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 /* Description strings. */
167 char *mfr_desc; /* Manufacturer. */
168 char *hw_desc; /* Hardware. */
169 char *sw_desc; /* Software version. */
170 char *serial_desc; /* Serial number. */
171 char *dp_desc; /* Datapath description. */
173 /* Kernel datapath information. */
174 struct dpif *dpif; /* Datapath. */
175 struct port_array ifaces; /* Indexed by kernel datapath port number. */
179 size_t n_ports, allocated_ports;
182 bool has_bonded_ports;
183 long long int bond_next_rebalance;
188 /* Flow statistics gathering. */
189 time_t next_stats_request;
191 /* Port mirroring. */
192 struct mirror *mirrors[MAX_MIRRORS];
194 /* This member is only valid *during* bridge_reconfigure(). */
195 const struct ovsrec_bridge *cfg;
198 /* List of all bridges. */
199 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
201 /* Maximum number of datapaths. */
202 enum { DP_MAX = 256 };
204 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
205 static void bridge_destroy(struct bridge *);
206 static struct bridge *bridge_lookup(const char *name);
207 static unixctl_cb_func bridge_unixctl_dump_flows;
208 static int bridge_run_one(struct bridge *);
209 static const struct ovsrec_controller *bridge_get_controller(
210 const struct ovsrec_open_vswitch *ovs_cfg,
211 const struct bridge *br);
212 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
214 static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
216 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
217 static void bridge_fetch_dp_ifaces(struct bridge *);
218 static void bridge_flush(struct bridge *);
219 static void bridge_pick_local_hw_addr(struct bridge *,
220 uint8_t ea[ETH_ADDR_LEN],
221 struct iface **hw_addr_iface);
222 static uint64_t bridge_pick_datapath_id(struct bridge *,
223 const uint8_t bridge_ea[ETH_ADDR_LEN],
224 struct iface *hw_addr_iface);
225 static struct iface *bridge_get_local_iface(struct bridge *);
226 static uint64_t dpid_from_hash(const void *, size_t nbytes);
228 static unixctl_cb_func bridge_unixctl_fdb_show;
230 static void bond_init(void);
231 static void bond_run(struct bridge *);
232 static void bond_wait(struct bridge *);
233 static void bond_rebalance_port(struct port *);
234 static void bond_send_learning_packets(struct port *);
235 static void bond_enable_slave(struct iface *iface, bool enable);
237 static struct port *port_create(struct bridge *, const char *name);
238 static void port_reconfigure(struct port *, const struct ovsrec_port *);
239 static void port_destroy(struct port *);
240 static struct port *port_lookup(const struct bridge *, const char *name);
241 static struct iface *port_lookup_iface(const struct port *, const char *name);
242 static struct port *port_from_dp_ifidx(const struct bridge *,
244 static void port_update_bond_compat(struct port *);
245 static void port_update_vlan_compat(struct port *);
246 static void port_update_bonding(struct port *);
248 static struct mirror *mirror_create(struct bridge *, const char *name);
249 static void mirror_destroy(struct mirror *);
250 static void mirror_reconfigure(struct bridge *);
251 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
252 static bool vlan_is_mirrored(const struct mirror *, int vlan);
254 static struct iface *iface_create(struct port *port,
255 const struct ovsrec_interface *if_cfg);
256 static void iface_destroy(struct iface *);
257 static struct iface *iface_lookup(const struct bridge *, const char *name);
258 static struct iface *iface_from_dp_ifidx(const struct bridge *,
260 static bool iface_is_internal(const struct bridge *, const char *name);
261 static void iface_set_mac(struct iface *);
263 /* Hooks into ofproto processing. */
264 static struct ofhooks bridge_ofhooks;
266 /* Public functions. */
268 /* Adds the name of each interface used by a bridge, including local and
269 * internal ports, to 'svec'. */
271 bridge_get_ifaces(struct svec *svec)
273 struct bridge *br, *next;
276 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
277 for (i = 0; i < br->n_ports; i++) {
278 struct port *port = br->ports[i];
280 for (j = 0; j < port->n_ifaces; j++) {
281 struct iface *iface = port->ifaces[j];
282 if (iface->dp_ifidx < 0) {
283 VLOG_ERR("%s interface not in datapath %s, ignoring",
284 iface->name, dpif_name(br->dpif));
286 if (iface->dp_ifidx != ODPP_LOCAL) {
287 svec_add(svec, iface->name);
296 bridge_init(const struct ovsrec_open_vswitch *cfg)
298 struct svec bridge_names;
299 struct svec dpif_names, dpif_types;
302 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
304 svec_init(&bridge_names);
305 for (i = 0; i < cfg->n_bridges; i++) {
306 svec_add(&bridge_names, cfg->bridges[i]->name);
308 svec_sort(&bridge_names);
310 svec_init(&dpif_names);
311 svec_init(&dpif_types);
312 dp_enumerate_types(&dpif_types);
313 for (i = 0; i < dpif_types.n; i++) {
318 dp_enumerate_names(dpif_types.names[i], &dpif_names);
320 for (j = 0; j < dpif_names.n; j++) {
321 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
323 struct svec all_names;
326 svec_init(&all_names);
327 dpif_get_all_names(dpif, &all_names);
328 for (k = 0; k < all_names.n; k++) {
329 if (svec_contains(&bridge_names, all_names.names[k])) {
335 svec_destroy(&all_names);
340 svec_destroy(&dpif_names);
341 svec_destroy(&dpif_types);
343 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
347 bridge_reconfigure(cfg);
352 config_string_change(const char *value, char **valuep)
354 if (value && (!*valuep || strcmp(value, *valuep))) {
356 *valuep = xstrdup(value);
364 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
366 /* XXX SSL should be configurable on a per-bridge basis.
367 * XXX should be possible to de-configure SSL. */
368 static char *private_key_file;
369 static char *certificate_file;
370 static char *cacert_file;
374 /* XXX We can't un-set SSL settings. */
378 if (config_string_change(ssl->private_key, &private_key_file)) {
379 stream_ssl_set_private_key_file(private_key_file);
382 if (config_string_change(ssl->certificate, &certificate_file)) {
383 stream_ssl_set_certificate_file(certificate_file);
386 /* We assume that even if the filename hasn't changed, if the CA cert
387 * file has been removed, that we want to move back into
388 * boot-strapping mode. This opens a small security hole, because
389 * the old certificate will still be trusted until vSwitch is
390 * restarted. We may want to address this in vconn's SSL library. */
391 if (config_string_change(ssl->ca_cert, &cacert_file)
392 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
393 stream_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert);
398 /* Attempt to create the network device 'iface_name' through the netdev
401 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
404 struct shash_node *node;
405 struct shash options;
409 shash_init(&options);
410 for (i = 0; i < iface_cfg->n_options; i++) {
411 shash_add(&options, iface_cfg->key_options[i],
412 xstrdup(iface_cfg->value_options[i]));
416 struct netdev_options netdev_options;
418 memset(&netdev_options, 0, sizeof netdev_options);
419 netdev_options.name = iface_cfg->name;
420 if (!strcmp(iface_cfg->type, "internal")) {
421 /* An "internal" config type maps to a netdev "system" type. */
422 netdev_options.type = "system";
424 netdev_options.type = iface_cfg->type;
426 netdev_options.args = &options;
427 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
428 netdev_options.may_create = true;
429 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
430 netdev_options.may_open = true;
433 error = netdev_open(&netdev_options, &iface->netdev);
436 netdev_get_carrier(iface->netdev, &iface->enabled);
438 } else if (iface->netdev) {
439 const char *netdev_type = netdev_get_type(iface->netdev);
440 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
441 ? iface_cfg->type : NULL;
443 /* An "internal" config type maps to a netdev "system" type. */
444 if (iface_type && !strcmp(iface_type, "internal")) {
445 iface_type = "system";
448 if (!iface_type || !strcmp(netdev_type, iface_type)) {
449 error = netdev_reconfigure(iface->netdev, &options);
451 VLOG_WARN("%s: attempting change device type from %s to %s",
452 iface_cfg->name, netdev_type, iface_type);
457 SHASH_FOR_EACH (node, &options) {
460 shash_destroy(&options);
466 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
468 return set_up_iface(iface_cfg, iface, false);
472 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
473 void *aux OVS_UNUSED)
475 if (!iface->netdev) {
476 int error = set_up_iface(iface->cfg, iface, true);
478 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
488 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
489 void *aux OVS_UNUSED)
491 if (iface->dp_ifidx >= 0) {
492 VLOG_DBG("%s has interface %s on port %d",
494 iface->name, iface->dp_ifidx);
497 VLOG_ERR("%s interface not in %s, dropping",
498 iface->name, dpif_name(br->dpif));
504 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
505 void *aux OVS_UNUSED)
507 /* Set policing attributes. */
508 netdev_set_policing(iface->netdev,
509 iface->cfg->ingress_policing_rate,
510 iface->cfg->ingress_policing_burst);
512 /* Set MAC address of internal interfaces other than the local
514 if (iface->dp_ifidx != ODPP_LOCAL
515 && iface_is_internal(br, iface->name)) {
516 iface_set_mac(iface);
522 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
523 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
524 * deletes from 'br' any ports that no longer have any interfaces. */
526 iterate_and_prune_ifaces(struct bridge *br,
527 bool (*cb)(struct bridge *, struct iface *,
533 for (i = 0; i < br->n_ports; ) {
534 struct port *port = br->ports[i];
535 for (j = 0; j < port->n_ifaces; ) {
536 struct iface *iface = port->ifaces[j];
537 if (cb(br, iface, aux)) {
540 iface_destroy(iface);
544 if (port->n_ifaces) {
547 VLOG_ERR("%s port has no interfaces, dropping", port->name);
554 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
556 struct ovsdb_idl_txn *txn;
557 struct shash old_br, new_br;
558 struct shash_node *node;
559 struct bridge *br, *next;
561 int sflow_bridge_number;
563 COVERAGE_INC(bridge_reconfigure);
565 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
567 /* Collect old and new bridges. */
570 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
571 shash_add(&old_br, br->name, br);
573 for (i = 0; i < ovs_cfg->n_bridges; i++) {
574 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
575 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
576 VLOG_WARN("more than one bridge named %s", br_cfg->name);
580 /* Get rid of deleted bridges and add new bridges. */
581 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
582 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
589 SHASH_FOR_EACH (node, &new_br) {
590 const char *br_name = node->name;
591 const struct ovsrec_bridge *br_cfg = node->data;
592 br = shash_find_data(&old_br, br_name);
594 /* If the bridge datapath type has changed, we need to tear it
595 * down and recreate. */
596 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
598 bridge_create(br_cfg);
601 bridge_create(br_cfg);
604 shash_destroy(&old_br);
605 shash_destroy(&new_br);
609 bridge_configure_ssl(ovs_cfg->ssl);
612 /* Reconfigure all bridges. */
613 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
614 bridge_reconfigure_one(ovs_cfg, br);
617 /* Add and delete ports on all datapaths.
619 * The kernel will reject any attempt to add a given port to a datapath if
620 * that port already belongs to a different datapath, so we must do all
621 * port deletions before any port additions. */
622 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
623 struct odp_port *dpif_ports;
625 struct shash want_ifaces;
627 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
628 bridge_get_all_ifaces(br, &want_ifaces);
629 for (i = 0; i < n_dpif_ports; i++) {
630 const struct odp_port *p = &dpif_ports[i];
631 if (!shash_find(&want_ifaces, p->devname)
632 && strcmp(p->devname, br->name)) {
633 int retval = dpif_port_del(br->dpif, p->port);
635 VLOG_ERR("failed to remove %s interface from %s: %s",
636 p->devname, dpif_name(br->dpif),
641 shash_destroy(&want_ifaces);
644 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
645 struct odp_port *dpif_ports;
647 struct shash cur_ifaces, want_ifaces;
648 struct shash_node *node;
650 /* Get the set of interfaces currently in this datapath. */
651 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
652 shash_init(&cur_ifaces);
653 for (i = 0; i < n_dpif_ports; i++) {
654 const char *name = dpif_ports[i].devname;
655 if (!shash_find(&cur_ifaces, name)) {
656 shash_add(&cur_ifaces, name, NULL);
661 /* Get the set of interfaces we want on this datapath. */
662 bridge_get_all_ifaces(br, &want_ifaces);
664 SHASH_FOR_EACH (node, &want_ifaces) {
665 const char *if_name = node->name;
666 struct iface *iface = node->data;
668 if (shash_find(&cur_ifaces, if_name)) {
669 /* Already exists, just reconfigure it. */
671 reconfigure_iface(iface->cfg, iface);
674 /* Need to add to datapath. */
678 /* Add to datapath. */
679 internal = iface_is_internal(br, if_name);
680 error = dpif_port_add(br->dpif, if_name,
681 internal ? ODP_PORT_INTERNAL : 0, NULL);
682 if (error == EFBIG) {
683 VLOG_ERR("ran out of valid port numbers on %s",
684 dpif_name(br->dpif));
687 VLOG_ERR("failed to add %s interface to %s: %s",
688 if_name, dpif_name(br->dpif), strerror(error));
692 shash_destroy(&cur_ifaces);
693 shash_destroy(&want_ifaces);
695 sflow_bridge_number = 0;
696 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
699 struct iface *local_iface;
700 struct iface *hw_addr_iface;
703 bridge_fetch_dp_ifaces(br);
705 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
706 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
708 /* Pick local port hardware address, datapath ID. */
709 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
710 local_iface = bridge_get_local_iface(br);
712 int error = netdev_set_etheraddr(local_iface->netdev, ea);
714 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
715 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
716 "Ethernet address: %s",
717 br->name, strerror(error));
721 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
722 ofproto_set_datapath_id(br->ofproto, dpid);
724 dpid_string = xasprintf("%012"PRIx64, dpid);
725 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
728 /* Set NetFlow configuration on this bridge. */
729 if (br->cfg->netflow) {
730 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
731 struct netflow_options opts;
733 memset(&opts, 0, sizeof opts);
735 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
736 if (nf_cfg->engine_type) {
737 opts.engine_type = *nf_cfg->engine_type;
739 if (nf_cfg->engine_id) {
740 opts.engine_id = *nf_cfg->engine_id;
743 opts.active_timeout = nf_cfg->active_timeout;
744 if (!opts.active_timeout) {
745 opts.active_timeout = -1;
746 } else if (opts.active_timeout < 0) {
747 VLOG_WARN("bridge %s: active timeout interval set to negative "
748 "value, using default instead (%d seconds)", br->name,
749 NF_ACTIVE_TIMEOUT_DEFAULT);
750 opts.active_timeout = -1;
753 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
754 if (opts.add_id_to_iface) {
755 if (opts.engine_id > 0x7f) {
756 VLOG_WARN("bridge %s: netflow port mangling may conflict "
757 "with another vswitch, choose an engine id less "
758 "than 128", br->name);
760 if (br->n_ports > 508) {
761 VLOG_WARN("bridge %s: netflow port mangling will conflict "
762 "with another port when more than 508 ports are "
767 opts.collectors.n = nf_cfg->n_targets;
768 opts.collectors.names = nf_cfg->targets;
769 if (ofproto_set_netflow(br->ofproto, &opts)) {
770 VLOG_ERR("bridge %s: problem setting netflow collectors",
774 ofproto_set_netflow(br->ofproto, NULL);
777 /* Set sFlow configuration on this bridge. */
778 if (br->cfg->sflow) {
779 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
780 const struct ovsrec_controller *ctrl;
781 struct ofproto_sflow_options oso;
783 memset(&oso, 0, sizeof oso);
785 oso.targets.n = sflow_cfg->n_targets;
786 oso.targets.names = sflow_cfg->targets;
788 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
789 if (sflow_cfg->sampling) {
790 oso.sampling_rate = *sflow_cfg->sampling;
793 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
794 if (sflow_cfg->polling) {
795 oso.polling_interval = *sflow_cfg->polling;
798 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
799 if (sflow_cfg->header) {
800 oso.header_len = *sflow_cfg->header;
803 oso.sub_id = sflow_bridge_number++;
804 oso.agent_device = sflow_cfg->agent;
806 ctrl = bridge_get_controller(ovs_cfg, br);
807 oso.control_ip = ctrl ? ctrl->local_ip : NULL;
808 ofproto_set_sflow(br->ofproto, &oso);
810 svec_destroy(&oso.targets);
812 ofproto_set_sflow(br->ofproto, NULL);
815 /* Update the controller and related settings. It would be more
816 * straightforward to call this from bridge_reconfigure_one(), but we
817 * can't do it there for two reasons. First, and most importantly, at
818 * that point we don't know the dp_ifidx of any interfaces that have
819 * been added to the bridge (because we haven't actually added them to
820 * the datapath). Second, at that point we haven't set the datapath ID
821 * yet; when a controller is configured, resetting the datapath ID will
822 * immediately disconnect from the controller, so it's better to set
823 * the datapath ID before the controller. */
824 bridge_reconfigure_controller(ovs_cfg, br);
826 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
827 for (i = 0; i < br->n_ports; i++) {
828 struct port *port = br->ports[i];
830 port_update_vlan_compat(port);
831 port_update_bonding(port);
834 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
835 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
838 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
840 ovsdb_idl_txn_commit(txn);
841 ovsdb_idl_txn_destroy(txn); /* XXX */
845 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
849 for (i = 0; i < br_cfg->n_other_config; i++) {
850 if (!strcmp(br_cfg->key_other_config[i], key)) {
851 return br_cfg->value_other_config[i];
858 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
859 struct iface **hw_addr_iface)
865 *hw_addr_iface = NULL;
867 /* Did the user request a particular MAC? */
868 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
869 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
870 if (eth_addr_is_multicast(ea)) {
871 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
872 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
873 } else if (eth_addr_is_zero(ea)) {
874 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
880 /* Otherwise choose the minimum non-local MAC address among all of the
882 memset(ea, 0xff, sizeof ea);
883 for (i = 0; i < br->n_ports; i++) {
884 struct port *port = br->ports[i];
885 uint8_t iface_ea[ETH_ADDR_LEN];
888 /* Mirror output ports don't participate. */
889 if (port->is_mirror_output_port) {
893 /* Choose the MAC address to represent the port. */
894 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
895 /* Find the interface with this Ethernet address (if any) so that
896 * we can provide the correct devname to the caller. */
898 for (j = 0; j < port->n_ifaces; j++) {
899 struct iface *candidate = port->ifaces[j];
900 uint8_t candidate_ea[ETH_ADDR_LEN];
901 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
902 && eth_addr_equals(iface_ea, candidate_ea)) {
907 /* Choose the interface whose MAC address will represent the port.
908 * The Linux kernel bonding code always chooses the MAC address of
909 * the first slave added to a bond, and the Fedora networking
910 * scripts always add slaves to a bond in alphabetical order, so
911 * for compatibility we choose the interface with the name that is
912 * first in alphabetical order. */
913 iface = port->ifaces[0];
914 for (j = 1; j < port->n_ifaces; j++) {
915 struct iface *candidate = port->ifaces[j];
916 if (strcmp(candidate->name, iface->name) < 0) {
921 /* The local port doesn't count (since we're trying to choose its
922 * MAC address anyway). */
923 if (iface->dp_ifidx == ODPP_LOCAL) {
928 error = netdev_get_etheraddr(iface->netdev, iface_ea);
930 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
931 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
932 iface->name, strerror(error));
937 /* Compare against our current choice. */
938 if (!eth_addr_is_multicast(iface_ea) &&
939 !eth_addr_is_local(iface_ea) &&
940 !eth_addr_is_reserved(iface_ea) &&
941 !eth_addr_is_zero(iface_ea) &&
942 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
944 memcpy(ea, iface_ea, ETH_ADDR_LEN);
945 *hw_addr_iface = iface;
948 if (eth_addr_is_multicast(ea)) {
949 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
950 *hw_addr_iface = NULL;
951 VLOG_WARN("bridge %s: using default bridge Ethernet "
952 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
954 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
955 br->name, ETH_ADDR_ARGS(ea));
959 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
960 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
961 * an interface on 'br', then that interface must be passed in as
962 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
963 * 'hw_addr_iface' must be passed in as a null pointer. */
965 bridge_pick_datapath_id(struct bridge *br,
966 const uint8_t bridge_ea[ETH_ADDR_LEN],
967 struct iface *hw_addr_iface)
970 * The procedure for choosing a bridge MAC address will, in the most
971 * ordinary case, also choose a unique MAC that we can use as a datapath
972 * ID. In some special cases, though, multiple bridges will end up with
973 * the same MAC address. This is OK for the bridges, but it will confuse
974 * the OpenFlow controller, because each datapath needs a unique datapath
977 * Datapath IDs must be unique. It is also very desirable that they be
978 * stable from one run to the next, so that policy set on a datapath
981 const char *datapath_id;
984 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
985 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
991 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
993 * A bridge whose MAC address is taken from a VLAN network device
994 * (that is, a network device created with vconfig(8) or similar
995 * tool) will have the same MAC address as a bridge on the VLAN
996 * device's physical network device.
998 * Handle this case by hashing the physical network device MAC
999 * along with the VLAN identifier.
1001 uint8_t buf[ETH_ADDR_LEN + 2];
1002 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1003 buf[ETH_ADDR_LEN] = vlan >> 8;
1004 buf[ETH_ADDR_LEN + 1] = vlan;
1005 return dpid_from_hash(buf, sizeof buf);
1008 * Assume that this bridge's MAC address is unique, since it
1009 * doesn't fit any of the cases we handle specially.
1014 * A purely internal bridge, that is, one that has no non-virtual
1015 * network devices on it at all, is more difficult because it has no
1016 * natural unique identifier at all.
1018 * When the host is a XenServer, we handle this case by hashing the
1019 * host's UUID with the name of the bridge. Names of bridges are
1020 * persistent across XenServer reboots, although they can be reused if
1021 * an internal network is destroyed and then a new one is later
1022 * created, so this is fairly effective.
1024 * When the host is not a XenServer, we punt by using a random MAC
1025 * address on each run.
1027 const char *host_uuid = xenserver_get_host_uuid();
1029 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1030 dpid = dpid_from_hash(combined, strlen(combined));
1036 return eth_addr_to_uint64(bridge_ea);
1040 dpid_from_hash(const void *data, size_t n)
1042 uint8_t hash[SHA1_DIGEST_SIZE];
1044 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1045 sha1_bytes(data, n, hash);
1046 eth_addr_mark_random(hash);
1047 return eth_addr_to_uint64(hash);
1053 struct bridge *br, *next;
1057 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1058 int error = bridge_run_one(br);
1060 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1061 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1062 "forcing reconfiguration", br->name);
1076 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1077 ofproto_wait(br->ofproto);
1078 if (br->controller) {
1082 mac_learning_wait(br->ml);
1087 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1088 * configuration changes. */
1090 bridge_flush(struct bridge *br)
1092 COVERAGE_INC(bridge_flush);
1094 mac_learning_flush(br->ml);
1097 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1098 * such interface. */
1099 static struct iface *
1100 bridge_get_local_iface(struct bridge *br)
1104 for (i = 0; i < br->n_ports; i++) {
1105 struct port *port = br->ports[i];
1106 for (j = 0; j < port->n_ifaces; j++) {
1107 struct iface *iface = port->ifaces[j];
1108 if (iface->dp_ifidx == ODPP_LOCAL) {
1117 /* Bridge unixctl user interface functions. */
1119 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1120 const char *args, void *aux OVS_UNUSED)
1122 struct ds ds = DS_EMPTY_INITIALIZER;
1123 const struct bridge *br;
1124 const struct mac_entry *e;
1126 br = bridge_lookup(args);
1128 unixctl_command_reply(conn, 501, "no such bridge");
1132 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1133 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1134 if (e->port < 0 || e->port >= br->n_ports) {
1137 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1138 br->ports[e->port]->ifaces[0]->dp_ifidx,
1139 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1141 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1145 /* Bridge reconfiguration functions. */
1146 static struct bridge *
1147 bridge_create(const struct ovsrec_bridge *br_cfg)
1152 assert(!bridge_lookup(br_cfg->name));
1153 br = xzalloc(sizeof *br);
1155 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1161 dpif_flow_flush(br->dpif);
1163 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1166 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1168 dpif_delete(br->dpif);
1169 dpif_close(br->dpif);
1174 br->name = xstrdup(br_cfg->name);
1176 br->ml = mac_learning_create();
1177 br->sent_config_request = false;
1178 eth_addr_nicira_random(br->default_ea);
1180 port_array_init(&br->ifaces);
1183 br->bond_next_rebalance = time_msec() + 10000;
1185 list_push_back(&all_bridges, &br->node);
1187 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1193 bridge_destroy(struct bridge *br)
1198 while (br->n_ports > 0) {
1199 port_destroy(br->ports[br->n_ports - 1]);
1201 list_remove(&br->node);
1202 error = dpif_delete(br->dpif);
1203 if (error && error != ENOENT) {
1204 VLOG_ERR("failed to delete %s: %s",
1205 dpif_name(br->dpif), strerror(error));
1207 dpif_close(br->dpif);
1208 ofproto_destroy(br->ofproto);
1209 free(br->controller);
1210 mac_learning_destroy(br->ml);
1211 port_array_destroy(&br->ifaces);
1218 static struct bridge *
1219 bridge_lookup(const char *name)
1223 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1224 if (!strcmp(br->name, name)) {
1232 bridge_exists(const char *name)
1234 return bridge_lookup(name) ? true : false;
1238 bridge_get_datapathid(const char *name)
1240 struct bridge *br = bridge_lookup(name);
1241 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1244 /* Handle requests for a listing of all flows known by the OpenFlow
1245 * stack, including those normally hidden. */
1247 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1248 const char *args, void *aux OVS_UNUSED)
1253 br = bridge_lookup(args);
1255 unixctl_command_reply(conn, 501, "Unknown bridge");
1260 ofproto_get_all_flows(br->ofproto, &results);
1262 unixctl_command_reply(conn, 200, ds_cstr(&results));
1263 ds_destroy(&results);
1267 bridge_run_one(struct bridge *br)
1271 error = ofproto_run1(br->ofproto);
1276 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1279 error = ofproto_run2(br->ofproto, br->flush);
1285 static const struct ovsrec_controller *
1286 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1287 const struct bridge *br)
1289 const struct ovsrec_controller *controller;
1291 controller = (br->cfg->controller ? br->cfg->controller
1292 : ovs_cfg->controller ? ovs_cfg->controller
1295 if (controller && !strcmp(controller->target, "none")) {
1303 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1305 struct svec *ifaces = ifaces_;
1306 if (!svec_contains(ifaces, iface->name)) {
1307 svec_add(ifaces, iface->name);
1311 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1313 br->name, iface->name, iface->port->name);
1319 bridge_update_desc(struct bridge *br)
1322 bool changed = false;
1325 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1326 if (desc != br->mfr_desc) {
1329 br->mfr_desc = xstrdup(desc);
1331 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1336 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1337 if (desc != br->hw_desc) {
1340 br->hw_desc = xstrdup(desc);
1342 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1347 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1348 if (desc != br->sw_desc) {
1351 br->sw_desc = xstrdup(desc);
1353 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1358 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1359 if (desc != br->serial_desc) {
1360 free(br->serial_desc);
1362 br->serial_desc = xstrdup(desc);
1364 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1369 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1370 if (desc != br->dp_desc) {
1373 br->dp_desc = xstrdup(desc);
1375 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1381 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1382 br->sw_desc, br->serial_desc, br->dp_desc);
1388 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1391 struct shash old_ports, new_ports;
1393 struct svec listeners, old_listeners;
1394 struct svec snoops, old_snoops;
1395 struct shash_node *node;
1398 /* Collect old ports. */
1399 shash_init(&old_ports);
1400 for (i = 0; i < br->n_ports; i++) {
1401 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1404 /* Collect new ports. */
1405 shash_init(&new_ports);
1406 for (i = 0; i < br->cfg->n_ports; i++) {
1407 const char *name = br->cfg->ports[i]->name;
1408 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1409 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1414 /* If we have a controller, then we need a local port. Complain if the
1415 * user didn't specify one.
1417 * XXX perhaps we should synthesize a port ourselves in this case. */
1418 if (bridge_get_controller(ovs_cfg, br)) {
1419 char local_name[IF_NAMESIZE];
1422 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1423 local_name, sizeof local_name);
1424 if (!error && !shash_find(&new_ports, local_name)) {
1425 VLOG_WARN("bridge %s: controller specified but no local port "
1426 "(port named %s) defined",
1427 br->name, local_name);
1431 /* Get rid of deleted ports and add new ports. */
1432 SHASH_FOR_EACH (node, &old_ports) {
1433 if (!shash_find(&new_ports, node->name)) {
1434 port_destroy(node->data);
1437 SHASH_FOR_EACH (node, &new_ports) {
1438 struct port *port = shash_find_data(&old_ports, node->name);
1440 port = port_create(br, node->name);
1442 port_reconfigure(port, node->data);
1444 shash_destroy(&old_ports);
1445 shash_destroy(&new_ports);
1447 /* Check and delete duplicate interfaces. */
1449 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1450 svec_destroy(&ifaces);
1452 /* Delete all flows if we're switching from connected to standalone or vice
1453 * versa. (XXX Should we delete all flows if we are switching from one
1454 * controller to another?) */
1457 /* Configure OpenFlow management listeners. */
1458 svec_init(&listeners);
1459 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1461 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1462 ovs_rundir, br->name));
1463 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1464 svec_clear(&listeners);
1466 svec_sort_unique(&listeners);
1468 svec_init(&old_listeners);
1469 ofproto_get_listeners(br->ofproto, &old_listeners);
1470 svec_sort_unique(&old_listeners);
1472 if (!svec_equal(&listeners, &old_listeners)) {
1473 ofproto_set_listeners(br->ofproto, &listeners);
1475 svec_destroy(&listeners);
1476 svec_destroy(&old_listeners);
1478 /* Configure OpenFlow controller connection snooping. */
1480 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1482 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1483 ovs_rundir, br->name));
1484 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1485 svec_clear(&snoops);
1487 svec_sort_unique(&snoops);
1489 svec_init(&old_snoops);
1490 ofproto_get_snoops(br->ofproto, &old_snoops);
1491 svec_sort_unique(&old_snoops);
1493 if (!svec_equal(&snoops, &old_snoops)) {
1494 ofproto_set_snoops(br->ofproto, &snoops);
1496 svec_destroy(&snoops);
1497 svec_destroy(&old_snoops);
1499 /* Default listener. */
1500 svec_init(&listeners);
1501 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1502 ovs_rundir, br->name));
1503 svec_init(&old_listeners);
1504 ofproto_get_listeners(br->ofproto, &old_listeners);
1505 if (!svec_equal(&listeners, &old_listeners)) {
1506 ofproto_set_listeners(br->ofproto, &listeners);
1508 svec_destroy(&listeners);
1509 svec_destroy(&old_listeners);
1511 /* Default snoop. */
1513 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1514 ovs_rundir, br->name));
1515 svec_init(&old_snoops);
1516 ofproto_get_snoops(br->ofproto, &old_snoops);
1517 if (!svec_equal(&snoops, &old_snoops)) {
1518 ofproto_set_snoops(br->ofproto, &snoops);
1520 svec_destroy(&snoops);
1521 svec_destroy(&old_snoops);
1524 mirror_reconfigure(br);
1526 bridge_update_desc(br);
1530 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1533 const struct ovsrec_controller *c;
1535 c = bridge_get_controller(ovs_cfg, br);
1536 if ((br->controller != NULL) != (c != NULL)) {
1537 ofproto_flush_flows(br->ofproto);
1539 free(br->controller);
1540 br->controller = c ? xstrdup(c->target) : NULL;
1543 int max_backoff, probe;
1544 int rate_limit, burst_limit;
1546 if (!strcmp(c->target, "discover")) {
1547 ofproto_set_discovery(br->ofproto, true,
1548 c->discover_accept_regex,
1549 c->discover_update_resolv_conf);
1551 struct iface *local_iface;
1555 in_band = (!c->connection_mode
1556 || !strcmp(c->connection_mode, "out-of-band"));
1557 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1558 ofproto_set_in_band(br->ofproto, in_band);
1560 local_iface = bridge_get_local_iface(br);
1561 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1562 struct netdev *netdev = local_iface->netdev;
1563 struct in_addr mask, gateway;
1565 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1568 if (!c->local_gateway
1569 || !inet_aton(c->local_gateway, &gateway)) {
1573 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1575 mask.s_addr = guess_netmask(ip.s_addr);
1577 if (!netdev_set_in4(netdev, ip, mask)) {
1578 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1580 br->name, IP_ARGS(&ip.s_addr),
1581 IP_ARGS(&mask.s_addr));
1584 if (gateway.s_addr) {
1585 if (!netdev_add_router(netdev, gateway)) {
1586 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1587 br->name, IP_ARGS(&gateway.s_addr));
1593 ofproto_set_failure(br->ofproto,
1595 || !strcmp(c->fail_mode, "standalone")
1596 || !strcmp(c->fail_mode, "open")));
1598 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1599 ofproto_set_probe_interval(br->ofproto, probe);
1601 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1602 ofproto_set_max_backoff(br->ofproto, max_backoff);
1604 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1605 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1606 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1608 union ofp_action action;
1611 /* Set up a flow that matches every packet and directs them to
1612 * OFPP_NORMAL (which goes to us). */
1613 memset(&action, 0, sizeof action);
1614 action.type = htons(OFPAT_OUTPUT);
1615 action.output.len = htons(sizeof action);
1616 action.output.port = htons(OFPP_NORMAL);
1617 memset(&flow, 0, sizeof flow);
1618 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1621 ofproto_set_in_band(br->ofproto, false);
1622 ofproto_set_max_backoff(br->ofproto, 1);
1623 ofproto_set_probe_interval(br->ofproto, 5);
1624 ofproto_set_failure(br->ofproto, false);
1627 ofproto_set_controller(br->ofproto, br->controller);
1631 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1636 for (i = 0; i < br->n_ports; i++) {
1637 struct port *port = br->ports[i];
1638 for (j = 0; j < port->n_ifaces; j++) {
1639 struct iface *iface = port->ifaces[j];
1640 shash_add_once(ifaces, iface->name, iface);
1642 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1643 shash_add_once(ifaces, port->name, NULL);
1648 /* For robustness, in case the administrator moves around datapath ports behind
1649 * our back, we re-check all the datapath port numbers here.
1651 * This function will set the 'dp_ifidx' members of interfaces that have
1652 * disappeared to -1, so only call this function from a context where those
1653 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1654 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1655 * datapath, which doesn't support UINT16_MAX+1 ports. */
1657 bridge_fetch_dp_ifaces(struct bridge *br)
1659 struct odp_port *dpif_ports;
1660 size_t n_dpif_ports;
1663 /* Reset all interface numbers. */
1664 for (i = 0; i < br->n_ports; i++) {
1665 struct port *port = br->ports[i];
1666 for (j = 0; j < port->n_ifaces; j++) {
1667 struct iface *iface = port->ifaces[j];
1668 iface->dp_ifidx = -1;
1671 port_array_clear(&br->ifaces);
1673 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1674 for (i = 0; i < n_dpif_ports; i++) {
1675 struct odp_port *p = &dpif_ports[i];
1676 struct iface *iface = iface_lookup(br, p->devname);
1678 if (iface->dp_ifidx >= 0) {
1679 VLOG_WARN("%s reported interface %s twice",
1680 dpif_name(br->dpif), p->devname);
1681 } else if (iface_from_dp_ifidx(br, p->port)) {
1682 VLOG_WARN("%s reported interface %"PRIu16" twice",
1683 dpif_name(br->dpif), p->port);
1685 port_array_set(&br->ifaces, p->port, iface);
1686 iface->dp_ifidx = p->port;
1690 int64_t ofport = (iface->dp_ifidx >= 0
1691 ? odp_port_to_ofp_port(iface->dp_ifidx)
1693 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1700 /* Bridge packet processing functions. */
1703 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1705 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1708 static struct bond_entry *
1709 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1711 return &port->bond_hash[bond_hash(mac)];
1715 bond_choose_iface(const struct port *port)
1717 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1718 size_t i, best_down_slave = -1;
1719 long long next_delay_expiration = LLONG_MAX;
1721 for (i = 0; i < port->n_ifaces; i++) {
1722 struct iface *iface = port->ifaces[i];
1724 if (iface->enabled) {
1726 } else if (iface->delay_expires < next_delay_expiration) {
1727 best_down_slave = i;
1728 next_delay_expiration = iface->delay_expires;
1732 if (best_down_slave != -1) {
1733 struct iface *iface = port->ifaces[best_down_slave];
1735 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1736 "since no other interface is up", iface->name,
1737 iface->delay_expires - time_msec());
1738 bond_enable_slave(iface, true);
1741 return best_down_slave;
1745 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1746 uint16_t *dp_ifidx, tag_type *tags)
1748 struct iface *iface;
1750 assert(port->n_ifaces);
1751 if (port->n_ifaces == 1) {
1752 iface = port->ifaces[0];
1754 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1755 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1756 || !port->ifaces[e->iface_idx]->enabled) {
1757 /* XXX select interface properly. The current interface selection
1758 * is only good for testing the rebalancing code. */
1759 e->iface_idx = bond_choose_iface(port);
1760 if (e->iface_idx < 0) {
1761 *tags |= port->no_ifaces_tag;
1764 e->iface_tag = tag_create_random();
1765 ((struct port *) port)->bond_compat_is_stale = true;
1767 *tags |= e->iface_tag;
1768 iface = port->ifaces[e->iface_idx];
1770 *dp_ifidx = iface->dp_ifidx;
1771 *tags |= iface->tag; /* Currently only used for bonding. */
1776 bond_link_status_update(struct iface *iface, bool carrier)
1778 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1779 struct port *port = iface->port;
1781 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1782 /* Nothing to do. */
1785 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1786 iface->name, carrier ? "detected" : "dropped");
1787 if (carrier == iface->enabled) {
1788 iface->delay_expires = LLONG_MAX;
1789 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1790 iface->name, carrier ? "disabled" : "enabled");
1791 } else if (carrier && port->active_iface < 0) {
1792 bond_enable_slave(iface, true);
1793 if (port->updelay) {
1794 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1795 "other interface is up", iface->name, port->updelay);
1798 int delay = carrier ? port->updelay : port->downdelay;
1799 iface->delay_expires = time_msec() + delay;
1802 "interface %s: will be %s if it stays %s for %d ms",
1804 carrier ? "enabled" : "disabled",
1805 carrier ? "up" : "down",
1812 bond_choose_active_iface(struct port *port)
1814 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1816 port->active_iface = bond_choose_iface(port);
1817 port->active_iface_tag = tag_create_random();
1818 if (port->active_iface >= 0) {
1819 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1820 port->name, port->ifaces[port->active_iface]->name);
1822 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1828 bond_enable_slave(struct iface *iface, bool enable)
1830 struct port *port = iface->port;
1831 struct bridge *br = port->bridge;
1833 /* This acts as a recursion check. If the act of disabling a slave
1834 * causes a different slave to be enabled, the flag will allow us to
1835 * skip redundant work when we reenter this function. It must be
1836 * cleared on exit to keep things safe with multiple bonds. */
1837 static bool moving_active_iface = false;
1839 iface->delay_expires = LLONG_MAX;
1840 if (enable == iface->enabled) {
1844 iface->enabled = enable;
1845 if (!iface->enabled) {
1846 VLOG_WARN("interface %s: disabled", iface->name);
1847 ofproto_revalidate(br->ofproto, iface->tag);
1848 if (iface->port_ifidx == port->active_iface) {
1849 ofproto_revalidate(br->ofproto,
1850 port->active_iface_tag);
1852 /* Disabling a slave can lead to another slave being immediately
1853 * enabled if there will be no active slaves but one is waiting
1854 * on an updelay. In this case we do not need to run most of the
1855 * code for the newly enabled slave since there was no period
1856 * without an active slave and it is redundant with the disabling
1858 moving_active_iface = true;
1859 bond_choose_active_iface(port);
1861 bond_send_learning_packets(port);
1863 VLOG_WARN("interface %s: enabled", iface->name);
1864 if (port->active_iface < 0 && !moving_active_iface) {
1865 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1866 bond_choose_active_iface(port);
1867 bond_send_learning_packets(port);
1869 iface->tag = tag_create_random();
1872 moving_active_iface = false;
1873 port->bond_compat_is_stale = true;
1877 bond_run(struct bridge *br)
1881 for (i = 0; i < br->n_ports; i++) {
1882 struct port *port = br->ports[i];
1884 if (port->n_ifaces >= 2) {
1885 for (j = 0; j < port->n_ifaces; j++) {
1886 struct iface *iface = port->ifaces[j];
1887 if (time_msec() >= iface->delay_expires) {
1888 bond_enable_slave(iface, !iface->enabled);
1893 if (port->bond_compat_is_stale) {
1894 port->bond_compat_is_stale = false;
1895 port_update_bond_compat(port);
1901 bond_wait(struct bridge *br)
1905 for (i = 0; i < br->n_ports; i++) {
1906 struct port *port = br->ports[i];
1907 if (port->n_ifaces < 2) {
1910 for (j = 0; j < port->n_ifaces; j++) {
1911 struct iface *iface = port->ifaces[j];
1912 if (iface->delay_expires != LLONG_MAX) {
1913 poll_timer_wait(iface->delay_expires - time_msec());
1920 set_dst(struct dst *p, const flow_t *flow,
1921 const struct port *in_port, const struct port *out_port,
1924 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1925 : in_port->vlan >= 0 ? in_port->vlan
1926 : ntohs(flow->dl_vlan));
1927 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1931 swap_dst(struct dst *p, struct dst *q)
1933 struct dst tmp = *p;
1938 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1939 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1940 * that we push to the datapath. We could in fact fully sort the array by
1941 * vlan, but in most cases there are at most two different vlan tags so that's
1942 * possibly overkill.) */
1944 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1946 struct dst *first = dsts;
1947 struct dst *last = dsts + n_dsts;
1949 while (first != last) {
1951 * - All dsts < first have vlan == 'vlan'.
1952 * - All dsts >= last have vlan != 'vlan'.
1953 * - first < last. */
1954 while (first->vlan == vlan) {
1955 if (++first == last) {
1960 /* Same invariants, plus one additional:
1961 * - first->vlan != vlan.
1963 while (last[-1].vlan != vlan) {
1964 if (--last == first) {
1969 /* Same invariants, plus one additional:
1970 * - last[-1].vlan == vlan.*/
1971 swap_dst(first++, --last);
1976 mirror_mask_ffs(mirror_mask_t mask)
1978 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1983 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1984 const struct dst *test)
1987 for (i = 0; i < n_dsts; i++) {
1988 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1996 port_trunks_vlan(const struct port *port, uint16_t vlan)
1998 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
2002 port_includes_vlan(const struct port *port, uint16_t vlan)
2004 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2008 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2009 const struct port *in_port, const struct port *out_port,
2010 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2012 mirror_mask_t mirrors = in_port->src_mirrors;
2013 struct dst *dst = dsts;
2016 if (out_port == FLOOD_PORT) {
2017 /* XXX use ODP_FLOOD if no vlans or bonding. */
2018 /* XXX even better, define each VLAN as a datapath port group */
2019 for (i = 0; i < br->n_ports; i++) {
2020 struct port *port = br->ports[i];
2021 if (port != in_port && port_includes_vlan(port, vlan)
2022 && !port->is_mirror_output_port
2023 && set_dst(dst, flow, in_port, port, tags)) {
2024 mirrors |= port->dst_mirrors;
2028 *nf_output_iface = NF_OUT_FLOOD;
2029 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2030 *nf_output_iface = dst->dp_ifidx;
2031 mirrors |= out_port->dst_mirrors;
2036 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2037 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2039 if (set_dst(dst, flow, in_port, m->out_port, tags)
2040 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2044 for (i = 0; i < br->n_ports; i++) {
2045 struct port *port = br->ports[i];
2046 if (port_includes_vlan(port, m->out_vlan)
2047 && set_dst(dst, flow, in_port, port, tags))
2051 if (port->vlan < 0) {
2052 dst->vlan = m->out_vlan;
2054 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2058 /* Use the vlan tag on the original flow instead of
2059 * the one passed in the vlan parameter. This ensures
2060 * that we compare the vlan from before any implicit
2061 * tagging tags place. This is necessary because
2062 * dst->vlan is the final vlan, after removing implicit
2064 flow_vlan = ntohs(flow->dl_vlan);
2065 if (flow_vlan == 0) {
2066 flow_vlan = OFP_VLAN_NONE;
2068 if (port == in_port && dst->vlan == flow_vlan) {
2069 /* Don't send out input port on same VLAN. */
2077 mirrors &= mirrors - 1;
2080 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2084 static void OVS_UNUSED
2085 print_dsts(const struct dst *dsts, size_t n)
2087 for (; n--; dsts++) {
2088 printf(">p%"PRIu16, dsts->dp_ifidx);
2089 if (dsts->vlan != OFP_VLAN_NONE) {
2090 printf("v%"PRIu16, dsts->vlan);
2096 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2097 const struct port *in_port, const struct port *out_port,
2098 tag_type *tags, struct odp_actions *actions,
2099 uint16_t *nf_output_iface)
2101 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2103 const struct dst *p;
2106 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2109 cur_vlan = ntohs(flow->dl_vlan);
2110 for (p = dsts; p < &dsts[n_dsts]; p++) {
2111 union odp_action *a;
2112 if (p->vlan != cur_vlan) {
2113 if (p->vlan == OFP_VLAN_NONE) {
2114 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2116 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2117 a->vlan_vid.vlan_vid = htons(p->vlan);
2121 a = odp_actions_add(actions, ODPAT_OUTPUT);
2122 a->output.port = p->dp_ifidx;
2126 /* Returns the effective vlan of a packet, taking into account both the
2127 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2128 * the packet is untagged and -1 indicates it has an invalid header and
2129 * should be dropped. */
2130 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2131 struct port *in_port, bool have_packet)
2133 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2134 * belongs to VLAN 0, so we should treat both cases identically. (In the
2135 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2136 * presumably to allow a priority to be specified. In the latter case, the
2137 * packet does not have any 802.1Q header.) */
2138 int vlan = ntohs(flow->dl_vlan);
2139 if (vlan == OFP_VLAN_NONE) {
2142 if (in_port->vlan >= 0) {
2144 /* XXX support double tagging? */
2146 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2147 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2148 "packet received on port %s configured with "
2149 "implicit VLAN %"PRIu16,
2150 br->name, ntohs(flow->dl_vlan),
2151 in_port->name, in_port->vlan);
2155 vlan = in_port->vlan;
2157 if (!port_includes_vlan(in_port, vlan)) {
2159 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2160 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2161 "packet received on port %s not configured for "
2163 br->name, vlan, in_port->name, vlan);
2173 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2174 struct port *in_port)
2176 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2177 vlan, in_port->port_idx);
2179 /* The log messages here could actually be useful in debugging,
2180 * so keep the rate limit relatively high. */
2181 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2183 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2184 "on port %s in VLAN %d",
2185 br->name, ETH_ADDR_ARGS(flow->dl_src),
2186 in_port->name, vlan);
2187 ofproto_revalidate(br->ofproto, rev_tag);
2192 is_bcast_arp_reply(const flow_t *flow)
2194 return (flow->dl_type == htons(ETH_TYPE_ARP)
2195 && flow->nw_proto == ARP_OP_REPLY
2196 && eth_addr_is_broadcast(flow->dl_dst));
2199 /* If the composed actions may be applied to any packet in the given 'flow',
2200 * returns true. Otherwise, the actions should only be applied to 'packet', or
2201 * not at all, if 'packet' was NULL. */
2203 process_flow(struct bridge *br, const flow_t *flow,
2204 const struct ofpbuf *packet, struct odp_actions *actions,
2205 tag_type *tags, uint16_t *nf_output_iface)
2207 struct iface *in_iface;
2208 struct port *in_port;
2209 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2213 /* Find the interface and port structure for the received packet. */
2214 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2216 /* No interface? Something fishy... */
2217 if (packet != NULL) {
2218 /* Odd. A few possible reasons here:
2220 * - We deleted an interface but there are still a few packets
2221 * queued up from it.
2223 * - Someone externally added an interface (e.g. with "ovs-dpctl
2224 * add-if") that we don't know about.
2226 * - Packet arrived on the local port but the local port is not
2227 * one of our bridge ports.
2229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2231 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2232 "interface %"PRIu16, br->name, flow->in_port);
2235 /* Return without adding any actions, to drop packets on this flow. */
2238 in_port = in_iface->port;
2239 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2244 /* Drop frames for reserved multicast addresses. */
2245 if (eth_addr_is_reserved(flow->dl_dst)) {
2249 /* Drop frames on ports reserved for mirroring. */
2250 if (in_port->is_mirror_output_port) {
2251 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2252 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2253 "which is reserved exclusively for mirroring",
2254 br->name, in_port->name);
2258 /* Packets received on bonds need special attention to avoid duplicates. */
2259 if (in_port->n_ifaces > 1) {
2262 if (eth_addr_is_multicast(flow->dl_dst)) {
2263 *tags |= in_port->active_iface_tag;
2264 if (in_port->active_iface != in_iface->port_ifidx) {
2265 /* Drop all multicast packets on inactive slaves. */
2270 /* Drop all packets for which we have learned a different input
2271 * port, because we probably sent the packet on one slave and got
2272 * it back on the other. Broadcast ARP replies are an exception
2273 * to this rule: the host has moved to another switch. */
2274 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2275 if (src_idx != -1 && src_idx != in_port->port_idx &&
2276 !is_bcast_arp_reply(flow)) {
2282 out_port = FLOOD_PORT;
2283 /* Learn source MAC (but don't try to learn from revalidation). */
2285 update_learning_table(br, flow, vlan, in_port);
2288 /* Determine output port. */
2289 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2291 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2292 out_port = br->ports[out_port_idx];
2293 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2294 /* If we are revalidating but don't have a learning entry then
2295 * eject the flow. Installing a flow that floods packets opens
2296 * up a window of time where we could learn from a packet reflected
2297 * on a bond and blackhole packets before the learning table is
2298 * updated to reflect the correct port. */
2302 /* Don't send packets out their input ports. */
2303 if (in_port == out_port) {
2308 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2314 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2317 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2318 const struct ofp_phy_port *opp,
2321 struct bridge *br = br_;
2322 struct iface *iface;
2325 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2331 if (reason == OFPPR_DELETE) {
2332 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2333 br->name, iface->name);
2334 iface_destroy(iface);
2335 if (!port->n_ifaces) {
2336 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2337 br->name, port->name);
2343 if (port->n_ifaces > 1) {
2344 bool up = !(opp->state & OFPPS_LINK_DOWN);
2345 bond_link_status_update(iface, up);
2346 port_update_bond_compat(port);
2352 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2353 struct odp_actions *actions, tag_type *tags,
2354 uint16_t *nf_output_iface, void *br_)
2356 struct bridge *br = br_;
2358 COVERAGE_INC(bridge_process_flow);
2359 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2363 bridge_account_flow_ofhook_cb(const flow_t *flow,
2364 const union odp_action *actions,
2365 size_t n_actions, unsigned long long int n_bytes,
2368 struct bridge *br = br_;
2369 struct port *in_port;
2370 const union odp_action *a;
2372 /* Feed information from the active flows back into the learning table
2373 * to ensure that table is always in sync with what is actually flowing
2374 * through the datapath. */
2375 in_port = port_from_dp_ifidx(br, flow->in_port);
2377 int vlan = flow_get_vlan(br, flow, in_port, false);
2379 update_learning_table(br, flow, vlan, in_port);
2383 if (!br->has_bonded_ports) {
2387 for (a = actions; a < &actions[n_actions]; a++) {
2388 if (a->type == ODPAT_OUTPUT) {
2389 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2390 if (out_port && out_port->n_ifaces >= 2) {
2391 struct bond_entry *e = lookup_bond_entry(out_port,
2393 e->tx_bytes += n_bytes;
2400 bridge_account_checkpoint_ofhook_cb(void *br_)
2402 struct bridge *br = br_;
2405 if (!br->has_bonded_ports) {
2409 /* The current ofproto implementation calls this callback at least once a
2410 * second, so this timer implementation is sufficient. */
2411 if (time_msec() < br->bond_next_rebalance) {
2414 br->bond_next_rebalance = time_msec() + 10000;
2416 for (i = 0; i < br->n_ports; i++) {
2417 struct port *port = br->ports[i];
2418 if (port->n_ifaces > 1) {
2419 bond_rebalance_port(port);
2424 static struct ofhooks bridge_ofhooks = {
2425 bridge_port_changed_ofhook_cb,
2426 bridge_normal_ofhook_cb,
2427 bridge_account_flow_ofhook_cb,
2428 bridge_account_checkpoint_ofhook_cb,
2431 /* Bonding functions. */
2433 /* Statistics for a single interface on a bonded port, used for load-based
2434 * bond rebalancing. */
2435 struct slave_balance {
2436 struct iface *iface; /* The interface. */
2437 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2439 /* All the "bond_entry"s that are assigned to this interface, in order of
2440 * increasing tx_bytes. */
2441 struct bond_entry **hashes;
2445 /* Sorts pointers to pointers to bond_entries in ascending order by the
2446 * interface to which they are assigned, and within a single interface in
2447 * ascending order of bytes transmitted. */
2449 compare_bond_entries(const void *a_, const void *b_)
2451 const struct bond_entry *const *ap = a_;
2452 const struct bond_entry *const *bp = b_;
2453 const struct bond_entry *a = *ap;
2454 const struct bond_entry *b = *bp;
2455 if (a->iface_idx != b->iface_idx) {
2456 return a->iface_idx > b->iface_idx ? 1 : -1;
2457 } else if (a->tx_bytes != b->tx_bytes) {
2458 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2464 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2465 * *descending* order by number of bytes transmitted. */
2467 compare_slave_balance(const void *a_, const void *b_)
2469 const struct slave_balance *a = a_;
2470 const struct slave_balance *b = b_;
2471 if (a->iface->enabled != b->iface->enabled) {
2472 return a->iface->enabled ? -1 : 1;
2473 } else if (a->tx_bytes != b->tx_bytes) {
2474 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2481 swap_bals(struct slave_balance *a, struct slave_balance *b)
2483 struct slave_balance tmp = *a;
2488 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2489 * given that 'p' (and only 'p') might be in the wrong location.
2491 * This function invalidates 'p', since it might now be in a different memory
2494 resort_bals(struct slave_balance *p,
2495 struct slave_balance bals[], size_t n_bals)
2498 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2499 swap_bals(p, p - 1);
2501 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2502 swap_bals(p, p + 1);
2508 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2510 if (VLOG_IS_DBG_ENABLED()) {
2511 struct ds ds = DS_EMPTY_INITIALIZER;
2512 const struct slave_balance *b;
2514 for (b = bals; b < bals + n_bals; b++) {
2518 ds_put_char(&ds, ',');
2520 ds_put_format(&ds, " %s %"PRIu64"kB",
2521 b->iface->name, b->tx_bytes / 1024);
2523 if (!b->iface->enabled) {
2524 ds_put_cstr(&ds, " (disabled)");
2526 if (b->n_hashes > 0) {
2527 ds_put_cstr(&ds, " (");
2528 for (i = 0; i < b->n_hashes; i++) {
2529 const struct bond_entry *e = b->hashes[i];
2531 ds_put_cstr(&ds, " + ");
2533 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2534 e - port->bond_hash, e->tx_bytes / 1024);
2536 ds_put_cstr(&ds, ")");
2539 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2544 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2546 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2549 struct bond_entry *hash = from->hashes[hash_idx];
2550 struct port *port = from->iface->port;
2551 uint64_t delta = hash->tx_bytes;
2553 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2554 "from %s to %s (now carrying %"PRIu64"kB and "
2555 "%"PRIu64"kB load, respectively)",
2556 port->name, delta / 1024, hash - port->bond_hash,
2557 from->iface->name, to->iface->name,
2558 (from->tx_bytes - delta) / 1024,
2559 (to->tx_bytes + delta) / 1024);
2561 /* Delete element from from->hashes.
2563 * We don't bother to add the element to to->hashes because not only would
2564 * it require more work, the only purpose it would be to allow that hash to
2565 * be migrated to another slave in this rebalancing run, and there is no
2566 * point in doing that. */
2567 if (hash_idx == 0) {
2570 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2571 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2575 /* Shift load away from 'from' to 'to'. */
2576 from->tx_bytes -= delta;
2577 to->tx_bytes += delta;
2579 /* Arrange for flows to be revalidated. */
2580 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2581 hash->iface_idx = to->iface->port_ifidx;
2582 hash->iface_tag = tag_create_random();
2586 bond_rebalance_port(struct port *port)
2588 struct slave_balance bals[DP_MAX_PORTS];
2590 struct bond_entry *hashes[BOND_MASK + 1];
2591 struct slave_balance *b, *from, *to;
2592 struct bond_entry *e;
2595 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2596 * descending order of tx_bytes, so that bals[0] represents the most
2597 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2600 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2601 * array for each slave_balance structure, we sort our local array of
2602 * hashes in order by slave, so that all of the hashes for a given slave
2603 * become contiguous in memory, and then we point each 'hashes' members of
2604 * a slave_balance structure to the start of a contiguous group. */
2605 n_bals = port->n_ifaces;
2606 for (b = bals; b < &bals[n_bals]; b++) {
2607 b->iface = port->ifaces[b - bals];
2612 for (i = 0; i <= BOND_MASK; i++) {
2613 hashes[i] = &port->bond_hash[i];
2615 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2616 for (i = 0; i <= BOND_MASK; i++) {
2618 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2619 b = &bals[e->iface_idx];
2620 b->tx_bytes += e->tx_bytes;
2622 b->hashes = &hashes[i];
2627 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2628 log_bals(bals, n_bals, port);
2630 /* Discard slaves that aren't enabled (which were sorted to the back of the
2631 * array earlier). */
2632 while (!bals[n_bals - 1].iface->enabled) {
2639 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2640 to = &bals[n_bals - 1];
2641 for (from = bals; from < to; ) {
2642 uint64_t overload = from->tx_bytes - to->tx_bytes;
2643 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2644 /* The extra load on 'from' (and all less-loaded slaves), compared
2645 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2646 * it is less than ~1Mbps. No point in rebalancing. */
2648 } else if (from->n_hashes == 1) {
2649 /* 'from' only carries a single MAC hash, so we can't shift any
2650 * load away from it, even though we want to. */
2653 /* 'from' is carrying significantly more load than 'to', and that
2654 * load is split across at least two different hashes. Pick a hash
2655 * to migrate to 'to' (the least-loaded slave), given that doing so
2656 * must decrease the ratio of the load on the two slaves by at
2659 * The sort order we use means that we prefer to shift away the
2660 * smallest hashes instead of the biggest ones. There is little
2661 * reason behind this decision; we could use the opposite sort
2662 * order to shift away big hashes ahead of small ones. */
2666 for (i = 0; i < from->n_hashes; i++) {
2667 double old_ratio, new_ratio;
2668 uint64_t delta = from->hashes[i]->tx_bytes;
2670 if (delta == 0 || from->tx_bytes - delta == 0) {
2671 /* Pointless move. */
2675 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2677 if (to->tx_bytes == 0) {
2678 /* Nothing on the new slave, move it. */
2682 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2683 new_ratio = (double)(from->tx_bytes - delta) /
2684 (to->tx_bytes + delta);
2686 if (new_ratio == 0) {
2687 /* Should already be covered but check to prevent division
2692 if (new_ratio < 1) {
2693 new_ratio = 1 / new_ratio;
2696 if (old_ratio - new_ratio > 0.1) {
2697 /* Would decrease the ratio, move it. */
2701 if (i < from->n_hashes) {
2702 bond_shift_load(from, to, i);
2703 port->bond_compat_is_stale = true;
2705 /* If the result of the migration changed the relative order of
2706 * 'from' and 'to' swap them back to maintain invariants. */
2707 if (order_swapped) {
2708 swap_bals(from, to);
2711 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2712 * point to different slave_balance structures. It is only
2713 * valid to do these two operations in a row at all because we
2714 * know that 'from' will not move past 'to' and vice versa. */
2715 resort_bals(from, bals, n_bals);
2716 resort_bals(to, bals, n_bals);
2723 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2724 * historical data to decay to <1% in 7 rebalancing runs. */
2725 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2731 bond_send_learning_packets(struct port *port)
2733 struct bridge *br = port->bridge;
2734 struct mac_entry *e;
2735 struct ofpbuf packet;
2736 int error, n_packets, n_errors;
2738 if (!port->n_ifaces || port->active_iface < 0) {
2742 ofpbuf_init(&packet, 128);
2743 error = n_packets = n_errors = 0;
2744 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2745 union ofp_action actions[2], *a;
2751 if (e->port == port->port_idx
2752 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2756 /* Compose actions. */
2757 memset(actions, 0, sizeof actions);
2760 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2761 a->vlan_vid.len = htons(sizeof *a);
2762 a->vlan_vid.vlan_vid = htons(e->vlan);
2765 a->output.type = htons(OFPAT_OUTPUT);
2766 a->output.len = htons(sizeof *a);
2767 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2772 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2774 flow_extract(&packet, ODPP_NONE, &flow);
2775 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2782 ofpbuf_uninit(&packet);
2785 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2786 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2787 "packets, last error was: %s",
2788 port->name, n_errors, n_packets, strerror(error));
2790 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2791 port->name, n_packets);
2795 /* Bonding unixctl user interface functions. */
2798 bond_unixctl_list(struct unixctl_conn *conn,
2799 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2801 struct ds ds = DS_EMPTY_INITIALIZER;
2802 const struct bridge *br;
2804 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2806 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2809 for (i = 0; i < br->n_ports; i++) {
2810 const struct port *port = br->ports[i];
2811 if (port->n_ifaces > 1) {
2814 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2815 for (j = 0; j < port->n_ifaces; j++) {
2816 const struct iface *iface = port->ifaces[j];
2818 ds_put_cstr(&ds, ", ");
2820 ds_put_cstr(&ds, iface->name);
2822 ds_put_char(&ds, '\n');
2826 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2830 static struct port *
2831 bond_find(const char *name)
2833 const struct bridge *br;
2835 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2838 for (i = 0; i < br->n_ports; i++) {
2839 struct port *port = br->ports[i];
2840 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2849 bond_unixctl_show(struct unixctl_conn *conn,
2850 const char *args, void *aux OVS_UNUSED)
2852 struct ds ds = DS_EMPTY_INITIALIZER;
2853 const struct port *port;
2856 port = bond_find(args);
2858 unixctl_command_reply(conn, 501, "no such bond");
2862 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2863 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2864 ds_put_format(&ds, "next rebalance: %lld ms\n",
2865 port->bridge->bond_next_rebalance - time_msec());
2866 for (j = 0; j < port->n_ifaces; j++) {
2867 const struct iface *iface = port->ifaces[j];
2868 struct bond_entry *be;
2871 ds_put_format(&ds, "slave %s: %s\n",
2872 iface->name, iface->enabled ? "enabled" : "disabled");
2873 if (j == port->active_iface) {
2874 ds_put_cstr(&ds, "\tactive slave\n");
2876 if (iface->delay_expires != LLONG_MAX) {
2877 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2878 iface->enabled ? "downdelay" : "updelay",
2879 iface->delay_expires - time_msec());
2883 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2884 int hash = be - port->bond_hash;
2885 struct mac_entry *me;
2887 if (be->iface_idx != j) {
2891 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2892 hash, be->tx_bytes / 1024);
2895 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2896 &port->bridge->ml->lrus) {
2899 if (bond_hash(me->mac) == hash
2900 && me->port != port->port_idx
2901 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2902 && dp_ifidx == iface->dp_ifidx)
2904 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2905 ETH_ADDR_ARGS(me->mac));
2910 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2915 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2916 void *aux OVS_UNUSED)
2918 char *args = (char *) args_;
2919 char *save_ptr = NULL;
2920 char *bond_s, *hash_s, *slave_s;
2921 uint8_t mac[ETH_ADDR_LEN];
2923 struct iface *iface;
2924 struct bond_entry *entry;
2927 bond_s = strtok_r(args, " ", &save_ptr);
2928 hash_s = strtok_r(NULL, " ", &save_ptr);
2929 slave_s = strtok_r(NULL, " ", &save_ptr);
2931 unixctl_command_reply(conn, 501,
2932 "usage: bond/migrate BOND HASH SLAVE");
2936 port = bond_find(bond_s);
2938 unixctl_command_reply(conn, 501, "no such bond");
2942 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2943 == ETH_ADDR_SCAN_COUNT) {
2944 hash = bond_hash(mac);
2945 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2946 hash = atoi(hash_s) & BOND_MASK;
2948 unixctl_command_reply(conn, 501, "bad hash");
2952 iface = port_lookup_iface(port, slave_s);
2954 unixctl_command_reply(conn, 501, "no such slave");
2958 if (!iface->enabled) {
2959 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2963 entry = &port->bond_hash[hash];
2964 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2965 entry->iface_idx = iface->port_ifidx;
2966 entry->iface_tag = tag_create_random();
2967 port->bond_compat_is_stale = true;
2968 unixctl_command_reply(conn, 200, "migrated");
2972 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2973 void *aux OVS_UNUSED)
2975 char *args = (char *) args_;
2976 char *save_ptr = NULL;
2977 char *bond_s, *slave_s;
2979 struct iface *iface;
2981 bond_s = strtok_r(args, " ", &save_ptr);
2982 slave_s = strtok_r(NULL, " ", &save_ptr);
2984 unixctl_command_reply(conn, 501,
2985 "usage: bond/set-active-slave BOND SLAVE");
2989 port = bond_find(bond_s);
2991 unixctl_command_reply(conn, 501, "no such bond");
2995 iface = port_lookup_iface(port, slave_s);
2997 unixctl_command_reply(conn, 501, "no such slave");
3001 if (!iface->enabled) {
3002 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3006 if (port->active_iface != iface->port_ifidx) {
3007 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3008 port->active_iface = iface->port_ifidx;
3009 port->active_iface_tag = tag_create_random();
3010 VLOG_INFO("port %s: active interface is now %s",
3011 port->name, iface->name);
3012 bond_send_learning_packets(port);
3013 unixctl_command_reply(conn, 200, "done");
3015 unixctl_command_reply(conn, 200, "no change");
3020 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3022 char *args = (char *) args_;
3023 char *save_ptr = NULL;
3024 char *bond_s, *slave_s;
3026 struct iface *iface;
3028 bond_s = strtok_r(args, " ", &save_ptr);
3029 slave_s = strtok_r(NULL, " ", &save_ptr);
3031 unixctl_command_reply(conn, 501,
3032 "usage: bond/enable/disable-slave BOND SLAVE");
3036 port = bond_find(bond_s);
3038 unixctl_command_reply(conn, 501, "no such bond");
3042 iface = port_lookup_iface(port, slave_s);
3044 unixctl_command_reply(conn, 501, "no such slave");
3048 bond_enable_slave(iface, enable);
3049 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3053 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3054 void *aux OVS_UNUSED)
3056 enable_slave(conn, args, true);
3060 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3061 void *aux OVS_UNUSED)
3063 enable_slave(conn, args, false);
3067 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3068 void *aux OVS_UNUSED)
3070 uint8_t mac[ETH_ADDR_LEN];
3074 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3075 == ETH_ADDR_SCAN_COUNT) {
3076 hash = bond_hash(mac);
3078 hash_cstr = xasprintf("%u", hash);
3079 unixctl_command_reply(conn, 200, hash_cstr);
3082 unixctl_command_reply(conn, 501, "invalid mac");
3089 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3090 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3091 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3092 unixctl_command_register("bond/set-active-slave",
3093 bond_unixctl_set_active_slave, NULL);
3094 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3096 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3098 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3101 /* Port functions. */
3103 static struct port *
3104 port_create(struct bridge *br, const char *name)
3108 port = xzalloc(sizeof *port);
3110 port->port_idx = br->n_ports;
3112 port->trunks = NULL;
3113 port->name = xstrdup(name);
3114 port->active_iface = -1;
3116 if (br->n_ports >= br->allocated_ports) {
3117 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3120 br->ports[br->n_ports++] = port;
3122 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3129 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3131 struct shash old_ifaces, new_ifaces;
3132 struct shash_node *node;
3133 unsigned long *trunks;
3139 /* Collect old and new interfaces. */
3140 shash_init(&old_ifaces);
3141 shash_init(&new_ifaces);
3142 for (i = 0; i < port->n_ifaces; i++) {
3143 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3145 for (i = 0; i < cfg->n_interfaces; i++) {
3146 const char *name = cfg->interfaces[i]->name;
3147 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3148 VLOG_WARN("port %s: %s specified twice as port interface",
3152 port->updelay = cfg->bond_updelay;
3153 if (port->updelay < 0) {
3156 port->updelay = cfg->bond_downdelay;
3157 if (port->downdelay < 0) {
3158 port->downdelay = 0;
3161 /* Get rid of deleted interfaces and add new interfaces. */
3162 SHASH_FOR_EACH (node, &old_ifaces) {
3163 if (!shash_find(&new_ifaces, node->name)) {
3164 iface_destroy(node->data);
3167 SHASH_FOR_EACH (node, &new_ifaces) {
3168 const struct ovsrec_interface *if_cfg = node->data;
3169 struct iface *iface;
3171 iface = shash_find_data(&old_ifaces, if_cfg->name);
3173 iface_create(port, if_cfg);
3175 iface->cfg = if_cfg;
3182 if (port->n_ifaces < 2) {
3184 if (vlan >= 0 && vlan <= 4095) {
3185 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3190 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3191 * they even work as-is. But they have not been tested. */
3192 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3196 if (port->vlan != vlan) {
3198 bridge_flush(port->bridge);
3201 /* Get trunked VLANs. */
3207 trunks = bitmap_allocate(4096);
3209 for (i = 0; i < cfg->n_trunks; i++) {
3210 int trunk = cfg->trunks[i];
3212 bitmap_set1(trunks, trunk);
3218 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3219 port->name, cfg->n_trunks);
3221 if (n_errors == cfg->n_trunks) {
3223 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3226 bitmap_set_multiple(trunks, 0, 4096, 1);
3229 if (cfg->n_trunks) {
3230 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3235 ? port->trunks != NULL
3236 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3237 bridge_flush(port->bridge);
3239 bitmap_free(port->trunks);
3240 port->trunks = trunks;
3242 shash_destroy(&old_ifaces);
3243 shash_destroy(&new_ifaces);
3247 port_destroy(struct port *port)
3250 struct bridge *br = port->bridge;
3254 proc_net_compat_update_vlan(port->name, NULL, 0);
3255 proc_net_compat_update_bond(port->name, NULL);
3257 for (i = 0; i < MAX_MIRRORS; i++) {
3258 struct mirror *m = br->mirrors[i];
3259 if (m && m->out_port == port) {
3264 while (port->n_ifaces > 0) {
3265 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3268 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3269 del->port_idx = port->port_idx;
3272 bitmap_free(port->trunks);
3279 static struct port *
3280 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3282 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3283 return iface ? iface->port : NULL;
3286 static struct port *
3287 port_lookup(const struct bridge *br, const char *name)
3291 for (i = 0; i < br->n_ports; i++) {
3292 struct port *port = br->ports[i];
3293 if (!strcmp(port->name, name)) {
3300 static struct iface *
3301 port_lookup_iface(const struct port *port, const char *name)
3305 for (j = 0; j < port->n_ifaces; j++) {
3306 struct iface *iface = port->ifaces[j];
3307 if (!strcmp(iface->name, name)) {
3315 port_update_bonding(struct port *port)
3317 if (port->n_ifaces < 2) {
3318 /* Not a bonded port. */
3319 if (port->bond_hash) {
3320 free(port->bond_hash);
3321 port->bond_hash = NULL;
3322 port->bond_compat_is_stale = true;
3323 port->bond_fake_iface = false;
3326 if (!port->bond_hash) {
3329 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3330 for (i = 0; i <= BOND_MASK; i++) {
3331 struct bond_entry *e = &port->bond_hash[i];
3335 port->no_ifaces_tag = tag_create_random();
3336 bond_choose_active_iface(port);
3338 port->bond_compat_is_stale = true;
3339 port->bond_fake_iface = port->cfg->bond_fake_iface;
3344 port_update_bond_compat(struct port *port)
3346 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3347 struct compat_bond bond;
3350 if (port->n_ifaces < 2) {
3351 proc_net_compat_update_bond(port->name, NULL);
3356 bond.updelay = port->updelay;
3357 bond.downdelay = port->downdelay;
3360 bond.hashes = compat_hashes;
3361 if (port->bond_hash) {
3362 const struct bond_entry *e;
3363 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3364 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3365 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3366 cbh->hash = e - port->bond_hash;
3367 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3372 bond.n_slaves = port->n_ifaces;
3373 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3374 for (i = 0; i < port->n_ifaces; i++) {
3375 struct iface *iface = port->ifaces[i];
3376 struct compat_bond_slave *slave = &bond.slaves[i];
3377 slave->name = iface->name;
3379 /* We need to make the same determination as the Linux bonding
3380 * code to determine whether a slave should be consider "up".
3381 * The Linux function bond_miimon_inspect() supports four
3382 * BOND_LINK_* states:
3384 * - BOND_LINK_UP: carrier detected, updelay has passed.
3385 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3386 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3387 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3389 * The function bond_info_show_slave() only considers BOND_LINK_UP
3390 * to be "up" and anything else to be "down".
3392 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3396 netdev_get_etheraddr(iface->netdev, slave->mac);
3399 if (port->bond_fake_iface) {
3400 struct netdev *bond_netdev;
3402 if (!netdev_open_default(port->name, &bond_netdev)) {
3404 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3406 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3408 netdev_close(bond_netdev);
3412 proc_net_compat_update_bond(port->name, &bond);
3417 port_update_vlan_compat(struct port *port)
3419 struct bridge *br = port->bridge;
3420 char *vlandev_name = NULL;
3422 if (port->vlan > 0) {
3423 /* Figure out the name that the VLAN device should actually have, if it
3424 * existed. This takes some work because the VLAN device would not
3425 * have port->name in its name; rather, it would have the trunk port's
3426 * name, and 'port' would be attached to a bridge that also had the
3427 * VLAN device one of its ports. So we need to find a trunk port that
3428 * includes port->vlan.
3430 * There might be more than one candidate. This doesn't happen on
3431 * XenServer, so if it happens we just pick the first choice in
3432 * alphabetical order instead of creating multiple VLAN devices. */
3434 for (i = 0; i < br->n_ports; i++) {
3435 struct port *p = br->ports[i];
3436 if (port_trunks_vlan(p, port->vlan)
3438 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3440 uint8_t ea[ETH_ADDR_LEN];
3441 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3442 if (!eth_addr_is_multicast(ea) &&
3443 !eth_addr_is_reserved(ea) &&
3444 !eth_addr_is_zero(ea)) {
3445 vlandev_name = p->name;
3450 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3453 /* Interface functions. */
3455 static struct iface *
3456 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3458 struct iface *iface;
3459 char *name = if_cfg->name;
3462 iface = xzalloc(sizeof *iface);
3464 iface->port_ifidx = port->n_ifaces;
3465 iface->name = xstrdup(name);
3466 iface->dp_ifidx = -1;
3467 iface->tag = tag_create_random();
3468 iface->delay_expires = LLONG_MAX;
3469 iface->netdev = NULL;
3470 iface->cfg = if_cfg;
3472 if (port->n_ifaces >= port->allocated_ifaces) {
3473 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3474 sizeof *port->ifaces);
3476 port->ifaces[port->n_ifaces++] = iface;
3477 if (port->n_ifaces > 1) {
3478 port->bridge->has_bonded_ports = true;
3481 /* Attempt to create the network interface in case it
3482 * doesn't exist yet. */
3483 if (!iface_is_internal(port->bridge, iface->name)) {
3484 error = set_up_iface(if_cfg, iface, true);
3486 VLOG_WARN("could not create iface %s: %s", iface->name,
3491 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3493 bridge_flush(port->bridge);
3499 iface_destroy(struct iface *iface)
3502 struct port *port = iface->port;
3503 struct bridge *br = port->bridge;
3504 bool del_active = port->active_iface == iface->port_ifidx;
3507 if (iface->dp_ifidx >= 0) {
3508 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3511 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3512 del->port_ifidx = iface->port_ifidx;
3514 netdev_close(iface->netdev);
3517 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3518 bond_choose_active_iface(port);
3519 bond_send_learning_packets(port);
3525 bridge_flush(port->bridge);
3529 static struct iface *
3530 iface_lookup(const struct bridge *br, const char *name)
3534 for (i = 0; i < br->n_ports; i++) {
3535 struct port *port = br->ports[i];
3536 for (j = 0; j < port->n_ifaces; j++) {
3537 struct iface *iface = port->ifaces[j];
3538 if (!strcmp(iface->name, name)) {
3546 static struct iface *
3547 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3549 return port_array_get(&br->ifaces, dp_ifidx);
3552 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3553 * 'br', that is, an interface that is entirely simulated within the datapath.
3554 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3555 * interfaces are created by setting "iface.<iface>.internal = true".
3557 * In addition, we have a kluge-y feature that creates an internal port with
3558 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3559 * This feature needs to go away in the long term. Until then, this is one
3560 * reason why this function takes a name instead of a struct iface: the fake
3561 * interfaces created this way do not have a struct iface. */
3563 iface_is_internal(const struct bridge *br, const char *if_name)
3565 /* XXX wastes time */
3566 struct iface *iface;
3569 if (!strcmp(if_name, br->name)) {
3573 iface = iface_lookup(br, if_name);
3574 if (iface && !strcmp(iface->cfg->type, "internal")) {
3578 port = port_lookup(br, if_name);
3579 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3585 /* Set Ethernet address of 'iface', if one is specified in the configuration
3588 iface_set_mac(struct iface *iface)
3590 uint8_t ea[ETH_ADDR_LEN];
3592 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3593 if (eth_addr_is_multicast(ea)) {
3594 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3596 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3597 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3598 iface->name, iface->name);
3600 int error = netdev_set_etheraddr(iface->netdev, ea);
3602 VLOG_ERR("interface %s: setting MAC failed (%s)",
3603 iface->name, strerror(error));
3609 /* Port mirroring. */
3612 mirror_reconfigure(struct bridge *br)
3614 struct shash old_mirrors, new_mirrors;
3615 struct shash_node *node;
3616 unsigned long *rspan_vlans;
3619 /* Collect old mirrors. */
3620 shash_init(&old_mirrors);
3621 for (i = 0; i < MAX_MIRRORS; i++) {
3622 if (br->mirrors[i]) {
3623 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3627 /* Collect new mirrors. */
3628 shash_init(&new_mirrors);
3629 for (i = 0; i < br->cfg->n_mirrors; i++) {
3630 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3631 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3632 VLOG_WARN("bridge %s: %s specified twice as mirror",
3633 br->name, cfg->name);
3637 /* Get rid of deleted mirrors and add new mirrors. */
3638 SHASH_FOR_EACH (node, &old_mirrors) {
3639 if (!shash_find(&new_mirrors, node->name)) {
3640 mirror_destroy(node->data);
3643 SHASH_FOR_EACH (node, &new_mirrors) {
3644 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3646 mirror = mirror_create(br, node->name);
3651 mirror_reconfigure_one(mirror, node->data);
3653 shash_destroy(&old_mirrors);
3654 shash_destroy(&new_mirrors);
3656 /* Update port reserved status. */
3657 for (i = 0; i < br->n_ports; i++) {
3658 br->ports[i]->is_mirror_output_port = false;
3660 for (i = 0; i < MAX_MIRRORS; i++) {
3661 struct mirror *m = br->mirrors[i];
3662 if (m && m->out_port) {
3663 m->out_port->is_mirror_output_port = true;
3667 /* Update flooded vlans (for RSPAN). */
3669 if (br->cfg->n_flood_vlans) {
3670 rspan_vlans = bitmap_allocate(4096);
3672 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3673 int64_t vlan = br->cfg->flood_vlans[i];
3674 if (vlan >= 0 && vlan < 4096) {
3675 bitmap_set1(rspan_vlans, vlan);
3676 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3679 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3684 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3689 static struct mirror *
3690 mirror_create(struct bridge *br, const char *name)
3695 for (i = 0; ; i++) {
3696 if (i >= MAX_MIRRORS) {
3697 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3698 "cannot create %s", br->name, MAX_MIRRORS, name);
3701 if (!br->mirrors[i]) {
3706 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3709 br->mirrors[i] = m = xzalloc(sizeof *m);
3712 m->name = xstrdup(name);
3713 shash_init(&m->src_ports);
3714 shash_init(&m->dst_ports);
3724 mirror_destroy(struct mirror *m)
3727 struct bridge *br = m->bridge;
3730 for (i = 0; i < br->n_ports; i++) {
3731 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3732 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3735 shash_destroy(&m->src_ports);
3736 shash_destroy(&m->dst_ports);
3739 m->bridge->mirrors[m->idx] = NULL;
3747 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3748 struct shash *names)
3752 for (i = 0; i < n_ports; i++) {
3753 const char *name = ports[i]->name;
3754 if (port_lookup(m->bridge, name)) {
3755 shash_add_once(names, name, NULL);
3757 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3758 "port %s", m->bridge->name, m->name, name);
3764 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3770 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3772 for (i = 0; i < cfg->n_select_vlan; i++) {
3773 int64_t vlan = cfg->select_vlan[i];
3774 if (vlan < 0 || vlan > 4095) {
3775 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3776 m->bridge->name, m->name, vlan);
3778 (*vlans)[n_vlans++] = vlan;
3785 vlan_is_mirrored(const struct mirror *m, int vlan)
3789 for (i = 0; i < m->n_vlans; i++) {
3790 if (m->vlans[i] == vlan) {
3798 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3802 for (i = 0; i < m->n_vlans; i++) {
3803 if (port_trunks_vlan(p, m->vlans[i])) {
3811 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3813 struct shash src_ports, dst_ports;
3814 mirror_mask_t mirror_bit;
3815 struct port *out_port;
3820 bool mirror_all_ports;
3821 bool any_ports_specified;
3822 bool any_vlans_specified;
3824 /* Get output port. */
3825 if (cfg->output_port) {
3826 out_port = port_lookup(m->bridge, cfg->output_port->name);
3828 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3829 m->bridge->name, m->name);
3835 if (cfg->output_vlan) {
3836 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3837 "output vlan; ignoring output vlan",
3838 m->bridge->name, m->name);
3840 } else if (cfg->output_vlan) {
3842 out_vlan = *cfg->output_vlan;
3844 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3845 m->bridge->name, m->name);
3850 /* Get all the ports, and drop duplicates and ports that don't exist. */
3851 shash_init(&src_ports);
3852 shash_init(&dst_ports);
3853 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3855 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3857 any_ports_specified = cfg->n_select_dst_port || cfg->n_select_dst_port;
3858 if (any_ports_specified
3859 && shash_is_empty(&src_ports) && shash_is_empty(&dst_ports)) {
3860 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3861 "selection ports exists", m->bridge->name, m->name);
3866 /* Get all the vlans, and drop duplicate and invalid vlans. */
3867 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3868 any_vlans_specified = cfg->n_select_vlan > 0;
3869 if (any_vlans_specified && !n_vlans) {
3870 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3871 "VLANs exists", m->bridge->name, m->name);
3876 /* Update mirror data. */
3877 if (!shash_equal_keys(&m->src_ports, &src_ports)
3878 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3879 || m->n_vlans != n_vlans
3880 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3881 || m->out_port != out_port
3882 || m->out_vlan != out_vlan) {
3883 bridge_flush(m->bridge);
3885 shash_swap(&m->src_ports, &src_ports);
3886 shash_swap(&m->dst_ports, &dst_ports);
3889 m->n_vlans = n_vlans;
3890 m->out_port = out_port;
3891 m->out_vlan = out_vlan;
3893 /* If no selection criteria have been given, mirror for all ports. */
3894 mirror_all_ports = !any_ports_specified && !any_vlans_specified;
3897 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3898 for (i = 0; i < m->bridge->n_ports; i++) {
3899 struct port *port = m->bridge->ports[i];
3901 if (mirror_all_ports
3902 || shash_find(&m->src_ports, port->name)
3905 ? port_trunks_any_mirrored_vlan(m, port)
3906 : vlan_is_mirrored(m, port->vlan)))) {
3907 port->src_mirrors |= mirror_bit;
3909 port->src_mirrors &= ~mirror_bit;
3912 if (mirror_all_ports || shash_find(&m->dst_ports, port->name)) {
3913 port->dst_mirrors |= mirror_bit;
3915 port->dst_mirrors &= ~mirror_bit;
3921 shash_destroy(&src_ports);
3922 shash_destroy(&dst_ports);