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 /* Kernel datapath information. */
167 struct dpif *dpif; /* Datapath. */
168 struct port_array ifaces; /* Indexed by kernel datapath port number. */
172 size_t n_ports, allocated_ports;
173 struct shash iface_by_name; /* "struct iface"s indexed by name. */
174 struct shash port_by_name; /* "struct port"s indexed by name. */
177 bool has_bonded_ports;
182 /* Port mirroring. */
183 struct mirror *mirrors[MAX_MIRRORS];
185 /* This member is only valid *during* bridge_reconfigure(). */
186 const struct ovsrec_bridge *cfg;
189 /* List of all bridges. */
190 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
192 /* Maximum number of datapaths. */
193 enum { DP_MAX = 256 };
195 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
196 static void bridge_destroy(struct bridge *);
197 static struct bridge *bridge_lookup(const char *name);
198 static unixctl_cb_func bridge_unixctl_dump_flows;
199 static int bridge_run_one(struct bridge *);
200 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
201 const struct bridge *br,
202 struct ovsrec_controller ***controllersp);
203 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
205 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
207 const struct sockaddr_in *managers,
209 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
210 static void bridge_fetch_dp_ifaces(struct bridge *);
211 static void bridge_flush(struct bridge *);
212 static void bridge_pick_local_hw_addr(struct bridge *,
213 uint8_t ea[ETH_ADDR_LEN],
214 struct iface **hw_addr_iface);
215 static uint64_t bridge_pick_datapath_id(struct bridge *,
216 const uint8_t bridge_ea[ETH_ADDR_LEN],
217 struct iface *hw_addr_iface);
218 static struct iface *bridge_get_local_iface(struct bridge *);
219 static uint64_t dpid_from_hash(const void *, size_t nbytes);
221 static unixctl_cb_func bridge_unixctl_fdb_show;
223 static void bond_init(void);
224 static void bond_run(struct bridge *);
225 static void bond_wait(struct bridge *);
226 static void bond_rebalance_port(struct port *);
227 static void bond_send_learning_packets(struct port *);
228 static void bond_enable_slave(struct iface *iface, bool enable);
230 static struct port *port_create(struct bridge *, const char *name);
231 static void port_reconfigure(struct port *, const struct ovsrec_port *);
232 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
233 static void port_destroy(struct port *);
234 static struct port *port_lookup(const struct bridge *, const char *name);
235 static struct iface *port_lookup_iface(const struct port *, const char *name);
236 static struct port *port_from_dp_ifidx(const struct bridge *,
238 static void port_update_bond_compat(struct port *);
239 static void port_update_vlan_compat(struct port *);
240 static void port_update_bonding(struct port *);
242 static struct mirror *mirror_create(struct bridge *, const char *name);
243 static void mirror_destroy(struct mirror *);
244 static void mirror_reconfigure(struct bridge *);
245 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
246 static bool vlan_is_mirrored(const struct mirror *, int vlan);
248 static struct iface *iface_create(struct port *port,
249 const struct ovsrec_interface *if_cfg);
250 static void iface_destroy(struct iface *);
251 static struct iface *iface_lookup(const struct bridge *, const char *name);
252 static struct iface *iface_from_dp_ifidx(const struct bridge *,
254 static bool iface_is_internal(const struct bridge *, const char *name);
255 static void iface_set_mac(struct iface *);
256 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
258 /* Hooks into ofproto processing. */
259 static struct ofhooks bridge_ofhooks;
261 /* Public functions. */
264 bridge_init(const struct ovsrec_open_vswitch *cfg)
266 struct svec bridge_names;
267 struct svec dpif_names, dpif_types;
270 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
272 svec_init(&bridge_names);
273 for (i = 0; i < cfg->n_bridges; i++) {
274 svec_add(&bridge_names, cfg->bridges[i]->name);
276 svec_sort(&bridge_names);
278 svec_init(&dpif_names);
279 svec_init(&dpif_types);
280 dp_enumerate_types(&dpif_types);
281 for (i = 0; i < dpif_types.n; i++) {
286 dp_enumerate_names(dpif_types.names[i], &dpif_names);
288 for (j = 0; j < dpif_names.n; j++) {
289 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
291 struct svec all_names;
294 svec_init(&all_names);
295 dpif_get_all_names(dpif, &all_names);
296 for (k = 0; k < all_names.n; k++) {
297 if (svec_contains(&bridge_names, all_names.names[k])) {
303 svec_destroy(&all_names);
308 svec_destroy(&bridge_names);
309 svec_destroy(&dpif_names);
310 svec_destroy(&dpif_types);
312 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
316 bridge_reconfigure(cfg);
321 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
323 /* XXX SSL should be configurable on a per-bridge basis. */
325 stream_ssl_set_private_key_file(ssl->private_key);
326 stream_ssl_set_certificate_file(ssl->certificate);
327 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
332 /* Attempt to create the network device 'iface_name' through the netdev
335 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
338 struct shash options;
342 shash_init(&options);
343 for (i = 0; i < iface_cfg->n_options; i++) {
344 shash_add(&options, iface_cfg->key_options[i],
345 xstrdup(iface_cfg->value_options[i]));
349 struct netdev_options netdev_options;
351 memset(&netdev_options, 0, sizeof netdev_options);
352 netdev_options.name = iface_cfg->name;
353 if (!strcmp(iface_cfg->type, "internal")) {
354 /* An "internal" config type maps to a netdev "system" type. */
355 netdev_options.type = "system";
357 netdev_options.type = iface_cfg->type;
359 netdev_options.args = &options;
360 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
362 error = netdev_open(&netdev_options, &iface->netdev);
365 netdev_get_carrier(iface->netdev, &iface->enabled);
367 } else if (iface->netdev) {
368 const char *netdev_type = netdev_get_type(iface->netdev);
369 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
370 ? iface_cfg->type : NULL;
372 /* An "internal" config type maps to a netdev "system" type. */
373 if (iface_type && !strcmp(iface_type, "internal")) {
374 iface_type = "system";
377 if (!iface_type || !strcmp(netdev_type, iface_type)) {
378 error = netdev_reconfigure(iface->netdev, &options);
380 VLOG_WARN("%s: attempting change device type from %s to %s",
381 iface_cfg->name, netdev_type, iface_type);
385 shash_destroy_free_data(&options);
391 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
393 return set_up_iface(iface_cfg, iface, false);
397 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
398 void *aux OVS_UNUSED)
400 if (!iface->netdev) {
401 int error = set_up_iface(iface->cfg, iface, true);
403 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
413 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
414 void *aux OVS_UNUSED)
416 if (iface->dp_ifidx >= 0) {
417 VLOG_DBG("%s has interface %s on port %d",
419 iface->name, iface->dp_ifidx);
422 VLOG_ERR("%s interface not in %s, dropping",
423 iface->name, dpif_name(br->dpif));
429 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
430 void *aux OVS_UNUSED)
432 /* Set policing attributes. */
433 netdev_set_policing(iface->netdev,
434 iface->cfg->ingress_policing_rate,
435 iface->cfg->ingress_policing_burst);
437 /* Set MAC address of internal interfaces other than the local
439 if (iface->dp_ifidx != ODPP_LOCAL
440 && iface_is_internal(br, iface->name)) {
441 iface_set_mac(iface);
447 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
448 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
449 * deletes from 'br' any ports that no longer have any interfaces. */
451 iterate_and_prune_ifaces(struct bridge *br,
452 bool (*cb)(struct bridge *, struct iface *,
458 for (i = 0; i < br->n_ports; ) {
459 struct port *port = br->ports[i];
460 for (j = 0; j < port->n_ifaces; ) {
461 struct iface *iface = port->ifaces[j];
462 if (cb(br, iface, aux)) {
465 iface_destroy(iface);
469 if (port->n_ifaces) {
472 VLOG_ERR("%s port has no interfaces, dropping", port->name);
478 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
479 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
480 * responsible for freeing '*managersp' (with free()).
482 * You may be asking yourself "why does ovs-vswitchd care?", because
483 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
484 * should not be and in fact is not directly involved in that. But
485 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
486 * it has to tell in-band control where the managers are to enable that.
489 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
490 struct sockaddr_in **managersp, size_t *n_managersp)
492 struct sockaddr_in *managers = NULL;
493 size_t n_managers = 0;
495 if (ovs_cfg->n_managers > 0) {
498 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
499 for (i = 0; i < ovs_cfg->n_managers; i++) {
500 const char *name = ovs_cfg->managers[i];
501 struct sockaddr_in *sin = &managers[i];
503 if ((!strncmp(name, "tcp:", 4)
504 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
505 (!strncmp(name, "ssl:", 4)
506 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
512 *managersp = managers;
513 *n_managersp = n_managers;
517 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
519 struct ovsdb_idl_txn *txn;
520 struct shash old_br, new_br;
521 struct shash_node *node;
522 struct bridge *br, *next;
523 struct sockaddr_in *managers;
526 int sflow_bridge_number;
528 COVERAGE_INC(bridge_reconfigure);
530 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
532 collect_managers(ovs_cfg, &managers, &n_managers);
534 /* Collect old and new bridges. */
537 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
538 shash_add(&old_br, br->name, br);
540 for (i = 0; i < ovs_cfg->n_bridges; i++) {
541 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
542 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
543 VLOG_WARN("more than one bridge named %s", br_cfg->name);
547 /* Get rid of deleted bridges and add new bridges. */
548 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
549 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
556 SHASH_FOR_EACH (node, &new_br) {
557 const char *br_name = node->name;
558 const struct ovsrec_bridge *br_cfg = node->data;
559 br = shash_find_data(&old_br, br_name);
561 /* If the bridge datapath type has changed, we need to tear it
562 * down and recreate. */
563 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
565 bridge_create(br_cfg);
568 bridge_create(br_cfg);
571 shash_destroy(&old_br);
572 shash_destroy(&new_br);
576 bridge_configure_ssl(ovs_cfg->ssl);
579 /* Reconfigure all bridges. */
580 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
581 bridge_reconfigure_one(ovs_cfg, br);
584 /* Add and delete ports on all datapaths.
586 * The kernel will reject any attempt to add a given port to a datapath if
587 * that port already belongs to a different datapath, so we must do all
588 * port deletions before any port additions. */
589 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
590 struct odp_port *dpif_ports;
592 struct shash want_ifaces;
594 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
595 bridge_get_all_ifaces(br, &want_ifaces);
596 for (i = 0; i < n_dpif_ports; i++) {
597 const struct odp_port *p = &dpif_ports[i];
598 if (!shash_find(&want_ifaces, p->devname)
599 && strcmp(p->devname, br->name)) {
600 int retval = dpif_port_del(br->dpif, p->port);
602 VLOG_ERR("failed to remove %s interface from %s: %s",
603 p->devname, dpif_name(br->dpif),
608 shash_destroy(&want_ifaces);
611 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
612 struct odp_port *dpif_ports;
614 struct shash cur_ifaces, want_ifaces;
615 struct shash_node *node;
617 /* Get the set of interfaces currently in this datapath. */
618 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
619 shash_init(&cur_ifaces);
620 for (i = 0; i < n_dpif_ports; i++) {
621 const char *name = dpif_ports[i].devname;
622 if (!shash_find(&cur_ifaces, name)) {
623 shash_add(&cur_ifaces, name, NULL);
628 /* Get the set of interfaces we want on this datapath. */
629 bridge_get_all_ifaces(br, &want_ifaces);
631 SHASH_FOR_EACH (node, &want_ifaces) {
632 const char *if_name = node->name;
633 struct iface *iface = node->data;
635 if (shash_find(&cur_ifaces, if_name)) {
636 /* Already exists, just reconfigure it. */
638 reconfigure_iface(iface->cfg, iface);
641 /* Need to add to datapath. */
645 /* Add to datapath. */
646 internal = iface_is_internal(br, if_name);
647 error = dpif_port_add(br->dpif, if_name,
648 internal ? ODP_PORT_INTERNAL : 0, NULL);
649 if (error == EFBIG) {
650 VLOG_ERR("ran out of valid port numbers on %s",
651 dpif_name(br->dpif));
654 VLOG_ERR("failed to add %s interface to %s: %s",
655 if_name, dpif_name(br->dpif), strerror(error));
659 shash_destroy(&cur_ifaces);
660 shash_destroy(&want_ifaces);
662 sflow_bridge_number = 0;
663 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
666 struct iface *local_iface;
667 struct iface *hw_addr_iface;
670 bridge_fetch_dp_ifaces(br);
672 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
673 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
675 /* Pick local port hardware address, datapath ID. */
676 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
677 local_iface = bridge_get_local_iface(br);
679 int error = netdev_set_etheraddr(local_iface->netdev, ea);
681 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
682 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
683 "Ethernet address: %s",
684 br->name, strerror(error));
688 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
689 ofproto_set_datapath_id(br->ofproto, dpid);
691 dpid_string = xasprintf("%016"PRIx64, dpid);
692 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
695 /* Set NetFlow configuration on this bridge. */
696 if (br->cfg->netflow) {
697 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
698 struct netflow_options opts;
700 memset(&opts, 0, sizeof opts);
702 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
703 if (nf_cfg->engine_type) {
704 opts.engine_type = *nf_cfg->engine_type;
706 if (nf_cfg->engine_id) {
707 opts.engine_id = *nf_cfg->engine_id;
710 opts.active_timeout = nf_cfg->active_timeout;
711 if (!opts.active_timeout) {
712 opts.active_timeout = -1;
713 } else if (opts.active_timeout < 0) {
714 VLOG_WARN("bridge %s: active timeout interval set to negative "
715 "value, using default instead (%d seconds)", br->name,
716 NF_ACTIVE_TIMEOUT_DEFAULT);
717 opts.active_timeout = -1;
720 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
721 if (opts.add_id_to_iface) {
722 if (opts.engine_id > 0x7f) {
723 VLOG_WARN("bridge %s: netflow port mangling may conflict "
724 "with another vswitch, choose an engine id less "
725 "than 128", br->name);
727 if (br->n_ports > 508) {
728 VLOG_WARN("bridge %s: netflow port mangling will conflict "
729 "with another port when more than 508 ports are "
734 opts.collectors.n = nf_cfg->n_targets;
735 opts.collectors.names = nf_cfg->targets;
736 if (ofproto_set_netflow(br->ofproto, &opts)) {
737 VLOG_ERR("bridge %s: problem setting netflow collectors",
741 ofproto_set_netflow(br->ofproto, NULL);
744 /* Set sFlow configuration on this bridge. */
745 if (br->cfg->sflow) {
746 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
747 struct ovsrec_controller **controllers;
748 struct ofproto_sflow_options oso;
749 size_t n_controllers;
752 memset(&oso, 0, sizeof oso);
754 oso.targets.n = sflow_cfg->n_targets;
755 oso.targets.names = sflow_cfg->targets;
757 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
758 if (sflow_cfg->sampling) {
759 oso.sampling_rate = *sflow_cfg->sampling;
762 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
763 if (sflow_cfg->polling) {
764 oso.polling_interval = *sflow_cfg->polling;
767 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
768 if (sflow_cfg->header) {
769 oso.header_len = *sflow_cfg->header;
772 oso.sub_id = sflow_bridge_number++;
773 oso.agent_device = sflow_cfg->agent;
775 oso.control_ip = NULL;
776 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
777 for (i = 0; i < n_controllers; i++) {
778 if (controllers[i]->local_ip) {
779 oso.control_ip = controllers[i]->local_ip;
783 ofproto_set_sflow(br->ofproto, &oso);
785 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
787 ofproto_set_sflow(br->ofproto, NULL);
790 /* Update the controller and related settings. It would be more
791 * straightforward to call this from bridge_reconfigure_one(), but we
792 * can't do it there for two reasons. First, and most importantly, at
793 * that point we don't know the dp_ifidx of any interfaces that have
794 * been added to the bridge (because we haven't actually added them to
795 * the datapath). Second, at that point we haven't set the datapath ID
796 * yet; when a controller is configured, resetting the datapath ID will
797 * immediately disconnect from the controller, so it's better to set
798 * the datapath ID before the controller. */
799 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
801 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
802 for (i = 0; i < br->n_ports; i++) {
803 struct port *port = br->ports[i];
806 port_update_vlan_compat(port);
807 port_update_bonding(port);
809 for (j = 0; j < port->n_ifaces; j++) {
810 iface_update_qos(port->ifaces[j], port->cfg->qos);
814 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
815 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
818 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
820 ovsdb_idl_txn_commit(txn);
821 ovsdb_idl_txn_destroy(txn); /* XXX */
827 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
831 for (i = 0; i < n; i++) {
832 if (!strcmp(keys[i], key)) {
840 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
842 return get_ovsrec_key_value(key,
843 br_cfg->key_other_config,
844 br_cfg->value_other_config,
845 br_cfg->n_other_config);
849 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
850 struct iface **hw_addr_iface)
856 *hw_addr_iface = NULL;
858 /* Did the user request a particular MAC? */
859 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
860 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
861 if (eth_addr_is_multicast(ea)) {
862 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
863 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
864 } else if (eth_addr_is_zero(ea)) {
865 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
871 /* Otherwise choose the minimum non-local MAC address among all of the
873 memset(ea, 0xff, sizeof ea);
874 for (i = 0; i < br->n_ports; i++) {
875 struct port *port = br->ports[i];
876 uint8_t iface_ea[ETH_ADDR_LEN];
879 /* Mirror output ports don't participate. */
880 if (port->is_mirror_output_port) {
884 /* Choose the MAC address to represent the port. */
885 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
886 /* Find the interface with this Ethernet address (if any) so that
887 * we can provide the correct devname to the caller. */
889 for (j = 0; j < port->n_ifaces; j++) {
890 struct iface *candidate = port->ifaces[j];
891 uint8_t candidate_ea[ETH_ADDR_LEN];
892 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
893 && eth_addr_equals(iface_ea, candidate_ea)) {
898 /* Choose the interface whose MAC address will represent the port.
899 * The Linux kernel bonding code always chooses the MAC address of
900 * the first slave added to a bond, and the Fedora networking
901 * scripts always add slaves to a bond in alphabetical order, so
902 * for compatibility we choose the interface with the name that is
903 * first in alphabetical order. */
904 iface = port->ifaces[0];
905 for (j = 1; j < port->n_ifaces; j++) {
906 struct iface *candidate = port->ifaces[j];
907 if (strcmp(candidate->name, iface->name) < 0) {
912 /* The local port doesn't count (since we're trying to choose its
913 * MAC address anyway). */
914 if (iface->dp_ifidx == ODPP_LOCAL) {
919 error = netdev_get_etheraddr(iface->netdev, iface_ea);
921 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
922 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
923 iface->name, strerror(error));
928 /* Compare against our current choice. */
929 if (!eth_addr_is_multicast(iface_ea) &&
930 !eth_addr_is_local(iface_ea) &&
931 !eth_addr_is_reserved(iface_ea) &&
932 !eth_addr_is_zero(iface_ea) &&
933 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
935 memcpy(ea, iface_ea, ETH_ADDR_LEN);
936 *hw_addr_iface = iface;
939 if (eth_addr_is_multicast(ea)) {
940 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
941 *hw_addr_iface = NULL;
942 VLOG_WARN("bridge %s: using default bridge Ethernet "
943 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
945 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
946 br->name, ETH_ADDR_ARGS(ea));
950 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
951 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
952 * an interface on 'br', then that interface must be passed in as
953 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
954 * 'hw_addr_iface' must be passed in as a null pointer. */
956 bridge_pick_datapath_id(struct bridge *br,
957 const uint8_t bridge_ea[ETH_ADDR_LEN],
958 struct iface *hw_addr_iface)
961 * The procedure for choosing a bridge MAC address will, in the most
962 * ordinary case, also choose a unique MAC that we can use as a datapath
963 * ID. In some special cases, though, multiple bridges will end up with
964 * the same MAC address. This is OK for the bridges, but it will confuse
965 * the OpenFlow controller, because each datapath needs a unique datapath
968 * Datapath IDs must be unique. It is also very desirable that they be
969 * stable from one run to the next, so that policy set on a datapath
972 const char *datapath_id;
975 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
976 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
982 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
984 * A bridge whose MAC address is taken from a VLAN network device
985 * (that is, a network device created with vconfig(8) or similar
986 * tool) will have the same MAC address as a bridge on the VLAN
987 * device's physical network device.
989 * Handle this case by hashing the physical network device MAC
990 * along with the VLAN identifier.
992 uint8_t buf[ETH_ADDR_LEN + 2];
993 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
994 buf[ETH_ADDR_LEN] = vlan >> 8;
995 buf[ETH_ADDR_LEN + 1] = vlan;
996 return dpid_from_hash(buf, sizeof buf);
999 * Assume that this bridge's MAC address is unique, since it
1000 * doesn't fit any of the cases we handle specially.
1005 * A purely internal bridge, that is, one that has no non-virtual
1006 * network devices on it at all, is more difficult because it has no
1007 * natural unique identifier at all.
1009 * When the host is a XenServer, we handle this case by hashing the
1010 * host's UUID with the name of the bridge. Names of bridges are
1011 * persistent across XenServer reboots, although they can be reused if
1012 * an internal network is destroyed and then a new one is later
1013 * created, so this is fairly effective.
1015 * When the host is not a XenServer, we punt by using a random MAC
1016 * address on each run.
1018 const char *host_uuid = xenserver_get_host_uuid();
1020 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1021 dpid = dpid_from_hash(combined, strlen(combined));
1027 return eth_addr_to_uint64(bridge_ea);
1031 dpid_from_hash(const void *data, size_t n)
1033 uint8_t hash[SHA1_DIGEST_SIZE];
1035 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1036 sha1_bytes(data, n, hash);
1037 eth_addr_mark_random(hash);
1038 return eth_addr_to_uint64(hash);
1044 struct bridge *br, *next;
1048 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1049 int error = bridge_run_one(br);
1051 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1052 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1053 "forcing reconfiguration", br->name);
1067 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1068 ofproto_wait(br->ofproto);
1069 if (ofproto_has_controller(br->ofproto)) {
1073 mac_learning_wait(br->ml);
1078 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1079 * configuration changes. */
1081 bridge_flush(struct bridge *br)
1083 COVERAGE_INC(bridge_flush);
1085 mac_learning_flush(br->ml);
1088 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1089 * such interface. */
1090 static struct iface *
1091 bridge_get_local_iface(struct bridge *br)
1095 for (i = 0; i < br->n_ports; i++) {
1096 struct port *port = br->ports[i];
1097 for (j = 0; j < port->n_ifaces; j++) {
1098 struct iface *iface = port->ifaces[j];
1099 if (iface->dp_ifidx == ODPP_LOCAL) {
1108 /* Bridge unixctl user interface functions. */
1110 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1111 const char *args, void *aux OVS_UNUSED)
1113 struct ds ds = DS_EMPTY_INITIALIZER;
1114 const struct bridge *br;
1115 const struct mac_entry *e;
1117 br = bridge_lookup(args);
1119 unixctl_command_reply(conn, 501, "no such bridge");
1123 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1124 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1125 if (e->port < 0 || e->port >= br->n_ports) {
1128 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1129 br->ports[e->port]->ifaces[0]->dp_ifidx,
1130 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1132 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1136 /* Bridge reconfiguration functions. */
1137 static struct bridge *
1138 bridge_create(const struct ovsrec_bridge *br_cfg)
1143 assert(!bridge_lookup(br_cfg->name));
1144 br = xzalloc(sizeof *br);
1146 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1152 dpif_flow_flush(br->dpif);
1154 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1157 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1159 dpif_delete(br->dpif);
1160 dpif_close(br->dpif);
1165 br->name = xstrdup(br_cfg->name);
1167 br->ml = mac_learning_create();
1168 eth_addr_nicira_random(br->default_ea);
1170 port_array_init(&br->ifaces);
1172 shash_init(&br->port_by_name);
1173 shash_init(&br->iface_by_name);
1177 list_push_back(&all_bridges, &br->node);
1179 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1185 bridge_destroy(struct bridge *br)
1190 while (br->n_ports > 0) {
1191 port_destroy(br->ports[br->n_ports - 1]);
1193 list_remove(&br->node);
1194 error = dpif_delete(br->dpif);
1195 if (error && error != ENOENT) {
1196 VLOG_ERR("failed to delete %s: %s",
1197 dpif_name(br->dpif), strerror(error));
1199 dpif_close(br->dpif);
1200 ofproto_destroy(br->ofproto);
1201 mac_learning_destroy(br->ml);
1202 port_array_destroy(&br->ifaces);
1203 shash_destroy(&br->port_by_name);
1204 shash_destroy(&br->iface_by_name);
1211 static struct bridge *
1212 bridge_lookup(const char *name)
1216 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1217 if (!strcmp(br->name, name)) {
1224 /* Handle requests for a listing of all flows known by the OpenFlow
1225 * stack, including those normally hidden. */
1227 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1228 const char *args, void *aux OVS_UNUSED)
1233 br = bridge_lookup(args);
1235 unixctl_command_reply(conn, 501, "Unknown bridge");
1240 ofproto_get_all_flows(br->ofproto, &results);
1242 unixctl_command_reply(conn, 200, ds_cstr(&results));
1243 ds_destroy(&results);
1247 bridge_run_one(struct bridge *br)
1251 error = ofproto_run1(br->ofproto);
1256 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1259 error = ofproto_run2(br->ofproto, br->flush);
1266 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1267 const struct bridge *br,
1268 struct ovsrec_controller ***controllersp)
1270 struct ovsrec_controller **controllers;
1271 size_t n_controllers;
1273 if (br->cfg->n_controller) {
1274 controllers = br->cfg->controller;
1275 n_controllers = br->cfg->n_controller;
1277 controllers = ovs_cfg->controller;
1278 n_controllers = ovs_cfg->n_controller;
1281 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1287 *controllersp = controllers;
1289 return n_controllers;
1293 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1296 struct shash old_ports, new_ports;
1297 struct svec listeners, old_listeners;
1298 struct svec snoops, old_snoops;
1299 struct shash_node *node;
1302 /* Collect old ports. */
1303 shash_init(&old_ports);
1304 for (i = 0; i < br->n_ports; i++) {
1305 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1308 /* Collect new ports. */
1309 shash_init(&new_ports);
1310 for (i = 0; i < br->cfg->n_ports; i++) {
1311 const char *name = br->cfg->ports[i]->name;
1312 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1313 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1318 /* If we have a controller, then we need a local port. Complain if the
1319 * user didn't specify one.
1321 * XXX perhaps we should synthesize a port ourselves in this case. */
1322 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1323 char local_name[IF_NAMESIZE];
1326 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1327 local_name, sizeof local_name);
1328 if (!error && !shash_find(&new_ports, local_name)) {
1329 VLOG_WARN("bridge %s: controller specified but no local port "
1330 "(port named %s) defined",
1331 br->name, local_name);
1335 /* Get rid of deleted ports.
1336 * Get rid of deleted interfaces on ports that still exist. */
1337 SHASH_FOR_EACH (node, &old_ports) {
1338 struct port *port = node->data;
1339 const struct ovsrec_port *port_cfg;
1341 port_cfg = shash_find_data(&new_ports, node->name);
1345 port_del_ifaces(port, port_cfg);
1349 /* Create new ports.
1350 * Add new interfaces to existing ports.
1351 * Reconfigure existing ports. */
1352 SHASH_FOR_EACH (node, &new_ports) {
1353 struct port *port = shash_find_data(&old_ports, node->name);
1355 port = port_create(br, node->name);
1358 port_reconfigure(port, node->data);
1359 if (!port->n_ifaces) {
1360 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1361 br->name, port->name);
1365 shash_destroy(&old_ports);
1366 shash_destroy(&new_ports);
1368 /* Delete all flows if we're switching from connected to standalone or vice
1369 * versa. (XXX Should we delete all flows if we are switching from one
1370 * controller to another?) */
1372 /* Configure OpenFlow management listener. */
1373 svec_init(&listeners);
1374 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1375 ovs_rundir, br->name));
1376 svec_init(&old_listeners);
1377 ofproto_get_listeners(br->ofproto, &old_listeners);
1378 if (!svec_equal(&listeners, &old_listeners)) {
1379 ofproto_set_listeners(br->ofproto, &listeners);
1381 svec_destroy(&listeners);
1382 svec_destroy(&old_listeners);
1384 /* Configure OpenFlow controller connection snooping. */
1386 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1387 ovs_rundir, br->name));
1388 svec_init(&old_snoops);
1389 ofproto_get_snoops(br->ofproto, &old_snoops);
1390 if (!svec_equal(&snoops, &old_snoops)) {
1391 ofproto_set_snoops(br->ofproto, &snoops);
1393 svec_destroy(&snoops);
1394 svec_destroy(&old_snoops);
1396 mirror_reconfigure(br);
1400 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1402 const struct sockaddr_in *managers,
1405 struct ovsrec_controller **controllers;
1406 size_t n_controllers;
1408 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1410 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1411 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1412 ofproto_flush_flows(br->ofproto);
1415 if (!n_controllers) {
1416 union ofp_action action;
1419 /* Clear out controllers. */
1420 ofproto_set_controllers(br->ofproto, NULL, 0);
1422 /* Set up a flow that matches every packet and directs them to
1423 * OFPP_NORMAL (which goes to us). */
1424 memset(&action, 0, sizeof action);
1425 action.type = htons(OFPAT_OUTPUT);
1426 action.output.len = htons(sizeof action);
1427 action.output.port = htons(OFPP_NORMAL);
1428 memset(&flow, 0, sizeof flow);
1429 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1431 struct ofproto_controller *ocs;
1434 ocs = xmalloc(n_controllers * sizeof *ocs);
1435 for (i = 0; i < n_controllers; i++) {
1436 struct ovsrec_controller *c = controllers[i];
1437 struct ofproto_controller *oc = &ocs[i];
1439 if (strcmp(c->target, "discover")) {
1440 struct iface *local_iface;
1443 local_iface = bridge_get_local_iface(br);
1444 if (local_iface && c->local_ip
1445 && inet_aton(c->local_ip, &ip)) {
1446 struct netdev *netdev = local_iface->netdev;
1447 struct in_addr mask, gateway;
1449 if (!c->local_netmask
1450 || !inet_aton(c->local_netmask, &mask)) {
1453 if (!c->local_gateway
1454 || !inet_aton(c->local_gateway, &gateway)) {
1458 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1460 mask.s_addr = guess_netmask(ip.s_addr);
1462 if (!netdev_set_in4(netdev, ip, mask)) {
1463 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1465 br->name, IP_ARGS(&ip.s_addr),
1466 IP_ARGS(&mask.s_addr));
1469 if (gateway.s_addr) {
1470 if (!netdev_add_router(netdev, gateway)) {
1471 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1472 br->name, IP_ARGS(&gateway.s_addr));
1478 oc->target = c->target;
1479 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1480 oc->probe_interval = (c->inactivity_probe
1481 ? *c->inactivity_probe / 1000 : 5);
1482 oc->fail = (!c->fail_mode
1483 || !strcmp(c->fail_mode, "standalone")
1484 || !strcmp(c->fail_mode, "open")
1485 ? OFPROTO_FAIL_STANDALONE
1486 : OFPROTO_FAIL_SECURE);
1487 oc->band = (!c->connection_mode
1488 || !strcmp(c->connection_mode, "in-band")
1490 : OFPROTO_OUT_OF_BAND);
1491 oc->accept_re = c->discover_accept_regex;
1492 oc->update_resolv_conf = c->discover_update_resolv_conf;
1493 oc->rate_limit = (c->controller_rate_limit
1494 ? *c->controller_rate_limit : 0);
1495 oc->burst_limit = (c->controller_burst_limit
1496 ? *c->controller_burst_limit : 0);
1498 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1504 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1509 for (i = 0; i < br->n_ports; i++) {
1510 struct port *port = br->ports[i];
1511 for (j = 0; j < port->n_ifaces; j++) {
1512 struct iface *iface = port->ifaces[j];
1513 shash_add_once(ifaces, iface->name, iface);
1515 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1516 shash_add_once(ifaces, port->name, NULL);
1521 /* For robustness, in case the administrator moves around datapath ports behind
1522 * our back, we re-check all the datapath port numbers here.
1524 * This function will set the 'dp_ifidx' members of interfaces that have
1525 * disappeared to -1, so only call this function from a context where those
1526 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1527 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1528 * datapath, which doesn't support UINT16_MAX+1 ports. */
1530 bridge_fetch_dp_ifaces(struct bridge *br)
1532 struct odp_port *dpif_ports;
1533 size_t n_dpif_ports;
1536 /* Reset all interface numbers. */
1537 for (i = 0; i < br->n_ports; i++) {
1538 struct port *port = br->ports[i];
1539 for (j = 0; j < port->n_ifaces; j++) {
1540 struct iface *iface = port->ifaces[j];
1541 iface->dp_ifidx = -1;
1544 port_array_clear(&br->ifaces);
1546 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1547 for (i = 0; i < n_dpif_ports; i++) {
1548 struct odp_port *p = &dpif_ports[i];
1549 struct iface *iface = iface_lookup(br, p->devname);
1551 if (iface->dp_ifidx >= 0) {
1552 VLOG_WARN("%s reported interface %s twice",
1553 dpif_name(br->dpif), p->devname);
1554 } else if (iface_from_dp_ifidx(br, p->port)) {
1555 VLOG_WARN("%s reported interface %"PRIu16" twice",
1556 dpif_name(br->dpif), p->port);
1558 port_array_set(&br->ifaces, p->port, iface);
1559 iface->dp_ifidx = p->port;
1563 int64_t ofport = (iface->dp_ifidx >= 0
1564 ? odp_port_to_ofp_port(iface->dp_ifidx)
1566 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1573 /* Bridge packet processing functions. */
1576 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1578 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1581 static struct bond_entry *
1582 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1584 return &port->bond_hash[bond_hash(mac)];
1588 bond_choose_iface(const struct port *port)
1590 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1591 size_t i, best_down_slave = -1;
1592 long long next_delay_expiration = LLONG_MAX;
1594 for (i = 0; i < port->n_ifaces; i++) {
1595 struct iface *iface = port->ifaces[i];
1597 if (iface->enabled) {
1599 } else if (iface->delay_expires < next_delay_expiration) {
1600 best_down_slave = i;
1601 next_delay_expiration = iface->delay_expires;
1605 if (best_down_slave != -1) {
1606 struct iface *iface = port->ifaces[best_down_slave];
1608 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1609 "since no other interface is up", iface->name,
1610 iface->delay_expires - time_msec());
1611 bond_enable_slave(iface, true);
1614 return best_down_slave;
1618 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1619 uint16_t *dp_ifidx, tag_type *tags)
1621 struct iface *iface;
1623 assert(port->n_ifaces);
1624 if (port->n_ifaces == 1) {
1625 iface = port->ifaces[0];
1627 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1628 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1629 || !port->ifaces[e->iface_idx]->enabled) {
1630 /* XXX select interface properly. The current interface selection
1631 * is only good for testing the rebalancing code. */
1632 e->iface_idx = bond_choose_iface(port);
1633 if (e->iface_idx < 0) {
1634 *tags |= port->no_ifaces_tag;
1637 e->iface_tag = tag_create_random();
1638 ((struct port *) port)->bond_compat_is_stale = true;
1640 *tags |= e->iface_tag;
1641 iface = port->ifaces[e->iface_idx];
1643 *dp_ifidx = iface->dp_ifidx;
1644 *tags |= iface->tag; /* Currently only used for bonding. */
1649 bond_link_status_update(struct iface *iface, bool carrier)
1651 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1652 struct port *port = iface->port;
1654 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1655 /* Nothing to do. */
1658 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1659 iface->name, carrier ? "detected" : "dropped");
1660 if (carrier == iface->enabled) {
1661 iface->delay_expires = LLONG_MAX;
1662 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1663 iface->name, carrier ? "disabled" : "enabled");
1664 } else if (carrier && port->active_iface < 0) {
1665 bond_enable_slave(iface, true);
1666 if (port->updelay) {
1667 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1668 "other interface is up", iface->name, port->updelay);
1671 int delay = carrier ? port->updelay : port->downdelay;
1672 iface->delay_expires = time_msec() + delay;
1675 "interface %s: will be %s if it stays %s for %d ms",
1677 carrier ? "enabled" : "disabled",
1678 carrier ? "up" : "down",
1685 bond_choose_active_iface(struct port *port)
1687 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1689 port->active_iface = bond_choose_iface(port);
1690 port->active_iface_tag = tag_create_random();
1691 if (port->active_iface >= 0) {
1692 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1693 port->name, port->ifaces[port->active_iface]->name);
1695 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1701 bond_enable_slave(struct iface *iface, bool enable)
1703 struct port *port = iface->port;
1704 struct bridge *br = port->bridge;
1706 /* This acts as a recursion check. If the act of disabling a slave
1707 * causes a different slave to be enabled, the flag will allow us to
1708 * skip redundant work when we reenter this function. It must be
1709 * cleared on exit to keep things safe with multiple bonds. */
1710 static bool moving_active_iface = false;
1712 iface->delay_expires = LLONG_MAX;
1713 if (enable == iface->enabled) {
1717 iface->enabled = enable;
1718 if (!iface->enabled) {
1719 VLOG_WARN("interface %s: disabled", iface->name);
1720 ofproto_revalidate(br->ofproto, iface->tag);
1721 if (iface->port_ifidx == port->active_iface) {
1722 ofproto_revalidate(br->ofproto,
1723 port->active_iface_tag);
1725 /* Disabling a slave can lead to another slave being immediately
1726 * enabled if there will be no active slaves but one is waiting
1727 * on an updelay. In this case we do not need to run most of the
1728 * code for the newly enabled slave since there was no period
1729 * without an active slave and it is redundant with the disabling
1731 moving_active_iface = true;
1732 bond_choose_active_iface(port);
1734 bond_send_learning_packets(port);
1736 VLOG_WARN("interface %s: enabled", iface->name);
1737 if (port->active_iface < 0 && !moving_active_iface) {
1738 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1739 bond_choose_active_iface(port);
1740 bond_send_learning_packets(port);
1742 iface->tag = tag_create_random();
1745 moving_active_iface = false;
1746 port->bond_compat_is_stale = true;
1749 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1750 * bond interface. */
1752 bond_update_fake_iface_stats(struct port *port)
1754 struct netdev_stats bond_stats;
1755 struct netdev *bond_dev;
1758 memset(&bond_stats, 0, sizeof bond_stats);
1760 for (i = 0; i < port->n_ifaces; i++) {
1761 struct netdev_stats slave_stats;
1763 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1764 /* XXX: We swap the stats here because they are swapped back when
1765 * reported by the internal device. The reason for this is
1766 * internal devices normally represent packets going into the system
1767 * but when used as fake bond device they represent packets leaving
1768 * the system. We really should do this in the internal device
1769 * itself because changing it here reverses the counts from the
1770 * perspective of the switch. However, the internal device doesn't
1771 * know what type of device it represents so we have to do it here
1773 bond_stats.tx_packets += slave_stats.rx_packets;
1774 bond_stats.tx_bytes += slave_stats.rx_bytes;
1775 bond_stats.rx_packets += slave_stats.tx_packets;
1776 bond_stats.rx_bytes += slave_stats.tx_bytes;
1780 if (!netdev_open_default(port->name, &bond_dev)) {
1781 netdev_set_stats(bond_dev, &bond_stats);
1782 netdev_close(bond_dev);
1787 bond_run(struct bridge *br)
1791 for (i = 0; i < br->n_ports; i++) {
1792 struct port *port = br->ports[i];
1794 if (port->n_ifaces >= 2) {
1795 for (j = 0; j < port->n_ifaces; j++) {
1796 struct iface *iface = port->ifaces[j];
1797 if (time_msec() >= iface->delay_expires) {
1798 bond_enable_slave(iface, !iface->enabled);
1802 if (port->bond_fake_iface
1803 && time_msec() >= port->bond_next_fake_iface_update) {
1804 bond_update_fake_iface_stats(port);
1805 port->bond_next_fake_iface_update = time_msec() + 1000;
1809 if (port->bond_compat_is_stale) {
1810 port->bond_compat_is_stale = false;
1811 port_update_bond_compat(port);
1817 bond_wait(struct bridge *br)
1821 for (i = 0; i < br->n_ports; i++) {
1822 struct port *port = br->ports[i];
1823 if (port->n_ifaces < 2) {
1826 for (j = 0; j < port->n_ifaces; j++) {
1827 struct iface *iface = port->ifaces[j];
1828 if (iface->delay_expires != LLONG_MAX) {
1829 poll_timer_wait_until(iface->delay_expires);
1832 if (port->bond_fake_iface) {
1833 poll_timer_wait_until(port->bond_next_fake_iface_update);
1839 set_dst(struct dst *p, const flow_t *flow,
1840 const struct port *in_port, const struct port *out_port,
1843 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1844 : in_port->vlan >= 0 ? in_port->vlan
1845 : ntohs(flow->dl_vlan));
1846 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1850 swap_dst(struct dst *p, struct dst *q)
1852 struct dst tmp = *p;
1857 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1858 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1859 * that we push to the datapath. We could in fact fully sort the array by
1860 * vlan, but in most cases there are at most two different vlan tags so that's
1861 * possibly overkill.) */
1863 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1865 struct dst *first = dsts;
1866 struct dst *last = dsts + n_dsts;
1868 while (first != last) {
1870 * - All dsts < first have vlan == 'vlan'.
1871 * - All dsts >= last have vlan != 'vlan'.
1872 * - first < last. */
1873 while (first->vlan == vlan) {
1874 if (++first == last) {
1879 /* Same invariants, plus one additional:
1880 * - first->vlan != vlan.
1882 while (last[-1].vlan != vlan) {
1883 if (--last == first) {
1888 /* Same invariants, plus one additional:
1889 * - last[-1].vlan == vlan.*/
1890 swap_dst(first++, --last);
1895 mirror_mask_ffs(mirror_mask_t mask)
1897 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1902 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1903 const struct dst *test)
1906 for (i = 0; i < n_dsts; i++) {
1907 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1915 port_trunks_vlan(const struct port *port, uint16_t vlan)
1917 return (port->vlan < 0
1918 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
1922 port_includes_vlan(const struct port *port, uint16_t vlan)
1924 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1928 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1929 const struct port *in_port, const struct port *out_port,
1930 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1932 mirror_mask_t mirrors = in_port->src_mirrors;
1933 struct dst *dst = dsts;
1936 if (out_port == FLOOD_PORT) {
1937 /* XXX use ODP_FLOOD if no vlans or bonding. */
1938 /* XXX even better, define each VLAN as a datapath port group */
1939 for (i = 0; i < br->n_ports; i++) {
1940 struct port *port = br->ports[i];
1941 if (port != in_port && port_includes_vlan(port, vlan)
1942 && !port->is_mirror_output_port
1943 && set_dst(dst, flow, in_port, port, tags)) {
1944 mirrors |= port->dst_mirrors;
1948 *nf_output_iface = NF_OUT_FLOOD;
1949 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1950 *nf_output_iface = dst->dp_ifidx;
1951 mirrors |= out_port->dst_mirrors;
1956 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1957 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1959 if (set_dst(dst, flow, in_port, m->out_port, tags)
1960 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1964 for (i = 0; i < br->n_ports; i++) {
1965 struct port *port = br->ports[i];
1966 if (port_includes_vlan(port, m->out_vlan)
1967 && set_dst(dst, flow, in_port, port, tags))
1971 if (port->vlan < 0) {
1972 dst->vlan = m->out_vlan;
1974 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1978 /* Use the vlan tag on the original flow instead of
1979 * the one passed in the vlan parameter. This ensures
1980 * that we compare the vlan from before any implicit
1981 * tagging tags place. This is necessary because
1982 * dst->vlan is the final vlan, after removing implicit
1984 flow_vlan = ntohs(flow->dl_vlan);
1985 if (flow_vlan == 0) {
1986 flow_vlan = OFP_VLAN_NONE;
1988 if (port == in_port && dst->vlan == flow_vlan) {
1989 /* Don't send out input port on same VLAN. */
1997 mirrors &= mirrors - 1;
2000 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2004 static void OVS_UNUSED
2005 print_dsts(const struct dst *dsts, size_t n)
2007 for (; n--; dsts++) {
2008 printf(">p%"PRIu16, dsts->dp_ifidx);
2009 if (dsts->vlan != OFP_VLAN_NONE) {
2010 printf("v%"PRIu16, dsts->vlan);
2016 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2017 const struct port *in_port, const struct port *out_port,
2018 tag_type *tags, struct odp_actions *actions,
2019 uint16_t *nf_output_iface)
2021 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2023 const struct dst *p;
2026 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2029 cur_vlan = ntohs(flow->dl_vlan);
2030 for (p = dsts; p < &dsts[n_dsts]; p++) {
2031 union odp_action *a;
2032 if (p->vlan != cur_vlan) {
2033 if (p->vlan == OFP_VLAN_NONE) {
2034 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2036 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2037 a->vlan_vid.vlan_vid = htons(p->vlan);
2041 a = odp_actions_add(actions, ODPAT_OUTPUT);
2042 a->output.port = p->dp_ifidx;
2046 /* Returns the effective vlan of a packet, taking into account both the
2047 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2048 * the packet is untagged and -1 indicates it has an invalid header and
2049 * should be dropped. */
2050 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2051 struct port *in_port, bool have_packet)
2053 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2054 * belongs to VLAN 0, so we should treat both cases identically. (In the
2055 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2056 * presumably to allow a priority to be specified. In the latter case, the
2057 * packet does not have any 802.1Q header.) */
2058 int vlan = ntohs(flow->dl_vlan);
2059 if (vlan == OFP_VLAN_NONE) {
2062 if (in_port->vlan >= 0) {
2064 /* XXX support double tagging? */
2066 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2067 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2068 "packet received on port %s configured with "
2069 "implicit VLAN %"PRIu16,
2070 br->name, ntohs(flow->dl_vlan),
2071 in_port->name, in_port->vlan);
2075 vlan = in_port->vlan;
2077 if (!port_includes_vlan(in_port, vlan)) {
2079 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2080 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2081 "packet received on port %s not configured for "
2083 br->name, vlan, in_port->name, vlan);
2092 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2093 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2094 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2096 is_gratuitous_arp(const flow_t *flow)
2098 return (flow->dl_type == htons(ETH_TYPE_ARP)
2099 && eth_addr_is_broadcast(flow->dl_dst)
2100 && (flow->nw_proto == ARP_OP_REPLY
2101 || (flow->nw_proto == ARP_OP_REQUEST
2102 && flow->nw_src == flow->nw_dst)));
2106 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2107 struct port *in_port)
2109 enum grat_arp_lock_type lock_type;
2112 /* We don't want to learn from gratuitous ARP packets that are reflected
2113 * back over bond slaves so we lock the learning table. */
2114 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2115 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2116 GRAT_ARP_LOCK_CHECK;
2118 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2121 /* The log messages here could actually be useful in debugging,
2122 * so keep the rate limit relatively high. */
2123 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2125 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2126 "on port %s in VLAN %d",
2127 br->name, ETH_ADDR_ARGS(flow->dl_src),
2128 in_port->name, vlan);
2129 ofproto_revalidate(br->ofproto, rev_tag);
2133 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2134 * dropped. Returns true if they may be forwarded, false if they should be
2137 * If 'have_packet' is true, it indicates that the caller is processing a
2138 * received packet. If 'have_packet' is false, then the caller is just
2139 * revalidating an existing flow because configuration has changed. Either
2140 * way, 'have_packet' only affects logging (there is no point in logging errors
2141 * during revalidation).
2143 * Sets '*in_portp' to the input port. This will be a null pointer if
2144 * flow->in_port does not designate a known input port (in which case
2145 * is_admissible() returns false).
2147 * When returning true, sets '*vlanp' to the effective VLAN of the input
2148 * packet, as returned by flow_get_vlan().
2150 * May also add tags to '*tags', although the current implementation only does
2151 * so in one special case.
2154 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2155 tag_type *tags, int *vlanp, struct port **in_portp)
2157 struct iface *in_iface;
2158 struct port *in_port;
2161 /* Find the interface and port structure for the received packet. */
2162 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2164 /* No interface? Something fishy... */
2166 /* Odd. A few possible reasons here:
2168 * - We deleted an interface but there are still a few packets
2169 * queued up from it.
2171 * - Someone externally added an interface (e.g. with "ovs-dpctl
2172 * add-if") that we don't know about.
2174 * - Packet arrived on the local port but the local port is not
2175 * one of our bridge ports.
2177 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2179 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2180 "interface %"PRIu16, br->name, flow->in_port);
2186 *in_portp = in_port = in_iface->port;
2187 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2192 /* Drop frames for reserved multicast addresses. */
2193 if (eth_addr_is_reserved(flow->dl_dst)) {
2197 /* Drop frames on ports reserved for mirroring. */
2198 if (in_port->is_mirror_output_port) {
2200 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2201 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2202 "%s, which is reserved exclusively for mirroring",
2203 br->name, in_port->name);
2208 /* Packets received on bonds need special attention to avoid duplicates. */
2209 if (in_port->n_ifaces > 1) {
2211 bool is_grat_arp_locked;
2213 if (eth_addr_is_multicast(flow->dl_dst)) {
2214 *tags |= in_port->active_iface_tag;
2215 if (in_port->active_iface != in_iface->port_ifidx) {
2216 /* Drop all multicast packets on inactive slaves. */
2221 /* Drop all packets for which we have learned a different input
2222 * port, because we probably sent the packet on one slave and got
2223 * it back on the other. Gratuitous ARP packets are an exception
2224 * to this rule: the host has moved to another switch. The exception
2225 * to the exception is if we locked the learning table to avoid
2226 * reflections on bond slaves. If this is the case, just drop the
2228 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2229 &is_grat_arp_locked);
2230 if (src_idx != -1 && src_idx != in_port->port_idx &&
2231 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2239 /* If the composed actions may be applied to any packet in the given 'flow',
2240 * returns true. Otherwise, the actions should only be applied to 'packet', or
2241 * not at all, if 'packet' was NULL. */
2243 process_flow(struct bridge *br, const flow_t *flow,
2244 const struct ofpbuf *packet, struct odp_actions *actions,
2245 tag_type *tags, uint16_t *nf_output_iface)
2247 struct port *in_port;
2248 struct port *out_port;
2252 /* Check whether we should drop packets in this flow. */
2253 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2258 /* Learn source MAC (but don't try to learn from revalidation). */
2260 update_learning_table(br, flow, vlan, in_port);
2263 /* Determine output port. */
2264 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2266 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2267 out_port = br->ports[out_port_idx];
2268 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2269 /* If we are revalidating but don't have a learning entry then
2270 * eject the flow. Installing a flow that floods packets opens
2271 * up a window of time where we could learn from a packet reflected
2272 * on a bond and blackhole packets before the learning table is
2273 * updated to reflect the correct port. */
2276 out_port = FLOOD_PORT;
2279 /* Don't send packets out their input ports. */
2280 if (in_port == out_port) {
2286 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2293 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2296 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2297 const struct ofp_phy_port *opp,
2300 struct bridge *br = br_;
2301 struct iface *iface;
2304 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2310 if (reason == OFPPR_DELETE) {
2311 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2312 br->name, iface->name);
2313 iface_destroy(iface);
2314 if (!port->n_ifaces) {
2315 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2316 br->name, port->name);
2322 if (port->n_ifaces > 1) {
2323 bool up = !(opp->state & OFPPS_LINK_DOWN);
2324 bond_link_status_update(iface, up);
2325 port_update_bond_compat(port);
2331 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2332 struct odp_actions *actions, tag_type *tags,
2333 uint16_t *nf_output_iface, void *br_)
2335 struct bridge *br = br_;
2337 COVERAGE_INC(bridge_process_flow);
2338 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2342 bridge_account_flow_ofhook_cb(const flow_t *flow,
2343 const union odp_action *actions,
2344 size_t n_actions, unsigned long long int n_bytes,
2347 struct bridge *br = br_;
2348 const union odp_action *a;
2349 struct port *in_port;
2353 /* Feed information from the active flows back into the learning table
2354 * to ensure that table is always in sync with what is actually flowing
2355 * through the datapath. */
2356 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2357 update_learning_table(br, flow, vlan, in_port);
2360 if (!br->has_bonded_ports) {
2364 for (a = actions; a < &actions[n_actions]; a++) {
2365 if (a->type == ODPAT_OUTPUT) {
2366 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2367 if (out_port && out_port->n_ifaces >= 2) {
2368 struct bond_entry *e = lookup_bond_entry(out_port,
2370 e->tx_bytes += n_bytes;
2377 bridge_account_checkpoint_ofhook_cb(void *br_)
2379 struct bridge *br = br_;
2383 if (!br->has_bonded_ports) {
2388 for (i = 0; i < br->n_ports; i++) {
2389 struct port *port = br->ports[i];
2390 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2391 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2392 bond_rebalance_port(port);
2397 static struct ofhooks bridge_ofhooks = {
2398 bridge_port_changed_ofhook_cb,
2399 bridge_normal_ofhook_cb,
2400 bridge_account_flow_ofhook_cb,
2401 bridge_account_checkpoint_ofhook_cb,
2404 /* Bonding functions. */
2406 /* Statistics for a single interface on a bonded port, used for load-based
2407 * bond rebalancing. */
2408 struct slave_balance {
2409 struct iface *iface; /* The interface. */
2410 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2412 /* All the "bond_entry"s that are assigned to this interface, in order of
2413 * increasing tx_bytes. */
2414 struct bond_entry **hashes;
2418 /* Sorts pointers to pointers to bond_entries in ascending order by the
2419 * interface to which they are assigned, and within a single interface in
2420 * ascending order of bytes transmitted. */
2422 compare_bond_entries(const void *a_, const void *b_)
2424 const struct bond_entry *const *ap = a_;
2425 const struct bond_entry *const *bp = b_;
2426 const struct bond_entry *a = *ap;
2427 const struct bond_entry *b = *bp;
2428 if (a->iface_idx != b->iface_idx) {
2429 return a->iface_idx > b->iface_idx ? 1 : -1;
2430 } else if (a->tx_bytes != b->tx_bytes) {
2431 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2437 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2438 * *descending* order by number of bytes transmitted. */
2440 compare_slave_balance(const void *a_, const void *b_)
2442 const struct slave_balance *a = a_;
2443 const struct slave_balance *b = b_;
2444 if (a->iface->enabled != b->iface->enabled) {
2445 return a->iface->enabled ? -1 : 1;
2446 } else if (a->tx_bytes != b->tx_bytes) {
2447 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2454 swap_bals(struct slave_balance *a, struct slave_balance *b)
2456 struct slave_balance tmp = *a;
2461 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2462 * given that 'p' (and only 'p') might be in the wrong location.
2464 * This function invalidates 'p', since it might now be in a different memory
2467 resort_bals(struct slave_balance *p,
2468 struct slave_balance bals[], size_t n_bals)
2471 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2472 swap_bals(p, p - 1);
2474 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2475 swap_bals(p, p + 1);
2481 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2483 if (VLOG_IS_DBG_ENABLED()) {
2484 struct ds ds = DS_EMPTY_INITIALIZER;
2485 const struct slave_balance *b;
2487 for (b = bals; b < bals + n_bals; b++) {
2491 ds_put_char(&ds, ',');
2493 ds_put_format(&ds, " %s %"PRIu64"kB",
2494 b->iface->name, b->tx_bytes / 1024);
2496 if (!b->iface->enabled) {
2497 ds_put_cstr(&ds, " (disabled)");
2499 if (b->n_hashes > 0) {
2500 ds_put_cstr(&ds, " (");
2501 for (i = 0; i < b->n_hashes; i++) {
2502 const struct bond_entry *e = b->hashes[i];
2504 ds_put_cstr(&ds, " + ");
2506 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2507 e - port->bond_hash, e->tx_bytes / 1024);
2509 ds_put_cstr(&ds, ")");
2512 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2517 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2519 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2522 struct bond_entry *hash = from->hashes[hash_idx];
2523 struct port *port = from->iface->port;
2524 uint64_t delta = hash->tx_bytes;
2526 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2527 "from %s to %s (now carrying %"PRIu64"kB and "
2528 "%"PRIu64"kB load, respectively)",
2529 port->name, delta / 1024, hash - port->bond_hash,
2530 from->iface->name, to->iface->name,
2531 (from->tx_bytes - delta) / 1024,
2532 (to->tx_bytes + delta) / 1024);
2534 /* Delete element from from->hashes.
2536 * We don't bother to add the element to to->hashes because not only would
2537 * it require more work, the only purpose it would be to allow that hash to
2538 * be migrated to another slave in this rebalancing run, and there is no
2539 * point in doing that. */
2540 if (hash_idx == 0) {
2543 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2544 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2548 /* Shift load away from 'from' to 'to'. */
2549 from->tx_bytes -= delta;
2550 to->tx_bytes += delta;
2552 /* Arrange for flows to be revalidated. */
2553 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2554 hash->iface_idx = to->iface->port_ifidx;
2555 hash->iface_tag = tag_create_random();
2559 bond_rebalance_port(struct port *port)
2561 struct slave_balance bals[DP_MAX_PORTS];
2563 struct bond_entry *hashes[BOND_MASK + 1];
2564 struct slave_balance *b, *from, *to;
2565 struct bond_entry *e;
2568 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2569 * descending order of tx_bytes, so that bals[0] represents the most
2570 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2573 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2574 * array for each slave_balance structure, we sort our local array of
2575 * hashes in order by slave, so that all of the hashes for a given slave
2576 * become contiguous in memory, and then we point each 'hashes' members of
2577 * a slave_balance structure to the start of a contiguous group. */
2578 n_bals = port->n_ifaces;
2579 for (b = bals; b < &bals[n_bals]; b++) {
2580 b->iface = port->ifaces[b - bals];
2585 for (i = 0; i <= BOND_MASK; i++) {
2586 hashes[i] = &port->bond_hash[i];
2588 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2589 for (i = 0; i <= BOND_MASK; i++) {
2591 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2592 b = &bals[e->iface_idx];
2593 b->tx_bytes += e->tx_bytes;
2595 b->hashes = &hashes[i];
2600 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2601 log_bals(bals, n_bals, port);
2603 /* Discard slaves that aren't enabled (which were sorted to the back of the
2604 * array earlier). */
2605 while (!bals[n_bals - 1].iface->enabled) {
2612 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2613 to = &bals[n_bals - 1];
2614 for (from = bals; from < to; ) {
2615 uint64_t overload = from->tx_bytes - to->tx_bytes;
2616 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2617 /* The extra load on 'from' (and all less-loaded slaves), compared
2618 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2619 * it is less than ~1Mbps. No point in rebalancing. */
2621 } else if (from->n_hashes == 1) {
2622 /* 'from' only carries a single MAC hash, so we can't shift any
2623 * load away from it, even though we want to. */
2626 /* 'from' is carrying significantly more load than 'to', and that
2627 * load is split across at least two different hashes. Pick a hash
2628 * to migrate to 'to' (the least-loaded slave), given that doing so
2629 * must decrease the ratio of the load on the two slaves by at
2632 * The sort order we use means that we prefer to shift away the
2633 * smallest hashes instead of the biggest ones. There is little
2634 * reason behind this decision; we could use the opposite sort
2635 * order to shift away big hashes ahead of small ones. */
2639 for (i = 0; i < from->n_hashes; i++) {
2640 double old_ratio, new_ratio;
2641 uint64_t delta = from->hashes[i]->tx_bytes;
2643 if (delta == 0 || from->tx_bytes - delta == 0) {
2644 /* Pointless move. */
2648 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2650 if (to->tx_bytes == 0) {
2651 /* Nothing on the new slave, move it. */
2655 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2656 new_ratio = (double)(from->tx_bytes - delta) /
2657 (to->tx_bytes + delta);
2659 if (new_ratio == 0) {
2660 /* Should already be covered but check to prevent division
2665 if (new_ratio < 1) {
2666 new_ratio = 1 / new_ratio;
2669 if (old_ratio - new_ratio > 0.1) {
2670 /* Would decrease the ratio, move it. */
2674 if (i < from->n_hashes) {
2675 bond_shift_load(from, to, i);
2676 port->bond_compat_is_stale = true;
2678 /* If the result of the migration changed the relative order of
2679 * 'from' and 'to' swap them back to maintain invariants. */
2680 if (order_swapped) {
2681 swap_bals(from, to);
2684 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2685 * point to different slave_balance structures. It is only
2686 * valid to do these two operations in a row at all because we
2687 * know that 'from' will not move past 'to' and vice versa. */
2688 resort_bals(from, bals, n_bals);
2689 resort_bals(to, bals, n_bals);
2696 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2697 * historical data to decay to <1% in 7 rebalancing runs. */
2698 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2704 bond_send_learning_packets(struct port *port)
2706 struct bridge *br = port->bridge;
2707 struct mac_entry *e;
2708 struct ofpbuf packet;
2709 int error, n_packets, n_errors;
2711 if (!port->n_ifaces || port->active_iface < 0) {
2715 ofpbuf_init(&packet, 128);
2716 error = n_packets = n_errors = 0;
2717 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2718 union ofp_action actions[2], *a;
2724 if (e->port == port->port_idx
2725 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2729 /* Compose actions. */
2730 memset(actions, 0, sizeof actions);
2733 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2734 a->vlan_vid.len = htons(sizeof *a);
2735 a->vlan_vid.vlan_vid = htons(e->vlan);
2738 a->output.type = htons(OFPAT_OUTPUT);
2739 a->output.len = htons(sizeof *a);
2740 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2745 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2747 flow_extract(&packet, 0, ODPP_NONE, &flow);
2748 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2755 ofpbuf_uninit(&packet);
2758 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2759 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2760 "packets, last error was: %s",
2761 port->name, n_errors, n_packets, strerror(error));
2763 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2764 port->name, n_packets);
2768 /* Bonding unixctl user interface functions. */
2771 bond_unixctl_list(struct unixctl_conn *conn,
2772 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2774 struct ds ds = DS_EMPTY_INITIALIZER;
2775 const struct bridge *br;
2777 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2779 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2782 for (i = 0; i < br->n_ports; i++) {
2783 const struct port *port = br->ports[i];
2784 if (port->n_ifaces > 1) {
2787 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2788 for (j = 0; j < port->n_ifaces; j++) {
2789 const struct iface *iface = port->ifaces[j];
2791 ds_put_cstr(&ds, ", ");
2793 ds_put_cstr(&ds, iface->name);
2795 ds_put_char(&ds, '\n');
2799 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2803 static struct port *
2804 bond_find(const char *name)
2806 const struct bridge *br;
2808 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2811 for (i = 0; i < br->n_ports; i++) {
2812 struct port *port = br->ports[i];
2813 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2822 bond_unixctl_show(struct unixctl_conn *conn,
2823 const char *args, void *aux OVS_UNUSED)
2825 struct ds ds = DS_EMPTY_INITIALIZER;
2826 const struct port *port;
2829 port = bond_find(args);
2831 unixctl_command_reply(conn, 501, "no such bond");
2835 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2836 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2837 ds_put_format(&ds, "next rebalance: %lld ms\n",
2838 port->bond_next_rebalance - time_msec());
2839 for (j = 0; j < port->n_ifaces; j++) {
2840 const struct iface *iface = port->ifaces[j];
2841 struct bond_entry *be;
2844 ds_put_format(&ds, "slave %s: %s\n",
2845 iface->name, iface->enabled ? "enabled" : "disabled");
2846 if (j == port->active_iface) {
2847 ds_put_cstr(&ds, "\tactive slave\n");
2849 if (iface->delay_expires != LLONG_MAX) {
2850 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2851 iface->enabled ? "downdelay" : "updelay",
2852 iface->delay_expires - time_msec());
2856 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2857 int hash = be - port->bond_hash;
2858 struct mac_entry *me;
2860 if (be->iface_idx != j) {
2864 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2865 hash, be->tx_bytes / 1024);
2868 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2869 &port->bridge->ml->lrus) {
2872 if (bond_hash(me->mac) == hash
2873 && me->port != port->port_idx
2874 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2875 && dp_ifidx == iface->dp_ifidx)
2877 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2878 ETH_ADDR_ARGS(me->mac));
2883 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2888 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2889 void *aux OVS_UNUSED)
2891 char *args = (char *) args_;
2892 char *save_ptr = NULL;
2893 char *bond_s, *hash_s, *slave_s;
2894 uint8_t mac[ETH_ADDR_LEN];
2896 struct iface *iface;
2897 struct bond_entry *entry;
2900 bond_s = strtok_r(args, " ", &save_ptr);
2901 hash_s = strtok_r(NULL, " ", &save_ptr);
2902 slave_s = strtok_r(NULL, " ", &save_ptr);
2904 unixctl_command_reply(conn, 501,
2905 "usage: bond/migrate BOND HASH SLAVE");
2909 port = bond_find(bond_s);
2911 unixctl_command_reply(conn, 501, "no such bond");
2915 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2916 == ETH_ADDR_SCAN_COUNT) {
2917 hash = bond_hash(mac);
2918 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2919 hash = atoi(hash_s) & BOND_MASK;
2921 unixctl_command_reply(conn, 501, "bad hash");
2925 iface = port_lookup_iface(port, slave_s);
2927 unixctl_command_reply(conn, 501, "no such slave");
2931 if (!iface->enabled) {
2932 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2936 entry = &port->bond_hash[hash];
2937 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2938 entry->iface_idx = iface->port_ifidx;
2939 entry->iface_tag = tag_create_random();
2940 port->bond_compat_is_stale = true;
2941 unixctl_command_reply(conn, 200, "migrated");
2945 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2946 void *aux OVS_UNUSED)
2948 char *args = (char *) args_;
2949 char *save_ptr = NULL;
2950 char *bond_s, *slave_s;
2952 struct iface *iface;
2954 bond_s = strtok_r(args, " ", &save_ptr);
2955 slave_s = strtok_r(NULL, " ", &save_ptr);
2957 unixctl_command_reply(conn, 501,
2958 "usage: bond/set-active-slave BOND SLAVE");
2962 port = bond_find(bond_s);
2964 unixctl_command_reply(conn, 501, "no such bond");
2968 iface = port_lookup_iface(port, slave_s);
2970 unixctl_command_reply(conn, 501, "no such slave");
2974 if (!iface->enabled) {
2975 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2979 if (port->active_iface != iface->port_ifidx) {
2980 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2981 port->active_iface = iface->port_ifidx;
2982 port->active_iface_tag = tag_create_random();
2983 VLOG_INFO("port %s: active interface is now %s",
2984 port->name, iface->name);
2985 bond_send_learning_packets(port);
2986 unixctl_command_reply(conn, 200, "done");
2988 unixctl_command_reply(conn, 200, "no change");
2993 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2995 char *args = (char *) args_;
2996 char *save_ptr = NULL;
2997 char *bond_s, *slave_s;
2999 struct iface *iface;
3001 bond_s = strtok_r(args, " ", &save_ptr);
3002 slave_s = strtok_r(NULL, " ", &save_ptr);
3004 unixctl_command_reply(conn, 501,
3005 "usage: bond/enable/disable-slave BOND SLAVE");
3009 port = bond_find(bond_s);
3011 unixctl_command_reply(conn, 501, "no such bond");
3015 iface = port_lookup_iface(port, slave_s);
3017 unixctl_command_reply(conn, 501, "no such slave");
3021 bond_enable_slave(iface, enable);
3022 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3026 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3027 void *aux OVS_UNUSED)
3029 enable_slave(conn, args, true);
3033 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3034 void *aux OVS_UNUSED)
3036 enable_slave(conn, args, false);
3040 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3041 void *aux OVS_UNUSED)
3043 uint8_t mac[ETH_ADDR_LEN];
3047 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3048 == ETH_ADDR_SCAN_COUNT) {
3049 hash = bond_hash(mac);
3051 hash_cstr = xasprintf("%u", hash);
3052 unixctl_command_reply(conn, 200, hash_cstr);
3055 unixctl_command_reply(conn, 501, "invalid mac");
3062 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3063 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3064 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3065 unixctl_command_register("bond/set-active-slave",
3066 bond_unixctl_set_active_slave, NULL);
3067 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3069 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3071 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3074 /* Port functions. */
3076 static struct port *
3077 port_create(struct bridge *br, const char *name)
3081 port = xzalloc(sizeof *port);
3083 port->port_idx = br->n_ports;
3085 port->trunks = NULL;
3086 port->name = xstrdup(name);
3087 port->active_iface = -1;
3089 if (br->n_ports >= br->allocated_ports) {
3090 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3093 br->ports[br->n_ports++] = port;
3094 shash_add_assert(&br->port_by_name, port->name, port);
3096 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3103 get_port_other_config(const struct ovsrec_port *port, const char *key,
3104 const char *default_value)
3106 const char *value = get_ovsrec_key_value(key,
3107 port->key_other_config,
3108 port->value_other_config,
3109 port->n_other_config);
3110 return value ? value : default_value;
3114 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3116 struct shash new_ifaces;
3119 /* Collect list of new interfaces. */
3120 shash_init(&new_ifaces);
3121 for (i = 0; i < cfg->n_interfaces; i++) {
3122 const char *name = cfg->interfaces[i]->name;
3123 shash_add_once(&new_ifaces, name, NULL);
3126 /* Get rid of deleted interfaces. */
3127 for (i = 0; i < port->n_ifaces; ) {
3128 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3129 iface_destroy(port->ifaces[i]);
3135 shash_destroy(&new_ifaces);
3139 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3141 struct shash new_ifaces;
3142 long long int next_rebalance;
3143 unsigned long *trunks;
3149 /* Update settings. */
3150 port->updelay = cfg->bond_updelay;
3151 if (port->updelay < 0) {
3154 port->updelay = cfg->bond_downdelay;
3155 if (port->downdelay < 0) {
3156 port->downdelay = 0;
3158 port->bond_rebalance_interval = atoi(
3159 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3160 if (port->bond_rebalance_interval < 1000) {
3161 port->bond_rebalance_interval = 1000;
3163 next_rebalance = time_msec() + port->bond_rebalance_interval;
3164 if (port->bond_next_rebalance > next_rebalance) {
3165 port->bond_next_rebalance = next_rebalance;
3168 /* Add new interfaces and update 'cfg' member of existing ones. */
3169 shash_init(&new_ifaces);
3170 for (i = 0; i < cfg->n_interfaces; i++) {
3171 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3172 struct iface *iface;
3174 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3175 VLOG_WARN("port %s: %s specified twice as port interface",
3176 port->name, if_cfg->name);
3180 iface = iface_lookup(port->bridge, if_cfg->name);
3182 if (iface->port != port) {
3183 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3185 port->bridge->name, if_cfg->name, iface->port->name);
3188 iface->cfg = if_cfg;
3190 iface_create(port, if_cfg);
3193 shash_destroy(&new_ifaces);
3198 if (port->n_ifaces < 2) {
3200 if (vlan >= 0 && vlan <= 4095) {
3201 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3206 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3207 * they even work as-is. But they have not been tested. */
3208 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3212 if (port->vlan != vlan) {
3214 bridge_flush(port->bridge);
3217 /* Get trunked VLANs. */
3219 if (vlan < 0 && cfg->n_trunks) {
3223 trunks = bitmap_allocate(4096);
3225 for (i = 0; i < cfg->n_trunks; i++) {
3226 int trunk = cfg->trunks[i];
3228 bitmap_set1(trunks, trunk);
3234 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3235 port->name, cfg->n_trunks);
3237 if (n_errors == cfg->n_trunks) {
3238 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3240 bitmap_free(trunks);
3243 } else if (vlan >= 0 && cfg->n_trunks) {
3244 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3248 ? port->trunks != NULL
3249 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3250 bridge_flush(port->bridge);
3252 bitmap_free(port->trunks);
3253 port->trunks = trunks;
3257 port_destroy(struct port *port)
3260 struct bridge *br = port->bridge;
3264 proc_net_compat_update_vlan(port->name, NULL, 0);
3265 proc_net_compat_update_bond(port->name, NULL);
3267 for (i = 0; i < MAX_MIRRORS; i++) {
3268 struct mirror *m = br->mirrors[i];
3269 if (m && m->out_port == port) {
3274 while (port->n_ifaces > 0) {
3275 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3278 shash_find_and_delete_assert(&br->port_by_name, port->name);
3280 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3281 del->port_idx = port->port_idx;
3284 bitmap_free(port->trunks);
3291 static struct port *
3292 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3294 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3295 return iface ? iface->port : NULL;
3298 static struct port *
3299 port_lookup(const struct bridge *br, const char *name)
3301 return shash_find_data(&br->port_by_name, name);
3304 static struct iface *
3305 port_lookup_iface(const struct port *port, const char *name)
3307 struct iface *iface = iface_lookup(port->bridge, name);
3308 return iface && iface->port == port ? iface : NULL;
3312 port_update_bonding(struct port *port)
3314 if (port->n_ifaces < 2) {
3315 /* Not a bonded port. */
3316 if (port->bond_hash) {
3317 free(port->bond_hash);
3318 port->bond_hash = NULL;
3319 port->bond_compat_is_stale = true;
3320 port->bond_fake_iface = false;
3323 if (!port->bond_hash) {
3326 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3327 for (i = 0; i <= BOND_MASK; i++) {
3328 struct bond_entry *e = &port->bond_hash[i];
3332 port->no_ifaces_tag = tag_create_random();
3333 bond_choose_active_iface(port);
3334 port->bond_next_rebalance
3335 = time_msec() + port->bond_rebalance_interval;
3337 if (port->cfg->bond_fake_iface) {
3338 port->bond_next_fake_iface_update = time_msec();
3341 port->bond_compat_is_stale = true;
3342 port->bond_fake_iface = port->cfg->bond_fake_iface;
3347 port_update_bond_compat(struct port *port)
3349 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3350 struct compat_bond bond;
3353 if (port->n_ifaces < 2) {
3354 proc_net_compat_update_bond(port->name, NULL);
3359 bond.updelay = port->updelay;
3360 bond.downdelay = port->downdelay;
3363 bond.hashes = compat_hashes;
3364 if (port->bond_hash) {
3365 const struct bond_entry *e;
3366 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3367 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3368 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3369 cbh->hash = e - port->bond_hash;
3370 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3375 bond.n_slaves = port->n_ifaces;
3376 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3377 for (i = 0; i < port->n_ifaces; i++) {
3378 struct iface *iface = port->ifaces[i];
3379 struct compat_bond_slave *slave = &bond.slaves[i];
3380 slave->name = iface->name;
3382 /* We need to make the same determination as the Linux bonding
3383 * code to determine whether a slave should be consider "up".
3384 * The Linux function bond_miimon_inspect() supports four
3385 * BOND_LINK_* states:
3387 * - BOND_LINK_UP: carrier detected, updelay has passed.
3388 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3389 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3390 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3392 * The function bond_info_show_slave() only considers BOND_LINK_UP
3393 * to be "up" and anything else to be "down".
3395 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3399 netdev_get_etheraddr(iface->netdev, slave->mac);
3402 if (port->bond_fake_iface) {
3403 struct netdev *bond_netdev;
3405 if (!netdev_open_default(port->name, &bond_netdev)) {
3407 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3409 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3411 netdev_close(bond_netdev);
3415 proc_net_compat_update_bond(port->name, &bond);
3420 port_update_vlan_compat(struct port *port)
3422 struct bridge *br = port->bridge;
3423 char *vlandev_name = NULL;
3425 if (port->vlan > 0) {
3426 /* Figure out the name that the VLAN device should actually have, if it
3427 * existed. This takes some work because the VLAN device would not
3428 * have port->name in its name; rather, it would have the trunk port's
3429 * name, and 'port' would be attached to a bridge that also had the
3430 * VLAN device one of its ports. So we need to find a trunk port that
3431 * includes port->vlan.
3433 * There might be more than one candidate. This doesn't happen on
3434 * XenServer, so if it happens we just pick the first choice in
3435 * alphabetical order instead of creating multiple VLAN devices. */
3437 for (i = 0; i < br->n_ports; i++) {
3438 struct port *p = br->ports[i];
3439 if (port_trunks_vlan(p, port->vlan)
3441 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3443 uint8_t ea[ETH_ADDR_LEN];
3444 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3445 if (!eth_addr_is_multicast(ea) &&
3446 !eth_addr_is_reserved(ea) &&
3447 !eth_addr_is_zero(ea)) {
3448 vlandev_name = p->name;
3453 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3456 /* Interface functions. */
3458 static struct iface *
3459 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3461 struct bridge *br = port->bridge;
3462 struct iface *iface;
3463 char *name = if_cfg->name;
3466 iface = xzalloc(sizeof *iface);
3468 iface->port_ifidx = port->n_ifaces;
3469 iface->name = xstrdup(name);
3470 iface->dp_ifidx = -1;
3471 iface->tag = tag_create_random();
3472 iface->delay_expires = LLONG_MAX;
3473 iface->netdev = NULL;
3474 iface->cfg = if_cfg;
3476 shash_add_assert(&br->iface_by_name, iface->name, iface);
3478 /* Attempt to create the network interface in case it doesn't exist yet. */
3479 if (!iface_is_internal(br, iface->name)) {
3480 error = set_up_iface(if_cfg, iface, true);
3482 VLOG_WARN("could not create iface %s: %s", iface->name,
3485 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3492 if (port->n_ifaces >= port->allocated_ifaces) {
3493 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3494 sizeof *port->ifaces);
3496 port->ifaces[port->n_ifaces++] = iface;
3497 if (port->n_ifaces > 1) {
3498 br->has_bonded_ports = true;
3501 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3509 iface_destroy(struct iface *iface)
3512 struct port *port = iface->port;
3513 struct bridge *br = port->bridge;
3514 bool del_active = port->active_iface == iface->port_ifidx;
3517 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3519 if (iface->dp_ifidx >= 0) {
3520 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3523 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3524 del->port_ifidx = iface->port_ifidx;
3526 netdev_close(iface->netdev);
3529 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3530 bond_choose_active_iface(port);
3531 bond_send_learning_packets(port);
3537 bridge_flush(port->bridge);
3541 static struct iface *
3542 iface_lookup(const struct bridge *br, const char *name)
3544 return shash_find_data(&br->iface_by_name, name);
3547 static struct iface *
3548 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3550 return port_array_get(&br->ifaces, dp_ifidx);
3553 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3554 * 'br', that is, an interface that is entirely simulated within the datapath.
3555 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3556 * interfaces are created by setting "iface.<iface>.internal = true".
3558 * In addition, we have a kluge-y feature that creates an internal port with
3559 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3560 * This feature needs to go away in the long term. Until then, this is one
3561 * reason why this function takes a name instead of a struct iface: the fake
3562 * interfaces created this way do not have a struct iface. */
3564 iface_is_internal(const struct bridge *br, const char *if_name)
3566 struct iface *iface;
3569 if (!strcmp(if_name, br->name)) {
3573 iface = iface_lookup(br, if_name);
3574 if (iface && !strcmp(iface->cfg->type, "internal")) {
3578 port = port_lookup(br, if_name);
3579 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3585 /* Set Ethernet address of 'iface', if one is specified in the configuration
3588 iface_set_mac(struct iface *iface)
3590 uint8_t ea[ETH_ADDR_LEN];
3592 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3593 if (eth_addr_is_multicast(ea)) {
3594 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3596 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3597 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3598 iface->name, iface->name);
3600 int error = netdev_set_etheraddr(iface->netdev, ea);
3602 VLOG_ERR("interface %s: setting MAC failed (%s)",
3603 iface->name, strerror(error));
3610 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3611 struct shash *shash)
3616 for (i = 0; i < n; i++) {
3617 shash_add(shash, keys[i], values[i]);
3621 struct iface_delete_queues_cbdata {
3622 struct netdev *netdev;
3623 const int64_t *queue_ids;
3628 queue_ids_include(const int64_t *ids, size_t n, int64_t target)
3633 while (low < high) {
3634 size_t mid = low + (high - low) / 2;
3635 if (target > ids[mid]) {
3637 } else if (target < ids[mid]) {
3647 iface_delete_queues(unsigned int queue_id,
3648 const struct shash *details OVS_UNUSED, void *cbdata_)
3650 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3652 if (!queue_ids_include(cbdata->queue_ids, cbdata->n_queue_ids, queue_id)) {
3653 netdev_delete_queue(cbdata->netdev, queue_id);
3658 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3660 if (!qos || qos->type[0] == '\0') {
3661 netdev_set_qos(iface->netdev, NULL, NULL);
3663 struct iface_delete_queues_cbdata cbdata;
3664 struct shash details;
3667 /* Configure top-level Qos for 'iface'. */
3668 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3669 qos->n_other_config, &details);
3670 netdev_set_qos(iface->netdev, qos->type, &details);
3671 shash_destroy(&details);
3673 /* Deconfigure queues that were deleted. */
3674 cbdata.netdev = iface->netdev;
3675 cbdata.queue_ids = qos->key_queues;
3676 cbdata.n_queue_ids = qos->n_queues;
3677 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3679 /* Configure queues for 'iface'. */
3680 for (i = 0; i < qos->n_queues; i++) {
3681 const struct ovsrec_queue *queue = qos->value_queues[i];
3682 unsigned int queue_id = qos->key_queues[i];
3684 shash_from_ovs_idl_map(queue->key_other_config,
3685 queue->value_other_config,
3686 queue->n_other_config, &details);
3687 netdev_set_queue(iface->netdev, queue_id, &details);
3688 shash_destroy(&details);
3693 /* Port mirroring. */
3696 mirror_reconfigure(struct bridge *br)
3698 struct shash old_mirrors, new_mirrors;
3699 struct shash_node *node;
3700 unsigned long *rspan_vlans;
3703 /* Collect old mirrors. */
3704 shash_init(&old_mirrors);
3705 for (i = 0; i < MAX_MIRRORS; i++) {
3706 if (br->mirrors[i]) {
3707 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3711 /* Collect new mirrors. */
3712 shash_init(&new_mirrors);
3713 for (i = 0; i < br->cfg->n_mirrors; i++) {
3714 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3715 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3716 VLOG_WARN("bridge %s: %s specified twice as mirror",
3717 br->name, cfg->name);
3721 /* Get rid of deleted mirrors and add new mirrors. */
3722 SHASH_FOR_EACH (node, &old_mirrors) {
3723 if (!shash_find(&new_mirrors, node->name)) {
3724 mirror_destroy(node->data);
3727 SHASH_FOR_EACH (node, &new_mirrors) {
3728 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3730 mirror = mirror_create(br, node->name);
3735 mirror_reconfigure_one(mirror, node->data);
3737 shash_destroy(&old_mirrors);
3738 shash_destroy(&new_mirrors);
3740 /* Update port reserved status. */
3741 for (i = 0; i < br->n_ports; i++) {
3742 br->ports[i]->is_mirror_output_port = false;
3744 for (i = 0; i < MAX_MIRRORS; i++) {
3745 struct mirror *m = br->mirrors[i];
3746 if (m && m->out_port) {
3747 m->out_port->is_mirror_output_port = true;
3751 /* Update flooded vlans (for RSPAN). */
3753 if (br->cfg->n_flood_vlans) {
3754 rspan_vlans = bitmap_allocate(4096);
3756 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3757 int64_t vlan = br->cfg->flood_vlans[i];
3758 if (vlan >= 0 && vlan < 4096) {
3759 bitmap_set1(rspan_vlans, vlan);
3760 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3763 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3768 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3773 static struct mirror *
3774 mirror_create(struct bridge *br, const char *name)
3779 for (i = 0; ; i++) {
3780 if (i >= MAX_MIRRORS) {
3781 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3782 "cannot create %s", br->name, MAX_MIRRORS, name);
3785 if (!br->mirrors[i]) {
3790 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3793 br->mirrors[i] = m = xzalloc(sizeof *m);
3796 m->name = xstrdup(name);
3797 shash_init(&m->src_ports);
3798 shash_init(&m->dst_ports);
3808 mirror_destroy(struct mirror *m)
3811 struct bridge *br = m->bridge;
3814 for (i = 0; i < br->n_ports; i++) {
3815 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3816 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3819 shash_destroy(&m->src_ports);
3820 shash_destroy(&m->dst_ports);
3823 m->bridge->mirrors[m->idx] = NULL;
3831 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3832 struct shash *names)
3836 for (i = 0; i < n_ports; i++) {
3837 const char *name = ports[i]->name;
3838 if (port_lookup(m->bridge, name)) {
3839 shash_add_once(names, name, NULL);
3841 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3842 "port %s", m->bridge->name, m->name, name);
3848 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3854 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3856 for (i = 0; i < cfg->n_select_vlan; i++) {
3857 int64_t vlan = cfg->select_vlan[i];
3858 if (vlan < 0 || vlan > 4095) {
3859 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3860 m->bridge->name, m->name, vlan);
3862 (*vlans)[n_vlans++] = vlan;
3869 vlan_is_mirrored(const struct mirror *m, int vlan)
3873 for (i = 0; i < m->n_vlans; i++) {
3874 if (m->vlans[i] == vlan) {
3882 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3886 for (i = 0; i < m->n_vlans; i++) {
3887 if (port_trunks_vlan(p, m->vlans[i])) {
3895 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3897 struct shash src_ports, dst_ports;
3898 mirror_mask_t mirror_bit;
3899 struct port *out_port;
3905 /* Get output port. */
3906 if (cfg->output_port) {
3907 out_port = port_lookup(m->bridge, cfg->output_port->name);
3909 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3910 m->bridge->name, m->name);
3916 if (cfg->output_vlan) {
3917 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3918 "output vlan; ignoring output vlan",
3919 m->bridge->name, m->name);
3921 } else if (cfg->output_vlan) {
3923 out_vlan = *cfg->output_vlan;
3925 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3926 m->bridge->name, m->name);
3931 shash_init(&src_ports);
3932 shash_init(&dst_ports);
3933 if (cfg->select_all) {
3934 for (i = 0; i < m->bridge->n_ports; i++) {
3935 const char *name = m->bridge->ports[i]->name;
3936 shash_add_once(&src_ports, name, NULL);
3937 shash_add_once(&dst_ports, name, NULL);
3942 /* Get ports, and drop duplicates and ports that don't exist. */
3943 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3945 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3948 /* Get all the vlans, and drop duplicate and invalid vlans. */
3949 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3952 /* Update mirror data. */
3953 if (!shash_equal_keys(&m->src_ports, &src_ports)
3954 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3955 || m->n_vlans != n_vlans
3956 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3957 || m->out_port != out_port
3958 || m->out_vlan != out_vlan) {
3959 bridge_flush(m->bridge);
3961 shash_swap(&m->src_ports, &src_ports);
3962 shash_swap(&m->dst_ports, &dst_ports);
3965 m->n_vlans = n_vlans;
3966 m->out_port = out_port;
3967 m->out_vlan = out_vlan;
3970 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3971 for (i = 0; i < m->bridge->n_ports; i++) {
3972 struct port *port = m->bridge->ports[i];
3974 if (shash_find(&m->src_ports, port->name)
3977 ? port_trunks_any_mirrored_vlan(m, port)
3978 : vlan_is_mirrored(m, port->vlan)))) {
3979 port->src_mirrors |= mirror_bit;
3981 port->src_mirrors &= ~mirror_bit;
3984 if (shash_find(&m->dst_ports, port->name)) {
3985 port->dst_mirrors |= mirror_bit;
3987 port->dst_mirrors &= ~mirror_bit;
3992 shash_destroy(&src_ports);
3993 shash_destroy(&dst_ports);