1 /* Copyright (c) 2008, 2009 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>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id;
74 struct port *port; /* Containing port. */
75 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 int dp_ifidx; /* Index within kernel datapath. */
80 uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */
82 tag_type tag; /* Tag associated with this interface. */
83 bool enabled; /* May be chosen for flows? */
84 long long delay_expires; /* Time after which 'enabled' may change. */
87 #define BOND_MASK 0xff
89 int iface_idx; /* Index of assigned iface, or -1 if none. */
90 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
91 tag_type iface_tag; /* Tag associated with iface_idx. */
94 #define MAX_MIRRORS 32
95 typedef uint32_t mirror_mask_t;
96 #define MIRROR_MASK_C(X) UINT32_C(X)
97 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
99 struct bridge *bridge;
103 /* Selection criteria. */
104 struct svec src_ports;
105 struct svec dst_ports;
110 struct port *out_port;
114 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
116 struct bridge *bridge;
118 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
119 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
122 /* An ordinary bridge port has 1 interface.
123 * A bridge port for bonding has at least 2 interfaces. */
124 struct iface **ifaces;
125 size_t n_ifaces, allocated_ifaces;
128 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
129 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
130 tag_type active_iface_tag; /* Tag for bcast flows. */
131 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
132 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
134 /* Port mirroring info. */
135 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
136 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
137 bool is_mirror_output_port; /* Does port mirroring send frames here? */
139 /* Spanning tree info. */
140 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
141 tag_type stp_state_tag; /* Tag for STP state change. */
144 #define DP_MAX_PORTS 255
146 struct list node; /* Node in global list of bridges. */
147 char *name; /* User-specified arbitrary name. */
148 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
149 bool sent_config_request; /* Successfully sent config request? */
150 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
152 /* Support for remote controllers. */
153 char *controller; /* NULL if there is no remote controller;
154 * "discover" to do controller discovery;
155 * otherwise a vconn name. */
157 /* OpenFlow switch processing. */
158 struct ofproto *ofproto; /* OpenFlow switch. */
160 /* Kernel datapath information. */
161 struct dpif *dpif; /* Datapath. */
162 struct port_array ifaces; /* Indexed by kernel datapath port number. */
166 size_t n_ports, allocated_ports;
169 bool has_bonded_ports;
170 long long int bond_next_rebalance;
175 /* Flow statistics gathering. */
176 time_t next_stats_request;
178 /* Port mirroring. */
179 struct mirror *mirrors[MAX_MIRRORS];
183 long long int stp_last_tick;
186 /* List of all bridges. */
187 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
189 /* Maximum number of datapaths. */
190 enum { DP_MAX = 256 };
192 static struct bridge *bridge_create(const char *name);
193 static void bridge_destroy(struct bridge *);
194 static struct bridge *bridge_lookup(const char *name);
195 static int bridge_run_one(struct bridge *);
196 static void bridge_reconfigure_one(struct bridge *);
197 static void bridge_reconfigure_controller(struct bridge *);
198 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
199 static void bridge_fetch_dp_ifaces(struct bridge *);
200 static void bridge_flush(struct bridge *);
201 static void bridge_pick_local_hw_addr(struct bridge *,
202 uint8_t ea[ETH_ADDR_LEN],
203 const char **devname);
204 static uint64_t bridge_pick_datapath_id(struct bridge *,
205 const uint8_t bridge_ea[ETH_ADDR_LEN],
206 const char *devname);
207 static uint64_t dpid_from_hash(const void *, size_t nbytes);
209 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
211 static void bond_init(void);
212 static void bond_run(struct bridge *);
213 static void bond_wait(struct bridge *);
214 static void bond_rebalance_port(struct port *);
215 static void bond_send_learning_packets(struct port *);
217 static void port_create(struct bridge *, const char *name);
218 static void port_reconfigure(struct port *);
219 static void port_destroy(struct port *);
220 static struct port *port_lookup(const struct bridge *, const char *name);
221 static struct iface *port_lookup_iface(const struct port *, const char *name);
222 static struct port *port_from_dp_ifidx(const struct bridge *,
224 static void port_update_bond_compat(struct port *);
225 static void port_update_vlan_compat(struct port *);
227 static void mirror_create(struct bridge *, const char *name);
228 static void mirror_destroy(struct mirror *);
229 static void mirror_reconfigure(struct bridge *);
230 static void mirror_reconfigure_one(struct mirror *);
231 static bool vlan_is_mirrored(const struct mirror *, int vlan);
233 static void brstp_reconfigure(struct bridge *);
234 static void brstp_adjust_timers(struct bridge *);
235 static void brstp_run(struct bridge *);
236 static void brstp_wait(struct bridge *);
238 static void iface_create(struct port *, const char *name);
239 static void iface_destroy(struct iface *);
240 static struct iface *iface_lookup(const struct bridge *, const char *name);
241 static struct iface *iface_from_dp_ifidx(const struct bridge *,
244 /* Hooks into ofproto processing. */
245 static struct ofhooks bridge_ofhooks;
247 /* Public functions. */
249 /* Adds the name of each interface used by a bridge, including local and
250 * internal ports, to 'svec'. */
252 bridge_get_ifaces(struct svec *svec)
254 struct bridge *br, *next;
257 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
258 for (i = 0; i < br->n_ports; i++) {
259 struct port *port = br->ports[i];
261 for (j = 0; j < port->n_ifaces; j++) {
262 struct iface *iface = port->ifaces[j];
263 if (iface->dp_ifidx < 0) {
264 VLOG_ERR("%s interface not in datapath %s, ignoring",
265 iface->name, dpif_name(br->dpif));
267 if (iface->dp_ifidx != ODPP_LOCAL) {
268 svec_add(svec, iface->name);
276 /* The caller must already have called cfg_read(). */
280 struct svec dpif_names;
283 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
285 dp_enumerate(&dpif_names);
286 for (i = 0; i < dpif_names.n; i++) {
287 const char *dpif_name = dpif_names.names[i];
291 retval = dpif_open(dpif_name, &dpif);
293 struct svec all_names;
296 svec_init(&all_names);
297 dpif_get_all_names(dpif, &all_names);
298 for (j = 0; j < all_names.n; j++) {
299 if (cfg_has("bridge.%s.port", all_names.names[j])) {
305 svec_destroy(&all_names);
311 bridge_reconfigure();
316 config_string_change(const char *key, char **valuep)
318 const char *value = cfg_get_string(0, "%s", key);
319 if (value && (!*valuep || strcmp(value, *valuep))) {
321 *valuep = xstrdup(value);
329 bridge_configure_ssl(void)
331 /* XXX SSL should be configurable on a per-bridge basis.
332 * XXX should be possible to de-configure SSL. */
333 static char *private_key_file;
334 static char *certificate_file;
335 static char *cacert_file;
338 if (config_string_change("ssl.private-key", &private_key_file)) {
339 vconn_ssl_set_private_key_file(private_key_file);
342 if (config_string_change("ssl.certificate", &certificate_file)) {
343 vconn_ssl_set_certificate_file(certificate_file);
346 /* We assume that even if the filename hasn't changed, if the CA cert
347 * file has been removed, that we want to move back into
348 * boot-strapping mode. This opens a small security hole, because
349 * the old certificate will still be trusted until vSwitch is
350 * restarted. We may want to address this in vconn's SSL library. */
351 if (config_string_change("ssl.ca-cert", &cacert_file)
352 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
353 vconn_ssl_set_ca_cert_file(cacert_file,
354 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
360 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
363 struct iface **local_ifacep = local_ifacep_;
365 if (iface->dp_ifidx >= 0) {
366 if (iface->dp_ifidx == ODPP_LOCAL) {
367 *local_ifacep = iface;
369 VLOG_DBG("%s has interface %s on port %d",
371 iface->name, iface->dp_ifidx);
374 VLOG_ERR("%s interface not in %s, dropping",
375 iface->name, dpif_name(br->dpif));
380 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
381 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
382 * deletes from 'br' any ports that no longer have any interfaces. */
384 iterate_and_prune_ifaces(struct bridge *br,
385 bool (*cb)(struct bridge *, struct iface *,
391 for (i = 0; i < br->n_ports; ) {
392 struct port *port = br->ports[i];
393 for (j = 0; j < port->n_ifaces; ) {
394 struct iface *iface = port->ifaces[j];
395 if (cb(br, iface, aux)) {
398 iface_destroy(iface);
402 if (port->n_ifaces) {
405 VLOG_ERR("%s port has no interfaces, dropping", port->name);
412 bridge_reconfigure(void)
414 struct svec old_br, new_br;
415 struct bridge *br, *next;
418 COVERAGE_INC(bridge_reconfigure);
420 /* Collect old and new bridges. */
423 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
424 svec_add(&old_br, br->name);
426 cfg_get_subsections(&new_br, "bridge");
428 /* Get rid of deleted bridges and add new bridges. */
431 assert(svec_is_unique(&old_br));
432 assert(svec_is_unique(&new_br));
433 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
434 if (!svec_contains(&new_br, br->name)) {
438 for (i = 0; i < new_br.n; i++) {
439 const char *name = new_br.names[i];
440 if (!svec_contains(&old_br, name)) {
444 svec_destroy(&old_br);
445 svec_destroy(&new_br);
449 bridge_configure_ssl();
452 /* Reconfigure all bridges. */
453 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
454 bridge_reconfigure_one(br);
457 /* Add and delete ports on all datapaths.
459 * The kernel will reject any attempt to add a given port to a datapath if
460 * that port already belongs to a different datapath, so we must do all
461 * port deletions before any port additions. */
462 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
463 struct odp_port *dpif_ports;
465 struct svec want_ifaces;
467 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
468 bridge_get_all_ifaces(br, &want_ifaces);
469 for (i = 0; i < n_dpif_ports; i++) {
470 const struct odp_port *p = &dpif_ports[i];
471 if (!svec_contains(&want_ifaces, p->devname)
472 && strcmp(p->devname, br->name)) {
473 int retval = dpif_port_del(br->dpif, p->port);
475 VLOG_ERR("failed to remove %s interface from %s: %s",
476 p->devname, dpif_name(br->dpif),
481 svec_destroy(&want_ifaces);
484 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
485 struct odp_port *dpif_ports;
487 struct svec cur_ifaces, want_ifaces, add_ifaces;
489 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
490 svec_init(&cur_ifaces);
491 for (i = 0; i < n_dpif_ports; i++) {
492 svec_add(&cur_ifaces, dpif_ports[i].devname);
495 svec_sort_unique(&cur_ifaces);
496 bridge_get_all_ifaces(br, &want_ifaces);
497 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
499 for (i = 0; i < add_ifaces.n; i++) {
500 const char *if_name = add_ifaces.names[i];
501 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
502 int flags = internal ? ODP_PORT_INTERNAL : 0;
503 int error = dpif_port_add(br->dpif, if_name, flags, NULL);
504 if (error == EXFULL) {
505 VLOG_ERR("ran out of valid port numbers on %s",
506 dpif_name(br->dpif));
509 VLOG_ERR("failed to add %s interface to %s: %s",
510 if_name, dpif_name(br->dpif), strerror(error));
513 svec_destroy(&cur_ifaces);
514 svec_destroy(&want_ifaces);
515 svec_destroy(&add_ifaces);
517 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
520 struct iface *local_iface = NULL;
522 uint8_t engine_type, engine_id;
523 bool add_id_to_iface = false;
524 struct svec nf_hosts;
526 bridge_fetch_dp_ifaces(br);
529 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, &local_iface);
531 /* Pick local port hardware address, datapath ID. */
532 bridge_pick_local_hw_addr(br, ea, &devname);
534 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
536 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
537 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
538 "Ethernet address: %s",
539 br->name, strerror(error));
543 dpid = bridge_pick_datapath_id(br, ea, devname);
544 ofproto_set_datapath_id(br->ofproto, dpid);
546 /* Set NetFlow configuration on this bridge. */
547 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
548 if (cfg_has("netflow.%s.engine-type", br->name)) {
549 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
552 if (cfg_has("netflow.%s.engine-id", br->name)) {
553 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
555 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
556 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
559 if (add_id_to_iface && engine_id > 0x7f) {
560 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
561 "another vswitch, choose an engine id less than 128",
564 if (add_id_to_iface && br->n_ports > 0x1ff) {
565 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
566 "another port when 512 or more ports are used",
569 svec_init(&nf_hosts);
570 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
571 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
572 engine_id, add_id_to_iface)) {
573 VLOG_ERR("bridge %s: problem setting netflow collectors",
577 /* Update the controller and related settings. It would be more
578 * straightforward to call this from bridge_reconfigure_one(), but we
579 * can't do it there for two reasons. First, and most importantly, at
580 * that point we don't know the dp_ifidx of any interfaces that have
581 * been added to the bridge (because we haven't actually added them to
582 * the datapath). Second, at that point we haven't set the datapath ID
583 * yet; when a controller is configured, resetting the datapath ID will
584 * immediately disconnect from the controller, so it's better to set
585 * the datapath ID before the controller. */
586 bridge_reconfigure_controller(br);
588 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
589 for (i = 0; i < br->n_ports; i++) {
590 struct port *port = br->ports[i];
591 port_update_vlan_compat(port);
594 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
595 brstp_reconfigure(br);
600 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
601 const char **devname)
603 uint64_t requested_ea;
609 /* Did the user request a particular MAC? */
610 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
612 eth_addr_from_uint64(requested_ea, ea);
613 if (eth_addr_is_multicast(ea)) {
614 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
615 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
616 } else if (eth_addr_is_zero(ea)) {
617 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
623 /* Otherwise choose the minimum MAC address among all of the interfaces.
624 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
625 * MAC of the physical interface in such an environment.) */
626 memset(ea, 0xff, sizeof ea);
627 for (i = 0; i < br->n_ports; i++) {
628 struct port *port = br->ports[i];
629 if (port->is_mirror_output_port) {
632 for (j = 0; j < port->n_ifaces; j++) {
633 struct iface *iface = port->ifaces[j];
634 uint8_t iface_ea[ETH_ADDR_LEN];
635 if (iface->dp_ifidx == ODPP_LOCAL
636 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
639 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
641 if (!eth_addr_is_multicast(iface_ea) &&
642 !eth_addr_is_reserved(iface_ea) &&
643 !eth_addr_is_zero(iface_ea) &&
644 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
645 memcpy(ea, iface_ea, ETH_ADDR_LEN);
646 *devname = iface->name;
649 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
650 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
651 iface->name, strerror(error));
655 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
656 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
658 VLOG_WARN("bridge %s: using default bridge Ethernet "
659 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
661 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
662 br->name, ETH_ADDR_ARGS(ea));
666 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
667 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
668 * a network device, then that network device's name must be passed in as
669 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
670 * passed in as a null pointer. */
672 bridge_pick_datapath_id(struct bridge *br,
673 const uint8_t bridge_ea[ETH_ADDR_LEN],
677 * The procedure for choosing a bridge MAC address will, in the most
678 * ordinary case, also choose a unique MAC that we can use as a datapath
679 * ID. In some special cases, though, multiple bridges will end up with
680 * the same MAC address. This is OK for the bridges, but it will confuse
681 * the OpenFlow controller, because each datapath needs a unique datapath
684 * Datapath IDs must be unique. It is also very desirable that they be
685 * stable from one run to the next, so that policy set on a datapath
690 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
697 if (!netdev_get_vlan_vid(devname, &vlan)) {
699 * A bridge whose MAC address is taken from a VLAN network device
700 * (that is, a network device created with vconfig(8) or similar
701 * tool) will have the same MAC address as a bridge on the VLAN
702 * device's physical network device.
704 * Handle this case by hashing the physical network device MAC
705 * along with the VLAN identifier.
707 uint8_t buf[ETH_ADDR_LEN + 2];
708 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
709 buf[ETH_ADDR_LEN] = vlan >> 8;
710 buf[ETH_ADDR_LEN + 1] = vlan;
711 return dpid_from_hash(buf, sizeof buf);
714 * Assume that this bridge's MAC address is unique, since it
715 * doesn't fit any of the cases we handle specially.
720 * A purely internal bridge, that is, one that has no non-virtual
721 * network devices on it at all, is more difficult because it has no
722 * natural unique identifier at all.
724 * When the host is a XenServer, we handle this case by hashing the
725 * host's UUID with the name of the bridge. Names of bridges are
726 * persistent across XenServer reboots, although they can be reused if
727 * an internal network is destroyed and then a new one is later
728 * created, so this is fairly effective.
730 * When the host is not a XenServer, we punt by using a random MAC
731 * address on each run.
733 const char *host_uuid = xenserver_get_host_uuid();
735 char *combined = xasprintf("%s,%s", host_uuid, br->name);
736 dpid = dpid_from_hash(combined, strlen(combined));
742 return eth_addr_to_uint64(bridge_ea);
746 dpid_from_hash(const void *data, size_t n)
748 uint8_t hash[SHA1_DIGEST_SIZE];
750 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
751 sha1_bytes(data, n, hash);
752 eth_addr_mark_random(hash);
753 return eth_addr_to_uint64(hash);
759 struct bridge *br, *next;
763 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
764 int error = bridge_run_one(br);
766 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
767 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
768 "forcing reconfiguration", br->name);
782 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
783 ofproto_wait(br->ofproto);
784 if (br->controller) {
789 mac_learning_wait(br->ml);
796 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
797 * configuration changes. */
799 bridge_flush(struct bridge *br)
801 COVERAGE_INC(bridge_flush);
804 mac_learning_flush(br->ml);
808 /* Bridge unixctl user interface functions. */
810 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
812 struct ds ds = DS_EMPTY_INITIALIZER;
813 const struct bridge *br;
815 br = bridge_lookup(args);
817 unixctl_command_reply(conn, 501, "no such bridge");
821 ds_put_cstr(&ds, " port VLAN MAC Age\n");
823 const struct mac_entry *e;
824 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
825 if (e->port < 0 || e->port >= br->n_ports) {
828 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
829 br->ports[e->port]->ifaces[0]->dp_ifidx,
830 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
833 unixctl_command_reply(conn, 200, ds_cstr(&ds));
837 /* Bridge reconfiguration functions. */
839 static struct bridge *
840 bridge_create(const char *name)
845 assert(!bridge_lookup(name));
846 br = xcalloc(1, sizeof *br);
848 error = dpif_create(name, &br->dpif);
849 if (error == EEXIST || error == EBUSY) {
850 error = dpif_open(name, &br->dpif);
852 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
853 name, strerror(error));
857 dpif_flow_flush(br->dpif);
859 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
864 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
866 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
867 dpif_delete(br->dpif);
868 dpif_close(br->dpif);
873 br->name = xstrdup(name);
874 br->ml = mac_learning_create();
875 br->sent_config_request = false;
876 eth_addr_random(br->default_ea);
878 port_array_init(&br->ifaces);
881 br->bond_next_rebalance = time_msec() + 10000;
883 list_push_back(&all_bridges, &br->node);
885 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
891 bridge_destroy(struct bridge *br)
896 while (br->n_ports > 0) {
897 port_destroy(br->ports[br->n_ports - 1]);
899 list_remove(&br->node);
900 error = dpif_delete(br->dpif);
901 if (error && error != ENOENT) {
902 VLOG_ERR("failed to delete %s: %s",
903 dpif_name(br->dpif), strerror(error));
905 dpif_close(br->dpif);
906 ofproto_destroy(br->ofproto);
907 free(br->controller);
908 mac_learning_destroy(br->ml);
909 port_array_destroy(&br->ifaces);
916 static struct bridge *
917 bridge_lookup(const char *name)
921 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
922 if (!strcmp(br->name, name)) {
930 bridge_exists(const char *name)
932 return bridge_lookup(name) ? true : false;
936 bridge_get_datapathid(const char *name)
938 struct bridge *br = bridge_lookup(name);
939 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
943 bridge_run_one(struct bridge *br)
947 error = ofproto_run1(br->ofproto);
953 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
958 error = ofproto_run2(br->ofproto, br->flush);
965 bridge_get_controller(const struct bridge *br)
967 const char *controller;
969 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
971 controller = cfg_get_string(0, "mgmt.controller");
973 return controller && controller[0] ? controller : NULL;
977 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
979 struct svec *ifaces = ifaces_;
980 if (!svec_contains(ifaces, iface->name)) {
981 svec_add(ifaces, iface->name);
985 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
987 br->name, iface->name, iface->port->name);
993 bridge_reconfigure_one(struct bridge *br)
995 struct svec old_ports, new_ports, ifaces;
996 struct svec listeners, old_listeners;
997 struct svec snoops, old_snoops;
1000 /* Collect old ports. */
1001 svec_init(&old_ports);
1002 for (i = 0; i < br->n_ports; i++) {
1003 svec_add(&old_ports, br->ports[i]->name);
1005 svec_sort(&old_ports);
1006 assert(svec_is_unique(&old_ports));
1008 /* Collect new ports. */
1009 svec_init(&new_ports);
1010 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1011 svec_sort(&new_ports);
1012 if (bridge_get_controller(br)) {
1013 char local_name[IF_NAMESIZE];
1016 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1017 local_name, sizeof local_name);
1018 if (!error && !svec_contains(&new_ports, local_name)) {
1019 svec_add(&new_ports, local_name);
1020 svec_sort(&new_ports);
1023 if (!svec_is_unique(&new_ports)) {
1024 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1025 br->name, svec_get_duplicate(&new_ports));
1026 svec_unique(&new_ports);
1029 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1031 /* Get rid of deleted ports and add new ports. */
1032 for (i = 0; i < br->n_ports; ) {
1033 struct port *port = br->ports[i];
1034 if (!svec_contains(&new_ports, port->name)) {
1040 for (i = 0; i < new_ports.n; i++) {
1041 const char *name = new_ports.names[i];
1042 if (!svec_contains(&old_ports, name)) {
1043 port_create(br, name);
1046 svec_destroy(&old_ports);
1047 svec_destroy(&new_ports);
1049 /* Reconfigure all ports. */
1050 for (i = 0; i < br->n_ports; i++) {
1051 port_reconfigure(br->ports[i]);
1054 /* Check and delete duplicate interfaces. */
1056 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1057 svec_destroy(&ifaces);
1059 /* Delete all flows if we're switching from connected to standalone or vice
1060 * versa. (XXX Should we delete all flows if we are switching from one
1061 * controller to another?) */
1063 /* Configure OpenFlow management listeners. */
1064 svec_init(&listeners);
1065 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1067 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1068 ovs_rundir, br->name));
1069 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1070 svec_clear(&listeners);
1072 svec_sort_unique(&listeners);
1074 svec_init(&old_listeners);
1075 ofproto_get_listeners(br->ofproto, &old_listeners);
1076 svec_sort_unique(&old_listeners);
1078 if (!svec_equal(&listeners, &old_listeners)) {
1079 ofproto_set_listeners(br->ofproto, &listeners);
1081 svec_destroy(&listeners);
1082 svec_destroy(&old_listeners);
1084 /* Configure OpenFlow controller connection snooping. */
1086 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1088 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1089 ovs_rundir, br->name));
1090 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1091 svec_clear(&snoops);
1093 svec_sort_unique(&snoops);
1095 svec_init(&old_snoops);
1096 ofproto_get_snoops(br->ofproto, &old_snoops);
1097 svec_sort_unique(&old_snoops);
1099 if (!svec_equal(&snoops, &old_snoops)) {
1100 ofproto_set_snoops(br->ofproto, &snoops);
1102 svec_destroy(&snoops);
1103 svec_destroy(&old_snoops);
1105 mirror_reconfigure(br);
1109 bridge_reconfigure_controller(struct bridge *br)
1111 char *pfx = xasprintf("bridge.%s.controller", br->name);
1112 const char *controller;
1114 controller = bridge_get_controller(br);
1115 if ((br->controller != NULL) != (controller != NULL)) {
1116 ofproto_flush_flows(br->ofproto);
1118 free(br->controller);
1119 br->controller = controller ? xstrdup(controller) : NULL;
1122 const char *fail_mode;
1123 int max_backoff, probe;
1124 int rate_limit, burst_limit;
1126 if (!strcmp(controller, "discover")) {
1127 bool update_resolv_conf = true;
1129 if (cfg_has("%s.update-resolv.conf", pfx)) {
1130 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1133 ofproto_set_discovery(br->ofproto, true,
1134 cfg_get_string(0, "%s.accept-regex", pfx),
1135 update_resolv_conf);
1137 char local_name[IF_NAMESIZE];
1138 struct netdev *netdev;
1142 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1144 || cfg_get_bool(0, "%s.in-band", pfx));
1145 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1146 ofproto_set_in_band(br->ofproto, in_band);
1148 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1149 local_name, sizeof local_name);
1151 error = netdev_open(local_name, NETDEV_ETH_TYPE_NONE, &netdev);
1154 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1155 struct in_addr ip, mask, gateway;
1156 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1157 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1158 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1160 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1162 mask.s_addr = guess_netmask(ip.s_addr);
1164 if (!netdev_set_in4(netdev, ip, mask)) {
1165 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1167 br->name, IP_ARGS(&ip.s_addr),
1168 IP_ARGS(&mask.s_addr));
1171 if (gateway.s_addr) {
1172 if (!netdev_add_router(netdev, gateway)) {
1173 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1174 br->name, IP_ARGS(&gateway.s_addr));
1178 netdev_close(netdev);
1182 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1184 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1186 ofproto_set_failure(br->ofproto,
1188 || !strcmp(fail_mode, "standalone")
1189 || !strcmp(fail_mode, "open")));
1191 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1193 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1198 ofproto_set_probe_interval(br->ofproto, probe);
1200 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1202 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1207 ofproto_set_max_backoff(br->ofproto, max_backoff);
1209 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1211 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1213 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1215 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1217 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1219 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1221 if (cfg_has("%s.commands.acl", pfx)) {
1222 struct svec command_acls;
1225 svec_init(&command_acls);
1226 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1227 command_acl = svec_join(&command_acls, ",", "");
1229 ofproto_set_remote_execution(br->ofproto, command_acl,
1230 cfg_get_string(0, "%s.commands.dir",
1233 svec_destroy(&command_acls);
1236 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1239 union ofp_action action;
1242 /* Set up a flow that matches every packet and directs them to
1243 * OFPP_NORMAL (which goes to us). */
1244 memset(&action, 0, sizeof action);
1245 action.type = htons(OFPAT_OUTPUT);
1246 action.output.len = htons(sizeof action);
1247 action.output.port = htons(OFPP_NORMAL);
1248 memset(&flow, 0, sizeof flow);
1249 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1252 ofproto_set_in_band(br->ofproto, false);
1253 ofproto_set_max_backoff(br->ofproto, 1);
1254 ofproto_set_probe_interval(br->ofproto, 5);
1255 ofproto_set_failure(br->ofproto, false);
1256 ofproto_set_stp(br->ofproto, false);
1260 ofproto_set_controller(br->ofproto, br->controller);
1264 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1269 for (i = 0; i < br->n_ports; i++) {
1270 struct port *port = br->ports[i];
1271 for (j = 0; j < port->n_ifaces; j++) {
1272 struct iface *iface = port->ifaces[j];
1273 svec_add(ifaces, iface->name);
1277 assert(svec_is_unique(ifaces));
1280 /* For robustness, in case the administrator moves around datapath ports behind
1281 * our back, we re-check all the datapath port numbers here.
1283 * This function will set the 'dp_ifidx' members of interfaces that have
1284 * disappeared to -1, so only call this function from a context where those
1285 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1286 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1287 * datapath, which doesn't support UINT16_MAX+1 ports. */
1289 bridge_fetch_dp_ifaces(struct bridge *br)
1291 struct odp_port *dpif_ports;
1292 size_t n_dpif_ports;
1295 /* Reset all interface numbers. */
1296 for (i = 0; i < br->n_ports; i++) {
1297 struct port *port = br->ports[i];
1298 for (j = 0; j < port->n_ifaces; j++) {
1299 struct iface *iface = port->ifaces[j];
1300 iface->dp_ifidx = -1;
1303 port_array_clear(&br->ifaces);
1305 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1306 for (i = 0; i < n_dpif_ports; i++) {
1307 struct odp_port *p = &dpif_ports[i];
1308 struct iface *iface = iface_lookup(br, p->devname);
1310 if (iface->dp_ifidx >= 0) {
1311 VLOG_WARN("%s reported interface %s twice",
1312 dpif_name(br->dpif), p->devname);
1313 } else if (iface_from_dp_ifidx(br, p->port)) {
1314 VLOG_WARN("%s reported interface %"PRIu16" twice",
1315 dpif_name(br->dpif), p->port);
1317 port_array_set(&br->ifaces, p->port, iface);
1318 iface->dp_ifidx = p->port;
1325 /* Bridge packet processing functions. */
1328 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1330 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1333 static struct bond_entry *
1334 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1336 return &port->bond_hash[bond_hash(mac)];
1340 bond_choose_iface(const struct port *port)
1343 for (i = 0; i < port->n_ifaces; i++) {
1344 if (port->ifaces[i]->enabled) {
1352 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1353 uint16_t *dp_ifidx, tag_type *tags)
1355 struct iface *iface;
1357 assert(port->n_ifaces);
1358 if (port->n_ifaces == 1) {
1359 iface = port->ifaces[0];
1361 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1362 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1363 || !port->ifaces[e->iface_idx]->enabled) {
1364 /* XXX select interface properly. The current interface selection
1365 * is only good for testing the rebalancing code. */
1366 e->iface_idx = bond_choose_iface(port);
1367 if (e->iface_idx < 0) {
1368 *tags |= port->no_ifaces_tag;
1371 e->iface_tag = tag_create_random();
1373 *tags |= e->iface_tag;
1374 iface = port->ifaces[e->iface_idx];
1376 *dp_ifidx = iface->dp_ifidx;
1377 *tags |= iface->tag; /* Currently only used for bonding. */
1382 bond_link_status_update(struct iface *iface, bool carrier)
1384 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1385 struct port *port = iface->port;
1387 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1388 /* Nothing to do. */
1391 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1392 iface->name, carrier ? "detected" : "dropped");
1393 if (carrier == iface->enabled) {
1394 iface->delay_expires = LLONG_MAX;
1395 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1396 iface->name, carrier ? "disabled" : "enabled");
1397 } else if (carrier && port->updelay && port->active_iface < 0) {
1398 iface->delay_expires = time_msec();
1399 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1400 "other interface is up", iface->name, port->updelay);
1402 int delay = carrier ? port->updelay : port->downdelay;
1403 iface->delay_expires = time_msec() + delay;
1406 "interface %s: will be %s if it stays %s for %d ms",
1408 carrier ? "enabled" : "disabled",
1409 carrier ? "up" : "down",
1416 bond_choose_active_iface(struct port *port)
1418 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1420 port->active_iface = bond_choose_iface(port);
1421 port->active_iface_tag = tag_create_random();
1422 if (port->active_iface >= 0) {
1423 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1424 port->name, port->ifaces[port->active_iface]->name);
1426 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1432 bond_enable_slave(struct iface *iface, bool enable)
1434 struct port *port = iface->port;
1435 struct bridge *br = port->bridge;
1437 iface->delay_expires = LLONG_MAX;
1438 if (enable == iface->enabled) {
1442 iface->enabled = enable;
1443 if (!iface->enabled) {
1444 VLOG_WARN("interface %s: disabled", iface->name);
1445 ofproto_revalidate(br->ofproto, iface->tag);
1446 if (iface->port_ifidx == port->active_iface) {
1447 ofproto_revalidate(br->ofproto,
1448 port->active_iface_tag);
1449 bond_choose_active_iface(port);
1451 bond_send_learning_packets(port);
1453 VLOG_WARN("interface %s: enabled", iface->name);
1454 if (port->active_iface < 0) {
1455 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1456 bond_choose_active_iface(port);
1457 bond_send_learning_packets(port);
1459 iface->tag = tag_create_random();
1464 bond_run(struct bridge *br)
1468 for (i = 0; i < br->n_ports; i++) {
1469 struct port *port = br->ports[i];
1470 if (port->n_ifaces < 2) {
1473 for (j = 0; j < port->n_ifaces; j++) {
1474 struct iface *iface = port->ifaces[j];
1475 if (time_msec() >= iface->delay_expires) {
1476 bond_enable_slave(iface, !iface->enabled);
1483 bond_wait(struct bridge *br)
1487 for (i = 0; i < br->n_ports; i++) {
1488 struct port *port = br->ports[i];
1489 if (port->n_ifaces < 2) {
1492 for (j = 0; j < port->n_ifaces; j++) {
1493 struct iface *iface = port->ifaces[j];
1494 if (iface->delay_expires != LLONG_MAX) {
1495 poll_timer_wait(iface->delay_expires - time_msec());
1502 set_dst(struct dst *p, const flow_t *flow,
1503 const struct port *in_port, const struct port *out_port,
1508 * XXX This uses too many tags: any broadcast flow will get one tag per
1509 * destination port, and thus a broadcast on a switch of any size is likely
1510 * to have all tag bits set. We should figure out a way to be smarter.
1512 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1513 *tags |= out_port->stp_state_tag;
1514 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1518 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1519 : in_port->vlan >= 0 ? in_port->vlan
1520 : ntohs(flow->dl_vlan));
1521 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1525 swap_dst(struct dst *p, struct dst *q)
1527 struct dst tmp = *p;
1532 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1533 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1534 * that we push to the datapath. We could in fact fully sort the array by
1535 * vlan, but in most cases there are at most two different vlan tags so that's
1536 * possibly overkill.) */
1538 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1540 struct dst *first = dsts;
1541 struct dst *last = dsts + n_dsts;
1543 while (first != last) {
1545 * - All dsts < first have vlan == 'vlan'.
1546 * - All dsts >= last have vlan != 'vlan'.
1547 * - first < last. */
1548 while (first->vlan == vlan) {
1549 if (++first == last) {
1554 /* Same invariants, plus one additional:
1555 * - first->vlan != vlan.
1557 while (last[-1].vlan != vlan) {
1558 if (--last == first) {
1563 /* Same invariants, plus one additional:
1564 * - last[-1].vlan == vlan.*/
1565 swap_dst(first++, --last);
1570 mirror_mask_ffs(mirror_mask_t mask)
1572 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1577 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1578 const struct dst *test)
1581 for (i = 0; i < n_dsts; i++) {
1582 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1590 port_trunks_vlan(const struct port *port, uint16_t vlan)
1592 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1596 port_includes_vlan(const struct port *port, uint16_t vlan)
1598 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1602 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1603 const struct port *in_port, const struct port *out_port,
1604 struct dst dsts[], tag_type *tags)
1606 mirror_mask_t mirrors = in_port->src_mirrors;
1607 struct dst *dst = dsts;
1610 *tags |= in_port->stp_state_tag;
1611 if (out_port == FLOOD_PORT) {
1612 /* XXX use ODP_FLOOD if no vlans or bonding. */
1613 /* XXX even better, define each VLAN as a datapath port group */
1614 for (i = 0; i < br->n_ports; i++) {
1615 struct port *port = br->ports[i];
1616 if (port != in_port && port_includes_vlan(port, vlan)
1617 && !port->is_mirror_output_port
1618 && set_dst(dst, flow, in_port, port, tags)) {
1619 mirrors |= port->dst_mirrors;
1623 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1624 mirrors |= out_port->dst_mirrors;
1629 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1630 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1632 if (set_dst(dst, flow, in_port, m->out_port, tags)
1633 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1637 for (i = 0; i < br->n_ports; i++) {
1638 struct port *port = br->ports[i];
1639 if (port_includes_vlan(port, m->out_vlan)
1640 && set_dst(dst, flow, in_port, port, tags)
1641 && !dst_is_duplicate(dsts, dst - dsts, dst))
1643 if (port->vlan < 0) {
1644 dst->vlan = m->out_vlan;
1646 if (dst->dp_ifidx == flow->in_port
1647 && dst->vlan == vlan) {
1648 /* Don't send out input port on same VLAN. */
1656 mirrors &= mirrors - 1;
1659 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1664 print_dsts(const struct dst *dsts, size_t n)
1666 for (; n--; dsts++) {
1667 printf(">p%"PRIu16, dsts->dp_ifidx);
1668 if (dsts->vlan != OFP_VLAN_NONE) {
1669 printf("v%"PRIu16, dsts->vlan);
1675 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1676 const struct port *in_port, const struct port *out_port,
1677 tag_type *tags, struct odp_actions *actions)
1679 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1681 const struct dst *p;
1684 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1686 cur_vlan = ntohs(flow->dl_vlan);
1687 for (p = dsts; p < &dsts[n_dsts]; p++) {
1688 union odp_action *a;
1689 if (p->vlan != cur_vlan) {
1690 if (p->vlan == OFP_VLAN_NONE) {
1691 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1693 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1694 a->vlan_vid.vlan_vid = htons(p->vlan);
1698 a = odp_actions_add(actions, ODPAT_OUTPUT);
1699 a->output.port = p->dp_ifidx;
1704 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1706 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1707 return (flow->dl_type == htons(ETH_TYPE_ARP)
1708 && eth_addr_is_broadcast(flow->dl_dst)
1709 && packet->size >= sizeof(struct arp_eth_header)
1710 && arp->ar_op == ARP_OP_REQUEST);
1713 /* If the composed actions may be applied to any packet in the given 'flow',
1714 * returns true. Otherwise, the actions should only be applied to 'packet', or
1715 * not at all, if 'packet' was NULL. */
1717 process_flow(struct bridge *br, const flow_t *flow,
1718 const struct ofpbuf *packet, struct odp_actions *actions,
1721 struct iface *in_iface;
1722 struct port *in_port;
1723 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1726 /* Find the interface and port structure for the received packet. */
1727 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1729 /* No interface? Something fishy... */
1730 if (packet != NULL) {
1731 /* Odd. A few possible reasons here:
1733 * - We deleted an interface but there are still a few packets
1734 * queued up from it.
1736 * - Someone externally added an interface (e.g. with "ovs-dpctl
1737 * add-if") that we don't know about.
1739 * - Packet arrived on the local port but the local port is not
1740 * one of our bridge ports.
1742 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1744 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1745 "interface %"PRIu16, br->name, flow->in_port);
1748 /* Return without adding any actions, to drop packets on this flow. */
1751 in_port = in_iface->port;
1753 /* Figure out what VLAN this packet belongs to.
1755 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1756 * belongs to VLAN 0, so we should treat both cases identically. (In the
1757 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1758 * presumably to allow a priority to be specified. In the latter case, the
1759 * packet does not have any 802.1Q header.) */
1760 vlan = ntohs(flow->dl_vlan);
1761 if (vlan == OFP_VLAN_NONE) {
1764 if (in_port->vlan >= 0) {
1766 /* XXX support double tagging? */
1767 if (packet != NULL) {
1768 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1769 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1770 "packet received on port %s configured with "
1771 "implicit VLAN %"PRIu16,
1772 br->name, ntohs(flow->dl_vlan),
1773 in_port->name, in_port->vlan);
1777 vlan = in_port->vlan;
1779 if (!port_includes_vlan(in_port, vlan)) {
1780 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1781 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1782 "packet received on port %s not configured for "
1784 br->name, vlan, in_port->name, vlan);
1789 /* Drop frames for ports that STP wants entirely killed (both for
1790 * forwarding and for learning). Later, after we do learning, we'll drop
1791 * the frames that STP wants to do learning but not forwarding on. */
1792 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1796 /* Drop frames for reserved multicast addresses. */
1797 if (eth_addr_is_reserved(flow->dl_dst)) {
1801 /* Drop frames on ports reserved for mirroring. */
1802 if (in_port->is_mirror_output_port) {
1803 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1804 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1805 "which is reserved exclusively for mirroring",
1806 br->name, in_port->name);
1810 /* Multicast (and broadcast) packets on bonds need special attention, to
1811 * avoid receiving duplicates. */
1812 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1813 *tags |= in_port->active_iface_tag;
1814 if (in_port->active_iface != in_iface->port_ifidx) {
1815 /* Drop all multicast packets on inactive slaves. */
1818 /* Drop all multicast packets for which we have learned a different
1819 * input port, because we probably sent the packet on one slaves
1820 * and got it back on the active slave. Broadcast ARP replies are
1821 * an exception to this rule: the host has moved to another
1823 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1824 if (src_idx != -1 && src_idx != in_port->port_idx) {
1826 if (!is_bcast_arp_reply(flow, packet)) {
1830 /* No way to know whether it's an ARP reply, because the
1831 * flow entry doesn't include enough information and we
1832 * don't have a packet. Punt. */
1840 out_port = FLOOD_PORT;
1844 /* Learn source MAC (but don't try to learn from revalidation). */
1846 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1847 vlan, in_port->port_idx);
1849 /* The log messages here could actually be useful in debugging,
1850 * so keep the rate limit relatively high. */
1851 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1853 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1854 "on port %s in VLAN %d",
1855 br->name, ETH_ADDR_ARGS(flow->dl_src),
1856 in_port->name, vlan);
1857 ofproto_revalidate(br->ofproto, rev_tag);
1861 /* Determine output port. */
1862 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1864 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1865 out_port = br->ports[out_port_idx];
1869 /* Don't send packets out their input ports. Don't forward frames that STP
1870 * wants us to discard. */
1871 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1876 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1879 * We send out only a single packet, instead of setting up a flow, if the
1880 * packet is an ARP directed to broadcast that arrived on a bonded
1881 * interface. In such a situation ARP requests and replies must be handled
1882 * differently, but OpenFlow unfortunately can't distinguish them.
1884 return (in_port->n_ifaces < 2
1885 || flow->dl_type != htons(ETH_TYPE_ARP)
1886 || !eth_addr_is_broadcast(flow->dl_dst));
1889 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1892 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1893 const struct ofp_phy_port *opp,
1896 struct bridge *br = br_;
1897 struct iface *iface;
1900 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1906 if (reason == OFPPR_DELETE) {
1907 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1908 br->name, iface->name);
1909 iface_destroy(iface);
1910 if (!port->n_ifaces) {
1911 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1912 br->name, port->name);
1918 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1919 if (port->n_ifaces > 1) {
1920 bool up = !(opp->state & OFPPS_LINK_DOWN);
1921 bond_link_status_update(iface, up);
1922 port_update_bond_compat(port);
1928 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1929 struct odp_actions *actions, tag_type *tags, void *br_)
1931 struct bridge *br = br_;
1934 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1935 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1936 brstp_receive(br, flow, payload);
1941 COVERAGE_INC(bridge_process_flow);
1942 return process_flow(br, flow, packet, actions, tags);
1946 bridge_account_flow_ofhook_cb(const flow_t *flow,
1947 const union odp_action *actions,
1948 size_t n_actions, unsigned long long int n_bytes,
1951 struct bridge *br = br_;
1952 const union odp_action *a;
1954 if (!br->has_bonded_ports) {
1958 for (a = actions; a < &actions[n_actions]; a++) {
1959 if (a->type == ODPAT_OUTPUT) {
1960 struct port *port = port_from_dp_ifidx(br, a->output.port);
1961 if (port && port->n_ifaces >= 2) {
1962 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1963 e->tx_bytes += n_bytes;
1970 bridge_account_checkpoint_ofhook_cb(void *br_)
1972 struct bridge *br = br_;
1975 if (!br->has_bonded_ports) {
1979 /* The current ofproto implementation calls this callback at least once a
1980 * second, so this timer implementation is sufficient. */
1981 if (time_msec() < br->bond_next_rebalance) {
1984 br->bond_next_rebalance = time_msec() + 10000;
1986 for (i = 0; i < br->n_ports; i++) {
1987 struct port *port = br->ports[i];
1988 if (port->n_ifaces > 1) {
1989 bond_rebalance_port(port);
1994 static struct ofhooks bridge_ofhooks = {
1995 bridge_port_changed_ofhook_cb,
1996 bridge_normal_ofhook_cb,
1997 bridge_account_flow_ofhook_cb,
1998 bridge_account_checkpoint_ofhook_cb,
2001 /* Bonding functions. */
2003 /* Statistics for a single interface on a bonded port, used for load-based
2004 * bond rebalancing. */
2005 struct slave_balance {
2006 struct iface *iface; /* The interface. */
2007 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2009 /* All the "bond_entry"s that are assigned to this interface, in order of
2010 * increasing tx_bytes. */
2011 struct bond_entry **hashes;
2015 /* Sorts pointers to pointers to bond_entries in ascending order by the
2016 * interface to which they are assigned, and within a single interface in
2017 * ascending order of bytes transmitted. */
2019 compare_bond_entries(const void *a_, const void *b_)
2021 const struct bond_entry *const *ap = a_;
2022 const struct bond_entry *const *bp = b_;
2023 const struct bond_entry *a = *ap;
2024 const struct bond_entry *b = *bp;
2025 if (a->iface_idx != b->iface_idx) {
2026 return a->iface_idx > b->iface_idx ? 1 : -1;
2027 } else if (a->tx_bytes != b->tx_bytes) {
2028 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2034 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2035 * *descending* order by number of bytes transmitted. */
2037 compare_slave_balance(const void *a_, const void *b_)
2039 const struct slave_balance *a = a_;
2040 const struct slave_balance *b = b_;
2041 if (a->iface->enabled != b->iface->enabled) {
2042 return a->iface->enabled ? -1 : 1;
2043 } else if (a->tx_bytes != b->tx_bytes) {
2044 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2051 swap_bals(struct slave_balance *a, struct slave_balance *b)
2053 struct slave_balance tmp = *a;
2058 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2059 * given that 'p' (and only 'p') might be in the wrong location.
2061 * This function invalidates 'p', since it might now be in a different memory
2064 resort_bals(struct slave_balance *p,
2065 struct slave_balance bals[], size_t n_bals)
2068 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2069 swap_bals(p, p - 1);
2071 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2072 swap_bals(p, p + 1);
2078 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2080 if (VLOG_IS_DBG_ENABLED()) {
2081 struct ds ds = DS_EMPTY_INITIALIZER;
2082 const struct slave_balance *b;
2084 for (b = bals; b < bals + n_bals; b++) {
2088 ds_put_char(&ds, ',');
2090 ds_put_format(&ds, " %s %"PRIu64"kB",
2091 b->iface->name, b->tx_bytes / 1024);
2093 if (!b->iface->enabled) {
2094 ds_put_cstr(&ds, " (disabled)");
2096 if (b->n_hashes > 0) {
2097 ds_put_cstr(&ds, " (");
2098 for (i = 0; i < b->n_hashes; i++) {
2099 const struct bond_entry *e = b->hashes[i];
2101 ds_put_cstr(&ds, " + ");
2103 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2104 e - port->bond_hash, e->tx_bytes / 1024);
2106 ds_put_cstr(&ds, ")");
2109 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2114 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2116 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2117 struct bond_entry *hash)
2119 struct port *port = from->iface->port;
2120 uint64_t delta = hash->tx_bytes;
2122 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2123 "from %s to %s (now carrying %"PRIu64"kB and "
2124 "%"PRIu64"kB load, respectively)",
2125 port->name, delta / 1024, hash - port->bond_hash,
2126 from->iface->name, to->iface->name,
2127 (from->tx_bytes - delta) / 1024,
2128 (to->tx_bytes + delta) / 1024);
2130 /* Delete element from from->hashes.
2132 * We don't bother to add the element to to->hashes because not only would
2133 * it require more work, the only purpose it would be to allow that hash to
2134 * be migrated to another slave in this rebalancing run, and there is no
2135 * point in doing that. */
2136 if (from->hashes[0] == hash) {
2139 int i = hash - from->hashes[0];
2140 memmove(from->hashes + i, from->hashes + i + 1,
2141 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2145 /* Shift load away from 'from' to 'to'. */
2146 from->tx_bytes -= delta;
2147 to->tx_bytes += delta;
2149 /* Arrange for flows to be revalidated. */
2150 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2151 hash->iface_idx = to->iface->port_ifidx;
2152 hash->iface_tag = tag_create_random();
2156 bond_rebalance_port(struct port *port)
2158 struct slave_balance bals[DP_MAX_PORTS];
2160 struct bond_entry *hashes[BOND_MASK + 1];
2161 struct slave_balance *b, *from, *to;
2162 struct bond_entry *e;
2165 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2166 * descending order of tx_bytes, so that bals[0] represents the most
2167 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2170 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2171 * array for each slave_balance structure, we sort our local array of
2172 * hashes in order by slave, so that all of the hashes for a given slave
2173 * become contiguous in memory, and then we point each 'hashes' members of
2174 * a slave_balance structure to the start of a contiguous group. */
2175 n_bals = port->n_ifaces;
2176 for (b = bals; b < &bals[n_bals]; b++) {
2177 b->iface = port->ifaces[b - bals];
2182 for (i = 0; i <= BOND_MASK; i++) {
2183 hashes[i] = &port->bond_hash[i];
2185 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2186 for (i = 0; i <= BOND_MASK; i++) {
2188 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2189 b = &bals[e->iface_idx];
2190 b->tx_bytes += e->tx_bytes;
2192 b->hashes = &hashes[i];
2197 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2198 log_bals(bals, n_bals, port);
2200 /* Discard slaves that aren't enabled (which were sorted to the back of the
2201 * array earlier). */
2202 while (!bals[n_bals - 1].iface->enabled) {
2209 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2210 to = &bals[n_bals - 1];
2211 for (from = bals; from < to; ) {
2212 uint64_t overload = from->tx_bytes - to->tx_bytes;
2213 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2214 /* The extra load on 'from' (and all less-loaded slaves), compared
2215 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2216 * it is less than ~1Mbps. No point in rebalancing. */
2218 } else if (from->n_hashes == 1) {
2219 /* 'from' only carries a single MAC hash, so we can't shift any
2220 * load away from it, even though we want to. */
2223 /* 'from' is carrying significantly more load than 'to', and that
2224 * load is split across at least two different hashes. Pick a hash
2225 * to migrate to 'to' (the least-loaded slave), given that doing so
2226 * must not cause 'to''s load to exceed 'from''s load.
2228 * The sort order we use means that we prefer to shift away the
2229 * smallest hashes instead of the biggest ones. There is little
2230 * reason behind this decision; we could use the opposite sort
2231 * order to shift away big hashes ahead of small ones. */
2234 for (i = 0; i < from->n_hashes; i++) {
2235 uint64_t delta = from->hashes[i]->tx_bytes;
2236 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2240 if (i < from->n_hashes) {
2241 bond_shift_load(from, to, from->hashes[i]);
2243 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2244 * point to different slave_balance structures. It is only
2245 * valid to do these two operations in a row at all because we
2246 * know that 'from' will not move past 'to' and vice versa. */
2247 resort_bals(from, bals, n_bals);
2248 resort_bals(to, bals, n_bals);
2255 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2256 * historical data to decay to <1% in 7 rebalancing runs. */
2257 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2263 bond_send_learning_packets(struct port *port)
2265 struct bridge *br = port->bridge;
2266 struct mac_entry *e;
2267 struct ofpbuf packet;
2268 int error, n_packets, n_errors;
2270 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2274 ofpbuf_init(&packet, 128);
2275 error = n_packets = n_errors = 0;
2276 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2277 static const char s[] = "Open vSwitch Bond Failover";
2278 union ofp_action actions[2], *a;
2279 struct eth_header *eth;
2280 struct llc_snap_header *llc_snap;
2286 if (e->port == port->port_idx
2287 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2291 /* Compose packet to send. */
2292 ofpbuf_clear(&packet);
2293 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2294 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2295 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2296 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2298 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2299 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2300 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2302 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2303 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2304 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2305 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2306 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2308 /* Compose actions. */
2309 memset(actions, 0, sizeof actions);
2312 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2313 a->vlan_vid.len = htons(sizeof *a);
2314 a->vlan_vid.vlan_vid = htons(e->vlan);
2317 a->output.type = htons(OFPAT_OUTPUT);
2318 a->output.len = htons(sizeof *a);
2319 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2324 flow_extract(&packet, ODPP_NONE, &flow);
2325 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2332 ofpbuf_uninit(&packet);
2335 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2336 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2337 "packets, last error was: %s",
2338 port->name, n_errors, n_packets, strerror(error));
2340 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2341 port->name, n_packets);
2345 /* Bonding unixctl user interface functions. */
2348 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2350 struct ds ds = DS_EMPTY_INITIALIZER;
2351 const struct bridge *br;
2353 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2355 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2358 for (i = 0; i < br->n_ports; i++) {
2359 const struct port *port = br->ports[i];
2360 if (port->n_ifaces > 1) {
2363 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2364 for (j = 0; j < port->n_ifaces; j++) {
2365 const struct iface *iface = port->ifaces[j];
2367 ds_put_cstr(&ds, ", ");
2369 ds_put_cstr(&ds, iface->name);
2371 ds_put_char(&ds, '\n');
2375 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2379 static struct port *
2380 bond_find(const char *name)
2382 const struct bridge *br;
2384 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2387 for (i = 0; i < br->n_ports; i++) {
2388 struct port *port = br->ports[i];
2389 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2398 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2400 struct ds ds = DS_EMPTY_INITIALIZER;
2401 const struct port *port;
2404 port = bond_find(args);
2406 unixctl_command_reply(conn, 501, "no such bond");
2410 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2411 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2412 ds_put_format(&ds, "next rebalance: %lld ms\n",
2413 port->bridge->bond_next_rebalance - time_msec());
2414 for (j = 0; j < port->n_ifaces; j++) {
2415 const struct iface *iface = port->ifaces[j];
2416 struct bond_entry *be;
2419 ds_put_format(&ds, "slave %s: %s\n",
2420 iface->name, iface->enabled ? "enabled" : "disabled");
2421 if (j == port->active_iface) {
2422 ds_put_cstr(&ds, "\tactive slave\n");
2424 if (iface->delay_expires != LLONG_MAX) {
2425 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2426 iface->enabled ? "downdelay" : "updelay",
2427 iface->delay_expires - time_msec());
2431 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2432 int hash = be - port->bond_hash;
2433 struct mac_entry *me;
2435 if (be->iface_idx != j) {
2439 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2440 hash, be->tx_bytes / 1024);
2443 if (!port->bridge->ml) {
2447 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2448 &port->bridge->ml->lrus) {
2451 if (bond_hash(me->mac) == hash
2452 && me->port != port->port_idx
2453 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2454 && dp_ifidx == iface->dp_ifidx)
2456 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2457 ETH_ADDR_ARGS(me->mac));
2462 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2467 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2469 char *args = (char *) args_;
2470 char *save_ptr = NULL;
2471 char *bond_s, *hash_s, *slave_s;
2472 uint8_t mac[ETH_ADDR_LEN];
2474 struct iface *iface;
2475 struct bond_entry *entry;
2478 bond_s = strtok_r(args, " ", &save_ptr);
2479 hash_s = strtok_r(NULL, " ", &save_ptr);
2480 slave_s = strtok_r(NULL, " ", &save_ptr);
2482 unixctl_command_reply(conn, 501,
2483 "usage: bond/migrate BOND HASH SLAVE");
2487 port = bond_find(bond_s);
2489 unixctl_command_reply(conn, 501, "no such bond");
2493 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2494 == ETH_ADDR_SCAN_COUNT) {
2495 hash = bond_hash(mac);
2496 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2497 hash = atoi(hash_s) & BOND_MASK;
2499 unixctl_command_reply(conn, 501, "bad hash");
2503 iface = port_lookup_iface(port, slave_s);
2505 unixctl_command_reply(conn, 501, "no such slave");
2509 if (!iface->enabled) {
2510 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2514 entry = &port->bond_hash[hash];
2515 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2516 entry->iface_idx = iface->port_ifidx;
2517 entry->iface_tag = tag_create_random();
2518 unixctl_command_reply(conn, 200, "migrated");
2522 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2524 char *args = (char *) args_;
2525 char *save_ptr = NULL;
2526 char *bond_s, *slave_s;
2528 struct iface *iface;
2530 bond_s = strtok_r(args, " ", &save_ptr);
2531 slave_s = strtok_r(NULL, " ", &save_ptr);
2533 unixctl_command_reply(conn, 501,
2534 "usage: bond/set-active-slave BOND SLAVE");
2538 port = bond_find(bond_s);
2540 unixctl_command_reply(conn, 501, "no such bond");
2544 iface = port_lookup_iface(port, slave_s);
2546 unixctl_command_reply(conn, 501, "no such slave");
2550 if (!iface->enabled) {
2551 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2555 if (port->active_iface != iface->port_ifidx) {
2556 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2557 port->active_iface = iface->port_ifidx;
2558 port->active_iface_tag = tag_create_random();
2559 VLOG_INFO("port %s: active interface is now %s",
2560 port->name, iface->name);
2561 bond_send_learning_packets(port);
2562 unixctl_command_reply(conn, 200, "done");
2564 unixctl_command_reply(conn, 200, "no change");
2569 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2571 char *args = (char *) args_;
2572 char *save_ptr = NULL;
2573 char *bond_s, *slave_s;
2575 struct iface *iface;
2577 bond_s = strtok_r(args, " ", &save_ptr);
2578 slave_s = strtok_r(NULL, " ", &save_ptr);
2580 unixctl_command_reply(conn, 501,
2581 "usage: bond/enable/disable-slave BOND SLAVE");
2585 port = bond_find(bond_s);
2587 unixctl_command_reply(conn, 501, "no such bond");
2591 iface = port_lookup_iface(port, slave_s);
2593 unixctl_command_reply(conn, 501, "no such slave");
2597 bond_enable_slave(iface, enable);
2598 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2602 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2604 enable_slave(conn, args, true);
2608 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2610 enable_slave(conn, args, false);
2616 unixctl_command_register("bond/list", bond_unixctl_list);
2617 unixctl_command_register("bond/show", bond_unixctl_show);
2618 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2619 unixctl_command_register("bond/set-active-slave",
2620 bond_unixctl_set_active_slave);
2621 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2622 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2625 /* Port functions. */
2628 port_create(struct bridge *br, const char *name)
2632 port = xcalloc(1, sizeof *port);
2634 port->port_idx = br->n_ports;
2636 port->trunks = NULL;
2637 port->name = xstrdup(name);
2638 port->active_iface = -1;
2639 port->stp_state = STP_DISABLED;
2640 port->stp_state_tag = 0;
2642 if (br->n_ports >= br->allocated_ports) {
2643 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2646 br->ports[br->n_ports++] = port;
2648 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2653 port_reconfigure(struct port *port)
2655 bool bonded = cfg_has_section("bonding.%s", port->name);
2656 struct svec old_ifaces, new_ifaces;
2657 unsigned long *trunks;
2661 /* Collect old and new interfaces. */
2662 svec_init(&old_ifaces);
2663 svec_init(&new_ifaces);
2664 for (i = 0; i < port->n_ifaces; i++) {
2665 svec_add(&old_ifaces, port->ifaces[i]->name);
2667 svec_sort(&old_ifaces);
2669 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2670 if (!new_ifaces.n) {
2671 VLOG_ERR("port %s: no interfaces specified for bonded port",
2673 } else if (new_ifaces.n == 1) {
2674 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2678 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2679 if (port->updelay < 0) {
2682 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2683 if (port->downdelay < 0) {
2684 port->downdelay = 0;
2687 svec_init(&new_ifaces);
2688 svec_add(&new_ifaces, port->name);
2691 /* Get rid of deleted interfaces and add new interfaces. */
2692 for (i = 0; i < port->n_ifaces; i++) {
2693 struct iface *iface = port->ifaces[i];
2694 if (!svec_contains(&new_ifaces, iface->name)) {
2695 iface_destroy(iface);
2700 for (i = 0; i < new_ifaces.n; i++) {
2701 const char *name = new_ifaces.names[i];
2702 if (!svec_contains(&old_ifaces, name)) {
2703 iface_create(port, name);
2709 if (cfg_has("vlan.%s.tag", port->name)) {
2711 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2712 if (vlan >= 0 && vlan <= 4095) {
2713 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2716 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2717 * they even work as-is. But they have not been tested. */
2718 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2722 if (port->vlan != vlan) {
2724 bridge_flush(port->bridge);
2727 /* Get trunked VLANs. */
2730 size_t n_trunks, n_errors;
2733 trunks = bitmap_allocate(4096);
2734 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2736 for (i = 0; i < n_trunks; i++) {
2737 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2739 bitmap_set1(trunks, trunk);
2745 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2746 port->name, n_trunks);
2748 if (n_errors == n_trunks) {
2750 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2753 bitmap_set_multiple(trunks, 0, 4096, 1);
2756 if (cfg_has("vlan.%s.trunks", port->name)) {
2757 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2758 port->name, port->name);
2762 ? port->trunks != NULL
2763 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2764 bridge_flush(port->bridge);
2766 bitmap_free(port->trunks);
2767 port->trunks = trunks;
2769 svec_destroy(&old_ifaces);
2770 svec_destroy(&new_ifaces);
2774 port_destroy(struct port *port)
2777 struct bridge *br = port->bridge;
2781 proc_net_compat_update_vlan(port->name, NULL, 0);
2783 for (i = 0; i < MAX_MIRRORS; i++) {
2784 struct mirror *m = br->mirrors[i];
2785 if (m && m->out_port == port) {
2790 while (port->n_ifaces > 0) {
2791 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2794 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2795 del->port_idx = port->port_idx;
2798 bitmap_free(port->trunks);
2805 static struct port *
2806 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2808 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2809 return iface ? iface->port : NULL;
2812 static struct port *
2813 port_lookup(const struct bridge *br, const char *name)
2817 for (i = 0; i < br->n_ports; i++) {
2818 struct port *port = br->ports[i];
2819 if (!strcmp(port->name, name)) {
2826 static struct iface *
2827 port_lookup_iface(const struct port *port, const char *name)
2831 for (j = 0; j < port->n_ifaces; j++) {
2832 struct iface *iface = port->ifaces[j];
2833 if (!strcmp(iface->name, name)) {
2841 port_update_bonding(struct port *port)
2843 if (port->n_ifaces < 2) {
2844 /* Not a bonded port. */
2845 if (port->bond_hash) {
2846 free(port->bond_hash);
2847 port->bond_hash = NULL;
2848 proc_net_compat_update_bond(port->name, NULL);
2851 if (!port->bond_hash) {
2854 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2855 for (i = 0; i <= BOND_MASK; i++) {
2856 struct bond_entry *e = &port->bond_hash[i];
2860 port->no_ifaces_tag = tag_create_random();
2861 bond_choose_active_iface(port);
2863 port_update_bond_compat(port);
2868 port_update_bond_compat(struct port *port)
2870 struct compat_bond bond;
2873 if (port->n_ifaces < 2) {
2878 bond.updelay = port->updelay;
2879 bond.downdelay = port->downdelay;
2880 bond.n_slaves = port->n_ifaces;
2881 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2882 for (i = 0; i < port->n_ifaces; i++) {
2883 struct iface *iface = port->ifaces[i];
2884 struct compat_bond_slave *slave = &bond.slaves[i];
2885 slave->name = iface->name;
2886 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2887 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2891 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2893 proc_net_compat_update_bond(port->name, &bond);
2898 port_update_vlan_compat(struct port *port)
2900 struct bridge *br = port->bridge;
2901 char *vlandev_name = NULL;
2903 if (port->vlan > 0) {
2904 /* Figure out the name that the VLAN device should actually have, if it
2905 * existed. This takes some work because the VLAN device would not
2906 * have port->name in its name; rather, it would have the trunk port's
2907 * name, and 'port' would be attached to a bridge that also had the
2908 * VLAN device one of its ports. So we need to find a trunk port that
2909 * includes port->vlan.
2911 * There might be more than one candidate. This doesn't happen on
2912 * XenServer, so if it happens we just pick the first choice in
2913 * alphabetical order instead of creating multiple VLAN devices. */
2915 for (i = 0; i < br->n_ports; i++) {
2916 struct port *p = br->ports[i];
2917 if (port_trunks_vlan(p, port->vlan)
2919 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2921 const uint8_t *ea = p->ifaces[0]->mac;
2922 if (!eth_addr_is_multicast(ea) &&
2923 !eth_addr_is_reserved(ea) &&
2924 !eth_addr_is_zero(ea)) {
2925 vlandev_name = p->name;
2930 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2933 /* Interface functions. */
2936 iface_create(struct port *port, const char *name)
2938 struct iface *iface;
2940 iface = xcalloc(1, sizeof *iface);
2942 iface->port_ifidx = port->n_ifaces;
2943 iface->name = xstrdup(name);
2944 iface->dp_ifidx = -1;
2945 iface->tag = tag_create_random();
2946 iface->delay_expires = LLONG_MAX;
2948 netdev_nodev_get_etheraddr(name, iface->mac);
2949 netdev_nodev_get_carrier(name, &iface->enabled);
2951 if (port->n_ifaces >= port->allocated_ifaces) {
2952 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2953 sizeof *port->ifaces);
2955 port->ifaces[port->n_ifaces++] = iface;
2956 if (port->n_ifaces > 1) {
2957 port->bridge->has_bonded_ports = true;
2960 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2962 port_update_bonding(port);
2963 bridge_flush(port->bridge);
2967 iface_destroy(struct iface *iface)
2970 struct port *port = iface->port;
2971 struct bridge *br = port->bridge;
2972 bool del_active = port->active_iface == iface->port_ifidx;
2975 if (iface->dp_ifidx >= 0) {
2976 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
2979 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
2980 del->port_ifidx = iface->port_ifidx;
2986 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2987 bond_choose_active_iface(port);
2988 bond_send_learning_packets(port);
2991 port_update_bonding(port);
2992 bridge_flush(port->bridge);
2996 static struct iface *
2997 iface_lookup(const struct bridge *br, const char *name)
3001 for (i = 0; i < br->n_ports; i++) {
3002 struct port *port = br->ports[i];
3003 for (j = 0; j < port->n_ifaces; j++) {
3004 struct iface *iface = port->ifaces[j];
3005 if (!strcmp(iface->name, name)) {
3013 static struct iface *
3014 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3016 return port_array_get(&br->ifaces, dp_ifidx);
3019 /* Port mirroring. */
3022 mirror_reconfigure(struct bridge *br)
3024 struct svec old_mirrors, new_mirrors;
3027 /* Collect old and new mirrors. */
3028 svec_init(&old_mirrors);
3029 svec_init(&new_mirrors);
3030 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3031 for (i = 0; i < MAX_MIRRORS; i++) {
3032 if (br->mirrors[i]) {
3033 svec_add(&old_mirrors, br->mirrors[i]->name);
3037 /* Get rid of deleted mirrors and add new mirrors. */
3038 svec_sort(&old_mirrors);
3039 assert(svec_is_unique(&old_mirrors));
3040 svec_sort(&new_mirrors);
3041 assert(svec_is_unique(&new_mirrors));
3042 for (i = 0; i < MAX_MIRRORS; i++) {
3043 struct mirror *m = br->mirrors[i];
3044 if (m && !svec_contains(&new_mirrors, m->name)) {
3048 for (i = 0; i < new_mirrors.n; i++) {
3049 const char *name = new_mirrors.names[i];
3050 if (!svec_contains(&old_mirrors, name)) {
3051 mirror_create(br, name);
3054 svec_destroy(&old_mirrors);
3055 svec_destroy(&new_mirrors);
3057 /* Reconfigure all mirrors. */
3058 for (i = 0; i < MAX_MIRRORS; i++) {
3059 if (br->mirrors[i]) {
3060 mirror_reconfigure_one(br->mirrors[i]);
3064 /* Update port reserved status. */
3065 for (i = 0; i < br->n_ports; i++) {
3066 br->ports[i]->is_mirror_output_port = false;
3068 for (i = 0; i < MAX_MIRRORS; i++) {
3069 struct mirror *m = br->mirrors[i];
3070 if (m && m->out_port) {
3071 m->out_port->is_mirror_output_port = true;
3077 mirror_create(struct bridge *br, const char *name)
3082 for (i = 0; ; i++) {
3083 if (i >= MAX_MIRRORS) {
3084 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3085 "cannot create %s", br->name, MAX_MIRRORS, name);
3088 if (!br->mirrors[i]) {
3093 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3096 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3099 m->name = xstrdup(name);
3100 svec_init(&m->src_ports);
3101 svec_init(&m->dst_ports);
3109 mirror_destroy(struct mirror *m)
3112 struct bridge *br = m->bridge;
3115 for (i = 0; i < br->n_ports; i++) {
3116 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3117 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3120 svec_destroy(&m->src_ports);
3121 svec_destroy(&m->dst_ports);
3124 m->bridge->mirrors[m->idx] = NULL;
3132 prune_ports(struct mirror *m, struct svec *ports)
3137 svec_sort_unique(ports);
3140 for (i = 0; i < ports->n; i++) {
3141 const char *name = ports->names[i];
3142 if (port_lookup(m->bridge, name)) {
3143 svec_add(&tmp, name);
3145 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3146 m->bridge->name, m->name, name);
3149 svec_swap(ports, &tmp);
3154 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3158 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3159 * order won't give us numeric sort order. But that's good enough for what
3160 * we need right now. */
3161 svec_sort_unique(vlan_strings);
3163 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3165 for (i = 0; i < vlan_strings->n; i++) {
3166 const char *name = vlan_strings->names[i];
3168 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3169 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3170 m->bridge->name, m->name, name);
3172 (*vlans)[n_vlans++] = vlan;
3179 vlan_is_mirrored(const struct mirror *m, int vlan)
3183 for (i = 0; i < m->n_vlans; i++) {
3184 if (m->vlans[i] == vlan) {
3192 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3196 for (i = 0; i < m->n_vlans; i++) {
3197 if (port_trunks_vlan(p, m->vlans[i])) {
3205 mirror_reconfigure_one(struct mirror *m)
3207 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3208 struct svec src_ports, dst_ports, ports;
3209 struct svec vlan_strings;
3210 mirror_mask_t mirror_bit;
3211 const char *out_port_name;
3212 struct port *out_port;
3217 bool mirror_all_ports;
3219 /* Get output port. */
3220 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3221 m->bridge->name, m->name);
3222 if (out_port_name) {
3223 out_port = port_lookup(m->bridge, out_port_name);
3225 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3226 "named %s", pfx, m->bridge->name, out_port_name);
3233 if (cfg_has("%s.output.vlan", pfx)) {
3234 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3235 "ignoring %s.output.vlan", pfx, pfx, pfx);
3237 } else if (cfg_has("%s.output.vlan", pfx)) {
3239 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3241 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3242 "but exactly one is required; disabling port mirror %s",
3243 pfx, pfx, pfx, pfx);
3249 /* Get all the ports, and drop duplicates and ports that don't exist. */
3250 svec_init(&src_ports);
3251 svec_init(&dst_ports);
3253 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3254 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3255 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3256 svec_append(&src_ports, &ports);
3257 svec_append(&dst_ports, &ports);
3258 svec_destroy(&ports);
3259 prune_ports(m, &src_ports);
3260 prune_ports(m, &dst_ports);
3262 /* Get all the vlans, and drop duplicate and invalid vlans. */
3263 svec_init(&vlan_strings);
3264 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3265 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3266 svec_destroy(&vlan_strings);
3268 /* Update mirror data. */
3269 if (!svec_equal(&m->src_ports, &src_ports)
3270 || !svec_equal(&m->dst_ports, &dst_ports)
3271 || m->n_vlans != n_vlans
3272 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3273 || m->out_port != out_port
3274 || m->out_vlan != out_vlan) {
3275 bridge_flush(m->bridge);
3277 svec_swap(&m->src_ports, &src_ports);
3278 svec_swap(&m->dst_ports, &dst_ports);
3281 m->n_vlans = n_vlans;
3282 m->out_port = out_port;
3283 m->out_vlan = out_vlan;
3285 /* If no selection criteria have been given, mirror for all ports. */
3286 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3289 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3290 for (i = 0; i < m->bridge->n_ports; i++) {
3291 struct port *port = m->bridge->ports[i];
3293 if (mirror_all_ports
3294 || svec_contains(&m->src_ports, port->name)
3297 ? port_trunks_any_mirrored_vlan(m, port)
3298 : vlan_is_mirrored(m, port->vlan)))) {
3299 port->src_mirrors |= mirror_bit;
3301 port->src_mirrors &= ~mirror_bit;
3304 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3305 port->dst_mirrors |= mirror_bit;
3307 port->dst_mirrors &= ~mirror_bit;
3312 svec_destroy(&src_ports);
3313 svec_destroy(&dst_ports);
3317 /* Spanning tree protocol. */
3319 static void brstp_update_port_state(struct port *);
3322 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3324 struct bridge *br = br_;
3325 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3326 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3328 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3330 } else if (eth_addr_is_zero(iface->mac)) {
3331 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3334 union ofp_action action;
3335 struct eth_header *eth = pkt->l2;
3338 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3340 memset(&action, 0, sizeof action);
3341 action.type = htons(OFPAT_OUTPUT);
3342 action.output.len = htons(sizeof action);
3343 action.output.port = htons(port_no);
3345 flow_extract(pkt, ODPP_NONE, &flow);
3346 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3352 brstp_reconfigure(struct bridge *br)
3356 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3358 stp_destroy(br->stp);
3364 uint64_t bridge_address, bridge_id;
3365 int bridge_priority;
3367 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3368 if (!bridge_address) {
3370 bridge_address = (stp_get_bridge_id(br->stp)
3371 & ((UINT64_C(1) << 48) - 1));
3373 uint8_t mac[ETH_ADDR_LEN];
3374 eth_addr_random(mac);
3375 bridge_address = eth_addr_to_uint64(mac);
3379 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3381 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3383 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3386 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3388 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3389 br->stp_last_tick = time_msec();
3392 if (bridge_id != stp_get_bridge_id(br->stp)) {
3393 stp_set_bridge_id(br->stp, bridge_id);
3398 for (i = 0; i < br->n_ports; i++) {
3399 struct port *p = br->ports[i];
3401 struct stp_port *sp;
3402 int path_cost, priority;
3408 dp_ifidx = p->ifaces[0]->dp_ifidx;
3409 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3413 sp = stp_get_port(br->stp, dp_ifidx);
3414 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3415 "stp.%s.port.%s.enabled",
3417 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3418 br->name, p->name));
3419 if (p->is_mirror_output_port) {
3422 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3423 bridge_flush(br); /* Might not be necessary. */
3425 stp_port_enable(sp);
3427 stp_port_disable(sp);
3431 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3433 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3435 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3436 "stp.%s.port.%s.priority",
3438 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3440 : STP_DEFAULT_PORT_PRIORITY);
3441 stp_port_set_priority(sp, priority);
3444 brstp_adjust_timers(br);
3446 for (i = 0; i < br->n_ports; i++) {
3447 brstp_update_port_state(br->ports[i]);
3452 brstp_update_port_state(struct port *p)
3454 struct bridge *br = p->bridge;
3455 enum stp_state state;
3457 /* Figure out new state. */
3458 state = STP_DISABLED;
3459 if (br->stp && p->n_ifaces > 0) {
3460 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3461 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3462 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3467 if (p->stp_state != state) {
3468 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3469 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3470 p->name, stp_state_name(p->stp_state),
3471 stp_state_name(state));
3472 if (p->stp_state == STP_DISABLED) {
3475 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3477 p->stp_state = state;
3478 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3479 : tag_create_random());
3484 brstp_adjust_timers(struct bridge *br)
3486 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3487 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3488 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3490 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3491 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3492 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3496 brstp_run(struct bridge *br)
3499 long long int now = time_msec();
3500 long long int elapsed = now - br->stp_last_tick;
3501 struct stp_port *sp;
3504 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3505 br->stp_last_tick = now;
3507 while (stp_get_changed_port(br->stp, &sp)) {
3508 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3510 brstp_update_port_state(p);
3517 brstp_wait(struct bridge *br)
3520 poll_timer_wait(1000);