1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
41 #include "dynamic-string.h"
47 #include "mac-learning.h"
51 #include "ofp-print.h"
53 #include "ofproto/netflow.h"
54 #include "ofproto/ofproto.h"
55 #include "ovsdb-data.h"
57 #include "poll-loop.h"
58 #include "proc-net-compat.h"
62 #include "socket-util.h"
63 #include "stream-ssl.h"
65 #include "system-stats.h"
70 #include "vswitchd/vswitch-idl.h"
71 #include "xenserver.h"
73 #include "sflow_api.h"
75 VLOG_DEFINE_THIS_MODULE(bridge);
77 COVERAGE_DEFINE(bridge_flush);
78 COVERAGE_DEFINE(bridge_process_flow);
79 COVERAGE_DEFINE(bridge_reconfigure);
80 COVERAGE_DEFINE(bridge_lacp_update);
88 struct dst builtin[32];
93 static void dst_set_init(struct dst_set *);
94 static void dst_set_add(struct dst_set *, const struct dst *);
95 static void dst_set_free(struct dst_set *);
98 LACP_CURRENT = 0x01, /* Current State. */
99 LACP_EXPIRED = 0x02, /* Expired State. */
100 LACP_DEFAULTED = 0x04, /* Partner is defaulted. */
101 LACP_ATTACHED = 0x08, /* Attached. Interface may be choosen for flows. */
105 /* These members are always valid. */
106 struct port *port; /* Containing port. */
107 size_t port_ifidx; /* Index within containing port. */
108 char *name; /* Host network device name. */
109 tag_type tag; /* Tag associated with this interface. */
110 long long delay_expires; /* Time after which 'enabled' may change. */
112 /* These members are valid only after bridge_reconfigure() causes them to
114 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
115 int dp_ifidx; /* Index within kernel datapath. */
116 struct netdev *netdev; /* Network device. */
117 bool enabled; /* May be chosen for flows? */
118 bool up; /* Is the interface up? */
119 const char *type; /* Usually same as cfg->type. */
120 struct cfm *cfm; /* Connectivity Fault Management */
121 const struct ovsrec_interface *cfg;
123 /* LACP information. */
124 enum lacp_status lacp_status; /* LACP status. */
125 uint16_t lacp_priority; /* LACP port priority. */
126 struct lacp_info lacp_actor; /* LACP actor information. */
127 struct lacp_info lacp_partner; /* LACP partner information. */
128 long long int lacp_tx; /* Next LACP message transmission time. */
129 long long int lacp_rx; /* Next LACP message receive time. */
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);
938 /* ovs-vswitchd has completed initialization, so allow the process that
939 * forked us to exit successfully. */
940 daemonize_complete();
944 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
945 const struct ovsdb_idl_column *column,
948 const struct ovsdb_datum *datum;
949 union ovsdb_atom atom;
952 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
953 atom.string = (char *) key;
954 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
955 return idx == UINT_MAX ? NULL : datum->values[idx].string;
959 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
961 return get_ovsrec_key_value(&br_cfg->header_,
962 &ovsrec_bridge_col_other_config, key);
966 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
967 struct iface **hw_addr_iface)
973 *hw_addr_iface = NULL;
975 /* Did the user request a particular MAC? */
976 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
977 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
978 if (eth_addr_is_multicast(ea)) {
979 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
980 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
981 } else if (eth_addr_is_zero(ea)) {
982 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
988 /* Otherwise choose the minimum non-local MAC address among all of the
990 memset(ea, 0xff, sizeof ea);
991 for (i = 0; i < br->n_ports; i++) {
992 struct port *port = br->ports[i];
993 uint8_t iface_ea[ETH_ADDR_LEN];
996 /* Mirror output ports don't participate. */
997 if (port->is_mirror_output_port) {
1001 /* Choose the MAC address to represent the port. */
1002 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1003 /* Find the interface with this Ethernet address (if any) so that
1004 * we can provide the correct devname to the caller. */
1006 for (j = 0; j < port->n_ifaces; j++) {
1007 struct iface *candidate = port->ifaces[j];
1008 uint8_t candidate_ea[ETH_ADDR_LEN];
1009 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1010 && eth_addr_equals(iface_ea, candidate_ea)) {
1015 /* Choose the interface whose MAC address will represent the port.
1016 * The Linux kernel bonding code always chooses the MAC address of
1017 * the first slave added to a bond, and the Fedora networking
1018 * scripts always add slaves to a bond in alphabetical order, so
1019 * for compatibility we choose the interface with the name that is
1020 * first in alphabetical order. */
1021 iface = port->ifaces[0];
1022 for (j = 1; j < port->n_ifaces; j++) {
1023 struct iface *candidate = port->ifaces[j];
1024 if (strcmp(candidate->name, iface->name) < 0) {
1029 /* The local port doesn't count (since we're trying to choose its
1030 * MAC address anyway). */
1031 if (iface->dp_ifidx == ODPP_LOCAL) {
1036 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1038 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1039 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1040 iface->name, strerror(error));
1045 /* Compare against our current choice. */
1046 if (!eth_addr_is_multicast(iface_ea) &&
1047 !eth_addr_is_local(iface_ea) &&
1048 !eth_addr_is_reserved(iface_ea) &&
1049 !eth_addr_is_zero(iface_ea) &&
1050 eth_addr_compare_3way(iface_ea, ea) < 0)
1052 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1053 *hw_addr_iface = iface;
1056 if (eth_addr_is_multicast(ea)) {
1057 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1058 *hw_addr_iface = NULL;
1059 VLOG_WARN("bridge %s: using default bridge Ethernet "
1060 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1062 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1063 br->name, ETH_ADDR_ARGS(ea));
1067 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1068 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1069 * an interface on 'br', then that interface must be passed in as
1070 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1071 * 'hw_addr_iface' must be passed in as a null pointer. */
1073 bridge_pick_datapath_id(struct bridge *br,
1074 const uint8_t bridge_ea[ETH_ADDR_LEN],
1075 struct iface *hw_addr_iface)
1078 * The procedure for choosing a bridge MAC address will, in the most
1079 * ordinary case, also choose a unique MAC that we can use as a datapath
1080 * ID. In some special cases, though, multiple bridges will end up with
1081 * the same MAC address. This is OK for the bridges, but it will confuse
1082 * the OpenFlow controller, because each datapath needs a unique datapath
1085 * Datapath IDs must be unique. It is also very desirable that they be
1086 * stable from one run to the next, so that policy set on a datapath
1089 const char *datapath_id;
1092 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1093 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1097 if (hw_addr_iface) {
1099 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1101 * A bridge whose MAC address is taken from a VLAN network device
1102 * (that is, a network device created with vconfig(8) or similar
1103 * tool) will have the same MAC address as a bridge on the VLAN
1104 * device's physical network device.
1106 * Handle this case by hashing the physical network device MAC
1107 * along with the VLAN identifier.
1109 uint8_t buf[ETH_ADDR_LEN + 2];
1110 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1111 buf[ETH_ADDR_LEN] = vlan >> 8;
1112 buf[ETH_ADDR_LEN + 1] = vlan;
1113 return dpid_from_hash(buf, sizeof buf);
1116 * Assume that this bridge's MAC address is unique, since it
1117 * doesn't fit any of the cases we handle specially.
1122 * A purely internal bridge, that is, one that has no non-virtual
1123 * network devices on it at all, is more difficult because it has no
1124 * natural unique identifier at all.
1126 * When the host is a XenServer, we handle this case by hashing the
1127 * host's UUID with the name of the bridge. Names of bridges are
1128 * persistent across XenServer reboots, although they can be reused if
1129 * an internal network is destroyed and then a new one is later
1130 * created, so this is fairly effective.
1132 * When the host is not a XenServer, we punt by using a random MAC
1133 * address on each run.
1135 const char *host_uuid = xenserver_get_host_uuid();
1137 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1138 dpid = dpid_from_hash(combined, strlen(combined));
1144 return eth_addr_to_uint64(bridge_ea);
1148 dpid_from_hash(const void *data, size_t n)
1150 uint8_t hash[SHA1_DIGEST_SIZE];
1152 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1153 sha1_bytes(data, n, hash);
1154 eth_addr_mark_random(hash);
1155 return eth_addr_to_uint64(hash);
1159 iface_refresh_status(struct iface *iface)
1163 enum netdev_flags flags;
1172 if (!netdev_get_status(iface->netdev, &sh)) {
1174 char **keys, **values;
1176 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1177 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1182 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1185 shash_destroy_free_data(&sh);
1187 error = netdev_get_flags(iface->netdev, &flags);
1189 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1192 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1195 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1197 ovsrec_interface_set_duplex(iface->cfg,
1198 netdev_features_is_full_duplex(current)
1200 /* warning: uint64_t -> int64_t conversion */
1201 bps = netdev_features_to_bps(current);
1202 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1205 ovsrec_interface_set_duplex(iface->cfg, NULL);
1206 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1210 ovsrec_interface_set_link_state(iface->cfg,
1211 netdev_get_carrier(iface->netdev)
1214 error = netdev_get_mtu(iface->netdev, &mtu);
1215 if (!error && mtu != INT_MAX) {
1217 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1220 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1225 iface_refresh_cfm_stats(struct iface *iface)
1229 const struct ovsrec_monitor *mon;
1231 mon = iface->cfg->monitor;
1238 for (i = 0; i < mon->n_remote_mps; i++) {
1239 const struct ovsrec_maintenance_point *mp;
1240 const struct remote_mp *rmp;
1242 mp = mon->remote_mps[i];
1243 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1245 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1248 if (hmap_is_empty(&cfm->x_remote_mps)) {
1249 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1252 struct remote_mp *rmp;
1253 int64_t *x_remote_mps;
1255 length = hmap_count(&cfm->x_remote_mps);
1256 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1259 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1260 x_remote_mps[i++] = rmp->mpid;
1263 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1267 if (hmap_is_empty(&cfm->x_remote_maids)) {
1268 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1271 char **x_remote_maids;
1272 struct remote_maid *rmaid;
1274 length = hmap_count(&cfm->x_remote_maids);
1275 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1278 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1281 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1283 for (j = 0; j < CCM_MAID_LEN; j++) {
1284 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1289 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1291 for (i = 0; i < length; i++) {
1292 free(x_remote_maids[i]);
1294 free(x_remote_maids);
1297 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1301 iface_refresh_stats(struct iface *iface)
1307 static const struct iface_stat iface_stats[] = {
1308 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1309 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1310 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1311 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1312 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1313 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1314 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1315 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1316 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1317 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1318 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1319 { "collisions", offsetof(struct netdev_stats, collisions) },
1321 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1322 const struct iface_stat *s;
1324 char *keys[N_STATS];
1325 int64_t values[N_STATS];
1328 struct netdev_stats stats;
1330 /* Intentionally ignore return value, since errors will set 'stats' to
1331 * all-1s, and we will deal with that correctly below. */
1332 netdev_get_stats(iface->netdev, &stats);
1335 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1336 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1337 if (value != UINT64_MAX) {
1344 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1348 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1350 struct ovsdb_datum datum;
1354 get_system_stats(&stats);
1356 ovsdb_datum_from_shash(&datum, &stats);
1357 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1361 static inline const char *
1362 nx_role_to_str(enum nx_role role)
1367 case NX_ROLE_MASTER:
1372 return "*** INVALID ROLE ***";
1377 bridge_refresh_controller_status(const struct bridge *br)
1380 const struct ovsrec_controller *cfg;
1382 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1384 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1385 struct ofproto_controller_info *cinfo =
1386 shash_find_data(&info, cfg->target);
1389 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1390 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1391 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1392 (char **) cinfo->pairs.values,
1395 ovsrec_controller_set_is_connected(cfg, false);
1396 ovsrec_controller_set_role(cfg, NULL);
1397 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1401 ofproto_free_ofproto_controller_info(&info);
1407 const struct ovsrec_open_vswitch *cfg;
1409 bool datapath_destroyed;
1410 bool database_changed;
1413 /* Let each bridge do the work that it needs to do. */
1414 datapath_destroyed = false;
1415 LIST_FOR_EACH (br, node, &all_bridges) {
1416 int error = bridge_run_one(br);
1418 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1419 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1420 "forcing reconfiguration", br->name);
1421 datapath_destroyed = true;
1425 /* (Re)configure if necessary. */
1426 database_changed = ovsdb_idl_run(idl);
1427 cfg = ovsrec_open_vswitch_first(idl);
1429 /* Re-configure SSL. We do this on every trip through the main loop,
1430 * instead of just when the database changes, because the contents of the
1431 * key and certificate files can change without the database changing.
1433 * We do this before bridge_reconfigure() because that function might
1434 * initiate SSL connections and thus requires SSL to be configured. */
1435 if (cfg && cfg->ssl) {
1436 const struct ovsrec_ssl *ssl = cfg->ssl;
1438 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1439 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1442 if (database_changed || datapath_destroyed) {
1444 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1446 bridge_configure_once(cfg);
1447 bridge_reconfigure(cfg);
1449 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1450 ovsdb_idl_txn_commit(txn);
1451 ovsdb_idl_txn_destroy(txn); /* XXX */
1453 /* We still need to reconfigure to avoid dangling pointers to
1454 * now-destroyed ovsrec structures inside bridge data. */
1455 static const struct ovsrec_open_vswitch null_cfg;
1457 bridge_reconfigure(&null_cfg);
1461 /* Refresh system and interface stats if necessary. */
1462 if (time_msec() >= stats_timer) {
1464 struct ovsdb_idl_txn *txn;
1466 txn = ovsdb_idl_txn_create(idl);
1467 LIST_FOR_EACH (br, node, &all_bridges) {
1470 for (i = 0; i < br->n_ports; i++) {
1471 struct port *port = br->ports[i];
1474 for (j = 0; j < port->n_ifaces; j++) {
1475 struct iface *iface = port->ifaces[j];
1476 iface_refresh_stats(iface);
1477 iface_refresh_cfm_stats(iface);
1478 iface_refresh_status(iface);
1481 bridge_refresh_controller_status(br);
1483 refresh_system_stats(cfg);
1484 ovsdb_idl_txn_commit(txn);
1485 ovsdb_idl_txn_destroy(txn); /* XXX */
1488 stats_timer = time_msec() + STATS_INTERVAL;
1496 struct iface *iface;
1498 LIST_FOR_EACH (br, node, &all_bridges) {
1499 ofproto_wait(br->ofproto);
1500 if (ofproto_has_primary_controller(br->ofproto)) {
1504 mac_learning_wait(br->ml);
1508 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1510 cfm_wait(iface->cfm);
1514 ovsdb_idl_wait(idl);
1515 poll_timer_wait_until(stats_timer);
1518 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1519 * configuration changes. */
1521 bridge_flush(struct bridge *br)
1523 COVERAGE_INC(bridge_flush);
1525 mac_learning_flush(br->ml);
1528 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1529 * such interface. */
1530 static struct iface *
1531 bridge_get_local_iface(struct bridge *br)
1535 for (i = 0; i < br->n_ports; i++) {
1536 struct port *port = br->ports[i];
1537 for (j = 0; j < port->n_ifaces; j++) {
1538 struct iface *iface = port->ifaces[j];
1539 if (iface->dp_ifidx == ODPP_LOCAL) {
1548 /* Bridge unixctl user interface functions. */
1550 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1551 const char *args, void *aux OVS_UNUSED)
1553 struct ds ds = DS_EMPTY_INITIALIZER;
1554 const struct bridge *br;
1555 const struct mac_entry *e;
1557 br = bridge_lookup(args);
1559 unixctl_command_reply(conn, 501, "no such bridge");
1563 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1564 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1565 if (e->port < 0 || e->port >= br->n_ports) {
1568 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1569 br->ports[e->port]->ifaces[0]->dp_ifidx,
1570 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1572 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1576 /* Bridge reconfiguration functions. */
1577 static struct bridge *
1578 bridge_create(const struct ovsrec_bridge *br_cfg)
1583 assert(!bridge_lookup(br_cfg->name));
1584 br = xzalloc(sizeof *br);
1586 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1592 dpif_flow_flush(br->dpif);
1594 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1597 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1599 dpif_delete(br->dpif);
1600 dpif_close(br->dpif);
1605 br->name = xstrdup(br_cfg->name);
1607 br->ml = mac_learning_create();
1608 eth_addr_nicira_random(br->default_ea);
1610 hmap_init(&br->ifaces);
1612 shash_init(&br->port_by_name);
1613 shash_init(&br->iface_by_name);
1617 list_push_back(&all_bridges, &br->node);
1619 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1625 bridge_destroy(struct bridge *br)
1630 while (br->n_ports > 0) {
1631 port_destroy(br->ports[br->n_ports - 1]);
1633 list_remove(&br->node);
1634 error = dpif_delete(br->dpif);
1635 if (error && error != ENOENT) {
1636 VLOG_ERR("failed to delete %s: %s",
1637 dpif_name(br->dpif), strerror(error));
1639 dpif_close(br->dpif);
1640 ofproto_destroy(br->ofproto);
1641 mac_learning_destroy(br->ml);
1642 hmap_destroy(&br->ifaces);
1643 shash_destroy(&br->port_by_name);
1644 shash_destroy(&br->iface_by_name);
1651 static struct bridge *
1652 bridge_lookup(const char *name)
1656 LIST_FOR_EACH (br, node, &all_bridges) {
1657 if (!strcmp(br->name, name)) {
1664 /* Handle requests for a listing of all flows known by the OpenFlow
1665 * stack, including those normally hidden. */
1667 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1668 const char *args, void *aux OVS_UNUSED)
1673 br = bridge_lookup(args);
1675 unixctl_command_reply(conn, 501, "Unknown bridge");
1680 ofproto_get_all_flows(br->ofproto, &results);
1682 unixctl_command_reply(conn, 200, ds_cstr(&results));
1683 ds_destroy(&results);
1686 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1687 * connections and reconnect. If BRIDGE is not specified, then all bridges
1688 * drop their controller connections and reconnect. */
1690 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1691 const char *args, void *aux OVS_UNUSED)
1694 if (args[0] != '\0') {
1695 br = bridge_lookup(args);
1697 unixctl_command_reply(conn, 501, "Unknown bridge");
1700 ofproto_reconnect_controllers(br->ofproto);
1702 LIST_FOR_EACH (br, node, &all_bridges) {
1703 ofproto_reconnect_controllers(br->ofproto);
1706 unixctl_command_reply(conn, 200, NULL);
1710 bridge_run_one(struct bridge *br)
1713 struct iface *iface;
1715 error = ofproto_run1(br->ofproto);
1720 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1724 error = ofproto_run2(br->ofproto, br->flush);
1727 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1728 struct ofpbuf *packet;
1734 packet = cfm_run(iface->cfm);
1736 iface_send_packet(iface, packet);
1737 ofpbuf_uninit(packet);
1746 bridge_get_controllers(const struct bridge *br,
1747 struct ovsrec_controller ***controllersp)
1749 struct ovsrec_controller **controllers;
1750 size_t n_controllers;
1752 controllers = br->cfg->controller;
1753 n_controllers = br->cfg->n_controller;
1755 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1761 *controllersp = controllers;
1763 return n_controllers;
1767 bridge_reconfigure_one(struct bridge *br)
1769 struct shash old_ports, new_ports;
1770 struct svec snoops, old_snoops;
1771 struct shash_node *node;
1772 enum ofproto_fail_mode fail_mode;
1775 /* Collect old ports. */
1776 shash_init(&old_ports);
1777 for (i = 0; i < br->n_ports; i++) {
1778 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1781 /* Collect new ports. */
1782 shash_init(&new_ports);
1783 for (i = 0; i < br->cfg->n_ports; i++) {
1784 const char *name = br->cfg->ports[i]->name;
1785 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1786 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1791 /* If we have a controller, then we need a local port. Complain if the
1792 * user didn't specify one.
1794 * XXX perhaps we should synthesize a port ourselves in this case. */
1795 if (bridge_get_controllers(br, NULL)) {
1796 char local_name[IF_NAMESIZE];
1799 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1800 local_name, sizeof local_name);
1801 if (!error && !shash_find(&new_ports, local_name)) {
1802 VLOG_WARN("bridge %s: controller specified but no local port "
1803 "(port named %s) defined",
1804 br->name, local_name);
1808 /* Get rid of deleted ports.
1809 * Get rid of deleted interfaces on ports that still exist. */
1810 SHASH_FOR_EACH (node, &old_ports) {
1811 struct port *port = node->data;
1812 const struct ovsrec_port *port_cfg;
1814 port_cfg = shash_find_data(&new_ports, node->name);
1818 port_del_ifaces(port, port_cfg);
1822 /* Create new ports.
1823 * Add new interfaces to existing ports.
1824 * Reconfigure existing ports. */
1825 SHASH_FOR_EACH (node, &new_ports) {
1826 struct port *port = shash_find_data(&old_ports, node->name);
1828 port = port_create(br, node->name);
1831 port_reconfigure(port, node->data);
1832 if (!port->n_ifaces) {
1833 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1834 br->name, port->name);
1838 shash_destroy(&old_ports);
1839 shash_destroy(&new_ports);
1841 /* Set the fail-mode */
1842 fail_mode = !br->cfg->fail_mode
1843 || !strcmp(br->cfg->fail_mode, "standalone")
1844 ? OFPROTO_FAIL_STANDALONE
1845 : OFPROTO_FAIL_SECURE;
1846 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1847 && !ofproto_has_primary_controller(br->ofproto)) {
1848 ofproto_flush_flows(br->ofproto);
1850 ofproto_set_fail_mode(br->ofproto, fail_mode);
1852 /* Delete all flows if we're switching from connected to standalone or vice
1853 * versa. (XXX Should we delete all flows if we are switching from one
1854 * controller to another?) */
1856 /* Configure OpenFlow controller connection snooping. */
1858 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1859 ovs_rundir(), br->name));
1860 svec_init(&old_snoops);
1861 ofproto_get_snoops(br->ofproto, &old_snoops);
1862 if (!svec_equal(&snoops, &old_snoops)) {
1863 ofproto_set_snoops(br->ofproto, &snoops);
1865 svec_destroy(&snoops);
1866 svec_destroy(&old_snoops);
1868 mirror_reconfigure(br);
1871 /* Initializes 'oc' appropriately as a management service controller for
1874 * The caller must free oc->target when it is no longer needed. */
1876 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1877 struct ofproto_controller *oc)
1879 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1880 oc->max_backoff = 0;
1881 oc->probe_interval = 60;
1882 oc->band = OFPROTO_OUT_OF_BAND;
1883 oc->accept_re = NULL;
1884 oc->update_resolv_conf = false;
1886 oc->burst_limit = 0;
1889 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1891 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1892 struct ofproto_controller *oc)
1894 oc->target = c->target;
1895 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1896 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1897 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1898 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1899 oc->accept_re = c->discover_accept_regex;
1900 oc->update_resolv_conf = c->discover_update_resolv_conf;
1901 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1902 oc->burst_limit = (c->controller_burst_limit
1903 ? *c->controller_burst_limit : 0);
1906 /* Configures the IP stack for 'br''s local interface properly according to the
1907 * configuration in 'c'. */
1909 bridge_configure_local_iface_netdev(struct bridge *br,
1910 struct ovsrec_controller *c)
1912 struct netdev *netdev;
1913 struct in_addr mask, gateway;
1915 struct iface *local_iface;
1918 /* Controller discovery does its own TCP/IP configuration later. */
1919 if (strcmp(c->target, "discover")) {
1923 /* If there's no local interface or no IP address, give up. */
1924 local_iface = bridge_get_local_iface(br);
1925 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1929 /* Bring up the local interface. */
1930 netdev = local_iface->netdev;
1931 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1933 /* Configure the IP address and netmask. */
1934 if (!c->local_netmask
1935 || !inet_aton(c->local_netmask, &mask)
1937 mask.s_addr = guess_netmask(ip.s_addr);
1939 if (!netdev_set_in4(netdev, ip, mask)) {
1940 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1941 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1944 /* Configure the default gateway. */
1945 if (c->local_gateway
1946 && inet_aton(c->local_gateway, &gateway)
1947 && gateway.s_addr) {
1948 if (!netdev_add_router(netdev, gateway)) {
1949 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1950 br->name, IP_ARGS(&gateway.s_addr));
1956 bridge_reconfigure_remotes(struct bridge *br,
1957 const struct sockaddr_in *managers,
1960 const char *disable_ib_str, *queue_id_str;
1961 bool disable_in_band = false;
1964 struct ovsrec_controller **controllers;
1965 size_t n_controllers;
1968 struct ofproto_controller *ocs;
1972 /* Check if we should disable in-band control on this bridge. */
1973 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1974 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1975 disable_in_band = true;
1978 /* Set OpenFlow queue ID for in-band control. */
1979 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1980 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1981 ofproto_set_in_band_queue(br->ofproto, queue_id);
1983 if (disable_in_band) {
1984 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1986 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1988 had_primary = ofproto_has_primary_controller(br->ofproto);
1990 n_controllers = bridge_get_controllers(br, &controllers);
1992 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1995 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1996 for (i = 0; i < n_controllers; i++) {
1997 struct ovsrec_controller *c = controllers[i];
1999 if (!strncmp(c->target, "punix:", 6)
2000 || !strncmp(c->target, "unix:", 5)) {
2001 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2003 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2004 * domain sockets and overwriting arbitrary local files. */
2005 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2006 "\"%s\" due to possibility for remote exploit",
2007 dpif_name(br->dpif), c->target);
2011 bridge_configure_local_iface_netdev(br, c);
2012 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2013 if (disable_in_band) {
2014 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2019 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2020 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2023 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2024 ofproto_flush_flows(br->ofproto);
2027 /* If there are no controllers and the bridge is in standalone
2028 * mode, set up a flow that matches every packet and directs
2029 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2030 * switch is in secure mode and we won't pass any traffic until
2031 * a controller has been defined and it tells us to do so. */
2033 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2034 union ofp_action action;
2035 struct cls_rule rule;
2037 memset(&action, 0, sizeof action);
2038 action.type = htons(OFPAT_OUTPUT);
2039 action.output.len = htons(sizeof action);
2040 action.output.port = htons(OFPP_NORMAL);
2041 cls_rule_init_catchall(&rule, 0);
2042 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2047 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2052 for (i = 0; i < br->n_ports; i++) {
2053 struct port *port = br->ports[i];
2054 for (j = 0; j < port->n_ifaces; j++) {
2055 struct iface *iface = port->ifaces[j];
2056 shash_add_once(ifaces, iface->name, iface);
2058 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2059 shash_add_once(ifaces, port->name, NULL);
2064 /* For robustness, in case the administrator moves around datapath ports behind
2065 * our back, we re-check all the datapath port numbers here.
2067 * This function will set the 'dp_ifidx' members of interfaces that have
2068 * disappeared to -1, so only call this function from a context where those
2069 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2070 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2071 * datapath, which doesn't support UINT16_MAX+1 ports. */
2073 bridge_fetch_dp_ifaces(struct bridge *br)
2075 struct dpif_port_dump dump;
2076 struct dpif_port dpif_port;
2079 /* Reset all interface numbers. */
2080 for (i = 0; i < br->n_ports; i++) {
2081 struct port *port = br->ports[i];
2082 for (j = 0; j < port->n_ifaces; j++) {
2083 struct iface *iface = port->ifaces[j];
2084 iface->dp_ifidx = -1;
2087 hmap_clear(&br->ifaces);
2089 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2090 struct iface *iface = iface_lookup(br, dpif_port.name);
2092 if (iface->dp_ifidx >= 0) {
2093 VLOG_WARN("%s reported interface %s twice",
2094 dpif_name(br->dpif), dpif_port.name);
2095 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2096 VLOG_WARN("%s reported interface %"PRIu16" twice",
2097 dpif_name(br->dpif), dpif_port.port_no);
2099 iface->dp_ifidx = dpif_port.port_no;
2100 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2101 hash_int(iface->dp_ifidx, 0));
2104 iface_set_ofport(iface->cfg,
2105 (iface->dp_ifidx >= 0
2106 ? odp_port_to_ofp_port(iface->dp_ifidx)
2112 /* Bridge packet processing functions. */
2115 bond_is_tcp_hash(const struct port *port)
2117 return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED;
2121 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2123 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2126 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2128 struct flow hash_flow;
2130 memcpy(&hash_flow, flow, sizeof hash_flow);
2131 hash_flow.vlan_tci = 0;
2133 /* The symmetric quality of this hash function is not required, but
2134 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2135 * purposes, so we use it out of convenience. */
2136 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2139 static struct bond_entry *
2140 lookup_bond_entry(const struct port *port, const struct flow *flow,
2143 assert(port->bond_mode != BM_AB);
2145 if (bond_is_tcp_hash(port)) {
2146 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2148 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2153 bond_choose_iface(const struct port *port)
2155 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2156 size_t i, best_down_slave = -1;
2157 long long next_delay_expiration = LLONG_MAX;
2159 for (i = 0; i < port->n_ifaces; i++) {
2160 struct iface *iface = port->ifaces[i];
2162 if (iface->enabled) {
2164 } else if (iface->delay_expires < next_delay_expiration
2165 && (iface->lacp_status & LACP_ATTACHED
2166 || !(port->lacp & LACP_NEGOTIATED))) {
2167 best_down_slave = i;
2168 next_delay_expiration = iface->delay_expires;
2172 if (best_down_slave != -1) {
2173 struct iface *iface = port->ifaces[best_down_slave];
2175 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2176 "since no other interface is up", iface->name,
2177 iface->delay_expires - time_msec());
2178 bond_enable_slave(iface, true);
2181 return best_down_slave;
2185 choose_output_iface(const struct port *port, const struct flow *flow,
2186 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2188 struct iface *iface;
2190 assert(port->n_ifaces);
2191 if (port->n_ifaces == 1) {
2192 iface = port->ifaces[0];
2193 } else if (port->bond_mode == BM_AB) {
2194 if (port->active_iface < 0) {
2195 *tags |= port->no_ifaces_tag;
2198 iface = port->ifaces[port->active_iface];
2200 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2201 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2202 || !port->ifaces[e->iface_idx]->enabled) {
2203 /* XXX select interface properly. The current interface selection
2204 * is only good for testing the rebalancing code. */
2205 e->iface_idx = bond_choose_iface(port);
2206 if (e->iface_idx < 0) {
2207 *tags |= port->no_ifaces_tag;
2210 e->iface_tag = tag_create_random();
2211 ((struct port *) port)->bond_compat_is_stale = true;
2213 *tags |= e->iface_tag;
2214 iface = port->ifaces[e->iface_idx];
2216 *dp_ifidx = iface->dp_ifidx;
2217 *tags |= iface->tag; /* Currently only used for bonding. */
2222 bond_link_status_update(struct iface *iface)
2224 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2225 struct port *port = iface->port;
2226 bool up = iface->up;
2227 int updelay, downdelay;
2229 updelay = port->updelay;
2230 downdelay = port->downdelay;
2232 if (iface->port->lacp & LACP_NEGOTIATED) {
2237 if (iface->port->lacp && up) {
2238 /* The interface is up if it's attached to an aggregator and its
2239 * partner is synchronized. The only exception is defaulted links.
2240 * They are not required to have synchronized partners because they
2241 * have no partners at all. However, they will only be attached if
2242 * negotiations failed on all interfaces in the bond. */
2243 up = iface->lacp_status & LACP_ATTACHED
2244 && (iface->lacp_partner.state & LACP_STATE_SYNC
2245 || iface->lacp_status & LACP_DEFAULTED);
2249 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2250 /* Nothing to do. */
2253 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2254 iface->name, up ? "up" : "down");
2255 if (up == iface->enabled) {
2256 iface->delay_expires = LLONG_MAX;
2257 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2258 iface->name, up ? "disabled" : "enabled");
2259 } else if (up && port->active_iface < 0) {
2260 bond_enable_slave(iface, true);
2262 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2263 "other interface is up", iface->name, updelay);
2266 int delay = up ? updelay : downdelay;
2267 iface->delay_expires = time_msec() + delay;
2270 "interface %s: will be %s if it stays %s for %d ms",
2272 up ? "enabled" : "disabled",
2280 bond_choose_active_iface(struct port *port)
2282 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2284 port->active_iface = bond_choose_iface(port);
2285 port->active_iface_tag = tag_create_random();
2286 if (port->active_iface >= 0) {
2287 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2288 port->name, port->ifaces[port->active_iface]->name);
2290 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2296 bond_enable_slave(struct iface *iface, bool enable)
2298 struct port *port = iface->port;
2299 struct bridge *br = port->bridge;
2301 /* This acts as a recursion check. If the act of disabling a slave
2302 * causes a different slave to be enabled, the flag will allow us to
2303 * skip redundant work when we reenter this function. It must be
2304 * cleared on exit to keep things safe with multiple bonds. */
2305 static bool moving_active_iface = false;
2307 iface->delay_expires = LLONG_MAX;
2308 if (enable == iface->enabled) {
2312 iface->enabled = enable;
2313 if (!iface->enabled) {
2314 VLOG_WARN("interface %s: disabled", iface->name);
2315 ofproto_revalidate(br->ofproto, iface->tag);
2316 if (iface->port_ifidx == port->active_iface) {
2317 ofproto_revalidate(br->ofproto,
2318 port->active_iface_tag);
2320 /* Disabling a slave can lead to another slave being immediately
2321 * enabled if there will be no active slaves but one is waiting
2322 * on an updelay. In this case we do not need to run most of the
2323 * code for the newly enabled slave since there was no period
2324 * without an active slave and it is redundant with the disabling
2326 moving_active_iface = true;
2327 bond_choose_active_iface(port);
2329 bond_send_learning_packets(port);
2331 VLOG_WARN("interface %s: enabled", iface->name);
2332 if (port->active_iface < 0 && !moving_active_iface) {
2333 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2334 bond_choose_active_iface(port);
2335 bond_send_learning_packets(port);
2337 iface->tag = tag_create_random();
2340 moving_active_iface = false;
2341 port->bond_compat_is_stale = true;
2344 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2345 * bond interface. */
2347 bond_update_fake_iface_stats(struct port *port)
2349 struct netdev_stats bond_stats;
2350 struct netdev *bond_dev;
2353 memset(&bond_stats, 0, sizeof bond_stats);
2355 for (i = 0; i < port->n_ifaces; i++) {
2356 struct netdev_stats slave_stats;
2358 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2359 /* XXX: We swap the stats here because they are swapped back when
2360 * reported by the internal device. The reason for this is
2361 * internal devices normally represent packets going into the system
2362 * but when used as fake bond device they represent packets leaving
2363 * the system. We really should do this in the internal device
2364 * itself because changing it here reverses the counts from the
2365 * perspective of the switch. However, the internal device doesn't
2366 * know what type of device it represents so we have to do it here
2368 bond_stats.tx_packets += slave_stats.rx_packets;
2369 bond_stats.tx_bytes += slave_stats.rx_bytes;
2370 bond_stats.rx_packets += slave_stats.tx_packets;
2371 bond_stats.rx_bytes += slave_stats.tx_bytes;
2375 if (!netdev_open_default(port->name, &bond_dev)) {
2376 netdev_set_stats(bond_dev, &bond_stats);
2377 netdev_close(bond_dev);
2382 bond_link_carrier_update(struct iface *iface, bool carrier)
2384 if (carrier == iface->up) {
2388 if (iface->lacp_status & LACP_CURRENT) {
2389 iface_set_lacp_expired(iface);
2392 iface->up = carrier;
2394 iface->port->bond_compat_is_stale = true;
2398 bond_run(struct bridge *br)
2402 for (i = 0; i < br->n_ports; i++) {
2403 struct port *port = br->ports[i];
2405 if (port->n_ifaces >= 2) {
2408 if (port->monitor) {
2409 assert(!port->miimon);
2411 /* Track carrier going up and down on interfaces. */
2412 while (!netdev_monitor_poll(port->monitor, &devname)) {
2413 struct iface *iface;
2415 iface = port_lookup_iface(port, devname);
2417 bool up = netdev_get_carrier(iface->netdev);
2418 bond_link_carrier_update(iface, up);
2423 assert(port->miimon);
2425 if (time_msec() >= port->bond_miimon_next_update) {
2426 for (j = 0; j < port->n_ifaces; j++) {
2427 struct iface *iface = port->ifaces[j];
2428 bool up = netdev_get_miimon(iface->netdev);
2429 bond_link_carrier_update(iface, up);
2431 port->bond_miimon_next_update = time_msec() +
2432 port->bond_miimon_interval;
2436 for (j = 0; j < port->n_ifaces; j++) {
2437 bond_link_status_update(port->ifaces[j]);
2440 for (j = 0; j < port->n_ifaces; j++) {
2441 struct iface *iface = port->ifaces[j];
2442 if (time_msec() >= iface->delay_expires) {
2443 bond_enable_slave(iface, !iface->enabled);
2447 if (port->bond_fake_iface
2448 && time_msec() >= port->bond_next_fake_iface_update) {
2449 bond_update_fake_iface_stats(port);
2450 port->bond_next_fake_iface_update = time_msec() + 1000;
2454 if (port->bond_compat_is_stale) {
2455 port->bond_compat_is_stale = false;
2456 port_update_bond_compat(port);
2462 bond_wait(struct bridge *br)
2466 for (i = 0; i < br->n_ports; i++) {
2467 struct port *port = br->ports[i];
2468 if (port->n_ifaces < 2) {
2472 if (port->monitor) {
2473 netdev_monitor_poll_wait(port->monitor);
2477 poll_timer_wait_until(port->bond_miimon_next_update);
2480 for (j = 0; j < port->n_ifaces; j++) {
2481 struct iface *iface = port->ifaces[j];
2482 if (iface->delay_expires != LLONG_MAX) {
2483 poll_timer_wait_until(iface->delay_expires);
2486 if (port->bond_fake_iface) {
2487 poll_timer_wait_until(port->bond_next_fake_iface_update);
2493 set_dst(struct dst *dst, const struct flow *flow,
2494 const struct port *in_port, const struct port *out_port,
2497 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2498 : in_port->vlan >= 0 ? in_port->vlan
2499 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2500 : vlan_tci_to_vid(flow->vlan_tci));
2501 return choose_output_iface(out_port, flow, dst->vlan,
2502 &dst->dp_ifidx, tags);
2506 swap_dst(struct dst *p, struct dst *q)
2508 struct dst tmp = *p;
2513 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2514 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2515 * that we push to the datapath. We could in fact fully sort the array by
2516 * vlan, but in most cases there are at most two different vlan tags so that's
2517 * possibly overkill.) */
2519 partition_dsts(struct dst_set *set, int vlan)
2521 struct dst *first = set->dsts;
2522 struct dst *last = set->dsts + set->n;
2524 while (first != last) {
2526 * - All dsts < first have vlan == 'vlan'.
2527 * - All dsts >= last have vlan != 'vlan'.
2528 * - first < last. */
2529 while (first->vlan == vlan) {
2530 if (++first == last) {
2535 /* Same invariants, plus one additional:
2536 * - first->vlan != vlan.
2538 while (last[-1].vlan != vlan) {
2539 if (--last == first) {
2544 /* Same invariants, plus one additional:
2545 * - last[-1].vlan == vlan.*/
2546 swap_dst(first++, --last);
2551 mirror_mask_ffs(mirror_mask_t mask)
2553 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2558 dst_set_init(struct dst_set *set)
2560 set->dsts = set->builtin;
2562 set->allocated = ARRAY_SIZE(set->builtin);
2566 dst_set_add(struct dst_set *set, const struct dst *dst)
2568 if (set->n >= set->allocated) {
2569 size_t new_allocated;
2570 struct dst *new_dsts;
2572 new_allocated = set->allocated * 2;
2573 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2574 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2578 set->dsts = new_dsts;
2579 set->allocated = new_allocated;
2581 set->dsts[set->n++] = *dst;
2585 dst_set_free(struct dst_set *set)
2587 if (set->dsts != set->builtin) {
2593 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2596 for (i = 0; i < set->n; i++) {
2597 if (set->dsts[i].vlan == test->vlan
2598 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2606 port_trunks_vlan(const struct port *port, uint16_t vlan)
2608 return (port->vlan < 0
2609 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2613 port_includes_vlan(const struct port *port, uint16_t vlan)
2615 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2619 port_is_floodable(const struct port *port)
2623 for (i = 0; i < port->n_ifaces; i++) {
2624 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2625 port->ifaces[i]->dp_ifidx)) {
2633 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2634 const struct port *in_port, const struct port *out_port,
2635 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2637 mirror_mask_t mirrors = in_port->src_mirrors;
2642 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2643 if (flow_vlan == 0) {
2644 flow_vlan = OFP_VLAN_NONE;
2647 if (out_port == FLOOD_PORT) {
2648 for (i = 0; i < br->n_ports; i++) {
2649 struct port *port = br->ports[i];
2651 && port_is_floodable(port)
2652 && port_includes_vlan(port, vlan)
2653 && !port->is_mirror_output_port
2654 && set_dst(&dst, flow, in_port, port, tags)) {
2655 mirrors |= port->dst_mirrors;
2656 dst_set_add(set, &dst);
2659 *nf_output_iface = NF_OUT_FLOOD;
2660 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2661 dst_set_add(set, &dst);
2662 *nf_output_iface = dst.dp_ifidx;
2663 mirrors |= out_port->dst_mirrors;
2667 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2668 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2670 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2671 && !dst_is_duplicate(set, &dst)) {
2672 dst_set_add(set, &dst);
2675 for (i = 0; i < br->n_ports; i++) {
2676 struct port *port = br->ports[i];
2677 if (port_includes_vlan(port, m->out_vlan)
2678 && set_dst(&dst, flow, in_port, port, tags))
2680 if (port->vlan < 0) {
2681 dst.vlan = m->out_vlan;
2683 if (dst_is_duplicate(set, &dst)) {
2687 /* Use the vlan tag on the original flow instead of
2688 * the one passed in the vlan parameter. This ensures
2689 * that we compare the vlan from before any implicit
2690 * tagging tags place. This is necessary because
2691 * dst->vlan is the final vlan, after removing implicit
2693 if (port == in_port && dst.vlan == flow_vlan) {
2694 /* Don't send out input port on same VLAN. */
2697 dst_set_add(set, &dst);
2702 mirrors &= mirrors - 1;
2705 partition_dsts(set, flow_vlan);
2708 static void OVS_UNUSED
2709 print_dsts(const struct dst_set *set)
2713 for (i = 0; i < set->n; i++) {
2714 const struct dst *dst = &set->dsts[i];
2716 printf(">p%"PRIu16, dst->dp_ifidx);
2717 if (dst->vlan != OFP_VLAN_NONE) {
2718 printf("v%"PRIu16, dst->vlan);
2724 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2725 const struct port *in_port, const struct port *out_port,
2726 tag_type *tags, struct ofpbuf *actions,
2727 uint16_t *nf_output_iface)
2734 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2737 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2738 if (cur_vlan == 0) {
2739 cur_vlan = OFP_VLAN_NONE;
2741 for (i = 0; i < set.n; i++) {
2742 const struct dst *dst = &set.dsts[i];
2743 if (dst->vlan != cur_vlan) {
2744 if (dst->vlan == OFP_VLAN_NONE) {
2745 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2748 tci = htons(dst->vlan & VLAN_VID_MASK);
2749 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2750 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2752 cur_vlan = dst->vlan;
2754 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2759 /* Returns the effective vlan of a packet, taking into account both the
2760 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2761 * the packet is untagged and -1 indicates it has an invalid header and
2762 * should be dropped. */
2763 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2764 struct port *in_port, bool have_packet)
2766 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2767 if (in_port->vlan >= 0) {
2769 /* XXX support double tagging? */
2771 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2772 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2773 "packet received on port %s configured with "
2774 "implicit VLAN %"PRIu16,
2775 br->name, vlan, in_port->name, in_port->vlan);
2779 vlan = in_port->vlan;
2781 if (!port_includes_vlan(in_port, vlan)) {
2783 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2784 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2785 "packet received on port %s not configured for "
2787 br->name, vlan, in_port->name, vlan);
2796 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2797 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2798 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2800 is_gratuitous_arp(const struct flow *flow)
2802 return (flow->dl_type == htons(ETH_TYPE_ARP)
2803 && eth_addr_is_broadcast(flow->dl_dst)
2804 && (flow->nw_proto == ARP_OP_REPLY
2805 || (flow->nw_proto == ARP_OP_REQUEST
2806 && flow->nw_src == flow->nw_dst)));
2810 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2811 struct port *in_port)
2813 enum grat_arp_lock_type lock_type;
2816 /* We don't want to learn from gratuitous ARP packets that are reflected
2817 * back over bond slaves so we lock the learning table. */
2818 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2819 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2820 GRAT_ARP_LOCK_CHECK;
2822 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2825 /* The log messages here could actually be useful in debugging,
2826 * so keep the rate limit relatively high. */
2827 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2829 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2830 "on port %s in VLAN %d",
2831 br->name, ETH_ADDR_ARGS(flow->dl_src),
2832 in_port->name, vlan);
2833 ofproto_revalidate(br->ofproto, rev_tag);
2837 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2838 * dropped. Returns true if they may be forwarded, false if they should be
2841 * If 'have_packet' is true, it indicates that the caller is processing a
2842 * received packet. If 'have_packet' is false, then the caller is just
2843 * revalidating an existing flow because configuration has changed. Either
2844 * way, 'have_packet' only affects logging (there is no point in logging errors
2845 * during revalidation).
2847 * Sets '*in_portp' to the input port. This will be a null pointer if
2848 * flow->in_port does not designate a known input port (in which case
2849 * is_admissible() returns false).
2851 * When returning true, sets '*vlanp' to the effective VLAN of the input
2852 * packet, as returned by flow_get_vlan().
2854 * May also add tags to '*tags', although the current implementation only does
2855 * so in one special case.
2858 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2859 tag_type *tags, int *vlanp, struct port **in_portp)
2861 struct iface *in_iface;
2862 struct port *in_port;
2865 /* Find the interface and port structure for the received packet. */
2866 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2868 /* No interface? Something fishy... */
2870 /* Odd. A few possible reasons here:
2872 * - We deleted an interface but there are still a few packets
2873 * queued up from it.
2875 * - Someone externally added an interface (e.g. with "ovs-dpctl
2876 * add-if") that we don't know about.
2878 * - Packet arrived on the local port but the local port is not
2879 * one of our bridge ports.
2881 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2883 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2884 "interface %"PRIu16, br->name, flow->in_port);
2890 *in_portp = in_port = in_iface->port;
2891 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2896 /* Drop frames for reserved multicast addresses. */
2897 if (eth_addr_is_reserved(flow->dl_dst)) {
2901 /* Drop frames on ports reserved for mirroring. */
2902 if (in_port->is_mirror_output_port) {
2904 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2905 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2906 "%s, which is reserved exclusively for mirroring",
2907 br->name, in_port->name);
2912 /* When using LACP, do not accept packets from disabled interfaces. */
2913 if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) {
2917 /* Packets received on non-LACP bonds need special attention to avoid
2919 if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) {
2921 bool is_grat_arp_locked;
2923 if (eth_addr_is_multicast(flow->dl_dst)) {
2924 *tags |= in_port->active_iface_tag;
2925 if (in_port->active_iface != in_iface->port_ifidx) {
2926 /* Drop all multicast packets on inactive slaves. */
2931 /* Drop all packets for which we have learned a different input
2932 * port, because we probably sent the packet on one slave and got
2933 * it back on the other. Gratuitous ARP packets are an exception
2934 * to this rule: the host has moved to another switch. The exception
2935 * to the exception is if we locked the learning table to avoid
2936 * reflections on bond slaves. If this is the case, just drop the
2938 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2939 &is_grat_arp_locked);
2940 if (src_idx != -1 && src_idx != in_port->port_idx &&
2941 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2949 /* If the composed actions may be applied to any packet in the given 'flow',
2950 * returns true. Otherwise, the actions should only be applied to 'packet', or
2951 * not at all, if 'packet' was NULL. */
2953 process_flow(struct bridge *br, const struct flow *flow,
2954 const struct ofpbuf *packet, struct ofpbuf *actions,
2955 tag_type *tags, uint16_t *nf_output_iface)
2957 struct port *in_port;
2958 struct port *out_port;
2962 /* Check whether we should drop packets in this flow. */
2963 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2968 /* Learn source MAC (but don't try to learn from revalidation). */
2970 update_learning_table(br, flow, vlan, in_port);
2973 /* Determine output port. */
2974 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2976 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2977 out_port = br->ports[out_port_idx];
2978 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2979 /* If we are revalidating but don't have a learning entry then
2980 * eject the flow. Installing a flow that floods packets opens
2981 * up a window of time where we could learn from a packet reflected
2982 * on a bond and blackhole packets before the learning table is
2983 * updated to reflect the correct port. */
2986 out_port = FLOOD_PORT;
2989 /* Don't send packets out their input ports. */
2990 if (in_port == out_port) {
2996 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3004 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3005 struct ofpbuf *actions, tag_type *tags,
3006 uint16_t *nf_output_iface, void *br_)
3008 struct iface *iface;
3009 struct bridge *br = br_;
3011 COVERAGE_INC(bridge_process_flow);
3013 iface = iface_from_dp_ifidx(br, flow->in_port);
3015 if (cfm_should_process_flow(flow)) {
3016 if (packet && iface->cfm) {
3017 cfm_process_heartbeat(iface->cfm, packet);
3020 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3022 lacp_process_packet(packet, iface);
3027 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3031 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3032 const struct nlattr *actions,
3034 unsigned long long int n_bytes, void *br_)
3036 struct bridge *br = br_;
3037 const struct nlattr *a;
3038 struct port *in_port;
3043 /* Feed information from the active flows back into the learning table to
3044 * ensure that table is always in sync with what is actually flowing
3045 * through the datapath.
3047 * We test that 'tags' is nonzero to ensure that only flows that include an
3048 * OFPP_NORMAL action are used for learning. This works because
3049 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3050 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3051 update_learning_table(br, flow, vlan, in_port);
3054 /* Account for bond slave utilization. */
3055 if (!br->has_bonded_ports) {
3058 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3059 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3060 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3061 if (out_port && out_port->n_ifaces >= 2 &&
3062 out_port->bond_mode == BM_SLB) {
3063 uint16_t vlan = (flow->vlan_tci
3064 ? vlan_tci_to_vid(flow->vlan_tci)
3066 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3067 e->tx_bytes += n_bytes;
3074 bridge_account_checkpoint_ofhook_cb(void *br_)
3076 struct bridge *br = br_;
3080 if (!br->has_bonded_ports) {
3085 for (i = 0; i < br->n_ports; i++) {
3086 struct port *port = br->ports[i];
3087 if (port->n_ifaces > 1 && port->bond_mode == BM_SLB
3088 && now >= port->bond_next_rebalance) {
3089 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3090 bond_rebalance_port(port);
3095 static struct ofhooks bridge_ofhooks = {
3096 bridge_normal_ofhook_cb,
3097 bridge_account_flow_ofhook_cb,
3098 bridge_account_checkpoint_ofhook_cb,
3101 /* LACP functions. */
3104 lacp_process_packet(const struct ofpbuf *packet, struct iface *iface)
3106 const struct lacp_pdu *pdu;
3108 if (!iface->port->lacp) {
3112 pdu = parse_lacp_packet(packet);
3117 iface->lacp_status = LACP_CURRENT;
3118 iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX;
3120 iface->lacp_actor.state = iface_get_lacp_state(iface);
3121 if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) {
3125 if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) {
3126 iface->port->lacp_need_update = true;
3127 iface->lacp_partner = pdu->actor;
3132 lacp_update_ifaces(struct port *port)
3136 struct lacp_info lead_pri;
3137 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3139 port->lacp_need_update = false;
3140 COVERAGE_INC(bridge_lacp_update);
3146 VLOG_DBG_RL(&rl, "port %s: re-evaluating LACP link status", port->name);
3149 for (i = 0; i < port->n_ifaces; i++) {
3150 struct iface *iface = port->ifaces[i];
3151 struct lacp_info pri;
3153 iface->lacp_status |= LACP_ATTACHED;
3154 ofproto_revalidate(port->bridge->ofproto, iface->tag);
3156 /* Don't allow loopback interfaces to send traffic or lead. */
3157 if (eth_addr_equals(iface->lacp_partner.sysid,
3158 iface->lacp_actor.sysid)) {
3159 VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is "
3160 "connected to its own bridge", iface->name);
3161 iface->lacp_status &= ~LACP_ATTACHED;
3165 if (iface->lacp_status & LACP_DEFAULTED) {
3169 iface_get_lacp_priority(iface, &pri);
3171 if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) {
3178 port->lacp &= ~LACP_NEGOTIATED;
3182 port->lacp |= LACP_NEGOTIATED;
3184 for (i = 0; i < port->n_ifaces; i++) {
3185 struct iface *iface = port->ifaces[i];
3187 if (iface->lacp_status & LACP_DEFAULTED
3188 || lead->lacp_partner.key != iface->lacp_partner.key
3189 || !eth_addr_equals(lead->lacp_partner.sysid,
3190 iface->lacp_partner.sysid)) {
3191 iface->lacp_status &= ~LACP_ATTACHED;
3197 lacp_iface_may_tx(const struct iface *iface)
3199 return iface->port->lacp & LACP_ACTIVE
3200 || iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED);
3204 lacp_run(struct bridge *br)
3207 struct ofpbuf packet;
3209 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3211 for (i = 0; i < br->n_ports; i++) {
3212 struct port *port = br->ports[i];
3218 for (j = 0; j < port->n_ifaces; j++) {
3219 struct iface *iface = port->ifaces[j];
3221 if (time_msec() > iface->lacp_rx) {
3222 if (iface->lacp_status & LACP_CURRENT) {
3223 iface_set_lacp_expired(iface);
3224 } else if (iface->lacp_status & LACP_EXPIRED) {
3225 iface_set_lacp_defaulted(iface);
3230 if (port->lacp_need_update) {
3231 lacp_update_ifaces(port);
3234 for (j = 0; j < port->n_ifaces; j++) {
3235 struct iface *iface = port->ifaces[j];
3236 uint8_t ea[ETH_ADDR_LEN];
3239 if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) {
3243 error = netdev_get_etheraddr(iface->netdev, ea);
3245 iface->lacp_actor.state = iface_get_lacp_state(iface);
3246 compose_lacp_packet(&packet, &iface->lacp_actor,
3247 &iface->lacp_partner, ea);
3248 iface_send_packet(iface, &packet);
3250 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3251 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3252 "(%s)", iface->name, strerror(error));
3255 iface->lacp_tx = time_msec() +
3256 (iface->lacp_partner.state & LACP_STATE_TIME
3258 : LACP_SLOW_TIME_TX);
3261 ofpbuf_uninit(&packet);
3265 lacp_wait(struct bridge *br)
3269 for (i = 0; i < br->n_ports; i++) {
3270 struct port *port = br->ports[i];
3276 for (j = 0; j < port->n_ifaces; j++) {
3277 struct iface *iface = port->ifaces[j];
3279 if (lacp_iface_may_tx(iface)) {
3280 poll_timer_wait_until(iface->lacp_tx);
3283 if (iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED)) {
3284 poll_timer_wait_until(iface->lacp_rx);
3290 /* Bonding functions. */
3292 /* Statistics for a single interface on a bonded port, used for load-based
3293 * bond rebalancing. */
3294 struct slave_balance {
3295 struct iface *iface; /* The interface. */
3296 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3298 /* All the "bond_entry"s that are assigned to this interface, in order of
3299 * increasing tx_bytes. */
3300 struct bond_entry **hashes;
3305 bond_mode_to_string(enum bond_mode bm) {
3306 static char *bm_slb = "balance-slb";
3307 static char *bm_ab = "active-backup";
3308 static char *bm_tcp = "balance-tcp";
3311 case BM_SLB: return bm_slb;
3312 case BM_AB: return bm_ab;
3313 case BM_TCP: return bm_tcp;
3320 /* Sorts pointers to pointers to bond_entries in ascending order by the
3321 * interface to which they are assigned, and within a single interface in
3322 * ascending order of bytes transmitted. */
3324 compare_bond_entries(const void *a_, const void *b_)
3326 const struct bond_entry *const *ap = a_;
3327 const struct bond_entry *const *bp = b_;
3328 const struct bond_entry *a = *ap;
3329 const struct bond_entry *b = *bp;
3330 if (a->iface_idx != b->iface_idx) {
3331 return a->iface_idx > b->iface_idx ? 1 : -1;
3332 } else if (a->tx_bytes != b->tx_bytes) {
3333 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3339 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3340 * *descending* order by number of bytes transmitted. */
3342 compare_slave_balance(const void *a_, const void *b_)
3344 const struct slave_balance *a = a_;
3345 const struct slave_balance *b = b_;
3346 if (a->iface->enabled != b->iface->enabled) {
3347 return a->iface->enabled ? -1 : 1;
3348 } else if (a->tx_bytes != b->tx_bytes) {
3349 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3356 swap_bals(struct slave_balance *a, struct slave_balance *b)
3358 struct slave_balance tmp = *a;
3363 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3364 * given that 'p' (and only 'p') might be in the wrong location.
3366 * This function invalidates 'p', since it might now be in a different memory
3369 resort_bals(struct slave_balance *p,
3370 struct slave_balance bals[], size_t n_bals)
3373 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3374 swap_bals(p, p - 1);
3376 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3377 swap_bals(p, p + 1);
3383 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3385 if (VLOG_IS_DBG_ENABLED()) {
3386 struct ds ds = DS_EMPTY_INITIALIZER;
3387 const struct slave_balance *b;
3389 for (b = bals; b < bals + n_bals; b++) {
3393 ds_put_char(&ds, ',');
3395 ds_put_format(&ds, " %s %"PRIu64"kB",
3396 b->iface->name, b->tx_bytes / 1024);
3398 if (!b->iface->enabled) {
3399 ds_put_cstr(&ds, " (disabled)");
3401 if (b->n_hashes > 0) {
3402 ds_put_cstr(&ds, " (");
3403 for (i = 0; i < b->n_hashes; i++) {
3404 const struct bond_entry *e = b->hashes[i];
3406 ds_put_cstr(&ds, " + ");
3408 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3409 e - port->bond_hash, e->tx_bytes / 1024);
3411 ds_put_cstr(&ds, ")");
3414 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3419 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3421 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3424 struct bond_entry *hash = from->hashes[hash_idx];
3425 struct port *port = from->iface->port;
3426 uint64_t delta = hash->tx_bytes;
3428 assert(port->bond_mode == BM_SLB);
3430 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3431 "from %s to %s (now carrying %"PRIu64"kB and "
3432 "%"PRIu64"kB load, respectively)",
3433 port->name, delta / 1024, hash - port->bond_hash,
3434 from->iface->name, to->iface->name,
3435 (from->tx_bytes - delta) / 1024,
3436 (to->tx_bytes + delta) / 1024);
3438 /* Delete element from from->hashes.
3440 * We don't bother to add the element to to->hashes because not only would
3441 * it require more work, the only purpose it would be to allow that hash to
3442 * be migrated to another slave in this rebalancing run, and there is no
3443 * point in doing that. */
3444 if (hash_idx == 0) {
3447 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3448 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3452 /* Shift load away from 'from' to 'to'. */
3453 from->tx_bytes -= delta;
3454 to->tx_bytes += delta;
3456 /* Arrange for flows to be revalidated. */
3457 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3458 hash->iface_idx = to->iface->port_ifidx;
3459 hash->iface_tag = tag_create_random();
3463 bond_rebalance_port(struct port *port)
3465 struct slave_balance *bals;
3467 struct bond_entry *hashes[BOND_MASK + 1];
3468 struct slave_balance *b, *from, *to;
3469 struct bond_entry *e;
3472 assert(port->bond_mode != BM_AB);
3474 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3475 * descending order of tx_bytes, so that bals[0] represents the most
3476 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3479 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3480 * array for each slave_balance structure, we sort our local array of
3481 * hashes in order by slave, so that all of the hashes for a given slave
3482 * become contiguous in memory, and then we point each 'hashes' members of
3483 * a slave_balance structure to the start of a contiguous group. */
3484 n_bals = port->n_ifaces;
3485 bals = xmalloc(n_bals * sizeof *bals);
3486 for (b = bals; b < &bals[n_bals]; b++) {
3487 b->iface = port->ifaces[b - bals];
3492 for (i = 0; i <= BOND_MASK; i++) {
3493 hashes[i] = &port->bond_hash[i];
3495 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3496 for (i = 0; i <= BOND_MASK; i++) {
3498 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3499 b = &bals[e->iface_idx];
3500 b->tx_bytes += e->tx_bytes;
3502 b->hashes = &hashes[i];
3507 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3508 log_bals(bals, n_bals, port);
3510 /* Discard slaves that aren't enabled (which were sorted to the back of the
3511 * array earlier). */
3512 while (!bals[n_bals - 1].iface->enabled) {
3519 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3520 to = &bals[n_bals - 1];
3521 for (from = bals; from < to; ) {
3522 uint64_t overload = from->tx_bytes - to->tx_bytes;
3523 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3524 /* The extra load on 'from' (and all less-loaded slaves), compared
3525 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3526 * it is less than ~1Mbps. No point in rebalancing. */
3528 } else if (from->n_hashes == 1) {
3529 /* 'from' only carries a single MAC hash, so we can't shift any
3530 * load away from it, even though we want to. */
3533 /* 'from' is carrying significantly more load than 'to', and that
3534 * load is split across at least two different hashes. Pick a hash
3535 * to migrate to 'to' (the least-loaded slave), given that doing so
3536 * must decrease the ratio of the load on the two slaves by at
3539 * The sort order we use means that we prefer to shift away the
3540 * smallest hashes instead of the biggest ones. There is little
3541 * reason behind this decision; we could use the opposite sort
3542 * order to shift away big hashes ahead of small ones. */
3545 for (i = 0; i < from->n_hashes; i++) {
3546 double old_ratio, new_ratio;
3547 uint64_t delta = from->hashes[i]->tx_bytes;
3549 if (delta == 0 || from->tx_bytes - delta == 0) {
3550 /* Pointless move. */
3554 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3556 if (to->tx_bytes == 0) {
3557 /* Nothing on the new slave, move it. */
3561 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3562 new_ratio = (double)(from->tx_bytes - delta) /
3563 (to->tx_bytes + delta);
3565 if (new_ratio == 0) {
3566 /* Should already be covered but check to prevent division
3571 if (new_ratio < 1) {
3572 new_ratio = 1 / new_ratio;
3575 if (old_ratio - new_ratio > 0.1) {
3576 /* Would decrease the ratio, move it. */
3580 if (i < from->n_hashes) {
3581 bond_shift_load(from, to, i);
3582 port->bond_compat_is_stale = true;
3584 /* If the result of the migration changed the relative order of
3585 * 'from' and 'to' swap them back to maintain invariants. */
3586 if (order_swapped) {
3587 swap_bals(from, to);
3590 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3591 * point to different slave_balance structures. It is only
3592 * valid to do these two operations in a row at all because we
3593 * know that 'from' will not move past 'to' and vice versa. */
3594 resort_bals(from, bals, n_bals);
3595 resort_bals(to, bals, n_bals);
3602 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3603 * historical data to decay to <1% in 7 rebalancing runs. */
3604 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3613 bond_send_learning_packets(struct port *port)
3615 struct bridge *br = port->bridge;
3616 struct mac_entry *e;
3617 struct ofpbuf packet;
3618 int error, n_packets, n_errors;
3620 if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
3624 ofpbuf_init(&packet, 128);
3625 error = n_packets = n_errors = 0;
3626 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3627 union ofp_action actions[2], *a;
3633 if (e->port == port->port_idx) {
3637 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3639 flow_extract(&packet, 0, ODPP_NONE, &flow);
3641 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3645 /* Compose actions. */
3646 memset(actions, 0, sizeof actions);
3649 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3650 a->vlan_vid.len = htons(sizeof *a);
3651 a->vlan_vid.vlan_vid = htons(e->vlan);
3654 a->output.type = htons(OFPAT_OUTPUT);
3655 a->output.len = htons(sizeof *a);
3656 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3661 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3668 ofpbuf_uninit(&packet);
3671 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3672 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3673 "packets, last error was: %s",
3674 port->name, n_errors, n_packets, strerror(error));
3676 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3677 port->name, n_packets);
3681 /* Bonding unixctl user interface functions. */
3684 bond_unixctl_list(struct unixctl_conn *conn,
3685 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3687 struct ds ds = DS_EMPTY_INITIALIZER;
3688 const struct bridge *br;
3690 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3692 LIST_FOR_EACH (br, node, &all_bridges) {
3695 for (i = 0; i < br->n_ports; i++) {
3696 const struct port *port = br->ports[i];
3697 if (port->n_ifaces > 1) {
3700 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3701 bond_mode_to_string(port->bond_mode));
3702 for (j = 0; j < port->n_ifaces; j++) {
3703 const struct iface *iface = port->ifaces[j];
3705 ds_put_cstr(&ds, ", ");
3707 ds_put_cstr(&ds, iface->name);
3709 ds_put_char(&ds, '\n');
3713 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3717 static struct port *
3718 bond_find(const char *name)
3720 const struct bridge *br;
3722 LIST_FOR_EACH (br, node, &all_bridges) {
3725 for (i = 0; i < br->n_ports; i++) {
3726 struct port *port = br->ports[i];
3727 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3736 ds_put_lacp_state(struct ds *ds, uint8_t state)
3738 if (state & LACP_STATE_ACT) {
3739 ds_put_cstr(ds, "activity ");
3742 if (state & LACP_STATE_TIME) {
3743 ds_put_cstr(ds, "timeout ");
3746 if (state & LACP_STATE_AGG) {
3747 ds_put_cstr(ds, "aggregation ");
3750 if (state & LACP_STATE_SYNC) {
3751 ds_put_cstr(ds, "synchronized ");
3754 if (state & LACP_STATE_COL) {
3755 ds_put_cstr(ds, "collecting ");
3758 if (state & LACP_STATE_DIST) {
3759 ds_put_cstr(ds, "distributing ");
3762 if (state & LACP_STATE_DEF) {
3763 ds_put_cstr(ds, "defaulted ");
3766 if (state & LACP_STATE_EXP) {
3767 ds_put_cstr(ds, "expired ");
3772 bond_unixctl_show(struct unixctl_conn *conn,
3773 const char *args, void *aux OVS_UNUSED)
3775 struct ds ds = DS_EMPTY_INITIALIZER;
3776 const struct port *port;
3779 port = bond_find(args);
3781 unixctl_command_reply(conn, 501, "no such bond");
3785 ds_put_format(&ds, "bond_mode: %s\n",
3786 bond_mode_to_string(port->bond_mode));
3789 ds_put_format(&ds, "lacp: %s\n",
3790 port->lacp & LACP_ACTIVE ? "active" : "passive");
3792 ds_put_cstr(&ds, "lacp: off\n");
3795 if (port->bond_mode != BM_AB) {
3796 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3797 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3801 ds_put_format(&ds, "bond-detect-mode: %s\n",
3802 port->miimon ? "miimon" : "carrier");
3805 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3806 port->bond_miimon_interval);
3809 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3810 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3812 if (port->bond_mode != BM_AB) {
3813 ds_put_format(&ds, "next rebalance: %lld ms\n",
3814 port->bond_next_rebalance - time_msec());
3817 for (j = 0; j < port->n_ifaces; j++) {
3818 const struct iface *iface = port->ifaces[j];
3819 struct bond_entry *be;
3823 ds_put_format(&ds, "\nslave %s: %s\n",
3824 iface->name, iface->enabled ? "enabled" : "disabled");
3825 if (j == port->active_iface) {
3826 ds_put_cstr(&ds, "\tactive slave\n");
3828 if (iface->delay_expires != LLONG_MAX) {
3829 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3830 iface->enabled ? "downdelay" : "updelay",
3831 iface->delay_expires - time_msec());
3835 ds_put_cstr(&ds, "\tstatus: ");
3837 if (iface->lacp_status & LACP_CURRENT) {
3838 ds_put_cstr(&ds, "current ");
3841 if (iface->lacp_status & LACP_EXPIRED) {
3842 ds_put_cstr(&ds, "expired ");
3845 if (iface->lacp_status & LACP_DEFAULTED) {
3846 ds_put_cstr(&ds, "defaulted ");
3849 if (iface->lacp_status & LACP_ATTACHED) {
3850 ds_put_cstr(&ds, "attached ");
3853 ds_put_cstr(&ds, "\n");
3855 ds_put_cstr(&ds, "\n\tactor sysid: ");
3856 ds_put_format(&ds, ETH_ADDR_FMT,
3857 ETH_ADDR_ARGS(iface->lacp_actor.sysid));
3858 ds_put_cstr(&ds, "\n");
3860 ds_put_format(&ds, "\tactor sys_priority: %u\n",
3861 ntohs(iface->lacp_actor.sys_priority));
3863 ds_put_format(&ds, "\tactor portid: %u\n",
3864 ntohs(iface->lacp_actor.portid));
3866 ds_put_format(&ds, "\tactor port_priority: %u\n",
3867 ntohs(iface->lacp_actor.port_priority));
3869 ds_put_format(&ds, "\tactor key: %u\n",
3870 ntohs(iface->lacp_actor.key));
3872 ds_put_cstr(&ds, "\tactor state: ");
3873 ds_put_lacp_state(&ds, iface_get_lacp_state(iface));
3874 ds_put_cstr(&ds, "\n\n");
3876 ds_put_cstr(&ds, "\tpartner sysid: ");
3877 ds_put_format(&ds, ETH_ADDR_FMT,
3878 ETH_ADDR_ARGS(iface->lacp_partner.sysid));
3879 ds_put_cstr(&ds, "\n");
3881 ds_put_format(&ds, "\tpartner sys_priority: %u\n",
3882 ntohs(iface->lacp_partner.sys_priority));
3884 ds_put_format(&ds, "\tpartner portid: %u\n",
3885 ntohs(iface->lacp_partner.portid));
3887 ds_put_format(&ds, "\tpartner port_priority: %u\n",
3888 ntohs(iface->lacp_partner.port_priority));
3890 ds_put_format(&ds, "\tpartner key: %u\n",
3891 ntohs(iface->lacp_partner.key));
3893 ds_put_cstr(&ds, "\tpartner state: ");
3894 ds_put_lacp_state(&ds, iface->lacp_partner.state);
3895 ds_put_cstr(&ds, "\n");
3898 if (port->bond_mode == BM_AB) {
3903 memset(&flow, 0, sizeof flow);
3904 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3905 int hash = be - port->bond_hash;
3906 struct mac_entry *me;
3908 if (be->iface_idx != j) {
3912 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3913 hash, be->tx_bytes / 1024);
3915 if (port->bond_mode != BM_SLB) {
3920 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3924 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3925 if (bond_hash_src(me->mac, me->vlan) == hash
3926 && me->port != port->port_idx
3927 && choose_output_iface(port, &flow, me->vlan,
3929 && dp_ifidx == iface->dp_ifidx)
3931 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3932 ETH_ADDR_ARGS(me->mac));
3937 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3942 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3943 void *aux OVS_UNUSED)
3945 char *args = (char *) args_;
3946 char *save_ptr = NULL;
3947 char *bond_s, *hash_s, *slave_s;
3949 struct iface *iface;
3950 struct bond_entry *entry;
3953 bond_s = strtok_r(args, " ", &save_ptr);
3954 hash_s = strtok_r(NULL, " ", &save_ptr);
3955 slave_s = strtok_r(NULL, " ", &save_ptr);
3957 unixctl_command_reply(conn, 501,
3958 "usage: bond/migrate BOND HASH SLAVE");
3962 port = bond_find(bond_s);
3964 unixctl_command_reply(conn, 501, "no such bond");
3968 if (port->bond_mode != BM_SLB) {
3969 unixctl_command_reply(conn, 501, "not an SLB bond");
3973 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3974 hash = atoi(hash_s) & BOND_MASK;
3976 unixctl_command_reply(conn, 501, "bad hash");
3980 iface = port_lookup_iface(port, slave_s);
3982 unixctl_command_reply(conn, 501, "no such slave");
3986 if (!iface->enabled) {
3987 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3991 entry = &port->bond_hash[hash];
3992 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3993 entry->iface_idx = iface->port_ifidx;
3994 entry->iface_tag = tag_create_random();
3995 port->bond_compat_is_stale = true;
3996 unixctl_command_reply(conn, 200, "migrated");
4000 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
4001 void *aux OVS_UNUSED)
4003 char *args = (char *) args_;
4004 char *save_ptr = NULL;
4005 char *bond_s, *slave_s;
4007 struct iface *iface;
4009 bond_s = strtok_r(args, " ", &save_ptr);
4010 slave_s = strtok_r(NULL, " ", &save_ptr);
4012 unixctl_command_reply(conn, 501,
4013 "usage: bond/set-active-slave BOND SLAVE");
4017 port = bond_find(bond_s);
4019 unixctl_command_reply(conn, 501, "no such bond");
4023 iface = port_lookup_iface(port, slave_s);
4025 unixctl_command_reply(conn, 501, "no such slave");
4029 if (!iface->enabled) {
4030 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
4034 if (port->active_iface != iface->port_ifidx) {
4035 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4036 port->active_iface = iface->port_ifidx;
4037 port->active_iface_tag = tag_create_random();
4038 VLOG_INFO("port %s: active interface is now %s",
4039 port->name, iface->name);
4040 bond_send_learning_packets(port);
4041 unixctl_command_reply(conn, 200, "done");
4043 unixctl_command_reply(conn, 200, "no change");
4048 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
4050 char *args = (char *) args_;
4051 char *save_ptr = NULL;
4052 char *bond_s, *slave_s;
4054 struct iface *iface;
4056 bond_s = strtok_r(args, " ", &save_ptr);
4057 slave_s = strtok_r(NULL, " ", &save_ptr);
4059 unixctl_command_reply(conn, 501,
4060 "usage: bond/enable/disable-slave BOND SLAVE");
4064 port = bond_find(bond_s);
4066 unixctl_command_reply(conn, 501, "no such bond");
4070 iface = port_lookup_iface(port, slave_s);
4072 unixctl_command_reply(conn, 501, "no such slave");
4076 bond_enable_slave(iface, enable);
4077 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
4081 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
4082 void *aux OVS_UNUSED)
4084 enable_slave(conn, args, true);
4088 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
4089 void *aux OVS_UNUSED)
4091 enable_slave(conn, args, false);
4095 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
4096 void *aux OVS_UNUSED)
4098 char *args = (char *) args_;
4099 uint8_t mac[ETH_ADDR_LEN];
4103 char *mac_s, *vlan_s;
4104 char *save_ptr = NULL;
4106 mac_s = strtok_r(args, " ", &save_ptr);
4107 vlan_s = strtok_r(NULL, " ", &save_ptr);
4110 if (sscanf(vlan_s, "%u", &vlan) != 1) {
4111 unixctl_command_reply(conn, 501, "invalid vlan");
4115 vlan = OFP_VLAN_NONE;
4118 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
4119 == ETH_ADDR_SCAN_COUNT) {
4120 hash = bond_hash_src(mac, vlan);
4122 hash_cstr = xasprintf("%u", hash);
4123 unixctl_command_reply(conn, 200, hash_cstr);
4126 unixctl_command_reply(conn, 501, "invalid mac");
4133 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
4134 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
4135 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
4136 unixctl_command_register("bond/set-active-slave",
4137 bond_unixctl_set_active_slave, NULL);
4138 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
4140 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
4142 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
4145 /* Port functions. */
4147 static struct port *
4148 port_create(struct bridge *br, const char *name)
4152 port = xzalloc(sizeof *port);
4154 port->port_idx = br->n_ports;
4156 port->trunks = NULL;
4157 port->name = xstrdup(name);
4158 port->active_iface = -1;
4160 if (br->n_ports >= br->allocated_ports) {
4161 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
4164 br->ports[br->n_ports++] = port;
4165 shash_add_assert(&br->port_by_name, port->name, port);
4167 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
4174 get_port_other_config(const struct ovsrec_port *port, const char *key,
4175 const char *default_value)
4179 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
4181 return value ? value : default_value;
4185 get_interface_other_config(const struct ovsrec_interface *iface,
4186 const char *key, const char *default_value)
4190 value = get_ovsrec_key_value(&iface->header_,
4191 &ovsrec_interface_col_other_config, key);
4192 return value ? value : default_value;
4196 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
4198 struct shash new_ifaces;
4201 /* Collect list of new interfaces. */
4202 shash_init(&new_ifaces);
4203 for (i = 0; i < cfg->n_interfaces; i++) {
4204 const char *name = cfg->interfaces[i]->name;
4205 shash_add_once(&new_ifaces, name, NULL);
4208 /* Get rid of deleted interfaces. */
4209 for (i = 0; i < port->n_ifaces; ) {
4210 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
4211 iface_destroy(port->ifaces[i]);
4217 shash_destroy(&new_ifaces);
4221 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4223 const char *detect_mode;
4224 struct shash new_ifaces;
4225 long long int next_rebalance, miimon_next_update, lacp_priority;
4226 unsigned long *trunks;
4232 /* Update settings. */
4233 port->updelay = cfg->bond_updelay;
4234 if (port->updelay < 0) {
4237 port->downdelay = cfg->bond_downdelay;
4238 if (port->downdelay < 0) {
4239 port->downdelay = 0;
4241 port->bond_rebalance_interval = atoi(
4242 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4243 if (port->bond_rebalance_interval < 1000) {
4244 port->bond_rebalance_interval = 1000;
4246 next_rebalance = time_msec() + port->bond_rebalance_interval;
4247 if (port->bond_next_rebalance > next_rebalance) {
4248 port->bond_next_rebalance = next_rebalance;
4251 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4254 if (!strcmp(detect_mode, "carrier")) {
4255 port->miimon = false;
4256 } else if (!strcmp(detect_mode, "miimon")) {
4257 port->miimon = true;
4259 port->miimon = false;
4260 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
4261 "carrier", port->name, detect_mode);
4264 port->bond_miimon_interval = atoi(
4265 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4266 if (port->bond_miimon_interval < 100) {
4267 port->bond_miimon_interval = 100;
4269 miimon_next_update = time_msec() + port->bond_miimon_interval;
4270 if (port->bond_miimon_next_update > miimon_next_update) {
4271 port->bond_miimon_next_update = miimon_next_update;
4274 if (!port->cfg->bond_mode ||
4275 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4276 port->bond_mode = BM_SLB;
4277 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4278 port->bond_mode = BM_AB;
4279 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4280 port->bond_mode = BM_TCP;
4282 port->bond_mode = BM_SLB;
4283 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4284 port->name, port->cfg->bond_mode,
4285 bond_mode_to_string(port->bond_mode));
4288 /* Add new interfaces and update 'cfg' member of existing ones. */
4289 shash_init(&new_ifaces);
4290 for (i = 0; i < cfg->n_interfaces; i++) {
4291 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4292 struct iface *iface;
4294 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4295 VLOG_WARN("port %s: %s specified twice as port interface",
4296 port->name, if_cfg->name);
4297 iface_set_ofport(if_cfg, -1);
4301 iface = iface_lookup(port->bridge, if_cfg->name);
4303 if (iface->port != port) {
4304 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4306 port->bridge->name, if_cfg->name, iface->port->name);
4309 iface->cfg = if_cfg;
4311 iface = iface_create(port, if_cfg);
4314 /* Determine interface type. The local port always has type
4315 * "internal". Other ports take their type from the database and
4316 * default to "system" if none is specified. */
4317 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4318 : if_cfg->type[0] ? if_cfg->type
4322 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4325 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4326 iface->lacp_priority = UINT16_MAX;
4328 iface->lacp_priority = lacp_priority;
4331 shash_destroy(&new_ifaces);
4334 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4336 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4337 /* Prefer bondable links if unspecified. */
4338 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4340 port->lacp_priority = lacp_priority;
4343 if (!port->cfg->lacp) {
4344 /* XXX when LACP implementation has been sufficiently tested, enable by
4345 * default and make active on bonded ports. */
4347 } else if (!strcmp(port->cfg->lacp, "off")) {
4349 } else if (!strcmp(port->cfg->lacp, "active")) {
4350 port->lacp = LACP_ACTIVE;
4351 } else if (!strcmp(port->cfg->lacp, "passive")) {
4352 port->lacp = LACP_PASSIVE;
4354 VLOG_WARN("port %s: unknown LACP mode %s",
4355 port->name, port->cfg->lacp);
4362 if (port->n_ifaces < 2) {
4364 if (vlan >= 0 && vlan <= 4095) {
4365 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4370 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4371 * they even work as-is. But they have not been tested. */
4372 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4376 if (port->vlan != vlan) {
4378 bridge_flush(port->bridge);
4381 /* Get trunked VLANs. */
4383 if (vlan < 0 && cfg->n_trunks) {
4386 trunks = bitmap_allocate(4096);
4388 for (i = 0; i < cfg->n_trunks; i++) {
4389 int trunk = cfg->trunks[i];
4391 bitmap_set1(trunks, trunk);
4397 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4398 port->name, cfg->n_trunks);
4400 if (n_errors == cfg->n_trunks) {
4401 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4403 bitmap_free(trunks);
4406 } else if (vlan >= 0 && cfg->n_trunks) {
4407 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4411 ? port->trunks != NULL
4412 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4413 bridge_flush(port->bridge);
4415 bitmap_free(port->trunks);
4416 port->trunks = trunks;
4420 port_destroy(struct port *port)
4423 struct bridge *br = port->bridge;
4427 proc_net_compat_update_vlan(port->name, NULL, 0);
4428 proc_net_compat_update_bond(port->name, NULL);
4430 for (i = 0; i < MAX_MIRRORS; i++) {
4431 struct mirror *m = br->mirrors[i];
4432 if (m && m->out_port == port) {
4437 while (port->n_ifaces > 0) {
4438 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4441 shash_find_and_delete_assert(&br->port_by_name, port->name);
4443 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4444 del->port_idx = port->port_idx;
4446 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4448 netdev_monitor_destroy(port->monitor);
4450 bitmap_free(port->trunks);
4457 static struct port *
4458 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4460 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4461 return iface ? iface->port : NULL;
4464 static struct port *
4465 port_lookup(const struct bridge *br, const char *name)
4467 return shash_find_data(&br->port_by_name, name);
4470 static struct iface *
4471 port_lookup_iface(const struct port *port, const char *name)
4473 struct iface *iface = iface_lookup(port->bridge, name);
4474 return iface && iface->port == port ? iface : NULL;
4478 port_update_lacp(struct port *port)
4483 if (!port->lacp || port->n_ifaces < 1) {
4484 for (i = 0; i < port->n_ifaces; i++) {
4485 iface_set_lacp_defaulted(port->ifaces[i]);
4491 for (i = 0; i < port->n_ifaces; i++) {
4492 struct iface *iface = port->ifaces[i];
4494 if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) {
4499 if (iface->dp_ifidx == port->lacp_key) {
4500 key_changed = false;
4505 port->lacp_key = port->ifaces[0]->dp_ifidx;
4508 for (i = 0; i < port->n_ifaces; i++) {
4509 struct iface *iface = port->ifaces[i];
4511 iface->lacp_actor.sys_priority = htons(port->lacp_priority);
4512 memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN);
4514 iface->lacp_actor.port_priority = htons(iface->lacp_priority);
4515 iface->lacp_actor.portid = htons(iface->dp_ifidx);
4516 iface->lacp_actor.key = htons(port->lacp_key);
4520 port->lacp_need_update = true;
4524 port_update_bonding(struct port *port)
4526 if (port->monitor) {
4527 netdev_monitor_destroy(port->monitor);
4528 port->monitor = NULL;
4530 if (port->n_ifaces < 2) {
4531 /* Not a bonded port. */
4532 if (port->bond_hash) {
4533 free(port->bond_hash);
4534 port->bond_hash = NULL;
4535 port->bond_compat_is_stale = true;
4538 port->bond_fake_iface = false;
4542 if (port->bond_mode != BM_AB && !port->bond_hash) {
4543 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4544 for (i = 0; i <= BOND_MASK; i++) {
4545 struct bond_entry *e = &port->bond_hash[i];
4549 port->no_ifaces_tag = tag_create_random();
4550 bond_choose_active_iface(port);
4551 port->bond_next_rebalance
4552 = time_msec() + port->bond_rebalance_interval;
4554 if (port->cfg->bond_fake_iface) {
4555 port->bond_next_fake_iface_update = time_msec();
4557 } else if (port->bond_mode == BM_AB) {
4558 free(port->bond_hash);
4559 port->bond_hash = NULL;
4561 port->bond_compat_is_stale = true;
4562 port->bond_fake_iface = port->cfg->bond_fake_iface;
4564 if (!port->miimon) {
4565 port->monitor = netdev_monitor_create();
4566 for (i = 0; i < port->n_ifaces; i++) {
4567 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4574 port_update_bond_compat(struct port *port)
4576 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
4577 struct compat_bond bond;
4580 if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) {
4581 proc_net_compat_update_bond(port->name, NULL);
4586 bond.updelay = port->updelay;
4587 bond.downdelay = port->downdelay;
4590 bond.hashes = compat_hashes;
4591 if (port->bond_hash) {
4592 const struct bond_entry *e;
4593 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4594 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
4595 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
4596 cbh->hash = e - port->bond_hash;
4597 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
4602 bond.n_slaves = port->n_ifaces;
4603 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
4604 for (i = 0; i < port->n_ifaces; i++) {
4605 struct iface *iface = port->ifaces[i];
4606 struct compat_bond_slave *slave = &bond.slaves[i];
4607 slave->name = iface->name;
4609 /* We need to make the same determination as the Linux bonding
4610 * code to determine whether a slave should be consider "up".
4611 * The Linux function bond_miimon_inspect() supports four
4612 * BOND_LINK_* states:
4614 * - BOND_LINK_UP: carrier detected, updelay has passed.
4615 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
4616 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
4617 * - BOND_LINK_BACK: carrier detected, updelay in progress.
4619 * The function bond_info_show_slave() only considers BOND_LINK_UP
4620 * to be "up" and anything else to be "down".
4622 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
4626 netdev_get_etheraddr(iface->netdev, slave->mac);
4629 if (port->bond_fake_iface) {
4630 struct netdev *bond_netdev;
4632 if (!netdev_open_default(port->name, &bond_netdev)) {
4634 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
4636 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
4638 netdev_close(bond_netdev);
4642 proc_net_compat_update_bond(port->name, &bond);
4647 port_update_vlan_compat(struct port *port)
4649 struct bridge *br = port->bridge;
4650 char *vlandev_name = NULL;
4652 if (port->vlan > 0) {
4653 /* Figure out the name that the VLAN device should actually have, if it
4654 * existed. This takes some work because the VLAN device would not
4655 * have port->name in its name; rather, it would have the trunk port's
4656 * name, and 'port' would be attached to a bridge that also had the
4657 * VLAN device one of its ports. So we need to find a trunk port that
4658 * includes port->vlan.
4660 * There might be more than one candidate. This doesn't happen on
4661 * XenServer, so if it happens we just pick the first choice in
4662 * alphabetical order instead of creating multiple VLAN devices. */
4664 for (i = 0; i < br->n_ports; i++) {
4665 struct port *p = br->ports[i];
4666 if (port_trunks_vlan(p, port->vlan)
4668 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
4670 uint8_t ea[ETH_ADDR_LEN];
4671 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
4672 if (!eth_addr_is_multicast(ea) &&
4673 !eth_addr_is_reserved(ea) &&
4674 !eth_addr_is_zero(ea)) {
4675 vlandev_name = p->name;
4680 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
4683 /* Interface functions. */
4686 iface_set_lacp_defaulted(struct iface *iface)
4688 memset(&iface->lacp_partner, 0, sizeof iface->lacp_partner);
4690 iface->lacp_status = LACP_DEFAULTED;
4692 iface->port->lacp_need_update = true;
4696 iface_set_lacp_expired(struct iface *iface)
4698 iface->lacp_status &= ~LACP_CURRENT;
4699 iface->lacp_status |= LACP_EXPIRED;
4700 iface->lacp_partner.state |= LACP_STATE_TIME;
4701 iface->lacp_partner.state &= ~LACP_STATE_SYNC;
4703 iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX;
4708 iface_get_lacp_state(const struct iface *iface)
4712 if (iface->port->lacp & LACP_ACTIVE) {
4713 state |= LACP_STATE_ACT;
4716 if (iface->lacp_status & LACP_ATTACHED) {
4717 state |= LACP_STATE_SYNC;
4720 if (iface->lacp_status & LACP_DEFAULTED) {
4721 state |= LACP_STATE_DEF;
4724 if (iface->lacp_status & LACP_EXPIRED) {
4725 state |= LACP_STATE_EXP;
4728 if (iface->port->n_ifaces > 1) {
4729 state |= LACP_STATE_AGG;
4732 if (iface->enabled) {
4733 state |= LACP_STATE_COL | LACP_STATE_DIST;
4739 /* Given 'iface', populates 'priority' with data representing its LACP link
4740 * priority. If two priority objects populated by this function are compared
4741 * using memcmp, the higher priority link will be less than the lower priority
4744 iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority)
4746 uint16_t partner_priority, actor_priority;
4748 /* Choose the lacp_info of the higher priority system by comparing their
4749 * system priorities and mac addresses. */
4750 actor_priority = ntohs(iface->lacp_actor.sys_priority);
4751 partner_priority = ntohs(iface->lacp_partner.sys_priority);
4752 if (actor_priority < partner_priority) {
4753 *priority = iface->lacp_actor;
4754 } else if (partner_priority < actor_priority) {
4755 *priority = iface->lacp_partner;
4756 } else if (eth_addr_compare_3way(iface->lacp_actor.sysid,
4757 iface->lacp_partner.sysid) < 0) {
4758 *priority = iface->lacp_actor;
4760 *priority = iface->lacp_partner;
4763 /* Key and state are not used in priority comparisons. */
4765 priority->state = 0;
4769 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4772 union ofp_action action;
4774 memset(&action, 0, sizeof action);
4775 action.output.type = htons(OFPAT_OUTPUT);
4776 action.output.len = htons(sizeof action);
4777 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4779 flow_extract(packet, 0, ODPP_NONE, &flow);
4781 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4783 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4784 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4788 static struct iface *
4789 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4791 struct bridge *br = port->bridge;
4792 struct iface *iface;
4793 char *name = if_cfg->name;
4795 iface = xzalloc(sizeof *iface);
4797 iface->port_ifidx = port->n_ifaces;
4798 iface->name = xstrdup(name);
4799 iface->dp_ifidx = -1;
4800 iface->tag = tag_create_random();
4801 iface->delay_expires = LLONG_MAX;
4802 iface->netdev = NULL;
4803 iface->cfg = if_cfg;
4804 iface_set_lacp_defaulted(iface);
4806 if (port->lacp & LACP_ACTIVE) {
4807 iface_set_lacp_expired(iface);
4810 shash_add_assert(&br->iface_by_name, iface->name, iface);
4812 if (port->n_ifaces >= port->allocated_ifaces) {
4813 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4814 sizeof *port->ifaces);
4816 port->ifaces[port->n_ifaces++] = iface;
4817 if (port->n_ifaces > 1) {
4818 br->has_bonded_ports = true;
4821 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4829 iface_destroy(struct iface *iface)
4832 struct port *port = iface->port;
4833 struct bridge *br = port->bridge;
4834 bool del_active = port->active_iface == iface->port_ifidx;
4837 if (port->monitor) {
4838 netdev_monitor_remove(port->monitor, iface->netdev);
4841 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4843 if (iface->dp_ifidx >= 0) {
4844 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4847 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4848 del->port_ifidx = iface->port_ifidx;
4850 netdev_close(iface->netdev);
4853 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4854 bond_choose_active_iface(port);
4855 bond_send_learning_packets(port);
4858 cfm_destroy(iface->cfm);
4863 bridge_flush(port->bridge);
4867 static struct iface *
4868 iface_lookup(const struct bridge *br, const char *name)
4870 return shash_find_data(&br->iface_by_name, name);
4873 static struct iface *
4874 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4876 struct iface *iface;
4878 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4879 hash_int(dp_ifidx, 0), &br->ifaces) {
4880 if (iface->dp_ifidx == dp_ifidx) {
4887 /* Set Ethernet address of 'iface', if one is specified in the configuration
4890 iface_set_mac(struct iface *iface)
4892 uint8_t ea[ETH_ADDR_LEN];
4894 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4895 if (eth_addr_is_multicast(ea)) {
4896 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4898 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4899 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4900 iface->name, iface->name);
4902 int error = netdev_set_etheraddr(iface->netdev, ea);
4904 VLOG_ERR("interface %s: setting MAC failed (%s)",
4905 iface->name, strerror(error));
4911 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4913 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4916 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4920 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4922 * The value strings in '*shash' are taken directly from values[], not copied,
4923 * so the caller should not modify or free them. */
4925 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4926 struct shash *shash)
4931 for (i = 0; i < n; i++) {
4932 shash_add(shash, keys[i], values[i]);
4936 /* Creates 'keys' and 'values' arrays from 'shash'.
4938 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4939 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4940 * are populated with with strings taken directly from 'shash' and thus have
4941 * the same ownership of the key-value pairs in shash.
4944 shash_to_ovs_idl_map(struct shash *shash,
4945 char ***keys, char ***values, size_t *n)
4949 struct shash_node *sn;
4951 count = shash_count(shash);
4953 k = xmalloc(count * sizeof *k);
4954 v = xmalloc(count * sizeof *v);
4957 SHASH_FOR_EACH(sn, shash) {
4968 struct iface_delete_queues_cbdata {
4969 struct netdev *netdev;
4970 const struct ovsdb_datum *queues;
4974 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4976 union ovsdb_atom atom;
4978 atom.integer = target;
4979 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4983 iface_delete_queues(unsigned int queue_id,
4984 const struct shash *details OVS_UNUSED, void *cbdata_)
4986 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4988 if (!queue_ids_include(cbdata->queues, queue_id)) {
4989 netdev_delete_queue(cbdata->netdev, queue_id);
4994 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4996 if (!qos || qos->type[0] == '\0') {
4997 netdev_set_qos(iface->netdev, NULL, NULL);
4999 struct iface_delete_queues_cbdata cbdata;
5000 struct shash details;
5003 /* Configure top-level Qos for 'iface'. */
5004 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
5005 qos->n_other_config, &details);
5006 netdev_set_qos(iface->netdev, qos->type, &details);
5007 shash_destroy(&details);
5009 /* Deconfigure queues that were deleted. */
5010 cbdata.netdev = iface->netdev;
5011 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
5013 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
5015 /* Configure queues for 'iface'. */
5016 for (i = 0; i < qos->n_queues; i++) {
5017 const struct ovsrec_queue *queue = qos->value_queues[i];
5018 unsigned int queue_id = qos->key_queues[i];
5020 shash_from_ovs_idl_map(queue->key_other_config,
5021 queue->value_other_config,
5022 queue->n_other_config, &details);
5023 netdev_set_queue(iface->netdev, queue_id, &details);
5024 shash_destroy(&details);
5030 iface_update_cfm(struct iface *iface)
5034 uint16_t *remote_mps;
5035 struct ovsrec_monitor *mon;
5036 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
5038 mon = iface->cfg->monitor;
5041 cfm_destroy(iface->cfm);
5046 if (netdev_get_etheraddr(iface->netdev, ea)) {
5047 VLOG_WARN("interface %s: Failed to get ethernet address. "
5048 "Skipping Monitor.", iface->name);
5052 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
5053 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
5058 iface->cfm = cfm_create();
5062 cfm->mpid = mon->mpid;
5063 cfm->interval = mon->interval ? *mon->interval : 1000;
5065 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
5066 memcpy(cfm->maid, maid, sizeof cfm->maid);
5068 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
5069 for(i = 0; i < mon->n_remote_mps; i++) {
5070 remote_mps[i] = mon->remote_mps[i]->mpid;
5072 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
5075 if (!cfm_configure(iface->cfm)) {
5076 cfm_destroy(iface->cfm);
5081 /* Port mirroring. */
5083 static struct mirror *
5084 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
5088 for (i = 0; i < MAX_MIRRORS; i++) {
5089 struct mirror *m = br->mirrors[i];
5090 if (m && uuid_equals(uuid, &m->uuid)) {
5098 mirror_reconfigure(struct bridge *br)
5100 unsigned long *rspan_vlans;
5103 /* Get rid of deleted mirrors. */
5104 for (i = 0; i < MAX_MIRRORS; i++) {
5105 struct mirror *m = br->mirrors[i];
5107 const struct ovsdb_datum *mc;
5108 union ovsdb_atom atom;
5110 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
5111 atom.uuid = br->mirrors[i]->uuid;
5112 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
5118 /* Add new mirrors and reconfigure existing ones. */
5119 for (i = 0; i < br->cfg->n_mirrors; i++) {
5120 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
5121 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
5123 mirror_reconfigure_one(m, cfg);
5125 mirror_create(br, cfg);
5129 /* Update port reserved status. */
5130 for (i = 0; i < br->n_ports; i++) {
5131 br->ports[i]->is_mirror_output_port = false;
5133 for (i = 0; i < MAX_MIRRORS; i++) {
5134 struct mirror *m = br->mirrors[i];
5135 if (m && m->out_port) {
5136 m->out_port->is_mirror_output_port = true;
5140 /* Update flooded vlans (for RSPAN). */
5142 if (br->cfg->n_flood_vlans) {
5143 rspan_vlans = bitmap_allocate(4096);
5145 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
5146 int64_t vlan = br->cfg->flood_vlans[i];
5147 if (vlan >= 0 && vlan < 4096) {
5148 bitmap_set1(rspan_vlans, vlan);
5149 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
5152 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
5157 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
5163 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
5168 for (i = 0; ; i++) {
5169 if (i >= MAX_MIRRORS) {
5170 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
5171 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
5174 if (!br->mirrors[i]) {
5179 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
5182 br->mirrors[i] = m = xzalloc(sizeof *m);
5185 m->name = xstrdup(cfg->name);
5186 shash_init(&m->src_ports);
5187 shash_init(&m->dst_ports);
5193 mirror_reconfigure_one(m, cfg);
5197 mirror_destroy(struct mirror *m)
5200 struct bridge *br = m->bridge;
5203 for (i = 0; i < br->n_ports; i++) {
5204 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5205 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5208 shash_destroy(&m->src_ports);
5209 shash_destroy(&m->dst_ports);
5212 m->bridge->mirrors[m->idx] = NULL;
5221 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
5222 struct shash *names)
5226 for (i = 0; i < n_ports; i++) {
5227 const char *name = ports[i]->name;
5228 if (port_lookup(m->bridge, name)) {
5229 shash_add_once(names, name, NULL);
5231 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
5232 "port %s", m->bridge->name, m->name, name);
5238 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
5244 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
5246 for (i = 0; i < cfg->n_select_vlan; i++) {
5247 int64_t vlan = cfg->select_vlan[i];
5248 if (vlan < 0 || vlan > 4095) {
5249 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
5250 m->bridge->name, m->name, vlan);
5252 (*vlans)[n_vlans++] = vlan;
5259 vlan_is_mirrored(const struct mirror *m, int vlan)
5263 for (i = 0; i < m->n_vlans; i++) {
5264 if (m->vlans[i] == vlan) {
5272 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
5276 for (i = 0; i < m->n_vlans; i++) {
5277 if (port_trunks_vlan(p, m->vlans[i])) {
5285 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
5287 struct shash src_ports, dst_ports;
5288 mirror_mask_t mirror_bit;
5289 struct port *out_port;
5296 if (strcmp(cfg->name, m->name)) {
5298 m->name = xstrdup(cfg->name);
5301 /* Get output port. */
5302 if (cfg->output_port) {
5303 out_port = port_lookup(m->bridge, cfg->output_port->name);
5305 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
5306 m->bridge->name, m->name);
5312 if (cfg->output_vlan) {
5313 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
5314 "output vlan; ignoring output vlan",
5315 m->bridge->name, m->name);
5317 } else if (cfg->output_vlan) {
5319 out_vlan = *cfg->output_vlan;
5321 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
5322 m->bridge->name, m->name);
5327 shash_init(&src_ports);
5328 shash_init(&dst_ports);
5329 if (cfg->select_all) {
5330 for (i = 0; i < m->bridge->n_ports; i++) {
5331 const char *name = m->bridge->ports[i]->name;
5332 shash_add_once(&src_ports, name, NULL);
5333 shash_add_once(&dst_ports, name, NULL);
5338 /* Get ports, and drop duplicates and ports that don't exist. */
5339 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
5341 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
5344 /* Get all the vlans, and drop duplicate and invalid vlans. */
5345 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
5348 /* Update mirror data. */
5349 if (!shash_equal_keys(&m->src_ports, &src_ports)
5350 || !shash_equal_keys(&m->dst_ports, &dst_ports)
5351 || m->n_vlans != n_vlans
5352 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
5353 || m->out_port != out_port
5354 || m->out_vlan != out_vlan) {
5355 bridge_flush(m->bridge);
5357 shash_swap(&m->src_ports, &src_ports);
5358 shash_swap(&m->dst_ports, &dst_ports);
5361 m->n_vlans = n_vlans;
5362 m->out_port = out_port;
5363 m->out_vlan = out_vlan;
5366 mirror_bit = MIRROR_MASK_C(1) << m->idx;
5367 for (i = 0; i < m->bridge->n_ports; i++) {
5368 struct port *port = m->bridge->ports[i];
5370 if (shash_find(&m->src_ports, port->name)
5373 ? port_trunks_any_mirrored_vlan(m, port)
5374 : vlan_is_mirrored(m, port->vlan)))) {
5375 port->src_mirrors |= mirror_bit;
5377 port->src_mirrors &= ~mirror_bit;
5380 if (shash_find(&m->dst_ports, port->name)) {
5381 port->dst_mirrors |= mirror_bit;
5383 port->dst_mirrors &= ~mirror_bit;
5388 shash_destroy(&src_ports);
5389 shash_destroy(&dst_ports);