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>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
36 #include "dynamic-string.h"
41 #include "mac-learning.h"
43 #include "xflow-util.h"
44 #include "ofp-print.h"
46 #include "ofproto/netflow.h"
47 #include "ofproto/ofproto.h"
49 #include "poll-loop.h"
50 #include "port-array.h"
51 #include "proc-net-compat.h"
55 #include "socket-util.h"
56 #include "stream-ssl.h"
62 #include "vswitchd/vswitch-idl.h"
63 #include "xenserver.h"
66 #include "sflow_api.h"
68 #define THIS_MODULE VLM_bridge
77 /* These members are always valid. */
78 struct port *port; /* Containing port. */
79 size_t port_ifidx; /* Index within containing port. */
80 char *name; /* Host network device name. */
81 tag_type tag; /* Tag associated with this interface. */
82 long long delay_expires; /* Time after which 'enabled' may change. */
84 /* These members are valid only after bridge_reconfigure() causes them to
86 int xf_ifidx; /* Index within kernel datapath. */
87 struct netdev *netdev; /* Network device. */
88 bool enabled; /* May be chosen for flows? */
90 /* This member is only valid *during* bridge_reconfigure(). */
91 const struct ovsrec_interface *cfg;
94 #define BOND_MASK 0xff
96 int iface_idx; /* Index of assigned iface, or -1 if none. */
97 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
98 tag_type iface_tag; /* Tag associated with iface_idx. */
101 #define MAX_MIRRORS 32
102 typedef uint32_t mirror_mask_t;
103 #define MIRROR_MASK_C(X) UINT32_C(X)
104 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
106 struct bridge *bridge;
110 /* Selection criteria. */
111 struct shash src_ports; /* Name is port name; data is always NULL. */
112 struct shash dst_ports; /* Name is port name; data is always NULL. */
117 struct port *out_port;
121 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
123 struct bridge *bridge;
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
130 /* An ordinary bridge port has 1 interface.
131 * A bridge port for bonding has at least 2 interfaces. */
132 struct iface **ifaces;
133 size_t n_ifaces, allocated_ifaces;
136 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
137 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
138 tag_type active_iface_tag; /* Tag for bcast flows. */
139 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
140 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
141 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
142 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
143 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
144 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
145 long long int bond_next_rebalance; /* Next rebalancing time. */
147 /* Port mirroring info. */
148 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
149 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
150 bool is_mirror_output_port; /* Does port mirroring send frames here? */
152 /* This member is only valid *during* bridge_reconfigure(). */
153 const struct ovsrec_port *cfg;
156 #define DP_MAX_PORTS 255
158 struct list node; /* Node in global list of bridges. */
159 char *name; /* User-specified arbitrary name. */
160 struct mac_learning *ml; /* MAC learning table. */
161 bool sent_config_request; /* Successfully sent config request? */
162 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
164 /* OpenFlow switch processing. */
165 struct ofproto *ofproto; /* OpenFlow switch. */
167 /* Description strings. */
168 char *mfr_desc; /* Manufacturer. */
169 char *hw_desc; /* Hardware. */
170 char *sw_desc; /* Software version. */
171 char *serial_desc; /* Serial number. */
172 char *dp_desc; /* Datapath description. */
174 /* Kernel datapath information. */
175 struct xfif *xfif; /* Datapath. */
176 struct port_array ifaces; /* Indexed by kernel datapath port number. */
180 size_t n_ports, allocated_ports;
181 struct shash iface_by_name; /* "struct iface"s indexed by name. */
182 struct shash port_by_name; /* "struct port"s indexed by name. */
185 bool has_bonded_ports;
190 /* Flow statistics gathering. */
191 time_t next_stats_request;
193 /* Port mirroring. */
194 struct mirror *mirrors[MAX_MIRRORS];
196 /* This member is only valid *during* bridge_reconfigure(). */
197 const struct ovsrec_bridge *cfg;
200 /* List of all bridges. */
201 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
203 /* Maximum number of datapaths. */
204 enum { DP_MAX = 256 };
206 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
207 static void bridge_destroy(struct bridge *);
208 static struct bridge *bridge_lookup(const char *name);
209 static unixctl_cb_func bridge_unixctl_dump_flows;
210 static int bridge_run_one(struct bridge *);
211 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
212 const struct bridge *br,
213 struct ovsrec_controller ***controllersp);
214 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
216 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
218 const struct sockaddr_in *managers,
220 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
221 static void bridge_fetch_dp_ifaces(struct bridge *);
222 static void bridge_flush(struct bridge *);
223 static void bridge_pick_local_hw_addr(struct bridge *,
224 uint8_t ea[ETH_ADDR_LEN],
225 struct iface **hw_addr_iface);
226 static uint64_t bridge_pick_datapath_id(struct bridge *,
227 const uint8_t bridge_ea[ETH_ADDR_LEN],
228 struct iface *hw_addr_iface);
229 static struct iface *bridge_get_local_iface(struct bridge *);
230 static uint64_t dpid_from_hash(const void *, size_t nbytes);
232 static unixctl_cb_func bridge_unixctl_fdb_show;
234 static void bond_init(void);
235 static void bond_run(struct bridge *);
236 static void bond_wait(struct bridge *);
237 static void bond_rebalance_port(struct port *);
238 static void bond_send_learning_packets(struct port *);
239 static void bond_enable_slave(struct iface *iface, bool enable);
241 static struct port *port_create(struct bridge *, const char *name);
242 static void port_reconfigure(struct port *, const struct ovsrec_port *);
243 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
244 static void port_destroy(struct port *);
245 static struct port *port_lookup(const struct bridge *, const char *name);
246 static struct iface *port_lookup_iface(const struct port *, const char *name);
247 static struct port *port_from_xf_ifidx(const struct bridge *,
249 static void port_update_bond_compat(struct port *);
250 static void port_update_vlan_compat(struct port *);
251 static void port_update_bonding(struct port *);
253 static struct mirror *mirror_create(struct bridge *, const char *name);
254 static void mirror_destroy(struct mirror *);
255 static void mirror_reconfigure(struct bridge *);
256 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
257 static bool vlan_is_mirrored(const struct mirror *, int vlan);
259 static struct iface *iface_create(struct port *port,
260 const struct ovsrec_interface *if_cfg);
261 static void iface_destroy(struct iface *);
262 static struct iface *iface_lookup(const struct bridge *, const char *name);
263 static struct iface *iface_from_xf_ifidx(const struct bridge *,
265 static bool iface_is_internal(const struct bridge *, const char *name);
266 static void iface_set_mac(struct iface *);
268 /* Hooks into ofproto processing. */
269 static struct ofhooks bridge_ofhooks;
271 /* Public functions. */
273 /* Adds the name of each interface used by a bridge, including local and
274 * internal ports, to 'svec'. */
276 bridge_get_ifaces(struct svec *svec)
278 struct bridge *br, *next;
281 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
282 for (i = 0; i < br->n_ports; i++) {
283 struct port *port = br->ports[i];
285 for (j = 0; j < port->n_ifaces; j++) {
286 struct iface *iface = port->ifaces[j];
287 if (iface->xf_ifidx < 0) {
288 VLOG_ERR("%s interface not in datapath %s, ignoring",
289 iface->name, xfif_name(br->xfif));
291 if (iface->xf_ifidx != XFLOWP_LOCAL) {
292 svec_add(svec, iface->name);
301 bridge_init(const struct ovsrec_open_vswitch *cfg)
303 struct svec bridge_names;
304 struct svec xfif_names, xfif_types;
307 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
309 svec_init(&bridge_names);
310 for (i = 0; i < cfg->n_bridges; i++) {
311 svec_add(&bridge_names, cfg->bridges[i]->name);
313 svec_sort(&bridge_names);
315 svec_init(&xfif_names);
316 svec_init(&xfif_types);
317 xf_enumerate_types(&xfif_types);
318 for (i = 0; i < xfif_types.n; i++) {
323 xf_enumerate_names(xfif_types.names[i], &xfif_names);
325 for (j = 0; j < xfif_names.n; j++) {
326 retval = xfif_open(xfif_names.names[j], xfif_types.names[i], &xfif);
328 struct svec all_names;
331 svec_init(&all_names);
332 xfif_get_all_names(xfif, &all_names);
333 for (k = 0; k < all_names.n; k++) {
334 if (svec_contains(&bridge_names, all_names.names[k])) {
340 svec_destroy(&all_names);
345 svec_destroy(&bridge_names);
346 svec_destroy(&xfif_names);
347 svec_destroy(&xfif_types);
349 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
353 bridge_reconfigure(cfg);
358 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
360 /* XXX SSL should be configurable on a per-bridge basis. */
362 stream_ssl_set_private_key_file(ssl->private_key);
363 stream_ssl_set_certificate_file(ssl->certificate);
364 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
369 /* Attempt to create the network device 'iface_name' through the netdev
372 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
375 struct shash_node *node;
376 struct shash options;
380 shash_init(&options);
381 for (i = 0; i < iface_cfg->n_options; i++) {
382 shash_add(&options, iface_cfg->key_options[i],
383 xstrdup(iface_cfg->value_options[i]));
387 struct netdev_options netdev_options;
389 memset(&netdev_options, 0, sizeof netdev_options);
390 netdev_options.name = iface_cfg->name;
391 if (!strcmp(iface_cfg->type, "internal")) {
392 /* An "internal" config type maps to a netdev "system" type. */
393 netdev_options.type = "system";
395 netdev_options.type = iface_cfg->type;
397 netdev_options.args = &options;
398 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
399 netdev_options.may_create = true;
400 netdev_options.may_open = true;
401 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
402 netdev_options.may_open = true;
405 error = netdev_open(&netdev_options, &iface->netdev);
408 netdev_get_carrier(iface->netdev, &iface->enabled);
410 } else if (iface->netdev) {
411 const char *netdev_type = netdev_get_type(iface->netdev);
412 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
413 ? iface_cfg->type : NULL;
415 /* An "internal" config type maps to a netdev "system" type. */
416 if (iface_type && !strcmp(iface_type, "internal")) {
417 iface_type = "system";
420 if (!iface_type || !strcmp(netdev_type, iface_type)) {
421 error = netdev_reconfigure(iface->netdev, &options);
423 VLOG_WARN("%s: attempting change device type from %s to %s",
424 iface_cfg->name, netdev_type, iface_type);
429 SHASH_FOR_EACH (node, &options) {
432 shash_destroy(&options);
438 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
440 return set_up_iface(iface_cfg, iface, false);
444 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
445 void *aux OVS_UNUSED)
447 if (!iface->netdev) {
448 int error = set_up_iface(iface->cfg, iface, true);
450 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
460 check_iface_xf_ifidx(struct bridge *br, struct iface *iface,
461 void *aux OVS_UNUSED)
463 if (iface->xf_ifidx >= 0) {
464 VLOG_DBG("%s has interface %s on port %d",
466 iface->name, iface->xf_ifidx);
469 VLOG_ERR("%s interface not in %s, dropping",
470 iface->name, xfif_name(br->xfif));
476 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
477 void *aux OVS_UNUSED)
479 /* Set policing attributes. */
480 netdev_set_policing(iface->netdev,
481 iface->cfg->ingress_policing_rate,
482 iface->cfg->ingress_policing_burst);
484 /* Set MAC address of internal interfaces other than the local
486 if (iface->xf_ifidx != XFLOWP_LOCAL
487 && iface_is_internal(br, iface->name)) {
488 iface_set_mac(iface);
494 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
495 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
496 * deletes from 'br' any ports that no longer have any interfaces. */
498 iterate_and_prune_ifaces(struct bridge *br,
499 bool (*cb)(struct bridge *, struct iface *,
505 for (i = 0; i < br->n_ports; ) {
506 struct port *port = br->ports[i];
507 for (j = 0; j < port->n_ifaces; ) {
508 struct iface *iface = port->ifaces[j];
509 if (cb(br, iface, aux)) {
512 iface_destroy(iface);
516 if (port->n_ifaces) {
519 VLOG_ERR("%s port has no interfaces, dropping", port->name);
525 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
526 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
527 * responsible for freeing '*managersp' (with free()).
529 * You may be asking yourself "why does ovs-vswitchd care?", because
530 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
531 * should not be and in fact is not directly involved in that. But
532 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
533 * it has to tell in-band control where the managers are to enable that.
536 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
537 struct sockaddr_in **managersp, size_t *n_managersp)
539 struct sockaddr_in *managers = NULL;
540 size_t n_managers = 0;
542 if (ovs_cfg->n_managers > 0) {
545 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
546 for (i = 0; i < ovs_cfg->n_managers; i++) {
547 const char *name = ovs_cfg->managers[i];
548 struct sockaddr_in *sin = &managers[i];
550 if ((!strncmp(name, "tcp:", 4)
551 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
552 (!strncmp(name, "ssl:", 4)
553 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
559 *managersp = managers;
560 *n_managersp = n_managers;
564 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
566 struct ovsdb_idl_txn *txn;
567 struct shash old_br, new_br;
568 struct shash_node *node;
569 struct bridge *br, *next;
570 struct sockaddr_in *managers;
573 int sflow_bridge_number;
575 COVERAGE_INC(bridge_reconfigure);
577 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
579 collect_managers(ovs_cfg, &managers, &n_managers);
581 /* Collect old and new bridges. */
584 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
585 shash_add(&old_br, br->name, br);
587 for (i = 0; i < ovs_cfg->n_bridges; i++) {
588 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
589 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
590 VLOG_WARN("more than one bridge named %s", br_cfg->name);
594 /* Get rid of deleted bridges and add new bridges. */
595 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
596 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
603 SHASH_FOR_EACH (node, &new_br) {
604 const char *br_name = node->name;
605 const struct ovsrec_bridge *br_cfg = node->data;
606 br = shash_find_data(&old_br, br_name);
608 /* If the bridge datapath type has changed, we need to tear it
609 * down and recreate. */
610 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
612 bridge_create(br_cfg);
615 bridge_create(br_cfg);
618 shash_destroy(&old_br);
619 shash_destroy(&new_br);
623 bridge_configure_ssl(ovs_cfg->ssl);
626 /* Reconfigure all bridges. */
627 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
628 bridge_reconfigure_one(ovs_cfg, br);
631 /* Add and delete ports on all datapaths.
633 * The kernel will reject any attempt to add a given port to a datapath if
634 * that port already belongs to a different datapath, so we must do all
635 * port deletions before any port additions. */
636 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
637 struct xflow_port *xfif_ports;
639 struct shash want_ifaces;
641 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
642 bridge_get_all_ifaces(br, &want_ifaces);
643 for (i = 0; i < n_xfif_ports; i++) {
644 const struct xflow_port *p = &xfif_ports[i];
645 if (!shash_find(&want_ifaces, p->devname)
646 && strcmp(p->devname, br->name)) {
647 int retval = xfif_port_del(br->xfif, p->port);
649 VLOG_ERR("failed to remove %s interface from %s: %s",
650 p->devname, xfif_name(br->xfif),
655 shash_destroy(&want_ifaces);
658 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
659 struct xflow_port *xfif_ports;
661 struct shash cur_ifaces, want_ifaces;
662 struct shash_node *node;
664 /* Get the set of interfaces currently in this datapath. */
665 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
666 shash_init(&cur_ifaces);
667 for (i = 0; i < n_xfif_ports; i++) {
668 const char *name = xfif_ports[i].devname;
669 if (!shash_find(&cur_ifaces, name)) {
670 shash_add(&cur_ifaces, name, NULL);
675 /* Get the set of interfaces we want on this datapath. */
676 bridge_get_all_ifaces(br, &want_ifaces);
678 SHASH_FOR_EACH (node, &want_ifaces) {
679 const char *if_name = node->name;
680 struct iface *iface = node->data;
682 if (shash_find(&cur_ifaces, if_name)) {
683 /* Already exists, just reconfigure it. */
685 reconfigure_iface(iface->cfg, iface);
688 /* Need to add to datapath. */
692 /* Add to datapath. */
693 internal = iface_is_internal(br, if_name);
694 error = xfif_port_add(br->xfif, if_name,
695 internal ? XFLOW_PORT_INTERNAL : 0, NULL);
696 if (error == EFBIG) {
697 VLOG_ERR("ran out of valid port numbers on %s",
698 xfif_name(br->xfif));
701 VLOG_ERR("failed to add %s interface to %s: %s",
702 if_name, xfif_name(br->xfif), strerror(error));
706 shash_destroy(&cur_ifaces);
707 shash_destroy(&want_ifaces);
709 sflow_bridge_number = 0;
710 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
713 struct iface *local_iface;
714 struct iface *hw_addr_iface;
717 bridge_fetch_dp_ifaces(br);
719 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
720 iterate_and_prune_ifaces(br, check_iface_xf_ifidx, NULL);
722 /* Pick local port hardware address, datapath ID. */
723 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
724 local_iface = bridge_get_local_iface(br);
726 int error = netdev_set_etheraddr(local_iface->netdev, ea);
728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
729 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
730 "Ethernet address: %s",
731 br->name, strerror(error));
735 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
736 ofproto_set_datapath_id(br->ofproto, dpid);
738 dpid_string = xasprintf("%012"PRIx64, dpid);
739 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
742 /* Set NetFlow configuration on this bridge. */
743 if (br->cfg->netflow) {
744 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
745 struct netflow_options opts;
747 memset(&opts, 0, sizeof opts);
749 xfif_get_netflow_ids(br->xfif, &opts.engine_type, &opts.engine_id);
750 if (nf_cfg->engine_type) {
751 opts.engine_type = *nf_cfg->engine_type;
753 if (nf_cfg->engine_id) {
754 opts.engine_id = *nf_cfg->engine_id;
757 opts.active_timeout = nf_cfg->active_timeout;
758 if (!opts.active_timeout) {
759 opts.active_timeout = -1;
760 } else if (opts.active_timeout < 0) {
761 VLOG_WARN("bridge %s: active timeout interval set to negative "
762 "value, using default instead (%d seconds)", br->name,
763 NF_ACTIVE_TIMEOUT_DEFAULT);
764 opts.active_timeout = -1;
767 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
768 if (opts.add_id_to_iface) {
769 if (opts.engine_id > 0x7f) {
770 VLOG_WARN("bridge %s: netflow port mangling may conflict "
771 "with another vswitch, choose an engine id less "
772 "than 128", br->name);
774 if (br->n_ports > 508) {
775 VLOG_WARN("bridge %s: netflow port mangling will conflict "
776 "with another port when more than 508 ports are "
781 opts.collectors.n = nf_cfg->n_targets;
782 opts.collectors.names = nf_cfg->targets;
783 if (ofproto_set_netflow(br->ofproto, &opts)) {
784 VLOG_ERR("bridge %s: problem setting netflow collectors",
788 ofproto_set_netflow(br->ofproto, NULL);
791 /* Set sFlow configuration on this bridge. */
792 if (br->cfg->sflow) {
793 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
794 struct ovsrec_controller **controllers;
795 struct ofproto_sflow_options oso;
796 size_t n_controllers;
799 memset(&oso, 0, sizeof oso);
801 oso.targets.n = sflow_cfg->n_targets;
802 oso.targets.names = sflow_cfg->targets;
804 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
805 if (sflow_cfg->sampling) {
806 oso.sampling_rate = *sflow_cfg->sampling;
809 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
810 if (sflow_cfg->polling) {
811 oso.polling_interval = *sflow_cfg->polling;
814 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
815 if (sflow_cfg->header) {
816 oso.header_len = *sflow_cfg->header;
819 oso.sub_id = sflow_bridge_number++;
820 oso.agent_device = sflow_cfg->agent;
822 oso.control_ip = NULL;
823 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
824 for (i = 0; i < n_controllers; i++) {
825 if (controllers[i]->local_ip) {
826 oso.control_ip = controllers[i]->local_ip;
830 ofproto_set_sflow(br->ofproto, &oso);
832 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
834 ofproto_set_sflow(br->ofproto, NULL);
837 /* Update the controller and related settings. It would be more
838 * straightforward to call this from bridge_reconfigure_one(), but we
839 * can't do it there for two reasons. First, and most importantly, at
840 * that point we don't know the xf_ifidx of any interfaces that have
841 * been added to the bridge (because we haven't actually added them to
842 * the datapath). Second, at that point we haven't set the datapath ID
843 * yet; when a controller is configured, resetting the datapath ID will
844 * immediately disconnect from the controller, so it's better to set
845 * the datapath ID before the controller. */
846 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
848 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
849 for (i = 0; i < br->n_ports; i++) {
850 struct port *port = br->ports[i];
852 port_update_vlan_compat(port);
853 port_update_bonding(port);
856 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
857 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
860 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
862 ovsdb_idl_txn_commit(txn);
863 ovsdb_idl_txn_destroy(txn); /* XXX */
869 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
873 for (i = 0; i < n; i++) {
874 if (!strcmp(keys[i], key)) {
882 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
884 return get_ovsrec_key_value(key,
885 br_cfg->key_other_config,
886 br_cfg->value_other_config,
887 br_cfg->n_other_config);
891 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
892 struct iface **hw_addr_iface)
898 *hw_addr_iface = NULL;
900 /* Did the user request a particular MAC? */
901 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
902 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
903 if (eth_addr_is_multicast(ea)) {
904 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
905 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
906 } else if (eth_addr_is_zero(ea)) {
907 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
913 /* Otherwise choose the minimum non-local MAC address among all of the
915 memset(ea, 0xff, sizeof ea);
916 for (i = 0; i < br->n_ports; i++) {
917 struct port *port = br->ports[i];
918 uint8_t iface_ea[ETH_ADDR_LEN];
921 /* Mirror output ports don't participate. */
922 if (port->is_mirror_output_port) {
926 /* Choose the MAC address to represent the port. */
927 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
928 /* Find the interface with this Ethernet address (if any) so that
929 * we can provide the correct devname to the caller. */
931 for (j = 0; j < port->n_ifaces; j++) {
932 struct iface *candidate = port->ifaces[j];
933 uint8_t candidate_ea[ETH_ADDR_LEN];
934 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
935 && eth_addr_equals(iface_ea, candidate_ea)) {
940 /* Choose the interface whose MAC address will represent the port.
941 * The Linux kernel bonding code always chooses the MAC address of
942 * the first slave added to a bond, and the Fedora networking
943 * scripts always add slaves to a bond in alphabetical order, so
944 * for compatibility we choose the interface with the name that is
945 * first in alphabetical order. */
946 iface = port->ifaces[0];
947 for (j = 1; j < port->n_ifaces; j++) {
948 struct iface *candidate = port->ifaces[j];
949 if (strcmp(candidate->name, iface->name) < 0) {
954 /* The local port doesn't count (since we're trying to choose its
955 * MAC address anyway). */
956 if (iface->xf_ifidx == XFLOWP_LOCAL) {
961 error = netdev_get_etheraddr(iface->netdev, iface_ea);
963 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
964 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
965 iface->name, strerror(error));
970 /* Compare against our current choice. */
971 if (!eth_addr_is_multicast(iface_ea) &&
972 !eth_addr_is_local(iface_ea) &&
973 !eth_addr_is_reserved(iface_ea) &&
974 !eth_addr_is_zero(iface_ea) &&
975 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
977 memcpy(ea, iface_ea, ETH_ADDR_LEN);
978 *hw_addr_iface = iface;
981 if (eth_addr_is_multicast(ea)) {
982 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
983 *hw_addr_iface = NULL;
984 VLOG_WARN("bridge %s: using default bridge Ethernet "
985 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
987 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
988 br->name, ETH_ADDR_ARGS(ea));
992 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
993 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
994 * an interface on 'br', then that interface must be passed in as
995 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
996 * 'hw_addr_iface' must be passed in as a null pointer. */
998 bridge_pick_datapath_id(struct bridge *br,
999 const uint8_t bridge_ea[ETH_ADDR_LEN],
1000 struct iface *hw_addr_iface)
1003 * The procedure for choosing a bridge MAC address will, in the most
1004 * ordinary case, also choose a unique MAC that we can use as a datapath
1005 * ID. In some special cases, though, multiple bridges will end up with
1006 * the same MAC address. This is OK for the bridges, but it will confuse
1007 * the OpenFlow controller, because each datapath needs a unique datapath
1010 * Datapath IDs must be unique. It is also very desirable that they be
1011 * stable from one run to the next, so that policy set on a datapath
1014 const char *datapath_id;
1017 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1018 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1022 if (hw_addr_iface) {
1024 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1026 * A bridge whose MAC address is taken from a VLAN network device
1027 * (that is, a network device created with vconfig(8) or similar
1028 * tool) will have the same MAC address as a bridge on the VLAN
1029 * device's physical network device.
1031 * Handle this case by hashing the physical network device MAC
1032 * along with the VLAN identifier.
1034 uint8_t buf[ETH_ADDR_LEN + 2];
1035 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1036 buf[ETH_ADDR_LEN] = vlan >> 8;
1037 buf[ETH_ADDR_LEN + 1] = vlan;
1038 return dpid_from_hash(buf, sizeof buf);
1041 * Assume that this bridge's MAC address is unique, since it
1042 * doesn't fit any of the cases we handle specially.
1047 * A purely internal bridge, that is, one that has no non-virtual
1048 * network devices on it at all, is more difficult because it has no
1049 * natural unique identifier at all.
1051 * When the host is a XenServer, we handle this case by hashing the
1052 * host's UUID with the name of the bridge. Names of bridges are
1053 * persistent across XenServer reboots, although they can be reused if
1054 * an internal network is destroyed and then a new one is later
1055 * created, so this is fairly effective.
1057 * When the host is not a XenServer, we punt by using a random MAC
1058 * address on each run.
1060 const char *host_uuid = xenserver_get_host_uuid();
1062 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1063 dpid = dpid_from_hash(combined, strlen(combined));
1069 return eth_addr_to_uint64(bridge_ea);
1073 dpid_from_hash(const void *data, size_t n)
1075 uint8_t hash[SHA1_DIGEST_SIZE];
1077 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1078 sha1_bytes(data, n, hash);
1079 eth_addr_mark_random(hash);
1080 return eth_addr_to_uint64(hash);
1086 struct bridge *br, *next;
1090 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1091 int error = bridge_run_one(br);
1093 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1094 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1095 "forcing reconfiguration", br->name);
1109 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1110 ofproto_wait(br->ofproto);
1111 if (ofproto_has_controller(br->ofproto)) {
1115 mac_learning_wait(br->ml);
1120 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1121 * configuration changes. */
1123 bridge_flush(struct bridge *br)
1125 COVERAGE_INC(bridge_flush);
1127 mac_learning_flush(br->ml);
1130 /* Returns the 'br' interface for the XFLOWP_LOCAL port, or null if 'br' has no
1131 * such interface. */
1132 static struct iface *
1133 bridge_get_local_iface(struct bridge *br)
1137 for (i = 0; i < br->n_ports; i++) {
1138 struct port *port = br->ports[i];
1139 for (j = 0; j < port->n_ifaces; j++) {
1140 struct iface *iface = port->ifaces[j];
1141 if (iface->xf_ifidx == XFLOWP_LOCAL) {
1150 /* Bridge unixctl user interface functions. */
1152 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1153 const char *args, void *aux OVS_UNUSED)
1155 struct ds ds = DS_EMPTY_INITIALIZER;
1156 const struct bridge *br;
1157 const struct mac_entry *e;
1159 br = bridge_lookup(args);
1161 unixctl_command_reply(conn, 501, "no such bridge");
1165 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1166 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1167 if (e->port < 0 || e->port >= br->n_ports) {
1170 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1171 br->ports[e->port]->ifaces[0]->xf_ifidx,
1172 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1174 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1178 /* Bridge reconfiguration functions. */
1179 static struct bridge *
1180 bridge_create(const struct ovsrec_bridge *br_cfg)
1185 assert(!bridge_lookup(br_cfg->name));
1186 br = xzalloc(sizeof *br);
1188 error = xfif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1194 xfif_flow_flush(br->xfif);
1196 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1199 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1201 xfif_delete(br->xfif);
1202 xfif_close(br->xfif);
1207 br->name = xstrdup(br_cfg->name);
1209 br->ml = mac_learning_create();
1210 br->sent_config_request = false;
1211 eth_addr_nicira_random(br->default_ea);
1213 port_array_init(&br->ifaces);
1215 shash_init(&br->port_by_name);
1216 shash_init(&br->iface_by_name);
1220 list_push_back(&all_bridges, &br->node);
1222 VLOG_INFO("created bridge %s on %s", br->name, xfif_name(br->xfif));
1228 bridge_destroy(struct bridge *br)
1233 while (br->n_ports > 0) {
1234 port_destroy(br->ports[br->n_ports - 1]);
1236 list_remove(&br->node);
1237 error = xfif_delete(br->xfif);
1238 if (error && error != ENOENT) {
1239 VLOG_ERR("failed to delete %s: %s",
1240 xfif_name(br->xfif), strerror(error));
1242 xfif_close(br->xfif);
1243 ofproto_destroy(br->ofproto);
1244 mac_learning_destroy(br->ml);
1245 port_array_destroy(&br->ifaces);
1246 shash_destroy(&br->port_by_name);
1247 shash_destroy(&br->iface_by_name);
1254 static struct bridge *
1255 bridge_lookup(const char *name)
1259 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1260 if (!strcmp(br->name, name)) {
1268 bridge_exists(const char *name)
1270 return bridge_lookup(name) ? true : false;
1274 bridge_get_datapathid(const char *name)
1276 struct bridge *br = bridge_lookup(name);
1277 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1280 /* Handle requests for a listing of all flows known by the OpenFlow
1281 * stack, including those normally hidden. */
1283 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1284 const char *args, void *aux OVS_UNUSED)
1289 br = bridge_lookup(args);
1291 unixctl_command_reply(conn, 501, "Unknown bridge");
1296 ofproto_get_all_flows(br->ofproto, &results);
1298 unixctl_command_reply(conn, 200, ds_cstr(&results));
1299 ds_destroy(&results);
1303 bridge_run_one(struct bridge *br)
1307 error = ofproto_run1(br->ofproto);
1312 //XXX mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1315 error = ofproto_run2(br->ofproto, br->flush);
1322 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1323 const struct bridge *br,
1324 struct ovsrec_controller ***controllersp)
1326 struct ovsrec_controller **controllers;
1327 size_t n_controllers;
1329 if (br->cfg->n_controller) {
1330 controllers = br->cfg->controller;
1331 n_controllers = br->cfg->n_controller;
1333 controllers = ovs_cfg->controller;
1334 n_controllers = ovs_cfg->n_controller;
1337 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1343 *controllersp = controllers;
1345 return n_controllers;
1349 bridge_update_desc(struct bridge *br OVS_UNUSED)
1352 bool changed = false;
1355 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1356 if (desc != br->mfr_desc) {
1359 br->mfr_desc = xstrdup(desc);
1361 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1366 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1367 if (desc != br->hw_desc) {
1370 br->hw_desc = xstrdup(desc);
1372 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1377 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1378 if (desc != br->sw_desc) {
1381 br->sw_desc = xstrdup(desc);
1383 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1388 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1389 if (desc != br->serial_desc) {
1390 free(br->serial_desc);
1392 br->serial_desc = xstrdup(desc);
1394 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1399 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1400 if (desc != br->dp_desc) {
1403 br->dp_desc = xstrdup(desc);
1405 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1411 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1412 br->sw_desc, br->serial_desc, br->dp_desc);
1418 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1421 struct shash old_ports, new_ports;
1422 struct svec listeners, old_listeners;
1423 struct svec snoops, old_snoops;
1424 struct shash_node *node;
1427 /* Collect old ports. */
1428 shash_init(&old_ports);
1429 for (i = 0; i < br->n_ports; i++) {
1430 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1433 /* Collect new ports. */
1434 shash_init(&new_ports);
1435 for (i = 0; i < br->cfg->n_ports; i++) {
1436 const char *name = br->cfg->ports[i]->name;
1437 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1438 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1443 /* If we have a controller, then we need a local port. Complain if the
1444 * user didn't specify one.
1446 * XXX perhaps we should synthesize a port ourselves in this case. */
1447 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1448 char local_name[IF_NAMESIZE];
1451 error = xfif_port_get_name(br->xfif, XFLOWP_LOCAL,
1452 local_name, sizeof local_name);
1453 if (!error && !shash_find(&new_ports, local_name)) {
1454 VLOG_WARN("bridge %s: controller specified but no local port "
1455 "(port named %s) defined",
1456 br->name, local_name);
1460 /* Get rid of deleted ports.
1461 * Get rid of deleted interfaces on ports that still exist. */
1462 SHASH_FOR_EACH (node, &old_ports) {
1463 struct port *port = node->data;
1464 const struct ovsrec_port *port_cfg;
1466 port_cfg = shash_find_data(&new_ports, node->name);
1470 port_del_ifaces(port, port_cfg);
1474 /* Create new ports.
1475 * Add new interfaces to existing ports.
1476 * Reconfigure existing ports. */
1477 SHASH_FOR_EACH (node, &new_ports) {
1478 struct port *port = shash_find_data(&old_ports, node->name);
1480 port = port_create(br, node->name);
1483 port_reconfigure(port, node->data);
1484 if (!port->n_ifaces) {
1485 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1486 br->name, port->name);
1490 shash_destroy(&old_ports);
1491 shash_destroy(&new_ports);
1493 /* Delete all flows if we're switching from connected to standalone or vice
1494 * versa. (XXX Should we delete all flows if we are switching from one
1495 * controller to another?) */
1498 /* Configure OpenFlow management listeners. */
1499 svec_init(&listeners);
1500 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1502 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1503 ovs_rundir, br->name));
1504 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1505 svec_clear(&listeners);
1507 svec_sort_unique(&listeners);
1509 svec_init(&old_listeners);
1510 ofproto_get_listeners(br->ofproto, &old_listeners);
1511 svec_sort_unique(&old_listeners);
1513 if (!svec_equal(&listeners, &old_listeners)) {
1514 ofproto_set_listeners(br->ofproto, &listeners);
1516 svec_destroy(&listeners);
1517 svec_destroy(&old_listeners);
1519 /* Configure OpenFlow controller connection snooping. */
1521 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1523 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1524 ovs_rundir, br->name));
1525 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1526 svec_clear(&snoops);
1528 svec_sort_unique(&snoops);
1530 svec_init(&old_snoops);
1531 ofproto_get_snoops(br->ofproto, &old_snoops);
1532 svec_sort_unique(&old_snoops);
1534 if (!svec_equal(&snoops, &old_snoops)) {
1535 ofproto_set_snoops(br->ofproto, &snoops);
1537 svec_destroy(&snoops);
1538 svec_destroy(&old_snoops);
1540 /* Default listener. */
1541 svec_init(&listeners);
1542 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1543 ovs_rundir, br->name));
1544 svec_init(&old_listeners);
1545 ofproto_get_listeners(br->ofproto, &old_listeners);
1546 if (!svec_equal(&listeners, &old_listeners)) {
1547 ofproto_set_listeners(br->ofproto, &listeners);
1549 svec_destroy(&listeners);
1550 svec_destroy(&old_listeners);
1552 /* Default snoop. */
1554 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1555 ovs_rundir, br->name));
1556 svec_init(&old_snoops);
1557 ofproto_get_snoops(br->ofproto, &old_snoops);
1558 if (!svec_equal(&snoops, &old_snoops)) {
1559 ofproto_set_snoops(br->ofproto, &snoops);
1561 svec_destroy(&snoops);
1562 svec_destroy(&old_snoops);
1565 mirror_reconfigure(br);
1567 bridge_update_desc(br);
1571 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1573 const struct sockaddr_in *managers,
1576 struct ovsrec_controller **controllers;
1577 size_t n_controllers;
1579 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1581 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1582 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1583 ofproto_flush_flows(br->ofproto);
1586 if (!n_controllers) {
1587 union ofp_action action;
1590 /* Clear out controllers. */
1591 ofproto_set_controllers(br->ofproto, NULL, 0);
1593 /* Set up a flow that matches every packet and directs them to
1594 * OFPP_NORMAL (which goes to us). */
1595 memset(&action, 0, sizeof action);
1596 action.type = htons(OFPAT_OUTPUT);
1597 action.output.len = htons(sizeof action);
1598 action.output.port = htons(OFPP_NORMAL);
1599 memset(&flow, 0, sizeof flow);
1600 flow.wildcards = OVSFW_ALL;
1601 ofproto_add_flow(br->ofproto, &flow, &action, 1, 0);
1603 struct ofproto_controller *ocs;
1606 ocs = xmalloc(n_controllers * sizeof *ocs);
1607 for (i = 0; i < n_controllers; i++) {
1608 struct ovsrec_controller *c = controllers[i];
1609 struct ofproto_controller *oc = &ocs[i];
1611 if (strcmp(c->target, "discover")) {
1612 struct iface *local_iface;
1615 local_iface = bridge_get_local_iface(br);
1616 if (local_iface && c->local_ip
1617 && inet_aton(c->local_ip, &ip)) {
1618 struct netdev *netdev = local_iface->netdev;
1619 struct in_addr mask, gateway;
1621 if (!c->local_netmask
1622 || !inet_aton(c->local_netmask, &mask)) {
1625 if (!c->local_gateway
1626 || !inet_aton(c->local_gateway, &gateway)) {
1630 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1632 mask.s_addr = guess_netmask(ip.s_addr);
1634 if (!netdev_set_in4(netdev, ip, mask)) {
1635 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1637 br->name, IP_ARGS(&ip.s_addr),
1638 IP_ARGS(&mask.s_addr));
1641 if (gateway.s_addr) {
1642 if (!netdev_add_router(netdev, gateway)) {
1643 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1644 br->name, IP_ARGS(&gateway.s_addr));
1650 oc->target = c->target;
1651 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1652 oc->probe_interval = (c->inactivity_probe
1653 ? *c->inactivity_probe / 1000 : 5);
1654 oc->fail = (!c->fail_mode
1655 || !strcmp(c->fail_mode, "standalone")
1656 || !strcmp(c->fail_mode, "open")
1657 ? OFPROTO_FAIL_STANDALONE
1658 : OFPROTO_FAIL_SECURE);
1659 oc->band = (!c->connection_mode
1660 || !strcmp(c->connection_mode, "in-band")
1662 : OFPROTO_OUT_OF_BAND);
1663 oc->accept_re = c->discover_accept_regex;
1664 oc->update_resolv_conf = c->discover_update_resolv_conf;
1665 oc->rate_limit = (c->controller_rate_limit
1666 ? *c->controller_rate_limit : 0);
1667 oc->burst_limit = (c->controller_burst_limit
1668 ? *c->controller_burst_limit : 0);
1670 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1676 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1681 for (i = 0; i < br->n_ports; i++) {
1682 struct port *port = br->ports[i];
1683 for (j = 0; j < port->n_ifaces; j++) {
1684 struct iface *iface = port->ifaces[j];
1685 shash_add_once(ifaces, iface->name, iface);
1687 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1688 shash_add_once(ifaces, port->name, NULL);
1693 /* For robustness, in case the administrator moves around datapath ports behind
1694 * our back, we re-check all the datapath port numbers here.
1696 * This function will set the 'xf_ifidx' members of interfaces that have
1697 * disappeared to -1, so only call this function from a context where those
1698 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1699 * 'xf_ifidx'es will cause trouble later when we try to send them to the
1700 * datapath, which doesn't support UINT16_MAX+1 ports. */
1702 bridge_fetch_dp_ifaces(struct bridge *br)
1704 struct xflow_port *xfif_ports;
1705 size_t n_xfif_ports;
1708 /* Reset all interface numbers. */
1709 for (i = 0; i < br->n_ports; i++) {
1710 struct port *port = br->ports[i];
1711 for (j = 0; j < port->n_ifaces; j++) {
1712 struct iface *iface = port->ifaces[j];
1713 iface->xf_ifidx = -1;
1716 port_array_clear(&br->ifaces);
1718 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
1719 for (i = 0; i < n_xfif_ports; i++) {
1720 struct xflow_port *p = &xfif_ports[i];
1721 struct iface *iface = iface_lookup(br, p->devname);
1723 if (iface->xf_ifidx >= 0) {
1724 VLOG_WARN("%s reported interface %s twice",
1725 xfif_name(br->xfif), p->devname);
1726 } else if (iface_from_xf_ifidx(br, p->port)) {
1727 VLOG_WARN("%s reported interface %"PRIu16" twice",
1728 xfif_name(br->xfif), p->port);
1730 port_array_set(&br->ifaces, p->port, iface);
1731 iface->xf_ifidx = p->port;
1735 int64_t ofport = (iface->xf_ifidx >= 0
1736 ? xflow_port_to_ofp_port(iface->xf_ifidx)
1738 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1745 /* Bridge packet processing functions. */
1748 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1750 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1753 static struct bond_entry *
1754 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1756 return &port->bond_hash[bond_hash(mac)];
1760 bond_choose_iface(const struct port *port)
1762 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1763 size_t i, best_down_slave = -1;
1764 long long next_delay_expiration = LLONG_MAX;
1766 for (i = 0; i < port->n_ifaces; i++) {
1767 struct iface *iface = port->ifaces[i];
1769 if (iface->enabled) {
1771 } else if (iface->delay_expires < next_delay_expiration) {
1772 best_down_slave = i;
1773 next_delay_expiration = iface->delay_expires;
1777 if (best_down_slave != -1) {
1778 struct iface *iface = port->ifaces[best_down_slave];
1780 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1781 "since no other interface is up", iface->name,
1782 iface->delay_expires - time_msec());
1783 bond_enable_slave(iface, true);
1786 return best_down_slave;
1790 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1791 uint16_t *xf_ifidx, tag_type *tags)
1793 struct iface *iface;
1795 assert(port->n_ifaces);
1796 if (port->n_ifaces == 1) {
1797 iface = port->ifaces[0];
1799 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1800 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1801 || !port->ifaces[e->iface_idx]->enabled) {
1802 /* XXX select interface properly. The current interface selection
1803 * is only good for testing the rebalancing code. */
1804 e->iface_idx = bond_choose_iface(port);
1805 if (e->iface_idx < 0) {
1806 *tags |= port->no_ifaces_tag;
1809 e->iface_tag = tag_create_random();
1810 ((struct port *) port)->bond_compat_is_stale = true;
1812 *tags |= e->iface_tag;
1813 iface = port->ifaces[e->iface_idx];
1815 *xf_ifidx = iface->xf_ifidx;
1816 *tags |= iface->tag; /* Currently only used for bonding. */
1821 bond_link_status_update(struct iface *iface, bool carrier)
1823 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1824 struct port *port = iface->port;
1826 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1827 /* Nothing to do. */
1830 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1831 iface->name, carrier ? "detected" : "dropped");
1832 if (carrier == iface->enabled) {
1833 iface->delay_expires = LLONG_MAX;
1834 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1835 iface->name, carrier ? "disabled" : "enabled");
1836 } else if (carrier && port->active_iface < 0) {
1837 bond_enable_slave(iface, true);
1838 if (port->updelay) {
1839 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1840 "other interface is up", iface->name, port->updelay);
1843 int delay = carrier ? port->updelay : port->downdelay;
1844 iface->delay_expires = time_msec() + delay;
1847 "interface %s: will be %s if it stays %s for %d ms",
1849 carrier ? "enabled" : "disabled",
1850 carrier ? "up" : "down",
1857 bond_choose_active_iface(struct port *port)
1859 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1861 port->active_iface = bond_choose_iface(port);
1862 port->active_iface_tag = tag_create_random();
1863 if (port->active_iface >= 0) {
1864 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1865 port->name, port->ifaces[port->active_iface]->name);
1867 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1873 bond_enable_slave(struct iface *iface, bool enable)
1875 struct port *port = iface->port;
1876 struct bridge *br = port->bridge;
1878 /* This acts as a recursion check. If the act of disabling a slave
1879 * causes a different slave to be enabled, the flag will allow us to
1880 * skip redundant work when we reenter this function. It must be
1881 * cleared on exit to keep things safe with multiple bonds. */
1882 static bool moving_active_iface = false;
1884 iface->delay_expires = LLONG_MAX;
1885 if (enable == iface->enabled) {
1889 iface->enabled = enable;
1890 if (!iface->enabled) {
1891 VLOG_WARN("interface %s: disabled", iface->name);
1892 ofproto_revalidate(br->ofproto, iface->tag);
1893 if (iface->port_ifidx == port->active_iface) {
1894 ofproto_revalidate(br->ofproto,
1895 port->active_iface_tag);
1897 /* Disabling a slave can lead to another slave being immediately
1898 * enabled if there will be no active slaves but one is waiting
1899 * on an updelay. In this case we do not need to run most of the
1900 * code for the newly enabled slave since there was no period
1901 * without an active slave and it is redundant with the disabling
1903 moving_active_iface = true;
1904 bond_choose_active_iface(port);
1906 bond_send_learning_packets(port);
1908 VLOG_WARN("interface %s: enabled", iface->name);
1909 if (port->active_iface < 0 && !moving_active_iface) {
1910 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1911 bond_choose_active_iface(port);
1912 bond_send_learning_packets(port);
1914 iface->tag = tag_create_random();
1917 moving_active_iface = false;
1918 port->bond_compat_is_stale = true;
1921 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1922 * bond interface. */
1924 bond_update_fake_iface_stats(struct port *port)
1926 struct netdev_stats bond_stats;
1927 struct netdev *bond_dev;
1930 memset(&bond_stats, 0, sizeof bond_stats);
1932 for (i = 0; i < port->n_ifaces; i++) {
1933 struct netdev_stats slave_stats;
1935 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1936 bond_stats.rx_packets += slave_stats.rx_packets;
1937 bond_stats.rx_bytes += slave_stats.rx_bytes;
1938 bond_stats.tx_packets += slave_stats.tx_packets;
1939 bond_stats.tx_bytes += slave_stats.tx_bytes;
1943 if (!netdev_open_default(port->name, &bond_dev)) {
1944 netdev_set_stats(bond_dev, &bond_stats);
1945 netdev_close(bond_dev);
1950 bond_run(struct bridge *br)
1954 for (i = 0; i < br->n_ports; i++) {
1955 struct port *port = br->ports[i];
1957 if (port->n_ifaces >= 2) {
1958 for (j = 0; j < port->n_ifaces; j++) {
1959 struct iface *iface = port->ifaces[j];
1960 if (time_msec() >= iface->delay_expires) {
1961 bond_enable_slave(iface, !iface->enabled);
1965 if (port->bond_fake_iface
1966 && time_msec() >= port->bond_next_fake_iface_update) {
1967 bond_update_fake_iface_stats(port);
1968 port->bond_next_fake_iface_update = time_msec() + 1000;
1972 if (port->bond_compat_is_stale) {
1973 port->bond_compat_is_stale = false;
1974 port_update_bond_compat(port);
1980 bond_wait(struct bridge *br)
1984 for (i = 0; i < br->n_ports; i++) {
1985 struct port *port = br->ports[i];
1986 if (port->n_ifaces < 2) {
1989 for (j = 0; j < port->n_ifaces; j++) {
1990 struct iface *iface = port->ifaces[j];
1991 if (iface->delay_expires != LLONG_MAX) {
1992 poll_timer_wait_until(iface->delay_expires);
1995 if (port->bond_fake_iface) {
1996 poll_timer_wait_until(port->bond_next_fake_iface_update);
2002 set_dst(struct dst *p, const flow_t *flow,
2003 const struct port *in_port, const struct port *out_port,
2006 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2007 : in_port->vlan >= 0 ? in_port->vlan
2008 : ntohs(flow->dl_vlan));
2009 return choose_output_iface(out_port, flow->dl_src, &p->xf_ifidx, tags);
2013 swap_dst(struct dst *p, struct dst *q)
2015 struct dst tmp = *p;
2020 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2021 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2022 * that we push to the datapath. We could in fact fully sort the array by
2023 * vlan, but in most cases there are at most two different vlan tags so that's
2024 * possibly overkill.) */
2026 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2028 struct dst *first = dsts;
2029 struct dst *last = dsts + n_dsts;
2031 while (first != last) {
2033 * - All dsts < first have vlan == 'vlan'.
2034 * - All dsts >= last have vlan != 'vlan'.
2035 * - first < last. */
2036 while (first->vlan == vlan) {
2037 if (++first == last) {
2042 /* Same invariants, plus one additional:
2043 * - first->vlan != vlan.
2045 while (last[-1].vlan != vlan) {
2046 if (--last == first) {
2051 /* Same invariants, plus one additional:
2052 * - last[-1].vlan == vlan.*/
2053 swap_dst(first++, --last);
2058 mirror_mask_ffs(mirror_mask_t mask)
2060 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2065 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2066 const struct dst *test)
2069 for (i = 0; i < n_dsts; i++) {
2070 if (dsts[i].vlan == test->vlan && dsts[i].xf_ifidx == test->xf_ifidx) {
2078 port_trunks_vlan(const struct port *port, uint16_t vlan)
2080 return (port->vlan < 0
2081 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2085 port_includes_vlan(const struct port *port, uint16_t vlan)
2087 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2091 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2092 const struct port *in_port, const struct port *out_port,
2093 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2095 mirror_mask_t mirrors = in_port->src_mirrors;
2096 struct dst *dst = dsts;
2099 if (out_port == FLOOD_PORT) {
2100 /* XXX use XFLOW_FLOOD if no vlans or bonding. */
2101 /* XXX even better, define each VLAN as a datapath port group */
2102 for (i = 0; i < br->n_ports; i++) {
2103 struct port *port = br->ports[i];
2104 if (port != in_port && port_includes_vlan(port, vlan)
2105 && !port->is_mirror_output_port
2106 && set_dst(dst, flow, in_port, port, tags)) {
2107 mirrors |= port->dst_mirrors;
2111 *nf_output_iface = NF_OUT_FLOOD;
2112 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2113 *nf_output_iface = dst->xf_ifidx;
2114 mirrors |= out_port->dst_mirrors;
2119 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2120 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2122 if (set_dst(dst, flow, in_port, m->out_port, tags)
2123 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2127 for (i = 0; i < br->n_ports; i++) {
2128 struct port *port = br->ports[i];
2129 if (port_includes_vlan(port, m->out_vlan)
2130 && set_dst(dst, flow, in_port, port, tags))
2134 if (port->vlan < 0) {
2135 dst->vlan = m->out_vlan;
2137 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2141 /* Use the vlan tag on the original flow instead of
2142 * the one passed in the vlan parameter. This ensures
2143 * that we compare the vlan from before any implicit
2144 * tagging tags place. This is necessary because
2145 * dst->vlan is the final vlan, after removing implicit
2147 flow_vlan = ntohs(flow->dl_vlan);
2148 if (flow_vlan == 0) {
2149 flow_vlan = OFP_VLAN_NONE;
2151 if (port == in_port && dst->vlan == flow_vlan) {
2152 /* Don't send out input port on same VLAN. */
2160 mirrors &= mirrors - 1;
2163 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2167 static void OVS_UNUSED
2168 print_dsts(const struct dst *dsts, size_t n)
2170 for (; n--; dsts++) {
2171 printf(">p%"PRIu16, dsts->xf_ifidx);
2172 if (dsts->vlan != OFP_VLAN_NONE) {
2173 printf("v%"PRIu16, dsts->vlan);
2179 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2180 const struct port *in_port, const struct port *out_port,
2181 tag_type *tags, struct xflow_actions *actions,
2182 uint16_t *nf_output_iface)
2184 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2186 const struct dst *p;
2189 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2192 cur_vlan = ntohs(flow->dl_vlan);
2193 for (p = dsts; p < &dsts[n_dsts]; p++) {
2194 union xflow_action *a;
2195 if (p->vlan != cur_vlan) {
2196 if (p->vlan == OFP_VLAN_NONE) {
2197 xflow_actions_add(actions, XFLOWAT_STRIP_VLAN);
2199 a = xflow_actions_add(actions, XFLOWAT_SET_DL_TCI);
2200 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2201 a->dl_tci.mask = htons(VLAN_VID_MASK);
2205 a = xflow_actions_add(actions, XFLOWAT_OUTPUT);
2206 a->output.port = p->xf_ifidx;
2210 /* Returns the effective vlan of a packet, taking into account both the
2211 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2212 * the packet is untagged and -1 indicates it has an invalid header and
2213 * should be dropped. */
2214 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2215 struct port *in_port, bool have_packet)
2217 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2218 * belongs to VLAN 0, so we should treat both cases identically. (In the
2219 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2220 * presumably to allow a priority to be specified. In the latter case, the
2221 * packet does not have any 802.1Q header.) */
2222 int vlan = ntohs(flow->dl_vlan);
2223 if (vlan == OFP_VLAN_NONE) {
2226 if (in_port->vlan >= 0) {
2228 /* XXX support double tagging? */
2230 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2231 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2232 "packet received on port %s configured with "
2233 "implicit VLAN %"PRIu16,
2234 br->name, ntohs(flow->dl_vlan),
2235 in_port->name, in_port->vlan);
2239 vlan = in_port->vlan;
2241 if (!port_includes_vlan(in_port, vlan)) {
2243 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2244 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2245 "packet received on port %s not configured for "
2247 br->name, vlan, in_port->name, vlan);
2257 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2258 struct port *in_port)
2260 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2261 vlan, in_port->port_idx);
2263 /* The log messages here could actually be useful in debugging,
2264 * so keep the rate limit relatively high. */
2265 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2267 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2268 "on port %s in VLAN %d",
2269 br->name, ETH_ADDR_ARGS(flow->dl_src),
2270 in_port->name, vlan);
2271 ofproto_revalidate(br->ofproto, rev_tag);
2275 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2276 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2277 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2279 is_gratuitous_arp(const flow_t *flow)
2281 return (flow->dl_type == htons(ETH_TYPE_ARP)
2282 && eth_addr_is_broadcast(flow->dl_dst)
2283 && (flow->nw_proto == ARP_OP_REPLY
2284 || (flow->nw_proto == ARP_OP_REQUEST
2285 && flow->nw_src == flow->nw_dst)));
2288 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2289 * dropped. Returns true if they may be forwarded, false if they should be
2292 * If 'have_packet' is true, it indicates that the caller is processing a
2293 * received packet. If 'have_packet' is false, then the caller is just
2294 * revalidating an existing flow because configuration has changed. Either
2295 * way, 'have_packet' only affects logging (there is no point in logging errors
2296 * during revalidation).
2298 * Sets '*in_portp' to the input port. This will be a null pointer if
2299 * flow->in_port does not designate a known input port (in which case
2300 * is_admissible() returns false).
2302 * When returning true, sets '*vlanp' to the effective VLAN of the input
2303 * packet, as returned by flow_get_vlan().
2305 * May also add tags to '*tags', although the current implementation only does
2306 * so in one special case.
2309 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2310 tag_type *tags, int *vlanp, struct port **in_portp)
2312 struct iface *in_iface;
2313 struct port *in_port;
2316 /* Find the interface and port structure for the received packet. */
2317 in_iface = iface_from_xf_ifidx(br, flow->in_port);
2319 /* No interface? Something fishy... */
2321 /* Odd. A few possible reasons here:
2323 * - We deleted an interface but there are still a few packets
2324 * queued up from it.
2326 * - Someone externally added an interface (e.g. with "ovs-dpctl
2327 * add-if") that we don't know about.
2329 * - Packet arrived on the local port but the local port is not
2330 * one of our bridge ports.
2332 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2334 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2335 "interface %"PRIu16, br->name, flow->in_port);
2341 *in_portp = in_port = in_iface->port;
2342 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2347 /* Drop frames for reserved multicast addresses. */
2348 if (eth_addr_is_reserved(flow->dl_dst)) {
2352 /* Drop frames on ports reserved for mirroring. */
2353 if (in_port->is_mirror_output_port) {
2355 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2356 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2357 "%s, which is reserved exclusively for mirroring",
2358 br->name, in_port->name);
2363 /* Packets received on bonds need special attention to avoid duplicates. */
2364 if (in_port->n_ifaces > 1) {
2367 if (eth_addr_is_multicast(flow->dl_dst)) {
2368 *tags |= in_port->active_iface_tag;
2369 if (in_port->active_iface != in_iface->port_ifidx) {
2370 /* Drop all multicast packets on inactive slaves. */
2375 /* Drop all packets for which we have learned a different input
2376 * port, because we probably sent the packet on one slave and got
2377 * it back on the other. Gratuitous ARP packets are an exception
2378 * to this rule: the host has moved to another switch. */
2379 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2380 if (src_idx != -1 && src_idx != in_port->port_idx &&
2381 !is_gratuitous_arp(flow)) {
2389 /* If the composed actions may be applied to any packet in the given 'flow',
2390 * returns true. Otherwise, the actions should only be applied to 'packet', or
2391 * not at all, if 'packet' was NULL. */
2393 process_flow(struct bridge *br, const flow_t *flow,
2394 const struct ofpbuf *packet, struct xflow_actions *actions,
2395 tag_type *tags, uint16_t *nf_output_iface)
2397 struct port *in_port;
2398 struct port *out_port;
2402 /* Check whether we should drop packets in this flow. */
2403 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2408 /* Learn source MAC (but don't try to learn from revalidation). */
2410 update_learning_table(br, flow, vlan, in_port);
2413 /* Determine output port. */
2414 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags);
2415 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2416 out_port = br->ports[out_port_idx];
2417 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2418 /* If we are revalidating but don't have a learning entry then
2419 * eject the flow. Installing a flow that floods packets opens
2420 * up a window of time where we could learn from a packet reflected
2421 * on a bond and blackhole packets before the learning table is
2422 * updated to reflect the correct port. */
2425 out_port = FLOOD_PORT;
2428 /* Don't send packets out their input ports. */
2429 if (in_port == out_port) {
2435 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2442 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2445 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2446 const struct ofp_phy_port *opp,
2449 struct bridge *br = br_;
2450 struct iface *iface;
2453 iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(opp->port_no));
2459 if (reason == OFPPR_DELETE) {
2460 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2461 br->name, iface->name);
2462 iface_destroy(iface);
2463 if (!port->n_ifaces) {
2464 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2465 br->name, port->name);
2471 if (port->n_ifaces > 1) {
2472 bool up = !(opp->state & OFPPS_LINK_DOWN);
2473 bond_link_status_update(iface, up);
2474 port_update_bond_compat(port);
2480 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2481 struct xflow_actions *actions, tag_type *tags,
2482 uint16_t *nf_output_iface, void *br_)
2484 struct bridge *br = br_;
2486 COVERAGE_INC(bridge_process_flow);
2487 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2491 bridge_account_flow_ofhook_cb(const flow_t *flow,
2492 const union xflow_action *actions,
2493 size_t n_actions, unsigned long long int n_bytes,
2496 struct bridge *br = br_;
2497 const union xflow_action *a;
2498 struct port *in_port;
2502 /* Feed information from the active flows back into the learning table
2503 * to ensure that table is always in sync with what is actually flowing
2504 * through the datapath. */
2505 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2506 update_learning_table(br, flow, vlan, in_port);
2509 if (!br->has_bonded_ports) {
2513 for (a = actions; a < &actions[n_actions]; a++) {
2514 if (a->type == XFLOWAT_OUTPUT) {
2515 struct port *out_port = port_from_xf_ifidx(br, a->output.port);
2516 if (out_port && out_port->n_ifaces >= 2) {
2517 struct bond_entry *e = lookup_bond_entry(out_port,
2519 e->tx_bytes += n_bytes;
2526 bridge_account_checkpoint_ofhook_cb(void *br_)
2528 struct bridge *br = br_;
2532 if (!br->has_bonded_ports) {
2537 for (i = 0; i < br->n_ports; i++) {
2538 struct port *port = br->ports[i];
2539 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2540 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2541 bond_rebalance_port(port);
2546 static struct ofhooks bridge_ofhooks = {
2547 bridge_port_changed_ofhook_cb,
2548 bridge_normal_ofhook_cb,
2549 bridge_account_flow_ofhook_cb,
2550 bridge_account_checkpoint_ofhook_cb,
2553 /* Bonding functions. */
2555 /* Statistics for a single interface on a bonded port, used for load-based
2556 * bond rebalancing. */
2557 struct slave_balance {
2558 struct iface *iface; /* The interface. */
2559 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2561 /* All the "bond_entry"s that are assigned to this interface, in order of
2562 * increasing tx_bytes. */
2563 struct bond_entry **hashes;
2567 /* Sorts pointers to pointers to bond_entries in ascending order by the
2568 * interface to which they are assigned, and within a single interface in
2569 * ascending order of bytes transmitted. */
2571 compare_bond_entries(const void *a_, const void *b_)
2573 const struct bond_entry *const *ap = a_;
2574 const struct bond_entry *const *bp = b_;
2575 const struct bond_entry *a = *ap;
2576 const struct bond_entry *b = *bp;
2577 if (a->iface_idx != b->iface_idx) {
2578 return a->iface_idx > b->iface_idx ? 1 : -1;
2579 } else if (a->tx_bytes != b->tx_bytes) {
2580 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2586 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2587 * *descending* order by number of bytes transmitted. */
2589 compare_slave_balance(const void *a_, const void *b_)
2591 const struct slave_balance *a = a_;
2592 const struct slave_balance *b = b_;
2593 if (a->iface->enabled != b->iface->enabled) {
2594 return a->iface->enabled ? -1 : 1;
2595 } else if (a->tx_bytes != b->tx_bytes) {
2596 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2603 swap_bals(struct slave_balance *a, struct slave_balance *b)
2605 struct slave_balance tmp = *a;
2610 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2611 * given that 'p' (and only 'p') might be in the wrong location.
2613 * This function invalidates 'p', since it might now be in a different memory
2616 resort_bals(struct slave_balance *p,
2617 struct slave_balance bals[], size_t n_bals)
2620 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2621 swap_bals(p, p - 1);
2623 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2624 swap_bals(p, p + 1);
2630 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2632 if (VLOG_IS_DBG_ENABLED()) {
2633 struct ds ds = DS_EMPTY_INITIALIZER;
2634 const struct slave_balance *b;
2636 for (b = bals; b < bals + n_bals; b++) {
2640 ds_put_char(&ds, ',');
2642 ds_put_format(&ds, " %s %"PRIu64"kB",
2643 b->iface->name, b->tx_bytes / 1024);
2645 if (!b->iface->enabled) {
2646 ds_put_cstr(&ds, " (disabled)");
2648 if (b->n_hashes > 0) {
2649 ds_put_cstr(&ds, " (");
2650 for (i = 0; i < b->n_hashes; i++) {
2651 const struct bond_entry *e = b->hashes[i];
2653 ds_put_cstr(&ds, " + ");
2655 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2656 e - port->bond_hash, e->tx_bytes / 1024);
2658 ds_put_cstr(&ds, ")");
2661 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2666 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2668 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2671 struct bond_entry *hash = from->hashes[hash_idx];
2672 struct port *port = from->iface->port;
2673 uint64_t delta = hash->tx_bytes;
2675 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2676 "from %s to %s (now carrying %"PRIu64"kB and "
2677 "%"PRIu64"kB load, respectively)",
2678 port->name, delta / 1024, hash - port->bond_hash,
2679 from->iface->name, to->iface->name,
2680 (from->tx_bytes - delta) / 1024,
2681 (to->tx_bytes + delta) / 1024);
2683 /* Delete element from from->hashes.
2685 * We don't bother to add the element to to->hashes because not only would
2686 * it require more work, the only purpose it would be to allow that hash to
2687 * be migrated to another slave in this rebalancing run, and there is no
2688 * point in doing that. */
2689 if (hash_idx == 0) {
2692 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2693 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2697 /* Shift load away from 'from' to 'to'. */
2698 from->tx_bytes -= delta;
2699 to->tx_bytes += delta;
2701 /* Arrange for flows to be revalidated. */
2702 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2703 hash->iface_idx = to->iface->port_ifidx;
2704 hash->iface_tag = tag_create_random();
2708 bond_rebalance_port(struct port *port)
2710 struct slave_balance bals[DP_MAX_PORTS];
2712 struct bond_entry *hashes[BOND_MASK + 1];
2713 struct slave_balance *b, *from, *to;
2714 struct bond_entry *e;
2717 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2718 * descending order of tx_bytes, so that bals[0] represents the most
2719 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2722 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2723 * array for each slave_balance structure, we sort our local array of
2724 * hashes in order by slave, so that all of the hashes for a given slave
2725 * become contiguous in memory, and then we point each 'hashes' members of
2726 * a slave_balance structure to the start of a contiguous group. */
2727 n_bals = port->n_ifaces;
2728 for (b = bals; b < &bals[n_bals]; b++) {
2729 b->iface = port->ifaces[b - bals];
2734 for (i = 0; i <= BOND_MASK; i++) {
2735 hashes[i] = &port->bond_hash[i];
2737 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2738 for (i = 0; i <= BOND_MASK; i++) {
2740 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2741 b = &bals[e->iface_idx];
2742 b->tx_bytes += e->tx_bytes;
2744 b->hashes = &hashes[i];
2749 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2750 log_bals(bals, n_bals, port);
2752 /* Discard slaves that aren't enabled (which were sorted to the back of the
2753 * array earlier). */
2754 while (!bals[n_bals - 1].iface->enabled) {
2761 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2762 to = &bals[n_bals - 1];
2763 for (from = bals; from < to; ) {
2764 uint64_t overload = from->tx_bytes - to->tx_bytes;
2765 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2766 /* The extra load on 'from' (and all less-loaded slaves), compared
2767 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2768 * it is less than ~1Mbps. No point in rebalancing. */
2770 } else if (from->n_hashes == 1) {
2771 /* 'from' only carries a single MAC hash, so we can't shift any
2772 * load away from it, even though we want to. */
2775 /* 'from' is carrying significantly more load than 'to', and that
2776 * load is split across at least two different hashes. Pick a hash
2777 * to migrate to 'to' (the least-loaded slave), given that doing so
2778 * must decrease the ratio of the load on the two slaves by at
2781 * The sort order we use means that we prefer to shift away the
2782 * smallest hashes instead of the biggest ones. There is little
2783 * reason behind this decision; we could use the opposite sort
2784 * order to shift away big hashes ahead of small ones. */
2788 for (i = 0; i < from->n_hashes; i++) {
2789 double old_ratio, new_ratio;
2790 uint64_t delta = from->hashes[i]->tx_bytes;
2792 if (delta == 0 || from->tx_bytes - delta == 0) {
2793 /* Pointless move. */
2797 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2799 if (to->tx_bytes == 0) {
2800 /* Nothing on the new slave, move it. */
2804 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2805 new_ratio = (double)(from->tx_bytes - delta) /
2806 (to->tx_bytes + delta);
2808 if (new_ratio == 0) {
2809 /* Should already be covered but check to prevent division
2814 if (new_ratio < 1) {
2815 new_ratio = 1 / new_ratio;
2818 if (old_ratio - new_ratio > 0.1) {
2819 /* Would decrease the ratio, move it. */
2823 if (i < from->n_hashes) {
2824 bond_shift_load(from, to, i);
2825 port->bond_compat_is_stale = true;
2827 /* If the result of the migration changed the relative order of
2828 * 'from' and 'to' swap them back to maintain invariants. */
2829 if (order_swapped) {
2830 swap_bals(from, to);
2833 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2834 * point to different slave_balance structures. It is only
2835 * valid to do these two operations in a row at all because we
2836 * know that 'from' will not move past 'to' and vice versa. */
2837 resort_bals(from, bals, n_bals);
2838 resort_bals(to, bals, n_bals);
2845 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2846 * historical data to decay to <1% in 7 rebalancing runs. */
2847 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2853 bond_send_learning_packets(struct port *port)
2855 struct bridge *br = port->bridge;
2856 struct mac_entry *e;
2857 struct ofpbuf packet;
2858 int error, n_packets, n_errors;
2860 if (!port->n_ifaces || port->active_iface < 0) {
2864 ofpbuf_init(&packet, 128);
2865 error = n_packets = n_errors = 0;
2866 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2867 union ofp_action actions[2], *a;
2873 if (e->port == port->port_idx
2874 || !choose_output_iface(port, e->mac, &xf_ifidx, &tags)) {
2878 /* Compose actions. */
2879 memset(actions, 0, sizeof actions);
2882 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2883 a->vlan_vid.len = htons(sizeof *a);
2884 a->vlan_vid.vlan_vid = htons(e->vlan);
2887 a->output.type = htons(OFPAT_OUTPUT);
2888 a->output.len = htons(sizeof *a);
2889 a->output.port = htons(xflow_port_to_ofp_port(xf_ifidx));
2894 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2896 flow_extract(&packet, 0, XFLOWP_NONE, &flow);
2897 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2904 ofpbuf_uninit(&packet);
2907 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2908 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2909 "packets, last error was: %s",
2910 port->name, n_errors, n_packets, strerror(error));
2912 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2913 port->name, n_packets);
2917 /* Bonding unixctl user interface functions. */
2920 bond_unixctl_list(struct unixctl_conn *conn,
2921 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2923 struct ds ds = DS_EMPTY_INITIALIZER;
2924 const struct bridge *br;
2926 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2928 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2931 for (i = 0; i < br->n_ports; i++) {
2932 const struct port *port = br->ports[i];
2933 if (port->n_ifaces > 1) {
2936 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2937 for (j = 0; j < port->n_ifaces; j++) {
2938 const struct iface *iface = port->ifaces[j];
2940 ds_put_cstr(&ds, ", ");
2942 ds_put_cstr(&ds, iface->name);
2944 ds_put_char(&ds, '\n');
2948 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2952 static struct port *
2953 bond_find(const char *name)
2955 const struct bridge *br;
2957 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2960 for (i = 0; i < br->n_ports; i++) {
2961 struct port *port = br->ports[i];
2962 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2971 bond_unixctl_show(struct unixctl_conn *conn,
2972 const char *args, void *aux OVS_UNUSED)
2974 struct ds ds = DS_EMPTY_INITIALIZER;
2975 const struct port *port;
2978 port = bond_find(args);
2980 unixctl_command_reply(conn, 501, "no such bond");
2984 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2985 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2986 ds_put_format(&ds, "next rebalance: %lld ms\n",
2987 port->bond_next_rebalance - time_msec());
2988 for (j = 0; j < port->n_ifaces; j++) {
2989 const struct iface *iface = port->ifaces[j];
2990 struct bond_entry *be;
2993 ds_put_format(&ds, "slave %s: %s\n",
2994 iface->name, iface->enabled ? "enabled" : "disabled");
2995 if (j == port->active_iface) {
2996 ds_put_cstr(&ds, "\tactive slave\n");
2998 if (iface->delay_expires != LLONG_MAX) {
2999 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3000 iface->enabled ? "downdelay" : "updelay",
3001 iface->delay_expires - time_msec());
3005 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3006 int hash = be - port->bond_hash;
3007 struct mac_entry *me;
3009 if (be->iface_idx != j) {
3013 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3014 hash, be->tx_bytes / 1024);
3017 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3018 &port->bridge->ml->lrus) {
3021 if (bond_hash(me->mac) == hash
3022 && me->port != port->port_idx
3023 && choose_output_iface(port, me->mac, &xf_ifidx, &tags)
3024 && xf_ifidx == iface->xf_ifidx)
3026 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3027 ETH_ADDR_ARGS(me->mac));
3032 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3037 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3038 void *aux OVS_UNUSED)
3040 char *args = (char *) args_;
3041 char *save_ptr = NULL;
3042 char *bond_s, *hash_s, *slave_s;
3043 uint8_t mac[ETH_ADDR_LEN];
3045 struct iface *iface;
3046 struct bond_entry *entry;
3049 bond_s = strtok_r(args, " ", &save_ptr);
3050 hash_s = strtok_r(NULL, " ", &save_ptr);
3051 slave_s = strtok_r(NULL, " ", &save_ptr);
3053 unixctl_command_reply(conn, 501,
3054 "usage: bond/migrate BOND HASH SLAVE");
3058 port = bond_find(bond_s);
3060 unixctl_command_reply(conn, 501, "no such bond");
3064 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3065 == ETH_ADDR_SCAN_COUNT) {
3066 hash = bond_hash(mac);
3067 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3068 hash = atoi(hash_s) & BOND_MASK;
3070 unixctl_command_reply(conn, 501, "bad hash");
3074 iface = port_lookup_iface(port, slave_s);
3076 unixctl_command_reply(conn, 501, "no such slave");
3080 if (!iface->enabled) {
3081 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3085 entry = &port->bond_hash[hash];
3086 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3087 entry->iface_idx = iface->port_ifidx;
3088 entry->iface_tag = tag_create_random();
3089 port->bond_compat_is_stale = true;
3090 unixctl_command_reply(conn, 200, "migrated");
3094 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3095 void *aux OVS_UNUSED)
3097 char *args = (char *) args_;
3098 char *save_ptr = NULL;
3099 char *bond_s, *slave_s;
3101 struct iface *iface;
3103 bond_s = strtok_r(args, " ", &save_ptr);
3104 slave_s = strtok_r(NULL, " ", &save_ptr);
3106 unixctl_command_reply(conn, 501,
3107 "usage: bond/set-active-slave BOND SLAVE");
3111 port = bond_find(bond_s);
3113 unixctl_command_reply(conn, 501, "no such bond");
3117 iface = port_lookup_iface(port, slave_s);
3119 unixctl_command_reply(conn, 501, "no such slave");
3123 if (!iface->enabled) {
3124 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3128 if (port->active_iface != iface->port_ifidx) {
3129 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3130 port->active_iface = iface->port_ifidx;
3131 port->active_iface_tag = tag_create_random();
3132 VLOG_INFO("port %s: active interface is now %s",
3133 port->name, iface->name);
3134 bond_send_learning_packets(port);
3135 unixctl_command_reply(conn, 200, "done");
3137 unixctl_command_reply(conn, 200, "no change");
3142 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3144 char *args = (char *) args_;
3145 char *save_ptr = NULL;
3146 char *bond_s, *slave_s;
3148 struct iface *iface;
3150 bond_s = strtok_r(args, " ", &save_ptr);
3151 slave_s = strtok_r(NULL, " ", &save_ptr);
3153 unixctl_command_reply(conn, 501,
3154 "usage: bond/enable/disable-slave BOND SLAVE");
3158 port = bond_find(bond_s);
3160 unixctl_command_reply(conn, 501, "no such bond");
3164 iface = port_lookup_iface(port, slave_s);
3166 unixctl_command_reply(conn, 501, "no such slave");
3170 bond_enable_slave(iface, enable);
3171 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3175 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3176 void *aux OVS_UNUSED)
3178 enable_slave(conn, args, true);
3182 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3183 void *aux OVS_UNUSED)
3185 enable_slave(conn, args, false);
3189 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3190 void *aux OVS_UNUSED)
3192 uint8_t mac[ETH_ADDR_LEN];
3196 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3197 == ETH_ADDR_SCAN_COUNT) {
3198 hash = bond_hash(mac);
3200 hash_cstr = xasprintf("%u", hash);
3201 unixctl_command_reply(conn, 200, hash_cstr);
3204 unixctl_command_reply(conn, 501, "invalid mac");
3211 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3212 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3213 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3214 unixctl_command_register("bond/set-active-slave",
3215 bond_unixctl_set_active_slave, NULL);
3216 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3218 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3220 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3223 /* Port functions. */
3225 static struct port *
3226 port_create(struct bridge *br, const char *name)
3230 port = xzalloc(sizeof *port);
3232 port->port_idx = br->n_ports;
3234 port->trunks = NULL;
3235 port->name = xstrdup(name);
3236 port->active_iface = -1;
3238 if (br->n_ports >= br->allocated_ports) {
3239 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3242 br->ports[br->n_ports++] = port;
3243 shash_add_assert(&br->port_by_name, port->name, port);
3245 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3252 get_port_other_config(const struct ovsrec_port *port, const char *key,
3253 const char *default_value)
3255 const char *value = get_ovsrec_key_value(key,
3256 port->key_other_config,
3257 port->value_other_config,
3258 port->n_other_config);
3259 return value ? value : default_value;
3263 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3265 struct shash new_ifaces;
3268 /* Collect list of new interfaces. */
3269 shash_init(&new_ifaces);
3270 for (i = 0; i < cfg->n_interfaces; i++) {
3271 const char *name = cfg->interfaces[i]->name;
3272 shash_add_once(&new_ifaces, name, NULL);
3275 /* Get rid of deleted interfaces. */
3276 for (i = 0; i < port->n_ifaces; ) {
3277 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3278 iface_destroy(port->ifaces[i]);
3284 shash_destroy(&new_ifaces);
3288 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3290 struct shash new_ifaces;
3291 long long int next_rebalance;
3292 unsigned long *trunks;
3298 /* Update settings. */
3299 port->updelay = cfg->bond_updelay;
3300 if (port->updelay < 0) {
3303 port->updelay = cfg->bond_downdelay;
3304 if (port->downdelay < 0) {
3305 port->downdelay = 0;
3307 port->bond_rebalance_interval = atoi(
3308 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3309 if (port->bond_rebalance_interval < 1000) {
3310 port->bond_rebalance_interval = 1000;
3312 next_rebalance = time_msec() + port->bond_rebalance_interval;
3313 if (port->bond_next_rebalance > next_rebalance) {
3314 port->bond_next_rebalance = next_rebalance;
3317 /* Add new interfaces and update 'cfg' member of existing ones. */
3318 shash_init(&new_ifaces);
3319 for (i = 0; i < cfg->n_interfaces; i++) {
3320 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3321 struct iface *iface;
3323 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3324 VLOG_WARN("port %s: %s specified twice as port interface",
3325 port->name, if_cfg->name);
3329 iface = iface_lookup(port->bridge, if_cfg->name);
3331 if (iface->port != port) {
3332 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3334 port->bridge->name, if_cfg->name, iface->port->name);
3337 iface->cfg = if_cfg;
3339 iface_create(port, if_cfg);
3342 shash_destroy(&new_ifaces);
3347 if (port->n_ifaces < 2) {
3349 if (vlan >= 0 && vlan <= 4095) {
3350 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3355 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3356 * they even work as-is. But they have not been tested. */
3357 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3361 if (port->vlan != vlan) {
3363 bridge_flush(port->bridge);
3366 /* Get trunked VLANs. */
3368 if (vlan < 0 && cfg->n_trunks) {
3372 trunks = bitmap_allocate(4096);
3374 for (i = 0; i < cfg->n_trunks; i++) {
3375 int trunk = cfg->trunks[i];
3377 bitmap_set1(trunks, trunk);
3383 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3384 port->name, cfg->n_trunks);
3386 if (n_errors == cfg->n_trunks) {
3387 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3389 bitmap_free(trunks);
3392 } else if (vlan >= 0 && cfg->n_trunks) {
3393 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3397 ? port->trunks != NULL
3398 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3399 bridge_flush(port->bridge);
3401 bitmap_free(port->trunks);
3402 port->trunks = trunks;
3406 port_destroy(struct port *port)
3409 struct bridge *br = port->bridge;
3413 proc_net_compat_update_vlan(port->name, NULL, 0);
3414 proc_net_compat_update_bond(port->name, NULL);
3416 for (i = 0; i < MAX_MIRRORS; i++) {
3417 struct mirror *m = br->mirrors[i];
3418 if (m && m->out_port == port) {
3423 while (port->n_ifaces > 0) {
3424 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3427 shash_find_and_delete_assert(&br->port_by_name, port->name);
3429 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3430 del->port_idx = port->port_idx;
3433 bitmap_free(port->trunks);
3440 static struct port *
3441 port_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3443 struct iface *iface = iface_from_xf_ifidx(br, xf_ifidx);
3444 return iface ? iface->port : NULL;
3447 static struct port *
3448 port_lookup(const struct bridge *br, const char *name)
3450 return shash_find_data(&br->port_by_name, name);
3453 static struct iface *
3454 port_lookup_iface(const struct port *port, const char *name)
3456 struct iface *iface = iface_lookup(port->bridge, name);
3457 return iface && iface->port == port ? iface : NULL;
3461 port_update_bonding(struct port *port)
3463 if (port->n_ifaces < 2) {
3464 /* Not a bonded port. */
3465 if (port->bond_hash) {
3466 free(port->bond_hash);
3467 port->bond_hash = NULL;
3468 port->bond_compat_is_stale = true;
3469 port->bond_fake_iface = false;
3472 if (!port->bond_hash) {
3475 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3476 for (i = 0; i <= BOND_MASK; i++) {
3477 struct bond_entry *e = &port->bond_hash[i];
3481 port->no_ifaces_tag = tag_create_random();
3482 bond_choose_active_iface(port);
3483 port->bond_next_rebalance
3484 = time_msec() + port->bond_rebalance_interval;
3486 if (port->cfg->bond_fake_iface) {
3487 port->bond_next_fake_iface_update = time_msec();
3490 port->bond_compat_is_stale = true;
3491 port->bond_fake_iface = port->cfg->bond_fake_iface;
3496 port_update_bond_compat(struct port *port)
3498 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3499 struct compat_bond bond;
3502 if (port->n_ifaces < 2) {
3503 proc_net_compat_update_bond(port->name, NULL);
3508 bond.updelay = port->updelay;
3509 bond.downdelay = port->downdelay;
3512 bond.hashes = compat_hashes;
3513 if (port->bond_hash) {
3514 const struct bond_entry *e;
3515 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3516 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3517 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3518 cbh->hash = e - port->bond_hash;
3519 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3524 bond.n_slaves = port->n_ifaces;
3525 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3526 for (i = 0; i < port->n_ifaces; i++) {
3527 struct iface *iface = port->ifaces[i];
3528 struct compat_bond_slave *slave = &bond.slaves[i];
3529 slave->name = iface->name;
3531 /* We need to make the same determination as the Linux bonding
3532 * code to determine whether a slave should be consider "up".
3533 * The Linux function bond_miimon_inspect() supports four
3534 * BOND_LINK_* states:
3536 * - BOND_LINK_UP: carrier detected, updelay has passed.
3537 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3538 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3539 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3541 * The function bond_info_show_slave() only considers BOND_LINK_UP
3542 * to be "up" and anything else to be "down".
3544 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3548 netdev_get_etheraddr(iface->netdev, slave->mac);
3551 if (port->bond_fake_iface) {
3552 struct netdev *bond_netdev;
3554 if (!netdev_open_default(port->name, &bond_netdev)) {
3556 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3558 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3560 netdev_close(bond_netdev);
3564 proc_net_compat_update_bond(port->name, &bond);
3569 port_update_vlan_compat(struct port *port)
3571 struct bridge *br = port->bridge;
3572 char *vlandev_name = NULL;
3574 if (port->vlan > 0) {
3575 /* Figure out the name that the VLAN device should actually have, if it
3576 * existed. This takes some work because the VLAN device would not
3577 * have port->name in its name; rather, it would have the trunk port's
3578 * name, and 'port' would be attached to a bridge that also had the
3579 * VLAN device one of its ports. So we need to find a trunk port that
3580 * includes port->vlan.
3582 * There might be more than one candidate. This doesn't happen on
3583 * XenServer, so if it happens we just pick the first choice in
3584 * alphabetical order instead of creating multiple VLAN devices. */
3586 for (i = 0; i < br->n_ports; i++) {
3587 struct port *p = br->ports[i];
3588 if (port_trunks_vlan(p, port->vlan)
3590 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3592 uint8_t ea[ETH_ADDR_LEN];
3593 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3594 if (!eth_addr_is_multicast(ea) &&
3595 !eth_addr_is_reserved(ea) &&
3596 !eth_addr_is_zero(ea)) {
3597 vlandev_name = p->name;
3602 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3605 /* Interface functions. */
3607 static struct iface *
3608 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3610 struct bridge *br = port->bridge;
3611 struct iface *iface;
3612 char *name = if_cfg->name;
3615 iface = xzalloc(sizeof *iface);
3617 iface->port_ifidx = port->n_ifaces;
3618 iface->name = xstrdup(name);
3619 iface->xf_ifidx = -1;
3620 iface->tag = tag_create_random();
3621 iface->delay_expires = LLONG_MAX;
3622 iface->netdev = NULL;
3623 iface->cfg = if_cfg;
3625 shash_add_assert(&br->iface_by_name, iface->name, iface);
3627 /* Attempt to create the network interface in case it doesn't exist yet. */
3628 if (!iface_is_internal(br, iface->name)) {
3629 error = set_up_iface(if_cfg, iface, true);
3631 VLOG_WARN("could not create iface %s: %s", iface->name,
3634 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3641 if (port->n_ifaces >= port->allocated_ifaces) {
3642 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3643 sizeof *port->ifaces);
3645 port->ifaces[port->n_ifaces++] = iface;
3646 if (port->n_ifaces > 1) {
3647 br->has_bonded_ports = true;
3650 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3658 iface_destroy(struct iface *iface)
3661 struct port *port = iface->port;
3662 struct bridge *br = port->bridge;
3663 bool del_active = port->active_iface == iface->port_ifidx;
3666 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3668 if (iface->xf_ifidx >= 0) {
3669 port_array_set(&br->ifaces, iface->xf_ifidx, NULL);
3672 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3673 del->port_ifidx = iface->port_ifidx;
3675 netdev_close(iface->netdev);
3678 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3679 bond_choose_active_iface(port);
3680 bond_send_learning_packets(port);
3686 bridge_flush(port->bridge);
3690 static struct iface *
3691 iface_lookup(const struct bridge *br, const char *name)
3693 return shash_find_data(&br->iface_by_name, name);
3696 static struct iface *
3697 iface_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3699 return port_array_get(&br->ifaces, xf_ifidx);
3702 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3703 * 'br', that is, an interface that is entirely simulated within the datapath.
3704 * The local port (XFLOWP_LOCAL) is always an internal interface. Other local
3705 * interfaces are created by setting "iface.<iface>.internal = true".
3707 * In addition, we have a kluge-y feature that creates an internal port with
3708 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3709 * This feature needs to go away in the long term. Until then, this is one
3710 * reason why this function takes a name instead of a struct iface: the fake
3711 * interfaces created this way do not have a struct iface. */
3713 iface_is_internal(const struct bridge *br, const char *if_name)
3715 struct iface *iface;
3718 if (!strcmp(if_name, br->name)) {
3722 iface = iface_lookup(br, if_name);
3723 if (iface && !strcmp(iface->cfg->type, "internal")) {
3727 port = port_lookup(br, if_name);
3728 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3734 /* Set Ethernet address of 'iface', if one is specified in the configuration
3737 iface_set_mac(struct iface *iface)
3739 uint8_t ea[ETH_ADDR_LEN];
3741 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3742 if (eth_addr_is_multicast(ea)) {
3743 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3745 } else if (iface->xf_ifidx == XFLOWP_LOCAL) {
3746 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3747 iface->name, iface->name);
3749 int error = netdev_set_etheraddr(iface->netdev, ea);
3751 VLOG_ERR("interface %s: setting MAC failed (%s)",
3752 iface->name, strerror(error));
3758 /* Port mirroring. */
3761 mirror_reconfigure(struct bridge *br)
3763 struct shash old_mirrors, new_mirrors;
3764 struct shash_node *node;
3765 unsigned long *rspan_vlans;
3768 /* Collect old mirrors. */
3769 shash_init(&old_mirrors);
3770 for (i = 0; i < MAX_MIRRORS; i++) {
3771 if (br->mirrors[i]) {
3772 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3776 /* Collect new mirrors. */
3777 shash_init(&new_mirrors);
3778 for (i = 0; i < br->cfg->n_mirrors; i++) {
3779 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3780 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3781 VLOG_WARN("bridge %s: %s specified twice as mirror",
3782 br->name, cfg->name);
3786 /* Get rid of deleted mirrors and add new mirrors. */
3787 SHASH_FOR_EACH (node, &old_mirrors) {
3788 if (!shash_find(&new_mirrors, node->name)) {
3789 mirror_destroy(node->data);
3792 SHASH_FOR_EACH (node, &new_mirrors) {
3793 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3795 mirror = mirror_create(br, node->name);
3800 mirror_reconfigure_one(mirror, node->data);
3802 shash_destroy(&old_mirrors);
3803 shash_destroy(&new_mirrors);
3805 /* Update port reserved status. */
3806 for (i = 0; i < br->n_ports; i++) {
3807 br->ports[i]->is_mirror_output_port = false;
3809 for (i = 0; i < MAX_MIRRORS; i++) {
3810 struct mirror *m = br->mirrors[i];
3811 if (m && m->out_port) {
3812 m->out_port->is_mirror_output_port = true;
3816 /* Update flooded vlans (for RSPAN). */
3818 if (br->cfg->n_flood_vlans) {
3819 rspan_vlans = bitmap_allocate(4096);
3821 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3822 int64_t vlan = br->cfg->flood_vlans[i];
3823 if (vlan >= 0 && vlan < 4096) {
3824 bitmap_set1(rspan_vlans, vlan);
3825 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3828 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3833 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3838 static struct mirror *
3839 mirror_create(struct bridge *br, const char *name)
3844 for (i = 0; ; i++) {
3845 if (i >= MAX_MIRRORS) {
3846 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3847 "cannot create %s", br->name, MAX_MIRRORS, name);
3850 if (!br->mirrors[i]) {
3855 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3858 br->mirrors[i] = m = xzalloc(sizeof *m);
3861 m->name = xstrdup(name);
3862 shash_init(&m->src_ports);
3863 shash_init(&m->dst_ports);
3873 mirror_destroy(struct mirror *m)
3876 struct bridge *br = m->bridge;
3879 for (i = 0; i < br->n_ports; i++) {
3880 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3881 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3884 shash_destroy(&m->src_ports);
3885 shash_destroy(&m->dst_ports);
3888 m->bridge->mirrors[m->idx] = NULL;
3896 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3897 struct shash *names)
3901 for (i = 0; i < n_ports; i++) {
3902 const char *name = ports[i]->name;
3903 if (port_lookup(m->bridge, name)) {
3904 shash_add_once(names, name, NULL);
3906 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3907 "port %s", m->bridge->name, m->name, name);
3913 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3919 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3921 for (i = 0; i < cfg->n_select_vlan; i++) {
3922 int64_t vlan = cfg->select_vlan[i];
3923 if (vlan < 0 || vlan > 4095) {
3924 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3925 m->bridge->name, m->name, vlan);
3927 (*vlans)[n_vlans++] = vlan;
3934 vlan_is_mirrored(const struct mirror *m, int vlan)
3938 for (i = 0; i < m->n_vlans; i++) {
3939 if (m->vlans[i] == vlan) {
3947 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3951 for (i = 0; i < m->n_vlans; i++) {
3952 if (port_trunks_vlan(p, m->vlans[i])) {
3960 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3962 struct shash src_ports, dst_ports;
3963 mirror_mask_t mirror_bit;
3964 struct port *out_port;
3970 /* Get output port. */
3971 if (cfg->output_port) {
3972 out_port = port_lookup(m->bridge, cfg->output_port->name);
3974 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3975 m->bridge->name, m->name);
3981 if (cfg->output_vlan) {
3982 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3983 "output vlan; ignoring output vlan",
3984 m->bridge->name, m->name);
3986 } else if (cfg->output_vlan) {
3988 out_vlan = *cfg->output_vlan;
3990 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3991 m->bridge->name, m->name);
3996 shash_init(&src_ports);
3997 shash_init(&dst_ports);
3998 if (cfg->select_all) {
3999 for (i = 0; i < m->bridge->n_ports; i++) {
4000 const char *name = m->bridge->ports[i]->name;
4001 shash_add_once(&src_ports, name, NULL);
4002 shash_add_once(&dst_ports, name, NULL);
4007 /* Get ports, and drop duplicates and ports that don't exist. */
4008 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4010 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4013 /* Get all the vlans, and drop duplicate and invalid vlans. */
4014 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4017 /* Update mirror data. */
4018 if (!shash_equal_keys(&m->src_ports, &src_ports)
4019 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4020 || m->n_vlans != n_vlans
4021 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4022 || m->out_port != out_port
4023 || m->out_vlan != out_vlan) {
4024 bridge_flush(m->bridge);
4026 shash_swap(&m->src_ports, &src_ports);
4027 shash_swap(&m->dst_ports, &dst_ports);
4030 m->n_vlans = n_vlans;
4031 m->out_port = out_port;
4032 m->out_vlan = out_vlan;
4035 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4036 for (i = 0; i < m->bridge->n_ports; i++) {
4037 struct port *port = m->bridge->ports[i];
4039 if (shash_find(&m->src_ports, port->name)
4042 ? port_trunks_any_mirrored_vlan(m, port)
4043 : vlan_is_mirrored(m, port->vlan)))) {
4044 port->src_mirrors |= mirror_bit;
4046 port->src_mirrors &= ~mirror_bit;
4049 if (shash_find(&m->dst_ports, port->name)) {
4050 port->dst_mirrors |= mirror_bit;
4052 port->dst_mirrors &= ~mirror_bit;
4057 shash_destroy(&src_ports);
4058 shash_destroy(&dst_ports);