1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id;
74 /* These members are always valid. */
75 struct port *port; /* Containing port. */
76 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 tag_type tag; /* Tag associated with this interface. */
79 long long delay_expires; /* Time after which 'enabled' may change. */
81 /* These members are valid only after bridge_reconfigure() causes them to
83 int dp_ifidx; /* Index within kernel datapath. */
84 struct netdev *netdev; /* Network device. */
85 bool enabled; /* May be chosen for flows? */
88 #define BOND_MASK 0xff
90 int iface_idx; /* Index of assigned iface, or -1 if none. */
91 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
92 tag_type iface_tag; /* Tag associated with iface_idx. */
95 #define MAX_MIRRORS 32
96 typedef uint32_t mirror_mask_t;
97 #define MIRROR_MASK_C(X) UINT32_C(X)
98 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
100 struct bridge *bridge;
104 /* Selection criteria. */
105 struct svec src_ports;
106 struct svec dst_ports;
111 struct port *out_port;
115 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
117 struct bridge *bridge;
119 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
120 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
123 /* An ordinary bridge port has 1 interface.
124 * A bridge port for bonding has at least 2 interfaces. */
125 struct iface **ifaces;
126 size_t n_ifaces, allocated_ifaces;
129 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
130 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
131 tag_type active_iface_tag; /* Tag for bcast flows. */
132 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
133 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
135 /* Port mirroring info. */
136 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
137 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
138 bool is_mirror_output_port; /* Does port mirroring send frames here? */
140 /* Spanning tree info. */
141 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
142 tag_type stp_state_tag; /* Tag for STP state change. */
145 #define DP_MAX_PORTS 255
147 struct list node; /* Node in global list of bridges. */
148 char *name; /* User-specified arbitrary name. */
149 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
150 bool sent_config_request; /* Successfully sent config request? */
151 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
153 /* Support for remote controllers. */
154 char *controller; /* NULL if there is no remote controller;
155 * "discover" to do controller discovery;
156 * otherwise a vconn name. */
158 /* OpenFlow switch processing. */
159 struct ofproto *ofproto; /* OpenFlow switch. */
161 /* Kernel datapath information. */
162 struct dpif *dpif; /* Datapath. */
163 struct port_array ifaces; /* Indexed by kernel datapath port number. */
167 size_t n_ports, allocated_ports;
170 bool has_bonded_ports;
171 long long int bond_next_rebalance;
176 /* Flow statistics gathering. */
177 time_t next_stats_request;
179 /* Port mirroring. */
180 struct mirror *mirrors[MAX_MIRRORS];
184 long long int stp_last_tick;
187 /* List of all bridges. */
188 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
190 /* Maximum number of datapaths. */
191 enum { DP_MAX = 256 };
193 static struct bridge *bridge_create(const char *name);
194 static void bridge_destroy(struct bridge *);
195 static struct bridge *bridge_lookup(const char *name);
196 static int bridge_run_one(struct bridge *);
197 static void bridge_reconfigure_one(struct bridge *);
198 static void bridge_reconfigure_controller(struct bridge *);
199 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
200 static void bridge_fetch_dp_ifaces(struct bridge *);
201 static void bridge_flush(struct bridge *);
202 static void bridge_pick_local_hw_addr(struct bridge *,
203 uint8_t ea[ETH_ADDR_LEN],
204 const char **devname);
205 static uint64_t bridge_pick_datapath_id(struct bridge *,
206 const uint8_t bridge_ea[ETH_ADDR_LEN],
207 const char *devname);
208 static uint64_t dpid_from_hash(const void *, size_t nbytes);
210 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
212 static void bond_init(void);
213 static void bond_run(struct bridge *);
214 static void bond_wait(struct bridge *);
215 static void bond_rebalance_port(struct port *);
216 static void bond_send_learning_packets(struct port *);
218 static void port_create(struct bridge *, const char *name);
219 static void port_reconfigure(struct port *);
220 static void port_destroy(struct port *);
221 static struct port *port_lookup(const struct bridge *, const char *name);
222 static struct iface *port_lookup_iface(const struct port *, const char *name);
223 static struct port *port_from_dp_ifidx(const struct bridge *,
225 static void port_update_bond_compat(struct port *);
226 static void port_update_vlan_compat(struct port *);
227 static void port_update_bonding(struct port *);
229 static void mirror_create(struct bridge *, const char *name);
230 static void mirror_destroy(struct mirror *);
231 static void mirror_reconfigure(struct bridge *);
232 static void mirror_reconfigure_one(struct mirror *);
233 static bool vlan_is_mirrored(const struct mirror *, int vlan);
235 static void brstp_reconfigure(struct bridge *);
236 static void brstp_adjust_timers(struct bridge *);
237 static void brstp_run(struct bridge *);
238 static void brstp_wait(struct bridge *);
240 static void iface_create(struct port *, const char *name);
241 static void iface_destroy(struct iface *);
242 static struct iface *iface_lookup(const struct bridge *, const char *name);
243 static struct iface *iface_from_dp_ifidx(const struct bridge *,
246 /* Hooks into ofproto processing. */
247 static struct ofhooks bridge_ofhooks;
249 /* Public functions. */
251 /* Adds the name of each interface used by a bridge, including local and
252 * internal ports, to 'svec'. */
254 bridge_get_ifaces(struct svec *svec)
256 struct bridge *br, *next;
259 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
260 for (i = 0; i < br->n_ports; i++) {
261 struct port *port = br->ports[i];
263 for (j = 0; j < port->n_ifaces; j++) {
264 struct iface *iface = port->ifaces[j];
265 if (iface->dp_ifidx < 0) {
266 VLOG_ERR("%s interface not in datapath %s, ignoring",
267 iface->name, dpif_name(br->dpif));
269 if (iface->dp_ifidx != ODPP_LOCAL) {
270 svec_add(svec, iface->name);
278 /* The caller must already have called cfg_read(). */
282 struct svec dpif_names;
285 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
287 dp_enumerate(&dpif_names);
288 for (i = 0; i < dpif_names.n; i++) {
289 const char *dpif_name = dpif_names.names[i];
293 retval = dpif_open(dpif_name, &dpif);
295 struct svec all_names;
298 svec_init(&all_names);
299 dpif_get_all_names(dpif, &all_names);
300 for (j = 0; j < all_names.n; j++) {
301 if (cfg_has("bridge.%s.port", all_names.names[j])) {
307 svec_destroy(&all_names);
313 bridge_reconfigure();
318 config_string_change(const char *key, char **valuep)
320 const char *value = cfg_get_string(0, "%s", key);
321 if (value && (!*valuep || strcmp(value, *valuep))) {
323 *valuep = xstrdup(value);
331 bridge_configure_ssl(void)
333 /* XXX SSL should be configurable on a per-bridge basis.
334 * XXX should be possible to de-configure SSL. */
335 static char *private_key_file;
336 static char *certificate_file;
337 static char *cacert_file;
340 if (config_string_change("ssl.private-key", &private_key_file)) {
341 vconn_ssl_set_private_key_file(private_key_file);
344 if (config_string_change("ssl.certificate", &certificate_file)) {
345 vconn_ssl_set_certificate_file(certificate_file);
348 /* We assume that even if the filename hasn't changed, if the CA cert
349 * file has been removed, that we want to move back into
350 * boot-strapping mode. This opens a small security hole, because
351 * the old certificate will still be trusted until vSwitch is
352 * restarted. We may want to address this in vconn's SSL library. */
353 if (config_string_change("ssl.ca-cert", &cacert_file)
354 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
355 vconn_ssl_set_ca_cert_file(cacert_file,
356 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
361 /* iterate_and_prune_ifaces() callback function that opens the network device
362 * for 'iface', if it is not already open, and retrieves the interface's MAC
363 * address and carrier status. */
365 init_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
370 } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE,
372 netdev_get_carrier(iface->netdev, &iface->enabled);
375 /* If the network device can't be opened, then we're not going to try
376 * to do anything with this interface. */
382 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
385 struct iface **local_ifacep = local_ifacep_;
387 if (iface->dp_ifidx >= 0) {
388 if (iface->dp_ifidx == ODPP_LOCAL) {
389 *local_ifacep = iface;
391 VLOG_DBG("%s has interface %s on port %d",
393 iface->name, iface->dp_ifidx);
396 VLOG_ERR("%s interface not in %s, dropping",
397 iface->name, dpif_name(br->dpif));
403 set_iface_policing(struct bridge *br UNUSED, struct iface *iface,
406 int rate = cfg_get_int(0, "port.%s.ingress.policing-rate", iface->name);
407 int burst = cfg_get_int(0, "port.%s.ingress.policing-burst", iface->name);
408 netdev_set_policing(iface->netdev, rate, burst);
412 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
413 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
414 * deletes from 'br' any ports that no longer have any interfaces. */
416 iterate_and_prune_ifaces(struct bridge *br,
417 bool (*cb)(struct bridge *, struct iface *,
423 for (i = 0; i < br->n_ports; ) {
424 struct port *port = br->ports[i];
425 for (j = 0; j < port->n_ifaces; ) {
426 struct iface *iface = port->ifaces[j];
427 if (cb(br, iface, aux)) {
430 iface_destroy(iface);
434 if (port->n_ifaces) {
437 VLOG_ERR("%s port has no interfaces, dropping", port->name);
444 bridge_reconfigure(void)
446 struct svec old_br, new_br;
447 struct bridge *br, *next;
450 COVERAGE_INC(bridge_reconfigure);
452 /* Collect old and new bridges. */
455 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
456 svec_add(&old_br, br->name);
458 cfg_get_subsections(&new_br, "bridge");
460 /* Get rid of deleted bridges and add new bridges. */
463 assert(svec_is_unique(&old_br));
464 assert(svec_is_unique(&new_br));
465 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
466 if (!svec_contains(&new_br, br->name)) {
470 for (i = 0; i < new_br.n; i++) {
471 const char *name = new_br.names[i];
472 if (!svec_contains(&old_br, name)) {
476 svec_destroy(&old_br);
477 svec_destroy(&new_br);
481 bridge_configure_ssl();
484 /* Reconfigure all bridges. */
485 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
486 bridge_reconfigure_one(br);
489 /* Add and delete ports on all datapaths.
491 * The kernel will reject any attempt to add a given port to a datapath if
492 * that port already belongs to a different datapath, so we must do all
493 * port deletions before any port additions. */
494 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
495 struct odp_port *dpif_ports;
497 struct svec want_ifaces;
499 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
500 bridge_get_all_ifaces(br, &want_ifaces);
501 for (i = 0; i < n_dpif_ports; i++) {
502 const struct odp_port *p = &dpif_ports[i];
503 if (!svec_contains(&want_ifaces, p->devname)
504 && strcmp(p->devname, br->name)) {
505 int retval = dpif_port_del(br->dpif, p->port);
507 VLOG_ERR("failed to remove %s interface from %s: %s",
508 p->devname, dpif_name(br->dpif),
513 svec_destroy(&want_ifaces);
516 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
517 struct odp_port *dpif_ports;
519 struct svec cur_ifaces, want_ifaces, add_ifaces;
521 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
522 svec_init(&cur_ifaces);
523 for (i = 0; i < n_dpif_ports; i++) {
524 svec_add(&cur_ifaces, dpif_ports[i].devname);
527 svec_sort_unique(&cur_ifaces);
528 bridge_get_all_ifaces(br, &want_ifaces);
529 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
531 for (i = 0; i < add_ifaces.n; i++) {
532 const char *if_name = add_ifaces.names[i];
533 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
534 int flags = internal ? ODP_PORT_INTERNAL : 0;
535 int error = dpif_port_add(br->dpif, if_name, flags, NULL);
536 if (error == EXFULL) {
537 VLOG_ERR("ran out of valid port numbers on %s",
538 dpif_name(br->dpif));
541 VLOG_ERR("failed to add %s interface to %s: %s",
542 if_name, dpif_name(br->dpif), strerror(error));
545 svec_destroy(&cur_ifaces);
546 svec_destroy(&want_ifaces);
547 svec_destroy(&add_ifaces);
549 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
552 struct iface *local_iface = NULL;
554 uint8_t engine_type, engine_id;
555 bool add_id_to_iface = false;
556 struct svec nf_hosts;
558 bridge_fetch_dp_ifaces(br);
559 iterate_and_prune_ifaces(br, init_iface_netdev, NULL);
562 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, &local_iface);
564 /* Pick local port hardware address, datapath ID. */
565 bridge_pick_local_hw_addr(br, ea, &devname);
567 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
569 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
570 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
571 "Ethernet address: %s",
572 br->name, strerror(error));
576 dpid = bridge_pick_datapath_id(br, ea, devname);
577 ofproto_set_datapath_id(br->ofproto, dpid);
579 /* Set NetFlow configuration on this bridge. */
580 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
581 if (cfg_has("netflow.%s.engine-type", br->name)) {
582 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
585 if (cfg_has("netflow.%s.engine-id", br->name)) {
586 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
588 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
589 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
592 if (add_id_to_iface && engine_id > 0x7f) {
593 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
594 "another vswitch, choose an engine id less than 128",
597 if (add_id_to_iface && br->n_ports > 0x1ff) {
598 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
599 "another port when 512 or more ports are used",
602 svec_init(&nf_hosts);
603 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
604 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
605 engine_id, add_id_to_iface)) {
606 VLOG_ERR("bridge %s: problem setting netflow collectors",
610 /* Update the controller and related settings. It would be more
611 * straightforward to call this from bridge_reconfigure_one(), but we
612 * can't do it there for two reasons. First, and most importantly, at
613 * that point we don't know the dp_ifidx of any interfaces that have
614 * been added to the bridge (because we haven't actually added them to
615 * the datapath). Second, at that point we haven't set the datapath ID
616 * yet; when a controller is configured, resetting the datapath ID will
617 * immediately disconnect from the controller, so it's better to set
618 * the datapath ID before the controller. */
619 bridge_reconfigure_controller(br);
621 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
622 for (i = 0; i < br->n_ports; i++) {
623 struct port *port = br->ports[i];
624 port_update_vlan_compat(port);
625 port_update_bonding(port);
628 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
629 brstp_reconfigure(br);
630 iterate_and_prune_ifaces(br, set_iface_policing, NULL);
635 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
636 const char **devname)
638 uint64_t requested_ea;
644 /* Did the user request a particular MAC? */
645 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
647 eth_addr_from_uint64(requested_ea, ea);
648 if (eth_addr_is_multicast(ea)) {
649 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
650 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
651 } else if (eth_addr_is_zero(ea)) {
652 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
658 /* Otherwise choose the minimum MAC address among all of the interfaces.
659 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
660 * MAC of the physical interface in such an environment.) */
661 memset(ea, 0xff, sizeof ea);
662 for (i = 0; i < br->n_ports; i++) {
663 struct port *port = br->ports[i];
664 if (port->is_mirror_output_port) {
667 for (j = 0; j < port->n_ifaces; j++) {
668 struct iface *iface = port->ifaces[j];
669 uint8_t iface_ea[ETH_ADDR_LEN];
670 if (iface->dp_ifidx == ODPP_LOCAL
671 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
674 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
676 if (!eth_addr_is_multicast(iface_ea) &&
677 !eth_addr_is_reserved(iface_ea) &&
678 !eth_addr_is_zero(iface_ea) &&
679 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
680 memcpy(ea, iface_ea, ETH_ADDR_LEN);
681 *devname = iface->name;
684 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
685 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
686 iface->name, strerror(error));
690 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
691 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
693 VLOG_WARN("bridge %s: using default bridge Ethernet "
694 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
696 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
697 br->name, ETH_ADDR_ARGS(ea));
701 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
702 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
703 * a network device, then that network device's name must be passed in as
704 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
705 * passed in as a null pointer. */
707 bridge_pick_datapath_id(struct bridge *br,
708 const uint8_t bridge_ea[ETH_ADDR_LEN],
712 * The procedure for choosing a bridge MAC address will, in the most
713 * ordinary case, also choose a unique MAC that we can use as a datapath
714 * ID. In some special cases, though, multiple bridges will end up with
715 * the same MAC address. This is OK for the bridges, but it will confuse
716 * the OpenFlow controller, because each datapath needs a unique datapath
719 * Datapath IDs must be unique. It is also very desirable that they be
720 * stable from one run to the next, so that policy set on a datapath
725 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
732 if (!netdev_get_vlan_vid(devname, &vlan)) {
734 * A bridge whose MAC address is taken from a VLAN network device
735 * (that is, a network device created with vconfig(8) or similar
736 * tool) will have the same MAC address as a bridge on the VLAN
737 * device's physical network device.
739 * Handle this case by hashing the physical network device MAC
740 * along with the VLAN identifier.
742 uint8_t buf[ETH_ADDR_LEN + 2];
743 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
744 buf[ETH_ADDR_LEN] = vlan >> 8;
745 buf[ETH_ADDR_LEN + 1] = vlan;
746 return dpid_from_hash(buf, sizeof buf);
749 * Assume that this bridge's MAC address is unique, since it
750 * doesn't fit any of the cases we handle specially.
755 * A purely internal bridge, that is, one that has no non-virtual
756 * network devices on it at all, is more difficult because it has no
757 * natural unique identifier at all.
759 * When the host is a XenServer, we handle this case by hashing the
760 * host's UUID with the name of the bridge. Names of bridges are
761 * persistent across XenServer reboots, although they can be reused if
762 * an internal network is destroyed and then a new one is later
763 * created, so this is fairly effective.
765 * When the host is not a XenServer, we punt by using a random MAC
766 * address on each run.
768 const char *host_uuid = xenserver_get_host_uuid();
770 char *combined = xasprintf("%s,%s", host_uuid, br->name);
771 dpid = dpid_from_hash(combined, strlen(combined));
777 return eth_addr_to_uint64(bridge_ea);
781 dpid_from_hash(const void *data, size_t n)
783 uint8_t hash[SHA1_DIGEST_SIZE];
785 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
786 sha1_bytes(data, n, hash);
787 eth_addr_mark_random(hash);
788 return eth_addr_to_uint64(hash);
794 struct bridge *br, *next;
798 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
799 int error = bridge_run_one(br);
801 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
802 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
803 "forcing reconfiguration", br->name);
817 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
818 ofproto_wait(br->ofproto);
819 if (br->controller) {
824 mac_learning_wait(br->ml);
831 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
832 * configuration changes. */
834 bridge_flush(struct bridge *br)
836 COVERAGE_INC(bridge_flush);
839 mac_learning_flush(br->ml);
843 /* Bridge unixctl user interface functions. */
845 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
847 struct ds ds = DS_EMPTY_INITIALIZER;
848 const struct bridge *br;
850 br = bridge_lookup(args);
852 unixctl_command_reply(conn, 501, "no such bridge");
856 ds_put_cstr(&ds, " port VLAN MAC Age\n");
858 const struct mac_entry *e;
859 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
860 if (e->port < 0 || e->port >= br->n_ports) {
863 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
864 br->ports[e->port]->ifaces[0]->dp_ifidx,
865 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
868 unixctl_command_reply(conn, 200, ds_cstr(&ds));
872 /* Bridge reconfiguration functions. */
874 static struct bridge *
875 bridge_create(const char *name)
880 assert(!bridge_lookup(name));
881 br = xcalloc(1, sizeof *br);
883 error = dpif_create(name, &br->dpif);
884 if (error == EEXIST || error == EBUSY) {
885 error = dpif_open(name, &br->dpif);
887 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
888 name, strerror(error));
892 dpif_flow_flush(br->dpif);
894 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
899 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
901 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
902 dpif_delete(br->dpif);
903 dpif_close(br->dpif);
908 br->name = xstrdup(name);
909 br->ml = mac_learning_create();
910 br->sent_config_request = false;
911 eth_addr_random(br->default_ea);
913 port_array_init(&br->ifaces);
916 br->bond_next_rebalance = time_msec() + 10000;
918 list_push_back(&all_bridges, &br->node);
920 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
926 bridge_destroy(struct bridge *br)
931 while (br->n_ports > 0) {
932 port_destroy(br->ports[br->n_ports - 1]);
934 list_remove(&br->node);
935 error = dpif_delete(br->dpif);
936 if (error && error != ENOENT) {
937 VLOG_ERR("failed to delete %s: %s",
938 dpif_name(br->dpif), strerror(error));
940 dpif_close(br->dpif);
941 ofproto_destroy(br->ofproto);
942 free(br->controller);
943 mac_learning_destroy(br->ml);
944 port_array_destroy(&br->ifaces);
951 static struct bridge *
952 bridge_lookup(const char *name)
956 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
957 if (!strcmp(br->name, name)) {
965 bridge_exists(const char *name)
967 return bridge_lookup(name) ? true : false;
971 bridge_get_datapathid(const char *name)
973 struct bridge *br = bridge_lookup(name);
974 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
978 bridge_run_one(struct bridge *br)
982 error = ofproto_run1(br->ofproto);
988 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
993 error = ofproto_run2(br->ofproto, br->flush);
1000 bridge_get_controller(const struct bridge *br)
1002 const char *controller;
1004 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1006 controller = cfg_get_string(0, "mgmt.controller");
1008 return controller && controller[0] ? controller : NULL;
1012 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1014 struct svec *ifaces = ifaces_;
1015 if (!svec_contains(ifaces, iface->name)) {
1016 svec_add(ifaces, iface->name);
1020 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1022 br->name, iface->name, iface->port->name);
1028 bridge_reconfigure_one(struct bridge *br)
1030 struct svec old_ports, new_ports, ifaces;
1031 struct svec listeners, old_listeners;
1032 struct svec snoops, old_snoops;
1035 /* Collect old ports. */
1036 svec_init(&old_ports);
1037 for (i = 0; i < br->n_ports; i++) {
1038 svec_add(&old_ports, br->ports[i]->name);
1040 svec_sort(&old_ports);
1041 assert(svec_is_unique(&old_ports));
1043 /* Collect new ports. */
1044 svec_init(&new_ports);
1045 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1046 svec_sort(&new_ports);
1047 if (bridge_get_controller(br)) {
1048 char local_name[IF_NAMESIZE];
1051 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1052 local_name, sizeof local_name);
1053 if (!error && !svec_contains(&new_ports, local_name)) {
1054 svec_add(&new_ports, local_name);
1055 svec_sort(&new_ports);
1058 if (!svec_is_unique(&new_ports)) {
1059 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1060 br->name, svec_get_duplicate(&new_ports));
1061 svec_unique(&new_ports);
1064 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1066 /* Get rid of deleted ports and add new ports. */
1067 for (i = 0; i < br->n_ports; ) {
1068 struct port *port = br->ports[i];
1069 if (!svec_contains(&new_ports, port->name)) {
1075 for (i = 0; i < new_ports.n; i++) {
1076 const char *name = new_ports.names[i];
1077 if (!svec_contains(&old_ports, name)) {
1078 port_create(br, name);
1081 svec_destroy(&old_ports);
1082 svec_destroy(&new_ports);
1084 /* Reconfigure all ports. */
1085 for (i = 0; i < br->n_ports; i++) {
1086 port_reconfigure(br->ports[i]);
1089 /* Check and delete duplicate interfaces. */
1091 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1092 svec_destroy(&ifaces);
1094 /* Delete all flows if we're switching from connected to standalone or vice
1095 * versa. (XXX Should we delete all flows if we are switching from one
1096 * controller to another?) */
1098 /* Configure OpenFlow management listeners. */
1099 svec_init(&listeners);
1100 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1102 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1103 ovs_rundir, br->name));
1104 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1105 svec_clear(&listeners);
1107 svec_sort_unique(&listeners);
1109 svec_init(&old_listeners);
1110 ofproto_get_listeners(br->ofproto, &old_listeners);
1111 svec_sort_unique(&old_listeners);
1113 if (!svec_equal(&listeners, &old_listeners)) {
1114 ofproto_set_listeners(br->ofproto, &listeners);
1116 svec_destroy(&listeners);
1117 svec_destroy(&old_listeners);
1119 /* Configure OpenFlow controller connection snooping. */
1121 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1123 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1124 ovs_rundir, br->name));
1125 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1126 svec_clear(&snoops);
1128 svec_sort_unique(&snoops);
1130 svec_init(&old_snoops);
1131 ofproto_get_snoops(br->ofproto, &old_snoops);
1132 svec_sort_unique(&old_snoops);
1134 if (!svec_equal(&snoops, &old_snoops)) {
1135 ofproto_set_snoops(br->ofproto, &snoops);
1137 svec_destroy(&snoops);
1138 svec_destroy(&old_snoops);
1140 mirror_reconfigure(br);
1144 bridge_reconfigure_controller(struct bridge *br)
1146 char *pfx = xasprintf("bridge.%s.controller", br->name);
1147 const char *controller;
1149 controller = bridge_get_controller(br);
1150 if ((br->controller != NULL) != (controller != NULL)) {
1151 ofproto_flush_flows(br->ofproto);
1153 free(br->controller);
1154 br->controller = controller ? xstrdup(controller) : NULL;
1157 const char *fail_mode;
1158 int max_backoff, probe;
1159 int rate_limit, burst_limit;
1161 if (!strcmp(controller, "discover")) {
1162 bool update_resolv_conf = true;
1164 if (cfg_has("%s.update-resolv.conf", pfx)) {
1165 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1168 ofproto_set_discovery(br->ofproto, true,
1169 cfg_get_string(0, "%s.accept-regex", pfx),
1170 update_resolv_conf);
1172 char local_name[IF_NAMESIZE];
1173 struct netdev *netdev;
1177 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1179 || cfg_get_bool(0, "%s.in-band", pfx));
1180 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1181 ofproto_set_in_band(br->ofproto, in_band);
1183 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1184 local_name, sizeof local_name);
1186 error = netdev_open(local_name, NETDEV_ETH_TYPE_NONE, &netdev);
1189 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1190 struct in_addr ip, mask, gateway;
1191 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1192 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1193 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1195 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1197 mask.s_addr = guess_netmask(ip.s_addr);
1199 if (!netdev_set_in4(netdev, ip, mask)) {
1200 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1202 br->name, IP_ARGS(&ip.s_addr),
1203 IP_ARGS(&mask.s_addr));
1206 if (gateway.s_addr) {
1207 if (!netdev_add_router(netdev, gateway)) {
1208 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1209 br->name, IP_ARGS(&gateway.s_addr));
1213 netdev_close(netdev);
1217 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1219 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1221 ofproto_set_failure(br->ofproto,
1223 || !strcmp(fail_mode, "standalone")
1224 || !strcmp(fail_mode, "open")));
1226 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1228 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1233 ofproto_set_probe_interval(br->ofproto, probe);
1235 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1237 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1242 ofproto_set_max_backoff(br->ofproto, max_backoff);
1244 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1246 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1248 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1250 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1252 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1254 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1256 if (cfg_has("%s.commands.acl", pfx)) {
1257 struct svec command_acls;
1260 svec_init(&command_acls);
1261 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1262 command_acl = svec_join(&command_acls, ",", "");
1264 ofproto_set_remote_execution(br->ofproto, command_acl,
1265 cfg_get_string(0, "%s.commands.dir",
1268 svec_destroy(&command_acls);
1271 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1274 union ofp_action action;
1277 /* Set up a flow that matches every packet and directs them to
1278 * OFPP_NORMAL (which goes to us). */
1279 memset(&action, 0, sizeof action);
1280 action.type = htons(OFPAT_OUTPUT);
1281 action.output.len = htons(sizeof action);
1282 action.output.port = htons(OFPP_NORMAL);
1283 memset(&flow, 0, sizeof flow);
1284 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1287 ofproto_set_in_band(br->ofproto, false);
1288 ofproto_set_max_backoff(br->ofproto, 1);
1289 ofproto_set_probe_interval(br->ofproto, 5);
1290 ofproto_set_failure(br->ofproto, false);
1291 ofproto_set_stp(br->ofproto, false);
1295 ofproto_set_controller(br->ofproto, br->controller);
1299 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1304 for (i = 0; i < br->n_ports; i++) {
1305 struct port *port = br->ports[i];
1306 for (j = 0; j < port->n_ifaces; j++) {
1307 struct iface *iface = port->ifaces[j];
1308 svec_add(ifaces, iface->name);
1312 assert(svec_is_unique(ifaces));
1315 /* For robustness, in case the administrator moves around datapath ports behind
1316 * our back, we re-check all the datapath port numbers here.
1318 * This function will set the 'dp_ifidx' members of interfaces that have
1319 * disappeared to -1, so only call this function from a context where those
1320 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1321 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1322 * datapath, which doesn't support UINT16_MAX+1 ports. */
1324 bridge_fetch_dp_ifaces(struct bridge *br)
1326 struct odp_port *dpif_ports;
1327 size_t n_dpif_ports;
1330 /* Reset all interface numbers. */
1331 for (i = 0; i < br->n_ports; i++) {
1332 struct port *port = br->ports[i];
1333 for (j = 0; j < port->n_ifaces; j++) {
1334 struct iface *iface = port->ifaces[j];
1335 iface->dp_ifidx = -1;
1338 port_array_clear(&br->ifaces);
1340 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1341 for (i = 0; i < n_dpif_ports; i++) {
1342 struct odp_port *p = &dpif_ports[i];
1343 struct iface *iface = iface_lookup(br, p->devname);
1345 if (iface->dp_ifidx >= 0) {
1346 VLOG_WARN("%s reported interface %s twice",
1347 dpif_name(br->dpif), p->devname);
1348 } else if (iface_from_dp_ifidx(br, p->port)) {
1349 VLOG_WARN("%s reported interface %"PRIu16" twice",
1350 dpif_name(br->dpif), p->port);
1352 port_array_set(&br->ifaces, p->port, iface);
1353 iface->dp_ifidx = p->port;
1360 /* Bridge packet processing functions. */
1363 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1365 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1368 static struct bond_entry *
1369 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1371 return &port->bond_hash[bond_hash(mac)];
1375 bond_choose_iface(const struct port *port)
1378 for (i = 0; i < port->n_ifaces; i++) {
1379 if (port->ifaces[i]->enabled) {
1387 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1388 uint16_t *dp_ifidx, tag_type *tags)
1390 struct iface *iface;
1392 assert(port->n_ifaces);
1393 if (port->n_ifaces == 1) {
1394 iface = port->ifaces[0];
1396 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1397 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1398 || !port->ifaces[e->iface_idx]->enabled) {
1399 /* XXX select interface properly. The current interface selection
1400 * is only good for testing the rebalancing code. */
1401 e->iface_idx = bond_choose_iface(port);
1402 if (e->iface_idx < 0) {
1403 *tags |= port->no_ifaces_tag;
1406 e->iface_tag = tag_create_random();
1408 *tags |= e->iface_tag;
1409 iface = port->ifaces[e->iface_idx];
1411 *dp_ifidx = iface->dp_ifidx;
1412 *tags |= iface->tag; /* Currently only used for bonding. */
1417 bond_link_status_update(struct iface *iface, bool carrier)
1419 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1420 struct port *port = iface->port;
1422 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1423 /* Nothing to do. */
1426 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1427 iface->name, carrier ? "detected" : "dropped");
1428 if (carrier == iface->enabled) {
1429 iface->delay_expires = LLONG_MAX;
1430 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1431 iface->name, carrier ? "disabled" : "enabled");
1432 } else if (carrier && port->updelay && port->active_iface < 0) {
1433 iface->delay_expires = time_msec();
1434 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1435 "other interface is up", iface->name, port->updelay);
1437 int delay = carrier ? port->updelay : port->downdelay;
1438 iface->delay_expires = time_msec() + delay;
1441 "interface %s: will be %s if it stays %s for %d ms",
1443 carrier ? "enabled" : "disabled",
1444 carrier ? "up" : "down",
1451 bond_choose_active_iface(struct port *port)
1453 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1455 port->active_iface = bond_choose_iface(port);
1456 port->active_iface_tag = tag_create_random();
1457 if (port->active_iface >= 0) {
1458 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1459 port->name, port->ifaces[port->active_iface]->name);
1461 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1467 bond_enable_slave(struct iface *iface, bool enable)
1469 struct port *port = iface->port;
1470 struct bridge *br = port->bridge;
1472 iface->delay_expires = LLONG_MAX;
1473 if (enable == iface->enabled) {
1477 iface->enabled = enable;
1478 if (!iface->enabled) {
1479 VLOG_WARN("interface %s: disabled", iface->name);
1480 ofproto_revalidate(br->ofproto, iface->tag);
1481 if (iface->port_ifidx == port->active_iface) {
1482 ofproto_revalidate(br->ofproto,
1483 port->active_iface_tag);
1484 bond_choose_active_iface(port);
1486 bond_send_learning_packets(port);
1488 VLOG_WARN("interface %s: enabled", iface->name);
1489 if (port->active_iface < 0) {
1490 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1491 bond_choose_active_iface(port);
1492 bond_send_learning_packets(port);
1494 iface->tag = tag_create_random();
1499 bond_run(struct bridge *br)
1503 for (i = 0; i < br->n_ports; i++) {
1504 struct port *port = br->ports[i];
1505 if (port->n_ifaces < 2) {
1508 for (j = 0; j < port->n_ifaces; j++) {
1509 struct iface *iface = port->ifaces[j];
1510 if (time_msec() >= iface->delay_expires) {
1511 bond_enable_slave(iface, !iface->enabled);
1518 bond_wait(struct bridge *br)
1522 for (i = 0; i < br->n_ports; i++) {
1523 struct port *port = br->ports[i];
1524 if (port->n_ifaces < 2) {
1527 for (j = 0; j < port->n_ifaces; j++) {
1528 struct iface *iface = port->ifaces[j];
1529 if (iface->delay_expires != LLONG_MAX) {
1530 poll_timer_wait(iface->delay_expires - time_msec());
1537 set_dst(struct dst *p, const flow_t *flow,
1538 const struct port *in_port, const struct port *out_port,
1543 * XXX This uses too many tags: any broadcast flow will get one tag per
1544 * destination port, and thus a broadcast on a switch of any size is likely
1545 * to have all tag bits set. We should figure out a way to be smarter.
1547 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1548 *tags |= out_port->stp_state_tag;
1549 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1553 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1554 : in_port->vlan >= 0 ? in_port->vlan
1555 : ntohs(flow->dl_vlan));
1556 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1560 swap_dst(struct dst *p, struct dst *q)
1562 struct dst tmp = *p;
1567 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1568 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1569 * that we push to the datapath. We could in fact fully sort the array by
1570 * vlan, but in most cases there are at most two different vlan tags so that's
1571 * possibly overkill.) */
1573 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1575 struct dst *first = dsts;
1576 struct dst *last = dsts + n_dsts;
1578 while (first != last) {
1580 * - All dsts < first have vlan == 'vlan'.
1581 * - All dsts >= last have vlan != 'vlan'.
1582 * - first < last. */
1583 while (first->vlan == vlan) {
1584 if (++first == last) {
1589 /* Same invariants, plus one additional:
1590 * - first->vlan != vlan.
1592 while (last[-1].vlan != vlan) {
1593 if (--last == first) {
1598 /* Same invariants, plus one additional:
1599 * - last[-1].vlan == vlan.*/
1600 swap_dst(first++, --last);
1605 mirror_mask_ffs(mirror_mask_t mask)
1607 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1612 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1613 const struct dst *test)
1616 for (i = 0; i < n_dsts; i++) {
1617 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1625 port_trunks_vlan(const struct port *port, uint16_t vlan)
1627 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1631 port_includes_vlan(const struct port *port, uint16_t vlan)
1633 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1637 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1638 const struct port *in_port, const struct port *out_port,
1639 struct dst dsts[], tag_type *tags)
1641 mirror_mask_t mirrors = in_port->src_mirrors;
1642 struct dst *dst = dsts;
1645 *tags |= in_port->stp_state_tag;
1646 if (out_port == FLOOD_PORT) {
1647 /* XXX use ODP_FLOOD if no vlans or bonding. */
1648 /* XXX even better, define each VLAN as a datapath port group */
1649 for (i = 0; i < br->n_ports; i++) {
1650 struct port *port = br->ports[i];
1651 if (port != in_port && port_includes_vlan(port, vlan)
1652 && !port->is_mirror_output_port
1653 && set_dst(dst, flow, in_port, port, tags)) {
1654 mirrors |= port->dst_mirrors;
1658 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1659 mirrors |= out_port->dst_mirrors;
1664 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1665 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1667 if (set_dst(dst, flow, in_port, m->out_port, tags)
1668 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1672 for (i = 0; i < br->n_ports; i++) {
1673 struct port *port = br->ports[i];
1674 if (port_includes_vlan(port, m->out_vlan)
1675 && set_dst(dst, flow, in_port, port, tags)
1676 && !dst_is_duplicate(dsts, dst - dsts, dst))
1678 if (port->vlan < 0) {
1679 dst->vlan = m->out_vlan;
1681 if (dst->dp_ifidx == flow->in_port
1682 && dst->vlan == vlan) {
1683 /* Don't send out input port on same VLAN. */
1691 mirrors &= mirrors - 1;
1694 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1699 print_dsts(const struct dst *dsts, size_t n)
1701 for (; n--; dsts++) {
1702 printf(">p%"PRIu16, dsts->dp_ifidx);
1703 if (dsts->vlan != OFP_VLAN_NONE) {
1704 printf("v%"PRIu16, dsts->vlan);
1710 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1711 const struct port *in_port, const struct port *out_port,
1712 tag_type *tags, struct odp_actions *actions)
1714 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1716 const struct dst *p;
1719 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1721 cur_vlan = ntohs(flow->dl_vlan);
1722 for (p = dsts; p < &dsts[n_dsts]; p++) {
1723 union odp_action *a;
1724 if (p->vlan != cur_vlan) {
1725 if (p->vlan == OFP_VLAN_NONE) {
1726 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1728 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1729 a->vlan_vid.vlan_vid = htons(p->vlan);
1733 a = odp_actions_add(actions, ODPAT_OUTPUT);
1734 a->output.port = p->dp_ifidx;
1739 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1741 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1742 return (flow->dl_type == htons(ETH_TYPE_ARP)
1743 && eth_addr_is_broadcast(flow->dl_dst)
1744 && packet->size >= sizeof(struct arp_eth_header)
1745 && arp->ar_op == ARP_OP_REQUEST);
1748 /* If the composed actions may be applied to any packet in the given 'flow',
1749 * returns true. Otherwise, the actions should only be applied to 'packet', or
1750 * not at all, if 'packet' was NULL. */
1752 process_flow(struct bridge *br, const flow_t *flow,
1753 const struct ofpbuf *packet, struct odp_actions *actions,
1756 struct iface *in_iface;
1757 struct port *in_port;
1758 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1761 /* Find the interface and port structure for the received packet. */
1762 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1764 /* No interface? Something fishy... */
1765 if (packet != NULL) {
1766 /* Odd. A few possible reasons here:
1768 * - We deleted an interface but there are still a few packets
1769 * queued up from it.
1771 * - Someone externally added an interface (e.g. with "ovs-dpctl
1772 * add-if") that we don't know about.
1774 * - Packet arrived on the local port but the local port is not
1775 * one of our bridge ports.
1777 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1779 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1780 "interface %"PRIu16, br->name, flow->in_port);
1783 /* Return without adding any actions, to drop packets on this flow. */
1786 in_port = in_iface->port;
1788 /* Figure out what VLAN this packet belongs to.
1790 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1791 * belongs to VLAN 0, so we should treat both cases identically. (In the
1792 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1793 * presumably to allow a priority to be specified. In the latter case, the
1794 * packet does not have any 802.1Q header.) */
1795 vlan = ntohs(flow->dl_vlan);
1796 if (vlan == OFP_VLAN_NONE) {
1799 if (in_port->vlan >= 0) {
1801 /* XXX support double tagging? */
1802 if (packet != NULL) {
1803 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1804 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1805 "packet received on port %s configured with "
1806 "implicit VLAN %"PRIu16,
1807 br->name, ntohs(flow->dl_vlan),
1808 in_port->name, in_port->vlan);
1812 vlan = in_port->vlan;
1814 if (!port_includes_vlan(in_port, vlan)) {
1815 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1816 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1817 "packet received on port %s not configured for "
1819 br->name, vlan, in_port->name, vlan);
1824 /* Drop frames for ports that STP wants entirely killed (both for
1825 * forwarding and for learning). Later, after we do learning, we'll drop
1826 * the frames that STP wants to do learning but not forwarding on. */
1827 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1831 /* Drop frames for reserved multicast addresses. */
1832 if (eth_addr_is_reserved(flow->dl_dst)) {
1836 /* Drop frames on ports reserved for mirroring. */
1837 if (in_port->is_mirror_output_port) {
1838 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1839 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1840 "which is reserved exclusively for mirroring",
1841 br->name, in_port->name);
1845 /* Multicast (and broadcast) packets on bonds need special attention, to
1846 * avoid receiving duplicates. */
1847 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1848 *tags |= in_port->active_iface_tag;
1849 if (in_port->active_iface != in_iface->port_ifidx) {
1850 /* Drop all multicast packets on inactive slaves. */
1853 /* Drop all multicast packets for which we have learned a different
1854 * input port, because we probably sent the packet on one slaves
1855 * and got it back on the active slave. Broadcast ARP replies are
1856 * an exception to this rule: the host has moved to another
1858 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1859 if (src_idx != -1 && src_idx != in_port->port_idx) {
1861 if (!is_bcast_arp_reply(flow, packet)) {
1865 /* No way to know whether it's an ARP reply, because the
1866 * flow entry doesn't include enough information and we
1867 * don't have a packet. Punt. */
1875 out_port = FLOOD_PORT;
1879 /* Learn source MAC (but don't try to learn from revalidation). */
1881 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1882 vlan, in_port->port_idx);
1884 /* The log messages here could actually be useful in debugging,
1885 * so keep the rate limit relatively high. */
1886 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1888 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1889 "on port %s in VLAN %d",
1890 br->name, ETH_ADDR_ARGS(flow->dl_src),
1891 in_port->name, vlan);
1892 ofproto_revalidate(br->ofproto, rev_tag);
1896 /* Determine output port. */
1897 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1899 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1900 out_port = br->ports[out_port_idx];
1904 /* Don't send packets out their input ports. Don't forward frames that STP
1905 * wants us to discard. */
1906 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1911 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1914 * We send out only a single packet, instead of setting up a flow, if the
1915 * packet is an ARP directed to broadcast that arrived on a bonded
1916 * interface. In such a situation ARP requests and replies must be handled
1917 * differently, but OpenFlow unfortunately can't distinguish them.
1919 return (in_port->n_ifaces < 2
1920 || flow->dl_type != htons(ETH_TYPE_ARP)
1921 || !eth_addr_is_broadcast(flow->dl_dst));
1924 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1927 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1928 const struct ofp_phy_port *opp,
1931 struct bridge *br = br_;
1932 struct iface *iface;
1935 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1941 if (reason == OFPPR_DELETE) {
1942 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1943 br->name, iface->name);
1944 iface_destroy(iface);
1945 if (!port->n_ifaces) {
1946 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1947 br->name, port->name);
1953 if (port->n_ifaces > 1) {
1954 bool up = !(opp->state & OFPPS_LINK_DOWN);
1955 bond_link_status_update(iface, up);
1956 port_update_bond_compat(port);
1962 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1963 struct odp_actions *actions, tag_type *tags, void *br_)
1965 struct bridge *br = br_;
1968 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1969 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1970 brstp_receive(br, flow, payload);
1975 COVERAGE_INC(bridge_process_flow);
1976 return process_flow(br, flow, packet, actions, tags);
1980 bridge_account_flow_ofhook_cb(const flow_t *flow,
1981 const union odp_action *actions,
1982 size_t n_actions, unsigned long long int n_bytes,
1985 struct bridge *br = br_;
1986 const union odp_action *a;
1988 if (!br->has_bonded_ports) {
1992 for (a = actions; a < &actions[n_actions]; a++) {
1993 if (a->type == ODPAT_OUTPUT) {
1994 struct port *port = port_from_dp_ifidx(br, a->output.port);
1995 if (port && port->n_ifaces >= 2) {
1996 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1997 e->tx_bytes += n_bytes;
2004 bridge_account_checkpoint_ofhook_cb(void *br_)
2006 struct bridge *br = br_;
2009 if (!br->has_bonded_ports) {
2013 /* The current ofproto implementation calls this callback at least once a
2014 * second, so this timer implementation is sufficient. */
2015 if (time_msec() < br->bond_next_rebalance) {
2018 br->bond_next_rebalance = time_msec() + 10000;
2020 for (i = 0; i < br->n_ports; i++) {
2021 struct port *port = br->ports[i];
2022 if (port->n_ifaces > 1) {
2023 bond_rebalance_port(port);
2028 static struct ofhooks bridge_ofhooks = {
2029 bridge_port_changed_ofhook_cb,
2030 bridge_normal_ofhook_cb,
2031 bridge_account_flow_ofhook_cb,
2032 bridge_account_checkpoint_ofhook_cb,
2035 /* Bonding functions. */
2037 /* Statistics for a single interface on a bonded port, used for load-based
2038 * bond rebalancing. */
2039 struct slave_balance {
2040 struct iface *iface; /* The interface. */
2041 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2043 /* All the "bond_entry"s that are assigned to this interface, in order of
2044 * increasing tx_bytes. */
2045 struct bond_entry **hashes;
2049 /* Sorts pointers to pointers to bond_entries in ascending order by the
2050 * interface to which they are assigned, and within a single interface in
2051 * ascending order of bytes transmitted. */
2053 compare_bond_entries(const void *a_, const void *b_)
2055 const struct bond_entry *const *ap = a_;
2056 const struct bond_entry *const *bp = b_;
2057 const struct bond_entry *a = *ap;
2058 const struct bond_entry *b = *bp;
2059 if (a->iface_idx != b->iface_idx) {
2060 return a->iface_idx > b->iface_idx ? 1 : -1;
2061 } else if (a->tx_bytes != b->tx_bytes) {
2062 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2068 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2069 * *descending* order by number of bytes transmitted. */
2071 compare_slave_balance(const void *a_, const void *b_)
2073 const struct slave_balance *a = a_;
2074 const struct slave_balance *b = b_;
2075 if (a->iface->enabled != b->iface->enabled) {
2076 return a->iface->enabled ? -1 : 1;
2077 } else if (a->tx_bytes != b->tx_bytes) {
2078 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2085 swap_bals(struct slave_balance *a, struct slave_balance *b)
2087 struct slave_balance tmp = *a;
2092 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2093 * given that 'p' (and only 'p') might be in the wrong location.
2095 * This function invalidates 'p', since it might now be in a different memory
2098 resort_bals(struct slave_balance *p,
2099 struct slave_balance bals[], size_t n_bals)
2102 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2103 swap_bals(p, p - 1);
2105 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2106 swap_bals(p, p + 1);
2112 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2114 if (VLOG_IS_DBG_ENABLED()) {
2115 struct ds ds = DS_EMPTY_INITIALIZER;
2116 const struct slave_balance *b;
2118 for (b = bals; b < bals + n_bals; b++) {
2122 ds_put_char(&ds, ',');
2124 ds_put_format(&ds, " %s %"PRIu64"kB",
2125 b->iface->name, b->tx_bytes / 1024);
2127 if (!b->iface->enabled) {
2128 ds_put_cstr(&ds, " (disabled)");
2130 if (b->n_hashes > 0) {
2131 ds_put_cstr(&ds, " (");
2132 for (i = 0; i < b->n_hashes; i++) {
2133 const struct bond_entry *e = b->hashes[i];
2135 ds_put_cstr(&ds, " + ");
2137 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2138 e - port->bond_hash, e->tx_bytes / 1024);
2140 ds_put_cstr(&ds, ")");
2143 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2148 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2150 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2151 struct bond_entry *hash)
2153 struct port *port = from->iface->port;
2154 uint64_t delta = hash->tx_bytes;
2156 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2157 "from %s to %s (now carrying %"PRIu64"kB and "
2158 "%"PRIu64"kB load, respectively)",
2159 port->name, delta / 1024, hash - port->bond_hash,
2160 from->iface->name, to->iface->name,
2161 (from->tx_bytes - delta) / 1024,
2162 (to->tx_bytes + delta) / 1024);
2164 /* Delete element from from->hashes.
2166 * We don't bother to add the element to to->hashes because not only would
2167 * it require more work, the only purpose it would be to allow that hash to
2168 * be migrated to another slave in this rebalancing run, and there is no
2169 * point in doing that. */
2170 if (from->hashes[0] == hash) {
2173 int i = hash - from->hashes[0];
2174 memmove(from->hashes + i, from->hashes + i + 1,
2175 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2179 /* Shift load away from 'from' to 'to'. */
2180 from->tx_bytes -= delta;
2181 to->tx_bytes += delta;
2183 /* Arrange for flows to be revalidated. */
2184 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2185 hash->iface_idx = to->iface->port_ifidx;
2186 hash->iface_tag = tag_create_random();
2190 bond_rebalance_port(struct port *port)
2192 struct slave_balance bals[DP_MAX_PORTS];
2194 struct bond_entry *hashes[BOND_MASK + 1];
2195 struct slave_balance *b, *from, *to;
2196 struct bond_entry *e;
2199 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2200 * descending order of tx_bytes, so that bals[0] represents the most
2201 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2204 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2205 * array for each slave_balance structure, we sort our local array of
2206 * hashes in order by slave, so that all of the hashes for a given slave
2207 * become contiguous in memory, and then we point each 'hashes' members of
2208 * a slave_balance structure to the start of a contiguous group. */
2209 n_bals = port->n_ifaces;
2210 for (b = bals; b < &bals[n_bals]; b++) {
2211 b->iface = port->ifaces[b - bals];
2216 for (i = 0; i <= BOND_MASK; i++) {
2217 hashes[i] = &port->bond_hash[i];
2219 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2220 for (i = 0; i <= BOND_MASK; i++) {
2222 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2223 b = &bals[e->iface_idx];
2224 b->tx_bytes += e->tx_bytes;
2226 b->hashes = &hashes[i];
2231 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2232 log_bals(bals, n_bals, port);
2234 /* Discard slaves that aren't enabled (which were sorted to the back of the
2235 * array earlier). */
2236 while (!bals[n_bals - 1].iface->enabled) {
2243 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2244 to = &bals[n_bals - 1];
2245 for (from = bals; from < to; ) {
2246 uint64_t overload = from->tx_bytes - to->tx_bytes;
2247 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2248 /* The extra load on 'from' (and all less-loaded slaves), compared
2249 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2250 * it is less than ~1Mbps. No point in rebalancing. */
2252 } else if (from->n_hashes == 1) {
2253 /* 'from' only carries a single MAC hash, so we can't shift any
2254 * load away from it, even though we want to. */
2257 /* 'from' is carrying significantly more load than 'to', and that
2258 * load is split across at least two different hashes. Pick a hash
2259 * to migrate to 'to' (the least-loaded slave), given that doing so
2260 * must not cause 'to''s load to exceed 'from''s load.
2262 * The sort order we use means that we prefer to shift away the
2263 * smallest hashes instead of the biggest ones. There is little
2264 * reason behind this decision; we could use the opposite sort
2265 * order to shift away big hashes ahead of small ones. */
2268 for (i = 0; i < from->n_hashes; i++) {
2269 uint64_t delta = from->hashes[i]->tx_bytes;
2270 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2274 if (i < from->n_hashes) {
2275 bond_shift_load(from, to, from->hashes[i]);
2277 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2278 * point to different slave_balance structures. It is only
2279 * valid to do these two operations in a row at all because we
2280 * know that 'from' will not move past 'to' and vice versa. */
2281 resort_bals(from, bals, n_bals);
2282 resort_bals(to, bals, n_bals);
2289 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2290 * historical data to decay to <1% in 7 rebalancing runs. */
2291 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2297 bond_send_learning_packets(struct port *port)
2299 struct bridge *br = port->bridge;
2300 struct mac_entry *e;
2301 struct ofpbuf packet;
2302 int error, n_packets, n_errors;
2304 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2308 ofpbuf_init(&packet, 128);
2309 error = n_packets = n_errors = 0;
2310 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2311 static const char s[] = "Open vSwitch Bond Failover";
2312 union ofp_action actions[2], *a;
2313 struct eth_header *eth;
2314 struct llc_snap_header *llc_snap;
2320 if (e->port == port->port_idx
2321 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2325 /* Compose packet to send. */
2326 ofpbuf_clear(&packet);
2327 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2328 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2329 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2330 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2332 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2333 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2334 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2336 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2337 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2338 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2339 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2340 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2342 /* Compose actions. */
2343 memset(actions, 0, sizeof actions);
2346 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2347 a->vlan_vid.len = htons(sizeof *a);
2348 a->vlan_vid.vlan_vid = htons(e->vlan);
2351 a->output.type = htons(OFPAT_OUTPUT);
2352 a->output.len = htons(sizeof *a);
2353 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2358 flow_extract(&packet, ODPP_NONE, &flow);
2359 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2366 ofpbuf_uninit(&packet);
2369 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2370 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2371 "packets, last error was: %s",
2372 port->name, n_errors, n_packets, strerror(error));
2374 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2375 port->name, n_packets);
2379 /* Bonding unixctl user interface functions. */
2382 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2384 struct ds ds = DS_EMPTY_INITIALIZER;
2385 const struct bridge *br;
2387 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2389 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2392 for (i = 0; i < br->n_ports; i++) {
2393 const struct port *port = br->ports[i];
2394 if (port->n_ifaces > 1) {
2397 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2398 for (j = 0; j < port->n_ifaces; j++) {
2399 const struct iface *iface = port->ifaces[j];
2401 ds_put_cstr(&ds, ", ");
2403 ds_put_cstr(&ds, iface->name);
2405 ds_put_char(&ds, '\n');
2409 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2413 static struct port *
2414 bond_find(const char *name)
2416 const struct bridge *br;
2418 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2421 for (i = 0; i < br->n_ports; i++) {
2422 struct port *port = br->ports[i];
2423 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2432 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2434 struct ds ds = DS_EMPTY_INITIALIZER;
2435 const struct port *port;
2438 port = bond_find(args);
2440 unixctl_command_reply(conn, 501, "no such bond");
2444 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2445 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2446 ds_put_format(&ds, "next rebalance: %lld ms\n",
2447 port->bridge->bond_next_rebalance - time_msec());
2448 for (j = 0; j < port->n_ifaces; j++) {
2449 const struct iface *iface = port->ifaces[j];
2450 struct bond_entry *be;
2453 ds_put_format(&ds, "slave %s: %s\n",
2454 iface->name, iface->enabled ? "enabled" : "disabled");
2455 if (j == port->active_iface) {
2456 ds_put_cstr(&ds, "\tactive slave\n");
2458 if (iface->delay_expires != LLONG_MAX) {
2459 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2460 iface->enabled ? "downdelay" : "updelay",
2461 iface->delay_expires - time_msec());
2465 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2466 int hash = be - port->bond_hash;
2467 struct mac_entry *me;
2469 if (be->iface_idx != j) {
2473 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2474 hash, be->tx_bytes / 1024);
2477 if (!port->bridge->ml) {
2481 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2482 &port->bridge->ml->lrus) {
2485 if (bond_hash(me->mac) == hash
2486 && me->port != port->port_idx
2487 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2488 && dp_ifidx == iface->dp_ifidx)
2490 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2491 ETH_ADDR_ARGS(me->mac));
2496 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2501 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2503 char *args = (char *) args_;
2504 char *save_ptr = NULL;
2505 char *bond_s, *hash_s, *slave_s;
2506 uint8_t mac[ETH_ADDR_LEN];
2508 struct iface *iface;
2509 struct bond_entry *entry;
2512 bond_s = strtok_r(args, " ", &save_ptr);
2513 hash_s = strtok_r(NULL, " ", &save_ptr);
2514 slave_s = strtok_r(NULL, " ", &save_ptr);
2516 unixctl_command_reply(conn, 501,
2517 "usage: bond/migrate BOND HASH SLAVE");
2521 port = bond_find(bond_s);
2523 unixctl_command_reply(conn, 501, "no such bond");
2527 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2528 == ETH_ADDR_SCAN_COUNT) {
2529 hash = bond_hash(mac);
2530 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2531 hash = atoi(hash_s) & BOND_MASK;
2533 unixctl_command_reply(conn, 501, "bad hash");
2537 iface = port_lookup_iface(port, slave_s);
2539 unixctl_command_reply(conn, 501, "no such slave");
2543 if (!iface->enabled) {
2544 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2548 entry = &port->bond_hash[hash];
2549 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2550 entry->iface_idx = iface->port_ifidx;
2551 entry->iface_tag = tag_create_random();
2552 unixctl_command_reply(conn, 200, "migrated");
2556 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2558 char *args = (char *) args_;
2559 char *save_ptr = NULL;
2560 char *bond_s, *slave_s;
2562 struct iface *iface;
2564 bond_s = strtok_r(args, " ", &save_ptr);
2565 slave_s = strtok_r(NULL, " ", &save_ptr);
2567 unixctl_command_reply(conn, 501,
2568 "usage: bond/set-active-slave BOND SLAVE");
2572 port = bond_find(bond_s);
2574 unixctl_command_reply(conn, 501, "no such bond");
2578 iface = port_lookup_iface(port, slave_s);
2580 unixctl_command_reply(conn, 501, "no such slave");
2584 if (!iface->enabled) {
2585 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2589 if (port->active_iface != iface->port_ifidx) {
2590 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2591 port->active_iface = iface->port_ifidx;
2592 port->active_iface_tag = tag_create_random();
2593 VLOG_INFO("port %s: active interface is now %s",
2594 port->name, iface->name);
2595 bond_send_learning_packets(port);
2596 unixctl_command_reply(conn, 200, "done");
2598 unixctl_command_reply(conn, 200, "no change");
2603 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2605 char *args = (char *) args_;
2606 char *save_ptr = NULL;
2607 char *bond_s, *slave_s;
2609 struct iface *iface;
2611 bond_s = strtok_r(args, " ", &save_ptr);
2612 slave_s = strtok_r(NULL, " ", &save_ptr);
2614 unixctl_command_reply(conn, 501,
2615 "usage: bond/enable/disable-slave BOND SLAVE");
2619 port = bond_find(bond_s);
2621 unixctl_command_reply(conn, 501, "no such bond");
2625 iface = port_lookup_iface(port, slave_s);
2627 unixctl_command_reply(conn, 501, "no such slave");
2631 bond_enable_slave(iface, enable);
2632 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2636 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2638 enable_slave(conn, args, true);
2642 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2644 enable_slave(conn, args, false);
2650 unixctl_command_register("bond/list", bond_unixctl_list);
2651 unixctl_command_register("bond/show", bond_unixctl_show);
2652 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2653 unixctl_command_register("bond/set-active-slave",
2654 bond_unixctl_set_active_slave);
2655 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2656 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2659 /* Port functions. */
2662 port_create(struct bridge *br, const char *name)
2666 port = xcalloc(1, sizeof *port);
2668 port->port_idx = br->n_ports;
2670 port->trunks = NULL;
2671 port->name = xstrdup(name);
2672 port->active_iface = -1;
2673 port->stp_state = STP_DISABLED;
2674 port->stp_state_tag = 0;
2676 if (br->n_ports >= br->allocated_ports) {
2677 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2680 br->ports[br->n_ports++] = port;
2682 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2687 port_reconfigure(struct port *port)
2689 bool bonded = cfg_has_section("bonding.%s", port->name);
2690 struct svec old_ifaces, new_ifaces;
2691 unsigned long *trunks;
2695 /* Collect old and new interfaces. */
2696 svec_init(&old_ifaces);
2697 svec_init(&new_ifaces);
2698 for (i = 0; i < port->n_ifaces; i++) {
2699 svec_add(&old_ifaces, port->ifaces[i]->name);
2701 svec_sort(&old_ifaces);
2703 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2704 if (!new_ifaces.n) {
2705 VLOG_ERR("port %s: no interfaces specified for bonded port",
2707 } else if (new_ifaces.n == 1) {
2708 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2712 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2713 if (port->updelay < 0) {
2716 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2717 if (port->downdelay < 0) {
2718 port->downdelay = 0;
2721 svec_init(&new_ifaces);
2722 svec_add(&new_ifaces, port->name);
2725 /* Get rid of deleted interfaces and add new interfaces. */
2726 for (i = 0; i < port->n_ifaces; i++) {
2727 struct iface *iface = port->ifaces[i];
2728 if (!svec_contains(&new_ifaces, iface->name)) {
2729 iface_destroy(iface);
2734 for (i = 0; i < new_ifaces.n; i++) {
2735 const char *name = new_ifaces.names[i];
2736 if (!svec_contains(&old_ifaces, name)) {
2737 iface_create(port, name);
2743 if (cfg_has("vlan.%s.tag", port->name)) {
2745 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2746 if (vlan >= 0 && vlan <= 4095) {
2747 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2750 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2751 * they even work as-is. But they have not been tested. */
2752 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2756 if (port->vlan != vlan) {
2758 bridge_flush(port->bridge);
2761 /* Get trunked VLANs. */
2764 size_t n_trunks, n_errors;
2767 trunks = bitmap_allocate(4096);
2768 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2770 for (i = 0; i < n_trunks; i++) {
2771 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2773 bitmap_set1(trunks, trunk);
2779 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2780 port->name, n_trunks);
2782 if (n_errors == n_trunks) {
2784 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2787 bitmap_set_multiple(trunks, 0, 4096, 1);
2790 if (cfg_has("vlan.%s.trunks", port->name)) {
2791 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2792 port->name, port->name);
2796 ? port->trunks != NULL
2797 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2798 bridge_flush(port->bridge);
2800 bitmap_free(port->trunks);
2801 port->trunks = trunks;
2803 svec_destroy(&old_ifaces);
2804 svec_destroy(&new_ifaces);
2808 port_destroy(struct port *port)
2811 struct bridge *br = port->bridge;
2815 proc_net_compat_update_vlan(port->name, NULL, 0);
2817 for (i = 0; i < MAX_MIRRORS; i++) {
2818 struct mirror *m = br->mirrors[i];
2819 if (m && m->out_port == port) {
2824 while (port->n_ifaces > 0) {
2825 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2828 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2829 del->port_idx = port->port_idx;
2832 bitmap_free(port->trunks);
2839 static struct port *
2840 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2842 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2843 return iface ? iface->port : NULL;
2846 static struct port *
2847 port_lookup(const struct bridge *br, const char *name)
2851 for (i = 0; i < br->n_ports; i++) {
2852 struct port *port = br->ports[i];
2853 if (!strcmp(port->name, name)) {
2860 static struct iface *
2861 port_lookup_iface(const struct port *port, const char *name)
2865 for (j = 0; j < port->n_ifaces; j++) {
2866 struct iface *iface = port->ifaces[j];
2867 if (!strcmp(iface->name, name)) {
2875 port_update_bonding(struct port *port)
2877 if (port->n_ifaces < 2) {
2878 /* Not a bonded port. */
2879 if (port->bond_hash) {
2880 free(port->bond_hash);
2881 port->bond_hash = NULL;
2882 proc_net_compat_update_bond(port->name, NULL);
2885 if (!port->bond_hash) {
2888 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2889 for (i = 0; i <= BOND_MASK; i++) {
2890 struct bond_entry *e = &port->bond_hash[i];
2894 port->no_ifaces_tag = tag_create_random();
2895 bond_choose_active_iface(port);
2897 port_update_bond_compat(port);
2902 port_update_bond_compat(struct port *port)
2904 struct compat_bond bond;
2907 if (port->n_ifaces < 2) {
2912 bond.updelay = port->updelay;
2913 bond.downdelay = port->downdelay;
2914 bond.n_slaves = port->n_ifaces;
2915 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2916 for (i = 0; i < port->n_ifaces; i++) {
2917 struct iface *iface = port->ifaces[i];
2918 struct compat_bond_slave *slave = &bond.slaves[i];
2919 slave->name = iface->name;
2920 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2921 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2925 netdev_get_etheraddr(iface->netdev, slave->mac);
2927 proc_net_compat_update_bond(port->name, &bond);
2932 port_update_vlan_compat(struct port *port)
2934 struct bridge *br = port->bridge;
2935 char *vlandev_name = NULL;
2937 if (port->vlan > 0) {
2938 /* Figure out the name that the VLAN device should actually have, if it
2939 * existed. This takes some work because the VLAN device would not
2940 * have port->name in its name; rather, it would have the trunk port's
2941 * name, and 'port' would be attached to a bridge that also had the
2942 * VLAN device one of its ports. So we need to find a trunk port that
2943 * includes port->vlan.
2945 * There might be more than one candidate. This doesn't happen on
2946 * XenServer, so if it happens we just pick the first choice in
2947 * alphabetical order instead of creating multiple VLAN devices. */
2949 for (i = 0; i < br->n_ports; i++) {
2950 struct port *p = br->ports[i];
2951 if (port_trunks_vlan(p, port->vlan)
2953 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2955 uint8_t ea[ETH_ADDR_LEN];
2956 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
2957 if (!eth_addr_is_multicast(ea) &&
2958 !eth_addr_is_reserved(ea) &&
2959 !eth_addr_is_zero(ea)) {
2960 vlandev_name = p->name;
2965 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2968 /* Interface functions. */
2971 iface_create(struct port *port, const char *name)
2973 struct iface *iface;
2975 iface = xcalloc(1, sizeof *iface);
2977 iface->port_ifidx = port->n_ifaces;
2978 iface->name = xstrdup(name);
2979 iface->dp_ifidx = -1;
2980 iface->tag = tag_create_random();
2981 iface->delay_expires = LLONG_MAX;
2982 iface->netdev = NULL;
2984 if (port->n_ifaces >= port->allocated_ifaces) {
2985 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2986 sizeof *port->ifaces);
2988 port->ifaces[port->n_ifaces++] = iface;
2989 if (port->n_ifaces > 1) {
2990 port->bridge->has_bonded_ports = true;
2993 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2995 bridge_flush(port->bridge);
2999 iface_destroy(struct iface *iface)
3002 struct port *port = iface->port;
3003 struct bridge *br = port->bridge;
3004 bool del_active = port->active_iface == iface->port_ifidx;
3007 if (iface->dp_ifidx >= 0) {
3008 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3011 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3012 del->port_ifidx = iface->port_ifidx;
3014 netdev_close(iface->netdev);
3019 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3020 bond_choose_active_iface(port);
3021 bond_send_learning_packets(port);
3024 bridge_flush(port->bridge);
3028 static struct iface *
3029 iface_lookup(const struct bridge *br, const char *name)
3033 for (i = 0; i < br->n_ports; i++) {
3034 struct port *port = br->ports[i];
3035 for (j = 0; j < port->n_ifaces; j++) {
3036 struct iface *iface = port->ifaces[j];
3037 if (!strcmp(iface->name, name)) {
3045 static struct iface *
3046 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3048 return port_array_get(&br->ifaces, dp_ifidx);
3051 /* Port mirroring. */
3054 mirror_reconfigure(struct bridge *br)
3056 struct svec old_mirrors, new_mirrors;
3059 /* Collect old and new mirrors. */
3060 svec_init(&old_mirrors);
3061 svec_init(&new_mirrors);
3062 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3063 for (i = 0; i < MAX_MIRRORS; i++) {
3064 if (br->mirrors[i]) {
3065 svec_add(&old_mirrors, br->mirrors[i]->name);
3069 /* Get rid of deleted mirrors and add new mirrors. */
3070 svec_sort(&old_mirrors);
3071 assert(svec_is_unique(&old_mirrors));
3072 svec_sort(&new_mirrors);
3073 assert(svec_is_unique(&new_mirrors));
3074 for (i = 0; i < MAX_MIRRORS; i++) {
3075 struct mirror *m = br->mirrors[i];
3076 if (m && !svec_contains(&new_mirrors, m->name)) {
3080 for (i = 0; i < new_mirrors.n; i++) {
3081 const char *name = new_mirrors.names[i];
3082 if (!svec_contains(&old_mirrors, name)) {
3083 mirror_create(br, name);
3086 svec_destroy(&old_mirrors);
3087 svec_destroy(&new_mirrors);
3089 /* Reconfigure all mirrors. */
3090 for (i = 0; i < MAX_MIRRORS; i++) {
3091 if (br->mirrors[i]) {
3092 mirror_reconfigure_one(br->mirrors[i]);
3096 /* Update port reserved status. */
3097 for (i = 0; i < br->n_ports; i++) {
3098 br->ports[i]->is_mirror_output_port = false;
3100 for (i = 0; i < MAX_MIRRORS; i++) {
3101 struct mirror *m = br->mirrors[i];
3102 if (m && m->out_port) {
3103 m->out_port->is_mirror_output_port = true;
3109 mirror_create(struct bridge *br, const char *name)
3114 for (i = 0; ; i++) {
3115 if (i >= MAX_MIRRORS) {
3116 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3117 "cannot create %s", br->name, MAX_MIRRORS, name);
3120 if (!br->mirrors[i]) {
3125 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3128 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3131 m->name = xstrdup(name);
3132 svec_init(&m->src_ports);
3133 svec_init(&m->dst_ports);
3141 mirror_destroy(struct mirror *m)
3144 struct bridge *br = m->bridge;
3147 for (i = 0; i < br->n_ports; i++) {
3148 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3149 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3152 svec_destroy(&m->src_ports);
3153 svec_destroy(&m->dst_ports);
3156 m->bridge->mirrors[m->idx] = NULL;
3164 prune_ports(struct mirror *m, struct svec *ports)
3169 svec_sort_unique(ports);
3172 for (i = 0; i < ports->n; i++) {
3173 const char *name = ports->names[i];
3174 if (port_lookup(m->bridge, name)) {
3175 svec_add(&tmp, name);
3177 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3178 m->bridge->name, m->name, name);
3181 svec_swap(ports, &tmp);
3186 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3190 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3191 * order won't give us numeric sort order. But that's good enough for what
3192 * we need right now. */
3193 svec_sort_unique(vlan_strings);
3195 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3197 for (i = 0; i < vlan_strings->n; i++) {
3198 const char *name = vlan_strings->names[i];
3200 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3201 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3202 m->bridge->name, m->name, name);
3204 (*vlans)[n_vlans++] = vlan;
3211 vlan_is_mirrored(const struct mirror *m, int vlan)
3215 for (i = 0; i < m->n_vlans; i++) {
3216 if (m->vlans[i] == vlan) {
3224 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3228 for (i = 0; i < m->n_vlans; i++) {
3229 if (port_trunks_vlan(p, m->vlans[i])) {
3237 mirror_reconfigure_one(struct mirror *m)
3239 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3240 struct svec src_ports, dst_ports, ports;
3241 struct svec vlan_strings;
3242 mirror_mask_t mirror_bit;
3243 const char *out_port_name;
3244 struct port *out_port;
3249 bool mirror_all_ports;
3251 /* Get output port. */
3252 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3253 m->bridge->name, m->name);
3254 if (out_port_name) {
3255 out_port = port_lookup(m->bridge, out_port_name);
3257 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3258 "named %s", pfx, m->bridge->name, out_port_name);
3265 if (cfg_has("%s.output.vlan", pfx)) {
3266 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3267 "ignoring %s.output.vlan", pfx, pfx, pfx);
3269 } else if (cfg_has("%s.output.vlan", pfx)) {
3271 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3273 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3274 "but exactly one is required; disabling port mirror %s",
3275 pfx, pfx, pfx, pfx);
3281 /* Get all the ports, and drop duplicates and ports that don't exist. */
3282 svec_init(&src_ports);
3283 svec_init(&dst_ports);
3285 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3286 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3287 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3288 svec_append(&src_ports, &ports);
3289 svec_append(&dst_ports, &ports);
3290 svec_destroy(&ports);
3291 prune_ports(m, &src_ports);
3292 prune_ports(m, &dst_ports);
3294 /* Get all the vlans, and drop duplicate and invalid vlans. */
3295 svec_init(&vlan_strings);
3296 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3297 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3298 svec_destroy(&vlan_strings);
3300 /* Update mirror data. */
3301 if (!svec_equal(&m->src_ports, &src_ports)
3302 || !svec_equal(&m->dst_ports, &dst_ports)
3303 || m->n_vlans != n_vlans
3304 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3305 || m->out_port != out_port
3306 || m->out_vlan != out_vlan) {
3307 bridge_flush(m->bridge);
3309 svec_swap(&m->src_ports, &src_ports);
3310 svec_swap(&m->dst_ports, &dst_ports);
3313 m->n_vlans = n_vlans;
3314 m->out_port = out_port;
3315 m->out_vlan = out_vlan;
3317 /* If no selection criteria have been given, mirror for all ports. */
3318 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3321 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3322 for (i = 0; i < m->bridge->n_ports; i++) {
3323 struct port *port = m->bridge->ports[i];
3325 if (mirror_all_ports
3326 || svec_contains(&m->src_ports, port->name)
3329 ? port_trunks_any_mirrored_vlan(m, port)
3330 : vlan_is_mirrored(m, port->vlan)))) {
3331 port->src_mirrors |= mirror_bit;
3333 port->src_mirrors &= ~mirror_bit;
3336 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3337 port->dst_mirrors |= mirror_bit;
3339 port->dst_mirrors &= ~mirror_bit;
3344 svec_destroy(&src_ports);
3345 svec_destroy(&dst_ports);
3349 /* Spanning tree protocol. */
3351 static void brstp_update_port_state(struct port *);
3354 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3356 struct bridge *br = br_;
3357 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3358 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3360 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3363 struct eth_header *eth = pkt->l2;
3365 netdev_get_etheraddr(iface->netdev, eth->eth_src);
3366 if (eth_addr_is_zero(eth->eth_src)) {
3367 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
3368 "with unknown MAC", br->name, port_no);
3370 union ofp_action action;
3373 memset(&action, 0, sizeof action);
3374 action.type = htons(OFPAT_OUTPUT);
3375 action.output.len = htons(sizeof action);
3376 action.output.port = htons(port_no);
3378 flow_extract(pkt, ODPP_NONE, &flow);
3379 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3386 brstp_reconfigure(struct bridge *br)
3390 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3392 stp_destroy(br->stp);
3398 uint64_t bridge_address, bridge_id;
3399 int bridge_priority;
3401 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3402 if (!bridge_address) {
3404 bridge_address = (stp_get_bridge_id(br->stp)
3405 & ((UINT64_C(1) << 48) - 1));
3407 uint8_t mac[ETH_ADDR_LEN];
3408 eth_addr_random(mac);
3409 bridge_address = eth_addr_to_uint64(mac);
3413 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3415 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3417 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3420 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3422 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3423 br->stp_last_tick = time_msec();
3426 if (bridge_id != stp_get_bridge_id(br->stp)) {
3427 stp_set_bridge_id(br->stp, bridge_id);
3432 for (i = 0; i < br->n_ports; i++) {
3433 struct port *p = br->ports[i];
3435 struct stp_port *sp;
3436 int path_cost, priority;
3442 dp_ifidx = p->ifaces[0]->dp_ifidx;
3443 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3447 sp = stp_get_port(br->stp, dp_ifidx);
3448 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3449 "stp.%s.port.%s.enabled",
3451 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3452 br->name, p->name));
3453 if (p->is_mirror_output_port) {
3456 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3457 bridge_flush(br); /* Might not be necessary. */
3459 stp_port_enable(sp);
3461 stp_port_disable(sp);
3465 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3467 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3469 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3470 "stp.%s.port.%s.priority",
3472 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3474 : STP_DEFAULT_PORT_PRIORITY);
3475 stp_port_set_priority(sp, priority);
3478 brstp_adjust_timers(br);
3480 for (i = 0; i < br->n_ports; i++) {
3481 brstp_update_port_state(br->ports[i]);
3486 brstp_update_port_state(struct port *p)
3488 struct bridge *br = p->bridge;
3489 enum stp_state state;
3491 /* Figure out new state. */
3492 state = STP_DISABLED;
3493 if (br->stp && p->n_ifaces > 0) {
3494 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3495 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3496 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3501 if (p->stp_state != state) {
3502 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3503 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3504 p->name, stp_state_name(p->stp_state),
3505 stp_state_name(state));
3506 if (p->stp_state == STP_DISABLED) {
3509 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3511 p->stp_state = state;
3512 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3513 : tag_create_random());
3518 brstp_adjust_timers(struct bridge *br)
3520 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3521 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3522 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3524 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3525 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3526 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3530 brstp_run(struct bridge *br)
3533 long long int now = time_msec();
3534 long long int elapsed = now - br->stp_last_tick;
3535 struct stp_port *sp;
3538 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3539 br->stp_last_tick = now;
3541 while (stp_get_changed_port(br->stp, &sp)) {
3542 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3544 brstp_update_port_state(p);
3551 brstp_wait(struct bridge *br)
3554 poll_timer_wait(1000);