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 /* These members are always valid. */
75 struct port *port; /* Containing port. */
76 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 tag_type tag; /* Tag associated with this interface. */
79 long long delay_expires; /* Time after which 'enabled' may change. */
81 /* These members are valid only after bridge_reconfigure() causes them to
83 int dp_ifidx; /* Index within kernel datapath. */
84 struct netdev *netdev; /* Network device. */
85 bool enabled; /* May be chosen for flows? */
88 #define BOND_MASK 0xff
90 int iface_idx; /* Index of assigned iface, or -1 if none. */
91 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
92 tag_type iface_tag; /* Tag associated with iface_idx. */
95 #define MAX_MIRRORS 32
96 typedef uint32_t mirror_mask_t;
97 #define MIRROR_MASK_C(X) UINT32_C(X)
98 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
100 struct bridge *bridge;
104 /* Selection criteria. */
105 struct svec src_ports;
106 struct svec dst_ports;
111 struct port *out_port;
115 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
117 struct bridge *bridge;
119 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
120 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
123 /* An ordinary bridge port has 1 interface.
124 * A bridge port for bonding has at least 2 interfaces. */
125 struct iface **ifaces;
126 size_t n_ifaces, allocated_ifaces;
129 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
130 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
131 tag_type active_iface_tag; /* Tag for bcast flows. */
132 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
133 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
135 /* Port mirroring info. */
136 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
137 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
138 bool is_mirror_output_port; /* Does port mirroring send frames here? */
140 /* Spanning tree info. */
141 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
142 tag_type stp_state_tag; /* Tag for STP state change. */
145 #define DP_MAX_PORTS 255
147 struct list node; /* Node in global list of bridges. */
148 char *name; /* User-specified arbitrary name. */
149 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
150 bool sent_config_request; /* Successfully sent config request? */
151 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
153 /* Support for remote controllers. */
154 char *controller; /* NULL if there is no remote controller;
155 * "discover" to do controller discovery;
156 * otherwise a vconn name. */
158 /* OpenFlow switch processing. */
159 struct ofproto *ofproto; /* OpenFlow switch. */
161 /* Kernel datapath information. */
162 struct dpif *dpif; /* Datapath. */
163 struct port_array ifaces; /* Indexed by kernel datapath port number. */
167 size_t n_ports, allocated_ports;
170 bool has_bonded_ports;
171 long long int bond_next_rebalance;
176 /* Flow statistics gathering. */
177 time_t next_stats_request;
179 /* Port mirroring. */
180 struct mirror *mirrors[MAX_MIRRORS];
184 long long int stp_last_tick;
187 /* List of all bridges. */
188 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
190 /* Maximum number of datapaths. */
191 enum { DP_MAX = 256 };
193 static struct bridge *bridge_create(const char *name);
194 static void bridge_destroy(struct bridge *);
195 static struct bridge *bridge_lookup(const char *name);
196 static int bridge_run_one(struct bridge *);
197 static void bridge_reconfigure_one(struct bridge *);
198 static void bridge_reconfigure_controller(struct bridge *);
199 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
200 static void bridge_fetch_dp_ifaces(struct bridge *);
201 static void bridge_flush(struct bridge *);
202 static void bridge_pick_local_hw_addr(struct bridge *,
203 uint8_t ea[ETH_ADDR_LEN],
204 const char **devname);
205 static uint64_t bridge_pick_datapath_id(struct bridge *,
206 const uint8_t bridge_ea[ETH_ADDR_LEN],
207 const char *devname);
208 static uint64_t dpid_from_hash(const void *, size_t nbytes);
210 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
212 static void bond_init(void);
213 static void bond_run(struct bridge *);
214 static void bond_wait(struct bridge *);
215 static void bond_rebalance_port(struct port *);
216 static void bond_send_learning_packets(struct port *);
218 static void port_create(struct bridge *, const char *name);
219 static void port_reconfigure(struct port *);
220 static void port_destroy(struct port *);
221 static struct port *port_lookup(const struct bridge *, const char *name);
222 static struct iface *port_lookup_iface(const struct port *, const char *name);
223 static struct port *port_from_dp_ifidx(const struct bridge *,
225 static void port_update_bond_compat(struct port *);
226 static void port_update_vlan_compat(struct port *);
227 static void port_update_bonding(struct port *);
229 static void mirror_create(struct bridge *, const char *name);
230 static void mirror_destroy(struct mirror *);
231 static void mirror_reconfigure(struct bridge *);
232 static void mirror_reconfigure_one(struct mirror *);
233 static bool vlan_is_mirrored(const struct mirror *, int vlan);
235 static void brstp_reconfigure(struct bridge *);
236 static void brstp_adjust_timers(struct bridge *);
237 static void brstp_run(struct bridge *);
238 static void brstp_wait(struct bridge *);
240 static void iface_create(struct port *, const char *name);
241 static void iface_destroy(struct iface *);
242 static struct iface *iface_lookup(const struct bridge *, const char *name);
243 static struct iface *iface_from_dp_ifidx(const struct bridge *,
246 /* Hooks into ofproto processing. */
247 static struct ofhooks bridge_ofhooks;
249 /* Public functions. */
251 /* Adds the name of each interface used by a bridge, including local and
252 * internal ports, to 'svec'. */
254 bridge_get_ifaces(struct svec *svec)
256 struct bridge *br, *next;
259 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
260 for (i = 0; i < br->n_ports; i++) {
261 struct port *port = br->ports[i];
263 for (j = 0; j < port->n_ifaces; j++) {
264 struct iface *iface = port->ifaces[j];
265 if (iface->dp_ifidx < 0) {
266 VLOG_ERR("%s interface not in datapath %s, ignoring",
267 iface->name, dpif_name(br->dpif));
269 if (iface->dp_ifidx != ODPP_LOCAL) {
270 svec_add(svec, iface->name);
278 /* The caller must already have called cfg_read(). */
282 struct svec dpif_names;
285 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
287 dp_enumerate(&dpif_names);
288 for (i = 0; i < dpif_names.n; i++) {
289 const char *dpif_name = dpif_names.names[i];
293 retval = dpif_open(dpif_name, &dpif);
295 struct svec all_names;
298 svec_init(&all_names);
299 dpif_get_all_names(dpif, &all_names);
300 for (j = 0; j < all_names.n; j++) {
301 if (cfg_has("bridge.%s.port", all_names.names[j])) {
307 svec_destroy(&all_names);
313 bridge_reconfigure();
318 config_string_change(const char *key, char **valuep)
320 const char *value = cfg_get_string(0, "%s", key);
321 if (value && (!*valuep || strcmp(value, *valuep))) {
323 *valuep = xstrdup(value);
331 bridge_configure_ssl(void)
333 /* XXX SSL should be configurable on a per-bridge basis.
334 * XXX should be possible to de-configure SSL. */
335 static char *private_key_file;
336 static char *certificate_file;
337 static char *cacert_file;
340 if (config_string_change("ssl.private-key", &private_key_file)) {
341 vconn_ssl_set_private_key_file(private_key_file);
344 if (config_string_change("ssl.certificate", &certificate_file)) {
345 vconn_ssl_set_certificate_file(certificate_file);
348 /* We assume that even if the filename hasn't changed, if the CA cert
349 * file has been removed, that we want to move back into
350 * boot-strapping mode. This opens a small security hole, because
351 * the old certificate will still be trusted until vSwitch is
352 * restarted. We may want to address this in vconn's SSL library. */
353 if (config_string_change("ssl.ca-cert", &cacert_file)
354 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
355 vconn_ssl_set_ca_cert_file(cacert_file,
356 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
361 /* iterate_and_prune_ifaces() callback function that opens the network device
362 * for 'iface', if it is not already open, and retrieves the interface's MAC
363 * address and carrier status. */
365 init_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
370 } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE,
372 netdev_get_carrier(iface->netdev, &iface->enabled);
375 /* If the network device can't be opened, then we're not going to try
376 * to do anything with this interface. */
382 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
385 struct iface **local_ifacep = local_ifacep_;
387 if (iface->dp_ifidx >= 0) {
388 if (iface->dp_ifidx == ODPP_LOCAL) {
389 *local_ifacep = iface;
391 VLOG_DBG("%s has interface %s on port %d",
393 iface->name, iface->dp_ifidx);
396 VLOG_ERR("%s interface not in %s, dropping",
397 iface->name, dpif_name(br->dpif));
402 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
403 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
404 * deletes from 'br' any ports that no longer have any interfaces. */
406 iterate_and_prune_ifaces(struct bridge *br,
407 bool (*cb)(struct bridge *, struct iface *,
413 for (i = 0; i < br->n_ports; ) {
414 struct port *port = br->ports[i];
415 for (j = 0; j < port->n_ifaces; ) {
416 struct iface *iface = port->ifaces[j];
417 if (cb(br, iface, aux)) {
420 iface_destroy(iface);
424 if (port->n_ifaces) {
427 VLOG_ERR("%s port has no interfaces, dropping", port->name);
434 bridge_reconfigure(void)
436 struct svec old_br, new_br;
437 struct bridge *br, *next;
440 COVERAGE_INC(bridge_reconfigure);
442 /* Collect old and new bridges. */
445 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
446 svec_add(&old_br, br->name);
448 cfg_get_subsections(&new_br, "bridge");
450 /* Get rid of deleted bridges and add new bridges. */
453 assert(svec_is_unique(&old_br));
454 assert(svec_is_unique(&new_br));
455 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
456 if (!svec_contains(&new_br, br->name)) {
460 for (i = 0; i < new_br.n; i++) {
461 const char *name = new_br.names[i];
462 if (!svec_contains(&old_br, name)) {
466 svec_destroy(&old_br);
467 svec_destroy(&new_br);
471 bridge_configure_ssl();
474 /* Reconfigure all bridges. */
475 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
476 bridge_reconfigure_one(br);
479 /* Add and delete ports on all datapaths.
481 * The kernel will reject any attempt to add a given port to a datapath if
482 * that port already belongs to a different datapath, so we must do all
483 * port deletions before any port additions. */
484 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
485 struct odp_port *dpif_ports;
487 struct svec want_ifaces;
489 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
490 bridge_get_all_ifaces(br, &want_ifaces);
491 for (i = 0; i < n_dpif_ports; i++) {
492 const struct odp_port *p = &dpif_ports[i];
493 if (!svec_contains(&want_ifaces, p->devname)
494 && strcmp(p->devname, br->name)) {
495 int retval = dpif_port_del(br->dpif, p->port);
497 VLOG_ERR("failed to remove %s interface from %s: %s",
498 p->devname, dpif_name(br->dpif),
503 svec_destroy(&want_ifaces);
506 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
507 struct odp_port *dpif_ports;
509 struct svec cur_ifaces, want_ifaces, add_ifaces;
511 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
512 svec_init(&cur_ifaces);
513 for (i = 0; i < n_dpif_ports; i++) {
514 svec_add(&cur_ifaces, dpif_ports[i].devname);
517 svec_sort_unique(&cur_ifaces);
518 bridge_get_all_ifaces(br, &want_ifaces);
519 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
521 for (i = 0; i < add_ifaces.n; i++) {
522 const char *if_name = add_ifaces.names[i];
523 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
524 int flags = internal ? ODP_PORT_INTERNAL : 0;
525 int error = dpif_port_add(br->dpif, if_name, flags, NULL);
526 if (error == EXFULL) {
527 VLOG_ERR("ran out of valid port numbers on %s",
528 dpif_name(br->dpif));
531 VLOG_ERR("failed to add %s interface to %s: %s",
532 if_name, dpif_name(br->dpif), strerror(error));
535 svec_destroy(&cur_ifaces);
536 svec_destroy(&want_ifaces);
537 svec_destroy(&add_ifaces);
539 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
542 struct iface *local_iface = NULL;
544 uint8_t engine_type, engine_id;
545 bool add_id_to_iface = false;
546 struct svec nf_hosts;
548 bridge_fetch_dp_ifaces(br);
549 iterate_and_prune_ifaces(br, init_iface_netdev, NULL);
552 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, &local_iface);
554 /* Pick local port hardware address, datapath ID. */
555 bridge_pick_local_hw_addr(br, ea, &devname);
557 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
559 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
560 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
561 "Ethernet address: %s",
562 br->name, strerror(error));
566 dpid = bridge_pick_datapath_id(br, ea, devname);
567 ofproto_set_datapath_id(br->ofproto, dpid);
569 /* Set NetFlow configuration on this bridge. */
570 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
571 if (cfg_has("netflow.%s.engine-type", br->name)) {
572 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
575 if (cfg_has("netflow.%s.engine-id", br->name)) {
576 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
578 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
579 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
582 if (add_id_to_iface && engine_id > 0x7f) {
583 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
584 "another vswitch, choose an engine id less than 128",
587 if (add_id_to_iface && br->n_ports > 0x1ff) {
588 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
589 "another port when 512 or more ports are used",
592 svec_init(&nf_hosts);
593 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
594 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
595 engine_id, add_id_to_iface)) {
596 VLOG_ERR("bridge %s: problem setting netflow collectors",
600 /* Update the controller and related settings. It would be more
601 * straightforward to call this from bridge_reconfigure_one(), but we
602 * can't do it there for two reasons. First, and most importantly, at
603 * that point we don't know the dp_ifidx of any interfaces that have
604 * been added to the bridge (because we haven't actually added them to
605 * the datapath). Second, at that point we haven't set the datapath ID
606 * yet; when a controller is configured, resetting the datapath ID will
607 * immediately disconnect from the controller, so it's better to set
608 * the datapath ID before the controller. */
609 bridge_reconfigure_controller(br);
611 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
612 for (i = 0; i < br->n_ports; i++) {
613 struct port *port = br->ports[i];
614 port_update_vlan_compat(port);
615 port_update_bonding(port);
618 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
619 brstp_reconfigure(br);
624 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
625 const char **devname)
627 uint64_t requested_ea;
633 /* Did the user request a particular MAC? */
634 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
636 eth_addr_from_uint64(requested_ea, ea);
637 if (eth_addr_is_multicast(ea)) {
638 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
639 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
640 } else if (eth_addr_is_zero(ea)) {
641 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
647 /* Otherwise choose the minimum MAC address among all of the interfaces.
648 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
649 * MAC of the physical interface in such an environment.) */
650 memset(ea, 0xff, sizeof ea);
651 for (i = 0; i < br->n_ports; i++) {
652 struct port *port = br->ports[i];
653 if (port->is_mirror_output_port) {
656 for (j = 0; j < port->n_ifaces; j++) {
657 struct iface *iface = port->ifaces[j];
658 uint8_t iface_ea[ETH_ADDR_LEN];
659 if (iface->dp_ifidx == ODPP_LOCAL
660 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
663 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
665 if (!eth_addr_is_multicast(iface_ea) &&
666 !eth_addr_is_reserved(iface_ea) &&
667 !eth_addr_is_zero(iface_ea) &&
668 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
669 memcpy(ea, iface_ea, ETH_ADDR_LEN);
670 *devname = iface->name;
673 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
674 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
675 iface->name, strerror(error));
679 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
680 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
682 VLOG_WARN("bridge %s: using default bridge Ethernet "
683 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
685 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
686 br->name, ETH_ADDR_ARGS(ea));
690 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
691 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
692 * a network device, then that network device's name must be passed in as
693 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
694 * passed in as a null pointer. */
696 bridge_pick_datapath_id(struct bridge *br,
697 const uint8_t bridge_ea[ETH_ADDR_LEN],
701 * The procedure for choosing a bridge MAC address will, in the most
702 * ordinary case, also choose a unique MAC that we can use as a datapath
703 * ID. In some special cases, though, multiple bridges will end up with
704 * the same MAC address. This is OK for the bridges, but it will confuse
705 * the OpenFlow controller, because each datapath needs a unique datapath
708 * Datapath IDs must be unique. It is also very desirable that they be
709 * stable from one run to the next, so that policy set on a datapath
714 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
721 if (!netdev_get_vlan_vid(devname, &vlan)) {
723 * A bridge whose MAC address is taken from a VLAN network device
724 * (that is, a network device created with vconfig(8) or similar
725 * tool) will have the same MAC address as a bridge on the VLAN
726 * device's physical network device.
728 * Handle this case by hashing the physical network device MAC
729 * along with the VLAN identifier.
731 uint8_t buf[ETH_ADDR_LEN + 2];
732 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
733 buf[ETH_ADDR_LEN] = vlan >> 8;
734 buf[ETH_ADDR_LEN + 1] = vlan;
735 return dpid_from_hash(buf, sizeof buf);
738 * Assume that this bridge's MAC address is unique, since it
739 * doesn't fit any of the cases we handle specially.
744 * A purely internal bridge, that is, one that has no non-virtual
745 * network devices on it at all, is more difficult because it has no
746 * natural unique identifier at all.
748 * When the host is a XenServer, we handle this case by hashing the
749 * host's UUID with the name of the bridge. Names of bridges are
750 * persistent across XenServer reboots, although they can be reused if
751 * an internal network is destroyed and then a new one is later
752 * created, so this is fairly effective.
754 * When the host is not a XenServer, we punt by using a random MAC
755 * address on each run.
757 const char *host_uuid = xenserver_get_host_uuid();
759 char *combined = xasprintf("%s,%s", host_uuid, br->name);
760 dpid = dpid_from_hash(combined, strlen(combined));
766 return eth_addr_to_uint64(bridge_ea);
770 dpid_from_hash(const void *data, size_t n)
772 uint8_t hash[SHA1_DIGEST_SIZE];
774 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
775 sha1_bytes(data, n, hash);
776 eth_addr_mark_random(hash);
777 return eth_addr_to_uint64(hash);
783 struct bridge *br, *next;
787 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
788 int error = bridge_run_one(br);
790 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
791 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
792 "forcing reconfiguration", br->name);
806 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
807 ofproto_wait(br->ofproto);
808 if (br->controller) {
813 mac_learning_wait(br->ml);
820 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
821 * configuration changes. */
823 bridge_flush(struct bridge *br)
825 COVERAGE_INC(bridge_flush);
828 mac_learning_flush(br->ml);
832 /* Bridge unixctl user interface functions. */
834 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
836 struct ds ds = DS_EMPTY_INITIALIZER;
837 const struct bridge *br;
839 br = bridge_lookup(args);
841 unixctl_command_reply(conn, 501, "no such bridge");
845 ds_put_cstr(&ds, " port VLAN MAC Age\n");
847 const struct mac_entry *e;
848 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
849 if (e->port < 0 || e->port >= br->n_ports) {
852 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
853 br->ports[e->port]->ifaces[0]->dp_ifidx,
854 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
857 unixctl_command_reply(conn, 200, ds_cstr(&ds));
861 /* Bridge reconfiguration functions. */
863 static struct bridge *
864 bridge_create(const char *name)
869 assert(!bridge_lookup(name));
870 br = xcalloc(1, sizeof *br);
872 error = dpif_create(name, &br->dpif);
873 if (error == EEXIST || error == EBUSY) {
874 error = dpif_open(name, &br->dpif);
876 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
877 name, strerror(error));
881 dpif_flow_flush(br->dpif);
883 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
888 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
890 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
891 dpif_delete(br->dpif);
892 dpif_close(br->dpif);
897 br->name = xstrdup(name);
898 br->ml = mac_learning_create();
899 br->sent_config_request = false;
900 eth_addr_random(br->default_ea);
902 port_array_init(&br->ifaces);
905 br->bond_next_rebalance = time_msec() + 10000;
907 list_push_back(&all_bridges, &br->node);
909 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
915 bridge_destroy(struct bridge *br)
920 while (br->n_ports > 0) {
921 port_destroy(br->ports[br->n_ports - 1]);
923 list_remove(&br->node);
924 error = dpif_delete(br->dpif);
925 if (error && error != ENOENT) {
926 VLOG_ERR("failed to delete %s: %s",
927 dpif_name(br->dpif), strerror(error));
929 dpif_close(br->dpif);
930 ofproto_destroy(br->ofproto);
931 free(br->controller);
932 mac_learning_destroy(br->ml);
933 port_array_destroy(&br->ifaces);
940 static struct bridge *
941 bridge_lookup(const char *name)
945 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
946 if (!strcmp(br->name, name)) {
954 bridge_exists(const char *name)
956 return bridge_lookup(name) ? true : false;
960 bridge_get_datapathid(const char *name)
962 struct bridge *br = bridge_lookup(name);
963 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
967 bridge_run_one(struct bridge *br)
971 error = ofproto_run1(br->ofproto);
977 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
982 error = ofproto_run2(br->ofproto, br->flush);
989 bridge_get_controller(const struct bridge *br)
991 const char *controller;
993 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
995 controller = cfg_get_string(0, "mgmt.controller");
997 return controller && controller[0] ? controller : NULL;
1001 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1003 struct svec *ifaces = ifaces_;
1004 if (!svec_contains(ifaces, iface->name)) {
1005 svec_add(ifaces, iface->name);
1009 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1011 br->name, iface->name, iface->port->name);
1017 bridge_reconfigure_one(struct bridge *br)
1019 struct svec old_ports, new_ports, ifaces;
1020 struct svec listeners, old_listeners;
1021 struct svec snoops, old_snoops;
1024 /* Collect old ports. */
1025 svec_init(&old_ports);
1026 for (i = 0; i < br->n_ports; i++) {
1027 svec_add(&old_ports, br->ports[i]->name);
1029 svec_sort(&old_ports);
1030 assert(svec_is_unique(&old_ports));
1032 /* Collect new ports. */
1033 svec_init(&new_ports);
1034 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1035 svec_sort(&new_ports);
1036 if (bridge_get_controller(br)) {
1037 char local_name[IF_NAMESIZE];
1040 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1041 local_name, sizeof local_name);
1042 if (!error && !svec_contains(&new_ports, local_name)) {
1043 svec_add(&new_ports, local_name);
1044 svec_sort(&new_ports);
1047 if (!svec_is_unique(&new_ports)) {
1048 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1049 br->name, svec_get_duplicate(&new_ports));
1050 svec_unique(&new_ports);
1053 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1055 /* Get rid of deleted ports and add new ports. */
1056 for (i = 0; i < br->n_ports; ) {
1057 struct port *port = br->ports[i];
1058 if (!svec_contains(&new_ports, port->name)) {
1064 for (i = 0; i < new_ports.n; i++) {
1065 const char *name = new_ports.names[i];
1066 if (!svec_contains(&old_ports, name)) {
1067 port_create(br, name);
1070 svec_destroy(&old_ports);
1071 svec_destroy(&new_ports);
1073 /* Reconfigure all ports. */
1074 for (i = 0; i < br->n_ports; i++) {
1075 port_reconfigure(br->ports[i]);
1078 /* Check and delete duplicate interfaces. */
1080 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1081 svec_destroy(&ifaces);
1083 /* Delete all flows if we're switching from connected to standalone or vice
1084 * versa. (XXX Should we delete all flows if we are switching from one
1085 * controller to another?) */
1087 /* Configure OpenFlow management listeners. */
1088 svec_init(&listeners);
1089 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1091 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1092 ovs_rundir, br->name));
1093 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1094 svec_clear(&listeners);
1096 svec_sort_unique(&listeners);
1098 svec_init(&old_listeners);
1099 ofproto_get_listeners(br->ofproto, &old_listeners);
1100 svec_sort_unique(&old_listeners);
1102 if (!svec_equal(&listeners, &old_listeners)) {
1103 ofproto_set_listeners(br->ofproto, &listeners);
1105 svec_destroy(&listeners);
1106 svec_destroy(&old_listeners);
1108 /* Configure OpenFlow controller connection snooping. */
1110 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1112 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1113 ovs_rundir, br->name));
1114 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1115 svec_clear(&snoops);
1117 svec_sort_unique(&snoops);
1119 svec_init(&old_snoops);
1120 ofproto_get_snoops(br->ofproto, &old_snoops);
1121 svec_sort_unique(&old_snoops);
1123 if (!svec_equal(&snoops, &old_snoops)) {
1124 ofproto_set_snoops(br->ofproto, &snoops);
1126 svec_destroy(&snoops);
1127 svec_destroy(&old_snoops);
1129 mirror_reconfigure(br);
1133 bridge_reconfigure_controller(struct bridge *br)
1135 char *pfx = xasprintf("bridge.%s.controller", br->name);
1136 const char *controller;
1138 controller = bridge_get_controller(br);
1139 if ((br->controller != NULL) != (controller != NULL)) {
1140 ofproto_flush_flows(br->ofproto);
1142 free(br->controller);
1143 br->controller = controller ? xstrdup(controller) : NULL;
1146 const char *fail_mode;
1147 int max_backoff, probe;
1148 int rate_limit, burst_limit;
1150 if (!strcmp(controller, "discover")) {
1151 bool update_resolv_conf = true;
1153 if (cfg_has("%s.update-resolv.conf", pfx)) {
1154 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1157 ofproto_set_discovery(br->ofproto, true,
1158 cfg_get_string(0, "%s.accept-regex", pfx),
1159 update_resolv_conf);
1161 char local_name[IF_NAMESIZE];
1162 struct netdev *netdev;
1166 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1168 || cfg_get_bool(0, "%s.in-band", pfx));
1169 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1170 ofproto_set_in_band(br->ofproto, in_band);
1172 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1173 local_name, sizeof local_name);
1175 error = netdev_open(local_name, NETDEV_ETH_TYPE_NONE, &netdev);
1178 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1179 struct in_addr ip, mask, gateway;
1180 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1181 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1182 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1184 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1186 mask.s_addr = guess_netmask(ip.s_addr);
1188 if (!netdev_set_in4(netdev, ip, mask)) {
1189 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1191 br->name, IP_ARGS(&ip.s_addr),
1192 IP_ARGS(&mask.s_addr));
1195 if (gateway.s_addr) {
1196 if (!netdev_add_router(netdev, gateway)) {
1197 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1198 br->name, IP_ARGS(&gateway.s_addr));
1202 netdev_close(netdev);
1206 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1208 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1210 ofproto_set_failure(br->ofproto,
1212 || !strcmp(fail_mode, "standalone")
1213 || !strcmp(fail_mode, "open")));
1215 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1217 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1222 ofproto_set_probe_interval(br->ofproto, probe);
1224 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1226 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1231 ofproto_set_max_backoff(br->ofproto, max_backoff);
1233 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1235 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1237 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1239 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1241 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1243 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1245 if (cfg_has("%s.commands.acl", pfx)) {
1246 struct svec command_acls;
1249 svec_init(&command_acls);
1250 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1251 command_acl = svec_join(&command_acls, ",", "");
1253 ofproto_set_remote_execution(br->ofproto, command_acl,
1254 cfg_get_string(0, "%s.commands.dir",
1257 svec_destroy(&command_acls);
1260 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1263 union ofp_action action;
1266 /* Set up a flow that matches every packet and directs them to
1267 * OFPP_NORMAL (which goes to us). */
1268 memset(&action, 0, sizeof action);
1269 action.type = htons(OFPAT_OUTPUT);
1270 action.output.len = htons(sizeof action);
1271 action.output.port = htons(OFPP_NORMAL);
1272 memset(&flow, 0, sizeof flow);
1273 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1276 ofproto_set_in_band(br->ofproto, false);
1277 ofproto_set_max_backoff(br->ofproto, 1);
1278 ofproto_set_probe_interval(br->ofproto, 5);
1279 ofproto_set_failure(br->ofproto, false);
1280 ofproto_set_stp(br->ofproto, false);
1284 ofproto_set_controller(br->ofproto, br->controller);
1288 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1293 for (i = 0; i < br->n_ports; i++) {
1294 struct port *port = br->ports[i];
1295 for (j = 0; j < port->n_ifaces; j++) {
1296 struct iface *iface = port->ifaces[j];
1297 svec_add(ifaces, iface->name);
1301 assert(svec_is_unique(ifaces));
1304 /* For robustness, in case the administrator moves around datapath ports behind
1305 * our back, we re-check all the datapath port numbers here.
1307 * This function will set the 'dp_ifidx' members of interfaces that have
1308 * disappeared to -1, so only call this function from a context where those
1309 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1310 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1311 * datapath, which doesn't support UINT16_MAX+1 ports. */
1313 bridge_fetch_dp_ifaces(struct bridge *br)
1315 struct odp_port *dpif_ports;
1316 size_t n_dpif_ports;
1319 /* Reset all interface numbers. */
1320 for (i = 0; i < br->n_ports; i++) {
1321 struct port *port = br->ports[i];
1322 for (j = 0; j < port->n_ifaces; j++) {
1323 struct iface *iface = port->ifaces[j];
1324 iface->dp_ifidx = -1;
1327 port_array_clear(&br->ifaces);
1329 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1330 for (i = 0; i < n_dpif_ports; i++) {
1331 struct odp_port *p = &dpif_ports[i];
1332 struct iface *iface = iface_lookup(br, p->devname);
1334 if (iface->dp_ifidx >= 0) {
1335 VLOG_WARN("%s reported interface %s twice",
1336 dpif_name(br->dpif), p->devname);
1337 } else if (iface_from_dp_ifidx(br, p->port)) {
1338 VLOG_WARN("%s reported interface %"PRIu16" twice",
1339 dpif_name(br->dpif), p->port);
1341 port_array_set(&br->ifaces, p->port, iface);
1342 iface->dp_ifidx = p->port;
1349 /* Bridge packet processing functions. */
1352 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1354 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1357 static struct bond_entry *
1358 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1360 return &port->bond_hash[bond_hash(mac)];
1364 bond_choose_iface(const struct port *port)
1367 for (i = 0; i < port->n_ifaces; i++) {
1368 if (port->ifaces[i]->enabled) {
1376 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1377 uint16_t *dp_ifidx, tag_type *tags)
1379 struct iface *iface;
1381 assert(port->n_ifaces);
1382 if (port->n_ifaces == 1) {
1383 iface = port->ifaces[0];
1385 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1386 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1387 || !port->ifaces[e->iface_idx]->enabled) {
1388 /* XXX select interface properly. The current interface selection
1389 * is only good for testing the rebalancing code. */
1390 e->iface_idx = bond_choose_iface(port);
1391 if (e->iface_idx < 0) {
1392 *tags |= port->no_ifaces_tag;
1395 e->iface_tag = tag_create_random();
1397 *tags |= e->iface_tag;
1398 iface = port->ifaces[e->iface_idx];
1400 *dp_ifidx = iface->dp_ifidx;
1401 *tags |= iface->tag; /* Currently only used for bonding. */
1406 bond_link_status_update(struct iface *iface, bool carrier)
1408 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1409 struct port *port = iface->port;
1411 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1412 /* Nothing to do. */
1415 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1416 iface->name, carrier ? "detected" : "dropped");
1417 if (carrier == iface->enabled) {
1418 iface->delay_expires = LLONG_MAX;
1419 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1420 iface->name, carrier ? "disabled" : "enabled");
1421 } else if (carrier && port->updelay && port->active_iface < 0) {
1422 iface->delay_expires = time_msec();
1423 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1424 "other interface is up", iface->name, port->updelay);
1426 int delay = carrier ? port->updelay : port->downdelay;
1427 iface->delay_expires = time_msec() + delay;
1430 "interface %s: will be %s if it stays %s for %d ms",
1432 carrier ? "enabled" : "disabled",
1433 carrier ? "up" : "down",
1440 bond_choose_active_iface(struct port *port)
1442 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1444 port->active_iface = bond_choose_iface(port);
1445 port->active_iface_tag = tag_create_random();
1446 if (port->active_iface >= 0) {
1447 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1448 port->name, port->ifaces[port->active_iface]->name);
1450 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1456 bond_enable_slave(struct iface *iface, bool enable)
1458 struct port *port = iface->port;
1459 struct bridge *br = port->bridge;
1461 iface->delay_expires = LLONG_MAX;
1462 if (enable == iface->enabled) {
1466 iface->enabled = enable;
1467 if (!iface->enabled) {
1468 VLOG_WARN("interface %s: disabled", iface->name);
1469 ofproto_revalidate(br->ofproto, iface->tag);
1470 if (iface->port_ifidx == port->active_iface) {
1471 ofproto_revalidate(br->ofproto,
1472 port->active_iface_tag);
1473 bond_choose_active_iface(port);
1475 bond_send_learning_packets(port);
1477 VLOG_WARN("interface %s: enabled", iface->name);
1478 if (port->active_iface < 0) {
1479 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1480 bond_choose_active_iface(port);
1481 bond_send_learning_packets(port);
1483 iface->tag = tag_create_random();
1488 bond_run(struct bridge *br)
1492 for (i = 0; i < br->n_ports; i++) {
1493 struct port *port = br->ports[i];
1494 if (port->n_ifaces < 2) {
1497 for (j = 0; j < port->n_ifaces; j++) {
1498 struct iface *iface = port->ifaces[j];
1499 if (time_msec() >= iface->delay_expires) {
1500 bond_enable_slave(iface, !iface->enabled);
1507 bond_wait(struct bridge *br)
1511 for (i = 0; i < br->n_ports; i++) {
1512 struct port *port = br->ports[i];
1513 if (port->n_ifaces < 2) {
1516 for (j = 0; j < port->n_ifaces; j++) {
1517 struct iface *iface = port->ifaces[j];
1518 if (iface->delay_expires != LLONG_MAX) {
1519 poll_timer_wait(iface->delay_expires - time_msec());
1526 set_dst(struct dst *p, const flow_t *flow,
1527 const struct port *in_port, const struct port *out_port,
1532 * XXX This uses too many tags: any broadcast flow will get one tag per
1533 * destination port, and thus a broadcast on a switch of any size is likely
1534 * to have all tag bits set. We should figure out a way to be smarter.
1536 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1537 *tags |= out_port->stp_state_tag;
1538 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1542 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1543 : in_port->vlan >= 0 ? in_port->vlan
1544 : ntohs(flow->dl_vlan));
1545 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1549 swap_dst(struct dst *p, struct dst *q)
1551 struct dst tmp = *p;
1556 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1557 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1558 * that we push to the datapath. We could in fact fully sort the array by
1559 * vlan, but in most cases there are at most two different vlan tags so that's
1560 * possibly overkill.) */
1562 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1564 struct dst *first = dsts;
1565 struct dst *last = dsts + n_dsts;
1567 while (first != last) {
1569 * - All dsts < first have vlan == 'vlan'.
1570 * - All dsts >= last have vlan != 'vlan'.
1571 * - first < last. */
1572 while (first->vlan == vlan) {
1573 if (++first == last) {
1578 /* Same invariants, plus one additional:
1579 * - first->vlan != vlan.
1581 while (last[-1].vlan != vlan) {
1582 if (--last == first) {
1587 /* Same invariants, plus one additional:
1588 * - last[-1].vlan == vlan.*/
1589 swap_dst(first++, --last);
1594 mirror_mask_ffs(mirror_mask_t mask)
1596 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1601 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1602 const struct dst *test)
1605 for (i = 0; i < n_dsts; i++) {
1606 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1614 port_trunks_vlan(const struct port *port, uint16_t vlan)
1616 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1620 port_includes_vlan(const struct port *port, uint16_t vlan)
1622 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1626 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1627 const struct port *in_port, const struct port *out_port,
1628 struct dst dsts[], tag_type *tags)
1630 mirror_mask_t mirrors = in_port->src_mirrors;
1631 struct dst *dst = dsts;
1634 *tags |= in_port->stp_state_tag;
1635 if (out_port == FLOOD_PORT) {
1636 /* XXX use ODP_FLOOD if no vlans or bonding. */
1637 /* XXX even better, define each VLAN as a datapath port group */
1638 for (i = 0; i < br->n_ports; i++) {
1639 struct port *port = br->ports[i];
1640 if (port != in_port && port_includes_vlan(port, vlan)
1641 && !port->is_mirror_output_port
1642 && set_dst(dst, flow, in_port, port, tags)) {
1643 mirrors |= port->dst_mirrors;
1647 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1648 mirrors |= out_port->dst_mirrors;
1653 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1654 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1656 if (set_dst(dst, flow, in_port, m->out_port, tags)
1657 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1661 for (i = 0; i < br->n_ports; i++) {
1662 struct port *port = br->ports[i];
1663 if (port_includes_vlan(port, m->out_vlan)
1664 && set_dst(dst, flow, in_port, port, tags)
1665 && !dst_is_duplicate(dsts, dst - dsts, dst))
1667 if (port->vlan < 0) {
1668 dst->vlan = m->out_vlan;
1670 if (dst->dp_ifidx == flow->in_port
1671 && dst->vlan == vlan) {
1672 /* Don't send out input port on same VLAN. */
1680 mirrors &= mirrors - 1;
1683 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1688 print_dsts(const struct dst *dsts, size_t n)
1690 for (; n--; dsts++) {
1691 printf(">p%"PRIu16, dsts->dp_ifidx);
1692 if (dsts->vlan != OFP_VLAN_NONE) {
1693 printf("v%"PRIu16, dsts->vlan);
1699 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1700 const struct port *in_port, const struct port *out_port,
1701 tag_type *tags, struct odp_actions *actions)
1703 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1705 const struct dst *p;
1708 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1710 cur_vlan = ntohs(flow->dl_vlan);
1711 for (p = dsts; p < &dsts[n_dsts]; p++) {
1712 union odp_action *a;
1713 if (p->vlan != cur_vlan) {
1714 if (p->vlan == OFP_VLAN_NONE) {
1715 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1717 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1718 a->vlan_vid.vlan_vid = htons(p->vlan);
1722 a = odp_actions_add(actions, ODPAT_OUTPUT);
1723 a->output.port = p->dp_ifidx;
1728 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1730 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1731 return (flow->dl_type == htons(ETH_TYPE_ARP)
1732 && eth_addr_is_broadcast(flow->dl_dst)
1733 && packet->size >= sizeof(struct arp_eth_header)
1734 && arp->ar_op == ARP_OP_REQUEST);
1737 /* If the composed actions may be applied to any packet in the given 'flow',
1738 * returns true. Otherwise, the actions should only be applied to 'packet', or
1739 * not at all, if 'packet' was NULL. */
1741 process_flow(struct bridge *br, const flow_t *flow,
1742 const struct ofpbuf *packet, struct odp_actions *actions,
1745 struct iface *in_iface;
1746 struct port *in_port;
1747 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1750 /* Find the interface and port structure for the received packet. */
1751 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1753 /* No interface? Something fishy... */
1754 if (packet != NULL) {
1755 /* Odd. A few possible reasons here:
1757 * - We deleted an interface but there are still a few packets
1758 * queued up from it.
1760 * - Someone externally added an interface (e.g. with "ovs-dpctl
1761 * add-if") that we don't know about.
1763 * - Packet arrived on the local port but the local port is not
1764 * one of our bridge ports.
1766 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1768 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1769 "interface %"PRIu16, br->name, flow->in_port);
1772 /* Return without adding any actions, to drop packets on this flow. */
1775 in_port = in_iface->port;
1777 /* Figure out what VLAN this packet belongs to.
1779 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1780 * belongs to VLAN 0, so we should treat both cases identically. (In the
1781 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1782 * presumably to allow a priority to be specified. In the latter case, the
1783 * packet does not have any 802.1Q header.) */
1784 vlan = ntohs(flow->dl_vlan);
1785 if (vlan == OFP_VLAN_NONE) {
1788 if (in_port->vlan >= 0) {
1790 /* XXX support double tagging? */
1791 if (packet != NULL) {
1792 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1793 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1794 "packet received on port %s configured with "
1795 "implicit VLAN %"PRIu16,
1796 br->name, ntohs(flow->dl_vlan),
1797 in_port->name, in_port->vlan);
1801 vlan = in_port->vlan;
1803 if (!port_includes_vlan(in_port, vlan)) {
1804 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1805 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1806 "packet received on port %s not configured for "
1808 br->name, vlan, in_port->name, vlan);
1813 /* Drop frames for ports that STP wants entirely killed (both for
1814 * forwarding and for learning). Later, after we do learning, we'll drop
1815 * the frames that STP wants to do learning but not forwarding on. */
1816 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1820 /* Drop frames for reserved multicast addresses. */
1821 if (eth_addr_is_reserved(flow->dl_dst)) {
1825 /* Drop frames on ports reserved for mirroring. */
1826 if (in_port->is_mirror_output_port) {
1827 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1828 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1829 "which is reserved exclusively for mirroring",
1830 br->name, in_port->name);
1834 /* Multicast (and broadcast) packets on bonds need special attention, to
1835 * avoid receiving duplicates. */
1836 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1837 *tags |= in_port->active_iface_tag;
1838 if (in_port->active_iface != in_iface->port_ifidx) {
1839 /* Drop all multicast packets on inactive slaves. */
1842 /* Drop all multicast packets for which we have learned a different
1843 * input port, because we probably sent the packet on one slaves
1844 * and got it back on the active slave. Broadcast ARP replies are
1845 * an exception to this rule: the host has moved to another
1847 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1848 if (src_idx != -1 && src_idx != in_port->port_idx) {
1850 if (!is_bcast_arp_reply(flow, packet)) {
1854 /* No way to know whether it's an ARP reply, because the
1855 * flow entry doesn't include enough information and we
1856 * don't have a packet. Punt. */
1864 out_port = FLOOD_PORT;
1868 /* Learn source MAC (but don't try to learn from revalidation). */
1870 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1871 vlan, in_port->port_idx);
1873 /* The log messages here could actually be useful in debugging,
1874 * so keep the rate limit relatively high. */
1875 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1877 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1878 "on port %s in VLAN %d",
1879 br->name, ETH_ADDR_ARGS(flow->dl_src),
1880 in_port->name, vlan);
1881 ofproto_revalidate(br->ofproto, rev_tag);
1885 /* Determine output port. */
1886 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1888 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1889 out_port = br->ports[out_port_idx];
1893 /* Don't send packets out their input ports. Don't forward frames that STP
1894 * wants us to discard. */
1895 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1900 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1903 * We send out only a single packet, instead of setting up a flow, if the
1904 * packet is an ARP directed to broadcast that arrived on a bonded
1905 * interface. In such a situation ARP requests and replies must be handled
1906 * differently, but OpenFlow unfortunately can't distinguish them.
1908 return (in_port->n_ifaces < 2
1909 || flow->dl_type != htons(ETH_TYPE_ARP)
1910 || !eth_addr_is_broadcast(flow->dl_dst));
1913 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1916 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1917 const struct ofp_phy_port *opp,
1920 struct bridge *br = br_;
1921 struct iface *iface;
1924 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1930 if (reason == OFPPR_DELETE) {
1931 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1932 br->name, iface->name);
1933 iface_destroy(iface);
1934 if (!port->n_ifaces) {
1935 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1936 br->name, port->name);
1942 if (port->n_ifaces > 1) {
1943 bool up = !(opp->state & OFPPS_LINK_DOWN);
1944 bond_link_status_update(iface, up);
1945 port_update_bond_compat(port);
1951 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1952 struct odp_actions *actions, tag_type *tags, void *br_)
1954 struct bridge *br = br_;
1957 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1958 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1959 brstp_receive(br, flow, payload);
1964 COVERAGE_INC(bridge_process_flow);
1965 return process_flow(br, flow, packet, actions, tags);
1969 bridge_account_flow_ofhook_cb(const flow_t *flow,
1970 const union odp_action *actions,
1971 size_t n_actions, unsigned long long int n_bytes,
1974 struct bridge *br = br_;
1975 const union odp_action *a;
1977 if (!br->has_bonded_ports) {
1981 for (a = actions; a < &actions[n_actions]; a++) {
1982 if (a->type == ODPAT_OUTPUT) {
1983 struct port *port = port_from_dp_ifidx(br, a->output.port);
1984 if (port && port->n_ifaces >= 2) {
1985 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1986 e->tx_bytes += n_bytes;
1993 bridge_account_checkpoint_ofhook_cb(void *br_)
1995 struct bridge *br = br_;
1998 if (!br->has_bonded_ports) {
2002 /* The current ofproto implementation calls this callback at least once a
2003 * second, so this timer implementation is sufficient. */
2004 if (time_msec() < br->bond_next_rebalance) {
2007 br->bond_next_rebalance = time_msec() + 10000;
2009 for (i = 0; i < br->n_ports; i++) {
2010 struct port *port = br->ports[i];
2011 if (port->n_ifaces > 1) {
2012 bond_rebalance_port(port);
2017 static struct ofhooks bridge_ofhooks = {
2018 bridge_port_changed_ofhook_cb,
2019 bridge_normal_ofhook_cb,
2020 bridge_account_flow_ofhook_cb,
2021 bridge_account_checkpoint_ofhook_cb,
2024 /* Bonding functions. */
2026 /* Statistics for a single interface on a bonded port, used for load-based
2027 * bond rebalancing. */
2028 struct slave_balance {
2029 struct iface *iface; /* The interface. */
2030 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2032 /* All the "bond_entry"s that are assigned to this interface, in order of
2033 * increasing tx_bytes. */
2034 struct bond_entry **hashes;
2038 /* Sorts pointers to pointers to bond_entries in ascending order by the
2039 * interface to which they are assigned, and within a single interface in
2040 * ascending order of bytes transmitted. */
2042 compare_bond_entries(const void *a_, const void *b_)
2044 const struct bond_entry *const *ap = a_;
2045 const struct bond_entry *const *bp = b_;
2046 const struct bond_entry *a = *ap;
2047 const struct bond_entry *b = *bp;
2048 if (a->iface_idx != b->iface_idx) {
2049 return a->iface_idx > b->iface_idx ? 1 : -1;
2050 } else if (a->tx_bytes != b->tx_bytes) {
2051 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2057 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2058 * *descending* order by number of bytes transmitted. */
2060 compare_slave_balance(const void *a_, const void *b_)
2062 const struct slave_balance *a = a_;
2063 const struct slave_balance *b = b_;
2064 if (a->iface->enabled != b->iface->enabled) {
2065 return a->iface->enabled ? -1 : 1;
2066 } else if (a->tx_bytes != b->tx_bytes) {
2067 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2074 swap_bals(struct slave_balance *a, struct slave_balance *b)
2076 struct slave_balance tmp = *a;
2081 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2082 * given that 'p' (and only 'p') might be in the wrong location.
2084 * This function invalidates 'p', since it might now be in a different memory
2087 resort_bals(struct slave_balance *p,
2088 struct slave_balance bals[], size_t n_bals)
2091 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2092 swap_bals(p, p - 1);
2094 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2095 swap_bals(p, p + 1);
2101 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2103 if (VLOG_IS_DBG_ENABLED()) {
2104 struct ds ds = DS_EMPTY_INITIALIZER;
2105 const struct slave_balance *b;
2107 for (b = bals; b < bals + n_bals; b++) {
2111 ds_put_char(&ds, ',');
2113 ds_put_format(&ds, " %s %"PRIu64"kB",
2114 b->iface->name, b->tx_bytes / 1024);
2116 if (!b->iface->enabled) {
2117 ds_put_cstr(&ds, " (disabled)");
2119 if (b->n_hashes > 0) {
2120 ds_put_cstr(&ds, " (");
2121 for (i = 0; i < b->n_hashes; i++) {
2122 const struct bond_entry *e = b->hashes[i];
2124 ds_put_cstr(&ds, " + ");
2126 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2127 e - port->bond_hash, e->tx_bytes / 1024);
2129 ds_put_cstr(&ds, ")");
2132 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2137 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2139 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2140 struct bond_entry *hash)
2142 struct port *port = from->iface->port;
2143 uint64_t delta = hash->tx_bytes;
2145 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2146 "from %s to %s (now carrying %"PRIu64"kB and "
2147 "%"PRIu64"kB load, respectively)",
2148 port->name, delta / 1024, hash - port->bond_hash,
2149 from->iface->name, to->iface->name,
2150 (from->tx_bytes - delta) / 1024,
2151 (to->tx_bytes + delta) / 1024);
2153 /* Delete element from from->hashes.
2155 * We don't bother to add the element to to->hashes because not only would
2156 * it require more work, the only purpose it would be to allow that hash to
2157 * be migrated to another slave in this rebalancing run, and there is no
2158 * point in doing that. */
2159 if (from->hashes[0] == hash) {
2162 int i = hash - from->hashes[0];
2163 memmove(from->hashes + i, from->hashes + i + 1,
2164 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2168 /* Shift load away from 'from' to 'to'. */
2169 from->tx_bytes -= delta;
2170 to->tx_bytes += delta;
2172 /* Arrange for flows to be revalidated. */
2173 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2174 hash->iface_idx = to->iface->port_ifidx;
2175 hash->iface_tag = tag_create_random();
2179 bond_rebalance_port(struct port *port)
2181 struct slave_balance bals[DP_MAX_PORTS];
2183 struct bond_entry *hashes[BOND_MASK + 1];
2184 struct slave_balance *b, *from, *to;
2185 struct bond_entry *e;
2188 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2189 * descending order of tx_bytes, so that bals[0] represents the most
2190 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2193 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2194 * array for each slave_balance structure, we sort our local array of
2195 * hashes in order by slave, so that all of the hashes for a given slave
2196 * become contiguous in memory, and then we point each 'hashes' members of
2197 * a slave_balance structure to the start of a contiguous group. */
2198 n_bals = port->n_ifaces;
2199 for (b = bals; b < &bals[n_bals]; b++) {
2200 b->iface = port->ifaces[b - bals];
2205 for (i = 0; i <= BOND_MASK; i++) {
2206 hashes[i] = &port->bond_hash[i];
2208 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2209 for (i = 0; i <= BOND_MASK; i++) {
2211 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2212 b = &bals[e->iface_idx];
2213 b->tx_bytes += e->tx_bytes;
2215 b->hashes = &hashes[i];
2220 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2221 log_bals(bals, n_bals, port);
2223 /* Discard slaves that aren't enabled (which were sorted to the back of the
2224 * array earlier). */
2225 while (!bals[n_bals - 1].iface->enabled) {
2232 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2233 to = &bals[n_bals - 1];
2234 for (from = bals; from < to; ) {
2235 uint64_t overload = from->tx_bytes - to->tx_bytes;
2236 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2237 /* The extra load on 'from' (and all less-loaded slaves), compared
2238 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2239 * it is less than ~1Mbps. No point in rebalancing. */
2241 } else if (from->n_hashes == 1) {
2242 /* 'from' only carries a single MAC hash, so we can't shift any
2243 * load away from it, even though we want to. */
2246 /* 'from' is carrying significantly more load than 'to', and that
2247 * load is split across at least two different hashes. Pick a hash
2248 * to migrate to 'to' (the least-loaded slave), given that doing so
2249 * must not cause 'to''s load to exceed 'from''s load.
2251 * The sort order we use means that we prefer to shift away the
2252 * smallest hashes instead of the biggest ones. There is little
2253 * reason behind this decision; we could use the opposite sort
2254 * order to shift away big hashes ahead of small ones. */
2257 for (i = 0; i < from->n_hashes; i++) {
2258 uint64_t delta = from->hashes[i]->tx_bytes;
2259 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2263 if (i < from->n_hashes) {
2264 bond_shift_load(from, to, from->hashes[i]);
2266 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2267 * point to different slave_balance structures. It is only
2268 * valid to do these two operations in a row at all because we
2269 * know that 'from' will not move past 'to' and vice versa. */
2270 resort_bals(from, bals, n_bals);
2271 resort_bals(to, bals, n_bals);
2278 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2279 * historical data to decay to <1% in 7 rebalancing runs. */
2280 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2286 bond_send_learning_packets(struct port *port)
2288 struct bridge *br = port->bridge;
2289 struct mac_entry *e;
2290 struct ofpbuf packet;
2291 int error, n_packets, n_errors;
2293 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2297 ofpbuf_init(&packet, 128);
2298 error = n_packets = n_errors = 0;
2299 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2300 static const char s[] = "Open vSwitch Bond Failover";
2301 union ofp_action actions[2], *a;
2302 struct eth_header *eth;
2303 struct llc_snap_header *llc_snap;
2309 if (e->port == port->port_idx
2310 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2314 /* Compose packet to send. */
2315 ofpbuf_clear(&packet);
2316 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2317 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2318 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2319 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2321 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2322 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2323 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2325 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2326 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2327 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2328 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2329 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2331 /* Compose actions. */
2332 memset(actions, 0, sizeof actions);
2335 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2336 a->vlan_vid.len = htons(sizeof *a);
2337 a->vlan_vid.vlan_vid = htons(e->vlan);
2340 a->output.type = htons(OFPAT_OUTPUT);
2341 a->output.len = htons(sizeof *a);
2342 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2347 flow_extract(&packet, ODPP_NONE, &flow);
2348 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2355 ofpbuf_uninit(&packet);
2358 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2359 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2360 "packets, last error was: %s",
2361 port->name, n_errors, n_packets, strerror(error));
2363 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2364 port->name, n_packets);
2368 /* Bonding unixctl user interface functions. */
2371 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2373 struct ds ds = DS_EMPTY_INITIALIZER;
2374 const struct bridge *br;
2376 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2378 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2381 for (i = 0; i < br->n_ports; i++) {
2382 const struct port *port = br->ports[i];
2383 if (port->n_ifaces > 1) {
2386 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2387 for (j = 0; j < port->n_ifaces; j++) {
2388 const struct iface *iface = port->ifaces[j];
2390 ds_put_cstr(&ds, ", ");
2392 ds_put_cstr(&ds, iface->name);
2394 ds_put_char(&ds, '\n');
2398 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2402 static struct port *
2403 bond_find(const char *name)
2405 const struct bridge *br;
2407 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2410 for (i = 0; i < br->n_ports; i++) {
2411 struct port *port = br->ports[i];
2412 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2421 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2423 struct ds ds = DS_EMPTY_INITIALIZER;
2424 const struct port *port;
2427 port = bond_find(args);
2429 unixctl_command_reply(conn, 501, "no such bond");
2433 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2434 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2435 ds_put_format(&ds, "next rebalance: %lld ms\n",
2436 port->bridge->bond_next_rebalance - time_msec());
2437 for (j = 0; j < port->n_ifaces; j++) {
2438 const struct iface *iface = port->ifaces[j];
2439 struct bond_entry *be;
2442 ds_put_format(&ds, "slave %s: %s\n",
2443 iface->name, iface->enabled ? "enabled" : "disabled");
2444 if (j == port->active_iface) {
2445 ds_put_cstr(&ds, "\tactive slave\n");
2447 if (iface->delay_expires != LLONG_MAX) {
2448 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2449 iface->enabled ? "downdelay" : "updelay",
2450 iface->delay_expires - time_msec());
2454 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2455 int hash = be - port->bond_hash;
2456 struct mac_entry *me;
2458 if (be->iface_idx != j) {
2462 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2463 hash, be->tx_bytes / 1024);
2466 if (!port->bridge->ml) {
2470 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2471 &port->bridge->ml->lrus) {
2474 if (bond_hash(me->mac) == hash
2475 && me->port != port->port_idx
2476 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2477 && dp_ifidx == iface->dp_ifidx)
2479 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2480 ETH_ADDR_ARGS(me->mac));
2485 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2490 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2492 char *args = (char *) args_;
2493 char *save_ptr = NULL;
2494 char *bond_s, *hash_s, *slave_s;
2495 uint8_t mac[ETH_ADDR_LEN];
2497 struct iface *iface;
2498 struct bond_entry *entry;
2501 bond_s = strtok_r(args, " ", &save_ptr);
2502 hash_s = strtok_r(NULL, " ", &save_ptr);
2503 slave_s = strtok_r(NULL, " ", &save_ptr);
2505 unixctl_command_reply(conn, 501,
2506 "usage: bond/migrate BOND HASH SLAVE");
2510 port = bond_find(bond_s);
2512 unixctl_command_reply(conn, 501, "no such bond");
2516 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2517 == ETH_ADDR_SCAN_COUNT) {
2518 hash = bond_hash(mac);
2519 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2520 hash = atoi(hash_s) & BOND_MASK;
2522 unixctl_command_reply(conn, 501, "bad hash");
2526 iface = port_lookup_iface(port, slave_s);
2528 unixctl_command_reply(conn, 501, "no such slave");
2532 if (!iface->enabled) {
2533 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2537 entry = &port->bond_hash[hash];
2538 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2539 entry->iface_idx = iface->port_ifidx;
2540 entry->iface_tag = tag_create_random();
2541 unixctl_command_reply(conn, 200, "migrated");
2545 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2547 char *args = (char *) args_;
2548 char *save_ptr = NULL;
2549 char *bond_s, *slave_s;
2551 struct iface *iface;
2553 bond_s = strtok_r(args, " ", &save_ptr);
2554 slave_s = strtok_r(NULL, " ", &save_ptr);
2556 unixctl_command_reply(conn, 501,
2557 "usage: bond/set-active-slave BOND SLAVE");
2561 port = bond_find(bond_s);
2563 unixctl_command_reply(conn, 501, "no such bond");
2567 iface = port_lookup_iface(port, slave_s);
2569 unixctl_command_reply(conn, 501, "no such slave");
2573 if (!iface->enabled) {
2574 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2578 if (port->active_iface != iface->port_ifidx) {
2579 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2580 port->active_iface = iface->port_ifidx;
2581 port->active_iface_tag = tag_create_random();
2582 VLOG_INFO("port %s: active interface is now %s",
2583 port->name, iface->name);
2584 bond_send_learning_packets(port);
2585 unixctl_command_reply(conn, 200, "done");
2587 unixctl_command_reply(conn, 200, "no change");
2592 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2594 char *args = (char *) args_;
2595 char *save_ptr = NULL;
2596 char *bond_s, *slave_s;
2598 struct iface *iface;
2600 bond_s = strtok_r(args, " ", &save_ptr);
2601 slave_s = strtok_r(NULL, " ", &save_ptr);
2603 unixctl_command_reply(conn, 501,
2604 "usage: bond/enable/disable-slave BOND SLAVE");
2608 port = bond_find(bond_s);
2610 unixctl_command_reply(conn, 501, "no such bond");
2614 iface = port_lookup_iface(port, slave_s);
2616 unixctl_command_reply(conn, 501, "no such slave");
2620 bond_enable_slave(iface, enable);
2621 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2625 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2627 enable_slave(conn, args, true);
2631 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2633 enable_slave(conn, args, false);
2639 unixctl_command_register("bond/list", bond_unixctl_list);
2640 unixctl_command_register("bond/show", bond_unixctl_show);
2641 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2642 unixctl_command_register("bond/set-active-slave",
2643 bond_unixctl_set_active_slave);
2644 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2645 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2648 /* Port functions. */
2651 port_create(struct bridge *br, const char *name)
2655 port = xcalloc(1, sizeof *port);
2657 port->port_idx = br->n_ports;
2659 port->trunks = NULL;
2660 port->name = xstrdup(name);
2661 port->active_iface = -1;
2662 port->stp_state = STP_DISABLED;
2663 port->stp_state_tag = 0;
2665 if (br->n_ports >= br->allocated_ports) {
2666 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2669 br->ports[br->n_ports++] = port;
2671 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2676 port_reconfigure(struct port *port)
2678 bool bonded = cfg_has_section("bonding.%s", port->name);
2679 struct svec old_ifaces, new_ifaces;
2680 unsigned long *trunks;
2684 /* Collect old and new interfaces. */
2685 svec_init(&old_ifaces);
2686 svec_init(&new_ifaces);
2687 for (i = 0; i < port->n_ifaces; i++) {
2688 svec_add(&old_ifaces, port->ifaces[i]->name);
2690 svec_sort(&old_ifaces);
2692 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2693 if (!new_ifaces.n) {
2694 VLOG_ERR("port %s: no interfaces specified for bonded port",
2696 } else if (new_ifaces.n == 1) {
2697 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2701 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2702 if (port->updelay < 0) {
2705 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2706 if (port->downdelay < 0) {
2707 port->downdelay = 0;
2710 svec_init(&new_ifaces);
2711 svec_add(&new_ifaces, port->name);
2714 /* Get rid of deleted interfaces and add new interfaces. */
2715 for (i = 0; i < port->n_ifaces; i++) {
2716 struct iface *iface = port->ifaces[i];
2717 if (!svec_contains(&new_ifaces, iface->name)) {
2718 iface_destroy(iface);
2723 for (i = 0; i < new_ifaces.n; i++) {
2724 const char *name = new_ifaces.names[i];
2725 if (!svec_contains(&old_ifaces, name)) {
2726 iface_create(port, name);
2732 if (cfg_has("vlan.%s.tag", port->name)) {
2734 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2735 if (vlan >= 0 && vlan <= 4095) {
2736 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2739 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2740 * they even work as-is. But they have not been tested. */
2741 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2745 if (port->vlan != vlan) {
2747 bridge_flush(port->bridge);
2750 /* Get trunked VLANs. */
2753 size_t n_trunks, n_errors;
2756 trunks = bitmap_allocate(4096);
2757 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2759 for (i = 0; i < n_trunks; i++) {
2760 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2762 bitmap_set1(trunks, trunk);
2768 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2769 port->name, n_trunks);
2771 if (n_errors == n_trunks) {
2773 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2776 bitmap_set_multiple(trunks, 0, 4096, 1);
2779 if (cfg_has("vlan.%s.trunks", port->name)) {
2780 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2781 port->name, port->name);
2785 ? port->trunks != NULL
2786 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2787 bridge_flush(port->bridge);
2789 bitmap_free(port->trunks);
2790 port->trunks = trunks;
2792 svec_destroy(&old_ifaces);
2793 svec_destroy(&new_ifaces);
2797 port_destroy(struct port *port)
2800 struct bridge *br = port->bridge;
2804 proc_net_compat_update_vlan(port->name, NULL, 0);
2806 for (i = 0; i < MAX_MIRRORS; i++) {
2807 struct mirror *m = br->mirrors[i];
2808 if (m && m->out_port == port) {
2813 while (port->n_ifaces > 0) {
2814 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2817 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2818 del->port_idx = port->port_idx;
2821 bitmap_free(port->trunks);
2828 static struct port *
2829 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2831 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2832 return iface ? iface->port : NULL;
2835 static struct port *
2836 port_lookup(const struct bridge *br, const char *name)
2840 for (i = 0; i < br->n_ports; i++) {
2841 struct port *port = br->ports[i];
2842 if (!strcmp(port->name, name)) {
2849 static struct iface *
2850 port_lookup_iface(const struct port *port, const char *name)
2854 for (j = 0; j < port->n_ifaces; j++) {
2855 struct iface *iface = port->ifaces[j];
2856 if (!strcmp(iface->name, name)) {
2864 port_update_bonding(struct port *port)
2866 if (port->n_ifaces < 2) {
2867 /* Not a bonded port. */
2868 if (port->bond_hash) {
2869 free(port->bond_hash);
2870 port->bond_hash = NULL;
2871 proc_net_compat_update_bond(port->name, NULL);
2874 if (!port->bond_hash) {
2877 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2878 for (i = 0; i <= BOND_MASK; i++) {
2879 struct bond_entry *e = &port->bond_hash[i];
2883 port->no_ifaces_tag = tag_create_random();
2884 bond_choose_active_iface(port);
2886 port_update_bond_compat(port);
2891 port_update_bond_compat(struct port *port)
2893 struct compat_bond bond;
2896 if (port->n_ifaces < 2) {
2901 bond.updelay = port->updelay;
2902 bond.downdelay = port->downdelay;
2903 bond.n_slaves = port->n_ifaces;
2904 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2905 for (i = 0; i < port->n_ifaces; i++) {
2906 struct iface *iface = port->ifaces[i];
2907 struct compat_bond_slave *slave = &bond.slaves[i];
2908 slave->name = iface->name;
2909 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2910 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2914 netdev_get_etheraddr(iface->netdev, slave->mac);
2916 proc_net_compat_update_bond(port->name, &bond);
2921 port_update_vlan_compat(struct port *port)
2923 struct bridge *br = port->bridge;
2924 char *vlandev_name = NULL;
2926 if (port->vlan > 0) {
2927 /* Figure out the name that the VLAN device should actually have, if it
2928 * existed. This takes some work because the VLAN device would not
2929 * have port->name in its name; rather, it would have the trunk port's
2930 * name, and 'port' would be attached to a bridge that also had the
2931 * VLAN device one of its ports. So we need to find a trunk port that
2932 * includes port->vlan.
2934 * There might be more than one candidate. This doesn't happen on
2935 * XenServer, so if it happens we just pick the first choice in
2936 * alphabetical order instead of creating multiple VLAN devices. */
2938 for (i = 0; i < br->n_ports; i++) {
2939 struct port *p = br->ports[i];
2940 if (port_trunks_vlan(p, port->vlan)
2942 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2944 uint8_t ea[ETH_ADDR_LEN];
2945 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
2946 if (!eth_addr_is_multicast(ea) &&
2947 !eth_addr_is_reserved(ea) &&
2948 !eth_addr_is_zero(ea)) {
2949 vlandev_name = p->name;
2954 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2957 /* Interface functions. */
2960 iface_create(struct port *port, const char *name)
2962 struct iface *iface;
2964 iface = xcalloc(1, sizeof *iface);
2966 iface->port_ifidx = port->n_ifaces;
2967 iface->name = xstrdup(name);
2968 iface->dp_ifidx = -1;
2969 iface->tag = tag_create_random();
2970 iface->delay_expires = LLONG_MAX;
2971 iface->netdev = NULL;
2973 if (port->n_ifaces >= port->allocated_ifaces) {
2974 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2975 sizeof *port->ifaces);
2977 port->ifaces[port->n_ifaces++] = iface;
2978 if (port->n_ifaces > 1) {
2979 port->bridge->has_bonded_ports = true;
2982 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2984 bridge_flush(port->bridge);
2988 iface_destroy(struct iface *iface)
2991 struct port *port = iface->port;
2992 struct bridge *br = port->bridge;
2993 bool del_active = port->active_iface == iface->port_ifidx;
2996 if (iface->dp_ifidx >= 0) {
2997 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3000 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3001 del->port_ifidx = iface->port_ifidx;
3003 netdev_close(iface->netdev);
3008 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3009 bond_choose_active_iface(port);
3010 bond_send_learning_packets(port);
3013 bridge_flush(port->bridge);
3017 static struct iface *
3018 iface_lookup(const struct bridge *br, const char *name)
3022 for (i = 0; i < br->n_ports; i++) {
3023 struct port *port = br->ports[i];
3024 for (j = 0; j < port->n_ifaces; j++) {
3025 struct iface *iface = port->ifaces[j];
3026 if (!strcmp(iface->name, name)) {
3034 static struct iface *
3035 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3037 return port_array_get(&br->ifaces, dp_ifidx);
3040 /* Port mirroring. */
3043 mirror_reconfigure(struct bridge *br)
3045 struct svec old_mirrors, new_mirrors;
3048 /* Collect old and new mirrors. */
3049 svec_init(&old_mirrors);
3050 svec_init(&new_mirrors);
3051 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3052 for (i = 0; i < MAX_MIRRORS; i++) {
3053 if (br->mirrors[i]) {
3054 svec_add(&old_mirrors, br->mirrors[i]->name);
3058 /* Get rid of deleted mirrors and add new mirrors. */
3059 svec_sort(&old_mirrors);
3060 assert(svec_is_unique(&old_mirrors));
3061 svec_sort(&new_mirrors);
3062 assert(svec_is_unique(&new_mirrors));
3063 for (i = 0; i < MAX_MIRRORS; i++) {
3064 struct mirror *m = br->mirrors[i];
3065 if (m && !svec_contains(&new_mirrors, m->name)) {
3069 for (i = 0; i < new_mirrors.n; i++) {
3070 const char *name = new_mirrors.names[i];
3071 if (!svec_contains(&old_mirrors, name)) {
3072 mirror_create(br, name);
3075 svec_destroy(&old_mirrors);
3076 svec_destroy(&new_mirrors);
3078 /* Reconfigure all mirrors. */
3079 for (i = 0; i < MAX_MIRRORS; i++) {
3080 if (br->mirrors[i]) {
3081 mirror_reconfigure_one(br->mirrors[i]);
3085 /* Update port reserved status. */
3086 for (i = 0; i < br->n_ports; i++) {
3087 br->ports[i]->is_mirror_output_port = false;
3089 for (i = 0; i < MAX_MIRRORS; i++) {
3090 struct mirror *m = br->mirrors[i];
3091 if (m && m->out_port) {
3092 m->out_port->is_mirror_output_port = true;
3098 mirror_create(struct bridge *br, const char *name)
3103 for (i = 0; ; i++) {
3104 if (i >= MAX_MIRRORS) {
3105 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3106 "cannot create %s", br->name, MAX_MIRRORS, name);
3109 if (!br->mirrors[i]) {
3114 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3117 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3120 m->name = xstrdup(name);
3121 svec_init(&m->src_ports);
3122 svec_init(&m->dst_ports);
3130 mirror_destroy(struct mirror *m)
3133 struct bridge *br = m->bridge;
3136 for (i = 0; i < br->n_ports; i++) {
3137 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3138 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3141 svec_destroy(&m->src_ports);
3142 svec_destroy(&m->dst_ports);
3145 m->bridge->mirrors[m->idx] = NULL;
3153 prune_ports(struct mirror *m, struct svec *ports)
3158 svec_sort_unique(ports);
3161 for (i = 0; i < ports->n; i++) {
3162 const char *name = ports->names[i];
3163 if (port_lookup(m->bridge, name)) {
3164 svec_add(&tmp, name);
3166 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3167 m->bridge->name, m->name, name);
3170 svec_swap(ports, &tmp);
3175 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3179 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3180 * order won't give us numeric sort order. But that's good enough for what
3181 * we need right now. */
3182 svec_sort_unique(vlan_strings);
3184 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3186 for (i = 0; i < vlan_strings->n; i++) {
3187 const char *name = vlan_strings->names[i];
3189 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3190 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3191 m->bridge->name, m->name, name);
3193 (*vlans)[n_vlans++] = vlan;
3200 vlan_is_mirrored(const struct mirror *m, int vlan)
3204 for (i = 0; i < m->n_vlans; i++) {
3205 if (m->vlans[i] == vlan) {
3213 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3217 for (i = 0; i < m->n_vlans; i++) {
3218 if (port_trunks_vlan(p, m->vlans[i])) {
3226 mirror_reconfigure_one(struct mirror *m)
3228 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3229 struct svec src_ports, dst_ports, ports;
3230 struct svec vlan_strings;
3231 mirror_mask_t mirror_bit;
3232 const char *out_port_name;
3233 struct port *out_port;
3238 bool mirror_all_ports;
3240 /* Get output port. */
3241 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3242 m->bridge->name, m->name);
3243 if (out_port_name) {
3244 out_port = port_lookup(m->bridge, out_port_name);
3246 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3247 "named %s", pfx, m->bridge->name, out_port_name);
3254 if (cfg_has("%s.output.vlan", pfx)) {
3255 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3256 "ignoring %s.output.vlan", pfx, pfx, pfx);
3258 } else if (cfg_has("%s.output.vlan", pfx)) {
3260 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3262 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3263 "but exactly one is required; disabling port mirror %s",
3264 pfx, pfx, pfx, pfx);
3270 /* Get all the ports, and drop duplicates and ports that don't exist. */
3271 svec_init(&src_ports);
3272 svec_init(&dst_ports);
3274 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3275 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3276 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3277 svec_append(&src_ports, &ports);
3278 svec_append(&dst_ports, &ports);
3279 svec_destroy(&ports);
3280 prune_ports(m, &src_ports);
3281 prune_ports(m, &dst_ports);
3283 /* Get all the vlans, and drop duplicate and invalid vlans. */
3284 svec_init(&vlan_strings);
3285 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3286 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3287 svec_destroy(&vlan_strings);
3289 /* Update mirror data. */
3290 if (!svec_equal(&m->src_ports, &src_ports)
3291 || !svec_equal(&m->dst_ports, &dst_ports)
3292 || m->n_vlans != n_vlans
3293 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3294 || m->out_port != out_port
3295 || m->out_vlan != out_vlan) {
3296 bridge_flush(m->bridge);
3298 svec_swap(&m->src_ports, &src_ports);
3299 svec_swap(&m->dst_ports, &dst_ports);
3302 m->n_vlans = n_vlans;
3303 m->out_port = out_port;
3304 m->out_vlan = out_vlan;
3306 /* If no selection criteria have been given, mirror for all ports. */
3307 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3310 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3311 for (i = 0; i < m->bridge->n_ports; i++) {
3312 struct port *port = m->bridge->ports[i];
3314 if (mirror_all_ports
3315 || svec_contains(&m->src_ports, port->name)
3318 ? port_trunks_any_mirrored_vlan(m, port)
3319 : vlan_is_mirrored(m, port->vlan)))) {
3320 port->src_mirrors |= mirror_bit;
3322 port->src_mirrors &= ~mirror_bit;
3325 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3326 port->dst_mirrors |= mirror_bit;
3328 port->dst_mirrors &= ~mirror_bit;
3333 svec_destroy(&src_ports);
3334 svec_destroy(&dst_ports);
3338 /* Spanning tree protocol. */
3340 static void brstp_update_port_state(struct port *);
3343 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3345 struct bridge *br = br_;
3346 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3347 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3349 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3352 struct eth_header *eth = pkt->l2;
3354 netdev_get_etheraddr(iface->netdev, eth->eth_src);
3355 if (eth_addr_is_zero(eth->eth_src)) {
3356 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
3357 "with unknown MAC", br->name, port_no);
3359 union ofp_action action;
3362 memset(&action, 0, sizeof action);
3363 action.type = htons(OFPAT_OUTPUT);
3364 action.output.len = htons(sizeof action);
3365 action.output.port = htons(port_no);
3367 flow_extract(pkt, ODPP_NONE, &flow);
3368 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3375 brstp_reconfigure(struct bridge *br)
3379 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3381 stp_destroy(br->stp);
3387 uint64_t bridge_address, bridge_id;
3388 int bridge_priority;
3390 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3391 if (!bridge_address) {
3393 bridge_address = (stp_get_bridge_id(br->stp)
3394 & ((UINT64_C(1) << 48) - 1));
3396 uint8_t mac[ETH_ADDR_LEN];
3397 eth_addr_random(mac);
3398 bridge_address = eth_addr_to_uint64(mac);
3402 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3404 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3406 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3409 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3411 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3412 br->stp_last_tick = time_msec();
3415 if (bridge_id != stp_get_bridge_id(br->stp)) {
3416 stp_set_bridge_id(br->stp, bridge_id);
3421 for (i = 0; i < br->n_ports; i++) {
3422 struct port *p = br->ports[i];
3424 struct stp_port *sp;
3425 int path_cost, priority;
3431 dp_ifidx = p->ifaces[0]->dp_ifidx;
3432 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3436 sp = stp_get_port(br->stp, dp_ifidx);
3437 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3438 "stp.%s.port.%s.enabled",
3440 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3441 br->name, p->name));
3442 if (p->is_mirror_output_port) {
3445 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3446 bridge_flush(br); /* Might not be necessary. */
3448 stp_port_enable(sp);
3450 stp_port_disable(sp);
3454 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3456 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3458 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3459 "stp.%s.port.%s.priority",
3461 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3463 : STP_DEFAULT_PORT_PRIORITY);
3464 stp_port_set_priority(sp, priority);
3467 brstp_adjust_timers(br);
3469 for (i = 0; i < br->n_ports; i++) {
3470 brstp_update_port_state(br->ports[i]);
3475 brstp_update_port_state(struct port *p)
3477 struct bridge *br = p->bridge;
3478 enum stp_state state;
3480 /* Figure out new state. */
3481 state = STP_DISABLED;
3482 if (br->stp && p->n_ifaces > 0) {
3483 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3484 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3485 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3490 if (p->stp_state != state) {
3491 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3492 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3493 p->name, stp_state_name(p->stp_state),
3494 stp_state_name(state));
3495 if (p->stp_state == STP_DISABLED) {
3498 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3500 p->stp_state = state;
3501 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3502 : tag_create_random());
3507 brstp_adjust_timers(struct bridge *br)
3509 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3510 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3511 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3513 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3514 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3515 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3519 brstp_run(struct bridge *br)
3522 long long int now = time_msec();
3523 long long int elapsed = now - br->stp_last_tick;
3524 struct stp_port *sp;
3527 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3528 br->stp_last_tick = now;
3530 while (stp_get_changed_port(br->stp, &sp)) {
3531 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3533 brstp_update_port_state(p);
3540 brstp_wait(struct bridge *br)
3543 poll_timer_wait(1000);