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 shash_add_once(&cur_ifaces, name, NULL);
652 /* Get the set of interfaces we want on this datapath. */
653 bridge_get_all_ifaces(br, &want_ifaces);
655 SHASH_FOR_EACH (node, &want_ifaces) {
656 const char *if_name = node->name;
657 struct iface *iface = node->data;
659 if (shash_find(&cur_ifaces, if_name)) {
660 /* Already exists, just reconfigure it. */
662 reconfigure_iface(iface->cfg, iface);
665 /* Need to add to datapath. */
669 /* Add to datapath. */
670 internal = iface_is_internal(br, if_name);
671 error = dpif_port_add(br->dpif, if_name,
672 internal ? ODP_PORT_INTERNAL : 0, NULL);
673 if (error == EFBIG) {
674 VLOG_ERR("ran out of valid port numbers on %s",
675 dpif_name(br->dpif));
678 VLOG_ERR("failed to add %s interface to %s: %s",
679 if_name, dpif_name(br->dpif), strerror(error));
683 shash_destroy(&cur_ifaces);
684 shash_destroy(&want_ifaces);
686 sflow_bridge_number = 0;
687 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
690 struct iface *local_iface;
691 struct iface *hw_addr_iface;
694 bridge_fetch_dp_ifaces(br);
696 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
697 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
699 /* Pick local port hardware address, datapath ID. */
700 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
701 local_iface = bridge_get_local_iface(br);
703 int error = netdev_set_etheraddr(local_iface->netdev, ea);
705 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
706 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
707 "Ethernet address: %s",
708 br->name, strerror(error));
712 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
713 ofproto_set_datapath_id(br->ofproto, dpid);
715 dpid_string = xasprintf("%016"PRIx64, dpid);
716 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
719 /* Set NetFlow configuration on this bridge. */
720 if (br->cfg->netflow) {
721 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
722 struct netflow_options opts;
724 memset(&opts, 0, sizeof opts);
726 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
727 if (nf_cfg->engine_type) {
728 opts.engine_type = *nf_cfg->engine_type;
730 if (nf_cfg->engine_id) {
731 opts.engine_id = *nf_cfg->engine_id;
734 opts.active_timeout = nf_cfg->active_timeout;
735 if (!opts.active_timeout) {
736 opts.active_timeout = -1;
737 } else if (opts.active_timeout < 0) {
738 VLOG_WARN("bridge %s: active timeout interval set to negative "
739 "value, using default instead (%d seconds)", br->name,
740 NF_ACTIVE_TIMEOUT_DEFAULT);
741 opts.active_timeout = -1;
744 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
745 if (opts.add_id_to_iface) {
746 if (opts.engine_id > 0x7f) {
747 VLOG_WARN("bridge %s: netflow port mangling may conflict "
748 "with another vswitch, choose an engine id less "
749 "than 128", br->name);
751 if (br->n_ports > 508) {
752 VLOG_WARN("bridge %s: netflow port mangling will conflict "
753 "with another port when more than 508 ports are "
758 opts.collectors.n = nf_cfg->n_targets;
759 opts.collectors.names = nf_cfg->targets;
760 if (ofproto_set_netflow(br->ofproto, &opts)) {
761 VLOG_ERR("bridge %s: problem setting netflow collectors",
765 ofproto_set_netflow(br->ofproto, NULL);
768 /* Set sFlow configuration on this bridge. */
769 if (br->cfg->sflow) {
770 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
771 struct ovsrec_controller **controllers;
772 struct ofproto_sflow_options oso;
773 size_t n_controllers;
776 memset(&oso, 0, sizeof oso);
778 oso.targets.n = sflow_cfg->n_targets;
779 oso.targets.names = sflow_cfg->targets;
781 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
782 if (sflow_cfg->sampling) {
783 oso.sampling_rate = *sflow_cfg->sampling;
786 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
787 if (sflow_cfg->polling) {
788 oso.polling_interval = *sflow_cfg->polling;
791 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
792 if (sflow_cfg->header) {
793 oso.header_len = *sflow_cfg->header;
796 oso.sub_id = sflow_bridge_number++;
797 oso.agent_device = sflow_cfg->agent;
799 oso.control_ip = NULL;
800 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
801 for (i = 0; i < n_controllers; i++) {
802 if (controllers[i]->local_ip) {
803 oso.control_ip = controllers[i]->local_ip;
807 ofproto_set_sflow(br->ofproto, &oso);
809 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
811 ofproto_set_sflow(br->ofproto, NULL);
814 /* Update the controller and related settings. It would be more
815 * straightforward to call this from bridge_reconfigure_one(), but we
816 * can't do it there for two reasons. First, and most importantly, at
817 * that point we don't know the dp_ifidx of any interfaces that have
818 * been added to the bridge (because we haven't actually added them to
819 * the datapath). Second, at that point we haven't set the datapath ID
820 * yet; when a controller is configured, resetting the datapath ID will
821 * immediately disconnect from the controller, so it's better to set
822 * the datapath ID before the controller. */
823 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
825 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
826 for (i = 0; i < br->n_ports; i++) {
827 struct port *port = br->ports[i];
830 port_update_vlan_compat(port);
831 port_update_bonding(port);
833 for (j = 0; j < port->n_ifaces; j++) {
834 iface_update_qos(port->ifaces[j], port->cfg->qos);
838 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
839 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
846 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
847 const struct ovsdb_idl_column *column,
850 const struct ovsdb_datum *datum;
851 union ovsdb_atom atom;
854 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
855 atom.string = (char *) key;
856 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
857 return idx == UINT_MAX ? NULL : datum->values[idx].string;
861 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
863 return get_ovsrec_key_value(&br_cfg->header_,
864 &ovsrec_bridge_col_other_config, key);
868 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
869 struct iface **hw_addr_iface)
875 *hw_addr_iface = NULL;
877 /* Did the user request a particular MAC? */
878 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
879 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
880 if (eth_addr_is_multicast(ea)) {
881 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
882 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
883 } else if (eth_addr_is_zero(ea)) {
884 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
890 /* Otherwise choose the minimum non-local MAC address among all of the
892 memset(ea, 0xff, sizeof ea);
893 for (i = 0; i < br->n_ports; i++) {
894 struct port *port = br->ports[i];
895 uint8_t iface_ea[ETH_ADDR_LEN];
898 /* Mirror output ports don't participate. */
899 if (port->is_mirror_output_port) {
903 /* Choose the MAC address to represent the port. */
904 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
905 /* Find the interface with this Ethernet address (if any) so that
906 * we can provide the correct devname to the caller. */
908 for (j = 0; j < port->n_ifaces; j++) {
909 struct iface *candidate = port->ifaces[j];
910 uint8_t candidate_ea[ETH_ADDR_LEN];
911 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
912 && eth_addr_equals(iface_ea, candidate_ea)) {
917 /* Choose the interface whose MAC address will represent the port.
918 * The Linux kernel bonding code always chooses the MAC address of
919 * the first slave added to a bond, and the Fedora networking
920 * scripts always add slaves to a bond in alphabetical order, so
921 * for compatibility we choose the interface with the name that is
922 * first in alphabetical order. */
923 iface = port->ifaces[0];
924 for (j = 1; j < port->n_ifaces; j++) {
925 struct iface *candidate = port->ifaces[j];
926 if (strcmp(candidate->name, iface->name) < 0) {
931 /* The local port doesn't count (since we're trying to choose its
932 * MAC address anyway). */
933 if (iface->dp_ifidx == ODPP_LOCAL) {
938 error = netdev_get_etheraddr(iface->netdev, iface_ea);
940 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
941 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
942 iface->name, strerror(error));
947 /* Compare against our current choice. */
948 if (!eth_addr_is_multicast(iface_ea) &&
949 !eth_addr_is_local(iface_ea) &&
950 !eth_addr_is_reserved(iface_ea) &&
951 !eth_addr_is_zero(iface_ea) &&
952 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
954 memcpy(ea, iface_ea, ETH_ADDR_LEN);
955 *hw_addr_iface = iface;
958 if (eth_addr_is_multicast(ea)) {
959 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
960 *hw_addr_iface = NULL;
961 VLOG_WARN("bridge %s: using default bridge Ethernet "
962 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
964 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
965 br->name, ETH_ADDR_ARGS(ea));
969 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
970 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
971 * an interface on 'br', then that interface must be passed in as
972 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
973 * 'hw_addr_iface' must be passed in as a null pointer. */
975 bridge_pick_datapath_id(struct bridge *br,
976 const uint8_t bridge_ea[ETH_ADDR_LEN],
977 struct iface *hw_addr_iface)
980 * The procedure for choosing a bridge MAC address will, in the most
981 * ordinary case, also choose a unique MAC that we can use as a datapath
982 * ID. In some special cases, though, multiple bridges will end up with
983 * the same MAC address. This is OK for the bridges, but it will confuse
984 * the OpenFlow controller, because each datapath needs a unique datapath
987 * Datapath IDs must be unique. It is also very desirable that they be
988 * stable from one run to the next, so that policy set on a datapath
991 const char *datapath_id;
994 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
995 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1001 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1003 * A bridge whose MAC address is taken from a VLAN network device
1004 * (that is, a network device created with vconfig(8) or similar
1005 * tool) will have the same MAC address as a bridge on the VLAN
1006 * device's physical network device.
1008 * Handle this case by hashing the physical network device MAC
1009 * along with the VLAN identifier.
1011 uint8_t buf[ETH_ADDR_LEN + 2];
1012 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1013 buf[ETH_ADDR_LEN] = vlan >> 8;
1014 buf[ETH_ADDR_LEN + 1] = vlan;
1015 return dpid_from_hash(buf, sizeof buf);
1018 * Assume that this bridge's MAC address is unique, since it
1019 * doesn't fit any of the cases we handle specially.
1024 * A purely internal bridge, that is, one that has no non-virtual
1025 * network devices on it at all, is more difficult because it has no
1026 * natural unique identifier at all.
1028 * When the host is a XenServer, we handle this case by hashing the
1029 * host's UUID with the name of the bridge. Names of bridges are
1030 * persistent across XenServer reboots, although they can be reused if
1031 * an internal network is destroyed and then a new one is later
1032 * created, so this is fairly effective.
1034 * When the host is not a XenServer, we punt by using a random MAC
1035 * address on each run.
1037 const char *host_uuid = xenserver_get_host_uuid();
1039 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1040 dpid = dpid_from_hash(combined, strlen(combined));
1046 return eth_addr_to_uint64(bridge_ea);
1050 dpid_from_hash(const void *data, size_t n)
1052 uint8_t hash[SHA1_DIGEST_SIZE];
1054 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1055 sha1_bytes(data, n, hash);
1056 eth_addr_mark_random(hash);
1057 return eth_addr_to_uint64(hash);
1061 iface_refresh_stats(struct iface *iface)
1067 static const struct iface_stat iface_stats[] = {
1068 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1069 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1070 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1071 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1072 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1073 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1074 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1075 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1076 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1077 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1078 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1079 { "collisions", offsetof(struct netdev_stats, collisions) },
1081 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1082 const struct iface_stat *s;
1084 char *keys[N_STATS];
1085 int64_t values[N_STATS];
1088 struct netdev_stats stats;
1090 /* Intentionally ignore return value, since errors will set 'stats' to
1091 * all-1s, and we will deal with that correctly below. */
1092 netdev_get_stats(iface->netdev, &stats);
1095 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1096 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1097 if (value != UINT64_MAX) {
1104 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1110 bool datapath_destroyed;
1113 /* Let each bridge do the work that it needs to do. */
1114 datapath_destroyed = false;
1115 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1116 int error = bridge_run_one(br);
1118 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1119 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1120 "forcing reconfiguration", br->name);
1121 datapath_destroyed = true;
1125 /* (Re)configure if necessary. */
1126 if (ovsdb_idl_run(idl) || datapath_destroyed) {
1127 const struct ovsrec_open_vswitch *cfg = ovsrec_open_vswitch_first(idl);
1129 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1131 bridge_configure_once(cfg);
1132 bridge_reconfigure(cfg);
1134 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1135 ovsdb_idl_txn_commit(txn);
1136 ovsdb_idl_txn_destroy(txn); /* XXX */
1138 /* We still need to reconfigure to avoid dangling pointers to
1139 * now-destroyed ovsrec structures inside bridge data. */
1140 static const struct ovsrec_open_vswitch null_cfg;
1142 bridge_reconfigure(&null_cfg);
1146 /* Refresh interface stats if necessary. */
1147 if (time_msec() >= iface_stats_timer) {
1148 struct ovsdb_idl_txn *txn;
1150 txn = ovsdb_idl_txn_create(idl);
1151 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1154 for (i = 0; i < br->n_ports; i++) {
1155 struct port *port = br->ports[i];
1158 for (j = 0; j < port->n_ifaces; j++) {
1159 struct iface *iface = port->ifaces[j];
1160 iface_refresh_stats(iface);
1164 ovsdb_idl_txn_commit(txn);
1165 ovsdb_idl_txn_destroy(txn); /* XXX */
1167 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
1176 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1177 ofproto_wait(br->ofproto);
1178 if (ofproto_has_controller(br->ofproto)) {
1182 mac_learning_wait(br->ml);
1185 ovsdb_idl_wait(idl);
1186 poll_timer_wait_until(iface_stats_timer);
1189 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1190 * configuration changes. */
1192 bridge_flush(struct bridge *br)
1194 COVERAGE_INC(bridge_flush);
1196 mac_learning_flush(br->ml);
1199 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1200 * such interface. */
1201 static struct iface *
1202 bridge_get_local_iface(struct bridge *br)
1206 for (i = 0; i < br->n_ports; i++) {
1207 struct port *port = br->ports[i];
1208 for (j = 0; j < port->n_ifaces; j++) {
1209 struct iface *iface = port->ifaces[j];
1210 if (iface->dp_ifidx == ODPP_LOCAL) {
1219 /* Bridge unixctl user interface functions. */
1221 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1222 const char *args, void *aux OVS_UNUSED)
1224 struct ds ds = DS_EMPTY_INITIALIZER;
1225 const struct bridge *br;
1226 const struct mac_entry *e;
1228 br = bridge_lookup(args);
1230 unixctl_command_reply(conn, 501, "no such bridge");
1234 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1235 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1236 if (e->port < 0 || e->port >= br->n_ports) {
1239 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1240 br->ports[e->port]->ifaces[0]->dp_ifidx,
1241 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1243 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1247 /* Bridge reconfiguration functions. */
1248 static struct bridge *
1249 bridge_create(const struct ovsrec_bridge *br_cfg)
1254 assert(!bridge_lookup(br_cfg->name));
1255 br = xzalloc(sizeof *br);
1257 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1263 dpif_flow_flush(br->dpif);
1265 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1268 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1270 dpif_delete(br->dpif);
1271 dpif_close(br->dpif);
1276 br->name = xstrdup(br_cfg->name);
1278 br->ml = mac_learning_create();
1279 eth_addr_nicira_random(br->default_ea);
1281 port_array_init(&br->ifaces);
1283 shash_init(&br->port_by_name);
1284 shash_init(&br->iface_by_name);
1288 list_push_back(&all_bridges, &br->node);
1290 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1296 bridge_destroy(struct bridge *br)
1301 while (br->n_ports > 0) {
1302 port_destroy(br->ports[br->n_ports - 1]);
1304 list_remove(&br->node);
1305 error = dpif_delete(br->dpif);
1306 if (error && error != ENOENT) {
1307 VLOG_ERR("failed to delete %s: %s",
1308 dpif_name(br->dpif), strerror(error));
1310 dpif_close(br->dpif);
1311 ofproto_destroy(br->ofproto);
1312 mac_learning_destroy(br->ml);
1313 port_array_destroy(&br->ifaces);
1314 shash_destroy(&br->port_by_name);
1315 shash_destroy(&br->iface_by_name);
1322 static struct bridge *
1323 bridge_lookup(const char *name)
1327 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1328 if (!strcmp(br->name, name)) {
1335 /* Handle requests for a listing of all flows known by the OpenFlow
1336 * stack, including those normally hidden. */
1338 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1339 const char *args, void *aux OVS_UNUSED)
1344 br = bridge_lookup(args);
1346 unixctl_command_reply(conn, 501, "Unknown bridge");
1351 ofproto_get_all_flows(br->ofproto, &results);
1353 unixctl_command_reply(conn, 200, ds_cstr(&results));
1354 ds_destroy(&results);
1357 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1358 * connections and reconnect. If BRIDGE is not specified, then all bridges
1359 * drop their controller connections and reconnect. */
1361 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1362 const char *args, void *aux OVS_UNUSED)
1365 if (args[0] != '\0') {
1366 br = bridge_lookup(args);
1368 unixctl_command_reply(conn, 501, "Unknown bridge");
1371 ofproto_reconnect_controllers(br->ofproto);
1373 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1374 ofproto_reconnect_controllers(br->ofproto);
1377 unixctl_command_reply(conn, 200, NULL);
1381 bridge_run_one(struct bridge *br)
1385 error = ofproto_run1(br->ofproto);
1390 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1393 error = ofproto_run2(br->ofproto, br->flush);
1400 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1401 const struct bridge *br,
1402 struct ovsrec_controller ***controllersp)
1404 struct ovsrec_controller **controllers;
1405 size_t n_controllers;
1407 if (br->cfg->n_controller) {
1408 controllers = br->cfg->controller;
1409 n_controllers = br->cfg->n_controller;
1411 controllers = ovs_cfg->controller;
1412 n_controllers = ovs_cfg->n_controller;
1415 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1421 *controllersp = controllers;
1423 return n_controllers;
1427 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1430 struct shash old_ports, new_ports;
1431 struct svec listeners, old_listeners;
1432 struct svec snoops, old_snoops;
1433 struct shash_node *node;
1436 /* Collect old ports. */
1437 shash_init(&old_ports);
1438 for (i = 0; i < br->n_ports; i++) {
1439 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1442 /* Collect new ports. */
1443 shash_init(&new_ports);
1444 for (i = 0; i < br->cfg->n_ports; i++) {
1445 const char *name = br->cfg->ports[i]->name;
1446 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1447 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1452 /* If we have a controller, then we need a local port. Complain if the
1453 * user didn't specify one.
1455 * XXX perhaps we should synthesize a port ourselves in this case. */
1456 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1457 char local_name[IF_NAMESIZE];
1460 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1461 local_name, sizeof local_name);
1462 if (!error && !shash_find(&new_ports, local_name)) {
1463 VLOG_WARN("bridge %s: controller specified but no local port "
1464 "(port named %s) defined",
1465 br->name, local_name);
1469 /* Get rid of deleted ports.
1470 * Get rid of deleted interfaces on ports that still exist. */
1471 SHASH_FOR_EACH (node, &old_ports) {
1472 struct port *port = node->data;
1473 const struct ovsrec_port *port_cfg;
1475 port_cfg = shash_find_data(&new_ports, node->name);
1479 port_del_ifaces(port, port_cfg);
1483 /* Create new ports.
1484 * Add new interfaces to existing ports.
1485 * Reconfigure existing ports. */
1486 SHASH_FOR_EACH (node, &new_ports) {
1487 struct port *port = shash_find_data(&old_ports, node->name);
1489 port = port_create(br, node->name);
1492 port_reconfigure(port, node->data);
1493 if (!port->n_ifaces) {
1494 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1495 br->name, port->name);
1499 shash_destroy(&old_ports);
1500 shash_destroy(&new_ports);
1502 /* Delete all flows if we're switching from connected to standalone or vice
1503 * versa. (XXX Should we delete all flows if we are switching from one
1504 * controller to another?) */
1506 /* Configure OpenFlow management listener. */
1507 svec_init(&listeners);
1508 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1509 ovs_rundir, br->name));
1510 svec_init(&old_listeners);
1511 ofproto_get_listeners(br->ofproto, &old_listeners);
1512 if (!svec_equal(&listeners, &old_listeners)) {
1513 ofproto_set_listeners(br->ofproto, &listeners);
1515 svec_destroy(&listeners);
1516 svec_destroy(&old_listeners);
1518 /* Configure OpenFlow controller connection snooping. */
1520 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1521 ovs_rundir, br->name));
1522 svec_init(&old_snoops);
1523 ofproto_get_snoops(br->ofproto, &old_snoops);
1524 if (!svec_equal(&snoops, &old_snoops)) {
1525 ofproto_set_snoops(br->ofproto, &snoops);
1527 svec_destroy(&snoops);
1528 svec_destroy(&old_snoops);
1530 mirror_reconfigure(br);
1534 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1536 const struct sockaddr_in *managers,
1539 struct ovsrec_controller **controllers;
1540 size_t n_controllers;
1542 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1544 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1545 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1546 ofproto_flush_flows(br->ofproto);
1549 if (!n_controllers) {
1550 union ofp_action action;
1553 /* Clear out controllers. */
1554 ofproto_set_controllers(br->ofproto, NULL, 0);
1556 /* Set up a flow that matches every packet and directs them to
1557 * OFPP_NORMAL (which goes to us). */
1558 memset(&action, 0, sizeof action);
1559 action.type = htons(OFPAT_OUTPUT);
1560 action.output.len = htons(sizeof action);
1561 action.output.port = htons(OFPP_NORMAL);
1562 memset(&flow, 0, sizeof flow);
1563 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1565 struct ofproto_controller *ocs;
1568 ocs = xmalloc(n_controllers * sizeof *ocs);
1569 for (i = 0; i < n_controllers; i++) {
1570 struct ovsrec_controller *c = controllers[i];
1571 struct ofproto_controller *oc = &ocs[i];
1573 if (strcmp(c->target, "discover")) {
1574 struct iface *local_iface;
1577 local_iface = bridge_get_local_iface(br);
1578 if (local_iface && c->local_ip
1579 && inet_aton(c->local_ip, &ip)) {
1580 struct netdev *netdev = local_iface->netdev;
1581 struct in_addr mask, gateway;
1583 if (!c->local_netmask
1584 || !inet_aton(c->local_netmask, &mask)) {
1587 if (!c->local_gateway
1588 || !inet_aton(c->local_gateway, &gateway)) {
1592 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1594 mask.s_addr = guess_netmask(ip.s_addr);
1596 if (!netdev_set_in4(netdev, ip, mask)) {
1597 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1599 br->name, IP_ARGS(&ip.s_addr),
1600 IP_ARGS(&mask.s_addr));
1603 if (gateway.s_addr) {
1604 if (!netdev_add_router(netdev, gateway)) {
1605 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1606 br->name, IP_ARGS(&gateway.s_addr));
1612 oc->target = c->target;
1613 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1614 oc->probe_interval = (c->inactivity_probe
1615 ? *c->inactivity_probe / 1000 : 5);
1616 oc->fail = (!c->fail_mode
1617 || !strcmp(c->fail_mode, "standalone")
1618 || !strcmp(c->fail_mode, "open")
1619 ? OFPROTO_FAIL_STANDALONE
1620 : OFPROTO_FAIL_SECURE);
1621 oc->band = (!c->connection_mode
1622 || !strcmp(c->connection_mode, "in-band")
1624 : OFPROTO_OUT_OF_BAND);
1625 oc->accept_re = c->discover_accept_regex;
1626 oc->update_resolv_conf = c->discover_update_resolv_conf;
1627 oc->rate_limit = (c->controller_rate_limit
1628 ? *c->controller_rate_limit : 0);
1629 oc->burst_limit = (c->controller_burst_limit
1630 ? *c->controller_burst_limit : 0);
1632 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1638 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1643 for (i = 0; i < br->n_ports; i++) {
1644 struct port *port = br->ports[i];
1645 for (j = 0; j < port->n_ifaces; j++) {
1646 struct iface *iface = port->ifaces[j];
1647 shash_add_once(ifaces, iface->name, iface);
1649 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1650 shash_add_once(ifaces, port->name, NULL);
1655 /* For robustness, in case the administrator moves around datapath ports behind
1656 * our back, we re-check all the datapath port numbers here.
1658 * This function will set the 'dp_ifidx' members of interfaces that have
1659 * disappeared to -1, so only call this function from a context where those
1660 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1661 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1662 * datapath, which doesn't support UINT16_MAX+1 ports. */
1664 bridge_fetch_dp_ifaces(struct bridge *br)
1666 struct odp_port *dpif_ports;
1667 size_t n_dpif_ports;
1670 /* Reset all interface numbers. */
1671 for (i = 0; i < br->n_ports; i++) {
1672 struct port *port = br->ports[i];
1673 for (j = 0; j < port->n_ifaces; j++) {
1674 struct iface *iface = port->ifaces[j];
1675 iface->dp_ifidx = -1;
1678 port_array_clear(&br->ifaces);
1680 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1681 for (i = 0; i < n_dpif_ports; i++) {
1682 struct odp_port *p = &dpif_ports[i];
1683 struct iface *iface = iface_lookup(br, p->devname);
1685 if (iface->dp_ifidx >= 0) {
1686 VLOG_WARN("%s reported interface %s twice",
1687 dpif_name(br->dpif), p->devname);
1688 } else if (iface_from_dp_ifidx(br, p->port)) {
1689 VLOG_WARN("%s reported interface %"PRIu16" twice",
1690 dpif_name(br->dpif), p->port);
1692 port_array_set(&br->ifaces, p->port, iface);
1693 iface->dp_ifidx = p->port;
1697 int64_t ofport = (iface->dp_ifidx >= 0
1698 ? odp_port_to_ofp_port(iface->dp_ifidx)
1700 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1707 /* Bridge packet processing functions. */
1710 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1712 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1715 static struct bond_entry *
1716 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1718 return &port->bond_hash[bond_hash(mac)];
1722 bond_choose_iface(const struct port *port)
1724 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1725 size_t i, best_down_slave = -1;
1726 long long next_delay_expiration = LLONG_MAX;
1728 for (i = 0; i < port->n_ifaces; i++) {
1729 struct iface *iface = port->ifaces[i];
1731 if (iface->enabled) {
1733 } else if (iface->delay_expires < next_delay_expiration) {
1734 best_down_slave = i;
1735 next_delay_expiration = iface->delay_expires;
1739 if (best_down_slave != -1) {
1740 struct iface *iface = port->ifaces[best_down_slave];
1742 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1743 "since no other interface is up", iface->name,
1744 iface->delay_expires - time_msec());
1745 bond_enable_slave(iface, true);
1748 return best_down_slave;
1752 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1753 uint16_t *dp_ifidx, tag_type *tags)
1755 struct iface *iface;
1757 assert(port->n_ifaces);
1758 if (port->n_ifaces == 1) {
1759 iface = port->ifaces[0];
1761 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1762 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1763 || !port->ifaces[e->iface_idx]->enabled) {
1764 /* XXX select interface properly. The current interface selection
1765 * is only good for testing the rebalancing code. */
1766 e->iface_idx = bond_choose_iface(port);
1767 if (e->iface_idx < 0) {
1768 *tags |= port->no_ifaces_tag;
1771 e->iface_tag = tag_create_random();
1772 ((struct port *) port)->bond_compat_is_stale = true;
1774 *tags |= e->iface_tag;
1775 iface = port->ifaces[e->iface_idx];
1777 *dp_ifidx = iface->dp_ifidx;
1778 *tags |= iface->tag; /* Currently only used for bonding. */
1783 bond_link_status_update(struct iface *iface, bool carrier)
1785 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1786 struct port *port = iface->port;
1788 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1789 /* Nothing to do. */
1792 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1793 iface->name, carrier ? "detected" : "dropped");
1794 if (carrier == iface->enabled) {
1795 iface->delay_expires = LLONG_MAX;
1796 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1797 iface->name, carrier ? "disabled" : "enabled");
1798 } else if (carrier && port->active_iface < 0) {
1799 bond_enable_slave(iface, true);
1800 if (port->updelay) {
1801 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1802 "other interface is up", iface->name, port->updelay);
1805 int delay = carrier ? port->updelay : port->downdelay;
1806 iface->delay_expires = time_msec() + delay;
1809 "interface %s: will be %s if it stays %s for %d ms",
1811 carrier ? "enabled" : "disabled",
1812 carrier ? "up" : "down",
1819 bond_choose_active_iface(struct port *port)
1821 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1823 port->active_iface = bond_choose_iface(port);
1824 port->active_iface_tag = tag_create_random();
1825 if (port->active_iface >= 0) {
1826 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1827 port->name, port->ifaces[port->active_iface]->name);
1829 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1835 bond_enable_slave(struct iface *iface, bool enable)
1837 struct port *port = iface->port;
1838 struct bridge *br = port->bridge;
1840 /* This acts as a recursion check. If the act of disabling a slave
1841 * causes a different slave to be enabled, the flag will allow us to
1842 * skip redundant work when we reenter this function. It must be
1843 * cleared on exit to keep things safe with multiple bonds. */
1844 static bool moving_active_iface = false;
1846 iface->delay_expires = LLONG_MAX;
1847 if (enable == iface->enabled) {
1851 iface->enabled = enable;
1852 if (!iface->enabled) {
1853 VLOG_WARN("interface %s: disabled", iface->name);
1854 ofproto_revalidate(br->ofproto, iface->tag);
1855 if (iface->port_ifidx == port->active_iface) {
1856 ofproto_revalidate(br->ofproto,
1857 port->active_iface_tag);
1859 /* Disabling a slave can lead to another slave being immediately
1860 * enabled if there will be no active slaves but one is waiting
1861 * on an updelay. In this case we do not need to run most of the
1862 * code for the newly enabled slave since there was no period
1863 * without an active slave and it is redundant with the disabling
1865 moving_active_iface = true;
1866 bond_choose_active_iface(port);
1868 bond_send_learning_packets(port);
1870 VLOG_WARN("interface %s: enabled", iface->name);
1871 if (port->active_iface < 0 && !moving_active_iface) {
1872 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1873 bond_choose_active_iface(port);
1874 bond_send_learning_packets(port);
1876 iface->tag = tag_create_random();
1879 moving_active_iface = false;
1880 port->bond_compat_is_stale = true;
1883 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1884 * bond interface. */
1886 bond_update_fake_iface_stats(struct port *port)
1888 struct netdev_stats bond_stats;
1889 struct netdev *bond_dev;
1892 memset(&bond_stats, 0, sizeof bond_stats);
1894 for (i = 0; i < port->n_ifaces; i++) {
1895 struct netdev_stats slave_stats;
1897 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1898 /* XXX: We swap the stats here because they are swapped back when
1899 * reported by the internal device. The reason for this is
1900 * internal devices normally represent packets going into the system
1901 * but when used as fake bond device they represent packets leaving
1902 * the system. We really should do this in the internal device
1903 * itself because changing it here reverses the counts from the
1904 * perspective of the switch. However, the internal device doesn't
1905 * know what type of device it represents so we have to do it here
1907 bond_stats.tx_packets += slave_stats.rx_packets;
1908 bond_stats.tx_bytes += slave_stats.rx_bytes;
1909 bond_stats.rx_packets += slave_stats.tx_packets;
1910 bond_stats.rx_bytes += slave_stats.tx_bytes;
1914 if (!netdev_open_default(port->name, &bond_dev)) {
1915 netdev_set_stats(bond_dev, &bond_stats);
1916 netdev_close(bond_dev);
1921 bond_run(struct bridge *br)
1925 for (i = 0; i < br->n_ports; i++) {
1926 struct port *port = br->ports[i];
1928 if (port->n_ifaces >= 2) {
1929 for (j = 0; j < port->n_ifaces; j++) {
1930 struct iface *iface = port->ifaces[j];
1931 if (time_msec() >= iface->delay_expires) {
1932 bond_enable_slave(iface, !iface->enabled);
1936 if (port->bond_fake_iface
1937 && time_msec() >= port->bond_next_fake_iface_update) {
1938 bond_update_fake_iface_stats(port);
1939 port->bond_next_fake_iface_update = time_msec() + 1000;
1943 if (port->bond_compat_is_stale) {
1944 port->bond_compat_is_stale = false;
1945 port_update_bond_compat(port);
1951 bond_wait(struct bridge *br)
1955 for (i = 0; i < br->n_ports; i++) {
1956 struct port *port = br->ports[i];
1957 if (port->n_ifaces < 2) {
1960 for (j = 0; j < port->n_ifaces; j++) {
1961 struct iface *iface = port->ifaces[j];
1962 if (iface->delay_expires != LLONG_MAX) {
1963 poll_timer_wait_until(iface->delay_expires);
1966 if (port->bond_fake_iface) {
1967 poll_timer_wait_until(port->bond_next_fake_iface_update);
1973 set_dst(struct dst *p, const flow_t *flow,
1974 const struct port *in_port, const struct port *out_port,
1977 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1978 : in_port->vlan >= 0 ? in_port->vlan
1979 : ntohs(flow->dl_vlan));
1980 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1984 swap_dst(struct dst *p, struct dst *q)
1986 struct dst tmp = *p;
1991 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1992 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1993 * that we push to the datapath. We could in fact fully sort the array by
1994 * vlan, but in most cases there are at most two different vlan tags so that's
1995 * possibly overkill.) */
1997 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1999 struct dst *first = dsts;
2000 struct dst *last = dsts + n_dsts;
2002 while (first != last) {
2004 * - All dsts < first have vlan == 'vlan'.
2005 * - All dsts >= last have vlan != 'vlan'.
2006 * - first < last. */
2007 while (first->vlan == vlan) {
2008 if (++first == last) {
2013 /* Same invariants, plus one additional:
2014 * - first->vlan != vlan.
2016 while (last[-1].vlan != vlan) {
2017 if (--last == first) {
2022 /* Same invariants, plus one additional:
2023 * - last[-1].vlan == vlan.*/
2024 swap_dst(first++, --last);
2029 mirror_mask_ffs(mirror_mask_t mask)
2031 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2036 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2037 const struct dst *test)
2040 for (i = 0; i < n_dsts; i++) {
2041 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2049 port_trunks_vlan(const struct port *port, uint16_t vlan)
2051 return (port->vlan < 0
2052 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2056 port_includes_vlan(const struct port *port, uint16_t vlan)
2058 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2062 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2063 const struct port *in_port, const struct port *out_port,
2064 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2066 mirror_mask_t mirrors = in_port->src_mirrors;
2067 struct dst *dst = dsts;
2070 if (out_port == FLOOD_PORT) {
2071 /* XXX use ODP_FLOOD if no vlans or bonding. */
2072 /* XXX even better, define each VLAN as a datapath port group */
2073 for (i = 0; i < br->n_ports; i++) {
2074 struct port *port = br->ports[i];
2075 if (port != in_port && port_includes_vlan(port, vlan)
2076 && !port->is_mirror_output_port
2077 && set_dst(dst, flow, in_port, port, tags)) {
2078 mirrors |= port->dst_mirrors;
2082 *nf_output_iface = NF_OUT_FLOOD;
2083 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2084 *nf_output_iface = dst->dp_ifidx;
2085 mirrors |= out_port->dst_mirrors;
2090 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2091 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2093 if (set_dst(dst, flow, in_port, m->out_port, tags)
2094 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2098 for (i = 0; i < br->n_ports; i++) {
2099 struct port *port = br->ports[i];
2100 if (port_includes_vlan(port, m->out_vlan)
2101 && set_dst(dst, flow, in_port, port, tags))
2105 if (port->vlan < 0) {
2106 dst->vlan = m->out_vlan;
2108 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2112 /* Use the vlan tag on the original flow instead of
2113 * the one passed in the vlan parameter. This ensures
2114 * that we compare the vlan from before any implicit
2115 * tagging tags place. This is necessary because
2116 * dst->vlan is the final vlan, after removing implicit
2118 flow_vlan = ntohs(flow->dl_vlan);
2119 if (flow_vlan == 0) {
2120 flow_vlan = OFP_VLAN_NONE;
2122 if (port == in_port && dst->vlan == flow_vlan) {
2123 /* Don't send out input port on same VLAN. */
2131 mirrors &= mirrors - 1;
2134 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2138 static void OVS_UNUSED
2139 print_dsts(const struct dst *dsts, size_t n)
2141 for (; n--; dsts++) {
2142 printf(">p%"PRIu16, dsts->dp_ifidx);
2143 if (dsts->vlan != OFP_VLAN_NONE) {
2144 printf("v%"PRIu16, dsts->vlan);
2150 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2151 const struct port *in_port, const struct port *out_port,
2152 tag_type *tags, struct odp_actions *actions,
2153 uint16_t *nf_output_iface)
2155 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2157 const struct dst *p;
2160 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2163 cur_vlan = ntohs(flow->dl_vlan);
2164 for (p = dsts; p < &dsts[n_dsts]; p++) {
2165 union odp_action *a;
2166 if (p->vlan != cur_vlan) {
2167 if (p->vlan == OFP_VLAN_NONE) {
2168 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2170 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2171 a->vlan_vid.vlan_vid = htons(p->vlan);
2175 a = odp_actions_add(actions, ODPAT_OUTPUT);
2176 a->output.port = p->dp_ifidx;
2180 /* Returns the effective vlan of a packet, taking into account both the
2181 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2182 * the packet is untagged and -1 indicates it has an invalid header and
2183 * should be dropped. */
2184 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2185 struct port *in_port, bool have_packet)
2187 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2188 * belongs to VLAN 0, so we should treat both cases identically. (In the
2189 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2190 * presumably to allow a priority to be specified. In the latter case, the
2191 * packet does not have any 802.1Q header.) */
2192 int vlan = ntohs(flow->dl_vlan);
2193 if (vlan == OFP_VLAN_NONE) {
2196 if (in_port->vlan >= 0) {
2198 /* XXX support double tagging? */
2200 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2201 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2202 "packet received on port %s configured with "
2203 "implicit VLAN %"PRIu16,
2204 br->name, ntohs(flow->dl_vlan),
2205 in_port->name, in_port->vlan);
2209 vlan = in_port->vlan;
2211 if (!port_includes_vlan(in_port, vlan)) {
2213 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2214 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2215 "packet received on port %s not configured for "
2217 br->name, vlan, in_port->name, vlan);
2226 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2227 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2228 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2230 is_gratuitous_arp(const flow_t *flow)
2232 return (flow->dl_type == htons(ETH_TYPE_ARP)
2233 && eth_addr_is_broadcast(flow->dl_dst)
2234 && (flow->nw_proto == ARP_OP_REPLY
2235 || (flow->nw_proto == ARP_OP_REQUEST
2236 && flow->nw_src == flow->nw_dst)));
2240 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2241 struct port *in_port)
2243 enum grat_arp_lock_type lock_type;
2246 /* We don't want to learn from gratuitous ARP packets that are reflected
2247 * back over bond slaves so we lock the learning table. */
2248 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2249 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2250 GRAT_ARP_LOCK_CHECK;
2252 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2255 /* The log messages here could actually be useful in debugging,
2256 * so keep the rate limit relatively high. */
2257 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2259 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2260 "on port %s in VLAN %d",
2261 br->name, ETH_ADDR_ARGS(flow->dl_src),
2262 in_port->name, vlan);
2263 ofproto_revalidate(br->ofproto, rev_tag);
2267 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2268 * dropped. Returns true if they may be forwarded, false if they should be
2271 * If 'have_packet' is true, it indicates that the caller is processing a
2272 * received packet. If 'have_packet' is false, then the caller is just
2273 * revalidating an existing flow because configuration has changed. Either
2274 * way, 'have_packet' only affects logging (there is no point in logging errors
2275 * during revalidation).
2277 * Sets '*in_portp' to the input port. This will be a null pointer if
2278 * flow->in_port does not designate a known input port (in which case
2279 * is_admissible() returns false).
2281 * When returning true, sets '*vlanp' to the effective VLAN of the input
2282 * packet, as returned by flow_get_vlan().
2284 * May also add tags to '*tags', although the current implementation only does
2285 * so in one special case.
2288 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2289 tag_type *tags, int *vlanp, struct port **in_portp)
2291 struct iface *in_iface;
2292 struct port *in_port;
2295 /* Find the interface and port structure for the received packet. */
2296 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2298 /* No interface? Something fishy... */
2300 /* Odd. A few possible reasons here:
2302 * - We deleted an interface but there are still a few packets
2303 * queued up from it.
2305 * - Someone externally added an interface (e.g. with "ovs-dpctl
2306 * add-if") that we don't know about.
2308 * - Packet arrived on the local port but the local port is not
2309 * one of our bridge ports.
2311 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2313 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2314 "interface %"PRIu16, br->name, flow->in_port);
2320 *in_portp = in_port = in_iface->port;
2321 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2326 /* Drop frames for reserved multicast addresses. */
2327 if (eth_addr_is_reserved(flow->dl_dst)) {
2331 /* Drop frames on ports reserved for mirroring. */
2332 if (in_port->is_mirror_output_port) {
2334 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2335 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2336 "%s, which is reserved exclusively for mirroring",
2337 br->name, in_port->name);
2342 /* Packets received on bonds need special attention to avoid duplicates. */
2343 if (in_port->n_ifaces > 1) {
2345 bool is_grat_arp_locked;
2347 if (eth_addr_is_multicast(flow->dl_dst)) {
2348 *tags |= in_port->active_iface_tag;
2349 if (in_port->active_iface != in_iface->port_ifidx) {
2350 /* Drop all multicast packets on inactive slaves. */
2355 /* Drop all packets for which we have learned a different input
2356 * port, because we probably sent the packet on one slave and got
2357 * it back on the other. Gratuitous ARP packets are an exception
2358 * to this rule: the host has moved to another switch. The exception
2359 * to the exception is if we locked the learning table to avoid
2360 * reflections on bond slaves. If this is the case, just drop the
2362 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2363 &is_grat_arp_locked);
2364 if (src_idx != -1 && src_idx != in_port->port_idx &&
2365 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2373 /* If the composed actions may be applied to any packet in the given 'flow',
2374 * returns true. Otherwise, the actions should only be applied to 'packet', or
2375 * not at all, if 'packet' was NULL. */
2377 process_flow(struct bridge *br, const flow_t *flow,
2378 const struct ofpbuf *packet, struct odp_actions *actions,
2379 tag_type *tags, uint16_t *nf_output_iface)
2381 struct port *in_port;
2382 struct port *out_port;
2386 /* Check whether we should drop packets in this flow. */
2387 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2392 /* Learn source MAC (but don't try to learn from revalidation). */
2394 update_learning_table(br, flow, vlan, in_port);
2397 /* Determine output port. */
2398 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2400 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2401 out_port = br->ports[out_port_idx];
2402 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2403 /* If we are revalidating but don't have a learning entry then
2404 * eject the flow. Installing a flow that floods packets opens
2405 * up a window of time where we could learn from a packet reflected
2406 * on a bond and blackhole packets before the learning table is
2407 * updated to reflect the correct port. */
2410 out_port = FLOOD_PORT;
2413 /* Don't send packets out their input ports. */
2414 if (in_port == out_port) {
2420 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2427 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2430 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2431 const struct ofp_phy_port *opp,
2434 struct bridge *br = br_;
2435 struct iface *iface;
2438 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2444 if (reason == OFPPR_DELETE) {
2445 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2446 br->name, iface->name);
2447 iface_destroy(iface);
2448 if (!port->n_ifaces) {
2449 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2450 br->name, port->name);
2456 if (port->n_ifaces > 1) {
2457 bool up = !(opp->state & OFPPS_LINK_DOWN);
2458 bond_link_status_update(iface, up);
2459 port_update_bond_compat(port);
2465 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2466 struct odp_actions *actions, tag_type *tags,
2467 uint16_t *nf_output_iface, void *br_)
2469 struct bridge *br = br_;
2471 COVERAGE_INC(bridge_process_flow);
2472 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2476 bridge_account_flow_ofhook_cb(const flow_t *flow,
2477 const union odp_action *actions,
2478 size_t n_actions, unsigned long long int n_bytes,
2481 struct bridge *br = br_;
2482 const union odp_action *a;
2483 struct port *in_port;
2487 /* Feed information from the active flows back into the learning table
2488 * to ensure that table is always in sync with what is actually flowing
2489 * through the datapath. */
2490 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2491 update_learning_table(br, flow, vlan, in_port);
2494 if (!br->has_bonded_ports) {
2498 for (a = actions; a < &actions[n_actions]; a++) {
2499 if (a->type == ODPAT_OUTPUT) {
2500 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2501 if (out_port && out_port->n_ifaces >= 2) {
2502 struct bond_entry *e = lookup_bond_entry(out_port,
2504 e->tx_bytes += n_bytes;
2511 bridge_account_checkpoint_ofhook_cb(void *br_)
2513 struct bridge *br = br_;
2517 if (!br->has_bonded_ports) {
2522 for (i = 0; i < br->n_ports; i++) {
2523 struct port *port = br->ports[i];
2524 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2525 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2526 bond_rebalance_port(port);
2531 static struct ofhooks bridge_ofhooks = {
2532 bridge_port_changed_ofhook_cb,
2533 bridge_normal_ofhook_cb,
2534 bridge_account_flow_ofhook_cb,
2535 bridge_account_checkpoint_ofhook_cb,
2538 /* Bonding functions. */
2540 /* Statistics for a single interface on a bonded port, used for load-based
2541 * bond rebalancing. */
2542 struct slave_balance {
2543 struct iface *iface; /* The interface. */
2544 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2546 /* All the "bond_entry"s that are assigned to this interface, in order of
2547 * increasing tx_bytes. */
2548 struct bond_entry **hashes;
2552 /* Sorts pointers to pointers to bond_entries in ascending order by the
2553 * interface to which they are assigned, and within a single interface in
2554 * ascending order of bytes transmitted. */
2556 compare_bond_entries(const void *a_, const void *b_)
2558 const struct bond_entry *const *ap = a_;
2559 const struct bond_entry *const *bp = b_;
2560 const struct bond_entry *a = *ap;
2561 const struct bond_entry *b = *bp;
2562 if (a->iface_idx != b->iface_idx) {
2563 return a->iface_idx > b->iface_idx ? 1 : -1;
2564 } else if (a->tx_bytes != b->tx_bytes) {
2565 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2571 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2572 * *descending* order by number of bytes transmitted. */
2574 compare_slave_balance(const void *a_, const void *b_)
2576 const struct slave_balance *a = a_;
2577 const struct slave_balance *b = b_;
2578 if (a->iface->enabled != b->iface->enabled) {
2579 return a->iface->enabled ? -1 : 1;
2580 } else if (a->tx_bytes != b->tx_bytes) {
2581 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2588 swap_bals(struct slave_balance *a, struct slave_balance *b)
2590 struct slave_balance tmp = *a;
2595 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2596 * given that 'p' (and only 'p') might be in the wrong location.
2598 * This function invalidates 'p', since it might now be in a different memory
2601 resort_bals(struct slave_balance *p,
2602 struct slave_balance bals[], size_t n_bals)
2605 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2606 swap_bals(p, p - 1);
2608 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2609 swap_bals(p, p + 1);
2615 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2617 if (VLOG_IS_DBG_ENABLED()) {
2618 struct ds ds = DS_EMPTY_INITIALIZER;
2619 const struct slave_balance *b;
2621 for (b = bals; b < bals + n_bals; b++) {
2625 ds_put_char(&ds, ',');
2627 ds_put_format(&ds, " %s %"PRIu64"kB",
2628 b->iface->name, b->tx_bytes / 1024);
2630 if (!b->iface->enabled) {
2631 ds_put_cstr(&ds, " (disabled)");
2633 if (b->n_hashes > 0) {
2634 ds_put_cstr(&ds, " (");
2635 for (i = 0; i < b->n_hashes; i++) {
2636 const struct bond_entry *e = b->hashes[i];
2638 ds_put_cstr(&ds, " + ");
2640 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2641 e - port->bond_hash, e->tx_bytes / 1024);
2643 ds_put_cstr(&ds, ")");
2646 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2651 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2653 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2656 struct bond_entry *hash = from->hashes[hash_idx];
2657 struct port *port = from->iface->port;
2658 uint64_t delta = hash->tx_bytes;
2660 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2661 "from %s to %s (now carrying %"PRIu64"kB and "
2662 "%"PRIu64"kB load, respectively)",
2663 port->name, delta / 1024, hash - port->bond_hash,
2664 from->iface->name, to->iface->name,
2665 (from->tx_bytes - delta) / 1024,
2666 (to->tx_bytes + delta) / 1024);
2668 /* Delete element from from->hashes.
2670 * We don't bother to add the element to to->hashes because not only would
2671 * it require more work, the only purpose it would be to allow that hash to
2672 * be migrated to another slave in this rebalancing run, and there is no
2673 * point in doing that. */
2674 if (hash_idx == 0) {
2677 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2678 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2682 /* Shift load away from 'from' to 'to'. */
2683 from->tx_bytes -= delta;
2684 to->tx_bytes += delta;
2686 /* Arrange for flows to be revalidated. */
2687 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2688 hash->iface_idx = to->iface->port_ifidx;
2689 hash->iface_tag = tag_create_random();
2693 bond_rebalance_port(struct port *port)
2695 struct slave_balance bals[DP_MAX_PORTS];
2697 struct bond_entry *hashes[BOND_MASK + 1];
2698 struct slave_balance *b, *from, *to;
2699 struct bond_entry *e;
2702 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2703 * descending order of tx_bytes, so that bals[0] represents the most
2704 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2707 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2708 * array for each slave_balance structure, we sort our local array of
2709 * hashes in order by slave, so that all of the hashes for a given slave
2710 * become contiguous in memory, and then we point each 'hashes' members of
2711 * a slave_balance structure to the start of a contiguous group. */
2712 n_bals = port->n_ifaces;
2713 for (b = bals; b < &bals[n_bals]; b++) {
2714 b->iface = port->ifaces[b - bals];
2719 for (i = 0; i <= BOND_MASK; i++) {
2720 hashes[i] = &port->bond_hash[i];
2722 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2723 for (i = 0; i <= BOND_MASK; i++) {
2725 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2726 b = &bals[e->iface_idx];
2727 b->tx_bytes += e->tx_bytes;
2729 b->hashes = &hashes[i];
2734 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2735 log_bals(bals, n_bals, port);
2737 /* Discard slaves that aren't enabled (which were sorted to the back of the
2738 * array earlier). */
2739 while (!bals[n_bals - 1].iface->enabled) {
2746 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2747 to = &bals[n_bals - 1];
2748 for (from = bals; from < to; ) {
2749 uint64_t overload = from->tx_bytes - to->tx_bytes;
2750 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2751 /* The extra load on 'from' (and all less-loaded slaves), compared
2752 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2753 * it is less than ~1Mbps. No point in rebalancing. */
2755 } else if (from->n_hashes == 1) {
2756 /* 'from' only carries a single MAC hash, so we can't shift any
2757 * load away from it, even though we want to. */
2760 /* 'from' is carrying significantly more load than 'to', and that
2761 * load is split across at least two different hashes. Pick a hash
2762 * to migrate to 'to' (the least-loaded slave), given that doing so
2763 * must decrease the ratio of the load on the two slaves by at
2766 * The sort order we use means that we prefer to shift away the
2767 * smallest hashes instead of the biggest ones. There is little
2768 * reason behind this decision; we could use the opposite sort
2769 * order to shift away big hashes ahead of small ones. */
2773 for (i = 0; i < from->n_hashes; i++) {
2774 double old_ratio, new_ratio;
2775 uint64_t delta = from->hashes[i]->tx_bytes;
2777 if (delta == 0 || from->tx_bytes - delta == 0) {
2778 /* Pointless move. */
2782 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2784 if (to->tx_bytes == 0) {
2785 /* Nothing on the new slave, move it. */
2789 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2790 new_ratio = (double)(from->tx_bytes - delta) /
2791 (to->tx_bytes + delta);
2793 if (new_ratio == 0) {
2794 /* Should already be covered but check to prevent division
2799 if (new_ratio < 1) {
2800 new_ratio = 1 / new_ratio;
2803 if (old_ratio - new_ratio > 0.1) {
2804 /* Would decrease the ratio, move it. */
2808 if (i < from->n_hashes) {
2809 bond_shift_load(from, to, i);
2810 port->bond_compat_is_stale = true;
2812 /* If the result of the migration changed the relative order of
2813 * 'from' and 'to' swap them back to maintain invariants. */
2814 if (order_swapped) {
2815 swap_bals(from, to);
2818 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2819 * point to different slave_balance structures. It is only
2820 * valid to do these two operations in a row at all because we
2821 * know that 'from' will not move past 'to' and vice versa. */
2822 resort_bals(from, bals, n_bals);
2823 resort_bals(to, bals, n_bals);
2830 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2831 * historical data to decay to <1% in 7 rebalancing runs. */
2832 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2838 bond_send_learning_packets(struct port *port)
2840 struct bridge *br = port->bridge;
2841 struct mac_entry *e;
2842 struct ofpbuf packet;
2843 int error, n_packets, n_errors;
2845 if (!port->n_ifaces || port->active_iface < 0) {
2849 ofpbuf_init(&packet, 128);
2850 error = n_packets = n_errors = 0;
2851 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2852 union ofp_action actions[2], *a;
2858 if (e->port == port->port_idx
2859 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2863 /* Compose actions. */
2864 memset(actions, 0, sizeof actions);
2867 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2868 a->vlan_vid.len = htons(sizeof *a);
2869 a->vlan_vid.vlan_vid = htons(e->vlan);
2872 a->output.type = htons(OFPAT_OUTPUT);
2873 a->output.len = htons(sizeof *a);
2874 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2879 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2881 flow_extract(&packet, 0, ODPP_NONE, &flow);
2882 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2889 ofpbuf_uninit(&packet);
2892 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2893 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2894 "packets, last error was: %s",
2895 port->name, n_errors, n_packets, strerror(error));
2897 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2898 port->name, n_packets);
2902 /* Bonding unixctl user interface functions. */
2905 bond_unixctl_list(struct unixctl_conn *conn,
2906 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2908 struct ds ds = DS_EMPTY_INITIALIZER;
2909 const struct bridge *br;
2911 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2913 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2916 for (i = 0; i < br->n_ports; i++) {
2917 const struct port *port = br->ports[i];
2918 if (port->n_ifaces > 1) {
2921 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2922 for (j = 0; j < port->n_ifaces; j++) {
2923 const struct iface *iface = port->ifaces[j];
2925 ds_put_cstr(&ds, ", ");
2927 ds_put_cstr(&ds, iface->name);
2929 ds_put_char(&ds, '\n');
2933 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2937 static struct port *
2938 bond_find(const char *name)
2940 const struct bridge *br;
2942 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2945 for (i = 0; i < br->n_ports; i++) {
2946 struct port *port = br->ports[i];
2947 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2956 bond_unixctl_show(struct unixctl_conn *conn,
2957 const char *args, void *aux OVS_UNUSED)
2959 struct ds ds = DS_EMPTY_INITIALIZER;
2960 const struct port *port;
2963 port = bond_find(args);
2965 unixctl_command_reply(conn, 501, "no such bond");
2969 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2970 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2971 ds_put_format(&ds, "next rebalance: %lld ms\n",
2972 port->bond_next_rebalance - time_msec());
2973 for (j = 0; j < port->n_ifaces; j++) {
2974 const struct iface *iface = port->ifaces[j];
2975 struct bond_entry *be;
2978 ds_put_format(&ds, "slave %s: %s\n",
2979 iface->name, iface->enabled ? "enabled" : "disabled");
2980 if (j == port->active_iface) {
2981 ds_put_cstr(&ds, "\tactive slave\n");
2983 if (iface->delay_expires != LLONG_MAX) {
2984 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2985 iface->enabled ? "downdelay" : "updelay",
2986 iface->delay_expires - time_msec());
2990 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2991 int hash = be - port->bond_hash;
2992 struct mac_entry *me;
2994 if (be->iface_idx != j) {
2998 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2999 hash, be->tx_bytes / 1024);
3002 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3003 &port->bridge->ml->lrus) {
3006 if (bond_hash(me->mac) == hash
3007 && me->port != port->port_idx
3008 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3009 && dp_ifidx == iface->dp_ifidx)
3011 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3012 ETH_ADDR_ARGS(me->mac));
3017 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3022 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3023 void *aux OVS_UNUSED)
3025 char *args = (char *) args_;
3026 char *save_ptr = NULL;
3027 char *bond_s, *hash_s, *slave_s;
3028 uint8_t mac[ETH_ADDR_LEN];
3030 struct iface *iface;
3031 struct bond_entry *entry;
3034 bond_s = strtok_r(args, " ", &save_ptr);
3035 hash_s = strtok_r(NULL, " ", &save_ptr);
3036 slave_s = strtok_r(NULL, " ", &save_ptr);
3038 unixctl_command_reply(conn, 501,
3039 "usage: bond/migrate BOND HASH SLAVE");
3043 port = bond_find(bond_s);
3045 unixctl_command_reply(conn, 501, "no such bond");
3049 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3050 == ETH_ADDR_SCAN_COUNT) {
3051 hash = bond_hash(mac);
3052 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3053 hash = atoi(hash_s) & BOND_MASK;
3055 unixctl_command_reply(conn, 501, "bad hash");
3059 iface = port_lookup_iface(port, slave_s);
3061 unixctl_command_reply(conn, 501, "no such slave");
3065 if (!iface->enabled) {
3066 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3070 entry = &port->bond_hash[hash];
3071 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3072 entry->iface_idx = iface->port_ifidx;
3073 entry->iface_tag = tag_create_random();
3074 port->bond_compat_is_stale = true;
3075 unixctl_command_reply(conn, 200, "migrated");
3079 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3080 void *aux OVS_UNUSED)
3082 char *args = (char *) args_;
3083 char *save_ptr = NULL;
3084 char *bond_s, *slave_s;
3086 struct iface *iface;
3088 bond_s = strtok_r(args, " ", &save_ptr);
3089 slave_s = strtok_r(NULL, " ", &save_ptr);
3091 unixctl_command_reply(conn, 501,
3092 "usage: bond/set-active-slave BOND SLAVE");
3096 port = bond_find(bond_s);
3098 unixctl_command_reply(conn, 501, "no such bond");
3102 iface = port_lookup_iface(port, slave_s);
3104 unixctl_command_reply(conn, 501, "no such slave");
3108 if (!iface->enabled) {
3109 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3113 if (port->active_iface != iface->port_ifidx) {
3114 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3115 port->active_iface = iface->port_ifidx;
3116 port->active_iface_tag = tag_create_random();
3117 VLOG_INFO("port %s: active interface is now %s",
3118 port->name, iface->name);
3119 bond_send_learning_packets(port);
3120 unixctl_command_reply(conn, 200, "done");
3122 unixctl_command_reply(conn, 200, "no change");
3127 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3129 char *args = (char *) args_;
3130 char *save_ptr = NULL;
3131 char *bond_s, *slave_s;
3133 struct iface *iface;
3135 bond_s = strtok_r(args, " ", &save_ptr);
3136 slave_s = strtok_r(NULL, " ", &save_ptr);
3138 unixctl_command_reply(conn, 501,
3139 "usage: bond/enable/disable-slave BOND SLAVE");
3143 port = bond_find(bond_s);
3145 unixctl_command_reply(conn, 501, "no such bond");
3149 iface = port_lookup_iface(port, slave_s);
3151 unixctl_command_reply(conn, 501, "no such slave");
3155 bond_enable_slave(iface, enable);
3156 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3160 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3161 void *aux OVS_UNUSED)
3163 enable_slave(conn, args, true);
3167 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3168 void *aux OVS_UNUSED)
3170 enable_slave(conn, args, false);
3174 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3175 void *aux OVS_UNUSED)
3177 uint8_t mac[ETH_ADDR_LEN];
3181 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3182 == ETH_ADDR_SCAN_COUNT) {
3183 hash = bond_hash(mac);
3185 hash_cstr = xasprintf("%u", hash);
3186 unixctl_command_reply(conn, 200, hash_cstr);
3189 unixctl_command_reply(conn, 501, "invalid mac");
3196 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3197 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3198 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3199 unixctl_command_register("bond/set-active-slave",
3200 bond_unixctl_set_active_slave, NULL);
3201 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3203 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3205 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3208 /* Port functions. */
3210 static struct port *
3211 port_create(struct bridge *br, const char *name)
3215 port = xzalloc(sizeof *port);
3217 port->port_idx = br->n_ports;
3219 port->trunks = NULL;
3220 port->name = xstrdup(name);
3221 port->active_iface = -1;
3223 if (br->n_ports >= br->allocated_ports) {
3224 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3227 br->ports[br->n_ports++] = port;
3228 shash_add_assert(&br->port_by_name, port->name, port);
3230 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3237 get_port_other_config(const struct ovsrec_port *port, const char *key,
3238 const char *default_value)
3242 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3244 return value ? value : default_value;
3248 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3250 struct shash new_ifaces;
3253 /* Collect list of new interfaces. */
3254 shash_init(&new_ifaces);
3255 for (i = 0; i < cfg->n_interfaces; i++) {
3256 const char *name = cfg->interfaces[i]->name;
3257 shash_add_once(&new_ifaces, name, NULL);
3260 /* Get rid of deleted interfaces. */
3261 for (i = 0; i < port->n_ifaces; ) {
3262 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3263 iface_destroy(port->ifaces[i]);
3269 shash_destroy(&new_ifaces);
3273 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3275 struct shash new_ifaces;
3276 long long int next_rebalance;
3277 unsigned long *trunks;
3283 /* Update settings. */
3284 port->updelay = cfg->bond_updelay;
3285 if (port->updelay < 0) {
3288 port->updelay = cfg->bond_downdelay;
3289 if (port->downdelay < 0) {
3290 port->downdelay = 0;
3292 port->bond_rebalance_interval = atoi(
3293 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3294 if (port->bond_rebalance_interval < 1000) {
3295 port->bond_rebalance_interval = 1000;
3297 next_rebalance = time_msec() + port->bond_rebalance_interval;
3298 if (port->bond_next_rebalance > next_rebalance) {
3299 port->bond_next_rebalance = next_rebalance;
3302 /* Add new interfaces and update 'cfg' member of existing ones. */
3303 shash_init(&new_ifaces);
3304 for (i = 0; i < cfg->n_interfaces; i++) {
3305 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3306 struct iface *iface;
3308 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3309 VLOG_WARN("port %s: %s specified twice as port interface",
3310 port->name, if_cfg->name);
3314 iface = iface_lookup(port->bridge, if_cfg->name);
3316 if (iface->port != port) {
3317 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3319 port->bridge->name, if_cfg->name, iface->port->name);
3322 iface->cfg = if_cfg;
3324 iface_create(port, if_cfg);
3327 shash_destroy(&new_ifaces);
3332 if (port->n_ifaces < 2) {
3334 if (vlan >= 0 && vlan <= 4095) {
3335 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3340 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3341 * they even work as-is. But they have not been tested. */
3342 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3346 if (port->vlan != vlan) {
3348 bridge_flush(port->bridge);
3351 /* Get trunked VLANs. */
3353 if (vlan < 0 && cfg->n_trunks) {
3357 trunks = bitmap_allocate(4096);
3359 for (i = 0; i < cfg->n_trunks; i++) {
3360 int trunk = cfg->trunks[i];
3362 bitmap_set1(trunks, trunk);
3368 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3369 port->name, cfg->n_trunks);
3371 if (n_errors == cfg->n_trunks) {
3372 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3374 bitmap_free(trunks);
3377 } else if (vlan >= 0 && cfg->n_trunks) {
3378 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3382 ? port->trunks != NULL
3383 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3384 bridge_flush(port->bridge);
3386 bitmap_free(port->trunks);
3387 port->trunks = trunks;
3391 port_destroy(struct port *port)
3394 struct bridge *br = port->bridge;
3398 proc_net_compat_update_vlan(port->name, NULL, 0);
3399 proc_net_compat_update_bond(port->name, NULL);
3401 for (i = 0; i < MAX_MIRRORS; i++) {
3402 struct mirror *m = br->mirrors[i];
3403 if (m && m->out_port == port) {
3408 while (port->n_ifaces > 0) {
3409 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3412 shash_find_and_delete_assert(&br->port_by_name, port->name);
3414 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3415 del->port_idx = port->port_idx;
3418 bitmap_free(port->trunks);
3425 static struct port *
3426 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3428 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3429 return iface ? iface->port : NULL;
3432 static struct port *
3433 port_lookup(const struct bridge *br, const char *name)
3435 return shash_find_data(&br->port_by_name, name);
3438 static struct iface *
3439 port_lookup_iface(const struct port *port, const char *name)
3441 struct iface *iface = iface_lookup(port->bridge, name);
3442 return iface && iface->port == port ? iface : NULL;
3446 port_update_bonding(struct port *port)
3448 if (port->n_ifaces < 2) {
3449 /* Not a bonded port. */
3450 if (port->bond_hash) {
3451 free(port->bond_hash);
3452 port->bond_hash = NULL;
3453 port->bond_compat_is_stale = true;
3454 port->bond_fake_iface = false;
3457 if (!port->bond_hash) {
3460 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3461 for (i = 0; i <= BOND_MASK; i++) {
3462 struct bond_entry *e = &port->bond_hash[i];
3466 port->no_ifaces_tag = tag_create_random();
3467 bond_choose_active_iface(port);
3468 port->bond_next_rebalance
3469 = time_msec() + port->bond_rebalance_interval;
3471 if (port->cfg->bond_fake_iface) {
3472 port->bond_next_fake_iface_update = time_msec();
3475 port->bond_compat_is_stale = true;
3476 port->bond_fake_iface = port->cfg->bond_fake_iface;
3481 port_update_bond_compat(struct port *port)
3483 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3484 struct compat_bond bond;
3487 if (port->n_ifaces < 2) {
3488 proc_net_compat_update_bond(port->name, NULL);
3493 bond.updelay = port->updelay;
3494 bond.downdelay = port->downdelay;
3497 bond.hashes = compat_hashes;
3498 if (port->bond_hash) {
3499 const struct bond_entry *e;
3500 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3501 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3502 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3503 cbh->hash = e - port->bond_hash;
3504 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3509 bond.n_slaves = port->n_ifaces;
3510 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3511 for (i = 0; i < port->n_ifaces; i++) {
3512 struct iface *iface = port->ifaces[i];
3513 struct compat_bond_slave *slave = &bond.slaves[i];
3514 slave->name = iface->name;
3516 /* We need to make the same determination as the Linux bonding
3517 * code to determine whether a slave should be consider "up".
3518 * The Linux function bond_miimon_inspect() supports four
3519 * BOND_LINK_* states:
3521 * - BOND_LINK_UP: carrier detected, updelay has passed.
3522 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3523 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3524 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3526 * The function bond_info_show_slave() only considers BOND_LINK_UP
3527 * to be "up" and anything else to be "down".
3529 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3533 netdev_get_etheraddr(iface->netdev, slave->mac);
3536 if (port->bond_fake_iface) {
3537 struct netdev *bond_netdev;
3539 if (!netdev_open_default(port->name, &bond_netdev)) {
3541 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3543 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3545 netdev_close(bond_netdev);
3549 proc_net_compat_update_bond(port->name, &bond);
3554 port_update_vlan_compat(struct port *port)
3556 struct bridge *br = port->bridge;
3557 char *vlandev_name = NULL;
3559 if (port->vlan > 0) {
3560 /* Figure out the name that the VLAN device should actually have, if it
3561 * existed. This takes some work because the VLAN device would not
3562 * have port->name in its name; rather, it would have the trunk port's
3563 * name, and 'port' would be attached to a bridge that also had the
3564 * VLAN device one of its ports. So we need to find a trunk port that
3565 * includes port->vlan.
3567 * There might be more than one candidate. This doesn't happen on
3568 * XenServer, so if it happens we just pick the first choice in
3569 * alphabetical order instead of creating multiple VLAN devices. */
3571 for (i = 0; i < br->n_ports; i++) {
3572 struct port *p = br->ports[i];
3573 if (port_trunks_vlan(p, port->vlan)
3575 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3577 uint8_t ea[ETH_ADDR_LEN];
3578 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3579 if (!eth_addr_is_multicast(ea) &&
3580 !eth_addr_is_reserved(ea) &&
3581 !eth_addr_is_zero(ea)) {
3582 vlandev_name = p->name;
3587 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3590 /* Interface functions. */
3592 static struct iface *
3593 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3595 struct bridge *br = port->bridge;
3596 struct iface *iface;
3597 char *name = if_cfg->name;
3600 iface = xzalloc(sizeof *iface);
3602 iface->port_ifidx = port->n_ifaces;
3603 iface->name = xstrdup(name);
3604 iface->dp_ifidx = -1;
3605 iface->tag = tag_create_random();
3606 iface->delay_expires = LLONG_MAX;
3607 iface->netdev = NULL;
3608 iface->cfg = if_cfg;
3610 shash_add_assert(&br->iface_by_name, iface->name, iface);
3612 /* Attempt to create the network interface in case it doesn't exist yet. */
3613 if (!iface_is_internal(br, iface->name)) {
3614 error = set_up_iface(if_cfg, iface, true);
3616 VLOG_WARN("could not create iface %s: %s", iface->name,
3619 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3626 if (port->n_ifaces >= port->allocated_ifaces) {
3627 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3628 sizeof *port->ifaces);
3630 port->ifaces[port->n_ifaces++] = iface;
3631 if (port->n_ifaces > 1) {
3632 br->has_bonded_ports = true;
3635 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3643 iface_destroy(struct iface *iface)
3646 struct port *port = iface->port;
3647 struct bridge *br = port->bridge;
3648 bool del_active = port->active_iface == iface->port_ifidx;
3651 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3653 if (iface->dp_ifidx >= 0) {
3654 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3657 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3658 del->port_ifidx = iface->port_ifidx;
3660 netdev_close(iface->netdev);
3663 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3664 bond_choose_active_iface(port);
3665 bond_send_learning_packets(port);
3671 bridge_flush(port->bridge);
3675 static struct iface *
3676 iface_lookup(const struct bridge *br, const char *name)
3678 return shash_find_data(&br->iface_by_name, name);
3681 static struct iface *
3682 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3684 return port_array_get(&br->ifaces, dp_ifidx);
3687 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3688 * 'br', that is, an interface that is entirely simulated within the datapath.
3689 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3690 * interfaces are created by setting "iface.<iface>.internal = true".
3692 * In addition, we have a kluge-y feature that creates an internal port with
3693 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3694 * This feature needs to go away in the long term. Until then, this is one
3695 * reason why this function takes a name instead of a struct iface: the fake
3696 * interfaces created this way do not have a struct iface. */
3698 iface_is_internal(const struct bridge *br, const char *if_name)
3700 struct iface *iface;
3703 if (!strcmp(if_name, br->name)) {
3707 iface = iface_lookup(br, if_name);
3708 if (iface && !strcmp(iface->cfg->type, "internal")) {
3712 port = port_lookup(br, if_name);
3713 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3719 /* Set Ethernet address of 'iface', if one is specified in the configuration
3722 iface_set_mac(struct iface *iface)
3724 uint8_t ea[ETH_ADDR_LEN];
3726 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3727 if (eth_addr_is_multicast(ea)) {
3728 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3730 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3731 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3732 iface->name, iface->name);
3734 int error = netdev_set_etheraddr(iface->netdev, ea);
3736 VLOG_ERR("interface %s: setting MAC failed (%s)",
3737 iface->name, strerror(error));
3744 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3745 struct shash *shash)
3750 for (i = 0; i < n; i++) {
3751 shash_add(shash, keys[i], values[i]);
3755 struct iface_delete_queues_cbdata {
3756 struct netdev *netdev;
3757 const int64_t *queue_ids;
3762 queue_ids_include(const int64_t *ids, size_t n, int64_t target)
3767 while (low < high) {
3768 size_t mid = low + (high - low) / 2;
3769 if (target > ids[mid]) {
3771 } else if (target < ids[mid]) {
3781 iface_delete_queues(unsigned int queue_id,
3782 const struct shash *details OVS_UNUSED, void *cbdata_)
3784 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3786 if (!queue_ids_include(cbdata->queue_ids, cbdata->n_queue_ids, queue_id)) {
3787 netdev_delete_queue(cbdata->netdev, queue_id);
3792 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3794 if (!qos || qos->type[0] == '\0') {
3795 netdev_set_qos(iface->netdev, NULL, NULL);
3797 struct iface_delete_queues_cbdata cbdata;
3798 struct shash details;
3801 /* Configure top-level Qos for 'iface'. */
3802 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3803 qos->n_other_config, &details);
3804 netdev_set_qos(iface->netdev, qos->type, &details);
3805 shash_destroy(&details);
3807 /* Deconfigure queues that were deleted. */
3808 cbdata.netdev = iface->netdev;
3809 cbdata.queue_ids = qos->key_queues;
3810 cbdata.n_queue_ids = qos->n_queues;
3811 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3813 /* Configure queues for 'iface'. */
3814 for (i = 0; i < qos->n_queues; i++) {
3815 const struct ovsrec_queue *queue = qos->value_queues[i];
3816 unsigned int queue_id = qos->key_queues[i];
3818 shash_from_ovs_idl_map(queue->key_other_config,
3819 queue->value_other_config,
3820 queue->n_other_config, &details);
3821 netdev_set_queue(iface->netdev, queue_id, &details);
3822 shash_destroy(&details);
3827 /* Port mirroring. */
3830 mirror_reconfigure(struct bridge *br)
3832 struct shash old_mirrors, new_mirrors;
3833 struct shash_node *node;
3834 unsigned long *rspan_vlans;
3837 /* Collect old mirrors. */
3838 shash_init(&old_mirrors);
3839 for (i = 0; i < MAX_MIRRORS; i++) {
3840 if (br->mirrors[i]) {
3841 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3845 /* Collect new mirrors. */
3846 shash_init(&new_mirrors);
3847 for (i = 0; i < br->cfg->n_mirrors; i++) {
3848 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3849 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3850 VLOG_WARN("bridge %s: %s specified twice as mirror",
3851 br->name, cfg->name);
3855 /* Get rid of deleted mirrors and add new mirrors. */
3856 SHASH_FOR_EACH (node, &old_mirrors) {
3857 if (!shash_find(&new_mirrors, node->name)) {
3858 mirror_destroy(node->data);
3861 SHASH_FOR_EACH (node, &new_mirrors) {
3862 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3864 mirror = mirror_create(br, node->name);
3869 mirror_reconfigure_one(mirror, node->data);
3871 shash_destroy(&old_mirrors);
3872 shash_destroy(&new_mirrors);
3874 /* Update port reserved status. */
3875 for (i = 0; i < br->n_ports; i++) {
3876 br->ports[i]->is_mirror_output_port = false;
3878 for (i = 0; i < MAX_MIRRORS; i++) {
3879 struct mirror *m = br->mirrors[i];
3880 if (m && m->out_port) {
3881 m->out_port->is_mirror_output_port = true;
3885 /* Update flooded vlans (for RSPAN). */
3887 if (br->cfg->n_flood_vlans) {
3888 rspan_vlans = bitmap_allocate(4096);
3890 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3891 int64_t vlan = br->cfg->flood_vlans[i];
3892 if (vlan >= 0 && vlan < 4096) {
3893 bitmap_set1(rspan_vlans, vlan);
3894 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3897 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3902 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3907 static struct mirror *
3908 mirror_create(struct bridge *br, const char *name)
3913 for (i = 0; ; i++) {
3914 if (i >= MAX_MIRRORS) {
3915 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3916 "cannot create %s", br->name, MAX_MIRRORS, name);
3919 if (!br->mirrors[i]) {
3924 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3927 br->mirrors[i] = m = xzalloc(sizeof *m);
3930 m->name = xstrdup(name);
3931 shash_init(&m->src_ports);
3932 shash_init(&m->dst_ports);
3942 mirror_destroy(struct mirror *m)
3945 struct bridge *br = m->bridge;
3948 for (i = 0; i < br->n_ports; i++) {
3949 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3950 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3953 shash_destroy(&m->src_ports);
3954 shash_destroy(&m->dst_ports);
3957 m->bridge->mirrors[m->idx] = NULL;
3965 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3966 struct shash *names)
3970 for (i = 0; i < n_ports; i++) {
3971 const char *name = ports[i]->name;
3972 if (port_lookup(m->bridge, name)) {
3973 shash_add_once(names, name, NULL);
3975 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3976 "port %s", m->bridge->name, m->name, name);
3982 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3988 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3990 for (i = 0; i < cfg->n_select_vlan; i++) {
3991 int64_t vlan = cfg->select_vlan[i];
3992 if (vlan < 0 || vlan > 4095) {
3993 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3994 m->bridge->name, m->name, vlan);
3996 (*vlans)[n_vlans++] = vlan;
4003 vlan_is_mirrored(const struct mirror *m, int vlan)
4007 for (i = 0; i < m->n_vlans; i++) {
4008 if (m->vlans[i] == vlan) {
4016 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4020 for (i = 0; i < m->n_vlans; i++) {
4021 if (port_trunks_vlan(p, m->vlans[i])) {
4029 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4031 struct shash src_ports, dst_ports;
4032 mirror_mask_t mirror_bit;
4033 struct port *out_port;
4039 /* Get output port. */
4040 if (cfg->output_port) {
4041 out_port = port_lookup(m->bridge, cfg->output_port->name);
4043 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4044 m->bridge->name, m->name);
4050 if (cfg->output_vlan) {
4051 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4052 "output vlan; ignoring output vlan",
4053 m->bridge->name, m->name);
4055 } else if (cfg->output_vlan) {
4057 out_vlan = *cfg->output_vlan;
4059 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4060 m->bridge->name, m->name);
4065 shash_init(&src_ports);
4066 shash_init(&dst_ports);
4067 if (cfg->select_all) {
4068 for (i = 0; i < m->bridge->n_ports; i++) {
4069 const char *name = m->bridge->ports[i]->name;
4070 shash_add_once(&src_ports, name, NULL);
4071 shash_add_once(&dst_ports, name, NULL);
4076 /* Get ports, and drop duplicates and ports that don't exist. */
4077 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4079 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4082 /* Get all the vlans, and drop duplicate and invalid vlans. */
4083 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4086 /* Update mirror data. */
4087 if (!shash_equal_keys(&m->src_ports, &src_ports)
4088 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4089 || m->n_vlans != n_vlans
4090 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4091 || m->out_port != out_port
4092 || m->out_vlan != out_vlan) {
4093 bridge_flush(m->bridge);
4095 shash_swap(&m->src_ports, &src_ports);
4096 shash_swap(&m->dst_ports, &dst_ports);
4099 m->n_vlans = n_vlans;
4100 m->out_port = out_port;
4101 m->out_vlan = out_vlan;
4104 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4105 for (i = 0; i < m->bridge->n_ports; i++) {
4106 struct port *port = m->bridge->ports[i];
4108 if (shash_find(&m->src_ports, port->name)
4111 ? port_trunks_any_mirrored_vlan(m, port)
4112 : vlan_is_mirrored(m, port->vlan)))) {
4113 port->src_mirrors |= mirror_bit;
4115 port->src_mirrors &= ~mirror_bit;
4118 if (shash_find(&m->dst_ports, port->name)) {
4119 port->dst_mirrors |= mirror_bit;
4121 port->dst_mirrors &= ~mirror_bit;
4126 shash_destroy(&src_ports);
4127 shash_destroy(&dst_ports);