1 /* Copyright (c) 2008, 2009, 2010 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.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
36 #include "dynamic-string.h"
40 #include "mac-learning.h"
43 #include "ofp-print.h"
45 #include "ofproto/netflow.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
54 #include "socket-util.h"
55 #include "stream-ssl.h"
61 #include "vswitchd/vswitch-idl.h"
62 #include "xenserver.h"
64 #include "sflow_api.h"
66 #define THIS_MODULE VLM_bridge
75 /* These members are always valid. */
76 struct port *port; /* Containing port. */
77 size_t port_ifidx; /* Index within containing port. */
78 char *name; /* Host network device name. */
79 tag_type tag; /* Tag associated with this interface. */
80 long long delay_expires; /* Time after which 'enabled' may change. */
82 /* These members are valid only after bridge_reconfigure() causes them to
84 int dp_ifidx; /* Index within kernel datapath. */
85 struct netdev *netdev; /* Network device. */
86 bool enabled; /* May be chosen for flows? */
88 /* This member is only valid *during* bridge_reconfigure(). */
89 const struct ovsrec_interface *cfg;
92 #define BOND_MASK 0xff
94 int iface_idx; /* Index of assigned iface, or -1 if none. */
95 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
96 tag_type iface_tag; /* Tag associated with iface_idx. */
99 #define MAX_MIRRORS 32
100 typedef uint32_t mirror_mask_t;
101 #define MIRROR_MASK_C(X) UINT32_C(X)
102 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
104 struct bridge *bridge;
108 /* Selection criteria. */
109 struct shash src_ports; /* Name is port name; data is always NULL. */
110 struct shash dst_ports; /* Name is port name; data is always NULL. */
115 struct port *out_port;
119 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
121 struct bridge *bridge;
123 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
124 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
127 /* An ordinary bridge port has 1 interface.
128 * A bridge port for bonding has at least 2 interfaces. */
129 struct iface **ifaces;
130 size_t n_ifaces, allocated_ifaces;
133 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
134 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
135 tag_type active_iface_tag; /* Tag for bcast flows. */
136 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
137 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
138 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
139 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
141 /* Port mirroring info. */
142 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
143 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
144 bool is_mirror_output_port; /* Does port mirroring send frames here? */
146 /* This member is only valid *during* bridge_reconfigure(). */
147 const struct ovsrec_port *cfg;
150 #define DP_MAX_PORTS 255
152 struct list node; /* Node in global list of bridges. */
153 char *name; /* User-specified arbitrary name. */
154 struct mac_learning *ml; /* MAC learning table. */
155 bool sent_config_request; /* Successfully sent config request? */
156 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
158 /* Support for remote controllers. */
159 char *controller; /* NULL if there is no remote controller;
160 * "discover" to do controller discovery;
161 * otherwise a vconn name. */
163 /* OpenFlow switch processing. */
164 struct ofproto *ofproto; /* OpenFlow switch. */
166 /* Kernel datapath information. */
167 struct dpif *dpif; /* Datapath. */
168 struct port_array ifaces; /* Indexed by kernel datapath port number. */
172 size_t n_ports, allocated_ports;
175 bool has_bonded_ports;
176 long long int bond_next_rebalance;
181 /* Flow statistics gathering. */
182 time_t next_stats_request;
184 /* Port mirroring. */
185 struct mirror *mirrors[MAX_MIRRORS];
187 /* This member is only valid *during* bridge_reconfigure(). */
188 const struct ovsrec_bridge *cfg;
191 /* List of all bridges. */
192 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
194 /* Maximum number of datapaths. */
195 enum { DP_MAX = 256 };
197 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
198 static void bridge_destroy(struct bridge *);
199 static struct bridge *bridge_lookup(const char *name);
200 static unixctl_cb_func bridge_unixctl_dump_flows;
201 static int bridge_run_one(struct bridge *);
202 static const struct ovsrec_controller *bridge_get_controller(
203 const struct ovsrec_open_vswitch *ovs_cfg,
204 const struct bridge *br);
205 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
207 static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
209 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
210 static void bridge_fetch_dp_ifaces(struct bridge *);
211 static void bridge_flush(struct bridge *);
212 static void bridge_pick_local_hw_addr(struct bridge *,
213 uint8_t ea[ETH_ADDR_LEN],
214 struct iface **hw_addr_iface);
215 static uint64_t bridge_pick_datapath_id(struct bridge *,
216 const uint8_t bridge_ea[ETH_ADDR_LEN],
217 struct iface *hw_addr_iface);
218 static struct iface *bridge_get_local_iface(struct bridge *);
219 static uint64_t dpid_from_hash(const void *, size_t nbytes);
221 static unixctl_cb_func bridge_unixctl_fdb_show;
223 static void bond_init(void);
224 static void bond_run(struct bridge *);
225 static void bond_wait(struct bridge *);
226 static void bond_rebalance_port(struct port *);
227 static void bond_send_learning_packets(struct port *);
228 static void bond_enable_slave(struct iface *iface, bool enable);
230 static struct port *port_create(struct bridge *, const char *name);
231 static void port_reconfigure(struct port *, const struct ovsrec_port *);
232 static void port_destroy(struct port *);
233 static struct port *port_lookup(const struct bridge *, const char *name);
234 static struct iface *port_lookup_iface(const struct port *, const char *name);
235 static struct port *port_from_dp_ifidx(const struct bridge *,
237 static void port_update_bond_compat(struct port *);
238 static void port_update_vlan_compat(struct port *);
239 static void port_update_bonding(struct port *);
241 static struct mirror *mirror_create(struct bridge *, const char *name);
242 static void mirror_destroy(struct mirror *);
243 static void mirror_reconfigure(struct bridge *);
244 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
245 static bool vlan_is_mirrored(const struct mirror *, int vlan);
247 static struct iface *iface_create(struct port *port,
248 const struct ovsrec_interface *if_cfg);
249 static void iface_destroy(struct iface *);
250 static struct iface *iface_lookup(const struct bridge *, const char *name);
251 static struct iface *iface_from_dp_ifidx(const struct bridge *,
253 static bool iface_is_internal(const struct bridge *, const char *name);
254 static void iface_set_mac(struct iface *);
256 /* Hooks into ofproto processing. */
257 static struct ofhooks bridge_ofhooks;
259 /* Public functions. */
261 /* Adds the name of each interface used by a bridge, including local and
262 * internal ports, to 'svec'. */
264 bridge_get_ifaces(struct svec *svec)
266 struct bridge *br, *next;
269 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
270 for (i = 0; i < br->n_ports; i++) {
271 struct port *port = br->ports[i];
273 for (j = 0; j < port->n_ifaces; j++) {
274 struct iface *iface = port->ifaces[j];
275 if (iface->dp_ifidx < 0) {
276 VLOG_ERR("%s interface not in datapath %s, ignoring",
277 iface->name, dpif_name(br->dpif));
279 if (iface->dp_ifidx != ODPP_LOCAL) {
280 svec_add(svec, iface->name);
289 bridge_init(const struct ovsrec_open_vswitch *cfg)
291 struct svec bridge_names;
292 struct svec dpif_names, dpif_types;
295 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
297 svec_init(&bridge_names);
298 for (i = 0; i < cfg->n_bridges; i++) {
299 svec_add(&bridge_names, cfg->bridges[i]->name);
301 svec_sort(&bridge_names);
303 svec_init(&dpif_names);
304 svec_init(&dpif_types);
305 dp_enumerate_types(&dpif_types);
306 for (i = 0; i < dpif_types.n; i++) {
311 dp_enumerate_names(dpif_types.names[i], &dpif_names);
313 for (j = 0; j < dpif_names.n; j++) {
314 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
316 struct svec all_names;
319 svec_init(&all_names);
320 dpif_get_all_names(dpif, &all_names);
321 for (k = 0; k < all_names.n; k++) {
322 if (svec_contains(&bridge_names, all_names.names[k])) {
328 svec_destroy(&all_names);
333 svec_destroy(&dpif_names);
334 svec_destroy(&dpif_types);
336 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
340 bridge_reconfigure(cfg);
345 config_string_change(const char *value, char **valuep)
347 if (value && (!*valuep || strcmp(value, *valuep))) {
349 *valuep = xstrdup(value);
357 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
359 /* XXX SSL should be configurable on a per-bridge basis.
360 * XXX should be possible to de-configure SSL. */
361 static char *private_key_file;
362 static char *certificate_file;
363 static char *cacert_file;
367 /* XXX We can't un-set SSL settings. */
371 if (config_string_change(ssl->private_key, &private_key_file)) {
372 stream_ssl_set_private_key_file(private_key_file);
375 if (config_string_change(ssl->certificate, &certificate_file)) {
376 stream_ssl_set_certificate_file(certificate_file);
379 /* We assume that even if the filename hasn't changed, if the CA cert
380 * file has been removed, that we want to move back into
381 * boot-strapping mode. This opens a small security hole, because
382 * the old certificate will still be trusted until vSwitch is
383 * restarted. We may want to address this in vconn's SSL library. */
384 if (config_string_change(ssl->ca_cert, &cacert_file)
385 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
386 stream_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert);
391 /* Attempt to create the network device 'iface_name' through the netdev
394 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
397 struct shash_node *node;
398 struct shash options;
402 shash_init(&options);
403 for (i = 0; i < iface_cfg->n_options; i++) {
404 shash_add(&options, iface_cfg->key_options[i],
405 xstrdup(iface_cfg->value_options[i]));
409 struct netdev_options netdev_options;
411 memset(&netdev_options, 0, sizeof netdev_options);
412 netdev_options.name = iface_cfg->name;
413 netdev_options.type = iface_cfg->type;
414 netdev_options.args = &options;
415 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
416 netdev_options.may_create = true;
417 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
418 netdev_options.may_open = true;
421 error = netdev_open(&netdev_options, &iface->netdev);
424 netdev_get_carrier(iface->netdev, &iface->enabled);
426 } else if (iface->netdev) {
427 const char *netdev_type = netdev_get_type(iface->netdev);
428 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
429 ? iface_cfg->type : NULL;
431 if (!iface_type || !strcmp(netdev_type, iface_type)) {
432 error = netdev_reconfigure(iface->netdev, &options);
434 VLOG_WARN("%s: attempting change device type from %s to %s",
435 iface_cfg->name, netdev_type, iface_type);
440 SHASH_FOR_EACH (node, &options) {
443 shash_destroy(&options);
449 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
451 return set_up_iface(iface_cfg, iface, false);
455 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
456 void *aux OVS_UNUSED)
458 if (!iface->netdev) {
459 int error = set_up_iface(iface->cfg, iface, true);
461 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
471 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
472 void *aux OVS_UNUSED)
474 if (iface->dp_ifidx >= 0) {
475 VLOG_DBG("%s has interface %s on port %d",
477 iface->name, iface->dp_ifidx);
480 VLOG_ERR("%s interface not in %s, dropping",
481 iface->name, dpif_name(br->dpif));
487 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
488 void *aux OVS_UNUSED)
490 /* Set policing attributes. */
491 netdev_set_policing(iface->netdev,
492 iface->cfg->ingress_policing_rate,
493 iface->cfg->ingress_policing_burst);
495 /* Set MAC address of internal interfaces other than the local
497 if (iface->dp_ifidx != ODPP_LOCAL
498 && iface_is_internal(br, iface->name)) {
499 iface_set_mac(iface);
505 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
506 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
507 * deletes from 'br' any ports that no longer have any interfaces. */
509 iterate_and_prune_ifaces(struct bridge *br,
510 bool (*cb)(struct bridge *, struct iface *,
516 for (i = 0; i < br->n_ports; ) {
517 struct port *port = br->ports[i];
518 for (j = 0; j < port->n_ifaces; ) {
519 struct iface *iface = port->ifaces[j];
520 if (cb(br, iface, aux)) {
523 iface_destroy(iface);
527 if (port->n_ifaces) {
530 VLOG_ERR("%s port has no interfaces, dropping", port->name);
537 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
539 struct ovsdb_idl_txn *txn;
540 struct shash old_br, new_br;
541 struct shash_node *node;
542 struct bridge *br, *next;
544 int sflow_bridge_number;
546 COVERAGE_INC(bridge_reconfigure);
548 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
550 /* Collect old and new bridges. */
553 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
554 shash_add(&old_br, br->name, br);
556 for (i = 0; i < ovs_cfg->n_bridges; i++) {
557 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
558 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
559 VLOG_WARN("more than one bridge named %s", br_cfg->name);
563 /* Get rid of deleted bridges and add new bridges. */
564 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
565 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
572 SHASH_FOR_EACH (node, &new_br) {
573 const char *br_name = node->name;
574 const struct ovsrec_bridge *br_cfg = node->data;
575 br = shash_find_data(&old_br, br_name);
577 /* If the bridge datapath type has changed, we need to tear it
578 * down and recreate. */
579 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
581 bridge_create(br_cfg);
584 bridge_create(br_cfg);
587 shash_destroy(&old_br);
588 shash_destroy(&new_br);
592 bridge_configure_ssl(ovs_cfg->ssl);
595 /* Reconfigure all bridges. */
596 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
597 bridge_reconfigure_one(ovs_cfg, br);
600 /* Add and delete ports on all datapaths.
602 * The kernel will reject any attempt to add a given port to a datapath if
603 * that port already belongs to a different datapath, so we must do all
604 * port deletions before any port additions. */
605 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
606 struct odp_port *dpif_ports;
608 struct shash want_ifaces;
610 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
611 bridge_get_all_ifaces(br, &want_ifaces);
612 for (i = 0; i < n_dpif_ports; i++) {
613 const struct odp_port *p = &dpif_ports[i];
614 if (!shash_find(&want_ifaces, p->devname)
615 && strcmp(p->devname, br->name)) {
616 int retval = dpif_port_del(br->dpif, p->port);
618 VLOG_ERR("failed to remove %s interface from %s: %s",
619 p->devname, dpif_name(br->dpif),
624 shash_destroy(&want_ifaces);
627 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
628 struct odp_port *dpif_ports;
630 struct shash cur_ifaces, want_ifaces;
631 struct shash_node *node;
633 /* Get the set of interfaces currently in this datapath. */
634 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
635 shash_init(&cur_ifaces);
636 for (i = 0; i < n_dpif_ports; i++) {
637 const char *name = dpif_ports[i].devname;
638 if (!shash_find(&cur_ifaces, name)) {
639 shash_add(&cur_ifaces, name, NULL);
644 /* Get the set of interfaces we want on this datapath. */
645 bridge_get_all_ifaces(br, &want_ifaces);
647 SHASH_FOR_EACH (node, &want_ifaces) {
648 const char *if_name = node->name;
649 struct iface *iface = node->data;
651 if (shash_find(&cur_ifaces, if_name)) {
652 /* Already exists, just reconfigure it. */
654 reconfigure_iface(iface->cfg, iface);
657 /* Need to add to datapath. */
661 /* Add to datapath. */
662 internal = iface_is_internal(br, if_name);
663 error = dpif_port_add(br->dpif, if_name,
664 internal ? ODP_PORT_INTERNAL : 0, NULL);
665 if (error == EFBIG) {
666 VLOG_ERR("ran out of valid port numbers on %s",
667 dpif_name(br->dpif));
670 VLOG_ERR("failed to add %s interface to %s: %s",
671 if_name, dpif_name(br->dpif), strerror(error));
675 shash_destroy(&cur_ifaces);
676 shash_destroy(&want_ifaces);
678 sflow_bridge_number = 0;
679 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
682 struct iface *local_iface;
683 struct iface *hw_addr_iface;
686 bridge_fetch_dp_ifaces(br);
688 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
689 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
691 /* Pick local port hardware address, datapath ID. */
692 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
693 local_iface = bridge_get_local_iface(br);
695 int error = netdev_set_etheraddr(local_iface->netdev, ea);
697 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
698 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
699 "Ethernet address: %s",
700 br->name, strerror(error));
704 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
705 ofproto_set_datapath_id(br->ofproto, dpid);
707 dpid_string = xasprintf("%012"PRIx64, dpid);
708 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
711 /* Set NetFlow configuration on this bridge. */
712 if (br->cfg->netflow) {
713 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
714 struct netflow_options opts;
716 memset(&opts, 0, sizeof opts);
718 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
719 if (nf_cfg->engine_type) {
720 opts.engine_type = *nf_cfg->engine_type;
722 if (nf_cfg->engine_id) {
723 opts.engine_id = *nf_cfg->engine_id;
726 opts.active_timeout = nf_cfg->active_timeout;
727 if (!opts.active_timeout) {
728 opts.active_timeout = -1;
729 } else if (opts.active_timeout < 0) {
730 VLOG_WARN("bridge %s: active timeout interval set to negative "
731 "value, using default instead (%d seconds)", br->name,
732 NF_ACTIVE_TIMEOUT_DEFAULT);
733 opts.active_timeout = -1;
736 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
737 if (opts.add_id_to_iface) {
738 if (opts.engine_id > 0x7f) {
739 VLOG_WARN("bridge %s: netflow port mangling may conflict "
740 "with another vswitch, choose an engine id less "
741 "than 128", br->name);
743 if (br->n_ports > 508) {
744 VLOG_WARN("bridge %s: netflow port mangling will conflict "
745 "with another port when more than 508 ports are "
750 opts.collectors.n = nf_cfg->n_targets;
751 opts.collectors.names = nf_cfg->targets;
752 if (ofproto_set_netflow(br->ofproto, &opts)) {
753 VLOG_ERR("bridge %s: problem setting netflow collectors",
757 ofproto_set_netflow(br->ofproto, NULL);
760 /* Set sFlow configuration on this bridge. */
761 if (br->cfg->sflow) {
762 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
763 const struct ovsrec_controller *ctrl;
764 struct ofproto_sflow_options oso;
766 memset(&oso, 0, sizeof oso);
768 oso.targets.n = sflow_cfg->n_targets;
769 oso.targets.names = sflow_cfg->targets;
771 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
772 if (sflow_cfg->sampling) {
773 oso.sampling_rate = *sflow_cfg->sampling;
776 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
777 if (sflow_cfg->polling) {
778 oso.polling_interval = *sflow_cfg->polling;
781 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
782 if (sflow_cfg->header) {
783 oso.header_len = *sflow_cfg->header;
786 oso.sub_id = sflow_bridge_number++;
787 oso.agent_device = sflow_cfg->agent;
789 ctrl = bridge_get_controller(ovs_cfg, br);
790 oso.control_ip = ctrl ? ctrl->local_ip : NULL;
791 ofproto_set_sflow(br->ofproto, &oso);
793 svec_destroy(&oso.targets);
795 ofproto_set_sflow(br->ofproto, NULL);
798 /* Update the controller and related settings. It would be more
799 * straightforward to call this from bridge_reconfigure_one(), but we
800 * can't do it there for two reasons. First, and most importantly, at
801 * that point we don't know the dp_ifidx of any interfaces that have
802 * been added to the bridge (because we haven't actually added them to
803 * the datapath). Second, at that point we haven't set the datapath ID
804 * yet; when a controller is configured, resetting the datapath ID will
805 * immediately disconnect from the controller, so it's better to set
806 * the datapath ID before the controller. */
807 bridge_reconfigure_controller(ovs_cfg, br);
809 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
810 for (i = 0; i < br->n_ports; i++) {
811 struct port *port = br->ports[i];
813 port_update_vlan_compat(port);
814 port_update_bonding(port);
817 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
818 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
821 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
823 ovsdb_idl_txn_commit(txn);
824 ovsdb_idl_txn_destroy(txn); /* XXX */
828 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
832 for (i = 0; i < br_cfg->n_other_config; i++) {
833 if (!strcmp(br_cfg->key_other_config[i], key)) {
834 return br_cfg->value_other_config[i];
841 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
842 struct iface **hw_addr_iface)
848 *hw_addr_iface = NULL;
850 /* Did the user request a particular MAC? */
851 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
852 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
853 if (eth_addr_is_multicast(ea)) {
854 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
855 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
856 } else if (eth_addr_is_zero(ea)) {
857 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
863 /* Otherwise choose the minimum non-local MAC address among all of the
865 memset(ea, 0xff, sizeof ea);
866 for (i = 0; i < br->n_ports; i++) {
867 struct port *port = br->ports[i];
868 uint8_t iface_ea[ETH_ADDR_LEN];
871 /* Mirror output ports don't participate. */
872 if (port->is_mirror_output_port) {
876 /* Choose the MAC address to represent the port. */
877 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
878 /* Find the interface with this Ethernet address (if any) so that
879 * we can provide the correct devname to the caller. */
881 for (j = 0; j < port->n_ifaces; j++) {
882 struct iface *candidate = port->ifaces[j];
883 uint8_t candidate_ea[ETH_ADDR_LEN];
884 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
885 && eth_addr_equals(iface_ea, candidate_ea)) {
890 /* Choose the interface whose MAC address will represent the port.
891 * The Linux kernel bonding code always chooses the MAC address of
892 * the first slave added to a bond, and the Fedora networking
893 * scripts always add slaves to a bond in alphabetical order, so
894 * for compatibility we choose the interface with the name that is
895 * first in alphabetical order. */
896 iface = port->ifaces[0];
897 for (j = 1; j < port->n_ifaces; j++) {
898 struct iface *candidate = port->ifaces[j];
899 if (strcmp(candidate->name, iface->name) < 0) {
904 /* The local port doesn't count (since we're trying to choose its
905 * MAC address anyway). */
906 if (iface->dp_ifidx == ODPP_LOCAL) {
911 error = netdev_get_etheraddr(iface->netdev, iface_ea);
913 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
914 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
915 iface->name, strerror(error));
920 /* Compare against our current choice. */
921 if (!eth_addr_is_multicast(iface_ea) &&
922 !eth_addr_is_local(iface_ea) &&
923 !eth_addr_is_reserved(iface_ea) &&
924 !eth_addr_is_zero(iface_ea) &&
925 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
927 memcpy(ea, iface_ea, ETH_ADDR_LEN);
928 *hw_addr_iface = iface;
931 if (eth_addr_is_multicast(ea)) {
932 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
933 *hw_addr_iface = NULL;
934 VLOG_WARN("bridge %s: using default bridge Ethernet "
935 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
937 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
938 br->name, ETH_ADDR_ARGS(ea));
942 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
943 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
944 * an interface on 'br', then that interface must be passed in as
945 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
946 * 'hw_addr_iface' must be passed in as a null pointer. */
948 bridge_pick_datapath_id(struct bridge *br,
949 const uint8_t bridge_ea[ETH_ADDR_LEN],
950 struct iface *hw_addr_iface)
953 * The procedure for choosing a bridge MAC address will, in the most
954 * ordinary case, also choose a unique MAC that we can use as a datapath
955 * ID. In some special cases, though, multiple bridges will end up with
956 * the same MAC address. This is OK for the bridges, but it will confuse
957 * the OpenFlow controller, because each datapath needs a unique datapath
960 * Datapath IDs must be unique. It is also very desirable that they be
961 * stable from one run to the next, so that policy set on a datapath
964 const char *datapath_id;
967 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
968 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
974 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
976 * A bridge whose MAC address is taken from a VLAN network device
977 * (that is, a network device created with vconfig(8) or similar
978 * tool) will have the same MAC address as a bridge on the VLAN
979 * device's physical network device.
981 * Handle this case by hashing the physical network device MAC
982 * along with the VLAN identifier.
984 uint8_t buf[ETH_ADDR_LEN + 2];
985 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
986 buf[ETH_ADDR_LEN] = vlan >> 8;
987 buf[ETH_ADDR_LEN + 1] = vlan;
988 return dpid_from_hash(buf, sizeof buf);
991 * Assume that this bridge's MAC address is unique, since it
992 * doesn't fit any of the cases we handle specially.
997 * A purely internal bridge, that is, one that has no non-virtual
998 * network devices on it at all, is more difficult because it has no
999 * natural unique identifier at all.
1001 * When the host is a XenServer, we handle this case by hashing the
1002 * host's UUID with the name of the bridge. Names of bridges are
1003 * persistent across XenServer reboots, although they can be reused if
1004 * an internal network is destroyed and then a new one is later
1005 * created, so this is fairly effective.
1007 * When the host is not a XenServer, we punt by using a random MAC
1008 * address on each run.
1010 const char *host_uuid = xenserver_get_host_uuid();
1012 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1013 dpid = dpid_from_hash(combined, strlen(combined));
1019 return eth_addr_to_uint64(bridge_ea);
1023 dpid_from_hash(const void *data, size_t n)
1025 uint8_t hash[SHA1_DIGEST_SIZE];
1027 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1028 sha1_bytes(data, n, hash);
1029 eth_addr_mark_random(hash);
1030 return eth_addr_to_uint64(hash);
1036 struct bridge *br, *next;
1040 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1041 int error = bridge_run_one(br);
1043 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1044 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1045 "forcing reconfiguration", br->name);
1059 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1060 ofproto_wait(br->ofproto);
1061 if (br->controller) {
1065 mac_learning_wait(br->ml);
1070 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1071 * configuration changes. */
1073 bridge_flush(struct bridge *br)
1075 COVERAGE_INC(bridge_flush);
1077 mac_learning_flush(br->ml);
1080 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1081 * such interface. */
1082 static struct iface *
1083 bridge_get_local_iface(struct bridge *br)
1087 for (i = 0; i < br->n_ports; i++) {
1088 struct port *port = br->ports[i];
1089 for (j = 0; j < port->n_ifaces; j++) {
1090 struct iface *iface = port->ifaces[j];
1091 if (iface->dp_ifidx == ODPP_LOCAL) {
1100 /* Bridge unixctl user interface functions. */
1102 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1103 const char *args, void *aux OVS_UNUSED)
1105 struct ds ds = DS_EMPTY_INITIALIZER;
1106 const struct bridge *br;
1107 const struct mac_entry *e;
1109 br = bridge_lookup(args);
1111 unixctl_command_reply(conn, 501, "no such bridge");
1115 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1116 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1117 if (e->port < 0 || e->port >= br->n_ports) {
1120 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1121 br->ports[e->port]->ifaces[0]->dp_ifidx,
1122 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1124 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1128 /* Bridge reconfiguration functions. */
1129 static struct bridge *
1130 bridge_create(const struct ovsrec_bridge *br_cfg)
1135 assert(!bridge_lookup(br_cfg->name));
1136 br = xzalloc(sizeof *br);
1138 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1144 dpif_flow_flush(br->dpif);
1146 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1149 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1151 dpif_delete(br->dpif);
1152 dpif_close(br->dpif);
1157 br->name = xstrdup(br_cfg->name);
1159 br->ml = mac_learning_create();
1160 br->sent_config_request = false;
1161 eth_addr_nicira_random(br->default_ea);
1163 port_array_init(&br->ifaces);
1166 br->bond_next_rebalance = time_msec() + 10000;
1168 list_push_back(&all_bridges, &br->node);
1170 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1176 bridge_destroy(struct bridge *br)
1181 while (br->n_ports > 0) {
1182 port_destroy(br->ports[br->n_ports - 1]);
1184 list_remove(&br->node);
1185 error = dpif_delete(br->dpif);
1186 if (error && error != ENOENT) {
1187 VLOG_ERR("failed to delete %s: %s",
1188 dpif_name(br->dpif), strerror(error));
1190 dpif_close(br->dpif);
1191 ofproto_destroy(br->ofproto);
1192 free(br->controller);
1193 mac_learning_destroy(br->ml);
1194 port_array_destroy(&br->ifaces);
1201 static struct bridge *
1202 bridge_lookup(const char *name)
1206 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1207 if (!strcmp(br->name, name)) {
1215 bridge_exists(const char *name)
1217 return bridge_lookup(name) ? true : false;
1221 bridge_get_datapathid(const char *name)
1223 struct bridge *br = bridge_lookup(name);
1224 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1227 /* Handle requests for a listing of all flows known by the OpenFlow
1228 * stack, including those normally hidden. */
1230 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1231 const char *args, void *aux OVS_UNUSED)
1236 br = bridge_lookup(args);
1238 unixctl_command_reply(conn, 501, "Unknown bridge");
1243 ofproto_get_all_flows(br->ofproto, &results);
1245 unixctl_command_reply(conn, 200, ds_cstr(&results));
1246 ds_destroy(&results);
1250 bridge_run_one(struct bridge *br)
1254 error = ofproto_run1(br->ofproto);
1259 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1262 error = ofproto_run2(br->ofproto, br->flush);
1268 static const struct ovsrec_controller *
1269 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1270 const struct bridge *br)
1272 const struct ovsrec_controller *controller;
1274 controller = (br->cfg->controller ? br->cfg->controller
1275 : ovs_cfg->controller ? ovs_cfg->controller
1278 if (controller && !strcmp(controller->target, "none")) {
1286 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1288 struct svec *ifaces = ifaces_;
1289 if (!svec_contains(ifaces, iface->name)) {
1290 svec_add(ifaces, iface->name);
1294 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1296 br->name, iface->name, iface->port->name);
1302 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1305 struct shash old_ports, new_ports;
1307 struct svec listeners, old_listeners;
1308 struct svec snoops, old_snoops;
1309 struct shash_node *node;
1312 /* Collect old ports. */
1313 shash_init(&old_ports);
1314 for (i = 0; i < br->n_ports; i++) {
1315 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1318 /* Collect new ports. */
1319 shash_init(&new_ports);
1320 for (i = 0; i < br->cfg->n_ports; i++) {
1321 const char *name = br->cfg->ports[i]->name;
1322 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1323 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1328 /* If we have a controller, then we need a local port. Complain if the
1329 * user didn't specify one.
1331 * XXX perhaps we should synthesize a port ourselves in this case. */
1332 if (bridge_get_controller(ovs_cfg, br)) {
1333 char local_name[IF_NAMESIZE];
1336 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1337 local_name, sizeof local_name);
1338 if (!error && !shash_find(&new_ports, local_name)) {
1339 VLOG_WARN("bridge %s: controller specified but no local port "
1340 "(port named %s) defined",
1341 br->name, local_name);
1345 /* Get rid of deleted ports and add new ports. */
1346 SHASH_FOR_EACH (node, &old_ports) {
1347 if (!shash_find(&new_ports, node->name)) {
1348 port_destroy(node->data);
1351 SHASH_FOR_EACH (node, &new_ports) {
1352 struct port *port = shash_find_data(&old_ports, node->name);
1354 port = port_create(br, node->name);
1356 port_reconfigure(port, node->data);
1358 shash_destroy(&old_ports);
1359 shash_destroy(&new_ports);
1361 /* Check and delete duplicate interfaces. */
1363 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1364 svec_destroy(&ifaces);
1366 /* Delete all flows if we're switching from connected to standalone or vice
1367 * versa. (XXX Should we delete all flows if we are switching from one
1368 * controller to another?) */
1371 /* Configure OpenFlow management listeners. */
1372 svec_init(&listeners);
1373 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1375 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1376 ovs_rundir, br->name));
1377 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1378 svec_clear(&listeners);
1380 svec_sort_unique(&listeners);
1382 svec_init(&old_listeners);
1383 ofproto_get_listeners(br->ofproto, &old_listeners);
1384 svec_sort_unique(&old_listeners);
1386 if (!svec_equal(&listeners, &old_listeners)) {
1387 ofproto_set_listeners(br->ofproto, &listeners);
1389 svec_destroy(&listeners);
1390 svec_destroy(&old_listeners);
1392 /* Configure OpenFlow controller connection snooping. */
1394 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1396 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1397 ovs_rundir, br->name));
1398 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1399 svec_clear(&snoops);
1401 svec_sort_unique(&snoops);
1403 svec_init(&old_snoops);
1404 ofproto_get_snoops(br->ofproto, &old_snoops);
1405 svec_sort_unique(&old_snoops);
1407 if (!svec_equal(&snoops, &old_snoops)) {
1408 ofproto_set_snoops(br->ofproto, &snoops);
1410 svec_destroy(&snoops);
1411 svec_destroy(&old_snoops);
1413 /* Default listener. */
1414 svec_init(&listeners);
1415 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1416 ovs_rundir, br->name));
1417 svec_init(&old_listeners);
1418 ofproto_get_listeners(br->ofproto, &old_listeners);
1419 if (!svec_equal(&listeners, &old_listeners)) {
1420 ofproto_set_listeners(br->ofproto, &listeners);
1422 svec_destroy(&listeners);
1423 svec_destroy(&old_listeners);
1425 /* Default snoop. */
1427 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1428 ovs_rundir, br->name));
1429 svec_init(&old_snoops);
1430 ofproto_get_snoops(br->ofproto, &old_snoops);
1431 if (!svec_equal(&snoops, &old_snoops)) {
1432 ofproto_set_snoops(br->ofproto, &snoops);
1434 svec_destroy(&snoops);
1435 svec_destroy(&old_snoops);
1438 mirror_reconfigure(br);
1442 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1445 char *pfx = xasprintf("bridge.%s.controller", br->name);
1446 const struct ovsrec_controller *c;
1448 c = bridge_get_controller(ovs_cfg, br);
1449 if ((br->controller != NULL) != (c != NULL)) {
1450 ofproto_flush_flows(br->ofproto);
1452 free(br->controller);
1453 br->controller = c ? xstrdup(c->target) : NULL;
1456 int max_backoff, probe;
1457 int rate_limit, burst_limit;
1459 if (!strcmp(c->target, "discover")) {
1460 ofproto_set_discovery(br->ofproto, true,
1461 c->discover_accept_regex,
1462 c->discover_update_resolv_conf);
1464 struct iface *local_iface;
1468 in_band = (!c->connection_mode
1469 || !strcmp(c->connection_mode, "out-of-band"));
1470 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1471 ofproto_set_in_band(br->ofproto, in_band);
1473 local_iface = bridge_get_local_iface(br);
1474 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1475 struct netdev *netdev = local_iface->netdev;
1476 struct in_addr mask, gateway;
1478 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1481 if (!c->local_gateway
1482 || !inet_aton(c->local_gateway, &gateway)) {
1486 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1488 mask.s_addr = guess_netmask(ip.s_addr);
1490 if (!netdev_set_in4(netdev, ip, mask)) {
1491 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1493 br->name, IP_ARGS(&ip.s_addr),
1494 IP_ARGS(&mask.s_addr));
1497 if (gateway.s_addr) {
1498 if (!netdev_add_router(netdev, gateway)) {
1499 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1500 br->name, IP_ARGS(&gateway.s_addr));
1506 ofproto_set_failure(br->ofproto,
1508 || !strcmp(c->fail_mode, "standalone")
1509 || !strcmp(c->fail_mode, "open")));
1511 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1512 ofproto_set_probe_interval(br->ofproto, probe);
1514 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1515 ofproto_set_max_backoff(br->ofproto, max_backoff);
1517 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1518 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1519 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1521 union ofp_action action;
1524 /* Set up a flow that matches every packet and directs them to
1525 * OFPP_NORMAL (which goes to us). */
1526 memset(&action, 0, sizeof action);
1527 action.type = htons(OFPAT_OUTPUT);
1528 action.output.len = htons(sizeof action);
1529 action.output.port = htons(OFPP_NORMAL);
1530 memset(&flow, 0, sizeof flow);
1531 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1534 ofproto_set_in_band(br->ofproto, false);
1535 ofproto_set_max_backoff(br->ofproto, 1);
1536 ofproto_set_probe_interval(br->ofproto, 5);
1537 ofproto_set_failure(br->ofproto, false);
1541 ofproto_set_controller(br->ofproto, br->controller);
1545 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1550 for (i = 0; i < br->n_ports; i++) {
1551 struct port *port = br->ports[i];
1552 for (j = 0; j < port->n_ifaces; j++) {
1553 struct iface *iface = port->ifaces[j];
1554 shash_add_once(ifaces, iface->name, iface);
1556 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1557 shash_add_once(ifaces, port->name, NULL);
1562 /* For robustness, in case the administrator moves around datapath ports behind
1563 * our back, we re-check all the datapath port numbers here.
1565 * This function will set the 'dp_ifidx' members of interfaces that have
1566 * disappeared to -1, so only call this function from a context where those
1567 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1568 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1569 * datapath, which doesn't support UINT16_MAX+1 ports. */
1571 bridge_fetch_dp_ifaces(struct bridge *br)
1573 struct odp_port *dpif_ports;
1574 size_t n_dpif_ports;
1577 /* Reset all interface numbers. */
1578 for (i = 0; i < br->n_ports; i++) {
1579 struct port *port = br->ports[i];
1580 for (j = 0; j < port->n_ifaces; j++) {
1581 struct iface *iface = port->ifaces[j];
1582 iface->dp_ifidx = -1;
1585 port_array_clear(&br->ifaces);
1587 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1588 for (i = 0; i < n_dpif_ports; i++) {
1589 struct odp_port *p = &dpif_ports[i];
1590 struct iface *iface = iface_lookup(br, p->devname);
1592 if (iface->dp_ifidx >= 0) {
1593 VLOG_WARN("%s reported interface %s twice",
1594 dpif_name(br->dpif), p->devname);
1595 } else if (iface_from_dp_ifidx(br, p->port)) {
1596 VLOG_WARN("%s reported interface %"PRIu16" twice",
1597 dpif_name(br->dpif), p->port);
1599 port_array_set(&br->ifaces, p->port, iface);
1600 iface->dp_ifidx = p->port;
1604 int64_t ofport = (iface->dp_ifidx >= 0
1605 ? odp_port_to_ofp_port(iface->dp_ifidx)
1607 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1614 /* Bridge packet processing functions. */
1617 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1619 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1622 static struct bond_entry *
1623 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1625 return &port->bond_hash[bond_hash(mac)];
1629 bond_choose_iface(const struct port *port)
1631 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1632 size_t i, best_down_slave = -1;
1633 long long next_delay_expiration = LLONG_MAX;
1635 for (i = 0; i < port->n_ifaces; i++) {
1636 struct iface *iface = port->ifaces[i];
1638 if (iface->enabled) {
1640 } else if (iface->delay_expires < next_delay_expiration) {
1641 best_down_slave = i;
1642 next_delay_expiration = iface->delay_expires;
1646 if (best_down_slave != -1) {
1647 struct iface *iface = port->ifaces[best_down_slave];
1649 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1650 "since no other interface is up", iface->name,
1651 iface->delay_expires - time_msec());
1652 bond_enable_slave(iface, true);
1655 return best_down_slave;
1659 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1660 uint16_t *dp_ifidx, tag_type *tags)
1662 struct iface *iface;
1664 assert(port->n_ifaces);
1665 if (port->n_ifaces == 1) {
1666 iface = port->ifaces[0];
1668 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1669 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1670 || !port->ifaces[e->iface_idx]->enabled) {
1671 /* XXX select interface properly. The current interface selection
1672 * is only good for testing the rebalancing code. */
1673 e->iface_idx = bond_choose_iface(port);
1674 if (e->iface_idx < 0) {
1675 *tags |= port->no_ifaces_tag;
1678 e->iface_tag = tag_create_random();
1679 ((struct port *) port)->bond_compat_is_stale = true;
1681 *tags |= e->iface_tag;
1682 iface = port->ifaces[e->iface_idx];
1684 *dp_ifidx = iface->dp_ifidx;
1685 *tags |= iface->tag; /* Currently only used for bonding. */
1690 bond_link_status_update(struct iface *iface, bool carrier)
1692 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1693 struct port *port = iface->port;
1695 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1696 /* Nothing to do. */
1699 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1700 iface->name, carrier ? "detected" : "dropped");
1701 if (carrier == iface->enabled) {
1702 iface->delay_expires = LLONG_MAX;
1703 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1704 iface->name, carrier ? "disabled" : "enabled");
1705 } else if (carrier && port->active_iface < 0) {
1706 bond_enable_slave(iface, true);
1707 if (port->updelay) {
1708 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1709 "other interface is up", iface->name, port->updelay);
1712 int delay = carrier ? port->updelay : port->downdelay;
1713 iface->delay_expires = time_msec() + delay;
1716 "interface %s: will be %s if it stays %s for %d ms",
1718 carrier ? "enabled" : "disabled",
1719 carrier ? "up" : "down",
1726 bond_choose_active_iface(struct port *port)
1728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1730 port->active_iface = bond_choose_iface(port);
1731 port->active_iface_tag = tag_create_random();
1732 if (port->active_iface >= 0) {
1733 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1734 port->name, port->ifaces[port->active_iface]->name);
1736 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1742 bond_enable_slave(struct iface *iface, bool enable)
1744 struct port *port = iface->port;
1745 struct bridge *br = port->bridge;
1747 /* This acts as a recursion check. If the act of disabling a slave
1748 * causes a different slave to be enabled, the flag will allow us to
1749 * skip redundant work when we reenter this function. It must be
1750 * cleared on exit to keep things safe with multiple bonds. */
1751 static bool moving_active_iface = false;
1753 iface->delay_expires = LLONG_MAX;
1754 if (enable == iface->enabled) {
1758 iface->enabled = enable;
1759 if (!iface->enabled) {
1760 VLOG_WARN("interface %s: disabled", iface->name);
1761 ofproto_revalidate(br->ofproto, iface->tag);
1762 if (iface->port_ifidx == port->active_iface) {
1763 ofproto_revalidate(br->ofproto,
1764 port->active_iface_tag);
1766 /* Disabling a slave can lead to another slave being immediately
1767 * enabled if there will be no active slaves but one is waiting
1768 * on an updelay. In this case we do not need to run most of the
1769 * code for the newly enabled slave since there was no period
1770 * without an active slave and it is redundant with the disabling
1772 moving_active_iface = true;
1773 bond_choose_active_iface(port);
1775 bond_send_learning_packets(port);
1777 VLOG_WARN("interface %s: enabled", iface->name);
1778 if (port->active_iface < 0 && !moving_active_iface) {
1779 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1780 bond_choose_active_iface(port);
1781 bond_send_learning_packets(port);
1783 iface->tag = tag_create_random();
1786 moving_active_iface = false;
1787 port->bond_compat_is_stale = true;
1791 bond_run(struct bridge *br)
1795 for (i = 0; i < br->n_ports; i++) {
1796 struct port *port = br->ports[i];
1798 if (port->n_ifaces >= 2) {
1799 for (j = 0; j < port->n_ifaces; j++) {
1800 struct iface *iface = port->ifaces[j];
1801 if (time_msec() >= iface->delay_expires) {
1802 bond_enable_slave(iface, !iface->enabled);
1807 if (port->bond_compat_is_stale) {
1808 port->bond_compat_is_stale = false;
1809 port_update_bond_compat(port);
1815 bond_wait(struct bridge *br)
1819 for (i = 0; i < br->n_ports; i++) {
1820 struct port *port = br->ports[i];
1821 if (port->n_ifaces < 2) {
1824 for (j = 0; j < port->n_ifaces; j++) {
1825 struct iface *iface = port->ifaces[j];
1826 if (iface->delay_expires != LLONG_MAX) {
1827 poll_timer_wait(iface->delay_expires - time_msec());
1834 set_dst(struct dst *p, const flow_t *flow,
1835 const struct port *in_port, const struct port *out_port,
1838 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1839 : in_port->vlan >= 0 ? in_port->vlan
1840 : ntohs(flow->dl_vlan));
1841 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1845 swap_dst(struct dst *p, struct dst *q)
1847 struct dst tmp = *p;
1852 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1853 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1854 * that we push to the datapath. We could in fact fully sort the array by
1855 * vlan, but in most cases there are at most two different vlan tags so that's
1856 * possibly overkill.) */
1858 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1860 struct dst *first = dsts;
1861 struct dst *last = dsts + n_dsts;
1863 while (first != last) {
1865 * - All dsts < first have vlan == 'vlan'.
1866 * - All dsts >= last have vlan != 'vlan'.
1867 * - first < last. */
1868 while (first->vlan == vlan) {
1869 if (++first == last) {
1874 /* Same invariants, plus one additional:
1875 * - first->vlan != vlan.
1877 while (last[-1].vlan != vlan) {
1878 if (--last == first) {
1883 /* Same invariants, plus one additional:
1884 * - last[-1].vlan == vlan.*/
1885 swap_dst(first++, --last);
1890 mirror_mask_ffs(mirror_mask_t mask)
1892 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1897 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1898 const struct dst *test)
1901 for (i = 0; i < n_dsts; i++) {
1902 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1910 port_trunks_vlan(const struct port *port, uint16_t vlan)
1912 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1916 port_includes_vlan(const struct port *port, uint16_t vlan)
1918 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1922 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1923 const struct port *in_port, const struct port *out_port,
1924 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1926 mirror_mask_t mirrors = in_port->src_mirrors;
1927 struct dst *dst = dsts;
1930 if (out_port == FLOOD_PORT) {
1931 /* XXX use ODP_FLOOD if no vlans or bonding. */
1932 /* XXX even better, define each VLAN as a datapath port group */
1933 for (i = 0; i < br->n_ports; i++) {
1934 struct port *port = br->ports[i];
1935 if (port != in_port && port_includes_vlan(port, vlan)
1936 && !port->is_mirror_output_port
1937 && set_dst(dst, flow, in_port, port, tags)) {
1938 mirrors |= port->dst_mirrors;
1942 *nf_output_iface = NF_OUT_FLOOD;
1943 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1944 *nf_output_iface = dst->dp_ifidx;
1945 mirrors |= out_port->dst_mirrors;
1950 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1951 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1953 if (set_dst(dst, flow, in_port, m->out_port, tags)
1954 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1958 for (i = 0; i < br->n_ports; i++) {
1959 struct port *port = br->ports[i];
1960 if (port_includes_vlan(port, m->out_vlan)
1961 && set_dst(dst, flow, in_port, port, tags))
1965 if (port->vlan < 0) {
1966 dst->vlan = m->out_vlan;
1968 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1972 /* Use the vlan tag on the original flow instead of
1973 * the one passed in the vlan parameter. This ensures
1974 * that we compare the vlan from before any implicit
1975 * tagging tags place. This is necessary because
1976 * dst->vlan is the final vlan, after removing implicit
1978 flow_vlan = ntohs(flow->dl_vlan);
1979 if (flow_vlan == 0) {
1980 flow_vlan = OFP_VLAN_NONE;
1982 if (port == in_port && dst->vlan == flow_vlan) {
1983 /* Don't send out input port on same VLAN. */
1991 mirrors &= mirrors - 1;
1994 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1998 static void OVS_UNUSED
1999 print_dsts(const struct dst *dsts, size_t n)
2001 for (; n--; dsts++) {
2002 printf(">p%"PRIu16, dsts->dp_ifidx);
2003 if (dsts->vlan != OFP_VLAN_NONE) {
2004 printf("v%"PRIu16, dsts->vlan);
2010 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2011 const struct port *in_port, const struct port *out_port,
2012 tag_type *tags, struct odp_actions *actions,
2013 uint16_t *nf_output_iface)
2015 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2017 const struct dst *p;
2020 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2023 cur_vlan = ntohs(flow->dl_vlan);
2024 for (p = dsts; p < &dsts[n_dsts]; p++) {
2025 union odp_action *a;
2026 if (p->vlan != cur_vlan) {
2027 if (p->vlan == OFP_VLAN_NONE) {
2028 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2030 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2031 a->vlan_vid.vlan_vid = htons(p->vlan);
2035 a = odp_actions_add(actions, ODPAT_OUTPUT);
2036 a->output.port = p->dp_ifidx;
2040 /* Returns the effective vlan of a packet, taking into account both the
2041 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2042 * the packet is untagged and -1 indicates it has an invalid header and
2043 * should be dropped. */
2044 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2045 struct port *in_port, bool have_packet)
2047 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2048 * belongs to VLAN 0, so we should treat both cases identically. (In the
2049 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2050 * presumably to allow a priority to be specified. In the latter case, the
2051 * packet does not have any 802.1Q header.) */
2052 int vlan = ntohs(flow->dl_vlan);
2053 if (vlan == OFP_VLAN_NONE) {
2056 if (in_port->vlan >= 0) {
2058 /* XXX support double tagging? */
2060 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2061 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2062 "packet received on port %s configured with "
2063 "implicit VLAN %"PRIu16,
2064 br->name, ntohs(flow->dl_vlan),
2065 in_port->name, in_port->vlan);
2069 vlan = in_port->vlan;
2071 if (!port_includes_vlan(in_port, vlan)) {
2073 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2074 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2075 "packet received on port %s not configured for "
2077 br->name, vlan, in_port->name, vlan);
2087 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2088 struct port *in_port)
2090 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2091 vlan, in_port->port_idx);
2093 /* The log messages here could actually be useful in debugging,
2094 * so keep the rate limit relatively high. */
2095 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2097 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2098 "on port %s in VLAN %d",
2099 br->name, ETH_ADDR_ARGS(flow->dl_src),
2100 in_port->name, vlan);
2101 ofproto_revalidate(br->ofproto, rev_tag);
2106 is_bcast_arp_reply(const flow_t *flow)
2108 return (flow->dl_type == htons(ETH_TYPE_ARP)
2109 && flow->nw_proto == ARP_OP_REPLY
2110 && eth_addr_is_broadcast(flow->dl_dst));
2113 /* If the composed actions may be applied to any packet in the given 'flow',
2114 * returns true. Otherwise, the actions should only be applied to 'packet', or
2115 * not at all, if 'packet' was NULL. */
2117 process_flow(struct bridge *br, const flow_t *flow,
2118 const struct ofpbuf *packet, struct odp_actions *actions,
2119 tag_type *tags, uint16_t *nf_output_iface)
2121 struct iface *in_iface;
2122 struct port *in_port;
2123 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2127 /* Find the interface and port structure for the received packet. */
2128 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2130 /* No interface? Something fishy... */
2131 if (packet != NULL) {
2132 /* Odd. A few possible reasons here:
2134 * - We deleted an interface but there are still a few packets
2135 * queued up from it.
2137 * - Someone externally added an interface (e.g. with "ovs-dpctl
2138 * add-if") that we don't know about.
2140 * - Packet arrived on the local port but the local port is not
2141 * one of our bridge ports.
2143 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2145 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2146 "interface %"PRIu16, br->name, flow->in_port);
2149 /* Return without adding any actions, to drop packets on this flow. */
2152 in_port = in_iface->port;
2153 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2158 /* Drop frames for reserved multicast addresses. */
2159 if (eth_addr_is_reserved(flow->dl_dst)) {
2163 /* Drop frames on ports reserved for mirroring. */
2164 if (in_port->is_mirror_output_port) {
2165 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2166 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2167 "which is reserved exclusively for mirroring",
2168 br->name, in_port->name);
2172 /* Packets received on bonds need special attention to avoid duplicates. */
2173 if (in_port->n_ifaces > 1) {
2176 if (eth_addr_is_multicast(flow->dl_dst)) {
2177 *tags |= in_port->active_iface_tag;
2178 if (in_port->active_iface != in_iface->port_ifidx) {
2179 /* Drop all multicast packets on inactive slaves. */
2184 /* Drop all packets for which we have learned a different input
2185 * port, because we probably sent the packet on one slave and got
2186 * it back on the other. Broadcast ARP replies are an exception
2187 * to this rule: the host has moved to another switch. */
2188 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2189 if (src_idx != -1 && src_idx != in_port->port_idx &&
2190 !is_bcast_arp_reply(flow)) {
2196 out_port = FLOOD_PORT;
2197 /* Learn source MAC (but don't try to learn from revalidation). */
2199 update_learning_table(br, flow, vlan, in_port);
2202 /* Determine output port. */
2203 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2205 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2206 out_port = br->ports[out_port_idx];
2207 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2208 /* If we are revalidating but don't have a learning entry then
2209 * eject the flow. Installing a flow that floods packets opens
2210 * up a window of time where we could learn from a packet reflected
2211 * on a bond and blackhole packets before the learning table is
2212 * updated to reflect the correct port. */
2216 /* Don't send packets out their input ports. */
2217 if (in_port == out_port) {
2222 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2228 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2231 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2232 const struct ofp_phy_port *opp,
2235 struct bridge *br = br_;
2236 struct iface *iface;
2239 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2245 if (reason == OFPPR_DELETE) {
2246 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2247 br->name, iface->name);
2248 iface_destroy(iface);
2249 if (!port->n_ifaces) {
2250 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2251 br->name, port->name);
2257 if (port->n_ifaces > 1) {
2258 bool up = !(opp->state & OFPPS_LINK_DOWN);
2259 bond_link_status_update(iface, up);
2260 port_update_bond_compat(port);
2266 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2267 struct odp_actions *actions, tag_type *tags,
2268 uint16_t *nf_output_iface, void *br_)
2270 struct bridge *br = br_;
2272 COVERAGE_INC(bridge_process_flow);
2273 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2277 bridge_account_flow_ofhook_cb(const flow_t *flow,
2278 const union odp_action *actions,
2279 size_t n_actions, unsigned long long int n_bytes,
2282 struct bridge *br = br_;
2283 struct port *in_port;
2284 const union odp_action *a;
2286 /* Feed information from the active flows back into the learning table
2287 * to ensure that table is always in sync with what is actually flowing
2288 * through the datapath. */
2289 in_port = port_from_dp_ifidx(br, flow->in_port);
2291 int vlan = flow_get_vlan(br, flow, in_port, false);
2293 update_learning_table(br, flow, vlan, in_port);
2297 if (!br->has_bonded_ports) {
2301 for (a = actions; a < &actions[n_actions]; a++) {
2302 if (a->type == ODPAT_OUTPUT) {
2303 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2304 if (out_port && out_port->n_ifaces >= 2) {
2305 struct bond_entry *e = lookup_bond_entry(out_port,
2307 e->tx_bytes += n_bytes;
2314 bridge_account_checkpoint_ofhook_cb(void *br_)
2316 struct bridge *br = br_;
2319 if (!br->has_bonded_ports) {
2323 /* The current ofproto implementation calls this callback at least once a
2324 * second, so this timer implementation is sufficient. */
2325 if (time_msec() < br->bond_next_rebalance) {
2328 br->bond_next_rebalance = time_msec() + 10000;
2330 for (i = 0; i < br->n_ports; i++) {
2331 struct port *port = br->ports[i];
2332 if (port->n_ifaces > 1) {
2333 bond_rebalance_port(port);
2338 static struct ofhooks bridge_ofhooks = {
2339 bridge_port_changed_ofhook_cb,
2340 bridge_normal_ofhook_cb,
2341 bridge_account_flow_ofhook_cb,
2342 bridge_account_checkpoint_ofhook_cb,
2345 /* Bonding functions. */
2347 /* Statistics for a single interface on a bonded port, used for load-based
2348 * bond rebalancing. */
2349 struct slave_balance {
2350 struct iface *iface; /* The interface. */
2351 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2353 /* All the "bond_entry"s that are assigned to this interface, in order of
2354 * increasing tx_bytes. */
2355 struct bond_entry **hashes;
2359 /* Sorts pointers to pointers to bond_entries in ascending order by the
2360 * interface to which they are assigned, and within a single interface in
2361 * ascending order of bytes transmitted. */
2363 compare_bond_entries(const void *a_, const void *b_)
2365 const struct bond_entry *const *ap = a_;
2366 const struct bond_entry *const *bp = b_;
2367 const struct bond_entry *a = *ap;
2368 const struct bond_entry *b = *bp;
2369 if (a->iface_idx != b->iface_idx) {
2370 return a->iface_idx > b->iface_idx ? 1 : -1;
2371 } else if (a->tx_bytes != b->tx_bytes) {
2372 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2378 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2379 * *descending* order by number of bytes transmitted. */
2381 compare_slave_balance(const void *a_, const void *b_)
2383 const struct slave_balance *a = a_;
2384 const struct slave_balance *b = b_;
2385 if (a->iface->enabled != b->iface->enabled) {
2386 return a->iface->enabled ? -1 : 1;
2387 } else if (a->tx_bytes != b->tx_bytes) {
2388 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2395 swap_bals(struct slave_balance *a, struct slave_balance *b)
2397 struct slave_balance tmp = *a;
2402 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2403 * given that 'p' (and only 'p') might be in the wrong location.
2405 * This function invalidates 'p', since it might now be in a different memory
2408 resort_bals(struct slave_balance *p,
2409 struct slave_balance bals[], size_t n_bals)
2412 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2413 swap_bals(p, p - 1);
2415 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2416 swap_bals(p, p + 1);
2422 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2424 if (VLOG_IS_DBG_ENABLED()) {
2425 struct ds ds = DS_EMPTY_INITIALIZER;
2426 const struct slave_balance *b;
2428 for (b = bals; b < bals + n_bals; b++) {
2432 ds_put_char(&ds, ',');
2434 ds_put_format(&ds, " %s %"PRIu64"kB",
2435 b->iface->name, b->tx_bytes / 1024);
2437 if (!b->iface->enabled) {
2438 ds_put_cstr(&ds, " (disabled)");
2440 if (b->n_hashes > 0) {
2441 ds_put_cstr(&ds, " (");
2442 for (i = 0; i < b->n_hashes; i++) {
2443 const struct bond_entry *e = b->hashes[i];
2445 ds_put_cstr(&ds, " + ");
2447 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2448 e - port->bond_hash, e->tx_bytes / 1024);
2450 ds_put_cstr(&ds, ")");
2453 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2458 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2460 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2463 struct bond_entry *hash = from->hashes[hash_idx];
2464 struct port *port = from->iface->port;
2465 uint64_t delta = hash->tx_bytes;
2467 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2468 "from %s to %s (now carrying %"PRIu64"kB and "
2469 "%"PRIu64"kB load, respectively)",
2470 port->name, delta / 1024, hash - port->bond_hash,
2471 from->iface->name, to->iface->name,
2472 (from->tx_bytes - delta) / 1024,
2473 (to->tx_bytes + delta) / 1024);
2475 /* Delete element from from->hashes.
2477 * We don't bother to add the element to to->hashes because not only would
2478 * it require more work, the only purpose it would be to allow that hash to
2479 * be migrated to another slave in this rebalancing run, and there is no
2480 * point in doing that. */
2481 if (hash_idx == 0) {
2484 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2485 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2489 /* Shift load away from 'from' to 'to'. */
2490 from->tx_bytes -= delta;
2491 to->tx_bytes += delta;
2493 /* Arrange for flows to be revalidated. */
2494 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2495 hash->iface_idx = to->iface->port_ifidx;
2496 hash->iface_tag = tag_create_random();
2500 bond_rebalance_port(struct port *port)
2502 struct slave_balance bals[DP_MAX_PORTS];
2504 struct bond_entry *hashes[BOND_MASK + 1];
2505 struct slave_balance *b, *from, *to;
2506 struct bond_entry *e;
2509 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2510 * descending order of tx_bytes, so that bals[0] represents the most
2511 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2514 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2515 * array for each slave_balance structure, we sort our local array of
2516 * hashes in order by slave, so that all of the hashes for a given slave
2517 * become contiguous in memory, and then we point each 'hashes' members of
2518 * a slave_balance structure to the start of a contiguous group. */
2519 n_bals = port->n_ifaces;
2520 for (b = bals; b < &bals[n_bals]; b++) {
2521 b->iface = port->ifaces[b - bals];
2526 for (i = 0; i <= BOND_MASK; i++) {
2527 hashes[i] = &port->bond_hash[i];
2529 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2530 for (i = 0; i <= BOND_MASK; i++) {
2532 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2533 b = &bals[e->iface_idx];
2534 b->tx_bytes += e->tx_bytes;
2536 b->hashes = &hashes[i];
2541 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2542 log_bals(bals, n_bals, port);
2544 /* Discard slaves that aren't enabled (which were sorted to the back of the
2545 * array earlier). */
2546 while (!bals[n_bals - 1].iface->enabled) {
2553 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2554 to = &bals[n_bals - 1];
2555 for (from = bals; from < to; ) {
2556 uint64_t overload = from->tx_bytes - to->tx_bytes;
2557 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2558 /* The extra load on 'from' (and all less-loaded slaves), compared
2559 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2560 * it is less than ~1Mbps. No point in rebalancing. */
2562 } else if (from->n_hashes == 1) {
2563 /* 'from' only carries a single MAC hash, so we can't shift any
2564 * load away from it, even though we want to. */
2567 /* 'from' is carrying significantly more load than 'to', and that
2568 * load is split across at least two different hashes. Pick a hash
2569 * to migrate to 'to' (the least-loaded slave), given that doing so
2570 * must decrease the ratio of the load on the two slaves by at
2573 * The sort order we use means that we prefer to shift away the
2574 * smallest hashes instead of the biggest ones. There is little
2575 * reason behind this decision; we could use the opposite sort
2576 * order to shift away big hashes ahead of small ones. */
2580 for (i = 0; i < from->n_hashes; i++) {
2581 double old_ratio, new_ratio;
2582 uint64_t delta = from->hashes[i]->tx_bytes;
2584 if (delta == 0 || from->tx_bytes - delta == 0) {
2585 /* Pointless move. */
2589 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2591 if (to->tx_bytes == 0) {
2592 /* Nothing on the new slave, move it. */
2596 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2597 new_ratio = (double)(from->tx_bytes - delta) /
2598 (to->tx_bytes + delta);
2600 if (new_ratio == 0) {
2601 /* Should already be covered but check to prevent division
2606 if (new_ratio < 1) {
2607 new_ratio = 1 / new_ratio;
2610 if (old_ratio - new_ratio > 0.1) {
2611 /* Would decrease the ratio, move it. */
2615 if (i < from->n_hashes) {
2616 bond_shift_load(from, to, i);
2617 port->bond_compat_is_stale = true;
2619 /* If the result of the migration changed the relative order of
2620 * 'from' and 'to' swap them back to maintain invariants. */
2621 if (order_swapped) {
2622 swap_bals(from, to);
2625 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2626 * point to different slave_balance structures. It is only
2627 * valid to do these two operations in a row at all because we
2628 * know that 'from' will not move past 'to' and vice versa. */
2629 resort_bals(from, bals, n_bals);
2630 resort_bals(to, bals, n_bals);
2637 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2638 * historical data to decay to <1% in 7 rebalancing runs. */
2639 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2645 bond_send_learning_packets(struct port *port)
2647 struct bridge *br = port->bridge;
2648 struct mac_entry *e;
2649 struct ofpbuf packet;
2650 int error, n_packets, n_errors;
2652 if (!port->n_ifaces || port->active_iface < 0) {
2656 ofpbuf_init(&packet, 128);
2657 error = n_packets = n_errors = 0;
2658 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2659 union ofp_action actions[2], *a;
2665 if (e->port == port->port_idx
2666 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2670 /* Compose actions. */
2671 memset(actions, 0, sizeof actions);
2674 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2675 a->vlan_vid.len = htons(sizeof *a);
2676 a->vlan_vid.vlan_vid = htons(e->vlan);
2679 a->output.type = htons(OFPAT_OUTPUT);
2680 a->output.len = htons(sizeof *a);
2681 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2686 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2688 flow_extract(&packet, ODPP_NONE, &flow);
2689 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2696 ofpbuf_uninit(&packet);
2699 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2700 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2701 "packets, last error was: %s",
2702 port->name, n_errors, n_packets, strerror(error));
2704 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2705 port->name, n_packets);
2709 /* Bonding unixctl user interface functions. */
2712 bond_unixctl_list(struct unixctl_conn *conn,
2713 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2715 struct ds ds = DS_EMPTY_INITIALIZER;
2716 const struct bridge *br;
2718 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2720 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2723 for (i = 0; i < br->n_ports; i++) {
2724 const struct port *port = br->ports[i];
2725 if (port->n_ifaces > 1) {
2728 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2729 for (j = 0; j < port->n_ifaces; j++) {
2730 const struct iface *iface = port->ifaces[j];
2732 ds_put_cstr(&ds, ", ");
2734 ds_put_cstr(&ds, iface->name);
2736 ds_put_char(&ds, '\n');
2740 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2744 static struct port *
2745 bond_find(const char *name)
2747 const struct bridge *br;
2749 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2752 for (i = 0; i < br->n_ports; i++) {
2753 struct port *port = br->ports[i];
2754 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2763 bond_unixctl_show(struct unixctl_conn *conn,
2764 const char *args, void *aux OVS_UNUSED)
2766 struct ds ds = DS_EMPTY_INITIALIZER;
2767 const struct port *port;
2770 port = bond_find(args);
2772 unixctl_command_reply(conn, 501, "no such bond");
2776 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2777 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2778 ds_put_format(&ds, "next rebalance: %lld ms\n",
2779 port->bridge->bond_next_rebalance - time_msec());
2780 for (j = 0; j < port->n_ifaces; j++) {
2781 const struct iface *iface = port->ifaces[j];
2782 struct bond_entry *be;
2785 ds_put_format(&ds, "slave %s: %s\n",
2786 iface->name, iface->enabled ? "enabled" : "disabled");
2787 if (j == port->active_iface) {
2788 ds_put_cstr(&ds, "\tactive slave\n");
2790 if (iface->delay_expires != LLONG_MAX) {
2791 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2792 iface->enabled ? "downdelay" : "updelay",
2793 iface->delay_expires - time_msec());
2797 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2798 int hash = be - port->bond_hash;
2799 struct mac_entry *me;
2801 if (be->iface_idx != j) {
2805 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2806 hash, be->tx_bytes / 1024);
2809 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2810 &port->bridge->ml->lrus) {
2813 if (bond_hash(me->mac) == hash
2814 && me->port != port->port_idx
2815 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2816 && dp_ifidx == iface->dp_ifidx)
2818 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2819 ETH_ADDR_ARGS(me->mac));
2824 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2829 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2830 void *aux OVS_UNUSED)
2832 char *args = (char *) args_;
2833 char *save_ptr = NULL;
2834 char *bond_s, *hash_s, *slave_s;
2835 uint8_t mac[ETH_ADDR_LEN];
2837 struct iface *iface;
2838 struct bond_entry *entry;
2841 bond_s = strtok_r(args, " ", &save_ptr);
2842 hash_s = strtok_r(NULL, " ", &save_ptr);
2843 slave_s = strtok_r(NULL, " ", &save_ptr);
2845 unixctl_command_reply(conn, 501,
2846 "usage: bond/migrate BOND HASH SLAVE");
2850 port = bond_find(bond_s);
2852 unixctl_command_reply(conn, 501, "no such bond");
2856 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2857 == ETH_ADDR_SCAN_COUNT) {
2858 hash = bond_hash(mac);
2859 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2860 hash = atoi(hash_s) & BOND_MASK;
2862 unixctl_command_reply(conn, 501, "bad hash");
2866 iface = port_lookup_iface(port, slave_s);
2868 unixctl_command_reply(conn, 501, "no such slave");
2872 if (!iface->enabled) {
2873 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2877 entry = &port->bond_hash[hash];
2878 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2879 entry->iface_idx = iface->port_ifidx;
2880 entry->iface_tag = tag_create_random();
2881 port->bond_compat_is_stale = true;
2882 unixctl_command_reply(conn, 200, "migrated");
2886 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2887 void *aux OVS_UNUSED)
2889 char *args = (char *) args_;
2890 char *save_ptr = NULL;
2891 char *bond_s, *slave_s;
2893 struct iface *iface;
2895 bond_s = strtok_r(args, " ", &save_ptr);
2896 slave_s = strtok_r(NULL, " ", &save_ptr);
2898 unixctl_command_reply(conn, 501,
2899 "usage: bond/set-active-slave BOND SLAVE");
2903 port = bond_find(bond_s);
2905 unixctl_command_reply(conn, 501, "no such bond");
2909 iface = port_lookup_iface(port, slave_s);
2911 unixctl_command_reply(conn, 501, "no such slave");
2915 if (!iface->enabled) {
2916 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2920 if (port->active_iface != iface->port_ifidx) {
2921 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2922 port->active_iface = iface->port_ifidx;
2923 port->active_iface_tag = tag_create_random();
2924 VLOG_INFO("port %s: active interface is now %s",
2925 port->name, iface->name);
2926 bond_send_learning_packets(port);
2927 unixctl_command_reply(conn, 200, "done");
2929 unixctl_command_reply(conn, 200, "no change");
2934 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2936 char *args = (char *) args_;
2937 char *save_ptr = NULL;
2938 char *bond_s, *slave_s;
2940 struct iface *iface;
2942 bond_s = strtok_r(args, " ", &save_ptr);
2943 slave_s = strtok_r(NULL, " ", &save_ptr);
2945 unixctl_command_reply(conn, 501,
2946 "usage: bond/enable/disable-slave BOND SLAVE");
2950 port = bond_find(bond_s);
2952 unixctl_command_reply(conn, 501, "no such bond");
2956 iface = port_lookup_iface(port, slave_s);
2958 unixctl_command_reply(conn, 501, "no such slave");
2962 bond_enable_slave(iface, enable);
2963 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2967 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
2968 void *aux OVS_UNUSED)
2970 enable_slave(conn, args, true);
2974 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
2975 void *aux OVS_UNUSED)
2977 enable_slave(conn, args, false);
2981 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
2982 void *aux OVS_UNUSED)
2984 uint8_t mac[ETH_ADDR_LEN];
2988 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2989 == ETH_ADDR_SCAN_COUNT) {
2990 hash = bond_hash(mac);
2992 hash_cstr = xasprintf("%u", hash);
2993 unixctl_command_reply(conn, 200, hash_cstr);
2996 unixctl_command_reply(conn, 501, "invalid mac");
3003 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3004 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3005 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3006 unixctl_command_register("bond/set-active-slave",
3007 bond_unixctl_set_active_slave, NULL);
3008 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3010 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3012 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3015 /* Port functions. */
3017 static struct port *
3018 port_create(struct bridge *br, const char *name)
3022 port = xzalloc(sizeof *port);
3024 port->port_idx = br->n_ports;
3026 port->trunks = NULL;
3027 port->name = xstrdup(name);
3028 port->active_iface = -1;
3030 if (br->n_ports >= br->allocated_ports) {
3031 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3034 br->ports[br->n_ports++] = port;
3036 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3043 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3045 struct shash old_ifaces, new_ifaces;
3046 struct shash_node *node;
3047 unsigned long *trunks;
3053 /* Collect old and new interfaces. */
3054 shash_init(&old_ifaces);
3055 shash_init(&new_ifaces);
3056 for (i = 0; i < port->n_ifaces; i++) {
3057 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3059 for (i = 0; i < cfg->n_interfaces; i++) {
3060 const char *name = cfg->interfaces[i]->name;
3061 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3062 VLOG_WARN("port %s: %s specified twice as port interface",
3066 port->updelay = cfg->bond_updelay;
3067 if (port->updelay < 0) {
3070 port->updelay = cfg->bond_downdelay;
3071 if (port->downdelay < 0) {
3072 port->downdelay = 0;
3075 /* Get rid of deleted interfaces and add new interfaces. */
3076 SHASH_FOR_EACH (node, &old_ifaces) {
3077 if (!shash_find(&new_ifaces, node->name)) {
3078 iface_destroy(node->data);
3081 SHASH_FOR_EACH (node, &new_ifaces) {
3082 const struct ovsrec_interface *if_cfg = node->data;
3083 struct iface *iface;
3085 iface = shash_find_data(&old_ifaces, if_cfg->name);
3087 iface_create(port, if_cfg);
3089 iface->cfg = if_cfg;
3096 if (port->n_ifaces < 2) {
3098 if (vlan >= 0 && vlan <= 4095) {
3099 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3104 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3105 * they even work as-is. But they have not been tested. */
3106 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3110 if (port->vlan != vlan) {
3112 bridge_flush(port->bridge);
3115 /* Get trunked VLANs. */
3121 trunks = bitmap_allocate(4096);
3123 for (i = 0; i < cfg->n_trunks; i++) {
3124 int trunk = cfg->trunks[i];
3126 bitmap_set1(trunks, trunk);
3132 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3133 port->name, cfg->n_trunks);
3135 if (n_errors == cfg->n_trunks) {
3137 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3140 bitmap_set_multiple(trunks, 0, 4096, 1);
3143 if (cfg->n_trunks) {
3144 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3149 ? port->trunks != NULL
3150 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3151 bridge_flush(port->bridge);
3153 bitmap_free(port->trunks);
3154 port->trunks = trunks;
3156 shash_destroy(&old_ifaces);
3157 shash_destroy(&new_ifaces);
3161 port_destroy(struct port *port)
3164 struct bridge *br = port->bridge;
3168 proc_net_compat_update_vlan(port->name, NULL, 0);
3169 proc_net_compat_update_bond(port->name, NULL);
3171 for (i = 0; i < MAX_MIRRORS; i++) {
3172 struct mirror *m = br->mirrors[i];
3173 if (m && m->out_port == port) {
3178 while (port->n_ifaces > 0) {
3179 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3182 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3183 del->port_idx = port->port_idx;
3186 bitmap_free(port->trunks);
3193 static struct port *
3194 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3196 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3197 return iface ? iface->port : NULL;
3200 static struct port *
3201 port_lookup(const struct bridge *br, const char *name)
3205 for (i = 0; i < br->n_ports; i++) {
3206 struct port *port = br->ports[i];
3207 if (!strcmp(port->name, name)) {
3214 static struct iface *
3215 port_lookup_iface(const struct port *port, const char *name)
3219 for (j = 0; j < port->n_ifaces; j++) {
3220 struct iface *iface = port->ifaces[j];
3221 if (!strcmp(iface->name, name)) {
3229 port_update_bonding(struct port *port)
3231 if (port->n_ifaces < 2) {
3232 /* Not a bonded port. */
3233 if (port->bond_hash) {
3234 free(port->bond_hash);
3235 port->bond_hash = NULL;
3236 port->bond_compat_is_stale = true;
3237 port->bond_fake_iface = false;
3240 if (!port->bond_hash) {
3243 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3244 for (i = 0; i <= BOND_MASK; i++) {
3245 struct bond_entry *e = &port->bond_hash[i];
3249 port->no_ifaces_tag = tag_create_random();
3250 bond_choose_active_iface(port);
3252 port->bond_compat_is_stale = true;
3253 port->bond_fake_iface = port->cfg->bond_fake_iface;
3258 port_update_bond_compat(struct port *port)
3260 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3261 struct compat_bond bond;
3264 if (port->n_ifaces < 2) {
3265 proc_net_compat_update_bond(port->name, NULL);
3270 bond.updelay = port->updelay;
3271 bond.downdelay = port->downdelay;
3274 bond.hashes = compat_hashes;
3275 if (port->bond_hash) {
3276 const struct bond_entry *e;
3277 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3278 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3279 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3280 cbh->hash = e - port->bond_hash;
3281 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3286 bond.n_slaves = port->n_ifaces;
3287 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3288 for (i = 0; i < port->n_ifaces; i++) {
3289 struct iface *iface = port->ifaces[i];
3290 struct compat_bond_slave *slave = &bond.slaves[i];
3291 slave->name = iface->name;
3293 /* We need to make the same determination as the Linux bonding
3294 * code to determine whether a slave should be consider "up".
3295 * The Linux function bond_miimon_inspect() supports four
3296 * BOND_LINK_* states:
3298 * - BOND_LINK_UP: carrier detected, updelay has passed.
3299 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3300 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3301 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3303 * The function bond_info_show_slave() only considers BOND_LINK_UP
3304 * to be "up" and anything else to be "down".
3306 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3310 netdev_get_etheraddr(iface->netdev, slave->mac);
3313 if (port->bond_fake_iface) {
3314 struct netdev *bond_netdev;
3316 if (!netdev_open_default(port->name, &bond_netdev)) {
3318 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3320 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3322 netdev_close(bond_netdev);
3326 proc_net_compat_update_bond(port->name, &bond);
3331 port_update_vlan_compat(struct port *port)
3333 struct bridge *br = port->bridge;
3334 char *vlandev_name = NULL;
3336 if (port->vlan > 0) {
3337 /* Figure out the name that the VLAN device should actually have, if it
3338 * existed. This takes some work because the VLAN device would not
3339 * have port->name in its name; rather, it would have the trunk port's
3340 * name, and 'port' would be attached to a bridge that also had the
3341 * VLAN device one of its ports. So we need to find a trunk port that
3342 * includes port->vlan.
3344 * There might be more than one candidate. This doesn't happen on
3345 * XenServer, so if it happens we just pick the first choice in
3346 * alphabetical order instead of creating multiple VLAN devices. */
3348 for (i = 0; i < br->n_ports; i++) {
3349 struct port *p = br->ports[i];
3350 if (port_trunks_vlan(p, port->vlan)
3352 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3354 uint8_t ea[ETH_ADDR_LEN];
3355 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3356 if (!eth_addr_is_multicast(ea) &&
3357 !eth_addr_is_reserved(ea) &&
3358 !eth_addr_is_zero(ea)) {
3359 vlandev_name = p->name;
3364 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3367 /* Interface functions. */
3369 static struct iface *
3370 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3372 struct iface *iface;
3373 char *name = if_cfg->name;
3376 iface = xzalloc(sizeof *iface);
3378 iface->port_ifidx = port->n_ifaces;
3379 iface->name = xstrdup(name);
3380 iface->dp_ifidx = -1;
3381 iface->tag = tag_create_random();
3382 iface->delay_expires = LLONG_MAX;
3383 iface->netdev = NULL;
3384 iface->cfg = if_cfg;
3386 if (port->n_ifaces >= port->allocated_ifaces) {
3387 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3388 sizeof *port->ifaces);
3390 port->ifaces[port->n_ifaces++] = iface;
3391 if (port->n_ifaces > 1) {
3392 port->bridge->has_bonded_ports = true;
3395 /* Attempt to create the network interface in case it
3396 * doesn't exist yet. */
3397 if (!iface_is_internal(port->bridge, iface->name)) {
3398 error = set_up_iface(if_cfg, iface, true);
3400 VLOG_WARN("could not create iface %s: %s", iface->name,
3405 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3407 bridge_flush(port->bridge);
3413 iface_destroy(struct iface *iface)
3416 struct port *port = iface->port;
3417 struct bridge *br = port->bridge;
3418 bool del_active = port->active_iface == iface->port_ifidx;
3421 if (iface->dp_ifidx >= 0) {
3422 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3425 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3426 del->port_ifidx = iface->port_ifidx;
3428 netdev_close(iface->netdev);
3431 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3432 bond_choose_active_iface(port);
3433 bond_send_learning_packets(port);
3439 bridge_flush(port->bridge);
3443 static struct iface *
3444 iface_lookup(const struct bridge *br, const char *name)
3448 for (i = 0; i < br->n_ports; i++) {
3449 struct port *port = br->ports[i];
3450 for (j = 0; j < port->n_ifaces; j++) {
3451 struct iface *iface = port->ifaces[j];
3452 if (!strcmp(iface->name, name)) {
3460 static struct iface *
3461 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3463 return port_array_get(&br->ifaces, dp_ifidx);
3466 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3467 * 'br', that is, an interface that is entirely simulated within the datapath.
3468 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3469 * interfaces are created by setting "iface.<iface>.internal = true".
3471 * In addition, we have a kluge-y feature that creates an internal port with
3472 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3473 * This feature needs to go away in the long term. Until then, this is one
3474 * reason why this function takes a name instead of a struct iface: the fake
3475 * interfaces created this way do not have a struct iface. */
3477 iface_is_internal(const struct bridge *br, const char *if_name)
3479 /* XXX wastes time */
3480 struct iface *iface;
3483 if (!strcmp(if_name, br->name)) {
3487 iface = iface_lookup(br, if_name);
3488 if (iface && !strcmp(iface->cfg->type, "internal")) {
3492 port = port_lookup(br, if_name);
3493 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3499 /* Set Ethernet address of 'iface', if one is specified in the configuration
3502 iface_set_mac(struct iface *iface)
3504 uint8_t ea[ETH_ADDR_LEN];
3506 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3507 if (eth_addr_is_multicast(ea)) {
3508 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3510 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3511 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3512 iface->name, iface->name);
3514 int error = netdev_set_etheraddr(iface->netdev, ea);
3516 VLOG_ERR("interface %s: setting MAC failed (%s)",
3517 iface->name, strerror(error));
3523 /* Port mirroring. */
3526 mirror_reconfigure(struct bridge *br)
3528 struct shash old_mirrors, new_mirrors;
3529 struct shash_node *node;
3530 unsigned long *rspan_vlans;
3533 /* Collect old mirrors. */
3534 shash_init(&old_mirrors);
3535 for (i = 0; i < MAX_MIRRORS; i++) {
3536 if (br->mirrors[i]) {
3537 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3541 /* Collect new mirrors. */
3542 shash_init(&new_mirrors);
3543 for (i = 0; i < br->cfg->n_mirrors; i++) {
3544 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3545 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3546 VLOG_WARN("bridge %s: %s specified twice as mirror",
3547 br->name, cfg->name);
3551 /* Get rid of deleted mirrors and add new mirrors. */
3552 SHASH_FOR_EACH (node, &old_mirrors) {
3553 if (!shash_find(&new_mirrors, node->name)) {
3554 mirror_destroy(node->data);
3557 SHASH_FOR_EACH (node, &new_mirrors) {
3558 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3560 mirror = mirror_create(br, node->name);
3565 mirror_reconfigure_one(mirror, node->data);
3567 shash_destroy(&old_mirrors);
3568 shash_destroy(&new_mirrors);
3570 /* Update port reserved status. */
3571 for (i = 0; i < br->n_ports; i++) {
3572 br->ports[i]->is_mirror_output_port = false;
3574 for (i = 0; i < MAX_MIRRORS; i++) {
3575 struct mirror *m = br->mirrors[i];
3576 if (m && m->out_port) {
3577 m->out_port->is_mirror_output_port = true;
3581 /* Update flooded vlans (for RSPAN). */
3583 if (br->cfg->n_flood_vlans) {
3584 rspan_vlans = bitmap_allocate(4096);
3586 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3587 int64_t vlan = br->cfg->flood_vlans[i];
3588 if (vlan >= 0 && vlan < 4096) {
3589 bitmap_set1(rspan_vlans, vlan);
3590 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3593 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3598 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3603 static struct mirror *
3604 mirror_create(struct bridge *br, const char *name)
3609 for (i = 0; ; i++) {
3610 if (i >= MAX_MIRRORS) {
3611 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3612 "cannot create %s", br->name, MAX_MIRRORS, name);
3615 if (!br->mirrors[i]) {
3620 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3623 br->mirrors[i] = m = xzalloc(sizeof *m);
3626 m->name = xstrdup(name);
3627 shash_init(&m->src_ports);
3628 shash_init(&m->dst_ports);
3638 mirror_destroy(struct mirror *m)
3641 struct bridge *br = m->bridge;
3644 for (i = 0; i < br->n_ports; i++) {
3645 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3646 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3649 shash_destroy(&m->src_ports);
3650 shash_destroy(&m->dst_ports);
3653 m->bridge->mirrors[m->idx] = NULL;
3661 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3662 struct shash *names)
3666 for (i = 0; i < n_ports; i++) {
3667 const char *name = ports[i]->name;
3668 if (port_lookup(m->bridge, name)) {
3669 shash_add_once(names, name, NULL);
3671 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3672 "port %s", m->bridge->name, m->name, name);
3678 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3684 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3686 for (i = 0; i < cfg->n_select_vlan; i++) {
3687 int64_t vlan = cfg->select_vlan[i];
3688 if (vlan < 0 || vlan > 4095) {
3689 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3690 m->bridge->name, m->name, vlan);
3692 (*vlans)[n_vlans++] = vlan;
3699 vlan_is_mirrored(const struct mirror *m, int vlan)
3703 for (i = 0; i < m->n_vlans; i++) {
3704 if (m->vlans[i] == vlan) {
3712 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3716 for (i = 0; i < m->n_vlans; i++) {
3717 if (port_trunks_vlan(p, m->vlans[i])) {
3725 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3727 struct shash src_ports, dst_ports;
3728 mirror_mask_t mirror_bit;
3729 struct port *out_port;
3734 bool mirror_all_ports;
3735 bool any_ports_specified;
3736 bool any_vlans_specified;
3738 /* Get output port. */
3739 if (cfg->output_port) {
3740 out_port = port_lookup(m->bridge, cfg->output_port->name);
3742 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3743 m->bridge->name, m->name);
3749 if (cfg->output_vlan) {
3750 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3751 "output vlan; ignoring output vlan",
3752 m->bridge->name, m->name);
3754 } else if (cfg->output_vlan) {
3756 out_vlan = *cfg->output_vlan;
3758 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3759 m->bridge->name, m->name);
3764 /* Get all the ports, and drop duplicates and ports that don't exist. */
3765 shash_init(&src_ports);
3766 shash_init(&dst_ports);
3767 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3769 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3771 any_ports_specified = cfg->n_select_dst_port || cfg->n_select_dst_port;
3772 if (any_ports_specified
3773 && shash_is_empty(&src_ports) && shash_is_empty(&dst_ports)) {
3774 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3775 "selection ports exists", m->bridge->name, m->name);
3780 /* Get all the vlans, and drop duplicate and invalid vlans. */
3781 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3782 any_vlans_specified = cfg->n_select_vlan > 0;
3783 if (any_vlans_specified && !n_vlans) {
3784 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3785 "VLANs exists", m->bridge->name, m->name);
3790 /* Update mirror data. */
3791 if (!shash_equal_keys(&m->src_ports, &src_ports)
3792 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3793 || m->n_vlans != n_vlans
3794 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3795 || m->out_port != out_port
3796 || m->out_vlan != out_vlan) {
3797 bridge_flush(m->bridge);
3799 shash_swap(&m->src_ports, &src_ports);
3800 shash_swap(&m->dst_ports, &dst_ports);
3803 m->n_vlans = n_vlans;
3804 m->out_port = out_port;
3805 m->out_vlan = out_vlan;
3807 /* If no selection criteria have been given, mirror for all ports. */
3808 mirror_all_ports = !any_ports_specified && !any_vlans_specified;
3811 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3812 for (i = 0; i < m->bridge->n_ports; i++) {
3813 struct port *port = m->bridge->ports[i];
3815 if (mirror_all_ports
3816 || shash_find(&m->src_ports, port->name)
3819 ? port_trunks_any_mirrored_vlan(m, port)
3820 : vlan_is_mirrored(m, port->vlan)))) {
3821 port->src_mirrors |= mirror_bit;
3823 port->src_mirrors &= ~mirror_bit;
3826 if (mirror_all_ports || shash_find(&m->dst_ports, port->name)) {
3827 port->dst_mirrors |= mirror_bit;
3829 port->dst_mirrors &= ~mirror_bit;
3835 shash_destroy(&src_ports);
3836 shash_destroy(&dst_ports);