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 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
193 static void bridge_destroy(struct bridge *);
194 static struct bridge *bridge_lookup(const char *name);
195 static unixctl_cb_func bridge_unixctl_dump_flows;
196 static int bridge_run_one(struct bridge *);
197 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
198 const struct bridge *br,
199 struct ovsrec_controller ***controllersp);
200 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
202 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
204 const struct sockaddr_in *managers,
206 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
207 static void bridge_fetch_dp_ifaces(struct bridge *);
208 static void bridge_flush(struct bridge *);
209 static void bridge_pick_local_hw_addr(struct bridge *,
210 uint8_t ea[ETH_ADDR_LEN],
211 struct iface **hw_addr_iface);
212 static uint64_t bridge_pick_datapath_id(struct bridge *,
213 const uint8_t bridge_ea[ETH_ADDR_LEN],
214 struct iface *hw_addr_iface);
215 static struct iface *bridge_get_local_iface(struct bridge *);
216 static uint64_t dpid_from_hash(const void *, size_t nbytes);
218 static unixctl_cb_func bridge_unixctl_fdb_show;
220 static void bond_init(void);
221 static void bond_run(struct bridge *);
222 static void bond_wait(struct bridge *);
223 static void bond_rebalance_port(struct port *);
224 static void bond_send_learning_packets(struct port *);
225 static void bond_enable_slave(struct iface *iface, bool enable);
227 static struct port *port_create(struct bridge *, const char *name);
228 static void port_reconfigure(struct port *, const struct ovsrec_port *);
229 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
230 static void port_destroy(struct port *);
231 static struct port *port_lookup(const struct bridge *, const char *name);
232 static struct iface *port_lookup_iface(const struct port *, const char *name);
233 static struct port *port_from_dp_ifidx(const struct bridge *,
235 static void port_update_bond_compat(struct port *);
236 static void port_update_vlan_compat(struct port *);
237 static void port_update_bonding(struct port *);
239 static struct mirror *mirror_create(struct bridge *, const char *name);
240 static void mirror_destroy(struct mirror *);
241 static void mirror_reconfigure(struct bridge *);
242 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
243 static bool vlan_is_mirrored(const struct mirror *, int vlan);
245 static struct iface *iface_create(struct port *port,
246 const struct ovsrec_interface *if_cfg);
247 static void iface_destroy(struct iface *);
248 static struct iface *iface_lookup(const struct bridge *, const char *name);
249 static struct iface *iface_from_dp_ifidx(const struct bridge *,
251 static bool iface_is_internal(const struct bridge *, const char *name);
252 static void iface_set_mac(struct iface *);
253 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
255 /* Hooks into ofproto processing. */
256 static struct ofhooks bridge_ofhooks;
258 /* Public functions. */
261 bridge_init(const struct ovsrec_open_vswitch *cfg)
263 struct svec bridge_names;
264 struct svec dpif_names, dpif_types;
267 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
269 svec_init(&bridge_names);
270 for (i = 0; i < cfg->n_bridges; i++) {
271 svec_add(&bridge_names, cfg->bridges[i]->name);
273 svec_sort(&bridge_names);
275 svec_init(&dpif_names);
276 svec_init(&dpif_types);
277 dp_enumerate_types(&dpif_types);
278 for (i = 0; i < dpif_types.n; i++) {
283 dp_enumerate_names(dpif_types.names[i], &dpif_names);
285 for (j = 0; j < dpif_names.n; j++) {
286 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
288 struct svec all_names;
291 svec_init(&all_names);
292 dpif_get_all_names(dpif, &all_names);
293 for (k = 0; k < all_names.n; k++) {
294 if (svec_contains(&bridge_names, all_names.names[k])) {
300 svec_destroy(&all_names);
305 svec_destroy(&bridge_names);
306 svec_destroy(&dpif_names);
307 svec_destroy(&dpif_types);
309 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
313 bridge_reconfigure(cfg);
318 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
320 /* XXX SSL should be configurable on a per-bridge basis. */
322 stream_ssl_set_private_key_file(ssl->private_key);
323 stream_ssl_set_certificate_file(ssl->certificate);
324 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
329 /* Attempt to create the network device 'iface_name' through the netdev
332 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
335 struct shash options;
339 shash_init(&options);
340 for (i = 0; i < iface_cfg->n_options; i++) {
341 shash_add(&options, iface_cfg->key_options[i],
342 xstrdup(iface_cfg->value_options[i]));
346 struct netdev_options netdev_options;
348 memset(&netdev_options, 0, sizeof netdev_options);
349 netdev_options.name = iface_cfg->name;
350 if (!strcmp(iface_cfg->type, "internal")) {
351 /* An "internal" config type maps to a netdev "system" type. */
352 netdev_options.type = "system";
354 netdev_options.type = iface_cfg->type;
356 netdev_options.args = &options;
357 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
359 error = netdev_open(&netdev_options, &iface->netdev);
362 netdev_get_carrier(iface->netdev, &iface->enabled);
364 } else if (iface->netdev) {
365 const char *netdev_type = netdev_get_type(iface->netdev);
366 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
367 ? iface_cfg->type : NULL;
369 /* An "internal" config type maps to a netdev "system" type. */
370 if (iface_type && !strcmp(iface_type, "internal")) {
371 iface_type = "system";
374 if (!iface_type || !strcmp(netdev_type, iface_type)) {
375 error = netdev_reconfigure(iface->netdev, &options);
377 VLOG_WARN("%s: attempting change device type from %s to %s",
378 iface_cfg->name, netdev_type, iface_type);
382 shash_destroy_free_data(&options);
388 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
390 return set_up_iface(iface_cfg, iface, false);
394 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
395 void *aux OVS_UNUSED)
397 if (!iface->netdev) {
398 int error = set_up_iface(iface->cfg, iface, true);
400 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
410 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
411 void *aux OVS_UNUSED)
413 if (iface->dp_ifidx >= 0) {
414 VLOG_DBG("%s has interface %s on port %d",
416 iface->name, iface->dp_ifidx);
419 VLOG_ERR("%s interface not in %s, dropping",
420 iface->name, dpif_name(br->dpif));
426 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
427 void *aux OVS_UNUSED)
429 /* Set policing attributes. */
430 netdev_set_policing(iface->netdev,
431 iface->cfg->ingress_policing_rate,
432 iface->cfg->ingress_policing_burst);
434 /* Set MAC address of internal interfaces other than the local
436 if (iface->dp_ifidx != ODPP_LOCAL
437 && iface_is_internal(br, iface->name)) {
438 iface_set_mac(iface);
444 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
445 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
446 * deletes from 'br' any ports that no longer have any interfaces. */
448 iterate_and_prune_ifaces(struct bridge *br,
449 bool (*cb)(struct bridge *, struct iface *,
455 for (i = 0; i < br->n_ports; ) {
456 struct port *port = br->ports[i];
457 for (j = 0; j < port->n_ifaces; ) {
458 struct iface *iface = port->ifaces[j];
459 if (cb(br, iface, aux)) {
462 iface_destroy(iface);
466 if (port->n_ifaces) {
469 VLOG_ERR("%s port has no interfaces, dropping", port->name);
475 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
476 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
477 * responsible for freeing '*managersp' (with free()).
479 * You may be asking yourself "why does ovs-vswitchd care?", because
480 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
481 * should not be and in fact is not directly involved in that. But
482 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
483 * it has to tell in-band control where the managers are to enable that.
486 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
487 struct sockaddr_in **managersp, size_t *n_managersp)
489 struct sockaddr_in *managers = NULL;
490 size_t n_managers = 0;
492 if (ovs_cfg->n_managers > 0) {
495 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
496 for (i = 0; i < ovs_cfg->n_managers; i++) {
497 const char *name = ovs_cfg->managers[i];
498 struct sockaddr_in *sin = &managers[i];
500 if ((!strncmp(name, "tcp:", 4)
501 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
502 (!strncmp(name, "ssl:", 4)
503 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
509 *managersp = managers;
510 *n_managersp = n_managers;
514 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
516 struct ovsdb_idl_txn *txn;
517 struct shash old_br, new_br;
518 struct shash_node *node;
519 struct bridge *br, *next;
520 struct sockaddr_in *managers;
523 int sflow_bridge_number;
525 COVERAGE_INC(bridge_reconfigure);
527 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
529 collect_managers(ovs_cfg, &managers, &n_managers);
531 /* Collect old and new bridges. */
534 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
535 shash_add(&old_br, br->name, br);
537 for (i = 0; i < ovs_cfg->n_bridges; i++) {
538 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
539 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
540 VLOG_WARN("more than one bridge named %s", br_cfg->name);
544 /* Get rid of deleted bridges and add new bridges. */
545 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
546 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
553 SHASH_FOR_EACH (node, &new_br) {
554 const char *br_name = node->name;
555 const struct ovsrec_bridge *br_cfg = node->data;
556 br = shash_find_data(&old_br, br_name);
558 /* If the bridge datapath type has changed, we need to tear it
559 * down and recreate. */
560 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
562 bridge_create(br_cfg);
565 bridge_create(br_cfg);
568 shash_destroy(&old_br);
569 shash_destroy(&new_br);
573 bridge_configure_ssl(ovs_cfg->ssl);
576 /* Reconfigure all bridges. */
577 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
578 bridge_reconfigure_one(ovs_cfg, br);
581 /* Add and delete ports on all datapaths.
583 * The kernel will reject any attempt to add a given port to a datapath if
584 * that port already belongs to a different datapath, so we must do all
585 * port deletions before any port additions. */
586 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
587 struct odp_port *dpif_ports;
589 struct shash want_ifaces;
591 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
592 bridge_get_all_ifaces(br, &want_ifaces);
593 for (i = 0; i < n_dpif_ports; i++) {
594 const struct odp_port *p = &dpif_ports[i];
595 if (!shash_find(&want_ifaces, p->devname)
596 && strcmp(p->devname, br->name)) {
597 int retval = dpif_port_del(br->dpif, p->port);
599 VLOG_ERR("failed to remove %s interface from %s: %s",
600 p->devname, dpif_name(br->dpif),
605 shash_destroy(&want_ifaces);
608 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
609 struct odp_port *dpif_ports;
611 struct shash cur_ifaces, want_ifaces;
612 struct shash_node *node;
614 /* Get the set of interfaces currently in this datapath. */
615 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
616 shash_init(&cur_ifaces);
617 for (i = 0; i < n_dpif_ports; i++) {
618 const char *name = dpif_ports[i].devname;
619 if (!shash_find(&cur_ifaces, name)) {
620 shash_add(&cur_ifaces, name, NULL);
625 /* Get the set of interfaces we want on this datapath. */
626 bridge_get_all_ifaces(br, &want_ifaces);
628 SHASH_FOR_EACH (node, &want_ifaces) {
629 const char *if_name = node->name;
630 struct iface *iface = node->data;
632 if (shash_find(&cur_ifaces, if_name)) {
633 /* Already exists, just reconfigure it. */
635 reconfigure_iface(iface->cfg, iface);
638 /* Need to add to datapath. */
642 /* Add to datapath. */
643 internal = iface_is_internal(br, if_name);
644 error = dpif_port_add(br->dpif, if_name,
645 internal ? ODP_PORT_INTERNAL : 0, NULL);
646 if (error == EFBIG) {
647 VLOG_ERR("ran out of valid port numbers on %s",
648 dpif_name(br->dpif));
651 VLOG_ERR("failed to add %s interface to %s: %s",
652 if_name, dpif_name(br->dpif), strerror(error));
656 shash_destroy(&cur_ifaces);
657 shash_destroy(&want_ifaces);
659 sflow_bridge_number = 0;
660 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
663 struct iface *local_iface;
664 struct iface *hw_addr_iface;
667 bridge_fetch_dp_ifaces(br);
669 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
670 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
672 /* Pick local port hardware address, datapath ID. */
673 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
674 local_iface = bridge_get_local_iface(br);
676 int error = netdev_set_etheraddr(local_iface->netdev, ea);
678 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
679 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
680 "Ethernet address: %s",
681 br->name, strerror(error));
685 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
686 ofproto_set_datapath_id(br->ofproto, dpid);
688 dpid_string = xasprintf("%016"PRIx64, dpid);
689 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
692 /* Set NetFlow configuration on this bridge. */
693 if (br->cfg->netflow) {
694 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
695 struct netflow_options opts;
697 memset(&opts, 0, sizeof opts);
699 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
700 if (nf_cfg->engine_type) {
701 opts.engine_type = *nf_cfg->engine_type;
703 if (nf_cfg->engine_id) {
704 opts.engine_id = *nf_cfg->engine_id;
707 opts.active_timeout = nf_cfg->active_timeout;
708 if (!opts.active_timeout) {
709 opts.active_timeout = -1;
710 } else if (opts.active_timeout < 0) {
711 VLOG_WARN("bridge %s: active timeout interval set to negative "
712 "value, using default instead (%d seconds)", br->name,
713 NF_ACTIVE_TIMEOUT_DEFAULT);
714 opts.active_timeout = -1;
717 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
718 if (opts.add_id_to_iface) {
719 if (opts.engine_id > 0x7f) {
720 VLOG_WARN("bridge %s: netflow port mangling may conflict "
721 "with another vswitch, choose an engine id less "
722 "than 128", br->name);
724 if (br->n_ports > 508) {
725 VLOG_WARN("bridge %s: netflow port mangling will conflict "
726 "with another port when more than 508 ports are "
731 opts.collectors.n = nf_cfg->n_targets;
732 opts.collectors.names = nf_cfg->targets;
733 if (ofproto_set_netflow(br->ofproto, &opts)) {
734 VLOG_ERR("bridge %s: problem setting netflow collectors",
738 ofproto_set_netflow(br->ofproto, NULL);
741 /* Set sFlow configuration on this bridge. */
742 if (br->cfg->sflow) {
743 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
744 struct ovsrec_controller **controllers;
745 struct ofproto_sflow_options oso;
746 size_t n_controllers;
749 memset(&oso, 0, sizeof oso);
751 oso.targets.n = sflow_cfg->n_targets;
752 oso.targets.names = sflow_cfg->targets;
754 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
755 if (sflow_cfg->sampling) {
756 oso.sampling_rate = *sflow_cfg->sampling;
759 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
760 if (sflow_cfg->polling) {
761 oso.polling_interval = *sflow_cfg->polling;
764 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
765 if (sflow_cfg->header) {
766 oso.header_len = *sflow_cfg->header;
769 oso.sub_id = sflow_bridge_number++;
770 oso.agent_device = sflow_cfg->agent;
772 oso.control_ip = NULL;
773 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
774 for (i = 0; i < n_controllers; i++) {
775 if (controllers[i]->local_ip) {
776 oso.control_ip = controllers[i]->local_ip;
780 ofproto_set_sflow(br->ofproto, &oso);
782 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
784 ofproto_set_sflow(br->ofproto, NULL);
787 /* Update the controller and related settings. It would be more
788 * straightforward to call this from bridge_reconfigure_one(), but we
789 * can't do it there for two reasons. First, and most importantly, at
790 * that point we don't know the dp_ifidx of any interfaces that have
791 * been added to the bridge (because we haven't actually added them to
792 * the datapath). Second, at that point we haven't set the datapath ID
793 * yet; when a controller is configured, resetting the datapath ID will
794 * immediately disconnect from the controller, so it's better to set
795 * the datapath ID before the controller. */
796 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
798 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
799 for (i = 0; i < br->n_ports; i++) {
800 struct port *port = br->ports[i];
803 port_update_vlan_compat(port);
804 port_update_bonding(port);
806 for (j = 0; j < port->n_ifaces; j++) {
807 iface_update_qos(port->ifaces[j], port->cfg->qos);
811 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
812 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
815 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
817 ovsdb_idl_txn_commit(txn);
818 ovsdb_idl_txn_destroy(txn); /* XXX */
824 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
828 for (i = 0; i < n; i++) {
829 if (!strcmp(keys[i], key)) {
837 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
839 return get_ovsrec_key_value(key,
840 br_cfg->key_other_config,
841 br_cfg->value_other_config,
842 br_cfg->n_other_config);
846 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
847 struct iface **hw_addr_iface)
853 *hw_addr_iface = NULL;
855 /* Did the user request a particular MAC? */
856 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
857 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
858 if (eth_addr_is_multicast(ea)) {
859 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
860 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
861 } else if (eth_addr_is_zero(ea)) {
862 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
868 /* Otherwise choose the minimum non-local MAC address among all of the
870 memset(ea, 0xff, sizeof ea);
871 for (i = 0; i < br->n_ports; i++) {
872 struct port *port = br->ports[i];
873 uint8_t iface_ea[ETH_ADDR_LEN];
876 /* Mirror output ports don't participate. */
877 if (port->is_mirror_output_port) {
881 /* Choose the MAC address to represent the port. */
882 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
883 /* Find the interface with this Ethernet address (if any) so that
884 * we can provide the correct devname to the caller. */
886 for (j = 0; j < port->n_ifaces; j++) {
887 struct iface *candidate = port->ifaces[j];
888 uint8_t candidate_ea[ETH_ADDR_LEN];
889 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
890 && eth_addr_equals(iface_ea, candidate_ea)) {
895 /* Choose the interface whose MAC address will represent the port.
896 * The Linux kernel bonding code always chooses the MAC address of
897 * the first slave added to a bond, and the Fedora networking
898 * scripts always add slaves to a bond in alphabetical order, so
899 * for compatibility we choose the interface with the name that is
900 * first in alphabetical order. */
901 iface = port->ifaces[0];
902 for (j = 1; j < port->n_ifaces; j++) {
903 struct iface *candidate = port->ifaces[j];
904 if (strcmp(candidate->name, iface->name) < 0) {
909 /* The local port doesn't count (since we're trying to choose its
910 * MAC address anyway). */
911 if (iface->dp_ifidx == ODPP_LOCAL) {
916 error = netdev_get_etheraddr(iface->netdev, iface_ea);
918 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
919 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
920 iface->name, strerror(error));
925 /* Compare against our current choice. */
926 if (!eth_addr_is_multicast(iface_ea) &&
927 !eth_addr_is_local(iface_ea) &&
928 !eth_addr_is_reserved(iface_ea) &&
929 !eth_addr_is_zero(iface_ea) &&
930 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
932 memcpy(ea, iface_ea, ETH_ADDR_LEN);
933 *hw_addr_iface = iface;
936 if (eth_addr_is_multicast(ea)) {
937 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
938 *hw_addr_iface = NULL;
939 VLOG_WARN("bridge %s: using default bridge Ethernet "
940 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
942 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
943 br->name, ETH_ADDR_ARGS(ea));
947 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
948 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
949 * an interface on 'br', then that interface must be passed in as
950 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
951 * 'hw_addr_iface' must be passed in as a null pointer. */
953 bridge_pick_datapath_id(struct bridge *br,
954 const uint8_t bridge_ea[ETH_ADDR_LEN],
955 struct iface *hw_addr_iface)
958 * The procedure for choosing a bridge MAC address will, in the most
959 * ordinary case, also choose a unique MAC that we can use as a datapath
960 * ID. In some special cases, though, multiple bridges will end up with
961 * the same MAC address. This is OK for the bridges, but it will confuse
962 * the OpenFlow controller, because each datapath needs a unique datapath
965 * Datapath IDs must be unique. It is also very desirable that they be
966 * stable from one run to the next, so that policy set on a datapath
969 const char *datapath_id;
972 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
973 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
979 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
981 * A bridge whose MAC address is taken from a VLAN network device
982 * (that is, a network device created with vconfig(8) or similar
983 * tool) will have the same MAC address as a bridge on the VLAN
984 * device's physical network device.
986 * Handle this case by hashing the physical network device MAC
987 * along with the VLAN identifier.
989 uint8_t buf[ETH_ADDR_LEN + 2];
990 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
991 buf[ETH_ADDR_LEN] = vlan >> 8;
992 buf[ETH_ADDR_LEN + 1] = vlan;
993 return dpid_from_hash(buf, sizeof buf);
996 * Assume that this bridge's MAC address is unique, since it
997 * doesn't fit any of the cases we handle specially.
1002 * A purely internal bridge, that is, one that has no non-virtual
1003 * network devices on it at all, is more difficult because it has no
1004 * natural unique identifier at all.
1006 * When the host is a XenServer, we handle this case by hashing the
1007 * host's UUID with the name of the bridge. Names of bridges are
1008 * persistent across XenServer reboots, although they can be reused if
1009 * an internal network is destroyed and then a new one is later
1010 * created, so this is fairly effective.
1012 * When the host is not a XenServer, we punt by using a random MAC
1013 * address on each run.
1015 const char *host_uuid = xenserver_get_host_uuid();
1017 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1018 dpid = dpid_from_hash(combined, strlen(combined));
1024 return eth_addr_to_uint64(bridge_ea);
1028 dpid_from_hash(const void *data, size_t n)
1030 uint8_t hash[SHA1_DIGEST_SIZE];
1032 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1033 sha1_bytes(data, n, hash);
1034 eth_addr_mark_random(hash);
1035 return eth_addr_to_uint64(hash);
1041 struct bridge *br, *next;
1045 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1046 int error = bridge_run_one(br);
1048 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1049 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1050 "forcing reconfiguration", br->name);
1064 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1065 ofproto_wait(br->ofproto);
1066 if (ofproto_has_controller(br->ofproto)) {
1070 mac_learning_wait(br->ml);
1075 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1076 * configuration changes. */
1078 bridge_flush(struct bridge *br)
1080 COVERAGE_INC(bridge_flush);
1082 mac_learning_flush(br->ml);
1085 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1086 * such interface. */
1087 static struct iface *
1088 bridge_get_local_iface(struct bridge *br)
1092 for (i = 0; i < br->n_ports; i++) {
1093 struct port *port = br->ports[i];
1094 for (j = 0; j < port->n_ifaces; j++) {
1095 struct iface *iface = port->ifaces[j];
1096 if (iface->dp_ifidx == ODPP_LOCAL) {
1105 /* Bridge unixctl user interface functions. */
1107 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1108 const char *args, void *aux OVS_UNUSED)
1110 struct ds ds = DS_EMPTY_INITIALIZER;
1111 const struct bridge *br;
1112 const struct mac_entry *e;
1114 br = bridge_lookup(args);
1116 unixctl_command_reply(conn, 501, "no such bridge");
1120 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1121 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1122 if (e->port < 0 || e->port >= br->n_ports) {
1125 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1126 br->ports[e->port]->ifaces[0]->dp_ifidx,
1127 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1129 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1133 /* Bridge reconfiguration functions. */
1134 static struct bridge *
1135 bridge_create(const struct ovsrec_bridge *br_cfg)
1140 assert(!bridge_lookup(br_cfg->name));
1141 br = xzalloc(sizeof *br);
1143 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1149 dpif_flow_flush(br->dpif);
1151 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1154 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1156 dpif_delete(br->dpif);
1157 dpif_close(br->dpif);
1162 br->name = xstrdup(br_cfg->name);
1164 br->ml = mac_learning_create();
1165 eth_addr_nicira_random(br->default_ea);
1167 port_array_init(&br->ifaces);
1169 shash_init(&br->port_by_name);
1170 shash_init(&br->iface_by_name);
1174 list_push_back(&all_bridges, &br->node);
1176 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1182 bridge_destroy(struct bridge *br)
1187 while (br->n_ports > 0) {
1188 port_destroy(br->ports[br->n_ports - 1]);
1190 list_remove(&br->node);
1191 error = dpif_delete(br->dpif);
1192 if (error && error != ENOENT) {
1193 VLOG_ERR("failed to delete %s: %s",
1194 dpif_name(br->dpif), strerror(error));
1196 dpif_close(br->dpif);
1197 ofproto_destroy(br->ofproto);
1198 mac_learning_destroy(br->ml);
1199 port_array_destroy(&br->ifaces);
1200 shash_destroy(&br->port_by_name);
1201 shash_destroy(&br->iface_by_name);
1208 static struct bridge *
1209 bridge_lookup(const char *name)
1213 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1214 if (!strcmp(br->name, name)) {
1221 /* Handle requests for a listing of all flows known by the OpenFlow
1222 * stack, including those normally hidden. */
1224 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1225 const char *args, void *aux OVS_UNUSED)
1230 br = bridge_lookup(args);
1232 unixctl_command_reply(conn, 501, "Unknown bridge");
1237 ofproto_get_all_flows(br->ofproto, &results);
1239 unixctl_command_reply(conn, 200, ds_cstr(&results));
1240 ds_destroy(&results);
1244 bridge_run_one(struct bridge *br)
1248 error = ofproto_run1(br->ofproto);
1253 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1256 error = ofproto_run2(br->ofproto, br->flush);
1263 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1264 const struct bridge *br,
1265 struct ovsrec_controller ***controllersp)
1267 struct ovsrec_controller **controllers;
1268 size_t n_controllers;
1270 if (br->cfg->n_controller) {
1271 controllers = br->cfg->controller;
1272 n_controllers = br->cfg->n_controller;
1274 controllers = ovs_cfg->controller;
1275 n_controllers = ovs_cfg->n_controller;
1278 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1284 *controllersp = controllers;
1286 return n_controllers;
1290 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1293 struct shash old_ports, new_ports;
1294 struct svec listeners, old_listeners;
1295 struct svec snoops, old_snoops;
1296 struct shash_node *node;
1299 /* Collect old ports. */
1300 shash_init(&old_ports);
1301 for (i = 0; i < br->n_ports; i++) {
1302 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1305 /* Collect new ports. */
1306 shash_init(&new_ports);
1307 for (i = 0; i < br->cfg->n_ports; i++) {
1308 const char *name = br->cfg->ports[i]->name;
1309 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1310 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1315 /* If we have a controller, then we need a local port. Complain if the
1316 * user didn't specify one.
1318 * XXX perhaps we should synthesize a port ourselves in this case. */
1319 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1320 char local_name[IF_NAMESIZE];
1323 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1324 local_name, sizeof local_name);
1325 if (!error && !shash_find(&new_ports, local_name)) {
1326 VLOG_WARN("bridge %s: controller specified but no local port "
1327 "(port named %s) defined",
1328 br->name, local_name);
1332 /* Get rid of deleted ports.
1333 * Get rid of deleted interfaces on ports that still exist. */
1334 SHASH_FOR_EACH (node, &old_ports) {
1335 struct port *port = node->data;
1336 const struct ovsrec_port *port_cfg;
1338 port_cfg = shash_find_data(&new_ports, node->name);
1342 port_del_ifaces(port, port_cfg);
1346 /* Create new ports.
1347 * Add new interfaces to existing ports.
1348 * Reconfigure existing ports. */
1349 SHASH_FOR_EACH (node, &new_ports) {
1350 struct port *port = shash_find_data(&old_ports, node->name);
1352 port = port_create(br, node->name);
1355 port_reconfigure(port, node->data);
1356 if (!port->n_ifaces) {
1357 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1358 br->name, port->name);
1362 shash_destroy(&old_ports);
1363 shash_destroy(&new_ports);
1365 /* Delete all flows if we're switching from connected to standalone or vice
1366 * versa. (XXX Should we delete all flows if we are switching from one
1367 * controller to another?) */
1369 /* Configure OpenFlow management listener. */
1370 svec_init(&listeners);
1371 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1372 ovs_rundir, br->name));
1373 svec_init(&old_listeners);
1374 ofproto_get_listeners(br->ofproto, &old_listeners);
1375 if (!svec_equal(&listeners, &old_listeners)) {
1376 ofproto_set_listeners(br->ofproto, &listeners);
1378 svec_destroy(&listeners);
1379 svec_destroy(&old_listeners);
1381 /* Configure OpenFlow controller connection snooping. */
1383 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1384 ovs_rundir, br->name));
1385 svec_init(&old_snoops);
1386 ofproto_get_snoops(br->ofproto, &old_snoops);
1387 if (!svec_equal(&snoops, &old_snoops)) {
1388 ofproto_set_snoops(br->ofproto, &snoops);
1390 svec_destroy(&snoops);
1391 svec_destroy(&old_snoops);
1393 mirror_reconfigure(br);
1397 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1399 const struct sockaddr_in *managers,
1402 struct ovsrec_controller **controllers;
1403 size_t n_controllers;
1405 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1407 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1408 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1409 ofproto_flush_flows(br->ofproto);
1412 if (!n_controllers) {
1413 union ofp_action action;
1416 /* Clear out controllers. */
1417 ofproto_set_controllers(br->ofproto, NULL, 0);
1419 /* Set up a flow that matches every packet and directs them to
1420 * OFPP_NORMAL (which goes to us). */
1421 memset(&action, 0, sizeof action);
1422 action.type = htons(OFPAT_OUTPUT);
1423 action.output.len = htons(sizeof action);
1424 action.output.port = htons(OFPP_NORMAL);
1425 memset(&flow, 0, sizeof flow);
1426 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1428 struct ofproto_controller *ocs;
1431 ocs = xmalloc(n_controllers * sizeof *ocs);
1432 for (i = 0; i < n_controllers; i++) {
1433 struct ovsrec_controller *c = controllers[i];
1434 struct ofproto_controller *oc = &ocs[i];
1436 if (strcmp(c->target, "discover")) {
1437 struct iface *local_iface;
1440 local_iface = bridge_get_local_iface(br);
1441 if (local_iface && c->local_ip
1442 && inet_aton(c->local_ip, &ip)) {
1443 struct netdev *netdev = local_iface->netdev;
1444 struct in_addr mask, gateway;
1446 if (!c->local_netmask
1447 || !inet_aton(c->local_netmask, &mask)) {
1450 if (!c->local_gateway
1451 || !inet_aton(c->local_gateway, &gateway)) {
1455 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1457 mask.s_addr = guess_netmask(ip.s_addr);
1459 if (!netdev_set_in4(netdev, ip, mask)) {
1460 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1462 br->name, IP_ARGS(&ip.s_addr),
1463 IP_ARGS(&mask.s_addr));
1466 if (gateway.s_addr) {
1467 if (!netdev_add_router(netdev, gateway)) {
1468 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1469 br->name, IP_ARGS(&gateway.s_addr));
1475 oc->target = c->target;
1476 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1477 oc->probe_interval = (c->inactivity_probe
1478 ? *c->inactivity_probe / 1000 : 5);
1479 oc->fail = (!c->fail_mode
1480 || !strcmp(c->fail_mode, "standalone")
1481 || !strcmp(c->fail_mode, "open")
1482 ? OFPROTO_FAIL_STANDALONE
1483 : OFPROTO_FAIL_SECURE);
1484 oc->band = (!c->connection_mode
1485 || !strcmp(c->connection_mode, "in-band")
1487 : OFPROTO_OUT_OF_BAND);
1488 oc->accept_re = c->discover_accept_regex;
1489 oc->update_resolv_conf = c->discover_update_resolv_conf;
1490 oc->rate_limit = (c->controller_rate_limit
1491 ? *c->controller_rate_limit : 0);
1492 oc->burst_limit = (c->controller_burst_limit
1493 ? *c->controller_burst_limit : 0);
1495 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1501 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1506 for (i = 0; i < br->n_ports; i++) {
1507 struct port *port = br->ports[i];
1508 for (j = 0; j < port->n_ifaces; j++) {
1509 struct iface *iface = port->ifaces[j];
1510 shash_add_once(ifaces, iface->name, iface);
1512 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1513 shash_add_once(ifaces, port->name, NULL);
1518 /* For robustness, in case the administrator moves around datapath ports behind
1519 * our back, we re-check all the datapath port numbers here.
1521 * This function will set the 'dp_ifidx' members of interfaces that have
1522 * disappeared to -1, so only call this function from a context where those
1523 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1524 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1525 * datapath, which doesn't support UINT16_MAX+1 ports. */
1527 bridge_fetch_dp_ifaces(struct bridge *br)
1529 struct odp_port *dpif_ports;
1530 size_t n_dpif_ports;
1533 /* Reset all interface numbers. */
1534 for (i = 0; i < br->n_ports; i++) {
1535 struct port *port = br->ports[i];
1536 for (j = 0; j < port->n_ifaces; j++) {
1537 struct iface *iface = port->ifaces[j];
1538 iface->dp_ifidx = -1;
1541 port_array_clear(&br->ifaces);
1543 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1544 for (i = 0; i < n_dpif_ports; i++) {
1545 struct odp_port *p = &dpif_ports[i];
1546 struct iface *iface = iface_lookup(br, p->devname);
1548 if (iface->dp_ifidx >= 0) {
1549 VLOG_WARN("%s reported interface %s twice",
1550 dpif_name(br->dpif), p->devname);
1551 } else if (iface_from_dp_ifidx(br, p->port)) {
1552 VLOG_WARN("%s reported interface %"PRIu16" twice",
1553 dpif_name(br->dpif), p->port);
1555 port_array_set(&br->ifaces, p->port, iface);
1556 iface->dp_ifidx = p->port;
1560 int64_t ofport = (iface->dp_ifidx >= 0
1561 ? odp_port_to_ofp_port(iface->dp_ifidx)
1563 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1570 /* Bridge packet processing functions. */
1573 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1575 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1578 static struct bond_entry *
1579 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1581 return &port->bond_hash[bond_hash(mac)];
1585 bond_choose_iface(const struct port *port)
1587 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1588 size_t i, best_down_slave = -1;
1589 long long next_delay_expiration = LLONG_MAX;
1591 for (i = 0; i < port->n_ifaces; i++) {
1592 struct iface *iface = port->ifaces[i];
1594 if (iface->enabled) {
1596 } else if (iface->delay_expires < next_delay_expiration) {
1597 best_down_slave = i;
1598 next_delay_expiration = iface->delay_expires;
1602 if (best_down_slave != -1) {
1603 struct iface *iface = port->ifaces[best_down_slave];
1605 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1606 "since no other interface is up", iface->name,
1607 iface->delay_expires - time_msec());
1608 bond_enable_slave(iface, true);
1611 return best_down_slave;
1615 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1616 uint16_t *dp_ifidx, tag_type *tags)
1618 struct iface *iface;
1620 assert(port->n_ifaces);
1621 if (port->n_ifaces == 1) {
1622 iface = port->ifaces[0];
1624 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1625 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1626 || !port->ifaces[e->iface_idx]->enabled) {
1627 /* XXX select interface properly. The current interface selection
1628 * is only good for testing the rebalancing code. */
1629 e->iface_idx = bond_choose_iface(port);
1630 if (e->iface_idx < 0) {
1631 *tags |= port->no_ifaces_tag;
1634 e->iface_tag = tag_create_random();
1635 ((struct port *) port)->bond_compat_is_stale = true;
1637 *tags |= e->iface_tag;
1638 iface = port->ifaces[e->iface_idx];
1640 *dp_ifidx = iface->dp_ifidx;
1641 *tags |= iface->tag; /* Currently only used for bonding. */
1646 bond_link_status_update(struct iface *iface, bool carrier)
1648 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1649 struct port *port = iface->port;
1651 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1652 /* Nothing to do. */
1655 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1656 iface->name, carrier ? "detected" : "dropped");
1657 if (carrier == iface->enabled) {
1658 iface->delay_expires = LLONG_MAX;
1659 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1660 iface->name, carrier ? "disabled" : "enabled");
1661 } else if (carrier && port->active_iface < 0) {
1662 bond_enable_slave(iface, true);
1663 if (port->updelay) {
1664 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1665 "other interface is up", iface->name, port->updelay);
1668 int delay = carrier ? port->updelay : port->downdelay;
1669 iface->delay_expires = time_msec() + delay;
1672 "interface %s: will be %s if it stays %s for %d ms",
1674 carrier ? "enabled" : "disabled",
1675 carrier ? "up" : "down",
1682 bond_choose_active_iface(struct port *port)
1684 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1686 port->active_iface = bond_choose_iface(port);
1687 port->active_iface_tag = tag_create_random();
1688 if (port->active_iface >= 0) {
1689 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1690 port->name, port->ifaces[port->active_iface]->name);
1692 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1698 bond_enable_slave(struct iface *iface, bool enable)
1700 struct port *port = iface->port;
1701 struct bridge *br = port->bridge;
1703 /* This acts as a recursion check. If the act of disabling a slave
1704 * causes a different slave to be enabled, the flag will allow us to
1705 * skip redundant work when we reenter this function. It must be
1706 * cleared on exit to keep things safe with multiple bonds. */
1707 static bool moving_active_iface = false;
1709 iface->delay_expires = LLONG_MAX;
1710 if (enable == iface->enabled) {
1714 iface->enabled = enable;
1715 if (!iface->enabled) {
1716 VLOG_WARN("interface %s: disabled", iface->name);
1717 ofproto_revalidate(br->ofproto, iface->tag);
1718 if (iface->port_ifidx == port->active_iface) {
1719 ofproto_revalidate(br->ofproto,
1720 port->active_iface_tag);
1722 /* Disabling a slave can lead to another slave being immediately
1723 * enabled if there will be no active slaves but one is waiting
1724 * on an updelay. In this case we do not need to run most of the
1725 * code for the newly enabled slave since there was no period
1726 * without an active slave and it is redundant with the disabling
1728 moving_active_iface = true;
1729 bond_choose_active_iface(port);
1731 bond_send_learning_packets(port);
1733 VLOG_WARN("interface %s: enabled", iface->name);
1734 if (port->active_iface < 0 && !moving_active_iface) {
1735 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1736 bond_choose_active_iface(port);
1737 bond_send_learning_packets(port);
1739 iface->tag = tag_create_random();
1742 moving_active_iface = false;
1743 port->bond_compat_is_stale = true;
1746 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1747 * bond interface. */
1749 bond_update_fake_iface_stats(struct port *port)
1751 struct netdev_stats bond_stats;
1752 struct netdev *bond_dev;
1755 memset(&bond_stats, 0, sizeof bond_stats);
1757 for (i = 0; i < port->n_ifaces; i++) {
1758 struct netdev_stats slave_stats;
1760 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1761 /* XXX: We swap the stats here because they are swapped back when
1762 * reported by the internal device. The reason for this is
1763 * internal devices normally represent packets going into the system
1764 * but when used as fake bond device they represent packets leaving
1765 * the system. We really should do this in the internal device
1766 * itself because changing it here reverses the counts from the
1767 * perspective of the switch. However, the internal device doesn't
1768 * know what type of device it represents so we have to do it here
1770 bond_stats.tx_packets += slave_stats.rx_packets;
1771 bond_stats.tx_bytes += slave_stats.rx_bytes;
1772 bond_stats.rx_packets += slave_stats.tx_packets;
1773 bond_stats.rx_bytes += slave_stats.tx_bytes;
1777 if (!netdev_open_default(port->name, &bond_dev)) {
1778 netdev_set_stats(bond_dev, &bond_stats);
1779 netdev_close(bond_dev);
1784 bond_run(struct bridge *br)
1788 for (i = 0; i < br->n_ports; i++) {
1789 struct port *port = br->ports[i];
1791 if (port->n_ifaces >= 2) {
1792 for (j = 0; j < port->n_ifaces; j++) {
1793 struct iface *iface = port->ifaces[j];
1794 if (time_msec() >= iface->delay_expires) {
1795 bond_enable_slave(iface, !iface->enabled);
1799 if (port->bond_fake_iface
1800 && time_msec() >= port->bond_next_fake_iface_update) {
1801 bond_update_fake_iface_stats(port);
1802 port->bond_next_fake_iface_update = time_msec() + 1000;
1806 if (port->bond_compat_is_stale) {
1807 port->bond_compat_is_stale = false;
1808 port_update_bond_compat(port);
1814 bond_wait(struct bridge *br)
1818 for (i = 0; i < br->n_ports; i++) {
1819 struct port *port = br->ports[i];
1820 if (port->n_ifaces < 2) {
1823 for (j = 0; j < port->n_ifaces; j++) {
1824 struct iface *iface = port->ifaces[j];
1825 if (iface->delay_expires != LLONG_MAX) {
1826 poll_timer_wait_until(iface->delay_expires);
1829 if (port->bond_fake_iface) {
1830 poll_timer_wait_until(port->bond_next_fake_iface_update);
1836 set_dst(struct dst *p, const flow_t *flow,
1837 const struct port *in_port, const struct port *out_port,
1840 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1841 : in_port->vlan >= 0 ? in_port->vlan
1842 : ntohs(flow->dl_vlan));
1843 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1847 swap_dst(struct dst *p, struct dst *q)
1849 struct dst tmp = *p;
1854 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1855 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1856 * that we push to the datapath. We could in fact fully sort the array by
1857 * vlan, but in most cases there are at most two different vlan tags so that's
1858 * possibly overkill.) */
1860 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1862 struct dst *first = dsts;
1863 struct dst *last = dsts + n_dsts;
1865 while (first != last) {
1867 * - All dsts < first have vlan == 'vlan'.
1868 * - All dsts >= last have vlan != 'vlan'.
1869 * - first < last. */
1870 while (first->vlan == vlan) {
1871 if (++first == last) {
1876 /* Same invariants, plus one additional:
1877 * - first->vlan != vlan.
1879 while (last[-1].vlan != vlan) {
1880 if (--last == first) {
1885 /* Same invariants, plus one additional:
1886 * - last[-1].vlan == vlan.*/
1887 swap_dst(first++, --last);
1892 mirror_mask_ffs(mirror_mask_t mask)
1894 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1899 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1900 const struct dst *test)
1903 for (i = 0; i < n_dsts; i++) {
1904 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1912 port_trunks_vlan(const struct port *port, uint16_t vlan)
1914 return (port->vlan < 0
1915 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
1919 port_includes_vlan(const struct port *port, uint16_t vlan)
1921 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1925 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1926 const struct port *in_port, const struct port *out_port,
1927 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1929 mirror_mask_t mirrors = in_port->src_mirrors;
1930 struct dst *dst = dsts;
1933 if (out_port == FLOOD_PORT) {
1934 /* XXX use ODP_FLOOD if no vlans or bonding. */
1935 /* XXX even better, define each VLAN as a datapath port group */
1936 for (i = 0; i < br->n_ports; i++) {
1937 struct port *port = br->ports[i];
1938 if (port != in_port && port_includes_vlan(port, vlan)
1939 && !port->is_mirror_output_port
1940 && set_dst(dst, flow, in_port, port, tags)) {
1941 mirrors |= port->dst_mirrors;
1945 *nf_output_iface = NF_OUT_FLOOD;
1946 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1947 *nf_output_iface = dst->dp_ifidx;
1948 mirrors |= out_port->dst_mirrors;
1953 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1954 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1956 if (set_dst(dst, flow, in_port, m->out_port, tags)
1957 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1961 for (i = 0; i < br->n_ports; i++) {
1962 struct port *port = br->ports[i];
1963 if (port_includes_vlan(port, m->out_vlan)
1964 && set_dst(dst, flow, in_port, port, tags))
1968 if (port->vlan < 0) {
1969 dst->vlan = m->out_vlan;
1971 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1975 /* Use the vlan tag on the original flow instead of
1976 * the one passed in the vlan parameter. This ensures
1977 * that we compare the vlan from before any implicit
1978 * tagging tags place. This is necessary because
1979 * dst->vlan is the final vlan, after removing implicit
1981 flow_vlan = ntohs(flow->dl_vlan);
1982 if (flow_vlan == 0) {
1983 flow_vlan = OFP_VLAN_NONE;
1985 if (port == in_port && dst->vlan == flow_vlan) {
1986 /* Don't send out input port on same VLAN. */
1994 mirrors &= mirrors - 1;
1997 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2001 static void OVS_UNUSED
2002 print_dsts(const struct dst *dsts, size_t n)
2004 for (; n--; dsts++) {
2005 printf(">p%"PRIu16, dsts->dp_ifidx);
2006 if (dsts->vlan != OFP_VLAN_NONE) {
2007 printf("v%"PRIu16, dsts->vlan);
2013 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2014 const struct port *in_port, const struct port *out_port,
2015 tag_type *tags, struct odp_actions *actions,
2016 uint16_t *nf_output_iface)
2018 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2020 const struct dst *p;
2023 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2026 cur_vlan = ntohs(flow->dl_vlan);
2027 for (p = dsts; p < &dsts[n_dsts]; p++) {
2028 union odp_action *a;
2029 if (p->vlan != cur_vlan) {
2030 if (p->vlan == OFP_VLAN_NONE) {
2031 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2033 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2034 a->vlan_vid.vlan_vid = htons(p->vlan);
2038 a = odp_actions_add(actions, ODPAT_OUTPUT);
2039 a->output.port = p->dp_ifidx;
2043 /* Returns the effective vlan of a packet, taking into account both the
2044 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2045 * the packet is untagged and -1 indicates it has an invalid header and
2046 * should be dropped. */
2047 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2048 struct port *in_port, bool have_packet)
2050 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2051 * belongs to VLAN 0, so we should treat both cases identically. (In the
2052 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2053 * presumably to allow a priority to be specified. In the latter case, the
2054 * packet does not have any 802.1Q header.) */
2055 int vlan = ntohs(flow->dl_vlan);
2056 if (vlan == OFP_VLAN_NONE) {
2059 if (in_port->vlan >= 0) {
2061 /* XXX support double tagging? */
2063 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2064 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2065 "packet received on port %s configured with "
2066 "implicit VLAN %"PRIu16,
2067 br->name, ntohs(flow->dl_vlan),
2068 in_port->name, in_port->vlan);
2072 vlan = in_port->vlan;
2074 if (!port_includes_vlan(in_port, vlan)) {
2076 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2077 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2078 "packet received on port %s not configured for "
2080 br->name, vlan, in_port->name, vlan);
2089 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2090 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2091 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2093 is_gratuitous_arp(const flow_t *flow)
2095 return (flow->dl_type == htons(ETH_TYPE_ARP)
2096 && eth_addr_is_broadcast(flow->dl_dst)
2097 && (flow->nw_proto == ARP_OP_REPLY
2098 || (flow->nw_proto == ARP_OP_REQUEST
2099 && flow->nw_src == flow->nw_dst)));
2103 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2104 struct port *in_port)
2106 enum grat_arp_lock_type lock_type;
2109 /* We don't want to learn from gratuitous ARP packets that are reflected
2110 * back over bond slaves so we lock the learning table. */
2111 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2112 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2113 GRAT_ARP_LOCK_CHECK;
2115 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2118 /* The log messages here could actually be useful in debugging,
2119 * so keep the rate limit relatively high. */
2120 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2122 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2123 "on port %s in VLAN %d",
2124 br->name, ETH_ADDR_ARGS(flow->dl_src),
2125 in_port->name, vlan);
2126 ofproto_revalidate(br->ofproto, rev_tag);
2130 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2131 * dropped. Returns true if they may be forwarded, false if they should be
2134 * If 'have_packet' is true, it indicates that the caller is processing a
2135 * received packet. If 'have_packet' is false, then the caller is just
2136 * revalidating an existing flow because configuration has changed. Either
2137 * way, 'have_packet' only affects logging (there is no point in logging errors
2138 * during revalidation).
2140 * Sets '*in_portp' to the input port. This will be a null pointer if
2141 * flow->in_port does not designate a known input port (in which case
2142 * is_admissible() returns false).
2144 * When returning true, sets '*vlanp' to the effective VLAN of the input
2145 * packet, as returned by flow_get_vlan().
2147 * May also add tags to '*tags', although the current implementation only does
2148 * so in one special case.
2151 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2152 tag_type *tags, int *vlanp, struct port **in_portp)
2154 struct iface *in_iface;
2155 struct port *in_port;
2158 /* Find the interface and port structure for the received packet. */
2159 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2161 /* No interface? Something fishy... */
2163 /* Odd. A few possible reasons here:
2165 * - We deleted an interface but there are still a few packets
2166 * queued up from it.
2168 * - Someone externally added an interface (e.g. with "ovs-dpctl
2169 * add-if") that we don't know about.
2171 * - Packet arrived on the local port but the local port is not
2172 * one of our bridge ports.
2174 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2176 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2177 "interface %"PRIu16, br->name, flow->in_port);
2183 *in_portp = in_port = in_iface->port;
2184 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2189 /* Drop frames for reserved multicast addresses. */
2190 if (eth_addr_is_reserved(flow->dl_dst)) {
2194 /* Drop frames on ports reserved for mirroring. */
2195 if (in_port->is_mirror_output_port) {
2197 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2198 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2199 "%s, which is reserved exclusively for mirroring",
2200 br->name, in_port->name);
2205 /* Packets received on bonds need special attention to avoid duplicates. */
2206 if (in_port->n_ifaces > 1) {
2208 bool is_grat_arp_locked;
2210 if (eth_addr_is_multicast(flow->dl_dst)) {
2211 *tags |= in_port->active_iface_tag;
2212 if (in_port->active_iface != in_iface->port_ifidx) {
2213 /* Drop all multicast packets on inactive slaves. */
2218 /* Drop all packets for which we have learned a different input
2219 * port, because we probably sent the packet on one slave and got
2220 * it back on the other. Gratuitous ARP packets are an exception
2221 * to this rule: the host has moved to another switch. The exception
2222 * to the exception is if we locked the learning table to avoid
2223 * reflections on bond slaves. If this is the case, just drop the
2225 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2226 &is_grat_arp_locked);
2227 if (src_idx != -1 && src_idx != in_port->port_idx &&
2228 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2236 /* If the composed actions may be applied to any packet in the given 'flow',
2237 * returns true. Otherwise, the actions should only be applied to 'packet', or
2238 * not at all, if 'packet' was NULL. */
2240 process_flow(struct bridge *br, const flow_t *flow,
2241 const struct ofpbuf *packet, struct odp_actions *actions,
2242 tag_type *tags, uint16_t *nf_output_iface)
2244 struct port *in_port;
2245 struct port *out_port;
2249 /* Check whether we should drop packets in this flow. */
2250 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2255 /* Learn source MAC (but don't try to learn from revalidation). */
2257 update_learning_table(br, flow, vlan, in_port);
2260 /* Determine output port. */
2261 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2263 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2264 out_port = br->ports[out_port_idx];
2265 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2266 /* If we are revalidating but don't have a learning entry then
2267 * eject the flow. Installing a flow that floods packets opens
2268 * up a window of time where we could learn from a packet reflected
2269 * on a bond and blackhole packets before the learning table is
2270 * updated to reflect the correct port. */
2273 out_port = FLOOD_PORT;
2276 /* Don't send packets out their input ports. */
2277 if (in_port == out_port) {
2283 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2290 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2293 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2294 const struct ofp_phy_port *opp,
2297 struct bridge *br = br_;
2298 struct iface *iface;
2301 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2307 if (reason == OFPPR_DELETE) {
2308 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2309 br->name, iface->name);
2310 iface_destroy(iface);
2311 if (!port->n_ifaces) {
2312 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2313 br->name, port->name);
2319 if (port->n_ifaces > 1) {
2320 bool up = !(opp->state & OFPPS_LINK_DOWN);
2321 bond_link_status_update(iface, up);
2322 port_update_bond_compat(port);
2328 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2329 struct odp_actions *actions, tag_type *tags,
2330 uint16_t *nf_output_iface, void *br_)
2332 struct bridge *br = br_;
2334 COVERAGE_INC(bridge_process_flow);
2335 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2339 bridge_account_flow_ofhook_cb(const flow_t *flow,
2340 const union odp_action *actions,
2341 size_t n_actions, unsigned long long int n_bytes,
2344 struct bridge *br = br_;
2345 const union odp_action *a;
2346 struct port *in_port;
2350 /* Feed information from the active flows back into the learning table
2351 * to ensure that table is always in sync with what is actually flowing
2352 * through the datapath. */
2353 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2354 update_learning_table(br, flow, vlan, in_port);
2357 if (!br->has_bonded_ports) {
2361 for (a = actions; a < &actions[n_actions]; a++) {
2362 if (a->type == ODPAT_OUTPUT) {
2363 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2364 if (out_port && out_port->n_ifaces >= 2) {
2365 struct bond_entry *e = lookup_bond_entry(out_port,
2367 e->tx_bytes += n_bytes;
2374 bridge_account_checkpoint_ofhook_cb(void *br_)
2376 struct bridge *br = br_;
2380 if (!br->has_bonded_ports) {
2385 for (i = 0; i < br->n_ports; i++) {
2386 struct port *port = br->ports[i];
2387 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2388 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2389 bond_rebalance_port(port);
2394 static struct ofhooks bridge_ofhooks = {
2395 bridge_port_changed_ofhook_cb,
2396 bridge_normal_ofhook_cb,
2397 bridge_account_flow_ofhook_cb,
2398 bridge_account_checkpoint_ofhook_cb,
2401 /* Bonding functions. */
2403 /* Statistics for a single interface on a bonded port, used for load-based
2404 * bond rebalancing. */
2405 struct slave_balance {
2406 struct iface *iface; /* The interface. */
2407 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2409 /* All the "bond_entry"s that are assigned to this interface, in order of
2410 * increasing tx_bytes. */
2411 struct bond_entry **hashes;
2415 /* Sorts pointers to pointers to bond_entries in ascending order by the
2416 * interface to which they are assigned, and within a single interface in
2417 * ascending order of bytes transmitted. */
2419 compare_bond_entries(const void *a_, const void *b_)
2421 const struct bond_entry *const *ap = a_;
2422 const struct bond_entry *const *bp = b_;
2423 const struct bond_entry *a = *ap;
2424 const struct bond_entry *b = *bp;
2425 if (a->iface_idx != b->iface_idx) {
2426 return a->iface_idx > b->iface_idx ? 1 : -1;
2427 } else if (a->tx_bytes != b->tx_bytes) {
2428 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2434 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2435 * *descending* order by number of bytes transmitted. */
2437 compare_slave_balance(const void *a_, const void *b_)
2439 const struct slave_balance *a = a_;
2440 const struct slave_balance *b = b_;
2441 if (a->iface->enabled != b->iface->enabled) {
2442 return a->iface->enabled ? -1 : 1;
2443 } else if (a->tx_bytes != b->tx_bytes) {
2444 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2451 swap_bals(struct slave_balance *a, struct slave_balance *b)
2453 struct slave_balance tmp = *a;
2458 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2459 * given that 'p' (and only 'p') might be in the wrong location.
2461 * This function invalidates 'p', since it might now be in a different memory
2464 resort_bals(struct slave_balance *p,
2465 struct slave_balance bals[], size_t n_bals)
2468 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2469 swap_bals(p, p - 1);
2471 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2472 swap_bals(p, p + 1);
2478 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2480 if (VLOG_IS_DBG_ENABLED()) {
2481 struct ds ds = DS_EMPTY_INITIALIZER;
2482 const struct slave_balance *b;
2484 for (b = bals; b < bals + n_bals; b++) {
2488 ds_put_char(&ds, ',');
2490 ds_put_format(&ds, " %s %"PRIu64"kB",
2491 b->iface->name, b->tx_bytes / 1024);
2493 if (!b->iface->enabled) {
2494 ds_put_cstr(&ds, " (disabled)");
2496 if (b->n_hashes > 0) {
2497 ds_put_cstr(&ds, " (");
2498 for (i = 0; i < b->n_hashes; i++) {
2499 const struct bond_entry *e = b->hashes[i];
2501 ds_put_cstr(&ds, " + ");
2503 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2504 e - port->bond_hash, e->tx_bytes / 1024);
2506 ds_put_cstr(&ds, ")");
2509 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2514 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2516 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2519 struct bond_entry *hash = from->hashes[hash_idx];
2520 struct port *port = from->iface->port;
2521 uint64_t delta = hash->tx_bytes;
2523 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2524 "from %s to %s (now carrying %"PRIu64"kB and "
2525 "%"PRIu64"kB load, respectively)",
2526 port->name, delta / 1024, hash - port->bond_hash,
2527 from->iface->name, to->iface->name,
2528 (from->tx_bytes - delta) / 1024,
2529 (to->tx_bytes + delta) / 1024);
2531 /* Delete element from from->hashes.
2533 * We don't bother to add the element to to->hashes because not only would
2534 * it require more work, the only purpose it would be to allow that hash to
2535 * be migrated to another slave in this rebalancing run, and there is no
2536 * point in doing that. */
2537 if (hash_idx == 0) {
2540 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2541 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2545 /* Shift load away from 'from' to 'to'. */
2546 from->tx_bytes -= delta;
2547 to->tx_bytes += delta;
2549 /* Arrange for flows to be revalidated. */
2550 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2551 hash->iface_idx = to->iface->port_ifidx;
2552 hash->iface_tag = tag_create_random();
2556 bond_rebalance_port(struct port *port)
2558 struct slave_balance bals[DP_MAX_PORTS];
2560 struct bond_entry *hashes[BOND_MASK + 1];
2561 struct slave_balance *b, *from, *to;
2562 struct bond_entry *e;
2565 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2566 * descending order of tx_bytes, so that bals[0] represents the most
2567 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2570 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2571 * array for each slave_balance structure, we sort our local array of
2572 * hashes in order by slave, so that all of the hashes for a given slave
2573 * become contiguous in memory, and then we point each 'hashes' members of
2574 * a slave_balance structure to the start of a contiguous group. */
2575 n_bals = port->n_ifaces;
2576 for (b = bals; b < &bals[n_bals]; b++) {
2577 b->iface = port->ifaces[b - bals];
2582 for (i = 0; i <= BOND_MASK; i++) {
2583 hashes[i] = &port->bond_hash[i];
2585 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2586 for (i = 0; i <= BOND_MASK; i++) {
2588 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2589 b = &bals[e->iface_idx];
2590 b->tx_bytes += e->tx_bytes;
2592 b->hashes = &hashes[i];
2597 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2598 log_bals(bals, n_bals, port);
2600 /* Discard slaves that aren't enabled (which were sorted to the back of the
2601 * array earlier). */
2602 while (!bals[n_bals - 1].iface->enabled) {
2609 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2610 to = &bals[n_bals - 1];
2611 for (from = bals; from < to; ) {
2612 uint64_t overload = from->tx_bytes - to->tx_bytes;
2613 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2614 /* The extra load on 'from' (and all less-loaded slaves), compared
2615 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2616 * it is less than ~1Mbps. No point in rebalancing. */
2618 } else if (from->n_hashes == 1) {
2619 /* 'from' only carries a single MAC hash, so we can't shift any
2620 * load away from it, even though we want to. */
2623 /* 'from' is carrying significantly more load than 'to', and that
2624 * load is split across at least two different hashes. Pick a hash
2625 * to migrate to 'to' (the least-loaded slave), given that doing so
2626 * must decrease the ratio of the load on the two slaves by at
2629 * The sort order we use means that we prefer to shift away the
2630 * smallest hashes instead of the biggest ones. There is little
2631 * reason behind this decision; we could use the opposite sort
2632 * order to shift away big hashes ahead of small ones. */
2636 for (i = 0; i < from->n_hashes; i++) {
2637 double old_ratio, new_ratio;
2638 uint64_t delta = from->hashes[i]->tx_bytes;
2640 if (delta == 0 || from->tx_bytes - delta == 0) {
2641 /* Pointless move. */
2645 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2647 if (to->tx_bytes == 0) {
2648 /* Nothing on the new slave, move it. */
2652 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2653 new_ratio = (double)(from->tx_bytes - delta) /
2654 (to->tx_bytes + delta);
2656 if (new_ratio == 0) {
2657 /* Should already be covered but check to prevent division
2662 if (new_ratio < 1) {
2663 new_ratio = 1 / new_ratio;
2666 if (old_ratio - new_ratio > 0.1) {
2667 /* Would decrease the ratio, move it. */
2671 if (i < from->n_hashes) {
2672 bond_shift_load(from, to, i);
2673 port->bond_compat_is_stale = true;
2675 /* If the result of the migration changed the relative order of
2676 * 'from' and 'to' swap them back to maintain invariants. */
2677 if (order_swapped) {
2678 swap_bals(from, to);
2681 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2682 * point to different slave_balance structures. It is only
2683 * valid to do these two operations in a row at all because we
2684 * know that 'from' will not move past 'to' and vice versa. */
2685 resort_bals(from, bals, n_bals);
2686 resort_bals(to, bals, n_bals);
2693 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2694 * historical data to decay to <1% in 7 rebalancing runs. */
2695 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2701 bond_send_learning_packets(struct port *port)
2703 struct bridge *br = port->bridge;
2704 struct mac_entry *e;
2705 struct ofpbuf packet;
2706 int error, n_packets, n_errors;
2708 if (!port->n_ifaces || port->active_iface < 0) {
2712 ofpbuf_init(&packet, 128);
2713 error = n_packets = n_errors = 0;
2714 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2715 union ofp_action actions[2], *a;
2721 if (e->port == port->port_idx
2722 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2726 /* Compose actions. */
2727 memset(actions, 0, sizeof actions);
2730 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2731 a->vlan_vid.len = htons(sizeof *a);
2732 a->vlan_vid.vlan_vid = htons(e->vlan);
2735 a->output.type = htons(OFPAT_OUTPUT);
2736 a->output.len = htons(sizeof *a);
2737 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2742 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2744 flow_extract(&packet, 0, ODPP_NONE, &flow);
2745 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2752 ofpbuf_uninit(&packet);
2755 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2756 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2757 "packets, last error was: %s",
2758 port->name, n_errors, n_packets, strerror(error));
2760 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2761 port->name, n_packets);
2765 /* Bonding unixctl user interface functions. */
2768 bond_unixctl_list(struct unixctl_conn *conn,
2769 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2771 struct ds ds = DS_EMPTY_INITIALIZER;
2772 const struct bridge *br;
2774 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2776 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2779 for (i = 0; i < br->n_ports; i++) {
2780 const struct port *port = br->ports[i];
2781 if (port->n_ifaces > 1) {
2784 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2785 for (j = 0; j < port->n_ifaces; j++) {
2786 const struct iface *iface = port->ifaces[j];
2788 ds_put_cstr(&ds, ", ");
2790 ds_put_cstr(&ds, iface->name);
2792 ds_put_char(&ds, '\n');
2796 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2800 static struct port *
2801 bond_find(const char *name)
2803 const struct bridge *br;
2805 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2808 for (i = 0; i < br->n_ports; i++) {
2809 struct port *port = br->ports[i];
2810 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2819 bond_unixctl_show(struct unixctl_conn *conn,
2820 const char *args, void *aux OVS_UNUSED)
2822 struct ds ds = DS_EMPTY_INITIALIZER;
2823 const struct port *port;
2826 port = bond_find(args);
2828 unixctl_command_reply(conn, 501, "no such bond");
2832 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2833 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2834 ds_put_format(&ds, "next rebalance: %lld ms\n",
2835 port->bond_next_rebalance - time_msec());
2836 for (j = 0; j < port->n_ifaces; j++) {
2837 const struct iface *iface = port->ifaces[j];
2838 struct bond_entry *be;
2841 ds_put_format(&ds, "slave %s: %s\n",
2842 iface->name, iface->enabled ? "enabled" : "disabled");
2843 if (j == port->active_iface) {
2844 ds_put_cstr(&ds, "\tactive slave\n");
2846 if (iface->delay_expires != LLONG_MAX) {
2847 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2848 iface->enabled ? "downdelay" : "updelay",
2849 iface->delay_expires - time_msec());
2853 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2854 int hash = be - port->bond_hash;
2855 struct mac_entry *me;
2857 if (be->iface_idx != j) {
2861 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2862 hash, be->tx_bytes / 1024);
2865 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2866 &port->bridge->ml->lrus) {
2869 if (bond_hash(me->mac) == hash
2870 && me->port != port->port_idx
2871 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2872 && dp_ifidx == iface->dp_ifidx)
2874 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2875 ETH_ADDR_ARGS(me->mac));
2880 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2885 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2886 void *aux OVS_UNUSED)
2888 char *args = (char *) args_;
2889 char *save_ptr = NULL;
2890 char *bond_s, *hash_s, *slave_s;
2891 uint8_t mac[ETH_ADDR_LEN];
2893 struct iface *iface;
2894 struct bond_entry *entry;
2897 bond_s = strtok_r(args, " ", &save_ptr);
2898 hash_s = strtok_r(NULL, " ", &save_ptr);
2899 slave_s = strtok_r(NULL, " ", &save_ptr);
2901 unixctl_command_reply(conn, 501,
2902 "usage: bond/migrate BOND HASH SLAVE");
2906 port = bond_find(bond_s);
2908 unixctl_command_reply(conn, 501, "no such bond");
2912 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2913 == ETH_ADDR_SCAN_COUNT) {
2914 hash = bond_hash(mac);
2915 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2916 hash = atoi(hash_s) & BOND_MASK;
2918 unixctl_command_reply(conn, 501, "bad hash");
2922 iface = port_lookup_iface(port, slave_s);
2924 unixctl_command_reply(conn, 501, "no such slave");
2928 if (!iface->enabled) {
2929 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2933 entry = &port->bond_hash[hash];
2934 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2935 entry->iface_idx = iface->port_ifidx;
2936 entry->iface_tag = tag_create_random();
2937 port->bond_compat_is_stale = true;
2938 unixctl_command_reply(conn, 200, "migrated");
2942 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2943 void *aux OVS_UNUSED)
2945 char *args = (char *) args_;
2946 char *save_ptr = NULL;
2947 char *bond_s, *slave_s;
2949 struct iface *iface;
2951 bond_s = strtok_r(args, " ", &save_ptr);
2952 slave_s = strtok_r(NULL, " ", &save_ptr);
2954 unixctl_command_reply(conn, 501,
2955 "usage: bond/set-active-slave BOND SLAVE");
2959 port = bond_find(bond_s);
2961 unixctl_command_reply(conn, 501, "no such bond");
2965 iface = port_lookup_iface(port, slave_s);
2967 unixctl_command_reply(conn, 501, "no such slave");
2971 if (!iface->enabled) {
2972 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2976 if (port->active_iface != iface->port_ifidx) {
2977 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2978 port->active_iface = iface->port_ifidx;
2979 port->active_iface_tag = tag_create_random();
2980 VLOG_INFO("port %s: active interface is now %s",
2981 port->name, iface->name);
2982 bond_send_learning_packets(port);
2983 unixctl_command_reply(conn, 200, "done");
2985 unixctl_command_reply(conn, 200, "no change");
2990 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2992 char *args = (char *) args_;
2993 char *save_ptr = NULL;
2994 char *bond_s, *slave_s;
2996 struct iface *iface;
2998 bond_s = strtok_r(args, " ", &save_ptr);
2999 slave_s = strtok_r(NULL, " ", &save_ptr);
3001 unixctl_command_reply(conn, 501,
3002 "usage: bond/enable/disable-slave BOND SLAVE");
3006 port = bond_find(bond_s);
3008 unixctl_command_reply(conn, 501, "no such bond");
3012 iface = port_lookup_iface(port, slave_s);
3014 unixctl_command_reply(conn, 501, "no such slave");
3018 bond_enable_slave(iface, enable);
3019 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3023 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3024 void *aux OVS_UNUSED)
3026 enable_slave(conn, args, true);
3030 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3031 void *aux OVS_UNUSED)
3033 enable_slave(conn, args, false);
3037 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3038 void *aux OVS_UNUSED)
3040 uint8_t mac[ETH_ADDR_LEN];
3044 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3045 == ETH_ADDR_SCAN_COUNT) {
3046 hash = bond_hash(mac);
3048 hash_cstr = xasprintf("%u", hash);
3049 unixctl_command_reply(conn, 200, hash_cstr);
3052 unixctl_command_reply(conn, 501, "invalid mac");
3059 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3060 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3061 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3062 unixctl_command_register("bond/set-active-slave",
3063 bond_unixctl_set_active_slave, NULL);
3064 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3066 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3068 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3071 /* Port functions. */
3073 static struct port *
3074 port_create(struct bridge *br, const char *name)
3078 port = xzalloc(sizeof *port);
3080 port->port_idx = br->n_ports;
3082 port->trunks = NULL;
3083 port->name = xstrdup(name);
3084 port->active_iface = -1;
3086 if (br->n_ports >= br->allocated_ports) {
3087 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3090 br->ports[br->n_ports++] = port;
3091 shash_add_assert(&br->port_by_name, port->name, port);
3093 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3100 get_port_other_config(const struct ovsrec_port *port, const char *key,
3101 const char *default_value)
3103 const char *value = get_ovsrec_key_value(key,
3104 port->key_other_config,
3105 port->value_other_config,
3106 port->n_other_config);
3107 return value ? value : default_value;
3111 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3113 struct shash new_ifaces;
3116 /* Collect list of new interfaces. */
3117 shash_init(&new_ifaces);
3118 for (i = 0; i < cfg->n_interfaces; i++) {
3119 const char *name = cfg->interfaces[i]->name;
3120 shash_add_once(&new_ifaces, name, NULL);
3123 /* Get rid of deleted interfaces. */
3124 for (i = 0; i < port->n_ifaces; ) {
3125 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3126 iface_destroy(port->ifaces[i]);
3132 shash_destroy(&new_ifaces);
3136 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3138 struct shash new_ifaces;
3139 long long int next_rebalance;
3140 unsigned long *trunks;
3146 /* Update settings. */
3147 port->updelay = cfg->bond_updelay;
3148 if (port->updelay < 0) {
3151 port->updelay = cfg->bond_downdelay;
3152 if (port->downdelay < 0) {
3153 port->downdelay = 0;
3155 port->bond_rebalance_interval = atoi(
3156 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3157 if (port->bond_rebalance_interval < 1000) {
3158 port->bond_rebalance_interval = 1000;
3160 next_rebalance = time_msec() + port->bond_rebalance_interval;
3161 if (port->bond_next_rebalance > next_rebalance) {
3162 port->bond_next_rebalance = next_rebalance;
3165 /* Add new interfaces and update 'cfg' member of existing ones. */
3166 shash_init(&new_ifaces);
3167 for (i = 0; i < cfg->n_interfaces; i++) {
3168 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3169 struct iface *iface;
3171 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3172 VLOG_WARN("port %s: %s specified twice as port interface",
3173 port->name, if_cfg->name);
3177 iface = iface_lookup(port->bridge, if_cfg->name);
3179 if (iface->port != port) {
3180 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3182 port->bridge->name, if_cfg->name, iface->port->name);
3185 iface->cfg = if_cfg;
3187 iface_create(port, if_cfg);
3190 shash_destroy(&new_ifaces);
3195 if (port->n_ifaces < 2) {
3197 if (vlan >= 0 && vlan <= 4095) {
3198 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3203 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3204 * they even work as-is. But they have not been tested. */
3205 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3209 if (port->vlan != vlan) {
3211 bridge_flush(port->bridge);
3214 /* Get trunked VLANs. */
3216 if (vlan < 0 && cfg->n_trunks) {
3220 trunks = bitmap_allocate(4096);
3222 for (i = 0; i < cfg->n_trunks; i++) {
3223 int trunk = cfg->trunks[i];
3225 bitmap_set1(trunks, trunk);
3231 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3232 port->name, cfg->n_trunks);
3234 if (n_errors == cfg->n_trunks) {
3235 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3237 bitmap_free(trunks);
3240 } else if (vlan >= 0 && cfg->n_trunks) {
3241 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3245 ? port->trunks != NULL
3246 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3247 bridge_flush(port->bridge);
3249 bitmap_free(port->trunks);
3250 port->trunks = trunks;
3254 port_destroy(struct port *port)
3257 struct bridge *br = port->bridge;
3261 proc_net_compat_update_vlan(port->name, NULL, 0);
3262 proc_net_compat_update_bond(port->name, NULL);
3264 for (i = 0; i < MAX_MIRRORS; i++) {
3265 struct mirror *m = br->mirrors[i];
3266 if (m && m->out_port == port) {
3271 while (port->n_ifaces > 0) {
3272 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3275 shash_find_and_delete_assert(&br->port_by_name, port->name);
3277 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3278 del->port_idx = port->port_idx;
3281 bitmap_free(port->trunks);
3288 static struct port *
3289 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3291 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3292 return iface ? iface->port : NULL;
3295 static struct port *
3296 port_lookup(const struct bridge *br, const char *name)
3298 return shash_find_data(&br->port_by_name, name);
3301 static struct iface *
3302 port_lookup_iface(const struct port *port, const char *name)
3304 struct iface *iface = iface_lookup(port->bridge, name);
3305 return iface && iface->port == port ? iface : NULL;
3309 port_update_bonding(struct port *port)
3311 if (port->n_ifaces < 2) {
3312 /* Not a bonded port. */
3313 if (port->bond_hash) {
3314 free(port->bond_hash);
3315 port->bond_hash = NULL;
3316 port->bond_compat_is_stale = true;
3317 port->bond_fake_iface = false;
3320 if (!port->bond_hash) {
3323 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3324 for (i = 0; i <= BOND_MASK; i++) {
3325 struct bond_entry *e = &port->bond_hash[i];
3329 port->no_ifaces_tag = tag_create_random();
3330 bond_choose_active_iface(port);
3331 port->bond_next_rebalance
3332 = time_msec() + port->bond_rebalance_interval;
3334 if (port->cfg->bond_fake_iface) {
3335 port->bond_next_fake_iface_update = time_msec();
3338 port->bond_compat_is_stale = true;
3339 port->bond_fake_iface = port->cfg->bond_fake_iface;
3344 port_update_bond_compat(struct port *port)
3346 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3347 struct compat_bond bond;
3350 if (port->n_ifaces < 2) {
3351 proc_net_compat_update_bond(port->name, NULL);
3356 bond.updelay = port->updelay;
3357 bond.downdelay = port->downdelay;
3360 bond.hashes = compat_hashes;
3361 if (port->bond_hash) {
3362 const struct bond_entry *e;
3363 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3364 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3365 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3366 cbh->hash = e - port->bond_hash;
3367 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3372 bond.n_slaves = port->n_ifaces;
3373 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3374 for (i = 0; i < port->n_ifaces; i++) {
3375 struct iface *iface = port->ifaces[i];
3376 struct compat_bond_slave *slave = &bond.slaves[i];
3377 slave->name = iface->name;
3379 /* We need to make the same determination as the Linux bonding
3380 * code to determine whether a slave should be consider "up".
3381 * The Linux function bond_miimon_inspect() supports four
3382 * BOND_LINK_* states:
3384 * - BOND_LINK_UP: carrier detected, updelay has passed.
3385 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3386 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3387 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3389 * The function bond_info_show_slave() only considers BOND_LINK_UP
3390 * to be "up" and anything else to be "down".
3392 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3396 netdev_get_etheraddr(iface->netdev, slave->mac);
3399 if (port->bond_fake_iface) {
3400 struct netdev *bond_netdev;
3402 if (!netdev_open_default(port->name, &bond_netdev)) {
3404 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3406 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3408 netdev_close(bond_netdev);
3412 proc_net_compat_update_bond(port->name, &bond);
3417 port_update_vlan_compat(struct port *port)
3419 struct bridge *br = port->bridge;
3420 char *vlandev_name = NULL;
3422 if (port->vlan > 0) {
3423 /* Figure out the name that the VLAN device should actually have, if it
3424 * existed. This takes some work because the VLAN device would not
3425 * have port->name in its name; rather, it would have the trunk port's
3426 * name, and 'port' would be attached to a bridge that also had the
3427 * VLAN device one of its ports. So we need to find a trunk port that
3428 * includes port->vlan.
3430 * There might be more than one candidate. This doesn't happen on
3431 * XenServer, so if it happens we just pick the first choice in
3432 * alphabetical order instead of creating multiple VLAN devices. */
3434 for (i = 0; i < br->n_ports; i++) {
3435 struct port *p = br->ports[i];
3436 if (port_trunks_vlan(p, port->vlan)
3438 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3440 uint8_t ea[ETH_ADDR_LEN];
3441 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3442 if (!eth_addr_is_multicast(ea) &&
3443 !eth_addr_is_reserved(ea) &&
3444 !eth_addr_is_zero(ea)) {
3445 vlandev_name = p->name;
3450 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3453 /* Interface functions. */
3455 static struct iface *
3456 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3458 struct bridge *br = port->bridge;
3459 struct iface *iface;
3460 char *name = if_cfg->name;
3463 iface = xzalloc(sizeof *iface);
3465 iface->port_ifidx = port->n_ifaces;
3466 iface->name = xstrdup(name);
3467 iface->dp_ifidx = -1;
3468 iface->tag = tag_create_random();
3469 iface->delay_expires = LLONG_MAX;
3470 iface->netdev = NULL;
3471 iface->cfg = if_cfg;
3473 shash_add_assert(&br->iface_by_name, iface->name, iface);
3475 /* Attempt to create the network interface in case it doesn't exist yet. */
3476 if (!iface_is_internal(br, iface->name)) {
3477 error = set_up_iface(if_cfg, iface, true);
3479 VLOG_WARN("could not create iface %s: %s", iface->name,
3482 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3489 if (port->n_ifaces >= port->allocated_ifaces) {
3490 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3491 sizeof *port->ifaces);
3493 port->ifaces[port->n_ifaces++] = iface;
3494 if (port->n_ifaces > 1) {
3495 br->has_bonded_ports = true;
3498 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3506 iface_destroy(struct iface *iface)
3509 struct port *port = iface->port;
3510 struct bridge *br = port->bridge;
3511 bool del_active = port->active_iface == iface->port_ifidx;
3514 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3516 if (iface->dp_ifidx >= 0) {
3517 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3520 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3521 del->port_ifidx = iface->port_ifidx;
3523 netdev_close(iface->netdev);
3526 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3527 bond_choose_active_iface(port);
3528 bond_send_learning_packets(port);
3534 bridge_flush(port->bridge);
3538 static struct iface *
3539 iface_lookup(const struct bridge *br, const char *name)
3541 return shash_find_data(&br->iface_by_name, name);
3544 static struct iface *
3545 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3547 return port_array_get(&br->ifaces, dp_ifidx);
3550 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3551 * 'br', that is, an interface that is entirely simulated within the datapath.
3552 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3553 * interfaces are created by setting "iface.<iface>.internal = true".
3555 * In addition, we have a kluge-y feature that creates an internal port with
3556 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3557 * This feature needs to go away in the long term. Until then, this is one
3558 * reason why this function takes a name instead of a struct iface: the fake
3559 * interfaces created this way do not have a struct iface. */
3561 iface_is_internal(const struct bridge *br, const char *if_name)
3563 struct iface *iface;
3566 if (!strcmp(if_name, br->name)) {
3570 iface = iface_lookup(br, if_name);
3571 if (iface && !strcmp(iface->cfg->type, "internal")) {
3575 port = port_lookup(br, if_name);
3576 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3582 /* Set Ethernet address of 'iface', if one is specified in the configuration
3585 iface_set_mac(struct iface *iface)
3587 uint8_t ea[ETH_ADDR_LEN];
3589 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3590 if (eth_addr_is_multicast(ea)) {
3591 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3593 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3594 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3595 iface->name, iface->name);
3597 int error = netdev_set_etheraddr(iface->netdev, ea);
3599 VLOG_ERR("interface %s: setting MAC failed (%s)",
3600 iface->name, strerror(error));
3607 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3608 struct shash *shash)
3613 for (i = 0; i < n; i++) {
3614 shash_add(shash, keys[i], values[i]);
3618 struct iface_delete_queues_cbdata {
3619 struct netdev *netdev;
3620 const int64_t *queue_ids;
3625 queue_ids_include(const int64_t *ids, size_t n, int64_t target)
3630 while (low < high) {
3631 size_t mid = low + (high - low) / 2;
3632 if (target > ids[mid]) {
3634 } else if (target < ids[mid]) {
3644 iface_delete_queues(unsigned int queue_id,
3645 const struct shash *details OVS_UNUSED, void *cbdata_)
3647 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3649 if (!queue_ids_include(cbdata->queue_ids, cbdata->n_queue_ids, queue_id)) {
3650 netdev_delete_queue(cbdata->netdev, queue_id);
3655 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3657 if (!qos || qos->type[0] == '\0') {
3658 netdev_set_qos(iface->netdev, NULL, NULL);
3660 struct iface_delete_queues_cbdata cbdata;
3661 struct shash details;
3664 /* Configure top-level Qos for 'iface'. */
3665 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3666 qos->n_other_config, &details);
3667 netdev_set_qos(iface->netdev, qos->type, &details);
3668 shash_destroy(&details);
3670 /* Deconfigure queues that were deleted. */
3671 cbdata.netdev = iface->netdev;
3672 cbdata.queue_ids = qos->key_queues;
3673 cbdata.n_queue_ids = qos->n_queues;
3674 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3676 /* Configure queues for 'iface'. */
3677 for (i = 0; i < qos->n_queues; i++) {
3678 const struct ovsrec_queue *queue = qos->value_queues[i];
3679 unsigned int queue_id = qos->key_queues[i];
3681 shash_from_ovs_idl_map(queue->key_other_config,
3682 queue->value_other_config,
3683 queue->n_other_config, &details);
3684 netdev_set_queue(iface->netdev, queue_id, &details);
3685 shash_destroy(&details);
3690 /* Port mirroring. */
3693 mirror_reconfigure(struct bridge *br)
3695 struct shash old_mirrors, new_mirrors;
3696 struct shash_node *node;
3697 unsigned long *rspan_vlans;
3700 /* Collect old mirrors. */
3701 shash_init(&old_mirrors);
3702 for (i = 0; i < MAX_MIRRORS; i++) {
3703 if (br->mirrors[i]) {
3704 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3708 /* Collect new mirrors. */
3709 shash_init(&new_mirrors);
3710 for (i = 0; i < br->cfg->n_mirrors; i++) {
3711 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3712 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3713 VLOG_WARN("bridge %s: %s specified twice as mirror",
3714 br->name, cfg->name);
3718 /* Get rid of deleted mirrors and add new mirrors. */
3719 SHASH_FOR_EACH (node, &old_mirrors) {
3720 if (!shash_find(&new_mirrors, node->name)) {
3721 mirror_destroy(node->data);
3724 SHASH_FOR_EACH (node, &new_mirrors) {
3725 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3727 mirror = mirror_create(br, node->name);
3732 mirror_reconfigure_one(mirror, node->data);
3734 shash_destroy(&old_mirrors);
3735 shash_destroy(&new_mirrors);
3737 /* Update port reserved status. */
3738 for (i = 0; i < br->n_ports; i++) {
3739 br->ports[i]->is_mirror_output_port = false;
3741 for (i = 0; i < MAX_MIRRORS; i++) {
3742 struct mirror *m = br->mirrors[i];
3743 if (m && m->out_port) {
3744 m->out_port->is_mirror_output_port = true;
3748 /* Update flooded vlans (for RSPAN). */
3750 if (br->cfg->n_flood_vlans) {
3751 rspan_vlans = bitmap_allocate(4096);
3753 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3754 int64_t vlan = br->cfg->flood_vlans[i];
3755 if (vlan >= 0 && vlan < 4096) {
3756 bitmap_set1(rspan_vlans, vlan);
3757 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3760 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3765 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3770 static struct mirror *
3771 mirror_create(struct bridge *br, const char *name)
3776 for (i = 0; ; i++) {
3777 if (i >= MAX_MIRRORS) {
3778 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3779 "cannot create %s", br->name, MAX_MIRRORS, name);
3782 if (!br->mirrors[i]) {
3787 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3790 br->mirrors[i] = m = xzalloc(sizeof *m);
3793 m->name = xstrdup(name);
3794 shash_init(&m->src_ports);
3795 shash_init(&m->dst_ports);
3805 mirror_destroy(struct mirror *m)
3808 struct bridge *br = m->bridge;
3811 for (i = 0; i < br->n_ports; i++) {
3812 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3813 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3816 shash_destroy(&m->src_ports);
3817 shash_destroy(&m->dst_ports);
3820 m->bridge->mirrors[m->idx] = NULL;
3828 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3829 struct shash *names)
3833 for (i = 0; i < n_ports; i++) {
3834 const char *name = ports[i]->name;
3835 if (port_lookup(m->bridge, name)) {
3836 shash_add_once(names, name, NULL);
3838 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3839 "port %s", m->bridge->name, m->name, name);
3845 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3851 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3853 for (i = 0; i < cfg->n_select_vlan; i++) {
3854 int64_t vlan = cfg->select_vlan[i];
3855 if (vlan < 0 || vlan > 4095) {
3856 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3857 m->bridge->name, m->name, vlan);
3859 (*vlans)[n_vlans++] = vlan;
3866 vlan_is_mirrored(const struct mirror *m, int vlan)
3870 for (i = 0; i < m->n_vlans; i++) {
3871 if (m->vlans[i] == vlan) {
3879 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3883 for (i = 0; i < m->n_vlans; i++) {
3884 if (port_trunks_vlan(p, m->vlans[i])) {
3892 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3894 struct shash src_ports, dst_ports;
3895 mirror_mask_t mirror_bit;
3896 struct port *out_port;
3902 /* Get output port. */
3903 if (cfg->output_port) {
3904 out_port = port_lookup(m->bridge, cfg->output_port->name);
3906 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3907 m->bridge->name, m->name);
3913 if (cfg->output_vlan) {
3914 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3915 "output vlan; ignoring output vlan",
3916 m->bridge->name, m->name);
3918 } else if (cfg->output_vlan) {
3920 out_vlan = *cfg->output_vlan;
3922 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3923 m->bridge->name, m->name);
3928 shash_init(&src_ports);
3929 shash_init(&dst_ports);
3930 if (cfg->select_all) {
3931 for (i = 0; i < m->bridge->n_ports; i++) {
3932 const char *name = m->bridge->ports[i]->name;
3933 shash_add_once(&src_ports, name, NULL);
3934 shash_add_once(&dst_ports, name, NULL);
3939 /* Get ports, and drop duplicates and ports that don't exist. */
3940 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3942 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3945 /* Get all the vlans, and drop duplicate and invalid vlans. */
3946 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3949 /* Update mirror data. */
3950 if (!shash_equal_keys(&m->src_ports, &src_ports)
3951 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3952 || m->n_vlans != n_vlans
3953 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3954 || m->out_port != out_port
3955 || m->out_vlan != out_vlan) {
3956 bridge_flush(m->bridge);
3958 shash_swap(&m->src_ports, &src_ports);
3959 shash_swap(&m->dst_ports, &dst_ports);
3962 m->n_vlans = n_vlans;
3963 m->out_port = out_port;
3964 m->out_vlan = out_vlan;
3967 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3968 for (i = 0; i < m->bridge->n_ports; i++) {
3969 struct port *port = m->bridge->ports[i];
3971 if (shash_find(&m->src_ports, port->name)
3974 ? port_trunks_any_mirrored_vlan(m, port)
3975 : vlan_is_mirrored(m, port->vlan)))) {
3976 port->src_mirrors |= mirror_bit;
3978 port->src_mirrors &= ~mirror_bit;
3981 if (shash_find(&m->dst_ports, port->name)) {
3982 port->dst_mirrors |= mirror_bit;
3984 port->dst_mirrors &= ~mirror_bit;
3989 shash_destroy(&src_ports);
3990 shash_destroy(&dst_ports);