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 bond_init(void);
210 static void bond_run(struct bridge *);
211 static void bond_wait(struct bridge *);
212 static void bond_rebalance_port(struct port *);
213 static void bond_send_learning_packets(struct port *);
215 static void port_create(struct bridge *, const char *name);
216 static void port_reconfigure(struct port *);
217 static void port_destroy(struct port *);
218 static struct port *port_lookup(const struct bridge *, const char *name);
219 static struct iface *port_lookup_iface(const struct port *, const char *name);
220 static struct port *port_from_dp_ifidx(const struct bridge *,
222 static void port_update_bond_compat(struct port *);
223 static void port_update_vlan_compat(struct port *);
225 static void mirror_create(struct bridge *, const char *name);
226 static void mirror_destroy(struct mirror *);
227 static void mirror_reconfigure(struct bridge *);
228 static void mirror_reconfigure_one(struct mirror *);
229 static bool vlan_is_mirrored(const struct mirror *, int vlan);
231 static void brstp_reconfigure(struct bridge *);
232 static void brstp_adjust_timers(struct bridge *);
233 static void brstp_run(struct bridge *);
234 static void brstp_wait(struct bridge *);
236 static void iface_create(struct port *, const char *name);
237 static void iface_destroy(struct iface *);
238 static struct iface *iface_lookup(const struct bridge *, const char *name);
239 static struct iface *iface_from_dp_ifidx(const struct bridge *,
242 /* Hooks into ofproto processing. */
243 static struct ofhooks bridge_ofhooks;
245 /* Public functions. */
247 /* Adds the name of each interface used by a bridge, including local and
248 * internal ports, to 'svec'. */
250 bridge_get_ifaces(struct svec *svec)
252 struct bridge *br, *next;
255 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
256 for (i = 0; i < br->n_ports; i++) {
257 struct port *port = br->ports[i];
259 for (j = 0; j < port->n_ifaces; j++) {
260 struct iface *iface = port->ifaces[j];
261 if (iface->dp_ifidx < 0) {
262 VLOG_ERR("%s interface not in dp%u, ignoring",
263 iface->name, dpif_id(&br->dpif));
265 if (iface->dp_ifidx != ODPP_LOCAL) {
266 svec_add(svec, iface->name);
274 /* The caller must already have called cfg_read(). */
283 for (i = 0; i < DP_MAX; i++) {
287 sprintf(devname, "dp%d", i);
288 retval = dpif_open(devname, &dpif);
290 char dpif_name[IF_NAMESIZE];
291 if (dpif_get_name(&dpif, dpif_name, sizeof dpif_name)
292 || !cfg_has("bridge.%s.port", dpif_name)) {
296 } else if (retval != ENODEV) {
297 VLOG_ERR("failed to delete datapath dp%d: %s",
298 i, strerror(retval));
302 bridge_reconfigure();
307 config_string_change(const char *key, char **valuep)
309 const char *value = cfg_get_string(0, "%s", key);
310 if (value && (!*valuep || strcmp(value, *valuep))) {
312 *valuep = xstrdup(value);
320 bridge_configure_ssl(void)
322 /* XXX SSL should be configurable on a per-bridge basis.
323 * XXX should be possible to de-configure SSL. */
324 static char *private_key_file;
325 static char *certificate_file;
326 static char *cacert_file;
329 if (config_string_change("ssl.private-key", &private_key_file)) {
330 vconn_ssl_set_private_key_file(private_key_file);
333 if (config_string_change("ssl.certificate", &certificate_file)) {
334 vconn_ssl_set_certificate_file(certificate_file);
337 /* We assume that even if the filename hasn't changed, if the CA cert
338 * file has been removed, that we want to move back into
339 * boot-strapping mode. This opens a small security hole, because
340 * the old certificate will still be trusted until vSwitch is
341 * restarted. We may want to address this in vconn's SSL library. */
342 if (config_string_change("ssl.ca-cert", &cacert_file)
343 || (stat(cacert_file, &s) && errno == ENOENT)) {
344 vconn_ssl_set_ca_cert_file(cacert_file,
345 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
351 bridge_reconfigure(void)
353 struct svec old_br, new_br, raw_new_br;
354 struct bridge *br, *next;
357 COVERAGE_INC(bridge_reconfigure);
359 /* Collect old bridges. */
361 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
362 svec_add(&old_br, br->name);
365 /* Collect new bridges. */
366 svec_init(&raw_new_br);
367 cfg_get_subsections(&raw_new_br, "bridge");
369 for (i = 0; i < raw_new_br.n; i++) {
370 const char *name = raw_new_br.names[i];
371 if ((!strncmp(name, "dp", 2) && isdigit(name[2])) ||
372 (!strncmp(name, "nl:", 3) && isdigit(name[3]))) {
373 VLOG_ERR("%s is not a valid bridge name (bridges may not be "
374 "named \"dp\" or \"nl:\" followed by a digit)", name);
376 svec_add(&new_br, name);
379 svec_destroy(&raw_new_br);
381 /* Get rid of deleted bridges and add new bridges. */
384 assert(svec_is_unique(&old_br));
385 assert(svec_is_unique(&new_br));
386 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
387 if (!svec_contains(&new_br, br->name)) {
391 for (i = 0; i < new_br.n; i++) {
392 const char *name = new_br.names[i];
393 if (!svec_contains(&old_br, name)) {
397 svec_destroy(&old_br);
398 svec_destroy(&new_br);
402 bridge_configure_ssl();
405 /* Reconfigure all bridges. */
406 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
407 bridge_reconfigure_one(br);
410 /* Add and delete ports on all datapaths.
412 * The kernel will reject any attempt to add a given port to a datapath if
413 * that port already belongs to a different datapath, so we must do all
414 * port deletions before any port additions. */
415 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
416 struct odp_port *dpif_ports;
418 struct svec want_ifaces;
420 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
421 bridge_get_all_ifaces(br, &want_ifaces);
422 for (i = 0; i < n_dpif_ports; i++) {
423 const struct odp_port *p = &dpif_ports[i];
424 if (!svec_contains(&want_ifaces, p->devname)
425 && strcmp(p->devname, br->name)) {
426 int retval = dpif_port_del(&br->dpif, p->port);
428 VLOG_ERR("failed to remove %s interface from dp%u: %s",
429 p->devname, dpif_id(&br->dpif), strerror(retval));
433 svec_destroy(&want_ifaces);
436 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
437 struct odp_port *dpif_ports;
439 struct svec cur_ifaces, want_ifaces, add_ifaces;
442 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
443 svec_init(&cur_ifaces);
444 for (i = 0; i < n_dpif_ports; i++) {
445 svec_add(&cur_ifaces, dpif_ports[i].devname);
448 svec_sort_unique(&cur_ifaces);
449 bridge_get_all_ifaces(br, &want_ifaces);
450 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
453 for (i = 0; i < add_ifaces.n; i++) {
454 const char *if_name = add_ifaces.names[i];
456 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
457 int error = dpif_port_add(&br->dpif, if_name, next_port_no++,
458 internal ? ODP_PORT_INTERNAL : 0);
459 if (error != EEXIST) {
460 if (next_port_no >= 256) {
461 VLOG_ERR("ran out of valid port numbers on dp%u",
466 VLOG_ERR("failed to add %s interface to dp%u: %s",
467 if_name, dpif_id(&br->dpif), strerror(error));
474 svec_destroy(&cur_ifaces);
475 svec_destroy(&want_ifaces);
476 svec_destroy(&add_ifaces);
478 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
481 struct iface *local_iface = NULL;
483 uint8_t engine_type = br->dpif.minor;
484 uint8_t engine_id = br->dpif.minor;
485 bool add_id_to_iface = false;
486 struct svec nf_hosts;
488 bridge_fetch_dp_ifaces(br);
489 for (i = 0; i < br->n_ports; ) {
490 struct port *port = br->ports[i];
492 for (j = 0; j < port->n_ifaces; ) {
493 struct iface *iface = port->ifaces[j];
494 if (iface->dp_ifidx < 0) {
495 VLOG_ERR("%s interface not in dp%u, dropping",
496 iface->name, dpif_id(&br->dpif));
497 iface_destroy(iface);
499 if (iface->dp_ifidx == ODPP_LOCAL) {
502 VLOG_DBG("dp%u has interface %s on port %d",
503 dpif_id(&br->dpif), iface->name, iface->dp_ifidx);
507 if (!port->n_ifaces) {
508 VLOG_ERR("%s port has no interfaces, dropping", port->name);
515 /* Pick local port hardware address, datapath ID. */
516 bridge_pick_local_hw_addr(br, ea, &devname);
518 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
520 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
521 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
522 "Ethernet address: %s",
523 br->name, strerror(error));
527 dpid = bridge_pick_datapath_id(br, ea, devname);
528 ofproto_set_datapath_id(br->ofproto, dpid);
530 /* Set NetFlow configuration on this bridge. */
531 if (cfg_has("netflow.%s.engine-type", br->name)) {
532 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
535 if (cfg_has("netflow.%s.engine-id", br->name)) {
536 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
538 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
539 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
542 if (add_id_to_iface && engine_id > 0x7f) {
543 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
544 "another vswitch, choose an engine id less than 128",
547 if (add_id_to_iface && br->n_ports > 0x1ff) {
548 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
549 "another port when 512 or more ports are used",
552 svec_init(&nf_hosts);
553 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
554 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
555 engine_id, add_id_to_iface)) {
556 VLOG_ERR("bridge %s: problem setting netflow collectors",
560 /* Update the controller and related settings. It would be more
561 * straightforward to call this from bridge_reconfigure_one(), but we
562 * can't do it there for two reasons. First, and most importantly, at
563 * that point we don't know the dp_ifidx of any interfaces that have
564 * been added to the bridge (because we haven't actually added them to
565 * the datapath). Second, at that point we haven't set the datapath ID
566 * yet; when a controller is configured, resetting the datapath ID will
567 * immediately disconnect from the controller, so it's better to set
568 * the datapath ID before the controller. */
569 bridge_reconfigure_controller(br);
571 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
572 for (i = 0; i < br->n_ports; i++) {
573 struct port *port = br->ports[i];
574 port_update_vlan_compat(port);
577 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
578 brstp_reconfigure(br);
583 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
584 const char **devname)
586 uint64_t requested_ea;
592 /* Did the user request a particular MAC? */
593 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
595 eth_addr_from_uint64(requested_ea, ea);
596 if (eth_addr_is_multicast(ea)) {
597 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
598 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
599 } else if (eth_addr_is_zero(ea)) {
600 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
606 /* Otherwise choose the minimum MAC address among all of the interfaces.
607 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
608 * MAC of the physical interface in such an environment.) */
609 memset(ea, 0xff, sizeof ea);
610 for (i = 0; i < br->n_ports; i++) {
611 struct port *port = br->ports[i];
612 if (port->is_mirror_output_port) {
615 for (j = 0; j < port->n_ifaces; j++) {
616 struct iface *iface = port->ifaces[j];
617 uint8_t iface_ea[ETH_ADDR_LEN];
618 if (iface->dp_ifidx == ODPP_LOCAL
619 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
622 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
624 if (!eth_addr_is_multicast(iface_ea) &&
625 !eth_addr_is_reserved(iface_ea) &&
626 !eth_addr_is_zero(iface_ea) &&
627 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
628 memcpy(ea, iface_ea, ETH_ADDR_LEN);
629 *devname = iface->name;
632 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
633 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
634 iface->name, strerror(error));
638 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
639 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
641 VLOG_WARN("bridge %s: using default bridge Ethernet "
642 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
644 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
645 br->name, ETH_ADDR_ARGS(ea));
649 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
650 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
651 * a network device, then that network device's name must be passed in as
652 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
653 * passed in as a null pointer. */
655 bridge_pick_datapath_id(struct bridge *br,
656 const uint8_t bridge_ea[ETH_ADDR_LEN],
660 * The procedure for choosing a bridge MAC address will, in the most
661 * ordinary case, also choose a unique MAC that we can use as a datapath
662 * ID. In some special cases, though, multiple bridges will end up with
663 * the same MAC address. This is OK for the bridges, but it will confuse
664 * the OpenFlow controller, because each datapath needs a unique datapath
667 * Datapath IDs must be unique. It is also very desirable that they be
668 * stable from one run to the next, so that policy set on a datapath
673 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
680 if (!netdev_get_vlan_vid(devname, &vlan)) {
682 * A bridge whose MAC address is taken from a VLAN network device
683 * (that is, a network device created with vconfig(8) or similar
684 * tool) will have the same MAC address as a bridge on the VLAN
685 * device's physical network device.
687 * Handle this case by hashing the physical network device MAC
688 * along with the VLAN identifier.
690 uint8_t buf[ETH_ADDR_LEN + 2];
691 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
692 buf[ETH_ADDR_LEN] = vlan >> 8;
693 buf[ETH_ADDR_LEN + 1] = vlan;
694 return dpid_from_hash(buf, sizeof buf);
697 * Assume that this bridge's MAC address is unique, since it
698 * doesn't fit any of the cases we handle specially.
703 * A purely internal bridge, that is, one that has no non-virtual
704 * network devices on it at all, is more difficult because it has no
705 * natural unique identifier at all.
707 * When the host is a XenServer, we handle this case by hashing the
708 * host's UUID with the name of the bridge. Names of bridges are
709 * persistent across XenServer reboots, although they can be reused if
710 * an internal network is destroyed and then a new one is later
711 * created, so this is fairly effective.
713 * When the host is not a XenServer, we punt by using a random MAC
714 * address on each run.
716 const char *host_uuid = xenserver_get_host_uuid();
718 char *combined = xasprintf("%s,%s", host_uuid, br->name);
719 dpid = dpid_from_hash(combined, strlen(combined));
725 return eth_addr_to_uint64(bridge_ea);
729 dpid_from_hash(const void *data, size_t n)
731 uint8_t hash[SHA1_DIGEST_SIZE];
733 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
734 sha1_bytes(data, n, hash);
735 eth_addr_mark_random(hash);
736 return eth_addr_to_uint64(hash);
742 struct bridge *br, *next;
746 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
747 int error = bridge_run_one(br);
749 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
750 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
751 "forcing reconfiguration", br->name);
765 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
766 ofproto_wait(br->ofproto);
767 if (br->controller) {
772 mac_learning_wait(br->ml);
779 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
780 * configuration changes. */
782 bridge_flush(struct bridge *br)
784 COVERAGE_INC(bridge_flush);
787 mac_learning_flush(br->ml);
791 /* Bridge reconfiguration functions. */
793 static struct bridge *
794 bridge_create(const char *name)
799 assert(!bridge_lookup(name));
800 br = xcalloc(1, sizeof *br);
802 error = dpif_create(name, &br->dpif);
803 if (error == EEXIST) {
804 error = dpif_open(name, &br->dpif);
806 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
807 name, strerror(error));
811 dpif_flow_flush(&br->dpif);
813 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
818 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
820 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
821 dpif_delete(&br->dpif);
822 dpif_close(&br->dpif);
827 br->name = xstrdup(name);
828 br->ml = mac_learning_create();
829 br->sent_config_request = false;
830 eth_addr_random(br->default_ea);
832 port_array_init(&br->ifaces);
835 br->bond_next_rebalance = time_msec() + 10000;
837 list_push_back(&all_bridges, &br->node);
839 VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif));
845 bridge_destroy(struct bridge *br)
850 while (br->n_ports > 0) {
851 port_destroy(br->ports[br->n_ports - 1]);
853 list_remove(&br->node);
854 error = dpif_delete(&br->dpif);
855 if (error && error != ENOENT) {
856 VLOG_ERR("failed to delete dp%u: %s",
857 dpif_id(&br->dpif), strerror(error));
859 dpif_close(&br->dpif);
860 ofproto_destroy(br->ofproto);
861 free(br->controller);
862 mac_learning_destroy(br->ml);
863 port_array_destroy(&br->ifaces);
870 static struct bridge *
871 bridge_lookup(const char *name)
875 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
876 if (!strcmp(br->name, name)) {
884 bridge_exists(const char *name)
886 return bridge_lookup(name) ? true : false;
890 bridge_get_datapathid(const char *name)
892 struct bridge *br = bridge_lookup(name);
893 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
897 bridge_run_one(struct bridge *br)
901 error = ofproto_run1(br->ofproto);
907 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
912 error = ofproto_run2(br->ofproto, br->flush);
919 bridge_get_controller(const struct bridge *br)
921 const char *controller;
923 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
925 controller = cfg_get_string(0, "mgmt.controller");
927 return controller && controller[0] ? controller : NULL;
931 bridge_reconfigure_one(struct bridge *br)
933 struct svec old_ports, new_ports, ifaces;
934 struct svec listeners, old_listeners;
935 struct svec snoops, old_snoops;
938 /* Collect old ports. */
939 svec_init(&old_ports);
940 for (i = 0; i < br->n_ports; i++) {
941 svec_add(&old_ports, br->ports[i]->name);
943 svec_sort(&old_ports);
944 assert(svec_is_unique(&old_ports));
946 /* Collect new ports. */
947 svec_init(&new_ports);
948 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
949 svec_sort(&new_ports);
950 if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) {
951 svec_add(&new_ports, br->name);
952 svec_sort(&new_ports);
954 if (!svec_is_unique(&new_ports)) {
955 VLOG_WARN("bridge %s: %s specified twice as bridge port",
956 br->name, svec_get_duplicate(&new_ports));
957 svec_unique(&new_ports);
960 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
962 /* Get rid of deleted ports and add new ports. */
963 for (i = 0; i < br->n_ports; ) {
964 struct port *port = br->ports[i];
965 if (!svec_contains(&new_ports, port->name)) {
971 for (i = 0; i < new_ports.n; i++) {
972 const char *name = new_ports.names[i];
973 if (!svec_contains(&old_ports, name)) {
974 port_create(br, name);
977 svec_destroy(&old_ports);
978 svec_destroy(&new_ports);
980 /* Reconfigure all ports. */
981 for (i = 0; i < br->n_ports; i++) {
982 port_reconfigure(br->ports[i]);
985 /* Check and delete duplicate interfaces. */
987 for (i = 0; i < br->n_ports; ) {
988 struct port *port = br->ports[i];
989 for (j = 0; j < port->n_ifaces; ) {
990 struct iface *iface = port->ifaces[j];
991 if (svec_contains(&ifaces, iface->name)) {
992 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
994 br->name, iface->name, port->name);
995 iface_destroy(iface);
997 svec_add(&ifaces, iface->name);
1002 if (!port->n_ifaces) {
1003 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1009 svec_destroy(&ifaces);
1011 /* Delete all flows if we're switching from connected to standalone or vice
1012 * versa. (XXX Should we delete all flows if we are switching from one
1013 * controller to another?) */
1015 /* Configure OpenFlow management listeners. */
1016 svec_init(&listeners);
1017 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1019 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1020 ovs_rundir, br->name));
1021 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1022 svec_clear(&listeners);
1024 svec_sort_unique(&listeners);
1026 svec_init(&old_listeners);
1027 ofproto_get_listeners(br->ofproto, &old_listeners);
1028 svec_sort_unique(&old_listeners);
1030 if (!svec_equal(&listeners, &old_listeners)) {
1031 ofproto_set_listeners(br->ofproto, &listeners);
1033 svec_destroy(&listeners);
1034 svec_destroy(&old_listeners);
1036 /* Configure OpenFlow controller connection snooping. */
1038 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1040 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1041 ovs_rundir, br->name));
1042 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1043 svec_clear(&snoops);
1045 svec_sort_unique(&snoops);
1047 svec_init(&old_snoops);
1048 ofproto_get_snoops(br->ofproto, &old_snoops);
1049 svec_sort_unique(&old_snoops);
1051 if (!svec_equal(&snoops, &old_snoops)) {
1052 ofproto_set_snoops(br->ofproto, &snoops);
1054 svec_destroy(&snoops);
1055 svec_destroy(&old_snoops);
1057 mirror_reconfigure(br);
1061 bridge_reconfigure_controller(struct bridge *br)
1063 char *pfx = xasprintf("bridge.%s.controller", br->name);
1064 const char *controller;
1066 controller = bridge_get_controller(br);
1067 if ((br->controller != NULL) != (controller != NULL)) {
1068 ofproto_flush_flows(br->ofproto);
1070 free(br->controller);
1071 br->controller = controller ? xstrdup(controller) : NULL;
1074 const char *fail_mode;
1075 int max_backoff, probe;
1076 int rate_limit, burst_limit;
1078 if (!strcmp(controller, "discover")) {
1079 ofproto_set_discovery(br->ofproto, true,
1080 cfg_get_string(0, "%s.accept-regex", pfx),
1081 cfg_get_bool(0, "%s.update-resolv.conf",
1084 struct netdev *netdev;
1088 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1090 || cfg_get_bool(0, "%s.in-band", pfx));
1091 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1092 ofproto_set_in_band(br->ofproto, in_band);
1094 error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev);
1096 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1097 struct in_addr ip, mask, gateway;
1098 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1099 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1100 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1102 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1104 mask.s_addr = guess_netmask(ip.s_addr);
1106 if (!netdev_set_in4(netdev, ip, mask)) {
1107 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1109 br->name, IP_ARGS(&ip.s_addr),
1110 IP_ARGS(&mask.s_addr));
1113 if (gateway.s_addr) {
1114 if (!netdev_add_router(gateway)) {
1115 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1116 br->name, IP_ARGS(&gateway.s_addr));
1120 netdev_close(netdev);
1124 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1126 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1128 ofproto_set_failure(br->ofproto,
1130 || !strcmp(fail_mode, "standalone")
1131 || !strcmp(fail_mode, "open")));
1133 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1134 ofproto_set_probe_interval(br->ofproto,
1135 probe ? probe : cfg_get_int(0, "mgmt.inactivity-probe"));
1137 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1139 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1144 ofproto_set_max_backoff(br->ofproto, max_backoff);
1146 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1148 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1150 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1152 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1154 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1156 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1158 if (cfg_has("%s.commands.acl", pfx)) {
1159 struct svec command_acls;
1162 svec_init(&command_acls);
1163 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1164 command_acl = svec_join(&command_acls, ",", "");
1166 ofproto_set_remote_execution(br->ofproto, command_acl,
1167 cfg_get_string(0, "%s.commands.dir",
1170 svec_destroy(&command_acls);
1173 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1176 union ofp_action action;
1179 /* Set up a flow that matches every packet and directs them to
1180 * OFPP_NORMAL (which goes to us). */
1181 memset(&action, 0, sizeof action);
1182 action.type = htons(OFPAT_OUTPUT);
1183 action.output.len = htons(sizeof action);
1184 action.output.port = htons(OFPP_NORMAL);
1185 memset(&flow, 0, sizeof flow);
1186 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1189 ofproto_set_in_band(br->ofproto, false);
1190 ofproto_set_max_backoff(br->ofproto, 1);
1191 ofproto_set_probe_interval(br->ofproto, 5);
1192 ofproto_set_failure(br->ofproto, false);
1193 ofproto_set_stp(br->ofproto, false);
1197 ofproto_set_controller(br->ofproto, br->controller);
1201 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1206 for (i = 0; i < br->n_ports; i++) {
1207 struct port *port = br->ports[i];
1208 for (j = 0; j < port->n_ifaces; j++) {
1209 struct iface *iface = port->ifaces[j];
1210 svec_add(ifaces, iface->name);
1214 assert(svec_is_unique(ifaces));
1217 /* For robustness, in case the administrator moves around datapath ports behind
1218 * our back, we re-check all the datapath port numbers here.
1220 * This function will set the 'dp_ifidx' members of interfaces that have
1221 * disappeared to -1, so only call this function from a context where those
1222 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1223 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1224 * datapath, which doesn't support UINT16_MAX+1 ports. */
1226 bridge_fetch_dp_ifaces(struct bridge *br)
1228 struct odp_port *dpif_ports;
1229 size_t n_dpif_ports;
1232 /* Reset all interface numbers. */
1233 for (i = 0; i < br->n_ports; i++) {
1234 struct port *port = br->ports[i];
1235 for (j = 0; j < port->n_ifaces; j++) {
1236 struct iface *iface = port->ifaces[j];
1237 iface->dp_ifidx = -1;
1240 port_array_clear(&br->ifaces);
1242 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
1243 for (i = 0; i < n_dpif_ports; i++) {
1244 struct odp_port *p = &dpif_ports[i];
1245 struct iface *iface = iface_lookup(br, p->devname);
1247 if (iface->dp_ifidx >= 0) {
1248 VLOG_WARN("dp%u reported interface %s twice",
1249 dpif_id(&br->dpif), p->devname);
1250 } else if (iface_from_dp_ifidx(br, p->port)) {
1251 VLOG_WARN("dp%u reported interface %"PRIu16" twice",
1252 dpif_id(&br->dpif), p->port);
1254 port_array_set(&br->ifaces, p->port, iface);
1255 iface->dp_ifidx = p->port;
1262 /* Bridge packet processing functions. */
1265 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1267 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1270 static struct bond_entry *
1271 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1273 return &port->bond_hash[bond_hash(mac)];
1277 bond_choose_iface(const struct port *port)
1280 for (i = 0; i < port->n_ifaces; i++) {
1281 if (port->ifaces[i]->enabled) {
1289 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1290 uint16_t *dp_ifidx, tag_type *tags)
1292 struct iface *iface;
1294 assert(port->n_ifaces);
1295 if (port->n_ifaces == 1) {
1296 iface = port->ifaces[0];
1298 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1299 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1300 || !port->ifaces[e->iface_idx]->enabled) {
1301 /* XXX select interface properly. The current interface selection
1302 * is only good for testing the rebalancing code. */
1303 e->iface_idx = bond_choose_iface(port);
1304 if (e->iface_idx < 0) {
1305 *tags |= port->no_ifaces_tag;
1308 e->iface_tag = tag_create_random();
1310 *tags |= e->iface_tag;
1311 iface = port->ifaces[e->iface_idx];
1313 *dp_ifidx = iface->dp_ifidx;
1314 *tags |= iface->tag; /* Currently only used for bonding. */
1319 bond_link_status_update(struct iface *iface, bool carrier)
1321 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1322 struct port *port = iface->port;
1324 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1325 /* Nothing to do. */
1328 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1329 iface->name, carrier ? "detected" : "dropped");
1330 if (carrier == iface->enabled) {
1331 iface->delay_expires = LLONG_MAX;
1332 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1333 iface->name, carrier ? "disabled" : "enabled");
1335 int delay = carrier ? port->updelay : port->downdelay;
1336 iface->delay_expires = time_msec() + delay;
1339 "interface %s: will be %s if it stays %s for %d ms",
1341 carrier ? "enabled" : "disabled",
1342 carrier ? "up" : "down",
1349 bond_choose_active_iface(struct port *port)
1351 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1353 port->active_iface = bond_choose_iface(port);
1354 port->active_iface_tag = tag_create_random();
1355 if (port->active_iface >= 0) {
1356 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1357 port->name, port->ifaces[port->active_iface]->name);
1359 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1365 bond_enable_slave(struct iface *iface, bool enable)
1367 struct port *port = iface->port;
1368 struct bridge *br = port->bridge;
1370 iface->delay_expires = LLONG_MAX;
1371 if (enable == iface->enabled) {
1375 iface->enabled = enable;
1376 if (!iface->enabled) {
1377 VLOG_WARN("interface %s: enabled", iface->name);
1378 ofproto_revalidate(br->ofproto, iface->tag);
1379 if (iface->port_ifidx == port->active_iface) {
1380 ofproto_revalidate(br->ofproto,
1381 port->active_iface_tag);
1382 bond_choose_active_iface(port);
1384 bond_send_learning_packets(port);
1386 VLOG_WARN("interface %s: disabled", iface->name);
1387 if (port->active_iface < 0) {
1388 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1389 bond_choose_active_iface(port);
1390 bond_send_learning_packets(port);
1392 iface->tag = tag_create_random();
1397 bond_run(struct bridge *br)
1401 for (i = 0; i < br->n_ports; i++) {
1402 struct port *port = br->ports[i];
1403 if (port->n_ifaces < 2) {
1406 for (j = 0; j < port->n_ifaces; j++) {
1407 struct iface *iface = port->ifaces[j];
1408 if (time_msec() >= iface->delay_expires) {
1409 bond_enable_slave(iface, !iface->enabled);
1416 bond_wait(struct bridge *br)
1420 for (i = 0; i < br->n_ports; i++) {
1421 struct port *port = br->ports[i];
1422 if (port->n_ifaces < 2) {
1425 for (j = 0; j < port->n_ifaces; j++) {
1426 struct iface *iface = port->ifaces[j];
1427 if (iface->delay_expires != LLONG_MAX) {
1428 poll_timer_wait(iface->delay_expires - time_msec());
1435 set_dst(struct dst *p, const flow_t *flow,
1436 const struct port *in_port, const struct port *out_port,
1441 * XXX This uses too many tags: any broadcast flow will get one tag per
1442 * destination port, and thus a broadcast on a switch of any size is likely
1443 * to have all tag bits set. We should figure out a way to be smarter.
1445 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1446 *tags |= out_port->stp_state_tag;
1447 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1451 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1452 : in_port->vlan >= 0 ? in_port->vlan
1453 : ntohs(flow->dl_vlan));
1454 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1458 swap_dst(struct dst *p, struct dst *q)
1460 struct dst tmp = *p;
1465 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1466 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1467 * that we push to the datapath. We could in fact fully sort the array by
1468 * vlan, but in most cases there are at most two different vlan tags so that's
1469 * possibly overkill.) */
1471 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1473 struct dst *first = dsts;
1474 struct dst *last = dsts + n_dsts;
1476 while (first != last) {
1478 * - All dsts < first have vlan == 'vlan'.
1479 * - All dsts >= last have vlan != 'vlan'.
1480 * - first < last. */
1481 while (first->vlan == vlan) {
1482 if (++first == last) {
1487 /* Same invariants, plus one additional:
1488 * - first->vlan != vlan.
1490 while (last[-1].vlan != vlan) {
1491 if (--last == first) {
1496 /* Same invariants, plus one additional:
1497 * - last[-1].vlan == vlan.*/
1498 swap_dst(first++, --last);
1503 mirror_mask_ffs(mirror_mask_t mask)
1505 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1510 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1511 const struct dst *test)
1514 for (i = 0; i < n_dsts; i++) {
1515 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1523 port_trunks_vlan(const struct port *port, uint16_t vlan)
1525 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1529 port_includes_vlan(const struct port *port, uint16_t vlan)
1531 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1535 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1536 const struct port *in_port, const struct port *out_port,
1537 struct dst dsts[], tag_type *tags)
1539 mirror_mask_t mirrors = in_port->src_mirrors;
1540 struct dst *dst = dsts;
1543 *tags |= in_port->stp_state_tag;
1544 if (out_port == FLOOD_PORT) {
1545 /* XXX use ODP_FLOOD if no vlans or bonding. */
1546 /* XXX even better, define each VLAN as a datapath port group */
1547 for (i = 0; i < br->n_ports; i++) {
1548 struct port *port = br->ports[i];
1549 if (port != in_port && port_includes_vlan(port, vlan)
1550 && !port->is_mirror_output_port
1551 && set_dst(dst, flow, in_port, port, tags)) {
1552 mirrors |= port->dst_mirrors;
1556 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1557 mirrors |= out_port->dst_mirrors;
1562 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1563 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1565 if (set_dst(dst, flow, in_port, m->out_port, tags)
1566 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1570 for (i = 0; i < br->n_ports; i++) {
1571 struct port *port = br->ports[i];
1572 if (port_includes_vlan(port, m->out_vlan)
1573 && set_dst(dst, flow, in_port, port, tags)
1574 && !dst_is_duplicate(dsts, dst - dsts, dst))
1576 if (port->vlan < 0) {
1577 dst->vlan = m->out_vlan;
1579 if (dst->dp_ifidx == flow->in_port
1580 && dst->vlan == vlan) {
1581 /* Don't send out input port on same VLAN. */
1589 mirrors &= mirrors - 1;
1592 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1597 print_dsts(const struct dst *dsts, size_t n)
1599 for (; n--; dsts++) {
1600 printf(">p%"PRIu16, dsts->dp_ifidx);
1601 if (dsts->vlan != OFP_VLAN_NONE) {
1602 printf("v%"PRIu16, dsts->vlan);
1608 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1609 const struct port *in_port, const struct port *out_port,
1610 tag_type *tags, struct odp_actions *actions)
1612 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1614 const struct dst *p;
1617 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1619 cur_vlan = ntohs(flow->dl_vlan);
1620 for (p = dsts; p < &dsts[n_dsts]; p++) {
1621 union odp_action *a;
1622 if (p->vlan != cur_vlan) {
1623 if (p->vlan == OFP_VLAN_NONE) {
1624 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1626 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1627 a->vlan_vid.vlan_vid = htons(p->vlan);
1631 a = odp_actions_add(actions, ODPAT_OUTPUT);
1632 a->output.port = p->dp_ifidx;
1637 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1639 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1640 return (flow->dl_type == htons(ETH_TYPE_ARP)
1641 && eth_addr_is_broadcast(flow->dl_dst)
1642 && packet->size >= sizeof(struct arp_eth_header)
1643 && arp->ar_op == ARP_OP_REQUEST);
1646 /* If the composed actions may be applied to any packet in the given 'flow',
1647 * returns true. Otherwise, the actions should only be applied to 'packet', or
1648 * not at all, if 'packet' was NULL. */
1650 process_flow(struct bridge *br, const flow_t *flow,
1651 const struct ofpbuf *packet, struct odp_actions *actions,
1654 struct iface *in_iface;
1655 struct port *in_port;
1656 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1659 /* Find the interface and port structure for the received packet. */
1660 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1662 /* No interface? Something fishy... */
1663 if (packet != NULL) {
1664 /* Odd. A few possible reasons here:
1666 * - We deleted an interface but there are still a few packets
1667 * queued up from it.
1669 * - Someone externally added an interface (e.g. with "ovs-dpctl
1670 * add-if") that we don't know about.
1672 * - Packet arrived on the local port but the local port is not
1673 * one of our bridge ports.
1675 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1677 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1678 "interface %"PRIu16, br->name, flow->in_port);
1681 /* Return without adding any actions, to drop packets on this flow. */
1684 in_port = in_iface->port;
1686 /* Figure out what VLAN this packet belongs to.
1688 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1689 * belongs to VLAN 0, so we should treat both cases identically. (In the
1690 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1691 * presumably to allow a priority to be specified. In the latter case, the
1692 * packet does not have any 802.1Q header.) */
1693 vlan = ntohs(flow->dl_vlan);
1694 if (vlan == OFP_VLAN_NONE) {
1697 if (in_port->vlan >= 0) {
1699 /* XXX support double tagging? */
1700 if (packet != NULL) {
1701 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1702 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1703 "packet received on port %s configured with "
1704 "implicit VLAN %"PRIu16,
1705 br->name, ntohs(flow->dl_vlan),
1706 in_port->name, in_port->vlan);
1710 vlan = in_port->vlan;
1712 if (!port_includes_vlan(in_port, vlan)) {
1713 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1714 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1715 "packet received on port %s not configured for "
1717 br->name, vlan, in_port->name, vlan);
1722 /* Drop frames for ports that STP wants entirely killed (both for
1723 * forwarding and for learning). Later, after we do learning, we'll drop
1724 * the frames that STP wants to do learning but not forwarding on. */
1725 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1729 /* Drop frames for reserved multicast addresses. */
1730 if (eth_addr_is_reserved(flow->dl_dst)) {
1734 /* Drop frames on ports reserved for mirroring. */
1735 if (in_port->is_mirror_output_port) {
1736 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1737 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1738 "which is reserved exclusively for mirroring",
1739 br->name, in_port->name);
1743 /* Multicast (and broadcast) packets on bonds need special attention, to
1744 * avoid receiving duplicates. */
1745 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1746 *tags |= in_port->active_iface_tag;
1747 if (in_port->active_iface != in_iface->port_ifidx) {
1748 /* Drop all multicast packets on inactive slaves. */
1751 /* Drop all multicast packets for which we have learned a different
1752 * input port, because we probably sent the packet on one slaves
1753 * and got it back on the active slave. Broadcast ARP replies are
1754 * an exception to this rule: the host has moved to another
1756 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1757 if (src_idx != -1 && src_idx != in_port->port_idx) {
1759 if (!is_bcast_arp_reply(flow, packet)) {
1763 /* No way to know whether it's an ARP reply, because the
1764 * flow entry doesn't include enough information and we
1765 * don't have a packet. Punt. */
1773 out_port = FLOOD_PORT;
1777 /* Learn source MAC (but don't try to learn from revalidation). */
1779 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1780 vlan, in_port->port_idx);
1782 /* The log messages here could actually be useful in debugging,
1783 * so keep the rate limit relatively high. */
1784 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1786 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1787 "on port %s in VLAN %d",
1788 br->name, ETH_ADDR_ARGS(flow->dl_src),
1789 in_port->name, vlan);
1790 ofproto_revalidate(br->ofproto, rev_tag);
1794 /* Determine output port. */
1795 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1797 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1798 out_port = br->ports[out_port_idx];
1802 /* Don't send packets out their input ports. Don't forward frames that STP
1803 * wants us to discard. */
1804 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1809 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1812 * We send out only a single packet, instead of setting up a flow, if the
1813 * packet is an ARP directed to broadcast that arrived on a bonded
1814 * interface. In such a situation ARP requests and replies must be handled
1815 * differently, but OpenFlow unfortunately can't distinguish them.
1817 return (in_port->n_ifaces < 2
1818 || flow->dl_type != htons(ETH_TYPE_ARP)
1819 || !eth_addr_is_broadcast(flow->dl_dst));
1822 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1825 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1826 const struct ofp_phy_port *opp,
1829 struct bridge *br = br_;
1830 struct iface *iface;
1833 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1839 if (reason == OFPPR_DELETE) {
1840 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1841 br->name, iface->name);
1842 iface_destroy(iface);
1843 if (!port->n_ifaces) {
1844 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1845 br->name, port->name);
1851 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1852 if (port->n_ifaces > 1) {
1853 bool up = !(opp->state & OFPPS_LINK_DOWN);
1854 bond_link_status_update(iface, up);
1855 port_update_bond_compat(port);
1861 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1862 struct odp_actions *actions, tag_type *tags, void *br_)
1864 struct bridge *br = br_;
1867 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1868 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1869 brstp_receive(br, flow, payload);
1874 COVERAGE_INC(bridge_process_flow);
1875 return process_flow(br, flow, packet, actions, tags);
1879 bridge_account_flow_ofhook_cb(const flow_t *flow,
1880 const union odp_action *actions,
1881 size_t n_actions, unsigned long long int n_bytes,
1884 struct bridge *br = br_;
1885 const union odp_action *a;
1887 if (!br->has_bonded_ports) {
1891 for (a = actions; a < &actions[n_actions]; a++) {
1892 if (a->type == ODPAT_OUTPUT) {
1893 struct port *port = port_from_dp_ifidx(br, a->output.port);
1894 if (port && port->n_ifaces >= 2) {
1895 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1896 e->tx_bytes += n_bytes;
1903 bridge_account_checkpoint_ofhook_cb(void *br_)
1905 struct bridge *br = br_;
1908 if (!br->has_bonded_ports) {
1912 /* The current ofproto implementation calls this callback at least once a
1913 * second, so this timer implementation is sufficient. */
1914 if (time_msec() < br->bond_next_rebalance) {
1917 br->bond_next_rebalance = time_msec() + 10000;
1919 for (i = 0; i < br->n_ports; i++) {
1920 struct port *port = br->ports[i];
1921 if (port->n_ifaces > 1) {
1922 bond_rebalance_port(port);
1927 static struct ofhooks bridge_ofhooks = {
1928 bridge_port_changed_ofhook_cb,
1929 bridge_normal_ofhook_cb,
1930 bridge_account_flow_ofhook_cb,
1931 bridge_account_checkpoint_ofhook_cb,
1934 /* Bonding functions. */
1936 /* Statistics for a single interface on a bonded port, used for load-based
1937 * bond rebalancing. */
1938 struct slave_balance {
1939 struct iface *iface; /* The interface. */
1940 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
1942 /* All the "bond_entry"s that are assigned to this interface, in order of
1943 * increasing tx_bytes. */
1944 struct bond_entry **hashes;
1948 /* Sorts pointers to pointers to bond_entries in ascending order by the
1949 * interface to which they are assigned, and within a single interface in
1950 * ascending order of bytes transmitted. */
1952 compare_bond_entries(const void *a_, const void *b_)
1954 const struct bond_entry *const *ap = a_;
1955 const struct bond_entry *const *bp = b_;
1956 const struct bond_entry *a = *ap;
1957 const struct bond_entry *b = *bp;
1958 if (a->iface_idx != b->iface_idx) {
1959 return a->iface_idx > b->iface_idx ? 1 : -1;
1960 } else if (a->tx_bytes != b->tx_bytes) {
1961 return a->tx_bytes > b->tx_bytes ? 1 : -1;
1967 /* Sorts slave_balances so that enabled ports come first, and otherwise in
1968 * *descending* order by number of bytes transmitted. */
1970 compare_slave_balance(const void *a_, const void *b_)
1972 const struct slave_balance *a = a_;
1973 const struct slave_balance *b = b_;
1974 if (a->iface->enabled != b->iface->enabled) {
1975 return a->iface->enabled ? -1 : 1;
1976 } else if (a->tx_bytes != b->tx_bytes) {
1977 return a->tx_bytes > b->tx_bytes ? -1 : 1;
1984 swap_bals(struct slave_balance *a, struct slave_balance *b)
1986 struct slave_balance tmp = *a;
1991 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
1992 * given that 'p' (and only 'p') might be in the wrong location.
1994 * This function invalidates 'p', since it might now be in a different memory
1997 resort_bals(struct slave_balance *p,
1998 struct slave_balance bals[], size_t n_bals)
2001 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2002 swap_bals(p, p - 1);
2004 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2005 swap_bals(p, p + 1);
2011 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2013 if (VLOG_IS_DBG_ENABLED()) {
2014 struct ds ds = DS_EMPTY_INITIALIZER;
2015 const struct slave_balance *b;
2017 for (b = bals; b < bals + n_bals; b++) {
2021 ds_put_char(&ds, ',');
2023 ds_put_format(&ds, " %s %"PRIu64"kB",
2024 b->iface->name, b->tx_bytes / 1024);
2026 if (!b->iface->enabled) {
2027 ds_put_cstr(&ds, " (disabled)");
2029 if (b->n_hashes > 0) {
2030 ds_put_cstr(&ds, " (");
2031 for (i = 0; i < b->n_hashes; i++) {
2032 const struct bond_entry *e = b->hashes[i];
2034 ds_put_cstr(&ds, " + ");
2036 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2037 e - port->bond_hash, e->tx_bytes / 1024);
2039 ds_put_cstr(&ds, ")");
2042 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2047 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2049 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2050 struct bond_entry *hash)
2052 struct port *port = from->iface->port;
2053 uint64_t delta = hash->tx_bytes;
2055 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2056 "from %s to %s (now carrying %"PRIu64"kB and "
2057 "%"PRIu64"kB load, respectively)",
2058 port->name, delta / 1024, hash - port->bond_hash,
2059 from->iface->name, to->iface->name,
2060 (from->tx_bytes - delta) / 1024,
2061 (to->tx_bytes + delta) / 1024);
2063 /* Delete element from from->hashes.
2065 * We don't bother to add the element to to->hashes because not only would
2066 * it require more work, the only purpose it would be to allow that hash to
2067 * be migrated to another slave in this rebalancing run, and there is no
2068 * point in doing that. */
2069 if (from->hashes[0] == hash) {
2072 int i = hash - from->hashes[0];
2073 memmove(from->hashes + i, from->hashes + i + 1,
2074 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2078 /* Shift load away from 'from' to 'to'. */
2079 from->tx_bytes -= delta;
2080 to->tx_bytes += delta;
2082 /* Arrange for flows to be revalidated. */
2083 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2084 hash->iface_idx = to->iface->port_ifidx;
2085 hash->iface_tag = tag_create_random();
2089 bond_rebalance_port(struct port *port)
2091 struct slave_balance bals[DP_MAX_PORTS];
2093 struct bond_entry *hashes[BOND_MASK + 1];
2094 struct slave_balance *b, *from, *to;
2095 struct bond_entry *e;
2098 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2099 * descending order of tx_bytes, so that bals[0] represents the most
2100 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2103 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2104 * array for each slave_balance structure, we sort our local array of
2105 * hashes in order by slave, so that all of the hashes for a given slave
2106 * become contiguous in memory, and then we point each 'hashes' members of
2107 * a slave_balance structure to the start of a contiguous group. */
2108 n_bals = port->n_ifaces;
2109 for (b = bals; b < &bals[n_bals]; b++) {
2110 b->iface = port->ifaces[b - bals];
2115 for (i = 0; i <= BOND_MASK; i++) {
2116 hashes[i] = &port->bond_hash[i];
2118 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2119 for (i = 0; i <= BOND_MASK; i++) {
2121 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2122 b = &bals[e->iface_idx];
2123 b->tx_bytes += e->tx_bytes;
2125 b->hashes = &hashes[i];
2130 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2131 log_bals(bals, n_bals, port);
2133 /* Discard slaves that aren't enabled (which were sorted to the back of the
2134 * array earlier). */
2135 while (!bals[n_bals - 1].iface->enabled) {
2142 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2143 to = &bals[n_bals - 1];
2144 for (from = bals; from < to; ) {
2145 uint64_t overload = from->tx_bytes - to->tx_bytes;
2146 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2147 /* The extra load on 'from' (and all less-loaded slaves), compared
2148 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2149 * it is less than ~1Mbps. No point in rebalancing. */
2151 } else if (from->n_hashes == 1) {
2152 /* 'from' only carries a single MAC hash, so we can't shift any
2153 * load away from it, even though we want to. */
2156 /* 'from' is carrying significantly more load than 'to', and that
2157 * load is split across at least two different hashes. Pick a hash
2158 * to migrate to 'to' (the least-loaded slave), given that doing so
2159 * must not cause 'to''s load to exceed 'from''s load.
2161 * The sort order we use means that we prefer to shift away the
2162 * smallest hashes instead of the biggest ones. There is little
2163 * reason behind this decision; we could use the opposite sort
2164 * order to shift away big hashes ahead of small ones. */
2167 for (i = 0; i < from->n_hashes; i++) {
2168 uint64_t delta = from->hashes[i]->tx_bytes;
2169 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2173 if (i < from->n_hashes) {
2174 bond_shift_load(from, to, from->hashes[i]);
2176 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2177 * point to different slave_balance structures. It is only
2178 * valid to do these two operations in a row at all because we
2179 * know that 'from' will not move past 'to' and vice versa. */
2180 resort_bals(from, bals, n_bals);
2181 resort_bals(to, bals, n_bals);
2188 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2189 * historical data to decay to <1% in 7 rebalancing runs. */
2190 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2196 bond_send_learning_packets(struct port *port)
2198 struct bridge *br = port->bridge;
2199 struct mac_entry *e;
2200 struct ofpbuf packet;
2201 int error, n_packets, n_errors;
2203 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2207 ofpbuf_init(&packet, 128);
2208 error = n_packets = n_errors = 0;
2209 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2210 static const char s[] = "Open vSwitch Bond Failover";
2211 union ofp_action actions[2], *a;
2212 struct eth_header *eth;
2213 struct llc_snap_header *llc_snap;
2219 if (e->port == port->port_idx
2220 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2224 /* Compose packet to send. */
2225 ofpbuf_clear(&packet);
2226 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2227 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2228 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2229 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2231 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2232 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2233 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2235 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2236 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2237 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2238 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2239 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2241 /* Compose actions. */
2242 memset(actions, 0, sizeof actions);
2245 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2246 a->vlan_vid.len = htons(sizeof *a);
2247 a->vlan_vid.vlan_vid = htons(e->vlan);
2250 a->output.type = htons(OFPAT_OUTPUT);
2251 a->output.len = htons(sizeof *a);
2252 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2257 flow_extract(&packet, ODPP_NONE, &flow);
2258 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2265 ofpbuf_uninit(&packet);
2268 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2269 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2270 "packets, last error was: %s",
2271 port->name, n_errors, n_packets, strerror(error));
2273 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2274 port->name, n_packets);
2278 /* Bonding unixctl user interface functions. */
2281 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2283 struct ds ds = DS_EMPTY_INITIALIZER;
2284 const struct bridge *br;
2286 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2288 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2291 for (i = 0; i < br->n_ports; i++) {
2292 const struct port *port = br->ports[i];
2293 if (port->n_ifaces > 1) {
2296 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2297 for (j = 0; j < port->n_ifaces; j++) {
2298 const struct iface *iface = port->ifaces[j];
2300 ds_put_cstr(&ds, ", ");
2302 ds_put_cstr(&ds, iface->name);
2304 ds_put_char(&ds, '\n');
2308 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2312 static struct port *
2313 bond_find(const char *name)
2315 const struct bridge *br;
2317 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2320 for (i = 0; i < br->n_ports; i++) {
2321 struct port *port = br->ports[i];
2322 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2331 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2333 struct ds ds = DS_EMPTY_INITIALIZER;
2334 const struct port *port;
2337 port = bond_find(args);
2339 unixctl_command_reply(conn, 501, "no such bond");
2343 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2344 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2345 ds_put_format(&ds, "next rebalance: %lld ms\n",
2346 port->bridge->bond_next_rebalance - time_msec());
2347 for (j = 0; j < port->n_ifaces; j++) {
2348 const struct iface *iface = port->ifaces[j];
2349 struct bond_entry *be;
2352 ds_put_format(&ds, "slave %s: %s\n",
2353 iface->name, iface->enabled ? "enabled" : "disabled");
2354 if (j == port->active_iface) {
2355 ds_put_cstr(&ds, "\tactive slave\n");
2357 if (iface->delay_expires != LLONG_MAX) {
2358 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2359 iface->enabled ? "downdelay" : "updelay",
2360 iface->delay_expires - time_msec());
2364 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2365 int hash = be - port->bond_hash;
2366 struct mac_entry *me;
2368 if (be->iface_idx != j) {
2372 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2373 hash, be->tx_bytes / 1024);
2376 if (!port->bridge->ml) {
2380 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2381 &port->bridge->ml->lrus) {
2384 if (bond_hash(me->mac) == hash
2385 && me->port != port->port_idx
2386 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2387 && dp_ifidx == iface->dp_ifidx)
2389 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2390 ETH_ADDR_ARGS(me->mac));
2395 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2400 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2402 char *args = (char *) args_;
2403 char *save_ptr = NULL;
2404 char *bond_s, *hash_s, *slave_s;
2405 uint8_t mac[ETH_ADDR_LEN];
2407 struct iface *iface;
2408 struct bond_entry *entry;
2411 bond_s = strtok_r(args, " ", &save_ptr);
2412 hash_s = strtok_r(NULL, " ", &save_ptr);
2413 slave_s = strtok_r(NULL, " ", &save_ptr);
2415 unixctl_command_reply(conn, 501,
2416 "usage: bond/migrate BOND HASH SLAVE");
2420 port = bond_find(bond_s);
2422 unixctl_command_reply(conn, 501, "no such bond");
2426 if (sscanf(hash_s, "%"SCNx8":%"SCNx8":%"SCNx8":%"SCNx8":%"SCNx8":%"SCNx8,
2427 &mac[0], &mac[1], &mac[2], &mac[3], &mac[4], &mac[5]) == 6) {
2428 hash = bond_hash(mac);
2429 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2430 hash = atoi(hash_s) & BOND_MASK;
2432 unixctl_command_reply(conn, 501, "bad hash");
2436 iface = port_lookup_iface(port, slave_s);
2438 unixctl_command_reply(conn, 501, "no such slave");
2442 if (!iface->enabled) {
2443 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2447 entry = &port->bond_hash[hash];
2448 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2449 entry->iface_idx = iface->port_ifidx;
2450 entry->iface_tag = tag_create_random();
2451 unixctl_command_reply(conn, 200, "migrated");
2455 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2457 char *args = (char *) args_;
2458 char *save_ptr = NULL;
2459 char *bond_s, *slave_s;
2461 struct iface *iface;
2463 bond_s = strtok_r(args, " ", &save_ptr);
2464 slave_s = strtok_r(NULL, " ", &save_ptr);
2466 unixctl_command_reply(conn, 501,
2467 "usage: bond/set-active-slave BOND SLAVE");
2471 port = bond_find(bond_s);
2473 unixctl_command_reply(conn, 501, "no such bond");
2477 iface = port_lookup_iface(port, slave_s);
2479 unixctl_command_reply(conn, 501, "no such slave");
2483 if (!iface->enabled) {
2484 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2488 if (port->active_iface != iface->port_ifidx) {
2489 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2490 port->active_iface = iface->port_ifidx;
2491 port->active_iface_tag = tag_create_random();
2492 VLOG_INFO("port %s: active interface is now %s",
2493 port->name, iface->name);
2494 bond_send_learning_packets(port);
2495 unixctl_command_reply(conn, 200, "done");
2497 unixctl_command_reply(conn, 200, "no change");
2502 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
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/enable/disable-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 bond_enable_slave(iface, enable);
2531 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2535 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2537 enable_slave(conn, args, true);
2541 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2543 enable_slave(conn, args, false);
2549 unixctl_command_register("bond/list", bond_unixctl_list);
2550 unixctl_command_register("bond/show", bond_unixctl_show);
2551 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2552 unixctl_command_register("bond/set-active-slave",
2553 bond_unixctl_set_active_slave);
2554 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2555 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2558 /* Port functions. */
2561 port_create(struct bridge *br, const char *name)
2565 port = xcalloc(1, sizeof *port);
2567 port->port_idx = br->n_ports;
2569 port->trunks = NULL;
2570 port->name = xstrdup(name);
2571 port->active_iface = -1;
2572 port->stp_state = STP_DISABLED;
2573 port->stp_state_tag = 0;
2575 if (br->n_ports >= br->allocated_ports) {
2576 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2579 br->ports[br->n_ports++] = port;
2581 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2586 port_reconfigure(struct port *port)
2588 bool bonded = cfg_has_section("bonding.%s", port->name);
2589 struct svec old_ifaces, new_ifaces;
2590 unsigned long *trunks;
2594 /* Collect old and new interfaces. */
2595 svec_init(&old_ifaces);
2596 svec_init(&new_ifaces);
2597 for (i = 0; i < port->n_ifaces; i++) {
2598 svec_add(&old_ifaces, port->ifaces[i]->name);
2600 svec_sort(&old_ifaces);
2602 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2603 if (!new_ifaces.n) {
2604 VLOG_ERR("port %s: no interfaces specified for bonded port",
2606 } else if (new_ifaces.n == 1) {
2607 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2611 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2612 if (port->updelay < 0) {
2615 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2616 if (port->downdelay < 0) {
2617 port->downdelay = 0;
2620 svec_init(&new_ifaces);
2621 svec_add(&new_ifaces, port->name);
2624 /* Get rid of deleted interfaces and add new interfaces. */
2625 for (i = 0; i < port->n_ifaces; i++) {
2626 struct iface *iface = port->ifaces[i];
2627 if (!svec_contains(&new_ifaces, iface->name)) {
2628 iface_destroy(iface);
2633 for (i = 0; i < new_ifaces.n; i++) {
2634 const char *name = new_ifaces.names[i];
2635 if (!svec_contains(&old_ifaces, name)) {
2636 iface_create(port, name);
2642 if (cfg_has("vlan.%s.tag", port->name)) {
2644 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2645 if (vlan >= 0 && vlan <= 4095) {
2646 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2649 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2650 * they even work as-is. But they have not been tested. */
2651 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2655 if (port->vlan != vlan) {
2657 bridge_flush(port->bridge);
2660 /* Get trunked VLANs. */
2663 size_t n_trunks, n_errors;
2666 trunks = bitmap_allocate(4096);
2667 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2669 for (i = 0; i < n_trunks; i++) {
2670 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2672 bitmap_set1(trunks, trunk);
2678 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2679 port->name, n_trunks);
2681 if (n_errors == n_trunks) {
2683 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2686 bitmap_set_multiple(trunks, 0, 4096, 1);
2689 if (cfg_has("vlan.%s.trunks", port->name)) {
2690 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2691 port->name, port->name);
2695 ? port->trunks != NULL
2696 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2697 bridge_flush(port->bridge);
2699 bitmap_free(port->trunks);
2700 port->trunks = trunks;
2702 svec_destroy(&old_ifaces);
2703 svec_destroy(&new_ifaces);
2707 port_destroy(struct port *port)
2710 struct bridge *br = port->bridge;
2714 proc_net_compat_update_vlan(port->name, NULL, 0);
2716 for (i = 0; i < MAX_MIRRORS; i++) {
2717 struct mirror *m = br->mirrors[i];
2718 if (m && m->out_port == port) {
2723 while (port->n_ifaces > 0) {
2724 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2727 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2728 del->port_idx = port->port_idx;
2731 bitmap_free(port->trunks);
2738 static struct port *
2739 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2741 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2742 return iface ? iface->port : NULL;
2745 static struct port *
2746 port_lookup(const struct bridge *br, const char *name)
2750 for (i = 0; i < br->n_ports; i++) {
2751 struct port *port = br->ports[i];
2752 if (!strcmp(port->name, name)) {
2759 static struct iface *
2760 port_lookup_iface(const struct port *port, const char *name)
2764 for (j = 0; j < port->n_ifaces; j++) {
2765 struct iface *iface = port->ifaces[j];
2766 if (!strcmp(iface->name, name)) {
2774 port_update_bonding(struct port *port)
2776 if (port->n_ifaces < 2) {
2777 /* Not a bonded port. */
2778 if (port->bond_hash) {
2779 free(port->bond_hash);
2780 port->bond_hash = NULL;
2781 proc_net_compat_update_bond(port->name, NULL);
2784 if (!port->bond_hash) {
2787 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2788 for (i = 0; i <= BOND_MASK; i++) {
2789 struct bond_entry *e = &port->bond_hash[i];
2793 port->no_ifaces_tag = tag_create_random();
2794 bond_choose_active_iface(port);
2796 port_update_bond_compat(port);
2801 port_update_bond_compat(struct port *port)
2803 struct compat_bond bond;
2806 if (port->n_ifaces < 2) {
2811 bond.updelay = port->updelay;
2812 bond.downdelay = port->downdelay;
2813 bond.n_slaves = port->n_ifaces;
2814 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2815 for (i = 0; i < port->n_ifaces; i++) {
2816 struct iface *iface = port->ifaces[i];
2817 struct compat_bond_slave *slave = &bond.slaves[i];
2818 slave->name = iface->name;
2819 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2820 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2824 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2826 proc_net_compat_update_bond(port->name, &bond);
2831 port_update_vlan_compat(struct port *port)
2833 struct bridge *br = port->bridge;
2834 char *vlandev_name = NULL;
2836 if (port->vlan > 0) {
2837 /* Figure out the name that the VLAN device should actually have, if it
2838 * existed. This takes some work because the VLAN device would not
2839 * have port->name in its name; rather, it would have the trunk port's
2840 * name, and 'port' would be attached to a bridge that also had the
2841 * VLAN device one of its ports. So we need to find a trunk port that
2842 * includes port->vlan.
2844 * There might be more than one candidate. This doesn't happen on
2845 * XenServer, so if it happens we just pick the first choice in
2846 * alphabetical order instead of creating multiple VLAN devices. */
2848 for (i = 0; i < br->n_ports; i++) {
2849 struct port *p = br->ports[i];
2850 if (port_trunks_vlan(p, port->vlan)
2852 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2854 const uint8_t *ea = p->ifaces[0]->mac;
2855 if (!eth_addr_is_multicast(ea) &&
2856 !eth_addr_is_reserved(ea) &&
2857 !eth_addr_is_zero(ea)) {
2858 vlandev_name = p->name;
2863 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2866 /* Interface functions. */
2869 iface_create(struct port *port, const char *name)
2871 struct iface *iface;
2873 iface = xcalloc(1, sizeof *iface);
2875 iface->port_ifidx = port->n_ifaces;
2876 iface->name = xstrdup(name);
2877 iface->dp_ifidx = -1;
2878 iface->tag = tag_create_random();
2879 iface->delay_expires = LLONG_MAX;
2881 netdev_nodev_get_etheraddr(name, iface->mac);
2882 netdev_nodev_get_carrier(name, &iface->enabled);
2884 if (port->n_ifaces >= port->allocated_ifaces) {
2885 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2886 sizeof *port->ifaces);
2888 port->ifaces[port->n_ifaces++] = iface;
2889 if (port->n_ifaces > 1) {
2890 port->bridge->has_bonded_ports = true;
2893 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2895 port_update_bonding(port);
2896 bridge_flush(port->bridge);
2900 iface_destroy(struct iface *iface)
2903 struct port *port = iface->port;
2904 struct bridge *br = port->bridge;
2905 bool del_active = port->active_iface == iface->port_ifidx;
2908 if (iface->dp_ifidx >= 0) {
2909 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
2912 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
2913 del->port_ifidx = iface->port_ifidx;
2919 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2920 bond_choose_active_iface(port);
2921 bond_send_learning_packets(port);
2924 port_update_bonding(port);
2925 bridge_flush(port->bridge);
2929 static struct iface *
2930 iface_lookup(const struct bridge *br, const char *name)
2934 for (i = 0; i < br->n_ports; i++) {
2935 struct port *port = br->ports[i];
2936 for (j = 0; j < port->n_ifaces; j++) {
2937 struct iface *iface = port->ifaces[j];
2938 if (!strcmp(iface->name, name)) {
2946 static struct iface *
2947 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2949 return port_array_get(&br->ifaces, dp_ifidx);
2952 /* Port mirroring. */
2955 mirror_reconfigure(struct bridge *br)
2957 struct svec old_mirrors, new_mirrors;
2960 /* Collect old and new mirrors. */
2961 svec_init(&old_mirrors);
2962 svec_init(&new_mirrors);
2963 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
2964 for (i = 0; i < MAX_MIRRORS; i++) {
2965 if (br->mirrors[i]) {
2966 svec_add(&old_mirrors, br->mirrors[i]->name);
2970 /* Get rid of deleted mirrors and add new mirrors. */
2971 svec_sort(&old_mirrors);
2972 assert(svec_is_unique(&old_mirrors));
2973 svec_sort(&new_mirrors);
2974 assert(svec_is_unique(&new_mirrors));
2975 for (i = 0; i < MAX_MIRRORS; i++) {
2976 struct mirror *m = br->mirrors[i];
2977 if (m && !svec_contains(&new_mirrors, m->name)) {
2981 for (i = 0; i < new_mirrors.n; i++) {
2982 const char *name = new_mirrors.names[i];
2983 if (!svec_contains(&old_mirrors, name)) {
2984 mirror_create(br, name);
2987 svec_destroy(&old_mirrors);
2988 svec_destroy(&new_mirrors);
2990 /* Reconfigure all mirrors. */
2991 for (i = 0; i < MAX_MIRRORS; i++) {
2992 if (br->mirrors[i]) {
2993 mirror_reconfigure_one(br->mirrors[i]);
2997 /* Update port reserved status. */
2998 for (i = 0; i < br->n_ports; i++) {
2999 br->ports[i]->is_mirror_output_port = false;
3001 for (i = 0; i < MAX_MIRRORS; i++) {
3002 struct mirror *m = br->mirrors[i];
3003 if (m && m->out_port) {
3004 m->out_port->is_mirror_output_port = true;
3010 mirror_create(struct bridge *br, const char *name)
3015 for (i = 0; ; i++) {
3016 if (i >= MAX_MIRRORS) {
3017 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3018 "cannot create %s", br->name, MAX_MIRRORS, name);
3021 if (!br->mirrors[i]) {
3026 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3029 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3032 m->name = xstrdup(name);
3033 svec_init(&m->src_ports);
3034 svec_init(&m->dst_ports);
3042 mirror_destroy(struct mirror *m)
3045 struct bridge *br = m->bridge;
3048 for (i = 0; i < br->n_ports; i++) {
3049 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3050 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3053 svec_destroy(&m->src_ports);
3054 svec_destroy(&m->dst_ports);
3057 m->bridge->mirrors[m->idx] = NULL;
3065 prune_ports(struct mirror *m, struct svec *ports)
3070 svec_sort_unique(ports);
3073 for (i = 0; i < ports->n; i++) {
3074 const char *name = ports->names[i];
3075 if (port_lookup(m->bridge, name)) {
3076 svec_add(&tmp, name);
3078 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3079 m->bridge->name, m->name, name);
3082 svec_swap(ports, &tmp);
3087 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3091 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3092 * order won't give us numeric sort order. But that's good enough for what
3093 * we need right now. */
3094 svec_sort_unique(vlan_strings);
3096 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3098 for (i = 0; i < vlan_strings->n; i++) {
3099 const char *name = vlan_strings->names[i];
3101 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3102 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3103 m->bridge->name, m->name, name);
3105 (*vlans)[n_vlans++] = vlan;
3112 vlan_is_mirrored(const struct mirror *m, int vlan)
3116 for (i = 0; i < m->n_vlans; i++) {
3117 if (m->vlans[i] == vlan) {
3125 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3129 for (i = 0; i < m->n_vlans; i++) {
3130 if (port_trunks_vlan(p, m->vlans[i])) {
3138 mirror_reconfigure_one(struct mirror *m)
3140 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3141 struct svec src_ports, dst_ports, ports;
3142 struct svec vlan_strings;
3143 mirror_mask_t mirror_bit;
3144 const char *out_port_name;
3145 struct port *out_port;
3150 bool mirror_all_ports;
3152 /* Get output port. */
3153 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3154 m->bridge->name, m->name);
3155 if (out_port_name) {
3156 out_port = port_lookup(m->bridge, out_port_name);
3158 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3159 "named %s", pfx, m->bridge->name, out_port_name);
3166 if (cfg_has("%s.output.vlan", pfx)) {
3167 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3168 "ignoring %s.output.vlan", pfx, pfx, pfx);
3170 } else if (cfg_has("%s.output.vlan", pfx)) {
3172 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3174 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3175 "but exactly one is required; disabling port mirror %s",
3176 pfx, pfx, pfx, pfx);
3182 /* Get all the ports, and drop duplicates and ports that don't exist. */
3183 svec_init(&src_ports);
3184 svec_init(&dst_ports);
3186 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3187 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3188 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3189 svec_append(&src_ports, &ports);
3190 svec_append(&dst_ports, &ports);
3191 svec_destroy(&ports);
3192 prune_ports(m, &src_ports);
3193 prune_ports(m, &dst_ports);
3195 /* Get all the vlans, and drop duplicate and invalid vlans. */
3196 svec_init(&vlan_strings);
3197 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3198 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3199 svec_destroy(&vlan_strings);
3201 /* Update mirror data. */
3202 if (!svec_equal(&m->src_ports, &src_ports)
3203 || !svec_equal(&m->dst_ports, &dst_ports)
3204 || m->n_vlans != n_vlans
3205 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3206 || m->out_port != out_port
3207 || m->out_vlan != out_vlan) {
3208 bridge_flush(m->bridge);
3210 svec_swap(&m->src_ports, &src_ports);
3211 svec_swap(&m->dst_ports, &dst_ports);
3214 m->n_vlans = n_vlans;
3215 m->out_port = out_port;
3216 m->out_vlan = out_vlan;
3218 /* If no selection criteria have been given, mirror for all ports. */
3219 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3222 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3223 for (i = 0; i < m->bridge->n_ports; i++) {
3224 struct port *port = m->bridge->ports[i];
3226 if (mirror_all_ports
3227 || svec_contains(&m->src_ports, port->name)
3230 ? port_trunks_any_mirrored_vlan(m, port)
3231 : vlan_is_mirrored(m, port->vlan)))) {
3232 port->src_mirrors |= mirror_bit;
3234 port->src_mirrors &= ~mirror_bit;
3237 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3238 port->dst_mirrors |= mirror_bit;
3240 port->dst_mirrors &= ~mirror_bit;
3245 svec_destroy(&src_ports);
3246 svec_destroy(&dst_ports);
3250 /* Spanning tree protocol. */
3252 static void brstp_update_port_state(struct port *);
3255 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3257 struct bridge *br = br_;
3258 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3259 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3261 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3263 } else if (eth_addr_is_zero(iface->mac)) {
3264 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3267 union ofp_action action;
3268 struct eth_header *eth = pkt->l2;
3271 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3273 memset(&action, 0, sizeof action);
3274 action.type = htons(OFPAT_OUTPUT);
3275 action.output.len = htons(sizeof action);
3276 action.output.port = htons(port_no);
3278 flow_extract(pkt, ODPP_NONE, &flow);
3279 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3285 brstp_reconfigure(struct bridge *br)
3289 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3291 stp_destroy(br->stp);
3297 uint64_t bridge_address, bridge_id;
3298 int bridge_priority;
3300 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3301 if (!bridge_address) {
3303 bridge_address = (stp_get_bridge_id(br->stp)
3304 & ((UINT64_C(1) << 48) - 1));
3306 uint8_t mac[ETH_ADDR_LEN];
3307 eth_addr_random(mac);
3308 bridge_address = eth_addr_to_uint64(mac);
3312 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3314 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3316 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3319 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3321 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3322 br->stp_last_tick = time_msec();
3325 if (bridge_id != stp_get_bridge_id(br->stp)) {
3326 stp_set_bridge_id(br->stp, bridge_id);
3331 for (i = 0; i < br->n_ports; i++) {
3332 struct port *p = br->ports[i];
3334 struct stp_port *sp;
3335 int path_cost, priority;
3341 dp_ifidx = p->ifaces[0]->dp_ifidx;
3342 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3346 sp = stp_get_port(br->stp, dp_ifidx);
3347 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3348 "stp.%s.port.%s.enabled",
3350 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3351 br->name, p->name));
3352 if (p->is_mirror_output_port) {
3355 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3356 bridge_flush(br); /* Might not be necessary. */
3358 stp_port_enable(sp);
3360 stp_port_disable(sp);
3364 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3366 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3368 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3369 "stp.%s.port.%s.priority",
3371 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3373 : STP_DEFAULT_PORT_PRIORITY);
3374 stp_port_set_priority(sp, priority);
3377 brstp_adjust_timers(br);
3379 for (i = 0; i < br->n_ports; i++) {
3380 brstp_update_port_state(br->ports[i]);
3385 brstp_update_port_state(struct port *p)
3387 struct bridge *br = p->bridge;
3388 enum stp_state state;
3390 /* Figure out new state. */
3391 state = STP_DISABLED;
3392 if (br->stp && p->n_ifaces > 0) {
3393 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3394 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3395 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3400 if (p->stp_state != state) {
3401 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3402 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3403 p->name, stp_state_name(p->stp_state),
3404 stp_state_name(state));
3405 if (p->stp_state == STP_DISABLED) {
3408 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3410 p->stp_state = state;
3411 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3412 : tag_create_random());
3417 brstp_adjust_timers(struct bridge *br)
3419 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3420 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3421 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3423 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3424 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3425 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3429 brstp_run(struct bridge *br)
3432 long long int now = time_msec();
3433 long long int elapsed = now - br->stp_last_tick;
3434 struct stp_port *sp;
3437 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3438 br->stp_last_tick = now;
3440 while (stp_get_changed_port(br->stp, &sp)) {
3441 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3443 brstp_update_port_state(p);
3450 brstp_wait(struct bridge *br)
3453 poll_timer_wait(1000);