimon_interval = 100;
4032 miimon_next_update = time_msec() + port->miimon_interval;
4033 if (port->miimon_next_update > miimon_next_update) {
4034 port->miimon_next_update = miimon_next_update;
4037 if (!port->cfg->bond_mode ||
4038 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4039 port->bond_mode = BM_SLB;
4040 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4041 port->bond_mode = BM_AB;
4042 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4043 port->bond_mode = BM_TCP;
4045 port->bond_mode = BM_SLB;
4046 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4047 port->name, port->cfg->bond_mode,
4048 bond_mode_to_string(port->bond_mode));
4051 /* Add new interfaces and update 'cfg' member of existing ones. */
4052 shash_init(&new_ifaces);
4053 for (i = 0; i < cfg->n_interfaces; i++) {
4054 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4055 struct iface *iface;
4057 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4058 VLOG_WARN("port %s: %s specified twice as port interface",
4059 port->name, if_cfg->name);
4060 iface_set_ofport(if_cfg, -1);
4064 iface = iface_lookup(port->bridge, if_cfg->name);
4066 if (iface->port != port) {
4067 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4069 port->bridge->name, if_cfg->name, iface->port->name);
4072 iface->cfg = if_cfg;
4074 iface = iface_create(port, if_cfg);
4077 /* Determine interface type. The local port always has type
4078 * "internal". Other ports take their type from the database and
4079 * default to "system" if none is specified. */
4080 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4081 : if_cfg->type[0] ? if_cfg->type
4085 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4088 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4089 iface->lacp_priority = UINT16_MAX;
4091 iface->lacp_priority = lacp_priority;
4094 shash_destroy(&new_ifaces);
4096 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4100 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4102 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4103 /* Prefer bondable links if unspecified. */
4104 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4106 port->lacp_priority = lacp_priority;
4109 if (!port->cfg->lacp) {
4110 /* XXX when LACP implementation has been sufficiently tested, enable by
4111 * default and make active on bonded ports. */
4112 lacp_destroy(port->lacp);
4114 } else if (!strcmp(port->cfg->lacp, "off")) {
4115 lacp_destroy(port->lacp);
4117 } else if (!strcmp(port->cfg->lacp, "active")) {
4119 port->lacp = lacp_create();
4121 port->lacp_active = true;
4122 } else if (!strcmp(port->cfg->lacp, "passive")) {
4124 port->lacp = lacp_create();
4126 port->lacp_active = false;
4128 VLOG_WARN("port %s: unknown LACP mode %s",
4129 port->name, port->cfg->lacp);
4130 lacp_destroy(port->lacp);
4137 if (port->n_ifaces < 2) {
4139 if (vlan >= 0 && vlan <= 4095) {
4140 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4145 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4146 * they even work as-is. But they have not been tested. */
4147 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4151 if (port->vlan != vlan) {
4153 bridge_flush(port->bridge);
4156 /* Get trunked VLANs. */
4158 if (vlan < 0 && cfg->n_trunks) {
4161 trunks = bitmap_allocate(4096);
4163 for (i = 0; i < cfg->n_trunks; i++) {
4164 int trunk = cfg->trunks[i];
4166 bitmap_set1(trunks, trunk);
4172 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4173 port->name, cfg->n_trunks);
4175 if (n_errors == cfg->n_trunks) {
4176 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4178 bitmap_free(trunks);
4181 } else if (vlan >= 0 && cfg->n_trunks) {
4182 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4186 ? port->trunks != NULL
4187 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4188 bridge_flush(port->bridge);
4190 bitmap_free(port->trunks);
4191 port->trunks = trunks;
4195 port_destroy(struct port *port)
4198 struct bridge *br = port->bridge;
4202 for (i = 0; i < MAX_MIRRORS; i++) {
4203 struct mirror *m = br->mirrors[i];
4204 if (m && m->out_port == port) {
4209 while (port->n_ifaces > 0) {
4210 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4213 shash_find_and_delete_assert(&br->port_by_name, port->name);
4215 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4216 del->port_idx = port->port_idx;
4218 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4220 netdev_monitor_destroy(port->monitor);
4222 bitmap_free(port->trunks);
4229 static struct port *
4230 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4232 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4233 return iface ? iface->port : NULL;
4236 static struct port *
4237 port_lookup(const struct bridge *br, const char *name)
4239 return shash_find_data(&br->port_by_name, name);
4242 static struct iface *
4243 port_lookup_iface(const struct port *port, const char *name)
4245 struct iface *iface = iface_lookup(port->bridge, name);
4246 return iface && iface->port == port ? iface : NULL;
4250 port_update_lacp(struct port *port)
4255 lacp_configure(port->lacp, port->name,
4256 port->bridge->ea, port->lacp_priority,
4257 port->lacp_active, port->lacp_fast);
4259 for (i = 0; i < port->n_ifaces; i++) {
4260 struct iface *iface = port->ifaces[i];
4261 lacp_slave_register(port->lacp, iface, iface->name,
4262 iface->dp_ifidx, iface->lacp_priority);
4268 port_update_bonding(struct port *port)
4270 if (port->n_ifaces < 2) {
4271 /* Not a bonded port. */
4272 free(port->bond_hash);
4273 port->bond_hash = NULL;
4274 port->bond_fake_iface = false;
4275 port->active_iface = -1;
4276 port->no_ifaces_tag = 0;
4280 if (port->bond_mode != BM_AB && !port->bond_hash) {
4281 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4282 for (i = 0; i <= BOND_MASK; i++) {
4283 struct bond_entry *e = &port->bond_hash[i];
4287 port->bond_next_rebalance
4288 = time_msec() + port->bond_rebalance_interval;
4289 } else if (port->bond_mode == BM_AB) {
4290 free(port->bond_hash);
4291 port->bond_hash = NULL;
4294 if (!port->no_ifaces_tag) {
4295 port->no_ifaces_tag = tag_create_random();
4298 if (port->active_iface < 0) {
4299 bond_choose_active_iface(port);
4302 port->bond_fake_iface = port->cfg->bond_fake_iface;
4303 if (port->bond_fake_iface) {
4304 port->bond_next_fake_iface_update = time_msec();
4310 /* Interface functions. */
4312 static struct iface *
4313 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4315 struct bridge *br = port->bridge;
4316 struct iface *iface;
4317 char *name = if_cfg->name;
4319 iface = xzalloc(sizeof *iface);
4321 iface->port_ifidx = port->n_ifaces;
4322 iface->name = xstrdup(name);
4323 iface->dp_ifidx = -1;
4324 iface->tag = tag_create_random();
4325 iface->delay_expires = LLONG_MAX;
4326 iface->netdev = NULL;
4327 iface->cfg = if_cfg;
4329 shash_add_assert(&br->iface_by_name, iface->name, iface);
4331 if (port->n_ifaces >= port->allocated_ifaces) {
4332 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4333 sizeof *port->ifaces);
4335 port->ifaces[port->n_ifaces++] = iface;
4336 if (port->n_ifaces > 1) {
4337 br->has_bonded_ports = true;
4340 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4348 iface_destroy(struct iface *iface)
4351 struct port *port = iface->port;
4352 struct bridge *br = port->bridge;
4353 bool del_active = port->active_iface == iface->port_ifidx;
4356 if (iface->port->lacp) {
4357 lacp_slave_unregister(iface->port->lacp, iface);
4360 if (port->monitor && iface->netdev) {
4361 netdev_monitor_remove(port->monitor, iface->netdev);
4364 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4366 if (iface->dp_ifidx >= 0) {
4367 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4370 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4371 del->port_ifidx = iface->port_ifidx;
4373 netdev_close(iface->netdev);
4376 bond_choose_active_iface(port);
4377 bond_send_learning_packets(port);
4383 bridge_flush(port->bridge);
4387 static struct iface *
4388 iface_lookup(const struct bridge *br, const char *name)
4390 return shash_find_data(&br->iface_by_name, name);
4393 static struct iface *
4394 iface_find(const char *name)
4396 const struct bridge *br;
4398 LIST_FOR_EACH (br, node, &all_bridges) {
4399 struct iface *iface = iface_lookup(br, name);
4408 static struct iface *
4409 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4411 struct iface *iface;
4413 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4414 hash_int(dp_ifidx, 0), &br->ifaces) {
4415 if (iface->dp_ifidx == dp_ifidx) {
4422 /* Set Ethernet address of 'iface', if one is specified in the configuration
4425 iface_set_mac(struct iface *iface)
4427 uint8_t ea[ETH_ADDR_LEN];
4429 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4430 if (eth_addr_is_multicast(ea)) {
4431 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4433 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4434 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4435 iface->name, iface->name);
4437 int error = netdev_set_etheraddr(iface->netdev, ea);
4439 VLOG_ERR("interface %s: setting MAC failed (%s)",
4440 iface->name, strerror(error));
4446 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4448 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4451 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4455 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4457 * The value strings in '*shash' are taken directly from values[], not copied,
4458 * so the caller should not modify or free them. */
4460 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4461 struct shash *shash)
4466 for (i = 0; i < n; i++) {
4467 shash_add(shash, keys[i], values[i]);
4471 /* Creates 'keys' and 'values' arrays from 'shash'.
4473 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4474 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4475 * are populated with with strings taken directly from 'shash' and thus have
4476 * the same ownership of the key-value pairs in shash.
4479 shash_to_ovs_idl_map(struct shash *shash,
4480 char ***keys, char ***values, size_t *n)
4484 struct shash_node *sn;
4486 count = shash_count(shash);
4488 k = xmalloc(count * sizeof *k);
4489 v = xmalloc(count * sizeof *v);
4492 SHASH_FOR_EACH(sn, shash) {
4503 struct iface_delete_queues_cbdata {
4504 struct netdev *netdev;
4505 const struct ovsdb_datum *queues;
4509 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4511 union ovsdb_atom atom;
4513 atom.integer = target;
4514 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4518 iface_delete_queues(unsigned int queue_id,
4519 const struct shash *details OVS_UNUSED, void *cbdata_)
4521 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4523 if (!queue_ids_include(cbdata->queues, queue_id)) {
4524 netdev_delete_queue(cbdata->netdev, queue_id);
4529 iface_update_carrier(struct iface *iface)
4531 bool carrier = iface_get_carrier(iface);
4532 if (carrier == iface->up) {
4536 iface->up = carrier;
4537 if (iface->port->lacp) {
4538 lacp_slave_carrier_changed(iface->port->lacp, iface);
4543 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4545 if (!qos || qos->type[0] == '\0') {
4546 netdev_set_qos(iface->netdev, NULL, NULL);
4548 struct iface_delete_queues_cbdata cbdata;
4549 struct shash details;
4552 /* Configure top-level Qos for 'iface'. */
4553 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4554 qos->n_other_config, &details);
4555 netdev_set_qos(iface->netdev, qos->type, &details);
4556 shash_destroy(&details);
4558 /* Deconfigure queues that were deleted. */
4559 cbdata.netdev = iface->netdev;
4560 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4562 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4564 /* Configure queues for 'iface'. */
4565 for (i = 0; i < qos->n_queues; i++) {
4566 const struct ovsrec_queue *queue = qos->value_queues[i];
4567 unsigned int queue_id = qos->key_queues[i];
4569 shash_from_ovs_idl_map(queue->key_other_config,
4570 queue->value_other_config,
4571 queue->n_other_config, &details);
4572 netdev_set_queue(iface->netdev, queue_id, &details);
4573 shash_destroy(&details);
4579 iface_update_cfm(struct iface *iface)
4583 uint16_t *remote_mps;
4584 struct ovsrec_monitor *mon;
4585 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4587 mon = iface->cfg->monitor;
4590 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4594 if (netdev_get_etheraddr(iface->netdev, ea)) {
4595 VLOG_WARN("interface %s: Failed to get ethernet address. "
4596 "Skipping Monitor.", iface->name);
4600 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4601 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4605 cfm.mpid = mon->mpid;
4606 cfm.interval = mon->interval ? *mon->interval : 1000;
4608 memcpy(cfm.eth_src, ea, sizeof cfm.eth_src);
4609 memcpy(cfm.maid, maid, sizeof cfm.maid);
4611 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4612 for(i = 0; i < mon->n_remote_mps; i++) {
4613 remote_mps[i] = mon->remote_mps[i]->mpid;
4616 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4617 &cfm, remote_mps, mon->n_remote_mps);
4621 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4622 * how 'iface''s port is configured.
4624 * Returns true if 'iface' is up, false otherwise. */
4626 iface_get_carrier(const struct iface *iface)
4628 return (iface->port->monitor
4629 ? netdev_get_carrier(iface->netdev)
4630 : netdev_get_miimon(iface->netdev));
4633 /* Port mirroring. */
4635 static struct mirror *
4636 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4640 for (i = 0; i < MAX_MIRRORS; i++) {
4641 struct mirror *m = br->mirrors[i];
4642 if (m && uuid_equals(uuid, &m->uuid)) {
4650 mirror_reconfigure(struct bridge *br)
4652 unsigned long *rspan_vlans;
4655 /* Get rid of deleted mirrors. */
4656 for (i = 0; i < MAX_MIRRORS; i++) {
4657 struct mirror *m = br->mirrors[i];
4659 const struct ovsdb_datum *mc;
4660 union ovsdb_atom atom;
4662 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4663 atom.uuid = br->mirrors[i]->uuid;
4664 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4670 /* Add new mirrors and reconfigure existing ones. */
4671 for (i = 0; i < br->cfg->n_mirrors; i++) {
4672 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4673 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4675 mirror_reconfigure_one(m, cfg);
4677 mirror_create(br, cfg);
4681 /* Update port reserved status. */
4682 for (i = 0; i < br->n_ports; i++) {
4683 br->ports[i]->is_mirror_output_port = false;
4685 for (i = 0; i < MAX_MIRRORS; i++) {
4686 struct mirror *m = br->mirrors[i];
4687 if (m && m->out_port) {
4688 m->out_port->is_mirror_output_port = true;
4692 /* Update flooded vlans (for RSPAN). */
4694 if (br->cfg->n_flood_vlans) {
4695 rspan_vlans = bitmap_allocate(4096);
4697 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4698 int64_t vlan = br->cfg->flood_vlans[i];
4699 if (vlan >= 0 && vlan < 4096) {
4700 bitmap_set1(rspan_vlans, vlan);
4701 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4704 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4709 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4715 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4720 for (i = 0; ; i++) {
4721 if (i >= MAX_MIRRORS) {
4722 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4723 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4726 if (!br->mirrors[i]) {
4731 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4734 br->mirrors[i] = m = xzalloc(sizeof *m);
4737 m->name = xstrdup(cfg->name);
4738 shash_init(&m->src_ports);
4739 shash_init(&m->dst_ports);
4745 mirror_reconfigure_one(m, cfg);
4749 mirror_destroy(struct mirror *m)
4752 struct bridge *br = m->bridge;
4755 for (i = 0; i < br->n_ports; i++) {
4756 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4757 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4760 shash_destroy(&m->src_ports);
4761 shash_destroy(&m->dst_ports);
4764 m->bridge->mirrors[m->idx] = NULL;
4773 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4774 struct shash *names)
4778 for (i = 0; i < n_ports; i++) {
4779 const char *name = ports[i]->name;
4780 if (port_lookup(m->bridge, name)) {
4781 shash_add_once(names, name, NULL);
4783 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4784 "port %s", m->bridge->name, m->name, name);
4790 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4796 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4798 for (i = 0; i < cfg->n_select_vlan; i++) {
4799 int64_t vlan = cfg->select_vlan[i];
4800 if (vlan < 0 || vlan > 4095) {
4801 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4802 m->bridge->name, m->name, vlan);
4804 (*vlans)[n_vlans++] = vlan;
4811 vlan_is_mirrored(const struct mirror *m, int vlan)
4815 for (i = 0; i < m->n_vlans; i++) {
4816 if (m->vlans[i] == vlan) {
4824 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4828 for (i = 0; i < m->n_vlans; i++) {
4829 if (port_trunks_vlan(p, m->vlans[i])) {
4837 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4839 struct shash src_ports, dst_ports;
4840 mirror_mask_t mirror_bit;
4841 struct port *out_port;
4848 if (strcmp(cfg->name, m->name)) {
4850 m->name = xstrdup(cfg->name);
4853 /* Get output port. */
4854 if (cfg->output_port) {
4855 out_port = port_lookup(m->bridge, cfg->output_port->name);
4857 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4858 m->bridge->name, m->name);
4864 if (cfg->output_vlan) {
4865 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4866 "output vlan; ignoring output vlan",
4867 m->bridge->name, m->name);
4869 } else if (cfg->output_vlan) {
4871 out_vlan = *cfg->output_vlan;
4873 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4874 m->bridge->name, m->name);
4879 shash_init(&src_ports);
4880 shash_init(&dst_ports);
4881 if (cfg->select_all) {
4882 for (i = 0; i < m->bridge->n_ports; i++) {
4883 const char *name = m->bridge->ports[i]->name;
4884 shash_add_once(&src_ports, name, NULL);
4885 shash_add_once(&dst_ports, name, NULL);
4890 /* Get ports, and drop duplicates and ports that don't exist. */
4891 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4893 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4896 /* Get all the vlans, and drop duplicate and invalid vlans. */
4897 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4900 /* Update mirror data. */
4901 if (!shash_equal_keys(&m->src_ports, &src_ports)
4902 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4903 || m->n_vlans != n_vlans
4904 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4905 || m->out_port != out_port
4906 || m->out_vlan != out_vlan) {
4907 bridge_flush(m->bridge);
4909 shash_swap(&m->src_ports, &src_ports);
4910 shash_swap(&m->dst_ports, &dst_ports);
4913 m->n_vlans = n_vlans;
4914 m->out_port = out_port;
4915 m->out_vlan = out_vlan;
4918 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4919 for (i = 0; i < m->bridge->n_ports; i++) {
4920 struct port *port = m->bridge->ports[i];
4922 if (shash_find(&m->src_ports, port->name)
4925 ? port_trunks_any_mirrored_vlan(m, port)
4926 : vlan_is_mirrored(m, port->vlan)))) {
4927 port->src_mirrors |= mirror_bit;
4929 port->src_mirrors &= ~mirror_bit;
4932 if (shash_find(&m->dst_ports, port->name)) {
4933 port->dst_mirrors |= mirror_bit;
4935 port->dst_mirrors &= ~mirror_bit;
4940 shash_destroy(&src_ports);
4941 shash_destroy(&dst_ports);