1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
49 #include "ovsdb-data.h"
51 #include "poll-loop.h"
52 #include "port-array.h"
53 #include "proc-net-compat.h"
57 #include "socket-util.h"
58 #include "stream-ssl.h"
64 #include "vswitchd/vswitch-idl.h"
65 #include "xenserver.h"
68 #include "sflow_api.h"
70 VLOG_DEFINE_THIS_MODULE(bridge)
78 /* These members are always valid. */
79 struct port *port; /* Containing port. */
80 size_t port_ifidx; /* Index within containing port. */
81 char *name; /* Host network device name. */
82 tag_type tag; /* Tag associated with this interface. */
83 long long delay_expires; /* Time after which 'enabled' may change. */
85 /* These members are valid only after bridge_reconfigure() causes them to
87 int dp_ifidx; /* Index within kernel datapath. */
88 struct netdev *netdev; /* Network device. */
89 bool enabled; /* May be chosen for flows? */
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;
108 struct uuid uuid; /* UUID of this "mirror" record in database. */
110 /* Selection criteria. */
111 struct shash src_ports; /* Name is port name; data is always NULL. */
112 struct shash dst_ports; /* Name is port name; data is always NULL. */
117 struct port *out_port;
121 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
123 struct bridge *bridge;
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
128 const struct ovsrec_port *cfg;
131 /* An ordinary bridge port has 1 interface.
132 * A bridge port for bonding has at least 2 interfaces. */
133 struct iface **ifaces;
134 size_t n_ifaces, allocated_ifaces;
137 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
138 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
139 tag_type active_iface_tag; /* Tag for bcast flows. */
140 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
141 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
142 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
143 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
144 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
145 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
146 long long int bond_next_rebalance; /* Next rebalancing time. */
148 /* Port mirroring info. */
149 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
150 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
151 bool is_mirror_output_port; /* Does port mirroring send frames here? */
154 #define DP_MAX_PORTS 255
156 struct list node; /* Node in global list of bridges. */
157 char *name; /* User-specified arbitrary name. */
158 struct mac_learning *ml; /* MAC learning table. */
159 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
160 const struct ovsrec_bridge *cfg;
162 /* OpenFlow switch processing. */
163 struct ofproto *ofproto; /* OpenFlow switch. */
165 /* Kernel datapath information. */
166 struct dpif *dpif; /* Datapath. */
167 struct port_array ifaces; /* Indexed by kernel datapath port number. */
171 size_t n_ports, allocated_ports;
172 struct shash iface_by_name; /* "struct iface"s indexed by name. */
173 struct shash port_by_name; /* "struct port"s indexed by name. */
176 bool has_bonded_ports;
181 /* Port mirroring. */
182 struct mirror *mirrors[MAX_MIRRORS];
185 /* List of all bridges. */
186 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
188 /* OVSDB IDL used to obtain configuration. */
189 static struct ovsdb_idl *idl;
191 /* Each time this timer expires, the bridge fetches statistics for every
192 * interface and pushes them into the database. */
193 #define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
194 static long long int iface_stats_timer = LLONG_MIN;
196 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
197 static void bridge_destroy(struct bridge *);
198 static struct bridge *bridge_lookup(const char *name);
199 static unixctl_cb_func bridge_unixctl_dump_flows;
200 static unixctl_cb_func bridge_unixctl_reconnect;
201 static int bridge_run_one(struct bridge *);
202 static size_t bridge_get_controllers(const struct bridge *br,
203 struct ovsrec_controller ***controllersp);
204 static void bridge_reconfigure_one(struct bridge *);
205 static void bridge_reconfigure_remotes(struct bridge *,
206 const struct sockaddr_in *managers,
208 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
209 static void bridge_fetch_dp_ifaces(struct bridge *);
210 static void bridge_flush(struct bridge *);
211 static void bridge_pick_local_hw_addr(struct bridge *,
212 uint8_t ea[ETH_ADDR_LEN],
213 struct iface **hw_addr_iface);
214 static uint64_t bridge_pick_datapath_id(struct bridge *,
215 const uint8_t bridge_ea[ETH_ADDR_LEN],
216 struct iface *hw_addr_iface);
217 static struct iface *bridge_get_local_iface(struct bridge *);
218 static uint64_t dpid_from_hash(const void *, size_t nbytes);
220 static unixctl_cb_func bridge_unixctl_fdb_show;
222 static void bond_init(void);
223 static void bond_run(struct bridge *);
224 static void bond_wait(struct bridge *);
225 static void bond_rebalance_port(struct port *);
226 static void bond_send_learning_packets(struct port *);
227 static void bond_enable_slave(struct iface *iface, bool enable);
229 static struct port *port_create(struct bridge *, const char *name);
230 static void port_reconfigure(struct port *, const struct ovsrec_port *);
231 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
232 static void port_destroy(struct port *);
233 static struct port *port_lookup(const struct bridge *, const char *name);
234 static struct iface *port_lookup_iface(const struct port *, const char *name);
235 static struct port *port_from_dp_ifidx(const struct bridge *,
237 static void port_update_bond_compat(struct port *);
238 static void port_update_vlan_compat(struct port *);
239 static void port_update_bonding(struct port *);
241 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
242 static void mirror_destroy(struct mirror *);
243 static void mirror_reconfigure(struct bridge *);
244 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
245 static bool vlan_is_mirrored(const struct mirror *, int vlan);
247 static struct iface *iface_create(struct port *port,
248 const struct ovsrec_interface *if_cfg);
249 static void iface_destroy(struct iface *);
250 static struct iface *iface_lookup(const struct bridge *, const char *name);
251 static struct iface *iface_from_dp_ifidx(const struct bridge *,
253 static bool iface_is_internal(const struct bridge *, const char *name);
254 static void iface_set_mac(struct iface *);
255 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
257 /* Hooks into ofproto processing. */
258 static struct ofhooks bridge_ofhooks;
260 /* Public functions. */
262 /* Initializes the bridge module, configuring it to obtain its configuration
263 * from an OVSDB server accessed over 'remote', which should be a string in a
264 * form acceptable to ovsdb_idl_create(). */
266 bridge_init(const char *remote)
268 /* Create connection to database. */
269 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
271 /* Register unixctl commands. */
272 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
273 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
275 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
280 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
281 * but for which the ovs-vswitchd configuration 'cfg' is required. */
283 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
285 static bool already_configured_once;
286 struct svec bridge_names;
287 struct svec dpif_names, dpif_types;
290 /* Only do this once per ovs-vswitchd run. */
291 if (already_configured_once) {
294 already_configured_once = true;
296 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
298 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
299 svec_init(&bridge_names);
300 for (i = 0; i < cfg->n_bridges; i++) {
301 svec_add(&bridge_names, cfg->bridges[i]->name);
303 svec_sort(&bridge_names);
305 /* Iterate over all system dpifs and delete any of them that do not appear
307 svec_init(&dpif_names);
308 svec_init(&dpif_types);
309 dp_enumerate_types(&dpif_types);
310 for (i = 0; i < dpif_types.n; i++) {
315 dp_enumerate_names(dpif_types.names[i], &dpif_names);
317 /* For each dpif... */
318 for (j = 0; j < dpif_names.n; j++) {
319 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
321 struct svec all_names;
324 /* ...check whether any of its names is in 'bridge_names'. */
325 svec_init(&all_names);
326 dpif_get_all_names(dpif, &all_names);
327 for (k = 0; k < all_names.n; k++) {
328 if (svec_contains(&bridge_names, all_names.names[k])) {
333 /* No. Delete the dpif. */
337 svec_destroy(&all_names);
342 svec_destroy(&bridge_names);
343 svec_destroy(&dpif_names);
344 svec_destroy(&dpif_types);
349 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
351 /* XXX SSL should be configurable on a per-bridge basis. */
353 stream_ssl_set_private_key_file(ssl->private_key);
354 stream_ssl_set_certificate_file(ssl->certificate);
355 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
360 /* Attempt to create the network device 'iface_name' through the netdev
363 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
366 struct shash options;
370 shash_init(&options);
371 for (i = 0; i < iface_cfg->n_options; i++) {
372 shash_add(&options, iface_cfg->key_options[i],
373 xstrdup(iface_cfg->value_options[i]));
377 struct netdev_options netdev_options;
379 memset(&netdev_options, 0, sizeof netdev_options);
380 netdev_options.name = iface_cfg->name;
381 if (!strcmp(iface_cfg->type, "internal")) {
382 /* An "internal" config type maps to a netdev "system" type. */
383 netdev_options.type = "system";
385 netdev_options.type = iface_cfg->type;
387 netdev_options.args = &options;
388 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
390 error = netdev_open(&netdev_options, &iface->netdev);
393 netdev_get_carrier(iface->netdev, &iface->enabled);
395 } else if (iface->netdev) {
396 const char *netdev_type = netdev_get_type(iface->netdev);
397 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
398 ? iface_cfg->type : NULL;
400 /* An "internal" config type maps to a netdev "system" type. */
401 if (iface_type && !strcmp(iface_type, "internal")) {
402 iface_type = "system";
405 if (!iface_type || !strcmp(netdev_type, iface_type)) {
406 error = netdev_reconfigure(iface->netdev, &options);
408 VLOG_WARN("%s: attempting change device type from %s to %s",
409 iface_cfg->name, netdev_type, iface_type);
413 shash_destroy_free_data(&options);
419 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
421 return set_up_iface(iface_cfg, iface, false);
425 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
426 void *aux OVS_UNUSED)
428 if (!iface->netdev) {
429 int error = set_up_iface(iface->cfg, iface, true);
431 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
441 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
442 void *aux OVS_UNUSED)
444 if (iface->dp_ifidx >= 0) {
445 VLOG_DBG("%s has interface %s on port %d",
447 iface->name, iface->dp_ifidx);
450 VLOG_ERR("%s interface not in %s, dropping",
451 iface->name, dpif_name(br->dpif));
457 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
458 void *aux OVS_UNUSED)
460 /* Set policing attributes. */
461 netdev_set_policing(iface->netdev,
462 iface->cfg->ingress_policing_rate,
463 iface->cfg->ingress_policing_burst);
465 /* Set MAC address of internal interfaces other than the local
467 if (iface->dp_ifidx != ODPP_LOCAL
468 && iface_is_internal(br, iface->name)) {
469 iface_set_mac(iface);
475 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
476 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
477 * deletes from 'br' any ports that no longer have any interfaces. */
479 iterate_and_prune_ifaces(struct bridge *br,
480 bool (*cb)(struct bridge *, struct iface *,
486 for (i = 0; i < br->n_ports; ) {
487 struct port *port = br->ports[i];
488 for (j = 0; j < port->n_ifaces; ) {
489 struct iface *iface = port->ifaces[j];
490 if (cb(br, iface, aux)) {
493 iface_destroy(iface);
497 if (port->n_ifaces) {
500 VLOG_ERR("%s port has no interfaces, dropping", port->name);
506 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
507 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
508 * responsible for freeing '*managersp' (with free()).
510 * You may be asking yourself "why does ovs-vswitchd care?", because
511 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
512 * should not be and in fact is not directly involved in that. But
513 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
514 * it has to tell in-band control where the managers are to enable that.
517 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
518 struct sockaddr_in **managersp, size_t *n_managersp)
520 struct sockaddr_in *managers = NULL;
521 size_t n_managers = 0;
523 if (ovs_cfg->n_managers > 0) {
526 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
527 for (i = 0; i < ovs_cfg->n_managers; i++) {
528 const char *name = ovs_cfg->managers[i];
529 struct sockaddr_in *sin = &managers[i];
531 if ((!strncmp(name, "tcp:", 4)
532 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
533 (!strncmp(name, "ssl:", 4)
534 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
540 *managersp = managers;
541 *n_managersp = n_managers;
545 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
547 struct shash old_br, new_br;
548 struct shash_node *node;
549 struct bridge *br, *next;
550 struct sockaddr_in *managers;
553 int sflow_bridge_number;
555 COVERAGE_INC(bridge_reconfigure);
557 collect_managers(ovs_cfg, &managers, &n_managers);
559 /* Collect old and new bridges. */
562 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
563 shash_add(&old_br, br->name, br);
565 for (i = 0; i < ovs_cfg->n_bridges; i++) {
566 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
567 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
568 VLOG_WARN("more than one bridge named %s", br_cfg->name);
572 /* Get rid of deleted bridges and add new bridges. */
573 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
574 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
581 SHASH_FOR_EACH (node, &new_br) {
582 const char *br_name = node->name;
583 const struct ovsrec_bridge *br_cfg = node->data;
584 br = shash_find_data(&old_br, br_name);
586 /* If the bridge datapath type has changed, we need to tear it
587 * down and recreate. */
588 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
590 bridge_create(br_cfg);
593 bridge_create(br_cfg);
596 shash_destroy(&old_br);
597 shash_destroy(&new_br);
601 bridge_configure_ssl(ovs_cfg->ssl);
604 /* Reconfigure all bridges. */
605 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
606 bridge_reconfigure_one(br);
609 /* Add and delete ports on all datapaths.
611 * The kernel will reject any attempt to add a given port to a datapath if
612 * that port already belongs to a different datapath, so we must do all
613 * port deletions before any port additions. */
614 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
615 struct odp_port *dpif_ports;
617 struct shash want_ifaces;
619 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
620 bridge_get_all_ifaces(br, &want_ifaces);
621 for (i = 0; i < n_dpif_ports; i++) {
622 const struct odp_port *p = &dpif_ports[i];
623 if (!shash_find(&want_ifaces, p->devname)
624 && strcmp(p->devname, br->name)) {
625 int retval = dpif_port_del(br->dpif, p->port);
627 VLOG_ERR("failed to remove %s interface from %s: %s",
628 p->devname, dpif_name(br->dpif),
633 shash_destroy(&want_ifaces);
636 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
637 struct odp_port *dpif_ports;
639 struct shash cur_ifaces, want_ifaces;
640 struct shash_node *node;
642 /* Get the set of interfaces currently in this datapath. */
643 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
644 shash_init(&cur_ifaces);
645 for (i = 0; i < n_dpif_ports; i++) {
646 const char *name = dpif_ports[i].devname;
647 shash_add_once(&cur_ifaces, name, NULL);
651 /* Get the set of interfaces we want on this datapath. */
652 bridge_get_all_ifaces(br, &want_ifaces);
654 SHASH_FOR_EACH (node, &want_ifaces) {
655 const char *if_name = node->name;
656 struct iface *iface = node->data;
658 if (shash_find(&cur_ifaces, if_name)) {
659 /* Already exists, just reconfigure it. */
661 reconfigure_iface(iface->cfg, iface);
664 /* Need to add to datapath. */
668 /* Add to datapath. */
669 internal = iface_is_internal(br, if_name);
670 error = dpif_port_add(br->dpif, if_name,
671 internal ? ODP_PORT_INTERNAL : 0, NULL);
672 if (error == EFBIG) {
673 VLOG_ERR("ran out of valid port numbers on %s",
674 dpif_name(br->dpif));
677 VLOG_ERR("failed to add %s interface to %s: %s",
678 if_name, dpif_name(br->dpif), strerror(error));
682 shash_destroy(&cur_ifaces);
683 shash_destroy(&want_ifaces);
685 sflow_bridge_number = 0;
686 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
689 struct iface *local_iface;
690 struct iface *hw_addr_iface;
693 bridge_fetch_dp_ifaces(br);
695 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
696 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
698 /* Pick local port hardware address, datapath ID. */
699 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
700 local_iface = bridge_get_local_iface(br);
702 int error = netdev_set_etheraddr(local_iface->netdev, ea);
704 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
705 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
706 "Ethernet address: %s",
707 br->name, strerror(error));
711 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
712 ofproto_set_datapath_id(br->ofproto, dpid);
714 dpid_string = xasprintf("%016"PRIx64, dpid);
715 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
718 /* Set NetFlow configuration on this bridge. */
719 if (br->cfg->netflow) {
720 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
721 struct netflow_options opts;
723 memset(&opts, 0, sizeof opts);
725 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
726 if (nf_cfg->engine_type) {
727 opts.engine_type = *nf_cfg->engine_type;
729 if (nf_cfg->engine_id) {
730 opts.engine_id = *nf_cfg->engine_id;
733 opts.active_timeout = nf_cfg->active_timeout;
734 if (!opts.active_timeout) {
735 opts.active_timeout = -1;
736 } else if (opts.active_timeout < 0) {
737 VLOG_WARN("bridge %s: active timeout interval set to negative "
738 "value, using default instead (%d seconds)", br->name,
739 NF_ACTIVE_TIMEOUT_DEFAULT);
740 opts.active_timeout = -1;
743 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
744 if (opts.add_id_to_iface) {
745 if (opts.engine_id > 0x7f) {
746 VLOG_WARN("bridge %s: netflow port mangling may conflict "
747 "with another vswitch, choose an engine id less "
748 "than 128", br->name);
750 if (br->n_ports > 508) {
751 VLOG_WARN("bridge %s: netflow port mangling will conflict "
752 "with another port when more than 508 ports are "
757 opts.collectors.n = nf_cfg->n_targets;
758 opts.collectors.names = nf_cfg->targets;
759 if (ofproto_set_netflow(br->ofproto, &opts)) {
760 VLOG_ERR("bridge %s: problem setting netflow collectors",
764 ofproto_set_netflow(br->ofproto, NULL);
767 /* Set sFlow configuration on this bridge. */
768 if (br->cfg->sflow) {
769 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
770 struct ovsrec_controller **controllers;
771 struct ofproto_sflow_options oso;
772 size_t n_controllers;
775 memset(&oso, 0, sizeof oso);
777 oso.targets.n = sflow_cfg->n_targets;
778 oso.targets.names = sflow_cfg->targets;
780 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
781 if (sflow_cfg->sampling) {
782 oso.sampling_rate = *sflow_cfg->sampling;
785 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
786 if (sflow_cfg->polling) {
787 oso.polling_interval = *sflow_cfg->polling;
790 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
791 if (sflow_cfg->header) {
792 oso.header_len = *sflow_cfg->header;
795 oso.sub_id = sflow_bridge_number++;
796 oso.agent_device = sflow_cfg->agent;
798 oso.control_ip = NULL;
799 n_controllers = bridge_get_controllers(br, &controllers);
800 for (i = 0; i < n_controllers; i++) {
801 if (controllers[i]->local_ip) {
802 oso.control_ip = controllers[i]->local_ip;
806 ofproto_set_sflow(br->ofproto, &oso);
808 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
810 ofproto_set_sflow(br->ofproto, NULL);
813 /* Update the controller and related settings. It would be more
814 * straightforward to call this from bridge_reconfigure_one(), but we
815 * can't do it there for two reasons. First, and most importantly, at
816 * that point we don't know the dp_ifidx of any interfaces that have
817 * been added to the bridge (because we haven't actually added them to
818 * the datapath). Second, at that point we haven't set the datapath ID
819 * yet; when a controller is configured, resetting the datapath ID will
820 * immediately disconnect from the controller, so it's better to set
821 * the datapath ID before the controller. */
822 bridge_reconfigure_remotes(br, managers, n_managers);
824 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
825 for (i = 0; i < br->n_ports; i++) {
826 struct port *port = br->ports[i];
829 port_update_vlan_compat(port);
830 port_update_bonding(port);
832 for (j = 0; j < port->n_ifaces; j++) {
833 iface_update_qos(port->ifaces[j], port->cfg->qos);
837 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
838 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
845 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
846 const struct ovsdb_idl_column *column,
849 const struct ovsdb_datum *datum;
850 union ovsdb_atom atom;
853 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
854 atom.string = (char *) key;
855 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
856 return idx == UINT_MAX ? NULL : datum->values[idx].string;
860 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
862 return get_ovsrec_key_value(&br_cfg->header_,
863 &ovsrec_bridge_col_other_config, key);
867 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
868 struct iface **hw_addr_iface)
874 *hw_addr_iface = NULL;
876 /* Did the user request a particular MAC? */
877 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
878 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
879 if (eth_addr_is_multicast(ea)) {
880 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
881 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
882 } else if (eth_addr_is_zero(ea)) {
883 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
889 /* Otherwise choose the minimum non-local MAC address among all of the
891 memset(ea, 0xff, sizeof ea);
892 for (i = 0; i < br->n_ports; i++) {
893 struct port *port = br->ports[i];
894 uint8_t iface_ea[ETH_ADDR_LEN];
897 /* Mirror output ports don't participate. */
898 if (port->is_mirror_output_port) {
902 /* Choose the MAC address to represent the port. */
903 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
904 /* Find the interface with this Ethernet address (if any) so that
905 * we can provide the correct devname to the caller. */
907 for (j = 0; j < port->n_ifaces; j++) {
908 struct iface *candidate = port->ifaces[j];
909 uint8_t candidate_ea[ETH_ADDR_LEN];
910 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
911 && eth_addr_equals(iface_ea, candidate_ea)) {
916 /* Choose the interface whose MAC address will represent the port.
917 * The Linux kernel bonding code always chooses the MAC address of
918 * the first slave added to a bond, and the Fedora networking
919 * scripts always add slaves to a bond in alphabetical order, so
920 * for compatibility we choose the interface with the name that is
921 * first in alphabetical order. */
922 iface = port->ifaces[0];
923 for (j = 1; j < port->n_ifaces; j++) {
924 struct iface *candidate = port->ifaces[j];
925 if (strcmp(candidate->name, iface->name) < 0) {
930 /* The local port doesn't count (since we're trying to choose its
931 * MAC address anyway). */
932 if (iface->dp_ifidx == ODPP_LOCAL) {
937 error = netdev_get_etheraddr(iface->netdev, iface_ea);
939 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
940 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
941 iface->name, strerror(error));
946 /* Compare against our current choice. */
947 if (!eth_addr_is_multicast(iface_ea) &&
948 !eth_addr_is_local(iface_ea) &&
949 !eth_addr_is_reserved(iface_ea) &&
950 !eth_addr_is_zero(iface_ea) &&
951 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
953 memcpy(ea, iface_ea, ETH_ADDR_LEN);
954 *hw_addr_iface = iface;
957 if (eth_addr_is_multicast(ea)) {
958 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
959 *hw_addr_iface = NULL;
960 VLOG_WARN("bridge %s: using default bridge Ethernet "
961 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
963 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
964 br->name, ETH_ADDR_ARGS(ea));
968 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
969 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
970 * an interface on 'br', then that interface must be passed in as
971 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
972 * 'hw_addr_iface' must be passed in as a null pointer. */
974 bridge_pick_datapath_id(struct bridge *br,
975 const uint8_t bridge_ea[ETH_ADDR_LEN],
976 struct iface *hw_addr_iface)
979 * The procedure for choosing a bridge MAC address will, in the most
980 * ordinary case, also choose a unique MAC that we can use as a datapath
981 * ID. In some special cases, though, multiple bridges will end up with
982 * the same MAC address. This is OK for the bridges, but it will confuse
983 * the OpenFlow controller, because each datapath needs a unique datapath
986 * Datapath IDs must be unique. It is also very desirable that they be
987 * stable from one run to the next, so that policy set on a datapath
990 const char *datapath_id;
993 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
994 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1000 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1002 * A bridge whose MAC address is taken from a VLAN network device
1003 * (that is, a network device created with vconfig(8) or similar
1004 * tool) will have the same MAC address as a bridge on the VLAN
1005 * device's physical network device.
1007 * Handle this case by hashing the physical network device MAC
1008 * along with the VLAN identifier.
1010 uint8_t buf[ETH_ADDR_LEN + 2];
1011 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1012 buf[ETH_ADDR_LEN] = vlan >> 8;
1013 buf[ETH_ADDR_LEN + 1] = vlan;
1014 return dpid_from_hash(buf, sizeof buf);
1017 * Assume that this bridge's MAC address is unique, since it
1018 * doesn't fit any of the cases we handle specially.
1023 * A purely internal bridge, that is, one that has no non-virtual
1024 * network devices on it at all, is more difficult because it has no
1025 * natural unique identifier at all.
1027 * When the host is a XenServer, we handle this case by hashing the
1028 * host's UUID with the name of the bridge. Names of bridges are
1029 * persistent across XenServer reboots, although they can be reused if
1030 * an internal network is destroyed and then a new one is later
1031 * created, so this is fairly effective.
1033 * When the host is not a XenServer, we punt by using a random MAC
1034 * address on each run.
1036 const char *host_uuid = xenserver_get_host_uuid();
1038 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1039 dpid = dpid_from_hash(combined, strlen(combined));
1045 return eth_addr_to_uint64(bridge_ea);
1049 dpid_from_hash(const void *data, size_t n)
1051 uint8_t hash[SHA1_DIGEST_SIZE];
1053 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1054 sha1_bytes(data, n, hash);
1055 eth_addr_mark_random(hash);
1056 return eth_addr_to_uint64(hash);
1060 iface_refresh_stats(struct iface *iface)
1066 static const struct iface_stat iface_stats[] = {
1067 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1068 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1069 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1070 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1071 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1072 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1073 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1074 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1075 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1076 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1077 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1078 { "collisions", offsetof(struct netdev_stats, collisions) },
1080 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1081 const struct iface_stat *s;
1083 char *keys[N_STATS];
1084 int64_t values[N_STATS];
1087 struct netdev_stats stats;
1089 /* Intentionally ignore return value, since errors will set 'stats' to
1090 * all-1s, and we will deal with that correctly below. */
1091 netdev_get_stats(iface->netdev, &stats);
1094 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1095 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1096 if (value != UINT64_MAX) {
1103 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1109 bool datapath_destroyed;
1112 /* Let each bridge do the work that it needs to do. */
1113 datapath_destroyed = false;
1114 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1115 int error = bridge_run_one(br);
1117 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1118 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1119 "forcing reconfiguration", br->name);
1120 datapath_destroyed = true;
1124 /* (Re)configure if necessary. */
1125 if (ovsdb_idl_run(idl) || datapath_destroyed) {
1126 const struct ovsrec_open_vswitch *cfg = ovsrec_open_vswitch_first(idl);
1128 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1130 bridge_configure_once(cfg);
1131 bridge_reconfigure(cfg);
1133 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1134 ovsdb_idl_txn_commit(txn);
1135 ovsdb_idl_txn_destroy(txn); /* XXX */
1137 /* We still need to reconfigure to avoid dangling pointers to
1138 * now-destroyed ovsrec structures inside bridge data. */
1139 static const struct ovsrec_open_vswitch null_cfg;
1141 bridge_reconfigure(&null_cfg);
1145 /* Refresh interface stats if necessary. */
1146 if (time_msec() >= iface_stats_timer) {
1147 struct ovsdb_idl_txn *txn;
1149 txn = ovsdb_idl_txn_create(idl);
1150 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1153 for (i = 0; i < br->n_ports; i++) {
1154 struct port *port = br->ports[i];
1157 for (j = 0; j < port->n_ifaces; j++) {
1158 struct iface *iface = port->ifaces[j];
1159 iface_refresh_stats(iface);
1163 ovsdb_idl_txn_commit(txn);
1164 ovsdb_idl_txn_destroy(txn); /* XXX */
1166 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
1175 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1176 ofproto_wait(br->ofproto);
1177 if (ofproto_has_controller(br->ofproto)) {
1181 mac_learning_wait(br->ml);
1184 ovsdb_idl_wait(idl);
1185 poll_timer_wait_until(iface_stats_timer);
1188 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1189 * configuration changes. */
1191 bridge_flush(struct bridge *br)
1193 COVERAGE_INC(bridge_flush);
1195 mac_learning_flush(br->ml);
1198 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1199 * such interface. */
1200 static struct iface *
1201 bridge_get_local_iface(struct bridge *br)
1205 for (i = 0; i < br->n_ports; i++) {
1206 struct port *port = br->ports[i];
1207 for (j = 0; j < port->n_ifaces; j++) {
1208 struct iface *iface = port->ifaces[j];
1209 if (iface->dp_ifidx == ODPP_LOCAL) {
1218 /* Bridge unixctl user interface functions. */
1220 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1221 const char *args, void *aux OVS_UNUSED)
1223 struct ds ds = DS_EMPTY_INITIALIZER;
1224 const struct bridge *br;
1225 const struct mac_entry *e;
1227 br = bridge_lookup(args);
1229 unixctl_command_reply(conn, 501, "no such bridge");
1233 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1234 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1235 if (e->port < 0 || e->port >= br->n_ports) {
1238 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1239 br->ports[e->port]->ifaces[0]->dp_ifidx,
1240 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1242 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1246 /* Bridge reconfiguration functions. */
1247 static struct bridge *
1248 bridge_create(const struct ovsrec_bridge *br_cfg)
1253 assert(!bridge_lookup(br_cfg->name));
1254 br = xzalloc(sizeof *br);
1256 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1262 dpif_flow_flush(br->dpif);
1264 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1267 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1269 dpif_delete(br->dpif);
1270 dpif_close(br->dpif);
1275 br->name = xstrdup(br_cfg->name);
1277 br->ml = mac_learning_create();
1278 eth_addr_nicira_random(br->default_ea);
1280 port_array_init(&br->ifaces);
1282 shash_init(&br->port_by_name);
1283 shash_init(&br->iface_by_name);
1287 list_push_back(&all_bridges, &br->node);
1289 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1295 bridge_destroy(struct bridge *br)
1300 while (br->n_ports > 0) {
1301 port_destroy(br->ports[br->n_ports - 1]);
1303 list_remove(&br->node);
1304 error = dpif_delete(br->dpif);
1305 if (error && error != ENOENT) {
1306 VLOG_ERR("failed to delete %s: %s",
1307 dpif_name(br->dpif), strerror(error));
1309 dpif_close(br->dpif);
1310 ofproto_destroy(br->ofproto);
1311 mac_learning_destroy(br->ml);
1312 port_array_destroy(&br->ifaces);
1313 shash_destroy(&br->port_by_name);
1314 shash_destroy(&br->iface_by_name);
1321 static struct bridge *
1322 bridge_lookup(const char *name)
1326 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1327 if (!strcmp(br->name, name)) {
1334 /* Handle requests for a listing of all flows known by the OpenFlow
1335 * stack, including those normally hidden. */
1337 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1338 const char *args, void *aux OVS_UNUSED)
1343 br = bridge_lookup(args);
1345 unixctl_command_reply(conn, 501, "Unknown bridge");
1350 ofproto_get_all_flows(br->ofproto, &results);
1352 unixctl_command_reply(conn, 200, ds_cstr(&results));
1353 ds_destroy(&results);
1356 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1357 * connections and reconnect. If BRIDGE is not specified, then all bridges
1358 * drop their controller connections and reconnect. */
1360 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1361 const char *args, void *aux OVS_UNUSED)
1364 if (args[0] != '\0') {
1365 br = bridge_lookup(args);
1367 unixctl_command_reply(conn, 501, "Unknown bridge");
1370 ofproto_reconnect_controllers(br->ofproto);
1372 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1373 ofproto_reconnect_controllers(br->ofproto);
1376 unixctl_command_reply(conn, 200, NULL);
1380 bridge_run_one(struct bridge *br)
1384 error = ofproto_run1(br->ofproto);
1389 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1392 error = ofproto_run2(br->ofproto, br->flush);
1399 bridge_get_controllers(const struct bridge *br,
1400 struct ovsrec_controller ***controllersp)
1402 struct ovsrec_controller **controllers;
1403 size_t n_controllers;
1405 controllers = br->cfg->controller;
1406 n_controllers = br->cfg->n_controller;
1408 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1414 *controllersp = controllers;
1416 return n_controllers;
1420 bridge_reconfigure_one(struct bridge *br)
1422 struct shash old_ports, new_ports;
1423 struct svec listeners, old_listeners;
1424 struct svec snoops, old_snoops;
1425 struct shash_node *node;
1428 /* Collect old ports. */
1429 shash_init(&old_ports);
1430 for (i = 0; i < br->n_ports; i++) {
1431 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1434 /* Collect new ports. */
1435 shash_init(&new_ports);
1436 for (i = 0; i < br->cfg->n_ports; i++) {
1437 const char *name = br->cfg->ports[i]->name;
1438 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1439 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1444 /* If we have a controller, then we need a local port. Complain if the
1445 * user didn't specify one.
1447 * XXX perhaps we should synthesize a port ourselves in this case. */
1448 if (bridge_get_controllers(br, NULL)) {
1449 char local_name[IF_NAMESIZE];
1452 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1453 local_name, sizeof local_name);
1454 if (!error && !shash_find(&new_ports, local_name)) {
1455 VLOG_WARN("bridge %s: controller specified but no local port "
1456 "(port named %s) defined",
1457 br->name, local_name);
1461 /* Get rid of deleted ports.
1462 * Get rid of deleted interfaces on ports that still exist. */
1463 SHASH_FOR_EACH (node, &old_ports) {
1464 struct port *port = node->data;
1465 const struct ovsrec_port *port_cfg;
1467 port_cfg = shash_find_data(&new_ports, node->name);
1471 port_del_ifaces(port, port_cfg);
1475 /* Create new ports.
1476 * Add new interfaces to existing ports.
1477 * Reconfigure existing ports. */
1478 SHASH_FOR_EACH (node, &new_ports) {
1479 struct port *port = shash_find_data(&old_ports, node->name);
1481 port = port_create(br, node->name);
1484 port_reconfigure(port, node->data);
1485 if (!port->n_ifaces) {
1486 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1487 br->name, port->name);
1491 shash_destroy(&old_ports);
1492 shash_destroy(&new_ports);
1494 /* Delete all flows if we're switching from connected to standalone or vice
1495 * versa. (XXX Should we delete all flows if we are switching from one
1496 * controller to another?) */
1498 /* Configure OpenFlow management listener. */
1499 svec_init(&listeners);
1500 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1501 ovs_rundir, br->name));
1502 svec_init(&old_listeners);
1503 ofproto_get_listeners(br->ofproto, &old_listeners);
1504 if (!svec_equal(&listeners, &old_listeners)) {
1505 ofproto_set_listeners(br->ofproto, &listeners);
1507 svec_destroy(&listeners);
1508 svec_destroy(&old_listeners);
1510 /* Configure OpenFlow controller connection snooping. */
1512 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1513 ovs_rundir, br->name));
1514 svec_init(&old_snoops);
1515 ofproto_get_snoops(br->ofproto, &old_snoops);
1516 if (!svec_equal(&snoops, &old_snoops)) {
1517 ofproto_set_snoops(br->ofproto, &snoops);
1519 svec_destroy(&snoops);
1520 svec_destroy(&old_snoops);
1522 mirror_reconfigure(br);
1526 bridge_reconfigure_remotes(struct bridge *br,
1527 const struct sockaddr_in *managers,
1530 struct ovsrec_controller **controllers;
1531 size_t n_controllers;
1533 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1535 n_controllers = bridge_get_controllers(br, &controllers);
1536 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1537 ofproto_flush_flows(br->ofproto);
1540 if (!n_controllers) {
1541 union ofp_action action;
1544 /* Clear out controllers. */
1545 ofproto_set_controllers(br->ofproto, NULL, 0);
1547 /* Set up a flow that matches every packet and directs them to
1548 * OFPP_NORMAL (which goes to us). */
1549 memset(&action, 0, sizeof action);
1550 action.type = htons(OFPAT_OUTPUT);
1551 action.output.len = htons(sizeof action);
1552 action.output.port = htons(OFPP_NORMAL);
1553 memset(&flow, 0, sizeof flow);
1554 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1556 struct ofproto_controller *ocs;
1559 ocs = xmalloc(n_controllers * sizeof *ocs);
1560 for (i = 0; i < n_controllers; i++) {
1561 struct ovsrec_controller *c = controllers[i];
1562 struct ofproto_controller *oc = &ocs[i];
1564 if (strcmp(c->target, "discover")) {
1565 struct iface *local_iface;
1568 local_iface = bridge_get_local_iface(br);
1569 if (local_iface && c->local_ip
1570 && inet_aton(c->local_ip, &ip)) {
1571 struct netdev *netdev = local_iface->netdev;
1572 struct in_addr mask, gateway;
1574 if (!c->local_netmask
1575 || !inet_aton(c->local_netmask, &mask)) {
1578 if (!c->local_gateway
1579 || !inet_aton(c->local_gateway, &gateway)) {
1583 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1585 mask.s_addr = guess_netmask(ip.s_addr);
1587 if (!netdev_set_in4(netdev, ip, mask)) {
1588 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1590 br->name, IP_ARGS(&ip.s_addr),
1591 IP_ARGS(&mask.s_addr));
1594 if (gateway.s_addr) {
1595 if (!netdev_add_router(netdev, gateway)) {
1596 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1597 br->name, IP_ARGS(&gateway.s_addr));
1603 oc->target = c->target;
1604 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1605 oc->probe_interval = (c->inactivity_probe
1606 ? *c->inactivity_probe / 1000 : 5);
1607 oc->fail = (!c->fail_mode
1608 || !strcmp(c->fail_mode, "standalone")
1609 || !strcmp(c->fail_mode, "open")
1610 ? OFPROTO_FAIL_STANDALONE
1611 : OFPROTO_FAIL_SECURE);
1612 oc->band = (!c->connection_mode
1613 || !strcmp(c->connection_mode, "in-band")
1615 : OFPROTO_OUT_OF_BAND);
1616 oc->accept_re = c->discover_accept_regex;
1617 oc->update_resolv_conf = c->discover_update_resolv_conf;
1618 oc->rate_limit = (c->controller_rate_limit
1619 ? *c->controller_rate_limit : 0);
1620 oc->burst_limit = (c->controller_burst_limit
1621 ? *c->controller_burst_limit : 0);
1623 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1629 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1634 for (i = 0; i < br->n_ports; i++) {
1635 struct port *port = br->ports[i];
1636 for (j = 0; j < port->n_ifaces; j++) {
1637 struct iface *iface = port->ifaces[j];
1638 shash_add_once(ifaces, iface->name, iface);
1640 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1641 shash_add_once(ifaces, port->name, NULL);
1646 /* For robustness, in case the administrator moves around datapath ports behind
1647 * our back, we re-check all the datapath port numbers here.
1649 * This function will set the 'dp_ifidx' members of interfaces that have
1650 * disappeared to -1, so only call this function from a context where those
1651 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1652 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1653 * datapath, which doesn't support UINT16_MAX+1 ports. */
1655 bridge_fetch_dp_ifaces(struct bridge *br)
1657 struct odp_port *dpif_ports;
1658 size_t n_dpif_ports;
1661 /* Reset all interface numbers. */
1662 for (i = 0; i < br->n_ports; i++) {
1663 struct port *port = br->ports[i];
1664 for (j = 0; j < port->n_ifaces; j++) {
1665 struct iface *iface = port->ifaces[j];
1666 iface->dp_ifidx = -1;
1669 port_array_clear(&br->ifaces);
1671 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1672 for (i = 0; i < n_dpif_ports; i++) {
1673 struct odp_port *p = &dpif_ports[i];
1674 struct iface *iface = iface_lookup(br, p->devname);
1676 if (iface->dp_ifidx >= 0) {
1677 VLOG_WARN("%s reported interface %s twice",
1678 dpif_name(br->dpif), p->devname);
1679 } else if (iface_from_dp_ifidx(br, p->port)) {
1680 VLOG_WARN("%s reported interface %"PRIu16" twice",
1681 dpif_name(br->dpif), p->port);
1683 port_array_set(&br->ifaces, p->port, iface);
1684 iface->dp_ifidx = p->port;
1688 int64_t ofport = (iface->dp_ifidx >= 0
1689 ? odp_port_to_ofp_port(iface->dp_ifidx)
1691 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1698 /* Bridge packet processing functions. */
1701 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1703 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1706 static struct bond_entry *
1707 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1709 return &port->bond_hash[bond_hash(mac)];
1713 bond_choose_iface(const struct port *port)
1715 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1716 size_t i, best_down_slave = -1;
1717 long long next_delay_expiration = LLONG_MAX;
1719 for (i = 0; i < port->n_ifaces; i++) {
1720 struct iface *iface = port->ifaces[i];
1722 if (iface->enabled) {
1724 } else if (iface->delay_expires < next_delay_expiration) {
1725 best_down_slave = i;
1726 next_delay_expiration = iface->delay_expires;
1730 if (best_down_slave != -1) {
1731 struct iface *iface = port->ifaces[best_down_slave];
1733 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1734 "since no other interface is up", iface->name,
1735 iface->delay_expires - time_msec());
1736 bond_enable_slave(iface, true);
1739 return best_down_slave;
1743 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1744 uint16_t *dp_ifidx, tag_type *tags)
1746 struct iface *iface;
1748 assert(port->n_ifaces);
1749 if (port->n_ifaces == 1) {
1750 iface = port->ifaces[0];
1752 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1753 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1754 || !port->ifaces[e->iface_idx]->enabled) {
1755 /* XXX select interface properly. The current interface selection
1756 * is only good for testing the rebalancing code. */
1757 e->iface_idx = bond_choose_iface(port);
1758 if (e->iface_idx < 0) {
1759 *tags |= port->no_ifaces_tag;
1762 e->iface_tag = tag_create_random();
1763 ((struct port *) port)->bond_compat_is_stale = true;
1765 *tags |= e->iface_tag;
1766 iface = port->ifaces[e->iface_idx];
1768 *dp_ifidx = iface->dp_ifidx;
1769 *tags |= iface->tag; /* Currently only used for bonding. */
1774 bond_link_status_update(struct iface *iface, bool carrier)
1776 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1777 struct port *port = iface->port;
1779 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1780 /* Nothing to do. */
1783 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1784 iface->name, carrier ? "detected" : "dropped");
1785 if (carrier == iface->enabled) {
1786 iface->delay_expires = LLONG_MAX;
1787 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1788 iface->name, carrier ? "disabled" : "enabled");
1789 } else if (carrier && port->active_iface < 0) {
1790 bond_enable_slave(iface, true);
1791 if (port->updelay) {
1792 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1793 "other interface is up", iface->name, port->updelay);
1796 int delay = carrier ? port->updelay : port->downdelay;
1797 iface->delay_expires = time_msec() + delay;
1800 "interface %s: will be %s if it stays %s for %d ms",
1802 carrier ? "enabled" : "disabled",
1803 carrier ? "up" : "down",
1810 bond_choose_active_iface(struct port *port)
1812 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1814 port->active_iface = bond_choose_iface(port);
1815 port->active_iface_tag = tag_create_random();
1816 if (port->active_iface >= 0) {
1817 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1818 port->name, port->ifaces[port->active_iface]->name);
1820 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1826 bond_enable_slave(struct iface *iface, bool enable)
1828 struct port *port = iface->port;
1829 struct bridge *br = port->bridge;
1831 /* This acts as a recursion check. If the act of disabling a slave
1832 * causes a different slave to be enabled, the flag will allow us to
1833 * skip redundant work when we reenter this function. It must be
1834 * cleared on exit to keep things safe with multiple bonds. */
1835 static bool moving_active_iface = false;
1837 iface->delay_expires = LLONG_MAX;
1838 if (enable == iface->enabled) {
1842 iface->enabled = enable;
1843 if (!iface->enabled) {
1844 VLOG_WARN("interface %s: disabled", iface->name);
1845 ofproto_revalidate(br->ofproto, iface->tag);
1846 if (iface->port_ifidx == port->active_iface) {
1847 ofproto_revalidate(br->ofproto,
1848 port->active_iface_tag);
1850 /* Disabling a slave can lead to another slave being immediately
1851 * enabled if there will be no active slaves but one is waiting
1852 * on an updelay. In this case we do not need to run most of the
1853 * code for the newly enabled slave since there was no period
1854 * without an active slave and it is redundant with the disabling
1856 moving_active_iface = true;
1857 bond_choose_active_iface(port);
1859 bond_send_learning_packets(port);
1861 VLOG_WARN("interface %s: enabled", iface->name);
1862 if (port->active_iface < 0 && !moving_active_iface) {
1863 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1864 bond_choose_active_iface(port);
1865 bond_send_learning_packets(port);
1867 iface->tag = tag_create_random();
1870 moving_active_iface = false;
1871 port->bond_compat_is_stale = true;
1874 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1875 * bond interface. */
1877 bond_update_fake_iface_stats(struct port *port)
1879 struct netdev_stats bond_stats;
1880 struct netdev *bond_dev;
1883 memset(&bond_stats, 0, sizeof bond_stats);
1885 for (i = 0; i < port->n_ifaces; i++) {
1886 struct netdev_stats slave_stats;
1888 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1889 /* XXX: We swap the stats here because they are swapped back when
1890 * reported by the internal device. The reason for this is
1891 * internal devices normally represent packets going into the system
1892 * but when used as fake bond device they represent packets leaving
1893 * the system. We really should do this in the internal device
1894 * itself because changing it here reverses the counts from the
1895 * perspective of the switch. However, the internal device doesn't
1896 * know what type of device it represents so we have to do it here
1898 bond_stats.tx_packets += slave_stats.rx_packets;
1899 bond_stats.tx_bytes += slave_stats.rx_bytes;
1900 bond_stats.rx_packets += slave_stats.tx_packets;
1901 bond_stats.rx_bytes += slave_stats.tx_bytes;
1905 if (!netdev_open_default(port->name, &bond_dev)) {
1906 netdev_set_stats(bond_dev, &bond_stats);
1907 netdev_close(bond_dev);
1912 bond_run(struct bridge *br)
1916 for (i = 0; i < br->n_ports; i++) {
1917 struct port *port = br->ports[i];
1919 if (port->n_ifaces >= 2) {
1920 for (j = 0; j < port->n_ifaces; j++) {
1921 struct iface *iface = port->ifaces[j];
1922 if (time_msec() >= iface->delay_expires) {
1923 bond_enable_slave(iface, !iface->enabled);
1927 if (port->bond_fake_iface
1928 && time_msec() >= port->bond_next_fake_iface_update) {
1929 bond_update_fake_iface_stats(port);
1930 port->bond_next_fake_iface_update = time_msec() + 1000;
1934 if (port->bond_compat_is_stale) {
1935 port->bond_compat_is_stale = false;
1936 port_update_bond_compat(port);
1942 bond_wait(struct bridge *br)
1946 for (i = 0; i < br->n_ports; i++) {
1947 struct port *port = br->ports[i];
1948 if (port->n_ifaces < 2) {
1951 for (j = 0; j < port->n_ifaces; j++) {
1952 struct iface *iface = port->ifaces[j];
1953 if (iface->delay_expires != LLONG_MAX) {
1954 poll_timer_wait_until(iface->delay_expires);
1957 if (port->bond_fake_iface) {
1958 poll_timer_wait_until(port->bond_next_fake_iface_update);
1964 set_dst(struct dst *p, const flow_t *flow,
1965 const struct port *in_port, const struct port *out_port,
1968 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1969 : in_port->vlan >= 0 ? in_port->vlan
1970 : ntohs(flow->dl_vlan));
1971 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1975 swap_dst(struct dst *p, struct dst *q)
1977 struct dst tmp = *p;
1982 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1983 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1984 * that we push to the datapath. We could in fact fully sort the array by
1985 * vlan, but in most cases there are at most two different vlan tags so that's
1986 * possibly overkill.) */
1988 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1990 struct dst *first = dsts;
1991 struct dst *last = dsts + n_dsts;
1993 while (first != last) {
1995 * - All dsts < first have vlan == 'vlan'.
1996 * - All dsts >= last have vlan != 'vlan'.
1997 * - first < last. */
1998 while (first->vlan == vlan) {
1999 if (++first == last) {
2004 /* Same invariants, plus one additional:
2005 * - first->vlan != vlan.
2007 while (last[-1].vlan != vlan) {
2008 if (--last == first) {
2013 /* Same invariants, plus one additional:
2014 * - last[-1].vlan == vlan.*/
2015 swap_dst(first++, --last);
2020 mirror_mask_ffs(mirror_mask_t mask)
2022 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2027 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2028 const struct dst *test)
2031 for (i = 0; i < n_dsts; i++) {
2032 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2040 port_trunks_vlan(const struct port *port, uint16_t vlan)
2042 return (port->vlan < 0
2043 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2047 port_includes_vlan(const struct port *port, uint16_t vlan)
2049 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2053 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2054 const struct port *in_port, const struct port *out_port,
2055 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2057 mirror_mask_t mirrors = in_port->src_mirrors;
2058 struct dst *dst = dsts;
2061 if (out_port == FLOOD_PORT) {
2062 /* XXX use ODP_FLOOD if no vlans or bonding. */
2063 /* XXX even better, define each VLAN as a datapath port group */
2064 for (i = 0; i < br->n_ports; i++) {
2065 struct port *port = br->ports[i];
2066 if (port != in_port && port_includes_vlan(port, vlan)
2067 && !port->is_mirror_output_port
2068 && set_dst(dst, flow, in_port, port, tags)) {
2069 mirrors |= port->dst_mirrors;
2073 *nf_output_iface = NF_OUT_FLOOD;
2074 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2075 *nf_output_iface = dst->dp_ifidx;
2076 mirrors |= out_port->dst_mirrors;
2081 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2082 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2084 if (set_dst(dst, flow, in_port, m->out_port, tags)
2085 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2089 for (i = 0; i < br->n_ports; i++) {
2090 struct port *port = br->ports[i];
2091 if (port_includes_vlan(port, m->out_vlan)
2092 && set_dst(dst, flow, in_port, port, tags))
2096 if (port->vlan < 0) {
2097 dst->vlan = m->out_vlan;
2099 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2103 /* Use the vlan tag on the original flow instead of
2104 * the one passed in the vlan parameter. This ensures
2105 * that we compare the vlan from before any implicit
2106 * tagging tags place. This is necessary because
2107 * dst->vlan is the final vlan, after removing implicit
2109 flow_vlan = ntohs(flow->dl_vlan);
2110 if (flow_vlan == 0) {
2111 flow_vlan = OFP_VLAN_NONE;
2113 if (port == in_port && dst->vlan == flow_vlan) {
2114 /* Don't send out input port on same VLAN. */
2122 mirrors &= mirrors - 1;
2125 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2129 static void OVS_UNUSED
2130 print_dsts(const struct dst *dsts, size_t n)
2132 for (; n--; dsts++) {
2133 printf(">p%"PRIu16, dsts->dp_ifidx);
2134 if (dsts->vlan != OFP_VLAN_NONE) {
2135 printf("v%"PRIu16, dsts->vlan);
2141 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2142 const struct port *in_port, const struct port *out_port,
2143 tag_type *tags, struct odp_actions *actions,
2144 uint16_t *nf_output_iface)
2146 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2148 const struct dst *p;
2151 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2154 cur_vlan = ntohs(flow->dl_vlan);
2155 for (p = dsts; p < &dsts[n_dsts]; p++) {
2156 union odp_action *a;
2157 if (p->vlan != cur_vlan) {
2158 if (p->vlan == OFP_VLAN_NONE) {
2159 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2161 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2162 a->vlan_vid.vlan_vid = htons(p->vlan);
2166 a = odp_actions_add(actions, ODPAT_OUTPUT);
2167 a->output.port = p->dp_ifidx;
2171 /* Returns the effective vlan of a packet, taking into account both the
2172 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2173 * the packet is untagged and -1 indicates it has an invalid header and
2174 * should be dropped. */
2175 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2176 struct port *in_port, bool have_packet)
2178 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2179 * belongs to VLAN 0, so we should treat both cases identically. (In the
2180 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2181 * presumably to allow a priority to be specified. In the latter case, the
2182 * packet does not have any 802.1Q header.) */
2183 int vlan = ntohs(flow->dl_vlan);
2184 if (vlan == OFP_VLAN_NONE) {
2187 if (in_port->vlan >= 0) {
2189 /* XXX support double tagging? */
2191 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2192 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2193 "packet received on port %s configured with "
2194 "implicit VLAN %"PRIu16,
2195 br->name, ntohs(flow->dl_vlan),
2196 in_port->name, in_port->vlan);
2200 vlan = in_port->vlan;
2202 if (!port_includes_vlan(in_port, vlan)) {
2204 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2205 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2206 "packet received on port %s not configured for "
2208 br->name, vlan, in_port->name, vlan);
2217 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2218 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2219 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2221 is_gratuitous_arp(const flow_t *flow)
2223 return (flow->dl_type == htons(ETH_TYPE_ARP)
2224 && eth_addr_is_broadcast(flow->dl_dst)
2225 && (flow->nw_proto == ARP_OP_REPLY
2226 || (flow->nw_proto == ARP_OP_REQUEST
2227 && flow->nw_src == flow->nw_dst)));
2231 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2232 struct port *in_port)
2234 enum grat_arp_lock_type lock_type;
2237 /* We don't want to learn from gratuitous ARP packets that are reflected
2238 * back over bond slaves so we lock the learning table. */
2239 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2240 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2241 GRAT_ARP_LOCK_CHECK;
2243 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2246 /* The log messages here could actually be useful in debugging,
2247 * so keep the rate limit relatively high. */
2248 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2250 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2251 "on port %s in VLAN %d",
2252 br->name, ETH_ADDR_ARGS(flow->dl_src),
2253 in_port->name, vlan);
2254 ofproto_revalidate(br->ofproto, rev_tag);
2258 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2259 * dropped. Returns true if they may be forwarded, false if they should be
2262 * If 'have_packet' is true, it indicates that the caller is processing a
2263 * received packet. If 'have_packet' is false, then the caller is just
2264 * revalidating an existing flow because configuration has changed. Either
2265 * way, 'have_packet' only affects logging (there is no point in logging errors
2266 * during revalidation).
2268 * Sets '*in_portp' to the input port. This will be a null pointer if
2269 * flow->in_port does not designate a known input port (in which case
2270 * is_admissible() returns false).
2272 * When returning true, sets '*vlanp' to the effective VLAN of the input
2273 * packet, as returned by flow_get_vlan().
2275 * May also add tags to '*tags', although the current implementation only does
2276 * so in one special case.
2279 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2280 tag_type *tags, int *vlanp, struct port **in_portp)
2282 struct iface *in_iface;
2283 struct port *in_port;
2286 /* Find the interface and port structure for the received packet. */
2287 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2289 /* No interface? Something fishy... */
2291 /* Odd. A few possible reasons here:
2293 * - We deleted an interface but there are still a few packets
2294 * queued up from it.
2296 * - Someone externally added an interface (e.g. with "ovs-dpctl
2297 * add-if") that we don't know about.
2299 * - Packet arrived on the local port but the local port is not
2300 * one of our bridge ports.
2302 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2304 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2305 "interface %"PRIu16, br->name, flow->in_port);
2311 *in_portp = in_port = in_iface->port;
2312 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2317 /* Drop frames for reserved multicast addresses. */
2318 if (eth_addr_is_reserved(flow->dl_dst)) {
2322 /* Drop frames on ports reserved for mirroring. */
2323 if (in_port->is_mirror_output_port) {
2325 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2326 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2327 "%s, which is reserved exclusively for mirroring",
2328 br->name, in_port->name);
2333 /* Packets received on bonds need special attention to avoid duplicates. */
2334 if (in_port->n_ifaces > 1) {
2336 bool is_grat_arp_locked;
2338 if (eth_addr_is_multicast(flow->dl_dst)) {
2339 *tags |= in_port->active_iface_tag;
2340 if (in_port->active_iface != in_iface->port_ifidx) {
2341 /* Drop all multicast packets on inactive slaves. */
2346 /* Drop all packets for which we have learned a different input
2347 * port, because we probably sent the packet on one slave and got
2348 * it back on the other. Gratuitous ARP packets are an exception
2349 * to this rule: the host has moved to another switch. The exception
2350 * to the exception is if we locked the learning table to avoid
2351 * reflections on bond slaves. If this is the case, just drop the
2353 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2354 &is_grat_arp_locked);
2355 if (src_idx != -1 && src_idx != in_port->port_idx &&
2356 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2364 /* If the composed actions may be applied to any packet in the given 'flow',
2365 * returns true. Otherwise, the actions should only be applied to 'packet', or
2366 * not at all, if 'packet' was NULL. */
2368 process_flow(struct bridge *br, const flow_t *flow,
2369 const struct ofpbuf *packet, struct odp_actions *actions,
2370 tag_type *tags, uint16_t *nf_output_iface)
2372 struct port *in_port;
2373 struct port *out_port;
2377 /* Check whether we should drop packets in this flow. */
2378 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2383 /* Learn source MAC (but don't try to learn from revalidation). */
2385 update_learning_table(br, flow, vlan, in_port);
2388 /* Determine output port. */
2389 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2391 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2392 out_port = br->ports[out_port_idx];
2393 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2394 /* If we are revalidating but don't have a learning entry then
2395 * eject the flow. Installing a flow that floods packets opens
2396 * up a window of time where we could learn from a packet reflected
2397 * on a bond and blackhole packets before the learning table is
2398 * updated to reflect the correct port. */
2401 out_port = FLOOD_PORT;
2404 /* Don't send packets out their input ports. */
2405 if (in_port == out_port) {
2411 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2418 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2421 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2422 const struct ofp_phy_port *opp,
2425 struct bridge *br = br_;
2426 struct iface *iface;
2429 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2435 if (reason == OFPPR_DELETE) {
2436 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2437 br->name, iface->name);
2438 iface_destroy(iface);
2439 if (!port->n_ifaces) {
2440 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2441 br->name, port->name);
2447 if (port->n_ifaces > 1) {
2448 bool up = !(opp->state & OFPPS_LINK_DOWN);
2449 bond_link_status_update(iface, up);
2450 port_update_bond_compat(port);
2456 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2457 struct odp_actions *actions, tag_type *tags,
2458 uint16_t *nf_output_iface, void *br_)
2460 struct bridge *br = br_;
2462 COVERAGE_INC(bridge_process_flow);
2463 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2467 bridge_account_flow_ofhook_cb(const flow_t *flow,
2468 const union odp_action *actions,
2469 size_t n_actions, unsigned long long int n_bytes,
2472 struct bridge *br = br_;
2473 const union odp_action *a;
2474 struct port *in_port;
2478 /* Feed information from the active flows back into the learning table
2479 * to ensure that table is always in sync with what is actually flowing
2480 * through the datapath. */
2481 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2482 update_learning_table(br, flow, vlan, in_port);
2485 if (!br->has_bonded_ports) {
2489 for (a = actions; a < &actions[n_actions]; a++) {
2490 if (a->type == ODPAT_OUTPUT) {
2491 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2492 if (out_port && out_port->n_ifaces >= 2) {
2493 struct bond_entry *e = lookup_bond_entry(out_port,
2495 e->tx_bytes += n_bytes;
2502 bridge_account_checkpoint_ofhook_cb(void *br_)
2504 struct bridge *br = br_;
2508 if (!br->has_bonded_ports) {
2513 for (i = 0; i < br->n_ports; i++) {
2514 struct port *port = br->ports[i];
2515 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2516 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2517 bond_rebalance_port(port);
2522 static struct ofhooks bridge_ofhooks = {
2523 bridge_port_changed_ofhook_cb,
2524 bridge_normal_ofhook_cb,
2525 bridge_account_flow_ofhook_cb,
2526 bridge_account_checkpoint_ofhook_cb,
2529 /* Bonding functions. */
2531 /* Statistics for a single interface on a bonded port, used for load-based
2532 * bond rebalancing. */
2533 struct slave_balance {
2534 struct iface *iface; /* The interface. */
2535 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2537 /* All the "bond_entry"s that are assigned to this interface, in order of
2538 * increasing tx_bytes. */
2539 struct bond_entry **hashes;
2543 /* Sorts pointers to pointers to bond_entries in ascending order by the
2544 * interface to which they are assigned, and within a single interface in
2545 * ascending order of bytes transmitted. */
2547 compare_bond_entries(const void *a_, const void *b_)
2549 const struct bond_entry *const *ap = a_;
2550 const struct bond_entry *const *bp = b_;
2551 const struct bond_entry *a = *ap;
2552 const struct bond_entry *b = *bp;
2553 if (a->iface_idx != b->iface_idx) {
2554 return a->iface_idx > b->iface_idx ? 1 : -1;
2555 } else if (a->tx_bytes != b->tx_bytes) {
2556 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2562 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2563 * *descending* order by number of bytes transmitted. */
2565 compare_slave_balance(const void *a_, const void *b_)
2567 const struct slave_balance *a = a_;
2568 const struct slave_balance *b = b_;
2569 if (a->iface->enabled != b->iface->enabled) {
2570 return a->iface->enabled ? -1 : 1;
2571 } else if (a->tx_bytes != b->tx_bytes) {
2572 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2579 swap_bals(struct slave_balance *a, struct slave_balance *b)
2581 struct slave_balance tmp = *a;
2586 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2587 * given that 'p' (and only 'p') might be in the wrong location.
2589 * This function invalidates 'p', since it might now be in a different memory
2592 resort_bals(struct slave_balance *p,
2593 struct slave_balance bals[], size_t n_bals)
2596 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2597 swap_bals(p, p - 1);
2599 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2600 swap_bals(p, p + 1);
2606 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2608 if (VLOG_IS_DBG_ENABLED()) {
2609 struct ds ds = DS_EMPTY_INITIALIZER;
2610 const struct slave_balance *b;
2612 for (b = bals; b < bals + n_bals; b++) {
2616 ds_put_char(&ds, ',');
2618 ds_put_format(&ds, " %s %"PRIu64"kB",
2619 b->iface->name, b->tx_bytes / 1024);
2621 if (!b->iface->enabled) {
2622 ds_put_cstr(&ds, " (disabled)");
2624 if (b->n_hashes > 0) {
2625 ds_put_cstr(&ds, " (");
2626 for (i = 0; i < b->n_hashes; i++) {
2627 const struct bond_entry *e = b->hashes[i];
2629 ds_put_cstr(&ds, " + ");
2631 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2632 e - port->bond_hash, e->tx_bytes / 1024);
2634 ds_put_cstr(&ds, ")");
2637 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2642 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2644 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2647 struct bond_entry *hash = from->hashes[hash_idx];
2648 struct port *port = from->iface->port;
2649 uint64_t delta = hash->tx_bytes;
2651 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2652 "from %s to %s (now carrying %"PRIu64"kB and "
2653 "%"PRIu64"kB load, respectively)",
2654 port->name, delta / 1024, hash - port->bond_hash,
2655 from->iface->name, to->iface->name,
2656 (from->tx_bytes - delta) / 1024,
2657 (to->tx_bytes + delta) / 1024);
2659 /* Delete element from from->hashes.
2661 * We don't bother to add the element to to->hashes because not only would
2662 * it require more work, the only purpose it would be to allow that hash to
2663 * be migrated to another slave in this rebalancing run, and there is no
2664 * point in doing that. */
2665 if (hash_idx == 0) {
2668 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2669 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2673 /* Shift load away from 'from' to 'to'. */
2674 from->tx_bytes -= delta;
2675 to->tx_bytes += delta;
2677 /* Arrange for flows to be revalidated. */
2678 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2679 hash->iface_idx = to->iface->port_ifidx;
2680 hash->iface_tag = tag_create_random();
2684 bond_rebalance_port(struct port *port)
2686 struct slave_balance bals[DP_MAX_PORTS];
2688 struct bond_entry *hashes[BOND_MASK + 1];
2689 struct slave_balance *b, *from, *to;
2690 struct bond_entry *e;
2693 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2694 * descending order of tx_bytes, so that bals[0] represents the most
2695 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2698 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2699 * array for each slave_balance structure, we sort our local array of
2700 * hashes in order by slave, so that all of the hashes for a given slave
2701 * become contiguous in memory, and then we point each 'hashes' members of
2702 * a slave_balance structure to the start of a contiguous group. */
2703 n_bals = port->n_ifaces;
2704 for (b = bals; b < &bals[n_bals]; b++) {
2705 b->iface = port->ifaces[b - bals];
2710 for (i = 0; i <= BOND_MASK; i++) {
2711 hashes[i] = &port->bond_hash[i];
2713 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2714 for (i = 0; i <= BOND_MASK; i++) {
2716 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2717 b = &bals[e->iface_idx];
2718 b->tx_bytes += e->tx_bytes;
2720 b->hashes = &hashes[i];
2725 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2726 log_bals(bals, n_bals, port);
2728 /* Discard slaves that aren't enabled (which were sorted to the back of the
2729 * array earlier). */
2730 while (!bals[n_bals - 1].iface->enabled) {
2737 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2738 to = &bals[n_bals - 1];
2739 for (from = bals; from < to; ) {
2740 uint64_t overload = from->tx_bytes - to->tx_bytes;
2741 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2742 /* The extra load on 'from' (and all less-loaded slaves), compared
2743 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2744 * it is less than ~1Mbps. No point in rebalancing. */
2746 } else if (from->n_hashes == 1) {
2747 /* 'from' only carries a single MAC hash, so we can't shift any
2748 * load away from it, even though we want to. */
2751 /* 'from' is carrying significantly more load than 'to', and that
2752 * load is split across at least two different hashes. Pick a hash
2753 * to migrate to 'to' (the least-loaded slave), given that doing so
2754 * must decrease the ratio of the load on the two slaves by at
2757 * The sort order we use means that we prefer to shift away the
2758 * smallest hashes instead of the biggest ones. There is little
2759 * reason behind this decision; we could use the opposite sort
2760 * order to shift away big hashes ahead of small ones. */
2764 for (i = 0; i < from->n_hashes; i++) {
2765 double old_ratio, new_ratio;
2766 uint64_t delta = from->hashes[i]->tx_bytes;
2768 if (delta == 0 || from->tx_bytes - delta == 0) {
2769 /* Pointless move. */
2773 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2775 if (to->tx_bytes == 0) {
2776 /* Nothing on the new slave, move it. */
2780 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2781 new_ratio = (double)(from->tx_bytes - delta) /
2782 (to->tx_bytes + delta);
2784 if (new_ratio == 0) {
2785 /* Should already be covered but check to prevent division
2790 if (new_ratio < 1) {
2791 new_ratio = 1 / new_ratio;
2794 if (old_ratio - new_ratio > 0.1) {
2795 /* Would decrease the ratio, move it. */
2799 if (i < from->n_hashes) {
2800 bond_shift_load(from, to, i);
2801 port->bond_compat_is_stale = true;
2803 /* If the result of the migration changed the relative order of
2804 * 'from' and 'to' swap them back to maintain invariants. */
2805 if (order_swapped) {
2806 swap_bals(from, to);
2809 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2810 * point to different slave_balance structures. It is only
2811 * valid to do these two operations in a row at all because we
2812 * know that 'from' will not move past 'to' and vice versa. */
2813 resort_bals(from, bals, n_bals);
2814 resort_bals(to, bals, n_bals);
2821 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2822 * historical data to decay to <1% in 7 rebalancing runs. */
2823 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2829 bond_send_learning_packets(struct port *port)
2831 struct bridge *br = port->bridge;
2832 struct mac_entry *e;
2833 struct ofpbuf packet;
2834 int error, n_packets, n_errors;
2836 if (!port->n_ifaces || port->active_iface < 0) {
2840 ofpbuf_init(&packet, 128);
2841 error = n_packets = n_errors = 0;
2842 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2843 union ofp_action actions[2], *a;
2849 if (e->port == port->port_idx
2850 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2854 /* Compose actions. */
2855 memset(actions, 0, sizeof actions);
2858 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2859 a->vlan_vid.len = htons(sizeof *a);
2860 a->vlan_vid.vlan_vid = htons(e->vlan);
2863 a->output.type = htons(OFPAT_OUTPUT);
2864 a->output.len = htons(sizeof *a);
2865 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2870 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2872 flow_extract(&packet, 0, ODPP_NONE, &flow);
2873 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2880 ofpbuf_uninit(&packet);
2883 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2884 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2885 "packets, last error was: %s",
2886 port->name, n_errors, n_packets, strerror(error));
2888 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2889 port->name, n_packets);
2893 /* Bonding unixctl user interface functions. */
2896 bond_unixctl_list(struct unixctl_conn *conn,
2897 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2899 struct ds ds = DS_EMPTY_INITIALIZER;
2900 const struct bridge *br;
2902 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2904 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2907 for (i = 0; i < br->n_ports; i++) {
2908 const struct port *port = br->ports[i];
2909 if (port->n_ifaces > 1) {
2912 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2913 for (j = 0; j < port->n_ifaces; j++) {
2914 const struct iface *iface = port->ifaces[j];
2916 ds_put_cstr(&ds, ", ");
2918 ds_put_cstr(&ds, iface->name);
2920 ds_put_char(&ds, '\n');
2924 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2928 static struct port *
2929 bond_find(const char *name)
2931 const struct bridge *br;
2933 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2936 for (i = 0; i < br->n_ports; i++) {
2937 struct port *port = br->ports[i];
2938 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2947 bond_unixctl_show(struct unixctl_conn *conn,
2948 const char *args, void *aux OVS_UNUSED)
2950 struct ds ds = DS_EMPTY_INITIALIZER;
2951 const struct port *port;
2954 port = bond_find(args);
2956 unixctl_command_reply(conn, 501, "no such bond");
2960 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2961 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2962 ds_put_format(&ds, "next rebalance: %lld ms\n",
2963 port->bond_next_rebalance - time_msec());
2964 for (j = 0; j < port->n_ifaces; j++) {
2965 const struct iface *iface = port->ifaces[j];
2966 struct bond_entry *be;
2969 ds_put_format(&ds, "slave %s: %s\n",
2970 iface->name, iface->enabled ? "enabled" : "disabled");
2971 if (j == port->active_iface) {
2972 ds_put_cstr(&ds, "\tactive slave\n");
2974 if (iface->delay_expires != LLONG_MAX) {
2975 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2976 iface->enabled ? "downdelay" : "updelay",
2977 iface->delay_expires - time_msec());
2981 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2982 int hash = be - port->bond_hash;
2983 struct mac_entry *me;
2985 if (be->iface_idx != j) {
2989 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2990 hash, be->tx_bytes / 1024);
2993 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2994 &port->bridge->ml->lrus) {
2997 if (bond_hash(me->mac) == hash
2998 && me->port != port->port_idx
2999 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3000 && dp_ifidx == iface->dp_ifidx)
3002 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3003 ETH_ADDR_ARGS(me->mac));
3008 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3013 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3014 void *aux OVS_UNUSED)
3016 char *args = (char *) args_;
3017 char *save_ptr = NULL;
3018 char *bond_s, *hash_s, *slave_s;
3019 uint8_t mac[ETH_ADDR_LEN];
3021 struct iface *iface;
3022 struct bond_entry *entry;
3025 bond_s = strtok_r(args, " ", &save_ptr);
3026 hash_s = strtok_r(NULL, " ", &save_ptr);
3027 slave_s = strtok_r(NULL, " ", &save_ptr);
3029 unixctl_command_reply(conn, 501,
3030 "usage: bond/migrate BOND HASH SLAVE");
3034 port = bond_find(bond_s);
3036 unixctl_command_reply(conn, 501, "no such bond");
3040 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3041 == ETH_ADDR_SCAN_COUNT) {
3042 hash = bond_hash(mac);
3043 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3044 hash = atoi(hash_s) & BOND_MASK;
3046 unixctl_command_reply(conn, 501, "bad hash");
3050 iface = port_lookup_iface(port, slave_s);
3052 unixctl_command_reply(conn, 501, "no such slave");
3056 if (!iface->enabled) {
3057 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3061 entry = &port->bond_hash[hash];
3062 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3063 entry->iface_idx = iface->port_ifidx;
3064 entry->iface_tag = tag_create_random();
3065 port->bond_compat_is_stale = true;
3066 unixctl_command_reply(conn, 200, "migrated");
3070 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3071 void *aux OVS_UNUSED)
3073 char *args = (char *) args_;
3074 char *save_ptr = NULL;
3075 char *bond_s, *slave_s;
3077 struct iface *iface;
3079 bond_s = strtok_r(args, " ", &save_ptr);
3080 slave_s = strtok_r(NULL, " ", &save_ptr);
3082 unixctl_command_reply(conn, 501,
3083 "usage: bond/set-active-slave BOND SLAVE");
3087 port = bond_find(bond_s);
3089 unixctl_command_reply(conn, 501, "no such bond");
3093 iface = port_lookup_iface(port, slave_s);
3095 unixctl_command_reply(conn, 501, "no such slave");
3099 if (!iface->enabled) {
3100 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3104 if (port->active_iface != iface->port_ifidx) {
3105 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3106 port->active_iface = iface->port_ifidx;
3107 port->active_iface_tag = tag_create_random();
3108 VLOG_INFO("port %s: active interface is now %s",
3109 port->name, iface->name);
3110 bond_send_learning_packets(port);
3111 unixctl_command_reply(conn, 200, "done");
3113 unixctl_command_reply(conn, 200, "no change");
3118 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3120 char *args = (char *) args_;
3121 char *save_ptr = NULL;
3122 char *bond_s, *slave_s;
3124 struct iface *iface;
3126 bond_s = strtok_r(args, " ", &save_ptr);
3127 slave_s = strtok_r(NULL, " ", &save_ptr);
3129 unixctl_command_reply(conn, 501,
3130 "usage: bond/enable/disable-slave BOND SLAVE");
3134 port = bond_find(bond_s);
3136 unixctl_command_reply(conn, 501, "no such bond");
3140 iface = port_lookup_iface(port, slave_s);
3142 unixctl_command_reply(conn, 501, "no such slave");
3146 bond_enable_slave(iface, enable);
3147 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3151 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3152 void *aux OVS_UNUSED)
3154 enable_slave(conn, args, true);
3158 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3159 void *aux OVS_UNUSED)
3161 enable_slave(conn, args, false);
3165 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3166 void *aux OVS_UNUSED)
3168 uint8_t mac[ETH_ADDR_LEN];
3172 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3173 == ETH_ADDR_SCAN_COUNT) {
3174 hash = bond_hash(mac);
3176 hash_cstr = xasprintf("%u", hash);
3177 unixctl_command_reply(conn, 200, hash_cstr);
3180 unixctl_command_reply(conn, 501, "invalid mac");
3187 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3188 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3189 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3190 unixctl_command_register("bond/set-active-slave",
3191 bond_unixctl_set_active_slave, NULL);
3192 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3194 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3196 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3199 /* Port functions. */
3201 static struct port *
3202 port_create(struct bridge *br, const char *name)
3206 port = xzalloc(sizeof *port);
3208 port->port_idx = br->n_ports;
3210 port->trunks = NULL;
3211 port->name = xstrdup(name);
3212 port->active_iface = -1;
3214 if (br->n_ports >= br->allocated_ports) {
3215 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3218 br->ports[br->n_ports++] = port;
3219 shash_add_assert(&br->port_by_name, port->name, port);
3221 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3228 get_port_other_config(const struct ovsrec_port *port, const char *key,
3229 const char *default_value)
3233 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3235 return value ? value : default_value;
3239 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3241 struct shash new_ifaces;
3244 /* Collect list of new interfaces. */
3245 shash_init(&new_ifaces);
3246 for (i = 0; i < cfg->n_interfaces; i++) {
3247 const char *name = cfg->interfaces[i]->name;
3248 shash_add_once(&new_ifaces, name, NULL);
3251 /* Get rid of deleted interfaces. */
3252 for (i = 0; i < port->n_ifaces; ) {
3253 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3254 iface_destroy(port->ifaces[i]);
3260 shash_destroy(&new_ifaces);
3264 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3266 struct shash new_ifaces;
3267 long long int next_rebalance;
3268 unsigned long *trunks;
3274 /* Update settings. */
3275 port->updelay = cfg->bond_updelay;
3276 if (port->updelay < 0) {
3279 port->updelay = cfg->bond_downdelay;
3280 if (port->downdelay < 0) {
3281 port->downdelay = 0;
3283 port->bond_rebalance_interval = atoi(
3284 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3285 if (port->bond_rebalance_interval < 1000) {
3286 port->bond_rebalance_interval = 1000;
3288 next_rebalance = time_msec() + port->bond_rebalance_interval;
3289 if (port->bond_next_rebalance > next_rebalance) {
3290 port->bond_next_rebalance = next_rebalance;
3293 /* Add new interfaces and update 'cfg' member of existing ones. */
3294 shash_init(&new_ifaces);
3295 for (i = 0; i < cfg->n_interfaces; i++) {
3296 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3297 struct iface *iface;
3299 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3300 VLOG_WARN("port %s: %s specified twice as port interface",
3301 port->name, if_cfg->name);
3305 iface = iface_lookup(port->bridge, if_cfg->name);
3307 if (iface->port != port) {
3308 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3310 port->bridge->name, if_cfg->name, iface->port->name);
3313 iface->cfg = if_cfg;
3315 iface_create(port, if_cfg);
3318 shash_destroy(&new_ifaces);
3323 if (port->n_ifaces < 2) {
3325 if (vlan >= 0 && vlan <= 4095) {
3326 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3331 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3332 * they even work as-is. But they have not been tested. */
3333 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3337 if (port->vlan != vlan) {
3339 bridge_flush(port->bridge);
3342 /* Get trunked VLANs. */
3344 if (vlan < 0 && cfg->n_trunks) {
3348 trunks = bitmap_allocate(4096);
3350 for (i = 0; i < cfg->n_trunks; i++) {
3351 int trunk = cfg->trunks[i];
3353 bitmap_set1(trunks, trunk);
3359 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3360 port->name, cfg->n_trunks);
3362 if (n_errors == cfg->n_trunks) {
3363 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3365 bitmap_free(trunks);
3368 } else if (vlan >= 0 && cfg->n_trunks) {
3369 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3373 ? port->trunks != NULL
3374 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3375 bridge_flush(port->bridge);
3377 bitmap_free(port->trunks);
3378 port->trunks = trunks;
3382 port_destroy(struct port *port)
3385 struct bridge *br = port->bridge;
3389 proc_net_compat_update_vlan(port->name, NULL, 0);
3390 proc_net_compat_update_bond(port->name, NULL);
3392 for (i = 0; i < MAX_MIRRORS; i++) {
3393 struct mirror *m = br->mirrors[i];
3394 if (m && m->out_port == port) {
3399 while (port->n_ifaces > 0) {
3400 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3403 shash_find_and_delete_assert(&br->port_by_name, port->name);
3405 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3406 del->port_idx = port->port_idx;
3409 bitmap_free(port->trunks);
3416 static struct port *
3417 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3419 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3420 return iface ? iface->port : NULL;
3423 static struct port *
3424 port_lookup(const struct bridge *br, const char *name)
3426 return shash_find_data(&br->port_by_name, name);
3429 static struct iface *
3430 port_lookup_iface(const struct port *port, const char *name)
3432 struct iface *iface = iface_lookup(port->bridge, name);
3433 return iface && iface->port == port ? iface : NULL;
3437 port_update_bonding(struct port *port)
3439 if (port->n_ifaces < 2) {
3440 /* Not a bonded port. */
3441 if (port->bond_hash) {
3442 free(port->bond_hash);
3443 port->bond_hash = NULL;
3444 port->bond_compat_is_stale = true;
3445 port->bond_fake_iface = false;
3448 if (!port->bond_hash) {
3451 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3452 for (i = 0; i <= BOND_MASK; i++) {
3453 struct bond_entry *e = &port->bond_hash[i];
3457 port->no_ifaces_tag = tag_create_random();
3458 bond_choose_active_iface(port);
3459 port->bond_next_rebalance
3460 = time_msec() + port->bond_rebalance_interval;
3462 if (port->cfg->bond_fake_iface) {
3463 port->bond_next_fake_iface_update = time_msec();
3466 port->bond_compat_is_stale = true;
3467 port->bond_fake_iface = port->cfg->bond_fake_iface;
3472 port_update_bond_compat(struct port *port)
3474 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3475 struct compat_bond bond;
3478 if (port->n_ifaces < 2) {
3479 proc_net_compat_update_bond(port->name, NULL);
3484 bond.updelay = port->updelay;
3485 bond.downdelay = port->downdelay;
3488 bond.hashes = compat_hashes;
3489 if (port->bond_hash) {
3490 const struct bond_entry *e;
3491 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3492 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3493 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3494 cbh->hash = e - port->bond_hash;
3495 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3500 bond.n_slaves = port->n_ifaces;
3501 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3502 for (i = 0; i < port->n_ifaces; i++) {
3503 struct iface *iface = port->ifaces[i];
3504 struct compat_bond_slave *slave = &bond.slaves[i];
3505 slave->name = iface->name;
3507 /* We need to make the same determination as the Linux bonding
3508 * code to determine whether a slave should be consider "up".
3509 * The Linux function bond_miimon_inspect() supports four
3510 * BOND_LINK_* states:
3512 * - BOND_LINK_UP: carrier detected, updelay has passed.
3513 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3514 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3515 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3517 * The function bond_info_show_slave() only considers BOND_LINK_UP
3518 * to be "up" and anything else to be "down".
3520 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3524 netdev_get_etheraddr(iface->netdev, slave->mac);
3527 if (port->bond_fake_iface) {
3528 struct netdev *bond_netdev;
3530 if (!netdev_open_default(port->name, &bond_netdev)) {
3532 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3534 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3536 netdev_close(bond_netdev);
3540 proc_net_compat_update_bond(port->name, &bond);
3545 port_update_vlan_compat(struct port *port)
3547 struct bridge *br = port->bridge;
3548 char *vlandev_name = NULL;
3550 if (port->vlan > 0) {
3551 /* Figure out the name that the VLAN device should actually have, if it
3552 * existed. This takes some work because the VLAN device would not
3553 * have port->name in its name; rather, it would have the trunk port's
3554 * name, and 'port' would be attached to a bridge that also had the
3555 * VLAN device one of its ports. So we need to find a trunk port that
3556 * includes port->vlan.
3558 * There might be more than one candidate. This doesn't happen on
3559 * XenServer, so if it happens we just pick the first choice in
3560 * alphabetical order instead of creating multiple VLAN devices. */
3562 for (i = 0; i < br->n_ports; i++) {
3563 struct port *p = br->ports[i];
3564 if (port_trunks_vlan(p, port->vlan)
3566 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3568 uint8_t ea[ETH_ADDR_LEN];
3569 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3570 if (!eth_addr_is_multicast(ea) &&
3571 !eth_addr_is_reserved(ea) &&
3572 !eth_addr_is_zero(ea)) {
3573 vlandev_name = p->name;
3578 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3581 /* Interface functions. */
3583 static struct iface *
3584 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3586 struct bridge *br = port->bridge;
3587 struct iface *iface;
3588 char *name = if_cfg->name;
3591 iface = xzalloc(sizeof *iface);
3593 iface->port_ifidx = port->n_ifaces;
3594 iface->name = xstrdup(name);
3595 iface->dp_ifidx = -1;
3596 iface->tag = tag_create_random();
3597 iface->delay_expires = LLONG_MAX;
3598 iface->netdev = NULL;
3599 iface->cfg = if_cfg;
3601 shash_add_assert(&br->iface_by_name, iface->name, iface);
3603 /* Attempt to create the network interface in case it doesn't exist yet. */
3604 if (!iface_is_internal(br, iface->name)) {
3605 error = set_up_iface(if_cfg, iface, true);
3607 VLOG_WARN("could not create iface %s: %s", iface->name,
3610 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3617 if (port->n_ifaces >= port->allocated_ifaces) {
3618 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3619 sizeof *port->ifaces);
3621 port->ifaces[port->n_ifaces++] = iface;
3622 if (port->n_ifaces > 1) {
3623 br->has_bonded_ports = true;
3626 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3634 iface_destroy(struct iface *iface)
3637 struct port *port = iface->port;
3638 struct bridge *br = port->bridge;
3639 bool del_active = port->active_iface == iface->port_ifidx;
3642 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3644 if (iface->dp_ifidx >= 0) {
3645 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3648 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3649 del->port_ifidx = iface->port_ifidx;
3651 netdev_close(iface->netdev);
3654 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3655 bond_choose_active_iface(port);
3656 bond_send_learning_packets(port);
3662 bridge_flush(port->bridge);
3666 static struct iface *
3667 iface_lookup(const struct bridge *br, const char *name)
3669 return shash_find_data(&br->iface_by_name, name);
3672 static struct iface *
3673 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3675 return port_array_get(&br->ifaces, dp_ifidx);
3678 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3679 * 'br', that is, an interface that is entirely simulated within the datapath.
3680 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3681 * interfaces are created by setting "iface.<iface>.internal = true".
3683 * In addition, we have a kluge-y feature that creates an internal port with
3684 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3685 * This feature needs to go away in the long term. Until then, this is one
3686 * reason why this function takes a name instead of a struct iface: the fake
3687 * interfaces created this way do not have a struct iface. */
3689 iface_is_internal(const struct bridge *br, const char *if_name)
3691 struct iface *iface;
3694 if (!strcmp(if_name, br->name)) {
3698 iface = iface_lookup(br, if_name);
3699 if (iface && !strcmp(iface->cfg->type, "internal")) {
3703 port = port_lookup(br, if_name);
3704 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3710 /* Set Ethernet address of 'iface', if one is specified in the configuration
3713 iface_set_mac(struct iface *iface)
3715 uint8_t ea[ETH_ADDR_LEN];
3717 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3718 if (eth_addr_is_multicast(ea)) {
3719 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3721 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3722 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3723 iface->name, iface->name);
3725 int error = netdev_set_etheraddr(iface->netdev, ea);
3727 VLOG_ERR("interface %s: setting MAC failed (%s)",
3728 iface->name, strerror(error));
3735 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3736 struct shash *shash)
3741 for (i = 0; i < n; i++) {
3742 shash_add(shash, keys[i], values[i]);
3746 struct iface_delete_queues_cbdata {
3747 struct netdev *netdev;
3748 const struct ovsdb_datum *queues;
3752 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3754 union ovsdb_atom atom;
3756 atom.integer = target;
3757 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3761 iface_delete_queues(unsigned int queue_id,
3762 const struct shash *details OVS_UNUSED, void *cbdata_)
3764 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3766 if (!queue_ids_include(cbdata->queues, queue_id)) {
3767 netdev_delete_queue(cbdata->netdev, queue_id);
3772 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3774 if (!qos || qos->type[0] == '\0') {
3775 netdev_set_qos(iface->netdev, NULL, NULL);
3777 struct iface_delete_queues_cbdata cbdata;
3778 struct shash details;
3781 /* Configure top-level Qos for 'iface'. */
3782 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3783 qos->n_other_config, &details);
3784 netdev_set_qos(iface->netdev, qos->type, &details);
3785 shash_destroy(&details);
3787 /* Deconfigure queues that were deleted. */
3788 cbdata.netdev = iface->netdev;
3789 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3791 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3793 /* Configure queues for 'iface'. */
3794 for (i = 0; i < qos->n_queues; i++) {
3795 const struct ovsrec_queue *queue = qos->value_queues[i];
3796 unsigned int queue_id = qos->key_queues[i];
3798 shash_from_ovs_idl_map(queue->key_other_config,
3799 queue->value_other_config,
3800 queue->n_other_config, &details);
3801 netdev_set_queue(iface->netdev, queue_id, &details);
3802 shash_destroy(&details);
3807 /* Port mirroring. */
3809 static struct mirror *
3810 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3814 for (i = 0; i < MAX_MIRRORS; i++) {
3815 struct mirror *m = br->mirrors[i];
3816 if (m && uuid_equals(uuid, &m->uuid)) {
3824 mirror_reconfigure(struct bridge *br)
3826 unsigned long *rspan_vlans;
3829 /* Get rid of deleted mirrors. */
3830 for (i = 0; i < MAX_MIRRORS; i++) {
3831 struct mirror *m = br->mirrors[i];
3833 const struct ovsdb_datum *mc;
3834 union ovsdb_atom atom;
3836 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3837 atom.uuid = br->mirrors[i]->uuid;
3838 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3844 /* Add new mirrors and reconfigure existing ones. */
3845 for (i = 0; i < br->cfg->n_mirrors; i++) {
3846 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3847 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3849 mirror_reconfigure_one(m, cfg);
3851 mirror_create(br, cfg);
3855 /* Update port reserved status. */
3856 for (i = 0; i < br->n_ports; i++) {
3857 br->ports[i]->is_mirror_output_port = false;
3859 for (i = 0; i < MAX_MIRRORS; i++) {
3860 struct mirror *m = br->mirrors[i];
3861 if (m && m->out_port) {
3862 m->out_port->is_mirror_output_port = true;
3866 /* Update flooded vlans (for RSPAN). */
3868 if (br->cfg->n_flood_vlans) {
3869 rspan_vlans = bitmap_allocate(4096);
3871 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3872 int64_t vlan = br->cfg->flood_vlans[i];
3873 if (vlan >= 0 && vlan < 4096) {
3874 bitmap_set1(rspan_vlans, vlan);
3875 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3878 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3883 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3889 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3894 for (i = 0; ; i++) {
3895 if (i >= MAX_MIRRORS) {
3896 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3897 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3900 if (!br->mirrors[i]) {
3905 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
3908 br->mirrors[i] = m = xzalloc(sizeof *m);
3911 m->name = xstrdup(cfg->name);
3912 shash_init(&m->src_ports);
3913 shash_init(&m->dst_ports);
3919 mirror_reconfigure_one(m, cfg);
3923 mirror_destroy(struct mirror *m)
3926 struct bridge *br = m->bridge;
3929 for (i = 0; i < br->n_ports; i++) {
3930 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3931 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3934 shash_destroy(&m->src_ports);
3935 shash_destroy(&m->dst_ports);
3938 m->bridge->mirrors[m->idx] = NULL;
3947 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3948 struct shash *names)
3952 for (i = 0; i < n_ports; i++) {
3953 const char *name = ports[i]->name;
3954 if (port_lookup(m->bridge, name)) {
3955 shash_add_once(names, name, NULL);
3957 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3958 "port %s", m->bridge->name, m->name, name);
3964 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3970 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3972 for (i = 0; i < cfg->n_select_vlan; i++) {
3973 int64_t vlan = cfg->select_vlan[i];
3974 if (vlan < 0 || vlan > 4095) {
3975 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3976 m->bridge->name, m->name, vlan);
3978 (*vlans)[n_vlans++] = vlan;
3985 vlan_is_mirrored(const struct mirror *m, int vlan)
3989 for (i = 0; i < m->n_vlans; i++) {
3990 if (m->vlans[i] == vlan) {
3998 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4002 for (i = 0; i < m->n_vlans; i++) {
4003 if (port_trunks_vlan(p, m->vlans[i])) {
4011 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4013 struct shash src_ports, dst_ports;
4014 mirror_mask_t mirror_bit;
4015 struct port *out_port;
4022 if (strcmp(cfg->name, m->name)) {
4024 m->name = xstrdup(cfg->name);
4027 /* Get output port. */
4028 if (cfg->output_port) {
4029 out_port = port_lookup(m->bridge, cfg->output_port->name);
4031 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4032 m->bridge->name, m->name);
4038 if (cfg->output_vlan) {
4039 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4040 "output vlan; ignoring output vlan",
4041 m->bridge->name, m->name);
4043 } else if (cfg->output_vlan) {
4045 out_vlan = *cfg->output_vlan;
4047 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4048 m->bridge->name, m->name);
4053 shash_init(&src_ports);
4054 shash_init(&dst_ports);
4055 if (cfg->select_all) {
4056 for (i = 0; i < m->bridge->n_ports; i++) {
4057 const char *name = m->bridge->ports[i]->name;
4058 shash_add_once(&src_ports, name, NULL);
4059 shash_add_once(&dst_ports, name, NULL);
4064 /* Get ports, and drop duplicates and ports that don't exist. */
4065 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4067 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4070 /* Get all the vlans, and drop duplicate and invalid vlans. */
4071 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4074 /* Update mirror data. */
4075 if (!shash_equal_keys(&m->src_ports, &src_ports)
4076 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4077 || m->n_vlans != n_vlans
4078 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4079 || m->out_port != out_port
4080 || m->out_vlan != out_vlan) {
4081 bridge_flush(m->bridge);
4083 shash_swap(&m->src_ports, &src_ports);
4084 shash_swap(&m->dst_ports, &dst_ports);
4087 m->n_vlans = n_vlans;
4088 m->out_port = out_port;
4089 m->out_vlan = out_vlan;
4092 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4093 for (i = 0; i < m->bridge->n_ports; i++) {
4094 struct port *port = m->bridge->ports[i];
4096 if (shash_find(&m->src_ports, port->name)
4099 ? port_trunks_any_mirrored_vlan(m, port)
4100 : vlan_is_mirrored(m, port->vlan)))) {
4101 port->src_mirrors |= mirror_bit;
4103 port->src_mirrors &= ~mirror_bit;
4106 if (shash_find(&m->dst_ports, port->name)) {
4107 port->dst_mirrors |= mirror_bit;
4109 port->dst_mirrors &= ~mirror_bit;
4114 shash_destroy(&src_ports);
4115 shash_destroy(&dst_ports);