1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
36 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
46 #include "ofproto/netflow.h"
47 #include "ofproto/ofproto.h"
49 #include "poll-loop.h"
50 #include "port-array.h"
51 #include "proc-net-compat.h"
55 #include "socket-util.h"
56 #include "stream-ssl.h"
62 #include "vswitchd/vswitch-idl.h"
63 #include "xenserver.h"
65 #include "sflow_api.h"
67 #define THIS_MODULE VLM_bridge
76 /* These members are always valid. */
77 struct port *port; /* Containing port. */
78 size_t port_ifidx; /* Index within containing port. */
79 char *name; /* Host network device name. */
80 tag_type tag; /* Tag associated with this interface. */
81 long long delay_expires; /* Time after which 'enabled' may change. */
83 /* These members are valid only after bridge_reconfigure() causes them to
85 int dp_ifidx; /* Index within kernel datapath. */
86 struct netdev *netdev; /* Network device. */
87 bool enabled; /* May be chosen for flows? */
89 /* This member is only valid *during* bridge_reconfigure(). */
90 const struct ovsrec_interface *cfg;
93 #define BOND_MASK 0xff
95 int iface_idx; /* Index of assigned iface, or -1 if none. */
96 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
97 tag_type iface_tag; /* Tag associated with iface_idx. */
100 #define MAX_MIRRORS 32
101 typedef uint32_t mirror_mask_t;
102 #define MIRROR_MASK_C(X) UINT32_C(X)
103 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
105 struct bridge *bridge;
109 /* Selection criteria. */
110 struct shash src_ports; /* Name is port name; data is always NULL. */
111 struct shash dst_ports; /* Name is port name; data is always NULL. */
116 struct port *out_port;
120 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
122 struct bridge *bridge;
124 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
125 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
128 /* An ordinary bridge port has 1 interface.
129 * A bridge port for bonding has at least 2 interfaces. */
130 struct iface **ifaces;
131 size_t n_ifaces, allocated_ifaces;
134 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
135 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
136 tag_type active_iface_tag; /* Tag for bcast flows. */
137 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
138 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
139 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
140 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
141 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
142 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
143 long long int bond_next_rebalance; /* Next rebalancing time. */
145 /* Port mirroring info. */
146 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
147 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
148 bool is_mirror_output_port; /* Does port mirroring send frames here? */
150 /* This member is only valid *during* bridge_reconfigure(). */
151 const struct ovsrec_port *cfg;
154 #define DP_MAX_PORTS 255
156 struct list node; /* Node in global list of bridges. */
157 char *name; /* User-specified arbitrary name. */
158 struct mac_learning *ml; /* MAC learning table. */
159 bool sent_config_request; /* Successfully sent config request? */
160 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
162 /* OpenFlow switch processing. */
163 struct ofproto *ofproto; /* OpenFlow switch. */
165 /* Description strings. */
166 char *mfr_desc; /* Manufacturer. */
167 char *hw_desc; /* Hardware. */
168 char *sw_desc; /* Software version. */
169 char *serial_desc; /* Serial number. */
170 char *dp_desc; /* Datapath description. */
172 /* Kernel datapath information. */
173 struct dpif *dpif; /* Datapath. */
174 struct port_array ifaces; /* Indexed by kernel datapath port number. */
178 size_t n_ports, allocated_ports;
179 struct shash iface_by_name; /* "struct iface"s indexed by name. */
180 struct shash port_by_name; /* "struct port"s indexed by name. */
183 bool has_bonded_ports;
188 /* Flow statistics gathering. */
189 time_t next_stats_request;
191 /* Port mirroring. */
192 struct mirror *mirrors[MAX_MIRRORS];
194 /* This member is only valid *during* bridge_reconfigure(). */
195 const struct ovsrec_bridge *cfg;
198 /* List of all bridges. */
199 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
201 /* Maximum number of datapaths. */
202 enum { DP_MAX = 256 };
204 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
205 static void bridge_destroy(struct bridge *);
206 static struct bridge *bridge_lookup(const char *name);
207 static unixctl_cb_func bridge_unixctl_dump_flows;
208 static int bridge_run_one(struct bridge *);
209 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
210 const struct bridge *br,
211 struct ovsrec_controller ***controllersp);
212 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
214 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
216 const struct sockaddr_in *managers,
218 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
219 static void bridge_fetch_dp_ifaces(struct bridge *);
220 static void bridge_flush(struct bridge *);
221 static void bridge_pick_local_hw_addr(struct bridge *,
222 uint8_t ea[ETH_ADDR_LEN],
223 struct iface **hw_addr_iface);
224 static uint64_t bridge_pick_datapath_id(struct bridge *,
225 const uint8_t bridge_ea[ETH_ADDR_LEN],
226 struct iface *hw_addr_iface);
227 static struct iface *bridge_get_local_iface(struct bridge *);
228 static uint64_t dpid_from_hash(const void *, size_t nbytes);
230 static unixctl_cb_func bridge_unixctl_fdb_show;
232 static void bond_init(void);
233 static void bond_run(struct bridge *);
234 static void bond_wait(struct bridge *);
235 static void bond_rebalance_port(struct port *);
236 static void bond_send_learning_packets(struct port *);
237 static void bond_enable_slave(struct iface *iface, bool enable);
239 static struct port *port_create(struct bridge *, const char *name);
240 static void port_reconfigure(struct port *, const struct ovsrec_port *);
241 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
242 static void port_destroy(struct port *);
243 static struct port *port_lookup(const struct bridge *, const char *name);
244 static struct iface *port_lookup_iface(const struct port *, const char *name);
245 static struct port *port_from_dp_ifidx(const struct bridge *,
247 static void port_update_bond_compat(struct port *);
248 static void port_update_vlan_compat(struct port *);
249 static void port_update_bonding(struct port *);
251 static struct mirror *mirror_create(struct bridge *, const char *name);
252 static void mirror_destroy(struct mirror *);
253 static void mirror_reconfigure(struct bridge *);
254 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
255 static bool vlan_is_mirrored(const struct mirror *, int vlan);
257 static struct iface *iface_create(struct port *port,
258 const struct ovsrec_interface *if_cfg);
259 static void iface_destroy(struct iface *);
260 static struct iface *iface_lookup(const struct bridge *, const char *name);
261 static struct iface *iface_from_dp_ifidx(const struct bridge *,
263 static bool iface_is_internal(const struct bridge *, const char *name);
264 static void iface_set_mac(struct iface *);
266 /* Hooks into ofproto processing. */
267 static struct ofhooks bridge_ofhooks;
269 /* Public functions. */
271 /* Adds the name of each interface used by a bridge, including local and
272 * internal ports, to 'svec'. */
274 bridge_get_ifaces(struct svec *svec)
276 struct bridge *br, *next;
279 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
280 for (i = 0; i < br->n_ports; i++) {
281 struct port *port = br->ports[i];
283 for (j = 0; j < port->n_ifaces; j++) {
284 struct iface *iface = port->ifaces[j];
285 if (iface->dp_ifidx < 0) {
286 VLOG_ERR("%s interface not in datapath %s, ignoring",
287 iface->name, dpif_name(br->dpif));
289 if (iface->dp_ifidx != ODPP_LOCAL) {
290 svec_add(svec, iface->name);
299 bridge_init(const struct ovsrec_open_vswitch *cfg)
301 struct svec bridge_names;
302 struct svec dpif_names, dpif_types;
305 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
307 svec_init(&bridge_names);
308 for (i = 0; i < cfg->n_bridges; i++) {
309 svec_add(&bridge_names, cfg->bridges[i]->name);
311 svec_sort(&bridge_names);
313 svec_init(&dpif_names);
314 svec_init(&dpif_types);
315 dp_enumerate_types(&dpif_types);
316 for (i = 0; i < dpif_types.n; i++) {
321 dp_enumerate_names(dpif_types.names[i], &dpif_names);
323 for (j = 0; j < dpif_names.n; j++) {
324 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
326 struct svec all_names;
329 svec_init(&all_names);
330 dpif_get_all_names(dpif, &all_names);
331 for (k = 0; k < all_names.n; k++) {
332 if (svec_contains(&bridge_names, all_names.names[k])) {
338 svec_destroy(&all_names);
343 svec_destroy(&bridge_names);
344 svec_destroy(&dpif_names);
345 svec_destroy(&dpif_types);
347 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
351 bridge_reconfigure(cfg);
356 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
358 /* XXX SSL should be configurable on a per-bridge basis. */
360 stream_ssl_set_private_key_file(ssl->private_key);
361 stream_ssl_set_certificate_file(ssl->certificate);
362 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
367 /* Attempt to create the network device 'iface_name' through the netdev
370 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
373 struct shash_node *node;
374 struct shash options;
378 shash_init(&options);
379 for (i = 0; i < iface_cfg->n_options; i++) {
380 shash_add(&options, iface_cfg->key_options[i],
381 xstrdup(iface_cfg->value_options[i]));
385 struct netdev_options netdev_options;
387 memset(&netdev_options, 0, sizeof netdev_options);
388 netdev_options.name = iface_cfg->name;
389 if (!strcmp(iface_cfg->type, "internal")) {
390 /* An "internal" config type maps to a netdev "system" type. */
391 netdev_options.type = "system";
393 netdev_options.type = iface_cfg->type;
395 netdev_options.args = &options;
396 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
397 netdev_options.may_create = true;
398 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
399 netdev_options.may_open = true;
402 error = netdev_open(&netdev_options, &iface->netdev);
405 netdev_get_carrier(iface->netdev, &iface->enabled);
407 } else if (iface->netdev) {
408 const char *netdev_type = netdev_get_type(iface->netdev);
409 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
410 ? iface_cfg->type : NULL;
412 /* An "internal" config type maps to a netdev "system" type. */
413 if (iface_type && !strcmp(iface_type, "internal")) {
414 iface_type = "system";
417 if (!iface_type || !strcmp(netdev_type, iface_type)) {
418 error = netdev_reconfigure(iface->netdev, &options);
420 VLOG_WARN("%s: attempting change device type from %s to %s",
421 iface_cfg->name, netdev_type, iface_type);
426 SHASH_FOR_EACH (node, &options) {
429 shash_destroy(&options);
435 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
437 return set_up_iface(iface_cfg, iface, false);
441 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
442 void *aux OVS_UNUSED)
444 if (!iface->netdev) {
445 int error = set_up_iface(iface->cfg, iface, true);
447 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
457 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
458 void *aux OVS_UNUSED)
460 if (iface->dp_ifidx >= 0) {
461 VLOG_DBG("%s has interface %s on port %d",
463 iface->name, iface->dp_ifidx);
466 VLOG_ERR("%s interface not in %s, dropping",
467 iface->name, dpif_name(br->dpif));
473 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
474 void *aux OVS_UNUSED)
476 /* Set policing attributes. */
477 netdev_set_policing(iface->netdev,
478 iface->cfg->ingress_policing_rate,
479 iface->cfg->ingress_policing_burst);
481 /* Set MAC address of internal interfaces other than the local
483 if (iface->dp_ifidx != ODPP_LOCAL
484 && iface_is_internal(br, iface->name)) {
485 iface_set_mac(iface);
491 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
492 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
493 * deletes from 'br' any ports that no longer have any interfaces. */
495 iterate_and_prune_ifaces(struct bridge *br,
496 bool (*cb)(struct bridge *, struct iface *,
502 for (i = 0; i < br->n_ports; ) {
503 struct port *port = br->ports[i];
504 for (j = 0; j < port->n_ifaces; ) {
505 struct iface *iface = port->ifaces[j];
506 if (cb(br, iface, aux)) {
509 iface_destroy(iface);
513 if (port->n_ifaces) {
516 VLOG_ERR("%s port has no interfaces, dropping", port->name);
522 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
523 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
524 * responsible for freeing '*managersp' (with free()).
526 * You may be asking yourself "why does ovs-vswitchd care?", because
527 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
528 * should not be and in fact is not directly involved in that. But
529 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
530 * it has to tell in-band control where the managers are to enable that.
533 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
534 struct sockaddr_in **managersp, size_t *n_managersp)
536 struct sockaddr_in *managers = NULL;
537 size_t n_managers = 0;
539 if (ovs_cfg->n_managers > 0) {
542 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
543 for (i = 0; i < ovs_cfg->n_managers; i++) {
544 const char *name = ovs_cfg->managers[i];
545 struct sockaddr_in *sin = &managers[i];
547 if ((!strncmp(name, "tcp:", 4)
548 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
549 (!strncmp(name, "ssl:", 4)
550 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
556 *managersp = managers;
557 *n_managersp = n_managers;
561 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
563 struct ovsdb_idl_txn *txn;
564 struct shash old_br, new_br;
565 struct shash_node *node;
566 struct bridge *br, *next;
567 struct sockaddr_in *managers;
570 int sflow_bridge_number;
572 COVERAGE_INC(bridge_reconfigure);
574 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
576 collect_managers(ovs_cfg, &managers, &n_managers);
578 /* Collect old and new bridges. */
581 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
582 shash_add(&old_br, br->name, br);
584 for (i = 0; i < ovs_cfg->n_bridges; i++) {
585 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
586 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
587 VLOG_WARN("more than one bridge named %s", br_cfg->name);
591 /* Get rid of deleted bridges and add new bridges. */
592 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
593 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
600 SHASH_FOR_EACH (node, &new_br) {
601 const char *br_name = node->name;
602 const struct ovsrec_bridge *br_cfg = node->data;
603 br = shash_find_data(&old_br, br_name);
605 /* If the bridge datapath type has changed, we need to tear it
606 * down and recreate. */
607 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
609 bridge_create(br_cfg);
612 bridge_create(br_cfg);
615 shash_destroy(&old_br);
616 shash_destroy(&new_br);
620 bridge_configure_ssl(ovs_cfg->ssl);
623 /* Reconfigure all bridges. */
624 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
625 bridge_reconfigure_one(ovs_cfg, br);
628 /* Add and delete ports on all datapaths.
630 * The kernel will reject any attempt to add a given port to a datapath if
631 * that port already belongs to a different datapath, so we must do all
632 * port deletions before any port additions. */
633 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
634 struct odp_port *dpif_ports;
636 struct shash want_ifaces;
638 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
639 bridge_get_all_ifaces(br, &want_ifaces);
640 for (i = 0; i < n_dpif_ports; i++) {
641 const struct odp_port *p = &dpif_ports[i];
642 if (!shash_find(&want_ifaces, p->devname)
643 && strcmp(p->devname, br->name)) {
644 int retval = dpif_port_del(br->dpif, p->port);
646 VLOG_ERR("failed to remove %s interface from %s: %s",
647 p->devname, dpif_name(br->dpif),
652 shash_destroy(&want_ifaces);
655 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
656 struct odp_port *dpif_ports;
658 struct shash cur_ifaces, want_ifaces;
659 struct shash_node *node;
661 /* Get the set of interfaces currently in this datapath. */
662 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
663 shash_init(&cur_ifaces);
664 for (i = 0; i < n_dpif_ports; i++) {
665 const char *name = dpif_ports[i].devname;
666 if (!shash_find(&cur_ifaces, name)) {
667 shash_add(&cur_ifaces, name, NULL);
672 /* Get the set of interfaces we want on this datapath. */
673 bridge_get_all_ifaces(br, &want_ifaces);
675 SHASH_FOR_EACH (node, &want_ifaces) {
676 const char *if_name = node->name;
677 struct iface *iface = node->data;
679 if (shash_find(&cur_ifaces, if_name)) {
680 /* Already exists, just reconfigure it. */
682 reconfigure_iface(iface->cfg, iface);
685 /* Need to add to datapath. */
689 /* Add to datapath. */
690 internal = iface_is_internal(br, if_name);
691 error = dpif_port_add(br->dpif, if_name,
692 internal ? ODP_PORT_INTERNAL : 0, NULL);
693 if (error == EFBIG) {
694 VLOG_ERR("ran out of valid port numbers on %s",
695 dpif_name(br->dpif));
698 VLOG_ERR("failed to add %s interface to %s: %s",
699 if_name, dpif_name(br->dpif), strerror(error));
703 shash_destroy(&cur_ifaces);
704 shash_destroy(&want_ifaces);
706 sflow_bridge_number = 0;
707 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
710 struct iface *local_iface;
711 struct iface *hw_addr_iface;
714 bridge_fetch_dp_ifaces(br);
716 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
717 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
719 /* Pick local port hardware address, datapath ID. */
720 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
721 local_iface = bridge_get_local_iface(br);
723 int error = netdev_set_etheraddr(local_iface->netdev, ea);
725 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
726 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
727 "Ethernet address: %s",
728 br->name, strerror(error));
732 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
733 ofproto_set_datapath_id(br->ofproto, dpid);
735 dpid_string = xasprintf("%012"PRIx64, dpid);
736 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
739 /* Set NetFlow configuration on this bridge. */
740 if (br->cfg->netflow) {
741 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
742 struct netflow_options opts;
744 memset(&opts, 0, sizeof opts);
746 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
747 if (nf_cfg->engine_type) {
748 opts.engine_type = *nf_cfg->engine_type;
750 if (nf_cfg->engine_id) {
751 opts.engine_id = *nf_cfg->engine_id;
754 opts.active_timeout = nf_cfg->active_timeout;
755 if (!opts.active_timeout) {
756 opts.active_timeout = -1;
757 } else if (opts.active_timeout < 0) {
758 VLOG_WARN("bridge %s: active timeout interval set to negative "
759 "value, using default instead (%d seconds)", br->name,
760 NF_ACTIVE_TIMEOUT_DEFAULT);
761 opts.active_timeout = -1;
764 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
765 if (opts.add_id_to_iface) {
766 if (opts.engine_id > 0x7f) {
767 VLOG_WARN("bridge %s: netflow port mangling may conflict "
768 "with another vswitch, choose an engine id less "
769 "than 128", br->name);
771 if (br->n_ports > 508) {
772 VLOG_WARN("bridge %s: netflow port mangling will conflict "
773 "with another port when more than 508 ports are "
778 opts.collectors.n = nf_cfg->n_targets;
779 opts.collectors.names = nf_cfg->targets;
780 if (ofproto_set_netflow(br->ofproto, &opts)) {
781 VLOG_ERR("bridge %s: problem setting netflow collectors",
785 ofproto_set_netflow(br->ofproto, NULL);
788 /* Set sFlow configuration on this bridge. */
789 if (br->cfg->sflow) {
790 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
791 struct ovsrec_controller **controllers;
792 struct ofproto_sflow_options oso;
793 size_t n_controllers;
796 memset(&oso, 0, sizeof oso);
798 oso.targets.n = sflow_cfg->n_targets;
799 oso.targets.names = sflow_cfg->targets;
801 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
802 if (sflow_cfg->sampling) {
803 oso.sampling_rate = *sflow_cfg->sampling;
806 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
807 if (sflow_cfg->polling) {
808 oso.polling_interval = *sflow_cfg->polling;
811 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
812 if (sflow_cfg->header) {
813 oso.header_len = *sflow_cfg->header;
816 oso.sub_id = sflow_bridge_number++;
817 oso.agent_device = sflow_cfg->agent;
819 oso.control_ip = NULL;
820 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
821 for (i = 0; i < n_controllers; i++) {
822 if (controllers[i]->local_ip) {
823 oso.control_ip = controllers[i]->local_ip;
827 ofproto_set_sflow(br->ofproto, &oso);
829 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
831 ofproto_set_sflow(br->ofproto, NULL);
834 /* Update the controller and related settings. It would be more
835 * straightforward to call this from bridge_reconfigure_one(), but we
836 * can't do it there for two reasons. First, and most importantly, at
837 * that point we don't know the dp_ifidx of any interfaces that have
838 * been added to the bridge (because we haven't actually added them to
839 * the datapath). Second, at that point we haven't set the datapath ID
840 * yet; when a controller is configured, resetting the datapath ID will
841 * immediately disconnect from the controller, so it's better to set
842 * the datapath ID before the controller. */
843 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
845 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
846 for (i = 0; i < br->n_ports; i++) {
847 struct port *port = br->ports[i];
849 port_update_vlan_compat(port);
850 port_update_bonding(port);
853 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
854 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
857 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
859 ovsdb_idl_txn_commit(txn);
860 ovsdb_idl_txn_destroy(txn); /* XXX */
866 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
870 for (i = 0; i < n; i++) {
871 if (!strcmp(keys[i], key)) {
879 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
881 return get_ovsrec_key_value(key,
882 br_cfg->key_other_config,
883 br_cfg->value_other_config,
884 br_cfg->n_other_config);
888 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
889 struct iface **hw_addr_iface)
895 *hw_addr_iface = NULL;
897 /* Did the user request a particular MAC? */
898 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
899 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
900 if (eth_addr_is_multicast(ea)) {
901 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
902 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
903 } else if (eth_addr_is_zero(ea)) {
904 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
910 /* Otherwise choose the minimum non-local MAC address among all of the
912 memset(ea, 0xff, sizeof ea);
913 for (i = 0; i < br->n_ports; i++) {
914 struct port *port = br->ports[i];
915 uint8_t iface_ea[ETH_ADDR_LEN];
918 /* Mirror output ports don't participate. */
919 if (port->is_mirror_output_port) {
923 /* Choose the MAC address to represent the port. */
924 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
925 /* Find the interface with this Ethernet address (if any) so that
926 * we can provide the correct devname to the caller. */
928 for (j = 0; j < port->n_ifaces; j++) {
929 struct iface *candidate = port->ifaces[j];
930 uint8_t candidate_ea[ETH_ADDR_LEN];
931 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
932 && eth_addr_equals(iface_ea, candidate_ea)) {
937 /* Choose the interface whose MAC address will represent the port.
938 * The Linux kernel bonding code always chooses the MAC address of
939 * the first slave added to a bond, and the Fedora networking
940 * scripts always add slaves to a bond in alphabetical order, so
941 * for compatibility we choose the interface with the name that is
942 * first in alphabetical order. */
943 iface = port->ifaces[0];
944 for (j = 1; j < port->n_ifaces; j++) {
945 struct iface *candidate = port->ifaces[j];
946 if (strcmp(candidate->name, iface->name) < 0) {
951 /* The local port doesn't count (since we're trying to choose its
952 * MAC address anyway). */
953 if (iface->dp_ifidx == ODPP_LOCAL) {
958 error = netdev_get_etheraddr(iface->netdev, iface_ea);
960 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
961 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
962 iface->name, strerror(error));
967 /* Compare against our current choice. */
968 if (!eth_addr_is_multicast(iface_ea) &&
969 !eth_addr_is_local(iface_ea) &&
970 !eth_addr_is_reserved(iface_ea) &&
971 !eth_addr_is_zero(iface_ea) &&
972 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
974 memcpy(ea, iface_ea, ETH_ADDR_LEN);
975 *hw_addr_iface = iface;
978 if (eth_addr_is_multicast(ea)) {
979 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
980 *hw_addr_iface = NULL;
981 VLOG_WARN("bridge %s: using default bridge Ethernet "
982 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
984 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
985 br->name, ETH_ADDR_ARGS(ea));
989 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
990 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
991 * an interface on 'br', then that interface must be passed in as
992 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
993 * 'hw_addr_iface' must be passed in as a null pointer. */
995 bridge_pick_datapath_id(struct bridge *br,
996 const uint8_t bridge_ea[ETH_ADDR_LEN],
997 struct iface *hw_addr_iface)
1000 * The procedure for choosing a bridge MAC address will, in the most
1001 * ordinary case, also choose a unique MAC that we can use as a datapath
1002 * ID. In some special cases, though, multiple bridges will end up with
1003 * the same MAC address. This is OK for the bridges, but it will confuse
1004 * the OpenFlow controller, because each datapath needs a unique datapath
1007 * Datapath IDs must be unique. It is also very desirable that they be
1008 * stable from one run to the next, so that policy set on a datapath
1011 const char *datapath_id;
1014 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1015 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1019 if (hw_addr_iface) {
1021 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1023 * A bridge whose MAC address is taken from a VLAN network device
1024 * (that is, a network device created with vconfig(8) or similar
1025 * tool) will have the same MAC address as a bridge on the VLAN
1026 * device's physical network device.
1028 * Handle this case by hashing the physical network device MAC
1029 * along with the VLAN identifier.
1031 uint8_t buf[ETH_ADDR_LEN + 2];
1032 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1033 buf[ETH_ADDR_LEN] = vlan >> 8;
1034 buf[ETH_ADDR_LEN + 1] = vlan;
1035 return dpid_from_hash(buf, sizeof buf);
1038 * Assume that this bridge's MAC address is unique, since it
1039 * doesn't fit any of the cases we handle specially.
1044 * A purely internal bridge, that is, one that has no non-virtual
1045 * network devices on it at all, is more difficult because it has no
1046 * natural unique identifier at all.
1048 * When the host is a XenServer, we handle this case by hashing the
1049 * host's UUID with the name of the bridge. Names of bridges are
1050 * persistent across XenServer reboots, although they can be reused if
1051 * an internal network is destroyed and then a new one is later
1052 * created, so this is fairly effective.
1054 * When the host is not a XenServer, we punt by using a random MAC
1055 * address on each run.
1057 const char *host_uuid = xenserver_get_host_uuid();
1059 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1060 dpid = dpid_from_hash(combined, strlen(combined));
1066 return eth_addr_to_uint64(bridge_ea);
1070 dpid_from_hash(const void *data, size_t n)
1072 uint8_t hash[SHA1_DIGEST_SIZE];
1074 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1075 sha1_bytes(data, n, hash);
1076 eth_addr_mark_random(hash);
1077 return eth_addr_to_uint64(hash);
1083 struct bridge *br, *next;
1087 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1088 int error = bridge_run_one(br);
1090 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1091 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1092 "forcing reconfiguration", br->name);
1106 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1107 ofproto_wait(br->ofproto);
1108 if (ofproto_has_controller(br->ofproto)) {
1112 mac_learning_wait(br->ml);
1117 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1118 * configuration changes. */
1120 bridge_flush(struct bridge *br)
1122 COVERAGE_INC(bridge_flush);
1124 mac_learning_flush(br->ml);
1127 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1128 * such interface. */
1129 static struct iface *
1130 bridge_get_local_iface(struct bridge *br)
1134 for (i = 0; i < br->n_ports; i++) {
1135 struct port *port = br->ports[i];
1136 for (j = 0; j < port->n_ifaces; j++) {
1137 struct iface *iface = port->ifaces[j];
1138 if (iface->dp_ifidx == ODPP_LOCAL) {
1147 /* Bridge unixctl user interface functions. */
1149 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1150 const char *args, void *aux OVS_UNUSED)
1152 struct ds ds = DS_EMPTY_INITIALIZER;
1153 const struct bridge *br;
1154 const struct mac_entry *e;
1156 br = bridge_lookup(args);
1158 unixctl_command_reply(conn, 501, "no such bridge");
1162 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1163 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1164 if (e->port < 0 || e->port >= br->n_ports) {
1167 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1168 br->ports[e->port]->ifaces[0]->dp_ifidx,
1169 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1171 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1175 /* Bridge reconfiguration functions. */
1176 static struct bridge *
1177 bridge_create(const struct ovsrec_bridge *br_cfg)
1182 assert(!bridge_lookup(br_cfg->name));
1183 br = xzalloc(sizeof *br);
1185 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1191 dpif_flow_flush(br->dpif);
1193 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1196 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1198 dpif_delete(br->dpif);
1199 dpif_close(br->dpif);
1204 br->name = xstrdup(br_cfg->name);
1206 br->ml = mac_learning_create();
1207 br->sent_config_request = false;
1208 eth_addr_nicira_random(br->default_ea);
1210 port_array_init(&br->ifaces);
1212 shash_init(&br->port_by_name);
1213 shash_init(&br->iface_by_name);
1217 list_push_back(&all_bridges, &br->node);
1219 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1225 bridge_destroy(struct bridge *br)
1230 while (br->n_ports > 0) {
1231 port_destroy(br->ports[br->n_ports - 1]);
1233 list_remove(&br->node);
1234 error = dpif_delete(br->dpif);
1235 if (error && error != ENOENT) {
1236 VLOG_ERR("failed to delete %s: %s",
1237 dpif_name(br->dpif), strerror(error));
1239 dpif_close(br->dpif);
1240 ofproto_destroy(br->ofproto);
1241 mac_learning_destroy(br->ml);
1242 port_array_destroy(&br->ifaces);
1243 shash_destroy(&br->port_by_name);
1244 shash_destroy(&br->iface_by_name);
1251 static struct bridge *
1252 bridge_lookup(const char *name)
1256 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1257 if (!strcmp(br->name, name)) {
1265 bridge_exists(const char *name)
1267 return bridge_lookup(name) ? true : false;
1271 bridge_get_datapathid(const char *name)
1273 struct bridge *br = bridge_lookup(name);
1274 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1277 /* Handle requests for a listing of all flows known by the OpenFlow
1278 * stack, including those normally hidden. */
1280 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1281 const char *args, void *aux OVS_UNUSED)
1286 br = bridge_lookup(args);
1288 unixctl_command_reply(conn, 501, "Unknown bridge");
1293 ofproto_get_all_flows(br->ofproto, &results);
1295 unixctl_command_reply(conn, 200, ds_cstr(&results));
1296 ds_destroy(&results);
1300 bridge_run_one(struct bridge *br)
1304 error = ofproto_run1(br->ofproto);
1309 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1312 error = ofproto_run2(br->ofproto, br->flush);
1319 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1320 const struct bridge *br,
1321 struct ovsrec_controller ***controllersp)
1323 struct ovsrec_controller **controllers;
1324 size_t n_controllers;
1326 if (br->cfg->n_controller) {
1327 controllers = br->cfg->controller;
1328 n_controllers = br->cfg->n_controller;
1330 controllers = ovs_cfg->controller;
1331 n_controllers = ovs_cfg->n_controller;
1334 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1340 *controllersp = controllers;
1342 return n_controllers;
1346 bridge_update_desc(struct bridge *br OVS_UNUSED)
1349 bool changed = false;
1352 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1353 if (desc != br->mfr_desc) {
1356 br->mfr_desc = xstrdup(desc);
1358 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1363 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1364 if (desc != br->hw_desc) {
1367 br->hw_desc = xstrdup(desc);
1369 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1374 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1375 if (desc != br->sw_desc) {
1378 br->sw_desc = xstrdup(desc);
1380 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1385 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1386 if (desc != br->serial_desc) {
1387 free(br->serial_desc);
1389 br->serial_desc = xstrdup(desc);
1391 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1396 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1397 if (desc != br->dp_desc) {
1400 br->dp_desc = xstrdup(desc);
1402 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1408 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1409 br->sw_desc, br->serial_desc, br->dp_desc);
1415 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1418 struct shash old_ports, new_ports;
1419 struct svec listeners, old_listeners;
1420 struct svec snoops, old_snoops;
1421 struct shash_node *node;
1424 /* Collect old ports. */
1425 shash_init(&old_ports);
1426 for (i = 0; i < br->n_ports; i++) {
1427 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1430 /* Collect new ports. */
1431 shash_init(&new_ports);
1432 for (i = 0; i < br->cfg->n_ports; i++) {
1433 const char *name = br->cfg->ports[i]->name;
1434 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1435 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1440 /* If we have a controller, then we need a local port. Complain if the
1441 * user didn't specify one.
1443 * XXX perhaps we should synthesize a port ourselves in this case. */
1444 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1445 char local_name[IF_NAMESIZE];
1448 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1449 local_name, sizeof local_name);
1450 if (!error && !shash_find(&new_ports, local_name)) {
1451 VLOG_WARN("bridge %s: controller specified but no local port "
1452 "(port named %s) defined",
1453 br->name, local_name);
1457 /* Get rid of deleted ports.
1458 * Get rid of deleted interfaces on ports that still exist. */
1459 SHASH_FOR_EACH (node, &old_ports) {
1460 struct port *port = node->data;
1461 const struct ovsrec_port *port_cfg;
1463 port_cfg = shash_find_data(&new_ports, node->name);
1467 port_del_ifaces(port, port_cfg);
1471 /* Create new ports.
1472 * Add new interfaces to existing ports.
1473 * Reconfigure existing ports. */
1474 SHASH_FOR_EACH (node, &new_ports) {
1475 struct port *port = shash_find_data(&old_ports, node->name);
1477 port = port_create(br, node->name);
1480 port_reconfigure(port, node->data);
1481 if (!port->n_ifaces) {
1482 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1483 br->name, port->name);
1487 shash_destroy(&old_ports);
1488 shash_destroy(&new_ports);
1490 /* Delete all flows if we're switching from connected to standalone or vice
1491 * versa. (XXX Should we delete all flows if we are switching from one
1492 * controller to another?) */
1495 /* Configure OpenFlow management listeners. */
1496 svec_init(&listeners);
1497 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1499 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1500 ovs_rundir, br->name));
1501 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1502 svec_clear(&listeners);
1504 svec_sort_unique(&listeners);
1506 svec_init(&old_listeners);
1507 ofproto_get_listeners(br->ofproto, &old_listeners);
1508 svec_sort_unique(&old_listeners);
1510 if (!svec_equal(&listeners, &old_listeners)) {
1511 ofproto_set_listeners(br->ofproto, &listeners);
1513 svec_destroy(&listeners);
1514 svec_destroy(&old_listeners);
1516 /* Configure OpenFlow controller connection snooping. */
1518 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1520 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1521 ovs_rundir, br->name));
1522 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1523 svec_clear(&snoops);
1525 svec_sort_unique(&snoops);
1527 svec_init(&old_snoops);
1528 ofproto_get_snoops(br->ofproto, &old_snoops);
1529 svec_sort_unique(&old_snoops);
1531 if (!svec_equal(&snoops, &old_snoops)) {
1532 ofproto_set_snoops(br->ofproto, &snoops);
1534 svec_destroy(&snoops);
1535 svec_destroy(&old_snoops);
1537 /* Default listener. */
1538 svec_init(&listeners);
1539 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1540 ovs_rundir, br->name));
1541 svec_init(&old_listeners);
1542 ofproto_get_listeners(br->ofproto, &old_listeners);
1543 if (!svec_equal(&listeners, &old_listeners)) {
1544 ofproto_set_listeners(br->ofproto, &listeners);
1546 svec_destroy(&listeners);
1547 svec_destroy(&old_listeners);
1549 /* Default snoop. */
1551 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1552 ovs_rundir, br->name));
1553 svec_init(&old_snoops);
1554 ofproto_get_snoops(br->ofproto, &old_snoops);
1555 if (!svec_equal(&snoops, &old_snoops)) {
1556 ofproto_set_snoops(br->ofproto, &snoops);
1558 svec_destroy(&snoops);
1559 svec_destroy(&old_snoops);
1562 mirror_reconfigure(br);
1564 bridge_update_desc(br);
1568 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1570 const struct sockaddr_in *managers,
1573 struct ovsrec_controller **controllers;
1574 size_t n_controllers;
1576 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1578 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1579 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1580 ofproto_flush_flows(br->ofproto);
1583 if (!n_controllers) {
1584 union ofp_action action;
1587 /* Clear out controllers. */
1588 ofproto_set_controllers(br->ofproto, NULL, 0);
1590 /* Set up a flow that matches every packet and directs them to
1591 * OFPP_NORMAL (which goes to us). */
1592 memset(&action, 0, sizeof action);
1593 action.type = htons(OFPAT_OUTPUT);
1594 action.output.len = htons(sizeof action);
1595 action.output.port = htons(OFPP_NORMAL);
1596 memset(&flow, 0, sizeof flow);
1597 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1599 struct ofproto_controller *ocs;
1602 ocs = xmalloc(n_controllers * sizeof *ocs);
1603 for (i = 0; i < n_controllers; i++) {
1604 struct ovsrec_controller *c = controllers[i];
1605 struct ofproto_controller *oc = &ocs[i];
1607 if (strcmp(c->target, "discover")) {
1608 struct iface *local_iface;
1611 local_iface = bridge_get_local_iface(br);
1612 if (local_iface && c->local_ip
1613 && inet_aton(c->local_ip, &ip)) {
1614 struct netdev *netdev = local_iface->netdev;
1615 struct in_addr mask, gateway;
1617 if (!c->local_netmask
1618 || !inet_aton(c->local_netmask, &mask)) {
1621 if (!c->local_gateway
1622 || !inet_aton(c->local_gateway, &gateway)) {
1626 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1628 mask.s_addr = guess_netmask(ip.s_addr);
1630 if (!netdev_set_in4(netdev, ip, mask)) {
1631 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1633 br->name, IP_ARGS(&ip.s_addr),
1634 IP_ARGS(&mask.s_addr));
1637 if (gateway.s_addr) {
1638 if (!netdev_add_router(netdev, gateway)) {
1639 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1640 br->name, IP_ARGS(&gateway.s_addr));
1646 oc->target = c->target;
1647 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1648 oc->probe_interval = (c->inactivity_probe
1649 ? *c->inactivity_probe / 1000 : 5);
1650 oc->fail = (!c->fail_mode
1651 || !strcmp(c->fail_mode, "standalone")
1652 || !strcmp(c->fail_mode, "open")
1653 ? OFPROTO_FAIL_STANDALONE
1654 : OFPROTO_FAIL_SECURE);
1655 oc->band = (!c->connection_mode
1656 || !strcmp(c->connection_mode, "in-band")
1658 : OFPROTO_OUT_OF_BAND);
1659 oc->accept_re = c->discover_accept_regex;
1660 oc->update_resolv_conf = c->discover_update_resolv_conf;
1661 oc->rate_limit = (c->controller_rate_limit
1662 ? *c->controller_rate_limit : 0);
1663 oc->burst_limit = (c->controller_burst_limit
1664 ? *c->controller_burst_limit : 0);
1666 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1672 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1677 for (i = 0; i < br->n_ports; i++) {
1678 struct port *port = br->ports[i];
1679 for (j = 0; j < port->n_ifaces; j++) {
1680 struct iface *iface = port->ifaces[j];
1681 shash_add_once(ifaces, iface->name, iface);
1683 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1684 shash_add_once(ifaces, port->name, NULL);
1689 /* For robustness, in case the administrator moves around datapath ports behind
1690 * our back, we re-check all the datapath port numbers here.
1692 * This function will set the 'dp_ifidx' members of interfaces that have
1693 * disappeared to -1, so only call this function from a context where those
1694 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1695 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1696 * datapath, which doesn't support UINT16_MAX+1 ports. */
1698 bridge_fetch_dp_ifaces(struct bridge *br)
1700 struct odp_port *dpif_ports;
1701 size_t n_dpif_ports;
1704 /* Reset all interface numbers. */
1705 for (i = 0; i < br->n_ports; i++) {
1706 struct port *port = br->ports[i];
1707 for (j = 0; j < port->n_ifaces; j++) {
1708 struct iface *iface = port->ifaces[j];
1709 iface->dp_ifidx = -1;
1712 port_array_clear(&br->ifaces);
1714 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1715 for (i = 0; i < n_dpif_ports; i++) {
1716 struct odp_port *p = &dpif_ports[i];
1717 struct iface *iface = iface_lookup(br, p->devname);
1719 if (iface->dp_ifidx >= 0) {
1720 VLOG_WARN("%s reported interface %s twice",
1721 dpif_name(br->dpif), p->devname);
1722 } else if (iface_from_dp_ifidx(br, p->port)) {
1723 VLOG_WARN("%s reported interface %"PRIu16" twice",
1724 dpif_name(br->dpif), p->port);
1726 port_array_set(&br->ifaces, p->port, iface);
1727 iface->dp_ifidx = p->port;
1731 int64_t ofport = (iface->dp_ifidx >= 0
1732 ? odp_port_to_ofp_port(iface->dp_ifidx)
1734 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1741 /* Bridge packet processing functions. */
1744 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1746 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1749 static struct bond_entry *
1750 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1752 return &port->bond_hash[bond_hash(mac)];
1756 bond_choose_iface(const struct port *port)
1758 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1759 size_t i, best_down_slave = -1;
1760 long long next_delay_expiration = LLONG_MAX;
1762 for (i = 0; i < port->n_ifaces; i++) {
1763 struct iface *iface = port->ifaces[i];
1765 if (iface->enabled) {
1767 } else if (iface->delay_expires < next_delay_expiration) {
1768 best_down_slave = i;
1769 next_delay_expiration = iface->delay_expires;
1773 if (best_down_slave != -1) {
1774 struct iface *iface = port->ifaces[best_down_slave];
1776 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1777 "since no other interface is up", iface->name,
1778 iface->delay_expires - time_msec());
1779 bond_enable_slave(iface, true);
1782 return best_down_slave;
1786 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1787 uint16_t *dp_ifidx, tag_type *tags)
1789 struct iface *iface;
1791 assert(port->n_ifaces);
1792 if (port->n_ifaces == 1) {
1793 iface = port->ifaces[0];
1795 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1796 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1797 || !port->ifaces[e->iface_idx]->enabled) {
1798 /* XXX select interface properly. The current interface selection
1799 * is only good for testing the rebalancing code. */
1800 e->iface_idx = bond_choose_iface(port);
1801 if (e->iface_idx < 0) {
1802 *tags |= port->no_ifaces_tag;
1805 e->iface_tag = tag_create_random();
1806 ((struct port *) port)->bond_compat_is_stale = true;
1808 *tags |= e->iface_tag;
1809 iface = port->ifaces[e->iface_idx];
1811 *dp_ifidx = iface->dp_ifidx;
1812 *tags |= iface->tag; /* Currently only used for bonding. */
1817 bond_link_status_update(struct iface *iface, bool carrier)
1819 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1820 struct port *port = iface->port;
1822 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1823 /* Nothing to do. */
1826 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1827 iface->name, carrier ? "detected" : "dropped");
1828 if (carrier == iface->enabled) {
1829 iface->delay_expires = LLONG_MAX;
1830 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1831 iface->name, carrier ? "disabled" : "enabled");
1832 } else if (carrier && port->active_iface < 0) {
1833 bond_enable_slave(iface, true);
1834 if (port->updelay) {
1835 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1836 "other interface is up", iface->name, port->updelay);
1839 int delay = carrier ? port->updelay : port->downdelay;
1840 iface->delay_expires = time_msec() + delay;
1843 "interface %s: will be %s if it stays %s for %d ms",
1845 carrier ? "enabled" : "disabled",
1846 carrier ? "up" : "down",
1853 bond_choose_active_iface(struct port *port)
1855 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1857 port->active_iface = bond_choose_iface(port);
1858 port->active_iface_tag = tag_create_random();
1859 if (port->active_iface >= 0) {
1860 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1861 port->name, port->ifaces[port->active_iface]->name);
1863 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1869 bond_enable_slave(struct iface *iface, bool enable)
1871 struct port *port = iface->port;
1872 struct bridge *br = port->bridge;
1874 /* This acts as a recursion check. If the act of disabling a slave
1875 * causes a different slave to be enabled, the flag will allow us to
1876 * skip redundant work when we reenter this function. It must be
1877 * cleared on exit to keep things safe with multiple bonds. */
1878 static bool moving_active_iface = false;
1880 iface->delay_expires = LLONG_MAX;
1881 if (enable == iface->enabled) {
1885 iface->enabled = enable;
1886 if (!iface->enabled) {
1887 VLOG_WARN("interface %s: disabled", iface->name);
1888 ofproto_revalidate(br->ofproto, iface->tag);
1889 if (iface->port_ifidx == port->active_iface) {
1890 ofproto_revalidate(br->ofproto,
1891 port->active_iface_tag);
1893 /* Disabling a slave can lead to another slave being immediately
1894 * enabled if there will be no active slaves but one is waiting
1895 * on an updelay. In this case we do not need to run most of the
1896 * code for the newly enabled slave since there was no period
1897 * without an active slave and it is redundant with the disabling
1899 moving_active_iface = true;
1900 bond_choose_active_iface(port);
1902 bond_send_learning_packets(port);
1904 VLOG_WARN("interface %s: enabled", iface->name);
1905 if (port->active_iface < 0 && !moving_active_iface) {
1906 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1907 bond_choose_active_iface(port);
1908 bond_send_learning_packets(port);
1910 iface->tag = tag_create_random();
1913 moving_active_iface = false;
1914 port->bond_compat_is_stale = true;
1917 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1918 * bond interface. */
1920 bond_update_fake_iface_stats(struct port *port)
1922 struct netdev_stats bond_stats;
1923 struct netdev *bond_dev;
1926 memset(&bond_stats, 0, sizeof bond_stats);
1928 for (i = 0; i < port->n_ifaces; i++) {
1929 struct netdev_stats slave_stats;
1931 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1932 bond_stats.rx_packets += slave_stats.rx_packets;
1933 bond_stats.rx_bytes += slave_stats.rx_bytes;
1934 bond_stats.tx_packets += slave_stats.tx_packets;
1935 bond_stats.tx_bytes += slave_stats.tx_bytes;
1939 if (!netdev_open_default(port->name, &bond_dev)) {
1940 netdev_set_stats(bond_dev, &bond_stats);
1941 netdev_close(bond_dev);
1946 bond_run(struct bridge *br)
1950 for (i = 0; i < br->n_ports; i++) {
1951 struct port *port = br->ports[i];
1953 if (port->n_ifaces >= 2) {
1954 for (j = 0; j < port->n_ifaces; j++) {
1955 struct iface *iface = port->ifaces[j];
1956 if (time_msec() >= iface->delay_expires) {
1957 bond_enable_slave(iface, !iface->enabled);
1961 if (port->bond_fake_iface
1962 && time_msec() >= port->bond_next_fake_iface_update) {
1963 bond_update_fake_iface_stats(port);
1964 port->bond_next_fake_iface_update = time_msec() + 1000;
1968 if (port->bond_compat_is_stale) {
1969 port->bond_compat_is_stale = false;
1970 port_update_bond_compat(port);
1976 bond_wait(struct bridge *br)
1980 for (i = 0; i < br->n_ports; i++) {
1981 struct port *port = br->ports[i];
1982 if (port->n_ifaces < 2) {
1985 for (j = 0; j < port->n_ifaces; j++) {
1986 struct iface *iface = port->ifaces[j];
1987 if (iface->delay_expires != LLONG_MAX) {
1988 poll_timer_wait(iface->delay_expires - time_msec());
1991 if (port->bond_fake_iface) {
1992 poll_timer_wait(port->bond_next_fake_iface_update - time_msec());
1998 set_dst(struct dst *p, const flow_t *flow,
1999 const struct port *in_port, const struct port *out_port,
2002 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2003 : in_port->vlan >= 0 ? in_port->vlan
2004 : ntohs(flow->dl_vlan));
2005 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2009 swap_dst(struct dst *p, struct dst *q)
2011 struct dst tmp = *p;
2016 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2017 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2018 * that we push to the datapath. We could in fact fully sort the array by
2019 * vlan, but in most cases there are at most two different vlan tags so that's
2020 * possibly overkill.) */
2022 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2024 struct dst *first = dsts;
2025 struct dst *last = dsts + n_dsts;
2027 while (first != last) {
2029 * - All dsts < first have vlan == 'vlan'.
2030 * - All dsts >= last have vlan != 'vlan'.
2031 * - first < last. */
2032 while (first->vlan == vlan) {
2033 if (++first == last) {
2038 /* Same invariants, plus one additional:
2039 * - first->vlan != vlan.
2041 while (last[-1].vlan != vlan) {
2042 if (--last == first) {
2047 /* Same invariants, plus one additional:
2048 * - last[-1].vlan == vlan.*/
2049 swap_dst(first++, --last);
2054 mirror_mask_ffs(mirror_mask_t mask)
2056 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2061 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2062 const struct dst *test)
2065 for (i = 0; i < n_dsts; i++) {
2066 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2074 port_trunks_vlan(const struct port *port, uint16_t vlan)
2076 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
2080 port_includes_vlan(const struct port *port, uint16_t vlan)
2082 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2086 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2087 const struct port *in_port, const struct port *out_port,
2088 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2090 mirror_mask_t mirrors = in_port->src_mirrors;
2091 struct dst *dst = dsts;
2094 if (out_port == FLOOD_PORT) {
2095 /* XXX use ODP_FLOOD if no vlans or bonding. */
2096 /* XXX even better, define each VLAN as a datapath port group */
2097 for (i = 0; i < br->n_ports; i++) {
2098 struct port *port = br->ports[i];
2099 if (port != in_port && port_includes_vlan(port, vlan)
2100 && !port->is_mirror_output_port
2101 && set_dst(dst, flow, in_port, port, tags)) {
2102 mirrors |= port->dst_mirrors;
2106 *nf_output_iface = NF_OUT_FLOOD;
2107 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2108 *nf_output_iface = dst->dp_ifidx;
2109 mirrors |= out_port->dst_mirrors;
2114 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2115 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2117 if (set_dst(dst, flow, in_port, m->out_port, tags)
2118 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2122 for (i = 0; i < br->n_ports; i++) {
2123 struct port *port = br->ports[i];
2124 if (port_includes_vlan(port, m->out_vlan)
2125 && set_dst(dst, flow, in_port, port, tags))
2129 if (port->vlan < 0) {
2130 dst->vlan = m->out_vlan;
2132 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2136 /* Use the vlan tag on the original flow instead of
2137 * the one passed in the vlan parameter. This ensures
2138 * that we compare the vlan from before any implicit
2139 * tagging tags place. This is necessary because
2140 * dst->vlan is the final vlan, after removing implicit
2142 flow_vlan = ntohs(flow->dl_vlan);
2143 if (flow_vlan == 0) {
2144 flow_vlan = OFP_VLAN_NONE;
2146 if (port == in_port && dst->vlan == flow_vlan) {
2147 /* Don't send out input port on same VLAN. */
2155 mirrors &= mirrors - 1;
2158 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2162 static void OVS_UNUSED
2163 print_dsts(const struct dst *dsts, size_t n)
2165 for (; n--; dsts++) {
2166 printf(">p%"PRIu16, dsts->dp_ifidx);
2167 if (dsts->vlan != OFP_VLAN_NONE) {
2168 printf("v%"PRIu16, dsts->vlan);
2174 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2175 const struct port *in_port, const struct port *out_port,
2176 tag_type *tags, struct odp_actions *actions,
2177 uint16_t *nf_output_iface)
2179 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2181 const struct dst *p;
2184 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2187 cur_vlan = ntohs(flow->dl_vlan);
2188 for (p = dsts; p < &dsts[n_dsts]; p++) {
2189 union odp_action *a;
2190 if (p->vlan != cur_vlan) {
2191 if (p->vlan == OFP_VLAN_NONE) {
2192 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2194 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2195 a->vlan_vid.vlan_vid = htons(p->vlan);
2199 a = odp_actions_add(actions, ODPAT_OUTPUT);
2200 a->output.port = p->dp_ifidx;
2204 /* Returns the effective vlan of a packet, taking into account both the
2205 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2206 * the packet is untagged and -1 indicates it has an invalid header and
2207 * should be dropped. */
2208 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2209 struct port *in_port, bool have_packet)
2211 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2212 * belongs to VLAN 0, so we should treat both cases identically. (In the
2213 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2214 * presumably to allow a priority to be specified. In the latter case, the
2215 * packet does not have any 802.1Q header.) */
2216 int vlan = ntohs(flow->dl_vlan);
2217 if (vlan == OFP_VLAN_NONE) {
2220 if (in_port->vlan >= 0) {
2222 /* XXX support double tagging? */
2224 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2225 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2226 "packet received on port %s configured with "
2227 "implicit VLAN %"PRIu16,
2228 br->name, ntohs(flow->dl_vlan),
2229 in_port->name, in_port->vlan);
2233 vlan = in_port->vlan;
2235 if (!port_includes_vlan(in_port, vlan)) {
2237 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2238 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2239 "packet received on port %s not configured for "
2241 br->name, vlan, in_port->name, vlan);
2251 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2252 struct port *in_port)
2254 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2255 vlan, in_port->port_idx);
2257 /* The log messages here could actually be useful in debugging,
2258 * so keep the rate limit relatively high. */
2259 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2261 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2262 "on port %s in VLAN %d",
2263 br->name, ETH_ADDR_ARGS(flow->dl_src),
2264 in_port->name, vlan);
2265 ofproto_revalidate(br->ofproto, rev_tag);
2270 is_bcast_arp_reply(const flow_t *flow)
2272 return (flow->dl_type == htons(ETH_TYPE_ARP)
2273 && flow->nw_proto == ARP_OP_REPLY
2274 && eth_addr_is_broadcast(flow->dl_dst));
2277 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2278 * dropped. Returns true if they may be forwarded, false if they should be
2281 * If 'have_packet' is true, it indicates that the caller is processing a
2282 * received packet. If 'have_packet' is false, then the caller is just
2283 * revalidating an existing flow because configuration has changed. Either
2284 * way, 'have_packet' only affects logging (there is no point in logging errors
2285 * during revalidation).
2287 * Sets '*in_portp' to the input port. This will be a null pointer if
2288 * flow->in_port does not designate a known input port (in which case
2289 * is_admissible() returns false).
2291 * When returning true, sets '*vlanp' to the effective VLAN of the input
2292 * packet, as returned by flow_get_vlan().
2294 * May also add tags to '*tags', although the current implementation only does
2295 * so in one special case.
2298 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2299 tag_type *tags, int *vlanp, struct port **in_portp)
2301 struct iface *in_iface;
2302 struct port *in_port;
2305 /* Find the interface and port structure for the received packet. */
2306 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2308 /* No interface? Something fishy... */
2310 /* Odd. A few possible reasons here:
2312 * - We deleted an interface but there are still a few packets
2313 * queued up from it.
2315 * - Someone externally added an interface (e.g. with "ovs-dpctl
2316 * add-if") that we don't know about.
2318 * - Packet arrived on the local port but the local port is not
2319 * one of our bridge ports.
2321 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2323 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2324 "interface %"PRIu16, br->name, flow->in_port);
2330 *in_portp = in_port = in_iface->port;
2331 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2336 /* Drop frames for reserved multicast addresses. */
2337 if (eth_addr_is_reserved(flow->dl_dst)) {
2341 /* Drop frames on ports reserved for mirroring. */
2342 if (in_port->is_mirror_output_port) {
2344 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2345 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2346 "%s, which is reserved exclusively for mirroring",
2347 br->name, in_port->name);
2352 /* Packets received on bonds need special attention to avoid duplicates. */
2353 if (in_port->n_ifaces > 1) {
2356 if (eth_addr_is_multicast(flow->dl_dst)) {
2357 *tags |= in_port->active_iface_tag;
2358 if (in_port->active_iface != in_iface->port_ifidx) {
2359 /* Drop all multicast packets on inactive slaves. */
2364 /* Drop all packets for which we have learned a different input
2365 * port, because we probably sent the packet on one slave and got
2366 * it back on the other. Broadcast ARP replies are an exception
2367 * to this rule: the host has moved to another switch. */
2368 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2369 if (src_idx != -1 && src_idx != in_port->port_idx &&
2370 !is_bcast_arp_reply(flow)) {
2378 /* If the composed actions may be applied to any packet in the given 'flow',
2379 * returns true. Otherwise, the actions should only be applied to 'packet', or
2380 * not at all, if 'packet' was NULL. */
2382 process_flow(struct bridge *br, const flow_t *flow,
2383 const struct ofpbuf *packet, struct odp_actions *actions,
2384 tag_type *tags, uint16_t *nf_output_iface)
2386 struct port *in_port;
2387 struct port *out_port;
2391 /* Check whether we should drop packets in this flow. */
2392 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2397 /* Learn source MAC (but don't try to learn from revalidation). */
2399 update_learning_table(br, flow, vlan, in_port);
2402 /* Determine output port. */
2403 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags);
2404 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2405 out_port = br->ports[out_port_idx];
2406 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2407 /* If we are revalidating but don't have a learning entry then
2408 * eject the flow. Installing a flow that floods packets opens
2409 * up a window of time where we could learn from a packet reflected
2410 * on a bond and blackhole packets before the learning table is
2411 * updated to reflect the correct port. */
2414 out_port = FLOOD_PORT;
2417 /* Don't send packets out their input ports. */
2418 if (in_port == out_port) {
2424 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2431 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2434 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2435 const struct ofp_phy_port *opp,
2438 struct bridge *br = br_;
2439 struct iface *iface;
2442 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2448 if (reason == OFPPR_DELETE) {
2449 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2450 br->name, iface->name);
2451 iface_destroy(iface);
2452 if (!port->n_ifaces) {
2453 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2454 br->name, port->name);
2460 if (port->n_ifaces > 1) {
2461 bool up = !(opp->state & OFPPS_LINK_DOWN);
2462 bond_link_status_update(iface, up);
2463 port_update_bond_compat(port);
2469 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2470 struct odp_actions *actions, tag_type *tags,
2471 uint16_t *nf_output_iface, void *br_)
2473 struct bridge *br = br_;
2475 COVERAGE_INC(bridge_process_flow);
2476 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2480 bridge_account_flow_ofhook_cb(const flow_t *flow,
2481 const union odp_action *actions,
2482 size_t n_actions, unsigned long long int n_bytes,
2485 struct bridge *br = br_;
2486 const union odp_action *a;
2487 struct port *in_port;
2491 /* Feed information from the active flows back into the learning table
2492 * to ensure that table is always in sync with what is actually flowing
2493 * through the datapath. */
2494 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2495 update_learning_table(br, flow, vlan, in_port);
2498 if (!br->has_bonded_ports) {
2502 for (a = actions; a < &actions[n_actions]; a++) {
2503 if (a->type == ODPAT_OUTPUT) {
2504 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2505 if (out_port && out_port->n_ifaces >= 2) {
2506 struct bond_entry *e = lookup_bond_entry(out_port,
2508 e->tx_bytes += n_bytes;
2515 bridge_account_checkpoint_ofhook_cb(void *br_)
2517 struct bridge *br = br_;
2521 if (!br->has_bonded_ports) {
2526 for (i = 0; i < br->n_ports; i++) {
2527 struct port *port = br->ports[i];
2528 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2529 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2530 bond_rebalance_port(port);
2535 static struct ofhooks bridge_ofhooks = {
2536 bridge_port_changed_ofhook_cb,
2537 bridge_normal_ofhook_cb,
2538 bridge_account_flow_ofhook_cb,
2539 bridge_account_checkpoint_ofhook_cb,
2542 /* Bonding functions. */
2544 /* Statistics for a single interface on a bonded port, used for load-based
2545 * bond rebalancing. */
2546 struct slave_balance {
2547 struct iface *iface; /* The interface. */
2548 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2550 /* All the "bond_entry"s that are assigned to this interface, in order of
2551 * increasing tx_bytes. */
2552 struct bond_entry **hashes;
2556 /* Sorts pointers to pointers to bond_entries in ascending order by the
2557 * interface to which they are assigned, and within a single interface in
2558 * ascending order of bytes transmitted. */
2560 compare_bond_entries(const void *a_, const void *b_)
2562 const struct bond_entry *const *ap = a_;
2563 const struct bond_entry *const *bp = b_;
2564 const struct bond_entry *a = *ap;
2565 const struct bond_entry *b = *bp;
2566 if (a->iface_idx != b->iface_idx) {
2567 return a->iface_idx > b->iface_idx ? 1 : -1;
2568 } else if (a->tx_bytes != b->tx_bytes) {
2569 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2575 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2576 * *descending* order by number of bytes transmitted. */
2578 compare_slave_balance(const void *a_, const void *b_)
2580 const struct slave_balance *a = a_;
2581 const struct slave_balance *b = b_;
2582 if (a->iface->enabled != b->iface->enabled) {
2583 return a->iface->enabled ? -1 : 1;
2584 } else if (a->tx_bytes != b->tx_bytes) {
2585 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2592 swap_bals(struct slave_balance *a, struct slave_balance *b)
2594 struct slave_balance tmp = *a;
2599 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2600 * given that 'p' (and only 'p') might be in the wrong location.
2602 * This function invalidates 'p', since it might now be in a different memory
2605 resort_bals(struct slave_balance *p,
2606 struct slave_balance bals[], size_t n_bals)
2609 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2610 swap_bals(p, p - 1);
2612 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2613 swap_bals(p, p + 1);
2619 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2621 if (VLOG_IS_DBG_ENABLED()) {
2622 struct ds ds = DS_EMPTY_INITIALIZER;
2623 const struct slave_balance *b;
2625 for (b = bals; b < bals + n_bals; b++) {
2629 ds_put_char(&ds, ',');
2631 ds_put_format(&ds, " %s %"PRIu64"kB",
2632 b->iface->name, b->tx_bytes / 1024);
2634 if (!b->iface->enabled) {
2635 ds_put_cstr(&ds, " (disabled)");
2637 if (b->n_hashes > 0) {
2638 ds_put_cstr(&ds, " (");
2639 for (i = 0; i < b->n_hashes; i++) {
2640 const struct bond_entry *e = b->hashes[i];
2642 ds_put_cstr(&ds, " + ");
2644 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2645 e - port->bond_hash, e->tx_bytes / 1024);
2647 ds_put_cstr(&ds, ")");
2650 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2655 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2657 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2660 struct bond_entry *hash = from->hashes[hash_idx];
2661 struct port *port = from->iface->port;
2662 uint64_t delta = hash->tx_bytes;
2664 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2665 "from %s to %s (now carrying %"PRIu64"kB and "
2666 "%"PRIu64"kB load, respectively)",
2667 port->name, delta / 1024, hash - port->bond_hash,
2668 from->iface->name, to->iface->name,
2669 (from->tx_bytes - delta) / 1024,
2670 (to->tx_bytes + delta) / 1024);
2672 /* Delete element from from->hashes.
2674 * We don't bother to add the element to to->hashes because not only would
2675 * it require more work, the only purpose it would be to allow that hash to
2676 * be migrated to another slave in this rebalancing run, and there is no
2677 * point in doing that. */
2678 if (hash_idx == 0) {
2681 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2682 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2686 /* Shift load away from 'from' to 'to'. */
2687 from->tx_bytes -= delta;
2688 to->tx_bytes += delta;
2690 /* Arrange for flows to be revalidated. */
2691 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2692 hash->iface_idx = to->iface->port_ifidx;
2693 hash->iface_tag = tag_create_random();
2697 bond_rebalance_port(struct port *port)
2699 struct slave_balance bals[DP_MAX_PORTS];
2701 struct bond_entry *hashes[BOND_MASK + 1];
2702 struct slave_balance *b, *from, *to;
2703 struct bond_entry *e;
2706 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2707 * descending order of tx_bytes, so that bals[0] represents the most
2708 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2711 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2712 * array for each slave_balance structure, we sort our local array of
2713 * hashes in order by slave, so that all of the hashes for a given slave
2714 * become contiguous in memory, and then we point each 'hashes' members of
2715 * a slave_balance structure to the start of a contiguous group. */
2716 n_bals = port->n_ifaces;
2717 for (b = bals; b < &bals[n_bals]; b++) {
2718 b->iface = port->ifaces[b - bals];
2723 for (i = 0; i <= BOND_MASK; i++) {
2724 hashes[i] = &port->bond_hash[i];
2726 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2727 for (i = 0; i <= BOND_MASK; i++) {
2729 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2730 b = &bals[e->iface_idx];
2731 b->tx_bytes += e->tx_bytes;
2733 b->hashes = &hashes[i];
2738 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2739 log_bals(bals, n_bals, port);
2741 /* Discard slaves that aren't enabled (which were sorted to the back of the
2742 * array earlier). */
2743 while (!bals[n_bals - 1].iface->enabled) {
2750 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2751 to = &bals[n_bals - 1];
2752 for (from = bals; from < to; ) {
2753 uint64_t overload = from->tx_bytes - to->tx_bytes;
2754 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2755 /* The extra load on 'from' (and all less-loaded slaves), compared
2756 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2757 * it is less than ~1Mbps. No point in rebalancing. */
2759 } else if (from->n_hashes == 1) {
2760 /* 'from' only carries a single MAC hash, so we can't shift any
2761 * load away from it, even though we want to. */
2764 /* 'from' is carrying significantly more load than 'to', and that
2765 * load is split across at least two different hashes. Pick a hash
2766 * to migrate to 'to' (the least-loaded slave), given that doing so
2767 * must decrease the ratio of the load on the two slaves by at
2770 * The sort order we use means that we prefer to shift away the
2771 * smallest hashes instead of the biggest ones. There is little
2772 * reason behind this decision; we could use the opposite sort
2773 * order to shift away big hashes ahead of small ones. */
2777 for (i = 0; i < from->n_hashes; i++) {
2778 double old_ratio, new_ratio;
2779 uint64_t delta = from->hashes[i]->tx_bytes;
2781 if (delta == 0 || from->tx_bytes - delta == 0) {
2782 /* Pointless move. */
2786 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2788 if (to->tx_bytes == 0) {
2789 /* Nothing on the new slave, move it. */
2793 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2794 new_ratio = (double)(from->tx_bytes - delta) /
2795 (to->tx_bytes + delta);
2797 if (new_ratio == 0) {
2798 /* Should already be covered but check to prevent division
2803 if (new_ratio < 1) {
2804 new_ratio = 1 / new_ratio;
2807 if (old_ratio - new_ratio > 0.1) {
2808 /* Would decrease the ratio, move it. */
2812 if (i < from->n_hashes) {
2813 bond_shift_load(from, to, i);
2814 port->bond_compat_is_stale = true;
2816 /* If the result of the migration changed the relative order of
2817 * 'from' and 'to' swap them back to maintain invariants. */
2818 if (order_swapped) {
2819 swap_bals(from, to);
2822 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2823 * point to different slave_balance structures. It is only
2824 * valid to do these two operations in a row at all because we
2825 * know that 'from' will not move past 'to' and vice versa. */
2826 resort_bals(from, bals, n_bals);
2827 resort_bals(to, bals, n_bals);
2834 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2835 * historical data to decay to <1% in 7 rebalancing runs. */
2836 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2842 bond_send_learning_packets(struct port *port)
2844 struct bridge *br = port->bridge;
2845 struct mac_entry *e;
2846 struct ofpbuf packet;
2847 int error, n_packets, n_errors;
2849 if (!port->n_ifaces || port->active_iface < 0) {
2853 ofpbuf_init(&packet, 128);
2854 error = n_packets = n_errors = 0;
2855 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2856 union ofp_action actions[2], *a;
2862 if (e->port == port->port_idx
2863 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2867 /* Compose actions. */
2868 memset(actions, 0, sizeof actions);
2871 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2872 a->vlan_vid.len = htons(sizeof *a);
2873 a->vlan_vid.vlan_vid = htons(e->vlan);
2876 a->output.type = htons(OFPAT_OUTPUT);
2877 a->output.len = htons(sizeof *a);
2878 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2883 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2885 flow_extract(&packet, 0, ODPP_NONE, &flow);
2886 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2893 ofpbuf_uninit(&packet);
2896 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2897 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2898 "packets, last error was: %s",
2899 port->name, n_errors, n_packets, strerror(error));
2901 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2902 port->name, n_packets);
2906 /* Bonding unixctl user interface functions. */
2909 bond_unixctl_list(struct unixctl_conn *conn,
2910 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2912 struct ds ds = DS_EMPTY_INITIALIZER;
2913 const struct bridge *br;
2915 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2917 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2920 for (i = 0; i < br->n_ports; i++) {
2921 const struct port *port = br->ports[i];
2922 if (port->n_ifaces > 1) {
2925 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2926 for (j = 0; j < port->n_ifaces; j++) {
2927 const struct iface *iface = port->ifaces[j];
2929 ds_put_cstr(&ds, ", ");
2931 ds_put_cstr(&ds, iface->name);
2933 ds_put_char(&ds, '\n');
2937 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2941 static struct port *
2942 bond_find(const char *name)
2944 const struct bridge *br;
2946 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2949 for (i = 0; i < br->n_ports; i++) {
2950 struct port *port = br->ports[i];
2951 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2960 bond_unixctl_show(struct unixctl_conn *conn,
2961 const char *args, void *aux OVS_UNUSED)
2963 struct ds ds = DS_EMPTY_INITIALIZER;
2964 const struct port *port;
2967 port = bond_find(args);
2969 unixctl_command_reply(conn, 501, "no such bond");
2973 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2974 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2975 ds_put_format(&ds, "next rebalance: %lld ms\n",
2976 port->bond_next_rebalance - time_msec());
2977 for (j = 0; j < port->n_ifaces; j++) {
2978 const struct iface *iface = port->ifaces[j];
2979 struct bond_entry *be;
2982 ds_put_format(&ds, "slave %s: %s\n",
2983 iface->name, iface->enabled ? "enabled" : "disabled");
2984 if (j == port->active_iface) {
2985 ds_put_cstr(&ds, "\tactive slave\n");
2987 if (iface->delay_expires != LLONG_MAX) {
2988 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2989 iface->enabled ? "downdelay" : "updelay",
2990 iface->delay_expires - time_msec());
2994 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2995 int hash = be - port->bond_hash;
2996 struct mac_entry *me;
2998 if (be->iface_idx != j) {
3002 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3003 hash, be->tx_bytes / 1024);
3006 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3007 &port->bridge->ml->lrus) {
3010 if (bond_hash(me->mac) == hash
3011 && me->port != port->port_idx
3012 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3013 && dp_ifidx == iface->dp_ifidx)
3015 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3016 ETH_ADDR_ARGS(me->mac));
3021 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3026 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3027 void *aux OVS_UNUSED)
3029 char *args = (char *) args_;
3030 char *save_ptr = NULL;
3031 char *bond_s, *hash_s, *slave_s;
3032 uint8_t mac[ETH_ADDR_LEN];
3034 struct iface *iface;
3035 struct bond_entry *entry;
3038 bond_s = strtok_r(args, " ", &save_ptr);
3039 hash_s = strtok_r(NULL, " ", &save_ptr);
3040 slave_s = strtok_r(NULL, " ", &save_ptr);
3042 unixctl_command_reply(conn, 501,
3043 "usage: bond/migrate BOND HASH SLAVE");
3047 port = bond_find(bond_s);
3049 unixctl_command_reply(conn, 501, "no such bond");
3053 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3054 == ETH_ADDR_SCAN_COUNT) {
3055 hash = bond_hash(mac);
3056 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3057 hash = atoi(hash_s) & BOND_MASK;
3059 unixctl_command_reply(conn, 501, "bad hash");
3063 iface = port_lookup_iface(port, slave_s);
3065 unixctl_command_reply(conn, 501, "no such slave");
3069 if (!iface->enabled) {
3070 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3074 entry = &port->bond_hash[hash];
3075 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3076 entry->iface_idx = iface->port_ifidx;
3077 entry->iface_tag = tag_create_random();
3078 port->bond_compat_is_stale = true;
3079 unixctl_command_reply(conn, 200, "migrated");
3083 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3084 void *aux OVS_UNUSED)
3086 char *args = (char *) args_;
3087 char *save_ptr = NULL;
3088 char *bond_s, *slave_s;
3090 struct iface *iface;
3092 bond_s = strtok_r(args, " ", &save_ptr);
3093 slave_s = strtok_r(NULL, " ", &save_ptr);
3095 unixctl_command_reply(conn, 501,
3096 "usage: bond/set-active-slave BOND SLAVE");
3100 port = bond_find(bond_s);
3102 unixctl_command_reply(conn, 501, "no such bond");
3106 iface = port_lookup_iface(port, slave_s);
3108 unixctl_command_reply(conn, 501, "no such slave");
3112 if (!iface->enabled) {
3113 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3117 if (port->active_iface != iface->port_ifidx) {
3118 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3119 port->active_iface = iface->port_ifidx;
3120 port->active_iface_tag = tag_create_random();
3121 VLOG_INFO("port %s: active interface is now %s",
3122 port->name, iface->name);
3123 bond_send_learning_packets(port);
3124 unixctl_command_reply(conn, 200, "done");
3126 unixctl_command_reply(conn, 200, "no change");
3131 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3133 char *args = (char *) args_;
3134 char *save_ptr = NULL;
3135 char *bond_s, *slave_s;
3137 struct iface *iface;
3139 bond_s = strtok_r(args, " ", &save_ptr);
3140 slave_s = strtok_r(NULL, " ", &save_ptr);
3142 unixctl_command_reply(conn, 501,
3143 "usage: bond/enable/disable-slave BOND SLAVE");
3147 port = bond_find(bond_s);
3149 unixctl_command_reply(conn, 501, "no such bond");
3153 iface = port_lookup_iface(port, slave_s);
3155 unixctl_command_reply(conn, 501, "no such slave");
3159 bond_enable_slave(iface, enable);
3160 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3164 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3165 void *aux OVS_UNUSED)
3167 enable_slave(conn, args, true);
3171 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3172 void *aux OVS_UNUSED)
3174 enable_slave(conn, args, false);
3178 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3179 void *aux OVS_UNUSED)
3181 uint8_t mac[ETH_ADDR_LEN];
3185 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3186 == ETH_ADDR_SCAN_COUNT) {
3187 hash = bond_hash(mac);
3189 hash_cstr = xasprintf("%u", hash);
3190 unixctl_command_reply(conn, 200, hash_cstr);
3193 unixctl_command_reply(conn, 501, "invalid mac");
3200 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3201 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3202 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3203 unixctl_command_register("bond/set-active-slave",
3204 bond_unixctl_set_active_slave, NULL);
3205 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3207 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3209 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3212 /* Port functions. */
3214 static struct port *
3215 port_create(struct bridge *br, const char *name)
3219 port = xzalloc(sizeof *port);
3221 port->port_idx = br->n_ports;
3223 port->trunks = NULL;
3224 port->name = xstrdup(name);
3225 port->active_iface = -1;
3227 if (br->n_ports >= br->allocated_ports) {
3228 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3231 br->ports[br->n_ports++] = port;
3232 shash_add_assert(&br->port_by_name, port->name, port);
3234 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3241 get_port_other_config(const struct ovsrec_port *port, const char *key,
3242 const char *default_value)
3244 const char *value = get_ovsrec_key_value(key,
3245 port->key_other_config,
3246 port->value_other_config,
3247 port->n_other_config);
3248 return value ? value : default_value;
3252 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3254 struct shash new_ifaces;
3257 /* Collect list of new interfaces. */
3258 shash_init(&new_ifaces);
3259 for (i = 0; i < cfg->n_interfaces; i++) {
3260 const char *name = cfg->interfaces[i]->name;
3261 shash_add_once(&new_ifaces, name, NULL);
3264 /* Get rid of deleted interfaces. */
3265 for (i = 0; i < port->n_ifaces; ) {
3266 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3267 iface_destroy(port->ifaces[i]);
3273 shash_destroy(&new_ifaces);
3277 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3279 struct shash new_ifaces;
3280 long long int next_rebalance;
3281 unsigned long *trunks;
3287 /* Update settings. */
3288 port->updelay = cfg->bond_updelay;
3289 if (port->updelay < 0) {
3292 port->updelay = cfg->bond_downdelay;
3293 if (port->downdelay < 0) {
3294 port->downdelay = 0;
3296 port->bond_rebalance_interval = atoi(
3297 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3298 if (port->bond_rebalance_interval < 1000) {
3299 port->bond_rebalance_interval = 1000;
3301 next_rebalance = time_msec() + port->bond_rebalance_interval;
3302 if (port->bond_next_rebalance > next_rebalance) {
3303 port->bond_next_rebalance = next_rebalance;
3306 /* Add new interfaces and update 'cfg' member of existing ones. */
3307 shash_init(&new_ifaces);
3308 for (i = 0; i < cfg->n_interfaces; i++) {
3309 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3310 struct iface *iface;
3312 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3313 VLOG_WARN("port %s: %s specified twice as port interface",
3314 port->name, if_cfg->name);
3318 iface = iface_lookup(port->bridge, if_cfg->name);
3320 if (iface->port != port) {
3321 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3323 port->bridge->name, if_cfg->name, iface->port->name);
3326 iface->cfg = if_cfg;
3328 iface_create(port, if_cfg);
3331 shash_destroy(&new_ifaces);
3336 if (port->n_ifaces < 2) {
3338 if (vlan >= 0 && vlan <= 4095) {
3339 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3344 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3345 * they even work as-is. But they have not been tested. */
3346 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3350 if (port->vlan != vlan) {
3352 bridge_flush(port->bridge);
3355 /* Get trunked VLANs. */
3361 trunks = bitmap_allocate(4096);
3363 for (i = 0; i < cfg->n_trunks; i++) {
3364 int trunk = cfg->trunks[i];
3366 bitmap_set1(trunks, trunk);
3372 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3373 port->name, cfg->n_trunks);
3375 if (n_errors == cfg->n_trunks) {
3377 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3380 bitmap_set_multiple(trunks, 0, 4096, 1);
3383 if (cfg->n_trunks) {
3384 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3389 ? port->trunks != NULL
3390 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3391 bridge_flush(port->bridge);
3393 bitmap_free(port->trunks);
3394 port->trunks = trunks;
3396 shash_destroy(&new_ifaces);
3400 port_destroy(struct port *port)
3403 struct bridge *br = port->bridge;
3407 proc_net_compat_update_vlan(port->name, NULL, 0);
3408 proc_net_compat_update_bond(port->name, NULL);
3410 for (i = 0; i < MAX_MIRRORS; i++) {
3411 struct mirror *m = br->mirrors[i];
3412 if (m && m->out_port == port) {
3417 while (port->n_ifaces > 0) {
3418 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3421 shash_find_and_delete_assert(&br->port_by_name, port->name);
3423 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3424 del->port_idx = port->port_idx;
3427 bitmap_free(port->trunks);
3434 static struct port *
3435 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3437 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3438 return iface ? iface->port : NULL;
3441 static struct port *
3442 port_lookup(const struct bridge *br, const char *name)
3444 return shash_find_data(&br->port_by_name, name);
3447 static struct iface *
3448 port_lookup_iface(const struct port *port, const char *name)
3450 struct iface *iface = iface_lookup(port->bridge, name);
3451 return iface && iface->port == port ? iface : NULL;
3455 port_update_bonding(struct port *port)
3457 if (port->n_ifaces < 2) {
3458 /* Not a bonded port. */
3459 if (port->bond_hash) {
3460 free(port->bond_hash);
3461 port->bond_hash = NULL;
3462 port->bond_compat_is_stale = true;
3463 port->bond_fake_iface = false;
3466 if (!port->bond_hash) {
3469 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3470 for (i = 0; i <= BOND_MASK; i++) {
3471 struct bond_entry *e = &port->bond_hash[i];
3475 port->no_ifaces_tag = tag_create_random();
3476 bond_choose_active_iface(port);
3477 port->bond_next_rebalance
3478 = time_msec() + port->bond_rebalance_interval;
3480 if (port->cfg->bond_fake_iface) {
3481 port->bond_next_fake_iface_update = time_msec();
3484 port->bond_compat_is_stale = true;
3485 port->bond_fake_iface = port->cfg->bond_fake_iface;
3490 port_update_bond_compat(struct port *port)
3492 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3493 struct compat_bond bond;
3496 if (port->n_ifaces < 2) {
3497 proc_net_compat_update_bond(port->name, NULL);
3502 bond.updelay = port->updelay;
3503 bond.downdelay = port->downdelay;
3506 bond.hashes = compat_hashes;
3507 if (port->bond_hash) {
3508 const struct bond_entry *e;
3509 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3510 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3511 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3512 cbh->hash = e - port->bond_hash;
3513 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3518 bond.n_slaves = port->n_ifaces;
3519 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3520 for (i = 0; i < port->n_ifaces; i++) {
3521 struct iface *iface = port->ifaces[i];
3522 struct compat_bond_slave *slave = &bond.slaves[i];
3523 slave->name = iface->name;
3525 /* We need to make the same determination as the Linux bonding
3526 * code to determine whether a slave should be consider "up".
3527 * The Linux function bond_miimon_inspect() supports four
3528 * BOND_LINK_* states:
3530 * - BOND_LINK_UP: carrier detected, updelay has passed.
3531 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3532 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3533 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3535 * The function bond_info_show_slave() only considers BOND_LINK_UP
3536 * to be "up" and anything else to be "down".
3538 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3542 netdev_get_etheraddr(iface->netdev, slave->mac);
3545 if (port->bond_fake_iface) {
3546 struct netdev *bond_netdev;
3548 if (!netdev_open_default(port->name, &bond_netdev)) {
3550 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3552 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3554 netdev_close(bond_netdev);
3558 proc_net_compat_update_bond(port->name, &bond);
3563 port_update_vlan_compat(struct port *port)
3565 struct bridge *br = port->bridge;
3566 char *vlandev_name = NULL;
3568 if (port->vlan > 0) {
3569 /* Figure out the name that the VLAN device should actually have, if it
3570 * existed. This takes some work because the VLAN device would not
3571 * have port->name in its name; rather, it would have the trunk port's
3572 * name, and 'port' would be attached to a bridge that also had the
3573 * VLAN device one of its ports. So we need to find a trunk port that
3574 * includes port->vlan.
3576 * There might be more than one candidate. This doesn't happen on
3577 * XenServer, so if it happens we just pick the first choice in
3578 * alphabetical order instead of creating multiple VLAN devices. */
3580 for (i = 0; i < br->n_ports; i++) {
3581 struct port *p = br->ports[i];
3582 if (port_trunks_vlan(p, port->vlan)
3584 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3586 uint8_t ea[ETH_ADDR_LEN];
3587 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3588 if (!eth_addr_is_multicast(ea) &&
3589 !eth_addr_is_reserved(ea) &&
3590 !eth_addr_is_zero(ea)) {
3591 vlandev_name = p->name;
3596 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3599 /* Interface functions. */
3601 static struct iface *
3602 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3604 struct bridge *br = port->bridge;
3605 struct iface *iface;
3606 char *name = if_cfg->name;
3609 iface = xzalloc(sizeof *iface);
3611 iface->port_ifidx = port->n_ifaces;
3612 iface->name = xstrdup(name);
3613 iface->dp_ifidx = -1;
3614 iface->tag = tag_create_random();
3615 iface->delay_expires = LLONG_MAX;
3616 iface->netdev = NULL;
3617 iface->cfg = if_cfg;
3619 /* Attempt to create the network interface in case it doesn't exist yet. */
3620 if (!iface_is_internal(br, iface->name)) {
3621 error = set_up_iface(if_cfg, iface, true);
3623 VLOG_WARN("could not create iface %s: %s", iface->name,
3632 shash_add_assert(&br->iface_by_name, iface->name, iface);
3634 if (port->n_ifaces >= port->allocated_ifaces) {
3635 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3636 sizeof *port->ifaces);
3638 port->ifaces[port->n_ifaces++] = iface;
3639 if (port->n_ifaces > 1) {
3640 br->has_bonded_ports = true;
3643 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3651 iface_destroy(struct iface *iface)
3654 struct port *port = iface->port;
3655 struct bridge *br = port->bridge;
3656 bool del_active = port->active_iface == iface->port_ifidx;
3659 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3661 if (iface->dp_ifidx >= 0) {
3662 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3665 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3666 del->port_ifidx = iface->port_ifidx;
3668 netdev_close(iface->netdev);
3671 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3672 bond_choose_active_iface(port);
3673 bond_send_learning_packets(port);
3679 bridge_flush(port->bridge);
3683 static struct iface *
3684 iface_lookup(const struct bridge *br, const char *name)
3686 return shash_find_data(&br->iface_by_name, name);
3689 static struct iface *
3690 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3692 return port_array_get(&br->ifaces, dp_ifidx);
3695 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3696 * 'br', that is, an interface that is entirely simulated within the datapath.
3697 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3698 * interfaces are created by setting "iface.<iface>.internal = true".
3700 * In addition, we have a kluge-y feature that creates an internal port with
3701 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3702 * This feature needs to go away in the long term. Until then, this is one
3703 * reason why this function takes a name instead of a struct iface: the fake
3704 * interfaces created this way do not have a struct iface. */
3706 iface_is_internal(const struct bridge *br, const char *if_name)
3708 struct iface *iface;
3711 if (!strcmp(if_name, br->name)) {
3715 iface = iface_lookup(br, if_name);
3716 if (iface && !strcmp(iface->cfg->type, "internal")) {
3720 port = port_lookup(br, if_name);
3721 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3727 /* Set Ethernet address of 'iface', if one is specified in the configuration
3730 iface_set_mac(struct iface *iface)
3732 uint8_t ea[ETH_ADDR_LEN];
3734 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3735 if (eth_addr_is_multicast(ea)) {
3736 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3738 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3739 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3740 iface->name, iface->name);
3742 int error = netdev_set_etheraddr(iface->netdev, ea);
3744 VLOG_ERR("interface %s: setting MAC failed (%s)",
3745 iface->name, strerror(error));
3751 /* Port mirroring. */
3754 mirror_reconfigure(struct bridge *br)
3756 struct shash old_mirrors, new_mirrors;
3757 struct shash_node *node;
3758 unsigned long *rspan_vlans;
3761 /* Collect old mirrors. */
3762 shash_init(&old_mirrors);
3763 for (i = 0; i < MAX_MIRRORS; i++) {
3764 if (br->mirrors[i]) {
3765 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3769 /* Collect new mirrors. */
3770 shash_init(&new_mirrors);
3771 for (i = 0; i < br->cfg->n_mirrors; i++) {
3772 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3773 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3774 VLOG_WARN("bridge %s: %s specified twice as mirror",
3775 br->name, cfg->name);
3779 /* Get rid of deleted mirrors and add new mirrors. */
3780 SHASH_FOR_EACH (node, &old_mirrors) {
3781 if (!shash_find(&new_mirrors, node->name)) {
3782 mirror_destroy(node->data);
3785 SHASH_FOR_EACH (node, &new_mirrors) {
3786 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3788 mirror = mirror_create(br, node->name);
3793 mirror_reconfigure_one(mirror, node->data);
3795 shash_destroy(&old_mirrors);
3796 shash_destroy(&new_mirrors);
3798 /* Update port reserved status. */
3799 for (i = 0; i < br->n_ports; i++) {
3800 br->ports[i]->is_mirror_output_port = false;
3802 for (i = 0; i < MAX_MIRRORS; i++) {
3803 struct mirror *m = br->mirrors[i];
3804 if (m && m->out_port) {
3805 m->out_port->is_mirror_output_port = true;
3809 /* Update flooded vlans (for RSPAN). */
3811 if (br->cfg->n_flood_vlans) {
3812 rspan_vlans = bitmap_allocate(4096);
3814 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3815 int64_t vlan = br->cfg->flood_vlans[i];
3816 if (vlan >= 0 && vlan < 4096) {
3817 bitmap_set1(rspan_vlans, vlan);
3818 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3821 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3826 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3831 static struct mirror *
3832 mirror_create(struct bridge *br, const char *name)
3837 for (i = 0; ; i++) {
3838 if (i >= MAX_MIRRORS) {
3839 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3840 "cannot create %s", br->name, MAX_MIRRORS, name);
3843 if (!br->mirrors[i]) {
3848 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3851 br->mirrors[i] = m = xzalloc(sizeof *m);
3854 m->name = xstrdup(name);
3855 shash_init(&m->src_ports);
3856 shash_init(&m->dst_ports);
3866 mirror_destroy(struct mirror *m)
3869 struct bridge *br = m->bridge;
3872 for (i = 0; i < br->n_ports; i++) {
3873 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3874 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3877 shash_destroy(&m->src_ports);
3878 shash_destroy(&m->dst_ports);
3881 m->bridge->mirrors[m->idx] = NULL;
3889 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3890 struct shash *names)
3894 for (i = 0; i < n_ports; i++) {
3895 const char *name = ports[i]->name;
3896 if (port_lookup(m->bridge, name)) {
3897 shash_add_once(names, name, NULL);
3899 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3900 "port %s", m->bridge->name, m->name, name);
3906 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3912 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3914 for (i = 0; i < cfg->n_select_vlan; i++) {
3915 int64_t vlan = cfg->select_vlan[i];
3916 if (vlan < 0 || vlan > 4095) {
3917 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3918 m->bridge->name, m->name, vlan);
3920 (*vlans)[n_vlans++] = vlan;
3927 vlan_is_mirrored(const struct mirror *m, int vlan)
3931 for (i = 0; i < m->n_vlans; i++) {
3932 if (m->vlans[i] == vlan) {
3940 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3944 for (i = 0; i < m->n_vlans; i++) {
3945 if (port_trunks_vlan(p, m->vlans[i])) {
3953 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3955 struct shash src_ports, dst_ports;
3956 mirror_mask_t mirror_bit;
3957 struct port *out_port;
3963 /* Get output port. */
3964 if (cfg->output_port) {
3965 out_port = port_lookup(m->bridge, cfg->output_port->name);
3967 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3968 m->bridge->name, m->name);
3974 if (cfg->output_vlan) {
3975 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3976 "output vlan; ignoring output vlan",
3977 m->bridge->name, m->name);
3979 } else if (cfg->output_vlan) {
3981 out_vlan = *cfg->output_vlan;
3983 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3984 m->bridge->name, m->name);
3989 shash_init(&src_ports);
3990 shash_init(&dst_ports);
3991 if (cfg->select_all) {
3992 for (i = 0; i < m->bridge->n_ports; i++) {
3993 const char *name = m->bridge->ports[i]->name;
3994 shash_add_once(&src_ports, name, NULL);
3995 shash_add_once(&dst_ports, name, NULL);
4000 /* Get ports, and drop duplicates and ports that don't exist. */
4001 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4003 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4006 /* Get all the vlans, and drop duplicate and invalid vlans. */
4007 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4010 /* Update mirror data. */
4011 if (!shash_equal_keys(&m->src_ports, &src_ports)
4012 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4013 || m->n_vlans != n_vlans
4014 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4015 || m->out_port != out_port
4016 || m->out_vlan != out_vlan) {
4017 bridge_flush(m->bridge);
4019 shash_swap(&m->src_ports, &src_ports);
4020 shash_swap(&m->dst_ports, &dst_ports);
4023 m->n_vlans = n_vlans;
4024 m->out_port = out_port;
4025 m->out_vlan = out_vlan;
4028 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4029 for (i = 0; i < m->bridge->n_ports; i++) {
4030 struct port *port = m->bridge->ports[i];
4032 if (shash_find(&m->src_ports, port->name)
4035 ? port_trunks_any_mirrored_vlan(m, port)
4036 : vlan_is_mirrored(m, port->vlan)))) {
4037 port->src_mirrors |= mirror_bit;
4039 port->src_mirrors &= ~mirror_bit;
4042 if (shash_find(&m->dst_ports, port->name)) {
4043 port->dst_mirrors |= mirror_bit;
4045 port->dst_mirrors &= ~mirror_bit;
4050 shash_destroy(&src_ports);
4051 shash_destroy(&dst_ports);