1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
36 #include "dynamic-string.h"
40 #include "mac-learning.h"
43 #include "ofp-print.h"
45 #include "ofproto/netflow.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
54 #include "socket-util.h"
55 #include "stream-ssl.h"
61 #include "vswitchd/vswitch-idl.h"
62 #include "xenserver.h"
64 #include "sflow_api.h"
66 #define THIS_MODULE VLM_bridge
75 /* These members are always valid. */
76 struct port *port; /* Containing port. */
77 size_t port_ifidx; /* Index within containing port. */
78 char *name; /* Host network device name. */
79 tag_type tag; /* Tag associated with this interface. */
80 long long delay_expires; /* Time after which 'enabled' may change. */
82 /* These members are valid only after bridge_reconfigure() causes them to
84 int dp_ifidx; /* Index within kernel datapath. */
85 struct netdev *netdev; /* Network device. */
86 bool enabled; /* May be chosen for flows? */
88 /* This member is only valid *during* bridge_reconfigure(). */
89 const struct ovsrec_interface *cfg;
92 #define BOND_MASK 0xff
94 int iface_idx; /* Index of assigned iface, or -1 if none. */
95 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
96 tag_type iface_tag; /* Tag associated with iface_idx. */
99 #define MAX_MIRRORS 32
100 typedef uint32_t mirror_mask_t;
101 #define MIRROR_MASK_C(X) UINT32_C(X)
102 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
104 struct bridge *bridge;
108 /* Selection criteria. */
109 struct shash src_ports; /* Name is port name; data is always NULL. */
110 struct shash dst_ports; /* Name is port name; data is always NULL. */
115 struct port *out_port;
119 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
121 struct bridge *bridge;
123 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
124 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
127 /* An ordinary bridge port has 1 interface.
128 * A bridge port for bonding has at least 2 interfaces. */
129 struct iface **ifaces;
130 size_t n_ifaces, allocated_ifaces;
133 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
134 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
135 tag_type active_iface_tag; /* Tag for bcast flows. */
136 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
137 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
138 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
139 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
141 /* Port mirroring info. */
142 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
143 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
144 bool is_mirror_output_port; /* Does port mirroring send frames here? */
146 /* This member is only valid *during* bridge_reconfigure(). */
147 const struct ovsrec_port *cfg;
150 #define DP_MAX_PORTS 255
152 struct list node; /* Node in global list of bridges. */
153 char *name; /* User-specified arbitrary name. */
154 struct mac_learning *ml; /* MAC learning table. */
155 bool sent_config_request; /* Successfully sent config request? */
156 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
158 /* Support for remote controllers. */
159 char *controller; /* NULL if there is no remote controller;
160 * "discover" to do controller discovery;
161 * otherwise a vconn name. */
163 /* OpenFlow switch processing. */
164 struct ofproto *ofproto; /* OpenFlow switch. */
166 /* Kernel datapath information. */
167 struct dpif *dpif; /* Datapath. */
168 struct port_array ifaces; /* Indexed by kernel datapath port number. */
172 size_t n_ports, allocated_ports;
175 bool has_bonded_ports;
176 long long int bond_next_rebalance;
181 /* Flow statistics gathering. */
182 time_t next_stats_request;
184 /* Port mirroring. */
185 struct mirror *mirrors[MAX_MIRRORS];
187 /* This member is only valid *during* bridge_reconfigure(). */
188 const struct ovsrec_bridge *cfg;
191 /* List of all bridges. */
192 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
194 /* Maximum number of datapaths. */
195 enum { DP_MAX = 256 };
197 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
198 static void bridge_destroy(struct bridge *);
199 static struct bridge *bridge_lookup(const char *name);
200 static unixctl_cb_func bridge_unixctl_dump_flows;
201 static int bridge_run_one(struct bridge *);
202 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
204 static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
206 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
207 static void bridge_fetch_dp_ifaces(struct bridge *);
208 static void bridge_flush(struct bridge *);
209 static void bridge_pick_local_hw_addr(struct bridge *,
210 uint8_t ea[ETH_ADDR_LEN],
211 struct iface **hw_addr_iface);
212 static uint64_t bridge_pick_datapath_id(struct bridge *,
213 const uint8_t bridge_ea[ETH_ADDR_LEN],
214 struct iface *hw_addr_iface);
215 static struct iface *bridge_get_local_iface(struct bridge *);
216 static uint64_t dpid_from_hash(const void *, size_t nbytes);
218 static unixctl_cb_func bridge_unixctl_fdb_show;
220 static void bond_init(void);
221 static void bond_run(struct bridge *);
222 static void bond_wait(struct bridge *);
223 static void bond_rebalance_port(struct port *);
224 static void bond_send_learning_packets(struct port *);
225 static void bond_enable_slave(struct iface *iface, bool enable);
227 static struct port *port_create(struct bridge *, const char *name);
228 static void port_reconfigure(struct port *, const struct ovsrec_port *);
229 static void port_destroy(struct port *);
230 static struct port *port_lookup(const struct bridge *, const char *name);
231 static struct iface *port_lookup_iface(const struct port *, const char *name);
232 static struct port *port_from_dp_ifidx(const struct bridge *,
234 static void port_update_bond_compat(struct port *);
235 static void port_update_vlan_compat(struct port *);
236 static void port_update_bonding(struct port *);
238 static struct mirror *mirror_create(struct bridge *, const char *name);
239 static void mirror_destroy(struct mirror *);
240 static void mirror_reconfigure(struct bridge *);
241 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
242 static bool vlan_is_mirrored(const struct mirror *, int vlan);
244 static struct iface *iface_create(struct port *port,
245 const struct ovsrec_interface *if_cfg);
246 static void iface_destroy(struct iface *);
247 static struct iface *iface_lookup(const struct bridge *, const char *name);
248 static struct iface *iface_from_dp_ifidx(const struct bridge *,
250 static bool iface_is_internal(const struct bridge *, const char *name);
251 static void iface_set_mac(struct iface *);
253 /* Hooks into ofproto processing. */
254 static struct ofhooks bridge_ofhooks;
256 /* Public functions. */
258 /* Adds the name of each interface used by a bridge, including local and
259 * internal ports, to 'svec'. */
261 bridge_get_ifaces(struct svec *svec)
263 struct bridge *br, *next;
266 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
267 for (i = 0; i < br->n_ports; i++) {
268 struct port *port = br->ports[i];
270 for (j = 0; j < port->n_ifaces; j++) {
271 struct iface *iface = port->ifaces[j];
272 if (iface->dp_ifidx < 0) {
273 VLOG_ERR("%s interface not in datapath %s, ignoring",
274 iface->name, dpif_name(br->dpif));
276 if (iface->dp_ifidx != ODPP_LOCAL) {
277 svec_add(svec, iface->name);
286 bridge_init(const struct ovsrec_open_vswitch *cfg)
288 struct svec bridge_names;
289 struct svec dpif_names, dpif_types;
292 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
294 svec_init(&bridge_names);
295 for (i = 0; i < cfg->n_bridges; i++) {
296 svec_add(&bridge_names, cfg->bridges[i]->name);
298 svec_sort(&bridge_names);
300 svec_init(&dpif_names);
301 svec_init(&dpif_types);
302 dp_enumerate_types(&dpif_types);
303 for (i = 0; i < dpif_types.n; i++) {
308 dp_enumerate_names(dpif_types.names[i], &dpif_names);
310 for (j = 0; j < dpif_names.n; j++) {
311 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
313 struct svec all_names;
316 svec_init(&all_names);
317 dpif_get_all_names(dpif, &all_names);
318 for (k = 0; k < all_names.n; k++) {
319 if (svec_contains(&bridge_names, all_names.names[k])) {
325 svec_destroy(&all_names);
330 svec_destroy(&dpif_names);
331 svec_destroy(&dpif_types);
333 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
337 bridge_reconfigure(cfg);
342 config_string_change(const char *value, char **valuep)
344 if (value && (!*valuep || strcmp(value, *valuep))) {
346 *valuep = xstrdup(value);
354 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
356 /* XXX SSL should be configurable on a per-bridge basis.
357 * XXX should be possible to de-configure SSL. */
358 static char *private_key_file;
359 static char *certificate_file;
360 static char *cacert_file;
364 /* XXX We can't un-set SSL settings. */
368 if (config_string_change(ssl->private_key, &private_key_file)) {
369 stream_ssl_set_private_key_file(private_key_file);
372 if (config_string_change(ssl->certificate, &certificate_file)) {
373 stream_ssl_set_certificate_file(certificate_file);
376 /* We assume that even if the filename hasn't changed, if the CA cert
377 * file has been removed, that we want to move back into
378 * boot-strapping mode. This opens a small security hole, because
379 * the old certificate will still be trusted until vSwitch is
380 * restarted. We may want to address this in vconn's SSL library. */
381 if (config_string_change(ssl->ca_cert, &cacert_file)
382 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
383 stream_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert);
388 /* Attempt to create the network device 'iface_name' through the netdev
391 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
394 struct shash_node *node;
395 struct shash options;
399 shash_init(&options);
400 for (i = 0; i < iface_cfg->n_options; i++) {
401 shash_add(&options, iface_cfg->key_options[i],
402 xstrdup(iface_cfg->value_options[i]));
406 struct netdev_options netdev_options;
408 memset(&netdev_options, 0, sizeof netdev_options);
409 netdev_options.name = iface_cfg->name;
410 netdev_options.type = iface_cfg->type;
411 netdev_options.args = &options;
412 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
413 netdev_options.may_create = true;
414 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
415 netdev_options.may_open = true;
418 error = netdev_open(&netdev_options, &iface->netdev);
421 netdev_get_carrier(iface->netdev, &iface->enabled);
423 } else if (iface->netdev) {
424 const char *netdev_type = netdev_get_type(iface->netdev);
425 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
426 ? iface_cfg->type : NULL;
428 if (!iface_type || !strcmp(netdev_type, iface_type)) {
429 error = netdev_reconfigure(iface->netdev, &options);
431 VLOG_WARN("%s: attempting change device type from %s to %s",
432 iface_cfg->name, netdev_type, iface_type);
437 SHASH_FOR_EACH (node, &options) {
440 shash_destroy(&options);
446 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
448 return set_up_iface(iface_cfg, iface, false);
452 check_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
455 if (!iface->netdev) {
456 int error = set_up_iface(iface->cfg, iface, true);
458 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
468 check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED)
470 if (iface->dp_ifidx >= 0) {
471 VLOG_DBG("%s has interface %s on port %d",
473 iface->name, iface->dp_ifidx);
476 VLOG_ERR("%s interface not in %s, dropping",
477 iface->name, dpif_name(br->dpif));
483 set_iface_properties(struct bridge *br UNUSED, struct iface *iface,
486 /* Set policing attributes. */
487 netdev_set_policing(iface->netdev,
488 iface->cfg->ingress_policing_rate,
489 iface->cfg->ingress_policing_burst);
491 /* Set MAC address of internal interfaces other than the local
493 if (iface->dp_ifidx != ODPP_LOCAL
494 && iface_is_internal(br, iface->name)) {
495 iface_set_mac(iface);
501 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
502 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
503 * deletes from 'br' any ports that no longer have any interfaces. */
505 iterate_and_prune_ifaces(struct bridge *br,
506 bool (*cb)(struct bridge *, struct iface *,
512 for (i = 0; i < br->n_ports; ) {
513 struct port *port = br->ports[i];
514 for (j = 0; j < port->n_ifaces; ) {
515 struct iface *iface = port->ifaces[j];
516 if (cb(br, iface, aux)) {
519 iface_destroy(iface);
523 if (port->n_ifaces) {
526 VLOG_ERR("%s port has no interfaces, dropping", port->name);
533 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
535 struct ovsdb_idl_txn *txn;
536 struct shash old_br, new_br;
537 struct shash_node *node;
538 struct bridge *br, *next;
540 int sflow_bridge_number;
542 COVERAGE_INC(bridge_reconfigure);
544 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
546 /* Collect old and new bridges. */
549 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
550 shash_add(&old_br, br->name, br);
552 for (i = 0; i < ovs_cfg->n_bridges; i++) {
553 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
554 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
555 VLOG_WARN("more than one bridge named %s", br_cfg->name);
559 /* Get rid of deleted bridges and add new bridges. */
560 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
561 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
568 SHASH_FOR_EACH (node, &new_br) {
569 const char *br_name = node->name;
570 const struct ovsrec_bridge *br_cfg = node->data;
571 br = shash_find_data(&old_br, br_name);
573 /* If the bridge datapath type has changed, we need to tear it
574 * down and recreate. */
575 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
577 bridge_create(br_cfg);
580 bridge_create(br_cfg);
583 shash_destroy(&old_br);
584 shash_destroy(&new_br);
588 bridge_configure_ssl(ovs_cfg->ssl);
591 /* Reconfigure all bridges. */
592 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
593 bridge_reconfigure_one(ovs_cfg, br);
596 /* Add and delete ports on all datapaths.
598 * The kernel will reject any attempt to add a given port to a datapath if
599 * that port already belongs to a different datapath, so we must do all
600 * port deletions before any port additions. */
601 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
602 struct odp_port *dpif_ports;
604 struct shash want_ifaces;
606 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
607 bridge_get_all_ifaces(br, &want_ifaces);
608 for (i = 0; i < n_dpif_ports; i++) {
609 const struct odp_port *p = &dpif_ports[i];
610 if (!shash_find(&want_ifaces, p->devname)
611 && strcmp(p->devname, br->name)) {
612 int retval = dpif_port_del(br->dpif, p->port);
614 VLOG_ERR("failed to remove %s interface from %s: %s",
615 p->devname, dpif_name(br->dpif),
620 shash_destroy(&want_ifaces);
623 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
624 struct odp_port *dpif_ports;
626 struct shash cur_ifaces, want_ifaces;
627 struct shash_node *node;
629 /* Get the set of interfaces currently in this datapath. */
630 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
631 shash_init(&cur_ifaces);
632 for (i = 0; i < n_dpif_ports; i++) {
633 const char *name = dpif_ports[i].devname;
634 if (!shash_find(&cur_ifaces, name)) {
635 shash_add(&cur_ifaces, name, NULL);
640 /* Get the set of interfaces we want on this datapath. */
641 bridge_get_all_ifaces(br, &want_ifaces);
643 SHASH_FOR_EACH (node, &want_ifaces) {
644 const char *if_name = node->name;
645 struct iface *iface = node->data;
647 if (shash_find(&cur_ifaces, if_name)) {
648 /* Already exists, just reconfigure it. */
650 reconfigure_iface(iface->cfg, iface);
653 /* Need to add to datapath. */
657 /* Add to datapath. */
658 internal = iface_is_internal(br, if_name);
659 error = dpif_port_add(br->dpif, if_name,
660 internal ? ODP_PORT_INTERNAL : 0, NULL);
661 if (error == EFBIG) {
662 VLOG_ERR("ran out of valid port numbers on %s",
663 dpif_name(br->dpif));
666 VLOG_ERR("failed to add %s interface to %s: %s",
667 if_name, dpif_name(br->dpif), strerror(error));
671 shash_destroy(&cur_ifaces);
672 shash_destroy(&want_ifaces);
674 sflow_bridge_number = 0;
675 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
678 struct iface *local_iface;
679 struct iface *hw_addr_iface;
682 bridge_fetch_dp_ifaces(br);
684 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
685 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
687 /* Pick local port hardware address, datapath ID. */
688 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
689 local_iface = bridge_get_local_iface(br);
691 int error = netdev_set_etheraddr(local_iface->netdev, ea);
693 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
694 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
695 "Ethernet address: %s",
696 br->name, strerror(error));
700 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
701 ofproto_set_datapath_id(br->ofproto, dpid);
703 dpid_string = xasprintf("%012"PRIx64, dpid);
704 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
707 /* Set NetFlow configuration on this bridge. */
708 if (br->cfg->netflow) {
709 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
710 struct netflow_options opts;
712 memset(&opts, 0, sizeof opts);
714 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
715 if (nf_cfg->engine_type) {
716 opts.engine_type = *nf_cfg->engine_type;
718 if (nf_cfg->engine_id) {
719 opts.engine_id = *nf_cfg->engine_id;
722 opts.active_timeout = nf_cfg->active_timeout;
723 if (!opts.active_timeout) {
724 opts.active_timeout = -1;
725 } else if (opts.active_timeout < 0) {
726 VLOG_WARN("bridge %s: active timeout interval set to negative "
727 "value, using default instead (%d seconds)", br->name,
728 NF_ACTIVE_TIMEOUT_DEFAULT);
729 opts.active_timeout = -1;
732 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
733 if (opts.add_id_to_iface) {
734 if (opts.engine_id > 0x7f) {
735 VLOG_WARN("bridge %s: netflow port mangling may conflict "
736 "with another vswitch, choose an engine id less "
737 "than 128", br->name);
739 if (br->n_ports > 508) {
740 VLOG_WARN("bridge %s: netflow port mangling will conflict "
741 "with another port when more than 508 ports are "
746 opts.collectors.n = nf_cfg->n_targets;
747 opts.collectors.names = nf_cfg->targets;
748 if (ofproto_set_netflow(br->ofproto, &opts)) {
749 VLOG_ERR("bridge %s: problem setting netflow collectors",
753 ofproto_set_netflow(br->ofproto, NULL);
756 /* Set sFlow configuration on this bridge. */
757 if (br->cfg->sflow) {
758 struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
759 struct ofproto_sflow_options oso;
761 memset(&oso, 0, sizeof oso);
763 oso.targets.n = sflow_cfg->n_targets;
764 oso.targets.names = sflow_cfg->targets;
766 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
767 if (sflow_cfg->sampling) {
768 oso.sampling_rate = *sflow_cfg->sampling;
771 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
772 if (sflow_cfg->polling) {
773 oso.polling_interval = *sflow_cfg->polling;
776 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
777 if (sflow_cfg->header) {
778 oso.header_len = *sflow_cfg->header;
781 oso.sub_id = sflow_bridge_number++;
782 oso.agent_device = sflow_cfg->agent;
785 ctrl = bridge_get_controller(ovs_cfg, br);
786 oso.control_ip = ctrl ? ctrl->local_ip : NULL;
788 ofproto_set_sflow(br->ofproto, &oso);
790 svec_destroy(&oso.targets);
792 ofproto_set_sflow(br->ofproto, NULL);
795 /* Update the controller and related settings. It would be more
796 * straightforward to call this from bridge_reconfigure_one(), but we
797 * can't do it there for two reasons. First, and most importantly, at
798 * that point we don't know the dp_ifidx of any interfaces that have
799 * been added to the bridge (because we haven't actually added them to
800 * the datapath). Second, at that point we haven't set the datapath ID
801 * yet; when a controller is configured, resetting the datapath ID will
802 * immediately disconnect from the controller, so it's better to set
803 * the datapath ID before the controller. */
804 bridge_reconfigure_controller(ovs_cfg, br);
806 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
807 for (i = 0; i < br->n_ports; i++) {
808 struct port *port = br->ports[i];
810 port_update_vlan_compat(port);
811 port_update_bonding(port);
814 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
815 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
818 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
820 ovsdb_idl_txn_commit(txn);
821 ovsdb_idl_txn_destroy(txn); /* XXX */
825 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
829 for (i = 0; i < br_cfg->n_other_config; i++) {
830 if (!strcmp(br_cfg->key_other_config[i], key)) {
831 return br_cfg->value_other_config[i];
838 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
839 struct iface **hw_addr_iface)
845 *hw_addr_iface = NULL;
847 /* Did the user request a particular MAC? */
848 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
849 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
850 if (eth_addr_is_multicast(ea)) {
851 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
852 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
853 } else if (eth_addr_is_zero(ea)) {
854 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
860 /* Otherwise choose the minimum non-local MAC address among all of the
862 memset(ea, 0xff, sizeof ea);
863 for (i = 0; i < br->n_ports; i++) {
864 struct port *port = br->ports[i];
865 uint8_t iface_ea[ETH_ADDR_LEN];
868 /* Mirror output ports don't participate. */
869 if (port->is_mirror_output_port) {
873 /* Choose the MAC address to represent the port. */
874 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
875 /* Find the interface with this Ethernet address (if any) so that
876 * we can provide the correct devname to the caller. */
878 for (j = 0; j < port->n_ifaces; j++) {
879 struct iface *candidate = port->ifaces[j];
880 uint8_t candidate_ea[ETH_ADDR_LEN];
881 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
882 && eth_addr_equals(iface_ea, candidate_ea)) {
887 /* Choose the interface whose MAC address will represent the port.
888 * The Linux kernel bonding code always chooses the MAC address of
889 * the first slave added to a bond, and the Fedora networking
890 * scripts always add slaves to a bond in alphabetical order, so
891 * for compatibility we choose the interface with the name that is
892 * first in alphabetical order. */
893 iface = port->ifaces[0];
894 for (j = 1; j < port->n_ifaces; j++) {
895 struct iface *candidate = port->ifaces[j];
896 if (strcmp(candidate->name, iface->name) < 0) {
901 /* The local port doesn't count (since we're trying to choose its
902 * MAC address anyway). */
903 if (iface->dp_ifidx == ODPP_LOCAL) {
908 error = netdev_get_etheraddr(iface->netdev, iface_ea);
910 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
911 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
912 iface->name, strerror(error));
917 /* Compare against our current choice. */
918 if (!eth_addr_is_multicast(iface_ea) &&
919 !eth_addr_is_local(iface_ea) &&
920 !eth_addr_is_reserved(iface_ea) &&
921 !eth_addr_is_zero(iface_ea) &&
922 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
924 memcpy(ea, iface_ea, ETH_ADDR_LEN);
925 *hw_addr_iface = iface;
928 if (eth_addr_is_multicast(ea)) {
929 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
930 *hw_addr_iface = NULL;
931 VLOG_WARN("bridge %s: using default bridge Ethernet "
932 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
934 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
935 br->name, ETH_ADDR_ARGS(ea));
939 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
940 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
941 * an interface on 'br', then that interface must be passed in as
942 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
943 * 'hw_addr_iface' must be passed in as a null pointer. */
945 bridge_pick_datapath_id(struct bridge *br,
946 const uint8_t bridge_ea[ETH_ADDR_LEN],
947 struct iface *hw_addr_iface)
950 * The procedure for choosing a bridge MAC address will, in the most
951 * ordinary case, also choose a unique MAC that we can use as a datapath
952 * ID. In some special cases, though, multiple bridges will end up with
953 * the same MAC address. This is OK for the bridges, but it will confuse
954 * the OpenFlow controller, because each datapath needs a unique datapath
957 * Datapath IDs must be unique. It is also very desirable that they be
958 * stable from one run to the next, so that policy set on a datapath
961 const char *datapath_id;
964 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
965 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
971 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
973 * A bridge whose MAC address is taken from a VLAN network device
974 * (that is, a network device created with vconfig(8) or similar
975 * tool) will have the same MAC address as a bridge on the VLAN
976 * device's physical network device.
978 * Handle this case by hashing the physical network device MAC
979 * along with the VLAN identifier.
981 uint8_t buf[ETH_ADDR_LEN + 2];
982 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
983 buf[ETH_ADDR_LEN] = vlan >> 8;
984 buf[ETH_ADDR_LEN + 1] = vlan;
985 return dpid_from_hash(buf, sizeof buf);
988 * Assume that this bridge's MAC address is unique, since it
989 * doesn't fit any of the cases we handle specially.
994 * A purely internal bridge, that is, one that has no non-virtual
995 * network devices on it at all, is more difficult because it has no
996 * natural unique identifier at all.
998 * When the host is a XenServer, we handle this case by hashing the
999 * host's UUID with the name of the bridge. Names of bridges are
1000 * persistent across XenServer reboots, although they can be reused if
1001 * an internal network is destroyed and then a new one is later
1002 * created, so this is fairly effective.
1004 * When the host is not a XenServer, we punt by using a random MAC
1005 * address on each run.
1007 const char *host_uuid = xenserver_get_host_uuid();
1009 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1010 dpid = dpid_from_hash(combined, strlen(combined));
1016 return eth_addr_to_uint64(bridge_ea);
1020 dpid_from_hash(const void *data, size_t n)
1022 uint8_t hash[SHA1_DIGEST_SIZE];
1024 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1025 sha1_bytes(data, n, hash);
1026 eth_addr_mark_random(hash);
1027 return eth_addr_to_uint64(hash);
1033 struct bridge *br, *next;
1037 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1038 int error = bridge_run_one(br);
1040 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1041 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1042 "forcing reconfiguration", br->name);
1056 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1057 ofproto_wait(br->ofproto);
1058 if (br->controller) {
1062 mac_learning_wait(br->ml);
1067 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1068 * configuration changes. */
1070 bridge_flush(struct bridge *br)
1072 COVERAGE_INC(bridge_flush);
1074 mac_learning_flush(br->ml);
1077 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1078 * such interface. */
1079 static struct iface *
1080 bridge_get_local_iface(struct bridge *br)
1084 for (i = 0; i < br->n_ports; i++) {
1085 struct port *port = br->ports[i];
1086 for (j = 0; j < port->n_ifaces; j++) {
1087 struct iface *iface = port->ifaces[j];
1088 if (iface->dp_ifidx == ODPP_LOCAL) {
1097 /* Bridge unixctl user interface functions. */
1099 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1100 const char *args, void *aux UNUSED)
1102 struct ds ds = DS_EMPTY_INITIALIZER;
1103 const struct bridge *br;
1104 const struct mac_entry *e;
1106 br = bridge_lookup(args);
1108 unixctl_command_reply(conn, 501, "no such bridge");
1112 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1113 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1114 if (e->port < 0 || e->port >= br->n_ports) {
1117 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1118 br->ports[e->port]->ifaces[0]->dp_ifidx,
1119 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1121 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1125 /* Bridge reconfiguration functions. */
1126 static struct bridge *
1127 bridge_create(const struct ovsrec_bridge *br_cfg)
1132 assert(!bridge_lookup(br_cfg->name));
1133 br = xzalloc(sizeof *br);
1135 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1141 dpif_flow_flush(br->dpif);
1143 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1146 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1148 dpif_delete(br->dpif);
1149 dpif_close(br->dpif);
1154 br->name = xstrdup(br_cfg->name);
1156 br->ml = mac_learning_create();
1157 br->sent_config_request = false;
1158 eth_addr_random(br->default_ea);
1160 port_array_init(&br->ifaces);
1163 br->bond_next_rebalance = time_msec() + 10000;
1165 list_push_back(&all_bridges, &br->node);
1167 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1173 bridge_destroy(struct bridge *br)
1178 while (br->n_ports > 0) {
1179 port_destroy(br->ports[br->n_ports - 1]);
1181 list_remove(&br->node);
1182 error = dpif_delete(br->dpif);
1183 if (error && error != ENOENT) {
1184 VLOG_ERR("failed to delete %s: %s",
1185 dpif_name(br->dpif), strerror(error));
1187 dpif_close(br->dpif);
1188 ofproto_destroy(br->ofproto);
1189 free(br->controller);
1190 mac_learning_destroy(br->ml);
1191 port_array_destroy(&br->ifaces);
1198 static struct bridge *
1199 bridge_lookup(const char *name)
1203 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1204 if (!strcmp(br->name, name)) {
1212 bridge_exists(const char *name)
1214 return bridge_lookup(name) ? true : false;
1218 bridge_get_datapathid(const char *name)
1220 struct bridge *br = bridge_lookup(name);
1221 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1224 /* Handle requests for a listing of all flows known by the OpenFlow
1225 * stack, including those normally hidden. */
1227 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1228 const char *args, void *aux UNUSED)
1233 br = bridge_lookup(args);
1235 unixctl_command_reply(conn, 501, "Unknown bridge");
1240 ofproto_get_all_flows(br->ofproto, &results);
1242 unixctl_command_reply(conn, 200, ds_cstr(&results));
1243 ds_destroy(&results);
1247 bridge_run_one(struct bridge *br)
1251 error = ofproto_run1(br->ofproto);
1256 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1259 error = ofproto_run2(br->ofproto, br->flush);
1265 static const struct ovsrec_controller *
1266 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1267 const struct bridge *br)
1269 const struct ovsrec_controller *controller;
1271 controller = (br->cfg->controller ? br->cfg->controller
1272 : ovs_cfg->controller ? ovs_cfg->controller
1275 if (controller && !strcmp(controller->target, "none")) {
1283 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1285 struct svec *ifaces = ifaces_;
1286 if (!svec_contains(ifaces, iface->name)) {
1287 svec_add(ifaces, iface->name);
1291 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1293 br->name, iface->name, iface->port->name);
1299 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1302 struct shash old_ports, new_ports;
1304 struct svec listeners, old_listeners;
1305 struct svec snoops, old_snoops;
1306 struct shash_node *node;
1309 /* Collect old ports. */
1310 shash_init(&old_ports);
1311 for (i = 0; i < br->n_ports; i++) {
1312 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1315 /* Collect new ports. */
1316 shash_init(&new_ports);
1317 for (i = 0; i < br->cfg->n_ports; i++) {
1318 const char *name = br->cfg->ports[i]->name;
1319 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1320 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1325 /* If we have a controller, then we need a local port. Complain if the
1326 * user didn't specify one.
1328 * XXX perhaps we should synthesize a port ourselves in this case. */
1329 if (bridge_get_controller(ovs_cfg, br)) {
1330 char local_name[IF_NAMESIZE];
1333 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1334 local_name, sizeof local_name);
1335 if (!error && !shash_find(&new_ports, local_name)) {
1336 VLOG_WARN("bridge %s: controller specified but no local port "
1337 "(port named %s) defined",
1338 br->name, local_name);
1342 /* Get rid of deleted ports and add new ports. */
1343 SHASH_FOR_EACH (node, &old_ports) {
1344 if (!shash_find(&new_ports, node->name)) {
1345 port_destroy(node->data);
1348 SHASH_FOR_EACH (node, &new_ports) {
1349 struct port *port = shash_find_data(&old_ports, node->name);
1351 port = port_create(br, node->name);
1353 port_reconfigure(port, node->data);
1355 shash_destroy(&old_ports);
1356 shash_destroy(&new_ports);
1358 /* Check and delete duplicate interfaces. */
1360 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1361 svec_destroy(&ifaces);
1363 /* Delete all flows if we're switching from connected to standalone or vice
1364 * versa. (XXX Should we delete all flows if we are switching from one
1365 * controller to another?) */
1368 /* Configure OpenFlow management listeners. */
1369 svec_init(&listeners);
1370 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1372 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1373 ovs_rundir, br->name));
1374 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1375 svec_clear(&listeners);
1377 svec_sort_unique(&listeners);
1379 svec_init(&old_listeners);
1380 ofproto_get_listeners(br->ofproto, &old_listeners);
1381 svec_sort_unique(&old_listeners);
1383 if (!svec_equal(&listeners, &old_listeners)) {
1384 ofproto_set_listeners(br->ofproto, &listeners);
1386 svec_destroy(&listeners);
1387 svec_destroy(&old_listeners);
1389 /* Configure OpenFlow controller connection snooping. */
1391 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1393 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1394 ovs_rundir, br->name));
1395 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1396 svec_clear(&snoops);
1398 svec_sort_unique(&snoops);
1400 svec_init(&old_snoops);
1401 ofproto_get_snoops(br->ofproto, &old_snoops);
1402 svec_sort_unique(&old_snoops);
1404 if (!svec_equal(&snoops, &old_snoops)) {
1405 ofproto_set_snoops(br->ofproto, &snoops);
1407 svec_destroy(&snoops);
1408 svec_destroy(&old_snoops);
1410 /* Default listener. */
1411 svec_init(&listeners);
1412 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1413 ovs_rundir, br->name));
1414 svec_init(&old_listeners);
1415 ofproto_get_listeners(br->ofproto, &old_listeners);
1416 if (!svec_equal(&listeners, &old_listeners)) {
1417 ofproto_set_listeners(br->ofproto, &listeners);
1419 svec_destroy(&listeners);
1420 svec_destroy(&old_listeners);
1422 /* Default snoop. */
1424 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1425 ovs_rundir, br->name));
1426 svec_init(&old_snoops);
1427 ofproto_get_snoops(br->ofproto, &old_snoops);
1428 if (!svec_equal(&snoops, &old_snoops)) {
1429 ofproto_set_snoops(br->ofproto, &snoops);
1431 svec_destroy(&snoops);
1432 svec_destroy(&old_snoops);
1435 mirror_reconfigure(br);
1439 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1442 char *pfx = xasprintf("bridge.%s.controller", br->name);
1443 const struct ovsrec_controller *c;
1445 c = bridge_get_controller(ovs_cfg, br);
1446 if ((br->controller != NULL) != (c != NULL)) {
1447 ofproto_flush_flows(br->ofproto);
1449 free(br->controller);
1450 br->controller = c ? xstrdup(c->target) : NULL;
1453 int max_backoff, probe;
1454 int rate_limit, burst_limit;
1456 if (!strcmp(c->target, "discover")) {
1457 ofproto_set_discovery(br->ofproto, true,
1458 c->discover_accept_regex,
1459 c->discover_update_resolv_conf);
1461 struct iface *local_iface;
1465 in_band = (!c->connection_mode
1466 || !strcmp(c->connection_mode, "out-of-band"));
1467 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1468 ofproto_set_in_band(br->ofproto, in_band);
1470 local_iface = bridge_get_local_iface(br);
1471 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1472 struct netdev *netdev = local_iface->netdev;
1473 struct in_addr ip, mask, gateway;
1475 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1478 if (!c->local_gateway
1479 || !inet_aton(c->local_gateway, &gateway)) {
1483 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1485 mask.s_addr = guess_netmask(ip.s_addr);
1487 if (!netdev_set_in4(netdev, ip, mask)) {
1488 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1490 br->name, IP_ARGS(&ip.s_addr),
1491 IP_ARGS(&mask.s_addr));
1494 if (gateway.s_addr) {
1495 if (!netdev_add_router(netdev, gateway)) {
1496 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1497 br->name, IP_ARGS(&gateway.s_addr));
1503 ofproto_set_failure(br->ofproto,
1505 || !strcmp(c->fail_mode, "standalone")
1506 || !strcmp(c->fail_mode, "open")));
1508 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1509 ofproto_set_probe_interval(br->ofproto, probe);
1511 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1512 ofproto_set_max_backoff(br->ofproto, max_backoff);
1514 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1515 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1516 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1518 union ofp_action action;
1521 /* Set up a flow that matches every packet and directs them to
1522 * OFPP_NORMAL (which goes to us). */
1523 memset(&action, 0, sizeof action);
1524 action.type = htons(OFPAT_OUTPUT);
1525 action.output.len = htons(sizeof action);
1526 action.output.port = htons(OFPP_NORMAL);
1527 memset(&flow, 0, sizeof flow);
1528 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1531 ofproto_set_in_band(br->ofproto, false);
1532 ofproto_set_max_backoff(br->ofproto, 1);
1533 ofproto_set_probe_interval(br->ofproto, 5);
1534 ofproto_set_failure(br->ofproto, false);
1538 ofproto_set_controller(br->ofproto, br->controller);
1542 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1547 for (i = 0; i < br->n_ports; i++) {
1548 struct port *port = br->ports[i];
1549 for (j = 0; j < port->n_ifaces; j++) {
1550 struct iface *iface = port->ifaces[j];
1551 shash_add_once(ifaces, iface->name, iface);
1553 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1554 shash_add_once(ifaces, port->name, NULL);
1559 /* For robustness, in case the administrator moves around datapath ports behind
1560 * our back, we re-check all the datapath port numbers here.
1562 * This function will set the 'dp_ifidx' members of interfaces that have
1563 * disappeared to -1, so only call this function from a context where those
1564 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1565 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1566 * datapath, which doesn't support UINT16_MAX+1 ports. */
1568 bridge_fetch_dp_ifaces(struct bridge *br)
1570 struct odp_port *dpif_ports;
1571 size_t n_dpif_ports;
1574 /* Reset all interface numbers. */
1575 for (i = 0; i < br->n_ports; i++) {
1576 struct port *port = br->ports[i];
1577 for (j = 0; j < port->n_ifaces; j++) {
1578 struct iface *iface = port->ifaces[j];
1579 iface->dp_ifidx = -1;
1582 port_array_clear(&br->ifaces);
1584 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1585 for (i = 0; i < n_dpif_ports; i++) {
1586 struct odp_port *p = &dpif_ports[i];
1587 struct iface *iface = iface_lookup(br, p->devname);
1589 if (iface->dp_ifidx >= 0) {
1590 VLOG_WARN("%s reported interface %s twice",
1591 dpif_name(br->dpif), p->devname);
1592 } else if (iface_from_dp_ifidx(br, p->port)) {
1593 VLOG_WARN("%s reported interface %"PRIu16" twice",
1594 dpif_name(br->dpif), p->port);
1596 port_array_set(&br->ifaces, p->port, iface);
1597 iface->dp_ifidx = p->port;
1601 int64_t ofport = (iface->dp_ifidx >= 0
1602 ? odp_port_to_ofp_port(iface->dp_ifidx)
1604 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1611 /* Bridge packet processing functions. */
1614 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1616 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1619 static struct bond_entry *
1620 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1622 return &port->bond_hash[bond_hash(mac)];
1626 bond_choose_iface(const struct port *port)
1628 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1629 size_t i, best_down_slave = -1;
1630 long long next_delay_expiration = LLONG_MAX;
1632 for (i = 0; i < port->n_ifaces; i++) {
1633 struct iface *iface = port->ifaces[i];
1635 if (iface->enabled) {
1637 } else if (iface->delay_expires < next_delay_expiration) {
1638 best_down_slave = i;
1639 next_delay_expiration = iface->delay_expires;
1643 if (best_down_slave != -1) {
1644 struct iface *iface = port->ifaces[best_down_slave];
1646 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1647 "since no other interface is up", iface->name,
1648 iface->delay_expires - time_msec());
1649 bond_enable_slave(iface, true);
1652 return best_down_slave;
1656 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1657 uint16_t *dp_ifidx, tag_type *tags)
1659 struct iface *iface;
1661 assert(port->n_ifaces);
1662 if (port->n_ifaces == 1) {
1663 iface = port->ifaces[0];
1665 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1666 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1667 || !port->ifaces[e->iface_idx]->enabled) {
1668 /* XXX select interface properly. The current interface selection
1669 * is only good for testing the rebalancing code. */
1670 e->iface_idx = bond_choose_iface(port);
1671 if (e->iface_idx < 0) {
1672 *tags |= port->no_ifaces_tag;
1675 e->iface_tag = tag_create_random();
1676 ((struct port *) port)->bond_compat_is_stale = true;
1678 *tags |= e->iface_tag;
1679 iface = port->ifaces[e->iface_idx];
1681 *dp_ifidx = iface->dp_ifidx;
1682 *tags |= iface->tag; /* Currently only used for bonding. */
1687 bond_link_status_update(struct iface *iface, bool carrier)
1689 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1690 struct port *port = iface->port;
1692 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1693 /* Nothing to do. */
1696 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1697 iface->name, carrier ? "detected" : "dropped");
1698 if (carrier == iface->enabled) {
1699 iface->delay_expires = LLONG_MAX;
1700 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1701 iface->name, carrier ? "disabled" : "enabled");
1702 } else if (carrier && port->active_iface < 0) {
1703 bond_enable_slave(iface, true);
1704 if (port->updelay) {
1705 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1706 "other interface is up", iface->name, port->updelay);
1709 int delay = carrier ? port->updelay : port->downdelay;
1710 iface->delay_expires = time_msec() + delay;
1713 "interface %s: will be %s if it stays %s for %d ms",
1715 carrier ? "enabled" : "disabled",
1716 carrier ? "up" : "down",
1723 bond_choose_active_iface(struct port *port)
1725 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1727 port->active_iface = bond_choose_iface(port);
1728 port->active_iface_tag = tag_create_random();
1729 if (port->active_iface >= 0) {
1730 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1731 port->name, port->ifaces[port->active_iface]->name);
1733 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1739 bond_enable_slave(struct iface *iface, bool enable)
1741 struct port *port = iface->port;
1742 struct bridge *br = port->bridge;
1744 /* This acts as a recursion check. If the act of disabling a slave
1745 * causes a different slave to be enabled, the flag will allow us to
1746 * skip redundant work when we reenter this function. It must be
1747 * cleared on exit to keep things safe with multiple bonds. */
1748 static bool moving_active_iface = false;
1750 iface->delay_expires = LLONG_MAX;
1751 if (enable == iface->enabled) {
1755 iface->enabled = enable;
1756 if (!iface->enabled) {
1757 VLOG_WARN("interface %s: disabled", iface->name);
1758 ofproto_revalidate(br->ofproto, iface->tag);
1759 if (iface->port_ifidx == port->active_iface) {
1760 ofproto_revalidate(br->ofproto,
1761 port->active_iface_tag);
1763 /* Disabling a slave can lead to another slave being immediately
1764 * enabled if there will be no active slaves but one is waiting
1765 * on an updelay. In this case we do not need to run most of the
1766 * code for the newly enabled slave since there was no period
1767 * without an active slave and it is redundant with the disabling
1769 moving_active_iface = true;
1770 bond_choose_active_iface(port);
1772 bond_send_learning_packets(port);
1774 VLOG_WARN("interface %s: enabled", iface->name);
1775 if (port->active_iface < 0 && !moving_active_iface) {
1776 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1777 bond_choose_active_iface(port);
1778 bond_send_learning_packets(port);
1780 iface->tag = tag_create_random();
1783 moving_active_iface = false;
1784 port->bond_compat_is_stale = true;
1788 bond_run(struct bridge *br)
1792 for (i = 0; i < br->n_ports; i++) {
1793 struct port *port = br->ports[i];
1795 if (port->n_ifaces >= 2) {
1796 for (j = 0; j < port->n_ifaces; j++) {
1797 struct iface *iface = port->ifaces[j];
1798 if (time_msec() >= iface->delay_expires) {
1799 bond_enable_slave(iface, !iface->enabled);
1804 if (port->bond_compat_is_stale) {
1805 port->bond_compat_is_stale = false;
1806 port_update_bond_compat(port);
1812 bond_wait(struct bridge *br)
1816 for (i = 0; i < br->n_ports; i++) {
1817 struct port *port = br->ports[i];
1818 if (port->n_ifaces < 2) {
1821 for (j = 0; j < port->n_ifaces; j++) {
1822 struct iface *iface = port->ifaces[j];
1823 if (iface->delay_expires != LLONG_MAX) {
1824 poll_timer_wait(iface->delay_expires - time_msec());
1831 set_dst(struct dst *p, const flow_t *flow,
1832 const struct port *in_port, const struct port *out_port,
1835 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1836 : in_port->vlan >= 0 ? in_port->vlan
1837 : ntohs(flow->dl_vlan));
1838 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1842 swap_dst(struct dst *p, struct dst *q)
1844 struct dst tmp = *p;
1849 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1850 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1851 * that we push to the datapath. We could in fact fully sort the array by
1852 * vlan, but in most cases there are at most two different vlan tags so that's
1853 * possibly overkill.) */
1855 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1857 struct dst *first = dsts;
1858 struct dst *last = dsts + n_dsts;
1860 while (first != last) {
1862 * - All dsts < first have vlan == 'vlan'.
1863 * - All dsts >= last have vlan != 'vlan'.
1864 * - first < last. */
1865 while (first->vlan == vlan) {
1866 if (++first == last) {
1871 /* Same invariants, plus one additional:
1872 * - first->vlan != vlan.
1874 while (last[-1].vlan != vlan) {
1875 if (--last == first) {
1880 /* Same invariants, plus one additional:
1881 * - last[-1].vlan == vlan.*/
1882 swap_dst(first++, --last);
1887 mirror_mask_ffs(mirror_mask_t mask)
1889 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1894 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1895 const struct dst *test)
1898 for (i = 0; i < n_dsts; i++) {
1899 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1907 port_trunks_vlan(const struct port *port, uint16_t vlan)
1909 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1913 port_includes_vlan(const struct port *port, uint16_t vlan)
1915 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1919 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1920 const struct port *in_port, const struct port *out_port,
1921 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1923 mirror_mask_t mirrors = in_port->src_mirrors;
1924 struct dst *dst = dsts;
1927 if (out_port == FLOOD_PORT) {
1928 /* XXX use ODP_FLOOD if no vlans or bonding. */
1929 /* XXX even better, define each VLAN as a datapath port group */
1930 for (i = 0; i < br->n_ports; i++) {
1931 struct port *port = br->ports[i];
1932 if (port != in_port && port_includes_vlan(port, vlan)
1933 && !port->is_mirror_output_port
1934 && set_dst(dst, flow, in_port, port, tags)) {
1935 mirrors |= port->dst_mirrors;
1939 *nf_output_iface = NF_OUT_FLOOD;
1940 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1941 *nf_output_iface = dst->dp_ifidx;
1942 mirrors |= out_port->dst_mirrors;
1947 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1948 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1950 if (set_dst(dst, flow, in_port, m->out_port, tags)
1951 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1955 for (i = 0; i < br->n_ports; i++) {
1956 struct port *port = br->ports[i];
1957 if (port_includes_vlan(port, m->out_vlan)
1958 && set_dst(dst, flow, in_port, port, tags))
1962 if (port->vlan < 0) {
1963 dst->vlan = m->out_vlan;
1965 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1969 /* Use the vlan tag on the original flow instead of
1970 * the one passed in the vlan parameter. This ensures
1971 * that we compare the vlan from before any implicit
1972 * tagging tags place. This is necessary because
1973 * dst->vlan is the final vlan, after removing implicit
1975 flow_vlan = ntohs(flow->dl_vlan);
1976 if (flow_vlan == 0) {
1977 flow_vlan = OFP_VLAN_NONE;
1979 if (port == in_port && dst->vlan == flow_vlan) {
1980 /* Don't send out input port on same VLAN. */
1988 mirrors &= mirrors - 1;
1991 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1996 print_dsts(const struct dst *dsts, size_t n)
1998 for (; n--; dsts++) {
1999 printf(">p%"PRIu16, dsts->dp_ifidx);
2000 if (dsts->vlan != OFP_VLAN_NONE) {
2001 printf("v%"PRIu16, dsts->vlan);
2007 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2008 const struct port *in_port, const struct port *out_port,
2009 tag_type *tags, struct odp_actions *actions,
2010 uint16_t *nf_output_iface)
2012 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2014 const struct dst *p;
2017 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2020 cur_vlan = ntohs(flow->dl_vlan);
2021 for (p = dsts; p < &dsts[n_dsts]; p++) {
2022 union odp_action *a;
2023 if (p->vlan != cur_vlan) {
2024 if (p->vlan == OFP_VLAN_NONE) {
2025 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2027 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2028 a->vlan_vid.vlan_vid = htons(p->vlan);
2032 a = odp_actions_add(actions, ODPAT_OUTPUT);
2033 a->output.port = p->dp_ifidx;
2037 /* Returns the effective vlan of a packet, taking into account both the
2038 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2039 * the packet is untagged and -1 indicates it has an invalid header and
2040 * should be dropped. */
2041 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2042 struct port *in_port, bool have_packet)
2044 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2045 * belongs to VLAN 0, so we should treat both cases identically. (In the
2046 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2047 * presumably to allow a priority to be specified. In the latter case, the
2048 * packet does not have any 802.1Q header.) */
2049 int vlan = ntohs(flow->dl_vlan);
2050 if (vlan == OFP_VLAN_NONE) {
2053 if (in_port->vlan >= 0) {
2055 /* XXX support double tagging? */
2057 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2058 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2059 "packet received on port %s configured with "
2060 "implicit VLAN %"PRIu16,
2061 br->name, ntohs(flow->dl_vlan),
2062 in_port->name, in_port->vlan);
2066 vlan = in_port->vlan;
2068 if (!port_includes_vlan(in_port, vlan)) {
2070 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2071 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2072 "packet received on port %s not configured for "
2074 br->name, vlan, in_port->name, vlan);
2084 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2085 struct port *in_port)
2087 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2088 vlan, in_port->port_idx);
2090 /* The log messages here could actually be useful in debugging,
2091 * so keep the rate limit relatively high. */
2092 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2094 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2095 "on port %s in VLAN %d",
2096 br->name, ETH_ADDR_ARGS(flow->dl_src),
2097 in_port->name, vlan);
2098 ofproto_revalidate(br->ofproto, rev_tag);
2103 is_bcast_arp_reply(const flow_t *flow)
2105 return (flow->dl_type == htons(ETH_TYPE_ARP)
2106 && flow->nw_proto == ARP_OP_REPLY
2107 && eth_addr_is_broadcast(flow->dl_dst));
2110 /* If the composed actions may be applied to any packet in the given 'flow',
2111 * returns true. Otherwise, the actions should only be applied to 'packet', or
2112 * not at all, if 'packet' was NULL. */
2114 process_flow(struct bridge *br, const flow_t *flow,
2115 const struct ofpbuf *packet, struct odp_actions *actions,
2116 tag_type *tags, uint16_t *nf_output_iface)
2118 struct iface *in_iface;
2119 struct port *in_port;
2120 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2124 /* Find the interface and port structure for the received packet. */
2125 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2127 /* No interface? Something fishy... */
2128 if (packet != NULL) {
2129 /* Odd. A few possible reasons here:
2131 * - We deleted an interface but there are still a few packets
2132 * queued up from it.
2134 * - Someone externally added an interface (e.g. with "ovs-dpctl
2135 * add-if") that we don't know about.
2137 * - Packet arrived on the local port but the local port is not
2138 * one of our bridge ports.
2140 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2142 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2143 "interface %"PRIu16, br->name, flow->in_port);
2146 /* Return without adding any actions, to drop packets on this flow. */
2149 in_port = in_iface->port;
2150 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2155 /* Drop frames for reserved multicast addresses. */
2156 if (eth_addr_is_reserved(flow->dl_dst)) {
2160 /* Drop frames on ports reserved for mirroring. */
2161 if (in_port->is_mirror_output_port) {
2162 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2163 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2164 "which is reserved exclusively for mirroring",
2165 br->name, in_port->name);
2169 /* Packets received on bonds need special attention to avoid duplicates. */
2170 if (in_port->n_ifaces > 1) {
2173 if (eth_addr_is_multicast(flow->dl_dst)) {
2174 *tags |= in_port->active_iface_tag;
2175 if (in_port->active_iface != in_iface->port_ifidx) {
2176 /* Drop all multicast packets on inactive slaves. */
2181 /* Drop all packets for which we have learned a different input
2182 * port, because we probably sent the packet on one slave and got
2183 * it back on the other. Broadcast ARP replies are an exception
2184 * to this rule: the host has moved to another switch. */
2185 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2186 if (src_idx != -1 && src_idx != in_port->port_idx &&
2187 !is_bcast_arp_reply(flow)) {
2193 out_port = FLOOD_PORT;
2194 /* Learn source MAC (but don't try to learn from revalidation). */
2196 update_learning_table(br, flow, vlan, in_port);
2199 /* Determine output port. */
2200 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2202 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2203 out_port = br->ports[out_port_idx];
2204 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2205 /* If we are revalidating but don't have a learning entry then
2206 * eject the flow. Installing a flow that floods packets opens
2207 * up a window of time where we could learn from a packet reflected
2208 * on a bond and blackhole packets before the learning table is
2209 * updated to reflect the correct port. */
2213 /* Don't send packets out their input ports. */
2214 if (in_port == out_port) {
2219 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2225 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2228 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2229 const struct ofp_phy_port *opp,
2232 struct bridge *br = br_;
2233 struct iface *iface;
2236 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2242 if (reason == OFPPR_DELETE) {
2243 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2244 br->name, iface->name);
2245 iface_destroy(iface);
2246 if (!port->n_ifaces) {
2247 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2248 br->name, port->name);
2254 if (port->n_ifaces > 1) {
2255 bool up = !(opp->state & OFPPS_LINK_DOWN);
2256 bond_link_status_update(iface, up);
2257 port_update_bond_compat(port);
2263 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2264 struct odp_actions *actions, tag_type *tags,
2265 uint16_t *nf_output_iface, void *br_)
2267 struct bridge *br = br_;
2269 COVERAGE_INC(bridge_process_flow);
2270 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2274 bridge_account_flow_ofhook_cb(const flow_t *flow,
2275 const union odp_action *actions,
2276 size_t n_actions, unsigned long long int n_bytes,
2279 struct bridge *br = br_;
2280 struct port *in_port;
2281 const union odp_action *a;
2283 /* Feed information from the active flows back into the learning table
2284 * to ensure that table is always in sync with what is actually flowing
2285 * through the datapath. */
2286 in_port = port_from_dp_ifidx(br, flow->in_port);
2288 int vlan = flow_get_vlan(br, flow, in_port, false);
2290 update_learning_table(br, flow, vlan, in_port);
2294 if (!br->has_bonded_ports) {
2298 for (a = actions; a < &actions[n_actions]; a++) {
2299 if (a->type == ODPAT_OUTPUT) {
2300 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2301 if (out_port && out_port->n_ifaces >= 2) {
2302 struct bond_entry *e = lookup_bond_entry(out_port,
2304 e->tx_bytes += n_bytes;
2311 bridge_account_checkpoint_ofhook_cb(void *br_)
2313 struct bridge *br = br_;
2316 if (!br->has_bonded_ports) {
2320 /* The current ofproto implementation calls this callback at least once a
2321 * second, so this timer implementation is sufficient. */
2322 if (time_msec() < br->bond_next_rebalance) {
2325 br->bond_next_rebalance = time_msec() + 10000;
2327 for (i = 0; i < br->n_ports; i++) {
2328 struct port *port = br->ports[i];
2329 if (port->n_ifaces > 1) {
2330 bond_rebalance_port(port);
2335 static struct ofhooks bridge_ofhooks = {
2336 bridge_port_changed_ofhook_cb,
2337 bridge_normal_ofhook_cb,
2338 bridge_account_flow_ofhook_cb,
2339 bridge_account_checkpoint_ofhook_cb,
2342 /* Bonding functions. */
2344 /* Statistics for a single interface on a bonded port, used for load-based
2345 * bond rebalancing. */
2346 struct slave_balance {
2347 struct iface *iface; /* The interface. */
2348 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2350 /* All the "bond_entry"s that are assigned to this interface, in order of
2351 * increasing tx_bytes. */
2352 struct bond_entry **hashes;
2356 /* Sorts pointers to pointers to bond_entries in ascending order by the
2357 * interface to which they are assigned, and within a single interface in
2358 * ascending order of bytes transmitted. */
2360 compare_bond_entries(const void *a_, const void *b_)
2362 const struct bond_entry *const *ap = a_;
2363 const struct bond_entry *const *bp = b_;
2364 const struct bond_entry *a = *ap;
2365 const struct bond_entry *b = *bp;
2366 if (a->iface_idx != b->iface_idx) {
2367 return a->iface_idx > b->iface_idx ? 1 : -1;
2368 } else if (a->tx_bytes != b->tx_bytes) {
2369 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2375 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2376 * *descending* order by number of bytes transmitted. */
2378 compare_slave_balance(const void *a_, const void *b_)
2380 const struct slave_balance *a = a_;
2381 const struct slave_balance *b = b_;
2382 if (a->iface->enabled != b->iface->enabled) {
2383 return a->iface->enabled ? -1 : 1;
2384 } else if (a->tx_bytes != b->tx_bytes) {
2385 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2392 swap_bals(struct slave_balance *a, struct slave_balance *b)
2394 struct slave_balance tmp = *a;
2399 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2400 * given that 'p' (and only 'p') might be in the wrong location.
2402 * This function invalidates 'p', since it might now be in a different memory
2405 resort_bals(struct slave_balance *p,
2406 struct slave_balance bals[], size_t n_bals)
2409 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2410 swap_bals(p, p - 1);
2412 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2413 swap_bals(p, p + 1);
2419 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2421 if (VLOG_IS_DBG_ENABLED()) {
2422 struct ds ds = DS_EMPTY_INITIALIZER;
2423 const struct slave_balance *b;
2425 for (b = bals; b < bals + n_bals; b++) {
2429 ds_put_char(&ds, ',');
2431 ds_put_format(&ds, " %s %"PRIu64"kB",
2432 b->iface->name, b->tx_bytes / 1024);
2434 if (!b->iface->enabled) {
2435 ds_put_cstr(&ds, " (disabled)");
2437 if (b->n_hashes > 0) {
2438 ds_put_cstr(&ds, " (");
2439 for (i = 0; i < b->n_hashes; i++) {
2440 const struct bond_entry *e = b->hashes[i];
2442 ds_put_cstr(&ds, " + ");
2444 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2445 e - port->bond_hash, e->tx_bytes / 1024);
2447 ds_put_cstr(&ds, ")");
2450 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2455 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2457 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2460 struct bond_entry *hash = from->hashes[hash_idx];
2461 struct port *port = from->iface->port;
2462 uint64_t delta = hash->tx_bytes;
2464 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2465 "from %s to %s (now carrying %"PRIu64"kB and "
2466 "%"PRIu64"kB load, respectively)",
2467 port->name, delta / 1024, hash - port->bond_hash,
2468 from->iface->name, to->iface->name,
2469 (from->tx_bytes - delta) / 1024,
2470 (to->tx_bytes + delta) / 1024);
2472 /* Delete element from from->hashes.
2474 * We don't bother to add the element to to->hashes because not only would
2475 * it require more work, the only purpose it would be to allow that hash to
2476 * be migrated to another slave in this rebalancing run, and there is no
2477 * point in doing that. */
2478 if (hash_idx == 0) {
2481 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2482 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2486 /* Shift load away from 'from' to 'to'. */
2487 from->tx_bytes -= delta;
2488 to->tx_bytes += delta;
2490 /* Arrange for flows to be revalidated. */
2491 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2492 hash->iface_idx = to->iface->port_ifidx;
2493 hash->iface_tag = tag_create_random();
2497 bond_rebalance_port(struct port *port)
2499 struct slave_balance bals[DP_MAX_PORTS];
2501 struct bond_entry *hashes[BOND_MASK + 1];
2502 struct slave_balance *b, *from, *to;
2503 struct bond_entry *e;
2506 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2507 * descending order of tx_bytes, so that bals[0] represents the most
2508 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2511 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2512 * array for each slave_balance structure, we sort our local array of
2513 * hashes in order by slave, so that all of the hashes for a given slave
2514 * become contiguous in memory, and then we point each 'hashes' members of
2515 * a slave_balance structure to the start of a contiguous group. */
2516 n_bals = port->n_ifaces;
2517 for (b = bals; b < &bals[n_bals]; b++) {
2518 b->iface = port->ifaces[b - bals];
2523 for (i = 0; i <= BOND_MASK; i++) {
2524 hashes[i] = &port->bond_hash[i];
2526 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2527 for (i = 0; i <= BOND_MASK; i++) {
2529 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2530 b = &bals[e->iface_idx];
2531 b->tx_bytes += e->tx_bytes;
2533 b->hashes = &hashes[i];
2538 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2539 log_bals(bals, n_bals, port);
2541 /* Discard slaves that aren't enabled (which were sorted to the back of the
2542 * array earlier). */
2543 while (!bals[n_bals - 1].iface->enabled) {
2550 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2551 to = &bals[n_bals - 1];
2552 for (from = bals; from < to; ) {
2553 uint64_t overload = from->tx_bytes - to->tx_bytes;
2554 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2555 /* The extra load on 'from' (and all less-loaded slaves), compared
2556 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2557 * it is less than ~1Mbps. No point in rebalancing. */
2559 } else if (from->n_hashes == 1) {
2560 /* 'from' only carries a single MAC hash, so we can't shift any
2561 * load away from it, even though we want to. */
2564 /* 'from' is carrying significantly more load than 'to', and that
2565 * load is split across at least two different hashes. Pick a hash
2566 * to migrate to 'to' (the least-loaded slave), given that doing so
2567 * must decrease the ratio of the load on the two slaves by at
2570 * The sort order we use means that we prefer to shift away the
2571 * smallest hashes instead of the biggest ones. There is little
2572 * reason behind this decision; we could use the opposite sort
2573 * order to shift away big hashes ahead of small ones. */
2577 for (i = 0; i < from->n_hashes; i++) {
2578 double old_ratio, new_ratio;
2579 uint64_t delta = from->hashes[i]->tx_bytes;
2581 if (delta == 0 || from->tx_bytes - delta == 0) {
2582 /* Pointless move. */
2586 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2588 if (to->tx_bytes == 0) {
2589 /* Nothing on the new slave, move it. */
2593 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2594 new_ratio = (double)(from->tx_bytes - delta) /
2595 (to->tx_bytes + delta);
2597 if (new_ratio == 0) {
2598 /* Should already be covered but check to prevent division
2603 if (new_ratio < 1) {
2604 new_ratio = 1 / new_ratio;
2607 if (old_ratio - new_ratio > 0.1) {
2608 /* Would decrease the ratio, move it. */
2612 if (i < from->n_hashes) {
2613 bond_shift_load(from, to, i);
2614 port->bond_compat_is_stale = true;
2616 /* If the result of the migration changed the relative order of
2617 * 'from' and 'to' swap them back to maintain invariants. */
2618 if (order_swapped) {
2619 swap_bals(from, to);
2622 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2623 * point to different slave_balance structures. It is only
2624 * valid to do these two operations in a row at all because we
2625 * know that 'from' will not move past 'to' and vice versa. */
2626 resort_bals(from, bals, n_bals);
2627 resort_bals(to, bals, n_bals);
2634 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2635 * historical data to decay to <1% in 7 rebalancing runs. */
2636 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2642 bond_send_learning_packets(struct port *port)
2644 struct bridge *br = port->bridge;
2645 struct mac_entry *e;
2646 struct ofpbuf packet;
2647 int error, n_packets, n_errors;
2649 if (!port->n_ifaces || port->active_iface < 0) {
2653 ofpbuf_init(&packet, 128);
2654 error = n_packets = n_errors = 0;
2655 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2656 union ofp_action actions[2], *a;
2662 if (e->port == port->port_idx
2663 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2667 /* Compose actions. */
2668 memset(actions, 0, sizeof actions);
2671 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2672 a->vlan_vid.len = htons(sizeof *a);
2673 a->vlan_vid.vlan_vid = htons(e->vlan);
2676 a->output.type = htons(OFPAT_OUTPUT);
2677 a->output.len = htons(sizeof *a);
2678 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2683 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2685 flow_extract(&packet, ODPP_NONE, &flow);
2686 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2693 ofpbuf_uninit(&packet);
2696 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2697 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2698 "packets, last error was: %s",
2699 port->name, n_errors, n_packets, strerror(error));
2701 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2702 port->name, n_packets);
2706 /* Bonding unixctl user interface functions. */
2709 bond_unixctl_list(struct unixctl_conn *conn,
2710 const char *args UNUSED, void *aux UNUSED)
2712 struct ds ds = DS_EMPTY_INITIALIZER;
2713 const struct bridge *br;
2715 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2717 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2720 for (i = 0; i < br->n_ports; i++) {
2721 const struct port *port = br->ports[i];
2722 if (port->n_ifaces > 1) {
2725 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2726 for (j = 0; j < port->n_ifaces; j++) {
2727 const struct iface *iface = port->ifaces[j];
2729 ds_put_cstr(&ds, ", ");
2731 ds_put_cstr(&ds, iface->name);
2733 ds_put_char(&ds, '\n');
2737 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2741 static struct port *
2742 bond_find(const char *name)
2744 const struct bridge *br;
2746 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2749 for (i = 0; i < br->n_ports; i++) {
2750 struct port *port = br->ports[i];
2751 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2760 bond_unixctl_show(struct unixctl_conn *conn,
2761 const char *args, void *aux UNUSED)
2763 struct ds ds = DS_EMPTY_INITIALIZER;
2764 const struct port *port;
2767 port = bond_find(args);
2769 unixctl_command_reply(conn, 501, "no such bond");
2773 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2774 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2775 ds_put_format(&ds, "next rebalance: %lld ms\n",
2776 port->bridge->bond_next_rebalance - time_msec());
2777 for (j = 0; j < port->n_ifaces; j++) {
2778 const struct iface *iface = port->ifaces[j];
2779 struct bond_entry *be;
2782 ds_put_format(&ds, "slave %s: %s\n",
2783 iface->name, iface->enabled ? "enabled" : "disabled");
2784 if (j == port->active_iface) {
2785 ds_put_cstr(&ds, "\tactive slave\n");
2787 if (iface->delay_expires != LLONG_MAX) {
2788 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2789 iface->enabled ? "downdelay" : "updelay",
2790 iface->delay_expires - time_msec());
2794 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2795 int hash = be - port->bond_hash;
2796 struct mac_entry *me;
2798 if (be->iface_idx != j) {
2802 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2803 hash, be->tx_bytes / 1024);
2806 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2807 &port->bridge->ml->lrus) {
2810 if (bond_hash(me->mac) == hash
2811 && me->port != port->port_idx
2812 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2813 && dp_ifidx == iface->dp_ifidx)
2815 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2816 ETH_ADDR_ARGS(me->mac));
2821 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2826 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2829 char *args = (char *) args_;
2830 char *save_ptr = NULL;
2831 char *bond_s, *hash_s, *slave_s;
2832 uint8_t mac[ETH_ADDR_LEN];
2834 struct iface *iface;
2835 struct bond_entry *entry;
2838 bond_s = strtok_r(args, " ", &save_ptr);
2839 hash_s = strtok_r(NULL, " ", &save_ptr);
2840 slave_s = strtok_r(NULL, " ", &save_ptr);
2842 unixctl_command_reply(conn, 501,
2843 "usage: bond/migrate BOND HASH SLAVE");
2847 port = bond_find(bond_s);
2849 unixctl_command_reply(conn, 501, "no such bond");
2853 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2854 == ETH_ADDR_SCAN_COUNT) {
2855 hash = bond_hash(mac);
2856 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2857 hash = atoi(hash_s) & BOND_MASK;
2859 unixctl_command_reply(conn, 501, "bad hash");
2863 iface = port_lookup_iface(port, slave_s);
2865 unixctl_command_reply(conn, 501, "no such slave");
2869 if (!iface->enabled) {
2870 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2874 entry = &port->bond_hash[hash];
2875 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2876 entry->iface_idx = iface->port_ifidx;
2877 entry->iface_tag = tag_create_random();
2878 port->bond_compat_is_stale = true;
2879 unixctl_command_reply(conn, 200, "migrated");
2883 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2886 char *args = (char *) args_;
2887 char *save_ptr = NULL;
2888 char *bond_s, *slave_s;
2890 struct iface *iface;
2892 bond_s = strtok_r(args, " ", &save_ptr);
2893 slave_s = strtok_r(NULL, " ", &save_ptr);
2895 unixctl_command_reply(conn, 501,
2896 "usage: bond/set-active-slave BOND SLAVE");
2900 port = bond_find(bond_s);
2902 unixctl_command_reply(conn, 501, "no such bond");
2906 iface = port_lookup_iface(port, slave_s);
2908 unixctl_command_reply(conn, 501, "no such slave");
2912 if (!iface->enabled) {
2913 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2917 if (port->active_iface != iface->port_ifidx) {
2918 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2919 port->active_iface = iface->port_ifidx;
2920 port->active_iface_tag = tag_create_random();
2921 VLOG_INFO("port %s: active interface is now %s",
2922 port->name, iface->name);
2923 bond_send_learning_packets(port);
2924 unixctl_command_reply(conn, 200, "done");
2926 unixctl_command_reply(conn, 200, "no change");
2931 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2933 char *args = (char *) args_;
2934 char *save_ptr = NULL;
2935 char *bond_s, *slave_s;
2937 struct iface *iface;
2939 bond_s = strtok_r(args, " ", &save_ptr);
2940 slave_s = strtok_r(NULL, " ", &save_ptr);
2942 unixctl_command_reply(conn, 501,
2943 "usage: bond/enable/disable-slave BOND SLAVE");
2947 port = bond_find(bond_s);
2949 unixctl_command_reply(conn, 501, "no such bond");
2953 iface = port_lookup_iface(port, slave_s);
2955 unixctl_command_reply(conn, 501, "no such slave");
2959 bond_enable_slave(iface, enable);
2960 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2964 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
2967 enable_slave(conn, args, true);
2971 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
2974 enable_slave(conn, args, false);
2978 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
2981 uint8_t mac[ETH_ADDR_LEN];
2985 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2986 == ETH_ADDR_SCAN_COUNT) {
2987 hash = bond_hash(mac);
2989 hash_cstr = xasprintf("%u", hash);
2990 unixctl_command_reply(conn, 200, hash_cstr);
2993 unixctl_command_reply(conn, 501, "invalid mac");
3000 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3001 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3002 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3003 unixctl_command_register("bond/set-active-slave",
3004 bond_unixctl_set_active_slave, NULL);
3005 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3007 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3009 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3012 /* Port functions. */
3014 static struct port *
3015 port_create(struct bridge *br, const char *name)
3019 port = xzalloc(sizeof *port);
3021 port->port_idx = br->n_ports;
3023 port->trunks = NULL;
3024 port->name = xstrdup(name);
3025 port->active_iface = -1;
3027 if (br->n_ports >= br->allocated_ports) {
3028 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3031 br->ports[br->n_ports++] = port;
3033 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3040 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3042 struct shash old_ifaces, new_ifaces;
3043 struct shash_node *node;
3044 unsigned long *trunks;
3050 /* Collect old and new interfaces. */
3051 shash_init(&old_ifaces);
3052 shash_init(&new_ifaces);
3053 for (i = 0; i < port->n_ifaces; i++) {
3054 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3056 for (i = 0; i < cfg->n_interfaces; i++) {
3057 const char *name = cfg->interfaces[i]->name;
3058 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3059 VLOG_WARN("port %s: %s specified twice as port interface",
3063 port->updelay = cfg->bond_updelay;
3064 if (port->updelay < 0) {
3067 port->updelay = cfg->bond_downdelay;
3068 if (port->downdelay < 0) {
3069 port->downdelay = 0;
3072 /* Get rid of deleted interfaces and add new interfaces. */
3073 SHASH_FOR_EACH (node, &old_ifaces) {
3074 if (!shash_find(&new_ifaces, node->name)) {
3075 iface_destroy(node->data);
3078 SHASH_FOR_EACH (node, &new_ifaces) {
3079 const struct ovsrec_interface *if_cfg = node->data;
3080 struct iface *iface;
3082 iface = shash_find_data(&old_ifaces, if_cfg->name);
3084 iface = iface_create(port, if_cfg);
3086 iface->cfg = if_cfg;
3093 if (port->n_ifaces < 2) {
3095 if (vlan >= 0 && vlan <= 4095) {
3096 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3101 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3102 * they even work as-is. But they have not been tested. */
3103 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3107 if (port->vlan != vlan) {
3109 bridge_flush(port->bridge);
3112 /* Get trunked VLANs. */
3118 trunks = bitmap_allocate(4096);
3120 for (i = 0; i < cfg->n_trunks; i++) {
3121 int trunk = cfg->trunks[i];
3123 bitmap_set1(trunks, trunk);
3129 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3130 port->name, cfg->n_trunks);
3132 if (n_errors == cfg->n_trunks) {
3134 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3137 bitmap_set_multiple(trunks, 0, 4096, 1);
3140 if (cfg->n_trunks) {
3141 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3146 ? port->trunks != NULL
3147 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3148 bridge_flush(port->bridge);
3150 bitmap_free(port->trunks);
3151 port->trunks = trunks;
3153 shash_destroy(&old_ifaces);
3154 shash_destroy(&new_ifaces);
3158 port_destroy(struct port *port)
3161 struct bridge *br = port->bridge;
3165 proc_net_compat_update_vlan(port->name, NULL, 0);
3166 proc_net_compat_update_bond(port->name, NULL);
3168 for (i = 0; i < MAX_MIRRORS; i++) {
3169 struct mirror *m = br->mirrors[i];
3170 if (m && m->out_port == port) {
3175 while (port->n_ifaces > 0) {
3176 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3179 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3180 del->port_idx = port->port_idx;
3183 bitmap_free(port->trunks);
3190 static struct port *
3191 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3193 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3194 return iface ? iface->port : NULL;
3197 static struct port *
3198 port_lookup(const struct bridge *br, const char *name)
3202 for (i = 0; i < br->n_ports; i++) {
3203 struct port *port = br->ports[i];
3204 if (!strcmp(port->name, name)) {
3211 static struct iface *
3212 port_lookup_iface(const struct port *port, const char *name)
3216 for (j = 0; j < port->n_ifaces; j++) {
3217 struct iface *iface = port->ifaces[j];
3218 if (!strcmp(iface->name, name)) {
3226 port_update_bonding(struct port *port)
3228 if (port->n_ifaces < 2) {
3229 /* Not a bonded port. */
3230 if (port->bond_hash) {
3231 free(port->bond_hash);
3232 port->bond_hash = NULL;
3233 port->bond_compat_is_stale = true;
3234 port->bond_fake_iface = false;
3237 if (!port->bond_hash) {
3240 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3241 for (i = 0; i <= BOND_MASK; i++) {
3242 struct bond_entry *e = &port->bond_hash[i];
3246 port->no_ifaces_tag = tag_create_random();
3247 bond_choose_active_iface(port);
3249 port->bond_compat_is_stale = true;
3250 port->bond_fake_iface = port->cfg->bond_fake_iface;
3255 port_update_bond_compat(struct port *port)
3257 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3258 struct compat_bond bond;
3261 if (port->n_ifaces < 2) {
3262 proc_net_compat_update_bond(port->name, NULL);
3267 bond.updelay = port->updelay;
3268 bond.downdelay = port->downdelay;
3271 bond.hashes = compat_hashes;
3272 if (port->bond_hash) {
3273 const struct bond_entry *e;
3274 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3275 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3276 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3277 cbh->hash = e - port->bond_hash;
3278 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3283 bond.n_slaves = port->n_ifaces;
3284 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3285 for (i = 0; i < port->n_ifaces; i++) {
3286 struct iface *iface = port->ifaces[i];
3287 struct compat_bond_slave *slave = &bond.slaves[i];
3288 slave->name = iface->name;
3290 /* We need to make the same determination as the Linux bonding
3291 * code to determine whether a slave should be consider "up".
3292 * The Linux function bond_miimon_inspect() supports four
3293 * BOND_LINK_* states:
3295 * - BOND_LINK_UP: carrier detected, updelay has passed.
3296 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3297 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3298 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3300 * The function bond_info_show_slave() only considers BOND_LINK_UP
3301 * to be "up" and anything else to be "down".
3303 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3307 netdev_get_etheraddr(iface->netdev, slave->mac);
3310 if (port->bond_fake_iface) {
3311 struct netdev *bond_netdev;
3313 if (!netdev_open_default(port->name, &bond_netdev)) {
3315 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3317 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3319 netdev_close(bond_netdev);
3323 proc_net_compat_update_bond(port->name, &bond);
3328 port_update_vlan_compat(struct port *port)
3330 struct bridge *br = port->bridge;
3331 char *vlandev_name = NULL;
3333 if (port->vlan > 0) {
3334 /* Figure out the name that the VLAN device should actually have, if it
3335 * existed. This takes some work because the VLAN device would not
3336 * have port->name in its name; rather, it would have the trunk port's
3337 * name, and 'port' would be attached to a bridge that also had the
3338 * VLAN device one of its ports. So we need to find a trunk port that
3339 * includes port->vlan.
3341 * There might be more than one candidate. This doesn't happen on
3342 * XenServer, so if it happens we just pick the first choice in
3343 * alphabetical order instead of creating multiple VLAN devices. */
3345 for (i = 0; i < br->n_ports; i++) {
3346 struct port *p = br->ports[i];
3347 if (port_trunks_vlan(p, port->vlan)
3349 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3351 uint8_t ea[ETH_ADDR_LEN];
3352 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3353 if (!eth_addr_is_multicast(ea) &&
3354 !eth_addr_is_reserved(ea) &&
3355 !eth_addr_is_zero(ea)) {
3356 vlandev_name = p->name;
3361 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3364 /* Interface functions. */
3366 static struct iface *
3367 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3369 struct iface *iface;
3370 char *name = if_cfg->name;
3373 iface = xzalloc(sizeof *iface);
3375 iface->port_ifidx = port->n_ifaces;
3376 iface->name = xstrdup(name);
3377 iface->dp_ifidx = -1;
3378 iface->tag = tag_create_random();
3379 iface->delay_expires = LLONG_MAX;
3380 iface->netdev = NULL;
3381 iface->cfg = if_cfg;
3383 if (port->n_ifaces >= port->allocated_ifaces) {
3384 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3385 sizeof *port->ifaces);
3387 port->ifaces[port->n_ifaces++] = iface;
3388 if (port->n_ifaces > 1) {
3389 port->bridge->has_bonded_ports = true;
3392 /* Attempt to create the network interface in case it
3393 * doesn't exist yet. */
3394 if (!iface_is_internal(port->bridge, iface->name)) {
3395 error = set_up_iface(if_cfg, iface, true);
3397 VLOG_WARN("could not create iface %s: %s", iface->name,
3402 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3404 bridge_flush(port->bridge);
3410 iface_destroy(struct iface *iface)
3413 struct port *port = iface->port;
3414 struct bridge *br = port->bridge;
3415 bool del_active = port->active_iface == iface->port_ifidx;
3418 if (iface->dp_ifidx >= 0) {
3419 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3422 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3423 del->port_ifidx = iface->port_ifidx;
3425 netdev_close(iface->netdev);
3428 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3429 bond_choose_active_iface(port);
3430 bond_send_learning_packets(port);
3436 bridge_flush(port->bridge);
3440 static struct iface *
3441 iface_lookup(const struct bridge *br, const char *name)
3445 for (i = 0; i < br->n_ports; i++) {
3446 struct port *port = br->ports[i];
3447 for (j = 0; j < port->n_ifaces; j++) {
3448 struct iface *iface = port->ifaces[j];
3449 if (!strcmp(iface->name, name)) {
3457 static struct iface *
3458 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3460 return port_array_get(&br->ifaces, dp_ifidx);
3463 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3464 * 'br', that is, an interface that is entirely simulated within the datapath.
3465 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3466 * interfaces are created by setting "iface.<iface>.internal = true".
3468 * In addition, we have a kluge-y feature that creates an internal port with
3469 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3470 * This feature needs to go away in the long term. Until then, this is one
3471 * reason why this function takes a name instead of a struct iface: the fake
3472 * interfaces created this way do not have a struct iface. */
3474 iface_is_internal(const struct bridge *br, const char *if_name)
3476 /* XXX wastes time */
3477 struct iface *iface;
3480 if (!strcmp(if_name, br->name)) {
3484 iface = iface_lookup(br, if_name);
3485 if (iface && !strcmp(iface->cfg->type, "internal")) {
3489 port = port_lookup(br, if_name);
3490 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3496 /* Set Ethernet address of 'iface', if one is specified in the configuration
3499 iface_set_mac(struct iface *iface)
3501 uint8_t ea[ETH_ADDR_LEN];
3503 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3504 if (eth_addr_is_multicast(ea)) {
3505 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3507 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3508 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3509 iface->name, iface->name);
3511 int error = netdev_set_etheraddr(iface->netdev, ea);
3513 VLOG_ERR("interface %s: setting MAC failed (%s)",
3514 iface->name, strerror(error));
3520 /* Port mirroring. */
3523 mirror_reconfigure(struct bridge *br)
3525 struct shash old_mirrors, new_mirrors;
3526 struct shash_node *node;
3527 unsigned long *rspan_vlans;
3530 /* Collect old mirrors. */
3531 shash_init(&old_mirrors);
3532 for (i = 0; i < MAX_MIRRORS; i++) {
3533 if (br->mirrors[i]) {
3534 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3538 /* Collect new mirrors. */
3539 shash_init(&new_mirrors);
3540 for (i = 0; i < br->cfg->n_mirrors; i++) {
3541 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3542 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3543 VLOG_WARN("bridge %s: %s specified twice as mirror",
3544 br->name, cfg->name);
3548 /* Get rid of deleted mirrors and add new mirrors. */
3549 SHASH_FOR_EACH (node, &old_mirrors) {
3550 if (!shash_find(&new_mirrors, node->name)) {
3551 mirror_destroy(node->data);
3554 SHASH_FOR_EACH (node, &new_mirrors) {
3555 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3557 mirror = mirror_create(br, node->name);
3562 mirror_reconfigure_one(mirror, node->data);
3564 shash_destroy(&old_mirrors);
3565 shash_destroy(&new_mirrors);
3567 /* Update port reserved status. */
3568 for (i = 0; i < br->n_ports; i++) {
3569 br->ports[i]->is_mirror_output_port = false;
3571 for (i = 0; i < MAX_MIRRORS; i++) {
3572 struct mirror *m = br->mirrors[i];
3573 if (m && m->out_port) {
3574 m->out_port->is_mirror_output_port = true;
3578 /* Update flooded vlans (for RSPAN). */
3580 if (br->cfg->n_flood_vlans) {
3581 rspan_vlans = bitmap_allocate(4096);
3583 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3584 int64_t vlan = br->cfg->flood_vlans[i];
3585 if (vlan >= 0 && vlan < 4096) {
3586 bitmap_set1(rspan_vlans, vlan);
3587 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3590 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3595 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3600 static struct mirror *
3601 mirror_create(struct bridge *br, const char *name)
3606 for (i = 0; ; i++) {
3607 if (i >= MAX_MIRRORS) {
3608 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3609 "cannot create %s", br->name, MAX_MIRRORS, name);
3612 if (!br->mirrors[i]) {
3617 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3620 br->mirrors[i] = m = xzalloc(sizeof *m);
3623 m->name = xstrdup(name);
3624 shash_init(&m->src_ports);
3625 shash_init(&m->dst_ports);
3635 mirror_destroy(struct mirror *m)
3638 struct bridge *br = m->bridge;
3641 for (i = 0; i < br->n_ports; i++) {
3642 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3643 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3646 shash_destroy(&m->src_ports);
3647 shash_destroy(&m->dst_ports);
3650 m->bridge->mirrors[m->idx] = NULL;
3658 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3659 struct shash *names)
3663 for (i = 0; i < n_ports; i++) {
3664 const char *name = ports[i]->name;
3665 if (port_lookup(m->bridge, name)) {
3666 shash_add_once(names, name, NULL);
3668 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3669 "port %s", m->bridge->name, m->name, name);
3675 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3681 *vlans = xmalloc(sizeof *vlans * cfg->n_select_vlan);
3683 for (i = 0; i < cfg->n_select_vlan; i++) {
3684 int64_t vlan = cfg->select_vlan[i];
3685 if (vlan < 0 || vlan > 4095) {
3686 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3687 m->bridge->name, m->name, vlan);
3689 (*vlans)[n_vlans++] = vlan;
3696 vlan_is_mirrored(const struct mirror *m, int vlan)
3700 for (i = 0; i < m->n_vlans; i++) {
3701 if (m->vlans[i] == vlan) {
3709 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3713 for (i = 0; i < m->n_vlans; i++) {
3714 if (port_trunks_vlan(p, m->vlans[i])) {
3722 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3724 struct shash src_ports, dst_ports;
3725 mirror_mask_t mirror_bit;
3726 struct port *out_port;
3731 bool mirror_all_ports;
3732 bool any_ports_specified;
3733 bool any_vlans_specified;
3735 /* Get output port. */
3736 if (cfg->output_port) {
3737 out_port = port_lookup(m->bridge, cfg->output_port->name);
3739 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3740 m->bridge->name, m->name);
3746 if (cfg->output_vlan) {
3747 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3748 "output vlan; ignoring output vlan",
3749 m->bridge->name, m->name);
3751 } else if (cfg->output_vlan) {
3753 out_vlan = *cfg->output_vlan;
3755 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3756 m->bridge->name, m->name);
3761 /* Get all the ports, and drop duplicates and ports that don't exist. */
3762 shash_init(&src_ports);
3763 shash_init(&dst_ports);
3764 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3766 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3768 any_ports_specified = cfg->n_select_dst_port || cfg->n_select_dst_port;
3769 if (any_ports_specified
3770 && shash_is_empty(&src_ports) && shash_is_empty(&dst_ports)) {
3771 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3772 "selection ports exists", m->bridge->name, m->name);
3777 /* Get all the vlans, and drop duplicate and invalid vlans. */
3778 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3779 any_vlans_specified = cfg->n_select_vlan > 0;
3780 if (any_vlans_specified && !n_vlans) {
3781 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3782 "VLANs exists", m->bridge->name, m->name);
3787 /* Update mirror data. */
3788 if (!shash_equal_keys(&m->src_ports, &src_ports)
3789 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3790 || m->n_vlans != n_vlans
3791 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3792 || m->out_port != out_port
3793 || m->out_vlan != out_vlan) {
3794 bridge_flush(m->bridge);
3796 shash_swap(&m->src_ports, &src_ports);
3797 shash_swap(&m->dst_ports, &dst_ports);
3800 m->n_vlans = n_vlans;
3801 m->out_port = out_port;
3802 m->out_vlan = out_vlan;
3804 /* If no selection criteria have been given, mirror for all ports. */
3805 mirror_all_ports = !any_ports_specified && !any_vlans_specified;
3808 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3809 for (i = 0; i < m->bridge->n_ports; i++) {
3810 struct port *port = m->bridge->ports[i];
3812 if (mirror_all_ports
3813 || shash_find(&m->src_ports, port->name)
3816 ? port_trunks_any_mirrored_vlan(m, port)
3817 : vlan_is_mirrored(m, port->vlan)))) {
3818 port->src_mirrors |= mirror_bit;
3820 port->src_mirrors &= ~mirror_bit;
3823 if (mirror_all_ports || shash_find(&m->dst_ports, port->name)) {
3824 port->dst_mirrors |= mirror_bit;
3826 port->dst_mirrors &= ~mirror_bit;
3832 shash_destroy(&src_ports);
3833 shash_destroy(&dst_ports);