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 uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */
86 bool enabled; /* May be chosen for flows? */
89 #define BOND_MASK 0xff
91 int iface_idx; /* Index of assigned iface, or -1 if none. */
92 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
93 tag_type iface_tag; /* Tag associated with iface_idx. */
96 #define MAX_MIRRORS 32
97 typedef uint32_t mirror_mask_t;
98 #define MIRROR_MASK_C(X) UINT32_C(X)
99 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
101 struct bridge *bridge;
105 /* Selection criteria. */
106 struct svec src_ports;
107 struct svec dst_ports;
112 struct port *out_port;
116 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
118 struct bridge *bridge;
120 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
121 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
124 /* An ordinary bridge port has 1 interface.
125 * A bridge port for bonding has at least 2 interfaces. */
126 struct iface **ifaces;
127 size_t n_ifaces, allocated_ifaces;
130 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
131 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
132 tag_type active_iface_tag; /* Tag for bcast flows. */
133 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
134 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
136 /* Port mirroring info. */
137 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
138 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
139 bool is_mirror_output_port; /* Does port mirroring send frames here? */
141 /* Spanning tree info. */
142 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
143 tag_type stp_state_tag; /* Tag for STP state change. */
146 #define DP_MAX_PORTS 255
148 struct list node; /* Node in global list of bridges. */
149 char *name; /* User-specified arbitrary name. */
150 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
151 bool sent_config_request; /* Successfully sent config request? */
152 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
154 /* Support for remote controllers. */
155 char *controller; /* NULL if there is no remote controller;
156 * "discover" to do controller discovery;
157 * otherwise a vconn name. */
159 /* OpenFlow switch processing. */
160 struct ofproto *ofproto; /* OpenFlow switch. */
162 /* Kernel datapath information. */
163 struct dpif *dpif; /* Datapath. */
164 struct port_array ifaces; /* Indexed by kernel datapath port number. */
168 size_t n_ports, allocated_ports;
171 bool has_bonded_ports;
172 long long int bond_next_rebalance;
177 /* Flow statistics gathering. */
178 time_t next_stats_request;
180 /* Port mirroring. */
181 struct mirror *mirrors[MAX_MIRRORS];
185 long long int stp_last_tick;
188 /* List of all bridges. */
189 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
191 /* Maximum number of datapaths. */
192 enum { DP_MAX = 256 };
194 static struct bridge *bridge_create(const char *name);
195 static void bridge_destroy(struct bridge *);
196 static struct bridge *bridge_lookup(const char *name);
197 static int bridge_run_one(struct bridge *);
198 static void bridge_reconfigure_one(struct bridge *);
199 static void bridge_reconfigure_controller(struct bridge *);
200 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
201 static void bridge_fetch_dp_ifaces(struct bridge *);
202 static void bridge_flush(struct bridge *);
203 static void bridge_pick_local_hw_addr(struct bridge *,
204 uint8_t ea[ETH_ADDR_LEN],
205 const char **devname);
206 static uint64_t bridge_pick_datapath_id(struct bridge *,
207 const uint8_t bridge_ea[ETH_ADDR_LEN],
208 const char *devname);
209 static uint64_t dpid_from_hash(const void *, size_t nbytes);
211 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
213 static void bond_init(void);
214 static void bond_run(struct bridge *);
215 static void bond_wait(struct bridge *);
216 static void bond_rebalance_port(struct port *);
217 static void bond_send_learning_packets(struct port *);
219 static void port_create(struct bridge *, const char *name);
220 static void port_reconfigure(struct port *);
221 static void port_destroy(struct port *);
222 static struct port *port_lookup(const struct bridge *, const char *name);
223 static struct iface *port_lookup_iface(const struct port *, const char *name);
224 static struct port *port_from_dp_ifidx(const struct bridge *,
226 static void port_update_bond_compat(struct port *);
227 static void port_update_vlan_compat(struct port *);
228 static void port_update_bonding(struct port *);
230 static void mirror_create(struct bridge *, const char *name);
231 static void mirror_destroy(struct mirror *);
232 static void mirror_reconfigure(struct bridge *);
233 static void mirror_reconfigure_one(struct mirror *);
234 static bool vlan_is_mirrored(const struct mirror *, int vlan);
236 static void brstp_reconfigure(struct bridge *);
237 static void brstp_adjust_timers(struct bridge *);
238 static void brstp_run(struct bridge *);
239 static void brstp_wait(struct bridge *);
241 static void iface_create(struct port *, const char *name);
242 static void iface_destroy(struct iface *);
243 static struct iface *iface_lookup(const struct bridge *, const char *name);
244 static struct iface *iface_from_dp_ifidx(const struct bridge *,
247 /* Hooks into ofproto processing. */
248 static struct ofhooks bridge_ofhooks;
250 /* Public functions. */
252 /* Adds the name of each interface used by a bridge, including local and
253 * internal ports, to 'svec'. */
255 bridge_get_ifaces(struct svec *svec)
257 struct bridge *br, *next;
260 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
261 for (i = 0; i < br->n_ports; i++) {
262 struct port *port = br->ports[i];
264 for (j = 0; j < port->n_ifaces; j++) {
265 struct iface *iface = port->ifaces[j];
266 if (iface->dp_ifidx < 0) {
267 VLOG_ERR("%s interface not in datapath %s, ignoring",
268 iface->name, dpif_name(br->dpif));
270 if (iface->dp_ifidx != ODPP_LOCAL) {
271 svec_add(svec, iface->name);
279 /* The caller must already have called cfg_read(). */
283 struct svec dpif_names;
286 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
288 dp_enumerate(&dpif_names);
289 for (i = 0; i < dpif_names.n; i++) {
290 const char *dpif_name = dpif_names.names[i];
294 retval = dpif_open(dpif_name, &dpif);
296 struct svec all_names;
299 svec_init(&all_names);
300 dpif_get_all_names(dpif, &all_names);
301 for (j = 0; j < all_names.n; j++) {
302 if (cfg_has("bridge.%s.port", all_names.names[j])) {
308 svec_destroy(&all_names);
314 bridge_reconfigure();
319 config_string_change(const char *key, char **valuep)
321 const char *value = cfg_get_string(0, "%s", key);
322 if (value && (!*valuep || strcmp(value, *valuep))) {
324 *valuep = xstrdup(value);
332 bridge_configure_ssl(void)
334 /* XXX SSL should be configurable on a per-bridge basis.
335 * XXX should be possible to de-configure SSL. */
336 static char *private_key_file;
337 static char *certificate_file;
338 static char *cacert_file;
341 if (config_string_change("ssl.private-key", &private_key_file)) {
342 vconn_ssl_set_private_key_file(private_key_file);
345 if (config_string_change("ssl.certificate", &certificate_file)) {
346 vconn_ssl_set_certificate_file(certificate_file);
349 /* We assume that even if the filename hasn't changed, if the CA cert
350 * file has been removed, that we want to move back into
351 * boot-strapping mode. This opens a small security hole, because
352 * the old certificate will still be trusted until vSwitch is
353 * restarted. We may want to address this in vconn's SSL library. */
354 if (config_string_change("ssl.ca-cert", &cacert_file)
355 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
356 vconn_ssl_set_ca_cert_file(cacert_file,
357 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
362 /* iterate_and_prune_ifaces() callback function that opens the network device
363 * for 'iface', if it is not already open, and retrieves the interface's MAC
364 * address and carrier status. */
366 init_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
371 } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE,
373 netdev_get_etheraddr(iface->netdev, iface->mac);
374 netdev_get_carrier(iface->netdev, &iface->enabled);
377 /* If the network device can't be opened, then we're not going to try
378 * to do anything with this interface. */
384 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
387 struct iface **local_ifacep = local_ifacep_;
389 if (iface->dp_ifidx >= 0) {
390 if (iface->dp_ifidx == ODPP_LOCAL) {
391 *local_ifacep = iface;
393 VLOG_DBG("%s has interface %s on port %d",
395 iface->name, iface->dp_ifidx);
398 VLOG_ERR("%s interface not in %s, dropping",
399 iface->name, dpif_name(br->dpif));
404 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
405 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
406 * deletes from 'br' any ports that no longer have any interfaces. */
408 iterate_and_prune_ifaces(struct bridge *br,
409 bool (*cb)(struct bridge *, struct iface *,
415 for (i = 0; i < br->n_ports; ) {
416 struct port *port = br->ports[i];
417 for (j = 0; j < port->n_ifaces; ) {
418 struct iface *iface = port->ifaces[j];
419 if (cb(br, iface, aux)) {
422 iface_destroy(iface);
426 if (port->n_ifaces) {
429 VLOG_ERR("%s port has no interfaces, dropping", port->name);
436 bridge_reconfigure(void)
438 struct svec old_br, new_br;
439 struct bridge *br, *next;
442 COVERAGE_INC(bridge_reconfigure);
444 /* Collect old and new bridges. */
447 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
448 svec_add(&old_br, br->name);
450 cfg_get_subsections(&new_br, "bridge");
452 /* Get rid of deleted bridges and add new bridges. */
455 assert(svec_is_unique(&old_br));
456 assert(svec_is_unique(&new_br));
457 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
458 if (!svec_contains(&new_br, br->name)) {
462 for (i = 0; i < new_br.n; i++) {
463 const char *name = new_br.names[i];
464 if (!svec_contains(&old_br, name)) {
468 svec_destroy(&old_br);
469 svec_destroy(&new_br);
473 bridge_configure_ssl();
476 /* Reconfigure all bridges. */
477 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
478 bridge_reconfigure_one(br);
481 /* Add and delete ports on all datapaths.
483 * The kernel will reject any attempt to add a given port to a datapath if
484 * that port already belongs to a different datapath, so we must do all
485 * port deletions before any port additions. */
486 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
487 struct odp_port *dpif_ports;
489 struct svec want_ifaces;
491 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
492 bridge_get_all_ifaces(br, &want_ifaces);
493 for (i = 0; i < n_dpif_ports; i++) {
494 const struct odp_port *p = &dpif_ports[i];
495 if (!svec_contains(&want_ifaces, p->devname)
496 && strcmp(p->devname, br->name)) {
497 int retval = dpif_port_del(br->dpif, p->port);
499 VLOG_ERR("failed to remove %s interface from %s: %s",
500 p->devname, dpif_name(br->dpif),
505 svec_destroy(&want_ifaces);
508 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
509 struct odp_port *dpif_ports;
511 struct svec cur_ifaces, want_ifaces, add_ifaces;
513 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
514 svec_init(&cur_ifaces);
515 for (i = 0; i < n_dpif_ports; i++) {
516 svec_add(&cur_ifaces, dpif_ports[i].devname);
519 svec_sort_unique(&cur_ifaces);
520 bridge_get_all_ifaces(br, &want_ifaces);
521 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
523 for (i = 0; i < add_ifaces.n; i++) {
524 const char *if_name = add_ifaces.names[i];
525 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
526 int flags = internal ? ODP_PORT_INTERNAL : 0;
527 int error = dpif_port_add(br->dpif, if_name, flags, NULL);
528 if (error == EXFULL) {
529 VLOG_ERR("ran out of valid port numbers on %s",
530 dpif_name(br->dpif));
533 VLOG_ERR("failed to add %s interface to %s: %s",
534 if_name, dpif_name(br->dpif), strerror(error));
537 svec_destroy(&cur_ifaces);
538 svec_destroy(&want_ifaces);
539 svec_destroy(&add_ifaces);
541 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
544 struct iface *local_iface = NULL;
546 uint8_t engine_type, engine_id;
547 bool add_id_to_iface = false;
548 struct svec nf_hosts;
550 bridge_fetch_dp_ifaces(br);
551 iterate_and_prune_ifaces(br, init_iface_netdev, NULL);
554 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, &local_iface);
556 /* Pick local port hardware address, datapath ID. */
557 bridge_pick_local_hw_addr(br, ea, &devname);
559 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
561 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
562 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
563 "Ethernet address: %s",
564 br->name, strerror(error));
568 dpid = bridge_pick_datapath_id(br, ea, devname);
569 ofproto_set_datapath_id(br->ofproto, dpid);
571 /* Set NetFlow configuration on this bridge. */
572 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
573 if (cfg_has("netflow.%s.engine-type", br->name)) {
574 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
577 if (cfg_has("netflow.%s.engine-id", br->name)) {
578 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
580 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
581 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
584 if (add_id_to_iface && engine_id > 0x7f) {
585 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
586 "another vswitch, choose an engine id less than 128",
589 if (add_id_to_iface && br->n_ports > 0x1ff) {
590 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
591 "another port when 512 or more ports are used",
594 svec_init(&nf_hosts);
595 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
596 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
597 engine_id, add_id_to_iface)) {
598 VLOG_ERR("bridge %s: problem setting netflow collectors",
602 /* Update the controller and related settings. It would be more
603 * straightforward to call this from bridge_reconfigure_one(), but we
604 * can't do it there for two reasons. First, and most importantly, at
605 * that point we don't know the dp_ifidx of any interfaces that have
606 * been added to the bridge (because we haven't actually added them to
607 * the datapath). Second, at that point we haven't set the datapath ID
608 * yet; when a controller is configured, resetting the datapath ID will
609 * immediately disconnect from the controller, so it's better to set
610 * the datapath ID before the controller. */
611 bridge_reconfigure_controller(br);
613 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
614 for (i = 0; i < br->n_ports; i++) {
615 struct port *port = br->ports[i];
616 port_update_vlan_compat(port);
617 port_update_bonding(port);
620 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
621 brstp_reconfigure(br);
626 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
627 const char **devname)
629 uint64_t requested_ea;
635 /* Did the user request a particular MAC? */
636 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
638 eth_addr_from_uint64(requested_ea, ea);
639 if (eth_addr_is_multicast(ea)) {
640 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
641 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
642 } else if (eth_addr_is_zero(ea)) {
643 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
649 /* Otherwise choose the minimum MAC address among all of the interfaces.
650 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
651 * MAC of the physical interface in such an environment.) */
652 memset(ea, 0xff, sizeof ea);
653 for (i = 0; i < br->n_ports; i++) {
654 struct port *port = br->ports[i];
655 if (port->is_mirror_output_port) {
658 for (j = 0; j < port->n_ifaces; j++) {
659 struct iface *iface = port->ifaces[j];
660 uint8_t iface_ea[ETH_ADDR_LEN];
661 if (iface->dp_ifidx == ODPP_LOCAL
662 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
665 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
667 if (!eth_addr_is_multicast(iface_ea) &&
668 !eth_addr_is_reserved(iface_ea) &&
669 !eth_addr_is_zero(iface_ea) &&
670 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
671 memcpy(ea, iface_ea, ETH_ADDR_LEN);
672 *devname = iface->name;
675 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
676 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
677 iface->name, strerror(error));
681 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
682 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
684 VLOG_WARN("bridge %s: using default bridge Ethernet "
685 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
687 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
688 br->name, ETH_ADDR_ARGS(ea));
692 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
693 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
694 * a network device, then that network device's name must be passed in as
695 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
696 * passed in as a null pointer. */
698 bridge_pick_datapath_id(struct bridge *br,
699 const uint8_t bridge_ea[ETH_ADDR_LEN],
703 * The procedure for choosing a bridge MAC address will, in the most
704 * ordinary case, also choose a unique MAC that we can use as a datapath
705 * ID. In some special cases, though, multiple bridges will end up with
706 * the same MAC address. This is OK for the bridges, but it will confuse
707 * the OpenFlow controller, because each datapath needs a unique datapath
710 * Datapath IDs must be unique. It is also very desirable that they be
711 * stable from one run to the next, so that policy set on a datapath
716 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
723 if (!netdev_get_vlan_vid(devname, &vlan)) {
725 * A bridge whose MAC address is taken from a VLAN network device
726 * (that is, a network device created with vconfig(8) or similar
727 * tool) will have the same MAC address as a bridge on the VLAN
728 * device's physical network device.
730 * Handle this case by hashing the physical network device MAC
731 * along with the VLAN identifier.
733 uint8_t buf[ETH_ADDR_LEN + 2];
734 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
735 buf[ETH_ADDR_LEN] = vlan >> 8;
736 buf[ETH_ADDR_LEN + 1] = vlan;
737 return dpid_from_hash(buf, sizeof buf);
740 * Assume that this bridge's MAC address is unique, since it
741 * doesn't fit any of the cases we handle specially.
746 * A purely internal bridge, that is, one that has no non-virtual
747 * network devices on it at all, is more difficult because it has no
748 * natural unique identifier at all.
750 * When the host is a XenServer, we handle this case by hashing the
751 * host's UUID with the name of the bridge. Names of bridges are
752 * persistent across XenServer reboots, although they can be reused if
753 * an internal network is destroyed and then a new one is later
754 * created, so this is fairly effective.
756 * When the host is not a XenServer, we punt by using a random MAC
757 * address on each run.
759 const char *host_uuid = xenserver_get_host_uuid();
761 char *combined = xasprintf("%s,%s", host_uuid, br->name);
762 dpid = dpid_from_hash(combined, strlen(combined));
768 return eth_addr_to_uint64(bridge_ea);
772 dpid_from_hash(const void *data, size_t n)
774 uint8_t hash[SHA1_DIGEST_SIZE];
776 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
777 sha1_bytes(data, n, hash);
778 eth_addr_mark_random(hash);
779 return eth_addr_to_uint64(hash);
785 struct bridge *br, *next;
789 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
790 int error = bridge_run_one(br);
792 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
793 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
794 "forcing reconfiguration", br->name);
808 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
809 ofproto_wait(br->ofproto);
810 if (br->controller) {
815 mac_learning_wait(br->ml);
822 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
823 * configuration changes. */
825 bridge_flush(struct bridge *br)
827 COVERAGE_INC(bridge_flush);
830 mac_learning_flush(br->ml);
834 /* Bridge unixctl user interface functions. */
836 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
838 struct ds ds = DS_EMPTY_INITIALIZER;
839 const struct bridge *br;
841 br = bridge_lookup(args);
843 unixctl_command_reply(conn, 501, "no such bridge");
847 ds_put_cstr(&ds, " port VLAN MAC Age\n");
849 const struct mac_entry *e;
850 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
851 if (e->port < 0 || e->port >= br->n_ports) {
854 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
855 br->ports[e->port]->ifaces[0]->dp_ifidx,
856 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
859 unixctl_command_reply(conn, 200, ds_cstr(&ds));
863 /* Bridge reconfiguration functions. */
865 static struct bridge *
866 bridge_create(const char *name)
871 assert(!bridge_lookup(name));
872 br = xcalloc(1, sizeof *br);
874 error = dpif_create(name, &br->dpif);
875 if (error == EEXIST || error == EBUSY) {
876 error = dpif_open(name, &br->dpif);
878 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
879 name, strerror(error));
883 dpif_flow_flush(br->dpif);
885 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
890 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
892 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
893 dpif_delete(br->dpif);
894 dpif_close(br->dpif);
899 br->name = xstrdup(name);
900 br->ml = mac_learning_create();
901 br->sent_config_request = false;
902 eth_addr_random(br->default_ea);
904 port_array_init(&br->ifaces);
907 br->bond_next_rebalance = time_msec() + 10000;
909 list_push_back(&all_bridges, &br->node);
911 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
917 bridge_destroy(struct bridge *br)
922 while (br->n_ports > 0) {
923 port_destroy(br->ports[br->n_ports - 1]);
925 list_remove(&br->node);
926 error = dpif_delete(br->dpif);
927 if (error && error != ENOENT) {
928 VLOG_ERR("failed to delete %s: %s",
929 dpif_name(br->dpif), strerror(error));
931 dpif_close(br->dpif);
932 ofproto_destroy(br->ofproto);
933 free(br->controller);
934 mac_learning_destroy(br->ml);
935 port_array_destroy(&br->ifaces);
942 static struct bridge *
943 bridge_lookup(const char *name)
947 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
948 if (!strcmp(br->name, name)) {
956 bridge_exists(const char *name)
958 return bridge_lookup(name) ? true : false;
962 bridge_get_datapathid(const char *name)
964 struct bridge *br = bridge_lookup(name);
965 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
969 bridge_run_one(struct bridge *br)
973 error = ofproto_run1(br->ofproto);
979 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
984 error = ofproto_run2(br->ofproto, br->flush);
991 bridge_get_controller(const struct bridge *br)
993 const char *controller;
995 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
997 controller = cfg_get_string(0, "mgmt.controller");
999 return controller && controller[0] ? controller : NULL;
1003 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1005 struct svec *ifaces = ifaces_;
1006 if (!svec_contains(ifaces, iface->name)) {
1007 svec_add(ifaces, iface->name);
1011 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1013 br->name, iface->name, iface->port->name);
1019 bridge_reconfigure_one(struct bridge *br)
1021 struct svec old_ports, new_ports, ifaces;
1022 struct svec listeners, old_listeners;
1023 struct svec snoops, old_snoops;
1026 /* Collect old ports. */
1027 svec_init(&old_ports);
1028 for (i = 0; i < br->n_ports; i++) {
1029 svec_add(&old_ports, br->ports[i]->name);
1031 svec_sort(&old_ports);
1032 assert(svec_is_unique(&old_ports));
1034 /* Collect new ports. */
1035 svec_init(&new_ports);
1036 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1037 svec_sort(&new_ports);
1038 if (bridge_get_controller(br)) {
1039 char local_name[IF_NAMESIZE];
1042 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1043 local_name, sizeof local_name);
1044 if (!error && !svec_contains(&new_ports, local_name)) {
1045 svec_add(&new_ports, local_name);
1046 svec_sort(&new_ports);
1049 if (!svec_is_unique(&new_ports)) {
1050 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1051 br->name, svec_get_duplicate(&new_ports));
1052 svec_unique(&new_ports);
1055 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1057 /* Get rid of deleted ports and add new ports. */
1058 for (i = 0; i < br->n_ports; ) {
1059 struct port *port = br->ports[i];
1060 if (!svec_contains(&new_ports, port->name)) {
1066 for (i = 0; i < new_ports.n; i++) {
1067 const char *name = new_ports.names[i];
1068 if (!svec_contains(&old_ports, name)) {
1069 port_create(br, name);
1072 svec_destroy(&old_ports);
1073 svec_destroy(&new_ports);
1075 /* Reconfigure all ports. */
1076 for (i = 0; i < br->n_ports; i++) {
1077 port_reconfigure(br->ports[i]);
1080 /* Check and delete duplicate interfaces. */
1082 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1083 svec_destroy(&ifaces);
1085 /* Delete all flows if we're switching from connected to standalone or vice
1086 * versa. (XXX Should we delete all flows if we are switching from one
1087 * controller to another?) */
1089 /* Configure OpenFlow management listeners. */
1090 svec_init(&listeners);
1091 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1093 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1094 ovs_rundir, br->name));
1095 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1096 svec_clear(&listeners);
1098 svec_sort_unique(&listeners);
1100 svec_init(&old_listeners);
1101 ofproto_get_listeners(br->ofproto, &old_listeners);
1102 svec_sort_unique(&old_listeners);
1104 if (!svec_equal(&listeners, &old_listeners)) {
1105 ofproto_set_listeners(br->ofproto, &listeners);
1107 svec_destroy(&listeners);
1108 svec_destroy(&old_listeners);
1110 /* Configure OpenFlow controller connection snooping. */
1112 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1114 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1115 ovs_rundir, br->name));
1116 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1117 svec_clear(&snoops);
1119 svec_sort_unique(&snoops);
1121 svec_init(&old_snoops);
1122 ofproto_get_snoops(br->ofproto, &old_snoops);
1123 svec_sort_unique(&old_snoops);
1125 if (!svec_equal(&snoops, &old_snoops)) {
1126 ofproto_set_snoops(br->ofproto, &snoops);
1128 svec_destroy(&snoops);
1129 svec_destroy(&old_snoops);
1131 mirror_reconfigure(br);
1135 bridge_reconfigure_controller(struct bridge *br)
1137 char *pfx = xasprintf("bridge.%s.controller", br->name);
1138 const char *controller;
1140 controller = bridge_get_controller(br);
1141 if ((br->controller != NULL) != (controller != NULL)) {
1142 ofproto_flush_flows(br->ofproto);
1144 free(br->controller);
1145 br->controller = controller ? xstrdup(controller) : NULL;
1148 const char *fail_mode;
1149 int max_backoff, probe;
1150 int rate_limit, burst_limit;
1152 if (!strcmp(controller, "discover")) {
1153 bool update_resolv_conf = true;
1155 if (cfg_has("%s.update-resolv.conf", pfx)) {
1156 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1159 ofproto_set_discovery(br->ofproto, true,
1160 cfg_get_string(0, "%s.accept-regex", pfx),
1161 update_resolv_conf);
1163 char local_name[IF_NAMESIZE];
1164 struct netdev *netdev;
1168 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1170 || cfg_get_bool(0, "%s.in-band", pfx));
1171 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1172 ofproto_set_in_band(br->ofproto, in_band);
1174 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1175 local_name, sizeof local_name);
1177 error = netdev_open(local_name, NETDEV_ETH_TYPE_NONE, &netdev);
1180 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1181 struct in_addr ip, mask, gateway;
1182 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1183 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1184 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1186 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1188 mask.s_addr = guess_netmask(ip.s_addr);
1190 if (!netdev_set_in4(netdev, ip, mask)) {
1191 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1193 br->name, IP_ARGS(&ip.s_addr),
1194 IP_ARGS(&mask.s_addr));
1197 if (gateway.s_addr) {
1198 if (!netdev_add_router(netdev, gateway)) {
1199 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1200 br->name, IP_ARGS(&gateway.s_addr));
1204 netdev_close(netdev);
1208 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1210 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1212 ofproto_set_failure(br->ofproto,
1214 || !strcmp(fail_mode, "standalone")
1215 || !strcmp(fail_mode, "open")));
1217 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1219 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1224 ofproto_set_probe_interval(br->ofproto, probe);
1226 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1228 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1233 ofproto_set_max_backoff(br->ofproto, max_backoff);
1235 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1237 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1239 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1241 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1243 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1245 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1247 if (cfg_has("%s.commands.acl", pfx)) {
1248 struct svec command_acls;
1251 svec_init(&command_acls);
1252 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1253 command_acl = svec_join(&command_acls, ",", "");
1255 ofproto_set_remote_execution(br->ofproto, command_acl,
1256 cfg_get_string(0, "%s.commands.dir",
1259 svec_destroy(&command_acls);
1262 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1265 union ofp_action action;
1268 /* Set up a flow that matches every packet and directs them to
1269 * OFPP_NORMAL (which goes to us). */
1270 memset(&action, 0, sizeof action);
1271 action.type = htons(OFPAT_OUTPUT);
1272 action.output.len = htons(sizeof action);
1273 action.output.port = htons(OFPP_NORMAL);
1274 memset(&flow, 0, sizeof flow);
1275 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1278 ofproto_set_in_band(br->ofproto, false);
1279 ofproto_set_max_backoff(br->ofproto, 1);
1280 ofproto_set_probe_interval(br->ofproto, 5);
1281 ofproto_set_failure(br->ofproto, false);
1282 ofproto_set_stp(br->ofproto, false);
1286 ofproto_set_controller(br->ofproto, br->controller);
1290 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1295 for (i = 0; i < br->n_ports; i++) {
1296 struct port *port = br->ports[i];
1297 for (j = 0; j < port->n_ifaces; j++) {
1298 struct iface *iface = port->ifaces[j];
1299 svec_add(ifaces, iface->name);
1303 assert(svec_is_unique(ifaces));
1306 /* For robustness, in case the administrator moves around datapath ports behind
1307 * our back, we re-check all the datapath port numbers here.
1309 * This function will set the 'dp_ifidx' members of interfaces that have
1310 * disappeared to -1, so only call this function from a context where those
1311 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1312 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1313 * datapath, which doesn't support UINT16_MAX+1 ports. */
1315 bridge_fetch_dp_ifaces(struct bridge *br)
1317 struct odp_port *dpif_ports;
1318 size_t n_dpif_ports;
1321 /* Reset all interface numbers. */
1322 for (i = 0; i < br->n_ports; i++) {
1323 struct port *port = br->ports[i];
1324 for (j = 0; j < port->n_ifaces; j++) {
1325 struct iface *iface = port->ifaces[j];
1326 iface->dp_ifidx = -1;
1329 port_array_clear(&br->ifaces);
1331 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1332 for (i = 0; i < n_dpif_ports; i++) {
1333 struct odp_port *p = &dpif_ports[i];
1334 struct iface *iface = iface_lookup(br, p->devname);
1336 if (iface->dp_ifidx >= 0) {
1337 VLOG_WARN("%s reported interface %s twice",
1338 dpif_name(br->dpif), p->devname);
1339 } else if (iface_from_dp_ifidx(br, p->port)) {
1340 VLOG_WARN("%s reported interface %"PRIu16" twice",
1341 dpif_name(br->dpif), p->port);
1343 port_array_set(&br->ifaces, p->port, iface);
1344 iface->dp_ifidx = p->port;
1351 /* Bridge packet processing functions. */
1354 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1356 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1359 static struct bond_entry *
1360 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1362 return &port->bond_hash[bond_hash(mac)];
1366 bond_choose_iface(const struct port *port)
1369 for (i = 0; i < port->n_ifaces; i++) {
1370 if (port->ifaces[i]->enabled) {
1378 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1379 uint16_t *dp_ifidx, tag_type *tags)
1381 struct iface *iface;
1383 assert(port->n_ifaces);
1384 if (port->n_ifaces == 1) {
1385 iface = port->ifaces[0];
1387 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1388 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1389 || !port->ifaces[e->iface_idx]->enabled) {
1390 /* XXX select interface properly. The current interface selection
1391 * is only good for testing the rebalancing code. */
1392 e->iface_idx = bond_choose_iface(port);
1393 if (e->iface_idx < 0) {
1394 *tags |= port->no_ifaces_tag;
1397 e->iface_tag = tag_create_random();
1399 *tags |= e->iface_tag;
1400 iface = port->ifaces[e->iface_idx];
1402 *dp_ifidx = iface->dp_ifidx;
1403 *tags |= iface->tag; /* Currently only used for bonding. */
1408 bond_link_status_update(struct iface *iface, bool carrier)
1410 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1411 struct port *port = iface->port;
1413 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1414 /* Nothing to do. */
1417 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1418 iface->name, carrier ? "detected" : "dropped");
1419 if (carrier == iface->enabled) {
1420 iface->delay_expires = LLONG_MAX;
1421 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1422 iface->name, carrier ? "disabled" : "enabled");
1423 } else if (carrier && port->updelay && port->active_iface < 0) {
1424 iface->delay_expires = time_msec();
1425 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1426 "other interface is up", iface->name, port->updelay);
1428 int delay = carrier ? port->updelay : port->downdelay;
1429 iface->delay_expires = time_msec() + delay;
1432 "interface %s: will be %s if it stays %s for %d ms",
1434 carrier ? "enabled" : "disabled",
1435 carrier ? "up" : "down",
1442 bond_choose_active_iface(struct port *port)
1444 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1446 port->active_iface = bond_choose_iface(port);
1447 port->active_iface_tag = tag_create_random();
1448 if (port->active_iface >= 0) {
1449 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1450 port->name, port->ifaces[port->active_iface]->name);
1452 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1458 bond_enable_slave(struct iface *iface, bool enable)
1460 struct port *port = iface->port;
1461 struct bridge *br = port->bridge;
1463 iface->delay_expires = LLONG_MAX;
1464 if (enable == iface->enabled) {
1468 iface->enabled = enable;
1469 if (!iface->enabled) {
1470 VLOG_WARN("interface %s: disabled", iface->name);
1471 ofproto_revalidate(br->ofproto, iface->tag);
1472 if (iface->port_ifidx == port->active_iface) {
1473 ofproto_revalidate(br->ofproto,
1474 port->active_iface_tag);
1475 bond_choose_active_iface(port);
1477 bond_send_learning_packets(port);
1479 VLOG_WARN("interface %s: enabled", iface->name);
1480 if (port->active_iface < 0) {
1481 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1482 bond_choose_active_iface(port);
1483 bond_send_learning_packets(port);
1485 iface->tag = tag_create_random();
1490 bond_run(struct bridge *br)
1494 for (i = 0; i < br->n_ports; i++) {
1495 struct port *port = br->ports[i];
1496 if (port->n_ifaces < 2) {
1499 for (j = 0; j < port->n_ifaces; j++) {
1500 struct iface *iface = port->ifaces[j];
1501 if (time_msec() >= iface->delay_expires) {
1502 bond_enable_slave(iface, !iface->enabled);
1509 bond_wait(struct bridge *br)
1513 for (i = 0; i < br->n_ports; i++) {
1514 struct port *port = br->ports[i];
1515 if (port->n_ifaces < 2) {
1518 for (j = 0; j < port->n_ifaces; j++) {
1519 struct iface *iface = port->ifaces[j];
1520 if (iface->delay_expires != LLONG_MAX) {
1521 poll_timer_wait(iface->delay_expires - time_msec());
1528 set_dst(struct dst *p, const flow_t *flow,
1529 const struct port *in_port, const struct port *out_port,
1534 * XXX This uses too many tags: any broadcast flow will get one tag per
1535 * destination port, and thus a broadcast on a switch of any size is likely
1536 * to have all tag bits set. We should figure out a way to be smarter.
1538 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1539 *tags |= out_port->stp_state_tag;
1540 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1544 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1545 : in_port->vlan >= 0 ? in_port->vlan
1546 : ntohs(flow->dl_vlan));
1547 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1551 swap_dst(struct dst *p, struct dst *q)
1553 struct dst tmp = *p;
1558 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1559 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1560 * that we push to the datapath. We could in fact fully sort the array by
1561 * vlan, but in most cases there are at most two different vlan tags so that's
1562 * possibly overkill.) */
1564 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1566 struct dst *first = dsts;
1567 struct dst *last = dsts + n_dsts;
1569 while (first != last) {
1571 * - All dsts < first have vlan == 'vlan'.
1572 * - All dsts >= last have vlan != 'vlan'.
1573 * - first < last. */
1574 while (first->vlan == vlan) {
1575 if (++first == last) {
1580 /* Same invariants, plus one additional:
1581 * - first->vlan != vlan.
1583 while (last[-1].vlan != vlan) {
1584 if (--last == first) {
1589 /* Same invariants, plus one additional:
1590 * - last[-1].vlan == vlan.*/
1591 swap_dst(first++, --last);
1596 mirror_mask_ffs(mirror_mask_t mask)
1598 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1603 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1604 const struct dst *test)
1607 for (i = 0; i < n_dsts; i++) {
1608 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1616 port_trunks_vlan(const struct port *port, uint16_t vlan)
1618 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1622 port_includes_vlan(const struct port *port, uint16_t vlan)
1624 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1628 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1629 const struct port *in_port, const struct port *out_port,
1630 struct dst dsts[], tag_type *tags)
1632 mirror_mask_t mirrors = in_port->src_mirrors;
1633 struct dst *dst = dsts;
1636 *tags |= in_port->stp_state_tag;
1637 if (out_port == FLOOD_PORT) {
1638 /* XXX use ODP_FLOOD if no vlans or bonding. */
1639 /* XXX even better, define each VLAN as a datapath port group */
1640 for (i = 0; i < br->n_ports; i++) {
1641 struct port *port = br->ports[i];
1642 if (port != in_port && port_includes_vlan(port, vlan)
1643 && !port->is_mirror_output_port
1644 && set_dst(dst, flow, in_port, port, tags)) {
1645 mirrors |= port->dst_mirrors;
1649 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1650 mirrors |= out_port->dst_mirrors;
1655 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1656 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1658 if (set_dst(dst, flow, in_port, m->out_port, tags)
1659 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1663 for (i = 0; i < br->n_ports; i++) {
1664 struct port *port = br->ports[i];
1665 if (port_includes_vlan(port, m->out_vlan)
1666 && set_dst(dst, flow, in_port, port, tags)
1667 && !dst_is_duplicate(dsts, dst - dsts, dst))
1669 if (port->vlan < 0) {
1670 dst->vlan = m->out_vlan;
1672 if (dst->dp_ifidx == flow->in_port
1673 && dst->vlan == vlan) {
1674 /* Don't send out input port on same VLAN. */
1682 mirrors &= mirrors - 1;
1685 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1690 print_dsts(const struct dst *dsts, size_t n)
1692 for (; n--; dsts++) {
1693 printf(">p%"PRIu16, dsts->dp_ifidx);
1694 if (dsts->vlan != OFP_VLAN_NONE) {
1695 printf("v%"PRIu16, dsts->vlan);
1701 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1702 const struct port *in_port, const struct port *out_port,
1703 tag_type *tags, struct odp_actions *actions)
1705 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1707 const struct dst *p;
1710 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1712 cur_vlan = ntohs(flow->dl_vlan);
1713 for (p = dsts; p < &dsts[n_dsts]; p++) {
1714 union odp_action *a;
1715 if (p->vlan != cur_vlan) {
1716 if (p->vlan == OFP_VLAN_NONE) {
1717 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1719 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1720 a->vlan_vid.vlan_vid = htons(p->vlan);
1724 a = odp_actions_add(actions, ODPAT_OUTPUT);
1725 a->output.port = p->dp_ifidx;
1730 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1732 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1733 return (flow->dl_type == htons(ETH_TYPE_ARP)
1734 && eth_addr_is_broadcast(flow->dl_dst)
1735 && packet->size >= sizeof(struct arp_eth_header)
1736 && arp->ar_op == ARP_OP_REQUEST);
1739 /* If the composed actions may be applied to any packet in the given 'flow',
1740 * returns true. Otherwise, the actions should only be applied to 'packet', or
1741 * not at all, if 'packet' was NULL. */
1743 process_flow(struct bridge *br, const flow_t *flow,
1744 const struct ofpbuf *packet, struct odp_actions *actions,
1747 struct iface *in_iface;
1748 struct port *in_port;
1749 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1752 /* Find the interface and port structure for the received packet. */
1753 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1755 /* No interface? Something fishy... */
1756 if (packet != NULL) {
1757 /* Odd. A few possible reasons here:
1759 * - We deleted an interface but there are still a few packets
1760 * queued up from it.
1762 * - Someone externally added an interface (e.g. with "ovs-dpctl
1763 * add-if") that we don't know about.
1765 * - Packet arrived on the local port but the local port is not
1766 * one of our bridge ports.
1768 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1770 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1771 "interface %"PRIu16, br->name, flow->in_port);
1774 /* Return without adding any actions, to drop packets on this flow. */
1777 in_port = in_iface->port;
1779 /* Figure out what VLAN this packet belongs to.
1781 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1782 * belongs to VLAN 0, so we should treat both cases identically. (In the
1783 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1784 * presumably to allow a priority to be specified. In the latter case, the
1785 * packet does not have any 802.1Q header.) */
1786 vlan = ntohs(flow->dl_vlan);
1787 if (vlan == OFP_VLAN_NONE) {
1790 if (in_port->vlan >= 0) {
1792 /* XXX support double tagging? */
1793 if (packet != NULL) {
1794 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1795 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1796 "packet received on port %s configured with "
1797 "implicit VLAN %"PRIu16,
1798 br->name, ntohs(flow->dl_vlan),
1799 in_port->name, in_port->vlan);
1803 vlan = in_port->vlan;
1805 if (!port_includes_vlan(in_port, vlan)) {
1806 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1807 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1808 "packet received on port %s not configured for "
1810 br->name, vlan, in_port->name, vlan);
1815 /* Drop frames for ports that STP wants entirely killed (both for
1816 * forwarding and for learning). Later, after we do learning, we'll drop
1817 * the frames that STP wants to do learning but not forwarding on. */
1818 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1822 /* Drop frames for reserved multicast addresses. */
1823 if (eth_addr_is_reserved(flow->dl_dst)) {
1827 /* Drop frames on ports reserved for mirroring. */
1828 if (in_port->is_mirror_output_port) {
1829 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1830 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1831 "which is reserved exclusively for mirroring",
1832 br->name, in_port->name);
1836 /* Multicast (and broadcast) packets on bonds need special attention, to
1837 * avoid receiving duplicates. */
1838 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1839 *tags |= in_port->active_iface_tag;
1840 if (in_port->active_iface != in_iface->port_ifidx) {
1841 /* Drop all multicast packets on inactive slaves. */
1844 /* Drop all multicast packets for which we have learned a different
1845 * input port, because we probably sent the packet on one slaves
1846 * and got it back on the active slave. Broadcast ARP replies are
1847 * an exception to this rule: the host has moved to another
1849 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1850 if (src_idx != -1 && src_idx != in_port->port_idx) {
1852 if (!is_bcast_arp_reply(flow, packet)) {
1856 /* No way to know whether it's an ARP reply, because the
1857 * flow entry doesn't include enough information and we
1858 * don't have a packet. Punt. */
1866 out_port = FLOOD_PORT;
1870 /* Learn source MAC (but don't try to learn from revalidation). */
1872 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1873 vlan, in_port->port_idx);
1875 /* The log messages here could actually be useful in debugging,
1876 * so keep the rate limit relatively high. */
1877 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1879 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1880 "on port %s in VLAN %d",
1881 br->name, ETH_ADDR_ARGS(flow->dl_src),
1882 in_port->name, vlan);
1883 ofproto_revalidate(br->ofproto, rev_tag);
1887 /* Determine output port. */
1888 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1890 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1891 out_port = br->ports[out_port_idx];
1895 /* Don't send packets out their input ports. Don't forward frames that STP
1896 * wants us to discard. */
1897 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1902 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1905 * We send out only a single packet, instead of setting up a flow, if the
1906 * packet is an ARP directed to broadcast that arrived on a bonded
1907 * interface. In such a situation ARP requests and replies must be handled
1908 * differently, but OpenFlow unfortunately can't distinguish them.
1910 return (in_port->n_ifaces < 2
1911 || flow->dl_type != htons(ETH_TYPE_ARP)
1912 || !eth_addr_is_broadcast(flow->dl_dst));
1915 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1918 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1919 const struct ofp_phy_port *opp,
1922 struct bridge *br = br_;
1923 struct iface *iface;
1926 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1932 if (reason == OFPPR_DELETE) {
1933 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1934 br->name, iface->name);
1935 iface_destroy(iface);
1936 if (!port->n_ifaces) {
1937 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1938 br->name, port->name);
1944 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1945 if (port->n_ifaces > 1) {
1946 bool up = !(opp->state & OFPPS_LINK_DOWN);
1947 bond_link_status_update(iface, up);
1948 port_update_bond_compat(port);
1954 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1955 struct odp_actions *actions, tag_type *tags, void *br_)
1957 struct bridge *br = br_;
1960 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1961 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1962 brstp_receive(br, flow, payload);
1967 COVERAGE_INC(bridge_process_flow);
1968 return process_flow(br, flow, packet, actions, tags);
1972 bridge_account_flow_ofhook_cb(const flow_t *flow,
1973 const union odp_action *actions,
1974 size_t n_actions, unsigned long long int n_bytes,
1977 struct bridge *br = br_;
1978 const union odp_action *a;
1980 if (!br->has_bonded_ports) {
1984 for (a = actions; a < &actions[n_actions]; a++) {
1985 if (a->type == ODPAT_OUTPUT) {
1986 struct port *port = port_from_dp_ifidx(br, a->output.port);
1987 if (port && port->n_ifaces >= 2) {
1988 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1989 e->tx_bytes += n_bytes;
1996 bridge_account_checkpoint_ofhook_cb(void *br_)
1998 struct bridge *br = br_;
2001 if (!br->has_bonded_ports) {
2005 /* The current ofproto implementation calls this callback at least once a
2006 * second, so this timer implementation is sufficient. */
2007 if (time_msec() < br->bond_next_rebalance) {
2010 br->bond_next_rebalance = time_msec() + 10000;
2012 for (i = 0; i < br->n_ports; i++) {
2013 struct port *port = br->ports[i];
2014 if (port->n_ifaces > 1) {
2015 bond_rebalance_port(port);
2020 static struct ofhooks bridge_ofhooks = {
2021 bridge_port_changed_ofhook_cb,
2022 bridge_normal_ofhook_cb,
2023 bridge_account_flow_ofhook_cb,
2024 bridge_account_checkpoint_ofhook_cb,
2027 /* Bonding functions. */
2029 /* Statistics for a single interface on a bonded port, used for load-based
2030 * bond rebalancing. */
2031 struct slave_balance {
2032 struct iface *iface; /* The interface. */
2033 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2035 /* All the "bond_entry"s that are assigned to this interface, in order of
2036 * increasing tx_bytes. */
2037 struct bond_entry **hashes;
2041 /* Sorts pointers to pointers to bond_entries in ascending order by the
2042 * interface to which they are assigned, and within a single interface in
2043 * ascending order of bytes transmitted. */
2045 compare_bond_entries(const void *a_, const void *b_)
2047 const struct bond_entry *const *ap = a_;
2048 const struct bond_entry *const *bp = b_;
2049 const struct bond_entry *a = *ap;
2050 const struct bond_entry *b = *bp;
2051 if (a->iface_idx != b->iface_idx) {
2052 return a->iface_idx > b->iface_idx ? 1 : -1;
2053 } else if (a->tx_bytes != b->tx_bytes) {
2054 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2060 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2061 * *descending* order by number of bytes transmitted. */
2063 compare_slave_balance(const void *a_, const void *b_)
2065 const struct slave_balance *a = a_;
2066 const struct slave_balance *b = b_;
2067 if (a->iface->enabled != b->iface->enabled) {
2068 return a->iface->enabled ? -1 : 1;
2069 } else if (a->tx_bytes != b->tx_bytes) {
2070 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2077 swap_bals(struct slave_balance *a, struct slave_balance *b)
2079 struct slave_balance tmp = *a;
2084 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2085 * given that 'p' (and only 'p') might be in the wrong location.
2087 * This function invalidates 'p', since it might now be in a different memory
2090 resort_bals(struct slave_balance *p,
2091 struct slave_balance bals[], size_t n_bals)
2094 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2095 swap_bals(p, p - 1);
2097 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2098 swap_bals(p, p + 1);
2104 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2106 if (VLOG_IS_DBG_ENABLED()) {
2107 struct ds ds = DS_EMPTY_INITIALIZER;
2108 const struct slave_balance *b;
2110 for (b = bals; b < bals + n_bals; b++) {
2114 ds_put_char(&ds, ',');
2116 ds_put_format(&ds, " %s %"PRIu64"kB",
2117 b->iface->name, b->tx_bytes / 1024);
2119 if (!b->iface->enabled) {
2120 ds_put_cstr(&ds, " (disabled)");
2122 if (b->n_hashes > 0) {
2123 ds_put_cstr(&ds, " (");
2124 for (i = 0; i < b->n_hashes; i++) {
2125 const struct bond_entry *e = b->hashes[i];
2127 ds_put_cstr(&ds, " + ");
2129 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2130 e - port->bond_hash, e->tx_bytes / 1024);
2132 ds_put_cstr(&ds, ")");
2135 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2140 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2142 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2143 struct bond_entry *hash)
2145 struct port *port = from->iface->port;
2146 uint64_t delta = hash->tx_bytes;
2148 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2149 "from %s to %s (now carrying %"PRIu64"kB and "
2150 "%"PRIu64"kB load, respectively)",
2151 port->name, delta / 1024, hash - port->bond_hash,
2152 from->iface->name, to->iface->name,
2153 (from->tx_bytes - delta) / 1024,
2154 (to->tx_bytes + delta) / 1024);
2156 /* Delete element from from->hashes.
2158 * We don't bother to add the element to to->hashes because not only would
2159 * it require more work, the only purpose it would be to allow that hash to
2160 * be migrated to another slave in this rebalancing run, and there is no
2161 * point in doing that. */
2162 if (from->hashes[0] == hash) {
2165 int i = hash - from->hashes[0];
2166 memmove(from->hashes + i, from->hashes + i + 1,
2167 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2171 /* Shift load away from 'from' to 'to'. */
2172 from->tx_bytes -= delta;
2173 to->tx_bytes += delta;
2175 /* Arrange for flows to be revalidated. */
2176 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2177 hash->iface_idx = to->iface->port_ifidx;
2178 hash->iface_tag = tag_create_random();
2182 bond_rebalance_port(struct port *port)
2184 struct slave_balance bals[DP_MAX_PORTS];
2186 struct bond_entry *hashes[BOND_MASK + 1];
2187 struct slave_balance *b, *from, *to;
2188 struct bond_entry *e;
2191 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2192 * descending order of tx_bytes, so that bals[0] represents the most
2193 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2196 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2197 * array for each slave_balance structure, we sort our local array of
2198 * hashes in order by slave, so that all of the hashes for a given slave
2199 * become contiguous in memory, and then we point each 'hashes' members of
2200 * a slave_balance structure to the start of a contiguous group. */
2201 n_bals = port->n_ifaces;
2202 for (b = bals; b < &bals[n_bals]; b++) {
2203 b->iface = port->ifaces[b - bals];
2208 for (i = 0; i <= BOND_MASK; i++) {
2209 hashes[i] = &port->bond_hash[i];
2211 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2212 for (i = 0; i <= BOND_MASK; i++) {
2214 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2215 b = &bals[e->iface_idx];
2216 b->tx_bytes += e->tx_bytes;
2218 b->hashes = &hashes[i];
2223 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2224 log_bals(bals, n_bals, port);
2226 /* Discard slaves that aren't enabled (which were sorted to the back of the
2227 * array earlier). */
2228 while (!bals[n_bals - 1].iface->enabled) {
2235 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2236 to = &bals[n_bals - 1];
2237 for (from = bals; from < to; ) {
2238 uint64_t overload = from->tx_bytes - to->tx_bytes;
2239 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2240 /* The extra load on 'from' (and all less-loaded slaves), compared
2241 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2242 * it is less than ~1Mbps. No point in rebalancing. */
2244 } else if (from->n_hashes == 1) {
2245 /* 'from' only carries a single MAC hash, so we can't shift any
2246 * load away from it, even though we want to. */
2249 /* 'from' is carrying significantly more load than 'to', and that
2250 * load is split across at least two different hashes. Pick a hash
2251 * to migrate to 'to' (the least-loaded slave), given that doing so
2252 * must not cause 'to''s load to exceed 'from''s load.
2254 * The sort order we use means that we prefer to shift away the
2255 * smallest hashes instead of the biggest ones. There is little
2256 * reason behind this decision; we could use the opposite sort
2257 * order to shift away big hashes ahead of small ones. */
2260 for (i = 0; i < from->n_hashes; i++) {
2261 uint64_t delta = from->hashes[i]->tx_bytes;
2262 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2266 if (i < from->n_hashes) {
2267 bond_shift_load(from, to, from->hashes[i]);
2269 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2270 * point to different slave_balance structures. It is only
2271 * valid to do these two operations in a row at all because we
2272 * know that 'from' will not move past 'to' and vice versa. */
2273 resort_bals(from, bals, n_bals);
2274 resort_bals(to, bals, n_bals);
2281 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2282 * historical data to decay to <1% in 7 rebalancing runs. */
2283 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2289 bond_send_learning_packets(struct port *port)
2291 struct bridge *br = port->bridge;
2292 struct mac_entry *e;
2293 struct ofpbuf packet;
2294 int error, n_packets, n_errors;
2296 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2300 ofpbuf_init(&packet, 128);
2301 error = n_packets = n_errors = 0;
2302 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2303 static const char s[] = "Open vSwitch Bond Failover";
2304 union ofp_action actions[2], *a;
2305 struct eth_header *eth;
2306 struct llc_snap_header *llc_snap;
2312 if (e->port == port->port_idx
2313 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2317 /* Compose packet to send. */
2318 ofpbuf_clear(&packet);
2319 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2320 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2321 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2322 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2324 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2325 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2326 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2328 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2329 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2330 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2331 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2332 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2334 /* Compose actions. */
2335 memset(actions, 0, sizeof actions);
2338 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2339 a->vlan_vid.len = htons(sizeof *a);
2340 a->vlan_vid.vlan_vid = htons(e->vlan);
2343 a->output.type = htons(OFPAT_OUTPUT);
2344 a->output.len = htons(sizeof *a);
2345 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2350 flow_extract(&packet, ODPP_NONE, &flow);
2351 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2358 ofpbuf_uninit(&packet);
2361 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2362 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2363 "packets, last error was: %s",
2364 port->name, n_errors, n_packets, strerror(error));
2366 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2367 port->name, n_packets);
2371 /* Bonding unixctl user interface functions. */
2374 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2376 struct ds ds = DS_EMPTY_INITIALIZER;
2377 const struct bridge *br;
2379 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2381 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2384 for (i = 0; i < br->n_ports; i++) {
2385 const struct port *port = br->ports[i];
2386 if (port->n_ifaces > 1) {
2389 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2390 for (j = 0; j < port->n_ifaces; j++) {
2391 const struct iface *iface = port->ifaces[j];
2393 ds_put_cstr(&ds, ", ");
2395 ds_put_cstr(&ds, iface->name);
2397 ds_put_char(&ds, '\n');
2401 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2405 static struct port *
2406 bond_find(const char *name)
2408 const struct bridge *br;
2410 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2413 for (i = 0; i < br->n_ports; i++) {
2414 struct port *port = br->ports[i];
2415 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2424 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2426 struct ds ds = DS_EMPTY_INITIALIZER;
2427 const struct port *port;
2430 port = bond_find(args);
2432 unixctl_command_reply(conn, 501, "no such bond");
2436 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2437 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2438 ds_put_format(&ds, "next rebalance: %lld ms\n",
2439 port->bridge->bond_next_rebalance - time_msec());
2440 for (j = 0; j < port->n_ifaces; j++) {
2441 const struct iface *iface = port->ifaces[j];
2442 struct bond_entry *be;
2445 ds_put_format(&ds, "slave %s: %s\n",
2446 iface->name, iface->enabled ? "enabled" : "disabled");
2447 if (j == port->active_iface) {
2448 ds_put_cstr(&ds, "\tactive slave\n");
2450 if (iface->delay_expires != LLONG_MAX) {
2451 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2452 iface->enabled ? "downdelay" : "updelay",
2453 iface->delay_expires - time_msec());
2457 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2458 int hash = be - port->bond_hash;
2459 struct mac_entry *me;
2461 if (be->iface_idx != j) {
2465 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2466 hash, be->tx_bytes / 1024);
2469 if (!port->bridge->ml) {
2473 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2474 &port->bridge->ml->lrus) {
2477 if (bond_hash(me->mac) == hash
2478 && me->port != port->port_idx
2479 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2480 && dp_ifidx == iface->dp_ifidx)
2482 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2483 ETH_ADDR_ARGS(me->mac));
2488 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2493 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2495 char *args = (char *) args_;
2496 char *save_ptr = NULL;
2497 char *bond_s, *hash_s, *slave_s;
2498 uint8_t mac[ETH_ADDR_LEN];
2500 struct iface *iface;
2501 struct bond_entry *entry;
2504 bond_s = strtok_r(args, " ", &save_ptr);
2505 hash_s = strtok_r(NULL, " ", &save_ptr);
2506 slave_s = strtok_r(NULL, " ", &save_ptr);
2508 unixctl_command_reply(conn, 501,
2509 "usage: bond/migrate BOND HASH SLAVE");
2513 port = bond_find(bond_s);
2515 unixctl_command_reply(conn, 501, "no such bond");
2519 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2520 == ETH_ADDR_SCAN_COUNT) {
2521 hash = bond_hash(mac);
2522 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2523 hash = atoi(hash_s) & BOND_MASK;
2525 unixctl_command_reply(conn, 501, "bad hash");
2529 iface = port_lookup_iface(port, slave_s);
2531 unixctl_command_reply(conn, 501, "no such slave");
2535 if (!iface->enabled) {
2536 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2540 entry = &port->bond_hash[hash];
2541 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2542 entry->iface_idx = iface->port_ifidx;
2543 entry->iface_tag = tag_create_random();
2544 unixctl_command_reply(conn, 200, "migrated");
2548 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2550 char *args = (char *) args_;
2551 char *save_ptr = NULL;
2552 char *bond_s, *slave_s;
2554 struct iface *iface;
2556 bond_s = strtok_r(args, " ", &save_ptr);
2557 slave_s = strtok_r(NULL, " ", &save_ptr);
2559 unixctl_command_reply(conn, 501,
2560 "usage: bond/set-active-slave BOND SLAVE");
2564 port = bond_find(bond_s);
2566 unixctl_command_reply(conn, 501, "no such bond");
2570 iface = port_lookup_iface(port, slave_s);
2572 unixctl_command_reply(conn, 501, "no such slave");
2576 if (!iface->enabled) {
2577 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2581 if (port->active_iface != iface->port_ifidx) {
2582 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2583 port->active_iface = iface->port_ifidx;
2584 port->active_iface_tag = tag_create_random();
2585 VLOG_INFO("port %s: active interface is now %s",
2586 port->name, iface->name);
2587 bond_send_learning_packets(port);
2588 unixctl_command_reply(conn, 200, "done");
2590 unixctl_command_reply(conn, 200, "no change");
2595 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2597 char *args = (char *) args_;
2598 char *save_ptr = NULL;
2599 char *bond_s, *slave_s;
2601 struct iface *iface;
2603 bond_s = strtok_r(args, " ", &save_ptr);
2604 slave_s = strtok_r(NULL, " ", &save_ptr);
2606 unixctl_command_reply(conn, 501,
2607 "usage: bond/enable/disable-slave BOND SLAVE");
2611 port = bond_find(bond_s);
2613 unixctl_command_reply(conn, 501, "no such bond");
2617 iface = port_lookup_iface(port, slave_s);
2619 unixctl_command_reply(conn, 501, "no such slave");
2623 bond_enable_slave(iface, enable);
2624 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2628 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2630 enable_slave(conn, args, true);
2634 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2636 enable_slave(conn, args, false);
2642 unixctl_command_register("bond/list", bond_unixctl_list);
2643 unixctl_command_register("bond/show", bond_unixctl_show);
2644 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2645 unixctl_command_register("bond/set-active-slave",
2646 bond_unixctl_set_active_slave);
2647 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2648 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2651 /* Port functions. */
2654 port_create(struct bridge *br, const char *name)
2658 port = xcalloc(1, sizeof *port);
2660 port->port_idx = br->n_ports;
2662 port->trunks = NULL;
2663 port->name = xstrdup(name);
2664 port->active_iface = -1;
2665 port->stp_state = STP_DISABLED;
2666 port->stp_state_tag = 0;
2668 if (br->n_ports >= br->allocated_ports) {
2669 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2672 br->ports[br->n_ports++] = port;
2674 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2679 port_reconfigure(struct port *port)
2681 bool bonded = cfg_has_section("bonding.%s", port->name);
2682 struct svec old_ifaces, new_ifaces;
2683 unsigned long *trunks;
2687 /* Collect old and new interfaces. */
2688 svec_init(&old_ifaces);
2689 svec_init(&new_ifaces);
2690 for (i = 0; i < port->n_ifaces; i++) {
2691 svec_add(&old_ifaces, port->ifaces[i]->name);
2693 svec_sort(&old_ifaces);
2695 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2696 if (!new_ifaces.n) {
2697 VLOG_ERR("port %s: no interfaces specified for bonded port",
2699 } else if (new_ifaces.n == 1) {
2700 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2704 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2705 if (port->updelay < 0) {
2708 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2709 if (port->downdelay < 0) {
2710 port->downdelay = 0;
2713 svec_init(&new_ifaces);
2714 svec_add(&new_ifaces, port->name);
2717 /* Get rid of deleted interfaces and add new interfaces. */
2718 for (i = 0; i < port->n_ifaces; i++) {
2719 struct iface *iface = port->ifaces[i];
2720 if (!svec_contains(&new_ifaces, iface->name)) {
2721 iface_destroy(iface);
2726 for (i = 0; i < new_ifaces.n; i++) {
2727 const char *name = new_ifaces.names[i];
2728 if (!svec_contains(&old_ifaces, name)) {
2729 iface_create(port, name);
2735 if (cfg_has("vlan.%s.tag", port->name)) {
2737 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2738 if (vlan >= 0 && vlan <= 4095) {
2739 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2742 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2743 * they even work as-is. But they have not been tested. */
2744 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2748 if (port->vlan != vlan) {
2750 bridge_flush(port->bridge);
2753 /* Get trunked VLANs. */
2756 size_t n_trunks, n_errors;
2759 trunks = bitmap_allocate(4096);
2760 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2762 for (i = 0; i < n_trunks; i++) {
2763 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2765 bitmap_set1(trunks, trunk);
2771 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2772 port->name, n_trunks);
2774 if (n_errors == n_trunks) {
2776 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2779 bitmap_set_multiple(trunks, 0, 4096, 1);
2782 if (cfg_has("vlan.%s.trunks", port->name)) {
2783 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2784 port->name, port->name);
2788 ? port->trunks != NULL
2789 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2790 bridge_flush(port->bridge);
2792 bitmap_free(port->trunks);
2793 port->trunks = trunks;
2795 svec_destroy(&old_ifaces);
2796 svec_destroy(&new_ifaces);
2800 port_destroy(struct port *port)
2803 struct bridge *br = port->bridge;
2807 proc_net_compat_update_vlan(port->name, NULL, 0);
2809 for (i = 0; i < MAX_MIRRORS; i++) {
2810 struct mirror *m = br->mirrors[i];
2811 if (m && m->out_port == port) {
2816 while (port->n_ifaces > 0) {
2817 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2820 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2821 del->port_idx = port->port_idx;
2824 bitmap_free(port->trunks);
2831 static struct port *
2832 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2834 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2835 return iface ? iface->port : NULL;
2838 static struct port *
2839 port_lookup(const struct bridge *br, const char *name)
2843 for (i = 0; i < br->n_ports; i++) {
2844 struct port *port = br->ports[i];
2845 if (!strcmp(port->name, name)) {
2852 static struct iface *
2853 port_lookup_iface(const struct port *port, const char *name)
2857 for (j = 0; j < port->n_ifaces; j++) {
2858 struct iface *iface = port->ifaces[j];
2859 if (!strcmp(iface->name, name)) {
2867 port_update_bonding(struct port *port)
2869 if (port->n_ifaces < 2) {
2870 /* Not a bonded port. */
2871 if (port->bond_hash) {
2872 free(port->bond_hash);
2873 port->bond_hash = NULL;
2874 proc_net_compat_update_bond(port->name, NULL);
2877 if (!port->bond_hash) {
2880 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2881 for (i = 0; i <= BOND_MASK; i++) {
2882 struct bond_entry *e = &port->bond_hash[i];
2886 port->no_ifaces_tag = tag_create_random();
2887 bond_choose_active_iface(port);
2889 port_update_bond_compat(port);
2894 port_update_bond_compat(struct port *port)
2896 struct compat_bond bond;
2899 if (port->n_ifaces < 2) {
2904 bond.updelay = port->updelay;
2905 bond.downdelay = port->downdelay;
2906 bond.n_slaves = port->n_ifaces;
2907 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2908 for (i = 0; i < port->n_ifaces; i++) {
2909 struct iface *iface = port->ifaces[i];
2910 struct compat_bond_slave *slave = &bond.slaves[i];
2911 slave->name = iface->name;
2912 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2913 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2917 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2919 proc_net_compat_update_bond(port->name, &bond);
2924 port_update_vlan_compat(struct port *port)
2926 struct bridge *br = port->bridge;
2927 char *vlandev_name = NULL;
2929 if (port->vlan > 0) {
2930 /* Figure out the name that the VLAN device should actually have, if it
2931 * existed. This takes some work because the VLAN device would not
2932 * have port->name in its name; rather, it would have the trunk port's
2933 * name, and 'port' would be attached to a bridge that also had the
2934 * VLAN device one of its ports. So we need to find a trunk port that
2935 * includes port->vlan.
2937 * There might be more than one candidate. This doesn't happen on
2938 * XenServer, so if it happens we just pick the first choice in
2939 * alphabetical order instead of creating multiple VLAN devices. */
2941 for (i = 0; i < br->n_ports; i++) {
2942 struct port *p = br->ports[i];
2943 if (port_trunks_vlan(p, port->vlan)
2945 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2947 const uint8_t *ea = p->ifaces[0]->mac;
2948 if (!eth_addr_is_multicast(ea) &&
2949 !eth_addr_is_reserved(ea) &&
2950 !eth_addr_is_zero(ea)) {
2951 vlandev_name = p->name;
2956 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2959 /* Interface functions. */
2962 iface_create(struct port *port, const char *name)
2964 struct iface *iface;
2966 iface = xcalloc(1, sizeof *iface);
2968 iface->port_ifidx = port->n_ifaces;
2969 iface->name = xstrdup(name);
2970 iface->dp_ifidx = -1;
2971 iface->tag = tag_create_random();
2972 iface->delay_expires = LLONG_MAX;
2973 iface->netdev = NULL;
2975 if (port->n_ifaces >= port->allocated_ifaces) {
2976 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2977 sizeof *port->ifaces);
2979 port->ifaces[port->n_ifaces++] = iface;
2980 if (port->n_ifaces > 1) {
2981 port->bridge->has_bonded_ports = true;
2984 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2986 bridge_flush(port->bridge);
2990 iface_destroy(struct iface *iface)
2993 struct port *port = iface->port;
2994 struct bridge *br = port->bridge;
2995 bool del_active = port->active_iface == iface->port_ifidx;
2998 if (iface->dp_ifidx >= 0) {
2999 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3002 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3003 del->port_ifidx = iface->port_ifidx;
3005 netdev_close(iface->netdev);
3010 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3011 bond_choose_active_iface(port);
3012 bond_send_learning_packets(port);
3015 bridge_flush(port->bridge);
3019 static struct iface *
3020 iface_lookup(const struct bridge *br, const char *name)
3024 for (i = 0; i < br->n_ports; i++) {
3025 struct port *port = br->ports[i];
3026 for (j = 0; j < port->n_ifaces; j++) {
3027 struct iface *iface = port->ifaces[j];
3028 if (!strcmp(iface->name, name)) {
3036 static struct iface *
3037 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3039 return port_array_get(&br->ifaces, dp_ifidx);
3042 /* Port mirroring. */
3045 mirror_reconfigure(struct bridge *br)
3047 struct svec old_mirrors, new_mirrors;
3050 /* Collect old and new mirrors. */
3051 svec_init(&old_mirrors);
3052 svec_init(&new_mirrors);
3053 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3054 for (i = 0; i < MAX_MIRRORS; i++) {
3055 if (br->mirrors[i]) {
3056 svec_add(&old_mirrors, br->mirrors[i]->name);
3060 /* Get rid of deleted mirrors and add new mirrors. */
3061 svec_sort(&old_mirrors);
3062 assert(svec_is_unique(&old_mirrors));
3063 svec_sort(&new_mirrors);
3064 assert(svec_is_unique(&new_mirrors));
3065 for (i = 0; i < MAX_MIRRORS; i++) {
3066 struct mirror *m = br->mirrors[i];
3067 if (m && !svec_contains(&new_mirrors, m->name)) {
3071 for (i = 0; i < new_mirrors.n; i++) {
3072 const char *name = new_mirrors.names[i];
3073 if (!svec_contains(&old_mirrors, name)) {
3074 mirror_create(br, name);
3077 svec_destroy(&old_mirrors);
3078 svec_destroy(&new_mirrors);
3080 /* Reconfigure all mirrors. */
3081 for (i = 0; i < MAX_MIRRORS; i++) {
3082 if (br->mirrors[i]) {
3083 mirror_reconfigure_one(br->mirrors[i]);
3087 /* Update port reserved status. */
3088 for (i = 0; i < br->n_ports; i++) {
3089 br->ports[i]->is_mirror_output_port = false;
3091 for (i = 0; i < MAX_MIRRORS; i++) {
3092 struct mirror *m = br->mirrors[i];
3093 if (m && m->out_port) {
3094 m->out_port->is_mirror_output_port = true;
3100 mirror_create(struct bridge *br, const char *name)
3105 for (i = 0; ; i++) {
3106 if (i >= MAX_MIRRORS) {
3107 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3108 "cannot create %s", br->name, MAX_MIRRORS, name);
3111 if (!br->mirrors[i]) {
3116 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3119 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3122 m->name = xstrdup(name);
3123 svec_init(&m->src_ports);
3124 svec_init(&m->dst_ports);
3132 mirror_destroy(struct mirror *m)
3135 struct bridge *br = m->bridge;
3138 for (i = 0; i < br->n_ports; i++) {
3139 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3140 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3143 svec_destroy(&m->src_ports);
3144 svec_destroy(&m->dst_ports);
3147 m->bridge->mirrors[m->idx] = NULL;
3155 prune_ports(struct mirror *m, struct svec *ports)
3160 svec_sort_unique(ports);
3163 for (i = 0; i < ports->n; i++) {
3164 const char *name = ports->names[i];
3165 if (port_lookup(m->bridge, name)) {
3166 svec_add(&tmp, name);
3168 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3169 m->bridge->name, m->name, name);
3172 svec_swap(ports, &tmp);
3177 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3181 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3182 * order won't give us numeric sort order. But that's good enough for what
3183 * we need right now. */
3184 svec_sort_unique(vlan_strings);
3186 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3188 for (i = 0; i < vlan_strings->n; i++) {
3189 const char *name = vlan_strings->names[i];
3191 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3192 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3193 m->bridge->name, m->name, name);
3195 (*vlans)[n_vlans++] = vlan;
3202 vlan_is_mirrored(const struct mirror *m, int vlan)
3206 for (i = 0; i < m->n_vlans; i++) {
3207 if (m->vlans[i] == vlan) {
3215 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3219 for (i = 0; i < m->n_vlans; i++) {
3220 if (port_trunks_vlan(p, m->vlans[i])) {
3228 mirror_reconfigure_one(struct mirror *m)
3230 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3231 struct svec src_ports, dst_ports, ports;
3232 struct svec vlan_strings;
3233 mirror_mask_t mirror_bit;
3234 const char *out_port_name;
3235 struct port *out_port;
3240 bool mirror_all_ports;
3242 /* Get output port. */
3243 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3244 m->bridge->name, m->name);
3245 if (out_port_name) {
3246 out_port = port_lookup(m->bridge, out_port_name);
3248 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3249 "named %s", pfx, m->bridge->name, out_port_name);
3256 if (cfg_has("%s.output.vlan", pfx)) {
3257 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3258 "ignoring %s.output.vlan", pfx, pfx, pfx);
3260 } else if (cfg_has("%s.output.vlan", pfx)) {
3262 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3264 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3265 "but exactly one is required; disabling port mirror %s",
3266 pfx, pfx, pfx, pfx);
3272 /* Get all the ports, and drop duplicates and ports that don't exist. */
3273 svec_init(&src_ports);
3274 svec_init(&dst_ports);
3276 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3277 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3278 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3279 svec_append(&src_ports, &ports);
3280 svec_append(&dst_ports, &ports);
3281 svec_destroy(&ports);
3282 prune_ports(m, &src_ports);
3283 prune_ports(m, &dst_ports);
3285 /* Get all the vlans, and drop duplicate and invalid vlans. */
3286 svec_init(&vlan_strings);
3287 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3288 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3289 svec_destroy(&vlan_strings);
3291 /* Update mirror data. */
3292 if (!svec_equal(&m->src_ports, &src_ports)
3293 || !svec_equal(&m->dst_ports, &dst_ports)
3294 || m->n_vlans != n_vlans
3295 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3296 || m->out_port != out_port
3297 || m->out_vlan != out_vlan) {
3298 bridge_flush(m->bridge);
3300 svec_swap(&m->src_ports, &src_ports);
3301 svec_swap(&m->dst_ports, &dst_ports);
3304 m->n_vlans = n_vlans;
3305 m->out_port = out_port;
3306 m->out_vlan = out_vlan;
3308 /* If no selection criteria have been given, mirror for all ports. */
3309 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3312 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3313 for (i = 0; i < m->bridge->n_ports; i++) {
3314 struct port *port = m->bridge->ports[i];
3316 if (mirror_all_ports
3317 || svec_contains(&m->src_ports, port->name)
3320 ? port_trunks_any_mirrored_vlan(m, port)
3321 : vlan_is_mirrored(m, port->vlan)))) {
3322 port->src_mirrors |= mirror_bit;
3324 port->src_mirrors &= ~mirror_bit;
3327 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3328 port->dst_mirrors |= mirror_bit;
3330 port->dst_mirrors &= ~mirror_bit;
3335 svec_destroy(&src_ports);
3336 svec_destroy(&dst_ports);
3340 /* Spanning tree protocol. */
3342 static void brstp_update_port_state(struct port *);
3345 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3347 struct bridge *br = br_;
3348 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3349 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3351 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3353 } else if (eth_addr_is_zero(iface->mac)) {
3354 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3357 union ofp_action action;
3358 struct eth_header *eth = pkt->l2;
3361 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3363 memset(&action, 0, sizeof action);
3364 action.type = htons(OFPAT_OUTPUT);
3365 action.output.len = htons(sizeof action);
3366 action.output.port = htons(port_no);
3368 flow_extract(pkt, ODPP_NONE, &flow);
3369 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);