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 struct iface **hw_addr_iface);
205 static uint64_t bridge_pick_datapath_id(struct bridge *,
206 const uint8_t bridge_ea[ETH_ADDR_LEN],
207 struct iface *hw_addr_iface);
208 static struct iface *bridge_get_local_iface(struct bridge *);
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 svec_init(&dpif_names);
289 dp_enumerate(&dpif_names);
290 for (i = 0; i < dpif_names.n; i++) {
291 const char *dpif_name = dpif_names.names[i];
295 retval = dpif_open(dpif_name, &dpif);
297 struct svec all_names;
300 svec_init(&all_names);
301 dpif_get_all_names(dpif, &all_names);
302 for (j = 0; j < all_names.n; j++) {
303 if (cfg_has("bridge.%s.port", all_names.names[j])) {
309 svec_destroy(&all_names);
315 bridge_reconfigure();
320 config_string_change(const char *key, char **valuep)
322 const char *value = cfg_get_string(0, "%s", key);
323 if (value && (!*valuep || strcmp(value, *valuep))) {
325 *valuep = xstrdup(value);
333 bridge_configure_ssl(void)
335 /* XXX SSL should be configurable on a per-bridge basis.
336 * XXX should be possible to de-configure SSL. */
337 static char *private_key_file;
338 static char *certificate_file;
339 static char *cacert_file;
342 if (config_string_change("ssl.private-key", &private_key_file)) {
343 vconn_ssl_set_private_key_file(private_key_file);
346 if (config_string_change("ssl.certificate", &certificate_file)) {
347 vconn_ssl_set_certificate_file(certificate_file);
350 /* We assume that even if the filename hasn't changed, if the CA cert
351 * file has been removed, that we want to move back into
352 * boot-strapping mode. This opens a small security hole, because
353 * the old certificate will still be trusted until vSwitch is
354 * restarted. We may want to address this in vconn's SSL library. */
355 if (config_string_change("ssl.ca-cert", &cacert_file)
356 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
357 vconn_ssl_set_ca_cert_file(cacert_file,
358 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
363 /* iterate_and_prune_ifaces() callback function that opens the network device
364 * for 'iface', if it is not already open, and retrieves the interface's MAC
365 * address and carrier status. */
367 init_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
372 } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE,
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, void *aux UNUSED)
386 if (iface->dp_ifidx >= 0) {
387 VLOG_DBG("%s has interface %s on port %d",
389 iface->name, iface->dp_ifidx);
392 VLOG_ERR("%s interface not in %s, dropping",
393 iface->name, dpif_name(br->dpif));
399 set_iface_policing(struct bridge *br UNUSED, struct iface *iface,
402 int rate = cfg_get_int(0, "port.%s.ingress.policing-rate", iface->name);
403 int burst = cfg_get_int(0, "port.%s.ingress.policing-burst", iface->name);
404 netdev_set_policing(iface->netdev, rate, burst);
408 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
409 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
410 * deletes from 'br' any ports that no longer have any interfaces. */
412 iterate_and_prune_ifaces(struct bridge *br,
413 bool (*cb)(struct bridge *, struct iface *,
419 for (i = 0; i < br->n_ports; ) {
420 struct port *port = br->ports[i];
421 for (j = 0; j < port->n_ifaces; ) {
422 struct iface *iface = port->ifaces[j];
423 if (cb(br, iface, aux)) {
426 iface_destroy(iface);
430 if (port->n_ifaces) {
433 VLOG_ERR("%s port has no interfaces, dropping", port->name);
440 bridge_reconfigure(void)
442 struct svec old_br, new_br;
443 struct bridge *br, *next;
446 COVERAGE_INC(bridge_reconfigure);
448 /* Collect old and new bridges. */
451 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
452 svec_add(&old_br, br->name);
454 cfg_get_subsections(&new_br, "bridge");
456 /* Get rid of deleted bridges and add new bridges. */
459 assert(svec_is_unique(&old_br));
460 assert(svec_is_unique(&new_br));
461 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
462 if (!svec_contains(&new_br, br->name)) {
466 for (i = 0; i < new_br.n; i++) {
467 const char *name = new_br.names[i];
468 if (!svec_contains(&old_br, name)) {
472 svec_destroy(&old_br);
473 svec_destroy(&new_br);
477 bridge_configure_ssl();
480 /* Reconfigure all bridges. */
481 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
482 bridge_reconfigure_one(br);
485 /* Add and delete ports on all datapaths.
487 * The kernel will reject any attempt to add a given port to a datapath if
488 * that port already belongs to a different datapath, so we must do all
489 * port deletions before any port additions. */
490 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
491 struct odp_port *dpif_ports;
493 struct svec want_ifaces;
495 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
496 bridge_get_all_ifaces(br, &want_ifaces);
497 for (i = 0; i < n_dpif_ports; i++) {
498 const struct odp_port *p = &dpif_ports[i];
499 if (!svec_contains(&want_ifaces, p->devname)
500 && strcmp(p->devname, br->name)) {
501 int retval = dpif_port_del(br->dpif, p->port);
503 VLOG_ERR("failed to remove %s interface from %s: %s",
504 p->devname, dpif_name(br->dpif),
509 svec_destroy(&want_ifaces);
512 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
513 struct odp_port *dpif_ports;
515 struct svec cur_ifaces, want_ifaces, add_ifaces;
517 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
518 svec_init(&cur_ifaces);
519 for (i = 0; i < n_dpif_ports; i++) {
520 svec_add(&cur_ifaces, dpif_ports[i].devname);
523 svec_sort_unique(&cur_ifaces);
524 bridge_get_all_ifaces(br, &want_ifaces);
525 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
527 for (i = 0; i < add_ifaces.n; i++) {
528 const char *if_name = add_ifaces.names[i];
529 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
530 int flags = internal ? ODP_PORT_INTERNAL : 0;
531 int error = dpif_port_add(br->dpif, if_name, flags, NULL);
532 if (error == EXFULL) {
533 VLOG_ERR("ran out of valid port numbers on %s",
534 dpif_name(br->dpif));
537 VLOG_ERR("failed to add %s interface to %s: %s",
538 if_name, dpif_name(br->dpif), strerror(error));
541 svec_destroy(&cur_ifaces);
542 svec_destroy(&want_ifaces);
543 svec_destroy(&add_ifaces);
545 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
548 struct iface *local_iface;
549 struct iface *hw_addr_iface;
550 uint8_t engine_type, engine_id;
551 bool add_id_to_iface = false;
552 struct svec nf_hosts;
554 bridge_fetch_dp_ifaces(br);
555 iterate_and_prune_ifaces(br, init_iface_netdev, NULL);
557 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
559 /* Pick local port hardware address, datapath ID. */
560 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
561 local_iface = bridge_get_local_iface(br);
563 int error = netdev_set_etheraddr(local_iface->netdev, ea);
565 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
566 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
567 "Ethernet address: %s",
568 br->name, strerror(error));
572 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
573 ofproto_set_datapath_id(br->ofproto, dpid);
575 /* Set NetFlow configuration on this bridge. */
576 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
577 if (cfg_has("netflow.%s.engine-type", br->name)) {
578 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
581 if (cfg_has("netflow.%s.engine-id", br->name)) {
582 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
584 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
585 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
588 if (add_id_to_iface && engine_id > 0x7f) {
589 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
590 "another vswitch, choose an engine id less than 128",
593 if (add_id_to_iface && br->n_ports > 0x1ff) {
594 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
595 "another port when 512 or more ports are used",
598 svec_init(&nf_hosts);
599 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
600 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
601 engine_id, add_id_to_iface)) {
602 VLOG_ERR("bridge %s: problem setting netflow collectors",
606 /* Update the controller and related settings. It would be more
607 * straightforward to call this from bridge_reconfigure_one(), but we
608 * can't do it there for two reasons. First, and most importantly, at
609 * that point we don't know the dp_ifidx of any interfaces that have
610 * been added to the bridge (because we haven't actually added them to
611 * the datapath). Second, at that point we haven't set the datapath ID
612 * yet; when a controller is configured, resetting the datapath ID will
613 * immediately disconnect from the controller, so it's better to set
614 * the datapath ID before the controller. */
615 bridge_reconfigure_controller(br);
617 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
618 for (i = 0; i < br->n_ports; i++) {
619 struct port *port = br->ports[i];
620 port_update_vlan_compat(port);
621 port_update_bonding(port);
624 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
625 brstp_reconfigure(br);
626 iterate_and_prune_ifaces(br, set_iface_policing, NULL);
631 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
632 struct iface **hw_addr_iface)
634 uint64_t requested_ea;
638 *hw_addr_iface = NULL;
640 /* Did the user request a particular MAC? */
641 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
643 eth_addr_from_uint64(requested_ea, ea);
644 if (eth_addr_is_multicast(ea)) {
645 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
646 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
647 } else if (eth_addr_is_zero(ea)) {
648 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
654 /* Otherwise choose the minimum MAC address among all of the interfaces.
655 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
656 * MAC of the physical interface in such an environment.) */
657 memset(ea, 0xff, sizeof ea);
658 for (i = 0; i < br->n_ports; i++) {
659 struct port *port = br->ports[i];
660 if (port->is_mirror_output_port) {
663 for (j = 0; j < port->n_ifaces; j++) {
664 struct iface *iface = port->ifaces[j];
665 uint8_t iface_ea[ETH_ADDR_LEN];
666 if (iface->dp_ifidx == ODPP_LOCAL
667 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
670 error = netdev_get_etheraddr(iface->netdev, iface_ea);
672 if (!eth_addr_is_multicast(iface_ea) &&
673 !eth_addr_is_reserved(iface_ea) &&
674 !eth_addr_is_zero(iface_ea) &&
675 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
676 memcpy(ea, iface_ea, ETH_ADDR_LEN);
677 *hw_addr_iface = iface;
680 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
681 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
682 iface->name, strerror(error));
686 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
687 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
688 *hw_addr_iface = NULL;
689 VLOG_WARN("bridge %s: using default bridge Ethernet "
690 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
692 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
693 br->name, ETH_ADDR_ARGS(ea));
697 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
698 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
699 * an interface on 'br', then that interface must be passed in as
700 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
701 * 'hw_addr_iface' must be passed in as a null pointer. */
703 bridge_pick_datapath_id(struct bridge *br,
704 const uint8_t bridge_ea[ETH_ADDR_LEN],
705 struct iface *hw_addr_iface)
708 * The procedure for choosing a bridge MAC address will, in the most
709 * ordinary case, also choose a unique MAC that we can use as a datapath
710 * ID. In some special cases, though, multiple bridges will end up with
711 * the same MAC address. This is OK for the bridges, but it will confuse
712 * the OpenFlow controller, because each datapath needs a unique datapath
715 * Datapath IDs must be unique. It is also very desirable that they be
716 * stable from one run to the next, so that policy set on a datapath
721 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
728 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
730 * A bridge whose MAC address is taken from a VLAN network device
731 * (that is, a network device created with vconfig(8) or similar
732 * tool) will have the same MAC address as a bridge on the VLAN
733 * device's physical network device.
735 * Handle this case by hashing the physical network device MAC
736 * along with the VLAN identifier.
738 uint8_t buf[ETH_ADDR_LEN + 2];
739 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
740 buf[ETH_ADDR_LEN] = vlan >> 8;
741 buf[ETH_ADDR_LEN + 1] = vlan;
742 return dpid_from_hash(buf, sizeof buf);
745 * Assume that this bridge's MAC address is unique, since it
746 * doesn't fit any of the cases we handle specially.
751 * A purely internal bridge, that is, one that has no non-virtual
752 * network devices on it at all, is more difficult because it has no
753 * natural unique identifier at all.
755 * When the host is a XenServer, we handle this case by hashing the
756 * host's UUID with the name of the bridge. Names of bridges are
757 * persistent across XenServer reboots, although they can be reused if
758 * an internal network is destroyed and then a new one is later
759 * created, so this is fairly effective.
761 * When the host is not a XenServer, we punt by using a random MAC
762 * address on each run.
764 const char *host_uuid = xenserver_get_host_uuid();
766 char *combined = xasprintf("%s,%s", host_uuid, br->name);
767 dpid = dpid_from_hash(combined, strlen(combined));
773 return eth_addr_to_uint64(bridge_ea);
777 dpid_from_hash(const void *data, size_t n)
779 uint8_t hash[SHA1_DIGEST_SIZE];
781 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
782 sha1_bytes(data, n, hash);
783 eth_addr_mark_random(hash);
784 return eth_addr_to_uint64(hash);
790 struct bridge *br, *next;
794 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
795 int error = bridge_run_one(br);
797 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
798 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
799 "forcing reconfiguration", br->name);
813 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
814 ofproto_wait(br->ofproto);
815 if (br->controller) {
820 mac_learning_wait(br->ml);
827 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
828 * configuration changes. */
830 bridge_flush(struct bridge *br)
832 COVERAGE_INC(bridge_flush);
835 mac_learning_flush(br->ml);
839 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
841 static struct iface *
842 bridge_get_local_iface(struct bridge *br)
846 for (i = 0; i < br->n_ports; i++) {
847 struct port *port = br->ports[i];
848 for (j = 0; j < port->n_ifaces; j++) {
849 struct iface *iface = port->ifaces[j];
850 if (iface->dp_ifidx == ODPP_LOCAL) {
859 /* Bridge unixctl user interface functions. */
861 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
863 struct ds ds = DS_EMPTY_INITIALIZER;
864 const struct bridge *br;
866 br = bridge_lookup(args);
868 unixctl_command_reply(conn, 501, "no such bridge");
872 ds_put_cstr(&ds, " port VLAN MAC Age\n");
874 const struct mac_entry *e;
875 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
876 if (e->port < 0 || e->port >= br->n_ports) {
879 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
880 br->ports[e->port]->ifaces[0]->dp_ifidx,
881 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
884 unixctl_command_reply(conn, 200, ds_cstr(&ds));
888 /* Bridge reconfiguration functions. */
890 static struct bridge *
891 bridge_create(const char *name)
896 assert(!bridge_lookup(name));
897 br = xcalloc(1, sizeof *br);
899 error = dpif_create(name, &br->dpif);
900 if (error == EEXIST || error == EBUSY) {
901 error = dpif_open(name, &br->dpif);
903 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
904 name, strerror(error));
908 dpif_flow_flush(br->dpif);
910 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
915 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
917 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
918 dpif_delete(br->dpif);
919 dpif_close(br->dpif);
924 br->name = xstrdup(name);
925 br->ml = mac_learning_create();
926 br->sent_config_request = false;
927 eth_addr_random(br->default_ea);
929 port_array_init(&br->ifaces);
932 br->bond_next_rebalance = time_msec() + 10000;
934 list_push_back(&all_bridges, &br->node);
936 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
942 bridge_destroy(struct bridge *br)
947 while (br->n_ports > 0) {
948 port_destroy(br->ports[br->n_ports - 1]);
950 list_remove(&br->node);
951 error = dpif_delete(br->dpif);
952 if (error && error != ENOENT) {
953 VLOG_ERR("failed to delete %s: %s",
954 dpif_name(br->dpif), strerror(error));
956 dpif_close(br->dpif);
957 ofproto_destroy(br->ofproto);
958 free(br->controller);
959 mac_learning_destroy(br->ml);
960 port_array_destroy(&br->ifaces);
967 static struct bridge *
968 bridge_lookup(const char *name)
972 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
973 if (!strcmp(br->name, name)) {
981 bridge_exists(const char *name)
983 return bridge_lookup(name) ? true : false;
987 bridge_get_datapathid(const char *name)
989 struct bridge *br = bridge_lookup(name);
990 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
994 bridge_run_one(struct bridge *br)
998 error = ofproto_run1(br->ofproto);
1004 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1009 error = ofproto_run2(br->ofproto, br->flush);
1016 bridge_get_controller(const struct bridge *br)
1018 const char *controller;
1020 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1022 controller = cfg_get_string(0, "mgmt.controller");
1024 return controller && controller[0] ? controller : NULL;
1028 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1030 struct svec *ifaces = ifaces_;
1031 if (!svec_contains(ifaces, iface->name)) {
1032 svec_add(ifaces, iface->name);
1036 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1038 br->name, iface->name, iface->port->name);
1044 bridge_reconfigure_one(struct bridge *br)
1046 struct svec old_ports, new_ports, ifaces;
1047 struct svec listeners, old_listeners;
1048 struct svec snoops, old_snoops;
1051 /* Collect old ports. */
1052 svec_init(&old_ports);
1053 for (i = 0; i < br->n_ports; i++) {
1054 svec_add(&old_ports, br->ports[i]->name);
1056 svec_sort(&old_ports);
1057 assert(svec_is_unique(&old_ports));
1059 /* Collect new ports. */
1060 svec_init(&new_ports);
1061 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1062 svec_sort(&new_ports);
1063 if (bridge_get_controller(br)) {
1064 char local_name[IF_NAMESIZE];
1067 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1068 local_name, sizeof local_name);
1069 if (!error && !svec_contains(&new_ports, local_name)) {
1070 svec_add(&new_ports, local_name);
1071 svec_sort(&new_ports);
1074 if (!svec_is_unique(&new_ports)) {
1075 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1076 br->name, svec_get_duplicate(&new_ports));
1077 svec_unique(&new_ports);
1080 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1082 /* Get rid of deleted ports and add new ports. */
1083 for (i = 0; i < br->n_ports; ) {
1084 struct port *port = br->ports[i];
1085 if (!svec_contains(&new_ports, port->name)) {
1091 for (i = 0; i < new_ports.n; i++) {
1092 const char *name = new_ports.names[i];
1093 if (!svec_contains(&old_ports, name)) {
1094 port_create(br, name);
1097 svec_destroy(&old_ports);
1098 svec_destroy(&new_ports);
1100 /* Reconfigure all ports. */
1101 for (i = 0; i < br->n_ports; i++) {
1102 port_reconfigure(br->ports[i]);
1105 /* Check and delete duplicate interfaces. */
1107 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1108 svec_destroy(&ifaces);
1110 /* Delete all flows if we're switching from connected to standalone or vice
1111 * versa. (XXX Should we delete all flows if we are switching from one
1112 * controller to another?) */
1114 /* Configure OpenFlow management listeners. */
1115 svec_init(&listeners);
1116 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1118 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1119 ovs_rundir, br->name));
1120 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1121 svec_clear(&listeners);
1123 svec_sort_unique(&listeners);
1125 svec_init(&old_listeners);
1126 ofproto_get_listeners(br->ofproto, &old_listeners);
1127 svec_sort_unique(&old_listeners);
1129 if (!svec_equal(&listeners, &old_listeners)) {
1130 ofproto_set_listeners(br->ofproto, &listeners);
1132 svec_destroy(&listeners);
1133 svec_destroy(&old_listeners);
1135 /* Configure OpenFlow controller connection snooping. */
1137 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1139 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1140 ovs_rundir, br->name));
1141 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1142 svec_clear(&snoops);
1144 svec_sort_unique(&snoops);
1146 svec_init(&old_snoops);
1147 ofproto_get_snoops(br->ofproto, &old_snoops);
1148 svec_sort_unique(&old_snoops);
1150 if (!svec_equal(&snoops, &old_snoops)) {
1151 ofproto_set_snoops(br->ofproto, &snoops);
1153 svec_destroy(&snoops);
1154 svec_destroy(&old_snoops);
1156 mirror_reconfigure(br);
1160 bridge_reconfigure_controller(struct bridge *br)
1162 char *pfx = xasprintf("bridge.%s.controller", br->name);
1163 const char *controller;
1165 controller = bridge_get_controller(br);
1166 if ((br->controller != NULL) != (controller != NULL)) {
1167 ofproto_flush_flows(br->ofproto);
1169 free(br->controller);
1170 br->controller = controller ? xstrdup(controller) : NULL;
1173 const char *fail_mode;
1174 int max_backoff, probe;
1175 int rate_limit, burst_limit;
1177 if (!strcmp(controller, "discover")) {
1178 bool update_resolv_conf = true;
1180 if (cfg_has("%s.update-resolv.conf", pfx)) {
1181 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1184 ofproto_set_discovery(br->ofproto, true,
1185 cfg_get_string(0, "%s.accept-regex", pfx),
1186 update_resolv_conf);
1188 struct iface *local_iface;
1191 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1193 || cfg_get_bool(0, "%s.in-band", pfx));
1194 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1195 ofproto_set_in_band(br->ofproto, in_band);
1197 local_iface = bridge_get_local_iface(br);
1199 && cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1200 struct netdev *netdev = local_iface->netdev;
1201 struct in_addr ip, mask, gateway;
1202 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1203 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1204 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1206 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1208 mask.s_addr = guess_netmask(ip.s_addr);
1210 if (!netdev_set_in4(netdev, ip, mask)) {
1211 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1213 br->name, IP_ARGS(&ip.s_addr),
1214 IP_ARGS(&mask.s_addr));
1217 if (gateway.s_addr) {
1218 if (!netdev_add_router(netdev, gateway)) {
1219 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1220 br->name, IP_ARGS(&gateway.s_addr));
1226 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1228 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1230 ofproto_set_failure(br->ofproto,
1232 || !strcmp(fail_mode, "standalone")
1233 || !strcmp(fail_mode, "open")));
1235 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1237 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1242 ofproto_set_probe_interval(br->ofproto, probe);
1244 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1246 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1251 ofproto_set_max_backoff(br->ofproto, max_backoff);
1253 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1255 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1257 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1259 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1261 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1263 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1265 if (cfg_has("%s.commands.acl", pfx)) {
1266 struct svec command_acls;
1269 svec_init(&command_acls);
1270 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1271 command_acl = svec_join(&command_acls, ",", "");
1273 ofproto_set_remote_execution(br->ofproto, command_acl,
1274 cfg_get_string(0, "%s.commands.dir",
1277 svec_destroy(&command_acls);
1280 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1283 union ofp_action action;
1286 /* Set up a flow that matches every packet and directs them to
1287 * OFPP_NORMAL (which goes to us). */
1288 memset(&action, 0, sizeof action);
1289 action.type = htons(OFPAT_OUTPUT);
1290 action.output.len = htons(sizeof action);
1291 action.output.port = htons(OFPP_NORMAL);
1292 memset(&flow, 0, sizeof flow);
1293 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1296 ofproto_set_in_band(br->ofproto, false);
1297 ofproto_set_max_backoff(br->ofproto, 1);
1298 ofproto_set_probe_interval(br->ofproto, 5);
1299 ofproto_set_failure(br->ofproto, false);
1300 ofproto_set_stp(br->ofproto, false);
1304 ofproto_set_controller(br->ofproto, br->controller);
1308 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1313 for (i = 0; i < br->n_ports; i++) {
1314 struct port *port = br->ports[i];
1315 for (j = 0; j < port->n_ifaces; j++) {
1316 struct iface *iface = port->ifaces[j];
1317 svec_add(ifaces, iface->name);
1321 assert(svec_is_unique(ifaces));
1324 /* For robustness, in case the administrator moves around datapath ports behind
1325 * our back, we re-check all the datapath port numbers here.
1327 * This function will set the 'dp_ifidx' members of interfaces that have
1328 * disappeared to -1, so only call this function from a context where those
1329 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1330 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1331 * datapath, which doesn't support UINT16_MAX+1 ports. */
1333 bridge_fetch_dp_ifaces(struct bridge *br)
1335 struct odp_port *dpif_ports;
1336 size_t n_dpif_ports;
1339 /* Reset all interface numbers. */
1340 for (i = 0; i < br->n_ports; i++) {
1341 struct port *port = br->ports[i];
1342 for (j = 0; j < port->n_ifaces; j++) {
1343 struct iface *iface = port->ifaces[j];
1344 iface->dp_ifidx = -1;
1347 port_array_clear(&br->ifaces);
1349 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1350 for (i = 0; i < n_dpif_ports; i++) {
1351 struct odp_port *p = &dpif_ports[i];
1352 struct iface *iface = iface_lookup(br, p->devname);
1354 if (iface->dp_ifidx >= 0) {
1355 VLOG_WARN("%s reported interface %s twice",
1356 dpif_name(br->dpif), p->devname);
1357 } else if (iface_from_dp_ifidx(br, p->port)) {
1358 VLOG_WARN("%s reported interface %"PRIu16" twice",
1359 dpif_name(br->dpif), p->port);
1361 port_array_set(&br->ifaces, p->port, iface);
1362 iface->dp_ifidx = p->port;
1369 /* Bridge packet processing functions. */
1372 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1374 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1377 static struct bond_entry *
1378 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1380 return &port->bond_hash[bond_hash(mac)];
1384 bond_choose_iface(const struct port *port)
1387 for (i = 0; i < port->n_ifaces; i++) {
1388 if (port->ifaces[i]->enabled) {
1396 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1397 uint16_t *dp_ifidx, tag_type *tags)
1399 struct iface *iface;
1401 assert(port->n_ifaces);
1402 if (port->n_ifaces == 1) {
1403 iface = port->ifaces[0];
1405 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1406 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1407 || !port->ifaces[e->iface_idx]->enabled) {
1408 /* XXX select interface properly. The current interface selection
1409 * is only good for testing the rebalancing code. */
1410 e->iface_idx = bond_choose_iface(port);
1411 if (e->iface_idx < 0) {
1412 *tags |= port->no_ifaces_tag;
1415 e->iface_tag = tag_create_random();
1417 *tags |= e->iface_tag;
1418 iface = port->ifaces[e->iface_idx];
1420 *dp_ifidx = iface->dp_ifidx;
1421 *tags |= iface->tag; /* Currently only used for bonding. */
1426 bond_link_status_update(struct iface *iface, bool carrier)
1428 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1429 struct port *port = iface->port;
1431 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1432 /* Nothing to do. */
1435 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1436 iface->name, carrier ? "detected" : "dropped");
1437 if (carrier == iface->enabled) {
1438 iface->delay_expires = LLONG_MAX;
1439 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1440 iface->name, carrier ? "disabled" : "enabled");
1441 } else if (carrier && port->updelay && port->active_iface < 0) {
1442 iface->delay_expires = time_msec();
1443 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1444 "other interface is up", iface->name, port->updelay);
1446 int delay = carrier ? port->updelay : port->downdelay;
1447 iface->delay_expires = time_msec() + delay;
1450 "interface %s: will be %s if it stays %s for %d ms",
1452 carrier ? "enabled" : "disabled",
1453 carrier ? "up" : "down",
1460 bond_choose_active_iface(struct port *port)
1462 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1464 port->active_iface = bond_choose_iface(port);
1465 port->active_iface_tag = tag_create_random();
1466 if (port->active_iface >= 0) {
1467 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1468 port->name, port->ifaces[port->active_iface]->name);
1470 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1476 bond_enable_slave(struct iface *iface, bool enable)
1478 struct port *port = iface->port;
1479 struct bridge *br = port->bridge;
1481 iface->delay_expires = LLONG_MAX;
1482 if (enable == iface->enabled) {
1486 iface->enabled = enable;
1487 if (!iface->enabled) {
1488 VLOG_WARN("interface %s: disabled", iface->name);
1489 ofproto_revalidate(br->ofproto, iface->tag);
1490 if (iface->port_ifidx == port->active_iface) {
1491 ofproto_revalidate(br->ofproto,
1492 port->active_iface_tag);
1493 bond_choose_active_iface(port);
1495 bond_send_learning_packets(port);
1497 VLOG_WARN("interface %s: enabled", iface->name);
1498 if (port->active_iface < 0) {
1499 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1500 bond_choose_active_iface(port);
1501 bond_send_learning_packets(port);
1503 iface->tag = tag_create_random();
1508 bond_run(struct bridge *br)
1512 for (i = 0; i < br->n_ports; i++) {
1513 struct port *port = br->ports[i];
1514 if (port->n_ifaces < 2) {
1517 for (j = 0; j < port->n_ifaces; j++) {
1518 struct iface *iface = port->ifaces[j];
1519 if (time_msec() >= iface->delay_expires) {
1520 bond_enable_slave(iface, !iface->enabled);
1527 bond_wait(struct bridge *br)
1531 for (i = 0; i < br->n_ports; i++) {
1532 struct port *port = br->ports[i];
1533 if (port->n_ifaces < 2) {
1536 for (j = 0; j < port->n_ifaces; j++) {
1537 struct iface *iface = port->ifaces[j];
1538 if (iface->delay_expires != LLONG_MAX) {
1539 poll_timer_wait(iface->delay_expires - time_msec());
1546 set_dst(struct dst *p, const flow_t *flow,
1547 const struct port *in_port, const struct port *out_port,
1552 * XXX This uses too many tags: any broadcast flow will get one tag per
1553 * destination port, and thus a broadcast on a switch of any size is likely
1554 * to have all tag bits set. We should figure out a way to be smarter.
1556 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1557 *tags |= out_port->stp_state_tag;
1558 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1562 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1563 : in_port->vlan >= 0 ? in_port->vlan
1564 : ntohs(flow->dl_vlan));
1565 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1569 swap_dst(struct dst *p, struct dst *q)
1571 struct dst tmp = *p;
1576 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1577 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1578 * that we push to the datapath. We could in fact fully sort the array by
1579 * vlan, but in most cases there are at most two different vlan tags so that's
1580 * possibly overkill.) */
1582 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1584 struct dst *first = dsts;
1585 struct dst *last = dsts + n_dsts;
1587 while (first != last) {
1589 * - All dsts < first have vlan == 'vlan'.
1590 * - All dsts >= last have vlan != 'vlan'.
1591 * - first < last. */
1592 while (first->vlan == vlan) {
1593 if (++first == last) {
1598 /* Same invariants, plus one additional:
1599 * - first->vlan != vlan.
1601 while (last[-1].vlan != vlan) {
1602 if (--last == first) {
1607 /* Same invariants, plus one additional:
1608 * - last[-1].vlan == vlan.*/
1609 swap_dst(first++, --last);
1614 mirror_mask_ffs(mirror_mask_t mask)
1616 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1621 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1622 const struct dst *test)
1625 for (i = 0; i < n_dsts; i++) {
1626 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1634 port_trunks_vlan(const struct port *port, uint16_t vlan)
1636 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1640 port_includes_vlan(const struct port *port, uint16_t vlan)
1642 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1646 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1647 const struct port *in_port, const struct port *out_port,
1648 struct dst dsts[], tag_type *tags)
1650 mirror_mask_t mirrors = in_port->src_mirrors;
1651 struct dst *dst = dsts;
1654 *tags |= in_port->stp_state_tag;
1655 if (out_port == FLOOD_PORT) {
1656 /* XXX use ODP_FLOOD if no vlans or bonding. */
1657 /* XXX even better, define each VLAN as a datapath port group */
1658 for (i = 0; i < br->n_ports; i++) {
1659 struct port *port = br->ports[i];
1660 if (port != in_port && port_includes_vlan(port, vlan)
1661 && !port->is_mirror_output_port
1662 && set_dst(dst, flow, in_port, port, tags)) {
1663 mirrors |= port->dst_mirrors;
1667 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1668 mirrors |= out_port->dst_mirrors;
1673 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1674 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1676 if (set_dst(dst, flow, in_port, m->out_port, tags)
1677 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1681 for (i = 0; i < br->n_ports; i++) {
1682 struct port *port = br->ports[i];
1683 if (port_includes_vlan(port, m->out_vlan)
1684 && set_dst(dst, flow, in_port, port, tags)
1685 && !dst_is_duplicate(dsts, dst - dsts, dst))
1687 if (port->vlan < 0) {
1688 dst->vlan = m->out_vlan;
1690 if (dst->dp_ifidx == flow->in_port
1691 && dst->vlan == vlan) {
1692 /* Don't send out input port on same VLAN. */
1700 mirrors &= mirrors - 1;
1703 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1708 print_dsts(const struct dst *dsts, size_t n)
1710 for (; n--; dsts++) {
1711 printf(">p%"PRIu16, dsts->dp_ifidx);
1712 if (dsts->vlan != OFP_VLAN_NONE) {
1713 printf("v%"PRIu16, dsts->vlan);
1719 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1720 const struct port *in_port, const struct port *out_port,
1721 tag_type *tags, struct odp_actions *actions)
1723 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1725 const struct dst *p;
1728 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1730 cur_vlan = ntohs(flow->dl_vlan);
1731 for (p = dsts; p < &dsts[n_dsts]; p++) {
1732 union odp_action *a;
1733 if (p->vlan != cur_vlan) {
1734 if (p->vlan == OFP_VLAN_NONE) {
1735 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1737 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1738 a->vlan_vid.vlan_vid = htons(p->vlan);
1742 a = odp_actions_add(actions, ODPAT_OUTPUT);
1743 a->output.port = p->dp_ifidx;
1748 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1750 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1751 return (flow->dl_type == htons(ETH_TYPE_ARP)
1752 && eth_addr_is_broadcast(flow->dl_dst)
1753 && packet->size >= sizeof(struct arp_eth_header)
1754 && arp->ar_op == ARP_OP_REQUEST);
1757 /* If the composed actions may be applied to any packet in the given 'flow',
1758 * returns true. Otherwise, the actions should only be applied to 'packet', or
1759 * not at all, if 'packet' was NULL. */
1761 process_flow(struct bridge *br, const flow_t *flow,
1762 const struct ofpbuf *packet, struct odp_actions *actions,
1765 struct iface *in_iface;
1766 struct port *in_port;
1767 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1770 /* Find the interface and port structure for the received packet. */
1771 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1773 /* No interface? Something fishy... */
1774 if (packet != NULL) {
1775 /* Odd. A few possible reasons here:
1777 * - We deleted an interface but there are still a few packets
1778 * queued up from it.
1780 * - Someone externally added an interface (e.g. with "ovs-dpctl
1781 * add-if") that we don't know about.
1783 * - Packet arrived on the local port but the local port is not
1784 * one of our bridge ports.
1786 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1788 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1789 "interface %"PRIu16, br->name, flow->in_port);
1792 /* Return without adding any actions, to drop packets on this flow. */
1795 in_port = in_iface->port;
1797 /* Figure out what VLAN this packet belongs to.
1799 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1800 * belongs to VLAN 0, so we should treat both cases identically. (In the
1801 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1802 * presumably to allow a priority to be specified. In the latter case, the
1803 * packet does not have any 802.1Q header.) */
1804 vlan = ntohs(flow->dl_vlan);
1805 if (vlan == OFP_VLAN_NONE) {
1808 if (in_port->vlan >= 0) {
1810 /* XXX support double tagging? */
1811 if (packet != NULL) {
1812 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1813 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1814 "packet received on port %s configured with "
1815 "implicit VLAN %"PRIu16,
1816 br->name, ntohs(flow->dl_vlan),
1817 in_port->name, in_port->vlan);
1821 vlan = in_port->vlan;
1823 if (!port_includes_vlan(in_port, vlan)) {
1824 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1825 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1826 "packet received on port %s not configured for "
1828 br->name, vlan, in_port->name, vlan);
1833 /* Drop frames for ports that STP wants entirely killed (both for
1834 * forwarding and for learning). Later, after we do learning, we'll drop
1835 * the frames that STP wants to do learning but not forwarding on. */
1836 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1840 /* Drop frames for reserved multicast addresses. */
1841 if (eth_addr_is_reserved(flow->dl_dst)) {
1845 /* Drop frames on ports reserved for mirroring. */
1846 if (in_port->is_mirror_output_port) {
1847 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1848 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1849 "which is reserved exclusively for mirroring",
1850 br->name, in_port->name);
1854 /* Multicast (and broadcast) packets on bonds need special attention, to
1855 * avoid receiving duplicates. */
1856 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1857 *tags |= in_port->active_iface_tag;
1858 if (in_port->active_iface != in_iface->port_ifidx) {
1859 /* Drop all multicast packets on inactive slaves. */
1862 /* Drop all multicast packets for which we have learned a different
1863 * input port, because we probably sent the packet on one slaves
1864 * and got it back on the active slave. Broadcast ARP replies are
1865 * an exception to this rule: the host has moved to another
1867 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1868 if (src_idx != -1 && src_idx != in_port->port_idx) {
1870 if (!is_bcast_arp_reply(flow, packet)) {
1874 /* No way to know whether it's an ARP reply, because the
1875 * flow entry doesn't include enough information and we
1876 * don't have a packet. Punt. */
1884 out_port = FLOOD_PORT;
1888 /* Learn source MAC (but don't try to learn from revalidation). */
1890 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1891 vlan, in_port->port_idx);
1893 /* The log messages here could actually be useful in debugging,
1894 * so keep the rate limit relatively high. */
1895 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1897 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1898 "on port %s in VLAN %d",
1899 br->name, ETH_ADDR_ARGS(flow->dl_src),
1900 in_port->name, vlan);
1901 ofproto_revalidate(br->ofproto, rev_tag);
1905 /* Determine output port. */
1906 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1908 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1909 out_port = br->ports[out_port_idx];
1913 /* Don't send packets out their input ports. Don't forward frames that STP
1914 * wants us to discard. */
1915 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1920 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1923 * We send out only a single packet, instead of setting up a flow, if the
1924 * packet is an ARP directed to broadcast that arrived on a bonded
1925 * interface. In such a situation ARP requests and replies must be handled
1926 * differently, but OpenFlow unfortunately can't distinguish them.
1928 return (in_port->n_ifaces < 2
1929 || flow->dl_type != htons(ETH_TYPE_ARP)
1930 || !eth_addr_is_broadcast(flow->dl_dst));
1933 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1936 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1937 const struct ofp_phy_port *opp,
1940 struct bridge *br = br_;
1941 struct iface *iface;
1944 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1950 if (reason == OFPPR_DELETE) {
1951 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1952 br->name, iface->name);
1953 iface_destroy(iface);
1954 if (!port->n_ifaces) {
1955 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1956 br->name, port->name);
1962 if (port->n_ifaces > 1) {
1963 bool up = !(opp->state & OFPPS_LINK_DOWN);
1964 bond_link_status_update(iface, up);
1965 port_update_bond_compat(port);
1971 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1972 struct odp_actions *actions, tag_type *tags, void *br_)
1974 struct bridge *br = br_;
1977 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1978 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1979 brstp_receive(br, flow, payload);
1984 COVERAGE_INC(bridge_process_flow);
1985 return process_flow(br, flow, packet, actions, tags);
1989 bridge_account_flow_ofhook_cb(const flow_t *flow,
1990 const union odp_action *actions,
1991 size_t n_actions, unsigned long long int n_bytes,
1994 struct bridge *br = br_;
1995 const union odp_action *a;
1997 if (!br->has_bonded_ports) {
2001 for (a = actions; a < &actions[n_actions]; a++) {
2002 if (a->type == ODPAT_OUTPUT) {
2003 struct port *port = port_from_dp_ifidx(br, a->output.port);
2004 if (port && port->n_ifaces >= 2) {
2005 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2006 e->tx_bytes += n_bytes;
2013 bridge_account_checkpoint_ofhook_cb(void *br_)
2015 struct bridge *br = br_;
2018 if (!br->has_bonded_ports) {
2022 /* The current ofproto implementation calls this callback at least once a
2023 * second, so this timer implementation is sufficient. */
2024 if (time_msec() < br->bond_next_rebalance) {
2027 br->bond_next_rebalance = time_msec() + 10000;
2029 for (i = 0; i < br->n_ports; i++) {
2030 struct port *port = br->ports[i];
2031 if (port->n_ifaces > 1) {
2032 bond_rebalance_port(port);
2037 static struct ofhooks bridge_ofhooks = {
2038 bridge_port_changed_ofhook_cb,
2039 bridge_normal_ofhook_cb,
2040 bridge_account_flow_ofhook_cb,
2041 bridge_account_checkpoint_ofhook_cb,
2044 /* Bonding functions. */
2046 /* Statistics for a single interface on a bonded port, used for load-based
2047 * bond rebalancing. */
2048 struct slave_balance {
2049 struct iface *iface; /* The interface. */
2050 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2052 /* All the "bond_entry"s that are assigned to this interface, in order of
2053 * increasing tx_bytes. */
2054 struct bond_entry **hashes;
2058 /* Sorts pointers to pointers to bond_entries in ascending order by the
2059 * interface to which they are assigned, and within a single interface in
2060 * ascending order of bytes transmitted. */
2062 compare_bond_entries(const void *a_, const void *b_)
2064 const struct bond_entry *const *ap = a_;
2065 const struct bond_entry *const *bp = b_;
2066 const struct bond_entry *a = *ap;
2067 const struct bond_entry *b = *bp;
2068 if (a->iface_idx != b->iface_idx) {
2069 return a->iface_idx > b->iface_idx ? 1 : -1;
2070 } else if (a->tx_bytes != b->tx_bytes) {
2071 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2077 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2078 * *descending* order by number of bytes transmitted. */
2080 compare_slave_balance(const void *a_, const void *b_)
2082 const struct slave_balance *a = a_;
2083 const struct slave_balance *b = b_;
2084 if (a->iface->enabled != b->iface->enabled) {
2085 return a->iface->enabled ? -1 : 1;
2086 } else if (a->tx_bytes != b->tx_bytes) {
2087 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2094 swap_bals(struct slave_balance *a, struct slave_balance *b)
2096 struct slave_balance tmp = *a;
2101 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2102 * given that 'p' (and only 'p') might be in the wrong location.
2104 * This function invalidates 'p', since it might now be in a different memory
2107 resort_bals(struct slave_balance *p,
2108 struct slave_balance bals[], size_t n_bals)
2111 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2112 swap_bals(p, p - 1);
2114 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2115 swap_bals(p, p + 1);
2121 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2123 if (VLOG_IS_DBG_ENABLED()) {
2124 struct ds ds = DS_EMPTY_INITIALIZER;
2125 const struct slave_balance *b;
2127 for (b = bals; b < bals + n_bals; b++) {
2131 ds_put_char(&ds, ',');
2133 ds_put_format(&ds, " %s %"PRIu64"kB",
2134 b->iface->name, b->tx_bytes / 1024);
2136 if (!b->iface->enabled) {
2137 ds_put_cstr(&ds, " (disabled)");
2139 if (b->n_hashes > 0) {
2140 ds_put_cstr(&ds, " (");
2141 for (i = 0; i < b->n_hashes; i++) {
2142 const struct bond_entry *e = b->hashes[i];
2144 ds_put_cstr(&ds, " + ");
2146 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2147 e - port->bond_hash, e->tx_bytes / 1024);
2149 ds_put_cstr(&ds, ")");
2152 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2157 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2159 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2160 struct bond_entry *hash)
2162 struct port *port = from->iface->port;
2163 uint64_t delta = hash->tx_bytes;
2165 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2166 "from %s to %s (now carrying %"PRIu64"kB and "
2167 "%"PRIu64"kB load, respectively)",
2168 port->name, delta / 1024, hash - port->bond_hash,
2169 from->iface->name, to->iface->name,
2170 (from->tx_bytes - delta) / 1024,
2171 (to->tx_bytes + delta) / 1024);
2173 /* Delete element from from->hashes.
2175 * We don't bother to add the element to to->hashes because not only would
2176 * it require more work, the only purpose it would be to allow that hash to
2177 * be migrated to another slave in this rebalancing run, and there is no
2178 * point in doing that. */
2179 if (from->hashes[0] == hash) {
2182 int i = hash - from->hashes[0];
2183 memmove(from->hashes + i, from->hashes + i + 1,
2184 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2188 /* Shift load away from 'from' to 'to'. */
2189 from->tx_bytes -= delta;
2190 to->tx_bytes += delta;
2192 /* Arrange for flows to be revalidated. */
2193 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2194 hash->iface_idx = to->iface->port_ifidx;
2195 hash->iface_tag = tag_create_random();
2199 bond_rebalance_port(struct port *port)
2201 struct slave_balance bals[DP_MAX_PORTS];
2203 struct bond_entry *hashes[BOND_MASK + 1];
2204 struct slave_balance *b, *from, *to;
2205 struct bond_entry *e;
2208 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2209 * descending order of tx_bytes, so that bals[0] represents the most
2210 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2213 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2214 * array for each slave_balance structure, we sort our local array of
2215 * hashes in order by slave, so that all of the hashes for a given slave
2216 * become contiguous in memory, and then we point each 'hashes' members of
2217 * a slave_balance structure to the start of a contiguous group. */
2218 n_bals = port->n_ifaces;
2219 for (b = bals; b < &bals[n_bals]; b++) {
2220 b->iface = port->ifaces[b - bals];
2225 for (i = 0; i <= BOND_MASK; i++) {
2226 hashes[i] = &port->bond_hash[i];
2228 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2229 for (i = 0; i <= BOND_MASK; i++) {
2231 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2232 b = &bals[e->iface_idx];
2233 b->tx_bytes += e->tx_bytes;
2235 b->hashes = &hashes[i];
2240 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2241 log_bals(bals, n_bals, port);
2243 /* Discard slaves that aren't enabled (which were sorted to the back of the
2244 * array earlier). */
2245 while (!bals[n_bals - 1].iface->enabled) {
2252 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2253 to = &bals[n_bals - 1];
2254 for (from = bals; from < to; ) {
2255 uint64_t overload = from->tx_bytes - to->tx_bytes;
2256 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2257 /* The extra load on 'from' (and all less-loaded slaves), compared
2258 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2259 * it is less than ~1Mbps. No point in rebalancing. */
2261 } else if (from->n_hashes == 1) {
2262 /* 'from' only carries a single MAC hash, so we can't shift any
2263 * load away from it, even though we want to. */
2266 /* 'from' is carrying significantly more load than 'to', and that
2267 * load is split across at least two different hashes. Pick a hash
2268 * to migrate to 'to' (the least-loaded slave), given that doing so
2269 * must not cause 'to''s load to exceed 'from''s load.
2271 * The sort order we use means that we prefer to shift away the
2272 * smallest hashes instead of the biggest ones. There is little
2273 * reason behind this decision; we could use the opposite sort
2274 * order to shift away big hashes ahead of small ones. */
2277 for (i = 0; i < from->n_hashes; i++) {
2278 uint64_t delta = from->hashes[i]->tx_bytes;
2279 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2283 if (i < from->n_hashes) {
2284 bond_shift_load(from, to, from->hashes[i]);
2286 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2287 * point to different slave_balance structures. It is only
2288 * valid to do these two operations in a row at all because we
2289 * know that 'from' will not move past 'to' and vice versa. */
2290 resort_bals(from, bals, n_bals);
2291 resort_bals(to, bals, n_bals);
2298 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2299 * historical data to decay to <1% in 7 rebalancing runs. */
2300 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2306 bond_send_learning_packets(struct port *port)
2308 struct bridge *br = port->bridge;
2309 struct mac_entry *e;
2310 struct ofpbuf packet;
2311 int error, n_packets, n_errors;
2313 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2317 ofpbuf_init(&packet, 128);
2318 error = n_packets = n_errors = 0;
2319 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2320 static const char s[] = "Open vSwitch Bond Failover";
2321 union ofp_action actions[2], *a;
2322 struct eth_header *eth;
2323 struct llc_snap_header *llc_snap;
2329 if (e->port == port->port_idx
2330 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2334 /* Compose packet to send. */
2335 ofpbuf_clear(&packet);
2336 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2337 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2338 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2339 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2341 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2342 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2343 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2345 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2346 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2347 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2348 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2349 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2351 /* Compose actions. */
2352 memset(actions, 0, sizeof actions);
2355 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2356 a->vlan_vid.len = htons(sizeof *a);
2357 a->vlan_vid.vlan_vid = htons(e->vlan);
2360 a->output.type = htons(OFPAT_OUTPUT);
2361 a->output.len = htons(sizeof *a);
2362 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2367 flow_extract(&packet, ODPP_NONE, &flow);
2368 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2375 ofpbuf_uninit(&packet);
2378 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2379 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2380 "packets, last error was: %s",
2381 port->name, n_errors, n_packets, strerror(error));
2383 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2384 port->name, n_packets);
2388 /* Bonding unixctl user interface functions. */
2391 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2393 struct ds ds = DS_EMPTY_INITIALIZER;
2394 const struct bridge *br;
2396 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2398 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2401 for (i = 0; i < br->n_ports; i++) {
2402 const struct port *port = br->ports[i];
2403 if (port->n_ifaces > 1) {
2406 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2407 for (j = 0; j < port->n_ifaces; j++) {
2408 const struct iface *iface = port->ifaces[j];
2410 ds_put_cstr(&ds, ", ");
2412 ds_put_cstr(&ds, iface->name);
2414 ds_put_char(&ds, '\n');
2418 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2422 static struct port *
2423 bond_find(const char *name)
2425 const struct bridge *br;
2427 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2430 for (i = 0; i < br->n_ports; i++) {
2431 struct port *port = br->ports[i];
2432 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2441 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2443 struct ds ds = DS_EMPTY_INITIALIZER;
2444 const struct port *port;
2447 port = bond_find(args);
2449 unixctl_command_reply(conn, 501, "no such bond");
2453 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2454 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2455 ds_put_format(&ds, "next rebalance: %lld ms\n",
2456 port->bridge->bond_next_rebalance - time_msec());
2457 for (j = 0; j < port->n_ifaces; j++) {
2458 const struct iface *iface = port->ifaces[j];
2459 struct bond_entry *be;
2462 ds_put_format(&ds, "slave %s: %s\n",
2463 iface->name, iface->enabled ? "enabled" : "disabled");
2464 if (j == port->active_iface) {
2465 ds_put_cstr(&ds, "\tactive slave\n");
2467 if (iface->delay_expires != LLONG_MAX) {
2468 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2469 iface->enabled ? "downdelay" : "updelay",
2470 iface->delay_expires - time_msec());
2474 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2475 int hash = be - port->bond_hash;
2476 struct mac_entry *me;
2478 if (be->iface_idx != j) {
2482 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2483 hash, be->tx_bytes / 1024);
2486 if (!port->bridge->ml) {
2490 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2491 &port->bridge->ml->lrus) {
2494 if (bond_hash(me->mac) == hash
2495 && me->port != port->port_idx
2496 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2497 && dp_ifidx == iface->dp_ifidx)
2499 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2500 ETH_ADDR_ARGS(me->mac));
2505 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2510 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2512 char *args = (char *) args_;
2513 char *save_ptr = NULL;
2514 char *bond_s, *hash_s, *slave_s;
2515 uint8_t mac[ETH_ADDR_LEN];
2517 struct iface *iface;
2518 struct bond_entry *entry;
2521 bond_s = strtok_r(args, " ", &save_ptr);
2522 hash_s = strtok_r(NULL, " ", &save_ptr);
2523 slave_s = strtok_r(NULL, " ", &save_ptr);
2525 unixctl_command_reply(conn, 501,
2526 "usage: bond/migrate BOND HASH SLAVE");
2530 port = bond_find(bond_s);
2532 unixctl_command_reply(conn, 501, "no such bond");
2536 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2537 == ETH_ADDR_SCAN_COUNT) {
2538 hash = bond_hash(mac);
2539 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2540 hash = atoi(hash_s) & BOND_MASK;
2542 unixctl_command_reply(conn, 501, "bad hash");
2546 iface = port_lookup_iface(port, slave_s);
2548 unixctl_command_reply(conn, 501, "no such slave");
2552 if (!iface->enabled) {
2553 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2557 entry = &port->bond_hash[hash];
2558 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2559 entry->iface_idx = iface->port_ifidx;
2560 entry->iface_tag = tag_create_random();
2561 unixctl_command_reply(conn, 200, "migrated");
2565 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2567 char *args = (char *) args_;
2568 char *save_ptr = NULL;
2569 char *bond_s, *slave_s;
2571 struct iface *iface;
2573 bond_s = strtok_r(args, " ", &save_ptr);
2574 slave_s = strtok_r(NULL, " ", &save_ptr);
2576 unixctl_command_reply(conn, 501,
2577 "usage: bond/set-active-slave BOND SLAVE");
2581 port = bond_find(bond_s);
2583 unixctl_command_reply(conn, 501, "no such bond");
2587 iface = port_lookup_iface(port, slave_s);
2589 unixctl_command_reply(conn, 501, "no such slave");
2593 if (!iface->enabled) {
2594 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2598 if (port->active_iface != iface->port_ifidx) {
2599 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2600 port->active_iface = iface->port_ifidx;
2601 port->active_iface_tag = tag_create_random();
2602 VLOG_INFO("port %s: active interface is now %s",
2603 port->name, iface->name);
2604 bond_send_learning_packets(port);
2605 unixctl_command_reply(conn, 200, "done");
2607 unixctl_command_reply(conn, 200, "no change");
2612 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2614 char *args = (char *) args_;
2615 char *save_ptr = NULL;
2616 char *bond_s, *slave_s;
2618 struct iface *iface;
2620 bond_s = strtok_r(args, " ", &save_ptr);
2621 slave_s = strtok_r(NULL, " ", &save_ptr);
2623 unixctl_command_reply(conn, 501,
2624 "usage: bond/enable/disable-slave BOND SLAVE");
2628 port = bond_find(bond_s);
2630 unixctl_command_reply(conn, 501, "no such bond");
2634 iface = port_lookup_iface(port, slave_s);
2636 unixctl_command_reply(conn, 501, "no such slave");
2640 bond_enable_slave(iface, enable);
2641 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2645 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2647 enable_slave(conn, args, true);
2651 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2653 enable_slave(conn, args, false);
2659 unixctl_command_register("bond/list", bond_unixctl_list);
2660 unixctl_command_register("bond/show", bond_unixctl_show);
2661 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2662 unixctl_command_register("bond/set-active-slave",
2663 bond_unixctl_set_active_slave);
2664 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2665 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2668 /* Port functions. */
2671 port_create(struct bridge *br, const char *name)
2675 port = xcalloc(1, sizeof *port);
2677 port->port_idx = br->n_ports;
2679 port->trunks = NULL;
2680 port->name = xstrdup(name);
2681 port->active_iface = -1;
2682 port->stp_state = STP_DISABLED;
2683 port->stp_state_tag = 0;
2685 if (br->n_ports >= br->allocated_ports) {
2686 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2689 br->ports[br->n_ports++] = port;
2691 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2696 port_reconfigure(struct port *port)
2698 bool bonded = cfg_has_section("bonding.%s", port->name);
2699 struct svec old_ifaces, new_ifaces;
2700 unsigned long *trunks;
2704 /* Collect old and new interfaces. */
2705 svec_init(&old_ifaces);
2706 svec_init(&new_ifaces);
2707 for (i = 0; i < port->n_ifaces; i++) {
2708 svec_add(&old_ifaces, port->ifaces[i]->name);
2710 svec_sort(&old_ifaces);
2712 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2713 if (!new_ifaces.n) {
2714 VLOG_ERR("port %s: no interfaces specified for bonded port",
2716 } else if (new_ifaces.n == 1) {
2717 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2721 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2722 if (port->updelay < 0) {
2725 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2726 if (port->downdelay < 0) {
2727 port->downdelay = 0;
2730 svec_init(&new_ifaces);
2731 svec_add(&new_ifaces, port->name);
2734 /* Get rid of deleted interfaces and add new interfaces. */
2735 for (i = 0; i < port->n_ifaces; i++) {
2736 struct iface *iface = port->ifaces[i];
2737 if (!svec_contains(&new_ifaces, iface->name)) {
2738 iface_destroy(iface);
2743 for (i = 0; i < new_ifaces.n; i++) {
2744 const char *name = new_ifaces.names[i];
2745 if (!svec_contains(&old_ifaces, name)) {
2746 iface_create(port, name);
2752 if (cfg_has("vlan.%s.tag", port->name)) {
2754 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2755 if (vlan >= 0 && vlan <= 4095) {
2756 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2759 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2760 * they even work as-is. But they have not been tested. */
2761 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2765 if (port->vlan != vlan) {
2767 bridge_flush(port->bridge);
2770 /* Get trunked VLANs. */
2773 size_t n_trunks, n_errors;
2776 trunks = bitmap_allocate(4096);
2777 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2779 for (i = 0; i < n_trunks; i++) {
2780 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2782 bitmap_set1(trunks, trunk);
2788 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2789 port->name, n_trunks);
2791 if (n_errors == n_trunks) {
2793 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2796 bitmap_set_multiple(trunks, 0, 4096, 1);
2799 if (cfg_has("vlan.%s.trunks", port->name)) {
2800 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2801 port->name, port->name);
2805 ? port->trunks != NULL
2806 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2807 bridge_flush(port->bridge);
2809 bitmap_free(port->trunks);
2810 port->trunks = trunks;
2812 svec_destroy(&old_ifaces);
2813 svec_destroy(&new_ifaces);
2817 port_destroy(struct port *port)
2820 struct bridge *br = port->bridge;
2824 proc_net_compat_update_vlan(port->name, NULL, 0);
2826 for (i = 0; i < MAX_MIRRORS; i++) {
2827 struct mirror *m = br->mirrors[i];
2828 if (m && m->out_port == port) {
2833 while (port->n_ifaces > 0) {
2834 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2837 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2838 del->port_idx = port->port_idx;
2841 bitmap_free(port->trunks);
2848 static struct port *
2849 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2851 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2852 return iface ? iface->port : NULL;
2855 static struct port *
2856 port_lookup(const struct bridge *br, const char *name)
2860 for (i = 0; i < br->n_ports; i++) {
2861 struct port *port = br->ports[i];
2862 if (!strcmp(port->name, name)) {
2869 static struct iface *
2870 port_lookup_iface(const struct port *port, const char *name)
2874 for (j = 0; j < port->n_ifaces; j++) {
2875 struct iface *iface = port->ifaces[j];
2876 if (!strcmp(iface->name, name)) {
2884 port_update_bonding(struct port *port)
2886 if (port->n_ifaces < 2) {
2887 /* Not a bonded port. */
2888 if (port->bond_hash) {
2889 free(port->bond_hash);
2890 port->bond_hash = NULL;
2891 proc_net_compat_update_bond(port->name, NULL);
2894 if (!port->bond_hash) {
2897 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2898 for (i = 0; i <= BOND_MASK; i++) {
2899 struct bond_entry *e = &port->bond_hash[i];
2903 port->no_ifaces_tag = tag_create_random();
2904 bond_choose_active_iface(port);
2906 port_update_bond_compat(port);
2911 port_update_bond_compat(struct port *port)
2913 struct compat_bond bond;
2916 if (port->n_ifaces < 2) {
2921 bond.updelay = port->updelay;
2922 bond.downdelay = port->downdelay;
2923 bond.n_slaves = port->n_ifaces;
2924 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2925 for (i = 0; i < port->n_ifaces; i++) {
2926 struct iface *iface = port->ifaces[i];
2927 struct compat_bond_slave *slave = &bond.slaves[i];
2928 slave->name = iface->name;
2929 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2930 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2934 netdev_get_etheraddr(iface->netdev, slave->mac);
2936 proc_net_compat_update_bond(port->name, &bond);
2941 port_update_vlan_compat(struct port *port)
2943 struct bridge *br = port->bridge;
2944 char *vlandev_name = NULL;
2946 if (port->vlan > 0) {
2947 /* Figure out the name that the VLAN device should actually have, if it
2948 * existed. This takes some work because the VLAN device would not
2949 * have port->name in its name; rather, it would have the trunk port's
2950 * name, and 'port' would be attached to a bridge that also had the
2951 * VLAN device one of its ports. So we need to find a trunk port that
2952 * includes port->vlan.
2954 * There might be more than one candidate. This doesn't happen on
2955 * XenServer, so if it happens we just pick the first choice in
2956 * alphabetical order instead of creating multiple VLAN devices. */
2958 for (i = 0; i < br->n_ports; i++) {
2959 struct port *p = br->ports[i];
2960 if (port_trunks_vlan(p, port->vlan)
2962 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2964 uint8_t ea[ETH_ADDR_LEN];
2965 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
2966 if (!eth_addr_is_multicast(ea) &&
2967 !eth_addr_is_reserved(ea) &&
2968 !eth_addr_is_zero(ea)) {
2969 vlandev_name = p->name;
2974 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2977 /* Interface functions. */
2980 iface_create(struct port *port, const char *name)
2982 struct iface *iface;
2984 iface = xcalloc(1, sizeof *iface);
2986 iface->port_ifidx = port->n_ifaces;
2987 iface->name = xstrdup(name);
2988 iface->dp_ifidx = -1;
2989 iface->tag = tag_create_random();
2990 iface->delay_expires = LLONG_MAX;
2991 iface->netdev = NULL;
2993 if (port->n_ifaces >= port->allocated_ifaces) {
2994 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2995 sizeof *port->ifaces);
2997 port->ifaces[port->n_ifaces++] = iface;
2998 if (port->n_ifaces > 1) {
2999 port->bridge->has_bonded_ports = true;
3002 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3004 bridge_flush(port->bridge);
3008 iface_destroy(struct iface *iface)
3011 struct port *port = iface->port;
3012 struct bridge *br = port->bridge;
3013 bool del_active = port->active_iface == iface->port_ifidx;
3016 if (iface->dp_ifidx >= 0) {
3017 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3020 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3021 del->port_ifidx = iface->port_ifidx;
3023 netdev_close(iface->netdev);
3028 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3029 bond_choose_active_iface(port);
3030 bond_send_learning_packets(port);
3033 bridge_flush(port->bridge);
3037 static struct iface *
3038 iface_lookup(const struct bridge *br, const char *name)
3042 for (i = 0; i < br->n_ports; i++) {
3043 struct port *port = br->ports[i];
3044 for (j = 0; j < port->n_ifaces; j++) {
3045 struct iface *iface = port->ifaces[j];
3046 if (!strcmp(iface->name, name)) {
3054 static struct iface *
3055 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3057 return port_array_get(&br->ifaces, dp_ifidx);
3060 /* Port mirroring. */
3063 mirror_reconfigure(struct bridge *br)
3065 struct svec old_mirrors, new_mirrors;
3068 /* Collect old and new mirrors. */
3069 svec_init(&old_mirrors);
3070 svec_init(&new_mirrors);
3071 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3072 for (i = 0; i < MAX_MIRRORS; i++) {
3073 if (br->mirrors[i]) {
3074 svec_add(&old_mirrors, br->mirrors[i]->name);
3078 /* Get rid of deleted mirrors and add new mirrors. */
3079 svec_sort(&old_mirrors);
3080 assert(svec_is_unique(&old_mirrors));
3081 svec_sort(&new_mirrors);
3082 assert(svec_is_unique(&new_mirrors));
3083 for (i = 0; i < MAX_MIRRORS; i++) {
3084 struct mirror *m = br->mirrors[i];
3085 if (m && !svec_contains(&new_mirrors, m->name)) {
3089 for (i = 0; i < new_mirrors.n; i++) {
3090 const char *name = new_mirrors.names[i];
3091 if (!svec_contains(&old_mirrors, name)) {
3092 mirror_create(br, name);
3095 svec_destroy(&old_mirrors);
3096 svec_destroy(&new_mirrors);
3098 /* Reconfigure all mirrors. */
3099 for (i = 0; i < MAX_MIRRORS; i++) {
3100 if (br->mirrors[i]) {
3101 mirror_reconfigure_one(br->mirrors[i]);
3105 /* Update port reserved status. */
3106 for (i = 0; i < br->n_ports; i++) {
3107 br->ports[i]->is_mirror_output_port = false;
3109 for (i = 0; i < MAX_MIRRORS; i++) {
3110 struct mirror *m = br->mirrors[i];
3111 if (m && m->out_port) {
3112 m->out_port->is_mirror_output_port = true;
3118 mirror_create(struct bridge *br, const char *name)
3123 for (i = 0; ; i++) {
3124 if (i >= MAX_MIRRORS) {
3125 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3126 "cannot create %s", br->name, MAX_MIRRORS, name);
3129 if (!br->mirrors[i]) {
3134 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3137 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3140 m->name = xstrdup(name);
3141 svec_init(&m->src_ports);
3142 svec_init(&m->dst_ports);
3150 mirror_destroy(struct mirror *m)
3153 struct bridge *br = m->bridge;
3156 for (i = 0; i < br->n_ports; i++) {
3157 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3158 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3161 svec_destroy(&m->src_ports);
3162 svec_destroy(&m->dst_ports);
3165 m->bridge->mirrors[m->idx] = NULL;
3173 prune_ports(struct mirror *m, struct svec *ports)
3178 svec_sort_unique(ports);
3181 for (i = 0; i < ports->n; i++) {
3182 const char *name = ports->names[i];
3183 if (port_lookup(m->bridge, name)) {
3184 svec_add(&tmp, name);
3186 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3187 m->bridge->name, m->name, name);
3190 svec_swap(ports, &tmp);
3195 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3199 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3200 * order won't give us numeric sort order. But that's good enough for what
3201 * we need right now. */
3202 svec_sort_unique(vlan_strings);
3204 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3206 for (i = 0; i < vlan_strings->n; i++) {
3207 const char *name = vlan_strings->names[i];
3209 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3210 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3211 m->bridge->name, m->name, name);
3213 (*vlans)[n_vlans++] = vlan;
3220 vlan_is_mirrored(const struct mirror *m, int vlan)
3224 for (i = 0; i < m->n_vlans; i++) {
3225 if (m->vlans[i] == vlan) {
3233 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3237 for (i = 0; i < m->n_vlans; i++) {
3238 if (port_trunks_vlan(p, m->vlans[i])) {
3246 mirror_reconfigure_one(struct mirror *m)
3248 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3249 struct svec src_ports, dst_ports, ports;
3250 struct svec vlan_strings;
3251 mirror_mask_t mirror_bit;
3252 const char *out_port_name;
3253 struct port *out_port;
3258 bool mirror_all_ports;
3260 /* Get output port. */
3261 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3262 m->bridge->name, m->name);
3263 if (out_port_name) {
3264 out_port = port_lookup(m->bridge, out_port_name);
3266 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3267 "named %s", pfx, m->bridge->name, out_port_name);
3274 if (cfg_has("%s.output.vlan", pfx)) {
3275 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3276 "ignoring %s.output.vlan", pfx, pfx, pfx);
3278 } else if (cfg_has("%s.output.vlan", pfx)) {
3280 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3282 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3283 "but exactly one is required; disabling port mirror %s",
3284 pfx, pfx, pfx, pfx);
3290 /* Get all the ports, and drop duplicates and ports that don't exist. */
3291 svec_init(&src_ports);
3292 svec_init(&dst_ports);
3294 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3295 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3296 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3297 svec_append(&src_ports, &ports);
3298 svec_append(&dst_ports, &ports);
3299 svec_destroy(&ports);
3300 prune_ports(m, &src_ports);
3301 prune_ports(m, &dst_ports);
3303 /* Get all the vlans, and drop duplicate and invalid vlans. */
3304 svec_init(&vlan_strings);
3305 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3306 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3307 svec_destroy(&vlan_strings);
3309 /* Update mirror data. */
3310 if (!svec_equal(&m->src_ports, &src_ports)
3311 || !svec_equal(&m->dst_ports, &dst_ports)
3312 || m->n_vlans != n_vlans
3313 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3314 || m->out_port != out_port
3315 || m->out_vlan != out_vlan) {
3316 bridge_flush(m->bridge);
3318 svec_swap(&m->src_ports, &src_ports);
3319 svec_swap(&m->dst_ports, &dst_ports);
3322 m->n_vlans = n_vlans;
3323 m->out_port = out_port;
3324 m->out_vlan = out_vlan;
3326 /* If no selection criteria have been given, mirror for all ports. */
3327 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3330 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3331 for (i = 0; i < m->bridge->n_ports; i++) {
3332 struct port *port = m->bridge->ports[i];
3334 if (mirror_all_ports
3335 || svec_contains(&m->src_ports, port->name)
3338 ? port_trunks_any_mirrored_vlan(m, port)
3339 : vlan_is_mirrored(m, port->vlan)))) {
3340 port->src_mirrors |= mirror_bit;
3342 port->src_mirrors &= ~mirror_bit;
3345 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3346 port->dst_mirrors |= mirror_bit;
3348 port->dst_mirrors &= ~mirror_bit;
3353 svec_destroy(&src_ports);
3354 svec_destroy(&dst_ports);
3358 /* Spanning tree protocol. */
3360 static void brstp_update_port_state(struct port *);
3363 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3365 struct bridge *br = br_;
3366 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3367 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3369 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3372 struct eth_header *eth = pkt->l2;
3374 netdev_get_etheraddr(iface->netdev, eth->eth_src);
3375 if (eth_addr_is_zero(eth->eth_src)) {
3376 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
3377 "with unknown MAC", br->name, port_no);
3379 union ofp_action action;
3382 memset(&action, 0, sizeof action);
3383 action.type = htons(OFPAT_OUTPUT);
3384 action.output.len = htons(sizeof action);
3385 action.output.port = htons(port_no);
3387 flow_extract(pkt, ODPP_NONE, &flow);
3388 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3395 brstp_reconfigure(struct bridge *br)
3399 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3401 stp_destroy(br->stp);
3407 uint64_t bridge_address, bridge_id;
3408 int bridge_priority;
3410 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3411 if (!bridge_address) {
3413 bridge_address = (stp_get_bridge_id(br->stp)
3414 & ((UINT64_C(1) << 48) - 1));
3416 uint8_t mac[ETH_ADDR_LEN];
3417 eth_addr_random(mac);
3418 bridge_address = eth_addr_to_uint64(mac);
3422 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3424 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3426 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3429 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3431 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3432 br->stp_last_tick = time_msec();
3435 if (bridge_id != stp_get_bridge_id(br->stp)) {
3436 stp_set_bridge_id(br->stp, bridge_id);
3441 for (i = 0; i < br->n_ports; i++) {
3442 struct port *p = br->ports[i];
3444 struct stp_port *sp;
3445 int path_cost, priority;
3451 dp_ifidx = p->ifaces[0]->dp_ifidx;
3452 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3456 sp = stp_get_port(br->stp, dp_ifidx);
3457 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3458 "stp.%s.port.%s.enabled",
3460 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3461 br->name, p->name));
3462 if (p->is_mirror_output_port) {
3465 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3466 bridge_flush(br); /* Might not be necessary. */
3468 stp_port_enable(sp);
3470 stp_port_disable(sp);
3474 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3476 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3478 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3479 "stp.%s.port.%s.priority",
3481 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3483 : STP_DEFAULT_PORT_PRIORITY);
3484 stp_port_set_priority(sp, priority);
3487 brstp_adjust_timers(br);
3489 for (i = 0; i < br->n_ports; i++) {
3490 brstp_update_port_state(br->ports[i]);
3495 brstp_update_port_state(struct port *p)
3497 struct bridge *br = p->bridge;
3498 enum stp_state state;
3500 /* Figure out new state. */
3501 state = STP_DISABLED;
3502 if (br->stp && p->n_ifaces > 0) {
3503 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3504 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3505 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3510 if (p->stp_state != state) {
3511 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3512 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3513 p->name, stp_state_name(p->stp_state),
3514 stp_state_name(state));
3515 if (p->stp_state == STP_DISABLED) {
3518 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3520 p->stp_state = state;
3521 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3522 : tag_create_random());
3527 brstp_adjust_timers(struct bridge *br)
3529 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3530 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3531 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3533 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3534 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3535 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3539 brstp_run(struct bridge *br)
3542 long long int now = time_msec();
3543 long long int elapsed = now - br->stp_last_tick;
3544 struct stp_port *sp;
3547 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3548 br->stp_last_tick = now;
3550 while (stp_get_changed_port(br->stp, &sp)) {
3551 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3553 brstp_update_port_state(p);
3560 brstp_wait(struct bridge *br)
3563 poll_timer_wait(1000);