1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
40 #include "dynamic-string.h"
46 #include "mac-learning.h"
50 #include "ofp-print.h"
52 #include "ofproto/netflow.h"
53 #include "ofproto/ofproto.h"
54 #include "ovsdb-data.h"
56 #include "poll-loop.h"
57 #include "proc-net-compat.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_reconfigure);
81 LACP_STATUS_CURRENT, /* Partner is up to date. */
82 LACP_STATUS_EXPIRED, /* Partner is out of date. Attempt to re-sync. */
83 LACP_STATUS_DEFAULTED /* Partner information is unknown. */
92 struct dst builtin[32];
97 static void dst_set_init(struct dst_set *);
98 static void dst_set_add(struct dst_set *, const struct dst *);
99 static void dst_set_free(struct dst_set *);
102 /* These members are always valid. */
103 struct port *port; /* Containing port. */
104 size_t port_ifidx; /* Index within containing port. */
105 char *name; /* Host network device name. */
106 tag_type tag; /* Tag associated with this interface. */
107 long long delay_expires; /* Time after which 'enabled' may change. */
109 /* These members are valid only after bridge_reconfigure() causes them to
111 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
112 int dp_ifidx; /* Index within kernel datapath. */
113 struct netdev *netdev; /* Network device. */
114 bool enabled; /* May be chosen for flows? */
115 bool up; /* Is the interface up? */
116 const char *type; /* Usually same as cfg->type. */
117 struct cfm *cfm; /* Connectivity Fault Management */
118 const struct ovsrec_interface *cfg;
120 /* LACP information. */
121 enum lacp_status lacp_status; /* LACP state machine status. */
122 uint16_t lacp_priority; /* LACP port priority. */
123 struct lacp_info lacp_actor; /* LACP actor information. */
124 struct lacp_info lacp_partner; /* LACP partner information. */
125 long long int lacp_tx; /* Next LACP message transmission time. */
126 long long int lacp_rx; /* Next LACP message receive time. */
127 bool lacp_attached; /* Attached to its aggregator? LACP allows
128 this link to be chosen for flows. */
131 #define BOND_MASK 0xff
133 int iface_idx; /* Index of assigned iface, or -1 if none. */
134 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
135 tag_type iface_tag; /* Tag associated with iface_idx. */
139 BM_SLB, /* Source Load Balance (Default). */
140 BM_AB /* Active Backup. */
143 #define MAX_MIRRORS 32
144 typedef uint32_t mirror_mask_t;
145 #define MIRROR_MASK_C(X) UINT32_C(X)
146 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
148 struct bridge *bridge;
151 struct uuid uuid; /* UUID of this "mirror" record in database. */
153 /* Selection criteria. */
154 struct shash src_ports; /* Name is port name; data is always NULL. */
155 struct shash dst_ports; /* Name is port name; data is always NULL. */
160 struct port *out_port;
164 /* Flags for a port's lacp member. */
165 #define LACP_ACTIVE 0x01 /* LACP is in active mode. */
166 #define LACP_PASSIVE 0x02 /* LACP is in passive mode. */
167 #define LACP_NEGOTIATED 0x04 /* LACP has successfully negotiated. */
169 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
171 struct bridge *bridge;
173 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
174 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
175 * NULL if all VLANs are trunked. */
176 const struct ovsrec_port *cfg;
179 /* An ordinary bridge port has 1 interface.
180 * A bridge port for bonding has at least 2 interfaces. */
181 struct iface **ifaces;
182 size_t n_ifaces, allocated_ifaces;
185 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
186 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
187 tag_type active_iface_tag; /* Tag for bcast flows. */
188 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
189 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
190 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
191 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
192 bool miimon; /* Use miimon instead of carrier? */
193 long long int bond_miimon_interval; /* Miimon status refresh interval. */
194 long long int bond_miimon_next_update; /* Time of next miimon update. */
195 long long int bond_next_fake_iface_update; /* Time of next update. */
196 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
198 /* LACP information. */
199 int lacp; /* LACP status flags. 0 if LACP is off. */
200 uint16_t lacp_key; /* LACP aggregation key. */
201 uint16_t lacp_priority; /* LACP system priority. */
202 bool lacp_need_update; /* Need to update attached interfaces? */
204 /* SLB specific bonding info. */
205 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
206 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
207 long long int bond_next_rebalance; /* Next rebalancing time. */
209 /* Port mirroring info. */
210 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
211 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
212 bool is_mirror_output_port; /* Does port mirroring send frames here? */
216 struct list node; /* Node in global list of bridges. */
217 char *name; /* User-specified arbitrary name. */
218 struct mac_learning *ml; /* MAC learning table. */
219 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
220 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
221 const struct ovsrec_bridge *cfg;
223 /* OpenFlow switch processing. */
224 struct ofproto *ofproto; /* OpenFlow switch. */
226 /* Kernel datapath information. */
227 struct dpif *dpif; /* Datapath. */
228 struct hmap ifaces; /* Contains "struct iface"s. */
232 size_t n_ports, allocated_ports;
233 struct shash iface_by_name; /* "struct iface"s indexed by name. */
234 struct shash port_by_name; /* "struct port"s indexed by name. */
237 bool has_bonded_ports;
242 /* Port mirroring. */
243 struct mirror *mirrors[MAX_MIRRORS];
246 /* List of all bridges. */
247 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
249 /* OVSDB IDL used to obtain configuration. */
250 static struct ovsdb_idl *idl;
252 /* Each time this timer expires, the bridge fetches systems and interface
253 * statistics and pushes them into the database. */
254 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
255 static long long int stats_timer = LLONG_MIN;
257 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
258 static void bridge_destroy(struct bridge *);
259 static struct bridge *bridge_lookup(const char *name);
260 static unixctl_cb_func bridge_unixctl_dump_flows;
261 static unixctl_cb_func bridge_unixctl_reconnect;
262 static int bridge_run_one(struct bridge *);
263 static size_t bridge_get_controllers(const struct bridge *br,
264 struct ovsrec_controller ***controllersp);
265 static void bridge_reconfigure_one(struct bridge *);
266 static void bridge_reconfigure_remotes(struct bridge *,
267 const struct sockaddr_in *managers,
269 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
270 static void bridge_fetch_dp_ifaces(struct bridge *);
271 static void bridge_flush(struct bridge *);
272 static void bridge_pick_local_hw_addr(struct bridge *,
273 uint8_t ea[ETH_ADDR_LEN],
274 struct iface **hw_addr_iface);
275 static uint64_t bridge_pick_datapath_id(struct bridge *,
276 const uint8_t bridge_ea[ETH_ADDR_LEN],
277 struct iface *hw_addr_iface);
278 static struct iface *bridge_get_local_iface(struct bridge *);
279 static uint64_t dpid_from_hash(const void *, size_t nbytes);
281 static unixctl_cb_func bridge_unixctl_fdb_show;
283 static void lacp_run(struct bridge *);
284 static void lacp_wait(struct bridge *);
285 static void lacp_process_packet(const struct ofpbuf *, struct iface *);
287 static void bond_init(void);
288 static void bond_run(struct bridge *);
289 static void bond_wait(struct bridge *);
290 static void bond_rebalance_port(struct port *);
291 static void bond_send_learning_packets(struct port *);
292 static void bond_enable_slave(struct iface *iface, bool enable);
294 static struct port *port_create(struct bridge *, const char *name);
295 static void port_reconfigure(struct port *, const struct ovsrec_port *);
296 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
297 static void port_destroy(struct port *);
298 static struct port *port_lookup(const struct bridge *, const char *name);
299 static struct iface *port_lookup_iface(const struct port *, const char *name);
300 static struct port *port_from_dp_ifidx(const struct bridge *,
302 static void port_update_bond_compat(struct port *);
303 static void port_update_vlan_compat(struct port *);
304 static void port_update_bonding(struct port *);
305 static void port_update_lacp(struct port *);
307 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
308 static void mirror_destroy(struct mirror *);
309 static void mirror_reconfigure(struct bridge *);
310 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
311 static bool vlan_is_mirrored(const struct mirror *, int vlan);
313 static struct iface *iface_create(struct port *port,
314 const struct ovsrec_interface *if_cfg);
315 static void iface_destroy(struct iface *);
316 static struct iface *iface_lookup(const struct bridge *, const char *name);
317 static struct iface *iface_from_dp_ifidx(const struct bridge *,
319 static void iface_set_mac(struct iface *);
320 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
321 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
322 static void iface_update_cfm(struct iface *);
323 static void iface_refresh_cfm_stats(struct iface *iface);
324 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
325 static uint8_t iface_get_lacp_state(const struct iface *);
326 static void iface_get_lacp_priority(struct iface *, struct lacp_info *);
327 static void iface_set_lacp_defaulted(struct iface *);
328 static void iface_set_lacp_expired(struct iface *);
330 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
332 static void shash_to_ovs_idl_map(struct shash *,
333 char ***keys, char ***values, size_t *n);
336 /* Hooks into ofproto processing. */
337 static struct ofhooks bridge_ofhooks;
339 /* Public functions. */
341 /* Initializes the bridge module, configuring it to obtain its configuration
342 * from an OVSDB server accessed over 'remote', which should be a string in a
343 * form acceptable to ovsdb_idl_create(). */
345 bridge_init(const char *remote)
347 /* Create connection to database. */
348 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
350 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
351 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
352 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
354 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
356 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
357 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
359 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
360 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
361 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
363 /* Register unixctl commands. */
364 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
365 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
367 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
375 struct bridge *br, *next_br;
377 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
380 ovsdb_idl_destroy(idl);
383 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
384 * but for which the ovs-vswitchd configuration 'cfg' is required. */
386 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
388 static bool already_configured_once;
389 struct svec bridge_names;
390 struct svec dpif_names, dpif_types;
393 /* Only do this once per ovs-vswitchd run. */
394 if (already_configured_once) {
397 already_configured_once = true;
399 stats_timer = time_msec() + STATS_INTERVAL;
401 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
402 svec_init(&bridge_names);
403 for (i = 0; i < cfg->n_bridges; i++) {
404 svec_add(&bridge_names, cfg->bridges[i]->name);
406 svec_sort(&bridge_names);
408 /* Iterate over all system dpifs and delete any of them that do not appear
410 svec_init(&dpif_names);
411 svec_init(&dpif_types);
412 dp_enumerate_types(&dpif_types);
413 for (i = 0; i < dpif_types.n; i++) {
416 dp_enumerate_names(dpif_types.names[i], &dpif_names);
418 /* Delete each dpif whose name is not in 'bridge_names'. */
419 for (j = 0; j < dpif_names.n; j++) {
420 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
424 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
433 svec_destroy(&bridge_names);
434 svec_destroy(&dpif_names);
435 svec_destroy(&dpif_types);
438 /* Callback for iterate_and_prune_ifaces(). */
440 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
442 if (!iface->netdev) {
443 /* We already reported a related error, don't bother duplicating it. */
447 if (iface->dp_ifidx < 0) {
448 VLOG_ERR("%s interface not in %s, dropping",
449 iface->name, dpif_name(br->dpif));
453 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
454 iface->name, iface->dp_ifidx);
458 /* Callback for iterate_and_prune_ifaces(). */
460 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
461 void *aux OVS_UNUSED)
463 /* Set policing attributes. */
464 netdev_set_policing(iface->netdev,
465 iface->cfg->ingress_policing_rate,
466 iface->cfg->ingress_policing_burst);
468 /* Set MAC address of internal interfaces other than the local
470 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
471 iface_set_mac(iface);
477 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
478 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
479 * deletes from 'br' any ports that no longer have any interfaces. */
481 iterate_and_prune_ifaces(struct bridge *br,
482 bool (*cb)(struct bridge *, struct iface *,
488 for (i = 0; i < br->n_ports; ) {
489 struct port *port = br->ports[i];
490 for (j = 0; j < port->n_ifaces; ) {
491 struct iface *iface = port->ifaces[j];
492 if (cb(br, iface, aux)) {
495 iface_set_ofport(iface->cfg, -1);
496 iface_destroy(iface);
500 if (port->n_ifaces) {
503 VLOG_ERR("%s port has no interfaces, dropping", port->name);
509 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
510 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
511 * responsible for freeing '*managersp' (with free()).
513 * You may be asking yourself "why does ovs-vswitchd care?", because
514 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
515 * should not be and in fact is not directly involved in that. But
516 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
517 * it has to tell in-band control where the managers are to enable that.
518 * (Thus, only managers connected in-band are collected.)
521 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
522 struct sockaddr_in **managersp, size_t *n_managersp)
524 struct sockaddr_in *managers = NULL;
525 size_t n_managers = 0;
526 struct shash targets;
529 /* Collect all of the potential targets, as the union of the "managers"
530 * column and the "targets" columns of the rows pointed to by
531 * "manager_options", excluding any that are out-of-band. */
532 shash_init(&targets);
533 for (i = 0; i < ovs_cfg->n_managers; i++) {
534 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
536 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
537 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
539 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
540 shash_find_and_delete(&targets, m->target);
542 shash_add_once(&targets, m->target, NULL);
546 /* Now extract the targets' IP addresses. */
547 if (!shash_is_empty(&targets)) {
548 struct shash_node *node;
550 managers = xmalloc(shash_count(&targets) * sizeof *managers);
551 SHASH_FOR_EACH (node, &targets) {
552 const char *target = node->name;
553 struct sockaddr_in *sin = &managers[n_managers];
555 if ((!strncmp(target, "tcp:", 4)
556 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
557 (!strncmp(target, "ssl:", 4)
558 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
563 shash_destroy(&targets);
565 *managersp = managers;
566 *n_managersp = n_managers;
570 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
572 struct shash old_br, new_br;
573 struct shash_node *node;
574 struct bridge *br, *next;
575 struct sockaddr_in *managers;
578 int sflow_bridge_number;
580 COVERAGE_INC(bridge_reconfigure);
582 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
584 /* Collect old and new bridges. */
587 LIST_FOR_EACH (br, node, &all_bridges) {
588 shash_add(&old_br, br->name, br);
590 for (i = 0; i < ovs_cfg->n_bridges; i++) {
591 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
592 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
593 VLOG_WARN("more than one bridge named %s", br_cfg->name);
597 /* Get rid of deleted bridges and add new bridges. */
598 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
599 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
606 SHASH_FOR_EACH (node, &new_br) {
607 const char *br_name = node->name;
608 const struct ovsrec_bridge *br_cfg = node->data;
609 br = shash_find_data(&old_br, br_name);
611 /* If the bridge datapath type has changed, we need to tear it
612 * down and recreate. */
613 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
615 bridge_create(br_cfg);
618 bridge_create(br_cfg);
621 shash_destroy(&old_br);
622 shash_destroy(&new_br);
624 /* Reconfigure all bridges. */
625 LIST_FOR_EACH (br, node, &all_bridges) {
626 bridge_reconfigure_one(br);
629 /* Add and delete ports on all datapaths.
631 * The kernel will reject any attempt to add a given port to a datapath if
632 * that port already belongs to a different datapath, so we must do all
633 * port deletions before any port additions. */
634 LIST_FOR_EACH (br, node, &all_bridges) {
635 struct dpif_port_dump dump;
636 struct shash want_ifaces;
637 struct dpif_port dpif_port;
639 bridge_get_all_ifaces(br, &want_ifaces);
640 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
641 if (!shash_find(&want_ifaces, dpif_port.name)
642 && strcmp(dpif_port.name, br->name)) {
643 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
645 VLOG_ERR("failed to remove %s interface from %s: %s",
646 dpif_port.name, dpif_name(br->dpif),
651 shash_destroy(&want_ifaces);
653 LIST_FOR_EACH (br, node, &all_bridges) {
654 struct shash cur_ifaces, want_ifaces;
655 struct dpif_port_dump dump;
656 struct dpif_port dpif_port;
658 /* Get the set of interfaces currently in this datapath. */
659 shash_init(&cur_ifaces);
660 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
661 struct dpif_port *port_info = xmalloc(sizeof *port_info);
662 dpif_port_clone(port_info, &dpif_port);
663 shash_add(&cur_ifaces, dpif_port.name, port_info);
666 /* Get the set of interfaces we want on this datapath. */
667 bridge_get_all_ifaces(br, &want_ifaces);
669 hmap_clear(&br->ifaces);
670 SHASH_FOR_EACH (node, &want_ifaces) {
671 const char *if_name = node->name;
672 struct iface *iface = node->data;
673 struct dpif_port *dpif_port;
677 type = iface ? iface->type : "internal";
678 dpif_port = shash_find_data(&cur_ifaces, if_name);
680 /* If we have a port or a netdev already, and it's not the type we
681 * want, then delete the port (if any) and close the netdev (if
683 if ((dpif_port && strcmp(dpif_port->type, type))
684 || (iface && iface->netdev
685 && strcmp(type, netdev_get_type(iface->netdev)))) {
687 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
694 netdev_close(iface->netdev);
695 iface->netdev = NULL;
699 /* If the port doesn't exist or we don't have the netdev open,
700 * we need to do more work. */
701 if (!dpif_port || (iface && !iface->netdev)) {
702 struct netdev_options options;
703 struct netdev *netdev;
706 /* First open the network device. */
707 options.name = if_name;
709 options.args = &args;
710 options.ethertype = NETDEV_ETH_TYPE_NONE;
714 shash_from_ovs_idl_map(iface->cfg->key_options,
715 iface->cfg->value_options,
716 iface->cfg->n_options, &args);
718 error = netdev_open(&options, &netdev);
719 shash_destroy(&args);
722 VLOG_WARN("could not open network device %s (%s)",
723 if_name, strerror(error));
727 /* Then add the port if we haven't already. */
729 error = dpif_port_add(br->dpif, netdev, NULL);
731 netdev_close(netdev);
732 if (error == EFBIG) {
733 VLOG_ERR("ran out of valid port numbers on %s",
734 dpif_name(br->dpif));
737 VLOG_ERR("failed to add %s interface to %s: %s",
738 if_name, dpif_name(br->dpif),
745 /* Update 'iface'. */
747 iface->netdev = netdev;
748 iface->enabled = netdev_get_carrier(iface->netdev);
749 iface->up = iface->enabled;
751 } else if (iface && iface->netdev) {
755 shash_from_ovs_idl_map(iface->cfg->key_options,
756 iface->cfg->value_options,
757 iface->cfg->n_options, &args);
758 netdev_set_config(iface->netdev, &args);
759 shash_destroy(&args);
762 shash_destroy(&want_ifaces);
764 SHASH_FOR_EACH (node, &cur_ifaces) {
765 struct dpif_port *port_info = node->data;
766 dpif_port_destroy(port_info);
769 shash_destroy(&cur_ifaces);
771 sflow_bridge_number = 0;
772 LIST_FOR_EACH (br, node, &all_bridges) {
775 struct iface *local_iface;
776 struct iface *hw_addr_iface;
779 bridge_fetch_dp_ifaces(br);
781 iterate_and_prune_ifaces(br, check_iface, NULL);
783 /* Pick local port hardware address, datapath ID. */
784 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
785 local_iface = bridge_get_local_iface(br);
787 int error = netdev_set_etheraddr(local_iface->netdev, ea);
789 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
790 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
791 "Ethernet address: %s",
792 br->name, strerror(error));
795 memcpy(br->ea, ea, ETH_ADDR_LEN);
797 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
798 ofproto_set_datapath_id(br->ofproto, dpid);
800 dpid_string = xasprintf("%016"PRIx64, dpid);
801 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
804 /* Set NetFlow configuration on this bridge. */
805 if (br->cfg->netflow) {
806 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
807 struct netflow_options opts;
809 memset(&opts, 0, sizeof opts);
811 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
812 if (nf_cfg->engine_type) {
813 opts.engine_type = *nf_cfg->engine_type;
815 if (nf_cfg->engine_id) {
816 opts.engine_id = *nf_cfg->engine_id;
819 opts.active_timeout = nf_cfg->active_timeout;
820 if (!opts.active_timeout) {
821 opts.active_timeout = -1;
822 } else if (opts.active_timeout < 0) {
823 VLOG_WARN("bridge %s: active timeout interval set to negative "
824 "value, using default instead (%d seconds)", br->name,
825 NF_ACTIVE_TIMEOUT_DEFAULT);
826 opts.active_timeout = -1;
829 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
830 if (opts.add_id_to_iface) {
831 if (opts.engine_id > 0x7f) {
832 VLOG_WARN("bridge %s: netflow port mangling may conflict "
833 "with another vswitch, choose an engine id less "
834 "than 128", br->name);
836 if (br->n_ports > 508) {
837 VLOG_WARN("bridge %s: netflow port mangling will conflict "
838 "with another port when more than 508 ports are "
843 opts.collectors.n = nf_cfg->n_targets;
844 opts.collectors.names = nf_cfg->targets;
845 if (ofproto_set_netflow(br->ofproto, &opts)) {
846 VLOG_ERR("bridge %s: problem setting netflow collectors",
850 ofproto_set_netflow(br->ofproto, NULL);
853 /* Set sFlow configuration on this bridge. */
854 if (br->cfg->sflow) {
855 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
856 struct ovsrec_controller **controllers;
857 struct ofproto_sflow_options oso;
858 size_t n_controllers;
860 memset(&oso, 0, sizeof oso);
862 oso.targets.n = sflow_cfg->n_targets;
863 oso.targets.names = sflow_cfg->targets;
865 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
866 if (sflow_cfg->sampling) {
867 oso.sampling_rate = *sflow_cfg->sampling;
870 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
871 if (sflow_cfg->polling) {
872 oso.polling_interval = *sflow_cfg->polling;
875 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
876 if (sflow_cfg->header) {
877 oso.header_len = *sflow_cfg->header;
880 oso.sub_id = sflow_bridge_number++;
881 oso.agent_device = sflow_cfg->agent;
883 oso.control_ip = NULL;
884 n_controllers = bridge_get_controllers(br, &controllers);
885 for (i = 0; i < n_controllers; i++) {
886 if (controllers[i]->local_ip) {
887 oso.control_ip = controllers[i]->local_ip;
891 ofproto_set_sflow(br->ofproto, &oso);
893 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
895 ofproto_set_sflow(br->ofproto, NULL);
898 /* Update the controller and related settings. It would be more
899 * straightforward to call this from bridge_reconfigure_one(), but we
900 * can't do it there for two reasons. First, and most importantly, at
901 * that point we don't know the dp_ifidx of any interfaces that have
902 * been added to the bridge (because we haven't actually added them to
903 * the datapath). Second, at that point we haven't set the datapath ID
904 * yet; when a controller is configured, resetting the datapath ID will
905 * immediately disconnect from the controller, so it's better to set
906 * the datapath ID before the controller. */
907 bridge_reconfigure_remotes(br, managers, n_managers);
909 LIST_FOR_EACH (br, node, &all_bridges) {
910 for (i = 0; i < br->n_ports; i++) {
911 struct port *port = br->ports[i];
914 port_update_vlan_compat(port);
915 port_update_bonding(port);
916 port_update_lacp(port);
918 for (j = 0; j < port->n_ifaces; j++) {
919 iface_update_qos(port->ifaces[j], port->cfg->qos);
923 LIST_FOR_EACH (br, node, &all_bridges) {
924 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
927 LIST_FOR_EACH (br, node, &all_bridges) {
929 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
930 iface_update_cfm(iface);
938 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
939 const struct ovsdb_idl_column *column,
942 const struct ovsdb_datum *datum;
943 union ovsdb_atom atom;
946 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
947 atom.string = (char *) key;
948 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
949 return idx == UINT_MAX ? NULL : datum->values[idx].string;
953 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
955 return get_ovsrec_key_value(&br_cfg->header_,
956 &ovsrec_bridge_col_other_config, key);
960 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
961 struct iface **hw_addr_iface)
967 *hw_addr_iface = NULL;
969 /* Did the user request a particular MAC? */
970 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
971 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
972 if (eth_addr_is_multicast(ea)) {
973 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
974 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
975 } else if (eth_addr_is_zero(ea)) {
976 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
982 /* Otherwise choose the minimum non-local MAC address among all of the
984 memset(ea, 0xff, sizeof ea);
985 for (i = 0; i < br->n_ports; i++) {
986 struct port *port = br->ports[i];
987 uint8_t iface_ea[ETH_ADDR_LEN];
990 /* Mirror output ports don't participate. */
991 if (port->is_mirror_output_port) {
995 /* Choose the MAC address to represent the port. */
996 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
997 /* Find the interface with this Ethernet address (if any) so that
998 * we can provide the correct devname to the caller. */
1000 for (j = 0; j < port->n_ifaces; j++) {
1001 struct iface *candidate = port->ifaces[j];
1002 uint8_t candidate_ea[ETH_ADDR_LEN];
1003 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1004 && eth_addr_equals(iface_ea, candidate_ea)) {
1009 /* Choose the interface whose MAC address will represent the port.
1010 * The Linux kernel bonding code always chooses the MAC address of
1011 * the first slave added to a bond, and the Fedora networking
1012 * scripts always add slaves to a bond in alphabetical order, so
1013 * for compatibility we choose the interface with the name that is
1014 * first in alphabetical order. */
1015 iface = port->ifaces[0];
1016 for (j = 1; j < port->n_ifaces; j++) {
1017 struct iface *candidate = port->ifaces[j];
1018 if (strcmp(candidate->name, iface->name) < 0) {
1023 /* The local port doesn't count (since we're trying to choose its
1024 * MAC address anyway). */
1025 if (iface->dp_ifidx == ODPP_LOCAL) {
1030 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1032 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1033 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1034 iface->name, strerror(error));
1039 /* Compare against our current choice. */
1040 if (!eth_addr_is_multicast(iface_ea) &&
1041 !eth_addr_is_local(iface_ea) &&
1042 !eth_addr_is_reserved(iface_ea) &&
1043 !eth_addr_is_zero(iface_ea) &&
1044 eth_addr_compare_3way(iface_ea, ea) < 0)
1046 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1047 *hw_addr_iface = iface;
1050 if (eth_addr_is_multicast(ea)) {
1051 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1052 *hw_addr_iface = NULL;
1053 VLOG_WARN("bridge %s: using default bridge Ethernet "
1054 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1056 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1057 br->name, ETH_ADDR_ARGS(ea));
1061 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1062 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1063 * an interface on 'br', then that interface must be passed in as
1064 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1065 * 'hw_addr_iface' must be passed in as a null pointer. */
1067 bridge_pick_datapath_id(struct bridge *br,
1068 const uint8_t bridge_ea[ETH_ADDR_LEN],
1069 struct iface *hw_addr_iface)
1072 * The procedure for choosing a bridge MAC address will, in the most
1073 * ordinary case, also choose a unique MAC that we can use as a datapath
1074 * ID. In some special cases, though, multiple bridges will end up with
1075 * the same MAC address. This is OK for the bridges, but it will confuse
1076 * the OpenFlow controller, because each datapath needs a unique datapath
1079 * Datapath IDs must be unique. It is also very desirable that they be
1080 * stable from one run to the next, so that policy set on a datapath
1083 const char *datapath_id;
1086 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1087 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1091 if (hw_addr_iface) {
1093 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1095 * A bridge whose MAC address is taken from a VLAN network device
1096 * (that is, a network device created with vconfig(8) or similar
1097 * tool) will have the same MAC address as a bridge on the VLAN
1098 * device's physical network device.
1100 * Handle this case by hashing the physical network device MAC
1101 * along with the VLAN identifier.
1103 uint8_t buf[ETH_ADDR_LEN + 2];
1104 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1105 buf[ETH_ADDR_LEN] = vlan >> 8;
1106 buf[ETH_ADDR_LEN + 1] = vlan;
1107 return dpid_from_hash(buf, sizeof buf);
1110 * Assume that this bridge's MAC address is unique, since it
1111 * doesn't fit any of the cases we handle specially.
1116 * A purely internal bridge, that is, one that has no non-virtual
1117 * network devices on it at all, is more difficult because it has no
1118 * natural unique identifier at all.
1120 * When the host is a XenServer, we handle this case by hashing the
1121 * host's UUID with the name of the bridge. Names of bridges are
1122 * persistent across XenServer reboots, although they can be reused if
1123 * an internal network is destroyed and then a new one is later
1124 * created, so this is fairly effective.
1126 * When the host is not a XenServer, we punt by using a random MAC
1127 * address on each run.
1129 const char *host_uuid = xenserver_get_host_uuid();
1131 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1132 dpid = dpid_from_hash(combined, strlen(combined));
1138 return eth_addr_to_uint64(bridge_ea);
1142 dpid_from_hash(const void *data, size_t n)
1144 uint8_t hash[SHA1_DIGEST_SIZE];
1146 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1147 sha1_bytes(data, n, hash);
1148 eth_addr_mark_random(hash);
1149 return eth_addr_to_uint64(hash);
1153 iface_refresh_status(struct iface *iface)
1157 enum netdev_flags flags;
1166 if (!netdev_get_status(iface->netdev, &sh)) {
1168 char **keys, **values;
1170 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1171 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1176 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1179 shash_destroy_free_data(&sh);
1181 error = netdev_get_flags(iface->netdev, &flags);
1183 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1186 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1189 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1191 ovsrec_interface_set_duplex(iface->cfg,
1192 netdev_features_is_full_duplex(current)
1194 /* warning: uint64_t -> int64_t conversion */
1195 bps = netdev_features_to_bps(current);
1196 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1199 ovsrec_interface_set_duplex(iface->cfg, NULL);
1200 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1204 ovsrec_interface_set_link_state(iface->cfg,
1205 netdev_get_carrier(iface->netdev)
1208 error = netdev_get_mtu(iface->netdev, &mtu);
1211 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1214 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1219 iface_refresh_cfm_stats(struct iface *iface)
1223 const struct ovsrec_monitor *mon;
1225 mon = iface->cfg->monitor;
1232 for (i = 0; i < mon->n_remote_mps; i++) {
1233 const struct ovsrec_maintenance_point *mp;
1234 const struct remote_mp *rmp;
1236 mp = mon->remote_mps[i];
1237 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1239 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1242 if (hmap_is_empty(&cfm->x_remote_mps)) {
1243 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1246 struct remote_mp *rmp;
1247 int64_t *x_remote_mps;
1249 length = hmap_count(&cfm->x_remote_mps);
1250 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1253 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1254 x_remote_mps[i++] = rmp->mpid;
1257 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1261 if (hmap_is_empty(&cfm->x_remote_maids)) {
1262 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1265 char **x_remote_maids;
1266 struct remote_maid *rmaid;
1268 length = hmap_count(&cfm->x_remote_maids);
1269 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1272 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1275 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1277 for (j = 0; j < CCM_MAID_LEN; j++) {
1278 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1283 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1285 for (i = 0; i < length; i++) {
1286 free(x_remote_maids[i]);
1288 free(x_remote_maids);
1291 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1295 iface_refresh_stats(struct iface *iface)
1301 static const struct iface_stat iface_stats[] = {
1302 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1303 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1304 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1305 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1306 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1307 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1308 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1309 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1310 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1311 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1312 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1313 { "collisions", offsetof(struct netdev_stats, collisions) },
1315 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1316 const struct iface_stat *s;
1318 char *keys[N_STATS];
1319 int64_t values[N_STATS];
1322 struct netdev_stats stats;
1324 /* Intentionally ignore return value, since errors will set 'stats' to
1325 * all-1s, and we will deal with that correctly below. */
1326 netdev_get_stats(iface->netdev, &stats);
1329 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1330 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1331 if (value != UINT64_MAX) {
1338 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1342 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1344 struct ovsdb_datum datum;
1348 get_system_stats(&stats);
1350 ovsdb_datum_from_shash(&datum, &stats);
1351 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1355 static inline const char *
1356 nx_role_to_str(enum nx_role role)
1361 case NX_ROLE_MASTER:
1366 return "*** INVALID ROLE ***";
1371 bridge_refresh_controller_status(const struct bridge *br)
1374 const struct ovsrec_controller *cfg;
1376 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1378 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1379 struct ofproto_controller_info *cinfo =
1380 shash_find_data(&info, cfg->target);
1383 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1384 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1385 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1386 (char **) cinfo->pairs.values,
1389 ovsrec_controller_set_is_connected(cfg, false);
1390 ovsrec_controller_set_role(cfg, NULL);
1391 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1395 ofproto_free_ofproto_controller_info(&info);
1401 const struct ovsrec_open_vswitch *cfg;
1403 bool datapath_destroyed;
1404 bool database_changed;
1407 /* Let each bridge do the work that it needs to do. */
1408 datapath_destroyed = false;
1409 LIST_FOR_EACH (br, node, &all_bridges) {
1410 int error = bridge_run_one(br);
1412 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1413 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1414 "forcing reconfiguration", br->name);
1415 datapath_destroyed = true;
1419 /* (Re)configure if necessary. */
1420 database_changed = ovsdb_idl_run(idl);
1421 cfg = ovsrec_open_vswitch_first(idl);
1423 /* Re-configure SSL. We do this on every trip through the main loop,
1424 * instead of just when the database changes, because the contents of the
1425 * key and certificate files can change without the database changing.
1427 * We do this before bridge_reconfigure() because that function might
1428 * initiate SSL connections and thus requires SSL to be configured. */
1429 if (cfg && cfg->ssl) {
1430 const struct ovsrec_ssl *ssl = cfg->ssl;
1432 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1433 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1436 if (database_changed || datapath_destroyed) {
1438 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1440 bridge_configure_once(cfg);
1441 bridge_reconfigure(cfg);
1443 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1444 ovsdb_idl_txn_commit(txn);
1445 ovsdb_idl_txn_destroy(txn); /* XXX */
1447 /* We still need to reconfigure to avoid dangling pointers to
1448 * now-destroyed ovsrec structures inside bridge data. */
1449 static const struct ovsrec_open_vswitch null_cfg;
1451 bridge_reconfigure(&null_cfg);
1455 /* Refresh system and interface stats if necessary. */
1456 if (time_msec() >= stats_timer) {
1458 struct ovsdb_idl_txn *txn;
1460 txn = ovsdb_idl_txn_create(idl);
1461 LIST_FOR_EACH (br, node, &all_bridges) {
1464 for (i = 0; i < br->n_ports; i++) {
1465 struct port *port = br->ports[i];
1468 for (j = 0; j < port->n_ifaces; j++) {
1469 struct iface *iface = port->ifaces[j];
1470 iface_refresh_stats(iface);
1471 iface_refresh_cfm_stats(iface);
1472 iface_refresh_status(iface);
1475 bridge_refresh_controller_status(br);
1477 refresh_system_stats(cfg);
1478 ovsdb_idl_txn_commit(txn);
1479 ovsdb_idl_txn_destroy(txn); /* XXX */
1482 stats_timer = time_msec() + STATS_INTERVAL;
1490 struct iface *iface;
1492 LIST_FOR_EACH (br, node, &all_bridges) {
1493 ofproto_wait(br->ofproto);
1494 if (ofproto_has_primary_controller(br->ofproto)) {
1498 mac_learning_wait(br->ml);
1502 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1504 cfm_wait(iface->cfm);
1508 ovsdb_idl_wait(idl);
1509 poll_timer_wait_until(stats_timer);
1512 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1513 * configuration changes. */
1515 bridge_flush(struct bridge *br)
1517 COVERAGE_INC(bridge_flush);
1519 mac_learning_flush(br->ml);
1522 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1523 * such interface. */
1524 static struct iface *
1525 bridge_get_local_iface(struct bridge *br)
1529 for (i = 0; i < br->n_ports; i++) {
1530 struct port *port = br->ports[i];
1531 for (j = 0; j < port->n_ifaces; j++) {
1532 struct iface *iface = port->ifaces[j];
1533 if (iface->dp_ifidx == ODPP_LOCAL) {
1542 /* Bridge unixctl user interface functions. */
1544 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1545 const char *args, void *aux OVS_UNUSED)
1547 struct ds ds = DS_EMPTY_INITIALIZER;
1548 const struct bridge *br;
1549 const struct mac_entry *e;
1551 br = bridge_lookup(args);
1553 unixctl_command_reply(conn, 501, "no such bridge");
1557 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1558 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1559 if (e->port < 0 || e->port >= br->n_ports) {
1562 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1563 br->ports[e->port]->ifaces[0]->dp_ifidx,
1564 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1566 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1570 /* Bridge reconfiguration functions. */
1571 static struct bridge *
1572 bridge_create(const struct ovsrec_bridge *br_cfg)
1577 assert(!bridge_lookup(br_cfg->name));
1578 br = xzalloc(sizeof *br);
1580 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1586 dpif_flow_flush(br->dpif);
1588 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1591 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1593 dpif_delete(br->dpif);
1594 dpif_close(br->dpif);
1599 br->name = xstrdup(br_cfg->name);
1601 br->ml = mac_learning_create();
1602 eth_addr_nicira_random(br->default_ea);
1604 hmap_init(&br->ifaces);
1606 shash_init(&br->port_by_name);
1607 shash_init(&br->iface_by_name);
1611 list_push_back(&all_bridges, &br->node);
1613 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1619 bridge_destroy(struct bridge *br)
1624 while (br->n_ports > 0) {
1625 port_destroy(br->ports[br->n_ports - 1]);
1627 list_remove(&br->node);
1628 error = dpif_delete(br->dpif);
1629 if (error && error != ENOENT) {
1630 VLOG_ERR("failed to delete %s: %s",
1631 dpif_name(br->dpif), strerror(error));
1633 dpif_close(br->dpif);
1634 ofproto_destroy(br->ofproto);
1635 mac_learning_destroy(br->ml);
1636 hmap_destroy(&br->ifaces);
1637 shash_destroy(&br->port_by_name);
1638 shash_destroy(&br->iface_by_name);
1645 static struct bridge *
1646 bridge_lookup(const char *name)
1650 LIST_FOR_EACH (br, node, &all_bridges) {
1651 if (!strcmp(br->name, name)) {
1658 /* Handle requests for a listing of all flows known by the OpenFlow
1659 * stack, including those normally hidden. */
1661 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1662 const char *args, void *aux OVS_UNUSED)
1667 br = bridge_lookup(args);
1669 unixctl_command_reply(conn, 501, "Unknown bridge");
1674 ofproto_get_all_flows(br->ofproto, &results);
1676 unixctl_command_reply(conn, 200, ds_cstr(&results));
1677 ds_destroy(&results);
1680 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1681 * connections and reconnect. If BRIDGE is not specified, then all bridges
1682 * drop their controller connections and reconnect. */
1684 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1685 const char *args, void *aux OVS_UNUSED)
1688 if (args[0] != '\0') {
1689 br = bridge_lookup(args);
1691 unixctl_command_reply(conn, 501, "Unknown bridge");
1694 ofproto_reconnect_controllers(br->ofproto);
1696 LIST_FOR_EACH (br, node, &all_bridges) {
1697 ofproto_reconnect_controllers(br->ofproto);
1700 unixctl_command_reply(conn, 200, NULL);
1704 bridge_run_one(struct bridge *br)
1707 struct iface *iface;
1709 error = ofproto_run1(br->ofproto);
1714 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1718 error = ofproto_run2(br->ofproto, br->flush);
1721 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1722 struct ofpbuf *packet;
1728 packet = cfm_run(iface->cfm);
1730 iface_send_packet(iface, packet);
1731 ofpbuf_uninit(packet);
1740 bridge_get_controllers(const struct bridge *br,
1741 struct ovsrec_controller ***controllersp)
1743 struct ovsrec_controller **controllers;
1744 size_t n_controllers;
1746 controllers = br->cfg->controller;
1747 n_controllers = br->cfg->n_controller;
1749 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1755 *controllersp = controllers;
1757 return n_controllers;
1761 bridge_reconfigure_one(struct bridge *br)
1763 struct shash old_ports, new_ports;
1764 struct svec snoops, old_snoops;
1765 struct shash_node *node;
1766 enum ofproto_fail_mode fail_mode;
1769 /* Collect old ports. */
1770 shash_init(&old_ports);
1771 for (i = 0; i < br->n_ports; i++) {
1772 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1775 /* Collect new ports. */
1776 shash_init(&new_ports);
1777 for (i = 0; i < br->cfg->n_ports; i++) {
1778 const char *name = br->cfg->ports[i]->name;
1779 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1780 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1785 /* If we have a controller, then we need a local port. Complain if the
1786 * user didn't specify one.
1788 * XXX perhaps we should synthesize a port ourselves in this case. */
1789 if (bridge_get_controllers(br, NULL)) {
1790 char local_name[IF_NAMESIZE];
1793 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1794 local_name, sizeof local_name);
1795 if (!error && !shash_find(&new_ports, local_name)) {
1796 VLOG_WARN("bridge %s: controller specified but no local port "
1797 "(port named %s) defined",
1798 br->name, local_name);
1802 /* Get rid of deleted ports.
1803 * Get rid of deleted interfaces on ports that still exist. */
1804 SHASH_FOR_EACH (node, &old_ports) {
1805 struct port *port = node->data;
1806 const struct ovsrec_port *port_cfg;
1808 port_cfg = shash_find_data(&new_ports, node->name);
1812 port_del_ifaces(port, port_cfg);
1816 /* Create new ports.
1817 * Add new interfaces to existing ports.
1818 * Reconfigure existing ports. */
1819 SHASH_FOR_EACH (node, &new_ports) {
1820 struct port *port = shash_find_data(&old_ports, node->name);
1822 port = port_create(br, node->name);
1825 port_reconfigure(port, node->data);
1826 if (!port->n_ifaces) {
1827 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1828 br->name, port->name);
1832 shash_destroy(&old_ports);
1833 shash_destroy(&new_ports);
1835 /* Set the fail-mode */
1836 fail_mode = !br->cfg->fail_mode
1837 || !strcmp(br->cfg->fail_mode, "standalone")
1838 ? OFPROTO_FAIL_STANDALONE
1839 : OFPROTO_FAIL_SECURE;
1840 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1841 && !ofproto_has_primary_controller(br->ofproto)) {
1842 ofproto_flush_flows(br->ofproto);
1844 ofproto_set_fail_mode(br->ofproto, fail_mode);
1846 /* Delete all flows if we're switching from connected to standalone or vice
1847 * versa. (XXX Should we delete all flows if we are switching from one
1848 * controller to another?) */
1850 /* Configure OpenFlow controller connection snooping. */
1852 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1853 ovs_rundir(), br->name));
1854 svec_init(&old_snoops);
1855 ofproto_get_snoops(br->ofproto, &old_snoops);
1856 if (!svec_equal(&snoops, &old_snoops)) {
1857 ofproto_set_snoops(br->ofproto, &snoops);
1859 svec_destroy(&snoops);
1860 svec_destroy(&old_snoops);
1862 mirror_reconfigure(br);
1865 /* Initializes 'oc' appropriately as a management service controller for
1868 * The caller must free oc->target when it is no longer needed. */
1870 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1871 struct ofproto_controller *oc)
1873 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1874 oc->max_backoff = 0;
1875 oc->probe_interval = 60;
1876 oc->band = OFPROTO_OUT_OF_BAND;
1877 oc->accept_re = NULL;
1878 oc->update_resolv_conf = false;
1880 oc->burst_limit = 0;
1883 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1885 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1886 struct ofproto_controller *oc)
1888 oc->target = c->target;
1889 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1890 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1891 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1892 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1893 oc->accept_re = c->discover_accept_regex;
1894 oc->update_resolv_conf = c->discover_update_resolv_conf;
1895 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1896 oc->burst_limit = (c->controller_burst_limit
1897 ? *c->controller_burst_limit : 0);
1900 /* Configures the IP stack for 'br''s local interface properly according to the
1901 * configuration in 'c'. */
1903 bridge_configure_local_iface_netdev(struct bridge *br,
1904 struct ovsrec_controller *c)
1906 struct netdev *netdev;
1907 struct in_addr mask, gateway;
1909 struct iface *local_iface;
1912 /* Controller discovery does its own TCP/IP configuration later. */
1913 if (strcmp(c->target, "discover")) {
1917 /* If there's no local interface or no IP address, give up. */
1918 local_iface = bridge_get_local_iface(br);
1919 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1923 /* Bring up the local interface. */
1924 netdev = local_iface->netdev;
1925 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1927 /* Configure the IP address and netmask. */
1928 if (!c->local_netmask
1929 || !inet_aton(c->local_netmask, &mask)
1931 mask.s_addr = guess_netmask(ip.s_addr);
1933 if (!netdev_set_in4(netdev, ip, mask)) {
1934 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1935 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1938 /* Configure the default gateway. */
1939 if (c->local_gateway
1940 && inet_aton(c->local_gateway, &gateway)
1941 && gateway.s_addr) {
1942 if (!netdev_add_router(netdev, gateway)) {
1943 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1944 br->name, IP_ARGS(&gateway.s_addr));
1950 bridge_reconfigure_remotes(struct bridge *br,
1951 const struct sockaddr_in *managers,
1954 const char *disable_ib_str, *queue_id_str;
1955 bool disable_in_band = false;
1958 struct ovsrec_controller **controllers;
1959 size_t n_controllers;
1962 struct ofproto_controller *ocs;
1966 /* Check if we should disable in-band control on this bridge. */
1967 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1968 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1969 disable_in_band = true;
1972 /* Set OpenFlow queue ID for in-band control. */
1973 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1974 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1975 ofproto_set_in_band_queue(br->ofproto, queue_id);
1977 if (disable_in_band) {
1978 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1980 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1982 had_primary = ofproto_has_primary_controller(br->ofproto);
1984 n_controllers = bridge_get_controllers(br, &controllers);
1986 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1989 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1990 for (i = 0; i < n_controllers; i++) {
1991 struct ovsrec_controller *c = controllers[i];
1993 if (!strncmp(c->target, "punix:", 6)
1994 || !strncmp(c->target, "unix:", 5)) {
1995 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1997 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1998 * domain sockets and overwriting arbitrary local files. */
1999 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2000 "\"%s\" due to possibility for remote exploit",
2001 dpif_name(br->dpif), c->target);
2005 bridge_configure_local_iface_netdev(br, c);
2006 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2007 if (disable_in_band) {
2008 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2013 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2014 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2017 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2018 ofproto_flush_flows(br->ofproto);
2021 /* If there are no controllers and the bridge is in standalone
2022 * mode, set up a flow that matches every packet and directs
2023 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2024 * switch is in secure mode and we won't pass any traffic until
2025 * a controller has been defined and it tells us to do so. */
2027 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2028 union ofp_action action;
2029 struct cls_rule rule;
2031 memset(&action, 0, sizeof action);
2032 action.type = htons(OFPAT_OUTPUT);
2033 action.output.len = htons(sizeof action);
2034 action.output.port = htons(OFPP_NORMAL);
2035 cls_rule_init_catchall(&rule, 0);
2036 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2041 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2046 for (i = 0; i < br->n_ports; i++) {
2047 struct port *port = br->ports[i];
2048 for (j = 0; j < port->n_ifaces; j++) {
2049 struct iface *iface = port->ifaces[j];
2050 shash_add_once(ifaces, iface->name, iface);
2052 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2053 shash_add_once(ifaces, port->name, NULL);
2058 /* For robustness, in case the administrator moves around datapath ports behind
2059 * our back, we re-check all the datapath port numbers here.
2061 * This function will set the 'dp_ifidx' members of interfaces that have
2062 * disappeared to -1, so only call this function from a context where those
2063 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2064 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2065 * datapath, which doesn't support UINT16_MAX+1 ports. */
2067 bridge_fetch_dp_ifaces(struct bridge *br)
2069 struct dpif_port_dump dump;
2070 struct dpif_port dpif_port;
2073 /* Reset all interface numbers. */
2074 for (i = 0; i < br->n_ports; i++) {
2075 struct port *port = br->ports[i];
2076 for (j = 0; j < port->n_ifaces; j++) {
2077 struct iface *iface = port->ifaces[j];
2078 iface->dp_ifidx = -1;
2081 hmap_clear(&br->ifaces);
2083 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2084 struct iface *iface = iface_lookup(br, dpif_port.name);
2086 if (iface->dp_ifidx >= 0) {
2087 VLOG_WARN("%s reported interface %s twice",
2088 dpif_name(br->dpif), dpif_port.name);
2089 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2090 VLOG_WARN("%s reported interface %"PRIu16" twice",
2091 dpif_name(br->dpif), dpif_port.port_no);
2093 iface->dp_ifidx = dpif_port.port_no;
2094 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2095 hash_int(iface->dp_ifidx, 0));
2098 iface_set_ofport(iface->cfg,
2099 (iface->dp_ifidx >= 0
2100 ? odp_port_to_ofp_port(iface->dp_ifidx)
2106 /* Bridge packet processing functions. */
2109 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2111 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2114 static struct bond_entry *
2115 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
2118 assert(port->bond_mode == BM_SLB);
2119 return &port->bond_hash[bond_hash(mac, vlan)];
2123 bond_choose_iface(const struct port *port)
2125 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2126 size_t i, best_down_slave = -1;
2127 long long next_delay_expiration = LLONG_MAX;
2129 for (i = 0; i < port->n_ifaces; i++) {
2130 struct iface *iface = port->ifaces[i];
2132 if (iface->enabled) {
2134 } else if (iface->delay_expires < next_delay_expiration
2135 && (iface->lacp_attached
2136 || !(port->lacp & LACP_NEGOTIATED))) {
2137 best_down_slave = i;
2138 next_delay_expiration = iface->delay_expires;
2142 if (best_down_slave != -1) {
2143 struct iface *iface = port->ifaces[best_down_slave];
2145 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2146 "since no other interface is up", iface->name,
2147 iface->delay_expires - time_msec());
2148 bond_enable_slave(iface, true);
2151 return best_down_slave;
2155 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2156 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2158 struct iface *iface;
2160 assert(port->n_ifaces);
2161 if (port->n_ifaces == 1) {
2162 iface = port->ifaces[0];
2163 } else if (port->bond_mode == BM_AB) {
2164 if (port->active_iface < 0) {
2165 *tags |= port->no_ifaces_tag;
2168 iface = port->ifaces[port->active_iface];
2169 } else if (port->bond_mode == BM_SLB){
2170 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2171 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2172 || !port->ifaces[e->iface_idx]->enabled) {
2173 /* XXX select interface properly. The current interface selection
2174 * is only good for testing the rebalancing code. */
2175 e->iface_idx = bond_choose_iface(port);
2176 if (e->iface_idx < 0) {
2177 *tags |= port->no_ifaces_tag;
2180 e->iface_tag = tag_create_random();
2181 ((struct port *) port)->bond_compat_is_stale = true;
2183 *tags |= e->iface_tag;
2184 iface = port->ifaces[e->iface_idx];
2188 *dp_ifidx = iface->dp_ifidx;
2189 *tags |= iface->tag; /* Currently only used for bonding. */
2194 bond_link_status_update(struct iface *iface)
2196 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2197 struct port *port = iface->port;
2198 bool up = iface->up;
2199 int updelay, downdelay;
2201 updelay = port->updelay;
2202 downdelay = port->downdelay;
2204 if (iface->port->lacp & LACP_NEGOTIATED) {
2209 if (iface->port->lacp && up) {
2210 /* The interface is up if it's attached to an aggregator and its
2211 * partner is synchronized. The only exception is defaulted links.
2212 * They are not required to have synchronized partners because they
2213 * have no partners at all. However, they will only be attached if
2214 * negotiations failed on all interfaces in the bond. */
2215 up = iface->lacp_attached
2216 && (iface->lacp_partner.state & LACP_STATE_SYNC
2217 || iface->lacp_status == LACP_STATUS_DEFAULTED);
2221 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2222 /* Nothing to do. */
2225 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2226 iface->name, up ? "up" : "down");
2227 if (up == iface->enabled) {
2228 iface->delay_expires = LLONG_MAX;
2229 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2230 iface->name, up ? "disabled" : "enabled");
2231 } else if (up && port->active_iface < 0) {
2232 bond_enable_slave(iface, true);
2234 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2235 "other interface is up", iface->name, updelay);
2238 int delay = up ? updelay : downdelay;
2239 iface->delay_expires = time_msec() + delay;
2242 "interface %s: will be %s if it stays %s for %d ms",
2244 up ? "enabled" : "disabled",
2252 bond_choose_active_iface(struct port *port)
2254 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2256 port->active_iface = bond_choose_iface(port);
2257 port->active_iface_tag = tag_create_random();
2258 if (port->active_iface >= 0) {
2259 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2260 port->name, port->ifaces[port->active_iface]->name);
2262 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2268 bond_enable_slave(struct iface *iface, bool enable)
2270 struct port *port = iface->port;
2271 struct bridge *br = port->bridge;
2273 /* This acts as a recursion check. If the act of disabling a slave
2274 * causes a different slave to be enabled, the flag will allow us to
2275 * skip redundant work when we reenter this function. It must be
2276 * cleared on exit to keep things safe with multiple bonds. */
2277 static bool moving_active_iface = false;
2279 iface->delay_expires = LLONG_MAX;
2280 if (enable == iface->enabled) {
2284 iface->enabled = enable;
2285 if (!iface->enabled) {
2286 VLOG_WARN("interface %s: disabled", iface->name);
2287 ofproto_revalidate(br->ofproto, iface->tag);
2288 if (iface->port_ifidx == port->active_iface) {
2289 ofproto_revalidate(br->ofproto,
2290 port->active_iface_tag);
2292 /* Disabling a slave can lead to another slave being immediately
2293 * enabled if there will be no active slaves but one is waiting
2294 * on an updelay. In this case we do not need to run most of the
2295 * code for the newly enabled slave since there was no period
2296 * without an active slave and it is redundant with the disabling
2298 moving_active_iface = true;
2299 bond_choose_active_iface(port);
2301 bond_send_learning_packets(port);
2303 VLOG_WARN("interface %s: enabled", iface->name);
2304 if (port->active_iface < 0 && !moving_active_iface) {
2305 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2306 bond_choose_active_iface(port);
2307 bond_send_learning_packets(port);
2309 iface->tag = tag_create_random();
2312 moving_active_iface = false;
2313 port->bond_compat_is_stale = true;
2316 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2317 * bond interface. */
2319 bond_update_fake_iface_stats(struct port *port)
2321 struct netdev_stats bond_stats;
2322 struct netdev *bond_dev;
2325 memset(&bond_stats, 0, sizeof bond_stats);
2327 for (i = 0; i < port->n_ifaces; i++) {
2328 struct netdev_stats slave_stats;
2330 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2331 /* XXX: We swap the stats here because they are swapped back when
2332 * reported by the internal device. The reason for this is
2333 * internal devices normally represent packets going into the system
2334 * but when used as fake bond device they represent packets leaving
2335 * the system. We really should do this in the internal device
2336 * itself because changing it here reverses the counts from the
2337 * perspective of the switch. However, the internal device doesn't
2338 * know what type of device it represents so we have to do it here
2340 bond_stats.tx_packets += slave_stats.rx_packets;
2341 bond_stats.tx_bytes += slave_stats.rx_bytes;
2342 bond_stats.rx_packets += slave_stats.tx_packets;
2343 bond_stats.rx_bytes += slave_stats.tx_bytes;
2347 if (!netdev_open_default(port->name, &bond_dev)) {
2348 netdev_set_stats(bond_dev, &bond_stats);
2349 netdev_close(bond_dev);
2354 bond_link_carrier_update(struct iface *iface, bool carrier)
2356 if (carrier == iface->up) {
2360 if (iface->lacp_status == LACP_STATUS_CURRENT) {
2361 iface_set_lacp_expired(iface);
2364 iface->up = carrier;
2366 iface->port->bond_compat_is_stale = true;
2370 bond_run(struct bridge *br)
2374 for (i = 0; i < br->n_ports; i++) {
2375 struct port *port = br->ports[i];
2377 if (port->n_ifaces >= 2) {
2380 if (port->monitor) {
2381 assert(!port->miimon);
2383 /* Track carrier going up and down on interfaces. */
2384 while (!netdev_monitor_poll(port->monitor, &devname)) {
2385 struct iface *iface;
2387 iface = port_lookup_iface(port, devname);
2389 bool up = netdev_get_carrier(iface->netdev);
2390 bond_link_carrier_update(iface, up);
2395 assert(port->miimon);
2397 if (time_msec() >= port->bond_miimon_next_update) {
2398 for (j = 0; j < port->n_ifaces; j++) {
2399 struct iface *iface = port->ifaces[j];
2400 bool up = netdev_get_miimon(iface->netdev);
2401 bond_link_carrier_update(iface, up);
2403 port->bond_miimon_next_update = time_msec() +
2404 port->bond_miimon_interval;
2408 for (j = 0; j < port->n_ifaces; j++) {
2409 bond_link_status_update(port->ifaces[j]);
2412 for (j = 0; j < port->n_ifaces; j++) {
2413 struct iface *iface = port->ifaces[j];
2414 if (time_msec() >= iface->delay_expires) {
2415 bond_enable_slave(iface, !iface->enabled);
2419 if (port->bond_fake_iface
2420 && time_msec() >= port->bond_next_fake_iface_update) {
2421 bond_update_fake_iface_stats(port);
2422 port->bond_next_fake_iface_update = time_msec() + 1000;
2426 if (port->bond_compat_is_stale) {
2427 port->bond_compat_is_stale = false;
2428 port_update_bond_compat(port);
2434 bond_wait(struct bridge *br)
2438 for (i = 0; i < br->n_ports; i++) {
2439 struct port *port = br->ports[i];
2440 if (port->n_ifaces < 2) {
2444 if (port->monitor) {
2445 netdev_monitor_poll_wait(port->monitor);
2449 poll_timer_wait_until(port->bond_miimon_next_update);
2452 for (j = 0; j < port->n_ifaces; j++) {
2453 struct iface *iface = port->ifaces[j];
2454 if (iface->delay_expires != LLONG_MAX) {
2455 poll_timer_wait_until(iface->delay_expires);
2458 if (port->bond_fake_iface) {
2459 poll_timer_wait_until(port->bond_next_fake_iface_update);
2465 set_dst(struct dst *dst, const struct flow *flow,
2466 const struct port *in_port, const struct port *out_port,
2469 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2470 : in_port->vlan >= 0 ? in_port->vlan
2471 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2472 : vlan_tci_to_vid(flow->vlan_tci));
2473 return choose_output_iface(out_port, flow->dl_src, dst->vlan,
2474 &dst->dp_ifidx, tags);
2478 swap_dst(struct dst *p, struct dst *q)
2480 struct dst tmp = *p;
2485 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2486 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2487 * that we push to the datapath. We could in fact fully sort the array by
2488 * vlan, but in most cases there are at most two different vlan tags so that's
2489 * possibly overkill.) */
2491 partition_dsts(struct dst_set *set, int vlan)
2493 struct dst *first = set->dsts;
2494 struct dst *last = set->dsts + set->n;
2496 while (first != last) {
2498 * - All dsts < first have vlan == 'vlan'.
2499 * - All dsts >= last have vlan != 'vlan'.
2500 * - first < last. */
2501 while (first->vlan == vlan) {
2502 if (++first == last) {
2507 /* Same invariants, plus one additional:
2508 * - first->vlan != vlan.
2510 while (last[-1].vlan != vlan) {
2511 if (--last == first) {
2516 /* Same invariants, plus one additional:
2517 * - last[-1].vlan == vlan.*/
2518 swap_dst(first++, --last);
2523 mirror_mask_ffs(mirror_mask_t mask)
2525 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2530 dst_set_init(struct dst_set *set)
2532 set->dsts = set->builtin;
2534 set->allocated = ARRAY_SIZE(set->builtin);
2538 dst_set_add(struct dst_set *set, const struct dst *dst)
2540 if (set->n >= set->allocated) {
2541 size_t new_allocated;
2542 struct dst *new_dsts;
2544 new_allocated = set->allocated * 2;
2545 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2546 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2550 set->dsts = new_dsts;
2551 set->allocated = new_allocated;
2553 set->dsts[set->n++] = *dst;
2557 dst_set_free(struct dst_set *set)
2559 if (set->dsts != set->builtin) {
2565 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2568 for (i = 0; i < set->n; i++) {
2569 if (set->dsts[i].vlan == test->vlan
2570 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2578 port_trunks_vlan(const struct port *port, uint16_t vlan)
2580 return (port->vlan < 0
2581 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2585 port_includes_vlan(const struct port *port, uint16_t vlan)
2587 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2591 port_is_floodable(const struct port *port)
2595 for (i = 0; i < port->n_ifaces; i++) {
2596 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2597 port->ifaces[i]->dp_ifidx)) {
2605 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2606 const struct port *in_port, const struct port *out_port,
2607 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2609 mirror_mask_t mirrors = in_port->src_mirrors;
2614 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2615 if (flow_vlan == 0) {
2616 flow_vlan = OFP_VLAN_NONE;
2619 if (out_port == FLOOD_PORT) {
2620 for (i = 0; i < br->n_ports; i++) {
2621 struct port *port = br->ports[i];
2623 && port_is_floodable(port)
2624 && port_includes_vlan(port, vlan)
2625 && !port->is_mirror_output_port
2626 && set_dst(&dst, flow, in_port, port, tags)) {
2627 mirrors |= port->dst_mirrors;
2628 dst_set_add(set, &dst);
2631 *nf_output_iface = NF_OUT_FLOOD;
2632 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2633 dst_set_add(set, &dst);
2634 *nf_output_iface = dst.dp_ifidx;
2635 mirrors |= out_port->dst_mirrors;
2639 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2640 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2642 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2643 && !dst_is_duplicate(set, &dst)) {
2644 dst_set_add(set, &dst);
2647 for (i = 0; i < br->n_ports; i++) {
2648 struct port *port = br->ports[i];
2649 if (port_includes_vlan(port, m->out_vlan)
2650 && set_dst(&dst, flow, in_port, port, tags))
2652 if (port->vlan < 0) {
2653 dst.vlan = m->out_vlan;
2655 if (dst_is_duplicate(set, &dst)) {
2659 /* Use the vlan tag on the original flow instead of
2660 * the one passed in the vlan parameter. This ensures
2661 * that we compare the vlan from before any implicit
2662 * tagging tags place. This is necessary because
2663 * dst->vlan is the final vlan, after removing implicit
2665 if (port == in_port && dst.vlan == flow_vlan) {
2666 /* Don't send out input port on same VLAN. */
2669 dst_set_add(set, &dst);
2674 mirrors &= mirrors - 1;
2677 partition_dsts(set, flow_vlan);
2680 static void OVS_UNUSED
2681 print_dsts(const struct dst_set *set)
2685 for (i = 0; i < set->n; i++) {
2686 const struct dst *dst = &set->dsts[i];
2688 printf(">p%"PRIu16, dst->dp_ifidx);
2689 if (dst->vlan != OFP_VLAN_NONE) {
2690 printf("v%"PRIu16, dst->vlan);
2696 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2697 const struct port *in_port, const struct port *out_port,
2698 tag_type *tags, struct ofpbuf *actions,
2699 uint16_t *nf_output_iface)
2706 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2709 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2710 if (cur_vlan == 0) {
2711 cur_vlan = OFP_VLAN_NONE;
2713 for (i = 0; i < set.n; i++) {
2714 const struct dst *dst = &set.dsts[i];
2715 if (dst->vlan != cur_vlan) {
2716 if (dst->vlan == OFP_VLAN_NONE) {
2717 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2720 tci = htons(dst->vlan & VLAN_VID_MASK);
2721 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2722 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2724 cur_vlan = dst->vlan;
2726 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2731 /* Returns the effective vlan of a packet, taking into account both the
2732 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2733 * the packet is untagged and -1 indicates it has an invalid header and
2734 * should be dropped. */
2735 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2736 struct port *in_port, bool have_packet)
2738 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2739 if (in_port->vlan >= 0) {
2741 /* XXX support double tagging? */
2743 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2744 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2745 "packet received on port %s configured with "
2746 "implicit VLAN %"PRIu16,
2747 br->name, vlan, in_port->name, in_port->vlan);
2751 vlan = in_port->vlan;
2753 if (!port_includes_vlan(in_port, vlan)) {
2755 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2756 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2757 "packet received on port %s not configured for "
2759 br->name, vlan, in_port->name, vlan);
2768 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2769 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2770 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2772 is_gratuitous_arp(const struct flow *flow)
2774 return (flow->dl_type == htons(ETH_TYPE_ARP)
2775 && eth_addr_is_broadcast(flow->dl_dst)
2776 && (flow->nw_proto == ARP_OP_REPLY
2777 || (flow->nw_proto == ARP_OP_REQUEST
2778 && flow->nw_src == flow->nw_dst)));
2782 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2783 struct port *in_port)
2785 enum grat_arp_lock_type lock_type;
2788 /* We don't want to learn from gratuitous ARP packets that are reflected
2789 * back over bond slaves so we lock the learning table. */
2790 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2791 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2792 GRAT_ARP_LOCK_CHECK;
2794 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2797 /* The log messages here could actually be useful in debugging,
2798 * so keep the rate limit relatively high. */
2799 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2801 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2802 "on port %s in VLAN %d",
2803 br->name, ETH_ADDR_ARGS(flow->dl_src),
2804 in_port->name, vlan);
2805 ofproto_revalidate(br->ofproto, rev_tag);
2809 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2810 * dropped. Returns true if they may be forwarded, false if they should be
2813 * If 'have_packet' is true, it indicates that the caller is processing a
2814 * received packet. If 'have_packet' is false, then the caller is just
2815 * revalidating an existing flow because configuration has changed. Either
2816 * way, 'have_packet' only affects logging (there is no point in logging errors
2817 * during revalidation).
2819 * Sets '*in_portp' to the input port. This will be a null pointer if
2820 * flow->in_port does not designate a known input port (in which case
2821 * is_admissible() returns false).
2823 * When returning true, sets '*vlanp' to the effective VLAN of the input
2824 * packet, as returned by flow_get_vlan().
2826 * May also add tags to '*tags', although the current implementation only does
2827 * so in one special case.
2830 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2831 tag_type *tags, int *vlanp, struct port **in_portp)
2833 struct iface *in_iface;
2834 struct port *in_port;
2837 /* Find the interface and port structure for the received packet. */
2838 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2840 /* No interface? Something fishy... */
2842 /* Odd. A few possible reasons here:
2844 * - We deleted an interface but there are still a few packets
2845 * queued up from it.
2847 * - Someone externally added an interface (e.g. with "ovs-dpctl
2848 * add-if") that we don't know about.
2850 * - Packet arrived on the local port but the local port is not
2851 * one of our bridge ports.
2853 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2855 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2856 "interface %"PRIu16, br->name, flow->in_port);
2862 *in_portp = in_port = in_iface->port;
2863 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2868 /* Drop frames for reserved multicast addresses. */
2869 if (eth_addr_is_reserved(flow->dl_dst)) {
2873 /* Drop frames on ports reserved for mirroring. */
2874 if (in_port->is_mirror_output_port) {
2876 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2877 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2878 "%s, which is reserved exclusively for mirroring",
2879 br->name, in_port->name);
2884 /* When using LACP, do not accept packets from disabled interfaces. */
2885 if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) {
2889 /* Packets received on non-LACP bonds need special attention to avoid
2891 if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) {
2893 bool is_grat_arp_locked;
2895 if (eth_addr_is_multicast(flow->dl_dst)) {
2896 *tags |= in_port->active_iface_tag;
2897 if (in_port->active_iface != in_iface->port_ifidx) {
2898 /* Drop all multicast packets on inactive slaves. */
2903 /* Drop all packets for which we have learned a different input
2904 * port, because we probably sent the packet on one slave and got
2905 * it back on the other. Gratuitous ARP packets are an exception
2906 * to this rule: the host has moved to another switch. The exception
2907 * to the exception is if we locked the learning table to avoid
2908 * reflections on bond slaves. If this is the case, just drop the
2910 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2911 &is_grat_arp_locked);
2912 if (src_idx != -1 && src_idx != in_port->port_idx &&
2913 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2921 /* If the composed actions may be applied to any packet in the given 'flow',
2922 * returns true. Otherwise, the actions should only be applied to 'packet', or
2923 * not at all, if 'packet' was NULL. */
2925 process_flow(struct bridge *br, const struct flow *flow,
2926 const struct ofpbuf *packet, struct ofpbuf *actions,
2927 tag_type *tags, uint16_t *nf_output_iface)
2929 struct port *in_port;
2930 struct port *out_port;
2934 /* Check whether we should drop packets in this flow. */
2935 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2940 /* Learn source MAC (but don't try to learn from revalidation). */
2942 update_learning_table(br, flow, vlan, in_port);
2945 /* Determine output port. */
2946 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2948 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2949 out_port = br->ports[out_port_idx];
2950 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2951 /* If we are revalidating but don't have a learning entry then
2952 * eject the flow. Installing a flow that floods packets opens
2953 * up a window of time where we could learn from a packet reflected
2954 * on a bond and blackhole packets before the learning table is
2955 * updated to reflect the correct port. */
2958 out_port = FLOOD_PORT;
2961 /* Don't send packets out their input ports. */
2962 if (in_port == out_port) {
2968 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2976 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2977 struct ofpbuf *actions, tag_type *tags,
2978 uint16_t *nf_output_iface, void *br_)
2980 struct iface *iface;
2981 struct bridge *br = br_;
2983 COVERAGE_INC(bridge_process_flow);
2985 iface = iface_from_dp_ifidx(br, flow->in_port);
2987 if (cfm_should_process_flow(flow)) {
2988 if (packet && iface->cfm) {
2989 cfm_process_heartbeat(iface->cfm, packet);
2992 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
2994 lacp_process_packet(packet, iface);
2999 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3003 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3004 const struct nlattr *actions,
3006 unsigned long long int n_bytes, void *br_)
3008 struct bridge *br = br_;
3009 const struct nlattr *a;
3010 struct port *in_port;
3015 /* Feed information from the active flows back into the learning table to
3016 * ensure that table is always in sync with what is actually flowing
3017 * through the datapath.
3019 * We test that 'tags' is nonzero to ensure that only flows that include an
3020 * OFPP_NORMAL action are used for learning. This works because
3021 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3022 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3023 update_learning_table(br, flow, vlan, in_port);
3026 /* Account for bond slave utilization. */
3027 if (!br->has_bonded_ports) {
3030 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3031 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3032 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3033 if (out_port && out_port->n_ifaces >= 2 &&
3034 out_port->bond_mode == BM_SLB) {
3035 uint16_t vlan = (flow->vlan_tci
3036 ? vlan_tci_to_vid(flow->vlan_tci)
3038 struct bond_entry *e = lookup_bond_entry(out_port,
3039 flow->dl_src, vlan);
3040 e->tx_bytes += n_bytes;
3047 bridge_account_checkpoint_ofhook_cb(void *br_)
3049 struct bridge *br = br_;
3053 if (!br->has_bonded_ports) {
3058 for (i = 0; i < br->n_ports; i++) {
3059 struct port *port = br->ports[i];
3060 if (port->n_ifaces > 1 && port->bond_mode == BM_SLB
3061 && now >= port->bond_next_rebalance) {
3062 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3063 bond_rebalance_port(port);
3068 static struct ofhooks bridge_ofhooks = {
3069 bridge_normal_ofhook_cb,
3070 bridge_account_flow_ofhook_cb,
3071 bridge_account_checkpoint_ofhook_cb,
3074 /* LACP functions. */
3077 lacp_process_packet(const struct ofpbuf *packet, struct iface *iface)
3079 const struct lacp_pdu *pdu;
3081 if (!iface->port->lacp) {
3085 pdu = parse_lacp_packet(packet);
3090 iface->lacp_status = LACP_STATUS_CURRENT;
3091 iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX;
3093 iface->lacp_actor.state = iface_get_lacp_state(iface);
3094 if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) {
3098 if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) {
3099 iface->port->lacp_need_update = true;
3100 iface->lacp_partner = pdu->actor;
3105 lacp_update_ifaces(struct port *port)
3109 struct lacp_info lead_pri;
3111 port->lacp_need_update = false;
3118 for (i = 0; i < port->n_ifaces; i++) {
3119 struct iface *iface = port->ifaces[i];
3120 struct lacp_info pri;
3122 iface->lacp_attached = true;
3123 ofproto_revalidate(port->bridge->ofproto, iface->tag);
3125 /* Don't allow loopback interfaces to send traffic or lead. */
3126 if (eth_addr_equals(iface->lacp_partner.sysid,
3127 iface->lacp_actor.sysid)) {
3128 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3129 VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is "
3130 "connected to its own bridge", iface->name);
3131 iface->lacp_attached = false;
3135 if (iface->lacp_status == LACP_STATUS_DEFAULTED) {
3139 iface_get_lacp_priority(iface, &pri);
3141 if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) {
3148 port->lacp &= ~LACP_NEGOTIATED;
3152 port->lacp |= LACP_NEGOTIATED;
3154 for (i = 0; i < port->n_ifaces; i++) {
3155 struct iface *iface = port->ifaces[i];
3157 if (iface->lacp_status == LACP_STATUS_DEFAULTED
3158 || lead->lacp_partner.key != iface->lacp_partner.key
3159 || !eth_addr_equals(lead->lacp_partner.sysid,
3160 iface->lacp_partner.sysid)) {
3161 iface->lacp_attached = false;
3167 lacp_iface_may_tx(const struct iface *iface)
3169 return iface->port->lacp & LACP_ACTIVE
3170 || iface->lacp_status != LACP_STATUS_DEFAULTED;
3174 lacp_run(struct bridge *br)
3177 struct ofpbuf packet;
3179 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3181 for (i = 0; i < br->n_ports; i++) {
3182 struct port *port = br->ports[i];
3188 for (j = 0; j < port->n_ifaces; j++) {
3189 struct iface *iface = port->ifaces[j];
3191 if (time_msec() > iface->lacp_rx) {
3192 if (iface->lacp_status == LACP_STATUS_CURRENT) {
3193 iface_set_lacp_expired(iface);
3194 } else if (iface->lacp_status == LACP_STATUS_EXPIRED) {
3195 iface_set_lacp_defaulted(iface);
3200 if (port->lacp_need_update) {
3201 lacp_update_ifaces(port);
3204 for (j = 0; j < port->n_ifaces; j++) {
3205 struct iface *iface = port->ifaces[j];
3206 uint8_t ea[ETH_ADDR_LEN];
3209 if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) {
3213 error = netdev_get_etheraddr(iface->netdev, ea);
3215 iface->lacp_actor.state = iface_get_lacp_state(iface);
3216 compose_lacp_packet(&packet, &iface->lacp_actor,
3217 &iface->lacp_partner, ea);
3218 iface_send_packet(iface, &packet);
3220 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3221 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3222 "(%s)", iface->name, strerror(error));
3225 iface->lacp_tx = time_msec() +
3226 (iface->lacp_partner.state & LACP_STATE_TIME
3228 : LACP_SLOW_TIME_TX);
3231 ofpbuf_uninit(&packet);
3235 lacp_wait(struct bridge *br)
3239 for (i = 0; i < br->n_ports; i++) {
3240 struct port *port = br->ports[i];
3246 for (j = 0; j < port->n_ifaces; j++) {
3247 struct iface *iface = port->ifaces[j];
3249 if (lacp_iface_may_tx(iface)) {
3250 poll_timer_wait_until(iface->lacp_tx);
3253 if (iface->lacp_status != LACP_STATUS_DEFAULTED) {
3254 poll_timer_wait_until(iface->lacp_rx);
3260 /* Bonding functions. */
3262 /* Statistics for a single interface on a bonded port, used for load-based
3263 * bond rebalancing. */
3264 struct slave_balance {
3265 struct iface *iface; /* The interface. */
3266 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3268 /* All the "bond_entry"s that are assigned to this interface, in order of
3269 * increasing tx_bytes. */
3270 struct bond_entry **hashes;
3275 bond_mode_to_string(enum bond_mode bm) {
3276 static char *bm_slb = "balance-slb";
3277 static char *bm_ab = "active-backup";
3280 case BM_SLB: return bm_slb;
3281 case BM_AB: return bm_ab;
3288 /* Sorts pointers to pointers to bond_entries in ascending order by the
3289 * interface to which they are assigned, and within a single interface in
3290 * ascending order of bytes transmitted. */
3292 compare_bond_entries(const void *a_, const void *b_)
3294 const struct bond_entry *const *ap = a_;
3295 const struct bond_entry *const *bp = b_;
3296 const struct bond_entry *a = *ap;
3297 const struct bond_entry *b = *bp;
3298 if (a->iface_idx != b->iface_idx) {
3299 return a->iface_idx > b->iface_idx ? 1 : -1;
3300 } else if (a->tx_bytes != b->tx_bytes) {
3301 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3307 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3308 * *descending* order by number of bytes transmitted. */
3310 compare_slave_balance(const void *a_, const void *b_)
3312 const struct slave_balance *a = a_;
3313 const struct slave_balance *b = b_;
3314 if (a->iface->enabled != b->iface->enabled) {
3315 return a->iface->enabled ? -1 : 1;
3316 } else if (a->tx_bytes != b->tx_bytes) {
3317 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3324 swap_bals(struct slave_balance *a, struct slave_balance *b)
3326 struct slave_balance tmp = *a;
3331 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3332 * given that 'p' (and only 'p') might be in the wrong location.
3334 * This function invalidates 'p', since it might now be in a different memory
3337 resort_bals(struct slave_balance *p,
3338 struct slave_balance bals[], size_t n_bals)
3341 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3342 swap_bals(p, p - 1);
3344 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3345 swap_bals(p, p + 1);
3351 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3353 if (VLOG_IS_DBG_ENABLED()) {
3354 struct ds ds = DS_EMPTY_INITIALIZER;
3355 const struct slave_balance *b;
3357 for (b = bals; b < bals + n_bals; b++) {
3361 ds_put_char(&ds, ',');
3363 ds_put_format(&ds, " %s %"PRIu64"kB",
3364 b->iface->name, b->tx_bytes / 1024);
3366 if (!b->iface->enabled) {
3367 ds_put_cstr(&ds, " (disabled)");
3369 if (b->n_hashes > 0) {
3370 ds_put_cstr(&ds, " (");
3371 for (i = 0; i < b->n_hashes; i++) {
3372 const struct bond_entry *e = b->hashes[i];
3374 ds_put_cstr(&ds, " + ");
3376 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3377 e - port->bond_hash, e->tx_bytes / 1024);
3379 ds_put_cstr(&ds, ")");
3382 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3387 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3389 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3392 struct bond_entry *hash = from->hashes[hash_idx];
3393 struct port *port = from->iface->port;
3394 uint64_t delta = hash->tx_bytes;
3396 assert(port->bond_mode == BM_SLB);
3398 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3399 "from %s to %s (now carrying %"PRIu64"kB and "
3400 "%"PRIu64"kB load, respectively)",
3401 port->name, delta / 1024, hash - port->bond_hash,
3402 from->iface->name, to->iface->name,
3403 (from->tx_bytes - delta) / 1024,
3404 (to->tx_bytes + delta) / 1024);
3406 /* Delete element from from->hashes.
3408 * We don't bother to add the element to to->hashes because not only would
3409 * it require more work, the only purpose it would be to allow that hash to
3410 * be migrated to another slave in this rebalancing run, and there is no
3411 * point in doing that. */
3412 if (hash_idx == 0) {
3415 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3416 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3420 /* Shift load away from 'from' to 'to'. */
3421 from->tx_bytes -= delta;
3422 to->tx_bytes += delta;
3424 /* Arrange for flows to be revalidated. */
3425 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3426 hash->iface_idx = to->iface->port_ifidx;
3427 hash->iface_tag = tag_create_random();
3431 bond_rebalance_port(struct port *port)
3433 struct slave_balance *bals;
3435 struct bond_entry *hashes[BOND_MASK + 1];
3436 struct slave_balance *b, *from, *to;
3437 struct bond_entry *e;
3440 assert(port->bond_mode == BM_SLB);
3442 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3443 * descending order of tx_bytes, so that bals[0] represents the most
3444 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3447 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3448 * array for each slave_balance structure, we sort our local array of
3449 * hashes in order by slave, so that all of the hashes for a given slave
3450 * become contiguous in memory, and then we point each 'hashes' members of
3451 * a slave_balance structure to the start of a contiguous group. */
3452 n_bals = port->n_ifaces;
3453 bals = xmalloc(n_bals * sizeof *bals);
3454 for (b = bals; b < &bals[n_bals]; b++) {
3455 b->iface = port->ifaces[b - bals];
3460 for (i = 0; i <= BOND_MASK; i++) {
3461 hashes[i] = &port->bond_hash[i];
3463 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3464 for (i = 0; i <= BOND_MASK; i++) {
3466 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3467 b = &bals[e->iface_idx];
3468 b->tx_bytes += e->tx_bytes;
3470 b->hashes = &hashes[i];
3475 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3476 log_bals(bals, n_bals, port);
3478 /* Discard slaves that aren't enabled (which were sorted to the back of the
3479 * array earlier). */
3480 while (!bals[n_bals - 1].iface->enabled) {
3487 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3488 to = &bals[n_bals - 1];
3489 for (from = bals; from < to; ) {
3490 uint64_t overload = from->tx_bytes - to->tx_bytes;
3491 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3492 /* The extra load on 'from' (and all less-loaded slaves), compared
3493 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3494 * it is less than ~1Mbps. No point in rebalancing. */
3496 } else if (from->n_hashes == 1) {
3497 /* 'from' only carries a single MAC hash, so we can't shift any
3498 * load away from it, even though we want to. */
3501 /* 'from' is carrying significantly more load than 'to', and that
3502 * load is split across at least two different hashes. Pick a hash
3503 * to migrate to 'to' (the least-loaded slave), given that doing so
3504 * must decrease the ratio of the load on the two slaves by at
3507 * The sort order we use means that we prefer to shift away the
3508 * smallest hashes instead of the biggest ones. There is little
3509 * reason behind this decision; we could use the opposite sort
3510 * order to shift away big hashes ahead of small ones. */
3513 for (i = 0; i < from->n_hashes; i++) {
3514 double old_ratio, new_ratio;
3515 uint64_t delta = from->hashes[i]->tx_bytes;
3517 if (delta == 0 || from->tx_bytes - delta == 0) {
3518 /* Pointless move. */
3522 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3524 if (to->tx_bytes == 0) {
3525 /* Nothing on the new slave, move it. */
3529 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3530 new_ratio = (double)(from->tx_bytes - delta) /
3531 (to->tx_bytes + delta);
3533 if (new_ratio == 0) {
3534 /* Should already be covered but check to prevent division
3539 if (new_ratio < 1) {
3540 new_ratio = 1 / new_ratio;
3543 if (old_ratio - new_ratio > 0.1) {
3544 /* Would decrease the ratio, move it. */
3548 if (i < from->n_hashes) {
3549 bond_shift_load(from, to, i);
3550 port->bond_compat_is_stale = true;
3552 /* If the result of the migration changed the relative order of
3553 * 'from' and 'to' swap them back to maintain invariants. */
3554 if (order_swapped) {
3555 swap_bals(from, to);
3558 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3559 * point to different slave_balance structures. It is only
3560 * valid to do these two operations in a row at all because we
3561 * know that 'from' will not move past 'to' and vice versa. */
3562 resort_bals(from, bals, n_bals);
3563 resort_bals(to, bals, n_bals);
3570 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3571 * historical data to decay to <1% in 7 rebalancing runs. */
3572 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3581 bond_send_learning_packets(struct port *port)
3583 struct bridge *br = port->bridge;
3584 struct mac_entry *e;
3585 struct ofpbuf packet;
3586 int error, n_packets, n_errors;
3588 if (!port->n_ifaces || port->active_iface < 0) {
3592 ofpbuf_init(&packet, 128);
3593 error = n_packets = n_errors = 0;
3594 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3595 union ofp_action actions[2], *a;
3601 if (e->port == port->port_idx
3602 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3606 /* Compose actions. */
3607 memset(actions, 0, sizeof actions);
3610 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3611 a->vlan_vid.len = htons(sizeof *a);
3612 a->vlan_vid.vlan_vid = htons(e->vlan);
3615 a->output.type = htons(OFPAT_OUTPUT);
3616 a->output.len = htons(sizeof *a);
3617 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3622 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3624 flow_extract(&packet, 0, ODPP_NONE, &flow);
3625 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3632 ofpbuf_uninit(&packet);
3635 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3636 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3637 "packets, last error was: %s",
3638 port->name, n_errors, n_packets, strerror(error));
3640 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3641 port->name, n_packets);
3645 /* Bonding unixctl user interface functions. */
3648 bond_unixctl_list(struct unixctl_conn *conn,
3649 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3651 struct ds ds = DS_EMPTY_INITIALIZER;
3652 const struct bridge *br;
3654 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3656 LIST_FOR_EACH (br, node, &all_bridges) {
3659 for (i = 0; i < br->n_ports; i++) {
3660 const struct port *port = br->ports[i];
3661 if (port->n_ifaces > 1) {
3664 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3665 bond_mode_to_string(port->bond_mode));
3666 for (j = 0; j < port->n_ifaces; j++) {
3667 const struct iface *iface = port->ifaces[j];
3669 ds_put_cstr(&ds, ", ");
3671 ds_put_cstr(&ds, iface->name);
3673 ds_put_char(&ds, '\n');
3677 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3681 static struct port *
3682 bond_find(const char *name)
3684 const struct bridge *br;
3686 LIST_FOR_EACH (br, node, &all_bridges) {
3689 for (i = 0; i < br->n_ports; i++) {
3690 struct port *port = br->ports[i];
3691 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3700 ds_put_lacp_state(struct ds *ds, uint8_t state)
3702 if (state & LACP_STATE_ACT) {
3703 ds_put_cstr(ds, "activity ");
3706 if (state & LACP_STATE_TIME) {
3707 ds_put_cstr(ds, "timeout ");
3710 if (state & LACP_STATE_AGG) {
3711 ds_put_cstr(ds, "aggregation ");
3714 if (state & LACP_STATE_SYNC) {
3715 ds_put_cstr(ds, "synchronized ");
3718 if (state & LACP_STATE_COL) {
3719 ds_put_cstr(ds, "collecting ");
3722 if (state & LACP_STATE_DIST) {
3723 ds_put_cstr(ds, "distributing ");
3726 if (state & LACP_STATE_DEF) {
3727 ds_put_cstr(ds, "defaulted ");
3730 if (state & LACP_STATE_EXP) {
3731 ds_put_cstr(ds, "expired ");
3736 bond_unixctl_show(struct unixctl_conn *conn,
3737 const char *args, void *aux OVS_UNUSED)
3739 struct ds ds = DS_EMPTY_INITIALIZER;
3740 const struct port *port;
3743 port = bond_find(args);
3745 unixctl_command_reply(conn, 501, "no such bond");
3749 ds_put_format(&ds, "bond_mode: %s\n",
3750 bond_mode_to_string(port->bond_mode));
3753 ds_put_format(&ds, "\tlacp: %s\n",
3754 port->lacp & LACP_ACTIVE ? "active" : "passive");
3756 ds_put_cstr(&ds, "\tlacp: off\n");
3759 ds_put_format(&ds, "bond-detect-mode: %s\n",
3760 port->miimon ? "miimon" : "carrier");
3763 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3764 port->bond_miimon_interval);
3767 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3768 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3770 if (port->bond_mode == BM_SLB) {
3771 ds_put_format(&ds, "next rebalance: %lld ms\n",
3772 port->bond_next_rebalance - time_msec());
3775 for (j = 0; j < port->n_ifaces; j++) {
3776 const struct iface *iface = port->ifaces[j];
3777 struct bond_entry *be;
3780 ds_put_format(&ds, "slave %s: %s\n",
3781 iface->name, iface->enabled ? "enabled" : "disabled");
3782 if (j == port->active_iface) {
3783 ds_put_cstr(&ds, "\tactive slave\n");
3785 if (iface->delay_expires != LLONG_MAX) {
3786 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3787 iface->enabled ? "downdelay" : "updelay",
3788 iface->delay_expires - time_msec());
3792 ds_put_cstr(&ds, "\tstatus: ");
3794 if (iface->lacp_status == LACP_STATUS_CURRENT) {
3795 ds_put_cstr(&ds, "current ");
3796 } else if (iface->lacp_status == LACP_STATUS_EXPIRED) {
3797 ds_put_cstr(&ds, "expired ");
3799 ds_put_cstr(&ds, "defaulted ");
3802 if (iface->lacp_attached) {
3803 ds_put_cstr(&ds, "attached ");
3806 ds_put_cstr(&ds, "\n");
3808 ds_put_cstr(&ds, "\n\tactor sysid: ");
3809 ds_put_format(&ds, ETH_ADDR_FMT,
3810 ETH_ADDR_ARGS(iface->lacp_actor.sysid));
3811 ds_put_cstr(&ds, "\n");
3813 ds_put_format(&ds, "\tactor sys_priority: %u\n",
3814 ntohs(iface->lacp_actor.sys_priority));
3816 ds_put_format(&ds, "\tactor portid: %u\n",
3817 ntohs(iface->lacp_actor.portid));
3819 ds_put_format(&ds, "\tactor port_priority: %u\n",
3820 ntohs(iface->lacp_actor.port_priority));
3822 ds_put_format(&ds, "\tactor key: %u\n",
3823 ntohs(iface->lacp_actor.key));
3825 ds_put_cstr(&ds, "\tactor state: ");
3826 ds_put_lacp_state(&ds, iface_get_lacp_state(iface));
3827 ds_put_cstr(&ds, "\n\n");
3829 ds_put_cstr(&ds, "\tpartner sysid: ");
3830 ds_put_format(&ds, ETH_ADDR_FMT,
3831 ETH_ADDR_ARGS(iface->lacp_partner.sysid));
3832 ds_put_cstr(&ds, "\n");
3834 ds_put_format(&ds, "\tpartner sys_priority: %u\n",
3835 ntohs(iface->lacp_partner.sys_priority));
3837 ds_put_format(&ds, "\tpartner portid: %u\n",
3838 ntohs(iface->lacp_partner.portid));
3840 ds_put_format(&ds, "\tpartner port_priority: %u\n",
3841 ntohs(iface->lacp_partner.port_priority));
3843 ds_put_format(&ds, "\tpartner key: %u\n",
3844 ntohs(iface->lacp_partner.key));
3846 ds_put_cstr(&ds, "\tpartner state: ");
3847 ds_put_lacp_state(&ds, iface->lacp_partner.state);
3848 ds_put_cstr(&ds, "\n\n");
3851 if (port->bond_mode != BM_SLB) {
3856 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3857 int hash = be - port->bond_hash;
3858 struct mac_entry *me;
3860 if (be->iface_idx != j) {
3864 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3865 hash, be->tx_bytes / 1024);
3868 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3871 if (bond_hash(me->mac, me->vlan) == hash
3872 && me->port != port->port_idx
3873 && choose_output_iface(port, me->mac, me->vlan,
3875 && dp_ifidx == iface->dp_ifidx)
3877 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3878 ETH_ADDR_ARGS(me->mac));
3883 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3888 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3889 void *aux OVS_UNUSED)
3891 char *args = (char *) args_;
3892 char *save_ptr = NULL;
3893 char *bond_s, *hash_s, *slave_s;
3895 struct iface *iface;
3896 struct bond_entry *entry;
3899 bond_s = strtok_r(args, " ", &save_ptr);
3900 hash_s = strtok_r(NULL, " ", &save_ptr);
3901 slave_s = strtok_r(NULL, " ", &save_ptr);
3903 unixctl_command_reply(conn, 501,
3904 "usage: bond/migrate BOND HASH SLAVE");
3908 port = bond_find(bond_s);
3910 unixctl_command_reply(conn, 501, "no such bond");
3914 if (port->bond_mode != BM_SLB) {
3915 unixctl_command_reply(conn, 501, "not an SLB bond");
3919 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3920 hash = atoi(hash_s) & BOND_MASK;
3922 unixctl_command_reply(conn, 501, "bad hash");
3926 iface = port_lookup_iface(port, slave_s);
3928 unixctl_command_reply(conn, 501, "no such slave");
3932 if (!iface->enabled) {
3933 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3937 entry = &port->bond_hash[hash];
3938 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3939 entry->iface_idx = iface->port_ifidx;
3940 entry->iface_tag = tag_create_random();
3941 port->bond_compat_is_stale = true;
3942 unixctl_command_reply(conn, 200, "migrated");
3946 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3947 void *aux OVS_UNUSED)
3949 char *args = (char *) args_;
3950 char *save_ptr = NULL;
3951 char *bond_s, *slave_s;
3953 struct iface *iface;
3955 bond_s = strtok_r(args, " ", &save_ptr);
3956 slave_s = strtok_r(NULL, " ", &save_ptr);
3958 unixctl_command_reply(conn, 501,
3959 "usage: bond/set-active-slave BOND SLAVE");
3963 port = bond_find(bond_s);
3965 unixctl_command_reply(conn, 501, "no such bond");
3969 iface = port_lookup_iface(port, slave_s);
3971 unixctl_command_reply(conn, 501, "no such slave");
3975 if (!iface->enabled) {
3976 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3980 if (port->active_iface != iface->port_ifidx) {
3981 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3982 port->active_iface = iface->port_ifidx;
3983 port->active_iface_tag = tag_create_random();
3984 VLOG_INFO("port %s: active interface is now %s",
3985 port->name, iface->name);
3986 bond_send_learning_packets(port);
3987 unixctl_command_reply(conn, 200, "done");
3989 unixctl_command_reply(conn, 200, "no change");
3994 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3996 char *args = (char *) args_;
3997 char *save_ptr = NULL;
3998 char *bond_s, *slave_s;
4000 struct iface *iface;
4002 bond_s = strtok_r(args, " ", &save_ptr);
4003 slave_s = strtok_r(NULL, " ", &save_ptr);
4005 unixctl_command_reply(conn, 501,
4006 "usage: bond/enable/disable-slave BOND SLAVE");
4010 port = bond_find(bond_s);
4012 unixctl_command_reply(conn, 501, "no such bond");
4016 iface = port_lookup_iface(port, slave_s);
4018 unixctl_command_reply(conn, 501, "no such slave");
4022 bond_enable_slave(iface, enable);
4023 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
4027 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
4028 void *aux OVS_UNUSED)
4030 enable_slave(conn, args, true);
4034 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
4035 void *aux OVS_UNUSED)
4037 enable_slave(conn, args, false);
4041 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
4042 void *aux OVS_UNUSED)
4044 char *args = (char *) args_;
4045 uint8_t mac[ETH_ADDR_LEN];
4049 char *mac_s, *vlan_s;
4050 char *save_ptr = NULL;
4052 mac_s = strtok_r(args, " ", &save_ptr);
4053 vlan_s = strtok_r(NULL, " ", &save_ptr);
4056 if (sscanf(vlan_s, "%u", &vlan) != 1) {
4057 unixctl_command_reply(conn, 501, "invalid vlan");
4061 vlan = OFP_VLAN_NONE;
4064 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
4065 == ETH_ADDR_SCAN_COUNT) {
4066 hash = bond_hash(mac, vlan);
4068 hash_cstr = xasprintf("%u", hash);
4069 unixctl_command_reply(conn, 200, hash_cstr);
4072 unixctl_command_reply(conn, 501, "invalid mac");
4079 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
4080 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
4081 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
4082 unixctl_command_register("bond/set-active-slave",
4083 bond_unixctl_set_active_slave, NULL);
4084 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
4086 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
4088 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
4091 /* Port functions. */
4093 static struct port *
4094 port_create(struct bridge *br, const char *name)
4098 port = xzalloc(sizeof *port);
4100 port->port_idx = br->n_ports;
4102 port->trunks = NULL;
4103 port->name = xstrdup(name);
4104 port->active_iface = -1;
4106 if (br->n_ports >= br->allocated_ports) {
4107 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
4110 br->ports[br->n_ports++] = port;
4111 shash_add_assert(&br->port_by_name, port->name, port);
4113 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
4120 get_port_other_config(const struct ovsrec_port *port, const char *key,
4121 const char *default_value)
4125 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
4127 return value ? value : default_value;
4131 get_interface_other_config(const struct ovsrec_interface *iface,
4132 const char *key, const char *default_value)
4136 value = get_ovsrec_key_value(&iface->header_,
4137 &ovsrec_interface_col_other_config, key);
4138 return value ? value : default_value;
4142 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
4144 struct shash new_ifaces;
4147 /* Collect list of new interfaces. */
4148 shash_init(&new_ifaces);
4149 for (i = 0; i < cfg->n_interfaces; i++) {
4150 const char *name = cfg->interfaces[i]->name;
4151 shash_add_once(&new_ifaces, name, NULL);
4154 /* Get rid of deleted interfaces. */
4155 for (i = 0; i < port->n_ifaces; ) {
4156 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
4157 iface_destroy(port->ifaces[i]);
4163 shash_destroy(&new_ifaces);
4167 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4169 const char *detect_mode;
4170 struct shash new_ifaces;
4171 long long int next_rebalance, miimon_next_update, lacp_priority;
4172 unsigned long *trunks;
4178 /* Update settings. */
4179 port->updelay = cfg->bond_updelay;
4180 if (port->updelay < 0) {
4183 port->downdelay = cfg->bond_downdelay;
4184 if (port->downdelay < 0) {
4185 port->downdelay = 0;
4187 port->bond_rebalance_interval = atoi(
4188 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4189 if (port->bond_rebalance_interval < 1000) {
4190 port->bond_rebalance_interval = 1000;
4192 next_rebalance = time_msec() + port->bond_rebalance_interval;
4193 if (port->bond_next_rebalance > next_rebalance) {
4194 port->bond_next_rebalance = next_rebalance;
4197 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4200 if (!strcmp(detect_mode, "carrier")) {
4201 port->miimon = false;
4202 } else if (!strcmp(detect_mode, "miimon")) {
4203 port->miimon = true;
4205 port->miimon = false;
4206 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
4207 "carrier", port->name, detect_mode);
4210 port->bond_miimon_interval = atoi(
4211 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4212 if (port->bond_miimon_interval < 100) {
4213 port->bond_miimon_interval = 100;
4215 miimon_next_update = time_msec() + port->bond_miimon_interval;
4216 if (port->bond_miimon_next_update > miimon_next_update) {
4217 port->bond_miimon_next_update = miimon_next_update;
4220 if (!port->cfg->bond_mode ||
4221 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4222 port->bond_mode = BM_SLB;
4223 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4224 port->bond_mode = BM_AB;
4226 port->bond_mode = BM_SLB;
4227 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4228 port->name, port->cfg->bond_mode,
4229 bond_mode_to_string(port->bond_mode));
4232 /* Add new interfaces and update 'cfg' member of existing ones. */
4233 shash_init(&new_ifaces);
4234 for (i = 0; i < cfg->n_interfaces; i++) {
4235 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4236 struct iface *iface;
4238 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4239 VLOG_WARN("port %s: %s specified twice as port interface",
4240 port->name, if_cfg->name);
4241 iface_set_ofport(if_cfg, -1);
4245 iface = iface_lookup(port->bridge, if_cfg->name);
4247 if (iface->port != port) {
4248 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4250 port->bridge->name, if_cfg->name, iface->port->name);
4253 iface->cfg = if_cfg;
4255 iface = iface_create(port, if_cfg);
4258 /* Determine interface type. The local port always has type
4259 * "internal". Other ports take their type from the database and
4260 * default to "system" if none is specified. */
4261 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4262 : if_cfg->type[0] ? if_cfg->type
4266 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4269 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4270 iface->lacp_priority = UINT16_MAX;
4272 iface->lacp_priority = lacp_priority;
4275 shash_destroy(&new_ifaces);
4278 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4280 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4281 /* Prefer bondable links if unspecified. */
4282 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4284 port->lacp_priority = lacp_priority;
4287 if (!port->cfg->lacp) {
4288 /* XXX when LACP implementation has been sufficiently tested, enable by
4289 * default and make active on bonded ports. */
4291 } else if (!strcmp(port->cfg->lacp, "off")) {
4293 } else if (!strcmp(port->cfg->lacp, "active")) {
4294 port->lacp = LACP_ACTIVE;
4295 } else if (!strcmp(port->cfg->lacp, "passive")) {
4296 port->lacp = LACP_PASSIVE;
4298 VLOG_WARN("port %s: unknown LACP mode %s",
4299 port->name, port->cfg->lacp);
4306 if (port->n_ifaces < 2) {
4308 if (vlan >= 0 && vlan <= 4095) {
4309 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4314 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4315 * they even work as-is. But they have not been tested. */
4316 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4320 if (port->vlan != vlan) {
4322 bridge_flush(port->bridge);
4325 /* Get trunked VLANs. */
4327 if (vlan < 0 && cfg->n_trunks) {
4330 trunks = bitmap_allocate(4096);
4332 for (i = 0; i < cfg->n_trunks; i++) {
4333 int trunk = cfg->trunks[i];
4335 bitmap_set1(trunks, trunk);
4341 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4342 port->name, cfg->n_trunks);
4344 if (n_errors == cfg->n_trunks) {
4345 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4347 bitmap_free(trunks);
4350 } else if (vlan >= 0 && cfg->n_trunks) {
4351 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4355 ? port->trunks != NULL
4356 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4357 bridge_flush(port->bridge);
4359 bitmap_free(port->trunks);
4360 port->trunks = trunks;
4364 port_destroy(struct port *port)
4367 struct bridge *br = port->bridge;
4371 proc_net_compat_update_vlan(port->name, NULL, 0);
4372 proc_net_compat_update_bond(port->name, NULL);
4374 for (i = 0; i < MAX_MIRRORS; i++) {
4375 struct mirror *m = br->mirrors[i];
4376 if (m && m->out_port == port) {
4381 while (port->n_ifaces > 0) {
4382 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4385 shash_find_and_delete_assert(&br->port_by_name, port->name);
4387 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4388 del->port_idx = port->port_idx;
4390 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4392 netdev_monitor_destroy(port->monitor);
4394 bitmap_free(port->trunks);
4401 static struct port *
4402 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4404 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4405 return iface ? iface->port : NULL;
4408 static struct port *
4409 port_lookup(const struct bridge *br, const char *name)
4411 return shash_find_data(&br->port_by_name, name);
4414 static struct iface *
4415 port_lookup_iface(const struct port *port, const char *name)
4417 struct iface *iface = iface_lookup(port->bridge, name);
4418 return iface && iface->port == port ? iface : NULL;
4422 port_update_lacp(struct port *port)
4427 if (!port->lacp || port->n_ifaces < 1) {
4432 for (i = 0; i < port->n_ifaces; i++) {
4433 struct iface *iface = port->ifaces[i];
4435 if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) {
4440 if (iface->dp_ifidx == port->lacp_key) {
4441 key_changed = false;
4446 port->lacp_key = port->ifaces[0]->dp_ifidx;
4449 for (i = 0; i < port->n_ifaces; i++) {
4450 struct iface *iface = port->ifaces[i];
4452 iface->lacp_actor.sys_priority = htons(port->lacp_priority);
4453 memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN);
4455 iface->lacp_actor.port_priority = htons(iface->lacp_priority);
4456 iface->lacp_actor.portid = htons(iface->dp_ifidx);
4457 iface->lacp_actor.key = htons(port->lacp_key);
4461 port->lacp_need_update = true;
4465 port_update_bonding(struct port *port)
4467 if (port->monitor) {
4468 netdev_monitor_destroy(port->monitor);
4469 port->monitor = NULL;
4471 if (port->n_ifaces < 2) {
4472 /* Not a bonded port. */
4473 if (port->bond_hash) {
4474 free(port->bond_hash);
4475 port->bond_hash = NULL;
4476 port->bond_compat_is_stale = true;
4479 port->bond_fake_iface = false;
4483 if (port->bond_mode == BM_SLB && !port->bond_hash) {
4484 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4485 for (i = 0; i <= BOND_MASK; i++) {
4486 struct bond_entry *e = &port->bond_hash[i];
4490 port->no_ifaces_tag = tag_create_random();
4491 bond_choose_active_iface(port);
4492 port->bond_next_rebalance
4493 = time_msec() + port->bond_rebalance_interval;
4495 if (port->cfg->bond_fake_iface) {
4496 port->bond_next_fake_iface_update = time_msec();
4498 } else if (port->bond_mode != BM_SLB) {
4499 free(port->bond_hash);
4500 port->bond_hash = NULL;
4502 port->bond_compat_is_stale = true;
4503 port->bond_fake_iface = port->cfg->bond_fake_iface;
4505 if (!port->miimon) {
4506 port->monitor = netdev_monitor_create();
4507 for (i = 0; i < port->n_ifaces; i++) {
4508 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4515 port_update_bond_compat(struct port *port)
4517 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
4518 struct compat_bond bond;
4521 if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) {
4522 proc_net_compat_update_bond(port->name, NULL);
4527 bond.updelay = port->updelay;
4528 bond.downdelay = port->downdelay;
4531 bond.hashes = compat_hashes;
4532 if (port->bond_hash) {
4533 const struct bond_entry *e;
4534 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4535 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
4536 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
4537 cbh->hash = e - port->bond_hash;
4538 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
4543 bond.n_slaves = port->n_ifaces;
4544 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
4545 for (i = 0; i < port->n_ifaces; i++) {
4546 struct iface *iface = port->ifaces[i];
4547 struct compat_bond_slave *slave = &bond.slaves[i];
4548 slave->name = iface->name;
4550 /* We need to make the same determination as the Linux bonding
4551 * code to determine whether a slave should be consider "up".
4552 * The Linux function bond_miimon_inspect() supports four
4553 * BOND_LINK_* states:
4555 * - BOND_LINK_UP: carrier detected, updelay has passed.
4556 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
4557 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
4558 * - BOND_LINK_BACK: carrier detected, updelay in progress.
4560 * The function bond_info_show_slave() only considers BOND_LINK_UP
4561 * to be "up" and anything else to be "down".
4563 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
4567 netdev_get_etheraddr(iface->netdev, slave->mac);
4570 if (port->bond_fake_iface) {
4571 struct netdev *bond_netdev;
4573 if (!netdev_open_default(port->name, &bond_netdev)) {
4575 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
4577 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
4579 netdev_close(bond_netdev);
4583 proc_net_compat_update_bond(port->name, &bond);
4588 port_update_vlan_compat(struct port *port)
4590 struct bridge *br = port->bridge;
4591 char *vlandev_name = NULL;
4593 if (port->vlan > 0) {
4594 /* Figure out the name that the VLAN device should actually have, if it
4595 * existed. This takes some work because the VLAN device would not
4596 * have port->name in its name; rather, it would have the trunk port's
4597 * name, and 'port' would be attached to a bridge that also had the
4598 * VLAN device one of its ports. So we need to find a trunk port that
4599 * includes port->vlan.
4601 * There might be more than one candidate. This doesn't happen on
4602 * XenServer, so if it happens we just pick the first choice in
4603 * alphabetical order instead of creating multiple VLAN devices. */
4605 for (i = 0; i < br->n_ports; i++) {
4606 struct port *p = br->ports[i];
4607 if (port_trunks_vlan(p, port->vlan)
4609 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
4611 uint8_t ea[ETH_ADDR_LEN];
4612 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
4613 if (!eth_addr_is_multicast(ea) &&
4614 !eth_addr_is_reserved(ea) &&
4615 !eth_addr_is_zero(ea)) {
4616 vlandev_name = p->name;
4621 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
4624 /* Interface functions. */
4627 iface_set_lacp_defaulted(struct iface *iface)
4629 memset(&iface->lacp_partner, 0xff, sizeof iface->lacp_partner);
4630 iface->lacp_partner.state = 0;
4632 iface->lacp_status = LACP_STATUS_DEFAULTED;
4634 iface->port->lacp_need_update = true;
4638 iface_set_lacp_expired(struct iface *iface)
4640 iface->lacp_status = LACP_STATUS_EXPIRED;
4641 iface->lacp_partner.state |= LACP_STATE_TIME;
4642 iface->lacp_partner.state &= ~LACP_STATE_SYNC;
4644 iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX;
4649 iface_get_lacp_state(const struct iface *iface)
4653 if (iface->port->lacp & LACP_ACTIVE) {
4654 state |= LACP_STATE_ACT;
4657 if (iface->lacp_status == LACP_STATUS_DEFAULTED) {
4658 state |= LACP_STATE_DEF;
4659 } else if (iface->lacp_attached) {
4660 state |= LACP_STATE_SYNC;
4663 if (iface->lacp_status == LACP_STATUS_EXPIRED) {
4664 state |= LACP_STATE_EXP;
4667 if (iface->port->n_ifaces > 1) {
4668 state |= LACP_STATE_AGG;
4671 if (iface->enabled) {
4672 state |= LACP_STATE_COL | LACP_STATE_DIST;
4678 /* Given 'iface', populates 'priority' with data representing its LACP link
4679 * priority. If two priority objects populated by this function are compared
4680 * using memcmp, the higher priority link will be less than the lower priority
4683 iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority)
4685 uint16_t partner_priority, actor_priority;
4687 /* Choose the lacp_info of the higher priority system by comparing their
4688 * system priorities and mac addresses. */
4689 actor_priority = ntohs(iface->lacp_actor.sys_priority);
4690 partner_priority = ntohs(iface->lacp_partner.sys_priority);
4691 if (actor_priority < partner_priority) {
4692 *priority = iface->lacp_actor;
4693 } else if (partner_priority < actor_priority) {
4694 *priority = iface->lacp_partner;
4695 } else if (eth_addr_compare_3way(iface->lacp_actor.sysid,
4696 iface->lacp_partner.sysid) < 0) {
4697 *priority = iface->lacp_actor;
4699 *priority = iface->lacp_partner;
4702 /* Key and state are not used in priority comparisons. */
4704 priority->state = 0;
4708 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4711 union ofp_action action;
4713 memset(&action, 0, sizeof action);
4714 action.output.type = htons(OFPAT_OUTPUT);
4715 action.output.len = htons(sizeof action);
4716 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4718 flow_extract(packet, 0, ODPP_NONE, &flow);
4720 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4722 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4723 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4727 static struct iface *
4728 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4730 struct bridge *br = port->bridge;
4731 struct iface *iface;
4732 char *name = if_cfg->name;
4734 iface = xzalloc(sizeof *iface);
4736 iface->port_ifidx = port->n_ifaces;
4737 iface->name = xstrdup(name);
4738 iface->dp_ifidx = -1;
4739 iface->tag = tag_create_random();
4740 iface->delay_expires = LLONG_MAX;
4741 iface->netdev = NULL;
4742 iface->cfg = if_cfg;
4743 iface_set_lacp_defaulted(iface);
4745 if (port->lacp & LACP_ACTIVE) {
4746 iface_set_lacp_expired(iface);
4749 shash_add_assert(&br->iface_by_name, iface->name, iface);
4751 if (port->n_ifaces >= port->allocated_ifaces) {
4752 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4753 sizeof *port->ifaces);
4755 port->ifaces[port->n_ifaces++] = iface;
4756 if (port->n_ifaces > 1) {
4757 br->has_bonded_ports = true;
4760 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4768 iface_destroy(struct iface *iface)
4771 struct port *port = iface->port;
4772 struct bridge *br = port->bridge;
4773 bool del_active = port->active_iface == iface->port_ifidx;
4776 if (port->monitor) {
4777 netdev_monitor_remove(port->monitor, iface->netdev);
4780 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4782 if (iface->dp_ifidx >= 0) {
4783 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4786 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4787 del->port_ifidx = iface->port_ifidx;
4789 netdev_close(iface->netdev);
4792 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4793 bond_choose_active_iface(port);
4794 bond_send_learning_packets(port);
4797 cfm_destroy(iface->cfm);
4802 bridge_flush(port->bridge);
4806 static struct iface *
4807 iface_lookup(const struct bridge *br, const char *name)
4809 return shash_find_data(&br->iface_by_name, name);
4812 static struct iface *
4813 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4815 struct iface *iface;
4817 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4818 hash_int(dp_ifidx, 0), &br->ifaces) {
4819 if (iface->dp_ifidx == dp_ifidx) {
4826 /* Set Ethernet address of 'iface', if one is specified in the configuration
4829 iface_set_mac(struct iface *iface)
4831 uint8_t ea[ETH_ADDR_LEN];
4833 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4834 if (eth_addr_is_multicast(ea)) {
4835 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4837 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4838 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4839 iface->name, iface->name);
4841 int error = netdev_set_etheraddr(iface->netdev, ea);
4843 VLOG_ERR("interface %s: setting MAC failed (%s)",
4844 iface->name, strerror(error));
4850 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4852 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4855 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4859 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4861 * The value strings in '*shash' are taken directly from values[], not copied,
4862 * so the caller should not modify or free them. */
4864 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4865 struct shash *shash)
4870 for (i = 0; i < n; i++) {
4871 shash_add(shash, keys[i], values[i]);
4875 /* Creates 'keys' and 'values' arrays from 'shash'.
4877 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4878 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4879 * are populated with with strings taken directly from 'shash' and thus have
4880 * the same ownership of the key-value pairs in shash.
4883 shash_to_ovs_idl_map(struct shash *shash,
4884 char ***keys, char ***values, size_t *n)
4888 struct shash_node *sn;
4890 count = shash_count(shash);
4892 k = xmalloc(count * sizeof *k);
4893 v = xmalloc(count * sizeof *v);
4896 SHASH_FOR_EACH(sn, shash) {
4907 struct iface_delete_queues_cbdata {
4908 struct netdev *netdev;
4909 const struct ovsdb_datum *queues;
4913 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4915 union ovsdb_atom atom;
4917 atom.integer = target;
4918 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4922 iface_delete_queues(unsigned int queue_id,
4923 const struct shash *details OVS_UNUSED, void *cbdata_)
4925 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4927 if (!queue_ids_include(cbdata->queues, queue_id)) {
4928 netdev_delete_queue(cbdata->netdev, queue_id);
4933 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4935 if (!qos || qos->type[0] == '\0') {
4936 netdev_set_qos(iface->netdev, NULL, NULL);
4938 struct iface_delete_queues_cbdata cbdata;
4939 struct shash details;
4942 /* Configure top-level Qos for 'iface'. */
4943 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4944 qos->n_other_config, &details);
4945 netdev_set_qos(iface->netdev, qos->type, &details);
4946 shash_destroy(&details);
4948 /* Deconfigure queues that were deleted. */
4949 cbdata.netdev = iface->netdev;
4950 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4952 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4954 /* Configure queues for 'iface'. */
4955 for (i = 0; i < qos->n_queues; i++) {
4956 const struct ovsrec_queue *queue = qos->value_queues[i];
4957 unsigned int queue_id = qos->key_queues[i];
4959 shash_from_ovs_idl_map(queue->key_other_config,
4960 queue->value_other_config,
4961 queue->n_other_config, &details);
4962 netdev_set_queue(iface->netdev, queue_id, &details);
4963 shash_destroy(&details);
4969 iface_update_cfm(struct iface *iface)
4973 uint16_t *remote_mps;
4974 struct ovsrec_monitor *mon;
4975 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4977 mon = iface->cfg->monitor;
4983 if (netdev_get_etheraddr(iface->netdev, ea)) {
4984 VLOG_WARN("interface %s: Failed to get ethernet address. "
4985 "Skipping Monitor.", iface->name);
4989 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4990 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4995 iface->cfm = cfm_create();
4999 cfm->mpid = mon->mpid;
5000 cfm->interval = mon->interval ? *mon->interval : 1000;
5002 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
5003 memcpy(cfm->maid, maid, sizeof cfm->maid);
5005 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
5006 for(i = 0; i < mon->n_remote_mps; i++) {
5007 remote_mps[i] = mon->remote_mps[i]->mpid;
5009 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
5012 if (!cfm_configure(iface->cfm)) {
5013 cfm_destroy(iface->cfm);
5018 /* Port mirroring. */
5020 static struct mirror *
5021 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
5025 for (i = 0; i < MAX_MIRRORS; i++) {
5026 struct mirror *m = br->mirrors[i];
5027 if (m && uuid_equals(uuid, &m->uuid)) {
5035 mirror_reconfigure(struct bridge *br)
5037 unsigned long *rspan_vlans;
5040 /* Get rid of deleted mirrors. */
5041 for (i = 0; i < MAX_MIRRORS; i++) {
5042 struct mirror *m = br->mirrors[i];
5044 const struct ovsdb_datum *mc;
5045 union ovsdb_atom atom;
5047 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
5048 atom.uuid = br->mirrors[i]->uuid;
5049 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
5055 /* Add new mirrors and reconfigure existing ones. */
5056 for (i = 0; i < br->cfg->n_mirrors; i++) {
5057 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
5058 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
5060 mirror_reconfigure_one(m, cfg);
5062 mirror_create(br, cfg);
5066 /* Update port reserved status. */
5067 for (i = 0; i < br->n_ports; i++) {
5068 br->ports[i]->is_mirror_output_port = false;
5070 for (i = 0; i < MAX_MIRRORS; i++) {
5071 struct mirror *m = br->mirrors[i];
5072 if (m && m->out_port) {
5073 m->out_port->is_mirror_output_port = true;
5077 /* Update flooded vlans (for RSPAN). */
5079 if (br->cfg->n_flood_vlans) {
5080 rspan_vlans = bitmap_allocate(4096);
5082 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
5083 int64_t vlan = br->cfg->flood_vlans[i];
5084 if (vlan >= 0 && vlan < 4096) {
5085 bitmap_set1(rspan_vlans, vlan);
5086 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
5089 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
5094 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
5100 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
5105 for (i = 0; ; i++) {
5106 if (i >= MAX_MIRRORS) {
5107 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
5108 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
5111 if (!br->mirrors[i]) {
5116 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
5119 br->mirrors[i] = m = xzalloc(sizeof *m);
5122 m->name = xstrdup(cfg->name);
5123 shash_init(&m->src_ports);
5124 shash_init(&m->dst_ports);
5130 mirror_reconfigure_one(m, cfg);
5134 mirror_destroy(struct mirror *m)
5137 struct bridge *br = m->bridge;
5140 for (i = 0; i < br->n_ports; i++) {
5141 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5142 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5145 shash_destroy(&m->src_ports);
5146 shash_destroy(&m->dst_ports);
5149 m->bridge->mirrors[m->idx] = NULL;
5158 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
5159 struct shash *names)
5163 for (i = 0; i < n_ports; i++) {
5164 const char *name = ports[i]->name;
5165 if (port_lookup(m->bridge, name)) {
5166 shash_add_once(names, name, NULL);
5168 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
5169 "port %s", m->bridge->name, m->name, name);
5175 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
5181 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
5183 for (i = 0; i < cfg->n_select_vlan; i++) {
5184 int64_t vlan = cfg->select_vlan[i];
5185 if (vlan < 0 || vlan > 4095) {
5186 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
5187 m->bridge->name, m->name, vlan);
5189 (*vlans)[n_vlans++] = vlan;
5196 vlan_is_mirrored(const struct mirror *m, int vlan)
5200 for (i = 0; i < m->n_vlans; i++) {
5201 if (m->vlans[i] == vlan) {
5209 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
5213 for (i = 0; i < m->n_vlans; i++) {
5214 if (port_trunks_vlan(p, m->vlans[i])) {
5222 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
5224 struct shash src_ports, dst_ports;
5225 mirror_mask_t mirror_bit;
5226 struct port *out_port;
5233 if (strcmp(cfg->name, m->name)) {
5235 m->name = xstrdup(cfg->name);
5238 /* Get output port. */
5239 if (cfg->output_port) {
5240 out_port = port_lookup(m->bridge, cfg->output_port->name);
5242 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
5243 m->bridge->name, m->name);
5249 if (cfg->output_vlan) {
5250 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
5251 "output vlan; ignoring output vlan",
5252 m->bridge->name, m->name);
5254 } else if (cfg->output_vlan) {
5256 out_vlan = *cfg->output_vlan;
5258 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
5259 m->bridge->name, m->name);
5264 shash_init(&src_ports);
5265 shash_init(&dst_ports);
5266 if (cfg->select_all) {
5267 for (i = 0; i < m->bridge->n_ports; i++) {
5268 const char *name = m->bridge->ports[i]->name;
5269 shash_add_once(&src_ports, name, NULL);
5270 shash_add_once(&dst_ports, name, NULL);
5275 /* Get ports, and drop duplicates and ports that don't exist. */
5276 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
5278 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
5281 /* Get all the vlans, and drop duplicate and invalid vlans. */
5282 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
5285 /* Update mirror data. */
5286 if (!shash_equal_keys(&m->src_ports, &src_ports)
5287 || !shash_equal_keys(&m->dst_ports, &dst_ports)
5288 || m->n_vlans != n_vlans
5289 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
5290 || m->out_port != out_port
5291 || m->out_vlan != out_vlan) {
5292 bridge_flush(m->bridge);
5294 shash_swap(&m->src_ports, &src_ports);
5295 shash_swap(&m->dst_ports, &dst_ports);
5298 m->n_vlans = n_vlans;
5299 m->out_port = out_port;
5300 m->out_vlan = out_vlan;
5303 mirror_bit = MIRROR_MASK_C(1) << m->idx;
5304 for (i = 0; i < m->bridge->n_ports; i++) {
5305 struct port *port = m->bridge->ports[i];
5307 if (shash_find(&m->src_ports, port->name)
5310 ? port_trunks_any_mirrored_vlan(m, port)
5311 : vlan_is_mirrored(m, port->vlan)))) {
5312 port->src_mirrors |= mirror_bit;
5314 port->src_mirrors &= ~mirror_bit;
5317 if (shash_find(&m->dst_ports, port->name)) {
5318 port->dst_mirrors |= mirror_bit;
5320 port->dst_mirrors &= ~mirror_bit;
5325 shash_destroy(&src_ports);
5326 shash_destroy(&dst_ports);