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 unixctl_cb_func bridge_unixctl_dump_flows;
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 unixctl_cb_func bridge_unixctl_fdb_show;
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, NULL);
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,
323 bridge_reconfigure();
328 config_string_change(const char *key, char **valuep)
330 const char *value = cfg_get_string(0, "%s", key);
331 if (value && (!*valuep || strcmp(value, *valuep))) {
333 *valuep = xstrdup(value);
341 bridge_configure_ssl(void)
343 /* XXX SSL should be configurable on a per-bridge basis.
344 * XXX should be possible to de-configure SSL. */
345 static char *private_key_file;
346 static char *certificate_file;
347 static char *cacert_file;
350 if (config_string_change("ssl.private-key", &private_key_file)) {
351 vconn_ssl_set_private_key_file(private_key_file);
354 if (config_string_change("ssl.certificate", &certificate_file)) {
355 vconn_ssl_set_certificate_file(certificate_file);
358 /* We assume that even if the filename hasn't changed, if the CA cert
359 * file has been removed, that we want to move back into
360 * boot-strapping mode. This opens a small security hole, because
361 * the old certificate will still be trusted until vSwitch is
362 * restarted. We may want to address this in vconn's SSL library. */
363 if (config_string_change("ssl.ca-cert", &cacert_file)
364 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
365 vconn_ssl_set_ca_cert_file(cacert_file,
366 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
371 /* iterate_and_prune_ifaces() callback function that opens the network device
372 * for 'iface', if it is not already open, and retrieves the interface's MAC
373 * address and carrier status. */
375 init_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
380 } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE,
382 netdev_get_carrier(iface->netdev, &iface->enabled);
385 /* If the network device can't be opened, then we're not going to try
386 * to do anything with this interface. */
392 check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED)
394 if (iface->dp_ifidx >= 0) {
395 VLOG_DBG("%s has interface %s on port %d",
397 iface->name, iface->dp_ifidx);
400 VLOG_ERR("%s interface not in %s, dropping",
401 iface->name, dpif_name(br->dpif));
407 set_iface_properties(struct bridge *br UNUSED, struct iface *iface,
412 /* Set policing attributes. */
413 rate = cfg_get_int(0, "port.%s.ingress.policing-rate", iface->name);
414 burst = cfg_get_int(0, "port.%s.ingress.policing-burst", iface->name);
415 netdev_set_policing(iface->netdev, rate, burst);
417 /* Set MAC address of internal interfaces other than the local
419 if (iface->dp_ifidx != ODPP_LOCAL
420 && iface_is_internal(br, iface->name)) {
421 iface_set_mac(iface);
427 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
428 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
429 * deletes from 'br' any ports that no longer have any interfaces. */
431 iterate_and_prune_ifaces(struct bridge *br,
432 bool (*cb)(struct bridge *, struct iface *,
438 for (i = 0; i < br->n_ports; ) {
439 struct port *port = br->ports[i];
440 for (j = 0; j < port->n_ifaces; ) {
441 struct iface *iface = port->ifaces[j];
442 if (cb(br, iface, aux)) {
445 iface_destroy(iface);
449 if (port->n_ifaces) {
452 VLOG_ERR("%s port has no interfaces, dropping", port->name);
459 bridge_reconfigure(void)
461 struct svec old_br, new_br;
462 struct bridge *br, *next;
465 COVERAGE_INC(bridge_reconfigure);
467 /* Collect old and new bridges. */
470 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
471 svec_add(&old_br, br->name);
473 cfg_get_subsections(&new_br, "bridge");
475 /* Get rid of deleted bridges and add new bridges. */
478 assert(svec_is_unique(&old_br));
479 assert(svec_is_unique(&new_br));
480 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
481 if (!svec_contains(&new_br, br->name)) {
485 for (i = 0; i < new_br.n; i++) {
486 const char *name = new_br.names[i];
487 if (!svec_contains(&old_br, name)) {
491 svec_destroy(&old_br);
492 svec_destroy(&new_br);
496 bridge_configure_ssl();
499 /* Reconfigure all bridges. */
500 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
501 bridge_reconfigure_one(br);
504 /* Add and delete ports on all datapaths.
506 * The kernel will reject any attempt to add a given port to a datapath if
507 * that port already belongs to a different datapath, so we must do all
508 * port deletions before any port additions. */
509 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
510 struct odp_port *dpif_ports;
512 struct svec want_ifaces;
514 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
515 bridge_get_all_ifaces(br, &want_ifaces);
516 for (i = 0; i < n_dpif_ports; i++) {
517 const struct odp_port *p = &dpif_ports[i];
518 if (!svec_contains(&want_ifaces, p->devname)
519 && strcmp(p->devname, br->name)) {
520 int retval = dpif_port_del(br->dpif, p->port);
522 VLOG_ERR("failed to remove %s interface from %s: %s",
523 p->devname, dpif_name(br->dpif),
528 svec_destroy(&want_ifaces);
531 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
532 struct odp_port *dpif_ports;
534 struct svec cur_ifaces, want_ifaces, add_ifaces;
536 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
537 svec_init(&cur_ifaces);
538 for (i = 0; i < n_dpif_ports; i++) {
539 svec_add(&cur_ifaces, dpif_ports[i].devname);
542 svec_sort_unique(&cur_ifaces);
543 bridge_get_all_ifaces(br, &want_ifaces);
544 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
546 for (i = 0; i < add_ifaces.n; i++) {
547 const char *if_name = add_ifaces.names[i];
551 /* Add to datapath. */
552 internal = iface_is_internal(br, if_name);
553 error = dpif_port_add(br->dpif, if_name,
554 internal ? ODP_PORT_INTERNAL : 0, NULL);
555 if (error == EFBIG) {
556 VLOG_ERR("ran out of valid port numbers on %s",
557 dpif_name(br->dpif));
560 VLOG_ERR("failed to add %s interface to %s: %s",
561 if_name, dpif_name(br->dpif), strerror(error));
564 svec_destroy(&cur_ifaces);
565 svec_destroy(&want_ifaces);
566 svec_destroy(&add_ifaces);
568 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
571 struct iface *local_iface;
572 struct iface *hw_addr_iface;
573 uint8_t engine_type, engine_id;
574 bool add_id_to_iface = false;
575 struct svec nf_hosts;
577 bridge_fetch_dp_ifaces(br);
578 iterate_and_prune_ifaces(br, init_iface_netdev, NULL);
580 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
582 /* Pick local port hardware address, datapath ID. */
583 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
584 local_iface = bridge_get_local_iface(br);
586 int error = netdev_set_etheraddr(local_iface->netdev, ea);
588 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
589 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
590 "Ethernet address: %s",
591 br->name, strerror(error));
595 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
596 ofproto_set_datapath_id(br->ofproto, dpid);
598 /* Set NetFlow configuration on this bridge. */
599 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
600 if (cfg_has("netflow.%s.engine-type", br->name)) {
601 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
604 if (cfg_has("netflow.%s.engine-id", br->name)) {
605 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
607 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
608 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
611 if (add_id_to_iface && engine_id > 0x7f) {
612 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
613 "another vswitch, choose an engine id less than 128",
616 if (add_id_to_iface && br->n_ports > 0x1ff) {
617 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
618 "another port when 512 or more ports are used",
621 svec_init(&nf_hosts);
622 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
623 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
624 engine_id, add_id_to_iface)) {
625 VLOG_ERR("bridge %s: problem setting netflow collectors",
628 svec_destroy(&nf_hosts);
630 /* Update the controller and related settings. It would be more
631 * straightforward to call this from bridge_reconfigure_one(), but we
632 * can't do it there for two reasons. First, and most importantly, at
633 * that point we don't know the dp_ifidx of any interfaces that have
634 * been added to the bridge (because we haven't actually added them to
635 * the datapath). Second, at that point we haven't set the datapath ID
636 * yet; when a controller is configured, resetting the datapath ID will
637 * immediately disconnect from the controller, so it's better to set
638 * the datapath ID before the controller. */
639 bridge_reconfigure_controller(br);
641 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
642 for (i = 0; i < br->n_ports; i++) {
643 struct port *port = br->ports[i];
645 port_update_vlan_compat(port);
646 port_update_bonding(port);
649 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
650 brstp_reconfigure(br);
651 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
656 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
657 struct iface **hw_addr_iface)
659 uint64_t requested_ea;
663 *hw_addr_iface = NULL;
665 /* Did the user request a particular MAC? */
666 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
668 eth_addr_from_uint64(requested_ea, ea);
669 if (eth_addr_is_multicast(ea)) {
670 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
671 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
672 } else if (eth_addr_is_zero(ea)) {
673 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
679 /* Otherwise choose the minimum MAC address among all of the interfaces.
680 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
681 * MAC of the physical interface in such an environment.) */
682 memset(ea, 0xff, sizeof ea);
683 for (i = 0; i < br->n_ports; i++) {
684 struct port *port = br->ports[i];
685 uint8_t iface_ea[ETH_ADDR_LEN];
686 uint64_t iface_ea_u64;
689 /* Mirror output ports don't participate. */
690 if (port->is_mirror_output_port) {
694 /* Choose the MAC address to represent the port. */
695 iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name);
697 /* User specified explicitly. */
698 eth_addr_from_uint64(iface_ea_u64, iface_ea);
700 /* Find the interface with this Ethernet address (if any) so that
701 * we can provide the correct devname to the caller. */
703 for (j = 0; j < port->n_ifaces; j++) {
704 struct iface *candidate = port->ifaces[j];
705 uint8_t candidate_ea[ETH_ADDR_LEN];
706 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
707 && eth_addr_equals(iface_ea, candidate_ea)) {
712 /* Choose the interface whose MAC address will represent the port.
713 * The Linux kernel bonding code always chooses the MAC address of
714 * the first slave added to a bond, and the Fedora networking
715 * scripts always add slaves to a bond in alphabetical order, so
716 * for compatibility we choose the interface with the name that is
717 * first in alphabetical order. */
718 iface = port->ifaces[0];
719 for (j = 1; j < port->n_ifaces; j++) {
720 struct iface *candidate = port->ifaces[j];
721 if (strcmp(candidate->name, iface->name) < 0) {
726 /* The local port doesn't count (since we're trying to choose its
727 * MAC address anyway). Other internal ports don't count because
728 * we really want a physical MAC if we can get it, and internal
729 * ports typically have randomly generated MACs. */
730 if (iface->dp_ifidx == ODPP_LOCAL
731 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
736 error = netdev_get_etheraddr(iface->netdev, iface_ea);
738 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
739 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
740 iface->name, strerror(error));
745 /* Compare against our current choice. */
746 if (!eth_addr_is_multicast(iface_ea) &&
747 !eth_addr_is_reserved(iface_ea) &&
748 !eth_addr_is_zero(iface_ea) &&
749 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
751 memcpy(ea, iface_ea, ETH_ADDR_LEN);
752 *hw_addr_iface = iface;
755 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
756 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
757 *hw_addr_iface = NULL;
758 VLOG_WARN("bridge %s: using default bridge Ethernet "
759 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
761 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
762 br->name, ETH_ADDR_ARGS(ea));
766 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
767 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
768 * an interface on 'br', then that interface must be passed in as
769 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
770 * 'hw_addr_iface' must be passed in as a null pointer. */
772 bridge_pick_datapath_id(struct bridge *br,
773 const uint8_t bridge_ea[ETH_ADDR_LEN],
774 struct iface *hw_addr_iface)
777 * The procedure for choosing a bridge MAC address will, in the most
778 * ordinary case, also choose a unique MAC that we can use as a datapath
779 * ID. In some special cases, though, multiple bridges will end up with
780 * the same MAC address. This is OK for the bridges, but it will confuse
781 * the OpenFlow controller, because each datapath needs a unique datapath
784 * Datapath IDs must be unique. It is also very desirable that they be
785 * stable from one run to the next, so that policy set on a datapath
790 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
797 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
799 * A bridge whose MAC address is taken from a VLAN network device
800 * (that is, a network device created with vconfig(8) or similar
801 * tool) will have the same MAC address as a bridge on the VLAN
802 * device's physical network device.
804 * Handle this case by hashing the physical network device MAC
805 * along with the VLAN identifier.
807 uint8_t buf[ETH_ADDR_LEN + 2];
808 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
809 buf[ETH_ADDR_LEN] = vlan >> 8;
810 buf[ETH_ADDR_LEN + 1] = vlan;
811 return dpid_from_hash(buf, sizeof buf);
814 * Assume that this bridge's MAC address is unique, since it
815 * doesn't fit any of the cases we handle specially.
820 * A purely internal bridge, that is, one that has no non-virtual
821 * network devices on it at all, is more difficult because it has no
822 * natural unique identifier at all.
824 * When the host is a XenServer, we handle this case by hashing the
825 * host's UUID with the name of the bridge. Names of bridges are
826 * persistent across XenServer reboots, although they can be reused if
827 * an internal network is destroyed and then a new one is later
828 * created, so this is fairly effective.
830 * When the host is not a XenServer, we punt by using a random MAC
831 * address on each run.
833 const char *host_uuid = xenserver_get_host_uuid();
835 char *combined = xasprintf("%s,%s", host_uuid, br->name);
836 dpid = dpid_from_hash(combined, strlen(combined));
842 return eth_addr_to_uint64(bridge_ea);
846 dpid_from_hash(const void *data, size_t n)
848 uint8_t hash[SHA1_DIGEST_SIZE];
850 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
851 sha1_bytes(data, n, hash);
852 eth_addr_mark_random(hash);
853 return eth_addr_to_uint64(hash);
859 struct bridge *br, *next;
863 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
864 int error = bridge_run_one(br);
866 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
867 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
868 "forcing reconfiguration", br->name);
882 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
883 ofproto_wait(br->ofproto);
884 if (br->controller) {
889 mac_learning_wait(br->ml);
896 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
897 * configuration changes. */
899 bridge_flush(struct bridge *br)
901 COVERAGE_INC(bridge_flush);
904 mac_learning_flush(br->ml);
908 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
910 static struct iface *
911 bridge_get_local_iface(struct bridge *br)
915 for (i = 0; i < br->n_ports; i++) {
916 struct port *port = br->ports[i];
917 for (j = 0; j < port->n_ifaces; j++) {
918 struct iface *iface = port->ifaces[j];
919 if (iface->dp_ifidx == ODPP_LOCAL) {
928 /* Bridge unixctl user interface functions. */
930 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
931 const char *args, void *aux UNUSED)
933 struct ds ds = DS_EMPTY_INITIALIZER;
934 const struct bridge *br;
936 br = bridge_lookup(args);
938 unixctl_command_reply(conn, 501, "no such bridge");
942 ds_put_cstr(&ds, " port VLAN MAC Age\n");
944 const struct mac_entry *e;
945 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
946 if (e->port < 0 || e->port >= br->n_ports) {
949 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
950 br->ports[e->port]->ifaces[0]->dp_ifidx,
951 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
954 unixctl_command_reply(conn, 200, ds_cstr(&ds));
958 /* Bridge reconfiguration functions. */
960 static struct bridge *
961 bridge_create(const char *name)
966 assert(!bridge_lookup(name));
967 br = xzalloc(sizeof *br);
969 error = dpif_create(name, &br->dpif);
970 if (error == EEXIST || error == EBUSY) {
971 error = dpif_open(name, &br->dpif);
973 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
974 name, strerror(error));
978 dpif_flow_flush(br->dpif);
980 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
985 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
987 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
988 dpif_delete(br->dpif);
989 dpif_close(br->dpif);
994 br->name = xstrdup(name);
995 br->ml = mac_learning_create();
996 br->sent_config_request = false;
997 eth_addr_random(br->default_ea);
999 port_array_init(&br->ifaces);
1002 br->bond_next_rebalance = time_msec() + 10000;
1004 list_push_back(&all_bridges, &br->node);
1006 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1012 bridge_destroy(struct bridge *br)
1017 while (br->n_ports > 0) {
1018 port_destroy(br->ports[br->n_ports - 1]);
1020 list_remove(&br->node);
1021 error = dpif_delete(br->dpif);
1022 if (error && error != ENOENT) {
1023 VLOG_ERR("failed to delete %s: %s",
1024 dpif_name(br->dpif), strerror(error));
1026 dpif_close(br->dpif);
1027 ofproto_destroy(br->ofproto);
1028 free(br->controller);
1029 mac_learning_destroy(br->ml);
1030 port_array_destroy(&br->ifaces);
1037 static struct bridge *
1038 bridge_lookup(const char *name)
1042 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1043 if (!strcmp(br->name, name)) {
1051 bridge_exists(const char *name)
1053 return bridge_lookup(name) ? true : false;
1057 bridge_get_datapathid(const char *name)
1059 struct bridge *br = bridge_lookup(name);
1060 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1063 /* Handle requests for a listing of all flows known by the OpenFlow
1064 * stack, including those normally hidden. */
1066 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1067 const char *args, void *aux UNUSED)
1072 br = bridge_lookup(args);
1074 unixctl_command_reply(conn, 501, "Unknown bridge");
1079 ofproto_get_all_flows(br->ofproto, &results);
1081 unixctl_command_reply(conn, 200, ds_cstr(&results));
1082 ds_destroy(&results);
1086 bridge_run_one(struct bridge *br)
1090 error = ofproto_run1(br->ofproto);
1096 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1101 error = ofproto_run2(br->ofproto, br->flush);
1108 bridge_get_controller(const struct bridge *br)
1110 const char *controller;
1112 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1114 controller = cfg_get_string(0, "mgmt.controller");
1116 return controller && controller[0] ? controller : NULL;
1120 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1122 struct svec *ifaces = ifaces_;
1123 if (!svec_contains(ifaces, iface->name)) {
1124 svec_add(ifaces, iface->name);
1128 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1130 br->name, iface->name, iface->port->name);
1136 bridge_reconfigure_one(struct bridge *br)
1138 struct svec old_ports, new_ports, ifaces;
1139 struct svec listeners, old_listeners;
1140 struct svec snoops, old_snoops;
1143 /* Collect old ports. */
1144 svec_init(&old_ports);
1145 for (i = 0; i < br->n_ports; i++) {
1146 svec_add(&old_ports, br->ports[i]->name);
1148 svec_sort(&old_ports);
1149 assert(svec_is_unique(&old_ports));
1151 /* Collect new ports. */
1152 svec_init(&new_ports);
1153 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1154 svec_sort(&new_ports);
1155 if (bridge_get_controller(br)) {
1156 char local_name[IF_NAMESIZE];
1159 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1160 local_name, sizeof local_name);
1161 if (!error && !svec_contains(&new_ports, local_name)) {
1162 svec_add(&new_ports, local_name);
1163 svec_sort(&new_ports);
1166 if (!svec_is_unique(&new_ports)) {
1167 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1168 br->name, svec_get_duplicate(&new_ports));
1169 svec_unique(&new_ports);
1172 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1174 /* Get rid of deleted ports and add new ports. */
1175 for (i = 0; i < br->n_ports; ) {
1176 struct port *port = br->ports[i];
1177 if (!svec_contains(&new_ports, port->name)) {
1183 for (i = 0; i < new_ports.n; i++) {
1184 const char *name = new_ports.names[i];
1185 if (!svec_contains(&old_ports, name)) {
1186 port_create(br, name);
1189 svec_destroy(&old_ports);
1190 svec_destroy(&new_ports);
1192 /* Reconfigure all ports. */
1193 for (i = 0; i < br->n_ports; i++) {
1194 port_reconfigure(br->ports[i]);
1197 /* Check and delete duplicate interfaces. */
1199 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1200 svec_destroy(&ifaces);
1202 /* Delete all flows if we're switching from connected to standalone or vice
1203 * versa. (XXX Should we delete all flows if we are switching from one
1204 * controller to another?) */
1206 /* Configure OpenFlow management listeners. */
1207 svec_init(&listeners);
1208 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1210 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1211 ovs_rundir, br->name));
1212 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1213 svec_clear(&listeners);
1215 svec_sort_unique(&listeners);
1217 svec_init(&old_listeners);
1218 ofproto_get_listeners(br->ofproto, &old_listeners);
1219 svec_sort_unique(&old_listeners);
1221 if (!svec_equal(&listeners, &old_listeners)) {
1222 ofproto_set_listeners(br->ofproto, &listeners);
1224 svec_destroy(&listeners);
1225 svec_destroy(&old_listeners);
1227 /* Configure OpenFlow controller connection snooping. */
1229 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1231 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1232 ovs_rundir, br->name));
1233 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1234 svec_clear(&snoops);
1236 svec_sort_unique(&snoops);
1238 svec_init(&old_snoops);
1239 ofproto_get_snoops(br->ofproto, &old_snoops);
1240 svec_sort_unique(&old_snoops);
1242 if (!svec_equal(&snoops, &old_snoops)) {
1243 ofproto_set_snoops(br->ofproto, &snoops);
1245 svec_destroy(&snoops);
1246 svec_destroy(&old_snoops);
1248 mirror_reconfigure(br);
1252 bridge_reconfigure_controller(struct bridge *br)
1254 char *pfx = xasprintf("bridge.%s.controller", br->name);
1255 const char *controller;
1257 controller = bridge_get_controller(br);
1258 if ((br->controller != NULL) != (controller != NULL)) {
1259 ofproto_flush_flows(br->ofproto);
1261 free(br->controller);
1262 br->controller = controller ? xstrdup(controller) : NULL;
1265 const char *fail_mode;
1266 int max_backoff, probe;
1267 int rate_limit, burst_limit;
1269 if (!strcmp(controller, "discover")) {
1270 bool update_resolv_conf = true;
1272 if (cfg_has("%s.update-resolv.conf", pfx)) {
1273 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1276 ofproto_set_discovery(br->ofproto, true,
1277 cfg_get_string(0, "%s.accept-regex", pfx),
1278 update_resolv_conf);
1280 struct iface *local_iface;
1283 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1285 || cfg_get_bool(0, "%s.in-band", pfx));
1286 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1287 ofproto_set_in_band(br->ofproto, in_band);
1289 local_iface = bridge_get_local_iface(br);
1291 && cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1292 struct netdev *netdev = local_iface->netdev;
1293 struct in_addr ip, mask, gateway;
1294 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1295 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1296 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1298 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1300 mask.s_addr = guess_netmask(ip.s_addr);
1302 if (!netdev_set_in4(netdev, ip, mask)) {
1303 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1305 br->name, IP_ARGS(&ip.s_addr),
1306 IP_ARGS(&mask.s_addr));
1309 if (gateway.s_addr) {
1310 if (!netdev_add_router(netdev, gateway)) {
1311 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1312 br->name, IP_ARGS(&gateway.s_addr));
1318 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1320 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1322 ofproto_set_failure(br->ofproto,
1324 || !strcmp(fail_mode, "standalone")
1325 || !strcmp(fail_mode, "open")));
1327 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1329 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1334 ofproto_set_probe_interval(br->ofproto, probe);
1336 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1338 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1343 ofproto_set_max_backoff(br->ofproto, max_backoff);
1345 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1347 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1349 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1351 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1353 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1355 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1357 if (cfg_has("%s.commands.acl", pfx)) {
1358 struct svec command_acls;
1361 svec_init(&command_acls);
1362 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1363 command_acl = svec_join(&command_acls, ",", "");
1365 ofproto_set_remote_execution(br->ofproto, command_acl,
1366 cfg_get_string(0, "%s.commands.dir",
1369 svec_destroy(&command_acls);
1372 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1375 union ofp_action action;
1378 /* Set up a flow that matches every packet and directs them to
1379 * OFPP_NORMAL (which goes to us). */
1380 memset(&action, 0, sizeof action);
1381 action.type = htons(OFPAT_OUTPUT);
1382 action.output.len = htons(sizeof action);
1383 action.output.port = htons(OFPP_NORMAL);
1384 memset(&flow, 0, sizeof flow);
1385 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1388 ofproto_set_in_band(br->ofproto, false);
1389 ofproto_set_max_backoff(br->ofproto, 1);
1390 ofproto_set_probe_interval(br->ofproto, 5);
1391 ofproto_set_failure(br->ofproto, false);
1392 ofproto_set_stp(br->ofproto, false);
1396 ofproto_set_controller(br->ofproto, br->controller);
1400 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1405 for (i = 0; i < br->n_ports; i++) {
1406 struct port *port = br->ports[i];
1407 for (j = 0; j < port->n_ifaces; j++) {
1408 struct iface *iface = port->ifaces[j];
1409 svec_add(ifaces, iface->name);
1411 if (port->n_ifaces > 1
1412 && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
1413 svec_add(ifaces, port->name);
1416 svec_sort_unique(ifaces);
1419 /* For robustness, in case the administrator moves around datapath ports behind
1420 * our back, we re-check all the datapath port numbers here.
1422 * This function will set the 'dp_ifidx' members of interfaces that have
1423 * disappeared to -1, so only call this function from a context where those
1424 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1425 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1426 * datapath, which doesn't support UINT16_MAX+1 ports. */
1428 bridge_fetch_dp_ifaces(struct bridge *br)
1430 struct odp_port *dpif_ports;
1431 size_t n_dpif_ports;
1434 /* Reset all interface numbers. */
1435 for (i = 0; i < br->n_ports; i++) {
1436 struct port *port = br->ports[i];
1437 for (j = 0; j < port->n_ifaces; j++) {
1438 struct iface *iface = port->ifaces[j];
1439 iface->dp_ifidx = -1;
1442 port_array_clear(&br->ifaces);
1444 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1445 for (i = 0; i < n_dpif_ports; i++) {
1446 struct odp_port *p = &dpif_ports[i];
1447 struct iface *iface = iface_lookup(br, p->devname);
1449 if (iface->dp_ifidx >= 0) {
1450 VLOG_WARN("%s reported interface %s twice",
1451 dpif_name(br->dpif), p->devname);
1452 } else if (iface_from_dp_ifidx(br, p->port)) {
1453 VLOG_WARN("%s reported interface %"PRIu16" twice",
1454 dpif_name(br->dpif), p->port);
1456 port_array_set(&br->ifaces, p->port, iface);
1457 iface->dp_ifidx = p->port;
1464 /* Bridge packet processing functions. */
1467 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1469 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1472 static struct bond_entry *
1473 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1475 return &port->bond_hash[bond_hash(mac)];
1479 bond_choose_iface(const struct port *port)
1482 for (i = 0; i < port->n_ifaces; i++) {
1483 if (port->ifaces[i]->enabled) {
1491 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1492 uint16_t *dp_ifidx, tag_type *tags)
1494 struct iface *iface;
1496 assert(port->n_ifaces);
1497 if (port->n_ifaces == 1) {
1498 iface = port->ifaces[0];
1500 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1501 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1502 || !port->ifaces[e->iface_idx]->enabled) {
1503 /* XXX select interface properly. The current interface selection
1504 * is only good for testing the rebalancing code. */
1505 e->iface_idx = bond_choose_iface(port);
1506 if (e->iface_idx < 0) {
1507 *tags |= port->no_ifaces_tag;
1510 e->iface_tag = tag_create_random();
1511 ((struct port *) port)->bond_compat_is_stale = true;
1513 *tags |= e->iface_tag;
1514 iface = port->ifaces[e->iface_idx];
1516 *dp_ifidx = iface->dp_ifidx;
1517 *tags |= iface->tag; /* Currently only used for bonding. */
1522 bond_link_status_update(struct iface *iface, bool carrier)
1524 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1525 struct port *port = iface->port;
1527 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1528 /* Nothing to do. */
1531 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1532 iface->name, carrier ? "detected" : "dropped");
1533 if (carrier == iface->enabled) {
1534 iface->delay_expires = LLONG_MAX;
1535 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1536 iface->name, carrier ? "disabled" : "enabled");
1537 } else if (carrier && port->updelay && port->active_iface < 0) {
1538 iface->delay_expires = time_msec();
1539 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1540 "other interface is up", iface->name, port->updelay);
1542 int delay = carrier ? port->updelay : port->downdelay;
1543 iface->delay_expires = time_msec() + delay;
1546 "interface %s: will be %s if it stays %s for %d ms",
1548 carrier ? "enabled" : "disabled",
1549 carrier ? "up" : "down",
1556 bond_choose_active_iface(struct port *port)
1558 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1560 port->active_iface = bond_choose_iface(port);
1561 port->active_iface_tag = tag_create_random();
1562 if (port->active_iface >= 0) {
1563 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1564 port->name, port->ifaces[port->active_iface]->name);
1566 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1572 bond_enable_slave(struct iface *iface, bool enable)
1574 struct port *port = iface->port;
1575 struct bridge *br = port->bridge;
1577 iface->delay_expires = LLONG_MAX;
1578 if (enable == iface->enabled) {
1582 iface->enabled = enable;
1583 if (!iface->enabled) {
1584 VLOG_WARN("interface %s: disabled", iface->name);
1585 ofproto_revalidate(br->ofproto, iface->tag);
1586 if (iface->port_ifidx == port->active_iface) {
1587 ofproto_revalidate(br->ofproto,
1588 port->active_iface_tag);
1589 bond_choose_active_iface(port);
1591 bond_send_learning_packets(port);
1593 VLOG_WARN("interface %s: enabled", iface->name);
1594 if (port->active_iface < 0) {
1595 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1596 bond_choose_active_iface(port);
1597 bond_send_learning_packets(port);
1599 iface->tag = tag_create_random();
1601 port_update_bond_compat(port);
1605 bond_run(struct bridge *br)
1609 for (i = 0; i < br->n_ports; i++) {
1610 struct port *port = br->ports[i];
1612 if (port->bond_compat_is_stale) {
1613 port->bond_compat_is_stale = false;
1614 port_update_bond_compat(port);
1617 if (port->n_ifaces < 2) {
1620 for (j = 0; j < port->n_ifaces; j++) {
1621 struct iface *iface = port->ifaces[j];
1622 if (time_msec() >= iface->delay_expires) {
1623 bond_enable_slave(iface, !iface->enabled);
1630 bond_wait(struct bridge *br)
1634 for (i = 0; i < br->n_ports; i++) {
1635 struct port *port = br->ports[i];
1636 if (port->n_ifaces < 2) {
1639 for (j = 0; j < port->n_ifaces; j++) {
1640 struct iface *iface = port->ifaces[j];
1641 if (iface->delay_expires != LLONG_MAX) {
1642 poll_timer_wait(iface->delay_expires - time_msec());
1649 set_dst(struct dst *p, const flow_t *flow,
1650 const struct port *in_port, const struct port *out_port,
1655 * XXX This uses too many tags: any broadcast flow will get one tag per
1656 * destination port, and thus a broadcast on a switch of any size is likely
1657 * to have all tag bits set. We should figure out a way to be smarter.
1659 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1660 *tags |= out_port->stp_state_tag;
1661 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1665 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1666 : in_port->vlan >= 0 ? in_port->vlan
1667 : ntohs(flow->dl_vlan));
1668 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1672 swap_dst(struct dst *p, struct dst *q)
1674 struct dst tmp = *p;
1679 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1680 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1681 * that we push to the datapath. We could in fact fully sort the array by
1682 * vlan, but in most cases there are at most two different vlan tags so that's
1683 * possibly overkill.) */
1685 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1687 struct dst *first = dsts;
1688 struct dst *last = dsts + n_dsts;
1690 while (first != last) {
1692 * - All dsts < first have vlan == 'vlan'.
1693 * - All dsts >= last have vlan != 'vlan'.
1694 * - first < last. */
1695 while (first->vlan == vlan) {
1696 if (++first == last) {
1701 /* Same invariants, plus one additional:
1702 * - first->vlan != vlan.
1704 while (last[-1].vlan != vlan) {
1705 if (--last == first) {
1710 /* Same invariants, plus one additional:
1711 * - last[-1].vlan == vlan.*/
1712 swap_dst(first++, --last);
1717 mirror_mask_ffs(mirror_mask_t mask)
1719 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1724 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1725 const struct dst *test)
1728 for (i = 0; i < n_dsts; i++) {
1729 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1737 port_trunks_vlan(const struct port *port, uint16_t vlan)
1739 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1743 port_includes_vlan(const struct port *port, uint16_t vlan)
1745 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1749 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1750 const struct port *in_port, const struct port *out_port,
1751 struct dst dsts[], tag_type *tags)
1753 mirror_mask_t mirrors = in_port->src_mirrors;
1754 struct dst *dst = dsts;
1757 *tags |= in_port->stp_state_tag;
1758 if (out_port == FLOOD_PORT) {
1759 /* XXX use ODP_FLOOD if no vlans or bonding. */
1760 /* XXX even better, define each VLAN as a datapath port group */
1761 for (i = 0; i < br->n_ports; i++) {
1762 struct port *port = br->ports[i];
1763 if (port != in_port && port_includes_vlan(port, vlan)
1764 && !port->is_mirror_output_port
1765 && set_dst(dst, flow, in_port, port, tags)) {
1766 mirrors |= port->dst_mirrors;
1770 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1771 mirrors |= out_port->dst_mirrors;
1776 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1777 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1779 if (set_dst(dst, flow, in_port, m->out_port, tags)
1780 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1784 for (i = 0; i < br->n_ports; i++) {
1785 struct port *port = br->ports[i];
1786 if (port_includes_vlan(port, m->out_vlan)
1787 && set_dst(dst, flow, in_port, port, tags))
1791 if (port->vlan < 0) {
1792 dst->vlan = m->out_vlan;
1794 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1798 /* Use the vlan tag on the original flow instead of
1799 * the one passed in the vlan parameter. This ensures
1800 * that we compare the vlan from before any implicit
1801 * tagging tags place. This is necessary because
1802 * dst->vlan is the final vlan, after removing implicit
1804 flow_vlan = ntohs(flow->dl_vlan);
1805 if (flow_vlan == 0) {
1806 flow_vlan = OFP_VLAN_NONE;
1808 if (port == in_port && dst->vlan == flow_vlan) {
1809 /* Don't send out input port on same VLAN. */
1817 mirrors &= mirrors - 1;
1820 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1825 print_dsts(const struct dst *dsts, size_t n)
1827 for (; n--; dsts++) {
1828 printf(">p%"PRIu16, dsts->dp_ifidx);
1829 if (dsts->vlan != OFP_VLAN_NONE) {
1830 printf("v%"PRIu16, dsts->vlan);
1836 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1837 const struct port *in_port, const struct port *out_port,
1838 tag_type *tags, struct odp_actions *actions)
1840 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1842 const struct dst *p;
1845 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1847 cur_vlan = ntohs(flow->dl_vlan);
1848 for (p = dsts; p < &dsts[n_dsts]; p++) {
1849 union odp_action *a;
1850 if (p->vlan != cur_vlan) {
1851 if (p->vlan == OFP_VLAN_NONE) {
1852 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1854 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1855 a->vlan_vid.vlan_vid = htons(p->vlan);
1859 a = odp_actions_add(actions, ODPAT_OUTPUT);
1860 a->output.port = p->dp_ifidx;
1865 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1867 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1868 return (flow->dl_type == htons(ETH_TYPE_ARP)
1869 && eth_addr_is_broadcast(flow->dl_dst)
1870 && packet->size >= sizeof(struct arp_eth_header)
1871 && arp->ar_op == ARP_OP_REQUEST);
1874 /* If the composed actions may be applied to any packet in the given 'flow',
1875 * returns true. Otherwise, the actions should only be applied to 'packet', or
1876 * not at all, if 'packet' was NULL. */
1878 process_flow(struct bridge *br, const flow_t *flow,
1879 const struct ofpbuf *packet, struct odp_actions *actions,
1882 struct iface *in_iface;
1883 struct port *in_port;
1884 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1887 /* Find the interface and port structure for the received packet. */
1888 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1890 /* No interface? Something fishy... */
1891 if (packet != NULL) {
1892 /* Odd. A few possible reasons here:
1894 * - We deleted an interface but there are still a few packets
1895 * queued up from it.
1897 * - Someone externally added an interface (e.g. with "ovs-dpctl
1898 * add-if") that we don't know about.
1900 * - Packet arrived on the local port but the local port is not
1901 * one of our bridge ports.
1903 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1905 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1906 "interface %"PRIu16, br->name, flow->in_port);
1909 /* Return without adding any actions, to drop packets on this flow. */
1912 in_port = in_iface->port;
1914 /* Figure out what VLAN this packet belongs to.
1916 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1917 * belongs to VLAN 0, so we should treat both cases identically. (In the
1918 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1919 * presumably to allow a priority to be specified. In the latter case, the
1920 * packet does not have any 802.1Q header.) */
1921 vlan = ntohs(flow->dl_vlan);
1922 if (vlan == OFP_VLAN_NONE) {
1925 if (in_port->vlan >= 0) {
1927 /* XXX support double tagging? */
1928 if (packet != NULL) {
1929 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1930 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1931 "packet received on port %s configured with "
1932 "implicit VLAN %"PRIu16,
1933 br->name, ntohs(flow->dl_vlan),
1934 in_port->name, in_port->vlan);
1938 vlan = in_port->vlan;
1940 if (!port_includes_vlan(in_port, vlan)) {
1941 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1942 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1943 "packet received on port %s not configured for "
1945 br->name, vlan, in_port->name, vlan);
1950 /* Drop frames for ports that STP wants entirely killed (both for
1951 * forwarding and for learning). Later, after we do learning, we'll drop
1952 * the frames that STP wants to do learning but not forwarding on. */
1953 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1957 /* Drop frames for reserved multicast addresses. */
1958 if (eth_addr_is_reserved(flow->dl_dst)) {
1962 /* Drop frames on ports reserved for mirroring. */
1963 if (in_port->is_mirror_output_port) {
1964 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1965 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1966 "which is reserved exclusively for mirroring",
1967 br->name, in_port->name);
1971 /* Packets received on bonds need special attention to avoid duplicates. */
1972 if (in_port->n_ifaces > 1) {
1975 if (eth_addr_is_multicast(flow->dl_dst)) {
1976 *tags |= in_port->active_iface_tag;
1977 if (in_port->active_iface != in_iface->port_ifidx) {
1978 /* Drop all multicast packets on inactive slaves. */
1983 /* Drop all packets for which we have learned a different input
1984 * port, because we probably sent the packet on one slave and got
1985 * it back on the other. Broadcast ARP replies are an exception
1986 * to this rule: the host has moved to another switch. */
1987 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1988 if (src_idx != -1 && src_idx != in_port->port_idx &&
1989 (!packet || !is_bcast_arp_reply(flow, packet))) {
1995 out_port = FLOOD_PORT;
1999 /* Learn source MAC (but don't try to learn from revalidation). */
2001 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2002 vlan, in_port->port_idx);
2004 /* The log messages here could actually be useful in debugging,
2005 * so keep the rate limit relatively high. */
2006 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2008 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2009 "on port %s in VLAN %d",
2010 br->name, ETH_ADDR_ARGS(flow->dl_src),
2011 in_port->name, vlan);
2012 ofproto_revalidate(br->ofproto, rev_tag);
2016 /* Determine output port. */
2017 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2019 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2020 out_port = br->ports[out_port_idx];
2021 } else if (!packet) {
2022 /* If we are revalidating but don't have a learning entry then
2023 * eject the flow. Installing a flow that floods packets will
2024 * prevent us from seeing future packets and learning properly. */
2029 /* Don't send packets out their input ports. Don't forward frames that STP
2030 * wants us to discard. */
2031 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
2036 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
2039 * We send out only a single packet, instead of setting up a flow, if the
2040 * packet is an ARP directed to broadcast that arrived on a bonded
2041 * interface. In such a situation ARP requests and replies must be handled
2042 * differently, but OpenFlow unfortunately can't distinguish them.
2044 return (in_port->n_ifaces < 2
2045 || flow->dl_type != htons(ETH_TYPE_ARP)
2046 || !eth_addr_is_broadcast(flow->dl_dst));
2049 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2052 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2053 const struct ofp_phy_port *opp,
2056 struct bridge *br = br_;
2057 struct iface *iface;
2060 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2066 if (reason == OFPPR_DELETE) {
2067 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2068 br->name, iface->name);
2069 iface_destroy(iface);
2070 if (!port->n_ifaces) {
2071 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2072 br->name, port->name);
2078 if (port->n_ifaces > 1) {
2079 bool up = !(opp->state & OFPPS_LINK_DOWN);
2080 bond_link_status_update(iface, up);
2081 port_update_bond_compat(port);
2087 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2088 struct odp_actions *actions, tag_type *tags, void *br_)
2090 struct bridge *br = br_;
2093 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
2094 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
2095 brstp_receive(br, flow, payload);
2100 COVERAGE_INC(bridge_process_flow);
2101 return process_flow(br, flow, packet, actions, tags);
2105 bridge_account_flow_ofhook_cb(const flow_t *flow,
2106 const union odp_action *actions,
2107 size_t n_actions, unsigned long long int n_bytes,
2110 struct bridge *br = br_;
2111 const union odp_action *a;
2113 if (!br->has_bonded_ports) {
2117 for (a = actions; a < &actions[n_actions]; a++) {
2118 if (a->type == ODPAT_OUTPUT) {
2119 struct port *port = port_from_dp_ifidx(br, a->output.port);
2120 if (port && port->n_ifaces >= 2) {
2121 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2122 e->tx_bytes += n_bytes;
2129 bridge_account_checkpoint_ofhook_cb(void *br_)
2131 struct bridge *br = br_;
2134 if (!br->has_bonded_ports) {
2138 /* The current ofproto implementation calls this callback at least once a
2139 * second, so this timer implementation is sufficient. */
2140 if (time_msec() < br->bond_next_rebalance) {
2143 br->bond_next_rebalance = time_msec() + 10000;
2145 for (i = 0; i < br->n_ports; i++) {
2146 struct port *port = br->ports[i];
2147 if (port->n_ifaces > 1) {
2148 bond_rebalance_port(port);
2153 static struct ofhooks bridge_ofhooks = {
2154 bridge_port_changed_ofhook_cb,
2155 bridge_normal_ofhook_cb,
2156 bridge_account_flow_ofhook_cb,
2157 bridge_account_checkpoint_ofhook_cb,
2160 /* Bonding functions. */
2162 /* Statistics for a single interface on a bonded port, used for load-based
2163 * bond rebalancing. */
2164 struct slave_balance {
2165 struct iface *iface; /* The interface. */
2166 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2168 /* All the "bond_entry"s that are assigned to this interface, in order of
2169 * increasing tx_bytes. */
2170 struct bond_entry **hashes;
2174 /* Sorts pointers to pointers to bond_entries in ascending order by the
2175 * interface to which they are assigned, and within a single interface in
2176 * ascending order of bytes transmitted. */
2178 compare_bond_entries(const void *a_, const void *b_)
2180 const struct bond_entry *const *ap = a_;
2181 const struct bond_entry *const *bp = b_;
2182 const struct bond_entry *a = *ap;
2183 const struct bond_entry *b = *bp;
2184 if (a->iface_idx != b->iface_idx) {
2185 return a->iface_idx > b->iface_idx ? 1 : -1;
2186 } else if (a->tx_bytes != b->tx_bytes) {
2187 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2193 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2194 * *descending* order by number of bytes transmitted. */
2196 compare_slave_balance(const void *a_, const void *b_)
2198 const struct slave_balance *a = a_;
2199 const struct slave_balance *b = b_;
2200 if (a->iface->enabled != b->iface->enabled) {
2201 return a->iface->enabled ? -1 : 1;
2202 } else if (a->tx_bytes != b->tx_bytes) {
2203 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2210 swap_bals(struct slave_balance *a, struct slave_balance *b)
2212 struct slave_balance tmp = *a;
2217 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2218 * given that 'p' (and only 'p') might be in the wrong location.
2220 * This function invalidates 'p', since it might now be in a different memory
2223 resort_bals(struct slave_balance *p,
2224 struct slave_balance bals[], size_t n_bals)
2227 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2228 swap_bals(p, p - 1);
2230 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2231 swap_bals(p, p + 1);
2237 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2239 if (VLOG_IS_DBG_ENABLED()) {
2240 struct ds ds = DS_EMPTY_INITIALIZER;
2241 const struct slave_balance *b;
2243 for (b = bals; b < bals + n_bals; b++) {
2247 ds_put_char(&ds, ',');
2249 ds_put_format(&ds, " %s %"PRIu64"kB",
2250 b->iface->name, b->tx_bytes / 1024);
2252 if (!b->iface->enabled) {
2253 ds_put_cstr(&ds, " (disabled)");
2255 if (b->n_hashes > 0) {
2256 ds_put_cstr(&ds, " (");
2257 for (i = 0; i < b->n_hashes; i++) {
2258 const struct bond_entry *e = b->hashes[i];
2260 ds_put_cstr(&ds, " + ");
2262 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2263 e - port->bond_hash, e->tx_bytes / 1024);
2265 ds_put_cstr(&ds, ")");
2268 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2273 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2275 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2278 struct bond_entry *hash = from->hashes[hash_idx];
2279 struct port *port = from->iface->port;
2280 uint64_t delta = hash->tx_bytes;
2282 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2283 "from %s to %s (now carrying %"PRIu64"kB and "
2284 "%"PRIu64"kB load, respectively)",
2285 port->name, delta / 1024, hash - port->bond_hash,
2286 from->iface->name, to->iface->name,
2287 (from->tx_bytes - delta) / 1024,
2288 (to->tx_bytes + delta) / 1024);
2290 /* Delete element from from->hashes.
2292 * We don't bother to add the element to to->hashes because not only would
2293 * it require more work, the only purpose it would be to allow that hash to
2294 * be migrated to another slave in this rebalancing run, and there is no
2295 * point in doing that. */
2296 if (hash_idx == 0) {
2299 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2300 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2304 /* Shift load away from 'from' to 'to'. */
2305 from->tx_bytes -= delta;
2306 to->tx_bytes += delta;
2308 /* Arrange for flows to be revalidated. */
2309 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2310 hash->iface_idx = to->iface->port_ifidx;
2311 hash->iface_tag = tag_create_random();
2315 bond_rebalance_port(struct port *port)
2317 struct slave_balance bals[DP_MAX_PORTS];
2319 struct bond_entry *hashes[BOND_MASK + 1];
2320 struct slave_balance *b, *from, *to;
2321 struct bond_entry *e;
2324 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2325 * descending order of tx_bytes, so that bals[0] represents the most
2326 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2329 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2330 * array for each slave_balance structure, we sort our local array of
2331 * hashes in order by slave, so that all of the hashes for a given slave
2332 * become contiguous in memory, and then we point each 'hashes' members of
2333 * a slave_balance structure to the start of a contiguous group. */
2334 n_bals = port->n_ifaces;
2335 for (b = bals; b < &bals[n_bals]; b++) {
2336 b->iface = port->ifaces[b - bals];
2341 for (i = 0; i <= BOND_MASK; i++) {
2342 hashes[i] = &port->bond_hash[i];
2344 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2345 for (i = 0; i <= BOND_MASK; i++) {
2347 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2348 b = &bals[e->iface_idx];
2349 b->tx_bytes += e->tx_bytes;
2351 b->hashes = &hashes[i];
2356 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2357 log_bals(bals, n_bals, port);
2359 /* Discard slaves that aren't enabled (which were sorted to the back of the
2360 * array earlier). */
2361 while (!bals[n_bals - 1].iface->enabled) {
2368 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2369 to = &bals[n_bals - 1];
2370 for (from = bals; from < to; ) {
2371 uint64_t overload = from->tx_bytes - to->tx_bytes;
2372 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2373 /* The extra load on 'from' (and all less-loaded slaves), compared
2374 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2375 * it is less than ~1Mbps. No point in rebalancing. */
2377 } else if (from->n_hashes == 1) {
2378 /* 'from' only carries a single MAC hash, so we can't shift any
2379 * load away from it, even though we want to. */
2382 /* 'from' is carrying significantly more load than 'to', and that
2383 * load is split across at least two different hashes. Pick a hash
2384 * to migrate to 'to' (the least-loaded slave), given that doing so
2385 * must decrease the ratio of the load on the two slaves by at
2388 * The sort order we use means that we prefer to shift away the
2389 * smallest hashes instead of the biggest ones. There is little
2390 * reason behind this decision; we could use the opposite sort
2391 * order to shift away big hashes ahead of small ones. */
2395 for (i = 0; i < from->n_hashes; i++) {
2396 double old_ratio, new_ratio;
2397 uint64_t delta = from->hashes[i]->tx_bytes;
2399 if (delta == 0 || from->tx_bytes - delta == 0) {
2400 /* Pointless move. */
2404 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2406 if (to->tx_bytes == 0) {
2407 /* Nothing on the new slave, move it. */
2411 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2412 new_ratio = (double)(from->tx_bytes - delta) /
2413 (to->tx_bytes + delta);
2415 if (new_ratio == 0) {
2416 /* Should already be covered but check to prevent division
2421 if (new_ratio < 1) {
2422 new_ratio = 1 / new_ratio;
2425 if (old_ratio - new_ratio > 0.1) {
2426 /* Would decrease the ratio, move it. */
2430 if (i < from->n_hashes) {
2431 bond_shift_load(from, to, i);
2432 port->bond_compat_is_stale = true;
2434 /* If the result of the migration changed the relative order of
2435 * 'from' and 'to' swap them back to maintain invariants. */
2436 if (order_swapped) {
2437 swap_bals(from, to);
2440 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2441 * point to different slave_balance structures. It is only
2442 * valid to do these two operations in a row at all because we
2443 * know that 'from' will not move past 'to' and vice versa. */
2444 resort_bals(from, bals, n_bals);
2445 resort_bals(to, bals, n_bals);
2452 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2453 * historical data to decay to <1% in 7 rebalancing runs. */
2454 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2460 bond_send_learning_packets(struct port *port)
2462 struct bridge *br = port->bridge;
2463 struct mac_entry *e;
2464 struct ofpbuf packet;
2465 int error, n_packets, n_errors;
2467 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2471 ofpbuf_init(&packet, 128);
2472 error = n_packets = n_errors = 0;
2473 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2474 union ofp_action actions[2], *a;
2480 if (e->port == port->port_idx
2481 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2485 /* Compose actions. */
2486 memset(actions, 0, sizeof actions);
2489 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2490 a->vlan_vid.len = htons(sizeof *a);
2491 a->vlan_vid.vlan_vid = htons(e->vlan);
2494 a->output.type = htons(OFPAT_OUTPUT);
2495 a->output.len = htons(sizeof *a);
2496 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2501 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2503 flow_extract(&packet, ODPP_NONE, &flow);
2504 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2511 ofpbuf_uninit(&packet);
2514 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2515 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2516 "packets, last error was: %s",
2517 port->name, n_errors, n_packets, strerror(error));
2519 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2520 port->name, n_packets);
2524 /* Bonding unixctl user interface functions. */
2527 bond_unixctl_list(struct unixctl_conn *conn,
2528 const char *args UNUSED, void *aux UNUSED)
2530 struct ds ds = DS_EMPTY_INITIALIZER;
2531 const struct bridge *br;
2533 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2535 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2538 for (i = 0; i < br->n_ports; i++) {
2539 const struct port *port = br->ports[i];
2540 if (port->n_ifaces > 1) {
2543 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2544 for (j = 0; j < port->n_ifaces; j++) {
2545 const struct iface *iface = port->ifaces[j];
2547 ds_put_cstr(&ds, ", ");
2549 ds_put_cstr(&ds, iface->name);
2551 ds_put_char(&ds, '\n');
2555 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2559 static struct port *
2560 bond_find(const char *name)
2562 const struct bridge *br;
2564 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2567 for (i = 0; i < br->n_ports; i++) {
2568 struct port *port = br->ports[i];
2569 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2578 bond_unixctl_show(struct unixctl_conn *conn,
2579 const char *args, void *aux UNUSED)
2581 struct ds ds = DS_EMPTY_INITIALIZER;
2582 const struct port *port;
2585 port = bond_find(args);
2587 unixctl_command_reply(conn, 501, "no such bond");
2591 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2592 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2593 ds_put_format(&ds, "next rebalance: %lld ms\n",
2594 port->bridge->bond_next_rebalance - time_msec());
2595 for (j = 0; j < port->n_ifaces; j++) {
2596 const struct iface *iface = port->ifaces[j];
2597 struct bond_entry *be;
2600 ds_put_format(&ds, "slave %s: %s\n",
2601 iface->name, iface->enabled ? "enabled" : "disabled");
2602 if (j == port->active_iface) {
2603 ds_put_cstr(&ds, "\tactive slave\n");
2605 if (iface->delay_expires != LLONG_MAX) {
2606 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2607 iface->enabled ? "downdelay" : "updelay",
2608 iface->delay_expires - time_msec());
2612 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2613 int hash = be - port->bond_hash;
2614 struct mac_entry *me;
2616 if (be->iface_idx != j) {
2620 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2621 hash, be->tx_bytes / 1024);
2624 if (!port->bridge->ml) {
2628 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2629 &port->bridge->ml->lrus) {
2632 if (bond_hash(me->mac) == hash
2633 && me->port != port->port_idx
2634 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2635 && dp_ifidx == iface->dp_ifidx)
2637 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2638 ETH_ADDR_ARGS(me->mac));
2643 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2648 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2651 char *args = (char *) args_;
2652 char *save_ptr = NULL;
2653 char *bond_s, *hash_s, *slave_s;
2654 uint8_t mac[ETH_ADDR_LEN];
2656 struct iface *iface;
2657 struct bond_entry *entry;
2660 bond_s = strtok_r(args, " ", &save_ptr);
2661 hash_s = strtok_r(NULL, " ", &save_ptr);
2662 slave_s = strtok_r(NULL, " ", &save_ptr);
2664 unixctl_command_reply(conn, 501,
2665 "usage: bond/migrate BOND HASH SLAVE");
2669 port = bond_find(bond_s);
2671 unixctl_command_reply(conn, 501, "no such bond");
2675 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2676 == ETH_ADDR_SCAN_COUNT) {
2677 hash = bond_hash(mac);
2678 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2679 hash = atoi(hash_s) & BOND_MASK;
2681 unixctl_command_reply(conn, 501, "bad hash");
2685 iface = port_lookup_iface(port, slave_s);
2687 unixctl_command_reply(conn, 501, "no such slave");
2691 if (!iface->enabled) {
2692 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2696 entry = &port->bond_hash[hash];
2697 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2698 entry->iface_idx = iface->port_ifidx;
2699 entry->iface_tag = tag_create_random();
2700 port->bond_compat_is_stale = true;
2701 unixctl_command_reply(conn, 200, "migrated");
2705 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2708 char *args = (char *) args_;
2709 char *save_ptr = NULL;
2710 char *bond_s, *slave_s;
2712 struct iface *iface;
2714 bond_s = strtok_r(args, " ", &save_ptr);
2715 slave_s = strtok_r(NULL, " ", &save_ptr);
2717 unixctl_command_reply(conn, 501,
2718 "usage: bond/set-active-slave BOND SLAVE");
2722 port = bond_find(bond_s);
2724 unixctl_command_reply(conn, 501, "no such bond");
2728 iface = port_lookup_iface(port, slave_s);
2730 unixctl_command_reply(conn, 501, "no such slave");
2734 if (!iface->enabled) {
2735 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2739 if (port->active_iface != iface->port_ifidx) {
2740 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2741 port->active_iface = iface->port_ifidx;
2742 port->active_iface_tag = tag_create_random();
2743 VLOG_INFO("port %s: active interface is now %s",
2744 port->name, iface->name);
2745 bond_send_learning_packets(port);
2746 unixctl_command_reply(conn, 200, "done");
2748 unixctl_command_reply(conn, 200, "no change");
2753 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2755 char *args = (char *) args_;
2756 char *save_ptr = NULL;
2757 char *bond_s, *slave_s;
2759 struct iface *iface;
2761 bond_s = strtok_r(args, " ", &save_ptr);
2762 slave_s = strtok_r(NULL, " ", &save_ptr);
2764 unixctl_command_reply(conn, 501,
2765 "usage: bond/enable/disable-slave BOND SLAVE");
2769 port = bond_find(bond_s);
2771 unixctl_command_reply(conn, 501, "no such bond");
2775 iface = port_lookup_iface(port, slave_s);
2777 unixctl_command_reply(conn, 501, "no such slave");
2781 bond_enable_slave(iface, enable);
2782 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2786 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
2789 enable_slave(conn, args, true);
2793 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
2796 enable_slave(conn, args, false);
2800 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
2803 uint8_t mac[ETH_ADDR_LEN];
2807 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2808 == ETH_ADDR_SCAN_COUNT) {
2809 hash = bond_hash(mac);
2811 hash_cstr = xasprintf("%u", hash);
2812 unixctl_command_reply(conn, 200, hash_cstr);
2815 unixctl_command_reply(conn, 501, "invalid mac");
2822 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
2823 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
2824 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
2825 unixctl_command_register("bond/set-active-slave",
2826 bond_unixctl_set_active_slave, NULL);
2827 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
2829 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
2831 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
2834 /* Port functions. */
2837 port_create(struct bridge *br, const char *name)
2841 port = xzalloc(sizeof *port);
2843 port->port_idx = br->n_ports;
2845 port->trunks = NULL;
2846 port->name = xstrdup(name);
2847 port->active_iface = -1;
2848 port->stp_state = STP_DISABLED;
2849 port->stp_state_tag = 0;
2851 if (br->n_ports >= br->allocated_ports) {
2852 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2855 br->ports[br->n_ports++] = port;
2857 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2862 port_reconfigure(struct port *port)
2864 bool bonded = cfg_has_section("bonding.%s", port->name);
2865 struct svec old_ifaces, new_ifaces;
2866 unsigned long *trunks;
2870 /* Collect old and new interfaces. */
2871 svec_init(&old_ifaces);
2872 svec_init(&new_ifaces);
2873 for (i = 0; i < port->n_ifaces; i++) {
2874 svec_add(&old_ifaces, port->ifaces[i]->name);
2876 svec_sort(&old_ifaces);
2878 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2879 if (!new_ifaces.n) {
2880 VLOG_ERR("port %s: no interfaces specified for bonded port",
2882 } else if (new_ifaces.n == 1) {
2883 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2887 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2888 if (port->updelay < 0) {
2891 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2892 if (port->downdelay < 0) {
2893 port->downdelay = 0;
2896 svec_init(&new_ifaces);
2897 svec_add(&new_ifaces, port->name);
2900 /* Get rid of deleted interfaces and add new interfaces. */
2901 for (i = 0; i < port->n_ifaces; i++) {
2902 struct iface *iface = port->ifaces[i];
2903 if (!svec_contains(&new_ifaces, iface->name)) {
2904 iface_destroy(iface);
2909 for (i = 0; i < new_ifaces.n; i++) {
2910 const char *name = new_ifaces.names[i];
2911 if (!svec_contains(&old_ifaces, name)) {
2912 iface_create(port, name);
2918 if (cfg_has("vlan.%s.tag", port->name)) {
2920 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2921 if (vlan >= 0 && vlan <= 4095) {
2922 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2925 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2926 * they even work as-is. But they have not been tested. */
2927 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2931 if (port->vlan != vlan) {
2933 bridge_flush(port->bridge);
2936 /* Get trunked VLANs. */
2939 size_t n_trunks, n_errors;
2942 trunks = bitmap_allocate(4096);
2943 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2945 for (i = 0; i < n_trunks; i++) {
2946 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2948 bitmap_set1(trunks, trunk);
2954 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2955 port->name, n_trunks);
2957 if (n_errors == n_trunks) {
2959 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2962 bitmap_set_multiple(trunks, 0, 4096, 1);
2965 if (cfg_has("vlan.%s.trunks", port->name)) {
2966 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2967 port->name, port->name);
2971 ? port->trunks != NULL
2972 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2973 bridge_flush(port->bridge);
2975 bitmap_free(port->trunks);
2976 port->trunks = trunks;
2978 svec_destroy(&old_ifaces);
2979 svec_destroy(&new_ifaces);
2983 port_destroy(struct port *port)
2986 struct bridge *br = port->bridge;
2990 proc_net_compat_update_vlan(port->name, NULL, 0);
2991 proc_net_compat_update_bond(port->name, NULL);
2993 for (i = 0; i < MAX_MIRRORS; i++) {
2994 struct mirror *m = br->mirrors[i];
2995 if (m && m->out_port == port) {
3000 while (port->n_ifaces > 0) {
3001 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3004 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3005 del->port_idx = port->port_idx;
3008 bitmap_free(port->trunks);
3015 static struct port *
3016 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3018 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3019 return iface ? iface->port : NULL;
3022 static struct port *
3023 port_lookup(const struct bridge *br, const char *name)
3027 for (i = 0; i < br->n_ports; i++) {
3028 struct port *port = br->ports[i];
3029 if (!strcmp(port->name, name)) {
3036 static struct iface *
3037 port_lookup_iface(const struct port *port, const char *name)
3041 for (j = 0; j < port->n_ifaces; j++) {
3042 struct iface *iface = port->ifaces[j];
3043 if (!strcmp(iface->name, name)) {
3051 port_update_bonding(struct port *port)
3053 if (port->n_ifaces < 2) {
3054 /* Not a bonded port. */
3055 if (port->bond_hash) {
3056 free(port->bond_hash);
3057 port->bond_hash = NULL;
3058 port->bond_compat_is_stale = true;
3061 if (!port->bond_hash) {
3064 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3065 for (i = 0; i <= BOND_MASK; i++) {
3066 struct bond_entry *e = &port->bond_hash[i];
3070 port->no_ifaces_tag = tag_create_random();
3071 bond_choose_active_iface(port);
3073 port->bond_compat_is_stale = true;
3078 port_update_bond_compat(struct port *port)
3080 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3081 struct compat_bond bond;
3084 if (port->n_ifaces < 2) {
3085 proc_net_compat_update_bond(port->name, NULL);
3090 bond.updelay = port->updelay;
3091 bond.downdelay = port->downdelay;
3094 bond.hashes = compat_hashes;
3095 if (port->bond_hash) {
3096 const struct bond_entry *e;
3097 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3098 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3099 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3100 cbh->hash = e - port->bond_hash;
3101 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3106 bond.n_slaves = port->n_ifaces;
3107 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3108 for (i = 0; i < port->n_ifaces; i++) {
3109 struct iface *iface = port->ifaces[i];
3110 struct compat_bond_slave *slave = &bond.slaves[i];
3111 slave->name = iface->name;
3113 /* We need to make the same determination as the Linux bonding
3114 * code to determine whether a slave should be consider "up".
3115 * The Linux function bond_miimon_inspect() supports four
3116 * BOND_LINK_* states:
3118 * - BOND_LINK_UP: carrier detected, updelay has passed.
3119 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3120 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3121 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3123 * The function bond_info_show_slave() only considers BOND_LINK_UP
3124 * to be "up" and anything else to be "down".
3126 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3130 netdev_get_etheraddr(iface->netdev, slave->mac);
3133 if (cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
3134 struct netdev *bond_netdev;
3136 if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) {
3138 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3140 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3142 netdev_close(bond_netdev);
3146 proc_net_compat_update_bond(port->name, &bond);
3151 port_update_vlan_compat(struct port *port)
3153 struct bridge *br = port->bridge;
3154 char *vlandev_name = NULL;
3156 if (port->vlan > 0) {
3157 /* Figure out the name that the VLAN device should actually have, if it
3158 * existed. This takes some work because the VLAN device would not
3159 * have port->name in its name; rather, it would have the trunk port's
3160 * name, and 'port' would be attached to a bridge that also had the
3161 * VLAN device one of its ports. So we need to find a trunk port that
3162 * includes port->vlan.
3164 * There might be more than one candidate. This doesn't happen on
3165 * XenServer, so if it happens we just pick the first choice in
3166 * alphabetical order instead of creating multiple VLAN devices. */
3168 for (i = 0; i < br->n_ports; i++) {
3169 struct port *p = br->ports[i];
3170 if (port_trunks_vlan(p, port->vlan)
3172 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3174 uint8_t ea[ETH_ADDR_LEN];
3175 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3176 if (!eth_addr_is_multicast(ea) &&
3177 !eth_addr_is_reserved(ea) &&
3178 !eth_addr_is_zero(ea)) {
3179 vlandev_name = p->name;
3184 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3187 /* Interface functions. */
3190 iface_create(struct port *port, const char *name)
3192 struct iface *iface;
3194 iface = xzalloc(sizeof *iface);
3196 iface->port_ifidx = port->n_ifaces;
3197 iface->name = xstrdup(name);
3198 iface->dp_ifidx = -1;
3199 iface->tag = tag_create_random();
3200 iface->delay_expires = LLONG_MAX;
3201 iface->netdev = NULL;
3203 if (port->n_ifaces >= port->allocated_ifaces) {
3204 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3205 sizeof *port->ifaces);
3207 port->ifaces[port->n_ifaces++] = iface;
3208 if (port->n_ifaces > 1) {
3209 port->bridge->has_bonded_ports = true;
3212 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3214 bridge_flush(port->bridge);
3218 iface_destroy(struct iface *iface)
3221 struct port *port = iface->port;
3222 struct bridge *br = port->bridge;
3223 bool del_active = port->active_iface == iface->port_ifidx;
3226 if (iface->dp_ifidx >= 0) {
3227 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3230 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3231 del->port_ifidx = iface->port_ifidx;
3233 netdev_close(iface->netdev);
3238 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3239 bond_choose_active_iface(port);
3240 bond_send_learning_packets(port);
3243 bridge_flush(port->bridge);
3247 static struct iface *
3248 iface_lookup(const struct bridge *br, const char *name)
3252 for (i = 0; i < br->n_ports; i++) {
3253 struct port *port = br->ports[i];
3254 for (j = 0; j < port->n_ifaces; j++) {
3255 struct iface *iface = port->ifaces[j];
3256 if (!strcmp(iface->name, name)) {
3264 static struct iface *
3265 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3267 return port_array_get(&br->ifaces, dp_ifidx);
3270 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3271 * 'br', that is, an interface that is entirely simulated within the datapath.
3272 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3273 * interfaces are created by setting "iface.<iface>.internal = true".
3275 * In addition, we have a kluge-y feature that creates an internal port with
3276 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3277 * This feature needs to go away in the long term. Until then, this is one
3278 * reason why this function takes a name instead of a struct iface: the fake
3279 * interfaces created this way do not have a struct iface. */
3281 iface_is_internal(const struct bridge *br, const char *iface)
3283 if (!strcmp(iface, br->name)
3284 || cfg_get_bool(0, "iface.%s.internal", iface)) {
3288 if (cfg_get_bool(0, "bonding.%s.fake-iface", iface)) {
3289 struct port *port = port_lookup(br, iface);
3290 if (port && port->n_ifaces > 1) {
3298 /* Set Ethernet address of 'iface', if one is specified in the configuration
3301 iface_set_mac(struct iface *iface)
3303 uint64_t mac = cfg_get_mac(0, "iface.%s.mac", iface->name);
3305 static uint8_t ea[ETH_ADDR_LEN];
3307 eth_addr_from_uint64(mac, ea);
3308 if (eth_addr_is_multicast(ea)) {
3309 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3311 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3312 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3313 iface->name, iface->name);
3315 int error = netdev_set_etheraddr(iface->netdev, ea);
3317 VLOG_ERR("interface %s: setting MAC failed (%s)",
3318 iface->name, strerror(error));
3324 /* Port mirroring. */
3327 mirror_reconfigure(struct bridge *br)
3329 struct svec old_mirrors, new_mirrors;
3332 /* Collect old and new mirrors. */
3333 svec_init(&old_mirrors);
3334 svec_init(&new_mirrors);
3335 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3336 for (i = 0; i < MAX_MIRRORS; i++) {
3337 if (br->mirrors[i]) {
3338 svec_add(&old_mirrors, br->mirrors[i]->name);
3342 /* Get rid of deleted mirrors and add new mirrors. */
3343 svec_sort(&old_mirrors);
3344 assert(svec_is_unique(&old_mirrors));
3345 svec_sort(&new_mirrors);
3346 assert(svec_is_unique(&new_mirrors));
3347 for (i = 0; i < MAX_MIRRORS; i++) {
3348 struct mirror *m = br->mirrors[i];
3349 if (m && !svec_contains(&new_mirrors, m->name)) {
3353 for (i = 0; i < new_mirrors.n; i++) {
3354 const char *name = new_mirrors.names[i];
3355 if (!svec_contains(&old_mirrors, name)) {
3356 mirror_create(br, name);
3359 svec_destroy(&old_mirrors);
3360 svec_destroy(&new_mirrors);
3362 /* Reconfigure all mirrors. */
3363 for (i = 0; i < MAX_MIRRORS; i++) {
3364 if (br->mirrors[i]) {
3365 mirror_reconfigure_one(br->mirrors[i]);
3369 /* Update port reserved status. */
3370 for (i = 0; i < br->n_ports; i++) {
3371 br->ports[i]->is_mirror_output_port = false;
3373 for (i = 0; i < MAX_MIRRORS; i++) {
3374 struct mirror *m = br->mirrors[i];
3375 if (m && m->out_port) {
3376 m->out_port->is_mirror_output_port = true;
3382 mirror_create(struct bridge *br, const char *name)
3387 for (i = 0; ; i++) {
3388 if (i >= MAX_MIRRORS) {
3389 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3390 "cannot create %s", br->name, MAX_MIRRORS, name);
3393 if (!br->mirrors[i]) {
3398 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3401 br->mirrors[i] = m = xzalloc(sizeof *m);
3404 m->name = xstrdup(name);
3405 svec_init(&m->src_ports);
3406 svec_init(&m->dst_ports);
3414 mirror_destroy(struct mirror *m)
3417 struct bridge *br = m->bridge;
3420 for (i = 0; i < br->n_ports; i++) {
3421 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3422 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3425 svec_destroy(&m->src_ports);
3426 svec_destroy(&m->dst_ports);
3429 m->bridge->mirrors[m->idx] = NULL;
3437 prune_ports(struct mirror *m, struct svec *ports)
3442 svec_sort_unique(ports);
3445 for (i = 0; i < ports->n; i++) {
3446 const char *name = ports->names[i];
3447 if (port_lookup(m->bridge, name)) {
3448 svec_add(&tmp, name);
3450 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3451 m->bridge->name, m->name, name);
3454 svec_swap(ports, &tmp);
3459 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3463 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3464 * order won't give us numeric sort order. But that's good enough for what
3465 * we need right now. */
3466 svec_sort_unique(vlan_strings);
3468 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3470 for (i = 0; i < vlan_strings->n; i++) {
3471 const char *name = vlan_strings->names[i];
3473 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3474 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3475 m->bridge->name, m->name, name);
3477 (*vlans)[n_vlans++] = vlan;
3484 vlan_is_mirrored(const struct mirror *m, int vlan)
3488 for (i = 0; i < m->n_vlans; i++) {
3489 if (m->vlans[i] == vlan) {
3497 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3501 for (i = 0; i < m->n_vlans; i++) {
3502 if (port_trunks_vlan(p, m->vlans[i])) {
3510 mirror_reconfigure_one(struct mirror *m)
3512 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3513 struct svec src_ports, dst_ports, ports;
3514 struct svec vlan_strings;
3515 mirror_mask_t mirror_bit;
3516 const char *out_port_name;
3517 struct port *out_port;
3522 bool mirror_all_ports;
3523 bool any_ports_specified;
3525 /* Get output port. */
3526 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3527 m->bridge->name, m->name);
3528 if (out_port_name) {
3529 out_port = port_lookup(m->bridge, out_port_name);
3531 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3532 "named %s", pfx, m->bridge->name, out_port_name);
3539 if (cfg_has("%s.output.vlan", pfx)) {
3540 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3541 "ignoring %s.output.vlan", pfx, pfx, pfx);
3543 } else if (cfg_has("%s.output.vlan", pfx)) {
3545 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3547 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3548 "but exactly one is required; disabling port mirror %s",
3549 pfx, pfx, pfx, pfx);
3555 /* Get all the ports, and drop duplicates and ports that don't exist. */
3556 svec_init(&src_ports);
3557 svec_init(&dst_ports);
3559 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3560 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3561 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3562 any_ports_specified = src_ports.n || dst_ports.n || ports.n;
3563 svec_append(&src_ports, &ports);
3564 svec_append(&dst_ports, &ports);
3565 svec_destroy(&ports);
3566 prune_ports(m, &src_ports);
3567 prune_ports(m, &dst_ports);
3568 if (any_ports_specified && !src_ports.n && !dst_ports.n) {
3569 VLOG_ERR("%s: none of the specified ports exist; "
3570 "disabling port mirror %s", pfx, pfx);
3575 /* Get all the vlans, and drop duplicate and invalid vlans. */
3576 svec_init(&vlan_strings);
3577 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3578 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3579 svec_destroy(&vlan_strings);
3581 /* Update mirror data. */
3582 if (!svec_equal(&m->src_ports, &src_ports)
3583 || !svec_equal(&m->dst_ports, &dst_ports)
3584 || m->n_vlans != n_vlans
3585 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3586 || m->out_port != out_port
3587 || m->out_vlan != out_vlan) {
3588 bridge_flush(m->bridge);
3590 svec_swap(&m->src_ports, &src_ports);
3591 svec_swap(&m->dst_ports, &dst_ports);
3594 m->n_vlans = n_vlans;
3595 m->out_port = out_port;
3596 m->out_vlan = out_vlan;
3598 /* If no selection criteria have been given, mirror for all ports. */
3599 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3602 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3603 for (i = 0; i < m->bridge->n_ports; i++) {
3604 struct port *port = m->bridge->ports[i];
3606 if (mirror_all_ports
3607 || svec_contains(&m->src_ports, port->name)
3610 ? port_trunks_any_mirrored_vlan(m, port)
3611 : vlan_is_mirrored(m, port->vlan)))) {
3612 port->src_mirrors |= mirror_bit;
3614 port->src_mirrors &= ~mirror_bit;
3617 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3618 port->dst_mirrors |= mirror_bit;
3620 port->dst_mirrors &= ~mirror_bit;
3626 svec_destroy(&src_ports);
3627 svec_destroy(&dst_ports);
3631 /* Spanning tree protocol. */
3633 static void brstp_update_port_state(struct port *);
3636 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3638 struct bridge *br = br_;
3639 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3640 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3642 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3645 struct eth_header *eth = pkt->l2;
3647 netdev_get_etheraddr(iface->netdev, eth->eth_src);
3648 if (eth_addr_is_zero(eth->eth_src)) {
3649 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
3650 "with unknown MAC", br->name, port_no);
3652 union ofp_action action;
3655 memset(&action, 0, sizeof action);
3656 action.type = htons(OFPAT_OUTPUT);
3657 action.output.len = htons(sizeof action);
3658 action.output.port = htons(port_no);
3660 flow_extract(pkt, ODPP_NONE, &flow);
3661 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3668 brstp_reconfigure(struct bridge *br)
3672 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3674 stp_destroy(br->stp);
3680 uint64_t bridge_address, bridge_id;
3681 int bridge_priority;
3683 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3684 if (!bridge_address) {
3686 bridge_address = (stp_get_bridge_id(br->stp)
3687 & ((UINT64_C(1) << 48) - 1));
3689 uint8_t mac[ETH_ADDR_LEN];
3690 eth_addr_random(mac);
3691 bridge_address = eth_addr_to_uint64(mac);
3695 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3697 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3699 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3702 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3704 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3705 br->stp_last_tick = time_msec();
3708 if (bridge_id != stp_get_bridge_id(br->stp)) {
3709 stp_set_bridge_id(br->stp, bridge_id);
3714 for (i = 0; i < br->n_ports; i++) {
3715 struct port *p = br->ports[i];
3717 struct stp_port *sp;
3718 int path_cost, priority;
3724 dp_ifidx = p->ifaces[0]->dp_ifidx;
3725 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3729 sp = stp_get_port(br->stp, dp_ifidx);
3730 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3731 "stp.%s.port.%s.enabled",
3733 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3734 br->name, p->name));
3735 if (p->is_mirror_output_port) {
3738 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3739 bridge_flush(br); /* Might not be necessary. */
3741 stp_port_enable(sp);
3743 stp_port_disable(sp);
3747 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3749 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3751 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3752 "stp.%s.port.%s.priority",
3754 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3756 : STP_DEFAULT_PORT_PRIORITY);
3757 stp_port_set_priority(sp, priority);
3760 brstp_adjust_timers(br);
3762 for (i = 0; i < br->n_ports; i++) {
3763 brstp_update_port_state(br->ports[i]);
3768 brstp_update_port_state(struct port *p)
3770 struct bridge *br = p->bridge;
3771 enum stp_state state;
3773 /* Figure out new state. */
3774 state = STP_DISABLED;
3775 if (br->stp && p->n_ifaces > 0) {
3776 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3777 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3778 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3783 if (p->stp_state != state) {
3784 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3785 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3786 p->name, stp_state_name(p->stp_state),
3787 stp_state_name(state));
3788 if (p->stp_state == STP_DISABLED) {
3791 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3793 p->stp_state = state;
3794 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3795 : tag_create_random());
3800 brstp_adjust_timers(struct bridge *br)
3802 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3803 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3804 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3806 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3807 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3808 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3812 brstp_run(struct bridge *br)
3815 long long int now = time_msec();
3816 long long int elapsed = now - br->stp_last_tick;
3817 struct stp_port *sp;
3820 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3821 br->stp_last_tick = now;
3823 while (stp_get_changed_port(br->stp, &sp)) {
3824 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3826 brstp_update_port_state(p);
3833 brstp_wait(struct bridge *br)
3836 poll_timer_wait(1000);