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"
40 #include "dynamic-string.h"
46 #include "mac-learning.h"
50 #include "ofp-print.h"
52 #include "ofproto/netflow.h"
53 #include "ofproto/ofproto.h"
54 #include "ovsdb-data.h"
56 #include "poll-loop.h"
57 #include "proc-net-compat.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_reconfigure);
79 COVERAGE_DEFINE(bridge_lacp_update);
82 LACP_STATUS_CURRENT, /* Partner is up to date. */
83 LACP_STATUS_EXPIRED, /* Partner is out of date. Attempt to re-sync. */
84 LACP_STATUS_DEFAULTED /* Partner information is unknown. */
93 struct dst builtin[32];
98 static void dst_set_init(struct dst_set *);
99 static void dst_set_add(struct dst_set *, const struct dst *);
100 static void dst_set_free(struct dst_set *);
103 /* These members are always valid. */
104 struct port *port; /* Containing port. */
105 size_t port_ifidx; /* Index within containing port. */
106 char *name; /* Host network device name. */
107 tag_type tag; /* Tag associated with this interface. */
108 long long delay_expires; /* Time after which 'enabled' may change. */
110 /* These members are valid only after bridge_reconfigure() causes them to
112 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
113 int dp_ifidx; /* Index within kernel datapath. */
114 struct netdev *netdev; /* Network device. */
115 bool enabled; /* May be chosen for flows? */
116 bool up; /* Is the interface up? */
117 const char *type; /* Usually same as cfg->type. */
118 struct cfm *cfm; /* Connectivity Fault Management */
119 const struct ovsrec_interface *cfg;
121 /* LACP information. */
122 enum lacp_status lacp_status; /* LACP state machine status. */
123 uint16_t lacp_priority; /* LACP port priority. */
124 struct lacp_info lacp_actor; /* LACP actor information. */
125 struct lacp_info lacp_partner; /* LACP partner information. */
126 long long int lacp_tx; /* Next LACP message transmission time. */
127 long long int lacp_rx; /* Next LACP message receive time. */
128 bool lacp_attached; /* Attached to its aggregator? LACP allows
129 this link to be chosen for flows. */
132 #define BOND_MASK 0xff
134 int iface_idx; /* Index of assigned iface, or -1 if none. */
135 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
136 tag_type iface_tag; /* Tag associated with iface_idx. */
140 BM_TCP, /* Transport Layer Load Balance. */
141 BM_SLB, /* Source Load Balance. */
142 BM_AB /* Active Backup. */
145 #define MAX_MIRRORS 32
146 typedef uint32_t mirror_mask_t;
147 #define MIRROR_MASK_C(X) UINT32_C(X)
148 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
150 struct bridge *bridge;
153 struct uuid uuid; /* UUID of this "mirror" record in database. */
155 /* Selection criteria. */
156 struct shash src_ports; /* Name is port name; data is always NULL. */
157 struct shash dst_ports; /* Name is port name; data is always NULL. */
162 struct port *out_port;
166 /* Flags for a port's lacp member. */
167 #define LACP_ACTIVE 0x01 /* LACP is in active mode. */
168 #define LACP_PASSIVE 0x02 /* LACP is in passive mode. */
169 #define LACP_NEGOTIATED 0x04 /* LACP has successfully negotiated. */
171 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
173 struct bridge *bridge;
175 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
176 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
177 * NULL if all VLANs are trunked. */
178 const struct ovsrec_port *cfg;
181 /* An ordinary bridge port has 1 interface.
182 * A bridge port for bonding has at least 2 interfaces. */
183 struct iface **ifaces;
184 size_t n_ifaces, allocated_ifaces;
187 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
188 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
189 tag_type active_iface_tag; /* Tag for bcast flows. */
190 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
191 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
192 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
193 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
194 bool miimon; /* Use miimon instead of carrier? */
195 long long int bond_miimon_interval; /* Miimon status refresh interval. */
196 long long int bond_miimon_next_update; /* Time of next miimon update. */
197 long long int bond_next_fake_iface_update; /* Time of next update. */
198 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
200 /* LACP information. */
201 int lacp; /* LACP status flags. 0 if LACP is off. */
202 uint16_t lacp_key; /* LACP aggregation key. */
203 uint16_t lacp_priority; /* LACP system priority. */
204 bool lacp_need_update; /* Need to update attached interfaces? */
206 /* SLB specific bonding info. */
207 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
208 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
209 long long int bond_next_rebalance; /* Next rebalancing time. */
211 /* Port mirroring info. */
212 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
213 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
214 bool is_mirror_output_port; /* Does port mirroring send frames here? */
218 struct list node; /* Node in global list of bridges. */
219 char *name; /* User-specified arbitrary name. */
220 struct mac_learning *ml; /* MAC learning table. */
221 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
222 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
223 const struct ovsrec_bridge *cfg;
225 /* OpenFlow switch processing. */
226 struct ofproto *ofproto; /* OpenFlow switch. */
228 /* Kernel datapath information. */
229 struct dpif *dpif; /* Datapath. */
230 struct hmap ifaces; /* Contains "struct iface"s. */
234 size_t n_ports, allocated_ports;
235 struct shash iface_by_name; /* "struct iface"s indexed by name. */
236 struct shash port_by_name; /* "struct port"s indexed by name. */
239 bool has_bonded_ports;
244 /* Port mirroring. */
245 struct mirror *mirrors[MAX_MIRRORS];
248 /* List of all bridges. */
249 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
251 /* OVSDB IDL used to obtain configuration. */
252 static struct ovsdb_idl *idl;
254 /* Each time this timer expires, the bridge fetches systems and interface
255 * statistics and pushes them into the database. */
256 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
257 static long long int stats_timer = LLONG_MIN;
259 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
260 static void bridge_destroy(struct bridge *);
261 static struct bridge *bridge_lookup(const char *name);
262 static unixctl_cb_func bridge_unixctl_dump_flows;
263 static unixctl_cb_func bridge_unixctl_reconnect;
264 static int bridge_run_one(struct bridge *);
265 static size_t bridge_get_controllers(const struct bridge *br,
266 struct ovsrec_controller ***controllersp);
267 static void bridge_reconfigure_one(struct bridge *);
268 static void bridge_reconfigure_remotes(struct bridge *,
269 const struct sockaddr_in *managers,
271 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
272 static void bridge_fetch_dp_ifaces(struct bridge *);
273 static void bridge_flush(struct bridge *);
274 static void bridge_pick_local_hw_addr(struct bridge *,
275 uint8_t ea[ETH_ADDR_LEN],
276 struct iface **hw_addr_iface);
277 static uint64_t bridge_pick_datapath_id(struct bridge *,
278 const uint8_t bridge_ea[ETH_ADDR_LEN],
279 struct iface *hw_addr_iface);
280 static struct iface *bridge_get_local_iface(struct bridge *);
281 static uint64_t dpid_from_hash(const void *, size_t nbytes);
283 static unixctl_cb_func bridge_unixctl_fdb_show;
285 static void lacp_run(struct bridge *);
286 static void lacp_wait(struct bridge *);
287 static void lacp_process_packet(const struct ofpbuf *, struct iface *);
289 static void bond_init(void);
290 static void bond_run(struct bridge *);
291 static void bond_wait(struct bridge *);
292 static void bond_rebalance_port(struct port *);
293 static void bond_send_learning_packets(struct port *);
294 static void bond_enable_slave(struct iface *iface, bool enable);
296 static struct port *port_create(struct bridge *, const char *name);
297 static void port_reconfigure(struct port *, const struct ovsrec_port *);
298 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
299 static void port_destroy(struct port *);
300 static struct port *port_lookup(const struct bridge *, const char *name);
301 static struct iface *port_lookup_iface(const struct port *, const char *name);
302 static struct port *port_from_dp_ifidx(const struct bridge *,
304 static void port_update_bond_compat(struct port *);
305 static void port_update_vlan_compat(struct port *);
306 static void port_update_bonding(struct port *);
307 static void port_update_lacp(struct port *);
309 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
310 static void mirror_destroy(struct mirror *);
311 static void mirror_reconfigure(struct bridge *);
312 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
313 static bool vlan_is_mirrored(const struct mirror *, int vlan);
315 static struct iface *iface_create(struct port *port,
316 const struct ovsrec_interface *if_cfg);
317 static void iface_destroy(struct iface *);
318 static struct iface *iface_lookup(const struct bridge *, const char *name);
319 static struct iface *iface_from_dp_ifidx(const struct bridge *,
321 static void iface_set_mac(struct iface *);
322 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
323 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
324 static void iface_update_cfm(struct iface *);
325 static void iface_refresh_cfm_stats(struct iface *iface);
326 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
327 static uint8_t iface_get_lacp_state(const struct iface *);
328 static void iface_get_lacp_priority(struct iface *, struct lacp_info *);
329 static void iface_set_lacp_defaulted(struct iface *);
330 static void iface_set_lacp_expired(struct iface *);
332 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
334 static void shash_to_ovs_idl_map(struct shash *,
335 char ***keys, char ***values, size_t *n);
338 /* Hooks into ofproto processing. */
339 static struct ofhooks bridge_ofhooks;
341 /* Public functions. */
343 /* Initializes the bridge module, configuring it to obtain its configuration
344 * from an OVSDB server accessed over 'remote', which should be a string in a
345 * form acceptable to ovsdb_idl_create(). */
347 bridge_init(const char *remote)
349 /* Create connection to database. */
350 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
352 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
353 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
354 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
356 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
358 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
359 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
361 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
362 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
363 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
365 /* Register unixctl commands. */
366 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
367 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
369 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
377 struct bridge *br, *next_br;
379 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
382 ovsdb_idl_destroy(idl);
385 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
386 * but for which the ovs-vswitchd configuration 'cfg' is required. */
388 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
390 static bool already_configured_once;
391 struct svec bridge_names;
392 struct svec dpif_names, dpif_types;
395 /* Only do this once per ovs-vswitchd run. */
396 if (already_configured_once) {
399 already_configured_once = true;
401 stats_timer = time_msec() + STATS_INTERVAL;
403 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
404 svec_init(&bridge_names);
405 for (i = 0; i < cfg->n_bridges; i++) {
406 svec_add(&bridge_names, cfg->bridges[i]->name);
408 svec_sort(&bridge_names);
410 /* Iterate over all system dpifs and delete any of them that do not appear
412 svec_init(&dpif_names);
413 svec_init(&dpif_types);
414 dp_enumerate_types(&dpif_types);
415 for (i = 0; i < dpif_types.n; i++) {
418 dp_enumerate_names(dpif_types.names[i], &dpif_names);
420 /* Delete each dpif whose name is not in 'bridge_names'. */
421 for (j = 0; j < dpif_names.n; j++) {
422 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
426 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
435 svec_destroy(&bridge_names);
436 svec_destroy(&dpif_names);
437 svec_destroy(&dpif_types);
440 /* Callback for iterate_and_prune_ifaces(). */
442 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
444 if (!iface->netdev) {
445 /* We already reported a related error, don't bother duplicating it. */
449 if (iface->dp_ifidx < 0) {
450 VLOG_ERR("%s interface not in %s, dropping",
451 iface->name, dpif_name(br->dpif));
455 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
456 iface->name, iface->dp_ifidx);
460 /* Callback for iterate_and_prune_ifaces(). */
462 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
463 void *aux OVS_UNUSED)
465 /* Set policing attributes. */
466 netdev_set_policing(iface->netdev,
467 iface->cfg->ingress_policing_rate,
468 iface->cfg->ingress_policing_burst);
470 /* Set MAC address of internal interfaces other than the local
472 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
473 iface_set_mac(iface);
479 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
480 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
481 * deletes from 'br' any ports that no longer have any interfaces. */
483 iterate_and_prune_ifaces(struct bridge *br,
484 bool (*cb)(struct bridge *, struct iface *,
490 for (i = 0; i < br->n_ports; ) {
491 struct port *port = br->ports[i];
492 for (j = 0; j < port->n_ifaces; ) {
493 struct iface *iface = port->ifaces[j];
494 if (cb(br, iface, aux)) {
497 iface_set_ofport(iface->cfg, -1);
498 iface_destroy(iface);
502 if (port->n_ifaces) {
505 VLOG_ERR("%s port has no interfaces, dropping", port->name);
511 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
512 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
513 * responsible for freeing '*managersp' (with free()).
515 * You may be asking yourself "why does ovs-vswitchd care?", because
516 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
517 * should not be and in fact is not directly involved in that. But
518 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
519 * it has to tell in-band control where the managers are to enable that.
520 * (Thus, only managers connected in-band are collected.)
523 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
524 struct sockaddr_in **managersp, size_t *n_managersp)
526 struct sockaddr_in *managers = NULL;
527 size_t n_managers = 0;
528 struct shash targets;
531 /* Collect all of the potential targets, as the union of the "managers"
532 * column and the "targets" columns of the rows pointed to by
533 * "manager_options", excluding any that are out-of-band. */
534 shash_init(&targets);
535 for (i = 0; i < ovs_cfg->n_managers; i++) {
536 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
538 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
539 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
541 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
542 shash_find_and_delete(&targets, m->target);
544 shash_add_once(&targets, m->target, NULL);
548 /* Now extract the targets' IP addresses. */
549 if (!shash_is_empty(&targets)) {
550 struct shash_node *node;
552 managers = xmalloc(shash_count(&targets) * sizeof *managers);
553 SHASH_FOR_EACH (node, &targets) {
554 const char *target = node->name;
555 struct sockaddr_in *sin = &managers[n_managers];
557 if ((!strncmp(target, "tcp:", 4)
558 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
559 (!strncmp(target, "ssl:", 4)
560 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
565 shash_destroy(&targets);
567 *managersp = managers;
568 *n_managersp = n_managers;
572 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
574 struct shash old_br, new_br;
575 struct shash_node *node;
576 struct bridge *br, *next;
577 struct sockaddr_in *managers;
580 int sflow_bridge_number;
582 COVERAGE_INC(bridge_reconfigure);
584 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
586 /* Collect old and new bridges. */
589 LIST_FOR_EACH (br, node, &all_bridges) {
590 shash_add(&old_br, br->name, br);
592 for (i = 0; i < ovs_cfg->n_bridges; i++) {
593 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
594 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
595 VLOG_WARN("more than one bridge named %s", br_cfg->name);
599 /* Get rid of deleted bridges and add new bridges. */
600 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
601 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
608 SHASH_FOR_EACH (node, &new_br) {
609 const char *br_name = node->name;
610 const struct ovsrec_bridge *br_cfg = node->data;
611 br = shash_find_data(&old_br, br_name);
613 /* If the bridge datapath type has changed, we need to tear it
614 * down and recreate. */
615 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
617 bridge_create(br_cfg);
620 bridge_create(br_cfg);
623 shash_destroy(&old_br);
624 shash_destroy(&new_br);
626 /* Reconfigure all bridges. */
627 LIST_FOR_EACH (br, node, &all_bridges) {
628 bridge_reconfigure_one(br);
631 /* Add and delete ports on all datapaths.
633 * The kernel will reject any attempt to add a given port to a datapath if
634 * that port already belongs to a different datapath, so we must do all
635 * port deletions before any port additions. */
636 LIST_FOR_EACH (br, node, &all_bridges) {
637 struct dpif_port_dump dump;
638 struct shash want_ifaces;
639 struct dpif_port dpif_port;
641 bridge_get_all_ifaces(br, &want_ifaces);
642 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
643 if (!shash_find(&want_ifaces, dpif_port.name)
644 && strcmp(dpif_port.name, br->name)) {
645 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
647 VLOG_ERR("failed to remove %s interface from %s: %s",
648 dpif_port.name, dpif_name(br->dpif),
653 shash_destroy(&want_ifaces);
655 LIST_FOR_EACH (br, node, &all_bridges) {
656 struct shash cur_ifaces, want_ifaces;
657 struct dpif_port_dump dump;
658 struct dpif_port dpif_port;
660 /* Get the set of interfaces currently in this datapath. */
661 shash_init(&cur_ifaces);
662 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
663 struct dpif_port *port_info = xmalloc(sizeof *port_info);
664 dpif_port_clone(port_info, &dpif_port);
665 shash_add(&cur_ifaces, dpif_port.name, port_info);
668 /* Get the set of interfaces we want on this datapath. */
669 bridge_get_all_ifaces(br, &want_ifaces);
671 hmap_clear(&br->ifaces);
672 SHASH_FOR_EACH (node, &want_ifaces) {
673 const char *if_name = node->name;
674 struct iface *iface = node->data;
675 struct dpif_port *dpif_port;
679 type = iface ? iface->type : "internal";
680 dpif_port = shash_find_data(&cur_ifaces, if_name);
682 /* If we have a port or a netdev already, and it's not the type we
683 * want, then delete the port (if any) and close the netdev (if
685 if ((dpif_port && strcmp(dpif_port->type, type))
686 || (iface && iface->netdev
687 && strcmp(type, netdev_get_type(iface->netdev)))) {
689 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
696 netdev_close(iface->netdev);
697 iface->netdev = NULL;
701 /* If the port doesn't exist or we don't have the netdev open,
702 * we need to do more work. */
703 if (!dpif_port || (iface && !iface->netdev)) {
704 struct netdev_options options;
705 struct netdev *netdev;
708 /* First open the network device. */
709 options.name = if_name;
711 options.args = &args;
712 options.ethertype = NETDEV_ETH_TYPE_NONE;
716 shash_from_ovs_idl_map(iface->cfg->key_options,
717 iface->cfg->value_options,
718 iface->cfg->n_options, &args);
720 error = netdev_open(&options, &netdev);
721 shash_destroy(&args);
724 VLOG_WARN("could not open network device %s (%s)",
725 if_name, strerror(error));
729 /* Then add the port if we haven't already. */
731 error = dpif_port_add(br->dpif, netdev, NULL);
733 netdev_close(netdev);
734 if (error == EFBIG) {
735 VLOG_ERR("ran out of valid port numbers on %s",
736 dpif_name(br->dpif));
739 VLOG_ERR("failed to add %s interface to %s: %s",
740 if_name, dpif_name(br->dpif),
747 /* Update 'iface'. */
749 iface->netdev = netdev;
750 iface->enabled = netdev_get_carrier(iface->netdev);
751 iface->up = iface->enabled;
753 } else if (iface && iface->netdev) {
757 shash_from_ovs_idl_map(iface->cfg->key_options,
758 iface->cfg->value_options,
759 iface->cfg->n_options, &args);
760 netdev_set_config(iface->netdev, &args);
761 shash_destroy(&args);
764 shash_destroy(&want_ifaces);
766 SHASH_FOR_EACH (node, &cur_ifaces) {
767 struct dpif_port *port_info = node->data;
768 dpif_port_destroy(port_info);
771 shash_destroy(&cur_ifaces);
773 sflow_bridge_number = 0;
774 LIST_FOR_EACH (br, node, &all_bridges) {
777 struct iface *local_iface;
778 struct iface *hw_addr_iface;
781 bridge_fetch_dp_ifaces(br);
783 iterate_and_prune_ifaces(br, check_iface, NULL);
785 /* Pick local port hardware address, datapath ID. */
786 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
787 local_iface = bridge_get_local_iface(br);
789 int error = netdev_set_etheraddr(local_iface->netdev, ea);
791 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
792 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
793 "Ethernet address: %s",
794 br->name, strerror(error));
797 memcpy(br->ea, ea, ETH_ADDR_LEN);
799 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
800 ofproto_set_datapath_id(br->ofproto, dpid);
802 dpid_string = xasprintf("%016"PRIx64, dpid);
803 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
806 /* Set NetFlow configuration on this bridge. */
807 if (br->cfg->netflow) {
808 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
809 struct netflow_options opts;
811 memset(&opts, 0, sizeof opts);
813 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
814 if (nf_cfg->engine_type) {
815 opts.engine_type = *nf_cfg->engine_type;
817 if (nf_cfg->engine_id) {
818 opts.engine_id = *nf_cfg->engine_id;
821 opts.active_timeout = nf_cfg->active_timeout;
822 if (!opts.active_timeout) {
823 opts.active_timeout = -1;
824 } else if (opts.active_timeout < 0) {
825 VLOG_WARN("bridge %s: active timeout interval set to negative "
826 "value, using default instead (%d seconds)", br->name,
827 NF_ACTIVE_TIMEOUT_DEFAULT);
828 opts.active_timeout = -1;
831 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
832 if (opts.add_id_to_iface) {
833 if (opts.engine_id > 0x7f) {
834 VLOG_WARN("bridge %s: netflow port mangling may conflict "
835 "with another vswitch, choose an engine id less "
836 "than 128", br->name);
838 if (br->n_ports > 508) {
839 VLOG_WARN("bridge %s: netflow port mangling will conflict "
840 "with another port when more than 508 ports are "
845 opts.collectors.n = nf_cfg->n_targets;
846 opts.collectors.names = nf_cfg->targets;
847 if (ofproto_set_netflow(br->ofproto, &opts)) {
848 VLOG_ERR("bridge %s: problem setting netflow collectors",
852 ofproto_set_netflow(br->ofproto, NULL);
855 /* Set sFlow configuration on this bridge. */
856 if (br->cfg->sflow) {
857 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
858 struct ovsrec_controller **controllers;
859 struct ofproto_sflow_options oso;
860 size_t n_controllers;
862 memset(&oso, 0, sizeof oso);
864 oso.targets.n = sflow_cfg->n_targets;
865 oso.targets.names = sflow_cfg->targets;
867 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
868 if (sflow_cfg->sampling) {
869 oso.sampling_rate = *sflow_cfg->sampling;
872 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
873 if (sflow_cfg->polling) {
874 oso.polling_interval = *sflow_cfg->polling;
877 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
878 if (sflow_cfg->header) {
879 oso.header_len = *sflow_cfg->header;
882 oso.sub_id = sflow_bridge_number++;
883 oso.agent_device = sflow_cfg->agent;
885 oso.control_ip = NULL;
886 n_controllers = bridge_get_controllers(br, &controllers);
887 for (i = 0; i < n_controllers; i++) {
888 if (controllers[i]->local_ip) {
889 oso.control_ip = controllers[i]->local_ip;
893 ofproto_set_sflow(br->ofproto, &oso);
895 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
897 ofproto_set_sflow(br->ofproto, NULL);
900 /* Update the controller and related settings. It would be more
901 * straightforward to call this from bridge_reconfigure_one(), but we
902 * can't do it there for two reasons. First, and most importantly, at
903 * that point we don't know the dp_ifidx of any interfaces that have
904 * been added to the bridge (because we haven't actually added them to
905 * the datapath). Second, at that point we haven't set the datapath ID
906 * yet; when a controller is configured, resetting the datapath ID will
907 * immediately disconnect from the controller, so it's better to set
908 * the datapath ID before the controller. */
909 bridge_reconfigure_remotes(br, managers, n_managers);
911 LIST_FOR_EACH (br, node, &all_bridges) {
912 for (i = 0; i < br->n_ports; i++) {
913 struct port *port = br->ports[i];
916 port_update_vlan_compat(port);
917 port_update_bonding(port);
918 port_update_lacp(port);
920 for (j = 0; j < port->n_ifaces; j++) {
921 iface_update_qos(port->ifaces[j], port->cfg->qos);
925 LIST_FOR_EACH (br, node, &all_bridges) {
926 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
929 LIST_FOR_EACH (br, node, &all_bridges) {
931 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
932 iface_update_cfm(iface);
940 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
941 const struct ovsdb_idl_column *column,
944 const struct ovsdb_datum *datum;
945 union ovsdb_atom atom;
948 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
949 atom.string = (char *) key;
950 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
951 return idx == UINT_MAX ? NULL : datum->values[idx].string;
955 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
957 return get_ovsrec_key_value(&br_cfg->header_,
958 &ovsrec_bridge_col_other_config, key);
962 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
963 struct iface **hw_addr_iface)
969 *hw_addr_iface = NULL;
971 /* Did the user request a particular MAC? */
972 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
973 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
974 if (eth_addr_is_multicast(ea)) {
975 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
976 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
977 } else if (eth_addr_is_zero(ea)) {
978 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
984 /* Otherwise choose the minimum non-local MAC address among all of the
986 memset(ea, 0xff, sizeof ea);
987 for (i = 0; i < br->n_ports; i++) {
988 struct port *port = br->ports[i];
989 uint8_t iface_ea[ETH_ADDR_LEN];
992 /* Mirror output ports don't participate. */
993 if (port->is_mirror_output_port) {
997 /* Choose the MAC address to represent the port. */
998 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
999 /* Find the interface with this Ethernet address (if any) so that
1000 * we can provide the correct devname to the caller. */
1002 for (j = 0; j < port->n_ifaces; j++) {
1003 struct iface *candidate = port->ifaces[j];
1004 uint8_t candidate_ea[ETH_ADDR_LEN];
1005 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1006 && eth_addr_equals(iface_ea, candidate_ea)) {
1011 /* Choose the interface whose MAC address will represent the port.
1012 * The Linux kernel bonding code always chooses the MAC address of
1013 * the first slave added to a bond, and the Fedora networking
1014 * scripts always add slaves to a bond in alphabetical order, so
1015 * for compatibility we choose the interface with the name that is
1016 * first in alphabetical order. */
1017 iface = port->ifaces[0];
1018 for (j = 1; j < port->n_ifaces; j++) {
1019 struct iface *candidate = port->ifaces[j];
1020 if (strcmp(candidate->name, iface->name) < 0) {
1025 /* The local port doesn't count (since we're trying to choose its
1026 * MAC address anyway). */
1027 if (iface->dp_ifidx == ODPP_LOCAL) {
1032 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1034 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1035 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1036 iface->name, strerror(error));
1041 /* Compare against our current choice. */
1042 if (!eth_addr_is_multicast(iface_ea) &&
1043 !eth_addr_is_local(iface_ea) &&
1044 !eth_addr_is_reserved(iface_ea) &&
1045 !eth_addr_is_zero(iface_ea) &&
1046 eth_addr_compare_3way(iface_ea, ea) < 0)
1048 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1049 *hw_addr_iface = iface;
1052 if (eth_addr_is_multicast(ea)) {
1053 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1054 *hw_addr_iface = NULL;
1055 VLOG_WARN("bridge %s: using default bridge Ethernet "
1056 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1058 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1059 br->name, ETH_ADDR_ARGS(ea));
1063 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1064 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1065 * an interface on 'br', then that interface must be passed in as
1066 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1067 * 'hw_addr_iface' must be passed in as a null pointer. */
1069 bridge_pick_datapath_id(struct bridge *br,
1070 const uint8_t bridge_ea[ETH_ADDR_LEN],
1071 struct iface *hw_addr_iface)
1074 * The procedure for choosing a bridge MAC address will, in the most
1075 * ordinary case, also choose a unique MAC that we can use as a datapath
1076 * ID. In some special cases, though, multiple bridges will end up with
1077 * the same MAC address. This is OK for the bridges, but it will confuse
1078 * the OpenFlow controller, because each datapath needs a unique datapath
1081 * Datapath IDs must be unique. It is also very desirable that they be
1082 * stable from one run to the next, so that policy set on a datapath
1085 const char *datapath_id;
1088 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1089 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1093 if (hw_addr_iface) {
1095 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1097 * A bridge whose MAC address is taken from a VLAN network device
1098 * (that is, a network device created with vconfig(8) or similar
1099 * tool) will have the same MAC address as a bridge on the VLAN
1100 * device's physical network device.
1102 * Handle this case by hashing the physical network device MAC
1103 * along with the VLAN identifier.
1105 uint8_t buf[ETH_ADDR_LEN + 2];
1106 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1107 buf[ETH_ADDR_LEN] = vlan >> 8;
1108 buf[ETH_ADDR_LEN + 1] = vlan;
1109 return dpid_from_hash(buf, sizeof buf);
1112 * Assume that this bridge's MAC address is unique, since it
1113 * doesn't fit any of the cases we handle specially.
1118 * A purely internal bridge, that is, one that has no non-virtual
1119 * network devices on it at all, is more difficult because it has no
1120 * natural unique identifier at all.
1122 * When the host is a XenServer, we handle this case by hashing the
1123 * host's UUID with the name of the bridge. Names of bridges are
1124 * persistent across XenServer reboots, although they can be reused if
1125 * an internal network is destroyed and then a new one is later
1126 * created, so this is fairly effective.
1128 * When the host is not a XenServer, we punt by using a random MAC
1129 * address on each run.
1131 const char *host_uuid = xenserver_get_host_uuid();
1133 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1134 dpid = dpid_from_hash(combined, strlen(combined));
1140 return eth_addr_to_uint64(bridge_ea);
1144 dpid_from_hash(const void *data, size_t n)
1146 uint8_t hash[SHA1_DIGEST_SIZE];
1148 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1149 sha1_bytes(data, n, hash);
1150 eth_addr_mark_random(hash);
1151 return eth_addr_to_uint64(hash);
1155 iface_refresh_status(struct iface *iface)
1159 enum netdev_flags flags;
1168 if (!netdev_get_status(iface->netdev, &sh)) {
1170 char **keys, **values;
1172 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1173 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1178 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1181 shash_destroy_free_data(&sh);
1183 error = netdev_get_flags(iface->netdev, &flags);
1185 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1188 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1191 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1193 ovsrec_interface_set_duplex(iface->cfg,
1194 netdev_features_is_full_duplex(current)
1196 /* warning: uint64_t -> int64_t conversion */
1197 bps = netdev_features_to_bps(current);
1198 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1201 ovsrec_interface_set_duplex(iface->cfg, NULL);
1202 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1206 ovsrec_interface_set_link_state(iface->cfg,
1207 netdev_get_carrier(iface->netdev)
1210 error = netdev_get_mtu(iface->netdev, &mtu);
1211 if (!error && mtu != INT_MAX) {
1213 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1216 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1221 iface_refresh_cfm_stats(struct iface *iface)
1225 const struct ovsrec_monitor *mon;
1227 mon = iface->cfg->monitor;
1234 for (i = 0; i < mon->n_remote_mps; i++) {
1235 const struct ovsrec_maintenance_point *mp;
1236 const struct remote_mp *rmp;
1238 mp = mon->remote_mps[i];
1239 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1241 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1244 if (hmap_is_empty(&cfm->x_remote_mps)) {
1245 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1248 struct remote_mp *rmp;
1249 int64_t *x_remote_mps;
1251 length = hmap_count(&cfm->x_remote_mps);
1252 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1255 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1256 x_remote_mps[i++] = rmp->mpid;
1259 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1263 if (hmap_is_empty(&cfm->x_remote_maids)) {
1264 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1267 char **x_remote_maids;
1268 struct remote_maid *rmaid;
1270 length = hmap_count(&cfm->x_remote_maids);
1271 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1274 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1277 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1279 for (j = 0; j < CCM_MAID_LEN; j++) {
1280 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1285 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1287 for (i = 0; i < length; i++) {
1288 free(x_remote_maids[i]);
1290 free(x_remote_maids);
1293 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1297 iface_refresh_stats(struct iface *iface)
1303 static const struct iface_stat iface_stats[] = {
1304 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1305 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1306 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1307 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1308 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1309 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1310 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1311 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1312 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1313 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1314 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1315 { "collisions", offsetof(struct netdev_stats, collisions) },
1317 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1318 const struct iface_stat *s;
1320 char *keys[N_STATS];
1321 int64_t values[N_STATS];
1324 struct netdev_stats stats;
1326 /* Intentionally ignore return value, since errors will set 'stats' to
1327 * all-1s, and we will deal with that correctly below. */
1328 netdev_get_stats(iface->netdev, &stats);
1331 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1332 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1333 if (value != UINT64_MAX) {
1340 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1344 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1346 struct ovsdb_datum datum;
1350 get_system_stats(&stats);
1352 ovsdb_datum_from_shash(&datum, &stats);
1353 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1357 static inline const char *
1358 nx_role_to_str(enum nx_role role)
1363 case NX_ROLE_MASTER:
1368 return "*** INVALID ROLE ***";
1373 bridge_refresh_controller_status(const struct bridge *br)
1376 const struct ovsrec_controller *cfg;
1378 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1380 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1381 struct ofproto_controller_info *cinfo =
1382 shash_find_data(&info, cfg->target);
1385 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1386 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1387 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1388 (char **) cinfo->pairs.values,
1391 ovsrec_controller_set_is_connected(cfg, false);
1392 ovsrec_controller_set_role(cfg, NULL);
1393 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1397 ofproto_free_ofproto_controller_info(&info);
1403 const struct ovsrec_open_vswitch *cfg;
1405 bool datapath_destroyed;
1406 bool database_changed;
1409 /* Let each bridge do the work that it needs to do. */
1410 datapath_destroyed = false;
1411 LIST_FOR_EACH (br, node, &all_bridges) {
1412 int error = bridge_run_one(br);
1414 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1415 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1416 "forcing reconfiguration", br->name);
1417 datapath_destroyed = true;
1421 /* (Re)configure if necessary. */
1422 database_changed = ovsdb_idl_run(idl);
1423 cfg = ovsrec_open_vswitch_first(idl);
1425 /* Re-configure SSL. We do this on every trip through the main loop,
1426 * instead of just when the database changes, because the contents of the
1427 * key and certificate files can change without the database changing.
1429 * We do this before bridge_reconfigure() because that function might
1430 * initiate SSL connections and thus requires SSL to be configured. */
1431 if (cfg && cfg->ssl) {
1432 const struct ovsrec_ssl *ssl = cfg->ssl;
1434 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1435 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1438 if (database_changed || datapath_destroyed) {
1440 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1442 bridge_configure_once(cfg);
1443 bridge_reconfigure(cfg);
1445 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1446 ovsdb_idl_txn_commit(txn);
1447 ovsdb_idl_txn_destroy(txn); /* XXX */
1449 /* We still need to reconfigure to avoid dangling pointers to
1450 * now-destroyed ovsrec structures inside bridge data. */
1451 static const struct ovsrec_open_vswitch null_cfg;
1453 bridge_reconfigure(&null_cfg);
1457 /* Refresh system and interface stats if necessary. */
1458 if (time_msec() >= stats_timer) {
1460 struct ovsdb_idl_txn *txn;
1462 txn = ovsdb_idl_txn_create(idl);
1463 LIST_FOR_EACH (br, node, &all_bridges) {
1466 for (i = 0; i < br->n_ports; i++) {
1467 struct port *port = br->ports[i];
1470 for (j = 0; j < port->n_ifaces; j++) {
1471 struct iface *iface = port->ifaces[j];
1472 iface_refresh_stats(iface);
1473 iface_refresh_cfm_stats(iface);
1474 iface_refresh_status(iface);
1477 bridge_refresh_controller_status(br);
1479 refresh_system_stats(cfg);
1480 ovsdb_idl_txn_commit(txn);
1481 ovsdb_idl_txn_destroy(txn); /* XXX */
1484 stats_timer = time_msec() + STATS_INTERVAL;
1492 struct iface *iface;
1494 LIST_FOR_EACH (br, node, &all_bridges) {
1495 ofproto_wait(br->ofproto);
1496 if (ofproto_has_primary_controller(br->ofproto)) {
1500 mac_learning_wait(br->ml);
1504 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1506 cfm_wait(iface->cfm);
1510 ovsdb_idl_wait(idl);
1511 poll_timer_wait_until(stats_timer);
1514 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1515 * configuration changes. */
1517 bridge_flush(struct bridge *br)
1519 COVERAGE_INC(bridge_flush);
1521 mac_learning_flush(br->ml);
1524 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1525 * such interface. */
1526 static struct iface *
1527 bridge_get_local_iface(struct bridge *br)
1531 for (i = 0; i < br->n_ports; i++) {
1532 struct port *port = br->ports[i];
1533 for (j = 0; j < port->n_ifaces; j++) {
1534 struct iface *iface = port->ifaces[j];
1535 if (iface->dp_ifidx == ODPP_LOCAL) {
1544 /* Bridge unixctl user interface functions. */
1546 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1547 const char *args, void *aux OVS_UNUSED)
1549 struct ds ds = DS_EMPTY_INITIALIZER;
1550 const struct bridge *br;
1551 const struct mac_entry *e;
1553 br = bridge_lookup(args);
1555 unixctl_command_reply(conn, 501, "no such bridge");
1559 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1560 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1561 if (e->port < 0 || e->port >= br->n_ports) {
1564 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1565 br->ports[e->port]->ifaces[0]->dp_ifidx,
1566 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1568 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1572 /* Bridge reconfiguration functions. */
1573 static struct bridge *
1574 bridge_create(const struct ovsrec_bridge *br_cfg)
1579 assert(!bridge_lookup(br_cfg->name));
1580 br = xzalloc(sizeof *br);
1582 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1588 dpif_flow_flush(br->dpif);
1590 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1593 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1595 dpif_delete(br->dpif);
1596 dpif_close(br->dpif);
1601 br->name = xstrdup(br_cfg->name);
1603 br->ml = mac_learning_create();
1604 eth_addr_nicira_random(br->default_ea);
1606 hmap_init(&br->ifaces);
1608 shash_init(&br->port_by_name);
1609 shash_init(&br->iface_by_name);
1613 list_push_back(&all_bridges, &br->node);
1615 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1621 bridge_destroy(struct bridge *br)
1626 while (br->n_ports > 0) {
1627 port_destroy(br->ports[br->n_ports - 1]);
1629 list_remove(&br->node);
1630 error = dpif_delete(br->dpif);
1631 if (error && error != ENOENT) {
1632 VLOG_ERR("failed to delete %s: %s",
1633 dpif_name(br->dpif), strerror(error));
1635 dpif_close(br->dpif);
1636 ofproto_destroy(br->ofproto);
1637 mac_learning_destroy(br->ml);
1638 hmap_destroy(&br->ifaces);
1639 shash_destroy(&br->port_by_name);
1640 shash_destroy(&br->iface_by_name);
1647 static struct bridge *
1648 bridge_lookup(const char *name)
1652 LIST_FOR_EACH (br, node, &all_bridges) {
1653 if (!strcmp(br->name, name)) {
1660 /* Handle requests for a listing of all flows known by the OpenFlow
1661 * stack, including those normally hidden. */
1663 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1664 const char *args, void *aux OVS_UNUSED)
1669 br = bridge_lookup(args);
1671 unixctl_command_reply(conn, 501, "Unknown bridge");
1676 ofproto_get_all_flows(br->ofproto, &results);
1678 unixctl_command_reply(conn, 200, ds_cstr(&results));
1679 ds_destroy(&results);
1682 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1683 * connections and reconnect. If BRIDGE is not specified, then all bridges
1684 * drop their controller connections and reconnect. */
1686 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1687 const char *args, void *aux OVS_UNUSED)
1690 if (args[0] != '\0') {
1691 br = bridge_lookup(args);
1693 unixctl_command_reply(conn, 501, "Unknown bridge");
1696 ofproto_reconnect_controllers(br->ofproto);
1698 LIST_FOR_EACH (br, node, &all_bridges) {
1699 ofproto_reconnect_controllers(br->ofproto);
1702 unixctl_command_reply(conn, 200, NULL);
1706 bridge_run_one(struct bridge *br)
1709 struct iface *iface;
1711 error = ofproto_run1(br->ofproto);
1716 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1720 error = ofproto_run2(br->ofproto, br->flush);
1723 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1724 struct ofpbuf *packet;
1730 packet = cfm_run(iface->cfm);
1732 iface_send_packet(iface, packet);
1733 ofpbuf_uninit(packet);
1742 bridge_get_controllers(const struct bridge *br,
1743 struct ovsrec_controller ***controllersp)
1745 struct ovsrec_controller **controllers;
1746 size_t n_controllers;
1748 controllers = br->cfg->controller;
1749 n_controllers = br->cfg->n_controller;
1751 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1757 *controllersp = controllers;
1759 return n_controllers;
1763 bridge_reconfigure_one(struct bridge *br)
1765 struct shash old_ports, new_ports;
1766 struct svec snoops, old_snoops;
1767 struct shash_node *node;
1768 enum ofproto_fail_mode fail_mode;
1771 /* Collect old ports. */
1772 shash_init(&old_ports);
1773 for (i = 0; i < br->n_ports; i++) {
1774 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1777 /* Collect new ports. */
1778 shash_init(&new_ports);
1779 for (i = 0; i < br->cfg->n_ports; i++) {
1780 const char *name = br->cfg->ports[i]->name;
1781 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1782 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1787 /* If we have a controller, then we need a local port. Complain if the
1788 * user didn't specify one.
1790 * XXX perhaps we should synthesize a port ourselves in this case. */
1791 if (bridge_get_controllers(br, NULL)) {
1792 char local_name[IF_NAMESIZE];
1795 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1796 local_name, sizeof local_name);
1797 if (!error && !shash_find(&new_ports, local_name)) {
1798 VLOG_WARN("bridge %s: controller specified but no local port "
1799 "(port named %s) defined",
1800 br->name, local_name);
1804 /* Get rid of deleted ports.
1805 * Get rid of deleted interfaces on ports that still exist. */
1806 SHASH_FOR_EACH (node, &old_ports) {
1807 struct port *port = node->data;
1808 const struct ovsrec_port *port_cfg;
1810 port_cfg = shash_find_data(&new_ports, node->name);
1814 port_del_ifaces(port, port_cfg);
1818 /* Create new ports.
1819 * Add new interfaces to existing ports.
1820 * Reconfigure existing ports. */
1821 SHASH_FOR_EACH (node, &new_ports) {
1822 struct port *port = shash_find_data(&old_ports, node->name);
1824 port = port_create(br, node->name);
1827 port_reconfigure(port, node->data);
1828 if (!port->n_ifaces) {
1829 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1830 br->name, port->name);
1834 shash_destroy(&old_ports);
1835 shash_destroy(&new_ports);
1837 /* Set the fail-mode */
1838 fail_mode = !br->cfg->fail_mode
1839 || !strcmp(br->cfg->fail_mode, "standalone")
1840 ? OFPROTO_FAIL_STANDALONE
1841 : OFPROTO_FAIL_SECURE;
1842 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1843 && !ofproto_has_primary_controller(br->ofproto)) {
1844 ofproto_flush_flows(br->ofproto);
1846 ofproto_set_fail_mode(br->ofproto, fail_mode);
1848 /* Delete all flows if we're switching from connected to standalone or vice
1849 * versa. (XXX Should we delete all flows if we are switching from one
1850 * controller to another?) */
1852 /* Configure OpenFlow controller connection snooping. */
1854 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1855 ovs_rundir(), br->name));
1856 svec_init(&old_snoops);
1857 ofproto_get_snoops(br->ofproto, &old_snoops);
1858 if (!svec_equal(&snoops, &old_snoops)) {
1859 ofproto_set_snoops(br->ofproto, &snoops);
1861 svec_destroy(&snoops);
1862 svec_destroy(&old_snoops);
1864 mirror_reconfigure(br);
1867 /* Initializes 'oc' appropriately as a management service controller for
1870 * The caller must free oc->target when it is no longer needed. */
1872 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1873 struct ofproto_controller *oc)
1875 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1876 oc->max_backoff = 0;
1877 oc->probe_interval = 60;
1878 oc->band = OFPROTO_OUT_OF_BAND;
1879 oc->accept_re = NULL;
1880 oc->update_resolv_conf = false;
1882 oc->burst_limit = 0;
1885 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1887 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1888 struct ofproto_controller *oc)
1890 oc->target = c->target;
1891 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1892 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1893 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1894 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1895 oc->accept_re = c->discover_accept_regex;
1896 oc->update_resolv_conf = c->discover_update_resolv_conf;
1897 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1898 oc->burst_limit = (c->controller_burst_limit
1899 ? *c->controller_burst_limit : 0);
1902 /* Configures the IP stack for 'br''s local interface properly according to the
1903 * configuration in 'c'. */
1905 bridge_configure_local_iface_netdev(struct bridge *br,
1906 struct ovsrec_controller *c)
1908 struct netdev *netdev;
1909 struct in_addr mask, gateway;
1911 struct iface *local_iface;
1914 /* Controller discovery does its own TCP/IP configuration later. */
1915 if (strcmp(c->target, "discover")) {
1919 /* If there's no local interface or no IP address, give up. */
1920 local_iface = bridge_get_local_iface(br);
1921 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1925 /* Bring up the local interface. */
1926 netdev = local_iface->netdev;
1927 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1929 /* Configure the IP address and netmask. */
1930 if (!c->local_netmask
1931 || !inet_aton(c->local_netmask, &mask)
1933 mask.s_addr = guess_netmask(ip.s_addr);
1935 if (!netdev_set_in4(netdev, ip, mask)) {
1936 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1937 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1940 /* Configure the default gateway. */
1941 if (c->local_gateway
1942 && inet_aton(c->local_gateway, &gateway)
1943 && gateway.s_addr) {
1944 if (!netdev_add_router(netdev, gateway)) {
1945 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1946 br->name, IP_ARGS(&gateway.s_addr));
1952 bridge_reconfigure_remotes(struct bridge *br,
1953 const struct sockaddr_in *managers,
1956 const char *disable_ib_str, *queue_id_str;
1957 bool disable_in_band = false;
1960 struct ovsrec_controller **controllers;
1961 size_t n_controllers;
1964 struct ofproto_controller *ocs;
1968 /* Check if we should disable in-band control on this bridge. */
1969 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1970 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1971 disable_in_band = true;
1974 /* Set OpenFlow queue ID for in-band control. */
1975 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1976 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1977 ofproto_set_in_band_queue(br->ofproto, queue_id);
1979 if (disable_in_band) {
1980 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1982 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1984 had_primary = ofproto_has_primary_controller(br->ofproto);
1986 n_controllers = bridge_get_controllers(br, &controllers);
1988 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1991 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1992 for (i = 0; i < n_controllers; i++) {
1993 struct ovsrec_controller *c = controllers[i];
1995 if (!strncmp(c->target, "punix:", 6)
1996 || !strncmp(c->target, "unix:", 5)) {
1997 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1999 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2000 * domain sockets and overwriting arbitrary local files. */
2001 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2002 "\"%s\" due to possibility for remote exploit",
2003 dpif_name(br->dpif), c->target);
2007 bridge_configure_local_iface_netdev(br, c);
2008 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2009 if (disable_in_band) {
2010 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2015 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2016 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2019 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2020 ofproto_flush_flows(br->ofproto);
2023 /* If there are no controllers and the bridge is in standalone
2024 * mode, set up a flow that matches every packet and directs
2025 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2026 * switch is in secure mode and we won't pass any traffic until
2027 * a controller has been defined and it tells us to do so. */
2029 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2030 union ofp_action action;
2031 struct cls_rule rule;
2033 memset(&action, 0, sizeof action);
2034 action.type = htons(OFPAT_OUTPUT);
2035 action.output.len = htons(sizeof action);
2036 action.output.port = htons(OFPP_NORMAL);
2037 cls_rule_init_catchall(&rule, 0);
2038 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2043 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2048 for (i = 0; i < br->n_ports; i++) {
2049 struct port *port = br->ports[i];
2050 for (j = 0; j < port->n_ifaces; j++) {
2051 struct iface *iface = port->ifaces[j];
2052 shash_add_once(ifaces, iface->name, iface);
2054 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2055 shash_add_once(ifaces, port->name, NULL);
2060 /* For robustness, in case the administrator moves around datapath ports behind
2061 * our back, we re-check all the datapath port numbers here.
2063 * This function will set the 'dp_ifidx' members of interfaces that have
2064 * disappeared to -1, so only call this function from a context where those
2065 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2066 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2067 * datapath, which doesn't support UINT16_MAX+1 ports. */
2069 bridge_fetch_dp_ifaces(struct bridge *br)
2071 struct dpif_port_dump dump;
2072 struct dpif_port dpif_port;
2075 /* Reset all interface numbers. */
2076 for (i = 0; i < br->n_ports; i++) {
2077 struct port *port = br->ports[i];
2078 for (j = 0; j < port->n_ifaces; j++) {
2079 struct iface *iface = port->ifaces[j];
2080 iface->dp_ifidx = -1;
2083 hmap_clear(&br->ifaces);
2085 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2086 struct iface *iface = iface_lookup(br, dpif_port.name);
2088 if (iface->dp_ifidx >= 0) {
2089 VLOG_WARN("%s reported interface %s twice",
2090 dpif_name(br->dpif), dpif_port.name);
2091 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2092 VLOG_WARN("%s reported interface %"PRIu16" twice",
2093 dpif_name(br->dpif), dpif_port.port_no);
2095 iface->dp_ifidx = dpif_port.port_no;
2096 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2097 hash_int(iface->dp_ifidx, 0));
2100 iface_set_ofport(iface->cfg,
2101 (iface->dp_ifidx >= 0
2102 ? odp_port_to_ofp_port(iface->dp_ifidx)
2108 /* Bridge packet processing functions. */
2111 bond_is_tcp_hash(const struct port *port)
2113 return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED;
2117 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2119 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2122 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2124 struct flow hash_flow;
2126 memcpy(&hash_flow, flow, sizeof hash_flow);
2127 hash_flow.vlan_tci = 0;
2129 /* The symmetric quality of this hash function is not required, but
2130 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2131 * purposes, so we use it out of convenience. */
2132 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2135 static struct bond_entry *
2136 lookup_bond_entry(const struct port *port, const struct flow *flow,
2139 assert(port->bond_mode != BM_AB);
2141 if (bond_is_tcp_hash(port)) {
2142 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2144 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2149 bond_choose_iface(const struct port *port)
2151 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2152 size_t i, best_down_slave = -1;
2153 long long next_delay_expiration = LLONG_MAX;
2155 for (i = 0; i < port->n_ifaces; i++) {
2156 struct iface *iface = port->ifaces[i];
2158 if (iface->enabled) {
2160 } else if (iface->delay_expires < next_delay_expiration
2161 && (iface->lacp_attached
2162 || !(port->lacp & LACP_NEGOTIATED))) {
2163 best_down_slave = i;
2164 next_delay_expiration = iface->delay_expires;
2168 if (best_down_slave != -1) {
2169 struct iface *iface = port->ifaces[best_down_slave];
2171 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2172 "since no other interface is up", iface->name,
2173 iface->delay_expires - time_msec());
2174 bond_enable_slave(iface, true);
2177 return best_down_slave;
2181 choose_output_iface(const struct port *port, const struct flow *flow,
2182 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2184 struct iface *iface;
2186 assert(port->n_ifaces);
2187 if (port->n_ifaces == 1) {
2188 iface = port->ifaces[0];
2189 } else if (port->bond_mode == BM_AB) {
2190 if (port->active_iface < 0) {
2191 *tags |= port->no_ifaces_tag;
2194 iface = port->ifaces[port->active_iface];
2196 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2197 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2198 || !port->ifaces[e->iface_idx]->enabled) {
2199 /* XXX select interface properly. The current interface selection
2200 * is only good for testing the rebalancing code. */
2201 e->iface_idx = bond_choose_iface(port);
2202 if (e->iface_idx < 0) {
2203 *tags |= port->no_ifaces_tag;
2206 e->iface_tag = tag_create_random();
2207 ((struct port *) port)->bond_compat_is_stale = true;
2209 *tags |= e->iface_tag;
2210 iface = port->ifaces[e->iface_idx];
2212 *dp_ifidx = iface->dp_ifidx;
2213 *tags |= iface->tag; /* Currently only used for bonding. */
2218 bond_link_status_update(struct iface *iface)
2220 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2221 struct port *port = iface->port;
2222 bool up = iface->up;
2223 int updelay, downdelay;
2225 updelay = port->updelay;
2226 downdelay = port->downdelay;
2228 if (iface->port->lacp & LACP_NEGOTIATED) {
2233 if (iface->port->lacp && up) {
2234 /* The interface is up if it's attached to an aggregator and its
2235 * partner is synchronized. The only exception is defaulted links.
2236 * They are not required to have synchronized partners because they
2237 * have no partners at all. However, they will only be attached if
2238 * negotiations failed on all interfaces in the bond. */
2239 up = iface->lacp_attached
2240 && (iface->lacp_partner.state & LACP_STATE_SYNC
2241 || iface->lacp_status == LACP_STATUS_DEFAULTED);
2245 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2246 /* Nothing to do. */
2249 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2250 iface->name, up ? "up" : "down");
2251 if (up == iface->enabled) {
2252 iface->delay_expires = LLONG_MAX;
2253 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2254 iface->name, up ? "disabled" : "enabled");
2255 } else if (up && port->active_iface < 0) {
2256 bond_enable_slave(iface, true);
2258 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2259 "other interface is up", iface->name, updelay);
2262 int delay = up ? updelay : downdelay;
2263 iface->delay_expires = time_msec() + delay;
2266 "interface %s: will be %s if it stays %s for %d ms",
2268 up ? "enabled" : "disabled",
2276 bond_choose_active_iface(struct port *port)
2278 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2280 port->active_iface = bond_choose_iface(port);
2281 port->active_iface_tag = tag_create_random();
2282 if (port->active_iface >= 0) {
2283 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2284 port->name, port->ifaces[port->active_iface]->name);
2286 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2292 bond_enable_slave(struct iface *iface, bool enable)
2294 struct port *port = iface->port;
2295 struct bridge *br = port->bridge;
2297 /* This acts as a recursion check. If the act of disabling a slave
2298 * causes a different slave to be enabled, the flag will allow us to
2299 * skip redundant work when we reenter this function. It must be
2300 * cleared on exit to keep things safe with multiple bonds. */
2301 static bool moving_active_iface = false;
2303 iface->delay_expires = LLONG_MAX;
2304 if (enable == iface->enabled) {
2308 iface->enabled = enable;
2309 if (!iface->enabled) {
2310 VLOG_WARN("interface %s: disabled", iface->name);
2311 ofproto_revalidate(br->ofproto, iface->tag);
2312 if (iface->port_ifidx == port->active_iface) {
2313 ofproto_revalidate(br->ofproto,
2314 port->active_iface_tag);
2316 /* Disabling a slave can lead to another slave being immediately
2317 * enabled if there will be no active slaves but one is waiting
2318 * on an updelay. In this case we do not need to run most of the
2319 * code for the newly enabled slave since there was no period
2320 * without an active slave and it is redundant with the disabling
2322 moving_active_iface = true;
2323 bond_choose_active_iface(port);
2325 bond_send_learning_packets(port);
2327 VLOG_WARN("interface %s: enabled", iface->name);
2328 if (port->active_iface < 0 && !moving_active_iface) {
2329 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2330 bond_choose_active_iface(port);
2331 bond_send_learning_packets(port);
2333 iface->tag = tag_create_random();
2336 moving_active_iface = false;
2337 port->bond_compat_is_stale = true;
2340 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2341 * bond interface. */
2343 bond_update_fake_iface_stats(struct port *port)
2345 struct netdev_stats bond_stats;
2346 struct netdev *bond_dev;
2349 memset(&bond_stats, 0, sizeof bond_stats);
2351 for (i = 0; i < port->n_ifaces; i++) {
2352 struct netdev_stats slave_stats;
2354 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2355 /* XXX: We swap the stats here because they are swapped back when
2356 * reported by the internal device. The reason for this is
2357 * internal devices normally represent packets going into the system
2358 * but when used as fake bond device they represent packets leaving
2359 * the system. We really should do this in the internal device
2360 * itself because changing it here reverses the counts from the
2361 * perspective of the switch. However, the internal device doesn't
2362 * know what type of device it represents so we have to do it here
2364 bond_stats.tx_packets += slave_stats.rx_packets;
2365 bond_stats.tx_bytes += slave_stats.rx_bytes;
2366 bond_stats.rx_packets += slave_stats.tx_packets;
2367 bond_stats.rx_bytes += slave_stats.tx_bytes;
2371 if (!netdev_open_default(port->name, &bond_dev)) {
2372 netdev_set_stats(bond_dev, &bond_stats);
2373 netdev_close(bond_dev);
2378 bond_link_carrier_update(struct iface *iface, bool carrier)
2380 if (carrier == iface->up) {
2384 if (iface->lacp_status == LACP_STATUS_CURRENT) {
2385 iface_set_lacp_expired(iface);
2388 iface->up = carrier;
2390 iface->port->bond_compat_is_stale = true;
2394 bond_run(struct bridge *br)
2398 for (i = 0; i < br->n_ports; i++) {
2399 struct port *port = br->ports[i];
2401 if (port->n_ifaces >= 2) {
2404 if (port->monitor) {
2405 assert(!port->miimon);
2407 /* Track carrier going up and down on interfaces. */
2408 while (!netdev_monitor_poll(port->monitor, &devname)) {
2409 struct iface *iface;
2411 iface = port_lookup_iface(port, devname);
2413 bool up = netdev_get_carrier(iface->netdev);
2414 bond_link_carrier_update(iface, up);
2419 assert(port->miimon);
2421 if (time_msec() >= port->bond_miimon_next_update) {
2422 for (j = 0; j < port->n_ifaces; j++) {
2423 struct iface *iface = port->ifaces[j];
2424 bool up = netdev_get_miimon(iface->netdev);
2425 bond_link_carrier_update(iface, up);
2427 port->bond_miimon_next_update = time_msec() +
2428 port->bond_miimon_interval;
2432 for (j = 0; j < port->n_ifaces; j++) {
2433 bond_link_status_update(port->ifaces[j]);
2436 for (j = 0; j < port->n_ifaces; j++) {
2437 struct iface *iface = port->ifaces[j];
2438 if (time_msec() >= iface->delay_expires) {
2439 bond_enable_slave(iface, !iface->enabled);
2443 if (port->bond_fake_iface
2444 && time_msec() >= port->bond_next_fake_iface_update) {
2445 bond_update_fake_iface_stats(port);
2446 port->bond_next_fake_iface_update = time_msec() + 1000;
2450 if (port->bond_compat_is_stale) {
2451 port->bond_compat_is_stale = false;
2452 port_update_bond_compat(port);
2458 bond_wait(struct bridge *br)
2462 for (i = 0; i < br->n_ports; i++) {
2463 struct port *port = br->ports[i];
2464 if (port->n_ifaces < 2) {
2468 if (port->monitor) {
2469 netdev_monitor_poll_wait(port->monitor);
2473 poll_timer_wait_until(port->bond_miimon_next_update);
2476 for (j = 0; j < port->n_ifaces; j++) {
2477 struct iface *iface = port->ifaces[j];
2478 if (iface->delay_expires != LLONG_MAX) {
2479 poll_timer_wait_until(iface->delay_expires);
2482 if (port->bond_fake_iface) {
2483 poll_timer_wait_until(port->bond_next_fake_iface_update);
2489 set_dst(struct dst *dst, const struct flow *flow,
2490 const struct port *in_port, const struct port *out_port,
2493 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2494 : in_port->vlan >= 0 ? in_port->vlan
2495 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2496 : vlan_tci_to_vid(flow->vlan_tci));
2497 return choose_output_iface(out_port, flow, dst->vlan,
2498 &dst->dp_ifidx, tags);
2502 swap_dst(struct dst *p, struct dst *q)
2504 struct dst tmp = *p;
2509 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2510 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2511 * that we push to the datapath. We could in fact fully sort the array by
2512 * vlan, but in most cases there are at most two different vlan tags so that's
2513 * possibly overkill.) */
2515 partition_dsts(struct dst_set *set, int vlan)
2517 struct dst *first = set->dsts;
2518 struct dst *last = set->dsts + set->n;
2520 while (first != last) {
2522 * - All dsts < first have vlan == 'vlan'.
2523 * - All dsts >= last have vlan != 'vlan'.
2524 * - first < last. */
2525 while (first->vlan == vlan) {
2526 if (++first == last) {
2531 /* Same invariants, plus one additional:
2532 * - first->vlan != vlan.
2534 while (last[-1].vlan != vlan) {
2535 if (--last == first) {
2540 /* Same invariants, plus one additional:
2541 * - last[-1].vlan == vlan.*/
2542 swap_dst(first++, --last);
2547 mirror_mask_ffs(mirror_mask_t mask)
2549 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2554 dst_set_init(struct dst_set *set)
2556 set->dsts = set->builtin;
2558 set->allocated = ARRAY_SIZE(set->builtin);
2562 dst_set_add(struct dst_set *set, const struct dst *dst)
2564 if (set->n >= set->allocated) {
2565 size_t new_allocated;
2566 struct dst *new_dsts;
2568 new_allocated = set->allocated * 2;
2569 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2570 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2574 set->dsts = new_dsts;
2575 set->allocated = new_allocated;
2577 set->dsts[set->n++] = *dst;
2581 dst_set_free(struct dst_set *set)
2583 if (set->dsts != set->builtin) {
2589 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2592 for (i = 0; i < set->n; i++) {
2593 if (set->dsts[i].vlan == test->vlan
2594 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2602 port_trunks_vlan(const struct port *port, uint16_t vlan)
2604 return (port->vlan < 0
2605 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2609 port_includes_vlan(const struct port *port, uint16_t vlan)
2611 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2615 port_is_floodable(const struct port *port)
2619 for (i = 0; i < port->n_ifaces; i++) {
2620 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2621 port->ifaces[i]->dp_ifidx)) {
2629 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2630 const struct port *in_port, const struct port *out_port,
2631 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2633 mirror_mask_t mirrors = in_port->src_mirrors;
2638 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2639 if (flow_vlan == 0) {
2640 flow_vlan = OFP_VLAN_NONE;
2643 if (out_port == FLOOD_PORT) {
2644 for (i = 0; i < br->n_ports; i++) {
2645 struct port *port = br->ports[i];
2647 && port_is_floodable(port)
2648 && port_includes_vlan(port, vlan)
2649 && !port->is_mirror_output_port
2650 && set_dst(&dst, flow, in_port, port, tags)) {
2651 mirrors |= port->dst_mirrors;
2652 dst_set_add(set, &dst);
2655 *nf_output_iface = NF_OUT_FLOOD;
2656 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2657 dst_set_add(set, &dst);
2658 *nf_output_iface = dst.dp_ifidx;
2659 mirrors |= out_port->dst_mirrors;
2663 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2664 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2666 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2667 && !dst_is_duplicate(set, &dst)) {
2668 dst_set_add(set, &dst);
2671 for (i = 0; i < br->n_ports; i++) {
2672 struct port *port = br->ports[i];
2673 if (port_includes_vlan(port, m->out_vlan)
2674 && set_dst(&dst, flow, in_port, port, tags))
2676 if (port->vlan < 0) {
2677 dst.vlan = m->out_vlan;
2679 if (dst_is_duplicate(set, &dst)) {
2683 /* Use the vlan tag on the original flow instead of
2684 * the one passed in the vlan parameter. This ensures
2685 * that we compare the vlan from before any implicit
2686 * tagging tags place. This is necessary because
2687 * dst->vlan is the final vlan, after removing implicit
2689 if (port == in_port && dst.vlan == flow_vlan) {
2690 /* Don't send out input port on same VLAN. */
2693 dst_set_add(set, &dst);
2698 mirrors &= mirrors - 1;
2701 partition_dsts(set, flow_vlan);
2704 static void OVS_UNUSED
2705 print_dsts(const struct dst_set *set)
2709 for (i = 0; i < set->n; i++) {
2710 const struct dst *dst = &set->dsts[i];
2712 printf(">p%"PRIu16, dst->dp_ifidx);
2713 if (dst->vlan != OFP_VLAN_NONE) {
2714 printf("v%"PRIu16, dst->vlan);
2720 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2721 const struct port *in_port, const struct port *out_port,
2722 tag_type *tags, struct ofpbuf *actions,
2723 uint16_t *nf_output_iface)
2730 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2733 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2734 if (cur_vlan == 0) {
2735 cur_vlan = OFP_VLAN_NONE;
2737 for (i = 0; i < set.n; i++) {
2738 const struct dst *dst = &set.dsts[i];
2739 if (dst->vlan != cur_vlan) {
2740 if (dst->vlan == OFP_VLAN_NONE) {
2741 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2744 tci = htons(dst->vlan & VLAN_VID_MASK);
2745 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2746 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2748 cur_vlan = dst->vlan;
2750 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2755 /* Returns the effective vlan of a packet, taking into account both the
2756 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2757 * the packet is untagged and -1 indicates it has an invalid header and
2758 * should be dropped. */
2759 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2760 struct port *in_port, bool have_packet)
2762 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2763 if (in_port->vlan >= 0) {
2765 /* XXX support double tagging? */
2767 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2768 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2769 "packet received on port %s configured with "
2770 "implicit VLAN %"PRIu16,
2771 br->name, vlan, in_port->name, in_port->vlan);
2775 vlan = in_port->vlan;
2777 if (!port_includes_vlan(in_port, vlan)) {
2779 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2780 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2781 "packet received on port %s not configured for "
2783 br->name, vlan, in_port->name, vlan);
2792 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2793 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2794 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2796 is_gratuitous_arp(const struct flow *flow)
2798 return (flow->dl_type == htons(ETH_TYPE_ARP)
2799 && eth_addr_is_broadcast(flow->dl_dst)
2800 && (flow->nw_proto == ARP_OP_REPLY
2801 || (flow->nw_proto == ARP_OP_REQUEST
2802 && flow->nw_src == flow->nw_dst)));
2806 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2807 struct port *in_port)
2809 enum grat_arp_lock_type lock_type;
2812 /* We don't want to learn from gratuitous ARP packets that are reflected
2813 * back over bond slaves so we lock the learning table. */
2814 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2815 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2816 GRAT_ARP_LOCK_CHECK;
2818 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2821 /* The log messages here could actually be useful in debugging,
2822 * so keep the rate limit relatively high. */
2823 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2825 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2826 "on port %s in VLAN %d",
2827 br->name, ETH_ADDR_ARGS(flow->dl_src),
2828 in_port->name, vlan);
2829 ofproto_revalidate(br->ofproto, rev_tag);
2833 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2834 * dropped. Returns true if they may be forwarded, false if they should be
2837 * If 'have_packet' is true, it indicates that the caller is processing a
2838 * received packet. If 'have_packet' is false, then the caller is just
2839 * revalidating an existing flow because configuration has changed. Either
2840 * way, 'have_packet' only affects logging (there is no point in logging errors
2841 * during revalidation).
2843 * Sets '*in_portp' to the input port. This will be a null pointer if
2844 * flow->in_port does not designate a known input port (in which case
2845 * is_admissible() returns false).
2847 * When returning true, sets '*vlanp' to the effective VLAN of the input
2848 * packet, as returned by flow_get_vlan().
2850 * May also add tags to '*tags', although the current implementation only does
2851 * so in one special case.
2854 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2855 tag_type *tags, int *vlanp, struct port **in_portp)
2857 struct iface *in_iface;
2858 struct port *in_port;
2861 /* Find the interface and port structure for the received packet. */
2862 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2864 /* No interface? Something fishy... */
2866 /* Odd. A few possible reasons here:
2868 * - We deleted an interface but there are still a few packets
2869 * queued up from it.
2871 * - Someone externally added an interface (e.g. with "ovs-dpctl
2872 * add-if") that we don't know about.
2874 * - Packet arrived on the local port but the local port is not
2875 * one of our bridge ports.
2877 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2879 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2880 "interface %"PRIu16, br->name, flow->in_port);
2886 *in_portp = in_port = in_iface->port;
2887 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2892 /* Drop frames for reserved multicast addresses. */
2893 if (eth_addr_is_reserved(flow->dl_dst)) {
2897 /* Drop frames on ports reserved for mirroring. */
2898 if (in_port->is_mirror_output_port) {
2900 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2901 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2902 "%s, which is reserved exclusively for mirroring",
2903 br->name, in_port->name);
2908 /* When using LACP, do not accept packets from disabled interfaces. */
2909 if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) {
2913 /* Packets received on non-LACP bonds need special attention to avoid
2915 if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) {
2917 bool is_grat_arp_locked;
2919 if (eth_addr_is_multicast(flow->dl_dst)) {
2920 *tags |= in_port->active_iface_tag;
2921 if (in_port->active_iface != in_iface->port_ifidx) {
2922 /* Drop all multicast packets on inactive slaves. */
2927 /* Drop all packets for which we have learned a different input
2928 * port, because we probably sent the packet on one slave and got
2929 * it back on the other. Gratuitous ARP packets are an exception
2930 * to this rule: the host has moved to another switch. The exception
2931 * to the exception is if we locked the learning table to avoid
2932 * reflections on bond slaves. If this is the case, just drop the
2934 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2935 &is_grat_arp_locked);
2936 if (src_idx != -1 && src_idx != in_port->port_idx &&
2937 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2945 /* If the composed actions may be applied to any packet in the given 'flow',
2946 * returns true. Otherwise, the actions should only be applied to 'packet', or
2947 * not at all, if 'packet' was NULL. */
2949 process_flow(struct bridge *br, const struct flow *flow,
2950 const struct ofpbuf *packet, struct ofpbuf *actions,
2951 tag_type *tags, uint16_t *nf_output_iface)
2953 struct port *in_port;
2954 struct port *out_port;
2958 /* Check whether we should drop packets in this flow. */
2959 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2964 /* Learn source MAC (but don't try to learn from revalidation). */
2966 update_learning_table(br, flow, vlan, in_port);
2969 /* Determine output port. */
2970 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2972 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2973 out_port = br->ports[out_port_idx];
2974 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2975 /* If we are revalidating but don't have a learning entry then
2976 * eject the flow. Installing a flow that floods packets opens
2977 * up a window of time where we could learn from a packet reflected
2978 * on a bond and blackhole packets before the learning table is
2979 * updated to reflect the correct port. */
2982 out_port = FLOOD_PORT;
2985 /* Don't send packets out their input ports. */
2986 if (in_port == out_port) {
2992 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3000 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3001 struct ofpbuf *actions, tag_type *tags,
3002 uint16_t *nf_output_iface, void *br_)
3004 struct iface *iface;
3005 struct bridge *br = br_;
3007 COVERAGE_INC(bridge_process_flow);
3009 iface = iface_from_dp_ifidx(br, flow->in_port);
3011 if (cfm_should_process_flow(flow)) {
3012 if (packet && iface->cfm) {
3013 cfm_process_heartbeat(iface->cfm, packet);
3016 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3018 lacp_process_packet(packet, iface);
3023 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3027 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3028 const struct nlattr *actions,
3030 unsigned long long int n_bytes, void *br_)
3032 struct bridge *br = br_;
3033 const struct nlattr *a;
3034 struct port *in_port;
3039 /* Feed information from the active flows back into the learning table to
3040 * ensure that table is always in sync with what is actually flowing
3041 * through the datapath.
3043 * We test that 'tags' is nonzero to ensure that only flows that include an
3044 * OFPP_NORMAL action are used for learning. This works because
3045 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3046 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3047 update_learning_table(br, flow, vlan, in_port);
3050 /* Account for bond slave utilization. */
3051 if (!br->has_bonded_ports) {
3054 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3055 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3056 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3057 if (out_port && out_port->n_ifaces >= 2 &&
3058 out_port->bond_mode == BM_SLB) {
3059 uint16_t vlan = (flow->vlan_tci
3060 ? vlan_tci_to_vid(flow->vlan_tci)
3062 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3063 e->tx_bytes += n_bytes;
3070 bridge_account_checkpoint_ofhook_cb(void *br_)
3072 struct bridge *br = br_;
3076 if (!br->has_bonded_ports) {
3081 for (i = 0; i < br->n_ports; i++) {
3082 struct port *port = br->ports[i];
3083 if (port->n_ifaces > 1 && port->bond_mode == BM_SLB
3084 && now >= port->bond_next_rebalance) {
3085 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3086 bond_rebalance_port(port);
3091 static struct ofhooks bridge_ofhooks = {
3092 bridge_normal_ofhook_cb,
3093 bridge_account_flow_ofhook_cb,
3094 bridge_account_checkpoint_ofhook_cb,
3097 /* LACP functions. */
3100 lacp_process_packet(const struct ofpbuf *packet, struct iface *iface)
3102 const struct lacp_pdu *pdu;
3104 if (!iface->port->lacp) {
3108 pdu = parse_lacp_packet(packet);
3113 iface->lacp_status = LACP_STATUS_CURRENT;
3114 iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX;
3116 iface->lacp_actor.state = iface_get_lacp_state(iface);
3117 if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) {
3121 if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) {
3122 iface->port->lacp_need_update = true;
3123 iface->lacp_partner = pdu->actor;
3128 lacp_update_ifaces(struct port *port)
3132 struct lacp_info lead_pri;
3133 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3135 port->lacp_need_update = false;
3136 COVERAGE_INC(bridge_lacp_update);
3142 VLOG_DBG_RL(&rl, "port %s: re-evaluating LACP link status", port->name);
3145 for (i = 0; i < port->n_ifaces; i++) {
3146 struct iface *iface = port->ifaces[i];
3147 struct lacp_info pri;
3149 iface->lacp_attached = true;
3150 ofproto_revalidate(port->bridge->ofproto, iface->tag);
3152 /* Don't allow loopback interfaces to send traffic or lead. */
3153 if (eth_addr_equals(iface->lacp_partner.sysid,
3154 iface->lacp_actor.sysid)) {
3155 VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is "
3156 "connected to its own bridge", iface->name);
3157 iface->lacp_attached = false;
3161 if (iface->lacp_status == LACP_STATUS_DEFAULTED) {
3165 iface_get_lacp_priority(iface, &pri);
3167 if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) {
3174 port->lacp &= ~LACP_NEGOTIATED;
3178 port->lacp |= LACP_NEGOTIATED;
3180 for (i = 0; i < port->n_ifaces; i++) {
3181 struct iface *iface = port->ifaces[i];
3183 if (iface->lacp_status == LACP_STATUS_DEFAULTED
3184 || lead->lacp_partner.key != iface->lacp_partner.key
3185 || !eth_addr_equals(lead->lacp_partner.sysid,
3186 iface->lacp_partner.sysid)) {
3187 iface->lacp_attached = false;
3193 lacp_iface_may_tx(const struct iface *iface)
3195 return iface->port->lacp & LACP_ACTIVE
3196 || iface->lacp_status != LACP_STATUS_DEFAULTED;
3200 lacp_run(struct bridge *br)
3203 struct ofpbuf packet;
3205 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3207 for (i = 0; i < br->n_ports; i++) {
3208 struct port *port = br->ports[i];
3214 for (j = 0; j < port->n_ifaces; j++) {
3215 struct iface *iface = port->ifaces[j];
3217 if (time_msec() > iface->lacp_rx) {
3218 if (iface->lacp_status == LACP_STATUS_CURRENT) {
3219 iface_set_lacp_expired(iface);
3220 } else if (iface->lacp_status == LACP_STATUS_EXPIRED) {
3221 iface_set_lacp_defaulted(iface);
3226 if (port->lacp_need_update) {
3227 lacp_update_ifaces(port);
3230 for (j = 0; j < port->n_ifaces; j++) {
3231 struct iface *iface = port->ifaces[j];
3232 uint8_t ea[ETH_ADDR_LEN];
3235 if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) {
3239 error = netdev_get_etheraddr(iface->netdev, ea);
3241 iface->lacp_actor.state = iface_get_lacp_state(iface);
3242 compose_lacp_packet(&packet, &iface->lacp_actor,
3243 &iface->lacp_partner, ea);
3244 iface_send_packet(iface, &packet);
3246 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3247 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3248 "(%s)", iface->name, strerror(error));
3251 iface->lacp_tx = time_msec() +
3252 (iface->lacp_partner.state & LACP_STATE_TIME
3254 : LACP_SLOW_TIME_TX);
3257 ofpbuf_uninit(&packet);
3261 lacp_wait(struct bridge *br)
3265 for (i = 0; i < br->n_ports; i++) {
3266 struct port *port = br->ports[i];
3272 for (j = 0; j < port->n_ifaces; j++) {
3273 struct iface *iface = port->ifaces[j];
3275 if (lacp_iface_may_tx(iface)) {
3276 poll_timer_wait_until(iface->lacp_tx);
3279 if (iface->lacp_status != LACP_STATUS_DEFAULTED) {
3280 poll_timer_wait_until(iface->lacp_rx);
3286 /* Bonding functions. */
3288 /* Statistics for a single interface on a bonded port, used for load-based
3289 * bond rebalancing. */
3290 struct slave_balance {
3291 struct iface *iface; /* The interface. */
3292 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3294 /* All the "bond_entry"s that are assigned to this interface, in order of
3295 * increasing tx_bytes. */
3296 struct bond_entry **hashes;
3301 bond_mode_to_string(enum bond_mode bm) {
3302 static char *bm_slb = "balance-slb";
3303 static char *bm_ab = "active-backup";
3304 static char *bm_tcp = "balance-tcp";
3307 case BM_SLB: return bm_slb;
3308 case BM_AB: return bm_ab;
3309 case BM_TCP: return bm_tcp;
3316 /* Sorts pointers to pointers to bond_entries in ascending order by the
3317 * interface to which they are assigned, and within a single interface in
3318 * ascending order of bytes transmitted. */
3320 compare_bond_entries(const void *a_, const void *b_)
3322 const struct bond_entry *const *ap = a_;
3323 const struct bond_entry *const *bp = b_;
3324 const struct bond_entry *a = *ap;
3325 const struct bond_entry *b = *bp;
3326 if (a->iface_idx != b->iface_idx) {
3327 return a->iface_idx > b->iface_idx ? 1 : -1;
3328 } else if (a->tx_bytes != b->tx_bytes) {
3329 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3335 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3336 * *descending* order by number of bytes transmitted. */
3338 compare_slave_balance(const void *a_, const void *b_)
3340 const struct slave_balance *a = a_;
3341 const struct slave_balance *b = b_;
3342 if (a->iface->enabled != b->iface->enabled) {
3343 return a->iface->enabled ? -1 : 1;
3344 } else if (a->tx_bytes != b->tx_bytes) {
3345 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3352 swap_bals(struct slave_balance *a, struct slave_balance *b)
3354 struct slave_balance tmp = *a;
3359 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3360 * given that 'p' (and only 'p') might be in the wrong location.
3362 * This function invalidates 'p', since it might now be in a different memory
3365 resort_bals(struct slave_balance *p,
3366 struct slave_balance bals[], size_t n_bals)
3369 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3370 swap_bals(p, p - 1);
3372 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3373 swap_bals(p, p + 1);
3379 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3381 if (VLOG_IS_DBG_ENABLED()) {
3382 struct ds ds = DS_EMPTY_INITIALIZER;
3383 const struct slave_balance *b;
3385 for (b = bals; b < bals + n_bals; b++) {
3389 ds_put_char(&ds, ',');
3391 ds_put_format(&ds, " %s %"PRIu64"kB",
3392 b->iface->name, b->tx_bytes / 1024);
3394 if (!b->iface->enabled) {
3395 ds_put_cstr(&ds, " (disabled)");
3397 if (b->n_hashes > 0) {
3398 ds_put_cstr(&ds, " (");
3399 for (i = 0; i < b->n_hashes; i++) {
3400 const struct bond_entry *e = b->hashes[i];
3402 ds_put_cstr(&ds, " + ");
3404 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3405 e - port->bond_hash, e->tx_bytes / 1024);
3407 ds_put_cstr(&ds, ")");
3410 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3415 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3417 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3420 struct bond_entry *hash = from->hashes[hash_idx];
3421 struct port *port = from->iface->port;
3422 uint64_t delta = hash->tx_bytes;
3424 assert(port->bond_mode == BM_SLB);
3426 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3427 "from %s to %s (now carrying %"PRIu64"kB and "
3428 "%"PRIu64"kB load, respectively)",
3429 port->name, delta / 1024, hash - port->bond_hash,
3430 from->iface->name, to->iface->name,
3431 (from->tx_bytes - delta) / 1024,
3432 (to->tx_bytes + delta) / 1024);
3434 /* Delete element from from->hashes.
3436 * We don't bother to add the element to to->hashes because not only would
3437 * it require more work, the only purpose it would be to allow that hash to
3438 * be migrated to another slave in this rebalancing run, and there is no
3439 * point in doing that. */
3440 if (hash_idx == 0) {
3443 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3444 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3448 /* Shift load away from 'from' to 'to'. */
3449 from->tx_bytes -= delta;
3450 to->tx_bytes += delta;
3452 /* Arrange for flows to be revalidated. */
3453 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3454 hash->iface_idx = to->iface->port_ifidx;
3455 hash->iface_tag = tag_create_random();
3459 bond_rebalance_port(struct port *port)
3461 struct slave_balance *bals;
3463 struct bond_entry *hashes[BOND_MASK + 1];
3464 struct slave_balance *b, *from, *to;
3465 struct bond_entry *e;
3468 assert(port->bond_mode != BM_AB);
3470 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3471 * descending order of tx_bytes, so that bals[0] represents the most
3472 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3475 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3476 * array for each slave_balance structure, we sort our local array of
3477 * hashes in order by slave, so that all of the hashes for a given slave
3478 * become contiguous in memory, and then we point each 'hashes' members of
3479 * a slave_balance structure to the start of a contiguous group. */
3480 n_bals = port->n_ifaces;
3481 bals = xmalloc(n_bals * sizeof *bals);
3482 for (b = bals; b < &bals[n_bals]; b++) {
3483 b->iface = port->ifaces[b - bals];
3488 for (i = 0; i <= BOND_MASK; i++) {
3489 hashes[i] = &port->bond_hash[i];
3491 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3492 for (i = 0; i <= BOND_MASK; i++) {
3494 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3495 b = &bals[e->iface_idx];
3496 b->tx_bytes += e->tx_bytes;
3498 b->hashes = &hashes[i];
3503 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3504 log_bals(bals, n_bals, port);
3506 /* Discard slaves that aren't enabled (which were sorted to the back of the
3507 * array earlier). */
3508 while (!bals[n_bals - 1].iface->enabled) {
3515 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3516 to = &bals[n_bals - 1];
3517 for (from = bals; from < to; ) {
3518 uint64_t overload = from->tx_bytes - to->tx_bytes;
3519 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3520 /* The extra load on 'from' (and all less-loaded slaves), compared
3521 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3522 * it is less than ~1Mbps. No point in rebalancing. */
3524 } else if (from->n_hashes == 1) {
3525 /* 'from' only carries a single MAC hash, so we can't shift any
3526 * load away from it, even though we want to. */
3529 /* 'from' is carrying significantly more load than 'to', and that
3530 * load is split across at least two different hashes. Pick a hash
3531 * to migrate to 'to' (the least-loaded slave), given that doing so
3532 * must decrease the ratio of the load on the two slaves by at
3535 * The sort order we use means that we prefer to shift away the
3536 * smallest hashes instead of the biggest ones. There is little
3537 * reason behind this decision; we could use the opposite sort
3538 * order to shift away big hashes ahead of small ones. */
3541 for (i = 0; i < from->n_hashes; i++) {
3542 double old_ratio, new_ratio;
3543 uint64_t delta = from->hashes[i]->tx_bytes;
3545 if (delta == 0 || from->tx_bytes - delta == 0) {
3546 /* Pointless move. */
3550 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3552 if (to->tx_bytes == 0) {
3553 /* Nothing on the new slave, move it. */
3557 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3558 new_ratio = (double)(from->tx_bytes - delta) /
3559 (to->tx_bytes + delta);
3561 if (new_ratio == 0) {
3562 /* Should already be covered but check to prevent division
3567 if (new_ratio < 1) {
3568 new_ratio = 1 / new_ratio;
3571 if (old_ratio - new_ratio > 0.1) {
3572 /* Would decrease the ratio, move it. */
3576 if (i < from->n_hashes) {
3577 bond_shift_load(from, to, i);
3578 port->bond_compat_is_stale = true;
3580 /* If the result of the migration changed the relative order of
3581 * 'from' and 'to' swap them back to maintain invariants. */
3582 if (order_swapped) {
3583 swap_bals(from, to);
3586 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3587 * point to different slave_balance structures. It is only
3588 * valid to do these two operations in a row at all because we
3589 * know that 'from' will not move past 'to' and vice versa. */
3590 resort_bals(from, bals, n_bals);
3591 resort_bals(to, bals, n_bals);
3598 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3599 * historical data to decay to <1% in 7 rebalancing runs. */
3600 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3609 bond_send_learning_packets(struct port *port)
3611 struct bridge *br = port->bridge;
3612 struct mac_entry *e;
3613 struct ofpbuf packet;
3614 int error, n_packets, n_errors;
3616 if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
3620 ofpbuf_init(&packet, 128);
3621 error = n_packets = n_errors = 0;
3622 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3623 union ofp_action actions[2], *a;
3629 if (e->port == port->port_idx) {
3633 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3635 flow_extract(&packet, 0, ODPP_NONE, &flow);
3637 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3641 /* Compose actions. */
3642 memset(actions, 0, sizeof actions);
3645 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3646 a->vlan_vid.len = htons(sizeof *a);
3647 a->vlan_vid.vlan_vid = htons(e->vlan);
3650 a->output.type = htons(OFPAT_OUTPUT);
3651 a->output.len = htons(sizeof *a);
3652 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3657 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3664 ofpbuf_uninit(&packet);
3667 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3668 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3669 "packets, last error was: %s",
3670 port->name, n_errors, n_packets, strerror(error));
3672 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3673 port->name, n_packets);
3677 /* Bonding unixctl user interface functions. */
3680 bond_unixctl_list(struct unixctl_conn *conn,
3681 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3683 struct ds ds = DS_EMPTY_INITIALIZER;
3684 const struct bridge *br;
3686 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3688 LIST_FOR_EACH (br, node, &all_bridges) {
3691 for (i = 0; i < br->n_ports; i++) {
3692 const struct port *port = br->ports[i];
3693 if (port->n_ifaces > 1) {
3696 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3697 bond_mode_to_string(port->bond_mode));
3698 for (j = 0; j < port->n_ifaces; j++) {
3699 const struct iface *iface = port->ifaces[j];
3701 ds_put_cstr(&ds, ", ");
3703 ds_put_cstr(&ds, iface->name);
3705 ds_put_char(&ds, '\n');
3709 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3713 static struct port *
3714 bond_find(const char *name)
3716 const struct bridge *br;
3718 LIST_FOR_EACH (br, node, &all_bridges) {
3721 for (i = 0; i < br->n_ports; i++) {
3722 struct port *port = br->ports[i];
3723 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3732 ds_put_lacp_state(struct ds *ds, uint8_t state)
3734 if (state & LACP_STATE_ACT) {
3735 ds_put_cstr(ds, "activity ");
3738 if (state & LACP_STATE_TIME) {
3739 ds_put_cstr(ds, "timeout ");
3742 if (state & LACP_STATE_AGG) {
3743 ds_put_cstr(ds, "aggregation ");
3746 if (state & LACP_STATE_SYNC) {
3747 ds_put_cstr(ds, "synchronized ");
3750 if (state & LACP_STATE_COL) {
3751 ds_put_cstr(ds, "collecting ");
3754 if (state & LACP_STATE_DIST) {
3755 ds_put_cstr(ds, "distributing ");
3758 if (state & LACP_STATE_DEF) {
3759 ds_put_cstr(ds, "defaulted ");
3762 if (state & LACP_STATE_EXP) {
3763 ds_put_cstr(ds, "expired ");
3768 bond_unixctl_show(struct unixctl_conn *conn,
3769 const char *args, void *aux OVS_UNUSED)
3771 struct ds ds = DS_EMPTY_INITIALIZER;
3772 const struct port *port;
3775 port = bond_find(args);
3777 unixctl_command_reply(conn, 501, "no such bond");
3781 ds_put_format(&ds, "bond_mode: %s\n",
3782 bond_mode_to_string(port->bond_mode));
3785 ds_put_format(&ds, "lacp: %s\n",
3786 port->lacp & LACP_ACTIVE ? "active" : "passive");
3788 ds_put_cstr(&ds, "lacp: off\n");
3791 if (port->bond_mode != BM_AB) {
3792 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3793 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3797 ds_put_format(&ds, "bond-detect-mode: %s\n",
3798 port->miimon ? "miimon" : "carrier");
3801 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3802 port->bond_miimon_interval);
3805 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3806 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3808 if (port->bond_mode != BM_AB) {
3809 ds_put_format(&ds, "next rebalance: %lld ms\n",
3810 port->bond_next_rebalance - time_msec());
3813 for (j = 0; j < port->n_ifaces; j++) {
3814 const struct iface *iface = port->ifaces[j];
3815 struct bond_entry *be;
3819 ds_put_format(&ds, "\nslave %s: %s\n",
3820 iface->name, iface->enabled ? "enabled" : "disabled");
3821 if (j == port->active_iface) {
3822 ds_put_cstr(&ds, "\tactive slave\n");
3824 if (iface->delay_expires != LLONG_MAX) {
3825 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3826 iface->enabled ? "downdelay" : "updelay",
3827 iface->delay_expires - time_msec());
3831 ds_put_cstr(&ds, "\tstatus: ");
3833 if (iface->lacp_status == LACP_STATUS_CURRENT) {
3834 ds_put_cstr(&ds, "current ");
3835 } else if (iface->lacp_status == LACP_STATUS_EXPIRED) {
3836 ds_put_cstr(&ds, "expired ");
3838 ds_put_cstr(&ds, "defaulted ");
3841 if (iface->lacp_attached) {
3842 ds_put_cstr(&ds, "attached ");
3845 ds_put_cstr(&ds, "\n");
3847 ds_put_cstr(&ds, "\n\tactor sysid: ");
3848 ds_put_format(&ds, ETH_ADDR_FMT,
3849 ETH_ADDR_ARGS(iface->lacp_actor.sysid));
3850 ds_put_cstr(&ds, "\n");
3852 ds_put_format(&ds, "\tactor sys_priority: %u\n",
3853 ntohs(iface->lacp_actor.sys_priority));
3855 ds_put_format(&ds, "\tactor portid: %u\n",
3856 ntohs(iface->lacp_actor.portid));
3858 ds_put_format(&ds, "\tactor port_priority: %u\n",
3859 ntohs(iface->lacp_actor.port_priority));
3861 ds_put_format(&ds, "\tactor key: %u\n",
3862 ntohs(iface->lacp_actor.key));
3864 ds_put_cstr(&ds, "\tactor state: ");
3865 ds_put_lacp_state(&ds, iface_get_lacp_state(iface));
3866 ds_put_cstr(&ds, "\n\n");
3868 ds_put_cstr(&ds, "\tpartner sysid: ");
3869 ds_put_format(&ds, ETH_ADDR_FMT,
3870 ETH_ADDR_ARGS(iface->lacp_partner.sysid));
3871 ds_put_cstr(&ds, "\n");
3873 ds_put_format(&ds, "\tpartner sys_priority: %u\n",
3874 ntohs(iface->lacp_partner.sys_priority));
3876 ds_put_format(&ds, "\tpartner portid: %u\n",
3877 ntohs(iface->lacp_partner.portid));
3879 ds_put_format(&ds, "\tpartner port_priority: %u\n",
3880 ntohs(iface->lacp_partner.port_priority));
3882 ds_put_format(&ds, "\tpartner key: %u\n",
3883 ntohs(iface->lacp_partner.key));
3885 ds_put_cstr(&ds, "\tpartner state: ");
3886 ds_put_lacp_state(&ds, iface->lacp_partner.state);
3887 ds_put_cstr(&ds, "\n");
3890 if (port->bond_mode == BM_AB) {
3895 memset(&flow, 0, sizeof flow);
3896 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3897 int hash = be - port->bond_hash;
3898 struct mac_entry *me;
3900 if (be->iface_idx != j) {
3904 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3905 hash, be->tx_bytes / 1024);
3907 if (port->bond_mode != BM_SLB) {
3912 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3916 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3917 if (bond_hash_src(me->mac, me->vlan) == hash
3918 && me->port != port->port_idx
3919 && choose_output_iface(port, &flow, me->vlan,
3921 && dp_ifidx == iface->dp_ifidx)
3923 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3924 ETH_ADDR_ARGS(me->mac));
3929 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3934 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3935 void *aux OVS_UNUSED)
3937 char *args = (char *) args_;
3938 char *save_ptr = NULL;
3939 char *bond_s, *hash_s, *slave_s;
3941 struct iface *iface;
3942 struct bond_entry *entry;
3945 bond_s = strtok_r(args, " ", &save_ptr);
3946 hash_s = strtok_r(NULL, " ", &save_ptr);
3947 slave_s = strtok_r(NULL, " ", &save_ptr);
3949 unixctl_command_reply(conn, 501,
3950 "usage: bond/migrate BOND HASH SLAVE");
3954 port = bond_find(bond_s);
3956 unixctl_command_reply(conn, 501, "no such bond");
3960 if (port->bond_mode != BM_SLB) {
3961 unixctl_command_reply(conn, 501, "not an SLB bond");
3965 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3966 hash = atoi(hash_s) & BOND_MASK;
3968 unixctl_command_reply(conn, 501, "bad hash");
3972 iface = port_lookup_iface(port, slave_s);
3974 unixctl_command_reply(conn, 501, "no such slave");
3978 if (!iface->enabled) {
3979 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3983 entry = &port->bond_hash[hash];
3984 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3985 entry->iface_idx = iface->port_ifidx;
3986 entry->iface_tag = tag_create_random();
3987 port->bond_compat_is_stale = true;
3988 unixctl_command_reply(conn, 200, "migrated");
3992 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3993 void *aux OVS_UNUSED)
3995 char *args = (char *) args_;
3996 char *save_ptr = NULL;
3997 char *bond_s, *slave_s;
3999 struct iface *iface;
4001 bond_s = strtok_r(args, " ", &save_ptr);
4002 slave_s = strtok_r(NULL, " ", &save_ptr);
4004 unixctl_command_reply(conn, 501,
4005 "usage: bond/set-active-slave BOND SLAVE");
4009 port = bond_find(bond_s);
4011 unixctl_command_reply(conn, 501, "no such bond");
4015 iface = port_lookup_iface(port, slave_s);
4017 unixctl_command_reply(conn, 501, "no such slave");
4021 if (!iface->enabled) {
4022 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
4026 if (port->active_iface != iface->port_ifidx) {
4027 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4028 port->active_iface = iface->port_ifidx;
4029 port->active_iface_tag = tag_create_random();
4030 VLOG_INFO("port %s: active interface is now %s",
4031 port->name, iface->name);
4032 bond_send_learning_packets(port);
4033 unixctl_command_reply(conn, 200, "done");
4035 unixctl_command_reply(conn, 200, "no change");
4040 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
4042 char *args = (char *) args_;
4043 char *save_ptr = NULL;
4044 char *bond_s, *slave_s;
4046 struct iface *iface;
4048 bond_s = strtok_r(args, " ", &save_ptr);
4049 slave_s = strtok_r(NULL, " ", &save_ptr);
4051 unixctl_command_reply(conn, 501,
4052 "usage: bond/enable/disable-slave BOND SLAVE");
4056 port = bond_find(bond_s);
4058 unixctl_command_reply(conn, 501, "no such bond");
4062 iface = port_lookup_iface(port, slave_s);
4064 unixctl_command_reply(conn, 501, "no such slave");
4068 bond_enable_slave(iface, enable);
4069 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
4073 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
4074 void *aux OVS_UNUSED)
4076 enable_slave(conn, args, true);
4080 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
4081 void *aux OVS_UNUSED)
4083 enable_slave(conn, args, false);
4087 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
4088 void *aux OVS_UNUSED)
4090 char *args = (char *) args_;
4091 uint8_t mac[ETH_ADDR_LEN];
4095 char *mac_s, *vlan_s;
4096 char *save_ptr = NULL;
4098 mac_s = strtok_r(args, " ", &save_ptr);
4099 vlan_s = strtok_r(NULL, " ", &save_ptr);
4102 if (sscanf(vlan_s, "%u", &vlan) != 1) {
4103 unixctl_command_reply(conn, 501, "invalid vlan");
4107 vlan = OFP_VLAN_NONE;
4110 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
4111 == ETH_ADDR_SCAN_COUNT) {
4112 hash = bond_hash_src(mac, vlan);
4114 hash_cstr = xasprintf("%u", hash);
4115 unixctl_command_reply(conn, 200, hash_cstr);
4118 unixctl_command_reply(conn, 501, "invalid mac");
4125 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
4126 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
4127 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
4128 unixctl_command_register("bond/set-active-slave",
4129 bond_unixctl_set_active_slave, NULL);
4130 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
4132 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
4134 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
4137 /* Port functions. */
4139 static struct port *
4140 port_create(struct bridge *br, const char *name)
4144 port = xzalloc(sizeof *port);
4146 port->port_idx = br->n_ports;
4148 port->trunks = NULL;
4149 port->name = xstrdup(name);
4150 port->active_iface = -1;
4152 if (br->n_ports >= br->allocated_ports) {
4153 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
4156 br->ports[br->n_ports++] = port;
4157 shash_add_assert(&br->port_by_name, port->name, port);
4159 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
4166 get_port_other_config(const struct ovsrec_port *port, const char *key,
4167 const char *default_value)
4171 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
4173 return value ? value : default_value;
4177 get_interface_other_config(const struct ovsrec_interface *iface,
4178 const char *key, const char *default_value)
4182 value = get_ovsrec_key_value(&iface->header_,
4183 &ovsrec_interface_col_other_config, key);
4184 return value ? value : default_value;
4188 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
4190 struct shash new_ifaces;
4193 /* Collect list of new interfaces. */
4194 shash_init(&new_ifaces);
4195 for (i = 0; i < cfg->n_interfaces; i++) {
4196 const char *name = cfg->interfaces[i]->name;
4197 shash_add_once(&new_ifaces, name, NULL);
4200 /* Get rid of deleted interfaces. */
4201 for (i = 0; i < port->n_ifaces; ) {
4202 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
4203 iface_destroy(port->ifaces[i]);
4209 shash_destroy(&new_ifaces);
4213 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4215 const char *detect_mode;
4216 struct shash new_ifaces;
4217 long long int next_rebalance, miimon_next_update, lacp_priority;
4218 unsigned long *trunks;
4224 /* Update settings. */
4225 port->updelay = cfg->bond_updelay;
4226 if (port->updelay < 0) {
4229 port->downdelay = cfg->bond_downdelay;
4230 if (port->downdelay < 0) {
4231 port->downdelay = 0;
4233 port->bond_rebalance_interval = atoi(
4234 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4235 if (port->bond_rebalance_interval < 1000) {
4236 port->bond_rebalance_interval = 1000;
4238 next_rebalance = time_msec() + port->bond_rebalance_interval;
4239 if (port->bond_next_rebalance > next_rebalance) {
4240 port->bond_next_rebalance = next_rebalance;
4243 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4246 if (!strcmp(detect_mode, "carrier")) {
4247 port->miimon = false;
4248 } else if (!strcmp(detect_mode, "miimon")) {
4249 port->miimon = true;
4251 port->miimon = false;
4252 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
4253 "carrier", port->name, detect_mode);
4256 port->bond_miimon_interval = atoi(
4257 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4258 if (port->bond_miimon_interval < 100) {
4259 port->bond_miimon_interval = 100;
4261 miimon_next_update = time_msec() + port->bond_miimon_interval;
4262 if (port->bond_miimon_next_update > miimon_next_update) {
4263 port->bond_miimon_next_update = miimon_next_update;
4266 if (!port->cfg->bond_mode ||
4267 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4268 port->bond_mode = BM_SLB;
4269 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4270 port->bond_mode = BM_AB;
4271 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4272 port->bond_mode = BM_TCP;
4274 port->bond_mode = BM_SLB;
4275 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4276 port->name, port->cfg->bond_mode,
4277 bond_mode_to_string(port->bond_mode));
4280 /* Add new interfaces and update 'cfg' member of existing ones. */
4281 shash_init(&new_ifaces);
4282 for (i = 0; i < cfg->n_interfaces; i++) {
4283 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4284 struct iface *iface;
4286 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4287 VLOG_WARN("port %s: %s specified twice as port interface",
4288 port->name, if_cfg->name);
4289 iface_set_ofport(if_cfg, -1);
4293 iface = iface_lookup(port->bridge, if_cfg->name);
4295 if (iface->port != port) {
4296 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4298 port->bridge->name, if_cfg->name, iface->port->name);
4301 iface->cfg = if_cfg;
4303 iface = iface_create(port, if_cfg);
4306 /* Determine interface type. The local port always has type
4307 * "internal". Other ports take their type from the database and
4308 * default to "system" if none is specified. */
4309 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4310 : if_cfg->type[0] ? if_cfg->type
4314 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4317 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4318 iface->lacp_priority = UINT16_MAX;
4320 iface->lacp_priority = lacp_priority;
4323 shash_destroy(&new_ifaces);
4326 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4328 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4329 /* Prefer bondable links if unspecified. */
4330 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4332 port->lacp_priority = lacp_priority;
4335 if (!port->cfg->lacp) {
4336 /* XXX when LACP implementation has been sufficiently tested, enable by
4337 * default and make active on bonded ports. */
4339 } else if (!strcmp(port->cfg->lacp, "off")) {
4341 } else if (!strcmp(port->cfg->lacp, "active")) {
4342 port->lacp = LACP_ACTIVE;
4343 } else if (!strcmp(port->cfg->lacp, "passive")) {
4344 port->lacp = LACP_PASSIVE;
4346 VLOG_WARN("port %s: unknown LACP mode %s",
4347 port->name, port->cfg->lacp);
4354 if (port->n_ifaces < 2) {
4356 if (vlan >= 0 && vlan <= 4095) {
4357 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4362 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4363 * they even work as-is. But they have not been tested. */
4364 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4368 if (port->vlan != vlan) {
4370 bridge_flush(port->bridge);
4373 /* Get trunked VLANs. */
4375 if (vlan < 0 && cfg->n_trunks) {
4378 trunks = bitmap_allocate(4096);
4380 for (i = 0; i < cfg->n_trunks; i++) {
4381 int trunk = cfg->trunks[i];
4383 bitmap_set1(trunks, trunk);
4389 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4390 port->name, cfg->n_trunks);
4392 if (n_errors == cfg->n_trunks) {
4393 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4395 bitmap_free(trunks);
4398 } else if (vlan >= 0 && cfg->n_trunks) {
4399 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4403 ? port->trunks != NULL
4404 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4405 bridge_flush(port->bridge);
4407 bitmap_free(port->trunks);
4408 port->trunks = trunks;
4412 port_destroy(struct port *port)
4415 struct bridge *br = port->bridge;
4419 proc_net_compat_update_vlan(port->name, NULL, 0);
4420 proc_net_compat_update_bond(port->name, NULL);
4422 for (i = 0; i < MAX_MIRRORS; i++) {
4423 struct mirror *m = br->mirrors[i];
4424 if (m && m->out_port == port) {
4429 while (port->n_ifaces > 0) {
4430 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4433 shash_find_and_delete_assert(&br->port_by_name, port->name);
4435 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4436 del->port_idx = port->port_idx;
4438 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4440 netdev_monitor_destroy(port->monitor);
4442 bitmap_free(port->trunks);
4449 static struct port *
4450 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4452 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4453 return iface ? iface->port : NULL;
4456 static struct port *
4457 port_lookup(const struct bridge *br, const char *name)
4459 return shash_find_data(&br->port_by_name, name);
4462 static struct iface *
4463 port_lookup_iface(const struct port *port, const char *name)
4465 struct iface *iface = iface_lookup(port->bridge, name);
4466 return iface && iface->port == port ? iface : NULL;
4470 port_update_lacp(struct port *port)
4475 if (!port->lacp || port->n_ifaces < 1) {
4476 for (i = 0; i < port->n_ifaces; i++) {
4477 iface_set_lacp_defaulted(port->ifaces[i]);
4483 for (i = 0; i < port->n_ifaces; i++) {
4484 struct iface *iface = port->ifaces[i];
4486 if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) {
4491 if (iface->dp_ifidx == port->lacp_key) {
4492 key_changed = false;
4497 port->lacp_key = port->ifaces[0]->dp_ifidx;
4500 for (i = 0; i < port->n_ifaces; i++) {
4501 struct iface *iface = port->ifaces[i];
4503 iface->lacp_actor.sys_priority = htons(port->lacp_priority);
4504 memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN);
4506 iface->lacp_actor.port_priority = htons(iface->lacp_priority);
4507 iface->lacp_actor.portid = htons(iface->dp_ifidx);
4508 iface->lacp_actor.key = htons(port->lacp_key);
4512 port->lacp_need_update = true;
4516 port_update_bonding(struct port *port)
4518 if (port->monitor) {
4519 netdev_monitor_destroy(port->monitor);
4520 port->monitor = NULL;
4522 if (port->n_ifaces < 2) {
4523 /* Not a bonded port. */
4524 if (port->bond_hash) {
4525 free(port->bond_hash);
4526 port->bond_hash = NULL;
4527 port->bond_compat_is_stale = true;
4530 port->bond_fake_iface = false;
4534 if (port->bond_mode != BM_AB && !port->bond_hash) {
4535 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4536 for (i = 0; i <= BOND_MASK; i++) {
4537 struct bond_entry *e = &port->bond_hash[i];
4541 port->no_ifaces_tag = tag_create_random();
4542 bond_choose_active_iface(port);
4543 port->bond_next_rebalance
4544 = time_msec() + port->bond_rebalance_interval;
4546 if (port->cfg->bond_fake_iface) {
4547 port->bond_next_fake_iface_update = time_msec();
4549 } else if (port->bond_mode == BM_AB) {
4550 free(port->bond_hash);
4551 port->bond_hash = NULL;
4553 port->bond_compat_is_stale = true;
4554 port->bond_fake_iface = port->cfg->bond_fake_iface;
4556 if (!port->miimon) {
4557 port->monitor = netdev_monitor_create();
4558 for (i = 0; i < port->n_ifaces; i++) {
4559 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4566 port_update_bond_compat(struct port *port)
4568 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
4569 struct compat_bond bond;
4572 if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) {
4573 proc_net_compat_update_bond(port->name, NULL);
4578 bond.updelay = port->updelay;
4579 bond.downdelay = port->downdelay;
4582 bond.hashes = compat_hashes;
4583 if (port->bond_hash) {
4584 const struct bond_entry *e;
4585 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4586 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
4587 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
4588 cbh->hash = e - port->bond_hash;
4589 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
4594 bond.n_slaves = port->n_ifaces;
4595 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
4596 for (i = 0; i < port->n_ifaces; i++) {
4597 struct iface *iface = port->ifaces[i];
4598 struct compat_bond_slave *slave = &bond.slaves[i];
4599 slave->name = iface->name;
4601 /* We need to make the same determination as the Linux bonding
4602 * code to determine whether a slave should be consider "up".
4603 * The Linux function bond_miimon_inspect() supports four
4604 * BOND_LINK_* states:
4606 * - BOND_LINK_UP: carrier detected, updelay has passed.
4607 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
4608 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
4609 * - BOND_LINK_BACK: carrier detected, updelay in progress.
4611 * The function bond_info_show_slave() only considers BOND_LINK_UP
4612 * to be "up" and anything else to be "down".
4614 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
4618 netdev_get_etheraddr(iface->netdev, slave->mac);
4621 if (port->bond_fake_iface) {
4622 struct netdev *bond_netdev;
4624 if (!netdev_open_default(port->name, &bond_netdev)) {
4626 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
4628 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
4630 netdev_close(bond_netdev);
4634 proc_net_compat_update_bond(port->name, &bond);
4639 port_update_vlan_compat(struct port *port)
4641 struct bridge *br = port->bridge;
4642 char *vlandev_name = NULL;
4644 if (port->vlan > 0) {
4645 /* Figure out the name that the VLAN device should actually have, if it
4646 * existed. This takes some work because the VLAN device would not
4647 * have port->name in its name; rather, it would have the trunk port's
4648 * name, and 'port' would be attached to a bridge that also had the
4649 * VLAN device one of its ports. So we need to find a trunk port that
4650 * includes port->vlan.
4652 * There might be more than one candidate. This doesn't happen on
4653 * XenServer, so if it happens we just pick the first choice in
4654 * alphabetical order instead of creating multiple VLAN devices. */
4656 for (i = 0; i < br->n_ports; i++) {
4657 struct port *p = br->ports[i];
4658 if (port_trunks_vlan(p, port->vlan)
4660 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
4662 uint8_t ea[ETH_ADDR_LEN];
4663 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
4664 if (!eth_addr_is_multicast(ea) &&
4665 !eth_addr_is_reserved(ea) &&
4666 !eth_addr_is_zero(ea)) {
4667 vlandev_name = p->name;
4672 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
4675 /* Interface functions. */
4678 iface_set_lacp_defaulted(struct iface *iface)
4680 memset(&iface->lacp_partner, 0, sizeof iface->lacp_partner);
4682 iface->lacp_status = LACP_STATUS_DEFAULTED;
4684 iface->port->lacp_need_update = true;
4688 iface_set_lacp_expired(struct iface *iface)
4690 iface->lacp_status = LACP_STATUS_EXPIRED;
4691 iface->lacp_partner.state |= LACP_STATE_TIME;
4692 iface->lacp_partner.state &= ~LACP_STATE_SYNC;
4694 iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX;
4699 iface_get_lacp_state(const struct iface *iface)
4703 if (iface->port->lacp & LACP_ACTIVE) {
4704 state |= LACP_STATE_ACT;
4707 if (iface->lacp_status == LACP_STATUS_DEFAULTED) {
4708 state |= LACP_STATE_DEF;
4709 } else if (iface->lacp_attached) {
4710 state |= LACP_STATE_SYNC;
4713 if (iface->lacp_status == LACP_STATUS_EXPIRED) {
4714 state |= LACP_STATE_EXP;
4717 if (iface->port->n_ifaces > 1) {
4718 state |= LACP_STATE_AGG;
4721 if (iface->enabled) {
4722 state |= LACP_STATE_COL | LACP_STATE_DIST;
4728 /* Given 'iface', populates 'priority' with data representing its LACP link
4729 * priority. If two priority objects populated by this function are compared
4730 * using memcmp, the higher priority link will be less than the lower priority
4733 iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority)
4735 uint16_t partner_priority, actor_priority;
4737 /* Choose the lacp_info of the higher priority system by comparing their
4738 * system priorities and mac addresses. */
4739 actor_priority = ntohs(iface->lacp_actor.sys_priority);
4740 partner_priority = ntohs(iface->lacp_partner.sys_priority);
4741 if (actor_priority < partner_priority) {
4742 *priority = iface->lacp_actor;
4743 } else if (partner_priority < actor_priority) {
4744 *priority = iface->lacp_partner;
4745 } else if (eth_addr_compare_3way(iface->lacp_actor.sysid,
4746 iface->lacp_partner.sysid) < 0) {
4747 *priority = iface->lacp_actor;
4749 *priority = iface->lacp_partner;
4752 /* Key and state are not used in priority comparisons. */
4754 priority->state = 0;
4758 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4761 union ofp_action action;
4763 memset(&action, 0, sizeof action);
4764 action.output.type = htons(OFPAT_OUTPUT);
4765 action.output.len = htons(sizeof action);
4766 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4768 flow_extract(packet, 0, ODPP_NONE, &flow);
4770 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4772 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4773 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4777 static struct iface *
4778 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4780 struct bridge *br = port->bridge;
4781 struct iface *iface;
4782 char *name = if_cfg->name;
4784 iface = xzalloc(sizeof *iface);
4786 iface->port_ifidx = port->n_ifaces;
4787 iface->name = xstrdup(name);
4788 iface->dp_ifidx = -1;
4789 iface->tag = tag_create_random();
4790 iface->delay_expires = LLONG_MAX;
4791 iface->netdev = NULL;
4792 iface->cfg = if_cfg;
4793 iface_set_lacp_defaulted(iface);
4795 if (port->lacp & LACP_ACTIVE) {
4796 iface_set_lacp_expired(iface);
4799 shash_add_assert(&br->iface_by_name, iface->name, iface);
4801 if (port->n_ifaces >= port->allocated_ifaces) {
4802 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4803 sizeof *port->ifaces);
4805 port->ifaces[port->n_ifaces++] = iface;
4806 if (port->n_ifaces > 1) {
4807 br->has_bonded_ports = true;
4810 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4818 iface_destroy(struct iface *iface)
4821 struct port *port = iface->port;
4822 struct bridge *br = port->bridge;
4823 bool del_active = port->active_iface == iface->port_ifidx;
4826 if (port->monitor) {
4827 netdev_monitor_remove(port->monitor, iface->netdev);
4830 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4832 if (iface->dp_ifidx >= 0) {
4833 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4836 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4837 del->port_ifidx = iface->port_ifidx;
4839 netdev_close(iface->netdev);
4842 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4843 bond_choose_active_iface(port);
4844 bond_send_learning_packets(port);
4847 cfm_destroy(iface->cfm);
4852 bridge_flush(port->bridge);
4856 static struct iface *
4857 iface_lookup(const struct bridge *br, const char *name)
4859 return shash_find_data(&br->iface_by_name, name);
4862 static struct iface *
4863 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4865 struct iface *iface;
4867 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4868 hash_int(dp_ifidx, 0), &br->ifaces) {
4869 if (iface->dp_ifidx == dp_ifidx) {
4876 /* Set Ethernet address of 'iface', if one is specified in the configuration
4879 iface_set_mac(struct iface *iface)
4881 uint8_t ea[ETH_ADDR_LEN];
4883 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4884 if (eth_addr_is_multicast(ea)) {
4885 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4887 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4888 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4889 iface->name, iface->name);
4891 int error = netdev_set_etheraddr(iface->netdev, ea);
4893 VLOG_ERR("interface %s: setting MAC failed (%s)",
4894 iface->name, strerror(error));
4900 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4902 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4905 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4909 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4911 * The value strings in '*shash' are taken directly from values[], not copied,
4912 * so the caller should not modify or free them. */
4914 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4915 struct shash *shash)
4920 for (i = 0; i < n; i++) {
4921 shash_add(shash, keys[i], values[i]);
4925 /* Creates 'keys' and 'values' arrays from 'shash'.
4927 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4928 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4929 * are populated with with strings taken directly from 'shash' and thus have
4930 * the same ownership of the key-value pairs in shash.
4933 shash_to_ovs_idl_map(struct shash *shash,
4934 char ***keys, char ***values, size_t *n)
4938 struct shash_node *sn;
4940 count = shash_count(shash);
4942 k = xmalloc(count * sizeof *k);
4943 v = xmalloc(count * sizeof *v);
4946 SHASH_FOR_EACH(sn, shash) {
4957 struct iface_delete_queues_cbdata {
4958 struct netdev *netdev;
4959 const struct ovsdb_datum *queues;
4963 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4965 union ovsdb_atom atom;
4967 atom.integer = target;
4968 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4972 iface_delete_queues(unsigned int queue_id,
4973 const struct shash *details OVS_UNUSED, void *cbdata_)
4975 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4977 if (!queue_ids_include(cbdata->queues, queue_id)) {
4978 netdev_delete_queue(cbdata->netdev, queue_id);
4983 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4985 if (!qos || qos->type[0] == '\0') {
4986 netdev_set_qos(iface->netdev, NULL, NULL);
4988 struct iface_delete_queues_cbdata cbdata;
4989 struct shash details;
4992 /* Configure top-level Qos for 'iface'. */
4993 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4994 qos->n_other_config, &details);
4995 netdev_set_qos(iface->netdev, qos->type, &details);
4996 shash_destroy(&details);
4998 /* Deconfigure queues that were deleted. */
4999 cbdata.netdev = iface->netdev;
5000 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
5002 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
5004 /* Configure queues for 'iface'. */
5005 for (i = 0; i < qos->n_queues; i++) {
5006 const struct ovsrec_queue *queue = qos->value_queues[i];
5007 unsigned int queue_id = qos->key_queues[i];
5009 shash_from_ovs_idl_map(queue->key_other_config,
5010 queue->value_other_config,
5011 queue->n_other_config, &details);
5012 netdev_set_queue(iface->netdev, queue_id, &details);
5013 shash_destroy(&details);
5019 iface_update_cfm(struct iface *iface)
5023 uint16_t *remote_mps;
5024 struct ovsrec_monitor *mon;
5025 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
5027 mon = iface->cfg->monitor;
5030 cfm_destroy(iface->cfm);
5035 if (netdev_get_etheraddr(iface->netdev, ea)) {
5036 VLOG_WARN("interface %s: Failed to get ethernet address. "
5037 "Skipping Monitor.", iface->name);
5041 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
5042 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
5047 iface->cfm = cfm_create();
5051 cfm->mpid = mon->mpid;
5052 cfm->interval = mon->interval ? *mon->interval : 1000;
5054 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
5055 memcpy(cfm->maid, maid, sizeof cfm->maid);
5057 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
5058 for(i = 0; i < mon->n_remote_mps; i++) {
5059 remote_mps[i] = mon->remote_mps[i]->mpid;
5061 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
5064 if (!cfm_configure(iface->cfm)) {
5065 cfm_destroy(iface->cfm);
5070 /* Port mirroring. */
5072 static struct mirror *
5073 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
5077 for (i = 0; i < MAX_MIRRORS; i++) {
5078 struct mirror *m = br->mirrors[i];
5079 if (m && uuid_equals(uuid, &m->uuid)) {
5087 mirror_reconfigure(struct bridge *br)
5089 unsigned long *rspan_vlans;
5092 /* Get rid of deleted mirrors. */
5093 for (i = 0; i < MAX_MIRRORS; i++) {
5094 struct mirror *m = br->mirrors[i];
5096 const struct ovsdb_datum *mc;
5097 union ovsdb_atom atom;
5099 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
5100 atom.uuid = br->mirrors[i]->uuid;
5101 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
5107 /* Add new mirrors and reconfigure existing ones. */
5108 for (i = 0; i < br->cfg->n_mirrors; i++) {
5109 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
5110 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
5112 mirror_reconfigure_one(m, cfg);
5114 mirror_create(br, cfg);
5118 /* Update port reserved status. */
5119 for (i = 0; i < br->n_ports; i++) {
5120 br->ports[i]->is_mirror_output_port = false;
5122 for (i = 0; i < MAX_MIRRORS; i++) {
5123 struct mirror *m = br->mirrors[i];
5124 if (m && m->out_port) {
5125 m->out_port->is_mirror_output_port = true;
5129 /* Update flooded vlans (for RSPAN). */
5131 if (br->cfg->n_flood_vlans) {
5132 rspan_vlans = bitmap_allocate(4096);
5134 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
5135 int64_t vlan = br->cfg->flood_vlans[i];
5136 if (vlan >= 0 && vlan < 4096) {
5137 bitmap_set1(rspan_vlans, vlan);
5138 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
5141 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
5146 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
5152 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
5157 for (i = 0; ; i++) {
5158 if (i >= MAX_MIRRORS) {
5159 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
5160 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
5163 if (!br->mirrors[i]) {
5168 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
5171 br->mirrors[i] = m = xzalloc(sizeof *m);
5174 m->name = xstrdup(cfg->name);
5175 shash_init(&m->src_ports);
5176 shash_init(&m->dst_ports);
5182 mirror_reconfigure_one(m, cfg);
5186 mirror_destroy(struct mirror *m)
5189 struct bridge *br = m->bridge;
5192 for (i = 0; i < br->n_ports; i++) {
5193 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5194 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5197 shash_destroy(&m->src_ports);
5198 shash_destroy(&m->dst_ports);
5201 m->bridge->mirrors[m->idx] = NULL;
5210 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
5211 struct shash *names)
5215 for (i = 0; i < n_ports; i++) {
5216 const char *name = ports[i]->name;
5217 if (port_lookup(m->bridge, name)) {
5218 shash_add_once(names, name, NULL);
5220 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
5221 "port %s", m->bridge->name, m->name, name);
5227 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
5233 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
5235 for (i = 0; i < cfg->n_select_vlan; i++) {
5236 int64_t vlan = cfg->select_vlan[i];
5237 if (vlan < 0 || vlan > 4095) {
5238 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
5239 m->bridge->name, m->name, vlan);
5241 (*vlans)[n_vlans++] = vlan;
5248 vlan_is_mirrored(const struct mirror *m, int vlan)
5252 for (i = 0; i < m->n_vlans; i++) {
5253 if (m->vlans[i] == vlan) {
5261 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
5265 for (i = 0; i < m->n_vlans; i++) {
5266 if (port_trunks_vlan(p, m->vlans[i])) {
5274 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
5276 struct shash src_ports, dst_ports;
5277 mirror_mask_t mirror_bit;
5278 struct port *out_port;
5285 if (strcmp(cfg->name, m->name)) {
5287 m->name = xstrdup(cfg->name);
5290 /* Get output port. */
5291 if (cfg->output_port) {
5292 out_port = port_lookup(m->bridge, cfg->output_port->name);
5294 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
5295 m->bridge->name, m->name);
5301 if (cfg->output_vlan) {
5302 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
5303 "output vlan; ignoring output vlan",
5304 m->bridge->name, m->name);
5306 } else if (cfg->output_vlan) {
5308 out_vlan = *cfg->output_vlan;
5310 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
5311 m->bridge->name, m->name);
5316 shash_init(&src_ports);
5317 shash_init(&dst_ports);
5318 if (cfg->select_all) {
5319 for (i = 0; i < m->bridge->n_ports; i++) {
5320 const char *name = m->bridge->ports[i]->name;
5321 shash_add_once(&src_ports, name, NULL);
5322 shash_add_once(&dst_ports, name, NULL);
5327 /* Get ports, and drop duplicates and ports that don't exist. */
5328 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
5330 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
5333 /* Get all the vlans, and drop duplicate and invalid vlans. */
5334 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
5337 /* Update mirror data. */
5338 if (!shash_equal_keys(&m->src_ports, &src_ports)
5339 || !shash_equal_keys(&m->dst_ports, &dst_ports)
5340 || m->n_vlans != n_vlans
5341 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
5342 || m->out_port != out_port
5343 || m->out_vlan != out_vlan) {
5344 bridge_flush(m->bridge);
5346 shash_swap(&m->src_ports, &src_ports);
5347 shash_swap(&m->dst_ports, &dst_ports);
5350 m->n_vlans = n_vlans;
5351 m->out_port = out_port;
5352 m->out_vlan = out_vlan;
5355 mirror_bit = MIRROR_MASK_C(1) << m->idx;
5356 for (i = 0; i < m->bridge->n_ports; i++) {
5357 struct port *port = m->bridge->ports[i];
5359 if (shash_find(&m->src_ports, port->name)
5362 ? port_trunks_any_mirrored_vlan(m, port)
5363 : vlan_is_mirrored(m, port->vlan)))) {
5364 port->src_mirrors |= mirror_bit;
5366 port->src_mirrors &= ~mirror_bit;
5369 if (shash_find(&m->dst_ports, port->name)) {
5370 port->dst_mirrors |= mirror_bit;
5372 port->dst_mirrors &= ~mirror_bit;
5377 shash_destroy(&src_ports);
5378 shash_destroy(&dst_ports);