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 UNUSED, struct iface *iface,
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, void *aux UNUSED)
473 if (iface->dp_ifidx >= 0) {
474 VLOG_DBG("%s has interface %s on port %d",
476 iface->name, iface->dp_ifidx);
479 VLOG_ERR("%s interface not in %s, dropping",
480 iface->name, dpif_name(br->dpif));
486 set_iface_properties(struct bridge *br UNUSED, struct iface *iface,
489 /* Set policing attributes. */
490 netdev_set_policing(iface->netdev,
491 iface->cfg->ingress_policing_rate,
492 iface->cfg->ingress_policing_burst);
494 /* Set MAC address of internal interfaces other than the local
496 if (iface->dp_ifidx != ODPP_LOCAL
497 && iface_is_internal(br, iface->name)) {
498 iface_set_mac(iface);
504 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
505 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
506 * deletes from 'br' any ports that no longer have any interfaces. */
508 iterate_and_prune_ifaces(struct bridge *br,
509 bool (*cb)(struct bridge *, struct iface *,
515 for (i = 0; i < br->n_ports; ) {
516 struct port *port = br->ports[i];
517 for (j = 0; j < port->n_ifaces; ) {
518 struct iface *iface = port->ifaces[j];
519 if (cb(br, iface, aux)) {
522 iface_destroy(iface);
526 if (port->n_ifaces) {
529 VLOG_ERR("%s port has no interfaces, dropping", port->name);
536 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
538 struct ovsdb_idl_txn *txn;
539 struct shash old_br, new_br;
540 struct shash_node *node;
541 struct bridge *br, *next;
543 int sflow_bridge_number;
545 COVERAGE_INC(bridge_reconfigure);
547 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
549 /* Collect old and new bridges. */
552 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
553 shash_add(&old_br, br->name, br);
555 for (i = 0; i < ovs_cfg->n_bridges; i++) {
556 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
557 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
558 VLOG_WARN("more than one bridge named %s", br_cfg->name);
562 /* Get rid of deleted bridges and add new bridges. */
563 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
564 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
571 SHASH_FOR_EACH (node, &new_br) {
572 const char *br_name = node->name;
573 const struct ovsrec_bridge *br_cfg = node->data;
574 br = shash_find_data(&old_br, br_name);
576 /* If the bridge datapath type has changed, we need to tear it
577 * down and recreate. */
578 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
580 bridge_create(br_cfg);
583 bridge_create(br_cfg);
586 shash_destroy(&old_br);
587 shash_destroy(&new_br);
591 bridge_configure_ssl(ovs_cfg->ssl);
594 /* Reconfigure all bridges. */
595 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
596 bridge_reconfigure_one(ovs_cfg, br);
599 /* Add and delete ports on all datapaths.
601 * The kernel will reject any attempt to add a given port to a datapath if
602 * that port already belongs to a different datapath, so we must do all
603 * port deletions before any port additions. */
604 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
605 struct odp_port *dpif_ports;
607 struct shash want_ifaces;
609 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
610 bridge_get_all_ifaces(br, &want_ifaces);
611 for (i = 0; i < n_dpif_ports; i++) {
612 const struct odp_port *p = &dpif_ports[i];
613 if (!shash_find(&want_ifaces, p->devname)
614 && strcmp(p->devname, br->name)) {
615 int retval = dpif_port_del(br->dpif, p->port);
617 VLOG_ERR("failed to remove %s interface from %s: %s",
618 p->devname, dpif_name(br->dpif),
623 shash_destroy(&want_ifaces);
626 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
627 struct odp_port *dpif_ports;
629 struct shash cur_ifaces, want_ifaces;
630 struct shash_node *node;
632 /* Get the set of interfaces currently in this datapath. */
633 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
634 shash_init(&cur_ifaces);
635 for (i = 0; i < n_dpif_ports; i++) {
636 const char *name = dpif_ports[i].devname;
637 if (!shash_find(&cur_ifaces, name)) {
638 shash_add(&cur_ifaces, name, NULL);
643 /* Get the set of interfaces we want on this datapath. */
644 bridge_get_all_ifaces(br, &want_ifaces);
646 SHASH_FOR_EACH (node, &want_ifaces) {
647 const char *if_name = node->name;
648 struct iface *iface = node->data;
650 if (shash_find(&cur_ifaces, if_name)) {
651 /* Already exists, just reconfigure it. */
653 reconfigure_iface(iface->cfg, iface);
656 /* Need to add to datapath. */
660 /* Add to datapath. */
661 internal = iface_is_internal(br, if_name);
662 error = dpif_port_add(br->dpif, if_name,
663 internal ? ODP_PORT_INTERNAL : 0, NULL);
664 if (error == EFBIG) {
665 VLOG_ERR("ran out of valid port numbers on %s",
666 dpif_name(br->dpif));
669 VLOG_ERR("failed to add %s interface to %s: %s",
670 if_name, dpif_name(br->dpif), strerror(error));
674 shash_destroy(&cur_ifaces);
675 shash_destroy(&want_ifaces);
677 sflow_bridge_number = 0;
678 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
681 struct iface *local_iface;
682 struct iface *hw_addr_iface;
685 bridge_fetch_dp_ifaces(br);
687 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
688 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
690 /* Pick local port hardware address, datapath ID. */
691 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
692 local_iface = bridge_get_local_iface(br);
694 int error = netdev_set_etheraddr(local_iface->netdev, ea);
696 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
697 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
698 "Ethernet address: %s",
699 br->name, strerror(error));
703 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
704 ofproto_set_datapath_id(br->ofproto, dpid);
706 dpid_string = xasprintf("%012"PRIx64, dpid);
707 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
710 /* Set NetFlow configuration on this bridge. */
711 if (br->cfg->netflow) {
712 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
713 struct netflow_options opts;
715 memset(&opts, 0, sizeof opts);
717 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
718 if (nf_cfg->engine_type) {
719 opts.engine_type = *nf_cfg->engine_type;
721 if (nf_cfg->engine_id) {
722 opts.engine_id = *nf_cfg->engine_id;
725 opts.active_timeout = nf_cfg->active_timeout;
726 if (!opts.active_timeout) {
727 opts.active_timeout = -1;
728 } else if (opts.active_timeout < 0) {
729 VLOG_WARN("bridge %s: active timeout interval set to negative "
730 "value, using default instead (%d seconds)", br->name,
731 NF_ACTIVE_TIMEOUT_DEFAULT);
732 opts.active_timeout = -1;
735 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
736 if (opts.add_id_to_iface) {
737 if (opts.engine_id > 0x7f) {
738 VLOG_WARN("bridge %s: netflow port mangling may conflict "
739 "with another vswitch, choose an engine id less "
740 "than 128", br->name);
742 if (br->n_ports > 508) {
743 VLOG_WARN("bridge %s: netflow port mangling will conflict "
744 "with another port when more than 508 ports are "
749 opts.collectors.n = nf_cfg->n_targets;
750 opts.collectors.names = nf_cfg->targets;
751 if (ofproto_set_netflow(br->ofproto, &opts)) {
752 VLOG_ERR("bridge %s: problem setting netflow collectors",
756 ofproto_set_netflow(br->ofproto, NULL);
759 /* Set sFlow configuration on this bridge. */
760 if (br->cfg->sflow) {
761 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
762 const struct ovsrec_controller *ctrl;
763 struct ofproto_sflow_options oso;
765 memset(&oso, 0, sizeof oso);
767 oso.targets.n = sflow_cfg->n_targets;
768 oso.targets.names = sflow_cfg->targets;
770 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
771 if (sflow_cfg->sampling) {
772 oso.sampling_rate = *sflow_cfg->sampling;
775 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
776 if (sflow_cfg->polling) {
777 oso.polling_interval = *sflow_cfg->polling;
780 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
781 if (sflow_cfg->header) {
782 oso.header_len = *sflow_cfg->header;
785 oso.sub_id = sflow_bridge_number++;
786 oso.agent_device = sflow_cfg->agent;
788 ctrl = bridge_get_controller(ovs_cfg, br);
789 oso.control_ip = ctrl ? ctrl->local_ip : NULL;
790 ofproto_set_sflow(br->ofproto, &oso);
792 svec_destroy(&oso.targets);
794 ofproto_set_sflow(br->ofproto, NULL);
797 /* Update the controller and related settings. It would be more
798 * straightforward to call this from bridge_reconfigure_one(), but we
799 * can't do it there for two reasons. First, and most importantly, at
800 * that point we don't know the dp_ifidx of any interfaces that have
801 * been added to the bridge (because we haven't actually added them to
802 * the datapath). Second, at that point we haven't set the datapath ID
803 * yet; when a controller is configured, resetting the datapath ID will
804 * immediately disconnect from the controller, so it's better to set
805 * the datapath ID before the controller. */
806 bridge_reconfigure_controller(ovs_cfg, br);
808 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
809 for (i = 0; i < br->n_ports; i++) {
810 struct port *port = br->ports[i];
812 port_update_vlan_compat(port);
813 port_update_bonding(port);
816 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
817 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
820 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
822 ovsdb_idl_txn_commit(txn);
823 ovsdb_idl_txn_destroy(txn); /* XXX */
827 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
831 for (i = 0; i < br_cfg->n_other_config; i++) {
832 if (!strcmp(br_cfg->key_other_config[i], key)) {
833 return br_cfg->value_other_config[i];
840 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
841 struct iface **hw_addr_iface)
847 *hw_addr_iface = NULL;
849 /* Did the user request a particular MAC? */
850 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
851 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
852 if (eth_addr_is_multicast(ea)) {
853 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
854 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
855 } else if (eth_addr_is_zero(ea)) {
856 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
862 /* Otherwise choose the minimum non-local MAC address among all of the
864 memset(ea, 0xff, sizeof ea);
865 for (i = 0; i < br->n_ports; i++) {
866 struct port *port = br->ports[i];
867 uint8_t iface_ea[ETH_ADDR_LEN];
870 /* Mirror output ports don't participate. */
871 if (port->is_mirror_output_port) {
875 /* Choose the MAC address to represent the port. */
876 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
877 /* Find the interface with this Ethernet address (if any) so that
878 * we can provide the correct devname to the caller. */
880 for (j = 0; j < port->n_ifaces; j++) {
881 struct iface *candidate = port->ifaces[j];
882 uint8_t candidate_ea[ETH_ADDR_LEN];
883 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
884 && eth_addr_equals(iface_ea, candidate_ea)) {
889 /* Choose the interface whose MAC address will represent the port.
890 * The Linux kernel bonding code always chooses the MAC address of
891 * the first slave added to a bond, and the Fedora networking
892 * scripts always add slaves to a bond in alphabetical order, so
893 * for compatibility we choose the interface with the name that is
894 * first in alphabetical order. */
895 iface = port->ifaces[0];
896 for (j = 1; j < port->n_ifaces; j++) {
897 struct iface *candidate = port->ifaces[j];
898 if (strcmp(candidate->name, iface->name) < 0) {
903 /* The local port doesn't count (since we're trying to choose its
904 * MAC address anyway). */
905 if (iface->dp_ifidx == ODPP_LOCAL) {
910 error = netdev_get_etheraddr(iface->netdev, iface_ea);
912 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
913 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
914 iface->name, strerror(error));
919 /* Compare against our current choice. */
920 if (!eth_addr_is_multicast(iface_ea) &&
921 !eth_addr_is_local(iface_ea) &&
922 !eth_addr_is_reserved(iface_ea) &&
923 !eth_addr_is_zero(iface_ea) &&
924 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
926 memcpy(ea, iface_ea, ETH_ADDR_LEN);
927 *hw_addr_iface = iface;
930 if (eth_addr_is_multicast(ea)) {
931 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
932 *hw_addr_iface = NULL;
933 VLOG_WARN("bridge %s: using default bridge Ethernet "
934 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
936 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
937 br->name, ETH_ADDR_ARGS(ea));
941 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
942 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
943 * an interface on 'br', then that interface must be passed in as
944 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
945 * 'hw_addr_iface' must be passed in as a null pointer. */
947 bridge_pick_datapath_id(struct bridge *br,
948 const uint8_t bridge_ea[ETH_ADDR_LEN],
949 struct iface *hw_addr_iface)
952 * The procedure for choosing a bridge MAC address will, in the most
953 * ordinary case, also choose a unique MAC that we can use as a datapath
954 * ID. In some special cases, though, multiple bridges will end up with
955 * the same MAC address. This is OK for the bridges, but it will confuse
956 * the OpenFlow controller, because each datapath needs a unique datapath
959 * Datapath IDs must be unique. It is also very desirable that they be
960 * stable from one run to the next, so that policy set on a datapath
963 const char *datapath_id;
966 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
967 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
973 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
975 * A bridge whose MAC address is taken from a VLAN network device
976 * (that is, a network device created with vconfig(8) or similar
977 * tool) will have the same MAC address as a bridge on the VLAN
978 * device's physical network device.
980 * Handle this case by hashing the physical network device MAC
981 * along with the VLAN identifier.
983 uint8_t buf[ETH_ADDR_LEN + 2];
984 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
985 buf[ETH_ADDR_LEN] = vlan >> 8;
986 buf[ETH_ADDR_LEN + 1] = vlan;
987 return dpid_from_hash(buf, sizeof buf);
990 * Assume that this bridge's MAC address is unique, since it
991 * doesn't fit any of the cases we handle specially.
996 * A purely internal bridge, that is, one that has no non-virtual
997 * network devices on it at all, is more difficult because it has no
998 * natural unique identifier at all.
1000 * When the host is a XenServer, we handle this case by hashing the
1001 * host's UUID with the name of the bridge. Names of bridges are
1002 * persistent across XenServer reboots, although they can be reused if
1003 * an internal network is destroyed and then a new one is later
1004 * created, so this is fairly effective.
1006 * When the host is not a XenServer, we punt by using a random MAC
1007 * address on each run.
1009 const char *host_uuid = xenserver_get_host_uuid();
1011 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1012 dpid = dpid_from_hash(combined, strlen(combined));
1018 return eth_addr_to_uint64(bridge_ea);
1022 dpid_from_hash(const void *data, size_t n)
1024 uint8_t hash[SHA1_DIGEST_SIZE];
1026 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1027 sha1_bytes(data, n, hash);
1028 eth_addr_mark_random(hash);
1029 return eth_addr_to_uint64(hash);
1035 struct bridge *br, *next;
1039 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1040 int error = bridge_run_one(br);
1042 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1043 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1044 "forcing reconfiguration", br->name);
1058 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1059 ofproto_wait(br->ofproto);
1060 if (br->controller) {
1064 mac_learning_wait(br->ml);
1069 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1070 * configuration changes. */
1072 bridge_flush(struct bridge *br)
1074 COVERAGE_INC(bridge_flush);
1076 mac_learning_flush(br->ml);
1079 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1080 * such interface. */
1081 static struct iface *
1082 bridge_get_local_iface(struct bridge *br)
1086 for (i = 0; i < br->n_ports; i++) {
1087 struct port *port = br->ports[i];
1088 for (j = 0; j < port->n_ifaces; j++) {
1089 struct iface *iface = port->ifaces[j];
1090 if (iface->dp_ifidx == ODPP_LOCAL) {
1099 /* Bridge unixctl user interface functions. */
1101 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1102 const char *args, void *aux UNUSED)
1104 struct ds ds = DS_EMPTY_INITIALIZER;
1105 const struct bridge *br;
1106 const struct mac_entry *e;
1108 br = bridge_lookup(args);
1110 unixctl_command_reply(conn, 501, "no such bridge");
1114 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1115 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1116 if (e->port < 0 || e->port >= br->n_ports) {
1119 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1120 br->ports[e->port]->ifaces[0]->dp_ifidx,
1121 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1123 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1127 /* Bridge reconfiguration functions. */
1128 static struct bridge *
1129 bridge_create(const struct ovsrec_bridge *br_cfg)
1134 assert(!bridge_lookup(br_cfg->name));
1135 br = xzalloc(sizeof *br);
1137 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1143 dpif_flow_flush(br->dpif);
1145 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1148 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1150 dpif_delete(br->dpif);
1151 dpif_close(br->dpif);
1156 br->name = xstrdup(br_cfg->name);
1158 br->ml = mac_learning_create();
1159 br->sent_config_request = false;
1160 eth_addr_nicira_random(br->default_ea);
1162 port_array_init(&br->ifaces);
1165 br->bond_next_rebalance = time_msec() + 10000;
1167 list_push_back(&all_bridges, &br->node);
1169 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1175 bridge_destroy(struct bridge *br)
1180 while (br->n_ports > 0) {
1181 port_destroy(br->ports[br->n_ports - 1]);
1183 list_remove(&br->node);
1184 error = dpif_delete(br->dpif);
1185 if (error && error != ENOENT) {
1186 VLOG_ERR("failed to delete %s: %s",
1187 dpif_name(br->dpif), strerror(error));
1189 dpif_close(br->dpif);
1190 ofproto_destroy(br->ofproto);
1191 free(br->controller);
1192 mac_learning_destroy(br->ml);
1193 port_array_destroy(&br->ifaces);
1200 static struct bridge *
1201 bridge_lookup(const char *name)
1205 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1206 if (!strcmp(br->name, name)) {
1214 bridge_exists(const char *name)
1216 return bridge_lookup(name) ? true : false;
1220 bridge_get_datapathid(const char *name)
1222 struct bridge *br = bridge_lookup(name);
1223 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1226 /* Handle requests for a listing of all flows known by the OpenFlow
1227 * stack, including those normally hidden. */
1229 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1230 const char *args, void *aux UNUSED)
1235 br = bridge_lookup(args);
1237 unixctl_command_reply(conn, 501, "Unknown bridge");
1242 ofproto_get_all_flows(br->ofproto, &results);
1244 unixctl_command_reply(conn, 200, ds_cstr(&results));
1245 ds_destroy(&results);
1249 bridge_run_one(struct bridge *br)
1253 error = ofproto_run1(br->ofproto);
1258 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1261 error = ofproto_run2(br->ofproto, br->flush);
1267 static const struct ovsrec_controller *
1268 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1269 const struct bridge *br)
1271 const struct ovsrec_controller *controller;
1273 controller = (br->cfg->controller ? br->cfg->controller
1274 : ovs_cfg->controller ? ovs_cfg->controller
1277 if (controller && !strcmp(controller->target, "none")) {
1285 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1287 struct svec *ifaces = ifaces_;
1288 if (!svec_contains(ifaces, iface->name)) {
1289 svec_add(ifaces, iface->name);
1293 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1295 br->name, iface->name, iface->port->name);
1301 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1304 struct shash old_ports, new_ports;
1306 struct svec listeners, old_listeners;
1307 struct svec snoops, old_snoops;
1308 struct shash_node *node;
1311 /* Collect old ports. */
1312 shash_init(&old_ports);
1313 for (i = 0; i < br->n_ports; i++) {
1314 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1317 /* Collect new ports. */
1318 shash_init(&new_ports);
1319 for (i = 0; i < br->cfg->n_ports; i++) {
1320 const char *name = br->cfg->ports[i]->name;
1321 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1322 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1327 /* If we have a controller, then we need a local port. Complain if the
1328 * user didn't specify one.
1330 * XXX perhaps we should synthesize a port ourselves in this case. */
1331 if (bridge_get_controller(ovs_cfg, br)) {
1332 char local_name[IF_NAMESIZE];
1335 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1336 local_name, sizeof local_name);
1337 if (!error && !shash_find(&new_ports, local_name)) {
1338 VLOG_WARN("bridge %s: controller specified but no local port "
1339 "(port named %s) defined",
1340 br->name, local_name);
1344 /* Get rid of deleted ports and add new ports. */
1345 SHASH_FOR_EACH (node, &old_ports) {
1346 if (!shash_find(&new_ports, node->name)) {
1347 port_destroy(node->data);
1350 SHASH_FOR_EACH (node, &new_ports) {
1351 struct port *port = shash_find_data(&old_ports, node->name);
1353 port = port_create(br, node->name);
1355 port_reconfigure(port, node->data);
1357 shash_destroy(&old_ports);
1358 shash_destroy(&new_ports);
1360 /* Check and delete duplicate interfaces. */
1362 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1363 svec_destroy(&ifaces);
1365 /* Delete all flows if we're switching from connected to standalone or vice
1366 * versa. (XXX Should we delete all flows if we are switching from one
1367 * controller to another?) */
1370 /* Configure OpenFlow management listeners. */
1371 svec_init(&listeners);
1372 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1374 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1375 ovs_rundir, br->name));
1376 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1377 svec_clear(&listeners);
1379 svec_sort_unique(&listeners);
1381 svec_init(&old_listeners);
1382 ofproto_get_listeners(br->ofproto, &old_listeners);
1383 svec_sort_unique(&old_listeners);
1385 if (!svec_equal(&listeners, &old_listeners)) {
1386 ofproto_set_listeners(br->ofproto, &listeners);
1388 svec_destroy(&listeners);
1389 svec_destroy(&old_listeners);
1391 /* Configure OpenFlow controller connection snooping. */
1393 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1395 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1396 ovs_rundir, br->name));
1397 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1398 svec_clear(&snoops);
1400 svec_sort_unique(&snoops);
1402 svec_init(&old_snoops);
1403 ofproto_get_snoops(br->ofproto, &old_snoops);
1404 svec_sort_unique(&old_snoops);
1406 if (!svec_equal(&snoops, &old_snoops)) {
1407 ofproto_set_snoops(br->ofproto, &snoops);
1409 svec_destroy(&snoops);
1410 svec_destroy(&old_snoops);
1412 /* Default listener. */
1413 svec_init(&listeners);
1414 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1415 ovs_rundir, br->name));
1416 svec_init(&old_listeners);
1417 ofproto_get_listeners(br->ofproto, &old_listeners);
1418 if (!svec_equal(&listeners, &old_listeners)) {
1419 ofproto_set_listeners(br->ofproto, &listeners);
1421 svec_destroy(&listeners);
1422 svec_destroy(&old_listeners);
1424 /* Default snoop. */
1426 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1427 ovs_rundir, br->name));
1428 svec_init(&old_snoops);
1429 ofproto_get_snoops(br->ofproto, &old_snoops);
1430 if (!svec_equal(&snoops, &old_snoops)) {
1431 ofproto_set_snoops(br->ofproto, &snoops);
1433 svec_destroy(&snoops);
1434 svec_destroy(&old_snoops);
1437 mirror_reconfigure(br);
1441 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1444 char *pfx = xasprintf("bridge.%s.controller", br->name);
1445 const struct ovsrec_controller *c;
1447 c = bridge_get_controller(ovs_cfg, br);
1448 if ((br->controller != NULL) != (c != NULL)) {
1449 ofproto_flush_flows(br->ofproto);
1451 free(br->controller);
1452 br->controller = c ? xstrdup(c->target) : NULL;
1455 int max_backoff, probe;
1456 int rate_limit, burst_limit;
1458 if (!strcmp(c->target, "discover")) {
1459 ofproto_set_discovery(br->ofproto, true,
1460 c->discover_accept_regex,
1461 c->discover_update_resolv_conf);
1463 struct iface *local_iface;
1467 in_band = (!c->connection_mode
1468 || !strcmp(c->connection_mode, "out-of-band"));
1469 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1470 ofproto_set_in_band(br->ofproto, in_band);
1472 local_iface = bridge_get_local_iface(br);
1473 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1474 struct netdev *netdev = local_iface->netdev;
1475 struct in_addr mask, gateway;
1477 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1480 if (!c->local_gateway
1481 || !inet_aton(c->local_gateway, &gateway)) {
1485 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1487 mask.s_addr = guess_netmask(ip.s_addr);
1489 if (!netdev_set_in4(netdev, ip, mask)) {
1490 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1492 br->name, IP_ARGS(&ip.s_addr),
1493 IP_ARGS(&mask.s_addr));
1496 if (gateway.s_addr) {
1497 if (!netdev_add_router(netdev, gateway)) {
1498 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1499 br->name, IP_ARGS(&gateway.s_addr));
1505 ofproto_set_failure(br->ofproto,
1507 || !strcmp(c->fail_mode, "standalone")
1508 || !strcmp(c->fail_mode, "open")));
1510 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1511 ofproto_set_probe_interval(br->ofproto, probe);
1513 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1514 ofproto_set_max_backoff(br->ofproto, max_backoff);
1516 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1517 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1518 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1520 union ofp_action action;
1523 /* Set up a flow that matches every packet and directs them to
1524 * OFPP_NORMAL (which goes to us). */
1525 memset(&action, 0, sizeof action);
1526 action.type = htons(OFPAT_OUTPUT);
1527 action.output.len = htons(sizeof action);
1528 action.output.port = htons(OFPP_NORMAL);
1529 memset(&flow, 0, sizeof flow);
1530 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1533 ofproto_set_in_band(br->ofproto, false);
1534 ofproto_set_max_backoff(br->ofproto, 1);
1535 ofproto_set_probe_interval(br->ofproto, 5);
1536 ofproto_set_failure(br->ofproto, false);
1540 ofproto_set_controller(br->ofproto, br->controller);
1544 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1549 for (i = 0; i < br->n_ports; i++) {
1550 struct port *port = br->ports[i];
1551 for (j = 0; j < port->n_ifaces; j++) {
1552 struct iface *iface = port->ifaces[j];
1553 shash_add_once(ifaces, iface->name, iface);
1555 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1556 shash_add_once(ifaces, port->name, NULL);
1561 /* For robustness, in case the administrator moves around datapath ports behind
1562 * our back, we re-check all the datapath port numbers here.
1564 * This function will set the 'dp_ifidx' members of interfaces that have
1565 * disappeared to -1, so only call this function from a context where those
1566 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1567 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1568 * datapath, which doesn't support UINT16_MAX+1 ports. */
1570 bridge_fetch_dp_ifaces(struct bridge *br)
1572 struct odp_port *dpif_ports;
1573 size_t n_dpif_ports;
1576 /* Reset all interface numbers. */
1577 for (i = 0; i < br->n_ports; i++) {
1578 struct port *port = br->ports[i];
1579 for (j = 0; j < port->n_ifaces; j++) {
1580 struct iface *iface = port->ifaces[j];
1581 iface->dp_ifidx = -1;
1584 port_array_clear(&br->ifaces);
1586 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1587 for (i = 0; i < n_dpif_ports; i++) {
1588 struct odp_port *p = &dpif_ports[i];
1589 struct iface *iface = iface_lookup(br, p->devname);
1591 if (iface->dp_ifidx >= 0) {
1592 VLOG_WARN("%s reported interface %s twice",
1593 dpif_name(br->dpif), p->devname);
1594 } else if (iface_from_dp_ifidx(br, p->port)) {
1595 VLOG_WARN("%s reported interface %"PRIu16" twice",
1596 dpif_name(br->dpif), p->port);
1598 port_array_set(&br->ifaces, p->port, iface);
1599 iface->dp_ifidx = p->port;
1603 int64_t ofport = (iface->dp_ifidx >= 0
1604 ? odp_port_to_ofp_port(iface->dp_ifidx)
1606 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1613 /* Bridge packet processing functions. */
1616 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1618 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1621 static struct bond_entry *
1622 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1624 return &port->bond_hash[bond_hash(mac)];
1628 bond_choose_iface(const struct port *port)
1630 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1631 size_t i, best_down_slave = -1;
1632 long long next_delay_expiration = LLONG_MAX;
1634 for (i = 0; i < port->n_ifaces; i++) {
1635 struct iface *iface = port->ifaces[i];
1637 if (iface->enabled) {
1639 } else if (iface->delay_expires < next_delay_expiration) {
1640 best_down_slave = i;
1641 next_delay_expiration = iface->delay_expires;
1645 if (best_down_slave != -1) {
1646 struct iface *iface = port->ifaces[best_down_slave];
1648 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1649 "since no other interface is up", iface->name,
1650 iface->delay_expires - time_msec());
1651 bond_enable_slave(iface, true);
1654 return best_down_slave;
1658 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1659 uint16_t *dp_ifidx, tag_type *tags)
1661 struct iface *iface;
1663 assert(port->n_ifaces);
1664 if (port->n_ifaces == 1) {
1665 iface = port->ifaces[0];
1667 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1668 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1669 || !port->ifaces[e->iface_idx]->enabled) {
1670 /* XXX select interface properly. The current interface selection
1671 * is only good for testing the rebalancing code. */
1672 e->iface_idx = bond_choose_iface(port);
1673 if (e->iface_idx < 0) {
1674 *tags |= port->no_ifaces_tag;
1677 e->iface_tag = tag_create_random();
1678 ((struct port *) port)->bond_compat_is_stale = true;
1680 *tags |= e->iface_tag;
1681 iface = port->ifaces[e->iface_idx];
1683 *dp_ifidx = iface->dp_ifidx;
1684 *tags |= iface->tag; /* Currently only used for bonding. */
1689 bond_link_status_update(struct iface *iface, bool carrier)
1691 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1692 struct port *port = iface->port;
1694 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1695 /* Nothing to do. */
1698 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1699 iface->name, carrier ? "detected" : "dropped");
1700 if (carrier == iface->enabled) {
1701 iface->delay_expires = LLONG_MAX;
1702 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1703 iface->name, carrier ? "disabled" : "enabled");
1704 } else if (carrier && port->active_iface < 0) {
1705 bond_enable_slave(iface, true);
1706 if (port->updelay) {
1707 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1708 "other interface is up", iface->name, port->updelay);
1711 int delay = carrier ? port->updelay : port->downdelay;
1712 iface->delay_expires = time_msec() + delay;
1715 "interface %s: will be %s if it stays %s for %d ms",
1717 carrier ? "enabled" : "disabled",
1718 carrier ? "up" : "down",
1725 bond_choose_active_iface(struct port *port)
1727 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1729 port->active_iface = bond_choose_iface(port);
1730 port->active_iface_tag = tag_create_random();
1731 if (port->active_iface >= 0) {
1732 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1733 port->name, port->ifaces[port->active_iface]->name);
1735 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1741 bond_enable_slave(struct iface *iface, bool enable)
1743 struct port *port = iface->port;
1744 struct bridge *br = port->bridge;
1746 /* This acts as a recursion check. If the act of disabling a slave
1747 * causes a different slave to be enabled, the flag will allow us to
1748 * skip redundant work when we reenter this function. It must be
1749 * cleared on exit to keep things safe with multiple bonds. */
1750 static bool moving_active_iface = false;
1752 iface->delay_expires = LLONG_MAX;
1753 if (enable == iface->enabled) {
1757 iface->enabled = enable;
1758 if (!iface->enabled) {
1759 VLOG_WARN("interface %s: disabled", iface->name);
1760 ofproto_revalidate(br->ofproto, iface->tag);
1761 if (iface->port_ifidx == port->active_iface) {
1762 ofproto_revalidate(br->ofproto,
1763 port->active_iface_tag);
1765 /* Disabling a slave can lead to another slave being immediately
1766 * enabled if there will be no active slaves but one is waiting
1767 * on an updelay. In this case we do not need to run most of the
1768 * code for the newly enabled slave since there was no period
1769 * without an active slave and it is redundant with the disabling
1771 moving_active_iface = true;
1772 bond_choose_active_iface(port);
1774 bond_send_learning_packets(port);
1776 VLOG_WARN("interface %s: enabled", iface->name);
1777 if (port->active_iface < 0 && !moving_active_iface) {
1778 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1779 bond_choose_active_iface(port);
1780 bond_send_learning_packets(port);
1782 iface->tag = tag_create_random();
1785 moving_active_iface = false;
1786 port->bond_compat_is_stale = true;
1790 bond_run(struct bridge *br)
1794 for (i = 0; i < br->n_ports; i++) {
1795 struct port *port = br->ports[i];
1797 if (port->n_ifaces >= 2) {
1798 for (j = 0; j < port->n_ifaces; j++) {
1799 struct iface *iface = port->ifaces[j];
1800 if (time_msec() >= iface->delay_expires) {
1801 bond_enable_slave(iface, !iface->enabled);
1806 if (port->bond_compat_is_stale) {
1807 port->bond_compat_is_stale = false;
1808 port_update_bond_compat(port);
1814 bond_wait(struct bridge *br)
1818 for (i = 0; i < br->n_ports; i++) {
1819 struct port *port = br->ports[i];
1820 if (port->n_ifaces < 2) {
1823 for (j = 0; j < port->n_ifaces; j++) {
1824 struct iface *iface = port->ifaces[j];
1825 if (iface->delay_expires != LLONG_MAX) {
1826 poll_timer_wait(iface->delay_expires - time_msec());
1833 set_dst(struct dst *p, const flow_t *flow,
1834 const struct port *in_port, const struct port *out_port,
1837 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1838 : in_port->vlan >= 0 ? in_port->vlan
1839 : ntohs(flow->dl_vlan));
1840 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1844 swap_dst(struct dst *p, struct dst *q)
1846 struct dst tmp = *p;
1851 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1852 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1853 * that we push to the datapath. We could in fact fully sort the array by
1854 * vlan, but in most cases there are at most two different vlan tags so that's
1855 * possibly overkill.) */
1857 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1859 struct dst *first = dsts;
1860 struct dst *last = dsts + n_dsts;
1862 while (first != last) {
1864 * - All dsts < first have vlan == 'vlan'.
1865 * - All dsts >= last have vlan != 'vlan'.
1866 * - first < last. */
1867 while (first->vlan == vlan) {
1868 if (++first == last) {
1873 /* Same invariants, plus one additional:
1874 * - first->vlan != vlan.
1876 while (last[-1].vlan != vlan) {
1877 if (--last == first) {
1882 /* Same invariants, plus one additional:
1883 * - last[-1].vlan == vlan.*/
1884 swap_dst(first++, --last);
1889 mirror_mask_ffs(mirror_mask_t mask)
1891 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1896 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1897 const struct dst *test)
1900 for (i = 0; i < n_dsts; i++) {
1901 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1909 port_trunks_vlan(const struct port *port, uint16_t vlan)
1911 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1915 port_includes_vlan(const struct port *port, uint16_t vlan)
1917 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1921 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1922 const struct port *in_port, const struct port *out_port,
1923 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1925 mirror_mask_t mirrors = in_port->src_mirrors;
1926 struct dst *dst = dsts;
1929 if (out_port == FLOOD_PORT) {
1930 /* XXX use ODP_FLOOD if no vlans or bonding. */
1931 /* XXX even better, define each VLAN as a datapath port group */
1932 for (i = 0; i < br->n_ports; i++) {
1933 struct port *port = br->ports[i];
1934 if (port != in_port && port_includes_vlan(port, vlan)
1935 && !port->is_mirror_output_port
1936 && set_dst(dst, flow, in_port, port, tags)) {
1937 mirrors |= port->dst_mirrors;
1941 *nf_output_iface = NF_OUT_FLOOD;
1942 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1943 *nf_output_iface = dst->dp_ifidx;
1944 mirrors |= out_port->dst_mirrors;
1949 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1950 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1952 if (set_dst(dst, flow, in_port, m->out_port, tags)
1953 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1957 for (i = 0; i < br->n_ports; i++) {
1958 struct port *port = br->ports[i];
1959 if (port_includes_vlan(port, m->out_vlan)
1960 && set_dst(dst, flow, in_port, port, tags))
1964 if (port->vlan < 0) {
1965 dst->vlan = m->out_vlan;
1967 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1971 /* Use the vlan tag on the original flow instead of
1972 * the one passed in the vlan parameter. This ensures
1973 * that we compare the vlan from before any implicit
1974 * tagging tags place. This is necessary because
1975 * dst->vlan is the final vlan, after removing implicit
1977 flow_vlan = ntohs(flow->dl_vlan);
1978 if (flow_vlan == 0) {
1979 flow_vlan = OFP_VLAN_NONE;
1981 if (port == in_port && dst->vlan == flow_vlan) {
1982 /* Don't send out input port on same VLAN. */
1990 mirrors &= mirrors - 1;
1993 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1998 print_dsts(const struct dst *dsts, size_t n)
2000 for (; n--; dsts++) {
2001 printf(">p%"PRIu16, dsts->dp_ifidx);
2002 if (dsts->vlan != OFP_VLAN_NONE) {
2003 printf("v%"PRIu16, dsts->vlan);
2009 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2010 const struct port *in_port, const struct port *out_port,
2011 tag_type *tags, struct odp_actions *actions,
2012 uint16_t *nf_output_iface)
2014 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2016 const struct dst *p;
2019 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2022 cur_vlan = ntohs(flow->dl_vlan);
2023 for (p = dsts; p < &dsts[n_dsts]; p++) {
2024 union odp_action *a;
2025 if (p->vlan != cur_vlan) {
2026 if (p->vlan == OFP_VLAN_NONE) {
2027 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2029 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2030 a->vlan_vid.vlan_vid = htons(p->vlan);
2034 a = odp_actions_add(actions, ODPAT_OUTPUT);
2035 a->output.port = p->dp_ifidx;
2039 /* Returns the effective vlan of a packet, taking into account both the
2040 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2041 * the packet is untagged and -1 indicates it has an invalid header and
2042 * should be dropped. */
2043 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2044 struct port *in_port, bool have_packet)
2046 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2047 * belongs to VLAN 0, so we should treat both cases identically. (In the
2048 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2049 * presumably to allow a priority to be specified. In the latter case, the
2050 * packet does not have any 802.1Q header.) */
2051 int vlan = ntohs(flow->dl_vlan);
2052 if (vlan == OFP_VLAN_NONE) {
2055 if (in_port->vlan >= 0) {
2057 /* XXX support double tagging? */
2059 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2060 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2061 "packet received on port %s configured with "
2062 "implicit VLAN %"PRIu16,
2063 br->name, ntohs(flow->dl_vlan),
2064 in_port->name, in_port->vlan);
2068 vlan = in_port->vlan;
2070 if (!port_includes_vlan(in_port, vlan)) {
2072 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2073 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2074 "packet received on port %s not configured for "
2076 br->name, vlan, in_port->name, vlan);
2086 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2087 struct port *in_port)
2089 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2090 vlan, in_port->port_idx);
2092 /* The log messages here could actually be useful in debugging,
2093 * so keep the rate limit relatively high. */
2094 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2096 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2097 "on port %s in VLAN %d",
2098 br->name, ETH_ADDR_ARGS(flow->dl_src),
2099 in_port->name, vlan);
2100 ofproto_revalidate(br->ofproto, rev_tag);
2105 is_bcast_arp_reply(const flow_t *flow)
2107 return (flow->dl_type == htons(ETH_TYPE_ARP)
2108 && flow->nw_proto == ARP_OP_REPLY
2109 && eth_addr_is_broadcast(flow->dl_dst));
2112 /* If the composed actions may be applied to any packet in the given 'flow',
2113 * returns true. Otherwise, the actions should only be applied to 'packet', or
2114 * not at all, if 'packet' was NULL. */
2116 process_flow(struct bridge *br, const flow_t *flow,
2117 const struct ofpbuf *packet, struct odp_actions *actions,
2118 tag_type *tags, uint16_t *nf_output_iface)
2120 struct iface *in_iface;
2121 struct port *in_port;
2122 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2126 /* Find the interface and port structure for the received packet. */
2127 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2129 /* No interface? Something fishy... */
2130 if (packet != NULL) {
2131 /* Odd. A few possible reasons here:
2133 * - We deleted an interface but there are still a few packets
2134 * queued up from it.
2136 * - Someone externally added an interface (e.g. with "ovs-dpctl
2137 * add-if") that we don't know about.
2139 * - Packet arrived on the local port but the local port is not
2140 * one of our bridge ports.
2142 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2144 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2145 "interface %"PRIu16, br->name, flow->in_port);
2148 /* Return without adding any actions, to drop packets on this flow. */
2151 in_port = in_iface->port;
2152 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2157 /* Drop frames for reserved multicast addresses. */
2158 if (eth_addr_is_reserved(flow->dl_dst)) {
2162 /* Drop frames on ports reserved for mirroring. */
2163 if (in_port->is_mirror_output_port) {
2164 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2165 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2166 "which is reserved exclusively for mirroring",
2167 br->name, in_port->name);
2171 /* Packets received on bonds need special attention to avoid duplicates. */
2172 if (in_port->n_ifaces > 1) {
2175 if (eth_addr_is_multicast(flow->dl_dst)) {
2176 *tags |= in_port->active_iface_tag;
2177 if (in_port->active_iface != in_iface->port_ifidx) {
2178 /* Drop all multicast packets on inactive slaves. */
2183 /* Drop all packets for which we have learned a different input
2184 * port, because we probably sent the packet on one slave and got
2185 * it back on the other. Broadcast ARP replies are an exception
2186 * to this rule: the host has moved to another switch. */
2187 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2188 if (src_idx != -1 && src_idx != in_port->port_idx &&
2189 !is_bcast_arp_reply(flow)) {
2195 out_port = FLOOD_PORT;
2196 /* Learn source MAC (but don't try to learn from revalidation). */
2198 update_learning_table(br, flow, vlan, in_port);
2201 /* Determine output port. */
2202 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2204 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2205 out_port = br->ports[out_port_idx];
2206 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2207 /* If we are revalidating but don't have a learning entry then
2208 * eject the flow. Installing a flow that floods packets opens
2209 * up a window of time where we could learn from a packet reflected
2210 * on a bond and blackhole packets before the learning table is
2211 * updated to reflect the correct port. */
2215 /* Don't send packets out their input ports. */
2216 if (in_port == out_port) {
2221 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2227 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2230 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2231 const struct ofp_phy_port *opp,
2234 struct bridge *br = br_;
2235 struct iface *iface;
2238 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2244 if (reason == OFPPR_DELETE) {
2245 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2246 br->name, iface->name);
2247 iface_destroy(iface);
2248 if (!port->n_ifaces) {
2249 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2250 br->name, port->name);
2256 if (port->n_ifaces > 1) {
2257 bool up = !(opp->state & OFPPS_LINK_DOWN);
2258 bond_link_status_update(iface, up);
2259 port_update_bond_compat(port);
2265 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2266 struct odp_actions *actions, tag_type *tags,
2267 uint16_t *nf_output_iface, void *br_)
2269 struct bridge *br = br_;
2271 COVERAGE_INC(bridge_process_flow);
2272 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2276 bridge_account_flow_ofhook_cb(const flow_t *flow,
2277 const union odp_action *actions,
2278 size_t n_actions, unsigned long long int n_bytes,
2281 struct bridge *br = br_;
2282 struct port *in_port;
2283 const union odp_action *a;
2285 /* Feed information from the active flows back into the learning table
2286 * to ensure that table is always in sync with what is actually flowing
2287 * through the datapath. */
2288 in_port = port_from_dp_ifidx(br, flow->in_port);
2290 int vlan = flow_get_vlan(br, flow, in_port, false);
2292 update_learning_table(br, flow, vlan, in_port);
2296 if (!br->has_bonded_ports) {
2300 for (a = actions; a < &actions[n_actions]; a++) {
2301 if (a->type == ODPAT_OUTPUT) {
2302 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2303 if (out_port && out_port->n_ifaces >= 2) {
2304 struct bond_entry *e = lookup_bond_entry(out_port,
2306 e->tx_bytes += n_bytes;
2313 bridge_account_checkpoint_ofhook_cb(void *br_)
2315 struct bridge *br = br_;
2318 if (!br->has_bonded_ports) {
2322 /* The current ofproto implementation calls this callback at least once a
2323 * second, so this timer implementation is sufficient. */
2324 if (time_msec() < br->bond_next_rebalance) {
2327 br->bond_next_rebalance = time_msec() + 10000;
2329 for (i = 0; i < br->n_ports; i++) {
2330 struct port *port = br->ports[i];
2331 if (port->n_ifaces > 1) {
2332 bond_rebalance_port(port);
2337 static struct ofhooks bridge_ofhooks = {
2338 bridge_port_changed_ofhook_cb,
2339 bridge_normal_ofhook_cb,
2340 bridge_account_flow_ofhook_cb,
2341 bridge_account_checkpoint_ofhook_cb,
2344 /* Bonding functions. */
2346 /* Statistics for a single interface on a bonded port, used for load-based
2347 * bond rebalancing. */
2348 struct slave_balance {
2349 struct iface *iface; /* The interface. */
2350 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2352 /* All the "bond_entry"s that are assigned to this interface, in order of
2353 * increasing tx_bytes. */
2354 struct bond_entry **hashes;
2358 /* Sorts pointers to pointers to bond_entries in ascending order by the
2359 * interface to which they are assigned, and within a single interface in
2360 * ascending order of bytes transmitted. */
2362 compare_bond_entries(const void *a_, const void *b_)
2364 const struct bond_entry *const *ap = a_;
2365 const struct bond_entry *const *bp = b_;
2366 const struct bond_entry *a = *ap;
2367 const struct bond_entry *b = *bp;
2368 if (a->iface_idx != b->iface_idx) {
2369 return a->iface_idx > b->iface_idx ? 1 : -1;
2370 } else if (a->tx_bytes != b->tx_bytes) {
2371 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2377 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2378 * *descending* order by number of bytes transmitted. */
2380 compare_slave_balance(const void *a_, const void *b_)
2382 const struct slave_balance *a = a_;
2383 const struct slave_balance *b = b_;
2384 if (a->iface->enabled != b->iface->enabled) {
2385 return a->iface->enabled ? -1 : 1;
2386 } else if (a->tx_bytes != b->tx_bytes) {
2387 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2394 swap_bals(struct slave_balance *a, struct slave_balance *b)
2396 struct slave_balance tmp = *a;
2401 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2402 * given that 'p' (and only 'p') might be in the wrong location.
2404 * This function invalidates 'p', since it might now be in a different memory
2407 resort_bals(struct slave_balance *p,
2408 struct slave_balance bals[], size_t n_bals)
2411 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2412 swap_bals(p, p - 1);
2414 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2415 swap_bals(p, p + 1);
2421 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2423 if (VLOG_IS_DBG_ENABLED()) {
2424 struct ds ds = DS_EMPTY_INITIALIZER;
2425 const struct slave_balance *b;
2427 for (b = bals; b < bals + n_bals; b++) {
2431 ds_put_char(&ds, ',');
2433 ds_put_format(&ds, " %s %"PRIu64"kB",
2434 b->iface->name, b->tx_bytes / 1024);
2436 if (!b->iface->enabled) {
2437 ds_put_cstr(&ds, " (disabled)");
2439 if (b->n_hashes > 0) {
2440 ds_put_cstr(&ds, " (");
2441 for (i = 0; i < b->n_hashes; i++) {
2442 const struct bond_entry *e = b->hashes[i];
2444 ds_put_cstr(&ds, " + ");
2446 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2447 e - port->bond_hash, e->tx_bytes / 1024);
2449 ds_put_cstr(&ds, ")");
2452 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2457 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2459 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2462 struct bond_entry *hash = from->hashes[hash_idx];
2463 struct port *port = from->iface->port;
2464 uint64_t delta = hash->tx_bytes;
2466 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2467 "from %s to %s (now carrying %"PRIu64"kB and "
2468 "%"PRIu64"kB load, respectively)",
2469 port->name, delta / 1024, hash - port->bond_hash,
2470 from->iface->name, to->iface->name,
2471 (from->tx_bytes - delta) / 1024,
2472 (to->tx_bytes + delta) / 1024);
2474 /* Delete element from from->hashes.
2476 * We don't bother to add the element to to->hashes because not only would
2477 * it require more work, the only purpose it would be to allow that hash to
2478 * be migrated to another slave in this rebalancing run, and there is no
2479 * point in doing that. */
2480 if (hash_idx == 0) {
2483 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2484 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2488 /* Shift load away from 'from' to 'to'. */
2489 from->tx_bytes -= delta;
2490 to->tx_bytes += delta;
2492 /* Arrange for flows to be revalidated. */
2493 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2494 hash->iface_idx = to->iface->port_ifidx;
2495 hash->iface_tag = tag_create_random();
2499 bond_rebalance_port(struct port *port)
2501 struct slave_balance bals[DP_MAX_PORTS];
2503 struct bond_entry *hashes[BOND_MASK + 1];
2504 struct slave_balance *b, *from, *to;
2505 struct bond_entry *e;
2508 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2509 * descending order of tx_bytes, so that bals[0] represents the most
2510 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2513 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2514 * array for each slave_balance structure, we sort our local array of
2515 * hashes in order by slave, so that all of the hashes for a given slave
2516 * become contiguous in memory, and then we point each 'hashes' members of
2517 * a slave_balance structure to the start of a contiguous group. */
2518 n_bals = port->n_ifaces;
2519 for (b = bals; b < &bals[n_bals]; b++) {
2520 b->iface = port->ifaces[b - bals];
2525 for (i = 0; i <= BOND_MASK; i++) {
2526 hashes[i] = &port->bond_hash[i];
2528 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2529 for (i = 0; i <= BOND_MASK; i++) {
2531 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2532 b = &bals[e->iface_idx];
2533 b->tx_bytes += e->tx_bytes;
2535 b->hashes = &hashes[i];
2540 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2541 log_bals(bals, n_bals, port);
2543 /* Discard slaves that aren't enabled (which were sorted to the back of the
2544 * array earlier). */
2545 while (!bals[n_bals - 1].iface->enabled) {
2552 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2553 to = &bals[n_bals - 1];
2554 for (from = bals; from < to; ) {
2555 uint64_t overload = from->tx_bytes - to->tx_bytes;
2556 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2557 /* The extra load on 'from' (and all less-loaded slaves), compared
2558 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2559 * it is less than ~1Mbps. No point in rebalancing. */
2561 } else if (from->n_hashes == 1) {
2562 /* 'from' only carries a single MAC hash, so we can't shift any
2563 * load away from it, even though we want to. */
2566 /* 'from' is carrying significantly more load than 'to', and that
2567 * load is split across at least two different hashes. Pick a hash
2568 * to migrate to 'to' (the least-loaded slave), given that doing so
2569 * must decrease the ratio of the load on the two slaves by at
2572 * The sort order we use means that we prefer to shift away the
2573 * smallest hashes instead of the biggest ones. There is little
2574 * reason behind this decision; we could use the opposite sort
2575 * order to shift away big hashes ahead of small ones. */
2579 for (i = 0; i < from->n_hashes; i++) {
2580 double old_ratio, new_ratio;
2581 uint64_t delta = from->hashes[i]->tx_bytes;
2583 if (delta == 0 || from->tx_bytes - delta == 0) {
2584 /* Pointless move. */
2588 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2590 if (to->tx_bytes == 0) {
2591 /* Nothing on the new slave, move it. */
2595 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2596 new_ratio = (double)(from->tx_bytes - delta) /
2597 (to->tx_bytes + delta);
2599 if (new_ratio == 0) {
2600 /* Should already be covered but check to prevent division
2605 if (new_ratio < 1) {
2606 new_ratio = 1 / new_ratio;
2609 if (old_ratio - new_ratio > 0.1) {
2610 /* Would decrease the ratio, move it. */
2614 if (i < from->n_hashes) {
2615 bond_shift_load(from, to, i);
2616 port->bond_compat_is_stale = true;
2618 /* If the result of the migration changed the relative order of
2619 * 'from' and 'to' swap them back to maintain invariants. */
2620 if (order_swapped) {
2621 swap_bals(from, to);
2624 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2625 * point to different slave_balance structures. It is only
2626 * valid to do these two operations in a row at all because we
2627 * know that 'from' will not move past 'to' and vice versa. */
2628 resort_bals(from, bals, n_bals);
2629 resort_bals(to, bals, n_bals);
2636 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2637 * historical data to decay to <1% in 7 rebalancing runs. */
2638 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2644 bond_send_learning_packets(struct port *port)
2646 struct bridge *br = port->bridge;
2647 struct mac_entry *e;
2648 struct ofpbuf packet;
2649 int error, n_packets, n_errors;
2651 if (!port->n_ifaces || port->active_iface < 0) {
2655 ofpbuf_init(&packet, 128);
2656 error = n_packets = n_errors = 0;
2657 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2658 union ofp_action actions[2], *a;
2664 if (e->port == port->port_idx
2665 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2669 /* Compose actions. */
2670 memset(actions, 0, sizeof actions);
2673 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2674 a->vlan_vid.len = htons(sizeof *a);
2675 a->vlan_vid.vlan_vid = htons(e->vlan);
2678 a->output.type = htons(OFPAT_OUTPUT);
2679 a->output.len = htons(sizeof *a);
2680 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2685 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2687 flow_extract(&packet, ODPP_NONE, &flow);
2688 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2695 ofpbuf_uninit(&packet);
2698 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2699 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2700 "packets, last error was: %s",
2701 port->name, n_errors, n_packets, strerror(error));
2703 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2704 port->name, n_packets);
2708 /* Bonding unixctl user interface functions. */
2711 bond_unixctl_list(struct unixctl_conn *conn,
2712 const char *args UNUSED, void *aux UNUSED)
2714 struct ds ds = DS_EMPTY_INITIALIZER;
2715 const struct bridge *br;
2717 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2719 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2722 for (i = 0; i < br->n_ports; i++) {
2723 const struct port *port = br->ports[i];
2724 if (port->n_ifaces > 1) {
2727 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2728 for (j = 0; j < port->n_ifaces; j++) {
2729 const struct iface *iface = port->ifaces[j];
2731 ds_put_cstr(&ds, ", ");
2733 ds_put_cstr(&ds, iface->name);
2735 ds_put_char(&ds, '\n');
2739 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2743 static struct port *
2744 bond_find(const char *name)
2746 const struct bridge *br;
2748 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2751 for (i = 0; i < br->n_ports; i++) {
2752 struct port *port = br->ports[i];
2753 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2762 bond_unixctl_show(struct unixctl_conn *conn,
2763 const char *args, void *aux UNUSED)
2765 struct ds ds = DS_EMPTY_INITIALIZER;
2766 const struct port *port;
2769 port = bond_find(args);
2771 unixctl_command_reply(conn, 501, "no such bond");
2775 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2776 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2777 ds_put_format(&ds, "next rebalance: %lld ms\n",
2778 port->bridge->bond_next_rebalance - time_msec());
2779 for (j = 0; j < port->n_ifaces; j++) {
2780 const struct iface *iface = port->ifaces[j];
2781 struct bond_entry *be;
2784 ds_put_format(&ds, "slave %s: %s\n",
2785 iface->name, iface->enabled ? "enabled" : "disabled");
2786 if (j == port->active_iface) {
2787 ds_put_cstr(&ds, "\tactive slave\n");
2789 if (iface->delay_expires != LLONG_MAX) {
2790 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2791 iface->enabled ? "downdelay" : "updelay",
2792 iface->delay_expires - time_msec());
2796 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2797 int hash = be - port->bond_hash;
2798 struct mac_entry *me;
2800 if (be->iface_idx != j) {
2804 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2805 hash, be->tx_bytes / 1024);
2808 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2809 &port->bridge->ml->lrus) {
2812 if (bond_hash(me->mac) == hash
2813 && me->port != port->port_idx
2814 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2815 && dp_ifidx == iface->dp_ifidx)
2817 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2818 ETH_ADDR_ARGS(me->mac));
2823 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2828 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2831 char *args = (char *) args_;
2832 char *save_ptr = NULL;
2833 char *bond_s, *hash_s, *slave_s;
2834 uint8_t mac[ETH_ADDR_LEN];
2836 struct iface *iface;
2837 struct bond_entry *entry;
2840 bond_s = strtok_r(args, " ", &save_ptr);
2841 hash_s = strtok_r(NULL, " ", &save_ptr);
2842 slave_s = strtok_r(NULL, " ", &save_ptr);
2844 unixctl_command_reply(conn, 501,
2845 "usage: bond/migrate BOND HASH SLAVE");
2849 port = bond_find(bond_s);
2851 unixctl_command_reply(conn, 501, "no such bond");
2855 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2856 == ETH_ADDR_SCAN_COUNT) {
2857 hash = bond_hash(mac);
2858 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2859 hash = atoi(hash_s) & BOND_MASK;
2861 unixctl_command_reply(conn, 501, "bad hash");
2865 iface = port_lookup_iface(port, slave_s);
2867 unixctl_command_reply(conn, 501, "no such slave");
2871 if (!iface->enabled) {
2872 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2876 entry = &port->bond_hash[hash];
2877 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2878 entry->iface_idx = iface->port_ifidx;
2879 entry->iface_tag = tag_create_random();
2880 port->bond_compat_is_stale = true;
2881 unixctl_command_reply(conn, 200, "migrated");
2885 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2888 char *args = (char *) args_;
2889 char *save_ptr = NULL;
2890 char *bond_s, *slave_s;
2892 struct iface *iface;
2894 bond_s = strtok_r(args, " ", &save_ptr);
2895 slave_s = strtok_r(NULL, " ", &save_ptr);
2897 unixctl_command_reply(conn, 501,
2898 "usage: bond/set-active-slave BOND SLAVE");
2902 port = bond_find(bond_s);
2904 unixctl_command_reply(conn, 501, "no such bond");
2908 iface = port_lookup_iface(port, slave_s);
2910 unixctl_command_reply(conn, 501, "no such slave");
2914 if (!iface->enabled) {
2915 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2919 if (port->active_iface != iface->port_ifidx) {
2920 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2921 port->active_iface = iface->port_ifidx;
2922 port->active_iface_tag = tag_create_random();
2923 VLOG_INFO("port %s: active interface is now %s",
2924 port->name, iface->name);
2925 bond_send_learning_packets(port);
2926 unixctl_command_reply(conn, 200, "done");
2928 unixctl_command_reply(conn, 200, "no change");
2933 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2935 char *args = (char *) args_;
2936 char *save_ptr = NULL;
2937 char *bond_s, *slave_s;
2939 struct iface *iface;
2941 bond_s = strtok_r(args, " ", &save_ptr);
2942 slave_s = strtok_r(NULL, " ", &save_ptr);
2944 unixctl_command_reply(conn, 501,
2945 "usage: bond/enable/disable-slave BOND SLAVE");
2949 port = bond_find(bond_s);
2951 unixctl_command_reply(conn, 501, "no such bond");
2955 iface = port_lookup_iface(port, slave_s);
2957 unixctl_command_reply(conn, 501, "no such slave");
2961 bond_enable_slave(iface, enable);
2962 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2966 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
2969 enable_slave(conn, args, true);
2973 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
2976 enable_slave(conn, args, false);
2980 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
2983 uint8_t mac[ETH_ADDR_LEN];
2987 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2988 == ETH_ADDR_SCAN_COUNT) {
2989 hash = bond_hash(mac);
2991 hash_cstr = xasprintf("%u", hash);
2992 unixctl_command_reply(conn, 200, hash_cstr);
2995 unixctl_command_reply(conn, 501, "invalid mac");
3002 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3003 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3004 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3005 unixctl_command_register("bond/set-active-slave",
3006 bond_unixctl_set_active_slave, NULL);
3007 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3009 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3011 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3014 /* Port functions. */
3016 static struct port *
3017 port_create(struct bridge *br, const char *name)
3021 port = xzalloc(sizeof *port);
3023 port->port_idx = br->n_ports;
3025 port->trunks = NULL;
3026 port->name = xstrdup(name);
3027 port->active_iface = -1;
3029 if (br->n_ports >= br->allocated_ports) {
3030 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3033 br->ports[br->n_ports++] = port;
3035 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3042 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3044 struct shash old_ifaces, new_ifaces;
3045 struct shash_node *node;
3046 unsigned long *trunks;
3052 /* Collect old and new interfaces. */
3053 shash_init(&old_ifaces);
3054 shash_init(&new_ifaces);
3055 for (i = 0; i < port->n_ifaces; i++) {
3056 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3058 for (i = 0; i < cfg->n_interfaces; i++) {
3059 const char *name = cfg->interfaces[i]->name;
3060 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3061 VLOG_WARN("port %s: %s specified twice as port interface",
3065 port->updelay = cfg->bond_updelay;
3066 if (port->updelay < 0) {
3069 port->updelay = cfg->bond_downdelay;
3070 if (port->downdelay < 0) {
3071 port->downdelay = 0;
3074 /* Get rid of deleted interfaces and add new interfaces. */
3075 SHASH_FOR_EACH (node, &old_ifaces) {
3076 if (!shash_find(&new_ifaces, node->name)) {
3077 iface_destroy(node->data);
3080 SHASH_FOR_EACH (node, &new_ifaces) {
3081 const struct ovsrec_interface *if_cfg = node->data;
3082 struct iface *iface;
3084 iface = shash_find_data(&old_ifaces, if_cfg->name);
3086 iface_create(port, if_cfg);
3088 iface->cfg = if_cfg;
3095 if (port->n_ifaces < 2) {
3097 if (vlan >= 0 && vlan <= 4095) {
3098 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3103 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3104 * they even work as-is. But they have not been tested. */
3105 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3109 if (port->vlan != vlan) {
3111 bridge_flush(port->bridge);
3114 /* Get trunked VLANs. */
3120 trunks = bitmap_allocate(4096);
3122 for (i = 0; i < cfg->n_trunks; i++) {
3123 int trunk = cfg->trunks[i];
3125 bitmap_set1(trunks, trunk);
3131 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3132 port->name, cfg->n_trunks);
3134 if (n_errors == cfg->n_trunks) {
3136 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3139 bitmap_set_multiple(trunks, 0, 4096, 1);
3142 if (cfg->n_trunks) {
3143 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3148 ? port->trunks != NULL
3149 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3150 bridge_flush(port->bridge);
3152 bitmap_free(port->trunks);
3153 port->trunks = trunks;
3155 shash_destroy(&old_ifaces);
3156 shash_destroy(&new_ifaces);
3160 port_destroy(struct port *port)
3163 struct bridge *br = port->bridge;
3167 proc_net_compat_update_vlan(port->name, NULL, 0);
3168 proc_net_compat_update_bond(port->name, NULL);
3170 for (i = 0; i < MAX_MIRRORS; i++) {
3171 struct mirror *m = br->mirrors[i];
3172 if (m && m->out_port == port) {
3177 while (port->n_ifaces > 0) {
3178 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3181 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3182 del->port_idx = port->port_idx;
3185 bitmap_free(port->trunks);
3192 static struct port *
3193 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3195 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3196 return iface ? iface->port : NULL;
3199 static struct port *
3200 port_lookup(const struct bridge *br, const char *name)
3204 for (i = 0; i < br->n_ports; i++) {
3205 struct port *port = br->ports[i];
3206 if (!strcmp(port->name, name)) {
3213 static struct iface *
3214 port_lookup_iface(const struct port *port, const char *name)
3218 for (j = 0; j < port->n_ifaces; j++) {
3219 struct iface *iface = port->ifaces[j];
3220 if (!strcmp(iface->name, name)) {
3228 port_update_bonding(struct port *port)
3230 if (port->n_ifaces < 2) {
3231 /* Not a bonded port. */
3232 if (port->bond_hash) {
3233 free(port->bond_hash);
3234 port->bond_hash = NULL;
3235 port->bond_compat_is_stale = true;
3236 port->bond_fake_iface = false;
3239 if (!port->bond_hash) {
3242 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3243 for (i = 0; i <= BOND_MASK; i++) {
3244 struct bond_entry *e = &port->bond_hash[i];
3248 port->no_ifaces_tag = tag_create_random();
3249 bond_choose_active_iface(port);
3251 port->bond_compat_is_stale = true;
3252 port->bond_fake_iface = port->cfg->bond_fake_iface;
3257 port_update_bond_compat(struct port *port)
3259 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3260 struct compat_bond bond;
3263 if (port->n_ifaces < 2) {
3264 proc_net_compat_update_bond(port->name, NULL);
3269 bond.updelay = port->updelay;
3270 bond.downdelay = port->downdelay;
3273 bond.hashes = compat_hashes;
3274 if (port->bond_hash) {
3275 const struct bond_entry *e;
3276 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3277 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3278 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3279 cbh->hash = e - port->bond_hash;
3280 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3285 bond.n_slaves = port->n_ifaces;
3286 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3287 for (i = 0; i < port->n_ifaces; i++) {
3288 struct iface *iface = port->ifaces[i];
3289 struct compat_bond_slave *slave = &bond.slaves[i];
3290 slave->name = iface->name;
3292 /* We need to make the same determination as the Linux bonding
3293 * code to determine whether a slave should be consider "up".
3294 * The Linux function bond_miimon_inspect() supports four
3295 * BOND_LINK_* states:
3297 * - BOND_LINK_UP: carrier detected, updelay has passed.
3298 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3299 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3300 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3302 * The function bond_info_show_slave() only considers BOND_LINK_UP
3303 * to be "up" and anything else to be "down".
3305 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3309 netdev_get_etheraddr(iface->netdev, slave->mac);
3312 if (port->bond_fake_iface) {
3313 struct netdev *bond_netdev;
3315 if (!netdev_open_default(port->name, &bond_netdev)) {
3317 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3319 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3321 netdev_close(bond_netdev);
3325 proc_net_compat_update_bond(port->name, &bond);
3330 port_update_vlan_compat(struct port *port)
3332 struct bridge *br = port->bridge;
3333 char *vlandev_name = NULL;
3335 if (port->vlan > 0) {
3336 /* Figure out the name that the VLAN device should actually have, if it
3337 * existed. This takes some work because the VLAN device would not
3338 * have port->name in its name; rather, it would have the trunk port's
3339 * name, and 'port' would be attached to a bridge that also had the
3340 * VLAN device one of its ports. So we need to find a trunk port that
3341 * includes port->vlan.
3343 * There might be more than one candidate. This doesn't happen on
3344 * XenServer, so if it happens we just pick the first choice in
3345 * alphabetical order instead of creating multiple VLAN devices. */
3347 for (i = 0; i < br->n_ports; i++) {
3348 struct port *p = br->ports[i];
3349 if (port_trunks_vlan(p, port->vlan)
3351 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3353 uint8_t ea[ETH_ADDR_LEN];
3354 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3355 if (!eth_addr_is_multicast(ea) &&
3356 !eth_addr_is_reserved(ea) &&
3357 !eth_addr_is_zero(ea)) {
3358 vlandev_name = p->name;
3363 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3366 /* Interface functions. */
3368 static struct iface *
3369 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3371 struct iface *iface;
3372 char *name = if_cfg->name;
3375 iface = xzalloc(sizeof *iface);
3377 iface->port_ifidx = port->n_ifaces;
3378 iface->name = xstrdup(name);
3379 iface->dp_ifidx = -1;
3380 iface->tag = tag_create_random();
3381 iface->delay_expires = LLONG_MAX;
3382 iface->netdev = NULL;
3383 iface->cfg = if_cfg;
3385 if (port->n_ifaces >= port->allocated_ifaces) {
3386 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3387 sizeof *port->ifaces);
3389 port->ifaces[port->n_ifaces++] = iface;
3390 if (port->n_ifaces > 1) {
3391 port->bridge->has_bonded_ports = true;
3394 /* Attempt to create the network interface in case it
3395 * doesn't exist yet. */
3396 if (!iface_is_internal(port->bridge, iface->name)) {
3397 error = set_up_iface(if_cfg, iface, true);
3399 VLOG_WARN("could not create iface %s: %s", iface->name,
3404 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3406 bridge_flush(port->bridge);
3412 iface_destroy(struct iface *iface)
3415 struct port *port = iface->port;
3416 struct bridge *br = port->bridge;
3417 bool del_active = port->active_iface == iface->port_ifidx;
3420 if (iface->dp_ifidx >= 0) {
3421 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3424 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3425 del->port_ifidx = iface->port_ifidx;
3427 netdev_close(iface->netdev);
3430 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3431 bond_choose_active_iface(port);
3432 bond_send_learning_packets(port);
3438 bridge_flush(port->bridge);
3442 static struct iface *
3443 iface_lookup(const struct bridge *br, const char *name)
3447 for (i = 0; i < br->n_ports; i++) {
3448 struct port *port = br->ports[i];
3449 for (j = 0; j < port->n_ifaces; j++) {
3450 struct iface *iface = port->ifaces[j];
3451 if (!strcmp(iface->name, name)) {
3459 static struct iface *
3460 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3462 return port_array_get(&br->ifaces, dp_ifidx);
3465 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3466 * 'br', that is, an interface that is entirely simulated within the datapath.
3467 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3468 * interfaces are created by setting "iface.<iface>.internal = true".
3470 * In addition, we have a kluge-y feature that creates an internal port with
3471 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3472 * This feature needs to go away in the long term. Until then, this is one
3473 * reason why this function takes a name instead of a struct iface: the fake
3474 * interfaces created this way do not have a struct iface. */
3476 iface_is_internal(const struct bridge *br, const char *if_name)
3478 /* XXX wastes time */
3479 struct iface *iface;
3482 if (!strcmp(if_name, br->name)) {
3486 iface = iface_lookup(br, if_name);
3487 if (iface && !strcmp(iface->cfg->type, "internal")) {
3491 port = port_lookup(br, if_name);
3492 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3498 /* Set Ethernet address of 'iface', if one is specified in the configuration
3501 iface_set_mac(struct iface *iface)
3503 uint8_t ea[ETH_ADDR_LEN];
3505 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3506 if (eth_addr_is_multicast(ea)) {
3507 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3509 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3510 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3511 iface->name, iface->name);
3513 int error = netdev_set_etheraddr(iface->netdev, ea);
3515 VLOG_ERR("interface %s: setting MAC failed (%s)",
3516 iface->name, strerror(error));
3522 /* Port mirroring. */
3525 mirror_reconfigure(struct bridge *br)
3527 struct shash old_mirrors, new_mirrors;
3528 struct shash_node *node;
3529 unsigned long *rspan_vlans;
3532 /* Collect old mirrors. */
3533 shash_init(&old_mirrors);
3534 for (i = 0; i < MAX_MIRRORS; i++) {
3535 if (br->mirrors[i]) {
3536 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3540 /* Collect new mirrors. */
3541 shash_init(&new_mirrors);
3542 for (i = 0; i < br->cfg->n_mirrors; i++) {
3543 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3544 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3545 VLOG_WARN("bridge %s: %s specified twice as mirror",
3546 br->name, cfg->name);
3550 /* Get rid of deleted mirrors and add new mirrors. */
3551 SHASH_FOR_EACH (node, &old_mirrors) {
3552 if (!shash_find(&new_mirrors, node->name)) {
3553 mirror_destroy(node->data);
3556 SHASH_FOR_EACH (node, &new_mirrors) {
3557 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3559 mirror = mirror_create(br, node->name);
3564 mirror_reconfigure_one(mirror, node->data);
3566 shash_destroy(&old_mirrors);
3567 shash_destroy(&new_mirrors);
3569 /* Update port reserved status. */
3570 for (i = 0; i < br->n_ports; i++) {
3571 br->ports[i]->is_mirror_output_port = false;
3573 for (i = 0; i < MAX_MIRRORS; i++) {
3574 struct mirror *m = br->mirrors[i];
3575 if (m && m->out_port) {
3576 m->out_port->is_mirror_output_port = true;
3580 /* Update flooded vlans (for RSPAN). */
3582 if (br->cfg->n_flood_vlans) {
3583 rspan_vlans = bitmap_allocate(4096);
3585 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3586 int64_t vlan = br->cfg->flood_vlans[i];
3587 if (vlan >= 0 && vlan < 4096) {
3588 bitmap_set1(rspan_vlans, vlan);
3589 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3592 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3597 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3602 static struct mirror *
3603 mirror_create(struct bridge *br, const char *name)
3608 for (i = 0; ; i++) {
3609 if (i >= MAX_MIRRORS) {
3610 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3611 "cannot create %s", br->name, MAX_MIRRORS, name);
3614 if (!br->mirrors[i]) {
3619 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3622 br->mirrors[i] = m = xzalloc(sizeof *m);
3625 m->name = xstrdup(name);
3626 shash_init(&m->src_ports);
3627 shash_init(&m->dst_ports);
3637 mirror_destroy(struct mirror *m)
3640 struct bridge *br = m->bridge;
3643 for (i = 0; i < br->n_ports; i++) {
3644 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3645 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3648 shash_destroy(&m->src_ports);
3649 shash_destroy(&m->dst_ports);
3652 m->bridge->mirrors[m->idx] = NULL;
3660 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3661 struct shash *names)
3665 for (i = 0; i < n_ports; i++) {
3666 const char *name = ports[i]->name;
3667 if (port_lookup(m->bridge, name)) {
3668 shash_add_once(names, name, NULL);
3670 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3671 "port %s", m->bridge->name, m->name, name);
3677 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3683 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3685 for (i = 0; i < cfg->n_select_vlan; i++) {
3686 int64_t vlan = cfg->select_vlan[i];
3687 if (vlan < 0 || vlan > 4095) {
3688 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3689 m->bridge->name, m->name, vlan);
3691 (*vlans)[n_vlans++] = vlan;
3698 vlan_is_mirrored(const struct mirror *m, int vlan)
3702 for (i = 0; i < m->n_vlans; i++) {
3703 if (m->vlans[i] == vlan) {
3711 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3715 for (i = 0; i < m->n_vlans; i++) {
3716 if (port_trunks_vlan(p, m->vlans[i])) {
3724 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3726 struct shash src_ports, dst_ports;
3727 mirror_mask_t mirror_bit;
3728 struct port *out_port;
3733 bool mirror_all_ports;
3734 bool any_ports_specified;
3735 bool any_vlans_specified;
3737 /* Get output port. */
3738 if (cfg->output_port) {
3739 out_port = port_lookup(m->bridge, cfg->output_port->name);
3741 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3742 m->bridge->name, m->name);
3748 if (cfg->output_vlan) {
3749 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3750 "output vlan; ignoring output vlan",
3751 m->bridge->name, m->name);
3753 } else if (cfg->output_vlan) {
3755 out_vlan = *cfg->output_vlan;
3757 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3758 m->bridge->name, m->name);
3763 /* Get all the ports, and drop duplicates and ports that don't exist. */
3764 shash_init(&src_ports);
3765 shash_init(&dst_ports);
3766 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3768 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3770 any_ports_specified = cfg->n_select_dst_port || cfg->n_select_dst_port;
3771 if (any_ports_specified
3772 && shash_is_empty(&src_ports) && shash_is_empty(&dst_ports)) {
3773 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3774 "selection ports exists", m->bridge->name, m->name);
3779 /* Get all the vlans, and drop duplicate and invalid vlans. */
3780 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3781 any_vlans_specified = cfg->n_select_vlan > 0;
3782 if (any_vlans_specified && !n_vlans) {
3783 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3784 "VLANs exists", m->bridge->name, m->name);
3789 /* Update mirror data. */
3790 if (!shash_equal_keys(&m->src_ports, &src_ports)
3791 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3792 || m->n_vlans != n_vlans
3793 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3794 || m->out_port != out_port
3795 || m->out_vlan != out_vlan) {
3796 bridge_flush(m->bridge);
3798 shash_swap(&m->src_ports, &src_ports);
3799 shash_swap(&m->dst_ports, &dst_ports);
3802 m->n_vlans = n_vlans;
3803 m->out_port = out_port;
3804 m->out_vlan = out_vlan;
3806 /* If no selection criteria have been given, mirror for all ports. */
3807 mirror_all_ports = !any_ports_specified && !any_vlans_specified;
3810 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3811 for (i = 0; i < m->bridge->n_ports; i++) {
3812 struct port *port = m->bridge->ports[i];
3814 if (mirror_all_ports
3815 || shash_find(&m->src_ports, port->name)
3818 ? port_trunks_any_mirrored_vlan(m, port)
3819 : vlan_is_mirrored(m, port->vlan)))) {
3820 port->src_mirrors |= mirror_bit;
3822 port->src_mirrors &= ~mirror_bit;
3825 if (mirror_all_ports || shash_find(&m->dst_ports, port->name)) {
3826 port->dst_mirrors |= mirror_bit;
3828 port->dst_mirrors &= ~mirror_bit;
3834 shash_destroy(&src_ports);
3835 shash_destroy(&dst_ports);