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.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
40 #include "dynamic-string.h"
46 #include "mac-learning.h"
50 #include "ofp-print.h"
52 #include "ofproto/netflow.h"
53 #include "ofproto/ofproto.h"
54 #include "ovsdb-data.h"
56 #include "poll-loop.h"
57 #include "proc-net-compat.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_reconfigure);
86 struct dst builtin[32];
91 static void dst_set_init(struct dst_set *);
92 static void dst_set_add(struct dst_set *, const struct dst *);
93 static void dst_set_free(struct dst_set *);
96 /* These members are always valid. */
97 struct port *port; /* Containing port. */
98 size_t port_ifidx; /* Index within containing port. */
99 char *name; /* Host network device name. */
100 tag_type tag; /* Tag associated with this interface. */
101 long long delay_expires; /* Time after which 'enabled' may change. */
103 /* These members are valid only after bridge_reconfigure() causes them to
105 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
106 int dp_ifidx; /* Index within kernel datapath. */
107 struct netdev *netdev; /* Network device. */
108 bool enabled; /* May be chosen for flows? */
109 const char *type; /* Usually same as cfg->type. */
110 struct cfm *cfm; /* Connectivity Fault Management */
111 const struct ovsrec_interface *cfg;
114 #define BOND_MASK 0xff
116 int iface_idx; /* Index of assigned iface, or -1 if none. */
117 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
118 tag_type iface_tag; /* Tag associated with iface_idx. */
122 BM_SLB, /* Source Load Balance (Default). */
123 BM_AB /* Active Backup. */
126 #define MAX_MIRRORS 32
127 typedef uint32_t mirror_mask_t;
128 #define MIRROR_MASK_C(X) UINT32_C(X)
129 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
131 struct bridge *bridge;
134 struct uuid uuid; /* UUID of this "mirror" record in database. */
136 /* Selection criteria. */
137 struct shash src_ports; /* Name is port name; data is always NULL. */
138 struct shash dst_ports; /* Name is port name; data is always NULL. */
143 struct port *out_port;
147 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
149 struct bridge *bridge;
151 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
152 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
153 * NULL if all VLANs are trunked. */
154 const struct ovsrec_port *cfg;
157 /* An ordinary bridge port has 1 interface.
158 * A bridge port for bonding has at least 2 interfaces. */
159 struct iface **ifaces;
160 size_t n_ifaces, allocated_ifaces;
163 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
164 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
165 tag_type active_iface_tag; /* Tag for bcast flows. */
166 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
167 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
168 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
169 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
170 bool miimon; /* Use miimon instead of carrier? */
171 long long int bond_miimon_interval; /* Miimon status refresh interval. */
172 long long int bond_miimon_next_update; /* Time of next miimon update. */
173 long long int bond_next_fake_iface_update; /* Time of next update. */
174 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
176 /* SLB specific bonding info. */
177 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
178 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
179 long long int bond_next_rebalance; /* Next rebalancing time. */
181 /* Port mirroring info. */
182 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
183 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
184 bool is_mirror_output_port; /* Does port mirroring send frames here? */
188 struct list node; /* Node in global list of bridges. */
189 char *name; /* User-specified arbitrary name. */
190 struct mac_learning *ml; /* MAC learning table. */
191 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
192 const struct ovsrec_bridge *cfg;
194 /* OpenFlow switch processing. */
195 struct ofproto *ofproto; /* OpenFlow switch. */
197 /* Kernel datapath information. */
198 struct dpif *dpif; /* Datapath. */
199 struct hmap ifaces; /* Contains "struct iface"s. */
203 size_t n_ports, allocated_ports;
204 struct shash iface_by_name; /* "struct iface"s indexed by name. */
205 struct shash port_by_name; /* "struct port"s indexed by name. */
208 bool has_bonded_ports;
213 /* Port mirroring. */
214 struct mirror *mirrors[MAX_MIRRORS];
217 /* List of all bridges. */
218 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
220 /* OVSDB IDL used to obtain configuration. */
221 static struct ovsdb_idl *idl;
223 /* Each time this timer expires, the bridge fetches systems and interface
224 * statistics and pushes them into the database. */
225 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
226 static long long int stats_timer = LLONG_MIN;
228 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
229 static void bridge_destroy(struct bridge *);
230 static struct bridge *bridge_lookup(const char *name);
231 static unixctl_cb_func bridge_unixctl_dump_flows;
232 static unixctl_cb_func bridge_unixctl_reconnect;
233 static int bridge_run_one(struct bridge *);
234 static size_t bridge_get_controllers(const struct bridge *br,
235 struct ovsrec_controller ***controllersp);
236 static void bridge_reconfigure_one(struct bridge *);
237 static void bridge_reconfigure_remotes(struct bridge *,
238 const struct sockaddr_in *managers,
240 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
241 static void bridge_fetch_dp_ifaces(struct bridge *);
242 static void bridge_flush(struct bridge *);
243 static void bridge_pick_local_hw_addr(struct bridge *,
244 uint8_t ea[ETH_ADDR_LEN],
245 struct iface **hw_addr_iface);
246 static uint64_t bridge_pick_datapath_id(struct bridge *,
247 const uint8_t bridge_ea[ETH_ADDR_LEN],
248 struct iface *hw_addr_iface);
249 static struct iface *bridge_get_local_iface(struct bridge *);
250 static uint64_t dpid_from_hash(const void *, size_t nbytes);
252 static unixctl_cb_func bridge_unixctl_fdb_show;
254 static void bond_init(void);
255 static void bond_run(struct bridge *);
256 static void bond_wait(struct bridge *);
257 static void bond_rebalance_port(struct port *);
258 static void bond_send_learning_packets(struct port *);
259 static void bond_enable_slave(struct iface *iface, bool enable);
261 static struct port *port_create(struct bridge *, const char *name);
262 static void port_reconfigure(struct port *, const struct ovsrec_port *);
263 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
264 static void port_destroy(struct port *);
265 static struct port *port_lookup(const struct bridge *, const char *name);
266 static struct iface *port_lookup_iface(const struct port *, const char *name);
267 static struct port *port_from_dp_ifidx(const struct bridge *,
269 static void port_update_bond_compat(struct port *);
270 static void port_update_vlan_compat(struct port *);
271 static void port_update_bonding(struct port *);
273 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
274 static void mirror_destroy(struct mirror *);
275 static void mirror_reconfigure(struct bridge *);
276 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
277 static bool vlan_is_mirrored(const struct mirror *, int vlan);
279 static struct iface *iface_create(struct port *port,
280 const struct ovsrec_interface *if_cfg);
281 static void iface_destroy(struct iface *);
282 static struct iface *iface_lookup(const struct bridge *, const char *name);
283 static struct iface *iface_from_dp_ifidx(const struct bridge *,
285 static void iface_set_mac(struct iface *);
286 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
287 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
288 static void iface_update_cfm(struct iface *);
289 static void iface_refresh_cfm_stats(struct iface *iface);
290 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
292 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
294 static void shash_to_ovs_idl_map(struct shash *,
295 char ***keys, char ***values, size_t *n);
298 /* Hooks into ofproto processing. */
299 static struct ofhooks bridge_ofhooks;
301 /* Public functions. */
303 /* Initializes the bridge module, configuring it to obtain its configuration
304 * from an OVSDB server accessed over 'remote', which should be a string in a
305 * form acceptable to ovsdb_idl_create(). */
307 bridge_init(const char *remote)
309 /* Create connection to database. */
310 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
312 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
313 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
314 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
316 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
318 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
319 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
321 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
322 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
323 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
325 /* Register unixctl commands. */
326 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
327 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
329 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
337 struct bridge *br, *next_br;
339 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
342 ovsdb_idl_destroy(idl);
345 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
346 * but for which the ovs-vswitchd configuration 'cfg' is required. */
348 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
350 static bool already_configured_once;
351 struct svec bridge_names;
352 struct svec dpif_names, dpif_types;
355 /* Only do this once per ovs-vswitchd run. */
356 if (already_configured_once) {
359 already_configured_once = true;
361 stats_timer = time_msec() + STATS_INTERVAL;
363 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
364 svec_init(&bridge_names);
365 for (i = 0; i < cfg->n_bridges; i++) {
366 svec_add(&bridge_names, cfg->bridges[i]->name);
368 svec_sort(&bridge_names);
370 /* Iterate over all system dpifs and delete any of them that do not appear
372 svec_init(&dpif_names);
373 svec_init(&dpif_types);
374 dp_enumerate_types(&dpif_types);
375 for (i = 0; i < dpif_types.n; i++) {
380 dp_enumerate_names(dpif_types.names[i], &dpif_names);
382 /* For each dpif... */
383 for (j = 0; j < dpif_names.n; j++) {
384 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
386 struct svec all_names;
389 /* ...check whether any of its names is in 'bridge_names'. */
390 svec_init(&all_names);
391 dpif_get_all_names(dpif, &all_names);
392 for (k = 0; k < all_names.n; k++) {
393 if (svec_contains(&bridge_names, all_names.names[k])) {
398 /* No. Delete the dpif. */
402 svec_destroy(&all_names);
407 svec_destroy(&bridge_names);
408 svec_destroy(&dpif_names);
409 svec_destroy(&dpif_types);
412 /* Callback for iterate_and_prune_ifaces(). */
414 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
416 if (!iface->netdev) {
417 /* We already reported a related error, don't bother duplicating it. */
421 if (iface->dp_ifidx < 0) {
422 VLOG_ERR("%s interface not in %s, dropping",
423 iface->name, dpif_name(br->dpif));
427 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
428 iface->name, iface->dp_ifidx);
432 /* Callback for iterate_and_prune_ifaces(). */
434 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
435 void *aux OVS_UNUSED)
437 /* Set policing attributes. */
438 netdev_set_policing(iface->netdev,
439 iface->cfg->ingress_policing_rate,
440 iface->cfg->ingress_policing_burst);
442 /* Set MAC address of internal interfaces other than the local
444 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
445 iface_set_mac(iface);
451 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
452 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
453 * deletes from 'br' any ports that no longer have any interfaces. */
455 iterate_and_prune_ifaces(struct bridge *br,
456 bool (*cb)(struct bridge *, struct iface *,
462 for (i = 0; i < br->n_ports; ) {
463 struct port *port = br->ports[i];
464 for (j = 0; j < port->n_ifaces; ) {
465 struct iface *iface = port->ifaces[j];
466 if (cb(br, iface, aux)) {
469 iface_set_ofport(iface->cfg, -1);
470 iface_destroy(iface);
474 if (port->n_ifaces) {
477 VLOG_ERR("%s port has no interfaces, dropping", port->name);
483 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
484 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
485 * responsible for freeing '*managersp' (with free()).
487 * You may be asking yourself "why does ovs-vswitchd care?", because
488 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
489 * should not be and in fact is not directly involved in that. But
490 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
491 * it has to tell in-band control where the managers are to enable that.
492 * (Thus, only managers connected in-band are collected.)
495 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
496 struct sockaddr_in **managersp, size_t *n_managersp)
498 struct sockaddr_in *managers = NULL;
499 size_t n_managers = 0;
500 struct shash targets;
503 /* Collect all of the potential targets, as the union of the "managers"
504 * column and the "targets" columns of the rows pointed to by
505 * "manager_options", excluding any that are out-of-band. */
506 shash_init(&targets);
507 for (i = 0; i < ovs_cfg->n_managers; i++) {
508 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
510 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
511 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
513 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
514 shash_find_and_delete(&targets, m->target);
516 shash_add_once(&targets, m->target, NULL);
520 /* Now extract the targets' IP addresses. */
521 if (!shash_is_empty(&targets)) {
522 struct shash_node *node;
524 managers = xmalloc(shash_count(&targets) * sizeof *managers);
525 SHASH_FOR_EACH (node, &targets) {
526 const char *target = node->name;
527 struct sockaddr_in *sin = &managers[n_managers];
529 if ((!strncmp(target, "tcp:", 4)
530 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
531 (!strncmp(target, "ssl:", 4)
532 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
537 shash_destroy(&targets);
539 *managersp = managers;
540 *n_managersp = n_managers;
544 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
546 struct shash old_br, new_br;
547 struct shash_node *node;
548 struct bridge *br, *next;
549 struct sockaddr_in *managers;
552 int sflow_bridge_number;
554 COVERAGE_INC(bridge_reconfigure);
556 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
558 /* Collect old and new bridges. */
561 LIST_FOR_EACH (br, node, &all_bridges) {
562 shash_add(&old_br, br->name, br);
564 for (i = 0; i < ovs_cfg->n_bridges; i++) {
565 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
566 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
567 VLOG_WARN("more than one bridge named %s", br_cfg->name);
571 /* Get rid of deleted bridges and add new bridges. */
572 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
573 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
580 SHASH_FOR_EACH (node, &new_br) {
581 const char *br_name = node->name;
582 const struct ovsrec_bridge *br_cfg = node->data;
583 br = shash_find_data(&old_br, br_name);
585 /* If the bridge datapath type has changed, we need to tear it
586 * down and recreate. */
587 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
589 bridge_create(br_cfg);
592 bridge_create(br_cfg);
595 shash_destroy(&old_br);
596 shash_destroy(&new_br);
598 /* Reconfigure all bridges. */
599 LIST_FOR_EACH (br, node, &all_bridges) {
600 bridge_reconfigure_one(br);
603 /* Add and delete ports on all datapaths.
605 * The kernel will reject any attempt to add a given port to a datapath if
606 * that port already belongs to a different datapath, so we must do all
607 * port deletions before any port additions. */
608 LIST_FOR_EACH (br, node, &all_bridges) {
609 struct odp_port *dpif_ports;
611 struct shash want_ifaces;
613 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
614 bridge_get_all_ifaces(br, &want_ifaces);
615 for (i = 0; i < n_dpif_ports; i++) {
616 const struct odp_port *p = &dpif_ports[i];
617 if (!shash_find(&want_ifaces, p->devname)
618 && strcmp(p->devname, br->name)) {
619 int retval = dpif_port_del(br->dpif, p->port);
621 VLOG_ERR("failed to remove %s interface from %s: %s",
622 p->devname, dpif_name(br->dpif),
627 shash_destroy(&want_ifaces);
630 LIST_FOR_EACH (br, node, &all_bridges) {
631 struct odp_port *dpif_ports;
633 struct shash cur_ifaces, want_ifaces;
635 /* Get the set of interfaces currently in this datapath. */
636 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
637 shash_init(&cur_ifaces);
638 for (i = 0; i < n_dpif_ports; i++) {
639 const char *name = dpif_ports[i].devname;
640 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
643 /* Get the set of interfaces we want on this datapath. */
644 bridge_get_all_ifaces(br, &want_ifaces);
646 hmap_clear(&br->ifaces);
647 SHASH_FOR_EACH (node, &want_ifaces) {
648 const char *if_name = node->name;
649 struct iface *iface = node->data;
650 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
651 const char *type = iface ? iface->type : "internal";
654 /* If we have a port or a netdev already, and it's not the type we
655 * want, then delete the port (if any) and close the netdev (if
657 if ((dpif_port && strcmp(dpif_port->type, type))
658 || (iface && iface->netdev
659 && strcmp(type, netdev_get_type(iface->netdev)))) {
661 error = ofproto_port_del(br->ofproto, dpif_port->port);
668 netdev_close(iface->netdev);
669 iface->netdev = NULL;
673 /* If the port doesn't exist or we don't have the netdev open,
674 * we need to do more work. */
675 if (!dpif_port || (iface && !iface->netdev)) {
676 struct netdev_options options;
677 struct netdev *netdev;
680 /* First open the network device. */
681 options.name = if_name;
683 options.args = &args;
684 options.ethertype = NETDEV_ETH_TYPE_NONE;
688 shash_from_ovs_idl_map(iface->cfg->key_options,
689 iface->cfg->value_options,
690 iface->cfg->n_options, &args);
692 error = netdev_open(&options, &netdev);
693 shash_destroy(&args);
696 VLOG_WARN("could not open network device %s (%s)",
697 if_name, strerror(error));
701 /* Then add the port if we haven't already. */
703 error = dpif_port_add(br->dpif, netdev, NULL);
705 netdev_close(netdev);
706 if (error == EFBIG) {
707 VLOG_ERR("ran out of valid port numbers on %s",
708 dpif_name(br->dpif));
711 VLOG_ERR("failed to add %s interface to %s: %s",
712 if_name, dpif_name(br->dpif),
719 /* Update 'iface'. */
721 iface->netdev = netdev;
722 iface->enabled = netdev_get_carrier(iface->netdev);
724 } else if (iface && iface->netdev) {
728 shash_from_ovs_idl_map(iface->cfg->key_options,
729 iface->cfg->value_options,
730 iface->cfg->n_options, &args);
731 netdev_reconfigure(iface->netdev, &args);
732 shash_destroy(&args);
736 shash_destroy(&cur_ifaces);
737 shash_destroy(&want_ifaces);
739 sflow_bridge_number = 0;
740 LIST_FOR_EACH (br, node, &all_bridges) {
743 struct iface *local_iface;
744 struct iface *hw_addr_iface;
747 bridge_fetch_dp_ifaces(br);
749 iterate_and_prune_ifaces(br, check_iface, NULL);
751 /* Pick local port hardware address, datapath ID. */
752 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
753 local_iface = bridge_get_local_iface(br);
755 int error = netdev_set_etheraddr(local_iface->netdev, ea);
757 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
758 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
759 "Ethernet address: %s",
760 br->name, strerror(error));
764 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
765 ofproto_set_datapath_id(br->ofproto, dpid);
767 dpid_string = xasprintf("%016"PRIx64, dpid);
768 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
771 /* Set NetFlow configuration on this bridge. */
772 if (br->cfg->netflow) {
773 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
774 struct netflow_options opts;
776 memset(&opts, 0, sizeof opts);
778 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
779 if (nf_cfg->engine_type) {
780 opts.engine_type = *nf_cfg->engine_type;
782 if (nf_cfg->engine_id) {
783 opts.engine_id = *nf_cfg->engine_id;
786 opts.active_timeout = nf_cfg->active_timeout;
787 if (!opts.active_timeout) {
788 opts.active_timeout = -1;
789 } else if (opts.active_timeout < 0) {
790 VLOG_WARN("bridge %s: active timeout interval set to negative "
791 "value, using default instead (%d seconds)", br->name,
792 NF_ACTIVE_TIMEOUT_DEFAULT);
793 opts.active_timeout = -1;
796 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
797 if (opts.add_id_to_iface) {
798 if (opts.engine_id > 0x7f) {
799 VLOG_WARN("bridge %s: netflow port mangling may conflict "
800 "with another vswitch, choose an engine id less "
801 "than 128", br->name);
803 if (br->n_ports > 508) {
804 VLOG_WARN("bridge %s: netflow port mangling will conflict "
805 "with another port when more than 508 ports are "
810 opts.collectors.n = nf_cfg->n_targets;
811 opts.collectors.names = nf_cfg->targets;
812 if (ofproto_set_netflow(br->ofproto, &opts)) {
813 VLOG_ERR("bridge %s: problem setting netflow collectors",
817 ofproto_set_netflow(br->ofproto, NULL);
820 /* Set sFlow configuration on this bridge. */
821 if (br->cfg->sflow) {
822 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
823 struct ovsrec_controller **controllers;
824 struct ofproto_sflow_options oso;
825 size_t n_controllers;
827 memset(&oso, 0, sizeof oso);
829 oso.targets.n = sflow_cfg->n_targets;
830 oso.targets.names = sflow_cfg->targets;
832 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
833 if (sflow_cfg->sampling) {
834 oso.sampling_rate = *sflow_cfg->sampling;
837 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
838 if (sflow_cfg->polling) {
839 oso.polling_interval = *sflow_cfg->polling;
842 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
843 if (sflow_cfg->header) {
844 oso.header_len = *sflow_cfg->header;
847 oso.sub_id = sflow_bridge_number++;
848 oso.agent_device = sflow_cfg->agent;
850 oso.control_ip = NULL;
851 n_controllers = bridge_get_controllers(br, &controllers);
852 for (i = 0; i < n_controllers; i++) {
853 if (controllers[i]->local_ip) {
854 oso.control_ip = controllers[i]->local_ip;
858 ofproto_set_sflow(br->ofproto, &oso);
860 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
862 ofproto_set_sflow(br->ofproto, NULL);
865 /* Update the controller and related settings. It would be more
866 * straightforward to call this from bridge_reconfigure_one(), but we
867 * can't do it there for two reasons. First, and most importantly, at
868 * that point we don't know the dp_ifidx of any interfaces that have
869 * been added to the bridge (because we haven't actually added them to
870 * the datapath). Second, at that point we haven't set the datapath ID
871 * yet; when a controller is configured, resetting the datapath ID will
872 * immediately disconnect from the controller, so it's better to set
873 * the datapath ID before the controller. */
874 bridge_reconfigure_remotes(br, managers, n_managers);
876 LIST_FOR_EACH (br, node, &all_bridges) {
877 for (i = 0; i < br->n_ports; i++) {
878 struct port *port = br->ports[i];
881 port_update_vlan_compat(port);
882 port_update_bonding(port);
884 for (j = 0; j < port->n_ifaces; j++) {
885 iface_update_qos(port->ifaces[j], port->cfg->qos);
889 LIST_FOR_EACH (br, node, &all_bridges) {
890 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
893 LIST_FOR_EACH (br, node, &all_bridges) {
895 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
896 iface_update_cfm(iface);
904 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
905 const struct ovsdb_idl_column *column,
908 const struct ovsdb_datum *datum;
909 union ovsdb_atom atom;
912 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
913 atom.string = (char *) key;
914 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
915 return idx == UINT_MAX ? NULL : datum->values[idx].string;
919 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
921 return get_ovsrec_key_value(&br_cfg->header_,
922 &ovsrec_bridge_col_other_config, key);
926 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
927 struct iface **hw_addr_iface)
933 *hw_addr_iface = NULL;
935 /* Did the user request a particular MAC? */
936 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
937 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
938 if (eth_addr_is_multicast(ea)) {
939 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
940 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
941 } else if (eth_addr_is_zero(ea)) {
942 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
948 /* Otherwise choose the minimum non-local MAC address among all of the
950 memset(ea, 0xff, sizeof ea);
951 for (i = 0; i < br->n_ports; i++) {
952 struct port *port = br->ports[i];
953 uint8_t iface_ea[ETH_ADDR_LEN];
956 /* Mirror output ports don't participate. */
957 if (port->is_mirror_output_port) {
961 /* Choose the MAC address to represent the port. */
962 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
963 /* Find the interface with this Ethernet address (if any) so that
964 * we can provide the correct devname to the caller. */
966 for (j = 0; j < port->n_ifaces; j++) {
967 struct iface *candidate = port->ifaces[j];
968 uint8_t candidate_ea[ETH_ADDR_LEN];
969 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
970 && eth_addr_equals(iface_ea, candidate_ea)) {
975 /* Choose the interface whose MAC address will represent the port.
976 * The Linux kernel bonding code always chooses the MAC address of
977 * the first slave added to a bond, and the Fedora networking
978 * scripts always add slaves to a bond in alphabetical order, so
979 * for compatibility we choose the interface with the name that is
980 * first in alphabetical order. */
981 iface = port->ifaces[0];
982 for (j = 1; j < port->n_ifaces; j++) {
983 struct iface *candidate = port->ifaces[j];
984 if (strcmp(candidate->name, iface->name) < 0) {
989 /* The local port doesn't count (since we're trying to choose its
990 * MAC address anyway). */
991 if (iface->dp_ifidx == ODPP_LOCAL) {
996 error = netdev_get_etheraddr(iface->netdev, iface_ea);
998 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
999 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1000 iface->name, strerror(error));
1005 /* Compare against our current choice. */
1006 if (!eth_addr_is_multicast(iface_ea) &&
1007 !eth_addr_is_local(iface_ea) &&
1008 !eth_addr_is_reserved(iface_ea) &&
1009 !eth_addr_is_zero(iface_ea) &&
1010 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
1012 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1013 *hw_addr_iface = iface;
1016 if (eth_addr_is_multicast(ea)) {
1017 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1018 *hw_addr_iface = NULL;
1019 VLOG_WARN("bridge %s: using default bridge Ethernet "
1020 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1022 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1023 br->name, ETH_ADDR_ARGS(ea));
1027 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1028 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1029 * an interface on 'br', then that interface must be passed in as
1030 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1031 * 'hw_addr_iface' must be passed in as a null pointer. */
1033 bridge_pick_datapath_id(struct bridge *br,
1034 const uint8_t bridge_ea[ETH_ADDR_LEN],
1035 struct iface *hw_addr_iface)
1038 * The procedure for choosing a bridge MAC address will, in the most
1039 * ordinary case, also choose a unique MAC that we can use as a datapath
1040 * ID. In some special cases, though, multiple bridges will end up with
1041 * the same MAC address. This is OK for the bridges, but it will confuse
1042 * the OpenFlow controller, because each datapath needs a unique datapath
1045 * Datapath IDs must be unique. It is also very desirable that they be
1046 * stable from one run to the next, so that policy set on a datapath
1049 const char *datapath_id;
1052 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1053 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1057 if (hw_addr_iface) {
1059 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1061 * A bridge whose MAC address is taken from a VLAN network device
1062 * (that is, a network device created with vconfig(8) or similar
1063 * tool) will have the same MAC address as a bridge on the VLAN
1064 * device's physical network device.
1066 * Handle this case by hashing the physical network device MAC
1067 * along with the VLAN identifier.
1069 uint8_t buf[ETH_ADDR_LEN + 2];
1070 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1071 buf[ETH_ADDR_LEN] = vlan >> 8;
1072 buf[ETH_ADDR_LEN + 1] = vlan;
1073 return dpid_from_hash(buf, sizeof buf);
1076 * Assume that this bridge's MAC address is unique, since it
1077 * doesn't fit any of the cases we handle specially.
1082 * A purely internal bridge, that is, one that has no non-virtual
1083 * network devices on it at all, is more difficult because it has no
1084 * natural unique identifier at all.
1086 * When the host is a XenServer, we handle this case by hashing the
1087 * host's UUID with the name of the bridge. Names of bridges are
1088 * persistent across XenServer reboots, although they can be reused if
1089 * an internal network is destroyed and then a new one is later
1090 * created, so this is fairly effective.
1092 * When the host is not a XenServer, we punt by using a random MAC
1093 * address on each run.
1095 const char *host_uuid = xenserver_get_host_uuid();
1097 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1098 dpid = dpid_from_hash(combined, strlen(combined));
1104 return eth_addr_to_uint64(bridge_ea);
1108 dpid_from_hash(const void *data, size_t n)
1110 uint8_t hash[SHA1_DIGEST_SIZE];
1112 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1113 sha1_bytes(data, n, hash);
1114 eth_addr_mark_random(hash);
1115 return eth_addr_to_uint64(hash);
1119 iface_refresh_status(struct iface *iface)
1125 if (!netdev_get_status(iface->netdev, &sh)) {
1127 char **keys, **values;
1129 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1130 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1135 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1138 shash_destroy_free_data(&sh);
1142 iface_refresh_cfm_stats(struct iface *iface)
1146 const struct ovsrec_monitor *mon;
1148 mon = iface->cfg->monitor;
1155 for (i = 0; i < mon->n_remote_mps; i++) {
1156 const struct ovsrec_maintenance_point *mp;
1157 const struct remote_mp *rmp;
1159 mp = mon->remote_mps[i];
1160 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1162 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1165 if (hmap_is_empty(&cfm->x_remote_mps)) {
1166 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1169 struct remote_mp *rmp;
1170 int64_t *x_remote_mps;
1172 length = hmap_count(&cfm->x_remote_mps);
1173 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1176 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1177 x_remote_mps[i++] = rmp->mpid;
1180 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1184 if (hmap_is_empty(&cfm->x_remote_maids)) {
1185 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1188 char **x_remote_maids;
1189 struct remote_maid *rmaid;
1191 length = hmap_count(&cfm->x_remote_maids);
1192 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1195 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1198 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1200 for (j = 0; j < CCM_MAID_LEN; j++) {
1201 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1206 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1208 for (i = 0; i < length; i++) {
1209 free(x_remote_maids[i]);
1211 free(x_remote_maids);
1214 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1218 iface_refresh_stats(struct iface *iface)
1224 static const struct iface_stat iface_stats[] = {
1225 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1226 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1227 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1228 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1229 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1230 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1231 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1232 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1233 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1234 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1235 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1236 { "collisions", offsetof(struct netdev_stats, collisions) },
1238 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1239 const struct iface_stat *s;
1241 char *keys[N_STATS];
1242 int64_t values[N_STATS];
1245 struct netdev_stats stats;
1247 /* Intentionally ignore return value, since errors will set 'stats' to
1248 * all-1s, and we will deal with that correctly below. */
1249 netdev_get_stats(iface->netdev, &stats);
1252 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1253 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1254 if (value != UINT64_MAX) {
1261 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1265 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1267 struct ovsdb_datum datum;
1271 get_system_stats(&stats);
1273 ovsdb_datum_from_shash(&datum, &stats);
1274 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1281 const struct ovsrec_open_vswitch *cfg;
1283 bool datapath_destroyed;
1284 bool database_changed;
1287 /* Let each bridge do the work that it needs to do. */
1288 datapath_destroyed = false;
1289 LIST_FOR_EACH (br, node, &all_bridges) {
1290 int error = bridge_run_one(br);
1292 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1293 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1294 "forcing reconfiguration", br->name);
1295 datapath_destroyed = true;
1299 /* (Re)configure if necessary. */
1300 database_changed = ovsdb_idl_run(idl);
1301 cfg = ovsrec_open_vswitch_first(idl);
1302 if (database_changed || datapath_destroyed) {
1304 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1306 bridge_configure_once(cfg);
1307 bridge_reconfigure(cfg);
1309 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1310 ovsdb_idl_txn_commit(txn);
1311 ovsdb_idl_txn_destroy(txn); /* XXX */
1313 /* We still need to reconfigure to avoid dangling pointers to
1314 * now-destroyed ovsrec structures inside bridge data. */
1315 static const struct ovsrec_open_vswitch null_cfg;
1317 bridge_reconfigure(&null_cfg);
1322 /* Re-configure SSL. We do this on every trip through the main loop,
1323 * instead of just when the database changes, because the contents of the
1324 * key and certificate files can change without the database changing. */
1325 if (cfg && cfg->ssl) {
1326 const struct ovsrec_ssl *ssl = cfg->ssl;
1328 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1329 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1333 /* Refresh system and interface stats if necessary. */
1334 if (time_msec() >= stats_timer) {
1336 struct ovsdb_idl_txn *txn;
1338 txn = ovsdb_idl_txn_create(idl);
1339 LIST_FOR_EACH (br, node, &all_bridges) {
1342 for (i = 0; i < br->n_ports; i++) {
1343 struct port *port = br->ports[i];
1346 for (j = 0; j < port->n_ifaces; j++) {
1347 struct iface *iface = port->ifaces[j];
1348 iface_refresh_stats(iface);
1349 iface_refresh_cfm_stats(iface);
1350 iface_refresh_status(iface);
1354 refresh_system_stats(cfg);
1355 ovsdb_idl_txn_commit(txn);
1356 ovsdb_idl_txn_destroy(txn); /* XXX */
1359 stats_timer = time_msec() + STATS_INTERVAL;
1367 struct iface *iface;
1369 LIST_FOR_EACH (br, node, &all_bridges) {
1370 ofproto_wait(br->ofproto);
1371 if (ofproto_has_primary_controller(br->ofproto)) {
1375 mac_learning_wait(br->ml);
1378 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1380 cfm_wait(iface->cfm);
1384 ovsdb_idl_wait(idl);
1385 poll_timer_wait_until(stats_timer);
1388 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1389 * configuration changes. */
1391 bridge_flush(struct bridge *br)
1393 COVERAGE_INC(bridge_flush);
1395 mac_learning_flush(br->ml);
1398 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1399 * such interface. */
1400 static struct iface *
1401 bridge_get_local_iface(struct bridge *br)
1405 for (i = 0; i < br->n_ports; i++) {
1406 struct port *port = br->ports[i];
1407 for (j = 0; j < port->n_ifaces; j++) {
1408 struct iface *iface = port->ifaces[j];
1409 if (iface->dp_ifidx == ODPP_LOCAL) {
1418 /* Bridge unixctl user interface functions. */
1420 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1421 const char *args, void *aux OVS_UNUSED)
1423 struct ds ds = DS_EMPTY_INITIALIZER;
1424 const struct bridge *br;
1425 const struct mac_entry *e;
1427 br = bridge_lookup(args);
1429 unixctl_command_reply(conn, 501, "no such bridge");
1433 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1434 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1435 if (e->port < 0 || e->port >= br->n_ports) {
1438 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1439 br->ports[e->port]->ifaces[0]->dp_ifidx,
1440 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1442 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1446 /* Bridge reconfiguration functions. */
1447 static struct bridge *
1448 bridge_create(const struct ovsrec_bridge *br_cfg)
1453 assert(!bridge_lookup(br_cfg->name));
1454 br = xzalloc(sizeof *br);
1456 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1462 dpif_flow_flush(br->dpif);
1464 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1467 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1469 dpif_delete(br->dpif);
1470 dpif_close(br->dpif);
1475 br->name = xstrdup(br_cfg->name);
1477 br->ml = mac_learning_create();
1478 eth_addr_nicira_random(br->default_ea);
1480 hmap_init(&br->ifaces);
1482 shash_init(&br->port_by_name);
1483 shash_init(&br->iface_by_name);
1487 list_push_back(&all_bridges, &br->node);
1489 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1495 bridge_destroy(struct bridge *br)
1500 while (br->n_ports > 0) {
1501 port_destroy(br->ports[br->n_ports - 1]);
1503 list_remove(&br->node);
1504 error = dpif_delete(br->dpif);
1505 if (error && error != ENOENT) {
1506 VLOG_ERR("failed to delete %s: %s",
1507 dpif_name(br->dpif), strerror(error));
1509 dpif_close(br->dpif);
1510 ofproto_destroy(br->ofproto);
1511 mac_learning_destroy(br->ml);
1512 hmap_destroy(&br->ifaces);
1513 shash_destroy(&br->port_by_name);
1514 shash_destroy(&br->iface_by_name);
1521 static struct bridge *
1522 bridge_lookup(const char *name)
1526 LIST_FOR_EACH (br, node, &all_bridges) {
1527 if (!strcmp(br->name, name)) {
1534 /* Handle requests for a listing of all flows known by the OpenFlow
1535 * stack, including those normally hidden. */
1537 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1538 const char *args, void *aux OVS_UNUSED)
1543 br = bridge_lookup(args);
1545 unixctl_command_reply(conn, 501, "Unknown bridge");
1550 ofproto_get_all_flows(br->ofproto, &results);
1552 unixctl_command_reply(conn, 200, ds_cstr(&results));
1553 ds_destroy(&results);
1556 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1557 * connections and reconnect. If BRIDGE is not specified, then all bridges
1558 * drop their controller connections and reconnect. */
1560 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1561 const char *args, void *aux OVS_UNUSED)
1564 if (args[0] != '\0') {
1565 br = bridge_lookup(args);
1567 unixctl_command_reply(conn, 501, "Unknown bridge");
1570 ofproto_reconnect_controllers(br->ofproto);
1572 LIST_FOR_EACH (br, node, &all_bridges) {
1573 ofproto_reconnect_controllers(br->ofproto);
1576 unixctl_command_reply(conn, 200, NULL);
1580 bridge_run_one(struct bridge *br)
1583 struct iface *iface;
1585 error = ofproto_run1(br->ofproto);
1590 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1593 error = ofproto_run2(br->ofproto, br->flush);
1596 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1597 struct ofpbuf *packet;
1603 packet = cfm_run(iface->cfm);
1605 iface_send_packet(iface, packet);
1606 ofpbuf_uninit(packet);
1615 bridge_get_controllers(const struct bridge *br,
1616 struct ovsrec_controller ***controllersp)
1618 struct ovsrec_controller **controllers;
1619 size_t n_controllers;
1621 controllers = br->cfg->controller;
1622 n_controllers = br->cfg->n_controller;
1624 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1630 *controllersp = controllers;
1632 return n_controllers;
1636 bridge_reconfigure_one(struct bridge *br)
1638 struct shash old_ports, new_ports;
1639 struct svec snoops, old_snoops;
1640 struct shash_node *node;
1641 enum ofproto_fail_mode fail_mode;
1644 /* Collect old ports. */
1645 shash_init(&old_ports);
1646 for (i = 0; i < br->n_ports; i++) {
1647 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1650 /* Collect new ports. */
1651 shash_init(&new_ports);
1652 for (i = 0; i < br->cfg->n_ports; i++) {
1653 const char *name = br->cfg->ports[i]->name;
1654 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1655 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1660 /* If we have a controller, then we need a local port. Complain if the
1661 * user didn't specify one.
1663 * XXX perhaps we should synthesize a port ourselves in this case. */
1664 if (bridge_get_controllers(br, NULL)) {
1665 char local_name[IF_NAMESIZE];
1668 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1669 local_name, sizeof local_name);
1670 if (!error && !shash_find(&new_ports, local_name)) {
1671 VLOG_WARN("bridge %s: controller specified but no local port "
1672 "(port named %s) defined",
1673 br->name, local_name);
1677 /* Get rid of deleted ports.
1678 * Get rid of deleted interfaces on ports that still exist. */
1679 SHASH_FOR_EACH (node, &old_ports) {
1680 struct port *port = node->data;
1681 const struct ovsrec_port *port_cfg;
1683 port_cfg = shash_find_data(&new_ports, node->name);
1687 port_del_ifaces(port, port_cfg);
1691 /* Create new ports.
1692 * Add new interfaces to existing ports.
1693 * Reconfigure existing ports. */
1694 SHASH_FOR_EACH (node, &new_ports) {
1695 struct port *port = shash_find_data(&old_ports, node->name);
1697 port = port_create(br, node->name);
1700 port_reconfigure(port, node->data);
1701 if (!port->n_ifaces) {
1702 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1703 br->name, port->name);
1707 shash_destroy(&old_ports);
1708 shash_destroy(&new_ports);
1710 /* Set the fail-mode */
1711 fail_mode = !br->cfg->fail_mode
1712 || !strcmp(br->cfg->fail_mode, "standalone")
1713 ? OFPROTO_FAIL_STANDALONE
1714 : OFPROTO_FAIL_SECURE;
1715 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1716 && !ofproto_has_primary_controller(br->ofproto)) {
1717 ofproto_flush_flows(br->ofproto);
1719 ofproto_set_fail_mode(br->ofproto, fail_mode);
1721 /* Delete all flows if we're switching from connected to standalone or vice
1722 * versa. (XXX Should we delete all flows if we are switching from one
1723 * controller to another?) */
1725 /* Configure OpenFlow controller connection snooping. */
1727 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1728 ovs_rundir(), br->name));
1729 svec_init(&old_snoops);
1730 ofproto_get_snoops(br->ofproto, &old_snoops);
1731 if (!svec_equal(&snoops, &old_snoops)) {
1732 ofproto_set_snoops(br->ofproto, &snoops);
1734 svec_destroy(&snoops);
1735 svec_destroy(&old_snoops);
1737 mirror_reconfigure(br);
1740 /* Initializes 'oc' appropriately as a management service controller for
1743 * The caller must free oc->target when it is no longer needed. */
1745 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1746 struct ofproto_controller *oc)
1748 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1749 oc->max_backoff = 0;
1750 oc->probe_interval = 60;
1751 oc->band = OFPROTO_OUT_OF_BAND;
1752 oc->accept_re = NULL;
1753 oc->update_resolv_conf = false;
1755 oc->burst_limit = 0;
1758 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1760 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1761 struct ofproto_controller *oc)
1763 oc->target = c->target;
1764 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1765 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1766 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1767 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1768 oc->accept_re = c->discover_accept_regex;
1769 oc->update_resolv_conf = c->discover_update_resolv_conf;
1770 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1771 oc->burst_limit = (c->controller_burst_limit
1772 ? *c->controller_burst_limit : 0);
1775 /* Configures the IP stack for 'br''s local interface properly according to the
1776 * configuration in 'c'. */
1778 bridge_configure_local_iface_netdev(struct bridge *br,
1779 struct ovsrec_controller *c)
1781 struct netdev *netdev;
1782 struct in_addr mask, gateway;
1784 struct iface *local_iface;
1787 /* Controller discovery does its own TCP/IP configuration later. */
1788 if (strcmp(c->target, "discover")) {
1792 /* If there's no local interface or no IP address, give up. */
1793 local_iface = bridge_get_local_iface(br);
1794 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1798 /* Bring up the local interface. */
1799 netdev = local_iface->netdev;
1800 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1802 /* Configure the IP address and netmask. */
1803 if (!c->local_netmask
1804 || !inet_aton(c->local_netmask, &mask)
1806 mask.s_addr = guess_netmask(ip.s_addr);
1808 if (!netdev_set_in4(netdev, ip, mask)) {
1809 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1810 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1813 /* Configure the default gateway. */
1814 if (c->local_gateway
1815 && inet_aton(c->local_gateway, &gateway)
1816 && gateway.s_addr) {
1817 if (!netdev_add_router(netdev, gateway)) {
1818 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1819 br->name, IP_ARGS(&gateway.s_addr));
1825 bridge_reconfigure_remotes(struct bridge *br,
1826 const struct sockaddr_in *managers,
1829 const char *disable_ib_str, *queue_id_str;
1830 bool disable_in_band = false;
1833 struct ovsrec_controller **controllers;
1834 size_t n_controllers;
1837 struct ofproto_controller *ocs;
1841 /* Check if we should disable in-band control on this bridge. */
1842 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1843 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1844 disable_in_band = true;
1847 /* Set OpenFlow queue ID for in-band control. */
1848 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1849 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1850 ofproto_set_in_band_queue(br->ofproto, queue_id);
1852 if (disable_in_band) {
1853 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1855 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1857 had_primary = ofproto_has_primary_controller(br->ofproto);
1859 n_controllers = bridge_get_controllers(br, &controllers);
1861 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1864 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1865 for (i = 0; i < n_controllers; i++) {
1866 struct ovsrec_controller *c = controllers[i];
1868 if (!strncmp(c->target, "punix:", 6)
1869 || !strncmp(c->target, "unix:", 5)) {
1870 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1872 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1873 * domain sockets and overwriting arbitrary local files. */
1874 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1875 "\"%s\" due to possibility for remote exploit",
1876 dpif_name(br->dpif), c->target);
1880 bridge_configure_local_iface_netdev(br, c);
1881 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1882 if (disable_in_band) {
1883 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1888 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1889 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1892 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1893 ofproto_flush_flows(br->ofproto);
1896 /* If there are no controllers and the bridge is in standalone
1897 * mode, set up a flow that matches every packet and directs
1898 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1899 * switch is in secure mode and we won't pass any traffic until
1900 * a controller has been defined and it tells us to do so. */
1902 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1903 union ofp_action action;
1904 struct cls_rule rule;
1906 memset(&action, 0, sizeof action);
1907 action.type = htons(OFPAT_OUTPUT);
1908 action.output.len = htons(sizeof action);
1909 action.output.port = htons(OFPP_NORMAL);
1910 cls_rule_init_catchall(&rule, 0);
1911 ofproto_add_flow(br->ofproto, &rule, &action, 1);
1916 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1921 for (i = 0; i < br->n_ports; i++) {
1922 struct port *port = br->ports[i];
1923 for (j = 0; j < port->n_ifaces; j++) {
1924 struct iface *iface = port->ifaces[j];
1925 shash_add_once(ifaces, iface->name, iface);
1927 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1928 shash_add_once(ifaces, port->name, NULL);
1933 /* For robustness, in case the administrator moves around datapath ports behind
1934 * our back, we re-check all the datapath port numbers here.
1936 * This function will set the 'dp_ifidx' members of interfaces that have
1937 * disappeared to -1, so only call this function from a context where those
1938 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1939 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1940 * datapath, which doesn't support UINT16_MAX+1 ports. */
1942 bridge_fetch_dp_ifaces(struct bridge *br)
1944 struct odp_port *dpif_ports;
1945 size_t n_dpif_ports;
1948 /* Reset all interface numbers. */
1949 for (i = 0; i < br->n_ports; i++) {
1950 struct port *port = br->ports[i];
1951 for (j = 0; j < port->n_ifaces; j++) {
1952 struct iface *iface = port->ifaces[j];
1953 iface->dp_ifidx = -1;
1956 hmap_clear(&br->ifaces);
1958 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1959 for (i = 0; i < n_dpif_ports; i++) {
1960 struct odp_port *p = &dpif_ports[i];
1961 struct iface *iface = iface_lookup(br, p->devname);
1963 if (iface->dp_ifidx >= 0) {
1964 VLOG_WARN("%s reported interface %s twice",
1965 dpif_name(br->dpif), p->devname);
1966 } else if (iface_from_dp_ifidx(br, p->port)) {
1967 VLOG_WARN("%s reported interface %"PRIu16" twice",
1968 dpif_name(br->dpif), p->port);
1970 iface->dp_ifidx = p->port;
1971 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1972 hash_int(iface->dp_ifidx, 0));
1975 iface_set_ofport(iface->cfg,
1976 (iface->dp_ifidx >= 0
1977 ? odp_port_to_ofp_port(iface->dp_ifidx)
1984 /* Bridge packet processing functions. */
1987 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
1989 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
1992 static struct bond_entry *
1993 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
1996 assert(port->bond_mode == BM_SLB);
1997 return &port->bond_hash[bond_hash(mac, vlan)];
2001 bond_choose_iface(const struct port *port)
2003 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2004 size_t i, best_down_slave = -1;
2005 long long next_delay_expiration = LLONG_MAX;
2007 for (i = 0; i < port->n_ifaces; i++) {
2008 struct iface *iface = port->ifaces[i];
2010 if (iface->enabled) {
2012 } else if (iface->delay_expires < next_delay_expiration) {
2013 best_down_slave = i;
2014 next_delay_expiration = iface->delay_expires;
2018 if (best_down_slave != -1) {
2019 struct iface *iface = port->ifaces[best_down_slave];
2021 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2022 "since no other interface is up", iface->name,
2023 iface->delay_expires - time_msec());
2024 bond_enable_slave(iface, true);
2027 return best_down_slave;
2031 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2032 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2034 struct iface *iface;
2036 assert(port->n_ifaces);
2037 if (port->n_ifaces == 1) {
2038 iface = port->ifaces[0];
2039 } else if (port->bond_mode == BM_AB) {
2040 if (port->active_iface < 0) {
2041 *tags |= port->no_ifaces_tag;
2044 iface = port->ifaces[port->active_iface];
2045 } else if (port->bond_mode == BM_SLB){
2046 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2047 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2048 || !port->ifaces[e->iface_idx]->enabled) {
2049 /* XXX select interface properly. The current interface selection
2050 * is only good for testing the rebalancing code. */
2051 e->iface_idx = bond_choose_iface(port);
2052 if (e->iface_idx < 0) {
2053 *tags |= port->no_ifaces_tag;
2056 e->iface_tag = tag_create_random();
2057 ((struct port *) port)->bond_compat_is_stale = true;
2059 *tags |= e->iface_tag;
2060 iface = port->ifaces[e->iface_idx];
2064 *dp_ifidx = iface->dp_ifidx;
2065 *tags |= iface->tag; /* Currently only used for bonding. */
2070 bond_link_status_update(struct iface *iface, bool carrier)
2072 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2073 struct port *port = iface->port;
2075 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2076 /* Nothing to do. */
2079 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2080 iface->name, carrier ? "up" : "down");
2081 if (carrier == iface->enabled) {
2082 iface->delay_expires = LLONG_MAX;
2083 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2084 iface->name, carrier ? "disabled" : "enabled");
2085 } else if (carrier && port->active_iface < 0) {
2086 bond_enable_slave(iface, true);
2087 if (port->updelay) {
2088 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2089 "other interface is up", iface->name, port->updelay);
2092 int delay = carrier ? port->updelay : port->downdelay;
2093 iface->delay_expires = time_msec() + delay;
2096 "interface %s: will be %s if it stays %s for %d ms",
2098 carrier ? "enabled" : "disabled",
2099 carrier ? "up" : "down",
2106 bond_choose_active_iface(struct port *port)
2108 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2110 port->active_iface = bond_choose_iface(port);
2111 port->active_iface_tag = tag_create_random();
2112 if (port->active_iface >= 0) {
2113 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2114 port->name, port->ifaces[port->active_iface]->name);
2116 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2122 bond_enable_slave(struct iface *iface, bool enable)
2124 struct port *port = iface->port;
2125 struct bridge *br = port->bridge;
2127 /* This acts as a recursion check. If the act of disabling a slave
2128 * causes a different slave to be enabled, the flag will allow us to
2129 * skip redundant work when we reenter this function. It must be
2130 * cleared on exit to keep things safe with multiple bonds. */
2131 static bool moving_active_iface = false;
2133 iface->delay_expires = LLONG_MAX;
2134 if (enable == iface->enabled) {
2138 iface->enabled = enable;
2139 if (!iface->enabled) {
2140 VLOG_WARN("interface %s: disabled", iface->name);
2141 ofproto_revalidate(br->ofproto, iface->tag);
2142 if (iface->port_ifidx == port->active_iface) {
2143 ofproto_revalidate(br->ofproto,
2144 port->active_iface_tag);
2146 /* Disabling a slave can lead to another slave being immediately
2147 * enabled if there will be no active slaves but one is waiting
2148 * on an updelay. In this case we do not need to run most of the
2149 * code for the newly enabled slave since there was no period
2150 * without an active slave and it is redundant with the disabling
2152 moving_active_iface = true;
2153 bond_choose_active_iface(port);
2155 bond_send_learning_packets(port);
2157 VLOG_WARN("interface %s: enabled", iface->name);
2158 if (port->active_iface < 0 && !moving_active_iface) {
2159 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2160 bond_choose_active_iface(port);
2161 bond_send_learning_packets(port);
2163 iface->tag = tag_create_random();
2166 moving_active_iface = false;
2167 port->bond_compat_is_stale = true;
2170 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2171 * bond interface. */
2173 bond_update_fake_iface_stats(struct port *port)
2175 struct netdev_stats bond_stats;
2176 struct netdev *bond_dev;
2179 memset(&bond_stats, 0, sizeof bond_stats);
2181 for (i = 0; i < port->n_ifaces; i++) {
2182 struct netdev_stats slave_stats;
2184 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2185 /* XXX: We swap the stats here because they are swapped back when
2186 * reported by the internal device. The reason for this is
2187 * internal devices normally represent packets going into the system
2188 * but when used as fake bond device they represent packets leaving
2189 * the system. We really should do this in the internal device
2190 * itself because changing it here reverses the counts from the
2191 * perspective of the switch. However, the internal device doesn't
2192 * know what type of device it represents so we have to do it here
2194 bond_stats.tx_packets += slave_stats.rx_packets;
2195 bond_stats.tx_bytes += slave_stats.rx_bytes;
2196 bond_stats.rx_packets += slave_stats.tx_packets;
2197 bond_stats.rx_bytes += slave_stats.tx_bytes;
2201 if (!netdev_open_default(port->name, &bond_dev)) {
2202 netdev_set_stats(bond_dev, &bond_stats);
2203 netdev_close(bond_dev);
2208 bond_run(struct bridge *br)
2212 for (i = 0; i < br->n_ports; i++) {
2213 struct port *port = br->ports[i];
2215 if (port->n_ifaces >= 2) {
2218 if (port->monitor) {
2219 assert(!port->miimon);
2221 /* Track carrier going up and down on interfaces. */
2222 while (!netdev_monitor_poll(port->monitor, &devname)) {
2223 struct iface *iface;
2225 iface = port_lookup_iface(port, devname);
2227 bool up = netdev_get_carrier(iface->netdev);
2229 bond_link_status_update(iface, up);
2230 port_update_bond_compat(port);
2235 assert(port->miimon);
2237 if (time_msec() >= port->bond_miimon_next_update) {
2238 for (j = 0; j < port->n_ifaces; j++) {
2239 struct iface *iface = port->ifaces[j];
2240 bool up = netdev_get_miimon(iface->netdev);
2242 bond_link_status_update(iface, up);
2243 port_update_bond_compat(port);
2245 port->bond_miimon_next_update = time_msec() +
2246 port->bond_miimon_interval;
2250 for (j = 0; j < port->n_ifaces; j++) {
2251 struct iface *iface = port->ifaces[j];
2252 if (time_msec() >= iface->delay_expires) {
2253 bond_enable_slave(iface, !iface->enabled);
2257 if (port->bond_fake_iface
2258 && time_msec() >= port->bond_next_fake_iface_update) {
2259 bond_update_fake_iface_stats(port);
2260 port->bond_next_fake_iface_update = time_msec() + 1000;
2264 if (port->bond_compat_is_stale) {
2265 port->bond_compat_is_stale = false;
2266 port_update_bond_compat(port);
2272 bond_wait(struct bridge *br)
2276 for (i = 0; i < br->n_ports; i++) {
2277 struct port *port = br->ports[i];
2278 if (port->n_ifaces < 2) {
2282 if (port->monitor) {
2283 netdev_monitor_poll_wait(port->monitor);
2287 poll_timer_wait_until(port->bond_miimon_next_update);
2290 for (j = 0; j < port->n_ifaces; j++) {
2291 struct iface *iface = port->ifaces[j];
2292 if (iface->delay_expires != LLONG_MAX) {
2293 poll_timer_wait_until(iface->delay_expires);
2296 if (port->bond_fake_iface) {
2297 poll_timer_wait_until(port->bond_next_fake_iface_update);
2303 set_dst(struct dst *dst, const struct flow *flow,
2304 const struct port *in_port, const struct port *out_port,
2307 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2308 : in_port->vlan >= 0 ? in_port->vlan
2309 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2310 : vlan_tci_to_vid(flow->vlan_tci));
2311 return choose_output_iface(out_port, flow->dl_src, dst->vlan,
2312 &dst->dp_ifidx, tags);
2316 swap_dst(struct dst *p, struct dst *q)
2318 struct dst tmp = *p;
2323 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2324 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2325 * that we push to the datapath. We could in fact fully sort the array by
2326 * vlan, but in most cases there are at most two different vlan tags so that's
2327 * possibly overkill.) */
2329 partition_dsts(struct dst_set *set, int vlan)
2331 struct dst *first = set->dsts;
2332 struct dst *last = set->dsts + set->n;
2334 while (first != last) {
2336 * - All dsts < first have vlan == 'vlan'.
2337 * - All dsts >= last have vlan != 'vlan'.
2338 * - first < last. */
2339 while (first->vlan == vlan) {
2340 if (++first == last) {
2345 /* Same invariants, plus one additional:
2346 * - first->vlan != vlan.
2348 while (last[-1].vlan != vlan) {
2349 if (--last == first) {
2354 /* Same invariants, plus one additional:
2355 * - last[-1].vlan == vlan.*/
2356 swap_dst(first++, --last);
2361 mirror_mask_ffs(mirror_mask_t mask)
2363 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2368 dst_set_init(struct dst_set *set)
2370 set->dsts = set->builtin;
2372 set->allocated = ARRAY_SIZE(set->builtin);
2376 dst_set_add(struct dst_set *set, const struct dst *dst)
2378 if (set->n >= set->allocated) {
2379 size_t new_allocated;
2380 struct dst *new_dsts;
2382 new_allocated = set->allocated * 2;
2383 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2384 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2388 set->dsts = new_dsts;
2389 set->allocated = new_allocated;
2391 set->dsts[set->n++] = *dst;
2395 dst_set_free(struct dst_set *set)
2397 if (set->dsts != set->builtin) {
2403 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2406 for (i = 0; i < set->n; i++) {
2407 if (set->dsts[i].vlan == test->vlan
2408 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2416 port_trunks_vlan(const struct port *port, uint16_t vlan)
2418 return (port->vlan < 0
2419 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2423 port_includes_vlan(const struct port *port, uint16_t vlan)
2425 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2429 port_is_floodable(const struct port *port)
2433 for (i = 0; i < port->n_ifaces; i++) {
2434 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2435 port->ifaces[i]->dp_ifidx)) {
2443 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2444 const struct port *in_port, const struct port *out_port,
2445 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2447 mirror_mask_t mirrors = in_port->src_mirrors;
2452 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2453 if (flow_vlan == 0) {
2454 flow_vlan = OFP_VLAN_NONE;
2457 if (out_port == FLOOD_PORT) {
2458 for (i = 0; i < br->n_ports; i++) {
2459 struct port *port = br->ports[i];
2461 && port_is_floodable(port)
2462 && port_includes_vlan(port, vlan)
2463 && !port->is_mirror_output_port
2464 && set_dst(&dst, flow, in_port, port, tags)) {
2465 mirrors |= port->dst_mirrors;
2466 dst_set_add(set, &dst);
2469 *nf_output_iface = NF_OUT_FLOOD;
2470 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2471 dst_set_add(set, &dst);
2472 *nf_output_iface = dst.dp_ifidx;
2473 mirrors |= out_port->dst_mirrors;
2477 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2478 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2480 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2481 && !dst_is_duplicate(set, &dst)) {
2482 dst_set_add(set, &dst);
2485 for (i = 0; i < br->n_ports; i++) {
2486 struct port *port = br->ports[i];
2487 if (port_includes_vlan(port, m->out_vlan)
2488 && set_dst(&dst, flow, in_port, port, tags))
2490 if (port->vlan < 0) {
2491 dst.vlan = m->out_vlan;
2493 if (dst_is_duplicate(set, &dst)) {
2497 /* Use the vlan tag on the original flow instead of
2498 * the one passed in the vlan parameter. This ensures
2499 * that we compare the vlan from before any implicit
2500 * tagging tags place. This is necessary because
2501 * dst->vlan is the final vlan, after removing implicit
2503 if (port == in_port && dst.vlan == flow_vlan) {
2504 /* Don't send out input port on same VLAN. */
2507 dst_set_add(set, &dst);
2512 mirrors &= mirrors - 1;
2515 partition_dsts(set, flow_vlan);
2518 static void OVS_UNUSED
2519 print_dsts(const struct dst_set *set)
2523 for (i = 0; i < set->n; i++) {
2524 const struct dst *dst = &set->dsts[i];
2526 printf(">p%"PRIu16, dst->dp_ifidx);
2527 if (dst->vlan != OFP_VLAN_NONE) {
2528 printf("v%"PRIu16, dst->vlan);
2534 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2535 const struct port *in_port, const struct port *out_port,
2536 tag_type *tags, struct ofpbuf *actions,
2537 uint16_t *nf_output_iface)
2544 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2547 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2548 if (cur_vlan == 0) {
2549 cur_vlan = OFP_VLAN_NONE;
2551 for (i = 0; i < set.n; i++) {
2552 const struct dst *dst = &set.dsts[i];
2553 if (dst->vlan != cur_vlan) {
2554 if (dst->vlan == OFP_VLAN_NONE) {
2555 nl_msg_put_flag(actions, ODPAT_STRIP_VLAN);
2558 tci = htons(dst->vlan & VLAN_VID_MASK);
2559 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2560 nl_msg_put_be16(actions, ODPAT_SET_DL_TCI, tci);
2562 cur_vlan = dst->vlan;
2564 nl_msg_put_u32(actions, ODPAT_OUTPUT, dst->dp_ifidx);
2569 /* Returns the effective vlan of a packet, taking into account both the
2570 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2571 * the packet is untagged and -1 indicates it has an invalid header and
2572 * should be dropped. */
2573 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2574 struct port *in_port, bool have_packet)
2576 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2577 if (in_port->vlan >= 0) {
2579 /* XXX support double tagging? */
2581 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2582 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2583 "packet received on port %s configured with "
2584 "implicit VLAN %"PRIu16,
2585 br->name, vlan, in_port->name, in_port->vlan);
2589 vlan = in_port->vlan;
2591 if (!port_includes_vlan(in_port, vlan)) {
2593 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2594 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2595 "packet received on port %s not configured for "
2597 br->name, vlan, in_port->name, vlan);
2606 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2607 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2608 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2610 is_gratuitous_arp(const struct flow *flow)
2612 return (flow->dl_type == htons(ETH_TYPE_ARP)
2613 && eth_addr_is_broadcast(flow->dl_dst)
2614 && (flow->nw_proto == ARP_OP_REPLY
2615 || (flow->nw_proto == ARP_OP_REQUEST
2616 && flow->nw_src == flow->nw_dst)));
2620 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2621 struct port *in_port)
2623 enum grat_arp_lock_type lock_type;
2626 /* We don't want to learn from gratuitous ARP packets that are reflected
2627 * back over bond slaves so we lock the learning table. */
2628 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2629 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2630 GRAT_ARP_LOCK_CHECK;
2632 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2635 /* The log messages here could actually be useful in debugging,
2636 * so keep the rate limit relatively high. */
2637 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2639 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2640 "on port %s in VLAN %d",
2641 br->name, ETH_ADDR_ARGS(flow->dl_src),
2642 in_port->name, vlan);
2643 ofproto_revalidate(br->ofproto, rev_tag);
2647 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2648 * dropped. Returns true if they may be forwarded, false if they should be
2651 * If 'have_packet' is true, it indicates that the caller is processing a
2652 * received packet. If 'have_packet' is false, then the caller is just
2653 * revalidating an existing flow because configuration has changed. Either
2654 * way, 'have_packet' only affects logging (there is no point in logging errors
2655 * during revalidation).
2657 * Sets '*in_portp' to the input port. This will be a null pointer if
2658 * flow->in_port does not designate a known input port (in which case
2659 * is_admissible() returns false).
2661 * When returning true, sets '*vlanp' to the effective VLAN of the input
2662 * packet, as returned by flow_get_vlan().
2664 * May also add tags to '*tags', although the current implementation only does
2665 * so in one special case.
2668 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2669 tag_type *tags, int *vlanp, struct port **in_portp)
2671 struct iface *in_iface;
2672 struct port *in_port;
2675 /* Find the interface and port structure for the received packet. */
2676 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2678 /* No interface? Something fishy... */
2680 /* Odd. A few possible reasons here:
2682 * - We deleted an interface but there are still a few packets
2683 * queued up from it.
2685 * - Someone externally added an interface (e.g. with "ovs-dpctl
2686 * add-if") that we don't know about.
2688 * - Packet arrived on the local port but the local port is not
2689 * one of our bridge ports.
2691 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2693 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2694 "interface %"PRIu16, br->name, flow->in_port);
2700 *in_portp = in_port = in_iface->port;
2701 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2706 /* Drop frames for reserved multicast addresses. */
2707 if (eth_addr_is_reserved(flow->dl_dst)) {
2711 /* Drop frames on ports reserved for mirroring. */
2712 if (in_port->is_mirror_output_port) {
2714 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2715 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2716 "%s, which is reserved exclusively for mirroring",
2717 br->name, in_port->name);
2722 /* Packets received on bonds need special attention to avoid duplicates. */
2723 if (in_port->n_ifaces > 1) {
2725 bool is_grat_arp_locked;
2727 if (eth_addr_is_multicast(flow->dl_dst)) {
2728 *tags |= in_port->active_iface_tag;
2729 if (in_port->active_iface != in_iface->port_ifidx) {
2730 /* Drop all multicast packets on inactive slaves. */
2735 /* Drop all packets for which we have learned a different input
2736 * port, because we probably sent the packet on one slave and got
2737 * it back on the other. Gratuitous ARP packets are an exception
2738 * to this rule: the host has moved to another switch. The exception
2739 * to the exception is if we locked the learning table to avoid
2740 * reflections on bond slaves. If this is the case, just drop the
2742 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2743 &is_grat_arp_locked);
2744 if (src_idx != -1 && src_idx != in_port->port_idx &&
2745 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2753 /* If the composed actions may be applied to any packet in the given 'flow',
2754 * returns true. Otherwise, the actions should only be applied to 'packet', or
2755 * not at all, if 'packet' was NULL. */
2757 process_flow(struct bridge *br, const struct flow *flow,
2758 const struct ofpbuf *packet, struct ofpbuf *actions,
2759 tag_type *tags, uint16_t *nf_output_iface)
2761 struct port *in_port;
2762 struct port *out_port;
2766 /* Check whether we should drop packets in this flow. */
2767 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2772 /* Learn source MAC (but don't try to learn from revalidation). */
2774 update_learning_table(br, flow, vlan, in_port);
2777 /* Determine output port. */
2778 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2780 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2781 out_port = br->ports[out_port_idx];
2782 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2783 /* If we are revalidating but don't have a learning entry then
2784 * eject the flow. Installing a flow that floods packets opens
2785 * up a window of time where we could learn from a packet reflected
2786 * on a bond and blackhole packets before the learning table is
2787 * updated to reflect the correct port. */
2790 out_port = FLOOD_PORT;
2793 /* Don't send packets out their input ports. */
2794 if (in_port == out_port) {
2800 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2808 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2809 struct ofpbuf *actions, tag_type *tags,
2810 uint16_t *nf_output_iface, void *br_)
2812 struct iface *iface;
2813 struct bridge *br = br_;
2815 COVERAGE_INC(bridge_process_flow);
2817 iface = iface_from_dp_ifidx(br, flow->in_port);
2819 if (cfm_should_process_flow(flow)) {
2820 if (packet && iface->cfm) {
2821 cfm_process_heartbeat(iface->cfm, packet);
2826 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2830 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2831 const struct nlattr *actions,
2833 unsigned long long int n_bytes, void *br_)
2835 struct bridge *br = br_;
2836 const struct nlattr *a;
2837 struct port *in_port;
2842 /* Feed information from the active flows back into the learning table to
2843 * ensure that table is always in sync with what is actually flowing
2844 * through the datapath.
2846 * We test that 'tags' is nonzero to ensure that only flows that include an
2847 * OFPP_NORMAL action are used for learning. This works because
2848 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2849 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2850 update_learning_table(br, flow, vlan, in_port);
2853 /* Account for bond slave utilization. */
2854 if (!br->has_bonded_ports) {
2857 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2858 if (nl_attr_type(a) == ODPAT_OUTPUT) {
2859 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2860 if (out_port && out_port->n_ifaces >= 2 &&
2861 out_port->bond_mode == BM_SLB) {
2862 uint16_t vlan = (flow->vlan_tci
2863 ? vlan_tci_to_vid(flow->vlan_tci)
2865 struct bond_entry *e = lookup_bond_entry(out_port,
2866 flow->dl_src, vlan);
2867 e->tx_bytes += n_bytes;
2874 bridge_account_checkpoint_ofhook_cb(void *br_)
2876 struct bridge *br = br_;
2880 if (!br->has_bonded_ports) {
2885 for (i = 0; i < br->n_ports; i++) {
2886 struct port *port = br->ports[i];
2887 if (port->n_ifaces > 1 && port->bond_mode == BM_SLB
2888 && now >= port->bond_next_rebalance) {
2889 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2890 bond_rebalance_port(port);
2895 static struct ofhooks bridge_ofhooks = {
2896 bridge_normal_ofhook_cb,
2897 bridge_account_flow_ofhook_cb,
2898 bridge_account_checkpoint_ofhook_cb,
2901 /* Bonding functions. */
2903 /* Statistics for a single interface on a bonded port, used for load-based
2904 * bond rebalancing. */
2905 struct slave_balance {
2906 struct iface *iface; /* The interface. */
2907 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2909 /* All the "bond_entry"s that are assigned to this interface, in order of
2910 * increasing tx_bytes. */
2911 struct bond_entry **hashes;
2916 bond_mode_to_string(enum bond_mode bm) {
2917 static char *bm_slb = "balance-slb";
2918 static char *bm_ab = "active-backup";
2921 case BM_SLB: return bm_slb;
2922 case BM_AB: return bm_ab;
2929 /* Sorts pointers to pointers to bond_entries in ascending order by the
2930 * interface to which they are assigned, and within a single interface in
2931 * ascending order of bytes transmitted. */
2933 compare_bond_entries(const void *a_, const void *b_)
2935 const struct bond_entry *const *ap = a_;
2936 const struct bond_entry *const *bp = b_;
2937 const struct bond_entry *a = *ap;
2938 const struct bond_entry *b = *bp;
2939 if (a->iface_idx != b->iface_idx) {
2940 return a->iface_idx > b->iface_idx ? 1 : -1;
2941 } else if (a->tx_bytes != b->tx_bytes) {
2942 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2948 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2949 * *descending* order by number of bytes transmitted. */
2951 compare_slave_balance(const void *a_, const void *b_)
2953 const struct slave_balance *a = a_;
2954 const struct slave_balance *b = b_;
2955 if (a->iface->enabled != b->iface->enabled) {
2956 return a->iface->enabled ? -1 : 1;
2957 } else if (a->tx_bytes != b->tx_bytes) {
2958 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2965 swap_bals(struct slave_balance *a, struct slave_balance *b)
2967 struct slave_balance tmp = *a;
2972 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2973 * given that 'p' (and only 'p') might be in the wrong location.
2975 * This function invalidates 'p', since it might now be in a different memory
2978 resort_bals(struct slave_balance *p,
2979 struct slave_balance bals[], size_t n_bals)
2982 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2983 swap_bals(p, p - 1);
2985 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2986 swap_bals(p, p + 1);
2992 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2994 if (VLOG_IS_DBG_ENABLED()) {
2995 struct ds ds = DS_EMPTY_INITIALIZER;
2996 const struct slave_balance *b;
2998 for (b = bals; b < bals + n_bals; b++) {
3002 ds_put_char(&ds, ',');
3004 ds_put_format(&ds, " %s %"PRIu64"kB",
3005 b->iface->name, b->tx_bytes / 1024);
3007 if (!b->iface->enabled) {
3008 ds_put_cstr(&ds, " (disabled)");
3010 if (b->n_hashes > 0) {
3011 ds_put_cstr(&ds, " (");
3012 for (i = 0; i < b->n_hashes; i++) {
3013 const struct bond_entry *e = b->hashes[i];
3015 ds_put_cstr(&ds, " + ");
3017 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3018 e - port->bond_hash, e->tx_bytes / 1024);
3020 ds_put_cstr(&ds, ")");
3023 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3028 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3030 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3033 struct bond_entry *hash = from->hashes[hash_idx];
3034 struct port *port = from->iface->port;
3035 uint64_t delta = hash->tx_bytes;
3037 assert(port->bond_mode == BM_SLB);
3039 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3040 "from %s to %s (now carrying %"PRIu64"kB and "
3041 "%"PRIu64"kB load, respectively)",
3042 port->name, delta / 1024, hash - port->bond_hash,
3043 from->iface->name, to->iface->name,
3044 (from->tx_bytes - delta) / 1024,
3045 (to->tx_bytes + delta) / 1024);
3047 /* Delete element from from->hashes.
3049 * We don't bother to add the element to to->hashes because not only would
3050 * it require more work, the only purpose it would be to allow that hash to
3051 * be migrated to another slave in this rebalancing run, and there is no
3052 * point in doing that. */
3053 if (hash_idx == 0) {
3056 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3057 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3061 /* Shift load away from 'from' to 'to'. */
3062 from->tx_bytes -= delta;
3063 to->tx_bytes += delta;
3065 /* Arrange for flows to be revalidated. */
3066 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3067 hash->iface_idx = to->iface->port_ifidx;
3068 hash->iface_tag = tag_create_random();
3072 bond_rebalance_port(struct port *port)
3074 struct slave_balance *bals;
3076 struct bond_entry *hashes[BOND_MASK + 1];
3077 struct slave_balance *b, *from, *to;
3078 struct bond_entry *e;
3081 assert(port->bond_mode == BM_SLB);
3083 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3084 * descending order of tx_bytes, so that bals[0] represents the most
3085 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3088 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3089 * array for each slave_balance structure, we sort our local array of
3090 * hashes in order by slave, so that all of the hashes for a given slave
3091 * become contiguous in memory, and then we point each 'hashes' members of
3092 * a slave_balance structure to the start of a contiguous group. */
3093 n_bals = port->n_ifaces;
3094 bals = xmalloc(n_bals * sizeof *bals);
3095 for (b = bals; b < &bals[n_bals]; b++) {
3096 b->iface = port->ifaces[b - bals];
3101 for (i = 0; i <= BOND_MASK; i++) {
3102 hashes[i] = &port->bond_hash[i];
3104 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3105 for (i = 0; i <= BOND_MASK; i++) {
3107 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3108 b = &bals[e->iface_idx];
3109 b->tx_bytes += e->tx_bytes;
3111 b->hashes = &hashes[i];
3116 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3117 log_bals(bals, n_bals, port);
3119 /* Discard slaves that aren't enabled (which were sorted to the back of the
3120 * array earlier). */
3121 while (!bals[n_bals - 1].iface->enabled) {
3128 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3129 to = &bals[n_bals - 1];
3130 for (from = bals; from < to; ) {
3131 uint64_t overload = from->tx_bytes - to->tx_bytes;
3132 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3133 /* The extra load on 'from' (and all less-loaded slaves), compared
3134 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3135 * it is less than ~1Mbps. No point in rebalancing. */
3137 } else if (from->n_hashes == 1) {
3138 /* 'from' only carries a single MAC hash, so we can't shift any
3139 * load away from it, even though we want to. */
3142 /* 'from' is carrying significantly more load than 'to', and that
3143 * load is split across at least two different hashes. Pick a hash
3144 * to migrate to 'to' (the least-loaded slave), given that doing so
3145 * must decrease the ratio of the load on the two slaves by at
3148 * The sort order we use means that we prefer to shift away the
3149 * smallest hashes instead of the biggest ones. There is little
3150 * reason behind this decision; we could use the opposite sort
3151 * order to shift away big hashes ahead of small ones. */
3154 for (i = 0; i < from->n_hashes; i++) {
3155 double old_ratio, new_ratio;
3156 uint64_t delta = from->hashes[i]->tx_bytes;
3158 if (delta == 0 || from->tx_bytes - delta == 0) {
3159 /* Pointless move. */
3163 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3165 if (to->tx_bytes == 0) {
3166 /* Nothing on the new slave, move it. */
3170 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3171 new_ratio = (double)(from->tx_bytes - delta) /
3172 (to->tx_bytes + delta);
3174 if (new_ratio == 0) {
3175 /* Should already be covered but check to prevent division
3180 if (new_ratio < 1) {
3181 new_ratio = 1 / new_ratio;
3184 if (old_ratio - new_ratio > 0.1) {
3185 /* Would decrease the ratio, move it. */
3189 if (i < from->n_hashes) {
3190 bond_shift_load(from, to, i);
3191 port->bond_compat_is_stale = true;
3193 /* If the result of the migration changed the relative order of
3194 * 'from' and 'to' swap them back to maintain invariants. */
3195 if (order_swapped) {
3196 swap_bals(from, to);
3199 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3200 * point to different slave_balance structures. It is only
3201 * valid to do these two operations in a row at all because we
3202 * know that 'from' will not move past 'to' and vice versa. */
3203 resort_bals(from, bals, n_bals);
3204 resort_bals(to, bals, n_bals);
3211 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3212 * historical data to decay to <1% in 7 rebalancing runs. */
3213 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3222 bond_send_learning_packets(struct port *port)
3224 struct bridge *br = port->bridge;
3225 struct mac_entry *e;
3226 struct ofpbuf packet;
3227 int error, n_packets, n_errors;
3229 if (!port->n_ifaces || port->active_iface < 0) {
3233 ofpbuf_init(&packet, 128);
3234 error = n_packets = n_errors = 0;
3235 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3236 union ofp_action actions[2], *a;
3242 if (e->port == port->port_idx
3243 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3247 /* Compose actions. */
3248 memset(actions, 0, sizeof actions);
3251 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3252 a->vlan_vid.len = htons(sizeof *a);
3253 a->vlan_vid.vlan_vid = htons(e->vlan);
3256 a->output.type = htons(OFPAT_OUTPUT);
3257 a->output.len = htons(sizeof *a);
3258 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3263 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3265 flow_extract(&packet, 0, ODPP_NONE, &flow);
3266 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3273 ofpbuf_uninit(&packet);
3276 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3277 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3278 "packets, last error was: %s",
3279 port->name, n_errors, n_packets, strerror(error));
3281 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3282 port->name, n_packets);
3286 /* Bonding unixctl user interface functions. */
3289 bond_unixctl_list(struct unixctl_conn *conn,
3290 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3292 struct ds ds = DS_EMPTY_INITIALIZER;
3293 const struct bridge *br;
3295 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3297 LIST_FOR_EACH (br, node, &all_bridges) {
3300 for (i = 0; i < br->n_ports; i++) {
3301 const struct port *port = br->ports[i];
3302 if (port->n_ifaces > 1) {
3305 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3306 bond_mode_to_string(port->bond_mode));
3307 for (j = 0; j < port->n_ifaces; j++) {
3308 const struct iface *iface = port->ifaces[j];
3310 ds_put_cstr(&ds, ", ");
3312 ds_put_cstr(&ds, iface->name);
3314 ds_put_char(&ds, '\n');
3318 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3322 static struct port *
3323 bond_find(const char *name)
3325 const struct bridge *br;
3327 LIST_FOR_EACH (br, node, &all_bridges) {
3330 for (i = 0; i < br->n_ports; i++) {
3331 struct port *port = br->ports[i];
3332 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3341 bond_unixctl_show(struct unixctl_conn *conn,
3342 const char *args, void *aux OVS_UNUSED)
3344 struct ds ds = DS_EMPTY_INITIALIZER;
3345 const struct port *port;
3348 port = bond_find(args);
3350 unixctl_command_reply(conn, 501, "no such bond");
3354 ds_put_format(&ds, "bond_mode: %s\n",
3355 bond_mode_to_string(port->bond_mode));
3356 ds_put_format(&ds, "bond-detect-mode: %s\n",
3357 port->miimon ? "miimon" : "carrier");
3358 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3359 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3361 if (port->bond_mode == BM_SLB) {
3362 ds_put_format(&ds, "next rebalance: %lld ms\n",
3363 port->bond_next_rebalance - time_msec());
3366 for (j = 0; j < port->n_ifaces; j++) {
3367 const struct iface *iface = port->ifaces[j];
3368 struct bond_entry *be;
3371 ds_put_format(&ds, "slave %s: %s\n",
3372 iface->name, iface->enabled ? "enabled" : "disabled");
3373 if (j == port->active_iface) {
3374 ds_put_cstr(&ds, "\tactive slave\n");
3376 if (iface->delay_expires != LLONG_MAX) {
3377 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3378 iface->enabled ? "downdelay" : "updelay",
3379 iface->delay_expires - time_msec());
3382 if (port->bond_mode != BM_SLB) {
3387 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3388 int hash = be - port->bond_hash;
3389 struct mac_entry *me;
3391 if (be->iface_idx != j) {
3395 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3396 hash, be->tx_bytes / 1024);
3399 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3402 if (bond_hash(me->mac, me->vlan) == hash
3403 && me->port != port->port_idx
3404 && choose_output_iface(port, me->mac, me->vlan,
3406 && dp_ifidx == iface->dp_ifidx)
3408 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3409 ETH_ADDR_ARGS(me->mac));
3414 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3419 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3420 void *aux OVS_UNUSED)
3422 char *args = (char *) args_;
3423 char *save_ptr = NULL;
3424 char *bond_s, *hash_s, *slave_s;
3426 struct iface *iface;
3427 struct bond_entry *entry;
3430 bond_s = strtok_r(args, " ", &save_ptr);
3431 hash_s = strtok_r(NULL, " ", &save_ptr);
3432 slave_s = strtok_r(NULL, " ", &save_ptr);
3434 unixctl_command_reply(conn, 501,
3435 "usage: bond/migrate BOND HASH SLAVE");
3439 port = bond_find(bond_s);
3441 unixctl_command_reply(conn, 501, "no such bond");
3445 if (port->bond_mode != BM_SLB) {
3446 unixctl_command_reply(conn, 501, "not an SLB bond");
3450 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3451 hash = atoi(hash_s) & BOND_MASK;
3453 unixctl_command_reply(conn, 501, "bad hash");
3457 iface = port_lookup_iface(port, slave_s);
3459 unixctl_command_reply(conn, 501, "no such slave");
3463 if (!iface->enabled) {
3464 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3468 entry = &port->bond_hash[hash];
3469 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3470 entry->iface_idx = iface->port_ifidx;
3471 entry->iface_tag = tag_create_random();
3472 port->bond_compat_is_stale = true;
3473 unixctl_command_reply(conn, 200, "migrated");
3477 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3478 void *aux OVS_UNUSED)
3480 char *args = (char *) args_;
3481 char *save_ptr = NULL;
3482 char *bond_s, *slave_s;
3484 struct iface *iface;
3486 bond_s = strtok_r(args, " ", &save_ptr);
3487 slave_s = strtok_r(NULL, " ", &save_ptr);
3489 unixctl_command_reply(conn, 501,
3490 "usage: bond/set-active-slave BOND SLAVE");
3494 port = bond_find(bond_s);
3496 unixctl_command_reply(conn, 501, "no such bond");
3500 iface = port_lookup_iface(port, slave_s);
3502 unixctl_command_reply(conn, 501, "no such slave");
3506 if (!iface->enabled) {
3507 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3511 if (port->active_iface != iface->port_ifidx) {
3512 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3513 port->active_iface = iface->port_ifidx;
3514 port->active_iface_tag = tag_create_random();
3515 VLOG_INFO("port %s: active interface is now %s",
3516 port->name, iface->name);
3517 bond_send_learning_packets(port);
3518 unixctl_command_reply(conn, 200, "done");
3520 unixctl_command_reply(conn, 200, "no change");
3525 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3527 char *args = (char *) args_;
3528 char *save_ptr = NULL;
3529 char *bond_s, *slave_s;
3531 struct iface *iface;
3533 bond_s = strtok_r(args, " ", &save_ptr);
3534 slave_s = strtok_r(NULL, " ", &save_ptr);
3536 unixctl_command_reply(conn, 501,
3537 "usage: bond/enable/disable-slave BOND SLAVE");
3541 port = bond_find(bond_s);
3543 unixctl_command_reply(conn, 501, "no such bond");
3547 iface = port_lookup_iface(port, slave_s);
3549 unixctl_command_reply(conn, 501, "no such slave");
3553 bond_enable_slave(iface, enable);
3554 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3558 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3559 void *aux OVS_UNUSED)
3561 enable_slave(conn, args, true);
3565 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3566 void *aux OVS_UNUSED)
3568 enable_slave(conn, args, false);
3572 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3573 void *aux OVS_UNUSED)
3575 char *args = (char *) args_;
3576 uint8_t mac[ETH_ADDR_LEN];
3580 char *mac_s, *vlan_s;
3581 char *save_ptr = NULL;
3583 mac_s = strtok_r(args, " ", &save_ptr);
3584 vlan_s = strtok_r(NULL, " ", &save_ptr);
3587 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3588 unixctl_command_reply(conn, 501, "invalid vlan");
3592 vlan = OFP_VLAN_NONE;
3595 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3596 == ETH_ADDR_SCAN_COUNT) {
3597 hash = bond_hash(mac, vlan);
3599 hash_cstr = xasprintf("%u", hash);
3600 unixctl_command_reply(conn, 200, hash_cstr);
3603 unixctl_command_reply(conn, 501, "invalid mac");
3610 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3611 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3612 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3613 unixctl_command_register("bond/set-active-slave",
3614 bond_unixctl_set_active_slave, NULL);
3615 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3617 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3619 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3622 /* Port functions. */
3624 static struct port *
3625 port_create(struct bridge *br, const char *name)
3629 port = xzalloc(sizeof *port);
3631 port->port_idx = br->n_ports;
3633 port->trunks = NULL;
3634 port->name = xstrdup(name);
3635 port->active_iface = -1;
3637 if (br->n_ports >= br->allocated_ports) {
3638 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3641 br->ports[br->n_ports++] = port;
3642 shash_add_assert(&br->port_by_name, port->name, port);
3644 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3651 get_port_other_config(const struct ovsrec_port *port, const char *key,
3652 const char *default_value)
3656 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3658 return value ? value : default_value;
3662 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3664 struct shash new_ifaces;
3667 /* Collect list of new interfaces. */
3668 shash_init(&new_ifaces);
3669 for (i = 0; i < cfg->n_interfaces; i++) {
3670 const char *name = cfg->interfaces[i]->name;
3671 shash_add_once(&new_ifaces, name, NULL);
3674 /* Get rid of deleted interfaces. */
3675 for (i = 0; i < port->n_ifaces; ) {
3676 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3677 iface_destroy(port->ifaces[i]);
3683 shash_destroy(&new_ifaces);
3687 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3689 const char *detect_mode;
3690 struct shash new_ifaces;
3691 long long int next_rebalance, miimon_next_update;
3692 unsigned long *trunks;
3698 /* Update settings. */
3699 port->updelay = cfg->bond_updelay;
3700 if (port->updelay < 0) {
3703 port->downdelay = cfg->bond_downdelay;
3704 if (port->downdelay < 0) {
3705 port->downdelay = 0;
3707 port->bond_rebalance_interval = atoi(
3708 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3709 if (port->bond_rebalance_interval < 1000) {
3710 port->bond_rebalance_interval = 1000;
3712 next_rebalance = time_msec() + port->bond_rebalance_interval;
3713 if (port->bond_next_rebalance > next_rebalance) {
3714 port->bond_next_rebalance = next_rebalance;
3717 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
3720 if (!strcmp(detect_mode, "carrier")) {
3721 port->miimon = false;
3722 } else if (!strcmp(detect_mode, "miimon")) {
3723 port->miimon = true;
3725 port->miimon = false;
3726 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
3727 "carrier", port->name, detect_mode);
3730 port->bond_miimon_interval = atoi(
3731 get_port_other_config(cfg, "bond-miimon-interval", "200"));
3732 if (port->bond_miimon_interval < 100) {
3733 port->bond_miimon_interval = 100;
3735 miimon_next_update = time_msec() + port->bond_miimon_interval;
3736 if (port->bond_miimon_next_update > miimon_next_update) {
3737 port->bond_miimon_next_update = miimon_next_update;
3740 if (!port->cfg->bond_mode ||
3741 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
3742 port->bond_mode = BM_SLB;
3743 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
3744 port->bond_mode = BM_AB;
3746 port->bond_mode = BM_SLB;
3747 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
3748 port->name, port->cfg->bond_mode,
3749 bond_mode_to_string(port->bond_mode));
3752 /* Add new interfaces and update 'cfg' member of existing ones. */
3753 shash_init(&new_ifaces);
3754 for (i = 0; i < cfg->n_interfaces; i++) {
3755 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3756 struct iface *iface;
3758 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3759 VLOG_WARN("port %s: %s specified twice as port interface",
3760 port->name, if_cfg->name);
3761 iface_set_ofport(if_cfg, -1);
3765 iface = iface_lookup(port->bridge, if_cfg->name);
3767 if (iface->port != port) {
3768 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3770 port->bridge->name, if_cfg->name, iface->port->name);
3773 iface->cfg = if_cfg;
3775 iface = iface_create(port, if_cfg);
3778 /* Determine interface type. The local port always has type
3779 * "internal". Other ports take their type from the database and
3780 * default to "system" if none is specified. */
3781 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3782 : if_cfg->type[0] ? if_cfg->type
3785 shash_destroy(&new_ifaces);
3790 if (port->n_ifaces < 2) {
3792 if (vlan >= 0 && vlan <= 4095) {
3793 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3798 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3799 * they even work as-is. But they have not been tested. */
3800 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3804 if (port->vlan != vlan) {
3806 bridge_flush(port->bridge);
3809 /* Get trunked VLANs. */
3811 if (vlan < 0 && cfg->n_trunks) {
3814 trunks = bitmap_allocate(4096);
3816 for (i = 0; i < cfg->n_trunks; i++) {
3817 int trunk = cfg->trunks[i];
3819 bitmap_set1(trunks, trunk);
3825 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3826 port->name, cfg->n_trunks);
3828 if (n_errors == cfg->n_trunks) {
3829 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3831 bitmap_free(trunks);
3834 } else if (vlan >= 0 && cfg->n_trunks) {
3835 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3839 ? port->trunks != NULL
3840 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3841 bridge_flush(port->bridge);
3843 bitmap_free(port->trunks);
3844 port->trunks = trunks;
3848 port_destroy(struct port *port)
3851 struct bridge *br = port->bridge;
3855 proc_net_compat_update_vlan(port->name, NULL, 0);
3856 proc_net_compat_update_bond(port->name, NULL);
3858 for (i = 0; i < MAX_MIRRORS; i++) {
3859 struct mirror *m = br->mirrors[i];
3860 if (m && m->out_port == port) {
3865 while (port->n_ifaces > 0) {
3866 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3869 shash_find_and_delete_assert(&br->port_by_name, port->name);
3871 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3872 del->port_idx = port->port_idx;
3874 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3876 netdev_monitor_destroy(port->monitor);
3878 bitmap_free(port->trunks);
3885 static struct port *
3886 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3888 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3889 return iface ? iface->port : NULL;
3892 static struct port *
3893 port_lookup(const struct bridge *br, const char *name)
3895 return shash_find_data(&br->port_by_name, name);
3898 static struct iface *
3899 port_lookup_iface(const struct port *port, const char *name)
3901 struct iface *iface = iface_lookup(port->bridge, name);
3902 return iface && iface->port == port ? iface : NULL;
3906 port_update_bonding(struct port *port)
3908 if (port->monitor) {
3909 netdev_monitor_destroy(port->monitor);
3910 port->monitor = NULL;
3912 if (port->n_ifaces < 2) {
3913 /* Not a bonded port. */
3914 if (port->bond_hash) {
3915 free(port->bond_hash);
3916 port->bond_hash = NULL;
3917 port->bond_compat_is_stale = true;
3920 port->bond_fake_iface = false;
3924 if (port->bond_mode == BM_SLB && !port->bond_hash) {
3925 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3926 for (i = 0; i <= BOND_MASK; i++) {
3927 struct bond_entry *e = &port->bond_hash[i];
3931 port->no_ifaces_tag = tag_create_random();
3932 bond_choose_active_iface(port);
3933 port->bond_next_rebalance
3934 = time_msec() + port->bond_rebalance_interval;
3936 if (port->cfg->bond_fake_iface) {
3937 port->bond_next_fake_iface_update = time_msec();
3939 } else if (port->bond_mode != BM_SLB) {
3940 free(port->bond_hash);
3941 port->bond_hash = NULL;
3943 port->bond_compat_is_stale = true;
3944 port->bond_fake_iface = port->cfg->bond_fake_iface;
3946 if (!port->miimon) {
3947 port->monitor = netdev_monitor_create();
3948 for (i = 0; i < port->n_ifaces; i++) {
3949 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
3956 port_update_bond_compat(struct port *port)
3958 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3959 struct compat_bond bond;
3962 if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) {
3963 proc_net_compat_update_bond(port->name, NULL);
3968 bond.updelay = port->updelay;
3969 bond.downdelay = port->downdelay;
3972 bond.hashes = compat_hashes;
3973 if (port->bond_hash) {
3974 const struct bond_entry *e;
3975 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3976 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3977 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3978 cbh->hash = e - port->bond_hash;
3979 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3984 bond.n_slaves = port->n_ifaces;
3985 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3986 for (i = 0; i < port->n_ifaces; i++) {
3987 struct iface *iface = port->ifaces[i];
3988 struct compat_bond_slave *slave = &bond.slaves[i];
3989 slave->name = iface->name;
3991 /* We need to make the same determination as the Linux bonding
3992 * code to determine whether a slave should be consider "up".
3993 * The Linux function bond_miimon_inspect() supports four
3994 * BOND_LINK_* states:
3996 * - BOND_LINK_UP: carrier detected, updelay has passed.
3997 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3998 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3999 * - BOND_LINK_BACK: carrier detected, updelay in progress.
4001 * The function bond_info_show_slave() only considers BOND_LINK_UP
4002 * to be "up" and anything else to be "down".
4004 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
4008 netdev_get_etheraddr(iface->netdev, slave->mac);
4011 if (port->bond_fake_iface) {
4012 struct netdev *bond_netdev;
4014 if (!netdev_open_default(port->name, &bond_netdev)) {
4016 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
4018 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
4020 netdev_close(bond_netdev);
4024 proc_net_compat_update_bond(port->name, &bond);
4029 port_update_vlan_compat(struct port *port)
4031 struct bridge *br = port->bridge;
4032 char *vlandev_name = NULL;
4034 if (port->vlan > 0) {
4035 /* Figure out the name that the VLAN device should actually have, if it
4036 * existed. This takes some work because the VLAN device would not
4037 * have port->name in its name; rather, it would have the trunk port's
4038 * name, and 'port' would be attached to a bridge that also had the
4039 * VLAN device one of its ports. So we need to find a trunk port that
4040 * includes port->vlan.
4042 * There might be more than one candidate. This doesn't happen on
4043 * XenServer, so if it happens we just pick the first choice in
4044 * alphabetical order instead of creating multiple VLAN devices. */
4046 for (i = 0; i < br->n_ports; i++) {
4047 struct port *p = br->ports[i];
4048 if (port_trunks_vlan(p, port->vlan)
4050 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
4052 uint8_t ea[ETH_ADDR_LEN];
4053 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
4054 if (!eth_addr_is_multicast(ea) &&
4055 !eth_addr_is_reserved(ea) &&
4056 !eth_addr_is_zero(ea)) {
4057 vlandev_name = p->name;
4062 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
4065 /* Interface functions. */
4068 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4071 union ofp_action action;
4073 memset(&action, 0, sizeof action);
4074 action.output.type = htons(OFPAT_OUTPUT);
4075 action.output.len = htons(sizeof action);
4076 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4078 flow_extract(packet, 0, ODPP_NONE, &flow);
4080 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4082 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4083 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4087 static struct iface *
4088 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4090 struct bridge *br = port->bridge;
4091 struct iface *iface;
4092 char *name = if_cfg->name;
4094 iface = xzalloc(sizeof *iface);
4096 iface->port_ifidx = port->n_ifaces;
4097 iface->name = xstrdup(name);
4098 iface->dp_ifidx = -1;
4099 iface->tag = tag_create_random();
4100 iface->delay_expires = LLONG_MAX;
4101 iface->netdev = NULL;
4102 iface->cfg = if_cfg;
4104 shash_add_assert(&br->iface_by_name, iface->name, iface);
4106 if (port->n_ifaces >= port->allocated_ifaces) {
4107 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4108 sizeof *port->ifaces);
4110 port->ifaces[port->n_ifaces++] = iface;
4111 if (port->n_ifaces > 1) {
4112 br->has_bonded_ports = true;
4115 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4123 iface_destroy(struct iface *iface)
4126 struct port *port = iface->port;
4127 struct bridge *br = port->bridge;
4128 bool del_active = port->active_iface == iface->port_ifidx;
4131 if (port->monitor) {
4132 netdev_monitor_remove(port->monitor, iface->netdev);
4135 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4137 if (iface->dp_ifidx >= 0) {
4138 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4141 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4142 del->port_ifidx = iface->port_ifidx;
4144 netdev_close(iface->netdev);
4147 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4148 bond_choose_active_iface(port);
4149 bond_send_learning_packets(port);
4152 cfm_destroy(iface->cfm);
4157 bridge_flush(port->bridge);
4161 static struct iface *
4162 iface_lookup(const struct bridge *br, const char *name)
4164 return shash_find_data(&br->iface_by_name, name);
4167 static struct iface *
4168 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4170 struct iface *iface;
4172 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4173 hash_int(dp_ifidx, 0), &br->ifaces) {
4174 if (iface->dp_ifidx == dp_ifidx) {
4181 /* Set Ethernet address of 'iface', if one is specified in the configuration
4184 iface_set_mac(struct iface *iface)
4186 uint8_t ea[ETH_ADDR_LEN];
4188 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4189 if (eth_addr_is_multicast(ea)) {
4190 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4192 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4193 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4194 iface->name, iface->name);
4196 int error = netdev_set_etheraddr(iface->netdev, ea);
4198 VLOG_ERR("interface %s: setting MAC failed (%s)",
4199 iface->name, strerror(error));
4205 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4207 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4210 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4214 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4216 * The value strings in '*shash' are taken directly from values[], not copied,
4217 * so the caller should not modify or free them. */
4219 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4220 struct shash *shash)
4225 for (i = 0; i < n; i++) {
4226 shash_add(shash, keys[i], values[i]);
4230 /* Creates 'keys' and 'values' arrays from 'shash'.
4232 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4233 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4234 * are populated with with strings taken directly from 'shash' and thus have
4235 * the same ownership of the key-value pairs in shash.
4238 shash_to_ovs_idl_map(struct shash *shash,
4239 char ***keys, char ***values, size_t *n)
4243 struct shash_node *sn;
4245 count = shash_count(shash);
4247 k = xmalloc(count * sizeof *k);
4248 v = xmalloc(count * sizeof *v);
4251 SHASH_FOR_EACH(sn, shash) {
4262 struct iface_delete_queues_cbdata {
4263 struct netdev *netdev;
4264 const struct ovsdb_datum *queues;
4268 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4270 union ovsdb_atom atom;
4272 atom.integer = target;
4273 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4277 iface_delete_queues(unsigned int queue_id,
4278 const struct shash *details OVS_UNUSED, void *cbdata_)
4280 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4282 if (!queue_ids_include(cbdata->queues, queue_id)) {
4283 netdev_delete_queue(cbdata->netdev, queue_id);
4288 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4290 if (!qos || qos->type[0] == '\0') {
4291 netdev_set_qos(iface->netdev, NULL, NULL);
4293 struct iface_delete_queues_cbdata cbdata;
4294 struct shash details;
4297 /* Configure top-level Qos for 'iface'. */
4298 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4299 qos->n_other_config, &details);
4300 netdev_set_qos(iface->netdev, qos->type, &details);
4301 shash_destroy(&details);
4303 /* Deconfigure queues that were deleted. */
4304 cbdata.netdev = iface->netdev;
4305 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4307 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4309 /* Configure queues for 'iface'. */
4310 for (i = 0; i < qos->n_queues; i++) {
4311 const struct ovsrec_queue *queue = qos->value_queues[i];
4312 unsigned int queue_id = qos->key_queues[i];
4314 shash_from_ovs_idl_map(queue->key_other_config,
4315 queue->value_other_config,
4316 queue->n_other_config, &details);
4317 netdev_set_queue(iface->netdev, queue_id, &details);
4318 shash_destroy(&details);
4324 iface_update_cfm(struct iface *iface)
4328 uint16_t *remote_mps;
4329 struct ovsrec_monitor *mon;
4330 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4332 mon = iface->cfg->monitor;
4338 if (netdev_get_etheraddr(iface->netdev, ea)) {
4339 VLOG_WARN("interface %s: Failed to get ethernet address. "
4340 "Skipping Monitor.", iface->name);
4344 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4345 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4350 iface->cfm = cfm_create();
4354 cfm->mpid = mon->mpid;
4355 cfm->interval = mon->interval ? *mon->interval : 1000;
4357 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4358 memcpy(cfm->maid, maid, sizeof cfm->maid);
4360 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4361 for(i = 0; i < mon->n_remote_mps; i++) {
4362 remote_mps[i] = mon->remote_mps[i]->mpid;
4364 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4367 if (!cfm_configure(iface->cfm)) {
4368 cfm_destroy(iface->cfm);
4373 /* Port mirroring. */
4375 static struct mirror *
4376 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4380 for (i = 0; i < MAX_MIRRORS; i++) {
4381 struct mirror *m = br->mirrors[i];
4382 if (m && uuid_equals(uuid, &m->uuid)) {
4390 mirror_reconfigure(struct bridge *br)
4392 unsigned long *rspan_vlans;
4395 /* Get rid of deleted mirrors. */
4396 for (i = 0; i < MAX_MIRRORS; i++) {
4397 struct mirror *m = br->mirrors[i];
4399 const struct ovsdb_datum *mc;
4400 union ovsdb_atom atom;
4402 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4403 atom.uuid = br->mirrors[i]->uuid;
4404 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4410 /* Add new mirrors and reconfigure existing ones. */
4411 for (i = 0; i < br->cfg->n_mirrors; i++) {
4412 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4413 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4415 mirror_reconfigure_one(m, cfg);
4417 mirror_create(br, cfg);
4421 /* Update port reserved status. */
4422 for (i = 0; i < br->n_ports; i++) {
4423 br->ports[i]->is_mirror_output_port = false;
4425 for (i = 0; i < MAX_MIRRORS; i++) {
4426 struct mirror *m = br->mirrors[i];
4427 if (m && m->out_port) {
4428 m->out_port->is_mirror_output_port = true;
4432 /* Update flooded vlans (for RSPAN). */
4434 if (br->cfg->n_flood_vlans) {
4435 rspan_vlans = bitmap_allocate(4096);
4437 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4438 int64_t vlan = br->cfg->flood_vlans[i];
4439 if (vlan >= 0 && vlan < 4096) {
4440 bitmap_set1(rspan_vlans, vlan);
4441 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4444 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4449 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4455 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4460 for (i = 0; ; i++) {
4461 if (i >= MAX_MIRRORS) {
4462 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4463 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4466 if (!br->mirrors[i]) {
4471 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4474 br->mirrors[i] = m = xzalloc(sizeof *m);
4477 m->name = xstrdup(cfg->name);
4478 shash_init(&m->src_ports);
4479 shash_init(&m->dst_ports);
4485 mirror_reconfigure_one(m, cfg);
4489 mirror_destroy(struct mirror *m)
4492 struct bridge *br = m->bridge;
4495 for (i = 0; i < br->n_ports; i++) {
4496 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4497 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4500 shash_destroy(&m->src_ports);
4501 shash_destroy(&m->dst_ports);
4504 m->bridge->mirrors[m->idx] = NULL;
4513 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4514 struct shash *names)
4518 for (i = 0; i < n_ports; i++) {
4519 const char *name = ports[i]->name;
4520 if (port_lookup(m->bridge, name)) {
4521 shash_add_once(names, name, NULL);
4523 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4524 "port %s", m->bridge->name, m->name, name);
4530 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4536 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4538 for (i = 0; i < cfg->n_select_vlan; i++) {
4539 int64_t vlan = cfg->select_vlan[i];
4540 if (vlan < 0 || vlan > 4095) {
4541 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4542 m->bridge->name, m->name, vlan);
4544 (*vlans)[n_vlans++] = vlan;
4551 vlan_is_mirrored(const struct mirror *m, int vlan)
4555 for (i = 0; i < m->n_vlans; i++) {
4556 if (m->vlans[i] == vlan) {
4564 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4568 for (i = 0; i < m->n_vlans; i++) {
4569 if (port_trunks_vlan(p, m->vlans[i])) {
4577 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4579 struct shash src_ports, dst_ports;
4580 mirror_mask_t mirror_bit;
4581 struct port *out_port;
4588 if (strcmp(cfg->name, m->name)) {
4590 m->name = xstrdup(cfg->name);
4593 /* Get output port. */
4594 if (cfg->output_port) {
4595 out_port = port_lookup(m->bridge, cfg->output_port->name);
4597 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4598 m->bridge->name, m->name);
4604 if (cfg->output_vlan) {
4605 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4606 "output vlan; ignoring output vlan",
4607 m->bridge->name, m->name);
4609 } else if (cfg->output_vlan) {
4611 out_vlan = *cfg->output_vlan;
4613 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4614 m->bridge->name, m->name);
4619 shash_init(&src_ports);
4620 shash_init(&dst_ports);
4621 if (cfg->select_all) {
4622 for (i = 0; i < m->bridge->n_ports; i++) {
4623 const char *name = m->bridge->ports[i]->name;
4624 shash_add_once(&src_ports, name, NULL);
4625 shash_add_once(&dst_ports, name, NULL);
4630 /* Get ports, and drop duplicates and ports that don't exist. */
4631 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4633 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4636 /* Get all the vlans, and drop duplicate and invalid vlans. */
4637 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4640 /* Update mirror data. */
4641 if (!shash_equal_keys(&m->src_ports, &src_ports)
4642 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4643 || m->n_vlans != n_vlans
4644 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4645 || m->out_port != out_port
4646 || m->out_vlan != out_vlan) {
4647 bridge_flush(m->bridge);
4649 shash_swap(&m->src_ports, &src_ports);
4650 shash_swap(&m->dst_ports, &dst_ports);
4653 m->n_vlans = n_vlans;
4654 m->out_port = out_port;
4655 m->out_vlan = out_vlan;
4658 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4659 for (i = 0; i < m->bridge->n_ports; i++) {
4660 struct port *port = m->bridge->ports[i];
4662 if (shash_find(&m->src_ports, port->name)
4665 ? port_trunks_any_mirrored_vlan(m, port)
4666 : vlan_is_mirrored(m, port->vlan)))) {
4667 port->src_mirrors |= mirror_bit;
4669 port->src_mirrors &= ~mirror_bit;
4672 if (shash_find(&m->dst_ports, port->name)) {
4673 port->dst_mirrors |= mirror_bit;
4675 port->dst_mirrors &= ~mirror_bit;
4680 shash_destroy(&src_ports);
4681 shash_destroy(&dst_ports);