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>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
50 #include "poll-loop.h"
51 #include "port-array.h"
52 #include "proc-net-compat.h"
56 #include "socket-util.h"
57 #include "stream-ssl.h"
63 #include "vswitchd/vswitch-idl.h"
64 #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 dp_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 dpif *dpif; /* 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_dp_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_dp_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->dp_ifidx < 0) {
288 VLOG_ERR("%s interface not in datapath %s, ignoring",
289 iface->name, dpif_name(br->dpif));
291 if (iface->dp_ifidx != ODPP_LOCAL) {
292 svec_add(svec, iface->name);
301 bridge_init(const struct ovsrec_open_vswitch *cfg)
303 struct svec bridge_names;
304 struct svec dpif_names, dpif_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(&dpif_names);
316 svec_init(&dpif_types);
317 dp_enumerate_types(&dpif_types);
318 for (i = 0; i < dpif_types.n; i++) {
323 dp_enumerate_names(dpif_types.names[i], &dpif_names);
325 for (j = 0; j < dpif_names.n; j++) {
326 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
328 struct svec all_names;
331 svec_init(&all_names);
332 dpif_get_all_names(dpif, &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(&dpif_names);
347 svec_destroy(&dpif_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 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
401 netdev_options.may_open = true;
404 error = netdev_open(&netdev_options, &iface->netdev);
407 netdev_get_carrier(iface->netdev, &iface->enabled);
409 } else if (iface->netdev) {
410 const char *netdev_type = netdev_get_type(iface->netdev);
411 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
412 ? iface_cfg->type : NULL;
414 /* An "internal" config type maps to a netdev "system" type. */
415 if (iface_type && !strcmp(iface_type, "internal")) {
416 iface_type = "system";
419 if (!iface_type || !strcmp(netdev_type, iface_type)) {
420 error = netdev_reconfigure(iface->netdev, &options);
422 VLOG_WARN("%s: attempting change device type from %s to %s",
423 iface_cfg->name, netdev_type, iface_type);
428 SHASH_FOR_EACH (node, &options) {
431 shash_destroy(&options);
437 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
439 return set_up_iface(iface_cfg, iface, false);
443 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
444 void *aux OVS_UNUSED)
446 if (!iface->netdev) {
447 int error = set_up_iface(iface->cfg, iface, true);
449 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
459 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
460 void *aux OVS_UNUSED)
462 if (iface->dp_ifidx >= 0) {
463 VLOG_DBG("%s has interface %s on port %d",
465 iface->name, iface->dp_ifidx);
468 VLOG_ERR("%s interface not in %s, dropping",
469 iface->name, dpif_name(br->dpif));
475 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
476 void *aux OVS_UNUSED)
478 /* Set policing attributes. */
479 netdev_set_policing(iface->netdev,
480 iface->cfg->ingress_policing_rate,
481 iface->cfg->ingress_policing_burst);
483 /* Set MAC address of internal interfaces other than the local
485 if (iface->dp_ifidx != ODPP_LOCAL
486 && iface_is_internal(br, iface->name)) {
487 iface_set_mac(iface);
493 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
494 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
495 * deletes from 'br' any ports that no longer have any interfaces. */
497 iterate_and_prune_ifaces(struct bridge *br,
498 bool (*cb)(struct bridge *, struct iface *,
504 for (i = 0; i < br->n_ports; ) {
505 struct port *port = br->ports[i];
506 for (j = 0; j < port->n_ifaces; ) {
507 struct iface *iface = port->ifaces[j];
508 if (cb(br, iface, aux)) {
511 iface_destroy(iface);
515 if (port->n_ifaces) {
518 VLOG_ERR("%s port has no interfaces, dropping", port->name);
524 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
525 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
526 * responsible for freeing '*managersp' (with free()).
528 * You may be asking yourself "why does ovs-vswitchd care?", because
529 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
530 * should not be and in fact is not directly involved in that. But
531 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
532 * it has to tell in-band control where the managers are to enable that.
535 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
536 struct sockaddr_in **managersp, size_t *n_managersp)
538 struct sockaddr_in *managers = NULL;
539 size_t n_managers = 0;
541 if (ovs_cfg->n_managers > 0) {
544 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
545 for (i = 0; i < ovs_cfg->n_managers; i++) {
546 const char *name = ovs_cfg->managers[i];
547 struct sockaddr_in *sin = &managers[i];
549 if ((!strncmp(name, "tcp:", 4)
550 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
551 (!strncmp(name, "ssl:", 4)
552 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
558 *managersp = managers;
559 *n_managersp = n_managers;
563 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
565 struct ovsdb_idl_txn *txn;
566 struct shash old_br, new_br;
567 struct shash_node *node;
568 struct bridge *br, *next;
569 struct sockaddr_in *managers;
572 int sflow_bridge_number;
574 COVERAGE_INC(bridge_reconfigure);
576 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
578 collect_managers(ovs_cfg, &managers, &n_managers);
580 /* Collect old and new bridges. */
583 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
584 shash_add(&old_br, br->name, br);
586 for (i = 0; i < ovs_cfg->n_bridges; i++) {
587 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
588 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
589 VLOG_WARN("more than one bridge named %s", br_cfg->name);
593 /* Get rid of deleted bridges and add new bridges. */
594 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
595 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
602 SHASH_FOR_EACH (node, &new_br) {
603 const char *br_name = node->name;
604 const struct ovsrec_bridge *br_cfg = node->data;
605 br = shash_find_data(&old_br, br_name);
607 /* If the bridge datapath type has changed, we need to tear it
608 * down and recreate. */
609 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
611 bridge_create(br_cfg);
614 bridge_create(br_cfg);
617 shash_destroy(&old_br);
618 shash_destroy(&new_br);
622 bridge_configure_ssl(ovs_cfg->ssl);
625 /* Reconfigure all bridges. */
626 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
627 bridge_reconfigure_one(ovs_cfg, br);
630 /* Add and delete ports on all datapaths.
632 * The kernel will reject any attempt to add a given port to a datapath if
633 * that port already belongs to a different datapath, so we must do all
634 * port deletions before any port additions. */
635 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
636 struct odp_port *dpif_ports;
638 struct shash want_ifaces;
640 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
641 bridge_get_all_ifaces(br, &want_ifaces);
642 for (i = 0; i < n_dpif_ports; i++) {
643 const struct odp_port *p = &dpif_ports[i];
644 if (!shash_find(&want_ifaces, p->devname)
645 && strcmp(p->devname, br->name)) {
646 int retval = dpif_port_del(br->dpif, p->port);
648 VLOG_ERR("failed to remove %s interface from %s: %s",
649 p->devname, dpif_name(br->dpif),
654 shash_destroy(&want_ifaces);
657 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
658 struct odp_port *dpif_ports;
660 struct shash cur_ifaces, want_ifaces;
661 struct shash_node *node;
663 /* Get the set of interfaces currently in this datapath. */
664 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
665 shash_init(&cur_ifaces);
666 for (i = 0; i < n_dpif_ports; i++) {
667 const char *name = dpif_ports[i].devname;
668 if (!shash_find(&cur_ifaces, name)) {
669 shash_add(&cur_ifaces, name, NULL);
674 /* Get the set of interfaces we want on this datapath. */
675 bridge_get_all_ifaces(br, &want_ifaces);
677 SHASH_FOR_EACH (node, &want_ifaces) {
678 const char *if_name = node->name;
679 struct iface *iface = node->data;
681 if (shash_find(&cur_ifaces, if_name)) {
682 /* Already exists, just reconfigure it. */
684 reconfigure_iface(iface->cfg, iface);
687 /* Need to add to datapath. */
691 /* Add to datapath. */
692 internal = iface_is_internal(br, if_name);
693 error = dpif_port_add(br->dpif, if_name,
694 internal ? ODP_PORT_INTERNAL : 0, NULL);
695 if (error == EFBIG) {
696 VLOG_ERR("ran out of valid port numbers on %s",
697 dpif_name(br->dpif));
700 VLOG_ERR("failed to add %s interface to %s: %s",
701 if_name, dpif_name(br->dpif), strerror(error));
705 shash_destroy(&cur_ifaces);
706 shash_destroy(&want_ifaces);
708 sflow_bridge_number = 0;
709 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
712 struct iface *local_iface;
713 struct iface *hw_addr_iface;
716 bridge_fetch_dp_ifaces(br);
718 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
719 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
721 /* Pick local port hardware address, datapath ID. */
722 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
723 local_iface = bridge_get_local_iface(br);
725 int error = netdev_set_etheraddr(local_iface->netdev, ea);
727 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
728 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
729 "Ethernet address: %s",
730 br->name, strerror(error));
734 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
735 ofproto_set_datapath_id(br->ofproto, dpid);
737 dpid_string = xasprintf("%012"PRIx64, dpid);
738 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
741 /* Set NetFlow configuration on this bridge. */
742 if (br->cfg->netflow) {
743 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
744 struct netflow_options opts;
746 memset(&opts, 0, sizeof opts);
748 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
749 if (nf_cfg->engine_type) {
750 opts.engine_type = *nf_cfg->engine_type;
752 if (nf_cfg->engine_id) {
753 opts.engine_id = *nf_cfg->engine_id;
756 opts.active_timeout = nf_cfg->active_timeout;
757 if (!opts.active_timeout) {
758 opts.active_timeout = -1;
759 } else if (opts.active_timeout < 0) {
760 VLOG_WARN("bridge %s: active timeout interval set to negative "
761 "value, using default instead (%d seconds)", br->name,
762 NF_ACTIVE_TIMEOUT_DEFAULT);
763 opts.active_timeout = -1;
766 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
767 if (opts.add_id_to_iface) {
768 if (opts.engine_id > 0x7f) {
769 VLOG_WARN("bridge %s: netflow port mangling may conflict "
770 "with another vswitch, choose an engine id less "
771 "than 128", br->name);
773 if (br->n_ports > 508) {
774 VLOG_WARN("bridge %s: netflow port mangling will conflict "
775 "with another port when more than 508 ports are "
780 opts.collectors.n = nf_cfg->n_targets;
781 opts.collectors.names = nf_cfg->targets;
782 if (ofproto_set_netflow(br->ofproto, &opts)) {
783 VLOG_ERR("bridge %s: problem setting netflow collectors",
787 ofproto_set_netflow(br->ofproto, NULL);
790 /* Set sFlow configuration on this bridge. */
791 if (br->cfg->sflow) {
792 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
793 struct ovsrec_controller **controllers;
794 struct ofproto_sflow_options oso;
795 size_t n_controllers;
798 memset(&oso, 0, sizeof oso);
800 oso.targets.n = sflow_cfg->n_targets;
801 oso.targets.names = sflow_cfg->targets;
803 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
804 if (sflow_cfg->sampling) {
805 oso.sampling_rate = *sflow_cfg->sampling;
808 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
809 if (sflow_cfg->polling) {
810 oso.polling_interval = *sflow_cfg->polling;
813 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
814 if (sflow_cfg->header) {
815 oso.header_len = *sflow_cfg->header;
818 oso.sub_id = sflow_bridge_number++;
819 oso.agent_device = sflow_cfg->agent;
821 oso.control_ip = NULL;
822 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
823 for (i = 0; i < n_controllers; i++) {
824 if (controllers[i]->local_ip) {
825 oso.control_ip = controllers[i]->local_ip;
829 ofproto_set_sflow(br->ofproto, &oso);
831 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
833 ofproto_set_sflow(br->ofproto, NULL);
836 /* Update the controller and related settings. It would be more
837 * straightforward to call this from bridge_reconfigure_one(), but we
838 * can't do it there for two reasons. First, and most importantly, at
839 * that point we don't know the dp_ifidx of any interfaces that have
840 * been added to the bridge (because we haven't actually added them to
841 * the datapath). Second, at that point we haven't set the datapath ID
842 * yet; when a controller is configured, resetting the datapath ID will
843 * immediately disconnect from the controller, so it's better to set
844 * the datapath ID before the controller. */
845 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
847 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
848 for (i = 0; i < br->n_ports; i++) {
849 struct port *port = br->ports[i];
851 port_update_vlan_compat(port);
852 port_update_bonding(port);
855 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
856 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
859 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
861 ovsdb_idl_txn_commit(txn);
862 ovsdb_idl_txn_destroy(txn); /* XXX */
868 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
872 for (i = 0; i < n; i++) {
873 if (!strcmp(keys[i], key)) {
881 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
883 return get_ovsrec_key_value(key,
884 br_cfg->key_other_config,
885 br_cfg->value_other_config,
886 br_cfg->n_other_config);
890 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
891 struct iface **hw_addr_iface)
897 *hw_addr_iface = NULL;
899 /* Did the user request a particular MAC? */
900 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
901 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
902 if (eth_addr_is_multicast(ea)) {
903 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
904 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
905 } else if (eth_addr_is_zero(ea)) {
906 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
912 /* Otherwise choose the minimum non-local MAC address among all of the
914 memset(ea, 0xff, sizeof ea);
915 for (i = 0; i < br->n_ports; i++) {
916 struct port *port = br->ports[i];
917 uint8_t iface_ea[ETH_ADDR_LEN];
920 /* Mirror output ports don't participate. */
921 if (port->is_mirror_output_port) {
925 /* Choose the MAC address to represent the port. */
926 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
927 /* Find the interface with this Ethernet address (if any) so that
928 * we can provide the correct devname to the caller. */
930 for (j = 0; j < port->n_ifaces; j++) {
931 struct iface *candidate = port->ifaces[j];
932 uint8_t candidate_ea[ETH_ADDR_LEN];
933 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
934 && eth_addr_equals(iface_ea, candidate_ea)) {
939 /* Choose the interface whose MAC address will represent the port.
940 * The Linux kernel bonding code always chooses the MAC address of
941 * the first slave added to a bond, and the Fedora networking
942 * scripts always add slaves to a bond in alphabetical order, so
943 * for compatibility we choose the interface with the name that is
944 * first in alphabetical order. */
945 iface = port->ifaces[0];
946 for (j = 1; j < port->n_ifaces; j++) {
947 struct iface *candidate = port->ifaces[j];
948 if (strcmp(candidate->name, iface->name) < 0) {
953 /* The local port doesn't count (since we're trying to choose its
954 * MAC address anyway). */
955 if (iface->dp_ifidx == ODPP_LOCAL) {
960 error = netdev_get_etheraddr(iface->netdev, iface_ea);
962 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
963 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
964 iface->name, strerror(error));
969 /* Compare against our current choice. */
970 if (!eth_addr_is_multicast(iface_ea) &&
971 !eth_addr_is_local(iface_ea) &&
972 !eth_addr_is_reserved(iface_ea) &&
973 !eth_addr_is_zero(iface_ea) &&
974 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
976 memcpy(ea, iface_ea, ETH_ADDR_LEN);
977 *hw_addr_iface = iface;
980 if (eth_addr_is_multicast(ea)) {
981 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
982 *hw_addr_iface = NULL;
983 VLOG_WARN("bridge %s: using default bridge Ethernet "
984 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
986 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
987 br->name, ETH_ADDR_ARGS(ea));
991 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
992 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
993 * an interface on 'br', then that interface must be passed in as
994 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
995 * 'hw_addr_iface' must be passed in as a null pointer. */
997 bridge_pick_datapath_id(struct bridge *br,
998 const uint8_t bridge_ea[ETH_ADDR_LEN],
999 struct iface *hw_addr_iface)
1002 * The procedure for choosing a bridge MAC address will, in the most
1003 * ordinary case, also choose a unique MAC that we can use as a datapath
1004 * ID. In some special cases, though, multiple bridges will end up with
1005 * the same MAC address. This is OK for the bridges, but it will confuse
1006 * the OpenFlow controller, because each datapath needs a unique datapath
1009 * Datapath IDs must be unique. It is also very desirable that they be
1010 * stable from one run to the next, so that policy set on a datapath
1013 const char *datapath_id;
1016 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1017 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1021 if (hw_addr_iface) {
1023 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1025 * A bridge whose MAC address is taken from a VLAN network device
1026 * (that is, a network device created with vconfig(8) or similar
1027 * tool) will have the same MAC address as a bridge on the VLAN
1028 * device's physical network device.
1030 * Handle this case by hashing the physical network device MAC
1031 * along with the VLAN identifier.
1033 uint8_t buf[ETH_ADDR_LEN + 2];
1034 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1035 buf[ETH_ADDR_LEN] = vlan >> 8;
1036 buf[ETH_ADDR_LEN + 1] = vlan;
1037 return dpid_from_hash(buf, sizeof buf);
1040 * Assume that this bridge's MAC address is unique, since it
1041 * doesn't fit any of the cases we handle specially.
1046 * A purely internal bridge, that is, one that has no non-virtual
1047 * network devices on it at all, is more difficult because it has no
1048 * natural unique identifier at all.
1050 * When the host is a XenServer, we handle this case by hashing the
1051 * host's UUID with the name of the bridge. Names of bridges are
1052 * persistent across XenServer reboots, although they can be reused if
1053 * an internal network is destroyed and then a new one is later
1054 * created, so this is fairly effective.
1056 * When the host is not a XenServer, we punt by using a random MAC
1057 * address on each run.
1059 const char *host_uuid = xenserver_get_host_uuid();
1061 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1062 dpid = dpid_from_hash(combined, strlen(combined));
1068 return eth_addr_to_uint64(bridge_ea);
1072 dpid_from_hash(const void *data, size_t n)
1074 uint8_t hash[SHA1_DIGEST_SIZE];
1076 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1077 sha1_bytes(data, n, hash);
1078 eth_addr_mark_random(hash);
1079 return eth_addr_to_uint64(hash);
1085 struct bridge *br, *next;
1089 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1090 int error = bridge_run_one(br);
1092 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1093 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1094 "forcing reconfiguration", br->name);
1108 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1109 ofproto_wait(br->ofproto);
1110 if (ofproto_has_controller(br->ofproto)) {
1114 mac_learning_wait(br->ml);
1119 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1120 * configuration changes. */
1122 bridge_flush(struct bridge *br)
1124 COVERAGE_INC(bridge_flush);
1126 mac_learning_flush(br->ml);
1129 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1130 * such interface. */
1131 static struct iface *
1132 bridge_get_local_iface(struct bridge *br)
1136 for (i = 0; i < br->n_ports; i++) {
1137 struct port *port = br->ports[i];
1138 for (j = 0; j < port->n_ifaces; j++) {
1139 struct iface *iface = port->ifaces[j];
1140 if (iface->dp_ifidx == ODPP_LOCAL) {
1149 /* Bridge unixctl user interface functions. */
1151 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1152 const char *args, void *aux OVS_UNUSED)
1154 struct ds ds = DS_EMPTY_INITIALIZER;
1155 const struct bridge *br;
1156 const struct mac_entry *e;
1158 br = bridge_lookup(args);
1160 unixctl_command_reply(conn, 501, "no such bridge");
1164 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1165 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1166 if (e->port < 0 || e->port >= br->n_ports) {
1169 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1170 br->ports[e->port]->ifaces[0]->dp_ifidx,
1171 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1173 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1177 /* Bridge reconfiguration functions. */
1178 static struct bridge *
1179 bridge_create(const struct ovsrec_bridge *br_cfg)
1184 assert(!bridge_lookup(br_cfg->name));
1185 br = xzalloc(sizeof *br);
1187 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1193 dpif_flow_flush(br->dpif);
1195 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1198 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1200 dpif_delete(br->dpif);
1201 dpif_close(br->dpif);
1206 br->name = xstrdup(br_cfg->name);
1208 br->ml = mac_learning_create();
1209 br->sent_config_request = false;
1210 eth_addr_nicira_random(br->default_ea);
1212 port_array_init(&br->ifaces);
1214 shash_init(&br->port_by_name);
1215 shash_init(&br->iface_by_name);
1219 list_push_back(&all_bridges, &br->node);
1221 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1227 bridge_destroy(struct bridge *br)
1232 while (br->n_ports > 0) {
1233 port_destroy(br->ports[br->n_ports - 1]);
1235 list_remove(&br->node);
1236 error = dpif_delete(br->dpif);
1237 if (error && error != ENOENT) {
1238 VLOG_ERR("failed to delete %s: %s",
1239 dpif_name(br->dpif), strerror(error));
1241 dpif_close(br->dpif);
1242 ofproto_destroy(br->ofproto);
1243 mac_learning_destroy(br->ml);
1244 port_array_destroy(&br->ifaces);
1245 shash_destroy(&br->port_by_name);
1246 shash_destroy(&br->iface_by_name);
1253 static struct bridge *
1254 bridge_lookup(const char *name)
1258 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1259 if (!strcmp(br->name, name)) {
1267 bridge_exists(const char *name)
1269 return bridge_lookup(name) ? true : false;
1273 bridge_get_datapathid(const char *name)
1275 struct bridge *br = bridge_lookup(name);
1276 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1279 /* Handle requests for a listing of all flows known by the OpenFlow
1280 * stack, including those normally hidden. */
1282 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1283 const char *args, void *aux OVS_UNUSED)
1288 br = bridge_lookup(args);
1290 unixctl_command_reply(conn, 501, "Unknown bridge");
1295 ofproto_get_all_flows(br->ofproto, &results);
1297 unixctl_command_reply(conn, 200, ds_cstr(&results));
1298 ds_destroy(&results);
1302 bridge_run_one(struct bridge *br)
1306 error = ofproto_run1(br->ofproto);
1311 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1314 error = ofproto_run2(br->ofproto, br->flush);
1321 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1322 const struct bridge *br,
1323 struct ovsrec_controller ***controllersp)
1325 struct ovsrec_controller **controllers;
1326 size_t n_controllers;
1328 if (br->cfg->n_controller) {
1329 controllers = br->cfg->controller;
1330 n_controllers = br->cfg->n_controller;
1332 controllers = ovs_cfg->controller;
1333 n_controllers = ovs_cfg->n_controller;
1336 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1342 *controllersp = controllers;
1344 return n_controllers;
1348 bridge_update_desc(struct bridge *br OVS_UNUSED)
1351 bool changed = false;
1354 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1355 if (desc != br->mfr_desc) {
1358 br->mfr_desc = xstrdup(desc);
1360 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1365 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1366 if (desc != br->hw_desc) {
1369 br->hw_desc = xstrdup(desc);
1371 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1376 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1377 if (desc != br->sw_desc) {
1380 br->sw_desc = xstrdup(desc);
1382 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1387 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1388 if (desc != br->serial_desc) {
1389 free(br->serial_desc);
1391 br->serial_desc = xstrdup(desc);
1393 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1398 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1399 if (desc != br->dp_desc) {
1402 br->dp_desc = xstrdup(desc);
1404 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1410 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1411 br->sw_desc, br->serial_desc, br->dp_desc);
1417 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1420 struct shash old_ports, new_ports;
1421 struct svec listeners, old_listeners;
1422 struct svec snoops, old_snoops;
1423 struct shash_node *node;
1426 /* Collect old ports. */
1427 shash_init(&old_ports);
1428 for (i = 0; i < br->n_ports; i++) {
1429 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1432 /* Collect new ports. */
1433 shash_init(&new_ports);
1434 for (i = 0; i < br->cfg->n_ports; i++) {
1435 const char *name = br->cfg->ports[i]->name;
1436 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1437 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1442 /* If we have a controller, then we need a local port. Complain if the
1443 * user didn't specify one.
1445 * XXX perhaps we should synthesize a port ourselves in this case. */
1446 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1447 char local_name[IF_NAMESIZE];
1450 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1451 local_name, sizeof local_name);
1452 if (!error && !shash_find(&new_ports, local_name)) {
1453 VLOG_WARN("bridge %s: controller specified but no local port "
1454 "(port named %s) defined",
1455 br->name, local_name);
1459 /* Get rid of deleted ports.
1460 * Get rid of deleted interfaces on ports that still exist. */
1461 SHASH_FOR_EACH (node, &old_ports) {
1462 struct port *port = node->data;
1463 const struct ovsrec_port *port_cfg;
1465 port_cfg = shash_find_data(&new_ports, node->name);
1469 port_del_ifaces(port, port_cfg);
1473 /* Create new ports.
1474 * Add new interfaces to existing ports.
1475 * Reconfigure existing ports. */
1476 SHASH_FOR_EACH (node, &new_ports) {
1477 struct port *port = shash_find_data(&old_ports, node->name);
1479 port = port_create(br, node->name);
1482 port_reconfigure(port, node->data);
1483 if (!port->n_ifaces) {
1484 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1485 br->name, port->name);
1489 shash_destroy(&old_ports);
1490 shash_destroy(&new_ports);
1492 /* Delete all flows if we're switching from connected to standalone or vice
1493 * versa. (XXX Should we delete all flows if we are switching from one
1494 * controller to another?) */
1497 /* Configure OpenFlow management listeners. */
1498 svec_init(&listeners);
1499 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1501 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1502 ovs_rundir, br->name));
1503 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1504 svec_clear(&listeners);
1506 svec_sort_unique(&listeners);
1508 svec_init(&old_listeners);
1509 ofproto_get_listeners(br->ofproto, &old_listeners);
1510 svec_sort_unique(&old_listeners);
1512 if (!svec_equal(&listeners, &old_listeners)) {
1513 ofproto_set_listeners(br->ofproto, &listeners);
1515 svec_destroy(&listeners);
1516 svec_destroy(&old_listeners);
1518 /* Configure OpenFlow controller connection snooping. */
1520 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1522 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1523 ovs_rundir, br->name));
1524 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1525 svec_clear(&snoops);
1527 svec_sort_unique(&snoops);
1529 svec_init(&old_snoops);
1530 ofproto_get_snoops(br->ofproto, &old_snoops);
1531 svec_sort_unique(&old_snoops);
1533 if (!svec_equal(&snoops, &old_snoops)) {
1534 ofproto_set_snoops(br->ofproto, &snoops);
1536 svec_destroy(&snoops);
1537 svec_destroy(&old_snoops);
1539 /* Default listener. */
1540 svec_init(&listeners);
1541 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1542 ovs_rundir, br->name));
1543 svec_init(&old_listeners);
1544 ofproto_get_listeners(br->ofproto, &old_listeners);
1545 if (!svec_equal(&listeners, &old_listeners)) {
1546 ofproto_set_listeners(br->ofproto, &listeners);
1548 svec_destroy(&listeners);
1549 svec_destroy(&old_listeners);
1551 /* Default snoop. */
1553 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1554 ovs_rundir, br->name));
1555 svec_init(&old_snoops);
1556 ofproto_get_snoops(br->ofproto, &old_snoops);
1557 if (!svec_equal(&snoops, &old_snoops)) {
1558 ofproto_set_snoops(br->ofproto, &snoops);
1560 svec_destroy(&snoops);
1561 svec_destroy(&old_snoops);
1564 mirror_reconfigure(br);
1566 bridge_update_desc(br);
1570 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1572 const struct sockaddr_in *managers,
1575 struct ovsrec_controller **controllers;
1576 size_t n_controllers;
1578 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1580 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1581 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1582 ofproto_flush_flows(br->ofproto);
1585 if (!n_controllers) {
1586 union ofp_action action;
1589 /* Clear out controllers. */
1590 ofproto_set_controllers(br->ofproto, NULL, 0);
1592 /* Set up a flow that matches every packet and directs them to
1593 * OFPP_NORMAL (which goes to us). */
1594 memset(&action, 0, sizeof action);
1595 action.type = htons(OFPAT_OUTPUT);
1596 action.output.len = htons(sizeof action);
1597 action.output.port = htons(OFPP_NORMAL);
1598 memset(&flow, 0, sizeof flow);
1599 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1601 struct ofproto_controller *ocs;
1604 ocs = xmalloc(n_controllers * sizeof *ocs);
1605 for (i = 0; i < n_controllers; i++) {
1606 struct ovsrec_controller *c = controllers[i];
1607 struct ofproto_controller *oc = &ocs[i];
1609 if (strcmp(c->target, "discover")) {
1610 struct iface *local_iface;
1613 local_iface = bridge_get_local_iface(br);
1614 if (local_iface && c->local_ip
1615 && inet_aton(c->local_ip, &ip)) {
1616 struct netdev *netdev = local_iface->netdev;
1617 struct in_addr mask, gateway;
1619 if (!c->local_netmask
1620 || !inet_aton(c->local_netmask, &mask)) {
1623 if (!c->local_gateway
1624 || !inet_aton(c->local_gateway, &gateway)) {
1628 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1630 mask.s_addr = guess_netmask(ip.s_addr);
1632 if (!netdev_set_in4(netdev, ip, mask)) {
1633 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1635 br->name, IP_ARGS(&ip.s_addr),
1636 IP_ARGS(&mask.s_addr));
1639 if (gateway.s_addr) {
1640 if (!netdev_add_router(netdev, gateway)) {
1641 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1642 br->name, IP_ARGS(&gateway.s_addr));
1648 oc->target = c->target;
1649 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1650 oc->probe_interval = (c->inactivity_probe
1651 ? *c->inactivity_probe / 1000 : 5);
1652 oc->fail = (!c->fail_mode
1653 || !strcmp(c->fail_mode, "standalone")
1654 || !strcmp(c->fail_mode, "open")
1655 ? OFPROTO_FAIL_STANDALONE
1656 : OFPROTO_FAIL_SECURE);
1657 oc->band = (!c->connection_mode
1658 || !strcmp(c->connection_mode, "in-band")
1660 : OFPROTO_OUT_OF_BAND);
1661 oc->accept_re = c->discover_accept_regex;
1662 oc->update_resolv_conf = c->discover_update_resolv_conf;
1663 oc->rate_limit = (c->controller_rate_limit
1664 ? *c->controller_rate_limit : 0);
1665 oc->burst_limit = (c->controller_burst_limit
1666 ? *c->controller_burst_limit : 0);
1668 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1674 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1679 for (i = 0; i < br->n_ports; i++) {
1680 struct port *port = br->ports[i];
1681 for (j = 0; j < port->n_ifaces; j++) {
1682 struct iface *iface = port->ifaces[j];
1683 shash_add_once(ifaces, iface->name, iface);
1685 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1686 shash_add_once(ifaces, port->name, NULL);
1691 /* For robustness, in case the administrator moves around datapath ports behind
1692 * our back, we re-check all the datapath port numbers here.
1694 * This function will set the 'dp_ifidx' members of interfaces that have
1695 * disappeared to -1, so only call this function from a context where those
1696 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1697 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1698 * datapath, which doesn't support UINT16_MAX+1 ports. */
1700 bridge_fetch_dp_ifaces(struct bridge *br)
1702 struct odp_port *dpif_ports;
1703 size_t n_dpif_ports;
1706 /* Reset all interface numbers. */
1707 for (i = 0; i < br->n_ports; i++) {
1708 struct port *port = br->ports[i];
1709 for (j = 0; j < port->n_ifaces; j++) {
1710 struct iface *iface = port->ifaces[j];
1711 iface->dp_ifidx = -1;
1714 port_array_clear(&br->ifaces);
1716 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1717 for (i = 0; i < n_dpif_ports; i++) {
1718 struct odp_port *p = &dpif_ports[i];
1719 struct iface *iface = iface_lookup(br, p->devname);
1721 if (iface->dp_ifidx >= 0) {
1722 VLOG_WARN("%s reported interface %s twice",
1723 dpif_name(br->dpif), p->devname);
1724 } else if (iface_from_dp_ifidx(br, p->port)) {
1725 VLOG_WARN("%s reported interface %"PRIu16" twice",
1726 dpif_name(br->dpif), p->port);
1728 port_array_set(&br->ifaces, p->port, iface);
1729 iface->dp_ifidx = p->port;
1733 int64_t ofport = (iface->dp_ifidx >= 0
1734 ? odp_port_to_ofp_port(iface->dp_ifidx)
1736 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1743 /* Bridge packet processing functions. */
1746 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1748 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1751 static struct bond_entry *
1752 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1754 return &port->bond_hash[bond_hash(mac)];
1758 bond_choose_iface(const struct port *port)
1760 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1761 size_t i, best_down_slave = -1;
1762 long long next_delay_expiration = LLONG_MAX;
1764 for (i = 0; i < port->n_ifaces; i++) {
1765 struct iface *iface = port->ifaces[i];
1767 if (iface->enabled) {
1769 } else if (iface->delay_expires < next_delay_expiration) {
1770 best_down_slave = i;
1771 next_delay_expiration = iface->delay_expires;
1775 if (best_down_slave != -1) {
1776 struct iface *iface = port->ifaces[best_down_slave];
1778 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1779 "since no other interface is up", iface->name,
1780 iface->delay_expires - time_msec());
1781 bond_enable_slave(iface, true);
1784 return best_down_slave;
1788 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1789 uint16_t *dp_ifidx, tag_type *tags)
1791 struct iface *iface;
1793 assert(port->n_ifaces);
1794 if (port->n_ifaces == 1) {
1795 iface = port->ifaces[0];
1797 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1798 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1799 || !port->ifaces[e->iface_idx]->enabled) {
1800 /* XXX select interface properly. The current interface selection
1801 * is only good for testing the rebalancing code. */
1802 e->iface_idx = bond_choose_iface(port);
1803 if (e->iface_idx < 0) {
1804 *tags |= port->no_ifaces_tag;
1807 e->iface_tag = tag_create_random();
1808 ((struct port *) port)->bond_compat_is_stale = true;
1810 *tags |= e->iface_tag;
1811 iface = port->ifaces[e->iface_idx];
1813 *dp_ifidx = iface->dp_ifidx;
1814 *tags |= iface->tag; /* Currently only used for bonding. */
1819 bond_link_status_update(struct iface *iface, bool carrier)
1821 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1822 struct port *port = iface->port;
1824 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1825 /* Nothing to do. */
1828 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1829 iface->name, carrier ? "detected" : "dropped");
1830 if (carrier == iface->enabled) {
1831 iface->delay_expires = LLONG_MAX;
1832 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1833 iface->name, carrier ? "disabled" : "enabled");
1834 } else if (carrier && port->active_iface < 0) {
1835 bond_enable_slave(iface, true);
1836 if (port->updelay) {
1837 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1838 "other interface is up", iface->name, port->updelay);
1841 int delay = carrier ? port->updelay : port->downdelay;
1842 iface->delay_expires = time_msec() + delay;
1845 "interface %s: will be %s if it stays %s for %d ms",
1847 carrier ? "enabled" : "disabled",
1848 carrier ? "up" : "down",
1855 bond_choose_active_iface(struct port *port)
1857 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1859 port->active_iface = bond_choose_iface(port);
1860 port->active_iface_tag = tag_create_random();
1861 if (port->active_iface >= 0) {
1862 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1863 port->name, port->ifaces[port->active_iface]->name);
1865 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1871 bond_enable_slave(struct iface *iface, bool enable)
1873 struct port *port = iface->port;
1874 struct bridge *br = port->bridge;
1876 /* This acts as a recursion check. If the act of disabling a slave
1877 * causes a different slave to be enabled, the flag will allow us to
1878 * skip redundant work when we reenter this function. It must be
1879 * cleared on exit to keep things safe with multiple bonds. */
1880 static bool moving_active_iface = false;
1882 iface->delay_expires = LLONG_MAX;
1883 if (enable == iface->enabled) {
1887 iface->enabled = enable;
1888 if (!iface->enabled) {
1889 VLOG_WARN("interface %s: disabled", iface->name);
1890 ofproto_revalidate(br->ofproto, iface->tag);
1891 if (iface->port_ifidx == port->active_iface) {
1892 ofproto_revalidate(br->ofproto,
1893 port->active_iface_tag);
1895 /* Disabling a slave can lead to another slave being immediately
1896 * enabled if there will be no active slaves but one is waiting
1897 * on an updelay. In this case we do not need to run most of the
1898 * code for the newly enabled slave since there was no period
1899 * without an active slave and it is redundant with the disabling
1901 moving_active_iface = true;
1902 bond_choose_active_iface(port);
1904 bond_send_learning_packets(port);
1906 VLOG_WARN("interface %s: enabled", iface->name);
1907 if (port->active_iface < 0 && !moving_active_iface) {
1908 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1909 bond_choose_active_iface(port);
1910 bond_send_learning_packets(port);
1912 iface->tag = tag_create_random();
1915 moving_active_iface = false;
1916 port->bond_compat_is_stale = true;
1919 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1920 * bond interface. */
1922 bond_update_fake_iface_stats(struct port *port)
1924 struct netdev_stats bond_stats;
1925 struct netdev *bond_dev;
1928 memset(&bond_stats, 0, sizeof bond_stats);
1930 for (i = 0; i < port->n_ifaces; i++) {
1931 struct netdev_stats slave_stats;
1933 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1934 bond_stats.rx_packets += slave_stats.rx_packets;
1935 bond_stats.rx_bytes += slave_stats.rx_bytes;
1936 bond_stats.tx_packets += slave_stats.tx_packets;
1937 bond_stats.tx_bytes += slave_stats.tx_bytes;
1941 if (!netdev_open_default(port->name, &bond_dev)) {
1942 netdev_set_stats(bond_dev, &bond_stats);
1943 netdev_close(bond_dev);
1948 bond_run(struct bridge *br)
1952 for (i = 0; i < br->n_ports; i++) {
1953 struct port *port = br->ports[i];
1955 if (port->n_ifaces >= 2) {
1956 for (j = 0; j < port->n_ifaces; j++) {
1957 struct iface *iface = port->ifaces[j];
1958 if (time_msec() >= iface->delay_expires) {
1959 bond_enable_slave(iface, !iface->enabled);
1963 if (port->bond_fake_iface
1964 && time_msec() >= port->bond_next_fake_iface_update) {
1965 bond_update_fake_iface_stats(port);
1966 port->bond_next_fake_iface_update = time_msec() + 1000;
1970 if (port->bond_compat_is_stale) {
1971 port->bond_compat_is_stale = false;
1972 port_update_bond_compat(port);
1978 bond_wait(struct bridge *br)
1982 for (i = 0; i < br->n_ports; i++) {
1983 struct port *port = br->ports[i];
1984 if (port->n_ifaces < 2) {
1987 for (j = 0; j < port->n_ifaces; j++) {
1988 struct iface *iface = port->ifaces[j];
1989 if (iface->delay_expires != LLONG_MAX) {
1990 poll_timer_wait_until(iface->delay_expires);
1993 if (port->bond_fake_iface) {
1994 poll_timer_wait_until(port->bond_next_fake_iface_update);
2000 set_dst(struct dst *p, const flow_t *flow,
2001 const struct port *in_port, const struct port *out_port,
2004 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2005 : in_port->vlan >= 0 ? in_port->vlan
2006 : ntohs(flow->dl_vlan));
2007 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2011 swap_dst(struct dst *p, struct dst *q)
2013 struct dst tmp = *p;
2018 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2019 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2020 * that we push to the datapath. We could in fact fully sort the array by
2021 * vlan, but in most cases there are at most two different vlan tags so that's
2022 * possibly overkill.) */
2024 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2026 struct dst *first = dsts;
2027 struct dst *last = dsts + n_dsts;
2029 while (first != last) {
2031 * - All dsts < first have vlan == 'vlan'.
2032 * - All dsts >= last have vlan != 'vlan'.
2033 * - first < last. */
2034 while (first->vlan == vlan) {
2035 if (++first == last) {
2040 /* Same invariants, plus one additional:
2041 * - first->vlan != vlan.
2043 while (last[-1].vlan != vlan) {
2044 if (--last == first) {
2049 /* Same invariants, plus one additional:
2050 * - last[-1].vlan == vlan.*/
2051 swap_dst(first++, --last);
2056 mirror_mask_ffs(mirror_mask_t mask)
2058 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2063 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2064 const struct dst *test)
2067 for (i = 0; i < n_dsts; i++) {
2068 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2076 port_trunks_vlan(const struct port *port, uint16_t vlan)
2078 return (port->vlan < 0
2079 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2083 port_includes_vlan(const struct port *port, uint16_t vlan)
2085 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2089 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2090 const struct port *in_port, const struct port *out_port,
2091 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2093 mirror_mask_t mirrors = in_port->src_mirrors;
2094 struct dst *dst = dsts;
2097 if (out_port == FLOOD_PORT) {
2098 /* XXX use ODP_FLOOD if no vlans or bonding. */
2099 /* XXX even better, define each VLAN as a datapath port group */
2100 for (i = 0; i < br->n_ports; i++) {
2101 struct port *port = br->ports[i];
2102 if (port != in_port && port_includes_vlan(port, vlan)
2103 && !port->is_mirror_output_port
2104 && set_dst(dst, flow, in_port, port, tags)) {
2105 mirrors |= port->dst_mirrors;
2109 *nf_output_iface = NF_OUT_FLOOD;
2110 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2111 *nf_output_iface = dst->dp_ifidx;
2112 mirrors |= out_port->dst_mirrors;
2117 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2118 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2120 if (set_dst(dst, flow, in_port, m->out_port, tags)
2121 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2125 for (i = 0; i < br->n_ports; i++) {
2126 struct port *port = br->ports[i];
2127 if (port_includes_vlan(port, m->out_vlan)
2128 && set_dst(dst, flow, in_port, port, tags))
2132 if (port->vlan < 0) {
2133 dst->vlan = m->out_vlan;
2135 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2139 /* Use the vlan tag on the original flow instead of
2140 * the one passed in the vlan parameter. This ensures
2141 * that we compare the vlan from before any implicit
2142 * tagging tags place. This is necessary because
2143 * dst->vlan is the final vlan, after removing implicit
2145 flow_vlan = ntohs(flow->dl_vlan);
2146 if (flow_vlan == 0) {
2147 flow_vlan = OFP_VLAN_NONE;
2149 if (port == in_port && dst->vlan == flow_vlan) {
2150 /* Don't send out input port on same VLAN. */
2158 mirrors &= mirrors - 1;
2161 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2165 static void OVS_UNUSED
2166 print_dsts(const struct dst *dsts, size_t n)
2168 for (; n--; dsts++) {
2169 printf(">p%"PRIu16, dsts->dp_ifidx);
2170 if (dsts->vlan != OFP_VLAN_NONE) {
2171 printf("v%"PRIu16, dsts->vlan);
2177 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2178 const struct port *in_port, const struct port *out_port,
2179 tag_type *tags, struct odp_actions *actions,
2180 uint16_t *nf_output_iface)
2182 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2184 const struct dst *p;
2187 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2190 cur_vlan = ntohs(flow->dl_vlan);
2191 for (p = dsts; p < &dsts[n_dsts]; p++) {
2192 union odp_action *a;
2193 if (p->vlan != cur_vlan) {
2194 if (p->vlan == OFP_VLAN_NONE) {
2195 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2197 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2198 a->vlan_vid.vlan_vid = htons(p->vlan);
2202 a = odp_actions_add(actions, ODPAT_OUTPUT);
2203 a->output.port = p->dp_ifidx;
2207 /* Returns the effective vlan of a packet, taking into account both the
2208 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2209 * the packet is untagged and -1 indicates it has an invalid header and
2210 * should be dropped. */
2211 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2212 struct port *in_port, bool have_packet)
2214 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2215 * belongs to VLAN 0, so we should treat both cases identically. (In the
2216 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2217 * presumably to allow a priority to be specified. In the latter case, the
2218 * packet does not have any 802.1Q header.) */
2219 int vlan = ntohs(flow->dl_vlan);
2220 if (vlan == OFP_VLAN_NONE) {
2223 if (in_port->vlan >= 0) {
2225 /* XXX support double tagging? */
2227 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2228 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2229 "packet received on port %s configured with "
2230 "implicit VLAN %"PRIu16,
2231 br->name, ntohs(flow->dl_vlan),
2232 in_port->name, in_port->vlan);
2236 vlan = in_port->vlan;
2238 if (!port_includes_vlan(in_port, vlan)) {
2240 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2241 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2242 "packet received on port %s not configured for "
2244 br->name, vlan, in_port->name, vlan);
2254 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2255 struct port *in_port)
2257 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2258 vlan, in_port->port_idx);
2260 /* The log messages here could actually be useful in debugging,
2261 * so keep the rate limit relatively high. */
2262 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2264 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2265 "on port %s in VLAN %d",
2266 br->name, ETH_ADDR_ARGS(flow->dl_src),
2267 in_port->name, vlan);
2268 ofproto_revalidate(br->ofproto, rev_tag);
2272 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2273 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2274 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2276 is_gratuitous_arp(const flow_t *flow)
2278 return (flow->dl_type == htons(ETH_TYPE_ARP)
2279 && eth_addr_is_broadcast(flow->dl_dst)
2280 && (flow->nw_proto == ARP_OP_REPLY
2281 || (flow->nw_proto == ARP_OP_REQUEST
2282 && flow->nw_src == flow->nw_dst)));
2285 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2286 * dropped. Returns true if they may be forwarded, false if they should be
2289 * If 'have_packet' is true, it indicates that the caller is processing a
2290 * received packet. If 'have_packet' is false, then the caller is just
2291 * revalidating an existing flow because configuration has changed. Either
2292 * way, 'have_packet' only affects logging (there is no point in logging errors
2293 * during revalidation).
2295 * Sets '*in_portp' to the input port. This will be a null pointer if
2296 * flow->in_port does not designate a known input port (in which case
2297 * is_admissible() returns false).
2299 * When returning true, sets '*vlanp' to the effective VLAN of the input
2300 * packet, as returned by flow_get_vlan().
2302 * May also add tags to '*tags', although the current implementation only does
2303 * so in one special case.
2306 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2307 tag_type *tags, int *vlanp, struct port **in_portp)
2309 struct iface *in_iface;
2310 struct port *in_port;
2313 /* Find the interface and port structure for the received packet. */
2314 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2316 /* No interface? Something fishy... */
2318 /* Odd. A few possible reasons here:
2320 * - We deleted an interface but there are still a few packets
2321 * queued up from it.
2323 * - Someone externally added an interface (e.g. with "ovs-dpctl
2324 * add-if") that we don't know about.
2326 * - Packet arrived on the local port but the local port is not
2327 * one of our bridge ports.
2329 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2331 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2332 "interface %"PRIu16, br->name, flow->in_port);
2338 *in_portp = in_port = in_iface->port;
2339 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2344 /* Drop frames for reserved multicast addresses. */
2345 if (eth_addr_is_reserved(flow->dl_dst)) {
2349 /* Drop frames on ports reserved for mirroring. */
2350 if (in_port->is_mirror_output_port) {
2352 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2353 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2354 "%s, which is reserved exclusively for mirroring",
2355 br->name, in_port->name);
2360 /* Packets received on bonds need special attention to avoid duplicates. */
2361 if (in_port->n_ifaces > 1) {
2364 if (eth_addr_is_multicast(flow->dl_dst)) {
2365 *tags |= in_port->active_iface_tag;
2366 if (in_port->active_iface != in_iface->port_ifidx) {
2367 /* Drop all multicast packets on inactive slaves. */
2372 /* Drop all packets for which we have learned a different input
2373 * port, because we probably sent the packet on one slave and got
2374 * it back on the other. Gratuitous ARP packets are an exception
2375 * to this rule: the host has moved to another switch. */
2376 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2377 if (src_idx != -1 && src_idx != in_port->port_idx &&
2378 !is_gratuitous_arp(flow)) {
2386 /* If the composed actions may be applied to any packet in the given 'flow',
2387 * returns true. Otherwise, the actions should only be applied to 'packet', or
2388 * not at all, if 'packet' was NULL. */
2390 process_flow(struct bridge *br, const flow_t *flow,
2391 const struct ofpbuf *packet, struct odp_actions *actions,
2392 tag_type *tags, uint16_t *nf_output_iface)
2394 struct port *in_port;
2395 struct port *out_port;
2399 /* Check whether we should drop packets in this flow. */
2400 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2405 /* Learn source MAC (but don't try to learn from revalidation). */
2407 update_learning_table(br, flow, vlan, in_port);
2410 /* Determine output port. */
2411 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags);
2412 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2413 out_port = br->ports[out_port_idx];
2414 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2415 /* If we are revalidating but don't have a learning entry then
2416 * eject the flow. Installing a flow that floods packets opens
2417 * up a window of time where we could learn from a packet reflected
2418 * on a bond and blackhole packets before the learning table is
2419 * updated to reflect the correct port. */
2422 out_port = FLOOD_PORT;
2425 /* Don't send packets out their input ports. */
2426 if (in_port == out_port) {
2432 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2439 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2442 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2443 const struct ofp_phy_port *opp,
2446 struct bridge *br = br_;
2447 struct iface *iface;
2450 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2456 if (reason == OFPPR_DELETE) {
2457 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2458 br->name, iface->name);
2459 iface_destroy(iface);
2460 if (!port->n_ifaces) {
2461 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2462 br->name, port->name);
2468 if (port->n_ifaces > 1) {
2469 bool up = !(opp->state & OFPPS_LINK_DOWN);
2470 bond_link_status_update(iface, up);
2471 port_update_bond_compat(port);
2477 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2478 struct odp_actions *actions, tag_type *tags,
2479 uint16_t *nf_output_iface, void *br_)
2481 struct bridge *br = br_;
2483 COVERAGE_INC(bridge_process_flow);
2484 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2488 bridge_account_flow_ofhook_cb(const flow_t *flow,
2489 const union odp_action *actions,
2490 size_t n_actions, unsigned long long int n_bytes,
2493 struct bridge *br = br_;
2494 const union odp_action *a;
2495 struct port *in_port;
2499 /* Feed information from the active flows back into the learning table
2500 * to ensure that table is always in sync with what is actually flowing
2501 * through the datapath. */
2502 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2503 update_learning_table(br, flow, vlan, in_port);
2506 if (!br->has_bonded_ports) {
2510 for (a = actions; a < &actions[n_actions]; a++) {
2511 if (a->type == ODPAT_OUTPUT) {
2512 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2513 if (out_port && out_port->n_ifaces >= 2) {
2514 struct bond_entry *e = lookup_bond_entry(out_port,
2516 e->tx_bytes += n_bytes;
2523 bridge_account_checkpoint_ofhook_cb(void *br_)
2525 struct bridge *br = br_;
2529 if (!br->has_bonded_ports) {
2534 for (i = 0; i < br->n_ports; i++) {
2535 struct port *port = br->ports[i];
2536 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2537 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2538 bond_rebalance_port(port);
2543 static struct ofhooks bridge_ofhooks = {
2544 bridge_port_changed_ofhook_cb,
2545 bridge_normal_ofhook_cb,
2546 bridge_account_flow_ofhook_cb,
2547 bridge_account_checkpoint_ofhook_cb,
2550 /* Bonding functions. */
2552 /* Statistics for a single interface on a bonded port, used for load-based
2553 * bond rebalancing. */
2554 struct slave_balance {
2555 struct iface *iface; /* The interface. */
2556 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2558 /* All the "bond_entry"s that are assigned to this interface, in order of
2559 * increasing tx_bytes. */
2560 struct bond_entry **hashes;
2564 /* Sorts pointers to pointers to bond_entries in ascending order by the
2565 * interface to which they are assigned, and within a single interface in
2566 * ascending order of bytes transmitted. */
2568 compare_bond_entries(const void *a_, const void *b_)
2570 const struct bond_entry *const *ap = a_;
2571 const struct bond_entry *const *bp = b_;
2572 const struct bond_entry *a = *ap;
2573 const struct bond_entry *b = *bp;
2574 if (a->iface_idx != b->iface_idx) {
2575 return a->iface_idx > b->iface_idx ? 1 : -1;
2576 } else if (a->tx_bytes != b->tx_bytes) {
2577 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2583 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2584 * *descending* order by number of bytes transmitted. */
2586 compare_slave_balance(const void *a_, const void *b_)
2588 const struct slave_balance *a = a_;
2589 const struct slave_balance *b = b_;
2590 if (a->iface->enabled != b->iface->enabled) {
2591 return a->iface->enabled ? -1 : 1;
2592 } else if (a->tx_bytes != b->tx_bytes) {
2593 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2600 swap_bals(struct slave_balance *a, struct slave_balance *b)
2602 struct slave_balance tmp = *a;
2607 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2608 * given that 'p' (and only 'p') might be in the wrong location.
2610 * This function invalidates 'p', since it might now be in a different memory
2613 resort_bals(struct slave_balance *p,
2614 struct slave_balance bals[], size_t n_bals)
2617 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2618 swap_bals(p, p - 1);
2620 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2621 swap_bals(p, p + 1);
2627 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2629 if (VLOG_IS_DBG_ENABLED()) {
2630 struct ds ds = DS_EMPTY_INITIALIZER;
2631 const struct slave_balance *b;
2633 for (b = bals; b < bals + n_bals; b++) {
2637 ds_put_char(&ds, ',');
2639 ds_put_format(&ds, " %s %"PRIu64"kB",
2640 b->iface->name, b->tx_bytes / 1024);
2642 if (!b->iface->enabled) {
2643 ds_put_cstr(&ds, " (disabled)");
2645 if (b->n_hashes > 0) {
2646 ds_put_cstr(&ds, " (");
2647 for (i = 0; i < b->n_hashes; i++) {
2648 const struct bond_entry *e = b->hashes[i];
2650 ds_put_cstr(&ds, " + ");
2652 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2653 e - port->bond_hash, e->tx_bytes / 1024);
2655 ds_put_cstr(&ds, ")");
2658 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2663 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2665 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2668 struct bond_entry *hash = from->hashes[hash_idx];
2669 struct port *port = from->iface->port;
2670 uint64_t delta = hash->tx_bytes;
2672 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2673 "from %s to %s (now carrying %"PRIu64"kB and "
2674 "%"PRIu64"kB load, respectively)",
2675 port->name, delta / 1024, hash - port->bond_hash,
2676 from->iface->name, to->iface->name,
2677 (from->tx_bytes - delta) / 1024,
2678 (to->tx_bytes + delta) / 1024);
2680 /* Delete element from from->hashes.
2682 * We don't bother to add the element to to->hashes because not only would
2683 * it require more work, the only purpose it would be to allow that hash to
2684 * be migrated to another slave in this rebalancing run, and there is no
2685 * point in doing that. */
2686 if (hash_idx == 0) {
2689 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2690 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2694 /* Shift load away from 'from' to 'to'. */
2695 from->tx_bytes -= delta;
2696 to->tx_bytes += delta;
2698 /* Arrange for flows to be revalidated. */
2699 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2700 hash->iface_idx = to->iface->port_ifidx;
2701 hash->iface_tag = tag_create_random();
2705 bond_rebalance_port(struct port *port)
2707 struct slave_balance bals[DP_MAX_PORTS];
2709 struct bond_entry *hashes[BOND_MASK + 1];
2710 struct slave_balance *b, *from, *to;
2711 struct bond_entry *e;
2714 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2715 * descending order of tx_bytes, so that bals[0] represents the most
2716 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2719 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2720 * array for each slave_balance structure, we sort our local array of
2721 * hashes in order by slave, so that all of the hashes for a given slave
2722 * become contiguous in memory, and then we point each 'hashes' members of
2723 * a slave_balance structure to the start of a contiguous group. */
2724 n_bals = port->n_ifaces;
2725 for (b = bals; b < &bals[n_bals]; b++) {
2726 b->iface = port->ifaces[b - bals];
2731 for (i = 0; i <= BOND_MASK; i++) {
2732 hashes[i] = &port->bond_hash[i];
2734 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2735 for (i = 0; i <= BOND_MASK; i++) {
2737 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2738 b = &bals[e->iface_idx];
2739 b->tx_bytes += e->tx_bytes;
2741 b->hashes = &hashes[i];
2746 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2747 log_bals(bals, n_bals, port);
2749 /* Discard slaves that aren't enabled (which were sorted to the back of the
2750 * array earlier). */
2751 while (!bals[n_bals - 1].iface->enabled) {
2758 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2759 to = &bals[n_bals - 1];
2760 for (from = bals; from < to; ) {
2761 uint64_t overload = from->tx_bytes - to->tx_bytes;
2762 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2763 /* The extra load on 'from' (and all less-loaded slaves), compared
2764 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2765 * it is less than ~1Mbps. No point in rebalancing. */
2767 } else if (from->n_hashes == 1) {
2768 /* 'from' only carries a single MAC hash, so we can't shift any
2769 * load away from it, even though we want to. */
2772 /* 'from' is carrying significantly more load than 'to', and that
2773 * load is split across at least two different hashes. Pick a hash
2774 * to migrate to 'to' (the least-loaded slave), given that doing so
2775 * must decrease the ratio of the load on the two slaves by at
2778 * The sort order we use means that we prefer to shift away the
2779 * smallest hashes instead of the biggest ones. There is little
2780 * reason behind this decision; we could use the opposite sort
2781 * order to shift away big hashes ahead of small ones. */
2785 for (i = 0; i < from->n_hashes; i++) {
2786 double old_ratio, new_ratio;
2787 uint64_t delta = from->hashes[i]->tx_bytes;
2789 if (delta == 0 || from->tx_bytes - delta == 0) {
2790 /* Pointless move. */
2794 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2796 if (to->tx_bytes == 0) {
2797 /* Nothing on the new slave, move it. */
2801 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2802 new_ratio = (double)(from->tx_bytes - delta) /
2803 (to->tx_bytes + delta);
2805 if (new_ratio == 0) {
2806 /* Should already be covered but check to prevent division
2811 if (new_ratio < 1) {
2812 new_ratio = 1 / new_ratio;
2815 if (old_ratio - new_ratio > 0.1) {
2816 /* Would decrease the ratio, move it. */
2820 if (i < from->n_hashes) {
2821 bond_shift_load(from, to, i);
2822 port->bond_compat_is_stale = true;
2824 /* If the result of the migration changed the relative order of
2825 * 'from' and 'to' swap them back to maintain invariants. */
2826 if (order_swapped) {
2827 swap_bals(from, to);
2830 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2831 * point to different slave_balance structures. It is only
2832 * valid to do these two operations in a row at all because we
2833 * know that 'from' will not move past 'to' and vice versa. */
2834 resort_bals(from, bals, n_bals);
2835 resort_bals(to, bals, n_bals);
2842 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2843 * historical data to decay to <1% in 7 rebalancing runs. */
2844 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2850 bond_send_learning_packets(struct port *port)
2852 struct bridge *br = port->bridge;
2853 struct mac_entry *e;
2854 struct ofpbuf packet;
2855 int error, n_packets, n_errors;
2857 if (!port->n_ifaces || port->active_iface < 0) {
2861 ofpbuf_init(&packet, 128);
2862 error = n_packets = n_errors = 0;
2863 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2864 union ofp_action actions[2], *a;
2870 if (e->port == port->port_idx
2871 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2875 /* Compose actions. */
2876 memset(actions, 0, sizeof actions);
2879 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2880 a->vlan_vid.len = htons(sizeof *a);
2881 a->vlan_vid.vlan_vid = htons(e->vlan);
2884 a->output.type = htons(OFPAT_OUTPUT);
2885 a->output.len = htons(sizeof *a);
2886 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2891 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2893 flow_extract(&packet, 0, ODPP_NONE, &flow);
2894 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2901 ofpbuf_uninit(&packet);
2904 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2905 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2906 "packets, last error was: %s",
2907 port->name, n_errors, n_packets, strerror(error));
2909 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2910 port->name, n_packets);
2914 /* Bonding unixctl user interface functions. */
2917 bond_unixctl_list(struct unixctl_conn *conn,
2918 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2920 struct ds ds = DS_EMPTY_INITIALIZER;
2921 const struct bridge *br;
2923 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2925 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2928 for (i = 0; i < br->n_ports; i++) {
2929 const struct port *port = br->ports[i];
2930 if (port->n_ifaces > 1) {
2933 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2934 for (j = 0; j < port->n_ifaces; j++) {
2935 const struct iface *iface = port->ifaces[j];
2937 ds_put_cstr(&ds, ", ");
2939 ds_put_cstr(&ds, iface->name);
2941 ds_put_char(&ds, '\n');
2945 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2949 static struct port *
2950 bond_find(const char *name)
2952 const struct bridge *br;
2954 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2957 for (i = 0; i < br->n_ports; i++) {
2958 struct port *port = br->ports[i];
2959 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2968 bond_unixctl_show(struct unixctl_conn *conn,
2969 const char *args, void *aux OVS_UNUSED)
2971 struct ds ds = DS_EMPTY_INITIALIZER;
2972 const struct port *port;
2975 port = bond_find(args);
2977 unixctl_command_reply(conn, 501, "no such bond");
2981 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2982 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2983 ds_put_format(&ds, "next rebalance: %lld ms\n",
2984 port->bond_next_rebalance - time_msec());
2985 for (j = 0; j < port->n_ifaces; j++) {
2986 const struct iface *iface = port->ifaces[j];
2987 struct bond_entry *be;
2990 ds_put_format(&ds, "slave %s: %s\n",
2991 iface->name, iface->enabled ? "enabled" : "disabled");
2992 if (j == port->active_iface) {
2993 ds_put_cstr(&ds, "\tactive slave\n");
2995 if (iface->delay_expires != LLONG_MAX) {
2996 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2997 iface->enabled ? "downdelay" : "updelay",
2998 iface->delay_expires - time_msec());
3002 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3003 int hash = be - port->bond_hash;
3004 struct mac_entry *me;
3006 if (be->iface_idx != j) {
3010 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3011 hash, be->tx_bytes / 1024);
3014 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3015 &port->bridge->ml->lrus) {
3018 if (bond_hash(me->mac) == hash
3019 && me->port != port->port_idx
3020 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3021 && dp_ifidx == iface->dp_ifidx)
3023 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3024 ETH_ADDR_ARGS(me->mac));
3029 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3034 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3035 void *aux OVS_UNUSED)
3037 char *args = (char *) args_;
3038 char *save_ptr = NULL;
3039 char *bond_s, *hash_s, *slave_s;
3040 uint8_t mac[ETH_ADDR_LEN];
3042 struct iface *iface;
3043 struct bond_entry *entry;
3046 bond_s = strtok_r(args, " ", &save_ptr);
3047 hash_s = strtok_r(NULL, " ", &save_ptr);
3048 slave_s = strtok_r(NULL, " ", &save_ptr);
3050 unixctl_command_reply(conn, 501,
3051 "usage: bond/migrate BOND HASH SLAVE");
3055 port = bond_find(bond_s);
3057 unixctl_command_reply(conn, 501, "no such bond");
3061 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3062 == ETH_ADDR_SCAN_COUNT) {
3063 hash = bond_hash(mac);
3064 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3065 hash = atoi(hash_s) & BOND_MASK;
3067 unixctl_command_reply(conn, 501, "bad hash");
3071 iface = port_lookup_iface(port, slave_s);
3073 unixctl_command_reply(conn, 501, "no such slave");
3077 if (!iface->enabled) {
3078 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3082 entry = &port->bond_hash[hash];
3083 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3084 entry->iface_idx = iface->port_ifidx;
3085 entry->iface_tag = tag_create_random();
3086 port->bond_compat_is_stale = true;
3087 unixctl_command_reply(conn, 200, "migrated");
3091 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3092 void *aux OVS_UNUSED)
3094 char *args = (char *) args_;
3095 char *save_ptr = NULL;
3096 char *bond_s, *slave_s;
3098 struct iface *iface;
3100 bond_s = strtok_r(args, " ", &save_ptr);
3101 slave_s = strtok_r(NULL, " ", &save_ptr);
3103 unixctl_command_reply(conn, 501,
3104 "usage: bond/set-active-slave BOND SLAVE");
3108 port = bond_find(bond_s);
3110 unixctl_command_reply(conn, 501, "no such bond");
3114 iface = port_lookup_iface(port, slave_s);
3116 unixctl_command_reply(conn, 501, "no such slave");
3120 if (!iface->enabled) {
3121 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3125 if (port->active_iface != iface->port_ifidx) {
3126 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3127 port->active_iface = iface->port_ifidx;
3128 port->active_iface_tag = tag_create_random();
3129 VLOG_INFO("port %s: active interface is now %s",
3130 port->name, iface->name);
3131 bond_send_learning_packets(port);
3132 unixctl_command_reply(conn, 200, "done");
3134 unixctl_command_reply(conn, 200, "no change");
3139 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3141 char *args = (char *) args_;
3142 char *save_ptr = NULL;
3143 char *bond_s, *slave_s;
3145 struct iface *iface;
3147 bond_s = strtok_r(args, " ", &save_ptr);
3148 slave_s = strtok_r(NULL, " ", &save_ptr);
3150 unixctl_command_reply(conn, 501,
3151 "usage: bond/enable/disable-slave BOND SLAVE");
3155 port = bond_find(bond_s);
3157 unixctl_command_reply(conn, 501, "no such bond");
3161 iface = port_lookup_iface(port, slave_s);
3163 unixctl_command_reply(conn, 501, "no such slave");
3167 bond_enable_slave(iface, enable);
3168 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3172 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3173 void *aux OVS_UNUSED)
3175 enable_slave(conn, args, true);
3179 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3180 void *aux OVS_UNUSED)
3182 enable_slave(conn, args, false);
3186 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3187 void *aux OVS_UNUSED)
3189 uint8_t mac[ETH_ADDR_LEN];
3193 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3194 == ETH_ADDR_SCAN_COUNT) {
3195 hash = bond_hash(mac);
3197 hash_cstr = xasprintf("%u", hash);
3198 unixctl_command_reply(conn, 200, hash_cstr);
3201 unixctl_command_reply(conn, 501, "invalid mac");
3208 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3209 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3210 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3211 unixctl_command_register("bond/set-active-slave",
3212 bond_unixctl_set_active_slave, NULL);
3213 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3215 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3217 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3220 /* Port functions. */
3222 static struct port *
3223 port_create(struct bridge *br, const char *name)
3227 port = xzalloc(sizeof *port);
3229 port->port_idx = br->n_ports;
3231 port->trunks = NULL;
3232 port->name = xstrdup(name);
3233 port->active_iface = -1;
3235 if (br->n_ports >= br->allocated_ports) {
3236 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3239 br->ports[br->n_ports++] = port;
3240 shash_add_assert(&br->port_by_name, port->name, port);
3242 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3249 get_port_other_config(const struct ovsrec_port *port, const char *key,
3250 const char *default_value)
3252 const char *value = get_ovsrec_key_value(key,
3253 port->key_other_config,
3254 port->value_other_config,
3255 port->n_other_config);
3256 return value ? value : default_value;
3260 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3262 struct shash new_ifaces;
3265 /* Collect list of new interfaces. */
3266 shash_init(&new_ifaces);
3267 for (i = 0; i < cfg->n_interfaces; i++) {
3268 const char *name = cfg->interfaces[i]->name;
3269 shash_add_once(&new_ifaces, name, NULL);
3272 /* Get rid of deleted interfaces. */
3273 for (i = 0; i < port->n_ifaces; ) {
3274 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3275 iface_destroy(port->ifaces[i]);
3281 shash_destroy(&new_ifaces);
3285 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3287 struct shash new_ifaces;
3288 long long int next_rebalance;
3289 unsigned long *trunks;
3295 /* Update settings. */
3296 port->updelay = cfg->bond_updelay;
3297 if (port->updelay < 0) {
3300 port->updelay = cfg->bond_downdelay;
3301 if (port->downdelay < 0) {
3302 port->downdelay = 0;
3304 port->bond_rebalance_interval = atoi(
3305 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3306 if (port->bond_rebalance_interval < 1000) {
3307 port->bond_rebalance_interval = 1000;
3309 next_rebalance = time_msec() + port->bond_rebalance_interval;
3310 if (port->bond_next_rebalance > next_rebalance) {
3311 port->bond_next_rebalance = next_rebalance;
3314 /* Add new interfaces and update 'cfg' member of existing ones. */
3315 shash_init(&new_ifaces);
3316 for (i = 0; i < cfg->n_interfaces; i++) {
3317 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3318 struct iface *iface;
3320 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3321 VLOG_WARN("port %s: %s specified twice as port interface",
3322 port->name, if_cfg->name);
3326 iface = iface_lookup(port->bridge, if_cfg->name);
3328 if (iface->port != port) {
3329 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3331 port->bridge->name, if_cfg->name, iface->port->name);
3334 iface->cfg = if_cfg;
3336 iface_create(port, if_cfg);
3339 shash_destroy(&new_ifaces);
3344 if (port->n_ifaces < 2) {
3346 if (vlan >= 0 && vlan <= 4095) {
3347 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3352 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3353 * they even work as-is. But they have not been tested. */
3354 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3358 if (port->vlan != vlan) {
3360 bridge_flush(port->bridge);
3363 /* Get trunked VLANs. */
3365 if (vlan < 0 && cfg->n_trunks) {
3369 trunks = bitmap_allocate(4096);
3371 for (i = 0; i < cfg->n_trunks; i++) {
3372 int trunk = cfg->trunks[i];
3374 bitmap_set1(trunks, trunk);
3380 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3381 port->name, cfg->n_trunks);
3383 if (n_errors == cfg->n_trunks) {
3384 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3386 bitmap_free(trunks);
3389 } else if (vlan >= 0 && cfg->n_trunks) {
3390 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3394 ? port->trunks != NULL
3395 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3396 bridge_flush(port->bridge);
3398 bitmap_free(port->trunks);
3399 port->trunks = trunks;
3403 port_destroy(struct port *port)
3406 struct bridge *br = port->bridge;
3410 proc_net_compat_update_vlan(port->name, NULL, 0);
3411 proc_net_compat_update_bond(port->name, NULL);
3413 for (i = 0; i < MAX_MIRRORS; i++) {
3414 struct mirror *m = br->mirrors[i];
3415 if (m && m->out_port == port) {
3420 while (port->n_ifaces > 0) {
3421 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3424 shash_find_and_delete_assert(&br->port_by_name, port->name);
3426 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3427 del->port_idx = port->port_idx;
3430 bitmap_free(port->trunks);
3437 static struct port *
3438 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3440 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3441 return iface ? iface->port : NULL;
3444 static struct port *
3445 port_lookup(const struct bridge *br, const char *name)
3447 return shash_find_data(&br->port_by_name, name);
3450 static struct iface *
3451 port_lookup_iface(const struct port *port, const char *name)
3453 struct iface *iface = iface_lookup(port->bridge, name);
3454 return iface && iface->port == port ? iface : NULL;
3458 port_update_bonding(struct port *port)
3460 if (port->n_ifaces < 2) {
3461 /* Not a bonded port. */
3462 if (port->bond_hash) {
3463 free(port->bond_hash);
3464 port->bond_hash = NULL;
3465 port->bond_compat_is_stale = true;
3466 port->bond_fake_iface = false;
3469 if (!port->bond_hash) {
3472 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3473 for (i = 0; i <= BOND_MASK; i++) {
3474 struct bond_entry *e = &port->bond_hash[i];
3478 port->no_ifaces_tag = tag_create_random();
3479 bond_choose_active_iface(port);
3480 port->bond_next_rebalance
3481 = time_msec() + port->bond_rebalance_interval;
3483 if (port->cfg->bond_fake_iface) {
3484 port->bond_next_fake_iface_update = time_msec();
3487 port->bond_compat_is_stale = true;
3488 port->bond_fake_iface = port->cfg->bond_fake_iface;
3493 port_update_bond_compat(struct port *port)
3495 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3496 struct compat_bond bond;
3499 if (port->n_ifaces < 2) {
3500 proc_net_compat_update_bond(port->name, NULL);
3505 bond.updelay = port->updelay;
3506 bond.downdelay = port->downdelay;
3509 bond.hashes = compat_hashes;
3510 if (port->bond_hash) {
3511 const struct bond_entry *e;
3512 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3513 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3514 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3515 cbh->hash = e - port->bond_hash;
3516 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3521 bond.n_slaves = port->n_ifaces;
3522 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3523 for (i = 0; i < port->n_ifaces; i++) {
3524 struct iface *iface = port->ifaces[i];
3525 struct compat_bond_slave *slave = &bond.slaves[i];
3526 slave->name = iface->name;
3528 /* We need to make the same determination as the Linux bonding
3529 * code to determine whether a slave should be consider "up".
3530 * The Linux function bond_miimon_inspect() supports four
3531 * BOND_LINK_* states:
3533 * - BOND_LINK_UP: carrier detected, updelay has passed.
3534 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3535 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3536 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3538 * The function bond_info_show_slave() only considers BOND_LINK_UP
3539 * to be "up" and anything else to be "down".
3541 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3545 netdev_get_etheraddr(iface->netdev, slave->mac);
3548 if (port->bond_fake_iface) {
3549 struct netdev *bond_netdev;
3551 if (!netdev_open_default(port->name, &bond_netdev)) {
3553 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3555 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3557 netdev_close(bond_netdev);
3561 proc_net_compat_update_bond(port->name, &bond);
3566 port_update_vlan_compat(struct port *port)
3568 struct bridge *br = port->bridge;
3569 char *vlandev_name = NULL;
3571 if (port->vlan > 0) {
3572 /* Figure out the name that the VLAN device should actually have, if it
3573 * existed. This takes some work because the VLAN device would not
3574 * have port->name in its name; rather, it would have the trunk port's
3575 * name, and 'port' would be attached to a bridge that also had the
3576 * VLAN device one of its ports. So we need to find a trunk port that
3577 * includes port->vlan.
3579 * There might be more than one candidate. This doesn't happen on
3580 * XenServer, so if it happens we just pick the first choice in
3581 * alphabetical order instead of creating multiple VLAN devices. */
3583 for (i = 0; i < br->n_ports; i++) {
3584 struct port *p = br->ports[i];
3585 if (port_trunks_vlan(p, port->vlan)
3587 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3589 uint8_t ea[ETH_ADDR_LEN];
3590 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3591 if (!eth_addr_is_multicast(ea) &&
3592 !eth_addr_is_reserved(ea) &&
3593 !eth_addr_is_zero(ea)) {
3594 vlandev_name = p->name;
3599 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3602 /* Interface functions. */
3604 static struct iface *
3605 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3607 struct bridge *br = port->bridge;
3608 struct iface *iface;
3609 char *name = if_cfg->name;
3612 iface = xzalloc(sizeof *iface);
3614 iface->port_ifidx = port->n_ifaces;
3615 iface->name = xstrdup(name);
3616 iface->dp_ifidx = -1;
3617 iface->tag = tag_create_random();
3618 iface->delay_expires = LLONG_MAX;
3619 iface->netdev = NULL;
3620 iface->cfg = if_cfg;
3622 shash_add_assert(&br->iface_by_name, iface->name, iface);
3624 /* Attempt to create the network interface in case it doesn't exist yet. */
3625 if (!iface_is_internal(br, iface->name)) {
3626 error = set_up_iface(if_cfg, iface, true);
3628 VLOG_WARN("could not create iface %s: %s", iface->name,
3631 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3638 if (port->n_ifaces >= port->allocated_ifaces) {
3639 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3640 sizeof *port->ifaces);
3642 port->ifaces[port->n_ifaces++] = iface;
3643 if (port->n_ifaces > 1) {
3644 br->has_bonded_ports = true;
3647 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3655 iface_destroy(struct iface *iface)
3658 struct port *port = iface->port;
3659 struct bridge *br = port->bridge;
3660 bool del_active = port->active_iface == iface->port_ifidx;
3663 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3665 if (iface->dp_ifidx >= 0) {
3666 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3669 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3670 del->port_ifidx = iface->port_ifidx;
3672 netdev_close(iface->netdev);
3675 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3676 bond_choose_active_iface(port);
3677 bond_send_learning_packets(port);
3683 bridge_flush(port->bridge);
3687 static struct iface *
3688 iface_lookup(const struct bridge *br, const char *name)
3690 return shash_find_data(&br->iface_by_name, name);
3693 static struct iface *
3694 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3696 return port_array_get(&br->ifaces, dp_ifidx);
3699 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3700 * 'br', that is, an interface that is entirely simulated within the datapath.
3701 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3702 * interfaces are created by setting "iface.<iface>.internal = true".
3704 * In addition, we have a kluge-y feature that creates an internal port with
3705 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3706 * This feature needs to go away in the long term. Until then, this is one
3707 * reason why this function takes a name instead of a struct iface: the fake
3708 * interfaces created this way do not have a struct iface. */
3710 iface_is_internal(const struct bridge *br, const char *if_name)
3712 struct iface *iface;
3715 if (!strcmp(if_name, br->name)) {
3719 iface = iface_lookup(br, if_name);
3720 if (iface && !strcmp(iface->cfg->type, "internal")) {
3724 port = port_lookup(br, if_name);
3725 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3731 /* Set Ethernet address of 'iface', if one is specified in the configuration
3734 iface_set_mac(struct iface *iface)
3736 uint8_t ea[ETH_ADDR_LEN];
3738 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3739 if (eth_addr_is_multicast(ea)) {
3740 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3742 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3743 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3744 iface->name, iface->name);
3746 int error = netdev_set_etheraddr(iface->netdev, ea);
3748 VLOG_ERR("interface %s: setting MAC failed (%s)",
3749 iface->name, strerror(error));
3755 /* Port mirroring. */
3758 mirror_reconfigure(struct bridge *br)
3760 struct shash old_mirrors, new_mirrors;
3761 struct shash_node *node;
3762 unsigned long *rspan_vlans;
3765 /* Collect old mirrors. */
3766 shash_init(&old_mirrors);
3767 for (i = 0; i < MAX_MIRRORS; i++) {
3768 if (br->mirrors[i]) {
3769 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3773 /* Collect new mirrors. */
3774 shash_init(&new_mirrors);
3775 for (i = 0; i < br->cfg->n_mirrors; i++) {
3776 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3777 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3778 VLOG_WARN("bridge %s: %s specified twice as mirror",
3779 br->name, cfg->name);
3783 /* Get rid of deleted mirrors and add new mirrors. */
3784 SHASH_FOR_EACH (node, &old_mirrors) {
3785 if (!shash_find(&new_mirrors, node->name)) {
3786 mirror_destroy(node->data);
3789 SHASH_FOR_EACH (node, &new_mirrors) {
3790 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3792 mirror = mirror_create(br, node->name);
3797 mirror_reconfigure_one(mirror, node->data);
3799 shash_destroy(&old_mirrors);
3800 shash_destroy(&new_mirrors);
3802 /* Update port reserved status. */
3803 for (i = 0; i < br->n_ports; i++) {
3804 br->ports[i]->is_mirror_output_port = false;
3806 for (i = 0; i < MAX_MIRRORS; i++) {
3807 struct mirror *m = br->mirrors[i];
3808 if (m && m->out_port) {
3809 m->out_port->is_mirror_output_port = true;
3813 /* Update flooded vlans (for RSPAN). */
3815 if (br->cfg->n_flood_vlans) {
3816 rspan_vlans = bitmap_allocate(4096);
3818 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3819 int64_t vlan = br->cfg->flood_vlans[i];
3820 if (vlan >= 0 && vlan < 4096) {
3821 bitmap_set1(rspan_vlans, vlan);
3822 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3825 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3830 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3835 static struct mirror *
3836 mirror_create(struct bridge *br, const char *name)
3841 for (i = 0; ; i++) {
3842 if (i >= MAX_MIRRORS) {
3843 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3844 "cannot create %s", br->name, MAX_MIRRORS, name);
3847 if (!br->mirrors[i]) {
3852 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3855 br->mirrors[i] = m = xzalloc(sizeof *m);
3858 m->name = xstrdup(name);
3859 shash_init(&m->src_ports);
3860 shash_init(&m->dst_ports);
3870 mirror_destroy(struct mirror *m)
3873 struct bridge *br = m->bridge;
3876 for (i = 0; i < br->n_ports; i++) {
3877 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3878 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3881 shash_destroy(&m->src_ports);
3882 shash_destroy(&m->dst_ports);
3885 m->bridge->mirrors[m->idx] = NULL;
3893 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3894 struct shash *names)
3898 for (i = 0; i < n_ports; i++) {
3899 const char *name = ports[i]->name;
3900 if (port_lookup(m->bridge, name)) {
3901 shash_add_once(names, name, NULL);
3903 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3904 "port %s", m->bridge->name, m->name, name);
3910 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3916 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3918 for (i = 0; i < cfg->n_select_vlan; i++) {
3919 int64_t vlan = cfg->select_vlan[i];
3920 if (vlan < 0 || vlan > 4095) {
3921 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3922 m->bridge->name, m->name, vlan);
3924 (*vlans)[n_vlans++] = vlan;
3931 vlan_is_mirrored(const struct mirror *m, int vlan)
3935 for (i = 0; i < m->n_vlans; i++) {
3936 if (m->vlans[i] == vlan) {
3944 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3948 for (i = 0; i < m->n_vlans; i++) {
3949 if (port_trunks_vlan(p, m->vlans[i])) {
3957 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3959 struct shash src_ports, dst_ports;
3960 mirror_mask_t mirror_bit;
3961 struct port *out_port;
3967 /* Get output port. */
3968 if (cfg->output_port) {
3969 out_port = port_lookup(m->bridge, cfg->output_port->name);
3971 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3972 m->bridge->name, m->name);
3978 if (cfg->output_vlan) {
3979 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3980 "output vlan; ignoring output vlan",
3981 m->bridge->name, m->name);
3983 } else if (cfg->output_vlan) {
3985 out_vlan = *cfg->output_vlan;
3987 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3988 m->bridge->name, m->name);
3993 shash_init(&src_ports);
3994 shash_init(&dst_ports);
3995 if (cfg->select_all) {
3996 for (i = 0; i < m->bridge->n_ports; i++) {
3997 const char *name = m->bridge->ports[i]->name;
3998 shash_add_once(&src_ports, name, NULL);
3999 shash_add_once(&dst_ports, name, NULL);
4004 /* Get ports, and drop duplicates and ports that don't exist. */
4005 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4007 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4010 /* Get all the vlans, and drop duplicate and invalid vlans. */
4011 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4014 /* Update mirror data. */
4015 if (!shash_equal_keys(&m->src_ports, &src_ports)
4016 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4017 || m->n_vlans != n_vlans
4018 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4019 || m->out_port != out_port
4020 || m->out_vlan != out_vlan) {
4021 bridge_flush(m->bridge);
4023 shash_swap(&m->src_ports, &src_ports);
4024 shash_swap(&m->dst_ports, &dst_ports);
4027 m->n_vlans = n_vlans;
4028 m->out_port = out_port;
4029 m->out_vlan = out_vlan;
4032 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4033 for (i = 0; i < m->bridge->n_ports; i++) {
4034 struct port *port = m->bridge->ports[i];
4036 if (shash_find(&m->src_ports, port->name)
4039 ? port_trunks_any_mirrored_vlan(m, port)
4040 : vlan_is_mirrored(m, port->vlan)))) {
4041 port->src_mirrors |= mirror_bit;
4043 port->src_mirrors &= ~mirror_bit;
4046 if (shash_find(&m->dst_ports, port->name)) {
4047 port->dst_mirrors |= mirror_bit;
4049 port->dst_mirrors &= ~mirror_bit;
4054 shash_destroy(&src_ports);
4055 shash_destroy(&dst_ports);