1 /* Copyright (c) 2008, 2009, 2010 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>
36 #include "dynamic-string.h"
40 #include "mac-learning.h"
43 #include "ofp-print.h"
45 #include "ofproto/netflow.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
54 #include "socket-util.h"
55 #include "stream-ssl.h"
61 #include "vswitchd/vswitch-idl.h"
62 #include "xenserver.h"
64 #include "sflow_api.h"
66 #define THIS_MODULE VLM_bridge
75 /* These members are always valid. */
76 struct port *port; /* Containing port. */
77 size_t port_ifidx; /* Index within containing port. */
78 char *name; /* Host network device name. */
79 tag_type tag; /* Tag associated with this interface. */
80 long long delay_expires; /* Time after which 'enabled' may change. */
82 /* These members are valid only after bridge_reconfigure() causes them to
84 int dp_ifidx; /* Index within kernel datapath. */
85 struct netdev *netdev; /* Network device. */
86 bool enabled; /* May be chosen for flows? */
88 /* This member is only valid *during* bridge_reconfigure(). */
89 const struct ovsrec_interface *cfg;
92 #define BOND_MASK 0xff
94 int iface_idx; /* Index of assigned iface, or -1 if none. */
95 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
96 tag_type iface_tag; /* Tag associated with iface_idx. */
99 #define MAX_MIRRORS 32
100 typedef uint32_t mirror_mask_t;
101 #define MIRROR_MASK_C(X) UINT32_C(X)
102 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
104 struct bridge *bridge;
108 /* Selection criteria. */
109 struct shash src_ports; /* Name is port name; data is always NULL. */
110 struct shash dst_ports; /* Name is port name; data is always NULL. */
115 struct port *out_port;
119 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
121 struct bridge *bridge;
123 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
124 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
127 /* An ordinary bridge port has 1 interface.
128 * A bridge port for bonding has at least 2 interfaces. */
129 struct iface **ifaces;
130 size_t n_ifaces, allocated_ifaces;
133 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
134 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
135 tag_type active_iface_tag; /* Tag for bcast flows. */
136 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
137 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
138 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
139 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
141 /* Port mirroring info. */
142 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
143 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
144 bool is_mirror_output_port; /* Does port mirroring send frames here? */
146 /* This member is only valid *during* bridge_reconfigure(). */
147 const struct ovsrec_port *cfg;
150 #define DP_MAX_PORTS 255
152 struct list node; /* Node in global list of bridges. */
153 char *name; /* User-specified arbitrary name. */
154 struct mac_learning *ml; /* MAC learning table. */
155 bool sent_config_request; /* Successfully sent config request? */
156 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
158 /* Support for remote controllers. */
159 char *controller; /* NULL if there is no remote controller;
160 * "discover" to do controller discovery;
161 * otherwise a vconn name. */
163 /* OpenFlow switch processing. */
164 struct ofproto *ofproto; /* OpenFlow switch. */
166 /* Description strings. */
167 char *mfr_desc; /* Manufacturer. */
168 char *hw_desc; /* Hardware. */
169 char *sw_desc; /* Software version. */
170 char *serial_desc; /* Serial number. */
171 char *dp_desc; /* Datapath description. */
173 /* Kernel datapath information. */
174 struct dpif *dpif; /* Datapath. */
175 struct port_array ifaces; /* Indexed by kernel datapath port number. */
179 size_t n_ports, allocated_ports;
182 bool has_bonded_ports;
183 long long int bond_next_rebalance;
188 /* Flow statistics gathering. */
189 time_t next_stats_request;
191 /* Port mirroring. */
192 struct mirror *mirrors[MAX_MIRRORS];
194 /* This member is only valid *during* bridge_reconfigure(). */
195 const struct ovsrec_bridge *cfg;
198 /* List of all bridges. */
199 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
201 /* Maximum number of datapaths. */
202 enum { DP_MAX = 256 };
204 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
205 static void bridge_destroy(struct bridge *);
206 static struct bridge *bridge_lookup(const char *name);
207 static unixctl_cb_func bridge_unixctl_dump_flows;
208 static int bridge_run_one(struct bridge *);
209 static const struct ovsrec_controller *bridge_get_controller(
210 const struct ovsrec_open_vswitch *ovs_cfg,
211 const struct bridge *br);
212 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
214 static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
216 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
217 static void bridge_fetch_dp_ifaces(struct bridge *);
218 static void bridge_flush(struct bridge *);
219 static void bridge_pick_local_hw_addr(struct bridge *,
220 uint8_t ea[ETH_ADDR_LEN],
221 struct iface **hw_addr_iface);
222 static uint64_t bridge_pick_datapath_id(struct bridge *,
223 const uint8_t bridge_ea[ETH_ADDR_LEN],
224 struct iface *hw_addr_iface);
225 static struct iface *bridge_get_local_iface(struct bridge *);
226 static uint64_t dpid_from_hash(const void *, size_t nbytes);
228 static unixctl_cb_func bridge_unixctl_fdb_show;
230 static void bond_init(void);
231 static void bond_run(struct bridge *);
232 static void bond_wait(struct bridge *);
233 static void bond_rebalance_port(struct port *);
234 static void bond_send_learning_packets(struct port *);
235 static void bond_enable_slave(struct iface *iface, bool enable);
237 static struct port *port_create(struct bridge *, const char *name);
238 static void port_reconfigure(struct port *, const struct ovsrec_port *);
239 static void port_destroy(struct port *);
240 static struct port *port_lookup(const struct bridge *, const char *name);
241 static struct iface *port_lookup_iface(const struct port *, const char *name);
242 static struct port *port_from_dp_ifidx(const struct bridge *,
244 static void port_update_bond_compat(struct port *);
245 static void port_update_vlan_compat(struct port *);
246 static void port_update_bonding(struct port *);
248 static struct mirror *mirror_create(struct bridge *, const char *name);
249 static void mirror_destroy(struct mirror *);
250 static void mirror_reconfigure(struct bridge *);
251 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
252 static bool vlan_is_mirrored(const struct mirror *, int vlan);
254 static struct iface *iface_create(struct port *port,
255 const struct ovsrec_interface *if_cfg);
256 static void iface_destroy(struct iface *);
257 static struct iface *iface_lookup(const struct bridge *, const char *name);
258 static struct iface *iface_from_dp_ifidx(const struct bridge *,
260 static bool iface_is_internal(const struct bridge *, const char *name);
261 static void iface_set_mac(struct iface *);
263 /* Hooks into ofproto processing. */
264 static struct ofhooks bridge_ofhooks;
266 /* Public functions. */
268 /* Adds the name of each interface used by a bridge, including local and
269 * internal ports, to 'svec'. */
271 bridge_get_ifaces(struct svec *svec)
273 struct bridge *br, *next;
276 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
277 for (i = 0; i < br->n_ports; i++) {
278 struct port *port = br->ports[i];
280 for (j = 0; j < port->n_ifaces; j++) {
281 struct iface *iface = port->ifaces[j];
282 if (iface->dp_ifidx < 0) {
283 VLOG_ERR("%s interface not in datapath %s, ignoring",
284 iface->name, dpif_name(br->dpif));
286 if (iface->dp_ifidx != ODPP_LOCAL) {
287 svec_add(svec, iface->name);
296 bridge_init(const struct ovsrec_open_vswitch *cfg)
298 struct svec bridge_names;
299 struct svec dpif_names, dpif_types;
302 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
304 svec_init(&bridge_names);
305 for (i = 0; i < cfg->n_bridges; i++) {
306 svec_add(&bridge_names, cfg->bridges[i]->name);
308 svec_sort(&bridge_names);
310 svec_init(&dpif_names);
311 svec_init(&dpif_types);
312 dp_enumerate_types(&dpif_types);
313 for (i = 0; i < dpif_types.n; i++) {
318 dp_enumerate_names(dpif_types.names[i], &dpif_names);
320 for (j = 0; j < dpif_names.n; j++) {
321 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
323 struct svec all_names;
326 svec_init(&all_names);
327 dpif_get_all_names(dpif, &all_names);
328 for (k = 0; k < all_names.n; k++) {
329 if (svec_contains(&bridge_names, all_names.names[k])) {
335 svec_destroy(&all_names);
340 svec_destroy(&dpif_names);
341 svec_destroy(&dpif_types);
343 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
347 bridge_reconfigure(cfg);
352 config_string_change(const char *value, char **valuep)
354 if (value && (!*valuep || strcmp(value, *valuep))) {
356 *valuep = xstrdup(value);
364 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
366 /* XXX SSL should be configurable on a per-bridge basis.
367 * XXX should be possible to de-configure SSL. */
368 static char *private_key_file;
369 static char *certificate_file;
370 static char *cacert_file;
374 /* XXX We can't un-set SSL settings. */
378 if (config_string_change(ssl->private_key, &private_key_file)) {
379 stream_ssl_set_private_key_file(private_key_file);
382 if (config_string_change(ssl->certificate, &certificate_file)) {
383 stream_ssl_set_certificate_file(certificate_file);
386 /* We assume that even if the filename hasn't changed, if the CA cert
387 * file has been removed, that we want to move back into
388 * boot-strapping mode. This opens a small security hole, because
389 * the old certificate will still be trusted until vSwitch is
390 * restarted. We may want to address this in vconn's SSL library. */
391 if (config_string_change(ssl->ca_cert, &cacert_file)
392 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
393 stream_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert);
398 /* Attempt to create the network device 'iface_name' through the netdev
401 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
404 struct shash_node *node;
405 struct shash options;
409 shash_init(&options);
410 for (i = 0; i < iface_cfg->n_options; i++) {
411 shash_add(&options, iface_cfg->key_options[i],
412 xstrdup(iface_cfg->value_options[i]));
416 struct netdev_options netdev_options;
418 memset(&netdev_options, 0, sizeof netdev_options);
419 netdev_options.name = iface_cfg->name;
420 netdev_options.type = iface_cfg->type;
421 netdev_options.args = &options;
422 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
423 netdev_options.may_create = true;
424 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
425 netdev_options.may_open = true;
428 error = netdev_open(&netdev_options, &iface->netdev);
431 netdev_get_carrier(iface->netdev, &iface->enabled);
433 } else if (iface->netdev) {
434 const char *netdev_type = netdev_get_type(iface->netdev);
435 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
436 ? iface_cfg->type : NULL;
438 if (!iface_type || !strcmp(netdev_type, iface_type)) {
439 error = netdev_reconfigure(iface->netdev, &options);
441 VLOG_WARN("%s: attempting change device type from %s to %s",
442 iface_cfg->name, netdev_type, iface_type);
447 SHASH_FOR_EACH (node, &options) {
450 shash_destroy(&options);
456 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
458 return set_up_iface(iface_cfg, iface, false);
462 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
463 void *aux OVS_UNUSED)
465 if (!iface->netdev) {
466 int error = set_up_iface(iface->cfg, iface, true);
468 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
478 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
479 void *aux OVS_UNUSED)
481 if (iface->dp_ifidx >= 0) {
482 VLOG_DBG("%s has interface %s on port %d",
484 iface->name, iface->dp_ifidx);
487 VLOG_ERR("%s interface not in %s, dropping",
488 iface->name, dpif_name(br->dpif));
494 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
495 void *aux OVS_UNUSED)
497 /* Set policing attributes. */
498 netdev_set_policing(iface->netdev,
499 iface->cfg->ingress_policing_rate,
500 iface->cfg->ingress_policing_burst);
502 /* Set MAC address of internal interfaces other than the local
504 if (iface->dp_ifidx != ODPP_LOCAL
505 && iface_is_internal(br, iface->name)) {
506 iface_set_mac(iface);
512 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
513 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
514 * deletes from 'br' any ports that no longer have any interfaces. */
516 iterate_and_prune_ifaces(struct bridge *br,
517 bool (*cb)(struct bridge *, struct iface *,
523 for (i = 0; i < br->n_ports; ) {
524 struct port *port = br->ports[i];
525 for (j = 0; j < port->n_ifaces; ) {
526 struct iface *iface = port->ifaces[j];
527 if (cb(br, iface, aux)) {
530 iface_destroy(iface);
534 if (port->n_ifaces) {
537 VLOG_ERR("%s port has no interfaces, dropping", port->name);
544 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
546 struct ovsdb_idl_txn *txn;
547 struct shash old_br, new_br;
548 struct shash_node *node;
549 struct bridge *br, *next;
551 int sflow_bridge_number;
553 COVERAGE_INC(bridge_reconfigure);
555 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
557 /* Collect old and new bridges. */
560 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
561 shash_add(&old_br, br->name, br);
563 for (i = 0; i < ovs_cfg->n_bridges; i++) {
564 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
565 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
566 VLOG_WARN("more than one bridge named %s", br_cfg->name);
570 /* Get rid of deleted bridges and add new bridges. */
571 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
572 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
579 SHASH_FOR_EACH (node, &new_br) {
580 const char *br_name = node->name;
581 const struct ovsrec_bridge *br_cfg = node->data;
582 br = shash_find_data(&old_br, br_name);
584 /* If the bridge datapath type has changed, we need to tear it
585 * down and recreate. */
586 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
588 bridge_create(br_cfg);
591 bridge_create(br_cfg);
594 shash_destroy(&old_br);
595 shash_destroy(&new_br);
599 bridge_configure_ssl(ovs_cfg->ssl);
602 /* Reconfigure all bridges. */
603 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
604 bridge_reconfigure_one(ovs_cfg, br);
607 /* Add and delete ports on all datapaths.
609 * The kernel will reject any attempt to add a given port to a datapath if
610 * that port already belongs to a different datapath, so we must do all
611 * port deletions before any port additions. */
612 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
613 struct odp_port *dpif_ports;
615 struct shash want_ifaces;
617 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
618 bridge_get_all_ifaces(br, &want_ifaces);
619 for (i = 0; i < n_dpif_ports; i++) {
620 const struct odp_port *p = &dpif_ports[i];
621 if (!shash_find(&want_ifaces, p->devname)
622 && strcmp(p->devname, br->name)) {
623 int retval = dpif_port_del(br->dpif, p->port);
625 VLOG_ERR("failed to remove %s interface from %s: %s",
626 p->devname, dpif_name(br->dpif),
631 shash_destroy(&want_ifaces);
634 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
635 struct odp_port *dpif_ports;
637 struct shash cur_ifaces, want_ifaces;
638 struct shash_node *node;
640 /* Get the set of interfaces currently in this datapath. */
641 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
642 shash_init(&cur_ifaces);
643 for (i = 0; i < n_dpif_ports; i++) {
644 const char *name = dpif_ports[i].devname;
645 if (!shash_find(&cur_ifaces, name)) {
646 shash_add(&cur_ifaces, name, NULL);
651 /* Get the set of interfaces we want on this datapath. */
652 bridge_get_all_ifaces(br, &want_ifaces);
654 SHASH_FOR_EACH (node, &want_ifaces) {
655 const char *if_name = node->name;
656 struct iface *iface = node->data;
658 if (shash_find(&cur_ifaces, if_name)) {
659 /* Already exists, just reconfigure it. */
661 reconfigure_iface(iface->cfg, iface);
664 /* Need to add to datapath. */
668 /* Add to datapath. */
669 internal = iface_is_internal(br, if_name);
670 error = dpif_port_add(br->dpif, if_name,
671 internal ? ODP_PORT_INTERNAL : 0, NULL);
672 if (error == EFBIG) {
673 VLOG_ERR("ran out of valid port numbers on %s",
674 dpif_name(br->dpif));
677 VLOG_ERR("failed to add %s interface to %s: %s",
678 if_name, dpif_name(br->dpif), strerror(error));
682 shash_destroy(&cur_ifaces);
683 shash_destroy(&want_ifaces);
685 sflow_bridge_number = 0;
686 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
689 struct iface *local_iface;
690 struct iface *hw_addr_iface;
693 bridge_fetch_dp_ifaces(br);
695 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
696 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
698 /* Pick local port hardware address, datapath ID. */
699 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
700 local_iface = bridge_get_local_iface(br);
702 int error = netdev_set_etheraddr(local_iface->netdev, ea);
704 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
705 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
706 "Ethernet address: %s",
707 br->name, strerror(error));
711 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
712 ofproto_set_datapath_id(br->ofproto, dpid);
714 dpid_string = xasprintf("%012"PRIx64, dpid);
715 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
718 /* Set NetFlow configuration on this bridge. */
719 if (br->cfg->netflow) {
720 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
721 struct netflow_options opts;
723 memset(&opts, 0, sizeof opts);
725 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
726 if (nf_cfg->engine_type) {
727 opts.engine_type = *nf_cfg->engine_type;
729 if (nf_cfg->engine_id) {
730 opts.engine_id = *nf_cfg->engine_id;
733 opts.active_timeout = nf_cfg->active_timeout;
734 if (!opts.active_timeout) {
735 opts.active_timeout = -1;
736 } else if (opts.active_timeout < 0) {
737 VLOG_WARN("bridge %s: active timeout interval set to negative "
738 "value, using default instead (%d seconds)", br->name,
739 NF_ACTIVE_TIMEOUT_DEFAULT);
740 opts.active_timeout = -1;
743 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
744 if (opts.add_id_to_iface) {
745 if (opts.engine_id > 0x7f) {
746 VLOG_WARN("bridge %s: netflow port mangling may conflict "
747 "with another vswitch, choose an engine id less "
748 "than 128", br->name);
750 if (br->n_ports > 508) {
751 VLOG_WARN("bridge %s: netflow port mangling will conflict "
752 "with another port when more than 508 ports are "
757 opts.collectors.n = nf_cfg->n_targets;
758 opts.collectors.names = nf_cfg->targets;
759 if (ofproto_set_netflow(br->ofproto, &opts)) {
760 VLOG_ERR("bridge %s: problem setting netflow collectors",
764 ofproto_set_netflow(br->ofproto, NULL);
767 /* Set sFlow configuration on this bridge. */
768 if (br->cfg->sflow) {
769 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
770 const struct ovsrec_controller *ctrl;
771 struct ofproto_sflow_options oso;
773 memset(&oso, 0, sizeof oso);
775 oso.targets.n = sflow_cfg->n_targets;
776 oso.targets.names = sflow_cfg->targets;
778 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
779 if (sflow_cfg->sampling) {
780 oso.sampling_rate = *sflow_cfg->sampling;
783 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
784 if (sflow_cfg->polling) {
785 oso.polling_interval = *sflow_cfg->polling;
788 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
789 if (sflow_cfg->header) {
790 oso.header_len = *sflow_cfg->header;
793 oso.sub_id = sflow_bridge_number++;
794 oso.agent_device = sflow_cfg->agent;
796 ctrl = bridge_get_controller(ovs_cfg, br);
797 oso.control_ip = ctrl ? ctrl->local_ip : NULL;
798 ofproto_set_sflow(br->ofproto, &oso);
800 svec_destroy(&oso.targets);
802 ofproto_set_sflow(br->ofproto, NULL);
805 /* Update the controller and related settings. It would be more
806 * straightforward to call this from bridge_reconfigure_one(), but we
807 * can't do it there for two reasons. First, and most importantly, at
808 * that point we don't know the dp_ifidx of any interfaces that have
809 * been added to the bridge (because we haven't actually added them to
810 * the datapath). Second, at that point we haven't set the datapath ID
811 * yet; when a controller is configured, resetting the datapath ID will
812 * immediately disconnect from the controller, so it's better to set
813 * the datapath ID before the controller. */
814 bridge_reconfigure_controller(ovs_cfg, br);
816 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
817 for (i = 0; i < br->n_ports; i++) {
818 struct port *port = br->ports[i];
820 port_update_vlan_compat(port);
821 port_update_bonding(port);
824 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
825 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
828 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
830 ovsdb_idl_txn_commit(txn);
831 ovsdb_idl_txn_destroy(txn); /* XXX */
835 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
839 for (i = 0; i < br_cfg->n_other_config; i++) {
840 if (!strcmp(br_cfg->key_other_config[i], key)) {
841 return br_cfg->value_other_config[i];
848 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
849 struct iface **hw_addr_iface)
855 *hw_addr_iface = NULL;
857 /* Did the user request a particular MAC? */
858 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
859 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
860 if (eth_addr_is_multicast(ea)) {
861 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
862 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
863 } else if (eth_addr_is_zero(ea)) {
864 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
870 /* Otherwise choose the minimum non-local MAC address among all of the
872 memset(ea, 0xff, sizeof ea);
873 for (i = 0; i < br->n_ports; i++) {
874 struct port *port = br->ports[i];
875 uint8_t iface_ea[ETH_ADDR_LEN];
878 /* Mirror output ports don't participate. */
879 if (port->is_mirror_output_port) {
883 /* Choose the MAC address to represent the port. */
884 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
885 /* Find the interface with this Ethernet address (if any) so that
886 * we can provide the correct devname to the caller. */
888 for (j = 0; j < port->n_ifaces; j++) {
889 struct iface *candidate = port->ifaces[j];
890 uint8_t candidate_ea[ETH_ADDR_LEN];
891 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
892 && eth_addr_equals(iface_ea, candidate_ea)) {
897 /* Choose the interface whose MAC address will represent the port.
898 * The Linux kernel bonding code always chooses the MAC address of
899 * the first slave added to a bond, and the Fedora networking
900 * scripts always add slaves to a bond in alphabetical order, so
901 * for compatibility we choose the interface with the name that is
902 * first in alphabetical order. */
903 iface = port->ifaces[0];
904 for (j = 1; j < port->n_ifaces; j++) {
905 struct iface *candidate = port->ifaces[j];
906 if (strcmp(candidate->name, iface->name) < 0) {
911 /* The local port doesn't count (since we're trying to choose its
912 * MAC address anyway). */
913 if (iface->dp_ifidx == ODPP_LOCAL) {
918 error = netdev_get_etheraddr(iface->netdev, iface_ea);
920 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
921 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
922 iface->name, strerror(error));
927 /* Compare against our current choice. */
928 if (!eth_addr_is_multicast(iface_ea) &&
929 !eth_addr_is_local(iface_ea) &&
930 !eth_addr_is_reserved(iface_ea) &&
931 !eth_addr_is_zero(iface_ea) &&
932 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
934 memcpy(ea, iface_ea, ETH_ADDR_LEN);
935 *hw_addr_iface = iface;
938 if (eth_addr_is_multicast(ea)) {
939 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
940 *hw_addr_iface = NULL;
941 VLOG_WARN("bridge %s: using default bridge Ethernet "
942 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
944 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
945 br->name, ETH_ADDR_ARGS(ea));
949 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
950 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
951 * an interface on 'br', then that interface must be passed in as
952 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
953 * 'hw_addr_iface' must be passed in as a null pointer. */
955 bridge_pick_datapath_id(struct bridge *br,
956 const uint8_t bridge_ea[ETH_ADDR_LEN],
957 struct iface *hw_addr_iface)
960 * The procedure for choosing a bridge MAC address will, in the most
961 * ordinary case, also choose a unique MAC that we can use as a datapath
962 * ID. In some special cases, though, multiple bridges will end up with
963 * the same MAC address. This is OK for the bridges, but it will confuse
964 * the OpenFlow controller, because each datapath needs a unique datapath
967 * Datapath IDs must be unique. It is also very desirable that they be
968 * stable from one run to the next, so that policy set on a datapath
971 const char *datapath_id;
974 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
975 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
981 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
983 * A bridge whose MAC address is taken from a VLAN network device
984 * (that is, a network device created with vconfig(8) or similar
985 * tool) will have the same MAC address as a bridge on the VLAN
986 * device's physical network device.
988 * Handle this case by hashing the physical network device MAC
989 * along with the VLAN identifier.
991 uint8_t buf[ETH_ADDR_LEN + 2];
992 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
993 buf[ETH_ADDR_LEN] = vlan >> 8;
994 buf[ETH_ADDR_LEN + 1] = vlan;
995 return dpid_from_hash(buf, sizeof buf);
998 * Assume that this bridge's MAC address is unique, since it
999 * doesn't fit any of the cases we handle specially.
1004 * A purely internal bridge, that is, one that has no non-virtual
1005 * network devices on it at all, is more difficult because it has no
1006 * natural unique identifier at all.
1008 * When the host is a XenServer, we handle this case by hashing the
1009 * host's UUID with the name of the bridge. Names of bridges are
1010 * persistent across XenServer reboots, although they can be reused if
1011 * an internal network is destroyed and then a new one is later
1012 * created, so this is fairly effective.
1014 * When the host is not a XenServer, we punt by using a random MAC
1015 * address on each run.
1017 const char *host_uuid = xenserver_get_host_uuid();
1019 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1020 dpid = dpid_from_hash(combined, strlen(combined));
1026 return eth_addr_to_uint64(bridge_ea);
1030 dpid_from_hash(const void *data, size_t n)
1032 uint8_t hash[SHA1_DIGEST_SIZE];
1034 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1035 sha1_bytes(data, n, hash);
1036 eth_addr_mark_random(hash);
1037 return eth_addr_to_uint64(hash);
1043 struct bridge *br, *next;
1047 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1048 int error = bridge_run_one(br);
1050 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1051 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1052 "forcing reconfiguration", br->name);
1066 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1067 ofproto_wait(br->ofproto);
1068 if (br->controller) {
1072 mac_learning_wait(br->ml);
1077 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1078 * configuration changes. */
1080 bridge_flush(struct bridge *br)
1082 COVERAGE_INC(bridge_flush);
1084 mac_learning_flush(br->ml);
1087 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1088 * such interface. */
1089 static struct iface *
1090 bridge_get_local_iface(struct bridge *br)
1094 for (i = 0; i < br->n_ports; i++) {
1095 struct port *port = br->ports[i];
1096 for (j = 0; j < port->n_ifaces; j++) {
1097 struct iface *iface = port->ifaces[j];
1098 if (iface->dp_ifidx == ODPP_LOCAL) {
1107 /* Bridge unixctl user interface functions. */
1109 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1110 const char *args, void *aux OVS_UNUSED)
1112 struct ds ds = DS_EMPTY_INITIALIZER;
1113 const struct bridge *br;
1114 const struct mac_entry *e;
1116 br = bridge_lookup(args);
1118 unixctl_command_reply(conn, 501, "no such bridge");
1122 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1123 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1124 if (e->port < 0 || e->port >= br->n_ports) {
1127 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1128 br->ports[e->port]->ifaces[0]->dp_ifidx,
1129 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1131 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1135 /* Bridge reconfiguration functions. */
1136 static struct bridge *
1137 bridge_create(const struct ovsrec_bridge *br_cfg)
1142 assert(!bridge_lookup(br_cfg->name));
1143 br = xzalloc(sizeof *br);
1145 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1151 dpif_flow_flush(br->dpif);
1153 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1156 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1158 dpif_delete(br->dpif);
1159 dpif_close(br->dpif);
1164 br->name = xstrdup(br_cfg->name);
1166 br->ml = mac_learning_create();
1167 br->sent_config_request = false;
1168 eth_addr_nicira_random(br->default_ea);
1170 port_array_init(&br->ifaces);
1173 br->bond_next_rebalance = time_msec() + 10000;
1175 list_push_back(&all_bridges, &br->node);
1177 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1183 bridge_destroy(struct bridge *br)
1188 while (br->n_ports > 0) {
1189 port_destroy(br->ports[br->n_ports - 1]);
1191 list_remove(&br->node);
1192 error = dpif_delete(br->dpif);
1193 if (error && error != ENOENT) {
1194 VLOG_ERR("failed to delete %s: %s",
1195 dpif_name(br->dpif), strerror(error));
1197 dpif_close(br->dpif);
1198 ofproto_destroy(br->ofproto);
1199 free(br->controller);
1200 mac_learning_destroy(br->ml);
1201 port_array_destroy(&br->ifaces);
1208 static struct bridge *
1209 bridge_lookup(const char *name)
1213 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1214 if (!strcmp(br->name, name)) {
1222 bridge_exists(const char *name)
1224 return bridge_lookup(name) ? true : false;
1228 bridge_get_datapathid(const char *name)
1230 struct bridge *br = bridge_lookup(name);
1231 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1234 /* Handle requests for a listing of all flows known by the OpenFlow
1235 * stack, including those normally hidden. */
1237 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1238 const char *args, void *aux OVS_UNUSED)
1243 br = bridge_lookup(args);
1245 unixctl_command_reply(conn, 501, "Unknown bridge");
1250 ofproto_get_all_flows(br->ofproto, &results);
1252 unixctl_command_reply(conn, 200, ds_cstr(&results));
1253 ds_destroy(&results);
1257 bridge_run_one(struct bridge *br)
1261 error = ofproto_run1(br->ofproto);
1266 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1269 error = ofproto_run2(br->ofproto, br->flush);
1275 static const struct ovsrec_controller *
1276 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1277 const struct bridge *br)
1279 const struct ovsrec_controller *controller;
1281 controller = (br->cfg->controller ? br->cfg->controller
1282 : ovs_cfg->controller ? ovs_cfg->controller
1285 if (controller && !strcmp(controller->target, "none")) {
1293 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1295 struct svec *ifaces = ifaces_;
1296 if (!svec_contains(ifaces, iface->name)) {
1297 svec_add(ifaces, iface->name);
1301 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1303 br->name, iface->name, iface->port->name);
1309 bridge_update_desc(struct bridge *br)
1312 bool changed = false;
1315 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1316 if (desc != br->mfr_desc) {
1319 br->mfr_desc = xstrdup(desc);
1321 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1326 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1327 if (desc != br->hw_desc) {
1330 br->hw_desc = xstrdup(desc);
1332 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1337 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1338 if (desc != br->sw_desc) {
1341 br->sw_desc = xstrdup(desc);
1343 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1348 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1349 if (desc != br->serial_desc) {
1350 free(br->serial_desc);
1352 br->serial_desc = xstrdup(desc);
1354 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1359 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1360 if (desc != br->dp_desc) {
1363 br->dp_desc = xstrdup(desc);
1365 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1371 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1372 br->sw_desc, br->serial_desc, br->dp_desc);
1378 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1381 struct shash old_ports, new_ports;
1383 struct svec listeners, old_listeners;
1384 struct svec snoops, old_snoops;
1385 struct shash_node *node;
1388 /* Collect old ports. */
1389 shash_init(&old_ports);
1390 for (i = 0; i < br->n_ports; i++) {
1391 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1394 /* Collect new ports. */
1395 shash_init(&new_ports);
1396 for (i = 0; i < br->cfg->n_ports; i++) {
1397 const char *name = br->cfg->ports[i]->name;
1398 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1399 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1404 /* If we have a controller, then we need a local port. Complain if the
1405 * user didn't specify one.
1407 * XXX perhaps we should synthesize a port ourselves in this case. */
1408 if (bridge_get_controller(ovs_cfg, br)) {
1409 char local_name[IF_NAMESIZE];
1412 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1413 local_name, sizeof local_name);
1414 if (!error && !shash_find(&new_ports, local_name)) {
1415 VLOG_WARN("bridge %s: controller specified but no local port "
1416 "(port named %s) defined",
1417 br->name, local_name);
1421 /* Get rid of deleted ports and add new ports. */
1422 SHASH_FOR_EACH (node, &old_ports) {
1423 if (!shash_find(&new_ports, node->name)) {
1424 port_destroy(node->data);
1427 SHASH_FOR_EACH (node, &new_ports) {
1428 struct port *port = shash_find_data(&old_ports, node->name);
1430 port = port_create(br, node->name);
1432 port_reconfigure(port, node->data);
1434 shash_destroy(&old_ports);
1435 shash_destroy(&new_ports);
1437 /* Check and delete duplicate interfaces. */
1439 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1440 svec_destroy(&ifaces);
1442 /* Delete all flows if we're switching from connected to standalone or vice
1443 * versa. (XXX Should we delete all flows if we are switching from one
1444 * controller to another?) */
1447 /* Configure OpenFlow management listeners. */
1448 svec_init(&listeners);
1449 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1451 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1452 ovs_rundir, br->name));
1453 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1454 svec_clear(&listeners);
1456 svec_sort_unique(&listeners);
1458 svec_init(&old_listeners);
1459 ofproto_get_listeners(br->ofproto, &old_listeners);
1460 svec_sort_unique(&old_listeners);
1462 if (!svec_equal(&listeners, &old_listeners)) {
1463 ofproto_set_listeners(br->ofproto, &listeners);
1465 svec_destroy(&listeners);
1466 svec_destroy(&old_listeners);
1468 /* Configure OpenFlow controller connection snooping. */
1470 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1472 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1473 ovs_rundir, br->name));
1474 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1475 svec_clear(&snoops);
1477 svec_sort_unique(&snoops);
1479 svec_init(&old_snoops);
1480 ofproto_get_snoops(br->ofproto, &old_snoops);
1481 svec_sort_unique(&old_snoops);
1483 if (!svec_equal(&snoops, &old_snoops)) {
1484 ofproto_set_snoops(br->ofproto, &snoops);
1486 svec_destroy(&snoops);
1487 svec_destroy(&old_snoops);
1489 /* Default listener. */
1490 svec_init(&listeners);
1491 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1492 ovs_rundir, br->name));
1493 svec_init(&old_listeners);
1494 ofproto_get_listeners(br->ofproto, &old_listeners);
1495 if (!svec_equal(&listeners, &old_listeners)) {
1496 ofproto_set_listeners(br->ofproto, &listeners);
1498 svec_destroy(&listeners);
1499 svec_destroy(&old_listeners);
1501 /* Default snoop. */
1503 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1504 ovs_rundir, br->name));
1505 svec_init(&old_snoops);
1506 ofproto_get_snoops(br->ofproto, &old_snoops);
1507 if (!svec_equal(&snoops, &old_snoops)) {
1508 ofproto_set_snoops(br->ofproto, &snoops);
1510 svec_destroy(&snoops);
1511 svec_destroy(&old_snoops);
1514 mirror_reconfigure(br);
1516 bridge_update_desc(br);
1520 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1523 char *pfx = xasprintf("bridge.%s.controller", br->name);
1524 const struct ovsrec_controller *c;
1526 c = bridge_get_controller(ovs_cfg, br);
1527 if ((br->controller != NULL) != (c != NULL)) {
1528 ofproto_flush_flows(br->ofproto);
1530 free(br->controller);
1531 br->controller = c ? xstrdup(c->target) : NULL;
1534 int max_backoff, probe;
1535 int rate_limit, burst_limit;
1537 if (!strcmp(c->target, "discover")) {
1538 ofproto_set_discovery(br->ofproto, true,
1539 c->discover_accept_regex,
1540 c->discover_update_resolv_conf);
1542 struct iface *local_iface;
1546 in_band = (!c->connection_mode
1547 || !strcmp(c->connection_mode, "out-of-band"));
1548 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1549 ofproto_set_in_band(br->ofproto, in_band);
1551 local_iface = bridge_get_local_iface(br);
1552 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1553 struct netdev *netdev = local_iface->netdev;
1554 struct in_addr mask, gateway;
1556 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1559 if (!c->local_gateway
1560 || !inet_aton(c->local_gateway, &gateway)) {
1564 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1566 mask.s_addr = guess_netmask(ip.s_addr);
1568 if (!netdev_set_in4(netdev, ip, mask)) {
1569 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1571 br->name, IP_ARGS(&ip.s_addr),
1572 IP_ARGS(&mask.s_addr));
1575 if (gateway.s_addr) {
1576 if (!netdev_add_router(netdev, gateway)) {
1577 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1578 br->name, IP_ARGS(&gateway.s_addr));
1584 ofproto_set_failure(br->ofproto,
1586 || !strcmp(c->fail_mode, "standalone")
1587 || !strcmp(c->fail_mode, "open")));
1589 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1590 ofproto_set_probe_interval(br->ofproto, probe);
1592 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1593 ofproto_set_max_backoff(br->ofproto, max_backoff);
1595 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1596 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1597 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1599 union ofp_action action;
1602 /* Set up a flow that matches every packet and directs them to
1603 * OFPP_NORMAL (which goes to us). */
1604 memset(&action, 0, sizeof action);
1605 action.type = htons(OFPAT_OUTPUT);
1606 action.output.len = htons(sizeof action);
1607 action.output.port = htons(OFPP_NORMAL);
1608 memset(&flow, 0, sizeof flow);
1609 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1612 ofproto_set_in_band(br->ofproto, false);
1613 ofproto_set_max_backoff(br->ofproto, 1);
1614 ofproto_set_probe_interval(br->ofproto, 5);
1615 ofproto_set_failure(br->ofproto, false);
1619 ofproto_set_controller(br->ofproto, br->controller);
1623 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1628 for (i = 0; i < br->n_ports; i++) {
1629 struct port *port = br->ports[i];
1630 for (j = 0; j < port->n_ifaces; j++) {
1631 struct iface *iface = port->ifaces[j];
1632 shash_add_once(ifaces, iface->name, iface);
1634 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1635 shash_add_once(ifaces, port->name, NULL);
1640 /* For robustness, in case the administrator moves around datapath ports behind
1641 * our back, we re-check all the datapath port numbers here.
1643 * This function will set the 'dp_ifidx' members of interfaces that have
1644 * disappeared to -1, so only call this function from a context where those
1645 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1646 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1647 * datapath, which doesn't support UINT16_MAX+1 ports. */
1649 bridge_fetch_dp_ifaces(struct bridge *br)
1651 struct odp_port *dpif_ports;
1652 size_t n_dpif_ports;
1655 /* Reset all interface numbers. */
1656 for (i = 0; i < br->n_ports; i++) {
1657 struct port *port = br->ports[i];
1658 for (j = 0; j < port->n_ifaces; j++) {
1659 struct iface *iface = port->ifaces[j];
1660 iface->dp_ifidx = -1;
1663 port_array_clear(&br->ifaces);
1665 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1666 for (i = 0; i < n_dpif_ports; i++) {
1667 struct odp_port *p = &dpif_ports[i];
1668 struct iface *iface = iface_lookup(br, p->devname);
1670 if (iface->dp_ifidx >= 0) {
1671 VLOG_WARN("%s reported interface %s twice",
1672 dpif_name(br->dpif), p->devname);
1673 } else if (iface_from_dp_ifidx(br, p->port)) {
1674 VLOG_WARN("%s reported interface %"PRIu16" twice",
1675 dpif_name(br->dpif), p->port);
1677 port_array_set(&br->ifaces, p->port, iface);
1678 iface->dp_ifidx = p->port;
1682 int64_t ofport = (iface->dp_ifidx >= 0
1683 ? odp_port_to_ofp_port(iface->dp_ifidx)
1685 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1692 /* Bridge packet processing functions. */
1695 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1697 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1700 static struct bond_entry *
1701 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1703 return &port->bond_hash[bond_hash(mac)];
1707 bond_choose_iface(const struct port *port)
1709 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1710 size_t i, best_down_slave = -1;
1711 long long next_delay_expiration = LLONG_MAX;
1713 for (i = 0; i < port->n_ifaces; i++) {
1714 struct iface *iface = port->ifaces[i];
1716 if (iface->enabled) {
1718 } else if (iface->delay_expires < next_delay_expiration) {
1719 best_down_slave = i;
1720 next_delay_expiration = iface->delay_expires;
1724 if (best_down_slave != -1) {
1725 struct iface *iface = port->ifaces[best_down_slave];
1727 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1728 "since no other interface is up", iface->name,
1729 iface->delay_expires - time_msec());
1730 bond_enable_slave(iface, true);
1733 return best_down_slave;
1737 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1738 uint16_t *dp_ifidx, tag_type *tags)
1740 struct iface *iface;
1742 assert(port->n_ifaces);
1743 if (port->n_ifaces == 1) {
1744 iface = port->ifaces[0];
1746 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1747 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1748 || !port->ifaces[e->iface_idx]->enabled) {
1749 /* XXX select interface properly. The current interface selection
1750 * is only good for testing the rebalancing code. */
1751 e->iface_idx = bond_choose_iface(port);
1752 if (e->iface_idx < 0) {
1753 *tags |= port->no_ifaces_tag;
1756 e->iface_tag = tag_create_random();
1757 ((struct port *) port)->bond_compat_is_stale = true;
1759 *tags |= e->iface_tag;
1760 iface = port->ifaces[e->iface_idx];
1762 *dp_ifidx = iface->dp_ifidx;
1763 *tags |= iface->tag; /* Currently only used for bonding. */
1768 bond_link_status_update(struct iface *iface, bool carrier)
1770 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1771 struct port *port = iface->port;
1773 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1774 /* Nothing to do. */
1777 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1778 iface->name, carrier ? "detected" : "dropped");
1779 if (carrier == iface->enabled) {
1780 iface->delay_expires = LLONG_MAX;
1781 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1782 iface->name, carrier ? "disabled" : "enabled");
1783 } else if (carrier && port->active_iface < 0) {
1784 bond_enable_slave(iface, true);
1785 if (port->updelay) {
1786 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1787 "other interface is up", iface->name, port->updelay);
1790 int delay = carrier ? port->updelay : port->downdelay;
1791 iface->delay_expires = time_msec() + delay;
1794 "interface %s: will be %s if it stays %s for %d ms",
1796 carrier ? "enabled" : "disabled",
1797 carrier ? "up" : "down",
1804 bond_choose_active_iface(struct port *port)
1806 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1808 port->active_iface = bond_choose_iface(port);
1809 port->active_iface_tag = tag_create_random();
1810 if (port->active_iface >= 0) {
1811 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1812 port->name, port->ifaces[port->active_iface]->name);
1814 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1820 bond_enable_slave(struct iface *iface, bool enable)
1822 struct port *port = iface->port;
1823 struct bridge *br = port->bridge;
1825 /* This acts as a recursion check. If the act of disabling a slave
1826 * causes a different slave to be enabled, the flag will allow us to
1827 * skip redundant work when we reenter this function. It must be
1828 * cleared on exit to keep things safe with multiple bonds. */
1829 static bool moving_active_iface = false;
1831 iface->delay_expires = LLONG_MAX;
1832 if (enable == iface->enabled) {
1836 iface->enabled = enable;
1837 if (!iface->enabled) {
1838 VLOG_WARN("interface %s: disabled", iface->name);
1839 ofproto_revalidate(br->ofproto, iface->tag);
1840 if (iface->port_ifidx == port->active_iface) {
1841 ofproto_revalidate(br->ofproto,
1842 port->active_iface_tag);
1844 /* Disabling a slave can lead to another slave being immediately
1845 * enabled if there will be no active slaves but one is waiting
1846 * on an updelay. In this case we do not need to run most of the
1847 * code for the newly enabled slave since there was no period
1848 * without an active slave and it is redundant with the disabling
1850 moving_active_iface = true;
1851 bond_choose_active_iface(port);
1853 bond_send_learning_packets(port);
1855 VLOG_WARN("interface %s: enabled", iface->name);
1856 if (port->active_iface < 0 && !moving_active_iface) {
1857 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1858 bond_choose_active_iface(port);
1859 bond_send_learning_packets(port);
1861 iface->tag = tag_create_random();
1864 moving_active_iface = false;
1865 port->bond_compat_is_stale = true;
1869 bond_run(struct bridge *br)
1873 for (i = 0; i < br->n_ports; i++) {
1874 struct port *port = br->ports[i];
1876 if (port->n_ifaces >= 2) {
1877 for (j = 0; j < port->n_ifaces; j++) {
1878 struct iface *iface = port->ifaces[j];
1879 if (time_msec() >= iface->delay_expires) {
1880 bond_enable_slave(iface, !iface->enabled);
1885 if (port->bond_compat_is_stale) {
1886 port->bond_compat_is_stale = false;
1887 port_update_bond_compat(port);
1893 bond_wait(struct bridge *br)
1897 for (i = 0; i < br->n_ports; i++) {
1898 struct port *port = br->ports[i];
1899 if (port->n_ifaces < 2) {
1902 for (j = 0; j < port->n_ifaces; j++) {
1903 struct iface *iface = port->ifaces[j];
1904 if (iface->delay_expires != LLONG_MAX) {
1905 poll_timer_wait(iface->delay_expires - time_msec());
1912 set_dst(struct dst *p, const flow_t *flow,
1913 const struct port *in_port, const struct port *out_port,
1916 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1917 : in_port->vlan >= 0 ? in_port->vlan
1918 : ntohs(flow->dl_vlan));
1919 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1923 swap_dst(struct dst *p, struct dst *q)
1925 struct dst tmp = *p;
1930 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1931 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1932 * that we push to the datapath. We could in fact fully sort the array by
1933 * vlan, but in most cases there are at most two different vlan tags so that's
1934 * possibly overkill.) */
1936 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1938 struct dst *first = dsts;
1939 struct dst *last = dsts + n_dsts;
1941 while (first != last) {
1943 * - All dsts < first have vlan == 'vlan'.
1944 * - All dsts >= last have vlan != 'vlan'.
1945 * - first < last. */
1946 while (first->vlan == vlan) {
1947 if (++first == last) {
1952 /* Same invariants, plus one additional:
1953 * - first->vlan != vlan.
1955 while (last[-1].vlan != vlan) {
1956 if (--last == first) {
1961 /* Same invariants, plus one additional:
1962 * - last[-1].vlan == vlan.*/
1963 swap_dst(first++, --last);
1968 mirror_mask_ffs(mirror_mask_t mask)
1970 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1975 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1976 const struct dst *test)
1979 for (i = 0; i < n_dsts; i++) {
1980 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1988 port_trunks_vlan(const struct port *port, uint16_t vlan)
1990 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1994 port_includes_vlan(const struct port *port, uint16_t vlan)
1996 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2000 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2001 const struct port *in_port, const struct port *out_port,
2002 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2004 mirror_mask_t mirrors = in_port->src_mirrors;
2005 struct dst *dst = dsts;
2008 if (out_port == FLOOD_PORT) {
2009 /* XXX use ODP_FLOOD if no vlans or bonding. */
2010 /* XXX even better, define each VLAN as a datapath port group */
2011 for (i = 0; i < br->n_ports; i++) {
2012 struct port *port = br->ports[i];
2013 if (port != in_port && port_includes_vlan(port, vlan)
2014 && !port->is_mirror_output_port
2015 && set_dst(dst, flow, in_port, port, tags)) {
2016 mirrors |= port->dst_mirrors;
2020 *nf_output_iface = NF_OUT_FLOOD;
2021 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2022 *nf_output_iface = dst->dp_ifidx;
2023 mirrors |= out_port->dst_mirrors;
2028 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2029 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2031 if (set_dst(dst, flow, in_port, m->out_port, tags)
2032 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2036 for (i = 0; i < br->n_ports; i++) {
2037 struct port *port = br->ports[i];
2038 if (port_includes_vlan(port, m->out_vlan)
2039 && set_dst(dst, flow, in_port, port, tags))
2043 if (port->vlan < 0) {
2044 dst->vlan = m->out_vlan;
2046 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2050 /* Use the vlan tag on the original flow instead of
2051 * the one passed in the vlan parameter. This ensures
2052 * that we compare the vlan from before any implicit
2053 * tagging tags place. This is necessary because
2054 * dst->vlan is the final vlan, after removing implicit
2056 flow_vlan = ntohs(flow->dl_vlan);
2057 if (flow_vlan == 0) {
2058 flow_vlan = OFP_VLAN_NONE;
2060 if (port == in_port && dst->vlan == flow_vlan) {
2061 /* Don't send out input port on same VLAN. */
2069 mirrors &= mirrors - 1;
2072 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2076 static void OVS_UNUSED
2077 print_dsts(const struct dst *dsts, size_t n)
2079 for (; n--; dsts++) {
2080 printf(">p%"PRIu16, dsts->dp_ifidx);
2081 if (dsts->vlan != OFP_VLAN_NONE) {
2082 printf("v%"PRIu16, dsts->vlan);
2088 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2089 const struct port *in_port, const struct port *out_port,
2090 tag_type *tags, struct odp_actions *actions,
2091 uint16_t *nf_output_iface)
2093 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2095 const struct dst *p;
2098 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2101 cur_vlan = ntohs(flow->dl_vlan);
2102 for (p = dsts; p < &dsts[n_dsts]; p++) {
2103 union odp_action *a;
2104 if (p->vlan != cur_vlan) {
2105 if (p->vlan == OFP_VLAN_NONE) {
2106 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2108 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2109 a->vlan_vid.vlan_vid = htons(p->vlan);
2113 a = odp_actions_add(actions, ODPAT_OUTPUT);
2114 a->output.port = p->dp_ifidx;
2118 /* Returns the effective vlan of a packet, taking into account both the
2119 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2120 * the packet is untagged and -1 indicates it has an invalid header and
2121 * should be dropped. */
2122 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2123 struct port *in_port, bool have_packet)
2125 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2126 * belongs to VLAN 0, so we should treat both cases identically. (In the
2127 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2128 * presumably to allow a priority to be specified. In the latter case, the
2129 * packet does not have any 802.1Q header.) */
2130 int vlan = ntohs(flow->dl_vlan);
2131 if (vlan == OFP_VLAN_NONE) {
2134 if (in_port->vlan >= 0) {
2136 /* XXX support double tagging? */
2138 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2139 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2140 "packet received on port %s configured with "
2141 "implicit VLAN %"PRIu16,
2142 br->name, ntohs(flow->dl_vlan),
2143 in_port->name, in_port->vlan);
2147 vlan = in_port->vlan;
2149 if (!port_includes_vlan(in_port, vlan)) {
2151 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2152 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2153 "packet received on port %s not configured for "
2155 br->name, vlan, in_port->name, vlan);
2165 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2166 struct port *in_port)
2168 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2169 vlan, in_port->port_idx);
2171 /* The log messages here could actually be useful in debugging,
2172 * so keep the rate limit relatively high. */
2173 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2175 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2176 "on port %s in VLAN %d",
2177 br->name, ETH_ADDR_ARGS(flow->dl_src),
2178 in_port->name, vlan);
2179 ofproto_revalidate(br->ofproto, rev_tag);
2184 is_bcast_arp_reply(const flow_t *flow)
2186 return (flow->dl_type == htons(ETH_TYPE_ARP)
2187 && flow->nw_proto == ARP_OP_REPLY
2188 && eth_addr_is_broadcast(flow->dl_dst));
2191 /* If the composed actions may be applied to any packet in the given 'flow',
2192 * returns true. Otherwise, the actions should only be applied to 'packet', or
2193 * not at all, if 'packet' was NULL. */
2195 process_flow(struct bridge *br, const flow_t *flow,
2196 const struct ofpbuf *packet, struct odp_actions *actions,
2197 tag_type *tags, uint16_t *nf_output_iface)
2199 struct iface *in_iface;
2200 struct port *in_port;
2201 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2205 /* Find the interface and port structure for the received packet. */
2206 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2208 /* No interface? Something fishy... */
2209 if (packet != NULL) {
2210 /* Odd. A few possible reasons here:
2212 * - We deleted an interface but there are still a few packets
2213 * queued up from it.
2215 * - Someone externally added an interface (e.g. with "ovs-dpctl
2216 * add-if") that we don't know about.
2218 * - Packet arrived on the local port but the local port is not
2219 * one of our bridge ports.
2221 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2223 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2224 "interface %"PRIu16, br->name, flow->in_port);
2227 /* Return without adding any actions, to drop packets on this flow. */
2230 in_port = in_iface->port;
2231 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2236 /* Drop frames for reserved multicast addresses. */
2237 if (eth_addr_is_reserved(flow->dl_dst)) {
2241 /* Drop frames on ports reserved for mirroring. */
2242 if (in_port->is_mirror_output_port) {
2243 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2244 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2245 "which is reserved exclusively for mirroring",
2246 br->name, in_port->name);
2250 /* Packets received on bonds need special attention to avoid duplicates. */
2251 if (in_port->n_ifaces > 1) {
2254 if (eth_addr_is_multicast(flow->dl_dst)) {
2255 *tags |= in_port->active_iface_tag;
2256 if (in_port->active_iface != in_iface->port_ifidx) {
2257 /* Drop all multicast packets on inactive slaves. */
2262 /* Drop all packets for which we have learned a different input
2263 * port, because we probably sent the packet on one slave and got
2264 * it back on the other. Broadcast ARP replies are an exception
2265 * to this rule: the host has moved to another switch. */
2266 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2267 if (src_idx != -1 && src_idx != in_port->port_idx &&
2268 !is_bcast_arp_reply(flow)) {
2274 out_port = FLOOD_PORT;
2275 /* Learn source MAC (but don't try to learn from revalidation). */
2277 update_learning_table(br, flow, vlan, in_port);
2280 /* Determine output port. */
2281 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2283 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2284 out_port = br->ports[out_port_idx];
2285 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2286 /* If we are revalidating but don't have a learning entry then
2287 * eject the flow. Installing a flow that floods packets opens
2288 * up a window of time where we could learn from a packet reflected
2289 * on a bond and blackhole packets before the learning table is
2290 * updated to reflect the correct port. */
2294 /* Don't send packets out their input ports. */
2295 if (in_port == out_port) {
2300 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2306 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2309 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2310 const struct ofp_phy_port *opp,
2313 struct bridge *br = br_;
2314 struct iface *iface;
2317 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2323 if (reason == OFPPR_DELETE) {
2324 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2325 br->name, iface->name);
2326 iface_destroy(iface);
2327 if (!port->n_ifaces) {
2328 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2329 br->name, port->name);
2335 if (port->n_ifaces > 1) {
2336 bool up = !(opp->state & OFPPS_LINK_DOWN);
2337 bond_link_status_update(iface, up);
2338 port_update_bond_compat(port);
2344 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2345 struct odp_actions *actions, tag_type *tags,
2346 uint16_t *nf_output_iface, void *br_)
2348 struct bridge *br = br_;
2350 COVERAGE_INC(bridge_process_flow);
2351 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2355 bridge_account_flow_ofhook_cb(const flow_t *flow,
2356 const union odp_action *actions,
2357 size_t n_actions, unsigned long long int n_bytes,
2360 struct bridge *br = br_;
2361 struct port *in_port;
2362 const union odp_action *a;
2364 /* Feed information from the active flows back into the learning table
2365 * to ensure that table is always in sync with what is actually flowing
2366 * through the datapath. */
2367 in_port = port_from_dp_ifidx(br, flow->in_port);
2369 int vlan = flow_get_vlan(br, flow, in_port, false);
2371 update_learning_table(br, flow, vlan, in_port);
2375 if (!br->has_bonded_ports) {
2379 for (a = actions; a < &actions[n_actions]; a++) {
2380 if (a->type == ODPAT_OUTPUT) {
2381 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2382 if (out_port && out_port->n_ifaces >= 2) {
2383 struct bond_entry *e = lookup_bond_entry(out_port,
2385 e->tx_bytes += n_bytes;
2392 bridge_account_checkpoint_ofhook_cb(void *br_)
2394 struct bridge *br = br_;
2397 if (!br->has_bonded_ports) {
2401 /* The current ofproto implementation calls this callback at least once a
2402 * second, so this timer implementation is sufficient. */
2403 if (time_msec() < br->bond_next_rebalance) {
2406 br->bond_next_rebalance = time_msec() + 10000;
2408 for (i = 0; i < br->n_ports; i++) {
2409 struct port *port = br->ports[i];
2410 if (port->n_ifaces > 1) {
2411 bond_rebalance_port(port);
2416 static struct ofhooks bridge_ofhooks = {
2417 bridge_port_changed_ofhook_cb,
2418 bridge_normal_ofhook_cb,
2419 bridge_account_flow_ofhook_cb,
2420 bridge_account_checkpoint_ofhook_cb,
2423 /* Bonding functions. */
2425 /* Statistics for a single interface on a bonded port, used for load-based
2426 * bond rebalancing. */
2427 struct slave_balance {
2428 struct iface *iface; /* The interface. */
2429 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2431 /* All the "bond_entry"s that are assigned to this interface, in order of
2432 * increasing tx_bytes. */
2433 struct bond_entry **hashes;
2437 /* Sorts pointers to pointers to bond_entries in ascending order by the
2438 * interface to which they are assigned, and within a single interface in
2439 * ascending order of bytes transmitted. */
2441 compare_bond_entries(const void *a_, const void *b_)
2443 const struct bond_entry *const *ap = a_;
2444 const struct bond_entry *const *bp = b_;
2445 const struct bond_entry *a = *ap;
2446 const struct bond_entry *b = *bp;
2447 if (a->iface_idx != b->iface_idx) {
2448 return a->iface_idx > b->iface_idx ? 1 : -1;
2449 } else if (a->tx_bytes != b->tx_bytes) {
2450 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2456 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2457 * *descending* order by number of bytes transmitted. */
2459 compare_slave_balance(const void *a_, const void *b_)
2461 const struct slave_balance *a = a_;
2462 const struct slave_balance *b = b_;
2463 if (a->iface->enabled != b->iface->enabled) {
2464 return a->iface->enabled ? -1 : 1;
2465 } else if (a->tx_bytes != b->tx_bytes) {
2466 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2473 swap_bals(struct slave_balance *a, struct slave_balance *b)
2475 struct slave_balance tmp = *a;
2480 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2481 * given that 'p' (and only 'p') might be in the wrong location.
2483 * This function invalidates 'p', since it might now be in a different memory
2486 resort_bals(struct slave_balance *p,
2487 struct slave_balance bals[], size_t n_bals)
2490 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2491 swap_bals(p, p - 1);
2493 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2494 swap_bals(p, p + 1);
2500 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2502 if (VLOG_IS_DBG_ENABLED()) {
2503 struct ds ds = DS_EMPTY_INITIALIZER;
2504 const struct slave_balance *b;
2506 for (b = bals; b < bals + n_bals; b++) {
2510 ds_put_char(&ds, ',');
2512 ds_put_format(&ds, " %s %"PRIu64"kB",
2513 b->iface->name, b->tx_bytes / 1024);
2515 if (!b->iface->enabled) {
2516 ds_put_cstr(&ds, " (disabled)");
2518 if (b->n_hashes > 0) {
2519 ds_put_cstr(&ds, " (");
2520 for (i = 0; i < b->n_hashes; i++) {
2521 const struct bond_entry *e = b->hashes[i];
2523 ds_put_cstr(&ds, " + ");
2525 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2526 e - port->bond_hash, e->tx_bytes / 1024);
2528 ds_put_cstr(&ds, ")");
2531 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2536 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2538 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2541 struct bond_entry *hash = from->hashes[hash_idx];
2542 struct port *port = from->iface->port;
2543 uint64_t delta = hash->tx_bytes;
2545 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2546 "from %s to %s (now carrying %"PRIu64"kB and "
2547 "%"PRIu64"kB load, respectively)",
2548 port->name, delta / 1024, hash - port->bond_hash,
2549 from->iface->name, to->iface->name,
2550 (from->tx_bytes - delta) / 1024,
2551 (to->tx_bytes + delta) / 1024);
2553 /* Delete element from from->hashes.
2555 * We don't bother to add the element to to->hashes because not only would
2556 * it require more work, the only purpose it would be to allow that hash to
2557 * be migrated to another slave in this rebalancing run, and there is no
2558 * point in doing that. */
2559 if (hash_idx == 0) {
2562 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2563 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2567 /* Shift load away from 'from' to 'to'. */
2568 from->tx_bytes -= delta;
2569 to->tx_bytes += delta;
2571 /* Arrange for flows to be revalidated. */
2572 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2573 hash->iface_idx = to->iface->port_ifidx;
2574 hash->iface_tag = tag_create_random();
2578 bond_rebalance_port(struct port *port)
2580 struct slave_balance bals[DP_MAX_PORTS];
2582 struct bond_entry *hashes[BOND_MASK + 1];
2583 struct slave_balance *b, *from, *to;
2584 struct bond_entry *e;
2587 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2588 * descending order of tx_bytes, so that bals[0] represents the most
2589 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2592 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2593 * array for each slave_balance structure, we sort our local array of
2594 * hashes in order by slave, so that all of the hashes for a given slave
2595 * become contiguous in memory, and then we point each 'hashes' members of
2596 * a slave_balance structure to the start of a contiguous group. */
2597 n_bals = port->n_ifaces;
2598 for (b = bals; b < &bals[n_bals]; b++) {
2599 b->iface = port->ifaces[b - bals];
2604 for (i = 0; i <= BOND_MASK; i++) {
2605 hashes[i] = &port->bond_hash[i];
2607 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2608 for (i = 0; i <= BOND_MASK; i++) {
2610 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2611 b = &bals[e->iface_idx];
2612 b->tx_bytes += e->tx_bytes;
2614 b->hashes = &hashes[i];
2619 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2620 log_bals(bals, n_bals, port);
2622 /* Discard slaves that aren't enabled (which were sorted to the back of the
2623 * array earlier). */
2624 while (!bals[n_bals - 1].iface->enabled) {
2631 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2632 to = &bals[n_bals - 1];
2633 for (from = bals; from < to; ) {
2634 uint64_t overload = from->tx_bytes - to->tx_bytes;
2635 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2636 /* The extra load on 'from' (and all less-loaded slaves), compared
2637 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2638 * it is less than ~1Mbps. No point in rebalancing. */
2640 } else if (from->n_hashes == 1) {
2641 /* 'from' only carries a single MAC hash, so we can't shift any
2642 * load away from it, even though we want to. */
2645 /* 'from' is carrying significantly more load than 'to', and that
2646 * load is split across at least two different hashes. Pick a hash
2647 * to migrate to 'to' (the least-loaded slave), given that doing so
2648 * must decrease the ratio of the load on the two slaves by at
2651 * The sort order we use means that we prefer to shift away the
2652 * smallest hashes instead of the biggest ones. There is little
2653 * reason behind this decision; we could use the opposite sort
2654 * order to shift away big hashes ahead of small ones. */
2658 for (i = 0; i < from->n_hashes; i++) {
2659 double old_ratio, new_ratio;
2660 uint64_t delta = from->hashes[i]->tx_bytes;
2662 if (delta == 0 || from->tx_bytes - delta == 0) {
2663 /* Pointless move. */
2667 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2669 if (to->tx_bytes == 0) {
2670 /* Nothing on the new slave, move it. */
2674 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2675 new_ratio = (double)(from->tx_bytes - delta) /
2676 (to->tx_bytes + delta);
2678 if (new_ratio == 0) {
2679 /* Should already be covered but check to prevent division
2684 if (new_ratio < 1) {
2685 new_ratio = 1 / new_ratio;
2688 if (old_ratio - new_ratio > 0.1) {
2689 /* Would decrease the ratio, move it. */
2693 if (i < from->n_hashes) {
2694 bond_shift_load(from, to, i);
2695 port->bond_compat_is_stale = true;
2697 /* If the result of the migration changed the relative order of
2698 * 'from' and 'to' swap them back to maintain invariants. */
2699 if (order_swapped) {
2700 swap_bals(from, to);
2703 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2704 * point to different slave_balance structures. It is only
2705 * valid to do these two operations in a row at all because we
2706 * know that 'from' will not move past 'to' and vice versa. */
2707 resort_bals(from, bals, n_bals);
2708 resort_bals(to, bals, n_bals);
2715 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2716 * historical data to decay to <1% in 7 rebalancing runs. */
2717 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2723 bond_send_learning_packets(struct port *port)
2725 struct bridge *br = port->bridge;
2726 struct mac_entry *e;
2727 struct ofpbuf packet;
2728 int error, n_packets, n_errors;
2730 if (!port->n_ifaces || port->active_iface < 0) {
2734 ofpbuf_init(&packet, 128);
2735 error = n_packets = n_errors = 0;
2736 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2737 union ofp_action actions[2], *a;
2743 if (e->port == port->port_idx
2744 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2748 /* Compose actions. */
2749 memset(actions, 0, sizeof actions);
2752 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2753 a->vlan_vid.len = htons(sizeof *a);
2754 a->vlan_vid.vlan_vid = htons(e->vlan);
2757 a->output.type = htons(OFPAT_OUTPUT);
2758 a->output.len = htons(sizeof *a);
2759 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2764 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2766 flow_extract(&packet, ODPP_NONE, &flow);
2767 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2774 ofpbuf_uninit(&packet);
2777 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2778 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2779 "packets, last error was: %s",
2780 port->name, n_errors, n_packets, strerror(error));
2782 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2783 port->name, n_packets);
2787 /* Bonding unixctl user interface functions. */
2790 bond_unixctl_list(struct unixctl_conn *conn,
2791 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2793 struct ds ds = DS_EMPTY_INITIALIZER;
2794 const struct bridge *br;
2796 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2798 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2801 for (i = 0; i < br->n_ports; i++) {
2802 const struct port *port = br->ports[i];
2803 if (port->n_ifaces > 1) {
2806 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2807 for (j = 0; j < port->n_ifaces; j++) {
2808 const struct iface *iface = port->ifaces[j];
2810 ds_put_cstr(&ds, ", ");
2812 ds_put_cstr(&ds, iface->name);
2814 ds_put_char(&ds, '\n');
2818 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2822 static struct port *
2823 bond_find(const char *name)
2825 const struct bridge *br;
2827 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2830 for (i = 0; i < br->n_ports; i++) {
2831 struct port *port = br->ports[i];
2832 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2841 bond_unixctl_show(struct unixctl_conn *conn,
2842 const char *args, void *aux OVS_UNUSED)
2844 struct ds ds = DS_EMPTY_INITIALIZER;
2845 const struct port *port;
2848 port = bond_find(args);
2850 unixctl_command_reply(conn, 501, "no such bond");
2854 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2855 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2856 ds_put_format(&ds, "next rebalance: %lld ms\n",
2857 port->bridge->bond_next_rebalance - time_msec());
2858 for (j = 0; j < port->n_ifaces; j++) {
2859 const struct iface *iface = port->ifaces[j];
2860 struct bond_entry *be;
2863 ds_put_format(&ds, "slave %s: %s\n",
2864 iface->name, iface->enabled ? "enabled" : "disabled");
2865 if (j == port->active_iface) {
2866 ds_put_cstr(&ds, "\tactive slave\n");
2868 if (iface->delay_expires != LLONG_MAX) {
2869 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2870 iface->enabled ? "downdelay" : "updelay",
2871 iface->delay_expires - time_msec());
2875 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2876 int hash = be - port->bond_hash;
2877 struct mac_entry *me;
2879 if (be->iface_idx != j) {
2883 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2884 hash, be->tx_bytes / 1024);
2887 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2888 &port->bridge->ml->lrus) {
2891 if (bond_hash(me->mac) == hash
2892 && me->port != port->port_idx
2893 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2894 && dp_ifidx == iface->dp_ifidx)
2896 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2897 ETH_ADDR_ARGS(me->mac));
2902 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2907 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2908 void *aux OVS_UNUSED)
2910 char *args = (char *) args_;
2911 char *save_ptr = NULL;
2912 char *bond_s, *hash_s, *slave_s;
2913 uint8_t mac[ETH_ADDR_LEN];
2915 struct iface *iface;
2916 struct bond_entry *entry;
2919 bond_s = strtok_r(args, " ", &save_ptr);
2920 hash_s = strtok_r(NULL, " ", &save_ptr);
2921 slave_s = strtok_r(NULL, " ", &save_ptr);
2923 unixctl_command_reply(conn, 501,
2924 "usage: bond/migrate BOND HASH SLAVE");
2928 port = bond_find(bond_s);
2930 unixctl_command_reply(conn, 501, "no such bond");
2934 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2935 == ETH_ADDR_SCAN_COUNT) {
2936 hash = bond_hash(mac);
2937 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2938 hash = atoi(hash_s) & BOND_MASK;
2940 unixctl_command_reply(conn, 501, "bad hash");
2944 iface = port_lookup_iface(port, slave_s);
2946 unixctl_command_reply(conn, 501, "no such slave");
2950 if (!iface->enabled) {
2951 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2955 entry = &port->bond_hash[hash];
2956 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2957 entry->iface_idx = iface->port_ifidx;
2958 entry->iface_tag = tag_create_random();
2959 port->bond_compat_is_stale = true;
2960 unixctl_command_reply(conn, 200, "migrated");
2964 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2965 void *aux OVS_UNUSED)
2967 char *args = (char *) args_;
2968 char *save_ptr = NULL;
2969 char *bond_s, *slave_s;
2971 struct iface *iface;
2973 bond_s = strtok_r(args, " ", &save_ptr);
2974 slave_s = strtok_r(NULL, " ", &save_ptr);
2976 unixctl_command_reply(conn, 501,
2977 "usage: bond/set-active-slave BOND SLAVE");
2981 port = bond_find(bond_s);
2983 unixctl_command_reply(conn, 501, "no such bond");
2987 iface = port_lookup_iface(port, slave_s);
2989 unixctl_command_reply(conn, 501, "no such slave");
2993 if (!iface->enabled) {
2994 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2998 if (port->active_iface != iface->port_ifidx) {
2999 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3000 port->active_iface = iface->port_ifidx;
3001 port->active_iface_tag = tag_create_random();
3002 VLOG_INFO("port %s: active interface is now %s",
3003 port->name, iface->name);
3004 bond_send_learning_packets(port);
3005 unixctl_command_reply(conn, 200, "done");
3007 unixctl_command_reply(conn, 200, "no change");
3012 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3014 char *args = (char *) args_;
3015 char *save_ptr = NULL;
3016 char *bond_s, *slave_s;
3018 struct iface *iface;
3020 bond_s = strtok_r(args, " ", &save_ptr);
3021 slave_s = strtok_r(NULL, " ", &save_ptr);
3023 unixctl_command_reply(conn, 501,
3024 "usage: bond/enable/disable-slave BOND SLAVE");
3028 port = bond_find(bond_s);
3030 unixctl_command_reply(conn, 501, "no such bond");
3034 iface = port_lookup_iface(port, slave_s);
3036 unixctl_command_reply(conn, 501, "no such slave");
3040 bond_enable_slave(iface, enable);
3041 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3045 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3046 void *aux OVS_UNUSED)
3048 enable_slave(conn, args, true);
3052 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3053 void *aux OVS_UNUSED)
3055 enable_slave(conn, args, false);
3059 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3060 void *aux OVS_UNUSED)
3062 uint8_t mac[ETH_ADDR_LEN];
3066 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3067 == ETH_ADDR_SCAN_COUNT) {
3068 hash = bond_hash(mac);
3070 hash_cstr = xasprintf("%u", hash);
3071 unixctl_command_reply(conn, 200, hash_cstr);
3074 unixctl_command_reply(conn, 501, "invalid mac");
3081 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3082 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3083 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3084 unixctl_command_register("bond/set-active-slave",
3085 bond_unixctl_set_active_slave, NULL);
3086 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3088 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3090 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3093 /* Port functions. */
3095 static struct port *
3096 port_create(struct bridge *br, const char *name)
3100 port = xzalloc(sizeof *port);
3102 port->port_idx = br->n_ports;
3104 port->trunks = NULL;
3105 port->name = xstrdup(name);
3106 port->active_iface = -1;
3108 if (br->n_ports >= br->allocated_ports) {
3109 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3112 br->ports[br->n_ports++] = port;
3114 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3121 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3123 struct shash old_ifaces, new_ifaces;
3124 struct shash_node *node;
3125 unsigned long *trunks;
3131 /* Collect old and new interfaces. */
3132 shash_init(&old_ifaces);
3133 shash_init(&new_ifaces);
3134 for (i = 0; i < port->n_ifaces; i++) {
3135 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3137 for (i = 0; i < cfg->n_interfaces; i++) {
3138 const char *name = cfg->interfaces[i]->name;
3139 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3140 VLOG_WARN("port %s: %s specified twice as port interface",
3144 port->updelay = cfg->bond_updelay;
3145 if (port->updelay < 0) {
3148 port->updelay = cfg->bond_downdelay;
3149 if (port->downdelay < 0) {
3150 port->downdelay = 0;
3153 /* Get rid of deleted interfaces and add new interfaces. */
3154 SHASH_FOR_EACH (node, &old_ifaces) {
3155 if (!shash_find(&new_ifaces, node->name)) {
3156 iface_destroy(node->data);
3159 SHASH_FOR_EACH (node, &new_ifaces) {
3160 const struct ovsrec_interface *if_cfg = node->data;
3161 struct iface *iface;
3163 iface = shash_find_data(&old_ifaces, if_cfg->name);
3165 iface_create(port, if_cfg);
3167 iface->cfg = if_cfg;
3174 if (port->n_ifaces < 2) {
3176 if (vlan >= 0 && vlan <= 4095) {
3177 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3182 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3183 * they even work as-is. But they have not been tested. */
3184 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3188 if (port->vlan != vlan) {
3190 bridge_flush(port->bridge);
3193 /* Get trunked VLANs. */
3199 trunks = bitmap_allocate(4096);
3201 for (i = 0; i < cfg->n_trunks; i++) {
3202 int trunk = cfg->trunks[i];
3204 bitmap_set1(trunks, trunk);
3210 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3211 port->name, cfg->n_trunks);
3213 if (n_errors == cfg->n_trunks) {
3215 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3218 bitmap_set_multiple(trunks, 0, 4096, 1);
3221 if (cfg->n_trunks) {
3222 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3227 ? port->trunks != NULL
3228 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3229 bridge_flush(port->bridge);
3231 bitmap_free(port->trunks);
3232 port->trunks = trunks;
3234 shash_destroy(&old_ifaces);
3235 shash_destroy(&new_ifaces);
3239 port_destroy(struct port *port)
3242 struct bridge *br = port->bridge;
3246 proc_net_compat_update_vlan(port->name, NULL, 0);
3247 proc_net_compat_update_bond(port->name, NULL);
3249 for (i = 0; i < MAX_MIRRORS; i++) {
3250 struct mirror *m = br->mirrors[i];
3251 if (m && m->out_port == port) {
3256 while (port->n_ifaces > 0) {
3257 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3260 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3261 del->port_idx = port->port_idx;
3264 bitmap_free(port->trunks);
3271 static struct port *
3272 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3274 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3275 return iface ? iface->port : NULL;
3278 static struct port *
3279 port_lookup(const struct bridge *br, const char *name)
3283 for (i = 0; i < br->n_ports; i++) {
3284 struct port *port = br->ports[i];
3285 if (!strcmp(port->name, name)) {
3292 static struct iface *
3293 port_lookup_iface(const struct port *port, const char *name)
3297 for (j = 0; j < port->n_ifaces; j++) {
3298 struct iface *iface = port->ifaces[j];
3299 if (!strcmp(iface->name, name)) {
3307 port_update_bonding(struct port *port)
3309 if (port->n_ifaces < 2) {
3310 /* Not a bonded port. */
3311 if (port->bond_hash) {
3312 free(port->bond_hash);
3313 port->bond_hash = NULL;
3314 port->bond_compat_is_stale = true;
3315 port->bond_fake_iface = false;
3318 if (!port->bond_hash) {
3321 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3322 for (i = 0; i <= BOND_MASK; i++) {
3323 struct bond_entry *e = &port->bond_hash[i];
3327 port->no_ifaces_tag = tag_create_random();
3328 bond_choose_active_iface(port);
3330 port->bond_compat_is_stale = true;
3331 port->bond_fake_iface = port->cfg->bond_fake_iface;
3336 port_update_bond_compat(struct port *port)
3338 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3339 struct compat_bond bond;
3342 if (port->n_ifaces < 2) {
3343 proc_net_compat_update_bond(port->name, NULL);
3348 bond.updelay = port->updelay;
3349 bond.downdelay = port->downdelay;
3352 bond.hashes = compat_hashes;
3353 if (port->bond_hash) {
3354 const struct bond_entry *e;
3355 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3356 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3357 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3358 cbh->hash = e - port->bond_hash;
3359 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3364 bond.n_slaves = port->n_ifaces;
3365 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3366 for (i = 0; i < port->n_ifaces; i++) {
3367 struct iface *iface = port->ifaces[i];
3368 struct compat_bond_slave *slave = &bond.slaves[i];
3369 slave->name = iface->name;
3371 /* We need to make the same determination as the Linux bonding
3372 * code to determine whether a slave should be consider "up".
3373 * The Linux function bond_miimon_inspect() supports four
3374 * BOND_LINK_* states:
3376 * - BOND_LINK_UP: carrier detected, updelay has passed.
3377 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3378 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3379 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3381 * The function bond_info_show_slave() only considers BOND_LINK_UP
3382 * to be "up" and anything else to be "down".
3384 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3388 netdev_get_etheraddr(iface->netdev, slave->mac);
3391 if (port->bond_fake_iface) {
3392 struct netdev *bond_netdev;
3394 if (!netdev_open_default(port->name, &bond_netdev)) {
3396 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3398 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3400 netdev_close(bond_netdev);
3404 proc_net_compat_update_bond(port->name, &bond);
3409 port_update_vlan_compat(struct port *port)
3411 struct bridge *br = port->bridge;
3412 char *vlandev_name = NULL;
3414 if (port->vlan > 0) {
3415 /* Figure out the name that the VLAN device should actually have, if it
3416 * existed. This takes some work because the VLAN device would not
3417 * have port->name in its name; rather, it would have the trunk port's
3418 * name, and 'port' would be attached to a bridge that also had the
3419 * VLAN device one of its ports. So we need to find a trunk port that
3420 * includes port->vlan.
3422 * There might be more than one candidate. This doesn't happen on
3423 * XenServer, so if it happens we just pick the first choice in
3424 * alphabetical order instead of creating multiple VLAN devices. */
3426 for (i = 0; i < br->n_ports; i++) {
3427 struct port *p = br->ports[i];
3428 if (port_trunks_vlan(p, port->vlan)
3430 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3432 uint8_t ea[ETH_ADDR_LEN];
3433 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3434 if (!eth_addr_is_multicast(ea) &&
3435 !eth_addr_is_reserved(ea) &&
3436 !eth_addr_is_zero(ea)) {
3437 vlandev_name = p->name;
3442 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3445 /* Interface functions. */
3447 static struct iface *
3448 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3450 struct iface *iface;
3451 char *name = if_cfg->name;
3454 iface = xzalloc(sizeof *iface);
3456 iface->port_ifidx = port->n_ifaces;
3457 iface->name = xstrdup(name);
3458 iface->dp_ifidx = -1;
3459 iface->tag = tag_create_random();
3460 iface->delay_expires = LLONG_MAX;
3461 iface->netdev = NULL;
3462 iface->cfg = if_cfg;
3464 if (port->n_ifaces >= port->allocated_ifaces) {
3465 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3466 sizeof *port->ifaces);
3468 port->ifaces[port->n_ifaces++] = iface;
3469 if (port->n_ifaces > 1) {
3470 port->bridge->has_bonded_ports = true;
3473 /* Attempt to create the network interface in case it
3474 * doesn't exist yet. */
3475 if (!iface_is_internal(port->bridge, iface->name)) {
3476 error = set_up_iface(if_cfg, iface, true);
3478 VLOG_WARN("could not create iface %s: %s", iface->name,
3483 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3485 bridge_flush(port->bridge);
3491 iface_destroy(struct iface *iface)
3494 struct port *port = iface->port;
3495 struct bridge *br = port->bridge;
3496 bool del_active = port->active_iface == iface->port_ifidx;
3499 if (iface->dp_ifidx >= 0) {
3500 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3503 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3504 del->port_ifidx = iface->port_ifidx;
3506 netdev_close(iface->netdev);
3509 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3510 bond_choose_active_iface(port);
3511 bond_send_learning_packets(port);
3517 bridge_flush(port->bridge);
3521 static struct iface *
3522 iface_lookup(const struct bridge *br, const char *name)
3526 for (i = 0; i < br->n_ports; i++) {
3527 struct port *port = br->ports[i];
3528 for (j = 0; j < port->n_ifaces; j++) {
3529 struct iface *iface = port->ifaces[j];
3530 if (!strcmp(iface->name, name)) {
3538 static struct iface *
3539 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3541 return port_array_get(&br->ifaces, dp_ifidx);
3544 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3545 * 'br', that is, an interface that is entirely simulated within the datapath.
3546 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3547 * interfaces are created by setting "iface.<iface>.internal = true".
3549 * In addition, we have a kluge-y feature that creates an internal port with
3550 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3551 * This feature needs to go away in the long term. Until then, this is one
3552 * reason why this function takes a name instead of a struct iface: the fake
3553 * interfaces created this way do not have a struct iface. */
3555 iface_is_internal(const struct bridge *br, const char *if_name)
3557 /* XXX wastes time */
3558 struct iface *iface;
3561 if (!strcmp(if_name, br->name)) {
3565 iface = iface_lookup(br, if_name);
3566 if (iface && !strcmp(iface->cfg->type, "internal")) {
3570 port = port_lookup(br, if_name);
3571 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3577 /* Set Ethernet address of 'iface', if one is specified in the configuration
3580 iface_set_mac(struct iface *iface)
3582 uint8_t ea[ETH_ADDR_LEN];
3584 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3585 if (eth_addr_is_multicast(ea)) {
3586 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3588 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3589 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3590 iface->name, iface->name);
3592 int error = netdev_set_etheraddr(iface->netdev, ea);
3594 VLOG_ERR("interface %s: setting MAC failed (%s)",
3595 iface->name, strerror(error));
3601 /* Port mirroring. */
3604 mirror_reconfigure(struct bridge *br)
3606 struct shash old_mirrors, new_mirrors;
3607 struct shash_node *node;
3608 unsigned long *rspan_vlans;
3611 /* Collect old mirrors. */
3612 shash_init(&old_mirrors);
3613 for (i = 0; i < MAX_MIRRORS; i++) {
3614 if (br->mirrors[i]) {
3615 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3619 /* Collect new mirrors. */
3620 shash_init(&new_mirrors);
3621 for (i = 0; i < br->cfg->n_mirrors; i++) {
3622 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3623 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3624 VLOG_WARN("bridge %s: %s specified twice as mirror",
3625 br->name, cfg->name);
3629 /* Get rid of deleted mirrors and add new mirrors. */
3630 SHASH_FOR_EACH (node, &old_mirrors) {
3631 if (!shash_find(&new_mirrors, node->name)) {
3632 mirror_destroy(node->data);
3635 SHASH_FOR_EACH (node, &new_mirrors) {
3636 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3638 mirror = mirror_create(br, node->name);
3643 mirror_reconfigure_one(mirror, node->data);
3645 shash_destroy(&old_mirrors);
3646 shash_destroy(&new_mirrors);
3648 /* Update port reserved status. */
3649 for (i = 0; i < br->n_ports; i++) {
3650 br->ports[i]->is_mirror_output_port = false;
3652 for (i = 0; i < MAX_MIRRORS; i++) {
3653 struct mirror *m = br->mirrors[i];
3654 if (m && m->out_port) {
3655 m->out_port->is_mirror_output_port = true;
3659 /* Update flooded vlans (for RSPAN). */
3661 if (br->cfg->n_flood_vlans) {
3662 rspan_vlans = bitmap_allocate(4096);
3664 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3665 int64_t vlan = br->cfg->flood_vlans[i];
3666 if (vlan >= 0 && vlan < 4096) {
3667 bitmap_set1(rspan_vlans, vlan);
3668 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3671 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3676 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3681 static struct mirror *
3682 mirror_create(struct bridge *br, const char *name)
3687 for (i = 0; ; i++) {
3688 if (i >= MAX_MIRRORS) {
3689 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3690 "cannot create %s", br->name, MAX_MIRRORS, name);
3693 if (!br->mirrors[i]) {
3698 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3701 br->mirrors[i] = m = xzalloc(sizeof *m);
3704 m->name = xstrdup(name);
3705 shash_init(&m->src_ports);
3706 shash_init(&m->dst_ports);
3716 mirror_destroy(struct mirror *m)
3719 struct bridge *br = m->bridge;
3722 for (i = 0; i < br->n_ports; i++) {
3723 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3724 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3727 shash_destroy(&m->src_ports);
3728 shash_destroy(&m->dst_ports);
3731 m->bridge->mirrors[m->idx] = NULL;
3739 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3740 struct shash *names)
3744 for (i = 0; i < n_ports; i++) {
3745 const char *name = ports[i]->name;
3746 if (port_lookup(m->bridge, name)) {
3747 shash_add_once(names, name, NULL);
3749 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3750 "port %s", m->bridge->name, m->name, name);
3756 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3762 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3764 for (i = 0; i < cfg->n_select_vlan; i++) {
3765 int64_t vlan = cfg->select_vlan[i];
3766 if (vlan < 0 || vlan > 4095) {
3767 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3768 m->bridge->name, m->name, vlan);
3770 (*vlans)[n_vlans++] = vlan;
3777 vlan_is_mirrored(const struct mirror *m, int vlan)
3781 for (i = 0; i < m->n_vlans; i++) {
3782 if (m->vlans[i] == vlan) {
3790 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3794 for (i = 0; i < m->n_vlans; i++) {
3795 if (port_trunks_vlan(p, m->vlans[i])) {
3803 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3805 struct shash src_ports, dst_ports;
3806 mirror_mask_t mirror_bit;
3807 struct port *out_port;
3812 bool mirror_all_ports;
3813 bool any_ports_specified;
3814 bool any_vlans_specified;
3816 /* Get output port. */
3817 if (cfg->output_port) {
3818 out_port = port_lookup(m->bridge, cfg->output_port->name);
3820 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3821 m->bridge->name, m->name);
3827 if (cfg->output_vlan) {
3828 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3829 "output vlan; ignoring output vlan",
3830 m->bridge->name, m->name);
3832 } else if (cfg->output_vlan) {
3834 out_vlan = *cfg->output_vlan;
3836 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3837 m->bridge->name, m->name);
3842 /* Get all the ports, and drop duplicates and ports that don't exist. */
3843 shash_init(&src_ports);
3844 shash_init(&dst_ports);
3845 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3847 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3849 any_ports_specified = cfg->n_select_dst_port || cfg->n_select_dst_port;
3850 if (any_ports_specified
3851 && shash_is_empty(&src_ports) && shash_is_empty(&dst_ports)) {
3852 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3853 "selection ports exists", m->bridge->name, m->name);
3858 /* Get all the vlans, and drop duplicate and invalid vlans. */
3859 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3860 any_vlans_specified = cfg->n_select_vlan > 0;
3861 if (any_vlans_specified && !n_vlans) {
3862 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3863 "VLANs exists", m->bridge->name, m->name);
3868 /* Update mirror data. */
3869 if (!shash_equal_keys(&m->src_ports, &src_ports)
3870 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3871 || m->n_vlans != n_vlans
3872 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3873 || m->out_port != out_port
3874 || m->out_vlan != out_vlan) {
3875 bridge_flush(m->bridge);
3877 shash_swap(&m->src_ports, &src_ports);
3878 shash_swap(&m->dst_ports, &dst_ports);
3881 m->n_vlans = n_vlans;
3882 m->out_port = out_port;
3883 m->out_vlan = out_vlan;
3885 /* If no selection criteria have been given, mirror for all ports. */
3886 mirror_all_ports = !any_ports_specified && !any_vlans_specified;
3889 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3890 for (i = 0; i < m->bridge->n_ports; i++) {
3891 struct port *port = m->bridge->ports[i];
3893 if (mirror_all_ports
3894 || shash_find(&m->src_ports, port->name)
3897 ? port_trunks_any_mirrored_vlan(m, port)
3898 : vlan_is_mirrored(m, port->vlan)))) {
3899 port->src_mirrors |= mirror_bit;
3901 port->src_mirrors &= ~mirror_bit;
3904 if (mirror_all_ports || shash_find(&m->dst_ports, port->name)) {
3905 port->dst_mirrors |= mirror_bit;
3907 port->dst_mirrors &= ~mirror_bit;
3913 shash_destroy(&src_ports);
3914 shash_destroy(&dst_ports);