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"
47 #include "poll-loop.h"
48 #include "port-array.h"
49 #include "proc-net-compat.h"
51 #include "secchan/ofproto.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 struct port *port; /* Containing port. */
75 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 int dp_ifidx; /* Index within kernel datapath. */
80 uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */
82 tag_type tag; /* Tag associated with this interface. */
83 bool enabled; /* May be chosen for flows? */
84 long long delay_expires; /* Time after which 'enabled' may change. */
87 #define BOND_MASK 0xff
89 int iface_idx; /* Index of assigned iface, or -1 if none. */
90 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
91 tag_type iface_tag; /* Tag associated with iface_idx. */
94 #define MAX_MIRRORS 32
95 typedef uint32_t mirror_mask_t;
96 #define MIRROR_MASK_C(X) UINT32_C(X)
97 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
99 struct bridge *bridge;
103 /* Selection criteria. */
104 struct svec src_ports;
105 struct svec dst_ports;
110 struct port *out_port;
114 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
116 struct bridge *bridge;
118 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
119 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
122 /* An ordinary bridge port has 1 interface.
123 * A bridge port for bonding has at least 2 interfaces. */
124 struct iface **ifaces;
125 size_t n_ifaces, allocated_ifaces;
128 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
129 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
130 tag_type active_iface_tag; /* Tag for bcast flows. */
131 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
132 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
133 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
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; /* Kernel 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 void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *);
197 static int bridge_run_one(struct bridge *);
198 static void bridge_reconfigure_one(struct bridge *);
199 static void bridge_reconfigure_controller(struct bridge *);
200 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
201 static void bridge_fetch_dp_ifaces(struct bridge *);
202 static void bridge_flush(struct bridge *);
203 static void bridge_pick_local_hw_addr(struct bridge *,
204 uint8_t ea[ETH_ADDR_LEN],
205 const char **devname);
206 static uint64_t bridge_pick_datapath_id(struct bridge *,
207 const uint8_t bridge_ea[ETH_ADDR_LEN],
208 const char *devname);
209 static uint64_t dpid_from_hash(const void *, size_t nbytes);
211 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
213 static void bond_init(void);
214 static void bond_run(struct bridge *);
215 static void bond_wait(struct bridge *);
216 static void bond_rebalance_port(struct port *);
217 static void bond_send_learning_packets(struct port *);
219 static void port_create(struct bridge *, const char *name);
220 static void port_reconfigure(struct port *);
221 static void port_destroy(struct port *);
222 static struct port *port_lookup(const struct bridge *, const char *name);
223 static struct iface *port_lookup_iface(const struct port *, const char *name);
224 static struct port *port_from_dp_ifidx(const struct bridge *,
226 static void port_update_bond_compat(struct port *);
227 static void port_update_vlan_compat(struct port *);
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 *,
245 static bool iface_is_internal(const struct bridge *, const char *name);
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 dp%u, ignoring",
268 iface->name, dpif_id(&br->dpif));
270 if (iface->dp_ifidx != ODPP_LOCAL) {
271 svec_add(svec, iface->name);
279 /* The caller must already have called cfg_read(). */
288 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
290 for (i = 0; i < DP_MAX; i++) {
294 sprintf(devname, "dp%d", i);
295 retval = dpif_open(devname, &dpif);
297 char dpif_name[IF_NAMESIZE];
298 if (dpif_get_name(&dpif, dpif_name, sizeof dpif_name)
299 || !cfg_has("bridge.%s.port", dpif_name)) {
303 } else if (retval != ENODEV) {
304 VLOG_ERR("failed to delete datapath dp%d: %s",
305 i, strerror(retval));
309 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows);
311 bridge_reconfigure();
316 config_string_change(const char *key, char **valuep)
318 const char *value = cfg_get_string(0, "%s", key);
319 if (value && (!*valuep || strcmp(value, *valuep))) {
321 *valuep = xstrdup(value);
329 bridge_configure_ssl(void)
331 /* XXX SSL should be configurable on a per-bridge basis.
332 * XXX should be possible to de-configure SSL. */
333 static char *private_key_file;
334 static char *certificate_file;
335 static char *cacert_file;
338 if (config_string_change("ssl.private-key", &private_key_file)) {
339 vconn_ssl_set_private_key_file(private_key_file);
342 if (config_string_change("ssl.certificate", &certificate_file)) {
343 vconn_ssl_set_certificate_file(certificate_file);
346 /* We assume that even if the filename hasn't changed, if the CA cert
347 * file has been removed, that we want to move back into
348 * boot-strapping mode. This opens a small security hole, because
349 * the old certificate will still be trusted until vSwitch is
350 * restarted. We may want to address this in vconn's SSL library. */
351 if (config_string_change("ssl.ca-cert", &cacert_file)
352 || (stat(cacert_file, &s) && errno == ENOENT)) {
353 vconn_ssl_set_ca_cert_file(cacert_file,
354 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
360 bridge_reconfigure(void)
362 struct svec old_br, new_br, raw_new_br;
363 struct bridge *br, *next;
366 COVERAGE_INC(bridge_reconfigure);
368 /* Collect old bridges. */
370 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
371 svec_add(&old_br, br->name);
374 /* Collect new bridges. */
375 svec_init(&raw_new_br);
376 cfg_get_subsections(&raw_new_br, "bridge");
378 for (i = 0; i < raw_new_br.n; i++) {
379 const char *name = raw_new_br.names[i];
380 if ((!strncmp(name, "dp", 2) && isdigit(name[2])) ||
381 (!strncmp(name, "nl:", 3) && isdigit(name[3]))) {
382 VLOG_ERR("%s is not a valid bridge name (bridges may not be "
383 "named \"dp\" or \"nl:\" followed by a digit)", name);
385 svec_add(&new_br, name);
388 svec_destroy(&raw_new_br);
390 /* Get rid of deleted bridges and add new bridges. */
393 assert(svec_is_unique(&old_br));
394 assert(svec_is_unique(&new_br));
395 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
396 if (!svec_contains(&new_br, br->name)) {
400 for (i = 0; i < new_br.n; i++) {
401 const char *name = new_br.names[i];
402 if (!svec_contains(&old_br, name)) {
406 svec_destroy(&old_br);
407 svec_destroy(&new_br);
411 bridge_configure_ssl();
414 /* Reconfigure all bridges. */
415 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
416 bridge_reconfigure_one(br);
419 /* Add and delete ports on all datapaths.
421 * The kernel will reject any attempt to add a given port to a datapath if
422 * that port already belongs to a different datapath, so we must do all
423 * port deletions before any port additions. */
424 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
425 struct odp_port *dpif_ports;
427 struct svec want_ifaces;
429 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
430 bridge_get_all_ifaces(br, &want_ifaces);
431 for (i = 0; i < n_dpif_ports; i++) {
432 const struct odp_port *p = &dpif_ports[i];
433 if (!svec_contains(&want_ifaces, p->devname)
434 && strcmp(p->devname, br->name)) {
435 int retval = dpif_port_del(&br->dpif, p->port);
437 VLOG_ERR("failed to remove %s interface from dp%u: %s",
438 p->devname, dpif_id(&br->dpif), strerror(retval));
442 svec_destroy(&want_ifaces);
445 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
446 struct odp_port *dpif_ports;
448 struct svec cur_ifaces, want_ifaces, add_ifaces;
451 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
452 svec_init(&cur_ifaces);
453 for (i = 0; i < n_dpif_ports; i++) {
454 svec_add(&cur_ifaces, dpif_ports[i].devname);
457 svec_sort_unique(&cur_ifaces);
458 bridge_get_all_ifaces(br, &want_ifaces);
459 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
462 for (i = 0; i < add_ifaces.n; i++) {
463 const char *if_name = add_ifaces.names[i];
468 /* Add to datapath. */
469 internal = iface_is_internal(br, if_name);
470 error = dpif_port_add(&br->dpif, if_name, next_port_no++,
471 internal ? ODP_PORT_INTERNAL : 0);
472 if (error != EEXIST) {
473 if (next_port_no >= 256) {
474 VLOG_ERR("ran out of valid port numbers on dp%u",
479 VLOG_ERR("failed to add %s interface to dp%u: %s",
480 if_name, dpif_id(&br->dpif), strerror(error));
487 svec_destroy(&cur_ifaces);
488 svec_destroy(&want_ifaces);
489 svec_destroy(&add_ifaces);
491 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
494 struct iface *local_iface = NULL;
496 uint8_t engine_type = br->dpif.minor;
497 uint8_t engine_id = br->dpif.minor;
498 bool add_id_to_iface = false;
499 struct svec nf_hosts;
501 bridge_fetch_dp_ifaces(br);
502 for (i = 0; i < br->n_ports; ) {
503 struct port *port = br->ports[i];
505 for (j = 0; j < port->n_ifaces; ) {
506 struct iface *iface = port->ifaces[j];
507 if (iface->dp_ifidx < 0) {
508 VLOG_ERR("%s interface not in dp%u, dropping",
509 iface->name, dpif_id(&br->dpif));
510 iface_destroy(iface);
512 if (iface->dp_ifidx == ODPP_LOCAL) {
515 VLOG_DBG("dp%u has interface %s on port %d",
516 dpif_id(&br->dpif), iface->name, iface->dp_ifidx);
520 if (!port->n_ifaces) {
521 VLOG_ERR("%s port has no interfaces, dropping", port->name);
528 /* Pick local port hardware address, datapath ID. */
529 bridge_pick_local_hw_addr(br, ea, &devname);
531 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
533 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
534 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
535 "Ethernet address: %s",
536 br->name, strerror(error));
540 dpid = bridge_pick_datapath_id(br, ea, devname);
541 ofproto_set_datapath_id(br->ofproto, dpid);
543 /* Set NetFlow configuration on this bridge. */
544 if (cfg_has("netflow.%s.engine-type", br->name)) {
545 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
548 if (cfg_has("netflow.%s.engine-id", br->name)) {
549 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
551 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
552 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
555 if (add_id_to_iface && engine_id > 0x7f) {
556 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
557 "another vswitch, choose an engine id less than 128",
560 if (add_id_to_iface && br->n_ports > 0x1ff) {
561 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
562 "another port when 512 or more ports are used",
565 svec_init(&nf_hosts);
566 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
567 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
568 engine_id, add_id_to_iface)) {
569 VLOG_ERR("bridge %s: problem setting netflow collectors",
572 svec_destroy(&nf_hosts);
574 /* Update the controller and related settings. It would be more
575 * straightforward to call this from bridge_reconfigure_one(), but we
576 * can't do it there for two reasons. First, and most importantly, at
577 * that point we don't know the dp_ifidx of any interfaces that have
578 * been added to the bridge (because we haven't actually added them to
579 * the datapath). Second, at that point we haven't set the datapath ID
580 * yet; when a controller is configured, resetting the datapath ID will
581 * immediately disconnect from the controller, so it's better to set
582 * the datapath ID before the controller. */
583 bridge_reconfigure_controller(br);
585 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
586 for (i = 0; i < br->n_ports; i++) {
587 struct port *port = br->ports[i];
588 port_update_vlan_compat(port);
591 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
592 brstp_reconfigure(br);
597 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
598 const char **devname)
600 uint64_t requested_ea;
606 /* Did the user request a particular MAC? */
607 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
609 eth_addr_from_uint64(requested_ea, ea);
610 if (eth_addr_is_multicast(ea)) {
611 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
612 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
613 } else if (eth_addr_is_zero(ea)) {
614 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
620 /* Otherwise choose the minimum MAC address among all of the interfaces.
621 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
622 * MAC of the physical interface in such an environment.) */
623 memset(ea, 0xff, sizeof ea);
624 for (i = 0; i < br->n_ports; i++) {
625 struct port *port = br->ports[i];
626 uint8_t iface_ea[ETH_ADDR_LEN];
627 uint64_t iface_ea_u64;
630 /* Mirror output ports don't participate. */
631 if (port->is_mirror_output_port) {
635 /* Choose the MAC address to represent the port. */
636 iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name);
638 /* User specified explicitly. */
639 eth_addr_from_uint64(iface_ea_u64, iface_ea);
641 /* Find the interface with this Ethernet address (if any) so that
642 * we can provide the correct devname to the caller. */
644 for (j = 0; j < port->n_ifaces; j++) {
645 struct iface *candidate = port->ifaces[j];
646 uint8_t candidate_ea[ETH_ADDR_LEN];
647 if (!netdev_nodev_get_etheraddr(candidate->name, candidate_ea)
648 && eth_addr_equals(iface_ea, candidate_ea)) {
653 /* Choose the interface whose MAC address will represent the port.
654 * The Linux kernel bonding code always chooses the MAC address of
655 * the first slave added to a bond, and the Fedora networking
656 * scripts always add slaves to a bond in alphabetical order, so
657 * for compatibility we choose the interface with the name that is
658 * first in alphabetical order. */
659 iface = port->ifaces[0];
660 for (j = 1; j < port->n_ifaces; j++) {
661 struct iface *candidate = port->ifaces[j];
662 if (strcmp(candidate->name, iface->name) < 0) {
667 /* The local port doesn't count (since we're trying to choose its
668 * MAC address anyway). Other internal ports don't count because
669 * we really want a physical MAC if we can get it, and internal
670 * ports typically have randomly generated MACs. */
671 if (iface->dp_ifidx == ODPP_LOCAL
672 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
677 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
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));
686 /* Compare against our current choice. */
687 if (!eth_addr_is_multicast(iface_ea) &&
688 !eth_addr_is_reserved(iface_ea) &&
689 !eth_addr_is_zero(iface_ea) &&
690 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
692 memcpy(ea, iface_ea, ETH_ADDR_LEN);
693 *devname = iface ? iface->name : NULL;
696 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
697 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
699 VLOG_WARN("bridge %s: using default bridge Ethernet "
700 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
702 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
703 br->name, ETH_ADDR_ARGS(ea));
707 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
708 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
709 * a network device, then that network device's name must be passed in as
710 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
711 * passed in as a null pointer. */
713 bridge_pick_datapath_id(struct bridge *br,
714 const uint8_t bridge_ea[ETH_ADDR_LEN],
718 * The procedure for choosing a bridge MAC address will, in the most
719 * ordinary case, also choose a unique MAC that we can use as a datapath
720 * ID. In some special cases, though, multiple bridges will end up with
721 * the same MAC address. This is OK for the bridges, but it will confuse
722 * the OpenFlow controller, because each datapath needs a unique datapath
725 * Datapath IDs must be unique. It is also very desirable that they be
726 * stable from one run to the next, so that policy set on a datapath
731 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
738 if (!netdev_get_vlan_vid(devname, &vlan)) {
740 * A bridge whose MAC address is taken from a VLAN network device
741 * (that is, a network device created with vconfig(8) or similar
742 * tool) will have the same MAC address as a bridge on the VLAN
743 * device's physical network device.
745 * Handle this case by hashing the physical network device MAC
746 * along with the VLAN identifier.
748 uint8_t buf[ETH_ADDR_LEN + 2];
749 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
750 buf[ETH_ADDR_LEN] = vlan >> 8;
751 buf[ETH_ADDR_LEN + 1] = vlan;
752 return dpid_from_hash(buf, sizeof buf);
755 * Assume that this bridge's MAC address is unique, since it
756 * doesn't fit any of the cases we handle specially.
761 * A purely internal bridge, that is, one that has no non-virtual
762 * network devices on it at all, is more difficult because it has no
763 * natural unique identifier at all.
765 * When the host is a XenServer, we handle this case by hashing the
766 * host's UUID with the name of the bridge. Names of bridges are
767 * persistent across XenServer reboots, although they can be reused if
768 * an internal network is destroyed and then a new one is later
769 * created, so this is fairly effective.
771 * When the host is not a XenServer, we punt by using a random MAC
772 * address on each run.
774 const char *host_uuid = xenserver_get_host_uuid();
776 char *combined = xasprintf("%s,%s", host_uuid, br->name);
777 dpid = dpid_from_hash(combined, strlen(combined));
783 return eth_addr_to_uint64(bridge_ea);
787 dpid_from_hash(const void *data, size_t n)
789 uint8_t hash[SHA1_DIGEST_SIZE];
791 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
792 sha1_bytes(data, n, hash);
793 eth_addr_mark_random(hash);
794 return eth_addr_to_uint64(hash);
800 struct bridge *br, *next;
804 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
805 int error = bridge_run_one(br);
807 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
808 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
809 "forcing reconfiguration", br->name);
823 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
824 ofproto_wait(br->ofproto);
825 if (br->controller) {
830 mac_learning_wait(br->ml);
837 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
838 * configuration changes. */
840 bridge_flush(struct bridge *br)
842 COVERAGE_INC(bridge_flush);
845 mac_learning_flush(br->ml);
849 /* Bridge unixctl user interface functions. */
851 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
853 struct ds ds = DS_EMPTY_INITIALIZER;
854 const struct bridge *br;
856 br = bridge_lookup(args);
858 unixctl_command_reply(conn, 501, "no such bridge");
862 ds_put_cstr(&ds, " port VLAN MAC Age\n");
864 const struct mac_entry *e;
865 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
866 if (e->port < 0 || e->port >= br->n_ports) {
869 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
870 br->ports[e->port]->ifaces[0]->dp_ifidx,
871 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
874 unixctl_command_reply(conn, 200, ds_cstr(&ds));
878 /* Bridge reconfiguration functions. */
880 static struct bridge *
881 bridge_create(const char *name)
886 assert(!bridge_lookup(name));
887 br = xcalloc(1, sizeof *br);
889 error = dpif_create(name, &br->dpif);
890 if (error == EEXIST) {
891 error = dpif_open(name, &br->dpif);
893 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
894 name, strerror(error));
898 dpif_flow_flush(&br->dpif);
900 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
905 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
907 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
908 dpif_delete(&br->dpif);
909 dpif_close(&br->dpif);
914 br->name = xstrdup(name);
915 br->ml = mac_learning_create();
916 br->sent_config_request = false;
917 eth_addr_random(br->default_ea);
919 port_array_init(&br->ifaces);
922 br->bond_next_rebalance = time_msec() + 10000;
924 list_push_back(&all_bridges, &br->node);
926 VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif));
932 bridge_destroy(struct bridge *br)
937 while (br->n_ports > 0) {
938 port_destroy(br->ports[br->n_ports - 1]);
940 list_remove(&br->node);
941 error = dpif_delete(&br->dpif);
942 if (error && error != ENOENT) {
943 VLOG_ERR("failed to delete dp%u: %s",
944 dpif_id(&br->dpif), strerror(error));
946 dpif_close(&br->dpif);
947 ofproto_destroy(br->ofproto);
948 free(br->controller);
949 mac_learning_destroy(br->ml);
950 port_array_destroy(&br->ifaces);
957 static struct bridge *
958 bridge_lookup(const char *name)
962 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
963 if (!strcmp(br->name, name)) {
971 bridge_exists(const char *name)
973 return bridge_lookup(name) ? true : false;
977 bridge_get_datapathid(const char *name)
979 struct bridge *br = bridge_lookup(name);
980 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
983 /* Handle requests for a listing of all flows known by the OpenFlow
984 * stack, including those normally hidden. */
986 bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args)
991 br = bridge_lookup(args);
993 unixctl_command_reply(conn, 501, "Unknown bridge");
998 ofproto_get_all_flows(br->ofproto, &results);
1000 unixctl_command_reply(conn, 200, ds_cstr(&results));
1001 ds_destroy(&results);
1005 bridge_run_one(struct bridge *br)
1009 error = ofproto_run1(br->ofproto);
1015 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1020 error = ofproto_run2(br->ofproto, br->flush);
1027 bridge_get_controller(const struct bridge *br)
1029 const char *controller;
1031 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1033 controller = cfg_get_string(0, "mgmt.controller");
1035 return controller && controller[0] ? controller : NULL;
1039 bridge_reconfigure_one(struct bridge *br)
1041 struct svec old_ports, new_ports, ifaces;
1042 struct svec listeners, old_listeners;
1043 struct svec snoops, old_snoops;
1046 /* Collect old ports. */
1047 svec_init(&old_ports);
1048 for (i = 0; i < br->n_ports; i++) {
1049 svec_add(&old_ports, br->ports[i]->name);
1051 svec_sort(&old_ports);
1052 assert(svec_is_unique(&old_ports));
1054 /* Collect new ports. */
1055 svec_init(&new_ports);
1056 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1057 svec_sort(&new_ports);
1058 if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) {
1059 svec_add(&new_ports, br->name);
1060 svec_sort(&new_ports);
1062 if (!svec_is_unique(&new_ports)) {
1063 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1064 br->name, svec_get_duplicate(&new_ports));
1065 svec_unique(&new_ports);
1068 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1070 /* Get rid of deleted ports and add new ports. */
1071 for (i = 0; i < br->n_ports; ) {
1072 struct port *port = br->ports[i];
1073 if (!svec_contains(&new_ports, port->name)) {
1079 for (i = 0; i < new_ports.n; i++) {
1080 const char *name = new_ports.names[i];
1081 if (!svec_contains(&old_ports, name)) {
1082 port_create(br, name);
1085 svec_destroy(&old_ports);
1086 svec_destroy(&new_ports);
1088 /* Reconfigure all ports. */
1089 for (i = 0; i < br->n_ports; i++) {
1090 port_reconfigure(br->ports[i]);
1093 /* Check and delete duplicate interfaces. */
1095 for (i = 0; i < br->n_ports; ) {
1096 struct port *port = br->ports[i];
1097 for (j = 0; j < port->n_ifaces; ) {
1098 struct iface *iface = port->ifaces[j];
1099 if (svec_contains(&ifaces, iface->name)) {
1100 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1102 br->name, iface->name, port->name);
1103 iface_destroy(iface);
1105 svec_add(&ifaces, iface->name);
1110 if (!port->n_ifaces) {
1111 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1117 svec_destroy(&ifaces);
1119 /* Delete all flows if we're switching from connected to standalone or vice
1120 * versa. (XXX Should we delete all flows if we are switching from one
1121 * controller to another?) */
1123 /* Configure OpenFlow management listeners. */
1124 svec_init(&listeners);
1125 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1127 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1128 ovs_rundir, br->name));
1129 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1130 svec_clear(&listeners);
1132 svec_sort_unique(&listeners);
1134 svec_init(&old_listeners);
1135 ofproto_get_listeners(br->ofproto, &old_listeners);
1136 svec_sort_unique(&old_listeners);
1138 if (!svec_equal(&listeners, &old_listeners)) {
1139 ofproto_set_listeners(br->ofproto, &listeners);
1141 svec_destroy(&listeners);
1142 svec_destroy(&old_listeners);
1144 /* Configure OpenFlow controller connection snooping. */
1146 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1148 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1149 ovs_rundir, br->name));
1150 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1151 svec_clear(&snoops);
1153 svec_sort_unique(&snoops);
1155 svec_init(&old_snoops);
1156 ofproto_get_snoops(br->ofproto, &old_snoops);
1157 svec_sort_unique(&old_snoops);
1159 if (!svec_equal(&snoops, &old_snoops)) {
1160 ofproto_set_snoops(br->ofproto, &snoops);
1162 svec_destroy(&snoops);
1163 svec_destroy(&old_snoops);
1165 mirror_reconfigure(br);
1169 bridge_reconfigure_controller(struct bridge *br)
1171 char *pfx = xasprintf("bridge.%s.controller", br->name);
1172 const char *controller;
1174 controller = bridge_get_controller(br);
1175 if ((br->controller != NULL) != (controller != NULL)) {
1176 ofproto_flush_flows(br->ofproto);
1178 free(br->controller);
1179 br->controller = controller ? xstrdup(controller) : NULL;
1182 const char *fail_mode;
1183 int max_backoff, probe;
1184 int rate_limit, burst_limit;
1186 if (!strcmp(controller, "discover")) {
1187 bool update_resolv_conf = true;
1189 if (cfg_has("%s.update-resolv.conf", pfx)) {
1190 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1193 ofproto_set_discovery(br->ofproto, true,
1194 cfg_get_string(0, "%s.accept-regex", pfx),
1195 update_resolv_conf);
1197 struct netdev *netdev;
1201 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1203 || cfg_get_bool(0, "%s.in-band", pfx));
1204 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1205 ofproto_set_in_band(br->ofproto, in_band);
1207 error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev);
1209 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1210 struct in_addr ip, mask, gateway;
1211 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1212 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1213 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1215 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1217 mask.s_addr = guess_netmask(ip.s_addr);
1219 if (!netdev_set_in4(netdev, ip, mask)) {
1220 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1222 br->name, IP_ARGS(&ip.s_addr),
1223 IP_ARGS(&mask.s_addr));
1226 if (gateway.s_addr) {
1227 if (!netdev_add_router(gateway)) {
1228 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1229 br->name, IP_ARGS(&gateway.s_addr));
1233 netdev_close(netdev);
1237 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1239 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1241 ofproto_set_failure(br->ofproto,
1243 || !strcmp(fail_mode, "standalone")
1244 || !strcmp(fail_mode, "open")));
1246 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1248 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1253 ofproto_set_probe_interval(br->ofproto, probe);
1255 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1257 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1262 ofproto_set_max_backoff(br->ofproto, max_backoff);
1264 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1266 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1268 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1270 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1272 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1274 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1276 if (cfg_has("%s.commands.acl", pfx)) {
1277 struct svec command_acls;
1280 svec_init(&command_acls);
1281 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1282 command_acl = svec_join(&command_acls, ",", "");
1284 ofproto_set_remote_execution(br->ofproto, command_acl,
1285 cfg_get_string(0, "%s.commands.dir",
1288 svec_destroy(&command_acls);
1291 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1294 union ofp_action action;
1297 /* Set up a flow that matches every packet and directs them to
1298 * OFPP_NORMAL (which goes to us). */
1299 memset(&action, 0, sizeof action);
1300 action.type = htons(OFPAT_OUTPUT);
1301 action.output.len = htons(sizeof action);
1302 action.output.port = htons(OFPP_NORMAL);
1303 memset(&flow, 0, sizeof flow);
1304 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1307 ofproto_set_in_band(br->ofproto, false);
1308 ofproto_set_max_backoff(br->ofproto, 1);
1309 ofproto_set_probe_interval(br->ofproto, 5);
1310 ofproto_set_failure(br->ofproto, false);
1311 ofproto_set_stp(br->ofproto, false);
1315 ofproto_set_controller(br->ofproto, br->controller);
1319 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1324 for (i = 0; i < br->n_ports; i++) {
1325 struct port *port = br->ports[i];
1326 for (j = 0; j < port->n_ifaces; j++) {
1327 struct iface *iface = port->ifaces[j];
1328 svec_add(ifaces, iface->name);
1330 if (port->n_ifaces > 1
1331 && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
1332 svec_add(ifaces, port->name);
1335 svec_sort_unique(ifaces);
1338 /* For robustness, in case the administrator moves around datapath ports behind
1339 * our back, we re-check all the datapath port numbers here.
1341 * This function will set the 'dp_ifidx' members of interfaces that have
1342 * disappeared to -1, so only call this function from a context where those
1343 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1344 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1345 * datapath, which doesn't support UINT16_MAX+1 ports. */
1347 bridge_fetch_dp_ifaces(struct bridge *br)
1349 struct odp_port *dpif_ports;
1350 size_t n_dpif_ports;
1353 /* Reset all interface numbers. */
1354 for (i = 0; i < br->n_ports; i++) {
1355 struct port *port = br->ports[i];
1356 for (j = 0; j < port->n_ifaces; j++) {
1357 struct iface *iface = port->ifaces[j];
1358 iface->dp_ifidx = -1;
1361 port_array_clear(&br->ifaces);
1363 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
1364 for (i = 0; i < n_dpif_ports; i++) {
1365 struct odp_port *p = &dpif_ports[i];
1366 struct iface *iface = iface_lookup(br, p->devname);
1368 if (iface->dp_ifidx >= 0) {
1369 VLOG_WARN("dp%u reported interface %s twice",
1370 dpif_id(&br->dpif), p->devname);
1371 } else if (iface_from_dp_ifidx(br, p->port)) {
1372 VLOG_WARN("dp%u reported interface %"PRIu16" twice",
1373 dpif_id(&br->dpif), p->port);
1375 port_array_set(&br->ifaces, p->port, iface);
1376 iface->dp_ifidx = p->port;
1383 /* Bridge packet processing functions. */
1386 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1388 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1391 static struct bond_entry *
1392 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1394 return &port->bond_hash[bond_hash(mac)];
1398 bond_choose_iface(const struct port *port)
1401 for (i = 0; i < port->n_ifaces; i++) {
1402 if (port->ifaces[i]->enabled) {
1410 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1411 uint16_t *dp_ifidx, tag_type *tags)
1413 struct iface *iface;
1415 assert(port->n_ifaces);
1416 if (port->n_ifaces == 1) {
1417 iface = port->ifaces[0];
1419 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1420 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1421 || !port->ifaces[e->iface_idx]->enabled) {
1422 /* XXX select interface properly. The current interface selection
1423 * is only good for testing the rebalancing code. */
1424 e->iface_idx = bond_choose_iface(port);
1425 if (e->iface_idx < 0) {
1426 *tags |= port->no_ifaces_tag;
1429 e->iface_tag = tag_create_random();
1430 ((struct port *) port)->bond_compat_is_stale = true;
1432 *tags |= e->iface_tag;
1433 iface = port->ifaces[e->iface_idx];
1435 *dp_ifidx = iface->dp_ifidx;
1436 *tags |= iface->tag; /* Currently only used for bonding. */
1441 bond_link_status_update(struct iface *iface, bool carrier)
1443 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1444 struct port *port = iface->port;
1446 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1447 /* Nothing to do. */
1450 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1451 iface->name, carrier ? "detected" : "dropped");
1452 if (carrier == iface->enabled) {
1453 iface->delay_expires = LLONG_MAX;
1454 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1455 iface->name, carrier ? "disabled" : "enabled");
1456 } else if (carrier && port->updelay && port->active_iface < 0) {
1457 iface->delay_expires = time_msec();
1458 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1459 "other interface is up", iface->name, port->updelay);
1461 int delay = carrier ? port->updelay : port->downdelay;
1462 iface->delay_expires = time_msec() + delay;
1465 "interface %s: will be %s if it stays %s for %d ms",
1467 carrier ? "enabled" : "disabled",
1468 carrier ? "up" : "down",
1475 bond_choose_active_iface(struct port *port)
1477 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1479 port->active_iface = bond_choose_iface(port);
1480 port->active_iface_tag = tag_create_random();
1481 if (port->active_iface >= 0) {
1482 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1483 port->name, port->ifaces[port->active_iface]->name);
1485 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1491 bond_enable_slave(struct iface *iface, bool enable)
1493 struct port *port = iface->port;
1494 struct bridge *br = port->bridge;
1496 iface->delay_expires = LLONG_MAX;
1497 if (enable == iface->enabled) {
1501 iface->enabled = enable;
1502 if (!iface->enabled) {
1503 VLOG_WARN("interface %s: disabled", iface->name);
1504 ofproto_revalidate(br->ofproto, iface->tag);
1505 if (iface->port_ifidx == port->active_iface) {
1506 ofproto_revalidate(br->ofproto,
1507 port->active_iface_tag);
1508 bond_choose_active_iface(port);
1510 bond_send_learning_packets(port);
1512 VLOG_WARN("interface %s: enabled", iface->name);
1513 if (port->active_iface < 0) {
1514 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1515 bond_choose_active_iface(port);
1516 bond_send_learning_packets(port);
1518 iface->tag = tag_create_random();
1520 port_update_bond_compat(port);
1524 bond_run(struct bridge *br)
1528 for (i = 0; i < br->n_ports; i++) {
1529 struct port *port = br->ports[i];
1531 if (port->bond_compat_is_stale) {
1532 port->bond_compat_is_stale = false;
1533 port_update_bond_compat(port);
1536 if (port->n_ifaces < 2) {
1539 for (j = 0; j < port->n_ifaces; j++) {
1540 struct iface *iface = port->ifaces[j];
1541 if (time_msec() >= iface->delay_expires) {
1542 bond_enable_slave(iface, !iface->enabled);
1549 bond_wait(struct bridge *br)
1553 for (i = 0; i < br->n_ports; i++) {
1554 struct port *port = br->ports[i];
1555 if (port->n_ifaces < 2) {
1558 for (j = 0; j < port->n_ifaces; j++) {
1559 struct iface *iface = port->ifaces[j];
1560 if (iface->delay_expires != LLONG_MAX) {
1561 poll_timer_wait(iface->delay_expires - time_msec());
1568 set_dst(struct dst *p, const flow_t *flow,
1569 const struct port *in_port, const struct port *out_port,
1574 * XXX This uses too many tags: any broadcast flow will get one tag per
1575 * destination port, and thus a broadcast on a switch of any size is likely
1576 * to have all tag bits set. We should figure out a way to be smarter.
1578 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1579 *tags |= out_port->stp_state_tag;
1580 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1584 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1585 : in_port->vlan >= 0 ? in_port->vlan
1586 : ntohs(flow->dl_vlan));
1587 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1591 swap_dst(struct dst *p, struct dst *q)
1593 struct dst tmp = *p;
1598 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1599 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1600 * that we push to the datapath. We could in fact fully sort the array by
1601 * vlan, but in most cases there are at most two different vlan tags so that's
1602 * possibly overkill.) */
1604 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1606 struct dst *first = dsts;
1607 struct dst *last = dsts + n_dsts;
1609 while (first != last) {
1611 * - All dsts < first have vlan == 'vlan'.
1612 * - All dsts >= last have vlan != 'vlan'.
1613 * - first < last. */
1614 while (first->vlan == vlan) {
1615 if (++first == last) {
1620 /* Same invariants, plus one additional:
1621 * - first->vlan != vlan.
1623 while (last[-1].vlan != vlan) {
1624 if (--last == first) {
1629 /* Same invariants, plus one additional:
1630 * - last[-1].vlan == vlan.*/
1631 swap_dst(first++, --last);
1636 mirror_mask_ffs(mirror_mask_t mask)
1638 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1643 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1644 const struct dst *test)
1647 for (i = 0; i < n_dsts; i++) {
1648 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1656 port_trunks_vlan(const struct port *port, uint16_t vlan)
1658 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1662 port_includes_vlan(const struct port *port, uint16_t vlan)
1664 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1668 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1669 const struct port *in_port, const struct port *out_port,
1670 struct dst dsts[], tag_type *tags)
1672 mirror_mask_t mirrors = in_port->src_mirrors;
1673 struct dst *dst = dsts;
1676 *tags |= in_port->stp_state_tag;
1677 if (out_port == FLOOD_PORT) {
1678 /* XXX use ODP_FLOOD if no vlans or bonding. */
1679 /* XXX even better, define each VLAN as a datapath port group */
1680 for (i = 0; i < br->n_ports; i++) {
1681 struct port *port = br->ports[i];
1682 if (port != in_port && port_includes_vlan(port, vlan)
1683 && !port->is_mirror_output_port
1684 && set_dst(dst, flow, in_port, port, tags)) {
1685 mirrors |= port->dst_mirrors;
1689 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1690 mirrors |= out_port->dst_mirrors;
1695 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1696 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1698 if (set_dst(dst, flow, in_port, m->out_port, tags)
1699 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1703 for (i = 0; i < br->n_ports; i++) {
1704 struct port *port = br->ports[i];
1705 if (port_includes_vlan(port, m->out_vlan)
1706 && set_dst(dst, flow, in_port, port, tags))
1708 if (port->vlan < 0) {
1709 dst->vlan = m->out_vlan;
1711 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1714 if (dst->dp_ifidx == flow->in_port
1715 && dst->vlan == vlan) {
1716 /* Don't send out input port on same VLAN. */
1724 mirrors &= mirrors - 1;
1727 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1732 print_dsts(const struct dst *dsts, size_t n)
1734 for (; n--; dsts++) {
1735 printf(">p%"PRIu16, dsts->dp_ifidx);
1736 if (dsts->vlan != OFP_VLAN_NONE) {
1737 printf("v%"PRIu16, dsts->vlan);
1743 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1744 const struct port *in_port, const struct port *out_port,
1745 tag_type *tags, struct odp_actions *actions)
1747 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1749 const struct dst *p;
1752 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1754 cur_vlan = ntohs(flow->dl_vlan);
1755 for (p = dsts; p < &dsts[n_dsts]; p++) {
1756 union odp_action *a;
1757 if (p->vlan != cur_vlan) {
1758 if (p->vlan == OFP_VLAN_NONE) {
1759 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1761 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1762 a->vlan_vid.vlan_vid = htons(p->vlan);
1766 a = odp_actions_add(actions, ODPAT_OUTPUT);
1767 a->output.port = p->dp_ifidx;
1772 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1774 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1775 return (flow->dl_type == htons(ETH_TYPE_ARP)
1776 && eth_addr_is_broadcast(flow->dl_dst)
1777 && packet->size >= sizeof(struct arp_eth_header)
1778 && arp->ar_op == ARP_OP_REQUEST);
1781 /* If the composed actions may be applied to any packet in the given 'flow',
1782 * returns true. Otherwise, the actions should only be applied to 'packet', or
1783 * not at all, if 'packet' was NULL. */
1785 process_flow(struct bridge *br, const flow_t *flow,
1786 const struct ofpbuf *packet, struct odp_actions *actions,
1789 struct iface *in_iface;
1790 struct port *in_port;
1791 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1794 /* Find the interface and port structure for the received packet. */
1795 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1797 /* No interface? Something fishy... */
1798 if (packet != NULL) {
1799 /* Odd. A few possible reasons here:
1801 * - We deleted an interface but there are still a few packets
1802 * queued up from it.
1804 * - Someone externally added an interface (e.g. with "ovs-dpctl
1805 * add-if") that we don't know about.
1807 * - Packet arrived on the local port but the local port is not
1808 * one of our bridge ports.
1810 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1812 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1813 "interface %"PRIu16, br->name, flow->in_port);
1816 /* Return without adding any actions, to drop packets on this flow. */
1819 in_port = in_iface->port;
1821 /* Figure out what VLAN this packet belongs to.
1823 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1824 * belongs to VLAN 0, so we should treat both cases identically. (In the
1825 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1826 * presumably to allow a priority to be specified. In the latter case, the
1827 * packet does not have any 802.1Q header.) */
1828 vlan = ntohs(flow->dl_vlan);
1829 if (vlan == OFP_VLAN_NONE) {
1832 if (in_port->vlan >= 0) {
1834 /* XXX support double tagging? */
1835 if (packet != NULL) {
1836 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1837 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1838 "packet received on port %s configured with "
1839 "implicit VLAN %"PRIu16,
1840 br->name, ntohs(flow->dl_vlan),
1841 in_port->name, in_port->vlan);
1845 vlan = in_port->vlan;
1847 if (!port_includes_vlan(in_port, vlan)) {
1848 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1849 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1850 "packet received on port %s not configured for "
1852 br->name, vlan, in_port->name, vlan);
1857 /* Drop frames for ports that STP wants entirely killed (both for
1858 * forwarding and for learning). Later, after we do learning, we'll drop
1859 * the frames that STP wants to do learning but not forwarding on. */
1860 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1864 /* Drop frames for reserved multicast addresses. */
1865 if (eth_addr_is_reserved(flow->dl_dst)) {
1869 /* Drop frames on ports reserved for mirroring. */
1870 if (in_port->is_mirror_output_port) {
1871 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1872 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1873 "which is reserved exclusively for mirroring",
1874 br->name, in_port->name);
1878 /* Multicast (and broadcast) packets on bonds need special attention, to
1879 * avoid receiving duplicates. */
1880 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1881 *tags |= in_port->active_iface_tag;
1882 if (in_port->active_iface != in_iface->port_ifidx) {
1883 /* Drop all multicast packets on inactive slaves. */
1886 /* Drop all multicast packets for which we have learned a different
1887 * input port, because we probably sent the packet on one slave
1888 * and got it back on the active slave. Broadcast ARP replies are
1889 * an exception to this rule: the host has moved to another
1891 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1892 if (src_idx != -1 && src_idx != in_port->port_idx) {
1894 if (!is_bcast_arp_reply(flow, packet)) {
1898 /* No way to know whether it's an ARP reply, because the
1899 * flow entry doesn't include enough information and we
1900 * don't have a packet. Punt. */
1908 out_port = FLOOD_PORT;
1912 /* Learn source MAC (but don't try to learn from revalidation). */
1914 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1915 vlan, in_port->port_idx);
1917 /* The log messages here could actually be useful in debugging,
1918 * so keep the rate limit relatively high. */
1919 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1921 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1922 "on port %s in VLAN %d",
1923 br->name, ETH_ADDR_ARGS(flow->dl_src),
1924 in_port->name, vlan);
1925 ofproto_revalidate(br->ofproto, rev_tag);
1929 /* Determine output port. */
1930 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1932 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1933 out_port = br->ports[out_port_idx];
1937 /* Don't send packets out their input ports. Don't forward frames that STP
1938 * wants us to discard. */
1939 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1944 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1947 * We send out only a single packet, instead of setting up a flow, if the
1948 * packet is an ARP directed to broadcast that arrived on a bonded
1949 * interface. In such a situation ARP requests and replies must be handled
1950 * differently, but OpenFlow unfortunately can't distinguish them.
1952 return (in_port->n_ifaces < 2
1953 || flow->dl_type != htons(ETH_TYPE_ARP)
1954 || !eth_addr_is_broadcast(flow->dl_dst));
1957 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1960 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1961 const struct ofp_phy_port *opp,
1964 struct bridge *br = br_;
1965 struct iface *iface;
1968 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1974 if (reason == OFPPR_DELETE) {
1975 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1976 br->name, iface->name);
1977 iface_destroy(iface);
1978 if (!port->n_ifaces) {
1979 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1980 br->name, port->name);
1986 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1987 if (port->n_ifaces > 1) {
1988 bool up = !(opp->state & OFPPS_LINK_DOWN);
1989 bond_link_status_update(iface, up);
1990 port_update_bond_compat(port);
1996 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1997 struct odp_actions *actions, tag_type *tags, void *br_)
1999 struct bridge *br = br_;
2002 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
2003 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
2004 brstp_receive(br, flow, payload);
2009 COVERAGE_INC(bridge_process_flow);
2010 return process_flow(br, flow, packet, actions, tags);
2014 bridge_account_flow_ofhook_cb(const flow_t *flow,
2015 const union odp_action *actions,
2016 size_t n_actions, unsigned long long int n_bytes,
2019 struct bridge *br = br_;
2020 const union odp_action *a;
2022 if (!br->has_bonded_ports) {
2026 for (a = actions; a < &actions[n_actions]; a++) {
2027 if (a->type == ODPAT_OUTPUT) {
2028 struct port *port = port_from_dp_ifidx(br, a->output.port);
2029 if (port && port->n_ifaces >= 2) {
2030 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2031 e->tx_bytes += n_bytes;
2038 bridge_account_checkpoint_ofhook_cb(void *br_)
2040 struct bridge *br = br_;
2043 if (!br->has_bonded_ports) {
2047 /* The current ofproto implementation calls this callback at least once a
2048 * second, so this timer implementation is sufficient. */
2049 if (time_msec() < br->bond_next_rebalance) {
2052 br->bond_next_rebalance = time_msec() + 10000;
2054 for (i = 0; i < br->n_ports; i++) {
2055 struct port *port = br->ports[i];
2056 if (port->n_ifaces > 1) {
2057 bond_rebalance_port(port);
2062 static struct ofhooks bridge_ofhooks = {
2063 bridge_port_changed_ofhook_cb,
2064 bridge_normal_ofhook_cb,
2065 bridge_account_flow_ofhook_cb,
2066 bridge_account_checkpoint_ofhook_cb,
2069 /* Bonding functions. */
2071 /* Statistics for a single interface on a bonded port, used for load-based
2072 * bond rebalancing. */
2073 struct slave_balance {
2074 struct iface *iface; /* The interface. */
2075 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2077 /* All the "bond_entry"s that are assigned to this interface, in order of
2078 * increasing tx_bytes. */
2079 struct bond_entry **hashes;
2083 /* Sorts pointers to pointers to bond_entries in ascending order by the
2084 * interface to which they are assigned, and within a single interface in
2085 * ascending order of bytes transmitted. */
2087 compare_bond_entries(const void *a_, const void *b_)
2089 const struct bond_entry *const *ap = a_;
2090 const struct bond_entry *const *bp = b_;
2091 const struct bond_entry *a = *ap;
2092 const struct bond_entry *b = *bp;
2093 if (a->iface_idx != b->iface_idx) {
2094 return a->iface_idx > b->iface_idx ? 1 : -1;
2095 } else if (a->tx_bytes != b->tx_bytes) {
2096 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2102 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2103 * *descending* order by number of bytes transmitted. */
2105 compare_slave_balance(const void *a_, const void *b_)
2107 const struct slave_balance *a = a_;
2108 const struct slave_balance *b = b_;
2109 if (a->iface->enabled != b->iface->enabled) {
2110 return a->iface->enabled ? -1 : 1;
2111 } else if (a->tx_bytes != b->tx_bytes) {
2112 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2119 swap_bals(struct slave_balance *a, struct slave_balance *b)
2121 struct slave_balance tmp = *a;
2126 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2127 * given that 'p' (and only 'p') might be in the wrong location.
2129 * This function invalidates 'p', since it might now be in a different memory
2132 resort_bals(struct slave_balance *p,
2133 struct slave_balance bals[], size_t n_bals)
2136 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2137 swap_bals(p, p - 1);
2139 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2140 swap_bals(p, p + 1);
2146 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2148 if (VLOG_IS_DBG_ENABLED()) {
2149 struct ds ds = DS_EMPTY_INITIALIZER;
2150 const struct slave_balance *b;
2152 for (b = bals; b < bals + n_bals; b++) {
2156 ds_put_char(&ds, ',');
2158 ds_put_format(&ds, " %s %"PRIu64"kB",
2159 b->iface->name, b->tx_bytes / 1024);
2161 if (!b->iface->enabled) {
2162 ds_put_cstr(&ds, " (disabled)");
2164 if (b->n_hashes > 0) {
2165 ds_put_cstr(&ds, " (");
2166 for (i = 0; i < b->n_hashes; i++) {
2167 const struct bond_entry *e = b->hashes[i];
2169 ds_put_cstr(&ds, " + ");
2171 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2172 e - port->bond_hash, e->tx_bytes / 1024);
2174 ds_put_cstr(&ds, ")");
2177 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2182 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2184 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2185 struct bond_entry *hash)
2187 struct port *port = from->iface->port;
2188 uint64_t delta = hash->tx_bytes;
2190 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2191 "from %s to %s (now carrying %"PRIu64"kB and "
2192 "%"PRIu64"kB load, respectively)",
2193 port->name, delta / 1024, hash - port->bond_hash,
2194 from->iface->name, to->iface->name,
2195 (from->tx_bytes - delta) / 1024,
2196 (to->tx_bytes + delta) / 1024);
2198 /* Delete element from from->hashes.
2200 * We don't bother to add the element to to->hashes because not only would
2201 * it require more work, the only purpose it would be to allow that hash to
2202 * be migrated to another slave in this rebalancing run, and there is no
2203 * point in doing that. */
2204 if (from->hashes[0] == hash) {
2207 int i = hash - from->hashes[0];
2208 memmove(from->hashes + i, from->hashes + i + 1,
2209 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2213 /* Shift load away from 'from' to 'to'. */
2214 from->tx_bytes -= delta;
2215 to->tx_bytes += delta;
2217 /* Arrange for flows to be revalidated. */
2218 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2219 hash->iface_idx = to->iface->port_ifidx;
2220 hash->iface_tag = tag_create_random();
2224 bond_rebalance_port(struct port *port)
2226 struct slave_balance bals[DP_MAX_PORTS];
2228 struct bond_entry *hashes[BOND_MASK + 1];
2229 struct slave_balance *b, *from, *to;
2230 struct bond_entry *e;
2233 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2234 * descending order of tx_bytes, so that bals[0] represents the most
2235 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2238 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2239 * array for each slave_balance structure, we sort our local array of
2240 * hashes in order by slave, so that all of the hashes for a given slave
2241 * become contiguous in memory, and then we point each 'hashes' members of
2242 * a slave_balance structure to the start of a contiguous group. */
2243 n_bals = port->n_ifaces;
2244 for (b = bals; b < &bals[n_bals]; b++) {
2245 b->iface = port->ifaces[b - bals];
2250 for (i = 0; i <= BOND_MASK; i++) {
2251 hashes[i] = &port->bond_hash[i];
2253 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2254 for (i = 0; i <= BOND_MASK; i++) {
2256 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2257 b = &bals[e->iface_idx];
2258 b->tx_bytes += e->tx_bytes;
2260 b->hashes = &hashes[i];
2265 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2266 log_bals(bals, n_bals, port);
2268 /* Discard slaves that aren't enabled (which were sorted to the back of the
2269 * array earlier). */
2270 while (!bals[n_bals - 1].iface->enabled) {
2277 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2278 to = &bals[n_bals - 1];
2279 for (from = bals; from < to; ) {
2280 uint64_t overload = from->tx_bytes - to->tx_bytes;
2281 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2282 /* The extra load on 'from' (and all less-loaded slaves), compared
2283 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2284 * it is less than ~1Mbps. No point in rebalancing. */
2286 } else if (from->n_hashes == 1) {
2287 /* 'from' only carries a single MAC hash, so we can't shift any
2288 * load away from it, even though we want to. */
2291 /* 'from' is carrying significantly more load than 'to', and that
2292 * load is split across at least two different hashes. Pick a hash
2293 * to migrate to 'to' (the least-loaded slave), given that doing so
2294 * must not cause 'to''s load to exceed 'from''s load.
2296 * The sort order we use means that we prefer to shift away the
2297 * smallest hashes instead of the biggest ones. There is little
2298 * reason behind this decision; we could use the opposite sort
2299 * order to shift away big hashes ahead of small ones. */
2302 for (i = 0; i < from->n_hashes; i++) {
2303 uint64_t delta = from->hashes[i]->tx_bytes;
2304 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2308 if (i < from->n_hashes) {
2309 bond_shift_load(from, to, from->hashes[i]);
2311 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2312 * point to different slave_balance structures. It is only
2313 * valid to do these two operations in a row at all because we
2314 * know that 'from' will not move past 'to' and vice versa. */
2315 resort_bals(from, bals, n_bals);
2316 resort_bals(to, bals, n_bals);
2320 port->bond_compat_is_stale = true;
2324 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2325 * historical data to decay to <1% in 7 rebalancing runs. */
2326 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2332 bond_send_learning_packets(struct port *port)
2334 struct bridge *br = port->bridge;
2335 struct mac_entry *e;
2336 struct ofpbuf packet;
2337 int error, n_packets, n_errors;
2339 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2343 ofpbuf_init(&packet, 128);
2344 error = n_packets = n_errors = 0;
2345 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2346 union ofp_action actions[2], *a;
2352 if (e->port == port->port_idx
2353 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2357 /* Compose actions. */
2358 memset(actions, 0, sizeof actions);
2361 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2362 a->vlan_vid.len = htons(sizeof *a);
2363 a->vlan_vid.vlan_vid = htons(e->vlan);
2366 a->output.type = htons(OFPAT_OUTPUT);
2367 a->output.len = htons(sizeof *a);
2368 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2373 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2375 flow_extract(&packet, ODPP_NONE, &flow);
2376 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2383 ofpbuf_uninit(&packet);
2386 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2387 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2388 "packets, last error was: %s",
2389 port->name, n_errors, n_packets, strerror(error));
2391 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2392 port->name, n_packets);
2396 /* Bonding unixctl user interface functions. */
2399 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2401 struct ds ds = DS_EMPTY_INITIALIZER;
2402 const struct bridge *br;
2404 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2406 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2409 for (i = 0; i < br->n_ports; i++) {
2410 const struct port *port = br->ports[i];
2411 if (port->n_ifaces > 1) {
2414 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2415 for (j = 0; j < port->n_ifaces; j++) {
2416 const struct iface *iface = port->ifaces[j];
2418 ds_put_cstr(&ds, ", ");
2420 ds_put_cstr(&ds, iface->name);
2422 ds_put_char(&ds, '\n');
2426 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2430 static struct port *
2431 bond_find(const char *name)
2433 const struct bridge *br;
2435 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2438 for (i = 0; i < br->n_ports; i++) {
2439 struct port *port = br->ports[i];
2440 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2449 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2451 struct ds ds = DS_EMPTY_INITIALIZER;
2452 const struct port *port;
2455 port = bond_find(args);
2457 unixctl_command_reply(conn, 501, "no such bond");
2461 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2462 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2463 ds_put_format(&ds, "next rebalance: %lld ms\n",
2464 port->bridge->bond_next_rebalance - time_msec());
2465 for (j = 0; j < port->n_ifaces; j++) {
2466 const struct iface *iface = port->ifaces[j];
2467 struct bond_entry *be;
2470 ds_put_format(&ds, "slave %s: %s\n",
2471 iface->name, iface->enabled ? "enabled" : "disabled");
2472 if (j == port->active_iface) {
2473 ds_put_cstr(&ds, "\tactive slave\n");
2475 if (iface->delay_expires != LLONG_MAX) {
2476 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2477 iface->enabled ? "downdelay" : "updelay",
2478 iface->delay_expires - time_msec());
2482 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2483 int hash = be - port->bond_hash;
2484 struct mac_entry *me;
2486 if (be->iface_idx != j) {
2490 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2491 hash, be->tx_bytes / 1024);
2494 if (!port->bridge->ml) {
2498 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2499 &port->bridge->ml->lrus) {
2502 if (bond_hash(me->mac) == hash
2503 && me->port != port->port_idx
2504 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2505 && dp_ifidx == iface->dp_ifidx)
2507 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2508 ETH_ADDR_ARGS(me->mac));
2513 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2518 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2520 char *args = (char *) args_;
2521 char *save_ptr = NULL;
2522 char *bond_s, *hash_s, *slave_s;
2523 uint8_t mac[ETH_ADDR_LEN];
2525 struct iface *iface;
2526 struct bond_entry *entry;
2529 bond_s = strtok_r(args, " ", &save_ptr);
2530 hash_s = strtok_r(NULL, " ", &save_ptr);
2531 slave_s = strtok_r(NULL, " ", &save_ptr);
2533 unixctl_command_reply(conn, 501,
2534 "usage: bond/migrate BOND HASH SLAVE");
2538 port = bond_find(bond_s);
2540 unixctl_command_reply(conn, 501, "no such bond");
2544 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2545 == ETH_ADDR_SCAN_COUNT) {
2546 hash = bond_hash(mac);
2547 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2548 hash = atoi(hash_s) & BOND_MASK;
2550 unixctl_command_reply(conn, 501, "bad hash");
2554 iface = port_lookup_iface(port, slave_s);
2556 unixctl_command_reply(conn, 501, "no such slave");
2560 if (!iface->enabled) {
2561 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2565 entry = &port->bond_hash[hash];
2566 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2567 entry->iface_idx = iface->port_ifidx;
2568 entry->iface_tag = tag_create_random();
2569 port->bond_compat_is_stale = true;
2570 unixctl_command_reply(conn, 200, "migrated");
2574 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2576 char *args = (char *) args_;
2577 char *save_ptr = NULL;
2578 char *bond_s, *slave_s;
2580 struct iface *iface;
2582 bond_s = strtok_r(args, " ", &save_ptr);
2583 slave_s = strtok_r(NULL, " ", &save_ptr);
2585 unixctl_command_reply(conn, 501,
2586 "usage: bond/set-active-slave BOND SLAVE");
2590 port = bond_find(bond_s);
2592 unixctl_command_reply(conn, 501, "no such bond");
2596 iface = port_lookup_iface(port, slave_s);
2598 unixctl_command_reply(conn, 501, "no such slave");
2602 if (!iface->enabled) {
2603 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2607 if (port->active_iface != iface->port_ifidx) {
2608 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2609 port->active_iface = iface->port_ifidx;
2610 port->active_iface_tag = tag_create_random();
2611 VLOG_INFO("port %s: active interface is now %s",
2612 port->name, iface->name);
2613 bond_send_learning_packets(port);
2614 unixctl_command_reply(conn, 200, "done");
2616 unixctl_command_reply(conn, 200, "no change");
2621 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2623 char *args = (char *) args_;
2624 char *save_ptr = NULL;
2625 char *bond_s, *slave_s;
2627 struct iface *iface;
2629 bond_s = strtok_r(args, " ", &save_ptr);
2630 slave_s = strtok_r(NULL, " ", &save_ptr);
2632 unixctl_command_reply(conn, 501,
2633 "usage: bond/enable/disable-slave BOND SLAVE");
2637 port = bond_find(bond_s);
2639 unixctl_command_reply(conn, 501, "no such bond");
2643 iface = port_lookup_iface(port, slave_s);
2645 unixctl_command_reply(conn, 501, "no such slave");
2649 bond_enable_slave(iface, enable);
2650 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2654 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2656 enable_slave(conn, args, true);
2660 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2662 enable_slave(conn, args, false);
2668 unixctl_command_register("bond/list", bond_unixctl_list);
2669 unixctl_command_register("bond/show", bond_unixctl_show);
2670 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2671 unixctl_command_register("bond/set-active-slave",
2672 bond_unixctl_set_active_slave);
2673 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2674 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2677 /* Port functions. */
2680 port_create(struct bridge *br, const char *name)
2684 port = xcalloc(1, sizeof *port);
2686 port->port_idx = br->n_ports;
2688 port->trunks = NULL;
2689 port->name = xstrdup(name);
2690 port->active_iface = -1;
2691 port->stp_state = STP_DISABLED;
2692 port->stp_state_tag = 0;
2694 if (br->n_ports >= br->allocated_ports) {
2695 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2698 br->ports[br->n_ports++] = port;
2700 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2705 port_reconfigure(struct port *port)
2707 bool bonded = cfg_has_section("bonding.%s", port->name);
2708 struct svec old_ifaces, new_ifaces;
2709 unsigned long *trunks;
2713 /* Collect old and new interfaces. */
2714 svec_init(&old_ifaces);
2715 svec_init(&new_ifaces);
2716 for (i = 0; i < port->n_ifaces; i++) {
2717 svec_add(&old_ifaces, port->ifaces[i]->name);
2719 svec_sort(&old_ifaces);
2721 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2722 if (!new_ifaces.n) {
2723 VLOG_ERR("port %s: no interfaces specified for bonded port",
2725 } else if (new_ifaces.n == 1) {
2726 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2730 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2731 if (port->updelay < 0) {
2734 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2735 if (port->downdelay < 0) {
2736 port->downdelay = 0;
2739 svec_init(&new_ifaces);
2740 svec_add(&new_ifaces, port->name);
2743 /* Get rid of deleted interfaces and add new interfaces. */
2744 for (i = 0; i < port->n_ifaces; i++) {
2745 struct iface *iface = port->ifaces[i];
2746 if (!svec_contains(&new_ifaces, iface->name)) {
2747 iface_destroy(iface);
2752 for (i = 0; i < new_ifaces.n; i++) {
2753 const char *name = new_ifaces.names[i];
2754 if (!svec_contains(&old_ifaces, name)) {
2755 iface_create(port, name);
2761 if (cfg_has("vlan.%s.tag", port->name)) {
2763 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2764 if (vlan >= 0 && vlan <= 4095) {
2765 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2768 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2769 * they even work as-is. But they have not been tested. */
2770 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2774 if (port->vlan != vlan) {
2776 bridge_flush(port->bridge);
2779 /* Get trunked VLANs. */
2782 size_t n_trunks, n_errors;
2785 trunks = bitmap_allocate(4096);
2786 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2788 for (i = 0; i < n_trunks; i++) {
2789 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2791 bitmap_set1(trunks, trunk);
2797 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2798 port->name, n_trunks);
2800 if (n_errors == n_trunks) {
2802 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2805 bitmap_set_multiple(trunks, 0, 4096, 1);
2808 if (cfg_has("vlan.%s.trunks", port->name)) {
2809 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2810 port->name, port->name);
2814 ? port->trunks != NULL
2815 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2816 bridge_flush(port->bridge);
2818 bitmap_free(port->trunks);
2819 port->trunks = trunks;
2821 svec_destroy(&old_ifaces);
2822 svec_destroy(&new_ifaces);
2826 port_destroy(struct port *port)
2829 struct bridge *br = port->bridge;
2833 proc_net_compat_update_vlan(port->name, NULL, 0);
2834 proc_net_compat_update_bond(port->name, NULL);
2836 for (i = 0; i < MAX_MIRRORS; i++) {
2837 struct mirror *m = br->mirrors[i];
2838 if (m && m->out_port == port) {
2843 while (port->n_ifaces > 0) {
2844 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2847 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2848 del->port_idx = port->port_idx;
2851 bitmap_free(port->trunks);
2858 static struct port *
2859 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2861 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2862 return iface ? iface->port : NULL;
2865 static struct port *
2866 port_lookup(const struct bridge *br, const char *name)
2870 for (i = 0; i < br->n_ports; i++) {
2871 struct port *port = br->ports[i];
2872 if (!strcmp(port->name, name)) {
2879 static struct iface *
2880 port_lookup_iface(const struct port *port, const char *name)
2884 for (j = 0; j < port->n_ifaces; j++) {
2885 struct iface *iface = port->ifaces[j];
2886 if (!strcmp(iface->name, name)) {
2894 port_update_bonding(struct port *port)
2896 if (port->n_ifaces < 2) {
2897 /* Not a bonded port. */
2898 if (port->bond_hash) {
2899 free(port->bond_hash);
2900 port->bond_hash = NULL;
2901 port->bond_compat_is_stale = true;
2904 if (!port->bond_hash) {
2907 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2908 for (i = 0; i <= BOND_MASK; i++) {
2909 struct bond_entry *e = &port->bond_hash[i];
2913 port->no_ifaces_tag = tag_create_random();
2914 bond_choose_active_iface(port);
2916 port->bond_compat_is_stale = true;
2921 port_update_bond_compat(struct port *port)
2923 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
2924 struct compat_bond bond;
2927 if (port->n_ifaces < 2) {
2928 proc_net_compat_update_bond(port->name, NULL);
2933 bond.updelay = port->updelay;
2934 bond.downdelay = port->downdelay;
2937 bond.hashes = compat_hashes;
2938 if (port->bond_hash) {
2939 const struct bond_entry *e;
2940 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
2941 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2942 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
2943 cbh->hash = e - port->bond_hash;
2944 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
2949 bond.n_slaves = port->n_ifaces;
2950 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2951 for (i = 0; i < port->n_ifaces; i++) {
2952 struct iface *iface = port->ifaces[i];
2953 struct compat_bond_slave *slave = &bond.slaves[i];
2954 slave->name = iface->name;
2956 /* We need to make the same determination as the Linux bonding
2957 * code to determine whether a slave should be consider "up".
2958 * The Linux function bond_miimon_inspect() supports four
2959 * BOND_LINK_* states:
2961 * - BOND_LINK_UP: carrier detected, updelay has passed.
2962 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
2963 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
2964 * - BOND_LINK_BACK: carrier detected, updelay in progress.
2966 * The function bond_info_show_slave() only considers BOND_LINK_UP
2967 * to be "up" and anything else to be "down".
2969 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
2973 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2976 proc_net_compat_update_bond(port->name, &bond);
2981 port_update_vlan_compat(struct port *port)
2983 struct bridge *br = port->bridge;
2984 char *vlandev_name = NULL;
2986 if (port->vlan > 0) {
2987 /* Figure out the name that the VLAN device should actually have, if it
2988 * existed. This takes some work because the VLAN device would not
2989 * have port->name in its name; rather, it would have the trunk port's
2990 * name, and 'port' would be attached to a bridge that also had the
2991 * VLAN device one of its ports. So we need to find a trunk port that
2992 * includes port->vlan.
2994 * There might be more than one candidate. This doesn't happen on
2995 * XenServer, so if it happens we just pick the first choice in
2996 * alphabetical order instead of creating multiple VLAN devices. */
2998 for (i = 0; i < br->n_ports; i++) {
2999 struct port *p = br->ports[i];
3000 if (port_trunks_vlan(p, port->vlan)
3002 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3004 const uint8_t *ea = p->ifaces[0]->mac;
3005 if (!eth_addr_is_multicast(ea) &&
3006 !eth_addr_is_reserved(ea) &&
3007 !eth_addr_is_zero(ea)) {
3008 vlandev_name = p->name;
3013 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3016 /* Interface functions. */
3019 iface_create(struct port *port, const char *name)
3021 struct iface *iface;
3023 iface = xcalloc(1, sizeof *iface);
3025 iface->port_ifidx = port->n_ifaces;
3026 iface->name = xstrdup(name);
3027 iface->dp_ifidx = -1;
3028 iface->tag = tag_create_random();
3029 iface->delay_expires = LLONG_MAX;
3031 if (!cfg_get_bool(0, "iface.%s.internal", iface->name)) {
3032 netdev_nodev_get_etheraddr(name, iface->mac);
3033 netdev_nodev_get_carrier(name, &iface->enabled);
3035 /* Internal interfaces are created later by the call to dpif_port_add()
3036 * in bridge_reconfigure(). Until then, we can't obtain any
3037 * information about them. (There's no real value in doing so, anyway,
3038 * because the 'mac' and 'enabled' values are only used for interfaces
3039 * that are bond slaves, and it doesn't normally make sense to bond an
3040 * internal interface.) */
3043 if (port->n_ifaces >= port->allocated_ifaces) {
3044 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3045 sizeof *port->ifaces);
3047 port->ifaces[port->n_ifaces++] = iface;
3048 if (port->n_ifaces > 1) {
3049 port->bridge->has_bonded_ports = true;
3052 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3054 port_update_bonding(port);
3055 bridge_flush(port->bridge);
3059 iface_destroy(struct iface *iface)
3062 struct port *port = iface->port;
3063 struct bridge *br = port->bridge;
3064 bool del_active = port->active_iface == iface->port_ifidx;
3067 if (iface->dp_ifidx >= 0) {
3068 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3071 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3072 del->port_ifidx = iface->port_ifidx;
3078 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3079 bond_choose_active_iface(port);
3080 bond_send_learning_packets(port);
3083 port_update_bonding(port);
3084 bridge_flush(port->bridge);
3088 static struct iface *
3089 iface_lookup(const struct bridge *br, const char *name)
3093 for (i = 0; i < br->n_ports; i++) {
3094 struct port *port = br->ports[i];
3095 for (j = 0; j < port->n_ifaces; j++) {
3096 struct iface *iface = port->ifaces[j];
3097 if (!strcmp(iface->name, name)) {
3105 static struct iface *
3106 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3108 return port_array_get(&br->ifaces, dp_ifidx);
3111 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3112 * 'br', that is, an interface that is entirely simulated within the datapath.
3113 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3114 * interfaces are created by setting "iface.<iface>.internal = true".
3116 * In addition, we have a kluge-y feature that creates an internal port with
3117 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3118 * This feature needs to go away in the long term. Until then, this is one
3119 * reason why this function takes a name instead of a struct iface: the fake
3120 * interfaces created this way do not have a struct iface. */
3122 iface_is_internal(const struct bridge *br, const char *iface)
3124 if (!strcmp(iface, br->name)
3125 || cfg_get_bool(0, "iface.%s.internal", iface)) {
3129 if (cfg_get_bool(0, "bonding.%s.fake-iface", iface)) {
3130 struct port *port = port_lookup(br, iface);
3131 if (port && port->n_ifaces > 1) {
3139 /* Port mirroring. */
3142 mirror_reconfigure(struct bridge *br)
3144 struct svec old_mirrors, new_mirrors;
3147 /* Collect old and new mirrors. */
3148 svec_init(&old_mirrors);
3149 svec_init(&new_mirrors);
3150 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3151 for (i = 0; i < MAX_MIRRORS; i++) {
3152 if (br->mirrors[i]) {
3153 svec_add(&old_mirrors, br->mirrors[i]->name);
3157 /* Get rid of deleted mirrors and add new mirrors. */
3158 svec_sort(&old_mirrors);
3159 assert(svec_is_unique(&old_mirrors));
3160 svec_sort(&new_mirrors);
3161 assert(svec_is_unique(&new_mirrors));
3162 for (i = 0; i < MAX_MIRRORS; i++) {
3163 struct mirror *m = br->mirrors[i];
3164 if (m && !svec_contains(&new_mirrors, m->name)) {
3168 for (i = 0; i < new_mirrors.n; i++) {
3169 const char *name = new_mirrors.names[i];
3170 if (!svec_contains(&old_mirrors, name)) {
3171 mirror_create(br, name);
3174 svec_destroy(&old_mirrors);
3175 svec_destroy(&new_mirrors);
3177 /* Reconfigure all mirrors. */
3178 for (i = 0; i < MAX_MIRRORS; i++) {
3179 if (br->mirrors[i]) {
3180 mirror_reconfigure_one(br->mirrors[i]);
3184 /* Update port reserved status. */
3185 for (i = 0; i < br->n_ports; i++) {
3186 br->ports[i]->is_mirror_output_port = false;
3188 for (i = 0; i < MAX_MIRRORS; i++) {
3189 struct mirror *m = br->mirrors[i];
3190 if (m && m->out_port) {
3191 m->out_port->is_mirror_output_port = true;
3197 mirror_create(struct bridge *br, const char *name)
3202 for (i = 0; ; i++) {
3203 if (i >= MAX_MIRRORS) {
3204 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3205 "cannot create %s", br->name, MAX_MIRRORS, name);
3208 if (!br->mirrors[i]) {
3213 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3216 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3219 m->name = xstrdup(name);
3220 svec_init(&m->src_ports);
3221 svec_init(&m->dst_ports);
3229 mirror_destroy(struct mirror *m)
3232 struct bridge *br = m->bridge;
3235 for (i = 0; i < br->n_ports; i++) {
3236 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3237 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3240 svec_destroy(&m->src_ports);
3241 svec_destroy(&m->dst_ports);
3244 m->bridge->mirrors[m->idx] = NULL;
3252 prune_ports(struct mirror *m, struct svec *ports)
3257 svec_sort_unique(ports);
3260 for (i = 0; i < ports->n; i++) {
3261 const char *name = ports->names[i];
3262 if (port_lookup(m->bridge, name)) {
3263 svec_add(&tmp, name);
3265 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3266 m->bridge->name, m->name, name);
3269 svec_swap(ports, &tmp);
3274 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3278 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3279 * order won't give us numeric sort order. But that's good enough for what
3280 * we need right now. */
3281 svec_sort_unique(vlan_strings);
3283 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3285 for (i = 0; i < vlan_strings->n; i++) {
3286 const char *name = vlan_strings->names[i];
3288 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3289 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3290 m->bridge->name, m->name, name);
3292 (*vlans)[n_vlans++] = vlan;
3299 vlan_is_mirrored(const struct mirror *m, int vlan)
3303 for (i = 0; i < m->n_vlans; i++) {
3304 if (m->vlans[i] == vlan) {
3312 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3316 for (i = 0; i < m->n_vlans; i++) {
3317 if (port_trunks_vlan(p, m->vlans[i])) {
3325 mirror_reconfigure_one(struct mirror *m)
3327 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3328 struct svec src_ports, dst_ports, ports;
3329 struct svec vlan_strings;
3330 mirror_mask_t mirror_bit;
3331 const char *out_port_name;
3332 struct port *out_port;
3337 bool mirror_all_ports;
3338 bool any_ports_specified;
3340 /* Get output port. */
3341 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3342 m->bridge->name, m->name);
3343 if (out_port_name) {
3344 out_port = port_lookup(m->bridge, out_port_name);
3346 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3347 "named %s", pfx, m->bridge->name, out_port_name);
3354 if (cfg_has("%s.output.vlan", pfx)) {
3355 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3356 "ignoring %s.output.vlan", pfx, pfx, pfx);
3358 } else if (cfg_has("%s.output.vlan", pfx)) {
3360 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3362 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3363 "but exactly one is required; disabling port mirror %s",
3364 pfx, pfx, pfx, pfx);
3370 /* Get all the ports, and drop duplicates and ports that don't exist. */
3371 svec_init(&src_ports);
3372 svec_init(&dst_ports);
3374 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3375 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3376 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3377 any_ports_specified = src_ports.n || dst_ports.n || ports.n;
3378 svec_append(&src_ports, &ports);
3379 svec_append(&dst_ports, &ports);
3380 svec_destroy(&ports);
3381 prune_ports(m, &src_ports);
3382 prune_ports(m, &dst_ports);
3383 if (any_ports_specified && !src_ports.n && !dst_ports.n) {
3384 VLOG_ERR("%s: none of the specified ports exist; "
3385 "disabling port mirror %s", pfx, pfx);
3390 /* Get all the vlans, and drop duplicate and invalid vlans. */
3391 svec_init(&vlan_strings);
3392 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3393 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3394 svec_destroy(&vlan_strings);
3396 /* Update mirror data. */
3397 if (!svec_equal(&m->src_ports, &src_ports)
3398 || !svec_equal(&m->dst_ports, &dst_ports)
3399 || m->n_vlans != n_vlans
3400 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3401 || m->out_port != out_port
3402 || m->out_vlan != out_vlan) {
3403 bridge_flush(m->bridge);
3405 svec_swap(&m->src_ports, &src_ports);
3406 svec_swap(&m->dst_ports, &dst_ports);
3409 m->n_vlans = n_vlans;
3410 m->out_port = out_port;
3411 m->out_vlan = out_vlan;
3413 /* If no selection criteria have been given, mirror for all ports. */
3414 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3417 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3418 for (i = 0; i < m->bridge->n_ports; i++) {
3419 struct port *port = m->bridge->ports[i];
3421 if (mirror_all_ports
3422 || svec_contains(&m->src_ports, port->name)
3425 ? port_trunks_any_mirrored_vlan(m, port)
3426 : vlan_is_mirrored(m, port->vlan)))) {
3427 port->src_mirrors |= mirror_bit;
3429 port->src_mirrors &= ~mirror_bit;
3432 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3433 port->dst_mirrors |= mirror_bit;
3435 port->dst_mirrors &= ~mirror_bit;
3441 svec_destroy(&src_ports);
3442 svec_destroy(&dst_ports);
3446 /* Spanning tree protocol. */
3448 static void brstp_update_port_state(struct port *);
3451 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3453 struct bridge *br = br_;
3454 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3455 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3457 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3459 } else if (eth_addr_is_zero(iface->mac)) {
3460 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3463 union ofp_action action;
3464 struct eth_header *eth = pkt->l2;
3467 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3469 memset(&action, 0, sizeof action);
3470 action.type = htons(OFPAT_OUTPUT);
3471 action.output.len = htons(sizeof action);
3472 action.output.port = htons(port_no);
3474 flow_extract(pkt, ODPP_NONE, &flow);
3475 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3481 brstp_reconfigure(struct bridge *br)
3485 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3487 stp_destroy(br->stp);
3493 uint64_t bridge_address, bridge_id;
3494 int bridge_priority;
3496 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3497 if (!bridge_address) {
3499 bridge_address = (stp_get_bridge_id(br->stp)
3500 & ((UINT64_C(1) << 48) - 1));
3502 uint8_t mac[ETH_ADDR_LEN];
3503 eth_addr_random(mac);
3504 bridge_address = eth_addr_to_uint64(mac);
3508 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3510 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3512 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3515 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3517 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3518 br->stp_last_tick = time_msec();
3521 if (bridge_id != stp_get_bridge_id(br->stp)) {
3522 stp_set_bridge_id(br->stp, bridge_id);
3527 for (i = 0; i < br->n_ports; i++) {
3528 struct port *p = br->ports[i];
3530 struct stp_port *sp;
3531 int path_cost, priority;
3537 dp_ifidx = p->ifaces[0]->dp_ifidx;
3538 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3542 sp = stp_get_port(br->stp, dp_ifidx);
3543 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3544 "stp.%s.port.%s.enabled",
3546 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3547 br->name, p->name));
3548 if (p->is_mirror_output_port) {
3551 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3552 bridge_flush(br); /* Might not be necessary. */
3554 stp_port_enable(sp);
3556 stp_port_disable(sp);
3560 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3562 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3564 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3565 "stp.%s.port.%s.priority",
3567 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3569 : STP_DEFAULT_PORT_PRIORITY);
3570 stp_port_set_priority(sp, priority);
3573 brstp_adjust_timers(br);
3575 for (i = 0; i < br->n_ports; i++) {
3576 brstp_update_port_state(br->ports[i]);
3581 brstp_update_port_state(struct port *p)
3583 struct bridge *br = p->bridge;
3584 enum stp_state state;
3586 /* Figure out new state. */
3587 state = STP_DISABLED;
3588 if (br->stp && p->n_ifaces > 0) {
3589 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3590 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3591 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3596 if (p->stp_state != state) {
3597 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3598 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3599 p->name, stp_state_name(p->stp_state),
3600 stp_state_name(state));
3601 if (p->stp_state == STP_DISABLED) {
3604 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3606 p->stp_state = state;
3607 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3608 : tag_create_random());
3613 brstp_adjust_timers(struct bridge *br)
3615 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3616 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3617 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3619 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3620 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3621 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3625 brstp_run(struct bridge *br)
3628 long long int now = time_msec();
3629 long long int elapsed = now - br->stp_last_tick;
3630 struct stp_port *sp;
3633 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3634 br->stp_last_tick = now;
3636 while (stp_get_changed_port(br->stp, &sp)) {
3637 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3639 brstp_update_port_state(p);
3646 brstp_wait(struct bridge *br)
3649 poll_timer_wait(1000);