1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
36 #include "dynamic-string.h"
40 #include "mac-learning.h"
43 #include "ofp-print.h"
45 #include "ofproto/netflow.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
54 #include "socket-util.h"
55 #include "stream-ssl.h"
61 #include "vswitchd/vswitch-idl.h"
62 #include "xenserver.h"
64 #include "sflow_api.h"
66 #define THIS_MODULE VLM_bridge
75 /* These members are always valid. */
76 struct port *port; /* Containing port. */
77 size_t port_ifidx; /* Index within containing port. */
78 char *name; /* Host network device name. */
79 tag_type tag; /* Tag associated with this interface. */
80 long long delay_expires; /* Time after which 'enabled' may change. */
82 /* These members are valid only after bridge_reconfigure() causes them to
84 int dp_ifidx; /* Index within kernel datapath. */
85 struct netdev *netdev; /* Network device. */
86 bool enabled; /* May be chosen for flows? */
88 /* This member is only valid *during* bridge_reconfigure(). */
89 const struct ovsrec_interface *cfg;
92 #define BOND_MASK 0xff
94 int iface_idx; /* Index of assigned iface, or -1 if none. */
95 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
96 tag_type iface_tag; /* Tag associated with iface_idx. */
99 #define MAX_MIRRORS 32
100 typedef uint32_t mirror_mask_t;
101 #define MIRROR_MASK_C(X) UINT32_C(X)
102 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
104 struct bridge *bridge;
108 /* Selection criteria. */
109 struct shash src_ports; /* Name is port name; data is always NULL. */
110 struct shash dst_ports; /* Name is port name; data is always NULL. */
115 struct port *out_port;
119 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
121 struct bridge *bridge;
123 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
124 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
127 /* An ordinary bridge port has 1 interface.
128 * A bridge port for bonding has at least 2 interfaces. */
129 struct iface **ifaces;
130 size_t n_ifaces, allocated_ifaces;
133 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
134 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
135 tag_type active_iface_tag; /* Tag for bcast flows. */
136 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
137 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
138 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
139 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
140 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
141 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
142 long long int bond_next_rebalance; /* Next rebalancing time. */
144 /* Port mirroring info. */
145 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
146 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
147 bool is_mirror_output_port; /* Does port mirroring send frames here? */
149 /* This member is only valid *during* bridge_reconfigure(). */
150 const struct ovsrec_port *cfg;
153 #define DP_MAX_PORTS 255
155 struct list node; /* Node in global list of bridges. */
156 char *name; /* User-specified arbitrary name. */
157 struct mac_learning *ml; /* MAC learning table. */
158 bool sent_config_request; /* Successfully sent config request? */
159 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
161 /* OpenFlow switch processing. */
162 struct ofproto *ofproto; /* OpenFlow switch. */
164 /* Description strings. */
165 char *mfr_desc; /* Manufacturer. */
166 char *hw_desc; /* Hardware. */
167 char *sw_desc; /* Software version. */
168 char *serial_desc; /* Serial number. */
169 char *dp_desc; /* Datapath description. */
171 /* Kernel datapath information. */
172 struct dpif *dpif; /* Datapath. */
173 struct port_array ifaces; /* Indexed by kernel datapath port number. */
177 size_t n_ports, allocated_ports;
180 bool has_bonded_ports;
185 /* Flow statistics gathering. */
186 time_t next_stats_request;
188 /* Port mirroring. */
189 struct mirror *mirrors[MAX_MIRRORS];
191 /* This member is only valid *during* bridge_reconfigure(). */
192 const struct ovsrec_bridge *cfg;
195 /* List of all bridges. */
196 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
198 /* Maximum number of datapaths. */
199 enum { DP_MAX = 256 };
201 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
202 static void bridge_destroy(struct bridge *);
203 static struct bridge *bridge_lookup(const char *name);
204 static unixctl_cb_func bridge_unixctl_dump_flows;
205 static int bridge_run_one(struct bridge *);
206 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
207 const struct bridge *br,
208 struct ovsrec_controller ***controllersp);
209 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
211 static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
213 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
214 static void bridge_fetch_dp_ifaces(struct bridge *);
215 static void bridge_flush(struct bridge *);
216 static void bridge_pick_local_hw_addr(struct bridge *,
217 uint8_t ea[ETH_ADDR_LEN],
218 struct iface **hw_addr_iface);
219 static uint64_t bridge_pick_datapath_id(struct bridge *,
220 const uint8_t bridge_ea[ETH_ADDR_LEN],
221 struct iface *hw_addr_iface);
222 static struct iface *bridge_get_local_iface(struct bridge *);
223 static uint64_t dpid_from_hash(const void *, size_t nbytes);
225 static unixctl_cb_func bridge_unixctl_fdb_show;
227 static void bond_init(void);
228 static void bond_run(struct bridge *);
229 static void bond_wait(struct bridge *);
230 static void bond_rebalance_port(struct port *);
231 static void bond_send_learning_packets(struct port *);
232 static void bond_enable_slave(struct iface *iface, bool enable);
234 static struct port *port_create(struct bridge *, const char *name);
235 static void port_reconfigure(struct port *, const struct ovsrec_port *);
236 static void port_destroy(struct port *);
237 static struct port *port_lookup(const struct bridge *, const char *name);
238 static struct iface *port_lookup_iface(const struct port *, const char *name);
239 static struct port *port_from_dp_ifidx(const struct bridge *,
241 static void port_update_bond_compat(struct port *);
242 static void port_update_vlan_compat(struct port *);
243 static void port_update_bonding(struct port *);
245 static struct mirror *mirror_create(struct bridge *, const char *name);
246 static void mirror_destroy(struct mirror *);
247 static void mirror_reconfigure(struct bridge *);
248 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
249 static bool vlan_is_mirrored(const struct mirror *, int vlan);
251 static struct iface *iface_create(struct port *port,
252 const struct ovsrec_interface *if_cfg);
253 static void iface_destroy(struct iface *);
254 static struct iface *iface_lookup(const struct bridge *, const char *name);
255 static struct iface *iface_from_dp_ifidx(const struct bridge *,
257 static bool iface_is_internal(const struct bridge *, const char *name);
258 static void iface_set_mac(struct iface *);
260 /* Hooks into ofproto processing. */
261 static struct ofhooks bridge_ofhooks;
263 /* Public functions. */
265 /* Adds the name of each interface used by a bridge, including local and
266 * internal ports, to 'svec'. */
268 bridge_get_ifaces(struct svec *svec)
270 struct bridge *br, *next;
273 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
274 for (i = 0; i < br->n_ports; i++) {
275 struct port *port = br->ports[i];
277 for (j = 0; j < port->n_ifaces; j++) {
278 struct iface *iface = port->ifaces[j];
279 if (iface->dp_ifidx < 0) {
280 VLOG_ERR("%s interface not in datapath %s, ignoring",
281 iface->name, dpif_name(br->dpif));
283 if (iface->dp_ifidx != ODPP_LOCAL) {
284 svec_add(svec, iface->name);
293 bridge_init(const struct ovsrec_open_vswitch *cfg)
295 struct svec bridge_names;
296 struct svec dpif_names, dpif_types;
299 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
301 svec_init(&bridge_names);
302 for (i = 0; i < cfg->n_bridges; i++) {
303 svec_add(&bridge_names, cfg->bridges[i]->name);
305 svec_sort(&bridge_names);
307 svec_init(&dpif_names);
308 svec_init(&dpif_types);
309 dp_enumerate_types(&dpif_types);
310 for (i = 0; i < dpif_types.n; i++) {
315 dp_enumerate_names(dpif_types.names[i], &dpif_names);
317 for (j = 0; j < dpif_names.n; j++) {
318 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
320 struct svec all_names;
323 svec_init(&all_names);
324 dpif_get_all_names(dpif, &all_names);
325 for (k = 0; k < all_names.n; k++) {
326 if (svec_contains(&bridge_names, all_names.names[k])) {
332 svec_destroy(&all_names);
337 svec_destroy(&bridge_names);
338 svec_destroy(&dpif_names);
339 svec_destroy(&dpif_types);
341 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
345 bridge_reconfigure(cfg);
350 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
352 /* XXX SSL should be configurable on a per-bridge basis. */
354 stream_ssl_set_private_key_file(ssl->private_key);
355 stream_ssl_set_certificate_file(ssl->certificate);
356 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
361 /* Attempt to create the network device 'iface_name' through the netdev
364 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
367 struct shash_node *node;
368 struct shash options;
372 shash_init(&options);
373 for (i = 0; i < iface_cfg->n_options; i++) {
374 shash_add(&options, iface_cfg->key_options[i],
375 xstrdup(iface_cfg->value_options[i]));
379 struct netdev_options netdev_options;
381 memset(&netdev_options, 0, sizeof netdev_options);
382 netdev_options.name = iface_cfg->name;
383 if (!strcmp(iface_cfg->type, "internal")) {
384 /* An "internal" config type maps to a netdev "system" type. */
385 netdev_options.type = "system";
387 netdev_options.type = iface_cfg->type;
389 netdev_options.args = &options;
390 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
391 netdev_options.may_create = true;
392 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
393 netdev_options.may_open = true;
396 error = netdev_open(&netdev_options, &iface->netdev);
399 netdev_get_carrier(iface->netdev, &iface->enabled);
401 } else if (iface->netdev) {
402 const char *netdev_type = netdev_get_type(iface->netdev);
403 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
404 ? iface_cfg->type : NULL;
406 /* An "internal" config type maps to a netdev "system" type. */
407 if (iface_type && !strcmp(iface_type, "internal")) {
408 iface_type = "system";
411 if (!iface_type || !strcmp(netdev_type, iface_type)) {
412 error = netdev_reconfigure(iface->netdev, &options);
414 VLOG_WARN("%s: attempting change device type from %s to %s",
415 iface_cfg->name, netdev_type, iface_type);
420 SHASH_FOR_EACH (node, &options) {
423 shash_destroy(&options);
429 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
431 return set_up_iface(iface_cfg, iface, false);
435 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
436 void *aux OVS_UNUSED)
438 if (!iface->netdev) {
439 int error = set_up_iface(iface->cfg, iface, true);
441 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
451 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
452 void *aux OVS_UNUSED)
454 if (iface->dp_ifidx >= 0) {
455 VLOG_DBG("%s has interface %s on port %d",
457 iface->name, iface->dp_ifidx);
460 VLOG_ERR("%s interface not in %s, dropping",
461 iface->name, dpif_name(br->dpif));
467 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
468 void *aux OVS_UNUSED)
470 /* Set policing attributes. */
471 netdev_set_policing(iface->netdev,
472 iface->cfg->ingress_policing_rate,
473 iface->cfg->ingress_policing_burst);
475 /* Set MAC address of internal interfaces other than the local
477 if (iface->dp_ifidx != ODPP_LOCAL
478 && iface_is_internal(br, iface->name)) {
479 iface_set_mac(iface);
485 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
486 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
487 * deletes from 'br' any ports that no longer have any interfaces. */
489 iterate_and_prune_ifaces(struct bridge *br,
490 bool (*cb)(struct bridge *, struct iface *,
496 for (i = 0; i < br->n_ports; ) {
497 struct port *port = br->ports[i];
498 for (j = 0; j < port->n_ifaces; ) {
499 struct iface *iface = port->ifaces[j];
500 if (cb(br, iface, aux)) {
503 iface_destroy(iface);
507 if (port->n_ifaces) {
510 VLOG_ERR("%s port has no interfaces, dropping", port->name);
517 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
519 struct ovsdb_idl_txn *txn;
520 struct shash old_br, new_br;
521 struct shash_node *node;
522 struct bridge *br, *next;
524 int sflow_bridge_number;
526 COVERAGE_INC(bridge_reconfigure);
528 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
530 /* Collect old and new bridges. */
533 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
534 shash_add(&old_br, br->name, br);
536 for (i = 0; i < ovs_cfg->n_bridges; i++) {
537 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
538 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
539 VLOG_WARN("more than one bridge named %s", br_cfg->name);
543 /* Get rid of deleted bridges and add new bridges. */
544 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
545 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
552 SHASH_FOR_EACH (node, &new_br) {
553 const char *br_name = node->name;
554 const struct ovsrec_bridge *br_cfg = node->data;
555 br = shash_find_data(&old_br, br_name);
557 /* If the bridge datapath type has changed, we need to tear it
558 * down and recreate. */
559 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
561 bridge_create(br_cfg);
564 bridge_create(br_cfg);
567 shash_destroy(&old_br);
568 shash_destroy(&new_br);
572 bridge_configure_ssl(ovs_cfg->ssl);
575 /* Reconfigure all bridges. */
576 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
577 bridge_reconfigure_one(ovs_cfg, br);
580 /* Add and delete ports on all datapaths.
582 * The kernel will reject any attempt to add a given port to a datapath if
583 * that port already belongs to a different datapath, so we must do all
584 * port deletions before any port additions. */
585 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
586 struct odp_port *dpif_ports;
588 struct shash want_ifaces;
590 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
591 bridge_get_all_ifaces(br, &want_ifaces);
592 for (i = 0; i < n_dpif_ports; i++) {
593 const struct odp_port *p = &dpif_ports[i];
594 if (!shash_find(&want_ifaces, p->devname)
595 && strcmp(p->devname, br->name)) {
596 int retval = dpif_port_del(br->dpif, p->port);
598 VLOG_ERR("failed to remove %s interface from %s: %s",
599 p->devname, dpif_name(br->dpif),
604 shash_destroy(&want_ifaces);
607 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
608 struct odp_port *dpif_ports;
610 struct shash cur_ifaces, want_ifaces;
611 struct shash_node *node;
613 /* Get the set of interfaces currently in this datapath. */
614 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
615 shash_init(&cur_ifaces);
616 for (i = 0; i < n_dpif_ports; i++) {
617 const char *name = dpif_ports[i].devname;
618 if (!shash_find(&cur_ifaces, name)) {
619 shash_add(&cur_ifaces, name, NULL);
624 /* Get the set of interfaces we want on this datapath. */
625 bridge_get_all_ifaces(br, &want_ifaces);
627 SHASH_FOR_EACH (node, &want_ifaces) {
628 const char *if_name = node->name;
629 struct iface *iface = node->data;
631 if (shash_find(&cur_ifaces, if_name)) {
632 /* Already exists, just reconfigure it. */
634 reconfigure_iface(iface->cfg, iface);
637 /* Need to add to datapath. */
641 /* Add to datapath. */
642 internal = iface_is_internal(br, if_name);
643 error = dpif_port_add(br->dpif, if_name,
644 internal ? ODP_PORT_INTERNAL : 0, NULL);
645 if (error == EFBIG) {
646 VLOG_ERR("ran out of valid port numbers on %s",
647 dpif_name(br->dpif));
650 VLOG_ERR("failed to add %s interface to %s: %s",
651 if_name, dpif_name(br->dpif), strerror(error));
655 shash_destroy(&cur_ifaces);
656 shash_destroy(&want_ifaces);
658 sflow_bridge_number = 0;
659 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
662 struct iface *local_iface;
663 struct iface *hw_addr_iface;
666 bridge_fetch_dp_ifaces(br);
668 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
669 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
671 /* Pick local port hardware address, datapath ID. */
672 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
673 local_iface = bridge_get_local_iface(br);
675 int error = netdev_set_etheraddr(local_iface->netdev, ea);
677 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
678 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
679 "Ethernet address: %s",
680 br->name, strerror(error));
684 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
685 ofproto_set_datapath_id(br->ofproto, dpid);
687 dpid_string = xasprintf("%012"PRIx64, dpid);
688 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
691 /* Set NetFlow configuration on this bridge. */
692 if (br->cfg->netflow) {
693 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
694 struct netflow_options opts;
696 memset(&opts, 0, sizeof opts);
698 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
699 if (nf_cfg->engine_type) {
700 opts.engine_type = *nf_cfg->engine_type;
702 if (nf_cfg->engine_id) {
703 opts.engine_id = *nf_cfg->engine_id;
706 opts.active_timeout = nf_cfg->active_timeout;
707 if (!opts.active_timeout) {
708 opts.active_timeout = -1;
709 } else if (opts.active_timeout < 0) {
710 VLOG_WARN("bridge %s: active timeout interval set to negative "
711 "value, using default instead (%d seconds)", br->name,
712 NF_ACTIVE_TIMEOUT_DEFAULT);
713 opts.active_timeout = -1;
716 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
717 if (opts.add_id_to_iface) {
718 if (opts.engine_id > 0x7f) {
719 VLOG_WARN("bridge %s: netflow port mangling may conflict "
720 "with another vswitch, choose an engine id less "
721 "than 128", br->name);
723 if (br->n_ports > 508) {
724 VLOG_WARN("bridge %s: netflow port mangling will conflict "
725 "with another port when more than 508 ports are "
730 opts.collectors.n = nf_cfg->n_targets;
731 opts.collectors.names = nf_cfg->targets;
732 if (ofproto_set_netflow(br->ofproto, &opts)) {
733 VLOG_ERR("bridge %s: problem setting netflow collectors",
737 ofproto_set_netflow(br->ofproto, NULL);
740 /* Set sFlow configuration on this bridge. */
741 if (br->cfg->sflow) {
742 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
743 struct ovsrec_controller **controllers;
744 struct ofproto_sflow_options oso;
745 size_t n_controllers;
748 memset(&oso, 0, sizeof oso);
750 oso.targets.n = sflow_cfg->n_targets;
751 oso.targets.names = sflow_cfg->targets;
753 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
754 if (sflow_cfg->sampling) {
755 oso.sampling_rate = *sflow_cfg->sampling;
758 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
759 if (sflow_cfg->polling) {
760 oso.polling_interval = *sflow_cfg->polling;
763 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
764 if (sflow_cfg->header) {
765 oso.header_len = *sflow_cfg->header;
768 oso.sub_id = sflow_bridge_number++;
769 oso.agent_device = sflow_cfg->agent;
771 oso.control_ip = NULL;
772 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
773 for (i = 0; i < n_controllers; i++) {
774 if (controllers[i]->local_ip) {
775 oso.control_ip = controllers[i]->local_ip;
779 ofproto_set_sflow(br->ofproto, &oso);
781 svec_destroy(&oso.targets);
783 ofproto_set_sflow(br->ofproto, NULL);
786 /* Update the controller and related settings. It would be more
787 * straightforward to call this from bridge_reconfigure_one(), but we
788 * can't do it there for two reasons. First, and most importantly, at
789 * that point we don't know the dp_ifidx of any interfaces that have
790 * been added to the bridge (because we haven't actually added them to
791 * the datapath). Second, at that point we haven't set the datapath ID
792 * yet; when a controller is configured, resetting the datapath ID will
793 * immediately disconnect from the controller, so it's better to set
794 * the datapath ID before the controller. */
795 bridge_reconfigure_controller(ovs_cfg, br);
797 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
798 for (i = 0; i < br->n_ports; i++) {
799 struct port *port = br->ports[i];
801 port_update_vlan_compat(port);
802 port_update_bonding(port);
805 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
806 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
809 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
811 ovsdb_idl_txn_commit(txn);
812 ovsdb_idl_txn_destroy(txn); /* XXX */
816 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
820 for (i = 0; i < n; i++) {
821 if (!strcmp(keys[i], key)) {
829 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
831 return get_ovsrec_key_value(key,
832 br_cfg->key_other_config,
833 br_cfg->value_other_config,
834 br_cfg->n_other_config);
838 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
839 struct iface **hw_addr_iface)
845 *hw_addr_iface = NULL;
847 /* Did the user request a particular MAC? */
848 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
849 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
850 if (eth_addr_is_multicast(ea)) {
851 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
852 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
853 } else if (eth_addr_is_zero(ea)) {
854 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
860 /* Otherwise choose the minimum non-local MAC address among all of the
862 memset(ea, 0xff, sizeof ea);
863 for (i = 0; i < br->n_ports; i++) {
864 struct port *port = br->ports[i];
865 uint8_t iface_ea[ETH_ADDR_LEN];
868 /* Mirror output ports don't participate. */
869 if (port->is_mirror_output_port) {
873 /* Choose the MAC address to represent the port. */
874 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
875 /* Find the interface with this Ethernet address (if any) so that
876 * we can provide the correct devname to the caller. */
878 for (j = 0; j < port->n_ifaces; j++) {
879 struct iface *candidate = port->ifaces[j];
880 uint8_t candidate_ea[ETH_ADDR_LEN];
881 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
882 && eth_addr_equals(iface_ea, candidate_ea)) {
887 /* Choose the interface whose MAC address will represent the port.
888 * The Linux kernel bonding code always chooses the MAC address of
889 * the first slave added to a bond, and the Fedora networking
890 * scripts always add slaves to a bond in alphabetical order, so
891 * for compatibility we choose the interface with the name that is
892 * first in alphabetical order. */
893 iface = port->ifaces[0];
894 for (j = 1; j < port->n_ifaces; j++) {
895 struct iface *candidate = port->ifaces[j];
896 if (strcmp(candidate->name, iface->name) < 0) {
901 /* The local port doesn't count (since we're trying to choose its
902 * MAC address anyway). */
903 if (iface->dp_ifidx == ODPP_LOCAL) {
908 error = netdev_get_etheraddr(iface->netdev, iface_ea);
910 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
911 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
912 iface->name, strerror(error));
917 /* Compare against our current choice. */
918 if (!eth_addr_is_multicast(iface_ea) &&
919 !eth_addr_is_local(iface_ea) &&
920 !eth_addr_is_reserved(iface_ea) &&
921 !eth_addr_is_zero(iface_ea) &&
922 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
924 memcpy(ea, iface_ea, ETH_ADDR_LEN);
925 *hw_addr_iface = iface;
928 if (eth_addr_is_multicast(ea)) {
929 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
930 *hw_addr_iface = NULL;
931 VLOG_WARN("bridge %s: using default bridge Ethernet "
932 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
934 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
935 br->name, ETH_ADDR_ARGS(ea));
939 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
940 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
941 * an interface on 'br', then that interface must be passed in as
942 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
943 * 'hw_addr_iface' must be passed in as a null pointer. */
945 bridge_pick_datapath_id(struct bridge *br,
946 const uint8_t bridge_ea[ETH_ADDR_LEN],
947 struct iface *hw_addr_iface)
950 * The procedure for choosing a bridge MAC address will, in the most
951 * ordinary case, also choose a unique MAC that we can use as a datapath
952 * ID. In some special cases, though, multiple bridges will end up with
953 * the same MAC address. This is OK for the bridges, but it will confuse
954 * the OpenFlow controller, because each datapath needs a unique datapath
957 * Datapath IDs must be unique. It is also very desirable that they be
958 * stable from one run to the next, so that policy set on a datapath
961 const char *datapath_id;
964 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
965 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
971 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
973 * A bridge whose MAC address is taken from a VLAN network device
974 * (that is, a network device created with vconfig(8) or similar
975 * tool) will have the same MAC address as a bridge on the VLAN
976 * device's physical network device.
978 * Handle this case by hashing the physical network device MAC
979 * along with the VLAN identifier.
981 uint8_t buf[ETH_ADDR_LEN + 2];
982 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
983 buf[ETH_ADDR_LEN] = vlan >> 8;
984 buf[ETH_ADDR_LEN + 1] = vlan;
985 return dpid_from_hash(buf, sizeof buf);
988 * Assume that this bridge's MAC address is unique, since it
989 * doesn't fit any of the cases we handle specially.
994 * A purely internal bridge, that is, one that has no non-virtual
995 * network devices on it at all, is more difficult because it has no
996 * natural unique identifier at all.
998 * When the host is a XenServer, we handle this case by hashing the
999 * host's UUID with the name of the bridge. Names of bridges are
1000 * persistent across XenServer reboots, although they can be reused if
1001 * an internal network is destroyed and then a new one is later
1002 * created, so this is fairly effective.
1004 * When the host is not a XenServer, we punt by using a random MAC
1005 * address on each run.
1007 const char *host_uuid = xenserver_get_host_uuid();
1009 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1010 dpid = dpid_from_hash(combined, strlen(combined));
1016 return eth_addr_to_uint64(bridge_ea);
1020 dpid_from_hash(const void *data, size_t n)
1022 uint8_t hash[SHA1_DIGEST_SIZE];
1024 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1025 sha1_bytes(data, n, hash);
1026 eth_addr_mark_random(hash);
1027 return eth_addr_to_uint64(hash);
1033 struct bridge *br, *next;
1037 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1038 int error = bridge_run_one(br);
1040 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1041 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1042 "forcing reconfiguration", br->name);
1056 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1057 ofproto_wait(br->ofproto);
1058 if (ofproto_has_controller(br->ofproto)) {
1062 mac_learning_wait(br->ml);
1067 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1068 * configuration changes. */
1070 bridge_flush(struct bridge *br)
1072 COVERAGE_INC(bridge_flush);
1074 mac_learning_flush(br->ml);
1077 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1078 * such interface. */
1079 static struct iface *
1080 bridge_get_local_iface(struct bridge *br)
1084 for (i = 0; i < br->n_ports; i++) {
1085 struct port *port = br->ports[i];
1086 for (j = 0; j < port->n_ifaces; j++) {
1087 struct iface *iface = port->ifaces[j];
1088 if (iface->dp_ifidx == ODPP_LOCAL) {
1097 /* Bridge unixctl user interface functions. */
1099 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1100 const char *args, void *aux OVS_UNUSED)
1102 struct ds ds = DS_EMPTY_INITIALIZER;
1103 const struct bridge *br;
1104 const struct mac_entry *e;
1106 br = bridge_lookup(args);
1108 unixctl_command_reply(conn, 501, "no such bridge");
1112 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1113 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1114 if (e->port < 0 || e->port >= br->n_ports) {
1117 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1118 br->ports[e->port]->ifaces[0]->dp_ifidx,
1119 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1121 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1125 /* Bridge reconfiguration functions. */
1126 static struct bridge *
1127 bridge_create(const struct ovsrec_bridge *br_cfg)
1132 assert(!bridge_lookup(br_cfg->name));
1133 br = xzalloc(sizeof *br);
1135 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1141 dpif_flow_flush(br->dpif);
1143 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1146 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1148 dpif_delete(br->dpif);
1149 dpif_close(br->dpif);
1154 br->name = xstrdup(br_cfg->name);
1156 br->ml = mac_learning_create();
1157 br->sent_config_request = false;
1158 eth_addr_nicira_random(br->default_ea);
1160 port_array_init(&br->ifaces);
1164 list_push_back(&all_bridges, &br->node);
1166 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1172 bridge_destroy(struct bridge *br)
1177 while (br->n_ports > 0) {
1178 port_destroy(br->ports[br->n_ports - 1]);
1180 list_remove(&br->node);
1181 error = dpif_delete(br->dpif);
1182 if (error && error != ENOENT) {
1183 VLOG_ERR("failed to delete %s: %s",
1184 dpif_name(br->dpif), strerror(error));
1186 dpif_close(br->dpif);
1187 ofproto_destroy(br->ofproto);
1188 mac_learning_destroy(br->ml);
1189 port_array_destroy(&br->ifaces);
1196 static struct bridge *
1197 bridge_lookup(const char *name)
1201 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1202 if (!strcmp(br->name, name)) {
1210 bridge_exists(const char *name)
1212 return bridge_lookup(name) ? true : false;
1216 bridge_get_datapathid(const char *name)
1218 struct bridge *br = bridge_lookup(name);
1219 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1222 /* Handle requests for a listing of all flows known by the OpenFlow
1223 * stack, including those normally hidden. */
1225 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1226 const char *args, void *aux OVS_UNUSED)
1231 br = bridge_lookup(args);
1233 unixctl_command_reply(conn, 501, "Unknown bridge");
1238 ofproto_get_all_flows(br->ofproto, &results);
1240 unixctl_command_reply(conn, 200, ds_cstr(&results));
1241 ds_destroy(&results);
1245 bridge_run_one(struct bridge *br)
1249 error = ofproto_run1(br->ofproto);
1254 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1257 error = ofproto_run2(br->ofproto, br->flush);
1264 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1265 const struct bridge *br,
1266 struct ovsrec_controller ***controllersp)
1268 struct ovsrec_controller **controllers;
1269 size_t n_controllers;
1271 if (br->cfg->n_controller) {
1272 controllers = br->cfg->controller;
1273 n_controllers = br->cfg->n_controller;
1275 controllers = ovs_cfg->controller;
1276 n_controllers = ovs_cfg->n_controller;
1279 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1285 *controllersp = controllers;
1287 return n_controllers;
1291 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1293 struct svec *ifaces = ifaces_;
1294 if (!svec_contains(ifaces, iface->name)) {
1295 svec_add(ifaces, iface->name);
1299 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1301 br->name, iface->name, iface->port->name);
1307 bridge_update_desc(struct bridge *br OVS_UNUSED)
1310 bool changed = false;
1313 desc = cfg_get_string(0, "bridge.%s.mfr-desc", br->name);
1314 if (desc != br->mfr_desc) {
1317 br->mfr_desc = xstrdup(desc);
1319 br->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
1324 desc = cfg_get_string(0, "bridge.%s.hw-desc", br->name);
1325 if (desc != br->hw_desc) {
1328 br->hw_desc = xstrdup(desc);
1330 br->hw_desc = xstrdup(DEFAULT_HW_DESC);
1335 desc = cfg_get_string(0, "bridge.%s.sw-desc", br->name);
1336 if (desc != br->sw_desc) {
1339 br->sw_desc = xstrdup(desc);
1341 br->sw_desc = xstrdup(DEFAULT_SW_DESC);
1346 desc = cfg_get_string(0, "bridge.%s.serial-desc", br->name);
1347 if (desc != br->serial_desc) {
1348 free(br->serial_desc);
1350 br->serial_desc = xstrdup(desc);
1352 br->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
1357 desc = cfg_get_string(0, "bridge.%s.dp-desc", br->name);
1358 if (desc != br->dp_desc) {
1361 br->dp_desc = xstrdup(desc);
1363 br->dp_desc = xstrdup(DEFAULT_DP_DESC);
1369 ofproto_set_desc(br->ofproto, br->mfr_desc, br->hw_desc,
1370 br->sw_desc, br->serial_desc, br->dp_desc);
1376 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1379 struct shash old_ports, new_ports;
1381 struct svec listeners, old_listeners;
1382 struct svec snoops, old_snoops;
1383 struct shash_node *node;
1386 /* Collect old ports. */
1387 shash_init(&old_ports);
1388 for (i = 0; i < br->n_ports; i++) {
1389 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1392 /* Collect new ports. */
1393 shash_init(&new_ports);
1394 for (i = 0; i < br->cfg->n_ports; i++) {
1395 const char *name = br->cfg->ports[i]->name;
1396 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1397 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1402 /* If we have a controller, then we need a local port. Complain if the
1403 * user didn't specify one.
1405 * XXX perhaps we should synthesize a port ourselves in this case. */
1406 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1407 char local_name[IF_NAMESIZE];
1410 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1411 local_name, sizeof local_name);
1412 if (!error && !shash_find(&new_ports, local_name)) {
1413 VLOG_WARN("bridge %s: controller specified but no local port "
1414 "(port named %s) defined",
1415 br->name, local_name);
1419 /* Get rid of deleted ports and add new ports. */
1420 SHASH_FOR_EACH (node, &old_ports) {
1421 if (!shash_find(&new_ports, node->name)) {
1422 port_destroy(node->data);
1425 SHASH_FOR_EACH (node, &new_ports) {
1426 struct port *port = shash_find_data(&old_ports, node->name);
1428 port = port_create(br, node->name);
1430 port_reconfigure(port, node->data);
1432 shash_destroy(&old_ports);
1433 shash_destroy(&new_ports);
1435 /* Check and delete duplicate interfaces. */
1437 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1438 svec_destroy(&ifaces);
1440 /* Delete all flows if we're switching from connected to standalone or vice
1441 * versa. (XXX Should we delete all flows if we are switching from one
1442 * controller to another?) */
1445 /* Configure OpenFlow management listeners. */
1446 svec_init(&listeners);
1447 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1449 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1450 ovs_rundir, br->name));
1451 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1452 svec_clear(&listeners);
1454 svec_sort_unique(&listeners);
1456 svec_init(&old_listeners);
1457 ofproto_get_listeners(br->ofproto, &old_listeners);
1458 svec_sort_unique(&old_listeners);
1460 if (!svec_equal(&listeners, &old_listeners)) {
1461 ofproto_set_listeners(br->ofproto, &listeners);
1463 svec_destroy(&listeners);
1464 svec_destroy(&old_listeners);
1466 /* Configure OpenFlow controller connection snooping. */
1468 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1470 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1471 ovs_rundir, br->name));
1472 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1473 svec_clear(&snoops);
1475 svec_sort_unique(&snoops);
1477 svec_init(&old_snoops);
1478 ofproto_get_snoops(br->ofproto, &old_snoops);
1479 svec_sort_unique(&old_snoops);
1481 if (!svec_equal(&snoops, &old_snoops)) {
1482 ofproto_set_snoops(br->ofproto, &snoops);
1484 svec_destroy(&snoops);
1485 svec_destroy(&old_snoops);
1487 /* Default listener. */
1488 svec_init(&listeners);
1489 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1490 ovs_rundir, br->name));
1491 svec_init(&old_listeners);
1492 ofproto_get_listeners(br->ofproto, &old_listeners);
1493 if (!svec_equal(&listeners, &old_listeners)) {
1494 ofproto_set_listeners(br->ofproto, &listeners);
1496 svec_destroy(&listeners);
1497 svec_destroy(&old_listeners);
1499 /* Default snoop. */
1501 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1502 ovs_rundir, br->name));
1503 svec_init(&old_snoops);
1504 ofproto_get_snoops(br->ofproto, &old_snoops);
1505 if (!svec_equal(&snoops, &old_snoops)) {
1506 ofproto_set_snoops(br->ofproto, &snoops);
1508 svec_destroy(&snoops);
1509 svec_destroy(&old_snoops);
1512 mirror_reconfigure(br);
1514 bridge_update_desc(br);
1518 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1521 struct ovsrec_controller **controllers;
1522 size_t n_controllers;
1524 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1525 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1526 ofproto_flush_flows(br->ofproto);
1529 if (!n_controllers) {
1530 union ofp_action action;
1533 /* Clear out controllers. */
1534 ofproto_set_controllers(br->ofproto, NULL, 0);
1536 /* Set up a flow that matches every packet and directs them to
1537 * OFPP_NORMAL (which goes to us). */
1538 memset(&action, 0, sizeof action);
1539 action.type = htons(OFPAT_OUTPUT);
1540 action.output.len = htons(sizeof action);
1541 action.output.port = htons(OFPP_NORMAL);
1542 memset(&flow, 0, sizeof flow);
1543 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1545 struct ofproto_controller *ocs;
1548 ocs = xmalloc(n_controllers * sizeof *ocs);
1549 for (i = 0; i < n_controllers; i++) {
1550 struct ovsrec_controller *c = controllers[i];
1551 struct ofproto_controller *oc = &ocs[i];
1553 if (strcmp(c->target, "discover")) {
1554 struct iface *local_iface;
1557 local_iface = bridge_get_local_iface(br);
1558 if (local_iface && c->local_ip
1559 && inet_aton(c->local_ip, &ip)) {
1560 struct netdev *netdev = local_iface->netdev;
1561 struct in_addr mask, gateway;
1563 if (!c->local_netmask
1564 || !inet_aton(c->local_netmask, &mask)) {
1567 if (!c->local_gateway
1568 || !inet_aton(c->local_gateway, &gateway)) {
1572 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1574 mask.s_addr = guess_netmask(ip.s_addr);
1576 if (!netdev_set_in4(netdev, ip, mask)) {
1577 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1579 br->name, IP_ARGS(&ip.s_addr),
1580 IP_ARGS(&mask.s_addr));
1583 if (gateway.s_addr) {
1584 if (!netdev_add_router(netdev, gateway)) {
1585 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1586 br->name, IP_ARGS(&gateway.s_addr));
1592 oc->target = c->target;
1593 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1594 oc->probe_interval = (c->inactivity_probe
1595 ? *c->inactivity_probe / 1000 : 5);
1596 oc->fail = (!c->fail_mode
1597 || !strcmp(c->fail_mode, "standalone")
1598 || !strcmp(c->fail_mode, "open")
1599 ? OFPROTO_FAIL_STANDALONE
1600 : OFPROTO_FAIL_SECURE);
1601 oc->band = (!c->connection_mode
1602 || !strcmp(c->connection_mode, "in-band")
1604 : OFPROTO_OUT_OF_BAND);
1605 oc->accept_re = c->discover_accept_regex;
1606 oc->update_resolv_conf = c->discover_update_resolv_conf;
1607 oc->rate_limit = (c->controller_rate_limit
1608 ? *c->controller_rate_limit : 0);
1609 oc->burst_limit = (c->controller_burst_limit
1610 ? *c->controller_burst_limit : 0);
1612 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1618 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1623 for (i = 0; i < br->n_ports; i++) {
1624 struct port *port = br->ports[i];
1625 for (j = 0; j < port->n_ifaces; j++) {
1626 struct iface *iface = port->ifaces[j];
1627 shash_add_once(ifaces, iface->name, iface);
1629 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1630 shash_add_once(ifaces, port->name, NULL);
1635 /* For robustness, in case the administrator moves around datapath ports behind
1636 * our back, we re-check all the datapath port numbers here.
1638 * This function will set the 'dp_ifidx' members of interfaces that have
1639 * disappeared to -1, so only call this function from a context where those
1640 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1641 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1642 * datapath, which doesn't support UINT16_MAX+1 ports. */
1644 bridge_fetch_dp_ifaces(struct bridge *br)
1646 struct odp_port *dpif_ports;
1647 size_t n_dpif_ports;
1650 /* Reset all interface numbers. */
1651 for (i = 0; i < br->n_ports; i++) {
1652 struct port *port = br->ports[i];
1653 for (j = 0; j < port->n_ifaces; j++) {
1654 struct iface *iface = port->ifaces[j];
1655 iface->dp_ifidx = -1;
1658 port_array_clear(&br->ifaces);
1660 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1661 for (i = 0; i < n_dpif_ports; i++) {
1662 struct odp_port *p = &dpif_ports[i];
1663 struct iface *iface = iface_lookup(br, p->devname);
1665 if (iface->dp_ifidx >= 0) {
1666 VLOG_WARN("%s reported interface %s twice",
1667 dpif_name(br->dpif), p->devname);
1668 } else if (iface_from_dp_ifidx(br, p->port)) {
1669 VLOG_WARN("%s reported interface %"PRIu16" twice",
1670 dpif_name(br->dpif), p->port);
1672 port_array_set(&br->ifaces, p->port, iface);
1673 iface->dp_ifidx = p->port;
1677 int64_t ofport = (iface->dp_ifidx >= 0
1678 ? odp_port_to_ofp_port(iface->dp_ifidx)
1680 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1687 /* Bridge packet processing functions. */
1690 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1692 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1695 static struct bond_entry *
1696 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1698 return &port->bond_hash[bond_hash(mac)];
1702 bond_choose_iface(const struct port *port)
1704 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1705 size_t i, best_down_slave = -1;
1706 long long next_delay_expiration = LLONG_MAX;
1708 for (i = 0; i < port->n_ifaces; i++) {
1709 struct iface *iface = port->ifaces[i];
1711 if (iface->enabled) {
1713 } else if (iface->delay_expires < next_delay_expiration) {
1714 best_down_slave = i;
1715 next_delay_expiration = iface->delay_expires;
1719 if (best_down_slave != -1) {
1720 struct iface *iface = port->ifaces[best_down_slave];
1722 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1723 "since no other interface is up", iface->name,
1724 iface->delay_expires - time_msec());
1725 bond_enable_slave(iface, true);
1728 return best_down_slave;
1732 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1733 uint16_t *dp_ifidx, tag_type *tags)
1735 struct iface *iface;
1737 assert(port->n_ifaces);
1738 if (port->n_ifaces == 1) {
1739 iface = port->ifaces[0];
1741 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1742 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1743 || !port->ifaces[e->iface_idx]->enabled) {
1744 /* XXX select interface properly. The current interface selection
1745 * is only good for testing the rebalancing code. */
1746 e->iface_idx = bond_choose_iface(port);
1747 if (e->iface_idx < 0) {
1748 *tags |= port->no_ifaces_tag;
1751 e->iface_tag = tag_create_random();
1752 ((struct port *) port)->bond_compat_is_stale = true;
1754 *tags |= e->iface_tag;
1755 iface = port->ifaces[e->iface_idx];
1757 *dp_ifidx = iface->dp_ifidx;
1758 *tags |= iface->tag; /* Currently only used for bonding. */
1763 bond_link_status_update(struct iface *iface, bool carrier)
1765 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1766 struct port *port = iface->port;
1768 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1769 /* Nothing to do. */
1772 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1773 iface->name, carrier ? "detected" : "dropped");
1774 if (carrier == iface->enabled) {
1775 iface->delay_expires = LLONG_MAX;
1776 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1777 iface->name, carrier ? "disabled" : "enabled");
1778 } else if (carrier && port->active_iface < 0) {
1779 bond_enable_slave(iface, true);
1780 if (port->updelay) {
1781 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1782 "other interface is up", iface->name, port->updelay);
1785 int delay = carrier ? port->updelay : port->downdelay;
1786 iface->delay_expires = time_msec() + delay;
1789 "interface %s: will be %s if it stays %s for %d ms",
1791 carrier ? "enabled" : "disabled",
1792 carrier ? "up" : "down",
1799 bond_choose_active_iface(struct port *port)
1801 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1803 port->active_iface = bond_choose_iface(port);
1804 port->active_iface_tag = tag_create_random();
1805 if (port->active_iface >= 0) {
1806 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1807 port->name, port->ifaces[port->active_iface]->name);
1809 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1815 bond_enable_slave(struct iface *iface, bool enable)
1817 struct port *port = iface->port;
1818 struct bridge *br = port->bridge;
1820 /* This acts as a recursion check. If the act of disabling a slave
1821 * causes a different slave to be enabled, the flag will allow us to
1822 * skip redundant work when we reenter this function. It must be
1823 * cleared on exit to keep things safe with multiple bonds. */
1824 static bool moving_active_iface = false;
1826 iface->delay_expires = LLONG_MAX;
1827 if (enable == iface->enabled) {
1831 iface->enabled = enable;
1832 if (!iface->enabled) {
1833 VLOG_WARN("interface %s: disabled", iface->name);
1834 ofproto_revalidate(br->ofproto, iface->tag);
1835 if (iface->port_ifidx == port->active_iface) {
1836 ofproto_revalidate(br->ofproto,
1837 port->active_iface_tag);
1839 /* Disabling a slave can lead to another slave being immediately
1840 * enabled if there will be no active slaves but one is waiting
1841 * on an updelay. In this case we do not need to run most of the
1842 * code for the newly enabled slave since there was no period
1843 * without an active slave and it is redundant with the disabling
1845 moving_active_iface = true;
1846 bond_choose_active_iface(port);
1848 bond_send_learning_packets(port);
1850 VLOG_WARN("interface %s: enabled", iface->name);
1851 if (port->active_iface < 0 && !moving_active_iface) {
1852 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1853 bond_choose_active_iface(port);
1854 bond_send_learning_packets(port);
1856 iface->tag = tag_create_random();
1859 moving_active_iface = false;
1860 port->bond_compat_is_stale = true;
1863 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1864 * bond interface. */
1866 bond_update_fake_iface_stats(struct port *port)
1868 struct netdev_stats bond_stats;
1869 struct netdev *bond_dev;
1872 memset(&bond_stats, 0, sizeof bond_stats);
1874 for (i = 0; i < port->n_ifaces; i++) {
1875 struct netdev_stats slave_stats;
1877 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1878 bond_stats.rx_packets += slave_stats.rx_packets;
1879 bond_stats.rx_bytes += slave_stats.rx_bytes;
1880 bond_stats.tx_packets += slave_stats.tx_packets;
1881 bond_stats.tx_bytes += slave_stats.tx_bytes;
1885 if (!netdev_open_default(port->name, &bond_dev)) {
1886 netdev_set_stats(bond_dev, &bond_stats);
1887 netdev_close(bond_dev);
1892 bond_run(struct bridge *br)
1896 for (i = 0; i < br->n_ports; i++) {
1897 struct port *port = br->ports[i];
1899 if (port->n_ifaces >= 2) {
1900 for (j = 0; j < port->n_ifaces; j++) {
1901 struct iface *iface = port->ifaces[j];
1902 if (time_msec() >= iface->delay_expires) {
1903 bond_enable_slave(iface, !iface->enabled);
1907 if (port->bond_fake_iface
1908 && time_msec() >= port->bond_next_fake_iface_update) {
1909 bond_update_fake_iface_stats(port);
1910 port->bond_next_fake_iface_update = time_msec() + 1000;
1914 if (port->bond_compat_is_stale) {
1915 port->bond_compat_is_stale = false;
1916 port_update_bond_compat(port);
1922 bond_wait(struct bridge *br)
1926 for (i = 0; i < br->n_ports; i++) {
1927 struct port *port = br->ports[i];
1928 if (port->n_ifaces < 2) {
1931 for (j = 0; j < port->n_ifaces; j++) {
1932 struct iface *iface = port->ifaces[j];
1933 if (iface->delay_expires != LLONG_MAX) {
1934 poll_timer_wait(iface->delay_expires - time_msec());
1937 if (port->bond_fake_iface) {
1938 poll_timer_wait(port->bond_next_fake_iface_update - time_msec());
1944 set_dst(struct dst *p, const flow_t *flow,
1945 const struct port *in_port, const struct port *out_port,
1948 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1949 : in_port->vlan >= 0 ? in_port->vlan
1950 : ntohs(flow->dl_vlan));
1951 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1955 swap_dst(struct dst *p, struct dst *q)
1957 struct dst tmp = *p;
1962 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1963 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1964 * that we push to the datapath. We could in fact fully sort the array by
1965 * vlan, but in most cases there are at most two different vlan tags so that's
1966 * possibly overkill.) */
1968 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1970 struct dst *first = dsts;
1971 struct dst *last = dsts + n_dsts;
1973 while (first != last) {
1975 * - All dsts < first have vlan == 'vlan'.
1976 * - All dsts >= last have vlan != 'vlan'.
1977 * - first < last. */
1978 while (first->vlan == vlan) {
1979 if (++first == last) {
1984 /* Same invariants, plus one additional:
1985 * - first->vlan != vlan.
1987 while (last[-1].vlan != vlan) {
1988 if (--last == first) {
1993 /* Same invariants, plus one additional:
1994 * - last[-1].vlan == vlan.*/
1995 swap_dst(first++, --last);
2000 mirror_mask_ffs(mirror_mask_t mask)
2002 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2007 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2008 const struct dst *test)
2011 for (i = 0; i < n_dsts; i++) {
2012 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2020 port_trunks_vlan(const struct port *port, uint16_t vlan)
2022 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
2026 port_includes_vlan(const struct port *port, uint16_t vlan)
2028 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2032 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2033 const struct port *in_port, const struct port *out_port,
2034 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2036 mirror_mask_t mirrors = in_port->src_mirrors;
2037 struct dst *dst = dsts;
2040 if (out_port == FLOOD_PORT) {
2041 /* XXX use ODP_FLOOD if no vlans or bonding. */
2042 /* XXX even better, define each VLAN as a datapath port group */
2043 for (i = 0; i < br->n_ports; i++) {
2044 struct port *port = br->ports[i];
2045 if (port != in_port && port_includes_vlan(port, vlan)
2046 && !port->is_mirror_output_port
2047 && set_dst(dst, flow, in_port, port, tags)) {
2048 mirrors |= port->dst_mirrors;
2052 *nf_output_iface = NF_OUT_FLOOD;
2053 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2054 *nf_output_iface = dst->dp_ifidx;
2055 mirrors |= out_port->dst_mirrors;
2060 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2061 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2063 if (set_dst(dst, flow, in_port, m->out_port, tags)
2064 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2068 for (i = 0; i < br->n_ports; i++) {
2069 struct port *port = br->ports[i];
2070 if (port_includes_vlan(port, m->out_vlan)
2071 && set_dst(dst, flow, in_port, port, tags))
2075 if (port->vlan < 0) {
2076 dst->vlan = m->out_vlan;
2078 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2082 /* Use the vlan tag on the original flow instead of
2083 * the one passed in the vlan parameter. This ensures
2084 * that we compare the vlan from before any implicit
2085 * tagging tags place. This is necessary because
2086 * dst->vlan is the final vlan, after removing implicit
2088 flow_vlan = ntohs(flow->dl_vlan);
2089 if (flow_vlan == 0) {
2090 flow_vlan = OFP_VLAN_NONE;
2092 if (port == in_port && dst->vlan == flow_vlan) {
2093 /* Don't send out input port on same VLAN. */
2101 mirrors &= mirrors - 1;
2104 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2108 static void OVS_UNUSED
2109 print_dsts(const struct dst *dsts, size_t n)
2111 for (; n--; dsts++) {
2112 printf(">p%"PRIu16, dsts->dp_ifidx);
2113 if (dsts->vlan != OFP_VLAN_NONE) {
2114 printf("v%"PRIu16, dsts->vlan);
2120 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2121 const struct port *in_port, const struct port *out_port,
2122 tag_type *tags, struct odp_actions *actions,
2123 uint16_t *nf_output_iface)
2125 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2127 const struct dst *p;
2130 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2133 cur_vlan = ntohs(flow->dl_vlan);
2134 for (p = dsts; p < &dsts[n_dsts]; p++) {
2135 union odp_action *a;
2136 if (p->vlan != cur_vlan) {
2137 if (p->vlan == OFP_VLAN_NONE) {
2138 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2140 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2141 a->vlan_vid.vlan_vid = htons(p->vlan);
2145 a = odp_actions_add(actions, ODPAT_OUTPUT);
2146 a->output.port = p->dp_ifidx;
2150 /* Returns the effective vlan of a packet, taking into account both the
2151 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2152 * the packet is untagged and -1 indicates it has an invalid header and
2153 * should be dropped. */
2154 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2155 struct port *in_port, bool have_packet)
2157 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2158 * belongs to VLAN 0, so we should treat both cases identically. (In the
2159 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2160 * presumably to allow a priority to be specified. In the latter case, the
2161 * packet does not have any 802.1Q header.) */
2162 int vlan = ntohs(flow->dl_vlan);
2163 if (vlan == OFP_VLAN_NONE) {
2166 if (in_port->vlan >= 0) {
2168 /* XXX support double tagging? */
2170 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2171 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2172 "packet received on port %s configured with "
2173 "implicit VLAN %"PRIu16,
2174 br->name, ntohs(flow->dl_vlan),
2175 in_port->name, in_port->vlan);
2179 vlan = in_port->vlan;
2181 if (!port_includes_vlan(in_port, vlan)) {
2183 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2184 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2185 "packet received on port %s not configured for "
2187 br->name, vlan, in_port->name, vlan);
2197 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2198 struct port *in_port)
2200 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2201 vlan, in_port->port_idx);
2203 /* The log messages here could actually be useful in debugging,
2204 * so keep the rate limit relatively high. */
2205 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2207 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2208 "on port %s in VLAN %d",
2209 br->name, ETH_ADDR_ARGS(flow->dl_src),
2210 in_port->name, vlan);
2211 ofproto_revalidate(br->ofproto, rev_tag);
2216 is_bcast_arp_reply(const flow_t *flow)
2218 return (flow->dl_type == htons(ETH_TYPE_ARP)
2219 && flow->nw_proto == ARP_OP_REPLY
2220 && eth_addr_is_broadcast(flow->dl_dst));
2223 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2224 * dropped. Returns true if they may be forwarded, false if they should be
2227 * If 'have_packet' is true, it indicates that the caller is processing a
2228 * received packet. If 'have_packet' is false, then the caller is just
2229 * revalidating an existing flow because configuration has changed. Either
2230 * way, 'have_packet' only affects logging (there is no point in logging errors
2231 * during revalidation).
2233 * Sets '*in_portp' to the input port. This will be a null pointer if
2234 * flow->in_port does not designate a known input port (in which case
2235 * is_admissible() returns false).
2237 * When returning true, sets '*vlanp' to the effective VLAN of the input
2238 * packet, as returned by flow_get_vlan().
2240 * May also add tags to '*tags', although the current implementation only does
2241 * so in one special case.
2244 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2245 tag_type *tags, int *vlanp, struct port **in_portp)
2247 struct iface *in_iface;
2248 struct port *in_port;
2251 /* Find the interface and port structure for the received packet. */
2252 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2254 /* No interface? Something fishy... */
2256 /* Odd. A few possible reasons here:
2258 * - We deleted an interface but there are still a few packets
2259 * queued up from it.
2261 * - Someone externally added an interface (e.g. with "ovs-dpctl
2262 * add-if") that we don't know about.
2264 * - Packet arrived on the local port but the local port is not
2265 * one of our bridge ports.
2267 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2269 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2270 "interface %"PRIu16, br->name, flow->in_port);
2276 *in_portp = in_port = in_iface->port;
2277 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2282 /* Drop frames for reserved multicast addresses. */
2283 if (eth_addr_is_reserved(flow->dl_dst)) {
2287 /* Drop frames on ports reserved for mirroring. */
2288 if (in_port->is_mirror_output_port) {
2290 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2291 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2292 "%s, which is reserved exclusively for mirroring",
2293 br->name, in_port->name);
2298 /* Packets received on bonds need special attention to avoid duplicates. */
2299 if (in_port->n_ifaces > 1) {
2302 if (eth_addr_is_multicast(flow->dl_dst)) {
2303 *tags |= in_port->active_iface_tag;
2304 if (in_port->active_iface != in_iface->port_ifidx) {
2305 /* Drop all multicast packets on inactive slaves. */
2310 /* Drop all packets for which we have learned a different input
2311 * port, because we probably sent the packet on one slave and got
2312 * it back on the other. Broadcast ARP replies are an exception
2313 * to this rule: the host has moved to another switch. */
2314 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2315 if (src_idx != -1 && src_idx != in_port->port_idx &&
2316 !is_bcast_arp_reply(flow)) {
2324 /* If the composed actions may be applied to any packet in the given 'flow',
2325 * returns true. Otherwise, the actions should only be applied to 'packet', or
2326 * not at all, if 'packet' was NULL. */
2328 process_flow(struct bridge *br, const flow_t *flow,
2329 const struct ofpbuf *packet, struct odp_actions *actions,
2330 tag_type *tags, uint16_t *nf_output_iface)
2332 struct port *in_port;
2333 struct port *out_port;
2337 /* Check whether we should drop packets in this flow. */
2338 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2343 /* Learn source MAC (but don't try to learn from revalidation). */
2345 update_learning_table(br, flow, vlan, in_port);
2348 /* Determine output port. */
2349 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags);
2350 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2351 out_port = br->ports[out_port_idx];
2352 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2353 /* If we are revalidating but don't have a learning entry then
2354 * eject the flow. Installing a flow that floods packets opens
2355 * up a window of time where we could learn from a packet reflected
2356 * on a bond and blackhole packets before the learning table is
2357 * updated to reflect the correct port. */
2360 out_port = FLOOD_PORT;
2363 /* Don't send packets out their input ports. */
2364 if (in_port == out_port) {
2370 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2377 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2380 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2381 const struct ofp_phy_port *opp,
2384 struct bridge *br = br_;
2385 struct iface *iface;
2388 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2394 if (reason == OFPPR_DELETE) {
2395 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2396 br->name, iface->name);
2397 iface_destroy(iface);
2398 if (!port->n_ifaces) {
2399 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2400 br->name, port->name);
2406 if (port->n_ifaces > 1) {
2407 bool up = !(opp->state & OFPPS_LINK_DOWN);
2408 bond_link_status_update(iface, up);
2409 port_update_bond_compat(port);
2415 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2416 struct odp_actions *actions, tag_type *tags,
2417 uint16_t *nf_output_iface, void *br_)
2419 struct bridge *br = br_;
2421 COVERAGE_INC(bridge_process_flow);
2422 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2426 bridge_account_flow_ofhook_cb(const flow_t *flow,
2427 const union odp_action *actions,
2428 size_t n_actions, unsigned long long int n_bytes,
2431 struct bridge *br = br_;
2432 const union odp_action *a;
2433 struct port *in_port;
2437 /* Feed information from the active flows back into the learning table
2438 * to ensure that table is always in sync with what is actually flowing
2439 * through the datapath. */
2440 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2441 update_learning_table(br, flow, vlan, in_port);
2444 if (!br->has_bonded_ports) {
2448 for (a = actions; a < &actions[n_actions]; a++) {
2449 if (a->type == ODPAT_OUTPUT) {
2450 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2451 if (out_port && out_port->n_ifaces >= 2) {
2452 struct bond_entry *e = lookup_bond_entry(out_port,
2454 e->tx_bytes += n_bytes;
2461 bridge_account_checkpoint_ofhook_cb(void *br_)
2463 struct bridge *br = br_;
2467 if (!br->has_bonded_ports) {
2472 for (i = 0; i < br->n_ports; i++) {
2473 struct port *port = br->ports[i];
2474 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2475 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2476 bond_rebalance_port(port);
2481 static struct ofhooks bridge_ofhooks = {
2482 bridge_port_changed_ofhook_cb,
2483 bridge_normal_ofhook_cb,
2484 bridge_account_flow_ofhook_cb,
2485 bridge_account_checkpoint_ofhook_cb,
2488 /* Bonding functions. */
2490 /* Statistics for a single interface on a bonded port, used for load-based
2491 * bond rebalancing. */
2492 struct slave_balance {
2493 struct iface *iface; /* The interface. */
2494 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2496 /* All the "bond_entry"s that are assigned to this interface, in order of
2497 * increasing tx_bytes. */
2498 struct bond_entry **hashes;
2502 /* Sorts pointers to pointers to bond_entries in ascending order by the
2503 * interface to which they are assigned, and within a single interface in
2504 * ascending order of bytes transmitted. */
2506 compare_bond_entries(const void *a_, const void *b_)
2508 const struct bond_entry *const *ap = a_;
2509 const struct bond_entry *const *bp = b_;
2510 const struct bond_entry *a = *ap;
2511 const struct bond_entry *b = *bp;
2512 if (a->iface_idx != b->iface_idx) {
2513 return a->iface_idx > b->iface_idx ? 1 : -1;
2514 } else if (a->tx_bytes != b->tx_bytes) {
2515 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2521 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2522 * *descending* order by number of bytes transmitted. */
2524 compare_slave_balance(const void *a_, const void *b_)
2526 const struct slave_balance *a = a_;
2527 const struct slave_balance *b = b_;
2528 if (a->iface->enabled != b->iface->enabled) {
2529 return a->iface->enabled ? -1 : 1;
2530 } else if (a->tx_bytes != b->tx_bytes) {
2531 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2538 swap_bals(struct slave_balance *a, struct slave_balance *b)
2540 struct slave_balance tmp = *a;
2545 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2546 * given that 'p' (and only 'p') might be in the wrong location.
2548 * This function invalidates 'p', since it might now be in a different memory
2551 resort_bals(struct slave_balance *p,
2552 struct slave_balance bals[], size_t n_bals)
2555 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2556 swap_bals(p, p - 1);
2558 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2559 swap_bals(p, p + 1);
2565 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2567 if (VLOG_IS_DBG_ENABLED()) {
2568 struct ds ds = DS_EMPTY_INITIALIZER;
2569 const struct slave_balance *b;
2571 for (b = bals; b < bals + n_bals; b++) {
2575 ds_put_char(&ds, ',');
2577 ds_put_format(&ds, " %s %"PRIu64"kB",
2578 b->iface->name, b->tx_bytes / 1024);
2580 if (!b->iface->enabled) {
2581 ds_put_cstr(&ds, " (disabled)");
2583 if (b->n_hashes > 0) {
2584 ds_put_cstr(&ds, " (");
2585 for (i = 0; i < b->n_hashes; i++) {
2586 const struct bond_entry *e = b->hashes[i];
2588 ds_put_cstr(&ds, " + ");
2590 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2591 e - port->bond_hash, e->tx_bytes / 1024);
2593 ds_put_cstr(&ds, ")");
2596 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2601 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2603 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2606 struct bond_entry *hash = from->hashes[hash_idx];
2607 struct port *port = from->iface->port;
2608 uint64_t delta = hash->tx_bytes;
2610 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2611 "from %s to %s (now carrying %"PRIu64"kB and "
2612 "%"PRIu64"kB load, respectively)",
2613 port->name, delta / 1024, hash - port->bond_hash,
2614 from->iface->name, to->iface->name,
2615 (from->tx_bytes - delta) / 1024,
2616 (to->tx_bytes + delta) / 1024);
2618 /* Delete element from from->hashes.
2620 * We don't bother to add the element to to->hashes because not only would
2621 * it require more work, the only purpose it would be to allow that hash to
2622 * be migrated to another slave in this rebalancing run, and there is no
2623 * point in doing that. */
2624 if (hash_idx == 0) {
2627 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2628 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2632 /* Shift load away from 'from' to 'to'. */
2633 from->tx_bytes -= delta;
2634 to->tx_bytes += delta;
2636 /* Arrange for flows to be revalidated. */
2637 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2638 hash->iface_idx = to->iface->port_ifidx;
2639 hash->iface_tag = tag_create_random();
2643 bond_rebalance_port(struct port *port)
2645 struct slave_balance bals[DP_MAX_PORTS];
2647 struct bond_entry *hashes[BOND_MASK + 1];
2648 struct slave_balance *b, *from, *to;
2649 struct bond_entry *e;
2652 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2653 * descending order of tx_bytes, so that bals[0] represents the most
2654 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2657 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2658 * array for each slave_balance structure, we sort our local array of
2659 * hashes in order by slave, so that all of the hashes for a given slave
2660 * become contiguous in memory, and then we point each 'hashes' members of
2661 * a slave_balance structure to the start of a contiguous group. */
2662 n_bals = port->n_ifaces;
2663 for (b = bals; b < &bals[n_bals]; b++) {
2664 b->iface = port->ifaces[b - bals];
2669 for (i = 0; i <= BOND_MASK; i++) {
2670 hashes[i] = &port->bond_hash[i];
2672 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2673 for (i = 0; i <= BOND_MASK; i++) {
2675 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2676 b = &bals[e->iface_idx];
2677 b->tx_bytes += e->tx_bytes;
2679 b->hashes = &hashes[i];
2684 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2685 log_bals(bals, n_bals, port);
2687 /* Discard slaves that aren't enabled (which were sorted to the back of the
2688 * array earlier). */
2689 while (!bals[n_bals - 1].iface->enabled) {
2696 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2697 to = &bals[n_bals - 1];
2698 for (from = bals; from < to; ) {
2699 uint64_t overload = from->tx_bytes - to->tx_bytes;
2700 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2701 /* The extra load on 'from' (and all less-loaded slaves), compared
2702 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2703 * it is less than ~1Mbps. No point in rebalancing. */
2705 } else if (from->n_hashes == 1) {
2706 /* 'from' only carries a single MAC hash, so we can't shift any
2707 * load away from it, even though we want to. */
2710 /* 'from' is carrying significantly more load than 'to', and that
2711 * load is split across at least two different hashes. Pick a hash
2712 * to migrate to 'to' (the least-loaded slave), given that doing so
2713 * must decrease the ratio of the load on the two slaves by at
2716 * The sort order we use means that we prefer to shift away the
2717 * smallest hashes instead of the biggest ones. There is little
2718 * reason behind this decision; we could use the opposite sort
2719 * order to shift away big hashes ahead of small ones. */
2723 for (i = 0; i < from->n_hashes; i++) {
2724 double old_ratio, new_ratio;
2725 uint64_t delta = from->hashes[i]->tx_bytes;
2727 if (delta == 0 || from->tx_bytes - delta == 0) {
2728 /* Pointless move. */
2732 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2734 if (to->tx_bytes == 0) {
2735 /* Nothing on the new slave, move it. */
2739 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2740 new_ratio = (double)(from->tx_bytes - delta) /
2741 (to->tx_bytes + delta);
2743 if (new_ratio == 0) {
2744 /* Should already be covered but check to prevent division
2749 if (new_ratio < 1) {
2750 new_ratio = 1 / new_ratio;
2753 if (old_ratio - new_ratio > 0.1) {
2754 /* Would decrease the ratio, move it. */
2758 if (i < from->n_hashes) {
2759 bond_shift_load(from, to, i);
2760 port->bond_compat_is_stale = true;
2762 /* If the result of the migration changed the relative order of
2763 * 'from' and 'to' swap them back to maintain invariants. */
2764 if (order_swapped) {
2765 swap_bals(from, to);
2768 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2769 * point to different slave_balance structures. It is only
2770 * valid to do these two operations in a row at all because we
2771 * know that 'from' will not move past 'to' and vice versa. */
2772 resort_bals(from, bals, n_bals);
2773 resort_bals(to, bals, n_bals);
2780 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2781 * historical data to decay to <1% in 7 rebalancing runs. */
2782 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2788 bond_send_learning_packets(struct port *port)
2790 struct bridge *br = port->bridge;
2791 struct mac_entry *e;
2792 struct ofpbuf packet;
2793 int error, n_packets, n_errors;
2795 if (!port->n_ifaces || port->active_iface < 0) {
2799 ofpbuf_init(&packet, 128);
2800 error = n_packets = n_errors = 0;
2801 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2802 union ofp_action actions[2], *a;
2808 if (e->port == port->port_idx
2809 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2813 /* Compose actions. */
2814 memset(actions, 0, sizeof actions);
2817 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2818 a->vlan_vid.len = htons(sizeof *a);
2819 a->vlan_vid.vlan_vid = htons(e->vlan);
2822 a->output.type = htons(OFPAT_OUTPUT);
2823 a->output.len = htons(sizeof *a);
2824 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2829 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2831 flow_extract(&packet, 0, ODPP_NONE, &flow);
2832 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2839 ofpbuf_uninit(&packet);
2842 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2843 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2844 "packets, last error was: %s",
2845 port->name, n_errors, n_packets, strerror(error));
2847 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2848 port->name, n_packets);
2852 /* Bonding unixctl user interface functions. */
2855 bond_unixctl_list(struct unixctl_conn *conn,
2856 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2858 struct ds ds = DS_EMPTY_INITIALIZER;
2859 const struct bridge *br;
2861 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2863 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2866 for (i = 0; i < br->n_ports; i++) {
2867 const struct port *port = br->ports[i];
2868 if (port->n_ifaces > 1) {
2871 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2872 for (j = 0; j < port->n_ifaces; j++) {
2873 const struct iface *iface = port->ifaces[j];
2875 ds_put_cstr(&ds, ", ");
2877 ds_put_cstr(&ds, iface->name);
2879 ds_put_char(&ds, '\n');
2883 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2887 static struct port *
2888 bond_find(const char *name)
2890 const struct bridge *br;
2892 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2895 for (i = 0; i < br->n_ports; i++) {
2896 struct port *port = br->ports[i];
2897 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2906 bond_unixctl_show(struct unixctl_conn *conn,
2907 const char *args, void *aux OVS_UNUSED)
2909 struct ds ds = DS_EMPTY_INITIALIZER;
2910 const struct port *port;
2913 port = bond_find(args);
2915 unixctl_command_reply(conn, 501, "no such bond");
2919 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2920 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2921 ds_put_format(&ds, "next rebalance: %lld ms\n",
2922 port->bond_next_rebalance - time_msec());
2923 for (j = 0; j < port->n_ifaces; j++) {
2924 const struct iface *iface = port->ifaces[j];
2925 struct bond_entry *be;
2928 ds_put_format(&ds, "slave %s: %s\n",
2929 iface->name, iface->enabled ? "enabled" : "disabled");
2930 if (j == port->active_iface) {
2931 ds_put_cstr(&ds, "\tactive slave\n");
2933 if (iface->delay_expires != LLONG_MAX) {
2934 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2935 iface->enabled ? "downdelay" : "updelay",
2936 iface->delay_expires - time_msec());
2940 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2941 int hash = be - port->bond_hash;
2942 struct mac_entry *me;
2944 if (be->iface_idx != j) {
2948 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2949 hash, be->tx_bytes / 1024);
2952 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2953 &port->bridge->ml->lrus) {
2956 if (bond_hash(me->mac) == hash
2957 && me->port != port->port_idx
2958 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2959 && dp_ifidx == iface->dp_ifidx)
2961 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2962 ETH_ADDR_ARGS(me->mac));
2967 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2972 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2973 void *aux OVS_UNUSED)
2975 char *args = (char *) args_;
2976 char *save_ptr = NULL;
2977 char *bond_s, *hash_s, *slave_s;
2978 uint8_t mac[ETH_ADDR_LEN];
2980 struct iface *iface;
2981 struct bond_entry *entry;
2984 bond_s = strtok_r(args, " ", &save_ptr);
2985 hash_s = strtok_r(NULL, " ", &save_ptr);
2986 slave_s = strtok_r(NULL, " ", &save_ptr);
2988 unixctl_command_reply(conn, 501,
2989 "usage: bond/migrate BOND HASH SLAVE");
2993 port = bond_find(bond_s);
2995 unixctl_command_reply(conn, 501, "no such bond");
2999 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3000 == ETH_ADDR_SCAN_COUNT) {
3001 hash = bond_hash(mac);
3002 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3003 hash = atoi(hash_s) & BOND_MASK;
3005 unixctl_command_reply(conn, 501, "bad hash");
3009 iface = port_lookup_iface(port, slave_s);
3011 unixctl_command_reply(conn, 501, "no such slave");
3015 if (!iface->enabled) {
3016 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3020 entry = &port->bond_hash[hash];
3021 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3022 entry->iface_idx = iface->port_ifidx;
3023 entry->iface_tag = tag_create_random();
3024 port->bond_compat_is_stale = true;
3025 unixctl_command_reply(conn, 200, "migrated");
3029 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3030 void *aux OVS_UNUSED)
3032 char *args = (char *) args_;
3033 char *save_ptr = NULL;
3034 char *bond_s, *slave_s;
3036 struct iface *iface;
3038 bond_s = strtok_r(args, " ", &save_ptr);
3039 slave_s = strtok_r(NULL, " ", &save_ptr);
3041 unixctl_command_reply(conn, 501,
3042 "usage: bond/set-active-slave BOND SLAVE");
3046 port = bond_find(bond_s);
3048 unixctl_command_reply(conn, 501, "no such bond");
3052 iface = port_lookup_iface(port, slave_s);
3054 unixctl_command_reply(conn, 501, "no such slave");
3058 if (!iface->enabled) {
3059 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3063 if (port->active_iface != iface->port_ifidx) {
3064 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3065 port->active_iface = iface->port_ifidx;
3066 port->active_iface_tag = tag_create_random();
3067 VLOG_INFO("port %s: active interface is now %s",
3068 port->name, iface->name);
3069 bond_send_learning_packets(port);
3070 unixctl_command_reply(conn, 200, "done");
3072 unixctl_command_reply(conn, 200, "no change");
3077 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3079 char *args = (char *) args_;
3080 char *save_ptr = NULL;
3081 char *bond_s, *slave_s;
3083 struct iface *iface;
3085 bond_s = strtok_r(args, " ", &save_ptr);
3086 slave_s = strtok_r(NULL, " ", &save_ptr);
3088 unixctl_command_reply(conn, 501,
3089 "usage: bond/enable/disable-slave BOND SLAVE");
3093 port = bond_find(bond_s);
3095 unixctl_command_reply(conn, 501, "no such bond");
3099 iface = port_lookup_iface(port, slave_s);
3101 unixctl_command_reply(conn, 501, "no such slave");
3105 bond_enable_slave(iface, enable);
3106 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3110 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3111 void *aux OVS_UNUSED)
3113 enable_slave(conn, args, true);
3117 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3118 void *aux OVS_UNUSED)
3120 enable_slave(conn, args, false);
3124 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3125 void *aux OVS_UNUSED)
3127 uint8_t mac[ETH_ADDR_LEN];
3131 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3132 == ETH_ADDR_SCAN_COUNT) {
3133 hash = bond_hash(mac);
3135 hash_cstr = xasprintf("%u", hash);
3136 unixctl_command_reply(conn, 200, hash_cstr);
3139 unixctl_command_reply(conn, 501, "invalid mac");
3146 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3147 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3148 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3149 unixctl_command_register("bond/set-active-slave",
3150 bond_unixctl_set_active_slave, NULL);
3151 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3153 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3155 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3158 /* Port functions. */
3160 static struct port *
3161 port_create(struct bridge *br, const char *name)
3165 port = xzalloc(sizeof *port);
3167 port->port_idx = br->n_ports;
3169 port->trunks = NULL;
3170 port->name = xstrdup(name);
3171 port->active_iface = -1;
3173 if (br->n_ports >= br->allocated_ports) {
3174 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3177 br->ports[br->n_ports++] = port;
3179 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3186 get_port_other_config(const struct ovsrec_port *port, const char *key,
3187 const char *default_value)
3189 const char *value = get_ovsrec_key_value(key,
3190 port->key_other_config,
3191 port->value_other_config,
3192 port->n_other_config);
3193 return value ? value : default_value;
3197 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3199 struct shash old_ifaces, new_ifaces;
3200 long long int next_rebalance;
3201 struct shash_node *node;
3202 unsigned long *trunks;
3208 /* Collect old and new interfaces. */
3209 shash_init(&old_ifaces);
3210 shash_init(&new_ifaces);
3211 for (i = 0; i < port->n_ifaces; i++) {
3212 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3214 for (i = 0; i < cfg->n_interfaces; i++) {
3215 const char *name = cfg->interfaces[i]->name;
3216 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3217 VLOG_WARN("port %s: %s specified twice as port interface",
3221 port->updelay = cfg->bond_updelay;
3222 if (port->updelay < 0) {
3225 port->updelay = cfg->bond_downdelay;
3226 if (port->downdelay < 0) {
3227 port->downdelay = 0;
3229 port->bond_rebalance_interval = atoi(
3230 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3231 if (port->bond_rebalance_interval < 1000) {
3232 port->bond_rebalance_interval = 1000;
3234 next_rebalance = time_msec() + port->bond_rebalance_interval;
3235 if (port->bond_next_rebalance > next_rebalance) {
3236 port->bond_next_rebalance = next_rebalance;
3239 /* Get rid of deleted interfaces and add new interfaces. */
3240 SHASH_FOR_EACH (node, &old_ifaces) {
3241 if (!shash_find(&new_ifaces, node->name)) {
3242 iface_destroy(node->data);
3245 SHASH_FOR_EACH (node, &new_ifaces) {
3246 const struct ovsrec_interface *if_cfg = node->data;
3247 struct iface *iface;
3249 iface = shash_find_data(&old_ifaces, if_cfg->name);
3251 iface_create(port, if_cfg);
3253 iface->cfg = if_cfg;
3260 if (port->n_ifaces < 2) {
3262 if (vlan >= 0 && vlan <= 4095) {
3263 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3268 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3269 * they even work as-is. But they have not been tested. */
3270 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3274 if (port->vlan != vlan) {
3276 bridge_flush(port->bridge);
3279 /* Get trunked VLANs. */
3285 trunks = bitmap_allocate(4096);
3287 for (i = 0; i < cfg->n_trunks; i++) {
3288 int trunk = cfg->trunks[i];
3290 bitmap_set1(trunks, trunk);
3296 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3297 port->name, cfg->n_trunks);
3299 if (n_errors == cfg->n_trunks) {
3301 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3304 bitmap_set_multiple(trunks, 0, 4096, 1);
3307 if (cfg->n_trunks) {
3308 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3313 ? port->trunks != NULL
3314 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3315 bridge_flush(port->bridge);
3317 bitmap_free(port->trunks);
3318 port->trunks = trunks;
3320 shash_destroy(&old_ifaces);
3321 shash_destroy(&new_ifaces);
3325 port_destroy(struct port *port)
3328 struct bridge *br = port->bridge;
3332 proc_net_compat_update_vlan(port->name, NULL, 0);
3333 proc_net_compat_update_bond(port->name, NULL);
3335 for (i = 0; i < MAX_MIRRORS; i++) {
3336 struct mirror *m = br->mirrors[i];
3337 if (m && m->out_port == port) {
3342 while (port->n_ifaces > 0) {
3343 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3346 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3347 del->port_idx = port->port_idx;
3350 bitmap_free(port->trunks);
3357 static struct port *
3358 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3360 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3361 return iface ? iface->port : NULL;
3364 static struct port *
3365 port_lookup(const struct bridge *br, const char *name)
3369 for (i = 0; i < br->n_ports; i++) {
3370 struct port *port = br->ports[i];
3371 if (!strcmp(port->name, name)) {
3378 static struct iface *
3379 port_lookup_iface(const struct port *port, const char *name)
3383 for (j = 0; j < port->n_ifaces; j++) {
3384 struct iface *iface = port->ifaces[j];
3385 if (!strcmp(iface->name, name)) {
3393 port_update_bonding(struct port *port)
3395 if (port->n_ifaces < 2) {
3396 /* Not a bonded port. */
3397 if (port->bond_hash) {
3398 free(port->bond_hash);
3399 port->bond_hash = NULL;
3400 port->bond_compat_is_stale = true;
3401 port->bond_fake_iface = false;
3404 if (!port->bond_hash) {
3407 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3408 for (i = 0; i <= BOND_MASK; i++) {
3409 struct bond_entry *e = &port->bond_hash[i];
3413 port->no_ifaces_tag = tag_create_random();
3414 bond_choose_active_iface(port);
3415 port->bond_next_rebalance
3416 = time_msec() + port->bond_rebalance_interval;
3418 if (port->cfg->bond_fake_iface) {
3419 port->bond_next_fake_iface_update = time_msec();
3422 port->bond_compat_is_stale = true;
3423 port->bond_fake_iface = port->cfg->bond_fake_iface;
3428 port_update_bond_compat(struct port *port)
3430 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3431 struct compat_bond bond;
3434 if (port->n_ifaces < 2) {
3435 proc_net_compat_update_bond(port->name, NULL);
3440 bond.updelay = port->updelay;
3441 bond.downdelay = port->downdelay;
3444 bond.hashes = compat_hashes;
3445 if (port->bond_hash) {
3446 const struct bond_entry *e;
3447 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3448 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3449 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3450 cbh->hash = e - port->bond_hash;
3451 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3456 bond.n_slaves = port->n_ifaces;
3457 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3458 for (i = 0; i < port->n_ifaces; i++) {
3459 struct iface *iface = port->ifaces[i];
3460 struct compat_bond_slave *slave = &bond.slaves[i];
3461 slave->name = iface->name;
3463 /* We need to make the same determination as the Linux bonding
3464 * code to determine whether a slave should be consider "up".
3465 * The Linux function bond_miimon_inspect() supports four
3466 * BOND_LINK_* states:
3468 * - BOND_LINK_UP: carrier detected, updelay has passed.
3469 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3470 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3471 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3473 * The function bond_info_show_slave() only considers BOND_LINK_UP
3474 * to be "up" and anything else to be "down".
3476 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3480 netdev_get_etheraddr(iface->netdev, slave->mac);
3483 if (port->bond_fake_iface) {
3484 struct netdev *bond_netdev;
3486 if (!netdev_open_default(port->name, &bond_netdev)) {
3488 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3490 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3492 netdev_close(bond_netdev);
3496 proc_net_compat_update_bond(port->name, &bond);
3501 port_update_vlan_compat(struct port *port)
3503 struct bridge *br = port->bridge;
3504 char *vlandev_name = NULL;
3506 if (port->vlan > 0) {
3507 /* Figure out the name that the VLAN device should actually have, if it
3508 * existed. This takes some work because the VLAN device would not
3509 * have port->name in its name; rather, it would have the trunk port's
3510 * name, and 'port' would be attached to a bridge that also had the
3511 * VLAN device one of its ports. So we need to find a trunk port that
3512 * includes port->vlan.
3514 * There might be more than one candidate. This doesn't happen on
3515 * XenServer, so if it happens we just pick the first choice in
3516 * alphabetical order instead of creating multiple VLAN devices. */
3518 for (i = 0; i < br->n_ports; i++) {
3519 struct port *p = br->ports[i];
3520 if (port_trunks_vlan(p, port->vlan)
3522 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3524 uint8_t ea[ETH_ADDR_LEN];
3525 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3526 if (!eth_addr_is_multicast(ea) &&
3527 !eth_addr_is_reserved(ea) &&
3528 !eth_addr_is_zero(ea)) {
3529 vlandev_name = p->name;
3534 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3537 /* Interface functions. */
3539 static struct iface *
3540 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3542 struct iface *iface;
3543 char *name = if_cfg->name;
3546 iface = xzalloc(sizeof *iface);
3548 iface->port_ifidx = port->n_ifaces;
3549 iface->name = xstrdup(name);
3550 iface->dp_ifidx = -1;
3551 iface->tag = tag_create_random();
3552 iface->delay_expires = LLONG_MAX;
3553 iface->netdev = NULL;
3554 iface->cfg = if_cfg;
3556 if (port->n_ifaces >= port->allocated_ifaces) {
3557 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3558 sizeof *port->ifaces);
3560 port->ifaces[port->n_ifaces++] = iface;
3561 if (port->n_ifaces > 1) {
3562 port->bridge->has_bonded_ports = true;
3565 /* Attempt to create the network interface in case it
3566 * doesn't exist yet. */
3567 if (!iface_is_internal(port->bridge, iface->name)) {
3568 error = set_up_iface(if_cfg, iface, true);
3570 VLOG_WARN("could not create iface %s: %s", iface->name,
3575 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3577 bridge_flush(port->bridge);
3583 iface_destroy(struct iface *iface)
3586 struct port *port = iface->port;
3587 struct bridge *br = port->bridge;
3588 bool del_active = port->active_iface == iface->port_ifidx;
3591 if (iface->dp_ifidx >= 0) {
3592 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3595 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3596 del->port_ifidx = iface->port_ifidx;
3598 netdev_close(iface->netdev);
3601 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3602 bond_choose_active_iface(port);
3603 bond_send_learning_packets(port);
3609 bridge_flush(port->bridge);
3613 static struct iface *
3614 iface_lookup(const struct bridge *br, const char *name)
3618 for (i = 0; i < br->n_ports; i++) {
3619 struct port *port = br->ports[i];
3620 for (j = 0; j < port->n_ifaces; j++) {
3621 struct iface *iface = port->ifaces[j];
3622 if (!strcmp(iface->name, name)) {
3630 static struct iface *
3631 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3633 return port_array_get(&br->ifaces, dp_ifidx);
3636 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3637 * 'br', that is, an interface that is entirely simulated within the datapath.
3638 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3639 * interfaces are created by setting "iface.<iface>.internal = true".
3641 * In addition, we have a kluge-y feature that creates an internal port with
3642 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3643 * This feature needs to go away in the long term. Until then, this is one
3644 * reason why this function takes a name instead of a struct iface: the fake
3645 * interfaces created this way do not have a struct iface. */
3647 iface_is_internal(const struct bridge *br, const char *if_name)
3649 /* XXX wastes time */
3650 struct iface *iface;
3653 if (!strcmp(if_name, br->name)) {
3657 iface = iface_lookup(br, if_name);
3658 if (iface && !strcmp(iface->cfg->type, "internal")) {
3662 port = port_lookup(br, if_name);
3663 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3669 /* Set Ethernet address of 'iface', if one is specified in the configuration
3672 iface_set_mac(struct iface *iface)
3674 uint8_t ea[ETH_ADDR_LEN];
3676 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3677 if (eth_addr_is_multicast(ea)) {
3678 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3680 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3681 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3682 iface->name, iface->name);
3684 int error = netdev_set_etheraddr(iface->netdev, ea);
3686 VLOG_ERR("interface %s: setting MAC failed (%s)",
3687 iface->name, strerror(error));
3693 /* Port mirroring. */
3696 mirror_reconfigure(struct bridge *br)
3698 struct shash old_mirrors, new_mirrors;
3699 struct shash_node *node;
3700 unsigned long *rspan_vlans;
3703 /* Collect old mirrors. */
3704 shash_init(&old_mirrors);
3705 for (i = 0; i < MAX_MIRRORS; i++) {
3706 if (br->mirrors[i]) {
3707 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3711 /* Collect new mirrors. */
3712 shash_init(&new_mirrors);
3713 for (i = 0; i < br->cfg->n_mirrors; i++) {
3714 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3715 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3716 VLOG_WARN("bridge %s: %s specified twice as mirror",
3717 br->name, cfg->name);
3721 /* Get rid of deleted mirrors and add new mirrors. */
3722 SHASH_FOR_EACH (node, &old_mirrors) {
3723 if (!shash_find(&new_mirrors, node->name)) {
3724 mirror_destroy(node->data);
3727 SHASH_FOR_EACH (node, &new_mirrors) {
3728 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3730 mirror = mirror_create(br, node->name);
3735 mirror_reconfigure_one(mirror, node->data);
3737 shash_destroy(&old_mirrors);
3738 shash_destroy(&new_mirrors);
3740 /* Update port reserved status. */
3741 for (i = 0; i < br->n_ports; i++) {
3742 br->ports[i]->is_mirror_output_port = false;
3744 for (i = 0; i < MAX_MIRRORS; i++) {
3745 struct mirror *m = br->mirrors[i];
3746 if (m && m->out_port) {
3747 m->out_port->is_mirror_output_port = true;
3751 /* Update flooded vlans (for RSPAN). */
3753 if (br->cfg->n_flood_vlans) {
3754 rspan_vlans = bitmap_allocate(4096);
3756 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3757 int64_t vlan = br->cfg->flood_vlans[i];
3758 if (vlan >= 0 && vlan < 4096) {
3759 bitmap_set1(rspan_vlans, vlan);
3760 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3763 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3768 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3773 static struct mirror *
3774 mirror_create(struct bridge *br, const char *name)
3779 for (i = 0; ; i++) {
3780 if (i >= MAX_MIRRORS) {
3781 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3782 "cannot create %s", br->name, MAX_MIRRORS, name);
3785 if (!br->mirrors[i]) {
3790 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3793 br->mirrors[i] = m = xzalloc(sizeof *m);
3796 m->name = xstrdup(name);
3797 shash_init(&m->src_ports);
3798 shash_init(&m->dst_ports);
3808 mirror_destroy(struct mirror *m)
3811 struct bridge *br = m->bridge;
3814 for (i = 0; i < br->n_ports; i++) {
3815 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3816 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3819 shash_destroy(&m->src_ports);
3820 shash_destroy(&m->dst_ports);
3823 m->bridge->mirrors[m->idx] = NULL;
3831 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3832 struct shash *names)
3836 for (i = 0; i < n_ports; i++) {
3837 const char *name = ports[i]->name;
3838 if (port_lookup(m->bridge, name)) {
3839 shash_add_once(names, name, NULL);
3841 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3842 "port %s", m->bridge->name, m->name, name);
3848 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3854 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3856 for (i = 0; i < cfg->n_select_vlan; i++) {
3857 int64_t vlan = cfg->select_vlan[i];
3858 if (vlan < 0 || vlan > 4095) {
3859 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3860 m->bridge->name, m->name, vlan);
3862 (*vlans)[n_vlans++] = vlan;
3869 vlan_is_mirrored(const struct mirror *m, int vlan)
3873 for (i = 0; i < m->n_vlans; i++) {
3874 if (m->vlans[i] == vlan) {
3882 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3886 for (i = 0; i < m->n_vlans; i++) {
3887 if (port_trunks_vlan(p, m->vlans[i])) {
3895 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3897 struct shash src_ports, dst_ports;
3898 mirror_mask_t mirror_bit;
3899 struct port *out_port;
3905 /* Get output port. */
3906 if (cfg->output_port) {
3907 out_port = port_lookup(m->bridge, cfg->output_port->name);
3909 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3910 m->bridge->name, m->name);
3916 if (cfg->output_vlan) {
3917 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3918 "output vlan; ignoring output vlan",
3919 m->bridge->name, m->name);
3921 } else if (cfg->output_vlan) {
3923 out_vlan = *cfg->output_vlan;
3925 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3926 m->bridge->name, m->name);
3931 shash_init(&src_ports);
3932 shash_init(&dst_ports);
3933 if (cfg->select_all) {
3934 for (i = 0; i < m->bridge->n_ports; i++) {
3935 const char *name = m->bridge->ports[i]->name;
3936 shash_add_once(&src_ports, name, NULL);
3937 shash_add_once(&dst_ports, name, NULL);
3942 /* Get ports, and drop duplicates and ports that don't exist. */
3943 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3945 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3948 /* Get all the vlans, and drop duplicate and invalid vlans. */
3949 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3952 /* Update mirror data. */
3953 if (!shash_equal_keys(&m->src_ports, &src_ports)
3954 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3955 || m->n_vlans != n_vlans
3956 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3957 || m->out_port != out_port
3958 || m->out_vlan != out_vlan) {
3959 bridge_flush(m->bridge);
3961 shash_swap(&m->src_ports, &src_ports);
3962 shash_swap(&m->dst_ports, &dst_ports);
3965 m->n_vlans = n_vlans;
3966 m->out_port = out_port;
3967 m->out_vlan = out_vlan;
3970 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3971 for (i = 0; i < m->bridge->n_ports; i++) {
3972 struct port *port = m->bridge->ports[i];
3974 if (shash_find(&m->src_ports, port->name)
3977 ? port_trunks_any_mirrored_vlan(m, port)
3978 : vlan_is_mirrored(m, port->vlan)))) {
3979 port->src_mirrors |= mirror_bit;
3981 port->src_mirrors &= ~mirror_bit;
3984 if (shash_find(&m->dst_ports, port->name)) {
3985 port->dst_mirrors |= mirror_bit;
3987 port->dst_mirrors &= ~mirror_bit;
3992 shash_destroy(&src_ports);
3993 shash_destroy(&dst_ports);