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 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_carrier(iface->netdev, &iface->enabled);
376 /* If the network device can't be opened, then we're not going to try
377 * to do anything with this interface. */
383 check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED)
385 if (iface->dp_ifidx >= 0) {
386 VLOG_DBG("%s has interface %s on port %d",
388 iface->name, iface->dp_ifidx);
391 VLOG_ERR("%s interface not in %s, dropping",
392 iface->name, dpif_name(br->dpif));
398 set_iface_policing(struct bridge *br UNUSED, struct iface *iface,
401 int rate = cfg_get_int(0, "port.%s.ingress.policing-rate", iface->name);
402 int burst = cfg_get_int(0, "port.%s.ingress.policing-burst", iface->name);
403 netdev_set_policing(iface->netdev, rate, burst);
407 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
408 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
409 * deletes from 'br' any ports that no longer have any interfaces. */
411 iterate_and_prune_ifaces(struct bridge *br,
412 bool (*cb)(struct bridge *, struct iface *,
418 for (i = 0; i < br->n_ports; ) {
419 struct port *port = br->ports[i];
420 for (j = 0; j < port->n_ifaces; ) {
421 struct iface *iface = port->ifaces[j];
422 if (cb(br, iface, aux)) {
425 iface_destroy(iface);
429 if (port->n_ifaces) {
432 VLOG_ERR("%s port has no interfaces, dropping", port->name);
439 bridge_reconfigure(void)
441 struct svec old_br, new_br;
442 struct bridge *br, *next;
445 COVERAGE_INC(bridge_reconfigure);
447 /* Collect old and new bridges. */
450 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
451 svec_add(&old_br, br->name);
453 cfg_get_subsections(&new_br, "bridge");
455 /* Get rid of deleted bridges and add new bridges. */
458 assert(svec_is_unique(&old_br));
459 assert(svec_is_unique(&new_br));
460 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
461 if (!svec_contains(&new_br, br->name)) {
465 for (i = 0; i < new_br.n; i++) {
466 const char *name = new_br.names[i];
467 if (!svec_contains(&old_br, name)) {
471 svec_destroy(&old_br);
472 svec_destroy(&new_br);
476 bridge_configure_ssl();
479 /* Reconfigure all bridges. */
480 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
481 bridge_reconfigure_one(br);
484 /* Add and delete ports on all datapaths.
486 * The kernel will reject any attempt to add a given port to a datapath if
487 * that port already belongs to a different datapath, so we must do all
488 * port deletions before any port additions. */
489 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
490 struct odp_port *dpif_ports;
492 struct svec want_ifaces;
494 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
495 bridge_get_all_ifaces(br, &want_ifaces);
496 for (i = 0; i < n_dpif_ports; i++) {
497 const struct odp_port *p = &dpif_ports[i];
498 if (!svec_contains(&want_ifaces, p->devname)
499 && strcmp(p->devname, br->name)) {
500 int retval = dpif_port_del(br->dpif, p->port);
502 VLOG_ERR("failed to remove %s interface from %s: %s",
503 p->devname, dpif_name(br->dpif),
508 svec_destroy(&want_ifaces);
511 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
512 struct odp_port *dpif_ports;
514 struct svec cur_ifaces, want_ifaces, add_ifaces;
516 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
517 svec_init(&cur_ifaces);
518 for (i = 0; i < n_dpif_ports; i++) {
519 svec_add(&cur_ifaces, dpif_ports[i].devname);
522 svec_sort_unique(&cur_ifaces);
523 bridge_get_all_ifaces(br, &want_ifaces);
524 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
526 for (i = 0; i < add_ifaces.n; i++) {
527 const char *if_name = add_ifaces.names[i];
528 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
529 int flags = internal ? ODP_PORT_INTERNAL : 0;
530 int error = dpif_port_add(br->dpif, if_name, flags, NULL);
531 if (error == EXFULL) {
532 VLOG_ERR("ran out of valid port numbers on %s",
533 dpif_name(br->dpif));
536 VLOG_ERR("failed to add %s interface to %s: %s",
537 if_name, dpif_name(br->dpif), strerror(error));
540 svec_destroy(&cur_ifaces);
541 svec_destroy(&want_ifaces);
542 svec_destroy(&add_ifaces);
544 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
547 struct iface *local_iface;
548 struct iface *hw_addr_iface;
549 uint8_t engine_type, engine_id;
550 bool add_id_to_iface = false;
551 struct svec nf_hosts;
553 bridge_fetch_dp_ifaces(br);
554 iterate_and_prune_ifaces(br, init_iface_netdev, NULL);
556 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
558 /* Pick local port hardware address, datapath ID. */
559 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
560 local_iface = bridge_get_local_iface(br);
562 int error = netdev_set_etheraddr(local_iface->netdev, ea);
564 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
565 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
566 "Ethernet address: %s",
567 br->name, strerror(error));
571 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
572 ofproto_set_datapath_id(br->ofproto, dpid);
574 /* Set NetFlow configuration on this bridge. */
575 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
576 if (cfg_has("netflow.%s.engine-type", br->name)) {
577 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
580 if (cfg_has("netflow.%s.engine-id", br->name)) {
581 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
583 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
584 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
587 if (add_id_to_iface && engine_id > 0x7f) {
588 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
589 "another vswitch, choose an engine id less than 128",
592 if (add_id_to_iface && br->n_ports > 0x1ff) {
593 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
594 "another port when 512 or more ports are used",
597 svec_init(&nf_hosts);
598 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
599 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
600 engine_id, add_id_to_iface)) {
601 VLOG_ERR("bridge %s: problem setting netflow collectors",
605 /* Update the controller and related settings. It would be more
606 * straightforward to call this from bridge_reconfigure_one(), but we
607 * can't do it there for two reasons. First, and most importantly, at
608 * that point we don't know the dp_ifidx of any interfaces that have
609 * been added to the bridge (because we haven't actually added them to
610 * the datapath). Second, at that point we haven't set the datapath ID
611 * yet; when a controller is configured, resetting the datapath ID will
612 * immediately disconnect from the controller, so it's better to set
613 * the datapath ID before the controller. */
614 bridge_reconfigure_controller(br);
616 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
617 for (i = 0; i < br->n_ports; i++) {
618 struct port *port = br->ports[i];
619 port_update_vlan_compat(port);
620 port_update_bonding(port);
623 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
624 brstp_reconfigure(br);
625 iterate_and_prune_ifaces(br, set_iface_policing, NULL);
630 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
631 struct iface **hw_addr_iface)
633 uint64_t requested_ea;
637 *hw_addr_iface = NULL;
639 /* Did the user request a particular MAC? */
640 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
642 eth_addr_from_uint64(requested_ea, ea);
643 if (eth_addr_is_multicast(ea)) {
644 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
645 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
646 } else if (eth_addr_is_zero(ea)) {
647 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
653 /* Otherwise choose the minimum MAC address among all of the interfaces.
654 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
655 * MAC of the physical interface in such an environment.) */
656 memset(ea, 0xff, sizeof ea);
657 for (i = 0; i < br->n_ports; i++) {
658 struct port *port = br->ports[i];
659 if (port->is_mirror_output_port) {
662 for (j = 0; j < port->n_ifaces; j++) {
663 struct iface *iface = port->ifaces[j];
664 uint8_t iface_ea[ETH_ADDR_LEN];
665 if (iface->dp_ifidx == ODPP_LOCAL
666 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
669 error = netdev_get_etheraddr(iface->netdev, iface_ea);
671 if (!eth_addr_is_multicast(iface_ea) &&
672 !eth_addr_is_reserved(iface_ea) &&
673 !eth_addr_is_zero(iface_ea) &&
674 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
675 memcpy(ea, iface_ea, ETH_ADDR_LEN);
676 *hw_addr_iface = iface;
679 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
680 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
681 iface->name, strerror(error));
685 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
686 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
687 *hw_addr_iface = NULL;
688 VLOG_WARN("bridge %s: using default bridge Ethernet "
689 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
691 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
692 br->name, ETH_ADDR_ARGS(ea));
696 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
697 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
698 * an interface on 'br', then that interface must be passed in as
699 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
700 * 'hw_addr_iface' must be passed in as a null pointer. */
702 bridge_pick_datapath_id(struct bridge *br,
703 const uint8_t bridge_ea[ETH_ADDR_LEN],
704 struct iface *hw_addr_iface)
707 * The procedure for choosing a bridge MAC address will, in the most
708 * ordinary case, also choose a unique MAC that we can use as a datapath
709 * ID. In some special cases, though, multiple bridges will end up with
710 * the same MAC address. This is OK for the bridges, but it will confuse
711 * the OpenFlow controller, because each datapath needs a unique datapath
714 * Datapath IDs must be unique. It is also very desirable that they be
715 * stable from one run to the next, so that policy set on a datapath
720 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
727 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
729 * A bridge whose MAC address is taken from a VLAN network device
730 * (that is, a network device created with vconfig(8) or similar
731 * tool) will have the same MAC address as a bridge on the VLAN
732 * device's physical network device.
734 * Handle this case by hashing the physical network device MAC
735 * along with the VLAN identifier.
737 uint8_t buf[ETH_ADDR_LEN + 2];
738 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
739 buf[ETH_ADDR_LEN] = vlan >> 8;
740 buf[ETH_ADDR_LEN + 1] = vlan;
741 return dpid_from_hash(buf, sizeof buf);
744 * Assume that this bridge's MAC address is unique, since it
745 * doesn't fit any of the cases we handle specially.
750 * A purely internal bridge, that is, one that has no non-virtual
751 * network devices on it at all, is more difficult because it has no
752 * natural unique identifier at all.
754 * When the host is a XenServer, we handle this case by hashing the
755 * host's UUID with the name of the bridge. Names of bridges are
756 * persistent across XenServer reboots, although they can be reused if
757 * an internal network is destroyed and then a new one is later
758 * created, so this is fairly effective.
760 * When the host is not a XenServer, we punt by using a random MAC
761 * address on each run.
763 const char *host_uuid = xenserver_get_host_uuid();
765 char *combined = xasprintf("%s,%s", host_uuid, br->name);
766 dpid = dpid_from_hash(combined, strlen(combined));
772 return eth_addr_to_uint64(bridge_ea);
776 dpid_from_hash(const void *data, size_t n)
778 uint8_t hash[SHA1_DIGEST_SIZE];
780 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
781 sha1_bytes(data, n, hash);
782 eth_addr_mark_random(hash);
783 return eth_addr_to_uint64(hash);
789 struct bridge *br, *next;
793 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
794 int error = bridge_run_one(br);
796 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
797 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
798 "forcing reconfiguration", br->name);
812 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
813 ofproto_wait(br->ofproto);
814 if (br->controller) {
819 mac_learning_wait(br->ml);
826 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
827 * configuration changes. */
829 bridge_flush(struct bridge *br)
831 COVERAGE_INC(bridge_flush);
834 mac_learning_flush(br->ml);
838 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
840 static struct iface *
841 bridge_get_local_iface(struct bridge *br)
845 for (i = 0; i < br->n_ports; i++) {
846 struct port *port = br->ports[i];
847 for (j = 0; j < port->n_ifaces; j++) {
848 struct iface *iface = port->ifaces[j];
849 if (iface->dp_ifidx == ODPP_LOCAL) {
858 /* Bridge unixctl user interface functions. */
860 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
862 struct ds ds = DS_EMPTY_INITIALIZER;
863 const struct bridge *br;
865 br = bridge_lookup(args);
867 unixctl_command_reply(conn, 501, "no such bridge");
871 ds_put_cstr(&ds, " port VLAN MAC Age\n");
873 const struct mac_entry *e;
874 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
875 if (e->port < 0 || e->port >= br->n_ports) {
878 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
879 br->ports[e->port]->ifaces[0]->dp_ifidx,
880 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
883 unixctl_command_reply(conn, 200, ds_cstr(&ds));
887 /* Bridge reconfiguration functions. */
889 static struct bridge *
890 bridge_create(const char *name)
895 assert(!bridge_lookup(name));
896 br = xcalloc(1, sizeof *br);
898 error = dpif_create(name, &br->dpif);
899 if (error == EEXIST || error == EBUSY) {
900 error = dpif_open(name, &br->dpif);
902 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
903 name, strerror(error));
907 dpif_flow_flush(br->dpif);
909 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
914 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
916 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
917 dpif_delete(br->dpif);
918 dpif_close(br->dpif);
923 br->name = xstrdup(name);
924 br->ml = mac_learning_create();
925 br->sent_config_request = false;
926 eth_addr_random(br->default_ea);
928 port_array_init(&br->ifaces);
931 br->bond_next_rebalance = time_msec() + 10000;
933 list_push_back(&all_bridges, &br->node);
935 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
941 bridge_destroy(struct bridge *br)
946 while (br->n_ports > 0) {
947 port_destroy(br->ports[br->n_ports - 1]);
949 list_remove(&br->node);
950 error = dpif_delete(br->dpif);
951 if (error && error != ENOENT) {
952 VLOG_ERR("failed to delete %s: %s",
953 dpif_name(br->dpif), strerror(error));
955 dpif_close(br->dpif);
956 ofproto_destroy(br->ofproto);
957 free(br->controller);
958 mac_learning_destroy(br->ml);
959 port_array_destroy(&br->ifaces);
966 static struct bridge *
967 bridge_lookup(const char *name)
971 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
972 if (!strcmp(br->name, name)) {
980 bridge_exists(const char *name)
982 return bridge_lookup(name) ? true : false;
986 bridge_get_datapathid(const char *name)
988 struct bridge *br = bridge_lookup(name);
989 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
993 bridge_run_one(struct bridge *br)
997 error = ofproto_run1(br->ofproto);
1003 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1008 error = ofproto_run2(br->ofproto, br->flush);
1015 bridge_get_controller(const struct bridge *br)
1017 const char *controller;
1019 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1021 controller = cfg_get_string(0, "mgmt.controller");
1023 return controller && controller[0] ? controller : NULL;
1027 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1029 struct svec *ifaces = ifaces_;
1030 if (!svec_contains(ifaces, iface->name)) {
1031 svec_add(ifaces, iface->name);
1035 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1037 br->name, iface->name, iface->port->name);
1043 bridge_reconfigure_one(struct bridge *br)
1045 struct svec old_ports, new_ports, ifaces;
1046 struct svec listeners, old_listeners;
1047 struct svec snoops, old_snoops;
1050 /* Collect old ports. */
1051 svec_init(&old_ports);
1052 for (i = 0; i < br->n_ports; i++) {
1053 svec_add(&old_ports, br->ports[i]->name);
1055 svec_sort(&old_ports);
1056 assert(svec_is_unique(&old_ports));
1058 /* Collect new ports. */
1059 svec_init(&new_ports);
1060 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1061 svec_sort(&new_ports);
1062 if (bridge_get_controller(br)) {
1063 char local_name[IF_NAMESIZE];
1066 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1067 local_name, sizeof local_name);
1068 if (!error && !svec_contains(&new_ports, local_name)) {
1069 svec_add(&new_ports, local_name);
1070 svec_sort(&new_ports);
1073 if (!svec_is_unique(&new_ports)) {
1074 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1075 br->name, svec_get_duplicate(&new_ports));
1076 svec_unique(&new_ports);
1079 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1081 /* Get rid of deleted ports and add new ports. */
1082 for (i = 0; i < br->n_ports; ) {
1083 struct port *port = br->ports[i];
1084 if (!svec_contains(&new_ports, port->name)) {
1090 for (i = 0; i < new_ports.n; i++) {
1091 const char *name = new_ports.names[i];
1092 if (!svec_contains(&old_ports, name)) {
1093 port_create(br, name);
1096 svec_destroy(&old_ports);
1097 svec_destroy(&new_ports);
1099 /* Reconfigure all ports. */
1100 for (i = 0; i < br->n_ports; i++) {
1101 port_reconfigure(br->ports[i]);
1104 /* Check and delete duplicate interfaces. */
1106 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1107 svec_destroy(&ifaces);
1109 /* Delete all flows if we're switching from connected to standalone or vice
1110 * versa. (XXX Should we delete all flows if we are switching from one
1111 * controller to another?) */
1113 /* Configure OpenFlow management listeners. */
1114 svec_init(&listeners);
1115 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1117 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1118 ovs_rundir, br->name));
1119 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1120 svec_clear(&listeners);
1122 svec_sort_unique(&listeners);
1124 svec_init(&old_listeners);
1125 ofproto_get_listeners(br->ofproto, &old_listeners);
1126 svec_sort_unique(&old_listeners);
1128 if (!svec_equal(&listeners, &old_listeners)) {
1129 ofproto_set_listeners(br->ofproto, &listeners);
1131 svec_destroy(&listeners);
1132 svec_destroy(&old_listeners);
1134 /* Configure OpenFlow controller connection snooping. */
1136 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1138 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1139 ovs_rundir, br->name));
1140 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1141 svec_clear(&snoops);
1143 svec_sort_unique(&snoops);
1145 svec_init(&old_snoops);
1146 ofproto_get_snoops(br->ofproto, &old_snoops);
1147 svec_sort_unique(&old_snoops);
1149 if (!svec_equal(&snoops, &old_snoops)) {
1150 ofproto_set_snoops(br->ofproto, &snoops);
1152 svec_destroy(&snoops);
1153 svec_destroy(&old_snoops);
1155 mirror_reconfigure(br);
1159 bridge_reconfigure_controller(struct bridge *br)
1161 char *pfx = xasprintf("bridge.%s.controller", br->name);
1162 const char *controller;
1164 controller = bridge_get_controller(br);
1165 if ((br->controller != NULL) != (controller != NULL)) {
1166 ofproto_flush_flows(br->ofproto);
1168 free(br->controller);
1169 br->controller = controller ? xstrdup(controller) : NULL;
1172 const char *fail_mode;
1173 int max_backoff, probe;
1174 int rate_limit, burst_limit;
1176 if (!strcmp(controller, "discover")) {
1177 bool update_resolv_conf = true;
1179 if (cfg_has("%s.update-resolv.conf", pfx)) {
1180 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1183 ofproto_set_discovery(br->ofproto, true,
1184 cfg_get_string(0, "%s.accept-regex", pfx),
1185 update_resolv_conf);
1187 struct iface *local_iface;
1190 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1192 || cfg_get_bool(0, "%s.in-band", pfx));
1193 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1194 ofproto_set_in_band(br->ofproto, in_band);
1196 local_iface = bridge_get_local_iface(br);
1198 && cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1199 struct netdev *netdev = local_iface->netdev;
1200 struct in_addr ip, mask, gateway;
1201 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1202 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1203 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1205 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1207 mask.s_addr = guess_netmask(ip.s_addr);
1209 if (!netdev_set_in4(netdev, ip, mask)) {
1210 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1212 br->name, IP_ARGS(&ip.s_addr),
1213 IP_ARGS(&mask.s_addr));
1216 if (gateway.s_addr) {
1217 if (!netdev_add_router(netdev, gateway)) {
1218 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1219 br->name, IP_ARGS(&gateway.s_addr));
1225 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1227 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1229 ofproto_set_failure(br->ofproto,
1231 || !strcmp(fail_mode, "standalone")
1232 || !strcmp(fail_mode, "open")));
1234 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1236 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1241 ofproto_set_probe_interval(br->ofproto, probe);
1243 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1245 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1250 ofproto_set_max_backoff(br->ofproto, max_backoff);
1252 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1254 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1256 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1258 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1260 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1262 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1264 if (cfg_has("%s.commands.acl", pfx)) {
1265 struct svec command_acls;
1268 svec_init(&command_acls);
1269 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1270 command_acl = svec_join(&command_acls, ",", "");
1272 ofproto_set_remote_execution(br->ofproto, command_acl,
1273 cfg_get_string(0, "%s.commands.dir",
1276 svec_destroy(&command_acls);
1279 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1282 union ofp_action action;
1285 /* Set up a flow that matches every packet and directs them to
1286 * OFPP_NORMAL (which goes to us). */
1287 memset(&action, 0, sizeof action);
1288 action.type = htons(OFPAT_OUTPUT);
1289 action.output.len = htons(sizeof action);
1290 action.output.port = htons(OFPP_NORMAL);
1291 memset(&flow, 0, sizeof flow);
1292 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1295 ofproto_set_in_band(br->ofproto, false);
1296 ofproto_set_max_backoff(br->ofproto, 1);
1297 ofproto_set_probe_interval(br->ofproto, 5);
1298 ofproto_set_failure(br->ofproto, false);
1299 ofproto_set_stp(br->ofproto, false);
1303 ofproto_set_controller(br->ofproto, br->controller);
1307 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1312 for (i = 0; i < br->n_ports; i++) {
1313 struct port *port = br->ports[i];
1314 for (j = 0; j < port->n_ifaces; j++) {
1315 struct iface *iface = port->ifaces[j];
1316 svec_add(ifaces, iface->name);
1320 assert(svec_is_unique(ifaces));
1323 /* For robustness, in case the administrator moves around datapath ports behind
1324 * our back, we re-check all the datapath port numbers here.
1326 * This function will set the 'dp_ifidx' members of interfaces that have
1327 * disappeared to -1, so only call this function from a context where those
1328 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1329 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1330 * datapath, which doesn't support UINT16_MAX+1 ports. */
1332 bridge_fetch_dp_ifaces(struct bridge *br)
1334 struct odp_port *dpif_ports;
1335 size_t n_dpif_ports;
1338 /* Reset all interface numbers. */
1339 for (i = 0; i < br->n_ports; i++) {
1340 struct port *port = br->ports[i];
1341 for (j = 0; j < port->n_ifaces; j++) {
1342 struct iface *iface = port->ifaces[j];
1343 iface->dp_ifidx = -1;
1346 port_array_clear(&br->ifaces);
1348 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1349 for (i = 0; i < n_dpif_ports; i++) {
1350 struct odp_port *p = &dpif_ports[i];
1351 struct iface *iface = iface_lookup(br, p->devname);
1353 if (iface->dp_ifidx >= 0) {
1354 VLOG_WARN("%s reported interface %s twice",
1355 dpif_name(br->dpif), p->devname);
1356 } else if (iface_from_dp_ifidx(br, p->port)) {
1357 VLOG_WARN("%s reported interface %"PRIu16" twice",
1358 dpif_name(br->dpif), p->port);
1360 port_array_set(&br->ifaces, p->port, iface);
1361 iface->dp_ifidx = p->port;
1368 /* Bridge packet processing functions. */
1371 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1373 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1376 static struct bond_entry *
1377 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1379 return &port->bond_hash[bond_hash(mac)];
1383 bond_choose_iface(const struct port *port)
1386 for (i = 0; i < port->n_ifaces; i++) {
1387 if (port->ifaces[i]->enabled) {
1395 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1396 uint16_t *dp_ifidx, tag_type *tags)
1398 struct iface *iface;
1400 assert(port->n_ifaces);
1401 if (port->n_ifaces == 1) {
1402 iface = port->ifaces[0];
1404 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1405 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1406 || !port->ifaces[e->iface_idx]->enabled) {
1407 /* XXX select interface properly. The current interface selection
1408 * is only good for testing the rebalancing code. */
1409 e->iface_idx = bond_choose_iface(port);
1410 if (e->iface_idx < 0) {
1411 *tags |= port->no_ifaces_tag;
1414 e->iface_tag = tag_create_random();
1416 *tags |= e->iface_tag;
1417 iface = port->ifaces[e->iface_idx];
1419 *dp_ifidx = iface->dp_ifidx;
1420 *tags |= iface->tag; /* Currently only used for bonding. */
1425 bond_link_status_update(struct iface *iface, bool carrier)
1427 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1428 struct port *port = iface->port;
1430 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1431 /* Nothing to do. */
1434 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1435 iface->name, carrier ? "detected" : "dropped");
1436 if (carrier == iface->enabled) {
1437 iface->delay_expires = LLONG_MAX;
1438 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1439 iface->name, carrier ? "disabled" : "enabled");
1440 } else if (carrier && port->updelay && port->active_iface < 0) {
1441 iface->delay_expires = time_msec();
1442 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1443 "other interface is up", iface->name, port->updelay);
1445 int delay = carrier ? port->updelay : port->downdelay;
1446 iface->delay_expires = time_msec() + delay;
1449 "interface %s: will be %s if it stays %s for %d ms",
1451 carrier ? "enabled" : "disabled",
1452 carrier ? "up" : "down",
1459 bond_choose_active_iface(struct port *port)
1461 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1463 port->active_iface = bond_choose_iface(port);
1464 port->active_iface_tag = tag_create_random();
1465 if (port->active_iface >= 0) {
1466 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1467 port->name, port->ifaces[port->active_iface]->name);
1469 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1475 bond_enable_slave(struct iface *iface, bool enable)
1477 struct port *port = iface->port;
1478 struct bridge *br = port->bridge;
1480 iface->delay_expires = LLONG_MAX;
1481 if (enable == iface->enabled) {
1485 iface->enabled = enable;
1486 if (!iface->enabled) {
1487 VLOG_WARN("interface %s: disabled", iface->name);
1488 ofproto_revalidate(br->ofproto, iface->tag);
1489 if (iface->port_ifidx == port->active_iface) {
1490 ofproto_revalidate(br->ofproto,
1491 port->active_iface_tag);
1492 bond_choose_active_iface(port);
1494 bond_send_learning_packets(port);
1496 VLOG_WARN("interface %s: enabled", iface->name);
1497 if (port->active_iface < 0) {
1498 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1499 bond_choose_active_iface(port);
1500 bond_send_learning_packets(port);
1502 iface->tag = tag_create_random();
1507 bond_run(struct bridge *br)
1511 for (i = 0; i < br->n_ports; i++) {
1512 struct port *port = br->ports[i];
1513 if (port->n_ifaces < 2) {
1516 for (j = 0; j < port->n_ifaces; j++) {
1517 struct iface *iface = port->ifaces[j];
1518 if (time_msec() >= iface->delay_expires) {
1519 bond_enable_slave(iface, !iface->enabled);
1526 bond_wait(struct bridge *br)
1530 for (i = 0; i < br->n_ports; i++) {
1531 struct port *port = br->ports[i];
1532 if (port->n_ifaces < 2) {
1535 for (j = 0; j < port->n_ifaces; j++) {
1536 struct iface *iface = port->ifaces[j];
1537 if (iface->delay_expires != LLONG_MAX) {
1538 poll_timer_wait(iface->delay_expires - time_msec());
1545 set_dst(struct dst *p, const flow_t *flow,
1546 const struct port *in_port, const struct port *out_port,
1551 * XXX This uses too many tags: any broadcast flow will get one tag per
1552 * destination port, and thus a broadcast on a switch of any size is likely
1553 * to have all tag bits set. We should figure out a way to be smarter.
1555 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1556 *tags |= out_port->stp_state_tag;
1557 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1561 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1562 : in_port->vlan >= 0 ? in_port->vlan
1563 : ntohs(flow->dl_vlan));
1564 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1568 swap_dst(struct dst *p, struct dst *q)
1570 struct dst tmp = *p;
1575 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1576 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1577 * that we push to the datapath. We could in fact fully sort the array by
1578 * vlan, but in most cases there are at most two different vlan tags so that's
1579 * possibly overkill.) */
1581 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1583 struct dst *first = dsts;
1584 struct dst *last = dsts + n_dsts;
1586 while (first != last) {
1588 * - All dsts < first have vlan == 'vlan'.
1589 * - All dsts >= last have vlan != 'vlan'.
1590 * - first < last. */
1591 while (first->vlan == vlan) {
1592 if (++first == last) {
1597 /* Same invariants, plus one additional:
1598 * - first->vlan != vlan.
1600 while (last[-1].vlan != vlan) {
1601 if (--last == first) {
1606 /* Same invariants, plus one additional:
1607 * - last[-1].vlan == vlan.*/
1608 swap_dst(first++, --last);
1613 mirror_mask_ffs(mirror_mask_t mask)
1615 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1620 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1621 const struct dst *test)
1624 for (i = 0; i < n_dsts; i++) {
1625 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1633 port_trunks_vlan(const struct port *port, uint16_t vlan)
1635 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1639 port_includes_vlan(const struct port *port, uint16_t vlan)
1641 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1645 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1646 const struct port *in_port, const struct port *out_port,
1647 struct dst dsts[], tag_type *tags)
1649 mirror_mask_t mirrors = in_port->src_mirrors;
1650 struct dst *dst = dsts;
1653 *tags |= in_port->stp_state_tag;
1654 if (out_port == FLOOD_PORT) {
1655 /* XXX use ODP_FLOOD if no vlans or bonding. */
1656 /* XXX even better, define each VLAN as a datapath port group */
1657 for (i = 0; i < br->n_ports; i++) {
1658 struct port *port = br->ports[i];
1659 if (port != in_port && port_includes_vlan(port, vlan)
1660 && !port->is_mirror_output_port
1661 && set_dst(dst, flow, in_port, port, tags)) {
1662 mirrors |= port->dst_mirrors;
1666 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1667 mirrors |= out_port->dst_mirrors;
1672 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1673 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1675 if (set_dst(dst, flow, in_port, m->out_port, tags)
1676 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1680 for (i = 0; i < br->n_ports; i++) {
1681 struct port *port = br->ports[i];
1682 if (port_includes_vlan(port, m->out_vlan)
1683 && set_dst(dst, flow, in_port, port, tags)
1684 && !dst_is_duplicate(dsts, dst - dsts, dst))
1686 if (port->vlan < 0) {
1687 dst->vlan = m->out_vlan;
1689 if (dst->dp_ifidx == flow->in_port
1690 && dst->vlan == vlan) {
1691 /* Don't send out input port on same VLAN. */
1699 mirrors &= mirrors - 1;
1702 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1707 print_dsts(const struct dst *dsts, size_t n)
1709 for (; n--; dsts++) {
1710 printf(">p%"PRIu16, dsts->dp_ifidx);
1711 if (dsts->vlan != OFP_VLAN_NONE) {
1712 printf("v%"PRIu16, dsts->vlan);
1718 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1719 const struct port *in_port, const struct port *out_port,
1720 tag_type *tags, struct odp_actions *actions)
1722 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1724 const struct dst *p;
1727 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1729 cur_vlan = ntohs(flow->dl_vlan);
1730 for (p = dsts; p < &dsts[n_dsts]; p++) {
1731 union odp_action *a;
1732 if (p->vlan != cur_vlan) {
1733 if (p->vlan == OFP_VLAN_NONE) {
1734 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1736 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1737 a->vlan_vid.vlan_vid = htons(p->vlan);
1741 a = odp_actions_add(actions, ODPAT_OUTPUT);
1742 a->output.port = p->dp_ifidx;
1747 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1749 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1750 return (flow->dl_type == htons(ETH_TYPE_ARP)
1751 && eth_addr_is_broadcast(flow->dl_dst)
1752 && packet->size >= sizeof(struct arp_eth_header)
1753 && arp->ar_op == ARP_OP_REQUEST);
1756 /* If the composed actions may be applied to any packet in the given 'flow',
1757 * returns true. Otherwise, the actions should only be applied to 'packet', or
1758 * not at all, if 'packet' was NULL. */
1760 process_flow(struct bridge *br, const flow_t *flow,
1761 const struct ofpbuf *packet, struct odp_actions *actions,
1764 struct iface *in_iface;
1765 struct port *in_port;
1766 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1769 /* Find the interface and port structure for the received packet. */
1770 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1772 /* No interface? Something fishy... */
1773 if (packet != NULL) {
1774 /* Odd. A few possible reasons here:
1776 * - We deleted an interface but there are still a few packets
1777 * queued up from it.
1779 * - Someone externally added an interface (e.g. with "ovs-dpctl
1780 * add-if") that we don't know about.
1782 * - Packet arrived on the local port but the local port is not
1783 * one of our bridge ports.
1785 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1787 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1788 "interface %"PRIu16, br->name, flow->in_port);
1791 /* Return without adding any actions, to drop packets on this flow. */
1794 in_port = in_iface->port;
1796 /* Figure out what VLAN this packet belongs to.
1798 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1799 * belongs to VLAN 0, so we should treat both cases identically. (In the
1800 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1801 * presumably to allow a priority to be specified. In the latter case, the
1802 * packet does not have any 802.1Q header.) */
1803 vlan = ntohs(flow->dl_vlan);
1804 if (vlan == OFP_VLAN_NONE) {
1807 if (in_port->vlan >= 0) {
1809 /* XXX support double tagging? */
1810 if (packet != NULL) {
1811 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1812 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1813 "packet received on port %s configured with "
1814 "implicit VLAN %"PRIu16,
1815 br->name, ntohs(flow->dl_vlan),
1816 in_port->name, in_port->vlan);
1820 vlan = in_port->vlan;
1822 if (!port_includes_vlan(in_port, vlan)) {
1823 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1824 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1825 "packet received on port %s not configured for "
1827 br->name, vlan, in_port->name, vlan);
1832 /* Drop frames for ports that STP wants entirely killed (both for
1833 * forwarding and for learning). Later, after we do learning, we'll drop
1834 * the frames that STP wants to do learning but not forwarding on. */
1835 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1839 /* Drop frames for reserved multicast addresses. */
1840 if (eth_addr_is_reserved(flow->dl_dst)) {
1844 /* Drop frames on ports reserved for mirroring. */
1845 if (in_port->is_mirror_output_port) {
1846 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1847 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1848 "which is reserved exclusively for mirroring",
1849 br->name, in_port->name);
1853 /* Multicast (and broadcast) packets on bonds need special attention, to
1854 * avoid receiving duplicates. */
1855 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1856 *tags |= in_port->active_iface_tag;
1857 if (in_port->active_iface != in_iface->port_ifidx) {
1858 /* Drop all multicast packets on inactive slaves. */
1861 /* Drop all multicast packets for which we have learned a different
1862 * input port, because we probably sent the packet on one slaves
1863 * and got it back on the active slave. Broadcast ARP replies are
1864 * an exception to this rule: the host has moved to another
1866 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1867 if (src_idx != -1 && src_idx != in_port->port_idx) {
1869 if (!is_bcast_arp_reply(flow, packet)) {
1873 /* No way to know whether it's an ARP reply, because the
1874 * flow entry doesn't include enough information and we
1875 * don't have a packet. Punt. */
1883 out_port = FLOOD_PORT;
1887 /* Learn source MAC (but don't try to learn from revalidation). */
1889 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1890 vlan, in_port->port_idx);
1892 /* The log messages here could actually be useful in debugging,
1893 * so keep the rate limit relatively high. */
1894 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1896 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1897 "on port %s in VLAN %d",
1898 br->name, ETH_ADDR_ARGS(flow->dl_src),
1899 in_port->name, vlan);
1900 ofproto_revalidate(br->ofproto, rev_tag);
1904 /* Determine output port. */
1905 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1907 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1908 out_port = br->ports[out_port_idx];
1912 /* Don't send packets out their input ports. Don't forward frames that STP
1913 * wants us to discard. */
1914 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1919 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1922 * We send out only a single packet, instead of setting up a flow, if the
1923 * packet is an ARP directed to broadcast that arrived on a bonded
1924 * interface. In such a situation ARP requests and replies must be handled
1925 * differently, but OpenFlow unfortunately can't distinguish them.
1927 return (in_port->n_ifaces < 2
1928 || flow->dl_type != htons(ETH_TYPE_ARP)
1929 || !eth_addr_is_broadcast(flow->dl_dst));
1932 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1935 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1936 const struct ofp_phy_port *opp,
1939 struct bridge *br = br_;
1940 struct iface *iface;
1943 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1949 if (reason == OFPPR_DELETE) {
1950 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1951 br->name, iface->name);
1952 iface_destroy(iface);
1953 if (!port->n_ifaces) {
1954 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1955 br->name, port->name);
1961 if (port->n_ifaces > 1) {
1962 bool up = !(opp->state & OFPPS_LINK_DOWN);
1963 bond_link_status_update(iface, up);
1964 port_update_bond_compat(port);
1970 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1971 struct odp_actions *actions, tag_type *tags, void *br_)
1973 struct bridge *br = br_;
1976 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1977 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1978 brstp_receive(br, flow, payload);
1983 COVERAGE_INC(bridge_process_flow);
1984 return process_flow(br, flow, packet, actions, tags);
1988 bridge_account_flow_ofhook_cb(const flow_t *flow,
1989 const union odp_action *actions,
1990 size_t n_actions, unsigned long long int n_bytes,
1993 struct bridge *br = br_;
1994 const union odp_action *a;
1996 if (!br->has_bonded_ports) {
2000 for (a = actions; a < &actions[n_actions]; a++) {
2001 if (a->type == ODPAT_OUTPUT) {
2002 struct port *port = port_from_dp_ifidx(br, a->output.port);
2003 if (port && port->n_ifaces >= 2) {
2004 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2005 e->tx_bytes += n_bytes;
2012 bridge_account_checkpoint_ofhook_cb(void *br_)
2014 struct bridge *br = br_;
2017 if (!br->has_bonded_ports) {
2021 /* The current ofproto implementation calls this callback at least once a
2022 * second, so this timer implementation is sufficient. */
2023 if (time_msec() < br->bond_next_rebalance) {
2026 br->bond_next_rebalance = time_msec() + 10000;
2028 for (i = 0; i < br->n_ports; i++) {
2029 struct port *port = br->ports[i];
2030 if (port->n_ifaces > 1) {
2031 bond_rebalance_port(port);
2036 static struct ofhooks bridge_ofhooks = {
2037 bridge_port_changed_ofhook_cb,
2038 bridge_normal_ofhook_cb,
2039 bridge_account_flow_ofhook_cb,
2040 bridge_account_checkpoint_ofhook_cb,
2043 /* Bonding functions. */
2045 /* Statistics for a single interface on a bonded port, used for load-based
2046 * bond rebalancing. */
2047 struct slave_balance {
2048 struct iface *iface; /* The interface. */
2049 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2051 /* All the "bond_entry"s that are assigned to this interface, in order of
2052 * increasing tx_bytes. */
2053 struct bond_entry **hashes;
2057 /* Sorts pointers to pointers to bond_entries in ascending order by the
2058 * interface to which they are assigned, and within a single interface in
2059 * ascending order of bytes transmitted. */
2061 compare_bond_entries(const void *a_, const void *b_)
2063 const struct bond_entry *const *ap = a_;
2064 const struct bond_entry *const *bp = b_;
2065 const struct bond_entry *a = *ap;
2066 const struct bond_entry *b = *bp;
2067 if (a->iface_idx != b->iface_idx) {
2068 return a->iface_idx > b->iface_idx ? 1 : -1;
2069 } else if (a->tx_bytes != b->tx_bytes) {
2070 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2076 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2077 * *descending* order by number of bytes transmitted. */
2079 compare_slave_balance(const void *a_, const void *b_)
2081 const struct slave_balance *a = a_;
2082 const struct slave_balance *b = b_;
2083 if (a->iface->enabled != b->iface->enabled) {
2084 return a->iface->enabled ? -1 : 1;
2085 } else if (a->tx_bytes != b->tx_bytes) {
2086 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2093 swap_bals(struct slave_balance *a, struct slave_balance *b)
2095 struct slave_balance tmp = *a;
2100 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2101 * given that 'p' (and only 'p') might be in the wrong location.
2103 * This function invalidates 'p', since it might now be in a different memory
2106 resort_bals(struct slave_balance *p,
2107 struct slave_balance bals[], size_t n_bals)
2110 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2111 swap_bals(p, p - 1);
2113 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2114 swap_bals(p, p + 1);
2120 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2122 if (VLOG_IS_DBG_ENABLED()) {
2123 struct ds ds = DS_EMPTY_INITIALIZER;
2124 const struct slave_balance *b;
2126 for (b = bals; b < bals + n_bals; b++) {
2130 ds_put_char(&ds, ',');
2132 ds_put_format(&ds, " %s %"PRIu64"kB",
2133 b->iface->name, b->tx_bytes / 1024);
2135 if (!b->iface->enabled) {
2136 ds_put_cstr(&ds, " (disabled)");
2138 if (b->n_hashes > 0) {
2139 ds_put_cstr(&ds, " (");
2140 for (i = 0; i < b->n_hashes; i++) {
2141 const struct bond_entry *e = b->hashes[i];
2143 ds_put_cstr(&ds, " + ");
2145 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2146 e - port->bond_hash, e->tx_bytes / 1024);
2148 ds_put_cstr(&ds, ")");
2151 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2156 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2158 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2159 struct bond_entry *hash)
2161 struct port *port = from->iface->port;
2162 uint64_t delta = hash->tx_bytes;
2164 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2165 "from %s to %s (now carrying %"PRIu64"kB and "
2166 "%"PRIu64"kB load, respectively)",
2167 port->name, delta / 1024, hash - port->bond_hash,
2168 from->iface->name, to->iface->name,
2169 (from->tx_bytes - delta) / 1024,
2170 (to->tx_bytes + delta) / 1024);
2172 /* Delete element from from->hashes.
2174 * We don't bother to add the element to to->hashes because not only would
2175 * it require more work, the only purpose it would be to allow that hash to
2176 * be migrated to another slave in this rebalancing run, and there is no
2177 * point in doing that. */
2178 if (from->hashes[0] == hash) {
2181 int i = hash - from->hashes[0];
2182 memmove(from->hashes + i, from->hashes + i + 1,
2183 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2187 /* Shift load away from 'from' to 'to'. */
2188 from->tx_bytes -= delta;
2189 to->tx_bytes += delta;
2191 /* Arrange for flows to be revalidated. */
2192 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2193 hash->iface_idx = to->iface->port_ifidx;
2194 hash->iface_tag = tag_create_random();
2198 bond_rebalance_port(struct port *port)
2200 struct slave_balance bals[DP_MAX_PORTS];
2202 struct bond_entry *hashes[BOND_MASK + 1];
2203 struct slave_balance *b, *from, *to;
2204 struct bond_entry *e;
2207 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2208 * descending order of tx_bytes, so that bals[0] represents the most
2209 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2212 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2213 * array for each slave_balance structure, we sort our local array of
2214 * hashes in order by slave, so that all of the hashes for a given slave
2215 * become contiguous in memory, and then we point each 'hashes' members of
2216 * a slave_balance structure to the start of a contiguous group. */
2217 n_bals = port->n_ifaces;
2218 for (b = bals; b < &bals[n_bals]; b++) {
2219 b->iface = port->ifaces[b - bals];
2224 for (i = 0; i <= BOND_MASK; i++) {
2225 hashes[i] = &port->bond_hash[i];
2227 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2228 for (i = 0; i <= BOND_MASK; i++) {
2230 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2231 b = &bals[e->iface_idx];
2232 b->tx_bytes += e->tx_bytes;
2234 b->hashes = &hashes[i];
2239 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2240 log_bals(bals, n_bals, port);
2242 /* Discard slaves that aren't enabled (which were sorted to the back of the
2243 * array earlier). */
2244 while (!bals[n_bals - 1].iface->enabled) {
2251 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2252 to = &bals[n_bals - 1];
2253 for (from = bals; from < to; ) {
2254 uint64_t overload = from->tx_bytes - to->tx_bytes;
2255 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2256 /* The extra load on 'from' (and all less-loaded slaves), compared
2257 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2258 * it is less than ~1Mbps. No point in rebalancing. */
2260 } else if (from->n_hashes == 1) {
2261 /* 'from' only carries a single MAC hash, so we can't shift any
2262 * load away from it, even though we want to. */
2265 /* 'from' is carrying significantly more load than 'to', and that
2266 * load is split across at least two different hashes. Pick a hash
2267 * to migrate to 'to' (the least-loaded slave), given that doing so
2268 * must not cause 'to''s load to exceed 'from''s load.
2270 * The sort order we use means that we prefer to shift away the
2271 * smallest hashes instead of the biggest ones. There is little
2272 * reason behind this decision; we could use the opposite sort
2273 * order to shift away big hashes ahead of small ones. */
2276 for (i = 0; i < from->n_hashes; i++) {
2277 uint64_t delta = from->hashes[i]->tx_bytes;
2278 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2282 if (i < from->n_hashes) {
2283 bond_shift_load(from, to, from->hashes[i]);
2285 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2286 * point to different slave_balance structures. It is only
2287 * valid to do these two operations in a row at all because we
2288 * know that 'from' will not move past 'to' and vice versa. */
2289 resort_bals(from, bals, n_bals);
2290 resort_bals(to, bals, n_bals);
2297 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2298 * historical data to decay to <1% in 7 rebalancing runs. */
2299 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2305 bond_send_learning_packets(struct port *port)
2307 struct bridge *br = port->bridge;
2308 struct mac_entry *e;
2309 struct ofpbuf packet;
2310 int error, n_packets, n_errors;
2312 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2316 ofpbuf_init(&packet, 128);
2317 error = n_packets = n_errors = 0;
2318 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2319 static const char s[] = "Open vSwitch Bond Failover";
2320 union ofp_action actions[2], *a;
2321 struct eth_header *eth;
2322 struct llc_snap_header *llc_snap;
2328 if (e->port == port->port_idx
2329 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2333 /* Compose packet to send. */
2334 ofpbuf_clear(&packet);
2335 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2336 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2337 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2338 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2340 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2341 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2342 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2344 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2345 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2346 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2347 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2348 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2350 /* Compose actions. */
2351 memset(actions, 0, sizeof actions);
2354 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2355 a->vlan_vid.len = htons(sizeof *a);
2356 a->vlan_vid.vlan_vid = htons(e->vlan);
2359 a->output.type = htons(OFPAT_OUTPUT);
2360 a->output.len = htons(sizeof *a);
2361 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2366 flow_extract(&packet, ODPP_NONE, &flow);
2367 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2374 ofpbuf_uninit(&packet);
2377 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2378 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2379 "packets, last error was: %s",
2380 port->name, n_errors, n_packets, strerror(error));
2382 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2383 port->name, n_packets);
2387 /* Bonding unixctl user interface functions. */
2390 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2392 struct ds ds = DS_EMPTY_INITIALIZER;
2393 const struct bridge *br;
2395 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2397 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2400 for (i = 0; i < br->n_ports; i++) {
2401 const struct port *port = br->ports[i];
2402 if (port->n_ifaces > 1) {
2405 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2406 for (j = 0; j < port->n_ifaces; j++) {
2407 const struct iface *iface = port->ifaces[j];
2409 ds_put_cstr(&ds, ", ");
2411 ds_put_cstr(&ds, iface->name);
2413 ds_put_char(&ds, '\n');
2417 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2421 static struct port *
2422 bond_find(const char *name)
2424 const struct bridge *br;
2426 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2429 for (i = 0; i < br->n_ports; i++) {
2430 struct port *port = br->ports[i];
2431 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2440 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2442 struct ds ds = DS_EMPTY_INITIALIZER;
2443 const struct port *port;
2446 port = bond_find(args);
2448 unixctl_command_reply(conn, 501, "no such bond");
2452 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2453 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2454 ds_put_format(&ds, "next rebalance: %lld ms\n",
2455 port->bridge->bond_next_rebalance - time_msec());
2456 for (j = 0; j < port->n_ifaces; j++) {
2457 const struct iface *iface = port->ifaces[j];
2458 struct bond_entry *be;
2461 ds_put_format(&ds, "slave %s: %s\n",
2462 iface->name, iface->enabled ? "enabled" : "disabled");
2463 if (j == port->active_iface) {
2464 ds_put_cstr(&ds, "\tactive slave\n");
2466 if (iface->delay_expires != LLONG_MAX) {
2467 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2468 iface->enabled ? "downdelay" : "updelay",
2469 iface->delay_expires - time_msec());
2473 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2474 int hash = be - port->bond_hash;
2475 struct mac_entry *me;
2477 if (be->iface_idx != j) {
2481 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2482 hash, be->tx_bytes / 1024);
2485 if (!port->bridge->ml) {
2489 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2490 &port->bridge->ml->lrus) {
2493 if (bond_hash(me->mac) == hash
2494 && me->port != port->port_idx
2495 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2496 && dp_ifidx == iface->dp_ifidx)
2498 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2499 ETH_ADDR_ARGS(me->mac));
2504 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2509 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2511 char *args = (char *) args_;
2512 char *save_ptr = NULL;
2513 char *bond_s, *hash_s, *slave_s;
2514 uint8_t mac[ETH_ADDR_LEN];
2516 struct iface *iface;
2517 struct bond_entry *entry;
2520 bond_s = strtok_r(args, " ", &save_ptr);
2521 hash_s = strtok_r(NULL, " ", &save_ptr);
2522 slave_s = strtok_r(NULL, " ", &save_ptr);
2524 unixctl_command_reply(conn, 501,
2525 "usage: bond/migrate BOND HASH SLAVE");
2529 port = bond_find(bond_s);
2531 unixctl_command_reply(conn, 501, "no such bond");
2535 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2536 == ETH_ADDR_SCAN_COUNT) {
2537 hash = bond_hash(mac);
2538 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2539 hash = atoi(hash_s) & BOND_MASK;
2541 unixctl_command_reply(conn, 501, "bad hash");
2545 iface = port_lookup_iface(port, slave_s);
2547 unixctl_command_reply(conn, 501, "no such slave");
2551 if (!iface->enabled) {
2552 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2556 entry = &port->bond_hash[hash];
2557 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2558 entry->iface_idx = iface->port_ifidx;
2559 entry->iface_tag = tag_create_random();
2560 unixctl_command_reply(conn, 200, "migrated");
2564 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2566 char *args = (char *) args_;
2567 char *save_ptr = NULL;
2568 char *bond_s, *slave_s;
2570 struct iface *iface;
2572 bond_s = strtok_r(args, " ", &save_ptr);
2573 slave_s = strtok_r(NULL, " ", &save_ptr);
2575 unixctl_command_reply(conn, 501,
2576 "usage: bond/set-active-slave BOND SLAVE");
2580 port = bond_find(bond_s);
2582 unixctl_command_reply(conn, 501, "no such bond");
2586 iface = port_lookup_iface(port, slave_s);
2588 unixctl_command_reply(conn, 501, "no such slave");
2592 if (!iface->enabled) {
2593 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2597 if (port->active_iface != iface->port_ifidx) {
2598 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2599 port->active_iface = iface->port_ifidx;
2600 port->active_iface_tag = tag_create_random();
2601 VLOG_INFO("port %s: active interface is now %s",
2602 port->name, iface->name);
2603 bond_send_learning_packets(port);
2604 unixctl_command_reply(conn, 200, "done");
2606 unixctl_command_reply(conn, 200, "no change");
2611 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2613 char *args = (char *) args_;
2614 char *save_ptr = NULL;
2615 char *bond_s, *slave_s;
2617 struct iface *iface;
2619 bond_s = strtok_r(args, " ", &save_ptr);
2620 slave_s = strtok_r(NULL, " ", &save_ptr);
2622 unixctl_command_reply(conn, 501,
2623 "usage: bond/enable/disable-slave BOND SLAVE");
2627 port = bond_find(bond_s);
2629 unixctl_command_reply(conn, 501, "no such bond");
2633 iface = port_lookup_iface(port, slave_s);
2635 unixctl_command_reply(conn, 501, "no such slave");
2639 bond_enable_slave(iface, enable);
2640 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2644 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2646 enable_slave(conn, args, true);
2650 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2652 enable_slave(conn, args, false);
2658 unixctl_command_register("bond/list", bond_unixctl_list);
2659 unixctl_command_register("bond/show", bond_unixctl_show);
2660 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2661 unixctl_command_register("bond/set-active-slave",
2662 bond_unixctl_set_active_slave);
2663 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2664 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2667 /* Port functions. */
2670 port_create(struct bridge *br, const char *name)
2674 port = xcalloc(1, sizeof *port);
2676 port->port_idx = br->n_ports;
2678 port->trunks = NULL;
2679 port->name = xstrdup(name);
2680 port->active_iface = -1;
2681 port->stp_state = STP_DISABLED;
2682 port->stp_state_tag = 0;
2684 if (br->n_ports >= br->allocated_ports) {
2685 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2688 br->ports[br->n_ports++] = port;
2690 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2695 port_reconfigure(struct port *port)
2697 bool bonded = cfg_has_section("bonding.%s", port->name);
2698 struct svec old_ifaces, new_ifaces;
2699 unsigned long *trunks;
2703 /* Collect old and new interfaces. */
2704 svec_init(&old_ifaces);
2705 svec_init(&new_ifaces);
2706 for (i = 0; i < port->n_ifaces; i++) {
2707 svec_add(&old_ifaces, port->ifaces[i]->name);
2709 svec_sort(&old_ifaces);
2711 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2712 if (!new_ifaces.n) {
2713 VLOG_ERR("port %s: no interfaces specified for bonded port",
2715 } else if (new_ifaces.n == 1) {
2716 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2720 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2721 if (port->updelay < 0) {
2724 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2725 if (port->downdelay < 0) {
2726 port->downdelay = 0;
2729 svec_init(&new_ifaces);
2730 svec_add(&new_ifaces, port->name);
2733 /* Get rid of deleted interfaces and add new interfaces. */
2734 for (i = 0; i < port->n_ifaces; i++) {
2735 struct iface *iface = port->ifaces[i];
2736 if (!svec_contains(&new_ifaces, iface->name)) {
2737 iface_destroy(iface);
2742 for (i = 0; i < new_ifaces.n; i++) {
2743 const char *name = new_ifaces.names[i];
2744 if (!svec_contains(&old_ifaces, name)) {
2745 iface_create(port, name);
2751 if (cfg_has("vlan.%s.tag", port->name)) {
2753 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2754 if (vlan >= 0 && vlan <= 4095) {
2755 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2758 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2759 * they even work as-is. But they have not been tested. */
2760 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2764 if (port->vlan != vlan) {
2766 bridge_flush(port->bridge);
2769 /* Get trunked VLANs. */
2772 size_t n_trunks, n_errors;
2775 trunks = bitmap_allocate(4096);
2776 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2778 for (i = 0; i < n_trunks; i++) {
2779 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2781 bitmap_set1(trunks, trunk);
2787 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2788 port->name, n_trunks);
2790 if (n_errors == n_trunks) {
2792 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2795 bitmap_set_multiple(trunks, 0, 4096, 1);
2798 if (cfg_has("vlan.%s.trunks", port->name)) {
2799 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2800 port->name, port->name);
2804 ? port->trunks != NULL
2805 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2806 bridge_flush(port->bridge);
2808 bitmap_free(port->trunks);
2809 port->trunks = trunks;
2811 svec_destroy(&old_ifaces);
2812 svec_destroy(&new_ifaces);
2816 port_destroy(struct port *port)
2819 struct bridge *br = port->bridge;
2823 proc_net_compat_update_vlan(port->name, NULL, 0);
2825 for (i = 0; i < MAX_MIRRORS; i++) {
2826 struct mirror *m = br->mirrors[i];
2827 if (m && m->out_port == port) {
2832 while (port->n_ifaces > 0) {
2833 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2836 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2837 del->port_idx = port->port_idx;
2840 bitmap_free(port->trunks);
2847 static struct port *
2848 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2850 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2851 return iface ? iface->port : NULL;
2854 static struct port *
2855 port_lookup(const struct bridge *br, const char *name)
2859 for (i = 0; i < br->n_ports; i++) {
2860 struct port *port = br->ports[i];
2861 if (!strcmp(port->name, name)) {
2868 static struct iface *
2869 port_lookup_iface(const struct port *port, const char *name)
2873 for (j = 0; j < port->n_ifaces; j++) {
2874 struct iface *iface = port->ifaces[j];
2875 if (!strcmp(iface->name, name)) {
2883 port_update_bonding(struct port *port)
2885 if (port->n_ifaces < 2) {
2886 /* Not a bonded port. */
2887 if (port->bond_hash) {
2888 free(port->bond_hash);
2889 port->bond_hash = NULL;
2890 proc_net_compat_update_bond(port->name, NULL);
2893 if (!port->bond_hash) {
2896 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2897 for (i = 0; i <= BOND_MASK; i++) {
2898 struct bond_entry *e = &port->bond_hash[i];
2902 port->no_ifaces_tag = tag_create_random();
2903 bond_choose_active_iface(port);
2905 port_update_bond_compat(port);
2910 port_update_bond_compat(struct port *port)
2912 struct compat_bond bond;
2915 if (port->n_ifaces < 2) {
2920 bond.updelay = port->updelay;
2921 bond.downdelay = port->downdelay;
2922 bond.n_slaves = port->n_ifaces;
2923 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2924 for (i = 0; i < port->n_ifaces; i++) {
2925 struct iface *iface = port->ifaces[i];
2926 struct compat_bond_slave *slave = &bond.slaves[i];
2927 slave->name = iface->name;
2928 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2929 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2933 netdev_get_etheraddr(iface->netdev, slave->mac);
2935 proc_net_compat_update_bond(port->name, &bond);
2940 port_update_vlan_compat(struct port *port)
2942 struct bridge *br = port->bridge;
2943 char *vlandev_name = NULL;
2945 if (port->vlan > 0) {
2946 /* Figure out the name that the VLAN device should actually have, if it
2947 * existed. This takes some work because the VLAN device would not
2948 * have port->name in its name; rather, it would have the trunk port's
2949 * name, and 'port' would be attached to a bridge that also had the
2950 * VLAN device one of its ports. So we need to find a trunk port that
2951 * includes port->vlan.
2953 * There might be more than one candidate. This doesn't happen on
2954 * XenServer, so if it happens we just pick the first choice in
2955 * alphabetical order instead of creating multiple VLAN devices. */
2957 for (i = 0; i < br->n_ports; i++) {
2958 struct port *p = br->ports[i];
2959 if (port_trunks_vlan(p, port->vlan)
2961 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2963 uint8_t ea[ETH_ADDR_LEN];
2964 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
2965 if (!eth_addr_is_multicast(ea) &&
2966 !eth_addr_is_reserved(ea) &&
2967 !eth_addr_is_zero(ea)) {
2968 vlandev_name = p->name;
2973 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2976 /* Interface functions. */
2979 iface_create(struct port *port, const char *name)
2981 struct iface *iface;
2983 iface = xcalloc(1, sizeof *iface);
2985 iface->port_ifidx = port->n_ifaces;
2986 iface->name = xstrdup(name);
2987 iface->dp_ifidx = -1;
2988 iface->tag = tag_create_random();
2989 iface->delay_expires = LLONG_MAX;
2990 iface->netdev = NULL;
2992 if (port->n_ifaces >= port->allocated_ifaces) {
2993 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2994 sizeof *port->ifaces);
2996 port->ifaces[port->n_ifaces++] = iface;
2997 if (port->n_ifaces > 1) {
2998 port->bridge->has_bonded_ports = true;
3001 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3003 bridge_flush(port->bridge);
3007 iface_destroy(struct iface *iface)
3010 struct port *port = iface->port;
3011 struct bridge *br = port->bridge;
3012 bool del_active = port->active_iface == iface->port_ifidx;
3015 if (iface->dp_ifidx >= 0) {
3016 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3019 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3020 del->port_ifidx = iface->port_ifidx;
3022 netdev_close(iface->netdev);
3027 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3028 bond_choose_active_iface(port);
3029 bond_send_learning_packets(port);
3032 bridge_flush(port->bridge);
3036 static struct iface *
3037 iface_lookup(const struct bridge *br, const char *name)
3041 for (i = 0; i < br->n_ports; i++) {
3042 struct port *port = br->ports[i];
3043 for (j = 0; j < port->n_ifaces; j++) {
3044 struct iface *iface = port->ifaces[j];
3045 if (!strcmp(iface->name, name)) {
3053 static struct iface *
3054 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3056 return port_array_get(&br->ifaces, dp_ifidx);
3059 /* Port mirroring. */
3062 mirror_reconfigure(struct bridge *br)
3064 struct svec old_mirrors, new_mirrors;
3067 /* Collect old and new mirrors. */
3068 svec_init(&old_mirrors);
3069 svec_init(&new_mirrors);
3070 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3071 for (i = 0; i < MAX_MIRRORS; i++) {
3072 if (br->mirrors[i]) {
3073 svec_add(&old_mirrors, br->mirrors[i]->name);
3077 /* Get rid of deleted mirrors and add new mirrors. */
3078 svec_sort(&old_mirrors);
3079 assert(svec_is_unique(&old_mirrors));
3080 svec_sort(&new_mirrors);
3081 assert(svec_is_unique(&new_mirrors));
3082 for (i = 0; i < MAX_MIRRORS; i++) {
3083 struct mirror *m = br->mirrors[i];
3084 if (m && !svec_contains(&new_mirrors, m->name)) {
3088 for (i = 0; i < new_mirrors.n; i++) {
3089 const char *name = new_mirrors.names[i];
3090 if (!svec_contains(&old_mirrors, name)) {
3091 mirror_create(br, name);
3094 svec_destroy(&old_mirrors);
3095 svec_destroy(&new_mirrors);
3097 /* Reconfigure all mirrors. */
3098 for (i = 0; i < MAX_MIRRORS; i++) {
3099 if (br->mirrors[i]) {
3100 mirror_reconfigure_one(br->mirrors[i]);
3104 /* Update port reserved status. */
3105 for (i = 0; i < br->n_ports; i++) {
3106 br->ports[i]->is_mirror_output_port = false;
3108 for (i = 0; i < MAX_MIRRORS; i++) {
3109 struct mirror *m = br->mirrors[i];
3110 if (m && m->out_port) {
3111 m->out_port->is_mirror_output_port = true;
3117 mirror_create(struct bridge *br, const char *name)
3122 for (i = 0; ; i++) {
3123 if (i >= MAX_MIRRORS) {
3124 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3125 "cannot create %s", br->name, MAX_MIRRORS, name);
3128 if (!br->mirrors[i]) {
3133 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3136 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3139 m->name = xstrdup(name);
3140 svec_init(&m->src_ports);
3141 svec_init(&m->dst_ports);
3149 mirror_destroy(struct mirror *m)
3152 struct bridge *br = m->bridge;
3155 for (i = 0; i < br->n_ports; i++) {
3156 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3157 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3160 svec_destroy(&m->src_ports);
3161 svec_destroy(&m->dst_ports);
3164 m->bridge->mirrors[m->idx] = NULL;
3172 prune_ports(struct mirror *m, struct svec *ports)
3177 svec_sort_unique(ports);
3180 for (i = 0; i < ports->n; i++) {
3181 const char *name = ports->names[i];
3182 if (port_lookup(m->bridge, name)) {
3183 svec_add(&tmp, name);
3185 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3186 m->bridge->name, m->name, name);
3189 svec_swap(ports, &tmp);
3194 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3198 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3199 * order won't give us numeric sort order. But that's good enough for what
3200 * we need right now. */
3201 svec_sort_unique(vlan_strings);
3203 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3205 for (i = 0; i < vlan_strings->n; i++) {
3206 const char *name = vlan_strings->names[i];
3208 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3209 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3210 m->bridge->name, m->name, name);
3212 (*vlans)[n_vlans++] = vlan;
3219 vlan_is_mirrored(const struct mirror *m, int vlan)
3223 for (i = 0; i < m->n_vlans; i++) {
3224 if (m->vlans[i] == vlan) {
3232 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3236 for (i = 0; i < m->n_vlans; i++) {
3237 if (port_trunks_vlan(p, m->vlans[i])) {
3245 mirror_reconfigure_one(struct mirror *m)
3247 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3248 struct svec src_ports, dst_ports, ports;
3249 struct svec vlan_strings;
3250 mirror_mask_t mirror_bit;
3251 const char *out_port_name;
3252 struct port *out_port;
3257 bool mirror_all_ports;
3259 /* Get output port. */
3260 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3261 m->bridge->name, m->name);
3262 if (out_port_name) {
3263 out_port = port_lookup(m->bridge, out_port_name);
3265 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3266 "named %s", pfx, m->bridge->name, out_port_name);
3273 if (cfg_has("%s.output.vlan", pfx)) {
3274 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3275 "ignoring %s.output.vlan", pfx, pfx, pfx);
3277 } else if (cfg_has("%s.output.vlan", pfx)) {
3279 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3281 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3282 "but exactly one is required; disabling port mirror %s",
3283 pfx, pfx, pfx, pfx);
3289 /* Get all the ports, and drop duplicates and ports that don't exist. */
3290 svec_init(&src_ports);
3291 svec_init(&dst_ports);
3293 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3294 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3295 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3296 svec_append(&src_ports, &ports);
3297 svec_append(&dst_ports, &ports);
3298 svec_destroy(&ports);
3299 prune_ports(m, &src_ports);
3300 prune_ports(m, &dst_ports);
3302 /* Get all the vlans, and drop duplicate and invalid vlans. */
3303 svec_init(&vlan_strings);
3304 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3305 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3306 svec_destroy(&vlan_strings);
3308 /* Update mirror data. */
3309 if (!svec_equal(&m->src_ports, &src_ports)
3310 || !svec_equal(&m->dst_ports, &dst_ports)
3311 || m->n_vlans != n_vlans
3312 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3313 || m->out_port != out_port
3314 || m->out_vlan != out_vlan) {
3315 bridge_flush(m->bridge);
3317 svec_swap(&m->src_ports, &src_ports);
3318 svec_swap(&m->dst_ports, &dst_ports);
3321 m->n_vlans = n_vlans;
3322 m->out_port = out_port;
3323 m->out_vlan = out_vlan;
3325 /* If no selection criteria have been given, mirror for all ports. */
3326 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3329 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3330 for (i = 0; i < m->bridge->n_ports; i++) {
3331 struct port *port = m->bridge->ports[i];
3333 if (mirror_all_ports
3334 || svec_contains(&m->src_ports, port->name)
3337 ? port_trunks_any_mirrored_vlan(m, port)
3338 : vlan_is_mirrored(m, port->vlan)))) {
3339 port->src_mirrors |= mirror_bit;
3341 port->src_mirrors &= ~mirror_bit;
3344 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3345 port->dst_mirrors |= mirror_bit;
3347 port->dst_mirrors &= ~mirror_bit;
3352 svec_destroy(&src_ports);
3353 svec_destroy(&dst_ports);
3357 /* Spanning tree protocol. */
3359 static void brstp_update_port_state(struct port *);
3362 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3364 struct bridge *br = br_;
3365 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3366 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3368 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3371 struct eth_header *eth = pkt->l2;
3373 netdev_get_etheraddr(iface->netdev, eth->eth_src);
3374 if (eth_addr_is_zero(eth->eth_src)) {
3375 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
3376 "with unknown MAC", br->name, port_no);
3378 union ofp_action action;
3381 memset(&action, 0, sizeof action);
3382 action.type = htons(OFPAT_OUTPUT);
3383 action.output.len = htons(sizeof action);
3384 action.output.port = htons(port_no);
3386 flow_extract(pkt, ODPP_NONE, &flow);
3387 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3394 brstp_reconfigure(struct bridge *br)
3398 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3400 stp_destroy(br->stp);
3406 uint64_t bridge_address, bridge_id;
3407 int bridge_priority;
3409 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3410 if (!bridge_address) {
3412 bridge_address = (stp_get_bridge_id(br->stp)
3413 & ((UINT64_C(1) << 48) - 1));
3415 uint8_t mac[ETH_ADDR_LEN];
3416 eth_addr_random(mac);
3417 bridge_address = eth_addr_to_uint64(mac);
3421 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3423 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3425 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3428 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3430 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3431 br->stp_last_tick = time_msec();
3434 if (bridge_id != stp_get_bridge_id(br->stp)) {
3435 stp_set_bridge_id(br->stp, bridge_id);
3440 for (i = 0; i < br->n_ports; i++) {
3441 struct port *p = br->ports[i];
3443 struct stp_port *sp;
3444 int path_cost, priority;
3450 dp_ifidx = p->ifaces[0]->dp_ifidx;
3451 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3455 sp = stp_get_port(br->stp, dp_ifidx);
3456 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3457 "stp.%s.port.%s.enabled",
3459 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3460 br->name, p->name));
3461 if (p->is_mirror_output_port) {
3464 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3465 bridge_flush(br); /* Might not be necessary. */
3467 stp_port_enable(sp);
3469 stp_port_disable(sp);
3473 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3475 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3477 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3478 "stp.%s.port.%s.priority",
3480 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3482 : STP_DEFAULT_PORT_PRIORITY);
3483 stp_port_set_priority(sp, priority);
3486 brstp_adjust_timers(br);
3488 for (i = 0; i < br->n_ports; i++) {
3489 brstp_update_port_state(br->ports[i]);
3494 brstp_update_port_state(struct port *p)
3496 struct bridge *br = p->bridge;
3497 enum stp_state state;
3499 /* Figure out new state. */
3500 state = STP_DISABLED;
3501 if (br->stp && p->n_ifaces > 0) {
3502 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3503 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3504 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3509 if (p->stp_state != state) {
3510 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3511 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3512 p->name, stp_state_name(p->stp_state),
3513 stp_state_name(state));
3514 if (p->stp_state == STP_DISABLED) {
3517 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3519 p->stp_state = state;
3520 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3521 : tag_create_random());
3526 brstp_adjust_timers(struct bridge *br)
3528 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3529 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3530 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3532 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3533 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3534 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3538 brstp_run(struct bridge *br)
3541 long long int now = time_msec();
3542 long long int elapsed = now - br->stp_last_tick;
3543 struct stp_port *sp;
3546 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3547 br->stp_last_tick = now;
3549 while (stp_get_changed_port(br->stp, &sp)) {
3550 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3552 brstp_update_port_state(p);
3559 brstp_wait(struct bridge *br)
3562 poll_timer_wait(1000);