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;
1426 enum ofproto_fail_mode fail_mode;
1429 /* Collect old ports. */
1430 shash_init(&old_ports);
1431 for (i = 0; i < br->n_ports; i++) {
1432 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1435 /* Collect new ports. */
1436 shash_init(&new_ports);
1437 for (i = 0; i < br->cfg->n_ports; i++) {
1438 const char *name = br->cfg->ports[i]->name;
1439 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1440 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1445 /* If we have a controller, then we need a local port. Complain if the
1446 * user didn't specify one.
1448 * XXX perhaps we should synthesize a port ourselves in this case. */
1449 if (bridge_get_controllers(br, NULL)) {
1450 char local_name[IF_NAMESIZE];
1453 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1454 local_name, sizeof local_name);
1455 if (!error && !shash_find(&new_ports, local_name)) {
1456 VLOG_WARN("bridge %s: controller specified but no local port "
1457 "(port named %s) defined",
1458 br->name, local_name);
1462 /* Get rid of deleted ports.
1463 * Get rid of deleted interfaces on ports that still exist. */
1464 SHASH_FOR_EACH (node, &old_ports) {
1465 struct port *port = node->data;
1466 const struct ovsrec_port *port_cfg;
1468 port_cfg = shash_find_data(&new_ports, node->name);
1472 port_del_ifaces(port, port_cfg);
1476 /* Create new ports.
1477 * Add new interfaces to existing ports.
1478 * Reconfigure existing ports. */
1479 SHASH_FOR_EACH (node, &new_ports) {
1480 struct port *port = shash_find_data(&old_ports, node->name);
1482 port = port_create(br, node->name);
1485 port_reconfigure(port, node->data);
1486 if (!port->n_ifaces) {
1487 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1488 br->name, port->name);
1492 shash_destroy(&old_ports);
1493 shash_destroy(&new_ports);
1495 /* Set the fail-mode */
1496 fail_mode = !br->cfg->fail_mode
1497 || !strcmp(br->cfg->fail_mode, "standalone")
1498 ? OFPROTO_FAIL_STANDALONE
1499 : OFPROTO_FAIL_SECURE;
1500 if ((ofproto_get_fail_mode(br->ofproto) != fail_mode)
1501 && !ofproto_has_controller(br->ofproto)) {
1502 ofproto_flush_flows(br->ofproto);
1504 ofproto_set_fail_mode(br->ofproto, fail_mode);
1506 /* Delete all flows if we're switching from connected to standalone or vice
1507 * versa. (XXX Should we delete all flows if we are switching from one
1508 * controller to another?) */
1510 /* Configure OpenFlow management listener. */
1511 svec_init(&listeners);
1512 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1513 ovs_rundir, br->name));
1514 svec_init(&old_listeners);
1515 ofproto_get_listeners(br->ofproto, &old_listeners);
1516 if (!svec_equal(&listeners, &old_listeners)) {
1517 ofproto_set_listeners(br->ofproto, &listeners);
1519 svec_destroy(&listeners);
1520 svec_destroy(&old_listeners);
1522 /* Configure OpenFlow controller connection snooping. */
1524 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1525 ovs_rundir, br->name));
1526 svec_init(&old_snoops);
1527 ofproto_get_snoops(br->ofproto, &old_snoops);
1528 if (!svec_equal(&snoops, &old_snoops)) {
1529 ofproto_set_snoops(br->ofproto, &snoops);
1531 svec_destroy(&snoops);
1532 svec_destroy(&old_snoops);
1534 mirror_reconfigure(br);
1538 bridge_reconfigure_remotes(struct bridge *br,
1539 const struct sockaddr_in *managers,
1542 struct ovsrec_controller **controllers;
1543 size_t n_controllers;
1545 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1547 n_controllers = bridge_get_controllers(br, &controllers);
1548 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1549 ofproto_flush_flows(br->ofproto);
1552 if (!n_controllers) {
1553 union ofp_action action;
1556 /* Clear out controllers. */
1557 ofproto_set_controllers(br->ofproto, NULL, 0);
1559 /* If there are no controllers and the bridge is in standalone
1560 * mode, set up a flow that matches every packet and directs
1561 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1562 * switch is in secure mode and we won't pass any traffic until
1563 * a controller has been defined and it tells us to do so. */
1564 if (ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1565 memset(&action, 0, sizeof action);
1566 action.type = htons(OFPAT_OUTPUT);
1567 action.output.len = htons(sizeof action);
1568 action.output.port = htons(OFPP_NORMAL);
1569 memset(&flow, 0, sizeof flow);
1570 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1573 struct ofproto_controller *ocs;
1576 ocs = xmalloc(n_controllers * sizeof *ocs);
1577 for (i = 0; i < n_controllers; i++) {
1578 struct ovsrec_controller *c = controllers[i];
1579 struct ofproto_controller *oc = &ocs[i];
1581 if (strcmp(c->target, "discover")) {
1582 struct iface *local_iface;
1585 local_iface = bridge_get_local_iface(br);
1586 if (local_iface && c->local_ip
1587 && inet_aton(c->local_ip, &ip)) {
1588 struct netdev *netdev = local_iface->netdev;
1589 struct in_addr mask, gateway;
1591 if (!c->local_netmask
1592 || !inet_aton(c->local_netmask, &mask)) {
1595 if (!c->local_gateway
1596 || !inet_aton(c->local_gateway, &gateway)) {
1600 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1602 mask.s_addr = guess_netmask(ip.s_addr);
1604 if (!netdev_set_in4(netdev, ip, mask)) {
1605 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1607 br->name, IP_ARGS(&ip.s_addr),
1608 IP_ARGS(&mask.s_addr));
1611 if (gateway.s_addr) {
1612 if (!netdev_add_router(netdev, gateway)) {
1613 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1614 br->name, IP_ARGS(&gateway.s_addr));
1620 oc->target = c->target;
1621 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1622 oc->probe_interval = (c->inactivity_probe
1623 ? *c->inactivity_probe / 1000 : 5);
1624 oc->band = (!c->connection_mode
1625 || !strcmp(c->connection_mode, "in-band")
1627 : OFPROTO_OUT_OF_BAND);
1628 oc->accept_re = c->discover_accept_regex;
1629 oc->update_resolv_conf = c->discover_update_resolv_conf;
1630 oc->rate_limit = (c->controller_rate_limit
1631 ? *c->controller_rate_limit : 0);
1632 oc->burst_limit = (c->controller_burst_limit
1633 ? *c->controller_burst_limit : 0);
1635 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1641 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1646 for (i = 0; i < br->n_ports; i++) {
1647 struct port *port = br->ports[i];
1648 for (j = 0; j < port->n_ifaces; j++) {
1649 struct iface *iface = port->ifaces[j];
1650 shash_add_once(ifaces, iface->name, iface);
1652 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1653 shash_add_once(ifaces, port->name, NULL);
1658 /* For robustness, in case the administrator moves around datapath ports behind
1659 * our back, we re-check all the datapath port numbers here.
1661 * This function will set the 'dp_ifidx' members of interfaces that have
1662 * disappeared to -1, so only call this function from a context where those
1663 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1664 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1665 * datapath, which doesn't support UINT16_MAX+1 ports. */
1667 bridge_fetch_dp_ifaces(struct bridge *br)
1669 struct odp_port *dpif_ports;
1670 size_t n_dpif_ports;
1673 /* Reset all interface numbers. */
1674 for (i = 0; i < br->n_ports; i++) {
1675 struct port *port = br->ports[i];
1676 for (j = 0; j < port->n_ifaces; j++) {
1677 struct iface *iface = port->ifaces[j];
1678 iface->dp_ifidx = -1;
1681 port_array_clear(&br->ifaces);
1683 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1684 for (i = 0; i < n_dpif_ports; i++) {
1685 struct odp_port *p = &dpif_ports[i];
1686 struct iface *iface = iface_lookup(br, p->devname);
1688 if (iface->dp_ifidx >= 0) {
1689 VLOG_WARN("%s reported interface %s twice",
1690 dpif_name(br->dpif), p->devname);
1691 } else if (iface_from_dp_ifidx(br, p->port)) {
1692 VLOG_WARN("%s reported interface %"PRIu16" twice",
1693 dpif_name(br->dpif), p->port);
1695 port_array_set(&br->ifaces, p->port, iface);
1696 iface->dp_ifidx = p->port;
1700 int64_t ofport = (iface->dp_ifidx >= 0
1701 ? odp_port_to_ofp_port(iface->dp_ifidx)
1703 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1710 /* Bridge packet processing functions. */
1713 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1715 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1718 static struct bond_entry *
1719 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1721 return &port->bond_hash[bond_hash(mac)];
1725 bond_choose_iface(const struct port *port)
1727 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1728 size_t i, best_down_slave = -1;
1729 long long next_delay_expiration = LLONG_MAX;
1731 for (i = 0; i < port->n_ifaces; i++) {
1732 struct iface *iface = port->ifaces[i];
1734 if (iface->enabled) {
1736 } else if (iface->delay_expires < next_delay_expiration) {
1737 best_down_slave = i;
1738 next_delay_expiration = iface->delay_expires;
1742 if (best_down_slave != -1) {
1743 struct iface *iface = port->ifaces[best_down_slave];
1745 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1746 "since no other interface is up", iface->name,
1747 iface->delay_expires - time_msec());
1748 bond_enable_slave(iface, true);
1751 return best_down_slave;
1755 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1756 uint16_t *dp_ifidx, tag_type *tags)
1758 struct iface *iface;
1760 assert(port->n_ifaces);
1761 if (port->n_ifaces == 1) {
1762 iface = port->ifaces[0];
1764 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1765 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1766 || !port->ifaces[e->iface_idx]->enabled) {
1767 /* XXX select interface properly. The current interface selection
1768 * is only good for testing the rebalancing code. */
1769 e->iface_idx = bond_choose_iface(port);
1770 if (e->iface_idx < 0) {
1771 *tags |= port->no_ifaces_tag;
1774 e->iface_tag = tag_create_random();
1775 ((struct port *) port)->bond_compat_is_stale = true;
1777 *tags |= e->iface_tag;
1778 iface = port->ifaces[e->iface_idx];
1780 *dp_ifidx = iface->dp_ifidx;
1781 *tags |= iface->tag; /* Currently only used for bonding. */
1786 bond_link_status_update(struct iface *iface, bool carrier)
1788 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1789 struct port *port = iface->port;
1791 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1792 /* Nothing to do. */
1795 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1796 iface->name, carrier ? "detected" : "dropped");
1797 if (carrier == iface->enabled) {
1798 iface->delay_expires = LLONG_MAX;
1799 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1800 iface->name, carrier ? "disabled" : "enabled");
1801 } else if (carrier && port->active_iface < 0) {
1802 bond_enable_slave(iface, true);
1803 if (port->updelay) {
1804 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1805 "other interface is up", iface->name, port->updelay);
1808 int delay = carrier ? port->updelay : port->downdelay;
1809 iface->delay_expires = time_msec() + delay;
1812 "interface %s: will be %s if it stays %s for %d ms",
1814 carrier ? "enabled" : "disabled",
1815 carrier ? "up" : "down",
1822 bond_choose_active_iface(struct port *port)
1824 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1826 port->active_iface = bond_choose_iface(port);
1827 port->active_iface_tag = tag_create_random();
1828 if (port->active_iface >= 0) {
1829 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1830 port->name, port->ifaces[port->active_iface]->name);
1832 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1838 bond_enable_slave(struct iface *iface, bool enable)
1840 struct port *port = iface->port;
1841 struct bridge *br = port->bridge;
1843 /* This acts as a recursion check. If the act of disabling a slave
1844 * causes a different slave to be enabled, the flag will allow us to
1845 * skip redundant work when we reenter this function. It must be
1846 * cleared on exit to keep things safe with multiple bonds. */
1847 static bool moving_active_iface = false;
1849 iface->delay_expires = LLONG_MAX;
1850 if (enable == iface->enabled) {
1854 iface->enabled = enable;
1855 if (!iface->enabled) {
1856 VLOG_WARN("interface %s: disabled", iface->name);
1857 ofproto_revalidate(br->ofproto, iface->tag);
1858 if (iface->port_ifidx == port->active_iface) {
1859 ofproto_revalidate(br->ofproto,
1860 port->active_iface_tag);
1862 /* Disabling a slave can lead to another slave being immediately
1863 * enabled if there will be no active slaves but one is waiting
1864 * on an updelay. In this case we do not need to run most of the
1865 * code for the newly enabled slave since there was no period
1866 * without an active slave and it is redundant with the disabling
1868 moving_active_iface = true;
1869 bond_choose_active_iface(port);
1871 bond_send_learning_packets(port);
1873 VLOG_WARN("interface %s: enabled", iface->name);
1874 if (port->active_iface < 0 && !moving_active_iface) {
1875 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1876 bond_choose_active_iface(port);
1877 bond_send_learning_packets(port);
1879 iface->tag = tag_create_random();
1882 moving_active_iface = false;
1883 port->bond_compat_is_stale = true;
1886 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1887 * bond interface. */
1889 bond_update_fake_iface_stats(struct port *port)
1891 struct netdev_stats bond_stats;
1892 struct netdev *bond_dev;
1895 memset(&bond_stats, 0, sizeof bond_stats);
1897 for (i = 0; i < port->n_ifaces; i++) {
1898 struct netdev_stats slave_stats;
1900 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1901 /* XXX: We swap the stats here because they are swapped back when
1902 * reported by the internal device. The reason for this is
1903 * internal devices normally represent packets going into the system
1904 * but when used as fake bond device they represent packets leaving
1905 * the system. We really should do this in the internal device
1906 * itself because changing it here reverses the counts from the
1907 * perspective of the switch. However, the internal device doesn't
1908 * know what type of device it represents so we have to do it here
1910 bond_stats.tx_packets += slave_stats.rx_packets;
1911 bond_stats.tx_bytes += slave_stats.rx_bytes;
1912 bond_stats.rx_packets += slave_stats.tx_packets;
1913 bond_stats.rx_bytes += slave_stats.tx_bytes;
1917 if (!netdev_open_default(port->name, &bond_dev)) {
1918 netdev_set_stats(bond_dev, &bond_stats);
1919 netdev_close(bond_dev);
1924 bond_run(struct bridge *br)
1928 for (i = 0; i < br->n_ports; i++) {
1929 struct port *port = br->ports[i];
1931 if (port->n_ifaces >= 2) {
1932 for (j = 0; j < port->n_ifaces; j++) {
1933 struct iface *iface = port->ifaces[j];
1934 if (time_msec() >= iface->delay_expires) {
1935 bond_enable_slave(iface, !iface->enabled);
1939 if (port->bond_fake_iface
1940 && time_msec() >= port->bond_next_fake_iface_update) {
1941 bond_update_fake_iface_stats(port);
1942 port->bond_next_fake_iface_update = time_msec() + 1000;
1946 if (port->bond_compat_is_stale) {
1947 port->bond_compat_is_stale = false;
1948 port_update_bond_compat(port);
1954 bond_wait(struct bridge *br)
1958 for (i = 0; i < br->n_ports; i++) {
1959 struct port *port = br->ports[i];
1960 if (port->n_ifaces < 2) {
1963 for (j = 0; j < port->n_ifaces; j++) {
1964 struct iface *iface = port->ifaces[j];
1965 if (iface->delay_expires != LLONG_MAX) {
1966 poll_timer_wait_until(iface->delay_expires);
1969 if (port->bond_fake_iface) {
1970 poll_timer_wait_until(port->bond_next_fake_iface_update);
1976 set_dst(struct dst *p, const flow_t *flow,
1977 const struct port *in_port, const struct port *out_port,
1980 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1981 : in_port->vlan >= 0 ? in_port->vlan
1982 : ntohs(flow->dl_vlan));
1983 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1987 swap_dst(struct dst *p, struct dst *q)
1989 struct dst tmp = *p;
1994 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1995 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1996 * that we push to the datapath. We could in fact fully sort the array by
1997 * vlan, but in most cases there are at most two different vlan tags so that's
1998 * possibly overkill.) */
2000 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2002 struct dst *first = dsts;
2003 struct dst *last = dsts + n_dsts;
2005 while (first != last) {
2007 * - All dsts < first have vlan == 'vlan'.
2008 * - All dsts >= last have vlan != 'vlan'.
2009 * - first < last. */
2010 while (first->vlan == vlan) {
2011 if (++first == last) {
2016 /* Same invariants, plus one additional:
2017 * - first->vlan != vlan.
2019 while (last[-1].vlan != vlan) {
2020 if (--last == first) {
2025 /* Same invariants, plus one additional:
2026 * - last[-1].vlan == vlan.*/
2027 swap_dst(first++, --last);
2032 mirror_mask_ffs(mirror_mask_t mask)
2034 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2039 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2040 const struct dst *test)
2043 for (i = 0; i < n_dsts; i++) {
2044 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2052 port_trunks_vlan(const struct port *port, uint16_t vlan)
2054 return (port->vlan < 0
2055 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2059 port_includes_vlan(const struct port *port, uint16_t vlan)
2061 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2065 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2066 const struct port *in_port, const struct port *out_port,
2067 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2069 mirror_mask_t mirrors = in_port->src_mirrors;
2070 struct dst *dst = dsts;
2073 if (out_port == FLOOD_PORT) {
2074 /* XXX use ODP_FLOOD if no vlans or bonding. */
2075 /* XXX even better, define each VLAN as a datapath port group */
2076 for (i = 0; i < br->n_ports; i++) {
2077 struct port *port = br->ports[i];
2078 if (port != in_port && port_includes_vlan(port, vlan)
2079 && !port->is_mirror_output_port
2080 && set_dst(dst, flow, in_port, port, tags)) {
2081 mirrors |= port->dst_mirrors;
2085 *nf_output_iface = NF_OUT_FLOOD;
2086 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2087 *nf_output_iface = dst->dp_ifidx;
2088 mirrors |= out_port->dst_mirrors;
2093 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2094 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2096 if (set_dst(dst, flow, in_port, m->out_port, tags)
2097 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2101 for (i = 0; i < br->n_ports; i++) {
2102 struct port *port = br->ports[i];
2103 if (port_includes_vlan(port, m->out_vlan)
2104 && set_dst(dst, flow, in_port, port, tags))
2108 if (port->vlan < 0) {
2109 dst->vlan = m->out_vlan;
2111 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2115 /* Use the vlan tag on the original flow instead of
2116 * the one passed in the vlan parameter. This ensures
2117 * that we compare the vlan from before any implicit
2118 * tagging tags place. This is necessary because
2119 * dst->vlan is the final vlan, after removing implicit
2121 flow_vlan = ntohs(flow->dl_vlan);
2122 if (flow_vlan == 0) {
2123 flow_vlan = OFP_VLAN_NONE;
2125 if (port == in_port && dst->vlan == flow_vlan) {
2126 /* Don't send out input port on same VLAN. */
2134 mirrors &= mirrors - 1;
2137 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2141 static void OVS_UNUSED
2142 print_dsts(const struct dst *dsts, size_t n)
2144 for (; n--; dsts++) {
2145 printf(">p%"PRIu16, dsts->dp_ifidx);
2146 if (dsts->vlan != OFP_VLAN_NONE) {
2147 printf("v%"PRIu16, dsts->vlan);
2153 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2154 const struct port *in_port, const struct port *out_port,
2155 tag_type *tags, struct odp_actions *actions,
2156 uint16_t *nf_output_iface)
2158 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2160 const struct dst *p;
2163 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2166 cur_vlan = ntohs(flow->dl_vlan);
2167 for (p = dsts; p < &dsts[n_dsts]; p++) {
2168 union odp_action *a;
2169 if (p->vlan != cur_vlan) {
2170 if (p->vlan == OFP_VLAN_NONE) {
2171 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2173 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2174 a->vlan_vid.vlan_vid = htons(p->vlan);
2178 a = odp_actions_add(actions, ODPAT_OUTPUT);
2179 a->output.port = p->dp_ifidx;
2183 /* Returns the effective vlan of a packet, taking into account both the
2184 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2185 * the packet is untagged and -1 indicates it has an invalid header and
2186 * should be dropped. */
2187 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2188 struct port *in_port, bool have_packet)
2190 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2191 * belongs to VLAN 0, so we should treat both cases identically. (In the
2192 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2193 * presumably to allow a priority to be specified. In the latter case, the
2194 * packet does not have any 802.1Q header.) */
2195 int vlan = ntohs(flow->dl_vlan);
2196 if (vlan == OFP_VLAN_NONE) {
2199 if (in_port->vlan >= 0) {
2201 /* XXX support double tagging? */
2203 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2204 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2205 "packet received on port %s configured with "
2206 "implicit VLAN %"PRIu16,
2207 br->name, ntohs(flow->dl_vlan),
2208 in_port->name, in_port->vlan);
2212 vlan = in_port->vlan;
2214 if (!port_includes_vlan(in_port, vlan)) {
2216 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2217 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2218 "packet received on port %s not configured for "
2220 br->name, vlan, in_port->name, vlan);
2229 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2230 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2231 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2233 is_gratuitous_arp(const flow_t *flow)
2235 return (flow->dl_type == htons(ETH_TYPE_ARP)
2236 && eth_addr_is_broadcast(flow->dl_dst)
2237 && (flow->nw_proto == ARP_OP_REPLY
2238 || (flow->nw_proto == ARP_OP_REQUEST
2239 && flow->nw_src == flow->nw_dst)));
2243 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2244 struct port *in_port)
2246 enum grat_arp_lock_type lock_type;
2249 /* We don't want to learn from gratuitous ARP packets that are reflected
2250 * back over bond slaves so we lock the learning table. */
2251 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2252 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2253 GRAT_ARP_LOCK_CHECK;
2255 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2258 /* The log messages here could actually be useful in debugging,
2259 * so keep the rate limit relatively high. */
2260 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2262 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2263 "on port %s in VLAN %d",
2264 br->name, ETH_ADDR_ARGS(flow->dl_src),
2265 in_port->name, vlan);
2266 ofproto_revalidate(br->ofproto, rev_tag);
2270 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2271 * dropped. Returns true if they may be forwarded, false if they should be
2274 * If 'have_packet' is true, it indicates that the caller is processing a
2275 * received packet. If 'have_packet' is false, then the caller is just
2276 * revalidating an existing flow because configuration has changed. Either
2277 * way, 'have_packet' only affects logging (there is no point in logging errors
2278 * during revalidation).
2280 * Sets '*in_portp' to the input port. This will be a null pointer if
2281 * flow->in_port does not designate a known input port (in which case
2282 * is_admissible() returns false).
2284 * When returning true, sets '*vlanp' to the effective VLAN of the input
2285 * packet, as returned by flow_get_vlan().
2287 * May also add tags to '*tags', although the current implementation only does
2288 * so in one special case.
2291 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2292 tag_type *tags, int *vlanp, struct port **in_portp)
2294 struct iface *in_iface;
2295 struct port *in_port;
2298 /* Find the interface and port structure for the received packet. */
2299 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2301 /* No interface? Something fishy... */
2303 /* Odd. A few possible reasons here:
2305 * - We deleted an interface but there are still a few packets
2306 * queued up from it.
2308 * - Someone externally added an interface (e.g. with "ovs-dpctl
2309 * add-if") that we don't know about.
2311 * - Packet arrived on the local port but the local port is not
2312 * one of our bridge ports.
2314 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2316 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2317 "interface %"PRIu16, br->name, flow->in_port);
2323 *in_portp = in_port = in_iface->port;
2324 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2329 /* Drop frames for reserved multicast addresses. */
2330 if (eth_addr_is_reserved(flow->dl_dst)) {
2334 /* Drop frames on ports reserved for mirroring. */
2335 if (in_port->is_mirror_output_port) {
2337 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2338 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2339 "%s, which is reserved exclusively for mirroring",
2340 br->name, in_port->name);
2345 /* Packets received on bonds need special attention to avoid duplicates. */
2346 if (in_port->n_ifaces > 1) {
2348 bool is_grat_arp_locked;
2350 if (eth_addr_is_multicast(flow->dl_dst)) {
2351 *tags |= in_port->active_iface_tag;
2352 if (in_port->active_iface != in_iface->port_ifidx) {
2353 /* Drop all multicast packets on inactive slaves. */
2358 /* Drop all packets for which we have learned a different input
2359 * port, because we probably sent the packet on one slave and got
2360 * it back on the other. Gratuitous ARP packets are an exception
2361 * to this rule: the host has moved to another switch. The exception
2362 * to the exception is if we locked the learning table to avoid
2363 * reflections on bond slaves. If this is the case, just drop the
2365 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2366 &is_grat_arp_locked);
2367 if (src_idx != -1 && src_idx != in_port->port_idx &&
2368 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2376 /* If the composed actions may be applied to any packet in the given 'flow',
2377 * returns true. Otherwise, the actions should only be applied to 'packet', or
2378 * not at all, if 'packet' was NULL. */
2380 process_flow(struct bridge *br, const flow_t *flow,
2381 const struct ofpbuf *packet, struct odp_actions *actions,
2382 tag_type *tags, uint16_t *nf_output_iface)
2384 struct port *in_port;
2385 struct port *out_port;
2389 /* Check whether we should drop packets in this flow. */
2390 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2395 /* Learn source MAC (but don't try to learn from revalidation). */
2397 update_learning_table(br, flow, vlan, in_port);
2400 /* Determine output port. */
2401 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2403 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2404 out_port = br->ports[out_port_idx];
2405 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2406 /* If we are revalidating but don't have a learning entry then
2407 * eject the flow. Installing a flow that floods packets opens
2408 * up a window of time where we could learn from a packet reflected
2409 * on a bond and blackhole packets before the learning table is
2410 * updated to reflect the correct port. */
2413 out_port = FLOOD_PORT;
2416 /* Don't send packets out their input ports. */
2417 if (in_port == out_port) {
2423 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2430 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2433 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2434 const struct ofp_phy_port *opp,
2437 struct bridge *br = br_;
2438 struct iface *iface;
2441 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2447 if (reason == OFPPR_DELETE) {
2448 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2449 br->name, iface->name);
2450 iface_destroy(iface);
2451 if (!port->n_ifaces) {
2452 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2453 br->name, port->name);
2459 if (port->n_ifaces > 1) {
2460 bool up = !(opp->state & OFPPS_LINK_DOWN);
2461 bond_link_status_update(iface, up);
2462 port_update_bond_compat(port);
2468 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2469 struct odp_actions *actions, tag_type *tags,
2470 uint16_t *nf_output_iface, void *br_)
2472 struct bridge *br = br_;
2474 COVERAGE_INC(bridge_process_flow);
2475 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2479 bridge_account_flow_ofhook_cb(const flow_t *flow,
2480 const union odp_action *actions,
2481 size_t n_actions, unsigned long long int n_bytes,
2484 struct bridge *br = br_;
2485 const union odp_action *a;
2486 struct port *in_port;
2490 /* Feed information from the active flows back into the learning table
2491 * to ensure that table is always in sync with what is actually flowing
2492 * through the datapath. */
2493 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2494 update_learning_table(br, flow, vlan, in_port);
2497 if (!br->has_bonded_ports) {
2501 for (a = actions; a < &actions[n_actions]; a++) {
2502 if (a->type == ODPAT_OUTPUT) {
2503 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2504 if (out_port && out_port->n_ifaces >= 2) {
2505 struct bond_entry *e = lookup_bond_entry(out_port,
2507 e->tx_bytes += n_bytes;
2514 bridge_account_checkpoint_ofhook_cb(void *br_)
2516 struct bridge *br = br_;
2520 if (!br->has_bonded_ports) {
2525 for (i = 0; i < br->n_ports; i++) {
2526 struct port *port = br->ports[i];
2527 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2528 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2529 bond_rebalance_port(port);
2534 static struct ofhooks bridge_ofhooks = {
2535 bridge_port_changed_ofhook_cb,
2536 bridge_normal_ofhook_cb,
2537 bridge_account_flow_ofhook_cb,
2538 bridge_account_checkpoint_ofhook_cb,
2541 /* Bonding functions. */
2543 /* Statistics for a single interface on a bonded port, used for load-based
2544 * bond rebalancing. */
2545 struct slave_balance {
2546 struct iface *iface; /* The interface. */
2547 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2549 /* All the "bond_entry"s that are assigned to this interface, in order of
2550 * increasing tx_bytes. */
2551 struct bond_entry **hashes;
2555 /* Sorts pointers to pointers to bond_entries in ascending order by the
2556 * interface to which they are assigned, and within a single interface in
2557 * ascending order of bytes transmitted. */
2559 compare_bond_entries(const void *a_, const void *b_)
2561 const struct bond_entry *const *ap = a_;
2562 const struct bond_entry *const *bp = b_;
2563 const struct bond_entry *a = *ap;
2564 const struct bond_entry *b = *bp;
2565 if (a->iface_idx != b->iface_idx) {
2566 return a->iface_idx > b->iface_idx ? 1 : -1;
2567 } else if (a->tx_bytes != b->tx_bytes) {
2568 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2574 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2575 * *descending* order by number of bytes transmitted. */
2577 compare_slave_balance(const void *a_, const void *b_)
2579 const struct slave_balance *a = a_;
2580 const struct slave_balance *b = b_;
2581 if (a->iface->enabled != b->iface->enabled) {
2582 return a->iface->enabled ? -1 : 1;
2583 } else if (a->tx_bytes != b->tx_bytes) {
2584 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2591 swap_bals(struct slave_balance *a, struct slave_balance *b)
2593 struct slave_balance tmp = *a;
2598 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2599 * given that 'p' (and only 'p') might be in the wrong location.
2601 * This function invalidates 'p', since it might now be in a different memory
2604 resort_bals(struct slave_balance *p,
2605 struct slave_balance bals[], size_t n_bals)
2608 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2609 swap_bals(p, p - 1);
2611 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2612 swap_bals(p, p + 1);
2618 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2620 if (VLOG_IS_DBG_ENABLED()) {
2621 struct ds ds = DS_EMPTY_INITIALIZER;
2622 const struct slave_balance *b;
2624 for (b = bals; b < bals + n_bals; b++) {
2628 ds_put_char(&ds, ',');
2630 ds_put_format(&ds, " %s %"PRIu64"kB",
2631 b->iface->name, b->tx_bytes / 1024);
2633 if (!b->iface->enabled) {
2634 ds_put_cstr(&ds, " (disabled)");
2636 if (b->n_hashes > 0) {
2637 ds_put_cstr(&ds, " (");
2638 for (i = 0; i < b->n_hashes; i++) {
2639 const struct bond_entry *e = b->hashes[i];
2641 ds_put_cstr(&ds, " + ");
2643 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2644 e - port->bond_hash, e->tx_bytes / 1024);
2646 ds_put_cstr(&ds, ")");
2649 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2654 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2656 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2659 struct bond_entry *hash = from->hashes[hash_idx];
2660 struct port *port = from->iface->port;
2661 uint64_t delta = hash->tx_bytes;
2663 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2664 "from %s to %s (now carrying %"PRIu64"kB and "
2665 "%"PRIu64"kB load, respectively)",
2666 port->name, delta / 1024, hash - port->bond_hash,
2667 from->iface->name, to->iface->name,
2668 (from->tx_bytes - delta) / 1024,
2669 (to->tx_bytes + delta) / 1024);
2671 /* Delete element from from->hashes.
2673 * We don't bother to add the element to to->hashes because not only would
2674 * it require more work, the only purpose it would be to allow that hash to
2675 * be migrated to another slave in this rebalancing run, and there is no
2676 * point in doing that. */
2677 if (hash_idx == 0) {
2680 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2681 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2685 /* Shift load away from 'from' to 'to'. */
2686 from->tx_bytes -= delta;
2687 to->tx_bytes += delta;
2689 /* Arrange for flows to be revalidated. */
2690 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2691 hash->iface_idx = to->iface->port_ifidx;
2692 hash->iface_tag = tag_create_random();
2696 bond_rebalance_port(struct port *port)
2698 struct slave_balance bals[DP_MAX_PORTS];
2700 struct bond_entry *hashes[BOND_MASK + 1];
2701 struct slave_balance *b, *from, *to;
2702 struct bond_entry *e;
2705 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2706 * descending order of tx_bytes, so that bals[0] represents the most
2707 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2710 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2711 * array for each slave_balance structure, we sort our local array of
2712 * hashes in order by slave, so that all of the hashes for a given slave
2713 * become contiguous in memory, and then we point each 'hashes' members of
2714 * a slave_balance structure to the start of a contiguous group. */
2715 n_bals = port->n_ifaces;
2716 for (b = bals; b < &bals[n_bals]; b++) {
2717 b->iface = port->ifaces[b - bals];
2722 for (i = 0; i <= BOND_MASK; i++) {
2723 hashes[i] = &port->bond_hash[i];
2725 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2726 for (i = 0; i <= BOND_MASK; i++) {
2728 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2729 b = &bals[e->iface_idx];
2730 b->tx_bytes += e->tx_bytes;
2732 b->hashes = &hashes[i];
2737 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2738 log_bals(bals, n_bals, port);
2740 /* Discard slaves that aren't enabled (which were sorted to the back of the
2741 * array earlier). */
2742 while (!bals[n_bals - 1].iface->enabled) {
2749 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2750 to = &bals[n_bals - 1];
2751 for (from = bals; from < to; ) {
2752 uint64_t overload = from->tx_bytes - to->tx_bytes;
2753 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2754 /* The extra load on 'from' (and all less-loaded slaves), compared
2755 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2756 * it is less than ~1Mbps. No point in rebalancing. */
2758 } else if (from->n_hashes == 1) {
2759 /* 'from' only carries a single MAC hash, so we can't shift any
2760 * load away from it, even though we want to. */
2763 /* 'from' is carrying significantly more load than 'to', and that
2764 * load is split across at least two different hashes. Pick a hash
2765 * to migrate to 'to' (the least-loaded slave), given that doing so
2766 * must decrease the ratio of the load on the two slaves by at
2769 * The sort order we use means that we prefer to shift away the
2770 * smallest hashes instead of the biggest ones. There is little
2771 * reason behind this decision; we could use the opposite sort
2772 * order to shift away big hashes ahead of small ones. */
2776 for (i = 0; i < from->n_hashes; i++) {
2777 double old_ratio, new_ratio;
2778 uint64_t delta = from->hashes[i]->tx_bytes;
2780 if (delta == 0 || from->tx_bytes - delta == 0) {
2781 /* Pointless move. */
2785 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2787 if (to->tx_bytes == 0) {
2788 /* Nothing on the new slave, move it. */
2792 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2793 new_ratio = (double)(from->tx_bytes - delta) /
2794 (to->tx_bytes + delta);
2796 if (new_ratio == 0) {
2797 /* Should already be covered but check to prevent division
2802 if (new_ratio < 1) {
2803 new_ratio = 1 / new_ratio;
2806 if (old_ratio - new_ratio > 0.1) {
2807 /* Would decrease the ratio, move it. */
2811 if (i < from->n_hashes) {
2812 bond_shift_load(from, to, i);
2813 port->bond_compat_is_stale = true;
2815 /* If the result of the migration changed the relative order of
2816 * 'from' and 'to' swap them back to maintain invariants. */
2817 if (order_swapped) {
2818 swap_bals(from, to);
2821 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2822 * point to different slave_balance structures. It is only
2823 * valid to do these two operations in a row at all because we
2824 * know that 'from' will not move past 'to' and vice versa. */
2825 resort_bals(from, bals, n_bals);
2826 resort_bals(to, bals, n_bals);
2833 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2834 * historical data to decay to <1% in 7 rebalancing runs. */
2835 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2841 bond_send_learning_packets(struct port *port)
2843 struct bridge *br = port->bridge;
2844 struct mac_entry *e;
2845 struct ofpbuf packet;
2846 int error, n_packets, n_errors;
2848 if (!port->n_ifaces || port->active_iface < 0) {
2852 ofpbuf_init(&packet, 128);
2853 error = n_packets = n_errors = 0;
2854 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2855 union ofp_action actions[2], *a;
2861 if (e->port == port->port_idx
2862 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2866 /* Compose actions. */
2867 memset(actions, 0, sizeof actions);
2870 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2871 a->vlan_vid.len = htons(sizeof *a);
2872 a->vlan_vid.vlan_vid = htons(e->vlan);
2875 a->output.type = htons(OFPAT_OUTPUT);
2876 a->output.len = htons(sizeof *a);
2877 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2882 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2884 flow_extract(&packet, 0, ODPP_NONE, &flow);
2885 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2892 ofpbuf_uninit(&packet);
2895 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2896 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2897 "packets, last error was: %s",
2898 port->name, n_errors, n_packets, strerror(error));
2900 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2901 port->name, n_packets);
2905 /* Bonding unixctl user interface functions. */
2908 bond_unixctl_list(struct unixctl_conn *conn,
2909 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2911 struct ds ds = DS_EMPTY_INITIALIZER;
2912 const struct bridge *br;
2914 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2916 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2919 for (i = 0; i < br->n_ports; i++) {
2920 const struct port *port = br->ports[i];
2921 if (port->n_ifaces > 1) {
2924 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2925 for (j = 0; j < port->n_ifaces; j++) {
2926 const struct iface *iface = port->ifaces[j];
2928 ds_put_cstr(&ds, ", ");
2930 ds_put_cstr(&ds, iface->name);
2932 ds_put_char(&ds, '\n');
2936 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2940 static struct port *
2941 bond_find(const char *name)
2943 const struct bridge *br;
2945 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2948 for (i = 0; i < br->n_ports; i++) {
2949 struct port *port = br->ports[i];
2950 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2959 bond_unixctl_show(struct unixctl_conn *conn,
2960 const char *args, void *aux OVS_UNUSED)
2962 struct ds ds = DS_EMPTY_INITIALIZER;
2963 const struct port *port;
2966 port = bond_find(args);
2968 unixctl_command_reply(conn, 501, "no such bond");
2972 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2973 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2974 ds_put_format(&ds, "next rebalance: %lld ms\n",
2975 port->bond_next_rebalance - time_msec());
2976 for (j = 0; j < port->n_ifaces; j++) {
2977 const struct iface *iface = port->ifaces[j];
2978 struct bond_entry *be;
2981 ds_put_format(&ds, "slave %s: %s\n",
2982 iface->name, iface->enabled ? "enabled" : "disabled");
2983 if (j == port->active_iface) {
2984 ds_put_cstr(&ds, "\tactive slave\n");
2986 if (iface->delay_expires != LLONG_MAX) {
2987 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2988 iface->enabled ? "downdelay" : "updelay",
2989 iface->delay_expires - time_msec());
2993 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2994 int hash = be - port->bond_hash;
2995 struct mac_entry *me;
2997 if (be->iface_idx != j) {
3001 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3002 hash, be->tx_bytes / 1024);
3005 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3006 &port->bridge->ml->lrus) {
3009 if (bond_hash(me->mac) == hash
3010 && me->port != port->port_idx
3011 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3012 && dp_ifidx == iface->dp_ifidx)
3014 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3015 ETH_ADDR_ARGS(me->mac));
3020 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3025 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3026 void *aux OVS_UNUSED)
3028 char *args = (char *) args_;
3029 char *save_ptr = NULL;
3030 char *bond_s, *hash_s, *slave_s;
3031 uint8_t mac[ETH_ADDR_LEN];
3033 struct iface *iface;
3034 struct bond_entry *entry;
3037 bond_s = strtok_r(args, " ", &save_ptr);
3038 hash_s = strtok_r(NULL, " ", &save_ptr);
3039 slave_s = strtok_r(NULL, " ", &save_ptr);
3041 unixctl_command_reply(conn, 501,
3042 "usage: bond/migrate BOND HASH SLAVE");
3046 port = bond_find(bond_s);
3048 unixctl_command_reply(conn, 501, "no such bond");
3052 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3053 == ETH_ADDR_SCAN_COUNT) {
3054 hash = bond_hash(mac);
3055 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3056 hash = atoi(hash_s) & BOND_MASK;
3058 unixctl_command_reply(conn, 501, "bad hash");
3062 iface = port_lookup_iface(port, slave_s);
3064 unixctl_command_reply(conn, 501, "no such slave");
3068 if (!iface->enabled) {
3069 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3073 entry = &port->bond_hash[hash];
3074 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3075 entry->iface_idx = iface->port_ifidx;
3076 entry->iface_tag = tag_create_random();
3077 port->bond_compat_is_stale = true;
3078 unixctl_command_reply(conn, 200, "migrated");
3082 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3083 void *aux OVS_UNUSED)
3085 char *args = (char *) args_;
3086 char *save_ptr = NULL;
3087 char *bond_s, *slave_s;
3089 struct iface *iface;
3091 bond_s = strtok_r(args, " ", &save_ptr);
3092 slave_s = strtok_r(NULL, " ", &save_ptr);
3094 unixctl_command_reply(conn, 501,
3095 "usage: bond/set-active-slave BOND SLAVE");
3099 port = bond_find(bond_s);
3101 unixctl_command_reply(conn, 501, "no such bond");
3105 iface = port_lookup_iface(port, slave_s);
3107 unixctl_command_reply(conn, 501, "no such slave");
3111 if (!iface->enabled) {
3112 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3116 if (port->active_iface != iface->port_ifidx) {
3117 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3118 port->active_iface = iface->port_ifidx;
3119 port->active_iface_tag = tag_create_random();
3120 VLOG_INFO("port %s: active interface is now %s",
3121 port->name, iface->name);
3122 bond_send_learning_packets(port);
3123 unixctl_command_reply(conn, 200, "done");
3125 unixctl_command_reply(conn, 200, "no change");
3130 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3132 char *args = (char *) args_;
3133 char *save_ptr = NULL;
3134 char *bond_s, *slave_s;
3136 struct iface *iface;
3138 bond_s = strtok_r(args, " ", &save_ptr);
3139 slave_s = strtok_r(NULL, " ", &save_ptr);
3141 unixctl_command_reply(conn, 501,
3142 "usage: bond/enable/disable-slave BOND SLAVE");
3146 port = bond_find(bond_s);
3148 unixctl_command_reply(conn, 501, "no such bond");
3152 iface = port_lookup_iface(port, slave_s);
3154 unixctl_command_reply(conn, 501, "no such slave");
3158 bond_enable_slave(iface, enable);
3159 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3163 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3164 void *aux OVS_UNUSED)
3166 enable_slave(conn, args, true);
3170 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3171 void *aux OVS_UNUSED)
3173 enable_slave(conn, args, false);
3177 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3178 void *aux OVS_UNUSED)
3180 uint8_t mac[ETH_ADDR_LEN];
3184 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3185 == ETH_ADDR_SCAN_COUNT) {
3186 hash = bond_hash(mac);
3188 hash_cstr = xasprintf("%u", hash);
3189 unixctl_command_reply(conn, 200, hash_cstr);
3192 unixctl_command_reply(conn, 501, "invalid mac");
3199 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3200 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3201 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3202 unixctl_command_register("bond/set-active-slave",
3203 bond_unixctl_set_active_slave, NULL);
3204 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3206 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3208 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3211 /* Port functions. */
3213 static struct port *
3214 port_create(struct bridge *br, const char *name)
3218 port = xzalloc(sizeof *port);
3220 port->port_idx = br->n_ports;
3222 port->trunks = NULL;
3223 port->name = xstrdup(name);
3224 port->active_iface = -1;
3226 if (br->n_ports >= br->allocated_ports) {
3227 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3230 br->ports[br->n_ports++] = port;
3231 shash_add_assert(&br->port_by_name, port->name, port);
3233 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3240 get_port_other_config(const struct ovsrec_port *port, const char *key,
3241 const char *default_value)
3245 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3247 return value ? value : default_value;
3251 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3253 struct shash new_ifaces;
3256 /* Collect list of new interfaces. */
3257 shash_init(&new_ifaces);
3258 for (i = 0; i < cfg->n_interfaces; i++) {
3259 const char *name = cfg->interfaces[i]->name;
3260 shash_add_once(&new_ifaces, name, NULL);
3263 /* Get rid of deleted interfaces. */
3264 for (i = 0; i < port->n_ifaces; ) {
3265 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3266 iface_destroy(port->ifaces[i]);
3272 shash_destroy(&new_ifaces);
3276 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3278 struct shash new_ifaces;
3279 long long int next_rebalance;
3280 unsigned long *trunks;
3286 /* Update settings. */
3287 port->updelay = cfg->bond_updelay;
3288 if (port->updelay < 0) {
3291 port->updelay = cfg->bond_downdelay;
3292 if (port->downdelay < 0) {
3293 port->downdelay = 0;
3295 port->bond_rebalance_interval = atoi(
3296 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3297 if (port->bond_rebalance_interval < 1000) {
3298 port->bond_rebalance_interval = 1000;
3300 next_rebalance = time_msec() + port->bond_rebalance_interval;
3301 if (port->bond_next_rebalance > next_rebalance) {
3302 port->bond_next_rebalance = next_rebalance;
3305 /* Add new interfaces and update 'cfg' member of existing ones. */
3306 shash_init(&new_ifaces);
3307 for (i = 0; i < cfg->n_interfaces; i++) {
3308 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3309 struct iface *iface;
3311 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3312 VLOG_WARN("port %s: %s specified twice as port interface",
3313 port->name, if_cfg->name);
3317 iface = iface_lookup(port->bridge, if_cfg->name);
3319 if (iface->port != port) {
3320 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3322 port->bridge->name, if_cfg->name, iface->port->name);
3325 iface->cfg = if_cfg;
3327 iface_create(port, if_cfg);
3330 shash_destroy(&new_ifaces);
3335 if (port->n_ifaces < 2) {
3337 if (vlan >= 0 && vlan <= 4095) {
3338 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3343 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3344 * they even work as-is. But they have not been tested. */
3345 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3349 if (port->vlan != vlan) {
3351 bridge_flush(port->bridge);
3354 /* Get trunked VLANs. */
3356 if (vlan < 0 && cfg->n_trunks) {
3360 trunks = bitmap_allocate(4096);
3362 for (i = 0; i < cfg->n_trunks; i++) {
3363 int trunk = cfg->trunks[i];
3365 bitmap_set1(trunks, trunk);
3371 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3372 port->name, cfg->n_trunks);
3374 if (n_errors == cfg->n_trunks) {
3375 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3377 bitmap_free(trunks);
3380 } else if (vlan >= 0 && cfg->n_trunks) {
3381 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3385 ? port->trunks != NULL
3386 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3387 bridge_flush(port->bridge);
3389 bitmap_free(port->trunks);
3390 port->trunks = trunks;
3394 port_destroy(struct port *port)
3397 struct bridge *br = port->bridge;
3401 proc_net_compat_update_vlan(port->name, NULL, 0);
3402 proc_net_compat_update_bond(port->name, NULL);
3404 for (i = 0; i < MAX_MIRRORS; i++) {
3405 struct mirror *m = br->mirrors[i];
3406 if (m && m->out_port == port) {
3411 while (port->n_ifaces > 0) {
3412 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3415 shash_find_and_delete_assert(&br->port_by_name, port->name);
3417 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3418 del->port_idx = port->port_idx;
3421 bitmap_free(port->trunks);
3428 static struct port *
3429 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3431 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3432 return iface ? iface->port : NULL;
3435 static struct port *
3436 port_lookup(const struct bridge *br, const char *name)
3438 return shash_find_data(&br->port_by_name, name);
3441 static struct iface *
3442 port_lookup_iface(const struct port *port, const char *name)
3444 struct iface *iface = iface_lookup(port->bridge, name);
3445 return iface && iface->port == port ? iface : NULL;
3449 port_update_bonding(struct port *port)
3451 if (port->n_ifaces < 2) {
3452 /* Not a bonded port. */
3453 if (port->bond_hash) {
3454 free(port->bond_hash);
3455 port->bond_hash = NULL;
3456 port->bond_compat_is_stale = true;
3457 port->bond_fake_iface = false;
3460 if (!port->bond_hash) {
3463 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3464 for (i = 0; i <= BOND_MASK; i++) {
3465 struct bond_entry *e = &port->bond_hash[i];
3469 port->no_ifaces_tag = tag_create_random();
3470 bond_choose_active_iface(port);
3471 port->bond_next_rebalance
3472 = time_msec() + port->bond_rebalance_interval;
3474 if (port->cfg->bond_fake_iface) {
3475 port->bond_next_fake_iface_update = time_msec();
3478 port->bond_compat_is_stale = true;
3479 port->bond_fake_iface = port->cfg->bond_fake_iface;
3484 port_update_bond_compat(struct port *port)
3486 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3487 struct compat_bond bond;
3490 if (port->n_ifaces < 2) {
3491 proc_net_compat_update_bond(port->name, NULL);
3496 bond.updelay = port->updelay;
3497 bond.downdelay = port->downdelay;
3500 bond.hashes = compat_hashes;
3501 if (port->bond_hash) {
3502 const struct bond_entry *e;
3503 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3504 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3505 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3506 cbh->hash = e - port->bond_hash;
3507 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3512 bond.n_slaves = port->n_ifaces;
3513 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3514 for (i = 0; i < port->n_ifaces; i++) {
3515 struct iface *iface = port->ifaces[i];
3516 struct compat_bond_slave *slave = &bond.slaves[i];
3517 slave->name = iface->name;
3519 /* We need to make the same determination as the Linux bonding
3520 * code to determine whether a slave should be consider "up".
3521 * The Linux function bond_miimon_inspect() supports four
3522 * BOND_LINK_* states:
3524 * - BOND_LINK_UP: carrier detected, updelay has passed.
3525 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3526 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3527 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3529 * The function bond_info_show_slave() only considers BOND_LINK_UP
3530 * to be "up" and anything else to be "down".
3532 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3536 netdev_get_etheraddr(iface->netdev, slave->mac);
3539 if (port->bond_fake_iface) {
3540 struct netdev *bond_netdev;
3542 if (!netdev_open_default(port->name, &bond_netdev)) {
3544 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3546 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3548 netdev_close(bond_netdev);
3552 proc_net_compat_update_bond(port->name, &bond);
3557 port_update_vlan_compat(struct port *port)
3559 struct bridge *br = port->bridge;
3560 char *vlandev_name = NULL;
3562 if (port->vlan > 0) {
3563 /* Figure out the name that the VLAN device should actually have, if it
3564 * existed. This takes some work because the VLAN device would not
3565 * have port->name in its name; rather, it would have the trunk port's
3566 * name, and 'port' would be attached to a bridge that also had the
3567 * VLAN device one of its ports. So we need to find a trunk port that
3568 * includes port->vlan.
3570 * There might be more than one candidate. This doesn't happen on
3571 * XenServer, so if it happens we just pick the first choice in
3572 * alphabetical order instead of creating multiple VLAN devices. */
3574 for (i = 0; i < br->n_ports; i++) {
3575 struct port *p = br->ports[i];
3576 if (port_trunks_vlan(p, port->vlan)
3578 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3580 uint8_t ea[ETH_ADDR_LEN];
3581 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3582 if (!eth_addr_is_multicast(ea) &&
3583 !eth_addr_is_reserved(ea) &&
3584 !eth_addr_is_zero(ea)) {
3585 vlandev_name = p->name;
3590 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3593 /* Interface functions. */
3595 static struct iface *
3596 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3598 struct bridge *br = port->bridge;
3599 struct iface *iface;
3600 char *name = if_cfg->name;
3603 iface = xzalloc(sizeof *iface);
3605 iface->port_ifidx = port->n_ifaces;
3606 iface->name = xstrdup(name);
3607 iface->dp_ifidx = -1;
3608 iface->tag = tag_create_random();
3609 iface->delay_expires = LLONG_MAX;
3610 iface->netdev = NULL;
3611 iface->cfg = if_cfg;
3613 shash_add_assert(&br->iface_by_name, iface->name, iface);
3615 /* Attempt to create the network interface in case it doesn't exist yet. */
3616 if (!iface_is_internal(br, iface->name)) {
3617 error = set_up_iface(if_cfg, iface, true);
3619 VLOG_WARN("could not create iface %s: %s", iface->name,
3622 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3629 if (port->n_ifaces >= port->allocated_ifaces) {
3630 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3631 sizeof *port->ifaces);
3633 port->ifaces[port->n_ifaces++] = iface;
3634 if (port->n_ifaces > 1) {
3635 br->has_bonded_ports = true;
3638 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3646 iface_destroy(struct iface *iface)
3649 struct port *port = iface->port;
3650 struct bridge *br = port->bridge;
3651 bool del_active = port->active_iface == iface->port_ifidx;
3654 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3656 if (iface->dp_ifidx >= 0) {
3657 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3660 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3661 del->port_ifidx = iface->port_ifidx;
3663 netdev_close(iface->netdev);
3666 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3667 bond_choose_active_iface(port);
3668 bond_send_learning_packets(port);
3674 bridge_flush(port->bridge);
3678 static struct iface *
3679 iface_lookup(const struct bridge *br, const char *name)
3681 return shash_find_data(&br->iface_by_name, name);
3684 static struct iface *
3685 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3687 return port_array_get(&br->ifaces, dp_ifidx);
3690 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3691 * 'br', that is, an interface that is entirely simulated within the datapath.
3692 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3693 * interfaces are created by setting "iface.<iface>.internal = true".
3695 * In addition, we have a kluge-y feature that creates an internal port with
3696 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3697 * This feature needs to go away in the long term. Until then, this is one
3698 * reason why this function takes a name instead of a struct iface: the fake
3699 * interfaces created this way do not have a struct iface. */
3701 iface_is_internal(const struct bridge *br, const char *if_name)
3703 struct iface *iface;
3706 if (!strcmp(if_name, br->name)) {
3710 iface = iface_lookup(br, if_name);
3711 if (iface && !strcmp(iface->cfg->type, "internal")) {
3715 port = port_lookup(br, if_name);
3716 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3722 /* Set Ethernet address of 'iface', if one is specified in the configuration
3725 iface_set_mac(struct iface *iface)
3727 uint8_t ea[ETH_ADDR_LEN];
3729 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3730 if (eth_addr_is_multicast(ea)) {
3731 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3733 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3734 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3735 iface->name, iface->name);
3737 int error = netdev_set_etheraddr(iface->netdev, ea);
3739 VLOG_ERR("interface %s: setting MAC failed (%s)",
3740 iface->name, strerror(error));
3747 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3748 struct shash *shash)
3753 for (i = 0; i < n; i++) {
3754 shash_add(shash, keys[i], values[i]);
3758 struct iface_delete_queues_cbdata {
3759 struct netdev *netdev;
3760 const struct ovsdb_datum *queues;
3764 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3766 union ovsdb_atom atom;
3768 atom.integer = target;
3769 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3773 iface_delete_queues(unsigned int queue_id,
3774 const struct shash *details OVS_UNUSED, void *cbdata_)
3776 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3778 if (!queue_ids_include(cbdata->queues, queue_id)) {
3779 netdev_delete_queue(cbdata->netdev, queue_id);
3784 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3786 if (!qos || qos->type[0] == '\0') {
3787 netdev_set_qos(iface->netdev, NULL, NULL);
3789 struct iface_delete_queues_cbdata cbdata;
3790 struct shash details;
3793 /* Configure top-level Qos for 'iface'. */
3794 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3795 qos->n_other_config, &details);
3796 netdev_set_qos(iface->netdev, qos->type, &details);
3797 shash_destroy(&details);
3799 /* Deconfigure queues that were deleted. */
3800 cbdata.netdev = iface->netdev;
3801 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3803 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3805 /* Configure queues for 'iface'. */
3806 for (i = 0; i < qos->n_queues; i++) {
3807 const struct ovsrec_queue *queue = qos->value_queues[i];
3808 unsigned int queue_id = qos->key_queues[i];
3810 shash_from_ovs_idl_map(queue->key_other_config,
3811 queue->value_other_config,
3812 queue->n_other_config, &details);
3813 netdev_set_queue(iface->netdev, queue_id, &details);
3814 shash_destroy(&details);
3819 /* Port mirroring. */
3821 static struct mirror *
3822 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3826 for (i = 0; i < MAX_MIRRORS; i++) {
3827 struct mirror *m = br->mirrors[i];
3828 if (m && uuid_equals(uuid, &m->uuid)) {
3836 mirror_reconfigure(struct bridge *br)
3838 unsigned long *rspan_vlans;
3841 /* Get rid of deleted mirrors. */
3842 for (i = 0; i < MAX_MIRRORS; i++) {
3843 struct mirror *m = br->mirrors[i];
3845 const struct ovsdb_datum *mc;
3846 union ovsdb_atom atom;
3848 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3849 atom.uuid = br->mirrors[i]->uuid;
3850 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3856 /* Add new mirrors and reconfigure existing ones. */
3857 for (i = 0; i < br->cfg->n_mirrors; i++) {
3858 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3859 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3861 mirror_reconfigure_one(m, cfg);
3863 mirror_create(br, cfg);
3867 /* Update port reserved status. */
3868 for (i = 0; i < br->n_ports; i++) {
3869 br->ports[i]->is_mirror_output_port = false;
3871 for (i = 0; i < MAX_MIRRORS; i++) {
3872 struct mirror *m = br->mirrors[i];
3873 if (m && m->out_port) {
3874 m->out_port->is_mirror_output_port = true;
3878 /* Update flooded vlans (for RSPAN). */
3880 if (br->cfg->n_flood_vlans) {
3881 rspan_vlans = bitmap_allocate(4096);
3883 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3884 int64_t vlan = br->cfg->flood_vlans[i];
3885 if (vlan >= 0 && vlan < 4096) {
3886 bitmap_set1(rspan_vlans, vlan);
3887 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3890 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3895 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3901 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3906 for (i = 0; ; i++) {
3907 if (i >= MAX_MIRRORS) {
3908 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3909 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3912 if (!br->mirrors[i]) {
3917 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
3920 br->mirrors[i] = m = xzalloc(sizeof *m);
3923 m->name = xstrdup(cfg->name);
3924 shash_init(&m->src_ports);
3925 shash_init(&m->dst_ports);
3931 mirror_reconfigure_one(m, cfg);
3935 mirror_destroy(struct mirror *m)
3938 struct bridge *br = m->bridge;
3941 for (i = 0; i < br->n_ports; i++) {
3942 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3943 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3946 shash_destroy(&m->src_ports);
3947 shash_destroy(&m->dst_ports);
3950 m->bridge->mirrors[m->idx] = NULL;
3959 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3960 struct shash *names)
3964 for (i = 0; i < n_ports; i++) {
3965 const char *name = ports[i]->name;
3966 if (port_lookup(m->bridge, name)) {
3967 shash_add_once(names, name, NULL);
3969 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3970 "port %s", m->bridge->name, m->name, name);
3976 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3982 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3984 for (i = 0; i < cfg->n_select_vlan; i++) {
3985 int64_t vlan = cfg->select_vlan[i];
3986 if (vlan < 0 || vlan > 4095) {
3987 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3988 m->bridge->name, m->name, vlan);
3990 (*vlans)[n_vlans++] = vlan;
3997 vlan_is_mirrored(const struct mirror *m, int vlan)
4001 for (i = 0; i < m->n_vlans; i++) {
4002 if (m->vlans[i] == vlan) {
4010 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4014 for (i = 0; i < m->n_vlans; i++) {
4015 if (port_trunks_vlan(p, m->vlans[i])) {
4023 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4025 struct shash src_ports, dst_ports;
4026 mirror_mask_t mirror_bit;
4027 struct port *out_port;
4034 if (strcmp(cfg->name, m->name)) {
4036 m->name = xstrdup(cfg->name);
4039 /* Get output port. */
4040 if (cfg->output_port) {
4041 out_port = port_lookup(m->bridge, cfg->output_port->name);
4043 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4044 m->bridge->name, m->name);
4050 if (cfg->output_vlan) {
4051 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4052 "output vlan; ignoring output vlan",
4053 m->bridge->name, m->name);
4055 } else if (cfg->output_vlan) {
4057 out_vlan = *cfg->output_vlan;
4059 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4060 m->bridge->name, m->name);
4065 shash_init(&src_ports);
4066 shash_init(&dst_ports);
4067 if (cfg->select_all) {
4068 for (i = 0; i < m->bridge->n_ports; i++) {
4069 const char *name = m->bridge->ports[i]->name;
4070 shash_add_once(&src_ports, name, NULL);
4071 shash_add_once(&dst_ports, name, NULL);
4076 /* Get ports, and drop duplicates and ports that don't exist. */
4077 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4079 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4082 /* Get all the vlans, and drop duplicate and invalid vlans. */
4083 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4086 /* Update mirror data. */
4087 if (!shash_equal_keys(&m->src_ports, &src_ports)
4088 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4089 || m->n_vlans != n_vlans
4090 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4091 || m->out_port != out_port
4092 || m->out_vlan != out_vlan) {
4093 bridge_flush(m->bridge);
4095 shash_swap(&m->src_ports, &src_ports);
4096 shash_swap(&m->dst_ports, &dst_ports);
4099 m->n_vlans = n_vlans;
4100 m->out_port = out_port;
4101 m->out_vlan = out_vlan;
4104 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4105 for (i = 0; i < m->bridge->n_ports; i++) {
4106 struct port *port = m->bridge->ports[i];
4108 if (shash_find(&m->src_ports, port->name)
4111 ? port_trunks_any_mirrored_vlan(m, port)
4112 : vlan_is_mirrored(m, port->vlan)))) {
4113 port->src_mirrors |= mirror_bit;
4115 port->src_mirrors &= ~mirror_bit;
4118 if (shash_find(&m->dst_ports, port->name)) {
4119 port->dst_mirrors |= mirror_bit;
4121 port->dst_mirrors &= ~mirror_bit;
4126 shash_destroy(&src_ports);
4127 shash_destroy(&dst_ports);