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"
66 #include "system-stats.h"
71 #include "vswitchd/vswitch-idl.h"
72 #include "xenserver.h"
74 #include "sflow_api.h"
76 VLOG_DEFINE_THIS_MODULE(bridge);
78 COVERAGE_DEFINE(bridge_flush);
79 COVERAGE_DEFINE(bridge_process_flow);
80 COVERAGE_DEFINE(bridge_process_cfm);
81 COVERAGE_DEFINE(bridge_process_lacp);
82 COVERAGE_DEFINE(bridge_reconfigure);
83 COVERAGE_DEFINE(bridge_lacp_update);
91 struct dst builtin[32];
96 static void dst_set_init(struct dst_set *);
97 static void dst_set_add(struct dst_set *, const struct dst *);
98 static void dst_set_free(struct dst_set *);
101 /* These members are always valid. */
102 struct list port_elem; /* Element in struct port's "ifaces" list. */
103 struct port *port; /* Containing port. */
104 char *name; /* Host network device name. */
105 tag_type tag; /* Tag associated with this interface. */
106 long long delay_expires; /* Time after which 'enabled' may change. */
108 /* These members are valid only after bridge_reconfigure() causes them to
110 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
111 int dp_ifidx; /* Index within kernel datapath. */
112 struct netdev *netdev; /* Network device. */
113 bool enabled; /* May be chosen for flows? */
114 bool up; /* Is the interface up? */
115 const char *type; /* Usually same as cfg->type. */
116 const struct ovsrec_interface *cfg;
118 /* LACP information. */
119 uint16_t lacp_priority; /* LACP port priority. */
122 #define BOND_MASK 0xff
124 struct iface *iface; /* Assigned iface, or NULL if none. */
125 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
126 tag_type tag; /* Tag for bond_entry<->iface association. */
130 BM_TCP, /* Transport Layer Load Balance. */
131 BM_SLB, /* Source Load Balance. */
132 BM_AB /* Active Backup. */
135 #define MAX_MIRRORS 32
136 typedef uint32_t mirror_mask_t;
137 #define MIRROR_MASK_C(X) UINT32_C(X)
138 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
140 struct bridge *bridge;
143 struct uuid uuid; /* UUID of this "mirror" record in database. */
145 /* Selection criteria. */
146 struct sset src_ports; /* Source port names. */
147 struct sset dst_ports; /* Destination port names. */
152 struct port *out_port;
156 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
158 struct bridge *bridge;
159 struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */
162 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
163 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
164 * NULL if all VLANs are trunked. */
165 const struct ovsrec_port *cfg;
168 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
169 long long int miimon_interval; /* Miimon status refresh interval. */
170 long long int miimon_next_update; /* Time of next miimon update. */
172 /* An ordinary bridge port has 1 interface.
173 * A bridge port for bonding has at least 2 interfaces. */
174 struct list ifaces; /* List of "struct iface"s. */
175 size_t n_ifaces; /* list_size(ifaces). */
178 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
179 struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */
180 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
181 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
182 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
183 long long int bond_next_fake_iface_update; /* Time of next update. */
185 /* LACP information. */
186 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
187 bool lacp_active; /* True if LACP is active */
188 bool lacp_fast; /* True if LACP is in fast mode. */
189 uint16_t lacp_priority; /* LACP system priority. */
191 /* SLB specific bonding info. */
192 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
193 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
194 long long int bond_next_rebalance; /* Next rebalancing time. */
196 /* Port mirroring info. */
197 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
198 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
199 bool is_mirror_output_port; /* Does port mirroring send frames here? */
203 struct list node; /* Node in global list of bridges. */
204 char *name; /* User-specified arbitrary name. */
205 struct mac_learning *ml; /* MAC learning table. */
206 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
207 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
208 const struct ovsrec_bridge *cfg;
210 /* OpenFlow switch processing. */
211 struct ofproto *ofproto; /* OpenFlow switch. */
213 /* Kernel datapath information. */
214 struct dpif *dpif; /* Datapath. */
215 struct hmap ifaces; /* Contains "struct iface"s. */
218 struct hmap ports; /* "struct port"s indexed by name. */
219 struct shash iface_by_name; /* "struct iface"s indexed by name. */
222 bool has_bonded_ports;
227 /* Port mirroring. */
228 struct mirror *mirrors[MAX_MIRRORS];
231 /* List of all bridges. */
232 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
234 /* OVSDB IDL used to obtain configuration. */
235 static struct ovsdb_idl *idl;
237 /* Each time this timer expires, the bridge fetches systems and interface
238 * statistics and pushes them into the database. */
239 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
240 static long long int stats_timer = LLONG_MIN;
242 /* Stores the time after which CFM statistics may be written to the database.
243 * Only updated when changes to the database require rate limiting. */
244 #define CFM_LIMIT_INTERVAL (1 * 1000) /* In milliseconds. */
245 static long long int cfm_limiter = LLONG_MIN;
247 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
248 static void bridge_destroy(struct bridge *);
249 static struct bridge *bridge_lookup(const char *name);
250 static unixctl_cb_func bridge_unixctl_dump_flows;
251 static unixctl_cb_func bridge_unixctl_reconnect;
252 static int bridge_run_one(struct bridge *);
253 static size_t bridge_get_controllers(const struct bridge *br,
254 struct ovsrec_controller ***controllersp);
255 static void bridge_reconfigure_one(struct bridge *);
256 static void bridge_reconfigure_remotes(struct bridge *,
257 const struct sockaddr_in *managers,
259 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
260 static void bridge_fetch_dp_ifaces(struct bridge *);
261 static void bridge_flush(struct bridge *);
262 static void bridge_pick_local_hw_addr(struct bridge *,
263 uint8_t ea[ETH_ADDR_LEN],
264 struct iface **hw_addr_iface);
265 static uint64_t bridge_pick_datapath_id(struct bridge *,
266 const uint8_t bridge_ea[ETH_ADDR_LEN],
267 struct iface *hw_addr_iface);
268 static uint64_t dpid_from_hash(const void *, size_t nbytes);
270 static unixctl_cb_func bridge_unixctl_fdb_show;
271 static unixctl_cb_func cfm_unixctl_show;
272 static unixctl_cb_func qos_unixctl_show;
274 static void bond_init(void);
275 static void bond_run(struct port *);
276 static void bond_wait(struct port *);
277 static void bond_rebalance_port(struct port *);
278 static void bond_send_learning_packets(struct port *);
279 static void bond_enable_slave(struct iface *iface, bool enable);
281 static void port_run(struct port *);
282 static void port_wait(struct port *);
283 static struct port *port_create(struct bridge *, const char *name);
284 static void port_reconfigure(struct port *, const struct ovsrec_port *);
285 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
286 static void port_destroy(struct port *);
287 static struct port *port_lookup(const struct bridge *, const char *name);
288 static struct iface *port_lookup_iface(const struct port *, const char *name);
289 static struct iface *port_get_an_iface(const struct port *);
290 static struct port *port_from_dp_ifidx(const struct bridge *,
292 static void port_update_bonding(struct port *);
293 static void port_update_lacp(struct port *);
295 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
296 static void mirror_destroy(struct mirror *);
297 static void mirror_reconfigure(struct bridge *);
298 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
299 static bool vlan_is_mirrored(const struct mirror *, int vlan);
301 static struct iface *iface_create(struct port *port,
302 const struct ovsrec_interface *if_cfg);
303 static void iface_destroy(struct iface *);
304 static struct iface *iface_lookup(const struct bridge *, const char *name);
305 static struct iface *iface_find(const char *name);
306 static struct iface *iface_from_dp_ifidx(const struct bridge *,
308 static void iface_set_mac(struct iface *);
309 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
310 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
311 static void iface_update_cfm(struct iface *);
312 static bool iface_refresh_cfm_stats(struct iface *iface);
313 static void iface_update_carrier(struct iface *);
314 static bool iface_get_carrier(const struct iface *);
316 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
318 static void shash_to_ovs_idl_map(struct shash *,
319 char ***keys, char ***values, size_t *n);
321 /* Hooks into ofproto processing. */
322 static struct ofhooks bridge_ofhooks;
324 /* Public functions. */
326 /* Initializes the bridge module, configuring it to obtain its configuration
327 * from an OVSDB server accessed over 'remote', which should be a string in a
328 * form acceptable to ovsdb_idl_create(). */
330 bridge_init(const char *remote)
332 /* Create connection to database. */
333 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
335 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
336 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
337 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
338 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_ovs_version);
339 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_db_version);
340 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_system_type);
341 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_system_version);
343 ovsdb_idl_omit_alert(idl, &ovsrec_bridge_col_datapath_id);
344 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
346 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
347 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
349 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_admin_state);
350 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_duplex);
351 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_link_speed);
352 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_link_state);
353 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_mtu);
354 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
355 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
356 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_status);
357 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
359 ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_is_connected);
360 ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_role);
361 ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_status);
362 ovsdb_idl_omit(idl, &ovsrec_controller_col_external_ids);
364 ovsdb_idl_omit_alert(idl, &ovsrec_maintenance_point_col_fault);
366 ovsdb_idl_omit_alert(idl, &ovsrec_monitor_col_fault);
368 ovsdb_idl_omit(idl, &ovsrec_qos_col_external_ids);
370 ovsdb_idl_omit(idl, &ovsrec_queue_col_external_ids);
372 ovsdb_idl_omit(idl, &ovsrec_mirror_col_external_ids);
374 ovsdb_idl_omit(idl, &ovsrec_netflow_col_external_ids);
376 ovsdb_idl_omit(idl, &ovsrec_sflow_col_external_ids);
378 ovsdb_idl_omit(idl, &ovsrec_manager_col_external_ids);
379 ovsdb_idl_omit(idl, &ovsrec_manager_col_inactivity_probe);
380 ovsdb_idl_omit(idl, &ovsrec_manager_col_is_connected);
381 ovsdb_idl_omit(idl, &ovsrec_manager_col_max_backoff);
382 ovsdb_idl_omit(idl, &ovsrec_manager_col_status);
384 ovsdb_idl_omit(idl, &ovsrec_ssl_col_external_ids);
386 /* Register unixctl commands. */
387 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
388 unixctl_command_register("cfm/show", cfm_unixctl_show, NULL);
389 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
390 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
392 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
401 struct bridge *br, *next_br;
403 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
406 ovsdb_idl_destroy(idl);
409 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
410 * but for which the ovs-vswitchd configuration 'cfg' is required. */
412 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
414 static bool already_configured_once;
415 struct sset bridge_names;
416 struct sset dpif_names, dpif_types;
420 /* Only do this once per ovs-vswitchd run. */
421 if (already_configured_once) {
424 already_configured_once = true;
426 stats_timer = time_msec() + STATS_INTERVAL;
428 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
429 sset_init(&bridge_names);
430 for (i = 0; i < cfg->n_bridges; i++) {
431 sset_add(&bridge_names, cfg->bridges[i]->name);
434 /* Iterate over all system dpifs and delete any of them that do not appear
436 sset_init(&dpif_names);
437 sset_init(&dpif_types);
438 dp_enumerate_types(&dpif_types);
439 SSET_FOR_EACH (type, &dpif_types) {
442 dp_enumerate_names(type, &dpif_names);
444 /* Delete each dpif whose name is not in 'bridge_names'. */
445 SSET_FOR_EACH (name, &dpif_names) {
446 if (!sset_contains(&bridge_names, name)) {
450 retval = dpif_open(name, type, &dpif);
458 sset_destroy(&bridge_names);
459 sset_destroy(&dpif_names);
460 sset_destroy(&dpif_types);
463 /* Callback for iterate_and_prune_ifaces(). */
465 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
467 if (!iface->netdev) {
468 /* We already reported a related error, don't bother duplicating it. */
472 if (iface->dp_ifidx < 0) {
473 VLOG_ERR("%s interface not in %s, dropping",
474 iface->name, dpif_name(br->dpif));
478 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
479 iface->name, iface->dp_ifidx);
483 /* Callback for iterate_and_prune_ifaces(). */
485 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
486 void *aux OVS_UNUSED)
488 /* Set policing attributes. */
489 netdev_set_policing(iface->netdev,
490 iface->cfg->ingress_policing_rate,
491 iface->cfg->ingress_policing_burst);
493 /* Set MAC address of internal interfaces other than the local
495 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
496 iface_set_mac(iface);
502 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
503 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
504 * deletes from 'br' any ports that no longer have any interfaces. */
506 iterate_and_prune_ifaces(struct bridge *br,
507 bool (*cb)(struct bridge *, struct iface *,
511 struct port *port, *next_port;
513 HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) {
514 struct iface *iface, *next_iface;
516 LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) {
517 if (!cb(br, iface, aux)) {
518 iface_set_ofport(iface->cfg, -1);
519 iface_destroy(iface);
523 if (!port->n_ifaces) {
524 VLOG_WARN("%s port has no interfaces, dropping", port->name);
530 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
531 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
532 * responsible for freeing '*managersp' (with free()).
534 * You may be asking yourself "why does ovs-vswitchd care?", because
535 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
536 * should not be and in fact is not directly involved in that. But
537 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
538 * it has to tell in-band control where the managers are to enable that.
539 * (Thus, only managers connected in-band are collected.)
542 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
543 struct sockaddr_in **managersp, size_t *n_managersp)
545 struct sockaddr_in *managers = NULL;
546 size_t n_managers = 0;
550 /* Collect all of the potential targets from the "targets" columns of the
551 * rows pointed to by "manager_options", excluding any that are
554 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
555 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
557 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
558 sset_find_and_delete(&targets, m->target);
560 sset_add(&targets, m->target);
564 /* Now extract the targets' IP addresses. */
565 if (!sset_is_empty(&targets)) {
568 managers = xmalloc(sset_count(&targets) * sizeof *managers);
569 SSET_FOR_EACH (target, &targets) {
570 struct sockaddr_in *sin = &managers[n_managers];
572 if ((!strncmp(target, "tcp:", 4)
573 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
574 (!strncmp(target, "ssl:", 4)
575 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
580 sset_destroy(&targets);
582 *managersp = managers;
583 *n_managersp = n_managers;
587 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
589 struct shash old_br, new_br;
590 struct shash_node *node;
591 struct bridge *br, *next;
592 struct sockaddr_in *managers;
595 int sflow_bridge_number;
597 COVERAGE_INC(bridge_reconfigure);
599 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
601 /* Collect old and new bridges. */
604 LIST_FOR_EACH (br, node, &all_bridges) {
605 shash_add(&old_br, br->name, br);
607 for (i = 0; i < ovs_cfg->n_bridges; i++) {
608 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
609 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
610 VLOG_WARN("more than one bridge named %s", br_cfg->name);
614 /* Get rid of deleted bridges and add new bridges. */
615 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
616 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
623 SHASH_FOR_EACH (node, &new_br) {
624 const char *br_name = node->name;
625 const struct ovsrec_bridge *br_cfg = node->data;
626 br = shash_find_data(&old_br, br_name);
628 /* If the bridge datapath type has changed, we need to tear it
629 * down and recreate. */
630 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
632 bridge_create(br_cfg);
635 bridge_create(br_cfg);
638 shash_destroy(&old_br);
639 shash_destroy(&new_br);
641 /* Reconfigure all bridges. */
642 LIST_FOR_EACH (br, node, &all_bridges) {
643 bridge_reconfigure_one(br);
646 /* Add and delete ports on all datapaths.
648 * The kernel will reject any attempt to add a given port to a datapath if
649 * that port already belongs to a different datapath, so we must do all
650 * port deletions before any port additions. */
651 LIST_FOR_EACH (br, node, &all_bridges) {
652 struct dpif_port_dump dump;
653 struct shash want_ifaces;
654 struct dpif_port dpif_port;
656 bridge_get_all_ifaces(br, &want_ifaces);
657 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
658 if (!shash_find(&want_ifaces, dpif_port.name)
659 && strcmp(dpif_port.name, br->name)) {
660 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
662 VLOG_WARN("failed to remove %s interface from %s: %s",
663 dpif_port.name, dpif_name(br->dpif),
668 shash_destroy(&want_ifaces);
670 LIST_FOR_EACH (br, node, &all_bridges) {
671 struct shash cur_ifaces, want_ifaces;
672 struct dpif_port_dump dump;
673 struct dpif_port dpif_port;
675 /* Get the set of interfaces currently in this datapath. */
676 shash_init(&cur_ifaces);
677 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
678 struct dpif_port *port_info = xmalloc(sizeof *port_info);
679 dpif_port_clone(port_info, &dpif_port);
680 shash_add(&cur_ifaces, dpif_port.name, port_info);
683 /* Get the set of interfaces we want on this datapath. */
684 bridge_get_all_ifaces(br, &want_ifaces);
686 hmap_clear(&br->ifaces);
687 SHASH_FOR_EACH (node, &want_ifaces) {
688 const char *if_name = node->name;
689 struct iface *iface = node->data;
690 struct dpif_port *dpif_port;
694 type = iface ? iface->type : "internal";
695 dpif_port = shash_find_data(&cur_ifaces, if_name);
697 /* If we have a port or a netdev already, and it's not the type we
698 * want, then delete the port (if any) and close the netdev (if
700 if ((dpif_port && strcmp(dpif_port->type, type))
701 || (iface && iface->netdev
702 && strcmp(type, netdev_get_type(iface->netdev)))) {
704 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
711 netdev_close(iface->netdev);
712 iface->netdev = NULL;
716 /* If the port doesn't exist or we don't have the netdev open,
717 * we need to do more work. */
718 if (!dpif_port || (iface && !iface->netdev)) {
719 struct netdev_options options;
720 struct netdev *netdev;
723 /* First open the network device. */
724 options.name = if_name;
726 options.args = &args;
727 options.ethertype = NETDEV_ETH_TYPE_NONE;
731 shash_from_ovs_idl_map(iface->cfg->key_options,
732 iface->cfg->value_options,
733 iface->cfg->n_options, &args);
735 error = netdev_open(&options, &netdev);
736 shash_destroy(&args);
739 VLOG_WARN("could not open network device %s (%s)",
740 if_name, strerror(error));
744 /* Then add the port if we haven't already. */
746 error = dpif_port_add(br->dpif, netdev, NULL);
748 netdev_close(netdev);
749 if (error == EFBIG) {
750 VLOG_ERR("ran out of valid port numbers on %s",
751 dpif_name(br->dpif));
754 VLOG_WARN("failed to add %s interface to %s: %s",
755 if_name, dpif_name(br->dpif),
762 /* Update 'iface'. */
764 iface->netdev = netdev;
765 iface->enabled = iface_get_carrier(iface);
766 iface->up = iface->enabled;
768 } else if (iface && iface->netdev) {
772 shash_from_ovs_idl_map(iface->cfg->key_options,
773 iface->cfg->value_options,
774 iface->cfg->n_options, &args);
775 netdev_set_config(iface->netdev, &args);
776 shash_destroy(&args);
779 shash_destroy(&want_ifaces);
781 SHASH_FOR_EACH (node, &cur_ifaces) {
782 struct dpif_port *port_info = node->data;
783 dpif_port_destroy(port_info);
786 shash_destroy(&cur_ifaces);
788 sflow_bridge_number = 0;
789 LIST_FOR_EACH (br, node, &all_bridges) {
792 struct iface *local_iface;
793 struct iface *hw_addr_iface;
796 bridge_fetch_dp_ifaces(br);
798 iterate_and_prune_ifaces(br, check_iface, NULL);
800 /* Pick local port hardware address, datapath ID. */
801 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
802 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
804 int error = netdev_set_etheraddr(local_iface->netdev, ea);
806 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
807 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
808 "Ethernet address: %s",
809 br->name, strerror(error));
812 memcpy(br->ea, ea, ETH_ADDR_LEN);
814 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
815 ofproto_set_datapath_id(br->ofproto, dpid);
817 dpid_string = xasprintf("%016"PRIx64, dpid);
818 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
821 /* Set NetFlow configuration on this bridge. */
822 if (br->cfg->netflow) {
823 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
824 struct netflow_options opts;
826 memset(&opts, 0, sizeof opts);
828 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
829 if (nf_cfg->engine_type) {
830 opts.engine_type = *nf_cfg->engine_type;
832 if (nf_cfg->engine_id) {
833 opts.engine_id = *nf_cfg->engine_id;
836 opts.active_timeout = nf_cfg->active_timeout;
837 if (!opts.active_timeout) {
838 opts.active_timeout = -1;
839 } else if (opts.active_timeout < 0) {
840 VLOG_WARN("bridge %s: active timeout interval set to negative "
841 "value, using default instead (%d seconds)", br->name,
842 NF_ACTIVE_TIMEOUT_DEFAULT);
843 opts.active_timeout = -1;
846 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
847 if (opts.add_id_to_iface) {
848 if (opts.engine_id > 0x7f) {
849 VLOG_WARN("bridge %s: netflow port mangling may conflict "
850 "with another vswitch, choose an engine id less "
851 "than 128", br->name);
853 if (hmap_count(&br->ports) > 508) {
854 VLOG_WARN("bridge %s: netflow port mangling will conflict "
855 "with another port when more than 508 ports are "
860 sset_init(&opts.collectors);
861 sset_add_array(&opts.collectors,
862 nf_cfg->targets, nf_cfg->n_targets);
863 if (ofproto_set_netflow(br->ofproto, &opts)) {
864 VLOG_ERR("bridge %s: problem setting netflow collectors",
867 sset_destroy(&opts.collectors);
869 ofproto_set_netflow(br->ofproto, NULL);
872 /* Set sFlow configuration on this bridge. */
873 if (br->cfg->sflow) {
874 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
875 struct ovsrec_controller **controllers;
876 struct ofproto_sflow_options oso;
877 size_t n_controllers;
879 memset(&oso, 0, sizeof oso);
881 sset_init(&oso.targets);
882 sset_add_array(&oso.targets,
883 sflow_cfg->targets, sflow_cfg->n_targets);
885 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
886 if (sflow_cfg->sampling) {
887 oso.sampling_rate = *sflow_cfg->sampling;
890 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
891 if (sflow_cfg->polling) {
892 oso.polling_interval = *sflow_cfg->polling;
895 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
896 if (sflow_cfg->header) {
897 oso.header_len = *sflow_cfg->header;
900 oso.sub_id = sflow_bridge_number++;
901 oso.agent_device = sflow_cfg->agent;
903 oso.control_ip = NULL;
904 n_controllers = bridge_get_controllers(br, &controllers);
905 for (i = 0; i < n_controllers; i++) {
906 if (controllers[i]->local_ip) {
907 oso.control_ip = controllers[i]->local_ip;
911 ofproto_set_sflow(br->ofproto, &oso);
913 sset_destroy(&oso.targets);
915 ofproto_set_sflow(br->ofproto, NULL);
918 /* Update the controller and related settings. It would be more
919 * straightforward to call this from bridge_reconfigure_one(), but we
920 * can't do it there for two reasons. First, and most importantly, at
921 * that point we don't know the dp_ifidx of any interfaces that have
922 * been added to the bridge (because we haven't actually added them to
923 * the datapath). Second, at that point we haven't set the datapath ID
924 * yet; when a controller is configured, resetting the datapath ID will
925 * immediately disconnect from the controller, so it's better to set
926 * the datapath ID before the controller. */
927 bridge_reconfigure_remotes(br, managers, n_managers);
929 LIST_FOR_EACH (br, node, &all_bridges) {
932 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
936 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
937 netdev_monitor_add(port->monitor, iface->netdev);
940 port->miimon_next_update = 0;
943 port_update_lacp(port);
944 port_update_bonding(port);
946 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
947 iface_update_qos(iface, port->cfg->qos);
951 LIST_FOR_EACH (br, node, &all_bridges) {
952 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
955 /* Some reconfiguration operations require the bridge to have been run at
957 LIST_FOR_EACH (br, node, &all_bridges) {
962 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
963 iface_update_cfm(iface);
969 /* ovs-vswitchd has completed initialization, so allow the process that
970 * forked us to exit successfully. */
971 daemonize_complete();
975 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
976 const struct ovsdb_idl_column *column,
979 const struct ovsdb_datum *datum;
980 union ovsdb_atom atom;
983 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
984 atom.string = (char *) key;
985 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
986 return idx == UINT_MAX ? NULL : datum->values[idx].string;
990 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
992 return get_ovsrec_key_value(&br_cfg->header_,
993 &ovsrec_bridge_col_other_config, key);
997 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
998 struct iface **hw_addr_iface)
1004 *hw_addr_iface = NULL;
1006 /* Did the user request a particular MAC? */
1007 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
1008 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
1009 if (eth_addr_is_multicast(ea)) {
1010 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
1011 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1012 } else if (eth_addr_is_zero(ea)) {
1013 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
1019 /* Otherwise choose the minimum non-local MAC address among all of the
1021 memset(ea, 0xff, ETH_ADDR_LEN);
1022 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1023 uint8_t iface_ea[ETH_ADDR_LEN];
1024 struct iface *candidate;
1025 struct iface *iface;
1027 /* Mirror output ports don't participate. */
1028 if (port->is_mirror_output_port) {
1032 /* Choose the MAC address to represent the port. */
1034 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1035 /* Find the interface with this Ethernet address (if any) so that
1036 * we can provide the correct devname to the caller. */
1037 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1038 uint8_t candidate_ea[ETH_ADDR_LEN];
1039 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1040 && eth_addr_equals(iface_ea, candidate_ea)) {
1045 /* Choose the interface whose MAC address will represent the port.
1046 * The Linux kernel bonding code always chooses the MAC address of
1047 * the first slave added to a bond, and the Fedora networking
1048 * scripts always add slaves to a bond in alphabetical order, so
1049 * for compatibility we choose the interface with the name that is
1050 * first in alphabetical order. */
1051 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1052 if (!iface || strcmp(candidate->name, iface->name) < 0) {
1057 /* The local port doesn't count (since we're trying to choose its
1058 * MAC address anyway). */
1059 if (iface->dp_ifidx == ODPP_LOCAL) {
1064 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1066 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1067 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1068 iface->name, strerror(error));
1073 /* Compare against our current choice. */
1074 if (!eth_addr_is_multicast(iface_ea) &&
1075 !eth_addr_is_local(iface_ea) &&
1076 !eth_addr_is_reserved(iface_ea) &&
1077 !eth_addr_is_zero(iface_ea) &&
1078 eth_addr_compare_3way(iface_ea, ea) < 0)
1080 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1081 *hw_addr_iface = iface;
1084 if (eth_addr_is_multicast(ea)) {
1085 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1086 *hw_addr_iface = NULL;
1087 VLOG_WARN("bridge %s: using default bridge Ethernet "
1088 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1090 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1091 br->name, ETH_ADDR_ARGS(ea));
1095 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1096 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1097 * an interface on 'br', then that interface must be passed in as
1098 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1099 * 'hw_addr_iface' must be passed in as a null pointer. */
1101 bridge_pick_datapath_id(struct bridge *br,
1102 const uint8_t bridge_ea[ETH_ADDR_LEN],
1103 struct iface *hw_addr_iface)
1106 * The procedure for choosing a bridge MAC address will, in the most
1107 * ordinary case, also choose a unique MAC that we can use as a datapath
1108 * ID. In some special cases, though, multiple bridges will end up with
1109 * the same MAC address. This is OK for the bridges, but it will confuse
1110 * the OpenFlow controller, because each datapath needs a unique datapath
1113 * Datapath IDs must be unique. It is also very desirable that they be
1114 * stable from one run to the next, so that policy set on a datapath
1117 const char *datapath_id;
1120 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1121 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1125 if (hw_addr_iface) {
1127 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1129 * A bridge whose MAC address is taken from a VLAN network device
1130 * (that is, a network device created with vconfig(8) or similar
1131 * tool) will have the same MAC address as a bridge on the VLAN
1132 * device's physical network device.
1134 * Handle this case by hashing the physical network device MAC
1135 * along with the VLAN identifier.
1137 uint8_t buf[ETH_ADDR_LEN + 2];
1138 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1139 buf[ETH_ADDR_LEN] = vlan >> 8;
1140 buf[ETH_ADDR_LEN + 1] = vlan;
1141 return dpid_from_hash(buf, sizeof buf);
1144 * Assume that this bridge's MAC address is unique, since it
1145 * doesn't fit any of the cases we handle specially.
1150 * A purely internal bridge, that is, one that has no non-virtual
1151 * network devices on it at all, is more difficult because it has no
1152 * natural unique identifier at all.
1154 * When the host is a XenServer, we handle this case by hashing the
1155 * host's UUID with the name of the bridge. Names of bridges are
1156 * persistent across XenServer reboots, although they can be reused if
1157 * an internal network is destroyed and then a new one is later
1158 * created, so this is fairly effective.
1160 * When the host is not a XenServer, we punt by using a random MAC
1161 * address on each run.
1163 const char *host_uuid = xenserver_get_host_uuid();
1165 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1166 dpid = dpid_from_hash(combined, strlen(combined));
1172 return eth_addr_to_uint64(bridge_ea);
1176 dpid_from_hash(const void *data, size_t n)
1178 uint8_t hash[SHA1_DIGEST_SIZE];
1180 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1181 sha1_bytes(data, n, hash);
1182 eth_addr_mark_random(hash);
1183 return eth_addr_to_uint64(hash);
1187 iface_refresh_status(struct iface *iface)
1191 enum netdev_flags flags;
1200 if (!netdev_get_status(iface->netdev, &sh)) {
1202 char **keys, **values;
1204 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1205 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1210 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1213 shash_destroy_free_data(&sh);
1215 error = netdev_get_flags(iface->netdev, &flags);
1217 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1220 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1223 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1225 ovsrec_interface_set_duplex(iface->cfg,
1226 netdev_features_is_full_duplex(current)
1228 /* warning: uint64_t -> int64_t conversion */
1229 bps = netdev_features_to_bps(current);
1230 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1233 ovsrec_interface_set_duplex(iface->cfg, NULL);
1234 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1238 ovsrec_interface_set_link_state(iface->cfg,
1239 iface_get_carrier(iface) ? "up" : "down");
1241 error = netdev_get_mtu(iface->netdev, &mtu);
1242 if (!error && mtu != INT_MAX) {
1244 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1247 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1251 /* Writes 'iface''s CFM statistics to the database. Returns true if anything
1252 * changed, false otherwise. */
1254 iface_refresh_cfm_stats(struct iface *iface)
1256 const struct ovsrec_monitor *mon;
1257 const struct cfm *cfm;
1258 bool changed = false;
1261 mon = iface->cfg->monitor;
1262 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1268 for (i = 0; i < mon->n_remote_mps; i++) {
1269 const struct ovsrec_maintenance_point *mp;
1270 const struct remote_mp *rmp;
1272 mp = mon->remote_mps[i];
1273 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1275 if (mp->n_fault != 1 || mp->fault[0] != rmp->fault) {
1276 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1281 if (mon->n_fault != 1 || mon->fault[0] != cfm->fault) {
1282 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1290 iface_refresh_stats(struct iface *iface)
1296 static const struct iface_stat iface_stats[] = {
1297 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1298 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1299 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1300 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1301 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1302 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1303 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1304 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1305 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1306 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1307 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1308 { "collisions", offsetof(struct netdev_stats, collisions) },
1310 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1311 const struct iface_stat *s;
1313 char *keys[N_STATS];
1314 int64_t values[N_STATS];
1317 struct netdev_stats stats;
1319 /* Intentionally ignore return value, since errors will set 'stats' to
1320 * all-1s, and we will deal with that correctly below. */
1321 netdev_get_stats(iface->netdev, &stats);
1324 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1325 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1326 if (value != UINT64_MAX) {
1333 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1337 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1339 struct ovsdb_datum datum;
1343 get_system_stats(&stats);
1345 ovsdb_datum_from_shash(&datum, &stats);
1346 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1350 static inline const char *
1351 nx_role_to_str(enum nx_role role)
1356 case NX_ROLE_MASTER:
1361 return "*** INVALID ROLE ***";
1366 bridge_refresh_controller_status(const struct bridge *br)
1369 const struct ovsrec_controller *cfg;
1371 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1373 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1374 struct ofproto_controller_info *cinfo =
1375 shash_find_data(&info, cfg->target);
1378 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1379 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1380 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1381 (char **) cinfo->pairs.values,
1384 ovsrec_controller_set_is_connected(cfg, false);
1385 ovsrec_controller_set_role(cfg, NULL);
1386 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1390 ofproto_free_ofproto_controller_info(&info);
1396 const struct ovsrec_open_vswitch *cfg;
1398 bool datapath_destroyed;
1399 bool database_changed;
1402 /* Let each bridge do the work that it needs to do. */
1403 datapath_destroyed = false;
1404 LIST_FOR_EACH (br, node, &all_bridges) {
1405 int error = bridge_run_one(br);
1407 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1408 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1409 "forcing reconfiguration", br->name);
1410 datapath_destroyed = true;
1414 /* (Re)configure if necessary. */
1415 database_changed = ovsdb_idl_run(idl);
1416 cfg = ovsrec_open_vswitch_first(idl);
1418 /* Re-configure SSL. We do this on every trip through the main loop,
1419 * instead of just when the database changes, because the contents of the
1420 * key and certificate files can change without the database changing.
1422 * We do this before bridge_reconfigure() because that function might
1423 * initiate SSL connections and thus requires SSL to be configured. */
1424 if (cfg && cfg->ssl) {
1425 const struct ovsrec_ssl *ssl = cfg->ssl;
1427 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1428 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1431 if (database_changed || datapath_destroyed) {
1433 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1435 bridge_configure_once(cfg);
1436 bridge_reconfigure(cfg);
1438 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1439 ovsdb_idl_txn_commit(txn);
1440 ovsdb_idl_txn_destroy(txn); /* XXX */
1442 /* We still need to reconfigure to avoid dangling pointers to
1443 * now-destroyed ovsrec structures inside bridge data. */
1444 static const struct ovsrec_open_vswitch null_cfg;
1446 bridge_reconfigure(&null_cfg);
1450 /* Refresh system and interface stats if necessary. */
1451 if (time_msec() >= stats_timer) {
1453 struct ovsdb_idl_txn *txn;
1455 txn = ovsdb_idl_txn_create(idl);
1456 LIST_FOR_EACH (br, node, &all_bridges) {
1459 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1460 struct iface *iface;
1462 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1463 iface_refresh_stats(iface);
1464 iface_refresh_status(iface);
1467 bridge_refresh_controller_status(br);
1469 refresh_system_stats(cfg);
1470 ovsdb_idl_txn_commit(txn);
1471 ovsdb_idl_txn_destroy(txn); /* XXX */
1474 stats_timer = time_msec() + STATS_INTERVAL;
1477 if (time_msec() >= cfm_limiter) {
1478 struct ovsdb_idl_txn *txn;
1479 bool changed = false;
1481 txn = ovsdb_idl_txn_create(idl);
1482 LIST_FOR_EACH (br, node, &all_bridges) {
1485 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1486 struct iface *iface;
1488 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1489 changed = iface_refresh_cfm_stats(iface) || changed;
1495 cfm_limiter = time_msec() + CFM_LIMIT_INTERVAL;
1498 ovsdb_idl_txn_commit(txn);
1499 ovsdb_idl_txn_destroy(txn);
1508 LIST_FOR_EACH (br, node, &all_bridges) {
1511 ofproto_wait(br->ofproto);
1512 mac_learning_wait(br->ml);
1513 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1517 ovsdb_idl_wait(idl);
1518 poll_timer_wait_until(stats_timer);
1520 if (cfm_limiter > time_msec()) {
1521 poll_timer_wait_until(cfm_limiter);
1525 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1526 * configuration changes. */
1528 bridge_flush(struct bridge *br)
1530 COVERAGE_INC(bridge_flush);
1534 /* Bridge unixctl user interface functions. */
1536 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1537 const char *args, void *aux OVS_UNUSED)
1539 struct ds ds = DS_EMPTY_INITIALIZER;
1540 const struct bridge *br;
1541 const struct mac_entry *e;
1543 br = bridge_lookup(args);
1545 unixctl_command_reply(conn, 501, "no such bridge");
1549 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1550 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1551 struct port *port = e->port.p;
1552 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1553 port_get_an_iface(port)->dp_ifidx,
1554 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1556 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1560 /* CFM unixctl user interface functions. */
1562 cfm_unixctl_show(struct unixctl_conn *conn,
1563 const char *args, void *aux OVS_UNUSED)
1565 struct ds ds = DS_EMPTY_INITIALIZER;
1566 struct iface *iface;
1567 const struct cfm *cfm;
1569 iface = iface_find(args);
1571 unixctl_command_reply(conn, 501, "no such interface");
1575 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1578 unixctl_command_reply(conn, 501, "CFM not enabled");
1582 cfm_dump_ds(cfm, &ds);
1583 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1587 /* QoS unixctl user interface functions. */
1589 struct qos_unixctl_show_cbdata {
1591 struct iface *iface;
1595 qos_unixctl_show_cb(unsigned int queue_id,
1596 const struct shash *details,
1599 struct qos_unixctl_show_cbdata *data = aux;
1600 struct ds *ds = data->ds;
1601 struct iface *iface = data->iface;
1602 struct netdev_queue_stats stats;
1603 struct shash_node *node;
1606 ds_put_cstr(ds, "\n");
1608 ds_put_format(ds, "Queue %u:\n", queue_id);
1610 ds_put_cstr(ds, "Default:\n");
1613 SHASH_FOR_EACH (node, details) {
1614 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1617 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1619 if (stats.tx_packets != UINT64_MAX) {
1620 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1623 if (stats.tx_bytes != UINT64_MAX) {
1624 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1627 if (stats.tx_errors != UINT64_MAX) {
1628 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1631 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1632 queue_id, strerror(error));
1637 qos_unixctl_show(struct unixctl_conn *conn,
1638 const char *args, void *aux OVS_UNUSED)
1640 struct ds ds = DS_EMPTY_INITIALIZER;
1641 struct shash sh = SHASH_INITIALIZER(&sh);
1642 struct iface *iface;
1644 struct shash_node *node;
1645 struct qos_unixctl_show_cbdata data;
1648 iface = iface_find(args);
1650 unixctl_command_reply(conn, 501, "no such interface");
1654 netdev_get_qos(iface->netdev, &type, &sh);
1656 if (*type != '\0') {
1657 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1659 SHASH_FOR_EACH (node, &sh) {
1660 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1665 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1668 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1670 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1672 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1673 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1676 shash_destroy_free_data(&sh);
1680 /* Bridge reconfiguration functions. */
1681 static struct bridge *
1682 bridge_create(const struct ovsrec_bridge *br_cfg)
1687 assert(!bridge_lookup(br_cfg->name));
1688 br = xzalloc(sizeof *br);
1690 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1696 dpif_flow_flush(br->dpif);
1698 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1701 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1703 dpif_delete(br->dpif);
1704 dpif_close(br->dpif);
1709 br->name = xstrdup(br_cfg->name);
1711 br->ml = mac_learning_create();
1712 eth_addr_nicira_random(br->default_ea);
1714 hmap_init(&br->ports);
1715 hmap_init(&br->ifaces);
1716 shash_init(&br->iface_by_name);
1720 list_push_back(&all_bridges, &br->node);
1722 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1728 bridge_destroy(struct bridge *br)
1731 struct port *port, *next;
1734 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1737 list_remove(&br->node);
1738 ofproto_destroy(br->ofproto);
1739 error = dpif_delete(br->dpif);
1740 if (error && error != ENOENT) {
1741 VLOG_ERR("failed to delete %s: %s",
1742 dpif_name(br->dpif), strerror(error));
1744 dpif_close(br->dpif);
1745 mac_learning_destroy(br->ml);
1746 hmap_destroy(&br->ifaces);
1747 hmap_destroy(&br->ports);
1748 shash_destroy(&br->iface_by_name);
1754 static struct bridge *
1755 bridge_lookup(const char *name)
1759 LIST_FOR_EACH (br, node, &all_bridges) {
1760 if (!strcmp(br->name, name)) {
1767 /* Handle requests for a listing of all flows known by the OpenFlow
1768 * stack, including those normally hidden. */
1770 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1771 const char *args, void *aux OVS_UNUSED)
1776 br = bridge_lookup(args);
1778 unixctl_command_reply(conn, 501, "Unknown bridge");
1783 ofproto_get_all_flows(br->ofproto, &results);
1785 unixctl_command_reply(conn, 200, ds_cstr(&results));
1786 ds_destroy(&results);
1789 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1790 * connections and reconnect. If BRIDGE is not specified, then all bridges
1791 * drop their controller connections and reconnect. */
1793 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1794 const char *args, void *aux OVS_UNUSED)
1797 if (args[0] != '\0') {
1798 br = bridge_lookup(args);
1800 unixctl_command_reply(conn, 501, "Unknown bridge");
1803 ofproto_reconnect_controllers(br->ofproto);
1805 LIST_FOR_EACH (br, node, &all_bridges) {
1806 ofproto_reconnect_controllers(br->ofproto);
1809 unixctl_command_reply(conn, 200, NULL);
1813 bridge_run_one(struct bridge *br)
1818 error = ofproto_run1(br->ofproto);
1823 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1825 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1829 error = ofproto_run2(br->ofproto, br->flush);
1836 bridge_get_controllers(const struct bridge *br,
1837 struct ovsrec_controller ***controllersp)
1839 struct ovsrec_controller **controllers;
1840 size_t n_controllers;
1842 controllers = br->cfg->controller;
1843 n_controllers = br->cfg->n_controller;
1845 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1851 *controllersp = controllers;
1853 return n_controllers;
1857 bridge_reconfigure_one(struct bridge *br)
1859 enum ofproto_fail_mode fail_mode;
1860 struct port *port, *next;
1861 struct shash_node *node;
1862 struct shash new_ports;
1865 /* Collect new ports. */
1866 shash_init(&new_ports);
1867 for (i = 0; i < br->cfg->n_ports; i++) {
1868 const char *name = br->cfg->ports[i]->name;
1869 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1870 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1875 /* If we have a controller, then we need a local port. Complain if the
1876 * user didn't specify one.
1878 * XXX perhaps we should synthesize a port ourselves in this case. */
1879 if (bridge_get_controllers(br, NULL)) {
1880 char local_name[IF_NAMESIZE];
1883 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1884 local_name, sizeof local_name);
1885 if (!error && !shash_find(&new_ports, local_name)) {
1886 VLOG_WARN("bridge %s: controller specified but no local port "
1887 "(port named %s) defined",
1888 br->name, local_name);
1892 /* Get rid of deleted ports.
1893 * Get rid of deleted interfaces on ports that still exist. */
1894 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1895 const struct ovsrec_port *port_cfg;
1897 port_cfg = shash_find_data(&new_ports, port->name);
1901 port_del_ifaces(port, port_cfg);
1905 /* Create new ports.
1906 * Add new interfaces to existing ports.
1907 * Reconfigure existing ports. */
1908 SHASH_FOR_EACH (node, &new_ports) {
1909 struct port *port = port_lookup(br, node->name);
1911 port = port_create(br, node->name);
1914 port_reconfigure(port, node->data);
1915 if (!port->n_ifaces) {
1916 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1917 br->name, port->name);
1921 shash_destroy(&new_ports);
1923 /* Set the fail-mode */
1924 fail_mode = !br->cfg->fail_mode
1925 || !strcmp(br->cfg->fail_mode, "standalone")
1926 ? OFPROTO_FAIL_STANDALONE
1927 : OFPROTO_FAIL_SECURE;
1928 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1929 && !ofproto_has_primary_controller(br->ofproto)) {
1930 ofproto_flush_flows(br->ofproto);
1932 ofproto_set_fail_mode(br->ofproto, fail_mode);
1934 /* Delete all flows if we're switching from connected to standalone or vice
1935 * versa. (XXX Should we delete all flows if we are switching from one
1936 * controller to another?) */
1938 /* Configure OpenFlow controller connection snooping. */
1939 if (!ofproto_has_snoops(br->ofproto)) {
1943 sset_add_and_free(&snoops, xasprintf("punix:%s/%s.snoop",
1944 ovs_rundir(), br->name));
1945 ofproto_set_snoops(br->ofproto, &snoops);
1946 sset_destroy(&snoops);
1949 mirror_reconfigure(br);
1952 /* Initializes 'oc' appropriately as a management service controller for
1955 * The caller must free oc->target when it is no longer needed. */
1957 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1958 struct ofproto_controller *oc)
1960 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1961 oc->max_backoff = 0;
1962 oc->probe_interval = 60;
1963 oc->band = OFPROTO_OUT_OF_BAND;
1965 oc->burst_limit = 0;
1968 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1970 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1971 struct ofproto_controller *oc)
1973 oc->target = c->target;
1974 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1975 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1976 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1977 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1978 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1979 oc->burst_limit = (c->controller_burst_limit
1980 ? *c->controller_burst_limit : 0);
1983 /* Configures the IP stack for 'br''s local interface properly according to the
1984 * configuration in 'c'. */
1986 bridge_configure_local_iface_netdev(struct bridge *br,
1987 struct ovsrec_controller *c)
1989 struct netdev *netdev;
1990 struct in_addr mask, gateway;
1992 struct iface *local_iface;
1995 /* If there's no local interface or no IP address, give up. */
1996 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
1997 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
2001 /* Bring up the local interface. */
2002 netdev = local_iface->netdev;
2003 netdev_turn_flags_on(netdev, NETDEV_UP, true);
2005 /* Configure the IP address and netmask. */
2006 if (!c->local_netmask
2007 || !inet_aton(c->local_netmask, &mask)
2009 mask.s_addr = guess_netmask(ip.s_addr);
2011 if (!netdev_set_in4(netdev, ip, mask)) {
2012 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
2013 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
2016 /* Configure the default gateway. */
2017 if (c->local_gateway
2018 && inet_aton(c->local_gateway, &gateway)
2019 && gateway.s_addr) {
2020 if (!netdev_add_router(netdev, gateway)) {
2021 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2022 br->name, IP_ARGS(&gateway.s_addr));
2028 bridge_reconfigure_remotes(struct bridge *br,
2029 const struct sockaddr_in *managers,
2032 const char *disable_ib_str, *queue_id_str;
2033 bool disable_in_band = false;
2036 struct ovsrec_controller **controllers;
2037 size_t n_controllers;
2040 struct ofproto_controller *ocs;
2044 /* Check if we should disable in-band control on this bridge. */
2045 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2046 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2047 disable_in_band = true;
2050 /* Set OpenFlow queue ID for in-band control. */
2051 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2052 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2053 ofproto_set_in_band_queue(br->ofproto, queue_id);
2055 if (disable_in_band) {
2056 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2058 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2060 had_primary = ofproto_has_primary_controller(br->ofproto);
2062 n_controllers = bridge_get_controllers(br, &controllers);
2064 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2067 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2068 for (i = 0; i < n_controllers; i++) {
2069 struct ovsrec_controller *c = controllers[i];
2071 if (!strncmp(c->target, "punix:", 6)
2072 || !strncmp(c->target, "unix:", 5)) {
2073 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2075 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2076 * domain sockets and overwriting arbitrary local files. */
2077 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2078 "\"%s\" due to possibility for remote exploit",
2079 dpif_name(br->dpif), c->target);
2083 bridge_configure_local_iface_netdev(br, c);
2084 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2085 if (disable_in_band) {
2086 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2091 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2092 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2095 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2096 ofproto_flush_flows(br->ofproto);
2099 /* If there are no controllers and the bridge is in standalone
2100 * mode, set up a flow that matches every packet and directs
2101 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2102 * switch is in secure mode and we won't pass any traffic until
2103 * a controller has been defined and it tells us to do so. */
2105 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2106 union ofp_action action;
2107 struct cls_rule rule;
2109 memset(&action, 0, sizeof action);
2110 action.type = htons(OFPAT_OUTPUT);
2111 action.output.len = htons(sizeof action);
2112 action.output.port = htons(OFPP_NORMAL);
2113 cls_rule_init_catchall(&rule, 0);
2114 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2119 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2124 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2125 struct iface *iface;
2127 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2128 shash_add_once(ifaces, iface->name, iface);
2130 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2131 shash_add_once(ifaces, port->name, NULL);
2136 /* For robustness, in case the administrator moves around datapath ports behind
2137 * our back, we re-check all the datapath port numbers here.
2139 * This function will set the 'dp_ifidx' members of interfaces that have
2140 * disappeared to -1, so only call this function from a context where those
2141 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2142 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2143 * datapath, which doesn't support UINT16_MAX+1 ports. */
2145 bridge_fetch_dp_ifaces(struct bridge *br)
2147 struct dpif_port_dump dump;
2148 struct dpif_port dpif_port;
2151 /* Reset all interface numbers. */
2152 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2153 struct iface *iface;
2155 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2156 iface->dp_ifidx = -1;
2159 hmap_clear(&br->ifaces);
2161 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2162 struct iface *iface = iface_lookup(br, dpif_port.name);
2164 if (iface->dp_ifidx >= 0) {
2165 VLOG_WARN("%s reported interface %s twice",
2166 dpif_name(br->dpif), dpif_port.name);
2167 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2168 VLOG_WARN("%s reported interface %"PRIu16" twice",
2169 dpif_name(br->dpif), dpif_port.port_no);
2171 iface->dp_ifidx = dpif_port.port_no;
2172 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2173 hash_int(iface->dp_ifidx, 0));
2176 iface_set_ofport(iface->cfg,
2177 (iface->dp_ifidx >= 0
2178 ? odp_port_to_ofp_port(iface->dp_ifidx)
2184 /* Bridge packet processing functions. */
2187 bond_is_tcp_hash(const struct port *port)
2189 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2193 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2195 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2198 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2200 struct flow hash_flow;
2202 memcpy(&hash_flow, flow, sizeof hash_flow);
2203 hash_flow.vlan_tci = 0;
2205 /* The symmetric quality of this hash function is not required, but
2206 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2207 * purposes, so we use it out of convenience. */
2208 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2211 static struct bond_entry *
2212 lookup_bond_entry(const struct port *port, const struct flow *flow,
2215 assert(port->bond_mode != BM_AB);
2217 if (bond_is_tcp_hash(port)) {
2218 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2220 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2224 static struct iface *
2225 bond_choose_iface(const struct port *port)
2227 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2228 struct iface *best_down_slave;
2229 struct iface *iface;
2231 best_down_slave = NULL;
2232 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2233 if (iface->enabled) {
2235 } else if ((!best_down_slave
2236 || iface->delay_expires < best_down_slave->delay_expires)
2237 && lacp_slave_may_enable(port->lacp, iface)) {
2238 best_down_slave = iface;
2242 if (best_down_slave) {
2243 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2244 "since no other interface is up",
2245 best_down_slave->name,
2246 best_down_slave->delay_expires - time_msec());
2247 bond_enable_slave(best_down_slave, true);
2250 return best_down_slave;
2254 choose_output_iface(const struct port *port, const struct flow *flow,
2255 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2257 struct iface *iface;
2259 assert(port->n_ifaces);
2260 if (port->n_ifaces == 1) {
2261 iface = port_get_an_iface(port);
2262 } else if (port->bond_mode == BM_AB) {
2263 iface = port->active_iface;
2265 *tags |= port->no_ifaces_tag;
2269 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2270 if (!e->iface || !e->iface->enabled) {
2271 /* XXX select interface properly. The current interface selection
2272 * is only good for testing the rebalancing code. */
2273 e->iface = bond_choose_iface(port);
2275 *tags |= port->no_ifaces_tag;
2278 e->tag = tag_create_random();
2283 *dp_ifidx = iface->dp_ifidx;
2284 *tags |= iface->tag; /* Currently only used for bonding. */
2289 bond_link_status_update(struct iface *iface)
2291 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2292 struct port *port = iface->port;
2293 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2294 int updelay, downdelay;
2296 updelay = port->updelay;
2297 downdelay = port->downdelay;
2299 if (lacp_negotiated(port->lacp)) {
2304 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2305 /* Nothing to do. */
2308 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2309 iface->name, up ? "up" : "down");
2310 if (up == iface->enabled) {
2311 iface->delay_expires = LLONG_MAX;
2312 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2313 iface->name, up ? "disabled" : "enabled");
2314 } else if (up && !port->active_iface) {
2315 bond_enable_slave(iface, true);
2317 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2318 "other interface is up", iface->name, updelay);
2321 int delay = up ? updelay : downdelay;
2322 iface->delay_expires = time_msec() + delay;
2325 "interface %s: will be %s if it stays %s for %d ms",
2327 up ? "enabled" : "disabled",
2335 bond_choose_active_iface(struct port *port)
2337 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2339 port->active_iface = bond_choose_iface(port);
2340 if (port->active_iface) {
2341 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2342 port->name, port->active_iface->name);
2344 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2350 bond_enable_slave(struct iface *iface, bool enable)
2352 struct port *port = iface->port;
2353 struct bridge *br = port->bridge;
2355 /* This acts as a recursion check. If the act of disabling a slave
2356 * causes a different slave to be enabled, the flag will allow us to
2357 * skip redundant work when we reenter this function. It must be
2358 * cleared on exit to keep things safe with multiple bonds. */
2359 static bool moving_active_iface = false;
2361 iface->delay_expires = LLONG_MAX;
2362 if (enable == iface->enabled) {
2366 iface->enabled = enable;
2367 if (!iface->enabled) {
2368 VLOG_WARN("interface %s: disabled", iface->name);
2369 ofproto_revalidate(br->ofproto, iface->tag);
2370 if (iface == port->active_iface) {
2371 /* Disabling a slave can lead to another slave being immediately
2372 * enabled if there will be no active slaves but one is waiting
2373 * on an updelay. In this case we do not need to run most of the
2374 * code for the newly enabled slave since there was no period
2375 * without an active slave and it is redundant with the disabling
2377 moving_active_iface = true;
2378 bond_choose_active_iface(port);
2380 bond_send_learning_packets(port);
2382 VLOG_WARN("interface %s: enabled", iface->name);
2383 if (!port->active_iface && !moving_active_iface) {
2384 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2385 bond_choose_active_iface(port);
2386 bond_send_learning_packets(port);
2388 iface->tag = tag_create_random();
2391 moving_active_iface = false;
2394 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2395 * bond interface. */
2397 bond_update_fake_iface_stats(struct port *port)
2399 struct netdev_stats bond_stats;
2400 struct netdev *bond_dev;
2401 struct iface *iface;
2403 memset(&bond_stats, 0, sizeof bond_stats);
2405 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2406 struct netdev_stats slave_stats;
2408 if (!netdev_get_stats(iface->netdev, &slave_stats)) {
2409 /* XXX: We swap the stats here because they are swapped back when
2410 * reported by the internal device. The reason for this is
2411 * internal devices normally represent packets going into the system
2412 * but when used as fake bond device they represent packets leaving
2413 * the system. We really should do this in the internal device
2414 * itself because changing it here reverses the counts from the
2415 * perspective of the switch. However, the internal device doesn't
2416 * know what type of device it represents so we have to do it here
2418 bond_stats.tx_packets += slave_stats.rx_packets;
2419 bond_stats.tx_bytes += slave_stats.rx_bytes;
2420 bond_stats.rx_packets += slave_stats.tx_packets;
2421 bond_stats.rx_bytes += slave_stats.tx_bytes;
2425 if (!netdev_open_default(port->name, &bond_dev)) {
2426 netdev_set_stats(bond_dev, &bond_stats);
2427 netdev_close(bond_dev);
2432 bond_run(struct port *port)
2434 struct iface *iface;
2436 if (port->n_ifaces < 2) {
2440 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2441 bond_link_status_update(iface);
2444 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2445 if (time_msec() >= iface->delay_expires) {
2446 bond_enable_slave(iface, !iface->enabled);
2450 if (port->bond_fake_iface
2451 && time_msec() >= port->bond_next_fake_iface_update) {
2452 bond_update_fake_iface_stats(port);
2453 port->bond_next_fake_iface_update = time_msec() + 1000;
2458 bond_wait(struct port *port)
2460 struct iface *iface;
2462 if (port->n_ifaces < 2) {
2466 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2467 if (iface->delay_expires != LLONG_MAX) {
2468 poll_timer_wait_until(iface->delay_expires);
2472 if (port->bond_fake_iface) {
2473 poll_timer_wait_until(port->bond_next_fake_iface_update);
2478 set_dst(struct dst *dst, const struct flow *flow,
2479 const struct port *in_port, const struct port *out_port,
2482 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2483 : in_port->vlan >= 0 ? in_port->vlan
2484 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2485 : vlan_tci_to_vid(flow->vlan_tci));
2486 return choose_output_iface(out_port, flow, dst->vlan,
2487 &dst->dp_ifidx, tags);
2491 swap_dst(struct dst *p, struct dst *q)
2493 struct dst tmp = *p;
2498 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2499 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2500 * that we push to the datapath. We could in fact fully sort the array by
2501 * vlan, but in most cases there are at most two different vlan tags so that's
2502 * possibly overkill.) */
2504 partition_dsts(struct dst_set *set, int vlan)
2506 struct dst *first = set->dsts;
2507 struct dst *last = set->dsts + set->n;
2509 while (first != last) {
2511 * - All dsts < first have vlan == 'vlan'.
2512 * - All dsts >= last have vlan != 'vlan'.
2513 * - first < last. */
2514 while (first->vlan == vlan) {
2515 if (++first == last) {
2520 /* Same invariants, plus one additional:
2521 * - first->vlan != vlan.
2523 while (last[-1].vlan != vlan) {
2524 if (--last == first) {
2529 /* Same invariants, plus one additional:
2530 * - last[-1].vlan == vlan.*/
2531 swap_dst(first++, --last);
2536 mirror_mask_ffs(mirror_mask_t mask)
2538 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2543 dst_set_init(struct dst_set *set)
2545 set->dsts = set->builtin;
2547 set->allocated = ARRAY_SIZE(set->builtin);
2551 dst_set_add(struct dst_set *set, const struct dst *dst)
2553 if (set->n >= set->allocated) {
2554 size_t new_allocated;
2555 struct dst *new_dsts;
2557 new_allocated = set->allocated * 2;
2558 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2559 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2563 set->dsts = new_dsts;
2564 set->allocated = new_allocated;
2566 set->dsts[set->n++] = *dst;
2570 dst_set_free(struct dst_set *set)
2572 if (set->dsts != set->builtin) {
2578 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2581 for (i = 0; i < set->n; i++) {
2582 if (set->dsts[i].vlan == test->vlan
2583 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2591 port_trunks_vlan(const struct port *port, uint16_t vlan)
2593 return (port->vlan < 0
2594 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2598 port_includes_vlan(const struct port *port, uint16_t vlan)
2600 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2604 port_is_floodable(const struct port *port)
2606 struct iface *iface;
2608 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2609 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2617 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2618 * there is no active iface. */
2620 port_get_active_iface_tag(const struct port *port)
2622 return (port->active_iface
2623 ? port->active_iface->tag
2624 : port->no_ifaces_tag);
2627 /* Returns an arbitrary interface within 'port'.
2629 * 'port' must have at least one interface. */
2630 static struct iface *
2631 port_get_an_iface(const struct port *port)
2633 return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem);
2637 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2638 const struct port *in_port, const struct port *out_port,
2639 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2641 mirror_mask_t mirrors = in_port->src_mirrors;
2645 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2646 if (flow_vlan == 0) {
2647 flow_vlan = OFP_VLAN_NONE;
2650 if (out_port == FLOOD_PORT) {
2653 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2655 && port_is_floodable(port)
2656 && port_includes_vlan(port, vlan)
2657 && !port->is_mirror_output_port
2658 && set_dst(&dst, flow, in_port, port, tags)) {
2659 mirrors |= port->dst_mirrors;
2660 dst_set_add(set, &dst);
2663 *nf_output_iface = NF_OUT_FLOOD;
2664 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2665 dst_set_add(set, &dst);
2666 *nf_output_iface = dst.dp_ifidx;
2667 mirrors |= out_port->dst_mirrors;
2671 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2672 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2674 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2675 && !dst_is_duplicate(set, &dst)) {
2676 dst_set_add(set, &dst);
2681 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2682 if (port_includes_vlan(port, m->out_vlan)
2683 && set_dst(&dst, flow, in_port, port, tags))
2685 if (port->vlan < 0) {
2686 dst.vlan = m->out_vlan;
2688 if (dst_is_duplicate(set, &dst)) {
2692 /* Use the vlan tag on the original flow instead of
2693 * the one passed in the vlan parameter. This ensures
2694 * that we compare the vlan from before any implicit
2695 * tagging tags place. This is necessary because
2696 * dst->vlan is the final vlan, after removing implicit
2698 if (port == in_port && dst.vlan == flow_vlan) {
2699 /* Don't send out input port on same VLAN. */
2702 dst_set_add(set, &dst);
2707 mirrors &= mirrors - 1;
2710 partition_dsts(set, flow_vlan);
2713 static void OVS_UNUSED
2714 print_dsts(const struct dst_set *set)
2718 for (i = 0; i < set->n; i++) {
2719 const struct dst *dst = &set->dsts[i];
2721 printf(">p%"PRIu16, dst->dp_ifidx);
2722 if (dst->vlan != OFP_VLAN_NONE) {
2723 printf("v%"PRIu16, dst->vlan);
2729 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2730 const struct port *in_port, const struct port *out_port,
2731 tag_type *tags, struct ofpbuf *actions,
2732 uint16_t *nf_output_iface)
2739 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2742 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2743 if (cur_vlan == 0) {
2744 cur_vlan = OFP_VLAN_NONE;
2746 for (i = 0; i < set.n; i++) {
2747 const struct dst *dst = &set.dsts[i];
2748 if (dst->vlan != cur_vlan) {
2749 if (dst->vlan == OFP_VLAN_NONE) {
2750 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2753 tci = htons(dst->vlan & VLAN_VID_MASK);
2754 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2755 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2757 cur_vlan = dst->vlan;
2759 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2764 /* Returns the effective vlan of a packet, taking into account both the
2765 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2766 * the packet is untagged and -1 indicates it has an invalid header and
2767 * should be dropped. */
2768 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2769 struct port *in_port, bool have_packet)
2771 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2772 if (in_port->vlan >= 0) {
2774 /* XXX support double tagging? */
2776 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2777 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2778 "packet received on port %s configured with "
2779 "implicit VLAN %"PRIu16,
2780 br->name, vlan, in_port->name, in_port->vlan);
2784 vlan = in_port->vlan;
2786 if (!port_includes_vlan(in_port, vlan)) {
2788 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2789 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2790 "packet received on port %s not configured for "
2792 br->name, vlan, in_port->name, vlan);
2801 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2802 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2803 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2805 is_gratuitous_arp(const struct flow *flow)
2807 return (flow->dl_type == htons(ETH_TYPE_ARP)
2808 && eth_addr_is_broadcast(flow->dl_dst)
2809 && (flow->nw_proto == ARP_OP_REPLY
2810 || (flow->nw_proto == ARP_OP_REQUEST
2811 && flow->nw_src == flow->nw_dst)));
2815 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2816 struct port *in_port)
2818 struct mac_entry *mac;
2820 if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) {
2824 mac = mac_learning_insert(br->ml, flow->dl_src, vlan);
2825 if (is_gratuitous_arp(flow)) {
2826 /* We don't want to learn from gratuitous ARP packets that are
2827 * reflected back over bond slaves so we lock the learning table. */
2828 if (in_port->n_ifaces == 1) {
2829 mac_entry_set_grat_arp_lock(mac);
2830 } else if (mac_entry_is_grat_arp_locked(mac)) {
2835 if (mac_entry_is_new(mac) || mac->port.p != in_port) {
2836 /* The log messages here could actually be useful in debugging,
2837 * so keep the rate limit relatively high. */
2838 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
2839 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2840 "on port %s in VLAN %d",
2841 br->name, ETH_ADDR_ARGS(flow->dl_src),
2842 in_port->name, vlan);
2844 mac->port.p = in_port;
2845 ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac));
2849 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2850 * dropped. Returns true if they may be forwarded, false if they should be
2853 * If 'have_packet' is true, it indicates that the caller is processing a
2854 * received packet. If 'have_packet' is false, then the caller is just
2855 * revalidating an existing flow because configuration has changed. Either
2856 * way, 'have_packet' only affects logging (there is no point in logging errors
2857 * during revalidation).
2859 * Sets '*in_portp' to the input port. This will be a null pointer if
2860 * flow->in_port does not designate a known input port (in which case
2861 * is_admissible() returns false).
2863 * When returning true, sets '*vlanp' to the effective VLAN of the input
2864 * packet, as returned by flow_get_vlan().
2866 * May also add tags to '*tags', although the current implementation only does
2867 * so in one special case.
2870 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2871 tag_type *tags, int *vlanp, struct port **in_portp)
2873 struct iface *in_iface;
2874 struct port *in_port;
2877 /* Find the interface and port structure for the received packet. */
2878 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2880 /* No interface? Something fishy... */
2882 /* Odd. A few possible reasons here:
2884 * - We deleted an interface but there are still a few packets
2885 * queued up from it.
2887 * - Someone externally added an interface (e.g. with "ovs-dpctl
2888 * add-if") that we don't know about.
2890 * - Packet arrived on the local port but the local port is not
2891 * one of our bridge ports.
2893 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2895 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2896 "interface %"PRIu16, br->name, flow->in_port);
2902 *in_portp = in_port = in_iface->port;
2903 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2908 /* Drop frames for reserved multicast addresses. */
2909 if (eth_addr_is_reserved(flow->dl_dst)) {
2913 /* Drop frames on ports reserved for mirroring. */
2914 if (in_port->is_mirror_output_port) {
2916 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2917 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2918 "%s, which is reserved exclusively for mirroring",
2919 br->name, in_port->name);
2924 /* When using LACP, do not accept packets from disabled interfaces. */
2925 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2929 /* Packets received on non-LACP bonds need special attention to avoid
2931 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2932 struct mac_entry *mac;
2934 if (eth_addr_is_multicast(flow->dl_dst)) {
2935 *tags |= port_get_active_iface_tag(in_port);
2936 if (in_port->active_iface != in_iface) {
2937 /* Drop all multicast packets on inactive slaves. */
2942 /* Drop all packets for which we have learned a different input
2943 * port, because we probably sent the packet on one slave and got
2944 * it back on the other. Gratuitous ARP packets are an exception
2945 * to this rule: the host has moved to another switch. The exception
2946 * to the exception is if we locked the learning table to avoid
2947 * reflections on bond slaves. If this is the case, just drop the
2949 mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL);
2950 if (mac && mac->port.p != in_port &&
2951 (!is_gratuitous_arp(flow) || mac_entry_is_grat_arp_locked(mac))) {
2959 /* If the composed actions may be applied to any packet in the given 'flow',
2960 * returns true. Otherwise, the actions should only be applied to 'packet', or
2961 * not at all, if 'packet' was NULL. */
2963 process_flow(struct bridge *br, const struct flow *flow,
2964 const struct ofpbuf *packet, struct ofpbuf *actions,
2965 tag_type *tags, uint16_t *nf_output_iface)
2967 struct port *in_port;
2968 struct port *out_port;
2969 struct mac_entry *mac;
2972 /* Check whether we should drop packets in this flow. */
2973 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2978 /* Learn source MAC (but don't try to learn from revalidation). */
2980 update_learning_table(br, flow, vlan, in_port);
2983 /* Determine output port. */
2984 mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags);
2986 out_port = mac->port.p;
2987 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2988 /* If we are revalidating but don't have a learning entry then
2989 * eject the flow. Installing a flow that floods packets opens
2990 * up a window of time where we could learn from a packet reflected
2991 * on a bond and blackhole packets before the learning table is
2992 * updated to reflect the correct port. */
2995 out_port = FLOOD_PORT;
2998 /* Don't send packets out their input ports. */
2999 if (in_port == out_port) {
3005 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3013 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3014 struct ofpbuf *actions, tag_type *tags,
3015 uint16_t *nf_output_iface, void *br_)
3017 struct bridge *br = br_;
3019 COVERAGE_INC(bridge_process_flow);
3020 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3024 bridge_special_ofhook_cb(const struct flow *flow,
3025 const struct ofpbuf *packet, void *br_)
3027 struct iface *iface;
3028 struct bridge *br = br_;
3030 iface = iface_from_dp_ifidx(br, flow->in_port);
3032 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3034 if (iface && iface->port->lacp && packet) {
3035 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
3038 COVERAGE_INC(bridge_process_lacp);
3039 lacp_process_pdu(iface->port->lacp, iface, pdu);
3049 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3050 const struct nlattr *actions,
3052 uint64_t n_bytes, void *br_)
3054 struct bridge *br = br_;
3055 const struct nlattr *a;
3056 struct port *in_port;
3061 /* Feed information from the active flows back into the learning table to
3062 * ensure that table is always in sync with what is actually flowing
3063 * through the datapath.
3065 * We test that 'tags' is nonzero to ensure that only flows that include an
3066 * OFPP_NORMAL action are used for learning. This works because
3067 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3068 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3069 update_learning_table(br, flow, vlan, in_port);
3072 /* Account for bond slave utilization. */
3073 if (!br->has_bonded_ports) {
3076 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3077 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3078 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3079 if (out_port && out_port->n_ifaces >= 2 &&
3080 out_port->bond_mode != BM_AB) {
3081 uint16_t vlan = (flow->vlan_tci
3082 ? vlan_tci_to_vid(flow->vlan_tci)
3084 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3085 e->tx_bytes += n_bytes;
3092 bridge_account_checkpoint_ofhook_cb(void *br_)
3094 struct bridge *br = br_;
3098 if (!br->has_bonded_ports) {
3103 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3104 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3105 && now >= port->bond_next_rebalance) {
3106 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3107 bond_rebalance_port(port);
3112 static struct ofhooks bridge_ofhooks = {
3113 bridge_normal_ofhook_cb,
3114 bridge_special_ofhook_cb,
3115 bridge_account_flow_ofhook_cb,
3116 bridge_account_checkpoint_ofhook_cb,
3119 /* Bonding functions. */
3121 /* Statistics for a single interface on a bonded port, used for load-based
3122 * bond rebalancing. */
3123 struct slave_balance {
3124 struct iface *iface; /* The interface. */
3125 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3127 /* All the "bond_entry"s that are assigned to this interface, in order of
3128 * increasing tx_bytes. */
3129 struct bond_entry **hashes;
3134 bond_mode_to_string(enum bond_mode bm) {
3135 static char *bm_slb = "balance-slb";
3136 static char *bm_ab = "active-backup";
3137 static char *bm_tcp = "balance-tcp";
3140 case BM_SLB: return bm_slb;
3141 case BM_AB: return bm_ab;
3142 case BM_TCP: return bm_tcp;
3149 /* Sorts pointers to pointers to bond_entries in ascending order by the
3150 * interface to which they are assigned, and within a single interface in
3151 * ascending order of bytes transmitted. */
3153 compare_bond_entries(const void *a_, const void *b_)
3155 const struct bond_entry *const *ap = a_;
3156 const struct bond_entry *const *bp = b_;
3157 const struct bond_entry *a = *ap;
3158 const struct bond_entry *b = *bp;
3159 if (a->iface != b->iface) {
3160 return a->iface > b->iface ? 1 : -1;
3161 } else if (a->tx_bytes != b->tx_bytes) {
3162 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3168 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3169 * *descending* order by number of bytes transmitted. */
3171 compare_slave_balance(const void *a_, const void *b_)
3173 const struct slave_balance *a = a_;
3174 const struct slave_balance *b = b_;
3175 if (a->iface->enabled != b->iface->enabled) {
3176 return a->iface->enabled ? -1 : 1;
3177 } else if (a->tx_bytes != b->tx_bytes) {
3178 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3185 swap_bals(struct slave_balance *a, struct slave_balance *b)
3187 struct slave_balance tmp = *a;
3192 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3193 * given that 'p' (and only 'p') might be in the wrong location.
3195 * This function invalidates 'p', since it might now be in a different memory
3198 resort_bals(struct slave_balance *p,
3199 struct slave_balance bals[], size_t n_bals)
3202 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3203 swap_bals(p, p - 1);
3205 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3206 swap_bals(p, p + 1);
3212 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3214 if (VLOG_IS_DBG_ENABLED()) {
3215 struct ds ds = DS_EMPTY_INITIALIZER;
3216 const struct slave_balance *b;
3218 for (b = bals; b < bals + n_bals; b++) {
3222 ds_put_char(&ds, ',');
3224 ds_put_format(&ds, " %s %"PRIu64"kB",
3225 b->iface->name, b->tx_bytes / 1024);
3227 if (!b->iface->enabled) {
3228 ds_put_cstr(&ds, " (disabled)");
3230 if (b->n_hashes > 0) {
3231 ds_put_cstr(&ds, " (");
3232 for (i = 0; i < b->n_hashes; i++) {
3233 const struct bond_entry *e = b->hashes[i];
3235 ds_put_cstr(&ds, " + ");
3237 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3238 e - port->bond_hash, e->tx_bytes / 1024);
3240 ds_put_cstr(&ds, ")");
3243 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3248 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3250 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3253 struct bond_entry *hash = from->hashes[hash_idx];
3254 struct port *port = from->iface->port;
3255 uint64_t delta = hash->tx_bytes;
3257 assert(port->bond_mode != BM_AB);
3259 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3260 "from %s to %s (now carrying %"PRIu64"kB and "
3261 "%"PRIu64"kB load, respectively)",
3262 port->name, delta / 1024, hash - port->bond_hash,
3263 from->iface->name, to->iface->name,
3264 (from->tx_bytes - delta) / 1024,
3265 (to->tx_bytes + delta) / 1024);
3267 /* Delete element from from->hashes.
3269 * We don't bother to add the element to to->hashes because not only would
3270 * it require more work, the only purpose it would be to allow that hash to
3271 * be migrated to another slave in this rebalancing run, and there is no
3272 * point in doing that. */
3273 if (hash_idx == 0) {
3276 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3277 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3281 /* Shift load away from 'from' to 'to'. */
3282 from->tx_bytes -= delta;
3283 to->tx_bytes += delta;
3285 /* Arrange for flows to be revalidated. */
3286 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3287 hash->iface = to->iface;
3288 hash->tag = tag_create_random();
3292 bond_rebalance_port(struct port *port)
3294 struct slave_balance *bals;
3296 struct bond_entry *hashes[BOND_MASK + 1];
3297 struct slave_balance *b, *from, *to;
3298 struct bond_entry *e;
3299 struct iface *iface;
3302 assert(port->bond_mode != BM_AB);
3304 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3305 * descending order of tx_bytes, so that bals[0] represents the most
3306 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3309 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3310 * array for each slave_balance structure, we sort our local array of
3311 * hashes in order by slave, so that all of the hashes for a given slave
3312 * become contiguous in memory, and then we point each 'hashes' members of
3313 * a slave_balance structure to the start of a contiguous group. */
3314 n_bals = port->n_ifaces;
3315 b = bals = xmalloc(n_bals * sizeof *bals);
3316 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3323 assert(b == &bals[n_bals]);
3324 for (i = 0; i <= BOND_MASK; i++) {
3325 hashes[i] = &port->bond_hash[i];
3327 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3328 for (i = 0; i <= BOND_MASK; i++) {
3334 for (b = bals; b < &bals[n_bals]; b++) {
3335 if (b->iface == e->iface) {
3336 b->tx_bytes += e->tx_bytes;
3338 b->hashes = &hashes[i];
3345 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3346 log_bals(bals, n_bals, port);
3348 /* Discard slaves that aren't enabled (which were sorted to the back of the
3349 * array earlier). */
3350 while (!bals[n_bals - 1].iface->enabled) {
3357 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3358 to = &bals[n_bals - 1];
3359 for (from = bals; from < to; ) {
3360 uint64_t overload = from->tx_bytes - to->tx_bytes;
3361 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3362 /* The extra load on 'from' (and all less-loaded slaves), compared
3363 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3364 * it is less than ~1Mbps. No point in rebalancing. */
3366 } else if (from->n_hashes == 1) {
3367 /* 'from' only carries a single MAC hash, so we can't shift any
3368 * load away from it, even though we want to. */
3371 /* 'from' is carrying significantly more load than 'to', and that
3372 * load is split across at least two different hashes. Pick a hash
3373 * to migrate to 'to' (the least-loaded slave), given that doing so
3374 * must decrease the ratio of the load on the two slaves by at
3377 * The sort order we use means that we prefer to shift away the
3378 * smallest hashes instead of the biggest ones. There is little
3379 * reason behind this decision; we could use the opposite sort
3380 * order to shift away big hashes ahead of small ones. */
3383 for (i = 0; i < from->n_hashes; i++) {
3384 double old_ratio, new_ratio;
3385 uint64_t delta = from->hashes[i]->tx_bytes;
3387 if (delta == 0 || from->tx_bytes - delta == 0) {
3388 /* Pointless move. */
3392 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3394 if (to->tx_bytes == 0) {
3395 /* Nothing on the new slave, move it. */
3399 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3400 new_ratio = (double)(from->tx_bytes - delta) /
3401 (to->tx_bytes + delta);
3403 if (new_ratio == 0) {
3404 /* Should already be covered but check to prevent division
3409 if (new_ratio < 1) {
3410 new_ratio = 1 / new_ratio;
3413 if (old_ratio - new_ratio > 0.1) {
3414 /* Would decrease the ratio, move it. */
3418 if (i < from->n_hashes) {
3419 bond_shift_load(from, to, i);
3421 /* If the result of the migration changed the relative order of
3422 * 'from' and 'to' swap them back to maintain invariants. */
3423 if (order_swapped) {
3424 swap_bals(from, to);
3427 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3428 * point to different slave_balance structures. It is only
3429 * valid to do these two operations in a row at all because we
3430 * know that 'from' will not move past 'to' and vice versa. */
3431 resort_bals(from, bals, n_bals);
3432 resort_bals(to, bals, n_bals);
3439 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3440 * historical data to decay to <1% in 7 rebalancing runs. */
3441 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3453 bond_send_learning_packets(struct port *port)
3455 struct bridge *br = port->bridge;
3456 struct mac_entry *e;
3457 struct ofpbuf packet;
3458 int error, n_packets, n_errors;
3460 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3464 ofpbuf_init(&packet, 128);
3465 error = n_packets = n_errors = 0;
3466 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3472 if (e->port.p == port) {
3476 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3478 flow_extract(&packet, 0, ODPP_NONE, &flow);
3480 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3486 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3492 ofpbuf_uninit(&packet);
3495 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3496 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3497 "packets, last error was: %s",
3498 port->name, n_errors, n_packets, strerror(error));
3500 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3501 port->name, n_packets);
3505 /* Bonding unixctl user interface functions. */
3508 bond_unixctl_list(struct unixctl_conn *conn,
3509 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3511 struct ds ds = DS_EMPTY_INITIALIZER;
3512 const struct bridge *br;
3514 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3516 LIST_FOR_EACH (br, node, &all_bridges) {
3519 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3520 if (port->n_ifaces > 1) {
3521 struct iface *iface;
3523 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3524 bond_mode_to_string(port->bond_mode));
3525 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3526 if (&iface->port_elem != list_front(&port->ifaces)) {
3527 ds_put_cstr(&ds, ", ");
3529 ds_put_cstr(&ds, iface->name);
3531 ds_put_char(&ds, '\n');
3535 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3539 static struct port *
3540 bond_find(const char *name)
3542 const struct bridge *br;
3544 LIST_FOR_EACH (br, node, &all_bridges) {
3547 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3548 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3557 bond_unixctl_show(struct unixctl_conn *conn,
3558 const char *args, void *aux OVS_UNUSED)
3560 struct ds ds = DS_EMPTY_INITIALIZER;
3561 const struct port *port;
3562 struct iface *iface;
3564 port = bond_find(args);
3566 unixctl_command_reply(conn, 501, "no such bond");
3570 ds_put_format(&ds, "bond_mode: %s\n",
3571 bond_mode_to_string(port->bond_mode));
3574 ds_put_format(&ds, "lacp: %s\n",
3575 port->lacp_active ? "active" : "passive");
3577 ds_put_cstr(&ds, "lacp: off\n");
3580 if (port->bond_mode != BM_AB) {
3581 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3582 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3586 ds_put_format(&ds, "bond-detect-mode: %s\n",
3587 port->monitor ? "carrier" : "miimon");
3589 if (!port->monitor) {
3590 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3591 port->miimon_interval);
3594 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3595 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3597 if (port->bond_mode != BM_AB) {
3598 ds_put_format(&ds, "next rebalance: %lld ms\n",
3599 port->bond_next_rebalance - time_msec());
3602 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3603 struct bond_entry *be;
3607 ds_put_format(&ds, "\nslave %s: %s\n",
3608 iface->name, iface->enabled ? "enabled" : "disabled");
3609 if (iface == port->active_iface) {
3610 ds_put_cstr(&ds, "\tactive slave\n");
3612 if (iface->delay_expires != LLONG_MAX) {
3613 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3614 iface->enabled ? "downdelay" : "updelay",
3615 iface->delay_expires - time_msec());
3618 if (port->bond_mode == BM_AB) {
3623 memset(&flow, 0, sizeof flow);
3624 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3625 int hash = be - port->bond_hash;
3626 struct mac_entry *me;
3628 if (be->iface != iface) {
3632 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3633 hash, be->tx_bytes / 1024);
3635 if (port->bond_mode != BM_SLB) {
3640 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3644 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3645 if (bond_hash_src(me->mac, me->vlan) == hash
3646 && me->port.p != port
3647 && choose_output_iface(port, &flow, me->vlan,
3649 && dp_ifidx == iface->dp_ifidx)
3651 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3652 ETH_ADDR_ARGS(me->mac));
3657 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3662 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3663 void *aux OVS_UNUSED)
3665 char *args = (char *) args_;
3666 char *save_ptr = NULL;
3667 char *bond_s, *hash_s, *slave_s;
3669 struct iface *iface;
3670 struct bond_entry *entry;
3673 bond_s = strtok_r(args, " ", &save_ptr);
3674 hash_s = strtok_r(NULL, " ", &save_ptr);
3675 slave_s = strtok_r(NULL, " ", &save_ptr);
3677 unixctl_command_reply(conn, 501,
3678 "usage: bond/migrate BOND HASH SLAVE");
3682 port = bond_find(bond_s);
3684 unixctl_command_reply(conn, 501, "no such bond");
3688 if (port->bond_mode != BM_SLB) {
3689 unixctl_command_reply(conn, 501, "not an SLB bond");
3693 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3694 hash = atoi(hash_s) & BOND_MASK;
3696 unixctl_command_reply(conn, 501, "bad hash");
3700 iface = port_lookup_iface(port, slave_s);
3702 unixctl_command_reply(conn, 501, "no such slave");
3706 if (!iface->enabled) {
3707 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3711 entry = &port->bond_hash[hash];
3712 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3713 entry->iface = iface;
3714 entry->tag = tag_create_random();
3715 unixctl_command_reply(conn, 200, "migrated");
3719 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3720 void *aux OVS_UNUSED)
3722 char *args = (char *) args_;
3723 char *save_ptr = NULL;
3724 char *bond_s, *slave_s;
3726 struct iface *iface;
3728 bond_s = strtok_r(args, " ", &save_ptr);
3729 slave_s = strtok_r(NULL, " ", &save_ptr);
3731 unixctl_command_reply(conn, 501,
3732 "usage: bond/set-active-slave BOND SLAVE");
3736 port = bond_find(bond_s);
3738 unixctl_command_reply(conn, 501, "no such bond");
3742 iface = port_lookup_iface(port, slave_s);
3744 unixctl_command_reply(conn, 501, "no such slave");
3748 if (!iface->enabled) {
3749 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3753 if (port->active_iface != iface) {
3754 ofproto_revalidate(port->bridge->ofproto,
3755 port_get_active_iface_tag(port));
3756 port->active_iface = iface;
3757 VLOG_INFO("port %s: active interface is now %s",
3758 port->name, iface->name);
3759 bond_send_learning_packets(port);
3760 unixctl_command_reply(conn, 200, "done");
3762 unixctl_command_reply(conn, 200, "no change");
3767 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3769 char *args = (char *) args_;
3770 char *save_ptr = NULL;
3771 char *bond_s, *slave_s;
3773 struct iface *iface;
3775 bond_s = strtok_r(args, " ", &save_ptr);
3776 slave_s = strtok_r(NULL, " ", &save_ptr);
3778 unixctl_command_reply(conn, 501,
3779 "usage: bond/enable/disable-slave BOND SLAVE");
3783 port = bond_find(bond_s);
3785 unixctl_command_reply(conn, 501, "no such bond");
3789 iface = port_lookup_iface(port, slave_s);
3791 unixctl_command_reply(conn, 501, "no such slave");
3795 bond_enable_slave(iface, enable);
3796 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3800 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3801 void *aux OVS_UNUSED)
3803 enable_slave(conn, args, true);
3807 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3808 void *aux OVS_UNUSED)
3810 enable_slave(conn, args, false);
3814 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3815 void *aux OVS_UNUSED)
3817 char *args = (char *) args_;
3818 uint8_t mac[ETH_ADDR_LEN];
3822 char *mac_s, *vlan_s;
3823 char *save_ptr = NULL;
3825 mac_s = strtok_r(args, " ", &save_ptr);
3826 vlan_s = strtok_r(NULL, " ", &save_ptr);
3829 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3830 unixctl_command_reply(conn, 501, "invalid vlan");
3834 vlan = OFP_VLAN_NONE;
3837 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3838 == ETH_ADDR_SCAN_COUNT) {
3839 hash = bond_hash_src(mac, vlan);
3841 hash_cstr = xasprintf("%u", hash);
3842 unixctl_command_reply(conn, 200, hash_cstr);
3845 unixctl_command_reply(conn, 501, "invalid mac");
3852 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3853 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3854 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3855 unixctl_command_register("bond/set-active-slave",
3856 bond_unixctl_set_active_slave, NULL);
3857 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3859 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3861 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3864 /* Port functions. */
3867 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3869 struct iface *iface = aux;
3870 uint8_t ea[ETH_ADDR_LEN];
3873 error = netdev_get_etheraddr(iface->netdev, ea);
3875 struct ofpbuf packet;
3876 struct lacp_pdu *packet_pdu;
3878 ofpbuf_init(&packet, 0);
3879 packet_pdu = compose_packet(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3880 sizeof *packet_pdu);
3881 memcpy(packet_pdu, pdu, sizeof *packet_pdu);
3882 ofproto_send_packet(iface->port->bridge->ofproto,
3883 iface->dp_ifidx, 0, &packet);
3884 ofpbuf_uninit(&packet);
3886 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3887 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3888 "(%s)", iface->name, strerror(error));
3893 port_run(struct port *port)
3895 if (port->monitor) {
3898 /* Track carrier going up and down on interfaces. */
3899 while (!netdev_monitor_poll(port->monitor, &devname)) {
3900 struct iface *iface;
3902 iface = port_lookup_iface(port, devname);
3904 iface_update_carrier(iface);
3908 } else if (time_msec() >= port->miimon_next_update) {
3909 struct iface *iface;
3911 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3912 iface_update_carrier(iface);
3914 port->miimon_next_update = time_msec() + port->miimon_interval;
3918 struct iface *iface;
3920 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3921 lacp_slave_enable(port->lacp, iface, iface->enabled);
3924 lacp_run(port->lacp, lacp_send_pdu_cb);
3931 port_wait(struct port *port)
3933 if (port->monitor) {
3934 netdev_monitor_poll_wait(port->monitor);
3936 poll_timer_wait_until(port->miimon_next_update);
3940 lacp_wait(port->lacp);
3946 static struct port *
3947 port_create(struct bridge *br, const char *name)
3951 port = xzalloc(sizeof *port);
3954 port->trunks = NULL;
3955 port->name = xstrdup(name);
3956 port->active_iface = NULL;
3957 list_init(&port->ifaces);
3959 hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0));
3961 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3968 get_port_other_config(const struct ovsrec_port *port, const char *key,
3969 const char *default_value)
3973 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3975 return value ? value : default_value;
3979 get_interface_other_config(const struct ovsrec_interface *iface,
3980 const char *key, const char *default_value)
3984 value = get_ovsrec_key_value(&iface->header_,
3985 &ovsrec_interface_col_other_config, key);
3986 return value ? value : default_value;
3990 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3992 struct iface *iface, *next;
3993 struct sset new_ifaces;
3996 /* Collect list of new interfaces. */
3997 sset_init(&new_ifaces);
3998 for (i = 0; i < cfg->n_interfaces; i++) {
3999 const char *name = cfg->interfaces[i]->name;
4000 sset_add(&new_ifaces, name);
4003 /* Get rid of deleted interfaces. */
4004 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4005 if (!sset_contains(&new_ifaces, iface->name)) {
4006 iface_destroy(iface);
4010 sset_destroy(&new_ifaces);
4013 /* Expires all MAC learning entries associated with 'port' and forces ofproto
4014 * to revalidate every flow. */
4016 port_flush_macs(struct port *port)
4018 struct bridge *br = port->bridge;
4019 struct mac_learning *ml = br->ml;
4020 struct mac_entry *mac, *next_mac;
4023 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
4024 if (mac->port.p == port) {
4025 mac_learning_expire(ml, mac);
4031 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4033 const char *detect_mode;
4034 struct sset new_ifaces;
4035 long long int next_rebalance, miimon_next_update, lacp_priority;
4036 bool need_flush = false;
4037 unsigned long *trunks;
4043 /* Update settings. */
4044 port->updelay = cfg->bond_updelay;
4045 if (port->updelay < 0) {
4048 port->downdelay = cfg->bond_downdelay;
4049 if (port->downdelay < 0) {
4050 port->downdelay = 0;
4052 port->bond_rebalance_interval = atoi(
4053 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4054 if (port->bond_rebalance_interval < 1000) {
4055 port->bond_rebalance_interval = 1000;
4057 next_rebalance = time_msec() + port->bond_rebalance_interval;
4058 if (port->bond_next_rebalance > next_rebalance) {
4059 port->bond_next_rebalance = next_rebalance;
4062 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4065 netdev_monitor_destroy(port->monitor);
4066 port->monitor = NULL;
4068 if (strcmp(detect_mode, "miimon")) {
4069 port->monitor = netdev_monitor_create();
4071 if (strcmp(detect_mode, "carrier")) {
4072 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4073 "defaulting to carrier", port->name, detect_mode);
4077 port->miimon_interval = atoi(
4078 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4079 if (port->miimon_interval < 100) {
4080 port->miimon_interval = 100;
4082 miimon_next_update = time_msec() + port->miimon_interval;
4083 if (port->miimon_next_update > miimon_next_update) {
4084 port->miimon_next_update = miimon_next_update;
4087 if (!port->cfg->bond_mode ||
4088 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4089 port->bond_mode = BM_SLB;
4090 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4091 port->bond_mode = BM_AB;
4092 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4093 port->bond_mode = BM_TCP;
4095 port->bond_mode = BM_SLB;
4096 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4097 port->name, port->cfg->bond_mode,
4098 bond_mode_to_string(port->bond_mode));
4101 /* Add new interfaces and update 'cfg' member of existing ones. */
4102 sset_init(&new_ifaces);
4103 for (i = 0; i < cfg->n_interfaces; i++) {
4104 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4105 struct iface *iface;
4107 if (!sset_add(&new_ifaces, if_cfg->name)) {
4108 VLOG_WARN("port %s: %s specified twice as port interface",
4109 port->name, if_cfg->name);
4110 iface_set_ofport(if_cfg, -1);
4114 iface = iface_lookup(port->bridge, if_cfg->name);
4116 if (iface->port != port) {
4117 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4119 port->bridge->name, if_cfg->name, iface->port->name);
4122 iface->cfg = if_cfg;
4124 iface = iface_create(port, if_cfg);
4127 /* Determine interface type. The local port always has type
4128 * "internal". Other ports take their type from the database and
4129 * default to "system" if none is specified. */
4130 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4131 : if_cfg->type[0] ? if_cfg->type
4135 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4138 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4139 iface->lacp_priority = UINT16_MAX;
4141 iface->lacp_priority = lacp_priority;
4144 sset_destroy(&new_ifaces);
4146 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4150 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4152 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4153 /* Prefer bondable links if unspecified. */
4154 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4156 port->lacp_priority = lacp_priority;
4159 if (!port->cfg->lacp) {
4160 /* XXX when LACP implementation has been sufficiently tested, enable by
4161 * default and make active on bonded ports. */
4162 lacp_destroy(port->lacp);
4164 } else if (!strcmp(port->cfg->lacp, "off")) {
4165 lacp_destroy(port->lacp);
4167 } else if (!strcmp(port->cfg->lacp, "active")) {
4169 port->lacp = lacp_create();
4171 port->lacp_active = true;
4172 } else if (!strcmp(port->cfg->lacp, "passive")) {
4174 port->lacp = lacp_create();
4176 port->lacp_active = false;
4178 VLOG_WARN("port %s: unknown LACP mode %s",
4179 port->name, port->cfg->lacp);
4180 lacp_destroy(port->lacp);
4187 if (port->n_ifaces < 2) {
4189 if (vlan >= 0 && vlan <= 4095) {
4190 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4195 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4196 * they even work as-is. But they have not been tested. */
4197 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4201 if (port->vlan != vlan) {
4206 /* Get trunked VLANs. */
4208 if (vlan < 0 && cfg->n_trunks) {
4211 trunks = bitmap_allocate(4096);
4213 for (i = 0; i < cfg->n_trunks; i++) {
4214 int trunk = cfg->trunks[i];
4216 bitmap_set1(trunks, trunk);
4222 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4223 port->name, cfg->n_trunks);
4225 if (n_errors == cfg->n_trunks) {
4226 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4228 bitmap_free(trunks);
4231 } else if (vlan >= 0 && cfg->n_trunks) {
4232 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4236 ? port->trunks != NULL
4237 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4240 bitmap_free(port->trunks);
4241 port->trunks = trunks;
4244 port_flush_macs(port);
4249 port_destroy(struct port *port)
4252 struct bridge *br = port->bridge;
4253 struct iface *iface, *next;
4256 for (i = 0; i < MAX_MIRRORS; i++) {
4257 struct mirror *m = br->mirrors[i];
4258 if (m && m->out_port == port) {
4263 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4264 iface_destroy(iface);
4267 hmap_remove(&br->ports, &port->hmap_node);
4269 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4271 port_flush_macs(port);
4273 lacp_destroy(port->lacp);
4274 netdev_monitor_destroy(port->monitor);
4275 bitmap_free(port->trunks);
4276 free(port->bond_hash);
4282 static struct port *
4283 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4285 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4286 return iface ? iface->port : NULL;
4289 static struct port *
4290 port_lookup(const struct bridge *br, const char *name)
4294 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0),
4296 if (!strcmp(port->name, name)) {
4303 static struct iface *
4304 port_lookup_iface(const struct port *port, const char *name)
4306 struct iface *iface = iface_lookup(port->bridge, name);
4307 return iface && iface->port == port ? iface : NULL;
4311 port_update_lacp(struct port *port)
4314 struct iface *iface;
4316 lacp_configure(port->lacp, port->name,
4317 port->bridge->ea, port->lacp_priority,
4318 port->lacp_active, port->lacp_fast);
4320 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
4321 lacp_slave_register(port->lacp, iface, iface->name,
4322 iface->dp_ifidx, iface->lacp_priority);
4328 port_update_bonding(struct port *port)
4330 if (port->n_ifaces < 2) {
4331 /* Not a bonded port. */
4332 free(port->bond_hash);
4333 port->bond_hash = NULL;
4334 port->bond_fake_iface = false;
4335 port->active_iface = NULL;
4336 port->no_ifaces_tag = 0;
4340 if (port->bond_mode != BM_AB && !port->bond_hash) {
4341 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4342 for (i = 0; i <= BOND_MASK; i++) {
4343 struct bond_entry *e = &port->bond_hash[i];
4347 port->bond_next_rebalance
4348 = time_msec() + port->bond_rebalance_interval;
4349 } else if (port->bond_mode == BM_AB) {
4350 free(port->bond_hash);
4351 port->bond_hash = NULL;
4354 if (!port->no_ifaces_tag) {
4355 port->no_ifaces_tag = tag_create_random();
4358 if (!port->active_iface) {
4359 bond_choose_active_iface(port);
4362 port->bond_fake_iface = port->cfg->bond_fake_iface;
4363 if (port->bond_fake_iface) {
4364 port->bond_next_fake_iface_update = time_msec();
4370 /* Interface functions. */
4372 static struct iface *
4373 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4375 struct bridge *br = port->bridge;
4376 struct iface *iface;
4377 char *name = if_cfg->name;
4379 iface = xzalloc(sizeof *iface);
4381 iface->name = xstrdup(name);
4382 iface->dp_ifidx = -1;
4383 iface->tag = tag_create_random();
4384 iface->delay_expires = LLONG_MAX;
4385 iface->netdev = NULL;
4386 iface->cfg = if_cfg;
4388 shash_add_assert(&br->iface_by_name, iface->name, iface);
4390 list_push_back(&port->ifaces, &iface->port_elem);
4393 if (port->n_ifaces > 1) {
4394 br->has_bonded_ports = true;
4397 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4405 iface_destroy(struct iface *iface)
4408 struct port *port = iface->port;
4409 struct bridge *br = port->bridge;
4410 bool del_active = port->active_iface == iface;
4412 if (port->bond_hash) {
4413 struct bond_entry *e;
4414 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4415 if (e->iface == iface) {
4421 if (iface->port->lacp) {
4422 lacp_slave_unregister(iface->port->lacp, iface);
4425 if (port->monitor && iface->netdev) {
4426 netdev_monitor_remove(port->monitor, iface->netdev);
4429 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4431 if (iface->dp_ifidx >= 0) {
4432 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4435 list_remove(&iface->port_elem);
4438 netdev_close(iface->netdev);
4441 bond_choose_active_iface(port);
4442 bond_send_learning_packets(port);
4448 bridge_flush(port->bridge);
4452 static struct iface *
4453 iface_lookup(const struct bridge *br, const char *name)
4455 return shash_find_data(&br->iface_by_name, name);
4458 static struct iface *
4459 iface_find(const char *name)
4461 const struct bridge *br;
4463 LIST_FOR_EACH (br, node, &all_bridges) {
4464 struct iface *iface = iface_lookup(br, name);
4473 static struct iface *
4474 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4476 struct iface *iface;
4478 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4479 hash_int(dp_ifidx, 0), &br->ifaces) {
4480 if (iface->dp_ifidx == dp_ifidx) {
4487 /* Set Ethernet address of 'iface', if one is specified in the configuration
4490 iface_set_mac(struct iface *iface)
4492 uint8_t ea[ETH_ADDR_LEN];
4494 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4495 if (eth_addr_is_multicast(ea)) {
4496 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4498 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4499 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4500 iface->name, iface->name);
4502 int error = netdev_set_etheraddr(iface->netdev, ea);
4504 VLOG_ERR("interface %s: setting MAC failed (%s)",
4505 iface->name, strerror(error));
4511 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4513 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4516 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4520 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4522 * The value strings in '*shash' are taken directly from values[], not copied,
4523 * so the caller should not modify or free them. */
4525 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4526 struct shash *shash)
4531 for (i = 0; i < n; i++) {
4532 shash_add(shash, keys[i], values[i]);
4536 /* Creates 'keys' and 'values' arrays from 'shash'.
4538 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4539 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4540 * are populated with with strings taken directly from 'shash' and thus have
4541 * the same ownership of the key-value pairs in shash.
4544 shash_to_ovs_idl_map(struct shash *shash,
4545 char ***keys, char ***values, size_t *n)
4549 struct shash_node *sn;
4551 count = shash_count(shash);
4553 k = xmalloc(count * sizeof *k);
4554 v = xmalloc(count * sizeof *v);
4557 SHASH_FOR_EACH(sn, shash) {
4568 struct iface_delete_queues_cbdata {
4569 struct netdev *netdev;
4570 const struct ovsdb_datum *queues;
4574 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4576 union ovsdb_atom atom;
4578 atom.integer = target;
4579 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4583 iface_delete_queues(unsigned int queue_id,
4584 const struct shash *details OVS_UNUSED, void *cbdata_)
4586 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4588 if (!queue_ids_include(cbdata->queues, queue_id)) {
4589 netdev_delete_queue(cbdata->netdev, queue_id);
4594 iface_update_carrier(struct iface *iface)
4596 bool carrier = iface_get_carrier(iface);
4597 if (carrier == iface->up) {
4601 iface->up = carrier;
4602 if (iface->port->lacp) {
4603 lacp_slave_carrier_changed(iface->port->lacp, iface);
4608 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4610 if (!qos || qos->type[0] == '\0') {
4611 netdev_set_qos(iface->netdev, NULL, NULL);
4613 struct iface_delete_queues_cbdata cbdata;
4614 struct shash details;
4617 /* Configure top-level Qos for 'iface'. */
4618 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4619 qos->n_other_config, &details);
4620 netdev_set_qos(iface->netdev, qos->type, &details);
4621 shash_destroy(&details);
4623 /* Deconfigure queues that were deleted. */
4624 cbdata.netdev = iface->netdev;
4625 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4627 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4629 /* Configure queues for 'iface'. */
4630 for (i = 0; i < qos->n_queues; i++) {
4631 const struct ovsrec_queue *queue = qos->value_queues[i];
4632 unsigned int queue_id = qos->key_queues[i];
4634 shash_from_ovs_idl_map(queue->key_other_config,
4635 queue->value_other_config,
4636 queue->n_other_config, &details);
4637 netdev_set_queue(iface->netdev, queue_id, &details);
4638 shash_destroy(&details);
4644 iface_update_cfm(struct iface *iface)
4648 uint16_t *remote_mps;
4649 struct ovsrec_monitor *mon;
4650 uint8_t maid[CCM_MAID_LEN];
4652 mon = iface->cfg->monitor;
4655 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4659 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4660 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4664 cfm.mpid = mon->mpid;
4665 cfm.interval = mon->interval ? *mon->interval : 1000;
4667 memcpy(cfm.maid, maid, sizeof cfm.maid);
4669 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4670 for(i = 0; i < mon->n_remote_mps; i++) {
4671 remote_mps[i] = mon->remote_mps[i]->mpid;
4674 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4675 &cfm, remote_mps, mon->n_remote_mps);
4679 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4680 * how 'iface''s port is configured.
4682 * Returns true if 'iface' is up, false otherwise. */
4684 iface_get_carrier(const struct iface *iface)
4686 return (iface->port->monitor
4687 ? netdev_get_carrier(iface->netdev)
4688 : netdev_get_miimon(iface->netdev));
4691 /* Port mirroring. */
4693 static struct mirror *
4694 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4698 for (i = 0; i < MAX_MIRRORS; i++) {
4699 struct mirror *m = br->mirrors[i];
4700 if (m && uuid_equals(uuid, &m->uuid)) {
4708 mirror_reconfigure(struct bridge *br)
4710 unsigned long *rspan_vlans;
4714 /* Get rid of deleted mirrors. */
4715 for (i = 0; i < MAX_MIRRORS; i++) {
4716 struct mirror *m = br->mirrors[i];
4718 const struct ovsdb_datum *mc;
4719 union ovsdb_atom atom;
4721 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4722 atom.uuid = br->mirrors[i]->uuid;
4723 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4729 /* Add new mirrors and reconfigure existing ones. */
4730 for (i = 0; i < br->cfg->n_mirrors; i++) {
4731 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4732 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4734 mirror_reconfigure_one(m, cfg);
4736 mirror_create(br, cfg);
4740 /* Update port reserved status. */
4741 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4742 port->is_mirror_output_port = false;
4744 for (i = 0; i < MAX_MIRRORS; i++) {
4745 struct mirror *m = br->mirrors[i];
4746 if (m && m->out_port) {
4747 m->out_port->is_mirror_output_port = true;
4751 /* Update flooded vlans (for RSPAN). */
4753 if (br->cfg->n_flood_vlans) {
4754 rspan_vlans = bitmap_allocate(4096);
4756 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4757 int64_t vlan = br->cfg->flood_vlans[i];
4758 if (vlan >= 0 && vlan < 4096) {
4759 bitmap_set1(rspan_vlans, vlan);
4760 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4763 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4768 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4770 mac_learning_flush(br->ml);
4775 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4780 for (i = 0; ; i++) {
4781 if (i >= MAX_MIRRORS) {
4782 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4783 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4786 if (!br->mirrors[i]) {
4791 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4793 mac_learning_flush(br->ml);
4795 br->mirrors[i] = m = xzalloc(sizeof *m);
4798 m->name = xstrdup(cfg->name);
4799 sset_init(&m->src_ports);
4800 sset_init(&m->dst_ports);
4806 mirror_reconfigure_one(m, cfg);
4810 mirror_destroy(struct mirror *m)
4813 struct bridge *br = m->bridge;
4816 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4817 port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4818 port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4821 sset_destroy(&m->src_ports);
4822 sset_destroy(&m->dst_ports);
4825 m->bridge->mirrors[m->idx] = NULL;
4830 mac_learning_flush(br->ml);
4835 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4840 for (i = 0; i < n_ports; i++) {
4841 const char *name = ports[i]->name;
4842 if (port_lookup(m->bridge, name)) {
4843 sset_add(names, name);
4845 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4846 "port %s", m->bridge->name, m->name, name);
4852 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4858 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4860 for (i = 0; i < cfg->n_select_vlan; i++) {
4861 int64_t vlan = cfg->select_vlan[i];
4862 if (vlan < 0 || vlan > 4095) {
4863 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4864 m->bridge->name, m->name, vlan);
4866 (*vlans)[n_vlans++] = vlan;
4873 vlan_is_mirrored(const struct mirror *m, int vlan)
4877 for (i = 0; i < m->n_vlans; i++) {
4878 if (m->vlans[i] == vlan) {
4886 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4890 for (i = 0; i < m->n_vlans; i++) {
4891 if (port_trunks_vlan(p, m->vlans[i])) {
4899 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4901 struct sset src_ports, dst_ports;
4902 mirror_mask_t mirror_bit;
4903 struct port *out_port;
4910 if (strcmp(cfg->name, m->name)) {
4912 m->name = xstrdup(cfg->name);
4915 /* Get output port. */
4916 if (cfg->output_port) {
4917 out_port = port_lookup(m->bridge, cfg->output_port->name);
4919 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4920 m->bridge->name, m->name);
4926 if (cfg->output_vlan) {
4927 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4928 "output vlan; ignoring output vlan",
4929 m->bridge->name, m->name);
4931 } else if (cfg->output_vlan) {
4933 out_vlan = *cfg->output_vlan;
4935 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4936 m->bridge->name, m->name);
4941 sset_init(&src_ports);
4942 sset_init(&dst_ports);
4943 if (cfg->select_all) {
4944 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4945 sset_add(&src_ports, port->name);
4946 sset_add(&dst_ports, port->name);
4951 /* Get ports, and drop duplicates and ports that don't exist. */
4952 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4954 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4957 /* Get all the vlans, and drop duplicate and invalid vlans. */
4958 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4961 /* Update mirror data. */
4962 if (!sset_equals(&m->src_ports, &src_ports)
4963 || !sset_equals(&m->dst_ports, &dst_ports)
4964 || m->n_vlans != n_vlans
4965 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4966 || m->out_port != out_port
4967 || m->out_vlan != out_vlan) {
4968 bridge_flush(m->bridge);
4969 mac_learning_flush(m->bridge->ml);
4971 sset_swap(&m->src_ports, &src_ports);
4972 sset_swap(&m->dst_ports, &dst_ports);
4975 m->n_vlans = n_vlans;
4976 m->out_port = out_port;
4977 m->out_vlan = out_vlan;
4980 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4981 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4982 if (sset_contains(&m->src_ports, port->name)
4985 ? port_trunks_any_mirrored_vlan(m, port)
4986 : vlan_is_mirrored(m, port->vlan)))) {
4987 port->src_mirrors |= mirror_bit;
4989 port->src_mirrors &= ~mirror_bit;
4992 if (sset_contains(&m->dst_ports, port->name)) {
4993 port->dst_mirrors |= mirror_bit;
4995 port->dst_mirrors &= ~mirror_bit;
5000 sset_destroy(&src_ports);
5001 sset_destroy(&dst_ports);