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>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
50 #include "poll-loop.h"
51 #include "port-array.h"
52 #include "proc-net-compat.h"
56 #include "socket-util.h"
57 #include "stream-ssl.h"
63 #include "vswitchd/vswitch-idl.h"
64 #include "xenserver.h"
66 #include "sflow_api.h"
68 #define THIS_MODULE VLM_bridge
77 /* These members are always valid. */
78 struct port *port; /* Containing port. */
79 size_t port_ifidx; /* Index within containing port. */
80 char *name; /* Host network device name. */
81 tag_type tag; /* Tag associated with this interface. */
82 long long delay_expires; /* Time after which 'enabled' may change. */
84 /* These members are valid only after bridge_reconfigure() causes them to
86 int dp_ifidx; /* Index within kernel datapath. */
87 struct netdev *netdev; /* Network device. */
88 bool enabled; /* May be chosen for flows? */
90 /* This member is only valid *during* bridge_reconfigure(). */
91 const struct ovsrec_interface *cfg;
94 #define BOND_MASK 0xff
96 int iface_idx; /* Index of assigned iface, or -1 if none. */
97 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
98 tag_type iface_tag; /* Tag associated with iface_idx. */
101 #define MAX_MIRRORS 32
102 typedef uint32_t mirror_mask_t;
103 #define MIRROR_MASK_C(X) UINT32_C(X)
104 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
106 struct bridge *bridge;
110 /* Selection criteria. */
111 struct shash src_ports; /* Name is port name; data is always NULL. */
112 struct shash dst_ports; /* Name is port name; data is always NULL. */
117 struct port *out_port;
121 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
123 struct bridge *bridge;
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
130 /* An ordinary bridge port has 1 interface.
131 * A bridge port for bonding has at least 2 interfaces. */
132 struct iface **ifaces;
133 size_t n_ifaces, allocated_ifaces;
136 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
137 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
138 tag_type active_iface_tag; /* Tag for bcast flows. */
139 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
140 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
141 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
142 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
143 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
144 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
145 long long int bond_next_rebalance; /* Next rebalancing time. */
147 /* Port mirroring info. */
148 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
149 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
150 bool is_mirror_output_port; /* Does port mirroring send frames here? */
152 /* This member is only valid *during* bridge_reconfigure(). */
153 const struct ovsrec_port *cfg;
156 #define DP_MAX_PORTS 255
158 struct list node; /* Node in global list of bridges. */
159 char *name; /* User-specified arbitrary name. */
160 struct mac_learning *ml; /* MAC learning table. */
161 bool sent_config_request; /* Successfully sent config request? */
162 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
164 /* OpenFlow switch processing. */
165 struct ofproto *ofproto; /* OpenFlow switch. */
167 /* Description strings. */
168 char *mfr_desc; /* Manufacturer. */
169 char *hw_desc; /* Hardware. */
170 char *sw_desc; /* Software version. */
171 char *serial_desc; /* Serial number. */
172 char *dp_desc; /* Datapath description. */
174 /* Kernel datapath information. */
175 struct dpif *dpif; /* Datapath. */
176 struct port_array ifaces; /* Indexed by kernel datapath port number. */
180 size_t n_ports, allocated_ports;
181 struct shash iface_by_name; /* "struct iface"s indexed by name. */
182 struct shash port_by_name; /* "struct port"s indexed by name. */
185 bool has_bonded_ports;
190 /* Flow statistics gathering. */
191 time_t next_stats_request;
193 /* Port mirroring. */
194 struct mirror *mirrors[MAX_MIRRORS];
196 /* This member is only valid *during* bridge_reconfigure(). */
197 const struct ovsrec_bridge *cfg;
200 /* List of all bridges. */
201 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
203 /* Maximum number of datapaths. */
204 enum { DP_MAX = 256 };
206 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
207 static void bridge_destroy(struct bridge *);
208 static struct bridge *bridge_lookup(const char *name);
209 static unixctl_cb_func bridge_unixctl_dump_flows;
210 static int bridge_run_one(struct bridge *);
211 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
212 const struct bridge *br,
213 struct ovsrec_controller ***controllersp);
214 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
216 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
218 const struct sockaddr_in *managers,
220 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
221 static void bridge_fetch_dp_ifaces(struct bridge *);
222 static void bridge_flush(struct bridge *);
223 static void bridge_pick_local_hw_addr(struct bridge *,
224 uint8_t ea[ETH_ADDR_LEN],
225 struct iface **hw_addr_iface);
226 static uint64_t bridge_pick_datapath_id(struct bridge *,
227 const uint8_t bridge_ea[ETH_ADDR_LEN],
228 struct iface *hw_addr_iface);
229 static struct iface *bridge_get_local_iface(struct bridge *);
230 static uint64_t dpid_from_hash(const void *, size_t nbytes);
232 static unixctl_cb_func bridge_unixctl_fdb_show;
234 static void bond_init(void);
235 static void bond_run(struct bridge *);
236 static void bond_wait(struct bridge *);
237 static void bond_rebalance_port(struct port *);
238 static void bond_send_learning_packets(struct port *);
239 static void bond_enable_slave(struct iface *iface, bool enable);
241 static struct port *port_create(struct bridge *, const char *name);
242 static void port_reconfigure(struct port *, const struct ovsrec_port *);
243 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
244 static void port_destroy(struct port *);
245 static struct port *port_lookup(const struct bridge *, const char *name);
246 static struct iface *port_lookup_iface(const struct port *, const char *name);
247 static struct port *port_from_dp_ifidx(const struct bridge *,
249 static void port_update_bond_compat(struct port *);
250 static void port_update_vlan_compat(struct port *);
251 static void port_update_bonding(struct port *);
253 static struct mirror *mirror_create(struct bridge *, const char *name);
254 static void mirror_destroy(struct mirror *);
255 static void mirror_reconfigure(struct bridge *);
256 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
257 static bool vlan_is_mirrored(const struct mirror *, int vlan);
259 static struct iface *iface_create(struct port *port,
260 const struct ovsrec_interface *if_cfg);
261 static void iface_destroy(struct iface *);
262 static struct iface *iface_lookup(const struct bridge *, const char *name);
263 static struct iface *iface_from_dp_ifidx(const struct bridge *,
265 static bool iface_is_internal(const struct bridge *, const char *name);
266 static void iface_set_mac(struct iface *);
268 /* Hooks into ofproto processing. */
269 static struct ofhooks bridge_ofhooks;
271 /* Public functions. */
273 /* Adds the name of each interface used by a bridge, including local and
274 * internal ports, to 'svec'. */
276 bridge_get_ifaces(struct svec *svec)
278 struct bridge *br, *next;
281 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
282 for (i = 0; i < br->n_ports; i++) {
283 struct port *port = br->ports[i];
285 for (j = 0; j < port->n_ifaces; j++) {
286 struct iface *iface = port->ifaces[j];
287 if (iface->dp_ifidx < 0) {
288 VLOG_ERR("%s interface not in datapath %s, ignoring",
289 iface->name, dpif_name(br->dpif));
291 if (iface->dp_ifidx != ODPP_LOCAL) {
292 svec_add(svec, iface->name);
301 bridge_init(const struct ovsrec_open_vswitch *cfg)
303 struct svec bridge_names;
304 struct svec dpif_names, dpif_types;
307 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
309 svec_init(&bridge_names);
310 for (i = 0; i < cfg->n_bridges; i++) {
311 svec_add(&bridge_names, cfg->bridges[i]->name);
313 svec_sort(&bridge_names);
315 svec_init(&dpif_names);
316 svec_init(&dpif_types);
317 dp_enumerate_types(&dpif_types);
318 for (i = 0; i < dpif_types.n; i++) {
323 dp_enumerate_names(dpif_types.names[i], &dpif_names);
325 for (j = 0; j < dpif_names.n; j++) {
326 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
328 struct svec all_names;
331 svec_init(&all_names);
332 dpif_get_all_names(dpif, &all_names);
333 for (k = 0; k < all_names.n; k++) {
334 if (svec_contains(&bridge_names, all_names.names[k])) {
340 svec_destroy(&all_names);
345 svec_destroy(&bridge_names);
346 svec_destroy(&dpif_names);
347 svec_destroy(&dpif_types);
349 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
353 bridge_reconfigure(cfg);
358 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
360 /* XXX SSL should be configurable on a per-bridge basis. */
362 stream_ssl_set_private_key_file(ssl->private_key);
363 stream_ssl_set_certificate_file(ssl->certificate);
364 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
369 /* Attempt to create the network device 'iface_name' through the netdev
372 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
375 struct shash_node *node;
376 struct shash options;
380 shash_init(&options);
381 for (i = 0; i < iface_cfg->n_options; i++) {
382 shash_add(&options, iface_cfg->key_options[i],
383 xstrdup(iface_cfg->value_options[i]));
387 struct netdev_options netdev_options;
389 memset(&netdev_options, 0, sizeof netdev_options);
390 netdev_options.name = iface_cfg->name;
391 if (!strcmp(iface_cfg->type, "internal")) {
392 /* An "internal" config type maps to a netdev "system" type. */
393 netdev_options.type = "system";
395 netdev_options.type = iface_cfg->type;
397 netdev_options.args = &options;
398 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
400 error = netdev_open(&netdev_options, &iface->netdev);
403 netdev_get_carrier(iface->netdev, &iface->enabled);
405 } else if (iface->netdev) {
406 const char *netdev_type = netdev_get_type(iface->netdev);
407 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
408 ? iface_cfg->type : NULL;
410 /* An "internal" config type maps to a netdev "system" type. */
411 if (iface_type && !strcmp(iface_type, "internal")) {
412 iface_type = "system";
415 if (!iface_type || !strcmp(netdev_type, iface_type)) {
416 error = netdev_reconfigure(iface->netdev, &options);
418 VLOG_WARN("%s: attempting change device type from %s to %s",
419 iface_cfg->name, netdev_type, iface_type);
424 SHASH_FOR_EACH (node, &options) {
427 shash_destroy(&options);
433 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
435 return set_up_iface(iface_cfg, iface, false);
439 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
440 void *aux OVS_UNUSED)
442 if (!iface->netdev) {
443 int error = set_up_iface(iface->cfg, iface, true);
445 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
455 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
456 void *aux OVS_UNUSED)
458 if (iface->dp_ifidx >= 0) {
459 VLOG_DBG("%s has interface %s on port %d",
461 iface->name, iface->dp_ifidx);
464 VLOG_ERR("%s interface not in %s, dropping",
465 iface->name, dpif_name(br->dpif));
471 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
472 void *aux OVS_UNUSED)
474 /* Set policing attributes. */
475 netdev_set_policing(iface->netdev,
476 iface->cfg->ingress_policing_rate,
477 iface->cfg->ingress_policing_burst);
479 /* Set MAC address of internal interfaces other than the local
481 if (iface->dp_ifidx != ODPP_LOCAL
482 && iface_is_internal(br, iface->name)) {
483 iface_set_mac(iface);
489 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
490 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
491 * deletes from 'br' any ports that no longer have any interfaces. */
493 iterate_and_prune_ifaces(struct bridge *br,
494 bool (*cb)(struct bridge *, struct iface *,
500 for (i = 0; i < br->n_ports; ) {
501 struct port *port = br->ports[i];
502 for (j = 0; j < port->n_ifaces; ) {
503 struct iface *iface = port->ifaces[j];
504 if (cb(br, iface, aux)) {
507 iface_destroy(iface);
511 if (port->n_ifaces) {
514 VLOG_ERR("%s port has no interfaces, dropping", port->name);
520 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
521 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
522 * responsible for freeing '*managersp' (with free()).
524 * You may be asking yourself "why does ovs-vswitchd care?", because
525 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
526 * should not be and in fact is not directly involved in that. But
527 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
528 * it has to tell in-band control where the managers are to enable that.
531 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
532 struct sockaddr_in **managersp, size_t *n_managersp)
534 struct sockaddr_in *managers = NULL;
535 size_t n_managers = 0;
537 if (ovs_cfg->n_managers > 0) {
540 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
541 for (i = 0; i < ovs_cfg->n_managers; i++) {
542 const char *name = ovs_cfg->managers[i];
543 struct sockaddr_in *sin = &managers[i];
545 if ((!strncmp(name, "tcp:", 4)
546 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
547 (!strncmp(name, "ssl:", 4)
548 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
554 *managersp = managers;
555 *n_managersp = n_managers;
559 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
561 struct ovsdb_idl_txn *txn;
562 struct shash old_br, new_br;
563 struct shash_node *node;
564 struct bridge *br, *next;
565 struct sockaddr_in *managers;
568 int sflow_bridge_number;
570 COVERAGE_INC(bridge_reconfigure);
572 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
574 collect_managers(ovs_cfg, &managers, &n_managers);
576 /* Collect old and new bridges. */
579 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
580 shash_add(&old_br, br->name, br);
582 for (i = 0; i < ovs_cfg->n_bridges; i++) {
583 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
584 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
585 VLOG_WARN("more than one bridge named %s", br_cfg->name);
589 /* Get rid of deleted bridges and add new bridges. */
590 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
591 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
598 SHASH_FOR_EACH (node, &new_br) {
599 const char *br_name = node->name;
600 const struct ovsrec_bridge *br_cfg = node->data;
601 br = shash_find_data(&old_br, br_name);
603 /* If the bridge datapath type has changed, we need to tear it
604 * down and recreate. */
605 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
607 bridge_create(br_cfg);
610 bridge_create(br_cfg);
613 shash_destroy(&old_br);
614 shash_destroy(&new_br);
618 bridge_configure_ssl(ovs_cfg->ssl);
621 /* Reconfigure all bridges. */
622 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
623 bridge_reconfigure_one(ovs_cfg, br);
626 /* Add and delete ports on all datapaths.
628 * The kernel will reject any attempt to add a given port to a datapath if
629 * that port already belongs to a different datapath, so we must do all
630 * port deletions before any port additions. */
631 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
632 struct odp_port *dpif_ports;
634 struct shash want_ifaces;
636 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
637 bridge_get_all_ifaces(br, &want_ifaces);
638 for (i = 0; i < n_dpif_ports; i++) {
639 const struct odp_port *p = &dpif_ports[i];
640 if (!shash_find(&want_ifaces, p->devname)
641 && strcmp(p->devname, br->name)) {
642 int retval = dpif_port_del(br->dpif, p->port);
644 VLOG_ERR("failed to remove %s interface from %s: %s",
645 p->devname, dpif_name(br->dpif),
650 shash_destroy(&want_ifaces);
653 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
654 struct odp_port *dpif_ports;
656 struct shash cur_ifaces, want_ifaces;
657 struct shash_node *node;
659 /* Get the set of interfaces currently in this datapath. */
660 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
661 shash_init(&cur_ifaces);
662 for (i = 0; i < n_dpif_ports; i++) {
663 const char *name = dpif_ports[i].devname;
664 if (!shash_find(&cur_ifaces, name)) {
665 shash_add(&cur_ifaces, name, NULL);
670 /* Get the set of interfaces we want on this datapath. */
671 bridge_get_all_ifaces(br, &want_ifaces);
673 SHASH_FOR_EACH (node, &want_ifaces) {
674 const char *if_name = node->name;
675 struct iface *iface = node->data;
677 if (shash_find(&cur_ifaces, if_name)) {
678 /* Already exists, just reconfigure it. */
680 reconfigure_iface(iface->cfg, iface);
683 /* Need to add to datapath. */
687 /* Add to datapath. */
688 internal = iface_is_internal(br, if_name);
689 error = dpif_port_add(br->dpif, if_name,
690 internal ? ODP_PORT_INTERNAL : 0, NULL);
691 if (error == EFBIG) {
692 VLOG_ERR("ran out of valid port numbers on %s",
693 dpif_name(br->dpif));
696 VLOG_ERR("failed to add %s interface to %s: %s",
697 if_name, dpif_name(br->dpif), strerror(error));
701 shash_destroy(&cur_ifaces);
702 shash_destroy(&want_ifaces);
704 sflow_bridge_number = 0;
705 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
708 struct iface *local_iface;
709 struct iface *hw_addr_iface;
712 bridge_fetch_dp_ifaces(br);
714 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
715 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
717 /* Pick local port hardware address, datapath ID. */
718 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
719 local_iface = bridge_get_local_iface(br);
721 int error = netdev_set_etheraddr(local_iface->netdev, ea);
723 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
724 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
725 "Ethernet address: %s",
726 br->name, strerror(error));
730 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
731 ofproto_set_datapath_id(br->ofproto, dpid);
733 dpid_string = xasprintf("%012"PRIx64, dpid);
734 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
737 /* Set NetFlow configuration on this bridge. */
738 if (br->cfg->netflow) {
739 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
740 struct netflow_options opts;
742 memset(&opts, 0, sizeof opts);
744 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
745 if (nf_cfg->engine_type) {
746 opts.engine_type = *nf_cfg->engine_type;
748 if (nf_cfg->engine_id) {
749 opts.engine_id = *nf_cfg->engine_id;
752 opts.active_timeout = nf_cfg->active_timeout;
753 if (!opts.active_timeout) {
754 opts.active_timeout = -1;
755 } else if (opts.active_timeout < 0) {
756 VLOG_WARN("bridge %s: active timeout interval set to negative "
757 "value, using default instead (%d seconds)", br->name,
758 NF_ACTIVE_TIMEOUT_DEFAULT);
759 opts.active_timeout = -1;
762 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
763 if (opts.add_id_to_iface) {
764 if (opts.engine_id > 0x7f) {
765 VLOG_WARN("bridge %s: netflow port mangling may conflict "
766 "with another vswitch, choose an engine id less "
767 "than 128", br->name);
769 if (br->n_ports > 508) {
770 VLOG_WARN("bridge %s: netflow port mangling will conflict "
771 "with another port when more than 508 ports are "
776 opts.collectors.n = nf_cfg->n_targets;
777 opts.collectors.names = nf_cfg->targets;
778 if (ofproto_set_netflow(br->ofproto, &opts)) {
779 VLOG_ERR("bridge %s: problem setting netflow collectors",
783 ofproto_set_netflow(br->ofproto, NULL);
786 /* Set sFlow configuration on this bridge. */
787 if (br->cfg->sflow) {
788 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
789 struct ovsrec_controller **controllers;
790 struct ofproto_sflow_options oso;
791 size_t n_controllers;
794 memset(&oso, 0, sizeof oso);
796 oso.targets.n = sflow_cfg->n_targets;
797 oso.targets.names = sflow_cfg->targets;
799 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
800 if (sflow_cfg->sampling) {
801 oso.sampling_rate = *sflow_cfg->sampling;
804 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
805 if (sflow_cfg->polling) {
806 oso.polling_interval = *sflow_cfg->polling;
809 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
810 if (sflow_cfg->header) {
811 oso.header_len = *sflow_cfg->header;
814 oso.sub_id = sflow_bridge_number++;
815 oso.agent_device = sflow_cfg->agent;
817 oso.control_ip = NULL;
818 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
819 for (i = 0; i < n_controllers; i++) {
820 if (controllers[i]->local_ip) {
821 oso.control_ip = controllers[i]->local_ip;
825 ofproto_set_sflow(br->ofproto, &oso);
827 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
829 ofproto_set_sflow(br->ofproto, NULL);
832 /* Update the controller and related settings. It would be more
833 * straightforward to call this from bridge_reconfigure_one(), but we
834 * can't do it there for two reasons. First, and most importantly, at
835 * that point we don't know the dp_ifidx of any interfaces that have
836 * been added to the bridge (because we haven't actually added them to
837 * the datapath). Second, at that point we haven't set the datapath ID
838 * yet; when a controller is configured, resetting the datapath ID will
839 * immediately disconnect from the controller, so it's better to set
840 * the datapath ID before the controller. */
841 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
843 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
844 for (i = 0; i < br->n_ports; i++) {
845 struct port *port = br->ports[i];
847 port_update_vlan_compat(port);
848 port_update_bonding(port);
851 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
852 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
855 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
857 ovsdb_idl_txn_commit(txn);
858 ovsdb_idl_txn_destroy(txn); /* XXX */
864 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
868 for (i = 0; i < n; i++) {
869 if (!strcmp(keys[i], key)) {
877 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
879 return get_ovsrec_key_value(key,
880 br_cfg->key_other_config,
881 br_cfg->value_other_config,
882 br_cfg->n_other_config);
886 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
887 struct iface **hw_addr_iface)
893 *hw_addr_iface = NULL;
895 /* Did the user request a particular MAC? */
896 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
897 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
898 if (eth_addr_is_multicast(ea)) {
899 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
900 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
901 } else if (eth_addr_is_zero(ea)) {
902 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
908 /* Otherwise choose the minimum non-local MAC address among all of the
910 memset(ea, 0xff, sizeof ea);
911 for (i = 0; i < br->n_ports; i++) {
912 struct port *port = br->ports[i];
913 uint8_t iface_ea[ETH_ADDR_LEN];
916 /* Mirror output ports don't participate. */
917 if (port->is_mirror_output_port) {
921 /* Choose the MAC address to represent the port. */
922 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
923 /* Find the interface with this Ethernet address (if any) so that
924 * we can provide the correct devname to the caller. */
926 for (j = 0; j < port->n_ifaces; j++) {
927 struct iface *candidate = port->ifaces[j];
928 uint8_t candidate_ea[ETH_ADDR_LEN];
929 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
930 && eth_addr_equals(iface_ea, candidate_ea)) {
935 /* Choose the interface whose MAC address will represent the port.
936 * The Linux kernel bonding code always chooses the MAC address of
937 * the first slave added to a bond, and the Fedora networking
938 * scripts always add slaves to a bond in alphabetical order, so
939 * for compatibility we choose the interface with the name that is
940 * first in alphabetical order. */
941 iface = port->ifaces[0];
942 for (j = 1; j < port->n_ifaces; j++) {
943 struct iface *candidate = port->ifaces[j];
944 if (strcmp(candidate->name, iface->name) < 0) {
949 /* The local port doesn't count (since we're trying to choose its
950 * MAC address anyway). */
951 if (iface->dp_ifidx == ODPP_LOCAL) {
956 error = netdev_get_etheraddr(iface->netdev, iface_ea);
958 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
959 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
960 iface->name, strerror(error));
965 /* Compare against our current choice. */
966 if (!eth_addr_is_multicast(iface_ea) &&
967 !eth_addr_is_local(iface_ea) &&
968 !eth_addr_is_reserved(iface_ea) &&
969 !eth_addr_is_zero(iface_ea) &&
970 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
972 memcpy(ea, iface_ea, ETH_ADDR_LEN);
973 *hw_addr_iface = iface;
976 if (eth_addr_is_multicast(ea)) {
977 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
978 *hw_addr_iface = NULL;
979 VLOG_WARN("bridge %s: using default bridge Ethernet "
980 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
982 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
983 br->name, ETH_ADDR_ARGS(ea));
987 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
988 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
989 * an interface on 'br', then that interface must be passed in as
990 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
991 * 'hw_addr_iface' must be passed in as a null pointer. */
993 bridge_pick_datapath_id(struct bridge *br,
994 const uint8_t bridge_ea[ETH_ADDR_LEN],
995 struct iface *hw_addr_iface)
998 * The procedure for choosing a bridge MAC address will, in the most
999 * ordinary case, also choose a unique MAC that we can use as a datapath
1000 * ID. In some special cases, though, multiple bridges will end up with
1001 * the same MAC address. This is OK for the bridges, but it will confuse
1002 * the OpenFlow controller, because each datapath needs a unique datapath
1005 * Datapath IDs must be unique. It is also very desirable that they be
1006 * stable from one run to the next, so that policy set on a datapath
1009 const char *datapath_id;
1012 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1013 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1017 if (hw_addr_iface) {
1019 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1021 * A bridge whose MAC address is taken from a VLAN network device
1022 * (that is, a network device created with vconfig(8) or similar
1023 * tool) will have the same MAC address as a bridge on the VLAN
1024 * device's physical network device.
1026 * Handle this case by hashing the physical network device MAC
1027 * along with the VLAN identifier.
1029 uint8_t buf[ETH_ADDR_LEN + 2];
1030 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1031 buf[ETH_ADDR_LEN] = vlan >> 8;
1032 buf[ETH_ADDR_LEN + 1] = vlan;
1033 return dpid_from_hash(buf, sizeof buf);
1036 * Assume that this bridge's MAC address is unique, since it
1037 * doesn't fit any of the cases we handle specially.
1042 * A purely internal bridge, that is, one that has no non-virtual
1043 * network devices on it at all, is more difficult because it has no
1044 * natural unique identifier at all.
1046 * When the host is a XenServer, we handle this case by hashing the
1047 * host's UUID with the name of the bridge. Names of bridges are
1048 * persistent across XenServer reboots, although they can be reused if
1049 * an internal network is destroyed and then a new one is later
1050 * created, so this is fairly effective.
1052 * When the host is not a XenServer, we punt by using a random MAC
1053 * address on each run.
1055 const char *host_uuid = xenserver_get_host_uuid();
1057 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1058 dpid = dpid_from_hash(combined, strlen(combined));
1064 return eth_addr_to_uint64(bridge_ea);
1068 dpid_from_hash(const void *data, size_t n)
1070 uint8_t hash[SHA1_DIGEST_SIZE];
1072 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1073 sha1_bytes(data, n, hash);
1074 eth_addr_mark_random(hash);
1075 return eth_addr_to_uint64(hash);
1081 struct bridge *br, *next;
1085 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1086 int error = bridge_run_one(br);
1088 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1089 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1090 "forcing reconfiguration", br->name);
1104 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1105 ofproto_wait(br->ofproto);
1106 if (ofproto_has_controller(br->ofproto)) {
1110 mac_learning_wait(br->ml);
1115 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1116 * configuration changes. */
1118 bridge_flush(struct bridge *br)
1120 COVERAGE_INC(bridge_flush);
1122 mac_learning_flush(br->ml);
1125 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1126 * such interface. */
1127 static struct iface *
1128 bridge_get_local_iface(struct bridge *br)
1132 for (i = 0; i < br->n_ports; i++) {
1133 struct port *port = br->ports[i];
1134 for (j = 0; j < port->n_ifaces; j++) {
1135 struct iface *iface = port->ifaces[j];
1136 if (iface->dp_ifidx == ODPP_LOCAL) {
1145 /* Bridge unixctl user interface functions. */
1147 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1148 const char *args, void *aux OVS_UNUSED)
1150 struct ds ds = DS_EMPTY_INITIALIZER;
1151 const struct bridge *br;
1152 const struct mac_entry *e;
1154 br = bridge_lookup(args);
1156 unixctl_command_reply(conn, 501, "no such bridge");
1160 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1161 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1162 if (e->port < 0 || e->port >= br->n_ports) {
1165 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1166 br->ports[e->port]->ifaces[0]->dp_ifidx,
1167 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1169 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1173 /* Bridge reconfiguration functions. */
1174 static struct bridge *
1175 bridge_create(const struct ovsrec_bridge *br_cfg)
1180 assert(!bridge_lookup(br_cfg->name));
1181 br = xzalloc(sizeof *br);
1183 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1189 dpif_flow_flush(br->dpif);
1191 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1194 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1196 dpif_delete(br->dpif);
1197 dpif_close(br->dpif);
1202 br->name = xstrdup(br_cfg->name);
1204 br->ml = mac_learning_create();
1205 br->sent_config_request = false;
1206 eth_addr_nicira_random(br->default_ea);
1208 port_array_init(&br->ifaces);
1210 shash_init(&br->port_by_name);
1211 shash_init(&br->iface_by_name);
1215 list_push_back(&all_bridges, &br->node);
1217 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1223 bridge_destroy(struct bridge *br)
1228 while (br->n_ports > 0) {
1229 port_destroy(br->ports[br->n_ports - 1]);
1231 list_remove(&br->node);
1232 error = dpif_delete(br->dpif);
1233 if (error && error != ENOENT) {
1234 VLOG_ERR("failed to delete %s: %s",
1235 dpif_name(br->dpif), strerror(error));
1237 dpif_close(br->dpif);
1238 ofproto_destroy(br->ofproto);
1239 mac_learning_destroy(br->ml);
1240 port_array_destroy(&br->ifaces);
1241 shash_destroy(&br->port_by_name);
1242 shash_destroy(&br->iface_by_name);
1249 static struct bridge *
1250 bridge_lookup(const char *name)
1254 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1255 if (!strcmp(br->name, name)) {
1263 bridge_exists(const char *name)
1265 return bridge_lookup(name) ? true : false;
1269 bridge_get_datapathid(const char *name)
1271 struct bridge *br = bridge_lookup(name);
1272 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1275 /* Handle requests for a listing of all flows known by the OpenFlow
1276 * stack, including those normally hidden. */
1278 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1279 const char *args, void *aux OVS_UNUSED)
1284 br = bridge_lookup(args);
1286 unixctl_command_reply(conn, 501, "Unknown bridge");
1291 ofproto_get_all_flows(br->ofproto, &results);
1293 unixctl_command_reply(conn, 200, ds_cstr(&results));
1294 ds_destroy(&results);
1298 bridge_run_one(struct bridge *br)
1302 error = ofproto_run1(br->ofproto);
1307 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1310 error = ofproto_run2(br->ofproto, br->flush);
1317 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1318 const struct bridge *br,
1319 struct ovsrec_controller ***controllersp)
1321 struct ovsrec_controller **controllers;
1322 size_t n_controllers;
1324 if (br->cfg->n_controller) {
1325 controllers = br->cfg->controller;
1326 n_controllers = br->cfg->n_controller;
1328 controllers = ovs_cfg->controller;
1329 n_controllers = ovs_cfg->n_controller;
1332 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1338 *controllersp = controllers;
1340 return n_controllers;
1344 bridge_update_desc(struct bridge *br OVS_UNUSED)
1347 bool changed = false;
1350 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1351 if (desc != br->mfr_desc) {
1354 br->mfr_desc = xstrdup(desc);
1356 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1361 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1362 if (desc != br->hw_desc) {
1365 br->hw_desc = xstrdup(desc);
1367 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1372 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1373 if (desc != br->sw_desc) {
1376 br->sw_desc = xstrdup(desc);
1378 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1383 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1384 if (desc != br->serial_desc) {
1385 free(br->serial_desc);
1387 br->serial_desc = xstrdup(desc);
1389 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1394 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1395 if (desc != br->dp_desc) {
1398 br->dp_desc = xstrdup(desc);
1400 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1406 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1407 br->sw_desc, br->serial_desc, br->dp_desc);
1413 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1416 struct shash old_ports, new_ports;
1417 struct svec listeners, old_listeners;
1418 struct svec snoops, old_snoops;
1419 struct shash_node *node;
1422 /* Collect old ports. */
1423 shash_init(&old_ports);
1424 for (i = 0; i < br->n_ports; i++) {
1425 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1428 /* Collect new ports. */
1429 shash_init(&new_ports);
1430 for (i = 0; i < br->cfg->n_ports; i++) {
1431 const char *name = br->cfg->ports[i]->name;
1432 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1433 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1438 /* If we have a controller, then we need a local port. Complain if the
1439 * user didn't specify one.
1441 * XXX perhaps we should synthesize a port ourselves in this case. */
1442 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1443 char local_name[IF_NAMESIZE];
1446 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1447 local_name, sizeof local_name);
1448 if (!error && !shash_find(&new_ports, local_name)) {
1449 VLOG_WARN("bridge %s: controller specified but no local port "
1450 "(port named %s) defined",
1451 br->name, local_name);
1455 /* Get rid of deleted ports.
1456 * Get rid of deleted interfaces on ports that still exist. */
1457 SHASH_FOR_EACH (node, &old_ports) {
1458 struct port *port = node->data;
1459 const struct ovsrec_port *port_cfg;
1461 port_cfg = shash_find_data(&new_ports, node->name);
1465 port_del_ifaces(port, port_cfg);
1469 /* Create new ports.
1470 * Add new interfaces to existing ports.
1471 * Reconfigure existing ports. */
1472 SHASH_FOR_EACH (node, &new_ports) {
1473 struct port *port = shash_find_data(&old_ports, node->name);
1475 port = port_create(br, node->name);
1478 port_reconfigure(port, node->data);
1479 if (!port->n_ifaces) {
1480 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1481 br->name, port->name);
1485 shash_destroy(&old_ports);
1486 shash_destroy(&new_ports);
1488 /* Delete all flows if we're switching from connected to standalone or vice
1489 * versa. (XXX Should we delete all flows if we are switching from one
1490 * controller to another?) */
1493 /* Configure OpenFlow management listeners. */
1494 svec_init(&listeners);
1495 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1497 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1498 ovs_rundir, br->name));
1499 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1500 svec_clear(&listeners);
1502 svec_sort_unique(&listeners);
1504 svec_init(&old_listeners);
1505 ofproto_get_listeners(br->ofproto, &old_listeners);
1506 svec_sort_unique(&old_listeners);
1508 if (!svec_equal(&listeners, &old_listeners)) {
1509 ofproto_set_listeners(br->ofproto, &listeners);
1511 svec_destroy(&listeners);
1512 svec_destroy(&old_listeners);
1514 /* Configure OpenFlow controller connection snooping. */
1516 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1518 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1519 ovs_rundir, br->name));
1520 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1521 svec_clear(&snoops);
1523 svec_sort_unique(&snoops);
1525 svec_init(&old_snoops);
1526 ofproto_get_snoops(br->ofproto, &old_snoops);
1527 svec_sort_unique(&old_snoops);
1529 if (!svec_equal(&snoops, &old_snoops)) {
1530 ofproto_set_snoops(br->ofproto, &snoops);
1532 svec_destroy(&snoops);
1533 svec_destroy(&old_snoops);
1535 /* Default listener. */
1536 svec_init(&listeners);
1537 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1538 ovs_rundir, br->name));
1539 svec_init(&old_listeners);
1540 ofproto_get_listeners(br->ofproto, &old_listeners);
1541 if (!svec_equal(&listeners, &old_listeners)) {
1542 ofproto_set_listeners(br->ofproto, &listeners);
1544 svec_destroy(&listeners);
1545 svec_destroy(&old_listeners);
1547 /* Default snoop. */
1549 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1550 ovs_rundir, br->name));
1551 svec_init(&old_snoops);
1552 ofproto_get_snoops(br->ofproto, &old_snoops);
1553 if (!svec_equal(&snoops, &old_snoops)) {
1554 ofproto_set_snoops(br->ofproto, &snoops);
1556 svec_destroy(&snoops);
1557 svec_destroy(&old_snoops);
1560 mirror_reconfigure(br);
1562 bridge_update_desc(br);
1566 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1568 const struct sockaddr_in *managers,
1571 struct ovsrec_controller **controllers;
1572 size_t n_controllers;
1574 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1576 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1577 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1578 ofproto_flush_flows(br->ofproto);
1581 if (!n_controllers) {
1582 union ofp_action action;
1585 /* Clear out controllers. */
1586 ofproto_set_controllers(br->ofproto, NULL, 0);
1588 /* Set up a flow that matches every packet and directs them to
1589 * OFPP_NORMAL (which goes to us). */
1590 memset(&action, 0, sizeof action);
1591 action.type = htons(OFPAT_OUTPUT);
1592 action.output.len = htons(sizeof action);
1593 action.output.port = htons(OFPP_NORMAL);
1594 memset(&flow, 0, sizeof flow);
1595 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1597 struct ofproto_controller *ocs;
1600 ocs = xmalloc(n_controllers * sizeof *ocs);
1601 for (i = 0; i < n_controllers; i++) {
1602 struct ovsrec_controller *c = controllers[i];
1603 struct ofproto_controller *oc = &ocs[i];
1605 if (strcmp(c->target, "discover")) {
1606 struct iface *local_iface;
1609 local_iface = bridge_get_local_iface(br);
1610 if (local_iface && c->local_ip
1611 && inet_aton(c->local_ip, &ip)) {
1612 struct netdev *netdev = local_iface->netdev;
1613 struct in_addr mask, gateway;
1615 if (!c->local_netmask
1616 || !inet_aton(c->local_netmask, &mask)) {
1619 if (!c->local_gateway
1620 || !inet_aton(c->local_gateway, &gateway)) {
1624 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1626 mask.s_addr = guess_netmask(ip.s_addr);
1628 if (!netdev_set_in4(netdev, ip, mask)) {
1629 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1631 br->name, IP_ARGS(&ip.s_addr),
1632 IP_ARGS(&mask.s_addr));
1635 if (gateway.s_addr) {
1636 if (!netdev_add_router(netdev, gateway)) {
1637 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1638 br->name, IP_ARGS(&gateway.s_addr));
1644 oc->target = c->target;
1645 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1646 oc->probe_interval = (c->inactivity_probe
1647 ? *c->inactivity_probe / 1000 : 5);
1648 oc->fail = (!c->fail_mode
1649 || !strcmp(c->fail_mode, "standalone")
1650 || !strcmp(c->fail_mode, "open")
1651 ? OFPROTO_FAIL_STANDALONE
1652 : OFPROTO_FAIL_SECURE);
1653 oc->band = (!c->connection_mode
1654 || !strcmp(c->connection_mode, "in-band")
1656 : OFPROTO_OUT_OF_BAND);
1657 oc->accept_re = c->discover_accept_regex;
1658 oc->update_resolv_conf = c->discover_update_resolv_conf;
1659 oc->rate_limit = (c->controller_rate_limit
1660 ? *c->controller_rate_limit : 0);
1661 oc->burst_limit = (c->controller_burst_limit
1662 ? *c->controller_burst_limit : 0);
1664 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1670 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1675 for (i = 0; i < br->n_ports; i++) {
1676 struct port *port = br->ports[i];
1677 for (j = 0; j < port->n_ifaces; j++) {
1678 struct iface *iface = port->ifaces[j];
1679 shash_add_once(ifaces, iface->name, iface);
1681 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1682 shash_add_once(ifaces, port->name, NULL);
1687 /* For robustness, in case the administrator moves around datapath ports behind
1688 * our back, we re-check all the datapath port numbers here.
1690 * This function will set the 'dp_ifidx' members of interfaces that have
1691 * disappeared to -1, so only call this function from a context where those
1692 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1693 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1694 * datapath, which doesn't support UINT16_MAX+1 ports. */
1696 bridge_fetch_dp_ifaces(struct bridge *br)
1698 struct odp_port *dpif_ports;
1699 size_t n_dpif_ports;
1702 /* Reset all interface numbers. */
1703 for (i = 0; i < br->n_ports; i++) {
1704 struct port *port = br->ports[i];
1705 for (j = 0; j < port->n_ifaces; j++) {
1706 struct iface *iface = port->ifaces[j];
1707 iface->dp_ifidx = -1;
1710 port_array_clear(&br->ifaces);
1712 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1713 for (i = 0; i < n_dpif_ports; i++) {
1714 struct odp_port *p = &dpif_ports[i];
1715 struct iface *iface = iface_lookup(br, p->devname);
1717 if (iface->dp_ifidx >= 0) {
1718 VLOG_WARN("%s reported interface %s twice",
1719 dpif_name(br->dpif), p->devname);
1720 } else if (iface_from_dp_ifidx(br, p->port)) {
1721 VLOG_WARN("%s reported interface %"PRIu16" twice",
1722 dpif_name(br->dpif), p->port);
1724 port_array_set(&br->ifaces, p->port, iface);
1725 iface->dp_ifidx = p->port;
1729 int64_t ofport = (iface->dp_ifidx >= 0
1730 ? odp_port_to_ofp_port(iface->dp_ifidx)
1732 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1739 /* Bridge packet processing functions. */
1742 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1744 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1747 static struct bond_entry *
1748 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1750 return &port->bond_hash[bond_hash(mac)];
1754 bond_choose_iface(const struct port *port)
1756 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1757 size_t i, best_down_slave = -1;
1758 long long next_delay_expiration = LLONG_MAX;
1760 for (i = 0; i < port->n_ifaces; i++) {
1761 struct iface *iface = port->ifaces[i];
1763 if (iface->enabled) {
1765 } else if (iface->delay_expires < next_delay_expiration) {
1766 best_down_slave = i;
1767 next_delay_expiration = iface->delay_expires;
1771 if (best_down_slave != -1) {
1772 struct iface *iface = port->ifaces[best_down_slave];
1774 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1775 "since no other interface is up", iface->name,
1776 iface->delay_expires - time_msec());
1777 bond_enable_slave(iface, true);
1780 return best_down_slave;
1784 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1785 uint16_t *dp_ifidx, tag_type *tags)
1787 struct iface *iface;
1789 assert(port->n_ifaces);
1790 if (port->n_ifaces == 1) {
1791 iface = port->ifaces[0];
1793 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1794 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1795 || !port->ifaces[e->iface_idx]->enabled) {
1796 /* XXX select interface properly. The current interface selection
1797 * is only good for testing the rebalancing code. */
1798 e->iface_idx = bond_choose_iface(port);
1799 if (e->iface_idx < 0) {
1800 *tags |= port->no_ifaces_tag;
1803 e->iface_tag = tag_create_random();
1804 ((struct port *) port)->bond_compat_is_stale = true;
1806 *tags |= e->iface_tag;
1807 iface = port->ifaces[e->iface_idx];
1809 *dp_ifidx = iface->dp_ifidx;
1810 *tags |= iface->tag; /* Currently only used for bonding. */
1815 bond_link_status_update(struct iface *iface, bool carrier)
1817 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1818 struct port *port = iface->port;
1820 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1821 /* Nothing to do. */
1824 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1825 iface->name, carrier ? "detected" : "dropped");
1826 if (carrier == iface->enabled) {
1827 iface->delay_expires = LLONG_MAX;
1828 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1829 iface->name, carrier ? "disabled" : "enabled");
1830 } else if (carrier && port->active_iface < 0) {
1831 bond_enable_slave(iface, true);
1832 if (port->updelay) {
1833 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1834 "other interface is up", iface->name, port->updelay);
1837 int delay = carrier ? port->updelay : port->downdelay;
1838 iface->delay_expires = time_msec() + delay;
1841 "interface %s: will be %s if it stays %s for %d ms",
1843 carrier ? "enabled" : "disabled",
1844 carrier ? "up" : "down",
1851 bond_choose_active_iface(struct port *port)
1853 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1855 port->active_iface = bond_choose_iface(port);
1856 port->active_iface_tag = tag_create_random();
1857 if (port->active_iface >= 0) {
1858 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1859 port->name, port->ifaces[port->active_iface]->name);
1861 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1867 bond_enable_slave(struct iface *iface, bool enable)
1869 struct port *port = iface->port;
1870 struct bridge *br = port->bridge;
1872 /* This acts as a recursion check. If the act of disabling a slave
1873 * causes a different slave to be enabled, the flag will allow us to
1874 * skip redundant work when we reenter this function. It must be
1875 * cleared on exit to keep things safe with multiple bonds. */
1876 static bool moving_active_iface = false;
1878 iface->delay_expires = LLONG_MAX;
1879 if (enable == iface->enabled) {
1883 iface->enabled = enable;
1884 if (!iface->enabled) {
1885 VLOG_WARN("interface %s: disabled", iface->name);
1886 ofproto_revalidate(br->ofproto, iface->tag);
1887 if (iface->port_ifidx == port->active_iface) {
1888 ofproto_revalidate(br->ofproto,
1889 port->active_iface_tag);
1891 /* Disabling a slave can lead to another slave being immediately
1892 * enabled if there will be no active slaves but one is waiting
1893 * on an updelay. In this case we do not need to run most of the
1894 * code for the newly enabled slave since there was no period
1895 * without an active slave and it is redundant with the disabling
1897 moving_active_iface = true;
1898 bond_choose_active_iface(port);
1900 bond_send_learning_packets(port);
1902 VLOG_WARN("interface %s: enabled", iface->name);
1903 if (port->active_iface < 0 && !moving_active_iface) {
1904 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1905 bond_choose_active_iface(port);
1906 bond_send_learning_packets(port);
1908 iface->tag = tag_create_random();
1911 moving_active_iface = false;
1912 port->bond_compat_is_stale = true;
1915 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1916 * bond interface. */
1918 bond_update_fake_iface_stats(struct port *port)
1920 struct netdev_stats bond_stats;
1921 struct netdev *bond_dev;
1924 memset(&bond_stats, 0, sizeof bond_stats);
1926 for (i = 0; i < port->n_ifaces; i++) {
1927 struct netdev_stats slave_stats;
1929 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1930 /* XXX: We swap the stats here because they are swapped back when
1931 * reported by the internal device. The reason for this is
1932 * internal devices normally represent packets going into the system
1933 * but when used as fake bond device they represent packets leaving
1934 * the system. We really should do this in the internal device
1935 * itself because changing it here reverses the counts from the
1936 * perspective of the switch. However, the internal device doesn't
1937 * know what type of device it represents so we have to do it here
1939 bond_stats.tx_packets += slave_stats.rx_packets;
1940 bond_stats.tx_bytes += slave_stats.rx_bytes;
1941 bond_stats.rx_packets += slave_stats.tx_packets;
1942 bond_stats.rx_bytes += slave_stats.tx_bytes;
1946 if (!netdev_open_default(port->name, &bond_dev)) {
1947 netdev_set_stats(bond_dev, &bond_stats);
1948 netdev_close(bond_dev);
1953 bond_run(struct bridge *br)
1957 for (i = 0; i < br->n_ports; i++) {
1958 struct port *port = br->ports[i];
1960 if (port->n_ifaces >= 2) {
1961 for (j = 0; j < port->n_ifaces; j++) {
1962 struct iface *iface = port->ifaces[j];
1963 if (time_msec() >= iface->delay_expires) {
1964 bond_enable_slave(iface, !iface->enabled);
1968 if (port->bond_fake_iface
1969 && time_msec() >= port->bond_next_fake_iface_update) {
1970 bond_update_fake_iface_stats(port);
1971 port->bond_next_fake_iface_update = time_msec() + 1000;
1975 if (port->bond_compat_is_stale) {
1976 port->bond_compat_is_stale = false;
1977 port_update_bond_compat(port);
1983 bond_wait(struct bridge *br)
1987 for (i = 0; i < br->n_ports; i++) {
1988 struct port *port = br->ports[i];
1989 if (port->n_ifaces < 2) {
1992 for (j = 0; j < port->n_ifaces; j++) {
1993 struct iface *iface = port->ifaces[j];
1994 if (iface->delay_expires != LLONG_MAX) {
1995 poll_timer_wait_until(iface->delay_expires);
1998 if (port->bond_fake_iface) {
1999 poll_timer_wait_until(port->bond_next_fake_iface_update);
2005 set_dst(struct dst *p, const flow_t *flow,
2006 const struct port *in_port, const struct port *out_port,
2009 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2010 : in_port->vlan >= 0 ? in_port->vlan
2011 : ntohs(flow->dl_vlan));
2012 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2016 swap_dst(struct dst *p, struct dst *q)
2018 struct dst tmp = *p;
2023 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2024 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2025 * that we push to the datapath. We could in fact fully sort the array by
2026 * vlan, but in most cases there are at most two different vlan tags so that's
2027 * possibly overkill.) */
2029 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2031 struct dst *first = dsts;
2032 struct dst *last = dsts + n_dsts;
2034 while (first != last) {
2036 * - All dsts < first have vlan == 'vlan'.
2037 * - All dsts >= last have vlan != 'vlan'.
2038 * - first < last. */
2039 while (first->vlan == vlan) {
2040 if (++first == last) {
2045 /* Same invariants, plus one additional:
2046 * - first->vlan != vlan.
2048 while (last[-1].vlan != vlan) {
2049 if (--last == first) {
2054 /* Same invariants, plus one additional:
2055 * - last[-1].vlan == vlan.*/
2056 swap_dst(first++, --last);
2061 mirror_mask_ffs(mirror_mask_t mask)
2063 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2068 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2069 const struct dst *test)
2072 for (i = 0; i < n_dsts; i++) {
2073 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2081 port_trunks_vlan(const struct port *port, uint16_t vlan)
2083 return (port->vlan < 0
2084 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2088 port_includes_vlan(const struct port *port, uint16_t vlan)
2090 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2094 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2095 const struct port *in_port, const struct port *out_port,
2096 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2098 mirror_mask_t mirrors = in_port->src_mirrors;
2099 struct dst *dst = dsts;
2102 if (out_port == FLOOD_PORT) {
2103 /* XXX use ODP_FLOOD if no vlans or bonding. */
2104 /* XXX even better, define each VLAN as a datapath port group */
2105 for (i = 0; i < br->n_ports; i++) {
2106 struct port *port = br->ports[i];
2107 if (port != in_port && port_includes_vlan(port, vlan)
2108 && !port->is_mirror_output_port
2109 && set_dst(dst, flow, in_port, port, tags)) {
2110 mirrors |= port->dst_mirrors;
2114 *nf_output_iface = NF_OUT_FLOOD;
2115 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2116 *nf_output_iface = dst->dp_ifidx;
2117 mirrors |= out_port->dst_mirrors;
2122 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2123 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2125 if (set_dst(dst, flow, in_port, m->out_port, tags)
2126 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2130 for (i = 0; i < br->n_ports; i++) {
2131 struct port *port = br->ports[i];
2132 if (port_includes_vlan(port, m->out_vlan)
2133 && set_dst(dst, flow, in_port, port, tags))
2137 if (port->vlan < 0) {
2138 dst->vlan = m->out_vlan;
2140 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2144 /* Use the vlan tag on the original flow instead of
2145 * the one passed in the vlan parameter. This ensures
2146 * that we compare the vlan from before any implicit
2147 * tagging tags place. This is necessary because
2148 * dst->vlan is the final vlan, after removing implicit
2150 flow_vlan = ntohs(flow->dl_vlan);
2151 if (flow_vlan == 0) {
2152 flow_vlan = OFP_VLAN_NONE;
2154 if (port == in_port && dst->vlan == flow_vlan) {
2155 /* Don't send out input port on same VLAN. */
2163 mirrors &= mirrors - 1;
2166 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2170 static void OVS_UNUSED
2171 print_dsts(const struct dst *dsts, size_t n)
2173 for (; n--; dsts++) {
2174 printf(">p%"PRIu16, dsts->dp_ifidx);
2175 if (dsts->vlan != OFP_VLAN_NONE) {
2176 printf("v%"PRIu16, dsts->vlan);
2182 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2183 const struct port *in_port, const struct port *out_port,
2184 tag_type *tags, struct odp_actions *actions,
2185 uint16_t *nf_output_iface)
2187 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2189 const struct dst *p;
2192 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2195 cur_vlan = ntohs(flow->dl_vlan);
2196 for (p = dsts; p < &dsts[n_dsts]; p++) {
2197 union odp_action *a;
2198 if (p->vlan != cur_vlan) {
2199 if (p->vlan == OFP_VLAN_NONE) {
2200 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2202 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2203 a->vlan_vid.vlan_vid = htons(p->vlan);
2207 a = odp_actions_add(actions, ODPAT_OUTPUT);
2208 a->output.port = p->dp_ifidx;
2212 /* Returns the effective vlan of a packet, taking into account both the
2213 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2214 * the packet is untagged and -1 indicates it has an invalid header and
2215 * should be dropped. */
2216 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2217 struct port *in_port, bool have_packet)
2219 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2220 * belongs to VLAN 0, so we should treat both cases identically. (In the
2221 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2222 * presumably to allow a priority to be specified. In the latter case, the
2223 * packet does not have any 802.1Q header.) */
2224 int vlan = ntohs(flow->dl_vlan);
2225 if (vlan == OFP_VLAN_NONE) {
2228 if (in_port->vlan >= 0) {
2230 /* XXX support double tagging? */
2232 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2233 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2234 "packet received on port %s configured with "
2235 "implicit VLAN %"PRIu16,
2236 br->name, ntohs(flow->dl_vlan),
2237 in_port->name, in_port->vlan);
2241 vlan = in_port->vlan;
2243 if (!port_includes_vlan(in_port, vlan)) {
2245 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2246 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2247 "packet received on port %s not configured for "
2249 br->name, vlan, in_port->name, vlan);
2258 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2259 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2260 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2262 is_gratuitous_arp(const flow_t *flow)
2264 return (flow->dl_type == htons(ETH_TYPE_ARP)
2265 && eth_addr_is_broadcast(flow->dl_dst)
2266 && (flow->nw_proto == ARP_OP_REPLY
2267 || (flow->nw_proto == ARP_OP_REQUEST
2268 && flow->nw_src == flow->nw_dst)));
2272 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2273 struct port *in_port)
2275 enum grat_arp_lock_type lock_type;
2278 /* We don't want to learn from gratuitous ARP packets that are reflected
2279 * back over bond slaves so we lock the learning table. */
2280 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2281 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2282 GRAT_ARP_LOCK_CHECK;
2284 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2287 /* The log messages here could actually be useful in debugging,
2288 * so keep the rate limit relatively high. */
2289 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2291 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2292 "on port %s in VLAN %d",
2293 br->name, ETH_ADDR_ARGS(flow->dl_src),
2294 in_port->name, vlan);
2295 ofproto_revalidate(br->ofproto, rev_tag);
2299 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2300 * dropped. Returns true if they may be forwarded, false if they should be
2303 * If 'have_packet' is true, it indicates that the caller is processing a
2304 * received packet. If 'have_packet' is false, then the caller is just
2305 * revalidating an existing flow because configuration has changed. Either
2306 * way, 'have_packet' only affects logging (there is no point in logging errors
2307 * during revalidation).
2309 * Sets '*in_portp' to the input port. This will be a null pointer if
2310 * flow->in_port does not designate a known input port (in which case
2311 * is_admissible() returns false).
2313 * When returning true, sets '*vlanp' to the effective VLAN of the input
2314 * packet, as returned by flow_get_vlan().
2316 * May also add tags to '*tags', although the current implementation only does
2317 * so in one special case.
2320 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2321 tag_type *tags, int *vlanp, struct port **in_portp)
2323 struct iface *in_iface;
2324 struct port *in_port;
2327 /* Find the interface and port structure for the received packet. */
2328 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2330 /* No interface? Something fishy... */
2332 /* Odd. A few possible reasons here:
2334 * - We deleted an interface but there are still a few packets
2335 * queued up from it.
2337 * - Someone externally added an interface (e.g. with "ovs-dpctl
2338 * add-if") that we don't know about.
2340 * - Packet arrived on the local port but the local port is not
2341 * one of our bridge ports.
2343 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2345 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2346 "interface %"PRIu16, br->name, flow->in_port);
2352 *in_portp = in_port = in_iface->port;
2353 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2358 /* Drop frames for reserved multicast addresses. */
2359 if (eth_addr_is_reserved(flow->dl_dst)) {
2363 /* Drop frames on ports reserved for mirroring. */
2364 if (in_port->is_mirror_output_port) {
2366 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2367 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2368 "%s, which is reserved exclusively for mirroring",
2369 br->name, in_port->name);
2374 /* Packets received on bonds need special attention to avoid duplicates. */
2375 if (in_port->n_ifaces > 1) {
2377 bool is_grat_arp_locked;
2379 if (eth_addr_is_multicast(flow->dl_dst)) {
2380 *tags |= in_port->active_iface_tag;
2381 if (in_port->active_iface != in_iface->port_ifidx) {
2382 /* Drop all multicast packets on inactive slaves. */
2387 /* Drop all packets for which we have learned a different input
2388 * port, because we probably sent the packet on one slave and got
2389 * it back on the other. Gratuitous ARP packets are an exception
2390 * to this rule: the host has moved to another switch. The exception
2391 * to the exception is if we locked the learning table to avoid
2392 * reflections on bond slaves. If this is the case, just drop the
2394 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2395 &is_grat_arp_locked);
2396 if (src_idx != -1 && src_idx != in_port->port_idx &&
2397 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2405 /* If the composed actions may be applied to any packet in the given 'flow',
2406 * returns true. Otherwise, the actions should only be applied to 'packet', or
2407 * not at all, if 'packet' was NULL. */
2409 process_flow(struct bridge *br, const flow_t *flow,
2410 const struct ofpbuf *packet, struct odp_actions *actions,
2411 tag_type *tags, uint16_t *nf_output_iface)
2413 struct port *in_port;
2414 struct port *out_port;
2418 /* Check whether we should drop packets in this flow. */
2419 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2424 /* Learn source MAC (but don't try to learn from revalidation). */
2426 update_learning_table(br, flow, vlan, in_port);
2429 /* Determine output port. */
2430 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2432 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2433 out_port = br->ports[out_port_idx];
2434 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2435 /* If we are revalidating but don't have a learning entry then
2436 * eject the flow. Installing a flow that floods packets opens
2437 * up a window of time where we could learn from a packet reflected
2438 * on a bond and blackhole packets before the learning table is
2439 * updated to reflect the correct port. */
2442 out_port = FLOOD_PORT;
2445 /* Don't send packets out their input ports. */
2446 if (in_port == out_port) {
2452 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2459 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2462 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2463 const struct ofp_phy_port *opp,
2466 struct bridge *br = br_;
2467 struct iface *iface;
2470 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2476 if (reason == OFPPR_DELETE) {
2477 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2478 br->name, iface->name);
2479 iface_destroy(iface);
2480 if (!port->n_ifaces) {
2481 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2482 br->name, port->name);
2488 if (port->n_ifaces > 1) {
2489 bool up = !(opp->state & OFPPS_LINK_DOWN);
2490 bond_link_status_update(iface, up);
2491 port_update_bond_compat(port);
2497 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2498 struct odp_actions *actions, tag_type *tags,
2499 uint16_t *nf_output_iface, void *br_)
2501 struct bridge *br = br_;
2503 COVERAGE_INC(bridge_process_flow);
2504 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2508 bridge_account_flow_ofhook_cb(const flow_t *flow,
2509 const union odp_action *actions,
2510 size_t n_actions, unsigned long long int n_bytes,
2513 struct bridge *br = br_;
2514 const union odp_action *a;
2515 struct port *in_port;
2519 /* Feed information from the active flows back into the learning table
2520 * to ensure that table is always in sync with what is actually flowing
2521 * through the datapath. */
2522 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2523 update_learning_table(br, flow, vlan, in_port);
2526 if (!br->has_bonded_ports) {
2530 for (a = actions; a < &actions[n_actions]; a++) {
2531 if (a->type == ODPAT_OUTPUT) {
2532 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2533 if (out_port && out_port->n_ifaces >= 2) {
2534 struct bond_entry *e = lookup_bond_entry(out_port,
2536 e->tx_bytes += n_bytes;
2543 bridge_account_checkpoint_ofhook_cb(void *br_)
2545 struct bridge *br = br_;
2549 if (!br->has_bonded_ports) {
2554 for (i = 0; i < br->n_ports; i++) {
2555 struct port *port = br->ports[i];
2556 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2557 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2558 bond_rebalance_port(port);
2563 static struct ofhooks bridge_ofhooks = {
2564 bridge_port_changed_ofhook_cb,
2565 bridge_normal_ofhook_cb,
2566 bridge_account_flow_ofhook_cb,
2567 bridge_account_checkpoint_ofhook_cb,
2570 /* Bonding functions. */
2572 /* Statistics for a single interface on a bonded port, used for load-based
2573 * bond rebalancing. */
2574 struct slave_balance {
2575 struct iface *iface; /* The interface. */
2576 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2578 /* All the "bond_entry"s that are assigned to this interface, in order of
2579 * increasing tx_bytes. */
2580 struct bond_entry **hashes;
2584 /* Sorts pointers to pointers to bond_entries in ascending order by the
2585 * interface to which they are assigned, and within a single interface in
2586 * ascending order of bytes transmitted. */
2588 compare_bond_entries(const void *a_, const void *b_)
2590 const struct bond_entry *const *ap = a_;
2591 const struct bond_entry *const *bp = b_;
2592 const struct bond_entry *a = *ap;
2593 const struct bond_entry *b = *bp;
2594 if (a->iface_idx != b->iface_idx) {
2595 return a->iface_idx > b->iface_idx ? 1 : -1;
2596 } else if (a->tx_bytes != b->tx_bytes) {
2597 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2603 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2604 * *descending* order by number of bytes transmitted. */
2606 compare_slave_balance(const void *a_, const void *b_)
2608 const struct slave_balance *a = a_;
2609 const struct slave_balance *b = b_;
2610 if (a->iface->enabled != b->iface->enabled) {
2611 return a->iface->enabled ? -1 : 1;
2612 } else if (a->tx_bytes != b->tx_bytes) {
2613 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2620 swap_bals(struct slave_balance *a, struct slave_balance *b)
2622 struct slave_balance tmp = *a;
2627 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2628 * given that 'p' (and only 'p') might be in the wrong location.
2630 * This function invalidates 'p', since it might now be in a different memory
2633 resort_bals(struct slave_balance *p,
2634 struct slave_balance bals[], size_t n_bals)
2637 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2638 swap_bals(p, p - 1);
2640 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2641 swap_bals(p, p + 1);
2647 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2649 if (VLOG_IS_DBG_ENABLED()) {
2650 struct ds ds = DS_EMPTY_INITIALIZER;
2651 const struct slave_balance *b;
2653 for (b = bals; b < bals + n_bals; b++) {
2657 ds_put_char(&ds, ',');
2659 ds_put_format(&ds, " %s %"PRIu64"kB",
2660 b->iface->name, b->tx_bytes / 1024);
2662 if (!b->iface->enabled) {
2663 ds_put_cstr(&ds, " (disabled)");
2665 if (b->n_hashes > 0) {
2666 ds_put_cstr(&ds, " (");
2667 for (i = 0; i < b->n_hashes; i++) {
2668 const struct bond_entry *e = b->hashes[i];
2670 ds_put_cstr(&ds, " + ");
2672 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2673 e - port->bond_hash, e->tx_bytes / 1024);
2675 ds_put_cstr(&ds, ")");
2678 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2683 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2685 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2688 struct bond_entry *hash = from->hashes[hash_idx];
2689 struct port *port = from->iface->port;
2690 uint64_t delta = hash->tx_bytes;
2692 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2693 "from %s to %s (now carrying %"PRIu64"kB and "
2694 "%"PRIu64"kB load, respectively)",
2695 port->name, delta / 1024, hash - port->bond_hash,
2696 from->iface->name, to->iface->name,
2697 (from->tx_bytes - delta) / 1024,
2698 (to->tx_bytes + delta) / 1024);
2700 /* Delete element from from->hashes.
2702 * We don't bother to add the element to to->hashes because not only would
2703 * it require more work, the only purpose it would be to allow that hash to
2704 * be migrated to another slave in this rebalancing run, and there is no
2705 * point in doing that. */
2706 if (hash_idx == 0) {
2709 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2710 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2714 /* Shift load away from 'from' to 'to'. */
2715 from->tx_bytes -= delta;
2716 to->tx_bytes += delta;
2718 /* Arrange for flows to be revalidated. */
2719 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2720 hash->iface_idx = to->iface->port_ifidx;
2721 hash->iface_tag = tag_create_random();
2725 bond_rebalance_port(struct port *port)
2727 struct slave_balance bals[DP_MAX_PORTS];
2729 struct bond_entry *hashes[BOND_MASK + 1];
2730 struct slave_balance *b, *from, *to;
2731 struct bond_entry *e;
2734 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2735 * descending order of tx_bytes, so that bals[0] represents the most
2736 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2739 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2740 * array for each slave_balance structure, we sort our local array of
2741 * hashes in order by slave, so that all of the hashes for a given slave
2742 * become contiguous in memory, and then we point each 'hashes' members of
2743 * a slave_balance structure to the start of a contiguous group. */
2744 n_bals = port->n_ifaces;
2745 for (b = bals; b < &bals[n_bals]; b++) {
2746 b->iface = port->ifaces[b - bals];
2751 for (i = 0; i <= BOND_MASK; i++) {
2752 hashes[i] = &port->bond_hash[i];
2754 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2755 for (i = 0; i <= BOND_MASK; i++) {
2757 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2758 b = &bals[e->iface_idx];
2759 b->tx_bytes += e->tx_bytes;
2761 b->hashes = &hashes[i];
2766 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2767 log_bals(bals, n_bals, port);
2769 /* Discard slaves that aren't enabled (which were sorted to the back of the
2770 * array earlier). */
2771 while (!bals[n_bals - 1].iface->enabled) {
2778 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2779 to = &bals[n_bals - 1];
2780 for (from = bals; from < to; ) {
2781 uint64_t overload = from->tx_bytes - to->tx_bytes;
2782 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2783 /* The extra load on 'from' (and all less-loaded slaves), compared
2784 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2785 * it is less than ~1Mbps. No point in rebalancing. */
2787 } else if (from->n_hashes == 1) {
2788 /* 'from' only carries a single MAC hash, so we can't shift any
2789 * load away from it, even though we want to. */
2792 /* 'from' is carrying significantly more load than 'to', and that
2793 * load is split across at least two different hashes. Pick a hash
2794 * to migrate to 'to' (the least-loaded slave), given that doing so
2795 * must decrease the ratio of the load on the two slaves by at
2798 * The sort order we use means that we prefer to shift away the
2799 * smallest hashes instead of the biggest ones. There is little
2800 * reason behind this decision; we could use the opposite sort
2801 * order to shift away big hashes ahead of small ones. */
2805 for (i = 0; i < from->n_hashes; i++) {
2806 double old_ratio, new_ratio;
2807 uint64_t delta = from->hashes[i]->tx_bytes;
2809 if (delta == 0 || from->tx_bytes - delta == 0) {
2810 /* Pointless move. */
2814 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2816 if (to->tx_bytes == 0) {
2817 /* Nothing on the new slave, move it. */
2821 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2822 new_ratio = (double)(from->tx_bytes - delta) /
2823 (to->tx_bytes + delta);
2825 if (new_ratio == 0) {
2826 /* Should already be covered but check to prevent division
2831 if (new_ratio < 1) {
2832 new_ratio = 1 / new_ratio;
2835 if (old_ratio - new_ratio > 0.1) {
2836 /* Would decrease the ratio, move it. */
2840 if (i < from->n_hashes) {
2841 bond_shift_load(from, to, i);
2842 port->bond_compat_is_stale = true;
2844 /* If the result of the migration changed the relative order of
2845 * 'from' and 'to' swap them back to maintain invariants. */
2846 if (order_swapped) {
2847 swap_bals(from, to);
2850 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2851 * point to different slave_balance structures. It is only
2852 * valid to do these two operations in a row at all because we
2853 * know that 'from' will not move past 'to' and vice versa. */
2854 resort_bals(from, bals, n_bals);
2855 resort_bals(to, bals, n_bals);
2862 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2863 * historical data to decay to <1% in 7 rebalancing runs. */
2864 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2870 bond_send_learning_packets(struct port *port)
2872 struct bridge *br = port->bridge;
2873 struct mac_entry *e;
2874 struct ofpbuf packet;
2875 int error, n_packets, n_errors;
2877 if (!port->n_ifaces || port->active_iface < 0) {
2881 ofpbuf_init(&packet, 128);
2882 error = n_packets = n_errors = 0;
2883 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2884 union ofp_action actions[2], *a;
2890 if (e->port == port->port_idx
2891 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2895 /* Compose actions. */
2896 memset(actions, 0, sizeof actions);
2899 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2900 a->vlan_vid.len = htons(sizeof *a);
2901 a->vlan_vid.vlan_vid = htons(e->vlan);
2904 a->output.type = htons(OFPAT_OUTPUT);
2905 a->output.len = htons(sizeof *a);
2906 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2911 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2913 flow_extract(&packet, 0, ODPP_NONE, &flow);
2914 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2921 ofpbuf_uninit(&packet);
2924 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2925 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2926 "packets, last error was: %s",
2927 port->name, n_errors, n_packets, strerror(error));
2929 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2930 port->name, n_packets);
2934 /* Bonding unixctl user interface functions. */
2937 bond_unixctl_list(struct unixctl_conn *conn,
2938 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2940 struct ds ds = DS_EMPTY_INITIALIZER;
2941 const struct bridge *br;
2943 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2945 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2948 for (i = 0; i < br->n_ports; i++) {
2949 const struct port *port = br->ports[i];
2950 if (port->n_ifaces > 1) {
2953 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2954 for (j = 0; j < port->n_ifaces; j++) {
2955 const struct iface *iface = port->ifaces[j];
2957 ds_put_cstr(&ds, ", ");
2959 ds_put_cstr(&ds, iface->name);
2961 ds_put_char(&ds, '\n');
2965 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2969 static struct port *
2970 bond_find(const char *name)
2972 const struct bridge *br;
2974 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2977 for (i = 0; i < br->n_ports; i++) {
2978 struct port *port = br->ports[i];
2979 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2988 bond_unixctl_show(struct unixctl_conn *conn,
2989 const char *args, void *aux OVS_UNUSED)
2991 struct ds ds = DS_EMPTY_INITIALIZER;
2992 const struct port *port;
2995 port = bond_find(args);
2997 unixctl_command_reply(conn, 501, "no such bond");
3001 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3002 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3003 ds_put_format(&ds, "next rebalance: %lld ms\n",
3004 port->bond_next_rebalance - time_msec());
3005 for (j = 0; j < port->n_ifaces; j++) {
3006 const struct iface *iface = port->ifaces[j];
3007 struct bond_entry *be;
3010 ds_put_format(&ds, "slave %s: %s\n",
3011 iface->name, iface->enabled ? "enabled" : "disabled");
3012 if (j == port->active_iface) {
3013 ds_put_cstr(&ds, "\tactive slave\n");
3015 if (iface->delay_expires != LLONG_MAX) {
3016 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3017 iface->enabled ? "downdelay" : "updelay",
3018 iface->delay_expires - time_msec());
3022 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3023 int hash = be - port->bond_hash;
3024 struct mac_entry *me;
3026 if (be->iface_idx != j) {
3030 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3031 hash, be->tx_bytes / 1024);
3034 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3035 &port->bridge->ml->lrus) {
3038 if (bond_hash(me->mac) == hash
3039 && me->port != port->port_idx
3040 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3041 && dp_ifidx == iface->dp_ifidx)
3043 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3044 ETH_ADDR_ARGS(me->mac));
3049 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3054 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3055 void *aux OVS_UNUSED)
3057 char *args = (char *) args_;
3058 char *save_ptr = NULL;
3059 char *bond_s, *hash_s, *slave_s;
3060 uint8_t mac[ETH_ADDR_LEN];
3062 struct iface *iface;
3063 struct bond_entry *entry;
3066 bond_s = strtok_r(args, " ", &save_ptr);
3067 hash_s = strtok_r(NULL, " ", &save_ptr);
3068 slave_s = strtok_r(NULL, " ", &save_ptr);
3070 unixctl_command_reply(conn, 501,
3071 "usage: bond/migrate BOND HASH SLAVE");
3075 port = bond_find(bond_s);
3077 unixctl_command_reply(conn, 501, "no such bond");
3081 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3082 == ETH_ADDR_SCAN_COUNT) {
3083 hash = bond_hash(mac);
3084 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3085 hash = atoi(hash_s) & BOND_MASK;
3087 unixctl_command_reply(conn, 501, "bad hash");
3091 iface = port_lookup_iface(port, slave_s);
3093 unixctl_command_reply(conn, 501, "no such slave");
3097 if (!iface->enabled) {
3098 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3102 entry = &port->bond_hash[hash];
3103 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3104 entry->iface_idx = iface->port_ifidx;
3105 entry->iface_tag = tag_create_random();
3106 port->bond_compat_is_stale = true;
3107 unixctl_command_reply(conn, 200, "migrated");
3111 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3112 void *aux OVS_UNUSED)
3114 char *args = (char *) args_;
3115 char *save_ptr = NULL;
3116 char *bond_s, *slave_s;
3118 struct iface *iface;
3120 bond_s = strtok_r(args, " ", &save_ptr);
3121 slave_s = strtok_r(NULL, " ", &save_ptr);
3123 unixctl_command_reply(conn, 501,
3124 "usage: bond/set-active-slave BOND SLAVE");
3128 port = bond_find(bond_s);
3130 unixctl_command_reply(conn, 501, "no such bond");
3134 iface = port_lookup_iface(port, slave_s);
3136 unixctl_command_reply(conn, 501, "no such slave");
3140 if (!iface->enabled) {
3141 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3145 if (port->active_iface != iface->port_ifidx) {
3146 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3147 port->active_iface = iface->port_ifidx;
3148 port->active_iface_tag = tag_create_random();
3149 VLOG_INFO("port %s: active interface is now %s",
3150 port->name, iface->name);
3151 bond_send_learning_packets(port);
3152 unixctl_command_reply(conn, 200, "done");
3154 unixctl_command_reply(conn, 200, "no change");
3159 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3161 char *args = (char *) args_;
3162 char *save_ptr = NULL;
3163 char *bond_s, *slave_s;
3165 struct iface *iface;
3167 bond_s = strtok_r(args, " ", &save_ptr);
3168 slave_s = strtok_r(NULL, " ", &save_ptr);
3170 unixctl_command_reply(conn, 501,
3171 "usage: bond/enable/disable-slave BOND SLAVE");
3175 port = bond_find(bond_s);
3177 unixctl_command_reply(conn, 501, "no such bond");
3181 iface = port_lookup_iface(port, slave_s);
3183 unixctl_command_reply(conn, 501, "no such slave");
3187 bond_enable_slave(iface, enable);
3188 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3192 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3193 void *aux OVS_UNUSED)
3195 enable_slave(conn, args, true);
3199 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3200 void *aux OVS_UNUSED)
3202 enable_slave(conn, args, false);
3206 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3207 void *aux OVS_UNUSED)
3209 uint8_t mac[ETH_ADDR_LEN];
3213 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3214 == ETH_ADDR_SCAN_COUNT) {
3215 hash = bond_hash(mac);
3217 hash_cstr = xasprintf("%u", hash);
3218 unixctl_command_reply(conn, 200, hash_cstr);
3221 unixctl_command_reply(conn, 501, "invalid mac");
3228 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3229 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3230 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3231 unixctl_command_register("bond/set-active-slave",
3232 bond_unixctl_set_active_slave, NULL);
3233 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3235 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3237 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3240 /* Port functions. */
3242 static struct port *
3243 port_create(struct bridge *br, const char *name)
3247 port = xzalloc(sizeof *port);
3249 port->port_idx = br->n_ports;
3251 port->trunks = NULL;
3252 port->name = xstrdup(name);
3253 port->active_iface = -1;
3255 if (br->n_ports >= br->allocated_ports) {
3256 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3259 br->ports[br->n_ports++] = port;
3260 shash_add_assert(&br->port_by_name, port->name, port);
3262 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3269 get_port_other_config(const struct ovsrec_port *port, const char *key,
3270 const char *default_value)
3272 const char *value = get_ovsrec_key_value(key,
3273 port->key_other_config,
3274 port->value_other_config,
3275 port->n_other_config);
3276 return value ? value : default_value;
3280 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3282 struct shash new_ifaces;
3285 /* Collect list of new interfaces. */
3286 shash_init(&new_ifaces);
3287 for (i = 0; i < cfg->n_interfaces; i++) {
3288 const char *name = cfg->interfaces[i]->name;
3289 shash_add_once(&new_ifaces, name, NULL);
3292 /* Get rid of deleted interfaces. */
3293 for (i = 0; i < port->n_ifaces; ) {
3294 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3295 iface_destroy(port->ifaces[i]);
3301 shash_destroy(&new_ifaces);
3305 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3307 struct shash new_ifaces;
3308 long long int next_rebalance;
3309 unsigned long *trunks;
3315 /* Update settings. */
3316 port->updelay = cfg->bond_updelay;
3317 if (port->updelay < 0) {
3320 port->updelay = cfg->bond_downdelay;
3321 if (port->downdelay < 0) {
3322 port->downdelay = 0;
3324 port->bond_rebalance_interval = atoi(
3325 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3326 if (port->bond_rebalance_interval < 1000) {
3327 port->bond_rebalance_interval = 1000;
3329 next_rebalance = time_msec() + port->bond_rebalance_interval;
3330 if (port->bond_next_rebalance > next_rebalance) {
3331 port->bond_next_rebalance = next_rebalance;
3334 /* Add new interfaces and update 'cfg' member of existing ones. */
3335 shash_init(&new_ifaces);
3336 for (i = 0; i < cfg->n_interfaces; i++) {
3337 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3338 struct iface *iface;
3340 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3341 VLOG_WARN("port %s: %s specified twice as port interface",
3342 port->name, if_cfg->name);
3346 iface = iface_lookup(port->bridge, if_cfg->name);
3348 if (iface->port != port) {
3349 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3351 port->bridge->name, if_cfg->name, iface->port->name);
3354 iface->cfg = if_cfg;
3356 iface_create(port, if_cfg);
3359 shash_destroy(&new_ifaces);
3364 if (port->n_ifaces < 2) {
3366 if (vlan >= 0 && vlan <= 4095) {
3367 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3372 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3373 * they even work as-is. But they have not been tested. */
3374 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3378 if (port->vlan != vlan) {
3380 bridge_flush(port->bridge);
3383 /* Get trunked VLANs. */
3385 if (vlan < 0 && cfg->n_trunks) {
3389 trunks = bitmap_allocate(4096);
3391 for (i = 0; i < cfg->n_trunks; i++) {
3392 int trunk = cfg->trunks[i];
3394 bitmap_set1(trunks, trunk);
3400 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3401 port->name, cfg->n_trunks);
3403 if (n_errors == cfg->n_trunks) {
3404 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3406 bitmap_free(trunks);
3409 } else if (vlan >= 0 && cfg->n_trunks) {
3410 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3414 ? port->trunks != NULL
3415 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3416 bridge_flush(port->bridge);
3418 bitmap_free(port->trunks);
3419 port->trunks = trunks;
3423 port_destroy(struct port *port)
3426 struct bridge *br = port->bridge;
3430 proc_net_compat_update_vlan(port->name, NULL, 0);
3431 proc_net_compat_update_bond(port->name, NULL);
3433 for (i = 0; i < MAX_MIRRORS; i++) {
3434 struct mirror *m = br->mirrors[i];
3435 if (m && m->out_port == port) {
3440 while (port->n_ifaces > 0) {
3441 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3444 shash_find_and_delete_assert(&br->port_by_name, port->name);
3446 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3447 del->port_idx = port->port_idx;
3450 bitmap_free(port->trunks);
3457 static struct port *
3458 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3460 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3461 return iface ? iface->port : NULL;
3464 static struct port *
3465 port_lookup(const struct bridge *br, const char *name)
3467 return shash_find_data(&br->port_by_name, name);
3470 static struct iface *
3471 port_lookup_iface(const struct port *port, const char *name)
3473 struct iface *iface = iface_lookup(port->bridge, name);
3474 return iface && iface->port == port ? iface : NULL;
3478 port_update_bonding(struct port *port)
3480 if (port->n_ifaces < 2) {
3481 /* Not a bonded port. */
3482 if (port->bond_hash) {
3483 free(port->bond_hash);
3484 port->bond_hash = NULL;
3485 port->bond_compat_is_stale = true;
3486 port->bond_fake_iface = false;
3489 if (!port->bond_hash) {
3492 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3493 for (i = 0; i <= BOND_MASK; i++) {
3494 struct bond_entry *e = &port->bond_hash[i];
3498 port->no_ifaces_tag = tag_create_random();
3499 bond_choose_active_iface(port);
3500 port->bond_next_rebalance
3501 = time_msec() + port->bond_rebalance_interval;
3503 if (port->cfg->bond_fake_iface) {
3504 port->bond_next_fake_iface_update = time_msec();
3507 port->bond_compat_is_stale = true;
3508 port->bond_fake_iface = port->cfg->bond_fake_iface;
3513 port_update_bond_compat(struct port *port)
3515 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3516 struct compat_bond bond;
3519 if (port->n_ifaces < 2) {
3520 proc_net_compat_update_bond(port->name, NULL);
3525 bond.updelay = port->updelay;
3526 bond.downdelay = port->downdelay;
3529 bond.hashes = compat_hashes;
3530 if (port->bond_hash) {
3531 const struct bond_entry *e;
3532 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3533 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3534 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3535 cbh->hash = e - port->bond_hash;
3536 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3541 bond.n_slaves = port->n_ifaces;
3542 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3543 for (i = 0; i < port->n_ifaces; i++) {
3544 struct iface *iface = port->ifaces[i];
3545 struct compat_bond_slave *slave = &bond.slaves[i];
3546 slave->name = iface->name;
3548 /* We need to make the same determination as the Linux bonding
3549 * code to determine whether a slave should be consider "up".
3550 * The Linux function bond_miimon_inspect() supports four
3551 * BOND_LINK_* states:
3553 * - BOND_LINK_UP: carrier detected, updelay has passed.
3554 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3555 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3556 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3558 * The function bond_info_show_slave() only considers BOND_LINK_UP
3559 * to be "up" and anything else to be "down".
3561 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3565 netdev_get_etheraddr(iface->netdev, slave->mac);
3568 if (port->bond_fake_iface) {
3569 struct netdev *bond_netdev;
3571 if (!netdev_open_default(port->name, &bond_netdev)) {
3573 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3575 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3577 netdev_close(bond_netdev);
3581 proc_net_compat_update_bond(port->name, &bond);
3586 port_update_vlan_compat(struct port *port)
3588 struct bridge *br = port->bridge;
3589 char *vlandev_name = NULL;
3591 if (port->vlan > 0) {
3592 /* Figure out the name that the VLAN device should actually have, if it
3593 * existed. This takes some work because the VLAN device would not
3594 * have port->name in its name; rather, it would have the trunk port's
3595 * name, and 'port' would be attached to a bridge that also had the
3596 * VLAN device one of its ports. So we need to find a trunk port that
3597 * includes port->vlan.
3599 * There might be more than one candidate. This doesn't happen on
3600 * XenServer, so if it happens we just pick the first choice in
3601 * alphabetical order instead of creating multiple VLAN devices. */
3603 for (i = 0; i < br->n_ports; i++) {
3604 struct port *p = br->ports[i];
3605 if (port_trunks_vlan(p, port->vlan)
3607 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3609 uint8_t ea[ETH_ADDR_LEN];
3610 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3611 if (!eth_addr_is_multicast(ea) &&
3612 !eth_addr_is_reserved(ea) &&
3613 !eth_addr_is_zero(ea)) {
3614 vlandev_name = p->name;
3619 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3622 /* Interface functions. */
3624 static struct iface *
3625 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3627 struct bridge *br = port->bridge;
3628 struct iface *iface;
3629 char *name = if_cfg->name;
3632 iface = xzalloc(sizeof *iface);
3634 iface->port_ifidx = port->n_ifaces;
3635 iface->name = xstrdup(name);
3636 iface->dp_ifidx = -1;
3637 iface->tag = tag_create_random();
3638 iface->delay_expires = LLONG_MAX;
3639 iface->netdev = NULL;
3640 iface->cfg = if_cfg;
3642 shash_add_assert(&br->iface_by_name, iface->name, iface);
3644 /* Attempt to create the network interface in case it doesn't exist yet. */
3645 if (!iface_is_internal(br, iface->name)) {
3646 error = set_up_iface(if_cfg, iface, true);
3648 VLOG_WARN("could not create iface %s: %s", iface->name,
3651 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3658 if (port->n_ifaces >= port->allocated_ifaces) {
3659 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3660 sizeof *port->ifaces);
3662 port->ifaces[port->n_ifaces++] = iface;
3663 if (port->n_ifaces > 1) {
3664 br->has_bonded_ports = true;
3667 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3675 iface_destroy(struct iface *iface)
3678 struct port *port = iface->port;
3679 struct bridge *br = port->bridge;
3680 bool del_active = port->active_iface == iface->port_ifidx;
3683 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3685 if (iface->dp_ifidx >= 0) {
3686 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3689 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3690 del->port_ifidx = iface->port_ifidx;
3692 netdev_close(iface->netdev);
3695 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3696 bond_choose_active_iface(port);
3697 bond_send_learning_packets(port);
3703 bridge_flush(port->bridge);
3707 static struct iface *
3708 iface_lookup(const struct bridge *br, const char *name)
3710 return shash_find_data(&br->iface_by_name, name);
3713 static struct iface *
3714 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3716 return port_array_get(&br->ifaces, dp_ifidx);
3719 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3720 * 'br', that is, an interface that is entirely simulated within the datapath.
3721 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3722 * interfaces are created by setting "iface.<iface>.internal = true".
3724 * In addition, we have a kluge-y feature that creates an internal port with
3725 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3726 * This feature needs to go away in the long term. Until then, this is one
3727 * reason why this function takes a name instead of a struct iface: the fake
3728 * interfaces created this way do not have a struct iface. */
3730 iface_is_internal(const struct bridge *br, const char *if_name)
3732 struct iface *iface;
3735 if (!strcmp(if_name, br->name)) {
3739 iface = iface_lookup(br, if_name);
3740 if (iface && !strcmp(iface->cfg->type, "internal")) {
3744 port = port_lookup(br, if_name);
3745 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3751 /* Set Ethernet address of 'iface', if one is specified in the configuration
3754 iface_set_mac(struct iface *iface)
3756 uint8_t ea[ETH_ADDR_LEN];
3758 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3759 if (eth_addr_is_multicast(ea)) {
3760 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3762 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3763 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3764 iface->name, iface->name);
3766 int error = netdev_set_etheraddr(iface->netdev, ea);
3768 VLOG_ERR("interface %s: setting MAC failed (%s)",
3769 iface->name, strerror(error));
3775 /* Port mirroring. */
3778 mirror_reconfigure(struct bridge *br)
3780 struct shash old_mirrors, new_mirrors;
3781 struct shash_node *node;
3782 unsigned long *rspan_vlans;
3785 /* Collect old mirrors. */
3786 shash_init(&old_mirrors);
3787 for (i = 0; i < MAX_MIRRORS; i++) {
3788 if (br->mirrors[i]) {
3789 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3793 /* Collect new mirrors. */
3794 shash_init(&new_mirrors);
3795 for (i = 0; i < br->cfg->n_mirrors; i++) {
3796 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3797 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3798 VLOG_WARN("bridge %s: %s specified twice as mirror",
3799 br->name, cfg->name);
3803 /* Get rid of deleted mirrors and add new mirrors. */
3804 SHASH_FOR_EACH (node, &old_mirrors) {
3805 if (!shash_find(&new_mirrors, node->name)) {
3806 mirror_destroy(node->data);
3809 SHASH_FOR_EACH (node, &new_mirrors) {
3810 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3812 mirror = mirror_create(br, node->name);
3817 mirror_reconfigure_one(mirror, node->data);
3819 shash_destroy(&old_mirrors);
3820 shash_destroy(&new_mirrors);
3822 /* Update port reserved status. */
3823 for (i = 0; i < br->n_ports; i++) {
3824 br->ports[i]->is_mirror_output_port = false;
3826 for (i = 0; i < MAX_MIRRORS; i++) {
3827 struct mirror *m = br->mirrors[i];
3828 if (m && m->out_port) {
3829 m->out_port->is_mirror_output_port = true;
3833 /* Update flooded vlans (for RSPAN). */
3835 if (br->cfg->n_flood_vlans) {
3836 rspan_vlans = bitmap_allocate(4096);
3838 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3839 int64_t vlan = br->cfg->flood_vlans[i];
3840 if (vlan >= 0 && vlan < 4096) {
3841 bitmap_set1(rspan_vlans, vlan);
3842 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3845 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3850 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3855 static struct mirror *
3856 mirror_create(struct bridge *br, const char *name)
3861 for (i = 0; ; i++) {
3862 if (i >= MAX_MIRRORS) {
3863 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3864 "cannot create %s", br->name, MAX_MIRRORS, name);
3867 if (!br->mirrors[i]) {
3872 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3875 br->mirrors[i] = m = xzalloc(sizeof *m);
3878 m->name = xstrdup(name);
3879 shash_init(&m->src_ports);
3880 shash_init(&m->dst_ports);
3890 mirror_destroy(struct mirror *m)
3893 struct bridge *br = m->bridge;
3896 for (i = 0; i < br->n_ports; i++) {
3897 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3898 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3901 shash_destroy(&m->src_ports);
3902 shash_destroy(&m->dst_ports);
3905 m->bridge->mirrors[m->idx] = NULL;
3913 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3914 struct shash *names)
3918 for (i = 0; i < n_ports; i++) {
3919 const char *name = ports[i]->name;
3920 if (port_lookup(m->bridge, name)) {
3921 shash_add_once(names, name, NULL);
3923 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3924 "port %s", m->bridge->name, m->name, name);
3930 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3936 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3938 for (i = 0; i < cfg->n_select_vlan; i++) {
3939 int64_t vlan = cfg->select_vlan[i];
3940 if (vlan < 0 || vlan > 4095) {
3941 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3942 m->bridge->name, m->name, vlan);
3944 (*vlans)[n_vlans++] = vlan;
3951 vlan_is_mirrored(const struct mirror *m, int vlan)
3955 for (i = 0; i < m->n_vlans; i++) {
3956 if (m->vlans[i] == vlan) {
3964 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3968 for (i = 0; i < m->n_vlans; i++) {
3969 if (port_trunks_vlan(p, m->vlans[i])) {
3977 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3979 struct shash src_ports, dst_ports;
3980 mirror_mask_t mirror_bit;
3981 struct port *out_port;
3987 /* Get output port. */
3988 if (cfg->output_port) {
3989 out_port = port_lookup(m->bridge, cfg->output_port->name);
3991 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3992 m->bridge->name, m->name);
3998 if (cfg->output_vlan) {
3999 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4000 "output vlan; ignoring output vlan",
4001 m->bridge->name, m->name);
4003 } else if (cfg->output_vlan) {
4005 out_vlan = *cfg->output_vlan;
4007 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4008 m->bridge->name, m->name);
4013 shash_init(&src_ports);
4014 shash_init(&dst_ports);
4015 if (cfg->select_all) {
4016 for (i = 0; i < m->bridge->n_ports; i++) {
4017 const char *name = m->bridge->ports[i]->name;
4018 shash_add_once(&src_ports, name, NULL);
4019 shash_add_once(&dst_ports, name, NULL);
4024 /* Get ports, and drop duplicates and ports that don't exist. */
4025 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4027 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4030 /* Get all the vlans, and drop duplicate and invalid vlans. */
4031 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4034 /* Update mirror data. */
4035 if (!shash_equal_keys(&m->src_ports, &src_ports)
4036 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4037 || m->n_vlans != n_vlans
4038 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4039 || m->out_port != out_port
4040 || m->out_vlan != out_vlan) {
4041 bridge_flush(m->bridge);
4043 shash_swap(&m->src_ports, &src_ports);
4044 shash_swap(&m->dst_ports, &dst_ports);
4047 m->n_vlans = n_vlans;
4048 m->out_port = out_port;
4049 m->out_vlan = out_vlan;
4052 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4053 for (i = 0; i < m->bridge->n_ports; i++) {
4054 struct port *port = m->bridge->ports[i];
4056 if (shash_find(&m->src_ports, port->name)
4059 ? port_trunks_any_mirrored_vlan(m, port)
4060 : vlan_is_mirrored(m, port->vlan)))) {
4061 port->src_mirrors |= mirror_bit;
4063 port->src_mirrors &= ~mirror_bit;
4066 if (shash_find(&m->dst_ports, port->name)) {
4067 port->dst_mirrors |= mirror_bit;
4069 port->dst_mirrors &= ~mirror_bit;
4074 shash_destroy(&src_ports);
4075 shash_destroy(&dst_ports);