1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
50 #include "poll-loop.h"
51 #include "port-array.h"
52 #include "proc-net-compat.h"
56 #include "socket-util.h"
57 #include "stream-ssl.h"
63 #include "vswitchd/vswitch-idl.h"
64 #include "xenserver.h"
66 #include "sflow_api.h"
68 #define THIS_MODULE VLM_bridge
77 /* These members are always valid. */
78 struct port *port; /* Containing port. */
79 size_t port_ifidx; /* Index within containing port. */
80 char *name; /* Host network device name. */
81 tag_type tag; /* Tag associated with this interface. */
82 long long delay_expires; /* Time after which 'enabled' may change. */
84 /* These members are valid only after bridge_reconfigure() causes them to
86 int dp_ifidx; /* Index within kernel datapath. */
87 struct netdev *netdev; /* Network device. */
88 bool enabled; /* May be chosen for flows? */
90 /* This member is only valid *during* bridge_reconfigure(). */
91 const struct ovsrec_interface *cfg;
94 #define BOND_MASK 0xff
96 int iface_idx; /* Index of assigned iface, or -1 if none. */
97 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
98 tag_type iface_tag; /* Tag associated with iface_idx. */
101 #define MAX_MIRRORS 32
102 typedef uint32_t mirror_mask_t;
103 #define MIRROR_MASK_C(X) UINT32_C(X)
104 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
106 struct bridge *bridge;
110 /* Selection criteria. */
111 struct shash src_ports; /* Name is port name; data is always NULL. */
112 struct shash dst_ports; /* Name is port name; data is always NULL. */
117 struct port *out_port;
121 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
123 struct bridge *bridge;
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
130 /* An ordinary bridge port has 1 interface.
131 * A bridge port for bonding has at least 2 interfaces. */
132 struct iface **ifaces;
133 size_t n_ifaces, allocated_ifaces;
136 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
137 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
138 tag_type active_iface_tag; /* Tag for bcast flows. */
139 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
140 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
141 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
142 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
143 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
144 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
145 long long int bond_next_rebalance; /* Next rebalancing time. */
147 /* Port mirroring info. */
148 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
149 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
150 bool is_mirror_output_port; /* Does port mirroring send frames here? */
152 /* This member is only valid *during* bridge_reconfigure(). */
153 const struct ovsrec_port *cfg;
156 #define DP_MAX_PORTS 255
158 struct list node; /* Node in global list of bridges. */
159 char *name; /* User-specified arbitrary name. */
160 struct mac_learning *ml; /* MAC learning table. */
161 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
163 /* OpenFlow switch processing. */
164 struct ofproto *ofproto; /* OpenFlow switch. */
166 /* Kernel datapath information. */
167 struct dpif *dpif; /* Datapath. */
168 struct port_array ifaces; /* Indexed by kernel datapath port number. */
172 size_t n_ports, allocated_ports;
173 struct shash iface_by_name; /* "struct iface"s indexed by name. */
174 struct shash port_by_name; /* "struct port"s indexed by name. */
177 bool has_bonded_ports;
182 /* Port mirroring. */
183 struct mirror *mirrors[MAX_MIRRORS];
185 /* This member is only valid *during* bridge_reconfigure(). */
186 const struct ovsrec_bridge *cfg;
189 /* List of all bridges. */
190 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
192 /* Maximum number of datapaths. */
193 enum { DP_MAX = 256 };
195 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
196 static void bridge_destroy(struct bridge *);
197 static struct bridge *bridge_lookup(const char *name);
198 static unixctl_cb_func bridge_unixctl_dump_flows;
199 static int bridge_run_one(struct bridge *);
200 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
201 const struct bridge *br,
202 struct ovsrec_controller ***controllersp);
203 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
205 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
207 const struct sockaddr_in *managers,
209 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
210 static void bridge_fetch_dp_ifaces(struct bridge *);
211 static void bridge_flush(struct bridge *);
212 static void bridge_pick_local_hw_addr(struct bridge *,
213 uint8_t ea[ETH_ADDR_LEN],
214 struct iface **hw_addr_iface);
215 static uint64_t bridge_pick_datapath_id(struct bridge *,
216 const uint8_t bridge_ea[ETH_ADDR_LEN],
217 struct iface *hw_addr_iface);
218 static struct iface *bridge_get_local_iface(struct bridge *);
219 static uint64_t dpid_from_hash(const void *, size_t nbytes);
221 static unixctl_cb_func bridge_unixctl_fdb_show;
223 static void bond_init(void);
224 static void bond_run(struct bridge *);
225 static void bond_wait(struct bridge *);
226 static void bond_rebalance_port(struct port *);
227 static void bond_send_learning_packets(struct port *);
228 static void bond_enable_slave(struct iface *iface, bool enable);
230 static struct port *port_create(struct bridge *, const char *name);
231 static void port_reconfigure(struct port *, const struct ovsrec_port *);
232 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
233 static void port_destroy(struct port *);
234 static struct port *port_lookup(const struct bridge *, const char *name);
235 static struct iface *port_lookup_iface(const struct port *, const char *name);
236 static struct port *port_from_dp_ifidx(const struct bridge *,
238 static void port_update_bond_compat(struct port *);
239 static void port_update_vlan_compat(struct port *);
240 static void port_update_bonding(struct port *);
242 static struct mirror *mirror_create(struct bridge *, const char *name);
243 static void mirror_destroy(struct mirror *);
244 static void mirror_reconfigure(struct bridge *);
245 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
246 static bool vlan_is_mirrored(const struct mirror *, int vlan);
248 static struct iface *iface_create(struct port *port,
249 const struct ovsrec_interface *if_cfg);
250 static void iface_destroy(struct iface *);
251 static struct iface *iface_lookup(const struct bridge *, const char *name);
252 static struct iface *iface_from_dp_ifidx(const struct bridge *,
254 static bool iface_is_internal(const struct bridge *, const char *name);
255 static void iface_set_mac(struct iface *);
256 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
258 /* Hooks into ofproto processing. */
259 static struct ofhooks bridge_ofhooks;
261 /* Public functions. */
263 /* Adds the name of each interface used by a bridge, including local and
264 * internal ports, to 'svec'. */
266 bridge_get_ifaces(struct svec *svec)
268 struct bridge *br, *next;
271 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
272 for (i = 0; i < br->n_ports; i++) {
273 struct port *port = br->ports[i];
275 for (j = 0; j < port->n_ifaces; j++) {
276 struct iface *iface = port->ifaces[j];
277 if (iface->dp_ifidx < 0) {
278 VLOG_ERR("%s interface not in datapath %s, ignoring",
279 iface->name, dpif_name(br->dpif));
281 if (iface->dp_ifidx != ODPP_LOCAL) {
282 svec_add(svec, iface->name);
291 bridge_init(const struct ovsrec_open_vswitch *cfg)
293 struct svec bridge_names;
294 struct svec dpif_names, dpif_types;
297 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
299 svec_init(&bridge_names);
300 for (i = 0; i < cfg->n_bridges; i++) {
301 svec_add(&bridge_names, cfg->bridges[i]->name);
303 svec_sort(&bridge_names);
305 svec_init(&dpif_names);
306 svec_init(&dpif_types);
307 dp_enumerate_types(&dpif_types);
308 for (i = 0; i < dpif_types.n; i++) {
313 dp_enumerate_names(dpif_types.names[i], &dpif_names);
315 for (j = 0; j < dpif_names.n; j++) {
316 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
318 struct svec all_names;
321 svec_init(&all_names);
322 dpif_get_all_names(dpif, &all_names);
323 for (k = 0; k < all_names.n; k++) {
324 if (svec_contains(&bridge_names, all_names.names[k])) {
330 svec_destroy(&all_names);
335 svec_destroy(&bridge_names);
336 svec_destroy(&dpif_names);
337 svec_destroy(&dpif_types);
339 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
343 bridge_reconfigure(cfg);
348 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
350 /* XXX SSL should be configurable on a per-bridge basis. */
352 stream_ssl_set_private_key_file(ssl->private_key);
353 stream_ssl_set_certificate_file(ssl->certificate);
354 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
359 /* Attempt to create the network device 'iface_name' through the netdev
362 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
365 struct shash options;
369 shash_init(&options);
370 for (i = 0; i < iface_cfg->n_options; i++) {
371 shash_add(&options, iface_cfg->key_options[i],
372 xstrdup(iface_cfg->value_options[i]));
376 struct netdev_options netdev_options;
378 memset(&netdev_options, 0, sizeof netdev_options);
379 netdev_options.name = iface_cfg->name;
380 if (!strcmp(iface_cfg->type, "internal")) {
381 /* An "internal" config type maps to a netdev "system" type. */
382 netdev_options.type = "system";
384 netdev_options.type = iface_cfg->type;
386 netdev_options.args = &options;
387 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
389 error = netdev_open(&netdev_options, &iface->netdev);
392 netdev_get_carrier(iface->netdev, &iface->enabled);
394 } else if (iface->netdev) {
395 const char *netdev_type = netdev_get_type(iface->netdev);
396 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
397 ? iface_cfg->type : NULL;
399 /* An "internal" config type maps to a netdev "system" type. */
400 if (iface_type && !strcmp(iface_type, "internal")) {
401 iface_type = "system";
404 if (!iface_type || !strcmp(netdev_type, iface_type)) {
405 error = netdev_reconfigure(iface->netdev, &options);
407 VLOG_WARN("%s: attempting change device type from %s to %s",
408 iface_cfg->name, netdev_type, iface_type);
412 shash_destroy_free_data(&options);
418 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
420 return set_up_iface(iface_cfg, iface, false);
424 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
425 void *aux OVS_UNUSED)
427 if (!iface->netdev) {
428 int error = set_up_iface(iface->cfg, iface, true);
430 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
440 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
441 void *aux OVS_UNUSED)
443 if (iface->dp_ifidx >= 0) {
444 VLOG_DBG("%s has interface %s on port %d",
446 iface->name, iface->dp_ifidx);
449 VLOG_ERR("%s interface not in %s, dropping",
450 iface->name, dpif_name(br->dpif));
456 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
457 void *aux OVS_UNUSED)
459 /* Set policing attributes. */
460 netdev_set_policing(iface->netdev,
461 iface->cfg->ingress_policing_rate,
462 iface->cfg->ingress_policing_burst);
464 /* Set MAC address of internal interfaces other than the local
466 if (iface->dp_ifidx != ODPP_LOCAL
467 && iface_is_internal(br, iface->name)) {
468 iface_set_mac(iface);
474 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
475 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
476 * deletes from 'br' any ports that no longer have any interfaces. */
478 iterate_and_prune_ifaces(struct bridge *br,
479 bool (*cb)(struct bridge *, struct iface *,
485 for (i = 0; i < br->n_ports; ) {
486 struct port *port = br->ports[i];
487 for (j = 0; j < port->n_ifaces; ) {
488 struct iface *iface = port->ifaces[j];
489 if (cb(br, iface, aux)) {
492 iface_destroy(iface);
496 if (port->n_ifaces) {
499 VLOG_ERR("%s port has no interfaces, dropping", port->name);
505 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
506 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
507 * responsible for freeing '*managersp' (with free()).
509 * You may be asking yourself "why does ovs-vswitchd care?", because
510 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
511 * should not be and in fact is not directly involved in that. But
512 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
513 * it has to tell in-band control where the managers are to enable that.
516 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
517 struct sockaddr_in **managersp, size_t *n_managersp)
519 struct sockaddr_in *managers = NULL;
520 size_t n_managers = 0;
522 if (ovs_cfg->n_managers > 0) {
525 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
526 for (i = 0; i < ovs_cfg->n_managers; i++) {
527 const char *name = ovs_cfg->managers[i];
528 struct sockaddr_in *sin = &managers[i];
530 if ((!strncmp(name, "tcp:", 4)
531 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
532 (!strncmp(name, "ssl:", 4)
533 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
539 *managersp = managers;
540 *n_managersp = n_managers;
544 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
546 struct ovsdb_idl_txn *txn;
547 struct shash old_br, new_br;
548 struct shash_node *node;
549 struct bridge *br, *next;
550 struct sockaddr_in *managers;
553 int sflow_bridge_number;
555 COVERAGE_INC(bridge_reconfigure);
557 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
559 collect_managers(ovs_cfg, &managers, &n_managers);
561 /* Collect old and new bridges. */
564 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
565 shash_add(&old_br, br->name, br);
567 for (i = 0; i < ovs_cfg->n_bridges; i++) {
568 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
569 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
570 VLOG_WARN("more than one bridge named %s", br_cfg->name);
574 /* Get rid of deleted bridges and add new bridges. */
575 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
576 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
583 SHASH_FOR_EACH (node, &new_br) {
584 const char *br_name = node->name;
585 const struct ovsrec_bridge *br_cfg = node->data;
586 br = shash_find_data(&old_br, br_name);
588 /* If the bridge datapath type has changed, we need to tear it
589 * down and recreate. */
590 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
592 bridge_create(br_cfg);
595 bridge_create(br_cfg);
598 shash_destroy(&old_br);
599 shash_destroy(&new_br);
603 bridge_configure_ssl(ovs_cfg->ssl);
606 /* Reconfigure all bridges. */
607 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
608 bridge_reconfigure_one(ovs_cfg, br);
611 /* Add and delete ports on all datapaths.
613 * The kernel will reject any attempt to add a given port to a datapath if
614 * that port already belongs to a different datapath, so we must do all
615 * port deletions before any port additions. */
616 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
617 struct odp_port *dpif_ports;
619 struct shash want_ifaces;
621 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
622 bridge_get_all_ifaces(br, &want_ifaces);
623 for (i = 0; i < n_dpif_ports; i++) {
624 const struct odp_port *p = &dpif_ports[i];
625 if (!shash_find(&want_ifaces, p->devname)
626 && strcmp(p->devname, br->name)) {
627 int retval = dpif_port_del(br->dpif, p->port);
629 VLOG_ERR("failed to remove %s interface from %s: %s",
630 p->devname, dpif_name(br->dpif),
635 shash_destroy(&want_ifaces);
638 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
639 struct odp_port *dpif_ports;
641 struct shash cur_ifaces, want_ifaces;
642 struct shash_node *node;
644 /* Get the set of interfaces currently in this datapath. */
645 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
646 shash_init(&cur_ifaces);
647 for (i = 0; i < n_dpif_ports; i++) {
648 const char *name = dpif_ports[i].devname;
649 if (!shash_find(&cur_ifaces, name)) {
650 shash_add(&cur_ifaces, name, NULL);
655 /* Get the set of interfaces we want on this datapath. */
656 bridge_get_all_ifaces(br, &want_ifaces);
658 SHASH_FOR_EACH (node, &want_ifaces) {
659 const char *if_name = node->name;
660 struct iface *iface = node->data;
662 if (shash_find(&cur_ifaces, if_name)) {
663 /* Already exists, just reconfigure it. */
665 reconfigure_iface(iface->cfg, iface);
668 /* Need to add to datapath. */
672 /* Add to datapath. */
673 internal = iface_is_internal(br, if_name);
674 error = dpif_port_add(br->dpif, if_name,
675 internal ? ODP_PORT_INTERNAL : 0, NULL);
676 if (error == EFBIG) {
677 VLOG_ERR("ran out of valid port numbers on %s",
678 dpif_name(br->dpif));
681 VLOG_ERR("failed to add %s interface to %s: %s",
682 if_name, dpif_name(br->dpif), strerror(error));
686 shash_destroy(&cur_ifaces);
687 shash_destroy(&want_ifaces);
689 sflow_bridge_number = 0;
690 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
693 struct iface *local_iface;
694 struct iface *hw_addr_iface;
697 bridge_fetch_dp_ifaces(br);
699 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
700 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
702 /* Pick local port hardware address, datapath ID. */
703 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
704 local_iface = bridge_get_local_iface(br);
706 int error = netdev_set_etheraddr(local_iface->netdev, ea);
708 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
709 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
710 "Ethernet address: %s",
711 br->name, strerror(error));
715 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
716 ofproto_set_datapath_id(br->ofproto, dpid);
718 dpid_string = xasprintf("%016"PRIx64, dpid);
719 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
722 /* Set NetFlow configuration on this bridge. */
723 if (br->cfg->netflow) {
724 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
725 struct netflow_options opts;
727 memset(&opts, 0, sizeof opts);
729 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
730 if (nf_cfg->engine_type) {
731 opts.engine_type = *nf_cfg->engine_type;
733 if (nf_cfg->engine_id) {
734 opts.engine_id = *nf_cfg->engine_id;
737 opts.active_timeout = nf_cfg->active_timeout;
738 if (!opts.active_timeout) {
739 opts.active_timeout = -1;
740 } else if (opts.active_timeout < 0) {
741 VLOG_WARN("bridge %s: active timeout interval set to negative "
742 "value, using default instead (%d seconds)", br->name,
743 NF_ACTIVE_TIMEOUT_DEFAULT);
744 opts.active_timeout = -1;
747 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
748 if (opts.add_id_to_iface) {
749 if (opts.engine_id > 0x7f) {
750 VLOG_WARN("bridge %s: netflow port mangling may conflict "
751 "with another vswitch, choose an engine id less "
752 "than 128", br->name);
754 if (br->n_ports > 508) {
755 VLOG_WARN("bridge %s: netflow port mangling will conflict "
756 "with another port when more than 508 ports are "
761 opts.collectors.n = nf_cfg->n_targets;
762 opts.collectors.names = nf_cfg->targets;
763 if (ofproto_set_netflow(br->ofproto, &opts)) {
764 VLOG_ERR("bridge %s: problem setting netflow collectors",
768 ofproto_set_netflow(br->ofproto, NULL);
771 /* Set sFlow configuration on this bridge. */
772 if (br->cfg->sflow) {
773 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
774 struct ovsrec_controller **controllers;
775 struct ofproto_sflow_options oso;
776 size_t n_controllers;
779 memset(&oso, 0, sizeof oso);
781 oso.targets.n = sflow_cfg->n_targets;
782 oso.targets.names = sflow_cfg->targets;
784 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
785 if (sflow_cfg->sampling) {
786 oso.sampling_rate = *sflow_cfg->sampling;
789 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
790 if (sflow_cfg->polling) {
791 oso.polling_interval = *sflow_cfg->polling;
794 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
795 if (sflow_cfg->header) {
796 oso.header_len = *sflow_cfg->header;
799 oso.sub_id = sflow_bridge_number++;
800 oso.agent_device = sflow_cfg->agent;
802 oso.control_ip = NULL;
803 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
804 for (i = 0; i < n_controllers; i++) {
805 if (controllers[i]->local_ip) {
806 oso.control_ip = controllers[i]->local_ip;
810 ofproto_set_sflow(br->ofproto, &oso);
812 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
814 ofproto_set_sflow(br->ofproto, NULL);
817 /* Update the controller and related settings. It would be more
818 * straightforward to call this from bridge_reconfigure_one(), but we
819 * can't do it there for two reasons. First, and most importantly, at
820 * that point we don't know the dp_ifidx of any interfaces that have
821 * been added to the bridge (because we haven't actually added them to
822 * the datapath). Second, at that point we haven't set the datapath ID
823 * yet; when a controller is configured, resetting the datapath ID will
824 * immediately disconnect from the controller, so it's better to set
825 * the datapath ID before the controller. */
826 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
828 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
829 for (i = 0; i < br->n_ports; i++) {
830 struct port *port = br->ports[i];
833 port_update_vlan_compat(port);
834 port_update_bonding(port);
836 for (j = 0; j < port->n_ifaces; j++) {
837 iface_update_qos(port->ifaces[j], port->cfg->qos);
841 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
842 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
845 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
847 ovsdb_idl_txn_commit(txn);
848 ovsdb_idl_txn_destroy(txn); /* XXX */
854 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
858 for (i = 0; i < n; i++) {
859 if (!strcmp(keys[i], key)) {
867 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
869 return get_ovsrec_key_value(key,
870 br_cfg->key_other_config,
871 br_cfg->value_other_config,
872 br_cfg->n_other_config);
876 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
877 struct iface **hw_addr_iface)
883 *hw_addr_iface = NULL;
885 /* Did the user request a particular MAC? */
886 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
887 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
888 if (eth_addr_is_multicast(ea)) {
889 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
890 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
891 } else if (eth_addr_is_zero(ea)) {
892 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
898 /* Otherwise choose the minimum non-local MAC address among all of the
900 memset(ea, 0xff, sizeof ea);
901 for (i = 0; i < br->n_ports; i++) {
902 struct port *port = br->ports[i];
903 uint8_t iface_ea[ETH_ADDR_LEN];
906 /* Mirror output ports don't participate. */
907 if (port->is_mirror_output_port) {
911 /* Choose the MAC address to represent the port. */
912 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
913 /* Find the interface with this Ethernet address (if any) so that
914 * we can provide the correct devname to the caller. */
916 for (j = 0; j < port->n_ifaces; j++) {
917 struct iface *candidate = port->ifaces[j];
918 uint8_t candidate_ea[ETH_ADDR_LEN];
919 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
920 && eth_addr_equals(iface_ea, candidate_ea)) {
925 /* Choose the interface whose MAC address will represent the port.
926 * The Linux kernel bonding code always chooses the MAC address of
927 * the first slave added to a bond, and the Fedora networking
928 * scripts always add slaves to a bond in alphabetical order, so
929 * for compatibility we choose the interface with the name that is
930 * first in alphabetical order. */
931 iface = port->ifaces[0];
932 for (j = 1; j < port->n_ifaces; j++) {
933 struct iface *candidate = port->ifaces[j];
934 if (strcmp(candidate->name, iface->name) < 0) {
939 /* The local port doesn't count (since we're trying to choose its
940 * MAC address anyway). */
941 if (iface->dp_ifidx == ODPP_LOCAL) {
946 error = netdev_get_etheraddr(iface->netdev, iface_ea);
948 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
949 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
950 iface->name, strerror(error));
955 /* Compare against our current choice. */
956 if (!eth_addr_is_multicast(iface_ea) &&
957 !eth_addr_is_local(iface_ea) &&
958 !eth_addr_is_reserved(iface_ea) &&
959 !eth_addr_is_zero(iface_ea) &&
960 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
962 memcpy(ea, iface_ea, ETH_ADDR_LEN);
963 *hw_addr_iface = iface;
966 if (eth_addr_is_multicast(ea)) {
967 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
968 *hw_addr_iface = NULL;
969 VLOG_WARN("bridge %s: using default bridge Ethernet "
970 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
972 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
973 br->name, ETH_ADDR_ARGS(ea));
977 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
978 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
979 * an interface on 'br', then that interface must be passed in as
980 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
981 * 'hw_addr_iface' must be passed in as a null pointer. */
983 bridge_pick_datapath_id(struct bridge *br,
984 const uint8_t bridge_ea[ETH_ADDR_LEN],
985 struct iface *hw_addr_iface)
988 * The procedure for choosing a bridge MAC address will, in the most
989 * ordinary case, also choose a unique MAC that we can use as a datapath
990 * ID. In some special cases, though, multiple bridges will end up with
991 * the same MAC address. This is OK for the bridges, but it will confuse
992 * the OpenFlow controller, because each datapath needs a unique datapath
995 * Datapath IDs must be unique. It is also very desirable that they be
996 * stable from one run to the next, so that policy set on a datapath
999 const char *datapath_id;
1002 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1003 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1007 if (hw_addr_iface) {
1009 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1011 * A bridge whose MAC address is taken from a VLAN network device
1012 * (that is, a network device created with vconfig(8) or similar
1013 * tool) will have the same MAC address as a bridge on the VLAN
1014 * device's physical network device.
1016 * Handle this case by hashing the physical network device MAC
1017 * along with the VLAN identifier.
1019 uint8_t buf[ETH_ADDR_LEN + 2];
1020 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1021 buf[ETH_ADDR_LEN] = vlan >> 8;
1022 buf[ETH_ADDR_LEN + 1] = vlan;
1023 return dpid_from_hash(buf, sizeof buf);
1026 * Assume that this bridge's MAC address is unique, since it
1027 * doesn't fit any of the cases we handle specially.
1032 * A purely internal bridge, that is, one that has no non-virtual
1033 * network devices on it at all, is more difficult because it has no
1034 * natural unique identifier at all.
1036 * When the host is a XenServer, we handle this case by hashing the
1037 * host's UUID with the name of the bridge. Names of bridges are
1038 * persistent across XenServer reboots, although they can be reused if
1039 * an internal network is destroyed and then a new one is later
1040 * created, so this is fairly effective.
1042 * When the host is not a XenServer, we punt by using a random MAC
1043 * address on each run.
1045 const char *host_uuid = xenserver_get_host_uuid();
1047 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1048 dpid = dpid_from_hash(combined, strlen(combined));
1054 return eth_addr_to_uint64(bridge_ea);
1058 dpid_from_hash(const void *data, size_t n)
1060 uint8_t hash[SHA1_DIGEST_SIZE];
1062 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1063 sha1_bytes(data, n, hash);
1064 eth_addr_mark_random(hash);
1065 return eth_addr_to_uint64(hash);
1071 struct bridge *br, *next;
1075 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1076 int error = bridge_run_one(br);
1078 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1079 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1080 "forcing reconfiguration", br->name);
1094 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1095 ofproto_wait(br->ofproto);
1096 if (ofproto_has_controller(br->ofproto)) {
1100 mac_learning_wait(br->ml);
1105 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1106 * configuration changes. */
1108 bridge_flush(struct bridge *br)
1110 COVERAGE_INC(bridge_flush);
1112 mac_learning_flush(br->ml);
1115 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1116 * such interface. */
1117 static struct iface *
1118 bridge_get_local_iface(struct bridge *br)
1122 for (i = 0; i < br->n_ports; i++) {
1123 struct port *port = br->ports[i];
1124 for (j = 0; j < port->n_ifaces; j++) {
1125 struct iface *iface = port->ifaces[j];
1126 if (iface->dp_ifidx == ODPP_LOCAL) {
1135 /* Bridge unixctl user interface functions. */
1137 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1138 const char *args, void *aux OVS_UNUSED)
1140 struct ds ds = DS_EMPTY_INITIALIZER;
1141 const struct bridge *br;
1142 const struct mac_entry *e;
1144 br = bridge_lookup(args);
1146 unixctl_command_reply(conn, 501, "no such bridge");
1150 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1151 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1152 if (e->port < 0 || e->port >= br->n_ports) {
1155 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1156 br->ports[e->port]->ifaces[0]->dp_ifidx,
1157 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1159 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1163 /* Bridge reconfiguration functions. */
1164 static struct bridge *
1165 bridge_create(const struct ovsrec_bridge *br_cfg)
1170 assert(!bridge_lookup(br_cfg->name));
1171 br = xzalloc(sizeof *br);
1173 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1179 dpif_flow_flush(br->dpif);
1181 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1184 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1186 dpif_delete(br->dpif);
1187 dpif_close(br->dpif);
1192 br->name = xstrdup(br_cfg->name);
1194 br->ml = mac_learning_create();
1195 eth_addr_nicira_random(br->default_ea);
1197 port_array_init(&br->ifaces);
1199 shash_init(&br->port_by_name);
1200 shash_init(&br->iface_by_name);
1204 list_push_back(&all_bridges, &br->node);
1206 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1212 bridge_destroy(struct bridge *br)
1217 while (br->n_ports > 0) {
1218 port_destroy(br->ports[br->n_ports - 1]);
1220 list_remove(&br->node);
1221 error = dpif_delete(br->dpif);
1222 if (error && error != ENOENT) {
1223 VLOG_ERR("failed to delete %s: %s",
1224 dpif_name(br->dpif), strerror(error));
1226 dpif_close(br->dpif);
1227 ofproto_destroy(br->ofproto);
1228 mac_learning_destroy(br->ml);
1229 port_array_destroy(&br->ifaces);
1230 shash_destroy(&br->port_by_name);
1231 shash_destroy(&br->iface_by_name);
1238 static struct bridge *
1239 bridge_lookup(const char *name)
1243 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1244 if (!strcmp(br->name, name)) {
1252 bridge_exists(const char *name)
1254 return bridge_lookup(name) ? true : false;
1258 bridge_get_datapathid(const char *name)
1260 struct bridge *br = bridge_lookup(name);
1261 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1264 /* Handle requests for a listing of all flows known by the OpenFlow
1265 * stack, including those normally hidden. */
1267 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1268 const char *args, void *aux OVS_UNUSED)
1273 br = bridge_lookup(args);
1275 unixctl_command_reply(conn, 501, "Unknown bridge");
1280 ofproto_get_all_flows(br->ofproto, &results);
1282 unixctl_command_reply(conn, 200, ds_cstr(&results));
1283 ds_destroy(&results);
1287 bridge_run_one(struct bridge *br)
1291 error = ofproto_run1(br->ofproto);
1296 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1299 error = ofproto_run2(br->ofproto, br->flush);
1306 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1307 const struct bridge *br,
1308 struct ovsrec_controller ***controllersp)
1310 struct ovsrec_controller **controllers;
1311 size_t n_controllers;
1313 if (br->cfg->n_controller) {
1314 controllers = br->cfg->controller;
1315 n_controllers = br->cfg->n_controller;
1317 controllers = ovs_cfg->controller;
1318 n_controllers = ovs_cfg->n_controller;
1321 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1327 *controllersp = controllers;
1329 return n_controllers;
1333 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1336 struct shash old_ports, new_ports;
1337 struct svec listeners, old_listeners;
1338 struct svec snoops, old_snoops;
1339 struct shash_node *node;
1342 /* Collect old ports. */
1343 shash_init(&old_ports);
1344 for (i = 0; i < br->n_ports; i++) {
1345 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1348 /* Collect new ports. */
1349 shash_init(&new_ports);
1350 for (i = 0; i < br->cfg->n_ports; i++) {
1351 const char *name = br->cfg->ports[i]->name;
1352 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1353 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1358 /* If we have a controller, then we need a local port. Complain if the
1359 * user didn't specify one.
1361 * XXX perhaps we should synthesize a port ourselves in this case. */
1362 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1363 char local_name[IF_NAMESIZE];
1366 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1367 local_name, sizeof local_name);
1368 if (!error && !shash_find(&new_ports, local_name)) {
1369 VLOG_WARN("bridge %s: controller specified but no local port "
1370 "(port named %s) defined",
1371 br->name, local_name);
1375 /* Get rid of deleted ports.
1376 * Get rid of deleted interfaces on ports that still exist. */
1377 SHASH_FOR_EACH (node, &old_ports) {
1378 struct port *port = node->data;
1379 const struct ovsrec_port *port_cfg;
1381 port_cfg = shash_find_data(&new_ports, node->name);
1385 port_del_ifaces(port, port_cfg);
1389 /* Create new ports.
1390 * Add new interfaces to existing ports.
1391 * Reconfigure existing ports. */
1392 SHASH_FOR_EACH (node, &new_ports) {
1393 struct port *port = shash_find_data(&old_ports, node->name);
1395 port = port_create(br, node->name);
1398 port_reconfigure(port, node->data);
1399 if (!port->n_ifaces) {
1400 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1401 br->name, port->name);
1405 shash_destroy(&old_ports);
1406 shash_destroy(&new_ports);
1408 /* Delete all flows if we're switching from connected to standalone or vice
1409 * versa. (XXX Should we delete all flows if we are switching from one
1410 * controller to another?) */
1412 /* Configure OpenFlow management listener. */
1413 svec_init(&listeners);
1414 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1415 ovs_rundir, br->name));
1416 svec_init(&old_listeners);
1417 ofproto_get_listeners(br->ofproto, &old_listeners);
1418 if (!svec_equal(&listeners, &old_listeners)) {
1419 ofproto_set_listeners(br->ofproto, &listeners);
1421 svec_destroy(&listeners);
1422 svec_destroy(&old_listeners);
1424 /* Configure OpenFlow controller connection snooping. */
1426 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1427 ovs_rundir, br->name));
1428 svec_init(&old_snoops);
1429 ofproto_get_snoops(br->ofproto, &old_snoops);
1430 if (!svec_equal(&snoops, &old_snoops)) {
1431 ofproto_set_snoops(br->ofproto, &snoops);
1433 svec_destroy(&snoops);
1434 svec_destroy(&old_snoops);
1436 mirror_reconfigure(br);
1440 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1442 const struct sockaddr_in *managers,
1445 struct ovsrec_controller **controllers;
1446 size_t n_controllers;
1448 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1450 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1451 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1452 ofproto_flush_flows(br->ofproto);
1455 if (!n_controllers) {
1456 union ofp_action action;
1459 /* Clear out controllers. */
1460 ofproto_set_controllers(br->ofproto, NULL, 0);
1462 /* Set up a flow that matches every packet and directs them to
1463 * OFPP_NORMAL (which goes to us). */
1464 memset(&action, 0, sizeof action);
1465 action.type = htons(OFPAT_OUTPUT);
1466 action.output.len = htons(sizeof action);
1467 action.output.port = htons(OFPP_NORMAL);
1468 memset(&flow, 0, sizeof flow);
1469 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1471 struct ofproto_controller *ocs;
1474 ocs = xmalloc(n_controllers * sizeof *ocs);
1475 for (i = 0; i < n_controllers; i++) {
1476 struct ovsrec_controller *c = controllers[i];
1477 struct ofproto_controller *oc = &ocs[i];
1479 if (strcmp(c->target, "discover")) {
1480 struct iface *local_iface;
1483 local_iface = bridge_get_local_iface(br);
1484 if (local_iface && c->local_ip
1485 && inet_aton(c->local_ip, &ip)) {
1486 struct netdev *netdev = local_iface->netdev;
1487 struct in_addr mask, gateway;
1489 if (!c->local_netmask
1490 || !inet_aton(c->local_netmask, &mask)) {
1493 if (!c->local_gateway
1494 || !inet_aton(c->local_gateway, &gateway)) {
1498 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1500 mask.s_addr = guess_netmask(ip.s_addr);
1502 if (!netdev_set_in4(netdev, ip, mask)) {
1503 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1505 br->name, IP_ARGS(&ip.s_addr),
1506 IP_ARGS(&mask.s_addr));
1509 if (gateway.s_addr) {
1510 if (!netdev_add_router(netdev, gateway)) {
1511 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1512 br->name, IP_ARGS(&gateway.s_addr));
1518 oc->target = c->target;
1519 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1520 oc->probe_interval = (c->inactivity_probe
1521 ? *c->inactivity_probe / 1000 : 5);
1522 oc->fail = (!c->fail_mode
1523 || !strcmp(c->fail_mode, "standalone")
1524 || !strcmp(c->fail_mode, "open")
1525 ? OFPROTO_FAIL_STANDALONE
1526 : OFPROTO_FAIL_SECURE);
1527 oc->band = (!c->connection_mode
1528 || !strcmp(c->connection_mode, "in-band")
1530 : OFPROTO_OUT_OF_BAND);
1531 oc->accept_re = c->discover_accept_regex;
1532 oc->update_resolv_conf = c->discover_update_resolv_conf;
1533 oc->rate_limit = (c->controller_rate_limit
1534 ? *c->controller_rate_limit : 0);
1535 oc->burst_limit = (c->controller_burst_limit
1536 ? *c->controller_burst_limit : 0);
1538 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1544 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1549 for (i = 0; i < br->n_ports; i++) {
1550 struct port *port = br->ports[i];
1551 for (j = 0; j < port->n_ifaces; j++) {
1552 struct iface *iface = port->ifaces[j];
1553 shash_add_once(ifaces, iface->name, iface);
1555 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1556 shash_add_once(ifaces, port->name, NULL);
1561 /* For robustness, in case the administrator moves around datapath ports behind
1562 * our back, we re-check all the datapath port numbers here.
1564 * This function will set the 'dp_ifidx' members of interfaces that have
1565 * disappeared to -1, so only call this function from a context where those
1566 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1567 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1568 * datapath, which doesn't support UINT16_MAX+1 ports. */
1570 bridge_fetch_dp_ifaces(struct bridge *br)
1572 struct odp_port *dpif_ports;
1573 size_t n_dpif_ports;
1576 /* Reset all interface numbers. */
1577 for (i = 0; i < br->n_ports; i++) {
1578 struct port *port = br->ports[i];
1579 for (j = 0; j < port->n_ifaces; j++) {
1580 struct iface *iface = port->ifaces[j];
1581 iface->dp_ifidx = -1;
1584 port_array_clear(&br->ifaces);
1586 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1587 for (i = 0; i < n_dpif_ports; i++) {
1588 struct odp_port *p = &dpif_ports[i];
1589 struct iface *iface = iface_lookup(br, p->devname);
1591 if (iface->dp_ifidx >= 0) {
1592 VLOG_WARN("%s reported interface %s twice",
1593 dpif_name(br->dpif), p->devname);
1594 } else if (iface_from_dp_ifidx(br, p->port)) {
1595 VLOG_WARN("%s reported interface %"PRIu16" twice",
1596 dpif_name(br->dpif), p->port);
1598 port_array_set(&br->ifaces, p->port, iface);
1599 iface->dp_ifidx = p->port;
1603 int64_t ofport = (iface->dp_ifidx >= 0
1604 ? odp_port_to_ofp_port(iface->dp_ifidx)
1606 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1613 /* Bridge packet processing functions. */
1616 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1618 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1621 static struct bond_entry *
1622 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1624 return &port->bond_hash[bond_hash(mac)];
1628 bond_choose_iface(const struct port *port)
1630 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1631 size_t i, best_down_slave = -1;
1632 long long next_delay_expiration = LLONG_MAX;
1634 for (i = 0; i < port->n_ifaces; i++) {
1635 struct iface *iface = port->ifaces[i];
1637 if (iface->enabled) {
1639 } else if (iface->delay_expires < next_delay_expiration) {
1640 best_down_slave = i;
1641 next_delay_expiration = iface->delay_expires;
1645 if (best_down_slave != -1) {
1646 struct iface *iface = port->ifaces[best_down_slave];
1648 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1649 "since no other interface is up", iface->name,
1650 iface->delay_expires - time_msec());
1651 bond_enable_slave(iface, true);
1654 return best_down_slave;
1658 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1659 uint16_t *dp_ifidx, tag_type *tags)
1661 struct iface *iface;
1663 assert(port->n_ifaces);
1664 if (port->n_ifaces == 1) {
1665 iface = port->ifaces[0];
1667 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1668 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1669 || !port->ifaces[e->iface_idx]->enabled) {
1670 /* XXX select interface properly. The current interface selection
1671 * is only good for testing the rebalancing code. */
1672 e->iface_idx = bond_choose_iface(port);
1673 if (e->iface_idx < 0) {
1674 *tags |= port->no_ifaces_tag;
1677 e->iface_tag = tag_create_random();
1678 ((struct port *) port)->bond_compat_is_stale = true;
1680 *tags |= e->iface_tag;
1681 iface = port->ifaces[e->iface_idx];
1683 *dp_ifidx = iface->dp_ifidx;
1684 *tags |= iface->tag; /* Currently only used for bonding. */
1689 bond_link_status_update(struct iface *iface, bool carrier)
1691 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1692 struct port *port = iface->port;
1694 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1695 /* Nothing to do. */
1698 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1699 iface->name, carrier ? "detected" : "dropped");
1700 if (carrier == iface->enabled) {
1701 iface->delay_expires = LLONG_MAX;
1702 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1703 iface->name, carrier ? "disabled" : "enabled");
1704 } else if (carrier && port->active_iface < 0) {
1705 bond_enable_slave(iface, true);
1706 if (port->updelay) {
1707 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1708 "other interface is up", iface->name, port->updelay);
1711 int delay = carrier ? port->updelay : port->downdelay;
1712 iface->delay_expires = time_msec() + delay;
1715 "interface %s: will be %s if it stays %s for %d ms",
1717 carrier ? "enabled" : "disabled",
1718 carrier ? "up" : "down",
1725 bond_choose_active_iface(struct port *port)
1727 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1729 port->active_iface = bond_choose_iface(port);
1730 port->active_iface_tag = tag_create_random();
1731 if (port->active_iface >= 0) {
1732 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1733 port->name, port->ifaces[port->active_iface]->name);
1735 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1741 bond_enable_slave(struct iface *iface, bool enable)
1743 struct port *port = iface->port;
1744 struct bridge *br = port->bridge;
1746 /* This acts as a recursion check. If the act of disabling a slave
1747 * causes a different slave to be enabled, the flag will allow us to
1748 * skip redundant work when we reenter this function. It must be
1749 * cleared on exit to keep things safe with multiple bonds. */
1750 static bool moving_active_iface = false;
1752 iface->delay_expires = LLONG_MAX;
1753 if (enable == iface->enabled) {
1757 iface->enabled = enable;
1758 if (!iface->enabled) {
1759 VLOG_WARN("interface %s: disabled", iface->name);
1760 ofproto_revalidate(br->ofproto, iface->tag);
1761 if (iface->port_ifidx == port->active_iface) {
1762 ofproto_revalidate(br->ofproto,
1763 port->active_iface_tag);
1765 /* Disabling a slave can lead to another slave being immediately
1766 * enabled if there will be no active slaves but one is waiting
1767 * on an updelay. In this case we do not need to run most of the
1768 * code for the newly enabled slave since there was no period
1769 * without an active slave and it is redundant with the disabling
1771 moving_active_iface = true;
1772 bond_choose_active_iface(port);
1774 bond_send_learning_packets(port);
1776 VLOG_WARN("interface %s: enabled", iface->name);
1777 if (port->active_iface < 0 && !moving_active_iface) {
1778 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1779 bond_choose_active_iface(port);
1780 bond_send_learning_packets(port);
1782 iface->tag = tag_create_random();
1785 moving_active_iface = false;
1786 port->bond_compat_is_stale = true;
1789 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1790 * bond interface. */
1792 bond_update_fake_iface_stats(struct port *port)
1794 struct netdev_stats bond_stats;
1795 struct netdev *bond_dev;
1798 memset(&bond_stats, 0, sizeof bond_stats);
1800 for (i = 0; i < port->n_ifaces; i++) {
1801 struct netdev_stats slave_stats;
1803 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1804 /* XXX: We swap the stats here because they are swapped back when
1805 * reported by the internal device. The reason for this is
1806 * internal devices normally represent packets going into the system
1807 * but when used as fake bond device they represent packets leaving
1808 * the system. We really should do this in the internal device
1809 * itself because changing it here reverses the counts from the
1810 * perspective of the switch. However, the internal device doesn't
1811 * know what type of device it represents so we have to do it here
1813 bond_stats.tx_packets += slave_stats.rx_packets;
1814 bond_stats.tx_bytes += slave_stats.rx_bytes;
1815 bond_stats.rx_packets += slave_stats.tx_packets;
1816 bond_stats.rx_bytes += slave_stats.tx_bytes;
1820 if (!netdev_open_default(port->name, &bond_dev)) {
1821 netdev_set_stats(bond_dev, &bond_stats);
1822 netdev_close(bond_dev);
1827 bond_run(struct bridge *br)
1831 for (i = 0; i < br->n_ports; i++) {
1832 struct port *port = br->ports[i];
1834 if (port->n_ifaces >= 2) {
1835 for (j = 0; j < port->n_ifaces; j++) {
1836 struct iface *iface = port->ifaces[j];
1837 if (time_msec() >= iface->delay_expires) {
1838 bond_enable_slave(iface, !iface->enabled);
1842 if (port->bond_fake_iface
1843 && time_msec() >= port->bond_next_fake_iface_update) {
1844 bond_update_fake_iface_stats(port);
1845 port->bond_next_fake_iface_update = time_msec() + 1000;
1849 if (port->bond_compat_is_stale) {
1850 port->bond_compat_is_stale = false;
1851 port_update_bond_compat(port);
1857 bond_wait(struct bridge *br)
1861 for (i = 0; i < br->n_ports; i++) {
1862 struct port *port = br->ports[i];
1863 if (port->n_ifaces < 2) {
1866 for (j = 0; j < port->n_ifaces; j++) {
1867 struct iface *iface = port->ifaces[j];
1868 if (iface->delay_expires != LLONG_MAX) {
1869 poll_timer_wait_until(iface->delay_expires);
1872 if (port->bond_fake_iface) {
1873 poll_timer_wait_until(port->bond_next_fake_iface_update);
1879 set_dst(struct dst *p, const flow_t *flow,
1880 const struct port *in_port, const struct port *out_port,
1883 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1884 : in_port->vlan >= 0 ? in_port->vlan
1885 : ntohs(flow->dl_vlan));
1886 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1890 swap_dst(struct dst *p, struct dst *q)
1892 struct dst tmp = *p;
1897 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1898 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1899 * that we push to the datapath. We could in fact fully sort the array by
1900 * vlan, but in most cases there are at most two different vlan tags so that's
1901 * possibly overkill.) */
1903 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1905 struct dst *first = dsts;
1906 struct dst *last = dsts + n_dsts;
1908 while (first != last) {
1910 * - All dsts < first have vlan == 'vlan'.
1911 * - All dsts >= last have vlan != 'vlan'.
1912 * - first < last. */
1913 while (first->vlan == vlan) {
1914 if (++first == last) {
1919 /* Same invariants, plus one additional:
1920 * - first->vlan != vlan.
1922 while (last[-1].vlan != vlan) {
1923 if (--last == first) {
1928 /* Same invariants, plus one additional:
1929 * - last[-1].vlan == vlan.*/
1930 swap_dst(first++, --last);
1935 mirror_mask_ffs(mirror_mask_t mask)
1937 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1942 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1943 const struct dst *test)
1946 for (i = 0; i < n_dsts; i++) {
1947 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1955 port_trunks_vlan(const struct port *port, uint16_t vlan)
1957 return (port->vlan < 0
1958 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
1962 port_includes_vlan(const struct port *port, uint16_t vlan)
1964 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1968 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1969 const struct port *in_port, const struct port *out_port,
1970 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1972 mirror_mask_t mirrors = in_port->src_mirrors;
1973 struct dst *dst = dsts;
1976 if (out_port == FLOOD_PORT) {
1977 /* XXX use ODP_FLOOD if no vlans or bonding. */
1978 /* XXX even better, define each VLAN as a datapath port group */
1979 for (i = 0; i < br->n_ports; i++) {
1980 struct port *port = br->ports[i];
1981 if (port != in_port && port_includes_vlan(port, vlan)
1982 && !port->is_mirror_output_port
1983 && set_dst(dst, flow, in_port, port, tags)) {
1984 mirrors |= port->dst_mirrors;
1988 *nf_output_iface = NF_OUT_FLOOD;
1989 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1990 *nf_output_iface = dst->dp_ifidx;
1991 mirrors |= out_port->dst_mirrors;
1996 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1997 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1999 if (set_dst(dst, flow, in_port, m->out_port, tags)
2000 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2004 for (i = 0; i < br->n_ports; i++) {
2005 struct port *port = br->ports[i];
2006 if (port_includes_vlan(port, m->out_vlan)
2007 && set_dst(dst, flow, in_port, port, tags))
2011 if (port->vlan < 0) {
2012 dst->vlan = m->out_vlan;
2014 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2018 /* Use the vlan tag on the original flow instead of
2019 * the one passed in the vlan parameter. This ensures
2020 * that we compare the vlan from before any implicit
2021 * tagging tags place. This is necessary because
2022 * dst->vlan is the final vlan, after removing implicit
2024 flow_vlan = ntohs(flow->dl_vlan);
2025 if (flow_vlan == 0) {
2026 flow_vlan = OFP_VLAN_NONE;
2028 if (port == in_port && dst->vlan == flow_vlan) {
2029 /* Don't send out input port on same VLAN. */
2037 mirrors &= mirrors - 1;
2040 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2044 static void OVS_UNUSED
2045 print_dsts(const struct dst *dsts, size_t n)
2047 for (; n--; dsts++) {
2048 printf(">p%"PRIu16, dsts->dp_ifidx);
2049 if (dsts->vlan != OFP_VLAN_NONE) {
2050 printf("v%"PRIu16, dsts->vlan);
2056 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2057 const struct port *in_port, const struct port *out_port,
2058 tag_type *tags, struct odp_actions *actions,
2059 uint16_t *nf_output_iface)
2061 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2063 const struct dst *p;
2066 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2069 cur_vlan = ntohs(flow->dl_vlan);
2070 for (p = dsts; p < &dsts[n_dsts]; p++) {
2071 union odp_action *a;
2072 if (p->vlan != cur_vlan) {
2073 if (p->vlan == OFP_VLAN_NONE) {
2074 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2076 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2077 a->vlan_vid.vlan_vid = htons(p->vlan);
2081 a = odp_actions_add(actions, ODPAT_OUTPUT);
2082 a->output.port = p->dp_ifidx;
2086 /* Returns the effective vlan of a packet, taking into account both the
2087 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2088 * the packet is untagged and -1 indicates it has an invalid header and
2089 * should be dropped. */
2090 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2091 struct port *in_port, bool have_packet)
2093 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2094 * belongs to VLAN 0, so we should treat both cases identically. (In the
2095 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2096 * presumably to allow a priority to be specified. In the latter case, the
2097 * packet does not have any 802.1Q header.) */
2098 int vlan = ntohs(flow->dl_vlan);
2099 if (vlan == OFP_VLAN_NONE) {
2102 if (in_port->vlan >= 0) {
2104 /* XXX support double tagging? */
2106 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2107 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2108 "packet received on port %s configured with "
2109 "implicit VLAN %"PRIu16,
2110 br->name, ntohs(flow->dl_vlan),
2111 in_port->name, in_port->vlan);
2115 vlan = in_port->vlan;
2117 if (!port_includes_vlan(in_port, vlan)) {
2119 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2120 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2121 "packet received on port %s not configured for "
2123 br->name, vlan, in_port->name, vlan);
2132 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2133 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2134 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2136 is_gratuitous_arp(const flow_t *flow)
2138 return (flow->dl_type == htons(ETH_TYPE_ARP)
2139 && eth_addr_is_broadcast(flow->dl_dst)
2140 && (flow->nw_proto == ARP_OP_REPLY
2141 || (flow->nw_proto == ARP_OP_REQUEST
2142 && flow->nw_src == flow->nw_dst)));
2146 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2147 struct port *in_port)
2149 enum grat_arp_lock_type lock_type;
2152 /* We don't want to learn from gratuitous ARP packets that are reflected
2153 * back over bond slaves so we lock the learning table. */
2154 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2155 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2156 GRAT_ARP_LOCK_CHECK;
2158 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2161 /* The log messages here could actually be useful in debugging,
2162 * so keep the rate limit relatively high. */
2163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2165 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2166 "on port %s in VLAN %d",
2167 br->name, ETH_ADDR_ARGS(flow->dl_src),
2168 in_port->name, vlan);
2169 ofproto_revalidate(br->ofproto, rev_tag);
2173 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2174 * dropped. Returns true if they may be forwarded, false if they should be
2177 * If 'have_packet' is true, it indicates that the caller is processing a
2178 * received packet. If 'have_packet' is false, then the caller is just
2179 * revalidating an existing flow because configuration has changed. Either
2180 * way, 'have_packet' only affects logging (there is no point in logging errors
2181 * during revalidation).
2183 * Sets '*in_portp' to the input port. This will be a null pointer if
2184 * flow->in_port does not designate a known input port (in which case
2185 * is_admissible() returns false).
2187 * When returning true, sets '*vlanp' to the effective VLAN of the input
2188 * packet, as returned by flow_get_vlan().
2190 * May also add tags to '*tags', although the current implementation only does
2191 * so in one special case.
2194 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2195 tag_type *tags, int *vlanp, struct port **in_portp)
2197 struct iface *in_iface;
2198 struct port *in_port;
2201 /* Find the interface and port structure for the received packet. */
2202 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2204 /* No interface? Something fishy... */
2206 /* Odd. A few possible reasons here:
2208 * - We deleted an interface but there are still a few packets
2209 * queued up from it.
2211 * - Someone externally added an interface (e.g. with "ovs-dpctl
2212 * add-if") that we don't know about.
2214 * - Packet arrived on the local port but the local port is not
2215 * one of our bridge ports.
2217 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2219 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2220 "interface %"PRIu16, br->name, flow->in_port);
2226 *in_portp = in_port = in_iface->port;
2227 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2232 /* Drop frames for reserved multicast addresses. */
2233 if (eth_addr_is_reserved(flow->dl_dst)) {
2237 /* Drop frames on ports reserved for mirroring. */
2238 if (in_port->is_mirror_output_port) {
2240 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2241 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2242 "%s, which is reserved exclusively for mirroring",
2243 br->name, in_port->name);
2248 /* Packets received on bonds need special attention to avoid duplicates. */
2249 if (in_port->n_ifaces > 1) {
2251 bool is_grat_arp_locked;
2253 if (eth_addr_is_multicast(flow->dl_dst)) {
2254 *tags |= in_port->active_iface_tag;
2255 if (in_port->active_iface != in_iface->port_ifidx) {
2256 /* Drop all multicast packets on inactive slaves. */
2261 /* Drop all packets for which we have learned a different input
2262 * port, because we probably sent the packet on one slave and got
2263 * it back on the other. Gratuitous ARP packets are an exception
2264 * to this rule: the host has moved to another switch. The exception
2265 * to the exception is if we locked the learning table to avoid
2266 * reflections on bond slaves. If this is the case, just drop the
2268 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2269 &is_grat_arp_locked);
2270 if (src_idx != -1 && src_idx != in_port->port_idx &&
2271 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2279 /* If the composed actions may be applied to any packet in the given 'flow',
2280 * returns true. Otherwise, the actions should only be applied to 'packet', or
2281 * not at all, if 'packet' was NULL. */
2283 process_flow(struct bridge *br, const flow_t *flow,
2284 const struct ofpbuf *packet, struct odp_actions *actions,
2285 tag_type *tags, uint16_t *nf_output_iface)
2287 struct port *in_port;
2288 struct port *out_port;
2292 /* Check whether we should drop packets in this flow. */
2293 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2298 /* Learn source MAC (but don't try to learn from revalidation). */
2300 update_learning_table(br, flow, vlan, in_port);
2303 /* Determine output port. */
2304 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2306 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2307 out_port = br->ports[out_port_idx];
2308 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2309 /* If we are revalidating but don't have a learning entry then
2310 * eject the flow. Installing a flow that floods packets opens
2311 * up a window of time where we could learn from a packet reflected
2312 * on a bond and blackhole packets before the learning table is
2313 * updated to reflect the correct port. */
2316 out_port = FLOOD_PORT;
2319 /* Don't send packets out their input ports. */
2320 if (in_port == out_port) {
2326 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2333 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2336 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2337 const struct ofp_phy_port *opp,
2340 struct bridge *br = br_;
2341 struct iface *iface;
2344 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2350 if (reason == OFPPR_DELETE) {
2351 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2352 br->name, iface->name);
2353 iface_destroy(iface);
2354 if (!port->n_ifaces) {
2355 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2356 br->name, port->name);
2362 if (port->n_ifaces > 1) {
2363 bool up = !(opp->state & OFPPS_LINK_DOWN);
2364 bond_link_status_update(iface, up);
2365 port_update_bond_compat(port);
2371 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2372 struct odp_actions *actions, tag_type *tags,
2373 uint16_t *nf_output_iface, void *br_)
2375 struct bridge *br = br_;
2377 COVERAGE_INC(bridge_process_flow);
2378 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2382 bridge_account_flow_ofhook_cb(const flow_t *flow,
2383 const union odp_action *actions,
2384 size_t n_actions, unsigned long long int n_bytes,
2387 struct bridge *br = br_;
2388 const union odp_action *a;
2389 struct port *in_port;
2393 /* Feed information from the active flows back into the learning table
2394 * to ensure that table is always in sync with what is actually flowing
2395 * through the datapath. */
2396 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2397 update_learning_table(br, flow, vlan, in_port);
2400 if (!br->has_bonded_ports) {
2404 for (a = actions; a < &actions[n_actions]; a++) {
2405 if (a->type == ODPAT_OUTPUT) {
2406 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2407 if (out_port && out_port->n_ifaces >= 2) {
2408 struct bond_entry *e = lookup_bond_entry(out_port,
2410 e->tx_bytes += n_bytes;
2417 bridge_account_checkpoint_ofhook_cb(void *br_)
2419 struct bridge *br = br_;
2423 if (!br->has_bonded_ports) {
2428 for (i = 0; i < br->n_ports; i++) {
2429 struct port *port = br->ports[i];
2430 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2431 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2432 bond_rebalance_port(port);
2437 static struct ofhooks bridge_ofhooks = {
2438 bridge_port_changed_ofhook_cb,
2439 bridge_normal_ofhook_cb,
2440 bridge_account_flow_ofhook_cb,
2441 bridge_account_checkpoint_ofhook_cb,
2444 /* Bonding functions. */
2446 /* Statistics for a single interface on a bonded port, used for load-based
2447 * bond rebalancing. */
2448 struct slave_balance {
2449 struct iface *iface; /* The interface. */
2450 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2452 /* All the "bond_entry"s that are assigned to this interface, in order of
2453 * increasing tx_bytes. */
2454 struct bond_entry **hashes;
2458 /* Sorts pointers to pointers to bond_entries in ascending order by the
2459 * interface to which they are assigned, and within a single interface in
2460 * ascending order of bytes transmitted. */
2462 compare_bond_entries(const void *a_, const void *b_)
2464 const struct bond_entry *const *ap = a_;
2465 const struct bond_entry *const *bp = b_;
2466 const struct bond_entry *a = *ap;
2467 const struct bond_entry *b = *bp;
2468 if (a->iface_idx != b->iface_idx) {
2469 return a->iface_idx > b->iface_idx ? 1 : -1;
2470 } else if (a->tx_bytes != b->tx_bytes) {
2471 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2477 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2478 * *descending* order by number of bytes transmitted. */
2480 compare_slave_balance(const void *a_, const void *b_)
2482 const struct slave_balance *a = a_;
2483 const struct slave_balance *b = b_;
2484 if (a->iface->enabled != b->iface->enabled) {
2485 return a->iface->enabled ? -1 : 1;
2486 } else if (a->tx_bytes != b->tx_bytes) {
2487 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2494 swap_bals(struct slave_balance *a, struct slave_balance *b)
2496 struct slave_balance tmp = *a;
2501 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2502 * given that 'p' (and only 'p') might be in the wrong location.
2504 * This function invalidates 'p', since it might now be in a different memory
2507 resort_bals(struct slave_balance *p,
2508 struct slave_balance bals[], size_t n_bals)
2511 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2512 swap_bals(p, p - 1);
2514 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2515 swap_bals(p, p + 1);
2521 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2523 if (VLOG_IS_DBG_ENABLED()) {
2524 struct ds ds = DS_EMPTY_INITIALIZER;
2525 const struct slave_balance *b;
2527 for (b = bals; b < bals + n_bals; b++) {
2531 ds_put_char(&ds, ',');
2533 ds_put_format(&ds, " %s %"PRIu64"kB",
2534 b->iface->name, b->tx_bytes / 1024);
2536 if (!b->iface->enabled) {
2537 ds_put_cstr(&ds, " (disabled)");
2539 if (b->n_hashes > 0) {
2540 ds_put_cstr(&ds, " (");
2541 for (i = 0; i < b->n_hashes; i++) {
2542 const struct bond_entry *e = b->hashes[i];
2544 ds_put_cstr(&ds, " + ");
2546 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2547 e - port->bond_hash, e->tx_bytes / 1024);
2549 ds_put_cstr(&ds, ")");
2552 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2557 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2559 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2562 struct bond_entry *hash = from->hashes[hash_idx];
2563 struct port *port = from->iface->port;
2564 uint64_t delta = hash->tx_bytes;
2566 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2567 "from %s to %s (now carrying %"PRIu64"kB and "
2568 "%"PRIu64"kB load, respectively)",
2569 port->name, delta / 1024, hash - port->bond_hash,
2570 from->iface->name, to->iface->name,
2571 (from->tx_bytes - delta) / 1024,
2572 (to->tx_bytes + delta) / 1024);
2574 /* Delete element from from->hashes.
2576 * We don't bother to add the element to to->hashes because not only would
2577 * it require more work, the only purpose it would be to allow that hash to
2578 * be migrated to another slave in this rebalancing run, and there is no
2579 * point in doing that. */
2580 if (hash_idx == 0) {
2583 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2584 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2588 /* Shift load away from 'from' to 'to'. */
2589 from->tx_bytes -= delta;
2590 to->tx_bytes += delta;
2592 /* Arrange for flows to be revalidated. */
2593 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2594 hash->iface_idx = to->iface->port_ifidx;
2595 hash->iface_tag = tag_create_random();
2599 bond_rebalance_port(struct port *port)
2601 struct slave_balance bals[DP_MAX_PORTS];
2603 struct bond_entry *hashes[BOND_MASK + 1];
2604 struct slave_balance *b, *from, *to;
2605 struct bond_entry *e;
2608 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2609 * descending order of tx_bytes, so that bals[0] represents the most
2610 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2613 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2614 * array for each slave_balance structure, we sort our local array of
2615 * hashes in order by slave, so that all of the hashes for a given slave
2616 * become contiguous in memory, and then we point each 'hashes' members of
2617 * a slave_balance structure to the start of a contiguous group. */
2618 n_bals = port->n_ifaces;
2619 for (b = bals; b < &bals[n_bals]; b++) {
2620 b->iface = port->ifaces[b - bals];
2625 for (i = 0; i <= BOND_MASK; i++) {
2626 hashes[i] = &port->bond_hash[i];
2628 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2629 for (i = 0; i <= BOND_MASK; i++) {
2631 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2632 b = &bals[e->iface_idx];
2633 b->tx_bytes += e->tx_bytes;
2635 b->hashes = &hashes[i];
2640 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2641 log_bals(bals, n_bals, port);
2643 /* Discard slaves that aren't enabled (which were sorted to the back of the
2644 * array earlier). */
2645 while (!bals[n_bals - 1].iface->enabled) {
2652 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2653 to = &bals[n_bals - 1];
2654 for (from = bals; from < to; ) {
2655 uint64_t overload = from->tx_bytes - to->tx_bytes;
2656 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2657 /* The extra load on 'from' (and all less-loaded slaves), compared
2658 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2659 * it is less than ~1Mbps. No point in rebalancing. */
2661 } else if (from->n_hashes == 1) {
2662 /* 'from' only carries a single MAC hash, so we can't shift any
2663 * load away from it, even though we want to. */
2666 /* 'from' is carrying significantly more load than 'to', and that
2667 * load is split across at least two different hashes. Pick a hash
2668 * to migrate to 'to' (the least-loaded slave), given that doing so
2669 * must decrease the ratio of the load on the two slaves by at
2672 * The sort order we use means that we prefer to shift away the
2673 * smallest hashes instead of the biggest ones. There is little
2674 * reason behind this decision; we could use the opposite sort
2675 * order to shift away big hashes ahead of small ones. */
2679 for (i = 0; i < from->n_hashes; i++) {
2680 double old_ratio, new_ratio;
2681 uint64_t delta = from->hashes[i]->tx_bytes;
2683 if (delta == 0 || from->tx_bytes - delta == 0) {
2684 /* Pointless move. */
2688 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2690 if (to->tx_bytes == 0) {
2691 /* Nothing on the new slave, move it. */
2695 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2696 new_ratio = (double)(from->tx_bytes - delta) /
2697 (to->tx_bytes + delta);
2699 if (new_ratio == 0) {
2700 /* Should already be covered but check to prevent division
2705 if (new_ratio < 1) {
2706 new_ratio = 1 / new_ratio;
2709 if (old_ratio - new_ratio > 0.1) {
2710 /* Would decrease the ratio, move it. */
2714 if (i < from->n_hashes) {
2715 bond_shift_load(from, to, i);
2716 port->bond_compat_is_stale = true;
2718 /* If the result of the migration changed the relative order of
2719 * 'from' and 'to' swap them back to maintain invariants. */
2720 if (order_swapped) {
2721 swap_bals(from, to);
2724 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2725 * point to different slave_balance structures. It is only
2726 * valid to do these two operations in a row at all because we
2727 * know that 'from' will not move past 'to' and vice versa. */
2728 resort_bals(from, bals, n_bals);
2729 resort_bals(to, bals, n_bals);
2736 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2737 * historical data to decay to <1% in 7 rebalancing runs. */
2738 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2744 bond_send_learning_packets(struct port *port)
2746 struct bridge *br = port->bridge;
2747 struct mac_entry *e;
2748 struct ofpbuf packet;
2749 int error, n_packets, n_errors;
2751 if (!port->n_ifaces || port->active_iface < 0) {
2755 ofpbuf_init(&packet, 128);
2756 error = n_packets = n_errors = 0;
2757 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2758 union ofp_action actions[2], *a;
2764 if (e->port == port->port_idx
2765 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2769 /* Compose actions. */
2770 memset(actions, 0, sizeof actions);
2773 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2774 a->vlan_vid.len = htons(sizeof *a);
2775 a->vlan_vid.vlan_vid = htons(e->vlan);
2778 a->output.type = htons(OFPAT_OUTPUT);
2779 a->output.len = htons(sizeof *a);
2780 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2785 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2787 flow_extract(&packet, 0, ODPP_NONE, &flow);
2788 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2795 ofpbuf_uninit(&packet);
2798 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2799 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2800 "packets, last error was: %s",
2801 port->name, n_errors, n_packets, strerror(error));
2803 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2804 port->name, n_packets);
2808 /* Bonding unixctl user interface functions. */
2811 bond_unixctl_list(struct unixctl_conn *conn,
2812 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2814 struct ds ds = DS_EMPTY_INITIALIZER;
2815 const struct bridge *br;
2817 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2819 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2822 for (i = 0; i < br->n_ports; i++) {
2823 const struct port *port = br->ports[i];
2824 if (port->n_ifaces > 1) {
2827 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2828 for (j = 0; j < port->n_ifaces; j++) {
2829 const struct iface *iface = port->ifaces[j];
2831 ds_put_cstr(&ds, ", ");
2833 ds_put_cstr(&ds, iface->name);
2835 ds_put_char(&ds, '\n');
2839 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2843 static struct port *
2844 bond_find(const char *name)
2846 const struct bridge *br;
2848 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2851 for (i = 0; i < br->n_ports; i++) {
2852 struct port *port = br->ports[i];
2853 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2862 bond_unixctl_show(struct unixctl_conn *conn,
2863 const char *args, void *aux OVS_UNUSED)
2865 struct ds ds = DS_EMPTY_INITIALIZER;
2866 const struct port *port;
2869 port = bond_find(args);
2871 unixctl_command_reply(conn, 501, "no such bond");
2875 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2876 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2877 ds_put_format(&ds, "next rebalance: %lld ms\n",
2878 port->bond_next_rebalance - time_msec());
2879 for (j = 0; j < port->n_ifaces; j++) {
2880 const struct iface *iface = port->ifaces[j];
2881 struct bond_entry *be;
2884 ds_put_format(&ds, "slave %s: %s\n",
2885 iface->name, iface->enabled ? "enabled" : "disabled");
2886 if (j == port->active_iface) {
2887 ds_put_cstr(&ds, "\tactive slave\n");
2889 if (iface->delay_expires != LLONG_MAX) {
2890 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2891 iface->enabled ? "downdelay" : "updelay",
2892 iface->delay_expires - time_msec());
2896 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2897 int hash = be - port->bond_hash;
2898 struct mac_entry *me;
2900 if (be->iface_idx != j) {
2904 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2905 hash, be->tx_bytes / 1024);
2908 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2909 &port->bridge->ml->lrus) {
2912 if (bond_hash(me->mac) == hash
2913 && me->port != port->port_idx
2914 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2915 && dp_ifidx == iface->dp_ifidx)
2917 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2918 ETH_ADDR_ARGS(me->mac));
2923 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2928 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2929 void *aux OVS_UNUSED)
2931 char *args = (char *) args_;
2932 char *save_ptr = NULL;
2933 char *bond_s, *hash_s, *slave_s;
2934 uint8_t mac[ETH_ADDR_LEN];
2936 struct iface *iface;
2937 struct bond_entry *entry;
2940 bond_s = strtok_r(args, " ", &save_ptr);
2941 hash_s = strtok_r(NULL, " ", &save_ptr);
2942 slave_s = strtok_r(NULL, " ", &save_ptr);
2944 unixctl_command_reply(conn, 501,
2945 "usage: bond/migrate BOND HASH SLAVE");
2949 port = bond_find(bond_s);
2951 unixctl_command_reply(conn, 501, "no such bond");
2955 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2956 == ETH_ADDR_SCAN_COUNT) {
2957 hash = bond_hash(mac);
2958 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2959 hash = atoi(hash_s) & BOND_MASK;
2961 unixctl_command_reply(conn, 501, "bad hash");
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 migrate to disabled slave");
2976 entry = &port->bond_hash[hash];
2977 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2978 entry->iface_idx = iface->port_ifidx;
2979 entry->iface_tag = tag_create_random();
2980 port->bond_compat_is_stale = true;
2981 unixctl_command_reply(conn, 200, "migrated");
2985 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2986 void *aux OVS_UNUSED)
2988 char *args = (char *) args_;
2989 char *save_ptr = NULL;
2990 char *bond_s, *slave_s;
2992 struct iface *iface;
2994 bond_s = strtok_r(args, " ", &save_ptr);
2995 slave_s = strtok_r(NULL, " ", &save_ptr);
2997 unixctl_command_reply(conn, 501,
2998 "usage: bond/set-active-slave BOND SLAVE");
3002 port = bond_find(bond_s);
3004 unixctl_command_reply(conn, 501, "no such bond");
3008 iface = port_lookup_iface(port, slave_s);
3010 unixctl_command_reply(conn, 501, "no such slave");
3014 if (!iface->enabled) {
3015 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3019 if (port->active_iface != iface->port_ifidx) {
3020 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3021 port->active_iface = iface->port_ifidx;
3022 port->active_iface_tag = tag_create_random();
3023 VLOG_INFO("port %s: active interface is now %s",
3024 port->name, iface->name);
3025 bond_send_learning_packets(port);
3026 unixctl_command_reply(conn, 200, "done");
3028 unixctl_command_reply(conn, 200, "no change");
3033 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3035 char *args = (char *) args_;
3036 char *save_ptr = NULL;
3037 char *bond_s, *slave_s;
3039 struct iface *iface;
3041 bond_s = strtok_r(args, " ", &save_ptr);
3042 slave_s = strtok_r(NULL, " ", &save_ptr);
3044 unixctl_command_reply(conn, 501,
3045 "usage: bond/enable/disable-slave BOND SLAVE");
3049 port = bond_find(bond_s);
3051 unixctl_command_reply(conn, 501, "no such bond");
3055 iface = port_lookup_iface(port, slave_s);
3057 unixctl_command_reply(conn, 501, "no such slave");
3061 bond_enable_slave(iface, enable);
3062 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3066 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3067 void *aux OVS_UNUSED)
3069 enable_slave(conn, args, true);
3073 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3074 void *aux OVS_UNUSED)
3076 enable_slave(conn, args, false);
3080 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3081 void *aux OVS_UNUSED)
3083 uint8_t mac[ETH_ADDR_LEN];
3087 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3088 == ETH_ADDR_SCAN_COUNT) {
3089 hash = bond_hash(mac);
3091 hash_cstr = xasprintf("%u", hash);
3092 unixctl_command_reply(conn, 200, hash_cstr);
3095 unixctl_command_reply(conn, 501, "invalid mac");
3102 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3103 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3104 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3105 unixctl_command_register("bond/set-active-slave",
3106 bond_unixctl_set_active_slave, NULL);
3107 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3109 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3111 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3114 /* Port functions. */
3116 static struct port *
3117 port_create(struct bridge *br, const char *name)
3121 port = xzalloc(sizeof *port);
3123 port->port_idx = br->n_ports;
3125 port->trunks = NULL;
3126 port->name = xstrdup(name);
3127 port->active_iface = -1;
3129 if (br->n_ports >= br->allocated_ports) {
3130 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3133 br->ports[br->n_ports++] = port;
3134 shash_add_assert(&br->port_by_name, port->name, port);
3136 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3143 get_port_other_config(const struct ovsrec_port *port, const char *key,
3144 const char *default_value)
3146 const char *value = get_ovsrec_key_value(key,
3147 port->key_other_config,
3148 port->value_other_config,
3149 port->n_other_config);
3150 return value ? value : default_value;
3154 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3156 struct shash new_ifaces;
3159 /* Collect list of new interfaces. */
3160 shash_init(&new_ifaces);
3161 for (i = 0; i < cfg->n_interfaces; i++) {
3162 const char *name = cfg->interfaces[i]->name;
3163 shash_add_once(&new_ifaces, name, NULL);
3166 /* Get rid of deleted interfaces. */
3167 for (i = 0; i < port->n_ifaces; ) {
3168 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3169 iface_destroy(port->ifaces[i]);
3175 shash_destroy(&new_ifaces);
3179 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3181 struct shash new_ifaces;
3182 long long int next_rebalance;
3183 unsigned long *trunks;
3189 /* Update settings. */
3190 port->updelay = cfg->bond_updelay;
3191 if (port->updelay < 0) {
3194 port->updelay = cfg->bond_downdelay;
3195 if (port->downdelay < 0) {
3196 port->downdelay = 0;
3198 port->bond_rebalance_interval = atoi(
3199 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3200 if (port->bond_rebalance_interval < 1000) {
3201 port->bond_rebalance_interval = 1000;
3203 next_rebalance = time_msec() + port->bond_rebalance_interval;
3204 if (port->bond_next_rebalance > next_rebalance) {
3205 port->bond_next_rebalance = next_rebalance;
3208 /* Add new interfaces and update 'cfg' member of existing ones. */
3209 shash_init(&new_ifaces);
3210 for (i = 0; i < cfg->n_interfaces; i++) {
3211 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3212 struct iface *iface;
3214 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3215 VLOG_WARN("port %s: %s specified twice as port interface",
3216 port->name, if_cfg->name);
3220 iface = iface_lookup(port->bridge, if_cfg->name);
3222 if (iface->port != port) {
3223 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3225 port->bridge->name, if_cfg->name, iface->port->name);
3228 iface->cfg = if_cfg;
3230 iface_create(port, if_cfg);
3233 shash_destroy(&new_ifaces);
3238 if (port->n_ifaces < 2) {
3240 if (vlan >= 0 && vlan <= 4095) {
3241 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3246 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3247 * they even work as-is. But they have not been tested. */
3248 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3252 if (port->vlan != vlan) {
3254 bridge_flush(port->bridge);
3257 /* Get trunked VLANs. */
3259 if (vlan < 0 && cfg->n_trunks) {
3263 trunks = bitmap_allocate(4096);
3265 for (i = 0; i < cfg->n_trunks; i++) {
3266 int trunk = cfg->trunks[i];
3268 bitmap_set1(trunks, trunk);
3274 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3275 port->name, cfg->n_trunks);
3277 if (n_errors == cfg->n_trunks) {
3278 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3280 bitmap_free(trunks);
3283 } else if (vlan >= 0 && cfg->n_trunks) {
3284 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3288 ? port->trunks != NULL
3289 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3290 bridge_flush(port->bridge);
3292 bitmap_free(port->trunks);
3293 port->trunks = trunks;
3297 port_destroy(struct port *port)
3300 struct bridge *br = port->bridge;
3304 proc_net_compat_update_vlan(port->name, NULL, 0);
3305 proc_net_compat_update_bond(port->name, NULL);
3307 for (i = 0; i < MAX_MIRRORS; i++) {
3308 struct mirror *m = br->mirrors[i];
3309 if (m && m->out_port == port) {
3314 while (port->n_ifaces > 0) {
3315 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3318 shash_find_and_delete_assert(&br->port_by_name, port->name);
3320 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3321 del->port_idx = port->port_idx;
3324 bitmap_free(port->trunks);
3331 static struct port *
3332 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3334 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3335 return iface ? iface->port : NULL;
3338 static struct port *
3339 port_lookup(const struct bridge *br, const char *name)
3341 return shash_find_data(&br->port_by_name, name);
3344 static struct iface *
3345 port_lookup_iface(const struct port *port, const char *name)
3347 struct iface *iface = iface_lookup(port->bridge, name);
3348 return iface && iface->port == port ? iface : NULL;
3352 port_update_bonding(struct port *port)
3354 if (port->n_ifaces < 2) {
3355 /* Not a bonded port. */
3356 if (port->bond_hash) {
3357 free(port->bond_hash);
3358 port->bond_hash = NULL;
3359 port->bond_compat_is_stale = true;
3360 port->bond_fake_iface = false;
3363 if (!port->bond_hash) {
3366 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3367 for (i = 0; i <= BOND_MASK; i++) {
3368 struct bond_entry *e = &port->bond_hash[i];
3372 port->no_ifaces_tag = tag_create_random();
3373 bond_choose_active_iface(port);
3374 port->bond_next_rebalance
3375 = time_msec() + port->bond_rebalance_interval;
3377 if (port->cfg->bond_fake_iface) {
3378 port->bond_next_fake_iface_update = time_msec();
3381 port->bond_compat_is_stale = true;
3382 port->bond_fake_iface = port->cfg->bond_fake_iface;
3387 port_update_bond_compat(struct port *port)
3389 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3390 struct compat_bond bond;
3393 if (port->n_ifaces < 2) {
3394 proc_net_compat_update_bond(port->name, NULL);
3399 bond.updelay = port->updelay;
3400 bond.downdelay = port->downdelay;
3403 bond.hashes = compat_hashes;
3404 if (port->bond_hash) {
3405 const struct bond_entry *e;
3406 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3407 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3408 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3409 cbh->hash = e - port->bond_hash;
3410 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3415 bond.n_slaves = port->n_ifaces;
3416 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3417 for (i = 0; i < port->n_ifaces; i++) {
3418 struct iface *iface = port->ifaces[i];
3419 struct compat_bond_slave *slave = &bond.slaves[i];
3420 slave->name = iface->name;
3422 /* We need to make the same determination as the Linux bonding
3423 * code to determine whether a slave should be consider "up".
3424 * The Linux function bond_miimon_inspect() supports four
3425 * BOND_LINK_* states:
3427 * - BOND_LINK_UP: carrier detected, updelay has passed.
3428 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3429 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3430 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3432 * The function bond_info_show_slave() only considers BOND_LINK_UP
3433 * to be "up" and anything else to be "down".
3435 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3439 netdev_get_etheraddr(iface->netdev, slave->mac);
3442 if (port->bond_fake_iface) {
3443 struct netdev *bond_netdev;
3445 if (!netdev_open_default(port->name, &bond_netdev)) {
3447 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3449 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3451 netdev_close(bond_netdev);
3455 proc_net_compat_update_bond(port->name, &bond);
3460 port_update_vlan_compat(struct port *port)
3462 struct bridge *br = port->bridge;
3463 char *vlandev_name = NULL;
3465 if (port->vlan > 0) {
3466 /* Figure out the name that the VLAN device should actually have, if it
3467 * existed. This takes some work because the VLAN device would not
3468 * have port->name in its name; rather, it would have the trunk port's
3469 * name, and 'port' would be attached to a bridge that also had the
3470 * VLAN device one of its ports. So we need to find a trunk port that
3471 * includes port->vlan.
3473 * There might be more than one candidate. This doesn't happen on
3474 * XenServer, so if it happens we just pick the first choice in
3475 * alphabetical order instead of creating multiple VLAN devices. */
3477 for (i = 0; i < br->n_ports; i++) {
3478 struct port *p = br->ports[i];
3479 if (port_trunks_vlan(p, port->vlan)
3481 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3483 uint8_t ea[ETH_ADDR_LEN];
3484 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3485 if (!eth_addr_is_multicast(ea) &&
3486 !eth_addr_is_reserved(ea) &&
3487 !eth_addr_is_zero(ea)) {
3488 vlandev_name = p->name;
3493 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3496 /* Interface functions. */
3498 static struct iface *
3499 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3501 struct bridge *br = port->bridge;
3502 struct iface *iface;
3503 char *name = if_cfg->name;
3506 iface = xzalloc(sizeof *iface);
3508 iface->port_ifidx = port->n_ifaces;
3509 iface->name = xstrdup(name);
3510 iface->dp_ifidx = -1;
3511 iface->tag = tag_create_random();
3512 iface->delay_expires = LLONG_MAX;
3513 iface->netdev = NULL;
3514 iface->cfg = if_cfg;
3516 shash_add_assert(&br->iface_by_name, iface->name, iface);
3518 /* Attempt to create the network interface in case it doesn't exist yet. */
3519 if (!iface_is_internal(br, iface->name)) {
3520 error = set_up_iface(if_cfg, iface, true);
3522 VLOG_WARN("could not create iface %s: %s", iface->name,
3525 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3532 if (port->n_ifaces >= port->allocated_ifaces) {
3533 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3534 sizeof *port->ifaces);
3536 port->ifaces[port->n_ifaces++] = iface;
3537 if (port->n_ifaces > 1) {
3538 br->has_bonded_ports = true;
3541 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3549 iface_destroy(struct iface *iface)
3552 struct port *port = iface->port;
3553 struct bridge *br = port->bridge;
3554 bool del_active = port->active_iface == iface->port_ifidx;
3557 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3559 if (iface->dp_ifidx >= 0) {
3560 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3563 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3564 del->port_ifidx = iface->port_ifidx;
3566 netdev_close(iface->netdev);
3569 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3570 bond_choose_active_iface(port);
3571 bond_send_learning_packets(port);
3577 bridge_flush(port->bridge);
3581 static struct iface *
3582 iface_lookup(const struct bridge *br, const char *name)
3584 return shash_find_data(&br->iface_by_name, name);
3587 static struct iface *
3588 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3590 return port_array_get(&br->ifaces, dp_ifidx);
3593 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3594 * 'br', that is, an interface that is entirely simulated within the datapath.
3595 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3596 * interfaces are created by setting "iface.<iface>.internal = true".
3598 * In addition, we have a kluge-y feature that creates an internal port with
3599 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3600 * This feature needs to go away in the long term. Until then, this is one
3601 * reason why this function takes a name instead of a struct iface: the fake
3602 * interfaces created this way do not have a struct iface. */
3604 iface_is_internal(const struct bridge *br, const char *if_name)
3606 struct iface *iface;
3609 if (!strcmp(if_name, br->name)) {
3613 iface = iface_lookup(br, if_name);
3614 if (iface && !strcmp(iface->cfg->type, "internal")) {
3618 port = port_lookup(br, if_name);
3619 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3625 /* Set Ethernet address of 'iface', if one is specified in the configuration
3628 iface_set_mac(struct iface *iface)
3630 uint8_t ea[ETH_ADDR_LEN];
3632 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3633 if (eth_addr_is_multicast(ea)) {
3634 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3636 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3637 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3638 iface->name, iface->name);
3640 int error = netdev_set_etheraddr(iface->netdev, ea);
3642 VLOG_ERR("interface %s: setting MAC failed (%s)",
3643 iface->name, strerror(error));
3650 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3651 struct shash *shash)
3656 for (i = 0; i < n; i++) {
3657 shash_add(shash, keys[i], values[i]);
3661 struct iface_delete_queues_cbdata {
3662 struct netdev *netdev;
3663 const int64_t *queue_ids;
3668 queue_ids_include(const int64_t *ids, size_t n, int64_t target)
3673 while (low < high) {
3674 size_t mid = low + (high - low) / 2;
3675 if (target > ids[mid]) {
3677 } else if (target < ids[mid]) {
3687 iface_delete_queues(unsigned int queue_id,
3688 const struct shash *details OVS_UNUSED, void *cbdata_)
3690 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3692 if (!queue_ids_include(cbdata->queue_ids, cbdata->n_queue_ids, queue_id)) {
3693 netdev_delete_queue(cbdata->netdev, queue_id);
3698 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3700 if (!qos || qos->type[0] == '\0') {
3701 netdev_set_qos(iface->netdev, NULL, NULL);
3703 struct iface_delete_queues_cbdata cbdata;
3704 struct shash details;
3707 /* Configure top-level Qos for 'iface'. */
3708 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3709 qos->n_other_config, &details);
3710 netdev_set_qos(iface->netdev, qos->type, &details);
3711 shash_destroy(&details);
3713 /* Deconfigure queues that were deleted. */
3714 cbdata.netdev = iface->netdev;
3715 cbdata.queue_ids = qos->key_queues;
3716 cbdata.n_queue_ids = qos->n_queues;
3717 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3719 /* Configure queues for 'iface'. */
3720 for (i = 0; i < qos->n_queues; i++) {
3721 const struct ovsrec_queue *queue = qos->value_queues[i];
3722 unsigned int queue_id = qos->key_queues[i];
3724 shash_from_ovs_idl_map(queue->key_other_config,
3725 queue->value_other_config,
3726 queue->n_other_config, &details);
3727 netdev_set_queue(iface->netdev, queue_id, &details);
3728 shash_destroy(&details);
3733 /* Port mirroring. */
3736 mirror_reconfigure(struct bridge *br)
3738 struct shash old_mirrors, new_mirrors;
3739 struct shash_node *node;
3740 unsigned long *rspan_vlans;
3743 /* Collect old mirrors. */
3744 shash_init(&old_mirrors);
3745 for (i = 0; i < MAX_MIRRORS; i++) {
3746 if (br->mirrors[i]) {
3747 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3751 /* Collect new mirrors. */
3752 shash_init(&new_mirrors);
3753 for (i = 0; i < br->cfg->n_mirrors; i++) {
3754 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3755 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3756 VLOG_WARN("bridge %s: %s specified twice as mirror",
3757 br->name, cfg->name);
3761 /* Get rid of deleted mirrors and add new mirrors. */
3762 SHASH_FOR_EACH (node, &old_mirrors) {
3763 if (!shash_find(&new_mirrors, node->name)) {
3764 mirror_destroy(node->data);
3767 SHASH_FOR_EACH (node, &new_mirrors) {
3768 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3770 mirror = mirror_create(br, node->name);
3775 mirror_reconfigure_one(mirror, node->data);
3777 shash_destroy(&old_mirrors);
3778 shash_destroy(&new_mirrors);
3780 /* Update port reserved status. */
3781 for (i = 0; i < br->n_ports; i++) {
3782 br->ports[i]->is_mirror_output_port = false;
3784 for (i = 0; i < MAX_MIRRORS; i++) {
3785 struct mirror *m = br->mirrors[i];
3786 if (m && m->out_port) {
3787 m->out_port->is_mirror_output_port = true;
3791 /* Update flooded vlans (for RSPAN). */
3793 if (br->cfg->n_flood_vlans) {
3794 rspan_vlans = bitmap_allocate(4096);
3796 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3797 int64_t vlan = br->cfg->flood_vlans[i];
3798 if (vlan >= 0 && vlan < 4096) {
3799 bitmap_set1(rspan_vlans, vlan);
3800 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3803 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3808 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3813 static struct mirror *
3814 mirror_create(struct bridge *br, const char *name)
3819 for (i = 0; ; i++) {
3820 if (i >= MAX_MIRRORS) {
3821 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3822 "cannot create %s", br->name, MAX_MIRRORS, name);
3825 if (!br->mirrors[i]) {
3830 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3833 br->mirrors[i] = m = xzalloc(sizeof *m);
3836 m->name = xstrdup(name);
3837 shash_init(&m->src_ports);
3838 shash_init(&m->dst_ports);
3848 mirror_destroy(struct mirror *m)
3851 struct bridge *br = m->bridge;
3854 for (i = 0; i < br->n_ports; i++) {
3855 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3856 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3859 shash_destroy(&m->src_ports);
3860 shash_destroy(&m->dst_ports);
3863 m->bridge->mirrors[m->idx] = NULL;
3871 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3872 struct shash *names)
3876 for (i = 0; i < n_ports; i++) {
3877 const char *name = ports[i]->name;
3878 if (port_lookup(m->bridge, name)) {
3879 shash_add_once(names, name, NULL);
3881 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3882 "port %s", m->bridge->name, m->name, name);
3888 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3894 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3896 for (i = 0; i < cfg->n_select_vlan; i++) {
3897 int64_t vlan = cfg->select_vlan[i];
3898 if (vlan < 0 || vlan > 4095) {
3899 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3900 m->bridge->name, m->name, vlan);
3902 (*vlans)[n_vlans++] = vlan;
3909 vlan_is_mirrored(const struct mirror *m, int vlan)
3913 for (i = 0; i < m->n_vlans; i++) {
3914 if (m->vlans[i] == vlan) {
3922 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3926 for (i = 0; i < m->n_vlans; i++) {
3927 if (port_trunks_vlan(p, m->vlans[i])) {
3935 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3937 struct shash src_ports, dst_ports;
3938 mirror_mask_t mirror_bit;
3939 struct port *out_port;
3945 /* Get output port. */
3946 if (cfg->output_port) {
3947 out_port = port_lookup(m->bridge, cfg->output_port->name);
3949 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3950 m->bridge->name, m->name);
3956 if (cfg->output_vlan) {
3957 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3958 "output vlan; ignoring output vlan",
3959 m->bridge->name, m->name);
3961 } else if (cfg->output_vlan) {
3963 out_vlan = *cfg->output_vlan;
3965 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3966 m->bridge->name, m->name);
3971 shash_init(&src_ports);
3972 shash_init(&dst_ports);
3973 if (cfg->select_all) {
3974 for (i = 0; i < m->bridge->n_ports; i++) {
3975 const char *name = m->bridge->ports[i]->name;
3976 shash_add_once(&src_ports, name, NULL);
3977 shash_add_once(&dst_ports, name, NULL);
3982 /* Get ports, and drop duplicates and ports that don't exist. */
3983 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3985 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3988 /* Get all the vlans, and drop duplicate and invalid vlans. */
3989 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3992 /* Update mirror data. */
3993 if (!shash_equal_keys(&m->src_ports, &src_ports)
3994 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3995 || m->n_vlans != n_vlans
3996 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3997 || m->out_port != out_port
3998 || m->out_vlan != out_vlan) {
3999 bridge_flush(m->bridge);
4001 shash_swap(&m->src_ports, &src_ports);
4002 shash_swap(&m->dst_ports, &dst_ports);
4005 m->n_vlans = n_vlans;
4006 m->out_port = out_port;
4007 m->out_vlan = out_vlan;
4010 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4011 for (i = 0; i < m->bridge->n_ports; i++) {
4012 struct port *port = m->bridge->ports[i];
4014 if (shash_find(&m->src_ports, port->name)
4017 ? port_trunks_any_mirrored_vlan(m, port)
4018 : vlan_is_mirrored(m, port->vlan)))) {
4019 port->src_mirrors |= mirror_bit;
4021 port->src_mirrors &= ~mirror_bit;
4024 if (shash_find(&m->dst_ports, port->name)) {
4025 port->dst_mirrors |= mirror_bit;
4027 port->dst_mirrors &= ~mirror_bit;
4032 shash_destroy(&src_ports);
4033 shash_destroy(&dst_ports);