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"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.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 /* These members are always valid. */
75 struct port *port; /* Containing port. */
76 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 tag_type tag; /* Tag associated with this interface. */
79 long long delay_expires; /* Time after which 'enabled' may change. */
81 /* These members are valid only after bridge_reconfigure() causes them to
83 int dp_ifidx; /* Index within kernel datapath. */
84 struct netdev *netdev; /* Network device. */
85 bool enabled; /* May be chosen for flows? */
88 #define BOND_MASK 0xff
90 int iface_idx; /* Index of assigned iface, or -1 if none. */
91 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
92 tag_type iface_tag; /* Tag associated with iface_idx. */
95 #define MAX_MIRRORS 32
96 typedef uint32_t mirror_mask_t;
97 #define MIRROR_MASK_C(X) UINT32_C(X)
98 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
100 struct bridge *bridge;
104 /* Selection criteria. */
105 struct svec src_ports;
106 struct svec dst_ports;
111 struct port *out_port;
115 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
117 struct bridge *bridge;
119 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
120 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
123 /* An ordinary bridge port has 1 interface.
124 * A bridge port for bonding has at least 2 interfaces. */
125 struct iface **ifaces;
126 size_t n_ifaces, allocated_ifaces;
129 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
130 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
131 tag_type active_iface_tag; /* Tag for bcast flows. */
132 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
133 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
134 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
136 /* Port mirroring info. */
137 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
138 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
139 bool is_mirror_output_port; /* Does port mirroring send frames here? */
141 /* Spanning tree info. */
142 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
143 tag_type stp_state_tag; /* Tag for STP state change. */
146 #define DP_MAX_PORTS 255
148 struct list node; /* Node in global list of bridges. */
149 char *name; /* User-specified arbitrary name. */
150 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
151 bool sent_config_request; /* Successfully sent config request? */
152 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
154 /* Support for remote controllers. */
155 char *controller; /* NULL if there is no remote controller;
156 * "discover" to do controller discovery;
157 * otherwise a vconn name. */
159 /* OpenFlow switch processing. */
160 struct ofproto *ofproto; /* OpenFlow switch. */
162 /* Kernel datapath information. */
163 struct dpif *dpif; /* Datapath. */
164 struct port_array ifaces; /* Indexed by kernel datapath port number. */
168 size_t n_ports, allocated_ports;
171 bool has_bonded_ports;
172 long long int bond_next_rebalance;
177 /* Flow statistics gathering. */
178 time_t next_stats_request;
180 /* Port mirroring. */
181 struct mirror *mirrors[MAX_MIRRORS];
185 long long int stp_last_tick;
188 /* List of all bridges. */
189 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
191 /* Maximum number of datapaths. */
192 enum { DP_MAX = 256 };
194 static struct bridge *bridge_create(const char *name);
195 static void bridge_destroy(struct bridge *);
196 static struct bridge *bridge_lookup(const char *name);
197 static void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *);
198 static int bridge_run_one(struct bridge *);
199 static void bridge_reconfigure_one(struct bridge *);
200 static void bridge_reconfigure_controller(struct bridge *);
201 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
202 static void bridge_fetch_dp_ifaces(struct bridge *);
203 static void bridge_flush(struct bridge *);
204 static void bridge_pick_local_hw_addr(struct bridge *,
205 uint8_t ea[ETH_ADDR_LEN],
206 struct iface **hw_addr_iface);
207 static uint64_t bridge_pick_datapath_id(struct bridge *,
208 const uint8_t bridge_ea[ETH_ADDR_LEN],
209 struct iface *hw_addr_iface);
210 static struct iface *bridge_get_local_iface(struct bridge *);
211 static uint64_t dpid_from_hash(const void *, size_t nbytes);
213 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
215 static void bond_init(void);
216 static void bond_run(struct bridge *);
217 static void bond_wait(struct bridge *);
218 static void bond_rebalance_port(struct port *);
219 static void bond_send_learning_packets(struct port *);
221 static void port_create(struct bridge *, const char *name);
222 static void port_reconfigure(struct port *);
223 static void port_destroy(struct port *);
224 static struct port *port_lookup(const struct bridge *, const char *name);
225 static struct iface *port_lookup_iface(const struct port *, const char *name);
226 static struct port *port_from_dp_ifidx(const struct bridge *,
228 static void port_update_bond_compat(struct port *);
229 static void port_update_vlan_compat(struct port *);
230 static void port_update_bonding(struct port *);
232 static void mirror_create(struct bridge *, const char *name);
233 static void mirror_destroy(struct mirror *);
234 static void mirror_reconfigure(struct bridge *);
235 static void mirror_reconfigure_one(struct mirror *);
236 static bool vlan_is_mirrored(const struct mirror *, int vlan);
238 static void brstp_reconfigure(struct bridge *);
239 static void brstp_adjust_timers(struct bridge *);
240 static void brstp_run(struct bridge *);
241 static void brstp_wait(struct bridge *);
243 static void iface_create(struct port *, const char *name);
244 static void iface_destroy(struct iface *);
245 static struct iface *iface_lookup(const struct bridge *, const char *name);
246 static struct iface *iface_from_dp_ifidx(const struct bridge *,
248 static bool iface_is_internal(const struct bridge *, const char *name);
249 static void iface_set_mac(struct iface *);
251 /* Hooks into ofproto processing. */
252 static struct ofhooks bridge_ofhooks;
254 /* Public functions. */
256 /* Adds the name of each interface used by a bridge, including local and
257 * internal ports, to 'svec'. */
259 bridge_get_ifaces(struct svec *svec)
261 struct bridge *br, *next;
264 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
265 for (i = 0; i < br->n_ports; i++) {
266 struct port *port = br->ports[i];
268 for (j = 0; j < port->n_ifaces; j++) {
269 struct iface *iface = port->ifaces[j];
270 if (iface->dp_ifidx < 0) {
271 VLOG_ERR("%s interface not in datapath %s, ignoring",
272 iface->name, dpif_name(br->dpif));
274 if (iface->dp_ifidx != ODPP_LOCAL) {
275 svec_add(svec, iface->name);
283 /* The caller must already have called cfg_read(). */
287 struct svec dpif_names;
290 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
292 svec_init(&dpif_names);
293 dp_enumerate(&dpif_names);
294 for (i = 0; i < dpif_names.n; i++) {
295 const char *dpif_name = dpif_names.names[i];
299 retval = dpif_open(dpif_name, &dpif);
301 struct svec all_names;
304 svec_init(&all_names);
305 dpif_get_all_names(dpif, &all_names);
306 for (j = 0; j < all_names.n; j++) {
307 if (cfg_has("bridge.%s.port", all_names.names[j])) {
313 svec_destroy(&all_names);
317 svec_destroy(&dpif_names);
319 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows);
322 bridge_reconfigure();
327 config_string_change(const char *key, char **valuep)
329 const char *value = cfg_get_string(0, "%s", key);
330 if (value && (!*valuep || strcmp(value, *valuep))) {
332 *valuep = xstrdup(value);
340 bridge_configure_ssl(void)
342 /* XXX SSL should be configurable on a per-bridge basis.
343 * XXX should be possible to de-configure SSL. */
344 static char *private_key_file;
345 static char *certificate_file;
346 static char *cacert_file;
349 if (config_string_change("ssl.private-key", &private_key_file)) {
350 vconn_ssl_set_private_key_file(private_key_file);
353 if (config_string_change("ssl.certificate", &certificate_file)) {
354 vconn_ssl_set_certificate_file(certificate_file);
357 /* We assume that even if the filename hasn't changed, if the CA cert
358 * file has been removed, that we want to move back into
359 * boot-strapping mode. This opens a small security hole, because
360 * the old certificate will still be trusted until vSwitch is
361 * restarted. We may want to address this in vconn's SSL library. */
362 if (config_string_change("ssl.ca-cert", &cacert_file)
363 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
364 vconn_ssl_set_ca_cert_file(cacert_file,
365 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
370 /* iterate_and_prune_ifaces() callback function that opens the network device
371 * for 'iface', if it is not already open, and retrieves the interface's MAC
372 * address and carrier status. */
374 init_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
379 } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE,
381 netdev_get_carrier(iface->netdev, &iface->enabled);
384 /* If the network device can't be opened, then we're not going to try
385 * to do anything with this interface. */
391 check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED)
393 if (iface->dp_ifidx >= 0) {
394 VLOG_DBG("%s has interface %s on port %d",
396 iface->name, iface->dp_ifidx);
399 VLOG_ERR("%s interface not in %s, dropping",
400 iface->name, dpif_name(br->dpif));
406 set_iface_properties(struct bridge *br UNUSED, struct iface *iface,
411 /* Set policing attributes. */
412 rate = cfg_get_int(0, "port.%s.ingress.policing-rate", iface->name);
413 burst = cfg_get_int(0, "port.%s.ingress.policing-burst", iface->name);
414 netdev_set_policing(iface->netdev, rate, burst);
416 /* Set MAC address of internal interfaces other than the local
418 if (iface->dp_ifidx != ODPP_LOCAL
419 && iface_is_internal(br, iface->name)) {
420 iface_set_mac(iface);
426 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
427 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
428 * deletes from 'br' any ports that no longer have any interfaces. */
430 iterate_and_prune_ifaces(struct bridge *br,
431 bool (*cb)(struct bridge *, struct iface *,
437 for (i = 0; i < br->n_ports; ) {
438 struct port *port = br->ports[i];
439 for (j = 0; j < port->n_ifaces; ) {
440 struct iface *iface = port->ifaces[j];
441 if (cb(br, iface, aux)) {
444 iface_destroy(iface);
448 if (port->n_ifaces) {
451 VLOG_ERR("%s port has no interfaces, dropping", port->name);
458 bridge_reconfigure(void)
460 struct svec old_br, new_br;
461 struct bridge *br, *next;
464 COVERAGE_INC(bridge_reconfigure);
466 /* Collect old and new bridges. */
469 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
470 svec_add(&old_br, br->name);
472 cfg_get_subsections(&new_br, "bridge");
474 /* Get rid of deleted bridges and add new bridges. */
477 assert(svec_is_unique(&old_br));
478 assert(svec_is_unique(&new_br));
479 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
480 if (!svec_contains(&new_br, br->name)) {
484 for (i = 0; i < new_br.n; i++) {
485 const char *name = new_br.names[i];
486 if (!svec_contains(&old_br, name)) {
490 svec_destroy(&old_br);
491 svec_destroy(&new_br);
495 bridge_configure_ssl();
498 /* Reconfigure all bridges. */
499 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
500 bridge_reconfigure_one(br);
503 /* Add and delete ports on all datapaths.
505 * The kernel will reject any attempt to add a given port to a datapath if
506 * that port already belongs to a different datapath, so we must do all
507 * port deletions before any port additions. */
508 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
509 struct odp_port *dpif_ports;
511 struct svec want_ifaces;
513 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
514 bridge_get_all_ifaces(br, &want_ifaces);
515 for (i = 0; i < n_dpif_ports; i++) {
516 const struct odp_port *p = &dpif_ports[i];
517 if (!svec_contains(&want_ifaces, p->devname)
518 && strcmp(p->devname, br->name)) {
519 int retval = dpif_port_del(br->dpif, p->port);
521 VLOG_ERR("failed to remove %s interface from %s: %s",
522 p->devname, dpif_name(br->dpif),
527 svec_destroy(&want_ifaces);
530 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
531 struct odp_port *dpif_ports;
533 struct svec cur_ifaces, want_ifaces, add_ifaces;
535 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
536 svec_init(&cur_ifaces);
537 for (i = 0; i < n_dpif_ports; i++) {
538 svec_add(&cur_ifaces, dpif_ports[i].devname);
541 svec_sort_unique(&cur_ifaces);
542 bridge_get_all_ifaces(br, &want_ifaces);
543 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
545 for (i = 0; i < add_ifaces.n; i++) {
546 const char *if_name = add_ifaces.names[i];
550 /* Add to datapath. */
551 internal = iface_is_internal(br, if_name);
552 error = dpif_port_add(br->dpif, if_name,
553 internal ? ODP_PORT_INTERNAL : 0, NULL);
554 if (error == EFBIG) {
555 VLOG_ERR("ran out of valid port numbers on %s",
556 dpif_name(br->dpif));
559 VLOG_ERR("failed to add %s interface to %s: %s",
560 if_name, dpif_name(br->dpif), strerror(error));
563 svec_destroy(&cur_ifaces);
564 svec_destroy(&want_ifaces);
565 svec_destroy(&add_ifaces);
567 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
570 struct iface *local_iface;
571 struct iface *hw_addr_iface;
572 uint8_t engine_type, engine_id;
573 bool add_id_to_iface = false;
574 struct svec nf_hosts;
576 bridge_fetch_dp_ifaces(br);
577 iterate_and_prune_ifaces(br, init_iface_netdev, NULL);
579 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
581 /* Pick local port hardware address, datapath ID. */
582 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
583 local_iface = bridge_get_local_iface(br);
585 int error = netdev_set_etheraddr(local_iface->netdev, ea);
587 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
588 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
589 "Ethernet address: %s",
590 br->name, strerror(error));
594 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
595 ofproto_set_datapath_id(br->ofproto, dpid);
597 /* Set NetFlow configuration on this bridge. */
598 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
599 if (cfg_has("netflow.%s.engine-type", br->name)) {
600 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
603 if (cfg_has("netflow.%s.engine-id", br->name)) {
604 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
606 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
607 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
610 if (add_id_to_iface && engine_id > 0x7f) {
611 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
612 "another vswitch, choose an engine id less than 128",
615 if (add_id_to_iface && br->n_ports > 0x1ff) {
616 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
617 "another port when 512 or more ports are used",
620 svec_init(&nf_hosts);
621 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
622 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
623 engine_id, add_id_to_iface)) {
624 VLOG_ERR("bridge %s: problem setting netflow collectors",
627 svec_destroy(&nf_hosts);
629 /* Update the controller and related settings. It would be more
630 * straightforward to call this from bridge_reconfigure_one(), but we
631 * can't do it there for two reasons. First, and most importantly, at
632 * that point we don't know the dp_ifidx of any interfaces that have
633 * been added to the bridge (because we haven't actually added them to
634 * the datapath). Second, at that point we haven't set the datapath ID
635 * yet; when a controller is configured, resetting the datapath ID will
636 * immediately disconnect from the controller, so it's better to set
637 * the datapath ID before the controller. */
638 bridge_reconfigure_controller(br);
640 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
641 for (i = 0; i < br->n_ports; i++) {
642 struct port *port = br->ports[i];
644 port_update_vlan_compat(port);
645 port_update_bonding(port);
648 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
649 brstp_reconfigure(br);
650 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
655 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
656 struct iface **hw_addr_iface)
658 uint64_t requested_ea;
662 *hw_addr_iface = NULL;
664 /* Did the user request a particular MAC? */
665 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
667 eth_addr_from_uint64(requested_ea, ea);
668 if (eth_addr_is_multicast(ea)) {
669 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
670 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
671 } else if (eth_addr_is_zero(ea)) {
672 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
678 /* Otherwise choose the minimum MAC address among all of the interfaces.
679 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
680 * MAC of the physical interface in such an environment.) */
681 memset(ea, 0xff, sizeof ea);
682 for (i = 0; i < br->n_ports; i++) {
683 struct port *port = br->ports[i];
684 uint8_t iface_ea[ETH_ADDR_LEN];
685 uint64_t iface_ea_u64;
688 /* Mirror output ports don't participate. */
689 if (port->is_mirror_output_port) {
693 /* Choose the MAC address to represent the port. */
694 iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name);
696 /* User specified explicitly. */
697 eth_addr_from_uint64(iface_ea_u64, iface_ea);
699 /* Find the interface with this Ethernet address (if any) so that
700 * we can provide the correct devname to the caller. */
702 for (j = 0; j < port->n_ifaces; j++) {
703 struct iface *candidate = port->ifaces[j];
704 uint8_t candidate_ea[ETH_ADDR_LEN];
705 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
706 && eth_addr_equals(iface_ea, candidate_ea)) {
711 /* Choose the interface whose MAC address will represent the port.
712 * The Linux kernel bonding code always chooses the MAC address of
713 * the first slave added to a bond, and the Fedora networking
714 * scripts always add slaves to a bond in alphabetical order, so
715 * for compatibility we choose the interface with the name that is
716 * first in alphabetical order. */
717 iface = port->ifaces[0];
718 for (j = 1; j < port->n_ifaces; j++) {
719 struct iface *candidate = port->ifaces[j];
720 if (strcmp(candidate->name, iface->name) < 0) {
725 /* The local port doesn't count (since we're trying to choose its
726 * MAC address anyway). Other internal ports don't count because
727 * we really want a physical MAC if we can get it, and internal
728 * ports typically have randomly generated MACs. */
729 if (iface->dp_ifidx == ODPP_LOCAL
730 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
735 error = netdev_get_etheraddr(iface->netdev, iface_ea);
737 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
738 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
739 iface->name, strerror(error));
744 /* Compare against our current choice. */
745 if (!eth_addr_is_multicast(iface_ea) &&
746 !eth_addr_is_reserved(iface_ea) &&
747 !eth_addr_is_zero(iface_ea) &&
748 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
750 memcpy(ea, iface_ea, ETH_ADDR_LEN);
751 *hw_addr_iface = iface;
754 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
755 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
756 *hw_addr_iface = NULL;
757 VLOG_WARN("bridge %s: using default bridge Ethernet "
758 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
760 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
761 br->name, ETH_ADDR_ARGS(ea));
765 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
766 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
767 * an interface on 'br', then that interface must be passed in as
768 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
769 * 'hw_addr_iface' must be passed in as a null pointer. */
771 bridge_pick_datapath_id(struct bridge *br,
772 const uint8_t bridge_ea[ETH_ADDR_LEN],
773 struct iface *hw_addr_iface)
776 * The procedure for choosing a bridge MAC address will, in the most
777 * ordinary case, also choose a unique MAC that we can use as a datapath
778 * ID. In some special cases, though, multiple bridges will end up with
779 * the same MAC address. This is OK for the bridges, but it will confuse
780 * the OpenFlow controller, because each datapath needs a unique datapath
783 * Datapath IDs must be unique. It is also very desirable that they be
784 * stable from one run to the next, so that policy set on a datapath
789 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
796 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
798 * A bridge whose MAC address is taken from a VLAN network device
799 * (that is, a network device created with vconfig(8) or similar
800 * tool) will have the same MAC address as a bridge on the VLAN
801 * device's physical network device.
803 * Handle this case by hashing the physical network device MAC
804 * along with the VLAN identifier.
806 uint8_t buf[ETH_ADDR_LEN + 2];
807 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
808 buf[ETH_ADDR_LEN] = vlan >> 8;
809 buf[ETH_ADDR_LEN + 1] = vlan;
810 return dpid_from_hash(buf, sizeof buf);
813 * Assume that this bridge's MAC address is unique, since it
814 * doesn't fit any of the cases we handle specially.
819 * A purely internal bridge, that is, one that has no non-virtual
820 * network devices on it at all, is more difficult because it has no
821 * natural unique identifier at all.
823 * When the host is a XenServer, we handle this case by hashing the
824 * host's UUID with the name of the bridge. Names of bridges are
825 * persistent across XenServer reboots, although they can be reused if
826 * an internal network is destroyed and then a new one is later
827 * created, so this is fairly effective.
829 * When the host is not a XenServer, we punt by using a random MAC
830 * address on each run.
832 const char *host_uuid = xenserver_get_host_uuid();
834 char *combined = xasprintf("%s,%s", host_uuid, br->name);
835 dpid = dpid_from_hash(combined, strlen(combined));
841 return eth_addr_to_uint64(bridge_ea);
845 dpid_from_hash(const void *data, size_t n)
847 uint8_t hash[SHA1_DIGEST_SIZE];
849 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
850 sha1_bytes(data, n, hash);
851 eth_addr_mark_random(hash);
852 return eth_addr_to_uint64(hash);
858 struct bridge *br, *next;
862 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
863 int error = bridge_run_one(br);
865 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
866 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
867 "forcing reconfiguration", br->name);
881 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
882 ofproto_wait(br->ofproto);
883 if (br->controller) {
888 mac_learning_wait(br->ml);
895 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
896 * configuration changes. */
898 bridge_flush(struct bridge *br)
900 COVERAGE_INC(bridge_flush);
903 mac_learning_flush(br->ml);
907 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
909 static struct iface *
910 bridge_get_local_iface(struct bridge *br)
914 for (i = 0; i < br->n_ports; i++) {
915 struct port *port = br->ports[i];
916 for (j = 0; j < port->n_ifaces; j++) {
917 struct iface *iface = port->ifaces[j];
918 if (iface->dp_ifidx == ODPP_LOCAL) {
927 /* Bridge unixctl user interface functions. */
929 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
931 struct ds ds = DS_EMPTY_INITIALIZER;
932 const struct bridge *br;
934 br = bridge_lookup(args);
936 unixctl_command_reply(conn, 501, "no such bridge");
940 ds_put_cstr(&ds, " port VLAN MAC Age\n");
942 const struct mac_entry *e;
943 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
944 if (e->port < 0 || e->port >= br->n_ports) {
947 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
948 br->ports[e->port]->ifaces[0]->dp_ifidx,
949 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
952 unixctl_command_reply(conn, 200, ds_cstr(&ds));
956 /* Bridge reconfiguration functions. */
958 static struct bridge *
959 bridge_create(const char *name)
964 assert(!bridge_lookup(name));
965 br = xcalloc(1, sizeof *br);
967 error = dpif_create(name, &br->dpif);
968 if (error == EEXIST || error == EBUSY) {
969 error = dpif_open(name, &br->dpif);
971 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
972 name, strerror(error));
976 dpif_flow_flush(br->dpif);
978 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
983 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
985 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
986 dpif_delete(br->dpif);
987 dpif_close(br->dpif);
992 br->name = xstrdup(name);
993 br->ml = mac_learning_create();
994 br->sent_config_request = false;
995 eth_addr_random(br->default_ea);
997 port_array_init(&br->ifaces);
1000 br->bond_next_rebalance = time_msec() + 10000;
1002 list_push_back(&all_bridges, &br->node);
1004 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1010 bridge_destroy(struct bridge *br)
1015 while (br->n_ports > 0) {
1016 port_destroy(br->ports[br->n_ports - 1]);
1018 list_remove(&br->node);
1019 error = dpif_delete(br->dpif);
1020 if (error && error != ENOENT) {
1021 VLOG_ERR("failed to delete %s: %s",
1022 dpif_name(br->dpif), strerror(error));
1024 dpif_close(br->dpif);
1025 ofproto_destroy(br->ofproto);
1026 free(br->controller);
1027 mac_learning_destroy(br->ml);
1028 port_array_destroy(&br->ifaces);
1035 static struct bridge *
1036 bridge_lookup(const char *name)
1040 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1041 if (!strcmp(br->name, name)) {
1049 bridge_exists(const char *name)
1051 return bridge_lookup(name) ? true : false;
1055 bridge_get_datapathid(const char *name)
1057 struct bridge *br = bridge_lookup(name);
1058 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1061 /* Handle requests for a listing of all flows known by the OpenFlow
1062 * stack, including those normally hidden. */
1064 bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args)
1069 br = bridge_lookup(args);
1071 unixctl_command_reply(conn, 501, "Unknown bridge");
1076 ofproto_get_all_flows(br->ofproto, &results);
1078 unixctl_command_reply(conn, 200, ds_cstr(&results));
1079 ds_destroy(&results);
1083 bridge_run_one(struct bridge *br)
1087 error = ofproto_run1(br->ofproto);
1093 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1098 error = ofproto_run2(br->ofproto, br->flush);
1105 bridge_get_controller(const struct bridge *br)
1107 const char *controller;
1109 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1111 controller = cfg_get_string(0, "mgmt.controller");
1113 return controller && controller[0] ? controller : NULL;
1117 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1119 struct svec *ifaces = ifaces_;
1120 if (!svec_contains(ifaces, iface->name)) {
1121 svec_add(ifaces, iface->name);
1125 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1127 br->name, iface->name, iface->port->name);
1133 bridge_reconfigure_one(struct bridge *br)
1135 struct svec old_ports, new_ports, ifaces;
1136 struct svec listeners, old_listeners;
1137 struct svec snoops, old_snoops;
1140 /* Collect old ports. */
1141 svec_init(&old_ports);
1142 for (i = 0; i < br->n_ports; i++) {
1143 svec_add(&old_ports, br->ports[i]->name);
1145 svec_sort(&old_ports);
1146 assert(svec_is_unique(&old_ports));
1148 /* Collect new ports. */
1149 svec_init(&new_ports);
1150 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1151 svec_sort(&new_ports);
1152 if (bridge_get_controller(br)) {
1153 char local_name[IF_NAMESIZE];
1156 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1157 local_name, sizeof local_name);
1158 if (!error && !svec_contains(&new_ports, local_name)) {
1159 svec_add(&new_ports, local_name);
1160 svec_sort(&new_ports);
1163 if (!svec_is_unique(&new_ports)) {
1164 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1165 br->name, svec_get_duplicate(&new_ports));
1166 svec_unique(&new_ports);
1169 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1171 /* Get rid of deleted ports and add new ports. */
1172 for (i = 0; i < br->n_ports; ) {
1173 struct port *port = br->ports[i];
1174 if (!svec_contains(&new_ports, port->name)) {
1180 for (i = 0; i < new_ports.n; i++) {
1181 const char *name = new_ports.names[i];
1182 if (!svec_contains(&old_ports, name)) {
1183 port_create(br, name);
1186 svec_destroy(&old_ports);
1187 svec_destroy(&new_ports);
1189 /* Reconfigure all ports. */
1190 for (i = 0; i < br->n_ports; i++) {
1191 port_reconfigure(br->ports[i]);
1194 /* Check and delete duplicate interfaces. */
1196 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1197 svec_destroy(&ifaces);
1199 /* Delete all flows if we're switching from connected to standalone or vice
1200 * versa. (XXX Should we delete all flows if we are switching from one
1201 * controller to another?) */
1203 /* Configure OpenFlow management listeners. */
1204 svec_init(&listeners);
1205 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1207 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1208 ovs_rundir, br->name));
1209 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1210 svec_clear(&listeners);
1212 svec_sort_unique(&listeners);
1214 svec_init(&old_listeners);
1215 ofproto_get_listeners(br->ofproto, &old_listeners);
1216 svec_sort_unique(&old_listeners);
1218 if (!svec_equal(&listeners, &old_listeners)) {
1219 ofproto_set_listeners(br->ofproto, &listeners);
1221 svec_destroy(&listeners);
1222 svec_destroy(&old_listeners);
1224 /* Configure OpenFlow controller connection snooping. */
1226 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1228 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1229 ovs_rundir, br->name));
1230 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1231 svec_clear(&snoops);
1233 svec_sort_unique(&snoops);
1235 svec_init(&old_snoops);
1236 ofproto_get_snoops(br->ofproto, &old_snoops);
1237 svec_sort_unique(&old_snoops);
1239 if (!svec_equal(&snoops, &old_snoops)) {
1240 ofproto_set_snoops(br->ofproto, &snoops);
1242 svec_destroy(&snoops);
1243 svec_destroy(&old_snoops);
1245 mirror_reconfigure(br);
1249 bridge_reconfigure_controller(struct bridge *br)
1251 char *pfx = xasprintf("bridge.%s.controller", br->name);
1252 const char *controller;
1254 controller = bridge_get_controller(br);
1255 if ((br->controller != NULL) != (controller != NULL)) {
1256 ofproto_flush_flows(br->ofproto);
1258 free(br->controller);
1259 br->controller = controller ? xstrdup(controller) : NULL;
1262 const char *fail_mode;
1263 int max_backoff, probe;
1264 int rate_limit, burst_limit;
1266 if (!strcmp(controller, "discover")) {
1267 bool update_resolv_conf = true;
1269 if (cfg_has("%s.update-resolv.conf", pfx)) {
1270 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1273 ofproto_set_discovery(br->ofproto, true,
1274 cfg_get_string(0, "%s.accept-regex", pfx),
1275 update_resolv_conf);
1277 struct iface *local_iface;
1280 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1282 || cfg_get_bool(0, "%s.in-band", pfx));
1283 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1284 ofproto_set_in_band(br->ofproto, in_band);
1286 local_iface = bridge_get_local_iface(br);
1288 && cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1289 struct netdev *netdev = local_iface->netdev;
1290 struct in_addr ip, mask, gateway;
1291 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1292 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1293 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1295 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1297 mask.s_addr = guess_netmask(ip.s_addr);
1299 if (!netdev_set_in4(netdev, ip, mask)) {
1300 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1302 br->name, IP_ARGS(&ip.s_addr),
1303 IP_ARGS(&mask.s_addr));
1306 if (gateway.s_addr) {
1307 if (!netdev_add_router(netdev, gateway)) {
1308 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1309 br->name, IP_ARGS(&gateway.s_addr));
1315 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1317 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1319 ofproto_set_failure(br->ofproto,
1321 || !strcmp(fail_mode, "standalone")
1322 || !strcmp(fail_mode, "open")));
1324 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1326 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1331 ofproto_set_probe_interval(br->ofproto, probe);
1333 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1335 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1340 ofproto_set_max_backoff(br->ofproto, max_backoff);
1342 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1344 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1346 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1348 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1350 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1352 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1354 if (cfg_has("%s.commands.acl", pfx)) {
1355 struct svec command_acls;
1358 svec_init(&command_acls);
1359 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1360 command_acl = svec_join(&command_acls, ",", "");
1362 ofproto_set_remote_execution(br->ofproto, command_acl,
1363 cfg_get_string(0, "%s.commands.dir",
1366 svec_destroy(&command_acls);
1369 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1372 union ofp_action action;
1375 /* Set up a flow that matches every packet and directs them to
1376 * OFPP_NORMAL (which goes to us). */
1377 memset(&action, 0, sizeof action);
1378 action.type = htons(OFPAT_OUTPUT);
1379 action.output.len = htons(sizeof action);
1380 action.output.port = htons(OFPP_NORMAL);
1381 memset(&flow, 0, sizeof flow);
1382 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1385 ofproto_set_in_band(br->ofproto, false);
1386 ofproto_set_max_backoff(br->ofproto, 1);
1387 ofproto_set_probe_interval(br->ofproto, 5);
1388 ofproto_set_failure(br->ofproto, false);
1389 ofproto_set_stp(br->ofproto, false);
1393 ofproto_set_controller(br->ofproto, br->controller);
1397 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1402 for (i = 0; i < br->n_ports; i++) {
1403 struct port *port = br->ports[i];
1404 for (j = 0; j < port->n_ifaces; j++) {
1405 struct iface *iface = port->ifaces[j];
1406 svec_add(ifaces, iface->name);
1408 if (port->n_ifaces > 1
1409 && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
1410 svec_add(ifaces, port->name);
1413 svec_sort_unique(ifaces);
1416 /* For robustness, in case the administrator moves around datapath ports behind
1417 * our back, we re-check all the datapath port numbers here.
1419 * This function will set the 'dp_ifidx' members of interfaces that have
1420 * disappeared to -1, so only call this function from a context where those
1421 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1422 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1423 * datapath, which doesn't support UINT16_MAX+1 ports. */
1425 bridge_fetch_dp_ifaces(struct bridge *br)
1427 struct odp_port *dpif_ports;
1428 size_t n_dpif_ports;
1431 /* Reset all interface numbers. */
1432 for (i = 0; i < br->n_ports; i++) {
1433 struct port *port = br->ports[i];
1434 for (j = 0; j < port->n_ifaces; j++) {
1435 struct iface *iface = port->ifaces[j];
1436 iface->dp_ifidx = -1;
1439 port_array_clear(&br->ifaces);
1441 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1442 for (i = 0; i < n_dpif_ports; i++) {
1443 struct odp_port *p = &dpif_ports[i];
1444 struct iface *iface = iface_lookup(br, p->devname);
1446 if (iface->dp_ifidx >= 0) {
1447 VLOG_WARN("%s reported interface %s twice",
1448 dpif_name(br->dpif), p->devname);
1449 } else if (iface_from_dp_ifidx(br, p->port)) {
1450 VLOG_WARN("%s reported interface %"PRIu16" twice",
1451 dpif_name(br->dpif), p->port);
1453 port_array_set(&br->ifaces, p->port, iface);
1454 iface->dp_ifidx = p->port;
1461 /* Bridge packet processing functions. */
1464 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1466 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1469 static struct bond_entry *
1470 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1472 return &port->bond_hash[bond_hash(mac)];
1476 bond_choose_iface(const struct port *port)
1479 for (i = 0; i < port->n_ifaces; i++) {
1480 if (port->ifaces[i]->enabled) {
1488 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1489 uint16_t *dp_ifidx, tag_type *tags)
1491 struct iface *iface;
1493 assert(port->n_ifaces);
1494 if (port->n_ifaces == 1) {
1495 iface = port->ifaces[0];
1497 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1498 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1499 || !port->ifaces[e->iface_idx]->enabled) {
1500 /* XXX select interface properly. The current interface selection
1501 * is only good for testing the rebalancing code. */
1502 e->iface_idx = bond_choose_iface(port);
1503 if (e->iface_idx < 0) {
1504 *tags |= port->no_ifaces_tag;
1507 e->iface_tag = tag_create_random();
1508 ((struct port *) port)->bond_compat_is_stale = true;
1510 *tags |= e->iface_tag;
1511 iface = port->ifaces[e->iface_idx];
1513 *dp_ifidx = iface->dp_ifidx;
1514 *tags |= iface->tag; /* Currently only used for bonding. */
1519 bond_link_status_update(struct iface *iface, bool carrier)
1521 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1522 struct port *port = iface->port;
1524 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1525 /* Nothing to do. */
1528 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1529 iface->name, carrier ? "detected" : "dropped");
1530 if (carrier == iface->enabled) {
1531 iface->delay_expires = LLONG_MAX;
1532 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1533 iface->name, carrier ? "disabled" : "enabled");
1534 } else if (carrier && port->updelay && port->active_iface < 0) {
1535 iface->delay_expires = time_msec();
1536 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1537 "other interface is up", iface->name, port->updelay);
1539 int delay = carrier ? port->updelay : port->downdelay;
1540 iface->delay_expires = time_msec() + delay;
1543 "interface %s: will be %s if it stays %s for %d ms",
1545 carrier ? "enabled" : "disabled",
1546 carrier ? "up" : "down",
1553 bond_choose_active_iface(struct port *port)
1555 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1557 port->active_iface = bond_choose_iface(port);
1558 port->active_iface_tag = tag_create_random();
1559 if (port->active_iface >= 0) {
1560 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1561 port->name, port->ifaces[port->active_iface]->name);
1563 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1569 bond_enable_slave(struct iface *iface, bool enable)
1571 struct port *port = iface->port;
1572 struct bridge *br = port->bridge;
1574 iface->delay_expires = LLONG_MAX;
1575 if (enable == iface->enabled) {
1579 iface->enabled = enable;
1580 if (!iface->enabled) {
1581 VLOG_WARN("interface %s: disabled", iface->name);
1582 ofproto_revalidate(br->ofproto, iface->tag);
1583 if (iface->port_ifidx == port->active_iface) {
1584 ofproto_revalidate(br->ofproto,
1585 port->active_iface_tag);
1586 bond_choose_active_iface(port);
1588 bond_send_learning_packets(port);
1590 VLOG_WARN("interface %s: enabled", iface->name);
1591 if (port->active_iface < 0) {
1592 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1593 bond_choose_active_iface(port);
1594 bond_send_learning_packets(port);
1596 iface->tag = tag_create_random();
1598 port_update_bond_compat(port);
1602 bond_run(struct bridge *br)
1606 for (i = 0; i < br->n_ports; i++) {
1607 struct port *port = br->ports[i];
1609 if (port->bond_compat_is_stale) {
1610 port->bond_compat_is_stale = false;
1611 port_update_bond_compat(port);
1614 if (port->n_ifaces < 2) {
1617 for (j = 0; j < port->n_ifaces; j++) {
1618 struct iface *iface = port->ifaces[j];
1619 if (time_msec() >= iface->delay_expires) {
1620 bond_enable_slave(iface, !iface->enabled);
1627 bond_wait(struct bridge *br)
1631 for (i = 0; i < br->n_ports; i++) {
1632 struct port *port = br->ports[i];
1633 if (port->n_ifaces < 2) {
1636 for (j = 0; j < port->n_ifaces; j++) {
1637 struct iface *iface = port->ifaces[j];
1638 if (iface->delay_expires != LLONG_MAX) {
1639 poll_timer_wait(iface->delay_expires - time_msec());
1646 set_dst(struct dst *p, const flow_t *flow,
1647 const struct port *in_port, const struct port *out_port,
1652 * XXX This uses too many tags: any broadcast flow will get one tag per
1653 * destination port, and thus a broadcast on a switch of any size is likely
1654 * to have all tag bits set. We should figure out a way to be smarter.
1656 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1657 *tags |= out_port->stp_state_tag;
1658 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1662 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1663 : in_port->vlan >= 0 ? in_port->vlan
1664 : ntohs(flow->dl_vlan));
1665 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1669 swap_dst(struct dst *p, struct dst *q)
1671 struct dst tmp = *p;
1676 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1677 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1678 * that we push to the datapath. We could in fact fully sort the array by
1679 * vlan, but in most cases there are at most two different vlan tags so that's
1680 * possibly overkill.) */
1682 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1684 struct dst *first = dsts;
1685 struct dst *last = dsts + n_dsts;
1687 while (first != last) {
1689 * - All dsts < first have vlan == 'vlan'.
1690 * - All dsts >= last have vlan != 'vlan'.
1691 * - first < last. */
1692 while (first->vlan == vlan) {
1693 if (++first == last) {
1698 /* Same invariants, plus one additional:
1699 * - first->vlan != vlan.
1701 while (last[-1].vlan != vlan) {
1702 if (--last == first) {
1707 /* Same invariants, plus one additional:
1708 * - last[-1].vlan == vlan.*/
1709 swap_dst(first++, --last);
1714 mirror_mask_ffs(mirror_mask_t mask)
1716 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1721 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1722 const struct dst *test)
1725 for (i = 0; i < n_dsts; i++) {
1726 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1734 port_trunks_vlan(const struct port *port, uint16_t vlan)
1736 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1740 port_includes_vlan(const struct port *port, uint16_t vlan)
1742 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1746 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1747 const struct port *in_port, const struct port *out_port,
1748 struct dst dsts[], tag_type *tags)
1750 mirror_mask_t mirrors = in_port->src_mirrors;
1751 struct dst *dst = dsts;
1754 *tags |= in_port->stp_state_tag;
1755 if (out_port == FLOOD_PORT) {
1756 /* XXX use ODP_FLOOD if no vlans or bonding. */
1757 /* XXX even better, define each VLAN as a datapath port group */
1758 for (i = 0; i < br->n_ports; i++) {
1759 struct port *port = br->ports[i];
1760 if (port != in_port && port_includes_vlan(port, vlan)
1761 && !port->is_mirror_output_port
1762 && set_dst(dst, flow, in_port, port, tags)) {
1763 mirrors |= port->dst_mirrors;
1767 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1768 mirrors |= out_port->dst_mirrors;
1773 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1774 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1776 if (set_dst(dst, flow, in_port, m->out_port, tags)
1777 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1781 for (i = 0; i < br->n_ports; i++) {
1782 struct port *port = br->ports[i];
1783 if (port_includes_vlan(port, m->out_vlan)
1784 && set_dst(dst, flow, in_port, port, tags))
1788 if (port->vlan < 0) {
1789 dst->vlan = m->out_vlan;
1791 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1795 /* Use the vlan tag on the original flow instead of
1796 * the one passed in the vlan parameter. This ensures
1797 * that we compare the vlan from before any implicit
1798 * tagging tags place. This is necessary because
1799 * dst->vlan is the final vlan, after removing implicit
1801 flow_vlan = ntohs(flow->dl_vlan);
1802 if (flow_vlan == 0) {
1803 flow_vlan = OFP_VLAN_NONE;
1805 if (port == in_port && dst->vlan == flow_vlan) {
1806 /* Don't send out input port on same VLAN. */
1814 mirrors &= mirrors - 1;
1817 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1822 print_dsts(const struct dst *dsts, size_t n)
1824 for (; n--; dsts++) {
1825 printf(">p%"PRIu16, dsts->dp_ifidx);
1826 if (dsts->vlan != OFP_VLAN_NONE) {
1827 printf("v%"PRIu16, dsts->vlan);
1833 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1834 const struct port *in_port, const struct port *out_port,
1835 tag_type *tags, struct odp_actions *actions)
1837 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1839 const struct dst *p;
1842 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1844 cur_vlan = ntohs(flow->dl_vlan);
1845 for (p = dsts; p < &dsts[n_dsts]; p++) {
1846 union odp_action *a;
1847 if (p->vlan != cur_vlan) {
1848 if (p->vlan == OFP_VLAN_NONE) {
1849 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1851 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1852 a->vlan_vid.vlan_vid = htons(p->vlan);
1856 a = odp_actions_add(actions, ODPAT_OUTPUT);
1857 a->output.port = p->dp_ifidx;
1862 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1864 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1865 return (flow->dl_type == htons(ETH_TYPE_ARP)
1866 && eth_addr_is_broadcast(flow->dl_dst)
1867 && packet->size >= sizeof(struct arp_eth_header)
1868 && arp->ar_op == ARP_OP_REQUEST);
1871 /* If the composed actions may be applied to any packet in the given 'flow',
1872 * returns true. Otherwise, the actions should only be applied to 'packet', or
1873 * not at all, if 'packet' was NULL. */
1875 process_flow(struct bridge *br, const flow_t *flow,
1876 const struct ofpbuf *packet, struct odp_actions *actions,
1879 struct iface *in_iface;
1880 struct port *in_port;
1881 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1884 /* Find the interface and port structure for the received packet. */
1885 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1887 /* No interface? Something fishy... */
1888 if (packet != NULL) {
1889 /* Odd. A few possible reasons here:
1891 * - We deleted an interface but there are still a few packets
1892 * queued up from it.
1894 * - Someone externally added an interface (e.g. with "ovs-dpctl
1895 * add-if") that we don't know about.
1897 * - Packet arrived on the local port but the local port is not
1898 * one of our bridge ports.
1900 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1902 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1903 "interface %"PRIu16, br->name, flow->in_port);
1906 /* Return without adding any actions, to drop packets on this flow. */
1909 in_port = in_iface->port;
1911 /* Figure out what VLAN this packet belongs to.
1913 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1914 * belongs to VLAN 0, so we should treat both cases identically. (In the
1915 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1916 * presumably to allow a priority to be specified. In the latter case, the
1917 * packet does not have any 802.1Q header.) */
1918 vlan = ntohs(flow->dl_vlan);
1919 if (vlan == OFP_VLAN_NONE) {
1922 if (in_port->vlan >= 0) {
1924 /* XXX support double tagging? */
1925 if (packet != NULL) {
1926 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1927 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1928 "packet received on port %s configured with "
1929 "implicit VLAN %"PRIu16,
1930 br->name, ntohs(flow->dl_vlan),
1931 in_port->name, in_port->vlan);
1935 vlan = in_port->vlan;
1937 if (!port_includes_vlan(in_port, vlan)) {
1938 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1939 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1940 "packet received on port %s not configured for "
1942 br->name, vlan, in_port->name, vlan);
1947 /* Drop frames for ports that STP wants entirely killed (both for
1948 * forwarding and for learning). Later, after we do learning, we'll drop
1949 * the frames that STP wants to do learning but not forwarding on. */
1950 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1954 /* Drop frames for reserved multicast addresses. */
1955 if (eth_addr_is_reserved(flow->dl_dst)) {
1959 /* Drop frames on ports reserved for mirroring. */
1960 if (in_port->is_mirror_output_port) {
1961 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1962 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1963 "which is reserved exclusively for mirroring",
1964 br->name, in_port->name);
1968 /* Packets received on bonds need special attention to avoid duplicates. */
1969 if (in_port->n_ifaces > 1) {
1972 if (eth_addr_is_multicast(flow->dl_dst)) {
1973 *tags |= in_port->active_iface_tag;
1974 if (in_port->active_iface != in_iface->port_ifidx) {
1975 /* Drop all multicast packets on inactive slaves. */
1980 /* Drop all packets for which we have learned a different input
1981 * port, because we probably sent the packet on one slave and got
1982 * it back on the other. Broadcast ARP replies are an exception
1983 * to this rule: the host has moved to another switch. */
1984 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1985 if (src_idx != -1 && src_idx != in_port->port_idx &&
1986 (!packet || !is_bcast_arp_reply(flow, packet))) {
1992 out_port = FLOOD_PORT;
1996 /* Learn source MAC (but don't try to learn from revalidation). */
1998 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1999 vlan, in_port->port_idx);
2001 /* The log messages here could actually be useful in debugging,
2002 * so keep the rate limit relatively high. */
2003 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2005 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2006 "on port %s in VLAN %d",
2007 br->name, ETH_ADDR_ARGS(flow->dl_src),
2008 in_port->name, vlan);
2009 ofproto_revalidate(br->ofproto, rev_tag);
2013 /* Determine output port. */
2014 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2016 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2017 out_port = br->ports[out_port_idx];
2018 } else if (!packet) {
2019 /* If we are revalidating but don't have a learning entry then
2020 * eject the flow. Installing a flow that floods packets will
2021 * prevent us from seeing future packets and learning properly. */
2026 /* Don't send packets out their input ports. Don't forward frames that STP
2027 * wants us to discard. */
2028 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
2033 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
2036 * We send out only a single packet, instead of setting up a flow, if the
2037 * packet is an ARP directed to broadcast that arrived on a bonded
2038 * interface. In such a situation ARP requests and replies must be handled
2039 * differently, but OpenFlow unfortunately can't distinguish them.
2041 return (in_port->n_ifaces < 2
2042 || flow->dl_type != htons(ETH_TYPE_ARP)
2043 || !eth_addr_is_broadcast(flow->dl_dst));
2046 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2049 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2050 const struct ofp_phy_port *opp,
2053 struct bridge *br = br_;
2054 struct iface *iface;
2057 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2063 if (reason == OFPPR_DELETE) {
2064 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2065 br->name, iface->name);
2066 iface_destroy(iface);
2067 if (!port->n_ifaces) {
2068 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2069 br->name, port->name);
2075 if (port->n_ifaces > 1) {
2076 bool up = !(opp->state & OFPPS_LINK_DOWN);
2077 bond_link_status_update(iface, up);
2078 port_update_bond_compat(port);
2084 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2085 struct odp_actions *actions, tag_type *tags, void *br_)
2087 struct bridge *br = br_;
2090 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
2091 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
2092 brstp_receive(br, flow, payload);
2097 COVERAGE_INC(bridge_process_flow);
2098 return process_flow(br, flow, packet, actions, tags);
2102 bridge_account_flow_ofhook_cb(const flow_t *flow,
2103 const union odp_action *actions,
2104 size_t n_actions, unsigned long long int n_bytes,
2107 struct bridge *br = br_;
2108 const union odp_action *a;
2110 if (!br->has_bonded_ports) {
2114 for (a = actions; a < &actions[n_actions]; a++) {
2115 if (a->type == ODPAT_OUTPUT) {
2116 struct port *port = port_from_dp_ifidx(br, a->output.port);
2117 if (port && port->n_ifaces >= 2) {
2118 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2119 e->tx_bytes += n_bytes;
2126 bridge_account_checkpoint_ofhook_cb(void *br_)
2128 struct bridge *br = br_;
2131 if (!br->has_bonded_ports) {
2135 /* The current ofproto implementation calls this callback at least once a
2136 * second, so this timer implementation is sufficient. */
2137 if (time_msec() < br->bond_next_rebalance) {
2140 br->bond_next_rebalance = time_msec() + 10000;
2142 for (i = 0; i < br->n_ports; i++) {
2143 struct port *port = br->ports[i];
2144 if (port->n_ifaces > 1) {
2145 bond_rebalance_port(port);
2150 static struct ofhooks bridge_ofhooks = {
2151 bridge_port_changed_ofhook_cb,
2152 bridge_normal_ofhook_cb,
2153 bridge_account_flow_ofhook_cb,
2154 bridge_account_checkpoint_ofhook_cb,
2157 /* Bonding functions. */
2159 /* Statistics for a single interface on a bonded port, used for load-based
2160 * bond rebalancing. */
2161 struct slave_balance {
2162 struct iface *iface; /* The interface. */
2163 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2165 /* All the "bond_entry"s that are assigned to this interface, in order of
2166 * increasing tx_bytes. */
2167 struct bond_entry **hashes;
2171 /* Sorts pointers to pointers to bond_entries in ascending order by the
2172 * interface to which they are assigned, and within a single interface in
2173 * ascending order of bytes transmitted. */
2175 compare_bond_entries(const void *a_, const void *b_)
2177 const struct bond_entry *const *ap = a_;
2178 const struct bond_entry *const *bp = b_;
2179 const struct bond_entry *a = *ap;
2180 const struct bond_entry *b = *bp;
2181 if (a->iface_idx != b->iface_idx) {
2182 return a->iface_idx > b->iface_idx ? 1 : -1;
2183 } else if (a->tx_bytes != b->tx_bytes) {
2184 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2190 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2191 * *descending* order by number of bytes transmitted. */
2193 compare_slave_balance(const void *a_, const void *b_)
2195 const struct slave_balance *a = a_;
2196 const struct slave_balance *b = b_;
2197 if (a->iface->enabled != b->iface->enabled) {
2198 return a->iface->enabled ? -1 : 1;
2199 } else if (a->tx_bytes != b->tx_bytes) {
2200 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2207 swap_bals(struct slave_balance *a, struct slave_balance *b)
2209 struct slave_balance tmp = *a;
2214 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2215 * given that 'p' (and only 'p') might be in the wrong location.
2217 * This function invalidates 'p', since it might now be in a different memory
2220 resort_bals(struct slave_balance *p,
2221 struct slave_balance bals[], size_t n_bals)
2224 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2225 swap_bals(p, p - 1);
2227 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2228 swap_bals(p, p + 1);
2234 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2236 if (VLOG_IS_DBG_ENABLED()) {
2237 struct ds ds = DS_EMPTY_INITIALIZER;
2238 const struct slave_balance *b;
2240 for (b = bals; b < bals + n_bals; b++) {
2244 ds_put_char(&ds, ',');
2246 ds_put_format(&ds, " %s %"PRIu64"kB",
2247 b->iface->name, b->tx_bytes / 1024);
2249 if (!b->iface->enabled) {
2250 ds_put_cstr(&ds, " (disabled)");
2252 if (b->n_hashes > 0) {
2253 ds_put_cstr(&ds, " (");
2254 for (i = 0; i < b->n_hashes; i++) {
2255 const struct bond_entry *e = b->hashes[i];
2257 ds_put_cstr(&ds, " + ");
2259 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2260 e - port->bond_hash, e->tx_bytes / 1024);
2262 ds_put_cstr(&ds, ")");
2265 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2270 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2272 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2275 struct bond_entry *hash = from->hashes[hash_idx];
2276 struct port *port = from->iface->port;
2277 uint64_t delta = hash->tx_bytes;
2279 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2280 "from %s to %s (now carrying %"PRIu64"kB and "
2281 "%"PRIu64"kB load, respectively)",
2282 port->name, delta / 1024, hash - port->bond_hash,
2283 from->iface->name, to->iface->name,
2284 (from->tx_bytes - delta) / 1024,
2285 (to->tx_bytes + delta) / 1024);
2287 /* Delete element from from->hashes.
2289 * We don't bother to add the element to to->hashes because not only would
2290 * it require more work, the only purpose it would be to allow that hash to
2291 * be migrated to another slave in this rebalancing run, and there is no
2292 * point in doing that. */
2293 if (hash_idx == 0) {
2296 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2297 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2301 /* Shift load away from 'from' to 'to'. */
2302 from->tx_bytes -= delta;
2303 to->tx_bytes += delta;
2305 /* Arrange for flows to be revalidated. */
2306 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2307 hash->iface_idx = to->iface->port_ifidx;
2308 hash->iface_tag = tag_create_random();
2312 bond_rebalance_port(struct port *port)
2314 struct slave_balance bals[DP_MAX_PORTS];
2316 struct bond_entry *hashes[BOND_MASK + 1];
2317 struct slave_balance *b, *from, *to;
2318 struct bond_entry *e;
2321 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2322 * descending order of tx_bytes, so that bals[0] represents the most
2323 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2326 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2327 * array for each slave_balance structure, we sort our local array of
2328 * hashes in order by slave, so that all of the hashes for a given slave
2329 * become contiguous in memory, and then we point each 'hashes' members of
2330 * a slave_balance structure to the start of a contiguous group. */
2331 n_bals = port->n_ifaces;
2332 for (b = bals; b < &bals[n_bals]; b++) {
2333 b->iface = port->ifaces[b - bals];
2338 for (i = 0; i <= BOND_MASK; i++) {
2339 hashes[i] = &port->bond_hash[i];
2341 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2342 for (i = 0; i <= BOND_MASK; i++) {
2344 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2345 b = &bals[e->iface_idx];
2346 b->tx_bytes += e->tx_bytes;
2348 b->hashes = &hashes[i];
2353 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2354 log_bals(bals, n_bals, port);
2356 /* Discard slaves that aren't enabled (which were sorted to the back of the
2357 * array earlier). */
2358 while (!bals[n_bals - 1].iface->enabled) {
2365 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2366 to = &bals[n_bals - 1];
2367 for (from = bals; from < to; ) {
2368 uint64_t overload = from->tx_bytes - to->tx_bytes;
2369 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2370 /* The extra load on 'from' (and all less-loaded slaves), compared
2371 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2372 * it is less than ~1Mbps. No point in rebalancing. */
2374 } else if (from->n_hashes == 1) {
2375 /* 'from' only carries a single MAC hash, so we can't shift any
2376 * load away from it, even though we want to. */
2379 /* 'from' is carrying significantly more load than 'to', and that
2380 * load is split across at least two different hashes. Pick a hash
2381 * to migrate to 'to' (the least-loaded slave), given that doing so
2382 * must decrease the ratio of the load on the two slaves by at
2385 * The sort order we use means that we prefer to shift away the
2386 * smallest hashes instead of the biggest ones. There is little
2387 * reason behind this decision; we could use the opposite sort
2388 * order to shift away big hashes ahead of small ones. */
2392 for (i = 0; i < from->n_hashes; i++) {
2393 double old_ratio, new_ratio;
2394 uint64_t delta = from->hashes[i]->tx_bytes;
2396 if (delta == 0 || from->tx_bytes - delta == 0) {
2397 /* Pointless move. */
2401 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2403 if (to->tx_bytes == 0) {
2404 /* Nothing on the new slave, move it. */
2408 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2409 new_ratio = (double)(from->tx_bytes - delta) /
2410 (to->tx_bytes + delta);
2412 if (new_ratio == 0) {
2413 /* Should already be covered but check to prevent division
2418 if (new_ratio < 1) {
2419 new_ratio = 1 / new_ratio;
2422 if (old_ratio - new_ratio > 0.1) {
2423 /* Would decrease the ratio, move it. */
2427 if (i < from->n_hashes) {
2428 bond_shift_load(from, to, i);
2429 port->bond_compat_is_stale = true;
2431 /* If the result of the migration changed the relative order of
2432 * 'from' and 'to' swap them back to maintain invariants. */
2433 if (order_swapped) {
2434 swap_bals(from, to);
2437 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2438 * point to different slave_balance structures. It is only
2439 * valid to do these two operations in a row at all because we
2440 * know that 'from' will not move past 'to' and vice versa. */
2441 resort_bals(from, bals, n_bals);
2442 resort_bals(to, bals, n_bals);
2449 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2450 * historical data to decay to <1% in 7 rebalancing runs. */
2451 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2457 bond_send_learning_packets(struct port *port)
2459 struct bridge *br = port->bridge;
2460 struct mac_entry *e;
2461 struct ofpbuf packet;
2462 int error, n_packets, n_errors;
2464 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2468 ofpbuf_init(&packet, 128);
2469 error = n_packets = n_errors = 0;
2470 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2471 union ofp_action actions[2], *a;
2477 if (e->port == port->port_idx
2478 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2482 /* Compose actions. */
2483 memset(actions, 0, sizeof actions);
2486 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2487 a->vlan_vid.len = htons(sizeof *a);
2488 a->vlan_vid.vlan_vid = htons(e->vlan);
2491 a->output.type = htons(OFPAT_OUTPUT);
2492 a->output.len = htons(sizeof *a);
2493 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2498 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2500 flow_extract(&packet, ODPP_NONE, &flow);
2501 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2508 ofpbuf_uninit(&packet);
2511 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2512 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2513 "packets, last error was: %s",
2514 port->name, n_errors, n_packets, strerror(error));
2516 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2517 port->name, n_packets);
2521 /* Bonding unixctl user interface functions. */
2524 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2526 struct ds ds = DS_EMPTY_INITIALIZER;
2527 const struct bridge *br;
2529 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2531 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2534 for (i = 0; i < br->n_ports; i++) {
2535 const struct port *port = br->ports[i];
2536 if (port->n_ifaces > 1) {
2539 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2540 for (j = 0; j < port->n_ifaces; j++) {
2541 const struct iface *iface = port->ifaces[j];
2543 ds_put_cstr(&ds, ", ");
2545 ds_put_cstr(&ds, iface->name);
2547 ds_put_char(&ds, '\n');
2551 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2555 static struct port *
2556 bond_find(const char *name)
2558 const struct bridge *br;
2560 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2563 for (i = 0; i < br->n_ports; i++) {
2564 struct port *port = br->ports[i];
2565 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2574 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2576 struct ds ds = DS_EMPTY_INITIALIZER;
2577 const struct port *port;
2580 port = bond_find(args);
2582 unixctl_command_reply(conn, 501, "no such bond");
2586 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2587 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2588 ds_put_format(&ds, "next rebalance: %lld ms\n",
2589 port->bridge->bond_next_rebalance - time_msec());
2590 for (j = 0; j < port->n_ifaces; j++) {
2591 const struct iface *iface = port->ifaces[j];
2592 struct bond_entry *be;
2595 ds_put_format(&ds, "slave %s: %s\n",
2596 iface->name, iface->enabled ? "enabled" : "disabled");
2597 if (j == port->active_iface) {
2598 ds_put_cstr(&ds, "\tactive slave\n");
2600 if (iface->delay_expires != LLONG_MAX) {
2601 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2602 iface->enabled ? "downdelay" : "updelay",
2603 iface->delay_expires - time_msec());
2607 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2608 int hash = be - port->bond_hash;
2609 struct mac_entry *me;
2611 if (be->iface_idx != j) {
2615 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2616 hash, be->tx_bytes / 1024);
2619 if (!port->bridge->ml) {
2623 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2624 &port->bridge->ml->lrus) {
2627 if (bond_hash(me->mac) == hash
2628 && me->port != port->port_idx
2629 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2630 && dp_ifidx == iface->dp_ifidx)
2632 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2633 ETH_ADDR_ARGS(me->mac));
2638 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2643 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2645 char *args = (char *) args_;
2646 char *save_ptr = NULL;
2647 char *bond_s, *hash_s, *slave_s;
2648 uint8_t mac[ETH_ADDR_LEN];
2650 struct iface *iface;
2651 struct bond_entry *entry;
2654 bond_s = strtok_r(args, " ", &save_ptr);
2655 hash_s = strtok_r(NULL, " ", &save_ptr);
2656 slave_s = strtok_r(NULL, " ", &save_ptr);
2658 unixctl_command_reply(conn, 501,
2659 "usage: bond/migrate BOND HASH SLAVE");
2663 port = bond_find(bond_s);
2665 unixctl_command_reply(conn, 501, "no such bond");
2669 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2670 == ETH_ADDR_SCAN_COUNT) {
2671 hash = bond_hash(mac);
2672 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2673 hash = atoi(hash_s) & BOND_MASK;
2675 unixctl_command_reply(conn, 501, "bad hash");
2679 iface = port_lookup_iface(port, slave_s);
2681 unixctl_command_reply(conn, 501, "no such slave");
2685 if (!iface->enabled) {
2686 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2690 entry = &port->bond_hash[hash];
2691 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2692 entry->iface_idx = iface->port_ifidx;
2693 entry->iface_tag = tag_create_random();
2694 port->bond_compat_is_stale = true;
2695 unixctl_command_reply(conn, 200, "migrated");
2699 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2701 char *args = (char *) args_;
2702 char *save_ptr = NULL;
2703 char *bond_s, *slave_s;
2705 struct iface *iface;
2707 bond_s = strtok_r(args, " ", &save_ptr);
2708 slave_s = strtok_r(NULL, " ", &save_ptr);
2710 unixctl_command_reply(conn, 501,
2711 "usage: bond/set-active-slave BOND SLAVE");
2715 port = bond_find(bond_s);
2717 unixctl_command_reply(conn, 501, "no such bond");
2721 iface = port_lookup_iface(port, slave_s);
2723 unixctl_command_reply(conn, 501, "no such slave");
2727 if (!iface->enabled) {
2728 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2732 if (port->active_iface != iface->port_ifidx) {
2733 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2734 port->active_iface = iface->port_ifidx;
2735 port->active_iface_tag = tag_create_random();
2736 VLOG_INFO("port %s: active interface is now %s",
2737 port->name, iface->name);
2738 bond_send_learning_packets(port);
2739 unixctl_command_reply(conn, 200, "done");
2741 unixctl_command_reply(conn, 200, "no change");
2746 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2748 char *args = (char *) args_;
2749 char *save_ptr = NULL;
2750 char *bond_s, *slave_s;
2752 struct iface *iface;
2754 bond_s = strtok_r(args, " ", &save_ptr);
2755 slave_s = strtok_r(NULL, " ", &save_ptr);
2757 unixctl_command_reply(conn, 501,
2758 "usage: bond/enable/disable-slave BOND SLAVE");
2762 port = bond_find(bond_s);
2764 unixctl_command_reply(conn, 501, "no such bond");
2768 iface = port_lookup_iface(port, slave_s);
2770 unixctl_command_reply(conn, 501, "no such slave");
2774 bond_enable_slave(iface, enable);
2775 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2779 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2781 enable_slave(conn, args, true);
2785 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2787 enable_slave(conn, args, false);
2791 bond_unixctl_hash(struct unixctl_conn *conn, const char *args)
2793 uint8_t mac[ETH_ADDR_LEN];
2797 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2798 == ETH_ADDR_SCAN_COUNT) {
2799 hash = bond_hash(mac);
2801 hash_cstr = xasprintf("%u", hash);
2802 unixctl_command_reply(conn, 200, hash_cstr);
2805 unixctl_command_reply(conn, 501, "invalid mac");
2812 unixctl_command_register("bond/list", bond_unixctl_list);
2813 unixctl_command_register("bond/show", bond_unixctl_show);
2814 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2815 unixctl_command_register("bond/set-active-slave",
2816 bond_unixctl_set_active_slave);
2817 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2818 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2819 unixctl_command_register("bond/hash", bond_unixctl_hash);
2822 /* Port functions. */
2825 port_create(struct bridge *br, const char *name)
2829 port = xcalloc(1, sizeof *port);
2831 port->port_idx = br->n_ports;
2833 port->trunks = NULL;
2834 port->name = xstrdup(name);
2835 port->active_iface = -1;
2836 port->stp_state = STP_DISABLED;
2837 port->stp_state_tag = 0;
2839 if (br->n_ports >= br->allocated_ports) {
2840 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2843 br->ports[br->n_ports++] = port;
2845 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2850 port_reconfigure(struct port *port)
2852 bool bonded = cfg_has_section("bonding.%s", port->name);
2853 struct svec old_ifaces, new_ifaces;
2854 unsigned long *trunks;
2858 /* Collect old and new interfaces. */
2859 svec_init(&old_ifaces);
2860 svec_init(&new_ifaces);
2861 for (i = 0; i < port->n_ifaces; i++) {
2862 svec_add(&old_ifaces, port->ifaces[i]->name);
2864 svec_sort(&old_ifaces);
2866 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2867 if (!new_ifaces.n) {
2868 VLOG_ERR("port %s: no interfaces specified for bonded port",
2870 } else if (new_ifaces.n == 1) {
2871 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2875 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2876 if (port->updelay < 0) {
2879 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2880 if (port->downdelay < 0) {
2881 port->downdelay = 0;
2884 svec_init(&new_ifaces);
2885 svec_add(&new_ifaces, port->name);
2888 /* Get rid of deleted interfaces and add new interfaces. */
2889 for (i = 0; i < port->n_ifaces; i++) {
2890 struct iface *iface = port->ifaces[i];
2891 if (!svec_contains(&new_ifaces, iface->name)) {
2892 iface_destroy(iface);
2897 for (i = 0; i < new_ifaces.n; i++) {
2898 const char *name = new_ifaces.names[i];
2899 if (!svec_contains(&old_ifaces, name)) {
2900 iface_create(port, name);
2906 if (cfg_has("vlan.%s.tag", port->name)) {
2908 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2909 if (vlan >= 0 && vlan <= 4095) {
2910 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2913 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2914 * they even work as-is. But they have not been tested. */
2915 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2919 if (port->vlan != vlan) {
2921 bridge_flush(port->bridge);
2924 /* Get trunked VLANs. */
2927 size_t n_trunks, n_errors;
2930 trunks = bitmap_allocate(4096);
2931 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2933 for (i = 0; i < n_trunks; i++) {
2934 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2936 bitmap_set1(trunks, trunk);
2942 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2943 port->name, n_trunks);
2945 if (n_errors == n_trunks) {
2947 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2950 bitmap_set_multiple(trunks, 0, 4096, 1);
2953 if (cfg_has("vlan.%s.trunks", port->name)) {
2954 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2955 port->name, port->name);
2959 ? port->trunks != NULL
2960 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2961 bridge_flush(port->bridge);
2963 bitmap_free(port->trunks);
2964 port->trunks = trunks;
2966 svec_destroy(&old_ifaces);
2967 svec_destroy(&new_ifaces);
2971 port_destroy(struct port *port)
2974 struct bridge *br = port->bridge;
2978 proc_net_compat_update_vlan(port->name, NULL, 0);
2979 proc_net_compat_update_bond(port->name, NULL);
2981 for (i = 0; i < MAX_MIRRORS; i++) {
2982 struct mirror *m = br->mirrors[i];
2983 if (m && m->out_port == port) {
2988 while (port->n_ifaces > 0) {
2989 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2992 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2993 del->port_idx = port->port_idx;
2996 bitmap_free(port->trunks);
3003 static struct port *
3004 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3006 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3007 return iface ? iface->port : NULL;
3010 static struct port *
3011 port_lookup(const struct bridge *br, const char *name)
3015 for (i = 0; i < br->n_ports; i++) {
3016 struct port *port = br->ports[i];
3017 if (!strcmp(port->name, name)) {
3024 static struct iface *
3025 port_lookup_iface(const struct port *port, const char *name)
3029 for (j = 0; j < port->n_ifaces; j++) {
3030 struct iface *iface = port->ifaces[j];
3031 if (!strcmp(iface->name, name)) {
3039 port_update_bonding(struct port *port)
3041 if (port->n_ifaces < 2) {
3042 /* Not a bonded port. */
3043 if (port->bond_hash) {
3044 free(port->bond_hash);
3045 port->bond_hash = NULL;
3046 port->bond_compat_is_stale = true;
3049 if (!port->bond_hash) {
3052 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3053 for (i = 0; i <= BOND_MASK; i++) {
3054 struct bond_entry *e = &port->bond_hash[i];
3058 port->no_ifaces_tag = tag_create_random();
3059 bond_choose_active_iface(port);
3061 port->bond_compat_is_stale = true;
3066 port_update_bond_compat(struct port *port)
3068 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3069 struct compat_bond bond;
3072 if (port->n_ifaces < 2) {
3073 proc_net_compat_update_bond(port->name, NULL);
3078 bond.updelay = port->updelay;
3079 bond.downdelay = port->downdelay;
3082 bond.hashes = compat_hashes;
3083 if (port->bond_hash) {
3084 const struct bond_entry *e;
3085 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3086 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3087 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3088 cbh->hash = e - port->bond_hash;
3089 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3094 bond.n_slaves = port->n_ifaces;
3095 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3096 for (i = 0; i < port->n_ifaces; i++) {
3097 struct iface *iface = port->ifaces[i];
3098 struct compat_bond_slave *slave = &bond.slaves[i];
3099 slave->name = iface->name;
3101 /* We need to make the same determination as the Linux bonding
3102 * code to determine whether a slave should be consider "up".
3103 * The Linux function bond_miimon_inspect() supports four
3104 * BOND_LINK_* states:
3106 * - BOND_LINK_UP: carrier detected, updelay has passed.
3107 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3108 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3109 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3111 * The function bond_info_show_slave() only considers BOND_LINK_UP
3112 * to be "up" and anything else to be "down".
3114 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3118 netdev_get_etheraddr(iface->netdev, slave->mac);
3121 if (cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
3122 struct netdev *bond_netdev;
3124 if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) {
3126 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3128 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3130 netdev_close(bond_netdev);
3134 proc_net_compat_update_bond(port->name, &bond);
3139 port_update_vlan_compat(struct port *port)
3141 struct bridge *br = port->bridge;
3142 char *vlandev_name = NULL;
3144 if (port->vlan > 0) {
3145 /* Figure out the name that the VLAN device should actually have, if it
3146 * existed. This takes some work because the VLAN device would not
3147 * have port->name in its name; rather, it would have the trunk port's
3148 * name, and 'port' would be attached to a bridge that also had the
3149 * VLAN device one of its ports. So we need to find a trunk port that
3150 * includes port->vlan.
3152 * There might be more than one candidate. This doesn't happen on
3153 * XenServer, so if it happens we just pick the first choice in
3154 * alphabetical order instead of creating multiple VLAN devices. */
3156 for (i = 0; i < br->n_ports; i++) {
3157 struct port *p = br->ports[i];
3158 if (port_trunks_vlan(p, port->vlan)
3160 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3162 uint8_t ea[ETH_ADDR_LEN];
3163 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3164 if (!eth_addr_is_multicast(ea) &&
3165 !eth_addr_is_reserved(ea) &&
3166 !eth_addr_is_zero(ea)) {
3167 vlandev_name = p->name;
3172 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3175 /* Interface functions. */
3178 iface_create(struct port *port, const char *name)
3180 struct iface *iface;
3182 iface = xcalloc(1, sizeof *iface);
3184 iface->port_ifidx = port->n_ifaces;
3185 iface->name = xstrdup(name);
3186 iface->dp_ifidx = -1;
3187 iface->tag = tag_create_random();
3188 iface->delay_expires = LLONG_MAX;
3189 iface->netdev = NULL;
3191 if (port->n_ifaces >= port->allocated_ifaces) {
3192 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3193 sizeof *port->ifaces);
3195 port->ifaces[port->n_ifaces++] = iface;
3196 if (port->n_ifaces > 1) {
3197 port->bridge->has_bonded_ports = true;
3200 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3202 bridge_flush(port->bridge);
3206 iface_destroy(struct iface *iface)
3209 struct port *port = iface->port;
3210 struct bridge *br = port->bridge;
3211 bool del_active = port->active_iface == iface->port_ifidx;
3214 if (iface->dp_ifidx >= 0) {
3215 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3218 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3219 del->port_ifidx = iface->port_ifidx;
3221 netdev_close(iface->netdev);
3226 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3227 bond_choose_active_iface(port);
3228 bond_send_learning_packets(port);
3231 bridge_flush(port->bridge);
3235 static struct iface *
3236 iface_lookup(const struct bridge *br, const char *name)
3240 for (i = 0; i < br->n_ports; i++) {
3241 struct port *port = br->ports[i];
3242 for (j = 0; j < port->n_ifaces; j++) {
3243 struct iface *iface = port->ifaces[j];
3244 if (!strcmp(iface->name, name)) {
3252 static struct iface *
3253 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3255 return port_array_get(&br->ifaces, dp_ifidx);
3258 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3259 * 'br', that is, an interface that is entirely simulated within the datapath.
3260 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3261 * interfaces are created by setting "iface.<iface>.internal = true".
3263 * In addition, we have a kluge-y feature that creates an internal port with
3264 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3265 * This feature needs to go away in the long term. Until then, this is one
3266 * reason why this function takes a name instead of a struct iface: the fake
3267 * interfaces created this way do not have a struct iface. */
3269 iface_is_internal(const struct bridge *br, const char *iface)
3271 if (!strcmp(iface, br->name)
3272 || cfg_get_bool(0, "iface.%s.internal", iface)) {
3276 if (cfg_get_bool(0, "bonding.%s.fake-iface", iface)) {
3277 struct port *port = port_lookup(br, iface);
3278 if (port && port->n_ifaces > 1) {
3286 /* Set Ethernet address of 'iface', if one is specified in the configuration
3289 iface_set_mac(struct iface *iface)
3291 uint64_t mac = cfg_get_mac(0, "iface.%s.mac", iface->name);
3293 static uint8_t ea[ETH_ADDR_LEN];
3295 eth_addr_from_uint64(mac, ea);
3296 if (eth_addr_is_multicast(ea)) {
3297 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3299 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3300 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3301 iface->name, iface->name);
3303 int error = netdev_set_etheraddr(iface->netdev, ea);
3305 VLOG_ERR("interface %s: setting MAC failed (%s)",
3306 iface->name, strerror(error));
3312 /* Port mirroring. */
3315 mirror_reconfigure(struct bridge *br)
3317 struct svec old_mirrors, new_mirrors;
3320 /* Collect old and new mirrors. */
3321 svec_init(&old_mirrors);
3322 svec_init(&new_mirrors);
3323 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3324 for (i = 0; i < MAX_MIRRORS; i++) {
3325 if (br->mirrors[i]) {
3326 svec_add(&old_mirrors, br->mirrors[i]->name);
3330 /* Get rid of deleted mirrors and add new mirrors. */
3331 svec_sort(&old_mirrors);
3332 assert(svec_is_unique(&old_mirrors));
3333 svec_sort(&new_mirrors);
3334 assert(svec_is_unique(&new_mirrors));
3335 for (i = 0; i < MAX_MIRRORS; i++) {
3336 struct mirror *m = br->mirrors[i];
3337 if (m && !svec_contains(&new_mirrors, m->name)) {
3341 for (i = 0; i < new_mirrors.n; i++) {
3342 const char *name = new_mirrors.names[i];
3343 if (!svec_contains(&old_mirrors, name)) {
3344 mirror_create(br, name);
3347 svec_destroy(&old_mirrors);
3348 svec_destroy(&new_mirrors);
3350 /* Reconfigure all mirrors. */
3351 for (i = 0; i < MAX_MIRRORS; i++) {
3352 if (br->mirrors[i]) {
3353 mirror_reconfigure_one(br->mirrors[i]);
3357 /* Update port reserved status. */
3358 for (i = 0; i < br->n_ports; i++) {
3359 br->ports[i]->is_mirror_output_port = false;
3361 for (i = 0; i < MAX_MIRRORS; i++) {
3362 struct mirror *m = br->mirrors[i];
3363 if (m && m->out_port) {
3364 m->out_port->is_mirror_output_port = true;
3370 mirror_create(struct bridge *br, const char *name)
3375 for (i = 0; ; i++) {
3376 if (i >= MAX_MIRRORS) {
3377 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3378 "cannot create %s", br->name, MAX_MIRRORS, name);
3381 if (!br->mirrors[i]) {
3386 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3389 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3392 m->name = xstrdup(name);
3393 svec_init(&m->src_ports);
3394 svec_init(&m->dst_ports);
3402 mirror_destroy(struct mirror *m)
3405 struct bridge *br = m->bridge;
3408 for (i = 0; i < br->n_ports; i++) {
3409 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3410 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3413 svec_destroy(&m->src_ports);
3414 svec_destroy(&m->dst_ports);
3417 m->bridge->mirrors[m->idx] = NULL;
3425 prune_ports(struct mirror *m, struct svec *ports)
3430 svec_sort_unique(ports);
3433 for (i = 0; i < ports->n; i++) {
3434 const char *name = ports->names[i];
3435 if (port_lookup(m->bridge, name)) {
3436 svec_add(&tmp, name);
3438 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3439 m->bridge->name, m->name, name);
3442 svec_swap(ports, &tmp);
3447 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3451 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3452 * order won't give us numeric sort order. But that's good enough for what
3453 * we need right now. */
3454 svec_sort_unique(vlan_strings);
3456 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3458 for (i = 0; i < vlan_strings->n; i++) {
3459 const char *name = vlan_strings->names[i];
3461 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3462 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3463 m->bridge->name, m->name, name);
3465 (*vlans)[n_vlans++] = vlan;
3472 vlan_is_mirrored(const struct mirror *m, int vlan)
3476 for (i = 0; i < m->n_vlans; i++) {
3477 if (m->vlans[i] == vlan) {
3485 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3489 for (i = 0; i < m->n_vlans; i++) {
3490 if (port_trunks_vlan(p, m->vlans[i])) {
3498 mirror_reconfigure_one(struct mirror *m)
3500 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3501 struct svec src_ports, dst_ports, ports;
3502 struct svec vlan_strings;
3503 mirror_mask_t mirror_bit;
3504 const char *out_port_name;
3505 struct port *out_port;
3510 bool mirror_all_ports;
3511 bool any_ports_specified;
3513 /* Get output port. */
3514 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3515 m->bridge->name, m->name);
3516 if (out_port_name) {
3517 out_port = port_lookup(m->bridge, out_port_name);
3519 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3520 "named %s", pfx, m->bridge->name, out_port_name);
3527 if (cfg_has("%s.output.vlan", pfx)) {
3528 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3529 "ignoring %s.output.vlan", pfx, pfx, pfx);
3531 } else if (cfg_has("%s.output.vlan", pfx)) {
3533 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3535 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3536 "but exactly one is required; disabling port mirror %s",
3537 pfx, pfx, pfx, pfx);
3543 /* Get all the ports, and drop duplicates and ports that don't exist. */
3544 svec_init(&src_ports);
3545 svec_init(&dst_ports);
3547 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3548 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3549 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3550 any_ports_specified = src_ports.n || dst_ports.n || ports.n;
3551 svec_append(&src_ports, &ports);
3552 svec_append(&dst_ports, &ports);
3553 svec_destroy(&ports);
3554 prune_ports(m, &src_ports);
3555 prune_ports(m, &dst_ports);
3556 if (any_ports_specified && !src_ports.n && !dst_ports.n) {
3557 VLOG_ERR("%s: none of the specified ports exist; "
3558 "disabling port mirror %s", pfx, pfx);
3563 /* Get all the vlans, and drop duplicate and invalid vlans. */
3564 svec_init(&vlan_strings);
3565 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3566 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3567 svec_destroy(&vlan_strings);
3569 /* Update mirror data. */
3570 if (!svec_equal(&m->src_ports, &src_ports)
3571 || !svec_equal(&m->dst_ports, &dst_ports)
3572 || m->n_vlans != n_vlans
3573 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3574 || m->out_port != out_port
3575 || m->out_vlan != out_vlan) {
3576 bridge_flush(m->bridge);
3578 svec_swap(&m->src_ports, &src_ports);
3579 svec_swap(&m->dst_ports, &dst_ports);
3582 m->n_vlans = n_vlans;
3583 m->out_port = out_port;
3584 m->out_vlan = out_vlan;
3586 /* If no selection criteria have been given, mirror for all ports. */
3587 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3590 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3591 for (i = 0; i < m->bridge->n_ports; i++) {
3592 struct port *port = m->bridge->ports[i];
3594 if (mirror_all_ports
3595 || svec_contains(&m->src_ports, port->name)
3598 ? port_trunks_any_mirrored_vlan(m, port)
3599 : vlan_is_mirrored(m, port->vlan)))) {
3600 port->src_mirrors |= mirror_bit;
3602 port->src_mirrors &= ~mirror_bit;
3605 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3606 port->dst_mirrors |= mirror_bit;
3608 port->dst_mirrors &= ~mirror_bit;
3614 svec_destroy(&src_ports);
3615 svec_destroy(&dst_ports);
3619 /* Spanning tree protocol. */
3621 static void brstp_update_port_state(struct port *);
3624 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3626 struct bridge *br = br_;
3627 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3628 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3630 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3633 struct eth_header *eth = pkt->l2;
3635 netdev_get_etheraddr(iface->netdev, eth->eth_src);
3636 if (eth_addr_is_zero(eth->eth_src)) {
3637 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
3638 "with unknown MAC", br->name, port_no);
3640 union ofp_action action;
3643 memset(&action, 0, sizeof action);
3644 action.type = htons(OFPAT_OUTPUT);
3645 action.output.len = htons(sizeof action);
3646 action.output.port = htons(port_no);
3648 flow_extract(pkt, ODPP_NONE, &flow);
3649 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3656 brstp_reconfigure(struct bridge *br)
3660 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3662 stp_destroy(br->stp);
3668 uint64_t bridge_address, bridge_id;
3669 int bridge_priority;
3671 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3672 if (!bridge_address) {
3674 bridge_address = (stp_get_bridge_id(br->stp)
3675 & ((UINT64_C(1) << 48) - 1));
3677 uint8_t mac[ETH_ADDR_LEN];
3678 eth_addr_random(mac);
3679 bridge_address = eth_addr_to_uint64(mac);
3683 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3685 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3687 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3690 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3692 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3693 br->stp_last_tick = time_msec();
3696 if (bridge_id != stp_get_bridge_id(br->stp)) {
3697 stp_set_bridge_id(br->stp, bridge_id);
3702 for (i = 0; i < br->n_ports; i++) {
3703 struct port *p = br->ports[i];
3705 struct stp_port *sp;
3706 int path_cost, priority;
3712 dp_ifidx = p->ifaces[0]->dp_ifidx;
3713 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3717 sp = stp_get_port(br->stp, dp_ifidx);
3718 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3719 "stp.%s.port.%s.enabled",
3721 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3722 br->name, p->name));
3723 if (p->is_mirror_output_port) {
3726 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3727 bridge_flush(br); /* Might not be necessary. */
3729 stp_port_enable(sp);
3731 stp_port_disable(sp);
3735 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3737 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3739 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3740 "stp.%s.port.%s.priority",
3742 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3744 : STP_DEFAULT_PORT_PRIORITY);
3745 stp_port_set_priority(sp, priority);
3748 brstp_adjust_timers(br);
3750 for (i = 0; i < br->n_ports; i++) {
3751 brstp_update_port_state(br->ports[i]);
3756 brstp_update_port_state(struct port *p)
3758 struct bridge *br = p->bridge;
3759 enum stp_state state;
3761 /* Figure out new state. */
3762 state = STP_DISABLED;
3763 if (br->stp && p->n_ifaces > 0) {
3764 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3765 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3766 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3771 if (p->stp_state != state) {
3772 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3773 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3774 p->name, stp_state_name(p->stp_state),
3775 stp_state_name(state));
3776 if (p->stp_state == STP_DISABLED) {
3779 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3781 p->stp_state = state;
3782 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3783 : tag_create_random());
3788 brstp_adjust_timers(struct bridge *br)
3790 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3791 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3792 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3794 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3795 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3796 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3800 brstp_run(struct bridge *br)
3803 long long int now = time_msec();
3804 long long int elapsed = now - br->stp_last_tick;
3805 struct stp_port *sp;
3808 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3809 br->stp_last_tick = now;
3811 while (stp_get_changed_port(br->stp, &sp)) {
3812 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3814 brstp_update_port_state(p);
3821 brstp_wait(struct bridge *br)
3824 poll_timer_wait(1000);