1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
47 #include "poll-loop.h"
48 #include "port-array.h"
49 #include "proc-net-compat.h"
51 #include "secchan/ofproto.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id;
74 struct port *port; /* Containing port. */
75 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 int dp_ifidx; /* Index within kernel datapath. */
80 uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */
82 tag_type tag; /* Tag associated with this interface. */
83 bool enabled; /* May be chosen for flows? */
84 long long delay_expires; /* Time after which 'enabled' may change. */
87 #define BOND_MASK 0xff
89 int iface_idx; /* Index of assigned iface, or -1 if none. */
90 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
91 tag_type iface_tag; /* Tag associated with iface_idx. */
94 #define MAX_MIRRORS 32
95 typedef uint32_t mirror_mask_t;
96 #define MIRROR_MASK_C(X) UINT32_C(X)
97 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
99 struct bridge *bridge;
103 /* Selection criteria. */
104 struct svec src_ports;
105 struct svec dst_ports;
110 struct port *out_port;
114 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
116 struct bridge *bridge;
118 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
119 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
122 /* An ordinary bridge port has 1 interface.
123 * A bridge port for bonding has at least 2 interfaces. */
124 struct iface **ifaces;
125 size_t n_ifaces, allocated_ifaces;
128 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
129 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
130 tag_type active_iface_tag; /* Tag for bcast flows. */
131 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
132 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
133 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
135 /* Port mirroring info. */
136 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
137 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
138 bool is_mirror_output_port; /* Does port mirroring send frames here? */
140 /* Spanning tree info. */
141 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
142 tag_type stp_state_tag; /* Tag for STP state change. */
145 #define DP_MAX_PORTS 255
147 struct list node; /* Node in global list of bridges. */
148 char *name; /* User-specified arbitrary name. */
149 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
150 bool sent_config_request; /* Successfully sent config request? */
151 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
153 /* Support for remote controllers. */
154 char *controller; /* NULL if there is no remote controller;
155 * "discover" to do controller discovery;
156 * otherwise a vconn name. */
158 /* OpenFlow switch processing. */
159 struct ofproto *ofproto; /* OpenFlow switch. */
161 /* Kernel datapath information. */
162 struct dpif dpif; /* Kernel datapath. */
163 struct port_array ifaces; /* Indexed by kernel datapath port number. */
167 size_t n_ports, allocated_ports;
170 bool has_bonded_ports;
171 long long int bond_next_rebalance;
176 /* Flow statistics gathering. */
177 time_t next_stats_request;
179 /* Port mirroring. */
180 struct mirror *mirrors[MAX_MIRRORS];
184 long long int stp_last_tick;
187 /* List of all bridges. */
188 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
190 /* Maximum number of datapaths. */
191 enum { DP_MAX = 256 };
193 static struct bridge *bridge_create(const char *name);
194 static void bridge_destroy(struct bridge *);
195 static struct bridge *bridge_lookup(const char *name);
196 static void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *);
197 static int bridge_run_one(struct bridge *);
198 static void bridge_reconfigure_one(struct bridge *);
199 static void bridge_reconfigure_controller(struct bridge *);
200 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
201 static void bridge_fetch_dp_ifaces(struct bridge *);
202 static void bridge_flush(struct bridge *);
203 static void bridge_pick_local_hw_addr(struct bridge *,
204 uint8_t ea[ETH_ADDR_LEN],
205 const char **devname);
206 static uint64_t bridge_pick_datapath_id(struct bridge *,
207 const uint8_t bridge_ea[ETH_ADDR_LEN],
208 const char *devname);
209 static uint64_t dpid_from_hash(const void *, size_t nbytes);
211 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
213 static void bond_init(void);
214 static void bond_run(struct bridge *);
215 static void bond_wait(struct bridge *);
216 static void bond_rebalance_port(struct port *);
217 static void bond_send_learning_packets(struct port *);
219 static void port_create(struct bridge *, const char *name);
220 static void port_reconfigure(struct port *);
221 static void port_destroy(struct port *);
222 static struct port *port_lookup(const struct bridge *, const char *name);
223 static struct iface *port_lookup_iface(const struct port *, const char *name);
224 static struct port *port_from_dp_ifidx(const struct bridge *,
226 static void port_update_bond_compat(struct port *);
227 static void port_update_vlan_compat(struct port *);
229 static void mirror_create(struct bridge *, const char *name);
230 static void mirror_destroy(struct mirror *);
231 static void mirror_reconfigure(struct bridge *);
232 static void mirror_reconfigure_one(struct mirror *);
233 static bool vlan_is_mirrored(const struct mirror *, int vlan);
235 static void brstp_reconfigure(struct bridge *);
236 static void brstp_adjust_timers(struct bridge *);
237 static void brstp_run(struct bridge *);
238 static void brstp_wait(struct bridge *);
240 static void iface_create(struct port *, const char *name);
241 static void iface_destroy(struct iface *);
242 static struct iface *iface_lookup(const struct bridge *, const char *name);
243 static struct iface *iface_from_dp_ifidx(const struct bridge *,
246 /* Hooks into ofproto processing. */
247 static struct ofhooks bridge_ofhooks;
249 /* Public functions. */
251 /* Adds the name of each interface used by a bridge, including local and
252 * internal ports, to 'svec'. */
254 bridge_get_ifaces(struct svec *svec)
256 struct bridge *br, *next;
259 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
260 for (i = 0; i < br->n_ports; i++) {
261 struct port *port = br->ports[i];
263 for (j = 0; j < port->n_ifaces; j++) {
264 struct iface *iface = port->ifaces[j];
265 if (iface->dp_ifidx < 0) {
266 VLOG_ERR("%s interface not in dp%u, ignoring",
267 iface->name, dpif_id(&br->dpif));
269 if (iface->dp_ifidx != ODPP_LOCAL) {
270 svec_add(svec, iface->name);
278 /* The caller must already have called cfg_read(). */
287 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
289 for (i = 0; i < DP_MAX; i++) {
293 sprintf(devname, "dp%d", i);
294 retval = dpif_open(devname, &dpif);
296 char dpif_name[IF_NAMESIZE];
297 if (dpif_get_name(&dpif, dpif_name, sizeof dpif_name)
298 || !cfg_has("bridge.%s.port", dpif_name)) {
302 } else if (retval != ENODEV) {
303 VLOG_ERR("failed to delete datapath dp%d: %s",
304 i, strerror(retval));
308 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows);
310 bridge_reconfigure();
315 config_string_change(const char *key, char **valuep)
317 const char *value = cfg_get_string(0, "%s", key);
318 if (value && (!*valuep || strcmp(value, *valuep))) {
320 *valuep = xstrdup(value);
328 bridge_configure_ssl(void)
330 /* XXX SSL should be configurable on a per-bridge basis.
331 * XXX should be possible to de-configure SSL. */
332 static char *private_key_file;
333 static char *certificate_file;
334 static char *cacert_file;
337 if (config_string_change("ssl.private-key", &private_key_file)) {
338 vconn_ssl_set_private_key_file(private_key_file);
341 if (config_string_change("ssl.certificate", &certificate_file)) {
342 vconn_ssl_set_certificate_file(certificate_file);
345 /* We assume that even if the filename hasn't changed, if the CA cert
346 * file has been removed, that we want to move back into
347 * boot-strapping mode. This opens a small security hole, because
348 * the old certificate will still be trusted until vSwitch is
349 * restarted. We may want to address this in vconn's SSL library. */
350 if (config_string_change("ssl.ca-cert", &cacert_file)
351 || (stat(cacert_file, &s) && errno == ENOENT)) {
352 vconn_ssl_set_ca_cert_file(cacert_file,
353 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
359 bridge_reconfigure(void)
361 struct svec old_br, new_br, raw_new_br;
362 struct bridge *br, *next;
365 COVERAGE_INC(bridge_reconfigure);
367 /* Collect old bridges. */
369 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
370 svec_add(&old_br, br->name);
373 /* Collect new bridges. */
374 svec_init(&raw_new_br);
375 cfg_get_subsections(&raw_new_br, "bridge");
377 for (i = 0; i < raw_new_br.n; i++) {
378 const char *name = raw_new_br.names[i];
379 if ((!strncmp(name, "dp", 2) && isdigit(name[2])) ||
380 (!strncmp(name, "nl:", 3) && isdigit(name[3]))) {
381 VLOG_ERR("%s is not a valid bridge name (bridges may not be "
382 "named \"dp\" or \"nl:\" followed by a digit)", name);
384 svec_add(&new_br, name);
387 svec_destroy(&raw_new_br);
389 /* Get rid of deleted bridges and add new bridges. */
392 assert(svec_is_unique(&old_br));
393 assert(svec_is_unique(&new_br));
394 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
395 if (!svec_contains(&new_br, br->name)) {
399 for (i = 0; i < new_br.n; i++) {
400 const char *name = new_br.names[i];
401 if (!svec_contains(&old_br, name)) {
405 svec_destroy(&old_br);
406 svec_destroy(&new_br);
410 bridge_configure_ssl();
413 /* Reconfigure all bridges. */
414 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
415 bridge_reconfigure_one(br);
418 /* Add and delete ports on all datapaths.
420 * The kernel will reject any attempt to add a given port to a datapath if
421 * that port already belongs to a different datapath, so we must do all
422 * port deletions before any port additions. */
423 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
424 struct odp_port *dpif_ports;
426 struct svec want_ifaces;
428 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
429 bridge_get_all_ifaces(br, &want_ifaces);
430 for (i = 0; i < n_dpif_ports; i++) {
431 const struct odp_port *p = &dpif_ports[i];
432 if (!svec_contains(&want_ifaces, p->devname)
433 && strcmp(p->devname, br->name)) {
434 int retval = dpif_port_del(&br->dpif, p->port);
436 VLOG_ERR("failed to remove %s interface from dp%u: %s",
437 p->devname, dpif_id(&br->dpif), strerror(retval));
441 svec_destroy(&want_ifaces);
444 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
445 struct odp_port *dpif_ports;
447 struct svec cur_ifaces, want_ifaces, add_ifaces;
450 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
451 svec_init(&cur_ifaces);
452 for (i = 0; i < n_dpif_ports; i++) {
453 svec_add(&cur_ifaces, dpif_ports[i].devname);
456 svec_sort_unique(&cur_ifaces);
457 bridge_get_all_ifaces(br, &want_ifaces);
458 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
461 for (i = 0; i < add_ifaces.n; i++) {
462 const char *if_name = add_ifaces.names[i];
464 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
465 int error = dpif_port_add(&br->dpif, if_name, next_port_no++,
466 internal ? ODP_PORT_INTERNAL : 0);
467 if (error != EEXIST) {
468 if (next_port_no >= 256) {
469 VLOG_ERR("ran out of valid port numbers on dp%u",
474 VLOG_ERR("failed to add %s interface to dp%u: %s",
475 if_name, dpif_id(&br->dpif), strerror(error));
482 svec_destroy(&cur_ifaces);
483 svec_destroy(&want_ifaces);
484 svec_destroy(&add_ifaces);
486 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
489 struct iface *local_iface = NULL;
491 uint8_t engine_type = br->dpif.minor;
492 uint8_t engine_id = br->dpif.minor;
493 bool add_id_to_iface = false;
494 struct svec nf_hosts;
496 bridge_fetch_dp_ifaces(br);
497 for (i = 0; i < br->n_ports; ) {
498 struct port *port = br->ports[i];
500 for (j = 0; j < port->n_ifaces; ) {
501 struct iface *iface = port->ifaces[j];
502 if (iface->dp_ifidx < 0) {
503 VLOG_ERR("%s interface not in dp%u, dropping",
504 iface->name, dpif_id(&br->dpif));
505 iface_destroy(iface);
507 if (iface->dp_ifidx == ODPP_LOCAL) {
510 VLOG_DBG("dp%u has interface %s on port %d",
511 dpif_id(&br->dpif), iface->name, iface->dp_ifidx);
515 if (!port->n_ifaces) {
516 VLOG_ERR("%s port has no interfaces, dropping", port->name);
523 /* Pick local port hardware address, datapath ID. */
524 bridge_pick_local_hw_addr(br, ea, &devname);
526 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
528 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
529 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
530 "Ethernet address: %s",
531 br->name, strerror(error));
535 dpid = bridge_pick_datapath_id(br, ea, devname);
536 ofproto_set_datapath_id(br->ofproto, dpid);
538 /* Set NetFlow configuration on this bridge. */
539 if (cfg_has("netflow.%s.engine-type", br->name)) {
540 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
543 if (cfg_has("netflow.%s.engine-id", br->name)) {
544 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
546 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
547 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
550 if (add_id_to_iface && engine_id > 0x7f) {
551 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
552 "another vswitch, choose an engine id less than 128",
555 if (add_id_to_iface && br->n_ports > 0x1ff) {
556 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
557 "another port when 512 or more ports are used",
560 svec_init(&nf_hosts);
561 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
562 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
563 engine_id, add_id_to_iface)) {
564 VLOG_ERR("bridge %s: problem setting netflow collectors",
568 /* Update the controller and related settings. It would be more
569 * straightforward to call this from bridge_reconfigure_one(), but we
570 * can't do it there for two reasons. First, and most importantly, at
571 * that point we don't know the dp_ifidx of any interfaces that have
572 * been added to the bridge (because we haven't actually added them to
573 * the datapath). Second, at that point we haven't set the datapath ID
574 * yet; when a controller is configured, resetting the datapath ID will
575 * immediately disconnect from the controller, so it's better to set
576 * the datapath ID before the controller. */
577 bridge_reconfigure_controller(br);
579 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
580 for (i = 0; i < br->n_ports; i++) {
581 struct port *port = br->ports[i];
582 port_update_vlan_compat(port);
585 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
586 brstp_reconfigure(br);
591 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
592 const char **devname)
594 uint64_t requested_ea;
600 /* Did the user request a particular MAC? */
601 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
603 eth_addr_from_uint64(requested_ea, ea);
604 if (eth_addr_is_multicast(ea)) {
605 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
606 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
607 } else if (eth_addr_is_zero(ea)) {
608 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
614 /* Otherwise choose the minimum MAC address among all of the interfaces.
615 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
616 * MAC of the physical interface in such an environment.) */
617 memset(ea, 0xff, sizeof ea);
618 for (i = 0; i < br->n_ports; i++) {
619 struct port *port = br->ports[i];
620 uint8_t iface_ea[ETH_ADDR_LEN];
621 uint64_t iface_ea_u64;
624 /* Mirror output ports don't participate. */
625 if (port->is_mirror_output_port) {
629 /* Choose the MAC address to represent the port. */
630 iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name);
632 /* User specified explicitly. */
633 eth_addr_from_uint64(iface_ea_u64, iface_ea);
635 /* Find the interface with this Ethernet address (if any) so that
636 * we can provide the correct devname to the caller. */
638 for (j = 0; j < port->n_ifaces; j++) {
639 struct iface *candidate = port->ifaces[j];
640 uint8_t candidate_ea[ETH_ADDR_LEN];
641 if (!netdev_nodev_get_etheraddr(candidate->name, candidate_ea)
642 && eth_addr_equals(iface_ea, candidate_ea)) {
647 /* Choose the interface whose MAC address will represent the port.
648 * The Linux kernel bonding code always chooses the MAC address of
649 * the first slave added to a bond, and the Fedora networking
650 * scripts always add slaves to a bond in alphabetical order, so
651 * for compatibility we choose the interface with the name that is
652 * first in alphabetical order. */
653 iface = port->ifaces[0];
654 for (j = 1; j < port->n_ifaces; j++) {
655 struct iface *candidate = port->ifaces[j];
656 if (strcmp(candidate->name, iface->name) < 0) {
661 /* The local port doesn't count (since we're trying to choose its
662 * MAC address anyway). Other internal ports don't count because
663 * we really want a physical MAC if we can get it, and internal
664 * ports typically have randomly generated MACs. */
665 if (iface->dp_ifidx == ODPP_LOCAL
666 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
671 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
673 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
674 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
675 iface->name, strerror(error));
680 /* Compare against our current choice. */
681 if (!eth_addr_is_multicast(iface_ea) &&
682 !eth_addr_is_reserved(iface_ea) &&
683 !eth_addr_is_zero(iface_ea) &&
684 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
686 memcpy(ea, iface_ea, ETH_ADDR_LEN);
687 *devname = iface ? iface->name : NULL;
690 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
691 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
693 VLOG_WARN("bridge %s: using default bridge Ethernet "
694 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
696 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
697 br->name, ETH_ADDR_ARGS(ea));
701 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
702 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
703 * a network device, then that network device's name must be passed in as
704 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
705 * passed in as a null pointer. */
707 bridge_pick_datapath_id(struct bridge *br,
708 const uint8_t bridge_ea[ETH_ADDR_LEN],
712 * The procedure for choosing a bridge MAC address will, in the most
713 * ordinary case, also choose a unique MAC that we can use as a datapath
714 * ID. In some special cases, though, multiple bridges will end up with
715 * the same MAC address. This is OK for the bridges, but it will confuse
716 * the OpenFlow controller, because each datapath needs a unique datapath
719 * Datapath IDs must be unique. It is also very desirable that they be
720 * stable from one run to the next, so that policy set on a datapath
725 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
732 if (!netdev_get_vlan_vid(devname, &vlan)) {
734 * A bridge whose MAC address is taken from a VLAN network device
735 * (that is, a network device created with vconfig(8) or similar
736 * tool) will have the same MAC address as a bridge on the VLAN
737 * device's physical network device.
739 * Handle this case by hashing the physical network device MAC
740 * along with the VLAN identifier.
742 uint8_t buf[ETH_ADDR_LEN + 2];
743 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
744 buf[ETH_ADDR_LEN] = vlan >> 8;
745 buf[ETH_ADDR_LEN + 1] = vlan;
746 return dpid_from_hash(buf, sizeof buf);
749 * Assume that this bridge's MAC address is unique, since it
750 * doesn't fit any of the cases we handle specially.
755 * A purely internal bridge, that is, one that has no non-virtual
756 * network devices on it at all, is more difficult because it has no
757 * natural unique identifier at all.
759 * When the host is a XenServer, we handle this case by hashing the
760 * host's UUID with the name of the bridge. Names of bridges are
761 * persistent across XenServer reboots, although they can be reused if
762 * an internal network is destroyed and then a new one is later
763 * created, so this is fairly effective.
765 * When the host is not a XenServer, we punt by using a random MAC
766 * address on each run.
768 const char *host_uuid = xenserver_get_host_uuid();
770 char *combined = xasprintf("%s,%s", host_uuid, br->name);
771 dpid = dpid_from_hash(combined, strlen(combined));
777 return eth_addr_to_uint64(bridge_ea);
781 dpid_from_hash(const void *data, size_t n)
783 uint8_t hash[SHA1_DIGEST_SIZE];
785 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
786 sha1_bytes(data, n, hash);
787 eth_addr_mark_random(hash);
788 return eth_addr_to_uint64(hash);
794 struct bridge *br, *next;
798 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
799 int error = bridge_run_one(br);
801 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
802 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
803 "forcing reconfiguration", br->name);
817 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
818 ofproto_wait(br->ofproto);
819 if (br->controller) {
824 mac_learning_wait(br->ml);
831 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
832 * configuration changes. */
834 bridge_flush(struct bridge *br)
836 COVERAGE_INC(bridge_flush);
839 mac_learning_flush(br->ml);
843 /* Bridge unixctl user interface functions. */
845 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
847 struct ds ds = DS_EMPTY_INITIALIZER;
848 const struct bridge *br;
850 br = bridge_lookup(args);
852 unixctl_command_reply(conn, 501, "no such bridge");
856 ds_put_cstr(&ds, " port VLAN MAC Age\n");
858 const struct mac_entry *e;
859 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
860 if (e->port < 0 || e->port >= br->n_ports) {
863 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
864 br->ports[e->port]->ifaces[0]->dp_ifidx,
865 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
868 unixctl_command_reply(conn, 200, ds_cstr(&ds));
872 /* Bridge reconfiguration functions. */
874 static struct bridge *
875 bridge_create(const char *name)
880 assert(!bridge_lookup(name));
881 br = xcalloc(1, sizeof *br);
883 error = dpif_create(name, &br->dpif);
884 if (error == EEXIST) {
885 error = dpif_open(name, &br->dpif);
887 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
888 name, strerror(error));
892 dpif_flow_flush(&br->dpif);
894 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
899 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
901 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
902 dpif_delete(&br->dpif);
903 dpif_close(&br->dpif);
908 br->name = xstrdup(name);
909 br->ml = mac_learning_create();
910 br->sent_config_request = false;
911 eth_addr_random(br->default_ea);
913 port_array_init(&br->ifaces);
916 br->bond_next_rebalance = time_msec() + 10000;
918 list_push_back(&all_bridges, &br->node);
920 VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif));
926 bridge_destroy(struct bridge *br)
931 while (br->n_ports > 0) {
932 port_destroy(br->ports[br->n_ports - 1]);
934 list_remove(&br->node);
935 error = dpif_delete(&br->dpif);
936 if (error && error != ENOENT) {
937 VLOG_ERR("failed to delete dp%u: %s",
938 dpif_id(&br->dpif), strerror(error));
940 dpif_close(&br->dpif);
941 ofproto_destroy(br->ofproto);
942 free(br->controller);
943 mac_learning_destroy(br->ml);
944 port_array_destroy(&br->ifaces);
951 static struct bridge *
952 bridge_lookup(const char *name)
956 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
957 if (!strcmp(br->name, name)) {
965 bridge_exists(const char *name)
967 return bridge_lookup(name) ? true : false;
971 bridge_get_datapathid(const char *name)
973 struct bridge *br = bridge_lookup(name);
974 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
977 /* Handle requests for a listing of all flows known by the OpenFlow
978 * stack, including those normally hidden. */
980 bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args)
985 br = bridge_lookup(args);
987 unixctl_command_reply(conn, 501, "Unknown bridge");
992 ofproto_get_all_flows(br->ofproto, &results);
994 unixctl_command_reply(conn, 200, ds_cstr(&results));
995 ds_destroy(&results);
999 bridge_run_one(struct bridge *br)
1003 error = ofproto_run1(br->ofproto);
1009 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1014 error = ofproto_run2(br->ofproto, br->flush);
1021 bridge_get_controller(const struct bridge *br)
1023 const char *controller;
1025 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1027 controller = cfg_get_string(0, "mgmt.controller");
1029 return controller && controller[0] ? controller : NULL;
1033 bridge_reconfigure_one(struct bridge *br)
1035 struct svec old_ports, new_ports, ifaces;
1036 struct svec listeners, old_listeners;
1037 struct svec snoops, old_snoops;
1040 /* Collect old ports. */
1041 svec_init(&old_ports);
1042 for (i = 0; i < br->n_ports; i++) {
1043 svec_add(&old_ports, br->ports[i]->name);
1045 svec_sort(&old_ports);
1046 assert(svec_is_unique(&old_ports));
1048 /* Collect new ports. */
1049 svec_init(&new_ports);
1050 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1051 svec_sort(&new_ports);
1052 if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) {
1053 svec_add(&new_ports, br->name);
1054 svec_sort(&new_ports);
1056 if (!svec_is_unique(&new_ports)) {
1057 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1058 br->name, svec_get_duplicate(&new_ports));
1059 svec_unique(&new_ports);
1062 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1064 /* Get rid of deleted ports and add new ports. */
1065 for (i = 0; i < br->n_ports; ) {
1066 struct port *port = br->ports[i];
1067 if (!svec_contains(&new_ports, port->name)) {
1073 for (i = 0; i < new_ports.n; i++) {
1074 const char *name = new_ports.names[i];
1075 if (!svec_contains(&old_ports, name)) {
1076 port_create(br, name);
1079 svec_destroy(&old_ports);
1080 svec_destroy(&new_ports);
1082 /* Reconfigure all ports. */
1083 for (i = 0; i < br->n_ports; i++) {
1084 port_reconfigure(br->ports[i]);
1087 /* Check and delete duplicate interfaces. */
1089 for (i = 0; i < br->n_ports; ) {
1090 struct port *port = br->ports[i];
1091 for (j = 0; j < port->n_ifaces; ) {
1092 struct iface *iface = port->ifaces[j];
1093 if (svec_contains(&ifaces, iface->name)) {
1094 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1096 br->name, iface->name, port->name);
1097 iface_destroy(iface);
1099 svec_add(&ifaces, iface->name);
1104 if (!port->n_ifaces) {
1105 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1111 svec_destroy(&ifaces);
1113 /* Delete all flows if we're switching from connected to standalone or vice
1114 * versa. (XXX Should we delete all flows if we are switching from one
1115 * controller to another?) */
1117 /* Configure OpenFlow management listeners. */
1118 svec_init(&listeners);
1119 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1121 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1122 ovs_rundir, br->name));
1123 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1124 svec_clear(&listeners);
1126 svec_sort_unique(&listeners);
1128 svec_init(&old_listeners);
1129 ofproto_get_listeners(br->ofproto, &old_listeners);
1130 svec_sort_unique(&old_listeners);
1132 if (!svec_equal(&listeners, &old_listeners)) {
1133 ofproto_set_listeners(br->ofproto, &listeners);
1135 svec_destroy(&listeners);
1136 svec_destroy(&old_listeners);
1138 /* Configure OpenFlow controller connection snooping. */
1140 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1142 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1143 ovs_rundir, br->name));
1144 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1145 svec_clear(&snoops);
1147 svec_sort_unique(&snoops);
1149 svec_init(&old_snoops);
1150 ofproto_get_snoops(br->ofproto, &old_snoops);
1151 svec_sort_unique(&old_snoops);
1153 if (!svec_equal(&snoops, &old_snoops)) {
1154 ofproto_set_snoops(br->ofproto, &snoops);
1156 svec_destroy(&snoops);
1157 svec_destroy(&old_snoops);
1159 mirror_reconfigure(br);
1163 bridge_reconfigure_controller(struct bridge *br)
1165 char *pfx = xasprintf("bridge.%s.controller", br->name);
1166 const char *controller;
1168 controller = bridge_get_controller(br);
1169 if ((br->controller != NULL) != (controller != NULL)) {
1170 ofproto_flush_flows(br->ofproto);
1172 free(br->controller);
1173 br->controller = controller ? xstrdup(controller) : NULL;
1176 const char *fail_mode;
1177 int max_backoff, probe;
1178 int rate_limit, burst_limit;
1180 if (!strcmp(controller, "discover")) {
1181 bool update_resolv_conf = true;
1183 if (cfg_has("%s.update-resolv.conf", pfx)) {
1184 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1187 ofproto_set_discovery(br->ofproto, true,
1188 cfg_get_string(0, "%s.accept-regex", pfx),
1189 update_resolv_conf);
1191 struct netdev *netdev;
1195 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1197 || cfg_get_bool(0, "%s.in-band", pfx));
1198 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1199 ofproto_set_in_band(br->ofproto, in_band);
1201 error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev);
1203 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1204 struct in_addr ip, mask, gateway;
1205 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1206 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1207 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1209 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1211 mask.s_addr = guess_netmask(ip.s_addr);
1213 if (!netdev_set_in4(netdev, ip, mask)) {
1214 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1216 br->name, IP_ARGS(&ip.s_addr),
1217 IP_ARGS(&mask.s_addr));
1220 if (gateway.s_addr) {
1221 if (!netdev_add_router(gateway)) {
1222 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1223 br->name, IP_ARGS(&gateway.s_addr));
1227 netdev_close(netdev);
1231 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1233 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1235 ofproto_set_failure(br->ofproto,
1237 || !strcmp(fail_mode, "standalone")
1238 || !strcmp(fail_mode, "open")));
1240 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1242 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1247 ofproto_set_probe_interval(br->ofproto, probe);
1249 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1251 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1256 ofproto_set_max_backoff(br->ofproto, max_backoff);
1258 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1260 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1262 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1264 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1266 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1268 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1270 if (cfg_has("%s.commands.acl", pfx)) {
1271 struct svec command_acls;
1274 svec_init(&command_acls);
1275 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1276 command_acl = svec_join(&command_acls, ",", "");
1278 ofproto_set_remote_execution(br->ofproto, command_acl,
1279 cfg_get_string(0, "%s.commands.dir",
1282 svec_destroy(&command_acls);
1285 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1288 union ofp_action action;
1291 /* Set up a flow that matches every packet and directs them to
1292 * OFPP_NORMAL (which goes to us). */
1293 memset(&action, 0, sizeof action);
1294 action.type = htons(OFPAT_OUTPUT);
1295 action.output.len = htons(sizeof action);
1296 action.output.port = htons(OFPP_NORMAL);
1297 memset(&flow, 0, sizeof flow);
1298 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1301 ofproto_set_in_band(br->ofproto, false);
1302 ofproto_set_max_backoff(br->ofproto, 1);
1303 ofproto_set_probe_interval(br->ofproto, 5);
1304 ofproto_set_failure(br->ofproto, false);
1305 ofproto_set_stp(br->ofproto, false);
1309 ofproto_set_controller(br->ofproto, br->controller);
1313 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1318 for (i = 0; i < br->n_ports; i++) {
1319 struct port *port = br->ports[i];
1320 for (j = 0; j < port->n_ifaces; j++) {
1321 struct iface *iface = port->ifaces[j];
1322 svec_add(ifaces, iface->name);
1326 assert(svec_is_unique(ifaces));
1329 /* For robustness, in case the administrator moves around datapath ports behind
1330 * our back, we re-check all the datapath port numbers here.
1332 * This function will set the 'dp_ifidx' members of interfaces that have
1333 * disappeared to -1, so only call this function from a context where those
1334 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1335 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1336 * datapath, which doesn't support UINT16_MAX+1 ports. */
1338 bridge_fetch_dp_ifaces(struct bridge *br)
1340 struct odp_port *dpif_ports;
1341 size_t n_dpif_ports;
1344 /* Reset all interface numbers. */
1345 for (i = 0; i < br->n_ports; i++) {
1346 struct port *port = br->ports[i];
1347 for (j = 0; j < port->n_ifaces; j++) {
1348 struct iface *iface = port->ifaces[j];
1349 iface->dp_ifidx = -1;
1352 port_array_clear(&br->ifaces);
1354 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
1355 for (i = 0; i < n_dpif_ports; i++) {
1356 struct odp_port *p = &dpif_ports[i];
1357 struct iface *iface = iface_lookup(br, p->devname);
1359 if (iface->dp_ifidx >= 0) {
1360 VLOG_WARN("dp%u reported interface %s twice",
1361 dpif_id(&br->dpif), p->devname);
1362 } else if (iface_from_dp_ifidx(br, p->port)) {
1363 VLOG_WARN("dp%u reported interface %"PRIu16" twice",
1364 dpif_id(&br->dpif), p->port);
1366 port_array_set(&br->ifaces, p->port, iface);
1367 iface->dp_ifidx = p->port;
1374 /* Bridge packet processing functions. */
1377 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1379 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1382 static struct bond_entry *
1383 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1385 return &port->bond_hash[bond_hash(mac)];
1389 bond_choose_iface(const struct port *port)
1392 for (i = 0; i < port->n_ifaces; i++) {
1393 if (port->ifaces[i]->enabled) {
1401 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1402 uint16_t *dp_ifidx, tag_type *tags)
1404 struct iface *iface;
1406 assert(port->n_ifaces);
1407 if (port->n_ifaces == 1) {
1408 iface = port->ifaces[0];
1410 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1411 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1412 || !port->ifaces[e->iface_idx]->enabled) {
1413 /* XXX select interface properly. The current interface selection
1414 * is only good for testing the rebalancing code. */
1415 e->iface_idx = bond_choose_iface(port);
1416 if (e->iface_idx < 0) {
1417 *tags |= port->no_ifaces_tag;
1420 e->iface_tag = tag_create_random();
1421 ((struct port *) port)->bond_compat_is_stale = true;
1423 *tags |= e->iface_tag;
1424 iface = port->ifaces[e->iface_idx];
1426 *dp_ifidx = iface->dp_ifidx;
1427 *tags |= iface->tag; /* Currently only used for bonding. */
1432 bond_link_status_update(struct iface *iface, bool carrier)
1434 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1435 struct port *port = iface->port;
1437 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1438 /* Nothing to do. */
1441 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1442 iface->name, carrier ? "detected" : "dropped");
1443 if (carrier == iface->enabled) {
1444 iface->delay_expires = LLONG_MAX;
1445 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1446 iface->name, carrier ? "disabled" : "enabled");
1447 } else if (carrier && port->updelay && port->active_iface < 0) {
1448 iface->delay_expires = time_msec();
1449 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1450 "other interface is up", iface->name, port->updelay);
1452 int delay = carrier ? port->updelay : port->downdelay;
1453 iface->delay_expires = time_msec() + delay;
1456 "interface %s: will be %s if it stays %s for %d ms",
1458 carrier ? "enabled" : "disabled",
1459 carrier ? "up" : "down",
1466 bond_choose_active_iface(struct port *port)
1468 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1470 port->active_iface = bond_choose_iface(port);
1471 port->active_iface_tag = tag_create_random();
1472 if (port->active_iface >= 0) {
1473 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1474 port->name, port->ifaces[port->active_iface]->name);
1476 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1482 bond_enable_slave(struct iface *iface, bool enable)
1484 struct port *port = iface->port;
1485 struct bridge *br = port->bridge;
1487 iface->delay_expires = LLONG_MAX;
1488 if (enable == iface->enabled) {
1492 iface->enabled = enable;
1493 if (!iface->enabled) {
1494 VLOG_WARN("interface %s: disabled", iface->name);
1495 ofproto_revalidate(br->ofproto, iface->tag);
1496 if (iface->port_ifidx == port->active_iface) {
1497 ofproto_revalidate(br->ofproto,
1498 port->active_iface_tag);
1499 bond_choose_active_iface(port);
1501 bond_send_learning_packets(port);
1503 VLOG_WARN("interface %s: enabled", iface->name);
1504 if (port->active_iface < 0) {
1505 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1506 bond_choose_active_iface(port);
1507 bond_send_learning_packets(port);
1509 iface->tag = tag_create_random();
1514 bond_run(struct bridge *br)
1518 for (i = 0; i < br->n_ports; i++) {
1519 struct port *port = br->ports[i];
1521 if (port->bond_compat_is_stale) {
1522 port->bond_compat_is_stale = false;
1523 port_update_bond_compat(port);
1526 if (port->n_ifaces < 2) {
1529 for (j = 0; j < port->n_ifaces; j++) {
1530 struct iface *iface = port->ifaces[j];
1531 if (time_msec() >= iface->delay_expires) {
1532 bond_enable_slave(iface, !iface->enabled);
1539 bond_wait(struct bridge *br)
1543 for (i = 0; i < br->n_ports; i++) {
1544 struct port *port = br->ports[i];
1545 if (port->n_ifaces < 2) {
1548 for (j = 0; j < port->n_ifaces; j++) {
1549 struct iface *iface = port->ifaces[j];
1550 if (iface->delay_expires != LLONG_MAX) {
1551 poll_timer_wait(iface->delay_expires - time_msec());
1558 set_dst(struct dst *p, const flow_t *flow,
1559 const struct port *in_port, const struct port *out_port,
1564 * XXX This uses too many tags: any broadcast flow will get one tag per
1565 * destination port, and thus a broadcast on a switch of any size is likely
1566 * to have all tag bits set. We should figure out a way to be smarter.
1568 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1569 *tags |= out_port->stp_state_tag;
1570 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1574 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1575 : in_port->vlan >= 0 ? in_port->vlan
1576 : ntohs(flow->dl_vlan));
1577 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1581 swap_dst(struct dst *p, struct dst *q)
1583 struct dst tmp = *p;
1588 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1589 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1590 * that we push to the datapath. We could in fact fully sort the array by
1591 * vlan, but in most cases there are at most two different vlan tags so that's
1592 * possibly overkill.) */
1594 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1596 struct dst *first = dsts;
1597 struct dst *last = dsts + n_dsts;
1599 while (first != last) {
1601 * - All dsts < first have vlan == 'vlan'.
1602 * - All dsts >= last have vlan != 'vlan'.
1603 * - first < last. */
1604 while (first->vlan == vlan) {
1605 if (++first == last) {
1610 /* Same invariants, plus one additional:
1611 * - first->vlan != vlan.
1613 while (last[-1].vlan != vlan) {
1614 if (--last == first) {
1619 /* Same invariants, plus one additional:
1620 * - last[-1].vlan == vlan.*/
1621 swap_dst(first++, --last);
1626 mirror_mask_ffs(mirror_mask_t mask)
1628 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1633 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1634 const struct dst *test)
1637 for (i = 0; i < n_dsts; i++) {
1638 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1646 port_trunks_vlan(const struct port *port, uint16_t vlan)
1648 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1652 port_includes_vlan(const struct port *port, uint16_t vlan)
1654 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1658 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1659 const struct port *in_port, const struct port *out_port,
1660 struct dst dsts[], tag_type *tags)
1662 mirror_mask_t mirrors = in_port->src_mirrors;
1663 struct dst *dst = dsts;
1666 *tags |= in_port->stp_state_tag;
1667 if (out_port == FLOOD_PORT) {
1668 /* XXX use ODP_FLOOD if no vlans or bonding. */
1669 /* XXX even better, define each VLAN as a datapath port group */
1670 for (i = 0; i < br->n_ports; i++) {
1671 struct port *port = br->ports[i];
1672 if (port != in_port && port_includes_vlan(port, vlan)
1673 && !port->is_mirror_output_port
1674 && set_dst(dst, flow, in_port, port, tags)) {
1675 mirrors |= port->dst_mirrors;
1679 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1680 mirrors |= out_port->dst_mirrors;
1685 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1686 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1688 if (set_dst(dst, flow, in_port, m->out_port, tags)
1689 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1693 for (i = 0; i < br->n_ports; i++) {
1694 struct port *port = br->ports[i];
1695 if (port_includes_vlan(port, m->out_vlan)
1696 && set_dst(dst, flow, in_port, port, tags)
1697 && !dst_is_duplicate(dsts, dst - dsts, dst))
1699 if (port->vlan < 0) {
1700 dst->vlan = m->out_vlan;
1702 if (dst->dp_ifidx == flow->in_port
1703 && dst->vlan == vlan) {
1704 /* Don't send out input port on same VLAN. */
1712 mirrors &= mirrors - 1;
1715 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1720 print_dsts(const struct dst *dsts, size_t n)
1722 for (; n--; dsts++) {
1723 printf(">p%"PRIu16, dsts->dp_ifidx);
1724 if (dsts->vlan != OFP_VLAN_NONE) {
1725 printf("v%"PRIu16, dsts->vlan);
1731 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1732 const struct port *in_port, const struct port *out_port,
1733 tag_type *tags, struct odp_actions *actions)
1735 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1737 const struct dst *p;
1740 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1742 cur_vlan = ntohs(flow->dl_vlan);
1743 for (p = dsts; p < &dsts[n_dsts]; p++) {
1744 union odp_action *a;
1745 if (p->vlan != cur_vlan) {
1746 if (p->vlan == OFP_VLAN_NONE) {
1747 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1749 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1750 a->vlan_vid.vlan_vid = htons(p->vlan);
1754 a = odp_actions_add(actions, ODPAT_OUTPUT);
1755 a->output.port = p->dp_ifidx;
1760 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1762 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1763 return (flow->dl_type == htons(ETH_TYPE_ARP)
1764 && eth_addr_is_broadcast(flow->dl_dst)
1765 && packet->size >= sizeof(struct arp_eth_header)
1766 && arp->ar_op == ARP_OP_REQUEST);
1769 /* If the composed actions may be applied to any packet in the given 'flow',
1770 * returns true. Otherwise, the actions should only be applied to 'packet', or
1771 * not at all, if 'packet' was NULL. */
1773 process_flow(struct bridge *br, const flow_t *flow,
1774 const struct ofpbuf *packet, struct odp_actions *actions,
1777 struct iface *in_iface;
1778 struct port *in_port;
1779 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1782 /* Find the interface and port structure for the received packet. */
1783 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1785 /* No interface? Something fishy... */
1786 if (packet != NULL) {
1787 /* Odd. A few possible reasons here:
1789 * - We deleted an interface but there are still a few packets
1790 * queued up from it.
1792 * - Someone externally added an interface (e.g. with "ovs-dpctl
1793 * add-if") that we don't know about.
1795 * - Packet arrived on the local port but the local port is not
1796 * one of our bridge ports.
1798 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1800 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1801 "interface %"PRIu16, br->name, flow->in_port);
1804 /* Return without adding any actions, to drop packets on this flow. */
1807 in_port = in_iface->port;
1809 /* Figure out what VLAN this packet belongs to.
1811 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1812 * belongs to VLAN 0, so we should treat both cases identically. (In the
1813 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1814 * presumably to allow a priority to be specified. In the latter case, the
1815 * packet does not have any 802.1Q header.) */
1816 vlan = ntohs(flow->dl_vlan);
1817 if (vlan == OFP_VLAN_NONE) {
1820 if (in_port->vlan >= 0) {
1822 /* XXX support double tagging? */
1823 if (packet != NULL) {
1824 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1825 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1826 "packet received on port %s configured with "
1827 "implicit VLAN %"PRIu16,
1828 br->name, ntohs(flow->dl_vlan),
1829 in_port->name, in_port->vlan);
1833 vlan = in_port->vlan;
1835 if (!port_includes_vlan(in_port, vlan)) {
1836 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1837 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1838 "packet received on port %s not configured for "
1840 br->name, vlan, in_port->name, vlan);
1845 /* Drop frames for ports that STP wants entirely killed (both for
1846 * forwarding and for learning). Later, after we do learning, we'll drop
1847 * the frames that STP wants to do learning but not forwarding on. */
1848 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1852 /* Drop frames for reserved multicast addresses. */
1853 if (eth_addr_is_reserved(flow->dl_dst)) {
1857 /* Drop frames on ports reserved for mirroring. */
1858 if (in_port->is_mirror_output_port) {
1859 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1860 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1861 "which is reserved exclusively for mirroring",
1862 br->name, in_port->name);
1866 /* Multicast (and broadcast) packets on bonds need special attention, to
1867 * avoid receiving duplicates. */
1868 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1869 *tags |= in_port->active_iface_tag;
1870 if (in_port->active_iface != in_iface->port_ifidx) {
1871 /* Drop all multicast packets on inactive slaves. */
1874 /* Drop all multicast packets for which we have learned a different
1875 * input port, because we probably sent the packet on one slaves
1876 * and got it back on the active slave. Broadcast ARP replies are
1877 * an exception to this rule: the host has moved to another
1879 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1880 if (src_idx != -1 && src_idx != in_port->port_idx) {
1882 if (!is_bcast_arp_reply(flow, packet)) {
1886 /* No way to know whether it's an ARP reply, because the
1887 * flow entry doesn't include enough information and we
1888 * don't have a packet. Punt. */
1896 out_port = FLOOD_PORT;
1900 /* Learn source MAC (but don't try to learn from revalidation). */
1902 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1903 vlan, in_port->port_idx);
1905 /* The log messages here could actually be useful in debugging,
1906 * so keep the rate limit relatively high. */
1907 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1909 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1910 "on port %s in VLAN %d",
1911 br->name, ETH_ADDR_ARGS(flow->dl_src),
1912 in_port->name, vlan);
1913 ofproto_revalidate(br->ofproto, rev_tag);
1917 /* Determine output port. */
1918 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1920 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1921 out_port = br->ports[out_port_idx];
1925 /* Don't send packets out their input ports. Don't forward frames that STP
1926 * wants us to discard. */
1927 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1932 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1935 * We send out only a single packet, instead of setting up a flow, if the
1936 * packet is an ARP directed to broadcast that arrived on a bonded
1937 * interface. In such a situation ARP requests and replies must be handled
1938 * differently, but OpenFlow unfortunately can't distinguish them.
1940 return (in_port->n_ifaces < 2
1941 || flow->dl_type != htons(ETH_TYPE_ARP)
1942 || !eth_addr_is_broadcast(flow->dl_dst));
1945 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1948 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1949 const struct ofp_phy_port *opp,
1952 struct bridge *br = br_;
1953 struct iface *iface;
1956 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1962 if (reason == OFPPR_DELETE) {
1963 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1964 br->name, iface->name);
1965 iface_destroy(iface);
1966 if (!port->n_ifaces) {
1967 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1968 br->name, port->name);
1974 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1975 if (port->n_ifaces > 1) {
1976 bool up = !(opp->state & OFPPS_LINK_DOWN);
1977 bond_link_status_update(iface, up);
1978 port_update_bond_compat(port);
1984 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1985 struct odp_actions *actions, tag_type *tags, void *br_)
1987 struct bridge *br = br_;
1990 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1991 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1992 brstp_receive(br, flow, payload);
1997 COVERAGE_INC(bridge_process_flow);
1998 return process_flow(br, flow, packet, actions, tags);
2002 bridge_account_flow_ofhook_cb(const flow_t *flow,
2003 const union odp_action *actions,
2004 size_t n_actions, unsigned long long int n_bytes,
2007 struct bridge *br = br_;
2008 const union odp_action *a;
2010 if (!br->has_bonded_ports) {
2014 for (a = actions; a < &actions[n_actions]; a++) {
2015 if (a->type == ODPAT_OUTPUT) {
2016 struct port *port = port_from_dp_ifidx(br, a->output.port);
2017 if (port && port->n_ifaces >= 2) {
2018 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2019 e->tx_bytes += n_bytes;
2026 bridge_account_checkpoint_ofhook_cb(void *br_)
2028 struct bridge *br = br_;
2031 if (!br->has_bonded_ports) {
2035 /* The current ofproto implementation calls this callback at least once a
2036 * second, so this timer implementation is sufficient. */
2037 if (time_msec() < br->bond_next_rebalance) {
2040 br->bond_next_rebalance = time_msec() + 10000;
2042 for (i = 0; i < br->n_ports; i++) {
2043 struct port *port = br->ports[i];
2044 if (port->n_ifaces > 1) {
2045 bond_rebalance_port(port);
2050 static struct ofhooks bridge_ofhooks = {
2051 bridge_port_changed_ofhook_cb,
2052 bridge_normal_ofhook_cb,
2053 bridge_account_flow_ofhook_cb,
2054 bridge_account_checkpoint_ofhook_cb,
2057 /* Bonding functions. */
2059 /* Statistics for a single interface on a bonded port, used for load-based
2060 * bond rebalancing. */
2061 struct slave_balance {
2062 struct iface *iface; /* The interface. */
2063 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2065 /* All the "bond_entry"s that are assigned to this interface, in order of
2066 * increasing tx_bytes. */
2067 struct bond_entry **hashes;
2071 /* Sorts pointers to pointers to bond_entries in ascending order by the
2072 * interface to which they are assigned, and within a single interface in
2073 * ascending order of bytes transmitted. */
2075 compare_bond_entries(const void *a_, const void *b_)
2077 const struct bond_entry *const *ap = a_;
2078 const struct bond_entry *const *bp = b_;
2079 const struct bond_entry *a = *ap;
2080 const struct bond_entry *b = *bp;
2081 if (a->iface_idx != b->iface_idx) {
2082 return a->iface_idx > b->iface_idx ? 1 : -1;
2083 } else if (a->tx_bytes != b->tx_bytes) {
2084 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2090 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2091 * *descending* order by number of bytes transmitted. */
2093 compare_slave_balance(const void *a_, const void *b_)
2095 const struct slave_balance *a = a_;
2096 const struct slave_balance *b = b_;
2097 if (a->iface->enabled != b->iface->enabled) {
2098 return a->iface->enabled ? -1 : 1;
2099 } else if (a->tx_bytes != b->tx_bytes) {
2100 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2107 swap_bals(struct slave_balance *a, struct slave_balance *b)
2109 struct slave_balance tmp = *a;
2114 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2115 * given that 'p' (and only 'p') might be in the wrong location.
2117 * This function invalidates 'p', since it might now be in a different memory
2120 resort_bals(struct slave_balance *p,
2121 struct slave_balance bals[], size_t n_bals)
2124 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2125 swap_bals(p, p - 1);
2127 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2128 swap_bals(p, p + 1);
2134 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2136 if (VLOG_IS_DBG_ENABLED()) {
2137 struct ds ds = DS_EMPTY_INITIALIZER;
2138 const struct slave_balance *b;
2140 for (b = bals; b < bals + n_bals; b++) {
2144 ds_put_char(&ds, ',');
2146 ds_put_format(&ds, " %s %"PRIu64"kB",
2147 b->iface->name, b->tx_bytes / 1024);
2149 if (!b->iface->enabled) {
2150 ds_put_cstr(&ds, " (disabled)");
2152 if (b->n_hashes > 0) {
2153 ds_put_cstr(&ds, " (");
2154 for (i = 0; i < b->n_hashes; i++) {
2155 const struct bond_entry *e = b->hashes[i];
2157 ds_put_cstr(&ds, " + ");
2159 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2160 e - port->bond_hash, e->tx_bytes / 1024);
2162 ds_put_cstr(&ds, ")");
2165 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2170 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2172 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2173 struct bond_entry *hash)
2175 struct port *port = from->iface->port;
2176 uint64_t delta = hash->tx_bytes;
2178 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2179 "from %s to %s (now carrying %"PRIu64"kB and "
2180 "%"PRIu64"kB load, respectively)",
2181 port->name, delta / 1024, hash - port->bond_hash,
2182 from->iface->name, to->iface->name,
2183 (from->tx_bytes - delta) / 1024,
2184 (to->tx_bytes + delta) / 1024);
2186 /* Delete element from from->hashes.
2188 * We don't bother to add the element to to->hashes because not only would
2189 * it require more work, the only purpose it would be to allow that hash to
2190 * be migrated to another slave in this rebalancing run, and there is no
2191 * point in doing that. */
2192 if (from->hashes[0] == hash) {
2195 int i = hash - from->hashes[0];
2196 memmove(from->hashes + i, from->hashes + i + 1,
2197 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2201 /* Shift load away from 'from' to 'to'. */
2202 from->tx_bytes -= delta;
2203 to->tx_bytes += delta;
2205 /* Arrange for flows to be revalidated. */
2206 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2207 hash->iface_idx = to->iface->port_ifidx;
2208 hash->iface_tag = tag_create_random();
2212 bond_rebalance_port(struct port *port)
2214 struct slave_balance bals[DP_MAX_PORTS];
2216 struct bond_entry *hashes[BOND_MASK + 1];
2217 struct slave_balance *b, *from, *to;
2218 struct bond_entry *e;
2221 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2222 * descending order of tx_bytes, so that bals[0] represents the most
2223 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2226 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2227 * array for each slave_balance structure, we sort our local array of
2228 * hashes in order by slave, so that all of the hashes for a given slave
2229 * become contiguous in memory, and then we point each 'hashes' members of
2230 * a slave_balance structure to the start of a contiguous group. */
2231 n_bals = port->n_ifaces;
2232 for (b = bals; b < &bals[n_bals]; b++) {
2233 b->iface = port->ifaces[b - bals];
2238 for (i = 0; i <= BOND_MASK; i++) {
2239 hashes[i] = &port->bond_hash[i];
2241 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2242 for (i = 0; i <= BOND_MASK; i++) {
2244 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2245 b = &bals[e->iface_idx];
2246 b->tx_bytes += e->tx_bytes;
2248 b->hashes = &hashes[i];
2253 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2254 log_bals(bals, n_bals, port);
2256 /* Discard slaves that aren't enabled (which were sorted to the back of the
2257 * array earlier). */
2258 while (!bals[n_bals - 1].iface->enabled) {
2265 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2266 to = &bals[n_bals - 1];
2267 for (from = bals; from < to; ) {
2268 uint64_t overload = from->tx_bytes - to->tx_bytes;
2269 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2270 /* The extra load on 'from' (and all less-loaded slaves), compared
2271 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2272 * it is less than ~1Mbps. No point in rebalancing. */
2274 } else if (from->n_hashes == 1) {
2275 /* 'from' only carries a single MAC hash, so we can't shift any
2276 * load away from it, even though we want to. */
2279 /* 'from' is carrying significantly more load than 'to', and that
2280 * load is split across at least two different hashes. Pick a hash
2281 * to migrate to 'to' (the least-loaded slave), given that doing so
2282 * must not cause 'to''s load to exceed 'from''s load.
2284 * The sort order we use means that we prefer to shift away the
2285 * smallest hashes instead of the biggest ones. There is little
2286 * reason behind this decision; we could use the opposite sort
2287 * order to shift away big hashes ahead of small ones. */
2290 for (i = 0; i < from->n_hashes; i++) {
2291 uint64_t delta = from->hashes[i]->tx_bytes;
2292 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2296 if (i < from->n_hashes) {
2297 bond_shift_load(from, to, from->hashes[i]);
2299 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2300 * point to different slave_balance structures. It is only
2301 * valid to do these two operations in a row at all because we
2302 * know that 'from' will not move past 'to' and vice versa. */
2303 resort_bals(from, bals, n_bals);
2304 resort_bals(to, bals, n_bals);
2308 port->bond_compat_is_stale = true;
2312 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2313 * historical data to decay to <1% in 7 rebalancing runs. */
2314 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2320 bond_send_learning_packets(struct port *port)
2322 struct bridge *br = port->bridge;
2323 struct mac_entry *e;
2324 struct ofpbuf packet;
2325 int error, n_packets, n_errors;
2327 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2331 ofpbuf_init(&packet, 128);
2332 error = n_packets = n_errors = 0;
2333 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2334 static const char s[] = "Open vSwitch Bond Failover";
2335 union ofp_action actions[2], *a;
2336 struct eth_header *eth;
2337 struct llc_snap_header *llc_snap;
2343 if (e->port == port->port_idx
2344 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2348 /* Compose packet to send. */
2349 ofpbuf_clear(&packet);
2350 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2351 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2352 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2353 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2355 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2356 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2357 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2359 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2360 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2361 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2362 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2363 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2365 /* Compose actions. */
2366 memset(actions, 0, sizeof actions);
2369 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2370 a->vlan_vid.len = htons(sizeof *a);
2371 a->vlan_vid.vlan_vid = htons(e->vlan);
2374 a->output.type = htons(OFPAT_OUTPUT);
2375 a->output.len = htons(sizeof *a);
2376 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2381 flow_extract(&packet, ODPP_NONE, &flow);
2382 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2389 ofpbuf_uninit(&packet);
2392 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2393 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2394 "packets, last error was: %s",
2395 port->name, n_errors, n_packets, strerror(error));
2397 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2398 port->name, n_packets);
2402 /* Bonding unixctl user interface functions. */
2405 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2407 struct ds ds = DS_EMPTY_INITIALIZER;
2408 const struct bridge *br;
2410 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2412 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2415 for (i = 0; i < br->n_ports; i++) {
2416 const struct port *port = br->ports[i];
2417 if (port->n_ifaces > 1) {
2420 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2421 for (j = 0; j < port->n_ifaces; j++) {
2422 const struct iface *iface = port->ifaces[j];
2424 ds_put_cstr(&ds, ", ");
2426 ds_put_cstr(&ds, iface->name);
2428 ds_put_char(&ds, '\n');
2432 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2436 static struct port *
2437 bond_find(const char *name)
2439 const struct bridge *br;
2441 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2444 for (i = 0; i < br->n_ports; i++) {
2445 struct port *port = br->ports[i];
2446 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2455 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2457 struct ds ds = DS_EMPTY_INITIALIZER;
2458 const struct port *port;
2461 port = bond_find(args);
2463 unixctl_command_reply(conn, 501, "no such bond");
2467 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2468 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2469 ds_put_format(&ds, "next rebalance: %lld ms\n",
2470 port->bridge->bond_next_rebalance - time_msec());
2471 for (j = 0; j < port->n_ifaces; j++) {
2472 const struct iface *iface = port->ifaces[j];
2473 struct bond_entry *be;
2476 ds_put_format(&ds, "slave %s: %s\n",
2477 iface->name, iface->enabled ? "enabled" : "disabled");
2478 if (j == port->active_iface) {
2479 ds_put_cstr(&ds, "\tactive slave\n");
2481 if (iface->delay_expires != LLONG_MAX) {
2482 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2483 iface->enabled ? "downdelay" : "updelay",
2484 iface->delay_expires - time_msec());
2488 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2489 int hash = be - port->bond_hash;
2490 struct mac_entry *me;
2492 if (be->iface_idx != j) {
2496 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2497 hash, be->tx_bytes / 1024);
2500 if (!port->bridge->ml) {
2504 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2505 &port->bridge->ml->lrus) {
2508 if (bond_hash(me->mac) == hash
2509 && me->port != port->port_idx
2510 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2511 && dp_ifidx == iface->dp_ifidx)
2513 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2514 ETH_ADDR_ARGS(me->mac));
2519 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2524 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2526 char *args = (char *) args_;
2527 char *save_ptr = NULL;
2528 char *bond_s, *hash_s, *slave_s;
2529 uint8_t mac[ETH_ADDR_LEN];
2531 struct iface *iface;
2532 struct bond_entry *entry;
2535 bond_s = strtok_r(args, " ", &save_ptr);
2536 hash_s = strtok_r(NULL, " ", &save_ptr);
2537 slave_s = strtok_r(NULL, " ", &save_ptr);
2539 unixctl_command_reply(conn, 501,
2540 "usage: bond/migrate BOND HASH SLAVE");
2544 port = bond_find(bond_s);
2546 unixctl_command_reply(conn, 501, "no such bond");
2550 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2551 == ETH_ADDR_SCAN_COUNT) {
2552 hash = bond_hash(mac);
2553 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2554 hash = atoi(hash_s) & BOND_MASK;
2556 unixctl_command_reply(conn, 501, "bad hash");
2560 iface = port_lookup_iface(port, slave_s);
2562 unixctl_command_reply(conn, 501, "no such slave");
2566 if (!iface->enabled) {
2567 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2571 entry = &port->bond_hash[hash];
2572 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2573 entry->iface_idx = iface->port_ifidx;
2574 entry->iface_tag = tag_create_random();
2575 port->bond_compat_is_stale = true;
2576 unixctl_command_reply(conn, 200, "migrated");
2580 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2582 char *args = (char *) args_;
2583 char *save_ptr = NULL;
2584 char *bond_s, *slave_s;
2586 struct iface *iface;
2588 bond_s = strtok_r(args, " ", &save_ptr);
2589 slave_s = strtok_r(NULL, " ", &save_ptr);
2591 unixctl_command_reply(conn, 501,
2592 "usage: bond/set-active-slave BOND SLAVE");
2596 port = bond_find(bond_s);
2598 unixctl_command_reply(conn, 501, "no such bond");
2602 iface = port_lookup_iface(port, slave_s);
2604 unixctl_command_reply(conn, 501, "no such slave");
2608 if (!iface->enabled) {
2609 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2613 if (port->active_iface != iface->port_ifidx) {
2614 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2615 port->active_iface = iface->port_ifidx;
2616 port->active_iface_tag = tag_create_random();
2617 VLOG_INFO("port %s: active interface is now %s",
2618 port->name, iface->name);
2619 bond_send_learning_packets(port);
2620 unixctl_command_reply(conn, 200, "done");
2622 unixctl_command_reply(conn, 200, "no change");
2627 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2629 char *args = (char *) args_;
2630 char *save_ptr = NULL;
2631 char *bond_s, *slave_s;
2633 struct iface *iface;
2635 bond_s = strtok_r(args, " ", &save_ptr);
2636 slave_s = strtok_r(NULL, " ", &save_ptr);
2638 unixctl_command_reply(conn, 501,
2639 "usage: bond/enable/disable-slave BOND SLAVE");
2643 port = bond_find(bond_s);
2645 unixctl_command_reply(conn, 501, "no such bond");
2649 iface = port_lookup_iface(port, slave_s);
2651 unixctl_command_reply(conn, 501, "no such slave");
2655 bond_enable_slave(iface, enable);
2656 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2660 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2662 enable_slave(conn, args, true);
2666 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2668 enable_slave(conn, args, false);
2674 unixctl_command_register("bond/list", bond_unixctl_list);
2675 unixctl_command_register("bond/show", bond_unixctl_show);
2676 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2677 unixctl_command_register("bond/set-active-slave",
2678 bond_unixctl_set_active_slave);
2679 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2680 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2683 /* Port functions. */
2686 port_create(struct bridge *br, const char *name)
2690 port = xcalloc(1, sizeof *port);
2692 port->port_idx = br->n_ports;
2694 port->trunks = NULL;
2695 port->name = xstrdup(name);
2696 port->active_iface = -1;
2697 port->stp_state = STP_DISABLED;
2698 port->stp_state_tag = 0;
2700 if (br->n_ports >= br->allocated_ports) {
2701 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2704 br->ports[br->n_ports++] = port;
2706 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2711 port_reconfigure(struct port *port)
2713 bool bonded = cfg_has_section("bonding.%s", port->name);
2714 struct svec old_ifaces, new_ifaces;
2715 unsigned long *trunks;
2719 /* Collect old and new interfaces. */
2720 svec_init(&old_ifaces);
2721 svec_init(&new_ifaces);
2722 for (i = 0; i < port->n_ifaces; i++) {
2723 svec_add(&old_ifaces, port->ifaces[i]->name);
2725 svec_sort(&old_ifaces);
2727 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2728 if (!new_ifaces.n) {
2729 VLOG_ERR("port %s: no interfaces specified for bonded port",
2731 } else if (new_ifaces.n == 1) {
2732 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2736 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2737 if (port->updelay < 0) {
2740 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2741 if (port->downdelay < 0) {
2742 port->downdelay = 0;
2745 svec_init(&new_ifaces);
2746 svec_add(&new_ifaces, port->name);
2749 /* Get rid of deleted interfaces and add new interfaces. */
2750 for (i = 0; i < port->n_ifaces; i++) {
2751 struct iface *iface = port->ifaces[i];
2752 if (!svec_contains(&new_ifaces, iface->name)) {
2753 iface_destroy(iface);
2758 for (i = 0; i < new_ifaces.n; i++) {
2759 const char *name = new_ifaces.names[i];
2760 if (!svec_contains(&old_ifaces, name)) {
2761 iface_create(port, name);
2767 if (cfg_has("vlan.%s.tag", port->name)) {
2769 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2770 if (vlan >= 0 && vlan <= 4095) {
2771 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2774 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2775 * they even work as-is. But they have not been tested. */
2776 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2780 if (port->vlan != vlan) {
2782 bridge_flush(port->bridge);
2785 /* Get trunked VLANs. */
2788 size_t n_trunks, n_errors;
2791 trunks = bitmap_allocate(4096);
2792 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2794 for (i = 0; i < n_trunks; i++) {
2795 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2797 bitmap_set1(trunks, trunk);
2803 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2804 port->name, n_trunks);
2806 if (n_errors == n_trunks) {
2808 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2811 bitmap_set_multiple(trunks, 0, 4096, 1);
2814 if (cfg_has("vlan.%s.trunks", port->name)) {
2815 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2816 port->name, port->name);
2820 ? port->trunks != NULL
2821 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2822 bridge_flush(port->bridge);
2824 bitmap_free(port->trunks);
2825 port->trunks = trunks;
2827 svec_destroy(&old_ifaces);
2828 svec_destroy(&new_ifaces);
2832 port_destroy(struct port *port)
2835 struct bridge *br = port->bridge;
2839 proc_net_compat_update_vlan(port->name, NULL, 0);
2840 proc_net_compat_update_bond(port->name, NULL);
2842 for (i = 0; i < MAX_MIRRORS; i++) {
2843 struct mirror *m = br->mirrors[i];
2844 if (m && m->out_port == port) {
2849 while (port->n_ifaces > 0) {
2850 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2853 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2854 del->port_idx = port->port_idx;
2857 bitmap_free(port->trunks);
2864 static struct port *
2865 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2867 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2868 return iface ? iface->port : NULL;
2871 static struct port *
2872 port_lookup(const struct bridge *br, const char *name)
2876 for (i = 0; i < br->n_ports; i++) {
2877 struct port *port = br->ports[i];
2878 if (!strcmp(port->name, name)) {
2885 static struct iface *
2886 port_lookup_iface(const struct port *port, const char *name)
2890 for (j = 0; j < port->n_ifaces; j++) {
2891 struct iface *iface = port->ifaces[j];
2892 if (!strcmp(iface->name, name)) {
2900 port_update_bonding(struct port *port)
2902 if (port->n_ifaces < 2) {
2903 /* Not a bonded port. */
2904 if (port->bond_hash) {
2905 free(port->bond_hash);
2906 port->bond_hash = NULL;
2907 port->bond_compat_is_stale = true;
2910 if (!port->bond_hash) {
2913 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2914 for (i = 0; i <= BOND_MASK; i++) {
2915 struct bond_entry *e = &port->bond_hash[i];
2919 port->no_ifaces_tag = tag_create_random();
2920 bond_choose_active_iface(port);
2922 port->bond_compat_is_stale = true;
2927 port_update_bond_compat(struct port *port)
2929 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
2930 struct compat_bond bond;
2933 if (port->n_ifaces < 2) {
2934 proc_net_compat_update_bond(port->name, NULL);
2939 bond.updelay = port->updelay;
2940 bond.downdelay = port->downdelay;
2943 bond.hashes = compat_hashes;
2944 if (port->bond_hash) {
2945 const struct bond_entry *e;
2946 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
2947 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2948 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
2949 cbh->hash = e - port->bond_hash;
2950 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
2955 bond.n_slaves = port->n_ifaces;
2956 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2957 for (i = 0; i < port->n_ifaces; i++) {
2958 struct iface *iface = port->ifaces[i];
2959 struct compat_bond_slave *slave = &bond.slaves[i];
2960 slave->name = iface->name;
2961 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2962 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2966 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2969 proc_net_compat_update_bond(port->name, &bond);
2974 port_update_vlan_compat(struct port *port)
2976 struct bridge *br = port->bridge;
2977 char *vlandev_name = NULL;
2979 if (port->vlan > 0) {
2980 /* Figure out the name that the VLAN device should actually have, if it
2981 * existed. This takes some work because the VLAN device would not
2982 * have port->name in its name; rather, it would have the trunk port's
2983 * name, and 'port' would be attached to a bridge that also had the
2984 * VLAN device one of its ports. So we need to find a trunk port that
2985 * includes port->vlan.
2987 * There might be more than one candidate. This doesn't happen on
2988 * XenServer, so if it happens we just pick the first choice in
2989 * alphabetical order instead of creating multiple VLAN devices. */
2991 for (i = 0; i < br->n_ports; i++) {
2992 struct port *p = br->ports[i];
2993 if (port_trunks_vlan(p, port->vlan)
2995 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2997 const uint8_t *ea = p->ifaces[0]->mac;
2998 if (!eth_addr_is_multicast(ea) &&
2999 !eth_addr_is_reserved(ea) &&
3000 !eth_addr_is_zero(ea)) {
3001 vlandev_name = p->name;
3006 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3009 /* Interface functions. */
3012 iface_create(struct port *port, const char *name)
3014 struct iface *iface;
3016 iface = xcalloc(1, sizeof *iface);
3018 iface->port_ifidx = port->n_ifaces;
3019 iface->name = xstrdup(name);
3020 iface->dp_ifidx = -1;
3021 iface->tag = tag_create_random();
3022 iface->delay_expires = LLONG_MAX;
3024 if (!cfg_get_bool(0, "iface.%s.internal", iface->name)) {
3025 netdev_nodev_get_etheraddr(name, iface->mac);
3026 netdev_nodev_get_carrier(name, &iface->enabled);
3028 /* Internal interfaces are created later by the call to dpif_port_add()
3029 * in bridge_reconfigure(). Until then, we can't obtain any
3030 * information about them. (There's no real value in doing so, anyway,
3031 * because the 'mac' and 'enabled' values are only used for interfaces
3032 * that are bond slaves, and it doesn't normally make sense to bond an
3033 * internal interface.) */
3036 if (port->n_ifaces >= port->allocated_ifaces) {
3037 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3038 sizeof *port->ifaces);
3040 port->ifaces[port->n_ifaces++] = iface;
3041 if (port->n_ifaces > 1) {
3042 port->bridge->has_bonded_ports = true;
3045 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3047 port_update_bonding(port);
3048 bridge_flush(port->bridge);
3052 iface_destroy(struct iface *iface)
3055 struct port *port = iface->port;
3056 struct bridge *br = port->bridge;
3057 bool del_active = port->active_iface == iface->port_ifidx;
3060 if (iface->dp_ifidx >= 0) {
3061 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3064 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3065 del->port_ifidx = iface->port_ifidx;
3071 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3072 bond_choose_active_iface(port);
3073 bond_send_learning_packets(port);
3076 port_update_bonding(port);
3077 bridge_flush(port->bridge);
3081 static struct iface *
3082 iface_lookup(const struct bridge *br, const char *name)
3086 for (i = 0; i < br->n_ports; i++) {
3087 struct port *port = br->ports[i];
3088 for (j = 0; j < port->n_ifaces; j++) {
3089 struct iface *iface = port->ifaces[j];
3090 if (!strcmp(iface->name, name)) {
3098 static struct iface *
3099 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3101 return port_array_get(&br->ifaces, dp_ifidx);
3104 /* Port mirroring. */
3107 mirror_reconfigure(struct bridge *br)
3109 struct svec old_mirrors, new_mirrors;
3112 /* Collect old and new mirrors. */
3113 svec_init(&old_mirrors);
3114 svec_init(&new_mirrors);
3115 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3116 for (i = 0; i < MAX_MIRRORS; i++) {
3117 if (br->mirrors[i]) {
3118 svec_add(&old_mirrors, br->mirrors[i]->name);
3122 /* Get rid of deleted mirrors and add new mirrors. */
3123 svec_sort(&old_mirrors);
3124 assert(svec_is_unique(&old_mirrors));
3125 svec_sort(&new_mirrors);
3126 assert(svec_is_unique(&new_mirrors));
3127 for (i = 0; i < MAX_MIRRORS; i++) {
3128 struct mirror *m = br->mirrors[i];
3129 if (m && !svec_contains(&new_mirrors, m->name)) {
3133 for (i = 0; i < new_mirrors.n; i++) {
3134 const char *name = new_mirrors.names[i];
3135 if (!svec_contains(&old_mirrors, name)) {
3136 mirror_create(br, name);
3139 svec_destroy(&old_mirrors);
3140 svec_destroy(&new_mirrors);
3142 /* Reconfigure all mirrors. */
3143 for (i = 0; i < MAX_MIRRORS; i++) {
3144 if (br->mirrors[i]) {
3145 mirror_reconfigure_one(br->mirrors[i]);
3149 /* Update port reserved status. */
3150 for (i = 0; i < br->n_ports; i++) {
3151 br->ports[i]->is_mirror_output_port = false;
3153 for (i = 0; i < MAX_MIRRORS; i++) {
3154 struct mirror *m = br->mirrors[i];
3155 if (m && m->out_port) {
3156 m->out_port->is_mirror_output_port = true;
3162 mirror_create(struct bridge *br, const char *name)
3167 for (i = 0; ; i++) {
3168 if (i >= MAX_MIRRORS) {
3169 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3170 "cannot create %s", br->name, MAX_MIRRORS, name);
3173 if (!br->mirrors[i]) {
3178 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3181 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3184 m->name = xstrdup(name);
3185 svec_init(&m->src_ports);
3186 svec_init(&m->dst_ports);
3194 mirror_destroy(struct mirror *m)
3197 struct bridge *br = m->bridge;
3200 for (i = 0; i < br->n_ports; i++) {
3201 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3202 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3205 svec_destroy(&m->src_ports);
3206 svec_destroy(&m->dst_ports);
3209 m->bridge->mirrors[m->idx] = NULL;
3217 prune_ports(struct mirror *m, struct svec *ports)
3222 svec_sort_unique(ports);
3225 for (i = 0; i < ports->n; i++) {
3226 const char *name = ports->names[i];
3227 if (port_lookup(m->bridge, name)) {
3228 svec_add(&tmp, name);
3230 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3231 m->bridge->name, m->name, name);
3234 svec_swap(ports, &tmp);
3239 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3243 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3244 * order won't give us numeric sort order. But that's good enough for what
3245 * we need right now. */
3246 svec_sort_unique(vlan_strings);
3248 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3250 for (i = 0; i < vlan_strings->n; i++) {
3251 const char *name = vlan_strings->names[i];
3253 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3254 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3255 m->bridge->name, m->name, name);
3257 (*vlans)[n_vlans++] = vlan;
3264 vlan_is_mirrored(const struct mirror *m, int vlan)
3268 for (i = 0; i < m->n_vlans; i++) {
3269 if (m->vlans[i] == vlan) {
3277 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3281 for (i = 0; i < m->n_vlans; i++) {
3282 if (port_trunks_vlan(p, m->vlans[i])) {
3290 mirror_reconfigure_one(struct mirror *m)
3292 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3293 struct svec src_ports, dst_ports, ports;
3294 struct svec vlan_strings;
3295 mirror_mask_t mirror_bit;
3296 const char *out_port_name;
3297 struct port *out_port;
3302 bool mirror_all_ports;
3304 /* Get output port. */
3305 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3306 m->bridge->name, m->name);
3307 if (out_port_name) {
3308 out_port = port_lookup(m->bridge, out_port_name);
3310 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3311 "named %s", pfx, m->bridge->name, out_port_name);
3318 if (cfg_has("%s.output.vlan", pfx)) {
3319 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3320 "ignoring %s.output.vlan", pfx, pfx, pfx);
3322 } else if (cfg_has("%s.output.vlan", pfx)) {
3324 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3326 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3327 "but exactly one is required; disabling port mirror %s",
3328 pfx, pfx, pfx, pfx);
3334 /* Get all the ports, and drop duplicates and ports that don't exist. */
3335 svec_init(&src_ports);
3336 svec_init(&dst_ports);
3338 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3339 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3340 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3341 svec_append(&src_ports, &ports);
3342 svec_append(&dst_ports, &ports);
3343 svec_destroy(&ports);
3344 prune_ports(m, &src_ports);
3345 prune_ports(m, &dst_ports);
3347 /* Get all the vlans, and drop duplicate and invalid vlans. */
3348 svec_init(&vlan_strings);
3349 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3350 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3351 svec_destroy(&vlan_strings);
3353 /* Update mirror data. */
3354 if (!svec_equal(&m->src_ports, &src_ports)
3355 || !svec_equal(&m->dst_ports, &dst_ports)
3356 || m->n_vlans != n_vlans
3357 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3358 || m->out_port != out_port
3359 || m->out_vlan != out_vlan) {
3360 bridge_flush(m->bridge);
3362 svec_swap(&m->src_ports, &src_ports);
3363 svec_swap(&m->dst_ports, &dst_ports);
3366 m->n_vlans = n_vlans;
3367 m->out_port = out_port;
3368 m->out_vlan = out_vlan;
3370 /* If no selection criteria have been given, mirror for all ports. */
3371 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3374 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3375 for (i = 0; i < m->bridge->n_ports; i++) {
3376 struct port *port = m->bridge->ports[i];
3378 if (mirror_all_ports
3379 || svec_contains(&m->src_ports, port->name)
3382 ? port_trunks_any_mirrored_vlan(m, port)
3383 : vlan_is_mirrored(m, port->vlan)))) {
3384 port->src_mirrors |= mirror_bit;
3386 port->src_mirrors &= ~mirror_bit;
3389 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3390 port->dst_mirrors |= mirror_bit;
3392 port->dst_mirrors &= ~mirror_bit;
3397 svec_destroy(&src_ports);
3398 svec_destroy(&dst_ports);
3402 /* Spanning tree protocol. */
3404 static void brstp_update_port_state(struct port *);
3407 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3409 struct bridge *br = br_;
3410 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3411 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3413 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3415 } else if (eth_addr_is_zero(iface->mac)) {
3416 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3419 union ofp_action action;
3420 struct eth_header *eth = pkt->l2;
3423 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3425 memset(&action, 0, sizeof action);
3426 action.type = htons(OFPAT_OUTPUT);
3427 action.output.len = htons(sizeof action);
3428 action.output.port = htons(port_no);
3430 flow_extract(pkt, ODPP_NONE, &flow);
3431 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3437 brstp_reconfigure(struct bridge *br)
3441 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3443 stp_destroy(br->stp);
3449 uint64_t bridge_address, bridge_id;
3450 int bridge_priority;
3452 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3453 if (!bridge_address) {
3455 bridge_address = (stp_get_bridge_id(br->stp)
3456 & ((UINT64_C(1) << 48) - 1));
3458 uint8_t mac[ETH_ADDR_LEN];
3459 eth_addr_random(mac);
3460 bridge_address = eth_addr_to_uint64(mac);
3464 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3466 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3468 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3471 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3473 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3474 br->stp_last_tick = time_msec();
3477 if (bridge_id != stp_get_bridge_id(br->stp)) {
3478 stp_set_bridge_id(br->stp, bridge_id);
3483 for (i = 0; i < br->n_ports; i++) {
3484 struct port *p = br->ports[i];
3486 struct stp_port *sp;
3487 int path_cost, priority;
3493 dp_ifidx = p->ifaces[0]->dp_ifidx;
3494 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3498 sp = stp_get_port(br->stp, dp_ifidx);
3499 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3500 "stp.%s.port.%s.enabled",
3502 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3503 br->name, p->name));
3504 if (p->is_mirror_output_port) {
3507 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3508 bridge_flush(br); /* Might not be necessary. */
3510 stp_port_enable(sp);
3512 stp_port_disable(sp);
3516 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3518 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3520 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3521 "stp.%s.port.%s.priority",
3523 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3525 : STP_DEFAULT_PORT_PRIORITY);
3526 stp_port_set_priority(sp, priority);
3529 brstp_adjust_timers(br);
3531 for (i = 0; i < br->n_ports; i++) {
3532 brstp_update_port_state(br->ports[i]);
3537 brstp_update_port_state(struct port *p)
3539 struct bridge *br = p->bridge;
3540 enum stp_state state;
3542 /* Figure out new state. */
3543 state = STP_DISABLED;
3544 if (br->stp && p->n_ifaces > 0) {
3545 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3546 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3547 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3552 if (p->stp_state != state) {
3553 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3554 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3555 p->name, stp_state_name(p->stp_state),
3556 stp_state_name(state));
3557 if (p->stp_state == STP_DISABLED) {
3560 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3562 p->stp_state = state;
3563 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3564 : tag_create_random());
3569 brstp_adjust_timers(struct bridge *br)
3571 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3572 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3573 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3575 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3576 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3577 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3581 brstp_run(struct bridge *br)
3584 long long int now = time_msec();
3585 long long int elapsed = now - br->stp_last_tick;
3586 struct stp_port *sp;
3589 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3590 br->stp_last_tick = now;
3592 while (stp_get_changed_port(br->stp, &sp)) {
3593 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3595 brstp_update_port_state(p);
3602 brstp_wait(struct bridge *br)
3605 poll_timer_wait(1000);