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 netdev_options.may_open = true;
395 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
396 netdev_options.may_open = true;
399 error = netdev_open(&netdev_options, &iface->netdev);
402 netdev_get_carrier(iface->netdev, &iface->enabled);
404 } else if (iface->netdev) {
405 const char *netdev_type = netdev_get_type(iface->netdev);
406 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
407 ? iface_cfg->type : NULL;
409 /* An "internal" config type maps to a netdev "system" type. */
410 if (iface_type && !strcmp(iface_type, "internal")) {
411 iface_type = "system";
414 if (!iface_type || !strcmp(netdev_type, iface_type)) {
415 error = netdev_reconfigure(iface->netdev, &options);
417 VLOG_WARN("%s: attempting change device type from %s to %s",
418 iface_cfg->name, netdev_type, iface_type);
423 SHASH_FOR_EACH (node, &options) {
426 shash_destroy(&options);
432 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
434 return set_up_iface(iface_cfg, iface, false);
438 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
439 void *aux OVS_UNUSED)
441 if (!iface->netdev) {
442 int error = set_up_iface(iface->cfg, iface, true);
444 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
454 check_iface_xf_ifidx(struct bridge *br, struct iface *iface,
455 void *aux OVS_UNUSED)
457 if (iface->xf_ifidx >= 0) {
458 VLOG_DBG("%s has interface %s on port %d",
460 iface->name, iface->xf_ifidx);
463 VLOG_ERR("%s interface not in %s, dropping",
464 iface->name, xfif_name(br->xfif));
470 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
471 void *aux OVS_UNUSED)
473 /* Set policing attributes. */
474 netdev_set_policing(iface->netdev,
475 iface->cfg->ingress_policing_rate,
476 iface->cfg->ingress_policing_burst);
478 /* Set MAC address of internal interfaces other than the local
480 if (iface->xf_ifidx != XFLOWP_LOCAL
481 && iface_is_internal(br, iface->name)) {
482 iface_set_mac(iface);
488 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
489 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
490 * deletes from 'br' any ports that no longer have any interfaces. */
492 iterate_and_prune_ifaces(struct bridge *br,
493 bool (*cb)(struct bridge *, struct iface *,
499 for (i = 0; i < br->n_ports; ) {
500 struct port *port = br->ports[i];
501 for (j = 0; j < port->n_ifaces; ) {
502 struct iface *iface = port->ifaces[j];
503 if (cb(br, iface, aux)) {
506 iface_destroy(iface);
510 if (port->n_ifaces) {
513 VLOG_ERR("%s port has no interfaces, dropping", port->name);
520 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
522 struct ovsdb_idl_txn *txn;
523 struct shash old_br, new_br;
524 struct shash_node *node;
525 struct bridge *br, *next;
527 int sflow_bridge_number;
529 COVERAGE_INC(bridge_reconfigure);
531 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
533 /* Collect old and new bridges. */
536 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
537 shash_add(&old_br, br->name, br);
539 for (i = 0; i < ovs_cfg->n_bridges; i++) {
540 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
541 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
542 VLOG_WARN("more than one bridge named %s", br_cfg->name);
546 /* Get rid of deleted bridges and add new bridges. */
547 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
548 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
555 SHASH_FOR_EACH (node, &new_br) {
556 const char *br_name = node->name;
557 const struct ovsrec_bridge *br_cfg = node->data;
558 br = shash_find_data(&old_br, br_name);
560 /* If the bridge datapath type has changed, we need to tear it
561 * down and recreate. */
562 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
564 bridge_create(br_cfg);
567 bridge_create(br_cfg);
570 shash_destroy(&old_br);
571 shash_destroy(&new_br);
575 bridge_configure_ssl(ovs_cfg->ssl);
578 /* Reconfigure all bridges. */
579 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
580 bridge_reconfigure_one(ovs_cfg, br);
583 /* Add and delete ports on all datapaths.
585 * The kernel will reject any attempt to add a given port to a datapath if
586 * that port already belongs to a different datapath, so we must do all
587 * port deletions before any port additions. */
588 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
589 struct xflow_port *xfif_ports;
591 struct shash want_ifaces;
593 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
594 bridge_get_all_ifaces(br, &want_ifaces);
595 for (i = 0; i < n_xfif_ports; i++) {
596 const struct xflow_port *p = &xfif_ports[i];
597 if (!shash_find(&want_ifaces, p->devname)
598 && strcmp(p->devname, br->name)) {
599 int retval = xfif_port_del(br->xfif, p->port);
601 VLOG_ERR("failed to remove %s interface from %s: %s",
602 p->devname, xfif_name(br->xfif),
607 shash_destroy(&want_ifaces);
610 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
611 struct xflow_port *xfif_ports;
613 struct shash cur_ifaces, want_ifaces;
614 struct shash_node *node;
616 /* Get the set of interfaces currently in this datapath. */
617 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
618 shash_init(&cur_ifaces);
619 for (i = 0; i < n_xfif_ports; i++) {
620 const char *name = xfif_ports[i].devname;
621 if (!shash_find(&cur_ifaces, name)) {
622 shash_add(&cur_ifaces, name, NULL);
627 /* Get the set of interfaces we want on this datapath. */
628 bridge_get_all_ifaces(br, &want_ifaces);
630 SHASH_FOR_EACH (node, &want_ifaces) {
631 const char *if_name = node->name;
632 struct iface *iface = node->data;
634 if (shash_find(&cur_ifaces, if_name)) {
635 /* Already exists, just reconfigure it. */
637 reconfigure_iface(iface->cfg, iface);
640 /* Need to add to datapath. */
644 /* Add to datapath. */
645 internal = iface_is_internal(br, if_name);
646 error = xfif_port_add(br->xfif, if_name,
647 internal ? XFLOW_PORT_INTERNAL : 0, NULL);
648 if (error == EFBIG) {
649 VLOG_ERR("ran out of valid port numbers on %s",
650 xfif_name(br->xfif));
653 VLOG_ERR("failed to add %s interface to %s: %s",
654 if_name, xfif_name(br->xfif), strerror(error));
658 shash_destroy(&cur_ifaces);
659 shash_destroy(&want_ifaces);
661 sflow_bridge_number = 0;
662 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
665 struct iface *local_iface;
666 struct iface *hw_addr_iface;
669 bridge_fetch_dp_ifaces(br);
671 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
672 iterate_and_prune_ifaces(br, check_iface_xf_ifidx, NULL);
674 /* Pick local port hardware address, datapath ID. */
675 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
676 local_iface = bridge_get_local_iface(br);
678 int error = netdev_set_etheraddr(local_iface->netdev, ea);
680 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
681 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
682 "Ethernet address: %s",
683 br->name, strerror(error));
687 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
688 ofproto_set_datapath_id(br->ofproto, dpid);
690 dpid_string = xasprintf("%012"PRIx64, dpid);
691 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
694 /* Set NetFlow configuration on this bridge. */
695 if (br->cfg->netflow) {
696 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
697 struct netflow_options opts;
699 memset(&opts, 0, sizeof opts);
701 xfif_get_netflow_ids(br->xfif, &opts.engine_type, &opts.engine_id);
702 if (nf_cfg->engine_type) {
703 opts.engine_type = *nf_cfg->engine_type;
705 if (nf_cfg->engine_id) {
706 opts.engine_id = *nf_cfg->engine_id;
709 opts.active_timeout = nf_cfg->active_timeout;
710 if (!opts.active_timeout) {
711 opts.active_timeout = -1;
712 } else if (opts.active_timeout < 0) {
713 VLOG_WARN("bridge %s: active timeout interval set to negative "
714 "value, using default instead (%d seconds)", br->name,
715 NF_ACTIVE_TIMEOUT_DEFAULT);
716 opts.active_timeout = -1;
719 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
720 if (opts.add_id_to_iface) {
721 if (opts.engine_id > 0x7f) {
722 VLOG_WARN("bridge %s: netflow port mangling may conflict "
723 "with another vswitch, choose an engine id less "
724 "than 128", br->name);
726 if (br->n_ports > 508) {
727 VLOG_WARN("bridge %s: netflow port mangling will conflict "
728 "with another port when more than 508 ports are "
733 opts.collectors.n = nf_cfg->n_targets;
734 opts.collectors.names = nf_cfg->targets;
735 if (ofproto_set_netflow(br->ofproto, &opts)) {
736 VLOG_ERR("bridge %s: problem setting netflow collectors",
740 ofproto_set_netflow(br->ofproto, NULL);
743 /* Set sFlow configuration on this bridge. */
744 if (br->cfg->sflow) {
745 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
746 const struct ovsrec_controller *ctrl;
747 struct ofproto_sflow_options oso;
749 memset(&oso, 0, sizeof oso);
751 oso.targets.n = sflow_cfg->n_targets;
752 oso.targets.names = sflow_cfg->targets;
754 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
755 if (sflow_cfg->sampling) {
756 oso.sampling_rate = *sflow_cfg->sampling;
759 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
760 if (sflow_cfg->polling) {
761 oso.polling_interval = *sflow_cfg->polling;
764 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
765 if (sflow_cfg->header) {
766 oso.header_len = *sflow_cfg->header;
769 oso.sub_id = sflow_bridge_number++;
770 oso.agent_device = sflow_cfg->agent;
772 ctrl = bridge_get_controller(ovs_cfg, br);
773 oso.control_ip = ctrl ? ctrl->local_ip : NULL;
774 ofproto_set_sflow(br->ofproto, &oso);
776 svec_destroy(&oso.targets);
778 ofproto_set_sflow(br->ofproto, NULL);
781 /* Update the controller and related settings. It would be more
782 * straightforward to call this from bridge_reconfigure_one(), but we
783 * can't do it there for two reasons. First, and most importantly, at
784 * that point we don't know the xf_ifidx of any interfaces that have
785 * been added to the bridge (because we haven't actually added them to
786 * the datapath). Second, at that point we haven't set the datapath ID
787 * yet; when a controller is configured, resetting the datapath ID will
788 * immediately disconnect from the controller, so it's better to set
789 * the datapath ID before the controller. */
790 bridge_reconfigure_controller(ovs_cfg, br);
792 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
793 for (i = 0; i < br->n_ports; i++) {
794 struct port *port = br->ports[i];
796 port_update_vlan_compat(port);
797 port_update_bonding(port);
800 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
801 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
804 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
806 ovsdb_idl_txn_commit(txn);
807 ovsdb_idl_txn_destroy(txn); /* XXX */
811 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
815 for (i = 0; i < br_cfg->n_other_config; i++) {
816 if (!strcmp(br_cfg->key_other_config[i], key)) {
817 return br_cfg->value_other_config[i];
824 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
825 struct iface **hw_addr_iface)
831 *hw_addr_iface = NULL;
833 /* Did the user request a particular MAC? */
834 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
835 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
836 if (eth_addr_is_multicast(ea)) {
837 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
838 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
839 } else if (eth_addr_is_zero(ea)) {
840 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
846 /* Otherwise choose the minimum non-local MAC address among all of the
848 memset(ea, 0xff, sizeof ea);
849 for (i = 0; i < br->n_ports; i++) {
850 struct port *port = br->ports[i];
851 uint8_t iface_ea[ETH_ADDR_LEN];
854 /* Mirror output ports don't participate. */
855 if (port->is_mirror_output_port) {
859 /* Choose the MAC address to represent the port. */
860 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
861 /* Find the interface with this Ethernet address (if any) so that
862 * we can provide the correct devname to the caller. */
864 for (j = 0; j < port->n_ifaces; j++) {
865 struct iface *candidate = port->ifaces[j];
866 uint8_t candidate_ea[ETH_ADDR_LEN];
867 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
868 && eth_addr_equals(iface_ea, candidate_ea)) {
873 /* Choose the interface whose MAC address will represent the port.
874 * The Linux kernel bonding code always chooses the MAC address of
875 * the first slave added to a bond, and the Fedora networking
876 * scripts always add slaves to a bond in alphabetical order, so
877 * for compatibility we choose the interface with the name that is
878 * first in alphabetical order. */
879 iface = port->ifaces[0];
880 for (j = 1; j < port->n_ifaces; j++) {
881 struct iface *candidate = port->ifaces[j];
882 if (strcmp(candidate->name, iface->name) < 0) {
887 /* The local port doesn't count (since we're trying to choose its
888 * MAC address anyway). */
889 if (iface->xf_ifidx == XFLOWP_LOCAL) {
894 error = netdev_get_etheraddr(iface->netdev, iface_ea);
896 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
897 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
898 iface->name, strerror(error));
903 /* Compare against our current choice. */
904 if (!eth_addr_is_multicast(iface_ea) &&
905 !eth_addr_is_local(iface_ea) &&
906 !eth_addr_is_reserved(iface_ea) &&
907 !eth_addr_is_zero(iface_ea) &&
908 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
910 memcpy(ea, iface_ea, ETH_ADDR_LEN);
911 *hw_addr_iface = iface;
914 if (eth_addr_is_multicast(ea)) {
915 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
916 *hw_addr_iface = NULL;
917 VLOG_WARN("bridge %s: using default bridge Ethernet "
918 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
920 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
921 br->name, ETH_ADDR_ARGS(ea));
925 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
926 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
927 * an interface on 'br', then that interface must be passed in as
928 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
929 * 'hw_addr_iface' must be passed in as a null pointer. */
931 bridge_pick_datapath_id(struct bridge *br,
932 const uint8_t bridge_ea[ETH_ADDR_LEN],
933 struct iface *hw_addr_iface)
936 * The procedure for choosing a bridge MAC address will, in the most
937 * ordinary case, also choose a unique MAC that we can use as a datapath
938 * ID. In some special cases, though, multiple bridges will end up with
939 * the same MAC address. This is OK for the bridges, but it will confuse
940 * the OpenFlow controller, because each datapath needs a unique datapath
943 * Datapath IDs must be unique. It is also very desirable that they be
944 * stable from one run to the next, so that policy set on a datapath
947 const char *datapath_id;
950 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
951 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
957 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
959 * A bridge whose MAC address is taken from a VLAN network device
960 * (that is, a network device created with vconfig(8) or similar
961 * tool) will have the same MAC address as a bridge on the VLAN
962 * device's physical network device.
964 * Handle this case by hashing the physical network device MAC
965 * along with the VLAN identifier.
967 uint8_t buf[ETH_ADDR_LEN + 2];
968 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
969 buf[ETH_ADDR_LEN] = vlan >> 8;
970 buf[ETH_ADDR_LEN + 1] = vlan;
971 return dpid_from_hash(buf, sizeof buf);
974 * Assume that this bridge's MAC address is unique, since it
975 * doesn't fit any of the cases we handle specially.
980 * A purely internal bridge, that is, one that has no non-virtual
981 * network devices on it at all, is more difficult because it has no
982 * natural unique identifier at all.
984 * When the host is a XenServer, we handle this case by hashing the
985 * host's UUID with the name of the bridge. Names of bridges are
986 * persistent across XenServer reboots, although they can be reused if
987 * an internal network is destroyed and then a new one is later
988 * created, so this is fairly effective.
990 * When the host is not a XenServer, we punt by using a random MAC
991 * address on each run.
993 const char *host_uuid = xenserver_get_host_uuid();
995 char *combined = xasprintf("%s,%s", host_uuid, br->name);
996 dpid = dpid_from_hash(combined, strlen(combined));
1002 return eth_addr_to_uint64(bridge_ea);
1006 dpid_from_hash(const void *data, size_t n)
1008 uint8_t hash[SHA1_DIGEST_SIZE];
1010 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1011 sha1_bytes(data, n, hash);
1012 eth_addr_mark_random(hash);
1013 return eth_addr_to_uint64(hash);
1019 struct bridge *br, *next;
1023 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1024 int error = bridge_run_one(br);
1026 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1027 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1028 "forcing reconfiguration", br->name);
1042 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1043 ofproto_wait(br->ofproto);
1044 if (br->controller) {
1048 mac_learning_wait(br->ml);
1053 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1054 * configuration changes. */
1056 bridge_flush(struct bridge *br)
1058 COVERAGE_INC(bridge_flush);
1060 mac_learning_flush(br->ml);
1063 /* Returns the 'br' interface for the XFLOWP_LOCAL port, or null if 'br' has no
1064 * such interface. */
1065 static struct iface *
1066 bridge_get_local_iface(struct bridge *br)
1070 for (i = 0; i < br->n_ports; i++) {
1071 struct port *port = br->ports[i];
1072 for (j = 0; j < port->n_ifaces; j++) {
1073 struct iface *iface = port->ifaces[j];
1074 if (iface->xf_ifidx == XFLOWP_LOCAL) {
1083 /* Bridge unixctl user interface functions. */
1085 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1086 const char *args, void *aux OVS_UNUSED)
1088 struct ds ds = DS_EMPTY_INITIALIZER;
1089 const struct bridge *br;
1090 const struct mac_entry *e;
1092 br = bridge_lookup(args);
1094 unixctl_command_reply(conn, 501, "no such bridge");
1098 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1099 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1100 if (e->port < 0 || e->port >= br->n_ports) {
1103 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1104 br->ports[e->port]->ifaces[0]->xf_ifidx,
1105 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1107 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1111 /* Bridge reconfiguration functions. */
1112 static struct bridge *
1113 bridge_create(const struct ovsrec_bridge *br_cfg)
1118 assert(!bridge_lookup(br_cfg->name));
1119 br = xzalloc(sizeof *br);
1121 error = xfif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1127 xfif_flow_flush(br->xfif);
1129 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1132 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1134 xfif_delete(br->xfif);
1135 xfif_close(br->xfif);
1140 br->name = xstrdup(br_cfg->name);
1142 br->ml = mac_learning_create();
1143 br->sent_config_request = false;
1144 eth_addr_nicira_random(br->default_ea);
1146 port_array_init(&br->ifaces);
1149 br->bond_next_rebalance = time_msec() + 10000;
1151 list_push_back(&all_bridges, &br->node);
1153 VLOG_INFO("created bridge %s on %s", br->name, xfif_name(br->xfif));
1159 bridge_destroy(struct bridge *br)
1164 while (br->n_ports > 0) {
1165 port_destroy(br->ports[br->n_ports - 1]);
1167 list_remove(&br->node);
1168 error = xfif_delete(br->xfif);
1169 if (error && error != ENOENT) {
1170 VLOG_ERR("failed to delete %s: %s",
1171 xfif_name(br->xfif), strerror(error));
1173 xfif_close(br->xfif);
1174 ofproto_destroy(br->ofproto);
1175 free(br->controller);
1176 mac_learning_destroy(br->ml);
1177 port_array_destroy(&br->ifaces);
1184 static struct bridge *
1185 bridge_lookup(const char *name)
1189 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1190 if (!strcmp(br->name, name)) {
1198 bridge_exists(const char *name)
1200 return bridge_lookup(name) ? true : false;
1204 bridge_get_datapathid(const char *name)
1206 struct bridge *br = bridge_lookup(name);
1207 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1210 /* Handle requests for a listing of all flows known by the OpenFlow
1211 * stack, including those normally hidden. */
1213 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1214 const char *args, void *aux OVS_UNUSED)
1219 br = bridge_lookup(args);
1221 unixctl_command_reply(conn, 501, "Unknown bridge");
1226 ofproto_get_all_flows(br->ofproto, &results);
1228 unixctl_command_reply(conn, 200, ds_cstr(&results));
1229 ds_destroy(&results);
1233 bridge_run_one(struct bridge *br)
1237 error = ofproto_run1(br->ofproto);
1242 //XXX mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1245 error = ofproto_run2(br->ofproto, br->flush);
1251 static const struct ovsrec_controller *
1252 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1253 const struct bridge *br)
1255 const struct ovsrec_controller *controller;
1257 controller = (br->cfg->controller ? br->cfg->controller
1258 : ovs_cfg->controller ? ovs_cfg->controller
1261 if (controller && !strcmp(controller->target, "none")) {
1269 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1271 struct svec *ifaces = ifaces_;
1272 if (!svec_contains(ifaces, iface->name)) {
1273 svec_add(ifaces, iface->name);
1277 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1279 br->name, iface->name, iface->port->name);
1285 bridge_update_desc(struct bridge *br)
1288 bool changed = false;
1291 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1292 if (desc != br->mfr_desc) {
1295 br->mfr_desc = xstrdup(desc);
1297 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1302 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1303 if (desc != br->hw_desc) {
1306 br->hw_desc = xstrdup(desc);
1308 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1313 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1314 if (desc != br->sw_desc) {
1317 br->sw_desc = xstrdup(desc);
1319 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1324 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1325 if (desc != br->serial_desc) {
1326 free(br->serial_desc);
1328 br->serial_desc = xstrdup(desc);
1330 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1335 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1336 if (desc != br->dp_desc) {
1339 br->dp_desc = xstrdup(desc);
1341 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1347 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1348 br->sw_desc, br->serial_desc, br->dp_desc);
1354 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1357 struct shash old_ports, new_ports;
1359 struct svec listeners, old_listeners;
1360 struct svec snoops, old_snoops;
1361 struct shash_node *node;
1364 /* Collect old ports. */
1365 shash_init(&old_ports);
1366 for (i = 0; i < br->n_ports; i++) {
1367 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1370 /* Collect new ports. */
1371 shash_init(&new_ports);
1372 for (i = 0; i < br->cfg->n_ports; i++) {
1373 const char *name = br->cfg->ports[i]->name;
1374 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1375 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1380 /* If we have a controller, then we need a local port. Complain if the
1381 * user didn't specify one.
1383 * XXX perhaps we should synthesize a port ourselves in this case. */
1384 if (bridge_get_controller(ovs_cfg, br)) {
1385 char local_name[IF_NAMESIZE];
1388 error = xfif_port_get_name(br->xfif, XFLOWP_LOCAL,
1389 local_name, sizeof local_name);
1390 if (!error && !shash_find(&new_ports, local_name)) {
1391 VLOG_WARN("bridge %s: controller specified but no local port "
1392 "(port named %s) defined",
1393 br->name, local_name);
1397 /* Get rid of deleted ports and add new ports. */
1398 SHASH_FOR_EACH (node, &old_ports) {
1399 if (!shash_find(&new_ports, node->name)) {
1400 port_destroy(node->data);
1403 SHASH_FOR_EACH (node, &new_ports) {
1404 struct port *port = shash_find_data(&old_ports, node->name);
1406 port = port_create(br, node->name);
1408 port_reconfigure(port, node->data);
1410 shash_destroy(&old_ports);
1411 shash_destroy(&new_ports);
1413 /* Check and delete duplicate interfaces. */
1415 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1416 svec_destroy(&ifaces);
1418 /* Delete all flows if we're switching from connected to standalone or vice
1419 * versa. (XXX Should we delete all flows if we are switching from one
1420 * controller to another?) */
1423 /* Configure OpenFlow management listeners. */
1424 svec_init(&listeners);
1425 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1427 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1428 ovs_rundir, br->name));
1429 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1430 svec_clear(&listeners);
1432 svec_sort_unique(&listeners);
1434 svec_init(&old_listeners);
1435 ofproto_get_listeners(br->ofproto, &old_listeners);
1436 svec_sort_unique(&old_listeners);
1438 if (!svec_equal(&listeners, &old_listeners)) {
1439 ofproto_set_listeners(br->ofproto, &listeners);
1441 svec_destroy(&listeners);
1442 svec_destroy(&old_listeners);
1444 /* Configure OpenFlow controller connection snooping. */
1446 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1448 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1449 ovs_rundir, br->name));
1450 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1451 svec_clear(&snoops);
1453 svec_sort_unique(&snoops);
1455 svec_init(&old_snoops);
1456 ofproto_get_snoops(br->ofproto, &old_snoops);
1457 svec_sort_unique(&old_snoops);
1459 if (!svec_equal(&snoops, &old_snoops)) {
1460 ofproto_set_snoops(br->ofproto, &snoops);
1462 svec_destroy(&snoops);
1463 svec_destroy(&old_snoops);
1465 /* Default listener. */
1466 svec_init(&listeners);
1467 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1468 ovs_rundir, br->name));
1469 svec_init(&old_listeners);
1470 ofproto_get_listeners(br->ofproto, &old_listeners);
1471 if (!svec_equal(&listeners, &old_listeners)) {
1472 ofproto_set_listeners(br->ofproto, &listeners);
1474 svec_destroy(&listeners);
1475 svec_destroy(&old_listeners);
1477 /* Default snoop. */
1479 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1480 ovs_rundir, br->name));
1481 svec_init(&old_snoops);
1482 ofproto_get_snoops(br->ofproto, &old_snoops);
1483 if (!svec_equal(&snoops, &old_snoops)) {
1484 ofproto_set_snoops(br->ofproto, &snoops);
1486 svec_destroy(&snoops);
1487 svec_destroy(&old_snoops);
1490 mirror_reconfigure(br);
1492 bridge_update_desc(br);
1496 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1499 const struct ovsrec_controller *c;
1501 c = bridge_get_controller(ovs_cfg, br);
1502 if ((br->controller != NULL) != (c != NULL)) {
1503 ofproto_flush_flows(br->ofproto);
1505 free(br->controller);
1506 br->controller = c ? xstrdup(c->target) : NULL;
1509 int max_backoff, probe;
1510 int rate_limit, burst_limit;
1512 if (!strcmp(c->target, "discover")) {
1513 ofproto_set_discovery(br->ofproto, true,
1514 c->discover_accept_regex,
1515 c->discover_update_resolv_conf);
1517 struct iface *local_iface;
1521 in_band = (!c->connection_mode
1522 || !strcmp(c->connection_mode, "out-of-band"));
1523 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1524 ofproto_set_in_band(br->ofproto, in_band);
1526 local_iface = bridge_get_local_iface(br);
1527 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1528 struct netdev *netdev = local_iface->netdev;
1529 struct in_addr mask, gateway;
1531 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1534 if (!c->local_gateway
1535 || !inet_aton(c->local_gateway, &gateway)) {
1539 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1541 mask.s_addr = guess_netmask(ip.s_addr);
1543 if (!netdev_set_in4(netdev, ip, mask)) {
1544 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1546 br->name, IP_ARGS(&ip.s_addr),
1547 IP_ARGS(&mask.s_addr));
1550 if (gateway.s_addr) {
1551 if (!netdev_add_router(netdev, gateway)) {
1552 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1553 br->name, IP_ARGS(&gateway.s_addr));
1559 ofproto_set_failure(br->ofproto,
1561 || !strcmp(c->fail_mode, "standalone")
1562 || !strcmp(c->fail_mode, "open")));
1564 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1565 ofproto_set_probe_interval(br->ofproto, probe);
1567 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1568 ofproto_set_max_backoff(br->ofproto, max_backoff);
1570 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1571 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1572 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1574 union ofp_action action;
1577 /* Set up a flow that matches every packet and directs them to
1578 * OFPP_NORMAL (which goes to us). */
1579 memset(&action, 0, sizeof action);
1580 action.type = htons(OFPAT_OUTPUT);
1581 action.output.len = htons(sizeof action);
1582 action.output.port = htons(OFPP_NORMAL);
1583 memset(&flow, 0, sizeof flow);
1584 flow.wildcards = OFPFW_ALL;
1585 ofproto_add_flow(br->ofproto, &flow, &action, 1, 0);
1587 ofproto_set_in_band(br->ofproto, false);
1588 ofproto_set_max_backoff(br->ofproto, 1);
1589 ofproto_set_probe_interval(br->ofproto, 5);
1590 ofproto_set_failure(br->ofproto, false);
1593 ofproto_set_controller(br->ofproto, br->controller);
1597 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1602 for (i = 0; i < br->n_ports; i++) {
1603 struct port *port = br->ports[i];
1604 for (j = 0; j < port->n_ifaces; j++) {
1605 struct iface *iface = port->ifaces[j];
1606 shash_add_once(ifaces, iface->name, iface);
1608 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1609 shash_add_once(ifaces, port->name, NULL);
1614 /* For robustness, in case the administrator moves around datapath ports behind
1615 * our back, we re-check all the datapath port numbers here.
1617 * This function will set the 'xf_ifidx' members of interfaces that have
1618 * disappeared to -1, so only call this function from a context where those
1619 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1620 * 'xf_ifidx'es will cause trouble later when we try to send them to the
1621 * datapath, which doesn't support UINT16_MAX+1 ports. */
1623 bridge_fetch_dp_ifaces(struct bridge *br)
1625 struct xflow_port *xfif_ports;
1626 size_t n_xfif_ports;
1629 /* Reset all interface numbers. */
1630 for (i = 0; i < br->n_ports; i++) {
1631 struct port *port = br->ports[i];
1632 for (j = 0; j < port->n_ifaces; j++) {
1633 struct iface *iface = port->ifaces[j];
1634 iface->xf_ifidx = -1;
1637 port_array_clear(&br->ifaces);
1639 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
1640 for (i = 0; i < n_xfif_ports; i++) {
1641 struct xflow_port *p = &xfif_ports[i];
1642 struct iface *iface = iface_lookup(br, p->devname);
1644 if (iface->xf_ifidx >= 0) {
1645 VLOG_WARN("%s reported interface %s twice",
1646 xfif_name(br->xfif), p->devname);
1647 } else if (iface_from_xf_ifidx(br, p->port)) {
1648 VLOG_WARN("%s reported interface %"PRIu16" twice",
1649 xfif_name(br->xfif), p->port);
1651 port_array_set(&br->ifaces, p->port, iface);
1652 iface->xf_ifidx = p->port;
1656 int64_t ofport = (iface->xf_ifidx >= 0
1657 ? xflow_port_to_ofp_port(iface->xf_ifidx)
1659 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1666 /* Bridge packet processing functions. */
1669 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1671 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1674 static struct bond_entry *
1675 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1677 return &port->bond_hash[bond_hash(mac)];
1681 bond_choose_iface(const struct port *port)
1683 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1684 size_t i, best_down_slave = -1;
1685 long long next_delay_expiration = LLONG_MAX;
1687 for (i = 0; i < port->n_ifaces; i++) {
1688 struct iface *iface = port->ifaces[i];
1690 if (iface->enabled) {
1692 } else if (iface->delay_expires < next_delay_expiration) {
1693 best_down_slave = i;
1694 next_delay_expiration = iface->delay_expires;
1698 if (best_down_slave != -1) {
1699 struct iface *iface = port->ifaces[best_down_slave];
1701 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1702 "since no other interface is up", iface->name,
1703 iface->delay_expires - time_msec());
1704 bond_enable_slave(iface, true);
1707 return best_down_slave;
1711 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1712 uint16_t *xf_ifidx, tag_type *tags)
1714 struct iface *iface;
1716 assert(port->n_ifaces);
1717 if (port->n_ifaces == 1) {
1718 iface = port->ifaces[0];
1720 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1721 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1722 || !port->ifaces[e->iface_idx]->enabled) {
1723 /* XXX select interface properly. The current interface selection
1724 * is only good for testing the rebalancing code. */
1725 e->iface_idx = bond_choose_iface(port);
1726 if (e->iface_idx < 0) {
1727 *tags |= port->no_ifaces_tag;
1730 e->iface_tag = tag_create_random();
1731 ((struct port *) port)->bond_compat_is_stale = true;
1733 *tags |= e->iface_tag;
1734 iface = port->ifaces[e->iface_idx];
1736 *xf_ifidx = iface->xf_ifidx;
1737 *tags |= iface->tag; /* Currently only used for bonding. */
1742 bond_link_status_update(struct iface *iface, bool carrier)
1744 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1745 struct port *port = iface->port;
1747 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1748 /* Nothing to do. */
1751 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1752 iface->name, carrier ? "detected" : "dropped");
1753 if (carrier == iface->enabled) {
1754 iface->delay_expires = LLONG_MAX;
1755 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1756 iface->name, carrier ? "disabled" : "enabled");
1757 } else if (carrier && port->active_iface < 0) {
1758 bond_enable_slave(iface, true);
1759 if (port->updelay) {
1760 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1761 "other interface is up", iface->name, port->updelay);
1764 int delay = carrier ? port->updelay : port->downdelay;
1765 iface->delay_expires = time_msec() + delay;
1768 "interface %s: will be %s if it stays %s for %d ms",
1770 carrier ? "enabled" : "disabled",
1771 carrier ? "up" : "down",
1778 bond_choose_active_iface(struct port *port)
1780 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1782 port->active_iface = bond_choose_iface(port);
1783 port->active_iface_tag = tag_create_random();
1784 if (port->active_iface >= 0) {
1785 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1786 port->name, port->ifaces[port->active_iface]->name);
1788 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1794 bond_enable_slave(struct iface *iface, bool enable)
1796 struct port *port = iface->port;
1797 struct bridge *br = port->bridge;
1799 /* This acts as a recursion check. If the act of disabling a slave
1800 * causes a different slave to be enabled, the flag will allow us to
1801 * skip redundant work when we reenter this function. It must be
1802 * cleared on exit to keep things safe with multiple bonds. */
1803 static bool moving_active_iface = false;
1805 iface->delay_expires = LLONG_MAX;
1806 if (enable == iface->enabled) {
1810 iface->enabled = enable;
1811 if (!iface->enabled) {
1812 VLOG_WARN("interface %s: disabled", iface->name);
1813 ofproto_revalidate(br->ofproto, iface->tag);
1814 if (iface->port_ifidx == port->active_iface) {
1815 ofproto_revalidate(br->ofproto,
1816 port->active_iface_tag);
1818 /* Disabling a slave can lead to another slave being immediately
1819 * enabled if there will be no active slaves but one is waiting
1820 * on an updelay. In this case we do not need to run most of the
1821 * code for the newly enabled slave since there was no period
1822 * without an active slave and it is redundant with the disabling
1824 moving_active_iface = true;
1825 bond_choose_active_iface(port);
1827 bond_send_learning_packets(port);
1829 VLOG_WARN("interface %s: enabled", iface->name);
1830 if (port->active_iface < 0 && !moving_active_iface) {
1831 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1832 bond_choose_active_iface(port);
1833 bond_send_learning_packets(port);
1835 iface->tag = tag_create_random();
1838 moving_active_iface = false;
1839 port->bond_compat_is_stale = true;
1843 bond_run(struct bridge *br)
1847 for (i = 0; i < br->n_ports; i++) {
1848 struct port *port = br->ports[i];
1850 if (port->n_ifaces >= 2) {
1851 for (j = 0; j < port->n_ifaces; j++) {
1852 struct iface *iface = port->ifaces[j];
1853 if (time_msec() >= iface->delay_expires) {
1854 bond_enable_slave(iface, !iface->enabled);
1859 if (port->bond_compat_is_stale) {
1860 port->bond_compat_is_stale = false;
1861 port_update_bond_compat(port);
1867 bond_wait(struct bridge *br)
1871 for (i = 0; i < br->n_ports; i++) {
1872 struct port *port = br->ports[i];
1873 if (port->n_ifaces < 2) {
1876 for (j = 0; j < port->n_ifaces; j++) {
1877 struct iface *iface = port->ifaces[j];
1878 if (iface->delay_expires != LLONG_MAX) {
1879 poll_timer_wait(iface->delay_expires - time_msec());
1886 set_dst(struct dst *p, const flow_t *flow,
1887 const struct port *in_port, const struct port *out_port,
1890 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1891 : in_port->vlan >= 0 ? in_port->vlan
1892 : ntohs(flow->dl_vlan));
1893 return choose_output_iface(out_port, flow->dl_src, &p->xf_ifidx, tags);
1897 swap_dst(struct dst *p, struct dst *q)
1899 struct dst tmp = *p;
1904 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1905 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1906 * that we push to the datapath. We could in fact fully sort the array by
1907 * vlan, but in most cases there are at most two different vlan tags so that's
1908 * possibly overkill.) */
1910 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1912 struct dst *first = dsts;
1913 struct dst *last = dsts + n_dsts;
1915 while (first != last) {
1917 * - All dsts < first have vlan == 'vlan'.
1918 * - All dsts >= last have vlan != 'vlan'.
1919 * - first < last. */
1920 while (first->vlan == vlan) {
1921 if (++first == last) {
1926 /* Same invariants, plus one additional:
1927 * - first->vlan != vlan.
1929 while (last[-1].vlan != vlan) {
1930 if (--last == first) {
1935 /* Same invariants, plus one additional:
1936 * - last[-1].vlan == vlan.*/
1937 swap_dst(first++, --last);
1942 mirror_mask_ffs(mirror_mask_t mask)
1944 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1949 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1950 const struct dst *test)
1953 for (i = 0; i < n_dsts; i++) {
1954 if (dsts[i].vlan == test->vlan && dsts[i].xf_ifidx == test->xf_ifidx) {
1962 port_trunks_vlan(const struct port *port, uint16_t vlan)
1964 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1968 port_includes_vlan(const struct port *port, uint16_t vlan)
1970 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1974 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1975 const struct port *in_port, const struct port *out_port,
1976 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1978 mirror_mask_t mirrors = in_port->src_mirrors;
1979 struct dst *dst = dsts;
1982 if (out_port == FLOOD_PORT) {
1983 /* XXX use XFLOW_FLOOD if no vlans or bonding. */
1984 /* XXX even better, define each VLAN as a datapath port group */
1985 for (i = 0; i < br->n_ports; i++) {
1986 struct port *port = br->ports[i];
1987 if (port != in_port && port_includes_vlan(port, vlan)
1988 && !port->is_mirror_output_port
1989 && set_dst(dst, flow, in_port, port, tags)) {
1990 mirrors |= port->dst_mirrors;
1994 *nf_output_iface = NF_OUT_FLOOD;
1995 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1996 *nf_output_iface = dst->xf_ifidx;
1997 mirrors |= out_port->dst_mirrors;
2002 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2003 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2005 if (set_dst(dst, flow, in_port, m->out_port, tags)
2006 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2010 for (i = 0; i < br->n_ports; i++) {
2011 struct port *port = br->ports[i];
2012 if (port_includes_vlan(port, m->out_vlan)
2013 && set_dst(dst, flow, in_port, port, tags))
2017 if (port->vlan < 0) {
2018 dst->vlan = m->out_vlan;
2020 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2024 /* Use the vlan tag on the original flow instead of
2025 * the one passed in the vlan parameter. This ensures
2026 * that we compare the vlan from before any implicit
2027 * tagging tags place. This is necessary because
2028 * dst->vlan is the final vlan, after removing implicit
2030 flow_vlan = ntohs(flow->dl_vlan);
2031 if (flow_vlan == 0) {
2032 flow_vlan = OFP_VLAN_NONE;
2034 if (port == in_port && dst->vlan == flow_vlan) {
2035 /* Don't send out input port on same VLAN. */
2043 mirrors &= mirrors - 1;
2046 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2050 static void OVS_UNUSED
2051 print_dsts(const struct dst *dsts, size_t n)
2053 for (; n--; dsts++) {
2054 printf(">p%"PRIu16, dsts->xf_ifidx);
2055 if (dsts->vlan != OFP_VLAN_NONE) {
2056 printf("v%"PRIu16, dsts->vlan);
2062 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2063 const struct port *in_port, const struct port *out_port,
2064 tag_type *tags, struct xflow_actions *actions,
2065 uint16_t *nf_output_iface)
2067 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2069 const struct dst *p;
2072 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2075 cur_vlan = ntohs(flow->dl_vlan);
2076 for (p = dsts; p < &dsts[n_dsts]; p++) {
2077 union xflow_action *a;
2078 if (p->vlan != cur_vlan) {
2079 if (p->vlan == OFP_VLAN_NONE) {
2080 xflow_actions_add(actions, XFLOWAT_STRIP_VLAN);
2082 a = xflow_actions_add(actions, XFLOWAT_SET_DL_TCI);
2083 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2084 a->dl_tci.mask = htons(VLAN_VID_MASK);
2088 a = xflow_actions_add(actions, XFLOWAT_OUTPUT);
2089 a->output.port = p->xf_ifidx;
2093 /* Returns the effective vlan of a packet, taking into account both the
2094 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2095 * the packet is untagged and -1 indicates it has an invalid header and
2096 * should be dropped. */
2097 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2098 struct port *in_port, bool have_packet)
2100 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2101 * belongs to VLAN 0, so we should treat both cases identically. (In the
2102 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2103 * presumably to allow a priority to be specified. In the latter case, the
2104 * packet does not have any 802.1Q header.) */
2105 int vlan = ntohs(flow->dl_vlan);
2106 if (vlan == OFP_VLAN_NONE) {
2109 if (in_port->vlan >= 0) {
2111 /* XXX support double tagging? */
2113 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2114 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2115 "packet received on port %s configured with "
2116 "implicit VLAN %"PRIu16,
2117 br->name, ntohs(flow->dl_vlan),
2118 in_port->name, in_port->vlan);
2122 vlan = in_port->vlan;
2124 if (!port_includes_vlan(in_port, vlan)) {
2126 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2127 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2128 "packet received on port %s not configured for "
2130 br->name, vlan, in_port->name, vlan);
2140 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2141 struct port *in_port)
2143 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2144 vlan, in_port->port_idx);
2146 /* The log messages here could actually be useful in debugging,
2147 * so keep the rate limit relatively high. */
2148 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2150 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2151 "on port %s in VLAN %d",
2152 br->name, ETH_ADDR_ARGS(flow->dl_src),
2153 in_port->name, vlan);
2154 ofproto_revalidate(br->ofproto, rev_tag);
2159 is_bcast_arp_reply(const flow_t *flow)
2161 return (flow->dl_type == htons(ETH_TYPE_ARP)
2162 && flow->nw_proto == ARP_OP_REPLY
2163 && eth_addr_is_broadcast(flow->dl_dst));
2166 /* If the composed actions may be applied to any packet in the given 'flow',
2167 * returns true. Otherwise, the actions should only be applied to 'packet', or
2168 * not at all, if 'packet' was NULL. */
2170 process_flow(struct bridge *br, const flow_t *flow,
2171 const struct ofpbuf *packet, struct xflow_actions *actions,
2172 tag_type *tags, uint16_t *nf_output_iface)
2174 struct iface *in_iface;
2175 struct port *in_port;
2176 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2180 /* Find the interface and port structure for the received packet. */
2181 in_iface = iface_from_xf_ifidx(br, flow->in_port);
2183 /* No interface? Something fishy... */
2184 if (packet != NULL) {
2185 /* Odd. A few possible reasons here:
2187 * - We deleted an interface but there are still a few packets
2188 * queued up from it.
2190 * - Someone externally added an interface (e.g. with "ovs-dpctl
2191 * add-if") that we don't know about.
2193 * - Packet arrived on the local port but the local port is not
2194 * one of our bridge ports.
2196 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2198 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2199 "interface %"PRIu16, br->name, flow->in_port);
2202 /* Return without adding any actions, to drop packets on this flow. */
2205 in_port = in_iface->port;
2206 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2211 /* Drop frames for reserved multicast addresses. */
2212 if (eth_addr_is_reserved(flow->dl_dst)) {
2216 /* Drop frames on ports reserved for mirroring. */
2217 if (in_port->is_mirror_output_port) {
2218 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2219 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2220 "which is reserved exclusively for mirroring",
2221 br->name, in_port->name);
2225 /* Packets received on bonds need special attention to avoid duplicates. */
2226 if (in_port->n_ifaces > 1) {
2229 if (eth_addr_is_multicast(flow->dl_dst)) {
2230 *tags |= in_port->active_iface_tag;
2231 if (in_port->active_iface != in_iface->port_ifidx) {
2232 /* Drop all multicast packets on inactive slaves. */
2237 /* Drop all packets for which we have learned a different input
2238 * port, because we probably sent the packet on one slave and got
2239 * it back on the other. Broadcast ARP replies are an exception
2240 * to this rule: the host has moved to another switch. */
2241 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2242 if (src_idx != -1 && src_idx != in_port->port_idx &&
2243 !is_bcast_arp_reply(flow)) {
2249 out_port = FLOOD_PORT;
2250 /* Learn source MAC (but don't try to learn from revalidation). */
2252 update_learning_table(br, flow, vlan, in_port);
2255 /* Determine output port. */
2256 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2258 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2259 out_port = br->ports[out_port_idx];
2260 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2261 /* If we are revalidating but don't have a learning entry then
2262 * eject the flow. Installing a flow that floods packets opens
2263 * up a window of time where we could learn from a packet reflected
2264 * on a bond and blackhole packets before the learning table is
2265 * updated to reflect the correct port. */
2269 /* Don't send packets out their input ports. */
2270 if (in_port == out_port) {
2275 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2281 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2284 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2285 const struct ofp_phy_port *opp,
2288 struct bridge *br = br_;
2289 struct iface *iface;
2292 iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(opp->port_no));
2298 if (reason == OFPPR_DELETE) {
2299 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2300 br->name, iface->name);
2301 iface_destroy(iface);
2302 if (!port->n_ifaces) {
2303 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2304 br->name, port->name);
2310 if (port->n_ifaces > 1) {
2311 bool up = !(opp->state & OFPPS_LINK_DOWN);
2312 bond_link_status_update(iface, up);
2313 port_update_bond_compat(port);
2319 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2320 struct xflow_actions *actions, tag_type *tags,
2321 uint16_t *nf_output_iface, void *br_)
2323 struct bridge *br = br_;
2325 COVERAGE_INC(bridge_process_flow);
2326 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2330 bridge_account_flow_ofhook_cb(const flow_t *flow,
2331 const union xflow_action *actions,
2332 size_t n_actions, unsigned long long int n_bytes,
2335 struct bridge *br = br_;
2336 struct port *in_port;
2337 const union xflow_action *a;
2339 /* Feed information from the active flows back into the learning table
2340 * to ensure that table is always in sync with what is actually flowing
2341 * through the datapath. */
2342 in_port = port_from_xf_ifidx(br, flow->in_port);
2344 int vlan = flow_get_vlan(br, flow, in_port, false);
2346 update_learning_table(br, flow, vlan, in_port);
2350 if (!br->has_bonded_ports) {
2354 for (a = actions; a < &actions[n_actions]; a++) {
2355 if (a->type == XFLOWAT_OUTPUT) {
2356 struct port *out_port = port_from_xf_ifidx(br, a->output.port);
2357 if (out_port && out_port->n_ifaces >= 2) {
2358 struct bond_entry *e = lookup_bond_entry(out_port,
2360 e->tx_bytes += n_bytes;
2367 bridge_account_checkpoint_ofhook_cb(void *br_)
2369 struct bridge *br = br_;
2372 if (!br->has_bonded_ports) {
2376 /* The current ofproto implementation calls this callback at least once a
2377 * second, so this timer implementation is sufficient. */
2378 if (time_msec() < br->bond_next_rebalance) {
2381 br->bond_next_rebalance = time_msec() + 10000;
2383 for (i = 0; i < br->n_ports; i++) {
2384 struct port *port = br->ports[i];
2385 if (port->n_ifaces > 1) {
2386 bond_rebalance_port(port);
2391 static struct ofhooks bridge_ofhooks = {
2392 bridge_port_changed_ofhook_cb,
2393 bridge_normal_ofhook_cb,
2394 bridge_account_flow_ofhook_cb,
2395 bridge_account_checkpoint_ofhook_cb,
2398 /* Bonding functions. */
2400 /* Statistics for a single interface on a bonded port, used for load-based
2401 * bond rebalancing. */
2402 struct slave_balance {
2403 struct iface *iface; /* The interface. */
2404 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2406 /* All the "bond_entry"s that are assigned to this interface, in order of
2407 * increasing tx_bytes. */
2408 struct bond_entry **hashes;
2412 /* Sorts pointers to pointers to bond_entries in ascending order by the
2413 * interface to which they are assigned, and within a single interface in
2414 * ascending order of bytes transmitted. */
2416 compare_bond_entries(const void *a_, const void *b_)
2418 const struct bond_entry *const *ap = a_;
2419 const struct bond_entry *const *bp = b_;
2420 const struct bond_entry *a = *ap;
2421 const struct bond_entry *b = *bp;
2422 if (a->iface_idx != b->iface_idx) {
2423 return a->iface_idx > b->iface_idx ? 1 : -1;
2424 } else if (a->tx_bytes != b->tx_bytes) {
2425 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2431 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2432 * *descending* order by number of bytes transmitted. */
2434 compare_slave_balance(const void *a_, const void *b_)
2436 const struct slave_balance *a = a_;
2437 const struct slave_balance *b = b_;
2438 if (a->iface->enabled != b->iface->enabled) {
2439 return a->iface->enabled ? -1 : 1;
2440 } else if (a->tx_bytes != b->tx_bytes) {
2441 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2448 swap_bals(struct slave_balance *a, struct slave_balance *b)
2450 struct slave_balance tmp = *a;
2455 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2456 * given that 'p' (and only 'p') might be in the wrong location.
2458 * This function invalidates 'p', since it might now be in a different memory
2461 resort_bals(struct slave_balance *p,
2462 struct slave_balance bals[], size_t n_bals)
2465 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2466 swap_bals(p, p - 1);
2468 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2469 swap_bals(p, p + 1);
2475 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2477 if (VLOG_IS_DBG_ENABLED()) {
2478 struct ds ds = DS_EMPTY_INITIALIZER;
2479 const struct slave_balance *b;
2481 for (b = bals; b < bals + n_bals; b++) {
2485 ds_put_char(&ds, ',');
2487 ds_put_format(&ds, " %s %"PRIu64"kB",
2488 b->iface->name, b->tx_bytes / 1024);
2490 if (!b->iface->enabled) {
2491 ds_put_cstr(&ds, " (disabled)");
2493 if (b->n_hashes > 0) {
2494 ds_put_cstr(&ds, " (");
2495 for (i = 0; i < b->n_hashes; i++) {
2496 const struct bond_entry *e = b->hashes[i];
2498 ds_put_cstr(&ds, " + ");
2500 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2501 e - port->bond_hash, e->tx_bytes / 1024);
2503 ds_put_cstr(&ds, ")");
2506 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2511 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2513 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2516 struct bond_entry *hash = from->hashes[hash_idx];
2517 struct port *port = from->iface->port;
2518 uint64_t delta = hash->tx_bytes;
2520 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2521 "from %s to %s (now carrying %"PRIu64"kB and "
2522 "%"PRIu64"kB load, respectively)",
2523 port->name, delta / 1024, hash - port->bond_hash,
2524 from->iface->name, to->iface->name,
2525 (from->tx_bytes - delta) / 1024,
2526 (to->tx_bytes + delta) / 1024);
2528 /* Delete element from from->hashes.
2530 * We don't bother to add the element to to->hashes because not only would
2531 * it require more work, the only purpose it would be to allow that hash to
2532 * be migrated to another slave in this rebalancing run, and there is no
2533 * point in doing that. */
2534 if (hash_idx == 0) {
2537 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2538 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2542 /* Shift load away from 'from' to 'to'. */
2543 from->tx_bytes -= delta;
2544 to->tx_bytes += delta;
2546 /* Arrange for flows to be revalidated. */
2547 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2548 hash->iface_idx = to->iface->port_ifidx;
2549 hash->iface_tag = tag_create_random();
2553 bond_rebalance_port(struct port *port)
2555 struct slave_balance bals[DP_MAX_PORTS];
2557 struct bond_entry *hashes[BOND_MASK + 1];
2558 struct slave_balance *b, *from, *to;
2559 struct bond_entry *e;
2562 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2563 * descending order of tx_bytes, so that bals[0] represents the most
2564 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2567 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2568 * array for each slave_balance structure, we sort our local array of
2569 * hashes in order by slave, so that all of the hashes for a given slave
2570 * become contiguous in memory, and then we point each 'hashes' members of
2571 * a slave_balance structure to the start of a contiguous group. */
2572 n_bals = port->n_ifaces;
2573 for (b = bals; b < &bals[n_bals]; b++) {
2574 b->iface = port->ifaces[b - bals];
2579 for (i = 0; i <= BOND_MASK; i++) {
2580 hashes[i] = &port->bond_hash[i];
2582 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2583 for (i = 0; i <= BOND_MASK; i++) {
2585 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2586 b = &bals[e->iface_idx];
2587 b->tx_bytes += e->tx_bytes;
2589 b->hashes = &hashes[i];
2594 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2595 log_bals(bals, n_bals, port);
2597 /* Discard slaves that aren't enabled (which were sorted to the back of the
2598 * array earlier). */
2599 while (!bals[n_bals - 1].iface->enabled) {
2606 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2607 to = &bals[n_bals - 1];
2608 for (from = bals; from < to; ) {
2609 uint64_t overload = from->tx_bytes - to->tx_bytes;
2610 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2611 /* The extra load on 'from' (and all less-loaded slaves), compared
2612 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2613 * it is less than ~1Mbps. No point in rebalancing. */
2615 } else if (from->n_hashes == 1) {
2616 /* 'from' only carries a single MAC hash, so we can't shift any
2617 * load away from it, even though we want to. */
2620 /* 'from' is carrying significantly more load than 'to', and that
2621 * load is split across at least two different hashes. Pick a hash
2622 * to migrate to 'to' (the least-loaded slave), given that doing so
2623 * must decrease the ratio of the load on the two slaves by at
2626 * The sort order we use means that we prefer to shift away the
2627 * smallest hashes instead of the biggest ones. There is little
2628 * reason behind this decision; we could use the opposite sort
2629 * order to shift away big hashes ahead of small ones. */
2633 for (i = 0; i < from->n_hashes; i++) {
2634 double old_ratio, new_ratio;
2635 uint64_t delta = from->hashes[i]->tx_bytes;
2637 if (delta == 0 || from->tx_bytes - delta == 0) {
2638 /* Pointless move. */
2642 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2644 if (to->tx_bytes == 0) {
2645 /* Nothing on the new slave, move it. */
2649 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2650 new_ratio = (double)(from->tx_bytes - delta) /
2651 (to->tx_bytes + delta);
2653 if (new_ratio == 0) {
2654 /* Should already be covered but check to prevent division
2659 if (new_ratio < 1) {
2660 new_ratio = 1 / new_ratio;
2663 if (old_ratio - new_ratio > 0.1) {
2664 /* Would decrease the ratio, move it. */
2668 if (i < from->n_hashes) {
2669 bond_shift_load(from, to, i);
2670 port->bond_compat_is_stale = true;
2672 /* If the result of the migration changed the relative order of
2673 * 'from' and 'to' swap them back to maintain invariants. */
2674 if (order_swapped) {
2675 swap_bals(from, to);
2678 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2679 * point to different slave_balance structures. It is only
2680 * valid to do these two operations in a row at all because we
2681 * know that 'from' will not move past 'to' and vice versa. */
2682 resort_bals(from, bals, n_bals);
2683 resort_bals(to, bals, n_bals);
2690 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2691 * historical data to decay to <1% in 7 rebalancing runs. */
2692 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2698 bond_send_learning_packets(struct port *port)
2700 struct bridge *br = port->bridge;
2701 struct mac_entry *e;
2702 struct ofpbuf packet;
2703 int error, n_packets, n_errors;
2705 if (!port->n_ifaces || port->active_iface < 0) {
2709 ofpbuf_init(&packet, 128);
2710 error = n_packets = n_errors = 0;
2711 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2712 union ofp_action actions[2], *a;
2718 if (e->port == port->port_idx
2719 || !choose_output_iface(port, e->mac, &xf_ifidx, &tags)) {
2723 /* Compose actions. */
2724 memset(actions, 0, sizeof actions);
2727 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2728 a->vlan_vid.len = htons(sizeof *a);
2729 a->vlan_vid.vlan_vid = htons(e->vlan);
2732 a->output.type = htons(OFPAT_OUTPUT);
2733 a->output.len = htons(sizeof *a);
2734 a->output.port = htons(xflow_port_to_ofp_port(xf_ifidx));
2739 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2741 flow_extract(&packet, XFLOWP_NONE, &flow);
2742 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2749 ofpbuf_uninit(&packet);
2752 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2753 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2754 "packets, last error was: %s",
2755 port->name, n_errors, n_packets, strerror(error));
2757 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2758 port->name, n_packets);
2762 /* Bonding unixctl user interface functions. */
2765 bond_unixctl_list(struct unixctl_conn *conn,
2766 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2768 struct ds ds = DS_EMPTY_INITIALIZER;
2769 const struct bridge *br;
2771 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2773 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2776 for (i = 0; i < br->n_ports; i++) {
2777 const struct port *port = br->ports[i];
2778 if (port->n_ifaces > 1) {
2781 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2782 for (j = 0; j < port->n_ifaces; j++) {
2783 const struct iface *iface = port->ifaces[j];
2785 ds_put_cstr(&ds, ", ");
2787 ds_put_cstr(&ds, iface->name);
2789 ds_put_char(&ds, '\n');
2793 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2797 static struct port *
2798 bond_find(const char *name)
2800 const struct bridge *br;
2802 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2805 for (i = 0; i < br->n_ports; i++) {
2806 struct port *port = br->ports[i];
2807 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2816 bond_unixctl_show(struct unixctl_conn *conn,
2817 const char *args, void *aux OVS_UNUSED)
2819 struct ds ds = DS_EMPTY_INITIALIZER;
2820 const struct port *port;
2823 port = bond_find(args);
2825 unixctl_command_reply(conn, 501, "no such bond");
2829 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2830 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2831 ds_put_format(&ds, "next rebalance: %lld ms\n",
2832 port->bridge->bond_next_rebalance - time_msec());
2833 for (j = 0; j < port->n_ifaces; j++) {
2834 const struct iface *iface = port->ifaces[j];
2835 struct bond_entry *be;
2838 ds_put_format(&ds, "slave %s: %s\n",
2839 iface->name, iface->enabled ? "enabled" : "disabled");
2840 if (j == port->active_iface) {
2841 ds_put_cstr(&ds, "\tactive slave\n");
2843 if (iface->delay_expires != LLONG_MAX) {
2844 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2845 iface->enabled ? "downdelay" : "updelay",
2846 iface->delay_expires - time_msec());
2850 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2851 int hash = be - port->bond_hash;
2852 struct mac_entry *me;
2854 if (be->iface_idx != j) {
2858 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2859 hash, be->tx_bytes / 1024);
2862 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2863 &port->bridge->ml->lrus) {
2866 if (bond_hash(me->mac) == hash
2867 && me->port != port->port_idx
2868 && choose_output_iface(port, me->mac, &xf_ifidx, &tags)
2869 && xf_ifidx == iface->xf_ifidx)
2871 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2872 ETH_ADDR_ARGS(me->mac));
2877 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2882 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2883 void *aux OVS_UNUSED)
2885 char *args = (char *) args_;
2886 char *save_ptr = NULL;
2887 char *bond_s, *hash_s, *slave_s;
2888 uint8_t mac[ETH_ADDR_LEN];
2890 struct iface *iface;
2891 struct bond_entry *entry;
2894 bond_s = strtok_r(args, " ", &save_ptr);
2895 hash_s = strtok_r(NULL, " ", &save_ptr);
2896 slave_s = strtok_r(NULL, " ", &save_ptr);
2898 unixctl_command_reply(conn, 501,
2899 "usage: bond/migrate BOND HASH SLAVE");
2903 port = bond_find(bond_s);
2905 unixctl_command_reply(conn, 501, "no such bond");
2909 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2910 == ETH_ADDR_SCAN_COUNT) {
2911 hash = bond_hash(mac);
2912 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2913 hash = atoi(hash_s) & BOND_MASK;
2915 unixctl_command_reply(conn, 501, "bad hash");
2919 iface = port_lookup_iface(port, slave_s);
2921 unixctl_command_reply(conn, 501, "no such slave");
2925 if (!iface->enabled) {
2926 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2930 entry = &port->bond_hash[hash];
2931 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2932 entry->iface_idx = iface->port_ifidx;
2933 entry->iface_tag = tag_create_random();
2934 port->bond_compat_is_stale = true;
2935 unixctl_command_reply(conn, 200, "migrated");
2939 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2940 void *aux OVS_UNUSED)
2942 char *args = (char *) args_;
2943 char *save_ptr = NULL;
2944 char *bond_s, *slave_s;
2946 struct iface *iface;
2948 bond_s = strtok_r(args, " ", &save_ptr);
2949 slave_s = strtok_r(NULL, " ", &save_ptr);
2951 unixctl_command_reply(conn, 501,
2952 "usage: bond/set-active-slave BOND SLAVE");
2956 port = bond_find(bond_s);
2958 unixctl_command_reply(conn, 501, "no such bond");
2962 iface = port_lookup_iface(port, slave_s);
2964 unixctl_command_reply(conn, 501, "no such slave");
2968 if (!iface->enabled) {
2969 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2973 if (port->active_iface != iface->port_ifidx) {
2974 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2975 port->active_iface = iface->port_ifidx;
2976 port->active_iface_tag = tag_create_random();
2977 VLOG_INFO("port %s: active interface is now %s",
2978 port->name, iface->name);
2979 bond_send_learning_packets(port);
2980 unixctl_command_reply(conn, 200, "done");
2982 unixctl_command_reply(conn, 200, "no change");
2987 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2989 char *args = (char *) args_;
2990 char *save_ptr = NULL;
2991 char *bond_s, *slave_s;
2993 struct iface *iface;
2995 bond_s = strtok_r(args, " ", &save_ptr);
2996 slave_s = strtok_r(NULL, " ", &save_ptr);
2998 unixctl_command_reply(conn, 501,
2999 "usage: bond/enable/disable-slave BOND SLAVE");
3003 port = bond_find(bond_s);
3005 unixctl_command_reply(conn, 501, "no such bond");
3009 iface = port_lookup_iface(port, slave_s);
3011 unixctl_command_reply(conn, 501, "no such slave");
3015 bond_enable_slave(iface, enable);
3016 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3020 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3021 void *aux OVS_UNUSED)
3023 enable_slave(conn, args, true);
3027 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3028 void *aux OVS_UNUSED)
3030 enable_slave(conn, args, false);
3034 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3035 void *aux OVS_UNUSED)
3037 uint8_t mac[ETH_ADDR_LEN];
3041 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3042 == ETH_ADDR_SCAN_COUNT) {
3043 hash = bond_hash(mac);
3045 hash_cstr = xasprintf("%u", hash);
3046 unixctl_command_reply(conn, 200, hash_cstr);
3049 unixctl_command_reply(conn, 501, "invalid mac");
3056 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3057 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3058 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3059 unixctl_command_register("bond/set-active-slave",
3060 bond_unixctl_set_active_slave, NULL);
3061 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3063 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3065 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3068 /* Port functions. */
3070 static struct port *
3071 port_create(struct bridge *br, const char *name)
3075 port = xzalloc(sizeof *port);
3077 port->port_idx = br->n_ports;
3079 port->trunks = NULL;
3080 port->name = xstrdup(name);
3081 port->active_iface = -1;
3083 if (br->n_ports >= br->allocated_ports) {
3084 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3087 br->ports[br->n_ports++] = port;
3089 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3096 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3098 struct shash old_ifaces, new_ifaces;
3099 struct shash_node *node;
3100 unsigned long *trunks;
3106 /* Collect old and new interfaces. */
3107 shash_init(&old_ifaces);
3108 shash_init(&new_ifaces);
3109 for (i = 0; i < port->n_ifaces; i++) {
3110 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3112 for (i = 0; i < cfg->n_interfaces; i++) {
3113 const char *name = cfg->interfaces[i]->name;
3114 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3115 VLOG_WARN("port %s: %s specified twice as port interface",
3119 port->updelay = cfg->bond_updelay;
3120 if (port->updelay < 0) {
3123 port->updelay = cfg->bond_downdelay;
3124 if (port->downdelay < 0) {
3125 port->downdelay = 0;
3128 /* Get rid of deleted interfaces and add new interfaces. */
3129 SHASH_FOR_EACH (node, &old_ifaces) {
3130 if (!shash_find(&new_ifaces, node->name)) {
3131 iface_destroy(node->data);
3134 SHASH_FOR_EACH (node, &new_ifaces) {
3135 const struct ovsrec_interface *if_cfg = node->data;
3136 struct iface *iface;
3138 iface = shash_find_data(&old_ifaces, if_cfg->name);
3140 iface_create(port, if_cfg);
3142 iface->cfg = if_cfg;
3149 if (port->n_ifaces < 2) {
3151 if (vlan >= 0 && vlan <= 4095) {
3152 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3157 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3158 * they even work as-is. But they have not been tested. */
3159 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3163 if (port->vlan != vlan) {
3165 bridge_flush(port->bridge);
3168 /* Get trunked VLANs. */
3174 trunks = bitmap_allocate(4096);
3176 for (i = 0; i < cfg->n_trunks; i++) {
3177 int trunk = cfg->trunks[i];
3179 bitmap_set1(trunks, trunk);
3185 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3186 port->name, cfg->n_trunks);
3188 if (n_errors == cfg->n_trunks) {
3190 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3193 bitmap_set_multiple(trunks, 0, 4096, 1);
3196 if (cfg->n_trunks) {
3197 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3202 ? port->trunks != NULL
3203 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3204 bridge_flush(port->bridge);
3206 bitmap_free(port->trunks);
3207 port->trunks = trunks;
3209 shash_destroy(&old_ifaces);
3210 shash_destroy(&new_ifaces);
3214 port_destroy(struct port *port)
3217 struct bridge *br = port->bridge;
3221 proc_net_compat_update_vlan(port->name, NULL, 0);
3222 proc_net_compat_update_bond(port->name, NULL);
3224 for (i = 0; i < MAX_MIRRORS; i++) {
3225 struct mirror *m = br->mirrors[i];
3226 if (m && m->out_port == port) {
3231 while (port->n_ifaces > 0) {
3232 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3235 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3236 del->port_idx = port->port_idx;
3239 bitmap_free(port->trunks);
3246 static struct port *
3247 port_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3249 struct iface *iface = iface_from_xf_ifidx(br, xf_ifidx);
3250 return iface ? iface->port : NULL;
3253 static struct port *
3254 port_lookup(const struct bridge *br, const char *name)
3258 for (i = 0; i < br->n_ports; i++) {
3259 struct port *port = br->ports[i];
3260 if (!strcmp(port->name, name)) {
3267 static struct iface *
3268 port_lookup_iface(const struct port *port, const char *name)
3272 for (j = 0; j < port->n_ifaces; j++) {
3273 struct iface *iface = port->ifaces[j];
3274 if (!strcmp(iface->name, name)) {
3282 port_update_bonding(struct port *port)
3284 if (port->n_ifaces < 2) {
3285 /* Not a bonded port. */
3286 if (port->bond_hash) {
3287 free(port->bond_hash);
3288 port->bond_hash = NULL;
3289 port->bond_compat_is_stale = true;
3290 port->bond_fake_iface = false;
3293 if (!port->bond_hash) {
3296 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3297 for (i = 0; i <= BOND_MASK; i++) {
3298 struct bond_entry *e = &port->bond_hash[i];
3302 port->no_ifaces_tag = tag_create_random();
3303 bond_choose_active_iface(port);
3305 port->bond_compat_is_stale = true;
3306 port->bond_fake_iface = port->cfg->bond_fake_iface;
3311 port_update_bond_compat(struct port *port)
3313 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3314 struct compat_bond bond;
3317 if (port->n_ifaces < 2) {
3318 proc_net_compat_update_bond(port->name, NULL);
3323 bond.updelay = port->updelay;
3324 bond.downdelay = port->downdelay;
3327 bond.hashes = compat_hashes;
3328 if (port->bond_hash) {
3329 const struct bond_entry *e;
3330 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3331 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3332 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3333 cbh->hash = e - port->bond_hash;
3334 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3339 bond.n_slaves = port->n_ifaces;
3340 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3341 for (i = 0; i < port->n_ifaces; i++) {
3342 struct iface *iface = port->ifaces[i];
3343 struct compat_bond_slave *slave = &bond.slaves[i];
3344 slave->name = iface->name;
3346 /* We need to make the same determination as the Linux bonding
3347 * code to determine whether a slave should be consider "up".
3348 * The Linux function bond_miimon_inspect() supports four
3349 * BOND_LINK_* states:
3351 * - BOND_LINK_UP: carrier detected, updelay has passed.
3352 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3353 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3354 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3356 * The function bond_info_show_slave() only considers BOND_LINK_UP
3357 * to be "up" and anything else to be "down".
3359 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3363 netdev_get_etheraddr(iface->netdev, slave->mac);
3366 if (port->bond_fake_iface) {
3367 struct netdev *bond_netdev;
3369 if (!netdev_open_default(port->name, &bond_netdev)) {
3371 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3373 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3375 netdev_close(bond_netdev);
3379 proc_net_compat_update_bond(port->name, &bond);
3384 port_update_vlan_compat(struct port *port)
3386 struct bridge *br = port->bridge;
3387 char *vlandev_name = NULL;
3389 if (port->vlan > 0) {
3390 /* Figure out the name that the VLAN device should actually have, if it
3391 * existed. This takes some work because the VLAN device would not
3392 * have port->name in its name; rather, it would have the trunk port's
3393 * name, and 'port' would be attached to a bridge that also had the
3394 * VLAN device one of its ports. So we need to find a trunk port that
3395 * includes port->vlan.
3397 * There might be more than one candidate. This doesn't happen on
3398 * XenServer, so if it happens we just pick the first choice in
3399 * alphabetical order instead of creating multiple VLAN devices. */
3401 for (i = 0; i < br->n_ports; i++) {
3402 struct port *p = br->ports[i];
3403 if (port_trunks_vlan(p, port->vlan)
3405 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3407 uint8_t ea[ETH_ADDR_LEN];
3408 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3409 if (!eth_addr_is_multicast(ea) &&
3410 !eth_addr_is_reserved(ea) &&
3411 !eth_addr_is_zero(ea)) {
3412 vlandev_name = p->name;
3417 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3420 /* Interface functions. */
3422 static struct iface *
3423 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3425 struct iface *iface;
3426 char *name = if_cfg->name;
3429 iface = xzalloc(sizeof *iface);
3431 iface->port_ifidx = port->n_ifaces;
3432 iface->name = xstrdup(name);
3433 iface->xf_ifidx = -1;
3434 iface->tag = tag_create_random();
3435 iface->delay_expires = LLONG_MAX;
3436 iface->netdev = NULL;
3437 iface->cfg = if_cfg;
3439 if (port->n_ifaces >= port->allocated_ifaces) {
3440 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3441 sizeof *port->ifaces);
3443 port->ifaces[port->n_ifaces++] = iface;
3444 if (port->n_ifaces > 1) {
3445 port->bridge->has_bonded_ports = true;
3448 /* Attempt to create the network interface in case it
3449 * doesn't exist yet. */
3450 if (!iface_is_internal(port->bridge, iface->name)) {
3451 error = set_up_iface(if_cfg, iface, true);
3453 VLOG_WARN("could not create iface %s: %s", iface->name,
3458 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3460 bridge_flush(port->bridge);
3466 iface_destroy(struct iface *iface)
3469 struct port *port = iface->port;
3470 struct bridge *br = port->bridge;
3471 bool del_active = port->active_iface == iface->port_ifidx;
3474 if (iface->xf_ifidx >= 0) {
3475 port_array_set(&br->ifaces, iface->xf_ifidx, NULL);
3478 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3479 del->port_ifidx = iface->port_ifidx;
3481 netdev_close(iface->netdev);
3484 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3485 bond_choose_active_iface(port);
3486 bond_send_learning_packets(port);
3492 bridge_flush(port->bridge);
3496 static struct iface *
3497 iface_lookup(const struct bridge *br, const char *name)
3501 for (i = 0; i < br->n_ports; i++) {
3502 struct port *port = br->ports[i];
3503 for (j = 0; j < port->n_ifaces; j++) {
3504 struct iface *iface = port->ifaces[j];
3505 if (!strcmp(iface->name, name)) {
3513 static struct iface *
3514 iface_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3516 return port_array_get(&br->ifaces, xf_ifidx);
3519 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3520 * 'br', that is, an interface that is entirely simulated within the datapath.
3521 * The local port (XFLOWP_LOCAL) is always an internal interface. Other local
3522 * interfaces are created by setting "iface.<iface>.internal = true".
3524 * In addition, we have a kluge-y feature that creates an internal port with
3525 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3526 * This feature needs to go away in the long term. Until then, this is one
3527 * reason why this function takes a name instead of a struct iface: the fake
3528 * interfaces created this way do not have a struct iface. */
3530 iface_is_internal(const struct bridge *br, const char *if_name)
3532 /* XXX wastes time */
3533 struct iface *iface;
3536 if (!strcmp(if_name, br->name)) {
3540 iface = iface_lookup(br, if_name);
3541 if (iface && !strcmp(iface->cfg->type, "internal")) {
3545 port = port_lookup(br, if_name);
3546 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3552 /* Set Ethernet address of 'iface', if one is specified in the configuration
3555 iface_set_mac(struct iface *iface)
3557 uint8_t ea[ETH_ADDR_LEN];
3559 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3560 if (eth_addr_is_multicast(ea)) {
3561 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3563 } else if (iface->xf_ifidx == XFLOWP_LOCAL) {
3564 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3565 iface->name, iface->name);
3567 int error = netdev_set_etheraddr(iface->netdev, ea);
3569 VLOG_ERR("interface %s: setting MAC failed (%s)",
3570 iface->name, strerror(error));
3576 /* Port mirroring. */
3579 mirror_reconfigure(struct bridge *br)
3581 struct shash old_mirrors, new_mirrors;
3582 struct shash_node *node;
3583 unsigned long *rspan_vlans;
3586 /* Collect old mirrors. */
3587 shash_init(&old_mirrors);
3588 for (i = 0; i < MAX_MIRRORS; i++) {
3589 if (br->mirrors[i]) {
3590 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3594 /* Collect new mirrors. */
3595 shash_init(&new_mirrors);
3596 for (i = 0; i < br->cfg->n_mirrors; i++) {
3597 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3598 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3599 VLOG_WARN("bridge %s: %s specified twice as mirror",
3600 br->name, cfg->name);
3604 /* Get rid of deleted mirrors and add new mirrors. */
3605 SHASH_FOR_EACH (node, &old_mirrors) {
3606 if (!shash_find(&new_mirrors, node->name)) {
3607 mirror_destroy(node->data);
3610 SHASH_FOR_EACH (node, &new_mirrors) {
3611 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3613 mirror = mirror_create(br, node->name);
3618 mirror_reconfigure_one(mirror, node->data);
3620 shash_destroy(&old_mirrors);
3621 shash_destroy(&new_mirrors);
3623 /* Update port reserved status. */
3624 for (i = 0; i < br->n_ports; i++) {
3625 br->ports[i]->is_mirror_output_port = false;
3627 for (i = 0; i < MAX_MIRRORS; i++) {
3628 struct mirror *m = br->mirrors[i];
3629 if (m && m->out_port) {
3630 m->out_port->is_mirror_output_port = true;
3634 /* Update flooded vlans (for RSPAN). */
3636 if (br->cfg->n_flood_vlans) {
3637 rspan_vlans = bitmap_allocate(4096);
3639 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3640 int64_t vlan = br->cfg->flood_vlans[i];
3641 if (vlan >= 0 && vlan < 4096) {
3642 bitmap_set1(rspan_vlans, vlan);
3643 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3646 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3651 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3656 static struct mirror *
3657 mirror_create(struct bridge *br, const char *name)
3662 for (i = 0; ; i++) {
3663 if (i >= MAX_MIRRORS) {
3664 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3665 "cannot create %s", br->name, MAX_MIRRORS, name);
3668 if (!br->mirrors[i]) {
3673 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3676 br->mirrors[i] = m = xzalloc(sizeof *m);
3679 m->name = xstrdup(name);
3680 shash_init(&m->src_ports);
3681 shash_init(&m->dst_ports);
3691 mirror_destroy(struct mirror *m)
3694 struct bridge *br = m->bridge;
3697 for (i = 0; i < br->n_ports; i++) {
3698 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3699 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3702 shash_destroy(&m->src_ports);
3703 shash_destroy(&m->dst_ports);
3706 m->bridge->mirrors[m->idx] = NULL;
3714 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3715 struct shash *names)
3719 for (i = 0; i < n_ports; i++) {
3720 const char *name = ports[i]->name;
3721 if (port_lookup(m->bridge, name)) {
3722 shash_add_once(names, name, NULL);
3724 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3725 "port %s", m->bridge->name, m->name, name);
3731 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3737 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3739 for (i = 0; i < cfg->n_select_vlan; i++) {
3740 int64_t vlan = cfg->select_vlan[i];
3741 if (vlan < 0 || vlan > 4095) {
3742 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3743 m->bridge->name, m->name, vlan);
3745 (*vlans)[n_vlans++] = vlan;
3752 vlan_is_mirrored(const struct mirror *m, int vlan)
3756 for (i = 0; i < m->n_vlans; i++) {
3757 if (m->vlans[i] == vlan) {
3765 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3769 for (i = 0; i < m->n_vlans; i++) {
3770 if (port_trunks_vlan(p, m->vlans[i])) {
3778 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3780 struct shash src_ports, dst_ports;
3781 mirror_mask_t mirror_bit;
3782 struct port *out_port;
3788 /* Get output port. */
3789 if (cfg->output_port) {
3790 out_port = port_lookup(m->bridge, cfg->output_port->name);
3792 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3793 m->bridge->name, m->name);
3799 if (cfg->output_vlan) {
3800 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3801 "output vlan; ignoring output vlan",
3802 m->bridge->name, m->name);
3804 } else if (cfg->output_vlan) {
3806 out_vlan = *cfg->output_vlan;
3808 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3809 m->bridge->name, m->name);
3814 shash_init(&src_ports);
3815 shash_init(&dst_ports);
3816 if (cfg->select_all) {
3817 for (i = 0; i < m->bridge->n_ports; i++) {
3818 const char *name = m->bridge->ports[i]->name;
3819 shash_add_once(&src_ports, name, NULL);
3820 shash_add_once(&dst_ports, name, NULL);
3825 /* Get ports, and drop duplicates and ports that don't exist. */
3826 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3828 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3831 /* Get all the vlans, and drop duplicate and invalid vlans. */
3832 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3835 /* Update mirror data. */
3836 if (!shash_equal_keys(&m->src_ports, &src_ports)
3837 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3838 || m->n_vlans != n_vlans
3839 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3840 || m->out_port != out_port
3841 || m->out_vlan != out_vlan) {
3842 bridge_flush(m->bridge);
3844 shash_swap(&m->src_ports, &src_ports);
3845 shash_swap(&m->dst_ports, &dst_ports);
3848 m->n_vlans = n_vlans;
3849 m->out_port = out_port;
3850 m->out_vlan = out_vlan;
3853 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3854 for (i = 0; i < m->bridge->n_ports; i++) {
3855 struct port *port = m->bridge->ports[i];
3857 if (shash_find(&m->src_ports, port->name)
3860 ? port_trunks_any_mirrored_vlan(m, port)
3861 : vlan_is_mirrored(m, port->vlan)))) {
3862 port->src_mirrors |= mirror_bit;
3864 port->src_mirrors &= ~mirror_bit;
3867 if (shash_find(&m->dst_ports, port->name)) {
3868 port->dst_mirrors |= mirror_bit;
3870 port->dst_mirrors &= ~mirror_bit;
3875 shash_destroy(&src_ports);
3876 shash_destroy(&dst_ports);