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? */
140 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
141 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
142 long long int bond_next_rebalance; /* Next rebalancing time. */
144 /* Port mirroring info. */
145 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
146 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
147 bool is_mirror_output_port; /* Does port mirroring send frames here? */
149 /* This member is only valid *during* bridge_reconfigure(). */
150 const struct ovsrec_port *cfg;
153 #define DP_MAX_PORTS 255
155 struct list node; /* Node in global list of bridges. */
156 char *name; /* User-specified arbitrary name. */
157 struct mac_learning *ml; /* MAC learning table. */
158 bool sent_config_request; /* Successfully sent config request? */
159 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
161 /* Support for remote controllers. */
162 char *controller; /* NULL if there is no remote controller;
163 * "discover" to do controller discovery;
164 * otherwise a vconn name. */
166 /* OpenFlow switch processing. */
167 struct ofproto *ofproto; /* OpenFlow switch. */
169 /* Description strings. */
170 char *mfr_desc; /* Manufacturer. */
171 char *hw_desc; /* Hardware. */
172 char *sw_desc; /* Software version. */
173 char *serial_desc; /* Serial number. */
174 char *dp_desc; /* Datapath description. */
176 /* Kernel datapath information. */
177 struct dpif *dpif; /* Datapath. */
178 struct port_array ifaces; /* Indexed by kernel datapath port number. */
182 size_t n_ports, allocated_ports;
185 bool has_bonded_ports;
190 /* Flow statistics gathering. */
191 time_t next_stats_request;
193 /* Port mirroring. */
194 struct mirror *mirrors[MAX_MIRRORS];
196 /* This member is only valid *during* bridge_reconfigure(). */
197 const struct ovsrec_bridge *cfg;
200 /* List of all bridges. */
201 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
203 /* Maximum number of datapaths. */
204 enum { DP_MAX = 256 };
206 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
207 static void bridge_destroy(struct bridge *);
208 static struct bridge *bridge_lookup(const char *name);
209 static unixctl_cb_func bridge_unixctl_dump_flows;
210 static int bridge_run_one(struct bridge *);
211 static const struct ovsrec_controller *bridge_get_controller(
212 const struct ovsrec_open_vswitch *ovs_cfg,
213 const struct bridge *br);
214 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
216 static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
218 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
219 static void bridge_fetch_dp_ifaces(struct bridge *);
220 static void bridge_flush(struct bridge *);
221 static void bridge_pick_local_hw_addr(struct bridge *,
222 uint8_t ea[ETH_ADDR_LEN],
223 struct iface **hw_addr_iface);
224 static uint64_t bridge_pick_datapath_id(struct bridge *,
225 const uint8_t bridge_ea[ETH_ADDR_LEN],
226 struct iface *hw_addr_iface);
227 static struct iface *bridge_get_local_iface(struct bridge *);
228 static uint64_t dpid_from_hash(const void *, size_t nbytes);
230 static unixctl_cb_func bridge_unixctl_fdb_show;
232 static void bond_init(void);
233 static void bond_run(struct bridge *);
234 static void bond_wait(struct bridge *);
235 static void bond_rebalance_port(struct port *);
236 static void bond_send_learning_packets(struct port *);
237 static void bond_enable_slave(struct iface *iface, bool enable);
239 static struct port *port_create(struct bridge *, const char *name);
240 static void port_reconfigure(struct port *, const struct ovsrec_port *);
241 static void port_destroy(struct port *);
242 static struct port *port_lookup(const struct bridge *, const char *name);
243 static struct iface *port_lookup_iface(const struct port *, const char *name);
244 static struct port *port_from_dp_ifidx(const struct bridge *,
246 static void port_update_bond_compat(struct port *);
247 static void port_update_vlan_compat(struct port *);
248 static void port_update_bonding(struct port *);
250 static struct mirror *mirror_create(struct bridge *, const char *name);
251 static void mirror_destroy(struct mirror *);
252 static void mirror_reconfigure(struct bridge *);
253 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
254 static bool vlan_is_mirrored(const struct mirror *, int vlan);
256 static struct iface *iface_create(struct port *port,
257 const struct ovsrec_interface *if_cfg);
258 static void iface_destroy(struct iface *);
259 static struct iface *iface_lookup(const struct bridge *, const char *name);
260 static struct iface *iface_from_dp_ifidx(const struct bridge *,
262 static bool iface_is_internal(const struct bridge *, const char *name);
263 static void iface_set_mac(struct iface *);
265 /* Hooks into ofproto processing. */
266 static struct ofhooks bridge_ofhooks;
268 /* Public functions. */
270 /* Adds the name of each interface used by a bridge, including local and
271 * internal ports, to 'svec'. */
273 bridge_get_ifaces(struct svec *svec)
275 struct bridge *br, *next;
278 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
279 for (i = 0; i < br->n_ports; i++) {
280 struct port *port = br->ports[i];
282 for (j = 0; j < port->n_ifaces; j++) {
283 struct iface *iface = port->ifaces[j];
284 if (iface->dp_ifidx < 0) {
285 VLOG_ERR("%s interface not in datapath %s, ignoring",
286 iface->name, dpif_name(br->dpif));
288 if (iface->dp_ifidx != ODPP_LOCAL) {
289 svec_add(svec, iface->name);
298 bridge_init(const struct ovsrec_open_vswitch *cfg)
300 struct svec bridge_names;
301 struct svec dpif_names, dpif_types;
304 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
306 svec_init(&bridge_names);
307 for (i = 0; i < cfg->n_bridges; i++) {
308 svec_add(&bridge_names, cfg->bridges[i]->name);
310 svec_sort(&bridge_names);
312 svec_init(&dpif_names);
313 svec_init(&dpif_types);
314 dp_enumerate_types(&dpif_types);
315 for (i = 0; i < dpif_types.n; i++) {
320 dp_enumerate_names(dpif_types.names[i], &dpif_names);
322 for (j = 0; j < dpif_names.n; j++) {
323 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
325 struct svec all_names;
328 svec_init(&all_names);
329 dpif_get_all_names(dpif, &all_names);
330 for (k = 0; k < all_names.n; k++) {
331 if (svec_contains(&bridge_names, all_names.names[k])) {
337 svec_destroy(&all_names);
342 svec_destroy(&bridge_names);
343 svec_destroy(&dpif_names);
344 svec_destroy(&dpif_types);
346 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
350 bridge_reconfigure(cfg);
355 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
357 /* XXX SSL should be configurable on a per-bridge basis. */
359 stream_ssl_set_private_key_file(ssl->private_key);
360 stream_ssl_set_certificate_file(ssl->certificate);
361 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
366 /* Attempt to create the network device 'iface_name' through the netdev
369 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
372 struct shash_node *node;
373 struct shash options;
377 shash_init(&options);
378 for (i = 0; i < iface_cfg->n_options; i++) {
379 shash_add(&options, iface_cfg->key_options[i],
380 xstrdup(iface_cfg->value_options[i]));
384 struct netdev_options netdev_options;
386 memset(&netdev_options, 0, sizeof netdev_options);
387 netdev_options.name = iface_cfg->name;
388 if (!strcmp(iface_cfg->type, "internal")) {
389 /* An "internal" config type maps to a netdev "system" type. */
390 netdev_options.type = "system";
392 netdev_options.type = iface_cfg->type;
394 netdev_options.args = &options;
395 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
396 netdev_options.may_create = true;
397 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
398 netdev_options.may_open = true;
401 error = netdev_open(&netdev_options, &iface->netdev);
404 netdev_get_carrier(iface->netdev, &iface->enabled);
406 } else if (iface->netdev) {
407 const char *netdev_type = netdev_get_type(iface->netdev);
408 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
409 ? iface_cfg->type : NULL;
411 /* An "internal" config type maps to a netdev "system" type. */
412 if (iface_type && !strcmp(iface_type, "internal")) {
413 iface_type = "system";
416 if (!iface_type || !strcmp(netdev_type, iface_type)) {
417 error = netdev_reconfigure(iface->netdev, &options);
419 VLOG_WARN("%s: attempting change device type from %s to %s",
420 iface_cfg->name, netdev_type, iface_type);
425 SHASH_FOR_EACH (node, &options) {
428 shash_destroy(&options);
434 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
436 return set_up_iface(iface_cfg, iface, false);
440 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
441 void *aux OVS_UNUSED)
443 if (!iface->netdev) {
444 int error = set_up_iface(iface->cfg, iface, true);
446 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
456 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
457 void *aux OVS_UNUSED)
459 if (iface->dp_ifidx >= 0) {
460 VLOG_DBG("%s has interface %s on port %d",
462 iface->name, iface->dp_ifidx);
465 VLOG_ERR("%s interface not in %s, dropping",
466 iface->name, dpif_name(br->dpif));
472 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
473 void *aux OVS_UNUSED)
475 /* Set policing attributes. */
476 netdev_set_policing(iface->netdev,
477 iface->cfg->ingress_policing_rate,
478 iface->cfg->ingress_policing_burst);
480 /* Set MAC address of internal interfaces other than the local
482 if (iface->dp_ifidx != ODPP_LOCAL
483 && iface_is_internal(br, iface->name)) {
484 iface_set_mac(iface);
490 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
491 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
492 * deletes from 'br' any ports that no longer have any interfaces. */
494 iterate_and_prune_ifaces(struct bridge *br,
495 bool (*cb)(struct bridge *, struct iface *,
501 for (i = 0; i < br->n_ports; ) {
502 struct port *port = br->ports[i];
503 for (j = 0; j < port->n_ifaces; ) {
504 struct iface *iface = port->ifaces[j];
505 if (cb(br, iface, aux)) {
508 iface_destroy(iface);
512 if (port->n_ifaces) {
515 VLOG_ERR("%s port has no interfaces, dropping", port->name);
522 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
524 struct ovsdb_idl_txn *txn;
525 struct shash old_br, new_br;
526 struct shash_node *node;
527 struct bridge *br, *next;
529 int sflow_bridge_number;
531 COVERAGE_INC(bridge_reconfigure);
533 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
535 /* Collect old and new bridges. */
538 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
539 shash_add(&old_br, br->name, br);
541 for (i = 0; i < ovs_cfg->n_bridges; i++) {
542 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
543 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
544 VLOG_WARN("more than one bridge named %s", br_cfg->name);
548 /* Get rid of deleted bridges and add new bridges. */
549 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
550 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
557 SHASH_FOR_EACH (node, &new_br) {
558 const char *br_name = node->name;
559 const struct ovsrec_bridge *br_cfg = node->data;
560 br = shash_find_data(&old_br, br_name);
562 /* If the bridge datapath type has changed, we need to tear it
563 * down and recreate. */
564 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
566 bridge_create(br_cfg);
569 bridge_create(br_cfg);
572 shash_destroy(&old_br);
573 shash_destroy(&new_br);
577 bridge_configure_ssl(ovs_cfg->ssl);
580 /* Reconfigure all bridges. */
581 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
582 bridge_reconfigure_one(ovs_cfg, br);
585 /* Add and delete ports on all datapaths.
587 * The kernel will reject any attempt to add a given port to a datapath if
588 * that port already belongs to a different datapath, so we must do all
589 * port deletions before any port additions. */
590 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
591 struct odp_port *dpif_ports;
593 struct shash want_ifaces;
595 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
596 bridge_get_all_ifaces(br, &want_ifaces);
597 for (i = 0; i < n_dpif_ports; i++) {
598 const struct odp_port *p = &dpif_ports[i];
599 if (!shash_find(&want_ifaces, p->devname)
600 && strcmp(p->devname, br->name)) {
601 int retval = dpif_port_del(br->dpif, p->port);
603 VLOG_ERR("failed to remove %s interface from %s: %s",
604 p->devname, dpif_name(br->dpif),
609 shash_destroy(&want_ifaces);
612 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
613 struct odp_port *dpif_ports;
615 struct shash cur_ifaces, want_ifaces;
616 struct shash_node *node;
618 /* Get the set of interfaces currently in this datapath. */
619 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
620 shash_init(&cur_ifaces);
621 for (i = 0; i < n_dpif_ports; i++) {
622 const char *name = dpif_ports[i].devname;
623 if (!shash_find(&cur_ifaces, name)) {
624 shash_add(&cur_ifaces, name, NULL);
629 /* Get the set of interfaces we want on this datapath. */
630 bridge_get_all_ifaces(br, &want_ifaces);
632 SHASH_FOR_EACH (node, &want_ifaces) {
633 const char *if_name = node->name;
634 struct iface *iface = node->data;
636 if (shash_find(&cur_ifaces, if_name)) {
637 /* Already exists, just reconfigure it. */
639 reconfigure_iface(iface->cfg, iface);
642 /* Need to add to datapath. */
646 /* Add to datapath. */
647 internal = iface_is_internal(br, if_name);
648 error = dpif_port_add(br->dpif, if_name,
649 internal ? ODP_PORT_INTERNAL : 0, NULL);
650 if (error == EFBIG) {
651 VLOG_ERR("ran out of valid port numbers on %s",
652 dpif_name(br->dpif));
655 VLOG_ERR("failed to add %s interface to %s: %s",
656 if_name, dpif_name(br->dpif), strerror(error));
660 shash_destroy(&cur_ifaces);
661 shash_destroy(&want_ifaces);
663 sflow_bridge_number = 0;
664 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
667 struct iface *local_iface;
668 struct iface *hw_addr_iface;
671 bridge_fetch_dp_ifaces(br);
673 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
674 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
676 /* Pick local port hardware address, datapath ID. */
677 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
678 local_iface = bridge_get_local_iface(br);
680 int error = netdev_set_etheraddr(local_iface->netdev, ea);
682 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
683 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
684 "Ethernet address: %s",
685 br->name, strerror(error));
689 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
690 ofproto_set_datapath_id(br->ofproto, dpid);
692 dpid_string = xasprintf("%012"PRIx64, dpid);
693 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
696 /* Set NetFlow configuration on this bridge. */
697 if (br->cfg->netflow) {
698 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
699 struct netflow_options opts;
701 memset(&opts, 0, sizeof opts);
703 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
704 if (nf_cfg->engine_type) {
705 opts.engine_type = *nf_cfg->engine_type;
707 if (nf_cfg->engine_id) {
708 opts.engine_id = *nf_cfg->engine_id;
711 opts.active_timeout = nf_cfg->active_timeout;
712 if (!opts.active_timeout) {
713 opts.active_timeout = -1;
714 } else if (opts.active_timeout < 0) {
715 VLOG_WARN("bridge %s: active timeout interval set to negative "
716 "value, using default instead (%d seconds)", br->name,
717 NF_ACTIVE_TIMEOUT_DEFAULT);
718 opts.active_timeout = -1;
721 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
722 if (opts.add_id_to_iface) {
723 if (opts.engine_id > 0x7f) {
724 VLOG_WARN("bridge %s: netflow port mangling may conflict "
725 "with another vswitch, choose an engine id less "
726 "than 128", br->name);
728 if (br->n_ports > 508) {
729 VLOG_WARN("bridge %s: netflow port mangling will conflict "
730 "with another port when more than 508 ports are "
735 opts.collectors.n = nf_cfg->n_targets;
736 opts.collectors.names = nf_cfg->targets;
737 if (ofproto_set_netflow(br->ofproto, &opts)) {
738 VLOG_ERR("bridge %s: problem setting netflow collectors",
742 ofproto_set_netflow(br->ofproto, NULL);
745 /* Set sFlow configuration on this bridge. */
746 if (br->cfg->sflow) {
747 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
748 const struct ovsrec_controller *ctrl;
749 struct ofproto_sflow_options oso;
751 memset(&oso, 0, sizeof oso);
753 oso.targets.n = sflow_cfg->n_targets;
754 oso.targets.names = sflow_cfg->targets;
756 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
757 if (sflow_cfg->sampling) {
758 oso.sampling_rate = *sflow_cfg->sampling;
761 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
762 if (sflow_cfg->polling) {
763 oso.polling_interval = *sflow_cfg->polling;
766 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
767 if (sflow_cfg->header) {
768 oso.header_len = *sflow_cfg->header;
771 oso.sub_id = sflow_bridge_number++;
772 oso.agent_device = sflow_cfg->agent;
774 ctrl = bridge_get_controller(ovs_cfg, br);
775 oso.control_ip = ctrl ? ctrl->local_ip : NULL;
776 ofproto_set_sflow(br->ofproto, &oso);
778 svec_destroy(&oso.targets);
780 ofproto_set_sflow(br->ofproto, NULL);
783 /* Update the controller and related settings. It would be more
784 * straightforward to call this from bridge_reconfigure_one(), but we
785 * can't do it there for two reasons. First, and most importantly, at
786 * that point we don't know the dp_ifidx of any interfaces that have
787 * been added to the bridge (because we haven't actually added them to
788 * the datapath). Second, at that point we haven't set the datapath ID
789 * yet; when a controller is configured, resetting the datapath ID will
790 * immediately disconnect from the controller, so it's better to set
791 * the datapath ID before the controller. */
792 bridge_reconfigure_controller(ovs_cfg, br);
794 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
795 for (i = 0; i < br->n_ports; i++) {
796 struct port *port = br->ports[i];
798 port_update_vlan_compat(port);
799 port_update_bonding(port);
802 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
803 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
806 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
808 ovsdb_idl_txn_commit(txn);
809 ovsdb_idl_txn_destroy(txn); /* XXX */
813 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
817 for (i = 0; i < n; i++) {
818 if (!strcmp(keys[i], key)) {
826 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
828 return get_ovsrec_key_value(key,
829 br_cfg->key_other_config,
830 br_cfg->value_other_config,
831 br_cfg->n_other_config);
835 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
836 struct iface **hw_addr_iface)
842 *hw_addr_iface = NULL;
844 /* Did the user request a particular MAC? */
845 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
846 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
847 if (eth_addr_is_multicast(ea)) {
848 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
849 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
850 } else if (eth_addr_is_zero(ea)) {
851 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
857 /* Otherwise choose the minimum non-local MAC address among all of the
859 memset(ea, 0xff, sizeof ea);
860 for (i = 0; i < br->n_ports; i++) {
861 struct port *port = br->ports[i];
862 uint8_t iface_ea[ETH_ADDR_LEN];
865 /* Mirror output ports don't participate. */
866 if (port->is_mirror_output_port) {
870 /* Choose the MAC address to represent the port. */
871 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
872 /* Find the interface with this Ethernet address (if any) so that
873 * we can provide the correct devname to the caller. */
875 for (j = 0; j < port->n_ifaces; j++) {
876 struct iface *candidate = port->ifaces[j];
877 uint8_t candidate_ea[ETH_ADDR_LEN];
878 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
879 && eth_addr_equals(iface_ea, candidate_ea)) {
884 /* Choose the interface whose MAC address will represent the port.
885 * The Linux kernel bonding code always chooses the MAC address of
886 * the first slave added to a bond, and the Fedora networking
887 * scripts always add slaves to a bond in alphabetical order, so
888 * for compatibility we choose the interface with the name that is
889 * first in alphabetical order. */
890 iface = port->ifaces[0];
891 for (j = 1; j < port->n_ifaces; j++) {
892 struct iface *candidate = port->ifaces[j];
893 if (strcmp(candidate->name, iface->name) < 0) {
898 /* The local port doesn't count (since we're trying to choose its
899 * MAC address anyway). */
900 if (iface->dp_ifidx == ODPP_LOCAL) {
905 error = netdev_get_etheraddr(iface->netdev, iface_ea);
907 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
908 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
909 iface->name, strerror(error));
914 /* Compare against our current choice. */
915 if (!eth_addr_is_multicast(iface_ea) &&
916 !eth_addr_is_local(iface_ea) &&
917 !eth_addr_is_reserved(iface_ea) &&
918 !eth_addr_is_zero(iface_ea) &&
919 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
921 memcpy(ea, iface_ea, ETH_ADDR_LEN);
922 *hw_addr_iface = iface;
925 if (eth_addr_is_multicast(ea)) {
926 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
927 *hw_addr_iface = NULL;
928 VLOG_WARN("bridge %s: using default bridge Ethernet "
929 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
931 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
932 br->name, ETH_ADDR_ARGS(ea));
936 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
937 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
938 * an interface on 'br', then that interface must be passed in as
939 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
940 * 'hw_addr_iface' must be passed in as a null pointer. */
942 bridge_pick_datapath_id(struct bridge *br,
943 const uint8_t bridge_ea[ETH_ADDR_LEN],
944 struct iface *hw_addr_iface)
947 * The procedure for choosing a bridge MAC address will, in the most
948 * ordinary case, also choose a unique MAC that we can use as a datapath
949 * ID. In some special cases, though, multiple bridges will end up with
950 * the same MAC address. This is OK for the bridges, but it will confuse
951 * the OpenFlow controller, because each datapath needs a unique datapath
954 * Datapath IDs must be unique. It is also very desirable that they be
955 * stable from one run to the next, so that policy set on a datapath
958 const char *datapath_id;
961 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
962 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
968 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
970 * A bridge whose MAC address is taken from a VLAN network device
971 * (that is, a network device created with vconfig(8) or similar
972 * tool) will have the same MAC address as a bridge on the VLAN
973 * device's physical network device.
975 * Handle this case by hashing the physical network device MAC
976 * along with the VLAN identifier.
978 uint8_t buf[ETH_ADDR_LEN + 2];
979 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
980 buf[ETH_ADDR_LEN] = vlan >> 8;
981 buf[ETH_ADDR_LEN + 1] = vlan;
982 return dpid_from_hash(buf, sizeof buf);
985 * Assume that this bridge's MAC address is unique, since it
986 * doesn't fit any of the cases we handle specially.
991 * A purely internal bridge, that is, one that has no non-virtual
992 * network devices on it at all, is more difficult because it has no
993 * natural unique identifier at all.
995 * When the host is a XenServer, we handle this case by hashing the
996 * host's UUID with the name of the bridge. Names of bridges are
997 * persistent across XenServer reboots, although they can be reused if
998 * an internal network is destroyed and then a new one is later
999 * created, so this is fairly effective.
1001 * When the host is not a XenServer, we punt by using a random MAC
1002 * address on each run.
1004 const char *host_uuid = xenserver_get_host_uuid();
1006 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1007 dpid = dpid_from_hash(combined, strlen(combined));
1013 return eth_addr_to_uint64(bridge_ea);
1017 dpid_from_hash(const void *data, size_t n)
1019 uint8_t hash[SHA1_DIGEST_SIZE];
1021 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1022 sha1_bytes(data, n, hash);
1023 eth_addr_mark_random(hash);
1024 return eth_addr_to_uint64(hash);
1030 struct bridge *br, *next;
1034 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1035 int error = bridge_run_one(br);
1037 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1038 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1039 "forcing reconfiguration", br->name);
1053 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1054 ofproto_wait(br->ofproto);
1055 if (br->controller) {
1059 mac_learning_wait(br->ml);
1064 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1065 * configuration changes. */
1067 bridge_flush(struct bridge *br)
1069 COVERAGE_INC(bridge_flush);
1071 mac_learning_flush(br->ml);
1074 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1075 * such interface. */
1076 static struct iface *
1077 bridge_get_local_iface(struct bridge *br)
1081 for (i = 0; i < br->n_ports; i++) {
1082 struct port *port = br->ports[i];
1083 for (j = 0; j < port->n_ifaces; j++) {
1084 struct iface *iface = port->ifaces[j];
1085 if (iface->dp_ifidx == ODPP_LOCAL) {
1094 /* Bridge unixctl user interface functions. */
1096 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1097 const char *args, void *aux OVS_UNUSED)
1099 struct ds ds = DS_EMPTY_INITIALIZER;
1100 const struct bridge *br;
1101 const struct mac_entry *e;
1103 br = bridge_lookup(args);
1105 unixctl_command_reply(conn, 501, "no such bridge");
1109 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1110 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1111 if (e->port < 0 || e->port >= br->n_ports) {
1114 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1115 br->ports[e->port]->ifaces[0]->dp_ifidx,
1116 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1118 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1122 /* Bridge reconfiguration functions. */
1123 static struct bridge *
1124 bridge_create(const struct ovsrec_bridge *br_cfg)
1129 assert(!bridge_lookup(br_cfg->name));
1130 br = xzalloc(sizeof *br);
1132 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1138 dpif_flow_flush(br->dpif);
1140 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1143 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1145 dpif_delete(br->dpif);
1146 dpif_close(br->dpif);
1151 br->name = xstrdup(br_cfg->name);
1153 br->ml = mac_learning_create();
1154 br->sent_config_request = false;
1155 eth_addr_nicira_random(br->default_ea);
1157 port_array_init(&br->ifaces);
1161 list_push_back(&all_bridges, &br->node);
1163 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1169 bridge_destroy(struct bridge *br)
1174 while (br->n_ports > 0) {
1175 port_destroy(br->ports[br->n_ports - 1]);
1177 list_remove(&br->node);
1178 error = dpif_delete(br->dpif);
1179 if (error && error != ENOENT) {
1180 VLOG_ERR("failed to delete %s: %s",
1181 dpif_name(br->dpif), strerror(error));
1183 dpif_close(br->dpif);
1184 ofproto_destroy(br->ofproto);
1185 free(br->controller);
1186 mac_learning_destroy(br->ml);
1187 port_array_destroy(&br->ifaces);
1194 static struct bridge *
1195 bridge_lookup(const char *name)
1199 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1200 if (!strcmp(br->name, name)) {
1208 bridge_exists(const char *name)
1210 return bridge_lookup(name) ? true : false;
1214 bridge_get_datapathid(const char *name)
1216 struct bridge *br = bridge_lookup(name);
1217 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1220 /* Handle requests for a listing of all flows known by the OpenFlow
1221 * stack, including those normally hidden. */
1223 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1224 const char *args, void *aux OVS_UNUSED)
1229 br = bridge_lookup(args);
1231 unixctl_command_reply(conn, 501, "Unknown bridge");
1236 ofproto_get_all_flows(br->ofproto, &results);
1238 unixctl_command_reply(conn, 200, ds_cstr(&results));
1239 ds_destroy(&results);
1243 bridge_run_one(struct bridge *br)
1247 error = ofproto_run1(br->ofproto);
1252 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1255 error = ofproto_run2(br->ofproto, br->flush);
1261 static const struct ovsrec_controller *
1262 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1263 const struct bridge *br)
1265 const struct ovsrec_controller *controller;
1267 controller = (br->cfg->controller ? br->cfg->controller
1268 : ovs_cfg->controller ? ovs_cfg->controller
1271 if (controller && !strcmp(controller->target, "none")) {
1279 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1281 struct svec *ifaces = ifaces_;
1282 if (!svec_contains(ifaces, iface->name)) {
1283 svec_add(ifaces, iface->name);
1287 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1289 br->name, iface->name, iface->port->name);
1295 bridge_update_desc(struct bridge *br OVS_UNUSED)
1298 bool changed = false;
1301 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1302 if (desc != br->mfr_desc) {
1305 br->mfr_desc = xstrdup(desc);
1307 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1312 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1313 if (desc != br->hw_desc) {
1316 br->hw_desc = xstrdup(desc);
1318 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1323 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1324 if (desc != br->sw_desc) {
1327 br->sw_desc = xstrdup(desc);
1329 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1334 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1335 if (desc != br->serial_desc) {
1336 free(br->serial_desc);
1338 br->serial_desc = xstrdup(desc);
1340 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1345 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1346 if (desc != br->dp_desc) {
1349 br->dp_desc = xstrdup(desc);
1351 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1357 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1358 br->sw_desc, br->serial_desc, br->dp_desc);
1364 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1367 struct shash old_ports, new_ports;
1369 struct svec listeners, old_listeners;
1370 struct svec snoops, old_snoops;
1371 struct shash_node *node;
1374 /* Collect old ports. */
1375 shash_init(&old_ports);
1376 for (i = 0; i < br->n_ports; i++) {
1377 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1380 /* Collect new ports. */
1381 shash_init(&new_ports);
1382 for (i = 0; i < br->cfg->n_ports; i++) {
1383 const char *name = br->cfg->ports[i]->name;
1384 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1385 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1390 /* If we have a controller, then we need a local port. Complain if the
1391 * user didn't specify one.
1393 * XXX perhaps we should synthesize a port ourselves in this case. */
1394 if (bridge_get_controller(ovs_cfg, br)) {
1395 char local_name[IF_NAMESIZE];
1398 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1399 local_name, sizeof local_name);
1400 if (!error && !shash_find(&new_ports, local_name)) {
1401 VLOG_WARN("bridge %s: controller specified but no local port "
1402 "(port named %s) defined",
1403 br->name, local_name);
1407 /* Get rid of deleted ports and add new ports. */
1408 SHASH_FOR_EACH (node, &old_ports) {
1409 if (!shash_find(&new_ports, node->name)) {
1410 port_destroy(node->data);
1413 SHASH_FOR_EACH (node, &new_ports) {
1414 struct port *port = shash_find_data(&old_ports, node->name);
1416 port = port_create(br, node->name);
1418 port_reconfigure(port, node->data);
1420 shash_destroy(&old_ports);
1421 shash_destroy(&new_ports);
1423 /* Check and delete duplicate interfaces. */
1425 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1426 svec_destroy(&ifaces);
1428 /* Delete all flows if we're switching from connected to standalone or vice
1429 * versa. (XXX Should we delete all flows if we are switching from one
1430 * controller to another?) */
1433 /* Configure OpenFlow management listeners. */
1434 svec_init(&listeners);
1435 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1437 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1438 ovs_rundir, br->name));
1439 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1440 svec_clear(&listeners);
1442 svec_sort_unique(&listeners);
1444 svec_init(&old_listeners);
1445 ofproto_get_listeners(br->ofproto, &old_listeners);
1446 svec_sort_unique(&old_listeners);
1448 if (!svec_equal(&listeners, &old_listeners)) {
1449 ofproto_set_listeners(br->ofproto, &listeners);
1451 svec_destroy(&listeners);
1452 svec_destroy(&old_listeners);
1454 /* Configure OpenFlow controller connection snooping. */
1456 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1458 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1459 ovs_rundir, br->name));
1460 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1461 svec_clear(&snoops);
1463 svec_sort_unique(&snoops);
1465 svec_init(&old_snoops);
1466 ofproto_get_snoops(br->ofproto, &old_snoops);
1467 svec_sort_unique(&old_snoops);
1469 if (!svec_equal(&snoops, &old_snoops)) {
1470 ofproto_set_snoops(br->ofproto, &snoops);
1472 svec_destroy(&snoops);
1473 svec_destroy(&old_snoops);
1475 /* Default listener. */
1476 svec_init(&listeners);
1477 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1478 ovs_rundir, br->name));
1479 svec_init(&old_listeners);
1480 ofproto_get_listeners(br->ofproto, &old_listeners);
1481 if (!svec_equal(&listeners, &old_listeners)) {
1482 ofproto_set_listeners(br->ofproto, &listeners);
1484 svec_destroy(&listeners);
1485 svec_destroy(&old_listeners);
1487 /* Default snoop. */
1489 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1490 ovs_rundir, br->name));
1491 svec_init(&old_snoops);
1492 ofproto_get_snoops(br->ofproto, &old_snoops);
1493 if (!svec_equal(&snoops, &old_snoops)) {
1494 ofproto_set_snoops(br->ofproto, &snoops);
1496 svec_destroy(&snoops);
1497 svec_destroy(&old_snoops);
1500 mirror_reconfigure(br);
1502 bridge_update_desc(br);
1506 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1509 const struct ovsrec_controller *c;
1511 c = bridge_get_controller(ovs_cfg, br);
1512 if ((br->controller != NULL) != (c != NULL)) {
1513 ofproto_flush_flows(br->ofproto);
1515 free(br->controller);
1516 br->controller = c ? xstrdup(c->target) : NULL;
1519 int max_backoff, probe;
1520 int rate_limit, burst_limit;
1522 if (!strcmp(c->target, "discover")) {
1523 ofproto_set_discovery(br->ofproto, true,
1524 c->discover_accept_regex,
1525 c->discover_update_resolv_conf);
1527 struct iface *local_iface;
1531 in_band = (!c->connection_mode
1532 || !strcmp(c->connection_mode, "out-of-band"));
1533 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1534 ofproto_set_in_band(br->ofproto, in_band);
1536 local_iface = bridge_get_local_iface(br);
1537 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1538 struct netdev *netdev = local_iface->netdev;
1539 struct in_addr mask, gateway;
1541 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1544 if (!c->local_gateway
1545 || !inet_aton(c->local_gateway, &gateway)) {
1549 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1551 mask.s_addr = guess_netmask(ip.s_addr);
1553 if (!netdev_set_in4(netdev, ip, mask)) {
1554 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1556 br->name, IP_ARGS(&ip.s_addr),
1557 IP_ARGS(&mask.s_addr));
1560 if (gateway.s_addr) {
1561 if (!netdev_add_router(netdev, gateway)) {
1562 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1563 br->name, IP_ARGS(&gateway.s_addr));
1569 ofproto_set_failure(br->ofproto,
1571 || !strcmp(c->fail_mode, "standalone")
1572 || !strcmp(c->fail_mode, "open")));
1574 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1575 ofproto_set_probe_interval(br->ofproto, probe);
1577 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1578 ofproto_set_max_backoff(br->ofproto, max_backoff);
1580 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1581 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1582 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1584 union ofp_action action;
1587 /* Set up a flow that matches every packet and directs them to
1588 * OFPP_NORMAL (which goes to us). */
1589 memset(&action, 0, sizeof action);
1590 action.type = htons(OFPAT_OUTPUT);
1591 action.output.len = htons(sizeof action);
1592 action.output.port = htons(OFPP_NORMAL);
1593 memset(&flow, 0, sizeof flow);
1594 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1596 ofproto_set_in_band(br->ofproto, false);
1597 ofproto_set_max_backoff(br->ofproto, 1);
1598 ofproto_set_probe_interval(br->ofproto, 5);
1599 ofproto_set_failure(br->ofproto, false);
1602 ofproto_set_controller(br->ofproto, br->controller);
1606 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1611 for (i = 0; i < br->n_ports; i++) {
1612 struct port *port = br->ports[i];
1613 for (j = 0; j < port->n_ifaces; j++) {
1614 struct iface *iface = port->ifaces[j];
1615 shash_add_once(ifaces, iface->name, iface);
1617 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1618 shash_add_once(ifaces, port->name, NULL);
1623 /* For robustness, in case the administrator moves around datapath ports behind
1624 * our back, we re-check all the datapath port numbers here.
1626 * This function will set the 'dp_ifidx' members of interfaces that have
1627 * disappeared to -1, so only call this function from a context where those
1628 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1629 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1630 * datapath, which doesn't support UINT16_MAX+1 ports. */
1632 bridge_fetch_dp_ifaces(struct bridge *br)
1634 struct odp_port *dpif_ports;
1635 size_t n_dpif_ports;
1638 /* Reset all interface numbers. */
1639 for (i = 0; i < br->n_ports; i++) {
1640 struct port *port = br->ports[i];
1641 for (j = 0; j < port->n_ifaces; j++) {
1642 struct iface *iface = port->ifaces[j];
1643 iface->dp_ifidx = -1;
1646 port_array_clear(&br->ifaces);
1648 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1649 for (i = 0; i < n_dpif_ports; i++) {
1650 struct odp_port *p = &dpif_ports[i];
1651 struct iface *iface = iface_lookup(br, p->devname);
1653 if (iface->dp_ifidx >= 0) {
1654 VLOG_WARN("%s reported interface %s twice",
1655 dpif_name(br->dpif), p->devname);
1656 } else if (iface_from_dp_ifidx(br, p->port)) {
1657 VLOG_WARN("%s reported interface %"PRIu16" twice",
1658 dpif_name(br->dpif), p->port);
1660 port_array_set(&br->ifaces, p->port, iface);
1661 iface->dp_ifidx = p->port;
1665 int64_t ofport = (iface->dp_ifidx >= 0
1666 ? odp_port_to_ofp_port(iface->dp_ifidx)
1668 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1675 /* Bridge packet processing functions. */
1678 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1680 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1683 static struct bond_entry *
1684 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1686 return &port->bond_hash[bond_hash(mac)];
1690 bond_choose_iface(const struct port *port)
1692 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1693 size_t i, best_down_slave = -1;
1694 long long next_delay_expiration = LLONG_MAX;
1696 for (i = 0; i < port->n_ifaces; i++) {
1697 struct iface *iface = port->ifaces[i];
1699 if (iface->enabled) {
1701 } else if (iface->delay_expires < next_delay_expiration) {
1702 best_down_slave = i;
1703 next_delay_expiration = iface->delay_expires;
1707 if (best_down_slave != -1) {
1708 struct iface *iface = port->ifaces[best_down_slave];
1710 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1711 "since no other interface is up", iface->name,
1712 iface->delay_expires - time_msec());
1713 bond_enable_slave(iface, true);
1716 return best_down_slave;
1720 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1721 uint16_t *dp_ifidx, tag_type *tags)
1723 struct iface *iface;
1725 assert(port->n_ifaces);
1726 if (port->n_ifaces == 1) {
1727 iface = port->ifaces[0];
1729 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1730 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1731 || !port->ifaces[e->iface_idx]->enabled) {
1732 /* XXX select interface properly. The current interface selection
1733 * is only good for testing the rebalancing code. */
1734 e->iface_idx = bond_choose_iface(port);
1735 if (e->iface_idx < 0) {
1736 *tags |= port->no_ifaces_tag;
1739 e->iface_tag = tag_create_random();
1740 ((struct port *) port)->bond_compat_is_stale = true;
1742 *tags |= e->iface_tag;
1743 iface = port->ifaces[e->iface_idx];
1745 *dp_ifidx = iface->dp_ifidx;
1746 *tags |= iface->tag; /* Currently only used for bonding. */
1751 bond_link_status_update(struct iface *iface, bool carrier)
1753 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1754 struct port *port = iface->port;
1756 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1757 /* Nothing to do. */
1760 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1761 iface->name, carrier ? "detected" : "dropped");
1762 if (carrier == iface->enabled) {
1763 iface->delay_expires = LLONG_MAX;
1764 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1765 iface->name, carrier ? "disabled" : "enabled");
1766 } else if (carrier && port->active_iface < 0) {
1767 bond_enable_slave(iface, true);
1768 if (port->updelay) {
1769 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1770 "other interface is up", iface->name, port->updelay);
1773 int delay = carrier ? port->updelay : port->downdelay;
1774 iface->delay_expires = time_msec() + delay;
1777 "interface %s: will be %s if it stays %s for %d ms",
1779 carrier ? "enabled" : "disabled",
1780 carrier ? "up" : "down",
1787 bond_choose_active_iface(struct port *port)
1789 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1791 port->active_iface = bond_choose_iface(port);
1792 port->active_iface_tag = tag_create_random();
1793 if (port->active_iface >= 0) {
1794 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1795 port->name, port->ifaces[port->active_iface]->name);
1797 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1803 bond_enable_slave(struct iface *iface, bool enable)
1805 struct port *port = iface->port;
1806 struct bridge *br = port->bridge;
1808 /* This acts as a recursion check. If the act of disabling a slave
1809 * causes a different slave to be enabled, the flag will allow us to
1810 * skip redundant work when we reenter this function. It must be
1811 * cleared on exit to keep things safe with multiple bonds. */
1812 static bool moving_active_iface = false;
1814 iface->delay_expires = LLONG_MAX;
1815 if (enable == iface->enabled) {
1819 iface->enabled = enable;
1820 if (!iface->enabled) {
1821 VLOG_WARN("interface %s: disabled", iface->name);
1822 ofproto_revalidate(br->ofproto, iface->tag);
1823 if (iface->port_ifidx == port->active_iface) {
1824 ofproto_revalidate(br->ofproto,
1825 port->active_iface_tag);
1827 /* Disabling a slave can lead to another slave being immediately
1828 * enabled if there will be no active slaves but one is waiting
1829 * on an updelay. In this case we do not need to run most of the
1830 * code for the newly enabled slave since there was no period
1831 * without an active slave and it is redundant with the disabling
1833 moving_active_iface = true;
1834 bond_choose_active_iface(port);
1836 bond_send_learning_packets(port);
1838 VLOG_WARN("interface %s: enabled", iface->name);
1839 if (port->active_iface < 0 && !moving_active_iface) {
1840 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1841 bond_choose_active_iface(port);
1842 bond_send_learning_packets(port);
1844 iface->tag = tag_create_random();
1847 moving_active_iface = false;
1848 port->bond_compat_is_stale = true;
1851 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1852 * bond interface. */
1854 bond_update_fake_iface_stats(struct port *port)
1856 struct netdev_stats bond_stats;
1857 struct netdev *bond_dev;
1860 memset(&bond_stats, 0, sizeof bond_stats);
1862 for (i = 0; i < port->n_ifaces; i++) {
1863 struct netdev_stats slave_stats;
1865 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1866 bond_stats.rx_packets += slave_stats.rx_packets;
1867 bond_stats.rx_bytes += slave_stats.rx_bytes;
1868 bond_stats.tx_packets += slave_stats.tx_packets;
1869 bond_stats.tx_bytes += slave_stats.tx_bytes;
1873 if (!netdev_open_default(port->name, &bond_dev)) {
1874 netdev_set_stats(bond_dev, &bond_stats);
1875 netdev_close(bond_dev);
1880 bond_run(struct bridge *br)
1884 for (i = 0; i < br->n_ports; i++) {
1885 struct port *port = br->ports[i];
1887 if (port->n_ifaces >= 2) {
1888 for (j = 0; j < port->n_ifaces; j++) {
1889 struct iface *iface = port->ifaces[j];
1890 if (time_msec() >= iface->delay_expires) {
1891 bond_enable_slave(iface, !iface->enabled);
1895 if (port->bond_fake_iface
1896 && time_msec() >= port->bond_next_fake_iface_update) {
1897 bond_update_fake_iface_stats(port);
1898 port->bond_next_fake_iface_update = time_msec() + 1000;
1902 if (port->bond_compat_is_stale) {
1903 port->bond_compat_is_stale = false;
1904 port_update_bond_compat(port);
1910 bond_wait(struct bridge *br)
1914 for (i = 0; i < br->n_ports; i++) {
1915 struct port *port = br->ports[i];
1916 if (port->n_ifaces < 2) {
1919 for (j = 0; j < port->n_ifaces; j++) {
1920 struct iface *iface = port->ifaces[j];
1921 if (iface->delay_expires != LLONG_MAX) {
1922 poll_timer_wait(iface->delay_expires - time_msec());
1925 if (port->bond_fake_iface) {
1926 poll_timer_wait(port->bond_next_fake_iface_update - time_msec());
1932 set_dst(struct dst *p, const flow_t *flow,
1933 const struct port *in_port, const struct port *out_port,
1936 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1937 : in_port->vlan >= 0 ? in_port->vlan
1938 : ntohs(flow->dl_vlan));
1939 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1943 swap_dst(struct dst *p, struct dst *q)
1945 struct dst tmp = *p;
1950 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1951 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1952 * that we push to the datapath. We could in fact fully sort the array by
1953 * vlan, but in most cases there are at most two different vlan tags so that's
1954 * possibly overkill.) */
1956 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1958 struct dst *first = dsts;
1959 struct dst *last = dsts + n_dsts;
1961 while (first != last) {
1963 * - All dsts < first have vlan == 'vlan'.
1964 * - All dsts >= last have vlan != 'vlan'.
1965 * - first < last. */
1966 while (first->vlan == vlan) {
1967 if (++first == last) {
1972 /* Same invariants, plus one additional:
1973 * - first->vlan != vlan.
1975 while (last[-1].vlan != vlan) {
1976 if (--last == first) {
1981 /* Same invariants, plus one additional:
1982 * - last[-1].vlan == vlan.*/
1983 swap_dst(first++, --last);
1988 mirror_mask_ffs(mirror_mask_t mask)
1990 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1995 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1996 const struct dst *test)
1999 for (i = 0; i < n_dsts; i++) {
2000 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2008 port_trunks_vlan(const struct port *port, uint16_t vlan)
2010 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
2014 port_includes_vlan(const struct port *port, uint16_t vlan)
2016 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2020 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2021 const struct port *in_port, const struct port *out_port,
2022 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2024 mirror_mask_t mirrors = in_port->src_mirrors;
2025 struct dst *dst = dsts;
2028 if (out_port == FLOOD_PORT) {
2029 /* XXX use ODP_FLOOD if no vlans or bonding. */
2030 /* XXX even better, define each VLAN as a datapath port group */
2031 for (i = 0; i < br->n_ports; i++) {
2032 struct port *port = br->ports[i];
2033 if (port != in_port && port_includes_vlan(port, vlan)
2034 && !port->is_mirror_output_port
2035 && set_dst(dst, flow, in_port, port, tags)) {
2036 mirrors |= port->dst_mirrors;
2040 *nf_output_iface = NF_OUT_FLOOD;
2041 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2042 *nf_output_iface = dst->dp_ifidx;
2043 mirrors |= out_port->dst_mirrors;
2048 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2049 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2051 if (set_dst(dst, flow, in_port, m->out_port, tags)
2052 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2056 for (i = 0; i < br->n_ports; i++) {
2057 struct port *port = br->ports[i];
2058 if (port_includes_vlan(port, m->out_vlan)
2059 && set_dst(dst, flow, in_port, port, tags))
2063 if (port->vlan < 0) {
2064 dst->vlan = m->out_vlan;
2066 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2070 /* Use the vlan tag on the original flow instead of
2071 * the one passed in the vlan parameter. This ensures
2072 * that we compare the vlan from before any implicit
2073 * tagging tags place. This is necessary because
2074 * dst->vlan is the final vlan, after removing implicit
2076 flow_vlan = ntohs(flow->dl_vlan);
2077 if (flow_vlan == 0) {
2078 flow_vlan = OFP_VLAN_NONE;
2080 if (port == in_port && dst->vlan == flow_vlan) {
2081 /* Don't send out input port on same VLAN. */
2089 mirrors &= mirrors - 1;
2092 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2096 static void OVS_UNUSED
2097 print_dsts(const struct dst *dsts, size_t n)
2099 for (; n--; dsts++) {
2100 printf(">p%"PRIu16, dsts->dp_ifidx);
2101 if (dsts->vlan != OFP_VLAN_NONE) {
2102 printf("v%"PRIu16, dsts->vlan);
2108 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2109 const struct port *in_port, const struct port *out_port,
2110 tag_type *tags, struct odp_actions *actions,
2111 uint16_t *nf_output_iface)
2113 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2115 const struct dst *p;
2118 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2121 cur_vlan = ntohs(flow->dl_vlan);
2122 for (p = dsts; p < &dsts[n_dsts]; p++) {
2123 union odp_action *a;
2124 if (p->vlan != cur_vlan) {
2125 if (p->vlan == OFP_VLAN_NONE) {
2126 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2128 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2129 a->vlan_vid.vlan_vid = htons(p->vlan);
2133 a = odp_actions_add(actions, ODPAT_OUTPUT);
2134 a->output.port = p->dp_ifidx;
2138 /* Returns the effective vlan of a packet, taking into account both the
2139 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2140 * the packet is untagged and -1 indicates it has an invalid header and
2141 * should be dropped. */
2142 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2143 struct port *in_port, bool have_packet)
2145 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2146 * belongs to VLAN 0, so we should treat both cases identically. (In the
2147 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2148 * presumably to allow a priority to be specified. In the latter case, the
2149 * packet does not have any 802.1Q header.) */
2150 int vlan = ntohs(flow->dl_vlan);
2151 if (vlan == OFP_VLAN_NONE) {
2154 if (in_port->vlan >= 0) {
2156 /* XXX support double tagging? */
2158 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2159 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2160 "packet received on port %s configured with "
2161 "implicit VLAN %"PRIu16,
2162 br->name, ntohs(flow->dl_vlan),
2163 in_port->name, in_port->vlan);
2167 vlan = in_port->vlan;
2169 if (!port_includes_vlan(in_port, vlan)) {
2171 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2172 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2173 "packet received on port %s not configured for "
2175 br->name, vlan, in_port->name, vlan);
2185 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2186 struct port *in_port)
2188 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2189 vlan, in_port->port_idx);
2191 /* The log messages here could actually be useful in debugging,
2192 * so keep the rate limit relatively high. */
2193 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2195 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2196 "on port %s in VLAN %d",
2197 br->name, ETH_ADDR_ARGS(flow->dl_src),
2198 in_port->name, vlan);
2199 ofproto_revalidate(br->ofproto, rev_tag);
2204 is_bcast_arp_reply(const flow_t *flow)
2206 return (flow->dl_type == htons(ETH_TYPE_ARP)
2207 && flow->nw_proto == ARP_OP_REPLY
2208 && eth_addr_is_broadcast(flow->dl_dst));
2211 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2212 * dropped. Returns true if they may be forwarded, false if they should be
2215 * If 'have_packet' is true, it indicates that the caller is processing a
2216 * received packet. If 'have_packet' is false, then the caller is just
2217 * revalidating an existing flow because configuration has changed. Either
2218 * way, 'have_packet' only affects logging (there is no point in logging errors
2219 * during revalidation).
2221 * Sets '*in_portp' to the input port. This will be a null pointer if
2222 * flow->in_port does not designate a known input port (in which case
2223 * is_admissible() returns false).
2225 * When returning true, sets '*vlanp' to the effective VLAN of the input
2226 * packet, as returned by flow_get_vlan().
2228 * May also add tags to '*tags', although the current implementation only does
2229 * so in one special case.
2232 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2233 tag_type *tags, int *vlanp, struct port **in_portp)
2235 struct iface *in_iface;
2236 struct port *in_port;
2239 /* Find the interface and port structure for the received packet. */
2240 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2242 /* No interface? Something fishy... */
2244 /* Odd. A few possible reasons here:
2246 * - We deleted an interface but there are still a few packets
2247 * queued up from it.
2249 * - Someone externally added an interface (e.g. with "ovs-dpctl
2250 * add-if") that we don't know about.
2252 * - Packet arrived on the local port but the local port is not
2253 * one of our bridge ports.
2255 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2257 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2258 "interface %"PRIu16, br->name, flow->in_port);
2264 *in_portp = in_port = in_iface->port;
2265 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2270 /* Drop frames for reserved multicast addresses. */
2271 if (eth_addr_is_reserved(flow->dl_dst)) {
2275 /* Drop frames on ports reserved for mirroring. */
2276 if (in_port->is_mirror_output_port) {
2278 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2279 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2280 "%s, which is reserved exclusively for mirroring",
2281 br->name, in_port->name);
2286 /* Packets received on bonds need special attention to avoid duplicates. */
2287 if (in_port->n_ifaces > 1) {
2290 if (eth_addr_is_multicast(flow->dl_dst)) {
2291 *tags |= in_port->active_iface_tag;
2292 if (in_port->active_iface != in_iface->port_ifidx) {
2293 /* Drop all multicast packets on inactive slaves. */
2298 /* Drop all packets for which we have learned a different input
2299 * port, because we probably sent the packet on one slave and got
2300 * it back on the other. Broadcast ARP replies are an exception
2301 * to this rule: the host has moved to another switch. */
2302 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2303 if (src_idx != -1 && src_idx != in_port->port_idx &&
2304 !is_bcast_arp_reply(flow)) {
2312 /* If the composed actions may be applied to any packet in the given 'flow',
2313 * returns true. Otherwise, the actions should only be applied to 'packet', or
2314 * not at all, if 'packet' was NULL. */
2316 process_flow(struct bridge *br, const flow_t *flow,
2317 const struct ofpbuf *packet, struct odp_actions *actions,
2318 tag_type *tags, uint16_t *nf_output_iface)
2320 struct port *in_port;
2321 struct port *out_port;
2325 /* Check whether we should drop packets in this flow. */
2326 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2331 /* Learn source MAC (but don't try to learn from revalidation). */
2333 update_learning_table(br, flow, vlan, in_port);
2336 /* Determine output port. */
2337 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags);
2338 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2339 out_port = br->ports[out_port_idx];
2340 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2341 /* If we are revalidating but don't have a learning entry then
2342 * eject the flow. Installing a flow that floods packets opens
2343 * up a window of time where we could learn from a packet reflected
2344 * on a bond and blackhole packets before the learning table is
2345 * updated to reflect the correct port. */
2348 out_port = FLOOD_PORT;
2351 /* Don't send packets out their input ports. */
2352 if (in_port == out_port) {
2358 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2365 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2368 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2369 const struct ofp_phy_port *opp,
2372 struct bridge *br = br_;
2373 struct iface *iface;
2376 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2382 if (reason == OFPPR_DELETE) {
2383 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2384 br->name, iface->name);
2385 iface_destroy(iface);
2386 if (!port->n_ifaces) {
2387 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2388 br->name, port->name);
2394 if (port->n_ifaces > 1) {
2395 bool up = !(opp->state & OFPPS_LINK_DOWN);
2396 bond_link_status_update(iface, up);
2397 port_update_bond_compat(port);
2403 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2404 struct odp_actions *actions, tag_type *tags,
2405 uint16_t *nf_output_iface, void *br_)
2407 struct bridge *br = br_;
2409 COVERAGE_INC(bridge_process_flow);
2410 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2414 bridge_account_flow_ofhook_cb(const flow_t *flow,
2415 const union odp_action *actions,
2416 size_t n_actions, unsigned long long int n_bytes,
2419 struct bridge *br = br_;
2420 const union odp_action *a;
2421 struct port *in_port;
2425 /* Feed information from the active flows back into the learning table
2426 * to ensure that table is always in sync with what is actually flowing
2427 * through the datapath. */
2428 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2429 update_learning_table(br, flow, vlan, in_port);
2432 if (!br->has_bonded_ports) {
2436 for (a = actions; a < &actions[n_actions]; a++) {
2437 if (a->type == ODPAT_OUTPUT) {
2438 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2439 if (out_port && out_port->n_ifaces >= 2) {
2440 struct bond_entry *e = lookup_bond_entry(out_port,
2442 e->tx_bytes += n_bytes;
2449 bridge_account_checkpoint_ofhook_cb(void *br_)
2451 struct bridge *br = br_;
2455 if (!br->has_bonded_ports) {
2460 for (i = 0; i < br->n_ports; i++) {
2461 struct port *port = br->ports[i];
2462 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2463 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2464 bond_rebalance_port(port);
2469 static struct ofhooks bridge_ofhooks = {
2470 bridge_port_changed_ofhook_cb,
2471 bridge_normal_ofhook_cb,
2472 bridge_account_flow_ofhook_cb,
2473 bridge_account_checkpoint_ofhook_cb,
2476 /* Bonding functions. */
2478 /* Statistics for a single interface on a bonded port, used for load-based
2479 * bond rebalancing. */
2480 struct slave_balance {
2481 struct iface *iface; /* The interface. */
2482 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2484 /* All the "bond_entry"s that are assigned to this interface, in order of
2485 * increasing tx_bytes. */
2486 struct bond_entry **hashes;
2490 /* Sorts pointers to pointers to bond_entries in ascending order by the
2491 * interface to which they are assigned, and within a single interface in
2492 * ascending order of bytes transmitted. */
2494 compare_bond_entries(const void *a_, const void *b_)
2496 const struct bond_entry *const *ap = a_;
2497 const struct bond_entry *const *bp = b_;
2498 const struct bond_entry *a = *ap;
2499 const struct bond_entry *b = *bp;
2500 if (a->iface_idx != b->iface_idx) {
2501 return a->iface_idx > b->iface_idx ? 1 : -1;
2502 } else if (a->tx_bytes != b->tx_bytes) {
2503 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2509 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2510 * *descending* order by number of bytes transmitted. */
2512 compare_slave_balance(const void *a_, const void *b_)
2514 const struct slave_balance *a = a_;
2515 const struct slave_balance *b = b_;
2516 if (a->iface->enabled != b->iface->enabled) {
2517 return a->iface->enabled ? -1 : 1;
2518 } else if (a->tx_bytes != b->tx_bytes) {
2519 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2526 swap_bals(struct slave_balance *a, struct slave_balance *b)
2528 struct slave_balance tmp = *a;
2533 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2534 * given that 'p' (and only 'p') might be in the wrong location.
2536 * This function invalidates 'p', since it might now be in a different memory
2539 resort_bals(struct slave_balance *p,
2540 struct slave_balance bals[], size_t n_bals)
2543 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2544 swap_bals(p, p - 1);
2546 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2547 swap_bals(p, p + 1);
2553 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2555 if (VLOG_IS_DBG_ENABLED()) {
2556 struct ds ds = DS_EMPTY_INITIALIZER;
2557 const struct slave_balance *b;
2559 for (b = bals; b < bals + n_bals; b++) {
2563 ds_put_char(&ds, ',');
2565 ds_put_format(&ds, " %s %"PRIu64"kB",
2566 b->iface->name, b->tx_bytes / 1024);
2568 if (!b->iface->enabled) {
2569 ds_put_cstr(&ds, " (disabled)");
2571 if (b->n_hashes > 0) {
2572 ds_put_cstr(&ds, " (");
2573 for (i = 0; i < b->n_hashes; i++) {
2574 const struct bond_entry *e = b->hashes[i];
2576 ds_put_cstr(&ds, " + ");
2578 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2579 e - port->bond_hash, e->tx_bytes / 1024);
2581 ds_put_cstr(&ds, ")");
2584 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2589 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2591 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2594 struct bond_entry *hash = from->hashes[hash_idx];
2595 struct port *port = from->iface->port;
2596 uint64_t delta = hash->tx_bytes;
2598 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2599 "from %s to %s (now carrying %"PRIu64"kB and "
2600 "%"PRIu64"kB load, respectively)",
2601 port->name, delta / 1024, hash - port->bond_hash,
2602 from->iface->name, to->iface->name,
2603 (from->tx_bytes - delta) / 1024,
2604 (to->tx_bytes + delta) / 1024);
2606 /* Delete element from from->hashes.
2608 * We don't bother to add the element to to->hashes because not only would
2609 * it require more work, the only purpose it would be to allow that hash to
2610 * be migrated to another slave in this rebalancing run, and there is no
2611 * point in doing that. */
2612 if (hash_idx == 0) {
2615 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2616 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2620 /* Shift load away from 'from' to 'to'. */
2621 from->tx_bytes -= delta;
2622 to->tx_bytes += delta;
2624 /* Arrange for flows to be revalidated. */
2625 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2626 hash->iface_idx = to->iface->port_ifidx;
2627 hash->iface_tag = tag_create_random();
2631 bond_rebalance_port(struct port *port)
2633 struct slave_balance bals[DP_MAX_PORTS];
2635 struct bond_entry *hashes[BOND_MASK + 1];
2636 struct slave_balance *b, *from, *to;
2637 struct bond_entry *e;
2640 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2641 * descending order of tx_bytes, so that bals[0] represents the most
2642 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2645 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2646 * array for each slave_balance structure, we sort our local array of
2647 * hashes in order by slave, so that all of the hashes for a given slave
2648 * become contiguous in memory, and then we point each 'hashes' members of
2649 * a slave_balance structure to the start of a contiguous group. */
2650 n_bals = port->n_ifaces;
2651 for (b = bals; b < &bals[n_bals]; b++) {
2652 b->iface = port->ifaces[b - bals];
2657 for (i = 0; i <= BOND_MASK; i++) {
2658 hashes[i] = &port->bond_hash[i];
2660 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2661 for (i = 0; i <= BOND_MASK; i++) {
2663 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2664 b = &bals[e->iface_idx];
2665 b->tx_bytes += e->tx_bytes;
2667 b->hashes = &hashes[i];
2672 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2673 log_bals(bals, n_bals, port);
2675 /* Discard slaves that aren't enabled (which were sorted to the back of the
2676 * array earlier). */
2677 while (!bals[n_bals - 1].iface->enabled) {
2684 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2685 to = &bals[n_bals - 1];
2686 for (from = bals; from < to; ) {
2687 uint64_t overload = from->tx_bytes - to->tx_bytes;
2688 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2689 /* The extra load on 'from' (and all less-loaded slaves), compared
2690 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2691 * it is less than ~1Mbps. No point in rebalancing. */
2693 } else if (from->n_hashes == 1) {
2694 /* 'from' only carries a single MAC hash, so we can't shift any
2695 * load away from it, even though we want to. */
2698 /* 'from' is carrying significantly more load than 'to', and that
2699 * load is split across at least two different hashes. Pick a hash
2700 * to migrate to 'to' (the least-loaded slave), given that doing so
2701 * must decrease the ratio of the load on the two slaves by at
2704 * The sort order we use means that we prefer to shift away the
2705 * smallest hashes instead of the biggest ones. There is little
2706 * reason behind this decision; we could use the opposite sort
2707 * order to shift away big hashes ahead of small ones. */
2711 for (i = 0; i < from->n_hashes; i++) {
2712 double old_ratio, new_ratio;
2713 uint64_t delta = from->hashes[i]->tx_bytes;
2715 if (delta == 0 || from->tx_bytes - delta == 0) {
2716 /* Pointless move. */
2720 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2722 if (to->tx_bytes == 0) {
2723 /* Nothing on the new slave, move it. */
2727 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2728 new_ratio = (double)(from->tx_bytes - delta) /
2729 (to->tx_bytes + delta);
2731 if (new_ratio == 0) {
2732 /* Should already be covered but check to prevent division
2737 if (new_ratio < 1) {
2738 new_ratio = 1 / new_ratio;
2741 if (old_ratio - new_ratio > 0.1) {
2742 /* Would decrease the ratio, move it. */
2746 if (i < from->n_hashes) {
2747 bond_shift_load(from, to, i);
2748 port->bond_compat_is_stale = true;
2750 /* If the result of the migration changed the relative order of
2751 * 'from' and 'to' swap them back to maintain invariants. */
2752 if (order_swapped) {
2753 swap_bals(from, to);
2756 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2757 * point to different slave_balance structures. It is only
2758 * valid to do these two operations in a row at all because we
2759 * know that 'from' will not move past 'to' and vice versa. */
2760 resort_bals(from, bals, n_bals);
2761 resort_bals(to, bals, n_bals);
2768 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2769 * historical data to decay to <1% in 7 rebalancing runs. */
2770 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2776 bond_send_learning_packets(struct port *port)
2778 struct bridge *br = port->bridge;
2779 struct mac_entry *e;
2780 struct ofpbuf packet;
2781 int error, n_packets, n_errors;
2783 if (!port->n_ifaces || port->active_iface < 0) {
2787 ofpbuf_init(&packet, 128);
2788 error = n_packets = n_errors = 0;
2789 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2790 union ofp_action actions[2], *a;
2796 if (e->port == port->port_idx
2797 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2801 /* Compose actions. */
2802 memset(actions, 0, sizeof actions);
2805 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2806 a->vlan_vid.len = htons(sizeof *a);
2807 a->vlan_vid.vlan_vid = htons(e->vlan);
2810 a->output.type = htons(OFPAT_OUTPUT);
2811 a->output.len = htons(sizeof *a);
2812 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2817 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2819 flow_extract(&packet, 0, ODPP_NONE, &flow);
2820 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2827 ofpbuf_uninit(&packet);
2830 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2831 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2832 "packets, last error was: %s",
2833 port->name, n_errors, n_packets, strerror(error));
2835 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2836 port->name, n_packets);
2840 /* Bonding unixctl user interface functions. */
2843 bond_unixctl_list(struct unixctl_conn *conn,
2844 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2846 struct ds ds = DS_EMPTY_INITIALIZER;
2847 const struct bridge *br;
2849 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2851 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2854 for (i = 0; i < br->n_ports; i++) {
2855 const struct port *port = br->ports[i];
2856 if (port->n_ifaces > 1) {
2859 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2860 for (j = 0; j < port->n_ifaces; j++) {
2861 const struct iface *iface = port->ifaces[j];
2863 ds_put_cstr(&ds, ", ");
2865 ds_put_cstr(&ds, iface->name);
2867 ds_put_char(&ds, '\n');
2871 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2875 static struct port *
2876 bond_find(const char *name)
2878 const struct bridge *br;
2880 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2883 for (i = 0; i < br->n_ports; i++) {
2884 struct port *port = br->ports[i];
2885 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2894 bond_unixctl_show(struct unixctl_conn *conn,
2895 const char *args, void *aux OVS_UNUSED)
2897 struct ds ds = DS_EMPTY_INITIALIZER;
2898 const struct port *port;
2901 port = bond_find(args);
2903 unixctl_command_reply(conn, 501, "no such bond");
2907 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2908 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2909 ds_put_format(&ds, "next rebalance: %lld ms\n",
2910 port->bond_next_rebalance - time_msec());
2911 for (j = 0; j < port->n_ifaces; j++) {
2912 const struct iface *iface = port->ifaces[j];
2913 struct bond_entry *be;
2916 ds_put_format(&ds, "slave %s: %s\n",
2917 iface->name, iface->enabled ? "enabled" : "disabled");
2918 if (j == port->active_iface) {
2919 ds_put_cstr(&ds, "\tactive slave\n");
2921 if (iface->delay_expires != LLONG_MAX) {
2922 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2923 iface->enabled ? "downdelay" : "updelay",
2924 iface->delay_expires - time_msec());
2928 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2929 int hash = be - port->bond_hash;
2930 struct mac_entry *me;
2932 if (be->iface_idx != j) {
2936 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2937 hash, be->tx_bytes / 1024);
2940 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2941 &port->bridge->ml->lrus) {
2944 if (bond_hash(me->mac) == hash
2945 && me->port != port->port_idx
2946 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2947 && dp_ifidx == iface->dp_ifidx)
2949 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2950 ETH_ADDR_ARGS(me->mac));
2955 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2960 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2961 void *aux OVS_UNUSED)
2963 char *args = (char *) args_;
2964 char *save_ptr = NULL;
2965 char *bond_s, *hash_s, *slave_s;
2966 uint8_t mac[ETH_ADDR_LEN];
2968 struct iface *iface;
2969 struct bond_entry *entry;
2972 bond_s = strtok_r(args, " ", &save_ptr);
2973 hash_s = strtok_r(NULL, " ", &save_ptr);
2974 slave_s = strtok_r(NULL, " ", &save_ptr);
2976 unixctl_command_reply(conn, 501,
2977 "usage: bond/migrate BOND HASH SLAVE");
2981 port = bond_find(bond_s);
2983 unixctl_command_reply(conn, 501, "no such bond");
2987 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2988 == ETH_ADDR_SCAN_COUNT) {
2989 hash = bond_hash(mac);
2990 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2991 hash = atoi(hash_s) & BOND_MASK;
2993 unixctl_command_reply(conn, 501, "bad hash");
2997 iface = port_lookup_iface(port, slave_s);
2999 unixctl_command_reply(conn, 501, "no such slave");
3003 if (!iface->enabled) {
3004 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3008 entry = &port->bond_hash[hash];
3009 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3010 entry->iface_idx = iface->port_ifidx;
3011 entry->iface_tag = tag_create_random();
3012 port->bond_compat_is_stale = true;
3013 unixctl_command_reply(conn, 200, "migrated");
3017 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3018 void *aux OVS_UNUSED)
3020 char *args = (char *) args_;
3021 char *save_ptr = NULL;
3022 char *bond_s, *slave_s;
3024 struct iface *iface;
3026 bond_s = strtok_r(args, " ", &save_ptr);
3027 slave_s = strtok_r(NULL, " ", &save_ptr);
3029 unixctl_command_reply(conn, 501,
3030 "usage: bond/set-active-slave BOND SLAVE");
3034 port = bond_find(bond_s);
3036 unixctl_command_reply(conn, 501, "no such bond");
3040 iface = port_lookup_iface(port, slave_s);
3042 unixctl_command_reply(conn, 501, "no such slave");
3046 if (!iface->enabled) {
3047 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3051 if (port->active_iface != iface->port_ifidx) {
3052 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3053 port->active_iface = iface->port_ifidx;
3054 port->active_iface_tag = tag_create_random();
3055 VLOG_INFO("port %s: active interface is now %s",
3056 port->name, iface->name);
3057 bond_send_learning_packets(port);
3058 unixctl_command_reply(conn, 200, "done");
3060 unixctl_command_reply(conn, 200, "no change");
3065 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3067 char *args = (char *) args_;
3068 char *save_ptr = NULL;
3069 char *bond_s, *slave_s;
3071 struct iface *iface;
3073 bond_s = strtok_r(args, " ", &save_ptr);
3074 slave_s = strtok_r(NULL, " ", &save_ptr);
3076 unixctl_command_reply(conn, 501,
3077 "usage: bond/enable/disable-slave BOND SLAVE");
3081 port = bond_find(bond_s);
3083 unixctl_command_reply(conn, 501, "no such bond");
3087 iface = port_lookup_iface(port, slave_s);
3089 unixctl_command_reply(conn, 501, "no such slave");
3093 bond_enable_slave(iface, enable);
3094 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3098 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3099 void *aux OVS_UNUSED)
3101 enable_slave(conn, args, true);
3105 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3106 void *aux OVS_UNUSED)
3108 enable_slave(conn, args, false);
3112 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3113 void *aux OVS_UNUSED)
3115 uint8_t mac[ETH_ADDR_LEN];
3119 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3120 == ETH_ADDR_SCAN_COUNT) {
3121 hash = bond_hash(mac);
3123 hash_cstr = xasprintf("%u", hash);
3124 unixctl_command_reply(conn, 200, hash_cstr);
3127 unixctl_command_reply(conn, 501, "invalid mac");
3134 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3135 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3136 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3137 unixctl_command_register("bond/set-active-slave",
3138 bond_unixctl_set_active_slave, NULL);
3139 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3141 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3143 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3146 /* Port functions. */
3148 static struct port *
3149 port_create(struct bridge *br, const char *name)
3153 port = xzalloc(sizeof *port);
3155 port->port_idx = br->n_ports;
3157 port->trunks = NULL;
3158 port->name = xstrdup(name);
3159 port->active_iface = -1;
3161 if (br->n_ports >= br->allocated_ports) {
3162 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3165 br->ports[br->n_ports++] = port;
3167 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3174 get_port_other_config(const struct ovsrec_port *port, const char *key,
3175 const char *default_value)
3177 const char *value = get_ovsrec_key_value(key,
3178 port->key_other_config,
3179 port->value_other_config,
3180 port->n_other_config);
3181 return value ? value : default_value;
3185 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3187 struct shash old_ifaces, new_ifaces;
3188 long long int next_rebalance;
3189 struct shash_node *node;
3190 unsigned long *trunks;
3196 /* Collect old and new interfaces. */
3197 shash_init(&old_ifaces);
3198 shash_init(&new_ifaces);
3199 for (i = 0; i < port->n_ifaces; i++) {
3200 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3202 for (i = 0; i < cfg->n_interfaces; i++) {
3203 const char *name = cfg->interfaces[i]->name;
3204 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3205 VLOG_WARN("port %s: %s specified twice as port interface",
3209 port->updelay = cfg->bond_updelay;
3210 if (port->updelay < 0) {
3213 port->updelay = cfg->bond_downdelay;
3214 if (port->downdelay < 0) {
3215 port->downdelay = 0;
3217 port->bond_rebalance_interval = atoi(
3218 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3219 if (port->bond_rebalance_interval < 1000) {
3220 port->bond_rebalance_interval = 1000;
3222 next_rebalance = time_msec() + port->bond_rebalance_interval;
3223 if (port->bond_next_rebalance > next_rebalance) {
3224 port->bond_next_rebalance = next_rebalance;
3227 /* Get rid of deleted interfaces and add new interfaces. */
3228 SHASH_FOR_EACH (node, &old_ifaces) {
3229 if (!shash_find(&new_ifaces, node->name)) {
3230 iface_destroy(node->data);
3233 SHASH_FOR_EACH (node, &new_ifaces) {
3234 const struct ovsrec_interface *if_cfg = node->data;
3235 struct iface *iface;
3237 iface = shash_find_data(&old_ifaces, if_cfg->name);
3239 iface_create(port, if_cfg);
3241 iface->cfg = if_cfg;
3248 if (port->n_ifaces < 2) {
3250 if (vlan >= 0 && vlan <= 4095) {
3251 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3256 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3257 * they even work as-is. But they have not been tested. */
3258 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3262 if (port->vlan != vlan) {
3264 bridge_flush(port->bridge);
3267 /* Get trunked VLANs. */
3273 trunks = bitmap_allocate(4096);
3275 for (i = 0; i < cfg->n_trunks; i++) {
3276 int trunk = cfg->trunks[i];
3278 bitmap_set1(trunks, trunk);
3284 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3285 port->name, cfg->n_trunks);
3287 if (n_errors == cfg->n_trunks) {
3289 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3292 bitmap_set_multiple(trunks, 0, 4096, 1);
3295 if (cfg->n_trunks) {
3296 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3301 ? port->trunks != NULL
3302 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3303 bridge_flush(port->bridge);
3305 bitmap_free(port->trunks);
3306 port->trunks = trunks;
3308 shash_destroy(&old_ifaces);
3309 shash_destroy(&new_ifaces);
3313 port_destroy(struct port *port)
3316 struct bridge *br = port->bridge;
3320 proc_net_compat_update_vlan(port->name, NULL, 0);
3321 proc_net_compat_update_bond(port->name, NULL);
3323 for (i = 0; i < MAX_MIRRORS; i++) {
3324 struct mirror *m = br->mirrors[i];
3325 if (m && m->out_port == port) {
3330 while (port->n_ifaces > 0) {
3331 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3334 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3335 del->port_idx = port->port_idx;
3338 bitmap_free(port->trunks);
3345 static struct port *
3346 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3348 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3349 return iface ? iface->port : NULL;
3352 static struct port *
3353 port_lookup(const struct bridge *br, const char *name)
3357 for (i = 0; i < br->n_ports; i++) {
3358 struct port *port = br->ports[i];
3359 if (!strcmp(port->name, name)) {
3366 static struct iface *
3367 port_lookup_iface(const struct port *port, const char *name)
3371 for (j = 0; j < port->n_ifaces; j++) {
3372 struct iface *iface = port->ifaces[j];
3373 if (!strcmp(iface->name, name)) {
3381 port_update_bonding(struct port *port)
3383 if (port->n_ifaces < 2) {
3384 /* Not a bonded port. */
3385 if (port->bond_hash) {
3386 free(port->bond_hash);
3387 port->bond_hash = NULL;
3388 port->bond_compat_is_stale = true;
3389 port->bond_fake_iface = false;
3392 if (!port->bond_hash) {
3395 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3396 for (i = 0; i <= BOND_MASK; i++) {
3397 struct bond_entry *e = &port->bond_hash[i];
3401 port->no_ifaces_tag = tag_create_random();
3402 bond_choose_active_iface(port);
3403 port->bond_next_rebalance
3404 = time_msec() + port->bond_rebalance_interval;
3406 if (port->cfg->bond_fake_iface) {
3407 port->bond_next_fake_iface_update = time_msec();
3410 port->bond_compat_is_stale = true;
3411 port->bond_fake_iface = port->cfg->bond_fake_iface;
3416 port_update_bond_compat(struct port *port)
3418 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3419 struct compat_bond bond;
3422 if (port->n_ifaces < 2) {
3423 proc_net_compat_update_bond(port->name, NULL);
3428 bond.updelay = port->updelay;
3429 bond.downdelay = port->downdelay;
3432 bond.hashes = compat_hashes;
3433 if (port->bond_hash) {
3434 const struct bond_entry *e;
3435 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3436 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3437 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3438 cbh->hash = e - port->bond_hash;
3439 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3444 bond.n_slaves = port->n_ifaces;
3445 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3446 for (i = 0; i < port->n_ifaces; i++) {
3447 struct iface *iface = port->ifaces[i];
3448 struct compat_bond_slave *slave = &bond.slaves[i];
3449 slave->name = iface->name;
3451 /* We need to make the same determination as the Linux bonding
3452 * code to determine whether a slave should be consider "up".
3453 * The Linux function bond_miimon_inspect() supports four
3454 * BOND_LINK_* states:
3456 * - BOND_LINK_UP: carrier detected, updelay has passed.
3457 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3458 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3459 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3461 * The function bond_info_show_slave() only considers BOND_LINK_UP
3462 * to be "up" and anything else to be "down".
3464 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3468 netdev_get_etheraddr(iface->netdev, slave->mac);
3471 if (port->bond_fake_iface) {
3472 struct netdev *bond_netdev;
3474 if (!netdev_open_default(port->name, &bond_netdev)) {
3476 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3478 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3480 netdev_close(bond_netdev);
3484 proc_net_compat_update_bond(port->name, &bond);
3489 port_update_vlan_compat(struct port *port)
3491 struct bridge *br = port->bridge;
3492 char *vlandev_name = NULL;
3494 if (port->vlan > 0) {
3495 /* Figure out the name that the VLAN device should actually have, if it
3496 * existed. This takes some work because the VLAN device would not
3497 * have port->name in its name; rather, it would have the trunk port's
3498 * name, and 'port' would be attached to a bridge that also had the
3499 * VLAN device one of its ports. So we need to find a trunk port that
3500 * includes port->vlan.
3502 * There might be more than one candidate. This doesn't happen on
3503 * XenServer, so if it happens we just pick the first choice in
3504 * alphabetical order instead of creating multiple VLAN devices. */
3506 for (i = 0; i < br->n_ports; i++) {
3507 struct port *p = br->ports[i];
3508 if (port_trunks_vlan(p, port->vlan)
3510 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3512 uint8_t ea[ETH_ADDR_LEN];
3513 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3514 if (!eth_addr_is_multicast(ea) &&
3515 !eth_addr_is_reserved(ea) &&
3516 !eth_addr_is_zero(ea)) {
3517 vlandev_name = p->name;
3522 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3525 /* Interface functions. */
3527 static struct iface *
3528 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3530 struct iface *iface;
3531 char *name = if_cfg->name;
3534 iface = xzalloc(sizeof *iface);
3536 iface->port_ifidx = port->n_ifaces;
3537 iface->name = xstrdup(name);
3538 iface->dp_ifidx = -1;
3539 iface->tag = tag_create_random();
3540 iface->delay_expires = LLONG_MAX;
3541 iface->netdev = NULL;
3542 iface->cfg = if_cfg;
3544 if (port->n_ifaces >= port->allocated_ifaces) {
3545 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3546 sizeof *port->ifaces);
3548 port->ifaces[port->n_ifaces++] = iface;
3549 if (port->n_ifaces > 1) {
3550 port->bridge->has_bonded_ports = true;
3553 /* Attempt to create the network interface in case it
3554 * doesn't exist yet. */
3555 if (!iface_is_internal(port->bridge, iface->name)) {
3556 error = set_up_iface(if_cfg, iface, true);
3558 VLOG_WARN("could not create iface %s: %s", iface->name,
3563 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3565 bridge_flush(port->bridge);
3571 iface_destroy(struct iface *iface)
3574 struct port *port = iface->port;
3575 struct bridge *br = port->bridge;
3576 bool del_active = port->active_iface == iface->port_ifidx;
3579 if (iface->dp_ifidx >= 0) {
3580 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3583 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3584 del->port_ifidx = iface->port_ifidx;
3586 netdev_close(iface->netdev);
3589 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3590 bond_choose_active_iface(port);
3591 bond_send_learning_packets(port);
3597 bridge_flush(port->bridge);
3601 static struct iface *
3602 iface_lookup(const struct bridge *br, const char *name)
3606 for (i = 0; i < br->n_ports; i++) {
3607 struct port *port = br->ports[i];
3608 for (j = 0; j < port->n_ifaces; j++) {
3609 struct iface *iface = port->ifaces[j];
3610 if (!strcmp(iface->name, name)) {
3618 static struct iface *
3619 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3621 return port_array_get(&br->ifaces, dp_ifidx);
3624 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3625 * 'br', that is, an interface that is entirely simulated within the datapath.
3626 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3627 * interfaces are created by setting "iface.<iface>.internal = true".
3629 * In addition, we have a kluge-y feature that creates an internal port with
3630 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3631 * This feature needs to go away in the long term. Until then, this is one
3632 * reason why this function takes a name instead of a struct iface: the fake
3633 * interfaces created this way do not have a struct iface. */
3635 iface_is_internal(const struct bridge *br, const char *if_name)
3637 /* XXX wastes time */
3638 struct iface *iface;
3641 if (!strcmp(if_name, br->name)) {
3645 iface = iface_lookup(br, if_name);
3646 if (iface && !strcmp(iface->cfg->type, "internal")) {
3650 port = port_lookup(br, if_name);
3651 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3657 /* Set Ethernet address of 'iface', if one is specified in the configuration
3660 iface_set_mac(struct iface *iface)
3662 uint8_t ea[ETH_ADDR_LEN];
3664 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3665 if (eth_addr_is_multicast(ea)) {
3666 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3668 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3669 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3670 iface->name, iface->name);
3672 int error = netdev_set_etheraddr(iface->netdev, ea);
3674 VLOG_ERR("interface %s: setting MAC failed (%s)",
3675 iface->name, strerror(error));
3681 /* Port mirroring. */
3684 mirror_reconfigure(struct bridge *br)
3686 struct shash old_mirrors, new_mirrors;
3687 struct shash_node *node;
3688 unsigned long *rspan_vlans;
3691 /* Collect old mirrors. */
3692 shash_init(&old_mirrors);
3693 for (i = 0; i < MAX_MIRRORS; i++) {
3694 if (br->mirrors[i]) {
3695 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3699 /* Collect new mirrors. */
3700 shash_init(&new_mirrors);
3701 for (i = 0; i < br->cfg->n_mirrors; i++) {
3702 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3703 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3704 VLOG_WARN("bridge %s: %s specified twice as mirror",
3705 br->name, cfg->name);
3709 /* Get rid of deleted mirrors and add new mirrors. */
3710 SHASH_FOR_EACH (node, &old_mirrors) {
3711 if (!shash_find(&new_mirrors, node->name)) {
3712 mirror_destroy(node->data);
3715 SHASH_FOR_EACH (node, &new_mirrors) {
3716 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3718 mirror = mirror_create(br, node->name);
3723 mirror_reconfigure_one(mirror, node->data);
3725 shash_destroy(&old_mirrors);
3726 shash_destroy(&new_mirrors);
3728 /* Update port reserved status. */
3729 for (i = 0; i < br->n_ports; i++) {
3730 br->ports[i]->is_mirror_output_port = false;
3732 for (i = 0; i < MAX_MIRRORS; i++) {
3733 struct mirror *m = br->mirrors[i];
3734 if (m && m->out_port) {
3735 m->out_port->is_mirror_output_port = true;
3739 /* Update flooded vlans (for RSPAN). */
3741 if (br->cfg->n_flood_vlans) {
3742 rspan_vlans = bitmap_allocate(4096);
3744 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3745 int64_t vlan = br->cfg->flood_vlans[i];
3746 if (vlan >= 0 && vlan < 4096) {
3747 bitmap_set1(rspan_vlans, vlan);
3748 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3751 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3756 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3761 static struct mirror *
3762 mirror_create(struct bridge *br, const char *name)
3767 for (i = 0; ; i++) {
3768 if (i >= MAX_MIRRORS) {
3769 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3770 "cannot create %s", br->name, MAX_MIRRORS, name);
3773 if (!br->mirrors[i]) {
3778 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3781 br->mirrors[i] = m = xzalloc(sizeof *m);
3784 m->name = xstrdup(name);
3785 shash_init(&m->src_ports);
3786 shash_init(&m->dst_ports);
3796 mirror_destroy(struct mirror *m)
3799 struct bridge *br = m->bridge;
3802 for (i = 0; i < br->n_ports; i++) {
3803 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3804 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3807 shash_destroy(&m->src_ports);
3808 shash_destroy(&m->dst_ports);
3811 m->bridge->mirrors[m->idx] = NULL;
3819 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3820 struct shash *names)
3824 for (i = 0; i < n_ports; i++) {
3825 const char *name = ports[i]->name;
3826 if (port_lookup(m->bridge, name)) {
3827 shash_add_once(names, name, NULL);
3829 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3830 "port %s", m->bridge->name, m->name, name);
3836 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3842 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3844 for (i = 0; i < cfg->n_select_vlan; i++) {
3845 int64_t vlan = cfg->select_vlan[i];
3846 if (vlan < 0 || vlan > 4095) {
3847 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3848 m->bridge->name, m->name, vlan);
3850 (*vlans)[n_vlans++] = vlan;
3857 vlan_is_mirrored(const struct mirror *m, int vlan)
3861 for (i = 0; i < m->n_vlans; i++) {
3862 if (m->vlans[i] == vlan) {
3870 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3874 for (i = 0; i < m->n_vlans; i++) {
3875 if (port_trunks_vlan(p, m->vlans[i])) {
3883 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3885 struct shash src_ports, dst_ports;
3886 mirror_mask_t mirror_bit;
3887 struct port *out_port;
3893 /* Get output port. */
3894 if (cfg->output_port) {
3895 out_port = port_lookup(m->bridge, cfg->output_port->name);
3897 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3898 m->bridge->name, m->name);
3904 if (cfg->output_vlan) {
3905 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3906 "output vlan; ignoring output vlan",
3907 m->bridge->name, m->name);
3909 } else if (cfg->output_vlan) {
3911 out_vlan = *cfg->output_vlan;
3913 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3914 m->bridge->name, m->name);
3919 shash_init(&src_ports);
3920 shash_init(&dst_ports);
3921 if (cfg->select_all) {
3922 for (i = 0; i < m->bridge->n_ports; i++) {
3923 const char *name = m->bridge->ports[i]->name;
3924 shash_add_once(&src_ports, name, NULL);
3925 shash_add_once(&dst_ports, name, NULL);
3930 /* Get ports, and drop duplicates and ports that don't exist. */
3931 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3933 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3936 /* Get all the vlans, and drop duplicate and invalid vlans. */
3937 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3940 /* Update mirror data. */
3941 if (!shash_equal_keys(&m->src_ports, &src_ports)
3942 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3943 || m->n_vlans != n_vlans
3944 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3945 || m->out_port != out_port
3946 || m->out_vlan != out_vlan) {
3947 bridge_flush(m->bridge);
3949 shash_swap(&m->src_ports, &src_ports);
3950 shash_swap(&m->dst_ports, &dst_ports);
3953 m->n_vlans = n_vlans;
3954 m->out_port = out_port;
3955 m->out_vlan = out_vlan;
3958 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3959 for (i = 0; i < m->bridge->n_ports; i++) {
3960 struct port *port = m->bridge->ports[i];
3962 if (shash_find(&m->src_ports, port->name)
3965 ? port_trunks_any_mirrored_vlan(m, port)
3966 : vlan_is_mirrored(m, port->vlan)))) {
3967 port->src_mirrors |= mirror_bit;
3969 port->src_mirrors &= ~mirror_bit;
3972 if (shash_find(&m->dst_ports, port->name)) {
3973 port->dst_mirrors |= mirror_bit;
3975 port->dst_mirrors &= ~mirror_bit;
3980 shash_destroy(&src_ports);
3981 shash_destroy(&dst_ports);