1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
40 #include "dynamic-string.h"
46 #include "mac-learning.h"
50 #include "ofp-print.h"
52 #include "ofproto/netflow.h"
53 #include "ofproto/ofproto.h"
54 #include "ovsdb-data.h"
56 #include "poll-loop.h"
57 #include "proc-net-compat.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_reconfigure);
86 struct dst builtin[32];
91 static void dst_set_init(struct dst_set *);
92 static void dst_set_add(struct dst_set *, const struct dst *);
93 static void dst_set_free(struct dst_set *);
96 /* These members are always valid. */
97 struct port *port; /* Containing port. */
98 size_t port_ifidx; /* Index within containing port. */
99 char *name; /* Host network device name. */
100 tag_type tag; /* Tag associated with this interface. */
101 long long delay_expires; /* Time after which 'enabled' may change. */
103 /* These members are valid only after bridge_reconfigure() causes them to
105 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
106 int dp_ifidx; /* Index within kernel datapath. */
107 struct netdev *netdev; /* Network device. */
108 bool enabled; /* May be chosen for flows? */
109 const char *type; /* Usually same as cfg->type. */
110 struct cfm *cfm; /* Connectivity Fault Management */
111 const struct ovsrec_interface *cfg;
114 #define BOND_MASK 0xff
116 int iface_idx; /* Index of assigned iface, or -1 if none. */
117 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
118 tag_type iface_tag; /* Tag associated with iface_idx. */
122 BM_SLB, /* Source Load Balance (Default). */
123 BM_AB /* Active Backup. */
126 #define MAX_MIRRORS 32
127 typedef uint32_t mirror_mask_t;
128 #define MIRROR_MASK_C(X) UINT32_C(X)
129 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
131 struct bridge *bridge;
134 struct uuid uuid; /* UUID of this "mirror" record in database. */
136 /* Selection criteria. */
137 struct shash src_ports; /* Name is port name; data is always NULL. */
138 struct shash dst_ports; /* Name is port name; data is always NULL. */
143 struct port *out_port;
147 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
149 struct bridge *bridge;
151 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
152 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
153 * NULL if all VLANs are trunked. */
154 const struct ovsrec_port *cfg;
157 /* An ordinary bridge port has 1 interface.
158 * A bridge port for bonding has at least 2 interfaces. */
159 struct iface **ifaces;
160 size_t n_ifaces, allocated_ifaces;
163 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
164 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
165 tag_type active_iface_tag; /* Tag for bcast flows. */
166 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
167 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
168 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
169 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
170 bool miimon; /* Use miimon instead of carrier? */
171 long long int bond_miimon_interval; /* Miimon status refresh interval. */
172 long long int bond_miimon_next_update; /* Time of next miimon update. */
173 long long int bond_next_fake_iface_update; /* Time of next update. */
174 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
176 /* SLB specific bonding info. */
177 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
178 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
179 long long int bond_next_rebalance; /* Next rebalancing time. */
181 /* Port mirroring info. */
182 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
183 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
184 bool is_mirror_output_port; /* Does port mirroring send frames here? */
188 struct list node; /* Node in global list of bridges. */
189 char *name; /* User-specified arbitrary name. */
190 struct mac_learning *ml; /* MAC learning table. */
191 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
192 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
193 const struct ovsrec_bridge *cfg;
195 /* OpenFlow switch processing. */
196 struct ofproto *ofproto; /* OpenFlow switch. */
198 /* Kernel datapath information. */
199 struct dpif *dpif; /* Datapath. */
200 struct hmap ifaces; /* Contains "struct iface"s. */
204 size_t n_ports, allocated_ports;
205 struct shash iface_by_name; /* "struct iface"s indexed by name. */
206 struct shash port_by_name; /* "struct port"s indexed by name. */
209 bool has_bonded_ports;
214 /* Port mirroring. */
215 struct mirror *mirrors[MAX_MIRRORS];
218 /* List of all bridges. */
219 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
221 /* OVSDB IDL used to obtain configuration. */
222 static struct ovsdb_idl *idl;
224 /* Each time this timer expires, the bridge fetches systems and interface
225 * statistics and pushes them into the database. */
226 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
227 static long long int stats_timer = LLONG_MIN;
229 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
230 static void bridge_destroy(struct bridge *);
231 static struct bridge *bridge_lookup(const char *name);
232 static unixctl_cb_func bridge_unixctl_dump_flows;
233 static unixctl_cb_func bridge_unixctl_reconnect;
234 static int bridge_run_one(struct bridge *);
235 static size_t bridge_get_controllers(const struct bridge *br,
236 struct ovsrec_controller ***controllersp);
237 static void bridge_reconfigure_one(struct bridge *);
238 static void bridge_reconfigure_remotes(struct bridge *,
239 const struct sockaddr_in *managers,
241 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
242 static void bridge_fetch_dp_ifaces(struct bridge *);
243 static void bridge_flush(struct bridge *);
244 static void bridge_pick_local_hw_addr(struct bridge *,
245 uint8_t ea[ETH_ADDR_LEN],
246 struct iface **hw_addr_iface);
247 static uint64_t bridge_pick_datapath_id(struct bridge *,
248 const uint8_t bridge_ea[ETH_ADDR_LEN],
249 struct iface *hw_addr_iface);
250 static struct iface *bridge_get_local_iface(struct bridge *);
251 static uint64_t dpid_from_hash(const void *, size_t nbytes);
253 static unixctl_cb_func bridge_unixctl_fdb_show;
255 static void bond_init(void);
256 static void bond_run(struct bridge *);
257 static void bond_wait(struct bridge *);
258 static void bond_rebalance_port(struct port *);
259 static void bond_send_learning_packets(struct port *);
260 static void bond_enable_slave(struct iface *iface, bool enable);
262 static struct port *port_create(struct bridge *, const char *name);
263 static void port_reconfigure(struct port *, const struct ovsrec_port *);
264 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
265 static void port_destroy(struct port *);
266 static struct port *port_lookup(const struct bridge *, const char *name);
267 static struct iface *port_lookup_iface(const struct port *, const char *name);
268 static struct port *port_from_dp_ifidx(const struct bridge *,
270 static void port_update_bond_compat(struct port *);
271 static void port_update_vlan_compat(struct port *);
272 static void port_update_bonding(struct port *);
274 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
275 static void mirror_destroy(struct mirror *);
276 static void mirror_reconfigure(struct bridge *);
277 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
278 static bool vlan_is_mirrored(const struct mirror *, int vlan);
280 static struct iface *iface_create(struct port *port,
281 const struct ovsrec_interface *if_cfg);
282 static void iface_destroy(struct iface *);
283 static struct iface *iface_lookup(const struct bridge *, const char *name);
284 static struct iface *iface_from_dp_ifidx(const struct bridge *,
286 static void iface_set_mac(struct iface *);
287 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
288 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
289 static void iface_update_cfm(struct iface *);
290 static void iface_refresh_cfm_stats(struct iface *iface);
291 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
293 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
295 static void shash_to_ovs_idl_map(struct shash *,
296 char ***keys, char ***values, size_t *n);
299 /* Hooks into ofproto processing. */
300 static struct ofhooks bridge_ofhooks;
302 /* Public functions. */
304 /* Initializes the bridge module, configuring it to obtain its configuration
305 * from an OVSDB server accessed over 'remote', which should be a string in a
306 * form acceptable to ovsdb_idl_create(). */
308 bridge_init(const char *remote)
310 /* Create connection to database. */
311 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
313 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
314 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
315 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
317 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
319 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
320 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
322 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
323 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
324 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
326 /* Register unixctl commands. */
327 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
328 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
330 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
338 struct bridge *br, *next_br;
340 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
343 ovsdb_idl_destroy(idl);
346 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
347 * but for which the ovs-vswitchd configuration 'cfg' is required. */
349 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
351 static bool already_configured_once;
352 struct svec bridge_names;
353 struct svec dpif_names, dpif_types;
356 /* Only do this once per ovs-vswitchd run. */
357 if (already_configured_once) {
360 already_configured_once = true;
362 stats_timer = time_msec() + STATS_INTERVAL;
364 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
365 svec_init(&bridge_names);
366 for (i = 0; i < cfg->n_bridges; i++) {
367 svec_add(&bridge_names, cfg->bridges[i]->name);
369 svec_sort(&bridge_names);
371 /* Iterate over all system dpifs and delete any of them that do not appear
373 svec_init(&dpif_names);
374 svec_init(&dpif_types);
375 dp_enumerate_types(&dpif_types);
376 for (i = 0; i < dpif_types.n; i++) {
379 dp_enumerate_names(dpif_types.names[i], &dpif_names);
381 /* Delete each dpif whose name is not in 'bridge_names'. */
382 for (j = 0; j < dpif_names.n; j++) {
383 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
387 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
396 svec_destroy(&bridge_names);
397 svec_destroy(&dpif_names);
398 svec_destroy(&dpif_types);
401 /* Callback for iterate_and_prune_ifaces(). */
403 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
405 if (!iface->netdev) {
406 /* We already reported a related error, don't bother duplicating it. */
410 if (iface->dp_ifidx < 0) {
411 VLOG_ERR("%s interface not in %s, dropping",
412 iface->name, dpif_name(br->dpif));
416 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
417 iface->name, iface->dp_ifidx);
421 /* Callback for iterate_and_prune_ifaces(). */
423 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
424 void *aux OVS_UNUSED)
426 /* Set policing attributes. */
427 netdev_set_policing(iface->netdev,
428 iface->cfg->ingress_policing_rate,
429 iface->cfg->ingress_policing_burst);
431 /* Set MAC address of internal interfaces other than the local
433 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
434 iface_set_mac(iface);
440 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
441 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
442 * deletes from 'br' any ports that no longer have any interfaces. */
444 iterate_and_prune_ifaces(struct bridge *br,
445 bool (*cb)(struct bridge *, struct iface *,
451 for (i = 0; i < br->n_ports; ) {
452 struct port *port = br->ports[i];
453 for (j = 0; j < port->n_ifaces; ) {
454 struct iface *iface = port->ifaces[j];
455 if (cb(br, iface, aux)) {
458 iface_set_ofport(iface->cfg, -1);
459 iface_destroy(iface);
463 if (port->n_ifaces) {
466 VLOG_ERR("%s port has no interfaces, dropping", port->name);
472 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
473 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
474 * responsible for freeing '*managersp' (with free()).
476 * You may be asking yourself "why does ovs-vswitchd care?", because
477 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
478 * should not be and in fact is not directly involved in that. But
479 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
480 * it has to tell in-band control where the managers are to enable that.
481 * (Thus, only managers connected in-band are collected.)
484 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
485 struct sockaddr_in **managersp, size_t *n_managersp)
487 struct sockaddr_in *managers = NULL;
488 size_t n_managers = 0;
489 struct shash targets;
492 /* Collect all of the potential targets, as the union of the "managers"
493 * column and the "targets" columns of the rows pointed to by
494 * "manager_options", excluding any that are out-of-band. */
495 shash_init(&targets);
496 for (i = 0; i < ovs_cfg->n_managers; i++) {
497 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
499 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
500 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
502 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
503 shash_find_and_delete(&targets, m->target);
505 shash_add_once(&targets, m->target, NULL);
509 /* Now extract the targets' IP addresses. */
510 if (!shash_is_empty(&targets)) {
511 struct shash_node *node;
513 managers = xmalloc(shash_count(&targets) * sizeof *managers);
514 SHASH_FOR_EACH (node, &targets) {
515 const char *target = node->name;
516 struct sockaddr_in *sin = &managers[n_managers];
518 if ((!strncmp(target, "tcp:", 4)
519 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
520 (!strncmp(target, "ssl:", 4)
521 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
526 shash_destroy(&targets);
528 *managersp = managers;
529 *n_managersp = n_managers;
533 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
535 struct shash old_br, new_br;
536 struct shash_node *node;
537 struct bridge *br, *next;
538 struct sockaddr_in *managers;
541 int sflow_bridge_number;
543 COVERAGE_INC(bridge_reconfigure);
545 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
547 /* Collect old and new bridges. */
550 LIST_FOR_EACH (br, node, &all_bridges) {
551 shash_add(&old_br, br->name, br);
553 for (i = 0; i < ovs_cfg->n_bridges; i++) {
554 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
555 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
556 VLOG_WARN("more than one bridge named %s", br_cfg->name);
560 /* Get rid of deleted bridges and add new bridges. */
561 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
562 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
569 SHASH_FOR_EACH (node, &new_br) {
570 const char *br_name = node->name;
571 const struct ovsrec_bridge *br_cfg = node->data;
572 br = shash_find_data(&old_br, br_name);
574 /* If the bridge datapath type has changed, we need to tear it
575 * down and recreate. */
576 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
578 bridge_create(br_cfg);
581 bridge_create(br_cfg);
584 shash_destroy(&old_br);
585 shash_destroy(&new_br);
587 /* Reconfigure all bridges. */
588 LIST_FOR_EACH (br, node, &all_bridges) {
589 bridge_reconfigure_one(br);
592 /* Add and delete ports on all datapaths.
594 * The kernel will reject any attempt to add a given port to a datapath if
595 * that port already belongs to a different datapath, so we must do all
596 * port deletions before any port additions. */
597 LIST_FOR_EACH (br, node, &all_bridges) {
598 struct dpif_port_dump dump;
599 struct shash want_ifaces;
600 struct dpif_port dpif_port;
602 bridge_get_all_ifaces(br, &want_ifaces);
603 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
604 if (!shash_find(&want_ifaces, dpif_port.name)
605 && strcmp(dpif_port.name, br->name)) {
606 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
608 VLOG_ERR("failed to remove %s interface from %s: %s",
609 dpif_port.name, dpif_name(br->dpif),
614 shash_destroy(&want_ifaces);
616 LIST_FOR_EACH (br, node, &all_bridges) {
617 struct shash cur_ifaces, want_ifaces;
618 struct dpif_port_dump dump;
619 struct dpif_port dpif_port;
621 /* Get the set of interfaces currently in this datapath. */
622 shash_init(&cur_ifaces);
623 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
624 struct dpif_port *port_info = xmalloc(sizeof *port_info);
625 dpif_port_clone(port_info, &dpif_port);
626 shash_add(&cur_ifaces, dpif_port.name, port_info);
629 /* Get the set of interfaces we want on this datapath. */
630 bridge_get_all_ifaces(br, &want_ifaces);
632 hmap_clear(&br->ifaces);
633 SHASH_FOR_EACH (node, &want_ifaces) {
634 const char *if_name = node->name;
635 struct iface *iface = node->data;
636 struct dpif_port *dpif_port;
640 type = iface ? iface->type : "internal";
641 dpif_port = shash_find_data(&cur_ifaces, if_name);
643 /* If we have a port or a netdev already, and it's not the type we
644 * want, then delete the port (if any) and close the netdev (if
646 if ((dpif_port && strcmp(dpif_port->type, type))
647 || (iface && iface->netdev
648 && strcmp(type, netdev_get_type(iface->netdev)))) {
650 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
657 netdev_close(iface->netdev);
658 iface->netdev = NULL;
662 /* If the port doesn't exist or we don't have the netdev open,
663 * we need to do more work. */
664 if (!dpif_port || (iface && !iface->netdev)) {
665 struct netdev_options options;
666 struct netdev *netdev;
669 /* First open the network device. */
670 options.name = if_name;
672 options.args = &args;
673 options.ethertype = NETDEV_ETH_TYPE_NONE;
677 shash_from_ovs_idl_map(iface->cfg->key_options,
678 iface->cfg->value_options,
679 iface->cfg->n_options, &args);
681 error = netdev_open(&options, &netdev);
682 shash_destroy(&args);
685 VLOG_WARN("could not open network device %s (%s)",
686 if_name, strerror(error));
690 /* Then add the port if we haven't already. */
692 error = dpif_port_add(br->dpif, netdev, NULL);
694 netdev_close(netdev);
695 if (error == EFBIG) {
696 VLOG_ERR("ran out of valid port numbers on %s",
697 dpif_name(br->dpif));
700 VLOG_ERR("failed to add %s interface to %s: %s",
701 if_name, dpif_name(br->dpif),
708 /* Update 'iface'. */
710 iface->netdev = netdev;
711 iface->enabled = netdev_get_carrier(iface->netdev);
713 } else if (iface && iface->netdev) {
717 shash_from_ovs_idl_map(iface->cfg->key_options,
718 iface->cfg->value_options,
719 iface->cfg->n_options, &args);
720 netdev_set_config(iface->netdev, &args);
721 shash_destroy(&args);
724 shash_destroy(&want_ifaces);
726 SHASH_FOR_EACH (node, &cur_ifaces) {
727 struct dpif_port *port_info = node->data;
728 dpif_port_destroy(port_info);
731 shash_destroy(&cur_ifaces);
733 sflow_bridge_number = 0;
734 LIST_FOR_EACH (br, node, &all_bridges) {
737 struct iface *local_iface;
738 struct iface *hw_addr_iface;
741 bridge_fetch_dp_ifaces(br);
743 iterate_and_prune_ifaces(br, check_iface, NULL);
745 /* Pick local port hardware address, datapath ID. */
746 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
747 local_iface = bridge_get_local_iface(br);
749 int error = netdev_set_etheraddr(local_iface->netdev, ea);
751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
752 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
753 "Ethernet address: %s",
754 br->name, strerror(error));
757 memcpy(br->ea, ea, ETH_ADDR_LEN);
759 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
760 ofproto_set_datapath_id(br->ofproto, dpid);
762 dpid_string = xasprintf("%016"PRIx64, dpid);
763 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
766 /* Set NetFlow configuration on this bridge. */
767 if (br->cfg->netflow) {
768 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
769 struct netflow_options opts;
771 memset(&opts, 0, sizeof opts);
773 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
774 if (nf_cfg->engine_type) {
775 opts.engine_type = *nf_cfg->engine_type;
777 if (nf_cfg->engine_id) {
778 opts.engine_id = *nf_cfg->engine_id;
781 opts.active_timeout = nf_cfg->active_timeout;
782 if (!opts.active_timeout) {
783 opts.active_timeout = -1;
784 } else if (opts.active_timeout < 0) {
785 VLOG_WARN("bridge %s: active timeout interval set to negative "
786 "value, using default instead (%d seconds)", br->name,
787 NF_ACTIVE_TIMEOUT_DEFAULT);
788 opts.active_timeout = -1;
791 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
792 if (opts.add_id_to_iface) {
793 if (opts.engine_id > 0x7f) {
794 VLOG_WARN("bridge %s: netflow port mangling may conflict "
795 "with another vswitch, choose an engine id less "
796 "than 128", br->name);
798 if (br->n_ports > 508) {
799 VLOG_WARN("bridge %s: netflow port mangling will conflict "
800 "with another port when more than 508 ports are "
805 opts.collectors.n = nf_cfg->n_targets;
806 opts.collectors.names = nf_cfg->targets;
807 if (ofproto_set_netflow(br->ofproto, &opts)) {
808 VLOG_ERR("bridge %s: problem setting netflow collectors",
812 ofproto_set_netflow(br->ofproto, NULL);
815 /* Set sFlow configuration on this bridge. */
816 if (br->cfg->sflow) {
817 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
818 struct ovsrec_controller **controllers;
819 struct ofproto_sflow_options oso;
820 size_t n_controllers;
822 memset(&oso, 0, sizeof oso);
824 oso.targets.n = sflow_cfg->n_targets;
825 oso.targets.names = sflow_cfg->targets;
827 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
828 if (sflow_cfg->sampling) {
829 oso.sampling_rate = *sflow_cfg->sampling;
832 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
833 if (sflow_cfg->polling) {
834 oso.polling_interval = *sflow_cfg->polling;
837 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
838 if (sflow_cfg->header) {
839 oso.header_len = *sflow_cfg->header;
842 oso.sub_id = sflow_bridge_number++;
843 oso.agent_device = sflow_cfg->agent;
845 oso.control_ip = NULL;
846 n_controllers = bridge_get_controllers(br, &controllers);
847 for (i = 0; i < n_controllers; i++) {
848 if (controllers[i]->local_ip) {
849 oso.control_ip = controllers[i]->local_ip;
853 ofproto_set_sflow(br->ofproto, &oso);
855 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
857 ofproto_set_sflow(br->ofproto, NULL);
860 /* Update the controller and related settings. It would be more
861 * straightforward to call this from bridge_reconfigure_one(), but we
862 * can't do it there for two reasons. First, and most importantly, at
863 * that point we don't know the dp_ifidx of any interfaces that have
864 * been added to the bridge (because we haven't actually added them to
865 * the datapath). Second, at that point we haven't set the datapath ID
866 * yet; when a controller is configured, resetting the datapath ID will
867 * immediately disconnect from the controller, so it's better to set
868 * the datapath ID before the controller. */
869 bridge_reconfigure_remotes(br, managers, n_managers);
871 LIST_FOR_EACH (br, node, &all_bridges) {
872 for (i = 0; i < br->n_ports; i++) {
873 struct port *port = br->ports[i];
876 port_update_vlan_compat(port);
877 port_update_bonding(port);
879 for (j = 0; j < port->n_ifaces; j++) {
880 iface_update_qos(port->ifaces[j], port->cfg->qos);
884 LIST_FOR_EACH (br, node, &all_bridges) {
885 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
888 LIST_FOR_EACH (br, node, &all_bridges) {
890 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
891 iface_update_cfm(iface);
899 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
900 const struct ovsdb_idl_column *column,
903 const struct ovsdb_datum *datum;
904 union ovsdb_atom atom;
907 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
908 atom.string = (char *) key;
909 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
910 return idx == UINT_MAX ? NULL : datum->values[idx].string;
914 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
916 return get_ovsrec_key_value(&br_cfg->header_,
917 &ovsrec_bridge_col_other_config, key);
921 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
922 struct iface **hw_addr_iface)
928 *hw_addr_iface = NULL;
930 /* Did the user request a particular MAC? */
931 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
932 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
933 if (eth_addr_is_multicast(ea)) {
934 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
935 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
936 } else if (eth_addr_is_zero(ea)) {
937 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
943 /* Otherwise choose the minimum non-local MAC address among all of the
945 memset(ea, 0xff, sizeof ea);
946 for (i = 0; i < br->n_ports; i++) {
947 struct port *port = br->ports[i];
948 uint8_t iface_ea[ETH_ADDR_LEN];
951 /* Mirror output ports don't participate. */
952 if (port->is_mirror_output_port) {
956 /* Choose the MAC address to represent the port. */
957 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
958 /* Find the interface with this Ethernet address (if any) so that
959 * we can provide the correct devname to the caller. */
961 for (j = 0; j < port->n_ifaces; j++) {
962 struct iface *candidate = port->ifaces[j];
963 uint8_t candidate_ea[ETH_ADDR_LEN];
964 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
965 && eth_addr_equals(iface_ea, candidate_ea)) {
970 /* Choose the interface whose MAC address will represent the port.
971 * The Linux kernel bonding code always chooses the MAC address of
972 * the first slave added to a bond, and the Fedora networking
973 * scripts always add slaves to a bond in alphabetical order, so
974 * for compatibility we choose the interface with the name that is
975 * first in alphabetical order. */
976 iface = port->ifaces[0];
977 for (j = 1; j < port->n_ifaces; j++) {
978 struct iface *candidate = port->ifaces[j];
979 if (strcmp(candidate->name, iface->name) < 0) {
984 /* The local port doesn't count (since we're trying to choose its
985 * MAC address anyway). */
986 if (iface->dp_ifidx == ODPP_LOCAL) {
991 error = netdev_get_etheraddr(iface->netdev, iface_ea);
993 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
994 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
995 iface->name, strerror(error));
1000 /* Compare against our current choice. */
1001 if (!eth_addr_is_multicast(iface_ea) &&
1002 !eth_addr_is_local(iface_ea) &&
1003 !eth_addr_is_reserved(iface_ea) &&
1004 !eth_addr_is_zero(iface_ea) &&
1005 eth_addr_compare_3way(iface_ea, ea) < 0)
1007 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1008 *hw_addr_iface = iface;
1011 if (eth_addr_is_multicast(ea)) {
1012 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1013 *hw_addr_iface = NULL;
1014 VLOG_WARN("bridge %s: using default bridge Ethernet "
1015 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1017 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1018 br->name, ETH_ADDR_ARGS(ea));
1022 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1023 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1024 * an interface on 'br', then that interface must be passed in as
1025 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1026 * 'hw_addr_iface' must be passed in as a null pointer. */
1028 bridge_pick_datapath_id(struct bridge *br,
1029 const uint8_t bridge_ea[ETH_ADDR_LEN],
1030 struct iface *hw_addr_iface)
1033 * The procedure for choosing a bridge MAC address will, in the most
1034 * ordinary case, also choose a unique MAC that we can use as a datapath
1035 * ID. In some special cases, though, multiple bridges will end up with
1036 * the same MAC address. This is OK for the bridges, but it will confuse
1037 * the OpenFlow controller, because each datapath needs a unique datapath
1040 * Datapath IDs must be unique. It is also very desirable that they be
1041 * stable from one run to the next, so that policy set on a datapath
1044 const char *datapath_id;
1047 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1048 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1052 if (hw_addr_iface) {
1054 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1056 * A bridge whose MAC address is taken from a VLAN network device
1057 * (that is, a network device created with vconfig(8) or similar
1058 * tool) will have the same MAC address as a bridge on the VLAN
1059 * device's physical network device.
1061 * Handle this case by hashing the physical network device MAC
1062 * along with the VLAN identifier.
1064 uint8_t buf[ETH_ADDR_LEN + 2];
1065 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1066 buf[ETH_ADDR_LEN] = vlan >> 8;
1067 buf[ETH_ADDR_LEN + 1] = vlan;
1068 return dpid_from_hash(buf, sizeof buf);
1071 * Assume that this bridge's MAC address is unique, since it
1072 * doesn't fit any of the cases we handle specially.
1077 * A purely internal bridge, that is, one that has no non-virtual
1078 * network devices on it at all, is more difficult because it has no
1079 * natural unique identifier at all.
1081 * When the host is a XenServer, we handle this case by hashing the
1082 * host's UUID with the name of the bridge. Names of bridges are
1083 * persistent across XenServer reboots, although they can be reused if
1084 * an internal network is destroyed and then a new one is later
1085 * created, so this is fairly effective.
1087 * When the host is not a XenServer, we punt by using a random MAC
1088 * address on each run.
1090 const char *host_uuid = xenserver_get_host_uuid();
1092 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1093 dpid = dpid_from_hash(combined, strlen(combined));
1099 return eth_addr_to_uint64(bridge_ea);
1103 dpid_from_hash(const void *data, size_t n)
1105 uint8_t hash[SHA1_DIGEST_SIZE];
1107 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1108 sha1_bytes(data, n, hash);
1109 eth_addr_mark_random(hash);
1110 return eth_addr_to_uint64(hash);
1114 iface_refresh_status(struct iface *iface)
1118 enum netdev_flags flags;
1127 if (!netdev_get_status(iface->netdev, &sh)) {
1129 char **keys, **values;
1131 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1132 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1137 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1140 shash_destroy_free_data(&sh);
1142 error = netdev_get_flags(iface->netdev, &flags);
1144 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1147 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1150 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1152 ovsrec_interface_set_duplex(iface->cfg,
1153 netdev_features_is_full_duplex(current)
1155 /* warning: uint64_t -> int64_t conversion */
1156 bps = netdev_features_to_bps(current);
1157 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1160 ovsrec_interface_set_duplex(iface->cfg, NULL);
1161 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1165 ovsrec_interface_set_link_state(iface->cfg,
1166 netdev_get_carrier(iface->netdev)
1169 error = netdev_get_mtu(iface->netdev, &mtu);
1172 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1175 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1180 iface_refresh_cfm_stats(struct iface *iface)
1184 const struct ovsrec_monitor *mon;
1186 mon = iface->cfg->monitor;
1193 for (i = 0; i < mon->n_remote_mps; i++) {
1194 const struct ovsrec_maintenance_point *mp;
1195 const struct remote_mp *rmp;
1197 mp = mon->remote_mps[i];
1198 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1200 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1203 if (hmap_is_empty(&cfm->x_remote_mps)) {
1204 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1207 struct remote_mp *rmp;
1208 int64_t *x_remote_mps;
1210 length = hmap_count(&cfm->x_remote_mps);
1211 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1214 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1215 x_remote_mps[i++] = rmp->mpid;
1218 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1222 if (hmap_is_empty(&cfm->x_remote_maids)) {
1223 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1226 char **x_remote_maids;
1227 struct remote_maid *rmaid;
1229 length = hmap_count(&cfm->x_remote_maids);
1230 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1233 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1236 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1238 for (j = 0; j < CCM_MAID_LEN; j++) {
1239 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1244 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1246 for (i = 0; i < length; i++) {
1247 free(x_remote_maids[i]);
1249 free(x_remote_maids);
1252 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1256 iface_refresh_stats(struct iface *iface)
1262 static const struct iface_stat iface_stats[] = {
1263 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1264 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1265 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1266 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1267 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1268 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1269 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1270 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1271 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1272 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1273 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1274 { "collisions", offsetof(struct netdev_stats, collisions) },
1276 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1277 const struct iface_stat *s;
1279 char *keys[N_STATS];
1280 int64_t values[N_STATS];
1283 struct netdev_stats stats;
1285 /* Intentionally ignore return value, since errors will set 'stats' to
1286 * all-1s, and we will deal with that correctly below. */
1287 netdev_get_stats(iface->netdev, &stats);
1290 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1291 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1292 if (value != UINT64_MAX) {
1299 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1303 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1305 struct ovsdb_datum datum;
1309 get_system_stats(&stats);
1311 ovsdb_datum_from_shash(&datum, &stats);
1312 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1316 static inline const char *
1317 nx_role_to_str(enum nx_role role)
1322 case NX_ROLE_MASTER:
1327 return "*** INVALID ROLE ***";
1332 bridge_refresh_controller_status(const struct bridge *br)
1335 const struct ovsrec_controller *cfg;
1337 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1339 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1340 struct ofproto_controller_info *cinfo =
1341 shash_find_data(&info, cfg->target);
1344 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1345 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1346 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1347 (char **) cinfo->pairs.values,
1350 ovsrec_controller_set_is_connected(cfg, false);
1351 ovsrec_controller_set_role(cfg, NULL);
1352 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1356 ofproto_free_ofproto_controller_info(&info);
1362 const struct ovsrec_open_vswitch *cfg;
1364 bool datapath_destroyed;
1365 bool database_changed;
1368 /* Let each bridge do the work that it needs to do. */
1369 datapath_destroyed = false;
1370 LIST_FOR_EACH (br, node, &all_bridges) {
1371 int error = bridge_run_one(br);
1373 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1374 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1375 "forcing reconfiguration", br->name);
1376 datapath_destroyed = true;
1380 /* (Re)configure if necessary. */
1381 database_changed = ovsdb_idl_run(idl);
1382 cfg = ovsrec_open_vswitch_first(idl);
1384 /* Re-configure SSL. We do this on every trip through the main loop,
1385 * instead of just when the database changes, because the contents of the
1386 * key and certificate files can change without the database changing.
1388 * We do this before bridge_reconfigure() because that function might
1389 * initiate SSL connections and thus requires SSL to be configured. */
1390 if (cfg && cfg->ssl) {
1391 const struct ovsrec_ssl *ssl = cfg->ssl;
1393 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1394 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1397 if (database_changed || datapath_destroyed) {
1399 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1401 bridge_configure_once(cfg);
1402 bridge_reconfigure(cfg);
1404 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1405 ovsdb_idl_txn_commit(txn);
1406 ovsdb_idl_txn_destroy(txn); /* XXX */
1408 /* We still need to reconfigure to avoid dangling pointers to
1409 * now-destroyed ovsrec structures inside bridge data. */
1410 static const struct ovsrec_open_vswitch null_cfg;
1412 bridge_reconfigure(&null_cfg);
1416 /* Refresh system and interface stats if necessary. */
1417 if (time_msec() >= stats_timer) {
1419 struct ovsdb_idl_txn *txn;
1421 txn = ovsdb_idl_txn_create(idl);
1422 LIST_FOR_EACH (br, node, &all_bridges) {
1425 for (i = 0; i < br->n_ports; i++) {
1426 struct port *port = br->ports[i];
1429 for (j = 0; j < port->n_ifaces; j++) {
1430 struct iface *iface = port->ifaces[j];
1431 iface_refresh_stats(iface);
1432 iface_refresh_cfm_stats(iface);
1433 iface_refresh_status(iface);
1436 bridge_refresh_controller_status(br);
1438 refresh_system_stats(cfg);
1439 ovsdb_idl_txn_commit(txn);
1440 ovsdb_idl_txn_destroy(txn); /* XXX */
1443 stats_timer = time_msec() + STATS_INTERVAL;
1451 struct iface *iface;
1453 LIST_FOR_EACH (br, node, &all_bridges) {
1454 ofproto_wait(br->ofproto);
1455 if (ofproto_has_primary_controller(br->ofproto)) {
1459 mac_learning_wait(br->ml);
1462 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1464 cfm_wait(iface->cfm);
1468 ovsdb_idl_wait(idl);
1469 poll_timer_wait_until(stats_timer);
1472 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1473 * configuration changes. */
1475 bridge_flush(struct bridge *br)
1477 COVERAGE_INC(bridge_flush);
1479 mac_learning_flush(br->ml);
1482 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1483 * such interface. */
1484 static struct iface *
1485 bridge_get_local_iface(struct bridge *br)
1489 for (i = 0; i < br->n_ports; i++) {
1490 struct port *port = br->ports[i];
1491 for (j = 0; j < port->n_ifaces; j++) {
1492 struct iface *iface = port->ifaces[j];
1493 if (iface->dp_ifidx == ODPP_LOCAL) {
1502 /* Bridge unixctl user interface functions. */
1504 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1505 const char *args, void *aux OVS_UNUSED)
1507 struct ds ds = DS_EMPTY_INITIALIZER;
1508 const struct bridge *br;
1509 const struct mac_entry *e;
1511 br = bridge_lookup(args);
1513 unixctl_command_reply(conn, 501, "no such bridge");
1517 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1518 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1519 if (e->port < 0 || e->port >= br->n_ports) {
1522 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1523 br->ports[e->port]->ifaces[0]->dp_ifidx,
1524 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1526 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1530 /* Bridge reconfiguration functions. */
1531 static struct bridge *
1532 bridge_create(const struct ovsrec_bridge *br_cfg)
1537 assert(!bridge_lookup(br_cfg->name));
1538 br = xzalloc(sizeof *br);
1540 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1546 dpif_flow_flush(br->dpif);
1548 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1551 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1553 dpif_delete(br->dpif);
1554 dpif_close(br->dpif);
1559 br->name = xstrdup(br_cfg->name);
1561 br->ml = mac_learning_create();
1562 eth_addr_nicira_random(br->default_ea);
1564 hmap_init(&br->ifaces);
1566 shash_init(&br->port_by_name);
1567 shash_init(&br->iface_by_name);
1571 list_push_back(&all_bridges, &br->node);
1573 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1579 bridge_destroy(struct bridge *br)
1584 while (br->n_ports > 0) {
1585 port_destroy(br->ports[br->n_ports - 1]);
1587 list_remove(&br->node);
1588 error = dpif_delete(br->dpif);
1589 if (error && error != ENOENT) {
1590 VLOG_ERR("failed to delete %s: %s",
1591 dpif_name(br->dpif), strerror(error));
1593 dpif_close(br->dpif);
1594 ofproto_destroy(br->ofproto);
1595 mac_learning_destroy(br->ml);
1596 hmap_destroy(&br->ifaces);
1597 shash_destroy(&br->port_by_name);
1598 shash_destroy(&br->iface_by_name);
1605 static struct bridge *
1606 bridge_lookup(const char *name)
1610 LIST_FOR_EACH (br, node, &all_bridges) {
1611 if (!strcmp(br->name, name)) {
1618 /* Handle requests for a listing of all flows known by the OpenFlow
1619 * stack, including those normally hidden. */
1621 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1622 const char *args, void *aux OVS_UNUSED)
1627 br = bridge_lookup(args);
1629 unixctl_command_reply(conn, 501, "Unknown bridge");
1634 ofproto_get_all_flows(br->ofproto, &results);
1636 unixctl_command_reply(conn, 200, ds_cstr(&results));
1637 ds_destroy(&results);
1640 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1641 * connections and reconnect. If BRIDGE is not specified, then all bridges
1642 * drop their controller connections and reconnect. */
1644 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1645 const char *args, void *aux OVS_UNUSED)
1648 if (args[0] != '\0') {
1649 br = bridge_lookup(args);
1651 unixctl_command_reply(conn, 501, "Unknown bridge");
1654 ofproto_reconnect_controllers(br->ofproto);
1656 LIST_FOR_EACH (br, node, &all_bridges) {
1657 ofproto_reconnect_controllers(br->ofproto);
1660 unixctl_command_reply(conn, 200, NULL);
1664 bridge_run_one(struct bridge *br)
1667 struct iface *iface;
1669 error = ofproto_run1(br->ofproto);
1674 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1677 error = ofproto_run2(br->ofproto, br->flush);
1680 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1681 struct ofpbuf *packet;
1687 packet = cfm_run(iface->cfm);
1689 iface_send_packet(iface, packet);
1690 ofpbuf_uninit(packet);
1699 bridge_get_controllers(const struct bridge *br,
1700 struct ovsrec_controller ***controllersp)
1702 struct ovsrec_controller **controllers;
1703 size_t n_controllers;
1705 controllers = br->cfg->controller;
1706 n_controllers = br->cfg->n_controller;
1708 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1714 *controllersp = controllers;
1716 return n_controllers;
1720 bridge_reconfigure_one(struct bridge *br)
1722 struct shash old_ports, new_ports;
1723 struct svec snoops, old_snoops;
1724 struct shash_node *node;
1725 enum ofproto_fail_mode fail_mode;
1728 /* Collect old ports. */
1729 shash_init(&old_ports);
1730 for (i = 0; i < br->n_ports; i++) {
1731 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1734 /* Collect new ports. */
1735 shash_init(&new_ports);
1736 for (i = 0; i < br->cfg->n_ports; i++) {
1737 const char *name = br->cfg->ports[i]->name;
1738 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1739 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1744 /* If we have a controller, then we need a local port. Complain if the
1745 * user didn't specify one.
1747 * XXX perhaps we should synthesize a port ourselves in this case. */
1748 if (bridge_get_controllers(br, NULL)) {
1749 char local_name[IF_NAMESIZE];
1752 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1753 local_name, sizeof local_name);
1754 if (!error && !shash_find(&new_ports, local_name)) {
1755 VLOG_WARN("bridge %s: controller specified but no local port "
1756 "(port named %s) defined",
1757 br->name, local_name);
1761 /* Get rid of deleted ports.
1762 * Get rid of deleted interfaces on ports that still exist. */
1763 SHASH_FOR_EACH (node, &old_ports) {
1764 struct port *port = node->data;
1765 const struct ovsrec_port *port_cfg;
1767 port_cfg = shash_find_data(&new_ports, node->name);
1771 port_del_ifaces(port, port_cfg);
1775 /* Create new ports.
1776 * Add new interfaces to existing ports.
1777 * Reconfigure existing ports. */
1778 SHASH_FOR_EACH (node, &new_ports) {
1779 struct port *port = shash_find_data(&old_ports, node->name);
1781 port = port_create(br, node->name);
1784 port_reconfigure(port, node->data);
1785 if (!port->n_ifaces) {
1786 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1787 br->name, port->name);
1791 shash_destroy(&old_ports);
1792 shash_destroy(&new_ports);
1794 /* Set the fail-mode */
1795 fail_mode = !br->cfg->fail_mode
1796 || !strcmp(br->cfg->fail_mode, "standalone")
1797 ? OFPROTO_FAIL_STANDALONE
1798 : OFPROTO_FAIL_SECURE;
1799 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1800 && !ofproto_has_primary_controller(br->ofproto)) {
1801 ofproto_flush_flows(br->ofproto);
1803 ofproto_set_fail_mode(br->ofproto, fail_mode);
1805 /* Delete all flows if we're switching from connected to standalone or vice
1806 * versa. (XXX Should we delete all flows if we are switching from one
1807 * controller to another?) */
1809 /* Configure OpenFlow controller connection snooping. */
1811 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1812 ovs_rundir(), br->name));
1813 svec_init(&old_snoops);
1814 ofproto_get_snoops(br->ofproto, &old_snoops);
1815 if (!svec_equal(&snoops, &old_snoops)) {
1816 ofproto_set_snoops(br->ofproto, &snoops);
1818 svec_destroy(&snoops);
1819 svec_destroy(&old_snoops);
1821 mirror_reconfigure(br);
1824 /* Initializes 'oc' appropriately as a management service controller for
1827 * The caller must free oc->target when it is no longer needed. */
1829 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1830 struct ofproto_controller *oc)
1832 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1833 oc->max_backoff = 0;
1834 oc->probe_interval = 60;
1835 oc->band = OFPROTO_OUT_OF_BAND;
1836 oc->accept_re = NULL;
1837 oc->update_resolv_conf = false;
1839 oc->burst_limit = 0;
1842 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1844 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1845 struct ofproto_controller *oc)
1847 oc->target = c->target;
1848 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1849 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1850 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1851 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1852 oc->accept_re = c->discover_accept_regex;
1853 oc->update_resolv_conf = c->discover_update_resolv_conf;
1854 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1855 oc->burst_limit = (c->controller_burst_limit
1856 ? *c->controller_burst_limit : 0);
1859 /* Configures the IP stack for 'br''s local interface properly according to the
1860 * configuration in 'c'. */
1862 bridge_configure_local_iface_netdev(struct bridge *br,
1863 struct ovsrec_controller *c)
1865 struct netdev *netdev;
1866 struct in_addr mask, gateway;
1868 struct iface *local_iface;
1871 /* Controller discovery does its own TCP/IP configuration later. */
1872 if (strcmp(c->target, "discover")) {
1876 /* If there's no local interface or no IP address, give up. */
1877 local_iface = bridge_get_local_iface(br);
1878 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1882 /* Bring up the local interface. */
1883 netdev = local_iface->netdev;
1884 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1886 /* Configure the IP address and netmask. */
1887 if (!c->local_netmask
1888 || !inet_aton(c->local_netmask, &mask)
1890 mask.s_addr = guess_netmask(ip.s_addr);
1892 if (!netdev_set_in4(netdev, ip, mask)) {
1893 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1894 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1897 /* Configure the default gateway. */
1898 if (c->local_gateway
1899 && inet_aton(c->local_gateway, &gateway)
1900 && gateway.s_addr) {
1901 if (!netdev_add_router(netdev, gateway)) {
1902 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1903 br->name, IP_ARGS(&gateway.s_addr));
1909 bridge_reconfigure_remotes(struct bridge *br,
1910 const struct sockaddr_in *managers,
1913 const char *disable_ib_str, *queue_id_str;
1914 bool disable_in_band = false;
1917 struct ovsrec_controller **controllers;
1918 size_t n_controllers;
1921 struct ofproto_controller *ocs;
1925 /* Check if we should disable in-band control on this bridge. */
1926 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1927 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1928 disable_in_band = true;
1931 /* Set OpenFlow queue ID for in-band control. */
1932 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1933 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1934 ofproto_set_in_band_queue(br->ofproto, queue_id);
1936 if (disable_in_band) {
1937 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1939 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1941 had_primary = ofproto_has_primary_controller(br->ofproto);
1943 n_controllers = bridge_get_controllers(br, &controllers);
1945 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1948 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1949 for (i = 0; i < n_controllers; i++) {
1950 struct ovsrec_controller *c = controllers[i];
1952 if (!strncmp(c->target, "punix:", 6)
1953 || !strncmp(c->target, "unix:", 5)) {
1954 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1956 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1957 * domain sockets and overwriting arbitrary local files. */
1958 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1959 "\"%s\" due to possibility for remote exploit",
1960 dpif_name(br->dpif), c->target);
1964 bridge_configure_local_iface_netdev(br, c);
1965 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1966 if (disable_in_band) {
1967 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1972 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1973 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1976 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1977 ofproto_flush_flows(br->ofproto);
1980 /* If there are no controllers and the bridge is in standalone
1981 * mode, set up a flow that matches every packet and directs
1982 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1983 * switch is in secure mode and we won't pass any traffic until
1984 * a controller has been defined and it tells us to do so. */
1986 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1987 union ofp_action action;
1988 struct cls_rule rule;
1990 memset(&action, 0, sizeof action);
1991 action.type = htons(OFPAT_OUTPUT);
1992 action.output.len = htons(sizeof action);
1993 action.output.port = htons(OFPP_NORMAL);
1994 cls_rule_init_catchall(&rule, 0);
1995 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2000 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2005 for (i = 0; i < br->n_ports; i++) {
2006 struct port *port = br->ports[i];
2007 for (j = 0; j < port->n_ifaces; j++) {
2008 struct iface *iface = port->ifaces[j];
2009 shash_add_once(ifaces, iface->name, iface);
2011 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2012 shash_add_once(ifaces, port->name, NULL);
2017 /* For robustness, in case the administrator moves around datapath ports behind
2018 * our back, we re-check all the datapath port numbers here.
2020 * This function will set the 'dp_ifidx' members of interfaces that have
2021 * disappeared to -1, so only call this function from a context where those
2022 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2023 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2024 * datapath, which doesn't support UINT16_MAX+1 ports. */
2026 bridge_fetch_dp_ifaces(struct bridge *br)
2028 struct dpif_port_dump dump;
2029 struct dpif_port dpif_port;
2032 /* Reset all interface numbers. */
2033 for (i = 0; i < br->n_ports; i++) {
2034 struct port *port = br->ports[i];
2035 for (j = 0; j < port->n_ifaces; j++) {
2036 struct iface *iface = port->ifaces[j];
2037 iface->dp_ifidx = -1;
2040 hmap_clear(&br->ifaces);
2042 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2043 struct iface *iface = iface_lookup(br, dpif_port.name);
2045 if (iface->dp_ifidx >= 0) {
2046 VLOG_WARN("%s reported interface %s twice",
2047 dpif_name(br->dpif), dpif_port.name);
2048 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2049 VLOG_WARN("%s reported interface %"PRIu16" twice",
2050 dpif_name(br->dpif), dpif_port.port_no);
2052 iface->dp_ifidx = dpif_port.port_no;
2053 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2054 hash_int(iface->dp_ifidx, 0));
2057 iface_set_ofport(iface->cfg,
2058 (iface->dp_ifidx >= 0
2059 ? odp_port_to_ofp_port(iface->dp_ifidx)
2065 /* Bridge packet processing functions. */
2068 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2070 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2073 static struct bond_entry *
2074 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
2077 assert(port->bond_mode == BM_SLB);
2078 return &port->bond_hash[bond_hash(mac, vlan)];
2082 bond_choose_iface(const struct port *port)
2084 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2085 size_t i, best_down_slave = -1;
2086 long long next_delay_expiration = LLONG_MAX;
2088 for (i = 0; i < port->n_ifaces; i++) {
2089 struct iface *iface = port->ifaces[i];
2091 if (iface->enabled) {
2093 } else if (iface->delay_expires < next_delay_expiration) {
2094 best_down_slave = i;
2095 next_delay_expiration = iface->delay_expires;
2099 if (best_down_slave != -1) {
2100 struct iface *iface = port->ifaces[best_down_slave];
2102 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2103 "since no other interface is up", iface->name,
2104 iface->delay_expires - time_msec());
2105 bond_enable_slave(iface, true);
2108 return best_down_slave;
2112 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2113 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2115 struct iface *iface;
2117 assert(port->n_ifaces);
2118 if (port->n_ifaces == 1) {
2119 iface = port->ifaces[0];
2120 } else if (port->bond_mode == BM_AB) {
2121 if (port->active_iface < 0) {
2122 *tags |= port->no_ifaces_tag;
2125 iface = port->ifaces[port->active_iface];
2126 } else if (port->bond_mode == BM_SLB){
2127 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2128 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2129 || !port->ifaces[e->iface_idx]->enabled) {
2130 /* XXX select interface properly. The current interface selection
2131 * is only good for testing the rebalancing code. */
2132 e->iface_idx = bond_choose_iface(port);
2133 if (e->iface_idx < 0) {
2134 *tags |= port->no_ifaces_tag;
2137 e->iface_tag = tag_create_random();
2138 ((struct port *) port)->bond_compat_is_stale = true;
2140 *tags |= e->iface_tag;
2141 iface = port->ifaces[e->iface_idx];
2145 *dp_ifidx = iface->dp_ifidx;
2146 *tags |= iface->tag; /* Currently only used for bonding. */
2151 bond_link_status_update(struct iface *iface, bool carrier)
2153 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2154 struct port *port = iface->port;
2156 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2157 /* Nothing to do. */
2160 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2161 iface->name, carrier ? "up" : "down");
2162 if (carrier == iface->enabled) {
2163 iface->delay_expires = LLONG_MAX;
2164 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2165 iface->name, carrier ? "disabled" : "enabled");
2166 } else if (carrier && port->active_iface < 0) {
2167 bond_enable_slave(iface, true);
2168 if (port->updelay) {
2169 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2170 "other interface is up", iface->name, port->updelay);
2173 int delay = carrier ? port->updelay : port->downdelay;
2174 iface->delay_expires = time_msec() + delay;
2177 "interface %s: will be %s if it stays %s for %d ms",
2179 carrier ? "enabled" : "disabled",
2180 carrier ? "up" : "down",
2187 bond_choose_active_iface(struct port *port)
2189 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2191 port->active_iface = bond_choose_iface(port);
2192 port->active_iface_tag = tag_create_random();
2193 if (port->active_iface >= 0) {
2194 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2195 port->name, port->ifaces[port->active_iface]->name);
2197 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2203 bond_enable_slave(struct iface *iface, bool enable)
2205 struct port *port = iface->port;
2206 struct bridge *br = port->bridge;
2208 /* This acts as a recursion check. If the act of disabling a slave
2209 * causes a different slave to be enabled, the flag will allow us to
2210 * skip redundant work when we reenter this function. It must be
2211 * cleared on exit to keep things safe with multiple bonds. */
2212 static bool moving_active_iface = false;
2214 iface->delay_expires = LLONG_MAX;
2215 if (enable == iface->enabled) {
2219 iface->enabled = enable;
2220 if (!iface->enabled) {
2221 VLOG_WARN("interface %s: disabled", iface->name);
2222 ofproto_revalidate(br->ofproto, iface->tag);
2223 if (iface->port_ifidx == port->active_iface) {
2224 ofproto_revalidate(br->ofproto,
2225 port->active_iface_tag);
2227 /* Disabling a slave can lead to another slave being immediately
2228 * enabled if there will be no active slaves but one is waiting
2229 * on an updelay. In this case we do not need to run most of the
2230 * code for the newly enabled slave since there was no period
2231 * without an active slave and it is redundant with the disabling
2233 moving_active_iface = true;
2234 bond_choose_active_iface(port);
2236 bond_send_learning_packets(port);
2238 VLOG_WARN("interface %s: enabled", iface->name);
2239 if (port->active_iface < 0 && !moving_active_iface) {
2240 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2241 bond_choose_active_iface(port);
2242 bond_send_learning_packets(port);
2244 iface->tag = tag_create_random();
2247 moving_active_iface = false;
2248 port->bond_compat_is_stale = true;
2251 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2252 * bond interface. */
2254 bond_update_fake_iface_stats(struct port *port)
2256 struct netdev_stats bond_stats;
2257 struct netdev *bond_dev;
2260 memset(&bond_stats, 0, sizeof bond_stats);
2262 for (i = 0; i < port->n_ifaces; i++) {
2263 struct netdev_stats slave_stats;
2265 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2266 /* XXX: We swap the stats here because they are swapped back when
2267 * reported by the internal device. The reason for this is
2268 * internal devices normally represent packets going into the system
2269 * but when used as fake bond device they represent packets leaving
2270 * the system. We really should do this in the internal device
2271 * itself because changing it here reverses the counts from the
2272 * perspective of the switch. However, the internal device doesn't
2273 * know what type of device it represents so we have to do it here
2275 bond_stats.tx_packets += slave_stats.rx_packets;
2276 bond_stats.tx_bytes += slave_stats.rx_bytes;
2277 bond_stats.rx_packets += slave_stats.tx_packets;
2278 bond_stats.rx_bytes += slave_stats.tx_bytes;
2282 if (!netdev_open_default(port->name, &bond_dev)) {
2283 netdev_set_stats(bond_dev, &bond_stats);
2284 netdev_close(bond_dev);
2289 bond_run(struct bridge *br)
2293 for (i = 0; i < br->n_ports; i++) {
2294 struct port *port = br->ports[i];
2296 if (port->n_ifaces >= 2) {
2299 if (port->monitor) {
2300 assert(!port->miimon);
2302 /* Track carrier going up and down on interfaces. */
2303 while (!netdev_monitor_poll(port->monitor, &devname)) {
2304 struct iface *iface;
2306 iface = port_lookup_iface(port, devname);
2308 bool up = netdev_get_carrier(iface->netdev);
2310 bond_link_status_update(iface, up);
2311 port_update_bond_compat(port);
2316 assert(port->miimon);
2318 if (time_msec() >= port->bond_miimon_next_update) {
2319 for (j = 0; j < port->n_ifaces; j++) {
2320 struct iface *iface = port->ifaces[j];
2321 bool up = netdev_get_miimon(iface->netdev);
2323 bond_link_status_update(iface, up);
2324 port_update_bond_compat(port);
2326 port->bond_miimon_next_update = time_msec() +
2327 port->bond_miimon_interval;
2331 for (j = 0; j < port->n_ifaces; j++) {
2332 struct iface *iface = port->ifaces[j];
2333 if (time_msec() >= iface->delay_expires) {
2334 bond_enable_slave(iface, !iface->enabled);
2338 if (port->bond_fake_iface
2339 && time_msec() >= port->bond_next_fake_iface_update) {
2340 bond_update_fake_iface_stats(port);
2341 port->bond_next_fake_iface_update = time_msec() + 1000;
2345 if (port->bond_compat_is_stale) {
2346 port->bond_compat_is_stale = false;
2347 port_update_bond_compat(port);
2353 bond_wait(struct bridge *br)
2357 for (i = 0; i < br->n_ports; i++) {
2358 struct port *port = br->ports[i];
2359 if (port->n_ifaces < 2) {
2363 if (port->monitor) {
2364 netdev_monitor_poll_wait(port->monitor);
2368 poll_timer_wait_until(port->bond_miimon_next_update);
2371 for (j = 0; j < port->n_ifaces; j++) {
2372 struct iface *iface = port->ifaces[j];
2373 if (iface->delay_expires != LLONG_MAX) {
2374 poll_timer_wait_until(iface->delay_expires);
2377 if (port->bond_fake_iface) {
2378 poll_timer_wait_until(port->bond_next_fake_iface_update);
2384 set_dst(struct dst *dst, const struct flow *flow,
2385 const struct port *in_port, const struct port *out_port,
2388 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2389 : in_port->vlan >= 0 ? in_port->vlan
2390 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2391 : vlan_tci_to_vid(flow->vlan_tci));
2392 return choose_output_iface(out_port, flow->dl_src, dst->vlan,
2393 &dst->dp_ifidx, tags);
2397 swap_dst(struct dst *p, struct dst *q)
2399 struct dst tmp = *p;
2404 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2405 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2406 * that we push to the datapath. We could in fact fully sort the array by
2407 * vlan, but in most cases there are at most two different vlan tags so that's
2408 * possibly overkill.) */
2410 partition_dsts(struct dst_set *set, int vlan)
2412 struct dst *first = set->dsts;
2413 struct dst *last = set->dsts + set->n;
2415 while (first != last) {
2417 * - All dsts < first have vlan == 'vlan'.
2418 * - All dsts >= last have vlan != 'vlan'.
2419 * - first < last. */
2420 while (first->vlan == vlan) {
2421 if (++first == last) {
2426 /* Same invariants, plus one additional:
2427 * - first->vlan != vlan.
2429 while (last[-1].vlan != vlan) {
2430 if (--last == first) {
2435 /* Same invariants, plus one additional:
2436 * - last[-1].vlan == vlan.*/
2437 swap_dst(first++, --last);
2442 mirror_mask_ffs(mirror_mask_t mask)
2444 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2449 dst_set_init(struct dst_set *set)
2451 set->dsts = set->builtin;
2453 set->allocated = ARRAY_SIZE(set->builtin);
2457 dst_set_add(struct dst_set *set, const struct dst *dst)
2459 if (set->n >= set->allocated) {
2460 size_t new_allocated;
2461 struct dst *new_dsts;
2463 new_allocated = set->allocated * 2;
2464 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2465 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2469 set->dsts = new_dsts;
2470 set->allocated = new_allocated;
2472 set->dsts[set->n++] = *dst;
2476 dst_set_free(struct dst_set *set)
2478 if (set->dsts != set->builtin) {
2484 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2487 for (i = 0; i < set->n; i++) {
2488 if (set->dsts[i].vlan == test->vlan
2489 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2497 port_trunks_vlan(const struct port *port, uint16_t vlan)
2499 return (port->vlan < 0
2500 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2504 port_includes_vlan(const struct port *port, uint16_t vlan)
2506 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2510 port_is_floodable(const struct port *port)
2514 for (i = 0; i < port->n_ifaces; i++) {
2515 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2516 port->ifaces[i]->dp_ifidx)) {
2524 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2525 const struct port *in_port, const struct port *out_port,
2526 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2528 mirror_mask_t mirrors = in_port->src_mirrors;
2533 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2534 if (flow_vlan == 0) {
2535 flow_vlan = OFP_VLAN_NONE;
2538 if (out_port == FLOOD_PORT) {
2539 for (i = 0; i < br->n_ports; i++) {
2540 struct port *port = br->ports[i];
2542 && port_is_floodable(port)
2543 && port_includes_vlan(port, vlan)
2544 && !port->is_mirror_output_port
2545 && set_dst(&dst, flow, in_port, port, tags)) {
2546 mirrors |= port->dst_mirrors;
2547 dst_set_add(set, &dst);
2550 *nf_output_iface = NF_OUT_FLOOD;
2551 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2552 dst_set_add(set, &dst);
2553 *nf_output_iface = dst.dp_ifidx;
2554 mirrors |= out_port->dst_mirrors;
2558 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2559 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2561 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2562 && !dst_is_duplicate(set, &dst)) {
2563 dst_set_add(set, &dst);
2566 for (i = 0; i < br->n_ports; i++) {
2567 struct port *port = br->ports[i];
2568 if (port_includes_vlan(port, m->out_vlan)
2569 && set_dst(&dst, flow, in_port, port, tags))
2571 if (port->vlan < 0) {
2572 dst.vlan = m->out_vlan;
2574 if (dst_is_duplicate(set, &dst)) {
2578 /* Use the vlan tag on the original flow instead of
2579 * the one passed in the vlan parameter. This ensures
2580 * that we compare the vlan from before any implicit
2581 * tagging tags place. This is necessary because
2582 * dst->vlan is the final vlan, after removing implicit
2584 if (port == in_port && dst.vlan == flow_vlan) {
2585 /* Don't send out input port on same VLAN. */
2588 dst_set_add(set, &dst);
2593 mirrors &= mirrors - 1;
2596 partition_dsts(set, flow_vlan);
2599 static void OVS_UNUSED
2600 print_dsts(const struct dst_set *set)
2604 for (i = 0; i < set->n; i++) {
2605 const struct dst *dst = &set->dsts[i];
2607 printf(">p%"PRIu16, dst->dp_ifidx);
2608 if (dst->vlan != OFP_VLAN_NONE) {
2609 printf("v%"PRIu16, dst->vlan);
2615 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2616 const struct port *in_port, const struct port *out_port,
2617 tag_type *tags, struct ofpbuf *actions,
2618 uint16_t *nf_output_iface)
2625 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2628 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2629 if (cur_vlan == 0) {
2630 cur_vlan = OFP_VLAN_NONE;
2632 for (i = 0; i < set.n; i++) {
2633 const struct dst *dst = &set.dsts[i];
2634 if (dst->vlan != cur_vlan) {
2635 if (dst->vlan == OFP_VLAN_NONE) {
2636 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2639 tci = htons(dst->vlan & VLAN_VID_MASK);
2640 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2641 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2643 cur_vlan = dst->vlan;
2645 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2650 /* Returns the effective vlan of a packet, taking into account both the
2651 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2652 * the packet is untagged and -1 indicates it has an invalid header and
2653 * should be dropped. */
2654 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2655 struct port *in_port, bool have_packet)
2657 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2658 if (in_port->vlan >= 0) {
2660 /* XXX support double tagging? */
2662 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2663 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2664 "packet received on port %s configured with "
2665 "implicit VLAN %"PRIu16,
2666 br->name, vlan, in_port->name, in_port->vlan);
2670 vlan = in_port->vlan;
2672 if (!port_includes_vlan(in_port, vlan)) {
2674 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2675 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2676 "packet received on port %s not configured for "
2678 br->name, vlan, in_port->name, vlan);
2687 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2688 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2689 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2691 is_gratuitous_arp(const struct flow *flow)
2693 return (flow->dl_type == htons(ETH_TYPE_ARP)
2694 && eth_addr_is_broadcast(flow->dl_dst)
2695 && (flow->nw_proto == ARP_OP_REPLY
2696 || (flow->nw_proto == ARP_OP_REQUEST
2697 && flow->nw_src == flow->nw_dst)));
2701 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2702 struct port *in_port)
2704 enum grat_arp_lock_type lock_type;
2707 /* We don't want to learn from gratuitous ARP packets that are reflected
2708 * back over bond slaves so we lock the learning table. */
2709 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2710 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2711 GRAT_ARP_LOCK_CHECK;
2713 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2716 /* The log messages here could actually be useful in debugging,
2717 * so keep the rate limit relatively high. */
2718 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2720 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2721 "on port %s in VLAN %d",
2722 br->name, ETH_ADDR_ARGS(flow->dl_src),
2723 in_port->name, vlan);
2724 ofproto_revalidate(br->ofproto, rev_tag);
2728 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2729 * dropped. Returns true if they may be forwarded, false if they should be
2732 * If 'have_packet' is true, it indicates that the caller is processing a
2733 * received packet. If 'have_packet' is false, then the caller is just
2734 * revalidating an existing flow because configuration has changed. Either
2735 * way, 'have_packet' only affects logging (there is no point in logging errors
2736 * during revalidation).
2738 * Sets '*in_portp' to the input port. This will be a null pointer if
2739 * flow->in_port does not designate a known input port (in which case
2740 * is_admissible() returns false).
2742 * When returning true, sets '*vlanp' to the effective VLAN of the input
2743 * packet, as returned by flow_get_vlan().
2745 * May also add tags to '*tags', although the current implementation only does
2746 * so in one special case.
2749 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2750 tag_type *tags, int *vlanp, struct port **in_portp)
2752 struct iface *in_iface;
2753 struct port *in_port;
2756 /* Find the interface and port structure for the received packet. */
2757 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2759 /* No interface? Something fishy... */
2761 /* Odd. A few possible reasons here:
2763 * - We deleted an interface but there are still a few packets
2764 * queued up from it.
2766 * - Someone externally added an interface (e.g. with "ovs-dpctl
2767 * add-if") that we don't know about.
2769 * - Packet arrived on the local port but the local port is not
2770 * one of our bridge ports.
2772 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2774 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2775 "interface %"PRIu16, br->name, flow->in_port);
2781 *in_portp = in_port = in_iface->port;
2782 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2787 /* Drop frames for reserved multicast addresses. */
2788 if (eth_addr_is_reserved(flow->dl_dst)) {
2792 /* Drop frames on ports reserved for mirroring. */
2793 if (in_port->is_mirror_output_port) {
2795 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2796 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2797 "%s, which is reserved exclusively for mirroring",
2798 br->name, in_port->name);
2803 /* Packets received on bonds need special attention to avoid duplicates. */
2804 if (in_port->n_ifaces > 1) {
2806 bool is_grat_arp_locked;
2808 if (eth_addr_is_multicast(flow->dl_dst)) {
2809 *tags |= in_port->active_iface_tag;
2810 if (in_port->active_iface != in_iface->port_ifidx) {
2811 /* Drop all multicast packets on inactive slaves. */
2816 /* Drop all packets for which we have learned a different input
2817 * port, because we probably sent the packet on one slave and got
2818 * it back on the other. Gratuitous ARP packets are an exception
2819 * to this rule: the host has moved to another switch. The exception
2820 * to the exception is if we locked the learning table to avoid
2821 * reflections on bond slaves. If this is the case, just drop the
2823 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2824 &is_grat_arp_locked);
2825 if (src_idx != -1 && src_idx != in_port->port_idx &&
2826 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2834 /* If the composed actions may be applied to any packet in the given 'flow',
2835 * returns true. Otherwise, the actions should only be applied to 'packet', or
2836 * not at all, if 'packet' was NULL. */
2838 process_flow(struct bridge *br, const struct flow *flow,
2839 const struct ofpbuf *packet, struct ofpbuf *actions,
2840 tag_type *tags, uint16_t *nf_output_iface)
2842 struct port *in_port;
2843 struct port *out_port;
2847 /* Check whether we should drop packets in this flow. */
2848 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2853 /* Learn source MAC (but don't try to learn from revalidation). */
2855 update_learning_table(br, flow, vlan, in_port);
2858 /* Determine output port. */
2859 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2861 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2862 out_port = br->ports[out_port_idx];
2863 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2864 /* If we are revalidating but don't have a learning entry then
2865 * eject the flow. Installing a flow that floods packets opens
2866 * up a window of time where we could learn from a packet reflected
2867 * on a bond and blackhole packets before the learning table is
2868 * updated to reflect the correct port. */
2871 out_port = FLOOD_PORT;
2874 /* Don't send packets out their input ports. */
2875 if (in_port == out_port) {
2881 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2889 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2890 struct ofpbuf *actions, tag_type *tags,
2891 uint16_t *nf_output_iface, void *br_)
2893 struct iface *iface;
2894 struct bridge *br = br_;
2896 COVERAGE_INC(bridge_process_flow);
2898 iface = iface_from_dp_ifidx(br, flow->in_port);
2900 if (cfm_should_process_flow(flow)) {
2901 if (packet && iface->cfm) {
2902 cfm_process_heartbeat(iface->cfm, packet);
2907 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2911 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2912 const struct nlattr *actions,
2914 unsigned long long int n_bytes, void *br_)
2916 struct bridge *br = br_;
2917 const struct nlattr *a;
2918 struct port *in_port;
2923 /* Feed information from the active flows back into the learning table to
2924 * ensure that table is always in sync with what is actually flowing
2925 * through the datapath.
2927 * We test that 'tags' is nonzero to ensure that only flows that include an
2928 * OFPP_NORMAL action are used for learning. This works because
2929 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2930 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2931 update_learning_table(br, flow, vlan, in_port);
2934 /* Account for bond slave utilization. */
2935 if (!br->has_bonded_ports) {
2938 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2939 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
2940 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2941 if (out_port && out_port->n_ifaces >= 2 &&
2942 out_port->bond_mode == BM_SLB) {
2943 uint16_t vlan = (flow->vlan_tci
2944 ? vlan_tci_to_vid(flow->vlan_tci)
2946 struct bond_entry *e = lookup_bond_entry(out_port,
2947 flow->dl_src, vlan);
2948 e->tx_bytes += n_bytes;
2955 bridge_account_checkpoint_ofhook_cb(void *br_)
2957 struct bridge *br = br_;
2961 if (!br->has_bonded_ports) {
2966 for (i = 0; i < br->n_ports; i++) {
2967 struct port *port = br->ports[i];
2968 if (port->n_ifaces > 1 && port->bond_mode == BM_SLB
2969 && now >= port->bond_next_rebalance) {
2970 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2971 bond_rebalance_port(port);
2976 static struct ofhooks bridge_ofhooks = {
2977 bridge_normal_ofhook_cb,
2978 bridge_account_flow_ofhook_cb,
2979 bridge_account_checkpoint_ofhook_cb,
2982 /* Bonding functions. */
2984 /* Statistics for a single interface on a bonded port, used for load-based
2985 * bond rebalancing. */
2986 struct slave_balance {
2987 struct iface *iface; /* The interface. */
2988 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2990 /* All the "bond_entry"s that are assigned to this interface, in order of
2991 * increasing tx_bytes. */
2992 struct bond_entry **hashes;
2997 bond_mode_to_string(enum bond_mode bm) {
2998 static char *bm_slb = "balance-slb";
2999 static char *bm_ab = "active-backup";
3002 case BM_SLB: return bm_slb;
3003 case BM_AB: return bm_ab;
3010 /* Sorts pointers to pointers to bond_entries in ascending order by the
3011 * interface to which they are assigned, and within a single interface in
3012 * ascending order of bytes transmitted. */
3014 compare_bond_entries(const void *a_, const void *b_)
3016 const struct bond_entry *const *ap = a_;
3017 const struct bond_entry *const *bp = b_;
3018 const struct bond_entry *a = *ap;
3019 const struct bond_entry *b = *bp;
3020 if (a->iface_idx != b->iface_idx) {
3021 return a->iface_idx > b->iface_idx ? 1 : -1;
3022 } else if (a->tx_bytes != b->tx_bytes) {
3023 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3029 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3030 * *descending* order by number of bytes transmitted. */
3032 compare_slave_balance(const void *a_, const void *b_)
3034 const struct slave_balance *a = a_;
3035 const struct slave_balance *b = b_;
3036 if (a->iface->enabled != b->iface->enabled) {
3037 return a->iface->enabled ? -1 : 1;
3038 } else if (a->tx_bytes != b->tx_bytes) {
3039 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3046 swap_bals(struct slave_balance *a, struct slave_balance *b)
3048 struct slave_balance tmp = *a;
3053 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3054 * given that 'p' (and only 'p') might be in the wrong location.
3056 * This function invalidates 'p', since it might now be in a different memory
3059 resort_bals(struct slave_balance *p,
3060 struct slave_balance bals[], size_t n_bals)
3063 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3064 swap_bals(p, p - 1);
3066 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3067 swap_bals(p, p + 1);
3073 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3075 if (VLOG_IS_DBG_ENABLED()) {
3076 struct ds ds = DS_EMPTY_INITIALIZER;
3077 const struct slave_balance *b;
3079 for (b = bals; b < bals + n_bals; b++) {
3083 ds_put_char(&ds, ',');
3085 ds_put_format(&ds, " %s %"PRIu64"kB",
3086 b->iface->name, b->tx_bytes / 1024);
3088 if (!b->iface->enabled) {
3089 ds_put_cstr(&ds, " (disabled)");
3091 if (b->n_hashes > 0) {
3092 ds_put_cstr(&ds, " (");
3093 for (i = 0; i < b->n_hashes; i++) {
3094 const struct bond_entry *e = b->hashes[i];
3096 ds_put_cstr(&ds, " + ");
3098 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3099 e - port->bond_hash, e->tx_bytes / 1024);
3101 ds_put_cstr(&ds, ")");
3104 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3109 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3111 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3114 struct bond_entry *hash = from->hashes[hash_idx];
3115 struct port *port = from->iface->port;
3116 uint64_t delta = hash->tx_bytes;
3118 assert(port->bond_mode == BM_SLB);
3120 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3121 "from %s to %s (now carrying %"PRIu64"kB and "
3122 "%"PRIu64"kB load, respectively)",
3123 port->name, delta / 1024, hash - port->bond_hash,
3124 from->iface->name, to->iface->name,
3125 (from->tx_bytes - delta) / 1024,
3126 (to->tx_bytes + delta) / 1024);
3128 /* Delete element from from->hashes.
3130 * We don't bother to add the element to to->hashes because not only would
3131 * it require more work, the only purpose it would be to allow that hash to
3132 * be migrated to another slave in this rebalancing run, and there is no
3133 * point in doing that. */
3134 if (hash_idx == 0) {
3137 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3138 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3142 /* Shift load away from 'from' to 'to'. */
3143 from->tx_bytes -= delta;
3144 to->tx_bytes += delta;
3146 /* Arrange for flows to be revalidated. */
3147 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3148 hash->iface_idx = to->iface->port_ifidx;
3149 hash->iface_tag = tag_create_random();
3153 bond_rebalance_port(struct port *port)
3155 struct slave_balance *bals;
3157 struct bond_entry *hashes[BOND_MASK + 1];
3158 struct slave_balance *b, *from, *to;
3159 struct bond_entry *e;
3162 assert(port->bond_mode == BM_SLB);
3164 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3165 * descending order of tx_bytes, so that bals[0] represents the most
3166 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3169 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3170 * array for each slave_balance structure, we sort our local array of
3171 * hashes in order by slave, so that all of the hashes for a given slave
3172 * become contiguous in memory, and then we point each 'hashes' members of
3173 * a slave_balance structure to the start of a contiguous group. */
3174 n_bals = port->n_ifaces;
3175 bals = xmalloc(n_bals * sizeof *bals);
3176 for (b = bals; b < &bals[n_bals]; b++) {
3177 b->iface = port->ifaces[b - bals];
3182 for (i = 0; i <= BOND_MASK; i++) {
3183 hashes[i] = &port->bond_hash[i];
3185 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3186 for (i = 0; i <= BOND_MASK; i++) {
3188 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3189 b = &bals[e->iface_idx];
3190 b->tx_bytes += e->tx_bytes;
3192 b->hashes = &hashes[i];
3197 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3198 log_bals(bals, n_bals, port);
3200 /* Discard slaves that aren't enabled (which were sorted to the back of the
3201 * array earlier). */
3202 while (!bals[n_bals - 1].iface->enabled) {
3209 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3210 to = &bals[n_bals - 1];
3211 for (from = bals; from < to; ) {
3212 uint64_t overload = from->tx_bytes - to->tx_bytes;
3213 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3214 /* The extra load on 'from' (and all less-loaded slaves), compared
3215 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3216 * it is less than ~1Mbps. No point in rebalancing. */
3218 } else if (from->n_hashes == 1) {
3219 /* 'from' only carries a single MAC hash, so we can't shift any
3220 * load away from it, even though we want to. */
3223 /* 'from' is carrying significantly more load than 'to', and that
3224 * load is split across at least two different hashes. Pick a hash
3225 * to migrate to 'to' (the least-loaded slave), given that doing so
3226 * must decrease the ratio of the load on the two slaves by at
3229 * The sort order we use means that we prefer to shift away the
3230 * smallest hashes instead of the biggest ones. There is little
3231 * reason behind this decision; we could use the opposite sort
3232 * order to shift away big hashes ahead of small ones. */
3235 for (i = 0; i < from->n_hashes; i++) {
3236 double old_ratio, new_ratio;
3237 uint64_t delta = from->hashes[i]->tx_bytes;
3239 if (delta == 0 || from->tx_bytes - delta == 0) {
3240 /* Pointless move. */
3244 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3246 if (to->tx_bytes == 0) {
3247 /* Nothing on the new slave, move it. */
3251 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3252 new_ratio = (double)(from->tx_bytes - delta) /
3253 (to->tx_bytes + delta);
3255 if (new_ratio == 0) {
3256 /* Should already be covered but check to prevent division
3261 if (new_ratio < 1) {
3262 new_ratio = 1 / new_ratio;
3265 if (old_ratio - new_ratio > 0.1) {
3266 /* Would decrease the ratio, move it. */
3270 if (i < from->n_hashes) {
3271 bond_shift_load(from, to, i);
3272 port->bond_compat_is_stale = true;
3274 /* If the result of the migration changed the relative order of
3275 * 'from' and 'to' swap them back to maintain invariants. */
3276 if (order_swapped) {
3277 swap_bals(from, to);
3280 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3281 * point to different slave_balance structures. It is only
3282 * valid to do these two operations in a row at all because we
3283 * know that 'from' will not move past 'to' and vice versa. */
3284 resort_bals(from, bals, n_bals);
3285 resort_bals(to, bals, n_bals);
3292 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3293 * historical data to decay to <1% in 7 rebalancing runs. */
3294 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3303 bond_send_learning_packets(struct port *port)
3305 struct bridge *br = port->bridge;
3306 struct mac_entry *e;
3307 struct ofpbuf packet;
3308 int error, n_packets, n_errors;
3310 if (!port->n_ifaces || port->active_iface < 0) {
3314 ofpbuf_init(&packet, 128);
3315 error = n_packets = n_errors = 0;
3316 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3317 union ofp_action actions[2], *a;
3323 if (e->port == port->port_idx
3324 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3328 /* Compose actions. */
3329 memset(actions, 0, sizeof actions);
3332 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3333 a->vlan_vid.len = htons(sizeof *a);
3334 a->vlan_vid.vlan_vid = htons(e->vlan);
3337 a->output.type = htons(OFPAT_OUTPUT);
3338 a->output.len = htons(sizeof *a);
3339 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3344 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3346 flow_extract(&packet, 0, ODPP_NONE, &flow);
3347 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3354 ofpbuf_uninit(&packet);
3357 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3358 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3359 "packets, last error was: %s",
3360 port->name, n_errors, n_packets, strerror(error));
3362 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3363 port->name, n_packets);
3367 /* Bonding unixctl user interface functions. */
3370 bond_unixctl_list(struct unixctl_conn *conn,
3371 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3373 struct ds ds = DS_EMPTY_INITIALIZER;
3374 const struct bridge *br;
3376 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3378 LIST_FOR_EACH (br, node, &all_bridges) {
3381 for (i = 0; i < br->n_ports; i++) {
3382 const struct port *port = br->ports[i];
3383 if (port->n_ifaces > 1) {
3386 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3387 bond_mode_to_string(port->bond_mode));
3388 for (j = 0; j < port->n_ifaces; j++) {
3389 const struct iface *iface = port->ifaces[j];
3391 ds_put_cstr(&ds, ", ");
3393 ds_put_cstr(&ds, iface->name);
3395 ds_put_char(&ds, '\n');
3399 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3403 static struct port *
3404 bond_find(const char *name)
3406 const struct bridge *br;
3408 LIST_FOR_EACH (br, node, &all_bridges) {
3411 for (i = 0; i < br->n_ports; i++) {
3412 struct port *port = br->ports[i];
3413 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3422 bond_unixctl_show(struct unixctl_conn *conn,
3423 const char *args, void *aux OVS_UNUSED)
3425 struct ds ds = DS_EMPTY_INITIALIZER;
3426 const struct port *port;
3429 port = bond_find(args);
3431 unixctl_command_reply(conn, 501, "no such bond");
3435 ds_put_format(&ds, "bond_mode: %s\n",
3436 bond_mode_to_string(port->bond_mode));
3437 ds_put_format(&ds, "bond-detect-mode: %s\n",
3438 port->miimon ? "miimon" : "carrier");
3441 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3442 port->bond_miimon_interval);
3445 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3446 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3448 if (port->bond_mode == BM_SLB) {
3449 ds_put_format(&ds, "next rebalance: %lld ms\n",
3450 port->bond_next_rebalance - time_msec());
3453 for (j = 0; j < port->n_ifaces; j++) {
3454 const struct iface *iface = port->ifaces[j];
3455 struct bond_entry *be;
3458 ds_put_format(&ds, "slave %s: %s\n",
3459 iface->name, iface->enabled ? "enabled" : "disabled");
3460 if (j == port->active_iface) {
3461 ds_put_cstr(&ds, "\tactive slave\n");
3463 if (iface->delay_expires != LLONG_MAX) {
3464 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3465 iface->enabled ? "downdelay" : "updelay",
3466 iface->delay_expires - time_msec());
3469 if (port->bond_mode != BM_SLB) {
3474 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3475 int hash = be - port->bond_hash;
3476 struct mac_entry *me;
3478 if (be->iface_idx != j) {
3482 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3483 hash, be->tx_bytes / 1024);
3486 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3489 if (bond_hash(me->mac, me->vlan) == hash
3490 && me->port != port->port_idx
3491 && choose_output_iface(port, me->mac, me->vlan,
3493 && dp_ifidx == iface->dp_ifidx)
3495 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3496 ETH_ADDR_ARGS(me->mac));
3501 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3506 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3507 void *aux OVS_UNUSED)
3509 char *args = (char *) args_;
3510 char *save_ptr = NULL;
3511 char *bond_s, *hash_s, *slave_s;
3513 struct iface *iface;
3514 struct bond_entry *entry;
3517 bond_s = strtok_r(args, " ", &save_ptr);
3518 hash_s = strtok_r(NULL, " ", &save_ptr);
3519 slave_s = strtok_r(NULL, " ", &save_ptr);
3521 unixctl_command_reply(conn, 501,
3522 "usage: bond/migrate BOND HASH SLAVE");
3526 port = bond_find(bond_s);
3528 unixctl_command_reply(conn, 501, "no such bond");
3532 if (port->bond_mode != BM_SLB) {
3533 unixctl_command_reply(conn, 501, "not an SLB bond");
3537 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3538 hash = atoi(hash_s) & BOND_MASK;
3540 unixctl_command_reply(conn, 501, "bad hash");
3544 iface = port_lookup_iface(port, slave_s);
3546 unixctl_command_reply(conn, 501, "no such slave");
3550 if (!iface->enabled) {
3551 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3555 entry = &port->bond_hash[hash];
3556 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3557 entry->iface_idx = iface->port_ifidx;
3558 entry->iface_tag = tag_create_random();
3559 port->bond_compat_is_stale = true;
3560 unixctl_command_reply(conn, 200, "migrated");
3564 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3565 void *aux OVS_UNUSED)
3567 char *args = (char *) args_;
3568 char *save_ptr = NULL;
3569 char *bond_s, *slave_s;
3571 struct iface *iface;
3573 bond_s = strtok_r(args, " ", &save_ptr);
3574 slave_s = strtok_r(NULL, " ", &save_ptr);
3576 unixctl_command_reply(conn, 501,
3577 "usage: bond/set-active-slave BOND SLAVE");
3581 port = bond_find(bond_s);
3583 unixctl_command_reply(conn, 501, "no such bond");
3587 iface = port_lookup_iface(port, slave_s);
3589 unixctl_command_reply(conn, 501, "no such slave");
3593 if (!iface->enabled) {
3594 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3598 if (port->active_iface != iface->port_ifidx) {
3599 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3600 port->active_iface = iface->port_ifidx;
3601 port->active_iface_tag = tag_create_random();
3602 VLOG_INFO("port %s: active interface is now %s",
3603 port->name, iface->name);
3604 bond_send_learning_packets(port);
3605 unixctl_command_reply(conn, 200, "done");
3607 unixctl_command_reply(conn, 200, "no change");
3612 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3614 char *args = (char *) args_;
3615 char *save_ptr = NULL;
3616 char *bond_s, *slave_s;
3618 struct iface *iface;
3620 bond_s = strtok_r(args, " ", &save_ptr);
3621 slave_s = strtok_r(NULL, " ", &save_ptr);
3623 unixctl_command_reply(conn, 501,
3624 "usage: bond/enable/disable-slave BOND SLAVE");
3628 port = bond_find(bond_s);
3630 unixctl_command_reply(conn, 501, "no such bond");
3634 iface = port_lookup_iface(port, slave_s);
3636 unixctl_command_reply(conn, 501, "no such slave");
3640 bond_enable_slave(iface, enable);
3641 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3645 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3646 void *aux OVS_UNUSED)
3648 enable_slave(conn, args, true);
3652 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3653 void *aux OVS_UNUSED)
3655 enable_slave(conn, args, false);
3659 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3660 void *aux OVS_UNUSED)
3662 char *args = (char *) args_;
3663 uint8_t mac[ETH_ADDR_LEN];
3667 char *mac_s, *vlan_s;
3668 char *save_ptr = NULL;
3670 mac_s = strtok_r(args, " ", &save_ptr);
3671 vlan_s = strtok_r(NULL, " ", &save_ptr);
3674 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3675 unixctl_command_reply(conn, 501, "invalid vlan");
3679 vlan = OFP_VLAN_NONE;
3682 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3683 == ETH_ADDR_SCAN_COUNT) {
3684 hash = bond_hash(mac, vlan);
3686 hash_cstr = xasprintf("%u", hash);
3687 unixctl_command_reply(conn, 200, hash_cstr);
3690 unixctl_command_reply(conn, 501, "invalid mac");
3697 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3698 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3699 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3700 unixctl_command_register("bond/set-active-slave",
3701 bond_unixctl_set_active_slave, NULL);
3702 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3704 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3706 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3709 /* Port functions. */
3711 static struct port *
3712 port_create(struct bridge *br, const char *name)
3716 port = xzalloc(sizeof *port);
3718 port->port_idx = br->n_ports;
3720 port->trunks = NULL;
3721 port->name = xstrdup(name);
3722 port->active_iface = -1;
3724 if (br->n_ports >= br->allocated_ports) {
3725 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3728 br->ports[br->n_ports++] = port;
3729 shash_add_assert(&br->port_by_name, port->name, port);
3731 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3738 get_port_other_config(const struct ovsrec_port *port, const char *key,
3739 const char *default_value)
3743 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3745 return value ? value : default_value;
3749 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3751 struct shash new_ifaces;
3754 /* Collect list of new interfaces. */
3755 shash_init(&new_ifaces);
3756 for (i = 0; i < cfg->n_interfaces; i++) {
3757 const char *name = cfg->interfaces[i]->name;
3758 shash_add_once(&new_ifaces, name, NULL);
3761 /* Get rid of deleted interfaces. */
3762 for (i = 0; i < port->n_ifaces; ) {
3763 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3764 iface_destroy(port->ifaces[i]);
3770 shash_destroy(&new_ifaces);
3774 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3776 const char *detect_mode;
3777 struct shash new_ifaces;
3778 long long int next_rebalance, miimon_next_update;
3779 unsigned long *trunks;
3785 /* Update settings. */
3786 port->updelay = cfg->bond_updelay;
3787 if (port->updelay < 0) {
3790 port->downdelay = cfg->bond_downdelay;
3791 if (port->downdelay < 0) {
3792 port->downdelay = 0;
3794 port->bond_rebalance_interval = atoi(
3795 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3796 if (port->bond_rebalance_interval < 1000) {
3797 port->bond_rebalance_interval = 1000;
3799 next_rebalance = time_msec() + port->bond_rebalance_interval;
3800 if (port->bond_next_rebalance > next_rebalance) {
3801 port->bond_next_rebalance = next_rebalance;
3804 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
3807 if (!strcmp(detect_mode, "carrier")) {
3808 port->miimon = false;
3809 } else if (!strcmp(detect_mode, "miimon")) {
3810 port->miimon = true;
3812 port->miimon = false;
3813 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
3814 "carrier", port->name, detect_mode);
3817 port->bond_miimon_interval = atoi(
3818 get_port_other_config(cfg, "bond-miimon-interval", "200"));
3819 if (port->bond_miimon_interval < 100) {
3820 port->bond_miimon_interval = 100;
3822 miimon_next_update = time_msec() + port->bond_miimon_interval;
3823 if (port->bond_miimon_next_update > miimon_next_update) {
3824 port->bond_miimon_next_update = miimon_next_update;
3827 if (!port->cfg->bond_mode ||
3828 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
3829 port->bond_mode = BM_SLB;
3830 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
3831 port->bond_mode = BM_AB;
3833 port->bond_mode = BM_SLB;
3834 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
3835 port->name, port->cfg->bond_mode,
3836 bond_mode_to_string(port->bond_mode));
3839 /* Add new interfaces and update 'cfg' member of existing ones. */
3840 shash_init(&new_ifaces);
3841 for (i = 0; i < cfg->n_interfaces; i++) {
3842 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3843 struct iface *iface;
3845 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3846 VLOG_WARN("port %s: %s specified twice as port interface",
3847 port->name, if_cfg->name);
3848 iface_set_ofport(if_cfg, -1);
3852 iface = iface_lookup(port->bridge, if_cfg->name);
3854 if (iface->port != port) {
3855 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3857 port->bridge->name, if_cfg->name, iface->port->name);
3860 iface->cfg = if_cfg;
3862 iface = iface_create(port, if_cfg);
3865 /* Determine interface type. The local port always has type
3866 * "internal". Other ports take their type from the database and
3867 * default to "system" if none is specified. */
3868 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3869 : if_cfg->type[0] ? if_cfg->type
3872 shash_destroy(&new_ifaces);
3877 if (port->n_ifaces < 2) {
3879 if (vlan >= 0 && vlan <= 4095) {
3880 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3885 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3886 * they even work as-is. But they have not been tested. */
3887 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3891 if (port->vlan != vlan) {
3893 bridge_flush(port->bridge);
3896 /* Get trunked VLANs. */
3898 if (vlan < 0 && cfg->n_trunks) {
3901 trunks = bitmap_allocate(4096);
3903 for (i = 0; i < cfg->n_trunks; i++) {
3904 int trunk = cfg->trunks[i];
3906 bitmap_set1(trunks, trunk);
3912 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3913 port->name, cfg->n_trunks);
3915 if (n_errors == cfg->n_trunks) {
3916 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3918 bitmap_free(trunks);
3921 } else if (vlan >= 0 && cfg->n_trunks) {
3922 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3926 ? port->trunks != NULL
3927 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3928 bridge_flush(port->bridge);
3930 bitmap_free(port->trunks);
3931 port->trunks = trunks;
3935 port_destroy(struct port *port)
3938 struct bridge *br = port->bridge;
3942 proc_net_compat_update_vlan(port->name, NULL, 0);
3943 proc_net_compat_update_bond(port->name, NULL);
3945 for (i = 0; i < MAX_MIRRORS; i++) {
3946 struct mirror *m = br->mirrors[i];
3947 if (m && m->out_port == port) {
3952 while (port->n_ifaces > 0) {
3953 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3956 shash_find_and_delete_assert(&br->port_by_name, port->name);
3958 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3959 del->port_idx = port->port_idx;
3961 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3963 netdev_monitor_destroy(port->monitor);
3965 bitmap_free(port->trunks);
3972 static struct port *
3973 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3975 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3976 return iface ? iface->port : NULL;
3979 static struct port *
3980 port_lookup(const struct bridge *br, const char *name)
3982 return shash_find_data(&br->port_by_name, name);
3985 static struct iface *
3986 port_lookup_iface(const struct port *port, const char *name)
3988 struct iface *iface = iface_lookup(port->bridge, name);
3989 return iface && iface->port == port ? iface : NULL;
3993 port_update_bonding(struct port *port)
3995 if (port->monitor) {
3996 netdev_monitor_destroy(port->monitor);
3997 port->monitor = NULL;
3999 if (port->n_ifaces < 2) {
4000 /* Not a bonded port. */
4001 if (port->bond_hash) {
4002 free(port->bond_hash);
4003 port->bond_hash = NULL;
4004 port->bond_compat_is_stale = true;
4007 port->bond_fake_iface = false;
4011 if (port->bond_mode == BM_SLB && !port->bond_hash) {
4012 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4013 for (i = 0; i <= BOND_MASK; i++) {
4014 struct bond_entry *e = &port->bond_hash[i];
4018 port->no_ifaces_tag = tag_create_random();
4019 bond_choose_active_iface(port);
4020 port->bond_next_rebalance
4021 = time_msec() + port->bond_rebalance_interval;
4023 if (port->cfg->bond_fake_iface) {
4024 port->bond_next_fake_iface_update = time_msec();
4026 } else if (port->bond_mode != BM_SLB) {
4027 free(port->bond_hash);
4028 port->bond_hash = NULL;
4030 port->bond_compat_is_stale = true;
4031 port->bond_fake_iface = port->cfg->bond_fake_iface;
4033 if (!port->miimon) {
4034 port->monitor = netdev_monitor_create();
4035 for (i = 0; i < port->n_ifaces; i++) {
4036 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4043 port_update_bond_compat(struct port *port)
4045 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
4046 struct compat_bond bond;
4049 if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) {
4050 proc_net_compat_update_bond(port->name, NULL);
4055 bond.updelay = port->updelay;
4056 bond.downdelay = port->downdelay;
4059 bond.hashes = compat_hashes;
4060 if (port->bond_hash) {
4061 const struct bond_entry *e;
4062 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4063 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
4064 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
4065 cbh->hash = e - port->bond_hash;
4066 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
4071 bond.n_slaves = port->n_ifaces;
4072 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
4073 for (i = 0; i < port->n_ifaces; i++) {
4074 struct iface *iface = port->ifaces[i];
4075 struct compat_bond_slave *slave = &bond.slaves[i];
4076 slave->name = iface->name;
4078 /* We need to make the same determination as the Linux bonding
4079 * code to determine whether a slave should be consider "up".
4080 * The Linux function bond_miimon_inspect() supports four
4081 * BOND_LINK_* states:
4083 * - BOND_LINK_UP: carrier detected, updelay has passed.
4084 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
4085 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
4086 * - BOND_LINK_BACK: carrier detected, updelay in progress.
4088 * The function bond_info_show_slave() only considers BOND_LINK_UP
4089 * to be "up" and anything else to be "down".
4091 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
4095 netdev_get_etheraddr(iface->netdev, slave->mac);
4098 if (port->bond_fake_iface) {
4099 struct netdev *bond_netdev;
4101 if (!netdev_open_default(port->name, &bond_netdev)) {
4103 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
4105 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
4107 netdev_close(bond_netdev);
4111 proc_net_compat_update_bond(port->name, &bond);
4116 port_update_vlan_compat(struct port *port)
4118 struct bridge *br = port->bridge;
4119 char *vlandev_name = NULL;
4121 if (port->vlan > 0) {
4122 /* Figure out the name that the VLAN device should actually have, if it
4123 * existed. This takes some work because the VLAN device would not
4124 * have port->name in its name; rather, it would have the trunk port's
4125 * name, and 'port' would be attached to a bridge that also had the
4126 * VLAN device one of its ports. So we need to find a trunk port that
4127 * includes port->vlan.
4129 * There might be more than one candidate. This doesn't happen on
4130 * XenServer, so if it happens we just pick the first choice in
4131 * alphabetical order instead of creating multiple VLAN devices. */
4133 for (i = 0; i < br->n_ports; i++) {
4134 struct port *p = br->ports[i];
4135 if (port_trunks_vlan(p, port->vlan)
4137 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
4139 uint8_t ea[ETH_ADDR_LEN];
4140 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
4141 if (!eth_addr_is_multicast(ea) &&
4142 !eth_addr_is_reserved(ea) &&
4143 !eth_addr_is_zero(ea)) {
4144 vlandev_name = p->name;
4149 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
4152 /* Interface functions. */
4155 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4158 union ofp_action action;
4160 memset(&action, 0, sizeof action);
4161 action.output.type = htons(OFPAT_OUTPUT);
4162 action.output.len = htons(sizeof action);
4163 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4165 flow_extract(packet, 0, ODPP_NONE, &flow);
4167 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4169 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4170 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4174 static struct iface *
4175 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4177 struct bridge *br = port->bridge;
4178 struct iface *iface;
4179 char *name = if_cfg->name;
4181 iface = xzalloc(sizeof *iface);
4183 iface->port_ifidx = port->n_ifaces;
4184 iface->name = xstrdup(name);
4185 iface->dp_ifidx = -1;
4186 iface->tag = tag_create_random();
4187 iface->delay_expires = LLONG_MAX;
4188 iface->netdev = NULL;
4189 iface->cfg = if_cfg;
4191 shash_add_assert(&br->iface_by_name, iface->name, iface);
4193 if (port->n_ifaces >= port->allocated_ifaces) {
4194 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4195 sizeof *port->ifaces);
4197 port->ifaces[port->n_ifaces++] = iface;
4198 if (port->n_ifaces > 1) {
4199 br->has_bonded_ports = true;
4202 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4210 iface_destroy(struct iface *iface)
4213 struct port *port = iface->port;
4214 struct bridge *br = port->bridge;
4215 bool del_active = port->active_iface == iface->port_ifidx;
4218 if (port->monitor) {
4219 netdev_monitor_remove(port->monitor, iface->netdev);
4222 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4224 if (iface->dp_ifidx >= 0) {
4225 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4228 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4229 del->port_ifidx = iface->port_ifidx;
4231 netdev_close(iface->netdev);
4234 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4235 bond_choose_active_iface(port);
4236 bond_send_learning_packets(port);
4239 cfm_destroy(iface->cfm);
4244 bridge_flush(port->bridge);
4248 static struct iface *
4249 iface_lookup(const struct bridge *br, const char *name)
4251 return shash_find_data(&br->iface_by_name, name);
4254 static struct iface *
4255 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4257 struct iface *iface;
4259 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4260 hash_int(dp_ifidx, 0), &br->ifaces) {
4261 if (iface->dp_ifidx == dp_ifidx) {
4268 /* Set Ethernet address of 'iface', if one is specified in the configuration
4271 iface_set_mac(struct iface *iface)
4273 uint8_t ea[ETH_ADDR_LEN];
4275 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4276 if (eth_addr_is_multicast(ea)) {
4277 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4279 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4280 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4281 iface->name, iface->name);
4283 int error = netdev_set_etheraddr(iface->netdev, ea);
4285 VLOG_ERR("interface %s: setting MAC failed (%s)",
4286 iface->name, strerror(error));
4292 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4294 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4297 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4301 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4303 * The value strings in '*shash' are taken directly from values[], not copied,
4304 * so the caller should not modify or free them. */
4306 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4307 struct shash *shash)
4312 for (i = 0; i < n; i++) {
4313 shash_add(shash, keys[i], values[i]);
4317 /* Creates 'keys' and 'values' arrays from 'shash'.
4319 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4320 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4321 * are populated with with strings taken directly from 'shash' and thus have
4322 * the same ownership of the key-value pairs in shash.
4325 shash_to_ovs_idl_map(struct shash *shash,
4326 char ***keys, char ***values, size_t *n)
4330 struct shash_node *sn;
4332 count = shash_count(shash);
4334 k = xmalloc(count * sizeof *k);
4335 v = xmalloc(count * sizeof *v);
4338 SHASH_FOR_EACH(sn, shash) {
4349 struct iface_delete_queues_cbdata {
4350 struct netdev *netdev;
4351 const struct ovsdb_datum *queues;
4355 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4357 union ovsdb_atom atom;
4359 atom.integer = target;
4360 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4364 iface_delete_queues(unsigned int queue_id,
4365 const struct shash *details OVS_UNUSED, void *cbdata_)
4367 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4369 if (!queue_ids_include(cbdata->queues, queue_id)) {
4370 netdev_delete_queue(cbdata->netdev, queue_id);
4375 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4377 if (!qos || qos->type[0] == '\0') {
4378 netdev_set_qos(iface->netdev, NULL, NULL);
4380 struct iface_delete_queues_cbdata cbdata;
4381 struct shash details;
4384 /* Configure top-level Qos for 'iface'. */
4385 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4386 qos->n_other_config, &details);
4387 netdev_set_qos(iface->netdev, qos->type, &details);
4388 shash_destroy(&details);
4390 /* Deconfigure queues that were deleted. */
4391 cbdata.netdev = iface->netdev;
4392 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4394 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4396 /* Configure queues for 'iface'. */
4397 for (i = 0; i < qos->n_queues; i++) {
4398 const struct ovsrec_queue *queue = qos->value_queues[i];
4399 unsigned int queue_id = qos->key_queues[i];
4401 shash_from_ovs_idl_map(queue->key_other_config,
4402 queue->value_other_config,
4403 queue->n_other_config, &details);
4404 netdev_set_queue(iface->netdev, queue_id, &details);
4405 shash_destroy(&details);
4411 iface_update_cfm(struct iface *iface)
4415 uint16_t *remote_mps;
4416 struct ovsrec_monitor *mon;
4417 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4419 mon = iface->cfg->monitor;
4425 if (netdev_get_etheraddr(iface->netdev, ea)) {
4426 VLOG_WARN("interface %s: Failed to get ethernet address. "
4427 "Skipping Monitor.", iface->name);
4431 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4432 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4437 iface->cfm = cfm_create();
4441 cfm->mpid = mon->mpid;
4442 cfm->interval = mon->interval ? *mon->interval : 1000;
4444 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4445 memcpy(cfm->maid, maid, sizeof cfm->maid);
4447 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4448 for(i = 0; i < mon->n_remote_mps; i++) {
4449 remote_mps[i] = mon->remote_mps[i]->mpid;
4451 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4454 if (!cfm_configure(iface->cfm)) {
4455 cfm_destroy(iface->cfm);
4460 /* Port mirroring. */
4462 static struct mirror *
4463 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4467 for (i = 0; i < MAX_MIRRORS; i++) {
4468 struct mirror *m = br->mirrors[i];
4469 if (m && uuid_equals(uuid, &m->uuid)) {
4477 mirror_reconfigure(struct bridge *br)
4479 unsigned long *rspan_vlans;
4482 /* Get rid of deleted mirrors. */
4483 for (i = 0; i < MAX_MIRRORS; i++) {
4484 struct mirror *m = br->mirrors[i];
4486 const struct ovsdb_datum *mc;
4487 union ovsdb_atom atom;
4489 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4490 atom.uuid = br->mirrors[i]->uuid;
4491 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4497 /* Add new mirrors and reconfigure existing ones. */
4498 for (i = 0; i < br->cfg->n_mirrors; i++) {
4499 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4500 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4502 mirror_reconfigure_one(m, cfg);
4504 mirror_create(br, cfg);
4508 /* Update port reserved status. */
4509 for (i = 0; i < br->n_ports; i++) {
4510 br->ports[i]->is_mirror_output_port = false;
4512 for (i = 0; i < MAX_MIRRORS; i++) {
4513 struct mirror *m = br->mirrors[i];
4514 if (m && m->out_port) {
4515 m->out_port->is_mirror_output_port = true;
4519 /* Update flooded vlans (for RSPAN). */
4521 if (br->cfg->n_flood_vlans) {
4522 rspan_vlans = bitmap_allocate(4096);
4524 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4525 int64_t vlan = br->cfg->flood_vlans[i];
4526 if (vlan >= 0 && vlan < 4096) {
4527 bitmap_set1(rspan_vlans, vlan);
4528 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4531 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4536 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4542 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4547 for (i = 0; ; i++) {
4548 if (i >= MAX_MIRRORS) {
4549 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4550 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4553 if (!br->mirrors[i]) {
4558 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4561 br->mirrors[i] = m = xzalloc(sizeof *m);
4564 m->name = xstrdup(cfg->name);
4565 shash_init(&m->src_ports);
4566 shash_init(&m->dst_ports);
4572 mirror_reconfigure_one(m, cfg);
4576 mirror_destroy(struct mirror *m)
4579 struct bridge *br = m->bridge;
4582 for (i = 0; i < br->n_ports; i++) {
4583 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4584 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4587 shash_destroy(&m->src_ports);
4588 shash_destroy(&m->dst_ports);
4591 m->bridge->mirrors[m->idx] = NULL;
4600 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4601 struct shash *names)
4605 for (i = 0; i < n_ports; i++) {
4606 const char *name = ports[i]->name;
4607 if (port_lookup(m->bridge, name)) {
4608 shash_add_once(names, name, NULL);
4610 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4611 "port %s", m->bridge->name, m->name, name);
4617 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4623 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4625 for (i = 0; i < cfg->n_select_vlan; i++) {
4626 int64_t vlan = cfg->select_vlan[i];
4627 if (vlan < 0 || vlan > 4095) {
4628 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4629 m->bridge->name, m->name, vlan);
4631 (*vlans)[n_vlans++] = vlan;
4638 vlan_is_mirrored(const struct mirror *m, int vlan)
4642 for (i = 0; i < m->n_vlans; i++) {
4643 if (m->vlans[i] == vlan) {
4651 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4655 for (i = 0; i < m->n_vlans; i++) {
4656 if (port_trunks_vlan(p, m->vlans[i])) {
4664 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4666 struct shash src_ports, dst_ports;
4667 mirror_mask_t mirror_bit;
4668 struct port *out_port;
4675 if (strcmp(cfg->name, m->name)) {
4677 m->name = xstrdup(cfg->name);
4680 /* Get output port. */
4681 if (cfg->output_port) {
4682 out_port = port_lookup(m->bridge, cfg->output_port->name);
4684 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4685 m->bridge->name, m->name);
4691 if (cfg->output_vlan) {
4692 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4693 "output vlan; ignoring output vlan",
4694 m->bridge->name, m->name);
4696 } else if (cfg->output_vlan) {
4698 out_vlan = *cfg->output_vlan;
4700 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4701 m->bridge->name, m->name);
4706 shash_init(&src_ports);
4707 shash_init(&dst_ports);
4708 if (cfg->select_all) {
4709 for (i = 0; i < m->bridge->n_ports; i++) {
4710 const char *name = m->bridge->ports[i]->name;
4711 shash_add_once(&src_ports, name, NULL);
4712 shash_add_once(&dst_ports, name, NULL);
4717 /* Get ports, and drop duplicates and ports that don't exist. */
4718 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4720 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4723 /* Get all the vlans, and drop duplicate and invalid vlans. */
4724 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4727 /* Update mirror data. */
4728 if (!shash_equal_keys(&m->src_ports, &src_ports)
4729 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4730 || m->n_vlans != n_vlans
4731 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4732 || m->out_port != out_port
4733 || m->out_vlan != out_vlan) {
4734 bridge_flush(m->bridge);
4736 shash_swap(&m->src_ports, &src_ports);
4737 shash_swap(&m->dst_ports, &dst_ports);
4740 m->n_vlans = n_vlans;
4741 m->out_port = out_port;
4742 m->out_vlan = out_vlan;
4745 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4746 for (i = 0; i < m->bridge->n_ports; i++) {
4747 struct port *port = m->bridge->ports[i];
4749 if (shash_find(&m->src_ports, port->name)
4752 ? port_trunks_any_mirrored_vlan(m, port)
4753 : vlan_is_mirrored(m, port->vlan)))) {
4754 port->src_mirrors |= mirror_bit;
4756 port->src_mirrors &= ~mirror_bit;
4759 if (shash_find(&m->dst_ports, port->name)) {
4760 port->dst_mirrors |= mirror_bit;
4762 port->dst_mirrors &= ~mirror_bit;
4767 shash_destroy(&src_ports);
4768 shash_destroy(&dst_ports);