1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
50 #include "poll-loop.h"
51 #include "port-array.h"
52 #include "proc-net-compat.h"
56 #include "socket-util.h"
57 #include "stream-ssl.h"
63 #include "vswitchd/vswitch-idl.h"
64 #include "xenserver.h"
66 #include "sflow_api.h"
68 #define THIS_MODULE VLM_bridge
77 /* These members are always valid. */
78 struct port *port; /* Containing port. */
79 size_t port_ifidx; /* Index within containing port. */
80 char *name; /* Host network device name. */
81 tag_type tag; /* Tag associated with this interface. */
82 long long delay_expires; /* Time after which 'enabled' may change. */
84 /* These members are valid only after bridge_reconfigure() causes them to
86 int dp_ifidx; /* Index within kernel datapath. */
87 struct netdev *netdev; /* Network device. */
88 bool enabled; /* May be chosen for flows? */
90 /* This member is only valid *during* bridge_reconfigure(). */
91 const struct ovsrec_interface *cfg;
94 #define BOND_MASK 0xff
96 int iface_idx; /* Index of assigned iface, or -1 if none. */
97 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
98 tag_type iface_tag; /* Tag associated with iface_idx. */
101 #define MAX_MIRRORS 32
102 typedef uint32_t mirror_mask_t;
103 #define MIRROR_MASK_C(X) UINT32_C(X)
104 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
106 struct bridge *bridge;
110 /* Selection criteria. */
111 struct shash src_ports; /* Name is port name; data is always NULL. */
112 struct shash dst_ports; /* Name is port name; data is always NULL. */
117 struct port *out_port;
121 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
123 struct bridge *bridge;
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
130 /* An ordinary bridge port has 1 interface.
131 * A bridge port for bonding has at least 2 interfaces. */
132 struct iface **ifaces;
133 size_t n_ifaces, allocated_ifaces;
136 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
137 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
138 tag_type active_iface_tag; /* Tag for bcast flows. */
139 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
140 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
141 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
142 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
143 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
144 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
145 long long int bond_next_rebalance; /* Next rebalancing time. */
147 /* Port mirroring info. */
148 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
149 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
150 bool is_mirror_output_port; /* Does port mirroring send frames here? */
152 /* This member is only valid *during* bridge_reconfigure(). */
153 const struct ovsrec_port *cfg;
156 #define DP_MAX_PORTS 255
158 struct list node; /* Node in global list of bridges. */
159 char *name; /* User-specified arbitrary name. */
160 struct mac_learning *ml; /* MAC learning table. */
161 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
163 /* OpenFlow switch processing. */
164 struct ofproto *ofproto; /* OpenFlow switch. */
166 /* Description strings. */
167 char *mfr_desc; /* Manufacturer. */
168 char *hw_desc; /* Hardware. */
169 char *sw_desc; /* Software version. */
170 char *serial_desc; /* Serial number. */
171 char *dp_desc; /* Datapath description. */
173 /* Kernel datapath information. */
174 struct dpif *dpif; /* Datapath. */
175 struct port_array ifaces; /* Indexed by kernel datapath port number. */
179 size_t n_ports, allocated_ports;
180 struct shash iface_by_name; /* "struct iface"s indexed by name. */
181 struct shash port_by_name; /* "struct port"s indexed by name. */
184 bool has_bonded_ports;
189 /* Port mirroring. */
190 struct mirror *mirrors[MAX_MIRRORS];
192 /* This member is only valid *during* bridge_reconfigure(). */
193 const struct ovsrec_bridge *cfg;
196 /* List of all bridges. */
197 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
199 /* Maximum number of datapaths. */
200 enum { DP_MAX = 256 };
202 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
203 static void bridge_destroy(struct bridge *);
204 static struct bridge *bridge_lookup(const char *name);
205 static unixctl_cb_func bridge_unixctl_dump_flows;
206 static int bridge_run_one(struct bridge *);
207 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
208 const struct bridge *br,
209 struct ovsrec_controller ***controllersp);
210 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
212 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
214 const struct sockaddr_in *managers,
216 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
217 static void bridge_fetch_dp_ifaces(struct bridge *);
218 static void bridge_flush(struct bridge *);
219 static void bridge_pick_local_hw_addr(struct bridge *,
220 uint8_t ea[ETH_ADDR_LEN],
221 struct iface **hw_addr_iface);
222 static uint64_t bridge_pick_datapath_id(struct bridge *,
223 const uint8_t bridge_ea[ETH_ADDR_LEN],
224 struct iface *hw_addr_iface);
225 static struct iface *bridge_get_local_iface(struct bridge *);
226 static uint64_t dpid_from_hash(const void *, size_t nbytes);
228 static unixctl_cb_func bridge_unixctl_fdb_show;
230 static void bond_init(void);
231 static void bond_run(struct bridge *);
232 static void bond_wait(struct bridge *);
233 static void bond_rebalance_port(struct port *);
234 static void bond_send_learning_packets(struct port *);
235 static void bond_enable_slave(struct iface *iface, bool enable);
237 static struct port *port_create(struct bridge *, const char *name);
238 static void port_reconfigure(struct port *, const struct ovsrec_port *);
239 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
240 static void port_destroy(struct port *);
241 static struct port *port_lookup(const struct bridge *, const char *name);
242 static struct iface *port_lookup_iface(const struct port *, const char *name);
243 static struct port *port_from_dp_ifidx(const struct bridge *,
245 static void port_update_bond_compat(struct port *);
246 static void port_update_vlan_compat(struct port *);
247 static void port_update_bonding(struct port *);
249 static struct mirror *mirror_create(struct bridge *, const char *name);
250 static void mirror_destroy(struct mirror *);
251 static void mirror_reconfigure(struct bridge *);
252 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
253 static bool vlan_is_mirrored(const struct mirror *, int vlan);
255 static struct iface *iface_create(struct port *port,
256 const struct ovsrec_interface *if_cfg);
257 static void iface_destroy(struct iface *);
258 static struct iface *iface_lookup(const struct bridge *, const char *name);
259 static struct iface *iface_from_dp_ifidx(const struct bridge *,
261 static bool iface_is_internal(const struct bridge *, const char *name);
262 static void iface_set_mac(struct iface *);
264 /* Hooks into ofproto processing. */
265 static struct ofhooks bridge_ofhooks;
267 /* Public functions. */
269 /* Adds the name of each interface used by a bridge, including local and
270 * internal ports, to 'svec'. */
272 bridge_get_ifaces(struct svec *svec)
274 struct bridge *br, *next;
277 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
278 for (i = 0; i < br->n_ports; i++) {
279 struct port *port = br->ports[i];
281 for (j = 0; j < port->n_ifaces; j++) {
282 struct iface *iface = port->ifaces[j];
283 if (iface->dp_ifidx < 0) {
284 VLOG_ERR("%s interface not in datapath %s, ignoring",
285 iface->name, dpif_name(br->dpif));
287 if (iface->dp_ifidx != ODPP_LOCAL) {
288 svec_add(svec, iface->name);
297 bridge_init(const struct ovsrec_open_vswitch *cfg)
299 struct svec bridge_names;
300 struct svec dpif_names, dpif_types;
303 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
305 svec_init(&bridge_names);
306 for (i = 0; i < cfg->n_bridges; i++) {
307 svec_add(&bridge_names, cfg->bridges[i]->name);
309 svec_sort(&bridge_names);
311 svec_init(&dpif_names);
312 svec_init(&dpif_types);
313 dp_enumerate_types(&dpif_types);
314 for (i = 0; i < dpif_types.n; i++) {
319 dp_enumerate_names(dpif_types.names[i], &dpif_names);
321 for (j = 0; j < dpif_names.n; j++) {
322 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
324 struct svec all_names;
327 svec_init(&all_names);
328 dpif_get_all_names(dpif, &all_names);
329 for (k = 0; k < all_names.n; k++) {
330 if (svec_contains(&bridge_names, all_names.names[k])) {
336 svec_destroy(&all_names);
341 svec_destroy(&bridge_names);
342 svec_destroy(&dpif_names);
343 svec_destroy(&dpif_types);
345 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
349 bridge_reconfigure(cfg);
354 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
356 /* XXX SSL should be configurable on a per-bridge basis. */
358 stream_ssl_set_private_key_file(ssl->private_key);
359 stream_ssl_set_certificate_file(ssl->certificate);
360 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
365 /* Attempt to create the network device 'iface_name' through the netdev
368 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
371 struct shash_node *node;
372 struct shash options;
376 shash_init(&options);
377 for (i = 0; i < iface_cfg->n_options; i++) {
378 shash_add(&options, iface_cfg->key_options[i],
379 xstrdup(iface_cfg->value_options[i]));
383 struct netdev_options netdev_options;
385 memset(&netdev_options, 0, sizeof netdev_options);
386 netdev_options.name = iface_cfg->name;
387 if (!strcmp(iface_cfg->type, "internal")) {
388 /* An "internal" config type maps to a netdev "system" type. */
389 netdev_options.type = "system";
391 netdev_options.type = iface_cfg->type;
393 netdev_options.args = &options;
394 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
396 error = netdev_open(&netdev_options, &iface->netdev);
399 netdev_get_carrier(iface->netdev, &iface->enabled);
401 } else if (iface->netdev) {
402 const char *netdev_type = netdev_get_type(iface->netdev);
403 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
404 ? iface_cfg->type : NULL;
406 /* An "internal" config type maps to a netdev "system" type. */
407 if (iface_type && !strcmp(iface_type, "internal")) {
408 iface_type = "system";
411 if (!iface_type || !strcmp(netdev_type, iface_type)) {
412 error = netdev_reconfigure(iface->netdev, &options);
414 VLOG_WARN("%s: attempting change device type from %s to %s",
415 iface_cfg->name, netdev_type, iface_type);
420 SHASH_FOR_EACH (node, &options) {
423 shash_destroy(&options);
429 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
431 return set_up_iface(iface_cfg, iface, false);
435 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
436 void *aux OVS_UNUSED)
438 if (!iface->netdev) {
439 int error = set_up_iface(iface->cfg, iface, true);
441 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
451 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
452 void *aux OVS_UNUSED)
454 if (iface->dp_ifidx >= 0) {
455 VLOG_DBG("%s has interface %s on port %d",
457 iface->name, iface->dp_ifidx);
460 VLOG_ERR("%s interface not in %s, dropping",
461 iface->name, dpif_name(br->dpif));
467 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
468 void *aux OVS_UNUSED)
470 /* Set policing attributes. */
471 netdev_set_policing(iface->netdev,
472 iface->cfg->ingress_policing_rate,
473 iface->cfg->ingress_policing_burst);
475 /* Set MAC address of internal interfaces other than the local
477 if (iface->dp_ifidx != ODPP_LOCAL
478 && iface_is_internal(br, iface->name)) {
479 iface_set_mac(iface);
485 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
486 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
487 * deletes from 'br' any ports that no longer have any interfaces. */
489 iterate_and_prune_ifaces(struct bridge *br,
490 bool (*cb)(struct bridge *, struct iface *,
496 for (i = 0; i < br->n_ports; ) {
497 struct port *port = br->ports[i];
498 for (j = 0; j < port->n_ifaces; ) {
499 struct iface *iface = port->ifaces[j];
500 if (cb(br, iface, aux)) {
503 iface_destroy(iface);
507 if (port->n_ifaces) {
510 VLOG_ERR("%s port has no interfaces, dropping", port->name);
516 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
517 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
518 * responsible for freeing '*managersp' (with free()).
520 * You may be asking yourself "why does ovs-vswitchd care?", because
521 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
522 * should not be and in fact is not directly involved in that. But
523 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
524 * it has to tell in-band control where the managers are to enable that.
527 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
528 struct sockaddr_in **managersp, size_t *n_managersp)
530 struct sockaddr_in *managers = NULL;
531 size_t n_managers = 0;
533 if (ovs_cfg->n_managers > 0) {
536 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
537 for (i = 0; i < ovs_cfg->n_managers; i++) {
538 const char *name = ovs_cfg->managers[i];
539 struct sockaddr_in *sin = &managers[i];
541 if ((!strncmp(name, "tcp:", 4)
542 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
543 (!strncmp(name, "ssl:", 4)
544 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
550 *managersp = managers;
551 *n_managersp = n_managers;
555 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
557 struct ovsdb_idl_txn *txn;
558 struct shash old_br, new_br;
559 struct shash_node *node;
560 struct bridge *br, *next;
561 struct sockaddr_in *managers;
564 int sflow_bridge_number;
566 COVERAGE_INC(bridge_reconfigure);
568 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
570 collect_managers(ovs_cfg, &managers, &n_managers);
572 /* Collect old and new bridges. */
575 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
576 shash_add(&old_br, br->name, br);
578 for (i = 0; i < ovs_cfg->n_bridges; i++) {
579 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
580 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
581 VLOG_WARN("more than one bridge named %s", br_cfg->name);
585 /* Get rid of deleted bridges and add new bridges. */
586 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
587 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
594 SHASH_FOR_EACH (node, &new_br) {
595 const char *br_name = node->name;
596 const struct ovsrec_bridge *br_cfg = node->data;
597 br = shash_find_data(&old_br, br_name);
599 /* If the bridge datapath type has changed, we need to tear it
600 * down and recreate. */
601 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
603 bridge_create(br_cfg);
606 bridge_create(br_cfg);
609 shash_destroy(&old_br);
610 shash_destroy(&new_br);
614 bridge_configure_ssl(ovs_cfg->ssl);
617 /* Reconfigure all bridges. */
618 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
619 bridge_reconfigure_one(ovs_cfg, br);
622 /* Add and delete ports on all datapaths.
624 * The kernel will reject any attempt to add a given port to a datapath if
625 * that port already belongs to a different datapath, so we must do all
626 * port deletions before any port additions. */
627 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
628 struct odp_port *dpif_ports;
630 struct shash want_ifaces;
632 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
633 bridge_get_all_ifaces(br, &want_ifaces);
634 for (i = 0; i < n_dpif_ports; i++) {
635 const struct odp_port *p = &dpif_ports[i];
636 if (!shash_find(&want_ifaces, p->devname)
637 && strcmp(p->devname, br->name)) {
638 int retval = dpif_port_del(br->dpif, p->port);
640 VLOG_ERR("failed to remove %s interface from %s: %s",
641 p->devname, dpif_name(br->dpif),
646 shash_destroy(&want_ifaces);
649 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
650 struct odp_port *dpif_ports;
652 struct shash cur_ifaces, want_ifaces;
653 struct shash_node *node;
655 /* Get the set of interfaces currently in this datapath. */
656 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
657 shash_init(&cur_ifaces);
658 for (i = 0; i < n_dpif_ports; i++) {
659 const char *name = dpif_ports[i].devname;
660 if (!shash_find(&cur_ifaces, name)) {
661 shash_add(&cur_ifaces, name, NULL);
666 /* Get the set of interfaces we want on this datapath. */
667 bridge_get_all_ifaces(br, &want_ifaces);
669 SHASH_FOR_EACH (node, &want_ifaces) {
670 const char *if_name = node->name;
671 struct iface *iface = node->data;
673 if (shash_find(&cur_ifaces, if_name)) {
674 /* Already exists, just reconfigure it. */
676 reconfigure_iface(iface->cfg, iface);
679 /* Need to add to datapath. */
683 /* Add to datapath. */
684 internal = iface_is_internal(br, if_name);
685 error = dpif_port_add(br->dpif, if_name,
686 internal ? ODP_PORT_INTERNAL : 0, NULL);
687 if (error == EFBIG) {
688 VLOG_ERR("ran out of valid port numbers on %s",
689 dpif_name(br->dpif));
692 VLOG_ERR("failed to add %s interface to %s: %s",
693 if_name, dpif_name(br->dpif), strerror(error));
697 shash_destroy(&cur_ifaces);
698 shash_destroy(&want_ifaces);
700 sflow_bridge_number = 0;
701 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
704 struct iface *local_iface;
705 struct iface *hw_addr_iface;
708 bridge_fetch_dp_ifaces(br);
710 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
711 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
713 /* Pick local port hardware address, datapath ID. */
714 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
715 local_iface = bridge_get_local_iface(br);
717 int error = netdev_set_etheraddr(local_iface->netdev, ea);
719 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
720 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
721 "Ethernet address: %s",
722 br->name, strerror(error));
726 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
727 ofproto_set_datapath_id(br->ofproto, dpid);
729 dpid_string = xasprintf("%012"PRIx64, dpid);
730 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
733 /* Set NetFlow configuration on this bridge. */
734 if (br->cfg->netflow) {
735 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
736 struct netflow_options opts;
738 memset(&opts, 0, sizeof opts);
740 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
741 if (nf_cfg->engine_type) {
742 opts.engine_type = *nf_cfg->engine_type;
744 if (nf_cfg->engine_id) {
745 opts.engine_id = *nf_cfg->engine_id;
748 opts.active_timeout = nf_cfg->active_timeout;
749 if (!opts.active_timeout) {
750 opts.active_timeout = -1;
751 } else if (opts.active_timeout < 0) {
752 VLOG_WARN("bridge %s: active timeout interval set to negative "
753 "value, using default instead (%d seconds)", br->name,
754 NF_ACTIVE_TIMEOUT_DEFAULT);
755 opts.active_timeout = -1;
758 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
759 if (opts.add_id_to_iface) {
760 if (opts.engine_id > 0x7f) {
761 VLOG_WARN("bridge %s: netflow port mangling may conflict "
762 "with another vswitch, choose an engine id less "
763 "than 128", br->name);
765 if (br->n_ports > 508) {
766 VLOG_WARN("bridge %s: netflow port mangling will conflict "
767 "with another port when more than 508 ports are "
772 opts.collectors.n = nf_cfg->n_targets;
773 opts.collectors.names = nf_cfg->targets;
774 if (ofproto_set_netflow(br->ofproto, &opts)) {
775 VLOG_ERR("bridge %s: problem setting netflow collectors",
779 ofproto_set_netflow(br->ofproto, NULL);
782 /* Set sFlow configuration on this bridge. */
783 if (br->cfg->sflow) {
784 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
785 struct ovsrec_controller **controllers;
786 struct ofproto_sflow_options oso;
787 size_t n_controllers;
790 memset(&oso, 0, sizeof oso);
792 oso.targets.n = sflow_cfg->n_targets;
793 oso.targets.names = sflow_cfg->targets;
795 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
796 if (sflow_cfg->sampling) {
797 oso.sampling_rate = *sflow_cfg->sampling;
800 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
801 if (sflow_cfg->polling) {
802 oso.polling_interval = *sflow_cfg->polling;
805 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
806 if (sflow_cfg->header) {
807 oso.header_len = *sflow_cfg->header;
810 oso.sub_id = sflow_bridge_number++;
811 oso.agent_device = sflow_cfg->agent;
813 oso.control_ip = NULL;
814 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
815 for (i = 0; i < n_controllers; i++) {
816 if (controllers[i]->local_ip) {
817 oso.control_ip = controllers[i]->local_ip;
821 ofproto_set_sflow(br->ofproto, &oso);
823 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
825 ofproto_set_sflow(br->ofproto, NULL);
828 /* Update the controller and related settings. It would be more
829 * straightforward to call this from bridge_reconfigure_one(), but we
830 * can't do it there for two reasons. First, and most importantly, at
831 * that point we don't know the dp_ifidx of any interfaces that have
832 * been added to the bridge (because we haven't actually added them to
833 * the datapath). Second, at that point we haven't set the datapath ID
834 * yet; when a controller is configured, resetting the datapath ID will
835 * immediately disconnect from the controller, so it's better to set
836 * the datapath ID before the controller. */
837 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
839 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
840 for (i = 0; i < br->n_ports; i++) {
841 struct port *port = br->ports[i];
843 port_update_vlan_compat(port);
844 port_update_bonding(port);
847 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
848 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
851 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
853 ovsdb_idl_txn_commit(txn);
854 ovsdb_idl_txn_destroy(txn); /* XXX */
860 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
864 for (i = 0; i < n; i++) {
865 if (!strcmp(keys[i], key)) {
873 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
875 return get_ovsrec_key_value(key,
876 br_cfg->key_other_config,
877 br_cfg->value_other_config,
878 br_cfg->n_other_config);
882 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
883 struct iface **hw_addr_iface)
889 *hw_addr_iface = NULL;
891 /* Did the user request a particular MAC? */
892 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
893 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
894 if (eth_addr_is_multicast(ea)) {
895 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
896 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
897 } else if (eth_addr_is_zero(ea)) {
898 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
904 /* Otherwise choose the minimum non-local MAC address among all of the
906 memset(ea, 0xff, sizeof ea);
907 for (i = 0; i < br->n_ports; i++) {
908 struct port *port = br->ports[i];
909 uint8_t iface_ea[ETH_ADDR_LEN];
912 /* Mirror output ports don't participate. */
913 if (port->is_mirror_output_port) {
917 /* Choose the MAC address to represent the port. */
918 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
919 /* Find the interface with this Ethernet address (if any) so that
920 * we can provide the correct devname to the caller. */
922 for (j = 0; j < port->n_ifaces; j++) {
923 struct iface *candidate = port->ifaces[j];
924 uint8_t candidate_ea[ETH_ADDR_LEN];
925 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
926 && eth_addr_equals(iface_ea, candidate_ea)) {
931 /* Choose the interface whose MAC address will represent the port.
932 * The Linux kernel bonding code always chooses the MAC address of
933 * the first slave added to a bond, and the Fedora networking
934 * scripts always add slaves to a bond in alphabetical order, so
935 * for compatibility we choose the interface with the name that is
936 * first in alphabetical order. */
937 iface = port->ifaces[0];
938 for (j = 1; j < port->n_ifaces; j++) {
939 struct iface *candidate = port->ifaces[j];
940 if (strcmp(candidate->name, iface->name) < 0) {
945 /* The local port doesn't count (since we're trying to choose its
946 * MAC address anyway). */
947 if (iface->dp_ifidx == ODPP_LOCAL) {
952 error = netdev_get_etheraddr(iface->netdev, iface_ea);
954 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
955 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
956 iface->name, strerror(error));
961 /* Compare against our current choice. */
962 if (!eth_addr_is_multicast(iface_ea) &&
963 !eth_addr_is_local(iface_ea) &&
964 !eth_addr_is_reserved(iface_ea) &&
965 !eth_addr_is_zero(iface_ea) &&
966 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
968 memcpy(ea, iface_ea, ETH_ADDR_LEN);
969 *hw_addr_iface = iface;
972 if (eth_addr_is_multicast(ea)) {
973 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
974 *hw_addr_iface = NULL;
975 VLOG_WARN("bridge %s: using default bridge Ethernet "
976 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
978 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
979 br->name, ETH_ADDR_ARGS(ea));
983 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
984 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
985 * an interface on 'br', then that interface must be passed in as
986 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
987 * 'hw_addr_iface' must be passed in as a null pointer. */
989 bridge_pick_datapath_id(struct bridge *br,
990 const uint8_t bridge_ea[ETH_ADDR_LEN],
991 struct iface *hw_addr_iface)
994 * The procedure for choosing a bridge MAC address will, in the most
995 * ordinary case, also choose a unique MAC that we can use as a datapath
996 * ID. In some special cases, though, multiple bridges will end up with
997 * the same MAC address. This is OK for the bridges, but it will confuse
998 * the OpenFlow controller, because each datapath needs a unique datapath
1001 * Datapath IDs must be unique. It is also very desirable that they be
1002 * stable from one run to the next, so that policy set on a datapath
1005 const char *datapath_id;
1008 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1009 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1013 if (hw_addr_iface) {
1015 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1017 * A bridge whose MAC address is taken from a VLAN network device
1018 * (that is, a network device created with vconfig(8) or similar
1019 * tool) will have the same MAC address as a bridge on the VLAN
1020 * device's physical network device.
1022 * Handle this case by hashing the physical network device MAC
1023 * along with the VLAN identifier.
1025 uint8_t buf[ETH_ADDR_LEN + 2];
1026 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1027 buf[ETH_ADDR_LEN] = vlan >> 8;
1028 buf[ETH_ADDR_LEN + 1] = vlan;
1029 return dpid_from_hash(buf, sizeof buf);
1032 * Assume that this bridge's MAC address is unique, since it
1033 * doesn't fit any of the cases we handle specially.
1038 * A purely internal bridge, that is, one that has no non-virtual
1039 * network devices on it at all, is more difficult because it has no
1040 * natural unique identifier at all.
1042 * When the host is a XenServer, we handle this case by hashing the
1043 * host's UUID with the name of the bridge. Names of bridges are
1044 * persistent across XenServer reboots, although they can be reused if
1045 * an internal network is destroyed and then a new one is later
1046 * created, so this is fairly effective.
1048 * When the host is not a XenServer, we punt by using a random MAC
1049 * address on each run.
1051 const char *host_uuid = xenserver_get_host_uuid();
1053 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1054 dpid = dpid_from_hash(combined, strlen(combined));
1060 return eth_addr_to_uint64(bridge_ea);
1064 dpid_from_hash(const void *data, size_t n)
1066 uint8_t hash[SHA1_DIGEST_SIZE];
1068 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1069 sha1_bytes(data, n, hash);
1070 eth_addr_mark_random(hash);
1071 return eth_addr_to_uint64(hash);
1077 struct bridge *br, *next;
1081 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1082 int error = bridge_run_one(br);
1084 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1085 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1086 "forcing reconfiguration", br->name);
1100 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1101 ofproto_wait(br->ofproto);
1102 if (ofproto_has_controller(br->ofproto)) {
1106 mac_learning_wait(br->ml);
1111 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1112 * configuration changes. */
1114 bridge_flush(struct bridge *br)
1116 COVERAGE_INC(bridge_flush);
1118 mac_learning_flush(br->ml);
1121 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1122 * such interface. */
1123 static struct iface *
1124 bridge_get_local_iface(struct bridge *br)
1128 for (i = 0; i < br->n_ports; i++) {
1129 struct port *port = br->ports[i];
1130 for (j = 0; j < port->n_ifaces; j++) {
1131 struct iface *iface = port->ifaces[j];
1132 if (iface->dp_ifidx == ODPP_LOCAL) {
1141 /* Bridge unixctl user interface functions. */
1143 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1144 const char *args, void *aux OVS_UNUSED)
1146 struct ds ds = DS_EMPTY_INITIALIZER;
1147 const struct bridge *br;
1148 const struct mac_entry *e;
1150 br = bridge_lookup(args);
1152 unixctl_command_reply(conn, 501, "no such bridge");
1156 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1157 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1158 if (e->port < 0 || e->port >= br->n_ports) {
1161 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1162 br->ports[e->port]->ifaces[0]->dp_ifidx,
1163 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1165 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1169 /* Bridge reconfiguration functions. */
1170 static struct bridge *
1171 bridge_create(const struct ovsrec_bridge *br_cfg)
1176 assert(!bridge_lookup(br_cfg->name));
1177 br = xzalloc(sizeof *br);
1179 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1185 dpif_flow_flush(br->dpif);
1187 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1190 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1192 dpif_delete(br->dpif);
1193 dpif_close(br->dpif);
1198 br->name = xstrdup(br_cfg->name);
1200 br->ml = mac_learning_create();
1201 eth_addr_nicira_random(br->default_ea);
1203 port_array_init(&br->ifaces);
1205 shash_init(&br->port_by_name);
1206 shash_init(&br->iface_by_name);
1210 list_push_back(&all_bridges, &br->node);
1212 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1218 bridge_destroy(struct bridge *br)
1223 while (br->n_ports > 0) {
1224 port_destroy(br->ports[br->n_ports - 1]);
1226 list_remove(&br->node);
1227 error = dpif_delete(br->dpif);
1228 if (error && error != ENOENT) {
1229 VLOG_ERR("failed to delete %s: %s",
1230 dpif_name(br->dpif), strerror(error));
1232 dpif_close(br->dpif);
1233 ofproto_destroy(br->ofproto);
1234 mac_learning_destroy(br->ml);
1235 port_array_destroy(&br->ifaces);
1236 shash_destroy(&br->port_by_name);
1237 shash_destroy(&br->iface_by_name);
1244 static struct bridge *
1245 bridge_lookup(const char *name)
1249 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1250 if (!strcmp(br->name, name)) {
1258 bridge_exists(const char *name)
1260 return bridge_lookup(name) ? true : false;
1264 bridge_get_datapathid(const char *name)
1266 struct bridge *br = bridge_lookup(name);
1267 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1270 /* Handle requests for a listing of all flows known by the OpenFlow
1271 * stack, including those normally hidden. */
1273 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1274 const char *args, void *aux OVS_UNUSED)
1279 br = bridge_lookup(args);
1281 unixctl_command_reply(conn, 501, "Unknown bridge");
1286 ofproto_get_all_flows(br->ofproto, &results);
1288 unixctl_command_reply(conn, 200, ds_cstr(&results));
1289 ds_destroy(&results);
1293 bridge_run_one(struct bridge *br)
1297 error = ofproto_run1(br->ofproto);
1302 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1305 error = ofproto_run2(br->ofproto, br->flush);
1312 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1313 const struct bridge *br,
1314 struct ovsrec_controller ***controllersp)
1316 struct ovsrec_controller **controllers;
1317 size_t n_controllers;
1319 if (br->cfg->n_controller) {
1320 controllers = br->cfg->controller;
1321 n_controllers = br->cfg->n_controller;
1323 controllers = ovs_cfg->controller;
1324 n_controllers = ovs_cfg->n_controller;
1327 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1333 *controllersp = controllers;
1335 return n_controllers;
1339 bridge_update_desc(struct bridge *br OVS_UNUSED)
1342 bool changed = false;
1345 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1346 if (desc != br->mfr_desc) {
1349 br->mfr_desc = xstrdup(desc);
1351 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1356 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1357 if (desc != br->hw_desc) {
1360 br->hw_desc = xstrdup(desc);
1362 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1367 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1368 if (desc != br->sw_desc) {
1371 br->sw_desc = xstrdup(desc);
1373 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1378 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1379 if (desc != br->serial_desc) {
1380 free(br->serial_desc);
1382 br->serial_desc = xstrdup(desc);
1384 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1389 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1390 if (desc != br->dp_desc) {
1393 br->dp_desc = xstrdup(desc);
1395 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1401 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1402 br->sw_desc, br->serial_desc, br->dp_desc);
1408 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1411 struct shash old_ports, new_ports;
1412 struct svec listeners, old_listeners;
1413 struct svec snoops, old_snoops;
1414 struct shash_node *node;
1417 /* Collect old ports. */
1418 shash_init(&old_ports);
1419 for (i = 0; i < br->n_ports; i++) {
1420 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1423 /* Collect new ports. */
1424 shash_init(&new_ports);
1425 for (i = 0; i < br->cfg->n_ports; i++) {
1426 const char *name = br->cfg->ports[i]->name;
1427 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1428 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1433 /* If we have a controller, then we need a local port. Complain if the
1434 * user didn't specify one.
1436 * XXX perhaps we should synthesize a port ourselves in this case. */
1437 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1438 char local_name[IF_NAMESIZE];
1441 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1442 local_name, sizeof local_name);
1443 if (!error && !shash_find(&new_ports, local_name)) {
1444 VLOG_WARN("bridge %s: controller specified but no local port "
1445 "(port named %s) defined",
1446 br->name, local_name);
1450 /* Get rid of deleted ports.
1451 * Get rid of deleted interfaces on ports that still exist. */
1452 SHASH_FOR_EACH (node, &old_ports) {
1453 struct port *port = node->data;
1454 const struct ovsrec_port *port_cfg;
1456 port_cfg = shash_find_data(&new_ports, node->name);
1460 port_del_ifaces(port, port_cfg);
1464 /* Create new ports.
1465 * Add new interfaces to existing ports.
1466 * Reconfigure existing ports. */
1467 SHASH_FOR_EACH (node, &new_ports) {
1468 struct port *port = shash_find_data(&old_ports, node->name);
1470 port = port_create(br, node->name);
1473 port_reconfigure(port, node->data);
1474 if (!port->n_ifaces) {
1475 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1476 br->name, port->name);
1480 shash_destroy(&old_ports);
1481 shash_destroy(&new_ports);
1483 /* Delete all flows if we're switching from connected to standalone or vice
1484 * versa. (XXX Should we delete all flows if we are switching from one
1485 * controller to another?) */
1488 /* Configure OpenFlow management listeners. */
1489 svec_init(&listeners);
1490 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1492 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1493 ovs_rundir, br->name));
1494 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1495 svec_clear(&listeners);
1497 svec_sort_unique(&listeners);
1499 svec_init(&old_listeners);
1500 ofproto_get_listeners(br->ofproto, &old_listeners);
1501 svec_sort_unique(&old_listeners);
1503 if (!svec_equal(&listeners, &old_listeners)) {
1504 ofproto_set_listeners(br->ofproto, &listeners);
1506 svec_destroy(&listeners);
1507 svec_destroy(&old_listeners);
1509 /* Configure OpenFlow controller connection snooping. */
1511 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1513 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1514 ovs_rundir, br->name));
1515 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1516 svec_clear(&snoops);
1518 svec_sort_unique(&snoops);
1520 svec_init(&old_snoops);
1521 ofproto_get_snoops(br->ofproto, &old_snoops);
1522 svec_sort_unique(&old_snoops);
1524 if (!svec_equal(&snoops, &old_snoops)) {
1525 ofproto_set_snoops(br->ofproto, &snoops);
1527 svec_destroy(&snoops);
1528 svec_destroy(&old_snoops);
1530 /* Default listener. */
1531 svec_init(&listeners);
1532 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1533 ovs_rundir, br->name));
1534 svec_init(&old_listeners);
1535 ofproto_get_listeners(br->ofproto, &old_listeners);
1536 if (!svec_equal(&listeners, &old_listeners)) {
1537 ofproto_set_listeners(br->ofproto, &listeners);
1539 svec_destroy(&listeners);
1540 svec_destroy(&old_listeners);
1542 /* Default snoop. */
1544 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1545 ovs_rundir, br->name));
1546 svec_init(&old_snoops);
1547 ofproto_get_snoops(br->ofproto, &old_snoops);
1548 if (!svec_equal(&snoops, &old_snoops)) {
1549 ofproto_set_snoops(br->ofproto, &snoops);
1551 svec_destroy(&snoops);
1552 svec_destroy(&old_snoops);
1555 mirror_reconfigure(br);
1557 bridge_update_desc(br);
1561 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1563 const struct sockaddr_in *managers,
1566 struct ovsrec_controller **controllers;
1567 size_t n_controllers;
1569 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1571 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1572 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1573 ofproto_flush_flows(br->ofproto);
1576 if (!n_controllers) {
1577 union ofp_action action;
1580 /* Clear out controllers. */
1581 ofproto_set_controllers(br->ofproto, NULL, 0);
1583 /* Set up a flow that matches every packet and directs them to
1584 * OFPP_NORMAL (which goes to us). */
1585 memset(&action, 0, sizeof action);
1586 action.type = htons(OFPAT_OUTPUT);
1587 action.output.len = htons(sizeof action);
1588 action.output.port = htons(OFPP_NORMAL);
1589 memset(&flow, 0, sizeof flow);
1590 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1592 struct ofproto_controller *ocs;
1595 ocs = xmalloc(n_controllers * sizeof *ocs);
1596 for (i = 0; i < n_controllers; i++) {
1597 struct ovsrec_controller *c = controllers[i];
1598 struct ofproto_controller *oc = &ocs[i];
1600 if (strcmp(c->target, "discover")) {
1601 struct iface *local_iface;
1604 local_iface = bridge_get_local_iface(br);
1605 if (local_iface && c->local_ip
1606 && inet_aton(c->local_ip, &ip)) {
1607 struct netdev *netdev = local_iface->netdev;
1608 struct in_addr mask, gateway;
1610 if (!c->local_netmask
1611 || !inet_aton(c->local_netmask, &mask)) {
1614 if (!c->local_gateway
1615 || !inet_aton(c->local_gateway, &gateway)) {
1619 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1621 mask.s_addr = guess_netmask(ip.s_addr);
1623 if (!netdev_set_in4(netdev, ip, mask)) {
1624 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1626 br->name, IP_ARGS(&ip.s_addr),
1627 IP_ARGS(&mask.s_addr));
1630 if (gateway.s_addr) {
1631 if (!netdev_add_router(netdev, gateway)) {
1632 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1633 br->name, IP_ARGS(&gateway.s_addr));
1639 oc->target = c->target;
1640 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1641 oc->probe_interval = (c->inactivity_probe
1642 ? *c->inactivity_probe / 1000 : 5);
1643 oc->fail = (!c->fail_mode
1644 || !strcmp(c->fail_mode, "standalone")
1645 || !strcmp(c->fail_mode, "open")
1646 ? OFPROTO_FAIL_STANDALONE
1647 : OFPROTO_FAIL_SECURE);
1648 oc->band = (!c->connection_mode
1649 || !strcmp(c->connection_mode, "in-band")
1651 : OFPROTO_OUT_OF_BAND);
1652 oc->accept_re = c->discover_accept_regex;
1653 oc->update_resolv_conf = c->discover_update_resolv_conf;
1654 oc->rate_limit = (c->controller_rate_limit
1655 ? *c->controller_rate_limit : 0);
1656 oc->burst_limit = (c->controller_burst_limit
1657 ? *c->controller_burst_limit : 0);
1659 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1665 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1670 for (i = 0; i < br->n_ports; i++) {
1671 struct port *port = br->ports[i];
1672 for (j = 0; j < port->n_ifaces; j++) {
1673 struct iface *iface = port->ifaces[j];
1674 shash_add_once(ifaces, iface->name, iface);
1676 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1677 shash_add_once(ifaces, port->name, NULL);
1682 /* For robustness, in case the administrator moves around datapath ports behind
1683 * our back, we re-check all the datapath port numbers here.
1685 * This function will set the 'dp_ifidx' members of interfaces that have
1686 * disappeared to -1, so only call this function from a context where those
1687 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1688 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1689 * datapath, which doesn't support UINT16_MAX+1 ports. */
1691 bridge_fetch_dp_ifaces(struct bridge *br)
1693 struct odp_port *dpif_ports;
1694 size_t n_dpif_ports;
1697 /* Reset all interface numbers. */
1698 for (i = 0; i < br->n_ports; i++) {
1699 struct port *port = br->ports[i];
1700 for (j = 0; j < port->n_ifaces; j++) {
1701 struct iface *iface = port->ifaces[j];
1702 iface->dp_ifidx = -1;
1705 port_array_clear(&br->ifaces);
1707 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1708 for (i = 0; i < n_dpif_ports; i++) {
1709 struct odp_port *p = &dpif_ports[i];
1710 struct iface *iface = iface_lookup(br, p->devname);
1712 if (iface->dp_ifidx >= 0) {
1713 VLOG_WARN("%s reported interface %s twice",
1714 dpif_name(br->dpif), p->devname);
1715 } else if (iface_from_dp_ifidx(br, p->port)) {
1716 VLOG_WARN("%s reported interface %"PRIu16" twice",
1717 dpif_name(br->dpif), p->port);
1719 port_array_set(&br->ifaces, p->port, iface);
1720 iface->dp_ifidx = p->port;
1724 int64_t ofport = (iface->dp_ifidx >= 0
1725 ? odp_port_to_ofp_port(iface->dp_ifidx)
1727 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1734 /* Bridge packet processing functions. */
1737 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1739 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1742 static struct bond_entry *
1743 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1745 return &port->bond_hash[bond_hash(mac)];
1749 bond_choose_iface(const struct port *port)
1751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1752 size_t i, best_down_slave = -1;
1753 long long next_delay_expiration = LLONG_MAX;
1755 for (i = 0; i < port->n_ifaces; i++) {
1756 struct iface *iface = port->ifaces[i];
1758 if (iface->enabled) {
1760 } else if (iface->delay_expires < next_delay_expiration) {
1761 best_down_slave = i;
1762 next_delay_expiration = iface->delay_expires;
1766 if (best_down_slave != -1) {
1767 struct iface *iface = port->ifaces[best_down_slave];
1769 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1770 "since no other interface is up", iface->name,
1771 iface->delay_expires - time_msec());
1772 bond_enable_slave(iface, true);
1775 return best_down_slave;
1779 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1780 uint16_t *dp_ifidx, tag_type *tags)
1782 struct iface *iface;
1784 assert(port->n_ifaces);
1785 if (port->n_ifaces == 1) {
1786 iface = port->ifaces[0];
1788 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1789 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1790 || !port->ifaces[e->iface_idx]->enabled) {
1791 /* XXX select interface properly. The current interface selection
1792 * is only good for testing the rebalancing code. */
1793 e->iface_idx = bond_choose_iface(port);
1794 if (e->iface_idx < 0) {
1795 *tags |= port->no_ifaces_tag;
1798 e->iface_tag = tag_create_random();
1799 ((struct port *) port)->bond_compat_is_stale = true;
1801 *tags |= e->iface_tag;
1802 iface = port->ifaces[e->iface_idx];
1804 *dp_ifidx = iface->dp_ifidx;
1805 *tags |= iface->tag; /* Currently only used for bonding. */
1810 bond_link_status_update(struct iface *iface, bool carrier)
1812 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1813 struct port *port = iface->port;
1815 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1816 /* Nothing to do. */
1819 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1820 iface->name, carrier ? "detected" : "dropped");
1821 if (carrier == iface->enabled) {
1822 iface->delay_expires = LLONG_MAX;
1823 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1824 iface->name, carrier ? "disabled" : "enabled");
1825 } else if (carrier && port->active_iface < 0) {
1826 bond_enable_slave(iface, true);
1827 if (port->updelay) {
1828 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1829 "other interface is up", iface->name, port->updelay);
1832 int delay = carrier ? port->updelay : port->downdelay;
1833 iface->delay_expires = time_msec() + delay;
1836 "interface %s: will be %s if it stays %s for %d ms",
1838 carrier ? "enabled" : "disabled",
1839 carrier ? "up" : "down",
1846 bond_choose_active_iface(struct port *port)
1848 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1850 port->active_iface = bond_choose_iface(port);
1851 port->active_iface_tag = tag_create_random();
1852 if (port->active_iface >= 0) {
1853 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1854 port->name, port->ifaces[port->active_iface]->name);
1856 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1862 bond_enable_slave(struct iface *iface, bool enable)
1864 struct port *port = iface->port;
1865 struct bridge *br = port->bridge;
1867 /* This acts as a recursion check. If the act of disabling a slave
1868 * causes a different slave to be enabled, the flag will allow us to
1869 * skip redundant work when we reenter this function. It must be
1870 * cleared on exit to keep things safe with multiple bonds. */
1871 static bool moving_active_iface = false;
1873 iface->delay_expires = LLONG_MAX;
1874 if (enable == iface->enabled) {
1878 iface->enabled = enable;
1879 if (!iface->enabled) {
1880 VLOG_WARN("interface %s: disabled", iface->name);
1881 ofproto_revalidate(br->ofproto, iface->tag);
1882 if (iface->port_ifidx == port->active_iface) {
1883 ofproto_revalidate(br->ofproto,
1884 port->active_iface_tag);
1886 /* Disabling a slave can lead to another slave being immediately
1887 * enabled if there will be no active slaves but one is waiting
1888 * on an updelay. In this case we do not need to run most of the
1889 * code for the newly enabled slave since there was no period
1890 * without an active slave and it is redundant with the disabling
1892 moving_active_iface = true;
1893 bond_choose_active_iface(port);
1895 bond_send_learning_packets(port);
1897 VLOG_WARN("interface %s: enabled", iface->name);
1898 if (port->active_iface < 0 && !moving_active_iface) {
1899 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1900 bond_choose_active_iface(port);
1901 bond_send_learning_packets(port);
1903 iface->tag = tag_create_random();
1906 moving_active_iface = false;
1907 port->bond_compat_is_stale = true;
1910 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1911 * bond interface. */
1913 bond_update_fake_iface_stats(struct port *port)
1915 struct netdev_stats bond_stats;
1916 struct netdev *bond_dev;
1919 memset(&bond_stats, 0, sizeof bond_stats);
1921 for (i = 0; i < port->n_ifaces; i++) {
1922 struct netdev_stats slave_stats;
1924 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1925 /* XXX: We swap the stats here because they are swapped back when
1926 * reported by the internal device. The reason for this is
1927 * internal devices normally represent packets going into the system
1928 * but when used as fake bond device they represent packets leaving
1929 * the system. We really should do this in the internal device
1930 * itself because changing it here reverses the counts from the
1931 * perspective of the switch. However, the internal device doesn't
1932 * know what type of device it represents so we have to do it here
1934 bond_stats.tx_packets += slave_stats.rx_packets;
1935 bond_stats.tx_bytes += slave_stats.rx_bytes;
1936 bond_stats.rx_packets += slave_stats.tx_packets;
1937 bond_stats.rx_bytes += slave_stats.tx_bytes;
1941 if (!netdev_open_default(port->name, &bond_dev)) {
1942 netdev_set_stats(bond_dev, &bond_stats);
1943 netdev_close(bond_dev);
1948 bond_run(struct bridge *br)
1952 for (i = 0; i < br->n_ports; i++) {
1953 struct port *port = br->ports[i];
1955 if (port->n_ifaces >= 2) {
1956 for (j = 0; j < port->n_ifaces; j++) {
1957 struct iface *iface = port->ifaces[j];
1958 if (time_msec() >= iface->delay_expires) {
1959 bond_enable_slave(iface, !iface->enabled);
1963 if (port->bond_fake_iface
1964 && time_msec() >= port->bond_next_fake_iface_update) {
1965 bond_update_fake_iface_stats(port);
1966 port->bond_next_fake_iface_update = time_msec() + 1000;
1970 if (port->bond_compat_is_stale) {
1971 port->bond_compat_is_stale = false;
1972 port_update_bond_compat(port);
1978 bond_wait(struct bridge *br)
1982 for (i = 0; i < br->n_ports; i++) {
1983 struct port *port = br->ports[i];
1984 if (port->n_ifaces < 2) {
1987 for (j = 0; j < port->n_ifaces; j++) {
1988 struct iface *iface = port->ifaces[j];
1989 if (iface->delay_expires != LLONG_MAX) {
1990 poll_timer_wait_until(iface->delay_expires);
1993 if (port->bond_fake_iface) {
1994 poll_timer_wait_until(port->bond_next_fake_iface_update);
2000 set_dst(struct dst *p, const flow_t *flow,
2001 const struct port *in_port, const struct port *out_port,
2004 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2005 : in_port->vlan >= 0 ? in_port->vlan
2006 : ntohs(flow->dl_vlan));
2007 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2011 swap_dst(struct dst *p, struct dst *q)
2013 struct dst tmp = *p;
2018 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2019 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2020 * that we push to the datapath. We could in fact fully sort the array by
2021 * vlan, but in most cases there are at most two different vlan tags so that's
2022 * possibly overkill.) */
2024 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2026 struct dst *first = dsts;
2027 struct dst *last = dsts + n_dsts;
2029 while (first != last) {
2031 * - All dsts < first have vlan == 'vlan'.
2032 * - All dsts >= last have vlan != 'vlan'.
2033 * - first < last. */
2034 while (first->vlan == vlan) {
2035 if (++first == last) {
2040 /* Same invariants, plus one additional:
2041 * - first->vlan != vlan.
2043 while (last[-1].vlan != vlan) {
2044 if (--last == first) {
2049 /* Same invariants, plus one additional:
2050 * - last[-1].vlan == vlan.*/
2051 swap_dst(first++, --last);
2056 mirror_mask_ffs(mirror_mask_t mask)
2058 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2063 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2064 const struct dst *test)
2067 for (i = 0; i < n_dsts; i++) {
2068 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2076 port_trunks_vlan(const struct port *port, uint16_t vlan)
2078 return (port->vlan < 0
2079 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2083 port_includes_vlan(const struct port *port, uint16_t vlan)
2085 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2089 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2090 const struct port *in_port, const struct port *out_port,
2091 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2093 mirror_mask_t mirrors = in_port->src_mirrors;
2094 struct dst *dst = dsts;
2097 if (out_port == FLOOD_PORT) {
2098 /* XXX use ODP_FLOOD if no vlans or bonding. */
2099 /* XXX even better, define each VLAN as a datapath port group */
2100 for (i = 0; i < br->n_ports; i++) {
2101 struct port *port = br->ports[i];
2102 if (port != in_port && port_includes_vlan(port, vlan)
2103 && !port->is_mirror_output_port
2104 && set_dst(dst, flow, in_port, port, tags)) {
2105 mirrors |= port->dst_mirrors;
2109 *nf_output_iface = NF_OUT_FLOOD;
2110 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2111 *nf_output_iface = dst->dp_ifidx;
2112 mirrors |= out_port->dst_mirrors;
2117 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2118 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2120 if (set_dst(dst, flow, in_port, m->out_port, tags)
2121 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2125 for (i = 0; i < br->n_ports; i++) {
2126 struct port *port = br->ports[i];
2127 if (port_includes_vlan(port, m->out_vlan)
2128 && set_dst(dst, flow, in_port, port, tags))
2132 if (port->vlan < 0) {
2133 dst->vlan = m->out_vlan;
2135 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2139 /* Use the vlan tag on the original flow instead of
2140 * the one passed in the vlan parameter. This ensures
2141 * that we compare the vlan from before any implicit
2142 * tagging tags place. This is necessary because
2143 * dst->vlan is the final vlan, after removing implicit
2145 flow_vlan = ntohs(flow->dl_vlan);
2146 if (flow_vlan == 0) {
2147 flow_vlan = OFP_VLAN_NONE;
2149 if (port == in_port && dst->vlan == flow_vlan) {
2150 /* Don't send out input port on same VLAN. */
2158 mirrors &= mirrors - 1;
2161 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2165 static void OVS_UNUSED
2166 print_dsts(const struct dst *dsts, size_t n)
2168 for (; n--; dsts++) {
2169 printf(">p%"PRIu16, dsts->dp_ifidx);
2170 if (dsts->vlan != OFP_VLAN_NONE) {
2171 printf("v%"PRIu16, dsts->vlan);
2177 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2178 const struct port *in_port, const struct port *out_port,
2179 tag_type *tags, struct odp_actions *actions,
2180 uint16_t *nf_output_iface)
2182 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2184 const struct dst *p;
2187 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2190 cur_vlan = ntohs(flow->dl_vlan);
2191 for (p = dsts; p < &dsts[n_dsts]; p++) {
2192 union odp_action *a;
2193 if (p->vlan != cur_vlan) {
2194 if (p->vlan == OFP_VLAN_NONE) {
2195 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2197 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2198 a->vlan_vid.vlan_vid = htons(p->vlan);
2202 a = odp_actions_add(actions, ODPAT_OUTPUT);
2203 a->output.port = p->dp_ifidx;
2207 /* Returns the effective vlan of a packet, taking into account both the
2208 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2209 * the packet is untagged and -1 indicates it has an invalid header and
2210 * should be dropped. */
2211 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2212 struct port *in_port, bool have_packet)
2214 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2215 * belongs to VLAN 0, so we should treat both cases identically. (In the
2216 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2217 * presumably to allow a priority to be specified. In the latter case, the
2218 * packet does not have any 802.1Q header.) */
2219 int vlan = ntohs(flow->dl_vlan);
2220 if (vlan == OFP_VLAN_NONE) {
2223 if (in_port->vlan >= 0) {
2225 /* XXX support double tagging? */
2227 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2228 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2229 "packet received on port %s configured with "
2230 "implicit VLAN %"PRIu16,
2231 br->name, ntohs(flow->dl_vlan),
2232 in_port->name, in_port->vlan);
2236 vlan = in_port->vlan;
2238 if (!port_includes_vlan(in_port, vlan)) {
2240 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2241 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2242 "packet received on port %s not configured for "
2244 br->name, vlan, in_port->name, vlan);
2253 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2254 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2255 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2257 is_gratuitous_arp(const flow_t *flow)
2259 return (flow->dl_type == htons(ETH_TYPE_ARP)
2260 && eth_addr_is_broadcast(flow->dl_dst)
2261 && (flow->nw_proto == ARP_OP_REPLY
2262 || (flow->nw_proto == ARP_OP_REQUEST
2263 && flow->nw_src == flow->nw_dst)));
2267 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2268 struct port *in_port)
2270 enum grat_arp_lock_type lock_type;
2273 /* We don't want to learn from gratuitous ARP packets that are reflected
2274 * back over bond slaves so we lock the learning table. */
2275 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2276 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2277 GRAT_ARP_LOCK_CHECK;
2279 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2282 /* The log messages here could actually be useful in debugging,
2283 * so keep the rate limit relatively high. */
2284 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2286 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2287 "on port %s in VLAN %d",
2288 br->name, ETH_ADDR_ARGS(flow->dl_src),
2289 in_port->name, vlan);
2290 ofproto_revalidate(br->ofproto, rev_tag);
2294 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2295 * dropped. Returns true if they may be forwarded, false if they should be
2298 * If 'have_packet' is true, it indicates that the caller is processing a
2299 * received packet. If 'have_packet' is false, then the caller is just
2300 * revalidating an existing flow because configuration has changed. Either
2301 * way, 'have_packet' only affects logging (there is no point in logging errors
2302 * during revalidation).
2304 * Sets '*in_portp' to the input port. This will be a null pointer if
2305 * flow->in_port does not designate a known input port (in which case
2306 * is_admissible() returns false).
2308 * When returning true, sets '*vlanp' to the effective VLAN of the input
2309 * packet, as returned by flow_get_vlan().
2311 * May also add tags to '*tags', although the current implementation only does
2312 * so in one special case.
2315 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2316 tag_type *tags, int *vlanp, struct port **in_portp)
2318 struct iface *in_iface;
2319 struct port *in_port;
2322 /* Find the interface and port structure for the received packet. */
2323 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2325 /* No interface? Something fishy... */
2327 /* Odd. A few possible reasons here:
2329 * - We deleted an interface but there are still a few packets
2330 * queued up from it.
2332 * - Someone externally added an interface (e.g. with "ovs-dpctl
2333 * add-if") that we don't know about.
2335 * - Packet arrived on the local port but the local port is not
2336 * one of our bridge ports.
2338 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2340 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2341 "interface %"PRIu16, br->name, flow->in_port);
2347 *in_portp = in_port = in_iface->port;
2348 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2353 /* Drop frames for reserved multicast addresses. */
2354 if (eth_addr_is_reserved(flow->dl_dst)) {
2358 /* Drop frames on ports reserved for mirroring. */
2359 if (in_port->is_mirror_output_port) {
2361 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2362 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2363 "%s, which is reserved exclusively for mirroring",
2364 br->name, in_port->name);
2369 /* Packets received on bonds need special attention to avoid duplicates. */
2370 if (in_port->n_ifaces > 1) {
2372 bool is_grat_arp_locked;
2374 if (eth_addr_is_multicast(flow->dl_dst)) {
2375 *tags |= in_port->active_iface_tag;
2376 if (in_port->active_iface != in_iface->port_ifidx) {
2377 /* Drop all multicast packets on inactive slaves. */
2382 /* Drop all packets for which we have learned a different input
2383 * port, because we probably sent the packet on one slave and got
2384 * it back on the other. Gratuitous ARP packets are an exception
2385 * to this rule: the host has moved to another switch. The exception
2386 * to the exception is if we locked the learning table to avoid
2387 * reflections on bond slaves. If this is the case, just drop the
2389 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2390 &is_grat_arp_locked);
2391 if (src_idx != -1 && src_idx != in_port->port_idx &&
2392 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2400 /* If the composed actions may be applied to any packet in the given 'flow',
2401 * returns true. Otherwise, the actions should only be applied to 'packet', or
2402 * not at all, if 'packet' was NULL. */
2404 process_flow(struct bridge *br, const flow_t *flow,
2405 const struct ofpbuf *packet, struct odp_actions *actions,
2406 tag_type *tags, uint16_t *nf_output_iface)
2408 struct port *in_port;
2409 struct port *out_port;
2413 /* Check whether we should drop packets in this flow. */
2414 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2419 /* Learn source MAC (but don't try to learn from revalidation). */
2421 update_learning_table(br, flow, vlan, in_port);
2424 /* Determine output port. */
2425 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2427 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2428 out_port = br->ports[out_port_idx];
2429 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2430 /* If we are revalidating but don't have a learning entry then
2431 * eject the flow. Installing a flow that floods packets opens
2432 * up a window of time where we could learn from a packet reflected
2433 * on a bond and blackhole packets before the learning table is
2434 * updated to reflect the correct port. */
2437 out_port = FLOOD_PORT;
2440 /* Don't send packets out their input ports. */
2441 if (in_port == out_port) {
2447 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2454 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2457 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2458 const struct ofp_phy_port *opp,
2461 struct bridge *br = br_;
2462 struct iface *iface;
2465 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2471 if (reason == OFPPR_DELETE) {
2472 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2473 br->name, iface->name);
2474 iface_destroy(iface);
2475 if (!port->n_ifaces) {
2476 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2477 br->name, port->name);
2483 if (port->n_ifaces > 1) {
2484 bool up = !(opp->state & OFPPS_LINK_DOWN);
2485 bond_link_status_update(iface, up);
2486 port_update_bond_compat(port);
2492 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2493 struct odp_actions *actions, tag_type *tags,
2494 uint16_t *nf_output_iface, void *br_)
2496 struct bridge *br = br_;
2498 COVERAGE_INC(bridge_process_flow);
2499 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2503 bridge_account_flow_ofhook_cb(const flow_t *flow,
2504 const union odp_action *actions,
2505 size_t n_actions, unsigned long long int n_bytes,
2508 struct bridge *br = br_;
2509 const union odp_action *a;
2510 struct port *in_port;
2514 /* Feed information from the active flows back into the learning table
2515 * to ensure that table is always in sync with what is actually flowing
2516 * through the datapath. */
2517 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2518 update_learning_table(br, flow, vlan, in_port);
2521 if (!br->has_bonded_ports) {
2525 for (a = actions; a < &actions[n_actions]; a++) {
2526 if (a->type == ODPAT_OUTPUT) {
2527 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2528 if (out_port && out_port->n_ifaces >= 2) {
2529 struct bond_entry *e = lookup_bond_entry(out_port,
2531 e->tx_bytes += n_bytes;
2538 bridge_account_checkpoint_ofhook_cb(void *br_)
2540 struct bridge *br = br_;
2544 if (!br->has_bonded_ports) {
2549 for (i = 0; i < br->n_ports; i++) {
2550 struct port *port = br->ports[i];
2551 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2552 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2553 bond_rebalance_port(port);
2558 static struct ofhooks bridge_ofhooks = {
2559 bridge_port_changed_ofhook_cb,
2560 bridge_normal_ofhook_cb,
2561 bridge_account_flow_ofhook_cb,
2562 bridge_account_checkpoint_ofhook_cb,
2565 /* Bonding functions. */
2567 /* Statistics for a single interface on a bonded port, used for load-based
2568 * bond rebalancing. */
2569 struct slave_balance {
2570 struct iface *iface; /* The interface. */
2571 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2573 /* All the "bond_entry"s that are assigned to this interface, in order of
2574 * increasing tx_bytes. */
2575 struct bond_entry **hashes;
2579 /* Sorts pointers to pointers to bond_entries in ascending order by the
2580 * interface to which they are assigned, and within a single interface in
2581 * ascending order of bytes transmitted. */
2583 compare_bond_entries(const void *a_, const void *b_)
2585 const struct bond_entry *const *ap = a_;
2586 const struct bond_entry *const *bp = b_;
2587 const struct bond_entry *a = *ap;
2588 const struct bond_entry *b = *bp;
2589 if (a->iface_idx != b->iface_idx) {
2590 return a->iface_idx > b->iface_idx ? 1 : -1;
2591 } else if (a->tx_bytes != b->tx_bytes) {
2592 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2598 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2599 * *descending* order by number of bytes transmitted. */
2601 compare_slave_balance(const void *a_, const void *b_)
2603 const struct slave_balance *a = a_;
2604 const struct slave_balance *b = b_;
2605 if (a->iface->enabled != b->iface->enabled) {
2606 return a->iface->enabled ? -1 : 1;
2607 } else if (a->tx_bytes != b->tx_bytes) {
2608 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2615 swap_bals(struct slave_balance *a, struct slave_balance *b)
2617 struct slave_balance tmp = *a;
2622 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2623 * given that 'p' (and only 'p') might be in the wrong location.
2625 * This function invalidates 'p', since it might now be in a different memory
2628 resort_bals(struct slave_balance *p,
2629 struct slave_balance bals[], size_t n_bals)
2632 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2633 swap_bals(p, p - 1);
2635 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2636 swap_bals(p, p + 1);
2642 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2644 if (VLOG_IS_DBG_ENABLED()) {
2645 struct ds ds = DS_EMPTY_INITIALIZER;
2646 const struct slave_balance *b;
2648 for (b = bals; b < bals + n_bals; b++) {
2652 ds_put_char(&ds, ',');
2654 ds_put_format(&ds, " %s %"PRIu64"kB",
2655 b->iface->name, b->tx_bytes / 1024);
2657 if (!b->iface->enabled) {
2658 ds_put_cstr(&ds, " (disabled)");
2660 if (b->n_hashes > 0) {
2661 ds_put_cstr(&ds, " (");
2662 for (i = 0; i < b->n_hashes; i++) {
2663 const struct bond_entry *e = b->hashes[i];
2665 ds_put_cstr(&ds, " + ");
2667 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2668 e - port->bond_hash, e->tx_bytes / 1024);
2670 ds_put_cstr(&ds, ")");
2673 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2678 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2680 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2683 struct bond_entry *hash = from->hashes[hash_idx];
2684 struct port *port = from->iface->port;
2685 uint64_t delta = hash->tx_bytes;
2687 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2688 "from %s to %s (now carrying %"PRIu64"kB and "
2689 "%"PRIu64"kB load, respectively)",
2690 port->name, delta / 1024, hash - port->bond_hash,
2691 from->iface->name, to->iface->name,
2692 (from->tx_bytes - delta) / 1024,
2693 (to->tx_bytes + delta) / 1024);
2695 /* Delete element from from->hashes.
2697 * We don't bother to add the element to to->hashes because not only would
2698 * it require more work, the only purpose it would be to allow that hash to
2699 * be migrated to another slave in this rebalancing run, and there is no
2700 * point in doing that. */
2701 if (hash_idx == 0) {
2704 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2705 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2709 /* Shift load away from 'from' to 'to'. */
2710 from->tx_bytes -= delta;
2711 to->tx_bytes += delta;
2713 /* Arrange for flows to be revalidated. */
2714 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2715 hash->iface_idx = to->iface->port_ifidx;
2716 hash->iface_tag = tag_create_random();
2720 bond_rebalance_port(struct port *port)
2722 struct slave_balance bals[DP_MAX_PORTS];
2724 struct bond_entry *hashes[BOND_MASK + 1];
2725 struct slave_balance *b, *from, *to;
2726 struct bond_entry *e;
2729 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2730 * descending order of tx_bytes, so that bals[0] represents the most
2731 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2734 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2735 * array for each slave_balance structure, we sort our local array of
2736 * hashes in order by slave, so that all of the hashes for a given slave
2737 * become contiguous in memory, and then we point each 'hashes' members of
2738 * a slave_balance structure to the start of a contiguous group. */
2739 n_bals = port->n_ifaces;
2740 for (b = bals; b < &bals[n_bals]; b++) {
2741 b->iface = port->ifaces[b - bals];
2746 for (i = 0; i <= BOND_MASK; i++) {
2747 hashes[i] = &port->bond_hash[i];
2749 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2750 for (i = 0; i <= BOND_MASK; i++) {
2752 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2753 b = &bals[e->iface_idx];
2754 b->tx_bytes += e->tx_bytes;
2756 b->hashes = &hashes[i];
2761 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2762 log_bals(bals, n_bals, port);
2764 /* Discard slaves that aren't enabled (which were sorted to the back of the
2765 * array earlier). */
2766 while (!bals[n_bals - 1].iface->enabled) {
2773 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2774 to = &bals[n_bals - 1];
2775 for (from = bals; from < to; ) {
2776 uint64_t overload = from->tx_bytes - to->tx_bytes;
2777 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2778 /* The extra load on 'from' (and all less-loaded slaves), compared
2779 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2780 * it is less than ~1Mbps. No point in rebalancing. */
2782 } else if (from->n_hashes == 1) {
2783 /* 'from' only carries a single MAC hash, so we can't shift any
2784 * load away from it, even though we want to. */
2787 /* 'from' is carrying significantly more load than 'to', and that
2788 * load is split across at least two different hashes. Pick a hash
2789 * to migrate to 'to' (the least-loaded slave), given that doing so
2790 * must decrease the ratio of the load on the two slaves by at
2793 * The sort order we use means that we prefer to shift away the
2794 * smallest hashes instead of the biggest ones. There is little
2795 * reason behind this decision; we could use the opposite sort
2796 * order to shift away big hashes ahead of small ones. */
2800 for (i = 0; i < from->n_hashes; i++) {
2801 double old_ratio, new_ratio;
2802 uint64_t delta = from->hashes[i]->tx_bytes;
2804 if (delta == 0 || from->tx_bytes - delta == 0) {
2805 /* Pointless move. */
2809 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2811 if (to->tx_bytes == 0) {
2812 /* Nothing on the new slave, move it. */
2816 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2817 new_ratio = (double)(from->tx_bytes - delta) /
2818 (to->tx_bytes + delta);
2820 if (new_ratio == 0) {
2821 /* Should already be covered but check to prevent division
2826 if (new_ratio < 1) {
2827 new_ratio = 1 / new_ratio;
2830 if (old_ratio - new_ratio > 0.1) {
2831 /* Would decrease the ratio, move it. */
2835 if (i < from->n_hashes) {
2836 bond_shift_load(from, to, i);
2837 port->bond_compat_is_stale = true;
2839 /* If the result of the migration changed the relative order of
2840 * 'from' and 'to' swap them back to maintain invariants. */
2841 if (order_swapped) {
2842 swap_bals(from, to);
2845 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2846 * point to different slave_balance structures. It is only
2847 * valid to do these two operations in a row at all because we
2848 * know that 'from' will not move past 'to' and vice versa. */
2849 resort_bals(from, bals, n_bals);
2850 resort_bals(to, bals, n_bals);
2857 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2858 * historical data to decay to <1% in 7 rebalancing runs. */
2859 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2865 bond_send_learning_packets(struct port *port)
2867 struct bridge *br = port->bridge;
2868 struct mac_entry *e;
2869 struct ofpbuf packet;
2870 int error, n_packets, n_errors;
2872 if (!port->n_ifaces || port->active_iface < 0) {
2876 ofpbuf_init(&packet, 128);
2877 error = n_packets = n_errors = 0;
2878 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2879 union ofp_action actions[2], *a;
2885 if (e->port == port->port_idx
2886 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2890 /* Compose actions. */
2891 memset(actions, 0, sizeof actions);
2894 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2895 a->vlan_vid.len = htons(sizeof *a);
2896 a->vlan_vid.vlan_vid = htons(e->vlan);
2899 a->output.type = htons(OFPAT_OUTPUT);
2900 a->output.len = htons(sizeof *a);
2901 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2906 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2908 flow_extract(&packet, 0, ODPP_NONE, &flow);
2909 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2916 ofpbuf_uninit(&packet);
2919 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2920 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2921 "packets, last error was: %s",
2922 port->name, n_errors, n_packets, strerror(error));
2924 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2925 port->name, n_packets);
2929 /* Bonding unixctl user interface functions. */
2932 bond_unixctl_list(struct unixctl_conn *conn,
2933 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2935 struct ds ds = DS_EMPTY_INITIALIZER;
2936 const struct bridge *br;
2938 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2940 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2943 for (i = 0; i < br->n_ports; i++) {
2944 const struct port *port = br->ports[i];
2945 if (port->n_ifaces > 1) {
2948 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2949 for (j = 0; j < port->n_ifaces; j++) {
2950 const struct iface *iface = port->ifaces[j];
2952 ds_put_cstr(&ds, ", ");
2954 ds_put_cstr(&ds, iface->name);
2956 ds_put_char(&ds, '\n');
2960 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2964 static struct port *
2965 bond_find(const char *name)
2967 const struct bridge *br;
2969 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2972 for (i = 0; i < br->n_ports; i++) {
2973 struct port *port = br->ports[i];
2974 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2983 bond_unixctl_show(struct unixctl_conn *conn,
2984 const char *args, void *aux OVS_UNUSED)
2986 struct ds ds = DS_EMPTY_INITIALIZER;
2987 const struct port *port;
2990 port = bond_find(args);
2992 unixctl_command_reply(conn, 501, "no such bond");
2996 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2997 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2998 ds_put_format(&ds, "next rebalance: %lld ms\n",
2999 port->bond_next_rebalance - time_msec());
3000 for (j = 0; j < port->n_ifaces; j++) {
3001 const struct iface *iface = port->ifaces[j];
3002 struct bond_entry *be;
3005 ds_put_format(&ds, "slave %s: %s\n",
3006 iface->name, iface->enabled ? "enabled" : "disabled");
3007 if (j == port->active_iface) {
3008 ds_put_cstr(&ds, "\tactive slave\n");
3010 if (iface->delay_expires != LLONG_MAX) {
3011 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3012 iface->enabled ? "downdelay" : "updelay",
3013 iface->delay_expires - time_msec());
3017 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3018 int hash = be - port->bond_hash;
3019 struct mac_entry *me;
3021 if (be->iface_idx != j) {
3025 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3026 hash, be->tx_bytes / 1024);
3029 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3030 &port->bridge->ml->lrus) {
3033 if (bond_hash(me->mac) == hash
3034 && me->port != port->port_idx
3035 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3036 && dp_ifidx == iface->dp_ifidx)
3038 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3039 ETH_ADDR_ARGS(me->mac));
3044 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3049 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3050 void *aux OVS_UNUSED)
3052 char *args = (char *) args_;
3053 char *save_ptr = NULL;
3054 char *bond_s, *hash_s, *slave_s;
3055 uint8_t mac[ETH_ADDR_LEN];
3057 struct iface *iface;
3058 struct bond_entry *entry;
3061 bond_s = strtok_r(args, " ", &save_ptr);
3062 hash_s = strtok_r(NULL, " ", &save_ptr);
3063 slave_s = strtok_r(NULL, " ", &save_ptr);
3065 unixctl_command_reply(conn, 501,
3066 "usage: bond/migrate BOND HASH SLAVE");
3070 port = bond_find(bond_s);
3072 unixctl_command_reply(conn, 501, "no such bond");
3076 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3077 == ETH_ADDR_SCAN_COUNT) {
3078 hash = bond_hash(mac);
3079 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3080 hash = atoi(hash_s) & BOND_MASK;
3082 unixctl_command_reply(conn, 501, "bad hash");
3086 iface = port_lookup_iface(port, slave_s);
3088 unixctl_command_reply(conn, 501, "no such slave");
3092 if (!iface->enabled) {
3093 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3097 entry = &port->bond_hash[hash];
3098 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3099 entry->iface_idx = iface->port_ifidx;
3100 entry->iface_tag = tag_create_random();
3101 port->bond_compat_is_stale = true;
3102 unixctl_command_reply(conn, 200, "migrated");
3106 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3107 void *aux OVS_UNUSED)
3109 char *args = (char *) args_;
3110 char *save_ptr = NULL;
3111 char *bond_s, *slave_s;
3113 struct iface *iface;
3115 bond_s = strtok_r(args, " ", &save_ptr);
3116 slave_s = strtok_r(NULL, " ", &save_ptr);
3118 unixctl_command_reply(conn, 501,
3119 "usage: bond/set-active-slave BOND SLAVE");
3123 port = bond_find(bond_s);
3125 unixctl_command_reply(conn, 501, "no such bond");
3129 iface = port_lookup_iface(port, slave_s);
3131 unixctl_command_reply(conn, 501, "no such slave");
3135 if (!iface->enabled) {
3136 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3140 if (port->active_iface != iface->port_ifidx) {
3141 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3142 port->active_iface = iface->port_ifidx;
3143 port->active_iface_tag = tag_create_random();
3144 VLOG_INFO("port %s: active interface is now %s",
3145 port->name, iface->name);
3146 bond_send_learning_packets(port);
3147 unixctl_command_reply(conn, 200, "done");
3149 unixctl_command_reply(conn, 200, "no change");
3154 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3156 char *args = (char *) args_;
3157 char *save_ptr = NULL;
3158 char *bond_s, *slave_s;
3160 struct iface *iface;
3162 bond_s = strtok_r(args, " ", &save_ptr);
3163 slave_s = strtok_r(NULL, " ", &save_ptr);
3165 unixctl_command_reply(conn, 501,
3166 "usage: bond/enable/disable-slave BOND SLAVE");
3170 port = bond_find(bond_s);
3172 unixctl_command_reply(conn, 501, "no such bond");
3176 iface = port_lookup_iface(port, slave_s);
3178 unixctl_command_reply(conn, 501, "no such slave");
3182 bond_enable_slave(iface, enable);
3183 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3187 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3188 void *aux OVS_UNUSED)
3190 enable_slave(conn, args, true);
3194 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3195 void *aux OVS_UNUSED)
3197 enable_slave(conn, args, false);
3201 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3202 void *aux OVS_UNUSED)
3204 uint8_t mac[ETH_ADDR_LEN];
3208 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3209 == ETH_ADDR_SCAN_COUNT) {
3210 hash = bond_hash(mac);
3212 hash_cstr = xasprintf("%u", hash);
3213 unixctl_command_reply(conn, 200, hash_cstr);
3216 unixctl_command_reply(conn, 501, "invalid mac");
3223 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3224 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3225 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3226 unixctl_command_register("bond/set-active-slave",
3227 bond_unixctl_set_active_slave, NULL);
3228 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3230 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3232 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3235 /* Port functions. */
3237 static struct port *
3238 port_create(struct bridge *br, const char *name)
3242 port = xzalloc(sizeof *port);
3244 port->port_idx = br->n_ports;
3246 port->trunks = NULL;
3247 port->name = xstrdup(name);
3248 port->active_iface = -1;
3250 if (br->n_ports >= br->allocated_ports) {
3251 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3254 br->ports[br->n_ports++] = port;
3255 shash_add_assert(&br->port_by_name, port->name, port);
3257 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3264 get_port_other_config(const struct ovsrec_port *port, const char *key,
3265 const char *default_value)
3267 const char *value = get_ovsrec_key_value(key,
3268 port->key_other_config,
3269 port->value_other_config,
3270 port->n_other_config);
3271 return value ? value : default_value;
3275 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3277 struct shash new_ifaces;
3280 /* Collect list of new interfaces. */
3281 shash_init(&new_ifaces);
3282 for (i = 0; i < cfg->n_interfaces; i++) {
3283 const char *name = cfg->interfaces[i]->name;
3284 shash_add_once(&new_ifaces, name, NULL);
3287 /* Get rid of deleted interfaces. */
3288 for (i = 0; i < port->n_ifaces; ) {
3289 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3290 iface_destroy(port->ifaces[i]);
3296 shash_destroy(&new_ifaces);
3300 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3302 struct shash new_ifaces;
3303 long long int next_rebalance;
3304 unsigned long *trunks;
3310 /* Update settings. */
3311 port->updelay = cfg->bond_updelay;
3312 if (port->updelay < 0) {
3315 port->updelay = cfg->bond_downdelay;
3316 if (port->downdelay < 0) {
3317 port->downdelay = 0;
3319 port->bond_rebalance_interval = atoi(
3320 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3321 if (port->bond_rebalance_interval < 1000) {
3322 port->bond_rebalance_interval = 1000;
3324 next_rebalance = time_msec() + port->bond_rebalance_interval;
3325 if (port->bond_next_rebalance > next_rebalance) {
3326 port->bond_next_rebalance = next_rebalance;
3329 /* Add new interfaces and update 'cfg' member of existing ones. */
3330 shash_init(&new_ifaces);
3331 for (i = 0; i < cfg->n_interfaces; i++) {
3332 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3333 struct iface *iface;
3335 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3336 VLOG_WARN("port %s: %s specified twice as port interface",
3337 port->name, if_cfg->name);
3341 iface = iface_lookup(port->bridge, if_cfg->name);
3343 if (iface->port != port) {
3344 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3346 port->bridge->name, if_cfg->name, iface->port->name);
3349 iface->cfg = if_cfg;
3351 iface_create(port, if_cfg);
3354 shash_destroy(&new_ifaces);
3359 if (port->n_ifaces < 2) {
3361 if (vlan >= 0 && vlan <= 4095) {
3362 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3367 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3368 * they even work as-is. But they have not been tested. */
3369 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3373 if (port->vlan != vlan) {
3375 bridge_flush(port->bridge);
3378 /* Get trunked VLANs. */
3380 if (vlan < 0 && cfg->n_trunks) {
3384 trunks = bitmap_allocate(4096);
3386 for (i = 0; i < cfg->n_trunks; i++) {
3387 int trunk = cfg->trunks[i];
3389 bitmap_set1(trunks, trunk);
3395 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3396 port->name, cfg->n_trunks);
3398 if (n_errors == cfg->n_trunks) {
3399 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3401 bitmap_free(trunks);
3404 } else if (vlan >= 0 && cfg->n_trunks) {
3405 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3409 ? port->trunks != NULL
3410 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3411 bridge_flush(port->bridge);
3413 bitmap_free(port->trunks);
3414 port->trunks = trunks;
3418 port_destroy(struct port *port)
3421 struct bridge *br = port->bridge;
3425 proc_net_compat_update_vlan(port->name, NULL, 0);
3426 proc_net_compat_update_bond(port->name, NULL);
3428 for (i = 0; i < MAX_MIRRORS; i++) {
3429 struct mirror *m = br->mirrors[i];
3430 if (m && m->out_port == port) {
3435 while (port->n_ifaces > 0) {
3436 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3439 shash_find_and_delete_assert(&br->port_by_name, port->name);
3441 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3442 del->port_idx = port->port_idx;
3445 bitmap_free(port->trunks);
3452 static struct port *
3453 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3455 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3456 return iface ? iface->port : NULL;
3459 static struct port *
3460 port_lookup(const struct bridge *br, const char *name)
3462 return shash_find_data(&br->port_by_name, name);
3465 static struct iface *
3466 port_lookup_iface(const struct port *port, const char *name)
3468 struct iface *iface = iface_lookup(port->bridge, name);
3469 return iface && iface->port == port ? iface : NULL;
3473 port_update_bonding(struct port *port)
3475 if (port->n_ifaces < 2) {
3476 /* Not a bonded port. */
3477 if (port->bond_hash) {
3478 free(port->bond_hash);
3479 port->bond_hash = NULL;
3480 port->bond_compat_is_stale = true;
3481 port->bond_fake_iface = false;
3484 if (!port->bond_hash) {
3487 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3488 for (i = 0; i <= BOND_MASK; i++) {
3489 struct bond_entry *e = &port->bond_hash[i];
3493 port->no_ifaces_tag = tag_create_random();
3494 bond_choose_active_iface(port);
3495 port->bond_next_rebalance
3496 = time_msec() + port->bond_rebalance_interval;
3498 if (port->cfg->bond_fake_iface) {
3499 port->bond_next_fake_iface_update = time_msec();
3502 port->bond_compat_is_stale = true;
3503 port->bond_fake_iface = port->cfg->bond_fake_iface;
3508 port_update_bond_compat(struct port *port)
3510 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3511 struct compat_bond bond;
3514 if (port->n_ifaces < 2) {
3515 proc_net_compat_update_bond(port->name, NULL);
3520 bond.updelay = port->updelay;
3521 bond.downdelay = port->downdelay;
3524 bond.hashes = compat_hashes;
3525 if (port->bond_hash) {
3526 const struct bond_entry *e;
3527 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3528 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3529 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3530 cbh->hash = e - port->bond_hash;
3531 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3536 bond.n_slaves = port->n_ifaces;
3537 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3538 for (i = 0; i < port->n_ifaces; i++) {
3539 struct iface *iface = port->ifaces[i];
3540 struct compat_bond_slave *slave = &bond.slaves[i];
3541 slave->name = iface->name;
3543 /* We need to make the same determination as the Linux bonding
3544 * code to determine whether a slave should be consider "up".
3545 * The Linux function bond_miimon_inspect() supports four
3546 * BOND_LINK_* states:
3548 * - BOND_LINK_UP: carrier detected, updelay has passed.
3549 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3550 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3551 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3553 * The function bond_info_show_slave() only considers BOND_LINK_UP
3554 * to be "up" and anything else to be "down".
3556 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3560 netdev_get_etheraddr(iface->netdev, slave->mac);
3563 if (port->bond_fake_iface) {
3564 struct netdev *bond_netdev;
3566 if (!netdev_open_default(port->name, &bond_netdev)) {
3568 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3570 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3572 netdev_close(bond_netdev);
3576 proc_net_compat_update_bond(port->name, &bond);
3581 port_update_vlan_compat(struct port *port)
3583 struct bridge *br = port->bridge;
3584 char *vlandev_name = NULL;
3586 if (port->vlan > 0) {
3587 /* Figure out the name that the VLAN device should actually have, if it
3588 * existed. This takes some work because the VLAN device would not
3589 * have port->name in its name; rather, it would have the trunk port's
3590 * name, and 'port' would be attached to a bridge that also had the
3591 * VLAN device one of its ports. So we need to find a trunk port that
3592 * includes port->vlan.
3594 * There might be more than one candidate. This doesn't happen on
3595 * XenServer, so if it happens we just pick the first choice in
3596 * alphabetical order instead of creating multiple VLAN devices. */
3598 for (i = 0; i < br->n_ports; i++) {
3599 struct port *p = br->ports[i];
3600 if (port_trunks_vlan(p, port->vlan)
3602 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3604 uint8_t ea[ETH_ADDR_LEN];
3605 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3606 if (!eth_addr_is_multicast(ea) &&
3607 !eth_addr_is_reserved(ea) &&
3608 !eth_addr_is_zero(ea)) {
3609 vlandev_name = p->name;
3614 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3617 /* Interface functions. */
3619 static struct iface *
3620 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3622 struct bridge *br = port->bridge;
3623 struct iface *iface;
3624 char *name = if_cfg->name;
3627 iface = xzalloc(sizeof *iface);
3629 iface->port_ifidx = port->n_ifaces;
3630 iface->name = xstrdup(name);
3631 iface->dp_ifidx = -1;
3632 iface->tag = tag_create_random();
3633 iface->delay_expires = LLONG_MAX;
3634 iface->netdev = NULL;
3635 iface->cfg = if_cfg;
3637 shash_add_assert(&br->iface_by_name, iface->name, iface);
3639 /* Attempt to create the network interface in case it doesn't exist yet. */
3640 if (!iface_is_internal(br, iface->name)) {
3641 error = set_up_iface(if_cfg, iface, true);
3643 VLOG_WARN("could not create iface %s: %s", iface->name,
3646 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3653 if (port->n_ifaces >= port->allocated_ifaces) {
3654 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3655 sizeof *port->ifaces);
3657 port->ifaces[port->n_ifaces++] = iface;
3658 if (port->n_ifaces > 1) {
3659 br->has_bonded_ports = true;
3662 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3670 iface_destroy(struct iface *iface)
3673 struct port *port = iface->port;
3674 struct bridge *br = port->bridge;
3675 bool del_active = port->active_iface == iface->port_ifidx;
3678 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3680 if (iface->dp_ifidx >= 0) {
3681 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3684 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3685 del->port_ifidx = iface->port_ifidx;
3687 netdev_close(iface->netdev);
3690 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3691 bond_choose_active_iface(port);
3692 bond_send_learning_packets(port);
3698 bridge_flush(port->bridge);
3702 static struct iface *
3703 iface_lookup(const struct bridge *br, const char *name)
3705 return shash_find_data(&br->iface_by_name, name);
3708 static struct iface *
3709 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3711 return port_array_get(&br->ifaces, dp_ifidx);
3714 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3715 * 'br', that is, an interface that is entirely simulated within the datapath.
3716 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3717 * interfaces are created by setting "iface.<iface>.internal = true".
3719 * In addition, we have a kluge-y feature that creates an internal port with
3720 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3721 * This feature needs to go away in the long term. Until then, this is one
3722 * reason why this function takes a name instead of a struct iface: the fake
3723 * interfaces created this way do not have a struct iface. */
3725 iface_is_internal(const struct bridge *br, const char *if_name)
3727 struct iface *iface;
3730 if (!strcmp(if_name, br->name)) {
3734 iface = iface_lookup(br, if_name);
3735 if (iface && !strcmp(iface->cfg->type, "internal")) {
3739 port = port_lookup(br, if_name);
3740 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3746 /* Set Ethernet address of 'iface', if one is specified in the configuration
3749 iface_set_mac(struct iface *iface)
3751 uint8_t ea[ETH_ADDR_LEN];
3753 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3754 if (eth_addr_is_multicast(ea)) {
3755 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3757 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3758 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3759 iface->name, iface->name);
3761 int error = netdev_set_etheraddr(iface->netdev, ea);
3763 VLOG_ERR("interface %s: setting MAC failed (%s)",
3764 iface->name, strerror(error));
3770 /* Port mirroring. */
3773 mirror_reconfigure(struct bridge *br)
3775 struct shash old_mirrors, new_mirrors;
3776 struct shash_node *node;
3777 unsigned long *rspan_vlans;
3780 /* Collect old mirrors. */
3781 shash_init(&old_mirrors);
3782 for (i = 0; i < MAX_MIRRORS; i++) {
3783 if (br->mirrors[i]) {
3784 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3788 /* Collect new mirrors. */
3789 shash_init(&new_mirrors);
3790 for (i = 0; i < br->cfg->n_mirrors; i++) {
3791 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3792 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3793 VLOG_WARN("bridge %s: %s specified twice as mirror",
3794 br->name, cfg->name);
3798 /* Get rid of deleted mirrors and add new mirrors. */
3799 SHASH_FOR_EACH (node, &old_mirrors) {
3800 if (!shash_find(&new_mirrors, node->name)) {
3801 mirror_destroy(node->data);
3804 SHASH_FOR_EACH (node, &new_mirrors) {
3805 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3807 mirror = mirror_create(br, node->name);
3812 mirror_reconfigure_one(mirror, node->data);
3814 shash_destroy(&old_mirrors);
3815 shash_destroy(&new_mirrors);
3817 /* Update port reserved status. */
3818 for (i = 0; i < br->n_ports; i++) {
3819 br->ports[i]->is_mirror_output_port = false;
3821 for (i = 0; i < MAX_MIRRORS; i++) {
3822 struct mirror *m = br->mirrors[i];
3823 if (m && m->out_port) {
3824 m->out_port->is_mirror_output_port = true;
3828 /* Update flooded vlans (for RSPAN). */
3830 if (br->cfg->n_flood_vlans) {
3831 rspan_vlans = bitmap_allocate(4096);
3833 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3834 int64_t vlan = br->cfg->flood_vlans[i];
3835 if (vlan >= 0 && vlan < 4096) {
3836 bitmap_set1(rspan_vlans, vlan);
3837 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3840 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3845 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3850 static struct mirror *
3851 mirror_create(struct bridge *br, const char *name)
3856 for (i = 0; ; i++) {
3857 if (i >= MAX_MIRRORS) {
3858 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3859 "cannot create %s", br->name, MAX_MIRRORS, name);
3862 if (!br->mirrors[i]) {
3867 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3870 br->mirrors[i] = m = xzalloc(sizeof *m);
3873 m->name = xstrdup(name);
3874 shash_init(&m->src_ports);
3875 shash_init(&m->dst_ports);
3885 mirror_destroy(struct mirror *m)
3888 struct bridge *br = m->bridge;
3891 for (i = 0; i < br->n_ports; i++) {
3892 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3893 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3896 shash_destroy(&m->src_ports);
3897 shash_destroy(&m->dst_ports);
3900 m->bridge->mirrors[m->idx] = NULL;
3908 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3909 struct shash *names)
3913 for (i = 0; i < n_ports; i++) {
3914 const char *name = ports[i]->name;
3915 if (port_lookup(m->bridge, name)) {
3916 shash_add_once(names, name, NULL);
3918 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3919 "port %s", m->bridge->name, m->name, name);
3925 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3931 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3933 for (i = 0; i < cfg->n_select_vlan; i++) {
3934 int64_t vlan = cfg->select_vlan[i];
3935 if (vlan < 0 || vlan > 4095) {
3936 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3937 m->bridge->name, m->name, vlan);
3939 (*vlans)[n_vlans++] = vlan;
3946 vlan_is_mirrored(const struct mirror *m, int vlan)
3950 for (i = 0; i < m->n_vlans; i++) {
3951 if (m->vlans[i] == vlan) {
3959 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3963 for (i = 0; i < m->n_vlans; i++) {
3964 if (port_trunks_vlan(p, m->vlans[i])) {
3972 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3974 struct shash src_ports, dst_ports;
3975 mirror_mask_t mirror_bit;
3976 struct port *out_port;
3982 /* Get output port. */
3983 if (cfg->output_port) {
3984 out_port = port_lookup(m->bridge, cfg->output_port->name);
3986 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3987 m->bridge->name, m->name);
3993 if (cfg->output_vlan) {
3994 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3995 "output vlan; ignoring output vlan",
3996 m->bridge->name, m->name);
3998 } else if (cfg->output_vlan) {
4000 out_vlan = *cfg->output_vlan;
4002 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4003 m->bridge->name, m->name);
4008 shash_init(&src_ports);
4009 shash_init(&dst_ports);
4010 if (cfg->select_all) {
4011 for (i = 0; i < m->bridge->n_ports; i++) {
4012 const char *name = m->bridge->ports[i]->name;
4013 shash_add_once(&src_ports, name, NULL);
4014 shash_add_once(&dst_ports, name, NULL);
4019 /* Get ports, and drop duplicates and ports that don't exist. */
4020 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4022 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4025 /* Get all the vlans, and drop duplicate and invalid vlans. */
4026 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4029 /* Update mirror data. */
4030 if (!shash_equal_keys(&m->src_ports, &src_ports)
4031 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4032 || m->n_vlans != n_vlans
4033 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4034 || m->out_port != out_port
4035 || m->out_vlan != out_vlan) {
4036 bridge_flush(m->bridge);
4038 shash_swap(&m->src_ports, &src_ports);
4039 shash_swap(&m->dst_ports, &dst_ports);
4042 m->n_vlans = n_vlans;
4043 m->out_port = out_port;
4044 m->out_vlan = out_vlan;
4047 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4048 for (i = 0; i < m->bridge->n_ports; i++) {
4049 struct port *port = m->bridge->ports[i];
4051 if (shash_find(&m->src_ports, port->name)
4054 ? port_trunks_any_mirrored_vlan(m, port)
4055 : vlan_is_mirrored(m, port->vlan)))) {
4056 port->src_mirrors |= mirror_bit;
4058 port->src_mirrors &= ~mirror_bit;
4061 if (shash_find(&m->dst_ports, port->name)) {
4062 port->dst_mirrors |= mirror_bit;
4064 port->dst_mirrors &= ~mirror_bit;
4069 shash_destroy(&src_ports);
4070 shash_destroy(&dst_ports);