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)
1123 enum netdev_flags flags;
1132 if (!netdev_get_status(iface->netdev, &sh)) {
1134 char **keys, **values;
1136 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1137 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1142 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1145 shash_destroy_free_data(&sh);
1147 error = netdev_get_flags(iface->netdev, &flags);
1149 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1152 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1155 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1157 ovsrec_interface_set_duplex(iface->cfg,
1158 netdev_features_is_full_duplex(current)
1160 /* warning: uint64_t -> int64_t conversion */
1161 bps = netdev_features_to_bps(current);
1162 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1165 ovsrec_interface_set_duplex(iface->cfg, NULL);
1166 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1170 ovsrec_interface_set_link_state(iface->cfg,
1171 netdev_get_carrier(iface->netdev)
1174 error = netdev_get_mtu(iface->netdev, &mtu);
1177 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1180 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1185 iface_refresh_cfm_stats(struct iface *iface)
1189 const struct ovsrec_monitor *mon;
1191 mon = iface->cfg->monitor;
1198 for (i = 0; i < mon->n_remote_mps; i++) {
1199 const struct ovsrec_maintenance_point *mp;
1200 const struct remote_mp *rmp;
1202 mp = mon->remote_mps[i];
1203 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1205 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1208 if (hmap_is_empty(&cfm->x_remote_mps)) {
1209 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1212 struct remote_mp *rmp;
1213 int64_t *x_remote_mps;
1215 length = hmap_count(&cfm->x_remote_mps);
1216 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1219 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1220 x_remote_mps[i++] = rmp->mpid;
1223 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1227 if (hmap_is_empty(&cfm->x_remote_maids)) {
1228 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1231 char **x_remote_maids;
1232 struct remote_maid *rmaid;
1234 length = hmap_count(&cfm->x_remote_maids);
1235 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1238 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1241 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1243 for (j = 0; j < CCM_MAID_LEN; j++) {
1244 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1249 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1251 for (i = 0; i < length; i++) {
1252 free(x_remote_maids[i]);
1254 free(x_remote_maids);
1257 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1261 iface_refresh_stats(struct iface *iface)
1267 static const struct iface_stat iface_stats[] = {
1268 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1269 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1270 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1271 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1272 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1273 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1274 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1275 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1276 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1277 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1278 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1279 { "collisions", offsetof(struct netdev_stats, collisions) },
1281 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1282 const struct iface_stat *s;
1284 char *keys[N_STATS];
1285 int64_t values[N_STATS];
1288 struct netdev_stats stats;
1290 /* Intentionally ignore return value, since errors will set 'stats' to
1291 * all-1s, and we will deal with that correctly below. */
1292 netdev_get_stats(iface->netdev, &stats);
1295 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1296 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1297 if (value != UINT64_MAX) {
1304 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1308 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1310 struct ovsdb_datum datum;
1314 get_system_stats(&stats);
1316 ovsdb_datum_from_shash(&datum, &stats);
1317 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1321 static inline const char *
1322 nx_role_to_str(enum nx_role role)
1327 case NX_ROLE_MASTER:
1332 return "*** INVALID ROLE ***";
1337 bridge_refresh_controller_status(const struct bridge *br)
1340 const struct ovsrec_controller *cfg;
1342 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1344 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1345 struct ofproto_controller_info *cinfo =
1346 shash_find_data(&info, cfg->target);
1349 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1350 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1351 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1352 (char **) cinfo->pairs.values,
1355 ovsrec_controller_set_is_connected(cfg, false);
1356 ovsrec_controller_set_role(cfg, NULL);
1357 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1361 ofproto_free_ofproto_controller_info(&info);
1367 const struct ovsrec_open_vswitch *cfg;
1369 bool datapath_destroyed;
1370 bool database_changed;
1373 /* Let each bridge do the work that it needs to do. */
1374 datapath_destroyed = false;
1375 LIST_FOR_EACH (br, node, &all_bridges) {
1376 int error = bridge_run_one(br);
1378 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1379 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1380 "forcing reconfiguration", br->name);
1381 datapath_destroyed = true;
1385 /* (Re)configure if necessary. */
1386 database_changed = ovsdb_idl_run(idl);
1387 cfg = ovsrec_open_vswitch_first(idl);
1389 /* Re-configure SSL. We do this on every trip through the main loop,
1390 * instead of just when the database changes, because the contents of the
1391 * key and certificate files can change without the database changing.
1393 * We do this before bridge_reconfigure() because that function might
1394 * initiate SSL connections and thus requires SSL to be configured. */
1395 if (cfg && cfg->ssl) {
1396 const struct ovsrec_ssl *ssl = cfg->ssl;
1398 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1399 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1402 if (database_changed || datapath_destroyed) {
1404 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1406 bridge_configure_once(cfg);
1407 bridge_reconfigure(cfg);
1409 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1410 ovsdb_idl_txn_commit(txn);
1411 ovsdb_idl_txn_destroy(txn); /* XXX */
1413 /* We still need to reconfigure to avoid dangling pointers to
1414 * now-destroyed ovsrec structures inside bridge data. */
1415 static const struct ovsrec_open_vswitch null_cfg;
1417 bridge_reconfigure(&null_cfg);
1421 /* Refresh system and interface stats if necessary. */
1422 if (time_msec() >= stats_timer) {
1424 struct ovsdb_idl_txn *txn;
1426 txn = ovsdb_idl_txn_create(idl);
1427 LIST_FOR_EACH (br, node, &all_bridges) {
1430 for (i = 0; i < br->n_ports; i++) {
1431 struct port *port = br->ports[i];
1434 for (j = 0; j < port->n_ifaces; j++) {
1435 struct iface *iface = port->ifaces[j];
1436 iface_refresh_stats(iface);
1437 iface_refresh_cfm_stats(iface);
1438 iface_refresh_status(iface);
1441 bridge_refresh_controller_status(br);
1443 refresh_system_stats(cfg);
1444 ovsdb_idl_txn_commit(txn);
1445 ovsdb_idl_txn_destroy(txn); /* XXX */
1448 stats_timer = time_msec() + STATS_INTERVAL;
1456 struct iface *iface;
1458 LIST_FOR_EACH (br, node, &all_bridges) {
1459 ofproto_wait(br->ofproto);
1460 if (ofproto_has_primary_controller(br->ofproto)) {
1464 mac_learning_wait(br->ml);
1467 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1469 cfm_wait(iface->cfm);
1473 ovsdb_idl_wait(idl);
1474 poll_timer_wait_until(stats_timer);
1477 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1478 * configuration changes. */
1480 bridge_flush(struct bridge *br)
1482 COVERAGE_INC(bridge_flush);
1484 mac_learning_flush(br->ml);
1487 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1488 * such interface. */
1489 static struct iface *
1490 bridge_get_local_iface(struct bridge *br)
1494 for (i = 0; i < br->n_ports; i++) {
1495 struct port *port = br->ports[i];
1496 for (j = 0; j < port->n_ifaces; j++) {
1497 struct iface *iface = port->ifaces[j];
1498 if (iface->dp_ifidx == ODPP_LOCAL) {
1507 /* Bridge unixctl user interface functions. */
1509 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1510 const char *args, void *aux OVS_UNUSED)
1512 struct ds ds = DS_EMPTY_INITIALIZER;
1513 const struct bridge *br;
1514 const struct mac_entry *e;
1516 br = bridge_lookup(args);
1518 unixctl_command_reply(conn, 501, "no such bridge");
1522 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1523 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1524 if (e->port < 0 || e->port >= br->n_ports) {
1527 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1528 br->ports[e->port]->ifaces[0]->dp_ifidx,
1529 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1531 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1535 /* Bridge reconfiguration functions. */
1536 static struct bridge *
1537 bridge_create(const struct ovsrec_bridge *br_cfg)
1542 assert(!bridge_lookup(br_cfg->name));
1543 br = xzalloc(sizeof *br);
1545 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1551 dpif_flow_flush(br->dpif);
1553 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1556 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1558 dpif_delete(br->dpif);
1559 dpif_close(br->dpif);
1564 br->name = xstrdup(br_cfg->name);
1566 br->ml = mac_learning_create();
1567 eth_addr_nicira_random(br->default_ea);
1569 hmap_init(&br->ifaces);
1571 shash_init(&br->port_by_name);
1572 shash_init(&br->iface_by_name);
1576 list_push_back(&all_bridges, &br->node);
1578 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1584 bridge_destroy(struct bridge *br)
1589 while (br->n_ports > 0) {
1590 port_destroy(br->ports[br->n_ports - 1]);
1592 list_remove(&br->node);
1593 error = dpif_delete(br->dpif);
1594 if (error && error != ENOENT) {
1595 VLOG_ERR("failed to delete %s: %s",
1596 dpif_name(br->dpif), strerror(error));
1598 dpif_close(br->dpif);
1599 ofproto_destroy(br->ofproto);
1600 mac_learning_destroy(br->ml);
1601 hmap_destroy(&br->ifaces);
1602 shash_destroy(&br->port_by_name);
1603 shash_destroy(&br->iface_by_name);
1610 static struct bridge *
1611 bridge_lookup(const char *name)
1615 LIST_FOR_EACH (br, node, &all_bridges) {
1616 if (!strcmp(br->name, name)) {
1623 /* Handle requests for a listing of all flows known by the OpenFlow
1624 * stack, including those normally hidden. */
1626 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1627 const char *args, void *aux OVS_UNUSED)
1632 br = bridge_lookup(args);
1634 unixctl_command_reply(conn, 501, "Unknown bridge");
1639 ofproto_get_all_flows(br->ofproto, &results);
1641 unixctl_command_reply(conn, 200, ds_cstr(&results));
1642 ds_destroy(&results);
1645 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1646 * connections and reconnect. If BRIDGE is not specified, then all bridges
1647 * drop their controller connections and reconnect. */
1649 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1650 const char *args, void *aux OVS_UNUSED)
1653 if (args[0] != '\0') {
1654 br = bridge_lookup(args);
1656 unixctl_command_reply(conn, 501, "Unknown bridge");
1659 ofproto_reconnect_controllers(br->ofproto);
1661 LIST_FOR_EACH (br, node, &all_bridges) {
1662 ofproto_reconnect_controllers(br->ofproto);
1665 unixctl_command_reply(conn, 200, NULL);
1669 bridge_run_one(struct bridge *br)
1672 struct iface *iface;
1674 error = ofproto_run1(br->ofproto);
1679 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1682 error = ofproto_run2(br->ofproto, br->flush);
1685 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1686 struct ofpbuf *packet;
1692 packet = cfm_run(iface->cfm);
1694 iface_send_packet(iface, packet);
1695 ofpbuf_uninit(packet);
1704 bridge_get_controllers(const struct bridge *br,
1705 struct ovsrec_controller ***controllersp)
1707 struct ovsrec_controller **controllers;
1708 size_t n_controllers;
1710 controllers = br->cfg->controller;
1711 n_controllers = br->cfg->n_controller;
1713 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1719 *controllersp = controllers;
1721 return n_controllers;
1725 bridge_reconfigure_one(struct bridge *br)
1727 struct shash old_ports, new_ports;
1728 struct svec snoops, old_snoops;
1729 struct shash_node *node;
1730 enum ofproto_fail_mode fail_mode;
1733 /* Collect old ports. */
1734 shash_init(&old_ports);
1735 for (i = 0; i < br->n_ports; i++) {
1736 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1739 /* Collect new ports. */
1740 shash_init(&new_ports);
1741 for (i = 0; i < br->cfg->n_ports; i++) {
1742 const char *name = br->cfg->ports[i]->name;
1743 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1744 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1749 /* If we have a controller, then we need a local port. Complain if the
1750 * user didn't specify one.
1752 * XXX perhaps we should synthesize a port ourselves in this case. */
1753 if (bridge_get_controllers(br, NULL)) {
1754 char local_name[IF_NAMESIZE];
1757 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1758 local_name, sizeof local_name);
1759 if (!error && !shash_find(&new_ports, local_name)) {
1760 VLOG_WARN("bridge %s: controller specified but no local port "
1761 "(port named %s) defined",
1762 br->name, local_name);
1766 /* Get rid of deleted ports.
1767 * Get rid of deleted interfaces on ports that still exist. */
1768 SHASH_FOR_EACH (node, &old_ports) {
1769 struct port *port = node->data;
1770 const struct ovsrec_port *port_cfg;
1772 port_cfg = shash_find_data(&new_ports, node->name);
1776 port_del_ifaces(port, port_cfg);
1780 /* Create new ports.
1781 * Add new interfaces to existing ports.
1782 * Reconfigure existing ports. */
1783 SHASH_FOR_EACH (node, &new_ports) {
1784 struct port *port = shash_find_data(&old_ports, node->name);
1786 port = port_create(br, node->name);
1789 port_reconfigure(port, node->data);
1790 if (!port->n_ifaces) {
1791 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1792 br->name, port->name);
1796 shash_destroy(&old_ports);
1797 shash_destroy(&new_ports);
1799 /* Set the fail-mode */
1800 fail_mode = !br->cfg->fail_mode
1801 || !strcmp(br->cfg->fail_mode, "standalone")
1802 ? OFPROTO_FAIL_STANDALONE
1803 : OFPROTO_FAIL_SECURE;
1804 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1805 && !ofproto_has_primary_controller(br->ofproto)) {
1806 ofproto_flush_flows(br->ofproto);
1808 ofproto_set_fail_mode(br->ofproto, fail_mode);
1810 /* Delete all flows if we're switching from connected to standalone or vice
1811 * versa. (XXX Should we delete all flows if we are switching from one
1812 * controller to another?) */
1814 /* Configure OpenFlow controller connection snooping. */
1816 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1817 ovs_rundir(), br->name));
1818 svec_init(&old_snoops);
1819 ofproto_get_snoops(br->ofproto, &old_snoops);
1820 if (!svec_equal(&snoops, &old_snoops)) {
1821 ofproto_set_snoops(br->ofproto, &snoops);
1823 svec_destroy(&snoops);
1824 svec_destroy(&old_snoops);
1826 mirror_reconfigure(br);
1829 /* Initializes 'oc' appropriately as a management service controller for
1832 * The caller must free oc->target when it is no longer needed. */
1834 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1835 struct ofproto_controller *oc)
1837 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1838 oc->max_backoff = 0;
1839 oc->probe_interval = 60;
1840 oc->band = OFPROTO_OUT_OF_BAND;
1841 oc->accept_re = NULL;
1842 oc->update_resolv_conf = false;
1844 oc->burst_limit = 0;
1847 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1849 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1850 struct ofproto_controller *oc)
1852 oc->target = c->target;
1853 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1854 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1855 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1856 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1857 oc->accept_re = c->discover_accept_regex;
1858 oc->update_resolv_conf = c->discover_update_resolv_conf;
1859 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1860 oc->burst_limit = (c->controller_burst_limit
1861 ? *c->controller_burst_limit : 0);
1864 /* Configures the IP stack for 'br''s local interface properly according to the
1865 * configuration in 'c'. */
1867 bridge_configure_local_iface_netdev(struct bridge *br,
1868 struct ovsrec_controller *c)
1870 struct netdev *netdev;
1871 struct in_addr mask, gateway;
1873 struct iface *local_iface;
1876 /* Controller discovery does its own TCP/IP configuration later. */
1877 if (strcmp(c->target, "discover")) {
1881 /* If there's no local interface or no IP address, give up. */
1882 local_iface = bridge_get_local_iface(br);
1883 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1887 /* Bring up the local interface. */
1888 netdev = local_iface->netdev;
1889 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1891 /* Configure the IP address and netmask. */
1892 if (!c->local_netmask
1893 || !inet_aton(c->local_netmask, &mask)
1895 mask.s_addr = guess_netmask(ip.s_addr);
1897 if (!netdev_set_in4(netdev, ip, mask)) {
1898 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1899 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1902 /* Configure the default gateway. */
1903 if (c->local_gateway
1904 && inet_aton(c->local_gateway, &gateway)
1905 && gateway.s_addr) {
1906 if (!netdev_add_router(netdev, gateway)) {
1907 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1908 br->name, IP_ARGS(&gateway.s_addr));
1914 bridge_reconfigure_remotes(struct bridge *br,
1915 const struct sockaddr_in *managers,
1918 const char *disable_ib_str, *queue_id_str;
1919 bool disable_in_band = false;
1922 struct ovsrec_controller **controllers;
1923 size_t n_controllers;
1926 struct ofproto_controller *ocs;
1930 /* Check if we should disable in-band control on this bridge. */
1931 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1932 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1933 disable_in_band = true;
1936 /* Set OpenFlow queue ID for in-band control. */
1937 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1938 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1939 ofproto_set_in_band_queue(br->ofproto, queue_id);
1941 if (disable_in_band) {
1942 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1944 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1946 had_primary = ofproto_has_primary_controller(br->ofproto);
1948 n_controllers = bridge_get_controllers(br, &controllers);
1950 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1953 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1954 for (i = 0; i < n_controllers; i++) {
1955 struct ovsrec_controller *c = controllers[i];
1957 if (!strncmp(c->target, "punix:", 6)
1958 || !strncmp(c->target, "unix:", 5)) {
1959 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1961 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1962 * domain sockets and overwriting arbitrary local files. */
1963 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1964 "\"%s\" due to possibility for remote exploit",
1965 dpif_name(br->dpif), c->target);
1969 bridge_configure_local_iface_netdev(br, c);
1970 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1971 if (disable_in_band) {
1972 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1977 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1978 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1981 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1982 ofproto_flush_flows(br->ofproto);
1985 /* If there are no controllers and the bridge is in standalone
1986 * mode, set up a flow that matches every packet and directs
1987 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1988 * switch is in secure mode and we won't pass any traffic until
1989 * a controller has been defined and it tells us to do so. */
1991 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1992 union ofp_action action;
1993 struct cls_rule rule;
1995 memset(&action, 0, sizeof action);
1996 action.type = htons(OFPAT_OUTPUT);
1997 action.output.len = htons(sizeof action);
1998 action.output.port = htons(OFPP_NORMAL);
1999 cls_rule_init_catchall(&rule, 0);
2000 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2005 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2010 for (i = 0; i < br->n_ports; i++) {
2011 struct port *port = br->ports[i];
2012 for (j = 0; j < port->n_ifaces; j++) {
2013 struct iface *iface = port->ifaces[j];
2014 shash_add_once(ifaces, iface->name, iface);
2016 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2017 shash_add_once(ifaces, port->name, NULL);
2022 /* For robustness, in case the administrator moves around datapath ports behind
2023 * our back, we re-check all the datapath port numbers here.
2025 * This function will set the 'dp_ifidx' members of interfaces that have
2026 * disappeared to -1, so only call this function from a context where those
2027 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2028 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2029 * datapath, which doesn't support UINT16_MAX+1 ports. */
2031 bridge_fetch_dp_ifaces(struct bridge *br)
2033 struct odp_port *dpif_ports;
2034 size_t n_dpif_ports;
2037 /* Reset all interface numbers. */
2038 for (i = 0; i < br->n_ports; i++) {
2039 struct port *port = br->ports[i];
2040 for (j = 0; j < port->n_ifaces; j++) {
2041 struct iface *iface = port->ifaces[j];
2042 iface->dp_ifidx = -1;
2045 hmap_clear(&br->ifaces);
2047 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
2048 for (i = 0; i < n_dpif_ports; i++) {
2049 struct odp_port *p = &dpif_ports[i];
2050 struct iface *iface = iface_lookup(br, p->devname);
2052 if (iface->dp_ifidx >= 0) {
2053 VLOG_WARN("%s reported interface %s twice",
2054 dpif_name(br->dpif), p->devname);
2055 } else if (iface_from_dp_ifidx(br, p->port)) {
2056 VLOG_WARN("%s reported interface %"PRIu16" twice",
2057 dpif_name(br->dpif), p->port);
2059 iface->dp_ifidx = p->port;
2060 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2061 hash_int(iface->dp_ifidx, 0));
2064 iface_set_ofport(iface->cfg,
2065 (iface->dp_ifidx >= 0
2066 ? odp_port_to_ofp_port(iface->dp_ifidx)
2073 /* Bridge packet processing functions. */
2076 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2078 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2081 static struct bond_entry *
2082 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
2085 assert(port->bond_mode == BM_SLB);
2086 return &port->bond_hash[bond_hash(mac, vlan)];
2090 bond_choose_iface(const struct port *port)
2092 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2093 size_t i, best_down_slave = -1;
2094 long long next_delay_expiration = LLONG_MAX;
2096 for (i = 0; i < port->n_ifaces; i++) {
2097 struct iface *iface = port->ifaces[i];
2099 if (iface->enabled) {
2101 } else if (iface->delay_expires < next_delay_expiration) {
2102 best_down_slave = i;
2103 next_delay_expiration = iface->delay_expires;
2107 if (best_down_slave != -1) {
2108 struct iface *iface = port->ifaces[best_down_slave];
2110 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2111 "since no other interface is up", iface->name,
2112 iface->delay_expires - time_msec());
2113 bond_enable_slave(iface, true);
2116 return best_down_slave;
2120 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2121 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2123 struct iface *iface;
2125 assert(port->n_ifaces);
2126 if (port->n_ifaces == 1) {
2127 iface = port->ifaces[0];
2128 } else if (port->bond_mode == BM_AB) {
2129 if (port->active_iface < 0) {
2130 *tags |= port->no_ifaces_tag;
2133 iface = port->ifaces[port->active_iface];
2134 } else if (port->bond_mode == BM_SLB){
2135 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2136 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2137 || !port->ifaces[e->iface_idx]->enabled) {
2138 /* XXX select interface properly. The current interface selection
2139 * is only good for testing the rebalancing code. */
2140 e->iface_idx = bond_choose_iface(port);
2141 if (e->iface_idx < 0) {
2142 *tags |= port->no_ifaces_tag;
2145 e->iface_tag = tag_create_random();
2146 ((struct port *) port)->bond_compat_is_stale = true;
2148 *tags |= e->iface_tag;
2149 iface = port->ifaces[e->iface_idx];
2153 *dp_ifidx = iface->dp_ifidx;
2154 *tags |= iface->tag; /* Currently only used for bonding. */
2159 bond_link_status_update(struct iface *iface, bool carrier)
2161 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2162 struct port *port = iface->port;
2164 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2165 /* Nothing to do. */
2168 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2169 iface->name, carrier ? "up" : "down");
2170 if (carrier == iface->enabled) {
2171 iface->delay_expires = LLONG_MAX;
2172 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2173 iface->name, carrier ? "disabled" : "enabled");
2174 } else if (carrier && port->active_iface < 0) {
2175 bond_enable_slave(iface, true);
2176 if (port->updelay) {
2177 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2178 "other interface is up", iface->name, port->updelay);
2181 int delay = carrier ? port->updelay : port->downdelay;
2182 iface->delay_expires = time_msec() + delay;
2185 "interface %s: will be %s if it stays %s for %d ms",
2187 carrier ? "enabled" : "disabled",
2188 carrier ? "up" : "down",
2195 bond_choose_active_iface(struct port *port)
2197 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2199 port->active_iface = bond_choose_iface(port);
2200 port->active_iface_tag = tag_create_random();
2201 if (port->active_iface >= 0) {
2202 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2203 port->name, port->ifaces[port->active_iface]->name);
2205 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2211 bond_enable_slave(struct iface *iface, bool enable)
2213 struct port *port = iface->port;
2214 struct bridge *br = port->bridge;
2216 /* This acts as a recursion check. If the act of disabling a slave
2217 * causes a different slave to be enabled, the flag will allow us to
2218 * skip redundant work when we reenter this function. It must be
2219 * cleared on exit to keep things safe with multiple bonds. */
2220 static bool moving_active_iface = false;
2222 iface->delay_expires = LLONG_MAX;
2223 if (enable == iface->enabled) {
2227 iface->enabled = enable;
2228 if (!iface->enabled) {
2229 VLOG_WARN("interface %s: disabled", iface->name);
2230 ofproto_revalidate(br->ofproto, iface->tag);
2231 if (iface->port_ifidx == port->active_iface) {
2232 ofproto_revalidate(br->ofproto,
2233 port->active_iface_tag);
2235 /* Disabling a slave can lead to another slave being immediately
2236 * enabled if there will be no active slaves but one is waiting
2237 * on an updelay. In this case we do not need to run most of the
2238 * code for the newly enabled slave since there was no period
2239 * without an active slave and it is redundant with the disabling
2241 moving_active_iface = true;
2242 bond_choose_active_iface(port);
2244 bond_send_learning_packets(port);
2246 VLOG_WARN("interface %s: enabled", iface->name);
2247 if (port->active_iface < 0 && !moving_active_iface) {
2248 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2249 bond_choose_active_iface(port);
2250 bond_send_learning_packets(port);
2252 iface->tag = tag_create_random();
2255 moving_active_iface = false;
2256 port->bond_compat_is_stale = true;
2259 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2260 * bond interface. */
2262 bond_update_fake_iface_stats(struct port *port)
2264 struct netdev_stats bond_stats;
2265 struct netdev *bond_dev;
2268 memset(&bond_stats, 0, sizeof bond_stats);
2270 for (i = 0; i < port->n_ifaces; i++) {
2271 struct netdev_stats slave_stats;
2273 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2274 /* XXX: We swap the stats here because they are swapped back when
2275 * reported by the internal device. The reason for this is
2276 * internal devices normally represent packets going into the system
2277 * but when used as fake bond device they represent packets leaving
2278 * the system. We really should do this in the internal device
2279 * itself because changing it here reverses the counts from the
2280 * perspective of the switch. However, the internal device doesn't
2281 * know what type of device it represents so we have to do it here
2283 bond_stats.tx_packets += slave_stats.rx_packets;
2284 bond_stats.tx_bytes += slave_stats.rx_bytes;
2285 bond_stats.rx_packets += slave_stats.tx_packets;
2286 bond_stats.rx_bytes += slave_stats.tx_bytes;
2290 if (!netdev_open_default(port->name, &bond_dev)) {
2291 netdev_set_stats(bond_dev, &bond_stats);
2292 netdev_close(bond_dev);
2297 bond_run(struct bridge *br)
2301 for (i = 0; i < br->n_ports; i++) {
2302 struct port *port = br->ports[i];
2304 if (port->n_ifaces >= 2) {
2307 if (port->monitor) {
2308 assert(!port->miimon);
2310 /* Track carrier going up and down on interfaces. */
2311 while (!netdev_monitor_poll(port->monitor, &devname)) {
2312 struct iface *iface;
2314 iface = port_lookup_iface(port, devname);
2316 bool up = netdev_get_carrier(iface->netdev);
2318 bond_link_status_update(iface, up);
2319 port_update_bond_compat(port);
2324 assert(port->miimon);
2326 if (time_msec() >= port->bond_miimon_next_update) {
2327 for (j = 0; j < port->n_ifaces; j++) {
2328 struct iface *iface = port->ifaces[j];
2329 bool up = netdev_get_miimon(iface->netdev);
2331 bond_link_status_update(iface, up);
2332 port_update_bond_compat(port);
2334 port->bond_miimon_next_update = time_msec() +
2335 port->bond_miimon_interval;
2339 for (j = 0; j < port->n_ifaces; j++) {
2340 struct iface *iface = port->ifaces[j];
2341 if (time_msec() >= iface->delay_expires) {
2342 bond_enable_slave(iface, !iface->enabled);
2346 if (port->bond_fake_iface
2347 && time_msec() >= port->bond_next_fake_iface_update) {
2348 bond_update_fake_iface_stats(port);
2349 port->bond_next_fake_iface_update = time_msec() + 1000;
2353 if (port->bond_compat_is_stale) {
2354 port->bond_compat_is_stale = false;
2355 port_update_bond_compat(port);
2361 bond_wait(struct bridge *br)
2365 for (i = 0; i < br->n_ports; i++) {
2366 struct port *port = br->ports[i];
2367 if (port->n_ifaces < 2) {
2371 if (port->monitor) {
2372 netdev_monitor_poll_wait(port->monitor);
2376 poll_timer_wait_until(port->bond_miimon_next_update);
2379 for (j = 0; j < port->n_ifaces; j++) {
2380 struct iface *iface = port->ifaces[j];
2381 if (iface->delay_expires != LLONG_MAX) {
2382 poll_timer_wait_until(iface->delay_expires);
2385 if (port->bond_fake_iface) {
2386 poll_timer_wait_until(port->bond_next_fake_iface_update);
2392 set_dst(struct dst *dst, const struct flow *flow,
2393 const struct port *in_port, const struct port *out_port,
2396 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2397 : in_port->vlan >= 0 ? in_port->vlan
2398 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2399 : vlan_tci_to_vid(flow->vlan_tci));
2400 return choose_output_iface(out_port, flow->dl_src, dst->vlan,
2401 &dst->dp_ifidx, tags);
2405 swap_dst(struct dst *p, struct dst *q)
2407 struct dst tmp = *p;
2412 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2413 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2414 * that we push to the datapath. We could in fact fully sort the array by
2415 * vlan, but in most cases there are at most two different vlan tags so that's
2416 * possibly overkill.) */
2418 partition_dsts(struct dst_set *set, int vlan)
2420 struct dst *first = set->dsts;
2421 struct dst *last = set->dsts + set->n;
2423 while (first != last) {
2425 * - All dsts < first have vlan == 'vlan'.
2426 * - All dsts >= last have vlan != 'vlan'.
2427 * - first < last. */
2428 while (first->vlan == vlan) {
2429 if (++first == last) {
2434 /* Same invariants, plus one additional:
2435 * - first->vlan != vlan.
2437 while (last[-1].vlan != vlan) {
2438 if (--last == first) {
2443 /* Same invariants, plus one additional:
2444 * - last[-1].vlan == vlan.*/
2445 swap_dst(first++, --last);
2450 mirror_mask_ffs(mirror_mask_t mask)
2452 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2457 dst_set_init(struct dst_set *set)
2459 set->dsts = set->builtin;
2461 set->allocated = ARRAY_SIZE(set->builtin);
2465 dst_set_add(struct dst_set *set, const struct dst *dst)
2467 if (set->n >= set->allocated) {
2468 size_t new_allocated;
2469 struct dst *new_dsts;
2471 new_allocated = set->allocated * 2;
2472 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2473 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2477 set->dsts = new_dsts;
2478 set->allocated = new_allocated;
2480 set->dsts[set->n++] = *dst;
2484 dst_set_free(struct dst_set *set)
2486 if (set->dsts != set->builtin) {
2492 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2495 for (i = 0; i < set->n; i++) {
2496 if (set->dsts[i].vlan == test->vlan
2497 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2505 port_trunks_vlan(const struct port *port, uint16_t vlan)
2507 return (port->vlan < 0
2508 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2512 port_includes_vlan(const struct port *port, uint16_t vlan)
2514 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2518 port_is_floodable(const struct port *port)
2522 for (i = 0; i < port->n_ifaces; i++) {
2523 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2524 port->ifaces[i]->dp_ifidx)) {
2532 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2533 const struct port *in_port, const struct port *out_port,
2534 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2536 mirror_mask_t mirrors = in_port->src_mirrors;
2541 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2542 if (flow_vlan == 0) {
2543 flow_vlan = OFP_VLAN_NONE;
2546 if (out_port == FLOOD_PORT) {
2547 for (i = 0; i < br->n_ports; i++) {
2548 struct port *port = br->ports[i];
2550 && port_is_floodable(port)
2551 && port_includes_vlan(port, vlan)
2552 && !port->is_mirror_output_port
2553 && set_dst(&dst, flow, in_port, port, tags)) {
2554 mirrors |= port->dst_mirrors;
2555 dst_set_add(set, &dst);
2558 *nf_output_iface = NF_OUT_FLOOD;
2559 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2560 dst_set_add(set, &dst);
2561 *nf_output_iface = dst.dp_ifidx;
2562 mirrors |= out_port->dst_mirrors;
2566 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2567 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2569 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2570 && !dst_is_duplicate(set, &dst)) {
2571 dst_set_add(set, &dst);
2574 for (i = 0; i < br->n_ports; i++) {
2575 struct port *port = br->ports[i];
2576 if (port_includes_vlan(port, m->out_vlan)
2577 && set_dst(&dst, flow, in_port, port, tags))
2579 if (port->vlan < 0) {
2580 dst.vlan = m->out_vlan;
2582 if (dst_is_duplicate(set, &dst)) {
2586 /* Use the vlan tag on the original flow instead of
2587 * the one passed in the vlan parameter. This ensures
2588 * that we compare the vlan from before any implicit
2589 * tagging tags place. This is necessary because
2590 * dst->vlan is the final vlan, after removing implicit
2592 if (port == in_port && dst.vlan == flow_vlan) {
2593 /* Don't send out input port on same VLAN. */
2596 dst_set_add(set, &dst);
2601 mirrors &= mirrors - 1;
2604 partition_dsts(set, flow_vlan);
2607 static void OVS_UNUSED
2608 print_dsts(const struct dst_set *set)
2612 for (i = 0; i < set->n; i++) {
2613 const struct dst *dst = &set->dsts[i];
2615 printf(">p%"PRIu16, dst->dp_ifidx);
2616 if (dst->vlan != OFP_VLAN_NONE) {
2617 printf("v%"PRIu16, dst->vlan);
2623 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2624 const struct port *in_port, const struct port *out_port,
2625 tag_type *tags, struct ofpbuf *actions,
2626 uint16_t *nf_output_iface)
2633 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2636 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2637 if (cur_vlan == 0) {
2638 cur_vlan = OFP_VLAN_NONE;
2640 for (i = 0; i < set.n; i++) {
2641 const struct dst *dst = &set.dsts[i];
2642 if (dst->vlan != cur_vlan) {
2643 if (dst->vlan == OFP_VLAN_NONE) {
2644 nl_msg_put_flag(actions, ODPAT_STRIP_VLAN);
2647 tci = htons(dst->vlan & VLAN_VID_MASK);
2648 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2649 nl_msg_put_be16(actions, ODPAT_SET_DL_TCI, tci);
2651 cur_vlan = dst->vlan;
2653 nl_msg_put_u32(actions, ODPAT_OUTPUT, dst->dp_ifidx);
2658 /* Returns the effective vlan of a packet, taking into account both the
2659 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2660 * the packet is untagged and -1 indicates it has an invalid header and
2661 * should be dropped. */
2662 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2663 struct port *in_port, bool have_packet)
2665 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2666 if (in_port->vlan >= 0) {
2668 /* XXX support double tagging? */
2670 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2671 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2672 "packet received on port %s configured with "
2673 "implicit VLAN %"PRIu16,
2674 br->name, vlan, in_port->name, in_port->vlan);
2678 vlan = in_port->vlan;
2680 if (!port_includes_vlan(in_port, vlan)) {
2682 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2683 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2684 "packet received on port %s not configured for "
2686 br->name, vlan, in_port->name, vlan);
2695 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2696 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2697 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2699 is_gratuitous_arp(const struct flow *flow)
2701 return (flow->dl_type == htons(ETH_TYPE_ARP)
2702 && eth_addr_is_broadcast(flow->dl_dst)
2703 && (flow->nw_proto == ARP_OP_REPLY
2704 || (flow->nw_proto == ARP_OP_REQUEST
2705 && flow->nw_src == flow->nw_dst)));
2709 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2710 struct port *in_port)
2712 enum grat_arp_lock_type lock_type;
2715 /* We don't want to learn from gratuitous ARP packets that are reflected
2716 * back over bond slaves so we lock the learning table. */
2717 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2718 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2719 GRAT_ARP_LOCK_CHECK;
2721 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2724 /* The log messages here could actually be useful in debugging,
2725 * so keep the rate limit relatively high. */
2726 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2728 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2729 "on port %s in VLAN %d",
2730 br->name, ETH_ADDR_ARGS(flow->dl_src),
2731 in_port->name, vlan);
2732 ofproto_revalidate(br->ofproto, rev_tag);
2736 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2737 * dropped. Returns true if they may be forwarded, false if they should be
2740 * If 'have_packet' is true, it indicates that the caller is processing a
2741 * received packet. If 'have_packet' is false, then the caller is just
2742 * revalidating an existing flow because configuration has changed. Either
2743 * way, 'have_packet' only affects logging (there is no point in logging errors
2744 * during revalidation).
2746 * Sets '*in_portp' to the input port. This will be a null pointer if
2747 * flow->in_port does not designate a known input port (in which case
2748 * is_admissible() returns false).
2750 * When returning true, sets '*vlanp' to the effective VLAN of the input
2751 * packet, as returned by flow_get_vlan().
2753 * May also add tags to '*tags', although the current implementation only does
2754 * so in one special case.
2757 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2758 tag_type *tags, int *vlanp, struct port **in_portp)
2760 struct iface *in_iface;
2761 struct port *in_port;
2764 /* Find the interface and port structure for the received packet. */
2765 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2767 /* No interface? Something fishy... */
2769 /* Odd. A few possible reasons here:
2771 * - We deleted an interface but there are still a few packets
2772 * queued up from it.
2774 * - Someone externally added an interface (e.g. with "ovs-dpctl
2775 * add-if") that we don't know about.
2777 * - Packet arrived on the local port but the local port is not
2778 * one of our bridge ports.
2780 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2782 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2783 "interface %"PRIu16, br->name, flow->in_port);
2789 *in_portp = in_port = in_iface->port;
2790 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2795 /* Drop frames for reserved multicast addresses. */
2796 if (eth_addr_is_reserved(flow->dl_dst)) {
2800 /* Drop frames on ports reserved for mirroring. */
2801 if (in_port->is_mirror_output_port) {
2803 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2804 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2805 "%s, which is reserved exclusively for mirroring",
2806 br->name, in_port->name);
2811 /* Packets received on bonds need special attention to avoid duplicates. */
2812 if (in_port->n_ifaces > 1) {
2814 bool is_grat_arp_locked;
2816 if (eth_addr_is_multicast(flow->dl_dst)) {
2817 *tags |= in_port->active_iface_tag;
2818 if (in_port->active_iface != in_iface->port_ifidx) {
2819 /* Drop all multicast packets on inactive slaves. */
2824 /* Drop all packets for which we have learned a different input
2825 * port, because we probably sent the packet on one slave and got
2826 * it back on the other. Gratuitous ARP packets are an exception
2827 * to this rule: the host has moved to another switch. The exception
2828 * to the exception is if we locked the learning table to avoid
2829 * reflections on bond slaves. If this is the case, just drop the
2831 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2832 &is_grat_arp_locked);
2833 if (src_idx != -1 && src_idx != in_port->port_idx &&
2834 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2842 /* If the composed actions may be applied to any packet in the given 'flow',
2843 * returns true. Otherwise, the actions should only be applied to 'packet', or
2844 * not at all, if 'packet' was NULL. */
2846 process_flow(struct bridge *br, const struct flow *flow,
2847 const struct ofpbuf *packet, struct ofpbuf *actions,
2848 tag_type *tags, uint16_t *nf_output_iface)
2850 struct port *in_port;
2851 struct port *out_port;
2855 /* Check whether we should drop packets in this flow. */
2856 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2861 /* Learn source MAC (but don't try to learn from revalidation). */
2863 update_learning_table(br, flow, vlan, in_port);
2866 /* Determine output port. */
2867 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2869 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2870 out_port = br->ports[out_port_idx];
2871 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2872 /* If we are revalidating but don't have a learning entry then
2873 * eject the flow. Installing a flow that floods packets opens
2874 * up a window of time where we could learn from a packet reflected
2875 * on a bond and blackhole packets before the learning table is
2876 * updated to reflect the correct port. */
2879 out_port = FLOOD_PORT;
2882 /* Don't send packets out their input ports. */
2883 if (in_port == out_port) {
2889 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2897 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2898 struct ofpbuf *actions, tag_type *tags,
2899 uint16_t *nf_output_iface, void *br_)
2901 struct iface *iface;
2902 struct bridge *br = br_;
2904 COVERAGE_INC(bridge_process_flow);
2906 iface = iface_from_dp_ifidx(br, flow->in_port);
2908 if (cfm_should_process_flow(flow)) {
2909 if (packet && iface->cfm) {
2910 cfm_process_heartbeat(iface->cfm, packet);
2915 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2919 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2920 const struct nlattr *actions,
2922 unsigned long long int n_bytes, void *br_)
2924 struct bridge *br = br_;
2925 const struct nlattr *a;
2926 struct port *in_port;
2931 /* Feed information from the active flows back into the learning table to
2932 * ensure that table is always in sync with what is actually flowing
2933 * through the datapath.
2935 * We test that 'tags' is nonzero to ensure that only flows that include an
2936 * OFPP_NORMAL action are used for learning. This works because
2937 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2938 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2939 update_learning_table(br, flow, vlan, in_port);
2942 /* Account for bond slave utilization. */
2943 if (!br->has_bonded_ports) {
2946 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2947 if (nl_attr_type(a) == ODPAT_OUTPUT) {
2948 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2949 if (out_port && out_port->n_ifaces >= 2 &&
2950 out_port->bond_mode == BM_SLB) {
2951 uint16_t vlan = (flow->vlan_tci
2952 ? vlan_tci_to_vid(flow->vlan_tci)
2954 struct bond_entry *e = lookup_bond_entry(out_port,
2955 flow->dl_src, vlan);
2956 e->tx_bytes += n_bytes;
2963 bridge_account_checkpoint_ofhook_cb(void *br_)
2965 struct bridge *br = br_;
2969 if (!br->has_bonded_ports) {
2974 for (i = 0; i < br->n_ports; i++) {
2975 struct port *port = br->ports[i];
2976 if (port->n_ifaces > 1 && port->bond_mode == BM_SLB
2977 && now >= port->bond_next_rebalance) {
2978 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2979 bond_rebalance_port(port);
2984 static struct ofhooks bridge_ofhooks = {
2985 bridge_normal_ofhook_cb,
2986 bridge_account_flow_ofhook_cb,
2987 bridge_account_checkpoint_ofhook_cb,
2990 /* Bonding functions. */
2992 /* Statistics for a single interface on a bonded port, used for load-based
2993 * bond rebalancing. */
2994 struct slave_balance {
2995 struct iface *iface; /* The interface. */
2996 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2998 /* All the "bond_entry"s that are assigned to this interface, in order of
2999 * increasing tx_bytes. */
3000 struct bond_entry **hashes;
3005 bond_mode_to_string(enum bond_mode bm) {
3006 static char *bm_slb = "balance-slb";
3007 static char *bm_ab = "active-backup";
3010 case BM_SLB: return bm_slb;
3011 case BM_AB: return bm_ab;
3018 /* Sorts pointers to pointers to bond_entries in ascending order by the
3019 * interface to which they are assigned, and within a single interface in
3020 * ascending order of bytes transmitted. */
3022 compare_bond_entries(const void *a_, const void *b_)
3024 const struct bond_entry *const *ap = a_;
3025 const struct bond_entry *const *bp = b_;
3026 const struct bond_entry *a = *ap;
3027 const struct bond_entry *b = *bp;
3028 if (a->iface_idx != b->iface_idx) {
3029 return a->iface_idx > b->iface_idx ? 1 : -1;
3030 } else if (a->tx_bytes != b->tx_bytes) {
3031 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3037 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3038 * *descending* order by number of bytes transmitted. */
3040 compare_slave_balance(const void *a_, const void *b_)
3042 const struct slave_balance *a = a_;
3043 const struct slave_balance *b = b_;
3044 if (a->iface->enabled != b->iface->enabled) {
3045 return a->iface->enabled ? -1 : 1;
3046 } else if (a->tx_bytes != b->tx_bytes) {
3047 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3054 swap_bals(struct slave_balance *a, struct slave_balance *b)
3056 struct slave_balance tmp = *a;
3061 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3062 * given that 'p' (and only 'p') might be in the wrong location.
3064 * This function invalidates 'p', since it might now be in a different memory
3067 resort_bals(struct slave_balance *p,
3068 struct slave_balance bals[], size_t n_bals)
3071 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3072 swap_bals(p, p - 1);
3074 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3075 swap_bals(p, p + 1);
3081 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3083 if (VLOG_IS_DBG_ENABLED()) {
3084 struct ds ds = DS_EMPTY_INITIALIZER;
3085 const struct slave_balance *b;
3087 for (b = bals; b < bals + n_bals; b++) {
3091 ds_put_char(&ds, ',');
3093 ds_put_format(&ds, " %s %"PRIu64"kB",
3094 b->iface->name, b->tx_bytes / 1024);
3096 if (!b->iface->enabled) {
3097 ds_put_cstr(&ds, " (disabled)");
3099 if (b->n_hashes > 0) {
3100 ds_put_cstr(&ds, " (");
3101 for (i = 0; i < b->n_hashes; i++) {
3102 const struct bond_entry *e = b->hashes[i];
3104 ds_put_cstr(&ds, " + ");
3106 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3107 e - port->bond_hash, e->tx_bytes / 1024);
3109 ds_put_cstr(&ds, ")");
3112 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3117 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3119 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3122 struct bond_entry *hash = from->hashes[hash_idx];
3123 struct port *port = from->iface->port;
3124 uint64_t delta = hash->tx_bytes;
3126 assert(port->bond_mode == BM_SLB);
3128 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3129 "from %s to %s (now carrying %"PRIu64"kB and "
3130 "%"PRIu64"kB load, respectively)",
3131 port->name, delta / 1024, hash - port->bond_hash,
3132 from->iface->name, to->iface->name,
3133 (from->tx_bytes - delta) / 1024,
3134 (to->tx_bytes + delta) / 1024);
3136 /* Delete element from from->hashes.
3138 * We don't bother to add the element to to->hashes because not only would
3139 * it require more work, the only purpose it would be to allow that hash to
3140 * be migrated to another slave in this rebalancing run, and there is no
3141 * point in doing that. */
3142 if (hash_idx == 0) {
3145 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3146 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3150 /* Shift load away from 'from' to 'to'. */
3151 from->tx_bytes -= delta;
3152 to->tx_bytes += delta;
3154 /* Arrange for flows to be revalidated. */
3155 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3156 hash->iface_idx = to->iface->port_ifidx;
3157 hash->iface_tag = tag_create_random();
3161 bond_rebalance_port(struct port *port)
3163 struct slave_balance *bals;
3165 struct bond_entry *hashes[BOND_MASK + 1];
3166 struct slave_balance *b, *from, *to;
3167 struct bond_entry *e;
3170 assert(port->bond_mode == BM_SLB);
3172 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3173 * descending order of tx_bytes, so that bals[0] represents the most
3174 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3177 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3178 * array for each slave_balance structure, we sort our local array of
3179 * hashes in order by slave, so that all of the hashes for a given slave
3180 * become contiguous in memory, and then we point each 'hashes' members of
3181 * a slave_balance structure to the start of a contiguous group. */
3182 n_bals = port->n_ifaces;
3183 bals = xmalloc(n_bals * sizeof *bals);
3184 for (b = bals; b < &bals[n_bals]; b++) {
3185 b->iface = port->ifaces[b - bals];
3190 for (i = 0; i <= BOND_MASK; i++) {
3191 hashes[i] = &port->bond_hash[i];
3193 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3194 for (i = 0; i <= BOND_MASK; i++) {
3196 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3197 b = &bals[e->iface_idx];
3198 b->tx_bytes += e->tx_bytes;
3200 b->hashes = &hashes[i];
3205 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3206 log_bals(bals, n_bals, port);
3208 /* Discard slaves that aren't enabled (which were sorted to the back of the
3209 * array earlier). */
3210 while (!bals[n_bals - 1].iface->enabled) {
3217 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3218 to = &bals[n_bals - 1];
3219 for (from = bals; from < to; ) {
3220 uint64_t overload = from->tx_bytes - to->tx_bytes;
3221 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3222 /* The extra load on 'from' (and all less-loaded slaves), compared
3223 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3224 * it is less than ~1Mbps. No point in rebalancing. */
3226 } else if (from->n_hashes == 1) {
3227 /* 'from' only carries a single MAC hash, so we can't shift any
3228 * load away from it, even though we want to. */
3231 /* 'from' is carrying significantly more load than 'to', and that
3232 * load is split across at least two different hashes. Pick a hash
3233 * to migrate to 'to' (the least-loaded slave), given that doing so
3234 * must decrease the ratio of the load on the two slaves by at
3237 * The sort order we use means that we prefer to shift away the
3238 * smallest hashes instead of the biggest ones. There is little
3239 * reason behind this decision; we could use the opposite sort
3240 * order to shift away big hashes ahead of small ones. */
3243 for (i = 0; i < from->n_hashes; i++) {
3244 double old_ratio, new_ratio;
3245 uint64_t delta = from->hashes[i]->tx_bytes;
3247 if (delta == 0 || from->tx_bytes - delta == 0) {
3248 /* Pointless move. */
3252 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3254 if (to->tx_bytes == 0) {
3255 /* Nothing on the new slave, move it. */
3259 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3260 new_ratio = (double)(from->tx_bytes - delta) /
3261 (to->tx_bytes + delta);
3263 if (new_ratio == 0) {
3264 /* Should already be covered but check to prevent division
3269 if (new_ratio < 1) {
3270 new_ratio = 1 / new_ratio;
3273 if (old_ratio - new_ratio > 0.1) {
3274 /* Would decrease the ratio, move it. */
3278 if (i < from->n_hashes) {
3279 bond_shift_load(from, to, i);
3280 port->bond_compat_is_stale = true;
3282 /* If the result of the migration changed the relative order of
3283 * 'from' and 'to' swap them back to maintain invariants. */
3284 if (order_swapped) {
3285 swap_bals(from, to);
3288 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3289 * point to different slave_balance structures. It is only
3290 * valid to do these two operations in a row at all because we
3291 * know that 'from' will not move past 'to' and vice versa. */
3292 resort_bals(from, bals, n_bals);
3293 resort_bals(to, bals, n_bals);
3300 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3301 * historical data to decay to <1% in 7 rebalancing runs. */
3302 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3311 bond_send_learning_packets(struct port *port)
3313 struct bridge *br = port->bridge;
3314 struct mac_entry *e;
3315 struct ofpbuf packet;
3316 int error, n_packets, n_errors;
3318 if (!port->n_ifaces || port->active_iface < 0) {
3322 ofpbuf_init(&packet, 128);
3323 error = n_packets = n_errors = 0;
3324 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3325 union ofp_action actions[2], *a;
3331 if (e->port == port->port_idx
3332 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3336 /* Compose actions. */
3337 memset(actions, 0, sizeof actions);
3340 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3341 a->vlan_vid.len = htons(sizeof *a);
3342 a->vlan_vid.vlan_vid = htons(e->vlan);
3345 a->output.type = htons(OFPAT_OUTPUT);
3346 a->output.len = htons(sizeof *a);
3347 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3352 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3354 flow_extract(&packet, 0, ODPP_NONE, &flow);
3355 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3362 ofpbuf_uninit(&packet);
3365 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3366 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3367 "packets, last error was: %s",
3368 port->name, n_errors, n_packets, strerror(error));
3370 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3371 port->name, n_packets);
3375 /* Bonding unixctl user interface functions. */
3378 bond_unixctl_list(struct unixctl_conn *conn,
3379 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3381 struct ds ds = DS_EMPTY_INITIALIZER;
3382 const struct bridge *br;
3384 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3386 LIST_FOR_EACH (br, node, &all_bridges) {
3389 for (i = 0; i < br->n_ports; i++) {
3390 const struct port *port = br->ports[i];
3391 if (port->n_ifaces > 1) {
3394 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3395 bond_mode_to_string(port->bond_mode));
3396 for (j = 0; j < port->n_ifaces; j++) {
3397 const struct iface *iface = port->ifaces[j];
3399 ds_put_cstr(&ds, ", ");
3401 ds_put_cstr(&ds, iface->name);
3403 ds_put_char(&ds, '\n');
3407 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3411 static struct port *
3412 bond_find(const char *name)
3414 const struct bridge *br;
3416 LIST_FOR_EACH (br, node, &all_bridges) {
3419 for (i = 0; i < br->n_ports; i++) {
3420 struct port *port = br->ports[i];
3421 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3430 bond_unixctl_show(struct unixctl_conn *conn,
3431 const char *args, void *aux OVS_UNUSED)
3433 struct ds ds = DS_EMPTY_INITIALIZER;
3434 const struct port *port;
3437 port = bond_find(args);
3439 unixctl_command_reply(conn, 501, "no such bond");
3443 ds_put_format(&ds, "bond_mode: %s\n",
3444 bond_mode_to_string(port->bond_mode));
3445 ds_put_format(&ds, "bond-detect-mode: %s\n",
3446 port->miimon ? "miimon" : "carrier");
3449 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3450 port->bond_miimon_interval);
3453 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3454 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3456 if (port->bond_mode == BM_SLB) {
3457 ds_put_format(&ds, "next rebalance: %lld ms\n",
3458 port->bond_next_rebalance - time_msec());
3461 for (j = 0; j < port->n_ifaces; j++) {
3462 const struct iface *iface = port->ifaces[j];
3463 struct bond_entry *be;
3466 ds_put_format(&ds, "slave %s: %s\n",
3467 iface->name, iface->enabled ? "enabled" : "disabled");
3468 if (j == port->active_iface) {
3469 ds_put_cstr(&ds, "\tactive slave\n");
3471 if (iface->delay_expires != LLONG_MAX) {
3472 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3473 iface->enabled ? "downdelay" : "updelay",
3474 iface->delay_expires - time_msec());
3477 if (port->bond_mode != BM_SLB) {
3482 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3483 int hash = be - port->bond_hash;
3484 struct mac_entry *me;
3486 if (be->iface_idx != j) {
3490 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3491 hash, be->tx_bytes / 1024);
3494 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3497 if (bond_hash(me->mac, me->vlan) == hash
3498 && me->port != port->port_idx
3499 && choose_output_iface(port, me->mac, me->vlan,
3501 && dp_ifidx == iface->dp_ifidx)
3503 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3504 ETH_ADDR_ARGS(me->mac));
3509 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3514 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3515 void *aux OVS_UNUSED)
3517 char *args = (char *) args_;
3518 char *save_ptr = NULL;
3519 char *bond_s, *hash_s, *slave_s;
3521 struct iface *iface;
3522 struct bond_entry *entry;
3525 bond_s = strtok_r(args, " ", &save_ptr);
3526 hash_s = strtok_r(NULL, " ", &save_ptr);
3527 slave_s = strtok_r(NULL, " ", &save_ptr);
3529 unixctl_command_reply(conn, 501,
3530 "usage: bond/migrate BOND HASH SLAVE");
3534 port = bond_find(bond_s);
3536 unixctl_command_reply(conn, 501, "no such bond");
3540 if (port->bond_mode != BM_SLB) {
3541 unixctl_command_reply(conn, 501, "not an SLB bond");
3545 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3546 hash = atoi(hash_s) & BOND_MASK;
3548 unixctl_command_reply(conn, 501, "bad hash");
3552 iface = port_lookup_iface(port, slave_s);
3554 unixctl_command_reply(conn, 501, "no such slave");
3558 if (!iface->enabled) {
3559 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3563 entry = &port->bond_hash[hash];
3564 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3565 entry->iface_idx = iface->port_ifidx;
3566 entry->iface_tag = tag_create_random();
3567 port->bond_compat_is_stale = true;
3568 unixctl_command_reply(conn, 200, "migrated");
3572 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3573 void *aux OVS_UNUSED)
3575 char *args = (char *) args_;
3576 char *save_ptr = NULL;
3577 char *bond_s, *slave_s;
3579 struct iface *iface;
3581 bond_s = strtok_r(args, " ", &save_ptr);
3582 slave_s = strtok_r(NULL, " ", &save_ptr);
3584 unixctl_command_reply(conn, 501,
3585 "usage: bond/set-active-slave BOND SLAVE");
3589 port = bond_find(bond_s);
3591 unixctl_command_reply(conn, 501, "no such bond");
3595 iface = port_lookup_iface(port, slave_s);
3597 unixctl_command_reply(conn, 501, "no such slave");
3601 if (!iface->enabled) {
3602 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3606 if (port->active_iface != iface->port_ifidx) {
3607 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3608 port->active_iface = iface->port_ifidx;
3609 port->active_iface_tag = tag_create_random();
3610 VLOG_INFO("port %s: active interface is now %s",
3611 port->name, iface->name);
3612 bond_send_learning_packets(port);
3613 unixctl_command_reply(conn, 200, "done");
3615 unixctl_command_reply(conn, 200, "no change");
3620 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3622 char *args = (char *) args_;
3623 char *save_ptr = NULL;
3624 char *bond_s, *slave_s;
3626 struct iface *iface;
3628 bond_s = strtok_r(args, " ", &save_ptr);
3629 slave_s = strtok_r(NULL, " ", &save_ptr);
3631 unixctl_command_reply(conn, 501,
3632 "usage: bond/enable/disable-slave BOND SLAVE");
3636 port = bond_find(bond_s);
3638 unixctl_command_reply(conn, 501, "no such bond");
3642 iface = port_lookup_iface(port, slave_s);
3644 unixctl_command_reply(conn, 501, "no such slave");
3648 bond_enable_slave(iface, enable);
3649 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3653 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3654 void *aux OVS_UNUSED)
3656 enable_slave(conn, args, true);
3660 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3661 void *aux OVS_UNUSED)
3663 enable_slave(conn, args, false);
3667 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3668 void *aux OVS_UNUSED)
3670 char *args = (char *) args_;
3671 uint8_t mac[ETH_ADDR_LEN];
3675 char *mac_s, *vlan_s;
3676 char *save_ptr = NULL;
3678 mac_s = strtok_r(args, " ", &save_ptr);
3679 vlan_s = strtok_r(NULL, " ", &save_ptr);
3682 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3683 unixctl_command_reply(conn, 501, "invalid vlan");
3687 vlan = OFP_VLAN_NONE;
3690 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3691 == ETH_ADDR_SCAN_COUNT) {
3692 hash = bond_hash(mac, vlan);
3694 hash_cstr = xasprintf("%u", hash);
3695 unixctl_command_reply(conn, 200, hash_cstr);
3698 unixctl_command_reply(conn, 501, "invalid mac");
3705 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3706 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3707 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3708 unixctl_command_register("bond/set-active-slave",
3709 bond_unixctl_set_active_slave, NULL);
3710 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3712 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3714 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3717 /* Port functions. */
3719 static struct port *
3720 port_create(struct bridge *br, const char *name)
3724 port = xzalloc(sizeof *port);
3726 port->port_idx = br->n_ports;
3728 port->trunks = NULL;
3729 port->name = xstrdup(name);
3730 port->active_iface = -1;
3732 if (br->n_ports >= br->allocated_ports) {
3733 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3736 br->ports[br->n_ports++] = port;
3737 shash_add_assert(&br->port_by_name, port->name, port);
3739 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3746 get_port_other_config(const struct ovsrec_port *port, const char *key,
3747 const char *default_value)
3751 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3753 return value ? value : default_value;
3757 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3759 struct shash new_ifaces;
3762 /* Collect list of new interfaces. */
3763 shash_init(&new_ifaces);
3764 for (i = 0; i < cfg->n_interfaces; i++) {
3765 const char *name = cfg->interfaces[i]->name;
3766 shash_add_once(&new_ifaces, name, NULL);
3769 /* Get rid of deleted interfaces. */
3770 for (i = 0; i < port->n_ifaces; ) {
3771 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3772 iface_destroy(port->ifaces[i]);
3778 shash_destroy(&new_ifaces);
3782 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3784 const char *detect_mode;
3785 struct shash new_ifaces;
3786 long long int next_rebalance, miimon_next_update;
3787 unsigned long *trunks;
3793 /* Update settings. */
3794 port->updelay = cfg->bond_updelay;
3795 if (port->updelay < 0) {
3798 port->downdelay = cfg->bond_downdelay;
3799 if (port->downdelay < 0) {
3800 port->downdelay = 0;
3802 port->bond_rebalance_interval = atoi(
3803 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3804 if (port->bond_rebalance_interval < 1000) {
3805 port->bond_rebalance_interval = 1000;
3807 next_rebalance = time_msec() + port->bond_rebalance_interval;
3808 if (port->bond_next_rebalance > next_rebalance) {
3809 port->bond_next_rebalance = next_rebalance;
3812 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
3815 if (!strcmp(detect_mode, "carrier")) {
3816 port->miimon = false;
3817 } else if (!strcmp(detect_mode, "miimon")) {
3818 port->miimon = true;
3820 port->miimon = false;
3821 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
3822 "carrier", port->name, detect_mode);
3825 port->bond_miimon_interval = atoi(
3826 get_port_other_config(cfg, "bond-miimon-interval", "200"));
3827 if (port->bond_miimon_interval < 100) {
3828 port->bond_miimon_interval = 100;
3830 miimon_next_update = time_msec() + port->bond_miimon_interval;
3831 if (port->bond_miimon_next_update > miimon_next_update) {
3832 port->bond_miimon_next_update = miimon_next_update;
3835 if (!port->cfg->bond_mode ||
3836 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
3837 port->bond_mode = BM_SLB;
3838 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
3839 port->bond_mode = BM_AB;
3841 port->bond_mode = BM_SLB;
3842 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
3843 port->name, port->cfg->bond_mode,
3844 bond_mode_to_string(port->bond_mode));
3847 /* Add new interfaces and update 'cfg' member of existing ones. */
3848 shash_init(&new_ifaces);
3849 for (i = 0; i < cfg->n_interfaces; i++) {
3850 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3851 struct iface *iface;
3853 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3854 VLOG_WARN("port %s: %s specified twice as port interface",
3855 port->name, if_cfg->name);
3856 iface_set_ofport(if_cfg, -1);
3860 iface = iface_lookup(port->bridge, if_cfg->name);
3862 if (iface->port != port) {
3863 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3865 port->bridge->name, if_cfg->name, iface->port->name);
3868 iface->cfg = if_cfg;
3870 iface = iface_create(port, if_cfg);
3873 /* Determine interface type. The local port always has type
3874 * "internal". Other ports take their type from the database and
3875 * default to "system" if none is specified. */
3876 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3877 : if_cfg->type[0] ? if_cfg->type
3880 shash_destroy(&new_ifaces);
3885 if (port->n_ifaces < 2) {
3887 if (vlan >= 0 && vlan <= 4095) {
3888 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3893 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3894 * they even work as-is. But they have not been tested. */
3895 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3899 if (port->vlan != vlan) {
3901 bridge_flush(port->bridge);
3904 /* Get trunked VLANs. */
3906 if (vlan < 0 && cfg->n_trunks) {
3909 trunks = bitmap_allocate(4096);
3911 for (i = 0; i < cfg->n_trunks; i++) {
3912 int trunk = cfg->trunks[i];
3914 bitmap_set1(trunks, trunk);
3920 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3921 port->name, cfg->n_trunks);
3923 if (n_errors == cfg->n_trunks) {
3924 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3926 bitmap_free(trunks);
3929 } else if (vlan >= 0 && cfg->n_trunks) {
3930 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3934 ? port->trunks != NULL
3935 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3936 bridge_flush(port->bridge);
3938 bitmap_free(port->trunks);
3939 port->trunks = trunks;
3943 port_destroy(struct port *port)
3946 struct bridge *br = port->bridge;
3950 proc_net_compat_update_vlan(port->name, NULL, 0);
3951 proc_net_compat_update_bond(port->name, NULL);
3953 for (i = 0; i < MAX_MIRRORS; i++) {
3954 struct mirror *m = br->mirrors[i];
3955 if (m && m->out_port == port) {
3960 while (port->n_ifaces > 0) {
3961 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3964 shash_find_and_delete_assert(&br->port_by_name, port->name);
3966 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3967 del->port_idx = port->port_idx;
3969 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3971 netdev_monitor_destroy(port->monitor);
3973 bitmap_free(port->trunks);
3980 static struct port *
3981 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3983 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3984 return iface ? iface->port : NULL;
3987 static struct port *
3988 port_lookup(const struct bridge *br, const char *name)
3990 return shash_find_data(&br->port_by_name, name);
3993 static struct iface *
3994 port_lookup_iface(const struct port *port, const char *name)
3996 struct iface *iface = iface_lookup(port->bridge, name);
3997 return iface && iface->port == port ? iface : NULL;
4001 port_update_bonding(struct port *port)
4003 if (port->monitor) {
4004 netdev_monitor_destroy(port->monitor);
4005 port->monitor = NULL;
4007 if (port->n_ifaces < 2) {
4008 /* Not a bonded port. */
4009 if (port->bond_hash) {
4010 free(port->bond_hash);
4011 port->bond_hash = NULL;
4012 port->bond_compat_is_stale = true;
4015 port->bond_fake_iface = false;
4019 if (port->bond_mode == BM_SLB && !port->bond_hash) {
4020 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4021 for (i = 0; i <= BOND_MASK; i++) {
4022 struct bond_entry *e = &port->bond_hash[i];
4026 port->no_ifaces_tag = tag_create_random();
4027 bond_choose_active_iface(port);
4028 port->bond_next_rebalance
4029 = time_msec() + port->bond_rebalance_interval;
4031 if (port->cfg->bond_fake_iface) {
4032 port->bond_next_fake_iface_update = time_msec();
4034 } else if (port->bond_mode != BM_SLB) {
4035 free(port->bond_hash);
4036 port->bond_hash = NULL;
4038 port->bond_compat_is_stale = true;
4039 port->bond_fake_iface = port->cfg->bond_fake_iface;
4041 if (!port->miimon) {
4042 port->monitor = netdev_monitor_create();
4043 for (i = 0; i < port->n_ifaces; i++) {
4044 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4051 port_update_bond_compat(struct port *port)
4053 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
4054 struct compat_bond bond;
4057 if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) {
4058 proc_net_compat_update_bond(port->name, NULL);
4063 bond.updelay = port->updelay;
4064 bond.downdelay = port->downdelay;
4067 bond.hashes = compat_hashes;
4068 if (port->bond_hash) {
4069 const struct bond_entry *e;
4070 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4071 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
4072 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
4073 cbh->hash = e - port->bond_hash;
4074 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
4079 bond.n_slaves = port->n_ifaces;
4080 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
4081 for (i = 0; i < port->n_ifaces; i++) {
4082 struct iface *iface = port->ifaces[i];
4083 struct compat_bond_slave *slave = &bond.slaves[i];
4084 slave->name = iface->name;
4086 /* We need to make the same determination as the Linux bonding
4087 * code to determine whether a slave should be consider "up".
4088 * The Linux function bond_miimon_inspect() supports four
4089 * BOND_LINK_* states:
4091 * - BOND_LINK_UP: carrier detected, updelay has passed.
4092 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
4093 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
4094 * - BOND_LINK_BACK: carrier detected, updelay in progress.
4096 * The function bond_info_show_slave() only considers BOND_LINK_UP
4097 * to be "up" and anything else to be "down".
4099 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
4103 netdev_get_etheraddr(iface->netdev, slave->mac);
4106 if (port->bond_fake_iface) {
4107 struct netdev *bond_netdev;
4109 if (!netdev_open_default(port->name, &bond_netdev)) {
4111 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
4113 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
4115 netdev_close(bond_netdev);
4119 proc_net_compat_update_bond(port->name, &bond);
4124 port_update_vlan_compat(struct port *port)
4126 struct bridge *br = port->bridge;
4127 char *vlandev_name = NULL;
4129 if (port->vlan > 0) {
4130 /* Figure out the name that the VLAN device should actually have, if it
4131 * existed. This takes some work because the VLAN device would not
4132 * have port->name in its name; rather, it would have the trunk port's
4133 * name, and 'port' would be attached to a bridge that also had the
4134 * VLAN device one of its ports. So we need to find a trunk port that
4135 * includes port->vlan.
4137 * There might be more than one candidate. This doesn't happen on
4138 * XenServer, so if it happens we just pick the first choice in
4139 * alphabetical order instead of creating multiple VLAN devices. */
4141 for (i = 0; i < br->n_ports; i++) {
4142 struct port *p = br->ports[i];
4143 if (port_trunks_vlan(p, port->vlan)
4145 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
4147 uint8_t ea[ETH_ADDR_LEN];
4148 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
4149 if (!eth_addr_is_multicast(ea) &&
4150 !eth_addr_is_reserved(ea) &&
4151 !eth_addr_is_zero(ea)) {
4152 vlandev_name = p->name;
4157 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
4160 /* Interface functions. */
4163 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4166 union ofp_action action;
4168 memset(&action, 0, sizeof action);
4169 action.output.type = htons(OFPAT_OUTPUT);
4170 action.output.len = htons(sizeof action);
4171 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4173 flow_extract(packet, 0, ODPP_NONE, &flow);
4175 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4177 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4178 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4182 static struct iface *
4183 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4185 struct bridge *br = port->bridge;
4186 struct iface *iface;
4187 char *name = if_cfg->name;
4189 iface = xzalloc(sizeof *iface);
4191 iface->port_ifidx = port->n_ifaces;
4192 iface->name = xstrdup(name);
4193 iface->dp_ifidx = -1;
4194 iface->tag = tag_create_random();
4195 iface->delay_expires = LLONG_MAX;
4196 iface->netdev = NULL;
4197 iface->cfg = if_cfg;
4199 shash_add_assert(&br->iface_by_name, iface->name, iface);
4201 if (port->n_ifaces >= port->allocated_ifaces) {
4202 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4203 sizeof *port->ifaces);
4205 port->ifaces[port->n_ifaces++] = iface;
4206 if (port->n_ifaces > 1) {
4207 br->has_bonded_ports = true;
4210 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4218 iface_destroy(struct iface *iface)
4221 struct port *port = iface->port;
4222 struct bridge *br = port->bridge;
4223 bool del_active = port->active_iface == iface->port_ifidx;
4226 if (port->monitor) {
4227 netdev_monitor_remove(port->monitor, iface->netdev);
4230 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4232 if (iface->dp_ifidx >= 0) {
4233 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4236 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4237 del->port_ifidx = iface->port_ifidx;
4239 netdev_close(iface->netdev);
4242 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4243 bond_choose_active_iface(port);
4244 bond_send_learning_packets(port);
4247 cfm_destroy(iface->cfm);
4252 bridge_flush(port->bridge);
4256 static struct iface *
4257 iface_lookup(const struct bridge *br, const char *name)
4259 return shash_find_data(&br->iface_by_name, name);
4262 static struct iface *
4263 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4265 struct iface *iface;
4267 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4268 hash_int(dp_ifidx, 0), &br->ifaces) {
4269 if (iface->dp_ifidx == dp_ifidx) {
4276 /* Set Ethernet address of 'iface', if one is specified in the configuration
4279 iface_set_mac(struct iface *iface)
4281 uint8_t ea[ETH_ADDR_LEN];
4283 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4284 if (eth_addr_is_multicast(ea)) {
4285 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4287 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4288 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4289 iface->name, iface->name);
4291 int error = netdev_set_etheraddr(iface->netdev, ea);
4293 VLOG_ERR("interface %s: setting MAC failed (%s)",
4294 iface->name, strerror(error));
4300 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4302 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4305 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4309 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4311 * The value strings in '*shash' are taken directly from values[], not copied,
4312 * so the caller should not modify or free them. */
4314 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4315 struct shash *shash)
4320 for (i = 0; i < n; i++) {
4321 shash_add(shash, keys[i], values[i]);
4325 /* Creates 'keys' and 'values' arrays from 'shash'.
4327 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4328 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4329 * are populated with with strings taken directly from 'shash' and thus have
4330 * the same ownership of the key-value pairs in shash.
4333 shash_to_ovs_idl_map(struct shash *shash,
4334 char ***keys, char ***values, size_t *n)
4338 struct shash_node *sn;
4340 count = shash_count(shash);
4342 k = xmalloc(count * sizeof *k);
4343 v = xmalloc(count * sizeof *v);
4346 SHASH_FOR_EACH(sn, shash) {
4357 struct iface_delete_queues_cbdata {
4358 struct netdev *netdev;
4359 const struct ovsdb_datum *queues;
4363 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4365 union ovsdb_atom atom;
4367 atom.integer = target;
4368 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4372 iface_delete_queues(unsigned int queue_id,
4373 const struct shash *details OVS_UNUSED, void *cbdata_)
4375 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4377 if (!queue_ids_include(cbdata->queues, queue_id)) {
4378 netdev_delete_queue(cbdata->netdev, queue_id);
4383 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4385 if (!qos || qos->type[0] == '\0') {
4386 netdev_set_qos(iface->netdev, NULL, NULL);
4388 struct iface_delete_queues_cbdata cbdata;
4389 struct shash details;
4392 /* Configure top-level Qos for 'iface'. */
4393 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4394 qos->n_other_config, &details);
4395 netdev_set_qos(iface->netdev, qos->type, &details);
4396 shash_destroy(&details);
4398 /* Deconfigure queues that were deleted. */
4399 cbdata.netdev = iface->netdev;
4400 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4402 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4404 /* Configure queues for 'iface'. */
4405 for (i = 0; i < qos->n_queues; i++) {
4406 const struct ovsrec_queue *queue = qos->value_queues[i];
4407 unsigned int queue_id = qos->key_queues[i];
4409 shash_from_ovs_idl_map(queue->key_other_config,
4410 queue->value_other_config,
4411 queue->n_other_config, &details);
4412 netdev_set_queue(iface->netdev, queue_id, &details);
4413 shash_destroy(&details);
4419 iface_update_cfm(struct iface *iface)
4423 uint16_t *remote_mps;
4424 struct ovsrec_monitor *mon;
4425 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4427 mon = iface->cfg->monitor;
4433 if (netdev_get_etheraddr(iface->netdev, ea)) {
4434 VLOG_WARN("interface %s: Failed to get ethernet address. "
4435 "Skipping Monitor.", iface->name);
4439 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4440 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4445 iface->cfm = cfm_create();
4449 cfm->mpid = mon->mpid;
4450 cfm->interval = mon->interval ? *mon->interval : 1000;
4452 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4453 memcpy(cfm->maid, maid, sizeof cfm->maid);
4455 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4456 for(i = 0; i < mon->n_remote_mps; i++) {
4457 remote_mps[i] = mon->remote_mps[i]->mpid;
4459 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4462 if (!cfm_configure(iface->cfm)) {
4463 cfm_destroy(iface->cfm);
4468 /* Port mirroring. */
4470 static struct mirror *
4471 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4475 for (i = 0; i < MAX_MIRRORS; i++) {
4476 struct mirror *m = br->mirrors[i];
4477 if (m && uuid_equals(uuid, &m->uuid)) {
4485 mirror_reconfigure(struct bridge *br)
4487 unsigned long *rspan_vlans;
4490 /* Get rid of deleted mirrors. */
4491 for (i = 0; i < MAX_MIRRORS; i++) {
4492 struct mirror *m = br->mirrors[i];
4494 const struct ovsdb_datum *mc;
4495 union ovsdb_atom atom;
4497 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4498 atom.uuid = br->mirrors[i]->uuid;
4499 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4505 /* Add new mirrors and reconfigure existing ones. */
4506 for (i = 0; i < br->cfg->n_mirrors; i++) {
4507 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4508 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4510 mirror_reconfigure_one(m, cfg);
4512 mirror_create(br, cfg);
4516 /* Update port reserved status. */
4517 for (i = 0; i < br->n_ports; i++) {
4518 br->ports[i]->is_mirror_output_port = false;
4520 for (i = 0; i < MAX_MIRRORS; i++) {
4521 struct mirror *m = br->mirrors[i];
4522 if (m && m->out_port) {
4523 m->out_port->is_mirror_output_port = true;
4527 /* Update flooded vlans (for RSPAN). */
4529 if (br->cfg->n_flood_vlans) {
4530 rspan_vlans = bitmap_allocate(4096);
4532 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4533 int64_t vlan = br->cfg->flood_vlans[i];
4534 if (vlan >= 0 && vlan < 4096) {
4535 bitmap_set1(rspan_vlans, vlan);
4536 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4539 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4544 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4550 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4555 for (i = 0; ; i++) {
4556 if (i >= MAX_MIRRORS) {
4557 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4558 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4561 if (!br->mirrors[i]) {
4566 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4569 br->mirrors[i] = m = xzalloc(sizeof *m);
4572 m->name = xstrdup(cfg->name);
4573 shash_init(&m->src_ports);
4574 shash_init(&m->dst_ports);
4580 mirror_reconfigure_one(m, cfg);
4584 mirror_destroy(struct mirror *m)
4587 struct bridge *br = m->bridge;
4590 for (i = 0; i < br->n_ports; i++) {
4591 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4592 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4595 shash_destroy(&m->src_ports);
4596 shash_destroy(&m->dst_ports);
4599 m->bridge->mirrors[m->idx] = NULL;
4608 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4609 struct shash *names)
4613 for (i = 0; i < n_ports; i++) {
4614 const char *name = ports[i]->name;
4615 if (port_lookup(m->bridge, name)) {
4616 shash_add_once(names, name, NULL);
4618 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4619 "port %s", m->bridge->name, m->name, name);
4625 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4631 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4633 for (i = 0; i < cfg->n_select_vlan; i++) {
4634 int64_t vlan = cfg->select_vlan[i];
4635 if (vlan < 0 || vlan > 4095) {
4636 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4637 m->bridge->name, m->name, vlan);
4639 (*vlans)[n_vlans++] = vlan;
4646 vlan_is_mirrored(const struct mirror *m, int vlan)
4650 for (i = 0; i < m->n_vlans; i++) {
4651 if (m->vlans[i] == vlan) {
4659 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4663 for (i = 0; i < m->n_vlans; i++) {
4664 if (port_trunks_vlan(p, m->vlans[i])) {
4672 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4674 struct shash src_ports, dst_ports;
4675 mirror_mask_t mirror_bit;
4676 struct port *out_port;
4683 if (strcmp(cfg->name, m->name)) {
4685 m->name = xstrdup(cfg->name);
4688 /* Get output port. */
4689 if (cfg->output_port) {
4690 out_port = port_lookup(m->bridge, cfg->output_port->name);
4692 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4693 m->bridge->name, m->name);
4699 if (cfg->output_vlan) {
4700 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4701 "output vlan; ignoring output vlan",
4702 m->bridge->name, m->name);
4704 } else if (cfg->output_vlan) {
4706 out_vlan = *cfg->output_vlan;
4708 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4709 m->bridge->name, m->name);
4714 shash_init(&src_ports);
4715 shash_init(&dst_ports);
4716 if (cfg->select_all) {
4717 for (i = 0; i < m->bridge->n_ports; i++) {
4718 const char *name = m->bridge->ports[i]->name;
4719 shash_add_once(&src_ports, name, NULL);
4720 shash_add_once(&dst_ports, name, NULL);
4725 /* Get ports, and drop duplicates and ports that don't exist. */
4726 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4728 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4731 /* Get all the vlans, and drop duplicate and invalid vlans. */
4732 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4735 /* Update mirror data. */
4736 if (!shash_equal_keys(&m->src_ports, &src_ports)
4737 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4738 || m->n_vlans != n_vlans
4739 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4740 || m->out_port != out_port
4741 || m->out_vlan != out_vlan) {
4742 bridge_flush(m->bridge);
4744 shash_swap(&m->src_ports, &src_ports);
4745 shash_swap(&m->dst_ports, &dst_ports);
4748 m->n_vlans = n_vlans;
4749 m->out_port = out_port;
4750 m->out_vlan = out_vlan;
4753 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4754 for (i = 0; i < m->bridge->n_ports; i++) {
4755 struct port *port = m->bridge->ports[i];
4757 if (shash_find(&m->src_ports, port->name)
4760 ? port_trunks_any_mirrored_vlan(m, port)
4761 : vlan_is_mirrored(m, port->vlan)))) {
4762 port->src_mirrors |= mirror_bit;
4764 port->src_mirrors &= ~mirror_bit;
4767 if (shash_find(&m->dst_ports, port->name)) {
4768 port->dst_mirrors |= mirror_bit;
4770 port->dst_mirrors &= ~mirror_bit;
4775 shash_destroy(&src_ports);
4776 shash_destroy(&dst_ports);