1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
41 #include "dynamic-string.h"
47 #include "mac-learning.h"
51 #include "ofp-print.h"
53 #include "ofproto/netflow.h"
54 #include "ofproto/ofproto.h"
55 #include "ovsdb-data.h"
57 #include "poll-loop.h"
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_process_cfm);
79 COVERAGE_DEFINE(bridge_process_lacp);
80 COVERAGE_DEFINE(bridge_reconfigure);
81 COVERAGE_DEFINE(bridge_lacp_update);
89 struct dst builtin[32];
94 static void dst_set_init(struct dst_set *);
95 static void dst_set_add(struct dst_set *, const struct dst *);
96 static void dst_set_free(struct dst_set *);
99 LACP_CURRENT = 0x01, /* Current State. */
100 LACP_EXPIRED = 0x02, /* Expired State. */
101 LACP_DEFAULTED = 0x04, /* Partner is defaulted. */
102 LACP_ATTACHED = 0x08, /* Attached. Interface may be choosen for flows. */
106 /* These members are always valid. */
107 struct port *port; /* Containing port. */
108 size_t port_ifidx; /* Index within containing port. */
109 char *name; /* Host network device name. */
110 tag_type tag; /* Tag associated with this interface. */
111 long long delay_expires; /* Time after which 'enabled' may change. */
113 /* These members are valid only after bridge_reconfigure() causes them to
115 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
116 int dp_ifidx; /* Index within kernel datapath. */
117 struct netdev *netdev; /* Network device. */
118 bool enabled; /* May be chosen for flows? */
119 bool up; /* Is the interface up? */
120 const char *type; /* Usually same as cfg->type. */
121 struct cfm *cfm; /* Connectivity Fault Management */
122 const struct ovsrec_interface *cfg;
124 /* LACP information. */
125 enum lacp_status lacp_status; /* LACP status. */
126 uint16_t lacp_priority; /* LACP port priority. */
127 struct lacp_info lacp_actor; /* LACP actor information. */
128 struct lacp_info lacp_partner; /* LACP partner information. */
129 long long int lacp_tx; /* Next LACP message transmission time. */
130 long long int lacp_rx; /* Next LACP message receive time. */
133 #define BOND_MASK 0xff
135 int iface_idx; /* Index of assigned iface, or -1 if none. */
136 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
137 tag_type iface_tag; /* Tag associated with iface_idx. */
141 BM_TCP, /* Transport Layer Load Balance. */
142 BM_SLB, /* Source Load Balance. */
143 BM_AB /* Active Backup. */
146 #define MAX_MIRRORS 32
147 typedef uint32_t mirror_mask_t;
148 #define MIRROR_MASK_C(X) UINT32_C(X)
149 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
151 struct bridge *bridge;
154 struct uuid uuid; /* UUID of this "mirror" record in database. */
156 /* Selection criteria. */
157 struct shash src_ports; /* Name is port name; data is always NULL. */
158 struct shash dst_ports; /* Name is port name; data is always NULL. */
163 struct port *out_port;
167 /* Flags for a port's lacp member. */
168 #define LACP_ACTIVE 0x01 /* LACP is in active mode. */
169 #define LACP_PASSIVE 0x02 /* LACP is in passive mode. */
170 #define LACP_NEGOTIATED 0x04 /* LACP has successfully negotiated. */
172 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
174 struct bridge *bridge;
176 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
177 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
178 * NULL if all VLANs are trunked. */
179 const struct ovsrec_port *cfg;
182 /* An ordinary bridge port has 1 interface.
183 * A bridge port for bonding has at least 2 interfaces. */
184 struct iface **ifaces;
185 size_t n_ifaces, allocated_ifaces;
188 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
189 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
190 tag_type active_iface_tag; /* Tag for bcast flows. */
191 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
192 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
193 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
194 bool miimon; /* Use miimon instead of carrier? */
195 long long int bond_miimon_interval; /* Miimon status refresh interval. */
196 long long int bond_miimon_next_update; /* Time of next miimon update. */
197 long long int bond_next_fake_iface_update; /* Time of next update. */
198 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
200 /* LACP information. */
201 int lacp; /* LACP status flags. 0 if LACP is off. */
202 uint16_t lacp_key; /* LACP aggregation key. */
203 uint16_t lacp_priority; /* LACP system priority. */
204 bool lacp_need_update; /* Need to update attached interfaces? */
206 /* SLB specific bonding info. */
207 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
208 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
209 long long int bond_next_rebalance; /* Next rebalancing time. */
211 /* Port mirroring info. */
212 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
213 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
214 bool is_mirror_output_port; /* Does port mirroring send frames here? */
218 struct list node; /* Node in global list of bridges. */
219 char *name; /* User-specified arbitrary name. */
220 struct mac_learning *ml; /* MAC learning table. */
221 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
222 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
223 const struct ovsrec_bridge *cfg;
225 /* OpenFlow switch processing. */
226 struct ofproto *ofproto; /* OpenFlow switch. */
228 /* Kernel datapath information. */
229 struct dpif *dpif; /* Datapath. */
230 struct hmap ifaces; /* Contains "struct iface"s. */
234 size_t n_ports, allocated_ports;
235 struct shash iface_by_name; /* "struct iface"s indexed by name. */
236 struct shash port_by_name; /* "struct port"s indexed by name. */
239 bool has_bonded_ports;
244 /* Port mirroring. */
245 struct mirror *mirrors[MAX_MIRRORS];
248 /* List of all bridges. */
249 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
251 /* OVSDB IDL used to obtain configuration. */
252 static struct ovsdb_idl *idl;
254 /* Each time this timer expires, the bridge fetches systems and interface
255 * statistics and pushes them into the database. */
256 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
257 static long long int stats_timer = LLONG_MIN;
259 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
260 static void bridge_destroy(struct bridge *);
261 static struct bridge *bridge_lookup(const char *name);
262 static unixctl_cb_func bridge_unixctl_dump_flows;
263 static unixctl_cb_func bridge_unixctl_reconnect;
264 static int bridge_run_one(struct bridge *);
265 static size_t bridge_get_controllers(const struct bridge *br,
266 struct ovsrec_controller ***controllersp);
267 static void bridge_reconfigure_one(struct bridge *);
268 static void bridge_reconfigure_remotes(struct bridge *,
269 const struct sockaddr_in *managers,
271 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
272 static void bridge_fetch_dp_ifaces(struct bridge *);
273 static void bridge_flush(struct bridge *);
274 static void bridge_pick_local_hw_addr(struct bridge *,
275 uint8_t ea[ETH_ADDR_LEN],
276 struct iface **hw_addr_iface);
277 static uint64_t bridge_pick_datapath_id(struct bridge *,
278 const uint8_t bridge_ea[ETH_ADDR_LEN],
279 struct iface *hw_addr_iface);
280 static struct iface *bridge_get_local_iface(struct bridge *);
281 static uint64_t dpid_from_hash(const void *, size_t nbytes);
283 static unixctl_cb_func bridge_unixctl_fdb_show;
284 static unixctl_cb_func qos_unixctl_show;
286 static void lacp_run(struct port *);
287 static void lacp_wait(struct port *);
288 static void lacp_process_packet(const struct ofpbuf *, struct iface *);
290 static void bond_init(void);
291 static void bond_run(struct port *);
292 static void bond_wait(struct port *);
293 static void bond_rebalance_port(struct port *);
294 static void bond_send_learning_packets(struct port *);
295 static void bond_enable_slave(struct iface *iface, bool enable);
297 static void port_run(struct port *);
298 static void port_wait(struct port *);
299 static struct port *port_create(struct bridge *, const char *name);
300 static void port_reconfigure(struct port *, const struct ovsrec_port *);
301 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
302 static void port_destroy(struct port *);
303 static struct port *port_lookup(const struct bridge *, const char *name);
304 static struct iface *port_lookup_iface(const struct port *, const char *name);
305 static struct port *port_from_dp_ifidx(const struct bridge *,
307 static void port_update_bonding(struct port *);
308 static void port_update_lacp(struct port *);
310 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
311 static void mirror_destroy(struct mirror *);
312 static void mirror_reconfigure(struct bridge *);
313 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
314 static bool vlan_is_mirrored(const struct mirror *, int vlan);
316 static struct iface *iface_create(struct port *port,
317 const struct ovsrec_interface *if_cfg);
318 static void iface_destroy(struct iface *);
319 static struct iface *iface_lookup(const struct bridge *, const char *name);
320 static struct iface *iface_find(const char *name);
321 static struct iface *iface_from_dp_ifidx(const struct bridge *,
323 static void iface_set_mac(struct iface *);
324 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
325 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
326 static void iface_update_cfm(struct iface *);
327 static void iface_refresh_cfm_stats(struct iface *iface);
328 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
329 static uint8_t iface_get_lacp_state(const struct iface *);
330 static void iface_get_lacp_priority(struct iface *, struct lacp_info *);
331 static void iface_set_lacp_defaulted(struct iface *);
332 static void iface_set_lacp_expired(struct iface *);
334 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
336 static void shash_to_ovs_idl_map(struct shash *,
337 char ***keys, char ***values, size_t *n);
340 /* Hooks into ofproto processing. */
341 static struct ofhooks bridge_ofhooks;
343 /* Public functions. */
345 /* Initializes the bridge module, configuring it to obtain its configuration
346 * from an OVSDB server accessed over 'remote', which should be a string in a
347 * form acceptable to ovsdb_idl_create(). */
349 bridge_init(const char *remote)
351 /* Create connection to database. */
352 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
354 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
355 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
356 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
358 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
360 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
361 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
363 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
364 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
365 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
367 /* Register unixctl commands. */
368 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
369 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
370 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
372 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
380 struct bridge *br, *next_br;
382 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
385 ovsdb_idl_destroy(idl);
388 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
389 * but for which the ovs-vswitchd configuration 'cfg' is required. */
391 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
393 static bool already_configured_once;
394 struct svec bridge_names;
395 struct svec dpif_names, dpif_types;
398 /* Only do this once per ovs-vswitchd run. */
399 if (already_configured_once) {
402 already_configured_once = true;
404 stats_timer = time_msec() + STATS_INTERVAL;
406 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
407 svec_init(&bridge_names);
408 for (i = 0; i < cfg->n_bridges; i++) {
409 svec_add(&bridge_names, cfg->bridges[i]->name);
411 svec_sort(&bridge_names);
413 /* Iterate over all system dpifs and delete any of them that do not appear
415 svec_init(&dpif_names);
416 svec_init(&dpif_types);
417 dp_enumerate_types(&dpif_types);
418 for (i = 0; i < dpif_types.n; i++) {
421 dp_enumerate_names(dpif_types.names[i], &dpif_names);
423 /* Delete each dpif whose name is not in 'bridge_names'. */
424 for (j = 0; j < dpif_names.n; j++) {
425 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
429 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
438 svec_destroy(&bridge_names);
439 svec_destroy(&dpif_names);
440 svec_destroy(&dpif_types);
443 /* Callback for iterate_and_prune_ifaces(). */
445 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
447 if (!iface->netdev) {
448 /* We already reported a related error, don't bother duplicating it. */
452 if (iface->dp_ifidx < 0) {
453 VLOG_ERR("%s interface not in %s, dropping",
454 iface->name, dpif_name(br->dpif));
458 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
459 iface->name, iface->dp_ifidx);
463 /* Callback for iterate_and_prune_ifaces(). */
465 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
466 void *aux OVS_UNUSED)
468 /* Set policing attributes. */
469 netdev_set_policing(iface->netdev,
470 iface->cfg->ingress_policing_rate,
471 iface->cfg->ingress_policing_burst);
473 /* Set MAC address of internal interfaces other than the local
475 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
476 iface_set_mac(iface);
482 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
483 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
484 * deletes from 'br' any ports that no longer have any interfaces. */
486 iterate_and_prune_ifaces(struct bridge *br,
487 bool (*cb)(struct bridge *, struct iface *,
493 for (i = 0; i < br->n_ports; ) {
494 struct port *port = br->ports[i];
495 for (j = 0; j < port->n_ifaces; ) {
496 struct iface *iface = port->ifaces[j];
497 if (cb(br, iface, aux)) {
500 iface_set_ofport(iface->cfg, -1);
501 iface_destroy(iface);
505 if (port->n_ifaces) {
508 VLOG_WARN("%s port has no interfaces, dropping", port->name);
514 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
515 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
516 * responsible for freeing '*managersp' (with free()).
518 * You may be asking yourself "why does ovs-vswitchd care?", because
519 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
520 * should not be and in fact is not directly involved in that. But
521 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
522 * it has to tell in-band control where the managers are to enable that.
523 * (Thus, only managers connected in-band are collected.)
526 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
527 struct sockaddr_in **managersp, size_t *n_managersp)
529 struct sockaddr_in *managers = NULL;
530 size_t n_managers = 0;
531 struct shash targets;
534 /* Collect all of the potential targets from the "targets" columns of the
535 * rows pointed to by "manager_options", excluding any that are
537 shash_init(&targets);
538 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
539 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
541 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
542 shash_find_and_delete(&targets, m->target);
544 shash_add_once(&targets, m->target, NULL);
548 /* Now extract the targets' IP addresses. */
549 if (!shash_is_empty(&targets)) {
550 struct shash_node *node;
552 managers = xmalloc(shash_count(&targets) * sizeof *managers);
553 SHASH_FOR_EACH (node, &targets) {
554 const char *target = node->name;
555 struct sockaddr_in *sin = &managers[n_managers];
557 if ((!strncmp(target, "tcp:", 4)
558 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
559 (!strncmp(target, "ssl:", 4)
560 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
565 shash_destroy(&targets);
567 *managersp = managers;
568 *n_managersp = n_managers;
572 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
574 struct shash old_br, new_br;
575 struct shash_node *node;
576 struct bridge *br, *next;
577 struct sockaddr_in *managers;
580 int sflow_bridge_number;
582 COVERAGE_INC(bridge_reconfigure);
584 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
586 /* Collect old and new bridges. */
589 LIST_FOR_EACH (br, node, &all_bridges) {
590 shash_add(&old_br, br->name, br);
592 for (i = 0; i < ovs_cfg->n_bridges; i++) {
593 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
594 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
595 VLOG_WARN("more than one bridge named %s", br_cfg->name);
599 /* Get rid of deleted bridges and add new bridges. */
600 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
601 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
608 SHASH_FOR_EACH (node, &new_br) {
609 const char *br_name = node->name;
610 const struct ovsrec_bridge *br_cfg = node->data;
611 br = shash_find_data(&old_br, br_name);
613 /* If the bridge datapath type has changed, we need to tear it
614 * down and recreate. */
615 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
617 bridge_create(br_cfg);
620 bridge_create(br_cfg);
623 shash_destroy(&old_br);
624 shash_destroy(&new_br);
626 /* Reconfigure all bridges. */
627 LIST_FOR_EACH (br, node, &all_bridges) {
628 bridge_reconfigure_one(br);
631 /* Add and delete ports on all datapaths.
633 * The kernel will reject any attempt to add a given port to a datapath if
634 * that port already belongs to a different datapath, so we must do all
635 * port deletions before any port additions. */
636 LIST_FOR_EACH (br, node, &all_bridges) {
637 struct dpif_port_dump dump;
638 struct shash want_ifaces;
639 struct dpif_port dpif_port;
641 bridge_get_all_ifaces(br, &want_ifaces);
642 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
643 if (!shash_find(&want_ifaces, dpif_port.name)
644 && strcmp(dpif_port.name, br->name)) {
645 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
647 VLOG_WARN("failed to remove %s interface from %s: %s",
648 dpif_port.name, dpif_name(br->dpif),
653 shash_destroy(&want_ifaces);
655 LIST_FOR_EACH (br, node, &all_bridges) {
656 struct shash cur_ifaces, want_ifaces;
657 struct dpif_port_dump dump;
658 struct dpif_port dpif_port;
660 /* Get the set of interfaces currently in this datapath. */
661 shash_init(&cur_ifaces);
662 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
663 struct dpif_port *port_info = xmalloc(sizeof *port_info);
664 dpif_port_clone(port_info, &dpif_port);
665 shash_add(&cur_ifaces, dpif_port.name, port_info);
668 /* Get the set of interfaces we want on this datapath. */
669 bridge_get_all_ifaces(br, &want_ifaces);
671 hmap_clear(&br->ifaces);
672 SHASH_FOR_EACH (node, &want_ifaces) {
673 const char *if_name = node->name;
674 struct iface *iface = node->data;
675 struct dpif_port *dpif_port;
679 type = iface ? iface->type : "internal";
680 dpif_port = shash_find_data(&cur_ifaces, if_name);
682 /* If we have a port or a netdev already, and it's not the type we
683 * want, then delete the port (if any) and close the netdev (if
685 if ((dpif_port && strcmp(dpif_port->type, type))
686 || (iface && iface->netdev
687 && strcmp(type, netdev_get_type(iface->netdev)))) {
689 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
696 netdev_close(iface->netdev);
697 iface->netdev = NULL;
701 /* If the port doesn't exist or we don't have the netdev open,
702 * we need to do more work. */
703 if (!dpif_port || (iface && !iface->netdev)) {
704 struct netdev_options options;
705 struct netdev *netdev;
708 /* First open the network device. */
709 options.name = if_name;
711 options.args = &args;
712 options.ethertype = NETDEV_ETH_TYPE_NONE;
716 shash_from_ovs_idl_map(iface->cfg->key_options,
717 iface->cfg->value_options,
718 iface->cfg->n_options, &args);
720 error = netdev_open(&options, &netdev);
721 shash_destroy(&args);
724 VLOG_WARN("could not open network device %s (%s)",
725 if_name, strerror(error));
729 /* Then add the port if we haven't already. */
731 error = dpif_port_add(br->dpif, netdev, NULL);
733 netdev_close(netdev);
734 if (error == EFBIG) {
735 VLOG_ERR("ran out of valid port numbers on %s",
736 dpif_name(br->dpif));
739 VLOG_WARN("failed to add %s interface to %s: %s",
740 if_name, dpif_name(br->dpif),
747 /* Update 'iface'. */
749 iface->netdev = netdev;
750 iface->enabled = netdev_get_carrier(iface->netdev);
751 iface->up = iface->enabled;
753 } else if (iface && iface->netdev) {
757 shash_from_ovs_idl_map(iface->cfg->key_options,
758 iface->cfg->value_options,
759 iface->cfg->n_options, &args);
760 netdev_set_config(iface->netdev, &args);
761 shash_destroy(&args);
764 shash_destroy(&want_ifaces);
766 SHASH_FOR_EACH (node, &cur_ifaces) {
767 struct dpif_port *port_info = node->data;
768 dpif_port_destroy(port_info);
771 shash_destroy(&cur_ifaces);
773 sflow_bridge_number = 0;
774 LIST_FOR_EACH (br, node, &all_bridges) {
777 struct iface *local_iface;
778 struct iface *hw_addr_iface;
781 bridge_fetch_dp_ifaces(br);
783 iterate_and_prune_ifaces(br, check_iface, NULL);
785 /* Pick local port hardware address, datapath ID. */
786 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
787 local_iface = bridge_get_local_iface(br);
789 int error = netdev_set_etheraddr(local_iface->netdev, ea);
791 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
792 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
793 "Ethernet address: %s",
794 br->name, strerror(error));
797 memcpy(br->ea, ea, ETH_ADDR_LEN);
799 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
800 ofproto_set_datapath_id(br->ofproto, dpid);
802 dpid_string = xasprintf("%016"PRIx64, dpid);
803 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
806 /* Set NetFlow configuration on this bridge. */
807 if (br->cfg->netflow) {
808 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
809 struct netflow_options opts;
811 memset(&opts, 0, sizeof opts);
813 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
814 if (nf_cfg->engine_type) {
815 opts.engine_type = *nf_cfg->engine_type;
817 if (nf_cfg->engine_id) {
818 opts.engine_id = *nf_cfg->engine_id;
821 opts.active_timeout = nf_cfg->active_timeout;
822 if (!opts.active_timeout) {
823 opts.active_timeout = -1;
824 } else if (opts.active_timeout < 0) {
825 VLOG_WARN("bridge %s: active timeout interval set to negative "
826 "value, using default instead (%d seconds)", br->name,
827 NF_ACTIVE_TIMEOUT_DEFAULT);
828 opts.active_timeout = -1;
831 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
832 if (opts.add_id_to_iface) {
833 if (opts.engine_id > 0x7f) {
834 VLOG_WARN("bridge %s: netflow port mangling may conflict "
835 "with another vswitch, choose an engine id less "
836 "than 128", br->name);
838 if (br->n_ports > 508) {
839 VLOG_WARN("bridge %s: netflow port mangling will conflict "
840 "with another port when more than 508 ports are "
845 opts.collectors.n = nf_cfg->n_targets;
846 opts.collectors.names = nf_cfg->targets;
847 if (ofproto_set_netflow(br->ofproto, &opts)) {
848 VLOG_ERR("bridge %s: problem setting netflow collectors",
852 ofproto_set_netflow(br->ofproto, NULL);
855 /* Set sFlow configuration on this bridge. */
856 if (br->cfg->sflow) {
857 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
858 struct ovsrec_controller **controllers;
859 struct ofproto_sflow_options oso;
860 size_t n_controllers;
862 memset(&oso, 0, sizeof oso);
864 oso.targets.n = sflow_cfg->n_targets;
865 oso.targets.names = sflow_cfg->targets;
867 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
868 if (sflow_cfg->sampling) {
869 oso.sampling_rate = *sflow_cfg->sampling;
872 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
873 if (sflow_cfg->polling) {
874 oso.polling_interval = *sflow_cfg->polling;
877 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
878 if (sflow_cfg->header) {
879 oso.header_len = *sflow_cfg->header;
882 oso.sub_id = sflow_bridge_number++;
883 oso.agent_device = sflow_cfg->agent;
885 oso.control_ip = NULL;
886 n_controllers = bridge_get_controllers(br, &controllers);
887 for (i = 0; i < n_controllers; i++) {
888 if (controllers[i]->local_ip) {
889 oso.control_ip = controllers[i]->local_ip;
893 ofproto_set_sflow(br->ofproto, &oso);
895 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
897 ofproto_set_sflow(br->ofproto, NULL);
900 /* Update the controller and related settings. It would be more
901 * straightforward to call this from bridge_reconfigure_one(), but we
902 * can't do it there for two reasons. First, and most importantly, at
903 * that point we don't know the dp_ifidx of any interfaces that have
904 * been added to the bridge (because we haven't actually added them to
905 * the datapath). Second, at that point we haven't set the datapath ID
906 * yet; when a controller is configured, resetting the datapath ID will
907 * immediately disconnect from the controller, so it's better to set
908 * the datapath ID before the controller. */
909 bridge_reconfigure_remotes(br, managers, n_managers);
911 LIST_FOR_EACH (br, node, &all_bridges) {
912 for (i = 0; i < br->n_ports; i++) {
913 struct port *port = br->ports[i];
916 port_update_bonding(port);
917 port_update_lacp(port);
919 for (j = 0; j < port->n_ifaces; j++) {
920 iface_update_qos(port->ifaces[j], port->cfg->qos);
924 LIST_FOR_EACH (br, node, &all_bridges) {
925 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
928 LIST_FOR_EACH (br, node, &all_bridges) {
930 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
931 iface_update_cfm(iface);
937 /* ovs-vswitchd has completed initialization, so allow the process that
938 * forked us to exit successfully. */
939 daemonize_complete();
943 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
944 const struct ovsdb_idl_column *column,
947 const struct ovsdb_datum *datum;
948 union ovsdb_atom atom;
951 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
952 atom.string = (char *) key;
953 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
954 return idx == UINT_MAX ? NULL : datum->values[idx].string;
958 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
960 return get_ovsrec_key_value(&br_cfg->header_,
961 &ovsrec_bridge_col_other_config, key);
965 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
966 struct iface **hw_addr_iface)
972 *hw_addr_iface = NULL;
974 /* Did the user request a particular MAC? */
975 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
976 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
977 if (eth_addr_is_multicast(ea)) {
978 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
979 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
980 } else if (eth_addr_is_zero(ea)) {
981 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
987 /* Otherwise choose the minimum non-local MAC address among all of the
989 memset(ea, 0xff, ETH_ADDR_LEN);
990 for (i = 0; i < br->n_ports; i++) {
991 struct port *port = br->ports[i];
992 uint8_t iface_ea[ETH_ADDR_LEN];
995 /* Mirror output ports don't participate. */
996 if (port->is_mirror_output_port) {
1000 /* Choose the MAC address to represent the port. */
1001 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1002 /* Find the interface with this Ethernet address (if any) so that
1003 * we can provide the correct devname to the caller. */
1005 for (j = 0; j < port->n_ifaces; j++) {
1006 struct iface *candidate = port->ifaces[j];
1007 uint8_t candidate_ea[ETH_ADDR_LEN];
1008 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1009 && eth_addr_equals(iface_ea, candidate_ea)) {
1014 /* Choose the interface whose MAC address will represent the port.
1015 * The Linux kernel bonding code always chooses the MAC address of
1016 * the first slave added to a bond, and the Fedora networking
1017 * scripts always add slaves to a bond in alphabetical order, so
1018 * for compatibility we choose the interface with the name that is
1019 * first in alphabetical order. */
1020 iface = port->ifaces[0];
1021 for (j = 1; j < port->n_ifaces; j++) {
1022 struct iface *candidate = port->ifaces[j];
1023 if (strcmp(candidate->name, iface->name) < 0) {
1028 /* The local port doesn't count (since we're trying to choose its
1029 * MAC address anyway). */
1030 if (iface->dp_ifidx == ODPP_LOCAL) {
1035 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1037 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1038 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1039 iface->name, strerror(error));
1044 /* Compare against our current choice. */
1045 if (!eth_addr_is_multicast(iface_ea) &&
1046 !eth_addr_is_local(iface_ea) &&
1047 !eth_addr_is_reserved(iface_ea) &&
1048 !eth_addr_is_zero(iface_ea) &&
1049 eth_addr_compare_3way(iface_ea, ea) < 0)
1051 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1052 *hw_addr_iface = iface;
1055 if (eth_addr_is_multicast(ea)) {
1056 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1057 *hw_addr_iface = NULL;
1058 VLOG_WARN("bridge %s: using default bridge Ethernet "
1059 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1061 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1062 br->name, ETH_ADDR_ARGS(ea));
1066 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1067 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1068 * an interface on 'br', then that interface must be passed in as
1069 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1070 * 'hw_addr_iface' must be passed in as a null pointer. */
1072 bridge_pick_datapath_id(struct bridge *br,
1073 const uint8_t bridge_ea[ETH_ADDR_LEN],
1074 struct iface *hw_addr_iface)
1077 * The procedure for choosing a bridge MAC address will, in the most
1078 * ordinary case, also choose a unique MAC that we can use as a datapath
1079 * ID. In some special cases, though, multiple bridges will end up with
1080 * the same MAC address. This is OK for the bridges, but it will confuse
1081 * the OpenFlow controller, because each datapath needs a unique datapath
1084 * Datapath IDs must be unique. It is also very desirable that they be
1085 * stable from one run to the next, so that policy set on a datapath
1088 const char *datapath_id;
1091 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1092 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1096 if (hw_addr_iface) {
1098 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1100 * A bridge whose MAC address is taken from a VLAN network device
1101 * (that is, a network device created with vconfig(8) or similar
1102 * tool) will have the same MAC address as a bridge on the VLAN
1103 * device's physical network device.
1105 * Handle this case by hashing the physical network device MAC
1106 * along with the VLAN identifier.
1108 uint8_t buf[ETH_ADDR_LEN + 2];
1109 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1110 buf[ETH_ADDR_LEN] = vlan >> 8;
1111 buf[ETH_ADDR_LEN + 1] = vlan;
1112 return dpid_from_hash(buf, sizeof buf);
1115 * Assume that this bridge's MAC address is unique, since it
1116 * doesn't fit any of the cases we handle specially.
1121 * A purely internal bridge, that is, one that has no non-virtual
1122 * network devices on it at all, is more difficult because it has no
1123 * natural unique identifier at all.
1125 * When the host is a XenServer, we handle this case by hashing the
1126 * host's UUID with the name of the bridge. Names of bridges are
1127 * persistent across XenServer reboots, although they can be reused if
1128 * an internal network is destroyed and then a new one is later
1129 * created, so this is fairly effective.
1131 * When the host is not a XenServer, we punt by using a random MAC
1132 * address on each run.
1134 const char *host_uuid = xenserver_get_host_uuid();
1136 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1137 dpid = dpid_from_hash(combined, strlen(combined));
1143 return eth_addr_to_uint64(bridge_ea);
1147 dpid_from_hash(const void *data, size_t n)
1149 uint8_t hash[SHA1_DIGEST_SIZE];
1151 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1152 sha1_bytes(data, n, hash);
1153 eth_addr_mark_random(hash);
1154 return eth_addr_to_uint64(hash);
1158 iface_refresh_status(struct iface *iface)
1162 enum netdev_flags flags;
1171 if (!netdev_get_status(iface->netdev, &sh)) {
1173 char **keys, **values;
1175 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1176 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1181 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1184 shash_destroy_free_data(&sh);
1186 error = netdev_get_flags(iface->netdev, &flags);
1188 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1191 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1194 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1196 ovsrec_interface_set_duplex(iface->cfg,
1197 netdev_features_is_full_duplex(current)
1199 /* warning: uint64_t -> int64_t conversion */
1200 bps = netdev_features_to_bps(current);
1201 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1204 ovsrec_interface_set_duplex(iface->cfg, NULL);
1205 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1209 ovsrec_interface_set_link_state(iface->cfg,
1210 netdev_get_carrier(iface->netdev)
1213 error = netdev_get_mtu(iface->netdev, &mtu);
1214 if (!error && mtu != INT_MAX) {
1216 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1219 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1224 iface_refresh_cfm_stats(struct iface *iface)
1228 const struct ovsrec_monitor *mon;
1230 mon = iface->cfg->monitor;
1237 for (i = 0; i < mon->n_remote_mps; i++) {
1238 const struct ovsrec_maintenance_point *mp;
1239 const struct remote_mp *rmp;
1241 mp = mon->remote_mps[i];
1242 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1244 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1247 if (hmap_is_empty(&cfm->x_remote_mps)) {
1248 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1251 struct remote_mp *rmp;
1252 int64_t *x_remote_mps;
1254 length = hmap_count(&cfm->x_remote_mps);
1255 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1258 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1259 x_remote_mps[i++] = rmp->mpid;
1262 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1266 if (hmap_is_empty(&cfm->x_remote_maids)) {
1267 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1270 char **x_remote_maids;
1271 struct remote_maid *rmaid;
1273 length = hmap_count(&cfm->x_remote_maids);
1274 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1277 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1280 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1282 for (j = 0; j < CCM_MAID_LEN; j++) {
1283 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1288 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1290 for (i = 0; i < length; i++) {
1291 free(x_remote_maids[i]);
1293 free(x_remote_maids);
1296 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1300 iface_refresh_stats(struct iface *iface)
1306 static const struct iface_stat iface_stats[] = {
1307 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1308 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1309 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1310 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1311 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1312 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1313 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1314 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1315 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1316 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1317 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1318 { "collisions", offsetof(struct netdev_stats, collisions) },
1320 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1321 const struct iface_stat *s;
1323 char *keys[N_STATS];
1324 int64_t values[N_STATS];
1327 struct netdev_stats stats;
1329 /* Intentionally ignore return value, since errors will set 'stats' to
1330 * all-1s, and we will deal with that correctly below. */
1331 netdev_get_stats(iface->netdev, &stats);
1334 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1335 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1336 if (value != UINT64_MAX) {
1343 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1347 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1349 struct ovsdb_datum datum;
1353 get_system_stats(&stats);
1355 ovsdb_datum_from_shash(&datum, &stats);
1356 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1360 static inline const char *
1361 nx_role_to_str(enum nx_role role)
1366 case NX_ROLE_MASTER:
1371 return "*** INVALID ROLE ***";
1376 bridge_refresh_controller_status(const struct bridge *br)
1379 const struct ovsrec_controller *cfg;
1381 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1383 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1384 struct ofproto_controller_info *cinfo =
1385 shash_find_data(&info, cfg->target);
1388 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1389 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1390 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1391 (char **) cinfo->pairs.values,
1394 ovsrec_controller_set_is_connected(cfg, false);
1395 ovsrec_controller_set_role(cfg, NULL);
1396 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1400 ofproto_free_ofproto_controller_info(&info);
1406 const struct ovsrec_open_vswitch *cfg;
1408 bool datapath_destroyed;
1409 bool database_changed;
1412 /* Let each bridge do the work that it needs to do. */
1413 datapath_destroyed = false;
1414 LIST_FOR_EACH (br, node, &all_bridges) {
1415 int error = bridge_run_one(br);
1417 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1418 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1419 "forcing reconfiguration", br->name);
1420 datapath_destroyed = true;
1424 /* (Re)configure if necessary. */
1425 database_changed = ovsdb_idl_run(idl);
1426 cfg = ovsrec_open_vswitch_first(idl);
1428 /* Re-configure SSL. We do this on every trip through the main loop,
1429 * instead of just when the database changes, because the contents of the
1430 * key and certificate files can change without the database changing.
1432 * We do this before bridge_reconfigure() because that function might
1433 * initiate SSL connections and thus requires SSL to be configured. */
1434 if (cfg && cfg->ssl) {
1435 const struct ovsrec_ssl *ssl = cfg->ssl;
1437 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1438 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1441 if (database_changed || datapath_destroyed) {
1443 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1445 bridge_configure_once(cfg);
1446 bridge_reconfigure(cfg);
1448 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1449 ovsdb_idl_txn_commit(txn);
1450 ovsdb_idl_txn_destroy(txn); /* XXX */
1452 /* We still need to reconfigure to avoid dangling pointers to
1453 * now-destroyed ovsrec structures inside bridge data. */
1454 static const struct ovsrec_open_vswitch null_cfg;
1456 bridge_reconfigure(&null_cfg);
1460 /* Refresh system and interface stats if necessary. */
1461 if (time_msec() >= stats_timer) {
1463 struct ovsdb_idl_txn *txn;
1465 txn = ovsdb_idl_txn_create(idl);
1466 LIST_FOR_EACH (br, node, &all_bridges) {
1469 for (i = 0; i < br->n_ports; i++) {
1470 struct port *port = br->ports[i];
1473 for (j = 0; j < port->n_ifaces; j++) {
1474 struct iface *iface = port->ifaces[j];
1475 iface_refresh_stats(iface);
1476 iface_refresh_cfm_stats(iface);
1477 iface_refresh_status(iface);
1480 bridge_refresh_controller_status(br);
1482 refresh_system_stats(cfg);
1483 ovsdb_idl_txn_commit(txn);
1484 ovsdb_idl_txn_destroy(txn); /* XXX */
1487 stats_timer = time_msec() + STATS_INTERVAL;
1496 LIST_FOR_EACH (br, node, &all_bridges) {
1499 ofproto_wait(br->ofproto);
1500 if (ofproto_has_primary_controller(br->ofproto)) {
1504 mac_learning_wait(br->ml);
1506 for (i = 0; i < br->n_ports; i++) {
1507 port_wait(br->ports[i]);
1510 ovsdb_idl_wait(idl);
1511 poll_timer_wait_until(stats_timer);
1514 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1515 * configuration changes. */
1517 bridge_flush(struct bridge *br)
1519 COVERAGE_INC(bridge_flush);
1521 mac_learning_flush(br->ml);
1524 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1525 * such interface. */
1526 static struct iface *
1527 bridge_get_local_iface(struct bridge *br)
1531 for (i = 0; i < br->n_ports; i++) {
1532 struct port *port = br->ports[i];
1533 for (j = 0; j < port->n_ifaces; j++) {
1534 struct iface *iface = port->ifaces[j];
1535 if (iface->dp_ifidx == ODPP_LOCAL) {
1544 /* Bridge unixctl user interface functions. */
1546 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1547 const char *args, void *aux OVS_UNUSED)
1549 struct ds ds = DS_EMPTY_INITIALIZER;
1550 const struct bridge *br;
1551 const struct mac_entry *e;
1553 br = bridge_lookup(args);
1555 unixctl_command_reply(conn, 501, "no such bridge");
1559 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1560 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1561 if (e->port < 0 || e->port >= br->n_ports) {
1564 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1565 br->ports[e->port]->ifaces[0]->dp_ifidx,
1566 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1568 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1572 /* QoS unixctl user interface functions. */
1574 struct qos_unixctl_show_cbdata {
1576 struct iface *iface;
1580 qos_unixctl_show_cb(unsigned int queue_id,
1581 const struct shash *details,
1584 struct qos_unixctl_show_cbdata *data = aux;
1585 struct ds *ds = data->ds;
1586 struct iface *iface = data->iface;
1587 struct netdev_queue_stats stats;
1588 struct shash_node *node;
1591 ds_put_cstr(ds, "\n");
1593 ds_put_format(ds, "Queue %u:\n", queue_id);
1595 ds_put_cstr(ds, "Default:\n");
1598 SHASH_FOR_EACH (node, details) {
1599 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1602 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1604 if (stats.tx_packets != UINT64_MAX) {
1605 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1608 if (stats.tx_bytes != UINT64_MAX) {
1609 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1612 if (stats.tx_errors != UINT64_MAX) {
1613 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1616 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1617 queue_id, strerror(error));
1622 qos_unixctl_show(struct unixctl_conn *conn,
1623 const char *args, void *aux OVS_UNUSED)
1625 struct ds ds = DS_EMPTY_INITIALIZER;
1626 struct shash sh = SHASH_INITIALIZER(&sh);
1627 struct iface *iface;
1629 struct shash_node *node;
1630 struct qos_unixctl_show_cbdata data;
1633 iface = iface_find(args);
1635 unixctl_command_reply(conn, 501, "no such interface");
1639 netdev_get_qos(iface->netdev, &type, &sh);
1641 if (*type != '\0') {
1642 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1644 SHASH_FOR_EACH (node, &sh) {
1645 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1650 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1653 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1655 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1657 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1658 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1661 shash_destroy_free_data(&sh);
1665 /* Bridge reconfiguration functions. */
1666 static struct bridge *
1667 bridge_create(const struct ovsrec_bridge *br_cfg)
1672 assert(!bridge_lookup(br_cfg->name));
1673 br = xzalloc(sizeof *br);
1675 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1681 dpif_flow_flush(br->dpif);
1683 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1686 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1688 dpif_delete(br->dpif);
1689 dpif_close(br->dpif);
1694 br->name = xstrdup(br_cfg->name);
1696 br->ml = mac_learning_create();
1697 eth_addr_nicira_random(br->default_ea);
1699 hmap_init(&br->ifaces);
1701 shash_init(&br->port_by_name);
1702 shash_init(&br->iface_by_name);
1706 list_push_back(&all_bridges, &br->node);
1708 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1714 bridge_destroy(struct bridge *br)
1719 while (br->n_ports > 0) {
1720 port_destroy(br->ports[br->n_ports - 1]);
1722 list_remove(&br->node);
1723 error = dpif_delete(br->dpif);
1724 if (error && error != ENOENT) {
1725 VLOG_ERR("failed to delete %s: %s",
1726 dpif_name(br->dpif), strerror(error));
1728 dpif_close(br->dpif);
1729 ofproto_destroy(br->ofproto);
1730 mac_learning_destroy(br->ml);
1731 hmap_destroy(&br->ifaces);
1732 shash_destroy(&br->port_by_name);
1733 shash_destroy(&br->iface_by_name);
1740 static struct bridge *
1741 bridge_lookup(const char *name)
1745 LIST_FOR_EACH (br, node, &all_bridges) {
1746 if (!strcmp(br->name, name)) {
1753 /* Handle requests for a listing of all flows known by the OpenFlow
1754 * stack, including those normally hidden. */
1756 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1757 const char *args, void *aux OVS_UNUSED)
1762 br = bridge_lookup(args);
1764 unixctl_command_reply(conn, 501, "Unknown bridge");
1769 ofproto_get_all_flows(br->ofproto, &results);
1771 unixctl_command_reply(conn, 200, ds_cstr(&results));
1772 ds_destroy(&results);
1775 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1776 * connections and reconnect. If BRIDGE is not specified, then all bridges
1777 * drop their controller connections and reconnect. */
1779 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1780 const char *args, void *aux OVS_UNUSED)
1783 if (args[0] != '\0') {
1784 br = bridge_lookup(args);
1786 unixctl_command_reply(conn, 501, "Unknown bridge");
1789 ofproto_reconnect_controllers(br->ofproto);
1791 LIST_FOR_EACH (br, node, &all_bridges) {
1792 ofproto_reconnect_controllers(br->ofproto);
1795 unixctl_command_reply(conn, 200, NULL);
1799 bridge_run_one(struct bridge *br)
1804 error = ofproto_run1(br->ofproto);
1809 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1811 for (i = 0; i < br->n_ports; i++) {
1812 port_run(br->ports[i]);
1815 error = ofproto_run2(br->ofproto, br->flush);
1822 bridge_get_controllers(const struct bridge *br,
1823 struct ovsrec_controller ***controllersp)
1825 struct ovsrec_controller **controllers;
1826 size_t n_controllers;
1828 controllers = br->cfg->controller;
1829 n_controllers = br->cfg->n_controller;
1831 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1837 *controllersp = controllers;
1839 return n_controllers;
1843 bridge_reconfigure_one(struct bridge *br)
1845 struct shash old_ports, new_ports;
1846 struct svec snoops, old_snoops;
1847 struct shash_node *node;
1848 enum ofproto_fail_mode fail_mode;
1851 /* Collect old ports. */
1852 shash_init(&old_ports);
1853 for (i = 0; i < br->n_ports; i++) {
1854 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1857 /* Collect new ports. */
1858 shash_init(&new_ports);
1859 for (i = 0; i < br->cfg->n_ports; i++) {
1860 const char *name = br->cfg->ports[i]->name;
1861 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1862 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1867 /* If we have a controller, then we need a local port. Complain if the
1868 * user didn't specify one.
1870 * XXX perhaps we should synthesize a port ourselves in this case. */
1871 if (bridge_get_controllers(br, NULL)) {
1872 char local_name[IF_NAMESIZE];
1875 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1876 local_name, sizeof local_name);
1877 if (!error && !shash_find(&new_ports, local_name)) {
1878 VLOG_WARN("bridge %s: controller specified but no local port "
1879 "(port named %s) defined",
1880 br->name, local_name);
1884 /* Get rid of deleted ports.
1885 * Get rid of deleted interfaces on ports that still exist. */
1886 SHASH_FOR_EACH (node, &old_ports) {
1887 struct port *port = node->data;
1888 const struct ovsrec_port *port_cfg;
1890 port_cfg = shash_find_data(&new_ports, node->name);
1894 port_del_ifaces(port, port_cfg);
1898 /* Create new ports.
1899 * Add new interfaces to existing ports.
1900 * Reconfigure existing ports. */
1901 SHASH_FOR_EACH (node, &new_ports) {
1902 struct port *port = shash_find_data(&old_ports, node->name);
1904 port = port_create(br, node->name);
1907 port_reconfigure(port, node->data);
1908 if (!port->n_ifaces) {
1909 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1910 br->name, port->name);
1914 shash_destroy(&old_ports);
1915 shash_destroy(&new_ports);
1917 /* Set the fail-mode */
1918 fail_mode = !br->cfg->fail_mode
1919 || !strcmp(br->cfg->fail_mode, "standalone")
1920 ? OFPROTO_FAIL_STANDALONE
1921 : OFPROTO_FAIL_SECURE;
1922 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1923 && !ofproto_has_primary_controller(br->ofproto)) {
1924 ofproto_flush_flows(br->ofproto);
1926 ofproto_set_fail_mode(br->ofproto, fail_mode);
1928 /* Delete all flows if we're switching from connected to standalone or vice
1929 * versa. (XXX Should we delete all flows if we are switching from one
1930 * controller to another?) */
1932 /* Configure OpenFlow controller connection snooping. */
1934 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1935 ovs_rundir(), br->name));
1936 svec_init(&old_snoops);
1937 ofproto_get_snoops(br->ofproto, &old_snoops);
1938 if (!svec_equal(&snoops, &old_snoops)) {
1939 ofproto_set_snoops(br->ofproto, &snoops);
1941 svec_destroy(&snoops);
1942 svec_destroy(&old_snoops);
1944 mirror_reconfigure(br);
1947 /* Initializes 'oc' appropriately as a management service controller for
1950 * The caller must free oc->target when it is no longer needed. */
1952 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1953 struct ofproto_controller *oc)
1955 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1956 oc->max_backoff = 0;
1957 oc->probe_interval = 60;
1958 oc->band = OFPROTO_OUT_OF_BAND;
1960 oc->burst_limit = 0;
1963 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1965 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1966 struct ofproto_controller *oc)
1968 oc->target = c->target;
1969 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1970 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1971 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1972 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1973 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1974 oc->burst_limit = (c->controller_burst_limit
1975 ? *c->controller_burst_limit : 0);
1978 /* Configures the IP stack for 'br''s local interface properly according to the
1979 * configuration in 'c'. */
1981 bridge_configure_local_iface_netdev(struct bridge *br,
1982 struct ovsrec_controller *c)
1984 struct netdev *netdev;
1985 struct in_addr mask, gateway;
1987 struct iface *local_iface;
1990 /* If there's no local interface or no IP address, give up. */
1991 local_iface = bridge_get_local_iface(br);
1992 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1996 /* Bring up the local interface. */
1997 netdev = local_iface->netdev;
1998 netdev_turn_flags_on(netdev, NETDEV_UP, true);
2000 /* Configure the IP address and netmask. */
2001 if (!c->local_netmask
2002 || !inet_aton(c->local_netmask, &mask)
2004 mask.s_addr = guess_netmask(ip.s_addr);
2006 if (!netdev_set_in4(netdev, ip, mask)) {
2007 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
2008 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
2011 /* Configure the default gateway. */
2012 if (c->local_gateway
2013 && inet_aton(c->local_gateway, &gateway)
2014 && gateway.s_addr) {
2015 if (!netdev_add_router(netdev, gateway)) {
2016 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2017 br->name, IP_ARGS(&gateway.s_addr));
2023 bridge_reconfigure_remotes(struct bridge *br,
2024 const struct sockaddr_in *managers,
2027 const char *disable_ib_str, *queue_id_str;
2028 bool disable_in_band = false;
2031 struct ovsrec_controller **controllers;
2032 size_t n_controllers;
2035 struct ofproto_controller *ocs;
2039 /* Check if we should disable in-band control on this bridge. */
2040 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2041 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2042 disable_in_band = true;
2045 /* Set OpenFlow queue ID for in-band control. */
2046 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2047 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2048 ofproto_set_in_band_queue(br->ofproto, queue_id);
2050 if (disable_in_band) {
2051 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2053 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2055 had_primary = ofproto_has_primary_controller(br->ofproto);
2057 n_controllers = bridge_get_controllers(br, &controllers);
2059 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2062 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2063 for (i = 0; i < n_controllers; i++) {
2064 struct ovsrec_controller *c = controllers[i];
2066 if (!strncmp(c->target, "punix:", 6)
2067 || !strncmp(c->target, "unix:", 5)) {
2068 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2070 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2071 * domain sockets and overwriting arbitrary local files. */
2072 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2073 "\"%s\" due to possibility for remote exploit",
2074 dpif_name(br->dpif), c->target);
2078 bridge_configure_local_iface_netdev(br, c);
2079 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2080 if (disable_in_band) {
2081 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2086 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2087 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2090 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2091 ofproto_flush_flows(br->ofproto);
2094 /* If there are no controllers and the bridge is in standalone
2095 * mode, set up a flow that matches every packet and directs
2096 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2097 * switch is in secure mode and we won't pass any traffic until
2098 * a controller has been defined and it tells us to do so. */
2100 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2101 union ofp_action action;
2102 struct cls_rule rule;
2104 memset(&action, 0, sizeof action);
2105 action.type = htons(OFPAT_OUTPUT);
2106 action.output.len = htons(sizeof action);
2107 action.output.port = htons(OFPP_NORMAL);
2108 cls_rule_init_catchall(&rule, 0);
2109 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2114 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2119 for (i = 0; i < br->n_ports; i++) {
2120 struct port *port = br->ports[i];
2121 for (j = 0; j < port->n_ifaces; j++) {
2122 struct iface *iface = port->ifaces[j];
2123 shash_add_once(ifaces, iface->name, iface);
2125 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2126 shash_add_once(ifaces, port->name, NULL);
2131 /* For robustness, in case the administrator moves around datapath ports behind
2132 * our back, we re-check all the datapath port numbers here.
2134 * This function will set the 'dp_ifidx' members of interfaces that have
2135 * disappeared to -1, so only call this function from a context where those
2136 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2137 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2138 * datapath, which doesn't support UINT16_MAX+1 ports. */
2140 bridge_fetch_dp_ifaces(struct bridge *br)
2142 struct dpif_port_dump dump;
2143 struct dpif_port dpif_port;
2146 /* Reset all interface numbers. */
2147 for (i = 0; i < br->n_ports; i++) {
2148 struct port *port = br->ports[i];
2149 for (j = 0; j < port->n_ifaces; j++) {
2150 struct iface *iface = port->ifaces[j];
2151 iface->dp_ifidx = -1;
2154 hmap_clear(&br->ifaces);
2156 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2157 struct iface *iface = iface_lookup(br, dpif_port.name);
2159 if (iface->dp_ifidx >= 0) {
2160 VLOG_WARN("%s reported interface %s twice",
2161 dpif_name(br->dpif), dpif_port.name);
2162 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2163 VLOG_WARN("%s reported interface %"PRIu16" twice",
2164 dpif_name(br->dpif), dpif_port.port_no);
2166 iface->dp_ifidx = dpif_port.port_no;
2167 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2168 hash_int(iface->dp_ifidx, 0));
2171 iface_set_ofport(iface->cfg,
2172 (iface->dp_ifidx >= 0
2173 ? odp_port_to_ofp_port(iface->dp_ifidx)
2179 /* Bridge packet processing functions. */
2182 bond_is_tcp_hash(const struct port *port)
2184 return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED;
2188 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2190 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2193 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2195 struct flow hash_flow;
2197 memcpy(&hash_flow, flow, sizeof hash_flow);
2198 hash_flow.vlan_tci = 0;
2200 /* The symmetric quality of this hash function is not required, but
2201 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2202 * purposes, so we use it out of convenience. */
2203 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2206 static struct bond_entry *
2207 lookup_bond_entry(const struct port *port, const struct flow *flow,
2210 assert(port->bond_mode != BM_AB);
2212 if (bond_is_tcp_hash(port)) {
2213 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2215 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2220 bond_choose_iface(const struct port *port)
2222 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2223 size_t i, best_down_slave = -1;
2224 long long next_delay_expiration = LLONG_MAX;
2226 for (i = 0; i < port->n_ifaces; i++) {
2227 struct iface *iface = port->ifaces[i];
2229 if (iface->enabled) {
2231 } else if (iface->delay_expires < next_delay_expiration
2232 && (iface->lacp_status & LACP_ATTACHED
2233 || !(port->lacp & LACP_NEGOTIATED))) {
2234 best_down_slave = i;
2235 next_delay_expiration = iface->delay_expires;
2239 if (best_down_slave != -1) {
2240 struct iface *iface = port->ifaces[best_down_slave];
2242 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2243 "since no other interface is up", iface->name,
2244 iface->delay_expires - time_msec());
2245 bond_enable_slave(iface, true);
2248 return best_down_slave;
2252 choose_output_iface(const struct port *port, const struct flow *flow,
2253 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2255 struct iface *iface;
2257 assert(port->n_ifaces);
2258 if (port->n_ifaces == 1) {
2259 iface = port->ifaces[0];
2260 } else if (port->bond_mode == BM_AB) {
2261 if (port->active_iface < 0) {
2262 *tags |= port->no_ifaces_tag;
2265 iface = port->ifaces[port->active_iface];
2267 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2268 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2269 || !port->ifaces[e->iface_idx]->enabled) {
2270 /* XXX select interface properly. The current interface selection
2271 * is only good for testing the rebalancing code. */
2272 e->iface_idx = bond_choose_iface(port);
2273 if (e->iface_idx < 0) {
2274 *tags |= port->no_ifaces_tag;
2277 e->iface_tag = tag_create_random();
2279 *tags |= e->iface_tag;
2280 iface = port->ifaces[e->iface_idx];
2282 *dp_ifidx = iface->dp_ifidx;
2283 *tags |= iface->tag; /* Currently only used for bonding. */
2288 bond_link_status_update(struct iface *iface)
2290 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2291 struct port *port = iface->port;
2292 bool up = iface->up;
2293 int updelay, downdelay;
2295 updelay = port->updelay;
2296 downdelay = port->downdelay;
2298 if (iface->port->lacp & LACP_NEGOTIATED) {
2303 if (iface->port->lacp && up) {
2304 /* The interface is up if it's attached to an aggregator and its
2305 * partner is synchronized. The only exception is defaulted links.
2306 * They are not required to have synchronized partners because they
2307 * have no partners at all. However, they will only be attached if
2308 * negotiations failed on all interfaces in the bond. */
2309 up = iface->lacp_status & LACP_ATTACHED
2310 && (iface->lacp_partner.state & LACP_STATE_SYNC
2311 || iface->lacp_status & LACP_DEFAULTED);
2315 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2316 /* Nothing to do. */
2319 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2320 iface->name, up ? "up" : "down");
2321 if (up == iface->enabled) {
2322 iface->delay_expires = LLONG_MAX;
2323 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2324 iface->name, up ? "disabled" : "enabled");
2325 } else if (up && port->active_iface < 0) {
2326 bond_enable_slave(iface, true);
2328 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2329 "other interface is up", iface->name, updelay);
2332 int delay = up ? updelay : downdelay;
2333 iface->delay_expires = time_msec() + delay;
2336 "interface %s: will be %s if it stays %s for %d ms",
2338 up ? "enabled" : "disabled",
2346 bond_choose_active_iface(struct port *port)
2348 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2350 port->active_iface = bond_choose_iface(port);
2351 port->active_iface_tag = tag_create_random();
2352 if (port->active_iface >= 0) {
2353 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2354 port->name, port->ifaces[port->active_iface]->name);
2356 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2362 bond_enable_slave(struct iface *iface, bool enable)
2364 struct port *port = iface->port;
2365 struct bridge *br = port->bridge;
2367 /* This acts as a recursion check. If the act of disabling a slave
2368 * causes a different slave to be enabled, the flag will allow us to
2369 * skip redundant work when we reenter this function. It must be
2370 * cleared on exit to keep things safe with multiple bonds. */
2371 static bool moving_active_iface = false;
2373 iface->delay_expires = LLONG_MAX;
2374 if (enable == iface->enabled) {
2378 iface->enabled = enable;
2379 if (!iface->enabled) {
2380 VLOG_WARN("interface %s: disabled", iface->name);
2381 ofproto_revalidate(br->ofproto, iface->tag);
2382 if (iface->port_ifidx == port->active_iface) {
2383 ofproto_revalidate(br->ofproto,
2384 port->active_iface_tag);
2386 /* Disabling a slave can lead to another slave being immediately
2387 * enabled if there will be no active slaves but one is waiting
2388 * on an updelay. In this case we do not need to run most of the
2389 * code for the newly enabled slave since there was no period
2390 * without an active slave and it is redundant with the disabling
2392 moving_active_iface = true;
2393 bond_choose_active_iface(port);
2395 bond_send_learning_packets(port);
2397 VLOG_WARN("interface %s: enabled", iface->name);
2398 if (port->active_iface < 0 && !moving_active_iface) {
2399 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2400 bond_choose_active_iface(port);
2401 bond_send_learning_packets(port);
2403 iface->tag = tag_create_random();
2406 moving_active_iface = false;
2409 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2410 * bond interface. */
2412 bond_update_fake_iface_stats(struct port *port)
2414 struct netdev_stats bond_stats;
2415 struct netdev *bond_dev;
2418 memset(&bond_stats, 0, sizeof bond_stats);
2420 for (i = 0; i < port->n_ifaces; i++) {
2421 struct netdev_stats slave_stats;
2423 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2424 /* XXX: We swap the stats here because they are swapped back when
2425 * reported by the internal device. The reason for this is
2426 * internal devices normally represent packets going into the system
2427 * but when used as fake bond device they represent packets leaving
2428 * the system. We really should do this in the internal device
2429 * itself because changing it here reverses the counts from the
2430 * perspective of the switch. However, the internal device doesn't
2431 * know what type of device it represents so we have to do it here
2433 bond_stats.tx_packets += slave_stats.rx_packets;
2434 bond_stats.tx_bytes += slave_stats.rx_bytes;
2435 bond_stats.rx_packets += slave_stats.tx_packets;
2436 bond_stats.rx_bytes += slave_stats.tx_bytes;
2440 if (!netdev_open_default(port->name, &bond_dev)) {
2441 netdev_set_stats(bond_dev, &bond_stats);
2442 netdev_close(bond_dev);
2447 bond_link_carrier_update(struct iface *iface, bool carrier)
2449 if (carrier == iface->up) {
2453 if (iface->lacp_status & LACP_CURRENT) {
2454 iface_set_lacp_expired(iface);
2457 iface->up = carrier;
2462 bond_run(struct port *port)
2467 if (port->n_ifaces < 2) {
2471 if (port->monitor) {
2472 assert(!port->miimon);
2474 /* Track carrier going up and down on interfaces. */
2475 while (!netdev_monitor_poll(port->monitor, &devname)) {
2476 struct iface *iface;
2478 iface = port_lookup_iface(port, devname);
2480 bool up = netdev_get_carrier(iface->netdev);
2481 bond_link_carrier_update(iface, up);
2486 assert(port->miimon);
2488 if (time_msec() >= port->bond_miimon_next_update) {
2489 for (i = 0; i < port->n_ifaces; i++) {
2490 struct iface *iface = port->ifaces[i];
2491 bool up = netdev_get_miimon(iface->netdev);
2492 bond_link_carrier_update(iface, up);
2494 port->bond_miimon_next_update = time_msec() +
2495 port->bond_miimon_interval;
2499 for (i = 0; i < port->n_ifaces; i++) {
2500 bond_link_status_update(port->ifaces[i]);
2503 for (i = 0; i < port->n_ifaces; i++) {
2504 struct iface *iface = port->ifaces[i];
2505 if (time_msec() >= iface->delay_expires) {
2506 bond_enable_slave(iface, !iface->enabled);
2510 if (port->bond_fake_iface
2511 && time_msec() >= port->bond_next_fake_iface_update) {
2512 bond_update_fake_iface_stats(port);
2513 port->bond_next_fake_iface_update = time_msec() + 1000;
2518 bond_wait(struct port *port)
2522 if (port->n_ifaces < 2) {
2526 if (port->monitor) {
2527 netdev_monitor_poll_wait(port->monitor);
2531 poll_timer_wait_until(port->bond_miimon_next_update);
2534 for (i = 0; i < port->n_ifaces; i++) {
2535 struct iface *iface = port->ifaces[i];
2536 if (iface->delay_expires != LLONG_MAX) {
2537 poll_timer_wait_until(iface->delay_expires);
2541 if (port->bond_fake_iface) {
2542 poll_timer_wait_until(port->bond_next_fake_iface_update);
2547 set_dst(struct dst *dst, const struct flow *flow,
2548 const struct port *in_port, const struct port *out_port,
2551 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2552 : in_port->vlan >= 0 ? in_port->vlan
2553 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2554 : vlan_tci_to_vid(flow->vlan_tci));
2555 return choose_output_iface(out_port, flow, dst->vlan,
2556 &dst->dp_ifidx, tags);
2560 swap_dst(struct dst *p, struct dst *q)
2562 struct dst tmp = *p;
2567 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2568 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2569 * that we push to the datapath. We could in fact fully sort the array by
2570 * vlan, but in most cases there are at most two different vlan tags so that's
2571 * possibly overkill.) */
2573 partition_dsts(struct dst_set *set, int vlan)
2575 struct dst *first = set->dsts;
2576 struct dst *last = set->dsts + set->n;
2578 while (first != last) {
2580 * - All dsts < first have vlan == 'vlan'.
2581 * - All dsts >= last have vlan != 'vlan'.
2582 * - first < last. */
2583 while (first->vlan == vlan) {
2584 if (++first == last) {
2589 /* Same invariants, plus one additional:
2590 * - first->vlan != vlan.
2592 while (last[-1].vlan != vlan) {
2593 if (--last == first) {
2598 /* Same invariants, plus one additional:
2599 * - last[-1].vlan == vlan.*/
2600 swap_dst(first++, --last);
2605 mirror_mask_ffs(mirror_mask_t mask)
2607 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2612 dst_set_init(struct dst_set *set)
2614 set->dsts = set->builtin;
2616 set->allocated = ARRAY_SIZE(set->builtin);
2620 dst_set_add(struct dst_set *set, const struct dst *dst)
2622 if (set->n >= set->allocated) {
2623 size_t new_allocated;
2624 struct dst *new_dsts;
2626 new_allocated = set->allocated * 2;
2627 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2628 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2632 set->dsts = new_dsts;
2633 set->allocated = new_allocated;
2635 set->dsts[set->n++] = *dst;
2639 dst_set_free(struct dst_set *set)
2641 if (set->dsts != set->builtin) {
2647 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2650 for (i = 0; i < set->n; i++) {
2651 if (set->dsts[i].vlan == test->vlan
2652 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2660 port_trunks_vlan(const struct port *port, uint16_t vlan)
2662 return (port->vlan < 0
2663 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2667 port_includes_vlan(const struct port *port, uint16_t vlan)
2669 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2673 port_is_floodable(const struct port *port)
2677 for (i = 0; i < port->n_ifaces; i++) {
2678 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2679 port->ifaces[i]->dp_ifidx)) {
2687 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2688 const struct port *in_port, const struct port *out_port,
2689 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2691 mirror_mask_t mirrors = in_port->src_mirrors;
2696 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2697 if (flow_vlan == 0) {
2698 flow_vlan = OFP_VLAN_NONE;
2701 if (out_port == FLOOD_PORT) {
2702 for (i = 0; i < br->n_ports; i++) {
2703 struct port *port = br->ports[i];
2705 && port_is_floodable(port)
2706 && port_includes_vlan(port, vlan)
2707 && !port->is_mirror_output_port
2708 && set_dst(&dst, flow, in_port, port, tags)) {
2709 mirrors |= port->dst_mirrors;
2710 dst_set_add(set, &dst);
2713 *nf_output_iface = NF_OUT_FLOOD;
2714 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2715 dst_set_add(set, &dst);
2716 *nf_output_iface = dst.dp_ifidx;
2717 mirrors |= out_port->dst_mirrors;
2721 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2722 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2724 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2725 && !dst_is_duplicate(set, &dst)) {
2726 dst_set_add(set, &dst);
2729 for (i = 0; i < br->n_ports; i++) {
2730 struct port *port = br->ports[i];
2731 if (port_includes_vlan(port, m->out_vlan)
2732 && set_dst(&dst, flow, in_port, port, tags))
2734 if (port->vlan < 0) {
2735 dst.vlan = m->out_vlan;
2737 if (dst_is_duplicate(set, &dst)) {
2741 /* Use the vlan tag on the original flow instead of
2742 * the one passed in the vlan parameter. This ensures
2743 * that we compare the vlan from before any implicit
2744 * tagging tags place. This is necessary because
2745 * dst->vlan is the final vlan, after removing implicit
2747 if (port == in_port && dst.vlan == flow_vlan) {
2748 /* Don't send out input port on same VLAN. */
2751 dst_set_add(set, &dst);
2756 mirrors &= mirrors - 1;
2759 partition_dsts(set, flow_vlan);
2762 static void OVS_UNUSED
2763 print_dsts(const struct dst_set *set)
2767 for (i = 0; i < set->n; i++) {
2768 const struct dst *dst = &set->dsts[i];
2770 printf(">p%"PRIu16, dst->dp_ifidx);
2771 if (dst->vlan != OFP_VLAN_NONE) {
2772 printf("v%"PRIu16, dst->vlan);
2778 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2779 const struct port *in_port, const struct port *out_port,
2780 tag_type *tags, struct ofpbuf *actions,
2781 uint16_t *nf_output_iface)
2788 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2791 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2792 if (cur_vlan == 0) {
2793 cur_vlan = OFP_VLAN_NONE;
2795 for (i = 0; i < set.n; i++) {
2796 const struct dst *dst = &set.dsts[i];
2797 if (dst->vlan != cur_vlan) {
2798 if (dst->vlan == OFP_VLAN_NONE) {
2799 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2802 tci = htons(dst->vlan & VLAN_VID_MASK);
2803 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2804 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2806 cur_vlan = dst->vlan;
2808 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2813 /* Returns the effective vlan of a packet, taking into account both the
2814 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2815 * the packet is untagged and -1 indicates it has an invalid header and
2816 * should be dropped. */
2817 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2818 struct port *in_port, bool have_packet)
2820 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2821 if (in_port->vlan >= 0) {
2823 /* XXX support double tagging? */
2825 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2826 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2827 "packet received on port %s configured with "
2828 "implicit VLAN %"PRIu16,
2829 br->name, vlan, in_port->name, in_port->vlan);
2833 vlan = in_port->vlan;
2835 if (!port_includes_vlan(in_port, vlan)) {
2837 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2838 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2839 "packet received on port %s not configured for "
2841 br->name, vlan, in_port->name, vlan);
2850 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2851 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2852 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2854 is_gratuitous_arp(const struct flow *flow)
2856 return (flow->dl_type == htons(ETH_TYPE_ARP)
2857 && eth_addr_is_broadcast(flow->dl_dst)
2858 && (flow->nw_proto == ARP_OP_REPLY
2859 || (flow->nw_proto == ARP_OP_REQUEST
2860 && flow->nw_src == flow->nw_dst)));
2864 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2865 struct port *in_port)
2867 enum grat_arp_lock_type lock_type;
2870 /* We don't want to learn from gratuitous ARP packets that are reflected
2871 * back over bond slaves so we lock the learning table. */
2872 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2873 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2874 GRAT_ARP_LOCK_CHECK;
2876 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2879 /* The log messages here could actually be useful in debugging,
2880 * so keep the rate limit relatively high. */
2881 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2883 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2884 "on port %s in VLAN %d",
2885 br->name, ETH_ADDR_ARGS(flow->dl_src),
2886 in_port->name, vlan);
2887 ofproto_revalidate(br->ofproto, rev_tag);
2891 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2892 * dropped. Returns true if they may be forwarded, false if they should be
2895 * If 'have_packet' is true, it indicates that the caller is processing a
2896 * received packet. If 'have_packet' is false, then the caller is just
2897 * revalidating an existing flow because configuration has changed. Either
2898 * way, 'have_packet' only affects logging (there is no point in logging errors
2899 * during revalidation).
2901 * Sets '*in_portp' to the input port. This will be a null pointer if
2902 * flow->in_port does not designate a known input port (in which case
2903 * is_admissible() returns false).
2905 * When returning true, sets '*vlanp' to the effective VLAN of the input
2906 * packet, as returned by flow_get_vlan().
2908 * May also add tags to '*tags', although the current implementation only does
2909 * so in one special case.
2912 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2913 tag_type *tags, int *vlanp, struct port **in_portp)
2915 struct iface *in_iface;
2916 struct port *in_port;
2919 /* Find the interface and port structure for the received packet. */
2920 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2922 /* No interface? Something fishy... */
2924 /* Odd. A few possible reasons here:
2926 * - We deleted an interface but there are still a few packets
2927 * queued up from it.
2929 * - Someone externally added an interface (e.g. with "ovs-dpctl
2930 * add-if") that we don't know about.
2932 * - Packet arrived on the local port but the local port is not
2933 * one of our bridge ports.
2935 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2937 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2938 "interface %"PRIu16, br->name, flow->in_port);
2944 *in_portp = in_port = in_iface->port;
2945 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2950 /* Drop frames for reserved multicast addresses. */
2951 if (eth_addr_is_reserved(flow->dl_dst)) {
2955 /* Drop frames on ports reserved for mirroring. */
2956 if (in_port->is_mirror_output_port) {
2958 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2959 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2960 "%s, which is reserved exclusively for mirroring",
2961 br->name, in_port->name);
2966 /* When using LACP, do not accept packets from disabled interfaces. */
2967 if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) {
2971 /* Packets received on non-LACP bonds need special attention to avoid
2973 if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) {
2975 bool is_grat_arp_locked;
2977 if (eth_addr_is_multicast(flow->dl_dst)) {
2978 *tags |= in_port->active_iface_tag;
2979 if (in_port->active_iface != in_iface->port_ifidx) {
2980 /* Drop all multicast packets on inactive slaves. */
2985 /* Drop all packets for which we have learned a different input
2986 * port, because we probably sent the packet on one slave and got
2987 * it back on the other. Gratuitous ARP packets are an exception
2988 * to this rule: the host has moved to another switch. The exception
2989 * to the exception is if we locked the learning table to avoid
2990 * reflections on bond slaves. If this is the case, just drop the
2992 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2993 &is_grat_arp_locked);
2994 if (src_idx != -1 && src_idx != in_port->port_idx &&
2995 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
3003 /* If the composed actions may be applied to any packet in the given 'flow',
3004 * returns true. Otherwise, the actions should only be applied to 'packet', or
3005 * not at all, if 'packet' was NULL. */
3007 process_flow(struct bridge *br, const struct flow *flow,
3008 const struct ofpbuf *packet, struct ofpbuf *actions,
3009 tag_type *tags, uint16_t *nf_output_iface)
3011 struct port *in_port;
3012 struct port *out_port;
3016 /* Check whether we should drop packets in this flow. */
3017 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
3022 /* Learn source MAC (but don't try to learn from revalidation). */
3024 update_learning_table(br, flow, vlan, in_port);
3027 /* Determine output port. */
3028 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
3030 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
3031 out_port = br->ports[out_port_idx];
3032 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
3033 /* If we are revalidating but don't have a learning entry then
3034 * eject the flow. Installing a flow that floods packets opens
3035 * up a window of time where we could learn from a packet reflected
3036 * on a bond and blackhole packets before the learning table is
3037 * updated to reflect the correct port. */
3040 out_port = FLOOD_PORT;
3043 /* Don't send packets out their input ports. */
3044 if (in_port == out_port) {
3050 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3058 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3059 struct ofpbuf *actions, tag_type *tags,
3060 uint16_t *nf_output_iface, void *br_)
3062 struct bridge *br = br_;
3064 COVERAGE_INC(bridge_process_flow);
3065 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3069 bridge_special_ofhook_cb(const struct flow *flow,
3070 const struct ofpbuf *packet, void *br_)
3072 struct iface *iface;
3073 struct bridge *br = br_;
3075 iface = iface_from_dp_ifidx(br, flow->in_port);
3077 if (cfm_should_process_flow(flow)) {
3079 if (iface && packet && iface->cfm) {
3080 COVERAGE_INC(bridge_process_cfm);
3081 cfm_process_heartbeat(iface->cfm, packet);
3084 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3086 if (iface && packet) {
3087 COVERAGE_INC(bridge_process_lacp);
3088 lacp_process_packet(packet, iface);
3097 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3098 const struct nlattr *actions,
3100 uint64_t n_bytes, void *br_)
3102 struct bridge *br = br_;
3103 const struct nlattr *a;
3104 struct port *in_port;
3109 /* Feed information from the active flows back into the learning table to
3110 * ensure that table is always in sync with what is actually flowing
3111 * through the datapath.
3113 * We test that 'tags' is nonzero to ensure that only flows that include an
3114 * OFPP_NORMAL action are used for learning. This works because
3115 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3116 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3117 update_learning_table(br, flow, vlan, in_port);
3120 /* Account for bond slave utilization. */
3121 if (!br->has_bonded_ports) {
3124 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3125 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3126 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3127 if (out_port && out_port->n_ifaces >= 2 &&
3128 out_port->bond_mode != BM_AB) {
3129 uint16_t vlan = (flow->vlan_tci
3130 ? vlan_tci_to_vid(flow->vlan_tci)
3132 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3133 e->tx_bytes += n_bytes;
3140 bridge_account_checkpoint_ofhook_cb(void *br_)
3142 struct bridge *br = br_;
3146 if (!br->has_bonded_ports) {
3151 for (i = 0; i < br->n_ports; i++) {
3152 struct port *port = br->ports[i];
3153 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3154 && now >= port->bond_next_rebalance) {
3155 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3156 bond_rebalance_port(port);
3161 static struct ofhooks bridge_ofhooks = {
3162 bridge_normal_ofhook_cb,
3163 bridge_special_ofhook_cb,
3164 bridge_account_flow_ofhook_cb,
3165 bridge_account_checkpoint_ofhook_cb,
3168 /* LACP functions. */
3171 lacp_process_packet(const struct ofpbuf *packet, struct iface *iface)
3173 const struct lacp_pdu *pdu;
3175 if (!iface->port->lacp) {
3179 pdu = parse_lacp_packet(packet);
3184 iface->lacp_status |= LACP_CURRENT;
3185 iface->lacp_status &= ~(LACP_EXPIRED | LACP_DEFAULTED);
3186 iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX;
3188 iface->lacp_actor.state = iface_get_lacp_state(iface);
3189 if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) {
3193 if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) {
3194 iface->port->lacp_need_update = true;
3195 iface->lacp_partner = pdu->actor;
3200 lacp_update_ifaces(struct port *port)
3204 struct lacp_info lead_pri;
3205 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3207 port->lacp_need_update = false;
3208 COVERAGE_INC(bridge_lacp_update);
3214 VLOG_DBG_RL(&rl, "port %s: re-evaluating LACP link status", port->name);
3217 for (i = 0; i < port->n_ifaces; i++) {
3218 struct iface *iface = port->ifaces[i];
3219 struct lacp_info pri;
3221 iface->lacp_status |= LACP_ATTACHED;
3222 ofproto_revalidate(port->bridge->ofproto, iface->tag);
3224 /* Don't allow loopback interfaces to send traffic or lead. */
3225 if (eth_addr_equals(iface->lacp_partner.sysid,
3226 iface->lacp_actor.sysid)) {
3227 VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is "
3228 "connected to its own bridge", iface->name);
3229 iface->lacp_status &= ~LACP_ATTACHED;
3233 if (iface->lacp_status & LACP_DEFAULTED) {
3237 iface_get_lacp_priority(iface, &pri);
3239 if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) {
3246 port->lacp &= ~LACP_NEGOTIATED;
3250 port->lacp |= LACP_NEGOTIATED;
3252 for (i = 0; i < port->n_ifaces; i++) {
3253 struct iface *iface = port->ifaces[i];
3255 if (iface->lacp_status & LACP_DEFAULTED
3256 || lead->lacp_partner.key != iface->lacp_partner.key
3257 || !eth_addr_equals(lead->lacp_partner.sysid,
3258 iface->lacp_partner.sysid)) {
3259 iface->lacp_status &= ~LACP_ATTACHED;
3265 lacp_iface_may_tx(const struct iface *iface)
3267 return iface->port->lacp & LACP_ACTIVE
3268 || iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED);
3272 lacp_run(struct port *port)
3275 struct ofpbuf packet;
3281 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3283 for (i = 0; i < port->n_ifaces; i++) {
3284 struct iface *iface = port->ifaces[i];
3286 if (time_msec() > iface->lacp_rx) {
3287 if (iface->lacp_status & LACP_CURRENT) {
3288 iface_set_lacp_expired(iface);
3289 } else if (iface->lacp_status & LACP_EXPIRED) {
3290 iface_set_lacp_defaulted(iface);
3295 if (port->lacp_need_update) {
3296 lacp_update_ifaces(port);
3299 for (i = 0; i < port->n_ifaces; i++) {
3300 struct iface *iface = port->ifaces[i];
3301 uint8_t ea[ETH_ADDR_LEN];
3304 if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) {
3308 error = netdev_get_etheraddr(iface->netdev, ea);
3310 iface->lacp_actor.state = iface_get_lacp_state(iface);
3311 compose_lacp_packet(&packet, &iface->lacp_actor,
3312 &iface->lacp_partner, ea);
3313 iface_send_packet(iface, &packet);
3315 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3316 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3317 "(%s)", iface->name, strerror(error));
3320 iface->lacp_tx = time_msec() +
3321 (iface->lacp_partner.state & LACP_STATE_TIME
3323 : LACP_SLOW_TIME_TX);
3325 ofpbuf_uninit(&packet);
3329 lacp_wait(struct port *port)
3337 for (i = 0; i < port->n_ifaces; i++) {
3338 struct iface *iface = port->ifaces[i];
3340 if (lacp_iface_may_tx(iface)) {
3341 poll_timer_wait_until(iface->lacp_tx);
3344 if (iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED)) {
3345 poll_timer_wait_until(iface->lacp_rx);
3350 /* Bonding functions. */
3352 /* Statistics for a single interface on a bonded port, used for load-based
3353 * bond rebalancing. */
3354 struct slave_balance {
3355 struct iface *iface; /* The interface. */
3356 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3358 /* All the "bond_entry"s that are assigned to this interface, in order of
3359 * increasing tx_bytes. */
3360 struct bond_entry **hashes;
3365 bond_mode_to_string(enum bond_mode bm) {
3366 static char *bm_slb = "balance-slb";
3367 static char *bm_ab = "active-backup";
3368 static char *bm_tcp = "balance-tcp";
3371 case BM_SLB: return bm_slb;
3372 case BM_AB: return bm_ab;
3373 case BM_TCP: return bm_tcp;
3380 /* Sorts pointers to pointers to bond_entries in ascending order by the
3381 * interface to which they are assigned, and within a single interface in
3382 * ascending order of bytes transmitted. */
3384 compare_bond_entries(const void *a_, const void *b_)
3386 const struct bond_entry *const *ap = a_;
3387 const struct bond_entry *const *bp = b_;
3388 const struct bond_entry *a = *ap;
3389 const struct bond_entry *b = *bp;
3390 if (a->iface_idx != b->iface_idx) {
3391 return a->iface_idx > b->iface_idx ? 1 : -1;
3392 } else if (a->tx_bytes != b->tx_bytes) {
3393 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3399 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3400 * *descending* order by number of bytes transmitted. */
3402 compare_slave_balance(const void *a_, const void *b_)
3404 const struct slave_balance *a = a_;
3405 const struct slave_balance *b = b_;
3406 if (a->iface->enabled != b->iface->enabled) {
3407 return a->iface->enabled ? -1 : 1;
3408 } else if (a->tx_bytes != b->tx_bytes) {
3409 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3416 swap_bals(struct slave_balance *a, struct slave_balance *b)
3418 struct slave_balance tmp = *a;
3423 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3424 * given that 'p' (and only 'p') might be in the wrong location.
3426 * This function invalidates 'p', since it might now be in a different memory
3429 resort_bals(struct slave_balance *p,
3430 struct slave_balance bals[], size_t n_bals)
3433 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3434 swap_bals(p, p - 1);
3436 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3437 swap_bals(p, p + 1);
3443 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3445 if (VLOG_IS_DBG_ENABLED()) {
3446 struct ds ds = DS_EMPTY_INITIALIZER;
3447 const struct slave_balance *b;
3449 for (b = bals; b < bals + n_bals; b++) {
3453 ds_put_char(&ds, ',');
3455 ds_put_format(&ds, " %s %"PRIu64"kB",
3456 b->iface->name, b->tx_bytes / 1024);
3458 if (!b->iface->enabled) {
3459 ds_put_cstr(&ds, " (disabled)");
3461 if (b->n_hashes > 0) {
3462 ds_put_cstr(&ds, " (");
3463 for (i = 0; i < b->n_hashes; i++) {
3464 const struct bond_entry *e = b->hashes[i];
3466 ds_put_cstr(&ds, " + ");
3468 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3469 e - port->bond_hash, e->tx_bytes / 1024);
3471 ds_put_cstr(&ds, ")");
3474 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3479 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3481 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3484 struct bond_entry *hash = from->hashes[hash_idx];
3485 struct port *port = from->iface->port;
3486 uint64_t delta = hash->tx_bytes;
3488 assert(port->bond_mode != BM_AB);
3490 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3491 "from %s to %s (now carrying %"PRIu64"kB and "
3492 "%"PRIu64"kB load, respectively)",
3493 port->name, delta / 1024, hash - port->bond_hash,
3494 from->iface->name, to->iface->name,
3495 (from->tx_bytes - delta) / 1024,
3496 (to->tx_bytes + delta) / 1024);
3498 /* Delete element from from->hashes.
3500 * We don't bother to add the element to to->hashes because not only would
3501 * it require more work, the only purpose it would be to allow that hash to
3502 * be migrated to another slave in this rebalancing run, and there is no
3503 * point in doing that. */
3504 if (hash_idx == 0) {
3507 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3508 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3512 /* Shift load away from 'from' to 'to'. */
3513 from->tx_bytes -= delta;
3514 to->tx_bytes += delta;
3516 /* Arrange for flows to be revalidated. */
3517 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3518 hash->iface_idx = to->iface->port_ifidx;
3519 hash->iface_tag = tag_create_random();
3523 bond_rebalance_port(struct port *port)
3525 struct slave_balance *bals;
3527 struct bond_entry *hashes[BOND_MASK + 1];
3528 struct slave_balance *b, *from, *to;
3529 struct bond_entry *e;
3532 assert(port->bond_mode != BM_AB);
3534 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3535 * descending order of tx_bytes, so that bals[0] represents the most
3536 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3539 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3540 * array for each slave_balance structure, we sort our local array of
3541 * hashes in order by slave, so that all of the hashes for a given slave
3542 * become contiguous in memory, and then we point each 'hashes' members of
3543 * a slave_balance structure to the start of a contiguous group. */
3544 n_bals = port->n_ifaces;
3545 bals = xmalloc(n_bals * sizeof *bals);
3546 for (b = bals; b < &bals[n_bals]; b++) {
3547 b->iface = port->ifaces[b - bals];
3552 for (i = 0; i <= BOND_MASK; i++) {
3553 hashes[i] = &port->bond_hash[i];
3555 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3556 for (i = 0; i <= BOND_MASK; i++) {
3558 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3559 b = &bals[e->iface_idx];
3560 b->tx_bytes += e->tx_bytes;
3562 b->hashes = &hashes[i];
3567 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3568 log_bals(bals, n_bals, port);
3570 /* Discard slaves that aren't enabled (which were sorted to the back of the
3571 * array earlier). */
3572 while (!bals[n_bals - 1].iface->enabled) {
3579 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3580 to = &bals[n_bals - 1];
3581 for (from = bals; from < to; ) {
3582 uint64_t overload = from->tx_bytes - to->tx_bytes;
3583 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3584 /* The extra load on 'from' (and all less-loaded slaves), compared
3585 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3586 * it is less than ~1Mbps. No point in rebalancing. */
3588 } else if (from->n_hashes == 1) {
3589 /* 'from' only carries a single MAC hash, so we can't shift any
3590 * load away from it, even though we want to. */
3593 /* 'from' is carrying significantly more load than 'to', and that
3594 * load is split across at least two different hashes. Pick a hash
3595 * to migrate to 'to' (the least-loaded slave), given that doing so
3596 * must decrease the ratio of the load on the two slaves by at
3599 * The sort order we use means that we prefer to shift away the
3600 * smallest hashes instead of the biggest ones. There is little
3601 * reason behind this decision; we could use the opposite sort
3602 * order to shift away big hashes ahead of small ones. */
3605 for (i = 0; i < from->n_hashes; i++) {
3606 double old_ratio, new_ratio;
3607 uint64_t delta = from->hashes[i]->tx_bytes;
3609 if (delta == 0 || from->tx_bytes - delta == 0) {
3610 /* Pointless move. */
3614 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3616 if (to->tx_bytes == 0) {
3617 /* Nothing on the new slave, move it. */
3621 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3622 new_ratio = (double)(from->tx_bytes - delta) /
3623 (to->tx_bytes + delta);
3625 if (new_ratio == 0) {
3626 /* Should already be covered but check to prevent division
3631 if (new_ratio < 1) {
3632 new_ratio = 1 / new_ratio;
3635 if (old_ratio - new_ratio > 0.1) {
3636 /* Would decrease the ratio, move it. */
3640 if (i < from->n_hashes) {
3641 bond_shift_load(from, to, i);
3643 /* If the result of the migration changed the relative order of
3644 * 'from' and 'to' swap them back to maintain invariants. */
3645 if (order_swapped) {
3646 swap_bals(from, to);
3649 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3650 * point to different slave_balance structures. It is only
3651 * valid to do these two operations in a row at all because we
3652 * know that 'from' will not move past 'to' and vice versa. */
3653 resort_bals(from, bals, n_bals);
3654 resort_bals(to, bals, n_bals);
3661 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3662 * historical data to decay to <1% in 7 rebalancing runs. */
3663 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3672 bond_send_learning_packets(struct port *port)
3674 struct bridge *br = port->bridge;
3675 struct mac_entry *e;
3676 struct ofpbuf packet;
3677 int error, n_packets, n_errors;
3679 if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
3683 ofpbuf_init(&packet, 128);
3684 error = n_packets = n_errors = 0;
3685 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3686 union ofp_action actions[2], *a;
3692 if (e->port == port->port_idx) {
3696 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3698 flow_extract(&packet, 0, ODPP_NONE, &flow);
3700 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3704 /* Compose actions. */
3705 memset(actions, 0, sizeof actions);
3708 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3709 a->vlan_vid.len = htons(sizeof *a);
3710 a->vlan_vid.vlan_vid = htons(e->vlan);
3713 a->output.type = htons(OFPAT_OUTPUT);
3714 a->output.len = htons(sizeof *a);
3715 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3720 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3727 ofpbuf_uninit(&packet);
3730 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3731 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3732 "packets, last error was: %s",
3733 port->name, n_errors, n_packets, strerror(error));
3735 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3736 port->name, n_packets);
3740 /* Bonding unixctl user interface functions. */
3743 bond_unixctl_list(struct unixctl_conn *conn,
3744 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3746 struct ds ds = DS_EMPTY_INITIALIZER;
3747 const struct bridge *br;
3749 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3751 LIST_FOR_EACH (br, node, &all_bridges) {
3754 for (i = 0; i < br->n_ports; i++) {
3755 const struct port *port = br->ports[i];
3756 if (port->n_ifaces > 1) {
3759 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3760 bond_mode_to_string(port->bond_mode));
3761 for (j = 0; j < port->n_ifaces; j++) {
3762 const struct iface *iface = port->ifaces[j];
3764 ds_put_cstr(&ds, ", ");
3766 ds_put_cstr(&ds, iface->name);
3768 ds_put_char(&ds, '\n');
3772 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3776 static struct port *
3777 bond_find(const char *name)
3779 const struct bridge *br;
3781 LIST_FOR_EACH (br, node, &all_bridges) {
3784 for (i = 0; i < br->n_ports; i++) {
3785 struct port *port = br->ports[i];
3786 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3795 ds_put_lacp_state(struct ds *ds, uint8_t state)
3797 if (state & LACP_STATE_ACT) {
3798 ds_put_cstr(ds, "activity ");
3801 if (state & LACP_STATE_TIME) {
3802 ds_put_cstr(ds, "timeout ");
3805 if (state & LACP_STATE_AGG) {
3806 ds_put_cstr(ds, "aggregation ");
3809 if (state & LACP_STATE_SYNC) {
3810 ds_put_cstr(ds, "synchronized ");
3813 if (state & LACP_STATE_COL) {
3814 ds_put_cstr(ds, "collecting ");
3817 if (state & LACP_STATE_DIST) {
3818 ds_put_cstr(ds, "distributing ");
3821 if (state & LACP_STATE_DEF) {
3822 ds_put_cstr(ds, "defaulted ");
3825 if (state & LACP_STATE_EXP) {
3826 ds_put_cstr(ds, "expired ");
3831 bond_unixctl_show(struct unixctl_conn *conn,
3832 const char *args, void *aux OVS_UNUSED)
3834 struct ds ds = DS_EMPTY_INITIALIZER;
3835 const struct port *port;
3838 port = bond_find(args);
3840 unixctl_command_reply(conn, 501, "no such bond");
3844 ds_put_format(&ds, "bond_mode: %s\n",
3845 bond_mode_to_string(port->bond_mode));
3848 ds_put_format(&ds, "lacp: %s\n",
3849 port->lacp & LACP_ACTIVE ? "active" : "passive");
3851 ds_put_cstr(&ds, "lacp: off\n");
3854 if (port->bond_mode != BM_AB) {
3855 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3856 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3860 ds_put_format(&ds, "bond-detect-mode: %s\n",
3861 port->miimon ? "miimon" : "carrier");
3864 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3865 port->bond_miimon_interval);
3868 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3869 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3871 if (port->bond_mode != BM_AB) {
3872 ds_put_format(&ds, "next rebalance: %lld ms\n",
3873 port->bond_next_rebalance - time_msec());
3876 for (j = 0; j < port->n_ifaces; j++) {
3877 const struct iface *iface = port->ifaces[j];
3878 struct bond_entry *be;
3882 ds_put_format(&ds, "\nslave %s: %s\n",
3883 iface->name, iface->enabled ? "enabled" : "disabled");
3884 if (j == port->active_iface) {
3885 ds_put_cstr(&ds, "\tactive slave\n");
3887 if (iface->delay_expires != LLONG_MAX) {
3888 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3889 iface->enabled ? "downdelay" : "updelay",
3890 iface->delay_expires - time_msec());
3894 ds_put_cstr(&ds, "\tstatus: ");
3896 if (iface->lacp_status & LACP_CURRENT) {
3897 ds_put_cstr(&ds, "current ");
3900 if (iface->lacp_status & LACP_EXPIRED) {
3901 ds_put_cstr(&ds, "expired ");
3904 if (iface->lacp_status & LACP_DEFAULTED) {
3905 ds_put_cstr(&ds, "defaulted ");
3908 if (iface->lacp_status & LACP_ATTACHED) {
3909 ds_put_cstr(&ds, "attached ");
3912 ds_put_cstr(&ds, "\n");
3914 ds_put_cstr(&ds, "\n\tactor sysid: ");
3915 ds_put_format(&ds, ETH_ADDR_FMT,
3916 ETH_ADDR_ARGS(iface->lacp_actor.sysid));
3917 ds_put_cstr(&ds, "\n");
3919 ds_put_format(&ds, "\tactor sys_priority: %u\n",
3920 ntohs(iface->lacp_actor.sys_priority));
3922 ds_put_format(&ds, "\tactor portid: %u\n",
3923 ntohs(iface->lacp_actor.portid));
3925 ds_put_format(&ds, "\tactor port_priority: %u\n",
3926 ntohs(iface->lacp_actor.port_priority));
3928 ds_put_format(&ds, "\tactor key: %u\n",
3929 ntohs(iface->lacp_actor.key));
3931 ds_put_cstr(&ds, "\tactor state: ");
3932 ds_put_lacp_state(&ds, iface_get_lacp_state(iface));
3933 ds_put_cstr(&ds, "\n\n");
3935 ds_put_cstr(&ds, "\tpartner sysid: ");
3936 ds_put_format(&ds, ETH_ADDR_FMT,
3937 ETH_ADDR_ARGS(iface->lacp_partner.sysid));
3938 ds_put_cstr(&ds, "\n");
3940 ds_put_format(&ds, "\tpartner sys_priority: %u\n",
3941 ntohs(iface->lacp_partner.sys_priority));
3943 ds_put_format(&ds, "\tpartner portid: %u\n",
3944 ntohs(iface->lacp_partner.portid));
3946 ds_put_format(&ds, "\tpartner port_priority: %u\n",
3947 ntohs(iface->lacp_partner.port_priority));
3949 ds_put_format(&ds, "\tpartner key: %u\n",
3950 ntohs(iface->lacp_partner.key));
3952 ds_put_cstr(&ds, "\tpartner state: ");
3953 ds_put_lacp_state(&ds, iface->lacp_partner.state);
3954 ds_put_cstr(&ds, "\n");
3957 if (port->bond_mode == BM_AB) {
3962 memset(&flow, 0, sizeof flow);
3963 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3964 int hash = be - port->bond_hash;
3965 struct mac_entry *me;
3967 if (be->iface_idx != j) {
3971 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3972 hash, be->tx_bytes / 1024);
3974 if (port->bond_mode != BM_SLB) {
3979 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3983 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3984 if (bond_hash_src(me->mac, me->vlan) == hash
3985 && me->port != port->port_idx
3986 && choose_output_iface(port, &flow, me->vlan,
3988 && dp_ifidx == iface->dp_ifidx)
3990 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3991 ETH_ADDR_ARGS(me->mac));
3996 unixctl_command_reply(conn, 200, ds_cstr(&ds));
4001 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
4002 void *aux OVS_UNUSED)
4004 char *args = (char *) args_;
4005 char *save_ptr = NULL;
4006 char *bond_s, *hash_s, *slave_s;
4008 struct iface *iface;
4009 struct bond_entry *entry;
4012 bond_s = strtok_r(args, " ", &save_ptr);
4013 hash_s = strtok_r(NULL, " ", &save_ptr);
4014 slave_s = strtok_r(NULL, " ", &save_ptr);
4016 unixctl_command_reply(conn, 501,
4017 "usage: bond/migrate BOND HASH SLAVE");
4021 port = bond_find(bond_s);
4023 unixctl_command_reply(conn, 501, "no such bond");
4027 if (port->bond_mode != BM_SLB) {
4028 unixctl_command_reply(conn, 501, "not an SLB bond");
4032 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
4033 hash = atoi(hash_s) & BOND_MASK;
4035 unixctl_command_reply(conn, 501, "bad hash");
4039 iface = port_lookup_iface(port, slave_s);
4041 unixctl_command_reply(conn, 501, "no such slave");
4045 if (!iface->enabled) {
4046 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
4050 entry = &port->bond_hash[hash];
4051 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
4052 entry->iface_idx = iface->port_ifidx;
4053 entry->iface_tag = tag_create_random();
4054 unixctl_command_reply(conn, 200, "migrated");
4058 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
4059 void *aux OVS_UNUSED)
4061 char *args = (char *) args_;
4062 char *save_ptr = NULL;
4063 char *bond_s, *slave_s;
4065 struct iface *iface;
4067 bond_s = strtok_r(args, " ", &save_ptr);
4068 slave_s = strtok_r(NULL, " ", &save_ptr);
4070 unixctl_command_reply(conn, 501,
4071 "usage: bond/set-active-slave BOND SLAVE");
4075 port = bond_find(bond_s);
4077 unixctl_command_reply(conn, 501, "no such bond");
4081 iface = port_lookup_iface(port, slave_s);
4083 unixctl_command_reply(conn, 501, "no such slave");
4087 if (!iface->enabled) {
4088 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
4092 if (port->active_iface != iface->port_ifidx) {
4093 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4094 port->active_iface = iface->port_ifidx;
4095 port->active_iface_tag = tag_create_random();
4096 VLOG_INFO("port %s: active interface is now %s",
4097 port->name, iface->name);
4098 bond_send_learning_packets(port);
4099 unixctl_command_reply(conn, 200, "done");
4101 unixctl_command_reply(conn, 200, "no change");
4106 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
4108 char *args = (char *) args_;
4109 char *save_ptr = NULL;
4110 char *bond_s, *slave_s;
4112 struct iface *iface;
4114 bond_s = strtok_r(args, " ", &save_ptr);
4115 slave_s = strtok_r(NULL, " ", &save_ptr);
4117 unixctl_command_reply(conn, 501,
4118 "usage: bond/enable/disable-slave BOND SLAVE");
4122 port = bond_find(bond_s);
4124 unixctl_command_reply(conn, 501, "no such bond");
4128 iface = port_lookup_iface(port, slave_s);
4130 unixctl_command_reply(conn, 501, "no such slave");
4134 bond_enable_slave(iface, enable);
4135 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
4139 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
4140 void *aux OVS_UNUSED)
4142 enable_slave(conn, args, true);
4146 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
4147 void *aux OVS_UNUSED)
4149 enable_slave(conn, args, false);
4153 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
4154 void *aux OVS_UNUSED)
4156 char *args = (char *) args_;
4157 uint8_t mac[ETH_ADDR_LEN];
4161 char *mac_s, *vlan_s;
4162 char *save_ptr = NULL;
4164 mac_s = strtok_r(args, " ", &save_ptr);
4165 vlan_s = strtok_r(NULL, " ", &save_ptr);
4168 if (sscanf(vlan_s, "%u", &vlan) != 1) {
4169 unixctl_command_reply(conn, 501, "invalid vlan");
4173 vlan = OFP_VLAN_NONE;
4176 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
4177 == ETH_ADDR_SCAN_COUNT) {
4178 hash = bond_hash_src(mac, vlan);
4180 hash_cstr = xasprintf("%u", hash);
4181 unixctl_command_reply(conn, 200, hash_cstr);
4184 unixctl_command_reply(conn, 501, "invalid mac");
4191 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
4192 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
4193 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
4194 unixctl_command_register("bond/set-active-slave",
4195 bond_unixctl_set_active_slave, NULL);
4196 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
4198 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
4200 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
4203 /* Port functions. */
4206 port_run(struct port *port)
4213 for (i = 0; i < port->n_ifaces; i++) {
4214 struct iface *iface = port->ifaces[i];
4217 struct ofpbuf *packet = cfm_run(iface->cfm);
4219 iface_send_packet(iface, packet);
4220 ofpbuf_uninit(packet);
4228 port_wait(struct port *port)
4235 for (i = 0; i < port->n_ifaces; i++) {
4236 struct iface *iface = port->ifaces[i];
4238 cfm_wait(iface->cfm);
4243 static struct port *
4244 port_create(struct bridge *br, const char *name)
4248 port = xzalloc(sizeof *port);
4250 port->port_idx = br->n_ports;
4252 port->trunks = NULL;
4253 port->name = xstrdup(name);
4254 port->active_iface = -1;
4256 if (br->n_ports >= br->allocated_ports) {
4257 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
4260 br->ports[br->n_ports++] = port;
4261 shash_add_assert(&br->port_by_name, port->name, port);
4263 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
4270 get_port_other_config(const struct ovsrec_port *port, const char *key,
4271 const char *default_value)
4275 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
4277 return value ? value : default_value;
4281 get_interface_other_config(const struct ovsrec_interface *iface,
4282 const char *key, const char *default_value)
4286 value = get_ovsrec_key_value(&iface->header_,
4287 &ovsrec_interface_col_other_config, key);
4288 return value ? value : default_value;
4292 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
4294 struct shash new_ifaces;
4297 /* Collect list of new interfaces. */
4298 shash_init(&new_ifaces);
4299 for (i = 0; i < cfg->n_interfaces; i++) {
4300 const char *name = cfg->interfaces[i]->name;
4301 shash_add_once(&new_ifaces, name, NULL);
4304 /* Get rid of deleted interfaces. */
4305 for (i = 0; i < port->n_ifaces; ) {
4306 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
4307 iface_destroy(port->ifaces[i]);
4313 shash_destroy(&new_ifaces);
4317 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4319 const char *detect_mode;
4320 struct shash new_ifaces;
4321 long long int next_rebalance, miimon_next_update, lacp_priority;
4322 unsigned long *trunks;
4328 /* Update settings. */
4329 port->updelay = cfg->bond_updelay;
4330 if (port->updelay < 0) {
4333 port->downdelay = cfg->bond_downdelay;
4334 if (port->downdelay < 0) {
4335 port->downdelay = 0;
4337 port->bond_rebalance_interval = atoi(
4338 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4339 if (port->bond_rebalance_interval < 1000) {
4340 port->bond_rebalance_interval = 1000;
4342 next_rebalance = time_msec() + port->bond_rebalance_interval;
4343 if (port->bond_next_rebalance > next_rebalance) {
4344 port->bond_next_rebalance = next_rebalance;
4347 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4350 if (!strcmp(detect_mode, "carrier")) {
4351 port->miimon = false;
4352 } else if (!strcmp(detect_mode, "miimon")) {
4353 port->miimon = true;
4355 port->miimon = false;
4356 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
4357 "carrier", port->name, detect_mode);
4360 port->bond_miimon_interval = atoi(
4361 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4362 if (port->bond_miimon_interval < 100) {
4363 port->bond_miimon_interval = 100;
4365 miimon_next_update = time_msec() + port->bond_miimon_interval;
4366 if (port->bond_miimon_next_update > miimon_next_update) {
4367 port->bond_miimon_next_update = miimon_next_update;
4370 if (!port->cfg->bond_mode ||
4371 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4372 port->bond_mode = BM_SLB;
4373 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4374 port->bond_mode = BM_AB;
4375 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4376 port->bond_mode = BM_TCP;
4378 port->bond_mode = BM_SLB;
4379 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4380 port->name, port->cfg->bond_mode,
4381 bond_mode_to_string(port->bond_mode));
4384 /* Add new interfaces and update 'cfg' member of existing ones. */
4385 shash_init(&new_ifaces);
4386 for (i = 0; i < cfg->n_interfaces; i++) {
4387 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4388 struct iface *iface;
4390 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4391 VLOG_WARN("port %s: %s specified twice as port interface",
4392 port->name, if_cfg->name);
4393 iface_set_ofport(if_cfg, -1);
4397 iface = iface_lookup(port->bridge, if_cfg->name);
4399 if (iface->port != port) {
4400 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4402 port->bridge->name, if_cfg->name, iface->port->name);
4405 iface->cfg = if_cfg;
4407 iface = iface_create(port, if_cfg);
4410 /* Determine interface type. The local port always has type
4411 * "internal". Other ports take their type from the database and
4412 * default to "system" if none is specified. */
4413 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4414 : if_cfg->type[0] ? if_cfg->type
4418 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4421 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4422 iface->lacp_priority = UINT16_MAX;
4424 iface->lacp_priority = lacp_priority;
4427 shash_destroy(&new_ifaces);
4430 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4432 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4433 /* Prefer bondable links if unspecified. */
4434 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4436 port->lacp_priority = lacp_priority;
4439 if (!port->cfg->lacp) {
4440 /* XXX when LACP implementation has been sufficiently tested, enable by
4441 * default and make active on bonded ports. */
4443 } else if (!strcmp(port->cfg->lacp, "off")) {
4445 } else if (!strcmp(port->cfg->lacp, "active")) {
4446 port->lacp = LACP_ACTIVE;
4447 } else if (!strcmp(port->cfg->lacp, "passive")) {
4448 port->lacp = LACP_PASSIVE;
4450 VLOG_WARN("port %s: unknown LACP mode %s",
4451 port->name, port->cfg->lacp);
4458 if (port->n_ifaces < 2) {
4460 if (vlan >= 0 && vlan <= 4095) {
4461 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4466 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4467 * they even work as-is. But they have not been tested. */
4468 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4472 if (port->vlan != vlan) {
4474 bridge_flush(port->bridge);
4477 /* Get trunked VLANs. */
4479 if (vlan < 0 && cfg->n_trunks) {
4482 trunks = bitmap_allocate(4096);
4484 for (i = 0; i < cfg->n_trunks; i++) {
4485 int trunk = cfg->trunks[i];
4487 bitmap_set1(trunks, trunk);
4493 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4494 port->name, cfg->n_trunks);
4496 if (n_errors == cfg->n_trunks) {
4497 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4499 bitmap_free(trunks);
4502 } else if (vlan >= 0 && cfg->n_trunks) {
4503 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4507 ? port->trunks != NULL
4508 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4509 bridge_flush(port->bridge);
4511 bitmap_free(port->trunks);
4512 port->trunks = trunks;
4516 port_destroy(struct port *port)
4519 struct bridge *br = port->bridge;
4523 for (i = 0; i < MAX_MIRRORS; i++) {
4524 struct mirror *m = br->mirrors[i];
4525 if (m && m->out_port == port) {
4530 while (port->n_ifaces > 0) {
4531 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4534 shash_find_and_delete_assert(&br->port_by_name, port->name);
4536 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4537 del->port_idx = port->port_idx;
4539 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4541 netdev_monitor_destroy(port->monitor);
4543 bitmap_free(port->trunks);
4550 static struct port *
4551 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4553 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4554 return iface ? iface->port : NULL;
4557 static struct port *
4558 port_lookup(const struct bridge *br, const char *name)
4560 return shash_find_data(&br->port_by_name, name);
4563 static struct iface *
4564 port_lookup_iface(const struct port *port, const char *name)
4566 struct iface *iface = iface_lookup(port->bridge, name);
4567 return iface && iface->port == port ? iface : NULL;
4571 port_update_lacp(struct port *port)
4576 if (!port->lacp || port->n_ifaces < 1) {
4577 for (i = 0; i < port->n_ifaces; i++) {
4578 iface_set_lacp_defaulted(port->ifaces[i]);
4584 for (i = 0; i < port->n_ifaces; i++) {
4585 struct iface *iface = port->ifaces[i];
4587 if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) {
4592 if (iface->dp_ifidx == port->lacp_key) {
4593 key_changed = false;
4598 port->lacp_key = port->ifaces[0]->dp_ifidx;
4601 for (i = 0; i < port->n_ifaces; i++) {
4602 struct iface *iface = port->ifaces[i];
4604 iface->lacp_actor.sys_priority = htons(port->lacp_priority);
4605 memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN);
4607 iface->lacp_actor.port_priority = htons(iface->lacp_priority);
4608 iface->lacp_actor.portid = htons(iface->dp_ifidx);
4609 iface->lacp_actor.key = htons(port->lacp_key);
4613 port->lacp_need_update = true;
4617 port_update_bonding(struct port *port)
4619 if (port->monitor) {
4620 netdev_monitor_destroy(port->monitor);
4621 port->monitor = NULL;
4623 if (port->n_ifaces < 2) {
4624 /* Not a bonded port. */
4625 free(port->bond_hash);
4626 port->bond_hash = NULL;
4627 port->bond_fake_iface = false;
4628 port->active_iface = -1;
4629 port->no_ifaces_tag = 0;
4633 if (port->bond_mode != BM_AB && !port->bond_hash) {
4634 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4635 for (i = 0; i <= BOND_MASK; i++) {
4636 struct bond_entry *e = &port->bond_hash[i];
4640 port->bond_next_rebalance
4641 = time_msec() + port->bond_rebalance_interval;
4642 } else if (port->bond_mode == BM_AB) {
4643 free(port->bond_hash);
4644 port->bond_hash = NULL;
4647 if (!port->no_ifaces_tag) {
4648 port->no_ifaces_tag = tag_create_random();
4651 if (port->active_iface < 0) {
4652 bond_choose_active_iface(port);
4655 port->bond_fake_iface = port->cfg->bond_fake_iface;
4656 if (port->bond_fake_iface) {
4657 port->bond_next_fake_iface_update = time_msec();
4660 if (!port->miimon) {
4661 port->monitor = netdev_monitor_create();
4662 for (i = 0; i < port->n_ifaces; i++) {
4663 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4669 /* Interface functions. */
4672 iface_set_lacp_defaulted(struct iface *iface)
4674 memset(&iface->lacp_partner, 0, sizeof iface->lacp_partner);
4676 iface->lacp_status |= LACP_DEFAULTED;
4677 iface->lacp_status &= ~(LACP_CURRENT | LACP_EXPIRED);
4679 iface->port->lacp_need_update = true;
4683 iface_set_lacp_expired(struct iface *iface)
4685 iface->lacp_status &= ~LACP_CURRENT;
4686 iface->lacp_status |= LACP_EXPIRED;
4687 iface->lacp_partner.state |= LACP_STATE_TIME;
4688 iface->lacp_partner.state &= ~LACP_STATE_SYNC;
4690 iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX;
4695 iface_get_lacp_state(const struct iface *iface)
4699 if (iface->port->lacp & LACP_ACTIVE) {
4700 state |= LACP_STATE_ACT;
4703 if (iface->lacp_status & LACP_ATTACHED) {
4704 state |= LACP_STATE_SYNC;
4707 if (iface->lacp_status & LACP_DEFAULTED) {
4708 state |= LACP_STATE_DEF;
4711 if (iface->lacp_status & LACP_EXPIRED) {
4712 state |= LACP_STATE_EXP;
4715 if (iface->port->n_ifaces > 1) {
4716 state |= LACP_STATE_AGG;
4719 if (iface->enabled) {
4720 state |= LACP_STATE_COL | LACP_STATE_DIST;
4726 /* Given 'iface', populates 'priority' with data representing its LACP link
4727 * priority. If two priority objects populated by this function are compared
4728 * using memcmp, the higher priority link will be less than the lower priority
4731 iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority)
4733 uint16_t partner_priority, actor_priority;
4735 /* Choose the lacp_info of the higher priority system by comparing their
4736 * system priorities and mac addresses. */
4737 actor_priority = ntohs(iface->lacp_actor.sys_priority);
4738 partner_priority = ntohs(iface->lacp_partner.sys_priority);
4739 if (actor_priority < partner_priority) {
4740 *priority = iface->lacp_actor;
4741 } else if (partner_priority < actor_priority) {
4742 *priority = iface->lacp_partner;
4743 } else if (eth_addr_compare_3way(iface->lacp_actor.sysid,
4744 iface->lacp_partner.sysid) < 0) {
4745 *priority = iface->lacp_actor;
4747 *priority = iface->lacp_partner;
4750 /* Key and state are not used in priority comparisons. */
4752 priority->state = 0;
4756 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4759 union ofp_action action;
4761 memset(&action, 0, sizeof action);
4762 action.output.type = htons(OFPAT_OUTPUT);
4763 action.output.len = htons(sizeof action);
4764 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4766 flow_extract(packet, 0, ODPP_NONE, &flow);
4768 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4770 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4771 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4775 static struct iface *
4776 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4778 struct bridge *br = port->bridge;
4779 struct iface *iface;
4780 char *name = if_cfg->name;
4782 iface = xzalloc(sizeof *iface);
4784 iface->port_ifidx = port->n_ifaces;
4785 iface->name = xstrdup(name);
4786 iface->dp_ifidx = -1;
4787 iface->tag = tag_create_random();
4788 iface->delay_expires = LLONG_MAX;
4789 iface->netdev = NULL;
4790 iface->cfg = if_cfg;
4791 iface_set_lacp_defaulted(iface);
4793 if (port->lacp & LACP_ACTIVE) {
4794 iface_set_lacp_expired(iface);
4797 shash_add_assert(&br->iface_by_name, iface->name, iface);
4799 if (port->n_ifaces >= port->allocated_ifaces) {
4800 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4801 sizeof *port->ifaces);
4803 port->ifaces[port->n_ifaces++] = iface;
4804 if (port->n_ifaces > 1) {
4805 br->has_bonded_ports = true;
4808 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4816 iface_destroy(struct iface *iface)
4819 struct port *port = iface->port;
4820 struct bridge *br = port->bridge;
4821 bool del_active = port->active_iface == iface->port_ifidx;
4824 if (port->monitor) {
4825 netdev_monitor_remove(port->monitor, iface->netdev);
4828 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4830 if (iface->dp_ifidx >= 0) {
4831 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4834 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4835 del->port_ifidx = iface->port_ifidx;
4837 netdev_close(iface->netdev);
4840 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4841 bond_choose_active_iface(port);
4842 bond_send_learning_packets(port);
4845 cfm_destroy(iface->cfm);
4850 bridge_flush(port->bridge);
4854 static struct iface *
4855 iface_lookup(const struct bridge *br, const char *name)
4857 return shash_find_data(&br->iface_by_name, name);
4860 static struct iface *
4861 iface_find(const char *name)
4863 const struct bridge *br;
4865 LIST_FOR_EACH (br, node, &all_bridges) {
4866 struct iface *iface = iface_lookup(br, name);
4875 static struct iface *
4876 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4878 struct iface *iface;
4880 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4881 hash_int(dp_ifidx, 0), &br->ifaces) {
4882 if (iface->dp_ifidx == dp_ifidx) {
4889 /* Set Ethernet address of 'iface', if one is specified in the configuration
4892 iface_set_mac(struct iface *iface)
4894 uint8_t ea[ETH_ADDR_LEN];
4896 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4897 if (eth_addr_is_multicast(ea)) {
4898 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4900 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4901 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4902 iface->name, iface->name);
4904 int error = netdev_set_etheraddr(iface->netdev, ea);
4906 VLOG_ERR("interface %s: setting MAC failed (%s)",
4907 iface->name, strerror(error));
4913 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4915 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4918 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4922 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4924 * The value strings in '*shash' are taken directly from values[], not copied,
4925 * so the caller should not modify or free them. */
4927 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4928 struct shash *shash)
4933 for (i = 0; i < n; i++) {
4934 shash_add(shash, keys[i], values[i]);
4938 /* Creates 'keys' and 'values' arrays from 'shash'.
4940 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4941 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4942 * are populated with with strings taken directly from 'shash' and thus have
4943 * the same ownership of the key-value pairs in shash.
4946 shash_to_ovs_idl_map(struct shash *shash,
4947 char ***keys, char ***values, size_t *n)
4951 struct shash_node *sn;
4953 count = shash_count(shash);
4955 k = xmalloc(count * sizeof *k);
4956 v = xmalloc(count * sizeof *v);
4959 SHASH_FOR_EACH(sn, shash) {
4970 struct iface_delete_queues_cbdata {
4971 struct netdev *netdev;
4972 const struct ovsdb_datum *queues;
4976 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4978 union ovsdb_atom atom;
4980 atom.integer = target;
4981 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4985 iface_delete_queues(unsigned int queue_id,
4986 const struct shash *details OVS_UNUSED, void *cbdata_)
4988 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4990 if (!queue_ids_include(cbdata->queues, queue_id)) {
4991 netdev_delete_queue(cbdata->netdev, queue_id);
4996 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4998 if (!qos || qos->type[0] == '\0') {
4999 netdev_set_qos(iface->netdev, NULL, NULL);
5001 struct iface_delete_queues_cbdata cbdata;
5002 struct shash details;
5005 /* Configure top-level Qos for 'iface'. */
5006 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
5007 qos->n_other_config, &details);
5008 netdev_set_qos(iface->netdev, qos->type, &details);
5009 shash_destroy(&details);
5011 /* Deconfigure queues that were deleted. */
5012 cbdata.netdev = iface->netdev;
5013 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
5015 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
5017 /* Configure queues for 'iface'. */
5018 for (i = 0; i < qos->n_queues; i++) {
5019 const struct ovsrec_queue *queue = qos->value_queues[i];
5020 unsigned int queue_id = qos->key_queues[i];
5022 shash_from_ovs_idl_map(queue->key_other_config,
5023 queue->value_other_config,
5024 queue->n_other_config, &details);
5025 netdev_set_queue(iface->netdev, queue_id, &details);
5026 shash_destroy(&details);
5032 iface_update_cfm(struct iface *iface)
5036 uint16_t *remote_mps;
5037 struct ovsrec_monitor *mon;
5038 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
5040 mon = iface->cfg->monitor;
5043 cfm_destroy(iface->cfm);
5048 if (netdev_get_etheraddr(iface->netdev, ea)) {
5049 VLOG_WARN("interface %s: Failed to get ethernet address. "
5050 "Skipping Monitor.", iface->name);
5054 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
5055 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
5060 iface->cfm = cfm_create();
5064 cfm->mpid = mon->mpid;
5065 cfm->interval = mon->interval ? *mon->interval : 1000;
5067 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
5068 memcpy(cfm->maid, maid, sizeof cfm->maid);
5070 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
5071 for(i = 0; i < mon->n_remote_mps; i++) {
5072 remote_mps[i] = mon->remote_mps[i]->mpid;
5074 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
5077 if (!cfm_configure(iface->cfm)) {
5078 cfm_destroy(iface->cfm);
5083 /* Port mirroring. */
5085 static struct mirror *
5086 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
5090 for (i = 0; i < MAX_MIRRORS; i++) {
5091 struct mirror *m = br->mirrors[i];
5092 if (m && uuid_equals(uuid, &m->uuid)) {
5100 mirror_reconfigure(struct bridge *br)
5102 unsigned long *rspan_vlans;
5105 /* Get rid of deleted mirrors. */
5106 for (i = 0; i < MAX_MIRRORS; i++) {
5107 struct mirror *m = br->mirrors[i];
5109 const struct ovsdb_datum *mc;
5110 union ovsdb_atom atom;
5112 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
5113 atom.uuid = br->mirrors[i]->uuid;
5114 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
5120 /* Add new mirrors and reconfigure existing ones. */
5121 for (i = 0; i < br->cfg->n_mirrors; i++) {
5122 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
5123 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
5125 mirror_reconfigure_one(m, cfg);
5127 mirror_create(br, cfg);
5131 /* Update port reserved status. */
5132 for (i = 0; i < br->n_ports; i++) {
5133 br->ports[i]->is_mirror_output_port = false;
5135 for (i = 0; i < MAX_MIRRORS; i++) {
5136 struct mirror *m = br->mirrors[i];
5137 if (m && m->out_port) {
5138 m->out_port->is_mirror_output_port = true;
5142 /* Update flooded vlans (for RSPAN). */
5144 if (br->cfg->n_flood_vlans) {
5145 rspan_vlans = bitmap_allocate(4096);
5147 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
5148 int64_t vlan = br->cfg->flood_vlans[i];
5149 if (vlan >= 0 && vlan < 4096) {
5150 bitmap_set1(rspan_vlans, vlan);
5151 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
5154 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
5159 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
5165 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
5170 for (i = 0; ; i++) {
5171 if (i >= MAX_MIRRORS) {
5172 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
5173 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
5176 if (!br->mirrors[i]) {
5181 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
5184 br->mirrors[i] = m = xzalloc(sizeof *m);
5187 m->name = xstrdup(cfg->name);
5188 shash_init(&m->src_ports);
5189 shash_init(&m->dst_ports);
5195 mirror_reconfigure_one(m, cfg);
5199 mirror_destroy(struct mirror *m)
5202 struct bridge *br = m->bridge;
5205 for (i = 0; i < br->n_ports; i++) {
5206 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5207 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5210 shash_destroy(&m->src_ports);
5211 shash_destroy(&m->dst_ports);
5214 m->bridge->mirrors[m->idx] = NULL;
5223 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
5224 struct shash *names)
5228 for (i = 0; i < n_ports; i++) {
5229 const char *name = ports[i]->name;
5230 if (port_lookup(m->bridge, name)) {
5231 shash_add_once(names, name, NULL);
5233 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
5234 "port %s", m->bridge->name, m->name, name);
5240 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
5246 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
5248 for (i = 0; i < cfg->n_select_vlan; i++) {
5249 int64_t vlan = cfg->select_vlan[i];
5250 if (vlan < 0 || vlan > 4095) {
5251 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
5252 m->bridge->name, m->name, vlan);
5254 (*vlans)[n_vlans++] = vlan;
5261 vlan_is_mirrored(const struct mirror *m, int vlan)
5265 for (i = 0; i < m->n_vlans; i++) {
5266 if (m->vlans[i] == vlan) {
5274 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
5278 for (i = 0; i < m->n_vlans; i++) {
5279 if (port_trunks_vlan(p, m->vlans[i])) {
5287 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
5289 struct shash src_ports, dst_ports;
5290 mirror_mask_t mirror_bit;
5291 struct port *out_port;
5298 if (strcmp(cfg->name, m->name)) {
5300 m->name = xstrdup(cfg->name);
5303 /* Get output port. */
5304 if (cfg->output_port) {
5305 out_port = port_lookup(m->bridge, cfg->output_port->name);
5307 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
5308 m->bridge->name, m->name);
5314 if (cfg->output_vlan) {
5315 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
5316 "output vlan; ignoring output vlan",
5317 m->bridge->name, m->name);
5319 } else if (cfg->output_vlan) {
5321 out_vlan = *cfg->output_vlan;
5323 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
5324 m->bridge->name, m->name);
5329 shash_init(&src_ports);
5330 shash_init(&dst_ports);
5331 if (cfg->select_all) {
5332 for (i = 0; i < m->bridge->n_ports; i++) {
5333 const char *name = m->bridge->ports[i]->name;
5334 shash_add_once(&src_ports, name, NULL);
5335 shash_add_once(&dst_ports, name, NULL);
5340 /* Get ports, and drop duplicates and ports that don't exist. */
5341 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
5343 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
5346 /* Get all the vlans, and drop duplicate and invalid vlans. */
5347 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
5350 /* Update mirror data. */
5351 if (!shash_equal_keys(&m->src_ports, &src_ports)
5352 || !shash_equal_keys(&m->dst_ports, &dst_ports)
5353 || m->n_vlans != n_vlans
5354 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
5355 || m->out_port != out_port
5356 || m->out_vlan != out_vlan) {
5357 bridge_flush(m->bridge);
5359 shash_swap(&m->src_ports, &src_ports);
5360 shash_swap(&m->dst_ports, &dst_ports);
5363 m->n_vlans = n_vlans;
5364 m->out_port = out_port;
5365 m->out_vlan = out_vlan;
5368 mirror_bit = MIRROR_MASK_C(1) << m->idx;
5369 for (i = 0; i < m->bridge->n_ports; i++) {
5370 struct port *port = m->bridge->ports[i];
5372 if (shash_find(&m->src_ports, port->name)
5375 ? port_trunks_any_mirrored_vlan(m, port)
5376 : vlan_is_mirrored(m, port->vlan)))) {
5377 port->src_mirrors |= mirror_bit;
5379 port->src_mirrors &= ~mirror_bit;
5382 if (shash_find(&m->dst_ports, port->name)) {
5383 port->dst_mirrors |= mirror_bit;
5385 port->dst_mirrors &= ~mirror_bit;
5390 shash_destroy(&src_ports);
5391 shash_destroy(&dst_ports);