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 char *name; /* Host network device name. */
103 tag_type tag; /* Tag associated with this interface. */
104 long long delay_expires; /* Time after which 'enabled' may change. */
106 /* These members are valid only after bridge_reconfigure() causes them to
108 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
109 int dp_ifidx; /* Index within kernel datapath. */
110 struct netdev *netdev; /* Network device. */
111 bool enabled; /* May be chosen for flows? */
112 bool up; /* Is the interface up? */
113 const char *type; /* Usually same as cfg->type. */
114 const struct ovsrec_interface *cfg;
116 /* LACP information. */
117 uint16_t lacp_priority; /* LACP port priority. */
120 #define BOND_MASK 0xff
122 struct iface *iface; /* Assigned iface, or NULL if none. */
123 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
124 tag_type tag; /* Tag for bond_entry<->iface association. */
128 BM_TCP, /* Transport Layer Load Balance. */
129 BM_SLB, /* Source Load Balance. */
130 BM_AB /* Active Backup. */
133 #define MAX_MIRRORS 32
134 typedef uint32_t mirror_mask_t;
135 #define MIRROR_MASK_C(X) UINT32_C(X)
136 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
138 struct bridge *bridge;
141 struct uuid uuid; /* UUID of this "mirror" record in database. */
143 /* Selection criteria. */
144 struct shash src_ports; /* Name is port name; data is always NULL. */
145 struct shash dst_ports; /* Name is port name; data is always NULL. */
150 struct port *out_port;
154 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
156 struct bridge *bridge;
158 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
159 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
160 * NULL if all VLANs are trunked. */
161 const struct ovsrec_port *cfg;
165 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
166 long long int miimon_interval; /* Miimon status refresh interval. */
167 long long int miimon_next_update; /* Time of next miimon update. */
169 /* An ordinary bridge port has 1 interface.
170 * A bridge port for bonding has at least 2 interfaces. */
171 struct iface **ifaces;
172 size_t n_ifaces, allocated_ifaces;
175 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
176 struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */
177 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
178 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
179 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
180 long long int bond_next_fake_iface_update; /* Time of next update. */
182 /* LACP information. */
183 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
184 bool lacp_active; /* True if LACP is active */
185 bool lacp_fast; /* True if LACP is in fast mode. */
186 uint16_t lacp_priority; /* LACP system priority. */
188 /* SLB specific bonding info. */
189 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
190 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
191 long long int bond_next_rebalance; /* Next rebalancing time. */
193 /* Port mirroring info. */
194 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
195 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
196 bool is_mirror_output_port; /* Does port mirroring send frames here? */
200 struct list node; /* Node in global list of bridges. */
201 char *name; /* User-specified arbitrary name. */
202 struct mac_learning *ml; /* MAC learning table. */
203 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
204 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
205 const struct ovsrec_bridge *cfg;
207 /* OpenFlow switch processing. */
208 struct ofproto *ofproto; /* OpenFlow switch. */
210 /* Kernel datapath information. */
211 struct dpif *dpif; /* Datapath. */
212 struct hmap ifaces; /* Contains "struct iface"s. */
216 size_t n_ports, allocated_ports;
217 struct shash iface_by_name; /* "struct iface"s indexed by name. */
218 struct shash port_by_name; /* "struct port"s indexed by name. */
221 bool has_bonded_ports;
226 /* Port mirroring. */
227 struct mirror *mirrors[MAX_MIRRORS];
230 /* List of all bridges. */
231 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
233 /* OVSDB IDL used to obtain configuration. */
234 static struct ovsdb_idl *idl;
236 /* Each time this timer expires, the bridge fetches systems and interface
237 * statistics and pushes them into the database. */
238 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
239 static long long int stats_timer = LLONG_MIN;
241 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
242 static void bridge_destroy(struct bridge *);
243 static struct bridge *bridge_lookup(const char *name);
244 static unixctl_cb_func bridge_unixctl_dump_flows;
245 static unixctl_cb_func bridge_unixctl_reconnect;
246 static int bridge_run_one(struct bridge *);
247 static size_t bridge_get_controllers(const struct bridge *br,
248 struct ovsrec_controller ***controllersp);
249 static void bridge_reconfigure_one(struct bridge *);
250 static void bridge_reconfigure_remotes(struct bridge *,
251 const struct sockaddr_in *managers,
253 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
254 static void bridge_fetch_dp_ifaces(struct bridge *);
255 static void bridge_flush(struct bridge *);
256 static void bridge_pick_local_hw_addr(struct bridge *,
257 uint8_t ea[ETH_ADDR_LEN],
258 struct iface **hw_addr_iface);
259 static uint64_t bridge_pick_datapath_id(struct bridge *,
260 const uint8_t bridge_ea[ETH_ADDR_LEN],
261 struct iface *hw_addr_iface);
262 static uint64_t dpid_from_hash(const void *, size_t nbytes);
264 static unixctl_cb_func bridge_unixctl_fdb_show;
265 static unixctl_cb_func qos_unixctl_show;
267 static void bond_init(void);
268 static void bond_run(struct port *);
269 static void bond_wait(struct port *);
270 static void bond_rebalance_port(struct port *);
271 static void bond_send_learning_packets(struct port *);
272 static void bond_enable_slave(struct iface *iface, bool enable);
274 static void port_run(struct port *);
275 static void port_wait(struct port *);
276 static struct port *port_create(struct bridge *, const char *name);
277 static void port_reconfigure(struct port *, const struct ovsrec_port *);
278 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
279 static void port_destroy(struct port *);
280 static struct port *port_lookup(const struct bridge *, const char *name);
281 static struct iface *port_lookup_iface(const struct port *, const char *name);
282 static struct port *port_from_dp_ifidx(const struct bridge *,
284 static void port_update_bonding(struct port *);
285 static void port_update_lacp(struct port *);
287 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
288 static void mirror_destroy(struct mirror *);
289 static void mirror_reconfigure(struct bridge *);
290 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
291 static bool vlan_is_mirrored(const struct mirror *, int vlan);
293 static struct iface *iface_create(struct port *port,
294 const struct ovsrec_interface *if_cfg);
295 static void iface_destroy(struct iface *);
296 static struct iface *iface_lookup(const struct bridge *, const char *name);
297 static struct iface *iface_find(const char *name);
298 static struct iface *iface_from_dp_ifidx(const struct bridge *,
300 static void iface_set_mac(struct iface *);
301 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
302 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
303 static void iface_update_cfm(struct iface *);
304 static void iface_refresh_cfm_stats(struct iface *iface);
305 static void iface_update_carrier(struct iface *);
306 static bool iface_get_carrier(const struct iface *);
308 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
310 static void shash_to_ovs_idl_map(struct shash *,
311 char ***keys, char ***values, size_t *n);
313 /* Hooks into ofproto processing. */
314 static struct ofhooks bridge_ofhooks;
316 /* Public functions. */
318 /* Initializes the bridge module, configuring it to obtain its configuration
319 * from an OVSDB server accessed over 'remote', which should be a string in a
320 * form acceptable to ovsdb_idl_create(). */
322 bridge_init(const char *remote)
324 /* Create connection to database. */
325 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
327 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
328 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
329 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
331 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
333 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
334 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
336 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
337 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
338 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
340 /* Register unixctl commands. */
341 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
342 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
343 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
345 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
354 struct bridge *br, *next_br;
356 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
359 ovsdb_idl_destroy(idl);
362 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
363 * but for which the ovs-vswitchd configuration 'cfg' is required. */
365 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
367 static bool already_configured_once;
368 struct svec bridge_names;
369 struct svec dpif_names, dpif_types;
372 /* Only do this once per ovs-vswitchd run. */
373 if (already_configured_once) {
376 already_configured_once = true;
378 stats_timer = time_msec() + STATS_INTERVAL;
380 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
381 svec_init(&bridge_names);
382 for (i = 0; i < cfg->n_bridges; i++) {
383 svec_add(&bridge_names, cfg->bridges[i]->name);
385 svec_sort(&bridge_names);
387 /* Iterate over all system dpifs and delete any of them that do not appear
389 svec_init(&dpif_names);
390 svec_init(&dpif_types);
391 dp_enumerate_types(&dpif_types);
392 for (i = 0; i < dpif_types.n; i++) {
395 dp_enumerate_names(dpif_types.names[i], &dpif_names);
397 /* Delete each dpif whose name is not in 'bridge_names'. */
398 for (j = 0; j < dpif_names.n; j++) {
399 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
403 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
412 svec_destroy(&bridge_names);
413 svec_destroy(&dpif_names);
414 svec_destroy(&dpif_types);
417 /* Callback for iterate_and_prune_ifaces(). */
419 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
421 if (!iface->netdev) {
422 /* We already reported a related error, don't bother duplicating it. */
426 if (iface->dp_ifidx < 0) {
427 VLOG_ERR("%s interface not in %s, dropping",
428 iface->name, dpif_name(br->dpif));
432 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
433 iface->name, iface->dp_ifidx);
437 /* Callback for iterate_and_prune_ifaces(). */
439 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
440 void *aux OVS_UNUSED)
442 /* Set policing attributes. */
443 netdev_set_policing(iface->netdev,
444 iface->cfg->ingress_policing_rate,
445 iface->cfg->ingress_policing_burst);
447 /* Set MAC address of internal interfaces other than the local
449 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
450 iface_set_mac(iface);
456 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
457 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
458 * deletes from 'br' any ports that no longer have any interfaces. */
460 iterate_and_prune_ifaces(struct bridge *br,
461 bool (*cb)(struct bridge *, struct iface *,
467 for (i = 0; i < br->n_ports; ) {
468 struct port *port = br->ports[i];
469 for (j = 0; j < port->n_ifaces; ) {
470 struct iface *iface = port->ifaces[j];
471 if (cb(br, iface, aux)) {
474 iface_set_ofport(iface->cfg, -1);
475 iface_destroy(iface);
479 if (port->n_ifaces) {
482 VLOG_WARN("%s port has no interfaces, dropping", port->name);
488 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
489 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
490 * responsible for freeing '*managersp' (with free()).
492 * You may be asking yourself "why does ovs-vswitchd care?", because
493 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
494 * should not be and in fact is not directly involved in that. But
495 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
496 * it has to tell in-band control where the managers are to enable that.
497 * (Thus, only managers connected in-band are collected.)
500 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
501 struct sockaddr_in **managersp, size_t *n_managersp)
503 struct sockaddr_in *managers = NULL;
504 size_t n_managers = 0;
505 struct shash targets;
508 /* Collect all of the potential targets from the "targets" columns of the
509 * rows pointed to by "manager_options", excluding any that are
511 shash_init(&targets);
512 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
513 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
515 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
516 shash_find_and_delete(&targets, m->target);
518 shash_add_once(&targets, m->target, NULL);
522 /* Now extract the targets' IP addresses. */
523 if (!shash_is_empty(&targets)) {
524 struct shash_node *node;
526 managers = xmalloc(shash_count(&targets) * sizeof *managers);
527 SHASH_FOR_EACH (node, &targets) {
528 const char *target = node->name;
529 struct sockaddr_in *sin = &managers[n_managers];
531 if ((!strncmp(target, "tcp:", 4)
532 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
533 (!strncmp(target, "ssl:", 4)
534 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
539 shash_destroy(&targets);
541 *managersp = managers;
542 *n_managersp = n_managers;
546 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
548 struct shash old_br, new_br;
549 struct shash_node *node;
550 struct bridge *br, *next;
551 struct sockaddr_in *managers;
554 int sflow_bridge_number;
556 COVERAGE_INC(bridge_reconfigure);
558 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
560 /* Collect old and new bridges. */
563 LIST_FOR_EACH (br, node, &all_bridges) {
564 shash_add(&old_br, br->name, br);
566 for (i = 0; i < ovs_cfg->n_bridges; i++) {
567 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
568 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
569 VLOG_WARN("more than one bridge named %s", br_cfg->name);
573 /* Get rid of deleted bridges and add new bridges. */
574 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
575 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
582 SHASH_FOR_EACH (node, &new_br) {
583 const char *br_name = node->name;
584 const struct ovsrec_bridge *br_cfg = node->data;
585 br = shash_find_data(&old_br, br_name);
587 /* If the bridge datapath type has changed, we need to tear it
588 * down and recreate. */
589 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
591 bridge_create(br_cfg);
594 bridge_create(br_cfg);
597 shash_destroy(&old_br);
598 shash_destroy(&new_br);
600 /* Reconfigure all bridges. */
601 LIST_FOR_EACH (br, node, &all_bridges) {
602 bridge_reconfigure_one(br);
605 /* Add and delete ports on all datapaths.
607 * The kernel will reject any attempt to add a given port to a datapath if
608 * that port already belongs to a different datapath, so we must do all
609 * port deletions before any port additions. */
610 LIST_FOR_EACH (br, node, &all_bridges) {
611 struct dpif_port_dump dump;
612 struct shash want_ifaces;
613 struct dpif_port dpif_port;
615 bridge_get_all_ifaces(br, &want_ifaces);
616 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
617 if (!shash_find(&want_ifaces, dpif_port.name)
618 && strcmp(dpif_port.name, br->name)) {
619 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
621 VLOG_WARN("failed to remove %s interface from %s: %s",
622 dpif_port.name, dpif_name(br->dpif),
627 shash_destroy(&want_ifaces);
629 LIST_FOR_EACH (br, node, &all_bridges) {
630 struct shash cur_ifaces, want_ifaces;
631 struct dpif_port_dump dump;
632 struct dpif_port dpif_port;
634 /* Get the set of interfaces currently in this datapath. */
635 shash_init(&cur_ifaces);
636 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
637 struct dpif_port *port_info = xmalloc(sizeof *port_info);
638 dpif_port_clone(port_info, &dpif_port);
639 shash_add(&cur_ifaces, dpif_port.name, port_info);
642 /* Get the set of interfaces we want on this datapath. */
643 bridge_get_all_ifaces(br, &want_ifaces);
645 hmap_clear(&br->ifaces);
646 SHASH_FOR_EACH (node, &want_ifaces) {
647 const char *if_name = node->name;
648 struct iface *iface = node->data;
649 struct dpif_port *dpif_port;
653 type = iface ? iface->type : "internal";
654 dpif_port = shash_find_data(&cur_ifaces, if_name);
656 /* If we have a port or a netdev already, and it's not the type we
657 * want, then delete the port (if any) and close the netdev (if
659 if ((dpif_port && strcmp(dpif_port->type, type))
660 || (iface && iface->netdev
661 && strcmp(type, netdev_get_type(iface->netdev)))) {
663 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
670 netdev_close(iface->netdev);
671 iface->netdev = NULL;
675 /* If the port doesn't exist or we don't have the netdev open,
676 * we need to do more work. */
677 if (!dpif_port || (iface && !iface->netdev)) {
678 struct netdev_options options;
679 struct netdev *netdev;
682 /* First open the network device. */
683 options.name = if_name;
685 options.args = &args;
686 options.ethertype = NETDEV_ETH_TYPE_NONE;
690 shash_from_ovs_idl_map(iface->cfg->key_options,
691 iface->cfg->value_options,
692 iface->cfg->n_options, &args);
694 error = netdev_open(&options, &netdev);
695 shash_destroy(&args);
698 VLOG_WARN("could not open network device %s (%s)",
699 if_name, strerror(error));
703 /* Then add the port if we haven't already. */
705 error = dpif_port_add(br->dpif, netdev, NULL);
707 netdev_close(netdev);
708 if (error == EFBIG) {
709 VLOG_ERR("ran out of valid port numbers on %s",
710 dpif_name(br->dpif));
713 VLOG_WARN("failed to add %s interface to %s: %s",
714 if_name, dpif_name(br->dpif),
721 /* Update 'iface'. */
723 iface->netdev = netdev;
724 iface->enabled = iface_get_carrier(iface);
725 iface->up = iface->enabled;
727 } else if (iface && iface->netdev) {
731 shash_from_ovs_idl_map(iface->cfg->key_options,
732 iface->cfg->value_options,
733 iface->cfg->n_options, &args);
734 netdev_set_config(iface->netdev, &args);
735 shash_destroy(&args);
738 shash_destroy(&want_ifaces);
740 SHASH_FOR_EACH (node, &cur_ifaces) {
741 struct dpif_port *port_info = node->data;
742 dpif_port_destroy(port_info);
745 shash_destroy(&cur_ifaces);
747 sflow_bridge_number = 0;
748 LIST_FOR_EACH (br, node, &all_bridges) {
751 struct iface *local_iface;
752 struct iface *hw_addr_iface;
755 bridge_fetch_dp_ifaces(br);
757 iterate_and_prune_ifaces(br, check_iface, NULL);
759 /* Pick local port hardware address, datapath ID. */
760 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
761 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
763 int error = netdev_set_etheraddr(local_iface->netdev, ea);
765 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
766 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
767 "Ethernet address: %s",
768 br->name, strerror(error));
771 memcpy(br->ea, ea, ETH_ADDR_LEN);
773 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
774 ofproto_set_datapath_id(br->ofproto, dpid);
776 dpid_string = xasprintf("%016"PRIx64, dpid);
777 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
780 /* Set NetFlow configuration on this bridge. */
781 if (br->cfg->netflow) {
782 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
783 struct netflow_options opts;
785 memset(&opts, 0, sizeof opts);
787 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
788 if (nf_cfg->engine_type) {
789 opts.engine_type = *nf_cfg->engine_type;
791 if (nf_cfg->engine_id) {
792 opts.engine_id = *nf_cfg->engine_id;
795 opts.active_timeout = nf_cfg->active_timeout;
796 if (!opts.active_timeout) {
797 opts.active_timeout = -1;
798 } else if (opts.active_timeout < 0) {
799 VLOG_WARN("bridge %s: active timeout interval set to negative "
800 "value, using default instead (%d seconds)", br->name,
801 NF_ACTIVE_TIMEOUT_DEFAULT);
802 opts.active_timeout = -1;
805 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
806 if (opts.add_id_to_iface) {
807 if (opts.engine_id > 0x7f) {
808 VLOG_WARN("bridge %s: netflow port mangling may conflict "
809 "with another vswitch, choose an engine id less "
810 "than 128", br->name);
812 if (br->n_ports > 508) {
813 VLOG_WARN("bridge %s: netflow port mangling will conflict "
814 "with another port when more than 508 ports are "
819 opts.collectors.n = nf_cfg->n_targets;
820 opts.collectors.names = nf_cfg->targets;
821 if (ofproto_set_netflow(br->ofproto, &opts)) {
822 VLOG_ERR("bridge %s: problem setting netflow collectors",
826 ofproto_set_netflow(br->ofproto, NULL);
829 /* Set sFlow configuration on this bridge. */
830 if (br->cfg->sflow) {
831 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
832 struct ovsrec_controller **controllers;
833 struct ofproto_sflow_options oso;
834 size_t n_controllers;
836 memset(&oso, 0, sizeof oso);
838 oso.targets.n = sflow_cfg->n_targets;
839 oso.targets.names = sflow_cfg->targets;
841 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
842 if (sflow_cfg->sampling) {
843 oso.sampling_rate = *sflow_cfg->sampling;
846 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
847 if (sflow_cfg->polling) {
848 oso.polling_interval = *sflow_cfg->polling;
851 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
852 if (sflow_cfg->header) {
853 oso.header_len = *sflow_cfg->header;
856 oso.sub_id = sflow_bridge_number++;
857 oso.agent_device = sflow_cfg->agent;
859 oso.control_ip = NULL;
860 n_controllers = bridge_get_controllers(br, &controllers);
861 for (i = 0; i < n_controllers; i++) {
862 if (controllers[i]->local_ip) {
863 oso.control_ip = controllers[i]->local_ip;
867 ofproto_set_sflow(br->ofproto, &oso);
869 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
871 ofproto_set_sflow(br->ofproto, NULL);
874 /* Update the controller and related settings. It would be more
875 * straightforward to call this from bridge_reconfigure_one(), but we
876 * can't do it there for two reasons. First, and most importantly, at
877 * that point we don't know the dp_ifidx of any interfaces that have
878 * been added to the bridge (because we haven't actually added them to
879 * the datapath). Second, at that point we haven't set the datapath ID
880 * yet; when a controller is configured, resetting the datapath ID will
881 * immediately disconnect from the controller, so it's better to set
882 * the datapath ID before the controller. */
883 bridge_reconfigure_remotes(br, managers, n_managers);
885 LIST_FOR_EACH (br, node, &all_bridges) {
886 for (i = 0; i < br->n_ports; i++) {
887 struct port *port = br->ports[i];
891 for (j = 0; j < port->n_ifaces; j++) {
892 netdev_monitor_add(port->monitor, port->ifaces[j]->netdev);
895 port->miimon_next_update = 0;
898 port_update_lacp(port);
899 port_update_bonding(port);
901 for (j = 0; j < port->n_ifaces; j++) {
902 iface_update_qos(port->ifaces[j], port->cfg->qos);
906 LIST_FOR_EACH (br, node, &all_bridges) {
907 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
910 LIST_FOR_EACH (br, node, &all_bridges) {
912 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
913 iface_update_cfm(iface);
919 /* ovs-vswitchd has completed initialization, so allow the process that
920 * forked us to exit successfully. */
921 daemonize_complete();
925 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
926 const struct ovsdb_idl_column *column,
929 const struct ovsdb_datum *datum;
930 union ovsdb_atom atom;
933 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
934 atom.string = (char *) key;
935 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
936 return idx == UINT_MAX ? NULL : datum->values[idx].string;
940 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
942 return get_ovsrec_key_value(&br_cfg->header_,
943 &ovsrec_bridge_col_other_config, key);
947 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
948 struct iface **hw_addr_iface)
954 *hw_addr_iface = NULL;
956 /* Did the user request a particular MAC? */
957 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
958 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
959 if (eth_addr_is_multicast(ea)) {
960 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
961 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
962 } else if (eth_addr_is_zero(ea)) {
963 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
969 /* Otherwise choose the minimum non-local MAC address among all of the
971 memset(ea, 0xff, ETH_ADDR_LEN);
972 for (i = 0; i < br->n_ports; i++) {
973 struct port *port = br->ports[i];
974 uint8_t iface_ea[ETH_ADDR_LEN];
977 /* Mirror output ports don't participate. */
978 if (port->is_mirror_output_port) {
982 /* Choose the MAC address to represent the port. */
983 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
984 /* Find the interface with this Ethernet address (if any) so that
985 * we can provide the correct devname to the caller. */
987 for (j = 0; j < port->n_ifaces; j++) {
988 struct iface *candidate = port->ifaces[j];
989 uint8_t candidate_ea[ETH_ADDR_LEN];
990 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
991 && eth_addr_equals(iface_ea, candidate_ea)) {
996 /* Choose the interface whose MAC address will represent the port.
997 * The Linux kernel bonding code always chooses the MAC address of
998 * the first slave added to a bond, and the Fedora networking
999 * scripts always add slaves to a bond in alphabetical order, so
1000 * for compatibility we choose the interface with the name that is
1001 * first in alphabetical order. */
1002 iface = port->ifaces[0];
1003 for (j = 1; j < port->n_ifaces; j++) {
1004 struct iface *candidate = port->ifaces[j];
1005 if (strcmp(candidate->name, iface->name) < 0) {
1010 /* The local port doesn't count (since we're trying to choose its
1011 * MAC address anyway). */
1012 if (iface->dp_ifidx == ODPP_LOCAL) {
1017 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1019 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1020 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1021 iface->name, strerror(error));
1026 /* Compare against our current choice. */
1027 if (!eth_addr_is_multicast(iface_ea) &&
1028 !eth_addr_is_local(iface_ea) &&
1029 !eth_addr_is_reserved(iface_ea) &&
1030 !eth_addr_is_zero(iface_ea) &&
1031 eth_addr_compare_3way(iface_ea, ea) < 0)
1033 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1034 *hw_addr_iface = iface;
1037 if (eth_addr_is_multicast(ea)) {
1038 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1039 *hw_addr_iface = NULL;
1040 VLOG_WARN("bridge %s: using default bridge Ethernet "
1041 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1043 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1044 br->name, ETH_ADDR_ARGS(ea));
1048 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1049 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1050 * an interface on 'br', then that interface must be passed in as
1051 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1052 * 'hw_addr_iface' must be passed in as a null pointer. */
1054 bridge_pick_datapath_id(struct bridge *br,
1055 const uint8_t bridge_ea[ETH_ADDR_LEN],
1056 struct iface *hw_addr_iface)
1059 * The procedure for choosing a bridge MAC address will, in the most
1060 * ordinary case, also choose a unique MAC that we can use as a datapath
1061 * ID. In some special cases, though, multiple bridges will end up with
1062 * the same MAC address. This is OK for the bridges, but it will confuse
1063 * the OpenFlow controller, because each datapath needs a unique datapath
1066 * Datapath IDs must be unique. It is also very desirable that they be
1067 * stable from one run to the next, so that policy set on a datapath
1070 const char *datapath_id;
1073 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1074 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1078 if (hw_addr_iface) {
1080 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1082 * A bridge whose MAC address is taken from a VLAN network device
1083 * (that is, a network device created with vconfig(8) or similar
1084 * tool) will have the same MAC address as a bridge on the VLAN
1085 * device's physical network device.
1087 * Handle this case by hashing the physical network device MAC
1088 * along with the VLAN identifier.
1090 uint8_t buf[ETH_ADDR_LEN + 2];
1091 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1092 buf[ETH_ADDR_LEN] = vlan >> 8;
1093 buf[ETH_ADDR_LEN + 1] = vlan;
1094 return dpid_from_hash(buf, sizeof buf);
1097 * Assume that this bridge's MAC address is unique, since it
1098 * doesn't fit any of the cases we handle specially.
1103 * A purely internal bridge, that is, one that has no non-virtual
1104 * network devices on it at all, is more difficult because it has no
1105 * natural unique identifier at all.
1107 * When the host is a XenServer, we handle this case by hashing the
1108 * host's UUID with the name of the bridge. Names of bridges are
1109 * persistent across XenServer reboots, although they can be reused if
1110 * an internal network is destroyed and then a new one is later
1111 * created, so this is fairly effective.
1113 * When the host is not a XenServer, we punt by using a random MAC
1114 * address on each run.
1116 const char *host_uuid = xenserver_get_host_uuid();
1118 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1119 dpid = dpid_from_hash(combined, strlen(combined));
1125 return eth_addr_to_uint64(bridge_ea);
1129 dpid_from_hash(const void *data, size_t n)
1131 uint8_t hash[SHA1_DIGEST_SIZE];
1133 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1134 sha1_bytes(data, n, hash);
1135 eth_addr_mark_random(hash);
1136 return eth_addr_to_uint64(hash);
1140 iface_refresh_status(struct iface *iface)
1144 enum netdev_flags flags;
1153 if (!netdev_get_status(iface->netdev, &sh)) {
1155 char **keys, **values;
1157 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1158 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1163 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1166 shash_destroy_free_data(&sh);
1168 error = netdev_get_flags(iface->netdev, &flags);
1170 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1173 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1176 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1178 ovsrec_interface_set_duplex(iface->cfg,
1179 netdev_features_is_full_duplex(current)
1181 /* warning: uint64_t -> int64_t conversion */
1182 bps = netdev_features_to_bps(current);
1183 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1186 ovsrec_interface_set_duplex(iface->cfg, NULL);
1187 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1191 ovsrec_interface_set_link_state(iface->cfg,
1192 iface_get_carrier(iface) ? "up" : "down");
1194 error = netdev_get_mtu(iface->netdev, &mtu);
1195 if (!error && mtu != INT_MAX) {
1197 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1200 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1205 iface_refresh_cfm_stats(struct iface *iface)
1207 const struct ovsrec_monitor *mon;
1208 const struct cfm *cfm;
1211 mon = iface->cfg->monitor;
1212 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1218 for (i = 0; i < mon->n_remote_mps; i++) {
1219 const struct ovsrec_maintenance_point *mp;
1220 const struct remote_mp *rmp;
1222 mp = mon->remote_mps[i];
1223 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1225 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1228 if (hmap_is_empty(&cfm->x_remote_mps)) {
1229 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1232 struct remote_mp *rmp;
1233 int64_t *x_remote_mps;
1235 length = hmap_count(&cfm->x_remote_mps);
1236 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1239 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1240 x_remote_mps[i++] = rmp->mpid;
1243 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1247 if (hmap_is_empty(&cfm->x_remote_maids)) {
1248 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1251 char **x_remote_maids;
1252 struct remote_maid *rmaid;
1254 length = hmap_count(&cfm->x_remote_maids);
1255 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1258 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1261 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1263 for (j = 0; j < CCM_MAID_LEN; j++) {
1264 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1269 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1271 for (i = 0; i < length; i++) {
1272 free(x_remote_maids[i]);
1274 free(x_remote_maids);
1277 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1281 iface_refresh_stats(struct iface *iface)
1287 static const struct iface_stat iface_stats[] = {
1288 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1289 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1290 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1291 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1292 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1293 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1294 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1295 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1296 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1297 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1298 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1299 { "collisions", offsetof(struct netdev_stats, collisions) },
1301 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1302 const struct iface_stat *s;
1304 char *keys[N_STATS];
1305 int64_t values[N_STATS];
1308 struct netdev_stats stats;
1310 /* Intentionally ignore return value, since errors will set 'stats' to
1311 * all-1s, and we will deal with that correctly below. */
1312 netdev_get_stats(iface->netdev, &stats);
1315 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1316 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1317 if (value != UINT64_MAX) {
1324 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1328 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1330 struct ovsdb_datum datum;
1334 get_system_stats(&stats);
1336 ovsdb_datum_from_shash(&datum, &stats);
1337 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1341 static inline const char *
1342 nx_role_to_str(enum nx_role role)
1347 case NX_ROLE_MASTER:
1352 return "*** INVALID ROLE ***";
1357 bridge_refresh_controller_status(const struct bridge *br)
1360 const struct ovsrec_controller *cfg;
1362 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1364 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1365 struct ofproto_controller_info *cinfo =
1366 shash_find_data(&info, cfg->target);
1369 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1370 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1371 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1372 (char **) cinfo->pairs.values,
1375 ovsrec_controller_set_is_connected(cfg, false);
1376 ovsrec_controller_set_role(cfg, NULL);
1377 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1381 ofproto_free_ofproto_controller_info(&info);
1387 const struct ovsrec_open_vswitch *cfg;
1389 bool datapath_destroyed;
1390 bool database_changed;
1393 /* Let each bridge do the work that it needs to do. */
1394 datapath_destroyed = false;
1395 LIST_FOR_EACH (br, node, &all_bridges) {
1396 int error = bridge_run_one(br);
1398 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1399 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1400 "forcing reconfiguration", br->name);
1401 datapath_destroyed = true;
1405 /* (Re)configure if necessary. */
1406 database_changed = ovsdb_idl_run(idl);
1407 cfg = ovsrec_open_vswitch_first(idl);
1409 /* Re-configure SSL. We do this on every trip through the main loop,
1410 * instead of just when the database changes, because the contents of the
1411 * key and certificate files can change without the database changing.
1413 * We do this before bridge_reconfigure() because that function might
1414 * initiate SSL connections and thus requires SSL to be configured. */
1415 if (cfg && cfg->ssl) {
1416 const struct ovsrec_ssl *ssl = cfg->ssl;
1418 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1419 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1422 if (database_changed || datapath_destroyed) {
1424 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1426 bridge_configure_once(cfg);
1427 bridge_reconfigure(cfg);
1429 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1430 ovsdb_idl_txn_commit(txn);
1431 ovsdb_idl_txn_destroy(txn); /* XXX */
1433 /* We still need to reconfigure to avoid dangling pointers to
1434 * now-destroyed ovsrec structures inside bridge data. */
1435 static const struct ovsrec_open_vswitch null_cfg;
1437 bridge_reconfigure(&null_cfg);
1441 /* Refresh system and interface stats if necessary. */
1442 if (time_msec() >= stats_timer) {
1444 struct ovsdb_idl_txn *txn;
1446 txn = ovsdb_idl_txn_create(idl);
1447 LIST_FOR_EACH (br, node, &all_bridges) {
1450 for (i = 0; i < br->n_ports; i++) {
1451 struct port *port = br->ports[i];
1454 for (j = 0; j < port->n_ifaces; j++) {
1455 struct iface *iface = port->ifaces[j];
1456 iface_refresh_stats(iface);
1457 iface_refresh_cfm_stats(iface);
1458 iface_refresh_status(iface);
1461 bridge_refresh_controller_status(br);
1463 refresh_system_stats(cfg);
1464 ovsdb_idl_txn_commit(txn);
1465 ovsdb_idl_txn_destroy(txn); /* XXX */
1468 stats_timer = time_msec() + STATS_INTERVAL;
1477 LIST_FOR_EACH (br, node, &all_bridges) {
1480 ofproto_wait(br->ofproto);
1481 if (ofproto_has_primary_controller(br->ofproto)) {
1485 mac_learning_wait(br->ml);
1487 for (i = 0; i < br->n_ports; i++) {
1488 port_wait(br->ports[i]);
1491 ovsdb_idl_wait(idl);
1492 poll_timer_wait_until(stats_timer);
1495 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1496 * configuration changes. */
1498 bridge_flush(struct bridge *br)
1500 COVERAGE_INC(bridge_flush);
1502 mac_learning_flush(br->ml);
1505 /* Bridge unixctl user interface functions. */
1507 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1508 const char *args, void *aux OVS_UNUSED)
1510 struct ds ds = DS_EMPTY_INITIALIZER;
1511 const struct bridge *br;
1512 const struct mac_entry *e;
1514 br = bridge_lookup(args);
1516 unixctl_command_reply(conn, 501, "no such bridge");
1520 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1521 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1522 if (e->port < 0 || e->port >= br->n_ports) {
1525 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1526 br->ports[e->port]->ifaces[0]->dp_ifidx,
1527 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1529 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1533 /* QoS unixctl user interface functions. */
1535 struct qos_unixctl_show_cbdata {
1537 struct iface *iface;
1541 qos_unixctl_show_cb(unsigned int queue_id,
1542 const struct shash *details,
1545 struct qos_unixctl_show_cbdata *data = aux;
1546 struct ds *ds = data->ds;
1547 struct iface *iface = data->iface;
1548 struct netdev_queue_stats stats;
1549 struct shash_node *node;
1552 ds_put_cstr(ds, "\n");
1554 ds_put_format(ds, "Queue %u:\n", queue_id);
1556 ds_put_cstr(ds, "Default:\n");
1559 SHASH_FOR_EACH (node, details) {
1560 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1563 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1565 if (stats.tx_packets != UINT64_MAX) {
1566 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1569 if (stats.tx_bytes != UINT64_MAX) {
1570 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1573 if (stats.tx_errors != UINT64_MAX) {
1574 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1577 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1578 queue_id, strerror(error));
1583 qos_unixctl_show(struct unixctl_conn *conn,
1584 const char *args, void *aux OVS_UNUSED)
1586 struct ds ds = DS_EMPTY_INITIALIZER;
1587 struct shash sh = SHASH_INITIALIZER(&sh);
1588 struct iface *iface;
1590 struct shash_node *node;
1591 struct qos_unixctl_show_cbdata data;
1594 iface = iface_find(args);
1596 unixctl_command_reply(conn, 501, "no such interface");
1600 netdev_get_qos(iface->netdev, &type, &sh);
1602 if (*type != '\0') {
1603 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1605 SHASH_FOR_EACH (node, &sh) {
1606 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1611 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1614 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1616 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1618 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1619 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1622 shash_destroy_free_data(&sh);
1626 /* Bridge reconfiguration functions. */
1627 static struct bridge *
1628 bridge_create(const struct ovsrec_bridge *br_cfg)
1633 assert(!bridge_lookup(br_cfg->name));
1634 br = xzalloc(sizeof *br);
1636 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1642 dpif_flow_flush(br->dpif);
1644 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1647 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1649 dpif_delete(br->dpif);
1650 dpif_close(br->dpif);
1655 br->name = xstrdup(br_cfg->name);
1657 br->ml = mac_learning_create();
1658 eth_addr_nicira_random(br->default_ea);
1660 hmap_init(&br->ifaces);
1662 shash_init(&br->port_by_name);
1663 shash_init(&br->iface_by_name);
1667 list_push_back(&all_bridges, &br->node);
1669 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1675 bridge_destroy(struct bridge *br)
1680 while (br->n_ports > 0) {
1681 port_destroy(br->ports[br->n_ports - 1]);
1683 list_remove(&br->node);
1684 error = dpif_delete(br->dpif);
1685 if (error && error != ENOENT) {
1686 VLOG_ERR("failed to delete %s: %s",
1687 dpif_name(br->dpif), strerror(error));
1689 dpif_close(br->dpif);
1690 ofproto_destroy(br->ofproto);
1691 mac_learning_destroy(br->ml);
1692 hmap_destroy(&br->ifaces);
1693 shash_destroy(&br->port_by_name);
1694 shash_destroy(&br->iface_by_name);
1701 static struct bridge *
1702 bridge_lookup(const char *name)
1706 LIST_FOR_EACH (br, node, &all_bridges) {
1707 if (!strcmp(br->name, name)) {
1714 /* Handle requests for a listing of all flows known by the OpenFlow
1715 * stack, including those normally hidden. */
1717 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1718 const char *args, void *aux OVS_UNUSED)
1723 br = bridge_lookup(args);
1725 unixctl_command_reply(conn, 501, "Unknown bridge");
1730 ofproto_get_all_flows(br->ofproto, &results);
1732 unixctl_command_reply(conn, 200, ds_cstr(&results));
1733 ds_destroy(&results);
1736 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1737 * connections and reconnect. If BRIDGE is not specified, then all bridges
1738 * drop their controller connections and reconnect. */
1740 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1741 const char *args, void *aux OVS_UNUSED)
1744 if (args[0] != '\0') {
1745 br = bridge_lookup(args);
1747 unixctl_command_reply(conn, 501, "Unknown bridge");
1750 ofproto_reconnect_controllers(br->ofproto);
1752 LIST_FOR_EACH (br, node, &all_bridges) {
1753 ofproto_reconnect_controllers(br->ofproto);
1756 unixctl_command_reply(conn, 200, NULL);
1760 bridge_run_one(struct bridge *br)
1765 error = ofproto_run1(br->ofproto);
1770 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1772 for (i = 0; i < br->n_ports; i++) {
1773 port_run(br->ports[i]);
1776 error = ofproto_run2(br->ofproto, br->flush);
1783 bridge_get_controllers(const struct bridge *br,
1784 struct ovsrec_controller ***controllersp)
1786 struct ovsrec_controller **controllers;
1787 size_t n_controllers;
1789 controllers = br->cfg->controller;
1790 n_controllers = br->cfg->n_controller;
1792 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1798 *controllersp = controllers;
1800 return n_controllers;
1804 bridge_reconfigure_one(struct bridge *br)
1806 struct shash old_ports, new_ports;
1807 struct svec snoops, old_snoops;
1808 struct shash_node *node;
1809 enum ofproto_fail_mode fail_mode;
1812 /* Collect old ports. */
1813 shash_init(&old_ports);
1814 for (i = 0; i < br->n_ports; i++) {
1815 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1818 /* Collect new ports. */
1819 shash_init(&new_ports);
1820 for (i = 0; i < br->cfg->n_ports; i++) {
1821 const char *name = br->cfg->ports[i]->name;
1822 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1823 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1828 /* If we have a controller, then we need a local port. Complain if the
1829 * user didn't specify one.
1831 * XXX perhaps we should synthesize a port ourselves in this case. */
1832 if (bridge_get_controllers(br, NULL)) {
1833 char local_name[IF_NAMESIZE];
1836 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1837 local_name, sizeof local_name);
1838 if (!error && !shash_find(&new_ports, local_name)) {
1839 VLOG_WARN("bridge %s: controller specified but no local port "
1840 "(port named %s) defined",
1841 br->name, local_name);
1845 /* Get rid of deleted ports.
1846 * Get rid of deleted interfaces on ports that still exist. */
1847 SHASH_FOR_EACH (node, &old_ports) {
1848 struct port *port = node->data;
1849 const struct ovsrec_port *port_cfg;
1851 port_cfg = shash_find_data(&new_ports, node->name);
1855 port_del_ifaces(port, port_cfg);
1859 /* Create new ports.
1860 * Add new interfaces to existing ports.
1861 * Reconfigure existing ports. */
1862 SHASH_FOR_EACH (node, &new_ports) {
1863 struct port *port = shash_find_data(&old_ports, node->name);
1865 port = port_create(br, node->name);
1868 port_reconfigure(port, node->data);
1869 if (!port->n_ifaces) {
1870 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1871 br->name, port->name);
1875 shash_destroy(&old_ports);
1876 shash_destroy(&new_ports);
1878 /* Set the fail-mode */
1879 fail_mode = !br->cfg->fail_mode
1880 || !strcmp(br->cfg->fail_mode, "standalone")
1881 ? OFPROTO_FAIL_STANDALONE
1882 : OFPROTO_FAIL_SECURE;
1883 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1884 && !ofproto_has_primary_controller(br->ofproto)) {
1885 ofproto_flush_flows(br->ofproto);
1887 ofproto_set_fail_mode(br->ofproto, fail_mode);
1889 /* Delete all flows if we're switching from connected to standalone or vice
1890 * versa. (XXX Should we delete all flows if we are switching from one
1891 * controller to another?) */
1893 /* Configure OpenFlow controller connection snooping. */
1895 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1896 ovs_rundir(), br->name));
1897 svec_init(&old_snoops);
1898 ofproto_get_snoops(br->ofproto, &old_snoops);
1899 if (!svec_equal(&snoops, &old_snoops)) {
1900 ofproto_set_snoops(br->ofproto, &snoops);
1902 svec_destroy(&snoops);
1903 svec_destroy(&old_snoops);
1905 mirror_reconfigure(br);
1908 /* Initializes 'oc' appropriately as a management service controller for
1911 * The caller must free oc->target when it is no longer needed. */
1913 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1914 struct ofproto_controller *oc)
1916 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1917 oc->max_backoff = 0;
1918 oc->probe_interval = 60;
1919 oc->band = OFPROTO_OUT_OF_BAND;
1921 oc->burst_limit = 0;
1924 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1926 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1927 struct ofproto_controller *oc)
1929 oc->target = c->target;
1930 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1931 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1932 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1933 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1934 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1935 oc->burst_limit = (c->controller_burst_limit
1936 ? *c->controller_burst_limit : 0);
1939 /* Configures the IP stack for 'br''s local interface properly according to the
1940 * configuration in 'c'. */
1942 bridge_configure_local_iface_netdev(struct bridge *br,
1943 struct ovsrec_controller *c)
1945 struct netdev *netdev;
1946 struct in_addr mask, gateway;
1948 struct iface *local_iface;
1951 /* If there's no local interface or no IP address, give up. */
1952 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
1953 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1957 /* Bring up the local interface. */
1958 netdev = local_iface->netdev;
1959 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1961 /* Configure the IP address and netmask. */
1962 if (!c->local_netmask
1963 || !inet_aton(c->local_netmask, &mask)
1965 mask.s_addr = guess_netmask(ip.s_addr);
1967 if (!netdev_set_in4(netdev, ip, mask)) {
1968 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1969 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1972 /* Configure the default gateway. */
1973 if (c->local_gateway
1974 && inet_aton(c->local_gateway, &gateway)
1975 && gateway.s_addr) {
1976 if (!netdev_add_router(netdev, gateway)) {
1977 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1978 br->name, IP_ARGS(&gateway.s_addr));
1984 bridge_reconfigure_remotes(struct bridge *br,
1985 const struct sockaddr_in *managers,
1988 const char *disable_ib_str, *queue_id_str;
1989 bool disable_in_band = false;
1992 struct ovsrec_controller **controllers;
1993 size_t n_controllers;
1996 struct ofproto_controller *ocs;
2000 /* Check if we should disable in-band control on this bridge. */
2001 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2002 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2003 disable_in_band = true;
2006 /* Set OpenFlow queue ID for in-band control. */
2007 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2008 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2009 ofproto_set_in_band_queue(br->ofproto, queue_id);
2011 if (disable_in_band) {
2012 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2014 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2016 had_primary = ofproto_has_primary_controller(br->ofproto);
2018 n_controllers = bridge_get_controllers(br, &controllers);
2020 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2023 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2024 for (i = 0; i < n_controllers; i++) {
2025 struct ovsrec_controller *c = controllers[i];
2027 if (!strncmp(c->target, "punix:", 6)
2028 || !strncmp(c->target, "unix:", 5)) {
2029 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2031 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2032 * domain sockets and overwriting arbitrary local files. */
2033 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2034 "\"%s\" due to possibility for remote exploit",
2035 dpif_name(br->dpif), c->target);
2039 bridge_configure_local_iface_netdev(br, c);
2040 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2041 if (disable_in_band) {
2042 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2047 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2048 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2051 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2052 ofproto_flush_flows(br->ofproto);
2055 /* If there are no controllers and the bridge is in standalone
2056 * mode, set up a flow that matches every packet and directs
2057 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2058 * switch is in secure mode and we won't pass any traffic until
2059 * a controller has been defined and it tells us to do so. */
2061 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2062 union ofp_action action;
2063 struct cls_rule rule;
2065 memset(&action, 0, sizeof action);
2066 action.type = htons(OFPAT_OUTPUT);
2067 action.output.len = htons(sizeof action);
2068 action.output.port = htons(OFPP_NORMAL);
2069 cls_rule_init_catchall(&rule, 0);
2070 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2075 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2080 for (i = 0; i < br->n_ports; i++) {
2081 struct port *port = br->ports[i];
2082 for (j = 0; j < port->n_ifaces; j++) {
2083 struct iface *iface = port->ifaces[j];
2084 shash_add_once(ifaces, iface->name, iface);
2086 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2087 shash_add_once(ifaces, port->name, NULL);
2092 /* For robustness, in case the administrator moves around datapath ports behind
2093 * our back, we re-check all the datapath port numbers here.
2095 * This function will set the 'dp_ifidx' members of interfaces that have
2096 * disappeared to -1, so only call this function from a context where those
2097 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2098 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2099 * datapath, which doesn't support UINT16_MAX+1 ports. */
2101 bridge_fetch_dp_ifaces(struct bridge *br)
2103 struct dpif_port_dump dump;
2104 struct dpif_port dpif_port;
2107 /* Reset all interface numbers. */
2108 for (i = 0; i < br->n_ports; i++) {
2109 struct port *port = br->ports[i];
2110 for (j = 0; j < port->n_ifaces; j++) {
2111 struct iface *iface = port->ifaces[j];
2112 iface->dp_ifidx = -1;
2115 hmap_clear(&br->ifaces);
2117 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2118 struct iface *iface = iface_lookup(br, dpif_port.name);
2120 if (iface->dp_ifidx >= 0) {
2121 VLOG_WARN("%s reported interface %s twice",
2122 dpif_name(br->dpif), dpif_port.name);
2123 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2124 VLOG_WARN("%s reported interface %"PRIu16" twice",
2125 dpif_name(br->dpif), dpif_port.port_no);
2127 iface->dp_ifidx = dpif_port.port_no;
2128 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2129 hash_int(iface->dp_ifidx, 0));
2132 iface_set_ofport(iface->cfg,
2133 (iface->dp_ifidx >= 0
2134 ? odp_port_to_ofp_port(iface->dp_ifidx)
2140 /* Bridge packet processing functions. */
2143 bond_is_tcp_hash(const struct port *port)
2145 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2149 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2151 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2154 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2156 struct flow hash_flow;
2158 memcpy(&hash_flow, flow, sizeof hash_flow);
2159 hash_flow.vlan_tci = 0;
2161 /* The symmetric quality of this hash function is not required, but
2162 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2163 * purposes, so we use it out of convenience. */
2164 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2167 static struct bond_entry *
2168 lookup_bond_entry(const struct port *port, const struct flow *flow,
2171 assert(port->bond_mode != BM_AB);
2173 if (bond_is_tcp_hash(port)) {
2174 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2176 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2180 static struct iface *
2181 bond_choose_iface(const struct port *port)
2183 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2184 struct iface *best_down_slave;
2187 best_down_slave = NULL;
2188 for (i = 0; i < port->n_ifaces; i++) {
2189 struct iface *iface = port->ifaces[i];
2191 if (iface->enabled) {
2193 } else if ((!best_down_slave
2194 || iface->delay_expires < best_down_slave->delay_expires)
2195 && lacp_slave_may_enable(port->lacp, iface)) {
2196 best_down_slave = iface;
2200 if (best_down_slave) {
2201 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2202 "since no other interface is up",
2203 best_down_slave->name,
2204 best_down_slave->delay_expires - time_msec());
2205 bond_enable_slave(best_down_slave, true);
2208 return best_down_slave;
2212 choose_output_iface(const struct port *port, const struct flow *flow,
2213 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2215 struct iface *iface;
2217 assert(port->n_ifaces);
2218 if (port->n_ifaces == 1) {
2219 iface = port->ifaces[0];
2220 } else if (port->bond_mode == BM_AB) {
2221 iface = port->active_iface;
2223 *tags |= port->no_ifaces_tag;
2227 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2228 if (!e->iface || !e->iface->enabled) {
2229 /* XXX select interface properly. The current interface selection
2230 * is only good for testing the rebalancing code. */
2231 e->iface = bond_choose_iface(port);
2233 *tags |= port->no_ifaces_tag;
2236 e->tag = tag_create_random();
2241 *dp_ifidx = iface->dp_ifidx;
2242 *tags |= iface->tag; /* Currently only used for bonding. */
2247 bond_link_status_update(struct iface *iface)
2249 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2250 struct port *port = iface->port;
2251 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2252 int updelay, downdelay;
2254 updelay = port->updelay;
2255 downdelay = port->downdelay;
2257 if (lacp_negotiated(port->lacp)) {
2262 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2263 /* Nothing to do. */
2266 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2267 iface->name, up ? "up" : "down");
2268 if (up == iface->enabled) {
2269 iface->delay_expires = LLONG_MAX;
2270 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2271 iface->name, up ? "disabled" : "enabled");
2272 } else if (up && !port->active_iface) {
2273 bond_enable_slave(iface, true);
2275 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2276 "other interface is up", iface->name, updelay);
2279 int delay = up ? updelay : downdelay;
2280 iface->delay_expires = time_msec() + delay;
2283 "interface %s: will be %s if it stays %s for %d ms",
2285 up ? "enabled" : "disabled",
2293 bond_choose_active_iface(struct port *port)
2295 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2297 port->active_iface = bond_choose_iface(port);
2298 if (port->active_iface) {
2299 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2300 port->name, port->active_iface->name);
2302 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2308 bond_enable_slave(struct iface *iface, bool enable)
2310 struct port *port = iface->port;
2311 struct bridge *br = port->bridge;
2313 /* This acts as a recursion check. If the act of disabling a slave
2314 * causes a different slave to be enabled, the flag will allow us to
2315 * skip redundant work when we reenter this function. It must be
2316 * cleared on exit to keep things safe with multiple bonds. */
2317 static bool moving_active_iface = false;
2319 iface->delay_expires = LLONG_MAX;
2320 if (enable == iface->enabled) {
2324 iface->enabled = enable;
2325 if (!iface->enabled) {
2326 VLOG_WARN("interface %s: disabled", iface->name);
2327 ofproto_revalidate(br->ofproto, iface->tag);
2328 if (iface == port->active_iface) {
2329 /* Disabling a slave can lead to another slave being immediately
2330 * enabled if there will be no active slaves but one is waiting
2331 * on an updelay. In this case we do not need to run most of the
2332 * code for the newly enabled slave since there was no period
2333 * without an active slave and it is redundant with the disabling
2335 moving_active_iface = true;
2336 bond_choose_active_iface(port);
2338 bond_send_learning_packets(port);
2340 VLOG_WARN("interface %s: enabled", iface->name);
2341 if (!port->active_iface && !moving_active_iface) {
2342 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2343 bond_choose_active_iface(port);
2344 bond_send_learning_packets(port);
2346 iface->tag = tag_create_random();
2349 moving_active_iface = false;
2352 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2353 * bond interface. */
2355 bond_update_fake_iface_stats(struct port *port)
2357 struct netdev_stats bond_stats;
2358 struct netdev *bond_dev;
2361 memset(&bond_stats, 0, sizeof bond_stats);
2363 for (i = 0; i < port->n_ifaces; i++) {
2364 struct netdev_stats slave_stats;
2366 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2367 /* XXX: We swap the stats here because they are swapped back when
2368 * reported by the internal device. The reason for this is
2369 * internal devices normally represent packets going into the system
2370 * but when used as fake bond device they represent packets leaving
2371 * the system. We really should do this in the internal device
2372 * itself because changing it here reverses the counts from the
2373 * perspective of the switch. However, the internal device doesn't
2374 * know what type of device it represents so we have to do it here
2376 bond_stats.tx_packets += slave_stats.rx_packets;
2377 bond_stats.tx_bytes += slave_stats.rx_bytes;
2378 bond_stats.rx_packets += slave_stats.tx_packets;
2379 bond_stats.rx_bytes += slave_stats.tx_bytes;
2383 if (!netdev_open_default(port->name, &bond_dev)) {
2384 netdev_set_stats(bond_dev, &bond_stats);
2385 netdev_close(bond_dev);
2390 bond_run(struct port *port)
2394 if (port->n_ifaces < 2) {
2398 for (i = 0; i < port->n_ifaces; i++) {
2399 bond_link_status_update(port->ifaces[i]);
2402 for (i = 0; i < port->n_ifaces; i++) {
2403 struct iface *iface = port->ifaces[i];
2404 if (time_msec() >= iface->delay_expires) {
2405 bond_enable_slave(iface, !iface->enabled);
2409 if (port->bond_fake_iface
2410 && time_msec() >= port->bond_next_fake_iface_update) {
2411 bond_update_fake_iface_stats(port);
2412 port->bond_next_fake_iface_update = time_msec() + 1000;
2417 bond_wait(struct port *port)
2421 if (port->n_ifaces < 2) {
2425 for (i = 0; i < port->n_ifaces; i++) {
2426 struct iface *iface = port->ifaces[i];
2427 if (iface->delay_expires != LLONG_MAX) {
2428 poll_timer_wait_until(iface->delay_expires);
2432 if (port->bond_fake_iface) {
2433 poll_timer_wait_until(port->bond_next_fake_iface_update);
2438 set_dst(struct dst *dst, const struct flow *flow,
2439 const struct port *in_port, const struct port *out_port,
2442 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2443 : in_port->vlan >= 0 ? in_port->vlan
2444 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2445 : vlan_tci_to_vid(flow->vlan_tci));
2446 return choose_output_iface(out_port, flow, dst->vlan,
2447 &dst->dp_ifidx, tags);
2451 swap_dst(struct dst *p, struct dst *q)
2453 struct dst tmp = *p;
2458 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2459 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2460 * that we push to the datapath. We could in fact fully sort the array by
2461 * vlan, but in most cases there are at most two different vlan tags so that's
2462 * possibly overkill.) */
2464 partition_dsts(struct dst_set *set, int vlan)
2466 struct dst *first = set->dsts;
2467 struct dst *last = set->dsts + set->n;
2469 while (first != last) {
2471 * - All dsts < first have vlan == 'vlan'.
2472 * - All dsts >= last have vlan != 'vlan'.
2473 * - first < last. */
2474 while (first->vlan == vlan) {
2475 if (++first == last) {
2480 /* Same invariants, plus one additional:
2481 * - first->vlan != vlan.
2483 while (last[-1].vlan != vlan) {
2484 if (--last == first) {
2489 /* Same invariants, plus one additional:
2490 * - last[-1].vlan == vlan.*/
2491 swap_dst(first++, --last);
2496 mirror_mask_ffs(mirror_mask_t mask)
2498 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2503 dst_set_init(struct dst_set *set)
2505 set->dsts = set->builtin;
2507 set->allocated = ARRAY_SIZE(set->builtin);
2511 dst_set_add(struct dst_set *set, const struct dst *dst)
2513 if (set->n >= set->allocated) {
2514 size_t new_allocated;
2515 struct dst *new_dsts;
2517 new_allocated = set->allocated * 2;
2518 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2519 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2523 set->dsts = new_dsts;
2524 set->allocated = new_allocated;
2526 set->dsts[set->n++] = *dst;
2530 dst_set_free(struct dst_set *set)
2532 if (set->dsts != set->builtin) {
2538 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2541 for (i = 0; i < set->n; i++) {
2542 if (set->dsts[i].vlan == test->vlan
2543 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2551 port_trunks_vlan(const struct port *port, uint16_t vlan)
2553 return (port->vlan < 0
2554 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2558 port_includes_vlan(const struct port *port, uint16_t vlan)
2560 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2564 port_is_floodable(const struct port *port)
2568 for (i = 0; i < port->n_ifaces; i++) {
2569 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2570 port->ifaces[i]->dp_ifidx)) {
2577 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2578 * there is no active iface. */
2580 port_get_active_iface_tag(const struct port *port)
2582 return (port->active_iface
2583 ? port->active_iface->tag
2584 : port->no_ifaces_tag);
2588 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2589 const struct port *in_port, const struct port *out_port,
2590 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2592 mirror_mask_t mirrors = in_port->src_mirrors;
2597 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2598 if (flow_vlan == 0) {
2599 flow_vlan = OFP_VLAN_NONE;
2602 if (out_port == FLOOD_PORT) {
2603 for (i = 0; i < br->n_ports; i++) {
2604 struct port *port = br->ports[i];
2606 && port_is_floodable(port)
2607 && port_includes_vlan(port, vlan)
2608 && !port->is_mirror_output_port
2609 && set_dst(&dst, flow, in_port, port, tags)) {
2610 mirrors |= port->dst_mirrors;
2611 dst_set_add(set, &dst);
2614 *nf_output_iface = NF_OUT_FLOOD;
2615 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2616 dst_set_add(set, &dst);
2617 *nf_output_iface = dst.dp_ifidx;
2618 mirrors |= out_port->dst_mirrors;
2622 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2623 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2625 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2626 && !dst_is_duplicate(set, &dst)) {
2627 dst_set_add(set, &dst);
2630 for (i = 0; i < br->n_ports; i++) {
2631 struct port *port = br->ports[i];
2632 if (port_includes_vlan(port, m->out_vlan)
2633 && set_dst(&dst, flow, in_port, port, tags))
2635 if (port->vlan < 0) {
2636 dst.vlan = m->out_vlan;
2638 if (dst_is_duplicate(set, &dst)) {
2642 /* Use the vlan tag on the original flow instead of
2643 * the one passed in the vlan parameter. This ensures
2644 * that we compare the vlan from before any implicit
2645 * tagging tags place. This is necessary because
2646 * dst->vlan is the final vlan, after removing implicit
2648 if (port == in_port && dst.vlan == flow_vlan) {
2649 /* Don't send out input port on same VLAN. */
2652 dst_set_add(set, &dst);
2657 mirrors &= mirrors - 1;
2660 partition_dsts(set, flow_vlan);
2663 static void OVS_UNUSED
2664 print_dsts(const struct dst_set *set)
2668 for (i = 0; i < set->n; i++) {
2669 const struct dst *dst = &set->dsts[i];
2671 printf(">p%"PRIu16, dst->dp_ifidx);
2672 if (dst->vlan != OFP_VLAN_NONE) {
2673 printf("v%"PRIu16, dst->vlan);
2679 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2680 const struct port *in_port, const struct port *out_port,
2681 tag_type *tags, struct ofpbuf *actions,
2682 uint16_t *nf_output_iface)
2689 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2692 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2693 if (cur_vlan == 0) {
2694 cur_vlan = OFP_VLAN_NONE;
2696 for (i = 0; i < set.n; i++) {
2697 const struct dst *dst = &set.dsts[i];
2698 if (dst->vlan != cur_vlan) {
2699 if (dst->vlan == OFP_VLAN_NONE) {
2700 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2703 tci = htons(dst->vlan & VLAN_VID_MASK);
2704 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2705 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2707 cur_vlan = dst->vlan;
2709 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2714 /* Returns the effective vlan of a packet, taking into account both the
2715 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2716 * the packet is untagged and -1 indicates it has an invalid header and
2717 * should be dropped. */
2718 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2719 struct port *in_port, bool have_packet)
2721 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2722 if (in_port->vlan >= 0) {
2724 /* XXX support double tagging? */
2726 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2727 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2728 "packet received on port %s configured with "
2729 "implicit VLAN %"PRIu16,
2730 br->name, vlan, in_port->name, in_port->vlan);
2734 vlan = in_port->vlan;
2736 if (!port_includes_vlan(in_port, vlan)) {
2738 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2739 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2740 "packet received on port %s not configured for "
2742 br->name, vlan, in_port->name, vlan);
2751 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2752 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2753 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2755 is_gratuitous_arp(const struct flow *flow)
2757 return (flow->dl_type == htons(ETH_TYPE_ARP)
2758 && eth_addr_is_broadcast(flow->dl_dst)
2759 && (flow->nw_proto == ARP_OP_REPLY
2760 || (flow->nw_proto == ARP_OP_REQUEST
2761 && flow->nw_src == flow->nw_dst)));
2765 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2766 struct port *in_port)
2768 enum grat_arp_lock_type lock_type;
2771 /* We don't want to learn from gratuitous ARP packets that are reflected
2772 * back over bond slaves so we lock the learning table. */
2773 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2774 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2775 GRAT_ARP_LOCK_CHECK;
2777 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2780 /* The log messages here could actually be useful in debugging,
2781 * so keep the rate limit relatively high. */
2782 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2784 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2785 "on port %s in VLAN %d",
2786 br->name, ETH_ADDR_ARGS(flow->dl_src),
2787 in_port->name, vlan);
2788 ofproto_revalidate(br->ofproto, rev_tag);
2792 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2793 * dropped. Returns true if they may be forwarded, false if they should be
2796 * If 'have_packet' is true, it indicates that the caller is processing a
2797 * received packet. If 'have_packet' is false, then the caller is just
2798 * revalidating an existing flow because configuration has changed. Either
2799 * way, 'have_packet' only affects logging (there is no point in logging errors
2800 * during revalidation).
2802 * Sets '*in_portp' to the input port. This will be a null pointer if
2803 * flow->in_port does not designate a known input port (in which case
2804 * is_admissible() returns false).
2806 * When returning true, sets '*vlanp' to the effective VLAN of the input
2807 * packet, as returned by flow_get_vlan().
2809 * May also add tags to '*tags', although the current implementation only does
2810 * so in one special case.
2813 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2814 tag_type *tags, int *vlanp, struct port **in_portp)
2816 struct iface *in_iface;
2817 struct port *in_port;
2820 /* Find the interface and port structure for the received packet. */
2821 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2823 /* No interface? Something fishy... */
2825 /* Odd. A few possible reasons here:
2827 * - We deleted an interface but there are still a few packets
2828 * queued up from it.
2830 * - Someone externally added an interface (e.g. with "ovs-dpctl
2831 * add-if") that we don't know about.
2833 * - Packet arrived on the local port but the local port is not
2834 * one of our bridge ports.
2836 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2838 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2839 "interface %"PRIu16, br->name, flow->in_port);
2845 *in_portp = in_port = in_iface->port;
2846 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2851 /* Drop frames for reserved multicast addresses. */
2852 if (eth_addr_is_reserved(flow->dl_dst)) {
2856 /* Drop frames on ports reserved for mirroring. */
2857 if (in_port->is_mirror_output_port) {
2859 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2860 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2861 "%s, which is reserved exclusively for mirroring",
2862 br->name, in_port->name);
2867 /* When using LACP, do not accept packets from disabled interfaces. */
2868 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2872 /* Packets received on non-LACP bonds need special attention to avoid
2874 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2876 bool is_grat_arp_locked;
2878 if (eth_addr_is_multicast(flow->dl_dst)) {
2879 *tags |= port_get_active_iface_tag(in_port);
2880 if (in_port->active_iface != in_iface) {
2881 /* Drop all multicast packets on inactive slaves. */
2886 /* Drop all packets for which we have learned a different input
2887 * port, because we probably sent the packet on one slave and got
2888 * it back on the other. Gratuitous ARP packets are an exception
2889 * to this rule: the host has moved to another switch. The exception
2890 * to the exception is if we locked the learning table to avoid
2891 * reflections on bond slaves. If this is the case, just drop the
2893 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2894 &is_grat_arp_locked);
2895 if (src_idx != -1 && src_idx != in_port->port_idx &&
2896 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2904 /* If the composed actions may be applied to any packet in the given 'flow',
2905 * returns true. Otherwise, the actions should only be applied to 'packet', or
2906 * not at all, if 'packet' was NULL. */
2908 process_flow(struct bridge *br, const struct flow *flow,
2909 const struct ofpbuf *packet, struct ofpbuf *actions,
2910 tag_type *tags, uint16_t *nf_output_iface)
2912 struct port *in_port;
2913 struct port *out_port;
2917 /* Check whether we should drop packets in this flow. */
2918 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2923 /* Learn source MAC (but don't try to learn from revalidation). */
2925 update_learning_table(br, flow, vlan, in_port);
2928 /* Determine output port. */
2929 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2931 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2932 out_port = br->ports[out_port_idx];
2933 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2934 /* If we are revalidating but don't have a learning entry then
2935 * eject the flow. Installing a flow that floods packets opens
2936 * up a window of time where we could learn from a packet reflected
2937 * on a bond and blackhole packets before the learning table is
2938 * updated to reflect the correct port. */
2941 out_port = FLOOD_PORT;
2944 /* Don't send packets out their input ports. */
2945 if (in_port == out_port) {
2951 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2959 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2960 struct ofpbuf *actions, tag_type *tags,
2961 uint16_t *nf_output_iface, void *br_)
2963 struct bridge *br = br_;
2965 COVERAGE_INC(bridge_process_flow);
2966 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2970 bridge_special_ofhook_cb(const struct flow *flow,
2971 const struct ofpbuf *packet, void *br_)
2973 struct iface *iface;
2974 struct bridge *br = br_;
2976 iface = iface_from_dp_ifidx(br, flow->in_port);
2978 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
2980 if (iface && iface->port->lacp && packet) {
2981 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
2984 COVERAGE_INC(bridge_process_lacp);
2985 lacp_process_pdu(iface->port->lacp, iface, pdu);
2995 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2996 const struct nlattr *actions,
2998 uint64_t n_bytes, void *br_)
3000 struct bridge *br = br_;
3001 const struct nlattr *a;
3002 struct port *in_port;
3007 /* Feed information from the active flows back into the learning table to
3008 * ensure that table is always in sync with what is actually flowing
3009 * through the datapath.
3011 * We test that 'tags' is nonzero to ensure that only flows that include an
3012 * OFPP_NORMAL action are used for learning. This works because
3013 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3014 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3015 update_learning_table(br, flow, vlan, in_port);
3018 /* Account for bond slave utilization. */
3019 if (!br->has_bonded_ports) {
3022 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3023 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3024 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3025 if (out_port && out_port->n_ifaces >= 2 &&
3026 out_port->bond_mode != BM_AB) {
3027 uint16_t vlan = (flow->vlan_tci
3028 ? vlan_tci_to_vid(flow->vlan_tci)
3030 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3031 e->tx_bytes += n_bytes;
3038 bridge_account_checkpoint_ofhook_cb(void *br_)
3040 struct bridge *br = br_;
3044 if (!br->has_bonded_ports) {
3049 for (i = 0; i < br->n_ports; i++) {
3050 struct port *port = br->ports[i];
3051 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3052 && now >= port->bond_next_rebalance) {
3053 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3054 bond_rebalance_port(port);
3059 static struct ofhooks bridge_ofhooks = {
3060 bridge_normal_ofhook_cb,
3061 bridge_special_ofhook_cb,
3062 bridge_account_flow_ofhook_cb,
3063 bridge_account_checkpoint_ofhook_cb,
3066 /* Bonding functions. */
3068 /* Statistics for a single interface on a bonded port, used for load-based
3069 * bond rebalancing. */
3070 struct slave_balance {
3071 struct iface *iface; /* The interface. */
3072 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3074 /* All the "bond_entry"s that are assigned to this interface, in order of
3075 * increasing tx_bytes. */
3076 struct bond_entry **hashes;
3081 bond_mode_to_string(enum bond_mode bm) {
3082 static char *bm_slb = "balance-slb";
3083 static char *bm_ab = "active-backup";
3084 static char *bm_tcp = "balance-tcp";
3087 case BM_SLB: return bm_slb;
3088 case BM_AB: return bm_ab;
3089 case BM_TCP: return bm_tcp;
3096 /* Sorts pointers to pointers to bond_entries in ascending order by the
3097 * interface to which they are assigned, and within a single interface in
3098 * ascending order of bytes transmitted. */
3100 compare_bond_entries(const void *a_, const void *b_)
3102 const struct bond_entry *const *ap = a_;
3103 const struct bond_entry *const *bp = b_;
3104 const struct bond_entry *a = *ap;
3105 const struct bond_entry *b = *bp;
3106 if (a->iface != b->iface) {
3107 return a->iface > b->iface ? 1 : -1;
3108 } else if (a->tx_bytes != b->tx_bytes) {
3109 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3115 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3116 * *descending* order by number of bytes transmitted. */
3118 compare_slave_balance(const void *a_, const void *b_)
3120 const struct slave_balance *a = a_;
3121 const struct slave_balance *b = b_;
3122 if (a->iface->enabled != b->iface->enabled) {
3123 return a->iface->enabled ? -1 : 1;
3124 } else if (a->tx_bytes != b->tx_bytes) {
3125 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3132 swap_bals(struct slave_balance *a, struct slave_balance *b)
3134 struct slave_balance tmp = *a;
3139 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3140 * given that 'p' (and only 'p') might be in the wrong location.
3142 * This function invalidates 'p', since it might now be in a different memory
3145 resort_bals(struct slave_balance *p,
3146 struct slave_balance bals[], size_t n_bals)
3149 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3150 swap_bals(p, p - 1);
3152 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3153 swap_bals(p, p + 1);
3159 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3161 if (VLOG_IS_DBG_ENABLED()) {
3162 struct ds ds = DS_EMPTY_INITIALIZER;
3163 const struct slave_balance *b;
3165 for (b = bals; b < bals + n_bals; b++) {
3169 ds_put_char(&ds, ',');
3171 ds_put_format(&ds, " %s %"PRIu64"kB",
3172 b->iface->name, b->tx_bytes / 1024);
3174 if (!b->iface->enabled) {
3175 ds_put_cstr(&ds, " (disabled)");
3177 if (b->n_hashes > 0) {
3178 ds_put_cstr(&ds, " (");
3179 for (i = 0; i < b->n_hashes; i++) {
3180 const struct bond_entry *e = b->hashes[i];
3182 ds_put_cstr(&ds, " + ");
3184 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3185 e - port->bond_hash, e->tx_bytes / 1024);
3187 ds_put_cstr(&ds, ")");
3190 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3195 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3197 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3200 struct bond_entry *hash = from->hashes[hash_idx];
3201 struct port *port = from->iface->port;
3202 uint64_t delta = hash->tx_bytes;
3204 assert(port->bond_mode != BM_AB);
3206 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3207 "from %s to %s (now carrying %"PRIu64"kB and "
3208 "%"PRIu64"kB load, respectively)",
3209 port->name, delta / 1024, hash - port->bond_hash,
3210 from->iface->name, to->iface->name,
3211 (from->tx_bytes - delta) / 1024,
3212 (to->tx_bytes + delta) / 1024);
3214 /* Delete element from from->hashes.
3216 * We don't bother to add the element to to->hashes because not only would
3217 * it require more work, the only purpose it would be to allow that hash to
3218 * be migrated to another slave in this rebalancing run, and there is no
3219 * point in doing that. */
3220 if (hash_idx == 0) {
3223 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3224 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3228 /* Shift load away from 'from' to 'to'. */
3229 from->tx_bytes -= delta;
3230 to->tx_bytes += delta;
3232 /* Arrange for flows to be revalidated. */
3233 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3234 hash->iface = to->iface;
3235 hash->tag = tag_create_random();
3239 bond_rebalance_port(struct port *port)
3241 struct slave_balance *bals;
3243 struct bond_entry *hashes[BOND_MASK + 1];
3244 struct slave_balance *b, *from, *to;
3245 struct bond_entry *e;
3248 assert(port->bond_mode != BM_AB);
3250 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3251 * descending order of tx_bytes, so that bals[0] represents the most
3252 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3255 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3256 * array for each slave_balance structure, we sort our local array of
3257 * hashes in order by slave, so that all of the hashes for a given slave
3258 * become contiguous in memory, and then we point each 'hashes' members of
3259 * a slave_balance structure to the start of a contiguous group. */
3260 n_bals = port->n_ifaces;
3261 bals = xmalloc(n_bals * sizeof *bals);
3262 for (b = bals; b < &bals[n_bals]; b++) {
3263 b->iface = port->ifaces[b - bals];
3268 for (i = 0; i <= BOND_MASK; i++) {
3269 hashes[i] = &port->bond_hash[i];
3271 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3272 for (i = 0; i <= BOND_MASK; i++) {
3278 for (b = bals; b < &bals[n_bals]; b++) {
3279 if (b->iface == e->iface) {
3280 b->tx_bytes += e->tx_bytes;
3282 b->hashes = &hashes[i];
3289 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3290 log_bals(bals, n_bals, port);
3292 /* Discard slaves that aren't enabled (which were sorted to the back of the
3293 * array earlier). */
3294 while (!bals[n_bals - 1].iface->enabled) {
3301 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3302 to = &bals[n_bals - 1];
3303 for (from = bals; from < to; ) {
3304 uint64_t overload = from->tx_bytes - to->tx_bytes;
3305 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3306 /* The extra load on 'from' (and all less-loaded slaves), compared
3307 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3308 * it is less than ~1Mbps. No point in rebalancing. */
3310 } else if (from->n_hashes == 1) {
3311 /* 'from' only carries a single MAC hash, so we can't shift any
3312 * load away from it, even though we want to. */
3315 /* 'from' is carrying significantly more load than 'to', and that
3316 * load is split across at least two different hashes. Pick a hash
3317 * to migrate to 'to' (the least-loaded slave), given that doing so
3318 * must decrease the ratio of the load on the two slaves by at
3321 * The sort order we use means that we prefer to shift away the
3322 * smallest hashes instead of the biggest ones. There is little
3323 * reason behind this decision; we could use the opposite sort
3324 * order to shift away big hashes ahead of small ones. */
3327 for (i = 0; i < from->n_hashes; i++) {
3328 double old_ratio, new_ratio;
3329 uint64_t delta = from->hashes[i]->tx_bytes;
3331 if (delta == 0 || from->tx_bytes - delta == 0) {
3332 /* Pointless move. */
3336 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3338 if (to->tx_bytes == 0) {
3339 /* Nothing on the new slave, move it. */
3343 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3344 new_ratio = (double)(from->tx_bytes - delta) /
3345 (to->tx_bytes + delta);
3347 if (new_ratio == 0) {
3348 /* Should already be covered but check to prevent division
3353 if (new_ratio < 1) {
3354 new_ratio = 1 / new_ratio;
3357 if (old_ratio - new_ratio > 0.1) {
3358 /* Would decrease the ratio, move it. */
3362 if (i < from->n_hashes) {
3363 bond_shift_load(from, to, i);
3365 /* If the result of the migration changed the relative order of
3366 * 'from' and 'to' swap them back to maintain invariants. */
3367 if (order_swapped) {
3368 swap_bals(from, to);
3371 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3372 * point to different slave_balance structures. It is only
3373 * valid to do these two operations in a row at all because we
3374 * know that 'from' will not move past 'to' and vice versa. */
3375 resort_bals(from, bals, n_bals);
3376 resort_bals(to, bals, n_bals);
3383 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3384 * historical data to decay to <1% in 7 rebalancing runs. */
3385 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3397 bond_send_learning_packets(struct port *port)
3399 struct bridge *br = port->bridge;
3400 struct mac_entry *e;
3401 struct ofpbuf packet;
3402 int error, n_packets, n_errors;
3404 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3408 ofpbuf_init(&packet, 128);
3409 error = n_packets = n_errors = 0;
3410 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3416 if (e->port == port->port_idx) {
3420 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3422 flow_extract(&packet, 0, ODPP_NONE, &flow);
3424 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3430 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3436 ofpbuf_uninit(&packet);
3439 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3440 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3441 "packets, last error was: %s",
3442 port->name, n_errors, n_packets, strerror(error));
3444 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3445 port->name, n_packets);
3449 /* Bonding unixctl user interface functions. */
3452 bond_unixctl_list(struct unixctl_conn *conn,
3453 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3455 struct ds ds = DS_EMPTY_INITIALIZER;
3456 const struct bridge *br;
3458 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3460 LIST_FOR_EACH (br, node, &all_bridges) {
3463 for (i = 0; i < br->n_ports; i++) {
3464 const struct port *port = br->ports[i];
3465 if (port->n_ifaces > 1) {
3468 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3469 bond_mode_to_string(port->bond_mode));
3470 for (j = 0; j < port->n_ifaces; j++) {
3471 const struct iface *iface = port->ifaces[j];
3473 ds_put_cstr(&ds, ", ");
3475 ds_put_cstr(&ds, iface->name);
3477 ds_put_char(&ds, '\n');
3481 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3485 static struct port *
3486 bond_find(const char *name)
3488 const struct bridge *br;
3490 LIST_FOR_EACH (br, node, &all_bridges) {
3493 for (i = 0; i < br->n_ports; i++) {
3494 struct port *port = br->ports[i];
3495 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3504 bond_unixctl_show(struct unixctl_conn *conn,
3505 const char *args, void *aux OVS_UNUSED)
3507 struct ds ds = DS_EMPTY_INITIALIZER;
3508 const struct port *port;
3511 port = bond_find(args);
3513 unixctl_command_reply(conn, 501, "no such bond");
3517 ds_put_format(&ds, "bond_mode: %s\n",
3518 bond_mode_to_string(port->bond_mode));
3521 ds_put_format(&ds, "lacp: %s\n",
3522 port->lacp_active ? "active" : "passive");
3524 ds_put_cstr(&ds, "lacp: off\n");
3527 if (port->bond_mode != BM_AB) {
3528 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3529 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3533 ds_put_format(&ds, "bond-detect-mode: %s\n",
3534 port->monitor ? "carrier" : "miimon");
3536 if (!port->monitor) {
3537 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3538 port->miimon_interval);
3541 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3542 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3544 if (port->bond_mode != BM_AB) {
3545 ds_put_format(&ds, "next rebalance: %lld ms\n",
3546 port->bond_next_rebalance - time_msec());
3549 for (j = 0; j < port->n_ifaces; j++) {
3550 const struct iface *iface = port->ifaces[j];
3551 struct bond_entry *be;
3555 ds_put_format(&ds, "\nslave %s: %s\n",
3556 iface->name, iface->enabled ? "enabled" : "disabled");
3557 if (iface == port->active_iface) {
3558 ds_put_cstr(&ds, "\tactive slave\n");
3560 if (iface->delay_expires != LLONG_MAX) {
3561 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3562 iface->enabled ? "downdelay" : "updelay",
3563 iface->delay_expires - time_msec());
3566 if (port->bond_mode == BM_AB) {
3571 memset(&flow, 0, sizeof flow);
3572 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3573 int hash = be - port->bond_hash;
3574 struct mac_entry *me;
3576 if (be->iface != iface) {
3580 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3581 hash, be->tx_bytes / 1024);
3583 if (port->bond_mode != BM_SLB) {
3588 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3592 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3593 if (bond_hash_src(me->mac, me->vlan) == hash
3594 && me->port != port->port_idx
3595 && choose_output_iface(port, &flow, me->vlan,
3597 && dp_ifidx == iface->dp_ifidx)
3599 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3600 ETH_ADDR_ARGS(me->mac));
3605 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3610 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3611 void *aux OVS_UNUSED)
3613 char *args = (char *) args_;
3614 char *save_ptr = NULL;
3615 char *bond_s, *hash_s, *slave_s;
3617 struct iface *iface;
3618 struct bond_entry *entry;
3621 bond_s = strtok_r(args, " ", &save_ptr);
3622 hash_s = strtok_r(NULL, " ", &save_ptr);
3623 slave_s = strtok_r(NULL, " ", &save_ptr);
3625 unixctl_command_reply(conn, 501,
3626 "usage: bond/migrate BOND HASH SLAVE");
3630 port = bond_find(bond_s);
3632 unixctl_command_reply(conn, 501, "no such bond");
3636 if (port->bond_mode != BM_SLB) {
3637 unixctl_command_reply(conn, 501, "not an SLB bond");
3641 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3642 hash = atoi(hash_s) & BOND_MASK;
3644 unixctl_command_reply(conn, 501, "bad hash");
3648 iface = port_lookup_iface(port, slave_s);
3650 unixctl_command_reply(conn, 501, "no such slave");
3654 if (!iface->enabled) {
3655 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3659 entry = &port->bond_hash[hash];
3660 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3661 entry->iface = iface;
3662 entry->tag = tag_create_random();
3663 unixctl_command_reply(conn, 200, "migrated");
3667 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3668 void *aux OVS_UNUSED)
3670 char *args = (char *) args_;
3671 char *save_ptr = NULL;
3672 char *bond_s, *slave_s;
3674 struct iface *iface;
3676 bond_s = strtok_r(args, " ", &save_ptr);
3677 slave_s = strtok_r(NULL, " ", &save_ptr);
3679 unixctl_command_reply(conn, 501,
3680 "usage: bond/set-active-slave BOND SLAVE");
3684 port = bond_find(bond_s);
3686 unixctl_command_reply(conn, 501, "no such bond");
3690 iface = port_lookup_iface(port, slave_s);
3692 unixctl_command_reply(conn, 501, "no such slave");
3696 if (!iface->enabled) {
3697 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3701 if (port->active_iface != iface) {
3702 ofproto_revalidate(port->bridge->ofproto,
3703 port_get_active_iface_tag(port));
3704 port->active_iface = iface;
3705 VLOG_INFO("port %s: active interface is now %s",
3706 port->name, iface->name);
3707 bond_send_learning_packets(port);
3708 unixctl_command_reply(conn, 200, "done");
3710 unixctl_command_reply(conn, 200, "no change");
3715 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3717 char *args = (char *) args_;
3718 char *save_ptr = NULL;
3719 char *bond_s, *slave_s;
3721 struct iface *iface;
3723 bond_s = strtok_r(args, " ", &save_ptr);
3724 slave_s = strtok_r(NULL, " ", &save_ptr);
3726 unixctl_command_reply(conn, 501,
3727 "usage: bond/enable/disable-slave BOND SLAVE");
3731 port = bond_find(bond_s);
3733 unixctl_command_reply(conn, 501, "no such bond");
3737 iface = port_lookup_iface(port, slave_s);
3739 unixctl_command_reply(conn, 501, "no such slave");
3743 bond_enable_slave(iface, enable);
3744 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3748 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3749 void *aux OVS_UNUSED)
3751 enable_slave(conn, args, true);
3755 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3756 void *aux OVS_UNUSED)
3758 enable_slave(conn, args, false);
3762 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3763 void *aux OVS_UNUSED)
3765 char *args = (char *) args_;
3766 uint8_t mac[ETH_ADDR_LEN];
3770 char *mac_s, *vlan_s;
3771 char *save_ptr = NULL;
3773 mac_s = strtok_r(args, " ", &save_ptr);
3774 vlan_s = strtok_r(NULL, " ", &save_ptr);
3777 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3778 unixctl_command_reply(conn, 501, "invalid vlan");
3782 vlan = OFP_VLAN_NONE;
3785 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3786 == ETH_ADDR_SCAN_COUNT) {
3787 hash = bond_hash_src(mac, vlan);
3789 hash_cstr = xasprintf("%u", hash);
3790 unixctl_command_reply(conn, 200, hash_cstr);
3793 unixctl_command_reply(conn, 501, "invalid mac");
3800 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3801 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3802 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3803 unixctl_command_register("bond/set-active-slave",
3804 bond_unixctl_set_active_slave, NULL);
3805 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3807 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3809 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3812 /* Port functions. */
3815 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3817 struct iface *iface = aux;
3818 uint8_t ea[ETH_ADDR_LEN];
3821 error = netdev_get_etheraddr(iface->netdev, ea);
3823 struct ofpbuf packet;
3825 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3826 compose_lacp_packet(&packet, ea, pdu);
3827 ofproto_send_packet(iface->port->bridge->ofproto,
3828 iface->dp_ifidx, 0, &packet);
3829 ofpbuf_uninit(&packet);
3831 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3832 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3833 "(%s)", iface->name, strerror(error));
3838 port_run(struct port *port)
3840 if (port->monitor) {
3843 /* Track carrier going up and down on interfaces. */
3844 while (!netdev_monitor_poll(port->monitor, &devname)) {
3845 struct iface *iface;
3847 iface = port_lookup_iface(port, devname);
3849 iface_update_carrier(iface);
3853 } else if (time_msec() >= port->miimon_next_update) {
3856 for (i = 0; i < port->n_ifaces; i++) {
3857 struct iface *iface = port->ifaces[i];
3858 iface_update_carrier(iface);
3860 port->miimon_next_update = time_msec() + port->miimon_interval;
3866 for (i = 0; i < port->n_ifaces; i++) {
3867 struct iface *iface = port->ifaces[i];
3868 lacp_slave_enable(port->lacp, iface, iface->enabled);
3871 lacp_run(port->lacp, lacp_send_pdu_cb);
3878 port_wait(struct port *port)
3880 if (port->monitor) {
3881 netdev_monitor_poll_wait(port->monitor);
3883 poll_timer_wait_until(port->miimon_next_update);
3887 lacp_wait(port->lacp);
3893 static struct port *
3894 port_create(struct bridge *br, const char *name)
3898 port = xzalloc(sizeof *port);
3900 port->port_idx = br->n_ports;
3902 port->trunks = NULL;
3903 port->name = xstrdup(name);
3904 port->active_iface = NULL;
3906 if (br->n_ports >= br->allocated_ports) {
3907 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3910 br->ports[br->n_ports++] = port;
3911 shash_add_assert(&br->port_by_name, port->name, port);
3913 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3920 get_port_other_config(const struct ovsrec_port *port, const char *key,
3921 const char *default_value)
3925 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3927 return value ? value : default_value;
3931 get_interface_other_config(const struct ovsrec_interface *iface,
3932 const char *key, const char *default_value)
3936 value = get_ovsrec_key_value(&iface->header_,
3937 &ovsrec_interface_col_other_config, key);
3938 return value ? value : default_value;
3942 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3944 struct shash new_ifaces;
3947 /* Collect list of new interfaces. */
3948 shash_init(&new_ifaces);
3949 for (i = 0; i < cfg->n_interfaces; i++) {
3950 const char *name = cfg->interfaces[i]->name;
3951 shash_add_once(&new_ifaces, name, NULL);
3954 /* Get rid of deleted interfaces. */
3955 for (i = 0; i < port->n_ifaces; ) {
3956 struct iface *iface = port->ifaces[i];
3957 if (!shash_find(&new_ifaces, iface->name)) {
3958 iface_destroy(iface);
3964 shash_destroy(&new_ifaces);
3968 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3970 const char *detect_mode;
3971 struct shash new_ifaces;
3972 long long int next_rebalance, miimon_next_update, lacp_priority;
3973 unsigned long *trunks;
3979 /* Update settings. */
3980 port->updelay = cfg->bond_updelay;
3981 if (port->updelay < 0) {
3984 port->downdelay = cfg->bond_downdelay;
3985 if (port->downdelay < 0) {
3986 port->downdelay = 0;
3988 port->bond_rebalance_interval = atoi(
3989 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3990 if (port->bond_rebalance_interval < 1000) {
3991 port->bond_rebalance_interval = 1000;
3993 next_rebalance = time_msec() + port->bond_rebalance_interval;
3994 if (port->bond_next_rebalance > next_rebalance) {
3995 port->bond_next_rebalance = next_rebalance;
3998 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4001 netdev_monitor_destroy(port->monitor);
4002 port->monitor = NULL;
4004 if (strcmp(detect_mode, "miimon")) {
4005 port->monitor = netdev_monitor_create();
4007 if (strcmp(detect_mode, "carrier")) {
4008 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4009 "defaulting to carrier", port->name, detect_mode);
4013 port->miimon_interval = atoi(
4014 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4015 if (port->miimon_interval < 100) {
4016 port->miimon_interval = 100;
4018 miimon_next_update = time_msec() + port->miimon_interval;
4019 if (port->miimon_next_update > miimon_next_update) {
4020 port->miimon_next_update = miimon_next_update;
4023 if (!port->cfg->bond_mode ||
4024 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4025 port->bond_mode = BM_SLB;
4026 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4027 port->bond_mode = BM_AB;
4028 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4029 port->bond_mode = BM_TCP;
4031 port->bond_mode = BM_SLB;
4032 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4033 port->name, port->cfg->bond_mode,
4034 bond_mode_to_string(port->bond_mode));
4037 /* Add new interfaces and update 'cfg' member of existing ones. */
4038 shash_init(&new_ifaces);
4039 for (i = 0; i < cfg->n_interfaces; i++) {
4040 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4041 struct iface *iface;
4043 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4044 VLOG_WARN("port %s: %s specified twice as port interface",
4045 port->name, if_cfg->name);
4046 iface_set_ofport(if_cfg, -1);
4050 iface = iface_lookup(port->bridge, if_cfg->name);
4052 if (iface->port != port) {
4053 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4055 port->bridge->name, if_cfg->name, iface->port->name);
4058 iface->cfg = if_cfg;
4060 iface = iface_create(port, if_cfg);
4063 /* Determine interface type. The local port always has type
4064 * "internal". Other ports take their type from the database and
4065 * default to "system" if none is specified. */
4066 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4067 : if_cfg->type[0] ? if_cfg->type
4071 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4074 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4075 iface->lacp_priority = UINT16_MAX;
4077 iface->lacp_priority = lacp_priority;
4080 shash_destroy(&new_ifaces);
4082 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4086 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4088 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4089 /* Prefer bondable links if unspecified. */
4090 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4092 port->lacp_priority = lacp_priority;
4095 if (!port->cfg->lacp) {
4096 /* XXX when LACP implementation has been sufficiently tested, enable by
4097 * default and make active on bonded ports. */
4098 lacp_destroy(port->lacp);
4100 } else if (!strcmp(port->cfg->lacp, "off")) {
4101 lacp_destroy(port->lacp);
4103 } else if (!strcmp(port->cfg->lacp, "active")) {
4105 port->lacp = lacp_create();
4107 port->lacp_active = true;
4108 } else if (!strcmp(port->cfg->lacp, "passive")) {
4110 port->lacp = lacp_create();
4112 port->lacp_active = false;
4114 VLOG_WARN("port %s: unknown LACP mode %s",
4115 port->name, port->cfg->lacp);
4116 lacp_destroy(port->lacp);
4123 if (port->n_ifaces < 2) {
4125 if (vlan >= 0 && vlan <= 4095) {
4126 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4131 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4132 * they even work as-is. But they have not been tested. */
4133 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4137 if (port->vlan != vlan) {
4139 bridge_flush(port->bridge);
4142 /* Get trunked VLANs. */
4144 if (vlan < 0 && cfg->n_trunks) {
4147 trunks = bitmap_allocate(4096);
4149 for (i = 0; i < cfg->n_trunks; i++) {
4150 int trunk = cfg->trunks[i];
4152 bitmap_set1(trunks, trunk);
4158 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4159 port->name, cfg->n_trunks);
4161 if (n_errors == cfg->n_trunks) {
4162 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4164 bitmap_free(trunks);
4167 } else if (vlan >= 0 && cfg->n_trunks) {
4168 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4172 ? port->trunks != NULL
4173 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4174 bridge_flush(port->bridge);
4176 bitmap_free(port->trunks);
4177 port->trunks = trunks;
4181 port_destroy(struct port *port)
4184 struct bridge *br = port->bridge;
4188 for (i = 0; i < MAX_MIRRORS; i++) {
4189 struct mirror *m = br->mirrors[i];
4190 if (m && m->out_port == port) {
4195 while (port->n_ifaces > 0) {
4196 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4199 shash_find_and_delete_assert(&br->port_by_name, port->name);
4201 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4202 del->port_idx = port->port_idx;
4204 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4206 netdev_monitor_destroy(port->monitor);
4208 bitmap_free(port->trunks);
4215 static struct port *
4216 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4218 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4219 return iface ? iface->port : NULL;
4222 static struct port *
4223 port_lookup(const struct bridge *br, const char *name)
4225 return shash_find_data(&br->port_by_name, name);
4228 static struct iface *
4229 port_lookup_iface(const struct port *port, const char *name)
4231 struct iface *iface = iface_lookup(port->bridge, name);
4232 return iface && iface->port == port ? iface : NULL;
4236 port_update_lacp(struct port *port)
4241 lacp_configure(port->lacp, port->name,
4242 port->bridge->ea, port->lacp_priority,
4243 port->lacp_active, port->lacp_fast);
4245 for (i = 0; i < port->n_ifaces; i++) {
4246 struct iface *iface = port->ifaces[i];
4247 lacp_slave_register(port->lacp, iface, iface->name,
4248 iface->dp_ifidx, iface->lacp_priority);
4254 port_update_bonding(struct port *port)
4256 if (port->n_ifaces < 2) {
4257 /* Not a bonded port. */
4258 free(port->bond_hash);
4259 port->bond_hash = NULL;
4260 port->bond_fake_iface = false;
4261 port->active_iface = NULL;
4262 port->no_ifaces_tag = 0;
4266 if (port->bond_mode != BM_AB && !port->bond_hash) {
4267 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4268 for (i = 0; i <= BOND_MASK; i++) {
4269 struct bond_entry *e = &port->bond_hash[i];
4273 port->bond_next_rebalance
4274 = time_msec() + port->bond_rebalance_interval;
4275 } else if (port->bond_mode == BM_AB) {
4276 free(port->bond_hash);
4277 port->bond_hash = NULL;
4280 if (!port->no_ifaces_tag) {
4281 port->no_ifaces_tag = tag_create_random();
4284 if (!port->active_iface) {
4285 bond_choose_active_iface(port);
4288 port->bond_fake_iface = port->cfg->bond_fake_iface;
4289 if (port->bond_fake_iface) {
4290 port->bond_next_fake_iface_update = time_msec();
4296 /* Interface functions. */
4298 static struct iface *
4299 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4301 struct bridge *br = port->bridge;
4302 struct iface *iface;
4303 char *name = if_cfg->name;
4305 iface = xzalloc(sizeof *iface);
4307 iface->name = xstrdup(name);
4308 iface->dp_ifidx = -1;
4309 iface->tag = tag_create_random();
4310 iface->delay_expires = LLONG_MAX;
4311 iface->netdev = NULL;
4312 iface->cfg = if_cfg;
4314 shash_add_assert(&br->iface_by_name, iface->name, iface);
4316 if (port->n_ifaces >= port->allocated_ifaces) {
4317 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4318 sizeof *port->ifaces);
4320 port->ifaces[port->n_ifaces++] = iface;
4321 if (port->n_ifaces > 1) {
4322 br->has_bonded_ports = true;
4325 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4333 iface_destroy(struct iface *iface)
4336 struct port *port = iface->port;
4337 struct bridge *br = port->bridge;
4338 bool del_active = port->active_iface == iface;
4341 if (port->bond_hash) {
4342 struct bond_entry *e;
4343 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4344 if (e->iface == iface) {
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 for (i = 0; i < port->n_ifaces; i++) {
4365 if (iface == port->ifaces[i]) {
4366 port->ifaces[i] = port->ifaces[--port->n_ifaces];
4371 netdev_close(iface->netdev);
4374 bond_choose_active_iface(port);
4375 bond_send_learning_packets(port);
4381 bridge_flush(port->bridge);
4385 static struct iface *
4386 iface_lookup(const struct bridge *br, const char *name)
4388 return shash_find_data(&br->iface_by_name, name);
4391 static struct iface *
4392 iface_find(const char *name)
4394 const struct bridge *br;
4396 LIST_FOR_EACH (br, node, &all_bridges) {
4397 struct iface *iface = iface_lookup(br, name);
4406 static struct iface *
4407 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4409 struct iface *iface;
4411 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4412 hash_int(dp_ifidx, 0), &br->ifaces) {
4413 if (iface->dp_ifidx == dp_ifidx) {
4420 /* Set Ethernet address of 'iface', if one is specified in the configuration
4423 iface_set_mac(struct iface *iface)
4425 uint8_t ea[ETH_ADDR_LEN];
4427 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4428 if (eth_addr_is_multicast(ea)) {
4429 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4431 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4432 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4433 iface->name, iface->name);
4435 int error = netdev_set_etheraddr(iface->netdev, ea);
4437 VLOG_ERR("interface %s: setting MAC failed (%s)",
4438 iface->name, strerror(error));
4444 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4446 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4449 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4453 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4455 * The value strings in '*shash' are taken directly from values[], not copied,
4456 * so the caller should not modify or free them. */
4458 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4459 struct shash *shash)
4464 for (i = 0; i < n; i++) {
4465 shash_add(shash, keys[i], values[i]);
4469 /* Creates 'keys' and 'values' arrays from 'shash'.
4471 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4472 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4473 * are populated with with strings taken directly from 'shash' and thus have
4474 * the same ownership of the key-value pairs in shash.
4477 shash_to_ovs_idl_map(struct shash *shash,
4478 char ***keys, char ***values, size_t *n)
4482 struct shash_node *sn;
4484 count = shash_count(shash);
4486 k = xmalloc(count * sizeof *k);
4487 v = xmalloc(count * sizeof *v);
4490 SHASH_FOR_EACH(sn, shash) {
4501 struct iface_delete_queues_cbdata {
4502 struct netdev *netdev;
4503 const struct ovsdb_datum *queues;
4507 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4509 union ovsdb_atom atom;
4511 atom.integer = target;
4512 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4516 iface_delete_queues(unsigned int queue_id,
4517 const struct shash *details OVS_UNUSED, void *cbdata_)
4519 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4521 if (!queue_ids_include(cbdata->queues, queue_id)) {
4522 netdev_delete_queue(cbdata->netdev, queue_id);
4527 iface_update_carrier(struct iface *iface)
4529 bool carrier = iface_get_carrier(iface);
4530 if (carrier == iface->up) {
4534 iface->up = carrier;
4535 if (iface->port->lacp) {
4536 lacp_slave_carrier_changed(iface->port->lacp, iface);
4541 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4543 if (!qos || qos->type[0] == '\0') {
4544 netdev_set_qos(iface->netdev, NULL, NULL);
4546 struct iface_delete_queues_cbdata cbdata;
4547 struct shash details;
4550 /* Configure top-level Qos for 'iface'. */
4551 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4552 qos->n_other_config, &details);
4553 netdev_set_qos(iface->netdev, qos->type, &details);
4554 shash_destroy(&details);
4556 /* Deconfigure queues that were deleted. */
4557 cbdata.netdev = iface->netdev;
4558 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4560 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4562 /* Configure queues for 'iface'. */
4563 for (i = 0; i < qos->n_queues; i++) {
4564 const struct ovsrec_queue *queue = qos->value_queues[i];
4565 unsigned int queue_id = qos->key_queues[i];
4567 shash_from_ovs_idl_map(queue->key_other_config,
4568 queue->value_other_config,
4569 queue->n_other_config, &details);
4570 netdev_set_queue(iface->netdev, queue_id, &details);
4571 shash_destroy(&details);
4577 iface_update_cfm(struct iface *iface)
4581 uint16_t *remote_mps;
4582 struct ovsrec_monitor *mon;
4583 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4585 mon = iface->cfg->monitor;
4588 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4592 if (netdev_get_etheraddr(iface->netdev, ea)) {
4593 VLOG_WARN("interface %s: Failed to get ethernet address. "
4594 "Skipping Monitor.", iface->name);
4598 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4599 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4603 cfm.mpid = mon->mpid;
4604 cfm.interval = mon->interval ? *mon->interval : 1000;
4606 memcpy(cfm.eth_src, ea, sizeof cfm.eth_src);
4607 memcpy(cfm.maid, maid, sizeof cfm.maid);
4609 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4610 for(i = 0; i < mon->n_remote_mps; i++) {
4611 remote_mps[i] = mon->remote_mps[i]->mpid;
4614 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4615 &cfm, remote_mps, mon->n_remote_mps);
4619 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4620 * how 'iface''s port is configured.
4622 * Returns true if 'iface' is up, false otherwise. */
4624 iface_get_carrier(const struct iface *iface)
4626 return (iface->port->monitor
4627 ? netdev_get_carrier(iface->netdev)
4628 : netdev_get_miimon(iface->netdev));
4631 /* Port mirroring. */
4633 static struct mirror *
4634 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4638 for (i = 0; i < MAX_MIRRORS; i++) {
4639 struct mirror *m = br->mirrors[i];
4640 if (m && uuid_equals(uuid, &m->uuid)) {
4648 mirror_reconfigure(struct bridge *br)
4650 unsigned long *rspan_vlans;
4653 /* Get rid of deleted mirrors. */
4654 for (i = 0; i < MAX_MIRRORS; i++) {
4655 struct mirror *m = br->mirrors[i];
4657 const struct ovsdb_datum *mc;
4658 union ovsdb_atom atom;
4660 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4661 atom.uuid = br->mirrors[i]->uuid;
4662 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4668 /* Add new mirrors and reconfigure existing ones. */
4669 for (i = 0; i < br->cfg->n_mirrors; i++) {
4670 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4671 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4673 mirror_reconfigure_one(m, cfg);
4675 mirror_create(br, cfg);
4679 /* Update port reserved status. */
4680 for (i = 0; i < br->n_ports; i++) {
4681 br->ports[i]->is_mirror_output_port = false;
4683 for (i = 0; i < MAX_MIRRORS; i++) {
4684 struct mirror *m = br->mirrors[i];
4685 if (m && m->out_port) {
4686 m->out_port->is_mirror_output_port = true;
4690 /* Update flooded vlans (for RSPAN). */
4692 if (br->cfg->n_flood_vlans) {
4693 rspan_vlans = bitmap_allocate(4096);
4695 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4696 int64_t vlan = br->cfg->flood_vlans[i];
4697 if (vlan >= 0 && vlan < 4096) {
4698 bitmap_set1(rspan_vlans, vlan);
4699 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4702 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4707 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4713 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4718 for (i = 0; ; i++) {
4719 if (i >= MAX_MIRRORS) {
4720 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4721 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4724 if (!br->mirrors[i]) {
4729 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4732 br->mirrors[i] = m = xzalloc(sizeof *m);
4735 m->name = xstrdup(cfg->name);
4736 shash_init(&m->src_ports);
4737 shash_init(&m->dst_ports);
4743 mirror_reconfigure_one(m, cfg);
4747 mirror_destroy(struct mirror *m)
4750 struct bridge *br = m->bridge;
4753 for (i = 0; i < br->n_ports; i++) {
4754 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4755 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4758 shash_destroy(&m->src_ports);
4759 shash_destroy(&m->dst_ports);
4762 m->bridge->mirrors[m->idx] = NULL;
4771 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4772 struct shash *names)
4776 for (i = 0; i < n_ports; i++) {
4777 const char *name = ports[i]->name;
4778 if (port_lookup(m->bridge, name)) {
4779 shash_add_once(names, name, NULL);
4781 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4782 "port %s", m->bridge->name, m->name, name);
4788 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4794 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4796 for (i = 0; i < cfg->n_select_vlan; i++) {
4797 int64_t vlan = cfg->select_vlan[i];
4798 if (vlan < 0 || vlan > 4095) {
4799 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4800 m->bridge->name, m->name, vlan);
4802 (*vlans)[n_vlans++] = vlan;
4809 vlan_is_mirrored(const struct mirror *m, int vlan)
4813 for (i = 0; i < m->n_vlans; i++) {
4814 if (m->vlans[i] == vlan) {
4822 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4826 for (i = 0; i < m->n_vlans; i++) {
4827 if (port_trunks_vlan(p, m->vlans[i])) {
4835 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4837 struct shash src_ports, dst_ports;
4838 mirror_mask_t mirror_bit;
4839 struct port *out_port;
4846 if (strcmp(cfg->name, m->name)) {
4848 m->name = xstrdup(cfg->name);
4851 /* Get output port. */
4852 if (cfg->output_port) {
4853 out_port = port_lookup(m->bridge, cfg->output_port->name);
4855 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4856 m->bridge->name, m->name);
4862 if (cfg->output_vlan) {
4863 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4864 "output vlan; ignoring output vlan",
4865 m->bridge->name, m->name);
4867 } else if (cfg->output_vlan) {
4869 out_vlan = *cfg->output_vlan;
4871 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4872 m->bridge->name, m->name);
4877 shash_init(&src_ports);
4878 shash_init(&dst_ports);
4879 if (cfg->select_all) {
4880 for (i = 0; i < m->bridge->n_ports; i++) {
4881 const char *name = m->bridge->ports[i]->name;
4882 shash_add_once(&src_ports, name, NULL);
4883 shash_add_once(&dst_ports, name, NULL);
4888 /* Get ports, and drop duplicates and ports that don't exist. */
4889 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4891 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4894 /* Get all the vlans, and drop duplicate and invalid vlans. */
4895 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4898 /* Update mirror data. */
4899 if (!shash_equal_keys(&m->src_ports, &src_ports)
4900 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4901 || m->n_vlans != n_vlans
4902 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4903 || m->out_port != out_port
4904 || m->out_vlan != out_vlan) {
4905 bridge_flush(m->bridge);
4907 shash_swap(&m->src_ports, &src_ports);
4908 shash_swap(&m->dst_ports, &dst_ports);
4911 m->n_vlans = n_vlans;
4912 m->out_port = out_port;
4913 m->out_vlan = out_vlan;
4916 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4917 for (i = 0; i < m->bridge->n_ports; i++) {
4918 struct port *port = m->bridge->ports[i];
4920 if (shash_find(&m->src_ports, port->name)
4923 ? port_trunks_any_mirrored_vlan(m, port)
4924 : vlan_is_mirrored(m, port->vlan)))) {
4925 port->src_mirrors |= mirror_bit;
4927 port->src_mirrors &= ~mirror_bit;
4930 if (shash_find(&m->dst_ports, port->name)) {
4931 port->dst_mirrors |= mirror_bit;
4933 port->dst_mirrors &= ~mirror_bit;
4938 shash_destroy(&src_ports);
4939 shash_destroy(&dst_ports);