1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
47 #include "poll-loop.h"
48 #include "port-array.h"
49 #include "proc-net-compat.h"
51 #include "secchan/ofproto.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id;
74 struct port *port; /* Containing port. */
75 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 int dp_ifidx; /* Index within kernel datapath. */
80 uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */
82 tag_type tag; /* Tag associated with this interface. */
83 bool enabled; /* May be chosen for flows? */
84 long long delay_expires; /* Time after which 'enabled' may change. */
87 #define BOND_MASK 0xff
89 int iface_idx; /* Index of assigned iface, or -1 if none. */
90 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
91 tag_type iface_tag; /* Tag associated with iface_idx. */
94 #define MAX_MIRRORS 32
95 typedef uint32_t mirror_mask_t;
96 #define MIRROR_MASK_C(X) UINT32_C(X)
97 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
99 struct bridge *bridge;
103 /* Selection criteria. */
104 struct svec src_ports;
105 struct svec dst_ports;
110 struct port *out_port;
114 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
116 struct bridge *bridge;
118 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
119 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
122 /* An ordinary bridge port has 1 interface.
123 * A bridge port for bonding has at least 2 interfaces. */
124 struct iface **ifaces;
125 size_t n_ifaces, allocated_ifaces;
128 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
129 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
130 tag_type active_iface_tag; /* Tag for bcast flows. */
131 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
132 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
133 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
135 /* Port mirroring info. */
136 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
137 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
138 bool is_mirror_output_port; /* Does port mirroring send frames here? */
140 /* Spanning tree info. */
141 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
142 tag_type stp_state_tag; /* Tag for STP state change. */
145 #define DP_MAX_PORTS 255
147 struct list node; /* Node in global list of bridges. */
148 char *name; /* User-specified arbitrary name. */
149 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
150 bool sent_config_request; /* Successfully sent config request? */
151 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
153 /* Support for remote controllers. */
154 char *controller; /* NULL if there is no remote controller;
155 * "discover" to do controller discovery;
156 * otherwise a vconn name. */
158 /* OpenFlow switch processing. */
159 struct ofproto *ofproto; /* OpenFlow switch. */
161 /* Kernel datapath information. */
162 struct dpif dpif; /* Kernel datapath. */
163 struct port_array ifaces; /* Indexed by kernel datapath port number. */
167 size_t n_ports, allocated_ports;
170 bool has_bonded_ports;
171 long long int bond_next_rebalance;
176 /* Flow statistics gathering. */
177 time_t next_stats_request;
179 /* Port mirroring. */
180 struct mirror *mirrors[MAX_MIRRORS];
184 long long int stp_last_tick;
187 /* List of all bridges. */
188 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
190 /* Maximum number of datapaths. */
191 enum { DP_MAX = 256 };
193 static struct bridge *bridge_create(const char *name);
194 static void bridge_destroy(struct bridge *);
195 static struct bridge *bridge_lookup(const char *name);
196 static void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *);
197 static int bridge_run_one(struct bridge *);
198 static void bridge_reconfigure_one(struct bridge *);
199 static void bridge_reconfigure_controller(struct bridge *);
200 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
201 static void bridge_fetch_dp_ifaces(struct bridge *);
202 static void bridge_flush(struct bridge *);
203 static void bridge_pick_local_hw_addr(struct bridge *,
204 uint8_t ea[ETH_ADDR_LEN],
205 const char **devname);
206 static uint64_t bridge_pick_datapath_id(struct bridge *,
207 const uint8_t bridge_ea[ETH_ADDR_LEN],
208 const char *devname);
209 static uint64_t dpid_from_hash(const void *, size_t nbytes);
211 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
213 static void bond_init(void);
214 static void bond_run(struct bridge *);
215 static void bond_wait(struct bridge *);
216 static void bond_rebalance_port(struct port *);
217 static void bond_send_learning_packets(struct port *);
219 static void port_create(struct bridge *, const char *name);
220 static void port_reconfigure(struct port *);
221 static void port_destroy(struct port *);
222 static struct port *port_lookup(const struct bridge *, const char *name);
223 static struct iface *port_lookup_iface(const struct port *, const char *name);
224 static struct port *port_from_dp_ifidx(const struct bridge *,
226 static void port_update_bond_compat(struct port *);
227 static void port_update_vlan_compat(struct port *);
229 static void mirror_create(struct bridge *, const char *name);
230 static void mirror_destroy(struct mirror *);
231 static void mirror_reconfigure(struct bridge *);
232 static void mirror_reconfigure_one(struct mirror *);
233 static bool vlan_is_mirrored(const struct mirror *, int vlan);
235 static void brstp_reconfigure(struct bridge *);
236 static void brstp_adjust_timers(struct bridge *);
237 static void brstp_run(struct bridge *);
238 static void brstp_wait(struct bridge *);
240 static void iface_create(struct port *, const char *name);
241 static void iface_destroy(struct iface *);
242 static struct iface *iface_lookup(const struct bridge *, const char *name);
243 static struct iface *iface_from_dp_ifidx(const struct bridge *,
245 static bool iface_is_internal(const struct bridge *, const char *name);
246 static void iface_set_mac(struct iface *);
248 /* Hooks into ofproto processing. */
249 static struct ofhooks bridge_ofhooks;
251 /* Public functions. */
253 /* Adds the name of each interface used by a bridge, including local and
254 * internal ports, to 'svec'. */
256 bridge_get_ifaces(struct svec *svec)
258 struct bridge *br, *next;
261 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
262 for (i = 0; i < br->n_ports; i++) {
263 struct port *port = br->ports[i];
265 for (j = 0; j < port->n_ifaces; j++) {
266 struct iface *iface = port->ifaces[j];
267 if (iface->dp_ifidx < 0) {
268 VLOG_ERR("%s interface not in dp%u, ignoring",
269 iface->name, dpif_id(&br->dpif));
271 if (iface->dp_ifidx != ODPP_LOCAL) {
272 svec_add(svec, iface->name);
280 /* The caller must already have called cfg_read(). */
289 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
291 for (i = 0; i < DP_MAX; i++) {
295 sprintf(devname, "dp%d", i);
296 retval = dpif_open(devname, &dpif);
298 char dpif_name[IF_NAMESIZE];
299 if (dpif_get_name(&dpif, dpif_name, sizeof dpif_name)
300 || !cfg_has("bridge.%s.port", dpif_name)) {
304 } else if (retval != ENODEV) {
305 VLOG_ERR("failed to delete datapath dp%d: %s",
306 i, strerror(retval));
310 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows);
312 bridge_reconfigure();
317 config_string_change(const char *key, char **valuep)
319 const char *value = cfg_get_string(0, "%s", key);
320 if (value && (!*valuep || strcmp(value, *valuep))) {
322 *valuep = xstrdup(value);
330 bridge_configure_ssl(void)
332 /* XXX SSL should be configurable on a per-bridge basis.
333 * XXX should be possible to de-configure SSL. */
334 static char *private_key_file;
335 static char *certificate_file;
336 static char *cacert_file;
339 if (config_string_change("ssl.private-key", &private_key_file)) {
340 vconn_ssl_set_private_key_file(private_key_file);
343 if (config_string_change("ssl.certificate", &certificate_file)) {
344 vconn_ssl_set_certificate_file(certificate_file);
347 /* We assume that even if the filename hasn't changed, if the CA cert
348 * file has been removed, that we want to move back into
349 * boot-strapping mode. This opens a small security hole, because
350 * the old certificate will still be trusted until vSwitch is
351 * restarted. We may want to address this in vconn's SSL library. */
352 if (config_string_change("ssl.ca-cert", &cacert_file)
353 || (stat(cacert_file, &s) && errno == ENOENT)) {
354 vconn_ssl_set_ca_cert_file(cacert_file,
355 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
361 bridge_reconfigure(void)
363 struct svec old_br, new_br, raw_new_br;
364 struct bridge *br, *next;
367 COVERAGE_INC(bridge_reconfigure);
369 /* Collect old bridges. */
371 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
372 svec_add(&old_br, br->name);
375 /* Collect new bridges. */
376 svec_init(&raw_new_br);
377 cfg_get_subsections(&raw_new_br, "bridge");
379 for (i = 0; i < raw_new_br.n; i++) {
380 const char *name = raw_new_br.names[i];
381 if ((!strncmp(name, "dp", 2) && isdigit(name[2])) ||
382 (!strncmp(name, "nl:", 3) && isdigit(name[3]))) {
383 VLOG_ERR("%s is not a valid bridge name (bridges may not be "
384 "named \"dp\" or \"nl:\" followed by a digit)", name);
386 svec_add(&new_br, name);
389 svec_destroy(&raw_new_br);
391 /* Get rid of deleted bridges and add new bridges. */
394 assert(svec_is_unique(&old_br));
395 assert(svec_is_unique(&new_br));
396 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
397 if (!svec_contains(&new_br, br->name)) {
401 for (i = 0; i < new_br.n; i++) {
402 const char *name = new_br.names[i];
403 if (!svec_contains(&old_br, name)) {
407 svec_destroy(&old_br);
408 svec_destroy(&new_br);
412 bridge_configure_ssl();
415 /* Reconfigure all bridges. */
416 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
417 bridge_reconfigure_one(br);
420 /* Add and delete ports on all datapaths.
422 * The kernel will reject any attempt to add a given port to a datapath if
423 * that port already belongs to a different datapath, so we must do all
424 * port deletions before any port additions. */
425 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
426 struct odp_port *dpif_ports;
428 struct svec want_ifaces;
430 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
431 bridge_get_all_ifaces(br, &want_ifaces);
432 for (i = 0; i < n_dpif_ports; i++) {
433 const struct odp_port *p = &dpif_ports[i];
434 if (!svec_contains(&want_ifaces, p->devname)
435 && strcmp(p->devname, br->name)) {
436 int retval = dpif_port_del(&br->dpif, p->port);
438 VLOG_ERR("failed to remove %s interface from dp%u: %s",
439 p->devname, dpif_id(&br->dpif), strerror(retval));
443 svec_destroy(&want_ifaces);
446 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
447 struct odp_port *dpif_ports;
449 struct svec cur_ifaces, want_ifaces, add_ifaces;
452 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
453 svec_init(&cur_ifaces);
454 for (i = 0; i < n_dpif_ports; i++) {
455 svec_add(&cur_ifaces, dpif_ports[i].devname);
458 svec_sort_unique(&cur_ifaces);
459 bridge_get_all_ifaces(br, &want_ifaces);
460 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
463 for (i = 0; i < add_ifaces.n; i++) {
464 const char *if_name = add_ifaces.names[i];
469 /* Add to datapath. */
470 internal = iface_is_internal(br, if_name);
471 error = dpif_port_add(&br->dpif, if_name, next_port_no++,
472 internal ? ODP_PORT_INTERNAL : 0);
473 if (error != EEXIST) {
474 if (next_port_no >= 256) {
475 VLOG_ERR("ran out of valid port numbers on dp%u",
480 VLOG_ERR("failed to add %s interface to dp%u: %s",
481 if_name, dpif_id(&br->dpif), strerror(error));
488 svec_destroy(&cur_ifaces);
489 svec_destroy(&want_ifaces);
490 svec_destroy(&add_ifaces);
492 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
495 struct iface *local_iface = NULL;
497 uint8_t engine_type = br->dpif.minor;
498 uint8_t engine_id = br->dpif.minor;
499 bool add_id_to_iface = false;
500 struct svec nf_hosts;
502 bridge_fetch_dp_ifaces(br);
503 for (i = 0; i < br->n_ports; ) {
504 struct port *port = br->ports[i];
506 for (j = 0; j < port->n_ifaces; ) {
507 struct iface *iface = port->ifaces[j];
508 if (iface->dp_ifidx < 0) {
509 VLOG_ERR("%s interface not in dp%u, dropping",
510 iface->name, dpif_id(&br->dpif));
511 iface_destroy(iface);
513 if (iface->dp_ifidx == ODPP_LOCAL) {
516 VLOG_DBG("dp%u has interface %s on port %d",
517 dpif_id(&br->dpif), iface->name, iface->dp_ifidx);
521 if (!port->n_ifaces) {
522 VLOG_ERR("%s port has no interfaces, dropping", port->name);
529 /* Pick local port hardware address, datapath ID. */
530 bridge_pick_local_hw_addr(br, ea, &devname);
532 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
534 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
535 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
536 "Ethernet address: %s",
537 br->name, strerror(error));
541 dpid = bridge_pick_datapath_id(br, ea, devname);
542 ofproto_set_datapath_id(br->ofproto, dpid);
544 /* Set NetFlow configuration on this bridge. */
545 if (cfg_has("netflow.%s.engine-type", br->name)) {
546 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
549 if (cfg_has("netflow.%s.engine-id", br->name)) {
550 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
552 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
553 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
556 if (add_id_to_iface && engine_id > 0x7f) {
557 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
558 "another vswitch, choose an engine id less than 128",
561 if (add_id_to_iface && br->n_ports > 0x1ff) {
562 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
563 "another port when 512 or more ports are used",
566 svec_init(&nf_hosts);
567 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
568 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
569 engine_id, add_id_to_iface)) {
570 VLOG_ERR("bridge %s: problem setting netflow collectors",
573 svec_destroy(&nf_hosts);
575 /* Update the controller and related settings. It would be more
576 * straightforward to call this from bridge_reconfigure_one(), but we
577 * can't do it there for two reasons. First, and most importantly, at
578 * that point we don't know the dp_ifidx of any interfaces that have
579 * been added to the bridge (because we haven't actually added them to
580 * the datapath). Second, at that point we haven't set the datapath ID
581 * yet; when a controller is configured, resetting the datapath ID will
582 * immediately disconnect from the controller, so it's better to set
583 * the datapath ID before the controller. */
584 bridge_reconfigure_controller(br);
586 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
587 for (i = 0; i < br->n_ports; i++) {
588 struct port *port = br->ports[i];
590 port_update_vlan_compat(port);
592 for (j = 0; j < port->n_ifaces; j++) {
593 struct iface *iface = port->ifaces[j];
594 if (iface->dp_ifidx != ODPP_LOCAL
595 && iface_is_internal(br, iface->name)) {
596 iface_set_mac(iface);
601 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
602 brstp_reconfigure(br);
607 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
608 const char **devname)
610 uint64_t requested_ea;
616 /* Did the user request a particular MAC? */
617 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
619 eth_addr_from_uint64(requested_ea, ea);
620 if (eth_addr_is_multicast(ea)) {
621 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
622 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
623 } else if (eth_addr_is_zero(ea)) {
624 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
630 /* Otherwise choose the minimum MAC address among all of the interfaces.
631 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
632 * MAC of the physical interface in such an environment.) */
633 memset(ea, 0xff, sizeof ea);
634 for (i = 0; i < br->n_ports; i++) {
635 struct port *port = br->ports[i];
636 uint8_t iface_ea[ETH_ADDR_LEN];
637 uint64_t iface_ea_u64;
640 /* Mirror output ports don't participate. */
641 if (port->is_mirror_output_port) {
645 /* Choose the MAC address to represent the port. */
646 iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name);
648 /* User specified explicitly. */
649 eth_addr_from_uint64(iface_ea_u64, iface_ea);
651 /* Find the interface with this Ethernet address (if any) so that
652 * we can provide the correct devname to the caller. */
654 for (j = 0; j < port->n_ifaces; j++) {
655 struct iface *candidate = port->ifaces[j];
656 uint8_t candidate_ea[ETH_ADDR_LEN];
657 if (!netdev_nodev_get_etheraddr(candidate->name, candidate_ea)
658 && eth_addr_equals(iface_ea, candidate_ea)) {
663 /* Choose the interface whose MAC address will represent the port.
664 * The Linux kernel bonding code always chooses the MAC address of
665 * the first slave added to a bond, and the Fedora networking
666 * scripts always add slaves to a bond in alphabetical order, so
667 * for compatibility we choose the interface with the name that is
668 * first in alphabetical order. */
669 iface = port->ifaces[0];
670 for (j = 1; j < port->n_ifaces; j++) {
671 struct iface *candidate = port->ifaces[j];
672 if (strcmp(candidate->name, iface->name) < 0) {
677 /* The local port doesn't count (since we're trying to choose its
678 * MAC address anyway). Other internal ports don't count because
679 * we really want a physical MAC if we can get it, and internal
680 * ports typically have randomly generated MACs. */
681 if (iface->dp_ifidx == ODPP_LOCAL
682 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
687 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
689 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
690 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
691 iface->name, strerror(error));
696 /* Compare against our current choice. */
697 if (!eth_addr_is_multicast(iface_ea) &&
698 !eth_addr_is_reserved(iface_ea) &&
699 !eth_addr_is_zero(iface_ea) &&
700 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
702 memcpy(ea, iface_ea, ETH_ADDR_LEN);
703 *devname = iface ? iface->name : NULL;
706 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
707 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
709 VLOG_WARN("bridge %s: using default bridge Ethernet "
710 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
712 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
713 br->name, ETH_ADDR_ARGS(ea));
717 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
718 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
719 * a network device, then that network device's name must be passed in as
720 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
721 * passed in as a null pointer. */
723 bridge_pick_datapath_id(struct bridge *br,
724 const uint8_t bridge_ea[ETH_ADDR_LEN],
728 * The procedure for choosing a bridge MAC address will, in the most
729 * ordinary case, also choose a unique MAC that we can use as a datapath
730 * ID. In some special cases, though, multiple bridges will end up with
731 * the same MAC address. This is OK for the bridges, but it will confuse
732 * the OpenFlow controller, because each datapath needs a unique datapath
735 * Datapath IDs must be unique. It is also very desirable that they be
736 * stable from one run to the next, so that policy set on a datapath
741 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
748 if (!netdev_get_vlan_vid(devname, &vlan)) {
750 * A bridge whose MAC address is taken from a VLAN network device
751 * (that is, a network device created with vconfig(8) or similar
752 * tool) will have the same MAC address as a bridge on the VLAN
753 * device's physical network device.
755 * Handle this case by hashing the physical network device MAC
756 * along with the VLAN identifier.
758 uint8_t buf[ETH_ADDR_LEN + 2];
759 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
760 buf[ETH_ADDR_LEN] = vlan >> 8;
761 buf[ETH_ADDR_LEN + 1] = vlan;
762 return dpid_from_hash(buf, sizeof buf);
765 * Assume that this bridge's MAC address is unique, since it
766 * doesn't fit any of the cases we handle specially.
771 * A purely internal bridge, that is, one that has no non-virtual
772 * network devices on it at all, is more difficult because it has no
773 * natural unique identifier at all.
775 * When the host is a XenServer, we handle this case by hashing the
776 * host's UUID with the name of the bridge. Names of bridges are
777 * persistent across XenServer reboots, although they can be reused if
778 * an internal network is destroyed and then a new one is later
779 * created, so this is fairly effective.
781 * When the host is not a XenServer, we punt by using a random MAC
782 * address on each run.
784 const char *host_uuid = xenserver_get_host_uuid();
786 char *combined = xasprintf("%s,%s", host_uuid, br->name);
787 dpid = dpid_from_hash(combined, strlen(combined));
793 return eth_addr_to_uint64(bridge_ea);
797 dpid_from_hash(const void *data, size_t n)
799 uint8_t hash[SHA1_DIGEST_SIZE];
801 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
802 sha1_bytes(data, n, hash);
803 eth_addr_mark_random(hash);
804 return eth_addr_to_uint64(hash);
810 struct bridge *br, *next;
814 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
815 int error = bridge_run_one(br);
817 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
818 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
819 "forcing reconfiguration", br->name);
833 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
834 ofproto_wait(br->ofproto);
835 if (br->controller) {
840 mac_learning_wait(br->ml);
847 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
848 * configuration changes. */
850 bridge_flush(struct bridge *br)
852 COVERAGE_INC(bridge_flush);
855 mac_learning_flush(br->ml);
859 /* Bridge unixctl user interface functions. */
861 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
863 struct ds ds = DS_EMPTY_INITIALIZER;
864 const struct bridge *br;
866 br = bridge_lookup(args);
868 unixctl_command_reply(conn, 501, "no such bridge");
872 ds_put_cstr(&ds, " port VLAN MAC Age\n");
874 const struct mac_entry *e;
875 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
876 if (e->port < 0 || e->port >= br->n_ports) {
879 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
880 br->ports[e->port]->ifaces[0]->dp_ifidx,
881 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
884 unixctl_command_reply(conn, 200, ds_cstr(&ds));
888 /* Bridge reconfiguration functions. */
890 static struct bridge *
891 bridge_create(const char *name)
896 assert(!bridge_lookup(name));
897 br = xcalloc(1, sizeof *br);
899 error = dpif_create(name, &br->dpif);
900 if (error == EEXIST) {
901 error = dpif_open(name, &br->dpif);
903 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
904 name, strerror(error));
908 dpif_flow_flush(&br->dpif);
910 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
915 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
917 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
918 dpif_delete(&br->dpif);
919 dpif_close(&br->dpif);
924 br->name = xstrdup(name);
925 br->ml = mac_learning_create();
926 br->sent_config_request = false;
927 eth_addr_random(br->default_ea);
929 port_array_init(&br->ifaces);
932 br->bond_next_rebalance = time_msec() + 10000;
934 list_push_back(&all_bridges, &br->node);
936 VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif));
942 bridge_destroy(struct bridge *br)
947 while (br->n_ports > 0) {
948 port_destroy(br->ports[br->n_ports - 1]);
950 list_remove(&br->node);
951 error = dpif_delete(&br->dpif);
952 if (error && error != ENOENT) {
953 VLOG_ERR("failed to delete dp%u: %s",
954 dpif_id(&br->dpif), strerror(error));
956 dpif_close(&br->dpif);
957 ofproto_destroy(br->ofproto);
958 free(br->controller);
959 mac_learning_destroy(br->ml);
960 port_array_destroy(&br->ifaces);
967 static struct bridge *
968 bridge_lookup(const char *name)
972 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
973 if (!strcmp(br->name, name)) {
981 bridge_exists(const char *name)
983 return bridge_lookup(name) ? true : false;
987 bridge_get_datapathid(const char *name)
989 struct bridge *br = bridge_lookup(name);
990 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
993 /* Handle requests for a listing of all flows known by the OpenFlow
994 * stack, including those normally hidden. */
996 bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args)
1001 br = bridge_lookup(args);
1003 unixctl_command_reply(conn, 501, "Unknown bridge");
1008 ofproto_get_all_flows(br->ofproto, &results);
1010 unixctl_command_reply(conn, 200, ds_cstr(&results));
1011 ds_destroy(&results);
1015 bridge_run_one(struct bridge *br)
1019 error = ofproto_run1(br->ofproto);
1025 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1030 error = ofproto_run2(br->ofproto, br->flush);
1037 bridge_get_controller(const struct bridge *br)
1039 const char *controller;
1041 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1043 controller = cfg_get_string(0, "mgmt.controller");
1045 return controller && controller[0] ? controller : NULL;
1049 bridge_reconfigure_one(struct bridge *br)
1051 struct svec old_ports, new_ports, ifaces;
1052 struct svec listeners, old_listeners;
1053 struct svec snoops, old_snoops;
1056 /* Collect old ports. */
1057 svec_init(&old_ports);
1058 for (i = 0; i < br->n_ports; i++) {
1059 svec_add(&old_ports, br->ports[i]->name);
1061 svec_sort(&old_ports);
1062 assert(svec_is_unique(&old_ports));
1064 /* Collect new ports. */
1065 svec_init(&new_ports);
1066 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1067 svec_sort(&new_ports);
1068 if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) {
1069 svec_add(&new_ports, br->name);
1070 svec_sort(&new_ports);
1072 if (!svec_is_unique(&new_ports)) {
1073 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1074 br->name, svec_get_duplicate(&new_ports));
1075 svec_unique(&new_ports);
1078 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1080 /* Get rid of deleted ports and add new ports. */
1081 for (i = 0; i < br->n_ports; ) {
1082 struct port *port = br->ports[i];
1083 if (!svec_contains(&new_ports, port->name)) {
1089 for (i = 0; i < new_ports.n; i++) {
1090 const char *name = new_ports.names[i];
1091 if (!svec_contains(&old_ports, name)) {
1092 port_create(br, name);
1095 svec_destroy(&old_ports);
1096 svec_destroy(&new_ports);
1098 /* Reconfigure all ports. */
1099 for (i = 0; i < br->n_ports; i++) {
1100 port_reconfigure(br->ports[i]);
1103 /* Check and delete duplicate interfaces. */
1105 for (i = 0; i < br->n_ports; ) {
1106 struct port *port = br->ports[i];
1107 for (j = 0; j < port->n_ifaces; ) {
1108 struct iface *iface = port->ifaces[j];
1109 if (svec_contains(&ifaces, iface->name)) {
1110 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1112 br->name, iface->name, port->name);
1113 iface_destroy(iface);
1115 svec_add(&ifaces, iface->name);
1120 if (!port->n_ifaces) {
1121 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1127 svec_destroy(&ifaces);
1129 /* Delete all flows if we're switching from connected to standalone or vice
1130 * versa. (XXX Should we delete all flows if we are switching from one
1131 * controller to another?) */
1133 /* Configure OpenFlow management listeners. */
1134 svec_init(&listeners);
1135 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1137 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1138 ovs_rundir, br->name));
1139 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1140 svec_clear(&listeners);
1142 svec_sort_unique(&listeners);
1144 svec_init(&old_listeners);
1145 ofproto_get_listeners(br->ofproto, &old_listeners);
1146 svec_sort_unique(&old_listeners);
1148 if (!svec_equal(&listeners, &old_listeners)) {
1149 ofproto_set_listeners(br->ofproto, &listeners);
1151 svec_destroy(&listeners);
1152 svec_destroy(&old_listeners);
1154 /* Configure OpenFlow controller connection snooping. */
1156 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1158 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1159 ovs_rundir, br->name));
1160 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1161 svec_clear(&snoops);
1163 svec_sort_unique(&snoops);
1165 svec_init(&old_snoops);
1166 ofproto_get_snoops(br->ofproto, &old_snoops);
1167 svec_sort_unique(&old_snoops);
1169 if (!svec_equal(&snoops, &old_snoops)) {
1170 ofproto_set_snoops(br->ofproto, &snoops);
1172 svec_destroy(&snoops);
1173 svec_destroy(&old_snoops);
1175 mirror_reconfigure(br);
1179 bridge_reconfigure_controller(struct bridge *br)
1181 char *pfx = xasprintf("bridge.%s.controller", br->name);
1182 const char *controller;
1184 controller = bridge_get_controller(br);
1185 if ((br->controller != NULL) != (controller != NULL)) {
1186 ofproto_flush_flows(br->ofproto);
1188 free(br->controller);
1189 br->controller = controller ? xstrdup(controller) : NULL;
1192 const char *fail_mode;
1193 int max_backoff, probe;
1194 int rate_limit, burst_limit;
1196 if (!strcmp(controller, "discover")) {
1197 bool update_resolv_conf = true;
1199 if (cfg_has("%s.update-resolv.conf", pfx)) {
1200 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1203 ofproto_set_discovery(br->ofproto, true,
1204 cfg_get_string(0, "%s.accept-regex", pfx),
1205 update_resolv_conf);
1207 struct netdev *netdev;
1211 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1213 || cfg_get_bool(0, "%s.in-band", pfx));
1214 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1215 ofproto_set_in_band(br->ofproto, in_band);
1217 error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev);
1219 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1220 struct in_addr ip, mask, gateway;
1221 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1222 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1223 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1225 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1227 mask.s_addr = guess_netmask(ip.s_addr);
1229 if (!netdev_set_in4(netdev, ip, mask)) {
1230 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1232 br->name, IP_ARGS(&ip.s_addr),
1233 IP_ARGS(&mask.s_addr));
1236 if (gateway.s_addr) {
1237 if (!netdev_add_router(gateway)) {
1238 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1239 br->name, IP_ARGS(&gateway.s_addr));
1243 netdev_close(netdev);
1247 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1249 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1251 ofproto_set_failure(br->ofproto,
1253 || !strcmp(fail_mode, "standalone")
1254 || !strcmp(fail_mode, "open")));
1256 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1258 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1263 ofproto_set_probe_interval(br->ofproto, probe);
1265 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1267 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1272 ofproto_set_max_backoff(br->ofproto, max_backoff);
1274 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1276 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1278 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1280 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1282 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1284 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1286 if (cfg_has("%s.commands.acl", pfx)) {
1287 struct svec command_acls;
1290 svec_init(&command_acls);
1291 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1292 command_acl = svec_join(&command_acls, ",", "");
1294 ofproto_set_remote_execution(br->ofproto, command_acl,
1295 cfg_get_string(0, "%s.commands.dir",
1298 svec_destroy(&command_acls);
1301 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1304 union ofp_action action;
1307 /* Set up a flow that matches every packet and directs them to
1308 * OFPP_NORMAL (which goes to us). */
1309 memset(&action, 0, sizeof action);
1310 action.type = htons(OFPAT_OUTPUT);
1311 action.output.len = htons(sizeof action);
1312 action.output.port = htons(OFPP_NORMAL);
1313 memset(&flow, 0, sizeof flow);
1314 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1317 ofproto_set_in_band(br->ofproto, false);
1318 ofproto_set_max_backoff(br->ofproto, 1);
1319 ofproto_set_probe_interval(br->ofproto, 5);
1320 ofproto_set_failure(br->ofproto, false);
1321 ofproto_set_stp(br->ofproto, false);
1325 ofproto_set_controller(br->ofproto, br->controller);
1329 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1334 for (i = 0; i < br->n_ports; i++) {
1335 struct port *port = br->ports[i];
1336 for (j = 0; j < port->n_ifaces; j++) {
1337 struct iface *iface = port->ifaces[j];
1338 svec_add(ifaces, iface->name);
1340 if (port->n_ifaces > 1
1341 && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
1342 svec_add(ifaces, port->name);
1345 svec_sort_unique(ifaces);
1348 /* For robustness, in case the administrator moves around datapath ports behind
1349 * our back, we re-check all the datapath port numbers here.
1351 * This function will set the 'dp_ifidx' members of interfaces that have
1352 * disappeared to -1, so only call this function from a context where those
1353 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1354 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1355 * datapath, which doesn't support UINT16_MAX+1 ports. */
1357 bridge_fetch_dp_ifaces(struct bridge *br)
1359 struct odp_port *dpif_ports;
1360 size_t n_dpif_ports;
1363 /* Reset all interface numbers. */
1364 for (i = 0; i < br->n_ports; i++) {
1365 struct port *port = br->ports[i];
1366 for (j = 0; j < port->n_ifaces; j++) {
1367 struct iface *iface = port->ifaces[j];
1368 iface->dp_ifidx = -1;
1371 port_array_clear(&br->ifaces);
1373 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
1374 for (i = 0; i < n_dpif_ports; i++) {
1375 struct odp_port *p = &dpif_ports[i];
1376 struct iface *iface = iface_lookup(br, p->devname);
1378 if (iface->dp_ifidx >= 0) {
1379 VLOG_WARN("dp%u reported interface %s twice",
1380 dpif_id(&br->dpif), p->devname);
1381 } else if (iface_from_dp_ifidx(br, p->port)) {
1382 VLOG_WARN("dp%u reported interface %"PRIu16" twice",
1383 dpif_id(&br->dpif), p->port);
1385 port_array_set(&br->ifaces, p->port, iface);
1386 iface->dp_ifidx = p->port;
1393 /* Bridge packet processing functions. */
1396 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1398 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1401 static struct bond_entry *
1402 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1404 return &port->bond_hash[bond_hash(mac)];
1408 bond_choose_iface(const struct port *port)
1411 for (i = 0; i < port->n_ifaces; i++) {
1412 if (port->ifaces[i]->enabled) {
1420 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1421 uint16_t *dp_ifidx, tag_type *tags)
1423 struct iface *iface;
1425 assert(port->n_ifaces);
1426 if (port->n_ifaces == 1) {
1427 iface = port->ifaces[0];
1429 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1430 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1431 || !port->ifaces[e->iface_idx]->enabled) {
1432 /* XXX select interface properly. The current interface selection
1433 * is only good for testing the rebalancing code. */
1434 e->iface_idx = bond_choose_iface(port);
1435 if (e->iface_idx < 0) {
1436 *tags |= port->no_ifaces_tag;
1439 e->iface_tag = tag_create_random();
1440 ((struct port *) port)->bond_compat_is_stale = true;
1442 *tags |= e->iface_tag;
1443 iface = port->ifaces[e->iface_idx];
1445 *dp_ifidx = iface->dp_ifidx;
1446 *tags |= iface->tag; /* Currently only used for bonding. */
1451 bond_link_status_update(struct iface *iface, bool carrier)
1453 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1454 struct port *port = iface->port;
1456 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1457 /* Nothing to do. */
1460 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1461 iface->name, carrier ? "detected" : "dropped");
1462 if (carrier == iface->enabled) {
1463 iface->delay_expires = LLONG_MAX;
1464 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1465 iface->name, carrier ? "disabled" : "enabled");
1466 } else if (carrier && port->updelay && port->active_iface < 0) {
1467 iface->delay_expires = time_msec();
1468 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1469 "other interface is up", iface->name, port->updelay);
1471 int delay = carrier ? port->updelay : port->downdelay;
1472 iface->delay_expires = time_msec() + delay;
1475 "interface %s: will be %s if it stays %s for %d ms",
1477 carrier ? "enabled" : "disabled",
1478 carrier ? "up" : "down",
1485 bond_choose_active_iface(struct port *port)
1487 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1489 port->active_iface = bond_choose_iface(port);
1490 port->active_iface_tag = tag_create_random();
1491 if (port->active_iface >= 0) {
1492 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1493 port->name, port->ifaces[port->active_iface]->name);
1495 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1501 bond_enable_slave(struct iface *iface, bool enable)
1503 struct port *port = iface->port;
1504 struct bridge *br = port->bridge;
1506 iface->delay_expires = LLONG_MAX;
1507 if (enable == iface->enabled) {
1511 iface->enabled = enable;
1512 if (!iface->enabled) {
1513 VLOG_WARN("interface %s: disabled", iface->name);
1514 ofproto_revalidate(br->ofproto, iface->tag);
1515 if (iface->port_ifidx == port->active_iface) {
1516 ofproto_revalidate(br->ofproto,
1517 port->active_iface_tag);
1518 bond_choose_active_iface(port);
1520 bond_send_learning_packets(port);
1522 VLOG_WARN("interface %s: enabled", iface->name);
1523 if (port->active_iface < 0) {
1524 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1525 bond_choose_active_iface(port);
1526 bond_send_learning_packets(port);
1528 iface->tag = tag_create_random();
1530 port_update_bond_compat(port);
1534 bond_run(struct bridge *br)
1538 for (i = 0; i < br->n_ports; i++) {
1539 struct port *port = br->ports[i];
1541 if (port->bond_compat_is_stale) {
1542 port->bond_compat_is_stale = false;
1543 port_update_bond_compat(port);
1546 if (port->n_ifaces < 2) {
1549 for (j = 0; j < port->n_ifaces; j++) {
1550 struct iface *iface = port->ifaces[j];
1551 if (time_msec() >= iface->delay_expires) {
1552 bond_enable_slave(iface, !iface->enabled);
1559 bond_wait(struct bridge *br)
1563 for (i = 0; i < br->n_ports; i++) {
1564 struct port *port = br->ports[i];
1565 if (port->n_ifaces < 2) {
1568 for (j = 0; j < port->n_ifaces; j++) {
1569 struct iface *iface = port->ifaces[j];
1570 if (iface->delay_expires != LLONG_MAX) {
1571 poll_timer_wait(iface->delay_expires - time_msec());
1578 set_dst(struct dst *p, const flow_t *flow,
1579 const struct port *in_port, const struct port *out_port,
1584 * XXX This uses too many tags: any broadcast flow will get one tag per
1585 * destination port, and thus a broadcast on a switch of any size is likely
1586 * to have all tag bits set. We should figure out a way to be smarter.
1588 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1589 *tags |= out_port->stp_state_tag;
1590 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1594 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1595 : in_port->vlan >= 0 ? in_port->vlan
1596 : ntohs(flow->dl_vlan));
1597 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1601 swap_dst(struct dst *p, struct dst *q)
1603 struct dst tmp = *p;
1608 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1609 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1610 * that we push to the datapath. We could in fact fully sort the array by
1611 * vlan, but in most cases there are at most two different vlan tags so that's
1612 * possibly overkill.) */
1614 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1616 struct dst *first = dsts;
1617 struct dst *last = dsts + n_dsts;
1619 while (first != last) {
1621 * - All dsts < first have vlan == 'vlan'.
1622 * - All dsts >= last have vlan != 'vlan'.
1623 * - first < last. */
1624 while (first->vlan == vlan) {
1625 if (++first == last) {
1630 /* Same invariants, plus one additional:
1631 * - first->vlan != vlan.
1633 while (last[-1].vlan != vlan) {
1634 if (--last == first) {
1639 /* Same invariants, plus one additional:
1640 * - last[-1].vlan == vlan.*/
1641 swap_dst(first++, --last);
1646 mirror_mask_ffs(mirror_mask_t mask)
1648 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1653 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1654 const struct dst *test)
1657 for (i = 0; i < n_dsts; i++) {
1658 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1666 port_trunks_vlan(const struct port *port, uint16_t vlan)
1668 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1672 port_includes_vlan(const struct port *port, uint16_t vlan)
1674 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1678 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1679 const struct port *in_port, const struct port *out_port,
1680 struct dst dsts[], tag_type *tags)
1682 mirror_mask_t mirrors = in_port->src_mirrors;
1683 struct dst *dst = dsts;
1686 *tags |= in_port->stp_state_tag;
1687 if (out_port == FLOOD_PORT) {
1688 /* XXX use ODP_FLOOD if no vlans or bonding. */
1689 /* XXX even better, define each VLAN as a datapath port group */
1690 for (i = 0; i < br->n_ports; i++) {
1691 struct port *port = br->ports[i];
1692 if (port != in_port && port_includes_vlan(port, vlan)
1693 && !port->is_mirror_output_port
1694 && set_dst(dst, flow, in_port, port, tags)) {
1695 mirrors |= port->dst_mirrors;
1699 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1700 mirrors |= out_port->dst_mirrors;
1705 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1706 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1708 if (set_dst(dst, flow, in_port, m->out_port, tags)
1709 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1713 for (i = 0; i < br->n_ports; i++) {
1714 struct port *port = br->ports[i];
1715 if (port_includes_vlan(port, m->out_vlan)
1716 && set_dst(dst, flow, in_port, port, tags))
1718 if (port->vlan < 0) {
1719 dst->vlan = m->out_vlan;
1721 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1724 if (port == in_port && dst->vlan == vlan) {
1725 /* Don't send out input port on same VLAN. */
1733 mirrors &= mirrors - 1;
1736 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1741 print_dsts(const struct dst *dsts, size_t n)
1743 for (; n--; dsts++) {
1744 printf(">p%"PRIu16, dsts->dp_ifidx);
1745 if (dsts->vlan != OFP_VLAN_NONE) {
1746 printf("v%"PRIu16, dsts->vlan);
1752 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1753 const struct port *in_port, const struct port *out_port,
1754 tag_type *tags, struct odp_actions *actions)
1756 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1758 const struct dst *p;
1761 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1763 cur_vlan = ntohs(flow->dl_vlan);
1764 for (p = dsts; p < &dsts[n_dsts]; p++) {
1765 union odp_action *a;
1766 if (p->vlan != cur_vlan) {
1767 if (p->vlan == OFP_VLAN_NONE) {
1768 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1770 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1771 a->vlan_vid.vlan_vid = htons(p->vlan);
1775 a = odp_actions_add(actions, ODPAT_OUTPUT);
1776 a->output.port = p->dp_ifidx;
1781 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1783 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1784 return (flow->dl_type == htons(ETH_TYPE_ARP)
1785 && eth_addr_is_broadcast(flow->dl_dst)
1786 && packet->size >= sizeof(struct arp_eth_header)
1787 && arp->ar_op == ARP_OP_REQUEST);
1790 /* If the composed actions may be applied to any packet in the given 'flow',
1791 * returns true. Otherwise, the actions should only be applied to 'packet', or
1792 * not at all, if 'packet' was NULL. */
1794 process_flow(struct bridge *br, const flow_t *flow,
1795 const struct ofpbuf *packet, struct odp_actions *actions,
1798 struct iface *in_iface;
1799 struct port *in_port;
1800 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1803 /* Find the interface and port structure for the received packet. */
1804 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1806 /* No interface? Something fishy... */
1807 if (packet != NULL) {
1808 /* Odd. A few possible reasons here:
1810 * - We deleted an interface but there are still a few packets
1811 * queued up from it.
1813 * - Someone externally added an interface (e.g. with "ovs-dpctl
1814 * add-if") that we don't know about.
1816 * - Packet arrived on the local port but the local port is not
1817 * one of our bridge ports.
1819 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1821 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1822 "interface %"PRIu16, br->name, flow->in_port);
1825 /* Return without adding any actions, to drop packets on this flow. */
1828 in_port = in_iface->port;
1830 /* Figure out what VLAN this packet belongs to.
1832 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1833 * belongs to VLAN 0, so we should treat both cases identically. (In the
1834 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1835 * presumably to allow a priority to be specified. In the latter case, the
1836 * packet does not have any 802.1Q header.) */
1837 vlan = ntohs(flow->dl_vlan);
1838 if (vlan == OFP_VLAN_NONE) {
1841 if (in_port->vlan >= 0) {
1843 /* XXX support double tagging? */
1844 if (packet != NULL) {
1845 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1846 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1847 "packet received on port %s configured with "
1848 "implicit VLAN %"PRIu16,
1849 br->name, ntohs(flow->dl_vlan),
1850 in_port->name, in_port->vlan);
1854 vlan = in_port->vlan;
1856 if (!port_includes_vlan(in_port, vlan)) {
1857 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1858 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1859 "packet received on port %s not configured for "
1861 br->name, vlan, in_port->name, vlan);
1866 /* Drop frames for ports that STP wants entirely killed (both for
1867 * forwarding and for learning). Later, after we do learning, we'll drop
1868 * the frames that STP wants to do learning but not forwarding on. */
1869 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1873 /* Drop frames for reserved multicast addresses. */
1874 if (eth_addr_is_reserved(flow->dl_dst)) {
1878 /* Drop frames on ports reserved for mirroring. */
1879 if (in_port->is_mirror_output_port) {
1880 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1881 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1882 "which is reserved exclusively for mirroring",
1883 br->name, in_port->name);
1887 /* Multicast (and broadcast) packets on bonds need special attention, to
1888 * avoid receiving duplicates. */
1889 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1890 *tags |= in_port->active_iface_tag;
1891 if (in_port->active_iface != in_iface->port_ifidx) {
1892 /* Drop all multicast packets on inactive slaves. */
1895 /* Drop all multicast packets for which we have learned a different
1896 * input port, because we probably sent the packet on one slave
1897 * and got it back on the active slave. Broadcast ARP replies are
1898 * an exception to this rule: the host has moved to another
1900 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1901 if (src_idx != -1 && src_idx != in_port->port_idx) {
1903 if (!is_bcast_arp_reply(flow, packet)) {
1907 /* No way to know whether it's an ARP reply, because the
1908 * flow entry doesn't include enough information and we
1909 * don't have a packet. Punt. */
1917 out_port = FLOOD_PORT;
1921 /* Learn source MAC (but don't try to learn from revalidation). */
1923 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1924 vlan, in_port->port_idx);
1926 /* The log messages here could actually be useful in debugging,
1927 * so keep the rate limit relatively high. */
1928 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1930 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1931 "on port %s in VLAN %d",
1932 br->name, ETH_ADDR_ARGS(flow->dl_src),
1933 in_port->name, vlan);
1934 ofproto_revalidate(br->ofproto, rev_tag);
1938 /* Determine output port. */
1939 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1941 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1942 out_port = br->ports[out_port_idx];
1946 /* Don't send packets out their input ports. Don't forward frames that STP
1947 * wants us to discard. */
1948 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1953 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1956 * We send out only a single packet, instead of setting up a flow, if the
1957 * packet is an ARP directed to broadcast that arrived on a bonded
1958 * interface. In such a situation ARP requests and replies must be handled
1959 * differently, but OpenFlow unfortunately can't distinguish them.
1961 return (in_port->n_ifaces < 2
1962 || flow->dl_type != htons(ETH_TYPE_ARP)
1963 || !eth_addr_is_broadcast(flow->dl_dst));
1966 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1969 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1970 const struct ofp_phy_port *opp,
1973 struct bridge *br = br_;
1974 struct iface *iface;
1977 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1983 if (reason == OFPPR_DELETE) {
1984 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1985 br->name, iface->name);
1986 iface_destroy(iface);
1987 if (!port->n_ifaces) {
1988 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1989 br->name, port->name);
1995 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1996 if (port->n_ifaces > 1) {
1997 bool up = !(opp->state & OFPPS_LINK_DOWN);
1998 bond_link_status_update(iface, up);
1999 port_update_bond_compat(port);
2005 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2006 struct odp_actions *actions, tag_type *tags, void *br_)
2008 struct bridge *br = br_;
2011 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
2012 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
2013 brstp_receive(br, flow, payload);
2018 COVERAGE_INC(bridge_process_flow);
2019 return process_flow(br, flow, packet, actions, tags);
2023 bridge_account_flow_ofhook_cb(const flow_t *flow,
2024 const union odp_action *actions,
2025 size_t n_actions, unsigned long long int n_bytes,
2028 struct bridge *br = br_;
2029 const union odp_action *a;
2031 if (!br->has_bonded_ports) {
2035 for (a = actions; a < &actions[n_actions]; a++) {
2036 if (a->type == ODPAT_OUTPUT) {
2037 struct port *port = port_from_dp_ifidx(br, a->output.port);
2038 if (port && port->n_ifaces >= 2) {
2039 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2040 e->tx_bytes += n_bytes;
2047 bridge_account_checkpoint_ofhook_cb(void *br_)
2049 struct bridge *br = br_;
2052 if (!br->has_bonded_ports) {
2056 /* The current ofproto implementation calls this callback at least once a
2057 * second, so this timer implementation is sufficient. */
2058 if (time_msec() < br->bond_next_rebalance) {
2061 br->bond_next_rebalance = time_msec() + 10000;
2063 for (i = 0; i < br->n_ports; i++) {
2064 struct port *port = br->ports[i];
2065 if (port->n_ifaces > 1) {
2066 bond_rebalance_port(port);
2071 static struct ofhooks bridge_ofhooks = {
2072 bridge_port_changed_ofhook_cb,
2073 bridge_normal_ofhook_cb,
2074 bridge_account_flow_ofhook_cb,
2075 bridge_account_checkpoint_ofhook_cb,
2078 /* Bonding functions. */
2080 /* Statistics for a single interface on a bonded port, used for load-based
2081 * bond rebalancing. */
2082 struct slave_balance {
2083 struct iface *iface; /* The interface. */
2084 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2086 /* All the "bond_entry"s that are assigned to this interface, in order of
2087 * increasing tx_bytes. */
2088 struct bond_entry **hashes;
2092 /* Sorts pointers to pointers to bond_entries in ascending order by the
2093 * interface to which they are assigned, and within a single interface in
2094 * ascending order of bytes transmitted. */
2096 compare_bond_entries(const void *a_, const void *b_)
2098 const struct bond_entry *const *ap = a_;
2099 const struct bond_entry *const *bp = b_;
2100 const struct bond_entry *a = *ap;
2101 const struct bond_entry *b = *bp;
2102 if (a->iface_idx != b->iface_idx) {
2103 return a->iface_idx > b->iface_idx ? 1 : -1;
2104 } else if (a->tx_bytes != b->tx_bytes) {
2105 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2111 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2112 * *descending* order by number of bytes transmitted. */
2114 compare_slave_balance(const void *a_, const void *b_)
2116 const struct slave_balance *a = a_;
2117 const struct slave_balance *b = b_;
2118 if (a->iface->enabled != b->iface->enabled) {
2119 return a->iface->enabled ? -1 : 1;
2120 } else if (a->tx_bytes != b->tx_bytes) {
2121 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2128 swap_bals(struct slave_balance *a, struct slave_balance *b)
2130 struct slave_balance tmp = *a;
2135 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2136 * given that 'p' (and only 'p') might be in the wrong location.
2138 * This function invalidates 'p', since it might now be in a different memory
2141 resort_bals(struct slave_balance *p,
2142 struct slave_balance bals[], size_t n_bals)
2145 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2146 swap_bals(p, p - 1);
2148 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2149 swap_bals(p, p + 1);
2155 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2157 if (VLOG_IS_DBG_ENABLED()) {
2158 struct ds ds = DS_EMPTY_INITIALIZER;
2159 const struct slave_balance *b;
2161 for (b = bals; b < bals + n_bals; b++) {
2165 ds_put_char(&ds, ',');
2167 ds_put_format(&ds, " %s %"PRIu64"kB",
2168 b->iface->name, b->tx_bytes / 1024);
2170 if (!b->iface->enabled) {
2171 ds_put_cstr(&ds, " (disabled)");
2173 if (b->n_hashes > 0) {
2174 ds_put_cstr(&ds, " (");
2175 for (i = 0; i < b->n_hashes; i++) {
2176 const struct bond_entry *e = b->hashes[i];
2178 ds_put_cstr(&ds, " + ");
2180 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2181 e - port->bond_hash, e->tx_bytes / 1024);
2183 ds_put_cstr(&ds, ")");
2186 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2191 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2193 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2194 struct bond_entry *hash)
2196 struct port *port = from->iface->port;
2197 uint64_t delta = hash->tx_bytes;
2199 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2200 "from %s to %s (now carrying %"PRIu64"kB and "
2201 "%"PRIu64"kB load, respectively)",
2202 port->name, delta / 1024, hash - port->bond_hash,
2203 from->iface->name, to->iface->name,
2204 (from->tx_bytes - delta) / 1024,
2205 (to->tx_bytes + delta) / 1024);
2207 /* Delete element from from->hashes.
2209 * We don't bother to add the element to to->hashes because not only would
2210 * it require more work, the only purpose it would be to allow that hash to
2211 * be migrated to another slave in this rebalancing run, and there is no
2212 * point in doing that. */
2213 if (from->hashes[0] == hash) {
2216 int i = hash - from->hashes[0];
2217 memmove(from->hashes + i, from->hashes + i + 1,
2218 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2222 /* Shift load away from 'from' to 'to'. */
2223 from->tx_bytes -= delta;
2224 to->tx_bytes += delta;
2226 /* Arrange for flows to be revalidated. */
2227 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2228 hash->iface_idx = to->iface->port_ifidx;
2229 hash->iface_tag = tag_create_random();
2233 bond_rebalance_port(struct port *port)
2235 struct slave_balance bals[DP_MAX_PORTS];
2237 struct bond_entry *hashes[BOND_MASK + 1];
2238 struct slave_balance *b, *from, *to;
2239 struct bond_entry *e;
2242 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2243 * descending order of tx_bytes, so that bals[0] represents the most
2244 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2247 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2248 * array for each slave_balance structure, we sort our local array of
2249 * hashes in order by slave, so that all of the hashes for a given slave
2250 * become contiguous in memory, and then we point each 'hashes' members of
2251 * a slave_balance structure to the start of a contiguous group. */
2252 n_bals = port->n_ifaces;
2253 for (b = bals; b < &bals[n_bals]; b++) {
2254 b->iface = port->ifaces[b - bals];
2259 for (i = 0; i <= BOND_MASK; i++) {
2260 hashes[i] = &port->bond_hash[i];
2262 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2263 for (i = 0; i <= BOND_MASK; i++) {
2265 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2266 b = &bals[e->iface_idx];
2267 b->tx_bytes += e->tx_bytes;
2269 b->hashes = &hashes[i];
2274 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2275 log_bals(bals, n_bals, port);
2277 /* Discard slaves that aren't enabled (which were sorted to the back of the
2278 * array earlier). */
2279 while (!bals[n_bals - 1].iface->enabled) {
2286 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2287 to = &bals[n_bals - 1];
2288 for (from = bals; from < to; ) {
2289 uint64_t overload = from->tx_bytes - to->tx_bytes;
2290 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2291 /* The extra load on 'from' (and all less-loaded slaves), compared
2292 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2293 * it is less than ~1Mbps. No point in rebalancing. */
2295 } else if (from->n_hashes == 1) {
2296 /* 'from' only carries a single MAC hash, so we can't shift any
2297 * load away from it, even though we want to. */
2300 /* 'from' is carrying significantly more load than 'to', and that
2301 * load is split across at least two different hashes. Pick a hash
2302 * to migrate to 'to' (the least-loaded slave), given that doing so
2303 * must not cause 'to''s load to exceed 'from''s load.
2305 * The sort order we use means that we prefer to shift away the
2306 * smallest hashes instead of the biggest ones. There is little
2307 * reason behind this decision; we could use the opposite sort
2308 * order to shift away big hashes ahead of small ones. */
2311 for (i = 0; i < from->n_hashes; i++) {
2312 uint64_t delta = from->hashes[i]->tx_bytes;
2313 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2317 if (i < from->n_hashes) {
2318 bond_shift_load(from, to, from->hashes[i]);
2320 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2321 * point to different slave_balance structures. It is only
2322 * valid to do these two operations in a row at all because we
2323 * know that 'from' will not move past 'to' and vice versa. */
2324 resort_bals(from, bals, n_bals);
2325 resort_bals(to, bals, n_bals);
2329 port->bond_compat_is_stale = true;
2333 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2334 * historical data to decay to <1% in 7 rebalancing runs. */
2335 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2341 bond_send_learning_packets(struct port *port)
2343 struct bridge *br = port->bridge;
2344 struct mac_entry *e;
2345 struct ofpbuf packet;
2346 int error, n_packets, n_errors;
2348 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2352 ofpbuf_init(&packet, 128);
2353 error = n_packets = n_errors = 0;
2354 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2355 union ofp_action actions[2], *a;
2361 if (e->port == port->port_idx
2362 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2366 /* Compose actions. */
2367 memset(actions, 0, sizeof actions);
2370 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2371 a->vlan_vid.len = htons(sizeof *a);
2372 a->vlan_vid.vlan_vid = htons(e->vlan);
2375 a->output.type = htons(OFPAT_OUTPUT);
2376 a->output.len = htons(sizeof *a);
2377 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2382 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2384 flow_extract(&packet, ODPP_NONE, &flow);
2385 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2392 ofpbuf_uninit(&packet);
2395 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2396 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2397 "packets, last error was: %s",
2398 port->name, n_errors, n_packets, strerror(error));
2400 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2401 port->name, n_packets);
2405 /* Bonding unixctl user interface functions. */
2408 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2410 struct ds ds = DS_EMPTY_INITIALIZER;
2411 const struct bridge *br;
2413 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2415 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2418 for (i = 0; i < br->n_ports; i++) {
2419 const struct port *port = br->ports[i];
2420 if (port->n_ifaces > 1) {
2423 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2424 for (j = 0; j < port->n_ifaces; j++) {
2425 const struct iface *iface = port->ifaces[j];
2427 ds_put_cstr(&ds, ", ");
2429 ds_put_cstr(&ds, iface->name);
2431 ds_put_char(&ds, '\n');
2435 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2439 static struct port *
2440 bond_find(const char *name)
2442 const struct bridge *br;
2444 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2447 for (i = 0; i < br->n_ports; i++) {
2448 struct port *port = br->ports[i];
2449 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2458 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2460 struct ds ds = DS_EMPTY_INITIALIZER;
2461 const struct port *port;
2464 port = bond_find(args);
2466 unixctl_command_reply(conn, 501, "no such bond");
2470 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2471 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2472 ds_put_format(&ds, "next rebalance: %lld ms\n",
2473 port->bridge->bond_next_rebalance - time_msec());
2474 for (j = 0; j < port->n_ifaces; j++) {
2475 const struct iface *iface = port->ifaces[j];
2476 struct bond_entry *be;
2479 ds_put_format(&ds, "slave %s: %s\n",
2480 iface->name, iface->enabled ? "enabled" : "disabled");
2481 if (j == port->active_iface) {
2482 ds_put_cstr(&ds, "\tactive slave\n");
2484 if (iface->delay_expires != LLONG_MAX) {
2485 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2486 iface->enabled ? "downdelay" : "updelay",
2487 iface->delay_expires - time_msec());
2491 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2492 int hash = be - port->bond_hash;
2493 struct mac_entry *me;
2495 if (be->iface_idx != j) {
2499 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2500 hash, be->tx_bytes / 1024);
2503 if (!port->bridge->ml) {
2507 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2508 &port->bridge->ml->lrus) {
2511 if (bond_hash(me->mac) == hash
2512 && me->port != port->port_idx
2513 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2514 && dp_ifidx == iface->dp_ifidx)
2516 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2517 ETH_ADDR_ARGS(me->mac));
2522 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2527 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2529 char *args = (char *) args_;
2530 char *save_ptr = NULL;
2531 char *bond_s, *hash_s, *slave_s;
2532 uint8_t mac[ETH_ADDR_LEN];
2534 struct iface *iface;
2535 struct bond_entry *entry;
2538 bond_s = strtok_r(args, " ", &save_ptr);
2539 hash_s = strtok_r(NULL, " ", &save_ptr);
2540 slave_s = strtok_r(NULL, " ", &save_ptr);
2542 unixctl_command_reply(conn, 501,
2543 "usage: bond/migrate BOND HASH SLAVE");
2547 port = bond_find(bond_s);
2549 unixctl_command_reply(conn, 501, "no such bond");
2553 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2554 == ETH_ADDR_SCAN_COUNT) {
2555 hash = bond_hash(mac);
2556 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2557 hash = atoi(hash_s) & BOND_MASK;
2559 unixctl_command_reply(conn, 501, "bad hash");
2563 iface = port_lookup_iface(port, slave_s);
2565 unixctl_command_reply(conn, 501, "no such slave");
2569 if (!iface->enabled) {
2570 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2574 entry = &port->bond_hash[hash];
2575 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2576 entry->iface_idx = iface->port_ifidx;
2577 entry->iface_tag = tag_create_random();
2578 port->bond_compat_is_stale = true;
2579 unixctl_command_reply(conn, 200, "migrated");
2583 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2585 char *args = (char *) args_;
2586 char *save_ptr = NULL;
2587 char *bond_s, *slave_s;
2589 struct iface *iface;
2591 bond_s = strtok_r(args, " ", &save_ptr);
2592 slave_s = strtok_r(NULL, " ", &save_ptr);
2594 unixctl_command_reply(conn, 501,
2595 "usage: bond/set-active-slave BOND SLAVE");
2599 port = bond_find(bond_s);
2601 unixctl_command_reply(conn, 501, "no such bond");
2605 iface = port_lookup_iface(port, slave_s);
2607 unixctl_command_reply(conn, 501, "no such slave");
2611 if (!iface->enabled) {
2612 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2616 if (port->active_iface != iface->port_ifidx) {
2617 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2618 port->active_iface = iface->port_ifidx;
2619 port->active_iface_tag = tag_create_random();
2620 VLOG_INFO("port %s: active interface is now %s",
2621 port->name, iface->name);
2622 bond_send_learning_packets(port);
2623 unixctl_command_reply(conn, 200, "done");
2625 unixctl_command_reply(conn, 200, "no change");
2630 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2632 char *args = (char *) args_;
2633 char *save_ptr = NULL;
2634 char *bond_s, *slave_s;
2636 struct iface *iface;
2638 bond_s = strtok_r(args, " ", &save_ptr);
2639 slave_s = strtok_r(NULL, " ", &save_ptr);
2641 unixctl_command_reply(conn, 501,
2642 "usage: bond/enable/disable-slave BOND SLAVE");
2646 port = bond_find(bond_s);
2648 unixctl_command_reply(conn, 501, "no such bond");
2652 iface = port_lookup_iface(port, slave_s);
2654 unixctl_command_reply(conn, 501, "no such slave");
2658 bond_enable_slave(iface, enable);
2659 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2663 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2665 enable_slave(conn, args, true);
2669 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2671 enable_slave(conn, args, false);
2677 unixctl_command_register("bond/list", bond_unixctl_list);
2678 unixctl_command_register("bond/show", bond_unixctl_show);
2679 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2680 unixctl_command_register("bond/set-active-slave",
2681 bond_unixctl_set_active_slave);
2682 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2683 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2686 /* Port functions. */
2689 port_create(struct bridge *br, const char *name)
2693 port = xcalloc(1, sizeof *port);
2695 port->port_idx = br->n_ports;
2697 port->trunks = NULL;
2698 port->name = xstrdup(name);
2699 port->active_iface = -1;
2700 port->stp_state = STP_DISABLED;
2701 port->stp_state_tag = 0;
2703 if (br->n_ports >= br->allocated_ports) {
2704 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2707 br->ports[br->n_ports++] = port;
2709 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2714 port_reconfigure(struct port *port)
2716 bool bonded = cfg_has_section("bonding.%s", port->name);
2717 struct svec old_ifaces, new_ifaces;
2718 unsigned long *trunks;
2722 /* Collect old and new interfaces. */
2723 svec_init(&old_ifaces);
2724 svec_init(&new_ifaces);
2725 for (i = 0; i < port->n_ifaces; i++) {
2726 svec_add(&old_ifaces, port->ifaces[i]->name);
2728 svec_sort(&old_ifaces);
2730 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2731 if (!new_ifaces.n) {
2732 VLOG_ERR("port %s: no interfaces specified for bonded port",
2734 } else if (new_ifaces.n == 1) {
2735 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2739 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2740 if (port->updelay < 0) {
2743 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2744 if (port->downdelay < 0) {
2745 port->downdelay = 0;
2748 svec_init(&new_ifaces);
2749 svec_add(&new_ifaces, port->name);
2752 /* Get rid of deleted interfaces and add new interfaces. */
2753 for (i = 0; i < port->n_ifaces; i++) {
2754 struct iface *iface = port->ifaces[i];
2755 if (!svec_contains(&new_ifaces, iface->name)) {
2756 iface_destroy(iface);
2761 for (i = 0; i < new_ifaces.n; i++) {
2762 const char *name = new_ifaces.names[i];
2763 if (!svec_contains(&old_ifaces, name)) {
2764 iface_create(port, name);
2770 if (cfg_has("vlan.%s.tag", port->name)) {
2772 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2773 if (vlan >= 0 && vlan <= 4095) {
2774 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2777 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2778 * they even work as-is. But they have not been tested. */
2779 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2783 if (port->vlan != vlan) {
2785 bridge_flush(port->bridge);
2788 /* Get trunked VLANs. */
2791 size_t n_trunks, n_errors;
2794 trunks = bitmap_allocate(4096);
2795 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2797 for (i = 0; i < n_trunks; i++) {
2798 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2800 bitmap_set1(trunks, trunk);
2806 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2807 port->name, n_trunks);
2809 if (n_errors == n_trunks) {
2811 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2814 bitmap_set_multiple(trunks, 0, 4096, 1);
2817 if (cfg_has("vlan.%s.trunks", port->name)) {
2818 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2819 port->name, port->name);
2823 ? port->trunks != NULL
2824 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2825 bridge_flush(port->bridge);
2827 bitmap_free(port->trunks);
2828 port->trunks = trunks;
2830 svec_destroy(&old_ifaces);
2831 svec_destroy(&new_ifaces);
2835 port_destroy(struct port *port)
2838 struct bridge *br = port->bridge;
2842 proc_net_compat_update_vlan(port->name, NULL, 0);
2843 proc_net_compat_update_bond(port->name, NULL);
2845 for (i = 0; i < MAX_MIRRORS; i++) {
2846 struct mirror *m = br->mirrors[i];
2847 if (m && m->out_port == port) {
2852 while (port->n_ifaces > 0) {
2853 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2856 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2857 del->port_idx = port->port_idx;
2860 bitmap_free(port->trunks);
2867 static struct port *
2868 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2870 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2871 return iface ? iface->port : NULL;
2874 static struct port *
2875 port_lookup(const struct bridge *br, const char *name)
2879 for (i = 0; i < br->n_ports; i++) {
2880 struct port *port = br->ports[i];
2881 if (!strcmp(port->name, name)) {
2888 static struct iface *
2889 port_lookup_iface(const struct port *port, const char *name)
2893 for (j = 0; j < port->n_ifaces; j++) {
2894 struct iface *iface = port->ifaces[j];
2895 if (!strcmp(iface->name, name)) {
2903 port_update_bonding(struct port *port)
2905 if (port->n_ifaces < 2) {
2906 /* Not a bonded port. */
2907 if (port->bond_hash) {
2908 free(port->bond_hash);
2909 port->bond_hash = NULL;
2910 port->bond_compat_is_stale = true;
2913 if (!port->bond_hash) {
2916 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2917 for (i = 0; i <= BOND_MASK; i++) {
2918 struct bond_entry *e = &port->bond_hash[i];
2922 port->no_ifaces_tag = tag_create_random();
2923 bond_choose_active_iface(port);
2925 port->bond_compat_is_stale = true;
2930 port_update_bond_compat(struct port *port)
2932 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
2933 struct compat_bond bond;
2936 if (port->n_ifaces < 2) {
2937 proc_net_compat_update_bond(port->name, NULL);
2942 bond.updelay = port->updelay;
2943 bond.downdelay = port->downdelay;
2946 bond.hashes = compat_hashes;
2947 if (port->bond_hash) {
2948 const struct bond_entry *e;
2949 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
2950 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2951 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
2952 cbh->hash = e - port->bond_hash;
2953 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
2958 bond.n_slaves = port->n_ifaces;
2959 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2960 for (i = 0; i < port->n_ifaces; i++) {
2961 struct iface *iface = port->ifaces[i];
2962 struct compat_bond_slave *slave = &bond.slaves[i];
2963 slave->name = iface->name;
2965 /* We need to make the same determination as the Linux bonding
2966 * code to determine whether a slave should be consider "up".
2967 * The Linux function bond_miimon_inspect() supports four
2968 * BOND_LINK_* states:
2970 * - BOND_LINK_UP: carrier detected, updelay has passed.
2971 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
2972 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
2973 * - BOND_LINK_BACK: carrier detected, updelay in progress.
2975 * The function bond_info_show_slave() only considers BOND_LINK_UP
2976 * to be "up" and anything else to be "down".
2978 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
2982 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2985 if (cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
2986 struct netdev *bond_netdev;
2988 if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) {
2990 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
2992 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
2994 netdev_close(bond_netdev);
2998 proc_net_compat_update_bond(port->name, &bond);
3003 port_update_vlan_compat(struct port *port)
3005 struct bridge *br = port->bridge;
3006 char *vlandev_name = NULL;
3008 if (port->vlan > 0) {
3009 /* Figure out the name that the VLAN device should actually have, if it
3010 * existed. This takes some work because the VLAN device would not
3011 * have port->name in its name; rather, it would have the trunk port's
3012 * name, and 'port' would be attached to a bridge that also had the
3013 * VLAN device one of its ports. So we need to find a trunk port that
3014 * includes port->vlan.
3016 * There might be more than one candidate. This doesn't happen on
3017 * XenServer, so if it happens we just pick the first choice in
3018 * alphabetical order instead of creating multiple VLAN devices. */
3020 for (i = 0; i < br->n_ports; i++) {
3021 struct port *p = br->ports[i];
3022 if (port_trunks_vlan(p, port->vlan)
3024 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3026 const uint8_t *ea = p->ifaces[0]->mac;
3027 if (!eth_addr_is_multicast(ea) &&
3028 !eth_addr_is_reserved(ea) &&
3029 !eth_addr_is_zero(ea)) {
3030 vlandev_name = p->name;
3035 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3038 /* Interface functions. */
3041 iface_create(struct port *port, const char *name)
3043 struct iface *iface;
3045 iface = xcalloc(1, sizeof *iface);
3047 iface->port_ifidx = port->n_ifaces;
3048 iface->name = xstrdup(name);
3049 iface->dp_ifidx = -1;
3050 iface->tag = tag_create_random();
3051 iface->delay_expires = LLONG_MAX;
3053 if (!cfg_get_bool(0, "iface.%s.internal", iface->name)) {
3054 netdev_nodev_get_etheraddr(name, iface->mac);
3055 netdev_nodev_get_carrier(name, &iface->enabled);
3057 /* Internal interfaces are created later by the call to dpif_port_add()
3058 * in bridge_reconfigure(). Until then, we can't obtain any
3059 * information about them. (There's no real value in doing so, anyway,
3060 * because the 'mac' and 'enabled' values are only used for interfaces
3061 * that are bond slaves, and it doesn't normally make sense to bond an
3062 * internal interface.) */
3065 if (port->n_ifaces >= port->allocated_ifaces) {
3066 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3067 sizeof *port->ifaces);
3069 port->ifaces[port->n_ifaces++] = iface;
3070 if (port->n_ifaces > 1) {
3071 port->bridge->has_bonded_ports = true;
3074 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3076 port_update_bonding(port);
3077 bridge_flush(port->bridge);
3081 iface_destroy(struct iface *iface)
3084 struct port *port = iface->port;
3085 struct bridge *br = port->bridge;
3086 bool del_active = port->active_iface == iface->port_ifidx;
3089 if (iface->dp_ifidx >= 0) {
3090 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3093 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3094 del->port_ifidx = iface->port_ifidx;
3100 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3101 bond_choose_active_iface(port);
3102 bond_send_learning_packets(port);
3105 port_update_bonding(port);
3106 bridge_flush(port->bridge);
3110 static struct iface *
3111 iface_lookup(const struct bridge *br, const char *name)
3115 for (i = 0; i < br->n_ports; i++) {
3116 struct port *port = br->ports[i];
3117 for (j = 0; j < port->n_ifaces; j++) {
3118 struct iface *iface = port->ifaces[j];
3119 if (!strcmp(iface->name, name)) {
3127 static struct iface *
3128 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3130 return port_array_get(&br->ifaces, dp_ifidx);
3133 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3134 * 'br', that is, an interface that is entirely simulated within the datapath.
3135 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3136 * interfaces are created by setting "iface.<iface>.internal = true".
3138 * In addition, we have a kluge-y feature that creates an internal port with
3139 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3140 * This feature needs to go away in the long term. Until then, this is one
3141 * reason why this function takes a name instead of a struct iface: the fake
3142 * interfaces created this way do not have a struct iface. */
3144 iface_is_internal(const struct bridge *br, const char *iface)
3146 if (!strcmp(iface, br->name)
3147 || cfg_get_bool(0, "iface.%s.internal", iface)) {
3151 if (cfg_get_bool(0, "bonding.%s.fake-iface", iface)) {
3152 struct port *port = port_lookup(br, iface);
3153 if (port && port->n_ifaces > 1) {
3161 /* Set Ethernet address of 'iface', if one is specified in the configuration
3164 iface_set_mac(struct iface *iface)
3166 uint64_t mac = cfg_get_mac(0, "iface.%s.mac", iface->name);
3168 static uint8_t ea[ETH_ADDR_LEN];
3170 eth_addr_from_uint64(mac, ea);
3171 if (eth_addr_is_multicast(ea)) {
3172 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3174 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3175 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3176 iface->name, iface->name);
3178 int error = netdev_nodev_set_etheraddr(iface->name, ea);
3180 VLOG_ERR("interface %s: setting MAC failed (%s)",
3181 iface->name, strerror(error));
3187 /* Port mirroring. */
3190 mirror_reconfigure(struct bridge *br)
3192 struct svec old_mirrors, new_mirrors;
3195 /* Collect old and new mirrors. */
3196 svec_init(&old_mirrors);
3197 svec_init(&new_mirrors);
3198 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3199 for (i = 0; i < MAX_MIRRORS; i++) {
3200 if (br->mirrors[i]) {
3201 svec_add(&old_mirrors, br->mirrors[i]->name);
3205 /* Get rid of deleted mirrors and add new mirrors. */
3206 svec_sort(&old_mirrors);
3207 assert(svec_is_unique(&old_mirrors));
3208 svec_sort(&new_mirrors);
3209 assert(svec_is_unique(&new_mirrors));
3210 for (i = 0; i < MAX_MIRRORS; i++) {
3211 struct mirror *m = br->mirrors[i];
3212 if (m && !svec_contains(&new_mirrors, m->name)) {
3216 for (i = 0; i < new_mirrors.n; i++) {
3217 const char *name = new_mirrors.names[i];
3218 if (!svec_contains(&old_mirrors, name)) {
3219 mirror_create(br, name);
3222 svec_destroy(&old_mirrors);
3223 svec_destroy(&new_mirrors);
3225 /* Reconfigure all mirrors. */
3226 for (i = 0; i < MAX_MIRRORS; i++) {
3227 if (br->mirrors[i]) {
3228 mirror_reconfigure_one(br->mirrors[i]);
3232 /* Update port reserved status. */
3233 for (i = 0; i < br->n_ports; i++) {
3234 br->ports[i]->is_mirror_output_port = false;
3236 for (i = 0; i < MAX_MIRRORS; i++) {
3237 struct mirror *m = br->mirrors[i];
3238 if (m && m->out_port) {
3239 m->out_port->is_mirror_output_port = true;
3245 mirror_create(struct bridge *br, const char *name)
3250 for (i = 0; ; i++) {
3251 if (i >= MAX_MIRRORS) {
3252 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3253 "cannot create %s", br->name, MAX_MIRRORS, name);
3256 if (!br->mirrors[i]) {
3261 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3264 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3267 m->name = xstrdup(name);
3268 svec_init(&m->src_ports);
3269 svec_init(&m->dst_ports);
3277 mirror_destroy(struct mirror *m)
3280 struct bridge *br = m->bridge;
3283 for (i = 0; i < br->n_ports; i++) {
3284 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3285 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3288 svec_destroy(&m->src_ports);
3289 svec_destroy(&m->dst_ports);
3292 m->bridge->mirrors[m->idx] = NULL;
3300 prune_ports(struct mirror *m, struct svec *ports)
3305 svec_sort_unique(ports);
3308 for (i = 0; i < ports->n; i++) {
3309 const char *name = ports->names[i];
3310 if (port_lookup(m->bridge, name)) {
3311 svec_add(&tmp, name);
3313 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3314 m->bridge->name, m->name, name);
3317 svec_swap(ports, &tmp);
3322 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3326 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3327 * order won't give us numeric sort order. But that's good enough for what
3328 * we need right now. */
3329 svec_sort_unique(vlan_strings);
3331 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3333 for (i = 0; i < vlan_strings->n; i++) {
3334 const char *name = vlan_strings->names[i];
3336 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3337 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3338 m->bridge->name, m->name, name);
3340 (*vlans)[n_vlans++] = vlan;
3347 vlan_is_mirrored(const struct mirror *m, int vlan)
3351 for (i = 0; i < m->n_vlans; i++) {
3352 if (m->vlans[i] == vlan) {
3360 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3364 for (i = 0; i < m->n_vlans; i++) {
3365 if (port_trunks_vlan(p, m->vlans[i])) {
3373 mirror_reconfigure_one(struct mirror *m)
3375 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3376 struct svec src_ports, dst_ports, ports;
3377 struct svec vlan_strings;
3378 mirror_mask_t mirror_bit;
3379 const char *out_port_name;
3380 struct port *out_port;
3385 bool mirror_all_ports;
3386 bool any_ports_specified;
3388 /* Get output port. */
3389 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3390 m->bridge->name, m->name);
3391 if (out_port_name) {
3392 out_port = port_lookup(m->bridge, out_port_name);
3394 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3395 "named %s", pfx, m->bridge->name, out_port_name);
3402 if (cfg_has("%s.output.vlan", pfx)) {
3403 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3404 "ignoring %s.output.vlan", pfx, pfx, pfx);
3406 } else if (cfg_has("%s.output.vlan", pfx)) {
3408 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3410 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3411 "but exactly one is required; disabling port mirror %s",
3412 pfx, pfx, pfx, pfx);
3418 /* Get all the ports, and drop duplicates and ports that don't exist. */
3419 svec_init(&src_ports);
3420 svec_init(&dst_ports);
3422 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3423 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3424 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3425 any_ports_specified = src_ports.n || dst_ports.n || ports.n;
3426 svec_append(&src_ports, &ports);
3427 svec_append(&dst_ports, &ports);
3428 svec_destroy(&ports);
3429 prune_ports(m, &src_ports);
3430 prune_ports(m, &dst_ports);
3431 if (any_ports_specified && !src_ports.n && !dst_ports.n) {
3432 VLOG_ERR("%s: none of the specified ports exist; "
3433 "disabling port mirror %s", pfx, pfx);
3438 /* Get all the vlans, and drop duplicate and invalid vlans. */
3439 svec_init(&vlan_strings);
3440 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3441 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3442 svec_destroy(&vlan_strings);
3444 /* Update mirror data. */
3445 if (!svec_equal(&m->src_ports, &src_ports)
3446 || !svec_equal(&m->dst_ports, &dst_ports)
3447 || m->n_vlans != n_vlans
3448 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3449 || m->out_port != out_port
3450 || m->out_vlan != out_vlan) {
3451 bridge_flush(m->bridge);
3453 svec_swap(&m->src_ports, &src_ports);
3454 svec_swap(&m->dst_ports, &dst_ports);
3457 m->n_vlans = n_vlans;
3458 m->out_port = out_port;
3459 m->out_vlan = out_vlan;
3461 /* If no selection criteria have been given, mirror for all ports. */
3462 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3465 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3466 for (i = 0; i < m->bridge->n_ports; i++) {
3467 struct port *port = m->bridge->ports[i];
3469 if (mirror_all_ports
3470 || svec_contains(&m->src_ports, port->name)
3473 ? port_trunks_any_mirrored_vlan(m, port)
3474 : vlan_is_mirrored(m, port->vlan)))) {
3475 port->src_mirrors |= mirror_bit;
3477 port->src_mirrors &= ~mirror_bit;
3480 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3481 port->dst_mirrors |= mirror_bit;
3483 port->dst_mirrors &= ~mirror_bit;
3489 svec_destroy(&src_ports);
3490 svec_destroy(&dst_ports);
3494 /* Spanning tree protocol. */
3496 static void brstp_update_port_state(struct port *);
3499 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3501 struct bridge *br = br_;
3502 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3503 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3505 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3507 } else if (eth_addr_is_zero(iface->mac)) {
3508 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3511 union ofp_action action;
3512 struct eth_header *eth = pkt->l2;
3515 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3517 memset(&action, 0, sizeof action);
3518 action.type = htons(OFPAT_OUTPUT);
3519 action.output.len = htons(sizeof action);
3520 action.output.port = htons(port_no);
3522 flow_extract(pkt, ODPP_NONE, &flow);
3523 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3529 brstp_reconfigure(struct bridge *br)
3533 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3535 stp_destroy(br->stp);
3541 uint64_t bridge_address, bridge_id;
3542 int bridge_priority;
3544 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3545 if (!bridge_address) {
3547 bridge_address = (stp_get_bridge_id(br->stp)
3548 & ((UINT64_C(1) << 48) - 1));
3550 uint8_t mac[ETH_ADDR_LEN];
3551 eth_addr_random(mac);
3552 bridge_address = eth_addr_to_uint64(mac);
3556 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3558 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3560 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3563 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3565 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3566 br->stp_last_tick = time_msec();
3569 if (bridge_id != stp_get_bridge_id(br->stp)) {
3570 stp_set_bridge_id(br->stp, bridge_id);
3575 for (i = 0; i < br->n_ports; i++) {
3576 struct port *p = br->ports[i];
3578 struct stp_port *sp;
3579 int path_cost, priority;
3585 dp_ifidx = p->ifaces[0]->dp_ifidx;
3586 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3590 sp = stp_get_port(br->stp, dp_ifidx);
3591 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3592 "stp.%s.port.%s.enabled",
3594 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3595 br->name, p->name));
3596 if (p->is_mirror_output_port) {
3599 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3600 bridge_flush(br); /* Might not be necessary. */
3602 stp_port_enable(sp);
3604 stp_port_disable(sp);
3608 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3610 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3612 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3613 "stp.%s.port.%s.priority",
3615 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3617 : STP_DEFAULT_PORT_PRIORITY);
3618 stp_port_set_priority(sp, priority);
3621 brstp_adjust_timers(br);
3623 for (i = 0; i < br->n_ports; i++) {
3624 brstp_update_port_state(br->ports[i]);
3629 brstp_update_port_state(struct port *p)
3631 struct bridge *br = p->bridge;
3632 enum stp_state state;
3634 /* Figure out new state. */
3635 state = STP_DISABLED;
3636 if (br->stp && p->n_ifaces > 0) {
3637 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3638 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3639 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3644 if (p->stp_state != state) {
3645 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3646 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3647 p->name, stp_state_name(p->stp_state),
3648 stp_state_name(state));
3649 if (p->stp_state == STP_DISABLED) {
3652 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3654 p->stp_state = state;
3655 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3656 : tag_create_random());
3661 brstp_adjust_timers(struct bridge *br)
3663 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3664 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3665 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3667 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3668 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3669 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3673 brstp_run(struct bridge *br)
3676 long long int now = time_msec();
3677 long long int elapsed = now - br->stp_last_tick;
3678 struct stp_port *sp;
3681 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3682 br->stp_last_tick = now;
3684 while (stp_get_changed_port(br->stp, &sp)) {
3685 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3687 brstp_update_port_state(p);
3694 brstp_wait(struct bridge *br)
3697 poll_timer_wait(1000);