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>
35 #include "dynamic-string.h"
39 #include "mac-learning.h"
41 #include "xflow-util.h"
42 #include "ofp-print.h"
44 #include "ofproto/netflow.h"
45 #include "ofproto/ofproto.h"
47 #include "poll-loop.h"
48 #include "port-array.h"
49 #include "proc-net-compat.h"
53 #include "socket-util.h"
54 #include "stream-ssl.h"
60 #include "vswitchd/vswitch-idl.h"
61 #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 xf_ifidx; /* Index within kernel datapath. */
85 struct netdev *netdev; /* Network device. */
86 bool enabled; /* May be chosen for flows? */
88 /* This member is only valid *during* bridge_reconfigure(). */
89 const struct ovsrec_interface *cfg;
92 #define BOND_MASK 0xff
94 int iface_idx; /* Index of assigned iface, or -1 if none. */
95 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
96 tag_type iface_tag; /* Tag associated with iface_idx. */
99 #define MAX_MIRRORS 32
100 typedef uint32_t mirror_mask_t;
101 #define MIRROR_MASK_C(X) UINT32_C(X)
102 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
104 struct bridge *bridge;
108 /* Selection criteria. */
109 struct shash src_ports; /* Name is port name; data is always NULL. */
110 struct shash dst_ports; /* Name is port name; data is always NULL. */
115 struct port *out_port;
119 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
121 struct bridge *bridge;
123 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
124 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
127 /* An ordinary bridge port has 1 interface.
128 * A bridge port for bonding has at least 2 interfaces. */
129 struct iface **ifaces;
130 size_t n_ifaces, allocated_ifaces;
133 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
134 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
135 tag_type active_iface_tag; /* Tag for bcast flows. */
136 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
137 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
138 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
139 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
141 /* Port mirroring info. */
142 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
143 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
144 bool is_mirror_output_port; /* Does port mirroring send frames here? */
146 /* This member is only valid *during* bridge_reconfigure(). */
147 const struct ovsrec_port *cfg;
150 #define DP_MAX_PORTS 255
152 struct list node; /* Node in global list of bridges. */
153 char *name; /* User-specified arbitrary name. */
154 struct mac_learning *ml; /* MAC learning table. */
155 bool sent_config_request; /* Successfully sent config request? */
156 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
158 /* Support for remote controllers. */
159 char *controller; /* NULL if there is no remote controller;
160 * "discover" to do controller discovery;
161 * otherwise a vconn name. */
163 /* OpenFlow switch processing. */
164 struct ofproto *ofproto; /* OpenFlow switch. */
166 /* Description strings. */
167 char *mfr_desc; /* Manufacturer. */
168 char *hw_desc; /* Hardware. */
169 char *sw_desc; /* Software version. */
170 char *serial_desc; /* Serial number. */
171 char *dp_desc; /* Datapath description. */
173 /* Kernel datapath information. */
174 struct xfif *xfif; /* Datapath. */
175 struct port_array ifaces; /* Indexed by kernel datapath port number. */
179 size_t n_ports, allocated_ports;
182 bool has_bonded_ports;
183 long long int bond_next_rebalance;
188 /* Flow statistics gathering. */
189 time_t next_stats_request;
191 /* Port mirroring. */
192 struct mirror *mirrors[MAX_MIRRORS];
194 /* This member is only valid *during* bridge_reconfigure(). */
195 const struct ovsrec_bridge *cfg;
198 /* List of all bridges. */
199 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
201 /* Maximum number of datapaths. */
202 enum { DP_MAX = 256 };
204 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
205 static void bridge_destroy(struct bridge *);
206 static struct bridge *bridge_lookup(const char *name);
207 static unixctl_cb_func bridge_unixctl_dump_flows;
208 static int bridge_run_one(struct bridge *);
209 static const struct ovsrec_controller *bridge_get_controller(
210 const struct ovsrec_open_vswitch *ovs_cfg,
211 const struct bridge *br);
212 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
214 static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
216 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
217 static void bridge_fetch_dp_ifaces(struct bridge *);
218 static void bridge_flush(struct bridge *);
219 static void bridge_pick_local_hw_addr(struct bridge *,
220 uint8_t ea[ETH_ADDR_LEN],
221 struct iface **hw_addr_iface);
222 static uint64_t bridge_pick_datapath_id(struct bridge *,
223 const uint8_t bridge_ea[ETH_ADDR_LEN],
224 struct iface *hw_addr_iface);
225 static struct iface *bridge_get_local_iface(struct bridge *);
226 static uint64_t dpid_from_hash(const void *, size_t nbytes);
228 static unixctl_cb_func bridge_unixctl_fdb_show;
230 static void bond_init(void);
231 static void bond_run(struct bridge *);
232 static void bond_wait(struct bridge *);
233 static void bond_rebalance_port(struct port *);
234 static void bond_send_learning_packets(struct port *);
235 static void bond_enable_slave(struct iface *iface, bool enable);
237 static struct port *port_create(struct bridge *, const char *name);
238 static void port_reconfigure(struct port *, const struct ovsrec_port *);
239 static void port_destroy(struct port *);
240 static struct port *port_lookup(const struct bridge *, const char *name);
241 static struct iface *port_lookup_iface(const struct port *, const char *name);
242 static struct port *port_from_xf_ifidx(const struct bridge *,
244 static void port_update_bond_compat(struct port *);
245 static void port_update_vlan_compat(struct port *);
246 static void port_update_bonding(struct port *);
248 static struct mirror *mirror_create(struct bridge *, const char *name);
249 static void mirror_destroy(struct mirror *);
250 static void mirror_reconfigure(struct bridge *);
251 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
252 static bool vlan_is_mirrored(const struct mirror *, int vlan);
254 static struct iface *iface_create(struct port *port,
255 const struct ovsrec_interface *if_cfg);
256 static void iface_destroy(struct iface *);
257 static struct iface *iface_lookup(const struct bridge *, const char *name);
258 static struct iface *iface_from_xf_ifidx(const struct bridge *,
260 static bool iface_is_internal(const struct bridge *, const char *name);
261 static void iface_set_mac(struct iface *);
263 /* Hooks into ofproto processing. */
264 static struct ofhooks bridge_ofhooks;
266 /* Public functions. */
268 /* Adds the name of each interface used by a bridge, including local and
269 * internal ports, to 'svec'. */
271 bridge_get_ifaces(struct svec *svec)
273 struct bridge *br, *next;
276 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
277 for (i = 0; i < br->n_ports; i++) {
278 struct port *port = br->ports[i];
280 for (j = 0; j < port->n_ifaces; j++) {
281 struct iface *iface = port->ifaces[j];
282 if (iface->xf_ifidx < 0) {
283 VLOG_ERR("%s interface not in datapath %s, ignoring",
284 iface->name, xfif_name(br->xfif));
286 if (iface->xf_ifidx != XFLOWP_LOCAL) {
287 svec_add(svec, iface->name);
296 bridge_init(const struct ovsrec_open_vswitch *cfg)
298 struct svec bridge_names;
299 struct svec xfif_names, xfif_types;
302 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
304 svec_init(&bridge_names);
305 for (i = 0; i < cfg->n_bridges; i++) {
306 svec_add(&bridge_names, cfg->bridges[i]->name);
308 svec_sort(&bridge_names);
310 svec_init(&xfif_names);
311 svec_init(&xfif_types);
312 xf_enumerate_types(&xfif_types);
313 for (i = 0; i < xfif_types.n; i++) {
318 xf_enumerate_names(xfif_types.names[i], &xfif_names);
320 for (j = 0; j < xfif_names.n; j++) {
321 retval = xfif_open(xfif_names.names[j], xfif_types.names[i], &xfif);
323 struct svec all_names;
326 svec_init(&all_names);
327 xfif_get_all_names(xfif, &all_names);
328 for (k = 0; k < all_names.n; k++) {
329 if (svec_contains(&bridge_names, all_names.names[k])) {
335 svec_destroy(&all_names);
340 svec_destroy(&xfif_names);
341 svec_destroy(&xfif_types);
343 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
347 bridge_reconfigure(cfg);
352 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
354 /* XXX SSL should be configurable on a per-bridge basis. */
356 stream_ssl_set_private_key_file(ssl->private_key);
357 stream_ssl_set_certificate_file(ssl->certificate);
358 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
363 /* Attempt to create the network device 'iface_name' through the netdev
366 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
369 struct shash_node *node;
370 struct shash options;
374 shash_init(&options);
375 for (i = 0; i < iface_cfg->n_options; i++) {
376 shash_add(&options, iface_cfg->key_options[i],
377 xstrdup(iface_cfg->value_options[i]));
381 struct netdev_options netdev_options;
383 memset(&netdev_options, 0, sizeof netdev_options);
384 netdev_options.name = iface_cfg->name;
385 if (!strcmp(iface_cfg->type, "internal")) {
386 /* An "internal" config type maps to a netdev "system" type. */
387 netdev_options.type = "system";
389 netdev_options.type = iface_cfg->type;
391 netdev_options.args = &options;
392 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
393 netdev_options.may_create = true;
394 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
395 netdev_options.may_open = true;
398 error = netdev_open(&netdev_options, &iface->netdev);
401 netdev_get_carrier(iface->netdev, &iface->enabled);
403 } else if (iface->netdev) {
404 const char *netdev_type = netdev_get_type(iface->netdev);
405 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
406 ? iface_cfg->type : NULL;
408 /* An "internal" config type maps to a netdev "system" type. */
409 if (iface_type && !strcmp(iface_type, "internal")) {
410 iface_type = "system";
413 if (!iface_type || !strcmp(netdev_type, iface_type)) {
414 error = netdev_reconfigure(iface->netdev, &options);
416 VLOG_WARN("%s: attempting change device type from %s to %s",
417 iface_cfg->name, netdev_type, iface_type);
422 SHASH_FOR_EACH (node, &options) {
425 shash_destroy(&options);
431 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
433 return set_up_iface(iface_cfg, iface, false);
437 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
438 void *aux OVS_UNUSED)
440 if (!iface->netdev) {
441 int error = set_up_iface(iface->cfg, iface, true);
443 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
453 check_iface_xf_ifidx(struct bridge *br, struct iface *iface,
454 void *aux OVS_UNUSED)
456 if (iface->xf_ifidx >= 0) {
457 VLOG_DBG("%s has interface %s on port %d",
459 iface->name, iface->xf_ifidx);
462 VLOG_ERR("%s interface not in %s, dropping",
463 iface->name, xfif_name(br->xfif));
469 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
470 void *aux OVS_UNUSED)
472 /* Set policing attributes. */
473 netdev_set_policing(iface->netdev,
474 iface->cfg->ingress_policing_rate,
475 iface->cfg->ingress_policing_burst);
477 /* Set MAC address of internal interfaces other than the local
479 if (iface->xf_ifidx != XFLOWP_LOCAL
480 && iface_is_internal(br, iface->name)) {
481 iface_set_mac(iface);
487 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
488 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
489 * deletes from 'br' any ports that no longer have any interfaces. */
491 iterate_and_prune_ifaces(struct bridge *br,
492 bool (*cb)(struct bridge *, struct iface *,
498 for (i = 0; i < br->n_ports; ) {
499 struct port *port = br->ports[i];
500 for (j = 0; j < port->n_ifaces; ) {
501 struct iface *iface = port->ifaces[j];
502 if (cb(br, iface, aux)) {
505 iface_destroy(iface);
509 if (port->n_ifaces) {
512 VLOG_ERR("%s port has no interfaces, dropping", port->name);
519 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
521 struct ovsdb_idl_txn *txn;
522 struct shash old_br, new_br;
523 struct shash_node *node;
524 struct bridge *br, *next;
526 int sflow_bridge_number;
528 COVERAGE_INC(bridge_reconfigure);
530 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
532 /* Collect old and new bridges. */
535 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
536 shash_add(&old_br, br->name, br);
538 for (i = 0; i < ovs_cfg->n_bridges; i++) {
539 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
540 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
541 VLOG_WARN("more than one bridge named %s", br_cfg->name);
545 /* Get rid of deleted bridges and add new bridges. */
546 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
547 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
554 SHASH_FOR_EACH (node, &new_br) {
555 const char *br_name = node->name;
556 const struct ovsrec_bridge *br_cfg = node->data;
557 br = shash_find_data(&old_br, br_name);
559 /* If the bridge datapath type has changed, we need to tear it
560 * down and recreate. */
561 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
563 bridge_create(br_cfg);
566 bridge_create(br_cfg);
569 shash_destroy(&old_br);
570 shash_destroy(&new_br);
574 bridge_configure_ssl(ovs_cfg->ssl);
577 /* Reconfigure all bridges. */
578 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
579 bridge_reconfigure_one(ovs_cfg, br);
582 /* Add and delete ports on all datapaths.
584 * The kernel will reject any attempt to add a given port to a datapath if
585 * that port already belongs to a different datapath, so we must do all
586 * port deletions before any port additions. */
587 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
588 struct xflow_port *xfif_ports;
590 struct shash want_ifaces;
592 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
593 bridge_get_all_ifaces(br, &want_ifaces);
594 for (i = 0; i < n_xfif_ports; i++) {
595 const struct xflow_port *p = &xfif_ports[i];
596 if (!shash_find(&want_ifaces, p->devname)
597 && strcmp(p->devname, br->name)) {
598 int retval = xfif_port_del(br->xfif, p->port);
600 VLOG_ERR("failed to remove %s interface from %s: %s",
601 p->devname, xfif_name(br->xfif),
606 shash_destroy(&want_ifaces);
609 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
610 struct xflow_port *xfif_ports;
612 struct shash cur_ifaces, want_ifaces;
613 struct shash_node *node;
615 /* Get the set of interfaces currently in this datapath. */
616 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
617 shash_init(&cur_ifaces);
618 for (i = 0; i < n_xfif_ports; i++) {
619 const char *name = xfif_ports[i].devname;
620 if (!shash_find(&cur_ifaces, name)) {
621 shash_add(&cur_ifaces, name, NULL);
626 /* Get the set of interfaces we want on this datapath. */
627 bridge_get_all_ifaces(br, &want_ifaces);
629 SHASH_FOR_EACH (node, &want_ifaces) {
630 const char *if_name = node->name;
631 struct iface *iface = node->data;
633 if (shash_find(&cur_ifaces, if_name)) {
634 /* Already exists, just reconfigure it. */
636 reconfigure_iface(iface->cfg, iface);
639 /* Need to add to datapath. */
643 /* Add to datapath. */
644 internal = iface_is_internal(br, if_name);
645 error = xfif_port_add(br->xfif, if_name,
646 internal ? XFLOW_PORT_INTERNAL : 0, NULL);
647 if (error == EFBIG) {
648 VLOG_ERR("ran out of valid port numbers on %s",
649 xfif_name(br->xfif));
652 VLOG_ERR("failed to add %s interface to %s: %s",
653 if_name, xfif_name(br->xfif), strerror(error));
657 shash_destroy(&cur_ifaces);
658 shash_destroy(&want_ifaces);
660 sflow_bridge_number = 0;
661 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
664 struct iface *local_iface;
665 struct iface *hw_addr_iface;
668 bridge_fetch_dp_ifaces(br);
670 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
671 iterate_and_prune_ifaces(br, check_iface_xf_ifidx, NULL);
673 /* Pick local port hardware address, datapath ID. */
674 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
675 local_iface = bridge_get_local_iface(br);
677 int error = netdev_set_etheraddr(local_iface->netdev, ea);
679 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
680 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
681 "Ethernet address: %s",
682 br->name, strerror(error));
686 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
687 ofproto_set_datapath_id(br->ofproto, dpid);
689 dpid_string = xasprintf("%012"PRIx64, dpid);
690 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
693 /* Set NetFlow configuration on this bridge. */
694 if (br->cfg->netflow) {
695 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
696 struct netflow_options opts;
698 memset(&opts, 0, sizeof opts);
700 xfif_get_netflow_ids(br->xfif, &opts.engine_type, &opts.engine_id);
701 if (nf_cfg->engine_type) {
702 opts.engine_type = *nf_cfg->engine_type;
704 if (nf_cfg->engine_id) {
705 opts.engine_id = *nf_cfg->engine_id;
708 opts.active_timeout = nf_cfg->active_timeout;
709 if (!opts.active_timeout) {
710 opts.active_timeout = -1;
711 } else if (opts.active_timeout < 0) {
712 VLOG_WARN("bridge %s: active timeout interval set to negative "
713 "value, using default instead (%d seconds)", br->name,
714 NF_ACTIVE_TIMEOUT_DEFAULT);
715 opts.active_timeout = -1;
718 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
719 if (opts.add_id_to_iface) {
720 if (opts.engine_id > 0x7f) {
721 VLOG_WARN("bridge %s: netflow port mangling may conflict "
722 "with another vswitch, choose an engine id less "
723 "than 128", br->name);
725 if (br->n_ports > 508) {
726 VLOG_WARN("bridge %s: netflow port mangling will conflict "
727 "with another port when more than 508 ports are "
732 opts.collectors.n = nf_cfg->n_targets;
733 opts.collectors.names = nf_cfg->targets;
734 if (ofproto_set_netflow(br->ofproto, &opts)) {
735 VLOG_ERR("bridge %s: problem setting netflow collectors",
739 ofproto_set_netflow(br->ofproto, NULL);
742 /* Set sFlow configuration on this bridge. */
743 if (br->cfg->sflow) {
744 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
745 const struct ovsrec_controller *ctrl;
746 struct ofproto_sflow_options oso;
748 memset(&oso, 0, sizeof oso);
750 oso.targets.n = sflow_cfg->n_targets;
751 oso.targets.names = sflow_cfg->targets;
753 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
754 if (sflow_cfg->sampling) {
755 oso.sampling_rate = *sflow_cfg->sampling;
758 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
759 if (sflow_cfg->polling) {
760 oso.polling_interval = *sflow_cfg->polling;
763 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
764 if (sflow_cfg->header) {
765 oso.header_len = *sflow_cfg->header;
768 oso.sub_id = sflow_bridge_number++;
769 oso.agent_device = sflow_cfg->agent;
771 ctrl = bridge_get_controller(ovs_cfg, br);
772 oso.control_ip = ctrl ? ctrl->local_ip : NULL;
773 ofproto_set_sflow(br->ofproto, &oso);
775 svec_destroy(&oso.targets);
777 ofproto_set_sflow(br->ofproto, NULL);
780 /* Update the controller and related settings. It would be more
781 * straightforward to call this from bridge_reconfigure_one(), but we
782 * can't do it there for two reasons. First, and most importantly, at
783 * that point we don't know the xf_ifidx of any interfaces that have
784 * been added to the bridge (because we haven't actually added them to
785 * the datapath). Second, at that point we haven't set the datapath ID
786 * yet; when a controller is configured, resetting the datapath ID will
787 * immediately disconnect from the controller, so it's better to set
788 * the datapath ID before the controller. */
789 bridge_reconfigure_controller(ovs_cfg, br);
791 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
792 for (i = 0; i < br->n_ports; i++) {
793 struct port *port = br->ports[i];
795 port_update_vlan_compat(port);
796 port_update_bonding(port);
799 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
800 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
803 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
805 ovsdb_idl_txn_commit(txn);
806 ovsdb_idl_txn_destroy(txn); /* XXX */
810 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
814 for (i = 0; i < br_cfg->n_other_config; i++) {
815 if (!strcmp(br_cfg->key_other_config[i], key)) {
816 return br_cfg->value_other_config[i];
823 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
824 struct iface **hw_addr_iface)
830 *hw_addr_iface = NULL;
832 /* Did the user request a particular MAC? */
833 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
834 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
835 if (eth_addr_is_multicast(ea)) {
836 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
837 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
838 } else if (eth_addr_is_zero(ea)) {
839 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
845 /* Otherwise choose the minimum non-local MAC address among all of the
847 memset(ea, 0xff, sizeof ea);
848 for (i = 0; i < br->n_ports; i++) {
849 struct port *port = br->ports[i];
850 uint8_t iface_ea[ETH_ADDR_LEN];
853 /* Mirror output ports don't participate. */
854 if (port->is_mirror_output_port) {
858 /* Choose the MAC address to represent the port. */
859 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
860 /* Find the interface with this Ethernet address (if any) so that
861 * we can provide the correct devname to the caller. */
863 for (j = 0; j < port->n_ifaces; j++) {
864 struct iface *candidate = port->ifaces[j];
865 uint8_t candidate_ea[ETH_ADDR_LEN];
866 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
867 && eth_addr_equals(iface_ea, candidate_ea)) {
872 /* Choose the interface whose MAC address will represent the port.
873 * The Linux kernel bonding code always chooses the MAC address of
874 * the first slave added to a bond, and the Fedora networking
875 * scripts always add slaves to a bond in alphabetical order, so
876 * for compatibility we choose the interface with the name that is
877 * first in alphabetical order. */
878 iface = port->ifaces[0];
879 for (j = 1; j < port->n_ifaces; j++) {
880 struct iface *candidate = port->ifaces[j];
881 if (strcmp(candidate->name, iface->name) < 0) {
886 /* The local port doesn't count (since we're trying to choose its
887 * MAC address anyway). */
888 if (iface->xf_ifidx == XFLOWP_LOCAL) {
893 error = netdev_get_etheraddr(iface->netdev, iface_ea);
895 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
896 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
897 iface->name, strerror(error));
902 /* Compare against our current choice. */
903 if (!eth_addr_is_multicast(iface_ea) &&
904 !eth_addr_is_local(iface_ea) &&
905 !eth_addr_is_reserved(iface_ea) &&
906 !eth_addr_is_zero(iface_ea) &&
907 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
909 memcpy(ea, iface_ea, ETH_ADDR_LEN);
910 *hw_addr_iface = iface;
913 if (eth_addr_is_multicast(ea)) {
914 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
915 *hw_addr_iface = NULL;
916 VLOG_WARN("bridge %s: using default bridge Ethernet "
917 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
919 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
920 br->name, ETH_ADDR_ARGS(ea));
924 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
925 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
926 * an interface on 'br', then that interface must be passed in as
927 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
928 * 'hw_addr_iface' must be passed in as a null pointer. */
930 bridge_pick_datapath_id(struct bridge *br,
931 const uint8_t bridge_ea[ETH_ADDR_LEN],
932 struct iface *hw_addr_iface)
935 * The procedure for choosing a bridge MAC address will, in the most
936 * ordinary case, also choose a unique MAC that we can use as a datapath
937 * ID. In some special cases, though, multiple bridges will end up with
938 * the same MAC address. This is OK for the bridges, but it will confuse
939 * the OpenFlow controller, because each datapath needs a unique datapath
942 * Datapath IDs must be unique. It is also very desirable that they be
943 * stable from one run to the next, so that policy set on a datapath
946 const char *datapath_id;
949 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
950 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
956 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
958 * A bridge whose MAC address is taken from a VLAN network device
959 * (that is, a network device created with vconfig(8) or similar
960 * tool) will have the same MAC address as a bridge on the VLAN
961 * device's physical network device.
963 * Handle this case by hashing the physical network device MAC
964 * along with the VLAN identifier.
966 uint8_t buf[ETH_ADDR_LEN + 2];
967 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
968 buf[ETH_ADDR_LEN] = vlan >> 8;
969 buf[ETH_ADDR_LEN + 1] = vlan;
970 return dpid_from_hash(buf, sizeof buf);
973 * Assume that this bridge's MAC address is unique, since it
974 * doesn't fit any of the cases we handle specially.
979 * A purely internal bridge, that is, one that has no non-virtual
980 * network devices on it at all, is more difficult because it has no
981 * natural unique identifier at all.
983 * When the host is a XenServer, we handle this case by hashing the
984 * host's UUID with the name of the bridge. Names of bridges are
985 * persistent across XenServer reboots, although they can be reused if
986 * an internal network is destroyed and then a new one is later
987 * created, so this is fairly effective.
989 * When the host is not a XenServer, we punt by using a random MAC
990 * address on each run.
992 const char *host_uuid = xenserver_get_host_uuid();
994 char *combined = xasprintf("%s,%s", host_uuid, br->name);
995 dpid = dpid_from_hash(combined, strlen(combined));
1001 return eth_addr_to_uint64(bridge_ea);
1005 dpid_from_hash(const void *data, size_t n)
1007 uint8_t hash[SHA1_DIGEST_SIZE];
1009 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1010 sha1_bytes(data, n, hash);
1011 eth_addr_mark_random(hash);
1012 return eth_addr_to_uint64(hash);
1018 struct bridge *br, *next;
1022 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1023 int error = bridge_run_one(br);
1025 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1026 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1027 "forcing reconfiguration", br->name);
1041 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1042 ofproto_wait(br->ofproto);
1043 if (br->controller) {
1047 mac_learning_wait(br->ml);
1052 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1053 * configuration changes. */
1055 bridge_flush(struct bridge *br)
1057 COVERAGE_INC(bridge_flush);
1059 mac_learning_flush(br->ml);
1062 /* Returns the 'br' interface for the XFLOWP_LOCAL port, or null if 'br' has no
1063 * such interface. */
1064 static struct iface *
1065 bridge_get_local_iface(struct bridge *br)
1069 for (i = 0; i < br->n_ports; i++) {
1070 struct port *port = br->ports[i];
1071 for (j = 0; j < port->n_ifaces; j++) {
1072 struct iface *iface = port->ifaces[j];
1073 if (iface->xf_ifidx == XFLOWP_LOCAL) {
1082 /* Bridge unixctl user interface functions. */
1084 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1085 const char *args, void *aux OVS_UNUSED)
1087 struct ds ds = DS_EMPTY_INITIALIZER;
1088 const struct bridge *br;
1089 const struct mac_entry *e;
1091 br = bridge_lookup(args);
1093 unixctl_command_reply(conn, 501, "no such bridge");
1097 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1098 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1099 if (e->port < 0 || e->port >= br->n_ports) {
1102 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1103 br->ports[e->port]->ifaces[0]->xf_ifidx,
1104 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1106 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1110 /* Bridge reconfiguration functions. */
1111 static struct bridge *
1112 bridge_create(const struct ovsrec_bridge *br_cfg)
1117 assert(!bridge_lookup(br_cfg->name));
1118 br = xzalloc(sizeof *br);
1120 error = xfif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1126 xfif_flow_flush(br->xfif);
1128 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1131 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1133 xfif_delete(br->xfif);
1134 xfif_close(br->xfif);
1139 br->name = xstrdup(br_cfg->name);
1141 br->ml = mac_learning_create();
1142 br->sent_config_request = false;
1143 eth_addr_nicira_random(br->default_ea);
1145 port_array_init(&br->ifaces);
1148 br->bond_next_rebalance = time_msec() + 10000;
1150 list_push_back(&all_bridges, &br->node);
1152 VLOG_INFO("created bridge %s on %s", br->name, xfif_name(br->xfif));
1158 bridge_destroy(struct bridge *br)
1163 while (br->n_ports > 0) {
1164 port_destroy(br->ports[br->n_ports - 1]);
1166 list_remove(&br->node);
1167 error = xfif_delete(br->xfif);
1168 if (error && error != ENOENT) {
1169 VLOG_ERR("failed to delete %s: %s",
1170 xfif_name(br->xfif), strerror(error));
1172 xfif_close(br->xfif);
1173 ofproto_destroy(br->ofproto);
1174 free(br->controller);
1175 mac_learning_destroy(br->ml);
1176 port_array_destroy(&br->ifaces);
1183 static struct bridge *
1184 bridge_lookup(const char *name)
1188 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1189 if (!strcmp(br->name, name)) {
1197 bridge_exists(const char *name)
1199 return bridge_lookup(name) ? true : false;
1203 bridge_get_datapathid(const char *name)
1205 struct bridge *br = bridge_lookup(name);
1206 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1209 /* Handle requests for a listing of all flows known by the OpenFlow
1210 * stack, including those normally hidden. */
1212 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1213 const char *args, void *aux OVS_UNUSED)
1218 br = bridge_lookup(args);
1220 unixctl_command_reply(conn, 501, "Unknown bridge");
1225 ofproto_get_all_flows(br->ofproto, &results);
1227 unixctl_command_reply(conn, 200, ds_cstr(&results));
1228 ds_destroy(&results);
1232 bridge_run_one(struct bridge *br)
1236 error = ofproto_run1(br->ofproto);
1241 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1244 error = ofproto_run2(br->ofproto, br->flush);
1250 static const struct ovsrec_controller *
1251 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1252 const struct bridge *br)
1254 const struct ovsrec_controller *controller;
1256 controller = (br->cfg->controller ? br->cfg->controller
1257 : ovs_cfg->controller ? ovs_cfg->controller
1260 if (controller && !strcmp(controller->target, "none")) {
1268 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1270 struct svec *ifaces = ifaces_;
1271 if (!svec_contains(ifaces, iface->name)) {
1272 svec_add(ifaces, iface->name);
1276 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1278 br->name, iface->name, iface->port->name);
1284 bridge_update_desc(struct bridge *br)
1287 bool changed = false;
1290 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1291 if (desc != br->mfr_desc) {
1294 br->mfr_desc = xstrdup(desc);
1296 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1301 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1302 if (desc != br->hw_desc) {
1305 br->hw_desc = xstrdup(desc);
1307 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1312 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1313 if (desc != br->sw_desc) {
1316 br->sw_desc = xstrdup(desc);
1318 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1323 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1324 if (desc != br->serial_desc) {
1325 free(br->serial_desc);
1327 br->serial_desc = xstrdup(desc);
1329 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1334 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1335 if (desc != br->dp_desc) {
1338 br->dp_desc = xstrdup(desc);
1340 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1346 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1347 br->sw_desc, br->serial_desc, br->dp_desc);
1353 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1356 struct shash old_ports, new_ports;
1358 struct svec listeners, old_listeners;
1359 struct svec snoops, old_snoops;
1360 struct shash_node *node;
1363 /* Collect old ports. */
1364 shash_init(&old_ports);
1365 for (i = 0; i < br->n_ports; i++) {
1366 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1369 /* Collect new ports. */
1370 shash_init(&new_ports);
1371 for (i = 0; i < br->cfg->n_ports; i++) {
1372 const char *name = br->cfg->ports[i]->name;
1373 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1374 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1379 /* If we have a controller, then we need a local port. Complain if the
1380 * user didn't specify one.
1382 * XXX perhaps we should synthesize a port ourselves in this case. */
1383 if (bridge_get_controller(ovs_cfg, br)) {
1384 char local_name[IF_NAMESIZE];
1387 error = xfif_port_get_name(br->xfif, XFLOWP_LOCAL,
1388 local_name, sizeof local_name);
1389 if (!error && !shash_find(&new_ports, local_name)) {
1390 VLOG_WARN("bridge %s: controller specified but no local port "
1391 "(port named %s) defined",
1392 br->name, local_name);
1396 /* Get rid of deleted ports and add new ports. */
1397 SHASH_FOR_EACH (node, &old_ports) {
1398 if (!shash_find(&new_ports, node->name)) {
1399 port_destroy(node->data);
1402 SHASH_FOR_EACH (node, &new_ports) {
1403 struct port *port = shash_find_data(&old_ports, node->name);
1405 port = port_create(br, node->name);
1407 port_reconfigure(port, node->data);
1409 shash_destroy(&old_ports);
1410 shash_destroy(&new_ports);
1412 /* Check and delete duplicate interfaces. */
1414 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1415 svec_destroy(&ifaces);
1417 /* Delete all flows if we're switching from connected to standalone or vice
1418 * versa. (XXX Should we delete all flows if we are switching from one
1419 * controller to another?) */
1422 /* Configure OpenFlow management listeners. */
1423 svec_init(&listeners);
1424 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1426 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1427 ovs_rundir, br->name));
1428 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1429 svec_clear(&listeners);
1431 svec_sort_unique(&listeners);
1433 svec_init(&old_listeners);
1434 ofproto_get_listeners(br->ofproto, &old_listeners);
1435 svec_sort_unique(&old_listeners);
1437 if (!svec_equal(&listeners, &old_listeners)) {
1438 ofproto_set_listeners(br->ofproto, &listeners);
1440 svec_destroy(&listeners);
1441 svec_destroy(&old_listeners);
1443 /* Configure OpenFlow controller connection snooping. */
1445 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1447 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1448 ovs_rundir, br->name));
1449 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1450 svec_clear(&snoops);
1452 svec_sort_unique(&snoops);
1454 svec_init(&old_snoops);
1455 ofproto_get_snoops(br->ofproto, &old_snoops);
1456 svec_sort_unique(&old_snoops);
1458 if (!svec_equal(&snoops, &old_snoops)) {
1459 ofproto_set_snoops(br->ofproto, &snoops);
1461 svec_destroy(&snoops);
1462 svec_destroy(&old_snoops);
1464 /* Default listener. */
1465 svec_init(&listeners);
1466 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1467 ovs_rundir, br->name));
1468 svec_init(&old_listeners);
1469 ofproto_get_listeners(br->ofproto, &old_listeners);
1470 if (!svec_equal(&listeners, &old_listeners)) {
1471 ofproto_set_listeners(br->ofproto, &listeners);
1473 svec_destroy(&listeners);
1474 svec_destroy(&old_listeners);
1476 /* Default snoop. */
1478 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1479 ovs_rundir, br->name));
1480 svec_init(&old_snoops);
1481 ofproto_get_snoops(br->ofproto, &old_snoops);
1482 if (!svec_equal(&snoops, &old_snoops)) {
1483 ofproto_set_snoops(br->ofproto, &snoops);
1485 svec_destroy(&snoops);
1486 svec_destroy(&old_snoops);
1489 mirror_reconfigure(br);
1491 bridge_update_desc(br);
1495 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1498 const struct ovsrec_controller *c;
1500 c = bridge_get_controller(ovs_cfg, br);
1501 if ((br->controller != NULL) != (c != NULL)) {
1502 ofproto_flush_flows(br->ofproto);
1504 free(br->controller);
1505 br->controller = c ? xstrdup(c->target) : NULL;
1508 int max_backoff, probe;
1509 int rate_limit, burst_limit;
1511 if (!strcmp(c->target, "discover")) {
1512 ofproto_set_discovery(br->ofproto, true,
1513 c->discover_accept_regex,
1514 c->discover_update_resolv_conf);
1516 struct iface *local_iface;
1520 in_band = (!c->connection_mode
1521 || !strcmp(c->connection_mode, "out-of-band"));
1522 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1523 ofproto_set_in_band(br->ofproto, in_band);
1525 local_iface = bridge_get_local_iface(br);
1526 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1527 struct netdev *netdev = local_iface->netdev;
1528 struct in_addr mask, gateway;
1530 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1533 if (!c->local_gateway
1534 || !inet_aton(c->local_gateway, &gateway)) {
1538 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1540 mask.s_addr = guess_netmask(ip.s_addr);
1542 if (!netdev_set_in4(netdev, ip, mask)) {
1543 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1545 br->name, IP_ARGS(&ip.s_addr),
1546 IP_ARGS(&mask.s_addr));
1549 if (gateway.s_addr) {
1550 if (!netdev_add_router(netdev, gateway)) {
1551 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1552 br->name, IP_ARGS(&gateway.s_addr));
1558 ofproto_set_failure(br->ofproto,
1560 || !strcmp(c->fail_mode, "standalone")
1561 || !strcmp(c->fail_mode, "open")));
1563 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1564 ofproto_set_probe_interval(br->ofproto, probe);
1566 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1567 ofproto_set_max_backoff(br->ofproto, max_backoff);
1569 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1570 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1571 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1573 union ofp_action action;
1576 /* Set up a flow that matches every packet and directs them to
1577 * OFPP_NORMAL (which goes to us). */
1578 memset(&action, 0, sizeof action);
1579 action.type = htons(OFPAT_OUTPUT);
1580 action.output.len = htons(sizeof action);
1581 action.output.port = htons(OFPP_NORMAL);
1582 memset(&flow, 0, sizeof flow);
1583 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1586 ofproto_set_in_band(br->ofproto, false);
1587 ofproto_set_max_backoff(br->ofproto, 1);
1588 ofproto_set_probe_interval(br->ofproto, 5);
1589 ofproto_set_failure(br->ofproto, false);
1592 ofproto_set_controller(br->ofproto, br->controller);
1596 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1601 for (i = 0; i < br->n_ports; i++) {
1602 struct port *port = br->ports[i];
1603 for (j = 0; j < port->n_ifaces; j++) {
1604 struct iface *iface = port->ifaces[j];
1605 shash_add_once(ifaces, iface->name, iface);
1607 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1608 shash_add_once(ifaces, port->name, NULL);
1613 /* For robustness, in case the administrator moves around datapath ports behind
1614 * our back, we re-check all the datapath port numbers here.
1616 * This function will set the 'xf_ifidx' members of interfaces that have
1617 * disappeared to -1, so only call this function from a context where those
1618 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1619 * 'xf_ifidx'es will cause trouble later when we try to send them to the
1620 * datapath, which doesn't support UINT16_MAX+1 ports. */
1622 bridge_fetch_dp_ifaces(struct bridge *br)
1624 struct xflow_port *xfif_ports;
1625 size_t n_xfif_ports;
1628 /* Reset all interface numbers. */
1629 for (i = 0; i < br->n_ports; i++) {
1630 struct port *port = br->ports[i];
1631 for (j = 0; j < port->n_ifaces; j++) {
1632 struct iface *iface = port->ifaces[j];
1633 iface->xf_ifidx = -1;
1636 port_array_clear(&br->ifaces);
1638 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
1639 for (i = 0; i < n_xfif_ports; i++) {
1640 struct xflow_port *p = &xfif_ports[i];
1641 struct iface *iface = iface_lookup(br, p->devname);
1643 if (iface->xf_ifidx >= 0) {
1644 VLOG_WARN("%s reported interface %s twice",
1645 xfif_name(br->xfif), p->devname);
1646 } else if (iface_from_xf_ifidx(br, p->port)) {
1647 VLOG_WARN("%s reported interface %"PRIu16" twice",
1648 xfif_name(br->xfif), p->port);
1650 port_array_set(&br->ifaces, p->port, iface);
1651 iface->xf_ifidx = p->port;
1655 int64_t ofport = (iface->xf_ifidx >= 0
1656 ? xflow_port_to_ofp_port(iface->xf_ifidx)
1658 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1665 /* Bridge packet processing functions. */
1668 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1670 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1673 static struct bond_entry *
1674 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1676 return &port->bond_hash[bond_hash(mac)];
1680 bond_choose_iface(const struct port *port)
1682 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1683 size_t i, best_down_slave = -1;
1684 long long next_delay_expiration = LLONG_MAX;
1686 for (i = 0; i < port->n_ifaces; i++) {
1687 struct iface *iface = port->ifaces[i];
1689 if (iface->enabled) {
1691 } else if (iface->delay_expires < next_delay_expiration) {
1692 best_down_slave = i;
1693 next_delay_expiration = iface->delay_expires;
1697 if (best_down_slave != -1) {
1698 struct iface *iface = port->ifaces[best_down_slave];
1700 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1701 "since no other interface is up", iface->name,
1702 iface->delay_expires - time_msec());
1703 bond_enable_slave(iface, true);
1706 return best_down_slave;
1710 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1711 uint16_t *xf_ifidx, tag_type *tags)
1713 struct iface *iface;
1715 assert(port->n_ifaces);
1716 if (port->n_ifaces == 1) {
1717 iface = port->ifaces[0];
1719 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1720 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1721 || !port->ifaces[e->iface_idx]->enabled) {
1722 /* XXX select interface properly. The current interface selection
1723 * is only good for testing the rebalancing code. */
1724 e->iface_idx = bond_choose_iface(port);
1725 if (e->iface_idx < 0) {
1726 *tags |= port->no_ifaces_tag;
1729 e->iface_tag = tag_create_random();
1730 ((struct port *) port)->bond_compat_is_stale = true;
1732 *tags |= e->iface_tag;
1733 iface = port->ifaces[e->iface_idx];
1735 *xf_ifidx = iface->xf_ifidx;
1736 *tags |= iface->tag; /* Currently only used for bonding. */
1741 bond_link_status_update(struct iface *iface, bool carrier)
1743 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1744 struct port *port = iface->port;
1746 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1747 /* Nothing to do. */
1750 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1751 iface->name, carrier ? "detected" : "dropped");
1752 if (carrier == iface->enabled) {
1753 iface->delay_expires = LLONG_MAX;
1754 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1755 iface->name, carrier ? "disabled" : "enabled");
1756 } else if (carrier && port->active_iface < 0) {
1757 bond_enable_slave(iface, true);
1758 if (port->updelay) {
1759 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1760 "other interface is up", iface->name, port->updelay);
1763 int delay = carrier ? port->updelay : port->downdelay;
1764 iface->delay_expires = time_msec() + delay;
1767 "interface %s: will be %s if it stays %s for %d ms",
1769 carrier ? "enabled" : "disabled",
1770 carrier ? "up" : "down",
1777 bond_choose_active_iface(struct port *port)
1779 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1781 port->active_iface = bond_choose_iface(port);
1782 port->active_iface_tag = tag_create_random();
1783 if (port->active_iface >= 0) {
1784 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1785 port->name, port->ifaces[port->active_iface]->name);
1787 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1793 bond_enable_slave(struct iface *iface, bool enable)
1795 struct port *port = iface->port;
1796 struct bridge *br = port->bridge;
1798 /* This acts as a recursion check. If the act of disabling a slave
1799 * causes a different slave to be enabled, the flag will allow us to
1800 * skip redundant work when we reenter this function. It must be
1801 * cleared on exit to keep things safe with multiple bonds. */
1802 static bool moving_active_iface = false;
1804 iface->delay_expires = LLONG_MAX;
1805 if (enable == iface->enabled) {
1809 iface->enabled = enable;
1810 if (!iface->enabled) {
1811 VLOG_WARN("interface %s: disabled", iface->name);
1812 ofproto_revalidate(br->ofproto, iface->tag);
1813 if (iface->port_ifidx == port->active_iface) {
1814 ofproto_revalidate(br->ofproto,
1815 port->active_iface_tag);
1817 /* Disabling a slave can lead to another slave being immediately
1818 * enabled if there will be no active slaves but one is waiting
1819 * on an updelay. In this case we do not need to run most of the
1820 * code for the newly enabled slave since there was no period
1821 * without an active slave and it is redundant with the disabling
1823 moving_active_iface = true;
1824 bond_choose_active_iface(port);
1826 bond_send_learning_packets(port);
1828 VLOG_WARN("interface %s: enabled", iface->name);
1829 if (port->active_iface < 0 && !moving_active_iface) {
1830 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1831 bond_choose_active_iface(port);
1832 bond_send_learning_packets(port);
1834 iface->tag = tag_create_random();
1837 moving_active_iface = false;
1838 port->bond_compat_is_stale = true;
1842 bond_run(struct bridge *br)
1846 for (i = 0; i < br->n_ports; i++) {
1847 struct port *port = br->ports[i];
1849 if (port->n_ifaces >= 2) {
1850 for (j = 0; j < port->n_ifaces; j++) {
1851 struct iface *iface = port->ifaces[j];
1852 if (time_msec() >= iface->delay_expires) {
1853 bond_enable_slave(iface, !iface->enabled);
1858 if (port->bond_compat_is_stale) {
1859 port->bond_compat_is_stale = false;
1860 port_update_bond_compat(port);
1866 bond_wait(struct bridge *br)
1870 for (i = 0; i < br->n_ports; i++) {
1871 struct port *port = br->ports[i];
1872 if (port->n_ifaces < 2) {
1875 for (j = 0; j < port->n_ifaces; j++) {
1876 struct iface *iface = port->ifaces[j];
1877 if (iface->delay_expires != LLONG_MAX) {
1878 poll_timer_wait(iface->delay_expires - time_msec());
1885 set_dst(struct dst *p, const flow_t *flow,
1886 const struct port *in_port, const struct port *out_port,
1889 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1890 : in_port->vlan >= 0 ? in_port->vlan
1891 : ntohs(flow->dl_vlan));
1892 return choose_output_iface(out_port, flow->dl_src, &p->xf_ifidx, tags);
1896 swap_dst(struct dst *p, struct dst *q)
1898 struct dst tmp = *p;
1903 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1904 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1905 * that we push to the datapath. We could in fact fully sort the array by
1906 * vlan, but in most cases there are at most two different vlan tags so that's
1907 * possibly overkill.) */
1909 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1911 struct dst *first = dsts;
1912 struct dst *last = dsts + n_dsts;
1914 while (first != last) {
1916 * - All dsts < first have vlan == 'vlan'.
1917 * - All dsts >= last have vlan != 'vlan'.
1918 * - first < last. */
1919 while (first->vlan == vlan) {
1920 if (++first == last) {
1925 /* Same invariants, plus one additional:
1926 * - first->vlan != vlan.
1928 while (last[-1].vlan != vlan) {
1929 if (--last == first) {
1934 /* Same invariants, plus one additional:
1935 * - last[-1].vlan == vlan.*/
1936 swap_dst(first++, --last);
1941 mirror_mask_ffs(mirror_mask_t mask)
1943 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1948 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1949 const struct dst *test)
1952 for (i = 0; i < n_dsts; i++) {
1953 if (dsts[i].vlan == test->vlan && dsts[i].xf_ifidx == test->xf_ifidx) {
1961 port_trunks_vlan(const struct port *port, uint16_t vlan)
1963 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1967 port_includes_vlan(const struct port *port, uint16_t vlan)
1969 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1973 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1974 const struct port *in_port, const struct port *out_port,
1975 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1977 mirror_mask_t mirrors = in_port->src_mirrors;
1978 struct dst *dst = dsts;
1981 if (out_port == FLOOD_PORT) {
1982 /* XXX use XFLOW_FLOOD if no vlans or bonding. */
1983 /* XXX even better, define each VLAN as a datapath port group */
1984 for (i = 0; i < br->n_ports; i++) {
1985 struct port *port = br->ports[i];
1986 if (port != in_port && port_includes_vlan(port, vlan)
1987 && !port->is_mirror_output_port
1988 && set_dst(dst, flow, in_port, port, tags)) {
1989 mirrors |= port->dst_mirrors;
1993 *nf_output_iface = NF_OUT_FLOOD;
1994 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1995 *nf_output_iface = dst->xf_ifidx;
1996 mirrors |= out_port->dst_mirrors;
2001 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2002 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2004 if (set_dst(dst, flow, in_port, m->out_port, tags)
2005 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2009 for (i = 0; i < br->n_ports; i++) {
2010 struct port *port = br->ports[i];
2011 if (port_includes_vlan(port, m->out_vlan)
2012 && set_dst(dst, flow, in_port, port, tags))
2016 if (port->vlan < 0) {
2017 dst->vlan = m->out_vlan;
2019 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2023 /* Use the vlan tag on the original flow instead of
2024 * the one passed in the vlan parameter. This ensures
2025 * that we compare the vlan from before any implicit
2026 * tagging tags place. This is necessary because
2027 * dst->vlan is the final vlan, after removing implicit
2029 flow_vlan = ntohs(flow->dl_vlan);
2030 if (flow_vlan == 0) {
2031 flow_vlan = OFP_VLAN_NONE;
2033 if (port == in_port && dst->vlan == flow_vlan) {
2034 /* Don't send out input port on same VLAN. */
2042 mirrors &= mirrors - 1;
2045 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2049 static void OVS_UNUSED
2050 print_dsts(const struct dst *dsts, size_t n)
2052 for (; n--; dsts++) {
2053 printf(">p%"PRIu16, dsts->xf_ifidx);
2054 if (dsts->vlan != OFP_VLAN_NONE) {
2055 printf("v%"PRIu16, dsts->vlan);
2061 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2062 const struct port *in_port, const struct port *out_port,
2063 tag_type *tags, struct xflow_actions *actions,
2064 uint16_t *nf_output_iface)
2066 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2068 const struct dst *p;
2071 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2074 cur_vlan = ntohs(flow->dl_vlan);
2075 for (p = dsts; p < &dsts[n_dsts]; p++) {
2076 union xflow_action *a;
2077 if (p->vlan != cur_vlan) {
2078 if (p->vlan == OFP_VLAN_NONE) {
2079 xflow_actions_add(actions, XFLOWAT_STRIP_VLAN);
2081 a = xflow_actions_add(actions, XFLOWAT_SET_DL_TCI);
2082 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2083 a->dl_tci.mask = htons(VLAN_VID_MASK);
2087 a = xflow_actions_add(actions, XFLOWAT_OUTPUT);
2088 a->output.port = p->xf_ifidx;
2092 /* Returns the effective vlan of a packet, taking into account both the
2093 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2094 * the packet is untagged and -1 indicates it has an invalid header and
2095 * should be dropped. */
2096 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2097 struct port *in_port, bool have_packet)
2099 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2100 * belongs to VLAN 0, so we should treat both cases identically. (In the
2101 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2102 * presumably to allow a priority to be specified. In the latter case, the
2103 * packet does not have any 802.1Q header.) */
2104 int vlan = ntohs(flow->dl_vlan);
2105 if (vlan == OFP_VLAN_NONE) {
2108 if (in_port->vlan >= 0) {
2110 /* XXX support double tagging? */
2112 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2113 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2114 "packet received on port %s configured with "
2115 "implicit VLAN %"PRIu16,
2116 br->name, ntohs(flow->dl_vlan),
2117 in_port->name, in_port->vlan);
2121 vlan = in_port->vlan;
2123 if (!port_includes_vlan(in_port, vlan)) {
2125 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2126 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2127 "packet received on port %s not configured for "
2129 br->name, vlan, in_port->name, vlan);
2139 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2140 struct port *in_port)
2142 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2143 vlan, in_port->port_idx);
2145 /* The log messages here could actually be useful in debugging,
2146 * so keep the rate limit relatively high. */
2147 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2149 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2150 "on port %s in VLAN %d",
2151 br->name, ETH_ADDR_ARGS(flow->dl_src),
2152 in_port->name, vlan);
2153 ofproto_revalidate(br->ofproto, rev_tag);
2158 is_bcast_arp_reply(const flow_t *flow)
2160 return (flow->dl_type == htons(ETH_TYPE_ARP)
2161 && flow->nw_proto == ARP_OP_REPLY
2162 && eth_addr_is_broadcast(flow->dl_dst));
2165 /* If the composed actions may be applied to any packet in the given 'flow',
2166 * returns true. Otherwise, the actions should only be applied to 'packet', or
2167 * not at all, if 'packet' was NULL. */
2169 process_flow(struct bridge *br, const flow_t *flow,
2170 const struct ofpbuf *packet, struct xflow_actions *actions,
2171 tag_type *tags, uint16_t *nf_output_iface)
2173 struct iface *in_iface;
2174 struct port *in_port;
2175 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2179 /* Find the interface and port structure for the received packet. */
2180 in_iface = iface_from_xf_ifidx(br, flow->in_port);
2182 /* No interface? Something fishy... */
2183 if (packet != NULL) {
2184 /* Odd. A few possible reasons here:
2186 * - We deleted an interface but there are still a few packets
2187 * queued up from it.
2189 * - Someone externally added an interface (e.g. with "ovs-dpctl
2190 * add-if") that we don't know about.
2192 * - Packet arrived on the local port but the local port is not
2193 * one of our bridge ports.
2195 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2197 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2198 "interface %"PRIu16, br->name, flow->in_port);
2201 /* Return without adding any actions, to drop packets on this flow. */
2204 in_port = in_iface->port;
2205 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2210 /* Drop frames for reserved multicast addresses. */
2211 if (eth_addr_is_reserved(flow->dl_dst)) {
2215 /* Drop frames on ports reserved for mirroring. */
2216 if (in_port->is_mirror_output_port) {
2217 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2218 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2219 "which is reserved exclusively for mirroring",
2220 br->name, in_port->name);
2224 /* Packets received on bonds need special attention to avoid duplicates. */
2225 if (in_port->n_ifaces > 1) {
2228 if (eth_addr_is_multicast(flow->dl_dst)) {
2229 *tags |= in_port->active_iface_tag;
2230 if (in_port->active_iface != in_iface->port_ifidx) {
2231 /* Drop all multicast packets on inactive slaves. */
2236 /* Drop all packets for which we have learned a different input
2237 * port, because we probably sent the packet on one slave and got
2238 * it back on the other. Broadcast ARP replies are an exception
2239 * to this rule: the host has moved to another switch. */
2240 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2241 if (src_idx != -1 && src_idx != in_port->port_idx &&
2242 !is_bcast_arp_reply(flow)) {
2248 out_port = FLOOD_PORT;
2249 /* Learn source MAC (but don't try to learn from revalidation). */
2251 update_learning_table(br, flow, vlan, in_port);
2254 /* Determine output port. */
2255 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2257 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2258 out_port = br->ports[out_port_idx];
2259 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2260 /* If we are revalidating but don't have a learning entry then
2261 * eject the flow. Installing a flow that floods packets opens
2262 * up a window of time where we could learn from a packet reflected
2263 * on a bond and blackhole packets before the learning table is
2264 * updated to reflect the correct port. */
2268 /* Don't send packets out their input ports. */
2269 if (in_port == out_port) {
2274 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2280 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2283 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2284 const struct ofp_phy_port *opp,
2287 struct bridge *br = br_;
2288 struct iface *iface;
2291 iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(opp->port_no));
2297 if (reason == OFPPR_DELETE) {
2298 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2299 br->name, iface->name);
2300 iface_destroy(iface);
2301 if (!port->n_ifaces) {
2302 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2303 br->name, port->name);
2309 if (port->n_ifaces > 1) {
2310 bool up = !(opp->state & OFPPS_LINK_DOWN);
2311 bond_link_status_update(iface, up);
2312 port_update_bond_compat(port);
2318 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2319 struct xflow_actions *actions, tag_type *tags,
2320 uint16_t *nf_output_iface, void *br_)
2322 struct bridge *br = br_;
2324 COVERAGE_INC(bridge_process_flow);
2325 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2329 bridge_account_flow_ofhook_cb(const flow_t *flow,
2330 const union xflow_action *actions,
2331 size_t n_actions, unsigned long long int n_bytes,
2334 struct bridge *br = br_;
2335 struct port *in_port;
2336 const union xflow_action *a;
2338 /* Feed information from the active flows back into the learning table
2339 * to ensure that table is always in sync with what is actually flowing
2340 * through the datapath. */
2341 in_port = port_from_xf_ifidx(br, flow->in_port);
2343 int vlan = flow_get_vlan(br, flow, in_port, false);
2345 update_learning_table(br, flow, vlan, in_port);
2349 if (!br->has_bonded_ports) {
2353 for (a = actions; a < &actions[n_actions]; a++) {
2354 if (a->type == XFLOWAT_OUTPUT) {
2355 struct port *out_port = port_from_xf_ifidx(br, a->output.port);
2356 if (out_port && out_port->n_ifaces >= 2) {
2357 struct bond_entry *e = lookup_bond_entry(out_port,
2359 e->tx_bytes += n_bytes;
2366 bridge_account_checkpoint_ofhook_cb(void *br_)
2368 struct bridge *br = br_;
2371 if (!br->has_bonded_ports) {
2375 /* The current ofproto implementation calls this callback at least once a
2376 * second, so this timer implementation is sufficient. */
2377 if (time_msec() < br->bond_next_rebalance) {
2380 br->bond_next_rebalance = time_msec() + 10000;
2382 for (i = 0; i < br->n_ports; i++) {
2383 struct port *port = br->ports[i];
2384 if (port->n_ifaces > 1) {
2385 bond_rebalance_port(port);
2390 static struct ofhooks bridge_ofhooks = {
2391 bridge_port_changed_ofhook_cb,
2392 bridge_normal_ofhook_cb,
2393 bridge_account_flow_ofhook_cb,
2394 bridge_account_checkpoint_ofhook_cb,
2397 /* Bonding functions. */
2399 /* Statistics for a single interface on a bonded port, used for load-based
2400 * bond rebalancing. */
2401 struct slave_balance {
2402 struct iface *iface; /* The interface. */
2403 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2405 /* All the "bond_entry"s that are assigned to this interface, in order of
2406 * increasing tx_bytes. */
2407 struct bond_entry **hashes;
2411 /* Sorts pointers to pointers to bond_entries in ascending order by the
2412 * interface to which they are assigned, and within a single interface in
2413 * ascending order of bytes transmitted. */
2415 compare_bond_entries(const void *a_, const void *b_)
2417 const struct bond_entry *const *ap = a_;
2418 const struct bond_entry *const *bp = b_;
2419 const struct bond_entry *a = *ap;
2420 const struct bond_entry *b = *bp;
2421 if (a->iface_idx != b->iface_idx) {
2422 return a->iface_idx > b->iface_idx ? 1 : -1;
2423 } else if (a->tx_bytes != b->tx_bytes) {
2424 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2430 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2431 * *descending* order by number of bytes transmitted. */
2433 compare_slave_balance(const void *a_, const void *b_)
2435 const struct slave_balance *a = a_;
2436 const struct slave_balance *b = b_;
2437 if (a->iface->enabled != b->iface->enabled) {
2438 return a->iface->enabled ? -1 : 1;
2439 } else if (a->tx_bytes != b->tx_bytes) {
2440 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2447 swap_bals(struct slave_balance *a, struct slave_balance *b)
2449 struct slave_balance tmp = *a;
2454 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2455 * given that 'p' (and only 'p') might be in the wrong location.
2457 * This function invalidates 'p', since it might now be in a different memory
2460 resort_bals(struct slave_balance *p,
2461 struct slave_balance bals[], size_t n_bals)
2464 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2465 swap_bals(p, p - 1);
2467 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2468 swap_bals(p, p + 1);
2474 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2476 if (VLOG_IS_DBG_ENABLED()) {
2477 struct ds ds = DS_EMPTY_INITIALIZER;
2478 const struct slave_balance *b;
2480 for (b = bals; b < bals + n_bals; b++) {
2484 ds_put_char(&ds, ',');
2486 ds_put_format(&ds, " %s %"PRIu64"kB",
2487 b->iface->name, b->tx_bytes / 1024);
2489 if (!b->iface->enabled) {
2490 ds_put_cstr(&ds, " (disabled)");
2492 if (b->n_hashes > 0) {
2493 ds_put_cstr(&ds, " (");
2494 for (i = 0; i < b->n_hashes; i++) {
2495 const struct bond_entry *e = b->hashes[i];
2497 ds_put_cstr(&ds, " + ");
2499 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2500 e - port->bond_hash, e->tx_bytes / 1024);
2502 ds_put_cstr(&ds, ")");
2505 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2510 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2512 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2515 struct bond_entry *hash = from->hashes[hash_idx];
2516 struct port *port = from->iface->port;
2517 uint64_t delta = hash->tx_bytes;
2519 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2520 "from %s to %s (now carrying %"PRIu64"kB and "
2521 "%"PRIu64"kB load, respectively)",
2522 port->name, delta / 1024, hash - port->bond_hash,
2523 from->iface->name, to->iface->name,
2524 (from->tx_bytes - delta) / 1024,
2525 (to->tx_bytes + delta) / 1024);
2527 /* Delete element from from->hashes.
2529 * We don't bother to add the element to to->hashes because not only would
2530 * it require more work, the only purpose it would be to allow that hash to
2531 * be migrated to another slave in this rebalancing run, and there is no
2532 * point in doing that. */
2533 if (hash_idx == 0) {
2536 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2537 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2541 /* Shift load away from 'from' to 'to'. */
2542 from->tx_bytes -= delta;
2543 to->tx_bytes += delta;
2545 /* Arrange for flows to be revalidated. */
2546 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2547 hash->iface_idx = to->iface->port_ifidx;
2548 hash->iface_tag = tag_create_random();
2552 bond_rebalance_port(struct port *port)
2554 struct slave_balance bals[DP_MAX_PORTS];
2556 struct bond_entry *hashes[BOND_MASK + 1];
2557 struct slave_balance *b, *from, *to;
2558 struct bond_entry *e;
2561 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2562 * descending order of tx_bytes, so that bals[0] represents the most
2563 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2566 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2567 * array for each slave_balance structure, we sort our local array of
2568 * hashes in order by slave, so that all of the hashes for a given slave
2569 * become contiguous in memory, and then we point each 'hashes' members of
2570 * a slave_balance structure to the start of a contiguous group. */
2571 n_bals = port->n_ifaces;
2572 for (b = bals; b < &bals[n_bals]; b++) {
2573 b->iface = port->ifaces[b - bals];
2578 for (i = 0; i <= BOND_MASK; i++) {
2579 hashes[i] = &port->bond_hash[i];
2581 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2582 for (i = 0; i <= BOND_MASK; i++) {
2584 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2585 b = &bals[e->iface_idx];
2586 b->tx_bytes += e->tx_bytes;
2588 b->hashes = &hashes[i];
2593 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2594 log_bals(bals, n_bals, port);
2596 /* Discard slaves that aren't enabled (which were sorted to the back of the
2597 * array earlier). */
2598 while (!bals[n_bals - 1].iface->enabled) {
2605 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2606 to = &bals[n_bals - 1];
2607 for (from = bals; from < to; ) {
2608 uint64_t overload = from->tx_bytes - to->tx_bytes;
2609 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2610 /* The extra load on 'from' (and all less-loaded slaves), compared
2611 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2612 * it is less than ~1Mbps. No point in rebalancing. */
2614 } else if (from->n_hashes == 1) {
2615 /* 'from' only carries a single MAC hash, so we can't shift any
2616 * load away from it, even though we want to. */
2619 /* 'from' is carrying significantly more load than 'to', and that
2620 * load is split across at least two different hashes. Pick a hash
2621 * to migrate to 'to' (the least-loaded slave), given that doing so
2622 * must decrease the ratio of the load on the two slaves by at
2625 * The sort order we use means that we prefer to shift away the
2626 * smallest hashes instead of the biggest ones. There is little
2627 * reason behind this decision; we could use the opposite sort
2628 * order to shift away big hashes ahead of small ones. */
2632 for (i = 0; i < from->n_hashes; i++) {
2633 double old_ratio, new_ratio;
2634 uint64_t delta = from->hashes[i]->tx_bytes;
2636 if (delta == 0 || from->tx_bytes - delta == 0) {
2637 /* Pointless move. */
2641 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2643 if (to->tx_bytes == 0) {
2644 /* Nothing on the new slave, move it. */
2648 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2649 new_ratio = (double)(from->tx_bytes - delta) /
2650 (to->tx_bytes + delta);
2652 if (new_ratio == 0) {
2653 /* Should already be covered but check to prevent division
2658 if (new_ratio < 1) {
2659 new_ratio = 1 / new_ratio;
2662 if (old_ratio - new_ratio > 0.1) {
2663 /* Would decrease the ratio, move it. */
2667 if (i < from->n_hashes) {
2668 bond_shift_load(from, to, i);
2669 port->bond_compat_is_stale = true;
2671 /* If the result of the migration changed the relative order of
2672 * 'from' and 'to' swap them back to maintain invariants. */
2673 if (order_swapped) {
2674 swap_bals(from, to);
2677 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2678 * point to different slave_balance structures. It is only
2679 * valid to do these two operations in a row at all because we
2680 * know that 'from' will not move past 'to' and vice versa. */
2681 resort_bals(from, bals, n_bals);
2682 resort_bals(to, bals, n_bals);
2689 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2690 * historical data to decay to <1% in 7 rebalancing runs. */
2691 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2697 bond_send_learning_packets(struct port *port)
2699 struct bridge *br = port->bridge;
2700 struct mac_entry *e;
2701 struct ofpbuf packet;
2702 int error, n_packets, n_errors;
2704 if (!port->n_ifaces || port->active_iface < 0) {
2708 ofpbuf_init(&packet, 128);
2709 error = n_packets = n_errors = 0;
2710 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2711 union ofp_action actions[2], *a;
2717 if (e->port == port->port_idx
2718 || !choose_output_iface(port, e->mac, &xf_ifidx, &tags)) {
2722 /* Compose actions. */
2723 memset(actions, 0, sizeof actions);
2726 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2727 a->vlan_vid.len = htons(sizeof *a);
2728 a->vlan_vid.vlan_vid = htons(e->vlan);
2731 a->output.type = htons(OFPAT_OUTPUT);
2732 a->output.len = htons(sizeof *a);
2733 a->output.port = htons(xflow_port_to_ofp_port(xf_ifidx));
2738 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2740 flow_extract(&packet, XFLOWP_NONE, &flow);
2741 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2748 ofpbuf_uninit(&packet);
2751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2752 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2753 "packets, last error was: %s",
2754 port->name, n_errors, n_packets, strerror(error));
2756 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2757 port->name, n_packets);
2761 /* Bonding unixctl user interface functions. */
2764 bond_unixctl_list(struct unixctl_conn *conn,
2765 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2767 struct ds ds = DS_EMPTY_INITIALIZER;
2768 const struct bridge *br;
2770 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2772 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2775 for (i = 0; i < br->n_ports; i++) {
2776 const struct port *port = br->ports[i];
2777 if (port->n_ifaces > 1) {
2780 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2781 for (j = 0; j < port->n_ifaces; j++) {
2782 const struct iface *iface = port->ifaces[j];
2784 ds_put_cstr(&ds, ", ");
2786 ds_put_cstr(&ds, iface->name);
2788 ds_put_char(&ds, '\n');
2792 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2796 static struct port *
2797 bond_find(const char *name)
2799 const struct bridge *br;
2801 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2804 for (i = 0; i < br->n_ports; i++) {
2805 struct port *port = br->ports[i];
2806 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2815 bond_unixctl_show(struct unixctl_conn *conn,
2816 const char *args, void *aux OVS_UNUSED)
2818 struct ds ds = DS_EMPTY_INITIALIZER;
2819 const struct port *port;
2822 port = bond_find(args);
2824 unixctl_command_reply(conn, 501, "no such bond");
2828 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2829 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2830 ds_put_format(&ds, "next rebalance: %lld ms\n",
2831 port->bridge->bond_next_rebalance - time_msec());
2832 for (j = 0; j < port->n_ifaces; j++) {
2833 const struct iface *iface = port->ifaces[j];
2834 struct bond_entry *be;
2837 ds_put_format(&ds, "slave %s: %s\n",
2838 iface->name, iface->enabled ? "enabled" : "disabled");
2839 if (j == port->active_iface) {
2840 ds_put_cstr(&ds, "\tactive slave\n");
2842 if (iface->delay_expires != LLONG_MAX) {
2843 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2844 iface->enabled ? "downdelay" : "updelay",
2845 iface->delay_expires - time_msec());
2849 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2850 int hash = be - port->bond_hash;
2851 struct mac_entry *me;
2853 if (be->iface_idx != j) {
2857 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2858 hash, be->tx_bytes / 1024);
2861 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2862 &port->bridge->ml->lrus) {
2865 if (bond_hash(me->mac) == hash
2866 && me->port != port->port_idx
2867 && choose_output_iface(port, me->mac, &xf_ifidx, &tags)
2868 && xf_ifidx == iface->xf_ifidx)
2870 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2871 ETH_ADDR_ARGS(me->mac));
2876 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2881 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2882 void *aux OVS_UNUSED)
2884 char *args = (char *) args_;
2885 char *save_ptr = NULL;
2886 char *bond_s, *hash_s, *slave_s;
2887 uint8_t mac[ETH_ADDR_LEN];
2889 struct iface *iface;
2890 struct bond_entry *entry;
2893 bond_s = strtok_r(args, " ", &save_ptr);
2894 hash_s = strtok_r(NULL, " ", &save_ptr);
2895 slave_s = strtok_r(NULL, " ", &save_ptr);
2897 unixctl_command_reply(conn, 501,
2898 "usage: bond/migrate BOND HASH SLAVE");
2902 port = bond_find(bond_s);
2904 unixctl_command_reply(conn, 501, "no such bond");
2908 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2909 == ETH_ADDR_SCAN_COUNT) {
2910 hash = bond_hash(mac);
2911 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2912 hash = atoi(hash_s) & BOND_MASK;
2914 unixctl_command_reply(conn, 501, "bad hash");
2918 iface = port_lookup_iface(port, slave_s);
2920 unixctl_command_reply(conn, 501, "no such slave");
2924 if (!iface->enabled) {
2925 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2929 entry = &port->bond_hash[hash];
2930 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2931 entry->iface_idx = iface->port_ifidx;
2932 entry->iface_tag = tag_create_random();
2933 port->bond_compat_is_stale = true;
2934 unixctl_command_reply(conn, 200, "migrated");
2938 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2939 void *aux OVS_UNUSED)
2941 char *args = (char *) args_;
2942 char *save_ptr = NULL;
2943 char *bond_s, *slave_s;
2945 struct iface *iface;
2947 bond_s = strtok_r(args, " ", &save_ptr);
2948 slave_s = strtok_r(NULL, " ", &save_ptr);
2950 unixctl_command_reply(conn, 501,
2951 "usage: bond/set-active-slave BOND SLAVE");
2955 port = bond_find(bond_s);
2957 unixctl_command_reply(conn, 501, "no such bond");
2961 iface = port_lookup_iface(port, slave_s);
2963 unixctl_command_reply(conn, 501, "no such slave");
2967 if (!iface->enabled) {
2968 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2972 if (port->active_iface != iface->port_ifidx) {
2973 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2974 port->active_iface = iface->port_ifidx;
2975 port->active_iface_tag = tag_create_random();
2976 VLOG_INFO("port %s: active interface is now %s",
2977 port->name, iface->name);
2978 bond_send_learning_packets(port);
2979 unixctl_command_reply(conn, 200, "done");
2981 unixctl_command_reply(conn, 200, "no change");
2986 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2988 char *args = (char *) args_;
2989 char *save_ptr = NULL;
2990 char *bond_s, *slave_s;
2992 struct iface *iface;
2994 bond_s = strtok_r(args, " ", &save_ptr);
2995 slave_s = strtok_r(NULL, " ", &save_ptr);
2997 unixctl_command_reply(conn, 501,
2998 "usage: bond/enable/disable-slave BOND SLAVE");
3002 port = bond_find(bond_s);
3004 unixctl_command_reply(conn, 501, "no such bond");
3008 iface = port_lookup_iface(port, slave_s);
3010 unixctl_command_reply(conn, 501, "no such slave");
3014 bond_enable_slave(iface, enable);
3015 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3019 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3020 void *aux OVS_UNUSED)
3022 enable_slave(conn, args, true);
3026 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3027 void *aux OVS_UNUSED)
3029 enable_slave(conn, args, false);
3033 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3034 void *aux OVS_UNUSED)
3036 uint8_t mac[ETH_ADDR_LEN];
3040 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3041 == ETH_ADDR_SCAN_COUNT) {
3042 hash = bond_hash(mac);
3044 hash_cstr = xasprintf("%u", hash);
3045 unixctl_command_reply(conn, 200, hash_cstr);
3048 unixctl_command_reply(conn, 501, "invalid mac");
3055 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3056 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3057 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3058 unixctl_command_register("bond/set-active-slave",
3059 bond_unixctl_set_active_slave, NULL);
3060 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3062 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3064 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3067 /* Port functions. */
3069 static struct port *
3070 port_create(struct bridge *br, const char *name)
3074 port = xzalloc(sizeof *port);
3076 port->port_idx = br->n_ports;
3078 port->trunks = NULL;
3079 port->name = xstrdup(name);
3080 port->active_iface = -1;
3082 if (br->n_ports >= br->allocated_ports) {
3083 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3086 br->ports[br->n_ports++] = port;
3088 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3095 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3097 struct shash old_ifaces, new_ifaces;
3098 struct shash_node *node;
3099 unsigned long *trunks;
3105 /* Collect old and new interfaces. */
3106 shash_init(&old_ifaces);
3107 shash_init(&new_ifaces);
3108 for (i = 0; i < port->n_ifaces; i++) {
3109 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3111 for (i = 0; i < cfg->n_interfaces; i++) {
3112 const char *name = cfg->interfaces[i]->name;
3113 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3114 VLOG_WARN("port %s: %s specified twice as port interface",
3118 port->updelay = cfg->bond_updelay;
3119 if (port->updelay < 0) {
3122 port->updelay = cfg->bond_downdelay;
3123 if (port->downdelay < 0) {
3124 port->downdelay = 0;
3127 /* Get rid of deleted interfaces and add new interfaces. */
3128 SHASH_FOR_EACH (node, &old_ifaces) {
3129 if (!shash_find(&new_ifaces, node->name)) {
3130 iface_destroy(node->data);
3133 SHASH_FOR_EACH (node, &new_ifaces) {
3134 const struct ovsrec_interface *if_cfg = node->data;
3135 struct iface *iface;
3137 iface = shash_find_data(&old_ifaces, if_cfg->name);
3139 iface_create(port, if_cfg);
3141 iface->cfg = if_cfg;
3148 if (port->n_ifaces < 2) {
3150 if (vlan >= 0 && vlan <= 4095) {
3151 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3156 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3157 * they even work as-is. But they have not been tested. */
3158 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3162 if (port->vlan != vlan) {
3164 bridge_flush(port->bridge);
3167 /* Get trunked VLANs. */
3173 trunks = bitmap_allocate(4096);
3175 for (i = 0; i < cfg->n_trunks; i++) {
3176 int trunk = cfg->trunks[i];
3178 bitmap_set1(trunks, trunk);
3184 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3185 port->name, cfg->n_trunks);
3187 if (n_errors == cfg->n_trunks) {
3189 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3192 bitmap_set_multiple(trunks, 0, 4096, 1);
3195 if (cfg->n_trunks) {
3196 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3201 ? port->trunks != NULL
3202 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3203 bridge_flush(port->bridge);
3205 bitmap_free(port->trunks);
3206 port->trunks = trunks;
3208 shash_destroy(&old_ifaces);
3209 shash_destroy(&new_ifaces);
3213 port_destroy(struct port *port)
3216 struct bridge *br = port->bridge;
3220 proc_net_compat_update_vlan(port->name, NULL, 0);
3221 proc_net_compat_update_bond(port->name, NULL);
3223 for (i = 0; i < MAX_MIRRORS; i++) {
3224 struct mirror *m = br->mirrors[i];
3225 if (m && m->out_port == port) {
3230 while (port->n_ifaces > 0) {
3231 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3234 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3235 del->port_idx = port->port_idx;
3238 bitmap_free(port->trunks);
3245 static struct port *
3246 port_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3248 struct iface *iface = iface_from_xf_ifidx(br, xf_ifidx);
3249 return iface ? iface->port : NULL;
3252 static struct port *
3253 port_lookup(const struct bridge *br, const char *name)
3257 for (i = 0; i < br->n_ports; i++) {
3258 struct port *port = br->ports[i];
3259 if (!strcmp(port->name, name)) {
3266 static struct iface *
3267 port_lookup_iface(const struct port *port, const char *name)
3271 for (j = 0; j < port->n_ifaces; j++) {
3272 struct iface *iface = port->ifaces[j];
3273 if (!strcmp(iface->name, name)) {
3281 port_update_bonding(struct port *port)
3283 if (port->n_ifaces < 2) {
3284 /* Not a bonded port. */
3285 if (port->bond_hash) {
3286 free(port->bond_hash);
3287 port->bond_hash = NULL;
3288 port->bond_compat_is_stale = true;
3289 port->bond_fake_iface = false;
3292 if (!port->bond_hash) {
3295 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3296 for (i = 0; i <= BOND_MASK; i++) {
3297 struct bond_entry *e = &port->bond_hash[i];
3301 port->no_ifaces_tag = tag_create_random();
3302 bond_choose_active_iface(port);
3304 port->bond_compat_is_stale = true;
3305 port->bond_fake_iface = port->cfg->bond_fake_iface;
3310 port_update_bond_compat(struct port *port)
3312 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3313 struct compat_bond bond;
3316 if (port->n_ifaces < 2) {
3317 proc_net_compat_update_bond(port->name, NULL);
3322 bond.updelay = port->updelay;
3323 bond.downdelay = port->downdelay;
3326 bond.hashes = compat_hashes;
3327 if (port->bond_hash) {
3328 const struct bond_entry *e;
3329 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3330 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3331 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3332 cbh->hash = e - port->bond_hash;
3333 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3338 bond.n_slaves = port->n_ifaces;
3339 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3340 for (i = 0; i < port->n_ifaces; i++) {
3341 struct iface *iface = port->ifaces[i];
3342 struct compat_bond_slave *slave = &bond.slaves[i];
3343 slave->name = iface->name;
3345 /* We need to make the same determination as the Linux bonding
3346 * code to determine whether a slave should be consider "up".
3347 * The Linux function bond_miimon_inspect() supports four
3348 * BOND_LINK_* states:
3350 * - BOND_LINK_UP: carrier detected, updelay has passed.
3351 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3352 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3353 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3355 * The function bond_info_show_slave() only considers BOND_LINK_UP
3356 * to be "up" and anything else to be "down".
3358 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3362 netdev_get_etheraddr(iface->netdev, slave->mac);
3365 if (port->bond_fake_iface) {
3366 struct netdev *bond_netdev;
3368 if (!netdev_open_default(port->name, &bond_netdev)) {
3370 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3372 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3374 netdev_close(bond_netdev);
3378 proc_net_compat_update_bond(port->name, &bond);
3383 port_update_vlan_compat(struct port *port)
3385 struct bridge *br = port->bridge;
3386 char *vlandev_name = NULL;
3388 if (port->vlan > 0) {
3389 /* Figure out the name that the VLAN device should actually have, if it
3390 * existed. This takes some work because the VLAN device would not
3391 * have port->name in its name; rather, it would have the trunk port's
3392 * name, and 'port' would be attached to a bridge that also had the
3393 * VLAN device one of its ports. So we need to find a trunk port that
3394 * includes port->vlan.
3396 * There might be more than one candidate. This doesn't happen on
3397 * XenServer, so if it happens we just pick the first choice in
3398 * alphabetical order instead of creating multiple VLAN devices. */
3400 for (i = 0; i < br->n_ports; i++) {
3401 struct port *p = br->ports[i];
3402 if (port_trunks_vlan(p, port->vlan)
3404 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3406 uint8_t ea[ETH_ADDR_LEN];
3407 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3408 if (!eth_addr_is_multicast(ea) &&
3409 !eth_addr_is_reserved(ea) &&
3410 !eth_addr_is_zero(ea)) {
3411 vlandev_name = p->name;
3416 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3419 /* Interface functions. */
3421 static struct iface *
3422 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3424 struct iface *iface;
3425 char *name = if_cfg->name;
3428 iface = xzalloc(sizeof *iface);
3430 iface->port_ifidx = port->n_ifaces;
3431 iface->name = xstrdup(name);
3432 iface->xf_ifidx = -1;
3433 iface->tag = tag_create_random();
3434 iface->delay_expires = LLONG_MAX;
3435 iface->netdev = NULL;
3436 iface->cfg = if_cfg;
3438 if (port->n_ifaces >= port->allocated_ifaces) {
3439 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3440 sizeof *port->ifaces);
3442 port->ifaces[port->n_ifaces++] = iface;
3443 if (port->n_ifaces > 1) {
3444 port->bridge->has_bonded_ports = true;
3447 /* Attempt to create the network interface in case it
3448 * doesn't exist yet. */
3449 if (!iface_is_internal(port->bridge, iface->name)) {
3450 error = set_up_iface(if_cfg, iface, true);
3452 VLOG_WARN("could not create iface %s: %s", iface->name,
3457 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3459 bridge_flush(port->bridge);
3465 iface_destroy(struct iface *iface)
3468 struct port *port = iface->port;
3469 struct bridge *br = port->bridge;
3470 bool del_active = port->active_iface == iface->port_ifidx;
3473 if (iface->xf_ifidx >= 0) {
3474 port_array_set(&br->ifaces, iface->xf_ifidx, NULL);
3477 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3478 del->port_ifidx = iface->port_ifidx;
3480 netdev_close(iface->netdev);
3483 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3484 bond_choose_active_iface(port);
3485 bond_send_learning_packets(port);
3491 bridge_flush(port->bridge);
3495 static struct iface *
3496 iface_lookup(const struct bridge *br, const char *name)
3500 for (i = 0; i < br->n_ports; i++) {
3501 struct port *port = br->ports[i];
3502 for (j = 0; j < port->n_ifaces; j++) {
3503 struct iface *iface = port->ifaces[j];
3504 if (!strcmp(iface->name, name)) {
3512 static struct iface *
3513 iface_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3515 return port_array_get(&br->ifaces, xf_ifidx);
3518 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3519 * 'br', that is, an interface that is entirely simulated within the datapath.
3520 * The local port (XFLOWP_LOCAL) is always an internal interface. Other local
3521 * interfaces are created by setting "iface.<iface>.internal = true".
3523 * In addition, we have a kluge-y feature that creates an internal port with
3524 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3525 * This feature needs to go away in the long term. Until then, this is one
3526 * reason why this function takes a name instead of a struct iface: the fake
3527 * interfaces created this way do not have a struct iface. */
3529 iface_is_internal(const struct bridge *br, const char *if_name)
3531 /* XXX wastes time */
3532 struct iface *iface;
3535 if (!strcmp(if_name, br->name)) {
3539 iface = iface_lookup(br, if_name);
3540 if (iface && !strcmp(iface->cfg->type, "internal")) {
3544 port = port_lookup(br, if_name);
3545 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3551 /* Set Ethernet address of 'iface', if one is specified in the configuration
3554 iface_set_mac(struct iface *iface)
3556 uint8_t ea[ETH_ADDR_LEN];
3558 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3559 if (eth_addr_is_multicast(ea)) {
3560 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3562 } else if (iface->xf_ifidx == XFLOWP_LOCAL) {
3563 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3564 iface->name, iface->name);
3566 int error = netdev_set_etheraddr(iface->netdev, ea);
3568 VLOG_ERR("interface %s: setting MAC failed (%s)",
3569 iface->name, strerror(error));
3575 /* Port mirroring. */
3578 mirror_reconfigure(struct bridge *br)
3580 struct shash old_mirrors, new_mirrors;
3581 struct shash_node *node;
3582 unsigned long *rspan_vlans;
3585 /* Collect old mirrors. */
3586 shash_init(&old_mirrors);
3587 for (i = 0; i < MAX_MIRRORS; i++) {
3588 if (br->mirrors[i]) {
3589 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3593 /* Collect new mirrors. */
3594 shash_init(&new_mirrors);
3595 for (i = 0; i < br->cfg->n_mirrors; i++) {
3596 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3597 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3598 VLOG_WARN("bridge %s: %s specified twice as mirror",
3599 br->name, cfg->name);
3603 /* Get rid of deleted mirrors and add new mirrors. */
3604 SHASH_FOR_EACH (node, &old_mirrors) {
3605 if (!shash_find(&new_mirrors, node->name)) {
3606 mirror_destroy(node->data);
3609 SHASH_FOR_EACH (node, &new_mirrors) {
3610 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3612 mirror = mirror_create(br, node->name);
3617 mirror_reconfigure_one(mirror, node->data);
3619 shash_destroy(&old_mirrors);
3620 shash_destroy(&new_mirrors);
3622 /* Update port reserved status. */
3623 for (i = 0; i < br->n_ports; i++) {
3624 br->ports[i]->is_mirror_output_port = false;
3626 for (i = 0; i < MAX_MIRRORS; i++) {
3627 struct mirror *m = br->mirrors[i];
3628 if (m && m->out_port) {
3629 m->out_port->is_mirror_output_port = true;
3633 /* Update flooded vlans (for RSPAN). */
3635 if (br->cfg->n_flood_vlans) {
3636 rspan_vlans = bitmap_allocate(4096);
3638 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3639 int64_t vlan = br->cfg->flood_vlans[i];
3640 if (vlan >= 0 && vlan < 4096) {
3641 bitmap_set1(rspan_vlans, vlan);
3642 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3645 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3650 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3655 static struct mirror *
3656 mirror_create(struct bridge *br, const char *name)
3661 for (i = 0; ; i++) {
3662 if (i >= MAX_MIRRORS) {
3663 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3664 "cannot create %s", br->name, MAX_MIRRORS, name);
3667 if (!br->mirrors[i]) {
3672 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3675 br->mirrors[i] = m = xzalloc(sizeof *m);
3678 m->name = xstrdup(name);
3679 shash_init(&m->src_ports);
3680 shash_init(&m->dst_ports);
3690 mirror_destroy(struct mirror *m)
3693 struct bridge *br = m->bridge;
3696 for (i = 0; i < br->n_ports; i++) {
3697 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3698 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3701 shash_destroy(&m->src_ports);
3702 shash_destroy(&m->dst_ports);
3705 m->bridge->mirrors[m->idx] = NULL;
3713 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3714 struct shash *names)
3718 for (i = 0; i < n_ports; i++) {
3719 const char *name = ports[i]->name;
3720 if (port_lookup(m->bridge, name)) {
3721 shash_add_once(names, name, NULL);
3723 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3724 "port %s", m->bridge->name, m->name, name);
3730 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3736 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3738 for (i = 0; i < cfg->n_select_vlan; i++) {
3739 int64_t vlan = cfg->select_vlan[i];
3740 if (vlan < 0 || vlan > 4095) {
3741 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3742 m->bridge->name, m->name, vlan);
3744 (*vlans)[n_vlans++] = vlan;
3751 vlan_is_mirrored(const struct mirror *m, int vlan)
3755 for (i = 0; i < m->n_vlans; i++) {
3756 if (m->vlans[i] == vlan) {
3764 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3768 for (i = 0; i < m->n_vlans; i++) {
3769 if (port_trunks_vlan(p, m->vlans[i])) {
3777 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3779 struct shash src_ports, dst_ports;
3780 mirror_mask_t mirror_bit;
3781 struct port *out_port;
3787 /* Get output port. */
3788 if (cfg->output_port) {
3789 out_port = port_lookup(m->bridge, cfg->output_port->name);
3791 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3792 m->bridge->name, m->name);
3798 if (cfg->output_vlan) {
3799 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3800 "output vlan; ignoring output vlan",
3801 m->bridge->name, m->name);
3803 } else if (cfg->output_vlan) {
3805 out_vlan = *cfg->output_vlan;
3807 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3808 m->bridge->name, m->name);
3813 shash_init(&src_ports);
3814 shash_init(&dst_ports);
3815 if (cfg->select_all) {
3816 for (i = 0; i < m->bridge->n_ports; i++) {
3817 const char *name = m->bridge->ports[i]->name;
3818 shash_add_once(&src_ports, name, NULL);
3819 shash_add_once(&dst_ports, name, NULL);
3824 /* Get ports, and drop duplicates and ports that don't exist. */
3825 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3827 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3830 /* Get all the vlans, and drop duplicate and invalid vlans. */
3831 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3834 /* Update mirror data. */
3835 if (!shash_equal_keys(&m->src_ports, &src_ports)
3836 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3837 || m->n_vlans != n_vlans
3838 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3839 || m->out_port != out_port
3840 || m->out_vlan != out_vlan) {
3841 bridge_flush(m->bridge);
3843 shash_swap(&m->src_ports, &src_ports);
3844 shash_swap(&m->dst_ports, &dst_ports);
3847 m->n_vlans = n_vlans;
3848 m->out_port = out_port;
3849 m->out_vlan = out_vlan;
3852 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3853 for (i = 0; i < m->bridge->n_ports; i++) {
3854 struct port *port = m->bridge->ports[i];
3856 if (shash_find(&m->src_ports, port->name)
3859 ? port_trunks_any_mirrored_vlan(m, port)
3860 : vlan_is_mirrored(m, port->vlan)))) {
3861 port->src_mirrors |= mirror_bit;
3863 port->src_mirrors &= ~mirror_bit;
3866 if (shash_find(&m->dst_ports, port->name)) {
3867 port->dst_mirrors |= mirror_bit;
3869 port->dst_mirrors &= ~mirror_bit;
3874 shash_destroy(&src_ports);
3875 shash_destroy(&dst_ports);