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 unixctl_cb_func bridge_unixctl_reconnect;
199 static int bridge_run_one(struct bridge *);
200 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
201 const struct bridge *br,
202 struct ovsrec_controller ***controllersp);
203 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
205 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
207 const struct sockaddr_in *managers,
209 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
210 static void bridge_fetch_dp_ifaces(struct bridge *);
211 static void bridge_flush(struct bridge *);
212 static void bridge_pick_local_hw_addr(struct bridge *,
213 uint8_t ea[ETH_ADDR_LEN],
214 struct iface **hw_addr_iface);
215 static uint64_t bridge_pick_datapath_id(struct bridge *,
216 const uint8_t bridge_ea[ETH_ADDR_LEN],
217 struct iface *hw_addr_iface);
218 static struct iface *bridge_get_local_iface(struct bridge *);
219 static uint64_t dpid_from_hash(const void *, size_t nbytes);
221 static unixctl_cb_func bridge_unixctl_fdb_show;
223 static void bond_init(void);
224 static void bond_run(struct bridge *);
225 static void bond_wait(struct bridge *);
226 static void bond_rebalance_port(struct port *);
227 static void bond_send_learning_packets(struct port *);
228 static void bond_enable_slave(struct iface *iface, bool enable);
230 static struct port *port_create(struct bridge *, const char *name);
231 static void port_reconfigure(struct port *, const struct ovsrec_port *);
232 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
233 static void port_destroy(struct port *);
234 static struct port *port_lookup(const struct bridge *, const char *name);
235 static struct iface *port_lookup_iface(const struct port *, const char *name);
236 static struct port *port_from_dp_ifidx(const struct bridge *,
238 static void port_update_bond_compat(struct port *);
239 static void port_update_vlan_compat(struct port *);
240 static void port_update_bonding(struct port *);
242 static struct mirror *mirror_create(struct bridge *, const char *name);
243 static void mirror_destroy(struct mirror *);
244 static void mirror_reconfigure(struct bridge *);
245 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
246 static bool vlan_is_mirrored(const struct mirror *, int vlan);
248 static struct iface *iface_create(struct port *port,
249 const struct ovsrec_interface *if_cfg);
250 static void iface_destroy(struct iface *);
251 static struct iface *iface_lookup(const struct bridge *, const char *name);
252 static struct iface *iface_from_dp_ifidx(const struct bridge *,
254 static bool iface_is_internal(const struct bridge *, const char *name);
255 static void iface_set_mac(struct iface *);
256 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
258 /* Hooks into ofproto processing. */
259 static struct ofhooks bridge_ofhooks;
261 /* Public functions. */
263 /* Initializes the bridge module, configuring it to obtain its configuration
264 * from an OVSDB server accessed over 'remote', which should be a string in a
265 * form acceptable to ovsdb_idl_create(). */
267 bridge_init(const char *remote)
269 /* Create connection to database. */
270 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
272 /* Register unixctl commands. */
273 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
274 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
276 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
281 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
282 * but for which the ovs-vswitchd configuration 'cfg' is required. */
284 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
286 static bool already_configured_once;
287 struct svec bridge_names;
288 struct svec dpif_names, dpif_types;
291 /* Only do this once per ovs-vswitchd run. */
292 if (already_configured_once) {
295 already_configured_once = true;
297 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
299 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
300 svec_init(&bridge_names);
301 for (i = 0; i < cfg->n_bridges; i++) {
302 svec_add(&bridge_names, cfg->bridges[i]->name);
304 svec_sort(&bridge_names);
306 /* Iterate over all system dpifs and delete any of them that do not appear
308 svec_init(&dpif_names);
309 svec_init(&dpif_types);
310 dp_enumerate_types(&dpif_types);
311 for (i = 0; i < dpif_types.n; i++) {
316 dp_enumerate_names(dpif_types.names[i], &dpif_names);
318 /* For each dpif... */
319 for (j = 0; j < dpif_names.n; j++) {
320 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
322 struct svec all_names;
325 /* ...check whether any of its names is in 'bridge_names'. */
326 svec_init(&all_names);
327 dpif_get_all_names(dpif, &all_names);
328 for (k = 0; k < all_names.n; k++) {
329 if (svec_contains(&bridge_names, all_names.names[k])) {
334 /* No. Delete the dpif. */
338 svec_destroy(&all_names);
343 svec_destroy(&bridge_names);
344 svec_destroy(&dpif_names);
345 svec_destroy(&dpif_types);
350 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
352 /* XXX SSL should be configurable on a per-bridge basis. */
354 stream_ssl_set_private_key_file(ssl->private_key);
355 stream_ssl_set_certificate_file(ssl->certificate);
356 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
361 /* Attempt to create the network device 'iface_name' through the netdev
364 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
367 struct shash options;
371 shash_init(&options);
372 for (i = 0; i < iface_cfg->n_options; i++) {
373 shash_add(&options, iface_cfg->key_options[i],
374 xstrdup(iface_cfg->value_options[i]));
378 struct netdev_options netdev_options;
380 memset(&netdev_options, 0, sizeof netdev_options);
381 netdev_options.name = iface_cfg->name;
382 if (!strcmp(iface_cfg->type, "internal")) {
383 /* An "internal" config type maps to a netdev "system" type. */
384 netdev_options.type = "system";
386 netdev_options.type = iface_cfg->type;
388 netdev_options.args = &options;
389 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
391 error = netdev_open(&netdev_options, &iface->netdev);
394 netdev_get_carrier(iface->netdev, &iface->enabled);
396 } else if (iface->netdev) {
397 const char *netdev_type = netdev_get_type(iface->netdev);
398 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
399 ? iface_cfg->type : NULL;
401 /* An "internal" config type maps to a netdev "system" type. */
402 if (iface_type && !strcmp(iface_type, "internal")) {
403 iface_type = "system";
406 if (!iface_type || !strcmp(netdev_type, iface_type)) {
407 error = netdev_reconfigure(iface->netdev, &options);
409 VLOG_WARN("%s: attempting change device type from %s to %s",
410 iface_cfg->name, netdev_type, iface_type);
414 shash_destroy_free_data(&options);
420 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
422 return set_up_iface(iface_cfg, iface, false);
426 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
427 void *aux OVS_UNUSED)
429 if (!iface->netdev) {
430 int error = set_up_iface(iface->cfg, iface, true);
432 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
442 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
443 void *aux OVS_UNUSED)
445 if (iface->dp_ifidx >= 0) {
446 VLOG_DBG("%s has interface %s on port %d",
448 iface->name, iface->dp_ifidx);
451 VLOG_ERR("%s interface not in %s, dropping",
452 iface->name, dpif_name(br->dpif));
458 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
459 void *aux OVS_UNUSED)
461 /* Set policing attributes. */
462 netdev_set_policing(iface->netdev,
463 iface->cfg->ingress_policing_rate,
464 iface->cfg->ingress_policing_burst);
466 /* Set MAC address of internal interfaces other than the local
468 if (iface->dp_ifidx != ODPP_LOCAL
469 && iface_is_internal(br, iface->name)) {
470 iface_set_mac(iface);
476 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
477 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
478 * deletes from 'br' any ports that no longer have any interfaces. */
480 iterate_and_prune_ifaces(struct bridge *br,
481 bool (*cb)(struct bridge *, struct iface *,
487 for (i = 0; i < br->n_ports; ) {
488 struct port *port = br->ports[i];
489 for (j = 0; j < port->n_ifaces; ) {
490 struct iface *iface = port->ifaces[j];
491 if (cb(br, iface, aux)) {
494 iface_destroy(iface);
498 if (port->n_ifaces) {
501 VLOG_ERR("%s port has no interfaces, dropping", port->name);
507 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
508 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
509 * responsible for freeing '*managersp' (with free()).
511 * You may be asking yourself "why does ovs-vswitchd care?", because
512 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
513 * should not be and in fact is not directly involved in that. But
514 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
515 * it has to tell in-band control where the managers are to enable that.
518 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
519 struct sockaddr_in **managersp, size_t *n_managersp)
521 struct sockaddr_in *managers = NULL;
522 size_t n_managers = 0;
524 if (ovs_cfg->n_managers > 0) {
527 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
528 for (i = 0; i < ovs_cfg->n_managers; i++) {
529 const char *name = ovs_cfg->managers[i];
530 struct sockaddr_in *sin = &managers[i];
532 if ((!strncmp(name, "tcp:", 4)
533 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
534 (!strncmp(name, "ssl:", 4)
535 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
541 *managersp = managers;
542 *n_managersp = n_managers;
546 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
548 struct shash old_br, new_br;
549 struct shash_node *node;
550 struct bridge *br, *next;
551 struct sockaddr_in *managers;
554 int sflow_bridge_number;
556 COVERAGE_INC(bridge_reconfigure);
558 collect_managers(ovs_cfg, &managers, &n_managers);
560 /* Collect old and new bridges. */
563 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
564 shash_add(&old_br, br->name, br);
566 for (i = 0; i < ovs_cfg->n_bridges; i++) {
567 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
568 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
569 VLOG_WARN("more than one bridge named %s", br_cfg->name);
573 /* Get rid of deleted bridges and add new bridges. */
574 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
575 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
582 SHASH_FOR_EACH (node, &new_br) {
583 const char *br_name = node->name;
584 const struct ovsrec_bridge *br_cfg = node->data;
585 br = shash_find_data(&old_br, br_name);
587 /* If the bridge datapath type has changed, we need to tear it
588 * down and recreate. */
589 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
591 bridge_create(br_cfg);
594 bridge_create(br_cfg);
597 shash_destroy(&old_br);
598 shash_destroy(&new_br);
602 bridge_configure_ssl(ovs_cfg->ssl);
605 /* Reconfigure all bridges. */
606 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
607 bridge_reconfigure_one(ovs_cfg, br);
610 /* Add and delete ports on all datapaths.
612 * The kernel will reject any attempt to add a given port to a datapath if
613 * that port already belongs to a different datapath, so we must do all
614 * port deletions before any port additions. */
615 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
616 struct odp_port *dpif_ports;
618 struct shash want_ifaces;
620 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
621 bridge_get_all_ifaces(br, &want_ifaces);
622 for (i = 0; i < n_dpif_ports; i++) {
623 const struct odp_port *p = &dpif_ports[i];
624 if (!shash_find(&want_ifaces, p->devname)
625 && strcmp(p->devname, br->name)) {
626 int retval = dpif_port_del(br->dpif, p->port);
628 VLOG_ERR("failed to remove %s interface from %s: %s",
629 p->devname, dpif_name(br->dpif),
634 shash_destroy(&want_ifaces);
637 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
638 struct odp_port *dpif_ports;
640 struct shash cur_ifaces, want_ifaces;
641 struct shash_node *node;
643 /* Get the set of interfaces currently in this datapath. */
644 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
645 shash_init(&cur_ifaces);
646 for (i = 0; i < n_dpif_ports; i++) {
647 const char *name = dpif_ports[i].devname;
648 if (!shash_find(&cur_ifaces, name)) {
649 shash_add(&cur_ifaces, name, NULL);
654 /* Get the set of interfaces we want on this datapath. */
655 bridge_get_all_ifaces(br, &want_ifaces);
657 SHASH_FOR_EACH (node, &want_ifaces) {
658 const char *if_name = node->name;
659 struct iface *iface = node->data;
661 if (shash_find(&cur_ifaces, if_name)) {
662 /* Already exists, just reconfigure it. */
664 reconfigure_iface(iface->cfg, iface);
667 /* Need to add to datapath. */
671 /* Add to datapath. */
672 internal = iface_is_internal(br, if_name);
673 error = dpif_port_add(br->dpif, if_name,
674 internal ? ODP_PORT_INTERNAL : 0, NULL);
675 if (error == EFBIG) {
676 VLOG_ERR("ran out of valid port numbers on %s",
677 dpif_name(br->dpif));
680 VLOG_ERR("failed to add %s interface to %s: %s",
681 if_name, dpif_name(br->dpif), strerror(error));
685 shash_destroy(&cur_ifaces);
686 shash_destroy(&want_ifaces);
688 sflow_bridge_number = 0;
689 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
692 struct iface *local_iface;
693 struct iface *hw_addr_iface;
696 bridge_fetch_dp_ifaces(br);
698 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
699 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
701 /* Pick local port hardware address, datapath ID. */
702 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
703 local_iface = bridge_get_local_iface(br);
705 int error = netdev_set_etheraddr(local_iface->netdev, ea);
707 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
708 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
709 "Ethernet address: %s",
710 br->name, strerror(error));
714 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
715 ofproto_set_datapath_id(br->ofproto, dpid);
717 dpid_string = xasprintf("%016"PRIx64, dpid);
718 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
721 /* Set NetFlow configuration on this bridge. */
722 if (br->cfg->netflow) {
723 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
724 struct netflow_options opts;
726 memset(&opts, 0, sizeof opts);
728 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
729 if (nf_cfg->engine_type) {
730 opts.engine_type = *nf_cfg->engine_type;
732 if (nf_cfg->engine_id) {
733 opts.engine_id = *nf_cfg->engine_id;
736 opts.active_timeout = nf_cfg->active_timeout;
737 if (!opts.active_timeout) {
738 opts.active_timeout = -1;
739 } else if (opts.active_timeout < 0) {
740 VLOG_WARN("bridge %s: active timeout interval set to negative "
741 "value, using default instead (%d seconds)", br->name,
742 NF_ACTIVE_TIMEOUT_DEFAULT);
743 opts.active_timeout = -1;
746 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
747 if (opts.add_id_to_iface) {
748 if (opts.engine_id > 0x7f) {
749 VLOG_WARN("bridge %s: netflow port mangling may conflict "
750 "with another vswitch, choose an engine id less "
751 "than 128", br->name);
753 if (br->n_ports > 508) {
754 VLOG_WARN("bridge %s: netflow port mangling will conflict "
755 "with another port when more than 508 ports are "
760 opts.collectors.n = nf_cfg->n_targets;
761 opts.collectors.names = nf_cfg->targets;
762 if (ofproto_set_netflow(br->ofproto, &opts)) {
763 VLOG_ERR("bridge %s: problem setting netflow collectors",
767 ofproto_set_netflow(br->ofproto, NULL);
770 /* Set sFlow configuration on this bridge. */
771 if (br->cfg->sflow) {
772 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
773 struct ovsrec_controller **controllers;
774 struct ofproto_sflow_options oso;
775 size_t n_controllers;
778 memset(&oso, 0, sizeof oso);
780 oso.targets.n = sflow_cfg->n_targets;
781 oso.targets.names = sflow_cfg->targets;
783 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
784 if (sflow_cfg->sampling) {
785 oso.sampling_rate = *sflow_cfg->sampling;
788 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
789 if (sflow_cfg->polling) {
790 oso.polling_interval = *sflow_cfg->polling;
793 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
794 if (sflow_cfg->header) {
795 oso.header_len = *sflow_cfg->header;
798 oso.sub_id = sflow_bridge_number++;
799 oso.agent_device = sflow_cfg->agent;
801 oso.control_ip = NULL;
802 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
803 for (i = 0; i < n_controllers; i++) {
804 if (controllers[i]->local_ip) {
805 oso.control_ip = controllers[i]->local_ip;
809 ofproto_set_sflow(br->ofproto, &oso);
811 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
813 ofproto_set_sflow(br->ofproto, NULL);
816 /* Update the controller and related settings. It would be more
817 * straightforward to call this from bridge_reconfigure_one(), but we
818 * can't do it there for two reasons. First, and most importantly, at
819 * that point we don't know the dp_ifidx of any interfaces that have
820 * been added to the bridge (because we haven't actually added them to
821 * the datapath). Second, at that point we haven't set the datapath ID
822 * yet; when a controller is configured, resetting the datapath ID will
823 * immediately disconnect from the controller, so it's better to set
824 * the datapath ID before the controller. */
825 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
827 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
828 for (i = 0; i < br->n_ports; i++) {
829 struct port *port = br->ports[i];
832 port_update_vlan_compat(port);
833 port_update_bonding(port);
835 for (j = 0; j < port->n_ifaces; j++) {
836 iface_update_qos(port->ifaces[j], port->cfg->qos);
840 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
841 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
848 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
852 for (i = 0; i < n; i++) {
853 if (!strcmp(keys[i], key)) {
861 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
863 return get_ovsrec_key_value(key,
864 br_cfg->key_other_config,
865 br_cfg->value_other_config,
866 br_cfg->n_other_config);
870 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
871 struct iface **hw_addr_iface)
877 *hw_addr_iface = NULL;
879 /* Did the user request a particular MAC? */
880 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
881 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
882 if (eth_addr_is_multicast(ea)) {
883 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
884 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
885 } else if (eth_addr_is_zero(ea)) {
886 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
892 /* Otherwise choose the minimum non-local MAC address among all of the
894 memset(ea, 0xff, sizeof ea);
895 for (i = 0; i < br->n_ports; i++) {
896 struct port *port = br->ports[i];
897 uint8_t iface_ea[ETH_ADDR_LEN];
900 /* Mirror output ports don't participate. */
901 if (port->is_mirror_output_port) {
905 /* Choose the MAC address to represent the port. */
906 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
907 /* Find the interface with this Ethernet address (if any) so that
908 * we can provide the correct devname to the caller. */
910 for (j = 0; j < port->n_ifaces; j++) {
911 struct iface *candidate = port->ifaces[j];
912 uint8_t candidate_ea[ETH_ADDR_LEN];
913 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
914 && eth_addr_equals(iface_ea, candidate_ea)) {
919 /* Choose the interface whose MAC address will represent the port.
920 * The Linux kernel bonding code always chooses the MAC address of
921 * the first slave added to a bond, and the Fedora networking
922 * scripts always add slaves to a bond in alphabetical order, so
923 * for compatibility we choose the interface with the name that is
924 * first in alphabetical order. */
925 iface = port->ifaces[0];
926 for (j = 1; j < port->n_ifaces; j++) {
927 struct iface *candidate = port->ifaces[j];
928 if (strcmp(candidate->name, iface->name) < 0) {
933 /* The local port doesn't count (since we're trying to choose its
934 * MAC address anyway). */
935 if (iface->dp_ifidx == ODPP_LOCAL) {
940 error = netdev_get_etheraddr(iface->netdev, iface_ea);
942 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
943 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
944 iface->name, strerror(error));
949 /* Compare against our current choice. */
950 if (!eth_addr_is_multicast(iface_ea) &&
951 !eth_addr_is_local(iface_ea) &&
952 !eth_addr_is_reserved(iface_ea) &&
953 !eth_addr_is_zero(iface_ea) &&
954 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
956 memcpy(ea, iface_ea, ETH_ADDR_LEN);
957 *hw_addr_iface = iface;
960 if (eth_addr_is_multicast(ea)) {
961 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
962 *hw_addr_iface = NULL;
963 VLOG_WARN("bridge %s: using default bridge Ethernet "
964 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
966 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
967 br->name, ETH_ADDR_ARGS(ea));
971 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
972 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
973 * an interface on 'br', then that interface must be passed in as
974 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
975 * 'hw_addr_iface' must be passed in as a null pointer. */
977 bridge_pick_datapath_id(struct bridge *br,
978 const uint8_t bridge_ea[ETH_ADDR_LEN],
979 struct iface *hw_addr_iface)
982 * The procedure for choosing a bridge MAC address will, in the most
983 * ordinary case, also choose a unique MAC that we can use as a datapath
984 * ID. In some special cases, though, multiple bridges will end up with
985 * the same MAC address. This is OK for the bridges, but it will confuse
986 * the OpenFlow controller, because each datapath needs a unique datapath
989 * Datapath IDs must be unique. It is also very desirable that they be
990 * stable from one run to the next, so that policy set on a datapath
993 const char *datapath_id;
996 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
997 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1001 if (hw_addr_iface) {
1003 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1005 * A bridge whose MAC address is taken from a VLAN network device
1006 * (that is, a network device created with vconfig(8) or similar
1007 * tool) will have the same MAC address as a bridge on the VLAN
1008 * device's physical network device.
1010 * Handle this case by hashing the physical network device MAC
1011 * along with the VLAN identifier.
1013 uint8_t buf[ETH_ADDR_LEN + 2];
1014 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1015 buf[ETH_ADDR_LEN] = vlan >> 8;
1016 buf[ETH_ADDR_LEN + 1] = vlan;
1017 return dpid_from_hash(buf, sizeof buf);
1020 * Assume that this bridge's MAC address is unique, since it
1021 * doesn't fit any of the cases we handle specially.
1026 * A purely internal bridge, that is, one that has no non-virtual
1027 * network devices on it at all, is more difficult because it has no
1028 * natural unique identifier at all.
1030 * When the host is a XenServer, we handle this case by hashing the
1031 * host's UUID with the name of the bridge. Names of bridges are
1032 * persistent across XenServer reboots, although they can be reused if
1033 * an internal network is destroyed and then a new one is later
1034 * created, so this is fairly effective.
1036 * When the host is not a XenServer, we punt by using a random MAC
1037 * address on each run.
1039 const char *host_uuid = xenserver_get_host_uuid();
1041 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1042 dpid = dpid_from_hash(combined, strlen(combined));
1048 return eth_addr_to_uint64(bridge_ea);
1052 dpid_from_hash(const void *data, size_t n)
1054 uint8_t hash[SHA1_DIGEST_SIZE];
1056 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1057 sha1_bytes(data, n, hash);
1058 eth_addr_mark_random(hash);
1059 return eth_addr_to_uint64(hash);
1063 iface_refresh_stats(struct iface *iface)
1069 static const struct iface_stat iface_stats[] = {
1070 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1071 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1072 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1073 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1074 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1075 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1076 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1077 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1078 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1079 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1080 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1081 { "collisions", offsetof(struct netdev_stats, collisions) },
1083 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1084 const struct iface_stat *s;
1086 char *keys[N_STATS];
1087 int64_t values[N_STATS];
1090 struct netdev_stats stats;
1092 /* Intentionally ignore return value, since errors will set 'stats' to
1093 * all-1s, and we will deal with that correctly below. */
1094 netdev_get_stats(iface->netdev, &stats);
1097 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1098 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1099 if (value != UINT64_MAX) {
1106 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1112 bool datapath_destroyed;
1115 /* Let each bridge do the work that it needs to do. */
1116 datapath_destroyed = false;
1117 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1118 int error = bridge_run_one(br);
1120 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1121 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1122 "forcing reconfiguration", br->name);
1123 datapath_destroyed = true;
1127 /* (Re)configure if necessary. */
1128 if (ovsdb_idl_run(idl) || datapath_destroyed) {
1129 const struct ovsrec_open_vswitch *cfg = ovsrec_open_vswitch_first(idl);
1131 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1133 bridge_configure_once(cfg);
1134 bridge_reconfigure(cfg);
1136 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1137 ovsdb_idl_txn_commit(txn);
1138 ovsdb_idl_txn_destroy(txn); /* XXX */
1140 /* We still need to reconfigure to avoid dangling pointers to
1141 * now-destroyed ovsrec structures inside bridge data. */
1142 static const struct ovsrec_open_vswitch null_cfg;
1144 bridge_reconfigure(&null_cfg);
1148 /* Refresh interface stats if necessary. */
1149 if (time_msec() >= iface_stats_timer) {
1150 struct ovsdb_idl_txn *txn;
1152 txn = ovsdb_idl_txn_create(idl);
1153 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1156 for (i = 0; i < br->n_ports; i++) {
1157 struct port *port = br->ports[i];
1160 for (j = 0; j < port->n_ifaces; j++) {
1161 struct iface *iface = port->ifaces[j];
1162 iface_refresh_stats(iface);
1166 ovsdb_idl_txn_commit(txn);
1167 ovsdb_idl_txn_destroy(txn); /* XXX */
1169 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
1178 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1179 ofproto_wait(br->ofproto);
1180 if (ofproto_has_controller(br->ofproto)) {
1184 mac_learning_wait(br->ml);
1187 ovsdb_idl_wait(idl);
1188 poll_timer_wait_until(iface_stats_timer);
1191 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1192 * configuration changes. */
1194 bridge_flush(struct bridge *br)
1196 COVERAGE_INC(bridge_flush);
1198 mac_learning_flush(br->ml);
1201 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1202 * such interface. */
1203 static struct iface *
1204 bridge_get_local_iface(struct bridge *br)
1208 for (i = 0; i < br->n_ports; i++) {
1209 struct port *port = br->ports[i];
1210 for (j = 0; j < port->n_ifaces; j++) {
1211 struct iface *iface = port->ifaces[j];
1212 if (iface->dp_ifidx == ODPP_LOCAL) {
1221 /* Bridge unixctl user interface functions. */
1223 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1224 const char *args, void *aux OVS_UNUSED)
1226 struct ds ds = DS_EMPTY_INITIALIZER;
1227 const struct bridge *br;
1228 const struct mac_entry *e;
1230 br = bridge_lookup(args);
1232 unixctl_command_reply(conn, 501, "no such bridge");
1236 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1237 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1238 if (e->port < 0 || e->port >= br->n_ports) {
1241 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1242 br->ports[e->port]->ifaces[0]->dp_ifidx,
1243 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1245 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1249 /* Bridge reconfiguration functions. */
1250 static struct bridge *
1251 bridge_create(const struct ovsrec_bridge *br_cfg)
1256 assert(!bridge_lookup(br_cfg->name));
1257 br = xzalloc(sizeof *br);
1259 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1265 dpif_flow_flush(br->dpif);
1267 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1270 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1272 dpif_delete(br->dpif);
1273 dpif_close(br->dpif);
1278 br->name = xstrdup(br_cfg->name);
1280 br->ml = mac_learning_create();
1281 eth_addr_nicira_random(br->default_ea);
1283 port_array_init(&br->ifaces);
1285 shash_init(&br->port_by_name);
1286 shash_init(&br->iface_by_name);
1290 list_push_back(&all_bridges, &br->node);
1292 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1298 bridge_destroy(struct bridge *br)
1303 while (br->n_ports > 0) {
1304 port_destroy(br->ports[br->n_ports - 1]);
1306 list_remove(&br->node);
1307 error = dpif_delete(br->dpif);
1308 if (error && error != ENOENT) {
1309 VLOG_ERR("failed to delete %s: %s",
1310 dpif_name(br->dpif), strerror(error));
1312 dpif_close(br->dpif);
1313 ofproto_destroy(br->ofproto);
1314 mac_learning_destroy(br->ml);
1315 port_array_destroy(&br->ifaces);
1316 shash_destroy(&br->port_by_name);
1317 shash_destroy(&br->iface_by_name);
1324 static struct bridge *
1325 bridge_lookup(const char *name)
1329 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1330 if (!strcmp(br->name, name)) {
1337 /* Handle requests for a listing of all flows known by the OpenFlow
1338 * stack, including those normally hidden. */
1340 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1341 const char *args, void *aux OVS_UNUSED)
1346 br = bridge_lookup(args);
1348 unixctl_command_reply(conn, 501, "Unknown bridge");
1353 ofproto_get_all_flows(br->ofproto, &results);
1355 unixctl_command_reply(conn, 200, ds_cstr(&results));
1356 ds_destroy(&results);
1359 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1360 * connections and reconnect. If BRIDGE is not specified, then all bridges
1361 * drop their controller connections and reconnect. */
1363 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1364 const char *args, void *aux OVS_UNUSED)
1367 if (args[0] != '\0') {
1368 br = bridge_lookup(args);
1370 unixctl_command_reply(conn, 501, "Unknown bridge");
1373 ofproto_reconnect_controllers(br->ofproto);
1375 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1376 ofproto_reconnect_controllers(br->ofproto);
1379 unixctl_command_reply(conn, 200, NULL);
1383 bridge_run_one(struct bridge *br)
1387 error = ofproto_run1(br->ofproto);
1392 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1395 error = ofproto_run2(br->ofproto, br->flush);
1402 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1403 const struct bridge *br,
1404 struct ovsrec_controller ***controllersp)
1406 struct ovsrec_controller **controllers;
1407 size_t n_controllers;
1409 if (br->cfg->n_controller) {
1410 controllers = br->cfg->controller;
1411 n_controllers = br->cfg->n_controller;
1413 controllers = ovs_cfg->controller;
1414 n_controllers = ovs_cfg->n_controller;
1417 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1423 *controllersp = controllers;
1425 return n_controllers;
1429 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1432 struct shash old_ports, new_ports;
1433 struct svec listeners, old_listeners;
1434 struct svec snoops, old_snoops;
1435 struct shash_node *node;
1438 /* Collect old ports. */
1439 shash_init(&old_ports);
1440 for (i = 0; i < br->n_ports; i++) {
1441 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1444 /* Collect new ports. */
1445 shash_init(&new_ports);
1446 for (i = 0; i < br->cfg->n_ports; i++) {
1447 const char *name = br->cfg->ports[i]->name;
1448 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1449 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1454 /* If we have a controller, then we need a local port. Complain if the
1455 * user didn't specify one.
1457 * XXX perhaps we should synthesize a port ourselves in this case. */
1458 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1459 char local_name[IF_NAMESIZE];
1462 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1463 local_name, sizeof local_name);
1464 if (!error && !shash_find(&new_ports, local_name)) {
1465 VLOG_WARN("bridge %s: controller specified but no local port "
1466 "(port named %s) defined",
1467 br->name, local_name);
1471 /* Get rid of deleted ports.
1472 * Get rid of deleted interfaces on ports that still exist. */
1473 SHASH_FOR_EACH (node, &old_ports) {
1474 struct port *port = node->data;
1475 const struct ovsrec_port *port_cfg;
1477 port_cfg = shash_find_data(&new_ports, node->name);
1481 port_del_ifaces(port, port_cfg);
1485 /* Create new ports.
1486 * Add new interfaces to existing ports.
1487 * Reconfigure existing ports. */
1488 SHASH_FOR_EACH (node, &new_ports) {
1489 struct port *port = shash_find_data(&old_ports, node->name);
1491 port = port_create(br, node->name);
1494 port_reconfigure(port, node->data);
1495 if (!port->n_ifaces) {
1496 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1497 br->name, port->name);
1501 shash_destroy(&old_ports);
1502 shash_destroy(&new_ports);
1504 /* Delete all flows if we're switching from connected to standalone or vice
1505 * versa. (XXX Should we delete all flows if we are switching from one
1506 * controller to another?) */
1508 /* Configure OpenFlow management listener. */
1509 svec_init(&listeners);
1510 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1511 ovs_rundir, br->name));
1512 svec_init(&old_listeners);
1513 ofproto_get_listeners(br->ofproto, &old_listeners);
1514 if (!svec_equal(&listeners, &old_listeners)) {
1515 ofproto_set_listeners(br->ofproto, &listeners);
1517 svec_destroy(&listeners);
1518 svec_destroy(&old_listeners);
1520 /* Configure OpenFlow controller connection snooping. */
1522 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1523 ovs_rundir, br->name));
1524 svec_init(&old_snoops);
1525 ofproto_get_snoops(br->ofproto, &old_snoops);
1526 if (!svec_equal(&snoops, &old_snoops)) {
1527 ofproto_set_snoops(br->ofproto, &snoops);
1529 svec_destroy(&snoops);
1530 svec_destroy(&old_snoops);
1532 mirror_reconfigure(br);
1536 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1538 const struct sockaddr_in *managers,
1541 struct ovsrec_controller **controllers;
1542 size_t n_controllers;
1544 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1546 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1547 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1548 ofproto_flush_flows(br->ofproto);
1551 if (!n_controllers) {
1552 union ofp_action action;
1555 /* Clear out controllers. */
1556 ofproto_set_controllers(br->ofproto, NULL, 0);
1558 /* Set up a flow that matches every packet and directs them to
1559 * OFPP_NORMAL (which goes to us). */
1560 memset(&action, 0, sizeof action);
1561 action.type = htons(OFPAT_OUTPUT);
1562 action.output.len = htons(sizeof action);
1563 action.output.port = htons(OFPP_NORMAL);
1564 memset(&flow, 0, sizeof flow);
1565 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1567 struct ofproto_controller *ocs;
1570 ocs = xmalloc(n_controllers * sizeof *ocs);
1571 for (i = 0; i < n_controllers; i++) {
1572 struct ovsrec_controller *c = controllers[i];
1573 struct ofproto_controller *oc = &ocs[i];
1575 if (strcmp(c->target, "discover")) {
1576 struct iface *local_iface;
1579 local_iface = bridge_get_local_iface(br);
1580 if (local_iface && c->local_ip
1581 && inet_aton(c->local_ip, &ip)) {
1582 struct netdev *netdev = local_iface->netdev;
1583 struct in_addr mask, gateway;
1585 if (!c->local_netmask
1586 || !inet_aton(c->local_netmask, &mask)) {
1589 if (!c->local_gateway
1590 || !inet_aton(c->local_gateway, &gateway)) {
1594 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1596 mask.s_addr = guess_netmask(ip.s_addr);
1598 if (!netdev_set_in4(netdev, ip, mask)) {
1599 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1601 br->name, IP_ARGS(&ip.s_addr),
1602 IP_ARGS(&mask.s_addr));
1605 if (gateway.s_addr) {
1606 if (!netdev_add_router(netdev, gateway)) {
1607 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1608 br->name, IP_ARGS(&gateway.s_addr));
1614 oc->target = c->target;
1615 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1616 oc->probe_interval = (c->inactivity_probe
1617 ? *c->inactivity_probe / 1000 : 5);
1618 oc->fail = (!c->fail_mode
1619 || !strcmp(c->fail_mode, "standalone")
1620 || !strcmp(c->fail_mode, "open")
1621 ? OFPROTO_FAIL_STANDALONE
1622 : OFPROTO_FAIL_SECURE);
1623 oc->band = (!c->connection_mode
1624 || !strcmp(c->connection_mode, "in-band")
1626 : OFPROTO_OUT_OF_BAND);
1627 oc->accept_re = c->discover_accept_regex;
1628 oc->update_resolv_conf = c->discover_update_resolv_conf;
1629 oc->rate_limit = (c->controller_rate_limit
1630 ? *c->controller_rate_limit : 0);
1631 oc->burst_limit = (c->controller_burst_limit
1632 ? *c->controller_burst_limit : 0);
1634 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1640 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1645 for (i = 0; i < br->n_ports; i++) {
1646 struct port *port = br->ports[i];
1647 for (j = 0; j < port->n_ifaces; j++) {
1648 struct iface *iface = port->ifaces[j];
1649 shash_add_once(ifaces, iface->name, iface);
1651 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1652 shash_add_once(ifaces, port->name, NULL);
1657 /* For robustness, in case the administrator moves around datapath ports behind
1658 * our back, we re-check all the datapath port numbers here.
1660 * This function will set the 'dp_ifidx' members of interfaces that have
1661 * disappeared to -1, so only call this function from a context where those
1662 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1663 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1664 * datapath, which doesn't support UINT16_MAX+1 ports. */
1666 bridge_fetch_dp_ifaces(struct bridge *br)
1668 struct odp_port *dpif_ports;
1669 size_t n_dpif_ports;
1672 /* Reset all interface numbers. */
1673 for (i = 0; i < br->n_ports; i++) {
1674 struct port *port = br->ports[i];
1675 for (j = 0; j < port->n_ifaces; j++) {
1676 struct iface *iface = port->ifaces[j];
1677 iface->dp_ifidx = -1;
1680 port_array_clear(&br->ifaces);
1682 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1683 for (i = 0; i < n_dpif_ports; i++) {
1684 struct odp_port *p = &dpif_ports[i];
1685 struct iface *iface = iface_lookup(br, p->devname);
1687 if (iface->dp_ifidx >= 0) {
1688 VLOG_WARN("%s reported interface %s twice",
1689 dpif_name(br->dpif), p->devname);
1690 } else if (iface_from_dp_ifidx(br, p->port)) {
1691 VLOG_WARN("%s reported interface %"PRIu16" twice",
1692 dpif_name(br->dpif), p->port);
1694 port_array_set(&br->ifaces, p->port, iface);
1695 iface->dp_ifidx = p->port;
1699 int64_t ofport = (iface->dp_ifidx >= 0
1700 ? odp_port_to_ofp_port(iface->dp_ifidx)
1702 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1709 /* Bridge packet processing functions. */
1712 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1714 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1717 static struct bond_entry *
1718 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1720 return &port->bond_hash[bond_hash(mac)];
1724 bond_choose_iface(const struct port *port)
1726 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1727 size_t i, best_down_slave = -1;
1728 long long next_delay_expiration = LLONG_MAX;
1730 for (i = 0; i < port->n_ifaces; i++) {
1731 struct iface *iface = port->ifaces[i];
1733 if (iface->enabled) {
1735 } else if (iface->delay_expires < next_delay_expiration) {
1736 best_down_slave = i;
1737 next_delay_expiration = iface->delay_expires;
1741 if (best_down_slave != -1) {
1742 struct iface *iface = port->ifaces[best_down_slave];
1744 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1745 "since no other interface is up", iface->name,
1746 iface->delay_expires - time_msec());
1747 bond_enable_slave(iface, true);
1750 return best_down_slave;
1754 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1755 uint16_t *dp_ifidx, tag_type *tags)
1757 struct iface *iface;
1759 assert(port->n_ifaces);
1760 if (port->n_ifaces == 1) {
1761 iface = port->ifaces[0];
1763 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1764 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1765 || !port->ifaces[e->iface_idx]->enabled) {
1766 /* XXX select interface properly. The current interface selection
1767 * is only good for testing the rebalancing code. */
1768 e->iface_idx = bond_choose_iface(port);
1769 if (e->iface_idx < 0) {
1770 *tags |= port->no_ifaces_tag;
1773 e->iface_tag = tag_create_random();
1774 ((struct port *) port)->bond_compat_is_stale = true;
1776 *tags |= e->iface_tag;
1777 iface = port->ifaces[e->iface_idx];
1779 *dp_ifidx = iface->dp_ifidx;
1780 *tags |= iface->tag; /* Currently only used for bonding. */
1785 bond_link_status_update(struct iface *iface, bool carrier)
1787 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1788 struct port *port = iface->port;
1790 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1791 /* Nothing to do. */
1794 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1795 iface->name, carrier ? "detected" : "dropped");
1796 if (carrier == iface->enabled) {
1797 iface->delay_expires = LLONG_MAX;
1798 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1799 iface->name, carrier ? "disabled" : "enabled");
1800 } else if (carrier && port->active_iface < 0) {
1801 bond_enable_slave(iface, true);
1802 if (port->updelay) {
1803 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1804 "other interface is up", iface->name, port->updelay);
1807 int delay = carrier ? port->updelay : port->downdelay;
1808 iface->delay_expires = time_msec() + delay;
1811 "interface %s: will be %s if it stays %s for %d ms",
1813 carrier ? "enabled" : "disabled",
1814 carrier ? "up" : "down",
1821 bond_choose_active_iface(struct port *port)
1823 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1825 port->active_iface = bond_choose_iface(port);
1826 port->active_iface_tag = tag_create_random();
1827 if (port->active_iface >= 0) {
1828 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1829 port->name, port->ifaces[port->active_iface]->name);
1831 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1837 bond_enable_slave(struct iface *iface, bool enable)
1839 struct port *port = iface->port;
1840 struct bridge *br = port->bridge;
1842 /* This acts as a recursion check. If the act of disabling a slave
1843 * causes a different slave to be enabled, the flag will allow us to
1844 * skip redundant work when we reenter this function. It must be
1845 * cleared on exit to keep things safe with multiple bonds. */
1846 static bool moving_active_iface = false;
1848 iface->delay_expires = LLONG_MAX;
1849 if (enable == iface->enabled) {
1853 iface->enabled = enable;
1854 if (!iface->enabled) {
1855 VLOG_WARN("interface %s: disabled", iface->name);
1856 ofproto_revalidate(br->ofproto, iface->tag);
1857 if (iface->port_ifidx == port->active_iface) {
1858 ofproto_revalidate(br->ofproto,
1859 port->active_iface_tag);
1861 /* Disabling a slave can lead to another slave being immediately
1862 * enabled if there will be no active slaves but one is waiting
1863 * on an updelay. In this case we do not need to run most of the
1864 * code for the newly enabled slave since there was no period
1865 * without an active slave and it is redundant with the disabling
1867 moving_active_iface = true;
1868 bond_choose_active_iface(port);
1870 bond_send_learning_packets(port);
1872 VLOG_WARN("interface %s: enabled", iface->name);
1873 if (port->active_iface < 0 && !moving_active_iface) {
1874 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1875 bond_choose_active_iface(port);
1876 bond_send_learning_packets(port);
1878 iface->tag = tag_create_random();
1881 moving_active_iface = false;
1882 port->bond_compat_is_stale = true;
1885 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1886 * bond interface. */
1888 bond_update_fake_iface_stats(struct port *port)
1890 struct netdev_stats bond_stats;
1891 struct netdev *bond_dev;
1894 memset(&bond_stats, 0, sizeof bond_stats);
1896 for (i = 0; i < port->n_ifaces; i++) {
1897 struct netdev_stats slave_stats;
1899 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1900 /* XXX: We swap the stats here because they are swapped back when
1901 * reported by the internal device. The reason for this is
1902 * internal devices normally represent packets going into the system
1903 * but when used as fake bond device they represent packets leaving
1904 * the system. We really should do this in the internal device
1905 * itself because changing it here reverses the counts from the
1906 * perspective of the switch. However, the internal device doesn't
1907 * know what type of device it represents so we have to do it here
1909 bond_stats.tx_packets += slave_stats.rx_packets;
1910 bond_stats.tx_bytes += slave_stats.rx_bytes;
1911 bond_stats.rx_packets += slave_stats.tx_packets;
1912 bond_stats.rx_bytes += slave_stats.tx_bytes;
1916 if (!netdev_open_default(port->name, &bond_dev)) {
1917 netdev_set_stats(bond_dev, &bond_stats);
1918 netdev_close(bond_dev);
1923 bond_run(struct bridge *br)
1927 for (i = 0; i < br->n_ports; i++) {
1928 struct port *port = br->ports[i];
1930 if (port->n_ifaces >= 2) {
1931 for (j = 0; j < port->n_ifaces; j++) {
1932 struct iface *iface = port->ifaces[j];
1933 if (time_msec() >= iface->delay_expires) {
1934 bond_enable_slave(iface, !iface->enabled);
1938 if (port->bond_fake_iface
1939 && time_msec() >= port->bond_next_fake_iface_update) {
1940 bond_update_fake_iface_stats(port);
1941 port->bond_next_fake_iface_update = time_msec() + 1000;
1945 if (port->bond_compat_is_stale) {
1946 port->bond_compat_is_stale = false;
1947 port_update_bond_compat(port);
1953 bond_wait(struct bridge *br)
1957 for (i = 0; i < br->n_ports; i++) {
1958 struct port *port = br->ports[i];
1959 if (port->n_ifaces < 2) {
1962 for (j = 0; j < port->n_ifaces; j++) {
1963 struct iface *iface = port->ifaces[j];
1964 if (iface->delay_expires != LLONG_MAX) {
1965 poll_timer_wait_until(iface->delay_expires);
1968 if (port->bond_fake_iface) {
1969 poll_timer_wait_until(port->bond_next_fake_iface_update);
1975 set_dst(struct dst *p, const flow_t *flow,
1976 const struct port *in_port, const struct port *out_port,
1979 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1980 : in_port->vlan >= 0 ? in_port->vlan
1981 : ntohs(flow->dl_vlan));
1982 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1986 swap_dst(struct dst *p, struct dst *q)
1988 struct dst tmp = *p;
1993 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1994 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1995 * that we push to the datapath. We could in fact fully sort the array by
1996 * vlan, but in most cases there are at most two different vlan tags so that's
1997 * possibly overkill.) */
1999 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2001 struct dst *first = dsts;
2002 struct dst *last = dsts + n_dsts;
2004 while (first != last) {
2006 * - All dsts < first have vlan == 'vlan'.
2007 * - All dsts >= last have vlan != 'vlan'.
2008 * - first < last. */
2009 while (first->vlan == vlan) {
2010 if (++first == last) {
2015 /* Same invariants, plus one additional:
2016 * - first->vlan != vlan.
2018 while (last[-1].vlan != vlan) {
2019 if (--last == first) {
2024 /* Same invariants, plus one additional:
2025 * - last[-1].vlan == vlan.*/
2026 swap_dst(first++, --last);
2031 mirror_mask_ffs(mirror_mask_t mask)
2033 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2038 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2039 const struct dst *test)
2042 for (i = 0; i < n_dsts; i++) {
2043 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2051 port_trunks_vlan(const struct port *port, uint16_t vlan)
2053 return (port->vlan < 0
2054 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2058 port_includes_vlan(const struct port *port, uint16_t vlan)
2060 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2064 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2065 const struct port *in_port, const struct port *out_port,
2066 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2068 mirror_mask_t mirrors = in_port->src_mirrors;
2069 struct dst *dst = dsts;
2072 if (out_port == FLOOD_PORT) {
2073 /* XXX use ODP_FLOOD if no vlans or bonding. */
2074 /* XXX even better, define each VLAN as a datapath port group */
2075 for (i = 0; i < br->n_ports; i++) {
2076 struct port *port = br->ports[i];
2077 if (port != in_port && port_includes_vlan(port, vlan)
2078 && !port->is_mirror_output_port
2079 && set_dst(dst, flow, in_port, port, tags)) {
2080 mirrors |= port->dst_mirrors;
2084 *nf_output_iface = NF_OUT_FLOOD;
2085 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2086 *nf_output_iface = dst->dp_ifidx;
2087 mirrors |= out_port->dst_mirrors;
2092 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2093 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2095 if (set_dst(dst, flow, in_port, m->out_port, tags)
2096 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2100 for (i = 0; i < br->n_ports; i++) {
2101 struct port *port = br->ports[i];
2102 if (port_includes_vlan(port, m->out_vlan)
2103 && set_dst(dst, flow, in_port, port, tags))
2107 if (port->vlan < 0) {
2108 dst->vlan = m->out_vlan;
2110 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2114 /* Use the vlan tag on the original flow instead of
2115 * the one passed in the vlan parameter. This ensures
2116 * that we compare the vlan from before any implicit
2117 * tagging tags place. This is necessary because
2118 * dst->vlan is the final vlan, after removing implicit
2120 flow_vlan = ntohs(flow->dl_vlan);
2121 if (flow_vlan == 0) {
2122 flow_vlan = OFP_VLAN_NONE;
2124 if (port == in_port && dst->vlan == flow_vlan) {
2125 /* Don't send out input port on same VLAN. */
2133 mirrors &= mirrors - 1;
2136 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2140 static void OVS_UNUSED
2141 print_dsts(const struct dst *dsts, size_t n)
2143 for (; n--; dsts++) {
2144 printf(">p%"PRIu16, dsts->dp_ifidx);
2145 if (dsts->vlan != OFP_VLAN_NONE) {
2146 printf("v%"PRIu16, dsts->vlan);
2152 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2153 const struct port *in_port, const struct port *out_port,
2154 tag_type *tags, struct odp_actions *actions,
2155 uint16_t *nf_output_iface)
2157 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2159 const struct dst *p;
2162 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2165 cur_vlan = ntohs(flow->dl_vlan);
2166 for (p = dsts; p < &dsts[n_dsts]; p++) {
2167 union odp_action *a;
2168 if (p->vlan != cur_vlan) {
2169 if (p->vlan == OFP_VLAN_NONE) {
2170 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2172 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2173 a->vlan_vid.vlan_vid = htons(p->vlan);
2177 a = odp_actions_add(actions, ODPAT_OUTPUT);
2178 a->output.port = p->dp_ifidx;
2182 /* Returns the effective vlan of a packet, taking into account both the
2183 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2184 * the packet is untagged and -1 indicates it has an invalid header and
2185 * should be dropped. */
2186 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2187 struct port *in_port, bool have_packet)
2189 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2190 * belongs to VLAN 0, so we should treat both cases identically. (In the
2191 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2192 * presumably to allow a priority to be specified. In the latter case, the
2193 * packet does not have any 802.1Q header.) */
2194 int vlan = ntohs(flow->dl_vlan);
2195 if (vlan == OFP_VLAN_NONE) {
2198 if (in_port->vlan >= 0) {
2200 /* XXX support double tagging? */
2202 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2203 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2204 "packet received on port %s configured with "
2205 "implicit VLAN %"PRIu16,
2206 br->name, ntohs(flow->dl_vlan),
2207 in_port->name, in_port->vlan);
2211 vlan = in_port->vlan;
2213 if (!port_includes_vlan(in_port, vlan)) {
2215 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2216 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2217 "packet received on port %s not configured for "
2219 br->name, vlan, in_port->name, vlan);
2228 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2229 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2230 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2232 is_gratuitous_arp(const flow_t *flow)
2234 return (flow->dl_type == htons(ETH_TYPE_ARP)
2235 && eth_addr_is_broadcast(flow->dl_dst)
2236 && (flow->nw_proto == ARP_OP_REPLY
2237 || (flow->nw_proto == ARP_OP_REQUEST
2238 && flow->nw_src == flow->nw_dst)));
2242 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2243 struct port *in_port)
2245 enum grat_arp_lock_type lock_type;
2248 /* We don't want to learn from gratuitous ARP packets that are reflected
2249 * back over bond slaves so we lock the learning table. */
2250 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2251 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2252 GRAT_ARP_LOCK_CHECK;
2254 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2257 /* The log messages here could actually be useful in debugging,
2258 * so keep the rate limit relatively high. */
2259 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2261 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2262 "on port %s in VLAN %d",
2263 br->name, ETH_ADDR_ARGS(flow->dl_src),
2264 in_port->name, vlan);
2265 ofproto_revalidate(br->ofproto, rev_tag);
2269 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2270 * dropped. Returns true if they may be forwarded, false if they should be
2273 * If 'have_packet' is true, it indicates that the caller is processing a
2274 * received packet. If 'have_packet' is false, then the caller is just
2275 * revalidating an existing flow because configuration has changed. Either
2276 * way, 'have_packet' only affects logging (there is no point in logging errors
2277 * during revalidation).
2279 * Sets '*in_portp' to the input port. This will be a null pointer if
2280 * flow->in_port does not designate a known input port (in which case
2281 * is_admissible() returns false).
2283 * When returning true, sets '*vlanp' to the effective VLAN of the input
2284 * packet, as returned by flow_get_vlan().
2286 * May also add tags to '*tags', although the current implementation only does
2287 * so in one special case.
2290 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2291 tag_type *tags, int *vlanp, struct port **in_portp)
2293 struct iface *in_iface;
2294 struct port *in_port;
2297 /* Find the interface and port structure for the received packet. */
2298 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2300 /* No interface? Something fishy... */
2302 /* Odd. A few possible reasons here:
2304 * - We deleted an interface but there are still a few packets
2305 * queued up from it.
2307 * - Someone externally added an interface (e.g. with "ovs-dpctl
2308 * add-if") that we don't know about.
2310 * - Packet arrived on the local port but the local port is not
2311 * one of our bridge ports.
2313 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2315 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2316 "interface %"PRIu16, br->name, flow->in_port);
2322 *in_portp = in_port = in_iface->port;
2323 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2328 /* Drop frames for reserved multicast addresses. */
2329 if (eth_addr_is_reserved(flow->dl_dst)) {
2333 /* Drop frames on ports reserved for mirroring. */
2334 if (in_port->is_mirror_output_port) {
2336 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2337 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2338 "%s, which is reserved exclusively for mirroring",
2339 br->name, in_port->name);
2344 /* Packets received on bonds need special attention to avoid duplicates. */
2345 if (in_port->n_ifaces > 1) {
2347 bool is_grat_arp_locked;
2349 if (eth_addr_is_multicast(flow->dl_dst)) {
2350 *tags |= in_port->active_iface_tag;
2351 if (in_port->active_iface != in_iface->port_ifidx) {
2352 /* Drop all multicast packets on inactive slaves. */
2357 /* Drop all packets for which we have learned a different input
2358 * port, because we probably sent the packet on one slave and got
2359 * it back on the other. Gratuitous ARP packets are an exception
2360 * to this rule: the host has moved to another switch. The exception
2361 * to the exception is if we locked the learning table to avoid
2362 * reflections on bond slaves. If this is the case, just drop the
2364 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2365 &is_grat_arp_locked);
2366 if (src_idx != -1 && src_idx != in_port->port_idx &&
2367 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2375 /* If the composed actions may be applied to any packet in the given 'flow',
2376 * returns true. Otherwise, the actions should only be applied to 'packet', or
2377 * not at all, if 'packet' was NULL. */
2379 process_flow(struct bridge *br, const flow_t *flow,
2380 const struct ofpbuf *packet, struct odp_actions *actions,
2381 tag_type *tags, uint16_t *nf_output_iface)
2383 struct port *in_port;
2384 struct port *out_port;
2388 /* Check whether we should drop packets in this flow. */
2389 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2394 /* Learn source MAC (but don't try to learn from revalidation). */
2396 update_learning_table(br, flow, vlan, in_port);
2399 /* Determine output port. */
2400 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2402 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2403 out_port = br->ports[out_port_idx];
2404 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2405 /* If we are revalidating but don't have a learning entry then
2406 * eject the flow. Installing a flow that floods packets opens
2407 * up a window of time where we could learn from a packet reflected
2408 * on a bond and blackhole packets before the learning table is
2409 * updated to reflect the correct port. */
2412 out_port = FLOOD_PORT;
2415 /* Don't send packets out their input ports. */
2416 if (in_port == out_port) {
2422 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2429 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2432 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2433 const struct ofp_phy_port *opp,
2436 struct bridge *br = br_;
2437 struct iface *iface;
2440 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2446 if (reason == OFPPR_DELETE) {
2447 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2448 br->name, iface->name);
2449 iface_destroy(iface);
2450 if (!port->n_ifaces) {
2451 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2452 br->name, port->name);
2458 if (port->n_ifaces > 1) {
2459 bool up = !(opp->state & OFPPS_LINK_DOWN);
2460 bond_link_status_update(iface, up);
2461 port_update_bond_compat(port);
2467 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2468 struct odp_actions *actions, tag_type *tags,
2469 uint16_t *nf_output_iface, void *br_)
2471 struct bridge *br = br_;
2473 COVERAGE_INC(bridge_process_flow);
2474 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2478 bridge_account_flow_ofhook_cb(const flow_t *flow,
2479 const union odp_action *actions,
2480 size_t n_actions, unsigned long long int n_bytes,
2483 struct bridge *br = br_;
2484 const union odp_action *a;
2485 struct port *in_port;
2489 /* Feed information from the active flows back into the learning table
2490 * to ensure that table is always in sync with what is actually flowing
2491 * through the datapath. */
2492 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2493 update_learning_table(br, flow, vlan, in_port);
2496 if (!br->has_bonded_ports) {
2500 for (a = actions; a < &actions[n_actions]; a++) {
2501 if (a->type == ODPAT_OUTPUT) {
2502 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2503 if (out_port && out_port->n_ifaces >= 2) {
2504 struct bond_entry *e = lookup_bond_entry(out_port,
2506 e->tx_bytes += n_bytes;
2513 bridge_account_checkpoint_ofhook_cb(void *br_)
2515 struct bridge *br = br_;
2519 if (!br->has_bonded_ports) {
2524 for (i = 0; i < br->n_ports; i++) {
2525 struct port *port = br->ports[i];
2526 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2527 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2528 bond_rebalance_port(port);
2533 static struct ofhooks bridge_ofhooks = {
2534 bridge_port_changed_ofhook_cb,
2535 bridge_normal_ofhook_cb,
2536 bridge_account_flow_ofhook_cb,
2537 bridge_account_checkpoint_ofhook_cb,
2540 /* Bonding functions. */
2542 /* Statistics for a single interface on a bonded port, used for load-based
2543 * bond rebalancing. */
2544 struct slave_balance {
2545 struct iface *iface; /* The interface. */
2546 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2548 /* All the "bond_entry"s that are assigned to this interface, in order of
2549 * increasing tx_bytes. */
2550 struct bond_entry **hashes;
2554 /* Sorts pointers to pointers to bond_entries in ascending order by the
2555 * interface to which they are assigned, and within a single interface in
2556 * ascending order of bytes transmitted. */
2558 compare_bond_entries(const void *a_, const void *b_)
2560 const struct bond_entry *const *ap = a_;
2561 const struct bond_entry *const *bp = b_;
2562 const struct bond_entry *a = *ap;
2563 const struct bond_entry *b = *bp;
2564 if (a->iface_idx != b->iface_idx) {
2565 return a->iface_idx > b->iface_idx ? 1 : -1;
2566 } else if (a->tx_bytes != b->tx_bytes) {
2567 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2573 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2574 * *descending* order by number of bytes transmitted. */
2576 compare_slave_balance(const void *a_, const void *b_)
2578 const struct slave_balance *a = a_;
2579 const struct slave_balance *b = b_;
2580 if (a->iface->enabled != b->iface->enabled) {
2581 return a->iface->enabled ? -1 : 1;
2582 } else if (a->tx_bytes != b->tx_bytes) {
2583 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2590 swap_bals(struct slave_balance *a, struct slave_balance *b)
2592 struct slave_balance tmp = *a;
2597 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2598 * given that 'p' (and only 'p') might be in the wrong location.
2600 * This function invalidates 'p', since it might now be in a different memory
2603 resort_bals(struct slave_balance *p,
2604 struct slave_balance bals[], size_t n_bals)
2607 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2608 swap_bals(p, p - 1);
2610 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2611 swap_bals(p, p + 1);
2617 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2619 if (VLOG_IS_DBG_ENABLED()) {
2620 struct ds ds = DS_EMPTY_INITIALIZER;
2621 const struct slave_balance *b;
2623 for (b = bals; b < bals + n_bals; b++) {
2627 ds_put_char(&ds, ',');
2629 ds_put_format(&ds, " %s %"PRIu64"kB",
2630 b->iface->name, b->tx_bytes / 1024);
2632 if (!b->iface->enabled) {
2633 ds_put_cstr(&ds, " (disabled)");
2635 if (b->n_hashes > 0) {
2636 ds_put_cstr(&ds, " (");
2637 for (i = 0; i < b->n_hashes; i++) {
2638 const struct bond_entry *e = b->hashes[i];
2640 ds_put_cstr(&ds, " + ");
2642 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2643 e - port->bond_hash, e->tx_bytes / 1024);
2645 ds_put_cstr(&ds, ")");
2648 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2653 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2655 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2658 struct bond_entry *hash = from->hashes[hash_idx];
2659 struct port *port = from->iface->port;
2660 uint64_t delta = hash->tx_bytes;
2662 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2663 "from %s to %s (now carrying %"PRIu64"kB and "
2664 "%"PRIu64"kB load, respectively)",
2665 port->name, delta / 1024, hash - port->bond_hash,
2666 from->iface->name, to->iface->name,
2667 (from->tx_bytes - delta) / 1024,
2668 (to->tx_bytes + delta) / 1024);
2670 /* Delete element from from->hashes.
2672 * We don't bother to add the element to to->hashes because not only would
2673 * it require more work, the only purpose it would be to allow that hash to
2674 * be migrated to another slave in this rebalancing run, and there is no
2675 * point in doing that. */
2676 if (hash_idx == 0) {
2679 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2680 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2684 /* Shift load away from 'from' to 'to'. */
2685 from->tx_bytes -= delta;
2686 to->tx_bytes += delta;
2688 /* Arrange for flows to be revalidated. */
2689 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2690 hash->iface_idx = to->iface->port_ifidx;
2691 hash->iface_tag = tag_create_random();
2695 bond_rebalance_port(struct port *port)
2697 struct slave_balance bals[DP_MAX_PORTS];
2699 struct bond_entry *hashes[BOND_MASK + 1];
2700 struct slave_balance *b, *from, *to;
2701 struct bond_entry *e;
2704 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2705 * descending order of tx_bytes, so that bals[0] represents the most
2706 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2709 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2710 * array for each slave_balance structure, we sort our local array of
2711 * hashes in order by slave, so that all of the hashes for a given slave
2712 * become contiguous in memory, and then we point each 'hashes' members of
2713 * a slave_balance structure to the start of a contiguous group. */
2714 n_bals = port->n_ifaces;
2715 for (b = bals; b < &bals[n_bals]; b++) {
2716 b->iface = port->ifaces[b - bals];
2721 for (i = 0; i <= BOND_MASK; i++) {
2722 hashes[i] = &port->bond_hash[i];
2724 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2725 for (i = 0; i <= BOND_MASK; i++) {
2727 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2728 b = &bals[e->iface_idx];
2729 b->tx_bytes += e->tx_bytes;
2731 b->hashes = &hashes[i];
2736 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2737 log_bals(bals, n_bals, port);
2739 /* Discard slaves that aren't enabled (which were sorted to the back of the
2740 * array earlier). */
2741 while (!bals[n_bals - 1].iface->enabled) {
2748 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2749 to = &bals[n_bals - 1];
2750 for (from = bals; from < to; ) {
2751 uint64_t overload = from->tx_bytes - to->tx_bytes;
2752 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2753 /* The extra load on 'from' (and all less-loaded slaves), compared
2754 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2755 * it is less than ~1Mbps. No point in rebalancing. */
2757 } else if (from->n_hashes == 1) {
2758 /* 'from' only carries a single MAC hash, so we can't shift any
2759 * load away from it, even though we want to. */
2762 /* 'from' is carrying significantly more load than 'to', and that
2763 * load is split across at least two different hashes. Pick a hash
2764 * to migrate to 'to' (the least-loaded slave), given that doing so
2765 * must decrease the ratio of the load on the two slaves by at
2768 * The sort order we use means that we prefer to shift away the
2769 * smallest hashes instead of the biggest ones. There is little
2770 * reason behind this decision; we could use the opposite sort
2771 * order to shift away big hashes ahead of small ones. */
2775 for (i = 0; i < from->n_hashes; i++) {
2776 double old_ratio, new_ratio;
2777 uint64_t delta = from->hashes[i]->tx_bytes;
2779 if (delta == 0 || from->tx_bytes - delta == 0) {
2780 /* Pointless move. */
2784 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2786 if (to->tx_bytes == 0) {
2787 /* Nothing on the new slave, move it. */
2791 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2792 new_ratio = (double)(from->tx_bytes - delta) /
2793 (to->tx_bytes + delta);
2795 if (new_ratio == 0) {
2796 /* Should already be covered but check to prevent division
2801 if (new_ratio < 1) {
2802 new_ratio = 1 / new_ratio;
2805 if (old_ratio - new_ratio > 0.1) {
2806 /* Would decrease the ratio, move it. */
2810 if (i < from->n_hashes) {
2811 bond_shift_load(from, to, i);
2812 port->bond_compat_is_stale = true;
2814 /* If the result of the migration changed the relative order of
2815 * 'from' and 'to' swap them back to maintain invariants. */
2816 if (order_swapped) {
2817 swap_bals(from, to);
2820 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2821 * point to different slave_balance structures. It is only
2822 * valid to do these two operations in a row at all because we
2823 * know that 'from' will not move past 'to' and vice versa. */
2824 resort_bals(from, bals, n_bals);
2825 resort_bals(to, bals, n_bals);
2832 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2833 * historical data to decay to <1% in 7 rebalancing runs. */
2834 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2840 bond_send_learning_packets(struct port *port)
2842 struct bridge *br = port->bridge;
2843 struct mac_entry *e;
2844 struct ofpbuf packet;
2845 int error, n_packets, n_errors;
2847 if (!port->n_ifaces || port->active_iface < 0) {
2851 ofpbuf_init(&packet, 128);
2852 error = n_packets = n_errors = 0;
2853 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2854 union ofp_action actions[2], *a;
2860 if (e->port == port->port_idx
2861 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2865 /* Compose actions. */
2866 memset(actions, 0, sizeof actions);
2869 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2870 a->vlan_vid.len = htons(sizeof *a);
2871 a->vlan_vid.vlan_vid = htons(e->vlan);
2874 a->output.type = htons(OFPAT_OUTPUT);
2875 a->output.len = htons(sizeof *a);
2876 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2881 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2883 flow_extract(&packet, 0, ODPP_NONE, &flow);
2884 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2891 ofpbuf_uninit(&packet);
2894 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2895 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2896 "packets, last error was: %s",
2897 port->name, n_errors, n_packets, strerror(error));
2899 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2900 port->name, n_packets);
2904 /* Bonding unixctl user interface functions. */
2907 bond_unixctl_list(struct unixctl_conn *conn,
2908 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2910 struct ds ds = DS_EMPTY_INITIALIZER;
2911 const struct bridge *br;
2913 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2915 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2918 for (i = 0; i < br->n_ports; i++) {
2919 const struct port *port = br->ports[i];
2920 if (port->n_ifaces > 1) {
2923 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2924 for (j = 0; j < port->n_ifaces; j++) {
2925 const struct iface *iface = port->ifaces[j];
2927 ds_put_cstr(&ds, ", ");
2929 ds_put_cstr(&ds, iface->name);
2931 ds_put_char(&ds, '\n');
2935 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2939 static struct port *
2940 bond_find(const char *name)
2942 const struct bridge *br;
2944 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2947 for (i = 0; i < br->n_ports; i++) {
2948 struct port *port = br->ports[i];
2949 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2958 bond_unixctl_show(struct unixctl_conn *conn,
2959 const char *args, void *aux OVS_UNUSED)
2961 struct ds ds = DS_EMPTY_INITIALIZER;
2962 const struct port *port;
2965 port = bond_find(args);
2967 unixctl_command_reply(conn, 501, "no such bond");
2971 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2972 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2973 ds_put_format(&ds, "next rebalance: %lld ms\n",
2974 port->bond_next_rebalance - time_msec());
2975 for (j = 0; j < port->n_ifaces; j++) {
2976 const struct iface *iface = port->ifaces[j];
2977 struct bond_entry *be;
2980 ds_put_format(&ds, "slave %s: %s\n",
2981 iface->name, iface->enabled ? "enabled" : "disabled");
2982 if (j == port->active_iface) {
2983 ds_put_cstr(&ds, "\tactive slave\n");
2985 if (iface->delay_expires != LLONG_MAX) {
2986 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2987 iface->enabled ? "downdelay" : "updelay",
2988 iface->delay_expires - time_msec());
2992 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2993 int hash = be - port->bond_hash;
2994 struct mac_entry *me;
2996 if (be->iface_idx != j) {
3000 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3001 hash, be->tx_bytes / 1024);
3004 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3005 &port->bridge->ml->lrus) {
3008 if (bond_hash(me->mac) == hash
3009 && me->port != port->port_idx
3010 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3011 && dp_ifidx == iface->dp_ifidx)
3013 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3014 ETH_ADDR_ARGS(me->mac));
3019 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3024 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3025 void *aux OVS_UNUSED)
3027 char *args = (char *) args_;
3028 char *save_ptr = NULL;
3029 char *bond_s, *hash_s, *slave_s;
3030 uint8_t mac[ETH_ADDR_LEN];
3032 struct iface *iface;
3033 struct bond_entry *entry;
3036 bond_s = strtok_r(args, " ", &save_ptr);
3037 hash_s = strtok_r(NULL, " ", &save_ptr);
3038 slave_s = strtok_r(NULL, " ", &save_ptr);
3040 unixctl_command_reply(conn, 501,
3041 "usage: bond/migrate BOND HASH SLAVE");
3045 port = bond_find(bond_s);
3047 unixctl_command_reply(conn, 501, "no such bond");
3051 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3052 == ETH_ADDR_SCAN_COUNT) {
3053 hash = bond_hash(mac);
3054 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3055 hash = atoi(hash_s) & BOND_MASK;
3057 unixctl_command_reply(conn, 501, "bad hash");
3061 iface = port_lookup_iface(port, slave_s);
3063 unixctl_command_reply(conn, 501, "no such slave");
3067 if (!iface->enabled) {
3068 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3072 entry = &port->bond_hash[hash];
3073 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3074 entry->iface_idx = iface->port_ifidx;
3075 entry->iface_tag = tag_create_random();
3076 port->bond_compat_is_stale = true;
3077 unixctl_command_reply(conn, 200, "migrated");
3081 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3082 void *aux OVS_UNUSED)
3084 char *args = (char *) args_;
3085 char *save_ptr = NULL;
3086 char *bond_s, *slave_s;
3088 struct iface *iface;
3090 bond_s = strtok_r(args, " ", &save_ptr);
3091 slave_s = strtok_r(NULL, " ", &save_ptr);
3093 unixctl_command_reply(conn, 501,
3094 "usage: bond/set-active-slave BOND SLAVE");
3098 port = bond_find(bond_s);
3100 unixctl_command_reply(conn, 501, "no such bond");
3104 iface = port_lookup_iface(port, slave_s);
3106 unixctl_command_reply(conn, 501, "no such slave");
3110 if (!iface->enabled) {
3111 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3115 if (port->active_iface != iface->port_ifidx) {
3116 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3117 port->active_iface = iface->port_ifidx;
3118 port->active_iface_tag = tag_create_random();
3119 VLOG_INFO("port %s: active interface is now %s",
3120 port->name, iface->name);
3121 bond_send_learning_packets(port);
3122 unixctl_command_reply(conn, 200, "done");
3124 unixctl_command_reply(conn, 200, "no change");
3129 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3131 char *args = (char *) args_;
3132 char *save_ptr = NULL;
3133 char *bond_s, *slave_s;
3135 struct iface *iface;
3137 bond_s = strtok_r(args, " ", &save_ptr);
3138 slave_s = strtok_r(NULL, " ", &save_ptr);
3140 unixctl_command_reply(conn, 501,
3141 "usage: bond/enable/disable-slave BOND SLAVE");
3145 port = bond_find(bond_s);
3147 unixctl_command_reply(conn, 501, "no such bond");
3151 iface = port_lookup_iface(port, slave_s);
3153 unixctl_command_reply(conn, 501, "no such slave");
3157 bond_enable_slave(iface, enable);
3158 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3162 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3163 void *aux OVS_UNUSED)
3165 enable_slave(conn, args, true);
3169 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3170 void *aux OVS_UNUSED)
3172 enable_slave(conn, args, false);
3176 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3177 void *aux OVS_UNUSED)
3179 uint8_t mac[ETH_ADDR_LEN];
3183 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3184 == ETH_ADDR_SCAN_COUNT) {
3185 hash = bond_hash(mac);
3187 hash_cstr = xasprintf("%u", hash);
3188 unixctl_command_reply(conn, 200, hash_cstr);
3191 unixctl_command_reply(conn, 501, "invalid mac");
3198 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3199 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3200 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3201 unixctl_command_register("bond/set-active-slave",
3202 bond_unixctl_set_active_slave, NULL);
3203 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3205 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3207 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3210 /* Port functions. */
3212 static struct port *
3213 port_create(struct bridge *br, const char *name)
3217 port = xzalloc(sizeof *port);
3219 port->port_idx = br->n_ports;
3221 port->trunks = NULL;
3222 port->name = xstrdup(name);
3223 port->active_iface = -1;
3225 if (br->n_ports >= br->allocated_ports) {
3226 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3229 br->ports[br->n_ports++] = port;
3230 shash_add_assert(&br->port_by_name, port->name, port);
3232 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3239 get_port_other_config(const struct ovsrec_port *port, const char *key,
3240 const char *default_value)
3242 const char *value = get_ovsrec_key_value(key,
3243 port->key_other_config,
3244 port->value_other_config,
3245 port->n_other_config);
3246 return value ? value : default_value;
3250 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3252 struct shash new_ifaces;
3255 /* Collect list of new interfaces. */
3256 shash_init(&new_ifaces);
3257 for (i = 0; i < cfg->n_interfaces; i++) {
3258 const char *name = cfg->interfaces[i]->name;
3259 shash_add_once(&new_ifaces, name, NULL);
3262 /* Get rid of deleted interfaces. */
3263 for (i = 0; i < port->n_ifaces; ) {
3264 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3265 iface_destroy(port->ifaces[i]);
3271 shash_destroy(&new_ifaces);
3275 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3277 struct shash new_ifaces;
3278 long long int next_rebalance;
3279 unsigned long *trunks;
3285 /* Update settings. */
3286 port->updelay = cfg->bond_updelay;
3287 if (port->updelay < 0) {
3290 port->updelay = cfg->bond_downdelay;
3291 if (port->downdelay < 0) {
3292 port->downdelay = 0;
3294 port->bond_rebalance_interval = atoi(
3295 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3296 if (port->bond_rebalance_interval < 1000) {
3297 port->bond_rebalance_interval = 1000;
3299 next_rebalance = time_msec() + port->bond_rebalance_interval;
3300 if (port->bond_next_rebalance > next_rebalance) {
3301 port->bond_next_rebalance = next_rebalance;
3304 /* Add new interfaces and update 'cfg' member of existing ones. */
3305 shash_init(&new_ifaces);
3306 for (i = 0; i < cfg->n_interfaces; i++) {
3307 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3308 struct iface *iface;
3310 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3311 VLOG_WARN("port %s: %s specified twice as port interface",
3312 port->name, if_cfg->name);
3316 iface = iface_lookup(port->bridge, if_cfg->name);
3318 if (iface->port != port) {
3319 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3321 port->bridge->name, if_cfg->name, iface->port->name);
3324 iface->cfg = if_cfg;
3326 iface_create(port, if_cfg);
3329 shash_destroy(&new_ifaces);
3334 if (port->n_ifaces < 2) {
3336 if (vlan >= 0 && vlan <= 4095) {
3337 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3342 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3343 * they even work as-is. But they have not been tested. */
3344 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3348 if (port->vlan != vlan) {
3350 bridge_flush(port->bridge);
3353 /* Get trunked VLANs. */
3355 if (vlan < 0 && cfg->n_trunks) {
3359 trunks = bitmap_allocate(4096);
3361 for (i = 0; i < cfg->n_trunks; i++) {
3362 int trunk = cfg->trunks[i];
3364 bitmap_set1(trunks, trunk);
3370 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3371 port->name, cfg->n_trunks);
3373 if (n_errors == cfg->n_trunks) {
3374 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3376 bitmap_free(trunks);
3379 } else if (vlan >= 0 && cfg->n_trunks) {
3380 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3384 ? port->trunks != NULL
3385 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3386 bridge_flush(port->bridge);
3388 bitmap_free(port->trunks);
3389 port->trunks = trunks;
3393 port_destroy(struct port *port)
3396 struct bridge *br = port->bridge;
3400 proc_net_compat_update_vlan(port->name, NULL, 0);
3401 proc_net_compat_update_bond(port->name, NULL);
3403 for (i = 0; i < MAX_MIRRORS; i++) {
3404 struct mirror *m = br->mirrors[i];
3405 if (m && m->out_port == port) {
3410 while (port->n_ifaces > 0) {
3411 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3414 shash_find_and_delete_assert(&br->port_by_name, port->name);
3416 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3417 del->port_idx = port->port_idx;
3420 bitmap_free(port->trunks);
3427 static struct port *
3428 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3430 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3431 return iface ? iface->port : NULL;
3434 static struct port *
3435 port_lookup(const struct bridge *br, const char *name)
3437 return shash_find_data(&br->port_by_name, name);
3440 static struct iface *
3441 port_lookup_iface(const struct port *port, const char *name)
3443 struct iface *iface = iface_lookup(port->bridge, name);
3444 return iface && iface->port == port ? iface : NULL;
3448 port_update_bonding(struct port *port)
3450 if (port->n_ifaces < 2) {
3451 /* Not a bonded port. */
3452 if (port->bond_hash) {
3453 free(port->bond_hash);
3454 port->bond_hash = NULL;
3455 port->bond_compat_is_stale = true;
3456 port->bond_fake_iface = false;
3459 if (!port->bond_hash) {
3462 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3463 for (i = 0; i <= BOND_MASK; i++) {
3464 struct bond_entry *e = &port->bond_hash[i];
3468 port->no_ifaces_tag = tag_create_random();
3469 bond_choose_active_iface(port);
3470 port->bond_next_rebalance
3471 = time_msec() + port->bond_rebalance_interval;
3473 if (port->cfg->bond_fake_iface) {
3474 port->bond_next_fake_iface_update = time_msec();
3477 port->bond_compat_is_stale = true;
3478 port->bond_fake_iface = port->cfg->bond_fake_iface;
3483 port_update_bond_compat(struct port *port)
3485 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3486 struct compat_bond bond;
3489 if (port->n_ifaces < 2) {
3490 proc_net_compat_update_bond(port->name, NULL);
3495 bond.updelay = port->updelay;
3496 bond.downdelay = port->downdelay;
3499 bond.hashes = compat_hashes;
3500 if (port->bond_hash) {
3501 const struct bond_entry *e;
3502 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3503 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3504 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3505 cbh->hash = e - port->bond_hash;
3506 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3511 bond.n_slaves = port->n_ifaces;
3512 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3513 for (i = 0; i < port->n_ifaces; i++) {
3514 struct iface *iface = port->ifaces[i];
3515 struct compat_bond_slave *slave = &bond.slaves[i];
3516 slave->name = iface->name;
3518 /* We need to make the same determination as the Linux bonding
3519 * code to determine whether a slave should be consider "up".
3520 * The Linux function bond_miimon_inspect() supports four
3521 * BOND_LINK_* states:
3523 * - BOND_LINK_UP: carrier detected, updelay has passed.
3524 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3525 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3526 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3528 * The function bond_info_show_slave() only considers BOND_LINK_UP
3529 * to be "up" and anything else to be "down".
3531 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3535 netdev_get_etheraddr(iface->netdev, slave->mac);
3538 if (port->bond_fake_iface) {
3539 struct netdev *bond_netdev;
3541 if (!netdev_open_default(port->name, &bond_netdev)) {
3543 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3545 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3547 netdev_close(bond_netdev);
3551 proc_net_compat_update_bond(port->name, &bond);
3556 port_update_vlan_compat(struct port *port)
3558 struct bridge *br = port->bridge;
3559 char *vlandev_name = NULL;
3561 if (port->vlan > 0) {
3562 /* Figure out the name that the VLAN device should actually have, if it
3563 * existed. This takes some work because the VLAN device would not
3564 * have port->name in its name; rather, it would have the trunk port's
3565 * name, and 'port' would be attached to a bridge that also had the
3566 * VLAN device one of its ports. So we need to find a trunk port that
3567 * includes port->vlan.
3569 * There might be more than one candidate. This doesn't happen on
3570 * XenServer, so if it happens we just pick the first choice in
3571 * alphabetical order instead of creating multiple VLAN devices. */
3573 for (i = 0; i < br->n_ports; i++) {
3574 struct port *p = br->ports[i];
3575 if (port_trunks_vlan(p, port->vlan)
3577 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3579 uint8_t ea[ETH_ADDR_LEN];
3580 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3581 if (!eth_addr_is_multicast(ea) &&
3582 !eth_addr_is_reserved(ea) &&
3583 !eth_addr_is_zero(ea)) {
3584 vlandev_name = p->name;
3589 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3592 /* Interface functions. */
3594 static struct iface *
3595 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3597 struct bridge *br = port->bridge;
3598 struct iface *iface;
3599 char *name = if_cfg->name;
3602 iface = xzalloc(sizeof *iface);
3604 iface->port_ifidx = port->n_ifaces;
3605 iface->name = xstrdup(name);
3606 iface->dp_ifidx = -1;
3607 iface->tag = tag_create_random();
3608 iface->delay_expires = LLONG_MAX;
3609 iface->netdev = NULL;
3610 iface->cfg = if_cfg;
3612 shash_add_assert(&br->iface_by_name, iface->name, iface);
3614 /* Attempt to create the network interface in case it doesn't exist yet. */
3615 if (!iface_is_internal(br, iface->name)) {
3616 error = set_up_iface(if_cfg, iface, true);
3618 VLOG_WARN("could not create iface %s: %s", iface->name,
3621 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3628 if (port->n_ifaces >= port->allocated_ifaces) {
3629 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3630 sizeof *port->ifaces);
3632 port->ifaces[port->n_ifaces++] = iface;
3633 if (port->n_ifaces > 1) {
3634 br->has_bonded_ports = true;
3637 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3645 iface_destroy(struct iface *iface)
3648 struct port *port = iface->port;
3649 struct bridge *br = port->bridge;
3650 bool del_active = port->active_iface == iface->port_ifidx;
3653 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3655 if (iface->dp_ifidx >= 0) {
3656 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3659 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3660 del->port_ifidx = iface->port_ifidx;
3662 netdev_close(iface->netdev);
3665 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3666 bond_choose_active_iface(port);
3667 bond_send_learning_packets(port);
3673 bridge_flush(port->bridge);
3677 static struct iface *
3678 iface_lookup(const struct bridge *br, const char *name)
3680 return shash_find_data(&br->iface_by_name, name);
3683 static struct iface *
3684 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3686 return port_array_get(&br->ifaces, dp_ifidx);
3689 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3690 * 'br', that is, an interface that is entirely simulated within the datapath.
3691 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3692 * interfaces are created by setting "iface.<iface>.internal = true".
3694 * In addition, we have a kluge-y feature that creates an internal port with
3695 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3696 * This feature needs to go away in the long term. Until then, this is one
3697 * reason why this function takes a name instead of a struct iface: the fake
3698 * interfaces created this way do not have a struct iface. */
3700 iface_is_internal(const struct bridge *br, const char *if_name)
3702 struct iface *iface;
3705 if (!strcmp(if_name, br->name)) {
3709 iface = iface_lookup(br, if_name);
3710 if (iface && !strcmp(iface->cfg->type, "internal")) {
3714 port = port_lookup(br, if_name);
3715 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3721 /* Set Ethernet address of 'iface', if one is specified in the configuration
3724 iface_set_mac(struct iface *iface)
3726 uint8_t ea[ETH_ADDR_LEN];
3728 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3729 if (eth_addr_is_multicast(ea)) {
3730 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3732 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3733 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3734 iface->name, iface->name);
3736 int error = netdev_set_etheraddr(iface->netdev, ea);
3738 VLOG_ERR("interface %s: setting MAC failed (%s)",
3739 iface->name, strerror(error));
3746 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3747 struct shash *shash)
3752 for (i = 0; i < n; i++) {
3753 shash_add(shash, keys[i], values[i]);
3757 struct iface_delete_queues_cbdata {
3758 struct netdev *netdev;
3759 const int64_t *queue_ids;
3764 queue_ids_include(const int64_t *ids, size_t n, int64_t target)
3769 while (low < high) {
3770 size_t mid = low + (high - low) / 2;
3771 if (target > ids[mid]) {
3773 } else if (target < ids[mid]) {
3783 iface_delete_queues(unsigned int queue_id,
3784 const struct shash *details OVS_UNUSED, void *cbdata_)
3786 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3788 if (!queue_ids_include(cbdata->queue_ids, cbdata->n_queue_ids, queue_id)) {
3789 netdev_delete_queue(cbdata->netdev, queue_id);
3794 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3796 if (!qos || qos->type[0] == '\0') {
3797 netdev_set_qos(iface->netdev, NULL, NULL);
3799 struct iface_delete_queues_cbdata cbdata;
3800 struct shash details;
3803 /* Configure top-level Qos for 'iface'. */
3804 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3805 qos->n_other_config, &details);
3806 netdev_set_qos(iface->netdev, qos->type, &details);
3807 shash_destroy(&details);
3809 /* Deconfigure queues that were deleted. */
3810 cbdata.netdev = iface->netdev;
3811 cbdata.queue_ids = qos->key_queues;
3812 cbdata.n_queue_ids = qos->n_queues;
3813 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3815 /* Configure queues for 'iface'. */
3816 for (i = 0; i < qos->n_queues; i++) {
3817 const struct ovsrec_queue *queue = qos->value_queues[i];
3818 unsigned int queue_id = qos->key_queues[i];
3820 shash_from_ovs_idl_map(queue->key_other_config,
3821 queue->value_other_config,
3822 queue->n_other_config, &details);
3823 netdev_set_queue(iface->netdev, queue_id, &details);
3824 shash_destroy(&details);
3829 /* Port mirroring. */
3832 mirror_reconfigure(struct bridge *br)
3834 struct shash old_mirrors, new_mirrors;
3835 struct shash_node *node;
3836 unsigned long *rspan_vlans;
3839 /* Collect old mirrors. */
3840 shash_init(&old_mirrors);
3841 for (i = 0; i < MAX_MIRRORS; i++) {
3842 if (br->mirrors[i]) {
3843 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3847 /* Collect new mirrors. */
3848 shash_init(&new_mirrors);
3849 for (i = 0; i < br->cfg->n_mirrors; i++) {
3850 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3851 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3852 VLOG_WARN("bridge %s: %s specified twice as mirror",
3853 br->name, cfg->name);
3857 /* Get rid of deleted mirrors and add new mirrors. */
3858 SHASH_FOR_EACH (node, &old_mirrors) {
3859 if (!shash_find(&new_mirrors, node->name)) {
3860 mirror_destroy(node->data);
3863 SHASH_FOR_EACH (node, &new_mirrors) {
3864 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3866 mirror = mirror_create(br, node->name);
3871 mirror_reconfigure_one(mirror, node->data);
3873 shash_destroy(&old_mirrors);
3874 shash_destroy(&new_mirrors);
3876 /* Update port reserved status. */
3877 for (i = 0; i < br->n_ports; i++) {
3878 br->ports[i]->is_mirror_output_port = false;
3880 for (i = 0; i < MAX_MIRRORS; i++) {
3881 struct mirror *m = br->mirrors[i];
3882 if (m && m->out_port) {
3883 m->out_port->is_mirror_output_port = true;
3887 /* Update flooded vlans (for RSPAN). */
3889 if (br->cfg->n_flood_vlans) {
3890 rspan_vlans = bitmap_allocate(4096);
3892 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3893 int64_t vlan = br->cfg->flood_vlans[i];
3894 if (vlan >= 0 && vlan < 4096) {
3895 bitmap_set1(rspan_vlans, vlan);
3896 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3899 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3904 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3909 static struct mirror *
3910 mirror_create(struct bridge *br, const char *name)
3915 for (i = 0; ; i++) {
3916 if (i >= MAX_MIRRORS) {
3917 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3918 "cannot create %s", br->name, MAX_MIRRORS, name);
3921 if (!br->mirrors[i]) {
3926 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3929 br->mirrors[i] = m = xzalloc(sizeof *m);
3932 m->name = xstrdup(name);
3933 shash_init(&m->src_ports);
3934 shash_init(&m->dst_ports);
3944 mirror_destroy(struct mirror *m)
3947 struct bridge *br = m->bridge;
3950 for (i = 0; i < br->n_ports; i++) {
3951 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3952 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3955 shash_destroy(&m->src_ports);
3956 shash_destroy(&m->dst_ports);
3959 m->bridge->mirrors[m->idx] = NULL;
3967 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3968 struct shash *names)
3972 for (i = 0; i < n_ports; i++) {
3973 const char *name = ports[i]->name;
3974 if (port_lookup(m->bridge, name)) {
3975 shash_add_once(names, name, NULL);
3977 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3978 "port %s", m->bridge->name, m->name, name);
3984 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3990 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3992 for (i = 0; i < cfg->n_select_vlan; i++) {
3993 int64_t vlan = cfg->select_vlan[i];
3994 if (vlan < 0 || vlan > 4095) {
3995 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3996 m->bridge->name, m->name, vlan);
3998 (*vlans)[n_vlans++] = vlan;
4005 vlan_is_mirrored(const struct mirror *m, int vlan)
4009 for (i = 0; i < m->n_vlans; i++) {
4010 if (m->vlans[i] == vlan) {
4018 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4022 for (i = 0; i < m->n_vlans; i++) {
4023 if (port_trunks_vlan(p, m->vlans[i])) {
4031 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4033 struct shash src_ports, dst_ports;
4034 mirror_mask_t mirror_bit;
4035 struct port *out_port;
4041 /* Get output port. */
4042 if (cfg->output_port) {
4043 out_port = port_lookup(m->bridge, cfg->output_port->name);
4045 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4046 m->bridge->name, m->name);
4052 if (cfg->output_vlan) {
4053 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4054 "output vlan; ignoring output vlan",
4055 m->bridge->name, m->name);
4057 } else if (cfg->output_vlan) {
4059 out_vlan = *cfg->output_vlan;
4061 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4062 m->bridge->name, m->name);
4067 shash_init(&src_ports);
4068 shash_init(&dst_ports);
4069 if (cfg->select_all) {
4070 for (i = 0; i < m->bridge->n_ports; i++) {
4071 const char *name = m->bridge->ports[i]->name;
4072 shash_add_once(&src_ports, name, NULL);
4073 shash_add_once(&dst_ports, name, NULL);
4078 /* Get ports, and drop duplicates and ports that don't exist. */
4079 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4081 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4084 /* Get all the vlans, and drop duplicate and invalid vlans. */
4085 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4088 /* Update mirror data. */
4089 if (!shash_equal_keys(&m->src_ports, &src_ports)
4090 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4091 || m->n_vlans != n_vlans
4092 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4093 || m->out_port != out_port
4094 || m->out_vlan != out_vlan) {
4095 bridge_flush(m->bridge);
4097 shash_swap(&m->src_ports, &src_ports);
4098 shash_swap(&m->dst_ports, &dst_ports);
4101 m->n_vlans = n_vlans;
4102 m->out_port = out_port;
4103 m->out_vlan = out_vlan;
4106 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4107 for (i = 0; i < m->bridge->n_ports; i++) {
4108 struct port *port = m->bridge->ports[i];
4110 if (shash_find(&m->src_ports, port->name)
4113 ? port_trunks_any_mirrored_vlan(m, port)
4114 : vlan_is_mirrored(m, port->vlan)))) {
4115 port->src_mirrors |= mirror_bit;
4117 port->src_mirrors &= ~mirror_bit;
4120 if (shash_find(&m->dst_ports, port->name)) {
4121 port->dst_mirrors |= mirror_bit;
4123 port->dst_mirrors &= ~mirror_bit;
4128 shash_destroy(&src_ports);
4129 shash_destroy(&dst_ports);