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? */
140 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
141 long long int bond_next_rebalance; /* Next rebalancing time. */
143 /* Port mirroring info. */
144 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
145 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
146 bool is_mirror_output_port; /* Does port mirroring send frames here? */
148 /* This member is only valid *during* bridge_reconfigure(). */
149 const struct ovsrec_port *cfg;
152 #define DP_MAX_PORTS 255
154 struct list node; /* Node in global list of bridges. */
155 char *name; /* User-specified arbitrary name. */
156 struct mac_learning *ml; /* MAC learning table. */
157 bool sent_config_request; /* Successfully sent config request? */
158 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
160 /* Support for remote controllers. */
161 char *controller; /* NULL if there is no remote controller;
162 * "discover" to do controller discovery;
163 * otherwise a vconn name. */
165 /* OpenFlow switch processing. */
166 struct ofproto *ofproto; /* OpenFlow switch. */
168 /* Description strings. */
169 char *mfr_desc; /* Manufacturer. */
170 char *hw_desc; /* Hardware. */
171 char *sw_desc; /* Software version. */
172 char *serial_desc; /* Serial number. */
173 char *dp_desc; /* Datapath description. */
175 /* Kernel datapath information. */
176 struct dpif *dpif; /* Datapath. */
177 struct port_array ifaces; /* Indexed by kernel datapath port number. */
181 size_t n_ports, allocated_ports;
184 bool has_bonded_ports;
189 /* Flow statistics gathering. */
190 time_t next_stats_request;
192 /* Port mirroring. */
193 struct mirror *mirrors[MAX_MIRRORS];
195 /* This member is only valid *during* bridge_reconfigure(). */
196 const struct ovsrec_bridge *cfg;
199 /* List of all bridges. */
200 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
202 /* Maximum number of datapaths. */
203 enum { DP_MAX = 256 };
205 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
206 static void bridge_destroy(struct bridge *);
207 static struct bridge *bridge_lookup(const char *name);
208 static unixctl_cb_func bridge_unixctl_dump_flows;
209 static int bridge_run_one(struct bridge *);
210 static const struct ovsrec_controller *bridge_get_controller(
211 const struct ovsrec_open_vswitch *ovs_cfg,
212 const struct bridge *br);
213 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
215 static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
217 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
218 static void bridge_fetch_dp_ifaces(struct bridge *);
219 static void bridge_flush(struct bridge *);
220 static void bridge_pick_local_hw_addr(struct bridge *,
221 uint8_t ea[ETH_ADDR_LEN],
222 struct iface **hw_addr_iface);
223 static uint64_t bridge_pick_datapath_id(struct bridge *,
224 const uint8_t bridge_ea[ETH_ADDR_LEN],
225 struct iface *hw_addr_iface);
226 static struct iface *bridge_get_local_iface(struct bridge *);
227 static uint64_t dpid_from_hash(const void *, size_t nbytes);
229 static unixctl_cb_func bridge_unixctl_fdb_show;
231 static void bond_init(void);
232 static void bond_run(struct bridge *);
233 static void bond_wait(struct bridge *);
234 static void bond_rebalance_port(struct port *);
235 static void bond_send_learning_packets(struct port *);
236 static void bond_enable_slave(struct iface *iface, bool enable);
238 static struct port *port_create(struct bridge *, const char *name);
239 static void port_reconfigure(struct port *, const struct ovsrec_port *);
240 static void port_destroy(struct port *);
241 static struct port *port_lookup(const struct bridge *, const char *name);
242 static struct iface *port_lookup_iface(const struct port *, const char *name);
243 static struct port *port_from_dp_ifidx(const struct bridge *,
245 static void port_update_bond_compat(struct port *);
246 static void port_update_vlan_compat(struct port *);
247 static void port_update_bonding(struct port *);
249 static struct mirror *mirror_create(struct bridge *, const char *name);
250 static void mirror_destroy(struct mirror *);
251 static void mirror_reconfigure(struct bridge *);
252 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
253 static bool vlan_is_mirrored(const struct mirror *, int vlan);
255 static struct iface *iface_create(struct port *port,
256 const struct ovsrec_interface *if_cfg);
257 static void iface_destroy(struct iface *);
258 static struct iface *iface_lookup(const struct bridge *, const char *name);
259 static struct iface *iface_from_dp_ifidx(const struct bridge *,
261 static bool iface_is_internal(const struct bridge *, const char *name);
262 static void iface_set_mac(struct iface *);
264 /* Hooks into ofproto processing. */
265 static struct ofhooks bridge_ofhooks;
267 /* Public functions. */
269 /* Adds the name of each interface used by a bridge, including local and
270 * internal ports, to 'svec'. */
272 bridge_get_ifaces(struct svec *svec)
274 struct bridge *br, *next;
277 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
278 for (i = 0; i < br->n_ports; i++) {
279 struct port *port = br->ports[i];
281 for (j = 0; j < port->n_ifaces; j++) {
282 struct iface *iface = port->ifaces[j];
283 if (iface->dp_ifidx < 0) {
284 VLOG_ERR("%s interface not in datapath %s, ignoring",
285 iface->name, dpif_name(br->dpif));
287 if (iface->dp_ifidx != ODPP_LOCAL) {
288 svec_add(svec, iface->name);
297 bridge_init(const struct ovsrec_open_vswitch *cfg)
299 struct svec bridge_names;
300 struct svec dpif_names, dpif_types;
303 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
305 svec_init(&bridge_names);
306 for (i = 0; i < cfg->n_bridges; i++) {
307 svec_add(&bridge_names, cfg->bridges[i]->name);
309 svec_sort(&bridge_names);
311 svec_init(&dpif_names);
312 svec_init(&dpif_types);
313 dp_enumerate_types(&dpif_types);
314 for (i = 0; i < dpif_types.n; i++) {
319 dp_enumerate_names(dpif_types.names[i], &dpif_names);
321 for (j = 0; j < dpif_names.n; j++) {
322 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
324 struct svec all_names;
327 svec_init(&all_names);
328 dpif_get_all_names(dpif, &all_names);
329 for (k = 0; k < all_names.n; k++) {
330 if (svec_contains(&bridge_names, all_names.names[k])) {
336 svec_destroy(&all_names);
341 svec_destroy(&bridge_names);
342 svec_destroy(&dpif_names);
343 svec_destroy(&dpif_types);
345 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
349 bridge_reconfigure(cfg);
354 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
356 /* XXX SSL should be configurable on a per-bridge basis. */
358 stream_ssl_set_private_key_file(ssl->private_key);
359 stream_ssl_set_certificate_file(ssl->certificate);
360 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
365 /* Attempt to create the network device 'iface_name' through the netdev
368 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
371 struct shash_node *node;
372 struct shash options;
376 shash_init(&options);
377 for (i = 0; i < iface_cfg->n_options; i++) {
378 shash_add(&options, iface_cfg->key_options[i],
379 xstrdup(iface_cfg->value_options[i]));
383 struct netdev_options netdev_options;
385 memset(&netdev_options, 0, sizeof netdev_options);
386 netdev_options.name = iface_cfg->name;
387 if (!strcmp(iface_cfg->type, "internal")) {
388 /* An "internal" config type maps to a netdev "system" type. */
389 netdev_options.type = "system";
391 netdev_options.type = iface_cfg->type;
393 netdev_options.args = &options;
394 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
395 netdev_options.may_create = true;
396 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
397 netdev_options.may_open = true;
400 error = netdev_open(&netdev_options, &iface->netdev);
403 netdev_get_carrier(iface->netdev, &iface->enabled);
405 } else if (iface->netdev) {
406 const char *netdev_type = netdev_get_type(iface->netdev);
407 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
408 ? iface_cfg->type : NULL;
410 /* An "internal" config type maps to a netdev "system" type. */
411 if (iface_type && !strcmp(iface_type, "internal")) {
412 iface_type = "system";
415 if (!iface_type || !strcmp(netdev_type, iface_type)) {
416 error = netdev_reconfigure(iface->netdev, &options);
418 VLOG_WARN("%s: attempting change device type from %s to %s",
419 iface_cfg->name, netdev_type, iface_type);
424 SHASH_FOR_EACH (node, &options) {
427 shash_destroy(&options);
433 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
435 return set_up_iface(iface_cfg, iface, false);
439 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
440 void *aux OVS_UNUSED)
442 if (!iface->netdev) {
443 int error = set_up_iface(iface->cfg, iface, true);
445 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
455 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
456 void *aux OVS_UNUSED)
458 if (iface->dp_ifidx >= 0) {
459 VLOG_DBG("%s has interface %s on port %d",
461 iface->name, iface->dp_ifidx);
464 VLOG_ERR("%s interface not in %s, dropping",
465 iface->name, dpif_name(br->dpif));
471 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
472 void *aux OVS_UNUSED)
474 /* Set policing attributes. */
475 netdev_set_policing(iface->netdev,
476 iface->cfg->ingress_policing_rate,
477 iface->cfg->ingress_policing_burst);
479 /* Set MAC address of internal interfaces other than the local
481 if (iface->dp_ifidx != ODPP_LOCAL
482 && iface_is_internal(br, iface->name)) {
483 iface_set_mac(iface);
489 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
490 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
491 * deletes from 'br' any ports that no longer have any interfaces. */
493 iterate_and_prune_ifaces(struct bridge *br,
494 bool (*cb)(struct bridge *, struct iface *,
500 for (i = 0; i < br->n_ports; ) {
501 struct port *port = br->ports[i];
502 for (j = 0; j < port->n_ifaces; ) {
503 struct iface *iface = port->ifaces[j];
504 if (cb(br, iface, aux)) {
507 iface_destroy(iface);
511 if (port->n_ifaces) {
514 VLOG_ERR("%s port has no interfaces, dropping", port->name);
521 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
523 struct ovsdb_idl_txn *txn;
524 struct shash old_br, new_br;
525 struct shash_node *node;
526 struct bridge *br, *next;
528 int sflow_bridge_number;
530 COVERAGE_INC(bridge_reconfigure);
532 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
534 /* Collect old and new bridges. */
537 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
538 shash_add(&old_br, br->name, br);
540 for (i = 0; i < ovs_cfg->n_bridges; i++) {
541 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
542 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
543 VLOG_WARN("more than one bridge named %s", br_cfg->name);
547 /* Get rid of deleted bridges and add new bridges. */
548 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
549 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
556 SHASH_FOR_EACH (node, &new_br) {
557 const char *br_name = node->name;
558 const struct ovsrec_bridge *br_cfg = node->data;
559 br = shash_find_data(&old_br, br_name);
561 /* If the bridge datapath type has changed, we need to tear it
562 * down and recreate. */
563 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
565 bridge_create(br_cfg);
568 bridge_create(br_cfg);
571 shash_destroy(&old_br);
572 shash_destroy(&new_br);
576 bridge_configure_ssl(ovs_cfg->ssl);
579 /* Reconfigure all bridges. */
580 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
581 bridge_reconfigure_one(ovs_cfg, br);
584 /* Add and delete ports on all datapaths.
586 * The kernel will reject any attempt to add a given port to a datapath if
587 * that port already belongs to a different datapath, so we must do all
588 * port deletions before any port additions. */
589 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
590 struct odp_port *dpif_ports;
592 struct shash want_ifaces;
594 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
595 bridge_get_all_ifaces(br, &want_ifaces);
596 for (i = 0; i < n_dpif_ports; i++) {
597 const struct odp_port *p = &dpif_ports[i];
598 if (!shash_find(&want_ifaces, p->devname)
599 && strcmp(p->devname, br->name)) {
600 int retval = dpif_port_del(br->dpif, p->port);
602 VLOG_ERR("failed to remove %s interface from %s: %s",
603 p->devname, dpif_name(br->dpif),
608 shash_destroy(&want_ifaces);
611 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
612 struct odp_port *dpif_ports;
614 struct shash cur_ifaces, want_ifaces;
615 struct shash_node *node;
617 /* Get the set of interfaces currently in this datapath. */
618 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
619 shash_init(&cur_ifaces);
620 for (i = 0; i < n_dpif_ports; i++) {
621 const char *name = dpif_ports[i].devname;
622 if (!shash_find(&cur_ifaces, name)) {
623 shash_add(&cur_ifaces, name, NULL);
628 /* Get the set of interfaces we want on this datapath. */
629 bridge_get_all_ifaces(br, &want_ifaces);
631 SHASH_FOR_EACH (node, &want_ifaces) {
632 const char *if_name = node->name;
633 struct iface *iface = node->data;
635 if (shash_find(&cur_ifaces, if_name)) {
636 /* Already exists, just reconfigure it. */
638 reconfigure_iface(iface->cfg, iface);
641 /* Need to add to datapath. */
645 /* Add to datapath. */
646 internal = iface_is_internal(br, if_name);
647 error = dpif_port_add(br->dpif, if_name,
648 internal ? ODP_PORT_INTERNAL : 0, NULL);
649 if (error == EFBIG) {
650 VLOG_ERR("ran out of valid port numbers on %s",
651 dpif_name(br->dpif));
654 VLOG_ERR("failed to add %s interface to %s: %s",
655 if_name, dpif_name(br->dpif), strerror(error));
659 shash_destroy(&cur_ifaces);
660 shash_destroy(&want_ifaces);
662 sflow_bridge_number = 0;
663 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
666 struct iface *local_iface;
667 struct iface *hw_addr_iface;
670 bridge_fetch_dp_ifaces(br);
672 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
673 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
675 /* Pick local port hardware address, datapath ID. */
676 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
677 local_iface = bridge_get_local_iface(br);
679 int error = netdev_set_etheraddr(local_iface->netdev, ea);
681 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
682 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
683 "Ethernet address: %s",
684 br->name, strerror(error));
688 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
689 ofproto_set_datapath_id(br->ofproto, dpid);
691 dpid_string = xasprintf("%012"PRIx64, dpid);
692 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
695 /* Set NetFlow configuration on this bridge. */
696 if (br->cfg->netflow) {
697 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
698 struct netflow_options opts;
700 memset(&opts, 0, sizeof opts);
702 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
703 if (nf_cfg->engine_type) {
704 opts.engine_type = *nf_cfg->engine_type;
706 if (nf_cfg->engine_id) {
707 opts.engine_id = *nf_cfg->engine_id;
710 opts.active_timeout = nf_cfg->active_timeout;
711 if (!opts.active_timeout) {
712 opts.active_timeout = -1;
713 } else if (opts.active_timeout < 0) {
714 VLOG_WARN("bridge %s: active timeout interval set to negative "
715 "value, using default instead (%d seconds)", br->name,
716 NF_ACTIVE_TIMEOUT_DEFAULT);
717 opts.active_timeout = -1;
720 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
721 if (opts.add_id_to_iface) {
722 if (opts.engine_id > 0x7f) {
723 VLOG_WARN("bridge %s: netflow port mangling may conflict "
724 "with another vswitch, choose an engine id less "
725 "than 128", br->name);
727 if (br->n_ports > 508) {
728 VLOG_WARN("bridge %s: netflow port mangling will conflict "
729 "with another port when more than 508 ports are "
734 opts.collectors.n = nf_cfg->n_targets;
735 opts.collectors.names = nf_cfg->targets;
736 if (ofproto_set_netflow(br->ofproto, &opts)) {
737 VLOG_ERR("bridge %s: problem setting netflow collectors",
741 ofproto_set_netflow(br->ofproto, NULL);
744 /* Set sFlow configuration on this bridge. */
745 if (br->cfg->sflow) {
746 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
747 const struct ovsrec_controller *ctrl;
748 struct ofproto_sflow_options oso;
750 memset(&oso, 0, sizeof oso);
752 oso.targets.n = sflow_cfg->n_targets;
753 oso.targets.names = sflow_cfg->targets;
755 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
756 if (sflow_cfg->sampling) {
757 oso.sampling_rate = *sflow_cfg->sampling;
760 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
761 if (sflow_cfg->polling) {
762 oso.polling_interval = *sflow_cfg->polling;
765 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
766 if (sflow_cfg->header) {
767 oso.header_len = *sflow_cfg->header;
770 oso.sub_id = sflow_bridge_number++;
771 oso.agent_device = sflow_cfg->agent;
773 ctrl = bridge_get_controller(ovs_cfg, br);
774 oso.control_ip = ctrl ? ctrl->local_ip : NULL;
775 ofproto_set_sflow(br->ofproto, &oso);
777 svec_destroy(&oso.targets);
779 ofproto_set_sflow(br->ofproto, NULL);
782 /* Update the controller and related settings. It would be more
783 * straightforward to call this from bridge_reconfigure_one(), but we
784 * can't do it there for two reasons. First, and most importantly, at
785 * that point we don't know the dp_ifidx of any interfaces that have
786 * been added to the bridge (because we haven't actually added them to
787 * the datapath). Second, at that point we haven't set the datapath ID
788 * yet; when a controller is configured, resetting the datapath ID will
789 * immediately disconnect from the controller, so it's better to set
790 * the datapath ID before the controller. */
791 bridge_reconfigure_controller(ovs_cfg, br);
793 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
794 for (i = 0; i < br->n_ports; i++) {
795 struct port *port = br->ports[i];
797 port_update_vlan_compat(port);
798 port_update_bonding(port);
801 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
802 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
805 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
807 ovsdb_idl_txn_commit(txn);
808 ovsdb_idl_txn_destroy(txn); /* XXX */
812 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
816 for (i = 0; i < n; i++) {
817 if (!strcmp(keys[i], key)) {
825 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
827 return get_ovsrec_key_value(key,
828 br_cfg->key_other_config,
829 br_cfg->value_other_config,
830 br_cfg->n_other_config);
834 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
835 struct iface **hw_addr_iface)
841 *hw_addr_iface = NULL;
843 /* Did the user request a particular MAC? */
844 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
845 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
846 if (eth_addr_is_multicast(ea)) {
847 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
848 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
849 } else if (eth_addr_is_zero(ea)) {
850 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
856 /* Otherwise choose the minimum non-local MAC address among all of the
858 memset(ea, 0xff, sizeof ea);
859 for (i = 0; i < br->n_ports; i++) {
860 struct port *port = br->ports[i];
861 uint8_t iface_ea[ETH_ADDR_LEN];
864 /* Mirror output ports don't participate. */
865 if (port->is_mirror_output_port) {
869 /* Choose the MAC address to represent the port. */
870 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
871 /* Find the interface with this Ethernet address (if any) so that
872 * we can provide the correct devname to the caller. */
874 for (j = 0; j < port->n_ifaces; j++) {
875 struct iface *candidate = port->ifaces[j];
876 uint8_t candidate_ea[ETH_ADDR_LEN];
877 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
878 && eth_addr_equals(iface_ea, candidate_ea)) {
883 /* Choose the interface whose MAC address will represent the port.
884 * The Linux kernel bonding code always chooses the MAC address of
885 * the first slave added to a bond, and the Fedora networking
886 * scripts always add slaves to a bond in alphabetical order, so
887 * for compatibility we choose the interface with the name that is
888 * first in alphabetical order. */
889 iface = port->ifaces[0];
890 for (j = 1; j < port->n_ifaces; j++) {
891 struct iface *candidate = port->ifaces[j];
892 if (strcmp(candidate->name, iface->name) < 0) {
897 /* The local port doesn't count (since we're trying to choose its
898 * MAC address anyway). */
899 if (iface->dp_ifidx == ODPP_LOCAL) {
904 error = netdev_get_etheraddr(iface->netdev, iface_ea);
906 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
907 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
908 iface->name, strerror(error));
913 /* Compare against our current choice. */
914 if (!eth_addr_is_multicast(iface_ea) &&
915 !eth_addr_is_local(iface_ea) &&
916 !eth_addr_is_reserved(iface_ea) &&
917 !eth_addr_is_zero(iface_ea) &&
918 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
920 memcpy(ea, iface_ea, ETH_ADDR_LEN);
921 *hw_addr_iface = iface;
924 if (eth_addr_is_multicast(ea)) {
925 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
926 *hw_addr_iface = NULL;
927 VLOG_WARN("bridge %s: using default bridge Ethernet "
928 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
930 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
931 br->name, ETH_ADDR_ARGS(ea));
935 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
936 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
937 * an interface on 'br', then that interface must be passed in as
938 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
939 * 'hw_addr_iface' must be passed in as a null pointer. */
941 bridge_pick_datapath_id(struct bridge *br,
942 const uint8_t bridge_ea[ETH_ADDR_LEN],
943 struct iface *hw_addr_iface)
946 * The procedure for choosing a bridge MAC address will, in the most
947 * ordinary case, also choose a unique MAC that we can use as a datapath
948 * ID. In some special cases, though, multiple bridges will end up with
949 * the same MAC address. This is OK for the bridges, but it will confuse
950 * the OpenFlow controller, because each datapath needs a unique datapath
953 * Datapath IDs must be unique. It is also very desirable that they be
954 * stable from one run to the next, so that policy set on a datapath
957 const char *datapath_id;
960 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
961 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
967 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
969 * A bridge whose MAC address is taken from a VLAN network device
970 * (that is, a network device created with vconfig(8) or similar
971 * tool) will have the same MAC address as a bridge on the VLAN
972 * device's physical network device.
974 * Handle this case by hashing the physical network device MAC
975 * along with the VLAN identifier.
977 uint8_t buf[ETH_ADDR_LEN + 2];
978 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
979 buf[ETH_ADDR_LEN] = vlan >> 8;
980 buf[ETH_ADDR_LEN + 1] = vlan;
981 return dpid_from_hash(buf, sizeof buf);
984 * Assume that this bridge's MAC address is unique, since it
985 * doesn't fit any of the cases we handle specially.
990 * A purely internal bridge, that is, one that has no non-virtual
991 * network devices on it at all, is more difficult because it has no
992 * natural unique identifier at all.
994 * When the host is a XenServer, we handle this case by hashing the
995 * host's UUID with the name of the bridge. Names of bridges are
996 * persistent across XenServer reboots, although they can be reused if
997 * an internal network is destroyed and then a new one is later
998 * created, so this is fairly effective.
1000 * When the host is not a XenServer, we punt by using a random MAC
1001 * address on each run.
1003 const char *host_uuid = xenserver_get_host_uuid();
1005 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1006 dpid = dpid_from_hash(combined, strlen(combined));
1012 return eth_addr_to_uint64(bridge_ea);
1016 dpid_from_hash(const void *data, size_t n)
1018 uint8_t hash[SHA1_DIGEST_SIZE];
1020 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1021 sha1_bytes(data, n, hash);
1022 eth_addr_mark_random(hash);
1023 return eth_addr_to_uint64(hash);
1029 struct bridge *br, *next;
1033 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1034 int error = bridge_run_one(br);
1036 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1037 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1038 "forcing reconfiguration", br->name);
1052 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1053 ofproto_wait(br->ofproto);
1054 if (br->controller) {
1058 mac_learning_wait(br->ml);
1063 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1064 * configuration changes. */
1066 bridge_flush(struct bridge *br)
1068 COVERAGE_INC(bridge_flush);
1070 mac_learning_flush(br->ml);
1073 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1074 * such interface. */
1075 static struct iface *
1076 bridge_get_local_iface(struct bridge *br)
1080 for (i = 0; i < br->n_ports; i++) {
1081 struct port *port = br->ports[i];
1082 for (j = 0; j < port->n_ifaces; j++) {
1083 struct iface *iface = port->ifaces[j];
1084 if (iface->dp_ifidx == ODPP_LOCAL) {
1093 /* Bridge unixctl user interface functions. */
1095 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1096 const char *args, void *aux OVS_UNUSED)
1098 struct ds ds = DS_EMPTY_INITIALIZER;
1099 const struct bridge *br;
1100 const struct mac_entry *e;
1102 br = bridge_lookup(args);
1104 unixctl_command_reply(conn, 501, "no such bridge");
1108 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1109 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1110 if (e->port < 0 || e->port >= br->n_ports) {
1113 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1114 br->ports[e->port]->ifaces[0]->dp_ifidx,
1115 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1117 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1121 /* Bridge reconfiguration functions. */
1122 static struct bridge *
1123 bridge_create(const struct ovsrec_bridge *br_cfg)
1128 assert(!bridge_lookup(br_cfg->name));
1129 br = xzalloc(sizeof *br);
1131 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1137 dpif_flow_flush(br->dpif);
1139 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1142 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1144 dpif_delete(br->dpif);
1145 dpif_close(br->dpif);
1150 br->name = xstrdup(br_cfg->name);
1152 br->ml = mac_learning_create();
1153 br->sent_config_request = false;
1154 eth_addr_nicira_random(br->default_ea);
1156 port_array_init(&br->ifaces);
1160 list_push_back(&all_bridges, &br->node);
1162 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1168 bridge_destroy(struct bridge *br)
1173 while (br->n_ports > 0) {
1174 port_destroy(br->ports[br->n_ports - 1]);
1176 list_remove(&br->node);
1177 error = dpif_delete(br->dpif);
1178 if (error && error != ENOENT) {
1179 VLOG_ERR("failed to delete %s: %s",
1180 dpif_name(br->dpif), strerror(error));
1182 dpif_close(br->dpif);
1183 ofproto_destroy(br->ofproto);
1184 free(br->controller);
1185 mac_learning_destroy(br->ml);
1186 port_array_destroy(&br->ifaces);
1193 static struct bridge *
1194 bridge_lookup(const char *name)
1198 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1199 if (!strcmp(br->name, name)) {
1207 bridge_exists(const char *name)
1209 return bridge_lookup(name) ? true : false;
1213 bridge_get_datapathid(const char *name)
1215 struct bridge *br = bridge_lookup(name);
1216 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1219 /* Handle requests for a listing of all flows known by the OpenFlow
1220 * stack, including those normally hidden. */
1222 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1223 const char *args, void *aux OVS_UNUSED)
1228 br = bridge_lookup(args);
1230 unixctl_command_reply(conn, 501, "Unknown bridge");
1235 ofproto_get_all_flows(br->ofproto, &results);
1237 unixctl_command_reply(conn, 200, ds_cstr(&results));
1238 ds_destroy(&results);
1242 bridge_run_one(struct bridge *br)
1246 error = ofproto_run1(br->ofproto);
1251 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1254 error = ofproto_run2(br->ofproto, br->flush);
1260 static const struct ovsrec_controller *
1261 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1262 const struct bridge *br)
1264 const struct ovsrec_controller *controller;
1266 controller = (br->cfg->controller ? br->cfg->controller
1267 : ovs_cfg->controller ? ovs_cfg->controller
1270 if (controller && !strcmp(controller->target, "none")) {
1278 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1280 struct svec *ifaces = ifaces_;
1281 if (!svec_contains(ifaces, iface->name)) {
1282 svec_add(ifaces, iface->name);
1286 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1288 br->name, iface->name, iface->port->name);
1294 bridge_update_desc(struct bridge *br)
1297 bool changed = false;
1300 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1301 if (desc != br->mfr_desc) {
1304 br->mfr_desc = xstrdup(desc);
1306 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1311 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1312 if (desc != br->hw_desc) {
1315 br->hw_desc = xstrdup(desc);
1317 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1322 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1323 if (desc != br->sw_desc) {
1326 br->sw_desc = xstrdup(desc);
1328 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1333 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1334 if (desc != br->serial_desc) {
1335 free(br->serial_desc);
1337 br->serial_desc = xstrdup(desc);
1339 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1344 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1345 if (desc != br->dp_desc) {
1348 br->dp_desc = xstrdup(desc);
1350 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1356 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1357 br->sw_desc, br->serial_desc, br->dp_desc);
1363 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1366 struct shash old_ports, new_ports;
1368 struct svec listeners, old_listeners;
1369 struct svec snoops, old_snoops;
1370 struct shash_node *node;
1373 /* Collect old ports. */
1374 shash_init(&old_ports);
1375 for (i = 0; i < br->n_ports; i++) {
1376 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1379 /* Collect new ports. */
1380 shash_init(&new_ports);
1381 for (i = 0; i < br->cfg->n_ports; i++) {
1382 const char *name = br->cfg->ports[i]->name;
1383 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1384 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1389 /* If we have a controller, then we need a local port. Complain if the
1390 * user didn't specify one.
1392 * XXX perhaps we should synthesize a port ourselves in this case. */
1393 if (bridge_get_controller(ovs_cfg, br)) {
1394 char local_name[IF_NAMESIZE];
1397 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1398 local_name, sizeof local_name);
1399 if (!error && !shash_find(&new_ports, local_name)) {
1400 VLOG_WARN("bridge %s: controller specified but no local port "
1401 "(port named %s) defined",
1402 br->name, local_name);
1406 /* Get rid of deleted ports and add new ports. */
1407 SHASH_FOR_EACH (node, &old_ports) {
1408 if (!shash_find(&new_ports, node->name)) {
1409 port_destroy(node->data);
1412 SHASH_FOR_EACH (node, &new_ports) {
1413 struct port *port = shash_find_data(&old_ports, node->name);
1415 port = port_create(br, node->name);
1417 port_reconfigure(port, node->data);
1419 shash_destroy(&old_ports);
1420 shash_destroy(&new_ports);
1422 /* Check and delete duplicate interfaces. */
1424 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1425 svec_destroy(&ifaces);
1427 /* Delete all flows if we're switching from connected to standalone or vice
1428 * versa. (XXX Should we delete all flows if we are switching from one
1429 * controller to another?) */
1432 /* Configure OpenFlow management listeners. */
1433 svec_init(&listeners);
1434 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1436 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1437 ovs_rundir, br->name));
1438 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1439 svec_clear(&listeners);
1441 svec_sort_unique(&listeners);
1443 svec_init(&old_listeners);
1444 ofproto_get_listeners(br->ofproto, &old_listeners);
1445 svec_sort_unique(&old_listeners);
1447 if (!svec_equal(&listeners, &old_listeners)) {
1448 ofproto_set_listeners(br->ofproto, &listeners);
1450 svec_destroy(&listeners);
1451 svec_destroy(&old_listeners);
1453 /* Configure OpenFlow controller connection snooping. */
1455 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1457 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1458 ovs_rundir, br->name));
1459 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1460 svec_clear(&snoops);
1462 svec_sort_unique(&snoops);
1464 svec_init(&old_snoops);
1465 ofproto_get_snoops(br->ofproto, &old_snoops);
1466 svec_sort_unique(&old_snoops);
1468 if (!svec_equal(&snoops, &old_snoops)) {
1469 ofproto_set_snoops(br->ofproto, &snoops);
1471 svec_destroy(&snoops);
1472 svec_destroy(&old_snoops);
1474 /* Default listener. */
1475 svec_init(&listeners);
1476 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1477 ovs_rundir, br->name));
1478 svec_init(&old_listeners);
1479 ofproto_get_listeners(br->ofproto, &old_listeners);
1480 if (!svec_equal(&listeners, &old_listeners)) {
1481 ofproto_set_listeners(br->ofproto, &listeners);
1483 svec_destroy(&listeners);
1484 svec_destroy(&old_listeners);
1486 /* Default snoop. */
1488 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1489 ovs_rundir, br->name));
1490 svec_init(&old_snoops);
1491 ofproto_get_snoops(br->ofproto, &old_snoops);
1492 if (!svec_equal(&snoops, &old_snoops)) {
1493 ofproto_set_snoops(br->ofproto, &snoops);
1495 svec_destroy(&snoops);
1496 svec_destroy(&old_snoops);
1499 mirror_reconfigure(br);
1501 bridge_update_desc(br);
1505 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1508 const struct ovsrec_controller *c;
1510 c = bridge_get_controller(ovs_cfg, br);
1511 if ((br->controller != NULL) != (c != NULL)) {
1512 ofproto_flush_flows(br->ofproto);
1514 free(br->controller);
1515 br->controller = c ? xstrdup(c->target) : NULL;
1518 int max_backoff, probe;
1519 int rate_limit, burst_limit;
1521 if (!strcmp(c->target, "discover")) {
1522 ofproto_set_discovery(br->ofproto, true,
1523 c->discover_accept_regex,
1524 c->discover_update_resolv_conf);
1526 struct iface *local_iface;
1530 in_band = (!c->connection_mode
1531 || !strcmp(c->connection_mode, "out-of-band"));
1532 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1533 ofproto_set_in_band(br->ofproto, in_band);
1535 local_iface = bridge_get_local_iface(br);
1536 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1537 struct netdev *netdev = local_iface->netdev;
1538 struct in_addr mask, gateway;
1540 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1543 if (!c->local_gateway
1544 || !inet_aton(c->local_gateway, &gateway)) {
1548 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1550 mask.s_addr = guess_netmask(ip.s_addr);
1552 if (!netdev_set_in4(netdev, ip, mask)) {
1553 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1555 br->name, IP_ARGS(&ip.s_addr),
1556 IP_ARGS(&mask.s_addr));
1559 if (gateway.s_addr) {
1560 if (!netdev_add_router(netdev, gateway)) {
1561 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1562 br->name, IP_ARGS(&gateway.s_addr));
1568 ofproto_set_failure(br->ofproto,
1570 || !strcmp(c->fail_mode, "standalone")
1571 || !strcmp(c->fail_mode, "open")));
1573 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1574 ofproto_set_probe_interval(br->ofproto, probe);
1576 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1577 ofproto_set_max_backoff(br->ofproto, max_backoff);
1579 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1580 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1581 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1583 union ofp_action action;
1586 /* Set up a flow that matches every packet and directs them to
1587 * OFPP_NORMAL (which goes to us). */
1588 memset(&action, 0, sizeof action);
1589 action.type = htons(OFPAT_OUTPUT);
1590 action.output.len = htons(sizeof action);
1591 action.output.port = htons(OFPP_NORMAL);
1592 memset(&flow, 0, sizeof flow);
1593 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1596 ofproto_set_in_band(br->ofproto, false);
1597 ofproto_set_max_backoff(br->ofproto, 1);
1598 ofproto_set_probe_interval(br->ofproto, 5);
1599 ofproto_set_failure(br->ofproto, false);
1602 ofproto_set_controller(br->ofproto, br->controller);
1606 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1611 for (i = 0; i < br->n_ports; i++) {
1612 struct port *port = br->ports[i];
1613 for (j = 0; j < port->n_ifaces; j++) {
1614 struct iface *iface = port->ifaces[j];
1615 shash_add_once(ifaces, iface->name, iface);
1617 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1618 shash_add_once(ifaces, port->name, NULL);
1623 /* For robustness, in case the administrator moves around datapath ports behind
1624 * our back, we re-check all the datapath port numbers here.
1626 * This function will set the 'dp_ifidx' members of interfaces that have
1627 * disappeared to -1, so only call this function from a context where those
1628 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1629 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1630 * datapath, which doesn't support UINT16_MAX+1 ports. */
1632 bridge_fetch_dp_ifaces(struct bridge *br)
1634 struct odp_port *dpif_ports;
1635 size_t n_dpif_ports;
1638 /* Reset all interface numbers. */
1639 for (i = 0; i < br->n_ports; i++) {
1640 struct port *port = br->ports[i];
1641 for (j = 0; j < port->n_ifaces; j++) {
1642 struct iface *iface = port->ifaces[j];
1643 iface->dp_ifidx = -1;
1646 port_array_clear(&br->ifaces);
1648 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1649 for (i = 0; i < n_dpif_ports; i++) {
1650 struct odp_port *p = &dpif_ports[i];
1651 struct iface *iface = iface_lookup(br, p->devname);
1653 if (iface->dp_ifidx >= 0) {
1654 VLOG_WARN("%s reported interface %s twice",
1655 dpif_name(br->dpif), p->devname);
1656 } else if (iface_from_dp_ifidx(br, p->port)) {
1657 VLOG_WARN("%s reported interface %"PRIu16" twice",
1658 dpif_name(br->dpif), p->port);
1660 port_array_set(&br->ifaces, p->port, iface);
1661 iface->dp_ifidx = p->port;
1665 int64_t ofport = (iface->dp_ifidx >= 0
1666 ? odp_port_to_ofp_port(iface->dp_ifidx)
1668 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1675 /* Bridge packet processing functions. */
1678 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1680 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1683 static struct bond_entry *
1684 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1686 return &port->bond_hash[bond_hash(mac)];
1690 bond_choose_iface(const struct port *port)
1692 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1693 size_t i, best_down_slave = -1;
1694 long long next_delay_expiration = LLONG_MAX;
1696 for (i = 0; i < port->n_ifaces; i++) {
1697 struct iface *iface = port->ifaces[i];
1699 if (iface->enabled) {
1701 } else if (iface->delay_expires < next_delay_expiration) {
1702 best_down_slave = i;
1703 next_delay_expiration = iface->delay_expires;
1707 if (best_down_slave != -1) {
1708 struct iface *iface = port->ifaces[best_down_slave];
1710 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1711 "since no other interface is up", iface->name,
1712 iface->delay_expires - time_msec());
1713 bond_enable_slave(iface, true);
1716 return best_down_slave;
1720 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1721 uint16_t *dp_ifidx, tag_type *tags)
1723 struct iface *iface;
1725 assert(port->n_ifaces);
1726 if (port->n_ifaces == 1) {
1727 iface = port->ifaces[0];
1729 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1730 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1731 || !port->ifaces[e->iface_idx]->enabled) {
1732 /* XXX select interface properly. The current interface selection
1733 * is only good for testing the rebalancing code. */
1734 e->iface_idx = bond_choose_iface(port);
1735 if (e->iface_idx < 0) {
1736 *tags |= port->no_ifaces_tag;
1739 e->iface_tag = tag_create_random();
1740 ((struct port *) port)->bond_compat_is_stale = true;
1742 *tags |= e->iface_tag;
1743 iface = port->ifaces[e->iface_idx];
1745 *dp_ifidx = iface->dp_ifidx;
1746 *tags |= iface->tag; /* Currently only used for bonding. */
1751 bond_link_status_update(struct iface *iface, bool carrier)
1753 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1754 struct port *port = iface->port;
1756 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1757 /* Nothing to do. */
1760 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1761 iface->name, carrier ? "detected" : "dropped");
1762 if (carrier == iface->enabled) {
1763 iface->delay_expires = LLONG_MAX;
1764 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1765 iface->name, carrier ? "disabled" : "enabled");
1766 } else if (carrier && port->active_iface < 0) {
1767 bond_enable_slave(iface, true);
1768 if (port->updelay) {
1769 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1770 "other interface is up", iface->name, port->updelay);
1773 int delay = carrier ? port->updelay : port->downdelay;
1774 iface->delay_expires = time_msec() + delay;
1777 "interface %s: will be %s if it stays %s for %d ms",
1779 carrier ? "enabled" : "disabled",
1780 carrier ? "up" : "down",
1787 bond_choose_active_iface(struct port *port)
1789 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1791 port->active_iface = bond_choose_iface(port);
1792 port->active_iface_tag = tag_create_random();
1793 if (port->active_iface >= 0) {
1794 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1795 port->name, port->ifaces[port->active_iface]->name);
1797 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1803 bond_enable_slave(struct iface *iface, bool enable)
1805 struct port *port = iface->port;
1806 struct bridge *br = port->bridge;
1808 /* This acts as a recursion check. If the act of disabling a slave
1809 * causes a different slave to be enabled, the flag will allow us to
1810 * skip redundant work when we reenter this function. It must be
1811 * cleared on exit to keep things safe with multiple bonds. */
1812 static bool moving_active_iface = false;
1814 iface->delay_expires = LLONG_MAX;
1815 if (enable == iface->enabled) {
1819 iface->enabled = enable;
1820 if (!iface->enabled) {
1821 VLOG_WARN("interface %s: disabled", iface->name);
1822 ofproto_revalidate(br->ofproto, iface->tag);
1823 if (iface->port_ifidx == port->active_iface) {
1824 ofproto_revalidate(br->ofproto,
1825 port->active_iface_tag);
1827 /* Disabling a slave can lead to another slave being immediately
1828 * enabled if there will be no active slaves but one is waiting
1829 * on an updelay. In this case we do not need to run most of the
1830 * code for the newly enabled slave since there was no period
1831 * without an active slave and it is redundant with the disabling
1833 moving_active_iface = true;
1834 bond_choose_active_iface(port);
1836 bond_send_learning_packets(port);
1838 VLOG_WARN("interface %s: enabled", iface->name);
1839 if (port->active_iface < 0 && !moving_active_iface) {
1840 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1841 bond_choose_active_iface(port);
1842 bond_send_learning_packets(port);
1844 iface->tag = tag_create_random();
1847 moving_active_iface = false;
1848 port->bond_compat_is_stale = true;
1852 bond_run(struct bridge *br)
1856 for (i = 0; i < br->n_ports; i++) {
1857 struct port *port = br->ports[i];
1859 if (port->n_ifaces >= 2) {
1860 for (j = 0; j < port->n_ifaces; j++) {
1861 struct iface *iface = port->ifaces[j];
1862 if (time_msec() >= iface->delay_expires) {
1863 bond_enable_slave(iface, !iface->enabled);
1868 if (port->bond_compat_is_stale) {
1869 port->bond_compat_is_stale = false;
1870 port_update_bond_compat(port);
1876 bond_wait(struct bridge *br)
1880 for (i = 0; i < br->n_ports; i++) {
1881 struct port *port = br->ports[i];
1882 if (port->n_ifaces < 2) {
1885 for (j = 0; j < port->n_ifaces; j++) {
1886 struct iface *iface = port->ifaces[j];
1887 if (iface->delay_expires != LLONG_MAX) {
1888 poll_timer_wait(iface->delay_expires - time_msec());
1895 set_dst(struct dst *p, const flow_t *flow,
1896 const struct port *in_port, const struct port *out_port,
1899 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1900 : in_port->vlan >= 0 ? in_port->vlan
1901 : ntohs(flow->dl_vlan));
1902 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1906 swap_dst(struct dst *p, struct dst *q)
1908 struct dst tmp = *p;
1913 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1914 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1915 * that we push to the datapath. We could in fact fully sort the array by
1916 * vlan, but in most cases there are at most two different vlan tags so that's
1917 * possibly overkill.) */
1919 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1921 struct dst *first = dsts;
1922 struct dst *last = dsts + n_dsts;
1924 while (first != last) {
1926 * - All dsts < first have vlan == 'vlan'.
1927 * - All dsts >= last have vlan != 'vlan'.
1928 * - first < last. */
1929 while (first->vlan == vlan) {
1930 if (++first == last) {
1935 /* Same invariants, plus one additional:
1936 * - first->vlan != vlan.
1938 while (last[-1].vlan != vlan) {
1939 if (--last == first) {
1944 /* Same invariants, plus one additional:
1945 * - last[-1].vlan == vlan.*/
1946 swap_dst(first++, --last);
1951 mirror_mask_ffs(mirror_mask_t mask)
1953 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1958 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1959 const struct dst *test)
1962 for (i = 0; i < n_dsts; i++) {
1963 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1971 port_trunks_vlan(const struct port *port, uint16_t vlan)
1973 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1977 port_includes_vlan(const struct port *port, uint16_t vlan)
1979 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1983 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1984 const struct port *in_port, const struct port *out_port,
1985 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1987 mirror_mask_t mirrors = in_port->src_mirrors;
1988 struct dst *dst = dsts;
1991 if (out_port == FLOOD_PORT) {
1992 /* XXX use ODP_FLOOD if no vlans or bonding. */
1993 /* XXX even better, define each VLAN as a datapath port group */
1994 for (i = 0; i < br->n_ports; i++) {
1995 struct port *port = br->ports[i];
1996 if (port != in_port && port_includes_vlan(port, vlan)
1997 && !port->is_mirror_output_port
1998 && set_dst(dst, flow, in_port, port, tags)) {
1999 mirrors |= port->dst_mirrors;
2003 *nf_output_iface = NF_OUT_FLOOD;
2004 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2005 *nf_output_iface = dst->dp_ifidx;
2006 mirrors |= out_port->dst_mirrors;
2011 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2012 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2014 if (set_dst(dst, flow, in_port, m->out_port, tags)
2015 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2019 for (i = 0; i < br->n_ports; i++) {
2020 struct port *port = br->ports[i];
2021 if (port_includes_vlan(port, m->out_vlan)
2022 && set_dst(dst, flow, in_port, port, tags))
2026 if (port->vlan < 0) {
2027 dst->vlan = m->out_vlan;
2029 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2033 /* Use the vlan tag on the original flow instead of
2034 * the one passed in the vlan parameter. This ensures
2035 * that we compare the vlan from before any implicit
2036 * tagging tags place. This is necessary because
2037 * dst->vlan is the final vlan, after removing implicit
2039 flow_vlan = ntohs(flow->dl_vlan);
2040 if (flow_vlan == 0) {
2041 flow_vlan = OFP_VLAN_NONE;
2043 if (port == in_port && dst->vlan == flow_vlan) {
2044 /* Don't send out input port on same VLAN. */
2052 mirrors &= mirrors - 1;
2055 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2059 static void OVS_UNUSED
2060 print_dsts(const struct dst *dsts, size_t n)
2062 for (; n--; dsts++) {
2063 printf(">p%"PRIu16, dsts->dp_ifidx);
2064 if (dsts->vlan != OFP_VLAN_NONE) {
2065 printf("v%"PRIu16, dsts->vlan);
2071 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2072 const struct port *in_port, const struct port *out_port,
2073 tag_type *tags, struct odp_actions *actions,
2074 uint16_t *nf_output_iface)
2076 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2078 const struct dst *p;
2081 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2084 cur_vlan = ntohs(flow->dl_vlan);
2085 for (p = dsts; p < &dsts[n_dsts]; p++) {
2086 union odp_action *a;
2087 if (p->vlan != cur_vlan) {
2088 if (p->vlan == OFP_VLAN_NONE) {
2089 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2091 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2092 a->vlan_vid.vlan_vid = htons(p->vlan);
2096 a = odp_actions_add(actions, ODPAT_OUTPUT);
2097 a->output.port = p->dp_ifidx;
2101 /* Returns the effective vlan of a packet, taking into account both the
2102 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2103 * the packet is untagged and -1 indicates it has an invalid header and
2104 * should be dropped. */
2105 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2106 struct port *in_port, bool have_packet)
2108 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2109 * belongs to VLAN 0, so we should treat both cases identically. (In the
2110 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2111 * presumably to allow a priority to be specified. In the latter case, the
2112 * packet does not have any 802.1Q header.) */
2113 int vlan = ntohs(flow->dl_vlan);
2114 if (vlan == OFP_VLAN_NONE) {
2117 if (in_port->vlan >= 0) {
2119 /* XXX support double tagging? */
2121 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2122 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2123 "packet received on port %s configured with "
2124 "implicit VLAN %"PRIu16,
2125 br->name, ntohs(flow->dl_vlan),
2126 in_port->name, in_port->vlan);
2130 vlan = in_port->vlan;
2132 if (!port_includes_vlan(in_port, vlan)) {
2134 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2135 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2136 "packet received on port %s not configured for "
2138 br->name, vlan, in_port->name, vlan);
2148 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2149 struct port *in_port)
2151 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2152 vlan, in_port->port_idx);
2154 /* The log messages here could actually be useful in debugging,
2155 * so keep the rate limit relatively high. */
2156 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2158 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2159 "on port %s in VLAN %d",
2160 br->name, ETH_ADDR_ARGS(flow->dl_src),
2161 in_port->name, vlan);
2162 ofproto_revalidate(br->ofproto, rev_tag);
2167 is_bcast_arp_reply(const flow_t *flow)
2169 return (flow->dl_type == htons(ETH_TYPE_ARP)
2170 && flow->nw_proto == ARP_OP_REPLY
2171 && eth_addr_is_broadcast(flow->dl_dst));
2174 /* If the composed actions may be applied to any packet in the given 'flow',
2175 * returns true. Otherwise, the actions should only be applied to 'packet', or
2176 * not at all, if 'packet' was NULL. */
2178 process_flow(struct bridge *br, const flow_t *flow,
2179 const struct ofpbuf *packet, struct odp_actions *actions,
2180 tag_type *tags, uint16_t *nf_output_iface)
2182 struct iface *in_iface;
2183 struct port *in_port;
2184 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2188 /* Find the interface and port structure for the received packet. */
2189 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2191 /* No interface? Something fishy... */
2192 if (packet != NULL) {
2193 /* Odd. A few possible reasons here:
2195 * - We deleted an interface but there are still a few packets
2196 * queued up from it.
2198 * - Someone externally added an interface (e.g. with "ovs-dpctl
2199 * add-if") that we don't know about.
2201 * - Packet arrived on the local port but the local port is not
2202 * one of our bridge ports.
2204 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2206 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2207 "interface %"PRIu16, br->name, flow->in_port);
2210 /* Return without adding any actions, to drop packets on this flow. */
2213 in_port = in_iface->port;
2214 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2219 /* Drop frames for reserved multicast addresses. */
2220 if (eth_addr_is_reserved(flow->dl_dst)) {
2224 /* Drop frames on ports reserved for mirroring. */
2225 if (in_port->is_mirror_output_port) {
2226 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2227 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2228 "which is reserved exclusively for mirroring",
2229 br->name, in_port->name);
2233 /* Packets received on bonds need special attention to avoid duplicates. */
2234 if (in_port->n_ifaces > 1) {
2237 if (eth_addr_is_multicast(flow->dl_dst)) {
2238 *tags |= in_port->active_iface_tag;
2239 if (in_port->active_iface != in_iface->port_ifidx) {
2240 /* Drop all multicast packets on inactive slaves. */
2245 /* Drop all packets for which we have learned a different input
2246 * port, because we probably sent the packet on one slave and got
2247 * it back on the other. Broadcast ARP replies are an exception
2248 * to this rule: the host has moved to another switch. */
2249 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2250 if (src_idx != -1 && src_idx != in_port->port_idx &&
2251 !is_bcast_arp_reply(flow)) {
2257 out_port = FLOOD_PORT;
2258 /* Learn source MAC (but don't try to learn from revalidation). */
2260 update_learning_table(br, flow, vlan, in_port);
2263 /* Determine output port. */
2264 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2266 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2267 out_port = br->ports[out_port_idx];
2268 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2269 /* If we are revalidating but don't have a learning entry then
2270 * eject the flow. Installing a flow that floods packets opens
2271 * up a window of time where we could learn from a packet reflected
2272 * on a bond and blackhole packets before the learning table is
2273 * updated to reflect the correct port. */
2277 /* Don't send packets out their input ports. */
2278 if (in_port == out_port) {
2283 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2289 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2292 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2293 const struct ofp_phy_port *opp,
2296 struct bridge *br = br_;
2297 struct iface *iface;
2300 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2306 if (reason == OFPPR_DELETE) {
2307 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2308 br->name, iface->name);
2309 iface_destroy(iface);
2310 if (!port->n_ifaces) {
2311 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2312 br->name, port->name);
2318 if (port->n_ifaces > 1) {
2319 bool up = !(opp->state & OFPPS_LINK_DOWN);
2320 bond_link_status_update(iface, up);
2321 port_update_bond_compat(port);
2327 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2328 struct odp_actions *actions, tag_type *tags,
2329 uint16_t *nf_output_iface, void *br_)
2331 struct bridge *br = br_;
2333 COVERAGE_INC(bridge_process_flow);
2334 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2338 bridge_account_flow_ofhook_cb(const flow_t *flow,
2339 const union odp_action *actions,
2340 size_t n_actions, unsigned long long int n_bytes,
2343 struct bridge *br = br_;
2344 struct port *in_port;
2345 const union odp_action *a;
2347 /* Feed information from the active flows back into the learning table
2348 * to ensure that table is always in sync with what is actually flowing
2349 * through the datapath. */
2350 in_port = port_from_dp_ifidx(br, flow->in_port);
2352 int vlan = flow_get_vlan(br, flow, in_port, false);
2354 update_learning_table(br, flow, vlan, in_port);
2358 if (!br->has_bonded_ports) {
2362 for (a = actions; a < &actions[n_actions]; a++) {
2363 if (a->type == ODPAT_OUTPUT) {
2364 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2365 if (out_port && out_port->n_ifaces >= 2) {
2366 struct bond_entry *e = lookup_bond_entry(out_port,
2368 e->tx_bytes += n_bytes;
2375 bridge_account_checkpoint_ofhook_cb(void *br_)
2377 struct bridge *br = br_;
2381 if (!br->has_bonded_ports) {
2386 for (i = 0; i < br->n_ports; i++) {
2387 struct port *port = br->ports[i];
2388 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2389 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2390 bond_rebalance_port(port);
2395 static struct ofhooks bridge_ofhooks = {
2396 bridge_port_changed_ofhook_cb,
2397 bridge_normal_ofhook_cb,
2398 bridge_account_flow_ofhook_cb,
2399 bridge_account_checkpoint_ofhook_cb,
2402 /* Bonding functions. */
2404 /* Statistics for a single interface on a bonded port, used for load-based
2405 * bond rebalancing. */
2406 struct slave_balance {
2407 struct iface *iface; /* The interface. */
2408 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2410 /* All the "bond_entry"s that are assigned to this interface, in order of
2411 * increasing tx_bytes. */
2412 struct bond_entry **hashes;
2416 /* Sorts pointers to pointers to bond_entries in ascending order by the
2417 * interface to which they are assigned, and within a single interface in
2418 * ascending order of bytes transmitted. */
2420 compare_bond_entries(const void *a_, const void *b_)
2422 const struct bond_entry *const *ap = a_;
2423 const struct bond_entry *const *bp = b_;
2424 const struct bond_entry *a = *ap;
2425 const struct bond_entry *b = *bp;
2426 if (a->iface_idx != b->iface_idx) {
2427 return a->iface_idx > b->iface_idx ? 1 : -1;
2428 } else if (a->tx_bytes != b->tx_bytes) {
2429 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2435 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2436 * *descending* order by number of bytes transmitted. */
2438 compare_slave_balance(const void *a_, const void *b_)
2440 const struct slave_balance *a = a_;
2441 const struct slave_balance *b = b_;
2442 if (a->iface->enabled != b->iface->enabled) {
2443 return a->iface->enabled ? -1 : 1;
2444 } else if (a->tx_bytes != b->tx_bytes) {
2445 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2452 swap_bals(struct slave_balance *a, struct slave_balance *b)
2454 struct slave_balance tmp = *a;
2459 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2460 * given that 'p' (and only 'p') might be in the wrong location.
2462 * This function invalidates 'p', since it might now be in a different memory
2465 resort_bals(struct slave_balance *p,
2466 struct slave_balance bals[], size_t n_bals)
2469 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2470 swap_bals(p, p - 1);
2472 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2473 swap_bals(p, p + 1);
2479 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2481 if (VLOG_IS_DBG_ENABLED()) {
2482 struct ds ds = DS_EMPTY_INITIALIZER;
2483 const struct slave_balance *b;
2485 for (b = bals; b < bals + n_bals; b++) {
2489 ds_put_char(&ds, ',');
2491 ds_put_format(&ds, " %s %"PRIu64"kB",
2492 b->iface->name, b->tx_bytes / 1024);
2494 if (!b->iface->enabled) {
2495 ds_put_cstr(&ds, " (disabled)");
2497 if (b->n_hashes > 0) {
2498 ds_put_cstr(&ds, " (");
2499 for (i = 0; i < b->n_hashes; i++) {
2500 const struct bond_entry *e = b->hashes[i];
2502 ds_put_cstr(&ds, " + ");
2504 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2505 e - port->bond_hash, e->tx_bytes / 1024);
2507 ds_put_cstr(&ds, ")");
2510 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2515 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2517 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2520 struct bond_entry *hash = from->hashes[hash_idx];
2521 struct port *port = from->iface->port;
2522 uint64_t delta = hash->tx_bytes;
2524 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2525 "from %s to %s (now carrying %"PRIu64"kB and "
2526 "%"PRIu64"kB load, respectively)",
2527 port->name, delta / 1024, hash - port->bond_hash,
2528 from->iface->name, to->iface->name,
2529 (from->tx_bytes - delta) / 1024,
2530 (to->tx_bytes + delta) / 1024);
2532 /* Delete element from from->hashes.
2534 * We don't bother to add the element to to->hashes because not only would
2535 * it require more work, the only purpose it would be to allow that hash to
2536 * be migrated to another slave in this rebalancing run, and there is no
2537 * point in doing that. */
2538 if (hash_idx == 0) {
2541 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2542 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2546 /* Shift load away from 'from' to 'to'. */
2547 from->tx_bytes -= delta;
2548 to->tx_bytes += delta;
2550 /* Arrange for flows to be revalidated. */
2551 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2552 hash->iface_idx = to->iface->port_ifidx;
2553 hash->iface_tag = tag_create_random();
2557 bond_rebalance_port(struct port *port)
2559 struct slave_balance bals[DP_MAX_PORTS];
2561 struct bond_entry *hashes[BOND_MASK + 1];
2562 struct slave_balance *b, *from, *to;
2563 struct bond_entry *e;
2566 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2567 * descending order of tx_bytes, so that bals[0] represents the most
2568 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2571 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2572 * array for each slave_balance structure, we sort our local array of
2573 * hashes in order by slave, so that all of the hashes for a given slave
2574 * become contiguous in memory, and then we point each 'hashes' members of
2575 * a slave_balance structure to the start of a contiguous group. */
2576 n_bals = port->n_ifaces;
2577 for (b = bals; b < &bals[n_bals]; b++) {
2578 b->iface = port->ifaces[b - bals];
2583 for (i = 0; i <= BOND_MASK; i++) {
2584 hashes[i] = &port->bond_hash[i];
2586 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2587 for (i = 0; i <= BOND_MASK; i++) {
2589 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2590 b = &bals[e->iface_idx];
2591 b->tx_bytes += e->tx_bytes;
2593 b->hashes = &hashes[i];
2598 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2599 log_bals(bals, n_bals, port);
2601 /* Discard slaves that aren't enabled (which were sorted to the back of the
2602 * array earlier). */
2603 while (!bals[n_bals - 1].iface->enabled) {
2610 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2611 to = &bals[n_bals - 1];
2612 for (from = bals; from < to; ) {
2613 uint64_t overload = from->tx_bytes - to->tx_bytes;
2614 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2615 /* The extra load on 'from' (and all less-loaded slaves), compared
2616 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2617 * it is less than ~1Mbps. No point in rebalancing. */
2619 } else if (from->n_hashes == 1) {
2620 /* 'from' only carries a single MAC hash, so we can't shift any
2621 * load away from it, even though we want to. */
2624 /* 'from' is carrying significantly more load than 'to', and that
2625 * load is split across at least two different hashes. Pick a hash
2626 * to migrate to 'to' (the least-loaded slave), given that doing so
2627 * must decrease the ratio of the load on the two slaves by at
2630 * The sort order we use means that we prefer to shift away the
2631 * smallest hashes instead of the biggest ones. There is little
2632 * reason behind this decision; we could use the opposite sort
2633 * order to shift away big hashes ahead of small ones. */
2637 for (i = 0; i < from->n_hashes; i++) {
2638 double old_ratio, new_ratio;
2639 uint64_t delta = from->hashes[i]->tx_bytes;
2641 if (delta == 0 || from->tx_bytes - delta == 0) {
2642 /* Pointless move. */
2646 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2648 if (to->tx_bytes == 0) {
2649 /* Nothing on the new slave, move it. */
2653 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2654 new_ratio = (double)(from->tx_bytes - delta) /
2655 (to->tx_bytes + delta);
2657 if (new_ratio == 0) {
2658 /* Should already be covered but check to prevent division
2663 if (new_ratio < 1) {
2664 new_ratio = 1 / new_ratio;
2667 if (old_ratio - new_ratio > 0.1) {
2668 /* Would decrease the ratio, move it. */
2672 if (i < from->n_hashes) {
2673 bond_shift_load(from, to, i);
2674 port->bond_compat_is_stale = true;
2676 /* If the result of the migration changed the relative order of
2677 * 'from' and 'to' swap them back to maintain invariants. */
2678 if (order_swapped) {
2679 swap_bals(from, to);
2682 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2683 * point to different slave_balance structures. It is only
2684 * valid to do these two operations in a row at all because we
2685 * know that 'from' will not move past 'to' and vice versa. */
2686 resort_bals(from, bals, n_bals);
2687 resort_bals(to, bals, n_bals);
2694 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2695 * historical data to decay to <1% in 7 rebalancing runs. */
2696 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2702 bond_send_learning_packets(struct port *port)
2704 struct bridge *br = port->bridge;
2705 struct mac_entry *e;
2706 struct ofpbuf packet;
2707 int error, n_packets, n_errors;
2709 if (!port->n_ifaces || port->active_iface < 0) {
2713 ofpbuf_init(&packet, 128);
2714 error = n_packets = n_errors = 0;
2715 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2716 union ofp_action actions[2], *a;
2722 if (e->port == port->port_idx
2723 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2727 /* Compose actions. */
2728 memset(actions, 0, sizeof actions);
2731 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2732 a->vlan_vid.len = htons(sizeof *a);
2733 a->vlan_vid.vlan_vid = htons(e->vlan);
2736 a->output.type = htons(OFPAT_OUTPUT);
2737 a->output.len = htons(sizeof *a);
2738 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2743 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2745 flow_extract(&packet, ODPP_NONE, &flow);
2746 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2753 ofpbuf_uninit(&packet);
2756 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2757 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2758 "packets, last error was: %s",
2759 port->name, n_errors, n_packets, strerror(error));
2761 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2762 port->name, n_packets);
2766 /* Bonding unixctl user interface functions. */
2769 bond_unixctl_list(struct unixctl_conn *conn,
2770 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2772 struct ds ds = DS_EMPTY_INITIALIZER;
2773 const struct bridge *br;
2775 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2777 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2780 for (i = 0; i < br->n_ports; i++) {
2781 const struct port *port = br->ports[i];
2782 if (port->n_ifaces > 1) {
2785 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2786 for (j = 0; j < port->n_ifaces; j++) {
2787 const struct iface *iface = port->ifaces[j];
2789 ds_put_cstr(&ds, ", ");
2791 ds_put_cstr(&ds, iface->name);
2793 ds_put_char(&ds, '\n');
2797 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2801 static struct port *
2802 bond_find(const char *name)
2804 const struct bridge *br;
2806 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2809 for (i = 0; i < br->n_ports; i++) {
2810 struct port *port = br->ports[i];
2811 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2820 bond_unixctl_show(struct unixctl_conn *conn,
2821 const char *args, void *aux OVS_UNUSED)
2823 struct ds ds = DS_EMPTY_INITIALIZER;
2824 const struct port *port;
2827 port = bond_find(args);
2829 unixctl_command_reply(conn, 501, "no such bond");
2833 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2834 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2835 ds_put_format(&ds, "next rebalance: %lld ms\n",
2836 port->bond_next_rebalance - time_msec());
2837 for (j = 0; j < port->n_ifaces; j++) {
2838 const struct iface *iface = port->ifaces[j];
2839 struct bond_entry *be;
2842 ds_put_format(&ds, "slave %s: %s\n",
2843 iface->name, iface->enabled ? "enabled" : "disabled");
2844 if (j == port->active_iface) {
2845 ds_put_cstr(&ds, "\tactive slave\n");
2847 if (iface->delay_expires != LLONG_MAX) {
2848 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2849 iface->enabled ? "downdelay" : "updelay",
2850 iface->delay_expires - time_msec());
2854 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2855 int hash = be - port->bond_hash;
2856 struct mac_entry *me;
2858 if (be->iface_idx != j) {
2862 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2863 hash, be->tx_bytes / 1024);
2866 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2867 &port->bridge->ml->lrus) {
2870 if (bond_hash(me->mac) == hash
2871 && me->port != port->port_idx
2872 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2873 && dp_ifidx == iface->dp_ifidx)
2875 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2876 ETH_ADDR_ARGS(me->mac));
2881 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2886 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2887 void *aux OVS_UNUSED)
2889 char *args = (char *) args_;
2890 char *save_ptr = NULL;
2891 char *bond_s, *hash_s, *slave_s;
2892 uint8_t mac[ETH_ADDR_LEN];
2894 struct iface *iface;
2895 struct bond_entry *entry;
2898 bond_s = strtok_r(args, " ", &save_ptr);
2899 hash_s = strtok_r(NULL, " ", &save_ptr);
2900 slave_s = strtok_r(NULL, " ", &save_ptr);
2902 unixctl_command_reply(conn, 501,
2903 "usage: bond/migrate BOND HASH SLAVE");
2907 port = bond_find(bond_s);
2909 unixctl_command_reply(conn, 501, "no such bond");
2913 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2914 == ETH_ADDR_SCAN_COUNT) {
2915 hash = bond_hash(mac);
2916 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2917 hash = atoi(hash_s) & BOND_MASK;
2919 unixctl_command_reply(conn, 501, "bad hash");
2923 iface = port_lookup_iface(port, slave_s);
2925 unixctl_command_reply(conn, 501, "no such slave");
2929 if (!iface->enabled) {
2930 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2934 entry = &port->bond_hash[hash];
2935 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2936 entry->iface_idx = iface->port_ifidx;
2937 entry->iface_tag = tag_create_random();
2938 port->bond_compat_is_stale = true;
2939 unixctl_command_reply(conn, 200, "migrated");
2943 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2944 void *aux OVS_UNUSED)
2946 char *args = (char *) args_;
2947 char *save_ptr = NULL;
2948 char *bond_s, *slave_s;
2950 struct iface *iface;
2952 bond_s = strtok_r(args, " ", &save_ptr);
2953 slave_s = strtok_r(NULL, " ", &save_ptr);
2955 unixctl_command_reply(conn, 501,
2956 "usage: bond/set-active-slave BOND SLAVE");
2960 port = bond_find(bond_s);
2962 unixctl_command_reply(conn, 501, "no such bond");
2966 iface = port_lookup_iface(port, slave_s);
2968 unixctl_command_reply(conn, 501, "no such slave");
2972 if (!iface->enabled) {
2973 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2977 if (port->active_iface != iface->port_ifidx) {
2978 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2979 port->active_iface = iface->port_ifidx;
2980 port->active_iface_tag = tag_create_random();
2981 VLOG_INFO("port %s: active interface is now %s",
2982 port->name, iface->name);
2983 bond_send_learning_packets(port);
2984 unixctl_command_reply(conn, 200, "done");
2986 unixctl_command_reply(conn, 200, "no change");
2991 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2993 char *args = (char *) args_;
2994 char *save_ptr = NULL;
2995 char *bond_s, *slave_s;
2997 struct iface *iface;
2999 bond_s = strtok_r(args, " ", &save_ptr);
3000 slave_s = strtok_r(NULL, " ", &save_ptr);
3002 unixctl_command_reply(conn, 501,
3003 "usage: bond/enable/disable-slave BOND SLAVE");
3007 port = bond_find(bond_s);
3009 unixctl_command_reply(conn, 501, "no such bond");
3013 iface = port_lookup_iface(port, slave_s);
3015 unixctl_command_reply(conn, 501, "no such slave");
3019 bond_enable_slave(iface, enable);
3020 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3024 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3025 void *aux OVS_UNUSED)
3027 enable_slave(conn, args, true);
3031 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3032 void *aux OVS_UNUSED)
3034 enable_slave(conn, args, false);
3038 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3039 void *aux OVS_UNUSED)
3041 uint8_t mac[ETH_ADDR_LEN];
3045 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3046 == ETH_ADDR_SCAN_COUNT) {
3047 hash = bond_hash(mac);
3049 hash_cstr = xasprintf("%u", hash);
3050 unixctl_command_reply(conn, 200, hash_cstr);
3053 unixctl_command_reply(conn, 501, "invalid mac");
3060 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3061 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3062 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3063 unixctl_command_register("bond/set-active-slave",
3064 bond_unixctl_set_active_slave, NULL);
3065 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3067 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3069 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3072 /* Port functions. */
3074 static struct port *
3075 port_create(struct bridge *br, const char *name)
3079 port = xzalloc(sizeof *port);
3081 port->port_idx = br->n_ports;
3083 port->trunks = NULL;
3084 port->name = xstrdup(name);
3085 port->active_iface = -1;
3087 if (br->n_ports >= br->allocated_ports) {
3088 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3091 br->ports[br->n_ports++] = port;
3093 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3100 get_port_other_config(const struct ovsrec_port *port, const char *key,
3101 const char *default_value)
3103 const char *value = get_ovsrec_key_value(key,
3104 port->key_other_config,
3105 port->value_other_config,
3106 port->n_other_config);
3107 return value ? value : default_value;
3111 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3113 struct shash old_ifaces, new_ifaces;
3114 long long int next_rebalance;
3115 struct shash_node *node;
3116 unsigned long *trunks;
3122 /* Collect old and new interfaces. */
3123 shash_init(&old_ifaces);
3124 shash_init(&new_ifaces);
3125 for (i = 0; i < port->n_ifaces; i++) {
3126 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3128 for (i = 0; i < cfg->n_interfaces; i++) {
3129 const char *name = cfg->interfaces[i]->name;
3130 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3131 VLOG_WARN("port %s: %s specified twice as port interface",
3135 port->updelay = cfg->bond_updelay;
3136 if (port->updelay < 0) {
3139 port->updelay = cfg->bond_downdelay;
3140 if (port->downdelay < 0) {
3141 port->downdelay = 0;
3143 port->bond_rebalance_interval = atoi(
3144 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3145 if (port->bond_rebalance_interval < 1000) {
3146 port->bond_rebalance_interval = 1000;
3148 next_rebalance = time_msec() + port->bond_rebalance_interval;
3149 if (port->bond_next_rebalance > next_rebalance) {
3150 port->bond_next_rebalance = next_rebalance;
3153 /* Get rid of deleted interfaces and add new interfaces. */
3154 SHASH_FOR_EACH (node, &old_ifaces) {
3155 if (!shash_find(&new_ifaces, node->name)) {
3156 iface_destroy(node->data);
3159 SHASH_FOR_EACH (node, &new_ifaces) {
3160 const struct ovsrec_interface *if_cfg = node->data;
3161 struct iface *iface;
3163 iface = shash_find_data(&old_ifaces, if_cfg->name);
3165 iface_create(port, if_cfg);
3167 iface->cfg = if_cfg;
3174 if (port->n_ifaces < 2) {
3176 if (vlan >= 0 && vlan <= 4095) {
3177 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3182 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3183 * they even work as-is. But they have not been tested. */
3184 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3188 if (port->vlan != vlan) {
3190 bridge_flush(port->bridge);
3193 /* Get trunked VLANs. */
3199 trunks = bitmap_allocate(4096);
3201 for (i = 0; i < cfg->n_trunks; i++) {
3202 int trunk = cfg->trunks[i];
3204 bitmap_set1(trunks, trunk);
3210 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3211 port->name, cfg->n_trunks);
3213 if (n_errors == cfg->n_trunks) {
3215 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3218 bitmap_set_multiple(trunks, 0, 4096, 1);
3221 if (cfg->n_trunks) {
3222 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3227 ? port->trunks != NULL
3228 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3229 bridge_flush(port->bridge);
3231 bitmap_free(port->trunks);
3232 port->trunks = trunks;
3234 shash_destroy(&old_ifaces);
3235 shash_destroy(&new_ifaces);
3239 port_destroy(struct port *port)
3242 struct bridge *br = port->bridge;
3246 proc_net_compat_update_vlan(port->name, NULL, 0);
3247 proc_net_compat_update_bond(port->name, NULL);
3249 for (i = 0; i < MAX_MIRRORS; i++) {
3250 struct mirror *m = br->mirrors[i];
3251 if (m && m->out_port == port) {
3256 while (port->n_ifaces > 0) {
3257 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3260 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3261 del->port_idx = port->port_idx;
3264 bitmap_free(port->trunks);
3271 static struct port *
3272 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3274 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3275 return iface ? iface->port : NULL;
3278 static struct port *
3279 port_lookup(const struct bridge *br, const char *name)
3283 for (i = 0; i < br->n_ports; i++) {
3284 struct port *port = br->ports[i];
3285 if (!strcmp(port->name, name)) {
3292 static struct iface *
3293 port_lookup_iface(const struct port *port, const char *name)
3297 for (j = 0; j < port->n_ifaces; j++) {
3298 struct iface *iface = port->ifaces[j];
3299 if (!strcmp(iface->name, name)) {
3307 port_update_bonding(struct port *port)
3309 if (port->n_ifaces < 2) {
3310 /* Not a bonded port. */
3311 if (port->bond_hash) {
3312 free(port->bond_hash);
3313 port->bond_hash = NULL;
3314 port->bond_compat_is_stale = true;
3315 port->bond_fake_iface = false;
3318 if (!port->bond_hash) {
3321 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3322 for (i = 0; i <= BOND_MASK; i++) {
3323 struct bond_entry *e = &port->bond_hash[i];
3327 port->no_ifaces_tag = tag_create_random();
3328 bond_choose_active_iface(port);
3329 port->bond_next_rebalance
3330 = time_msec() + port->bond_rebalance_interval;
3332 port->bond_compat_is_stale = true;
3333 port->bond_fake_iface = port->cfg->bond_fake_iface;
3338 port_update_bond_compat(struct port *port)
3340 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3341 struct compat_bond bond;
3344 if (port->n_ifaces < 2) {
3345 proc_net_compat_update_bond(port->name, NULL);
3350 bond.updelay = port->updelay;
3351 bond.downdelay = port->downdelay;
3354 bond.hashes = compat_hashes;
3355 if (port->bond_hash) {
3356 const struct bond_entry *e;
3357 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3358 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3359 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3360 cbh->hash = e - port->bond_hash;
3361 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3366 bond.n_slaves = port->n_ifaces;
3367 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3368 for (i = 0; i < port->n_ifaces; i++) {
3369 struct iface *iface = port->ifaces[i];
3370 struct compat_bond_slave *slave = &bond.slaves[i];
3371 slave->name = iface->name;
3373 /* We need to make the same determination as the Linux bonding
3374 * code to determine whether a slave should be consider "up".
3375 * The Linux function bond_miimon_inspect() supports four
3376 * BOND_LINK_* states:
3378 * - BOND_LINK_UP: carrier detected, updelay has passed.
3379 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3380 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3381 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3383 * The function bond_info_show_slave() only considers BOND_LINK_UP
3384 * to be "up" and anything else to be "down".
3386 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3390 netdev_get_etheraddr(iface->netdev, slave->mac);
3393 if (port->bond_fake_iface) {
3394 struct netdev *bond_netdev;
3396 if (!netdev_open_default(port->name, &bond_netdev)) {
3398 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3400 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3402 netdev_close(bond_netdev);
3406 proc_net_compat_update_bond(port->name, &bond);
3411 port_update_vlan_compat(struct port *port)
3413 struct bridge *br = port->bridge;
3414 char *vlandev_name = NULL;
3416 if (port->vlan > 0) {
3417 /* Figure out the name that the VLAN device should actually have, if it
3418 * existed. This takes some work because the VLAN device would not
3419 * have port->name in its name; rather, it would have the trunk port's
3420 * name, and 'port' would be attached to a bridge that also had the
3421 * VLAN device one of its ports. So we need to find a trunk port that
3422 * includes port->vlan.
3424 * There might be more than one candidate. This doesn't happen on
3425 * XenServer, so if it happens we just pick the first choice in
3426 * alphabetical order instead of creating multiple VLAN devices. */
3428 for (i = 0; i < br->n_ports; i++) {
3429 struct port *p = br->ports[i];
3430 if (port_trunks_vlan(p, port->vlan)
3432 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3434 uint8_t ea[ETH_ADDR_LEN];
3435 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3436 if (!eth_addr_is_multicast(ea) &&
3437 !eth_addr_is_reserved(ea) &&
3438 !eth_addr_is_zero(ea)) {
3439 vlandev_name = p->name;
3444 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3447 /* Interface functions. */
3449 static struct iface *
3450 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3452 struct iface *iface;
3453 char *name = if_cfg->name;
3456 iface = xzalloc(sizeof *iface);
3458 iface->port_ifidx = port->n_ifaces;
3459 iface->name = xstrdup(name);
3460 iface->dp_ifidx = -1;
3461 iface->tag = tag_create_random();
3462 iface->delay_expires = LLONG_MAX;
3463 iface->netdev = NULL;
3464 iface->cfg = if_cfg;
3466 if (port->n_ifaces >= port->allocated_ifaces) {
3467 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3468 sizeof *port->ifaces);
3470 port->ifaces[port->n_ifaces++] = iface;
3471 if (port->n_ifaces > 1) {
3472 port->bridge->has_bonded_ports = true;
3475 /* Attempt to create the network interface in case it
3476 * doesn't exist yet. */
3477 if (!iface_is_internal(port->bridge, iface->name)) {
3478 error = set_up_iface(if_cfg, iface, true);
3480 VLOG_WARN("could not create iface %s: %s", iface->name,
3485 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3487 bridge_flush(port->bridge);
3493 iface_destroy(struct iface *iface)
3496 struct port *port = iface->port;
3497 struct bridge *br = port->bridge;
3498 bool del_active = port->active_iface == iface->port_ifidx;
3501 if (iface->dp_ifidx >= 0) {
3502 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3505 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3506 del->port_ifidx = iface->port_ifidx;
3508 netdev_close(iface->netdev);
3511 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3512 bond_choose_active_iface(port);
3513 bond_send_learning_packets(port);
3519 bridge_flush(port->bridge);
3523 static struct iface *
3524 iface_lookup(const struct bridge *br, const char *name)
3528 for (i = 0; i < br->n_ports; i++) {
3529 struct port *port = br->ports[i];
3530 for (j = 0; j < port->n_ifaces; j++) {
3531 struct iface *iface = port->ifaces[j];
3532 if (!strcmp(iface->name, name)) {
3540 static struct iface *
3541 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3543 return port_array_get(&br->ifaces, dp_ifidx);
3546 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3547 * 'br', that is, an interface that is entirely simulated within the datapath.
3548 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3549 * interfaces are created by setting "iface.<iface>.internal = true".
3551 * In addition, we have a kluge-y feature that creates an internal port with
3552 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3553 * This feature needs to go away in the long term. Until then, this is one
3554 * reason why this function takes a name instead of a struct iface: the fake
3555 * interfaces created this way do not have a struct iface. */
3557 iface_is_internal(const struct bridge *br, const char *if_name)
3559 /* XXX wastes time */
3560 struct iface *iface;
3563 if (!strcmp(if_name, br->name)) {
3567 iface = iface_lookup(br, if_name);
3568 if (iface && !strcmp(iface->cfg->type, "internal")) {
3572 port = port_lookup(br, if_name);
3573 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3579 /* Set Ethernet address of 'iface', if one is specified in the configuration
3582 iface_set_mac(struct iface *iface)
3584 uint8_t ea[ETH_ADDR_LEN];
3586 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3587 if (eth_addr_is_multicast(ea)) {
3588 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3590 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3591 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3592 iface->name, iface->name);
3594 int error = netdev_set_etheraddr(iface->netdev, ea);
3596 VLOG_ERR("interface %s: setting MAC failed (%s)",
3597 iface->name, strerror(error));
3603 /* Port mirroring. */
3606 mirror_reconfigure(struct bridge *br)
3608 struct shash old_mirrors, new_mirrors;
3609 struct shash_node *node;
3610 unsigned long *rspan_vlans;
3613 /* Collect old mirrors. */
3614 shash_init(&old_mirrors);
3615 for (i = 0; i < MAX_MIRRORS; i++) {
3616 if (br->mirrors[i]) {
3617 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3621 /* Collect new mirrors. */
3622 shash_init(&new_mirrors);
3623 for (i = 0; i < br->cfg->n_mirrors; i++) {
3624 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3625 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3626 VLOG_WARN("bridge %s: %s specified twice as mirror",
3627 br->name, cfg->name);
3631 /* Get rid of deleted mirrors and add new mirrors. */
3632 SHASH_FOR_EACH (node, &old_mirrors) {
3633 if (!shash_find(&new_mirrors, node->name)) {
3634 mirror_destroy(node->data);
3637 SHASH_FOR_EACH (node, &new_mirrors) {
3638 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3640 mirror = mirror_create(br, node->name);
3645 mirror_reconfigure_one(mirror, node->data);
3647 shash_destroy(&old_mirrors);
3648 shash_destroy(&new_mirrors);
3650 /* Update port reserved status. */
3651 for (i = 0; i < br->n_ports; i++) {
3652 br->ports[i]->is_mirror_output_port = false;
3654 for (i = 0; i < MAX_MIRRORS; i++) {
3655 struct mirror *m = br->mirrors[i];
3656 if (m && m->out_port) {
3657 m->out_port->is_mirror_output_port = true;
3661 /* Update flooded vlans (for RSPAN). */
3663 if (br->cfg->n_flood_vlans) {
3664 rspan_vlans = bitmap_allocate(4096);
3666 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3667 int64_t vlan = br->cfg->flood_vlans[i];
3668 if (vlan >= 0 && vlan < 4096) {
3669 bitmap_set1(rspan_vlans, vlan);
3670 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3673 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3678 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3683 static struct mirror *
3684 mirror_create(struct bridge *br, const char *name)
3689 for (i = 0; ; i++) {
3690 if (i >= MAX_MIRRORS) {
3691 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3692 "cannot create %s", br->name, MAX_MIRRORS, name);
3695 if (!br->mirrors[i]) {
3700 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3703 br->mirrors[i] = m = xzalloc(sizeof *m);
3706 m->name = xstrdup(name);
3707 shash_init(&m->src_ports);
3708 shash_init(&m->dst_ports);
3718 mirror_destroy(struct mirror *m)
3721 struct bridge *br = m->bridge;
3724 for (i = 0; i < br->n_ports; i++) {
3725 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3726 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3729 shash_destroy(&m->src_ports);
3730 shash_destroy(&m->dst_ports);
3733 m->bridge->mirrors[m->idx] = NULL;
3741 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3742 struct shash *names)
3746 for (i = 0; i < n_ports; i++) {
3747 const char *name = ports[i]->name;
3748 if (port_lookup(m->bridge, name)) {
3749 shash_add_once(names, name, NULL);
3751 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3752 "port %s", m->bridge->name, m->name, name);
3758 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3764 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3766 for (i = 0; i < cfg->n_select_vlan; i++) {
3767 int64_t vlan = cfg->select_vlan[i];
3768 if (vlan < 0 || vlan > 4095) {
3769 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3770 m->bridge->name, m->name, vlan);
3772 (*vlans)[n_vlans++] = vlan;
3779 vlan_is_mirrored(const struct mirror *m, int vlan)
3783 for (i = 0; i < m->n_vlans; i++) {
3784 if (m->vlans[i] == vlan) {
3792 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3796 for (i = 0; i < m->n_vlans; i++) {
3797 if (port_trunks_vlan(p, m->vlans[i])) {
3805 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3807 struct shash src_ports, dst_ports;
3808 mirror_mask_t mirror_bit;
3809 struct port *out_port;
3815 /* Get output port. */
3816 if (cfg->output_port) {
3817 out_port = port_lookup(m->bridge, cfg->output_port->name);
3819 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3820 m->bridge->name, m->name);
3826 if (cfg->output_vlan) {
3827 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3828 "output vlan; ignoring output vlan",
3829 m->bridge->name, m->name);
3831 } else if (cfg->output_vlan) {
3833 out_vlan = *cfg->output_vlan;
3835 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3836 m->bridge->name, m->name);
3841 shash_init(&src_ports);
3842 shash_init(&dst_ports);
3843 if (cfg->select_all) {
3844 for (i = 0; i < m->bridge->n_ports; i++) {
3845 const char *name = m->bridge->ports[i]->name;
3846 shash_add_once(&src_ports, name, NULL);
3847 shash_add_once(&dst_ports, name, NULL);
3852 /* Get ports, and drop duplicates and ports that don't exist. */
3853 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3855 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3858 /* Get all the vlans, and drop duplicate and invalid vlans. */
3859 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3862 /* Update mirror data. */
3863 if (!shash_equal_keys(&m->src_ports, &src_ports)
3864 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3865 || m->n_vlans != n_vlans
3866 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3867 || m->out_port != out_port
3868 || m->out_vlan != out_vlan) {
3869 bridge_flush(m->bridge);
3871 shash_swap(&m->src_ports, &src_ports);
3872 shash_swap(&m->dst_ports, &dst_ports);
3875 m->n_vlans = n_vlans;
3876 m->out_port = out_port;
3877 m->out_vlan = out_vlan;
3880 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3881 for (i = 0; i < m->bridge->n_ports; i++) {
3882 struct port *port = m->bridge->ports[i];
3884 if (shash_find(&m->src_ports, port->name)
3887 ? port_trunks_any_mirrored_vlan(m, port)
3888 : vlan_is_mirrored(m, port->vlan)))) {
3889 port->src_mirrors |= mirror_bit;
3891 port->src_mirrors &= ~mirror_bit;
3894 if (shash_find(&m->dst_ports, port->name)) {
3895 port->dst_mirrors |= mirror_bit;
3897 port->dst_mirrors &= ~mirror_bit;
3902 shash_destroy(&src_ports);
3903 shash_destroy(&dst_ports);