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/netflow.h"
52 #include "secchan/ofproto.h"
53 #include "socket-util.h"
60 #include "vconn-ssl.h"
61 #include "xenserver.h"
64 #define THIS_MODULE VLM_bridge
72 extern uint64_t mgmt_id;
75 struct port *port; /* Containing port. */
76 size_t port_ifidx; /* Index within containing port. */
78 char *name; /* Host network device name. */
79 int dp_ifidx; /* Index within kernel datapath. */
81 uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */
83 tag_type tag; /* Tag associated with this interface. */
84 bool enabled; /* May be chosen for flows? */
85 long long delay_expires; /* Time after which 'enabled' may change. */
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. */
134 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
136 /* Port mirroring info. */
137 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
138 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
139 bool is_mirror_output_port; /* Does port mirroring send frames here? */
141 /* Spanning tree info. */
142 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
143 tag_type stp_state_tag; /* Tag for STP state change. */
146 #define DP_MAX_PORTS 255
148 struct list node; /* Node in global list of bridges. */
149 char *name; /* User-specified arbitrary name. */
150 struct mac_learning *ml; /* MAC learning table. */
151 bool sent_config_request; /* Successfully sent config request? */
152 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
154 /* Support for remote controllers. */
155 char *controller; /* NULL if there is no remote controller;
156 * "discover" to do controller discovery;
157 * otherwise a vconn name. */
159 /* OpenFlow switch processing. */
160 struct ofproto *ofproto; /* OpenFlow switch. */
162 /* Kernel datapath information. */
163 struct dpif dpif; /* Kernel datapath. */
164 struct port_array ifaces; /* Indexed by kernel datapath port number. */
168 size_t n_ports, allocated_ports;
171 bool has_bonded_ports;
172 long long int bond_next_rebalance;
177 /* Flow statistics gathering. */
178 time_t next_stats_request;
180 /* Port mirroring. */
181 struct mirror *mirrors[MAX_MIRRORS];
185 long long int stp_last_tick;
188 /* List of all bridges. */
189 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
191 /* Maximum number of datapaths. */
192 enum { DP_MAX = 256 };
194 static struct bridge *bridge_create(const char *name);
195 static void bridge_destroy(struct bridge *);
196 static struct bridge *bridge_lookup(const char *name);
197 static void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *);
198 static int bridge_run_one(struct bridge *);
199 static void bridge_reconfigure_one(struct bridge *);
200 static void bridge_reconfigure_controller(struct bridge *);
201 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
202 static void bridge_fetch_dp_ifaces(struct bridge *);
203 static void bridge_flush(struct bridge *);
204 static void bridge_pick_local_hw_addr(struct bridge *,
205 uint8_t ea[ETH_ADDR_LEN],
206 const char **devname);
207 static uint64_t bridge_pick_datapath_id(struct bridge *,
208 const uint8_t bridge_ea[ETH_ADDR_LEN],
209 const char *devname);
210 static uint64_t dpid_from_hash(const void *, size_t nbytes);
212 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
214 static void bond_init(void);
215 static void bond_run(struct bridge *);
216 static void bond_wait(struct bridge *);
217 static void bond_rebalance_port(struct port *);
218 static void bond_send_learning_packets(struct port *);
219 static void bond_enable_slave(struct iface *iface, bool enable);
221 static void port_create(struct bridge *, const char *name);
222 static void port_reconfigure(struct port *);
223 static void port_destroy(struct port *);
224 static struct port *port_lookup(const struct bridge *, const char *name);
225 static struct iface *port_lookup_iface(const struct port *, const char *name);
226 static struct port *port_from_dp_ifidx(const struct bridge *,
228 static void port_update_bond_compat(struct port *);
229 static void port_update_vlan_compat(struct port *);
231 static void mirror_create(struct bridge *, const char *name);
232 static void mirror_destroy(struct mirror *);
233 static void mirror_reconfigure(struct bridge *);
234 static void mirror_reconfigure_one(struct mirror *);
235 static bool vlan_is_mirrored(const struct mirror *, int vlan);
237 static void brstp_reconfigure(struct bridge *);
238 static void brstp_adjust_timers(struct bridge *);
239 static void brstp_run(struct bridge *);
240 static void brstp_wait(struct bridge *);
242 static void iface_create(struct port *, const char *name);
243 static void iface_destroy(struct iface *);
244 static struct iface *iface_lookup(const struct bridge *, const char *name);
245 static struct iface *iface_from_dp_ifidx(const struct bridge *,
247 static bool iface_is_internal(const struct bridge *, const char *name);
248 static void iface_set_mac(struct iface *);
250 /* Hooks into ofproto processing. */
251 static struct ofhooks bridge_ofhooks;
253 /* Public functions. */
255 /* Adds the name of each interface used by a bridge, including local and
256 * internal ports, to 'svec'. */
258 bridge_get_ifaces(struct svec *svec)
260 struct bridge *br, *next;
263 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
264 for (i = 0; i < br->n_ports; i++) {
265 struct port *port = br->ports[i];
267 for (j = 0; j < port->n_ifaces; j++) {
268 struct iface *iface = port->ifaces[j];
269 if (iface->dp_ifidx < 0) {
270 VLOG_ERR("%s interface not in dp%u, ignoring",
271 iface->name, dpif_id(&br->dpif));
273 if (iface->dp_ifidx != ODPP_LOCAL) {
274 svec_add(svec, iface->name);
282 /* The caller must already have called cfg_read(). */
291 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
293 for (i = 0; i < DP_MAX; i++) {
297 sprintf(devname, "dp%d", i);
298 retval = dpif_open(devname, &dpif);
300 char dpif_name[IF_NAMESIZE];
301 if (dpif_get_name(&dpif, dpif_name, sizeof dpif_name)
302 || !cfg_has("bridge.%s.port", dpif_name)) {
306 } else if (retval != ENODEV) {
307 VLOG_ERR("failed to delete datapath dp%d: %s",
308 i, strerror(retval));
312 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows);
314 bridge_reconfigure();
319 config_string_change(const char *key, char **valuep)
321 const char *value = cfg_get_string(0, "%s", key);
322 if (value && (!*valuep || strcmp(value, *valuep))) {
324 *valuep = xstrdup(value);
332 bridge_configure_ssl(void)
334 /* XXX SSL should be configurable on a per-bridge basis.
335 * XXX should be possible to de-configure SSL. */
336 static char *private_key_file;
337 static char *certificate_file;
338 static char *cacert_file;
341 if (config_string_change("ssl.private-key", &private_key_file)) {
342 vconn_ssl_set_private_key_file(private_key_file);
345 if (config_string_change("ssl.certificate", &certificate_file)) {
346 vconn_ssl_set_certificate_file(certificate_file);
349 /* We assume that even if the filename hasn't changed, if the CA cert
350 * file has been removed, that we want to move back into
351 * boot-strapping mode. This opens a small security hole, because
352 * the old certificate will still be trusted until vSwitch is
353 * restarted. We may want to address this in vconn's SSL library. */
354 if (config_string_change("ssl.ca-cert", &cacert_file)
355 || (stat(cacert_file, &s) && errno == ENOENT)) {
356 vconn_ssl_set_ca_cert_file(cacert_file,
357 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
363 bridge_reconfigure(void)
365 struct svec old_br, new_br, raw_new_br;
366 struct bridge *br, *next;
369 COVERAGE_INC(bridge_reconfigure);
371 /* Collect old bridges. */
373 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
374 svec_add(&old_br, br->name);
377 /* Collect new bridges. */
378 svec_init(&raw_new_br);
379 cfg_get_subsections(&raw_new_br, "bridge");
381 for (i = 0; i < raw_new_br.n; i++) {
382 const char *name = raw_new_br.names[i];
383 if ((!strncmp(name, "dp", 2) && isdigit(name[2])) ||
384 (!strncmp(name, "nl:", 3) && isdigit(name[3]))) {
385 VLOG_ERR("%s is not a valid bridge name (bridges may not be "
386 "named \"dp\" or \"nl:\" followed by a digit)", name);
388 svec_add(&new_br, name);
391 svec_destroy(&raw_new_br);
393 /* Get rid of deleted bridges and add new bridges. */
396 assert(svec_is_unique(&old_br));
397 assert(svec_is_unique(&new_br));
398 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
399 if (!svec_contains(&new_br, br->name)) {
403 for (i = 0; i < new_br.n; i++) {
404 const char *name = new_br.names[i];
405 if (!svec_contains(&old_br, name)) {
409 svec_destroy(&old_br);
410 svec_destroy(&new_br);
414 bridge_configure_ssl();
417 /* Reconfigure all bridges. */
418 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
419 bridge_reconfigure_one(br);
422 /* Add and delete ports on all datapaths.
424 * The kernel will reject any attempt to add a given port to a datapath if
425 * that port already belongs to a different datapath, so we must do all
426 * port deletions before any port additions. */
427 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
428 struct odp_port *dpif_ports;
430 struct svec want_ifaces;
432 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
433 bridge_get_all_ifaces(br, &want_ifaces);
434 for (i = 0; i < n_dpif_ports; i++) {
435 const struct odp_port *p = &dpif_ports[i];
436 if (!svec_contains(&want_ifaces, p->devname)
437 && strcmp(p->devname, br->name)) {
438 int retval = dpif_port_del(&br->dpif, p->port);
440 VLOG_ERR("failed to remove %s interface from dp%u: %s",
441 p->devname, dpif_id(&br->dpif), strerror(retval));
445 svec_destroy(&want_ifaces);
448 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
449 struct odp_port *dpif_ports;
451 struct svec cur_ifaces, want_ifaces, add_ifaces;
454 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
455 svec_init(&cur_ifaces);
456 for (i = 0; i < n_dpif_ports; i++) {
457 svec_add(&cur_ifaces, dpif_ports[i].devname);
460 svec_sort_unique(&cur_ifaces);
461 bridge_get_all_ifaces(br, &want_ifaces);
462 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
465 for (i = 0; i < add_ifaces.n; i++) {
466 const char *if_name = add_ifaces.names[i];
471 /* Add to datapath. */
472 internal = iface_is_internal(br, if_name);
473 error = dpif_port_add(&br->dpif, if_name, next_port_no++,
474 internal ? ODP_PORT_INTERNAL : 0);
475 if (error != EEXIST) {
476 if (next_port_no >= 256) {
477 VLOG_ERR("ran out of valid port numbers on dp%u",
482 VLOG_ERR("failed to add %s interface to dp%u: %s",
483 if_name, dpif_id(&br->dpif), strerror(error));
490 svec_destroy(&cur_ifaces);
491 svec_destroy(&want_ifaces);
492 svec_destroy(&add_ifaces);
494 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
497 struct iface *local_iface = NULL;
499 struct netflow_options nf_options;
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 memset(&nf_options, 0, sizeof nf_options);
545 nf_options.engine_type = br->dpif.minor;
546 nf_options.engine_id = br->dpif.minor;
547 nf_options.active_timeout = -1;
549 if (cfg_has("netflow.%s.engine-type", br->name)) {
550 nf_options.engine_type = cfg_get_int(0, "netflow.%s.engine-type",
553 if (cfg_has("netflow.%s.engine-id", br->name)) {
554 nf_options.engine_id = cfg_get_int(0, "netflow.%s.engine-id",
557 if (cfg_has("netflow.%s.active-timeout", br->name)) {
558 nf_options.active_timeout = cfg_get_int(0,
559 "netflow.%s.active-timeout",
562 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
563 nf_options.add_id_to_iface = cfg_get_bool(0,
564 "netflow.%s.add-id-to-iface",
567 if (nf_options.add_id_to_iface && nf_options.engine_id > 0x7f) {
568 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
569 "another vswitch, choose an engine id less than 128",
572 if (nf_options.add_id_to_iface && br->n_ports > 508) {
573 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
574 "another port when more than 508 ports are used",
577 svec_init(&nf_options.collectors);
578 cfg_get_all_keys(&nf_options.collectors, "netflow.%s.host", br->name);
579 if (ofproto_set_netflow(br->ofproto, &nf_options)) {
580 VLOG_ERR("bridge %s: problem setting netflow collectors",
583 svec_destroy(&nf_options.collectors);
585 /* Update the controller and related settings. It would be more
586 * straightforward to call this from bridge_reconfigure_one(), but we
587 * can't do it there for two reasons. First, and most importantly, at
588 * that point we don't know the dp_ifidx of any interfaces that have
589 * been added to the bridge (because we haven't actually added them to
590 * the datapath). Second, at that point we haven't set the datapath ID
591 * yet; when a controller is configured, resetting the datapath ID will
592 * immediately disconnect from the controller, so it's better to set
593 * the datapath ID before the controller. */
594 bridge_reconfigure_controller(br);
596 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
597 for (i = 0; i < br->n_ports; i++) {
598 struct port *port = br->ports[i];
600 port_update_vlan_compat(port);
602 for (j = 0; j < port->n_ifaces; j++) {
603 struct iface *iface = port->ifaces[j];
604 if (iface->dp_ifidx != ODPP_LOCAL
605 && iface_is_internal(br, iface->name)) {
606 iface_set_mac(iface);
611 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
612 brstp_reconfigure(br);
617 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
618 const char **devname)
620 uint64_t requested_ea;
626 /* Did the user request a particular MAC? */
627 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
629 eth_addr_from_uint64(requested_ea, ea);
630 if (eth_addr_is_multicast(ea)) {
631 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
632 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
633 } else if (eth_addr_is_zero(ea)) {
634 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
640 /* Otherwise choose the minimum MAC address among all of the interfaces.
641 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
642 * MAC of the physical interface in such an environment.) */
643 memset(ea, 0xff, sizeof ea);
644 for (i = 0; i < br->n_ports; i++) {
645 struct port *port = br->ports[i];
646 uint8_t iface_ea[ETH_ADDR_LEN];
647 uint64_t iface_ea_u64;
650 /* Mirror output ports don't participate. */
651 if (port->is_mirror_output_port) {
655 /* Choose the MAC address to represent the port. */
656 iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name);
658 /* User specified explicitly. */
659 eth_addr_from_uint64(iface_ea_u64, iface_ea);
661 /* Find the interface with this Ethernet address (if any) so that
662 * we can provide the correct devname to the caller. */
664 for (j = 0; j < port->n_ifaces; j++) {
665 struct iface *candidate = port->ifaces[j];
666 uint8_t candidate_ea[ETH_ADDR_LEN];
667 if (!netdev_nodev_get_etheraddr(candidate->name, candidate_ea)
668 && eth_addr_equals(iface_ea, candidate_ea)) {
673 /* Choose the interface whose MAC address will represent the port.
674 * The Linux kernel bonding code always chooses the MAC address of
675 * the first slave added to a bond, and the Fedora networking
676 * scripts always add slaves to a bond in alphabetical order, so
677 * for compatibility we choose the interface with the name that is
678 * first in alphabetical order. */
679 iface = port->ifaces[0];
680 for (j = 1; j < port->n_ifaces; j++) {
681 struct iface *candidate = port->ifaces[j];
682 if (strcmp(candidate->name, iface->name) < 0) {
687 /* The local port doesn't count (since we're trying to choose its
688 * MAC address anyway). Other internal ports don't count because
689 * we really want a physical MAC if we can get it, and internal
690 * ports typically have randomly generated MACs. */
691 if (iface->dp_ifidx == ODPP_LOCAL
692 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
697 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
699 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
700 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
701 iface->name, strerror(error));
706 /* Compare against our current choice. */
707 if (!eth_addr_is_multicast(iface_ea) &&
708 !eth_addr_is_reserved(iface_ea) &&
709 !eth_addr_is_zero(iface_ea) &&
710 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
712 memcpy(ea, iface_ea, ETH_ADDR_LEN);
713 *devname = iface ? iface->name : NULL;
716 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
717 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
719 VLOG_WARN("bridge %s: using default bridge Ethernet "
720 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
722 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
723 br->name, ETH_ADDR_ARGS(ea));
727 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
728 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
729 * a network device, then that network device's name must be passed in as
730 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
731 * passed in as a null pointer. */
733 bridge_pick_datapath_id(struct bridge *br,
734 const uint8_t bridge_ea[ETH_ADDR_LEN],
738 * The procedure for choosing a bridge MAC address will, in the most
739 * ordinary case, also choose a unique MAC that we can use as a datapath
740 * ID. In some special cases, though, multiple bridges will end up with
741 * the same MAC address. This is OK for the bridges, but it will confuse
742 * the OpenFlow controller, because each datapath needs a unique datapath
745 * Datapath IDs must be unique. It is also very desirable that they be
746 * stable from one run to the next, so that policy set on a datapath
751 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
758 if (!netdev_get_vlan_vid(devname, &vlan)) {
760 * A bridge whose MAC address is taken from a VLAN network device
761 * (that is, a network device created with vconfig(8) or similar
762 * tool) will have the same MAC address as a bridge on the VLAN
763 * device's physical network device.
765 * Handle this case by hashing the physical network device MAC
766 * along with the VLAN identifier.
768 uint8_t buf[ETH_ADDR_LEN + 2];
769 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
770 buf[ETH_ADDR_LEN] = vlan >> 8;
771 buf[ETH_ADDR_LEN + 1] = vlan;
772 return dpid_from_hash(buf, sizeof buf);
775 * Assume that this bridge's MAC address is unique, since it
776 * doesn't fit any of the cases we handle specially.
781 * A purely internal bridge, that is, one that has no non-virtual
782 * network devices on it at all, is more difficult because it has no
783 * natural unique identifier at all.
785 * When the host is a XenServer, we handle this case by hashing the
786 * host's UUID with the name of the bridge. Names of bridges are
787 * persistent across XenServer reboots, although they can be reused if
788 * an internal network is destroyed and then a new one is later
789 * created, so this is fairly effective.
791 * When the host is not a XenServer, we punt by using a random MAC
792 * address on each run.
794 const char *host_uuid = xenserver_get_host_uuid();
796 char *combined = xasprintf("%s,%s", host_uuid, br->name);
797 dpid = dpid_from_hash(combined, strlen(combined));
803 return eth_addr_to_uint64(bridge_ea);
807 dpid_from_hash(const void *data, size_t n)
809 uint8_t hash[SHA1_DIGEST_SIZE];
811 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
812 sha1_bytes(data, n, hash);
813 eth_addr_mark_random(hash);
814 return eth_addr_to_uint64(hash);
820 struct bridge *br, *next;
824 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
825 int error = bridge_run_one(br);
827 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
828 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
829 "forcing reconfiguration", br->name);
843 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
844 ofproto_wait(br->ofproto);
845 if (br->controller) {
849 mac_learning_wait(br->ml);
855 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
856 * configuration changes. */
858 bridge_flush(struct bridge *br)
860 COVERAGE_INC(bridge_flush);
862 mac_learning_flush(br->ml);
865 /* Bridge unixctl user interface functions. */
867 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
869 struct ds ds = DS_EMPTY_INITIALIZER;
870 const struct bridge *br;
871 const struct mac_entry *e;
873 br = bridge_lookup(args);
875 unixctl_command_reply(conn, 501, "no such bridge");
879 ds_put_cstr(&ds, " port VLAN MAC Age\n");
880 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
881 if (e->port < 0 || e->port >= br->n_ports) {
884 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
885 br->ports[e->port]->ifaces[0]->dp_ifidx,
886 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
888 unixctl_command_reply(conn, 200, ds_cstr(&ds));
892 /* Bridge reconfiguration functions. */
894 static struct bridge *
895 bridge_create(const char *name)
900 assert(!bridge_lookup(name));
901 br = xcalloc(1, sizeof *br);
903 error = dpif_create(name, &br->dpif);
904 if (error == EEXIST) {
905 error = dpif_open(name, &br->dpif);
907 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
908 name, strerror(error));
912 dpif_flow_flush(&br->dpif);
914 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
919 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
921 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
922 dpif_delete(&br->dpif);
923 dpif_close(&br->dpif);
928 br->name = xstrdup(name);
929 br->ml = mac_learning_create();
930 br->sent_config_request = false;
931 eth_addr_random(br->default_ea);
933 port_array_init(&br->ifaces);
936 br->bond_next_rebalance = time_msec() + 10000;
938 list_push_back(&all_bridges, &br->node);
940 VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif));
946 bridge_destroy(struct bridge *br)
951 while (br->n_ports > 0) {
952 port_destroy(br->ports[br->n_ports - 1]);
954 list_remove(&br->node);
955 error = dpif_delete(&br->dpif);
956 if (error && error != ENOENT) {
957 VLOG_ERR("failed to delete dp%u: %s",
958 dpif_id(&br->dpif), strerror(error));
960 dpif_close(&br->dpif);
961 ofproto_destroy(br->ofproto);
962 free(br->controller);
963 mac_learning_destroy(br->ml);
964 port_array_destroy(&br->ifaces);
971 static struct bridge *
972 bridge_lookup(const char *name)
976 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
977 if (!strcmp(br->name, name)) {
985 bridge_exists(const char *name)
987 return bridge_lookup(name) ? true : false;
991 bridge_get_datapathid(const char *name)
993 struct bridge *br = bridge_lookup(name);
994 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
997 /* Handle requests for a listing of all flows known by the OpenFlow
998 * stack, including those normally hidden. */
1000 bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args)
1005 br = bridge_lookup(args);
1007 unixctl_command_reply(conn, 501, "Unknown bridge");
1012 ofproto_get_all_flows(br->ofproto, &results);
1014 unixctl_command_reply(conn, 200, ds_cstr(&results));
1015 ds_destroy(&results);
1019 bridge_run_one(struct bridge *br)
1023 error = ofproto_run1(br->ofproto);
1028 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1032 error = ofproto_run2(br->ofproto, br->flush);
1039 bridge_get_controller(const struct bridge *br)
1041 const char *controller;
1043 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1045 controller = cfg_get_string(0, "mgmt.controller");
1047 return controller && controller[0] ? controller : NULL;
1051 bridge_reconfigure_one(struct bridge *br)
1053 struct svec old_ports, new_ports, ifaces;
1054 struct svec listeners, old_listeners;
1055 struct svec snoops, old_snoops;
1058 /* Collect old ports. */
1059 svec_init(&old_ports);
1060 for (i = 0; i < br->n_ports; i++) {
1061 svec_add(&old_ports, br->ports[i]->name);
1063 svec_sort(&old_ports);
1064 assert(svec_is_unique(&old_ports));
1066 /* Collect new ports. */
1067 svec_init(&new_ports);
1068 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1069 svec_sort(&new_ports);
1070 if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) {
1071 svec_add(&new_ports, br->name);
1072 svec_sort(&new_ports);
1074 if (!svec_is_unique(&new_ports)) {
1075 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1076 br->name, svec_get_duplicate(&new_ports));
1077 svec_unique(&new_ports);
1080 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1082 /* Get rid of deleted ports and add new ports. */
1083 for (i = 0; i < br->n_ports; ) {
1084 struct port *port = br->ports[i];
1085 if (!svec_contains(&new_ports, port->name)) {
1091 for (i = 0; i < new_ports.n; i++) {
1092 const char *name = new_ports.names[i];
1093 if (!svec_contains(&old_ports, name)) {
1094 port_create(br, name);
1097 svec_destroy(&old_ports);
1098 svec_destroy(&new_ports);
1100 /* Reconfigure all ports. */
1101 for (i = 0; i < br->n_ports; i++) {
1102 port_reconfigure(br->ports[i]);
1105 /* Check and delete duplicate interfaces. */
1107 for (i = 0; i < br->n_ports; ) {
1108 struct port *port = br->ports[i];
1109 for (j = 0; j < port->n_ifaces; ) {
1110 struct iface *iface = port->ifaces[j];
1111 if (svec_contains(&ifaces, iface->name)) {
1112 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1114 br->name, iface->name, port->name);
1115 iface_destroy(iface);
1117 svec_add(&ifaces, iface->name);
1122 if (!port->n_ifaces) {
1123 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1129 svec_destroy(&ifaces);
1131 /* Delete all flows if we're switching from connected to standalone or vice
1132 * versa. (XXX Should we delete all flows if we are switching from one
1133 * controller to another?) */
1135 /* Configure OpenFlow management listeners. */
1136 svec_init(&listeners);
1137 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1139 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1140 ovs_rundir, br->name));
1141 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1142 svec_clear(&listeners);
1144 svec_sort_unique(&listeners);
1146 svec_init(&old_listeners);
1147 ofproto_get_listeners(br->ofproto, &old_listeners);
1148 svec_sort_unique(&old_listeners);
1150 if (!svec_equal(&listeners, &old_listeners)) {
1151 ofproto_set_listeners(br->ofproto, &listeners);
1153 svec_destroy(&listeners);
1154 svec_destroy(&old_listeners);
1156 /* Configure OpenFlow controller connection snooping. */
1158 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1160 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1161 ovs_rundir, br->name));
1162 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1163 svec_clear(&snoops);
1165 svec_sort_unique(&snoops);
1167 svec_init(&old_snoops);
1168 ofproto_get_snoops(br->ofproto, &old_snoops);
1169 svec_sort_unique(&old_snoops);
1171 if (!svec_equal(&snoops, &old_snoops)) {
1172 ofproto_set_snoops(br->ofproto, &snoops);
1174 svec_destroy(&snoops);
1175 svec_destroy(&old_snoops);
1177 mirror_reconfigure(br);
1181 bridge_reconfigure_controller(struct bridge *br)
1183 char *pfx = xasprintf("bridge.%s.controller", br->name);
1184 const char *controller;
1186 controller = bridge_get_controller(br);
1187 if ((br->controller != NULL) != (controller != NULL)) {
1188 ofproto_flush_flows(br->ofproto);
1190 free(br->controller);
1191 br->controller = controller ? xstrdup(controller) : NULL;
1194 const char *fail_mode;
1195 int max_backoff, probe;
1196 int rate_limit, burst_limit;
1198 if (!strcmp(controller, "discover")) {
1199 bool update_resolv_conf = true;
1201 if (cfg_has("%s.update-resolv.conf", pfx)) {
1202 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1205 ofproto_set_discovery(br->ofproto, true,
1206 cfg_get_string(0, "%s.accept-regex", pfx),
1207 update_resolv_conf);
1209 struct netdev *netdev;
1213 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1215 || cfg_get_bool(0, "%s.in-band", pfx));
1216 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1217 ofproto_set_in_band(br->ofproto, in_band);
1219 error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev);
1221 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1222 struct in_addr ip, mask, gateway;
1223 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1224 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1225 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1227 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1229 mask.s_addr = guess_netmask(ip.s_addr);
1231 if (!netdev_set_in4(netdev, ip, mask)) {
1232 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1234 br->name, IP_ARGS(&ip.s_addr),
1235 IP_ARGS(&mask.s_addr));
1238 if (gateway.s_addr) {
1239 if (!netdev_add_router(gateway)) {
1240 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1241 br->name, IP_ARGS(&gateway.s_addr));
1245 netdev_close(netdev);
1249 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1251 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1253 ofproto_set_failure(br->ofproto,
1255 || !strcmp(fail_mode, "standalone")
1256 || !strcmp(fail_mode, "open")));
1258 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1260 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1265 ofproto_set_probe_interval(br->ofproto, probe);
1267 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1269 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1274 ofproto_set_max_backoff(br->ofproto, max_backoff);
1276 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1278 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1280 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1282 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1284 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1286 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1288 if (cfg_has("%s.commands.acl", pfx)) {
1289 struct svec command_acls;
1292 svec_init(&command_acls);
1293 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1294 command_acl = svec_join(&command_acls, ",", "");
1296 ofproto_set_remote_execution(br->ofproto, command_acl,
1297 cfg_get_string(0, "%s.commands.dir",
1300 svec_destroy(&command_acls);
1303 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1306 union ofp_action action;
1309 /* Set up a flow that matches every packet and directs them to
1310 * OFPP_NORMAL (which goes to us). */
1311 memset(&action, 0, sizeof action);
1312 action.type = htons(OFPAT_OUTPUT);
1313 action.output.len = htons(sizeof action);
1314 action.output.port = htons(OFPP_NORMAL);
1315 memset(&flow, 0, sizeof flow);
1316 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1319 ofproto_set_in_band(br->ofproto, false);
1320 ofproto_set_max_backoff(br->ofproto, 1);
1321 ofproto_set_probe_interval(br->ofproto, 5);
1322 ofproto_set_failure(br->ofproto, false);
1323 ofproto_set_stp(br->ofproto, false);
1327 ofproto_set_controller(br->ofproto, br->controller);
1331 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1336 for (i = 0; i < br->n_ports; i++) {
1337 struct port *port = br->ports[i];
1338 for (j = 0; j < port->n_ifaces; j++) {
1339 struct iface *iface = port->ifaces[j];
1340 svec_add(ifaces, iface->name);
1342 if (port->n_ifaces > 1
1343 && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
1344 svec_add(ifaces, port->name);
1347 svec_sort_unique(ifaces);
1350 /* For robustness, in case the administrator moves around datapath ports behind
1351 * our back, we re-check all the datapath port numbers here.
1353 * This function will set the 'dp_ifidx' members of interfaces that have
1354 * disappeared to -1, so only call this function from a context where those
1355 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1356 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1357 * datapath, which doesn't support UINT16_MAX+1 ports. */
1359 bridge_fetch_dp_ifaces(struct bridge *br)
1361 struct odp_port *dpif_ports;
1362 size_t n_dpif_ports;
1365 /* Reset all interface numbers. */
1366 for (i = 0; i < br->n_ports; i++) {
1367 struct port *port = br->ports[i];
1368 for (j = 0; j < port->n_ifaces; j++) {
1369 struct iface *iface = port->ifaces[j];
1370 iface->dp_ifidx = -1;
1373 port_array_clear(&br->ifaces);
1375 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
1376 for (i = 0; i < n_dpif_ports; i++) {
1377 struct odp_port *p = &dpif_ports[i];
1378 struct iface *iface = iface_lookup(br, p->devname);
1380 if (iface->dp_ifidx >= 0) {
1381 VLOG_WARN("dp%u reported interface %s twice",
1382 dpif_id(&br->dpif), p->devname);
1383 } else if (iface_from_dp_ifidx(br, p->port)) {
1384 VLOG_WARN("dp%u reported interface %"PRIu16" twice",
1385 dpif_id(&br->dpif), p->port);
1387 port_array_set(&br->ifaces, p->port, iface);
1388 iface->dp_ifidx = p->port;
1395 /* Bridge packet processing functions. */
1398 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1400 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1403 static struct bond_entry *
1404 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1406 return &port->bond_hash[bond_hash(mac)];
1410 bond_choose_iface(const struct port *port)
1412 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1413 size_t i, best_down_slave = -1;
1414 long long next_delay_expiration = LLONG_MAX;
1416 for (i = 0; i < port->n_ifaces; i++) {
1417 struct iface *iface = port->ifaces[i];
1419 if (iface->enabled) {
1421 } else if (iface->delay_expires < next_delay_expiration) {
1422 best_down_slave = i;
1423 next_delay_expiration = iface->delay_expires;
1427 if (best_down_slave != -1) {
1428 struct iface *iface = port->ifaces[best_down_slave];
1430 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1431 "since no other interface is up", iface->name,
1432 iface->delay_expires - time_msec());
1433 bond_enable_slave(iface, true);
1436 return best_down_slave;
1440 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1441 uint16_t *dp_ifidx, tag_type *tags)
1443 struct iface *iface;
1445 assert(port->n_ifaces);
1446 if (port->n_ifaces == 1) {
1447 iface = port->ifaces[0];
1449 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1450 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1451 || !port->ifaces[e->iface_idx]->enabled) {
1452 /* XXX select interface properly. The current interface selection
1453 * is only good for testing the rebalancing code. */
1454 e->iface_idx = bond_choose_iface(port);
1455 if (e->iface_idx < 0) {
1456 *tags |= port->no_ifaces_tag;
1459 e->iface_tag = tag_create_random();
1460 ((struct port *) port)->bond_compat_is_stale = true;
1462 *tags |= e->iface_tag;
1463 iface = port->ifaces[e->iface_idx];
1465 *dp_ifidx = iface->dp_ifidx;
1466 *tags |= iface->tag; /* Currently only used for bonding. */
1471 bond_link_status_update(struct iface *iface, bool carrier)
1473 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1474 struct port *port = iface->port;
1476 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1477 /* Nothing to do. */
1480 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1481 iface->name, carrier ? "detected" : "dropped");
1482 if (carrier == iface->enabled) {
1483 iface->delay_expires = LLONG_MAX;
1484 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1485 iface->name, carrier ? "disabled" : "enabled");
1486 } else if (carrier && port->active_iface < 0) {
1487 bond_enable_slave(iface, true);
1488 if (port->updelay) {
1489 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1490 "other interface is up", iface->name, port->updelay);
1493 int delay = carrier ? port->updelay : port->downdelay;
1494 iface->delay_expires = time_msec() + delay;
1497 "interface %s: will be %s if it stays %s for %d ms",
1499 carrier ? "enabled" : "disabled",
1500 carrier ? "up" : "down",
1507 bond_choose_active_iface(struct port *port)
1509 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1511 port->active_iface = bond_choose_iface(port);
1512 port->active_iface_tag = tag_create_random();
1513 if (port->active_iface >= 0) {
1514 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1515 port->name, port->ifaces[port->active_iface]->name);
1517 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1523 bond_enable_slave(struct iface *iface, bool enable)
1525 struct port *port = iface->port;
1526 struct bridge *br = port->bridge;
1528 /* This acts as a recursion check. If the act of disabling a slave
1529 * causes a different slave to be enabled, the flag will allow us to
1530 * skip redundant work when we reenter this function. It must be
1531 * cleared on exit to keep things safe with multiple bonds. */
1532 static bool moving_active_iface = false;
1534 iface->delay_expires = LLONG_MAX;
1535 if (enable == iface->enabled) {
1539 iface->enabled = enable;
1540 if (!iface->enabled) {
1541 VLOG_WARN("interface %s: disabled", iface->name);
1542 ofproto_revalidate(br->ofproto, iface->tag);
1543 if (iface->port_ifidx == port->active_iface) {
1544 ofproto_revalidate(br->ofproto,
1545 port->active_iface_tag);
1547 /* Disabling a slave can lead to another slave being immediately
1548 * enabled if there will be no active slaves but one is waiting
1549 * on an updelay. In this case we do not need to run most of the
1550 * code for the newly enabled slave since there was no period
1551 * without an active slave and it is redundant with the disabling
1553 moving_active_iface = true;
1554 bond_choose_active_iface(port);
1556 bond_send_learning_packets(port);
1558 VLOG_WARN("interface %s: enabled", iface->name);
1559 if (port->active_iface < 0 && !moving_active_iface) {
1560 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1561 bond_choose_active_iface(port);
1562 bond_send_learning_packets(port);
1564 iface->tag = tag_create_random();
1567 moving_active_iface = false;
1568 port->bond_compat_is_stale = true;
1572 bond_run(struct bridge *br)
1576 for (i = 0; i < br->n_ports; i++) {
1577 struct port *port = br->ports[i];
1579 if (port->n_ifaces >= 2) {
1580 for (j = 0; j < port->n_ifaces; j++) {
1581 struct iface *iface = port->ifaces[j];
1582 if (time_msec() >= iface->delay_expires) {
1583 bond_enable_slave(iface, !iface->enabled);
1588 if (port->bond_compat_is_stale) {
1589 port->bond_compat_is_stale = false;
1590 port_update_bond_compat(port);
1596 bond_wait(struct bridge *br)
1600 for (i = 0; i < br->n_ports; i++) {
1601 struct port *port = br->ports[i];
1602 if (port->n_ifaces < 2) {
1605 for (j = 0; j < port->n_ifaces; j++) {
1606 struct iface *iface = port->ifaces[j];
1607 if (iface->delay_expires != LLONG_MAX) {
1608 poll_timer_wait(iface->delay_expires - time_msec());
1615 set_dst(struct dst *p, const flow_t *flow,
1616 const struct port *in_port, const struct port *out_port,
1621 * XXX This uses too many tags: any broadcast flow will get one tag per
1622 * destination port, and thus a broadcast on a switch of any size is likely
1623 * to have all tag bits set. We should figure out a way to be smarter.
1625 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1626 *tags |= out_port->stp_state_tag;
1627 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1631 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1632 : in_port->vlan >= 0 ? in_port->vlan
1633 : ntohs(flow->dl_vlan));
1634 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1638 swap_dst(struct dst *p, struct dst *q)
1640 struct dst tmp = *p;
1645 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1646 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1647 * that we push to the datapath. We could in fact fully sort the array by
1648 * vlan, but in most cases there are at most two different vlan tags so that's
1649 * possibly overkill.) */
1651 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1653 struct dst *first = dsts;
1654 struct dst *last = dsts + n_dsts;
1656 while (first != last) {
1658 * - All dsts < first have vlan == 'vlan'.
1659 * - All dsts >= last have vlan != 'vlan'.
1660 * - first < last. */
1661 while (first->vlan == vlan) {
1662 if (++first == last) {
1667 /* Same invariants, plus one additional:
1668 * - first->vlan != vlan.
1670 while (last[-1].vlan != vlan) {
1671 if (--last == first) {
1676 /* Same invariants, plus one additional:
1677 * - last[-1].vlan == vlan.*/
1678 swap_dst(first++, --last);
1683 mirror_mask_ffs(mirror_mask_t mask)
1685 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1690 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1691 const struct dst *test)
1694 for (i = 0; i < n_dsts; i++) {
1695 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1703 port_trunks_vlan(const struct port *port, uint16_t vlan)
1705 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1709 port_includes_vlan(const struct port *port, uint16_t vlan)
1711 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1715 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1716 const struct port *in_port, const struct port *out_port,
1717 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1719 mirror_mask_t mirrors = in_port->src_mirrors;
1720 struct dst *dst = dsts;
1723 *tags |= in_port->stp_state_tag;
1724 if (out_port == FLOOD_PORT) {
1725 /* XXX use ODP_FLOOD if no vlans or bonding. */
1726 /* XXX even better, define each VLAN as a datapath port group */
1727 for (i = 0; i < br->n_ports; i++) {
1728 struct port *port = br->ports[i];
1729 if (port != in_port && port_includes_vlan(port, vlan)
1730 && !port->is_mirror_output_port
1731 && set_dst(dst, flow, in_port, port, tags)) {
1732 mirrors |= port->dst_mirrors;
1736 *nf_output_iface = NF_OUT_FLOOD;
1737 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1738 *nf_output_iface = dst->dp_ifidx;
1739 mirrors |= out_port->dst_mirrors;
1744 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1745 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1747 if (set_dst(dst, flow, in_port, m->out_port, tags)
1748 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1752 for (i = 0; i < br->n_ports; i++) {
1753 struct port *port = br->ports[i];
1754 if (port_includes_vlan(port, m->out_vlan)
1755 && set_dst(dst, flow, in_port, port, tags))
1759 if (port->vlan < 0) {
1760 dst->vlan = m->out_vlan;
1762 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1766 /* Use the vlan tag on the original flow instead of
1767 * the one passed in the vlan parameter. This ensures
1768 * that we compare the vlan from before any implicit
1769 * tagging tags place. This is necessary because
1770 * dst->vlan is the final vlan, after removing implicit
1772 flow_vlan = ntohs(flow->dl_vlan);
1773 if (flow_vlan == 0) {
1774 flow_vlan = OFP_VLAN_NONE;
1776 if (port == in_port && dst->vlan == flow_vlan) {
1777 /* Don't send out input port on same VLAN. */
1785 mirrors &= mirrors - 1;
1788 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1793 print_dsts(const struct dst *dsts, size_t n)
1795 for (; n--; dsts++) {
1796 printf(">p%"PRIu16, dsts->dp_ifidx);
1797 if (dsts->vlan != OFP_VLAN_NONE) {
1798 printf("v%"PRIu16, dsts->vlan);
1804 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1805 const struct port *in_port, const struct port *out_port,
1806 tag_type *tags, struct odp_actions *actions,
1807 uint16_t *nf_output_iface)
1809 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1811 const struct dst *p;
1814 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
1817 cur_vlan = ntohs(flow->dl_vlan);
1818 for (p = dsts; p < &dsts[n_dsts]; p++) {
1819 union odp_action *a;
1820 if (p->vlan != cur_vlan) {
1821 if (p->vlan == OFP_VLAN_NONE) {
1822 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1824 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1825 a->vlan_vid.vlan_vid = htons(p->vlan);
1829 a = odp_actions_add(actions, ODPAT_OUTPUT);
1830 a->output.port = p->dp_ifidx;
1835 is_bcast_arp_reply(const flow_t *flow)
1837 return (flow->dl_type == htons(ETH_TYPE_ARP)
1838 && flow->nw_proto == ARP_OP_REPLY
1839 && eth_addr_is_broadcast(flow->dl_dst));
1842 /* If the composed actions may be applied to any packet in the given 'flow',
1843 * returns true. Otherwise, the actions should only be applied to 'packet', or
1844 * not at all, if 'packet' was NULL. */
1846 process_flow(struct bridge *br, const flow_t *flow,
1847 const struct ofpbuf *packet, struct odp_actions *actions,
1848 tag_type *tags, uint16_t *nf_output_iface)
1850 struct iface *in_iface;
1851 struct port *in_port;
1852 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1856 /* Find the interface and port structure for the received packet. */
1857 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1859 /* No interface? Something fishy... */
1860 if (packet != NULL) {
1861 /* Odd. A few possible reasons here:
1863 * - We deleted an interface but there are still a few packets
1864 * queued up from it.
1866 * - Someone externally added an interface (e.g. with "ovs-dpctl
1867 * add-if") that we don't know about.
1869 * - Packet arrived on the local port but the local port is not
1870 * one of our bridge ports.
1872 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1874 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1875 "interface %"PRIu16, br->name, flow->in_port);
1878 /* Return without adding any actions, to drop packets on this flow. */
1881 in_port = in_iface->port;
1883 /* Figure out what VLAN this packet belongs to.
1885 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1886 * belongs to VLAN 0, so we should treat both cases identically. (In the
1887 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1888 * presumably to allow a priority to be specified. In the latter case, the
1889 * packet does not have any 802.1Q header.) */
1890 vlan = ntohs(flow->dl_vlan);
1891 if (vlan == OFP_VLAN_NONE) {
1894 if (in_port->vlan >= 0) {
1896 /* XXX support double tagging? */
1897 if (packet != NULL) {
1898 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1899 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1900 "packet received on port %s configured with "
1901 "implicit VLAN %"PRIu16,
1902 br->name, ntohs(flow->dl_vlan),
1903 in_port->name, in_port->vlan);
1907 vlan = in_port->vlan;
1909 if (!port_includes_vlan(in_port, vlan)) {
1910 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1911 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1912 "packet received on port %s not configured for "
1914 br->name, vlan, in_port->name, vlan);
1919 /* Drop frames for ports that STP wants entirely killed (both for
1920 * forwarding and for learning). Later, after we do learning, we'll drop
1921 * the frames that STP wants to do learning but not forwarding on. */
1922 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1926 /* Drop frames for reserved multicast addresses. */
1927 if (eth_addr_is_reserved(flow->dl_dst)) {
1931 /* Drop frames on ports reserved for mirroring. */
1932 if (in_port->is_mirror_output_port) {
1933 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1934 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1935 "which is reserved exclusively for mirroring",
1936 br->name, in_port->name);
1940 /* Packets received on bonds need special attention to avoid duplicates. */
1941 if (in_port->n_ifaces > 1) {
1944 if (eth_addr_is_multicast(flow->dl_dst)) {
1945 *tags |= in_port->active_iface_tag;
1946 if (in_port->active_iface != in_iface->port_ifidx) {
1947 /* Drop all multicast packets on inactive slaves. */
1952 /* Drop all packets for which we have learned a different input
1953 * port, because we probably sent the packet on one slave and got
1954 * it back on the other. Broadcast ARP replies are an exception
1955 * to this rule: the host has moved to another switch. */
1956 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1957 if (src_idx != -1 && src_idx != in_port->port_idx &&
1958 !is_bcast_arp_reply(flow)) {
1964 out_port = FLOOD_PORT;
1965 /* Learn source MAC (but don't try to learn from revalidation). */
1967 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1968 vlan, in_port->port_idx);
1970 /* The log messages here could actually be useful in debugging,
1971 * so keep the rate limit relatively high. */
1972 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1974 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1975 "on port %s in VLAN %d",
1976 br->name, ETH_ADDR_ARGS(flow->dl_src),
1977 in_port->name, vlan);
1978 ofproto_revalidate(br->ofproto, rev_tag);
1982 /* Determine output port. */
1983 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1985 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1986 out_port = br->ports[out_port_idx];
1987 } else if (!packet) {
1988 /* If we are revalidating but don't have a learning entry then
1989 * eject the flow. Installing a flow that floods packets will
1990 * prevent us from seeing future packets and learning properly. */
1994 /* Don't send packets out their input ports. Don't forward frames that STP
1995 * wants us to discard. */
1996 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
2001 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2007 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2010 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2011 const struct ofp_phy_port *opp,
2014 struct bridge *br = br_;
2015 struct iface *iface;
2018 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2024 if (reason == OFPPR_DELETE) {
2025 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2026 br->name, iface->name);
2027 iface_destroy(iface);
2028 if (!port->n_ifaces) {
2029 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2030 br->name, port->name);
2036 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
2037 if (port->n_ifaces > 1) {
2038 bool up = !(opp->state & OFPPS_LINK_DOWN);
2039 bond_link_status_update(iface, up);
2040 port_update_bond_compat(port);
2046 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2047 struct odp_actions *actions, tag_type *tags,
2048 uint16_t *nf_output_iface, void *br_)
2050 struct bridge *br = br_;
2053 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
2054 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
2055 brstp_receive(br, flow, payload);
2060 COVERAGE_INC(bridge_process_flow);
2061 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2065 bridge_account_flow_ofhook_cb(const flow_t *flow,
2066 const union odp_action *actions,
2067 size_t n_actions, unsigned long long int n_bytes,
2070 struct bridge *br = br_;
2071 const union odp_action *a;
2073 if (!br->has_bonded_ports) {
2077 for (a = actions; a < &actions[n_actions]; a++) {
2078 if (a->type == ODPAT_OUTPUT) {
2079 struct port *port = port_from_dp_ifidx(br, a->output.port);
2080 if (port && port->n_ifaces >= 2) {
2081 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2082 e->tx_bytes += n_bytes;
2089 bridge_account_checkpoint_ofhook_cb(void *br_)
2091 struct bridge *br = br_;
2094 if (!br->has_bonded_ports) {
2098 /* The current ofproto implementation calls this callback at least once a
2099 * second, so this timer implementation is sufficient. */
2100 if (time_msec() < br->bond_next_rebalance) {
2103 br->bond_next_rebalance = time_msec() + 10000;
2105 for (i = 0; i < br->n_ports; i++) {
2106 struct port *port = br->ports[i];
2107 if (port->n_ifaces > 1) {
2108 bond_rebalance_port(port);
2113 static struct ofhooks bridge_ofhooks = {
2114 bridge_port_changed_ofhook_cb,
2115 bridge_normal_ofhook_cb,
2116 bridge_account_flow_ofhook_cb,
2117 bridge_account_checkpoint_ofhook_cb,
2120 /* Bonding functions. */
2122 /* Statistics for a single interface on a bonded port, used for load-based
2123 * bond rebalancing. */
2124 struct slave_balance {
2125 struct iface *iface; /* The interface. */
2126 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2128 /* All the "bond_entry"s that are assigned to this interface, in order of
2129 * increasing tx_bytes. */
2130 struct bond_entry **hashes;
2134 /* Sorts pointers to pointers to bond_entries in ascending order by the
2135 * interface to which they are assigned, and within a single interface in
2136 * ascending order of bytes transmitted. */
2138 compare_bond_entries(const void *a_, const void *b_)
2140 const struct bond_entry *const *ap = a_;
2141 const struct bond_entry *const *bp = b_;
2142 const struct bond_entry *a = *ap;
2143 const struct bond_entry *b = *bp;
2144 if (a->iface_idx != b->iface_idx) {
2145 return a->iface_idx > b->iface_idx ? 1 : -1;
2146 } else if (a->tx_bytes != b->tx_bytes) {
2147 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2153 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2154 * *descending* order by number of bytes transmitted. */
2156 compare_slave_balance(const void *a_, const void *b_)
2158 const struct slave_balance *a = a_;
2159 const struct slave_balance *b = b_;
2160 if (a->iface->enabled != b->iface->enabled) {
2161 return a->iface->enabled ? -1 : 1;
2162 } else if (a->tx_bytes != b->tx_bytes) {
2163 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2170 swap_bals(struct slave_balance *a, struct slave_balance *b)
2172 struct slave_balance tmp = *a;
2177 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2178 * given that 'p' (and only 'p') might be in the wrong location.
2180 * This function invalidates 'p', since it might now be in a different memory
2183 resort_bals(struct slave_balance *p,
2184 struct slave_balance bals[], size_t n_bals)
2187 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2188 swap_bals(p, p - 1);
2190 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2191 swap_bals(p, p + 1);
2197 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2199 if (VLOG_IS_DBG_ENABLED()) {
2200 struct ds ds = DS_EMPTY_INITIALIZER;
2201 const struct slave_balance *b;
2203 for (b = bals; b < bals + n_bals; b++) {
2207 ds_put_char(&ds, ',');
2209 ds_put_format(&ds, " %s %"PRIu64"kB",
2210 b->iface->name, b->tx_bytes / 1024);
2212 if (!b->iface->enabled) {
2213 ds_put_cstr(&ds, " (disabled)");
2215 if (b->n_hashes > 0) {
2216 ds_put_cstr(&ds, " (");
2217 for (i = 0; i < b->n_hashes; i++) {
2218 const struct bond_entry *e = b->hashes[i];
2220 ds_put_cstr(&ds, " + ");
2222 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2223 e - port->bond_hash, e->tx_bytes / 1024);
2225 ds_put_cstr(&ds, ")");
2228 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2233 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2235 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2238 struct bond_entry *hash = from->hashes[hash_idx];
2239 struct port *port = from->iface->port;
2240 uint64_t delta = hash->tx_bytes;
2242 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2243 "from %s to %s (now carrying %"PRIu64"kB and "
2244 "%"PRIu64"kB load, respectively)",
2245 port->name, delta / 1024, hash - port->bond_hash,
2246 from->iface->name, to->iface->name,
2247 (from->tx_bytes - delta) / 1024,
2248 (to->tx_bytes + delta) / 1024);
2250 /* Delete element from from->hashes.
2252 * We don't bother to add the element to to->hashes because not only would
2253 * it require more work, the only purpose it would be to allow that hash to
2254 * be migrated to another slave in this rebalancing run, and there is no
2255 * point in doing that. */
2256 if (hash_idx == 0) {
2259 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2260 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2264 /* Shift load away from 'from' to 'to'. */
2265 from->tx_bytes -= delta;
2266 to->tx_bytes += delta;
2268 /* Arrange for flows to be revalidated. */
2269 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2270 hash->iface_idx = to->iface->port_ifidx;
2271 hash->iface_tag = tag_create_random();
2275 bond_rebalance_port(struct port *port)
2277 struct slave_balance bals[DP_MAX_PORTS];
2279 struct bond_entry *hashes[BOND_MASK + 1];
2280 struct slave_balance *b, *from, *to;
2281 struct bond_entry *e;
2284 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2285 * descending order of tx_bytes, so that bals[0] represents the most
2286 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2289 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2290 * array for each slave_balance structure, we sort our local array of
2291 * hashes in order by slave, so that all of the hashes for a given slave
2292 * become contiguous in memory, and then we point each 'hashes' members of
2293 * a slave_balance structure to the start of a contiguous group. */
2294 n_bals = port->n_ifaces;
2295 for (b = bals; b < &bals[n_bals]; b++) {
2296 b->iface = port->ifaces[b - bals];
2301 for (i = 0; i <= BOND_MASK; i++) {
2302 hashes[i] = &port->bond_hash[i];
2304 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2305 for (i = 0; i <= BOND_MASK; i++) {
2307 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2308 b = &bals[e->iface_idx];
2309 b->tx_bytes += e->tx_bytes;
2311 b->hashes = &hashes[i];
2316 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2317 log_bals(bals, n_bals, port);
2319 /* Discard slaves that aren't enabled (which were sorted to the back of the
2320 * array earlier). */
2321 while (!bals[n_bals - 1].iface->enabled) {
2328 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2329 to = &bals[n_bals - 1];
2330 for (from = bals; from < to; ) {
2331 uint64_t overload = from->tx_bytes - to->tx_bytes;
2332 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2333 /* The extra load on 'from' (and all less-loaded slaves), compared
2334 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2335 * it is less than ~1Mbps. No point in rebalancing. */
2337 } else if (from->n_hashes == 1) {
2338 /* 'from' only carries a single MAC hash, so we can't shift any
2339 * load away from it, even though we want to. */
2342 /* 'from' is carrying significantly more load than 'to', and that
2343 * load is split across at least two different hashes. Pick a hash
2344 * to migrate to 'to' (the least-loaded slave), given that doing so
2345 * must decrease the ratio of the load on the two slaves by at
2348 * The sort order we use means that we prefer to shift away the
2349 * smallest hashes instead of the biggest ones. There is little
2350 * reason behind this decision; we could use the opposite sort
2351 * order to shift away big hashes ahead of small ones. */
2355 for (i = 0; i < from->n_hashes; i++) {
2356 double old_ratio, new_ratio;
2357 uint64_t delta = from->hashes[i]->tx_bytes;
2359 if (delta == 0 || from->tx_bytes - delta == 0) {
2360 /* Pointless move. */
2364 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2366 if (to->tx_bytes == 0) {
2367 /* Nothing on the new slave, move it. */
2371 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2372 new_ratio = (double)(from->tx_bytes - delta) /
2373 (to->tx_bytes + delta);
2375 if (new_ratio == 0) {
2376 /* Should already be covered but check to prevent division
2381 if (new_ratio < 1) {
2382 new_ratio = 1 / new_ratio;
2385 if (old_ratio - new_ratio > 0.1) {
2386 /* Would decrease the ratio, move it. */
2390 if (i < from->n_hashes) {
2391 bond_shift_load(from, to, i);
2392 port->bond_compat_is_stale = true;
2394 /* If the result of the migration changed the relative order of
2395 * 'from' and 'to' swap them back to maintain invariants. */
2396 if (order_swapped) {
2397 swap_bals(from, to);
2400 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2401 * point to different slave_balance structures. It is only
2402 * valid to do these two operations in a row at all because we
2403 * know that 'from' will not move past 'to' and vice versa. */
2404 resort_bals(from, bals, n_bals);
2405 resort_bals(to, bals, n_bals);
2412 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2413 * historical data to decay to <1% in 7 rebalancing runs. */
2414 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2420 bond_send_learning_packets(struct port *port)
2422 struct bridge *br = port->bridge;
2423 struct mac_entry *e;
2424 struct ofpbuf packet;
2425 int error, n_packets, n_errors;
2427 if (!port->n_ifaces || port->active_iface < 0) {
2431 ofpbuf_init(&packet, 128);
2432 error = n_packets = n_errors = 0;
2433 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2434 union ofp_action actions[2], *a;
2440 if (e->port == port->port_idx
2441 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2445 /* Compose actions. */
2446 memset(actions, 0, sizeof actions);
2449 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2450 a->vlan_vid.len = htons(sizeof *a);
2451 a->vlan_vid.vlan_vid = htons(e->vlan);
2454 a->output.type = htons(OFPAT_OUTPUT);
2455 a->output.len = htons(sizeof *a);
2456 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2461 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2463 flow_extract(&packet, ODPP_NONE, &flow);
2464 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2471 ofpbuf_uninit(&packet);
2474 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2475 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2476 "packets, last error was: %s",
2477 port->name, n_errors, n_packets, strerror(error));
2479 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2480 port->name, n_packets);
2484 /* Bonding unixctl user interface functions. */
2487 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2489 struct ds ds = DS_EMPTY_INITIALIZER;
2490 const struct bridge *br;
2492 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2494 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2497 for (i = 0; i < br->n_ports; i++) {
2498 const struct port *port = br->ports[i];
2499 if (port->n_ifaces > 1) {
2502 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2503 for (j = 0; j < port->n_ifaces; j++) {
2504 const struct iface *iface = port->ifaces[j];
2506 ds_put_cstr(&ds, ", ");
2508 ds_put_cstr(&ds, iface->name);
2510 ds_put_char(&ds, '\n');
2514 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2518 static struct port *
2519 bond_find(const char *name)
2521 const struct bridge *br;
2523 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2526 for (i = 0; i < br->n_ports; i++) {
2527 struct port *port = br->ports[i];
2528 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2537 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2539 struct ds ds = DS_EMPTY_INITIALIZER;
2540 const struct port *port;
2543 port = bond_find(args);
2545 unixctl_command_reply(conn, 501, "no such bond");
2549 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2550 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2551 ds_put_format(&ds, "next rebalance: %lld ms\n",
2552 port->bridge->bond_next_rebalance - time_msec());
2553 for (j = 0; j < port->n_ifaces; j++) {
2554 const struct iface *iface = port->ifaces[j];
2555 struct bond_entry *be;
2558 ds_put_format(&ds, "slave %s: %s\n",
2559 iface->name, iface->enabled ? "enabled" : "disabled");
2560 if (j == port->active_iface) {
2561 ds_put_cstr(&ds, "\tactive slave\n");
2563 if (iface->delay_expires != LLONG_MAX) {
2564 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2565 iface->enabled ? "downdelay" : "updelay",
2566 iface->delay_expires - time_msec());
2570 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2571 int hash = be - port->bond_hash;
2572 struct mac_entry *me;
2574 if (be->iface_idx != j) {
2578 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2579 hash, be->tx_bytes / 1024);
2582 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2583 &port->bridge->ml->lrus) {
2586 if (bond_hash(me->mac) == hash
2587 && me->port != port->port_idx
2588 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2589 && dp_ifidx == iface->dp_ifidx)
2591 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2592 ETH_ADDR_ARGS(me->mac));
2597 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2602 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2604 char *args = (char *) args_;
2605 char *save_ptr = NULL;
2606 char *bond_s, *hash_s, *slave_s;
2607 uint8_t mac[ETH_ADDR_LEN];
2609 struct iface *iface;
2610 struct bond_entry *entry;
2613 bond_s = strtok_r(args, " ", &save_ptr);
2614 hash_s = strtok_r(NULL, " ", &save_ptr);
2615 slave_s = strtok_r(NULL, " ", &save_ptr);
2617 unixctl_command_reply(conn, 501,
2618 "usage: bond/migrate BOND HASH SLAVE");
2622 port = bond_find(bond_s);
2624 unixctl_command_reply(conn, 501, "no such bond");
2628 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2629 == ETH_ADDR_SCAN_COUNT) {
2630 hash = bond_hash(mac);
2631 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2632 hash = atoi(hash_s) & BOND_MASK;
2634 unixctl_command_reply(conn, 501, "bad hash");
2638 iface = port_lookup_iface(port, slave_s);
2640 unixctl_command_reply(conn, 501, "no such slave");
2644 if (!iface->enabled) {
2645 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2649 entry = &port->bond_hash[hash];
2650 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2651 entry->iface_idx = iface->port_ifidx;
2652 entry->iface_tag = tag_create_random();
2653 port->bond_compat_is_stale = true;
2654 unixctl_command_reply(conn, 200, "migrated");
2658 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2660 char *args = (char *) args_;
2661 char *save_ptr = NULL;
2662 char *bond_s, *slave_s;
2664 struct iface *iface;
2666 bond_s = strtok_r(args, " ", &save_ptr);
2667 slave_s = strtok_r(NULL, " ", &save_ptr);
2669 unixctl_command_reply(conn, 501,
2670 "usage: bond/set-active-slave BOND SLAVE");
2674 port = bond_find(bond_s);
2676 unixctl_command_reply(conn, 501, "no such bond");
2680 iface = port_lookup_iface(port, slave_s);
2682 unixctl_command_reply(conn, 501, "no such slave");
2686 if (!iface->enabled) {
2687 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2691 if (port->active_iface != iface->port_ifidx) {
2692 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2693 port->active_iface = iface->port_ifidx;
2694 port->active_iface_tag = tag_create_random();
2695 VLOG_INFO("port %s: active interface is now %s",
2696 port->name, iface->name);
2697 bond_send_learning_packets(port);
2698 unixctl_command_reply(conn, 200, "done");
2700 unixctl_command_reply(conn, 200, "no change");
2705 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2707 char *args = (char *) args_;
2708 char *save_ptr = NULL;
2709 char *bond_s, *slave_s;
2711 struct iface *iface;
2713 bond_s = strtok_r(args, " ", &save_ptr);
2714 slave_s = strtok_r(NULL, " ", &save_ptr);
2716 unixctl_command_reply(conn, 501,
2717 "usage: bond/enable/disable-slave BOND SLAVE");
2721 port = bond_find(bond_s);
2723 unixctl_command_reply(conn, 501, "no such bond");
2727 iface = port_lookup_iface(port, slave_s);
2729 unixctl_command_reply(conn, 501, "no such slave");
2733 bond_enable_slave(iface, enable);
2734 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2738 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2740 enable_slave(conn, args, true);
2744 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2746 enable_slave(conn, args, false);
2752 unixctl_command_register("bond/list", bond_unixctl_list);
2753 unixctl_command_register("bond/show", bond_unixctl_show);
2754 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2755 unixctl_command_register("bond/set-active-slave",
2756 bond_unixctl_set_active_slave);
2757 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2758 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2761 /* Port functions. */
2764 port_create(struct bridge *br, const char *name)
2768 port = xcalloc(1, sizeof *port);
2770 port->port_idx = br->n_ports;
2772 port->trunks = NULL;
2773 port->name = xstrdup(name);
2774 port->active_iface = -1;
2775 port->stp_state = STP_DISABLED;
2776 port->stp_state_tag = 0;
2778 if (br->n_ports >= br->allocated_ports) {
2779 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2782 br->ports[br->n_ports++] = port;
2784 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2789 port_reconfigure(struct port *port)
2791 bool bonded = cfg_has_section("bonding.%s", port->name);
2792 struct svec old_ifaces, new_ifaces;
2793 unsigned long *trunks;
2797 /* Collect old and new interfaces. */
2798 svec_init(&old_ifaces);
2799 svec_init(&new_ifaces);
2800 for (i = 0; i < port->n_ifaces; i++) {
2801 svec_add(&old_ifaces, port->ifaces[i]->name);
2803 svec_sort(&old_ifaces);
2805 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2806 if (!new_ifaces.n) {
2807 VLOG_ERR("port %s: no interfaces specified for bonded port",
2809 } else if (new_ifaces.n == 1) {
2810 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2814 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2815 if (port->updelay < 0) {
2818 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2819 if (port->downdelay < 0) {
2820 port->downdelay = 0;
2823 svec_init(&new_ifaces);
2824 svec_add(&new_ifaces, port->name);
2827 /* Get rid of deleted interfaces and add new interfaces. */
2828 for (i = 0; i < port->n_ifaces; i++) {
2829 struct iface *iface = port->ifaces[i];
2830 if (!svec_contains(&new_ifaces, iface->name)) {
2831 iface_destroy(iface);
2836 for (i = 0; i < new_ifaces.n; i++) {
2837 const char *name = new_ifaces.names[i];
2838 if (!svec_contains(&old_ifaces, name)) {
2839 iface_create(port, name);
2845 if (cfg_has("vlan.%s.tag", port->name)) {
2847 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2848 if (vlan >= 0 && vlan <= 4095) {
2849 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2852 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2853 * they even work as-is. But they have not been tested. */
2854 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2858 if (port->vlan != vlan) {
2860 bridge_flush(port->bridge);
2863 /* Get trunked VLANs. */
2866 size_t n_trunks, n_errors;
2869 trunks = bitmap_allocate(4096);
2870 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2872 for (i = 0; i < n_trunks; i++) {
2873 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2875 bitmap_set1(trunks, trunk);
2881 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2882 port->name, n_trunks);
2884 if (n_errors == n_trunks) {
2886 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2889 bitmap_set_multiple(trunks, 0, 4096, 1);
2892 if (cfg_has("vlan.%s.trunks", port->name)) {
2893 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2894 port->name, port->name);
2898 ? port->trunks != NULL
2899 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2900 bridge_flush(port->bridge);
2902 bitmap_free(port->trunks);
2903 port->trunks = trunks;
2905 svec_destroy(&old_ifaces);
2906 svec_destroy(&new_ifaces);
2910 port_destroy(struct port *port)
2913 struct bridge *br = port->bridge;
2917 proc_net_compat_update_vlan(port->name, NULL, 0);
2918 proc_net_compat_update_bond(port->name, NULL);
2920 for (i = 0; i < MAX_MIRRORS; i++) {
2921 struct mirror *m = br->mirrors[i];
2922 if (m && m->out_port == port) {
2927 while (port->n_ifaces > 0) {
2928 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2931 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2932 del->port_idx = port->port_idx;
2935 bitmap_free(port->trunks);
2942 static struct port *
2943 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2945 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2946 return iface ? iface->port : NULL;
2949 static struct port *
2950 port_lookup(const struct bridge *br, const char *name)
2954 for (i = 0; i < br->n_ports; i++) {
2955 struct port *port = br->ports[i];
2956 if (!strcmp(port->name, name)) {
2963 static struct iface *
2964 port_lookup_iface(const struct port *port, const char *name)
2968 for (j = 0; j < port->n_ifaces; j++) {
2969 struct iface *iface = port->ifaces[j];
2970 if (!strcmp(iface->name, name)) {
2978 port_update_bonding(struct port *port)
2980 if (port->n_ifaces < 2) {
2981 /* Not a bonded port. */
2982 if (port->bond_hash) {
2983 free(port->bond_hash);
2984 port->bond_hash = NULL;
2985 port->bond_compat_is_stale = true;
2988 if (!port->bond_hash) {
2991 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2992 for (i = 0; i <= BOND_MASK; i++) {
2993 struct bond_entry *e = &port->bond_hash[i];
2997 port->no_ifaces_tag = tag_create_random();
2998 bond_choose_active_iface(port);
3000 port->bond_compat_is_stale = true;
3005 port_update_bond_compat(struct port *port)
3007 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3008 struct compat_bond bond;
3011 if (port->n_ifaces < 2) {
3012 proc_net_compat_update_bond(port->name, NULL);
3017 bond.updelay = port->updelay;
3018 bond.downdelay = port->downdelay;
3021 bond.hashes = compat_hashes;
3022 if (port->bond_hash) {
3023 const struct bond_entry *e;
3024 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3025 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3026 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3027 cbh->hash = e - port->bond_hash;
3028 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3033 bond.n_slaves = port->n_ifaces;
3034 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3035 for (i = 0; i < port->n_ifaces; i++) {
3036 struct iface *iface = port->ifaces[i];
3037 struct compat_bond_slave *slave = &bond.slaves[i];
3038 slave->name = iface->name;
3040 /* We need to make the same determination as the Linux bonding
3041 * code to determine whether a slave should be consider "up".
3042 * The Linux function bond_miimon_inspect() supports four
3043 * BOND_LINK_* states:
3045 * - BOND_LINK_UP: carrier detected, updelay has passed.
3046 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3047 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3048 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3050 * The function bond_info_show_slave() only considers BOND_LINK_UP
3051 * to be "up" and anything else to be "down".
3053 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3057 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
3060 if (cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
3061 struct netdev *bond_netdev;
3063 if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) {
3065 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3067 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3069 netdev_close(bond_netdev);
3073 proc_net_compat_update_bond(port->name, &bond);
3078 port_update_vlan_compat(struct port *port)
3080 struct bridge *br = port->bridge;
3081 char *vlandev_name = NULL;
3083 if (port->vlan > 0) {
3084 /* Figure out the name that the VLAN device should actually have, if it
3085 * existed. This takes some work because the VLAN device would not
3086 * have port->name in its name; rather, it would have the trunk port's
3087 * name, and 'port' would be attached to a bridge that also had the
3088 * VLAN device one of its ports. So we need to find a trunk port that
3089 * includes port->vlan.
3091 * There might be more than one candidate. This doesn't happen on
3092 * XenServer, so if it happens we just pick the first choice in
3093 * alphabetical order instead of creating multiple VLAN devices. */
3095 for (i = 0; i < br->n_ports; i++) {
3096 struct port *p = br->ports[i];
3097 if (port_trunks_vlan(p, port->vlan)
3099 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3101 const uint8_t *ea = p->ifaces[0]->mac;
3102 if (!eth_addr_is_multicast(ea) &&
3103 !eth_addr_is_reserved(ea) &&
3104 !eth_addr_is_zero(ea)) {
3105 vlandev_name = p->name;
3110 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3113 /* Interface functions. */
3116 iface_create(struct port *port, const char *name)
3118 struct iface *iface;
3120 iface = xcalloc(1, sizeof *iface);
3122 iface->port_ifidx = port->n_ifaces;
3123 iface->name = xstrdup(name);
3124 iface->dp_ifidx = -1;
3125 iface->tag = tag_create_random();
3126 iface->delay_expires = LLONG_MAX;
3128 if (!cfg_get_bool(0, "iface.%s.internal", iface->name)) {
3129 netdev_nodev_get_etheraddr(name, iface->mac);
3130 netdev_nodev_get_carrier(name, &iface->enabled);
3132 /* Internal interfaces are created later by the call to dpif_port_add()
3133 * in bridge_reconfigure(). Until then, we can't obtain any
3134 * information about them. (There's no real value in doing so, anyway,
3135 * because the 'mac' and 'enabled' values are only used for interfaces
3136 * that are bond slaves, and it doesn't normally make sense to bond an
3137 * internal interface.) */
3140 if (port->n_ifaces >= port->allocated_ifaces) {
3141 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3142 sizeof *port->ifaces);
3144 port->ifaces[port->n_ifaces++] = iface;
3145 if (port->n_ifaces > 1) {
3146 port->bridge->has_bonded_ports = true;
3149 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3151 port_update_bonding(port);
3152 bridge_flush(port->bridge);
3156 iface_destroy(struct iface *iface)
3159 struct port *port = iface->port;
3160 struct bridge *br = port->bridge;
3161 bool del_active = port->active_iface == iface->port_ifidx;
3164 if (iface->dp_ifidx >= 0) {
3165 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3168 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3169 del->port_ifidx = iface->port_ifidx;
3175 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3176 bond_choose_active_iface(port);
3177 bond_send_learning_packets(port);
3180 port_update_bonding(port);
3181 bridge_flush(port->bridge);
3185 static struct iface *
3186 iface_lookup(const struct bridge *br, const char *name)
3190 for (i = 0; i < br->n_ports; i++) {
3191 struct port *port = br->ports[i];
3192 for (j = 0; j < port->n_ifaces; j++) {
3193 struct iface *iface = port->ifaces[j];
3194 if (!strcmp(iface->name, name)) {
3202 static struct iface *
3203 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3205 return port_array_get(&br->ifaces, dp_ifidx);
3208 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3209 * 'br', that is, an interface that is entirely simulated within the datapath.
3210 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3211 * interfaces are created by setting "iface.<iface>.internal = true".
3213 * In addition, we have a kluge-y feature that creates an internal port with
3214 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3215 * This feature needs to go away in the long term. Until then, this is one
3216 * reason why this function takes a name instead of a struct iface: the fake
3217 * interfaces created this way do not have a struct iface. */
3219 iface_is_internal(const struct bridge *br, const char *iface)
3221 if (!strcmp(iface, br->name)
3222 || cfg_get_bool(0, "iface.%s.internal", iface)) {
3226 if (cfg_get_bool(0, "bonding.%s.fake-iface", iface)) {
3227 struct port *port = port_lookup(br, iface);
3228 if (port && port->n_ifaces > 1) {
3236 /* Set Ethernet address of 'iface', if one is specified in the configuration
3239 iface_set_mac(struct iface *iface)
3241 uint64_t mac = cfg_get_mac(0, "iface.%s.mac", iface->name);
3243 static uint8_t ea[ETH_ADDR_LEN];
3245 eth_addr_from_uint64(mac, ea);
3246 if (eth_addr_is_multicast(ea)) {
3247 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3249 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3250 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3251 iface->name, iface->name);
3253 int error = netdev_nodev_set_etheraddr(iface->name, ea);
3255 VLOG_ERR("interface %s: setting MAC failed (%s)",
3256 iface->name, strerror(error));
3262 /* Port mirroring. */
3265 mirror_reconfigure(struct bridge *br)
3267 struct svec old_mirrors, new_mirrors;
3270 /* Collect old and new mirrors. */
3271 svec_init(&old_mirrors);
3272 svec_init(&new_mirrors);
3273 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3274 for (i = 0; i < MAX_MIRRORS; i++) {
3275 if (br->mirrors[i]) {
3276 svec_add(&old_mirrors, br->mirrors[i]->name);
3280 /* Get rid of deleted mirrors and add new mirrors. */
3281 svec_sort(&old_mirrors);
3282 assert(svec_is_unique(&old_mirrors));
3283 svec_sort(&new_mirrors);
3284 assert(svec_is_unique(&new_mirrors));
3285 for (i = 0; i < MAX_MIRRORS; i++) {
3286 struct mirror *m = br->mirrors[i];
3287 if (m && !svec_contains(&new_mirrors, m->name)) {
3291 for (i = 0; i < new_mirrors.n; i++) {
3292 const char *name = new_mirrors.names[i];
3293 if (!svec_contains(&old_mirrors, name)) {
3294 mirror_create(br, name);
3297 svec_destroy(&old_mirrors);
3298 svec_destroy(&new_mirrors);
3300 /* Reconfigure all mirrors. */
3301 for (i = 0; i < MAX_MIRRORS; i++) {
3302 if (br->mirrors[i]) {
3303 mirror_reconfigure_one(br->mirrors[i]);
3307 /* Update port reserved status. */
3308 for (i = 0; i < br->n_ports; i++) {
3309 br->ports[i]->is_mirror_output_port = false;
3311 for (i = 0; i < MAX_MIRRORS; i++) {
3312 struct mirror *m = br->mirrors[i];
3313 if (m && m->out_port) {
3314 m->out_port->is_mirror_output_port = true;
3320 mirror_create(struct bridge *br, const char *name)
3325 for (i = 0; ; i++) {
3326 if (i >= MAX_MIRRORS) {
3327 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3328 "cannot create %s", br->name, MAX_MIRRORS, name);
3331 if (!br->mirrors[i]) {
3336 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3339 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3342 m->name = xstrdup(name);
3343 svec_init(&m->src_ports);
3344 svec_init(&m->dst_ports);
3352 mirror_destroy(struct mirror *m)
3355 struct bridge *br = m->bridge;
3358 for (i = 0; i < br->n_ports; i++) {
3359 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3360 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3363 svec_destroy(&m->src_ports);
3364 svec_destroy(&m->dst_ports);
3367 m->bridge->mirrors[m->idx] = NULL;
3375 prune_ports(struct mirror *m, struct svec *ports)
3380 svec_sort_unique(ports);
3383 for (i = 0; i < ports->n; i++) {
3384 const char *name = ports->names[i];
3385 if (port_lookup(m->bridge, name)) {
3386 svec_add(&tmp, name);
3388 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3389 m->bridge->name, m->name, name);
3392 svec_swap(ports, &tmp);
3397 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3401 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3402 * order won't give us numeric sort order. But that's good enough for what
3403 * we need right now. */
3404 svec_sort_unique(vlan_strings);
3406 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3408 for (i = 0; i < vlan_strings->n; i++) {
3409 const char *name = vlan_strings->names[i];
3411 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3412 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3413 m->bridge->name, m->name, name);
3415 (*vlans)[n_vlans++] = vlan;
3422 vlan_is_mirrored(const struct mirror *m, int vlan)
3426 for (i = 0; i < m->n_vlans; i++) {
3427 if (m->vlans[i] == vlan) {
3435 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3439 for (i = 0; i < m->n_vlans; i++) {
3440 if (port_trunks_vlan(p, m->vlans[i])) {
3448 mirror_reconfigure_one(struct mirror *m)
3450 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3451 struct svec src_ports, dst_ports, ports;
3452 struct svec vlan_strings;
3453 mirror_mask_t mirror_bit;
3454 const char *out_port_name;
3455 struct port *out_port;
3460 bool mirror_all_ports;
3461 bool any_ports_specified;
3463 /* Get output port. */
3464 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3465 m->bridge->name, m->name);
3466 if (out_port_name) {
3467 out_port = port_lookup(m->bridge, out_port_name);
3469 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3470 "named %s", pfx, m->bridge->name, out_port_name);
3477 if (cfg_has("%s.output.vlan", pfx)) {
3478 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3479 "ignoring %s.output.vlan", pfx, pfx, pfx);
3481 } else if (cfg_has("%s.output.vlan", pfx)) {
3483 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3485 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3486 "but exactly one is required; disabling port mirror %s",
3487 pfx, pfx, pfx, pfx);
3493 /* Get all the ports, and drop duplicates and ports that don't exist. */
3494 svec_init(&src_ports);
3495 svec_init(&dst_ports);
3497 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3498 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3499 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3500 any_ports_specified = src_ports.n || dst_ports.n || ports.n;
3501 svec_append(&src_ports, &ports);
3502 svec_append(&dst_ports, &ports);
3503 svec_destroy(&ports);
3504 prune_ports(m, &src_ports);
3505 prune_ports(m, &dst_ports);
3506 if (any_ports_specified && !src_ports.n && !dst_ports.n) {
3507 VLOG_ERR("%s: none of the specified ports exist; "
3508 "disabling port mirror %s", pfx, pfx);
3513 /* Get all the vlans, and drop duplicate and invalid vlans. */
3514 svec_init(&vlan_strings);
3515 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3516 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3517 svec_destroy(&vlan_strings);
3519 /* Update mirror data. */
3520 if (!svec_equal(&m->src_ports, &src_ports)
3521 || !svec_equal(&m->dst_ports, &dst_ports)
3522 || m->n_vlans != n_vlans
3523 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3524 || m->out_port != out_port
3525 || m->out_vlan != out_vlan) {
3526 bridge_flush(m->bridge);
3528 svec_swap(&m->src_ports, &src_ports);
3529 svec_swap(&m->dst_ports, &dst_ports);
3532 m->n_vlans = n_vlans;
3533 m->out_port = out_port;
3534 m->out_vlan = out_vlan;
3536 /* If no selection criteria have been given, mirror for all ports. */
3537 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3540 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3541 for (i = 0; i < m->bridge->n_ports; i++) {
3542 struct port *port = m->bridge->ports[i];
3544 if (mirror_all_ports
3545 || svec_contains(&m->src_ports, port->name)
3548 ? port_trunks_any_mirrored_vlan(m, port)
3549 : vlan_is_mirrored(m, port->vlan)))) {
3550 port->src_mirrors |= mirror_bit;
3552 port->src_mirrors &= ~mirror_bit;
3555 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3556 port->dst_mirrors |= mirror_bit;
3558 port->dst_mirrors &= ~mirror_bit;
3564 svec_destroy(&src_ports);
3565 svec_destroy(&dst_ports);
3569 /* Spanning tree protocol. */
3571 static void brstp_update_port_state(struct port *);
3574 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3576 struct bridge *br = br_;
3577 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3578 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3580 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3582 } else if (eth_addr_is_zero(iface->mac)) {
3583 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3586 union ofp_action action;
3587 struct eth_header *eth = pkt->l2;
3590 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3592 memset(&action, 0, sizeof action);
3593 action.type = htons(OFPAT_OUTPUT);
3594 action.output.len = htons(sizeof action);
3595 action.output.port = htons(port_no);
3597 flow_extract(pkt, ODPP_NONE, &flow);
3598 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3604 brstp_reconfigure(struct bridge *br)
3608 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3610 stp_destroy(br->stp);
3616 uint64_t bridge_address, bridge_id;
3617 int bridge_priority;
3619 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3620 if (!bridge_address) {
3622 bridge_address = (stp_get_bridge_id(br->stp)
3623 & ((UINT64_C(1) << 48) - 1));
3625 uint8_t mac[ETH_ADDR_LEN];
3626 eth_addr_random(mac);
3627 bridge_address = eth_addr_to_uint64(mac);
3631 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3633 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3635 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3638 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3640 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3641 br->stp_last_tick = time_msec();
3644 if (bridge_id != stp_get_bridge_id(br->stp)) {
3645 stp_set_bridge_id(br->stp, bridge_id);
3650 for (i = 0; i < br->n_ports; i++) {
3651 struct port *p = br->ports[i];
3653 struct stp_port *sp;
3654 int path_cost, priority;
3660 dp_ifidx = p->ifaces[0]->dp_ifidx;
3661 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3665 sp = stp_get_port(br->stp, dp_ifidx);
3666 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3667 "stp.%s.port.%s.enabled",
3669 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3670 br->name, p->name));
3671 if (p->is_mirror_output_port) {
3674 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3675 bridge_flush(br); /* Might not be necessary. */
3677 stp_port_enable(sp);
3679 stp_port_disable(sp);
3683 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3685 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3687 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3688 "stp.%s.port.%s.priority",
3690 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3692 : STP_DEFAULT_PORT_PRIORITY);
3693 stp_port_set_priority(sp, priority);
3696 brstp_adjust_timers(br);
3698 for (i = 0; i < br->n_ports; i++) {
3699 brstp_update_port_state(br->ports[i]);
3704 brstp_update_port_state(struct port *p)
3706 struct bridge *br = p->bridge;
3707 enum stp_state state;
3709 /* Figure out new state. */
3710 state = STP_DISABLED;
3711 if (br->stp && p->n_ifaces > 0) {
3712 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3713 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3714 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3719 if (p->stp_state != state) {
3720 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3721 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3722 p->name, stp_state_name(p->stp_state),
3723 stp_state_name(state));
3724 if (p->stp_state == STP_DISABLED) {
3727 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3729 p->stp_state = state;
3730 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3731 : tag_create_random());
3736 brstp_adjust_timers(struct bridge *br)
3738 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3739 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3740 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3742 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3743 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3744 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3748 brstp_run(struct bridge *br)
3751 long long int now = time_msec();
3752 long long int elapsed = now - br->stp_last_tick;
3753 struct stp_port *sp;
3756 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3757 br->stp_last_tick = now;
3759 while (stp_get_changed_port(br->stp, &sp)) {
3760 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3762 brstp_update_port_state(p);
3769 brstp_wait(struct bridge *br)
3772 poll_timer_wait(1000);