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 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
813 e->port, e->vlan, ETH_ADDR_ARGS(e->mac),
817 unixctl_command_reply(conn, 200, ds_cstr(&ds));
821 /* Bridge reconfiguration functions. */
823 static struct bridge *
824 bridge_create(const char *name)
829 assert(!bridge_lookup(name));
830 br = xcalloc(1, sizeof *br);
832 error = dpif_create(name, &br->dpif);
833 if (error == EEXIST) {
834 error = dpif_open(name, &br->dpif);
836 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
837 name, strerror(error));
841 dpif_flow_flush(&br->dpif);
843 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
848 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
850 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
851 dpif_delete(&br->dpif);
852 dpif_close(&br->dpif);
857 br->name = xstrdup(name);
858 br->ml = mac_learning_create();
859 br->sent_config_request = false;
860 eth_addr_random(br->default_ea);
862 port_array_init(&br->ifaces);
865 br->bond_next_rebalance = time_msec() + 10000;
867 list_push_back(&all_bridges, &br->node);
869 VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif));
875 bridge_destroy(struct bridge *br)
880 while (br->n_ports > 0) {
881 port_destroy(br->ports[br->n_ports - 1]);
883 list_remove(&br->node);
884 error = dpif_delete(&br->dpif);
885 if (error && error != ENOENT) {
886 VLOG_ERR("failed to delete dp%u: %s",
887 dpif_id(&br->dpif), strerror(error));
889 dpif_close(&br->dpif);
890 ofproto_destroy(br->ofproto);
891 free(br->controller);
892 mac_learning_destroy(br->ml);
893 port_array_destroy(&br->ifaces);
900 static struct bridge *
901 bridge_lookup(const char *name)
905 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
906 if (!strcmp(br->name, name)) {
914 bridge_exists(const char *name)
916 return bridge_lookup(name) ? true : false;
920 bridge_get_datapathid(const char *name)
922 struct bridge *br = bridge_lookup(name);
923 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
927 bridge_run_one(struct bridge *br)
931 error = ofproto_run1(br->ofproto);
937 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
942 error = ofproto_run2(br->ofproto, br->flush);
949 bridge_get_controller(const struct bridge *br)
951 const char *controller;
953 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
955 controller = cfg_get_string(0, "mgmt.controller");
957 return controller && controller[0] ? controller : NULL;
961 bridge_reconfigure_one(struct bridge *br)
963 struct svec old_ports, new_ports, ifaces;
964 struct svec listeners, old_listeners;
965 struct svec snoops, old_snoops;
968 /* Collect old ports. */
969 svec_init(&old_ports);
970 for (i = 0; i < br->n_ports; i++) {
971 svec_add(&old_ports, br->ports[i]->name);
973 svec_sort(&old_ports);
974 assert(svec_is_unique(&old_ports));
976 /* Collect new ports. */
977 svec_init(&new_ports);
978 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
979 svec_sort(&new_ports);
980 if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) {
981 svec_add(&new_ports, br->name);
982 svec_sort(&new_ports);
984 if (!svec_is_unique(&new_ports)) {
985 VLOG_WARN("bridge %s: %s specified twice as bridge port",
986 br->name, svec_get_duplicate(&new_ports));
987 svec_unique(&new_ports);
990 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
992 /* Get rid of deleted ports and add new ports. */
993 for (i = 0; i < br->n_ports; ) {
994 struct port *port = br->ports[i];
995 if (!svec_contains(&new_ports, port->name)) {
1001 for (i = 0; i < new_ports.n; i++) {
1002 const char *name = new_ports.names[i];
1003 if (!svec_contains(&old_ports, name)) {
1004 port_create(br, name);
1007 svec_destroy(&old_ports);
1008 svec_destroy(&new_ports);
1010 /* Reconfigure all ports. */
1011 for (i = 0; i < br->n_ports; i++) {
1012 port_reconfigure(br->ports[i]);
1015 /* Check and delete duplicate interfaces. */
1017 for (i = 0; i < br->n_ports; ) {
1018 struct port *port = br->ports[i];
1019 for (j = 0; j < port->n_ifaces; ) {
1020 struct iface *iface = port->ifaces[j];
1021 if (svec_contains(&ifaces, iface->name)) {
1022 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1024 br->name, iface->name, port->name);
1025 iface_destroy(iface);
1027 svec_add(&ifaces, iface->name);
1032 if (!port->n_ifaces) {
1033 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1039 svec_destroy(&ifaces);
1041 /* Delete all flows if we're switching from connected to standalone or vice
1042 * versa. (XXX Should we delete all flows if we are switching from one
1043 * controller to another?) */
1045 /* Configure OpenFlow management listeners. */
1046 svec_init(&listeners);
1047 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1049 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1050 ovs_rundir, br->name));
1051 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1052 svec_clear(&listeners);
1054 svec_sort_unique(&listeners);
1056 svec_init(&old_listeners);
1057 ofproto_get_listeners(br->ofproto, &old_listeners);
1058 svec_sort_unique(&old_listeners);
1060 if (!svec_equal(&listeners, &old_listeners)) {
1061 ofproto_set_listeners(br->ofproto, &listeners);
1063 svec_destroy(&listeners);
1064 svec_destroy(&old_listeners);
1066 /* Configure OpenFlow controller connection snooping. */
1068 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1070 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1071 ovs_rundir, br->name));
1072 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1073 svec_clear(&snoops);
1075 svec_sort_unique(&snoops);
1077 svec_init(&old_snoops);
1078 ofproto_get_snoops(br->ofproto, &old_snoops);
1079 svec_sort_unique(&old_snoops);
1081 if (!svec_equal(&snoops, &old_snoops)) {
1082 ofproto_set_snoops(br->ofproto, &snoops);
1084 svec_destroy(&snoops);
1085 svec_destroy(&old_snoops);
1087 mirror_reconfigure(br);
1091 bridge_reconfigure_controller(struct bridge *br)
1093 char *pfx = xasprintf("bridge.%s.controller", br->name);
1094 const char *controller;
1096 controller = bridge_get_controller(br);
1097 if ((br->controller != NULL) != (controller != NULL)) {
1098 ofproto_flush_flows(br->ofproto);
1100 free(br->controller);
1101 br->controller = controller ? xstrdup(controller) : NULL;
1104 const char *fail_mode;
1105 int max_backoff, probe;
1106 int rate_limit, burst_limit;
1108 if (!strcmp(controller, "discover")) {
1109 bool update_resolv_conf = true;
1111 if (cfg_has("%s.update-resolv.conf", pfx)) {
1112 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1115 ofproto_set_discovery(br->ofproto, true,
1116 cfg_get_string(0, "%s.accept-regex", pfx),
1117 update_resolv_conf);
1119 struct netdev *netdev;
1123 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1125 || cfg_get_bool(0, "%s.in-band", pfx));
1126 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1127 ofproto_set_in_band(br->ofproto, in_band);
1129 error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev);
1131 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1132 struct in_addr ip, mask, gateway;
1133 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1134 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1135 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1137 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1139 mask.s_addr = guess_netmask(ip.s_addr);
1141 if (!netdev_set_in4(netdev, ip, mask)) {
1142 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1144 br->name, IP_ARGS(&ip.s_addr),
1145 IP_ARGS(&mask.s_addr));
1148 if (gateway.s_addr) {
1149 if (!netdev_add_router(gateway)) {
1150 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1151 br->name, IP_ARGS(&gateway.s_addr));
1155 netdev_close(netdev);
1159 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1161 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1163 ofproto_set_failure(br->ofproto,
1165 || !strcmp(fail_mode, "standalone")
1166 || !strcmp(fail_mode, "open")));
1168 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1170 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1175 ofproto_set_probe_interval(br->ofproto, probe);
1177 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1179 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1184 ofproto_set_max_backoff(br->ofproto, max_backoff);
1186 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1188 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1190 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1192 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1194 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1196 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1198 if (cfg_has("%s.commands.acl", pfx)) {
1199 struct svec command_acls;
1202 svec_init(&command_acls);
1203 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1204 command_acl = svec_join(&command_acls, ",", "");
1206 ofproto_set_remote_execution(br->ofproto, command_acl,
1207 cfg_get_string(0, "%s.commands.dir",
1210 svec_destroy(&command_acls);
1213 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1216 union ofp_action action;
1219 /* Set up a flow that matches every packet and directs them to
1220 * OFPP_NORMAL (which goes to us). */
1221 memset(&action, 0, sizeof action);
1222 action.type = htons(OFPAT_OUTPUT);
1223 action.output.len = htons(sizeof action);
1224 action.output.port = htons(OFPP_NORMAL);
1225 memset(&flow, 0, sizeof flow);
1226 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1229 ofproto_set_in_band(br->ofproto, false);
1230 ofproto_set_max_backoff(br->ofproto, 1);
1231 ofproto_set_probe_interval(br->ofproto, 5);
1232 ofproto_set_failure(br->ofproto, false);
1233 ofproto_set_stp(br->ofproto, false);
1237 ofproto_set_controller(br->ofproto, br->controller);
1241 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1246 for (i = 0; i < br->n_ports; i++) {
1247 struct port *port = br->ports[i];
1248 for (j = 0; j < port->n_ifaces; j++) {
1249 struct iface *iface = port->ifaces[j];
1250 svec_add(ifaces, iface->name);
1254 assert(svec_is_unique(ifaces));
1257 /* For robustness, in case the administrator moves around datapath ports behind
1258 * our back, we re-check all the datapath port numbers here.
1260 * This function will set the 'dp_ifidx' members of interfaces that have
1261 * disappeared to -1, so only call this function from a context where those
1262 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1263 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1264 * datapath, which doesn't support UINT16_MAX+1 ports. */
1266 bridge_fetch_dp_ifaces(struct bridge *br)
1268 struct odp_port *dpif_ports;
1269 size_t n_dpif_ports;
1272 /* Reset all interface numbers. */
1273 for (i = 0; i < br->n_ports; i++) {
1274 struct port *port = br->ports[i];
1275 for (j = 0; j < port->n_ifaces; j++) {
1276 struct iface *iface = port->ifaces[j];
1277 iface->dp_ifidx = -1;
1280 port_array_clear(&br->ifaces);
1282 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
1283 for (i = 0; i < n_dpif_ports; i++) {
1284 struct odp_port *p = &dpif_ports[i];
1285 struct iface *iface = iface_lookup(br, p->devname);
1287 if (iface->dp_ifidx >= 0) {
1288 VLOG_WARN("dp%u reported interface %s twice",
1289 dpif_id(&br->dpif), p->devname);
1290 } else if (iface_from_dp_ifidx(br, p->port)) {
1291 VLOG_WARN("dp%u reported interface %"PRIu16" twice",
1292 dpif_id(&br->dpif), p->port);
1294 port_array_set(&br->ifaces, p->port, iface);
1295 iface->dp_ifidx = p->port;
1302 /* Bridge packet processing functions. */
1305 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1307 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1310 static struct bond_entry *
1311 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1313 return &port->bond_hash[bond_hash(mac)];
1317 bond_choose_iface(const struct port *port)
1320 for (i = 0; i < port->n_ifaces; i++) {
1321 if (port->ifaces[i]->enabled) {
1329 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1330 uint16_t *dp_ifidx, tag_type *tags)
1332 struct iface *iface;
1334 assert(port->n_ifaces);
1335 if (port->n_ifaces == 1) {
1336 iface = port->ifaces[0];
1338 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1339 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1340 || !port->ifaces[e->iface_idx]->enabled) {
1341 /* XXX select interface properly. The current interface selection
1342 * is only good for testing the rebalancing code. */
1343 e->iface_idx = bond_choose_iface(port);
1344 if (e->iface_idx < 0) {
1345 *tags |= port->no_ifaces_tag;
1348 e->iface_tag = tag_create_random();
1350 *tags |= e->iface_tag;
1351 iface = port->ifaces[e->iface_idx];
1353 *dp_ifidx = iface->dp_ifidx;
1354 *tags |= iface->tag; /* Currently only used for bonding. */
1359 bond_link_status_update(struct iface *iface, bool carrier)
1361 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1362 struct port *port = iface->port;
1364 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1365 /* Nothing to do. */
1368 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1369 iface->name, carrier ? "detected" : "dropped");
1370 if (carrier == iface->enabled) {
1371 iface->delay_expires = LLONG_MAX;
1372 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1373 iface->name, carrier ? "disabled" : "enabled");
1374 } else if (carrier && port->updelay && port->active_iface < 0) {
1375 iface->delay_expires = time_msec();
1376 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1377 "other interface is up", iface->name, port->updelay);
1379 int delay = carrier ? port->updelay : port->downdelay;
1380 iface->delay_expires = time_msec() + delay;
1383 "interface %s: will be %s if it stays %s for %d ms",
1385 carrier ? "enabled" : "disabled",
1386 carrier ? "up" : "down",
1393 bond_choose_active_iface(struct port *port)
1395 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1397 port->active_iface = bond_choose_iface(port);
1398 port->active_iface_tag = tag_create_random();
1399 if (port->active_iface >= 0) {
1400 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1401 port->name, port->ifaces[port->active_iface]->name);
1403 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1409 bond_enable_slave(struct iface *iface, bool enable)
1411 struct port *port = iface->port;
1412 struct bridge *br = port->bridge;
1414 iface->delay_expires = LLONG_MAX;
1415 if (enable == iface->enabled) {
1419 iface->enabled = enable;
1420 if (!iface->enabled) {
1421 VLOG_WARN("interface %s: disabled", iface->name);
1422 ofproto_revalidate(br->ofproto, iface->tag);
1423 if (iface->port_ifidx == port->active_iface) {
1424 ofproto_revalidate(br->ofproto,
1425 port->active_iface_tag);
1426 bond_choose_active_iface(port);
1428 bond_send_learning_packets(port);
1430 VLOG_WARN("interface %s: enabled", iface->name);
1431 if (port->active_iface < 0) {
1432 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1433 bond_choose_active_iface(port);
1434 bond_send_learning_packets(port);
1436 iface->tag = tag_create_random();
1441 bond_run(struct bridge *br)
1445 for (i = 0; i < br->n_ports; i++) {
1446 struct port *port = br->ports[i];
1447 if (port->n_ifaces < 2) {
1450 for (j = 0; j < port->n_ifaces; j++) {
1451 struct iface *iface = port->ifaces[j];
1452 if (time_msec() >= iface->delay_expires) {
1453 bond_enable_slave(iface, !iface->enabled);
1460 bond_wait(struct bridge *br)
1464 for (i = 0; i < br->n_ports; i++) {
1465 struct port *port = br->ports[i];
1466 if (port->n_ifaces < 2) {
1469 for (j = 0; j < port->n_ifaces; j++) {
1470 struct iface *iface = port->ifaces[j];
1471 if (iface->delay_expires != LLONG_MAX) {
1472 poll_timer_wait(iface->delay_expires - time_msec());
1479 set_dst(struct dst *p, const flow_t *flow,
1480 const struct port *in_port, const struct port *out_port,
1485 * XXX This uses too many tags: any broadcast flow will get one tag per
1486 * destination port, and thus a broadcast on a switch of any size is likely
1487 * to have all tag bits set. We should figure out a way to be smarter.
1489 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1490 *tags |= out_port->stp_state_tag;
1491 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1495 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1496 : in_port->vlan >= 0 ? in_port->vlan
1497 : ntohs(flow->dl_vlan));
1498 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1502 swap_dst(struct dst *p, struct dst *q)
1504 struct dst tmp = *p;
1509 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1510 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1511 * that we push to the datapath. We could in fact fully sort the array by
1512 * vlan, but in most cases there are at most two different vlan tags so that's
1513 * possibly overkill.) */
1515 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1517 struct dst *first = dsts;
1518 struct dst *last = dsts + n_dsts;
1520 while (first != last) {
1522 * - All dsts < first have vlan == 'vlan'.
1523 * - All dsts >= last have vlan != 'vlan'.
1524 * - first < last. */
1525 while (first->vlan == vlan) {
1526 if (++first == last) {
1531 /* Same invariants, plus one additional:
1532 * - first->vlan != vlan.
1534 while (last[-1].vlan != vlan) {
1535 if (--last == first) {
1540 /* Same invariants, plus one additional:
1541 * - last[-1].vlan == vlan.*/
1542 swap_dst(first++, --last);
1547 mirror_mask_ffs(mirror_mask_t mask)
1549 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1554 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1555 const struct dst *test)
1558 for (i = 0; i < n_dsts; i++) {
1559 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1567 port_trunks_vlan(const struct port *port, uint16_t vlan)
1569 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1573 port_includes_vlan(const struct port *port, uint16_t vlan)
1575 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1579 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1580 const struct port *in_port, const struct port *out_port,
1581 struct dst dsts[], tag_type *tags)
1583 mirror_mask_t mirrors = in_port->src_mirrors;
1584 struct dst *dst = dsts;
1587 *tags |= in_port->stp_state_tag;
1588 if (out_port == FLOOD_PORT) {
1589 /* XXX use ODP_FLOOD if no vlans or bonding. */
1590 /* XXX even better, define each VLAN as a datapath port group */
1591 for (i = 0; i < br->n_ports; i++) {
1592 struct port *port = br->ports[i];
1593 if (port != in_port && port_includes_vlan(port, vlan)
1594 && !port->is_mirror_output_port
1595 && set_dst(dst, flow, in_port, port, tags)) {
1596 mirrors |= port->dst_mirrors;
1600 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1601 mirrors |= out_port->dst_mirrors;
1606 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1607 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1609 if (set_dst(dst, flow, in_port, m->out_port, tags)
1610 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1614 for (i = 0; i < br->n_ports; i++) {
1615 struct port *port = br->ports[i];
1616 if (port_includes_vlan(port, m->out_vlan)
1617 && set_dst(dst, flow, in_port, port, tags)
1618 && !dst_is_duplicate(dsts, dst - dsts, dst))
1620 if (port->vlan < 0) {
1621 dst->vlan = m->out_vlan;
1623 if (dst->dp_ifidx == flow->in_port
1624 && dst->vlan == vlan) {
1625 /* Don't send out input port on same VLAN. */
1633 mirrors &= mirrors - 1;
1636 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1641 print_dsts(const struct dst *dsts, size_t n)
1643 for (; n--; dsts++) {
1644 printf(">p%"PRIu16, dsts->dp_ifidx);
1645 if (dsts->vlan != OFP_VLAN_NONE) {
1646 printf("v%"PRIu16, dsts->vlan);
1652 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1653 const struct port *in_port, const struct port *out_port,
1654 tag_type *tags, struct odp_actions *actions)
1656 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1658 const struct dst *p;
1661 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1663 cur_vlan = ntohs(flow->dl_vlan);
1664 for (p = dsts; p < &dsts[n_dsts]; p++) {
1665 union odp_action *a;
1666 if (p->vlan != cur_vlan) {
1667 if (p->vlan == OFP_VLAN_NONE) {
1668 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1670 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1671 a->vlan_vid.vlan_vid = htons(p->vlan);
1675 a = odp_actions_add(actions, ODPAT_OUTPUT);
1676 a->output.port = p->dp_ifidx;
1681 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1683 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1684 return (flow->dl_type == htons(ETH_TYPE_ARP)
1685 && eth_addr_is_broadcast(flow->dl_dst)
1686 && packet->size >= sizeof(struct arp_eth_header)
1687 && arp->ar_op == ARP_OP_REQUEST);
1690 /* If the composed actions may be applied to any packet in the given 'flow',
1691 * returns true. Otherwise, the actions should only be applied to 'packet', or
1692 * not at all, if 'packet' was NULL. */
1694 process_flow(struct bridge *br, const flow_t *flow,
1695 const struct ofpbuf *packet, struct odp_actions *actions,
1698 struct iface *in_iface;
1699 struct port *in_port;
1700 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1703 /* Find the interface and port structure for the received packet. */
1704 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1706 /* No interface? Something fishy... */
1707 if (packet != NULL) {
1708 /* Odd. A few possible reasons here:
1710 * - We deleted an interface but there are still a few packets
1711 * queued up from it.
1713 * - Someone externally added an interface (e.g. with "ovs-dpctl
1714 * add-if") that we don't know about.
1716 * - Packet arrived on the local port but the local port is not
1717 * one of our bridge ports.
1719 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1721 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1722 "interface %"PRIu16, br->name, flow->in_port);
1725 /* Return without adding any actions, to drop packets on this flow. */
1728 in_port = in_iface->port;
1730 /* Figure out what VLAN this packet belongs to.
1732 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1733 * belongs to VLAN 0, so we should treat both cases identically. (In the
1734 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1735 * presumably to allow a priority to be specified. In the latter case, the
1736 * packet does not have any 802.1Q header.) */
1737 vlan = ntohs(flow->dl_vlan);
1738 if (vlan == OFP_VLAN_NONE) {
1741 if (in_port->vlan >= 0) {
1743 /* XXX support double tagging? */
1744 if (packet != NULL) {
1745 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1746 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1747 "packet received on port %s configured with "
1748 "implicit VLAN %"PRIu16,
1749 br->name, ntohs(flow->dl_vlan),
1750 in_port->name, in_port->vlan);
1754 vlan = in_port->vlan;
1756 if (!port_includes_vlan(in_port, vlan)) {
1757 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1758 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1759 "packet received on port %s not configured for "
1761 br->name, vlan, in_port->name, vlan);
1766 /* Drop frames for ports that STP wants entirely killed (both for
1767 * forwarding and for learning). Later, after we do learning, we'll drop
1768 * the frames that STP wants to do learning but not forwarding on. */
1769 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1773 /* Drop frames for reserved multicast addresses. */
1774 if (eth_addr_is_reserved(flow->dl_dst)) {
1778 /* Drop frames on ports reserved for mirroring. */
1779 if (in_port->is_mirror_output_port) {
1780 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1781 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1782 "which is reserved exclusively for mirroring",
1783 br->name, in_port->name);
1787 /* Multicast (and broadcast) packets on bonds need special attention, to
1788 * avoid receiving duplicates. */
1789 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1790 *tags |= in_port->active_iface_tag;
1791 if (in_port->active_iface != in_iface->port_ifidx) {
1792 /* Drop all multicast packets on inactive slaves. */
1795 /* Drop all multicast packets for which we have learned a different
1796 * input port, because we probably sent the packet on one slaves
1797 * and got it back on the active slave. Broadcast ARP replies are
1798 * an exception to this rule: the host has moved to another
1800 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1801 if (src_idx != -1 && src_idx != in_port->port_idx) {
1803 if (!is_bcast_arp_reply(flow, packet)) {
1807 /* No way to know whether it's an ARP reply, because the
1808 * flow entry doesn't include enough information and we
1809 * don't have a packet. Punt. */
1817 out_port = FLOOD_PORT;
1821 /* Learn source MAC (but don't try to learn from revalidation). */
1823 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1824 vlan, in_port->port_idx);
1826 /* The log messages here could actually be useful in debugging,
1827 * so keep the rate limit relatively high. */
1828 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1830 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1831 "on port %s in VLAN %d",
1832 br->name, ETH_ADDR_ARGS(flow->dl_src),
1833 in_port->name, vlan);
1834 ofproto_revalidate(br->ofproto, rev_tag);
1838 /* Determine output port. */
1839 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1841 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1842 out_port = br->ports[out_port_idx];
1846 /* Don't send packets out their input ports. Don't forward frames that STP
1847 * wants us to discard. */
1848 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1853 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1856 * We send out only a single packet, instead of setting up a flow, if the
1857 * packet is an ARP directed to broadcast that arrived on a bonded
1858 * interface. In such a situation ARP requests and replies must be handled
1859 * differently, but OpenFlow unfortunately can't distinguish them.
1861 return (in_port->n_ifaces < 2
1862 || flow->dl_type != htons(ETH_TYPE_ARP)
1863 || !eth_addr_is_broadcast(flow->dl_dst));
1866 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1869 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1870 const struct ofp_phy_port *opp,
1873 struct bridge *br = br_;
1874 struct iface *iface;
1877 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1883 if (reason == OFPPR_DELETE) {
1884 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1885 br->name, iface->name);
1886 iface_destroy(iface);
1887 if (!port->n_ifaces) {
1888 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1889 br->name, port->name);
1895 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1896 if (port->n_ifaces > 1) {
1897 bool up = !(opp->state & OFPPS_LINK_DOWN);
1898 bond_link_status_update(iface, up);
1899 port_update_bond_compat(port);
1905 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1906 struct odp_actions *actions, tag_type *tags, void *br_)
1908 struct bridge *br = br_;
1911 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1912 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1913 brstp_receive(br, flow, payload);
1918 COVERAGE_INC(bridge_process_flow);
1919 return process_flow(br, flow, packet, actions, tags);
1923 bridge_account_flow_ofhook_cb(const flow_t *flow,
1924 const union odp_action *actions,
1925 size_t n_actions, unsigned long long int n_bytes,
1928 struct bridge *br = br_;
1929 const union odp_action *a;
1931 if (!br->has_bonded_ports) {
1935 for (a = actions; a < &actions[n_actions]; a++) {
1936 if (a->type == ODPAT_OUTPUT) {
1937 struct port *port = port_from_dp_ifidx(br, a->output.port);
1938 if (port && port->n_ifaces >= 2) {
1939 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1940 e->tx_bytes += n_bytes;
1947 bridge_account_checkpoint_ofhook_cb(void *br_)
1949 struct bridge *br = br_;
1952 if (!br->has_bonded_ports) {
1956 /* The current ofproto implementation calls this callback at least once a
1957 * second, so this timer implementation is sufficient. */
1958 if (time_msec() < br->bond_next_rebalance) {
1961 br->bond_next_rebalance = time_msec() + 10000;
1963 for (i = 0; i < br->n_ports; i++) {
1964 struct port *port = br->ports[i];
1965 if (port->n_ifaces > 1) {
1966 bond_rebalance_port(port);
1971 static struct ofhooks bridge_ofhooks = {
1972 bridge_port_changed_ofhook_cb,
1973 bridge_normal_ofhook_cb,
1974 bridge_account_flow_ofhook_cb,
1975 bridge_account_checkpoint_ofhook_cb,
1978 /* Bonding functions. */
1980 /* Statistics for a single interface on a bonded port, used for load-based
1981 * bond rebalancing. */
1982 struct slave_balance {
1983 struct iface *iface; /* The interface. */
1984 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
1986 /* All the "bond_entry"s that are assigned to this interface, in order of
1987 * increasing tx_bytes. */
1988 struct bond_entry **hashes;
1992 /* Sorts pointers to pointers to bond_entries in ascending order by the
1993 * interface to which they are assigned, and within a single interface in
1994 * ascending order of bytes transmitted. */
1996 compare_bond_entries(const void *a_, const void *b_)
1998 const struct bond_entry *const *ap = a_;
1999 const struct bond_entry *const *bp = b_;
2000 const struct bond_entry *a = *ap;
2001 const struct bond_entry *b = *bp;
2002 if (a->iface_idx != b->iface_idx) {
2003 return a->iface_idx > b->iface_idx ? 1 : -1;
2004 } else if (a->tx_bytes != b->tx_bytes) {
2005 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2011 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2012 * *descending* order by number of bytes transmitted. */
2014 compare_slave_balance(const void *a_, const void *b_)
2016 const struct slave_balance *a = a_;
2017 const struct slave_balance *b = b_;
2018 if (a->iface->enabled != b->iface->enabled) {
2019 return a->iface->enabled ? -1 : 1;
2020 } else if (a->tx_bytes != b->tx_bytes) {
2021 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2028 swap_bals(struct slave_balance *a, struct slave_balance *b)
2030 struct slave_balance tmp = *a;
2035 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2036 * given that 'p' (and only 'p') might be in the wrong location.
2038 * This function invalidates 'p', since it might now be in a different memory
2041 resort_bals(struct slave_balance *p,
2042 struct slave_balance bals[], size_t n_bals)
2045 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2046 swap_bals(p, p - 1);
2048 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2049 swap_bals(p, p + 1);
2055 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2057 if (VLOG_IS_DBG_ENABLED()) {
2058 struct ds ds = DS_EMPTY_INITIALIZER;
2059 const struct slave_balance *b;
2061 for (b = bals; b < bals + n_bals; b++) {
2065 ds_put_char(&ds, ',');
2067 ds_put_format(&ds, " %s %"PRIu64"kB",
2068 b->iface->name, b->tx_bytes / 1024);
2070 if (!b->iface->enabled) {
2071 ds_put_cstr(&ds, " (disabled)");
2073 if (b->n_hashes > 0) {
2074 ds_put_cstr(&ds, " (");
2075 for (i = 0; i < b->n_hashes; i++) {
2076 const struct bond_entry *e = b->hashes[i];
2078 ds_put_cstr(&ds, " + ");
2080 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2081 e - port->bond_hash, e->tx_bytes / 1024);
2083 ds_put_cstr(&ds, ")");
2086 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2091 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2093 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2094 struct bond_entry *hash)
2096 struct port *port = from->iface->port;
2097 uint64_t delta = hash->tx_bytes;
2099 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2100 "from %s to %s (now carrying %"PRIu64"kB and "
2101 "%"PRIu64"kB load, respectively)",
2102 port->name, delta / 1024, hash - port->bond_hash,
2103 from->iface->name, to->iface->name,
2104 (from->tx_bytes - delta) / 1024,
2105 (to->tx_bytes + delta) / 1024);
2107 /* Delete element from from->hashes.
2109 * We don't bother to add the element to to->hashes because not only would
2110 * it require more work, the only purpose it would be to allow that hash to
2111 * be migrated to another slave in this rebalancing run, and there is no
2112 * point in doing that. */
2113 if (from->hashes[0] == hash) {
2116 int i = hash - from->hashes[0];
2117 memmove(from->hashes + i, from->hashes + i + 1,
2118 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2122 /* Shift load away from 'from' to 'to'. */
2123 from->tx_bytes -= delta;
2124 to->tx_bytes += delta;
2126 /* Arrange for flows to be revalidated. */
2127 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2128 hash->iface_idx = to->iface->port_ifidx;
2129 hash->iface_tag = tag_create_random();
2133 bond_rebalance_port(struct port *port)
2135 struct slave_balance bals[DP_MAX_PORTS];
2137 struct bond_entry *hashes[BOND_MASK + 1];
2138 struct slave_balance *b, *from, *to;
2139 struct bond_entry *e;
2142 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2143 * descending order of tx_bytes, so that bals[0] represents the most
2144 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2147 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2148 * array for each slave_balance structure, we sort our local array of
2149 * hashes in order by slave, so that all of the hashes for a given slave
2150 * become contiguous in memory, and then we point each 'hashes' members of
2151 * a slave_balance structure to the start of a contiguous group. */
2152 n_bals = port->n_ifaces;
2153 for (b = bals; b < &bals[n_bals]; b++) {
2154 b->iface = port->ifaces[b - bals];
2159 for (i = 0; i <= BOND_MASK; i++) {
2160 hashes[i] = &port->bond_hash[i];
2162 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2163 for (i = 0; i <= BOND_MASK; i++) {
2165 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2166 b = &bals[e->iface_idx];
2167 b->tx_bytes += e->tx_bytes;
2169 b->hashes = &hashes[i];
2174 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2175 log_bals(bals, n_bals, port);
2177 /* Discard slaves that aren't enabled (which were sorted to the back of the
2178 * array earlier). */
2179 while (!bals[n_bals - 1].iface->enabled) {
2186 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2187 to = &bals[n_bals - 1];
2188 for (from = bals; from < to; ) {
2189 uint64_t overload = from->tx_bytes - to->tx_bytes;
2190 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2191 /* The extra load on 'from' (and all less-loaded slaves), compared
2192 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2193 * it is less than ~1Mbps. No point in rebalancing. */
2195 } else if (from->n_hashes == 1) {
2196 /* 'from' only carries a single MAC hash, so we can't shift any
2197 * load away from it, even though we want to. */
2200 /* 'from' is carrying significantly more load than 'to', and that
2201 * load is split across at least two different hashes. Pick a hash
2202 * to migrate to 'to' (the least-loaded slave), given that doing so
2203 * must not cause 'to''s load to exceed 'from''s load.
2205 * The sort order we use means that we prefer to shift away the
2206 * smallest hashes instead of the biggest ones. There is little
2207 * reason behind this decision; we could use the opposite sort
2208 * order to shift away big hashes ahead of small ones. */
2211 for (i = 0; i < from->n_hashes; i++) {
2212 uint64_t delta = from->hashes[i]->tx_bytes;
2213 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2217 if (i < from->n_hashes) {
2218 bond_shift_load(from, to, from->hashes[i]);
2220 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2221 * point to different slave_balance structures. It is only
2222 * valid to do these two operations in a row at all because we
2223 * know that 'from' will not move past 'to' and vice versa. */
2224 resort_bals(from, bals, n_bals);
2225 resort_bals(to, bals, n_bals);
2232 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2233 * historical data to decay to <1% in 7 rebalancing runs. */
2234 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2240 bond_send_learning_packets(struct port *port)
2242 struct bridge *br = port->bridge;
2243 struct mac_entry *e;
2244 struct ofpbuf packet;
2245 int error, n_packets, n_errors;
2247 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2251 ofpbuf_init(&packet, 128);
2252 error = n_packets = n_errors = 0;
2253 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2254 static const char s[] = "Open vSwitch Bond Failover";
2255 union ofp_action actions[2], *a;
2256 struct eth_header *eth;
2257 struct llc_snap_header *llc_snap;
2263 if (e->port == port->port_idx
2264 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2268 /* Compose packet to send. */
2269 ofpbuf_clear(&packet);
2270 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2271 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2272 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2273 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2275 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2276 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2277 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2279 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2280 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2281 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2282 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2283 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2285 /* Compose actions. */
2286 memset(actions, 0, sizeof actions);
2289 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2290 a->vlan_vid.len = htons(sizeof *a);
2291 a->vlan_vid.vlan_vid = htons(e->vlan);
2294 a->output.type = htons(OFPAT_OUTPUT);
2295 a->output.len = htons(sizeof *a);
2296 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2301 flow_extract(&packet, ODPP_NONE, &flow);
2302 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2309 ofpbuf_uninit(&packet);
2312 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2313 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2314 "packets, last error was: %s",
2315 port->name, n_errors, n_packets, strerror(error));
2317 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2318 port->name, n_packets);
2322 /* Bonding unixctl user interface functions. */
2325 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2327 struct ds ds = DS_EMPTY_INITIALIZER;
2328 const struct bridge *br;
2330 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2332 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2335 for (i = 0; i < br->n_ports; i++) {
2336 const struct port *port = br->ports[i];
2337 if (port->n_ifaces > 1) {
2340 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2341 for (j = 0; j < port->n_ifaces; j++) {
2342 const struct iface *iface = port->ifaces[j];
2344 ds_put_cstr(&ds, ", ");
2346 ds_put_cstr(&ds, iface->name);
2348 ds_put_char(&ds, '\n');
2352 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2356 static struct port *
2357 bond_find(const char *name)
2359 const struct bridge *br;
2361 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2364 for (i = 0; i < br->n_ports; i++) {
2365 struct port *port = br->ports[i];
2366 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2375 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2377 struct ds ds = DS_EMPTY_INITIALIZER;
2378 const struct port *port;
2381 port = bond_find(args);
2383 unixctl_command_reply(conn, 501, "no such bond");
2387 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2388 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2389 ds_put_format(&ds, "next rebalance: %lld ms\n",
2390 port->bridge->bond_next_rebalance - time_msec());
2391 for (j = 0; j < port->n_ifaces; j++) {
2392 const struct iface *iface = port->ifaces[j];
2393 struct bond_entry *be;
2396 ds_put_format(&ds, "slave %s: %s\n",
2397 iface->name, iface->enabled ? "enabled" : "disabled");
2398 if (j == port->active_iface) {
2399 ds_put_cstr(&ds, "\tactive slave\n");
2401 if (iface->delay_expires != LLONG_MAX) {
2402 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2403 iface->enabled ? "downdelay" : "updelay",
2404 iface->delay_expires - time_msec());
2408 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2409 int hash = be - port->bond_hash;
2410 struct mac_entry *me;
2412 if (be->iface_idx != j) {
2416 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2417 hash, be->tx_bytes / 1024);
2420 if (!port->bridge->ml) {
2424 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2425 &port->bridge->ml->lrus) {
2428 if (bond_hash(me->mac) == hash
2429 && me->port != port->port_idx
2430 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2431 && dp_ifidx == iface->dp_ifidx)
2433 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2434 ETH_ADDR_ARGS(me->mac));
2439 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2444 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2446 char *args = (char *) args_;
2447 char *save_ptr = NULL;
2448 char *bond_s, *hash_s, *slave_s;
2449 uint8_t mac[ETH_ADDR_LEN];
2451 struct iface *iface;
2452 struct bond_entry *entry;
2455 bond_s = strtok_r(args, " ", &save_ptr);
2456 hash_s = strtok_r(NULL, " ", &save_ptr);
2457 slave_s = strtok_r(NULL, " ", &save_ptr);
2459 unixctl_command_reply(conn, 501,
2460 "usage: bond/migrate BOND HASH SLAVE");
2464 port = bond_find(bond_s);
2466 unixctl_command_reply(conn, 501, "no such bond");
2470 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2471 == ETH_ADDR_SCAN_COUNT) {
2472 hash = bond_hash(mac);
2473 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2474 hash = atoi(hash_s) & BOND_MASK;
2476 unixctl_command_reply(conn, 501, "bad hash");
2480 iface = port_lookup_iface(port, slave_s);
2482 unixctl_command_reply(conn, 501, "no such slave");
2486 if (!iface->enabled) {
2487 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2491 entry = &port->bond_hash[hash];
2492 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2493 entry->iface_idx = iface->port_ifidx;
2494 entry->iface_tag = tag_create_random();
2495 unixctl_command_reply(conn, 200, "migrated");
2499 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2501 char *args = (char *) args_;
2502 char *save_ptr = NULL;
2503 char *bond_s, *slave_s;
2505 struct iface *iface;
2507 bond_s = strtok_r(args, " ", &save_ptr);
2508 slave_s = strtok_r(NULL, " ", &save_ptr);
2510 unixctl_command_reply(conn, 501,
2511 "usage: bond/set-active-slave BOND SLAVE");
2515 port = bond_find(bond_s);
2517 unixctl_command_reply(conn, 501, "no such bond");
2521 iface = port_lookup_iface(port, slave_s);
2523 unixctl_command_reply(conn, 501, "no such slave");
2527 if (!iface->enabled) {
2528 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2532 if (port->active_iface != iface->port_ifidx) {
2533 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2534 port->active_iface = iface->port_ifidx;
2535 port->active_iface_tag = tag_create_random();
2536 VLOG_INFO("port %s: active interface is now %s",
2537 port->name, iface->name);
2538 bond_send_learning_packets(port);
2539 unixctl_command_reply(conn, 200, "done");
2541 unixctl_command_reply(conn, 200, "no change");
2546 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2548 char *args = (char *) args_;
2549 char *save_ptr = NULL;
2550 char *bond_s, *slave_s;
2552 struct iface *iface;
2554 bond_s = strtok_r(args, " ", &save_ptr);
2555 slave_s = strtok_r(NULL, " ", &save_ptr);
2557 unixctl_command_reply(conn, 501,
2558 "usage: bond/enable/disable-slave BOND SLAVE");
2562 port = bond_find(bond_s);
2564 unixctl_command_reply(conn, 501, "no such bond");
2568 iface = port_lookup_iface(port, slave_s);
2570 unixctl_command_reply(conn, 501, "no such slave");
2574 bond_enable_slave(iface, enable);
2575 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2579 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2581 enable_slave(conn, args, true);
2585 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2587 enable_slave(conn, args, false);
2593 unixctl_command_register("bond/list", bond_unixctl_list);
2594 unixctl_command_register("bond/show", bond_unixctl_show);
2595 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2596 unixctl_command_register("bond/set-active-slave",
2597 bond_unixctl_set_active_slave);
2598 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2599 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2602 /* Port functions. */
2605 port_create(struct bridge *br, const char *name)
2609 port = xcalloc(1, sizeof *port);
2611 port->port_idx = br->n_ports;
2613 port->trunks = NULL;
2614 port->name = xstrdup(name);
2615 port->active_iface = -1;
2616 port->stp_state = STP_DISABLED;
2617 port->stp_state_tag = 0;
2619 if (br->n_ports >= br->allocated_ports) {
2620 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2623 br->ports[br->n_ports++] = port;
2625 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2630 port_reconfigure(struct port *port)
2632 bool bonded = cfg_has_section("bonding.%s", port->name);
2633 struct svec old_ifaces, new_ifaces;
2634 unsigned long *trunks;
2638 /* Collect old and new interfaces. */
2639 svec_init(&old_ifaces);
2640 svec_init(&new_ifaces);
2641 for (i = 0; i < port->n_ifaces; i++) {
2642 svec_add(&old_ifaces, port->ifaces[i]->name);
2644 svec_sort(&old_ifaces);
2646 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2647 if (!new_ifaces.n) {
2648 VLOG_ERR("port %s: no interfaces specified for bonded port",
2650 } else if (new_ifaces.n == 1) {
2651 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2655 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2656 if (port->updelay < 0) {
2659 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2660 if (port->downdelay < 0) {
2661 port->downdelay = 0;
2664 svec_init(&new_ifaces);
2665 svec_add(&new_ifaces, port->name);
2668 /* Get rid of deleted interfaces and add new interfaces. */
2669 for (i = 0; i < port->n_ifaces; i++) {
2670 struct iface *iface = port->ifaces[i];
2671 if (!svec_contains(&new_ifaces, iface->name)) {
2672 iface_destroy(iface);
2677 for (i = 0; i < new_ifaces.n; i++) {
2678 const char *name = new_ifaces.names[i];
2679 if (!svec_contains(&old_ifaces, name)) {
2680 iface_create(port, name);
2686 if (cfg_has("vlan.%s.tag", port->name)) {
2688 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2689 if (vlan >= 0 && vlan <= 4095) {
2690 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2693 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2694 * they even work as-is. But they have not been tested. */
2695 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2699 if (port->vlan != vlan) {
2701 bridge_flush(port->bridge);
2704 /* Get trunked VLANs. */
2707 size_t n_trunks, n_errors;
2710 trunks = bitmap_allocate(4096);
2711 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2713 for (i = 0; i < n_trunks; i++) {
2714 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2716 bitmap_set1(trunks, trunk);
2722 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2723 port->name, n_trunks);
2725 if (n_errors == n_trunks) {
2727 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2730 bitmap_set_multiple(trunks, 0, 4096, 1);
2733 if (cfg_has("vlan.%s.trunks", port->name)) {
2734 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2735 port->name, port->name);
2739 ? port->trunks != NULL
2740 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2741 bridge_flush(port->bridge);
2743 bitmap_free(port->trunks);
2744 port->trunks = trunks;
2746 svec_destroy(&old_ifaces);
2747 svec_destroy(&new_ifaces);
2751 port_destroy(struct port *port)
2754 struct bridge *br = port->bridge;
2758 proc_net_compat_update_vlan(port->name, NULL, 0);
2760 for (i = 0; i < MAX_MIRRORS; i++) {
2761 struct mirror *m = br->mirrors[i];
2762 if (m && m->out_port == port) {
2767 while (port->n_ifaces > 0) {
2768 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2771 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2772 del->port_idx = port->port_idx;
2775 bitmap_free(port->trunks);
2782 static struct port *
2783 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2785 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2786 return iface ? iface->port : NULL;
2789 static struct port *
2790 port_lookup(const struct bridge *br, const char *name)
2794 for (i = 0; i < br->n_ports; i++) {
2795 struct port *port = br->ports[i];
2796 if (!strcmp(port->name, name)) {
2803 static struct iface *
2804 port_lookup_iface(const struct port *port, const char *name)
2808 for (j = 0; j < port->n_ifaces; j++) {
2809 struct iface *iface = port->ifaces[j];
2810 if (!strcmp(iface->name, name)) {
2818 port_update_bonding(struct port *port)
2820 if (port->n_ifaces < 2) {
2821 /* Not a bonded port. */
2822 if (port->bond_hash) {
2823 free(port->bond_hash);
2824 port->bond_hash = NULL;
2825 proc_net_compat_update_bond(port->name, NULL);
2828 if (!port->bond_hash) {
2831 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2832 for (i = 0; i <= BOND_MASK; i++) {
2833 struct bond_entry *e = &port->bond_hash[i];
2837 port->no_ifaces_tag = tag_create_random();
2838 bond_choose_active_iface(port);
2840 port_update_bond_compat(port);
2845 port_update_bond_compat(struct port *port)
2847 struct compat_bond bond;
2850 if (port->n_ifaces < 2) {
2855 bond.updelay = port->updelay;
2856 bond.downdelay = port->downdelay;
2857 bond.n_slaves = port->n_ifaces;
2858 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2859 for (i = 0; i < port->n_ifaces; i++) {
2860 struct iface *iface = port->ifaces[i];
2861 struct compat_bond_slave *slave = &bond.slaves[i];
2862 slave->name = iface->name;
2863 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2864 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2868 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2870 proc_net_compat_update_bond(port->name, &bond);
2875 port_update_vlan_compat(struct port *port)
2877 struct bridge *br = port->bridge;
2878 char *vlandev_name = NULL;
2880 if (port->vlan > 0) {
2881 /* Figure out the name that the VLAN device should actually have, if it
2882 * existed. This takes some work because the VLAN device would not
2883 * have port->name in its name; rather, it would have the trunk port's
2884 * name, and 'port' would be attached to a bridge that also had the
2885 * VLAN device one of its ports. So we need to find a trunk port that
2886 * includes port->vlan.
2888 * There might be more than one candidate. This doesn't happen on
2889 * XenServer, so if it happens we just pick the first choice in
2890 * alphabetical order instead of creating multiple VLAN devices. */
2892 for (i = 0; i < br->n_ports; i++) {
2893 struct port *p = br->ports[i];
2894 if (port_trunks_vlan(p, port->vlan)
2896 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2898 const uint8_t *ea = p->ifaces[0]->mac;
2899 if (!eth_addr_is_multicast(ea) &&
2900 !eth_addr_is_reserved(ea) &&
2901 !eth_addr_is_zero(ea)) {
2902 vlandev_name = p->name;
2907 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2910 /* Interface functions. */
2913 iface_create(struct port *port, const char *name)
2915 struct iface *iface;
2917 iface = xcalloc(1, sizeof *iface);
2919 iface->port_ifidx = port->n_ifaces;
2920 iface->name = xstrdup(name);
2921 iface->dp_ifidx = -1;
2922 iface->tag = tag_create_random();
2923 iface->delay_expires = LLONG_MAX;
2925 netdev_nodev_get_etheraddr(name, iface->mac);
2926 netdev_nodev_get_carrier(name, &iface->enabled);
2928 if (port->n_ifaces >= port->allocated_ifaces) {
2929 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2930 sizeof *port->ifaces);
2932 port->ifaces[port->n_ifaces++] = iface;
2933 if (port->n_ifaces > 1) {
2934 port->bridge->has_bonded_ports = true;
2937 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2939 port_update_bonding(port);
2940 bridge_flush(port->bridge);
2944 iface_destroy(struct iface *iface)
2947 struct port *port = iface->port;
2948 struct bridge *br = port->bridge;
2949 bool del_active = port->active_iface == iface->port_ifidx;
2952 if (iface->dp_ifidx >= 0) {
2953 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
2956 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
2957 del->port_ifidx = iface->port_ifidx;
2963 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2964 bond_choose_active_iface(port);
2965 bond_send_learning_packets(port);
2968 port_update_bonding(port);
2969 bridge_flush(port->bridge);
2973 static struct iface *
2974 iface_lookup(const struct bridge *br, const char *name)
2978 for (i = 0; i < br->n_ports; i++) {
2979 struct port *port = br->ports[i];
2980 for (j = 0; j < port->n_ifaces; j++) {
2981 struct iface *iface = port->ifaces[j];
2982 if (!strcmp(iface->name, name)) {
2990 static struct iface *
2991 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2993 return port_array_get(&br->ifaces, dp_ifidx);
2996 /* Port mirroring. */
2999 mirror_reconfigure(struct bridge *br)
3001 struct svec old_mirrors, new_mirrors;
3004 /* Collect old and new mirrors. */
3005 svec_init(&old_mirrors);
3006 svec_init(&new_mirrors);
3007 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3008 for (i = 0; i < MAX_MIRRORS; i++) {
3009 if (br->mirrors[i]) {
3010 svec_add(&old_mirrors, br->mirrors[i]->name);
3014 /* Get rid of deleted mirrors and add new mirrors. */
3015 svec_sort(&old_mirrors);
3016 assert(svec_is_unique(&old_mirrors));
3017 svec_sort(&new_mirrors);
3018 assert(svec_is_unique(&new_mirrors));
3019 for (i = 0; i < MAX_MIRRORS; i++) {
3020 struct mirror *m = br->mirrors[i];
3021 if (m && !svec_contains(&new_mirrors, m->name)) {
3025 for (i = 0; i < new_mirrors.n; i++) {
3026 const char *name = new_mirrors.names[i];
3027 if (!svec_contains(&old_mirrors, name)) {
3028 mirror_create(br, name);
3031 svec_destroy(&old_mirrors);
3032 svec_destroy(&new_mirrors);
3034 /* Reconfigure all mirrors. */
3035 for (i = 0; i < MAX_MIRRORS; i++) {
3036 if (br->mirrors[i]) {
3037 mirror_reconfigure_one(br->mirrors[i]);
3041 /* Update port reserved status. */
3042 for (i = 0; i < br->n_ports; i++) {
3043 br->ports[i]->is_mirror_output_port = false;
3045 for (i = 0; i < MAX_MIRRORS; i++) {
3046 struct mirror *m = br->mirrors[i];
3047 if (m && m->out_port) {
3048 m->out_port->is_mirror_output_port = true;
3054 mirror_create(struct bridge *br, const char *name)
3059 for (i = 0; ; i++) {
3060 if (i >= MAX_MIRRORS) {
3061 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3062 "cannot create %s", br->name, MAX_MIRRORS, name);
3065 if (!br->mirrors[i]) {
3070 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3073 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3076 m->name = xstrdup(name);
3077 svec_init(&m->src_ports);
3078 svec_init(&m->dst_ports);
3086 mirror_destroy(struct mirror *m)
3089 struct bridge *br = m->bridge;
3092 for (i = 0; i < br->n_ports; i++) {
3093 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3094 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3097 svec_destroy(&m->src_ports);
3098 svec_destroy(&m->dst_ports);
3101 m->bridge->mirrors[m->idx] = NULL;
3109 prune_ports(struct mirror *m, struct svec *ports)
3114 svec_sort_unique(ports);
3117 for (i = 0; i < ports->n; i++) {
3118 const char *name = ports->names[i];
3119 if (port_lookup(m->bridge, name)) {
3120 svec_add(&tmp, name);
3122 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3123 m->bridge->name, m->name, name);
3126 svec_swap(ports, &tmp);
3131 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3135 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3136 * order won't give us numeric sort order. But that's good enough for what
3137 * we need right now. */
3138 svec_sort_unique(vlan_strings);
3140 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3142 for (i = 0; i < vlan_strings->n; i++) {
3143 const char *name = vlan_strings->names[i];
3145 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3146 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3147 m->bridge->name, m->name, name);
3149 (*vlans)[n_vlans++] = vlan;
3156 vlan_is_mirrored(const struct mirror *m, int vlan)
3160 for (i = 0; i < m->n_vlans; i++) {
3161 if (m->vlans[i] == vlan) {
3169 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3173 for (i = 0; i < m->n_vlans; i++) {
3174 if (port_trunks_vlan(p, m->vlans[i])) {
3182 mirror_reconfigure_one(struct mirror *m)
3184 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3185 struct svec src_ports, dst_ports, ports;
3186 struct svec vlan_strings;
3187 mirror_mask_t mirror_bit;
3188 const char *out_port_name;
3189 struct port *out_port;
3194 bool mirror_all_ports;
3196 /* Get output port. */
3197 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3198 m->bridge->name, m->name);
3199 if (out_port_name) {
3200 out_port = port_lookup(m->bridge, out_port_name);
3202 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3203 "named %s", pfx, m->bridge->name, out_port_name);
3210 if (cfg_has("%s.output.vlan", pfx)) {
3211 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3212 "ignoring %s.output.vlan", pfx, pfx, pfx);
3214 } else if (cfg_has("%s.output.vlan", pfx)) {
3216 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3218 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3219 "but exactly one is required; disabling port mirror %s",
3220 pfx, pfx, pfx, pfx);
3226 /* Get all the ports, and drop duplicates and ports that don't exist. */
3227 svec_init(&src_ports);
3228 svec_init(&dst_ports);
3230 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3231 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3232 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3233 svec_append(&src_ports, &ports);
3234 svec_append(&dst_ports, &ports);
3235 svec_destroy(&ports);
3236 prune_ports(m, &src_ports);
3237 prune_ports(m, &dst_ports);
3239 /* Get all the vlans, and drop duplicate and invalid vlans. */
3240 svec_init(&vlan_strings);
3241 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3242 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3243 svec_destroy(&vlan_strings);
3245 /* Update mirror data. */
3246 if (!svec_equal(&m->src_ports, &src_ports)
3247 || !svec_equal(&m->dst_ports, &dst_ports)
3248 || m->n_vlans != n_vlans
3249 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3250 || m->out_port != out_port
3251 || m->out_vlan != out_vlan) {
3252 bridge_flush(m->bridge);
3254 svec_swap(&m->src_ports, &src_ports);
3255 svec_swap(&m->dst_ports, &dst_ports);
3258 m->n_vlans = n_vlans;
3259 m->out_port = out_port;
3260 m->out_vlan = out_vlan;
3262 /* If no selection criteria have been given, mirror for all ports. */
3263 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3266 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3267 for (i = 0; i < m->bridge->n_ports; i++) {
3268 struct port *port = m->bridge->ports[i];
3270 if (mirror_all_ports
3271 || svec_contains(&m->src_ports, port->name)
3274 ? port_trunks_any_mirrored_vlan(m, port)
3275 : vlan_is_mirrored(m, port->vlan)))) {
3276 port->src_mirrors |= mirror_bit;
3278 port->src_mirrors &= ~mirror_bit;
3281 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3282 port->dst_mirrors |= mirror_bit;
3284 port->dst_mirrors &= ~mirror_bit;
3289 svec_destroy(&src_ports);
3290 svec_destroy(&dst_ports);
3294 /* Spanning tree protocol. */
3296 static void brstp_update_port_state(struct port *);
3299 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3301 struct bridge *br = br_;
3302 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3303 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3305 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3307 } else if (eth_addr_is_zero(iface->mac)) {
3308 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3311 union ofp_action action;
3312 struct eth_header *eth = pkt->l2;
3315 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3317 memset(&action, 0, sizeof action);
3318 action.type = htons(OFPAT_OUTPUT);
3319 action.output.len = htons(sizeof action);
3320 action.output.port = htons(port_no);
3322 flow_extract(pkt, ODPP_NONE, &flow);
3323 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3329 brstp_reconfigure(struct bridge *br)
3333 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3335 stp_destroy(br->stp);
3341 uint64_t bridge_address, bridge_id;
3342 int bridge_priority;
3344 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3345 if (!bridge_address) {
3347 bridge_address = (stp_get_bridge_id(br->stp)
3348 & ((UINT64_C(1) << 48) - 1));
3350 uint8_t mac[ETH_ADDR_LEN];
3351 eth_addr_random(mac);
3352 bridge_address = eth_addr_to_uint64(mac);
3356 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3358 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3360 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3363 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3365 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3366 br->stp_last_tick = time_msec();
3369 if (bridge_id != stp_get_bridge_id(br->stp)) {
3370 stp_set_bridge_id(br->stp, bridge_id);
3375 for (i = 0; i < br->n_ports; i++) {
3376 struct port *p = br->ports[i];
3378 struct stp_port *sp;
3379 int path_cost, priority;
3385 dp_ifidx = p->ifaces[0]->dp_ifidx;
3386 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3390 sp = stp_get_port(br->stp, dp_ifidx);
3391 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3392 "stp.%s.port.%s.enabled",
3394 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3395 br->name, p->name));
3396 if (p->is_mirror_output_port) {
3399 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3400 bridge_flush(br); /* Might not be necessary. */
3402 stp_port_enable(sp);
3404 stp_port_disable(sp);
3408 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3410 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3412 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3413 "stp.%s.port.%s.priority",
3415 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3417 : STP_DEFAULT_PORT_PRIORITY);
3418 stp_port_set_priority(sp, priority);
3421 brstp_adjust_timers(br);
3423 for (i = 0; i < br->n_ports; i++) {
3424 brstp_update_port_state(br->ports[i]);
3429 brstp_update_port_state(struct port *p)
3431 struct bridge *br = p->bridge;
3432 enum stp_state state;
3434 /* Figure out new state. */
3435 state = STP_DISABLED;
3436 if (br->stp && p->n_ifaces > 0) {
3437 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3438 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3439 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3444 if (p->stp_state != state) {
3445 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3446 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3447 p->name, stp_state_name(p->stp_state),
3448 stp_state_name(state));
3449 if (p->stp_state == STP_DISABLED) {
3452 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3454 p->stp_state = state;
3455 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3456 : tag_create_random());
3461 brstp_adjust_timers(struct bridge *br)
3463 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3464 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3465 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3467 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3468 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3469 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3473 brstp_run(struct bridge *br)
3476 long long int now = time_msec();
3477 long long int elapsed = now - br->stp_last_tick;
3478 struct stp_port *sp;
3481 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3482 br->stp_last_tick = now;
3484 while (stp_get_changed_port(br->stp, &sp)) {
3485 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3487 brstp_update_port_state(p);
3494 brstp_wait(struct bridge *br)
3497 poll_timer_wait(1000);