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 active_iface_tag; /* Tag for bcast flows. */
179 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
180 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
181 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
182 long long int bond_next_fake_iface_update; /* Time of next update. */
184 /* LACP information. */
185 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
186 bool lacp_active; /* True if LACP is active */
187 bool lacp_fast; /* True if LACP is in fast mode. */
188 uint16_t lacp_priority; /* LACP system priority. */
190 /* SLB specific bonding info. */
191 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
192 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
193 long long int bond_next_rebalance; /* Next rebalancing time. */
195 /* Port mirroring info. */
196 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
197 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
198 bool is_mirror_output_port; /* Does port mirroring send frames here? */
202 struct list node; /* Node in global list of bridges. */
203 char *name; /* User-specified arbitrary name. */
204 struct mac_learning *ml; /* MAC learning table. */
205 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
206 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
207 const struct ovsrec_bridge *cfg;
209 /* OpenFlow switch processing. */
210 struct ofproto *ofproto; /* OpenFlow switch. */
212 /* Kernel datapath information. */
213 struct dpif *dpif; /* Datapath. */
214 struct hmap ifaces; /* Contains "struct iface"s. */
218 size_t n_ports, allocated_ports;
219 struct shash iface_by_name; /* "struct iface"s indexed by name. */
220 struct shash port_by_name; /* "struct port"s indexed by name. */
223 bool has_bonded_ports;
228 /* Port mirroring. */
229 struct mirror *mirrors[MAX_MIRRORS];
232 /* List of all bridges. */
233 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
235 /* OVSDB IDL used to obtain configuration. */
236 static struct ovsdb_idl *idl;
238 /* Each time this timer expires, the bridge fetches systems and interface
239 * statistics and pushes them into the database. */
240 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
241 static long long int stats_timer = LLONG_MIN;
243 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
244 static void bridge_destroy(struct bridge *);
245 static struct bridge *bridge_lookup(const char *name);
246 static unixctl_cb_func bridge_unixctl_dump_flows;
247 static unixctl_cb_func bridge_unixctl_reconnect;
248 static int bridge_run_one(struct bridge *);
249 static size_t bridge_get_controllers(const struct bridge *br,
250 struct ovsrec_controller ***controllersp);
251 static void bridge_reconfigure_one(struct bridge *);
252 static void bridge_reconfigure_remotes(struct bridge *,
253 const struct sockaddr_in *managers,
255 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
256 static void bridge_fetch_dp_ifaces(struct bridge *);
257 static void bridge_flush(struct bridge *);
258 static void bridge_pick_local_hw_addr(struct bridge *,
259 uint8_t ea[ETH_ADDR_LEN],
260 struct iface **hw_addr_iface);
261 static uint64_t bridge_pick_datapath_id(struct bridge *,
262 const uint8_t bridge_ea[ETH_ADDR_LEN],
263 struct iface *hw_addr_iface);
264 static struct iface *bridge_get_local_iface(struct bridge *);
265 static uint64_t dpid_from_hash(const void *, size_t nbytes);
267 static unixctl_cb_func bridge_unixctl_fdb_show;
268 static unixctl_cb_func qos_unixctl_show;
270 static void bond_init(void);
271 static void bond_run(struct port *);
272 static void bond_wait(struct port *);
273 static void bond_rebalance_port(struct port *);
274 static void bond_send_learning_packets(struct port *);
275 static void bond_enable_slave(struct iface *iface, bool enable);
277 static void port_run(struct port *);
278 static void port_wait(struct port *);
279 static struct port *port_create(struct bridge *, const char *name);
280 static void port_reconfigure(struct port *, const struct ovsrec_port *);
281 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
282 static void port_destroy(struct port *);
283 static struct port *port_lookup(const struct bridge *, const char *name);
284 static struct iface *port_lookup_iface(const struct port *, const char *name);
285 static struct port *port_from_dp_ifidx(const struct bridge *,
287 static void port_update_bonding(struct port *);
288 static void port_update_lacp(struct port *);
290 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
291 static void mirror_destroy(struct mirror *);
292 static void mirror_reconfigure(struct bridge *);
293 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
294 static bool vlan_is_mirrored(const struct mirror *, int vlan);
296 static struct iface *iface_create(struct port *port,
297 const struct ovsrec_interface *if_cfg);
298 static void iface_destroy(struct iface *);
299 static struct iface *iface_lookup(const struct bridge *, const char *name);
300 static struct iface *iface_find(const char *name);
301 static struct iface *iface_from_dp_ifidx(const struct bridge *,
303 static void iface_set_mac(struct iface *);
304 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
305 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
306 static void iface_update_cfm(struct iface *);
307 static void iface_refresh_cfm_stats(struct iface *iface);
308 static void iface_update_carrier(struct iface *);
309 static bool iface_get_carrier(const struct iface *);
311 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
313 static void shash_to_ovs_idl_map(struct shash *,
314 char ***keys, char ***values, size_t *n);
316 /* Hooks into ofproto processing. */
317 static struct ofhooks bridge_ofhooks;
319 /* Public functions. */
321 /* Initializes the bridge module, configuring it to obtain its configuration
322 * from an OVSDB server accessed over 'remote', which should be a string in a
323 * form acceptable to ovsdb_idl_create(). */
325 bridge_init(const char *remote)
327 /* Create connection to database. */
328 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
330 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
331 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
332 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
334 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
336 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
337 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
339 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
340 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
341 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
343 /* Register unixctl commands. */
344 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
345 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
346 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
348 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
357 struct bridge *br, *next_br;
359 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
362 ovsdb_idl_destroy(idl);
365 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
366 * but for which the ovs-vswitchd configuration 'cfg' is required. */
368 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
370 static bool already_configured_once;
371 struct svec bridge_names;
372 struct svec dpif_names, dpif_types;
375 /* Only do this once per ovs-vswitchd run. */
376 if (already_configured_once) {
379 already_configured_once = true;
381 stats_timer = time_msec() + STATS_INTERVAL;
383 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
384 svec_init(&bridge_names);
385 for (i = 0; i < cfg->n_bridges; i++) {
386 svec_add(&bridge_names, cfg->bridges[i]->name);
388 svec_sort(&bridge_names);
390 /* Iterate over all system dpifs and delete any of them that do not appear
392 svec_init(&dpif_names);
393 svec_init(&dpif_types);
394 dp_enumerate_types(&dpif_types);
395 for (i = 0; i < dpif_types.n; i++) {
398 dp_enumerate_names(dpif_types.names[i], &dpif_names);
400 /* Delete each dpif whose name is not in 'bridge_names'. */
401 for (j = 0; j < dpif_names.n; j++) {
402 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
406 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
415 svec_destroy(&bridge_names);
416 svec_destroy(&dpif_names);
417 svec_destroy(&dpif_types);
420 /* Callback for iterate_and_prune_ifaces(). */
422 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
424 if (!iface->netdev) {
425 /* We already reported a related error, don't bother duplicating it. */
429 if (iface->dp_ifidx < 0) {
430 VLOG_ERR("%s interface not in %s, dropping",
431 iface->name, dpif_name(br->dpif));
435 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
436 iface->name, iface->dp_ifidx);
440 /* Callback for iterate_and_prune_ifaces(). */
442 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
443 void *aux OVS_UNUSED)
445 /* Set policing attributes. */
446 netdev_set_policing(iface->netdev,
447 iface->cfg->ingress_policing_rate,
448 iface->cfg->ingress_policing_burst);
450 /* Set MAC address of internal interfaces other than the local
452 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
453 iface_set_mac(iface);
459 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
460 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
461 * deletes from 'br' any ports that no longer have any interfaces. */
463 iterate_and_prune_ifaces(struct bridge *br,
464 bool (*cb)(struct bridge *, struct iface *,
470 for (i = 0; i < br->n_ports; ) {
471 struct port *port = br->ports[i];
472 for (j = 0; j < port->n_ifaces; ) {
473 struct iface *iface = port->ifaces[j];
474 if (cb(br, iface, aux)) {
477 iface_set_ofport(iface->cfg, -1);
478 iface_destroy(iface);
482 if (port->n_ifaces) {
485 VLOG_WARN("%s port has no interfaces, dropping", port->name);
491 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
492 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
493 * responsible for freeing '*managersp' (with free()).
495 * You may be asking yourself "why does ovs-vswitchd care?", because
496 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
497 * should not be and in fact is not directly involved in that. But
498 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
499 * it has to tell in-band control where the managers are to enable that.
500 * (Thus, only managers connected in-band are collected.)
503 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
504 struct sockaddr_in **managersp, size_t *n_managersp)
506 struct sockaddr_in *managers = NULL;
507 size_t n_managers = 0;
508 struct shash targets;
511 /* Collect all of the potential targets from the "targets" columns of the
512 * rows pointed to by "manager_options", excluding any that are
514 shash_init(&targets);
515 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
516 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
518 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
519 shash_find_and_delete(&targets, m->target);
521 shash_add_once(&targets, m->target, NULL);
525 /* Now extract the targets' IP addresses. */
526 if (!shash_is_empty(&targets)) {
527 struct shash_node *node;
529 managers = xmalloc(shash_count(&targets) * sizeof *managers);
530 SHASH_FOR_EACH (node, &targets) {
531 const char *target = node->name;
532 struct sockaddr_in *sin = &managers[n_managers];
534 if ((!strncmp(target, "tcp:", 4)
535 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
536 (!strncmp(target, "ssl:", 4)
537 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
542 shash_destroy(&targets);
544 *managersp = managers;
545 *n_managersp = n_managers;
549 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
551 struct shash old_br, new_br;
552 struct shash_node *node;
553 struct bridge *br, *next;
554 struct sockaddr_in *managers;
557 int sflow_bridge_number;
559 COVERAGE_INC(bridge_reconfigure);
561 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
563 /* Collect old and new bridges. */
566 LIST_FOR_EACH (br, node, &all_bridges) {
567 shash_add(&old_br, br->name, br);
569 for (i = 0; i < ovs_cfg->n_bridges; i++) {
570 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
571 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
572 VLOG_WARN("more than one bridge named %s", br_cfg->name);
576 /* Get rid of deleted bridges and add new bridges. */
577 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
578 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
585 SHASH_FOR_EACH (node, &new_br) {
586 const char *br_name = node->name;
587 const struct ovsrec_bridge *br_cfg = node->data;
588 br = shash_find_data(&old_br, br_name);
590 /* If the bridge datapath type has changed, we need to tear it
591 * down and recreate. */
592 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
594 bridge_create(br_cfg);
597 bridge_create(br_cfg);
600 shash_destroy(&old_br);
601 shash_destroy(&new_br);
603 /* Reconfigure all bridges. */
604 LIST_FOR_EACH (br, node, &all_bridges) {
605 bridge_reconfigure_one(br);
608 /* Add and delete ports on all datapaths.
610 * The kernel will reject any attempt to add a given port to a datapath if
611 * that port already belongs to a different datapath, so we must do all
612 * port deletions before any port additions. */
613 LIST_FOR_EACH (br, node, &all_bridges) {
614 struct dpif_port_dump dump;
615 struct shash want_ifaces;
616 struct dpif_port dpif_port;
618 bridge_get_all_ifaces(br, &want_ifaces);
619 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
620 if (!shash_find(&want_ifaces, dpif_port.name)
621 && strcmp(dpif_port.name, br->name)) {
622 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
624 VLOG_WARN("failed to remove %s interface from %s: %s",
625 dpif_port.name, dpif_name(br->dpif),
630 shash_destroy(&want_ifaces);
632 LIST_FOR_EACH (br, node, &all_bridges) {
633 struct shash cur_ifaces, want_ifaces;
634 struct dpif_port_dump dump;
635 struct dpif_port dpif_port;
637 /* Get the set of interfaces currently in this datapath. */
638 shash_init(&cur_ifaces);
639 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
640 struct dpif_port *port_info = xmalloc(sizeof *port_info);
641 dpif_port_clone(port_info, &dpif_port);
642 shash_add(&cur_ifaces, dpif_port.name, port_info);
645 /* Get the set of interfaces we want on this datapath. */
646 bridge_get_all_ifaces(br, &want_ifaces);
648 hmap_clear(&br->ifaces);
649 SHASH_FOR_EACH (node, &want_ifaces) {
650 const char *if_name = node->name;
651 struct iface *iface = node->data;
652 struct dpif_port *dpif_port;
656 type = iface ? iface->type : "internal";
657 dpif_port = shash_find_data(&cur_ifaces, if_name);
659 /* If we have a port or a netdev already, and it's not the type we
660 * want, then delete the port (if any) and close the netdev (if
662 if ((dpif_port && strcmp(dpif_port->type, type))
663 || (iface && iface->netdev
664 && strcmp(type, netdev_get_type(iface->netdev)))) {
666 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
673 netdev_close(iface->netdev);
674 iface->netdev = NULL;
678 /* If the port doesn't exist or we don't have the netdev open,
679 * we need to do more work. */
680 if (!dpif_port || (iface && !iface->netdev)) {
681 struct netdev_options options;
682 struct netdev *netdev;
685 /* First open the network device. */
686 options.name = if_name;
688 options.args = &args;
689 options.ethertype = NETDEV_ETH_TYPE_NONE;
693 shash_from_ovs_idl_map(iface->cfg->key_options,
694 iface->cfg->value_options,
695 iface->cfg->n_options, &args);
697 error = netdev_open(&options, &netdev);
698 shash_destroy(&args);
701 VLOG_WARN("could not open network device %s (%s)",
702 if_name, strerror(error));
706 /* Then add the port if we haven't already. */
708 error = dpif_port_add(br->dpif, netdev, NULL);
710 netdev_close(netdev);
711 if (error == EFBIG) {
712 VLOG_ERR("ran out of valid port numbers on %s",
713 dpif_name(br->dpif));
716 VLOG_WARN("failed to add %s interface to %s: %s",
717 if_name, dpif_name(br->dpif),
724 /* Update 'iface'. */
726 iface->netdev = netdev;
727 iface->enabled = iface_get_carrier(iface);
728 iface->up = iface->enabled;
730 } else if (iface && iface->netdev) {
734 shash_from_ovs_idl_map(iface->cfg->key_options,
735 iface->cfg->value_options,
736 iface->cfg->n_options, &args);
737 netdev_set_config(iface->netdev, &args);
738 shash_destroy(&args);
741 shash_destroy(&want_ifaces);
743 SHASH_FOR_EACH (node, &cur_ifaces) {
744 struct dpif_port *port_info = node->data;
745 dpif_port_destroy(port_info);
748 shash_destroy(&cur_ifaces);
750 sflow_bridge_number = 0;
751 LIST_FOR_EACH (br, node, &all_bridges) {
754 struct iface *local_iface;
755 struct iface *hw_addr_iface;
758 bridge_fetch_dp_ifaces(br);
760 iterate_and_prune_ifaces(br, check_iface, NULL);
762 /* Pick local port hardware address, datapath ID. */
763 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
764 local_iface = bridge_get_local_iface(br);
766 int error = netdev_set_etheraddr(local_iface->netdev, ea);
768 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
769 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
770 "Ethernet address: %s",
771 br->name, strerror(error));
774 memcpy(br->ea, ea, ETH_ADDR_LEN);
776 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
777 ofproto_set_datapath_id(br->ofproto, dpid);
779 dpid_string = xasprintf("%016"PRIx64, dpid);
780 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
783 /* Set NetFlow configuration on this bridge. */
784 if (br->cfg->netflow) {
785 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
786 struct netflow_options opts;
788 memset(&opts, 0, sizeof opts);
790 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
791 if (nf_cfg->engine_type) {
792 opts.engine_type = *nf_cfg->engine_type;
794 if (nf_cfg->engine_id) {
795 opts.engine_id = *nf_cfg->engine_id;
798 opts.active_timeout = nf_cfg->active_timeout;
799 if (!opts.active_timeout) {
800 opts.active_timeout = -1;
801 } else if (opts.active_timeout < 0) {
802 VLOG_WARN("bridge %s: active timeout interval set to negative "
803 "value, using default instead (%d seconds)", br->name,
804 NF_ACTIVE_TIMEOUT_DEFAULT);
805 opts.active_timeout = -1;
808 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
809 if (opts.add_id_to_iface) {
810 if (opts.engine_id > 0x7f) {
811 VLOG_WARN("bridge %s: netflow port mangling may conflict "
812 "with another vswitch, choose an engine id less "
813 "than 128", br->name);
815 if (br->n_ports > 508) {
816 VLOG_WARN("bridge %s: netflow port mangling will conflict "
817 "with another port when more than 508 ports are "
822 opts.collectors.n = nf_cfg->n_targets;
823 opts.collectors.names = nf_cfg->targets;
824 if (ofproto_set_netflow(br->ofproto, &opts)) {
825 VLOG_ERR("bridge %s: problem setting netflow collectors",
829 ofproto_set_netflow(br->ofproto, NULL);
832 /* Set sFlow configuration on this bridge. */
833 if (br->cfg->sflow) {
834 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
835 struct ovsrec_controller **controllers;
836 struct ofproto_sflow_options oso;
837 size_t n_controllers;
839 memset(&oso, 0, sizeof oso);
841 oso.targets.n = sflow_cfg->n_targets;
842 oso.targets.names = sflow_cfg->targets;
844 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
845 if (sflow_cfg->sampling) {
846 oso.sampling_rate = *sflow_cfg->sampling;
849 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
850 if (sflow_cfg->polling) {
851 oso.polling_interval = *sflow_cfg->polling;
854 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
855 if (sflow_cfg->header) {
856 oso.header_len = *sflow_cfg->header;
859 oso.sub_id = sflow_bridge_number++;
860 oso.agent_device = sflow_cfg->agent;
862 oso.control_ip = NULL;
863 n_controllers = bridge_get_controllers(br, &controllers);
864 for (i = 0; i < n_controllers; i++) {
865 if (controllers[i]->local_ip) {
866 oso.control_ip = controllers[i]->local_ip;
870 ofproto_set_sflow(br->ofproto, &oso);
872 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
874 ofproto_set_sflow(br->ofproto, NULL);
877 /* Update the controller and related settings. It would be more
878 * straightforward to call this from bridge_reconfigure_one(), but we
879 * can't do it there for two reasons. First, and most importantly, at
880 * that point we don't know the dp_ifidx of any interfaces that have
881 * been added to the bridge (because we haven't actually added them to
882 * the datapath). Second, at that point we haven't set the datapath ID
883 * yet; when a controller is configured, resetting the datapath ID will
884 * immediately disconnect from the controller, so it's better to set
885 * the datapath ID before the controller. */
886 bridge_reconfigure_remotes(br, managers, n_managers);
888 LIST_FOR_EACH (br, node, &all_bridges) {
889 for (i = 0; i < br->n_ports; i++) {
890 struct port *port = br->ports[i];
894 for (j = 0; j < port->n_ifaces; j++) {
895 netdev_monitor_add(port->monitor, port->ifaces[j]->netdev);
898 port->miimon_next_update = 0;
901 port_update_lacp(port);
902 port_update_bonding(port);
904 for (j = 0; j < port->n_ifaces; j++) {
905 iface_update_qos(port->ifaces[j], port->cfg->qos);
909 LIST_FOR_EACH (br, node, &all_bridges) {
910 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
913 LIST_FOR_EACH (br, node, &all_bridges) {
915 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
916 iface_update_cfm(iface);
922 /* ovs-vswitchd has completed initialization, so allow the process that
923 * forked us to exit successfully. */
924 daemonize_complete();
928 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
929 const struct ovsdb_idl_column *column,
932 const struct ovsdb_datum *datum;
933 union ovsdb_atom atom;
936 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
937 atom.string = (char *) key;
938 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
939 return idx == UINT_MAX ? NULL : datum->values[idx].string;
943 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
945 return get_ovsrec_key_value(&br_cfg->header_,
946 &ovsrec_bridge_col_other_config, key);
950 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
951 struct iface **hw_addr_iface)
957 *hw_addr_iface = NULL;
959 /* Did the user request a particular MAC? */
960 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
961 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
962 if (eth_addr_is_multicast(ea)) {
963 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
964 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
965 } else if (eth_addr_is_zero(ea)) {
966 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
972 /* Otherwise choose the minimum non-local MAC address among all of the
974 memset(ea, 0xff, ETH_ADDR_LEN);
975 for (i = 0; i < br->n_ports; i++) {
976 struct port *port = br->ports[i];
977 uint8_t iface_ea[ETH_ADDR_LEN];
980 /* Mirror output ports don't participate. */
981 if (port->is_mirror_output_port) {
985 /* Choose the MAC address to represent the port. */
986 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
987 /* Find the interface with this Ethernet address (if any) so that
988 * we can provide the correct devname to the caller. */
990 for (j = 0; j < port->n_ifaces; j++) {
991 struct iface *candidate = port->ifaces[j];
992 uint8_t candidate_ea[ETH_ADDR_LEN];
993 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
994 && eth_addr_equals(iface_ea, candidate_ea)) {
999 /* Choose the interface whose MAC address will represent the port.
1000 * The Linux kernel bonding code always chooses the MAC address of
1001 * the first slave added to a bond, and the Fedora networking
1002 * scripts always add slaves to a bond in alphabetical order, so
1003 * for compatibility we choose the interface with the name that is
1004 * first in alphabetical order. */
1005 iface = port->ifaces[0];
1006 for (j = 1; j < port->n_ifaces; j++) {
1007 struct iface *candidate = port->ifaces[j];
1008 if (strcmp(candidate->name, iface->name) < 0) {
1013 /* The local port doesn't count (since we're trying to choose its
1014 * MAC address anyway). */
1015 if (iface->dp_ifidx == ODPP_LOCAL) {
1020 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1022 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1023 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1024 iface->name, strerror(error));
1029 /* Compare against our current choice. */
1030 if (!eth_addr_is_multicast(iface_ea) &&
1031 !eth_addr_is_local(iface_ea) &&
1032 !eth_addr_is_reserved(iface_ea) &&
1033 !eth_addr_is_zero(iface_ea) &&
1034 eth_addr_compare_3way(iface_ea, ea) < 0)
1036 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1037 *hw_addr_iface = iface;
1040 if (eth_addr_is_multicast(ea)) {
1041 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1042 *hw_addr_iface = NULL;
1043 VLOG_WARN("bridge %s: using default bridge Ethernet "
1044 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1046 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1047 br->name, ETH_ADDR_ARGS(ea));
1051 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1052 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1053 * an interface on 'br', then that interface must be passed in as
1054 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1055 * 'hw_addr_iface' must be passed in as a null pointer. */
1057 bridge_pick_datapath_id(struct bridge *br,
1058 const uint8_t bridge_ea[ETH_ADDR_LEN],
1059 struct iface *hw_addr_iface)
1062 * The procedure for choosing a bridge MAC address will, in the most
1063 * ordinary case, also choose a unique MAC that we can use as a datapath
1064 * ID. In some special cases, though, multiple bridges will end up with
1065 * the same MAC address. This is OK for the bridges, but it will confuse
1066 * the OpenFlow controller, because each datapath needs a unique datapath
1069 * Datapath IDs must be unique. It is also very desirable that they be
1070 * stable from one run to the next, so that policy set on a datapath
1073 const char *datapath_id;
1076 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1077 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1081 if (hw_addr_iface) {
1083 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1085 * A bridge whose MAC address is taken from a VLAN network device
1086 * (that is, a network device created with vconfig(8) or similar
1087 * tool) will have the same MAC address as a bridge on the VLAN
1088 * device's physical network device.
1090 * Handle this case by hashing the physical network device MAC
1091 * along with the VLAN identifier.
1093 uint8_t buf[ETH_ADDR_LEN + 2];
1094 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1095 buf[ETH_ADDR_LEN] = vlan >> 8;
1096 buf[ETH_ADDR_LEN + 1] = vlan;
1097 return dpid_from_hash(buf, sizeof buf);
1100 * Assume that this bridge's MAC address is unique, since it
1101 * doesn't fit any of the cases we handle specially.
1106 * A purely internal bridge, that is, one that has no non-virtual
1107 * network devices on it at all, is more difficult because it has no
1108 * natural unique identifier at all.
1110 * When the host is a XenServer, we handle this case by hashing the
1111 * host's UUID with the name of the bridge. Names of bridges are
1112 * persistent across XenServer reboots, although they can be reused if
1113 * an internal network is destroyed and then a new one is later
1114 * created, so this is fairly effective.
1116 * When the host is not a XenServer, we punt by using a random MAC
1117 * address on each run.
1119 const char *host_uuid = xenserver_get_host_uuid();
1121 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1122 dpid = dpid_from_hash(combined, strlen(combined));
1128 return eth_addr_to_uint64(bridge_ea);
1132 dpid_from_hash(const void *data, size_t n)
1134 uint8_t hash[SHA1_DIGEST_SIZE];
1136 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1137 sha1_bytes(data, n, hash);
1138 eth_addr_mark_random(hash);
1139 return eth_addr_to_uint64(hash);
1143 iface_refresh_status(struct iface *iface)
1147 enum netdev_flags flags;
1156 if (!netdev_get_status(iface->netdev, &sh)) {
1158 char **keys, **values;
1160 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1161 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1166 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1169 shash_destroy_free_data(&sh);
1171 error = netdev_get_flags(iface->netdev, &flags);
1173 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1176 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1179 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1181 ovsrec_interface_set_duplex(iface->cfg,
1182 netdev_features_is_full_duplex(current)
1184 /* warning: uint64_t -> int64_t conversion */
1185 bps = netdev_features_to_bps(current);
1186 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1189 ovsrec_interface_set_duplex(iface->cfg, NULL);
1190 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1194 ovsrec_interface_set_link_state(iface->cfg,
1195 iface_get_carrier(iface) ? "up" : "down");
1197 error = netdev_get_mtu(iface->netdev, &mtu);
1198 if (!error && mtu != INT_MAX) {
1200 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1203 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1208 iface_refresh_cfm_stats(struct iface *iface)
1210 const struct ovsrec_monitor *mon;
1211 const struct cfm *cfm;
1214 mon = iface->cfg->monitor;
1215 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1221 for (i = 0; i < mon->n_remote_mps; i++) {
1222 const struct ovsrec_maintenance_point *mp;
1223 const struct remote_mp *rmp;
1225 mp = mon->remote_mps[i];
1226 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1228 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1231 if (hmap_is_empty(&cfm->x_remote_mps)) {
1232 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1235 struct remote_mp *rmp;
1236 int64_t *x_remote_mps;
1238 length = hmap_count(&cfm->x_remote_mps);
1239 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1242 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1243 x_remote_mps[i++] = rmp->mpid;
1246 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1250 if (hmap_is_empty(&cfm->x_remote_maids)) {
1251 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1254 char **x_remote_maids;
1255 struct remote_maid *rmaid;
1257 length = hmap_count(&cfm->x_remote_maids);
1258 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1261 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1264 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1266 for (j = 0; j < CCM_MAID_LEN; j++) {
1267 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1272 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1274 for (i = 0; i < length; i++) {
1275 free(x_remote_maids[i]);
1277 free(x_remote_maids);
1280 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1284 iface_refresh_stats(struct iface *iface)
1290 static const struct iface_stat iface_stats[] = {
1291 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1292 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1293 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1294 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1295 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1296 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1297 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1298 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1299 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1300 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1301 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1302 { "collisions", offsetof(struct netdev_stats, collisions) },
1304 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1305 const struct iface_stat *s;
1307 char *keys[N_STATS];
1308 int64_t values[N_STATS];
1311 struct netdev_stats stats;
1313 /* Intentionally ignore return value, since errors will set 'stats' to
1314 * all-1s, and we will deal with that correctly below. */
1315 netdev_get_stats(iface->netdev, &stats);
1318 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1319 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1320 if (value != UINT64_MAX) {
1327 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1331 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1333 struct ovsdb_datum datum;
1337 get_system_stats(&stats);
1339 ovsdb_datum_from_shash(&datum, &stats);
1340 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1344 static inline const char *
1345 nx_role_to_str(enum nx_role role)
1350 case NX_ROLE_MASTER:
1355 return "*** INVALID ROLE ***";
1360 bridge_refresh_controller_status(const struct bridge *br)
1363 const struct ovsrec_controller *cfg;
1365 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1367 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1368 struct ofproto_controller_info *cinfo =
1369 shash_find_data(&info, cfg->target);
1372 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1373 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1374 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1375 (char **) cinfo->pairs.values,
1378 ovsrec_controller_set_is_connected(cfg, false);
1379 ovsrec_controller_set_role(cfg, NULL);
1380 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1384 ofproto_free_ofproto_controller_info(&info);
1390 const struct ovsrec_open_vswitch *cfg;
1392 bool datapath_destroyed;
1393 bool database_changed;
1396 /* Let each bridge do the work that it needs to do. */
1397 datapath_destroyed = false;
1398 LIST_FOR_EACH (br, node, &all_bridges) {
1399 int error = bridge_run_one(br);
1401 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1402 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1403 "forcing reconfiguration", br->name);
1404 datapath_destroyed = true;
1408 /* (Re)configure if necessary. */
1409 database_changed = ovsdb_idl_run(idl);
1410 cfg = ovsrec_open_vswitch_first(idl);
1412 /* Re-configure SSL. We do this on every trip through the main loop,
1413 * instead of just when the database changes, because the contents of the
1414 * key and certificate files can change without the database changing.
1416 * We do this before bridge_reconfigure() because that function might
1417 * initiate SSL connections and thus requires SSL to be configured. */
1418 if (cfg && cfg->ssl) {
1419 const struct ovsrec_ssl *ssl = cfg->ssl;
1421 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1422 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1425 if (database_changed || datapath_destroyed) {
1427 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1429 bridge_configure_once(cfg);
1430 bridge_reconfigure(cfg);
1432 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1433 ovsdb_idl_txn_commit(txn);
1434 ovsdb_idl_txn_destroy(txn); /* XXX */
1436 /* We still need to reconfigure to avoid dangling pointers to
1437 * now-destroyed ovsrec structures inside bridge data. */
1438 static const struct ovsrec_open_vswitch null_cfg;
1440 bridge_reconfigure(&null_cfg);
1444 /* Refresh system and interface stats if necessary. */
1445 if (time_msec() >= stats_timer) {
1447 struct ovsdb_idl_txn *txn;
1449 txn = ovsdb_idl_txn_create(idl);
1450 LIST_FOR_EACH (br, node, &all_bridges) {
1453 for (i = 0; i < br->n_ports; i++) {
1454 struct port *port = br->ports[i];
1457 for (j = 0; j < port->n_ifaces; j++) {
1458 struct iface *iface = port->ifaces[j];
1459 iface_refresh_stats(iface);
1460 iface_refresh_cfm_stats(iface);
1461 iface_refresh_status(iface);
1464 bridge_refresh_controller_status(br);
1466 refresh_system_stats(cfg);
1467 ovsdb_idl_txn_commit(txn);
1468 ovsdb_idl_txn_destroy(txn); /* XXX */
1471 stats_timer = time_msec() + STATS_INTERVAL;
1480 LIST_FOR_EACH (br, node, &all_bridges) {
1483 ofproto_wait(br->ofproto);
1484 if (ofproto_has_primary_controller(br->ofproto)) {
1488 mac_learning_wait(br->ml);
1490 for (i = 0; i < br->n_ports; i++) {
1491 port_wait(br->ports[i]);
1494 ovsdb_idl_wait(idl);
1495 poll_timer_wait_until(stats_timer);
1498 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1499 * configuration changes. */
1501 bridge_flush(struct bridge *br)
1503 COVERAGE_INC(bridge_flush);
1505 mac_learning_flush(br->ml);
1508 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1509 * such interface. */
1510 static struct iface *
1511 bridge_get_local_iface(struct bridge *br)
1515 for (i = 0; i < br->n_ports; i++) {
1516 struct port *port = br->ports[i];
1517 for (j = 0; j < port->n_ifaces; j++) {
1518 struct iface *iface = port->ifaces[j];
1519 if (iface->dp_ifidx == ODPP_LOCAL) {
1528 /* Bridge unixctl user interface functions. */
1530 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1531 const char *args, void *aux OVS_UNUSED)
1533 struct ds ds = DS_EMPTY_INITIALIZER;
1534 const struct bridge *br;
1535 const struct mac_entry *e;
1537 br = bridge_lookup(args);
1539 unixctl_command_reply(conn, 501, "no such bridge");
1543 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1544 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1545 if (e->port < 0 || e->port >= br->n_ports) {
1548 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1549 br->ports[e->port]->ifaces[0]->dp_ifidx,
1550 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1552 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1556 /* QoS unixctl user interface functions. */
1558 struct qos_unixctl_show_cbdata {
1560 struct iface *iface;
1564 qos_unixctl_show_cb(unsigned int queue_id,
1565 const struct shash *details,
1568 struct qos_unixctl_show_cbdata *data = aux;
1569 struct ds *ds = data->ds;
1570 struct iface *iface = data->iface;
1571 struct netdev_queue_stats stats;
1572 struct shash_node *node;
1575 ds_put_cstr(ds, "\n");
1577 ds_put_format(ds, "Queue %u:\n", queue_id);
1579 ds_put_cstr(ds, "Default:\n");
1582 SHASH_FOR_EACH (node, details) {
1583 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1586 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1588 if (stats.tx_packets != UINT64_MAX) {
1589 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1592 if (stats.tx_bytes != UINT64_MAX) {
1593 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1596 if (stats.tx_errors != UINT64_MAX) {
1597 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1600 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1601 queue_id, strerror(error));
1606 qos_unixctl_show(struct unixctl_conn *conn,
1607 const char *args, void *aux OVS_UNUSED)
1609 struct ds ds = DS_EMPTY_INITIALIZER;
1610 struct shash sh = SHASH_INITIALIZER(&sh);
1611 struct iface *iface;
1613 struct shash_node *node;
1614 struct qos_unixctl_show_cbdata data;
1617 iface = iface_find(args);
1619 unixctl_command_reply(conn, 501, "no such interface");
1623 netdev_get_qos(iface->netdev, &type, &sh);
1625 if (*type != '\0') {
1626 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1628 SHASH_FOR_EACH (node, &sh) {
1629 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1634 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1637 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1639 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1641 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1642 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1645 shash_destroy_free_data(&sh);
1649 /* Bridge reconfiguration functions. */
1650 static struct bridge *
1651 bridge_create(const struct ovsrec_bridge *br_cfg)
1656 assert(!bridge_lookup(br_cfg->name));
1657 br = xzalloc(sizeof *br);
1659 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1665 dpif_flow_flush(br->dpif);
1667 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1670 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1672 dpif_delete(br->dpif);
1673 dpif_close(br->dpif);
1678 br->name = xstrdup(br_cfg->name);
1680 br->ml = mac_learning_create();
1681 eth_addr_nicira_random(br->default_ea);
1683 hmap_init(&br->ifaces);
1685 shash_init(&br->port_by_name);
1686 shash_init(&br->iface_by_name);
1690 list_push_back(&all_bridges, &br->node);
1692 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1698 bridge_destroy(struct bridge *br)
1703 while (br->n_ports > 0) {
1704 port_destroy(br->ports[br->n_ports - 1]);
1706 list_remove(&br->node);
1707 error = dpif_delete(br->dpif);
1708 if (error && error != ENOENT) {
1709 VLOG_ERR("failed to delete %s: %s",
1710 dpif_name(br->dpif), strerror(error));
1712 dpif_close(br->dpif);
1713 ofproto_destroy(br->ofproto);
1714 mac_learning_destroy(br->ml);
1715 hmap_destroy(&br->ifaces);
1716 shash_destroy(&br->port_by_name);
1717 shash_destroy(&br->iface_by_name);
1724 static struct bridge *
1725 bridge_lookup(const char *name)
1729 LIST_FOR_EACH (br, node, &all_bridges) {
1730 if (!strcmp(br->name, name)) {
1737 /* Handle requests for a listing of all flows known by the OpenFlow
1738 * stack, including those normally hidden. */
1740 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1741 const char *args, void *aux OVS_UNUSED)
1746 br = bridge_lookup(args);
1748 unixctl_command_reply(conn, 501, "Unknown bridge");
1753 ofproto_get_all_flows(br->ofproto, &results);
1755 unixctl_command_reply(conn, 200, ds_cstr(&results));
1756 ds_destroy(&results);
1759 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1760 * connections and reconnect. If BRIDGE is not specified, then all bridges
1761 * drop their controller connections and reconnect. */
1763 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1764 const char *args, void *aux OVS_UNUSED)
1767 if (args[0] != '\0') {
1768 br = bridge_lookup(args);
1770 unixctl_command_reply(conn, 501, "Unknown bridge");
1773 ofproto_reconnect_controllers(br->ofproto);
1775 LIST_FOR_EACH (br, node, &all_bridges) {
1776 ofproto_reconnect_controllers(br->ofproto);
1779 unixctl_command_reply(conn, 200, NULL);
1783 bridge_run_one(struct bridge *br)
1788 error = ofproto_run1(br->ofproto);
1793 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1795 for (i = 0; i < br->n_ports; i++) {
1796 port_run(br->ports[i]);
1799 error = ofproto_run2(br->ofproto, br->flush);
1806 bridge_get_controllers(const struct bridge *br,
1807 struct ovsrec_controller ***controllersp)
1809 struct ovsrec_controller **controllers;
1810 size_t n_controllers;
1812 controllers = br->cfg->controller;
1813 n_controllers = br->cfg->n_controller;
1815 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1821 *controllersp = controllers;
1823 return n_controllers;
1827 bridge_reconfigure_one(struct bridge *br)
1829 struct shash old_ports, new_ports;
1830 struct svec snoops, old_snoops;
1831 struct shash_node *node;
1832 enum ofproto_fail_mode fail_mode;
1835 /* Collect old ports. */
1836 shash_init(&old_ports);
1837 for (i = 0; i < br->n_ports; i++) {
1838 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1841 /* Collect new ports. */
1842 shash_init(&new_ports);
1843 for (i = 0; i < br->cfg->n_ports; i++) {
1844 const char *name = br->cfg->ports[i]->name;
1845 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1846 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1851 /* If we have a controller, then we need a local port. Complain if the
1852 * user didn't specify one.
1854 * XXX perhaps we should synthesize a port ourselves in this case. */
1855 if (bridge_get_controllers(br, NULL)) {
1856 char local_name[IF_NAMESIZE];
1859 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1860 local_name, sizeof local_name);
1861 if (!error && !shash_find(&new_ports, local_name)) {
1862 VLOG_WARN("bridge %s: controller specified but no local port "
1863 "(port named %s) defined",
1864 br->name, local_name);
1868 /* Get rid of deleted ports.
1869 * Get rid of deleted interfaces on ports that still exist. */
1870 SHASH_FOR_EACH (node, &old_ports) {
1871 struct port *port = node->data;
1872 const struct ovsrec_port *port_cfg;
1874 port_cfg = shash_find_data(&new_ports, node->name);
1878 port_del_ifaces(port, port_cfg);
1882 /* Create new ports.
1883 * Add new interfaces to existing ports.
1884 * Reconfigure existing ports. */
1885 SHASH_FOR_EACH (node, &new_ports) {
1886 struct port *port = shash_find_data(&old_ports, node->name);
1888 port = port_create(br, node->name);
1891 port_reconfigure(port, node->data);
1892 if (!port->n_ifaces) {
1893 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1894 br->name, port->name);
1898 shash_destroy(&old_ports);
1899 shash_destroy(&new_ports);
1901 /* Set the fail-mode */
1902 fail_mode = !br->cfg->fail_mode
1903 || !strcmp(br->cfg->fail_mode, "standalone")
1904 ? OFPROTO_FAIL_STANDALONE
1905 : OFPROTO_FAIL_SECURE;
1906 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1907 && !ofproto_has_primary_controller(br->ofproto)) {
1908 ofproto_flush_flows(br->ofproto);
1910 ofproto_set_fail_mode(br->ofproto, fail_mode);
1912 /* Delete all flows if we're switching from connected to standalone or vice
1913 * versa. (XXX Should we delete all flows if we are switching from one
1914 * controller to another?) */
1916 /* Configure OpenFlow controller connection snooping. */
1918 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1919 ovs_rundir(), br->name));
1920 svec_init(&old_snoops);
1921 ofproto_get_snoops(br->ofproto, &old_snoops);
1922 if (!svec_equal(&snoops, &old_snoops)) {
1923 ofproto_set_snoops(br->ofproto, &snoops);
1925 svec_destroy(&snoops);
1926 svec_destroy(&old_snoops);
1928 mirror_reconfigure(br);
1931 /* Initializes 'oc' appropriately as a management service controller for
1934 * The caller must free oc->target when it is no longer needed. */
1936 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1937 struct ofproto_controller *oc)
1939 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1940 oc->max_backoff = 0;
1941 oc->probe_interval = 60;
1942 oc->band = OFPROTO_OUT_OF_BAND;
1944 oc->burst_limit = 0;
1947 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1949 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1950 struct ofproto_controller *oc)
1952 oc->target = c->target;
1953 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1954 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1955 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1956 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1957 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1958 oc->burst_limit = (c->controller_burst_limit
1959 ? *c->controller_burst_limit : 0);
1962 /* Configures the IP stack for 'br''s local interface properly according to the
1963 * configuration in 'c'. */
1965 bridge_configure_local_iface_netdev(struct bridge *br,
1966 struct ovsrec_controller *c)
1968 struct netdev *netdev;
1969 struct in_addr mask, gateway;
1971 struct iface *local_iface;
1974 /* If there's no local interface or no IP address, give up. */
1975 local_iface = bridge_get_local_iface(br);
1976 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1980 /* Bring up the local interface. */
1981 netdev = local_iface->netdev;
1982 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1984 /* Configure the IP address and netmask. */
1985 if (!c->local_netmask
1986 || !inet_aton(c->local_netmask, &mask)
1988 mask.s_addr = guess_netmask(ip.s_addr);
1990 if (!netdev_set_in4(netdev, ip, mask)) {
1991 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1992 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1995 /* Configure the default gateway. */
1996 if (c->local_gateway
1997 && inet_aton(c->local_gateway, &gateway)
1998 && gateway.s_addr) {
1999 if (!netdev_add_router(netdev, gateway)) {
2000 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2001 br->name, IP_ARGS(&gateway.s_addr));
2007 bridge_reconfigure_remotes(struct bridge *br,
2008 const struct sockaddr_in *managers,
2011 const char *disable_ib_str, *queue_id_str;
2012 bool disable_in_band = false;
2015 struct ovsrec_controller **controllers;
2016 size_t n_controllers;
2019 struct ofproto_controller *ocs;
2023 /* Check if we should disable in-band control on this bridge. */
2024 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2025 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2026 disable_in_band = true;
2029 /* Set OpenFlow queue ID for in-band control. */
2030 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2031 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2032 ofproto_set_in_band_queue(br->ofproto, queue_id);
2034 if (disable_in_band) {
2035 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2037 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2039 had_primary = ofproto_has_primary_controller(br->ofproto);
2041 n_controllers = bridge_get_controllers(br, &controllers);
2043 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2046 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2047 for (i = 0; i < n_controllers; i++) {
2048 struct ovsrec_controller *c = controllers[i];
2050 if (!strncmp(c->target, "punix:", 6)
2051 || !strncmp(c->target, "unix:", 5)) {
2052 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2054 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2055 * domain sockets and overwriting arbitrary local files. */
2056 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2057 "\"%s\" due to possibility for remote exploit",
2058 dpif_name(br->dpif), c->target);
2062 bridge_configure_local_iface_netdev(br, c);
2063 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2064 if (disable_in_band) {
2065 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2070 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2071 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2074 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2075 ofproto_flush_flows(br->ofproto);
2078 /* If there are no controllers and the bridge is in standalone
2079 * mode, set up a flow that matches every packet and directs
2080 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2081 * switch is in secure mode and we won't pass any traffic until
2082 * a controller has been defined and it tells us to do so. */
2084 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2085 union ofp_action action;
2086 struct cls_rule rule;
2088 memset(&action, 0, sizeof action);
2089 action.type = htons(OFPAT_OUTPUT);
2090 action.output.len = htons(sizeof action);
2091 action.output.port = htons(OFPP_NORMAL);
2092 cls_rule_init_catchall(&rule, 0);
2093 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2098 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2103 for (i = 0; i < br->n_ports; i++) {
2104 struct port *port = br->ports[i];
2105 for (j = 0; j < port->n_ifaces; j++) {
2106 struct iface *iface = port->ifaces[j];
2107 shash_add_once(ifaces, iface->name, iface);
2109 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2110 shash_add_once(ifaces, port->name, NULL);
2115 /* For robustness, in case the administrator moves around datapath ports behind
2116 * our back, we re-check all the datapath port numbers here.
2118 * This function will set the 'dp_ifidx' members of interfaces that have
2119 * disappeared to -1, so only call this function from a context where those
2120 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2121 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2122 * datapath, which doesn't support UINT16_MAX+1 ports. */
2124 bridge_fetch_dp_ifaces(struct bridge *br)
2126 struct dpif_port_dump dump;
2127 struct dpif_port dpif_port;
2130 /* Reset all interface numbers. */
2131 for (i = 0; i < br->n_ports; i++) {
2132 struct port *port = br->ports[i];
2133 for (j = 0; j < port->n_ifaces; j++) {
2134 struct iface *iface = port->ifaces[j];
2135 iface->dp_ifidx = -1;
2138 hmap_clear(&br->ifaces);
2140 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2141 struct iface *iface = iface_lookup(br, dpif_port.name);
2143 if (iface->dp_ifidx >= 0) {
2144 VLOG_WARN("%s reported interface %s twice",
2145 dpif_name(br->dpif), dpif_port.name);
2146 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2147 VLOG_WARN("%s reported interface %"PRIu16" twice",
2148 dpif_name(br->dpif), dpif_port.port_no);
2150 iface->dp_ifidx = dpif_port.port_no;
2151 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2152 hash_int(iface->dp_ifidx, 0));
2155 iface_set_ofport(iface->cfg,
2156 (iface->dp_ifidx >= 0
2157 ? odp_port_to_ofp_port(iface->dp_ifidx)
2163 /* Bridge packet processing functions. */
2166 bond_is_tcp_hash(const struct port *port)
2168 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2172 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2174 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2177 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2179 struct flow hash_flow;
2181 memcpy(&hash_flow, flow, sizeof hash_flow);
2182 hash_flow.vlan_tci = 0;
2184 /* The symmetric quality of this hash function is not required, but
2185 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2186 * purposes, so we use it out of convenience. */
2187 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2190 static struct bond_entry *
2191 lookup_bond_entry(const struct port *port, const struct flow *flow,
2194 assert(port->bond_mode != BM_AB);
2196 if (bond_is_tcp_hash(port)) {
2197 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2199 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2204 bond_choose_iface(const struct port *port)
2206 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2207 size_t i, best_down_slave = -1;
2208 long long next_delay_expiration = LLONG_MAX;
2210 for (i = 0; i < port->n_ifaces; i++) {
2211 struct iface *iface = port->ifaces[i];
2213 if (iface->enabled) {
2215 } else if (iface->delay_expires < next_delay_expiration
2216 && lacp_slave_may_enable(port->lacp, iface)) {
2217 best_down_slave = i;
2218 next_delay_expiration = iface->delay_expires;
2222 if (best_down_slave != -1) {
2223 struct iface *iface = port->ifaces[best_down_slave];
2225 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2226 "since no other interface is up", iface->name,
2227 iface->delay_expires - time_msec());
2228 bond_enable_slave(iface, true);
2231 return best_down_slave;
2235 choose_output_iface(const struct port *port, const struct flow *flow,
2236 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2238 struct iface *iface;
2240 assert(port->n_ifaces);
2241 if (port->n_ifaces == 1) {
2242 iface = port->ifaces[0];
2243 } else if (port->bond_mode == BM_AB) {
2244 if (port->active_iface < 0) {
2245 *tags |= port->no_ifaces_tag;
2248 iface = port->ifaces[port->active_iface];
2250 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2251 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2252 || !port->ifaces[e->iface_idx]->enabled) {
2253 /* XXX select interface properly. The current interface selection
2254 * is only good for testing the rebalancing code. */
2255 e->iface_idx = bond_choose_iface(port);
2256 if (e->iface_idx < 0) {
2257 *tags |= port->no_ifaces_tag;
2260 e->iface_tag = tag_create_random();
2262 *tags |= e->iface_tag;
2263 iface = port->ifaces[e->iface_idx];
2265 *dp_ifidx = iface->dp_ifidx;
2266 *tags |= iface->tag; /* Currently only used for bonding. */
2271 bond_link_status_update(struct iface *iface)
2273 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2274 struct port *port = iface->port;
2275 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2276 int updelay, downdelay;
2278 updelay = port->updelay;
2279 downdelay = port->downdelay;
2281 if (lacp_negotiated(port->lacp)) {
2286 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2287 /* Nothing to do. */
2290 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2291 iface->name, up ? "up" : "down");
2292 if (up == iface->enabled) {
2293 iface->delay_expires = LLONG_MAX;
2294 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2295 iface->name, up ? "disabled" : "enabled");
2296 } else if (up && port->active_iface < 0) {
2297 bond_enable_slave(iface, true);
2299 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2300 "other interface is up", iface->name, updelay);
2303 int delay = up ? updelay : downdelay;
2304 iface->delay_expires = time_msec() + delay;
2307 "interface %s: will be %s if it stays %s for %d ms",
2309 up ? "enabled" : "disabled",
2317 bond_choose_active_iface(struct port *port)
2319 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2321 port->active_iface = bond_choose_iface(port);
2322 port->active_iface_tag = tag_create_random();
2323 if (port->active_iface >= 0) {
2324 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2325 port->name, port->ifaces[port->active_iface]->name);
2327 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2333 bond_enable_slave(struct iface *iface, bool enable)
2335 struct port *port = iface->port;
2336 struct bridge *br = port->bridge;
2338 /* This acts as a recursion check. If the act of disabling a slave
2339 * causes a different slave to be enabled, the flag will allow us to
2340 * skip redundant work when we reenter this function. It must be
2341 * cleared on exit to keep things safe with multiple bonds. */
2342 static bool moving_active_iface = false;
2344 iface->delay_expires = LLONG_MAX;
2345 if (enable == iface->enabled) {
2349 iface->enabled = enable;
2350 if (!iface->enabled) {
2351 VLOG_WARN("interface %s: disabled", iface->name);
2352 ofproto_revalidate(br->ofproto, iface->tag);
2353 if (iface->port_ifidx == port->active_iface) {
2354 ofproto_revalidate(br->ofproto,
2355 port->active_iface_tag);
2357 /* Disabling a slave can lead to another slave being immediately
2358 * enabled if there will be no active slaves but one is waiting
2359 * on an updelay. In this case we do not need to run most of the
2360 * code for the newly enabled slave since there was no period
2361 * without an active slave and it is redundant with the disabling
2363 moving_active_iface = true;
2364 bond_choose_active_iface(port);
2366 bond_send_learning_packets(port);
2368 VLOG_WARN("interface %s: enabled", iface->name);
2369 if (port->active_iface < 0 && !moving_active_iface) {
2370 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2371 bond_choose_active_iface(port);
2372 bond_send_learning_packets(port);
2374 iface->tag = tag_create_random();
2377 moving_active_iface = false;
2380 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2381 * bond interface. */
2383 bond_update_fake_iface_stats(struct port *port)
2385 struct netdev_stats bond_stats;
2386 struct netdev *bond_dev;
2389 memset(&bond_stats, 0, sizeof bond_stats);
2391 for (i = 0; i < port->n_ifaces; i++) {
2392 struct netdev_stats slave_stats;
2394 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2395 /* XXX: We swap the stats here because they are swapped back when
2396 * reported by the internal device. The reason for this is
2397 * internal devices normally represent packets going into the system
2398 * but when used as fake bond device they represent packets leaving
2399 * the system. We really should do this in the internal device
2400 * itself because changing it here reverses the counts from the
2401 * perspective of the switch. However, the internal device doesn't
2402 * know what type of device it represents so we have to do it here
2404 bond_stats.tx_packets += slave_stats.rx_packets;
2405 bond_stats.tx_bytes += slave_stats.rx_bytes;
2406 bond_stats.rx_packets += slave_stats.tx_packets;
2407 bond_stats.rx_bytes += slave_stats.tx_bytes;
2411 if (!netdev_open_default(port->name, &bond_dev)) {
2412 netdev_set_stats(bond_dev, &bond_stats);
2413 netdev_close(bond_dev);
2418 bond_run(struct port *port)
2422 if (port->n_ifaces < 2) {
2426 for (i = 0; i < port->n_ifaces; i++) {
2427 bond_link_status_update(port->ifaces[i]);
2430 for (i = 0; i < port->n_ifaces; i++) {
2431 struct iface *iface = port->ifaces[i];
2432 if (time_msec() >= iface->delay_expires) {
2433 bond_enable_slave(iface, !iface->enabled);
2437 if (port->bond_fake_iface
2438 && time_msec() >= port->bond_next_fake_iface_update) {
2439 bond_update_fake_iface_stats(port);
2440 port->bond_next_fake_iface_update = time_msec() + 1000;
2445 bond_wait(struct port *port)
2449 if (port->n_ifaces < 2) {
2453 for (i = 0; i < port->n_ifaces; i++) {
2454 struct iface *iface = port->ifaces[i];
2455 if (iface->delay_expires != LLONG_MAX) {
2456 poll_timer_wait_until(iface->delay_expires);
2460 if (port->bond_fake_iface) {
2461 poll_timer_wait_until(port->bond_next_fake_iface_update);
2466 set_dst(struct dst *dst, const struct flow *flow,
2467 const struct port *in_port, const struct port *out_port,
2470 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2471 : in_port->vlan >= 0 ? in_port->vlan
2472 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2473 : vlan_tci_to_vid(flow->vlan_tci));
2474 return choose_output_iface(out_port, flow, dst->vlan,
2475 &dst->dp_ifidx, tags);
2479 swap_dst(struct dst *p, struct dst *q)
2481 struct dst tmp = *p;
2486 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2487 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2488 * that we push to the datapath. We could in fact fully sort the array by
2489 * vlan, but in most cases there are at most two different vlan tags so that's
2490 * possibly overkill.) */
2492 partition_dsts(struct dst_set *set, int vlan)
2494 struct dst *first = set->dsts;
2495 struct dst *last = set->dsts + set->n;
2497 while (first != last) {
2499 * - All dsts < first have vlan == 'vlan'.
2500 * - All dsts >= last have vlan != 'vlan'.
2501 * - first < last. */
2502 while (first->vlan == vlan) {
2503 if (++first == last) {
2508 /* Same invariants, plus one additional:
2509 * - first->vlan != vlan.
2511 while (last[-1].vlan != vlan) {
2512 if (--last == first) {
2517 /* Same invariants, plus one additional:
2518 * - last[-1].vlan == vlan.*/
2519 swap_dst(first++, --last);
2524 mirror_mask_ffs(mirror_mask_t mask)
2526 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2531 dst_set_init(struct dst_set *set)
2533 set->dsts = set->builtin;
2535 set->allocated = ARRAY_SIZE(set->builtin);
2539 dst_set_add(struct dst_set *set, const struct dst *dst)
2541 if (set->n >= set->allocated) {
2542 size_t new_allocated;
2543 struct dst *new_dsts;
2545 new_allocated = set->allocated * 2;
2546 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2547 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2551 set->dsts = new_dsts;
2552 set->allocated = new_allocated;
2554 set->dsts[set->n++] = *dst;
2558 dst_set_free(struct dst_set *set)
2560 if (set->dsts != set->builtin) {
2566 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2569 for (i = 0; i < set->n; i++) {
2570 if (set->dsts[i].vlan == test->vlan
2571 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2579 port_trunks_vlan(const struct port *port, uint16_t vlan)
2581 return (port->vlan < 0
2582 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2586 port_includes_vlan(const struct port *port, uint16_t vlan)
2588 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2592 port_is_floodable(const struct port *port)
2596 for (i = 0; i < port->n_ifaces; i++) {
2597 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2598 port->ifaces[i]->dp_ifidx)) {
2606 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2607 const struct port *in_port, const struct port *out_port,
2608 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2610 mirror_mask_t mirrors = in_port->src_mirrors;
2615 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2616 if (flow_vlan == 0) {
2617 flow_vlan = OFP_VLAN_NONE;
2620 if (out_port == FLOOD_PORT) {
2621 for (i = 0; i < br->n_ports; i++) {
2622 struct port *port = br->ports[i];
2624 && port_is_floodable(port)
2625 && port_includes_vlan(port, vlan)
2626 && !port->is_mirror_output_port
2627 && set_dst(&dst, flow, in_port, port, tags)) {
2628 mirrors |= port->dst_mirrors;
2629 dst_set_add(set, &dst);
2632 *nf_output_iface = NF_OUT_FLOOD;
2633 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2634 dst_set_add(set, &dst);
2635 *nf_output_iface = dst.dp_ifidx;
2636 mirrors |= out_port->dst_mirrors;
2640 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2641 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2643 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2644 && !dst_is_duplicate(set, &dst)) {
2645 dst_set_add(set, &dst);
2648 for (i = 0; i < br->n_ports; i++) {
2649 struct port *port = br->ports[i];
2650 if (port_includes_vlan(port, m->out_vlan)
2651 && set_dst(&dst, flow, in_port, port, tags))
2653 if (port->vlan < 0) {
2654 dst.vlan = m->out_vlan;
2656 if (dst_is_duplicate(set, &dst)) {
2660 /* Use the vlan tag on the original flow instead of
2661 * the one passed in the vlan parameter. This ensures
2662 * that we compare the vlan from before any implicit
2663 * tagging tags place. This is necessary because
2664 * dst->vlan is the final vlan, after removing implicit
2666 if (port == in_port && dst.vlan == flow_vlan) {
2667 /* Don't send out input port on same VLAN. */
2670 dst_set_add(set, &dst);
2675 mirrors &= mirrors - 1;
2678 partition_dsts(set, flow_vlan);
2681 static void OVS_UNUSED
2682 print_dsts(const struct dst_set *set)
2686 for (i = 0; i < set->n; i++) {
2687 const struct dst *dst = &set->dsts[i];
2689 printf(">p%"PRIu16, dst->dp_ifidx);
2690 if (dst->vlan != OFP_VLAN_NONE) {
2691 printf("v%"PRIu16, dst->vlan);
2697 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2698 const struct port *in_port, const struct port *out_port,
2699 tag_type *tags, struct ofpbuf *actions,
2700 uint16_t *nf_output_iface)
2707 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2710 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2711 if (cur_vlan == 0) {
2712 cur_vlan = OFP_VLAN_NONE;
2714 for (i = 0; i < set.n; i++) {
2715 const struct dst *dst = &set.dsts[i];
2716 if (dst->vlan != cur_vlan) {
2717 if (dst->vlan == OFP_VLAN_NONE) {
2718 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2721 tci = htons(dst->vlan & VLAN_VID_MASK);
2722 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2723 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2725 cur_vlan = dst->vlan;
2727 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2732 /* Returns the effective vlan of a packet, taking into account both the
2733 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2734 * the packet is untagged and -1 indicates it has an invalid header and
2735 * should be dropped. */
2736 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2737 struct port *in_port, bool have_packet)
2739 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2740 if (in_port->vlan >= 0) {
2742 /* XXX support double tagging? */
2744 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2745 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2746 "packet received on port %s configured with "
2747 "implicit VLAN %"PRIu16,
2748 br->name, vlan, in_port->name, in_port->vlan);
2752 vlan = in_port->vlan;
2754 if (!port_includes_vlan(in_port, vlan)) {
2756 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2757 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2758 "packet received on port %s not configured for "
2760 br->name, vlan, in_port->name, vlan);
2769 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2770 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2771 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2773 is_gratuitous_arp(const struct flow *flow)
2775 return (flow->dl_type == htons(ETH_TYPE_ARP)
2776 && eth_addr_is_broadcast(flow->dl_dst)
2777 && (flow->nw_proto == ARP_OP_REPLY
2778 || (flow->nw_proto == ARP_OP_REQUEST
2779 && flow->nw_src == flow->nw_dst)));
2783 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2784 struct port *in_port)
2786 enum grat_arp_lock_type lock_type;
2789 /* We don't want to learn from gratuitous ARP packets that are reflected
2790 * back over bond slaves so we lock the learning table. */
2791 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2792 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2793 GRAT_ARP_LOCK_CHECK;
2795 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2798 /* The log messages here could actually be useful in debugging,
2799 * so keep the rate limit relatively high. */
2800 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2802 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2803 "on port %s in VLAN %d",
2804 br->name, ETH_ADDR_ARGS(flow->dl_src),
2805 in_port->name, vlan);
2806 ofproto_revalidate(br->ofproto, rev_tag);
2810 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2811 * dropped. Returns true if they may be forwarded, false if they should be
2814 * If 'have_packet' is true, it indicates that the caller is processing a
2815 * received packet. If 'have_packet' is false, then the caller is just
2816 * revalidating an existing flow because configuration has changed. Either
2817 * way, 'have_packet' only affects logging (there is no point in logging errors
2818 * during revalidation).
2820 * Sets '*in_portp' to the input port. This will be a null pointer if
2821 * flow->in_port does not designate a known input port (in which case
2822 * is_admissible() returns false).
2824 * When returning true, sets '*vlanp' to the effective VLAN of the input
2825 * packet, as returned by flow_get_vlan().
2827 * May also add tags to '*tags', although the current implementation only does
2828 * so in one special case.
2831 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2832 tag_type *tags, int *vlanp, struct port **in_portp)
2834 struct iface *in_iface;
2835 struct port *in_port;
2838 /* Find the interface and port structure for the received packet. */
2839 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2841 /* No interface? Something fishy... */
2843 /* Odd. A few possible reasons here:
2845 * - We deleted an interface but there are still a few packets
2846 * queued up from it.
2848 * - Someone externally added an interface (e.g. with "ovs-dpctl
2849 * add-if") that we don't know about.
2851 * - Packet arrived on the local port but the local port is not
2852 * one of our bridge ports.
2854 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2856 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2857 "interface %"PRIu16, br->name, flow->in_port);
2863 *in_portp = in_port = in_iface->port;
2864 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2869 /* Drop frames for reserved multicast addresses. */
2870 if (eth_addr_is_reserved(flow->dl_dst)) {
2874 /* Drop frames on ports reserved for mirroring. */
2875 if (in_port->is_mirror_output_port) {
2877 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2878 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2879 "%s, which is reserved exclusively for mirroring",
2880 br->name, in_port->name);
2885 /* When using LACP, do not accept packets from disabled interfaces. */
2886 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2890 /* Packets received on non-LACP bonds need special attention to avoid
2892 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2894 bool is_grat_arp_locked;
2896 if (eth_addr_is_multicast(flow->dl_dst)) {
2897 *tags |= in_port->active_iface_tag;
2898 if (in_port->active_iface != in_iface->port_ifidx) {
2899 /* Drop all multicast packets on inactive slaves. */
2904 /* Drop all packets for which we have learned a different input
2905 * port, because we probably sent the packet on one slave and got
2906 * it back on the other. Gratuitous ARP packets are an exception
2907 * to this rule: the host has moved to another switch. The exception
2908 * to the exception is if we locked the learning table to avoid
2909 * reflections on bond slaves. If this is the case, just drop the
2911 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2912 &is_grat_arp_locked);
2913 if (src_idx != -1 && src_idx != in_port->port_idx &&
2914 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2922 /* If the composed actions may be applied to any packet in the given 'flow',
2923 * returns true. Otherwise, the actions should only be applied to 'packet', or
2924 * not at all, if 'packet' was NULL. */
2926 process_flow(struct bridge *br, const struct flow *flow,
2927 const struct ofpbuf *packet, struct ofpbuf *actions,
2928 tag_type *tags, uint16_t *nf_output_iface)
2930 struct port *in_port;
2931 struct port *out_port;
2935 /* Check whether we should drop packets in this flow. */
2936 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2941 /* Learn source MAC (but don't try to learn from revalidation). */
2943 update_learning_table(br, flow, vlan, in_port);
2946 /* Determine output port. */
2947 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2949 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2950 out_port = br->ports[out_port_idx];
2951 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2952 /* If we are revalidating but don't have a learning entry then
2953 * eject the flow. Installing a flow that floods packets opens
2954 * up a window of time where we could learn from a packet reflected
2955 * on a bond and blackhole packets before the learning table is
2956 * updated to reflect the correct port. */
2959 out_port = FLOOD_PORT;
2962 /* Don't send packets out their input ports. */
2963 if (in_port == out_port) {
2969 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2977 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2978 struct ofpbuf *actions, tag_type *tags,
2979 uint16_t *nf_output_iface, void *br_)
2981 struct bridge *br = br_;
2983 COVERAGE_INC(bridge_process_flow);
2984 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2988 bridge_special_ofhook_cb(const struct flow *flow,
2989 const struct ofpbuf *packet, void *br_)
2991 struct iface *iface;
2992 struct bridge *br = br_;
2994 iface = iface_from_dp_ifidx(br, flow->in_port);
2996 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
2998 if (iface && iface->port->lacp && packet) {
2999 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
3002 COVERAGE_INC(bridge_process_lacp);
3003 lacp_process_pdu(iface->port->lacp, iface, pdu);
3013 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3014 const struct nlattr *actions,
3016 uint64_t n_bytes, void *br_)
3018 struct bridge *br = br_;
3019 const struct nlattr *a;
3020 struct port *in_port;
3025 /* Feed information from the active flows back into the learning table to
3026 * ensure that table is always in sync with what is actually flowing
3027 * through the datapath.
3029 * We test that 'tags' is nonzero to ensure that only flows that include an
3030 * OFPP_NORMAL action are used for learning. This works because
3031 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3032 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3033 update_learning_table(br, flow, vlan, in_port);
3036 /* Account for bond slave utilization. */
3037 if (!br->has_bonded_ports) {
3040 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3041 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3042 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3043 if (out_port && out_port->n_ifaces >= 2 &&
3044 out_port->bond_mode != BM_AB) {
3045 uint16_t vlan = (flow->vlan_tci
3046 ? vlan_tci_to_vid(flow->vlan_tci)
3048 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3049 e->tx_bytes += n_bytes;
3056 bridge_account_checkpoint_ofhook_cb(void *br_)
3058 struct bridge *br = br_;
3062 if (!br->has_bonded_ports) {
3067 for (i = 0; i < br->n_ports; i++) {
3068 struct port *port = br->ports[i];
3069 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3070 && now >= port->bond_next_rebalance) {
3071 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3072 bond_rebalance_port(port);
3077 static struct ofhooks bridge_ofhooks = {
3078 bridge_normal_ofhook_cb,
3079 bridge_special_ofhook_cb,
3080 bridge_account_flow_ofhook_cb,
3081 bridge_account_checkpoint_ofhook_cb,
3084 /* Bonding functions. */
3086 /* Statistics for a single interface on a bonded port, used for load-based
3087 * bond rebalancing. */
3088 struct slave_balance {
3089 struct iface *iface; /* The interface. */
3090 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3092 /* All the "bond_entry"s that are assigned to this interface, in order of
3093 * increasing tx_bytes. */
3094 struct bond_entry **hashes;
3099 bond_mode_to_string(enum bond_mode bm) {
3100 static char *bm_slb = "balance-slb";
3101 static char *bm_ab = "active-backup";
3102 static char *bm_tcp = "balance-tcp";
3105 case BM_SLB: return bm_slb;
3106 case BM_AB: return bm_ab;
3107 case BM_TCP: return bm_tcp;
3114 /* Sorts pointers to pointers to bond_entries in ascending order by the
3115 * interface to which they are assigned, and within a single interface in
3116 * ascending order of bytes transmitted. */
3118 compare_bond_entries(const void *a_, const void *b_)
3120 const struct bond_entry *const *ap = a_;
3121 const struct bond_entry *const *bp = b_;
3122 const struct bond_entry *a = *ap;
3123 const struct bond_entry *b = *bp;
3124 if (a->iface_idx != b->iface_idx) {
3125 return a->iface_idx > b->iface_idx ? 1 : -1;
3126 } else if (a->tx_bytes != b->tx_bytes) {
3127 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3133 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3134 * *descending* order by number of bytes transmitted. */
3136 compare_slave_balance(const void *a_, const void *b_)
3138 const struct slave_balance *a = a_;
3139 const struct slave_balance *b = b_;
3140 if (a->iface->enabled != b->iface->enabled) {
3141 return a->iface->enabled ? -1 : 1;
3142 } else if (a->tx_bytes != b->tx_bytes) {
3143 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3150 swap_bals(struct slave_balance *a, struct slave_balance *b)
3152 struct slave_balance tmp = *a;
3157 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3158 * given that 'p' (and only 'p') might be in the wrong location.
3160 * This function invalidates 'p', since it might now be in a different memory
3163 resort_bals(struct slave_balance *p,
3164 struct slave_balance bals[], size_t n_bals)
3167 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3168 swap_bals(p, p - 1);
3170 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3171 swap_bals(p, p + 1);
3177 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3179 if (VLOG_IS_DBG_ENABLED()) {
3180 struct ds ds = DS_EMPTY_INITIALIZER;
3181 const struct slave_balance *b;
3183 for (b = bals; b < bals + n_bals; b++) {
3187 ds_put_char(&ds, ',');
3189 ds_put_format(&ds, " %s %"PRIu64"kB",
3190 b->iface->name, b->tx_bytes / 1024);
3192 if (!b->iface->enabled) {
3193 ds_put_cstr(&ds, " (disabled)");
3195 if (b->n_hashes > 0) {
3196 ds_put_cstr(&ds, " (");
3197 for (i = 0; i < b->n_hashes; i++) {
3198 const struct bond_entry *e = b->hashes[i];
3200 ds_put_cstr(&ds, " + ");
3202 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3203 e - port->bond_hash, e->tx_bytes / 1024);
3205 ds_put_cstr(&ds, ")");
3208 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3213 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3215 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3218 struct bond_entry *hash = from->hashes[hash_idx];
3219 struct port *port = from->iface->port;
3220 uint64_t delta = hash->tx_bytes;
3222 assert(port->bond_mode != BM_AB);
3224 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3225 "from %s to %s (now carrying %"PRIu64"kB and "
3226 "%"PRIu64"kB load, respectively)",
3227 port->name, delta / 1024, hash - port->bond_hash,
3228 from->iface->name, to->iface->name,
3229 (from->tx_bytes - delta) / 1024,
3230 (to->tx_bytes + delta) / 1024);
3232 /* Delete element from from->hashes.
3234 * We don't bother to add the element to to->hashes because not only would
3235 * it require more work, the only purpose it would be to allow that hash to
3236 * be migrated to another slave in this rebalancing run, and there is no
3237 * point in doing that. */
3238 if (hash_idx == 0) {
3241 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3242 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3246 /* Shift load away from 'from' to 'to'. */
3247 from->tx_bytes -= delta;
3248 to->tx_bytes += delta;
3250 /* Arrange for flows to be revalidated. */
3251 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3252 hash->iface_idx = to->iface->port_ifidx;
3253 hash->iface_tag = tag_create_random();
3257 bond_rebalance_port(struct port *port)
3259 struct slave_balance *bals;
3261 struct bond_entry *hashes[BOND_MASK + 1];
3262 struct slave_balance *b, *from, *to;
3263 struct bond_entry *e;
3266 assert(port->bond_mode != BM_AB);
3268 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3269 * descending order of tx_bytes, so that bals[0] represents the most
3270 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3273 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3274 * array for each slave_balance structure, we sort our local array of
3275 * hashes in order by slave, so that all of the hashes for a given slave
3276 * become contiguous in memory, and then we point each 'hashes' members of
3277 * a slave_balance structure to the start of a contiguous group. */
3278 n_bals = port->n_ifaces;
3279 bals = xmalloc(n_bals * sizeof *bals);
3280 for (b = bals; b < &bals[n_bals]; b++) {
3281 b->iface = port->ifaces[b - bals];
3286 for (i = 0; i <= BOND_MASK; i++) {
3287 hashes[i] = &port->bond_hash[i];
3289 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3290 for (i = 0; i <= BOND_MASK; i++) {
3292 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3293 b = &bals[e->iface_idx];
3294 b->tx_bytes += e->tx_bytes;
3296 b->hashes = &hashes[i];
3301 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3302 log_bals(bals, n_bals, port);
3304 /* Discard slaves that aren't enabled (which were sorted to the back of the
3305 * array earlier). */
3306 while (!bals[n_bals - 1].iface->enabled) {
3313 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3314 to = &bals[n_bals - 1];
3315 for (from = bals; from < to; ) {
3316 uint64_t overload = from->tx_bytes - to->tx_bytes;
3317 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3318 /* The extra load on 'from' (and all less-loaded slaves), compared
3319 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3320 * it is less than ~1Mbps. No point in rebalancing. */
3322 } else if (from->n_hashes == 1) {
3323 /* 'from' only carries a single MAC hash, so we can't shift any
3324 * load away from it, even though we want to. */
3327 /* 'from' is carrying significantly more load than 'to', and that
3328 * load is split across at least two different hashes. Pick a hash
3329 * to migrate to 'to' (the least-loaded slave), given that doing so
3330 * must decrease the ratio of the load on the two slaves by at
3333 * The sort order we use means that we prefer to shift away the
3334 * smallest hashes instead of the biggest ones. There is little
3335 * reason behind this decision; we could use the opposite sort
3336 * order to shift away big hashes ahead of small ones. */
3339 for (i = 0; i < from->n_hashes; i++) {
3340 double old_ratio, new_ratio;
3341 uint64_t delta = from->hashes[i]->tx_bytes;
3343 if (delta == 0 || from->tx_bytes - delta == 0) {
3344 /* Pointless move. */
3348 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3350 if (to->tx_bytes == 0) {
3351 /* Nothing on the new slave, move it. */
3355 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3356 new_ratio = (double)(from->tx_bytes - delta) /
3357 (to->tx_bytes + delta);
3359 if (new_ratio == 0) {
3360 /* Should already be covered but check to prevent division
3365 if (new_ratio < 1) {
3366 new_ratio = 1 / new_ratio;
3369 if (old_ratio - new_ratio > 0.1) {
3370 /* Would decrease the ratio, move it. */
3374 if (i < from->n_hashes) {
3375 bond_shift_load(from, to, i);
3377 /* If the result of the migration changed the relative order of
3378 * 'from' and 'to' swap them back to maintain invariants. */
3379 if (order_swapped) {
3380 swap_bals(from, to);
3383 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3384 * point to different slave_balance structures. It is only
3385 * valid to do these two operations in a row at all because we
3386 * know that 'from' will not move past 'to' and vice versa. */
3387 resort_bals(from, bals, n_bals);
3388 resort_bals(to, bals, n_bals);
3395 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3396 * historical data to decay to <1% in 7 rebalancing runs. */
3397 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3406 bond_send_learning_packets(struct port *port)
3408 struct bridge *br = port->bridge;
3409 struct mac_entry *e;
3410 struct ofpbuf packet;
3411 int error, n_packets, n_errors;
3413 if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
3417 ofpbuf_init(&packet, 128);
3418 error = n_packets = n_errors = 0;
3419 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3425 if (e->port == port->port_idx) {
3429 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3431 flow_extract(&packet, 0, ODPP_NONE, &flow);
3433 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3439 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3445 ofpbuf_uninit(&packet);
3448 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3449 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3450 "packets, last error was: %s",
3451 port->name, n_errors, n_packets, strerror(error));
3453 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3454 port->name, n_packets);
3458 /* Bonding unixctl user interface functions. */
3461 bond_unixctl_list(struct unixctl_conn *conn,
3462 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3464 struct ds ds = DS_EMPTY_INITIALIZER;
3465 const struct bridge *br;
3467 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3469 LIST_FOR_EACH (br, node, &all_bridges) {
3472 for (i = 0; i < br->n_ports; i++) {
3473 const struct port *port = br->ports[i];
3474 if (port->n_ifaces > 1) {
3477 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3478 bond_mode_to_string(port->bond_mode));
3479 for (j = 0; j < port->n_ifaces; j++) {
3480 const struct iface *iface = port->ifaces[j];
3482 ds_put_cstr(&ds, ", ");
3484 ds_put_cstr(&ds, iface->name);
3486 ds_put_char(&ds, '\n');
3490 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3494 static struct port *
3495 bond_find(const char *name)
3497 const struct bridge *br;
3499 LIST_FOR_EACH (br, node, &all_bridges) {
3502 for (i = 0; i < br->n_ports; i++) {
3503 struct port *port = br->ports[i];
3504 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3513 bond_unixctl_show(struct unixctl_conn *conn,
3514 const char *args, void *aux OVS_UNUSED)
3516 struct ds ds = DS_EMPTY_INITIALIZER;
3517 const struct port *port;
3520 port = bond_find(args);
3522 unixctl_command_reply(conn, 501, "no such bond");
3526 ds_put_format(&ds, "bond_mode: %s\n",
3527 bond_mode_to_string(port->bond_mode));
3530 ds_put_format(&ds, "lacp: %s\n",
3531 port->lacp_active ? "active" : "passive");
3533 ds_put_cstr(&ds, "lacp: off\n");
3536 if (port->bond_mode != BM_AB) {
3537 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3538 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3542 ds_put_format(&ds, "bond-detect-mode: %s\n",
3543 port->monitor ? "carrier" : "miimon");
3545 if (!port->monitor) {
3546 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3547 port->miimon_interval);
3550 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3551 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3553 if (port->bond_mode != BM_AB) {
3554 ds_put_format(&ds, "next rebalance: %lld ms\n",
3555 port->bond_next_rebalance - time_msec());
3558 for (j = 0; j < port->n_ifaces; j++) {
3559 const struct iface *iface = port->ifaces[j];
3560 struct bond_entry *be;
3564 ds_put_format(&ds, "\nslave %s: %s\n",
3565 iface->name, iface->enabled ? "enabled" : "disabled");
3566 if (j == port->active_iface) {
3567 ds_put_cstr(&ds, "\tactive slave\n");
3569 if (iface->delay_expires != LLONG_MAX) {
3570 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3571 iface->enabled ? "downdelay" : "updelay",
3572 iface->delay_expires - time_msec());
3575 if (port->bond_mode == BM_AB) {
3580 memset(&flow, 0, sizeof flow);
3581 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3582 int hash = be - port->bond_hash;
3583 struct mac_entry *me;
3585 if (be->iface_idx != j) {
3589 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3590 hash, be->tx_bytes / 1024);
3592 if (port->bond_mode != BM_SLB) {
3597 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3601 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3602 if (bond_hash_src(me->mac, me->vlan) == hash
3603 && me->port != port->port_idx
3604 && choose_output_iface(port, &flow, me->vlan,
3606 && dp_ifidx == iface->dp_ifidx)
3608 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3609 ETH_ADDR_ARGS(me->mac));
3614 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3619 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3620 void *aux OVS_UNUSED)
3622 char *args = (char *) args_;
3623 char *save_ptr = NULL;
3624 char *bond_s, *hash_s, *slave_s;
3626 struct iface *iface;
3627 struct bond_entry *entry;
3630 bond_s = strtok_r(args, " ", &save_ptr);
3631 hash_s = strtok_r(NULL, " ", &save_ptr);
3632 slave_s = strtok_r(NULL, " ", &save_ptr);
3634 unixctl_command_reply(conn, 501,
3635 "usage: bond/migrate BOND HASH SLAVE");
3639 port = bond_find(bond_s);
3641 unixctl_command_reply(conn, 501, "no such bond");
3645 if (port->bond_mode != BM_SLB) {
3646 unixctl_command_reply(conn, 501, "not an SLB bond");
3650 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3651 hash = atoi(hash_s) & BOND_MASK;
3653 unixctl_command_reply(conn, 501, "bad hash");
3657 iface = port_lookup_iface(port, slave_s);
3659 unixctl_command_reply(conn, 501, "no such slave");
3663 if (!iface->enabled) {
3664 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3668 entry = &port->bond_hash[hash];
3669 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3670 entry->iface_idx = iface->port_ifidx;
3671 entry->iface_tag = tag_create_random();
3672 unixctl_command_reply(conn, 200, "migrated");
3676 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3677 void *aux OVS_UNUSED)
3679 char *args = (char *) args_;
3680 char *save_ptr = NULL;
3681 char *bond_s, *slave_s;
3683 struct iface *iface;
3685 bond_s = strtok_r(args, " ", &save_ptr);
3686 slave_s = strtok_r(NULL, " ", &save_ptr);
3688 unixctl_command_reply(conn, 501,
3689 "usage: bond/set-active-slave BOND SLAVE");
3693 port = bond_find(bond_s);
3695 unixctl_command_reply(conn, 501, "no such bond");
3699 iface = port_lookup_iface(port, slave_s);
3701 unixctl_command_reply(conn, 501, "no such slave");
3705 if (!iface->enabled) {
3706 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3710 if (port->active_iface != iface->port_ifidx) {
3711 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3712 port->active_iface = iface->port_ifidx;
3713 port->active_iface_tag = tag_create_random();
3714 VLOG_INFO("port %s: active interface is now %s",
3715 port->name, iface->name);
3716 bond_send_learning_packets(port);
3717 unixctl_command_reply(conn, 200, "done");
3719 unixctl_command_reply(conn, 200, "no change");
3724 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3726 char *args = (char *) args_;
3727 char *save_ptr = NULL;
3728 char *bond_s, *slave_s;
3730 struct iface *iface;
3732 bond_s = strtok_r(args, " ", &save_ptr);
3733 slave_s = strtok_r(NULL, " ", &save_ptr);
3735 unixctl_command_reply(conn, 501,
3736 "usage: bond/enable/disable-slave BOND SLAVE");
3740 port = bond_find(bond_s);
3742 unixctl_command_reply(conn, 501, "no such bond");
3746 iface = port_lookup_iface(port, slave_s);
3748 unixctl_command_reply(conn, 501, "no such slave");
3752 bond_enable_slave(iface, enable);
3753 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3757 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3758 void *aux OVS_UNUSED)
3760 enable_slave(conn, args, true);
3764 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3765 void *aux OVS_UNUSED)
3767 enable_slave(conn, args, false);
3771 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3772 void *aux OVS_UNUSED)
3774 char *args = (char *) args_;
3775 uint8_t mac[ETH_ADDR_LEN];
3779 char *mac_s, *vlan_s;
3780 char *save_ptr = NULL;
3782 mac_s = strtok_r(args, " ", &save_ptr);
3783 vlan_s = strtok_r(NULL, " ", &save_ptr);
3786 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3787 unixctl_command_reply(conn, 501, "invalid vlan");
3791 vlan = OFP_VLAN_NONE;
3794 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3795 == ETH_ADDR_SCAN_COUNT) {
3796 hash = bond_hash_src(mac, vlan);
3798 hash_cstr = xasprintf("%u", hash);
3799 unixctl_command_reply(conn, 200, hash_cstr);
3802 unixctl_command_reply(conn, 501, "invalid mac");
3809 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3810 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3811 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3812 unixctl_command_register("bond/set-active-slave",
3813 bond_unixctl_set_active_slave, NULL);
3814 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3816 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3818 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3821 /* Port functions. */
3824 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3826 struct iface *iface = aux;
3827 uint8_t ea[ETH_ADDR_LEN];
3830 error = netdev_get_etheraddr(iface->netdev, ea);
3832 struct ofpbuf packet;
3834 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3835 compose_lacp_packet(&packet, ea, pdu);
3836 ofproto_send_packet(iface->port->bridge->ofproto,
3837 iface->dp_ifidx, 0, &packet);
3838 ofpbuf_uninit(&packet);
3840 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3841 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3842 "(%s)", iface->name, strerror(error));
3847 port_run(struct port *port)
3849 if (port->monitor) {
3852 /* Track carrier going up and down on interfaces. */
3853 while (!netdev_monitor_poll(port->monitor, &devname)) {
3854 struct iface *iface;
3856 iface = port_lookup_iface(port, devname);
3858 iface_update_carrier(iface);
3862 } else if (time_msec() >= port->miimon_next_update) {
3865 for (i = 0; i < port->n_ifaces; i++) {
3866 struct iface *iface = port->ifaces[i];
3867 iface_update_carrier(iface);
3869 port->miimon_next_update = time_msec() + port->miimon_interval;
3875 for (i = 0; i < port->n_ifaces; i++) {
3876 struct iface *iface = port->ifaces[i];
3877 lacp_slave_enable(port->lacp, iface, iface->enabled);
3880 lacp_run(port->lacp, lacp_send_pdu_cb);
3887 port_wait(struct port *port)
3889 if (port->monitor) {
3890 netdev_monitor_poll_wait(port->monitor);
3892 poll_timer_wait_until(port->miimon_next_update);
3896 lacp_wait(port->lacp);
3902 static struct port *
3903 port_create(struct bridge *br, const char *name)
3907 port = xzalloc(sizeof *port);
3909 port->port_idx = br->n_ports;
3911 port->trunks = NULL;
3912 port->name = xstrdup(name);
3913 port->active_iface = -1;
3915 if (br->n_ports >= br->allocated_ports) {
3916 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3919 br->ports[br->n_ports++] = port;
3920 shash_add_assert(&br->port_by_name, port->name, port);
3922 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3929 get_port_other_config(const struct ovsrec_port *port, const char *key,
3930 const char *default_value)
3934 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3936 return value ? value : default_value;
3940 get_interface_other_config(const struct ovsrec_interface *iface,
3941 const char *key, const char *default_value)
3945 value = get_ovsrec_key_value(&iface->header_,
3946 &ovsrec_interface_col_other_config, key);
3947 return value ? value : default_value;
3951 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3953 struct shash new_ifaces;
3956 /* Collect list of new interfaces. */
3957 shash_init(&new_ifaces);
3958 for (i = 0; i < cfg->n_interfaces; i++) {
3959 const char *name = cfg->interfaces[i]->name;
3960 shash_add_once(&new_ifaces, name, NULL);
3963 /* Get rid of deleted interfaces. */
3964 for (i = 0; i < port->n_ifaces; ) {
3965 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3966 iface_destroy(port->ifaces[i]);
3972 shash_destroy(&new_ifaces);
3976 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3978 const char *detect_mode;
3979 struct shash new_ifaces;
3980 long long int next_rebalance, miimon_next_update, lacp_priority;
3981 unsigned long *trunks;
3987 /* Update settings. */
3988 port->updelay = cfg->bond_updelay;
3989 if (port->updelay < 0) {
3992 port->downdelay = cfg->bond_downdelay;
3993 if (port->downdelay < 0) {
3994 port->downdelay = 0;
3996 port->bond_rebalance_interval = atoi(
3997 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3998 if (port->bond_rebalance_interval < 1000) {
3999 port->bond_rebalance_interval = 1000;
4001 next_rebalance = time_msec() + port->bond_rebalance_interval;
4002 if (port->bond_next_rebalance > next_rebalance) {
4003 port->bond_next_rebalance = next_rebalance;
4006 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4009 netdev_monitor_destroy(port->monitor);
4010 port->monitor = NULL;
4012 if (strcmp(detect_mode, "miimon")) {
4013 port->monitor = netdev_monitor_create();
4015 if (strcmp(detect_mode, "carrier")) {
4016 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4017 "defaulting to carrier", port->name, detect_mode);
4021 port->miimon_interval = atoi(
4022 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4023 if (port->miimon_interval < 100) {
4024 port->miimon_interval = 100;
4026 miimon_next_update = time_msec() + port->miimon_interval;
4027 if (port->miimon_next_update > miimon_next_update) {
4028 port->miimon_next_update = miimon_next_update;
4031 if (!port->cfg->bond_mode ||
4032 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4033 port->bond_mode = BM_SLB;
4034 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4035 port->bond_mode = BM_AB;
4036 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4037 port->bond_mode = BM_TCP;
4039 port->bond_mode = BM_SLB;
4040 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4041 port->name, port->cfg->bond_mode,
4042 bond_mode_to_string(port->bond_mode));
4045 /* Add new interfaces and update 'cfg' member of existing ones. */
4046 shash_init(&new_ifaces);
4047 for (i = 0; i < cfg->n_interfaces; i++) {
4048 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4049 struct iface *iface;
4051 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4052 VLOG_WARN("port %s: %s specified twice as port interface",
4053 port->name, if_cfg->name);
4054 iface_set_ofport(if_cfg, -1);
4058 iface = iface_lookup(port->bridge, if_cfg->name);
4060 if (iface->port != port) {
4061 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4063 port->bridge->name, if_cfg->name, iface->port->name);
4066 iface->cfg = if_cfg;
4068 iface = iface_create(port, if_cfg);
4071 /* Determine interface type. The local port always has type
4072 * "internal". Other ports take their type from the database and
4073 * default to "system" if none is specified. */
4074 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4075 : if_cfg->type[0] ? if_cfg->type
4079 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4082 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4083 iface->lacp_priority = UINT16_MAX;
4085 iface->lacp_priority = lacp_priority;
4088 shash_destroy(&new_ifaces);
4090 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4094 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4096 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4097 /* Prefer bondable links if unspecified. */
4098 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4100 port->lacp_priority = lacp_priority;
4103 if (!port->cfg->lacp) {
4104 /* XXX when LACP implementation has been sufficiently tested, enable by
4105 * default and make active on bonded ports. */
4106 lacp_destroy(port->lacp);
4108 } else if (!strcmp(port->cfg->lacp, "off")) {
4109 lacp_destroy(port->lacp);
4111 } else if (!strcmp(port->cfg->lacp, "active")) {
4113 port->lacp = lacp_create();
4115 port->lacp_active = true;
4116 } else if (!strcmp(port->cfg->lacp, "passive")) {
4118 port->lacp = lacp_create();
4120 port->lacp_active = false;
4122 VLOG_WARN("port %s: unknown LACP mode %s",
4123 port->name, port->cfg->lacp);
4124 lacp_destroy(port->lacp);
4131 if (port->n_ifaces < 2) {
4133 if (vlan >= 0 && vlan <= 4095) {
4134 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4139 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4140 * they even work as-is. But they have not been tested. */
4141 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4145 if (port->vlan != vlan) {
4147 bridge_flush(port->bridge);
4150 /* Get trunked VLANs. */
4152 if (vlan < 0 && cfg->n_trunks) {
4155 trunks = bitmap_allocate(4096);
4157 for (i = 0; i < cfg->n_trunks; i++) {
4158 int trunk = cfg->trunks[i];
4160 bitmap_set1(trunks, trunk);
4166 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4167 port->name, cfg->n_trunks);
4169 if (n_errors == cfg->n_trunks) {
4170 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4172 bitmap_free(trunks);
4175 } else if (vlan >= 0 && cfg->n_trunks) {
4176 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4180 ? port->trunks != NULL
4181 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4182 bridge_flush(port->bridge);
4184 bitmap_free(port->trunks);
4185 port->trunks = trunks;
4189 port_destroy(struct port *port)
4192 struct bridge *br = port->bridge;
4196 for (i = 0; i < MAX_MIRRORS; i++) {
4197 struct mirror *m = br->mirrors[i];
4198 if (m && m->out_port == port) {
4203 while (port->n_ifaces > 0) {
4204 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4207 shash_find_and_delete_assert(&br->port_by_name, port->name);
4209 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4210 del->port_idx = port->port_idx;
4212 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4214 netdev_monitor_destroy(port->monitor);
4216 bitmap_free(port->trunks);
4223 static struct port *
4224 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4226 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4227 return iface ? iface->port : NULL;
4230 static struct port *
4231 port_lookup(const struct bridge *br, const char *name)
4233 return shash_find_data(&br->port_by_name, name);
4236 static struct iface *
4237 port_lookup_iface(const struct port *port, const char *name)
4239 struct iface *iface = iface_lookup(port->bridge, name);
4240 return iface && iface->port == port ? iface : NULL;
4244 port_update_lacp(struct port *port)
4249 lacp_configure(port->lacp, port->name,
4250 port->bridge->ea, port->lacp_priority,
4251 port->lacp_active, port->lacp_fast);
4253 for (i = 0; i < port->n_ifaces; i++) {
4254 struct iface *iface = port->ifaces[i];
4255 lacp_slave_register(port->lacp, iface, iface->name,
4256 iface->dp_ifidx, iface->lacp_priority);
4262 port_update_bonding(struct port *port)
4264 if (port->n_ifaces < 2) {
4265 /* Not a bonded port. */
4266 free(port->bond_hash);
4267 port->bond_hash = NULL;
4268 port->bond_fake_iface = false;
4269 port->active_iface = -1;
4270 port->no_ifaces_tag = 0;
4274 if (port->bond_mode != BM_AB && !port->bond_hash) {
4275 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4276 for (i = 0; i <= BOND_MASK; i++) {
4277 struct bond_entry *e = &port->bond_hash[i];
4281 port->bond_next_rebalance
4282 = time_msec() + port->bond_rebalance_interval;
4283 } else if (port->bond_mode == BM_AB) {
4284 free(port->bond_hash);
4285 port->bond_hash = NULL;
4288 if (!port->no_ifaces_tag) {
4289 port->no_ifaces_tag = tag_create_random();
4292 if (port->active_iface < 0) {
4293 bond_choose_active_iface(port);
4296 port->bond_fake_iface = port->cfg->bond_fake_iface;
4297 if (port->bond_fake_iface) {
4298 port->bond_next_fake_iface_update = time_msec();
4304 /* Interface functions. */
4306 static struct iface *
4307 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4309 struct bridge *br = port->bridge;
4310 struct iface *iface;
4311 char *name = if_cfg->name;
4313 iface = xzalloc(sizeof *iface);
4315 iface->port_ifidx = port->n_ifaces;
4316 iface->name = xstrdup(name);
4317 iface->dp_ifidx = -1;
4318 iface->tag = tag_create_random();
4319 iface->delay_expires = LLONG_MAX;
4320 iface->netdev = NULL;
4321 iface->cfg = if_cfg;
4323 shash_add_assert(&br->iface_by_name, iface->name, iface);
4325 if (port->n_ifaces >= port->allocated_ifaces) {
4326 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4327 sizeof *port->ifaces);
4329 port->ifaces[port->n_ifaces++] = iface;
4330 if (port->n_ifaces > 1) {
4331 br->has_bonded_ports = true;
4334 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4342 iface_destroy(struct iface *iface)
4345 struct port *port = iface->port;
4346 struct bridge *br = port->bridge;
4347 bool del_active = port->active_iface == iface->port_ifidx;
4350 if (iface->port->lacp) {
4351 lacp_slave_unregister(iface->port->lacp, iface);
4354 if (port->monitor && iface->netdev) {
4355 netdev_monitor_remove(port->monitor, iface->netdev);
4358 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4360 if (iface->dp_ifidx >= 0) {
4361 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4364 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4365 del->port_ifidx = iface->port_ifidx;
4367 netdev_close(iface->netdev);
4370 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4371 bond_choose_active_iface(port);
4372 bond_send_learning_packets(port);
4378 bridge_flush(port->bridge);
4382 static struct iface *
4383 iface_lookup(const struct bridge *br, const char *name)
4385 return shash_find_data(&br->iface_by_name, name);
4388 static struct iface *
4389 iface_find(const char *name)
4391 const struct bridge *br;
4393 LIST_FOR_EACH (br, node, &all_bridges) {
4394 struct iface *iface = iface_lookup(br, name);
4403 static struct iface *
4404 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4406 struct iface *iface;
4408 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4409 hash_int(dp_ifidx, 0), &br->ifaces) {
4410 if (iface->dp_ifidx == dp_ifidx) {
4417 /* Set Ethernet address of 'iface', if one is specified in the configuration
4420 iface_set_mac(struct iface *iface)
4422 uint8_t ea[ETH_ADDR_LEN];
4424 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4425 if (eth_addr_is_multicast(ea)) {
4426 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4428 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4429 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4430 iface->name, iface->name);
4432 int error = netdev_set_etheraddr(iface->netdev, ea);
4434 VLOG_ERR("interface %s: setting MAC failed (%s)",
4435 iface->name, strerror(error));
4441 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4443 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4446 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4450 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4452 * The value strings in '*shash' are taken directly from values[], not copied,
4453 * so the caller should not modify or free them. */
4455 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4456 struct shash *shash)
4461 for (i = 0; i < n; i++) {
4462 shash_add(shash, keys[i], values[i]);
4466 /* Creates 'keys' and 'values' arrays from 'shash'.
4468 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4469 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4470 * are populated with with strings taken directly from 'shash' and thus have
4471 * the same ownership of the key-value pairs in shash.
4474 shash_to_ovs_idl_map(struct shash *shash,
4475 char ***keys, char ***values, size_t *n)
4479 struct shash_node *sn;
4481 count = shash_count(shash);
4483 k = xmalloc(count * sizeof *k);
4484 v = xmalloc(count * sizeof *v);
4487 SHASH_FOR_EACH(sn, shash) {
4498 struct iface_delete_queues_cbdata {
4499 struct netdev *netdev;
4500 const struct ovsdb_datum *queues;
4504 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4506 union ovsdb_atom atom;
4508 atom.integer = target;
4509 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4513 iface_delete_queues(unsigned int queue_id,
4514 const struct shash *details OVS_UNUSED, void *cbdata_)
4516 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4518 if (!queue_ids_include(cbdata->queues, queue_id)) {
4519 netdev_delete_queue(cbdata->netdev, queue_id);
4524 iface_update_carrier(struct iface *iface)
4526 bool carrier = iface_get_carrier(iface);
4527 if (carrier == iface->up) {
4531 iface->up = carrier;
4532 if (iface->port->lacp) {
4533 lacp_slave_carrier_changed(iface->port->lacp, iface);
4538 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4540 if (!qos || qos->type[0] == '\0') {
4541 netdev_set_qos(iface->netdev, NULL, NULL);
4543 struct iface_delete_queues_cbdata cbdata;
4544 struct shash details;
4547 /* Configure top-level Qos for 'iface'. */
4548 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4549 qos->n_other_config, &details);
4550 netdev_set_qos(iface->netdev, qos->type, &details);
4551 shash_destroy(&details);
4553 /* Deconfigure queues that were deleted. */
4554 cbdata.netdev = iface->netdev;
4555 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4557 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4559 /* Configure queues for 'iface'. */
4560 for (i = 0; i < qos->n_queues; i++) {
4561 const struct ovsrec_queue *queue = qos->value_queues[i];
4562 unsigned int queue_id = qos->key_queues[i];
4564 shash_from_ovs_idl_map(queue->key_other_config,
4565 queue->value_other_config,
4566 queue->n_other_config, &details);
4567 netdev_set_queue(iface->netdev, queue_id, &details);
4568 shash_destroy(&details);
4574 iface_update_cfm(struct iface *iface)
4578 uint16_t *remote_mps;
4579 struct ovsrec_monitor *mon;
4580 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4582 mon = iface->cfg->monitor;
4585 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4589 if (netdev_get_etheraddr(iface->netdev, ea)) {
4590 VLOG_WARN("interface %s: Failed to get ethernet address. "
4591 "Skipping Monitor.", iface->name);
4595 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4596 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4600 cfm.mpid = mon->mpid;
4601 cfm.interval = mon->interval ? *mon->interval : 1000;
4603 memcpy(cfm.eth_src, ea, sizeof cfm.eth_src);
4604 memcpy(cfm.maid, maid, sizeof cfm.maid);
4606 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4607 for(i = 0; i < mon->n_remote_mps; i++) {
4608 remote_mps[i] = mon->remote_mps[i]->mpid;
4611 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4612 &cfm, remote_mps, mon->n_remote_mps);
4616 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4617 * how 'iface''s port is configured.
4619 * Returns true if 'iface' is up, false otherwise. */
4621 iface_get_carrier(const struct iface *iface)
4623 return (iface->port->monitor
4624 ? netdev_get_carrier(iface->netdev)
4625 : netdev_get_miimon(iface->netdev));
4628 /* Port mirroring. */
4630 static struct mirror *
4631 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4635 for (i = 0; i < MAX_MIRRORS; i++) {
4636 struct mirror *m = br->mirrors[i];
4637 if (m && uuid_equals(uuid, &m->uuid)) {
4645 mirror_reconfigure(struct bridge *br)
4647 unsigned long *rspan_vlans;
4650 /* Get rid of deleted mirrors. */
4651 for (i = 0; i < MAX_MIRRORS; i++) {
4652 struct mirror *m = br->mirrors[i];
4654 const struct ovsdb_datum *mc;
4655 union ovsdb_atom atom;
4657 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4658 atom.uuid = br->mirrors[i]->uuid;
4659 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4665 /* Add new mirrors and reconfigure existing ones. */
4666 for (i = 0; i < br->cfg->n_mirrors; i++) {
4667 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4668 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4670 mirror_reconfigure_one(m, cfg);
4672 mirror_create(br, cfg);
4676 /* Update port reserved status. */
4677 for (i = 0; i < br->n_ports; i++) {
4678 br->ports[i]->is_mirror_output_port = false;
4680 for (i = 0; i < MAX_MIRRORS; i++) {
4681 struct mirror *m = br->mirrors[i];
4682 if (m && m->out_port) {
4683 m->out_port->is_mirror_output_port = true;
4687 /* Update flooded vlans (for RSPAN). */
4689 if (br->cfg->n_flood_vlans) {
4690 rspan_vlans = bitmap_allocate(4096);
4692 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4693 int64_t vlan = br->cfg->flood_vlans[i];
4694 if (vlan >= 0 && vlan < 4096) {
4695 bitmap_set1(rspan_vlans, vlan);
4696 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4699 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4704 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4710 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4715 for (i = 0; ; i++) {
4716 if (i >= MAX_MIRRORS) {
4717 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4718 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4721 if (!br->mirrors[i]) {
4726 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4729 br->mirrors[i] = m = xzalloc(sizeof *m);
4732 m->name = xstrdup(cfg->name);
4733 shash_init(&m->src_ports);
4734 shash_init(&m->dst_ports);
4740 mirror_reconfigure_one(m, cfg);
4744 mirror_destroy(struct mirror *m)
4747 struct bridge *br = m->bridge;
4750 for (i = 0; i < br->n_ports; i++) {
4751 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4752 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4755 shash_destroy(&m->src_ports);
4756 shash_destroy(&m->dst_ports);
4759 m->bridge->mirrors[m->idx] = NULL;
4768 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4769 struct shash *names)
4773 for (i = 0; i < n_ports; i++) {
4774 const char *name = ports[i]->name;
4775 if (port_lookup(m->bridge, name)) {
4776 shash_add_once(names, name, NULL);
4778 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4779 "port %s", m->bridge->name, m->name, name);
4785 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4791 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4793 for (i = 0; i < cfg->n_select_vlan; i++) {
4794 int64_t vlan = cfg->select_vlan[i];
4795 if (vlan < 0 || vlan > 4095) {
4796 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4797 m->bridge->name, m->name, vlan);
4799 (*vlans)[n_vlans++] = vlan;
4806 vlan_is_mirrored(const struct mirror *m, int vlan)
4810 for (i = 0; i < m->n_vlans; i++) {
4811 if (m->vlans[i] == vlan) {
4819 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4823 for (i = 0; i < m->n_vlans; i++) {
4824 if (port_trunks_vlan(p, m->vlans[i])) {
4832 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4834 struct shash src_ports, dst_ports;
4835 mirror_mask_t mirror_bit;
4836 struct port *out_port;
4843 if (strcmp(cfg->name, m->name)) {
4845 m->name = xstrdup(cfg->name);
4848 /* Get output port. */
4849 if (cfg->output_port) {
4850 out_port = port_lookup(m->bridge, cfg->output_port->name);
4852 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4853 m->bridge->name, m->name);
4859 if (cfg->output_vlan) {
4860 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4861 "output vlan; ignoring output vlan",
4862 m->bridge->name, m->name);
4864 } else if (cfg->output_vlan) {
4866 out_vlan = *cfg->output_vlan;
4868 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4869 m->bridge->name, m->name);
4874 shash_init(&src_ports);
4875 shash_init(&dst_ports);
4876 if (cfg->select_all) {
4877 for (i = 0; i < m->bridge->n_ports; i++) {
4878 const char *name = m->bridge->ports[i]->name;
4879 shash_add_once(&src_ports, name, NULL);
4880 shash_add_once(&dst_ports, name, NULL);
4885 /* Get ports, and drop duplicates and ports that don't exist. */
4886 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4888 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4891 /* Get all the vlans, and drop duplicate and invalid vlans. */
4892 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4895 /* Update mirror data. */
4896 if (!shash_equal_keys(&m->src_ports, &src_ports)
4897 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4898 || m->n_vlans != n_vlans
4899 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4900 || m->out_port != out_port
4901 || m->out_vlan != out_vlan) {
4902 bridge_flush(m->bridge);
4904 shash_swap(&m->src_ports, &src_ports);
4905 shash_swap(&m->dst_ports, &dst_ports);
4908 m->n_vlans = n_vlans;
4909 m->out_port = out_port;
4910 m->out_vlan = out_vlan;
4913 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4914 for (i = 0; i < m->bridge->n_ports; i++) {
4915 struct port *port = m->bridge->ports[i];
4917 if (shash_find(&m->src_ports, port->name)
4920 ? port_trunks_any_mirrored_vlan(m, port)
4921 : vlan_is_mirrored(m, port->vlan)))) {
4922 port->src_mirrors |= mirror_bit;
4924 port->src_mirrors &= ~mirror_bit;
4927 if (shash_find(&m->dst_ports, port->name)) {
4928 port->dst_mirrors |= mirror_bit;
4930 port->dst_mirrors &= ~mirror_bit;
4935 shash_destroy(&src_ports);
4936 shash_destroy(&dst_ports);