1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id;
74 struct port *port; /* Containing port. */
75 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 int dp_ifidx; /* Index within kernel datapath. */
80 uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */
82 tag_type tag; /* Tag associated with this interface. */
83 bool enabled; /* May be chosen for flows? */
84 long long delay_expires; /* Time after which 'enabled' may change. */
87 #define BOND_MASK 0xff
89 int iface_idx; /* Index of assigned iface, or -1 if none. */
90 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
91 tag_type iface_tag; /* Tag associated with iface_idx. */
94 #define MAX_MIRRORS 32
95 typedef uint32_t mirror_mask_t;
96 #define MIRROR_MASK_C(X) UINT32_C(X)
97 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
99 struct bridge *bridge;
103 /* Selection criteria. */
104 struct svec src_ports;
105 struct svec dst_ports;
110 struct port *out_port;
114 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
116 struct bridge *bridge;
118 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
119 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
122 /* An ordinary bridge port has 1 interface.
123 * A bridge port for bonding has at least 2 interfaces. */
124 struct iface **ifaces;
125 size_t n_ifaces, allocated_ifaces;
128 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
129 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
130 tag_type active_iface_tag; /* Tag for bcast flows. */
131 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
132 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
134 /* Port mirroring info. */
135 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
136 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
137 bool is_mirror_output_port; /* Does port mirroring send frames here? */
139 /* Spanning tree info. */
140 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
141 tag_type stp_state_tag; /* Tag for STP state change. */
144 #define DP_MAX_PORTS 255
146 struct list node; /* Node in global list of bridges. */
147 char *name; /* User-specified arbitrary name. */
148 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
149 bool sent_config_request; /* Successfully sent config request? */
150 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
152 /* Support for remote controllers. */
153 char *controller; /* NULL if there is no remote controller;
154 * "discover" to do controller discovery;
155 * otherwise a vconn name. */
157 /* OpenFlow switch processing. */
158 struct ofproto *ofproto; /* OpenFlow switch. */
160 /* Kernel datapath information. */
161 struct dpif *dpif; /* Datapath. */
162 struct port_array ifaces; /* Indexed by kernel datapath port number. */
166 size_t n_ports, allocated_ports;
169 bool has_bonded_ports;
170 long long int bond_next_rebalance;
175 /* Flow statistics gathering. */
176 time_t next_stats_request;
178 /* Port mirroring. */
179 struct mirror *mirrors[MAX_MIRRORS];
183 long long int stp_last_tick;
186 /* List of all bridges. */
187 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
189 /* Maximum number of datapaths. */
190 enum { DP_MAX = 256 };
192 static struct bridge *bridge_create(const char *name);
193 static void bridge_destroy(struct bridge *);
194 static struct bridge *bridge_lookup(const char *name);
195 static int bridge_run_one(struct bridge *);
196 static void bridge_reconfigure_one(struct bridge *);
197 static void bridge_reconfigure_controller(struct bridge *);
198 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
199 static void bridge_fetch_dp_ifaces(struct bridge *);
200 static void bridge_flush(struct bridge *);
201 static void bridge_pick_local_hw_addr(struct bridge *,
202 uint8_t ea[ETH_ADDR_LEN],
203 const char **devname);
204 static uint64_t bridge_pick_datapath_id(struct bridge *,
205 const uint8_t bridge_ea[ETH_ADDR_LEN],
206 const char *devname);
207 static uint64_t dpid_from_hash(const void *, size_t nbytes);
209 static void 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 datapath %s, ignoring",
263 iface->name, dpif_name(br->dpif));
265 if (iface->dp_ifidx != ODPP_LOCAL) {
266 svec_add(svec, iface->name);
274 /* The caller must already have called cfg_read(). */
278 struct svec dpif_names;
281 dp_enumerate(&dpif_names);
282 for (i = 0; i < dpif_names.n; i++) {
283 const char *dpif_name = dpif_names.names[i];
287 retval = dpif_open(dpif_name, &dpif);
289 struct svec all_names;
292 svec_init(&all_names);
293 dpif_get_all_names(dpif, &all_names);
294 for (j = 0; j < all_names.n; j++) {
295 if (cfg_has("bridge.%s.port", all_names.names[j])) {
301 svec_destroy(&all_names);
307 bridge_reconfigure();
312 config_string_change(const char *key, char **valuep)
314 const char *value = cfg_get_string(0, "%s", key);
315 if (value && (!*valuep || strcmp(value, *valuep))) {
317 *valuep = xstrdup(value);
325 bridge_configure_ssl(void)
327 /* XXX SSL should be configurable on a per-bridge basis.
328 * XXX should be possible to de-configure SSL. */
329 static char *private_key_file;
330 static char *certificate_file;
331 static char *cacert_file;
334 if (config_string_change("ssl.private-key", &private_key_file)) {
335 vconn_ssl_set_private_key_file(private_key_file);
338 if (config_string_change("ssl.certificate", &certificate_file)) {
339 vconn_ssl_set_certificate_file(certificate_file);
342 /* We assume that even if the filename hasn't changed, if the CA cert
343 * file has been removed, that we want to move back into
344 * boot-strapping mode. This opens a small security hole, because
345 * the old certificate will still be trusted until vSwitch is
346 * restarted. We may want to address this in vconn's SSL library. */
347 if (config_string_change("ssl.ca-cert", &cacert_file)
348 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
349 vconn_ssl_set_ca_cert_file(cacert_file,
350 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
356 bridge_reconfigure(void)
358 struct svec old_br, new_br;
359 struct bridge *br, *next;
362 COVERAGE_INC(bridge_reconfigure);
364 /* Collect old and new bridges. */
367 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
368 svec_add(&old_br, br->name);
370 cfg_get_subsections(&new_br, "bridge");
372 /* Get rid of deleted bridges and add new bridges. */
375 assert(svec_is_unique(&old_br));
376 assert(svec_is_unique(&new_br));
377 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
378 if (!svec_contains(&new_br, br->name)) {
382 for (i = 0; i < new_br.n; i++) {
383 const char *name = new_br.names[i];
384 if (!svec_contains(&old_br, name)) {
388 svec_destroy(&old_br);
389 svec_destroy(&new_br);
393 bridge_configure_ssl();
396 /* Reconfigure all bridges. */
397 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
398 bridge_reconfigure_one(br);
401 /* Add and delete ports on all datapaths.
403 * The kernel will reject any attempt to add a given port to a datapath if
404 * that port already belongs to a different datapath, so we must do all
405 * port deletions before any port additions. */
406 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
407 struct odp_port *dpif_ports;
409 struct svec want_ifaces;
411 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
412 bridge_get_all_ifaces(br, &want_ifaces);
413 for (i = 0; i < n_dpif_ports; i++) {
414 const struct odp_port *p = &dpif_ports[i];
415 if (!svec_contains(&want_ifaces, p->devname)
416 && strcmp(p->devname, br->name)) {
417 int retval = dpif_port_del(br->dpif, p->port);
419 VLOG_ERR("failed to remove %s interface from %s: %s",
420 p->devname, dpif_name(br->dpif),
425 svec_destroy(&want_ifaces);
428 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
429 struct odp_port *dpif_ports;
431 struct svec cur_ifaces, want_ifaces, add_ifaces;
433 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
434 svec_init(&cur_ifaces);
435 for (i = 0; i < n_dpif_ports; i++) {
436 svec_add(&cur_ifaces, dpif_ports[i].devname);
439 svec_sort_unique(&cur_ifaces);
440 bridge_get_all_ifaces(br, &want_ifaces);
441 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
443 for (i = 0; i < add_ifaces.n; i++) {
444 const char *if_name = add_ifaces.names[i];
445 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
446 int flags = internal ? ODP_PORT_INTERNAL : 0;
447 int error = dpif_port_add(br->dpif, if_name, flags, NULL);
448 if (error == EXFULL) {
449 VLOG_ERR("ran out of valid port numbers on %s",
450 dpif_name(br->dpif));
453 VLOG_ERR("failed to add %s interface to %s: %s",
454 if_name, dpif_name(br->dpif), strerror(error));
457 svec_destroy(&cur_ifaces);
458 svec_destroy(&want_ifaces);
459 svec_destroy(&add_ifaces);
461 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
464 struct iface *local_iface = NULL;
466 uint8_t engine_type, engine_id;
467 bool add_id_to_iface = false;
468 struct svec nf_hosts;
470 bridge_fetch_dp_ifaces(br);
471 for (i = 0; i < br->n_ports; ) {
472 struct port *port = br->ports[i];
474 for (j = 0; j < port->n_ifaces; ) {
475 struct iface *iface = port->ifaces[j];
476 if (iface->dp_ifidx < 0) {
477 VLOG_ERR("%s interface not in %s, dropping",
478 iface->name, dpif_name(br->dpif));
479 iface_destroy(iface);
481 if (iface->dp_ifidx == ODPP_LOCAL) {
484 VLOG_DBG("%s has interface %s on port %d",
486 iface->name, iface->dp_ifidx);
490 if (!port->n_ifaces) {
491 VLOG_ERR("%s port has no interfaces, dropping", port->name);
498 /* Pick local port hardware address, datapath ID. */
499 bridge_pick_local_hw_addr(br, ea, &devname);
501 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
503 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
504 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
505 "Ethernet address: %s",
506 br->name, strerror(error));
510 dpid = bridge_pick_datapath_id(br, ea, devname);
511 ofproto_set_datapath_id(br->ofproto, dpid);
513 /* Set NetFlow configuration on this bridge. */
514 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
515 if (cfg_has("netflow.%s.engine-type", br->name)) {
516 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
519 if (cfg_has("netflow.%s.engine-id", br->name)) {
520 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
522 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
523 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
526 if (add_id_to_iface && engine_id > 0x7f) {
527 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
528 "another vswitch, choose an engine id less than 128",
531 if (add_id_to_iface && br->n_ports > 0x1ff) {
532 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
533 "another port when 512 or more ports are used",
536 svec_init(&nf_hosts);
537 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
538 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
539 engine_id, add_id_to_iface)) {
540 VLOG_ERR("bridge %s: problem setting netflow collectors",
544 /* Update the controller and related settings. It would be more
545 * straightforward to call this from bridge_reconfigure_one(), but we
546 * can't do it there for two reasons. First, and most importantly, at
547 * that point we don't know the dp_ifidx of any interfaces that have
548 * been added to the bridge (because we haven't actually added them to
549 * the datapath). Second, at that point we haven't set the datapath ID
550 * yet; when a controller is configured, resetting the datapath ID will
551 * immediately disconnect from the controller, so it's better to set
552 * the datapath ID before the controller. */
553 bridge_reconfigure_controller(br);
555 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
556 for (i = 0; i < br->n_ports; i++) {
557 struct port *port = br->ports[i];
558 port_update_vlan_compat(port);
561 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
562 brstp_reconfigure(br);
567 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
568 const char **devname)
570 uint64_t requested_ea;
576 /* Did the user request a particular MAC? */
577 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
579 eth_addr_from_uint64(requested_ea, ea);
580 if (eth_addr_is_multicast(ea)) {
581 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
582 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
583 } else if (eth_addr_is_zero(ea)) {
584 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
590 /* Otherwise choose the minimum MAC address among all of the interfaces.
591 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
592 * MAC of the physical interface in such an environment.) */
593 memset(ea, 0xff, sizeof ea);
594 for (i = 0; i < br->n_ports; i++) {
595 struct port *port = br->ports[i];
596 if (port->is_mirror_output_port) {
599 for (j = 0; j < port->n_ifaces; j++) {
600 struct iface *iface = port->ifaces[j];
601 uint8_t iface_ea[ETH_ADDR_LEN];
602 if (iface->dp_ifidx == ODPP_LOCAL
603 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
606 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
608 if (!eth_addr_is_multicast(iface_ea) &&
609 !eth_addr_is_reserved(iface_ea) &&
610 !eth_addr_is_zero(iface_ea) &&
611 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
612 memcpy(ea, iface_ea, ETH_ADDR_LEN);
613 *devname = iface->name;
616 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
617 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
618 iface->name, strerror(error));
622 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
623 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
625 VLOG_WARN("bridge %s: using default bridge Ethernet "
626 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
628 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
629 br->name, ETH_ADDR_ARGS(ea));
633 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
634 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
635 * a network device, then that network device's name must be passed in as
636 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
637 * passed in as a null pointer. */
639 bridge_pick_datapath_id(struct bridge *br,
640 const uint8_t bridge_ea[ETH_ADDR_LEN],
644 * The procedure for choosing a bridge MAC address will, in the most
645 * ordinary case, also choose a unique MAC that we can use as a datapath
646 * ID. In some special cases, though, multiple bridges will end up with
647 * the same MAC address. This is OK for the bridges, but it will confuse
648 * the OpenFlow controller, because each datapath needs a unique datapath
651 * Datapath IDs must be unique. It is also very desirable that they be
652 * stable from one run to the next, so that policy set on a datapath
657 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
664 if (!netdev_get_vlan_vid(devname, &vlan)) {
666 * A bridge whose MAC address is taken from a VLAN network device
667 * (that is, a network device created with vconfig(8) or similar
668 * tool) will have the same MAC address as a bridge on the VLAN
669 * device's physical network device.
671 * Handle this case by hashing the physical network device MAC
672 * along with the VLAN identifier.
674 uint8_t buf[ETH_ADDR_LEN + 2];
675 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
676 buf[ETH_ADDR_LEN] = vlan >> 8;
677 buf[ETH_ADDR_LEN + 1] = vlan;
678 return dpid_from_hash(buf, sizeof buf);
681 * Assume that this bridge's MAC address is unique, since it
682 * doesn't fit any of the cases we handle specially.
687 * A purely internal bridge, that is, one that has no non-virtual
688 * network devices on it at all, is more difficult because it has no
689 * natural unique identifier at all.
691 * When the host is a XenServer, we handle this case by hashing the
692 * host's UUID with the name of the bridge. Names of bridges are
693 * persistent across XenServer reboots, although they can be reused if
694 * an internal network is destroyed and then a new one is later
695 * created, so this is fairly effective.
697 * When the host is not a XenServer, we punt by using a random MAC
698 * address on each run.
700 const char *host_uuid = xenserver_get_host_uuid();
702 char *combined = xasprintf("%s,%s", host_uuid, br->name);
703 dpid = dpid_from_hash(combined, strlen(combined));
709 return eth_addr_to_uint64(bridge_ea);
713 dpid_from_hash(const void *data, size_t n)
715 uint8_t hash[SHA1_DIGEST_SIZE];
717 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
718 sha1_bytes(data, n, hash);
719 eth_addr_mark_random(hash);
720 return eth_addr_to_uint64(hash);
726 struct bridge *br, *next;
730 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
731 int error = bridge_run_one(br);
733 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
734 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
735 "forcing reconfiguration", br->name);
749 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
750 ofproto_wait(br->ofproto);
751 if (br->controller) {
756 mac_learning_wait(br->ml);
763 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
764 * configuration changes. */
766 bridge_flush(struct bridge *br)
768 COVERAGE_INC(bridge_flush);
771 mac_learning_flush(br->ml);
775 /* Bridge reconfiguration functions. */
777 static struct bridge *
778 bridge_create(const char *name)
783 assert(!bridge_lookup(name));
784 br = xcalloc(1, sizeof *br);
786 error = dpif_create(name, &br->dpif);
787 if (error == EEXIST || error == EBUSY) {
788 error = dpif_open(name, &br->dpif);
790 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
791 name, strerror(error));
795 dpif_flow_flush(br->dpif);
797 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
802 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
804 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
805 dpif_delete(br->dpif);
806 dpif_close(br->dpif);
811 br->name = xstrdup(name);
812 br->ml = mac_learning_create();
813 br->sent_config_request = false;
814 eth_addr_random(br->default_ea);
816 port_array_init(&br->ifaces);
819 br->bond_next_rebalance = time_msec() + 10000;
821 list_push_back(&all_bridges, &br->node);
823 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
829 bridge_destroy(struct bridge *br)
834 while (br->n_ports > 0) {
835 port_destroy(br->ports[br->n_ports - 1]);
837 list_remove(&br->node);
838 error = dpif_delete(br->dpif);
839 if (error && error != ENOENT) {
840 VLOG_ERR("failed to delete %s: %s",
841 dpif_name(br->dpif), strerror(error));
843 dpif_close(br->dpif);
844 ofproto_destroy(br->ofproto);
845 free(br->controller);
846 mac_learning_destroy(br->ml);
847 port_array_destroy(&br->ifaces);
854 static struct bridge *
855 bridge_lookup(const char *name)
859 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
860 if (!strcmp(br->name, name)) {
868 bridge_exists(const char *name)
870 return bridge_lookup(name) ? true : false;
874 bridge_get_datapathid(const char *name)
876 struct bridge *br = bridge_lookup(name);
877 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
881 bridge_run_one(struct bridge *br)
885 error = ofproto_run1(br->ofproto);
891 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
896 error = ofproto_run2(br->ofproto, br->flush);
903 bridge_get_controller(const struct bridge *br)
905 const char *controller;
907 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
909 controller = cfg_get_string(0, "mgmt.controller");
911 return controller && controller[0] ? controller : NULL;
915 bridge_reconfigure_one(struct bridge *br)
917 struct svec old_ports, new_ports, ifaces;
918 struct svec listeners, old_listeners;
919 struct svec snoops, old_snoops;
922 /* Collect old ports. */
923 svec_init(&old_ports);
924 for (i = 0; i < br->n_ports; i++) {
925 svec_add(&old_ports, br->ports[i]->name);
927 svec_sort(&old_ports);
928 assert(svec_is_unique(&old_ports));
930 /* Collect new ports. */
931 svec_init(&new_ports);
932 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
933 svec_sort(&new_ports);
934 if (bridge_get_controller(br)) {
935 char local_name[IF_NAMESIZE];
938 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
939 local_name, sizeof local_name);
940 if (!error && !svec_contains(&new_ports, local_name)) {
941 svec_add(&new_ports, local_name);
942 svec_sort(&new_ports);
945 if (!svec_is_unique(&new_ports)) {
946 VLOG_WARN("bridge %s: %s specified twice as bridge port",
947 br->name, svec_get_duplicate(&new_ports));
948 svec_unique(&new_ports);
951 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
953 /* Get rid of deleted ports and add new ports. */
954 for (i = 0; i < br->n_ports; ) {
955 struct port *port = br->ports[i];
956 if (!svec_contains(&new_ports, port->name)) {
962 for (i = 0; i < new_ports.n; i++) {
963 const char *name = new_ports.names[i];
964 if (!svec_contains(&old_ports, name)) {
965 port_create(br, name);
968 svec_destroy(&old_ports);
969 svec_destroy(&new_ports);
971 /* Reconfigure all ports. */
972 for (i = 0; i < br->n_ports; i++) {
973 port_reconfigure(br->ports[i]);
976 /* Check and delete duplicate interfaces. */
978 for (i = 0; i < br->n_ports; ) {
979 struct port *port = br->ports[i];
980 for (j = 0; j < port->n_ifaces; ) {
981 struct iface *iface = port->ifaces[j];
982 if (svec_contains(&ifaces, iface->name)) {
983 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
985 br->name, iface->name, port->name);
986 iface_destroy(iface);
988 svec_add(&ifaces, iface->name);
993 if (!port->n_ifaces) {
994 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1000 svec_destroy(&ifaces);
1002 /* Delete all flows if we're switching from connected to standalone or vice
1003 * versa. (XXX Should we delete all flows if we are switching from one
1004 * controller to another?) */
1006 /* Configure OpenFlow management listeners. */
1007 svec_init(&listeners);
1008 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1010 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1011 ovs_rundir, br->name));
1012 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1013 svec_clear(&listeners);
1015 svec_sort_unique(&listeners);
1017 svec_init(&old_listeners);
1018 ofproto_get_listeners(br->ofproto, &old_listeners);
1019 svec_sort_unique(&old_listeners);
1021 if (!svec_equal(&listeners, &old_listeners)) {
1022 ofproto_set_listeners(br->ofproto, &listeners);
1024 svec_destroy(&listeners);
1025 svec_destroy(&old_listeners);
1027 /* Configure OpenFlow controller connection snooping. */
1029 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1031 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1032 ovs_rundir, br->name));
1033 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1034 svec_clear(&snoops);
1036 svec_sort_unique(&snoops);
1038 svec_init(&old_snoops);
1039 ofproto_get_snoops(br->ofproto, &old_snoops);
1040 svec_sort_unique(&old_snoops);
1042 if (!svec_equal(&snoops, &old_snoops)) {
1043 ofproto_set_snoops(br->ofproto, &snoops);
1045 svec_destroy(&snoops);
1046 svec_destroy(&old_snoops);
1048 mirror_reconfigure(br);
1052 bridge_reconfigure_controller(struct bridge *br)
1054 char *pfx = xasprintf("bridge.%s.controller", br->name);
1055 const char *controller;
1057 controller = bridge_get_controller(br);
1058 if ((br->controller != NULL) != (controller != NULL)) {
1059 ofproto_flush_flows(br->ofproto);
1061 free(br->controller);
1062 br->controller = controller ? xstrdup(controller) : NULL;
1065 const char *fail_mode;
1066 int max_backoff, probe;
1067 int rate_limit, burst_limit;
1069 if (!strcmp(controller, "discover")) {
1070 ofproto_set_discovery(br->ofproto, true,
1071 cfg_get_string(0, "%s.accept-regex", pfx),
1072 cfg_get_bool(0, "%s.update-resolv.conf",
1075 char local_name[IF_NAMESIZE];
1076 struct netdev *netdev;
1080 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1082 || cfg_get_bool(0, "%s.in-band", pfx));
1083 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1084 ofproto_set_in_band(br->ofproto, in_band);
1086 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1087 local_name, sizeof local_name);
1089 error = netdev_open(local_name, NETDEV_ETH_TYPE_NONE, &netdev);
1092 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1093 struct in_addr ip, mask, gateway;
1094 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1095 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1096 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1098 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1100 mask.s_addr = guess_netmask(ip.s_addr);
1102 if (!netdev_set_in4(netdev, ip, mask)) {
1103 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1105 br->name, IP_ARGS(&ip.s_addr),
1106 IP_ARGS(&mask.s_addr));
1109 if (gateway.s_addr) {
1110 if (!netdev_add_router(gateway)) {
1111 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1112 br->name, IP_ARGS(&gateway.s_addr));
1116 netdev_close(netdev);
1120 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1122 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1124 ofproto_set_failure(br->ofproto,
1126 || !strcmp(fail_mode, "standalone")
1127 || !strcmp(fail_mode, "open")));
1129 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1131 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1136 ofproto_set_probe_interval(br->ofproto, probe);
1138 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1140 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1145 ofproto_set_max_backoff(br->ofproto, max_backoff);
1147 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1149 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1151 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1153 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1155 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1157 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1159 if (cfg_has("%s.commands.acl", pfx)) {
1160 struct svec command_acls;
1163 svec_init(&command_acls);
1164 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1165 command_acl = svec_join(&command_acls, ",", "");
1167 ofproto_set_remote_execution(br->ofproto, command_acl,
1168 cfg_get_string(0, "%s.commands.dir",
1171 svec_destroy(&command_acls);
1174 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1177 union ofp_action action;
1180 /* Set up a flow that matches every packet and directs them to
1181 * OFPP_NORMAL (which goes to us). */
1182 memset(&action, 0, sizeof action);
1183 action.type = htons(OFPAT_OUTPUT);
1184 action.output.len = htons(sizeof action);
1185 action.output.port = htons(OFPP_NORMAL);
1186 memset(&flow, 0, sizeof flow);
1187 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1190 ofproto_set_in_band(br->ofproto, false);
1191 ofproto_set_max_backoff(br->ofproto, 1);
1192 ofproto_set_probe_interval(br->ofproto, 5);
1193 ofproto_set_failure(br->ofproto, false);
1194 ofproto_set_stp(br->ofproto, false);
1198 ofproto_set_controller(br->ofproto, br->controller);
1202 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1207 for (i = 0; i < br->n_ports; i++) {
1208 struct port *port = br->ports[i];
1209 for (j = 0; j < port->n_ifaces; j++) {
1210 struct iface *iface = port->ifaces[j];
1211 svec_add(ifaces, iface->name);
1215 assert(svec_is_unique(ifaces));
1218 /* For robustness, in case the administrator moves around datapath ports behind
1219 * our back, we re-check all the datapath port numbers here.
1221 * This function will set the 'dp_ifidx' members of interfaces that have
1222 * disappeared to -1, so only call this function from a context where those
1223 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1224 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1225 * datapath, which doesn't support UINT16_MAX+1 ports. */
1227 bridge_fetch_dp_ifaces(struct bridge *br)
1229 struct odp_port *dpif_ports;
1230 size_t n_dpif_ports;
1233 /* Reset all interface numbers. */
1234 for (i = 0; i < br->n_ports; i++) {
1235 struct port *port = br->ports[i];
1236 for (j = 0; j < port->n_ifaces; j++) {
1237 struct iface *iface = port->ifaces[j];
1238 iface->dp_ifidx = -1;
1241 port_array_clear(&br->ifaces);
1243 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1244 for (i = 0; i < n_dpif_ports; i++) {
1245 struct odp_port *p = &dpif_ports[i];
1246 struct iface *iface = iface_lookup(br, p->devname);
1248 if (iface->dp_ifidx >= 0) {
1249 VLOG_WARN("%s reported interface %s twice",
1250 dpif_name(br->dpif), p->devname);
1251 } else if (iface_from_dp_ifidx(br, p->port)) {
1252 VLOG_WARN("%s reported interface %"PRIu16" twice",
1253 dpif_name(br->dpif), p->port);
1255 port_array_set(&br->ifaces, p->port, iface);
1256 iface->dp_ifidx = p->port;
1263 /* Bridge packet processing functions. */
1266 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1268 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1271 static struct bond_entry *
1272 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1274 return &port->bond_hash[bond_hash(mac)];
1278 bond_choose_iface(const struct port *port)
1281 for (i = 0; i < port->n_ifaces; i++) {
1282 if (port->ifaces[i]->enabled) {
1290 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1291 uint16_t *dp_ifidx, tag_type *tags)
1293 struct iface *iface;
1295 assert(port->n_ifaces);
1296 if (port->n_ifaces == 1) {
1297 iface = port->ifaces[0];
1299 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1300 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1301 || !port->ifaces[e->iface_idx]->enabled) {
1302 /* XXX select interface properly. The current interface selection
1303 * is only good for testing the rebalancing code. */
1304 e->iface_idx = bond_choose_iface(port);
1305 if (e->iface_idx < 0) {
1306 *tags |= port->no_ifaces_tag;
1309 e->iface_tag = tag_create_random();
1311 *tags |= e->iface_tag;
1312 iface = port->ifaces[e->iface_idx];
1314 *dp_ifidx = iface->dp_ifidx;
1315 *tags |= iface->tag; /* Currently only used for bonding. */
1320 bond_link_status_update(struct iface *iface, bool carrier)
1322 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1323 struct port *port = iface->port;
1325 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1326 /* Nothing to do. */
1329 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1330 iface->name, carrier ? "detected" : "dropped");
1331 if (carrier == iface->enabled) {
1332 iface->delay_expires = LLONG_MAX;
1333 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1334 iface->name, carrier ? "disabled" : "enabled");
1336 int delay = carrier ? port->updelay : port->downdelay;
1337 iface->delay_expires = time_msec() + delay;
1340 "interface %s: will be %s if it stays %s for %d ms",
1342 carrier ? "enabled" : "disabled",
1343 carrier ? "up" : "down",
1350 bond_choose_active_iface(struct port *port)
1352 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1354 port->active_iface = bond_choose_iface(port);
1355 port->active_iface_tag = tag_create_random();
1356 if (port->active_iface >= 0) {
1357 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1358 port->name, port->ifaces[port->active_iface]->name);
1360 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1366 bond_enable_slave(struct iface *iface, bool enable)
1368 struct port *port = iface->port;
1369 struct bridge *br = port->bridge;
1371 iface->delay_expires = LLONG_MAX;
1372 if (enable == iface->enabled) {
1376 iface->enabled = enable;
1377 if (!iface->enabled) {
1378 VLOG_WARN("interface %s: enabled", iface->name);
1379 ofproto_revalidate(br->ofproto, iface->tag);
1380 if (iface->port_ifidx == port->active_iface) {
1381 ofproto_revalidate(br->ofproto,
1382 port->active_iface_tag);
1383 bond_choose_active_iface(port);
1385 bond_send_learning_packets(port);
1387 VLOG_WARN("interface %s: disabled", iface->name);
1388 if (port->active_iface < 0) {
1389 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1390 bond_choose_active_iface(port);
1391 bond_send_learning_packets(port);
1393 iface->tag = tag_create_random();
1398 bond_run(struct bridge *br)
1402 for (i = 0; i < br->n_ports; i++) {
1403 struct port *port = br->ports[i];
1404 if (port->n_ifaces < 2) {
1407 for (j = 0; j < port->n_ifaces; j++) {
1408 struct iface *iface = port->ifaces[j];
1409 if (time_msec() >= iface->delay_expires) {
1410 bond_enable_slave(iface, !iface->enabled);
1417 bond_wait(struct bridge *br)
1421 for (i = 0; i < br->n_ports; i++) {
1422 struct port *port = br->ports[i];
1423 if (port->n_ifaces < 2) {
1426 for (j = 0; j < port->n_ifaces; j++) {
1427 struct iface *iface = port->ifaces[j];
1428 if (iface->delay_expires != LLONG_MAX) {
1429 poll_timer_wait(iface->delay_expires - time_msec());
1436 set_dst(struct dst *p, const flow_t *flow,
1437 const struct port *in_port, const struct port *out_port,
1442 * XXX This uses too many tags: any broadcast flow will get one tag per
1443 * destination port, and thus a broadcast on a switch of any size is likely
1444 * to have all tag bits set. We should figure out a way to be smarter.
1446 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1447 *tags |= out_port->stp_state_tag;
1448 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1452 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1453 : in_port->vlan >= 0 ? in_port->vlan
1454 : ntohs(flow->dl_vlan));
1455 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1459 swap_dst(struct dst *p, struct dst *q)
1461 struct dst tmp = *p;
1466 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1467 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1468 * that we push to the datapath. We could in fact fully sort the array by
1469 * vlan, but in most cases there are at most two different vlan tags so that's
1470 * possibly overkill.) */
1472 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1474 struct dst *first = dsts;
1475 struct dst *last = dsts + n_dsts;
1477 while (first != last) {
1479 * - All dsts < first have vlan == 'vlan'.
1480 * - All dsts >= last have vlan != 'vlan'.
1481 * - first < last. */
1482 while (first->vlan == vlan) {
1483 if (++first == last) {
1488 /* Same invariants, plus one additional:
1489 * - first->vlan != vlan.
1491 while (last[-1].vlan != vlan) {
1492 if (--last == first) {
1497 /* Same invariants, plus one additional:
1498 * - last[-1].vlan == vlan.*/
1499 swap_dst(first++, --last);
1504 mirror_mask_ffs(mirror_mask_t mask)
1506 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1511 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1512 const struct dst *test)
1515 for (i = 0; i < n_dsts; i++) {
1516 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1524 port_trunks_vlan(const struct port *port, uint16_t vlan)
1526 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1530 port_includes_vlan(const struct port *port, uint16_t vlan)
1532 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1536 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1537 const struct port *in_port, const struct port *out_port,
1538 struct dst dsts[], tag_type *tags)
1540 mirror_mask_t mirrors = in_port->src_mirrors;
1541 struct dst *dst = dsts;
1544 *tags |= in_port->stp_state_tag;
1545 if (out_port == FLOOD_PORT) {
1546 /* XXX use ODP_FLOOD if no vlans or bonding. */
1547 /* XXX even better, define each VLAN as a datapath port group */
1548 for (i = 0; i < br->n_ports; i++) {
1549 struct port *port = br->ports[i];
1550 if (port != in_port && port_includes_vlan(port, vlan)
1551 && !port->is_mirror_output_port
1552 && set_dst(dst, flow, in_port, port, tags)) {
1553 mirrors |= port->dst_mirrors;
1557 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1558 mirrors |= out_port->dst_mirrors;
1563 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1564 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1566 if (set_dst(dst, flow, in_port, m->out_port, tags)
1567 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1571 for (i = 0; i < br->n_ports; i++) {
1572 struct port *port = br->ports[i];
1573 if (port_includes_vlan(port, m->out_vlan)
1574 && set_dst(dst, flow, in_port, port, tags)
1575 && !dst_is_duplicate(dsts, dst - dsts, dst))
1577 if (port->vlan < 0) {
1578 dst->vlan = m->out_vlan;
1580 if (dst->dp_ifidx == flow->in_port
1581 && dst->vlan == vlan) {
1582 /* Don't send out input port on same VLAN. */
1590 mirrors &= mirrors - 1;
1593 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1598 print_dsts(const struct dst *dsts, size_t n)
1600 for (; n--; dsts++) {
1601 printf(">p%"PRIu16, dsts->dp_ifidx);
1602 if (dsts->vlan != OFP_VLAN_NONE) {
1603 printf("v%"PRIu16, dsts->vlan);
1609 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1610 const struct port *in_port, const struct port *out_port,
1611 tag_type *tags, struct odp_actions *actions)
1613 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1615 const struct dst *p;
1618 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1620 cur_vlan = ntohs(flow->dl_vlan);
1621 for (p = dsts; p < &dsts[n_dsts]; p++) {
1622 union odp_action *a;
1623 if (p->vlan != cur_vlan) {
1624 if (p->vlan == OFP_VLAN_NONE) {
1625 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1627 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1628 a->vlan_vid.vlan_vid = htons(p->vlan);
1632 a = odp_actions_add(actions, ODPAT_OUTPUT);
1633 a->output.port = p->dp_ifidx;
1638 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1640 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1641 return (flow->dl_type == htons(ETH_TYPE_ARP)
1642 && eth_addr_is_broadcast(flow->dl_dst)
1643 && packet->size >= sizeof(struct arp_eth_header)
1644 && arp->ar_op == ARP_OP_REQUEST);
1647 /* If the composed actions may be applied to any packet in the given 'flow',
1648 * returns true. Otherwise, the actions should only be applied to 'packet', or
1649 * not at all, if 'packet' was NULL. */
1651 process_flow(struct bridge *br, const flow_t *flow,
1652 const struct ofpbuf *packet, struct odp_actions *actions,
1655 struct iface *in_iface;
1656 struct port *in_port;
1657 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1660 /* Find the interface and port structure for the received packet. */
1661 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1663 /* No interface? Something fishy... */
1664 if (packet != NULL) {
1665 /* Odd. A few possible reasons here:
1667 * - We deleted an interface but there are still a few packets
1668 * queued up from it.
1670 * - Someone externally added an interface (e.g. with "ovs-dpctl
1671 * add-if") that we don't know about.
1673 * - Packet arrived on the local port but the local port is not
1674 * one of our bridge ports.
1676 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1678 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1679 "interface %"PRIu16, br->name, flow->in_port);
1682 /* Return without adding any actions, to drop packets on this flow. */
1685 in_port = in_iface->port;
1687 /* Figure out what VLAN this packet belongs to.
1689 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1690 * belongs to VLAN 0, so we should treat both cases identically. (In the
1691 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1692 * presumably to allow a priority to be specified. In the latter case, the
1693 * packet does not have any 802.1Q header.) */
1694 vlan = ntohs(flow->dl_vlan);
1695 if (vlan == OFP_VLAN_NONE) {
1698 if (in_port->vlan >= 0) {
1700 /* XXX support double tagging? */
1701 if (packet != NULL) {
1702 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1703 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1704 "packet received on port %s configured with "
1705 "implicit VLAN %"PRIu16,
1706 br->name, ntohs(flow->dl_vlan),
1707 in_port->name, in_port->vlan);
1711 vlan = in_port->vlan;
1713 if (!port_includes_vlan(in_port, vlan)) {
1714 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1715 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1716 "packet received on port %s not configured for "
1718 br->name, vlan, in_port->name, vlan);
1723 /* Drop frames for ports that STP wants entirely killed (both for
1724 * forwarding and for learning). Later, after we do learning, we'll drop
1725 * the frames that STP wants to do learning but not forwarding on. */
1726 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1730 /* Drop frames for reserved multicast addresses. */
1731 if (eth_addr_is_reserved(flow->dl_dst)) {
1735 /* Drop frames on ports reserved for mirroring. */
1736 if (in_port->is_mirror_output_port) {
1737 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1738 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1739 "which is reserved exclusively for mirroring",
1740 br->name, in_port->name);
1744 /* Multicast (and broadcast) packets on bonds need special attention, to
1745 * avoid receiving duplicates. */
1746 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1747 *tags |= in_port->active_iface_tag;
1748 if (in_port->active_iface != in_iface->port_ifidx) {
1749 /* Drop all multicast packets on inactive slaves. */
1752 /* Drop all multicast packets for which we have learned a different
1753 * input port, because we probably sent the packet on one slaves
1754 * and got it back on the active slave. Broadcast ARP replies are
1755 * an exception to this rule: the host has moved to another
1757 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1758 if (src_idx != -1 && src_idx != in_port->port_idx) {
1760 if (!is_bcast_arp_reply(flow, packet)) {
1764 /* No way to know whether it's an ARP reply, because the
1765 * flow entry doesn't include enough information and we
1766 * don't have a packet. Punt. */
1774 out_port = FLOOD_PORT;
1778 /* Learn source MAC (but don't try to learn from revalidation). */
1780 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1781 vlan, in_port->port_idx);
1783 /* The log messages here could actually be useful in debugging,
1784 * so keep the rate limit relatively high. */
1785 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1787 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1788 "on port %s in VLAN %d",
1789 br->name, ETH_ADDR_ARGS(flow->dl_src),
1790 in_port->name, vlan);
1791 ofproto_revalidate(br->ofproto, rev_tag);
1795 /* Determine output port. */
1796 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1798 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1799 out_port = br->ports[out_port_idx];
1803 /* Don't send packets out their input ports. Don't forward frames that STP
1804 * wants us to discard. */
1805 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1810 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1813 * We send out only a single packet, instead of setting up a flow, if the
1814 * packet is an ARP directed to broadcast that arrived on a bonded
1815 * interface. In such a situation ARP requests and replies must be handled
1816 * differently, but OpenFlow unfortunately can't distinguish them.
1818 return (in_port->n_ifaces < 2
1819 || flow->dl_type != htons(ETH_TYPE_ARP)
1820 || !eth_addr_is_broadcast(flow->dl_dst));
1823 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1826 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1827 const struct ofp_phy_port *opp,
1830 struct bridge *br = br_;
1831 struct iface *iface;
1834 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1840 if (reason == OFPPR_DELETE) {
1841 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1842 br->name, iface->name);
1843 iface_destroy(iface);
1844 if (!port->n_ifaces) {
1845 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1846 br->name, port->name);
1852 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1853 if (port->n_ifaces > 1) {
1854 bool up = !(opp->state & OFPPS_LINK_DOWN);
1855 bond_link_status_update(iface, up);
1856 port_update_bond_compat(port);
1862 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1863 struct odp_actions *actions, tag_type *tags, void *br_)
1865 struct bridge *br = br_;
1868 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1869 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1870 brstp_receive(br, flow, payload);
1875 COVERAGE_INC(bridge_process_flow);
1876 return process_flow(br, flow, packet, actions, tags);
1880 bridge_account_flow_ofhook_cb(const flow_t *flow,
1881 const union odp_action *actions,
1882 size_t n_actions, unsigned long long int n_bytes,
1885 struct bridge *br = br_;
1886 const union odp_action *a;
1888 if (!br->has_bonded_ports) {
1892 for (a = actions; a < &actions[n_actions]; a++) {
1893 if (a->type == ODPAT_OUTPUT) {
1894 struct port *port = port_from_dp_ifidx(br, a->output.port);
1895 if (port && port->n_ifaces >= 2) {
1896 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1897 e->tx_bytes += n_bytes;
1904 bridge_account_checkpoint_ofhook_cb(void *br_)
1906 struct bridge *br = br_;
1909 if (!br->has_bonded_ports) {
1913 /* The current ofproto implementation calls this callback at least once a
1914 * second, so this timer implementation is sufficient. */
1915 if (time_msec() < br->bond_next_rebalance) {
1918 br->bond_next_rebalance = time_msec() + 10000;
1920 for (i = 0; i < br->n_ports; i++) {
1921 struct port *port = br->ports[i];
1922 if (port->n_ifaces > 1) {
1923 bond_rebalance_port(port);
1928 static struct ofhooks bridge_ofhooks = {
1929 bridge_port_changed_ofhook_cb,
1930 bridge_normal_ofhook_cb,
1931 bridge_account_flow_ofhook_cb,
1932 bridge_account_checkpoint_ofhook_cb,
1935 /* Bonding functions. */
1937 /* Statistics for a single interface on a bonded port, used for load-based
1938 * bond rebalancing. */
1939 struct slave_balance {
1940 struct iface *iface; /* The interface. */
1941 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
1943 /* All the "bond_entry"s that are assigned to this interface, in order of
1944 * increasing tx_bytes. */
1945 struct bond_entry **hashes;
1949 /* Sorts pointers to pointers to bond_entries in ascending order by the
1950 * interface to which they are assigned, and within a single interface in
1951 * ascending order of bytes transmitted. */
1953 compare_bond_entries(const void *a_, const void *b_)
1955 const struct bond_entry *const *ap = a_;
1956 const struct bond_entry *const *bp = b_;
1957 const struct bond_entry *a = *ap;
1958 const struct bond_entry *b = *bp;
1959 if (a->iface_idx != b->iface_idx) {
1960 return a->iface_idx > b->iface_idx ? 1 : -1;
1961 } else if (a->tx_bytes != b->tx_bytes) {
1962 return a->tx_bytes > b->tx_bytes ? 1 : -1;
1968 /* Sorts slave_balances so that enabled ports come first, and otherwise in
1969 * *descending* order by number of bytes transmitted. */
1971 compare_slave_balance(const void *a_, const void *b_)
1973 const struct slave_balance *a = a_;
1974 const struct slave_balance *b = b_;
1975 if (a->iface->enabled != b->iface->enabled) {
1976 return a->iface->enabled ? -1 : 1;
1977 } else if (a->tx_bytes != b->tx_bytes) {
1978 return a->tx_bytes > b->tx_bytes ? -1 : 1;
1985 swap_bals(struct slave_balance *a, struct slave_balance *b)
1987 struct slave_balance tmp = *a;
1992 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
1993 * given that 'p' (and only 'p') might be in the wrong location.
1995 * This function invalidates 'p', since it might now be in a different memory
1998 resort_bals(struct slave_balance *p,
1999 struct slave_balance bals[], size_t n_bals)
2002 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2003 swap_bals(p, p - 1);
2005 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2006 swap_bals(p, p + 1);
2012 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2014 if (VLOG_IS_DBG_ENABLED()) {
2015 struct ds ds = DS_EMPTY_INITIALIZER;
2016 const struct slave_balance *b;
2018 for (b = bals; b < bals + n_bals; b++) {
2022 ds_put_char(&ds, ',');
2024 ds_put_format(&ds, " %s %"PRIu64"kB",
2025 b->iface->name, b->tx_bytes / 1024);
2027 if (!b->iface->enabled) {
2028 ds_put_cstr(&ds, " (disabled)");
2030 if (b->n_hashes > 0) {
2031 ds_put_cstr(&ds, " (");
2032 for (i = 0; i < b->n_hashes; i++) {
2033 const struct bond_entry *e = b->hashes[i];
2035 ds_put_cstr(&ds, " + ");
2037 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2038 e - port->bond_hash, e->tx_bytes / 1024);
2040 ds_put_cstr(&ds, ")");
2043 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2048 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2050 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2051 struct bond_entry *hash)
2053 struct port *port = from->iface->port;
2054 uint64_t delta = hash->tx_bytes;
2056 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2057 "from %s to %s (now carrying %"PRIu64"kB and "
2058 "%"PRIu64"kB load, respectively)",
2059 port->name, delta / 1024, hash - port->bond_hash,
2060 from->iface->name, to->iface->name,
2061 (from->tx_bytes - delta) / 1024,
2062 (to->tx_bytes + delta) / 1024);
2064 /* Delete element from from->hashes.
2066 * We don't bother to add the element to to->hashes because not only would
2067 * it require more work, the only purpose it would be to allow that hash to
2068 * be migrated to another slave in this rebalancing run, and there is no
2069 * point in doing that. */
2070 if (from->hashes[0] == hash) {
2073 int i = hash - from->hashes[0];
2074 memmove(from->hashes + i, from->hashes + i + 1,
2075 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2079 /* Shift load away from 'from' to 'to'. */
2080 from->tx_bytes -= delta;
2081 to->tx_bytes += delta;
2083 /* Arrange for flows to be revalidated. */
2084 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2085 hash->iface_idx = to->iface->port_ifidx;
2086 hash->iface_tag = tag_create_random();
2090 bond_rebalance_port(struct port *port)
2092 struct slave_balance bals[DP_MAX_PORTS];
2094 struct bond_entry *hashes[BOND_MASK + 1];
2095 struct slave_balance *b, *from, *to;
2096 struct bond_entry *e;
2099 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2100 * descending order of tx_bytes, so that bals[0] represents the most
2101 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2104 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2105 * array for each slave_balance structure, we sort our local array of
2106 * hashes in order by slave, so that all of the hashes for a given slave
2107 * become contiguous in memory, and then we point each 'hashes' members of
2108 * a slave_balance structure to the start of a contiguous group. */
2109 n_bals = port->n_ifaces;
2110 for (b = bals; b < &bals[n_bals]; b++) {
2111 b->iface = port->ifaces[b - bals];
2116 for (i = 0; i <= BOND_MASK; i++) {
2117 hashes[i] = &port->bond_hash[i];
2119 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2120 for (i = 0; i <= BOND_MASK; i++) {
2122 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2123 b = &bals[e->iface_idx];
2124 b->tx_bytes += e->tx_bytes;
2126 b->hashes = &hashes[i];
2131 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2132 log_bals(bals, n_bals, port);
2134 /* Discard slaves that aren't enabled (which were sorted to the back of the
2135 * array earlier). */
2136 while (!bals[n_bals - 1].iface->enabled) {
2143 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2144 to = &bals[n_bals - 1];
2145 for (from = bals; from < to; ) {
2146 uint64_t overload = from->tx_bytes - to->tx_bytes;
2147 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2148 /* The extra load on 'from' (and all less-loaded slaves), compared
2149 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2150 * it is less than ~1Mbps. No point in rebalancing. */
2152 } else if (from->n_hashes == 1) {
2153 /* 'from' only carries a single MAC hash, so we can't shift any
2154 * load away from it, even though we want to. */
2157 /* 'from' is carrying significantly more load than 'to', and that
2158 * load is split across at least two different hashes. Pick a hash
2159 * to migrate to 'to' (the least-loaded slave), given that doing so
2160 * must not cause 'to''s load to exceed 'from''s load.
2162 * The sort order we use means that we prefer to shift away the
2163 * smallest hashes instead of the biggest ones. There is little
2164 * reason behind this decision; we could use the opposite sort
2165 * order to shift away big hashes ahead of small ones. */
2168 for (i = 0; i < from->n_hashes; i++) {
2169 uint64_t delta = from->hashes[i]->tx_bytes;
2170 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2174 if (i < from->n_hashes) {
2175 bond_shift_load(from, to, from->hashes[i]);
2177 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2178 * point to different slave_balance structures. It is only
2179 * valid to do these two operations in a row at all because we
2180 * know that 'from' will not move past 'to' and vice versa. */
2181 resort_bals(from, bals, n_bals);
2182 resort_bals(to, bals, n_bals);
2189 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2190 * historical data to decay to <1% in 7 rebalancing runs. */
2191 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2197 bond_send_learning_packets(struct port *port)
2199 struct bridge *br = port->bridge;
2200 struct mac_entry *e;
2201 struct ofpbuf packet;
2202 int error, n_packets, n_errors;
2204 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2208 ofpbuf_init(&packet, 128);
2209 error = n_packets = n_errors = 0;
2210 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2211 static const char s[] = "Open vSwitch Bond Failover";
2212 union ofp_action actions[2], *a;
2213 struct eth_header *eth;
2214 struct llc_snap_header *llc_snap;
2220 if (e->port == port->port_idx
2221 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2225 /* Compose packet to send. */
2226 ofpbuf_clear(&packet);
2227 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2228 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2229 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2230 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2232 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2233 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2234 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2236 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2237 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2238 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2239 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2240 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2242 /* Compose actions. */
2243 memset(actions, 0, sizeof actions);
2246 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2247 a->vlan_vid.len = htons(sizeof *a);
2248 a->vlan_vid.vlan_vid = htons(e->vlan);
2251 a->output.type = htons(OFPAT_OUTPUT);
2252 a->output.len = htons(sizeof *a);
2253 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2258 flow_extract(&packet, ODPP_NONE, &flow);
2259 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2266 ofpbuf_uninit(&packet);
2269 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2270 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2271 "packets, last error was: %s",
2272 port->name, n_errors, n_packets, strerror(error));
2274 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2275 port->name, n_packets);
2279 /* Bonding unixctl user interface functions. */
2282 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2284 struct ds ds = DS_EMPTY_INITIALIZER;
2285 const struct bridge *br;
2287 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2289 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2292 for (i = 0; i < br->n_ports; i++) {
2293 const struct port *port = br->ports[i];
2294 if (port->n_ifaces > 1) {
2297 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2298 for (j = 0; j < port->n_ifaces; j++) {
2299 const struct iface *iface = port->ifaces[j];
2301 ds_put_cstr(&ds, ", ");
2303 ds_put_cstr(&ds, iface->name);
2305 ds_put_char(&ds, '\n');
2309 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2313 static struct port *
2314 bond_find(const char *name)
2316 const struct bridge *br;
2318 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2321 for (i = 0; i < br->n_ports; i++) {
2322 struct port *port = br->ports[i];
2323 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2332 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2334 struct ds ds = DS_EMPTY_INITIALIZER;
2335 const struct port *port;
2338 port = bond_find(args);
2340 unixctl_command_reply(conn, 501, "no such bond");
2344 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2345 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2346 ds_put_format(&ds, "next rebalance: %lld ms\n",
2347 port->bridge->bond_next_rebalance - time_msec());
2348 for (j = 0; j < port->n_ifaces; j++) {
2349 const struct iface *iface = port->ifaces[j];
2350 struct bond_entry *be;
2353 ds_put_format(&ds, "slave %s: %s\n",
2354 iface->name, iface->enabled ? "enabled" : "disabled");
2355 if (j == port->active_iface) {
2356 ds_put_cstr(&ds, "\tactive slave\n");
2358 if (iface->delay_expires != LLONG_MAX) {
2359 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2360 iface->enabled ? "downdelay" : "updelay",
2361 iface->delay_expires - time_msec());
2365 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2366 int hash = be - port->bond_hash;
2367 struct mac_entry *me;
2369 if (be->iface_idx != j) {
2373 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2374 hash, be->tx_bytes / 1024);
2377 if (!port->bridge->ml) {
2381 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2382 &port->bridge->ml->lrus) {
2385 if (bond_hash(me->mac) == hash
2386 && me->port != port->port_idx
2387 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2388 && dp_ifidx == iface->dp_ifidx)
2390 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2391 ETH_ADDR_ARGS(me->mac));
2396 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2401 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2403 char *args = (char *) args_;
2404 char *save_ptr = NULL;
2405 char *bond_s, *hash_s, *slave_s;
2406 uint8_t mac[ETH_ADDR_LEN];
2408 struct iface *iface;
2409 struct bond_entry *entry;
2412 bond_s = strtok_r(args, " ", &save_ptr);
2413 hash_s = strtok_r(NULL, " ", &save_ptr);
2414 slave_s = strtok_r(NULL, " ", &save_ptr);
2416 unixctl_command_reply(conn, 501,
2417 "usage: bond/migrate BOND HASH SLAVE");
2421 port = bond_find(bond_s);
2423 unixctl_command_reply(conn, 501, "no such bond");
2427 if (sscanf(hash_s, "%"SCNx8":%"SCNx8":%"SCNx8":%"SCNx8":%"SCNx8":%"SCNx8,
2428 &mac[0], &mac[1], &mac[2], &mac[3], &mac[4], &mac[5]) == 6) {
2429 hash = bond_hash(mac);
2430 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2431 hash = atoi(hash_s) & BOND_MASK;
2433 unixctl_command_reply(conn, 501, "bad hash");
2437 iface = port_lookup_iface(port, slave_s);
2439 unixctl_command_reply(conn, 501, "no such slave");
2443 if (!iface->enabled) {
2444 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2448 entry = &port->bond_hash[hash];
2449 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2450 entry->iface_idx = iface->port_ifidx;
2451 entry->iface_tag = tag_create_random();
2452 unixctl_command_reply(conn, 200, "migrated");
2456 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2458 char *args = (char *) args_;
2459 char *save_ptr = NULL;
2460 char *bond_s, *slave_s;
2462 struct iface *iface;
2464 bond_s = strtok_r(args, " ", &save_ptr);
2465 slave_s = strtok_r(NULL, " ", &save_ptr);
2467 unixctl_command_reply(conn, 501,
2468 "usage: bond/set-active-slave BOND SLAVE");
2472 port = bond_find(bond_s);
2474 unixctl_command_reply(conn, 501, "no such bond");
2478 iface = port_lookup_iface(port, slave_s);
2480 unixctl_command_reply(conn, 501, "no such slave");
2484 if (!iface->enabled) {
2485 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2489 if (port->active_iface != iface->port_ifidx) {
2490 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2491 port->active_iface = iface->port_ifidx;
2492 port->active_iface_tag = tag_create_random();
2493 VLOG_INFO("port %s: active interface is now %s",
2494 port->name, iface->name);
2495 bond_send_learning_packets(port);
2496 unixctl_command_reply(conn, 200, "done");
2498 unixctl_command_reply(conn, 200, "no change");
2503 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2505 char *args = (char *) args_;
2506 char *save_ptr = NULL;
2507 char *bond_s, *slave_s;
2509 struct iface *iface;
2511 bond_s = strtok_r(args, " ", &save_ptr);
2512 slave_s = strtok_r(NULL, " ", &save_ptr);
2514 unixctl_command_reply(conn, 501,
2515 "usage: bond/enable/disable-slave BOND SLAVE");
2519 port = bond_find(bond_s);
2521 unixctl_command_reply(conn, 501, "no such bond");
2525 iface = port_lookup_iface(port, slave_s);
2527 unixctl_command_reply(conn, 501, "no such slave");
2531 bond_enable_slave(iface, enable);
2532 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2536 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2538 enable_slave(conn, args, true);
2542 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2544 enable_slave(conn, args, false);
2550 unixctl_command_register("bond/list", bond_unixctl_list);
2551 unixctl_command_register("bond/show", bond_unixctl_show);
2552 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2553 unixctl_command_register("bond/set-active-slave",
2554 bond_unixctl_set_active_slave);
2555 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2556 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2559 /* Port functions. */
2562 port_create(struct bridge *br, const char *name)
2566 port = xcalloc(1, sizeof *port);
2568 port->port_idx = br->n_ports;
2570 port->trunks = NULL;
2571 port->name = xstrdup(name);
2572 port->active_iface = -1;
2573 port->stp_state = STP_DISABLED;
2574 port->stp_state_tag = 0;
2576 if (br->n_ports >= br->allocated_ports) {
2577 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2580 br->ports[br->n_ports++] = port;
2582 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2587 port_reconfigure(struct port *port)
2589 bool bonded = cfg_has_section("bonding.%s", port->name);
2590 struct svec old_ifaces, new_ifaces;
2591 unsigned long *trunks;
2595 /* Collect old and new interfaces. */
2596 svec_init(&old_ifaces);
2597 svec_init(&new_ifaces);
2598 for (i = 0; i < port->n_ifaces; i++) {
2599 svec_add(&old_ifaces, port->ifaces[i]->name);
2601 svec_sort(&old_ifaces);
2603 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2604 if (!new_ifaces.n) {
2605 VLOG_ERR("port %s: no interfaces specified for bonded port",
2607 } else if (new_ifaces.n == 1) {
2608 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2612 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2613 if (port->updelay < 0) {
2616 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2617 if (port->downdelay < 0) {
2618 port->downdelay = 0;
2621 svec_init(&new_ifaces);
2622 svec_add(&new_ifaces, port->name);
2625 /* Get rid of deleted interfaces and add new interfaces. */
2626 for (i = 0; i < port->n_ifaces; i++) {
2627 struct iface *iface = port->ifaces[i];
2628 if (!svec_contains(&new_ifaces, iface->name)) {
2629 iface_destroy(iface);
2634 for (i = 0; i < new_ifaces.n; i++) {
2635 const char *name = new_ifaces.names[i];
2636 if (!svec_contains(&old_ifaces, name)) {
2637 iface_create(port, name);
2643 if (cfg_has("vlan.%s.tag", port->name)) {
2645 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2646 if (vlan >= 0 && vlan <= 4095) {
2647 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2650 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2651 * they even work as-is. But they have not been tested. */
2652 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2656 if (port->vlan != vlan) {
2658 bridge_flush(port->bridge);
2661 /* Get trunked VLANs. */
2664 size_t n_trunks, n_errors;
2667 trunks = bitmap_allocate(4096);
2668 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2670 for (i = 0; i < n_trunks; i++) {
2671 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2673 bitmap_set1(trunks, trunk);
2679 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2680 port->name, n_trunks);
2682 if (n_errors == n_trunks) {
2684 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2687 bitmap_set_multiple(trunks, 0, 4096, 1);
2690 if (cfg_has("vlan.%s.trunks", port->name)) {
2691 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2692 port->name, port->name);
2696 ? port->trunks != NULL
2697 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2698 bridge_flush(port->bridge);
2700 bitmap_free(port->trunks);
2701 port->trunks = trunks;
2703 svec_destroy(&old_ifaces);
2704 svec_destroy(&new_ifaces);
2708 port_destroy(struct port *port)
2711 struct bridge *br = port->bridge;
2715 proc_net_compat_update_vlan(port->name, NULL, 0);
2717 for (i = 0; i < MAX_MIRRORS; i++) {
2718 struct mirror *m = br->mirrors[i];
2719 if (m && m->out_port == port) {
2724 while (port->n_ifaces > 0) {
2725 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2728 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2729 del->port_idx = port->port_idx;
2732 bitmap_free(port->trunks);
2739 static struct port *
2740 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2742 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2743 return iface ? iface->port : NULL;
2746 static struct port *
2747 port_lookup(const struct bridge *br, const char *name)
2751 for (i = 0; i < br->n_ports; i++) {
2752 struct port *port = br->ports[i];
2753 if (!strcmp(port->name, name)) {
2760 static struct iface *
2761 port_lookup_iface(const struct port *port, const char *name)
2765 for (j = 0; j < port->n_ifaces; j++) {
2766 struct iface *iface = port->ifaces[j];
2767 if (!strcmp(iface->name, name)) {
2775 port_update_bonding(struct port *port)
2777 if (port->n_ifaces < 2) {
2778 /* Not a bonded port. */
2779 if (port->bond_hash) {
2780 free(port->bond_hash);
2781 port->bond_hash = NULL;
2782 proc_net_compat_update_bond(port->name, NULL);
2785 if (!port->bond_hash) {
2788 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2789 for (i = 0; i <= BOND_MASK; i++) {
2790 struct bond_entry *e = &port->bond_hash[i];
2794 port->no_ifaces_tag = tag_create_random();
2795 bond_choose_active_iface(port);
2797 port_update_bond_compat(port);
2802 port_update_bond_compat(struct port *port)
2804 struct compat_bond bond;
2807 if (port->n_ifaces < 2) {
2812 bond.updelay = port->updelay;
2813 bond.downdelay = port->downdelay;
2814 bond.n_slaves = port->n_ifaces;
2815 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2816 for (i = 0; i < port->n_ifaces; i++) {
2817 struct iface *iface = port->ifaces[i];
2818 struct compat_bond_slave *slave = &bond.slaves[i];
2819 slave->name = iface->name;
2820 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2821 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2825 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2827 proc_net_compat_update_bond(port->name, &bond);
2832 port_update_vlan_compat(struct port *port)
2834 struct bridge *br = port->bridge;
2835 char *vlandev_name = NULL;
2837 if (port->vlan > 0) {
2838 /* Figure out the name that the VLAN device should actually have, if it
2839 * existed. This takes some work because the VLAN device would not
2840 * have port->name in its name; rather, it would have the trunk port's
2841 * name, and 'port' would be attached to a bridge that also had the
2842 * VLAN device one of its ports. So we need to find a trunk port that
2843 * includes port->vlan.
2845 * There might be more than one candidate. This doesn't happen on
2846 * XenServer, so if it happens we just pick the first choice in
2847 * alphabetical order instead of creating multiple VLAN devices. */
2849 for (i = 0; i < br->n_ports; i++) {
2850 struct port *p = br->ports[i];
2851 if (port_trunks_vlan(p, port->vlan)
2853 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2855 const uint8_t *ea = p->ifaces[0]->mac;
2856 if (!eth_addr_is_multicast(ea) &&
2857 !eth_addr_is_reserved(ea) &&
2858 !eth_addr_is_zero(ea)) {
2859 vlandev_name = p->name;
2864 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2867 /* Interface functions. */
2870 iface_create(struct port *port, const char *name)
2872 struct iface *iface;
2874 iface = xcalloc(1, sizeof *iface);
2876 iface->port_ifidx = port->n_ifaces;
2877 iface->name = xstrdup(name);
2878 iface->dp_ifidx = -1;
2879 iface->tag = tag_create_random();
2880 iface->delay_expires = LLONG_MAX;
2882 netdev_nodev_get_etheraddr(name, iface->mac);
2883 netdev_nodev_get_carrier(name, &iface->enabled);
2885 if (port->n_ifaces >= port->allocated_ifaces) {
2886 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2887 sizeof *port->ifaces);
2889 port->ifaces[port->n_ifaces++] = iface;
2890 if (port->n_ifaces > 1) {
2891 port->bridge->has_bonded_ports = true;
2894 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2896 port_update_bonding(port);
2897 bridge_flush(port->bridge);
2901 iface_destroy(struct iface *iface)
2904 struct port *port = iface->port;
2905 struct bridge *br = port->bridge;
2906 bool del_active = port->active_iface == iface->port_ifidx;
2909 if (iface->dp_ifidx >= 0) {
2910 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
2913 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
2914 del->port_ifidx = iface->port_ifidx;
2920 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2921 bond_choose_active_iface(port);
2922 bond_send_learning_packets(port);
2925 port_update_bonding(port);
2926 bridge_flush(port->bridge);
2930 static struct iface *
2931 iface_lookup(const struct bridge *br, const char *name)
2935 for (i = 0; i < br->n_ports; i++) {
2936 struct port *port = br->ports[i];
2937 for (j = 0; j < port->n_ifaces; j++) {
2938 struct iface *iface = port->ifaces[j];
2939 if (!strcmp(iface->name, name)) {
2947 static struct iface *
2948 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2950 return port_array_get(&br->ifaces, dp_ifidx);
2953 /* Port mirroring. */
2956 mirror_reconfigure(struct bridge *br)
2958 struct svec old_mirrors, new_mirrors;
2961 /* Collect old and new mirrors. */
2962 svec_init(&old_mirrors);
2963 svec_init(&new_mirrors);
2964 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
2965 for (i = 0; i < MAX_MIRRORS; i++) {
2966 if (br->mirrors[i]) {
2967 svec_add(&old_mirrors, br->mirrors[i]->name);
2971 /* Get rid of deleted mirrors and add new mirrors. */
2972 svec_sort(&old_mirrors);
2973 assert(svec_is_unique(&old_mirrors));
2974 svec_sort(&new_mirrors);
2975 assert(svec_is_unique(&new_mirrors));
2976 for (i = 0; i < MAX_MIRRORS; i++) {
2977 struct mirror *m = br->mirrors[i];
2978 if (m && !svec_contains(&new_mirrors, m->name)) {
2982 for (i = 0; i < new_mirrors.n; i++) {
2983 const char *name = new_mirrors.names[i];
2984 if (!svec_contains(&old_mirrors, name)) {
2985 mirror_create(br, name);
2988 svec_destroy(&old_mirrors);
2989 svec_destroy(&new_mirrors);
2991 /* Reconfigure all mirrors. */
2992 for (i = 0; i < MAX_MIRRORS; i++) {
2993 if (br->mirrors[i]) {
2994 mirror_reconfigure_one(br->mirrors[i]);
2998 /* Update port reserved status. */
2999 for (i = 0; i < br->n_ports; i++) {
3000 br->ports[i]->is_mirror_output_port = false;
3002 for (i = 0; i < MAX_MIRRORS; i++) {
3003 struct mirror *m = br->mirrors[i];
3004 if (m && m->out_port) {
3005 m->out_port->is_mirror_output_port = true;
3011 mirror_create(struct bridge *br, const char *name)
3016 for (i = 0; ; i++) {
3017 if (i >= MAX_MIRRORS) {
3018 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3019 "cannot create %s", br->name, MAX_MIRRORS, name);
3022 if (!br->mirrors[i]) {
3027 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3030 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3033 m->name = xstrdup(name);
3034 svec_init(&m->src_ports);
3035 svec_init(&m->dst_ports);
3043 mirror_destroy(struct mirror *m)
3046 struct bridge *br = m->bridge;
3049 for (i = 0; i < br->n_ports; i++) {
3050 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3051 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3054 svec_destroy(&m->src_ports);
3055 svec_destroy(&m->dst_ports);
3058 m->bridge->mirrors[m->idx] = NULL;
3066 prune_ports(struct mirror *m, struct svec *ports)
3071 svec_sort_unique(ports);
3074 for (i = 0; i < ports->n; i++) {
3075 const char *name = ports->names[i];
3076 if (port_lookup(m->bridge, name)) {
3077 svec_add(&tmp, name);
3079 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3080 m->bridge->name, m->name, name);
3083 svec_swap(ports, &tmp);
3088 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3092 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3093 * order won't give us numeric sort order. But that's good enough for what
3094 * we need right now. */
3095 svec_sort_unique(vlan_strings);
3097 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3099 for (i = 0; i < vlan_strings->n; i++) {
3100 const char *name = vlan_strings->names[i];
3102 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3103 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3104 m->bridge->name, m->name, name);
3106 (*vlans)[n_vlans++] = vlan;
3113 vlan_is_mirrored(const struct mirror *m, int vlan)
3117 for (i = 0; i < m->n_vlans; i++) {
3118 if (m->vlans[i] == vlan) {
3126 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3130 for (i = 0; i < m->n_vlans; i++) {
3131 if (port_trunks_vlan(p, m->vlans[i])) {
3139 mirror_reconfigure_one(struct mirror *m)
3141 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3142 struct svec src_ports, dst_ports, ports;
3143 struct svec vlan_strings;
3144 mirror_mask_t mirror_bit;
3145 const char *out_port_name;
3146 struct port *out_port;
3151 bool mirror_all_ports;
3153 /* Get output port. */
3154 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3155 m->bridge->name, m->name);
3156 if (out_port_name) {
3157 out_port = port_lookup(m->bridge, out_port_name);
3159 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3160 "named %s", pfx, m->bridge->name, out_port_name);
3167 if (cfg_has("%s.output.vlan", pfx)) {
3168 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3169 "ignoring %s.output.vlan", pfx, pfx, pfx);
3171 } else if (cfg_has("%s.output.vlan", pfx)) {
3173 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3175 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3176 "but exactly one is required; disabling port mirror %s",
3177 pfx, pfx, pfx, pfx);
3183 /* Get all the ports, and drop duplicates and ports that don't exist. */
3184 svec_init(&src_ports);
3185 svec_init(&dst_ports);
3187 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3188 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3189 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3190 svec_append(&src_ports, &ports);
3191 svec_append(&dst_ports, &ports);
3192 svec_destroy(&ports);
3193 prune_ports(m, &src_ports);
3194 prune_ports(m, &dst_ports);
3196 /* Get all the vlans, and drop duplicate and invalid vlans. */
3197 svec_init(&vlan_strings);
3198 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3199 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3200 svec_destroy(&vlan_strings);
3202 /* Update mirror data. */
3203 if (!svec_equal(&m->src_ports, &src_ports)
3204 || !svec_equal(&m->dst_ports, &dst_ports)
3205 || m->n_vlans != n_vlans
3206 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3207 || m->out_port != out_port
3208 || m->out_vlan != out_vlan) {
3209 bridge_flush(m->bridge);
3211 svec_swap(&m->src_ports, &src_ports);
3212 svec_swap(&m->dst_ports, &dst_ports);
3215 m->n_vlans = n_vlans;
3216 m->out_port = out_port;
3217 m->out_vlan = out_vlan;
3219 /* If no selection criteria have been given, mirror for all ports. */
3220 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3223 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3224 for (i = 0; i < m->bridge->n_ports; i++) {
3225 struct port *port = m->bridge->ports[i];
3227 if (mirror_all_ports
3228 || svec_contains(&m->src_ports, port->name)
3231 ? port_trunks_any_mirrored_vlan(m, port)
3232 : vlan_is_mirrored(m, port->vlan)))) {
3233 port->src_mirrors |= mirror_bit;
3235 port->src_mirrors &= ~mirror_bit;
3238 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3239 port->dst_mirrors |= mirror_bit;
3241 port->dst_mirrors &= ~mirror_bit;
3246 svec_destroy(&src_ports);
3247 svec_destroy(&dst_ports);
3251 /* Spanning tree protocol. */
3253 static void brstp_update_port_state(struct port *);
3256 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3258 struct bridge *br = br_;
3259 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3260 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3262 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3264 } else if (eth_addr_is_zero(iface->mac)) {
3265 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3268 union ofp_action action;
3269 struct eth_header *eth = pkt->l2;
3272 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3274 memset(&action, 0, sizeof action);
3275 action.type = htons(OFPAT_OUTPUT);
3276 action.output.len = htons(sizeof action);
3277 action.output.port = htons(port_no);
3279 flow_extract(pkt, ODPP_NONE, &flow);
3280 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3286 brstp_reconfigure(struct bridge *br)
3290 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3292 stp_destroy(br->stp);
3298 uint64_t bridge_address, bridge_id;
3299 int bridge_priority;
3301 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3302 if (!bridge_address) {
3304 bridge_address = (stp_get_bridge_id(br->stp)
3305 & ((UINT64_C(1) << 48) - 1));
3307 uint8_t mac[ETH_ADDR_LEN];
3308 eth_addr_random(mac);
3309 bridge_address = eth_addr_to_uint64(mac);
3313 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3315 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3317 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3320 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3322 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3323 br->stp_last_tick = time_msec();
3326 if (bridge_id != stp_get_bridge_id(br->stp)) {
3327 stp_set_bridge_id(br->stp, bridge_id);
3332 for (i = 0; i < br->n_ports; i++) {
3333 struct port *p = br->ports[i];
3335 struct stp_port *sp;
3336 int path_cost, priority;
3342 dp_ifidx = p->ifaces[0]->dp_ifidx;
3343 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3347 sp = stp_get_port(br->stp, dp_ifidx);
3348 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3349 "stp.%s.port.%s.enabled",
3351 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3352 br->name, p->name));
3353 if (p->is_mirror_output_port) {
3356 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3357 bridge_flush(br); /* Might not be necessary. */
3359 stp_port_enable(sp);
3361 stp_port_disable(sp);
3365 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3367 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3369 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3370 "stp.%s.port.%s.priority",
3372 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3374 : STP_DEFAULT_PORT_PRIORITY);
3375 stp_port_set_priority(sp, priority);
3378 brstp_adjust_timers(br);
3380 for (i = 0; i < br->n_ports; i++) {
3381 brstp_update_port_state(br->ports[i]);
3386 brstp_update_port_state(struct port *p)
3388 struct bridge *br = p->bridge;
3389 enum stp_state state;
3391 /* Figure out new state. */
3392 state = STP_DISABLED;
3393 if (br->stp && p->n_ifaces > 0) {
3394 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3395 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3396 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3401 if (p->stp_state != state) {
3402 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3403 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3404 p->name, stp_state_name(p->stp_state),
3405 stp_state_name(state));
3406 if (p->stp_state == STP_DISABLED) {
3409 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3411 p->stp_state = state;
3412 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3413 : tag_create_random());
3418 brstp_adjust_timers(struct bridge *br)
3420 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3421 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3422 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3424 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3425 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3426 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3430 brstp_run(struct bridge *br)
3433 long long int now = time_msec();
3434 long long int elapsed = now - br->stp_last_tick;
3435 struct stp_port *sp;
3438 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3439 br->stp_last_tick = now;
3441 while (stp_get_changed_port(br->stp, &sp)) {
3442 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3444 brstp_update_port_state(p);
3451 brstp_wait(struct bridge *br)
3454 poll_timer_wait(1000);