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"
47 #include "poll-loop.h"
48 #include "port-array.h"
49 #include "proc-net-compat.h"
51 #include "secchan/ofproto.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; /* Kernel 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 dp%u, ignoring",
265 iface->name, dpif_id(&br->dpif));
267 if (iface->dp_ifidx != ODPP_LOCAL) {
268 svec_add(svec, iface->name);
276 /* The caller must already have called cfg_read(). */
285 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
287 for (i = 0; i < DP_MAX; i++) {
291 sprintf(devname, "dp%d", i);
292 retval = dpif_open(devname, &dpif);
294 char dpif_name[IF_NAMESIZE];
295 if (dpif_get_name(&dpif, dpif_name, sizeof dpif_name)
296 || !cfg_has("bridge.%s.port", dpif_name)) {
300 } else if (retval != ENODEV) {
301 VLOG_ERR("failed to delete datapath dp%d: %s",
302 i, strerror(retval));
306 bridge_reconfigure();
311 config_string_change(const char *key, char **valuep)
313 const char *value = cfg_get_string(0, "%s", key);
314 if (value && (!*valuep || strcmp(value, *valuep))) {
316 *valuep = xstrdup(value);
324 bridge_configure_ssl(void)
326 /* XXX SSL should be configurable on a per-bridge basis.
327 * XXX should be possible to de-configure SSL. */
328 static char *private_key_file;
329 static char *certificate_file;
330 static char *cacert_file;
333 if (config_string_change("ssl.private-key", &private_key_file)) {
334 vconn_ssl_set_private_key_file(private_key_file);
337 if (config_string_change("ssl.certificate", &certificate_file)) {
338 vconn_ssl_set_certificate_file(certificate_file);
341 /* We assume that even if the filename hasn't changed, if the CA cert
342 * file has been removed, that we want to move back into
343 * boot-strapping mode. This opens a small security hole, because
344 * the old certificate will still be trusted until vSwitch is
345 * restarted. We may want to address this in vconn's SSL library. */
346 if (config_string_change("ssl.ca-cert", &cacert_file)
347 || (stat(cacert_file, &s) && errno == ENOENT)) {
348 vconn_ssl_set_ca_cert_file(cacert_file,
349 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
355 bridge_reconfigure(void)
357 struct svec old_br, new_br, raw_new_br;
358 struct bridge *br, *next;
361 COVERAGE_INC(bridge_reconfigure);
363 /* Collect old bridges. */
365 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
366 svec_add(&old_br, br->name);
369 /* Collect new bridges. */
370 svec_init(&raw_new_br);
371 cfg_get_subsections(&raw_new_br, "bridge");
373 for (i = 0; i < raw_new_br.n; i++) {
374 const char *name = raw_new_br.names[i];
375 if ((!strncmp(name, "dp", 2) && isdigit(name[2])) ||
376 (!strncmp(name, "nl:", 3) && isdigit(name[3]))) {
377 VLOG_ERR("%s is not a valid bridge name (bridges may not be "
378 "named \"dp\" or \"nl:\" followed by a digit)", name);
380 svec_add(&new_br, name);
383 svec_destroy(&raw_new_br);
385 /* Get rid of deleted bridges and add new bridges. */
388 assert(svec_is_unique(&old_br));
389 assert(svec_is_unique(&new_br));
390 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
391 if (!svec_contains(&new_br, br->name)) {
395 for (i = 0; i < new_br.n; i++) {
396 const char *name = new_br.names[i];
397 if (!svec_contains(&old_br, name)) {
401 svec_destroy(&old_br);
402 svec_destroy(&new_br);
406 bridge_configure_ssl();
409 /* Reconfigure all bridges. */
410 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
411 bridge_reconfigure_one(br);
414 /* Add and delete ports on all datapaths.
416 * The kernel will reject any attempt to add a given port to a datapath if
417 * that port already belongs to a different datapath, so we must do all
418 * port deletions before any port additions. */
419 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
420 struct odp_port *dpif_ports;
422 struct svec want_ifaces;
424 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
425 bridge_get_all_ifaces(br, &want_ifaces);
426 for (i = 0; i < n_dpif_ports; i++) {
427 const struct odp_port *p = &dpif_ports[i];
428 if (!svec_contains(&want_ifaces, p->devname)
429 && strcmp(p->devname, br->name)) {
430 int retval = dpif_port_del(&br->dpif, p->port);
432 VLOG_ERR("failed to remove %s interface from dp%u: %s",
433 p->devname, dpif_id(&br->dpif), strerror(retval));
437 svec_destroy(&want_ifaces);
440 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
441 struct odp_port *dpif_ports;
443 struct svec cur_ifaces, want_ifaces, add_ifaces;
446 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
447 svec_init(&cur_ifaces);
448 for (i = 0; i < n_dpif_ports; i++) {
449 svec_add(&cur_ifaces, dpif_ports[i].devname);
452 svec_sort_unique(&cur_ifaces);
453 bridge_get_all_ifaces(br, &want_ifaces);
454 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
457 for (i = 0; i < add_ifaces.n; i++) {
458 const char *if_name = add_ifaces.names[i];
460 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
461 int error = dpif_port_add(&br->dpif, if_name, next_port_no++,
462 internal ? ODP_PORT_INTERNAL : 0);
463 if (error != EEXIST) {
464 if (next_port_no >= 256) {
465 VLOG_ERR("ran out of valid port numbers on dp%u",
470 VLOG_ERR("failed to add %s interface to dp%u: %s",
471 if_name, dpif_id(&br->dpif), strerror(error));
478 svec_destroy(&cur_ifaces);
479 svec_destroy(&want_ifaces);
480 svec_destroy(&add_ifaces);
482 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
485 struct iface *local_iface = NULL;
487 uint8_t engine_type = br->dpif.minor;
488 uint8_t engine_id = br->dpif.minor;
489 bool add_id_to_iface = false;
490 struct svec nf_hosts;
492 bridge_fetch_dp_ifaces(br);
493 for (i = 0; i < br->n_ports; ) {
494 struct port *port = br->ports[i];
496 for (j = 0; j < port->n_ifaces; ) {
497 struct iface *iface = port->ifaces[j];
498 if (iface->dp_ifidx < 0) {
499 VLOG_ERR("%s interface not in dp%u, dropping",
500 iface->name, dpif_id(&br->dpif));
501 iface_destroy(iface);
503 if (iface->dp_ifidx == ODPP_LOCAL) {
506 VLOG_DBG("dp%u has interface %s on port %d",
507 dpif_id(&br->dpif), iface->name, iface->dp_ifidx);
511 if (!port->n_ifaces) {
512 VLOG_ERR("%s port has no interfaces, dropping", port->name);
519 /* Pick local port hardware address, datapath ID. */
520 bridge_pick_local_hw_addr(br, ea, &devname);
522 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
524 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
525 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
526 "Ethernet address: %s",
527 br->name, strerror(error));
531 dpid = bridge_pick_datapath_id(br, ea, devname);
532 ofproto_set_datapath_id(br->ofproto, dpid);
534 /* Set NetFlow configuration on this bridge. */
535 if (cfg_has("netflow.%s.engine-type", br->name)) {
536 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
539 if (cfg_has("netflow.%s.engine-id", br->name)) {
540 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
542 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
543 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
546 if (add_id_to_iface && engine_id > 0x7f) {
547 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
548 "another vswitch, choose an engine id less than 128",
551 if (add_id_to_iface && br->n_ports > 0x1ff) {
552 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
553 "another port when 512 or more ports are used",
556 svec_init(&nf_hosts);
557 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
558 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
559 engine_id, add_id_to_iface)) {
560 VLOG_ERR("bridge %s: problem setting netflow collectors",
564 /* Update the controller and related settings. It would be more
565 * straightforward to call this from bridge_reconfigure_one(), but we
566 * can't do it there for two reasons. First, and most importantly, at
567 * that point we don't know the dp_ifidx of any interfaces that have
568 * been added to the bridge (because we haven't actually added them to
569 * the datapath). Second, at that point we haven't set the datapath ID
570 * yet; when a controller is configured, resetting the datapath ID will
571 * immediately disconnect from the controller, so it's better to set
572 * the datapath ID before the controller. */
573 bridge_reconfigure_controller(br);
575 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
576 for (i = 0; i < br->n_ports; i++) {
577 struct port *port = br->ports[i];
578 port_update_vlan_compat(port);
581 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
582 brstp_reconfigure(br);
587 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
588 const char **devname)
590 uint64_t requested_ea;
596 /* Did the user request a particular MAC? */
597 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
599 eth_addr_from_uint64(requested_ea, ea);
600 if (eth_addr_is_multicast(ea)) {
601 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
602 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
603 } else if (eth_addr_is_zero(ea)) {
604 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
610 /* Otherwise choose the minimum MAC address among all of the interfaces.
611 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
612 * MAC of the physical interface in such an environment.) */
613 memset(ea, 0xff, sizeof ea);
614 for (i = 0; i < br->n_ports; i++) {
615 struct port *port = br->ports[i];
616 if (port->is_mirror_output_port) {
619 for (j = 0; j < port->n_ifaces; j++) {
620 struct iface *iface = port->ifaces[j];
621 uint8_t iface_ea[ETH_ADDR_LEN];
622 if (iface->dp_ifidx == ODPP_LOCAL
623 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
626 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
628 if (!eth_addr_is_multicast(iface_ea) &&
629 !eth_addr_is_reserved(iface_ea) &&
630 !eth_addr_is_zero(iface_ea) &&
631 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
632 memcpy(ea, iface_ea, ETH_ADDR_LEN);
633 *devname = iface->name;
636 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
637 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
638 iface->name, strerror(error));
642 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
643 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
645 VLOG_WARN("bridge %s: using default bridge Ethernet "
646 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
648 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
649 br->name, ETH_ADDR_ARGS(ea));
653 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
654 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
655 * a network device, then that network device's name must be passed in as
656 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
657 * passed in as a null pointer. */
659 bridge_pick_datapath_id(struct bridge *br,
660 const uint8_t bridge_ea[ETH_ADDR_LEN],
664 * The procedure for choosing a bridge MAC address will, in the most
665 * ordinary case, also choose a unique MAC that we can use as a datapath
666 * ID. In some special cases, though, multiple bridges will end up with
667 * the same MAC address. This is OK for the bridges, but it will confuse
668 * the OpenFlow controller, because each datapath needs a unique datapath
671 * Datapath IDs must be unique. It is also very desirable that they be
672 * stable from one run to the next, so that policy set on a datapath
677 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
684 if (!netdev_get_vlan_vid(devname, &vlan)) {
686 * A bridge whose MAC address is taken from a VLAN network device
687 * (that is, a network device created with vconfig(8) or similar
688 * tool) will have the same MAC address as a bridge on the VLAN
689 * device's physical network device.
691 * Handle this case by hashing the physical network device MAC
692 * along with the VLAN identifier.
694 uint8_t buf[ETH_ADDR_LEN + 2];
695 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
696 buf[ETH_ADDR_LEN] = vlan >> 8;
697 buf[ETH_ADDR_LEN + 1] = vlan;
698 return dpid_from_hash(buf, sizeof buf);
701 * Assume that this bridge's MAC address is unique, since it
702 * doesn't fit any of the cases we handle specially.
707 * A purely internal bridge, that is, one that has no non-virtual
708 * network devices on it at all, is more difficult because it has no
709 * natural unique identifier at all.
711 * When the host is a XenServer, we handle this case by hashing the
712 * host's UUID with the name of the bridge. Names of bridges are
713 * persistent across XenServer reboots, although they can be reused if
714 * an internal network is destroyed and then a new one is later
715 * created, so this is fairly effective.
717 * When the host is not a XenServer, we punt by using a random MAC
718 * address on each run.
720 const char *host_uuid = xenserver_get_host_uuid();
722 char *combined = xasprintf("%s,%s", host_uuid, br->name);
723 dpid = dpid_from_hash(combined, strlen(combined));
729 return eth_addr_to_uint64(bridge_ea);
733 dpid_from_hash(const void *data, size_t n)
735 uint8_t hash[SHA1_DIGEST_SIZE];
737 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
738 sha1_bytes(data, n, hash);
739 eth_addr_mark_random(hash);
740 return eth_addr_to_uint64(hash);
746 struct bridge *br, *next;
750 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
751 int error = bridge_run_one(br);
753 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
754 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
755 "forcing reconfiguration", br->name);
769 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
770 ofproto_wait(br->ofproto);
771 if (br->controller) {
776 mac_learning_wait(br->ml);
783 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
784 * configuration changes. */
786 bridge_flush(struct bridge *br)
788 COVERAGE_INC(bridge_flush);
791 mac_learning_flush(br->ml);
795 /* Bridge unixctl user interface functions. */
797 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
799 struct ds ds = DS_EMPTY_INITIALIZER;
800 const struct bridge *br;
802 br = bridge_lookup(args);
804 unixctl_command_reply(conn, 501, "no such bridge");
808 ds_put_cstr(&ds, " port VLAN MAC Age\n");
810 const struct mac_entry *e;
811 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
812 if (e->port < 0 || e->port >= br->n_ports) {
815 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
816 br->ports[e->port]->ifaces[0]->dp_ifidx,
817 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
820 unixctl_command_reply(conn, 200, ds_cstr(&ds));
824 /* Bridge reconfiguration functions. */
826 static struct bridge *
827 bridge_create(const char *name)
832 assert(!bridge_lookup(name));
833 br = xcalloc(1, sizeof *br);
835 error = dpif_create(name, &br->dpif);
836 if (error == EEXIST) {
837 error = dpif_open(name, &br->dpif);
839 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
840 name, strerror(error));
844 dpif_flow_flush(&br->dpif);
846 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
851 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
853 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
854 dpif_delete(&br->dpif);
855 dpif_close(&br->dpif);
860 br->name = xstrdup(name);
861 br->ml = mac_learning_create();
862 br->sent_config_request = false;
863 eth_addr_random(br->default_ea);
865 port_array_init(&br->ifaces);
868 br->bond_next_rebalance = time_msec() + 10000;
870 list_push_back(&all_bridges, &br->node);
872 VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif));
878 bridge_destroy(struct bridge *br)
883 while (br->n_ports > 0) {
884 port_destroy(br->ports[br->n_ports - 1]);
886 list_remove(&br->node);
887 error = dpif_delete(&br->dpif);
888 if (error && error != ENOENT) {
889 VLOG_ERR("failed to delete dp%u: %s",
890 dpif_id(&br->dpif), strerror(error));
892 dpif_close(&br->dpif);
893 ofproto_destroy(br->ofproto);
894 free(br->controller);
895 mac_learning_destroy(br->ml);
896 port_array_destroy(&br->ifaces);
903 static struct bridge *
904 bridge_lookup(const char *name)
908 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
909 if (!strcmp(br->name, name)) {
917 bridge_exists(const char *name)
919 return bridge_lookup(name) ? true : false;
923 bridge_get_datapathid(const char *name)
925 struct bridge *br = bridge_lookup(name);
926 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
930 bridge_run_one(struct bridge *br)
934 error = ofproto_run1(br->ofproto);
940 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
945 error = ofproto_run2(br->ofproto, br->flush);
952 bridge_get_controller(const struct bridge *br)
954 const char *controller;
956 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
958 controller = cfg_get_string(0, "mgmt.controller");
960 return controller && controller[0] ? controller : NULL;
964 bridge_reconfigure_one(struct bridge *br)
966 struct svec old_ports, new_ports, ifaces;
967 struct svec listeners, old_listeners;
968 struct svec snoops, old_snoops;
971 /* Collect old ports. */
972 svec_init(&old_ports);
973 for (i = 0; i < br->n_ports; i++) {
974 svec_add(&old_ports, br->ports[i]->name);
976 svec_sort(&old_ports);
977 assert(svec_is_unique(&old_ports));
979 /* Collect new ports. */
980 svec_init(&new_ports);
981 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
982 svec_sort(&new_ports);
983 if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) {
984 svec_add(&new_ports, br->name);
985 svec_sort(&new_ports);
987 if (!svec_is_unique(&new_ports)) {
988 VLOG_WARN("bridge %s: %s specified twice as bridge port",
989 br->name, svec_get_duplicate(&new_ports));
990 svec_unique(&new_ports);
993 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
995 /* Get rid of deleted ports and add new ports. */
996 for (i = 0; i < br->n_ports; ) {
997 struct port *port = br->ports[i];
998 if (!svec_contains(&new_ports, port->name)) {
1004 for (i = 0; i < new_ports.n; i++) {
1005 const char *name = new_ports.names[i];
1006 if (!svec_contains(&old_ports, name)) {
1007 port_create(br, name);
1010 svec_destroy(&old_ports);
1011 svec_destroy(&new_ports);
1013 /* Reconfigure all ports. */
1014 for (i = 0; i < br->n_ports; i++) {
1015 port_reconfigure(br->ports[i]);
1018 /* Check and delete duplicate interfaces. */
1020 for (i = 0; i < br->n_ports; ) {
1021 struct port *port = br->ports[i];
1022 for (j = 0; j < port->n_ifaces; ) {
1023 struct iface *iface = port->ifaces[j];
1024 if (svec_contains(&ifaces, iface->name)) {
1025 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1027 br->name, iface->name, port->name);
1028 iface_destroy(iface);
1030 svec_add(&ifaces, iface->name);
1035 if (!port->n_ifaces) {
1036 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1042 svec_destroy(&ifaces);
1044 /* Delete all flows if we're switching from connected to standalone or vice
1045 * versa. (XXX Should we delete all flows if we are switching from one
1046 * controller to another?) */
1048 /* Configure OpenFlow management listeners. */
1049 svec_init(&listeners);
1050 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1052 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1053 ovs_rundir, br->name));
1054 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1055 svec_clear(&listeners);
1057 svec_sort_unique(&listeners);
1059 svec_init(&old_listeners);
1060 ofproto_get_listeners(br->ofproto, &old_listeners);
1061 svec_sort_unique(&old_listeners);
1063 if (!svec_equal(&listeners, &old_listeners)) {
1064 ofproto_set_listeners(br->ofproto, &listeners);
1066 svec_destroy(&listeners);
1067 svec_destroy(&old_listeners);
1069 /* Configure OpenFlow controller connection snooping. */
1071 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1073 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1074 ovs_rundir, br->name));
1075 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1076 svec_clear(&snoops);
1078 svec_sort_unique(&snoops);
1080 svec_init(&old_snoops);
1081 ofproto_get_snoops(br->ofproto, &old_snoops);
1082 svec_sort_unique(&old_snoops);
1084 if (!svec_equal(&snoops, &old_snoops)) {
1085 ofproto_set_snoops(br->ofproto, &snoops);
1087 svec_destroy(&snoops);
1088 svec_destroy(&old_snoops);
1090 mirror_reconfigure(br);
1094 bridge_reconfigure_controller(struct bridge *br)
1096 char *pfx = xasprintf("bridge.%s.controller", br->name);
1097 const char *controller;
1099 controller = bridge_get_controller(br);
1100 if ((br->controller != NULL) != (controller != NULL)) {
1101 ofproto_flush_flows(br->ofproto);
1103 free(br->controller);
1104 br->controller = controller ? xstrdup(controller) : NULL;
1107 const char *fail_mode;
1108 int max_backoff, probe;
1109 int rate_limit, burst_limit;
1111 if (!strcmp(controller, "discover")) {
1112 bool update_resolv_conf = true;
1114 if (cfg_has("%s.update-resolv.conf", pfx)) {
1115 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1118 ofproto_set_discovery(br->ofproto, true,
1119 cfg_get_string(0, "%s.accept-regex", pfx),
1120 update_resolv_conf);
1122 struct netdev *netdev;
1126 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1128 || cfg_get_bool(0, "%s.in-band", pfx));
1129 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1130 ofproto_set_in_band(br->ofproto, in_band);
1132 error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev);
1134 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1135 struct in_addr ip, mask, gateway;
1136 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1137 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1138 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1140 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1142 mask.s_addr = guess_netmask(ip.s_addr);
1144 if (!netdev_set_in4(netdev, ip, mask)) {
1145 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1147 br->name, IP_ARGS(&ip.s_addr),
1148 IP_ARGS(&mask.s_addr));
1151 if (gateway.s_addr) {
1152 if (!netdev_add_router(gateway)) {
1153 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1154 br->name, IP_ARGS(&gateway.s_addr));
1158 netdev_close(netdev);
1162 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1164 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1166 ofproto_set_failure(br->ofproto,
1168 || !strcmp(fail_mode, "standalone")
1169 || !strcmp(fail_mode, "open")));
1171 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1173 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1178 ofproto_set_probe_interval(br->ofproto, probe);
1180 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1182 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1187 ofproto_set_max_backoff(br->ofproto, max_backoff);
1189 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1191 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1193 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1195 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1197 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1199 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1201 if (cfg_has("%s.commands.acl", pfx)) {
1202 struct svec command_acls;
1205 svec_init(&command_acls);
1206 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1207 command_acl = svec_join(&command_acls, ",", "");
1209 ofproto_set_remote_execution(br->ofproto, command_acl,
1210 cfg_get_string(0, "%s.commands.dir",
1213 svec_destroy(&command_acls);
1216 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1219 union ofp_action action;
1222 /* Set up a flow that matches every packet and directs them to
1223 * OFPP_NORMAL (which goes to us). */
1224 memset(&action, 0, sizeof action);
1225 action.type = htons(OFPAT_OUTPUT);
1226 action.output.len = htons(sizeof action);
1227 action.output.port = htons(OFPP_NORMAL);
1228 memset(&flow, 0, sizeof flow);
1229 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1232 ofproto_set_in_band(br->ofproto, false);
1233 ofproto_set_max_backoff(br->ofproto, 1);
1234 ofproto_set_probe_interval(br->ofproto, 5);
1235 ofproto_set_failure(br->ofproto, false);
1236 ofproto_set_stp(br->ofproto, false);
1240 ofproto_set_controller(br->ofproto, br->controller);
1244 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1249 for (i = 0; i < br->n_ports; i++) {
1250 struct port *port = br->ports[i];
1251 for (j = 0; j < port->n_ifaces; j++) {
1252 struct iface *iface = port->ifaces[j];
1253 svec_add(ifaces, iface->name);
1257 assert(svec_is_unique(ifaces));
1260 /* For robustness, in case the administrator moves around datapath ports behind
1261 * our back, we re-check all the datapath port numbers here.
1263 * This function will set the 'dp_ifidx' members of interfaces that have
1264 * disappeared to -1, so only call this function from a context where those
1265 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1266 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1267 * datapath, which doesn't support UINT16_MAX+1 ports. */
1269 bridge_fetch_dp_ifaces(struct bridge *br)
1271 struct odp_port *dpif_ports;
1272 size_t n_dpif_ports;
1275 /* Reset all interface numbers. */
1276 for (i = 0; i < br->n_ports; i++) {
1277 struct port *port = br->ports[i];
1278 for (j = 0; j < port->n_ifaces; j++) {
1279 struct iface *iface = port->ifaces[j];
1280 iface->dp_ifidx = -1;
1283 port_array_clear(&br->ifaces);
1285 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
1286 for (i = 0; i < n_dpif_ports; i++) {
1287 struct odp_port *p = &dpif_ports[i];
1288 struct iface *iface = iface_lookup(br, p->devname);
1290 if (iface->dp_ifidx >= 0) {
1291 VLOG_WARN("dp%u reported interface %s twice",
1292 dpif_id(&br->dpif), p->devname);
1293 } else if (iface_from_dp_ifidx(br, p->port)) {
1294 VLOG_WARN("dp%u reported interface %"PRIu16" twice",
1295 dpif_id(&br->dpif), p->port);
1297 port_array_set(&br->ifaces, p->port, iface);
1298 iface->dp_ifidx = p->port;
1305 /* Bridge packet processing functions. */
1308 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1310 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1313 static struct bond_entry *
1314 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1316 return &port->bond_hash[bond_hash(mac)];
1320 bond_choose_iface(const struct port *port)
1323 for (i = 0; i < port->n_ifaces; i++) {
1324 if (port->ifaces[i]->enabled) {
1332 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1333 uint16_t *dp_ifidx, tag_type *tags)
1335 struct iface *iface;
1337 assert(port->n_ifaces);
1338 if (port->n_ifaces == 1) {
1339 iface = port->ifaces[0];
1341 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1342 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1343 || !port->ifaces[e->iface_idx]->enabled) {
1344 /* XXX select interface properly. The current interface selection
1345 * is only good for testing the rebalancing code. */
1346 e->iface_idx = bond_choose_iface(port);
1347 if (e->iface_idx < 0) {
1348 *tags |= port->no_ifaces_tag;
1351 e->iface_tag = tag_create_random();
1353 *tags |= e->iface_tag;
1354 iface = port->ifaces[e->iface_idx];
1356 *dp_ifidx = iface->dp_ifidx;
1357 *tags |= iface->tag; /* Currently only used for bonding. */
1362 bond_link_status_update(struct iface *iface, bool carrier)
1364 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1365 struct port *port = iface->port;
1367 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1368 /* Nothing to do. */
1371 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1372 iface->name, carrier ? "detected" : "dropped");
1373 if (carrier == iface->enabled) {
1374 iface->delay_expires = LLONG_MAX;
1375 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1376 iface->name, carrier ? "disabled" : "enabled");
1377 } else if (carrier && port->updelay && port->active_iface < 0) {
1378 iface->delay_expires = time_msec();
1379 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1380 "other interface is up", iface->name, port->updelay);
1382 int delay = carrier ? port->updelay : port->downdelay;
1383 iface->delay_expires = time_msec() + delay;
1386 "interface %s: will be %s if it stays %s for %d ms",
1388 carrier ? "enabled" : "disabled",
1389 carrier ? "up" : "down",
1396 bond_choose_active_iface(struct port *port)
1398 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1400 port->active_iface = bond_choose_iface(port);
1401 port->active_iface_tag = tag_create_random();
1402 if (port->active_iface >= 0) {
1403 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1404 port->name, port->ifaces[port->active_iface]->name);
1406 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1412 bond_enable_slave(struct iface *iface, bool enable)
1414 struct port *port = iface->port;
1415 struct bridge *br = port->bridge;
1417 iface->delay_expires = LLONG_MAX;
1418 if (enable == iface->enabled) {
1422 iface->enabled = enable;
1423 if (!iface->enabled) {
1424 VLOG_WARN("interface %s: disabled", iface->name);
1425 ofproto_revalidate(br->ofproto, iface->tag);
1426 if (iface->port_ifidx == port->active_iface) {
1427 ofproto_revalidate(br->ofproto,
1428 port->active_iface_tag);
1429 bond_choose_active_iface(port);
1431 bond_send_learning_packets(port);
1433 VLOG_WARN("interface %s: enabled", iface->name);
1434 if (port->active_iface < 0) {
1435 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1436 bond_choose_active_iface(port);
1437 bond_send_learning_packets(port);
1439 iface->tag = tag_create_random();
1444 bond_run(struct bridge *br)
1448 for (i = 0; i < br->n_ports; i++) {
1449 struct port *port = br->ports[i];
1450 if (port->n_ifaces < 2) {
1453 for (j = 0; j < port->n_ifaces; j++) {
1454 struct iface *iface = port->ifaces[j];
1455 if (time_msec() >= iface->delay_expires) {
1456 bond_enable_slave(iface, !iface->enabled);
1463 bond_wait(struct bridge *br)
1467 for (i = 0; i < br->n_ports; i++) {
1468 struct port *port = br->ports[i];
1469 if (port->n_ifaces < 2) {
1472 for (j = 0; j < port->n_ifaces; j++) {
1473 struct iface *iface = port->ifaces[j];
1474 if (iface->delay_expires != LLONG_MAX) {
1475 poll_timer_wait(iface->delay_expires - time_msec());
1482 set_dst(struct dst *p, const flow_t *flow,
1483 const struct port *in_port, const struct port *out_port,
1488 * XXX This uses too many tags: any broadcast flow will get one tag per
1489 * destination port, and thus a broadcast on a switch of any size is likely
1490 * to have all tag bits set. We should figure out a way to be smarter.
1492 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1493 *tags |= out_port->stp_state_tag;
1494 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1498 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1499 : in_port->vlan >= 0 ? in_port->vlan
1500 : ntohs(flow->dl_vlan));
1501 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1505 swap_dst(struct dst *p, struct dst *q)
1507 struct dst tmp = *p;
1512 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1513 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1514 * that we push to the datapath. We could in fact fully sort the array by
1515 * vlan, but in most cases there are at most two different vlan tags so that's
1516 * possibly overkill.) */
1518 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1520 struct dst *first = dsts;
1521 struct dst *last = dsts + n_dsts;
1523 while (first != last) {
1525 * - All dsts < first have vlan == 'vlan'.
1526 * - All dsts >= last have vlan != 'vlan'.
1527 * - first < last. */
1528 while (first->vlan == vlan) {
1529 if (++first == last) {
1534 /* Same invariants, plus one additional:
1535 * - first->vlan != vlan.
1537 while (last[-1].vlan != vlan) {
1538 if (--last == first) {
1543 /* Same invariants, plus one additional:
1544 * - last[-1].vlan == vlan.*/
1545 swap_dst(first++, --last);
1550 mirror_mask_ffs(mirror_mask_t mask)
1552 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1557 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1558 const struct dst *test)
1561 for (i = 0; i < n_dsts; i++) {
1562 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1570 port_trunks_vlan(const struct port *port, uint16_t vlan)
1572 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1576 port_includes_vlan(const struct port *port, uint16_t vlan)
1578 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1582 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1583 const struct port *in_port, const struct port *out_port,
1584 struct dst dsts[], tag_type *tags)
1586 mirror_mask_t mirrors = in_port->src_mirrors;
1587 struct dst *dst = dsts;
1590 *tags |= in_port->stp_state_tag;
1591 if (out_port == FLOOD_PORT) {
1592 /* XXX use ODP_FLOOD if no vlans or bonding. */
1593 /* XXX even better, define each VLAN as a datapath port group */
1594 for (i = 0; i < br->n_ports; i++) {
1595 struct port *port = br->ports[i];
1596 if (port != in_port && port_includes_vlan(port, vlan)
1597 && !port->is_mirror_output_port
1598 && set_dst(dst, flow, in_port, port, tags)) {
1599 mirrors |= port->dst_mirrors;
1603 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1604 mirrors |= out_port->dst_mirrors;
1609 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1610 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1612 if (set_dst(dst, flow, in_port, m->out_port, tags)
1613 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1617 for (i = 0; i < br->n_ports; i++) {
1618 struct port *port = br->ports[i];
1619 if (port_includes_vlan(port, m->out_vlan)
1620 && set_dst(dst, flow, in_port, port, tags)
1621 && !dst_is_duplicate(dsts, dst - dsts, dst))
1623 if (port->vlan < 0) {
1624 dst->vlan = m->out_vlan;
1626 if (dst->dp_ifidx == flow->in_port
1627 && dst->vlan == vlan) {
1628 /* Don't send out input port on same VLAN. */
1636 mirrors &= mirrors - 1;
1639 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1644 print_dsts(const struct dst *dsts, size_t n)
1646 for (; n--; dsts++) {
1647 printf(">p%"PRIu16, dsts->dp_ifidx);
1648 if (dsts->vlan != OFP_VLAN_NONE) {
1649 printf("v%"PRIu16, dsts->vlan);
1655 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1656 const struct port *in_port, const struct port *out_port,
1657 tag_type *tags, struct odp_actions *actions)
1659 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1661 const struct dst *p;
1664 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1666 cur_vlan = ntohs(flow->dl_vlan);
1667 for (p = dsts; p < &dsts[n_dsts]; p++) {
1668 union odp_action *a;
1669 if (p->vlan != cur_vlan) {
1670 if (p->vlan == OFP_VLAN_NONE) {
1671 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1673 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1674 a->vlan_vid.vlan_vid = htons(p->vlan);
1678 a = odp_actions_add(actions, ODPAT_OUTPUT);
1679 a->output.port = p->dp_ifidx;
1684 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1686 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1687 return (flow->dl_type == htons(ETH_TYPE_ARP)
1688 && eth_addr_is_broadcast(flow->dl_dst)
1689 && packet->size >= sizeof(struct arp_eth_header)
1690 && arp->ar_op == ARP_OP_REQUEST);
1693 /* If the composed actions may be applied to any packet in the given 'flow',
1694 * returns true. Otherwise, the actions should only be applied to 'packet', or
1695 * not at all, if 'packet' was NULL. */
1697 process_flow(struct bridge *br, const flow_t *flow,
1698 const struct ofpbuf *packet, struct odp_actions *actions,
1701 struct iface *in_iface;
1702 struct port *in_port;
1703 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1706 /* Find the interface and port structure for the received packet. */
1707 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1709 /* No interface? Something fishy... */
1710 if (packet != NULL) {
1711 /* Odd. A few possible reasons here:
1713 * - We deleted an interface but there are still a few packets
1714 * queued up from it.
1716 * - Someone externally added an interface (e.g. with "ovs-dpctl
1717 * add-if") that we don't know about.
1719 * - Packet arrived on the local port but the local port is not
1720 * one of our bridge ports.
1722 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1724 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1725 "interface %"PRIu16, br->name, flow->in_port);
1728 /* Return without adding any actions, to drop packets on this flow. */
1731 in_port = in_iface->port;
1733 /* Figure out what VLAN this packet belongs to.
1735 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1736 * belongs to VLAN 0, so we should treat both cases identically. (In the
1737 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1738 * presumably to allow a priority to be specified. In the latter case, the
1739 * packet does not have any 802.1Q header.) */
1740 vlan = ntohs(flow->dl_vlan);
1741 if (vlan == OFP_VLAN_NONE) {
1744 if (in_port->vlan >= 0) {
1746 /* XXX support double tagging? */
1747 if (packet != NULL) {
1748 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1749 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1750 "packet received on port %s configured with "
1751 "implicit VLAN %"PRIu16,
1752 br->name, ntohs(flow->dl_vlan),
1753 in_port->name, in_port->vlan);
1757 vlan = in_port->vlan;
1759 if (!port_includes_vlan(in_port, vlan)) {
1760 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1761 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1762 "packet received on port %s not configured for "
1764 br->name, vlan, in_port->name, vlan);
1769 /* Drop frames for ports that STP wants entirely killed (both for
1770 * forwarding and for learning). Later, after we do learning, we'll drop
1771 * the frames that STP wants to do learning but not forwarding on. */
1772 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1776 /* Drop frames for reserved multicast addresses. */
1777 if (eth_addr_is_reserved(flow->dl_dst)) {
1781 /* Drop frames on ports reserved for mirroring. */
1782 if (in_port->is_mirror_output_port) {
1783 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1784 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1785 "which is reserved exclusively for mirroring",
1786 br->name, in_port->name);
1790 /* Multicast (and broadcast) packets on bonds need special attention, to
1791 * avoid receiving duplicates. */
1792 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1793 *tags |= in_port->active_iface_tag;
1794 if (in_port->active_iface != in_iface->port_ifidx) {
1795 /* Drop all multicast packets on inactive slaves. */
1798 /* Drop all multicast packets for which we have learned a different
1799 * input port, because we probably sent the packet on one slaves
1800 * and got it back on the active slave. Broadcast ARP replies are
1801 * an exception to this rule: the host has moved to another
1803 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1804 if (src_idx != -1 && src_idx != in_port->port_idx) {
1806 if (!is_bcast_arp_reply(flow, packet)) {
1810 /* No way to know whether it's an ARP reply, because the
1811 * flow entry doesn't include enough information and we
1812 * don't have a packet. Punt. */
1820 out_port = FLOOD_PORT;
1824 /* Learn source MAC (but don't try to learn from revalidation). */
1826 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1827 vlan, in_port->port_idx);
1829 /* The log messages here could actually be useful in debugging,
1830 * so keep the rate limit relatively high. */
1831 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1833 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1834 "on port %s in VLAN %d",
1835 br->name, ETH_ADDR_ARGS(flow->dl_src),
1836 in_port->name, vlan);
1837 ofproto_revalidate(br->ofproto, rev_tag);
1841 /* Determine output port. */
1842 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1844 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1845 out_port = br->ports[out_port_idx];
1849 /* Don't send packets out their input ports. Don't forward frames that STP
1850 * wants us to discard. */
1851 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1856 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1859 * We send out only a single packet, instead of setting up a flow, if the
1860 * packet is an ARP directed to broadcast that arrived on a bonded
1861 * interface. In such a situation ARP requests and replies must be handled
1862 * differently, but OpenFlow unfortunately can't distinguish them.
1864 return (in_port->n_ifaces < 2
1865 || flow->dl_type != htons(ETH_TYPE_ARP)
1866 || !eth_addr_is_broadcast(flow->dl_dst));
1869 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1872 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1873 const struct ofp_phy_port *opp,
1876 struct bridge *br = br_;
1877 struct iface *iface;
1880 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1886 if (reason == OFPPR_DELETE) {
1887 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1888 br->name, iface->name);
1889 iface_destroy(iface);
1890 if (!port->n_ifaces) {
1891 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1892 br->name, port->name);
1898 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1899 if (port->n_ifaces > 1) {
1900 bool up = !(opp->state & OFPPS_LINK_DOWN);
1901 bond_link_status_update(iface, up);
1902 port_update_bond_compat(port);
1908 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1909 struct odp_actions *actions, tag_type *tags, void *br_)
1911 struct bridge *br = br_;
1914 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1915 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1916 brstp_receive(br, flow, payload);
1921 COVERAGE_INC(bridge_process_flow);
1922 return process_flow(br, flow, packet, actions, tags);
1926 bridge_account_flow_ofhook_cb(const flow_t *flow,
1927 const union odp_action *actions,
1928 size_t n_actions, unsigned long long int n_bytes,
1931 struct bridge *br = br_;
1932 const union odp_action *a;
1934 if (!br->has_bonded_ports) {
1938 for (a = actions; a < &actions[n_actions]; a++) {
1939 if (a->type == ODPAT_OUTPUT) {
1940 struct port *port = port_from_dp_ifidx(br, a->output.port);
1941 if (port && port->n_ifaces >= 2) {
1942 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1943 e->tx_bytes += n_bytes;
1950 bridge_account_checkpoint_ofhook_cb(void *br_)
1952 struct bridge *br = br_;
1955 if (!br->has_bonded_ports) {
1959 /* The current ofproto implementation calls this callback at least once a
1960 * second, so this timer implementation is sufficient. */
1961 if (time_msec() < br->bond_next_rebalance) {
1964 br->bond_next_rebalance = time_msec() + 10000;
1966 for (i = 0; i < br->n_ports; i++) {
1967 struct port *port = br->ports[i];
1968 if (port->n_ifaces > 1) {
1969 bond_rebalance_port(port);
1974 static struct ofhooks bridge_ofhooks = {
1975 bridge_port_changed_ofhook_cb,
1976 bridge_normal_ofhook_cb,
1977 bridge_account_flow_ofhook_cb,
1978 bridge_account_checkpoint_ofhook_cb,
1981 /* Bonding functions. */
1983 /* Statistics for a single interface on a bonded port, used for load-based
1984 * bond rebalancing. */
1985 struct slave_balance {
1986 struct iface *iface; /* The interface. */
1987 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
1989 /* All the "bond_entry"s that are assigned to this interface, in order of
1990 * increasing tx_bytes. */
1991 struct bond_entry **hashes;
1995 /* Sorts pointers to pointers to bond_entries in ascending order by the
1996 * interface to which they are assigned, and within a single interface in
1997 * ascending order of bytes transmitted. */
1999 compare_bond_entries(const void *a_, const void *b_)
2001 const struct bond_entry *const *ap = a_;
2002 const struct bond_entry *const *bp = b_;
2003 const struct bond_entry *a = *ap;
2004 const struct bond_entry *b = *bp;
2005 if (a->iface_idx != b->iface_idx) {
2006 return a->iface_idx > b->iface_idx ? 1 : -1;
2007 } else if (a->tx_bytes != b->tx_bytes) {
2008 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2014 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2015 * *descending* order by number of bytes transmitted. */
2017 compare_slave_balance(const void *a_, const void *b_)
2019 const struct slave_balance *a = a_;
2020 const struct slave_balance *b = b_;
2021 if (a->iface->enabled != b->iface->enabled) {
2022 return a->iface->enabled ? -1 : 1;
2023 } else if (a->tx_bytes != b->tx_bytes) {
2024 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2031 swap_bals(struct slave_balance *a, struct slave_balance *b)
2033 struct slave_balance tmp = *a;
2038 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2039 * given that 'p' (and only 'p') might be in the wrong location.
2041 * This function invalidates 'p', since it might now be in a different memory
2044 resort_bals(struct slave_balance *p,
2045 struct slave_balance bals[], size_t n_bals)
2048 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2049 swap_bals(p, p - 1);
2051 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2052 swap_bals(p, p + 1);
2058 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2060 if (VLOG_IS_DBG_ENABLED()) {
2061 struct ds ds = DS_EMPTY_INITIALIZER;
2062 const struct slave_balance *b;
2064 for (b = bals; b < bals + n_bals; b++) {
2068 ds_put_char(&ds, ',');
2070 ds_put_format(&ds, " %s %"PRIu64"kB",
2071 b->iface->name, b->tx_bytes / 1024);
2073 if (!b->iface->enabled) {
2074 ds_put_cstr(&ds, " (disabled)");
2076 if (b->n_hashes > 0) {
2077 ds_put_cstr(&ds, " (");
2078 for (i = 0; i < b->n_hashes; i++) {
2079 const struct bond_entry *e = b->hashes[i];
2081 ds_put_cstr(&ds, " + ");
2083 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2084 e - port->bond_hash, e->tx_bytes / 1024);
2086 ds_put_cstr(&ds, ")");
2089 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2094 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2096 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2097 struct bond_entry *hash)
2099 struct port *port = from->iface->port;
2100 uint64_t delta = hash->tx_bytes;
2102 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2103 "from %s to %s (now carrying %"PRIu64"kB and "
2104 "%"PRIu64"kB load, respectively)",
2105 port->name, delta / 1024, hash - port->bond_hash,
2106 from->iface->name, to->iface->name,
2107 (from->tx_bytes - delta) / 1024,
2108 (to->tx_bytes + delta) / 1024);
2110 /* Delete element from from->hashes.
2112 * We don't bother to add the element to to->hashes because not only would
2113 * it require more work, the only purpose it would be to allow that hash to
2114 * be migrated to another slave in this rebalancing run, and there is no
2115 * point in doing that. */
2116 if (from->hashes[0] == hash) {
2119 int i = hash - from->hashes[0];
2120 memmove(from->hashes + i, from->hashes + i + 1,
2121 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2125 /* Shift load away from 'from' to 'to'. */
2126 from->tx_bytes -= delta;
2127 to->tx_bytes += delta;
2129 /* Arrange for flows to be revalidated. */
2130 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2131 hash->iface_idx = to->iface->port_ifidx;
2132 hash->iface_tag = tag_create_random();
2136 bond_rebalance_port(struct port *port)
2138 struct slave_balance bals[DP_MAX_PORTS];
2140 struct bond_entry *hashes[BOND_MASK + 1];
2141 struct slave_balance *b, *from, *to;
2142 struct bond_entry *e;
2145 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2146 * descending order of tx_bytes, so that bals[0] represents the most
2147 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2150 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2151 * array for each slave_balance structure, we sort our local array of
2152 * hashes in order by slave, so that all of the hashes for a given slave
2153 * become contiguous in memory, and then we point each 'hashes' members of
2154 * a slave_balance structure to the start of a contiguous group. */
2155 n_bals = port->n_ifaces;
2156 for (b = bals; b < &bals[n_bals]; b++) {
2157 b->iface = port->ifaces[b - bals];
2162 for (i = 0; i <= BOND_MASK; i++) {
2163 hashes[i] = &port->bond_hash[i];
2165 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2166 for (i = 0; i <= BOND_MASK; i++) {
2168 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2169 b = &bals[e->iface_idx];
2170 b->tx_bytes += e->tx_bytes;
2172 b->hashes = &hashes[i];
2177 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2178 log_bals(bals, n_bals, port);
2180 /* Discard slaves that aren't enabled (which were sorted to the back of the
2181 * array earlier). */
2182 while (!bals[n_bals - 1].iface->enabled) {
2189 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2190 to = &bals[n_bals - 1];
2191 for (from = bals; from < to; ) {
2192 uint64_t overload = from->tx_bytes - to->tx_bytes;
2193 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2194 /* The extra load on 'from' (and all less-loaded slaves), compared
2195 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2196 * it is less than ~1Mbps. No point in rebalancing. */
2198 } else if (from->n_hashes == 1) {
2199 /* 'from' only carries a single MAC hash, so we can't shift any
2200 * load away from it, even though we want to. */
2203 /* 'from' is carrying significantly more load than 'to', and that
2204 * load is split across at least two different hashes. Pick a hash
2205 * to migrate to 'to' (the least-loaded slave), given that doing so
2206 * must not cause 'to''s load to exceed 'from''s load.
2208 * The sort order we use means that we prefer to shift away the
2209 * smallest hashes instead of the biggest ones. There is little
2210 * reason behind this decision; we could use the opposite sort
2211 * order to shift away big hashes ahead of small ones. */
2214 for (i = 0; i < from->n_hashes; i++) {
2215 uint64_t delta = from->hashes[i]->tx_bytes;
2216 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2220 if (i < from->n_hashes) {
2221 bond_shift_load(from, to, from->hashes[i]);
2223 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2224 * point to different slave_balance structures. It is only
2225 * valid to do these two operations in a row at all because we
2226 * know that 'from' will not move past 'to' and vice versa. */
2227 resort_bals(from, bals, n_bals);
2228 resort_bals(to, bals, n_bals);
2235 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2236 * historical data to decay to <1% in 7 rebalancing runs. */
2237 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2243 bond_send_learning_packets(struct port *port)
2245 struct bridge *br = port->bridge;
2246 struct mac_entry *e;
2247 struct ofpbuf packet;
2248 int error, n_packets, n_errors;
2250 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2254 ofpbuf_init(&packet, 128);
2255 error = n_packets = n_errors = 0;
2256 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2257 static const char s[] = "Open vSwitch Bond Failover";
2258 union ofp_action actions[2], *a;
2259 struct eth_header *eth;
2260 struct llc_snap_header *llc_snap;
2266 if (e->port == port->port_idx
2267 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2271 /* Compose packet to send. */
2272 ofpbuf_clear(&packet);
2273 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2274 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2275 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2276 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2278 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2279 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2280 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2282 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2283 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2284 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2285 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2286 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2288 /* Compose actions. */
2289 memset(actions, 0, sizeof actions);
2292 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2293 a->vlan_vid.len = htons(sizeof *a);
2294 a->vlan_vid.vlan_vid = htons(e->vlan);
2297 a->output.type = htons(OFPAT_OUTPUT);
2298 a->output.len = htons(sizeof *a);
2299 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2304 flow_extract(&packet, ODPP_NONE, &flow);
2305 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2312 ofpbuf_uninit(&packet);
2315 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2316 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2317 "packets, last error was: %s",
2318 port->name, n_errors, n_packets, strerror(error));
2320 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2321 port->name, n_packets);
2325 /* Bonding unixctl user interface functions. */
2328 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2330 struct ds ds = DS_EMPTY_INITIALIZER;
2331 const struct bridge *br;
2333 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2335 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2338 for (i = 0; i < br->n_ports; i++) {
2339 const struct port *port = br->ports[i];
2340 if (port->n_ifaces > 1) {
2343 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2344 for (j = 0; j < port->n_ifaces; j++) {
2345 const struct iface *iface = port->ifaces[j];
2347 ds_put_cstr(&ds, ", ");
2349 ds_put_cstr(&ds, iface->name);
2351 ds_put_char(&ds, '\n');
2355 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2359 static struct port *
2360 bond_find(const char *name)
2362 const struct bridge *br;
2364 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2367 for (i = 0; i < br->n_ports; i++) {
2368 struct port *port = br->ports[i];
2369 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2378 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2380 struct ds ds = DS_EMPTY_INITIALIZER;
2381 const struct port *port;
2384 port = bond_find(args);
2386 unixctl_command_reply(conn, 501, "no such bond");
2390 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2391 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2392 ds_put_format(&ds, "next rebalance: %lld ms\n",
2393 port->bridge->bond_next_rebalance - time_msec());
2394 for (j = 0; j < port->n_ifaces; j++) {
2395 const struct iface *iface = port->ifaces[j];
2396 struct bond_entry *be;
2399 ds_put_format(&ds, "slave %s: %s\n",
2400 iface->name, iface->enabled ? "enabled" : "disabled");
2401 if (j == port->active_iface) {
2402 ds_put_cstr(&ds, "\tactive slave\n");
2404 if (iface->delay_expires != LLONG_MAX) {
2405 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2406 iface->enabled ? "downdelay" : "updelay",
2407 iface->delay_expires - time_msec());
2411 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2412 int hash = be - port->bond_hash;
2413 struct mac_entry *me;
2415 if (be->iface_idx != j) {
2419 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2420 hash, be->tx_bytes / 1024);
2423 if (!port->bridge->ml) {
2427 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2428 &port->bridge->ml->lrus) {
2431 if (bond_hash(me->mac) == hash
2432 && me->port != port->port_idx
2433 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2434 && dp_ifidx == iface->dp_ifidx)
2436 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2437 ETH_ADDR_ARGS(me->mac));
2442 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2447 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2449 char *args = (char *) args_;
2450 char *save_ptr = NULL;
2451 char *bond_s, *hash_s, *slave_s;
2452 uint8_t mac[ETH_ADDR_LEN];
2454 struct iface *iface;
2455 struct bond_entry *entry;
2458 bond_s = strtok_r(args, " ", &save_ptr);
2459 hash_s = strtok_r(NULL, " ", &save_ptr);
2460 slave_s = strtok_r(NULL, " ", &save_ptr);
2462 unixctl_command_reply(conn, 501,
2463 "usage: bond/migrate BOND HASH SLAVE");
2467 port = bond_find(bond_s);
2469 unixctl_command_reply(conn, 501, "no such bond");
2473 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2474 == ETH_ADDR_SCAN_COUNT) {
2475 hash = bond_hash(mac);
2476 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2477 hash = atoi(hash_s) & BOND_MASK;
2479 unixctl_command_reply(conn, 501, "bad hash");
2483 iface = port_lookup_iface(port, slave_s);
2485 unixctl_command_reply(conn, 501, "no such slave");
2489 if (!iface->enabled) {
2490 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2494 entry = &port->bond_hash[hash];
2495 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2496 entry->iface_idx = iface->port_ifidx;
2497 entry->iface_tag = tag_create_random();
2498 unixctl_command_reply(conn, 200, "migrated");
2502 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2504 char *args = (char *) args_;
2505 char *save_ptr = NULL;
2506 char *bond_s, *slave_s;
2508 struct iface *iface;
2510 bond_s = strtok_r(args, " ", &save_ptr);
2511 slave_s = strtok_r(NULL, " ", &save_ptr);
2513 unixctl_command_reply(conn, 501,
2514 "usage: bond/set-active-slave BOND SLAVE");
2518 port = bond_find(bond_s);
2520 unixctl_command_reply(conn, 501, "no such bond");
2524 iface = port_lookup_iface(port, slave_s);
2526 unixctl_command_reply(conn, 501, "no such slave");
2530 if (!iface->enabled) {
2531 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2535 if (port->active_iface != iface->port_ifidx) {
2536 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2537 port->active_iface = iface->port_ifidx;
2538 port->active_iface_tag = tag_create_random();
2539 VLOG_INFO("port %s: active interface is now %s",
2540 port->name, iface->name);
2541 bond_send_learning_packets(port);
2542 unixctl_command_reply(conn, 200, "done");
2544 unixctl_command_reply(conn, 200, "no change");
2549 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2551 char *args = (char *) args_;
2552 char *save_ptr = NULL;
2553 char *bond_s, *slave_s;
2555 struct iface *iface;
2557 bond_s = strtok_r(args, " ", &save_ptr);
2558 slave_s = strtok_r(NULL, " ", &save_ptr);
2560 unixctl_command_reply(conn, 501,
2561 "usage: bond/enable/disable-slave BOND SLAVE");
2565 port = bond_find(bond_s);
2567 unixctl_command_reply(conn, 501, "no such bond");
2571 iface = port_lookup_iface(port, slave_s);
2573 unixctl_command_reply(conn, 501, "no such slave");
2577 bond_enable_slave(iface, enable);
2578 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2582 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2584 enable_slave(conn, args, true);
2588 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2590 enable_slave(conn, args, false);
2596 unixctl_command_register("bond/list", bond_unixctl_list);
2597 unixctl_command_register("bond/show", bond_unixctl_show);
2598 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2599 unixctl_command_register("bond/set-active-slave",
2600 bond_unixctl_set_active_slave);
2601 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2602 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2605 /* Port functions. */
2608 port_create(struct bridge *br, const char *name)
2612 port = xcalloc(1, sizeof *port);
2614 port->port_idx = br->n_ports;
2616 port->trunks = NULL;
2617 port->name = xstrdup(name);
2618 port->active_iface = -1;
2619 port->stp_state = STP_DISABLED;
2620 port->stp_state_tag = 0;
2622 if (br->n_ports >= br->allocated_ports) {
2623 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2626 br->ports[br->n_ports++] = port;
2628 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2633 port_reconfigure(struct port *port)
2635 bool bonded = cfg_has_section("bonding.%s", port->name);
2636 struct svec old_ifaces, new_ifaces;
2637 unsigned long *trunks;
2641 /* Collect old and new interfaces. */
2642 svec_init(&old_ifaces);
2643 svec_init(&new_ifaces);
2644 for (i = 0; i < port->n_ifaces; i++) {
2645 svec_add(&old_ifaces, port->ifaces[i]->name);
2647 svec_sort(&old_ifaces);
2649 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2650 if (!new_ifaces.n) {
2651 VLOG_ERR("port %s: no interfaces specified for bonded port",
2653 } else if (new_ifaces.n == 1) {
2654 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2658 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2659 if (port->updelay < 0) {
2662 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2663 if (port->downdelay < 0) {
2664 port->downdelay = 0;
2667 svec_init(&new_ifaces);
2668 svec_add(&new_ifaces, port->name);
2671 /* Get rid of deleted interfaces and add new interfaces. */
2672 for (i = 0; i < port->n_ifaces; i++) {
2673 struct iface *iface = port->ifaces[i];
2674 if (!svec_contains(&new_ifaces, iface->name)) {
2675 iface_destroy(iface);
2680 for (i = 0; i < new_ifaces.n; i++) {
2681 const char *name = new_ifaces.names[i];
2682 if (!svec_contains(&old_ifaces, name)) {
2683 iface_create(port, name);
2689 if (cfg_has("vlan.%s.tag", port->name)) {
2691 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2692 if (vlan >= 0 && vlan <= 4095) {
2693 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2696 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2697 * they even work as-is. But they have not been tested. */
2698 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2702 if (port->vlan != vlan) {
2704 bridge_flush(port->bridge);
2707 /* Get trunked VLANs. */
2710 size_t n_trunks, n_errors;
2713 trunks = bitmap_allocate(4096);
2714 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2716 for (i = 0; i < n_trunks; i++) {
2717 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2719 bitmap_set1(trunks, trunk);
2725 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2726 port->name, n_trunks);
2728 if (n_errors == n_trunks) {
2730 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2733 bitmap_set_multiple(trunks, 0, 4096, 1);
2736 if (cfg_has("vlan.%s.trunks", port->name)) {
2737 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2738 port->name, port->name);
2742 ? port->trunks != NULL
2743 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2744 bridge_flush(port->bridge);
2746 bitmap_free(port->trunks);
2747 port->trunks = trunks;
2749 svec_destroy(&old_ifaces);
2750 svec_destroy(&new_ifaces);
2754 port_destroy(struct port *port)
2757 struct bridge *br = port->bridge;
2761 proc_net_compat_update_vlan(port->name, NULL, 0);
2763 for (i = 0; i < MAX_MIRRORS; i++) {
2764 struct mirror *m = br->mirrors[i];
2765 if (m && m->out_port == port) {
2770 while (port->n_ifaces > 0) {
2771 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2774 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2775 del->port_idx = port->port_idx;
2778 bitmap_free(port->trunks);
2785 static struct port *
2786 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2788 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2789 return iface ? iface->port : NULL;
2792 static struct port *
2793 port_lookup(const struct bridge *br, const char *name)
2797 for (i = 0; i < br->n_ports; i++) {
2798 struct port *port = br->ports[i];
2799 if (!strcmp(port->name, name)) {
2806 static struct iface *
2807 port_lookup_iface(const struct port *port, const char *name)
2811 for (j = 0; j < port->n_ifaces; j++) {
2812 struct iface *iface = port->ifaces[j];
2813 if (!strcmp(iface->name, name)) {
2821 port_update_bonding(struct port *port)
2823 if (port->n_ifaces < 2) {
2824 /* Not a bonded port. */
2825 if (port->bond_hash) {
2826 free(port->bond_hash);
2827 port->bond_hash = NULL;
2828 proc_net_compat_update_bond(port->name, NULL);
2831 if (!port->bond_hash) {
2834 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2835 for (i = 0; i <= BOND_MASK; i++) {
2836 struct bond_entry *e = &port->bond_hash[i];
2840 port->no_ifaces_tag = tag_create_random();
2841 bond_choose_active_iface(port);
2843 port_update_bond_compat(port);
2848 port_update_bond_compat(struct port *port)
2850 struct compat_bond bond;
2853 if (port->n_ifaces < 2) {
2858 bond.updelay = port->updelay;
2859 bond.downdelay = port->downdelay;
2860 bond.n_slaves = port->n_ifaces;
2861 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2862 for (i = 0; i < port->n_ifaces; i++) {
2863 struct iface *iface = port->ifaces[i];
2864 struct compat_bond_slave *slave = &bond.slaves[i];
2865 slave->name = iface->name;
2866 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2867 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2871 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2873 proc_net_compat_update_bond(port->name, &bond);
2878 port_update_vlan_compat(struct port *port)
2880 struct bridge *br = port->bridge;
2881 char *vlandev_name = NULL;
2883 if (port->vlan > 0) {
2884 /* Figure out the name that the VLAN device should actually have, if it
2885 * existed. This takes some work because the VLAN device would not
2886 * have port->name in its name; rather, it would have the trunk port's
2887 * name, and 'port' would be attached to a bridge that also had the
2888 * VLAN device one of its ports. So we need to find a trunk port that
2889 * includes port->vlan.
2891 * There might be more than one candidate. This doesn't happen on
2892 * XenServer, so if it happens we just pick the first choice in
2893 * alphabetical order instead of creating multiple VLAN devices. */
2895 for (i = 0; i < br->n_ports; i++) {
2896 struct port *p = br->ports[i];
2897 if (port_trunks_vlan(p, port->vlan)
2899 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2901 const uint8_t *ea = p->ifaces[0]->mac;
2902 if (!eth_addr_is_multicast(ea) &&
2903 !eth_addr_is_reserved(ea) &&
2904 !eth_addr_is_zero(ea)) {
2905 vlandev_name = p->name;
2910 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2913 /* Interface functions. */
2916 iface_create(struct port *port, const char *name)
2918 struct iface *iface;
2920 iface = xcalloc(1, sizeof *iface);
2922 iface->port_ifidx = port->n_ifaces;
2923 iface->name = xstrdup(name);
2924 iface->dp_ifidx = -1;
2925 iface->tag = tag_create_random();
2926 iface->delay_expires = LLONG_MAX;
2928 netdev_nodev_get_etheraddr(name, iface->mac);
2929 netdev_nodev_get_carrier(name, &iface->enabled);
2931 if (port->n_ifaces >= port->allocated_ifaces) {
2932 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2933 sizeof *port->ifaces);
2935 port->ifaces[port->n_ifaces++] = iface;
2936 if (port->n_ifaces > 1) {
2937 port->bridge->has_bonded_ports = true;
2940 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2942 port_update_bonding(port);
2943 bridge_flush(port->bridge);
2947 iface_destroy(struct iface *iface)
2950 struct port *port = iface->port;
2951 struct bridge *br = port->bridge;
2952 bool del_active = port->active_iface == iface->port_ifidx;
2955 if (iface->dp_ifidx >= 0) {
2956 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
2959 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
2960 del->port_ifidx = iface->port_ifidx;
2966 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2967 bond_choose_active_iface(port);
2968 bond_send_learning_packets(port);
2971 port_update_bonding(port);
2972 bridge_flush(port->bridge);
2976 static struct iface *
2977 iface_lookup(const struct bridge *br, const char *name)
2981 for (i = 0; i < br->n_ports; i++) {
2982 struct port *port = br->ports[i];
2983 for (j = 0; j < port->n_ifaces; j++) {
2984 struct iface *iface = port->ifaces[j];
2985 if (!strcmp(iface->name, name)) {
2993 static struct iface *
2994 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2996 return port_array_get(&br->ifaces, dp_ifidx);
2999 /* Port mirroring. */
3002 mirror_reconfigure(struct bridge *br)
3004 struct svec old_mirrors, new_mirrors;
3007 /* Collect old and new mirrors. */
3008 svec_init(&old_mirrors);
3009 svec_init(&new_mirrors);
3010 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3011 for (i = 0; i < MAX_MIRRORS; i++) {
3012 if (br->mirrors[i]) {
3013 svec_add(&old_mirrors, br->mirrors[i]->name);
3017 /* Get rid of deleted mirrors and add new mirrors. */
3018 svec_sort(&old_mirrors);
3019 assert(svec_is_unique(&old_mirrors));
3020 svec_sort(&new_mirrors);
3021 assert(svec_is_unique(&new_mirrors));
3022 for (i = 0; i < MAX_MIRRORS; i++) {
3023 struct mirror *m = br->mirrors[i];
3024 if (m && !svec_contains(&new_mirrors, m->name)) {
3028 for (i = 0; i < new_mirrors.n; i++) {
3029 const char *name = new_mirrors.names[i];
3030 if (!svec_contains(&old_mirrors, name)) {
3031 mirror_create(br, name);
3034 svec_destroy(&old_mirrors);
3035 svec_destroy(&new_mirrors);
3037 /* Reconfigure all mirrors. */
3038 for (i = 0; i < MAX_MIRRORS; i++) {
3039 if (br->mirrors[i]) {
3040 mirror_reconfigure_one(br->mirrors[i]);
3044 /* Update port reserved status. */
3045 for (i = 0; i < br->n_ports; i++) {
3046 br->ports[i]->is_mirror_output_port = false;
3048 for (i = 0; i < MAX_MIRRORS; i++) {
3049 struct mirror *m = br->mirrors[i];
3050 if (m && m->out_port) {
3051 m->out_port->is_mirror_output_port = true;
3057 mirror_create(struct bridge *br, const char *name)
3062 for (i = 0; ; i++) {
3063 if (i >= MAX_MIRRORS) {
3064 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3065 "cannot create %s", br->name, MAX_MIRRORS, name);
3068 if (!br->mirrors[i]) {
3073 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3076 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3079 m->name = xstrdup(name);
3080 svec_init(&m->src_ports);
3081 svec_init(&m->dst_ports);
3089 mirror_destroy(struct mirror *m)
3092 struct bridge *br = m->bridge;
3095 for (i = 0; i < br->n_ports; i++) {
3096 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3097 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3100 svec_destroy(&m->src_ports);
3101 svec_destroy(&m->dst_ports);
3104 m->bridge->mirrors[m->idx] = NULL;
3112 prune_ports(struct mirror *m, struct svec *ports)
3117 svec_sort_unique(ports);
3120 for (i = 0; i < ports->n; i++) {
3121 const char *name = ports->names[i];
3122 if (port_lookup(m->bridge, name)) {
3123 svec_add(&tmp, name);
3125 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3126 m->bridge->name, m->name, name);
3129 svec_swap(ports, &tmp);
3134 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3138 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3139 * order won't give us numeric sort order. But that's good enough for what
3140 * we need right now. */
3141 svec_sort_unique(vlan_strings);
3143 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3145 for (i = 0; i < vlan_strings->n; i++) {
3146 const char *name = vlan_strings->names[i];
3148 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3149 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3150 m->bridge->name, m->name, name);
3152 (*vlans)[n_vlans++] = vlan;
3159 vlan_is_mirrored(const struct mirror *m, int vlan)
3163 for (i = 0; i < m->n_vlans; i++) {
3164 if (m->vlans[i] == vlan) {
3172 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3176 for (i = 0; i < m->n_vlans; i++) {
3177 if (port_trunks_vlan(p, m->vlans[i])) {
3185 mirror_reconfigure_one(struct mirror *m)
3187 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3188 struct svec src_ports, dst_ports, ports;
3189 struct svec vlan_strings;
3190 mirror_mask_t mirror_bit;
3191 const char *out_port_name;
3192 struct port *out_port;
3197 bool mirror_all_ports;
3199 /* Get output port. */
3200 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3201 m->bridge->name, m->name);
3202 if (out_port_name) {
3203 out_port = port_lookup(m->bridge, out_port_name);
3205 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3206 "named %s", pfx, m->bridge->name, out_port_name);
3213 if (cfg_has("%s.output.vlan", pfx)) {
3214 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3215 "ignoring %s.output.vlan", pfx, pfx, pfx);
3217 } else if (cfg_has("%s.output.vlan", pfx)) {
3219 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3221 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3222 "but exactly one is required; disabling port mirror %s",
3223 pfx, pfx, pfx, pfx);
3229 /* Get all the ports, and drop duplicates and ports that don't exist. */
3230 svec_init(&src_ports);
3231 svec_init(&dst_ports);
3233 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3234 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3235 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3236 svec_append(&src_ports, &ports);
3237 svec_append(&dst_ports, &ports);
3238 svec_destroy(&ports);
3239 prune_ports(m, &src_ports);
3240 prune_ports(m, &dst_ports);
3242 /* Get all the vlans, and drop duplicate and invalid vlans. */
3243 svec_init(&vlan_strings);
3244 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3245 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3246 svec_destroy(&vlan_strings);
3248 /* Update mirror data. */
3249 if (!svec_equal(&m->src_ports, &src_ports)
3250 || !svec_equal(&m->dst_ports, &dst_ports)
3251 || m->n_vlans != n_vlans
3252 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3253 || m->out_port != out_port
3254 || m->out_vlan != out_vlan) {
3255 bridge_flush(m->bridge);
3257 svec_swap(&m->src_ports, &src_ports);
3258 svec_swap(&m->dst_ports, &dst_ports);
3261 m->n_vlans = n_vlans;
3262 m->out_port = out_port;
3263 m->out_vlan = out_vlan;
3265 /* If no selection criteria have been given, mirror for all ports. */
3266 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3269 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3270 for (i = 0; i < m->bridge->n_ports; i++) {
3271 struct port *port = m->bridge->ports[i];
3273 if (mirror_all_ports
3274 || svec_contains(&m->src_ports, port->name)
3277 ? port_trunks_any_mirrored_vlan(m, port)
3278 : vlan_is_mirrored(m, port->vlan)))) {
3279 port->src_mirrors |= mirror_bit;
3281 port->src_mirrors &= ~mirror_bit;
3284 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3285 port->dst_mirrors |= mirror_bit;
3287 port->dst_mirrors &= ~mirror_bit;
3292 svec_destroy(&src_ports);
3293 svec_destroy(&dst_ports);
3297 /* Spanning tree protocol. */
3299 static void brstp_update_port_state(struct port *);
3302 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3304 struct bridge *br = br_;
3305 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3306 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3308 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3310 } else if (eth_addr_is_zero(iface->mac)) {
3311 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3314 union ofp_action action;
3315 struct eth_header *eth = pkt->l2;
3318 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3320 memset(&action, 0, sizeof action);
3321 action.type = htons(OFPAT_OUTPUT);
3322 action.output.len = htons(sizeof action);
3323 action.output.port = htons(port_no);
3325 flow_extract(pkt, ODPP_NONE, &flow);
3326 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3332 brstp_reconfigure(struct bridge *br)
3336 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3338 stp_destroy(br->stp);
3344 uint64_t bridge_address, bridge_id;
3345 int bridge_priority;
3347 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3348 if (!bridge_address) {
3350 bridge_address = (stp_get_bridge_id(br->stp)
3351 & ((UINT64_C(1) << 48) - 1));
3353 uint8_t mac[ETH_ADDR_LEN];
3354 eth_addr_random(mac);
3355 bridge_address = eth_addr_to_uint64(mac);
3359 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3361 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3363 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3366 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3368 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3369 br->stp_last_tick = time_msec();
3372 if (bridge_id != stp_get_bridge_id(br->stp)) {
3373 stp_set_bridge_id(br->stp, bridge_id);
3378 for (i = 0; i < br->n_ports; i++) {
3379 struct port *p = br->ports[i];
3381 struct stp_port *sp;
3382 int path_cost, priority;
3388 dp_ifidx = p->ifaces[0]->dp_ifidx;
3389 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3393 sp = stp_get_port(br->stp, dp_ifidx);
3394 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3395 "stp.%s.port.%s.enabled",
3397 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3398 br->name, p->name));
3399 if (p->is_mirror_output_port) {
3402 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3403 bridge_flush(br); /* Might not be necessary. */
3405 stp_port_enable(sp);
3407 stp_port_disable(sp);
3411 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3413 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3415 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3416 "stp.%s.port.%s.priority",
3418 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3420 : STP_DEFAULT_PORT_PRIORITY);
3421 stp_port_set_priority(sp, priority);
3424 brstp_adjust_timers(br);
3426 for (i = 0; i < br->n_ports; i++) {
3427 brstp_update_port_state(br->ports[i]);
3432 brstp_update_port_state(struct port *p)
3434 struct bridge *br = p->bridge;
3435 enum stp_state state;
3437 /* Figure out new state. */
3438 state = STP_DISABLED;
3439 if (br->stp && p->n_ifaces > 0) {
3440 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3441 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3442 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3447 if (p->stp_state != state) {
3448 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3449 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3450 p->name, stp_state_name(p->stp_state),
3451 stp_state_name(state));
3452 if (p->stp_state == STP_DISABLED) {
3455 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3457 p->stp_state = state;
3458 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3459 : tag_create_random());
3464 brstp_adjust_timers(struct bridge *br)
3466 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3467 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3468 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3470 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3471 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3472 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3476 brstp_run(struct bridge *br)
3479 long long int now = time_msec();
3480 long long int elapsed = now - br->stp_last_tick;
3481 struct stp_port *sp;
3484 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3485 br->stp_last_tick = now;
3487 while (stp_get_changed_port(br->stp, &sp)) {
3488 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3490 brstp_update_port_state(p);
3497 brstp_wait(struct bridge *br)
3500 poll_timer_wait(1000);