1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
50 #include "poll-loop.h"
51 #include "port-array.h"
52 #include "proc-net-compat.h"
56 #include "socket-util.h"
57 #include "stream-ssl.h"
63 #include "vswitchd/vswitch-idl.h"
64 #include "xenserver.h"
66 #include "sflow_api.h"
68 #define THIS_MODULE VLM_bridge
77 /* These members are always valid. */
78 struct port *port; /* Containing port. */
79 size_t port_ifidx; /* Index within containing port. */
80 char *name; /* Host network device name. */
81 tag_type tag; /* Tag associated with this interface. */
82 long long delay_expires; /* Time after which 'enabled' may change. */
84 /* These members are valid only after bridge_reconfigure() causes them to
86 int dp_ifidx; /* Index within kernel datapath. */
87 struct netdev *netdev; /* Network device. */
88 bool enabled; /* May be chosen for flows? */
89 const struct ovsrec_interface *cfg;
92 #define BOND_MASK 0xff
94 int iface_idx; /* Index of assigned iface, or -1 if none. */
95 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
96 tag_type iface_tag; /* Tag associated with iface_idx. */
99 #define MAX_MIRRORS 32
100 typedef uint32_t mirror_mask_t;
101 #define MIRROR_MASK_C(X) UINT32_C(X)
102 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
104 struct bridge *bridge;
108 /* Selection criteria. */
109 struct shash src_ports; /* Name is port name; data is always NULL. */
110 struct shash dst_ports; /* Name is port name; data is always NULL. */
115 struct port *out_port;
119 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
121 struct bridge *bridge;
123 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
124 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
125 * NULL if all VLANs are trunked. */
126 const struct ovsrec_port *cfg;
129 /* An ordinary bridge port has 1 interface.
130 * A bridge port for bonding has at least 2 interfaces. */
131 struct iface **ifaces;
132 size_t n_ifaces, allocated_ifaces;
135 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
136 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
137 tag_type active_iface_tag; /* Tag for bcast flows. */
138 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
139 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
140 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
141 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
142 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
143 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
144 long long int bond_next_rebalance; /* Next rebalancing time. */
146 /* Port mirroring info. */
147 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
148 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
149 bool is_mirror_output_port; /* Does port mirroring send frames here? */
152 #define DP_MAX_PORTS 255
154 struct list node; /* Node in global list of bridges. */
155 char *name; /* User-specified arbitrary name. */
156 struct mac_learning *ml; /* MAC learning table. */
157 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
158 const struct ovsrec_bridge *cfg;
160 /* OpenFlow switch processing. */
161 struct ofproto *ofproto; /* OpenFlow switch. */
163 /* Kernel datapath information. */
164 struct dpif *dpif; /* Datapath. */
165 struct port_array ifaces; /* Indexed by kernel datapath port number. */
169 size_t n_ports, allocated_ports;
170 struct shash iface_by_name; /* "struct iface"s indexed by name. */
171 struct shash port_by_name; /* "struct port"s indexed by name. */
174 bool has_bonded_ports;
179 /* Port mirroring. */
180 struct mirror *mirrors[MAX_MIRRORS];
183 /* List of all bridges. */
184 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
186 /* OVSDB IDL used to obtain configuration. */
187 static struct ovsdb_idl *idl;
189 /* Each time this timer expires, the bridge fetches statistics for every
190 * interface and pushes them into the database. */
191 #define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
192 static long long int iface_stats_timer = LLONG_MIN;
194 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
195 static void bridge_destroy(struct bridge *);
196 static struct bridge *bridge_lookup(const char *name);
197 static unixctl_cb_func bridge_unixctl_dump_flows;
198 static int bridge_run_one(struct bridge *);
199 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
200 const struct bridge *br,
201 struct ovsrec_controller ***controllersp);
202 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
204 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
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 struct mirror *mirror_create(struct bridge *, const char *name);
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,
278 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
279 * but for which the ovs-vswitchd configuration 'cfg' is required. */
281 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
283 static bool already_configured_once;
284 struct svec bridge_names;
285 struct svec dpif_names, dpif_types;
288 /* Only do this once per ovs-vswitchd run. */
289 if (already_configured_once) {
292 already_configured_once = true;
294 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
296 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
297 svec_init(&bridge_names);
298 for (i = 0; i < cfg->n_bridges; i++) {
299 svec_add(&bridge_names, cfg->bridges[i]->name);
301 svec_sort(&bridge_names);
303 /* Iterate over all system dpifs and delete any of them that do not appear
305 svec_init(&dpif_names);
306 svec_init(&dpif_types);
307 dp_enumerate_types(&dpif_types);
308 for (i = 0; i < dpif_types.n; i++) {
313 dp_enumerate_names(dpif_types.names[i], &dpif_names);
315 /* For each dpif... */
316 for (j = 0; j < dpif_names.n; j++) {
317 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
319 struct svec all_names;
322 /* ...check whether any of its names is in 'bridge_names'. */
323 svec_init(&all_names);
324 dpif_get_all_names(dpif, &all_names);
325 for (k = 0; k < all_names.n; k++) {
326 if (svec_contains(&bridge_names, all_names.names[k])) {
331 /* No. Delete the dpif. */
335 svec_destroy(&all_names);
340 svec_destroy(&bridge_names);
341 svec_destroy(&dpif_names);
342 svec_destroy(&dpif_types);
347 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
349 /* XXX SSL should be configurable on a per-bridge basis. */
351 stream_ssl_set_private_key_file(ssl->private_key);
352 stream_ssl_set_certificate_file(ssl->certificate);
353 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
358 /* Attempt to create the network device 'iface_name' through the netdev
361 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
364 struct shash options;
368 shash_init(&options);
369 for (i = 0; i < iface_cfg->n_options; i++) {
370 shash_add(&options, iface_cfg->key_options[i],
371 xstrdup(iface_cfg->value_options[i]));
375 struct netdev_options netdev_options;
377 memset(&netdev_options, 0, sizeof netdev_options);
378 netdev_options.name = iface_cfg->name;
379 if (!strcmp(iface_cfg->type, "internal")) {
380 /* An "internal" config type maps to a netdev "system" type. */
381 netdev_options.type = "system";
383 netdev_options.type = iface_cfg->type;
385 netdev_options.args = &options;
386 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
388 error = netdev_open(&netdev_options, &iface->netdev);
391 netdev_get_carrier(iface->netdev, &iface->enabled);
393 } else if (iface->netdev) {
394 const char *netdev_type = netdev_get_type(iface->netdev);
395 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
396 ? iface_cfg->type : NULL;
398 /* An "internal" config type maps to a netdev "system" type. */
399 if (iface_type && !strcmp(iface_type, "internal")) {
400 iface_type = "system";
403 if (!iface_type || !strcmp(netdev_type, iface_type)) {
404 error = netdev_reconfigure(iface->netdev, &options);
406 VLOG_WARN("%s: attempting change device type from %s to %s",
407 iface_cfg->name, netdev_type, iface_type);
411 shash_destroy_free_data(&options);
417 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
419 return set_up_iface(iface_cfg, iface, false);
423 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
424 void *aux OVS_UNUSED)
426 if (!iface->netdev) {
427 int error = set_up_iface(iface->cfg, iface, true);
429 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
439 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
440 void *aux OVS_UNUSED)
442 if (iface->dp_ifidx >= 0) {
443 VLOG_DBG("%s has interface %s on port %d",
445 iface->name, iface->dp_ifidx);
448 VLOG_ERR("%s interface not in %s, dropping",
449 iface->name, dpif_name(br->dpif));
455 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
456 void *aux OVS_UNUSED)
458 /* Set policing attributes. */
459 netdev_set_policing(iface->netdev,
460 iface->cfg->ingress_policing_rate,
461 iface->cfg->ingress_policing_burst);
463 /* Set MAC address of internal interfaces other than the local
465 if (iface->dp_ifidx != ODPP_LOCAL
466 && iface_is_internal(br, iface->name)) {
467 iface_set_mac(iface);
473 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
474 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
475 * deletes from 'br' any ports that no longer have any interfaces. */
477 iterate_and_prune_ifaces(struct bridge *br,
478 bool (*cb)(struct bridge *, struct iface *,
484 for (i = 0; i < br->n_ports; ) {
485 struct port *port = br->ports[i];
486 for (j = 0; j < port->n_ifaces; ) {
487 struct iface *iface = port->ifaces[j];
488 if (cb(br, iface, aux)) {
491 iface_destroy(iface);
495 if (port->n_ifaces) {
498 VLOG_ERR("%s port has no interfaces, dropping", port->name);
504 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
505 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
506 * responsible for freeing '*managersp' (with free()).
508 * You may be asking yourself "why does ovs-vswitchd care?", because
509 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
510 * should not be and in fact is not directly involved in that. But
511 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
512 * it has to tell in-band control where the managers are to enable that.
515 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
516 struct sockaddr_in **managersp, size_t *n_managersp)
518 struct sockaddr_in *managers = NULL;
519 size_t n_managers = 0;
521 if (ovs_cfg->n_managers > 0) {
524 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
525 for (i = 0; i < ovs_cfg->n_managers; i++) {
526 const char *name = ovs_cfg->managers[i];
527 struct sockaddr_in *sin = &managers[i];
529 if ((!strncmp(name, "tcp:", 4)
530 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
531 (!strncmp(name, "ssl:", 4)
532 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
538 *managersp = managers;
539 *n_managersp = n_managers;
543 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
545 struct shash old_br, new_br;
546 struct shash_node *node;
547 struct bridge *br, *next;
548 struct sockaddr_in *managers;
551 int sflow_bridge_number;
553 COVERAGE_INC(bridge_reconfigure);
555 collect_managers(ovs_cfg, &managers, &n_managers);
557 /* Collect old and new bridges. */
560 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
561 shash_add(&old_br, br->name, br);
563 for (i = 0; i < ovs_cfg->n_bridges; i++) {
564 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
565 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
566 VLOG_WARN("more than one bridge named %s", br_cfg->name);
570 /* Get rid of deleted bridges and add new bridges. */
571 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
572 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
579 SHASH_FOR_EACH (node, &new_br) {
580 const char *br_name = node->name;
581 const struct ovsrec_bridge *br_cfg = node->data;
582 br = shash_find_data(&old_br, br_name);
584 /* If the bridge datapath type has changed, we need to tear it
585 * down and recreate. */
586 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
588 bridge_create(br_cfg);
591 bridge_create(br_cfg);
594 shash_destroy(&old_br);
595 shash_destroy(&new_br);
599 bridge_configure_ssl(ovs_cfg->ssl);
602 /* Reconfigure all bridges. */
603 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
604 bridge_reconfigure_one(ovs_cfg, br);
607 /* Add and delete ports on all datapaths.
609 * The kernel will reject any attempt to add a given port to a datapath if
610 * that port already belongs to a different datapath, so we must do all
611 * port deletions before any port additions. */
612 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
613 struct odp_port *dpif_ports;
615 struct shash want_ifaces;
617 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
618 bridge_get_all_ifaces(br, &want_ifaces);
619 for (i = 0; i < n_dpif_ports; i++) {
620 const struct odp_port *p = &dpif_ports[i];
621 if (!shash_find(&want_ifaces, p->devname)
622 && strcmp(p->devname, br->name)) {
623 int retval = dpif_port_del(br->dpif, p->port);
625 VLOG_ERR("failed to remove %s interface from %s: %s",
626 p->devname, dpif_name(br->dpif),
631 shash_destroy(&want_ifaces);
634 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
635 struct odp_port *dpif_ports;
637 struct shash cur_ifaces, want_ifaces;
638 struct shash_node *node;
640 /* Get the set of interfaces currently in this datapath. */
641 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
642 shash_init(&cur_ifaces);
643 for (i = 0; i < n_dpif_ports; i++) {
644 const char *name = dpif_ports[i].devname;
645 if (!shash_find(&cur_ifaces, name)) {
646 shash_add(&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(ovs_cfg, 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(ovs_cfg, 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 char *key, char **keys, char **values, size_t n)
849 for (i = 0; i < n; i++) {
850 if (!strcmp(keys[i], key)) {
858 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
860 return get_ovsrec_key_value(key,
861 br_cfg->key_other_config,
862 br_cfg->value_other_config,
863 br_cfg->n_other_config);
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);
1357 bridge_run_one(struct bridge *br)
1361 error = ofproto_run1(br->ofproto);
1366 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1369 error = ofproto_run2(br->ofproto, br->flush);
1376 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1377 const struct bridge *br,
1378 struct ovsrec_controller ***controllersp)
1380 struct ovsrec_controller **controllers;
1381 size_t n_controllers;
1383 if (br->cfg->n_controller) {
1384 controllers = br->cfg->controller;
1385 n_controllers = br->cfg->n_controller;
1387 controllers = ovs_cfg->controller;
1388 n_controllers = ovs_cfg->n_controller;
1391 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1397 *controllersp = controllers;
1399 return n_controllers;
1403 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1406 struct shash old_ports, new_ports;
1407 struct svec listeners, old_listeners;
1408 struct svec snoops, old_snoops;
1409 struct shash_node *node;
1412 /* Collect old ports. */
1413 shash_init(&old_ports);
1414 for (i = 0; i < br->n_ports; i++) {
1415 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1418 /* Collect new ports. */
1419 shash_init(&new_ports);
1420 for (i = 0; i < br->cfg->n_ports; i++) {
1421 const char *name = br->cfg->ports[i]->name;
1422 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1423 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1428 /* If we have a controller, then we need a local port. Complain if the
1429 * user didn't specify one.
1431 * XXX perhaps we should synthesize a port ourselves in this case. */
1432 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1433 char local_name[IF_NAMESIZE];
1436 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1437 local_name, sizeof local_name);
1438 if (!error && !shash_find(&new_ports, local_name)) {
1439 VLOG_WARN("bridge %s: controller specified but no local port "
1440 "(port named %s) defined",
1441 br->name, local_name);
1445 /* Get rid of deleted ports.
1446 * Get rid of deleted interfaces on ports that still exist. */
1447 SHASH_FOR_EACH (node, &old_ports) {
1448 struct port *port = node->data;
1449 const struct ovsrec_port *port_cfg;
1451 port_cfg = shash_find_data(&new_ports, node->name);
1455 port_del_ifaces(port, port_cfg);
1459 /* Create new ports.
1460 * Add new interfaces to existing ports.
1461 * Reconfigure existing ports. */
1462 SHASH_FOR_EACH (node, &new_ports) {
1463 struct port *port = shash_find_data(&old_ports, node->name);
1465 port = port_create(br, node->name);
1468 port_reconfigure(port, node->data);
1469 if (!port->n_ifaces) {
1470 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1471 br->name, port->name);
1475 shash_destroy(&old_ports);
1476 shash_destroy(&new_ports);
1478 /* Delete all flows if we're switching from connected to standalone or vice
1479 * versa. (XXX Should we delete all flows if we are switching from one
1480 * controller to another?) */
1482 /* Configure OpenFlow management listener. */
1483 svec_init(&listeners);
1484 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1485 ovs_rundir, br->name));
1486 svec_init(&old_listeners);
1487 ofproto_get_listeners(br->ofproto, &old_listeners);
1488 if (!svec_equal(&listeners, &old_listeners)) {
1489 ofproto_set_listeners(br->ofproto, &listeners);
1491 svec_destroy(&listeners);
1492 svec_destroy(&old_listeners);
1494 /* Configure OpenFlow controller connection snooping. */
1496 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1497 ovs_rundir, br->name));
1498 svec_init(&old_snoops);
1499 ofproto_get_snoops(br->ofproto, &old_snoops);
1500 if (!svec_equal(&snoops, &old_snoops)) {
1501 ofproto_set_snoops(br->ofproto, &snoops);
1503 svec_destroy(&snoops);
1504 svec_destroy(&old_snoops);
1506 mirror_reconfigure(br);
1510 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1512 const struct sockaddr_in *managers,
1515 struct ovsrec_controller **controllers;
1516 size_t n_controllers;
1518 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1520 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1521 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1522 ofproto_flush_flows(br->ofproto);
1525 if (!n_controllers) {
1526 union ofp_action action;
1529 /* Clear out controllers. */
1530 ofproto_set_controllers(br->ofproto, NULL, 0);
1532 /* Set up a flow that matches every packet and directs them to
1533 * OFPP_NORMAL (which goes to us). */
1534 memset(&action, 0, sizeof action);
1535 action.type = htons(OFPAT_OUTPUT);
1536 action.output.len = htons(sizeof action);
1537 action.output.port = htons(OFPP_NORMAL);
1538 memset(&flow, 0, sizeof flow);
1539 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1541 struct ofproto_controller *ocs;
1544 ocs = xmalloc(n_controllers * sizeof *ocs);
1545 for (i = 0; i < n_controllers; i++) {
1546 struct ovsrec_controller *c = controllers[i];
1547 struct ofproto_controller *oc = &ocs[i];
1549 if (strcmp(c->target, "discover")) {
1550 struct iface *local_iface;
1553 local_iface = bridge_get_local_iface(br);
1554 if (local_iface && c->local_ip
1555 && inet_aton(c->local_ip, &ip)) {
1556 struct netdev *netdev = local_iface->netdev;
1557 struct in_addr mask, gateway;
1559 if (!c->local_netmask
1560 || !inet_aton(c->local_netmask, &mask)) {
1563 if (!c->local_gateway
1564 || !inet_aton(c->local_gateway, &gateway)) {
1568 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1570 mask.s_addr = guess_netmask(ip.s_addr);
1572 if (!netdev_set_in4(netdev, ip, mask)) {
1573 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1575 br->name, IP_ARGS(&ip.s_addr),
1576 IP_ARGS(&mask.s_addr));
1579 if (gateway.s_addr) {
1580 if (!netdev_add_router(netdev, gateway)) {
1581 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1582 br->name, IP_ARGS(&gateway.s_addr));
1588 oc->target = c->target;
1589 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1590 oc->probe_interval = (c->inactivity_probe
1591 ? *c->inactivity_probe / 1000 : 5);
1592 oc->fail = (!c->fail_mode
1593 || !strcmp(c->fail_mode, "standalone")
1594 || !strcmp(c->fail_mode, "open")
1595 ? OFPROTO_FAIL_STANDALONE
1596 : OFPROTO_FAIL_SECURE);
1597 oc->band = (!c->connection_mode
1598 || !strcmp(c->connection_mode, "in-band")
1600 : OFPROTO_OUT_OF_BAND);
1601 oc->accept_re = c->discover_accept_regex;
1602 oc->update_resolv_conf = c->discover_update_resolv_conf;
1603 oc->rate_limit = (c->controller_rate_limit
1604 ? *c->controller_rate_limit : 0);
1605 oc->burst_limit = (c->controller_burst_limit
1606 ? *c->controller_burst_limit : 0);
1608 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1614 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1619 for (i = 0; i < br->n_ports; i++) {
1620 struct port *port = br->ports[i];
1621 for (j = 0; j < port->n_ifaces; j++) {
1622 struct iface *iface = port->ifaces[j];
1623 shash_add_once(ifaces, iface->name, iface);
1625 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1626 shash_add_once(ifaces, port->name, NULL);
1631 /* For robustness, in case the administrator moves around datapath ports behind
1632 * our back, we re-check all the datapath port numbers here.
1634 * This function will set the 'dp_ifidx' members of interfaces that have
1635 * disappeared to -1, so only call this function from a context where those
1636 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1637 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1638 * datapath, which doesn't support UINT16_MAX+1 ports. */
1640 bridge_fetch_dp_ifaces(struct bridge *br)
1642 struct odp_port *dpif_ports;
1643 size_t n_dpif_ports;
1646 /* Reset all interface numbers. */
1647 for (i = 0; i < br->n_ports; i++) {
1648 struct port *port = br->ports[i];
1649 for (j = 0; j < port->n_ifaces; j++) {
1650 struct iface *iface = port->ifaces[j];
1651 iface->dp_ifidx = -1;
1654 port_array_clear(&br->ifaces);
1656 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1657 for (i = 0; i < n_dpif_ports; i++) {
1658 struct odp_port *p = &dpif_ports[i];
1659 struct iface *iface = iface_lookup(br, p->devname);
1661 if (iface->dp_ifidx >= 0) {
1662 VLOG_WARN("%s reported interface %s twice",
1663 dpif_name(br->dpif), p->devname);
1664 } else if (iface_from_dp_ifidx(br, p->port)) {
1665 VLOG_WARN("%s reported interface %"PRIu16" twice",
1666 dpif_name(br->dpif), p->port);
1668 port_array_set(&br->ifaces, p->port, iface);
1669 iface->dp_ifidx = p->port;
1673 int64_t ofport = (iface->dp_ifidx >= 0
1674 ? odp_port_to_ofp_port(iface->dp_ifidx)
1676 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1683 /* Bridge packet processing functions. */
1686 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1688 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1691 static struct bond_entry *
1692 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1694 return &port->bond_hash[bond_hash(mac)];
1698 bond_choose_iface(const struct port *port)
1700 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1701 size_t i, best_down_slave = -1;
1702 long long next_delay_expiration = LLONG_MAX;
1704 for (i = 0; i < port->n_ifaces; i++) {
1705 struct iface *iface = port->ifaces[i];
1707 if (iface->enabled) {
1709 } else if (iface->delay_expires < next_delay_expiration) {
1710 best_down_slave = i;
1711 next_delay_expiration = iface->delay_expires;
1715 if (best_down_slave != -1) {
1716 struct iface *iface = port->ifaces[best_down_slave];
1718 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1719 "since no other interface is up", iface->name,
1720 iface->delay_expires - time_msec());
1721 bond_enable_slave(iface, true);
1724 return best_down_slave;
1728 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1729 uint16_t *dp_ifidx, tag_type *tags)
1731 struct iface *iface;
1733 assert(port->n_ifaces);
1734 if (port->n_ifaces == 1) {
1735 iface = port->ifaces[0];
1737 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1738 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1739 || !port->ifaces[e->iface_idx]->enabled) {
1740 /* XXX select interface properly. The current interface selection
1741 * is only good for testing the rebalancing code. */
1742 e->iface_idx = bond_choose_iface(port);
1743 if (e->iface_idx < 0) {
1744 *tags |= port->no_ifaces_tag;
1747 e->iface_tag = tag_create_random();
1748 ((struct port *) port)->bond_compat_is_stale = true;
1750 *tags |= e->iface_tag;
1751 iface = port->ifaces[e->iface_idx];
1753 *dp_ifidx = iface->dp_ifidx;
1754 *tags |= iface->tag; /* Currently only used for bonding. */
1759 bond_link_status_update(struct iface *iface, bool carrier)
1761 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1762 struct port *port = iface->port;
1764 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1765 /* Nothing to do. */
1768 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1769 iface->name, carrier ? "detected" : "dropped");
1770 if (carrier == iface->enabled) {
1771 iface->delay_expires = LLONG_MAX;
1772 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1773 iface->name, carrier ? "disabled" : "enabled");
1774 } else if (carrier && port->active_iface < 0) {
1775 bond_enable_slave(iface, true);
1776 if (port->updelay) {
1777 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1778 "other interface is up", iface->name, port->updelay);
1781 int delay = carrier ? port->updelay : port->downdelay;
1782 iface->delay_expires = time_msec() + delay;
1785 "interface %s: will be %s if it stays %s for %d ms",
1787 carrier ? "enabled" : "disabled",
1788 carrier ? "up" : "down",
1795 bond_choose_active_iface(struct port *port)
1797 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1799 port->active_iface = bond_choose_iface(port);
1800 port->active_iface_tag = tag_create_random();
1801 if (port->active_iface >= 0) {
1802 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1803 port->name, port->ifaces[port->active_iface]->name);
1805 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1811 bond_enable_slave(struct iface *iface, bool enable)
1813 struct port *port = iface->port;
1814 struct bridge *br = port->bridge;
1816 /* This acts as a recursion check. If the act of disabling a slave
1817 * causes a different slave to be enabled, the flag will allow us to
1818 * skip redundant work when we reenter this function. It must be
1819 * cleared on exit to keep things safe with multiple bonds. */
1820 static bool moving_active_iface = false;
1822 iface->delay_expires = LLONG_MAX;
1823 if (enable == iface->enabled) {
1827 iface->enabled = enable;
1828 if (!iface->enabled) {
1829 VLOG_WARN("interface %s: disabled", iface->name);
1830 ofproto_revalidate(br->ofproto, iface->tag);
1831 if (iface->port_ifidx == port->active_iface) {
1832 ofproto_revalidate(br->ofproto,
1833 port->active_iface_tag);
1835 /* Disabling a slave can lead to another slave being immediately
1836 * enabled if there will be no active slaves but one is waiting
1837 * on an updelay. In this case we do not need to run most of the
1838 * code for the newly enabled slave since there was no period
1839 * without an active slave and it is redundant with the disabling
1841 moving_active_iface = true;
1842 bond_choose_active_iface(port);
1844 bond_send_learning_packets(port);
1846 VLOG_WARN("interface %s: enabled", iface->name);
1847 if (port->active_iface < 0 && !moving_active_iface) {
1848 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1849 bond_choose_active_iface(port);
1850 bond_send_learning_packets(port);
1852 iface->tag = tag_create_random();
1855 moving_active_iface = false;
1856 port->bond_compat_is_stale = true;
1859 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1860 * bond interface. */
1862 bond_update_fake_iface_stats(struct port *port)
1864 struct netdev_stats bond_stats;
1865 struct netdev *bond_dev;
1868 memset(&bond_stats, 0, sizeof bond_stats);
1870 for (i = 0; i < port->n_ifaces; i++) {
1871 struct netdev_stats slave_stats;
1873 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1874 /* XXX: We swap the stats here because they are swapped back when
1875 * reported by the internal device. The reason for this is
1876 * internal devices normally represent packets going into the system
1877 * but when used as fake bond device they represent packets leaving
1878 * the system. We really should do this in the internal device
1879 * itself because changing it here reverses the counts from the
1880 * perspective of the switch. However, the internal device doesn't
1881 * know what type of device it represents so we have to do it here
1883 bond_stats.tx_packets += slave_stats.rx_packets;
1884 bond_stats.tx_bytes += slave_stats.rx_bytes;
1885 bond_stats.rx_packets += slave_stats.tx_packets;
1886 bond_stats.rx_bytes += slave_stats.tx_bytes;
1890 if (!netdev_open_default(port->name, &bond_dev)) {
1891 netdev_set_stats(bond_dev, &bond_stats);
1892 netdev_close(bond_dev);
1897 bond_run(struct bridge *br)
1901 for (i = 0; i < br->n_ports; i++) {
1902 struct port *port = br->ports[i];
1904 if (port->n_ifaces >= 2) {
1905 for (j = 0; j < port->n_ifaces; j++) {
1906 struct iface *iface = port->ifaces[j];
1907 if (time_msec() >= iface->delay_expires) {
1908 bond_enable_slave(iface, !iface->enabled);
1912 if (port->bond_fake_iface
1913 && time_msec() >= port->bond_next_fake_iface_update) {
1914 bond_update_fake_iface_stats(port);
1915 port->bond_next_fake_iface_update = time_msec() + 1000;
1919 if (port->bond_compat_is_stale) {
1920 port->bond_compat_is_stale = false;
1921 port_update_bond_compat(port);
1927 bond_wait(struct bridge *br)
1931 for (i = 0; i < br->n_ports; i++) {
1932 struct port *port = br->ports[i];
1933 if (port->n_ifaces < 2) {
1936 for (j = 0; j < port->n_ifaces; j++) {
1937 struct iface *iface = port->ifaces[j];
1938 if (iface->delay_expires != LLONG_MAX) {
1939 poll_timer_wait_until(iface->delay_expires);
1942 if (port->bond_fake_iface) {
1943 poll_timer_wait_until(port->bond_next_fake_iface_update);
1949 set_dst(struct dst *p, const flow_t *flow,
1950 const struct port *in_port, const struct port *out_port,
1953 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1954 : in_port->vlan >= 0 ? in_port->vlan
1955 : ntohs(flow->dl_vlan));
1956 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1960 swap_dst(struct dst *p, struct dst *q)
1962 struct dst tmp = *p;
1967 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1968 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1969 * that we push to the datapath. We could in fact fully sort the array by
1970 * vlan, but in most cases there are at most two different vlan tags so that's
1971 * possibly overkill.) */
1973 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1975 struct dst *first = dsts;
1976 struct dst *last = dsts + n_dsts;
1978 while (first != last) {
1980 * - All dsts < first have vlan == 'vlan'.
1981 * - All dsts >= last have vlan != 'vlan'.
1982 * - first < last. */
1983 while (first->vlan == vlan) {
1984 if (++first == last) {
1989 /* Same invariants, plus one additional:
1990 * - first->vlan != vlan.
1992 while (last[-1].vlan != vlan) {
1993 if (--last == first) {
1998 /* Same invariants, plus one additional:
1999 * - last[-1].vlan == vlan.*/
2000 swap_dst(first++, --last);
2005 mirror_mask_ffs(mirror_mask_t mask)
2007 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2012 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2013 const struct dst *test)
2016 for (i = 0; i < n_dsts; i++) {
2017 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2025 port_trunks_vlan(const struct port *port, uint16_t vlan)
2027 return (port->vlan < 0
2028 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2032 port_includes_vlan(const struct port *port, uint16_t vlan)
2034 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2038 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2039 const struct port *in_port, const struct port *out_port,
2040 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2042 mirror_mask_t mirrors = in_port->src_mirrors;
2043 struct dst *dst = dsts;
2046 if (out_port == FLOOD_PORT) {
2047 /* XXX use ODP_FLOOD if no vlans or bonding. */
2048 /* XXX even better, define each VLAN as a datapath port group */
2049 for (i = 0; i < br->n_ports; i++) {
2050 struct port *port = br->ports[i];
2051 if (port != in_port && port_includes_vlan(port, vlan)
2052 && !port->is_mirror_output_port
2053 && set_dst(dst, flow, in_port, port, tags)) {
2054 mirrors |= port->dst_mirrors;
2058 *nf_output_iface = NF_OUT_FLOOD;
2059 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2060 *nf_output_iface = dst->dp_ifidx;
2061 mirrors |= out_port->dst_mirrors;
2066 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2067 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2069 if (set_dst(dst, flow, in_port, m->out_port, tags)
2070 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2074 for (i = 0; i < br->n_ports; i++) {
2075 struct port *port = br->ports[i];
2076 if (port_includes_vlan(port, m->out_vlan)
2077 && set_dst(dst, flow, in_port, port, tags))
2081 if (port->vlan < 0) {
2082 dst->vlan = m->out_vlan;
2084 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2088 /* Use the vlan tag on the original flow instead of
2089 * the one passed in the vlan parameter. This ensures
2090 * that we compare the vlan from before any implicit
2091 * tagging tags place. This is necessary because
2092 * dst->vlan is the final vlan, after removing implicit
2094 flow_vlan = ntohs(flow->dl_vlan);
2095 if (flow_vlan == 0) {
2096 flow_vlan = OFP_VLAN_NONE;
2098 if (port == in_port && dst->vlan == flow_vlan) {
2099 /* Don't send out input port on same VLAN. */
2107 mirrors &= mirrors - 1;
2110 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2114 static void OVS_UNUSED
2115 print_dsts(const struct dst *dsts, size_t n)
2117 for (; n--; dsts++) {
2118 printf(">p%"PRIu16, dsts->dp_ifidx);
2119 if (dsts->vlan != OFP_VLAN_NONE) {
2120 printf("v%"PRIu16, dsts->vlan);
2126 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2127 const struct port *in_port, const struct port *out_port,
2128 tag_type *tags, struct odp_actions *actions,
2129 uint16_t *nf_output_iface)
2131 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2133 const struct dst *p;
2136 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2139 cur_vlan = ntohs(flow->dl_vlan);
2140 for (p = dsts; p < &dsts[n_dsts]; p++) {
2141 union odp_action *a;
2142 if (p->vlan != cur_vlan) {
2143 if (p->vlan == OFP_VLAN_NONE) {
2144 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2146 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2147 a->vlan_vid.vlan_vid = htons(p->vlan);
2151 a = odp_actions_add(actions, ODPAT_OUTPUT);
2152 a->output.port = p->dp_ifidx;
2156 /* Returns the effective vlan of a packet, taking into account both the
2157 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2158 * the packet is untagged and -1 indicates it has an invalid header and
2159 * should be dropped. */
2160 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2161 struct port *in_port, bool have_packet)
2163 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2164 * belongs to VLAN 0, so we should treat both cases identically. (In the
2165 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2166 * presumably to allow a priority to be specified. In the latter case, the
2167 * packet does not have any 802.1Q header.) */
2168 int vlan = ntohs(flow->dl_vlan);
2169 if (vlan == OFP_VLAN_NONE) {
2172 if (in_port->vlan >= 0) {
2174 /* XXX support double tagging? */
2176 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2177 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2178 "packet received on port %s configured with "
2179 "implicit VLAN %"PRIu16,
2180 br->name, ntohs(flow->dl_vlan),
2181 in_port->name, in_port->vlan);
2185 vlan = in_port->vlan;
2187 if (!port_includes_vlan(in_port, vlan)) {
2189 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2190 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2191 "packet received on port %s not configured for "
2193 br->name, vlan, in_port->name, vlan);
2202 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2203 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2204 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2206 is_gratuitous_arp(const flow_t *flow)
2208 return (flow->dl_type == htons(ETH_TYPE_ARP)
2209 && eth_addr_is_broadcast(flow->dl_dst)
2210 && (flow->nw_proto == ARP_OP_REPLY
2211 || (flow->nw_proto == ARP_OP_REQUEST
2212 && flow->nw_src == flow->nw_dst)));
2216 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2217 struct port *in_port)
2219 enum grat_arp_lock_type lock_type;
2222 /* We don't want to learn from gratuitous ARP packets that are reflected
2223 * back over bond slaves so we lock the learning table. */
2224 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2225 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2226 GRAT_ARP_LOCK_CHECK;
2228 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2231 /* The log messages here could actually be useful in debugging,
2232 * so keep the rate limit relatively high. */
2233 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2235 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2236 "on port %s in VLAN %d",
2237 br->name, ETH_ADDR_ARGS(flow->dl_src),
2238 in_port->name, vlan);
2239 ofproto_revalidate(br->ofproto, rev_tag);
2243 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2244 * dropped. Returns true if they may be forwarded, false if they should be
2247 * If 'have_packet' is true, it indicates that the caller is processing a
2248 * received packet. If 'have_packet' is false, then the caller is just
2249 * revalidating an existing flow because configuration has changed. Either
2250 * way, 'have_packet' only affects logging (there is no point in logging errors
2251 * during revalidation).
2253 * Sets '*in_portp' to the input port. This will be a null pointer if
2254 * flow->in_port does not designate a known input port (in which case
2255 * is_admissible() returns false).
2257 * When returning true, sets '*vlanp' to the effective VLAN of the input
2258 * packet, as returned by flow_get_vlan().
2260 * May also add tags to '*tags', although the current implementation only does
2261 * so in one special case.
2264 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2265 tag_type *tags, int *vlanp, struct port **in_portp)
2267 struct iface *in_iface;
2268 struct port *in_port;
2271 /* Find the interface and port structure for the received packet. */
2272 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2274 /* No interface? Something fishy... */
2276 /* Odd. A few possible reasons here:
2278 * - We deleted an interface but there are still a few packets
2279 * queued up from it.
2281 * - Someone externally added an interface (e.g. with "ovs-dpctl
2282 * add-if") that we don't know about.
2284 * - Packet arrived on the local port but the local port is not
2285 * one of our bridge ports.
2287 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2289 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2290 "interface %"PRIu16, br->name, flow->in_port);
2296 *in_portp = in_port = in_iface->port;
2297 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2302 /* Drop frames for reserved multicast addresses. */
2303 if (eth_addr_is_reserved(flow->dl_dst)) {
2307 /* Drop frames on ports reserved for mirroring. */
2308 if (in_port->is_mirror_output_port) {
2310 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2311 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2312 "%s, which is reserved exclusively for mirroring",
2313 br->name, in_port->name);
2318 /* Packets received on bonds need special attention to avoid duplicates. */
2319 if (in_port->n_ifaces > 1) {
2321 bool is_grat_arp_locked;
2323 if (eth_addr_is_multicast(flow->dl_dst)) {
2324 *tags |= in_port->active_iface_tag;
2325 if (in_port->active_iface != in_iface->port_ifidx) {
2326 /* Drop all multicast packets on inactive slaves. */
2331 /* Drop all packets for which we have learned a different input
2332 * port, because we probably sent the packet on one slave and got
2333 * it back on the other. Gratuitous ARP packets are an exception
2334 * to this rule: the host has moved to another switch. The exception
2335 * to the exception is if we locked the learning table to avoid
2336 * reflections on bond slaves. If this is the case, just drop the
2338 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2339 &is_grat_arp_locked);
2340 if (src_idx != -1 && src_idx != in_port->port_idx &&
2341 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2349 /* If the composed actions may be applied to any packet in the given 'flow',
2350 * returns true. Otherwise, the actions should only be applied to 'packet', or
2351 * not at all, if 'packet' was NULL. */
2353 process_flow(struct bridge *br, const flow_t *flow,
2354 const struct ofpbuf *packet, struct odp_actions *actions,
2355 tag_type *tags, uint16_t *nf_output_iface)
2357 struct port *in_port;
2358 struct port *out_port;
2362 /* Check whether we should drop packets in this flow. */
2363 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2368 /* Learn source MAC (but don't try to learn from revalidation). */
2370 update_learning_table(br, flow, vlan, in_port);
2373 /* Determine output port. */
2374 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2376 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2377 out_port = br->ports[out_port_idx];
2378 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2379 /* If we are revalidating but don't have a learning entry then
2380 * eject the flow. Installing a flow that floods packets opens
2381 * up a window of time where we could learn from a packet reflected
2382 * on a bond and blackhole packets before the learning table is
2383 * updated to reflect the correct port. */
2386 out_port = FLOOD_PORT;
2389 /* Don't send packets out their input ports. */
2390 if (in_port == out_port) {
2396 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2403 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2406 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2407 const struct ofp_phy_port *opp,
2410 struct bridge *br = br_;
2411 struct iface *iface;
2414 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2420 if (reason == OFPPR_DELETE) {
2421 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2422 br->name, iface->name);
2423 iface_destroy(iface);
2424 if (!port->n_ifaces) {
2425 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2426 br->name, port->name);
2432 if (port->n_ifaces > 1) {
2433 bool up = !(opp->state & OFPPS_LINK_DOWN);
2434 bond_link_status_update(iface, up);
2435 port_update_bond_compat(port);
2441 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2442 struct odp_actions *actions, tag_type *tags,
2443 uint16_t *nf_output_iface, void *br_)
2445 struct bridge *br = br_;
2447 COVERAGE_INC(bridge_process_flow);
2448 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2452 bridge_account_flow_ofhook_cb(const flow_t *flow,
2453 const union odp_action *actions,
2454 size_t n_actions, unsigned long long int n_bytes,
2457 struct bridge *br = br_;
2458 const union odp_action *a;
2459 struct port *in_port;
2463 /* Feed information from the active flows back into the learning table
2464 * to ensure that table is always in sync with what is actually flowing
2465 * through the datapath. */
2466 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2467 update_learning_table(br, flow, vlan, in_port);
2470 if (!br->has_bonded_ports) {
2474 for (a = actions; a < &actions[n_actions]; a++) {
2475 if (a->type == ODPAT_OUTPUT) {
2476 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2477 if (out_port && out_port->n_ifaces >= 2) {
2478 struct bond_entry *e = lookup_bond_entry(out_port,
2480 e->tx_bytes += n_bytes;
2487 bridge_account_checkpoint_ofhook_cb(void *br_)
2489 struct bridge *br = br_;
2493 if (!br->has_bonded_ports) {
2498 for (i = 0; i < br->n_ports; i++) {
2499 struct port *port = br->ports[i];
2500 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2501 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2502 bond_rebalance_port(port);
2507 static struct ofhooks bridge_ofhooks = {
2508 bridge_port_changed_ofhook_cb,
2509 bridge_normal_ofhook_cb,
2510 bridge_account_flow_ofhook_cb,
2511 bridge_account_checkpoint_ofhook_cb,
2514 /* Bonding functions. */
2516 /* Statistics for a single interface on a bonded port, used for load-based
2517 * bond rebalancing. */
2518 struct slave_balance {
2519 struct iface *iface; /* The interface. */
2520 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2522 /* All the "bond_entry"s that are assigned to this interface, in order of
2523 * increasing tx_bytes. */
2524 struct bond_entry **hashes;
2528 /* Sorts pointers to pointers to bond_entries in ascending order by the
2529 * interface to which they are assigned, and within a single interface in
2530 * ascending order of bytes transmitted. */
2532 compare_bond_entries(const void *a_, const void *b_)
2534 const struct bond_entry *const *ap = a_;
2535 const struct bond_entry *const *bp = b_;
2536 const struct bond_entry *a = *ap;
2537 const struct bond_entry *b = *bp;
2538 if (a->iface_idx != b->iface_idx) {
2539 return a->iface_idx > b->iface_idx ? 1 : -1;
2540 } else if (a->tx_bytes != b->tx_bytes) {
2541 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2547 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2548 * *descending* order by number of bytes transmitted. */
2550 compare_slave_balance(const void *a_, const void *b_)
2552 const struct slave_balance *a = a_;
2553 const struct slave_balance *b = b_;
2554 if (a->iface->enabled != b->iface->enabled) {
2555 return a->iface->enabled ? -1 : 1;
2556 } else if (a->tx_bytes != b->tx_bytes) {
2557 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2564 swap_bals(struct slave_balance *a, struct slave_balance *b)
2566 struct slave_balance tmp = *a;
2571 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2572 * given that 'p' (and only 'p') might be in the wrong location.
2574 * This function invalidates 'p', since it might now be in a different memory
2577 resort_bals(struct slave_balance *p,
2578 struct slave_balance bals[], size_t n_bals)
2581 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2582 swap_bals(p, p - 1);
2584 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2585 swap_bals(p, p + 1);
2591 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2593 if (VLOG_IS_DBG_ENABLED()) {
2594 struct ds ds = DS_EMPTY_INITIALIZER;
2595 const struct slave_balance *b;
2597 for (b = bals; b < bals + n_bals; b++) {
2601 ds_put_char(&ds, ',');
2603 ds_put_format(&ds, " %s %"PRIu64"kB",
2604 b->iface->name, b->tx_bytes / 1024);
2606 if (!b->iface->enabled) {
2607 ds_put_cstr(&ds, " (disabled)");
2609 if (b->n_hashes > 0) {
2610 ds_put_cstr(&ds, " (");
2611 for (i = 0; i < b->n_hashes; i++) {
2612 const struct bond_entry *e = b->hashes[i];
2614 ds_put_cstr(&ds, " + ");
2616 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2617 e - port->bond_hash, e->tx_bytes / 1024);
2619 ds_put_cstr(&ds, ")");
2622 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2627 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2629 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2632 struct bond_entry *hash = from->hashes[hash_idx];
2633 struct port *port = from->iface->port;
2634 uint64_t delta = hash->tx_bytes;
2636 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2637 "from %s to %s (now carrying %"PRIu64"kB and "
2638 "%"PRIu64"kB load, respectively)",
2639 port->name, delta / 1024, hash - port->bond_hash,
2640 from->iface->name, to->iface->name,
2641 (from->tx_bytes - delta) / 1024,
2642 (to->tx_bytes + delta) / 1024);
2644 /* Delete element from from->hashes.
2646 * We don't bother to add the element to to->hashes because not only would
2647 * it require more work, the only purpose it would be to allow that hash to
2648 * be migrated to another slave in this rebalancing run, and there is no
2649 * point in doing that. */
2650 if (hash_idx == 0) {
2653 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2654 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2658 /* Shift load away from 'from' to 'to'. */
2659 from->tx_bytes -= delta;
2660 to->tx_bytes += delta;
2662 /* Arrange for flows to be revalidated. */
2663 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2664 hash->iface_idx = to->iface->port_ifidx;
2665 hash->iface_tag = tag_create_random();
2669 bond_rebalance_port(struct port *port)
2671 struct slave_balance bals[DP_MAX_PORTS];
2673 struct bond_entry *hashes[BOND_MASK + 1];
2674 struct slave_balance *b, *from, *to;
2675 struct bond_entry *e;
2678 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2679 * descending order of tx_bytes, so that bals[0] represents the most
2680 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2683 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2684 * array for each slave_balance structure, we sort our local array of
2685 * hashes in order by slave, so that all of the hashes for a given slave
2686 * become contiguous in memory, and then we point each 'hashes' members of
2687 * a slave_balance structure to the start of a contiguous group. */
2688 n_bals = port->n_ifaces;
2689 for (b = bals; b < &bals[n_bals]; b++) {
2690 b->iface = port->ifaces[b - bals];
2695 for (i = 0; i <= BOND_MASK; i++) {
2696 hashes[i] = &port->bond_hash[i];
2698 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2699 for (i = 0; i <= BOND_MASK; i++) {
2701 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2702 b = &bals[e->iface_idx];
2703 b->tx_bytes += e->tx_bytes;
2705 b->hashes = &hashes[i];
2710 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2711 log_bals(bals, n_bals, port);
2713 /* Discard slaves that aren't enabled (which were sorted to the back of the
2714 * array earlier). */
2715 while (!bals[n_bals - 1].iface->enabled) {
2722 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2723 to = &bals[n_bals - 1];
2724 for (from = bals; from < to; ) {
2725 uint64_t overload = from->tx_bytes - to->tx_bytes;
2726 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2727 /* The extra load on 'from' (and all less-loaded slaves), compared
2728 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2729 * it is less than ~1Mbps. No point in rebalancing. */
2731 } else if (from->n_hashes == 1) {
2732 /* 'from' only carries a single MAC hash, so we can't shift any
2733 * load away from it, even though we want to. */
2736 /* 'from' is carrying significantly more load than 'to', and that
2737 * load is split across at least two different hashes. Pick a hash
2738 * to migrate to 'to' (the least-loaded slave), given that doing so
2739 * must decrease the ratio of the load on the two slaves by at
2742 * The sort order we use means that we prefer to shift away the
2743 * smallest hashes instead of the biggest ones. There is little
2744 * reason behind this decision; we could use the opposite sort
2745 * order to shift away big hashes ahead of small ones. */
2749 for (i = 0; i < from->n_hashes; i++) {
2750 double old_ratio, new_ratio;
2751 uint64_t delta = from->hashes[i]->tx_bytes;
2753 if (delta == 0 || from->tx_bytes - delta == 0) {
2754 /* Pointless move. */
2758 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2760 if (to->tx_bytes == 0) {
2761 /* Nothing on the new slave, move it. */
2765 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2766 new_ratio = (double)(from->tx_bytes - delta) /
2767 (to->tx_bytes + delta);
2769 if (new_ratio == 0) {
2770 /* Should already be covered but check to prevent division
2775 if (new_ratio < 1) {
2776 new_ratio = 1 / new_ratio;
2779 if (old_ratio - new_ratio > 0.1) {
2780 /* Would decrease the ratio, move it. */
2784 if (i < from->n_hashes) {
2785 bond_shift_load(from, to, i);
2786 port->bond_compat_is_stale = true;
2788 /* If the result of the migration changed the relative order of
2789 * 'from' and 'to' swap them back to maintain invariants. */
2790 if (order_swapped) {
2791 swap_bals(from, to);
2794 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2795 * point to different slave_balance structures. It is only
2796 * valid to do these two operations in a row at all because we
2797 * know that 'from' will not move past 'to' and vice versa. */
2798 resort_bals(from, bals, n_bals);
2799 resort_bals(to, bals, n_bals);
2806 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2807 * historical data to decay to <1% in 7 rebalancing runs. */
2808 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2814 bond_send_learning_packets(struct port *port)
2816 struct bridge *br = port->bridge;
2817 struct mac_entry *e;
2818 struct ofpbuf packet;
2819 int error, n_packets, n_errors;
2821 if (!port->n_ifaces || port->active_iface < 0) {
2825 ofpbuf_init(&packet, 128);
2826 error = n_packets = n_errors = 0;
2827 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2828 union ofp_action actions[2], *a;
2834 if (e->port == port->port_idx
2835 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2839 /* Compose actions. */
2840 memset(actions, 0, sizeof actions);
2843 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2844 a->vlan_vid.len = htons(sizeof *a);
2845 a->vlan_vid.vlan_vid = htons(e->vlan);
2848 a->output.type = htons(OFPAT_OUTPUT);
2849 a->output.len = htons(sizeof *a);
2850 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2855 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2857 flow_extract(&packet, 0, ODPP_NONE, &flow);
2858 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2865 ofpbuf_uninit(&packet);
2868 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2869 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2870 "packets, last error was: %s",
2871 port->name, n_errors, n_packets, strerror(error));
2873 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2874 port->name, n_packets);
2878 /* Bonding unixctl user interface functions. */
2881 bond_unixctl_list(struct unixctl_conn *conn,
2882 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2884 struct ds ds = DS_EMPTY_INITIALIZER;
2885 const struct bridge *br;
2887 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2889 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2892 for (i = 0; i < br->n_ports; i++) {
2893 const struct port *port = br->ports[i];
2894 if (port->n_ifaces > 1) {
2897 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2898 for (j = 0; j < port->n_ifaces; j++) {
2899 const struct iface *iface = port->ifaces[j];
2901 ds_put_cstr(&ds, ", ");
2903 ds_put_cstr(&ds, iface->name);
2905 ds_put_char(&ds, '\n');
2909 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2913 static struct port *
2914 bond_find(const char *name)
2916 const struct bridge *br;
2918 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2921 for (i = 0; i < br->n_ports; i++) {
2922 struct port *port = br->ports[i];
2923 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2932 bond_unixctl_show(struct unixctl_conn *conn,
2933 const char *args, void *aux OVS_UNUSED)
2935 struct ds ds = DS_EMPTY_INITIALIZER;
2936 const struct port *port;
2939 port = bond_find(args);
2941 unixctl_command_reply(conn, 501, "no such bond");
2945 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2946 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2947 ds_put_format(&ds, "next rebalance: %lld ms\n",
2948 port->bond_next_rebalance - time_msec());
2949 for (j = 0; j < port->n_ifaces; j++) {
2950 const struct iface *iface = port->ifaces[j];
2951 struct bond_entry *be;
2954 ds_put_format(&ds, "slave %s: %s\n",
2955 iface->name, iface->enabled ? "enabled" : "disabled");
2956 if (j == port->active_iface) {
2957 ds_put_cstr(&ds, "\tactive slave\n");
2959 if (iface->delay_expires != LLONG_MAX) {
2960 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2961 iface->enabled ? "downdelay" : "updelay",
2962 iface->delay_expires - time_msec());
2966 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2967 int hash = be - port->bond_hash;
2968 struct mac_entry *me;
2970 if (be->iface_idx != j) {
2974 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2975 hash, be->tx_bytes / 1024);
2978 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2979 &port->bridge->ml->lrus) {
2982 if (bond_hash(me->mac) == hash
2983 && me->port != port->port_idx
2984 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2985 && dp_ifidx == iface->dp_ifidx)
2987 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2988 ETH_ADDR_ARGS(me->mac));
2993 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2998 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2999 void *aux OVS_UNUSED)
3001 char *args = (char *) args_;
3002 char *save_ptr = NULL;
3003 char *bond_s, *hash_s, *slave_s;
3004 uint8_t mac[ETH_ADDR_LEN];
3006 struct iface *iface;
3007 struct bond_entry *entry;
3010 bond_s = strtok_r(args, " ", &save_ptr);
3011 hash_s = strtok_r(NULL, " ", &save_ptr);
3012 slave_s = strtok_r(NULL, " ", &save_ptr);
3014 unixctl_command_reply(conn, 501,
3015 "usage: bond/migrate BOND HASH SLAVE");
3019 port = bond_find(bond_s);
3021 unixctl_command_reply(conn, 501, "no such bond");
3025 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3026 == ETH_ADDR_SCAN_COUNT) {
3027 hash = bond_hash(mac);
3028 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3029 hash = atoi(hash_s) & BOND_MASK;
3031 unixctl_command_reply(conn, 501, "bad hash");
3035 iface = port_lookup_iface(port, slave_s);
3037 unixctl_command_reply(conn, 501, "no such slave");
3041 if (!iface->enabled) {
3042 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3046 entry = &port->bond_hash[hash];
3047 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3048 entry->iface_idx = iface->port_ifidx;
3049 entry->iface_tag = tag_create_random();
3050 port->bond_compat_is_stale = true;
3051 unixctl_command_reply(conn, 200, "migrated");
3055 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3056 void *aux OVS_UNUSED)
3058 char *args = (char *) args_;
3059 char *save_ptr = NULL;
3060 char *bond_s, *slave_s;
3062 struct iface *iface;
3064 bond_s = strtok_r(args, " ", &save_ptr);
3065 slave_s = strtok_r(NULL, " ", &save_ptr);
3067 unixctl_command_reply(conn, 501,
3068 "usage: bond/set-active-slave BOND SLAVE");
3072 port = bond_find(bond_s);
3074 unixctl_command_reply(conn, 501, "no such bond");
3078 iface = port_lookup_iface(port, slave_s);
3080 unixctl_command_reply(conn, 501, "no such slave");
3084 if (!iface->enabled) {
3085 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3089 if (port->active_iface != iface->port_ifidx) {
3090 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3091 port->active_iface = iface->port_ifidx;
3092 port->active_iface_tag = tag_create_random();
3093 VLOG_INFO("port %s: active interface is now %s",
3094 port->name, iface->name);
3095 bond_send_learning_packets(port);
3096 unixctl_command_reply(conn, 200, "done");
3098 unixctl_command_reply(conn, 200, "no change");
3103 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3105 char *args = (char *) args_;
3106 char *save_ptr = NULL;
3107 char *bond_s, *slave_s;
3109 struct iface *iface;
3111 bond_s = strtok_r(args, " ", &save_ptr);
3112 slave_s = strtok_r(NULL, " ", &save_ptr);
3114 unixctl_command_reply(conn, 501,
3115 "usage: bond/enable/disable-slave BOND SLAVE");
3119 port = bond_find(bond_s);
3121 unixctl_command_reply(conn, 501, "no such bond");
3125 iface = port_lookup_iface(port, slave_s);
3127 unixctl_command_reply(conn, 501, "no such slave");
3131 bond_enable_slave(iface, enable);
3132 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3136 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3137 void *aux OVS_UNUSED)
3139 enable_slave(conn, args, true);
3143 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3144 void *aux OVS_UNUSED)
3146 enable_slave(conn, args, false);
3150 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3151 void *aux OVS_UNUSED)
3153 uint8_t mac[ETH_ADDR_LEN];
3157 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3158 == ETH_ADDR_SCAN_COUNT) {
3159 hash = bond_hash(mac);
3161 hash_cstr = xasprintf("%u", hash);
3162 unixctl_command_reply(conn, 200, hash_cstr);
3165 unixctl_command_reply(conn, 501, "invalid mac");
3172 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3173 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3174 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3175 unixctl_command_register("bond/set-active-slave",
3176 bond_unixctl_set_active_slave, NULL);
3177 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3179 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3181 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3184 /* Port functions. */
3186 static struct port *
3187 port_create(struct bridge *br, const char *name)
3191 port = xzalloc(sizeof *port);
3193 port->port_idx = br->n_ports;
3195 port->trunks = NULL;
3196 port->name = xstrdup(name);
3197 port->active_iface = -1;
3199 if (br->n_ports >= br->allocated_ports) {
3200 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3203 br->ports[br->n_ports++] = port;
3204 shash_add_assert(&br->port_by_name, port->name, port);
3206 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3213 get_port_other_config(const struct ovsrec_port *port, const char *key,
3214 const char *default_value)
3216 const char *value = get_ovsrec_key_value(key,
3217 port->key_other_config,
3218 port->value_other_config,
3219 port->n_other_config);
3220 return value ? value : default_value;
3224 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3226 struct shash new_ifaces;
3229 /* Collect list of new interfaces. */
3230 shash_init(&new_ifaces);
3231 for (i = 0; i < cfg->n_interfaces; i++) {
3232 const char *name = cfg->interfaces[i]->name;
3233 shash_add_once(&new_ifaces, name, NULL);
3236 /* Get rid of deleted interfaces. */
3237 for (i = 0; i < port->n_ifaces; ) {
3238 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3239 iface_destroy(port->ifaces[i]);
3245 shash_destroy(&new_ifaces);
3249 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3251 struct shash new_ifaces;
3252 long long int next_rebalance;
3253 unsigned long *trunks;
3259 /* Update settings. */
3260 port->updelay = cfg->bond_updelay;
3261 if (port->updelay < 0) {
3264 port->updelay = cfg->bond_downdelay;
3265 if (port->downdelay < 0) {
3266 port->downdelay = 0;
3268 port->bond_rebalance_interval = atoi(
3269 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3270 if (port->bond_rebalance_interval < 1000) {
3271 port->bond_rebalance_interval = 1000;
3273 next_rebalance = time_msec() + port->bond_rebalance_interval;
3274 if (port->bond_next_rebalance > next_rebalance) {
3275 port->bond_next_rebalance = next_rebalance;
3278 /* Add new interfaces and update 'cfg' member of existing ones. */
3279 shash_init(&new_ifaces);
3280 for (i = 0; i < cfg->n_interfaces; i++) {
3281 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3282 struct iface *iface;
3284 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3285 VLOG_WARN("port %s: %s specified twice as port interface",
3286 port->name, if_cfg->name);
3290 iface = iface_lookup(port->bridge, if_cfg->name);
3292 if (iface->port != port) {
3293 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3295 port->bridge->name, if_cfg->name, iface->port->name);
3298 iface->cfg = if_cfg;
3300 iface_create(port, if_cfg);
3303 shash_destroy(&new_ifaces);
3308 if (port->n_ifaces < 2) {
3310 if (vlan >= 0 && vlan <= 4095) {
3311 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3316 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3317 * they even work as-is. But they have not been tested. */
3318 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3322 if (port->vlan != vlan) {
3324 bridge_flush(port->bridge);
3327 /* Get trunked VLANs. */
3329 if (vlan < 0 && cfg->n_trunks) {
3333 trunks = bitmap_allocate(4096);
3335 for (i = 0; i < cfg->n_trunks; i++) {
3336 int trunk = cfg->trunks[i];
3338 bitmap_set1(trunks, trunk);
3344 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3345 port->name, cfg->n_trunks);
3347 if (n_errors == cfg->n_trunks) {
3348 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3350 bitmap_free(trunks);
3353 } else if (vlan >= 0 && cfg->n_trunks) {
3354 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3358 ? port->trunks != NULL
3359 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3360 bridge_flush(port->bridge);
3362 bitmap_free(port->trunks);
3363 port->trunks = trunks;
3367 port_destroy(struct port *port)
3370 struct bridge *br = port->bridge;
3374 proc_net_compat_update_vlan(port->name, NULL, 0);
3375 proc_net_compat_update_bond(port->name, NULL);
3377 for (i = 0; i < MAX_MIRRORS; i++) {
3378 struct mirror *m = br->mirrors[i];
3379 if (m && m->out_port == port) {
3384 while (port->n_ifaces > 0) {
3385 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3388 shash_find_and_delete_assert(&br->port_by_name, port->name);
3390 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3391 del->port_idx = port->port_idx;
3394 bitmap_free(port->trunks);
3401 static struct port *
3402 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3404 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3405 return iface ? iface->port : NULL;
3408 static struct port *
3409 port_lookup(const struct bridge *br, const char *name)
3411 return shash_find_data(&br->port_by_name, name);
3414 static struct iface *
3415 port_lookup_iface(const struct port *port, const char *name)
3417 struct iface *iface = iface_lookup(port->bridge, name);
3418 return iface && iface->port == port ? iface : NULL;
3422 port_update_bonding(struct port *port)
3424 if (port->n_ifaces < 2) {
3425 /* Not a bonded port. */
3426 if (port->bond_hash) {
3427 free(port->bond_hash);
3428 port->bond_hash = NULL;
3429 port->bond_compat_is_stale = true;
3430 port->bond_fake_iface = false;
3433 if (!port->bond_hash) {
3436 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3437 for (i = 0; i <= BOND_MASK; i++) {
3438 struct bond_entry *e = &port->bond_hash[i];
3442 port->no_ifaces_tag = tag_create_random();
3443 bond_choose_active_iface(port);
3444 port->bond_next_rebalance
3445 = time_msec() + port->bond_rebalance_interval;
3447 if (port->cfg->bond_fake_iface) {
3448 port->bond_next_fake_iface_update = time_msec();
3451 port->bond_compat_is_stale = true;
3452 port->bond_fake_iface = port->cfg->bond_fake_iface;
3457 port_update_bond_compat(struct port *port)
3459 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3460 struct compat_bond bond;
3463 if (port->n_ifaces < 2) {
3464 proc_net_compat_update_bond(port->name, NULL);
3469 bond.updelay = port->updelay;
3470 bond.downdelay = port->downdelay;
3473 bond.hashes = compat_hashes;
3474 if (port->bond_hash) {
3475 const struct bond_entry *e;
3476 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3477 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3478 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3479 cbh->hash = e - port->bond_hash;
3480 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3485 bond.n_slaves = port->n_ifaces;
3486 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3487 for (i = 0; i < port->n_ifaces; i++) {
3488 struct iface *iface = port->ifaces[i];
3489 struct compat_bond_slave *slave = &bond.slaves[i];
3490 slave->name = iface->name;
3492 /* We need to make the same determination as the Linux bonding
3493 * code to determine whether a slave should be consider "up".
3494 * The Linux function bond_miimon_inspect() supports four
3495 * BOND_LINK_* states:
3497 * - BOND_LINK_UP: carrier detected, updelay has passed.
3498 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3499 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3500 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3502 * The function bond_info_show_slave() only considers BOND_LINK_UP
3503 * to be "up" and anything else to be "down".
3505 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3509 netdev_get_etheraddr(iface->netdev, slave->mac);
3512 if (port->bond_fake_iface) {
3513 struct netdev *bond_netdev;
3515 if (!netdev_open_default(port->name, &bond_netdev)) {
3517 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3519 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3521 netdev_close(bond_netdev);
3525 proc_net_compat_update_bond(port->name, &bond);
3530 port_update_vlan_compat(struct port *port)
3532 struct bridge *br = port->bridge;
3533 char *vlandev_name = NULL;
3535 if (port->vlan > 0) {
3536 /* Figure out the name that the VLAN device should actually have, if it
3537 * existed. This takes some work because the VLAN device would not
3538 * have port->name in its name; rather, it would have the trunk port's
3539 * name, and 'port' would be attached to a bridge that also had the
3540 * VLAN device one of its ports. So we need to find a trunk port that
3541 * includes port->vlan.
3543 * There might be more than one candidate. This doesn't happen on
3544 * XenServer, so if it happens we just pick the first choice in
3545 * alphabetical order instead of creating multiple VLAN devices. */
3547 for (i = 0; i < br->n_ports; i++) {
3548 struct port *p = br->ports[i];
3549 if (port_trunks_vlan(p, port->vlan)
3551 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3553 uint8_t ea[ETH_ADDR_LEN];
3554 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3555 if (!eth_addr_is_multicast(ea) &&
3556 !eth_addr_is_reserved(ea) &&
3557 !eth_addr_is_zero(ea)) {
3558 vlandev_name = p->name;
3563 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3566 /* Interface functions. */
3568 static struct iface *
3569 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3571 struct bridge *br = port->bridge;
3572 struct iface *iface;
3573 char *name = if_cfg->name;
3576 iface = xzalloc(sizeof *iface);
3578 iface->port_ifidx = port->n_ifaces;
3579 iface->name = xstrdup(name);
3580 iface->dp_ifidx = -1;
3581 iface->tag = tag_create_random();
3582 iface->delay_expires = LLONG_MAX;
3583 iface->netdev = NULL;
3584 iface->cfg = if_cfg;
3586 shash_add_assert(&br->iface_by_name, iface->name, iface);
3588 /* Attempt to create the network interface in case it doesn't exist yet. */
3589 if (!iface_is_internal(br, iface->name)) {
3590 error = set_up_iface(if_cfg, iface, true);
3592 VLOG_WARN("could not create iface %s: %s", iface->name,
3595 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3602 if (port->n_ifaces >= port->allocated_ifaces) {
3603 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3604 sizeof *port->ifaces);
3606 port->ifaces[port->n_ifaces++] = iface;
3607 if (port->n_ifaces > 1) {
3608 br->has_bonded_ports = true;
3611 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3619 iface_destroy(struct iface *iface)
3622 struct port *port = iface->port;
3623 struct bridge *br = port->bridge;
3624 bool del_active = port->active_iface == iface->port_ifidx;
3627 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3629 if (iface->dp_ifidx >= 0) {
3630 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3633 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3634 del->port_ifidx = iface->port_ifidx;
3636 netdev_close(iface->netdev);
3639 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3640 bond_choose_active_iface(port);
3641 bond_send_learning_packets(port);
3647 bridge_flush(port->bridge);
3651 static struct iface *
3652 iface_lookup(const struct bridge *br, const char *name)
3654 return shash_find_data(&br->iface_by_name, name);
3657 static struct iface *
3658 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3660 return port_array_get(&br->ifaces, dp_ifidx);
3663 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3664 * 'br', that is, an interface that is entirely simulated within the datapath.
3665 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3666 * interfaces are created by setting "iface.<iface>.internal = true".
3668 * In addition, we have a kluge-y feature that creates an internal port with
3669 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3670 * This feature needs to go away in the long term. Until then, this is one
3671 * reason why this function takes a name instead of a struct iface: the fake
3672 * interfaces created this way do not have a struct iface. */
3674 iface_is_internal(const struct bridge *br, const char *if_name)
3676 struct iface *iface;
3679 if (!strcmp(if_name, br->name)) {
3683 iface = iface_lookup(br, if_name);
3684 if (iface && !strcmp(iface->cfg->type, "internal")) {
3688 port = port_lookup(br, if_name);
3689 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3695 /* Set Ethernet address of 'iface', if one is specified in the configuration
3698 iface_set_mac(struct iface *iface)
3700 uint8_t ea[ETH_ADDR_LEN];
3702 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3703 if (eth_addr_is_multicast(ea)) {
3704 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3706 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3707 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3708 iface->name, iface->name);
3710 int error = netdev_set_etheraddr(iface->netdev, ea);
3712 VLOG_ERR("interface %s: setting MAC failed (%s)",
3713 iface->name, strerror(error));
3720 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3721 struct shash *shash)
3726 for (i = 0; i < n; i++) {
3727 shash_add(shash, keys[i], values[i]);
3731 struct iface_delete_queues_cbdata {
3732 struct netdev *netdev;
3733 const int64_t *queue_ids;
3738 queue_ids_include(const int64_t *ids, size_t n, int64_t target)
3743 while (low < high) {
3744 size_t mid = low + (high - low) / 2;
3745 if (target > ids[mid]) {
3747 } else if (target < ids[mid]) {
3757 iface_delete_queues(unsigned int queue_id,
3758 const struct shash *details OVS_UNUSED, void *cbdata_)
3760 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3762 if (!queue_ids_include(cbdata->queue_ids, cbdata->n_queue_ids, queue_id)) {
3763 netdev_delete_queue(cbdata->netdev, queue_id);
3768 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3770 if (!qos || qos->type[0] == '\0') {
3771 netdev_set_qos(iface->netdev, NULL, NULL);
3773 struct iface_delete_queues_cbdata cbdata;
3774 struct shash details;
3777 /* Configure top-level Qos for 'iface'. */
3778 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3779 qos->n_other_config, &details);
3780 netdev_set_qos(iface->netdev, qos->type, &details);
3781 shash_destroy(&details);
3783 /* Deconfigure queues that were deleted. */
3784 cbdata.netdev = iface->netdev;
3785 cbdata.queue_ids = qos->key_queues;
3786 cbdata.n_queue_ids = qos->n_queues;
3787 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3789 /* Configure queues for 'iface'. */
3790 for (i = 0; i < qos->n_queues; i++) {
3791 const struct ovsrec_queue *queue = qos->value_queues[i];
3792 unsigned int queue_id = qos->key_queues[i];
3794 shash_from_ovs_idl_map(queue->key_other_config,
3795 queue->value_other_config,
3796 queue->n_other_config, &details);
3797 netdev_set_queue(iface->netdev, queue_id, &details);
3798 shash_destroy(&details);
3803 /* Port mirroring. */
3806 mirror_reconfigure(struct bridge *br)
3808 struct shash old_mirrors, new_mirrors;
3809 struct shash_node *node;
3810 unsigned long *rspan_vlans;
3813 /* Collect old mirrors. */
3814 shash_init(&old_mirrors);
3815 for (i = 0; i < MAX_MIRRORS; i++) {
3816 if (br->mirrors[i]) {
3817 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3821 /* Collect new mirrors. */
3822 shash_init(&new_mirrors);
3823 for (i = 0; i < br->cfg->n_mirrors; i++) {
3824 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3825 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3826 VLOG_WARN("bridge %s: %s specified twice as mirror",
3827 br->name, cfg->name);
3831 /* Get rid of deleted mirrors and add new mirrors. */
3832 SHASH_FOR_EACH (node, &old_mirrors) {
3833 if (!shash_find(&new_mirrors, node->name)) {
3834 mirror_destroy(node->data);
3837 SHASH_FOR_EACH (node, &new_mirrors) {
3838 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3840 mirror = mirror_create(br, node->name);
3845 mirror_reconfigure_one(mirror, node->data);
3847 shash_destroy(&old_mirrors);
3848 shash_destroy(&new_mirrors);
3850 /* Update port reserved status. */
3851 for (i = 0; i < br->n_ports; i++) {
3852 br->ports[i]->is_mirror_output_port = false;
3854 for (i = 0; i < MAX_MIRRORS; i++) {
3855 struct mirror *m = br->mirrors[i];
3856 if (m && m->out_port) {
3857 m->out_port->is_mirror_output_port = true;
3861 /* Update flooded vlans (for RSPAN). */
3863 if (br->cfg->n_flood_vlans) {
3864 rspan_vlans = bitmap_allocate(4096);
3866 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3867 int64_t vlan = br->cfg->flood_vlans[i];
3868 if (vlan >= 0 && vlan < 4096) {
3869 bitmap_set1(rspan_vlans, vlan);
3870 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3873 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3878 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3883 static struct mirror *
3884 mirror_create(struct bridge *br, const char *name)
3889 for (i = 0; ; i++) {
3890 if (i >= MAX_MIRRORS) {
3891 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3892 "cannot create %s", br->name, MAX_MIRRORS, name);
3895 if (!br->mirrors[i]) {
3900 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3903 br->mirrors[i] = m = xzalloc(sizeof *m);
3906 m->name = xstrdup(name);
3907 shash_init(&m->src_ports);
3908 shash_init(&m->dst_ports);
3918 mirror_destroy(struct mirror *m)
3921 struct bridge *br = m->bridge;
3924 for (i = 0; i < br->n_ports; i++) {
3925 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3926 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3929 shash_destroy(&m->src_ports);
3930 shash_destroy(&m->dst_ports);
3933 m->bridge->mirrors[m->idx] = NULL;
3941 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3942 struct shash *names)
3946 for (i = 0; i < n_ports; i++) {
3947 const char *name = ports[i]->name;
3948 if (port_lookup(m->bridge, name)) {
3949 shash_add_once(names, name, NULL);
3951 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3952 "port %s", m->bridge->name, m->name, name);
3958 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3964 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3966 for (i = 0; i < cfg->n_select_vlan; i++) {
3967 int64_t vlan = cfg->select_vlan[i];
3968 if (vlan < 0 || vlan > 4095) {
3969 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3970 m->bridge->name, m->name, vlan);
3972 (*vlans)[n_vlans++] = vlan;
3979 vlan_is_mirrored(const struct mirror *m, int vlan)
3983 for (i = 0; i < m->n_vlans; i++) {
3984 if (m->vlans[i] == vlan) {
3992 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3996 for (i = 0; i < m->n_vlans; i++) {
3997 if (port_trunks_vlan(p, m->vlans[i])) {
4005 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4007 struct shash src_ports, dst_ports;
4008 mirror_mask_t mirror_bit;
4009 struct port *out_port;
4015 /* Get output port. */
4016 if (cfg->output_port) {
4017 out_port = port_lookup(m->bridge, cfg->output_port->name);
4019 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4020 m->bridge->name, m->name);
4026 if (cfg->output_vlan) {
4027 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4028 "output vlan; ignoring output vlan",
4029 m->bridge->name, m->name);
4031 } else if (cfg->output_vlan) {
4033 out_vlan = *cfg->output_vlan;
4035 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4036 m->bridge->name, m->name);
4041 shash_init(&src_ports);
4042 shash_init(&dst_ports);
4043 if (cfg->select_all) {
4044 for (i = 0; i < m->bridge->n_ports; i++) {
4045 const char *name = m->bridge->ports[i]->name;
4046 shash_add_once(&src_ports, name, NULL);
4047 shash_add_once(&dst_ports, name, NULL);
4052 /* Get ports, and drop duplicates and ports that don't exist. */
4053 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4055 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4058 /* Get all the vlans, and drop duplicate and invalid vlans. */
4059 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4062 /* Update mirror data. */
4063 if (!shash_equal_keys(&m->src_ports, &src_ports)
4064 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4065 || m->n_vlans != n_vlans
4066 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4067 || m->out_port != out_port
4068 || m->out_vlan != out_vlan) {
4069 bridge_flush(m->bridge);
4071 shash_swap(&m->src_ports, &src_ports);
4072 shash_swap(&m->dst_ports, &dst_ports);
4075 m->n_vlans = n_vlans;
4076 m->out_port = out_port;
4077 m->out_vlan = out_vlan;
4080 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4081 for (i = 0; i < m->bridge->n_ports; i++) {
4082 struct port *port = m->bridge->ports[i];
4084 if (shash_find(&m->src_ports, port->name)
4087 ? port_trunks_any_mirrored_vlan(m, port)
4088 : vlan_is_mirrored(m, port->vlan)))) {
4089 port->src_mirrors |= mirror_bit;
4091 port->src_mirrors &= ~mirror_bit;
4094 if (shash_find(&m->dst_ports, port->name)) {
4095 port->dst_mirrors |= mirror_bit;
4097 port->dst_mirrors &= ~mirror_bit;
4102 shash_destroy(&src_ports);
4103 shash_destroy(&dst_ports);