1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
36 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
46 #include "ofproto/netflow.h"
47 #include "ofproto/ofproto.h"
49 #include "poll-loop.h"
50 #include "port-array.h"
51 #include "proc-net-compat.h"
55 #include "socket-util.h"
56 #include "stream-ssl.h"
62 #include "vswitchd/vswitch-idl.h"
63 #include "xenserver.h"
65 #include "sflow_api.h"
67 #define THIS_MODULE VLM_bridge
76 /* These members are always valid. */
77 struct port *port; /* Containing port. */
78 size_t port_ifidx; /* Index within containing port. */
79 char *name; /* Host network device name. */
80 tag_type tag; /* Tag associated with this interface. */
81 long long delay_expires; /* Time after which 'enabled' may change. */
83 /* These members are valid only after bridge_reconfigure() causes them to
85 int dp_ifidx; /* Index within kernel datapath. */
86 struct netdev *netdev; /* Network device. */
87 bool enabled; /* May be chosen for flows? */
89 /* This member is only valid *during* bridge_reconfigure(). */
90 const struct ovsrec_interface *cfg;
93 #define BOND_MASK 0xff
95 int iface_idx; /* Index of assigned iface, or -1 if none. */
96 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
97 tag_type iface_tag; /* Tag associated with iface_idx. */
100 #define MAX_MIRRORS 32
101 typedef uint32_t mirror_mask_t;
102 #define MIRROR_MASK_C(X) UINT32_C(X)
103 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
105 struct bridge *bridge;
109 /* Selection criteria. */
110 struct shash src_ports; /* Name is port name; data is always NULL. */
111 struct shash dst_ports; /* Name is port name; data is always NULL. */
116 struct port *out_port;
120 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
122 struct bridge *bridge;
124 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
125 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
126 * NULL if all VLANs are trunked. */
129 /* An ordinary bridge port has 1 interface.
130 * A bridge port for bonding has at least 2 interfaces. */
131 struct iface **ifaces;
132 size_t n_ifaces, allocated_ifaces;
135 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
136 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
137 tag_type active_iface_tag; /* Tag for bcast flows. */
138 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
139 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
140 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
141 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
142 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
143 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
144 long long int bond_next_rebalance; /* Next rebalancing time. */
146 /* Port mirroring info. */
147 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
148 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
149 bool is_mirror_output_port; /* Does port mirroring send frames here? */
151 /* This member is only valid *during* bridge_reconfigure(). */
152 const struct ovsrec_port *cfg;
155 #define DP_MAX_PORTS 255
157 struct list node; /* Node in global list of bridges. */
158 char *name; /* User-specified arbitrary name. */
159 struct mac_learning *ml; /* MAC learning table. */
160 bool sent_config_request; /* Successfully sent config request? */
161 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
163 /* OpenFlow switch processing. */
164 struct ofproto *ofproto; /* OpenFlow switch. */
166 /* Description strings. */
167 char *mfr_desc; /* Manufacturer. */
168 char *hw_desc; /* Hardware. */
169 char *sw_desc; /* Software version. */
170 char *serial_desc; /* Serial number. */
171 char *dp_desc; /* Datapath description. */
173 /* Kernel datapath information. */
174 struct dpif *dpif; /* Datapath. */
175 struct port_array ifaces; /* Indexed by kernel datapath port number. */
179 size_t n_ports, allocated_ports;
180 struct shash iface_by_name; /* "struct iface"s indexed by name. */
181 struct shash port_by_name; /* "struct port"s indexed by name. */
184 bool has_bonded_ports;
189 /* Flow statistics gathering. */
190 time_t next_stats_request;
192 /* Port mirroring. */
193 struct mirror *mirrors[MAX_MIRRORS];
195 /* This member is only valid *during* bridge_reconfigure(). */
196 const struct ovsrec_bridge *cfg;
199 /* List of all bridges. */
200 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
202 /* Maximum number of datapaths. */
203 enum { DP_MAX = 256 };
205 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
206 static void bridge_destroy(struct bridge *);
207 static struct bridge *bridge_lookup(const char *name);
208 static unixctl_cb_func bridge_unixctl_dump_flows;
209 static int bridge_run_one(struct bridge *);
210 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
211 const struct bridge *br,
212 struct ovsrec_controller ***controllersp);
213 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
215 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
217 const struct sockaddr_in *managers,
219 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
220 static void bridge_fetch_dp_ifaces(struct bridge *);
221 static void bridge_flush(struct bridge *);
222 static void bridge_pick_local_hw_addr(struct bridge *,
223 uint8_t ea[ETH_ADDR_LEN],
224 struct iface **hw_addr_iface);
225 static uint64_t bridge_pick_datapath_id(struct bridge *,
226 const uint8_t bridge_ea[ETH_ADDR_LEN],
227 struct iface *hw_addr_iface);
228 static struct iface *bridge_get_local_iface(struct bridge *);
229 static uint64_t dpid_from_hash(const void *, size_t nbytes);
231 static unixctl_cb_func bridge_unixctl_fdb_show;
233 static void bond_init(void);
234 static void bond_run(struct bridge *);
235 static void bond_wait(struct bridge *);
236 static void bond_rebalance_port(struct port *);
237 static void bond_send_learning_packets(struct port *);
238 static void bond_enable_slave(struct iface *iface, bool enable);
240 static struct port *port_create(struct bridge *, const char *name);
241 static void port_reconfigure(struct port *, const struct ovsrec_port *);
242 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
243 static void port_destroy(struct port *);
244 static struct port *port_lookup(const struct bridge *, const char *name);
245 static struct iface *port_lookup_iface(const struct port *, const char *name);
246 static struct port *port_from_dp_ifidx(const struct bridge *,
248 static void port_update_bond_compat(struct port *);
249 static void port_update_vlan_compat(struct port *);
250 static void port_update_bonding(struct port *);
252 static struct mirror *mirror_create(struct bridge *, const char *name);
253 static void mirror_destroy(struct mirror *);
254 static void mirror_reconfigure(struct bridge *);
255 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
256 static bool vlan_is_mirrored(const struct mirror *, int vlan);
258 static struct iface *iface_create(struct port *port,
259 const struct ovsrec_interface *if_cfg);
260 static void iface_destroy(struct iface *);
261 static struct iface *iface_lookup(const struct bridge *, const char *name);
262 static struct iface *iface_from_dp_ifidx(const struct bridge *,
264 static bool iface_is_internal(const struct bridge *, const char *name);
265 static void iface_set_mac(struct iface *);
267 /* Hooks into ofproto processing. */
268 static struct ofhooks bridge_ofhooks;
270 /* Public functions. */
272 /* Adds the name of each interface used by a bridge, including local and
273 * internal ports, to 'svec'. */
275 bridge_get_ifaces(struct svec *svec)
277 struct bridge *br, *next;
280 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
281 for (i = 0; i < br->n_ports; i++) {
282 struct port *port = br->ports[i];
284 for (j = 0; j < port->n_ifaces; j++) {
285 struct iface *iface = port->ifaces[j];
286 if (iface->dp_ifidx < 0) {
287 VLOG_ERR("%s interface not in datapath %s, ignoring",
288 iface->name, dpif_name(br->dpif));
290 if (iface->dp_ifidx != ODPP_LOCAL) {
291 svec_add(svec, iface->name);
300 bridge_init(const struct ovsrec_open_vswitch *cfg)
302 struct svec bridge_names;
303 struct svec dpif_names, dpif_types;
306 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
308 svec_init(&bridge_names);
309 for (i = 0; i < cfg->n_bridges; i++) {
310 svec_add(&bridge_names, cfg->bridges[i]->name);
312 svec_sort(&bridge_names);
314 svec_init(&dpif_names);
315 svec_init(&dpif_types);
316 dp_enumerate_types(&dpif_types);
317 for (i = 0; i < dpif_types.n; i++) {
322 dp_enumerate_names(dpif_types.names[i], &dpif_names);
324 for (j = 0; j < dpif_names.n; j++) {
325 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
327 struct svec all_names;
330 svec_init(&all_names);
331 dpif_get_all_names(dpif, &all_names);
332 for (k = 0; k < all_names.n; k++) {
333 if (svec_contains(&bridge_names, all_names.names[k])) {
339 svec_destroy(&all_names);
344 svec_destroy(&bridge_names);
345 svec_destroy(&dpif_names);
346 svec_destroy(&dpif_types);
348 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
352 bridge_reconfigure(cfg);
357 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
359 /* XXX SSL should be configurable on a per-bridge basis. */
361 stream_ssl_set_private_key_file(ssl->private_key);
362 stream_ssl_set_certificate_file(ssl->certificate);
363 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
368 /* Attempt to create the network device 'iface_name' through the netdev
371 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
374 struct shash_node *node;
375 struct shash options;
379 shash_init(&options);
380 for (i = 0; i < iface_cfg->n_options; i++) {
381 shash_add(&options, iface_cfg->key_options[i],
382 xstrdup(iface_cfg->value_options[i]));
386 struct netdev_options netdev_options;
388 memset(&netdev_options, 0, sizeof netdev_options);
389 netdev_options.name = iface_cfg->name;
390 if (!strcmp(iface_cfg->type, "internal")) {
391 /* An "internal" config type maps to a netdev "system" type. */
392 netdev_options.type = "system";
394 netdev_options.type = iface_cfg->type;
396 netdev_options.args = &options;
397 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
398 netdev_options.may_create = true;
399 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
400 netdev_options.may_open = true;
403 error = netdev_open(&netdev_options, &iface->netdev);
406 netdev_get_carrier(iface->netdev, &iface->enabled);
408 } else if (iface->netdev) {
409 const char *netdev_type = netdev_get_type(iface->netdev);
410 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
411 ? iface_cfg->type : NULL;
413 /* An "internal" config type maps to a netdev "system" type. */
414 if (iface_type && !strcmp(iface_type, "internal")) {
415 iface_type = "system";
418 if (!iface_type || !strcmp(netdev_type, iface_type)) {
419 error = netdev_reconfigure(iface->netdev, &options);
421 VLOG_WARN("%s: attempting change device type from %s to %s",
422 iface_cfg->name, netdev_type, iface_type);
427 SHASH_FOR_EACH (node, &options) {
430 shash_destroy(&options);
436 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
438 return set_up_iface(iface_cfg, iface, false);
442 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
443 void *aux OVS_UNUSED)
445 if (!iface->netdev) {
446 int error = set_up_iface(iface->cfg, iface, true);
448 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
458 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
459 void *aux OVS_UNUSED)
461 if (iface->dp_ifidx >= 0) {
462 VLOG_DBG("%s has interface %s on port %d",
464 iface->name, iface->dp_ifidx);
467 VLOG_ERR("%s interface not in %s, dropping",
468 iface->name, dpif_name(br->dpif));
474 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
475 void *aux OVS_UNUSED)
477 /* Set policing attributes. */
478 netdev_set_policing(iface->netdev,
479 iface->cfg->ingress_policing_rate,
480 iface->cfg->ingress_policing_burst);
482 /* Set MAC address of internal interfaces other than the local
484 if (iface->dp_ifidx != ODPP_LOCAL
485 && iface_is_internal(br, iface->name)) {
486 iface_set_mac(iface);
492 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
493 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
494 * deletes from 'br' any ports that no longer have any interfaces. */
496 iterate_and_prune_ifaces(struct bridge *br,
497 bool (*cb)(struct bridge *, struct iface *,
503 for (i = 0; i < br->n_ports; ) {
504 struct port *port = br->ports[i];
505 for (j = 0; j < port->n_ifaces; ) {
506 struct iface *iface = port->ifaces[j];
507 if (cb(br, iface, aux)) {
510 iface_destroy(iface);
514 if (port->n_ifaces) {
517 VLOG_ERR("%s port has no interfaces, dropping", port->name);
523 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
524 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
525 * responsible for freeing '*managersp' (with free()).
527 * You may be asking yourself "why does ovs-vswitchd care?", because
528 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
529 * should not be and in fact is not directly involved in that. But
530 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
531 * it has to tell in-band control where the managers are to enable that.
534 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
535 struct sockaddr_in **managersp, size_t *n_managersp)
537 struct sockaddr_in *managers = NULL;
538 size_t n_managers = 0;
540 if (ovs_cfg->n_managers > 0) {
543 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
544 for (i = 0; i < ovs_cfg->n_managers; i++) {
545 const char *name = ovs_cfg->managers[i];
546 struct sockaddr_in *sin = &managers[i];
548 if ((!strncmp(name, "tcp:", 4)
549 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
550 (!strncmp(name, "ssl:", 4)
551 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
557 *managersp = managers;
558 *n_managersp = n_managers;
562 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
564 struct ovsdb_idl_txn *txn;
565 struct shash old_br, new_br;
566 struct shash_node *node;
567 struct bridge *br, *next;
568 struct sockaddr_in *managers;
571 int sflow_bridge_number;
573 COVERAGE_INC(bridge_reconfigure);
575 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
577 collect_managers(ovs_cfg, &managers, &n_managers);
579 /* Collect old and new bridges. */
582 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
583 shash_add(&old_br, br->name, br);
585 for (i = 0; i < ovs_cfg->n_bridges; i++) {
586 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
587 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
588 VLOG_WARN("more than one bridge named %s", br_cfg->name);
592 /* Get rid of deleted bridges and add new bridges. */
593 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
594 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
601 SHASH_FOR_EACH (node, &new_br) {
602 const char *br_name = node->name;
603 const struct ovsrec_bridge *br_cfg = node->data;
604 br = shash_find_data(&old_br, br_name);
606 /* If the bridge datapath type has changed, we need to tear it
607 * down and recreate. */
608 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
610 bridge_create(br_cfg);
613 bridge_create(br_cfg);
616 shash_destroy(&old_br);
617 shash_destroy(&new_br);
621 bridge_configure_ssl(ovs_cfg->ssl);
624 /* Reconfigure all bridges. */
625 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
626 bridge_reconfigure_one(ovs_cfg, br);
629 /* Add and delete ports on all datapaths.
631 * The kernel will reject any attempt to add a given port to a datapath if
632 * that port already belongs to a different datapath, so we must do all
633 * port deletions before any port additions. */
634 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
635 struct odp_port *dpif_ports;
637 struct shash want_ifaces;
639 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
640 bridge_get_all_ifaces(br, &want_ifaces);
641 for (i = 0; i < n_dpif_ports; i++) {
642 const struct odp_port *p = &dpif_ports[i];
643 if (!shash_find(&want_ifaces, p->devname)
644 && strcmp(p->devname, br->name)) {
645 int retval = dpif_port_del(br->dpif, p->port);
647 VLOG_ERR("failed to remove %s interface from %s: %s",
648 p->devname, dpif_name(br->dpif),
653 shash_destroy(&want_ifaces);
656 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
657 struct odp_port *dpif_ports;
659 struct shash cur_ifaces, want_ifaces;
660 struct shash_node *node;
662 /* Get the set of interfaces currently in this datapath. */
663 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
664 shash_init(&cur_ifaces);
665 for (i = 0; i < n_dpif_ports; i++) {
666 const char *name = dpif_ports[i].devname;
667 if (!shash_find(&cur_ifaces, name)) {
668 shash_add(&cur_ifaces, name, NULL);
673 /* Get the set of interfaces we want on this datapath. */
674 bridge_get_all_ifaces(br, &want_ifaces);
676 SHASH_FOR_EACH (node, &want_ifaces) {
677 const char *if_name = node->name;
678 struct iface *iface = node->data;
680 if (shash_find(&cur_ifaces, if_name)) {
681 /* Already exists, just reconfigure it. */
683 reconfigure_iface(iface->cfg, iface);
686 /* Need to add to datapath. */
690 /* Add to datapath. */
691 internal = iface_is_internal(br, if_name);
692 error = dpif_port_add(br->dpif, if_name,
693 internal ? ODP_PORT_INTERNAL : 0, NULL);
694 if (error == EFBIG) {
695 VLOG_ERR("ran out of valid port numbers on %s",
696 dpif_name(br->dpif));
699 VLOG_ERR("failed to add %s interface to %s: %s",
700 if_name, dpif_name(br->dpif), strerror(error));
704 shash_destroy(&cur_ifaces);
705 shash_destroy(&want_ifaces);
707 sflow_bridge_number = 0;
708 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
711 struct iface *local_iface;
712 struct iface *hw_addr_iface;
715 bridge_fetch_dp_ifaces(br);
717 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
718 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
720 /* Pick local port hardware address, datapath ID. */
721 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
722 local_iface = bridge_get_local_iface(br);
724 int error = netdev_set_etheraddr(local_iface->netdev, ea);
726 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
727 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
728 "Ethernet address: %s",
729 br->name, strerror(error));
733 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
734 ofproto_set_datapath_id(br->ofproto, dpid);
736 dpid_string = xasprintf("%012"PRIx64, dpid);
737 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
740 /* Set NetFlow configuration on this bridge. */
741 if (br->cfg->netflow) {
742 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
743 struct netflow_options opts;
745 memset(&opts, 0, sizeof opts);
747 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
748 if (nf_cfg->engine_type) {
749 opts.engine_type = *nf_cfg->engine_type;
751 if (nf_cfg->engine_id) {
752 opts.engine_id = *nf_cfg->engine_id;
755 opts.active_timeout = nf_cfg->active_timeout;
756 if (!opts.active_timeout) {
757 opts.active_timeout = -1;
758 } else if (opts.active_timeout < 0) {
759 VLOG_WARN("bridge %s: active timeout interval set to negative "
760 "value, using default instead (%d seconds)", br->name,
761 NF_ACTIVE_TIMEOUT_DEFAULT);
762 opts.active_timeout = -1;
765 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
766 if (opts.add_id_to_iface) {
767 if (opts.engine_id > 0x7f) {
768 VLOG_WARN("bridge %s: netflow port mangling may conflict "
769 "with another vswitch, choose an engine id less "
770 "than 128", br->name);
772 if (br->n_ports > 508) {
773 VLOG_WARN("bridge %s: netflow port mangling will conflict "
774 "with another port when more than 508 ports are "
779 opts.collectors.n = nf_cfg->n_targets;
780 opts.collectors.names = nf_cfg->targets;
781 if (ofproto_set_netflow(br->ofproto, &opts)) {
782 VLOG_ERR("bridge %s: problem setting netflow collectors",
786 ofproto_set_netflow(br->ofproto, NULL);
789 /* Set sFlow configuration on this bridge. */
790 if (br->cfg->sflow) {
791 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
792 struct ovsrec_controller **controllers;
793 struct ofproto_sflow_options oso;
794 size_t n_controllers;
797 memset(&oso, 0, sizeof oso);
799 oso.targets.n = sflow_cfg->n_targets;
800 oso.targets.names = sflow_cfg->targets;
802 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
803 if (sflow_cfg->sampling) {
804 oso.sampling_rate = *sflow_cfg->sampling;
807 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
808 if (sflow_cfg->polling) {
809 oso.polling_interval = *sflow_cfg->polling;
812 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
813 if (sflow_cfg->header) {
814 oso.header_len = *sflow_cfg->header;
817 oso.sub_id = sflow_bridge_number++;
818 oso.agent_device = sflow_cfg->agent;
820 oso.control_ip = NULL;
821 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
822 for (i = 0; i < n_controllers; i++) {
823 if (controllers[i]->local_ip) {
824 oso.control_ip = controllers[i]->local_ip;
828 ofproto_set_sflow(br->ofproto, &oso);
830 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
832 ofproto_set_sflow(br->ofproto, NULL);
835 /* Update the controller and related settings. It would be more
836 * straightforward to call this from bridge_reconfigure_one(), but we
837 * can't do it there for two reasons. First, and most importantly, at
838 * that point we don't know the dp_ifidx of any interfaces that have
839 * been added to the bridge (because we haven't actually added them to
840 * the datapath). Second, at that point we haven't set the datapath ID
841 * yet; when a controller is configured, resetting the datapath ID will
842 * immediately disconnect from the controller, so it's better to set
843 * the datapath ID before the controller. */
844 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
846 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
847 for (i = 0; i < br->n_ports; i++) {
848 struct port *port = br->ports[i];
850 port_update_vlan_compat(port);
851 port_update_bonding(port);
854 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
855 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
858 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
860 ovsdb_idl_txn_commit(txn);
861 ovsdb_idl_txn_destroy(txn); /* XXX */
867 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
871 for (i = 0; i < n; i++) {
872 if (!strcmp(keys[i], key)) {
880 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
882 return get_ovsrec_key_value(key,
883 br_cfg->key_other_config,
884 br_cfg->value_other_config,
885 br_cfg->n_other_config);
889 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
890 struct iface **hw_addr_iface)
896 *hw_addr_iface = NULL;
898 /* Did the user request a particular MAC? */
899 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
900 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
901 if (eth_addr_is_multicast(ea)) {
902 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
903 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
904 } else if (eth_addr_is_zero(ea)) {
905 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
911 /* Otherwise choose the minimum non-local MAC address among all of the
913 memset(ea, 0xff, sizeof ea);
914 for (i = 0; i < br->n_ports; i++) {
915 struct port *port = br->ports[i];
916 uint8_t iface_ea[ETH_ADDR_LEN];
919 /* Mirror output ports don't participate. */
920 if (port->is_mirror_output_port) {
924 /* Choose the MAC address to represent the port. */
925 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
926 /* Find the interface with this Ethernet address (if any) so that
927 * we can provide the correct devname to the caller. */
929 for (j = 0; j < port->n_ifaces; j++) {
930 struct iface *candidate = port->ifaces[j];
931 uint8_t candidate_ea[ETH_ADDR_LEN];
932 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
933 && eth_addr_equals(iface_ea, candidate_ea)) {
938 /* Choose the interface whose MAC address will represent the port.
939 * The Linux kernel bonding code always chooses the MAC address of
940 * the first slave added to a bond, and the Fedora networking
941 * scripts always add slaves to a bond in alphabetical order, so
942 * for compatibility we choose the interface with the name that is
943 * first in alphabetical order. */
944 iface = port->ifaces[0];
945 for (j = 1; j < port->n_ifaces; j++) {
946 struct iface *candidate = port->ifaces[j];
947 if (strcmp(candidate->name, iface->name) < 0) {
952 /* The local port doesn't count (since we're trying to choose its
953 * MAC address anyway). */
954 if (iface->dp_ifidx == ODPP_LOCAL) {
959 error = netdev_get_etheraddr(iface->netdev, iface_ea);
961 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
962 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
963 iface->name, strerror(error));
968 /* Compare against our current choice. */
969 if (!eth_addr_is_multicast(iface_ea) &&
970 !eth_addr_is_local(iface_ea) &&
971 !eth_addr_is_reserved(iface_ea) &&
972 !eth_addr_is_zero(iface_ea) &&
973 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
975 memcpy(ea, iface_ea, ETH_ADDR_LEN);
976 *hw_addr_iface = iface;
979 if (eth_addr_is_multicast(ea)) {
980 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
981 *hw_addr_iface = NULL;
982 VLOG_WARN("bridge %s: using default bridge Ethernet "
983 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
985 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
986 br->name, ETH_ADDR_ARGS(ea));
990 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
991 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
992 * an interface on 'br', then that interface must be passed in as
993 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
994 * 'hw_addr_iface' must be passed in as a null pointer. */
996 bridge_pick_datapath_id(struct bridge *br,
997 const uint8_t bridge_ea[ETH_ADDR_LEN],
998 struct iface *hw_addr_iface)
1001 * The procedure for choosing a bridge MAC address will, in the most
1002 * ordinary case, also choose a unique MAC that we can use as a datapath
1003 * ID. In some special cases, though, multiple bridges will end up with
1004 * the same MAC address. This is OK for the bridges, but it will confuse
1005 * the OpenFlow controller, because each datapath needs a unique datapath
1008 * Datapath IDs must be unique. It is also very desirable that they be
1009 * stable from one run to the next, so that policy set on a datapath
1012 const char *datapath_id;
1015 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1016 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1020 if (hw_addr_iface) {
1022 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1024 * A bridge whose MAC address is taken from a VLAN network device
1025 * (that is, a network device created with vconfig(8) or similar
1026 * tool) will have the same MAC address as a bridge on the VLAN
1027 * device's physical network device.
1029 * Handle this case by hashing the physical network device MAC
1030 * along with the VLAN identifier.
1032 uint8_t buf[ETH_ADDR_LEN + 2];
1033 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1034 buf[ETH_ADDR_LEN] = vlan >> 8;
1035 buf[ETH_ADDR_LEN + 1] = vlan;
1036 return dpid_from_hash(buf, sizeof buf);
1039 * Assume that this bridge's MAC address is unique, since it
1040 * doesn't fit any of the cases we handle specially.
1045 * A purely internal bridge, that is, one that has no non-virtual
1046 * network devices on it at all, is more difficult because it has no
1047 * natural unique identifier at all.
1049 * When the host is a XenServer, we handle this case by hashing the
1050 * host's UUID with the name of the bridge. Names of bridges are
1051 * persistent across XenServer reboots, although they can be reused if
1052 * an internal network is destroyed and then a new one is later
1053 * created, so this is fairly effective.
1055 * When the host is not a XenServer, we punt by using a random MAC
1056 * address on each run.
1058 const char *host_uuid = xenserver_get_host_uuid();
1060 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1061 dpid = dpid_from_hash(combined, strlen(combined));
1067 return eth_addr_to_uint64(bridge_ea);
1071 dpid_from_hash(const void *data, size_t n)
1073 uint8_t hash[SHA1_DIGEST_SIZE];
1075 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1076 sha1_bytes(data, n, hash);
1077 eth_addr_mark_random(hash);
1078 return eth_addr_to_uint64(hash);
1084 struct bridge *br, *next;
1088 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1089 int error = bridge_run_one(br);
1091 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1092 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1093 "forcing reconfiguration", br->name);
1107 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1108 ofproto_wait(br->ofproto);
1109 if (ofproto_has_controller(br->ofproto)) {
1113 mac_learning_wait(br->ml);
1118 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1119 * configuration changes. */
1121 bridge_flush(struct bridge *br)
1123 COVERAGE_INC(bridge_flush);
1125 mac_learning_flush(br->ml);
1128 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1129 * such interface. */
1130 static struct iface *
1131 bridge_get_local_iface(struct bridge *br)
1135 for (i = 0; i < br->n_ports; i++) {
1136 struct port *port = br->ports[i];
1137 for (j = 0; j < port->n_ifaces; j++) {
1138 struct iface *iface = port->ifaces[j];
1139 if (iface->dp_ifidx == ODPP_LOCAL) {
1148 /* Bridge unixctl user interface functions. */
1150 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1151 const char *args, void *aux OVS_UNUSED)
1153 struct ds ds = DS_EMPTY_INITIALIZER;
1154 const struct bridge *br;
1155 const struct mac_entry *e;
1157 br = bridge_lookup(args);
1159 unixctl_command_reply(conn, 501, "no such bridge");
1163 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1164 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1165 if (e->port < 0 || e->port >= br->n_ports) {
1168 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1169 br->ports[e->port]->ifaces[0]->dp_ifidx,
1170 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1172 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1176 /* Bridge reconfiguration functions. */
1177 static struct bridge *
1178 bridge_create(const struct ovsrec_bridge *br_cfg)
1183 assert(!bridge_lookup(br_cfg->name));
1184 br = xzalloc(sizeof *br);
1186 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1192 dpif_flow_flush(br->dpif);
1194 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1197 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1199 dpif_delete(br->dpif);
1200 dpif_close(br->dpif);
1205 br->name = xstrdup(br_cfg->name);
1207 br->ml = mac_learning_create();
1208 br->sent_config_request = false;
1209 eth_addr_nicira_random(br->default_ea);
1211 port_array_init(&br->ifaces);
1213 shash_init(&br->port_by_name);
1214 shash_init(&br->iface_by_name);
1218 list_push_back(&all_bridges, &br->node);
1220 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1226 bridge_destroy(struct bridge *br)
1231 while (br->n_ports > 0) {
1232 port_destroy(br->ports[br->n_ports - 1]);
1234 list_remove(&br->node);
1235 error = dpif_delete(br->dpif);
1236 if (error && error != ENOENT) {
1237 VLOG_ERR("failed to delete %s: %s",
1238 dpif_name(br->dpif), strerror(error));
1240 dpif_close(br->dpif);
1241 ofproto_destroy(br->ofproto);
1242 mac_learning_destroy(br->ml);
1243 port_array_destroy(&br->ifaces);
1244 shash_destroy(&br->port_by_name);
1245 shash_destroy(&br->iface_by_name);
1252 static struct bridge *
1253 bridge_lookup(const char *name)
1257 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1258 if (!strcmp(br->name, name)) {
1266 bridge_exists(const char *name)
1268 return bridge_lookup(name) ? true : false;
1272 bridge_get_datapathid(const char *name)
1274 struct bridge *br = bridge_lookup(name);
1275 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1278 /* Handle requests for a listing of all flows known by the OpenFlow
1279 * stack, including those normally hidden. */
1281 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1282 const char *args, void *aux OVS_UNUSED)
1287 br = bridge_lookup(args);
1289 unixctl_command_reply(conn, 501, "Unknown bridge");
1294 ofproto_get_all_flows(br->ofproto, &results);
1296 unixctl_command_reply(conn, 200, ds_cstr(&results));
1297 ds_destroy(&results);
1301 bridge_run_one(struct bridge *br)
1305 error = ofproto_run1(br->ofproto);
1310 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1313 error = ofproto_run2(br->ofproto, br->flush);
1320 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1321 const struct bridge *br,
1322 struct ovsrec_controller ***controllersp)
1324 struct ovsrec_controller **controllers;
1325 size_t n_controllers;
1327 if (br->cfg->n_controller) {
1328 controllers = br->cfg->controller;
1329 n_controllers = br->cfg->n_controller;
1331 controllers = ovs_cfg->controller;
1332 n_controllers = ovs_cfg->n_controller;
1335 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1341 *controllersp = controllers;
1343 return n_controllers;
1347 bridge_update_desc(struct bridge *br OVS_UNUSED)
1350 bool changed = false;
1353 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1354 if (desc != br->mfr_desc) {
1357 br->mfr_desc = xstrdup(desc);
1359 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1364 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1365 if (desc != br->hw_desc) {
1368 br->hw_desc = xstrdup(desc);
1370 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1375 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1376 if (desc != br->sw_desc) {
1379 br->sw_desc = xstrdup(desc);
1381 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1386 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1387 if (desc != br->serial_desc) {
1388 free(br->serial_desc);
1390 br->serial_desc = xstrdup(desc);
1392 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1397 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1398 if (desc != br->dp_desc) {
1401 br->dp_desc = xstrdup(desc);
1403 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1409 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1410 br->sw_desc, br->serial_desc, br->dp_desc);
1416 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1419 struct shash old_ports, new_ports;
1420 struct svec listeners, old_listeners;
1421 struct svec snoops, old_snoops;
1422 struct shash_node *node;
1425 /* Collect old ports. */
1426 shash_init(&old_ports);
1427 for (i = 0; i < br->n_ports; i++) {
1428 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1431 /* Collect new ports. */
1432 shash_init(&new_ports);
1433 for (i = 0; i < br->cfg->n_ports; i++) {
1434 const char *name = br->cfg->ports[i]->name;
1435 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1436 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1441 /* If we have a controller, then we need a local port. Complain if the
1442 * user didn't specify one.
1444 * XXX perhaps we should synthesize a port ourselves in this case. */
1445 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1446 char local_name[IF_NAMESIZE];
1449 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1450 local_name, sizeof local_name);
1451 if (!error && !shash_find(&new_ports, local_name)) {
1452 VLOG_WARN("bridge %s: controller specified but no local port "
1453 "(port named %s) defined",
1454 br->name, local_name);
1458 /* Get rid of deleted ports.
1459 * Get rid of deleted interfaces on ports that still exist. */
1460 SHASH_FOR_EACH (node, &old_ports) {
1461 struct port *port = node->data;
1462 const struct ovsrec_port *port_cfg;
1464 port_cfg = shash_find_data(&new_ports, node->name);
1468 port_del_ifaces(port, port_cfg);
1472 /* Create new ports.
1473 * Add new interfaces to existing ports.
1474 * Reconfigure existing ports. */
1475 SHASH_FOR_EACH (node, &new_ports) {
1476 struct port *port = shash_find_data(&old_ports, node->name);
1478 port = port_create(br, node->name);
1481 port_reconfigure(port, node->data);
1482 if (!port->n_ifaces) {
1483 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1484 br->name, port->name);
1488 shash_destroy(&old_ports);
1489 shash_destroy(&new_ports);
1491 /* Delete all flows if we're switching from connected to standalone or vice
1492 * versa. (XXX Should we delete all flows if we are switching from one
1493 * controller to another?) */
1496 /* Configure OpenFlow management listeners. */
1497 svec_init(&listeners);
1498 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1500 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1501 ovs_rundir, br->name));
1502 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1503 svec_clear(&listeners);
1505 svec_sort_unique(&listeners);
1507 svec_init(&old_listeners);
1508 ofproto_get_listeners(br->ofproto, &old_listeners);
1509 svec_sort_unique(&old_listeners);
1511 if (!svec_equal(&listeners, &old_listeners)) {
1512 ofproto_set_listeners(br->ofproto, &listeners);
1514 svec_destroy(&listeners);
1515 svec_destroy(&old_listeners);
1517 /* Configure OpenFlow controller connection snooping. */
1519 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1521 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1522 ovs_rundir, br->name));
1523 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1524 svec_clear(&snoops);
1526 svec_sort_unique(&snoops);
1528 svec_init(&old_snoops);
1529 ofproto_get_snoops(br->ofproto, &old_snoops);
1530 svec_sort_unique(&old_snoops);
1532 if (!svec_equal(&snoops, &old_snoops)) {
1533 ofproto_set_snoops(br->ofproto, &snoops);
1535 svec_destroy(&snoops);
1536 svec_destroy(&old_snoops);
1538 /* Default listener. */
1539 svec_init(&listeners);
1540 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1541 ovs_rundir, br->name));
1542 svec_init(&old_listeners);
1543 ofproto_get_listeners(br->ofproto, &old_listeners);
1544 if (!svec_equal(&listeners, &old_listeners)) {
1545 ofproto_set_listeners(br->ofproto, &listeners);
1547 svec_destroy(&listeners);
1548 svec_destroy(&old_listeners);
1550 /* Default snoop. */
1552 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1553 ovs_rundir, br->name));
1554 svec_init(&old_snoops);
1555 ofproto_get_snoops(br->ofproto, &old_snoops);
1556 if (!svec_equal(&snoops, &old_snoops)) {
1557 ofproto_set_snoops(br->ofproto, &snoops);
1559 svec_destroy(&snoops);
1560 svec_destroy(&old_snoops);
1563 mirror_reconfigure(br);
1565 bridge_update_desc(br);
1569 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1571 const struct sockaddr_in *managers,
1574 struct ovsrec_controller **controllers;
1575 size_t n_controllers;
1577 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1579 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1580 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1581 ofproto_flush_flows(br->ofproto);
1584 if (!n_controllers) {
1585 union ofp_action action;
1588 /* Clear out controllers. */
1589 ofproto_set_controllers(br->ofproto, NULL, 0);
1591 /* Set up a flow that matches every packet and directs them to
1592 * OFPP_NORMAL (which goes to us). */
1593 memset(&action, 0, sizeof action);
1594 action.type = htons(OFPAT_OUTPUT);
1595 action.output.len = htons(sizeof action);
1596 action.output.port = htons(OFPP_NORMAL);
1597 memset(&flow, 0, sizeof flow);
1598 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1600 struct ofproto_controller *ocs;
1603 ocs = xmalloc(n_controllers * sizeof *ocs);
1604 for (i = 0; i < n_controllers; i++) {
1605 struct ovsrec_controller *c = controllers[i];
1606 struct ofproto_controller *oc = &ocs[i];
1608 if (strcmp(c->target, "discover")) {
1609 struct iface *local_iface;
1612 local_iface = bridge_get_local_iface(br);
1613 if (local_iface && c->local_ip
1614 && inet_aton(c->local_ip, &ip)) {
1615 struct netdev *netdev = local_iface->netdev;
1616 struct in_addr mask, gateway;
1618 if (!c->local_netmask
1619 || !inet_aton(c->local_netmask, &mask)) {
1622 if (!c->local_gateway
1623 || !inet_aton(c->local_gateway, &gateway)) {
1627 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1629 mask.s_addr = guess_netmask(ip.s_addr);
1631 if (!netdev_set_in4(netdev, ip, mask)) {
1632 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1634 br->name, IP_ARGS(&ip.s_addr),
1635 IP_ARGS(&mask.s_addr));
1638 if (gateway.s_addr) {
1639 if (!netdev_add_router(netdev, gateway)) {
1640 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1641 br->name, IP_ARGS(&gateway.s_addr));
1647 oc->target = c->target;
1648 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1649 oc->probe_interval = (c->inactivity_probe
1650 ? *c->inactivity_probe / 1000 : 5);
1651 oc->fail = (!c->fail_mode
1652 || !strcmp(c->fail_mode, "standalone")
1653 || !strcmp(c->fail_mode, "open")
1654 ? OFPROTO_FAIL_STANDALONE
1655 : OFPROTO_FAIL_SECURE);
1656 oc->band = (!c->connection_mode
1657 || !strcmp(c->connection_mode, "in-band")
1659 : OFPROTO_OUT_OF_BAND);
1660 oc->accept_re = c->discover_accept_regex;
1661 oc->update_resolv_conf = c->discover_update_resolv_conf;
1662 oc->rate_limit = (c->controller_rate_limit
1663 ? *c->controller_rate_limit : 0);
1664 oc->burst_limit = (c->controller_burst_limit
1665 ? *c->controller_burst_limit : 0);
1667 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1673 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1678 for (i = 0; i < br->n_ports; i++) {
1679 struct port *port = br->ports[i];
1680 for (j = 0; j < port->n_ifaces; j++) {
1681 struct iface *iface = port->ifaces[j];
1682 shash_add_once(ifaces, iface->name, iface);
1684 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1685 shash_add_once(ifaces, port->name, NULL);
1690 /* For robustness, in case the administrator moves around datapath ports behind
1691 * our back, we re-check all the datapath port numbers here.
1693 * This function will set the 'dp_ifidx' members of interfaces that have
1694 * disappeared to -1, so only call this function from a context where those
1695 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1696 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1697 * datapath, which doesn't support UINT16_MAX+1 ports. */
1699 bridge_fetch_dp_ifaces(struct bridge *br)
1701 struct odp_port *dpif_ports;
1702 size_t n_dpif_ports;
1705 /* Reset all interface numbers. */
1706 for (i = 0; i < br->n_ports; i++) {
1707 struct port *port = br->ports[i];
1708 for (j = 0; j < port->n_ifaces; j++) {
1709 struct iface *iface = port->ifaces[j];
1710 iface->dp_ifidx = -1;
1713 port_array_clear(&br->ifaces);
1715 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1716 for (i = 0; i < n_dpif_ports; i++) {
1717 struct odp_port *p = &dpif_ports[i];
1718 struct iface *iface = iface_lookup(br, p->devname);
1720 if (iface->dp_ifidx >= 0) {
1721 VLOG_WARN("%s reported interface %s twice",
1722 dpif_name(br->dpif), p->devname);
1723 } else if (iface_from_dp_ifidx(br, p->port)) {
1724 VLOG_WARN("%s reported interface %"PRIu16" twice",
1725 dpif_name(br->dpif), p->port);
1727 port_array_set(&br->ifaces, p->port, iface);
1728 iface->dp_ifidx = p->port;
1732 int64_t ofport = (iface->dp_ifidx >= 0
1733 ? odp_port_to_ofp_port(iface->dp_ifidx)
1735 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1742 /* Bridge packet processing functions. */
1745 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1747 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1750 static struct bond_entry *
1751 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1753 return &port->bond_hash[bond_hash(mac)];
1757 bond_choose_iface(const struct port *port)
1759 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1760 size_t i, best_down_slave = -1;
1761 long long next_delay_expiration = LLONG_MAX;
1763 for (i = 0; i < port->n_ifaces; i++) {
1764 struct iface *iface = port->ifaces[i];
1766 if (iface->enabled) {
1768 } else if (iface->delay_expires < next_delay_expiration) {
1769 best_down_slave = i;
1770 next_delay_expiration = iface->delay_expires;
1774 if (best_down_slave != -1) {
1775 struct iface *iface = port->ifaces[best_down_slave];
1777 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1778 "since no other interface is up", iface->name,
1779 iface->delay_expires - time_msec());
1780 bond_enable_slave(iface, true);
1783 return best_down_slave;
1787 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1788 uint16_t *dp_ifidx, tag_type *tags)
1790 struct iface *iface;
1792 assert(port->n_ifaces);
1793 if (port->n_ifaces == 1) {
1794 iface = port->ifaces[0];
1796 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1797 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1798 || !port->ifaces[e->iface_idx]->enabled) {
1799 /* XXX select interface properly. The current interface selection
1800 * is only good for testing the rebalancing code. */
1801 e->iface_idx = bond_choose_iface(port);
1802 if (e->iface_idx < 0) {
1803 *tags |= port->no_ifaces_tag;
1806 e->iface_tag = tag_create_random();
1807 ((struct port *) port)->bond_compat_is_stale = true;
1809 *tags |= e->iface_tag;
1810 iface = port->ifaces[e->iface_idx];
1812 *dp_ifidx = iface->dp_ifidx;
1813 *tags |= iface->tag; /* Currently only used for bonding. */
1818 bond_link_status_update(struct iface *iface, bool carrier)
1820 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1821 struct port *port = iface->port;
1823 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1824 /* Nothing to do. */
1827 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1828 iface->name, carrier ? "detected" : "dropped");
1829 if (carrier == iface->enabled) {
1830 iface->delay_expires = LLONG_MAX;
1831 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1832 iface->name, carrier ? "disabled" : "enabled");
1833 } else if (carrier && port->active_iface < 0) {
1834 bond_enable_slave(iface, true);
1835 if (port->updelay) {
1836 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1837 "other interface is up", iface->name, port->updelay);
1840 int delay = carrier ? port->updelay : port->downdelay;
1841 iface->delay_expires = time_msec() + delay;
1844 "interface %s: will be %s if it stays %s for %d ms",
1846 carrier ? "enabled" : "disabled",
1847 carrier ? "up" : "down",
1854 bond_choose_active_iface(struct port *port)
1856 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1858 port->active_iface = bond_choose_iface(port);
1859 port->active_iface_tag = tag_create_random();
1860 if (port->active_iface >= 0) {
1861 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1862 port->name, port->ifaces[port->active_iface]->name);
1864 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1870 bond_enable_slave(struct iface *iface, bool enable)
1872 struct port *port = iface->port;
1873 struct bridge *br = port->bridge;
1875 /* This acts as a recursion check. If the act of disabling a slave
1876 * causes a different slave to be enabled, the flag will allow us to
1877 * skip redundant work when we reenter this function. It must be
1878 * cleared on exit to keep things safe with multiple bonds. */
1879 static bool moving_active_iface = false;
1881 iface->delay_expires = LLONG_MAX;
1882 if (enable == iface->enabled) {
1886 iface->enabled = enable;
1887 if (!iface->enabled) {
1888 VLOG_WARN("interface %s: disabled", iface->name);
1889 ofproto_revalidate(br->ofproto, iface->tag);
1890 if (iface->port_ifidx == port->active_iface) {
1891 ofproto_revalidate(br->ofproto,
1892 port->active_iface_tag);
1894 /* Disabling a slave can lead to another slave being immediately
1895 * enabled if there will be no active slaves but one is waiting
1896 * on an updelay. In this case we do not need to run most of the
1897 * code for the newly enabled slave since there was no period
1898 * without an active slave and it is redundant with the disabling
1900 moving_active_iface = true;
1901 bond_choose_active_iface(port);
1903 bond_send_learning_packets(port);
1905 VLOG_WARN("interface %s: enabled", iface->name);
1906 if (port->active_iface < 0 && !moving_active_iface) {
1907 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1908 bond_choose_active_iface(port);
1909 bond_send_learning_packets(port);
1911 iface->tag = tag_create_random();
1914 moving_active_iface = false;
1915 port->bond_compat_is_stale = true;
1918 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1919 * bond interface. */
1921 bond_update_fake_iface_stats(struct port *port)
1923 struct netdev_stats bond_stats;
1924 struct netdev *bond_dev;
1927 memset(&bond_stats, 0, sizeof bond_stats);
1929 for (i = 0; i < port->n_ifaces; i++) {
1930 struct netdev_stats slave_stats;
1932 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1933 bond_stats.rx_packets += slave_stats.rx_packets;
1934 bond_stats.rx_bytes += slave_stats.rx_bytes;
1935 bond_stats.tx_packets += slave_stats.tx_packets;
1936 bond_stats.tx_bytes += slave_stats.tx_bytes;
1940 if (!netdev_open_default(port->name, &bond_dev)) {
1941 netdev_set_stats(bond_dev, &bond_stats);
1942 netdev_close(bond_dev);
1947 bond_run(struct bridge *br)
1951 for (i = 0; i < br->n_ports; i++) {
1952 struct port *port = br->ports[i];
1954 if (port->n_ifaces >= 2) {
1955 for (j = 0; j < port->n_ifaces; j++) {
1956 struct iface *iface = port->ifaces[j];
1957 if (time_msec() >= iface->delay_expires) {
1958 bond_enable_slave(iface, !iface->enabled);
1962 if (port->bond_fake_iface
1963 && time_msec() >= port->bond_next_fake_iface_update) {
1964 bond_update_fake_iface_stats(port);
1965 port->bond_next_fake_iface_update = time_msec() + 1000;
1969 if (port->bond_compat_is_stale) {
1970 port->bond_compat_is_stale = false;
1971 port_update_bond_compat(port);
1977 bond_wait(struct bridge *br)
1981 for (i = 0; i < br->n_ports; i++) {
1982 struct port *port = br->ports[i];
1983 if (port->n_ifaces < 2) {
1986 for (j = 0; j < port->n_ifaces; j++) {
1987 struct iface *iface = port->ifaces[j];
1988 if (iface->delay_expires != LLONG_MAX) {
1989 poll_timer_wait(iface->delay_expires - time_msec());
1992 if (port->bond_fake_iface) {
1993 poll_timer_wait(port->bond_next_fake_iface_update - time_msec());
1999 set_dst(struct dst *p, const flow_t *flow,
2000 const struct port *in_port, const struct port *out_port,
2003 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2004 : in_port->vlan >= 0 ? in_port->vlan
2005 : ntohs(flow->dl_vlan));
2006 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2010 swap_dst(struct dst *p, struct dst *q)
2012 struct dst tmp = *p;
2017 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2018 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2019 * that we push to the datapath. We could in fact fully sort the array by
2020 * vlan, but in most cases there are at most two different vlan tags so that's
2021 * possibly overkill.) */
2023 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2025 struct dst *first = dsts;
2026 struct dst *last = dsts + n_dsts;
2028 while (first != last) {
2030 * - All dsts < first have vlan == 'vlan'.
2031 * - All dsts >= last have vlan != 'vlan'.
2032 * - first < last. */
2033 while (first->vlan == vlan) {
2034 if (++first == last) {
2039 /* Same invariants, plus one additional:
2040 * - first->vlan != vlan.
2042 while (last[-1].vlan != vlan) {
2043 if (--last == first) {
2048 /* Same invariants, plus one additional:
2049 * - last[-1].vlan == vlan.*/
2050 swap_dst(first++, --last);
2055 mirror_mask_ffs(mirror_mask_t mask)
2057 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2062 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2063 const struct dst *test)
2066 for (i = 0; i < n_dsts; i++) {
2067 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2075 port_trunks_vlan(const struct port *port, uint16_t vlan)
2077 return (port->vlan < 0
2078 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2082 port_includes_vlan(const struct port *port, uint16_t vlan)
2084 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2088 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2089 const struct port *in_port, const struct port *out_port,
2090 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2092 mirror_mask_t mirrors = in_port->src_mirrors;
2093 struct dst *dst = dsts;
2096 if (out_port == FLOOD_PORT) {
2097 /* XXX use ODP_FLOOD if no vlans or bonding. */
2098 /* XXX even better, define each VLAN as a datapath port group */
2099 for (i = 0; i < br->n_ports; i++) {
2100 struct port *port = br->ports[i];
2101 if (port != in_port && port_includes_vlan(port, vlan)
2102 && !port->is_mirror_output_port
2103 && set_dst(dst, flow, in_port, port, tags)) {
2104 mirrors |= port->dst_mirrors;
2108 *nf_output_iface = NF_OUT_FLOOD;
2109 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2110 *nf_output_iface = dst->dp_ifidx;
2111 mirrors |= out_port->dst_mirrors;
2116 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2117 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2119 if (set_dst(dst, flow, in_port, m->out_port, tags)
2120 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2124 for (i = 0; i < br->n_ports; i++) {
2125 struct port *port = br->ports[i];
2126 if (port_includes_vlan(port, m->out_vlan)
2127 && set_dst(dst, flow, in_port, port, tags))
2131 if (port->vlan < 0) {
2132 dst->vlan = m->out_vlan;
2134 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2138 /* Use the vlan tag on the original flow instead of
2139 * the one passed in the vlan parameter. This ensures
2140 * that we compare the vlan from before any implicit
2141 * tagging tags place. This is necessary because
2142 * dst->vlan is the final vlan, after removing implicit
2144 flow_vlan = ntohs(flow->dl_vlan);
2145 if (flow_vlan == 0) {
2146 flow_vlan = OFP_VLAN_NONE;
2148 if (port == in_port && dst->vlan == flow_vlan) {
2149 /* Don't send out input port on same VLAN. */
2157 mirrors &= mirrors - 1;
2160 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2164 static void OVS_UNUSED
2165 print_dsts(const struct dst *dsts, size_t n)
2167 for (; n--; dsts++) {
2168 printf(">p%"PRIu16, dsts->dp_ifidx);
2169 if (dsts->vlan != OFP_VLAN_NONE) {
2170 printf("v%"PRIu16, dsts->vlan);
2176 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2177 const struct port *in_port, const struct port *out_port,
2178 tag_type *tags, struct odp_actions *actions,
2179 uint16_t *nf_output_iface)
2181 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2183 const struct dst *p;
2186 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2189 cur_vlan = ntohs(flow->dl_vlan);
2190 for (p = dsts; p < &dsts[n_dsts]; p++) {
2191 union odp_action *a;
2192 if (p->vlan != cur_vlan) {
2193 if (p->vlan == OFP_VLAN_NONE) {
2194 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2196 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2197 a->vlan_vid.vlan_vid = htons(p->vlan);
2201 a = odp_actions_add(actions, ODPAT_OUTPUT);
2202 a->output.port = p->dp_ifidx;
2206 /* Returns the effective vlan of a packet, taking into account both the
2207 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2208 * the packet is untagged and -1 indicates it has an invalid header and
2209 * should be dropped. */
2210 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2211 struct port *in_port, bool have_packet)
2213 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2214 * belongs to VLAN 0, so we should treat both cases identically. (In the
2215 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2216 * presumably to allow a priority to be specified. In the latter case, the
2217 * packet does not have any 802.1Q header.) */
2218 int vlan = ntohs(flow->dl_vlan);
2219 if (vlan == OFP_VLAN_NONE) {
2222 if (in_port->vlan >= 0) {
2224 /* XXX support double tagging? */
2226 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2227 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2228 "packet received on port %s configured with "
2229 "implicit VLAN %"PRIu16,
2230 br->name, ntohs(flow->dl_vlan),
2231 in_port->name, in_port->vlan);
2235 vlan = in_port->vlan;
2237 if (!port_includes_vlan(in_port, vlan)) {
2239 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2240 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2241 "packet received on port %s not configured for "
2243 br->name, vlan, in_port->name, vlan);
2253 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2254 struct port *in_port)
2256 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2257 vlan, in_port->port_idx);
2259 /* The log messages here could actually be useful in debugging,
2260 * so keep the rate limit relatively high. */
2261 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2263 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2264 "on port %s in VLAN %d",
2265 br->name, ETH_ADDR_ARGS(flow->dl_src),
2266 in_port->name, vlan);
2267 ofproto_revalidate(br->ofproto, rev_tag);
2272 is_bcast_arp_reply(const flow_t *flow)
2274 return (flow->dl_type == htons(ETH_TYPE_ARP)
2275 && flow->nw_proto == ARP_OP_REPLY
2276 && eth_addr_is_broadcast(flow->dl_dst));
2279 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2280 * dropped. Returns true if they may be forwarded, false if they should be
2283 * If 'have_packet' is true, it indicates that the caller is processing a
2284 * received packet. If 'have_packet' is false, then the caller is just
2285 * revalidating an existing flow because configuration has changed. Either
2286 * way, 'have_packet' only affects logging (there is no point in logging errors
2287 * during revalidation).
2289 * Sets '*in_portp' to the input port. This will be a null pointer if
2290 * flow->in_port does not designate a known input port (in which case
2291 * is_admissible() returns false).
2293 * When returning true, sets '*vlanp' to the effective VLAN of the input
2294 * packet, as returned by flow_get_vlan().
2296 * May also add tags to '*tags', although the current implementation only does
2297 * so in one special case.
2300 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2301 tag_type *tags, int *vlanp, struct port **in_portp)
2303 struct iface *in_iface;
2304 struct port *in_port;
2307 /* Find the interface and port structure for the received packet. */
2308 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2310 /* No interface? Something fishy... */
2312 /* Odd. A few possible reasons here:
2314 * - We deleted an interface but there are still a few packets
2315 * queued up from it.
2317 * - Someone externally added an interface (e.g. with "ovs-dpctl
2318 * add-if") that we don't know about.
2320 * - Packet arrived on the local port but the local port is not
2321 * one of our bridge ports.
2323 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2325 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2326 "interface %"PRIu16, br->name, flow->in_port);
2332 *in_portp = in_port = in_iface->port;
2333 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2338 /* Drop frames for reserved multicast addresses. */
2339 if (eth_addr_is_reserved(flow->dl_dst)) {
2343 /* Drop frames on ports reserved for mirroring. */
2344 if (in_port->is_mirror_output_port) {
2346 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2347 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2348 "%s, which is reserved exclusively for mirroring",
2349 br->name, in_port->name);
2354 /* Packets received on bonds need special attention to avoid duplicates. */
2355 if (in_port->n_ifaces > 1) {
2358 if (eth_addr_is_multicast(flow->dl_dst)) {
2359 *tags |= in_port->active_iface_tag;
2360 if (in_port->active_iface != in_iface->port_ifidx) {
2361 /* Drop all multicast packets on inactive slaves. */
2366 /* Drop all packets for which we have learned a different input
2367 * port, because we probably sent the packet on one slave and got
2368 * it back on the other. Broadcast ARP replies are an exception
2369 * to this rule: the host has moved to another switch. */
2370 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2371 if (src_idx != -1 && src_idx != in_port->port_idx &&
2372 !is_bcast_arp_reply(flow)) {
2380 /* If the composed actions may be applied to any packet in the given 'flow',
2381 * returns true. Otherwise, the actions should only be applied to 'packet', or
2382 * not at all, if 'packet' was NULL. */
2384 process_flow(struct bridge *br, const flow_t *flow,
2385 const struct ofpbuf *packet, struct odp_actions *actions,
2386 tag_type *tags, uint16_t *nf_output_iface)
2388 struct port *in_port;
2389 struct port *out_port;
2393 /* Check whether we should drop packets in this flow. */
2394 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2399 /* Learn source MAC (but don't try to learn from revalidation). */
2401 update_learning_table(br, flow, vlan, in_port);
2404 /* Determine output port. */
2405 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags);
2406 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2407 out_port = br->ports[out_port_idx];
2408 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2409 /* If we are revalidating but don't have a learning entry then
2410 * eject the flow. Installing a flow that floods packets opens
2411 * up a window of time where we could learn from a packet reflected
2412 * on a bond and blackhole packets before the learning table is
2413 * updated to reflect the correct port. */
2416 out_port = FLOOD_PORT;
2419 /* Don't send packets out their input ports. */
2420 if (in_port == out_port) {
2426 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2433 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2436 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2437 const struct ofp_phy_port *opp,
2440 struct bridge *br = br_;
2441 struct iface *iface;
2444 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2450 if (reason == OFPPR_DELETE) {
2451 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2452 br->name, iface->name);
2453 iface_destroy(iface);
2454 if (!port->n_ifaces) {
2455 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2456 br->name, port->name);
2462 if (port->n_ifaces > 1) {
2463 bool up = !(opp->state & OFPPS_LINK_DOWN);
2464 bond_link_status_update(iface, up);
2465 port_update_bond_compat(port);
2471 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2472 struct odp_actions *actions, tag_type *tags,
2473 uint16_t *nf_output_iface, void *br_)
2475 struct bridge *br = br_;
2477 COVERAGE_INC(bridge_process_flow);
2478 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2482 bridge_account_flow_ofhook_cb(const flow_t *flow,
2483 const union odp_action *actions,
2484 size_t n_actions, unsigned long long int n_bytes,
2487 struct bridge *br = br_;
2488 const union odp_action *a;
2489 struct port *in_port;
2493 /* Feed information from the active flows back into the learning table
2494 * to ensure that table is always in sync with what is actually flowing
2495 * through the datapath. */
2496 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2497 update_learning_table(br, flow, vlan, in_port);
2500 if (!br->has_bonded_ports) {
2504 for (a = actions; a < &actions[n_actions]; a++) {
2505 if (a->type == ODPAT_OUTPUT) {
2506 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2507 if (out_port && out_port->n_ifaces >= 2) {
2508 struct bond_entry *e = lookup_bond_entry(out_port,
2510 e->tx_bytes += n_bytes;
2517 bridge_account_checkpoint_ofhook_cb(void *br_)
2519 struct bridge *br = br_;
2523 if (!br->has_bonded_ports) {
2528 for (i = 0; i < br->n_ports; i++) {
2529 struct port *port = br->ports[i];
2530 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2531 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2532 bond_rebalance_port(port);
2537 static struct ofhooks bridge_ofhooks = {
2538 bridge_port_changed_ofhook_cb,
2539 bridge_normal_ofhook_cb,
2540 bridge_account_flow_ofhook_cb,
2541 bridge_account_checkpoint_ofhook_cb,
2544 /* Bonding functions. */
2546 /* Statistics for a single interface on a bonded port, used for load-based
2547 * bond rebalancing. */
2548 struct slave_balance {
2549 struct iface *iface; /* The interface. */
2550 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2552 /* All the "bond_entry"s that are assigned to this interface, in order of
2553 * increasing tx_bytes. */
2554 struct bond_entry **hashes;
2558 /* Sorts pointers to pointers to bond_entries in ascending order by the
2559 * interface to which they are assigned, and within a single interface in
2560 * ascending order of bytes transmitted. */
2562 compare_bond_entries(const void *a_, const void *b_)
2564 const struct bond_entry *const *ap = a_;
2565 const struct bond_entry *const *bp = b_;
2566 const struct bond_entry *a = *ap;
2567 const struct bond_entry *b = *bp;
2568 if (a->iface_idx != b->iface_idx) {
2569 return a->iface_idx > b->iface_idx ? 1 : -1;
2570 } else if (a->tx_bytes != b->tx_bytes) {
2571 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2577 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2578 * *descending* order by number of bytes transmitted. */
2580 compare_slave_balance(const void *a_, const void *b_)
2582 const struct slave_balance *a = a_;
2583 const struct slave_balance *b = b_;
2584 if (a->iface->enabled != b->iface->enabled) {
2585 return a->iface->enabled ? -1 : 1;
2586 } else if (a->tx_bytes != b->tx_bytes) {
2587 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2594 swap_bals(struct slave_balance *a, struct slave_balance *b)
2596 struct slave_balance tmp = *a;
2601 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2602 * given that 'p' (and only 'p') might be in the wrong location.
2604 * This function invalidates 'p', since it might now be in a different memory
2607 resort_bals(struct slave_balance *p,
2608 struct slave_balance bals[], size_t n_bals)
2611 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2612 swap_bals(p, p - 1);
2614 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2615 swap_bals(p, p + 1);
2621 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2623 if (VLOG_IS_DBG_ENABLED()) {
2624 struct ds ds = DS_EMPTY_INITIALIZER;
2625 const struct slave_balance *b;
2627 for (b = bals; b < bals + n_bals; b++) {
2631 ds_put_char(&ds, ',');
2633 ds_put_format(&ds, " %s %"PRIu64"kB",
2634 b->iface->name, b->tx_bytes / 1024);
2636 if (!b->iface->enabled) {
2637 ds_put_cstr(&ds, " (disabled)");
2639 if (b->n_hashes > 0) {
2640 ds_put_cstr(&ds, " (");
2641 for (i = 0; i < b->n_hashes; i++) {
2642 const struct bond_entry *e = b->hashes[i];
2644 ds_put_cstr(&ds, " + ");
2646 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2647 e - port->bond_hash, e->tx_bytes / 1024);
2649 ds_put_cstr(&ds, ")");
2652 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2657 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2659 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2662 struct bond_entry *hash = from->hashes[hash_idx];
2663 struct port *port = from->iface->port;
2664 uint64_t delta = hash->tx_bytes;
2666 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2667 "from %s to %s (now carrying %"PRIu64"kB and "
2668 "%"PRIu64"kB load, respectively)",
2669 port->name, delta / 1024, hash - port->bond_hash,
2670 from->iface->name, to->iface->name,
2671 (from->tx_bytes - delta) / 1024,
2672 (to->tx_bytes + delta) / 1024);
2674 /* Delete element from from->hashes.
2676 * We don't bother to add the element to to->hashes because not only would
2677 * it require more work, the only purpose it would be to allow that hash to
2678 * be migrated to another slave in this rebalancing run, and there is no
2679 * point in doing that. */
2680 if (hash_idx == 0) {
2683 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2684 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2688 /* Shift load away from 'from' to 'to'. */
2689 from->tx_bytes -= delta;
2690 to->tx_bytes += delta;
2692 /* Arrange for flows to be revalidated. */
2693 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2694 hash->iface_idx = to->iface->port_ifidx;
2695 hash->iface_tag = tag_create_random();
2699 bond_rebalance_port(struct port *port)
2701 struct slave_balance bals[DP_MAX_PORTS];
2703 struct bond_entry *hashes[BOND_MASK + 1];
2704 struct slave_balance *b, *from, *to;
2705 struct bond_entry *e;
2708 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2709 * descending order of tx_bytes, so that bals[0] represents the most
2710 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2713 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2714 * array for each slave_balance structure, we sort our local array of
2715 * hashes in order by slave, so that all of the hashes for a given slave
2716 * become contiguous in memory, and then we point each 'hashes' members of
2717 * a slave_balance structure to the start of a contiguous group. */
2718 n_bals = port->n_ifaces;
2719 for (b = bals; b < &bals[n_bals]; b++) {
2720 b->iface = port->ifaces[b - bals];
2725 for (i = 0; i <= BOND_MASK; i++) {
2726 hashes[i] = &port->bond_hash[i];
2728 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2729 for (i = 0; i <= BOND_MASK; i++) {
2731 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2732 b = &bals[e->iface_idx];
2733 b->tx_bytes += e->tx_bytes;
2735 b->hashes = &hashes[i];
2740 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2741 log_bals(bals, n_bals, port);
2743 /* Discard slaves that aren't enabled (which were sorted to the back of the
2744 * array earlier). */
2745 while (!bals[n_bals - 1].iface->enabled) {
2752 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2753 to = &bals[n_bals - 1];
2754 for (from = bals; from < to; ) {
2755 uint64_t overload = from->tx_bytes - to->tx_bytes;
2756 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2757 /* The extra load on 'from' (and all less-loaded slaves), compared
2758 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2759 * it is less than ~1Mbps. No point in rebalancing. */
2761 } else if (from->n_hashes == 1) {
2762 /* 'from' only carries a single MAC hash, so we can't shift any
2763 * load away from it, even though we want to. */
2766 /* 'from' is carrying significantly more load than 'to', and that
2767 * load is split across at least two different hashes. Pick a hash
2768 * to migrate to 'to' (the least-loaded slave), given that doing so
2769 * must decrease the ratio of the load on the two slaves by at
2772 * The sort order we use means that we prefer to shift away the
2773 * smallest hashes instead of the biggest ones. There is little
2774 * reason behind this decision; we could use the opposite sort
2775 * order to shift away big hashes ahead of small ones. */
2779 for (i = 0; i < from->n_hashes; i++) {
2780 double old_ratio, new_ratio;
2781 uint64_t delta = from->hashes[i]->tx_bytes;
2783 if (delta == 0 || from->tx_bytes - delta == 0) {
2784 /* Pointless move. */
2788 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2790 if (to->tx_bytes == 0) {
2791 /* Nothing on the new slave, move it. */
2795 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2796 new_ratio = (double)(from->tx_bytes - delta) /
2797 (to->tx_bytes + delta);
2799 if (new_ratio == 0) {
2800 /* Should already be covered but check to prevent division
2805 if (new_ratio < 1) {
2806 new_ratio = 1 / new_ratio;
2809 if (old_ratio - new_ratio > 0.1) {
2810 /* Would decrease the ratio, move it. */
2814 if (i < from->n_hashes) {
2815 bond_shift_load(from, to, i);
2816 port->bond_compat_is_stale = true;
2818 /* If the result of the migration changed the relative order of
2819 * 'from' and 'to' swap them back to maintain invariants. */
2820 if (order_swapped) {
2821 swap_bals(from, to);
2824 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2825 * point to different slave_balance structures. It is only
2826 * valid to do these two operations in a row at all because we
2827 * know that 'from' will not move past 'to' and vice versa. */
2828 resort_bals(from, bals, n_bals);
2829 resort_bals(to, bals, n_bals);
2836 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2837 * historical data to decay to <1% in 7 rebalancing runs. */
2838 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2844 bond_send_learning_packets(struct port *port)
2846 struct bridge *br = port->bridge;
2847 struct mac_entry *e;
2848 struct ofpbuf packet;
2849 int error, n_packets, n_errors;
2851 if (!port->n_ifaces || port->active_iface < 0) {
2855 ofpbuf_init(&packet, 128);
2856 error = n_packets = n_errors = 0;
2857 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2858 union ofp_action actions[2], *a;
2864 if (e->port == port->port_idx
2865 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2869 /* Compose actions. */
2870 memset(actions, 0, sizeof actions);
2873 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2874 a->vlan_vid.len = htons(sizeof *a);
2875 a->vlan_vid.vlan_vid = htons(e->vlan);
2878 a->output.type = htons(OFPAT_OUTPUT);
2879 a->output.len = htons(sizeof *a);
2880 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2885 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2887 flow_extract(&packet, 0, ODPP_NONE, &flow);
2888 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2895 ofpbuf_uninit(&packet);
2898 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2899 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2900 "packets, last error was: %s",
2901 port->name, n_errors, n_packets, strerror(error));
2903 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2904 port->name, n_packets);
2908 /* Bonding unixctl user interface functions. */
2911 bond_unixctl_list(struct unixctl_conn *conn,
2912 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2914 struct ds ds = DS_EMPTY_INITIALIZER;
2915 const struct bridge *br;
2917 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2919 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2922 for (i = 0; i < br->n_ports; i++) {
2923 const struct port *port = br->ports[i];
2924 if (port->n_ifaces > 1) {
2927 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2928 for (j = 0; j < port->n_ifaces; j++) {
2929 const struct iface *iface = port->ifaces[j];
2931 ds_put_cstr(&ds, ", ");
2933 ds_put_cstr(&ds, iface->name);
2935 ds_put_char(&ds, '\n');
2939 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2943 static struct port *
2944 bond_find(const char *name)
2946 const struct bridge *br;
2948 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2951 for (i = 0; i < br->n_ports; i++) {
2952 struct port *port = br->ports[i];
2953 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2962 bond_unixctl_show(struct unixctl_conn *conn,
2963 const char *args, void *aux OVS_UNUSED)
2965 struct ds ds = DS_EMPTY_INITIALIZER;
2966 const struct port *port;
2969 port = bond_find(args);
2971 unixctl_command_reply(conn, 501, "no such bond");
2975 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2976 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2977 ds_put_format(&ds, "next rebalance: %lld ms\n",
2978 port->bond_next_rebalance - time_msec());
2979 for (j = 0; j < port->n_ifaces; j++) {
2980 const struct iface *iface = port->ifaces[j];
2981 struct bond_entry *be;
2984 ds_put_format(&ds, "slave %s: %s\n",
2985 iface->name, iface->enabled ? "enabled" : "disabled");
2986 if (j == port->active_iface) {
2987 ds_put_cstr(&ds, "\tactive slave\n");
2989 if (iface->delay_expires != LLONG_MAX) {
2990 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2991 iface->enabled ? "downdelay" : "updelay",
2992 iface->delay_expires - time_msec());
2996 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2997 int hash = be - port->bond_hash;
2998 struct mac_entry *me;
3000 if (be->iface_idx != j) {
3004 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3005 hash, be->tx_bytes / 1024);
3008 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3009 &port->bridge->ml->lrus) {
3012 if (bond_hash(me->mac) == hash
3013 && me->port != port->port_idx
3014 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3015 && dp_ifidx == iface->dp_ifidx)
3017 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3018 ETH_ADDR_ARGS(me->mac));
3023 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3028 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3029 void *aux OVS_UNUSED)
3031 char *args = (char *) args_;
3032 char *save_ptr = NULL;
3033 char *bond_s, *hash_s, *slave_s;
3034 uint8_t mac[ETH_ADDR_LEN];
3036 struct iface *iface;
3037 struct bond_entry *entry;
3040 bond_s = strtok_r(args, " ", &save_ptr);
3041 hash_s = strtok_r(NULL, " ", &save_ptr);
3042 slave_s = strtok_r(NULL, " ", &save_ptr);
3044 unixctl_command_reply(conn, 501,
3045 "usage: bond/migrate BOND HASH SLAVE");
3049 port = bond_find(bond_s);
3051 unixctl_command_reply(conn, 501, "no such bond");
3055 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3056 == ETH_ADDR_SCAN_COUNT) {
3057 hash = bond_hash(mac);
3058 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3059 hash = atoi(hash_s) & BOND_MASK;
3061 unixctl_command_reply(conn, 501, "bad hash");
3065 iface = port_lookup_iface(port, slave_s);
3067 unixctl_command_reply(conn, 501, "no such slave");
3071 if (!iface->enabled) {
3072 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3076 entry = &port->bond_hash[hash];
3077 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3078 entry->iface_idx = iface->port_ifidx;
3079 entry->iface_tag = tag_create_random();
3080 port->bond_compat_is_stale = true;
3081 unixctl_command_reply(conn, 200, "migrated");
3085 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3086 void *aux OVS_UNUSED)
3088 char *args = (char *) args_;
3089 char *save_ptr = NULL;
3090 char *bond_s, *slave_s;
3092 struct iface *iface;
3094 bond_s = strtok_r(args, " ", &save_ptr);
3095 slave_s = strtok_r(NULL, " ", &save_ptr);
3097 unixctl_command_reply(conn, 501,
3098 "usage: bond/set-active-slave BOND SLAVE");
3102 port = bond_find(bond_s);
3104 unixctl_command_reply(conn, 501, "no such bond");
3108 iface = port_lookup_iface(port, slave_s);
3110 unixctl_command_reply(conn, 501, "no such slave");
3114 if (!iface->enabled) {
3115 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3119 if (port->active_iface != iface->port_ifidx) {
3120 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3121 port->active_iface = iface->port_ifidx;
3122 port->active_iface_tag = tag_create_random();
3123 VLOG_INFO("port %s: active interface is now %s",
3124 port->name, iface->name);
3125 bond_send_learning_packets(port);
3126 unixctl_command_reply(conn, 200, "done");
3128 unixctl_command_reply(conn, 200, "no change");
3133 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3135 char *args = (char *) args_;
3136 char *save_ptr = NULL;
3137 char *bond_s, *slave_s;
3139 struct iface *iface;
3141 bond_s = strtok_r(args, " ", &save_ptr);
3142 slave_s = strtok_r(NULL, " ", &save_ptr);
3144 unixctl_command_reply(conn, 501,
3145 "usage: bond/enable/disable-slave BOND SLAVE");
3149 port = bond_find(bond_s);
3151 unixctl_command_reply(conn, 501, "no such bond");
3155 iface = port_lookup_iface(port, slave_s);
3157 unixctl_command_reply(conn, 501, "no such slave");
3161 bond_enable_slave(iface, enable);
3162 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3166 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3167 void *aux OVS_UNUSED)
3169 enable_slave(conn, args, true);
3173 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3174 void *aux OVS_UNUSED)
3176 enable_slave(conn, args, false);
3180 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3181 void *aux OVS_UNUSED)
3183 uint8_t mac[ETH_ADDR_LEN];
3187 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3188 == ETH_ADDR_SCAN_COUNT) {
3189 hash = bond_hash(mac);
3191 hash_cstr = xasprintf("%u", hash);
3192 unixctl_command_reply(conn, 200, hash_cstr);
3195 unixctl_command_reply(conn, 501, "invalid mac");
3202 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3203 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3204 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3205 unixctl_command_register("bond/set-active-slave",
3206 bond_unixctl_set_active_slave, NULL);
3207 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3209 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3211 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3214 /* Port functions. */
3216 static struct port *
3217 port_create(struct bridge *br, const char *name)
3221 port = xzalloc(sizeof *port);
3223 port->port_idx = br->n_ports;
3225 port->trunks = NULL;
3226 port->name = xstrdup(name);
3227 port->active_iface = -1;
3229 if (br->n_ports >= br->allocated_ports) {
3230 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3233 br->ports[br->n_ports++] = port;
3234 shash_add_assert(&br->port_by_name, port->name, port);
3236 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3243 get_port_other_config(const struct ovsrec_port *port, const char *key,
3244 const char *default_value)
3246 const char *value = get_ovsrec_key_value(key,
3247 port->key_other_config,
3248 port->value_other_config,
3249 port->n_other_config);
3250 return value ? value : default_value;
3254 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3256 struct shash new_ifaces;
3259 /* Collect list of new interfaces. */
3260 shash_init(&new_ifaces);
3261 for (i = 0; i < cfg->n_interfaces; i++) {
3262 const char *name = cfg->interfaces[i]->name;
3263 shash_add_once(&new_ifaces, name, NULL);
3266 /* Get rid of deleted interfaces. */
3267 for (i = 0; i < port->n_ifaces; ) {
3268 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3269 iface_destroy(port->ifaces[i]);
3275 shash_destroy(&new_ifaces);
3279 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3281 struct shash new_ifaces;
3282 long long int next_rebalance;
3283 unsigned long *trunks;
3289 /* Update settings. */
3290 port->updelay = cfg->bond_updelay;
3291 if (port->updelay < 0) {
3294 port->updelay = cfg->bond_downdelay;
3295 if (port->downdelay < 0) {
3296 port->downdelay = 0;
3298 port->bond_rebalance_interval = atoi(
3299 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3300 if (port->bond_rebalance_interval < 1000) {
3301 port->bond_rebalance_interval = 1000;
3303 next_rebalance = time_msec() + port->bond_rebalance_interval;
3304 if (port->bond_next_rebalance > next_rebalance) {
3305 port->bond_next_rebalance = next_rebalance;
3308 /* Add new interfaces and update 'cfg' member of existing ones. */
3309 shash_init(&new_ifaces);
3310 for (i = 0; i < cfg->n_interfaces; i++) {
3311 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3312 struct iface *iface;
3314 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3315 VLOG_WARN("port %s: %s specified twice as port interface",
3316 port->name, if_cfg->name);
3320 iface = iface_lookup(port->bridge, if_cfg->name);
3322 if (iface->port != port) {
3323 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3325 port->bridge->name, if_cfg->name, iface->port->name);
3328 iface->cfg = if_cfg;
3330 iface_create(port, if_cfg);
3333 shash_destroy(&new_ifaces);
3338 if (port->n_ifaces < 2) {
3340 if (vlan >= 0 && vlan <= 4095) {
3341 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3346 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3347 * they even work as-is. But they have not been tested. */
3348 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3352 if (port->vlan != vlan) {
3354 bridge_flush(port->bridge);
3357 /* Get trunked VLANs. */
3359 if (vlan < 0 && cfg->n_trunks) {
3363 trunks = bitmap_allocate(4096);
3365 for (i = 0; i < cfg->n_trunks; i++) {
3366 int trunk = cfg->trunks[i];
3368 bitmap_set1(trunks, trunk);
3374 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3375 port->name, cfg->n_trunks);
3377 if (n_errors == cfg->n_trunks) {
3378 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3380 bitmap_free(trunks);
3383 } else if (vlan >= 0 && cfg->n_trunks) {
3384 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3388 ? port->trunks != NULL
3389 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3390 bridge_flush(port->bridge);
3392 bitmap_free(port->trunks);
3393 port->trunks = trunks;
3395 shash_destroy(&new_ifaces);
3399 port_destroy(struct port *port)
3402 struct bridge *br = port->bridge;
3406 proc_net_compat_update_vlan(port->name, NULL, 0);
3407 proc_net_compat_update_bond(port->name, NULL);
3409 for (i = 0; i < MAX_MIRRORS; i++) {
3410 struct mirror *m = br->mirrors[i];
3411 if (m && m->out_port == port) {
3416 while (port->n_ifaces > 0) {
3417 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3420 shash_find_and_delete_assert(&br->port_by_name, port->name);
3422 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3423 del->port_idx = port->port_idx;
3426 bitmap_free(port->trunks);
3433 static struct port *
3434 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3436 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3437 return iface ? iface->port : NULL;
3440 static struct port *
3441 port_lookup(const struct bridge *br, const char *name)
3443 return shash_find_data(&br->port_by_name, name);
3446 static struct iface *
3447 port_lookup_iface(const struct port *port, const char *name)
3449 struct iface *iface = iface_lookup(port->bridge, name);
3450 return iface && iface->port == port ? iface : NULL;
3454 port_update_bonding(struct port *port)
3456 if (port->n_ifaces < 2) {
3457 /* Not a bonded port. */
3458 if (port->bond_hash) {
3459 free(port->bond_hash);
3460 port->bond_hash = NULL;
3461 port->bond_compat_is_stale = true;
3462 port->bond_fake_iface = false;
3465 if (!port->bond_hash) {
3468 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3469 for (i = 0; i <= BOND_MASK; i++) {
3470 struct bond_entry *e = &port->bond_hash[i];
3474 port->no_ifaces_tag = tag_create_random();
3475 bond_choose_active_iface(port);
3476 port->bond_next_rebalance
3477 = time_msec() + port->bond_rebalance_interval;
3479 if (port->cfg->bond_fake_iface) {
3480 port->bond_next_fake_iface_update = time_msec();
3483 port->bond_compat_is_stale = true;
3484 port->bond_fake_iface = port->cfg->bond_fake_iface;
3489 port_update_bond_compat(struct port *port)
3491 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3492 struct compat_bond bond;
3495 if (port->n_ifaces < 2) {
3496 proc_net_compat_update_bond(port->name, NULL);
3501 bond.updelay = port->updelay;
3502 bond.downdelay = port->downdelay;
3505 bond.hashes = compat_hashes;
3506 if (port->bond_hash) {
3507 const struct bond_entry *e;
3508 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3509 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3510 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3511 cbh->hash = e - port->bond_hash;
3512 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3517 bond.n_slaves = port->n_ifaces;
3518 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3519 for (i = 0; i < port->n_ifaces; i++) {
3520 struct iface *iface = port->ifaces[i];
3521 struct compat_bond_slave *slave = &bond.slaves[i];
3522 slave->name = iface->name;
3524 /* We need to make the same determination as the Linux bonding
3525 * code to determine whether a slave should be consider "up".
3526 * The Linux function bond_miimon_inspect() supports four
3527 * BOND_LINK_* states:
3529 * - BOND_LINK_UP: carrier detected, updelay has passed.
3530 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3531 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3532 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3534 * The function bond_info_show_slave() only considers BOND_LINK_UP
3535 * to be "up" and anything else to be "down".
3537 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3541 netdev_get_etheraddr(iface->netdev, slave->mac);
3544 if (port->bond_fake_iface) {
3545 struct netdev *bond_netdev;
3547 if (!netdev_open_default(port->name, &bond_netdev)) {
3549 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3551 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3553 netdev_close(bond_netdev);
3557 proc_net_compat_update_bond(port->name, &bond);
3562 port_update_vlan_compat(struct port *port)
3564 struct bridge *br = port->bridge;
3565 char *vlandev_name = NULL;
3567 if (port->vlan > 0) {
3568 /* Figure out the name that the VLAN device should actually have, if it
3569 * existed. This takes some work because the VLAN device would not
3570 * have port->name in its name; rather, it would have the trunk port's
3571 * name, and 'port' would be attached to a bridge that also had the
3572 * VLAN device one of its ports. So we need to find a trunk port that
3573 * includes port->vlan.
3575 * There might be more than one candidate. This doesn't happen on
3576 * XenServer, so if it happens we just pick the first choice in
3577 * alphabetical order instead of creating multiple VLAN devices. */
3579 for (i = 0; i < br->n_ports; i++) {
3580 struct port *p = br->ports[i];
3581 if (port_trunks_vlan(p, port->vlan)
3583 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3585 uint8_t ea[ETH_ADDR_LEN];
3586 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3587 if (!eth_addr_is_multicast(ea) &&
3588 !eth_addr_is_reserved(ea) &&
3589 !eth_addr_is_zero(ea)) {
3590 vlandev_name = p->name;
3595 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3598 /* Interface functions. */
3600 static struct iface *
3601 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3603 struct bridge *br = port->bridge;
3604 struct iface *iface;
3605 char *name = if_cfg->name;
3608 iface = xzalloc(sizeof *iface);
3610 iface->port_ifidx = port->n_ifaces;
3611 iface->name = xstrdup(name);
3612 iface->dp_ifidx = -1;
3613 iface->tag = tag_create_random();
3614 iface->delay_expires = LLONG_MAX;
3615 iface->netdev = NULL;
3616 iface->cfg = if_cfg;
3618 /* Attempt to create the network interface in case it doesn't exist yet. */
3619 if (!iface_is_internal(br, iface->name)) {
3620 error = set_up_iface(if_cfg, iface, true);
3622 VLOG_WARN("could not create iface %s: %s", iface->name,
3631 shash_add_assert(&br->iface_by_name, iface->name, iface);
3633 if (port->n_ifaces >= port->allocated_ifaces) {
3634 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3635 sizeof *port->ifaces);
3637 port->ifaces[port->n_ifaces++] = iface;
3638 if (port->n_ifaces > 1) {
3639 br->has_bonded_ports = true;
3642 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3650 iface_destroy(struct iface *iface)
3653 struct port *port = iface->port;
3654 struct bridge *br = port->bridge;
3655 bool del_active = port->active_iface == iface->port_ifidx;
3658 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3660 if (iface->dp_ifidx >= 0) {
3661 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3664 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3665 del->port_ifidx = iface->port_ifidx;
3667 netdev_close(iface->netdev);
3670 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3671 bond_choose_active_iface(port);
3672 bond_send_learning_packets(port);
3678 bridge_flush(port->bridge);
3682 static struct iface *
3683 iface_lookup(const struct bridge *br, const char *name)
3685 return shash_find_data(&br->iface_by_name, name);
3688 static struct iface *
3689 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3691 return port_array_get(&br->ifaces, dp_ifidx);
3694 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3695 * 'br', that is, an interface that is entirely simulated within the datapath.
3696 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3697 * interfaces are created by setting "iface.<iface>.internal = true".
3699 * In addition, we have a kluge-y feature that creates an internal port with
3700 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3701 * This feature needs to go away in the long term. Until then, this is one
3702 * reason why this function takes a name instead of a struct iface: the fake
3703 * interfaces created this way do not have a struct iface. */
3705 iface_is_internal(const struct bridge *br, const char *if_name)
3707 struct iface *iface;
3710 if (!strcmp(if_name, br->name)) {
3714 iface = iface_lookup(br, if_name);
3715 if (iface && !strcmp(iface->cfg->type, "internal")) {
3719 port = port_lookup(br, if_name);
3720 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3726 /* Set Ethernet address of 'iface', if one is specified in the configuration
3729 iface_set_mac(struct iface *iface)
3731 uint8_t ea[ETH_ADDR_LEN];
3733 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3734 if (eth_addr_is_multicast(ea)) {
3735 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3737 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3738 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3739 iface->name, iface->name);
3741 int error = netdev_set_etheraddr(iface->netdev, ea);
3743 VLOG_ERR("interface %s: setting MAC failed (%s)",
3744 iface->name, strerror(error));
3750 /* Port mirroring. */
3753 mirror_reconfigure(struct bridge *br)
3755 struct shash old_mirrors, new_mirrors;
3756 struct shash_node *node;
3757 unsigned long *rspan_vlans;
3760 /* Collect old mirrors. */
3761 shash_init(&old_mirrors);
3762 for (i = 0; i < MAX_MIRRORS; i++) {
3763 if (br->mirrors[i]) {
3764 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3768 /* Collect new mirrors. */
3769 shash_init(&new_mirrors);
3770 for (i = 0; i < br->cfg->n_mirrors; i++) {
3771 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3772 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3773 VLOG_WARN("bridge %s: %s specified twice as mirror",
3774 br->name, cfg->name);
3778 /* Get rid of deleted mirrors and add new mirrors. */
3779 SHASH_FOR_EACH (node, &old_mirrors) {
3780 if (!shash_find(&new_mirrors, node->name)) {
3781 mirror_destroy(node->data);
3784 SHASH_FOR_EACH (node, &new_mirrors) {
3785 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3787 mirror = mirror_create(br, node->name);
3792 mirror_reconfigure_one(mirror, node->data);
3794 shash_destroy(&old_mirrors);
3795 shash_destroy(&new_mirrors);
3797 /* Update port reserved status. */
3798 for (i = 0; i < br->n_ports; i++) {
3799 br->ports[i]->is_mirror_output_port = false;
3801 for (i = 0; i < MAX_MIRRORS; i++) {
3802 struct mirror *m = br->mirrors[i];
3803 if (m && m->out_port) {
3804 m->out_port->is_mirror_output_port = true;
3808 /* Update flooded vlans (for RSPAN). */
3810 if (br->cfg->n_flood_vlans) {
3811 rspan_vlans = bitmap_allocate(4096);
3813 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3814 int64_t vlan = br->cfg->flood_vlans[i];
3815 if (vlan >= 0 && vlan < 4096) {
3816 bitmap_set1(rspan_vlans, vlan);
3817 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3820 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3825 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3830 static struct mirror *
3831 mirror_create(struct bridge *br, const char *name)
3836 for (i = 0; ; i++) {
3837 if (i >= MAX_MIRRORS) {
3838 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3839 "cannot create %s", br->name, MAX_MIRRORS, name);
3842 if (!br->mirrors[i]) {
3847 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3850 br->mirrors[i] = m = xzalloc(sizeof *m);
3853 m->name = xstrdup(name);
3854 shash_init(&m->src_ports);
3855 shash_init(&m->dst_ports);
3865 mirror_destroy(struct mirror *m)
3868 struct bridge *br = m->bridge;
3871 for (i = 0; i < br->n_ports; i++) {
3872 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3873 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3876 shash_destroy(&m->src_ports);
3877 shash_destroy(&m->dst_ports);
3880 m->bridge->mirrors[m->idx] = NULL;
3888 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3889 struct shash *names)
3893 for (i = 0; i < n_ports; i++) {
3894 const char *name = ports[i]->name;
3895 if (port_lookup(m->bridge, name)) {
3896 shash_add_once(names, name, NULL);
3898 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3899 "port %s", m->bridge->name, m->name, name);
3905 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3911 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3913 for (i = 0; i < cfg->n_select_vlan; i++) {
3914 int64_t vlan = cfg->select_vlan[i];
3915 if (vlan < 0 || vlan > 4095) {
3916 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3917 m->bridge->name, m->name, vlan);
3919 (*vlans)[n_vlans++] = vlan;
3926 vlan_is_mirrored(const struct mirror *m, int vlan)
3930 for (i = 0; i < m->n_vlans; i++) {
3931 if (m->vlans[i] == vlan) {
3939 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3943 for (i = 0; i < m->n_vlans; i++) {
3944 if (port_trunks_vlan(p, m->vlans[i])) {
3952 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3954 struct shash src_ports, dst_ports;
3955 mirror_mask_t mirror_bit;
3956 struct port *out_port;
3962 /* Get output port. */
3963 if (cfg->output_port) {
3964 out_port = port_lookup(m->bridge, cfg->output_port->name);
3966 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3967 m->bridge->name, m->name);
3973 if (cfg->output_vlan) {
3974 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3975 "output vlan; ignoring output vlan",
3976 m->bridge->name, m->name);
3978 } else if (cfg->output_vlan) {
3980 out_vlan = *cfg->output_vlan;
3982 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3983 m->bridge->name, m->name);
3988 shash_init(&src_ports);
3989 shash_init(&dst_ports);
3990 if (cfg->select_all) {
3991 for (i = 0; i < m->bridge->n_ports; i++) {
3992 const char *name = m->bridge->ports[i]->name;
3993 shash_add_once(&src_ports, name, NULL);
3994 shash_add_once(&dst_ports, name, NULL);
3999 /* Get ports, and drop duplicates and ports that don't exist. */
4000 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4002 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4005 /* Get all the vlans, and drop duplicate and invalid vlans. */
4006 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4009 /* Update mirror data. */
4010 if (!shash_equal_keys(&m->src_ports, &src_ports)
4011 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4012 || m->n_vlans != n_vlans
4013 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4014 || m->out_port != out_port
4015 || m->out_vlan != out_vlan) {
4016 bridge_flush(m->bridge);
4018 shash_swap(&m->src_ports, &src_ports);
4019 shash_swap(&m->dst_ports, &dst_ports);
4022 m->n_vlans = n_vlans;
4023 m->out_port = out_port;
4024 m->out_vlan = out_vlan;
4027 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4028 for (i = 0; i < m->bridge->n_ports; i++) {
4029 struct port *port = m->bridge->ports[i];
4031 if (shash_find(&m->src_ports, port->name)
4034 ? port_trunks_any_mirrored_vlan(m, port)
4035 : vlan_is_mirrored(m, port->vlan)))) {
4036 port->src_mirrors |= mirror_bit;
4038 port->src_mirrors &= ~mirror_bit;
4041 if (shash_find(&m->dst_ports, port->name)) {
4042 port->dst_mirrors |= mirror_bit;
4044 port->dst_mirrors &= ~mirror_bit;
4049 shash_destroy(&src_ports);
4050 shash_destroy(&dst_ports);