1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
47 #include "poll-loop.h"
48 #include "port-array.h"
49 #include "proc-net-compat.h"
51 #include "secchan/ofproto.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id;
74 struct port *port; /* Containing port. */
75 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 int dp_ifidx; /* Index within kernel datapath. */
80 uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */
82 tag_type tag; /* Tag associated with this interface. */
83 bool enabled; /* May be chosen for flows? */
84 long long delay_expires; /* Time after which 'enabled' may change. */
87 #define BOND_MASK 0xff
89 int iface_idx; /* Index of assigned iface, or -1 if none. */
90 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
91 tag_type iface_tag; /* Tag associated with iface_idx. */
94 #define MAX_MIRRORS 32
95 typedef uint32_t mirror_mask_t;
96 #define MIRROR_MASK_C(X) UINT32_C(X)
97 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
99 struct bridge *bridge;
103 /* Selection criteria. */
104 struct svec src_ports;
105 struct svec dst_ports;
110 struct port *out_port;
114 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
116 struct bridge *bridge;
118 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
119 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
122 /* An ordinary bridge port has 1 interface.
123 * A bridge port for bonding has at least 2 interfaces. */
124 struct iface **ifaces;
125 size_t n_ifaces, allocated_ifaces;
128 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
129 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
130 tag_type active_iface_tag; /* Tag for bcast flows. */
131 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
132 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
133 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
135 /* Port mirroring info. */
136 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
137 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
138 bool is_mirror_output_port; /* Does port mirroring send frames here? */
140 /* Spanning tree info. */
141 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
142 tag_type stp_state_tag; /* Tag for STP state change. */
145 #define DP_MAX_PORTS 255
147 struct list node; /* Node in global list of bridges. */
148 char *name; /* User-specified arbitrary name. */
149 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
150 bool sent_config_request; /* Successfully sent config request? */
151 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
153 /* Support for remote controllers. */
154 char *controller; /* NULL if there is no remote controller;
155 * "discover" to do controller discovery;
156 * otherwise a vconn name. */
158 /* OpenFlow switch processing. */
159 struct ofproto *ofproto; /* OpenFlow switch. */
161 /* Kernel datapath information. */
162 struct dpif dpif; /* Kernel datapath. */
163 struct port_array ifaces; /* Indexed by kernel datapath port number. */
167 size_t n_ports, allocated_ports;
170 bool has_bonded_ports;
171 long long int bond_next_rebalance;
176 /* Flow statistics gathering. */
177 time_t next_stats_request;
179 /* Port mirroring. */
180 struct mirror *mirrors[MAX_MIRRORS];
184 long long int stp_last_tick;
187 /* List of all bridges. */
188 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
190 /* Maximum number of datapaths. */
191 enum { DP_MAX = 256 };
193 static struct bridge *bridge_create(const char *name);
194 static void bridge_destroy(struct bridge *);
195 static struct bridge *bridge_lookup(const char *name);
196 static void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *);
197 static int bridge_run_one(struct bridge *);
198 static void bridge_reconfigure_one(struct bridge *);
199 static void bridge_reconfigure_controller(struct bridge *);
200 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
201 static void bridge_fetch_dp_ifaces(struct bridge *);
202 static void bridge_flush(struct bridge *);
203 static void bridge_pick_local_hw_addr(struct bridge *,
204 uint8_t ea[ETH_ADDR_LEN],
205 const char **devname);
206 static uint64_t bridge_pick_datapath_id(struct bridge *,
207 const uint8_t bridge_ea[ETH_ADDR_LEN],
208 const char *devname);
209 static uint64_t dpid_from_hash(const void *, size_t nbytes);
211 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
213 static void bond_init(void);
214 static void bond_run(struct bridge *);
215 static void bond_wait(struct bridge *);
216 static void bond_rebalance_port(struct port *);
217 static void bond_send_learning_packets(struct port *);
219 static void port_create(struct bridge *, const char *name);
220 static void port_reconfigure(struct port *);
221 static void port_destroy(struct port *);
222 static struct port *port_lookup(const struct bridge *, const char *name);
223 static struct iface *port_lookup_iface(const struct port *, const char *name);
224 static struct port *port_from_dp_ifidx(const struct bridge *,
226 static void port_update_bond_compat(struct port *);
227 static void port_update_vlan_compat(struct port *);
229 static void mirror_create(struct bridge *, const char *name);
230 static void mirror_destroy(struct mirror *);
231 static void mirror_reconfigure(struct bridge *);
232 static void mirror_reconfigure_one(struct mirror *);
233 static bool vlan_is_mirrored(const struct mirror *, int vlan);
235 static void brstp_reconfigure(struct bridge *);
236 static void brstp_adjust_timers(struct bridge *);
237 static void brstp_run(struct bridge *);
238 static void brstp_wait(struct bridge *);
240 static void iface_create(struct port *, const char *name);
241 static void iface_destroy(struct iface *);
242 static struct iface *iface_lookup(const struct bridge *, const char *name);
243 static struct iface *iface_from_dp_ifidx(const struct bridge *,
246 /* Hooks into ofproto processing. */
247 static struct ofhooks bridge_ofhooks;
249 /* Public functions. */
251 /* Adds the name of each interface used by a bridge, including local and
252 * internal ports, to 'svec'. */
254 bridge_get_ifaces(struct svec *svec)
256 struct bridge *br, *next;
259 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
260 for (i = 0; i < br->n_ports; i++) {
261 struct port *port = br->ports[i];
263 for (j = 0; j < port->n_ifaces; j++) {
264 struct iface *iface = port->ifaces[j];
265 if (iface->dp_ifidx < 0) {
266 VLOG_ERR("%s interface not in dp%u, ignoring",
267 iface->name, dpif_id(&br->dpif));
269 if (iface->dp_ifidx != ODPP_LOCAL) {
270 svec_add(svec, iface->name);
278 /* The caller must already have called cfg_read(). */
287 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
289 for (i = 0; i < DP_MAX; i++) {
293 sprintf(devname, "dp%d", i);
294 retval = dpif_open(devname, &dpif);
296 char dpif_name[IF_NAMESIZE];
297 if (dpif_get_name(&dpif, dpif_name, sizeof dpif_name)
298 || !cfg_has("bridge.%s.port", dpif_name)) {
302 } else if (retval != ENODEV) {
303 VLOG_ERR("failed to delete datapath dp%d: %s",
304 i, strerror(retval));
308 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows);
310 bridge_reconfigure();
315 config_string_change(const char *key, char **valuep)
317 const char *value = cfg_get_string(0, "%s", key);
318 if (value && (!*valuep || strcmp(value, *valuep))) {
320 *valuep = xstrdup(value);
328 bridge_configure_ssl(void)
330 /* XXX SSL should be configurable on a per-bridge basis.
331 * XXX should be possible to de-configure SSL. */
332 static char *private_key_file;
333 static char *certificate_file;
334 static char *cacert_file;
337 if (config_string_change("ssl.private-key", &private_key_file)) {
338 vconn_ssl_set_private_key_file(private_key_file);
341 if (config_string_change("ssl.certificate", &certificate_file)) {
342 vconn_ssl_set_certificate_file(certificate_file);
345 /* We assume that even if the filename hasn't changed, if the CA cert
346 * file has been removed, that we want to move back into
347 * boot-strapping mode. This opens a small security hole, because
348 * the old certificate will still be trusted until vSwitch is
349 * restarted. We may want to address this in vconn's SSL library. */
350 if (config_string_change("ssl.ca-cert", &cacert_file)
351 || (stat(cacert_file, &s) && errno == ENOENT)) {
352 vconn_ssl_set_ca_cert_file(cacert_file,
353 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
359 bridge_reconfigure(void)
361 struct svec old_br, new_br, raw_new_br;
362 struct bridge *br, *next;
365 COVERAGE_INC(bridge_reconfigure);
367 /* Collect old bridges. */
369 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
370 svec_add(&old_br, br->name);
373 /* Collect new bridges. */
374 svec_init(&raw_new_br);
375 cfg_get_subsections(&raw_new_br, "bridge");
377 for (i = 0; i < raw_new_br.n; i++) {
378 const char *name = raw_new_br.names[i];
379 if ((!strncmp(name, "dp", 2) && isdigit(name[2])) ||
380 (!strncmp(name, "nl:", 3) && isdigit(name[3]))) {
381 VLOG_ERR("%s is not a valid bridge name (bridges may not be "
382 "named \"dp\" or \"nl:\" followed by a digit)", name);
384 svec_add(&new_br, name);
387 svec_destroy(&raw_new_br);
389 /* Get rid of deleted bridges and add new bridges. */
392 assert(svec_is_unique(&old_br));
393 assert(svec_is_unique(&new_br));
394 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
395 if (!svec_contains(&new_br, br->name)) {
399 for (i = 0; i < new_br.n; i++) {
400 const char *name = new_br.names[i];
401 if (!svec_contains(&old_br, name)) {
405 svec_destroy(&old_br);
406 svec_destroy(&new_br);
410 bridge_configure_ssl();
413 /* Reconfigure all bridges. */
414 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
415 bridge_reconfigure_one(br);
418 /* Add and delete ports on all datapaths.
420 * The kernel will reject any attempt to add a given port to a datapath if
421 * that port already belongs to a different datapath, so we must do all
422 * port deletions before any port additions. */
423 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
424 struct odp_port *dpif_ports;
426 struct svec want_ifaces;
428 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
429 bridge_get_all_ifaces(br, &want_ifaces);
430 for (i = 0; i < n_dpif_ports; i++) {
431 const struct odp_port *p = &dpif_ports[i];
432 if (!svec_contains(&want_ifaces, p->devname)
433 && strcmp(p->devname, br->name)) {
434 int retval = dpif_port_del(&br->dpif, p->port);
436 VLOG_ERR("failed to remove %s interface from dp%u: %s",
437 p->devname, dpif_id(&br->dpif), strerror(retval));
441 svec_destroy(&want_ifaces);
444 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
445 struct odp_port *dpif_ports;
447 struct svec cur_ifaces, want_ifaces, add_ifaces;
450 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
451 svec_init(&cur_ifaces);
452 for (i = 0; i < n_dpif_ports; i++) {
453 svec_add(&cur_ifaces, dpif_ports[i].devname);
456 svec_sort_unique(&cur_ifaces);
457 bridge_get_all_ifaces(br, &want_ifaces);
458 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
461 for (i = 0; i < add_ifaces.n; i++) {
462 const char *if_name = add_ifaces.names[i];
467 /* It's an internal interface if it's marked that way, or if
468 * it's a bonded interface for which we're faking up a network
470 internal = cfg_get_bool(0, "iface.%s.internal", if_name);
471 if (cfg_get_bool(0, "bonding.%s.fake-iface", if_name)) {
472 struct port *port = port_lookup(br, if_name);
473 if (port && port->n_ifaces > 1) {
478 /* Add to datapath. */
479 error = dpif_port_add(&br->dpif, if_name, next_port_no++,
480 internal ? ODP_PORT_INTERNAL : 0);
481 if (error != EEXIST) {
482 if (next_port_no >= 256) {
483 VLOG_ERR("ran out of valid port numbers on dp%u",
488 VLOG_ERR("failed to add %s interface to dp%u: %s",
489 if_name, dpif_id(&br->dpif), strerror(error));
496 svec_destroy(&cur_ifaces);
497 svec_destroy(&want_ifaces);
498 svec_destroy(&add_ifaces);
500 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
503 struct iface *local_iface = NULL;
505 uint8_t engine_type = br->dpif.minor;
506 uint8_t engine_id = br->dpif.minor;
507 bool add_id_to_iface = false;
508 struct svec nf_hosts;
510 bridge_fetch_dp_ifaces(br);
511 for (i = 0; i < br->n_ports; ) {
512 struct port *port = br->ports[i];
514 for (j = 0; j < port->n_ifaces; ) {
515 struct iface *iface = port->ifaces[j];
516 if (iface->dp_ifidx < 0) {
517 VLOG_ERR("%s interface not in dp%u, dropping",
518 iface->name, dpif_id(&br->dpif));
519 iface_destroy(iface);
521 if (iface->dp_ifidx == ODPP_LOCAL) {
524 VLOG_DBG("dp%u has interface %s on port %d",
525 dpif_id(&br->dpif), iface->name, iface->dp_ifidx);
529 if (!port->n_ifaces) {
530 VLOG_ERR("%s port has no interfaces, dropping", port->name);
537 /* Pick local port hardware address, datapath ID. */
538 bridge_pick_local_hw_addr(br, ea, &devname);
540 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
542 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
543 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
544 "Ethernet address: %s",
545 br->name, strerror(error));
549 dpid = bridge_pick_datapath_id(br, ea, devname);
550 ofproto_set_datapath_id(br->ofproto, dpid);
552 /* Set NetFlow configuration on this bridge. */
553 if (cfg_has("netflow.%s.engine-type", br->name)) {
554 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
557 if (cfg_has("netflow.%s.engine-id", br->name)) {
558 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
560 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
561 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
564 if (add_id_to_iface && engine_id > 0x7f) {
565 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
566 "another vswitch, choose an engine id less than 128",
569 if (add_id_to_iface && br->n_ports > 0x1ff) {
570 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
571 "another port when 512 or more ports are used",
574 svec_init(&nf_hosts);
575 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
576 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
577 engine_id, add_id_to_iface)) {
578 VLOG_ERR("bridge %s: problem setting netflow collectors",
582 /* Update the controller and related settings. It would be more
583 * straightforward to call this from bridge_reconfigure_one(), but we
584 * can't do it there for two reasons. First, and most importantly, at
585 * that point we don't know the dp_ifidx of any interfaces that have
586 * been added to the bridge (because we haven't actually added them to
587 * the datapath). Second, at that point we haven't set the datapath ID
588 * yet; when a controller is configured, resetting the datapath ID will
589 * immediately disconnect from the controller, so it's better to set
590 * the datapath ID before the controller. */
591 bridge_reconfigure_controller(br);
593 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
594 for (i = 0; i < br->n_ports; i++) {
595 struct port *port = br->ports[i];
596 port_update_vlan_compat(port);
599 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
600 brstp_reconfigure(br);
605 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
606 const char **devname)
608 uint64_t requested_ea;
614 /* Did the user request a particular MAC? */
615 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
617 eth_addr_from_uint64(requested_ea, ea);
618 if (eth_addr_is_multicast(ea)) {
619 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
620 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
621 } else if (eth_addr_is_zero(ea)) {
622 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
628 /* Otherwise choose the minimum MAC address among all of the interfaces.
629 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
630 * MAC of the physical interface in such an environment.) */
631 memset(ea, 0xff, sizeof ea);
632 for (i = 0; i < br->n_ports; i++) {
633 struct port *port = br->ports[i];
634 uint8_t iface_ea[ETH_ADDR_LEN];
635 uint64_t iface_ea_u64;
638 /* Mirror output ports don't participate. */
639 if (port->is_mirror_output_port) {
643 /* Choose the MAC address to represent the port. */
644 iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name);
646 /* User specified explicitly. */
647 eth_addr_from_uint64(iface_ea_u64, iface_ea);
649 /* Find the interface with this Ethernet address (if any) so that
650 * we can provide the correct devname to the caller. */
652 for (j = 0; j < port->n_ifaces; j++) {
653 struct iface *candidate = port->ifaces[j];
654 uint8_t candidate_ea[ETH_ADDR_LEN];
655 if (!netdev_nodev_get_etheraddr(candidate->name, candidate_ea)
656 && eth_addr_equals(iface_ea, candidate_ea)) {
661 /* Choose the interface whose MAC address will represent the port.
662 * The Linux kernel bonding code always chooses the MAC address of
663 * the first slave added to a bond, and the Fedora networking
664 * scripts always add slaves to a bond in alphabetical order, so
665 * for compatibility we choose the interface with the name that is
666 * first in alphabetical order. */
667 iface = port->ifaces[0];
668 for (j = 1; j < port->n_ifaces; j++) {
669 struct iface *candidate = port->ifaces[j];
670 if (strcmp(candidate->name, iface->name) < 0) {
675 /* The local port doesn't count (since we're trying to choose its
676 * MAC address anyway). Other internal ports don't count because
677 * we really want a physical MAC if we can get it, and internal
678 * ports typically have randomly generated MACs. */
679 if (iface->dp_ifidx == ODPP_LOCAL
680 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
685 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
687 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
688 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
689 iface->name, strerror(error));
694 /* Compare against our current choice. */
695 if (!eth_addr_is_multicast(iface_ea) &&
696 !eth_addr_is_reserved(iface_ea) &&
697 !eth_addr_is_zero(iface_ea) &&
698 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
700 memcpy(ea, iface_ea, ETH_ADDR_LEN);
701 *devname = iface ? iface->name : NULL;
704 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
705 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
707 VLOG_WARN("bridge %s: using default bridge Ethernet "
708 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
710 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
711 br->name, ETH_ADDR_ARGS(ea));
715 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
716 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
717 * a network device, then that network device's name must be passed in as
718 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
719 * passed in as a null pointer. */
721 bridge_pick_datapath_id(struct bridge *br,
722 const uint8_t bridge_ea[ETH_ADDR_LEN],
726 * The procedure for choosing a bridge MAC address will, in the most
727 * ordinary case, also choose a unique MAC that we can use as a datapath
728 * ID. In some special cases, though, multiple bridges will end up with
729 * the same MAC address. This is OK for the bridges, but it will confuse
730 * the OpenFlow controller, because each datapath needs a unique datapath
733 * Datapath IDs must be unique. It is also very desirable that they be
734 * stable from one run to the next, so that policy set on a datapath
739 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
746 if (!netdev_get_vlan_vid(devname, &vlan)) {
748 * A bridge whose MAC address is taken from a VLAN network device
749 * (that is, a network device created with vconfig(8) or similar
750 * tool) will have the same MAC address as a bridge on the VLAN
751 * device's physical network device.
753 * Handle this case by hashing the physical network device MAC
754 * along with the VLAN identifier.
756 uint8_t buf[ETH_ADDR_LEN + 2];
757 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
758 buf[ETH_ADDR_LEN] = vlan >> 8;
759 buf[ETH_ADDR_LEN + 1] = vlan;
760 return dpid_from_hash(buf, sizeof buf);
763 * Assume that this bridge's MAC address is unique, since it
764 * doesn't fit any of the cases we handle specially.
769 * A purely internal bridge, that is, one that has no non-virtual
770 * network devices on it at all, is more difficult because it has no
771 * natural unique identifier at all.
773 * When the host is a XenServer, we handle this case by hashing the
774 * host's UUID with the name of the bridge. Names of bridges are
775 * persistent across XenServer reboots, although they can be reused if
776 * an internal network is destroyed and then a new one is later
777 * created, so this is fairly effective.
779 * When the host is not a XenServer, we punt by using a random MAC
780 * address on each run.
782 const char *host_uuid = xenserver_get_host_uuid();
784 char *combined = xasprintf("%s,%s", host_uuid, br->name);
785 dpid = dpid_from_hash(combined, strlen(combined));
791 return eth_addr_to_uint64(bridge_ea);
795 dpid_from_hash(const void *data, size_t n)
797 uint8_t hash[SHA1_DIGEST_SIZE];
799 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
800 sha1_bytes(data, n, hash);
801 eth_addr_mark_random(hash);
802 return eth_addr_to_uint64(hash);
808 struct bridge *br, *next;
812 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
813 int error = bridge_run_one(br);
815 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
816 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
817 "forcing reconfiguration", br->name);
831 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
832 ofproto_wait(br->ofproto);
833 if (br->controller) {
838 mac_learning_wait(br->ml);
845 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
846 * configuration changes. */
848 bridge_flush(struct bridge *br)
850 COVERAGE_INC(bridge_flush);
853 mac_learning_flush(br->ml);
857 /* Bridge unixctl user interface functions. */
859 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
861 struct ds ds = DS_EMPTY_INITIALIZER;
862 const struct bridge *br;
864 br = bridge_lookup(args);
866 unixctl_command_reply(conn, 501, "no such bridge");
870 ds_put_cstr(&ds, " port VLAN MAC Age\n");
872 const struct mac_entry *e;
873 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
874 if (e->port < 0 || e->port >= br->n_ports) {
877 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
878 br->ports[e->port]->ifaces[0]->dp_ifidx,
879 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
882 unixctl_command_reply(conn, 200, ds_cstr(&ds));
886 /* Bridge reconfiguration functions. */
888 static struct bridge *
889 bridge_create(const char *name)
894 assert(!bridge_lookup(name));
895 br = xcalloc(1, sizeof *br);
897 error = dpif_create(name, &br->dpif);
898 if (error == EEXIST) {
899 error = dpif_open(name, &br->dpif);
901 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
902 name, strerror(error));
906 dpif_flow_flush(&br->dpif);
908 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
913 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
915 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
916 dpif_delete(&br->dpif);
917 dpif_close(&br->dpif);
922 br->name = xstrdup(name);
923 br->ml = mac_learning_create();
924 br->sent_config_request = false;
925 eth_addr_random(br->default_ea);
927 port_array_init(&br->ifaces);
930 br->bond_next_rebalance = time_msec() + 10000;
932 list_push_back(&all_bridges, &br->node);
934 VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif));
940 bridge_destroy(struct bridge *br)
945 while (br->n_ports > 0) {
946 port_destroy(br->ports[br->n_ports - 1]);
948 list_remove(&br->node);
949 error = dpif_delete(&br->dpif);
950 if (error && error != ENOENT) {
951 VLOG_ERR("failed to delete dp%u: %s",
952 dpif_id(&br->dpif), strerror(error));
954 dpif_close(&br->dpif);
955 ofproto_destroy(br->ofproto);
956 free(br->controller);
957 mac_learning_destroy(br->ml);
958 port_array_destroy(&br->ifaces);
965 static struct bridge *
966 bridge_lookup(const char *name)
970 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
971 if (!strcmp(br->name, name)) {
979 bridge_exists(const char *name)
981 return bridge_lookup(name) ? true : false;
985 bridge_get_datapathid(const char *name)
987 struct bridge *br = bridge_lookup(name);
988 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
991 /* Handle requests for a listing of all flows known by the OpenFlow
992 * stack, including those normally hidden. */
994 bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args)
999 br = bridge_lookup(args);
1001 unixctl_command_reply(conn, 501, "Unknown bridge");
1006 ofproto_get_all_flows(br->ofproto, &results);
1008 unixctl_command_reply(conn, 200, ds_cstr(&results));
1009 ds_destroy(&results);
1013 bridge_run_one(struct bridge *br)
1017 error = ofproto_run1(br->ofproto);
1023 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1028 error = ofproto_run2(br->ofproto, br->flush);
1035 bridge_get_controller(const struct bridge *br)
1037 const char *controller;
1039 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1041 controller = cfg_get_string(0, "mgmt.controller");
1043 return controller && controller[0] ? controller : NULL;
1047 bridge_reconfigure_one(struct bridge *br)
1049 struct svec old_ports, new_ports, ifaces;
1050 struct svec listeners, old_listeners;
1051 struct svec snoops, old_snoops;
1054 /* Collect old ports. */
1055 svec_init(&old_ports);
1056 for (i = 0; i < br->n_ports; i++) {
1057 svec_add(&old_ports, br->ports[i]->name);
1059 svec_sort(&old_ports);
1060 assert(svec_is_unique(&old_ports));
1062 /* Collect new ports. */
1063 svec_init(&new_ports);
1064 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1065 svec_sort(&new_ports);
1066 if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) {
1067 svec_add(&new_ports, br->name);
1068 svec_sort(&new_ports);
1070 if (!svec_is_unique(&new_ports)) {
1071 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1072 br->name, svec_get_duplicate(&new_ports));
1073 svec_unique(&new_ports);
1076 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1078 /* Get rid of deleted ports and add new ports. */
1079 for (i = 0; i < br->n_ports; ) {
1080 struct port *port = br->ports[i];
1081 if (!svec_contains(&new_ports, port->name)) {
1087 for (i = 0; i < new_ports.n; i++) {
1088 const char *name = new_ports.names[i];
1089 if (!svec_contains(&old_ports, name)) {
1090 port_create(br, name);
1093 svec_destroy(&old_ports);
1094 svec_destroy(&new_ports);
1096 /* Reconfigure all ports. */
1097 for (i = 0; i < br->n_ports; i++) {
1098 port_reconfigure(br->ports[i]);
1101 /* Check and delete duplicate interfaces. */
1103 for (i = 0; i < br->n_ports; ) {
1104 struct port *port = br->ports[i];
1105 for (j = 0; j < port->n_ifaces; ) {
1106 struct iface *iface = port->ifaces[j];
1107 if (svec_contains(&ifaces, iface->name)) {
1108 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1110 br->name, iface->name, port->name);
1111 iface_destroy(iface);
1113 svec_add(&ifaces, iface->name);
1118 if (!port->n_ifaces) {
1119 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1125 svec_destroy(&ifaces);
1127 /* Delete all flows if we're switching from connected to standalone or vice
1128 * versa. (XXX Should we delete all flows if we are switching from one
1129 * controller to another?) */
1131 /* Configure OpenFlow management listeners. */
1132 svec_init(&listeners);
1133 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1135 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1136 ovs_rundir, br->name));
1137 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1138 svec_clear(&listeners);
1140 svec_sort_unique(&listeners);
1142 svec_init(&old_listeners);
1143 ofproto_get_listeners(br->ofproto, &old_listeners);
1144 svec_sort_unique(&old_listeners);
1146 if (!svec_equal(&listeners, &old_listeners)) {
1147 ofproto_set_listeners(br->ofproto, &listeners);
1149 svec_destroy(&listeners);
1150 svec_destroy(&old_listeners);
1152 /* Configure OpenFlow controller connection snooping. */
1154 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1156 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1157 ovs_rundir, br->name));
1158 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1159 svec_clear(&snoops);
1161 svec_sort_unique(&snoops);
1163 svec_init(&old_snoops);
1164 ofproto_get_snoops(br->ofproto, &old_snoops);
1165 svec_sort_unique(&old_snoops);
1167 if (!svec_equal(&snoops, &old_snoops)) {
1168 ofproto_set_snoops(br->ofproto, &snoops);
1170 svec_destroy(&snoops);
1171 svec_destroy(&old_snoops);
1173 mirror_reconfigure(br);
1177 bridge_reconfigure_controller(struct bridge *br)
1179 char *pfx = xasprintf("bridge.%s.controller", br->name);
1180 const char *controller;
1182 controller = bridge_get_controller(br);
1183 if ((br->controller != NULL) != (controller != NULL)) {
1184 ofproto_flush_flows(br->ofproto);
1186 free(br->controller);
1187 br->controller = controller ? xstrdup(controller) : NULL;
1190 const char *fail_mode;
1191 int max_backoff, probe;
1192 int rate_limit, burst_limit;
1194 if (!strcmp(controller, "discover")) {
1195 bool update_resolv_conf = true;
1197 if (cfg_has("%s.update-resolv.conf", pfx)) {
1198 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1201 ofproto_set_discovery(br->ofproto, true,
1202 cfg_get_string(0, "%s.accept-regex", pfx),
1203 update_resolv_conf);
1205 struct netdev *netdev;
1209 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1211 || cfg_get_bool(0, "%s.in-band", pfx));
1212 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1213 ofproto_set_in_band(br->ofproto, in_band);
1215 error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev);
1217 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1218 struct in_addr ip, mask, gateway;
1219 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1220 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1221 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1223 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1225 mask.s_addr = guess_netmask(ip.s_addr);
1227 if (!netdev_set_in4(netdev, ip, mask)) {
1228 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1230 br->name, IP_ARGS(&ip.s_addr),
1231 IP_ARGS(&mask.s_addr));
1234 if (gateway.s_addr) {
1235 if (!netdev_add_router(gateway)) {
1236 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1237 br->name, IP_ARGS(&gateway.s_addr));
1241 netdev_close(netdev);
1245 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1247 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1249 ofproto_set_failure(br->ofproto,
1251 || !strcmp(fail_mode, "standalone")
1252 || !strcmp(fail_mode, "open")));
1254 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1256 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1261 ofproto_set_probe_interval(br->ofproto, probe);
1263 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1265 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1270 ofproto_set_max_backoff(br->ofproto, max_backoff);
1272 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1274 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1276 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1278 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1280 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1282 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1284 if (cfg_has("%s.commands.acl", pfx)) {
1285 struct svec command_acls;
1288 svec_init(&command_acls);
1289 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1290 command_acl = svec_join(&command_acls, ",", "");
1292 ofproto_set_remote_execution(br->ofproto, command_acl,
1293 cfg_get_string(0, "%s.commands.dir",
1296 svec_destroy(&command_acls);
1299 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1302 union ofp_action action;
1305 /* Set up a flow that matches every packet and directs them to
1306 * OFPP_NORMAL (which goes to us). */
1307 memset(&action, 0, sizeof action);
1308 action.type = htons(OFPAT_OUTPUT);
1309 action.output.len = htons(sizeof action);
1310 action.output.port = htons(OFPP_NORMAL);
1311 memset(&flow, 0, sizeof flow);
1312 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1315 ofproto_set_in_band(br->ofproto, false);
1316 ofproto_set_max_backoff(br->ofproto, 1);
1317 ofproto_set_probe_interval(br->ofproto, 5);
1318 ofproto_set_failure(br->ofproto, false);
1319 ofproto_set_stp(br->ofproto, false);
1323 ofproto_set_controller(br->ofproto, br->controller);
1327 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1332 for (i = 0; i < br->n_ports; i++) {
1333 struct port *port = br->ports[i];
1334 for (j = 0; j < port->n_ifaces; j++) {
1335 struct iface *iface = port->ifaces[j];
1336 svec_add(ifaces, iface->name);
1338 if (port->n_ifaces > 1
1339 && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
1340 svec_add(ifaces, port->name);
1343 svec_sort_unique(ifaces);
1346 /* For robustness, in case the administrator moves around datapath ports behind
1347 * our back, we re-check all the datapath port numbers here.
1349 * This function will set the 'dp_ifidx' members of interfaces that have
1350 * disappeared to -1, so only call this function from a context where those
1351 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1352 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1353 * datapath, which doesn't support UINT16_MAX+1 ports. */
1355 bridge_fetch_dp_ifaces(struct bridge *br)
1357 struct odp_port *dpif_ports;
1358 size_t n_dpif_ports;
1361 /* Reset all interface numbers. */
1362 for (i = 0; i < br->n_ports; i++) {
1363 struct port *port = br->ports[i];
1364 for (j = 0; j < port->n_ifaces; j++) {
1365 struct iface *iface = port->ifaces[j];
1366 iface->dp_ifidx = -1;
1369 port_array_clear(&br->ifaces);
1371 dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports);
1372 for (i = 0; i < n_dpif_ports; i++) {
1373 struct odp_port *p = &dpif_ports[i];
1374 struct iface *iface = iface_lookup(br, p->devname);
1376 if (iface->dp_ifidx >= 0) {
1377 VLOG_WARN("dp%u reported interface %s twice",
1378 dpif_id(&br->dpif), p->devname);
1379 } else if (iface_from_dp_ifidx(br, p->port)) {
1380 VLOG_WARN("dp%u reported interface %"PRIu16" twice",
1381 dpif_id(&br->dpif), p->port);
1383 port_array_set(&br->ifaces, p->port, iface);
1384 iface->dp_ifidx = p->port;
1391 /* Bridge packet processing functions. */
1394 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1396 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1399 static struct bond_entry *
1400 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1402 return &port->bond_hash[bond_hash(mac)];
1406 bond_choose_iface(const struct port *port)
1409 for (i = 0; i < port->n_ifaces; i++) {
1410 if (port->ifaces[i]->enabled) {
1418 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1419 uint16_t *dp_ifidx, tag_type *tags)
1421 struct iface *iface;
1423 assert(port->n_ifaces);
1424 if (port->n_ifaces == 1) {
1425 iface = port->ifaces[0];
1427 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1428 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1429 || !port->ifaces[e->iface_idx]->enabled) {
1430 /* XXX select interface properly. The current interface selection
1431 * is only good for testing the rebalancing code. */
1432 e->iface_idx = bond_choose_iface(port);
1433 if (e->iface_idx < 0) {
1434 *tags |= port->no_ifaces_tag;
1437 e->iface_tag = tag_create_random();
1438 ((struct port *) port)->bond_compat_is_stale = true;
1440 *tags |= e->iface_tag;
1441 iface = port->ifaces[e->iface_idx];
1443 *dp_ifidx = iface->dp_ifidx;
1444 *tags |= iface->tag; /* Currently only used for bonding. */
1449 bond_link_status_update(struct iface *iface, bool carrier)
1451 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1452 struct port *port = iface->port;
1454 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1455 /* Nothing to do. */
1458 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1459 iface->name, carrier ? "detected" : "dropped");
1460 if (carrier == iface->enabled) {
1461 iface->delay_expires = LLONG_MAX;
1462 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1463 iface->name, carrier ? "disabled" : "enabled");
1464 } else if (carrier && port->updelay && port->active_iface < 0) {
1465 iface->delay_expires = time_msec();
1466 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1467 "other interface is up", iface->name, port->updelay);
1469 int delay = carrier ? port->updelay : port->downdelay;
1470 iface->delay_expires = time_msec() + delay;
1473 "interface %s: will be %s if it stays %s for %d ms",
1475 carrier ? "enabled" : "disabled",
1476 carrier ? "up" : "down",
1483 bond_choose_active_iface(struct port *port)
1485 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1487 port->active_iface = bond_choose_iface(port);
1488 port->active_iface_tag = tag_create_random();
1489 if (port->active_iface >= 0) {
1490 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1491 port->name, port->ifaces[port->active_iface]->name);
1493 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1499 bond_enable_slave(struct iface *iface, bool enable)
1501 struct port *port = iface->port;
1502 struct bridge *br = port->bridge;
1504 iface->delay_expires = LLONG_MAX;
1505 if (enable == iface->enabled) {
1509 iface->enabled = enable;
1510 if (!iface->enabled) {
1511 VLOG_WARN("interface %s: disabled", iface->name);
1512 ofproto_revalidate(br->ofproto, iface->tag);
1513 if (iface->port_ifidx == port->active_iface) {
1514 ofproto_revalidate(br->ofproto,
1515 port->active_iface_tag);
1516 bond_choose_active_iface(port);
1518 bond_send_learning_packets(port);
1520 VLOG_WARN("interface %s: enabled", iface->name);
1521 if (port->active_iface < 0) {
1522 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1523 bond_choose_active_iface(port);
1524 bond_send_learning_packets(port);
1526 iface->tag = tag_create_random();
1531 bond_run(struct bridge *br)
1535 for (i = 0; i < br->n_ports; i++) {
1536 struct port *port = br->ports[i];
1538 if (port->bond_compat_is_stale) {
1539 port->bond_compat_is_stale = false;
1540 port_update_bond_compat(port);
1543 if (port->n_ifaces < 2) {
1546 for (j = 0; j < port->n_ifaces; j++) {
1547 struct iface *iface = port->ifaces[j];
1548 if (time_msec() >= iface->delay_expires) {
1549 bond_enable_slave(iface, !iface->enabled);
1556 bond_wait(struct bridge *br)
1560 for (i = 0; i < br->n_ports; i++) {
1561 struct port *port = br->ports[i];
1562 if (port->n_ifaces < 2) {
1565 for (j = 0; j < port->n_ifaces; j++) {
1566 struct iface *iface = port->ifaces[j];
1567 if (iface->delay_expires != LLONG_MAX) {
1568 poll_timer_wait(iface->delay_expires - time_msec());
1575 set_dst(struct dst *p, const flow_t *flow,
1576 const struct port *in_port, const struct port *out_port,
1581 * XXX This uses too many tags: any broadcast flow will get one tag per
1582 * destination port, and thus a broadcast on a switch of any size is likely
1583 * to have all tag bits set. We should figure out a way to be smarter.
1585 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1586 *tags |= out_port->stp_state_tag;
1587 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1591 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1592 : in_port->vlan >= 0 ? in_port->vlan
1593 : ntohs(flow->dl_vlan));
1594 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1598 swap_dst(struct dst *p, struct dst *q)
1600 struct dst tmp = *p;
1605 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1606 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1607 * that we push to the datapath. We could in fact fully sort the array by
1608 * vlan, but in most cases there are at most two different vlan tags so that's
1609 * possibly overkill.) */
1611 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1613 struct dst *first = dsts;
1614 struct dst *last = dsts + n_dsts;
1616 while (first != last) {
1618 * - All dsts < first have vlan == 'vlan'.
1619 * - All dsts >= last have vlan != 'vlan'.
1620 * - first < last. */
1621 while (first->vlan == vlan) {
1622 if (++first == last) {
1627 /* Same invariants, plus one additional:
1628 * - first->vlan != vlan.
1630 while (last[-1].vlan != vlan) {
1631 if (--last == first) {
1636 /* Same invariants, plus one additional:
1637 * - last[-1].vlan == vlan.*/
1638 swap_dst(first++, --last);
1643 mirror_mask_ffs(mirror_mask_t mask)
1645 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1650 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1651 const struct dst *test)
1654 for (i = 0; i < n_dsts; i++) {
1655 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1663 port_trunks_vlan(const struct port *port, uint16_t vlan)
1665 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1669 port_includes_vlan(const struct port *port, uint16_t vlan)
1671 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1675 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1676 const struct port *in_port, const struct port *out_port,
1677 struct dst dsts[], tag_type *tags)
1679 mirror_mask_t mirrors = in_port->src_mirrors;
1680 struct dst *dst = dsts;
1683 *tags |= in_port->stp_state_tag;
1684 if (out_port == FLOOD_PORT) {
1685 /* XXX use ODP_FLOOD if no vlans or bonding. */
1686 /* XXX even better, define each VLAN as a datapath port group */
1687 for (i = 0; i < br->n_ports; i++) {
1688 struct port *port = br->ports[i];
1689 if (port != in_port && port_includes_vlan(port, vlan)
1690 && !port->is_mirror_output_port
1691 && set_dst(dst, flow, in_port, port, tags)) {
1692 mirrors |= port->dst_mirrors;
1696 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1697 mirrors |= out_port->dst_mirrors;
1702 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1703 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1705 if (set_dst(dst, flow, in_port, m->out_port, tags)
1706 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1710 for (i = 0; i < br->n_ports; i++) {
1711 struct port *port = br->ports[i];
1712 if (port_includes_vlan(port, m->out_vlan)
1713 && set_dst(dst, flow, in_port, port, tags)
1714 && !dst_is_duplicate(dsts, dst - dsts, dst))
1716 if (port->vlan < 0) {
1717 dst->vlan = m->out_vlan;
1719 if (dst->dp_ifidx == flow->in_port
1720 && dst->vlan == vlan) {
1721 /* Don't send out input port on same VLAN. */
1729 mirrors &= mirrors - 1;
1732 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1737 print_dsts(const struct dst *dsts, size_t n)
1739 for (; n--; dsts++) {
1740 printf(">p%"PRIu16, dsts->dp_ifidx);
1741 if (dsts->vlan != OFP_VLAN_NONE) {
1742 printf("v%"PRIu16, dsts->vlan);
1748 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1749 const struct port *in_port, const struct port *out_port,
1750 tag_type *tags, struct odp_actions *actions)
1752 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1754 const struct dst *p;
1757 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1759 cur_vlan = ntohs(flow->dl_vlan);
1760 for (p = dsts; p < &dsts[n_dsts]; p++) {
1761 union odp_action *a;
1762 if (p->vlan != cur_vlan) {
1763 if (p->vlan == OFP_VLAN_NONE) {
1764 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1766 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1767 a->vlan_vid.vlan_vid = htons(p->vlan);
1771 a = odp_actions_add(actions, ODPAT_OUTPUT);
1772 a->output.port = p->dp_ifidx;
1777 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1779 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1780 return (flow->dl_type == htons(ETH_TYPE_ARP)
1781 && eth_addr_is_broadcast(flow->dl_dst)
1782 && packet->size >= sizeof(struct arp_eth_header)
1783 && arp->ar_op == ARP_OP_REQUEST);
1786 /* If the composed actions may be applied to any packet in the given 'flow',
1787 * returns true. Otherwise, the actions should only be applied to 'packet', or
1788 * not at all, if 'packet' was NULL. */
1790 process_flow(struct bridge *br, const flow_t *flow,
1791 const struct ofpbuf *packet, struct odp_actions *actions,
1794 struct iface *in_iface;
1795 struct port *in_port;
1796 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1799 /* Find the interface and port structure for the received packet. */
1800 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1802 /* No interface? Something fishy... */
1803 if (packet != NULL) {
1804 /* Odd. A few possible reasons here:
1806 * - We deleted an interface but there are still a few packets
1807 * queued up from it.
1809 * - Someone externally added an interface (e.g. with "ovs-dpctl
1810 * add-if") that we don't know about.
1812 * - Packet arrived on the local port but the local port is not
1813 * one of our bridge ports.
1815 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1817 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1818 "interface %"PRIu16, br->name, flow->in_port);
1821 /* Return without adding any actions, to drop packets on this flow. */
1824 in_port = in_iface->port;
1826 /* Figure out what VLAN this packet belongs to.
1828 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1829 * belongs to VLAN 0, so we should treat both cases identically. (In the
1830 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1831 * presumably to allow a priority to be specified. In the latter case, the
1832 * packet does not have any 802.1Q header.) */
1833 vlan = ntohs(flow->dl_vlan);
1834 if (vlan == OFP_VLAN_NONE) {
1837 if (in_port->vlan >= 0) {
1839 /* XXX support double tagging? */
1840 if (packet != NULL) {
1841 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1842 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1843 "packet received on port %s configured with "
1844 "implicit VLAN %"PRIu16,
1845 br->name, ntohs(flow->dl_vlan),
1846 in_port->name, in_port->vlan);
1850 vlan = in_port->vlan;
1852 if (!port_includes_vlan(in_port, vlan)) {
1853 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1854 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1855 "packet received on port %s not configured for "
1857 br->name, vlan, in_port->name, vlan);
1862 /* Drop frames for ports that STP wants entirely killed (both for
1863 * forwarding and for learning). Later, after we do learning, we'll drop
1864 * the frames that STP wants to do learning but not forwarding on. */
1865 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1869 /* Drop frames for reserved multicast addresses. */
1870 if (eth_addr_is_reserved(flow->dl_dst)) {
1874 /* Drop frames on ports reserved for mirroring. */
1875 if (in_port->is_mirror_output_port) {
1876 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1877 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1878 "which is reserved exclusively for mirroring",
1879 br->name, in_port->name);
1883 /* Multicast (and broadcast) packets on bonds need special attention, to
1884 * avoid receiving duplicates. */
1885 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1886 *tags |= in_port->active_iface_tag;
1887 if (in_port->active_iface != in_iface->port_ifidx) {
1888 /* Drop all multicast packets on inactive slaves. */
1891 /* Drop all multicast packets for which we have learned a different
1892 * input port, because we probably sent the packet on one slaves
1893 * and got it back on the active slave. Broadcast ARP replies are
1894 * an exception to this rule: the host has moved to another
1896 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1897 if (src_idx != -1 && src_idx != in_port->port_idx) {
1899 if (!is_bcast_arp_reply(flow, packet)) {
1903 /* No way to know whether it's an ARP reply, because the
1904 * flow entry doesn't include enough information and we
1905 * don't have a packet. Punt. */
1913 out_port = FLOOD_PORT;
1917 /* Learn source MAC (but don't try to learn from revalidation). */
1919 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1920 vlan, in_port->port_idx);
1922 /* The log messages here could actually be useful in debugging,
1923 * so keep the rate limit relatively high. */
1924 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1926 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1927 "on port %s in VLAN %d",
1928 br->name, ETH_ADDR_ARGS(flow->dl_src),
1929 in_port->name, vlan);
1930 ofproto_revalidate(br->ofproto, rev_tag);
1934 /* Determine output port. */
1935 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1937 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1938 out_port = br->ports[out_port_idx];
1942 /* Don't send packets out their input ports. Don't forward frames that STP
1943 * wants us to discard. */
1944 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1949 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1952 * We send out only a single packet, instead of setting up a flow, if the
1953 * packet is an ARP directed to broadcast that arrived on a bonded
1954 * interface. In such a situation ARP requests and replies must be handled
1955 * differently, but OpenFlow unfortunately can't distinguish them.
1957 return (in_port->n_ifaces < 2
1958 || flow->dl_type != htons(ETH_TYPE_ARP)
1959 || !eth_addr_is_broadcast(flow->dl_dst));
1962 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1965 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1966 const struct ofp_phy_port *opp,
1969 struct bridge *br = br_;
1970 struct iface *iface;
1973 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1979 if (reason == OFPPR_DELETE) {
1980 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1981 br->name, iface->name);
1982 iface_destroy(iface);
1983 if (!port->n_ifaces) {
1984 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1985 br->name, port->name);
1991 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1992 if (port->n_ifaces > 1) {
1993 bool up = !(opp->state & OFPPS_LINK_DOWN);
1994 bond_link_status_update(iface, up);
1995 port_update_bond_compat(port);
2001 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2002 struct odp_actions *actions, tag_type *tags, void *br_)
2004 struct bridge *br = br_;
2007 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
2008 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
2009 brstp_receive(br, flow, payload);
2014 COVERAGE_INC(bridge_process_flow);
2015 return process_flow(br, flow, packet, actions, tags);
2019 bridge_account_flow_ofhook_cb(const flow_t *flow,
2020 const union odp_action *actions,
2021 size_t n_actions, unsigned long long int n_bytes,
2024 struct bridge *br = br_;
2025 const union odp_action *a;
2027 if (!br->has_bonded_ports) {
2031 for (a = actions; a < &actions[n_actions]; a++) {
2032 if (a->type == ODPAT_OUTPUT) {
2033 struct port *port = port_from_dp_ifidx(br, a->output.port);
2034 if (port && port->n_ifaces >= 2) {
2035 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2036 e->tx_bytes += n_bytes;
2043 bridge_account_checkpoint_ofhook_cb(void *br_)
2045 struct bridge *br = br_;
2048 if (!br->has_bonded_ports) {
2052 /* The current ofproto implementation calls this callback at least once a
2053 * second, so this timer implementation is sufficient. */
2054 if (time_msec() < br->bond_next_rebalance) {
2057 br->bond_next_rebalance = time_msec() + 10000;
2059 for (i = 0; i < br->n_ports; i++) {
2060 struct port *port = br->ports[i];
2061 if (port->n_ifaces > 1) {
2062 bond_rebalance_port(port);
2067 static struct ofhooks bridge_ofhooks = {
2068 bridge_port_changed_ofhook_cb,
2069 bridge_normal_ofhook_cb,
2070 bridge_account_flow_ofhook_cb,
2071 bridge_account_checkpoint_ofhook_cb,
2074 /* Bonding functions. */
2076 /* Statistics for a single interface on a bonded port, used for load-based
2077 * bond rebalancing. */
2078 struct slave_balance {
2079 struct iface *iface; /* The interface. */
2080 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2082 /* All the "bond_entry"s that are assigned to this interface, in order of
2083 * increasing tx_bytes. */
2084 struct bond_entry **hashes;
2088 /* Sorts pointers to pointers to bond_entries in ascending order by the
2089 * interface to which they are assigned, and within a single interface in
2090 * ascending order of bytes transmitted. */
2092 compare_bond_entries(const void *a_, const void *b_)
2094 const struct bond_entry *const *ap = a_;
2095 const struct bond_entry *const *bp = b_;
2096 const struct bond_entry *a = *ap;
2097 const struct bond_entry *b = *bp;
2098 if (a->iface_idx != b->iface_idx) {
2099 return a->iface_idx > b->iface_idx ? 1 : -1;
2100 } else if (a->tx_bytes != b->tx_bytes) {
2101 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2107 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2108 * *descending* order by number of bytes transmitted. */
2110 compare_slave_balance(const void *a_, const void *b_)
2112 const struct slave_balance *a = a_;
2113 const struct slave_balance *b = b_;
2114 if (a->iface->enabled != b->iface->enabled) {
2115 return a->iface->enabled ? -1 : 1;
2116 } else if (a->tx_bytes != b->tx_bytes) {
2117 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2124 swap_bals(struct slave_balance *a, struct slave_balance *b)
2126 struct slave_balance tmp = *a;
2131 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2132 * given that 'p' (and only 'p') might be in the wrong location.
2134 * This function invalidates 'p', since it might now be in a different memory
2137 resort_bals(struct slave_balance *p,
2138 struct slave_balance bals[], size_t n_bals)
2141 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2142 swap_bals(p, p - 1);
2144 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2145 swap_bals(p, p + 1);
2151 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2153 if (VLOG_IS_DBG_ENABLED()) {
2154 struct ds ds = DS_EMPTY_INITIALIZER;
2155 const struct slave_balance *b;
2157 for (b = bals; b < bals + n_bals; b++) {
2161 ds_put_char(&ds, ',');
2163 ds_put_format(&ds, " %s %"PRIu64"kB",
2164 b->iface->name, b->tx_bytes / 1024);
2166 if (!b->iface->enabled) {
2167 ds_put_cstr(&ds, " (disabled)");
2169 if (b->n_hashes > 0) {
2170 ds_put_cstr(&ds, " (");
2171 for (i = 0; i < b->n_hashes; i++) {
2172 const struct bond_entry *e = b->hashes[i];
2174 ds_put_cstr(&ds, " + ");
2176 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2177 e - port->bond_hash, e->tx_bytes / 1024);
2179 ds_put_cstr(&ds, ")");
2182 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2187 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2189 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2190 struct bond_entry *hash)
2192 struct port *port = from->iface->port;
2193 uint64_t delta = hash->tx_bytes;
2195 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2196 "from %s to %s (now carrying %"PRIu64"kB and "
2197 "%"PRIu64"kB load, respectively)",
2198 port->name, delta / 1024, hash - port->bond_hash,
2199 from->iface->name, to->iface->name,
2200 (from->tx_bytes - delta) / 1024,
2201 (to->tx_bytes + delta) / 1024);
2203 /* Delete element from from->hashes.
2205 * We don't bother to add the element to to->hashes because not only would
2206 * it require more work, the only purpose it would be to allow that hash to
2207 * be migrated to another slave in this rebalancing run, and there is no
2208 * point in doing that. */
2209 if (from->hashes[0] == hash) {
2212 int i = hash - from->hashes[0];
2213 memmove(from->hashes + i, from->hashes + i + 1,
2214 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2218 /* Shift load away from 'from' to 'to'. */
2219 from->tx_bytes -= delta;
2220 to->tx_bytes += delta;
2222 /* Arrange for flows to be revalidated. */
2223 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2224 hash->iface_idx = to->iface->port_ifidx;
2225 hash->iface_tag = tag_create_random();
2229 bond_rebalance_port(struct port *port)
2231 struct slave_balance bals[DP_MAX_PORTS];
2233 struct bond_entry *hashes[BOND_MASK + 1];
2234 struct slave_balance *b, *from, *to;
2235 struct bond_entry *e;
2238 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2239 * descending order of tx_bytes, so that bals[0] represents the most
2240 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2243 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2244 * array for each slave_balance structure, we sort our local array of
2245 * hashes in order by slave, so that all of the hashes for a given slave
2246 * become contiguous in memory, and then we point each 'hashes' members of
2247 * a slave_balance structure to the start of a contiguous group. */
2248 n_bals = port->n_ifaces;
2249 for (b = bals; b < &bals[n_bals]; b++) {
2250 b->iface = port->ifaces[b - bals];
2255 for (i = 0; i <= BOND_MASK; i++) {
2256 hashes[i] = &port->bond_hash[i];
2258 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2259 for (i = 0; i <= BOND_MASK; i++) {
2261 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2262 b = &bals[e->iface_idx];
2263 b->tx_bytes += e->tx_bytes;
2265 b->hashes = &hashes[i];
2270 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2271 log_bals(bals, n_bals, port);
2273 /* Discard slaves that aren't enabled (which were sorted to the back of the
2274 * array earlier). */
2275 while (!bals[n_bals - 1].iface->enabled) {
2282 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2283 to = &bals[n_bals - 1];
2284 for (from = bals; from < to; ) {
2285 uint64_t overload = from->tx_bytes - to->tx_bytes;
2286 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2287 /* The extra load on 'from' (and all less-loaded slaves), compared
2288 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2289 * it is less than ~1Mbps. No point in rebalancing. */
2291 } else if (from->n_hashes == 1) {
2292 /* 'from' only carries a single MAC hash, so we can't shift any
2293 * load away from it, even though we want to. */
2296 /* 'from' is carrying significantly more load than 'to', and that
2297 * load is split across at least two different hashes. Pick a hash
2298 * to migrate to 'to' (the least-loaded slave), given that doing so
2299 * must not cause 'to''s load to exceed 'from''s load.
2301 * The sort order we use means that we prefer to shift away the
2302 * smallest hashes instead of the biggest ones. There is little
2303 * reason behind this decision; we could use the opposite sort
2304 * order to shift away big hashes ahead of small ones. */
2307 for (i = 0; i < from->n_hashes; i++) {
2308 uint64_t delta = from->hashes[i]->tx_bytes;
2309 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2313 if (i < from->n_hashes) {
2314 bond_shift_load(from, to, from->hashes[i]);
2316 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2317 * point to different slave_balance structures. It is only
2318 * valid to do these two operations in a row at all because we
2319 * know that 'from' will not move past 'to' and vice versa. */
2320 resort_bals(from, bals, n_bals);
2321 resort_bals(to, bals, n_bals);
2325 port->bond_compat_is_stale = true;
2329 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2330 * historical data to decay to <1% in 7 rebalancing runs. */
2331 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2337 bond_send_learning_packets(struct port *port)
2339 struct bridge *br = port->bridge;
2340 struct mac_entry *e;
2341 struct ofpbuf packet;
2342 int error, n_packets, n_errors;
2344 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2348 ofpbuf_init(&packet, 128);
2349 error = n_packets = n_errors = 0;
2350 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2351 static const char s[] = "Open vSwitch Bond Failover";
2352 union ofp_action actions[2], *a;
2353 struct eth_header *eth;
2354 struct llc_snap_header *llc_snap;
2360 if (e->port == port->port_idx
2361 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2365 /* Compose packet to send. */
2366 ofpbuf_clear(&packet);
2367 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2368 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2369 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2370 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2372 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2373 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2374 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2376 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2377 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2378 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2379 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2380 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2382 /* Compose actions. */
2383 memset(actions, 0, sizeof actions);
2386 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2387 a->vlan_vid.len = htons(sizeof *a);
2388 a->vlan_vid.vlan_vid = htons(e->vlan);
2391 a->output.type = htons(OFPAT_OUTPUT);
2392 a->output.len = htons(sizeof *a);
2393 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2398 flow_extract(&packet, ODPP_NONE, &flow);
2399 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2406 ofpbuf_uninit(&packet);
2409 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2410 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2411 "packets, last error was: %s",
2412 port->name, n_errors, n_packets, strerror(error));
2414 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2415 port->name, n_packets);
2419 /* Bonding unixctl user interface functions. */
2422 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2424 struct ds ds = DS_EMPTY_INITIALIZER;
2425 const struct bridge *br;
2427 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2429 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2432 for (i = 0; i < br->n_ports; i++) {
2433 const struct port *port = br->ports[i];
2434 if (port->n_ifaces > 1) {
2437 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2438 for (j = 0; j < port->n_ifaces; j++) {
2439 const struct iface *iface = port->ifaces[j];
2441 ds_put_cstr(&ds, ", ");
2443 ds_put_cstr(&ds, iface->name);
2445 ds_put_char(&ds, '\n');
2449 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2453 static struct port *
2454 bond_find(const char *name)
2456 const struct bridge *br;
2458 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2461 for (i = 0; i < br->n_ports; i++) {
2462 struct port *port = br->ports[i];
2463 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2472 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2474 struct ds ds = DS_EMPTY_INITIALIZER;
2475 const struct port *port;
2478 port = bond_find(args);
2480 unixctl_command_reply(conn, 501, "no such bond");
2484 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2485 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2486 ds_put_format(&ds, "next rebalance: %lld ms\n",
2487 port->bridge->bond_next_rebalance - time_msec());
2488 for (j = 0; j < port->n_ifaces; j++) {
2489 const struct iface *iface = port->ifaces[j];
2490 struct bond_entry *be;
2493 ds_put_format(&ds, "slave %s: %s\n",
2494 iface->name, iface->enabled ? "enabled" : "disabled");
2495 if (j == port->active_iface) {
2496 ds_put_cstr(&ds, "\tactive slave\n");
2498 if (iface->delay_expires != LLONG_MAX) {
2499 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2500 iface->enabled ? "downdelay" : "updelay",
2501 iface->delay_expires - time_msec());
2505 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2506 int hash = be - port->bond_hash;
2507 struct mac_entry *me;
2509 if (be->iface_idx != j) {
2513 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2514 hash, be->tx_bytes / 1024);
2517 if (!port->bridge->ml) {
2521 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2522 &port->bridge->ml->lrus) {
2525 if (bond_hash(me->mac) == hash
2526 && me->port != port->port_idx
2527 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2528 && dp_ifidx == iface->dp_ifidx)
2530 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2531 ETH_ADDR_ARGS(me->mac));
2536 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2541 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2543 char *args = (char *) args_;
2544 char *save_ptr = NULL;
2545 char *bond_s, *hash_s, *slave_s;
2546 uint8_t mac[ETH_ADDR_LEN];
2548 struct iface *iface;
2549 struct bond_entry *entry;
2552 bond_s = strtok_r(args, " ", &save_ptr);
2553 hash_s = strtok_r(NULL, " ", &save_ptr);
2554 slave_s = strtok_r(NULL, " ", &save_ptr);
2556 unixctl_command_reply(conn, 501,
2557 "usage: bond/migrate BOND HASH SLAVE");
2561 port = bond_find(bond_s);
2563 unixctl_command_reply(conn, 501, "no such bond");
2567 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2568 == ETH_ADDR_SCAN_COUNT) {
2569 hash = bond_hash(mac);
2570 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2571 hash = atoi(hash_s) & BOND_MASK;
2573 unixctl_command_reply(conn, 501, "bad hash");
2577 iface = port_lookup_iface(port, slave_s);
2579 unixctl_command_reply(conn, 501, "no such slave");
2583 if (!iface->enabled) {
2584 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2588 entry = &port->bond_hash[hash];
2589 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2590 entry->iface_idx = iface->port_ifidx;
2591 entry->iface_tag = tag_create_random();
2592 port->bond_compat_is_stale = true;
2593 unixctl_command_reply(conn, 200, "migrated");
2597 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2599 char *args = (char *) args_;
2600 char *save_ptr = NULL;
2601 char *bond_s, *slave_s;
2603 struct iface *iface;
2605 bond_s = strtok_r(args, " ", &save_ptr);
2606 slave_s = strtok_r(NULL, " ", &save_ptr);
2608 unixctl_command_reply(conn, 501,
2609 "usage: bond/set-active-slave BOND SLAVE");
2613 port = bond_find(bond_s);
2615 unixctl_command_reply(conn, 501, "no such bond");
2619 iface = port_lookup_iface(port, slave_s);
2621 unixctl_command_reply(conn, 501, "no such slave");
2625 if (!iface->enabled) {
2626 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2630 if (port->active_iface != iface->port_ifidx) {
2631 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2632 port->active_iface = iface->port_ifidx;
2633 port->active_iface_tag = tag_create_random();
2634 VLOG_INFO("port %s: active interface is now %s",
2635 port->name, iface->name);
2636 bond_send_learning_packets(port);
2637 unixctl_command_reply(conn, 200, "done");
2639 unixctl_command_reply(conn, 200, "no change");
2644 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2646 char *args = (char *) args_;
2647 char *save_ptr = NULL;
2648 char *bond_s, *slave_s;
2650 struct iface *iface;
2652 bond_s = strtok_r(args, " ", &save_ptr);
2653 slave_s = strtok_r(NULL, " ", &save_ptr);
2655 unixctl_command_reply(conn, 501,
2656 "usage: bond/enable/disable-slave BOND SLAVE");
2660 port = bond_find(bond_s);
2662 unixctl_command_reply(conn, 501, "no such bond");
2666 iface = port_lookup_iface(port, slave_s);
2668 unixctl_command_reply(conn, 501, "no such slave");
2672 bond_enable_slave(iface, enable);
2673 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2677 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2679 enable_slave(conn, args, true);
2683 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2685 enable_slave(conn, args, false);
2691 unixctl_command_register("bond/list", bond_unixctl_list);
2692 unixctl_command_register("bond/show", bond_unixctl_show);
2693 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2694 unixctl_command_register("bond/set-active-slave",
2695 bond_unixctl_set_active_slave);
2696 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2697 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2700 /* Port functions. */
2703 port_create(struct bridge *br, const char *name)
2707 port = xcalloc(1, sizeof *port);
2709 port->port_idx = br->n_ports;
2711 port->trunks = NULL;
2712 port->name = xstrdup(name);
2713 port->active_iface = -1;
2714 port->stp_state = STP_DISABLED;
2715 port->stp_state_tag = 0;
2717 if (br->n_ports >= br->allocated_ports) {
2718 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2721 br->ports[br->n_ports++] = port;
2723 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2728 port_reconfigure(struct port *port)
2730 bool bonded = cfg_has_section("bonding.%s", port->name);
2731 struct svec old_ifaces, new_ifaces;
2732 unsigned long *trunks;
2736 /* Collect old and new interfaces. */
2737 svec_init(&old_ifaces);
2738 svec_init(&new_ifaces);
2739 for (i = 0; i < port->n_ifaces; i++) {
2740 svec_add(&old_ifaces, port->ifaces[i]->name);
2742 svec_sort(&old_ifaces);
2744 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2745 if (!new_ifaces.n) {
2746 VLOG_ERR("port %s: no interfaces specified for bonded port",
2748 } else if (new_ifaces.n == 1) {
2749 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2753 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2754 if (port->updelay < 0) {
2757 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2758 if (port->downdelay < 0) {
2759 port->downdelay = 0;
2762 svec_init(&new_ifaces);
2763 svec_add(&new_ifaces, port->name);
2766 /* Get rid of deleted interfaces and add new interfaces. */
2767 for (i = 0; i < port->n_ifaces; i++) {
2768 struct iface *iface = port->ifaces[i];
2769 if (!svec_contains(&new_ifaces, iface->name)) {
2770 iface_destroy(iface);
2775 for (i = 0; i < new_ifaces.n; i++) {
2776 const char *name = new_ifaces.names[i];
2777 if (!svec_contains(&old_ifaces, name)) {
2778 iface_create(port, name);
2784 if (cfg_has("vlan.%s.tag", port->name)) {
2786 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2787 if (vlan >= 0 && vlan <= 4095) {
2788 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2791 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2792 * they even work as-is. But they have not been tested. */
2793 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2797 if (port->vlan != vlan) {
2799 bridge_flush(port->bridge);
2802 /* Get trunked VLANs. */
2805 size_t n_trunks, n_errors;
2808 trunks = bitmap_allocate(4096);
2809 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2811 for (i = 0; i < n_trunks; i++) {
2812 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2814 bitmap_set1(trunks, trunk);
2820 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2821 port->name, n_trunks);
2823 if (n_errors == n_trunks) {
2825 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2828 bitmap_set_multiple(trunks, 0, 4096, 1);
2831 if (cfg_has("vlan.%s.trunks", port->name)) {
2832 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2833 port->name, port->name);
2837 ? port->trunks != NULL
2838 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2839 bridge_flush(port->bridge);
2841 bitmap_free(port->trunks);
2842 port->trunks = trunks;
2844 svec_destroy(&old_ifaces);
2845 svec_destroy(&new_ifaces);
2849 port_destroy(struct port *port)
2852 struct bridge *br = port->bridge;
2856 proc_net_compat_update_vlan(port->name, NULL, 0);
2857 proc_net_compat_update_bond(port->name, NULL);
2859 for (i = 0; i < MAX_MIRRORS; i++) {
2860 struct mirror *m = br->mirrors[i];
2861 if (m && m->out_port == port) {
2866 while (port->n_ifaces > 0) {
2867 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2870 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2871 del->port_idx = port->port_idx;
2874 bitmap_free(port->trunks);
2881 static struct port *
2882 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2884 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2885 return iface ? iface->port : NULL;
2888 static struct port *
2889 port_lookup(const struct bridge *br, const char *name)
2893 for (i = 0; i < br->n_ports; i++) {
2894 struct port *port = br->ports[i];
2895 if (!strcmp(port->name, name)) {
2902 static struct iface *
2903 port_lookup_iface(const struct port *port, const char *name)
2907 for (j = 0; j < port->n_ifaces; j++) {
2908 struct iface *iface = port->ifaces[j];
2909 if (!strcmp(iface->name, name)) {
2917 port_update_bonding(struct port *port)
2919 if (port->n_ifaces < 2) {
2920 /* Not a bonded port. */
2921 if (port->bond_hash) {
2922 free(port->bond_hash);
2923 port->bond_hash = NULL;
2924 port->bond_compat_is_stale = true;
2927 if (!port->bond_hash) {
2930 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2931 for (i = 0; i <= BOND_MASK; i++) {
2932 struct bond_entry *e = &port->bond_hash[i];
2936 port->no_ifaces_tag = tag_create_random();
2937 bond_choose_active_iface(port);
2939 port->bond_compat_is_stale = true;
2944 port_update_bond_compat(struct port *port)
2946 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
2947 struct compat_bond bond;
2950 if (port->n_ifaces < 2) {
2951 proc_net_compat_update_bond(port->name, NULL);
2956 bond.updelay = port->updelay;
2957 bond.downdelay = port->downdelay;
2960 bond.hashes = compat_hashes;
2961 if (port->bond_hash) {
2962 const struct bond_entry *e;
2963 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
2964 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2965 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
2966 cbh->hash = e - port->bond_hash;
2967 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
2972 bond.n_slaves = port->n_ifaces;
2973 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2974 for (i = 0; i < port->n_ifaces; i++) {
2975 struct iface *iface = port->ifaces[i];
2976 struct compat_bond_slave *slave = &bond.slaves[i];
2977 slave->name = iface->name;
2978 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2979 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2983 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2986 proc_net_compat_update_bond(port->name, &bond);
2991 port_update_vlan_compat(struct port *port)
2993 struct bridge *br = port->bridge;
2994 char *vlandev_name = NULL;
2996 if (port->vlan > 0) {
2997 /* Figure out the name that the VLAN device should actually have, if it
2998 * existed. This takes some work because the VLAN device would not
2999 * have port->name in its name; rather, it would have the trunk port's
3000 * name, and 'port' would be attached to a bridge that also had the
3001 * VLAN device one of its ports. So we need to find a trunk port that
3002 * includes port->vlan.
3004 * There might be more than one candidate. This doesn't happen on
3005 * XenServer, so if it happens we just pick the first choice in
3006 * alphabetical order instead of creating multiple VLAN devices. */
3008 for (i = 0; i < br->n_ports; i++) {
3009 struct port *p = br->ports[i];
3010 if (port_trunks_vlan(p, port->vlan)
3012 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3014 const uint8_t *ea = p->ifaces[0]->mac;
3015 if (!eth_addr_is_multicast(ea) &&
3016 !eth_addr_is_reserved(ea) &&
3017 !eth_addr_is_zero(ea)) {
3018 vlandev_name = p->name;
3023 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3026 /* Interface functions. */
3029 iface_create(struct port *port, const char *name)
3031 struct iface *iface;
3033 iface = xcalloc(1, sizeof *iface);
3035 iface->port_ifidx = port->n_ifaces;
3036 iface->name = xstrdup(name);
3037 iface->dp_ifidx = -1;
3038 iface->tag = tag_create_random();
3039 iface->delay_expires = LLONG_MAX;
3041 if (!cfg_get_bool(0, "iface.%s.internal", iface->name)) {
3042 netdev_nodev_get_etheraddr(name, iface->mac);
3043 netdev_nodev_get_carrier(name, &iface->enabled);
3045 /* Internal interfaces are created later by the call to dpif_port_add()
3046 * in bridge_reconfigure(). Until then, we can't obtain any
3047 * information about them. (There's no real value in doing so, anyway,
3048 * because the 'mac' and 'enabled' values are only used for interfaces
3049 * that are bond slaves, and it doesn't normally make sense to bond an
3050 * internal interface.) */
3053 if (port->n_ifaces >= port->allocated_ifaces) {
3054 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3055 sizeof *port->ifaces);
3057 port->ifaces[port->n_ifaces++] = iface;
3058 if (port->n_ifaces > 1) {
3059 port->bridge->has_bonded_ports = true;
3062 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3064 port_update_bonding(port);
3065 bridge_flush(port->bridge);
3069 iface_destroy(struct iface *iface)
3072 struct port *port = iface->port;
3073 struct bridge *br = port->bridge;
3074 bool del_active = port->active_iface == iface->port_ifidx;
3077 if (iface->dp_ifidx >= 0) {
3078 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3081 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3082 del->port_ifidx = iface->port_ifidx;
3088 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3089 bond_choose_active_iface(port);
3090 bond_send_learning_packets(port);
3093 port_update_bonding(port);
3094 bridge_flush(port->bridge);
3098 static struct iface *
3099 iface_lookup(const struct bridge *br, const char *name)
3103 for (i = 0; i < br->n_ports; i++) {
3104 struct port *port = br->ports[i];
3105 for (j = 0; j < port->n_ifaces; j++) {
3106 struct iface *iface = port->ifaces[j];
3107 if (!strcmp(iface->name, name)) {
3115 static struct iface *
3116 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3118 return port_array_get(&br->ifaces, dp_ifidx);
3121 /* Port mirroring. */
3124 mirror_reconfigure(struct bridge *br)
3126 struct svec old_mirrors, new_mirrors;
3129 /* Collect old and new mirrors. */
3130 svec_init(&old_mirrors);
3131 svec_init(&new_mirrors);
3132 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3133 for (i = 0; i < MAX_MIRRORS; i++) {
3134 if (br->mirrors[i]) {
3135 svec_add(&old_mirrors, br->mirrors[i]->name);
3139 /* Get rid of deleted mirrors and add new mirrors. */
3140 svec_sort(&old_mirrors);
3141 assert(svec_is_unique(&old_mirrors));
3142 svec_sort(&new_mirrors);
3143 assert(svec_is_unique(&new_mirrors));
3144 for (i = 0; i < MAX_MIRRORS; i++) {
3145 struct mirror *m = br->mirrors[i];
3146 if (m && !svec_contains(&new_mirrors, m->name)) {
3150 for (i = 0; i < new_mirrors.n; i++) {
3151 const char *name = new_mirrors.names[i];
3152 if (!svec_contains(&old_mirrors, name)) {
3153 mirror_create(br, name);
3156 svec_destroy(&old_mirrors);
3157 svec_destroy(&new_mirrors);
3159 /* Reconfigure all mirrors. */
3160 for (i = 0; i < MAX_MIRRORS; i++) {
3161 if (br->mirrors[i]) {
3162 mirror_reconfigure_one(br->mirrors[i]);
3166 /* Update port reserved status. */
3167 for (i = 0; i < br->n_ports; i++) {
3168 br->ports[i]->is_mirror_output_port = false;
3170 for (i = 0; i < MAX_MIRRORS; i++) {
3171 struct mirror *m = br->mirrors[i];
3172 if (m && m->out_port) {
3173 m->out_port->is_mirror_output_port = true;
3179 mirror_create(struct bridge *br, const char *name)
3184 for (i = 0; ; i++) {
3185 if (i >= MAX_MIRRORS) {
3186 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3187 "cannot create %s", br->name, MAX_MIRRORS, name);
3190 if (!br->mirrors[i]) {
3195 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3198 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3201 m->name = xstrdup(name);
3202 svec_init(&m->src_ports);
3203 svec_init(&m->dst_ports);
3211 mirror_destroy(struct mirror *m)
3214 struct bridge *br = m->bridge;
3217 for (i = 0; i < br->n_ports; i++) {
3218 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3219 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3222 svec_destroy(&m->src_ports);
3223 svec_destroy(&m->dst_ports);
3226 m->bridge->mirrors[m->idx] = NULL;
3234 prune_ports(struct mirror *m, struct svec *ports)
3239 svec_sort_unique(ports);
3242 for (i = 0; i < ports->n; i++) {
3243 const char *name = ports->names[i];
3244 if (port_lookup(m->bridge, name)) {
3245 svec_add(&tmp, name);
3247 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3248 m->bridge->name, m->name, name);
3251 svec_swap(ports, &tmp);
3256 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3260 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3261 * order won't give us numeric sort order. But that's good enough for what
3262 * we need right now. */
3263 svec_sort_unique(vlan_strings);
3265 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3267 for (i = 0; i < vlan_strings->n; i++) {
3268 const char *name = vlan_strings->names[i];
3270 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3271 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3272 m->bridge->name, m->name, name);
3274 (*vlans)[n_vlans++] = vlan;
3281 vlan_is_mirrored(const struct mirror *m, int vlan)
3285 for (i = 0; i < m->n_vlans; i++) {
3286 if (m->vlans[i] == vlan) {
3294 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3298 for (i = 0; i < m->n_vlans; i++) {
3299 if (port_trunks_vlan(p, m->vlans[i])) {
3307 mirror_reconfigure_one(struct mirror *m)
3309 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3310 struct svec src_ports, dst_ports, ports;
3311 struct svec vlan_strings;
3312 mirror_mask_t mirror_bit;
3313 const char *out_port_name;
3314 struct port *out_port;
3319 bool mirror_all_ports;
3321 /* Get output port. */
3322 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3323 m->bridge->name, m->name);
3324 if (out_port_name) {
3325 out_port = port_lookup(m->bridge, out_port_name);
3327 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3328 "named %s", pfx, m->bridge->name, out_port_name);
3335 if (cfg_has("%s.output.vlan", pfx)) {
3336 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3337 "ignoring %s.output.vlan", pfx, pfx, pfx);
3339 } else if (cfg_has("%s.output.vlan", pfx)) {
3341 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3343 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3344 "but exactly one is required; disabling port mirror %s",
3345 pfx, pfx, pfx, pfx);
3351 /* Get all the ports, and drop duplicates and ports that don't exist. */
3352 svec_init(&src_ports);
3353 svec_init(&dst_ports);
3355 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3356 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3357 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3358 svec_append(&src_ports, &ports);
3359 svec_append(&dst_ports, &ports);
3360 svec_destroy(&ports);
3361 prune_ports(m, &src_ports);
3362 prune_ports(m, &dst_ports);
3364 /* Get all the vlans, and drop duplicate and invalid vlans. */
3365 svec_init(&vlan_strings);
3366 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3367 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3368 svec_destroy(&vlan_strings);
3370 /* Update mirror data. */
3371 if (!svec_equal(&m->src_ports, &src_ports)
3372 || !svec_equal(&m->dst_ports, &dst_ports)
3373 || m->n_vlans != n_vlans
3374 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3375 || m->out_port != out_port
3376 || m->out_vlan != out_vlan) {
3377 bridge_flush(m->bridge);
3379 svec_swap(&m->src_ports, &src_ports);
3380 svec_swap(&m->dst_ports, &dst_ports);
3383 m->n_vlans = n_vlans;
3384 m->out_port = out_port;
3385 m->out_vlan = out_vlan;
3387 /* If no selection criteria have been given, mirror for all ports. */
3388 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3391 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3392 for (i = 0; i < m->bridge->n_ports; i++) {
3393 struct port *port = m->bridge->ports[i];
3395 if (mirror_all_ports
3396 || svec_contains(&m->src_ports, port->name)
3399 ? port_trunks_any_mirrored_vlan(m, port)
3400 : vlan_is_mirrored(m, port->vlan)))) {
3401 port->src_mirrors |= mirror_bit;
3403 port->src_mirrors &= ~mirror_bit;
3406 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3407 port->dst_mirrors |= mirror_bit;
3409 port->dst_mirrors &= ~mirror_bit;
3414 svec_destroy(&src_ports);
3415 svec_destroy(&dst_ports);
3419 /* Spanning tree protocol. */
3421 static void brstp_update_port_state(struct port *);
3424 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3426 struct bridge *br = br_;
3427 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3428 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3430 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3432 } else if (eth_addr_is_zero(iface->mac)) {
3433 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3436 union ofp_action action;
3437 struct eth_header *eth = pkt->l2;
3440 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3442 memset(&action, 0, sizeof action);
3443 action.type = htons(OFPAT_OUTPUT);
3444 action.output.len = htons(sizeof action);
3445 action.output.port = htons(port_no);
3447 flow_extract(pkt, ODPP_NONE, &flow);
3448 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3454 brstp_reconfigure(struct bridge *br)
3458 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3460 stp_destroy(br->stp);
3466 uint64_t bridge_address, bridge_id;
3467 int bridge_priority;
3469 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3470 if (!bridge_address) {
3472 bridge_address = (stp_get_bridge_id(br->stp)
3473 & ((UINT64_C(1) << 48) - 1));
3475 uint8_t mac[ETH_ADDR_LEN];
3476 eth_addr_random(mac);
3477 bridge_address = eth_addr_to_uint64(mac);
3481 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3483 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3485 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3488 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3490 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3491 br->stp_last_tick = time_msec();
3494 if (bridge_id != stp_get_bridge_id(br->stp)) {
3495 stp_set_bridge_id(br->stp, bridge_id);
3500 for (i = 0; i < br->n_ports; i++) {
3501 struct port *p = br->ports[i];
3503 struct stp_port *sp;
3504 int path_cost, priority;
3510 dp_ifidx = p->ifaces[0]->dp_ifidx;
3511 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3515 sp = stp_get_port(br->stp, dp_ifidx);
3516 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3517 "stp.%s.port.%s.enabled",
3519 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3520 br->name, p->name));
3521 if (p->is_mirror_output_port) {
3524 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3525 bridge_flush(br); /* Might not be necessary. */
3527 stp_port_enable(sp);
3529 stp_port_disable(sp);
3533 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3535 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3537 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3538 "stp.%s.port.%s.priority",
3540 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3542 : STP_DEFAULT_PORT_PRIORITY);
3543 stp_port_set_priority(sp, priority);
3546 brstp_adjust_timers(br);
3548 for (i = 0; i < br->n_ports; i++) {
3549 brstp_update_port_state(br->ports[i]);
3554 brstp_update_port_state(struct port *p)
3556 struct bridge *br = p->bridge;
3557 enum stp_state state;
3559 /* Figure out new state. */
3560 state = STP_DISABLED;
3561 if (br->stp && p->n_ifaces > 0) {
3562 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3563 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3564 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3569 if (p->stp_state != state) {
3570 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3571 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3572 p->name, stp_state_name(p->stp_state),
3573 stp_state_name(state));
3574 if (p->stp_state == STP_DISABLED) {
3577 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3579 p->stp_state = state;
3580 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3581 : tag_create_random());
3586 brstp_adjust_timers(struct bridge *br)
3588 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3589 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3590 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3592 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3593 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3594 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3598 brstp_run(struct bridge *br)
3601 long long int now = time_msec();
3602 long long int elapsed = now - br->stp_last_tick;
3603 struct stp_port *sp;
3606 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3607 br->stp_last_tick = now;
3609 while (stp_get_changed_port(br->stp, &sp)) {
3610 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3612 brstp_update_port_state(p);
3619 brstp_wait(struct bridge *br)
3622 poll_timer_wait(1000);