1 /* Copyright (c) 2008, 2009, 2010, 2011 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 dpif_port_dump dump;
610 struct shash want_ifaces;
611 struct odp_port odp_port;
613 bridge_get_all_ifaces(br, &want_ifaces);
614 DPIF_PORT_FOR_EACH (&odp_port, &dump, br->dpif) {
615 if (!shash_find(&want_ifaces, odp_port.devname)
616 && strcmp(odp_port.devname, br->name)) {
617 int retval = dpif_port_del(br->dpif, odp_port.port);
619 VLOG_ERR("failed to remove %s interface from %s: %s",
620 odp_port.devname, dpif_name(br->dpif),
625 shash_destroy(&want_ifaces);
627 LIST_FOR_EACH (br, node, &all_bridges) {
629 char *type; /* Network device type, e.g. "system". */
630 uint32_t port_no; /* Port number within datapath. */
632 struct shash cur_ifaces, want_ifaces;
633 struct dpif_port_dump dump;
634 struct odp_port odp_port;
636 /* Get the set of interfaces currently in this datapath. */
637 shash_init(&cur_ifaces);
638 DPIF_PORT_FOR_EACH (&odp_port, &dump, br->dpif) {
639 struct dpif_port *port_info = xmalloc(sizeof *port_info);
640 port_info->port_no = odp_port.port;
641 port_info->type = xstrdup(odp_port.type);
642 shash_add(&cur_ifaces, odp_port.devname, port_info);
645 /* Get the set of interfaces we want on this datapath. */
646 bridge_get_all_ifaces(br, &want_ifaces);
648 hmap_clear(&br->ifaces);
649 SHASH_FOR_EACH (node, &want_ifaces) {
650 const char *if_name = node->name;
651 struct iface *iface = node->data;
652 struct dpif_port *dpif_port;
656 type = iface ? iface->type : "internal";
657 dpif_port = shash_find_data(&cur_ifaces, if_name);
659 /* If we have a port or a netdev already, and it's not the type we
660 * want, then delete the port (if any) and close the netdev (if
662 if ((dpif_port && strcmp(dpif_port->type, type))
663 || (iface && iface->netdev
664 && strcmp(type, netdev_get_type(iface->netdev)))) {
666 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
673 netdev_close(iface->netdev);
674 iface->netdev = NULL;
678 /* If the port doesn't exist or we don't have the netdev open,
679 * we need to do more work. */
680 if (!dpif_port || (iface && !iface->netdev)) {
681 struct netdev_options options;
682 struct netdev *netdev;
685 /* First open the network device. */
686 options.name = if_name;
688 options.args = &args;
689 options.ethertype = NETDEV_ETH_TYPE_NONE;
693 shash_from_ovs_idl_map(iface->cfg->key_options,
694 iface->cfg->value_options,
695 iface->cfg->n_options, &args);
697 error = netdev_open(&options, &netdev);
698 shash_destroy(&args);
701 VLOG_WARN("could not open network device %s (%s)",
702 if_name, strerror(error));
706 /* Then add the port if we haven't already. */
708 error = dpif_port_add(br->dpif, netdev, NULL);
710 netdev_close(netdev);
711 if (error == EFBIG) {
712 VLOG_ERR("ran out of valid port numbers on %s",
713 dpif_name(br->dpif));
716 VLOG_ERR("failed to add %s interface to %s: %s",
717 if_name, dpif_name(br->dpif),
724 /* Update 'iface'. */
726 iface->netdev = netdev;
727 iface->enabled = netdev_get_carrier(iface->netdev);
729 } else if (iface && iface->netdev) {
733 shash_from_ovs_idl_map(iface->cfg->key_options,
734 iface->cfg->value_options,
735 iface->cfg->n_options, &args);
736 netdev_set_config(iface->netdev, &args);
737 shash_destroy(&args);
740 shash_destroy(&want_ifaces);
742 SHASH_FOR_EACH (node, &cur_ifaces) {
743 struct dpif_port *port_info = node->data;
744 free(port_info->type);
747 shash_destroy(&cur_ifaces);
749 sflow_bridge_number = 0;
750 LIST_FOR_EACH (br, node, &all_bridges) {
753 struct iface *local_iface;
754 struct iface *hw_addr_iface;
757 bridge_fetch_dp_ifaces(br);
759 iterate_and_prune_ifaces(br, check_iface, NULL);
761 /* Pick local port hardware address, datapath ID. */
762 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
763 local_iface = bridge_get_local_iface(br);
765 int error = netdev_set_etheraddr(local_iface->netdev, ea);
767 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
768 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
769 "Ethernet address: %s",
770 br->name, strerror(error));
774 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
775 ofproto_set_datapath_id(br->ofproto, dpid);
777 dpid_string = xasprintf("%016"PRIx64, dpid);
778 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
781 /* Set NetFlow configuration on this bridge. */
782 if (br->cfg->netflow) {
783 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
784 struct netflow_options opts;
786 memset(&opts, 0, sizeof opts);
788 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
789 if (nf_cfg->engine_type) {
790 opts.engine_type = *nf_cfg->engine_type;
792 if (nf_cfg->engine_id) {
793 opts.engine_id = *nf_cfg->engine_id;
796 opts.active_timeout = nf_cfg->active_timeout;
797 if (!opts.active_timeout) {
798 opts.active_timeout = -1;
799 } else if (opts.active_timeout < 0) {
800 VLOG_WARN("bridge %s: active timeout interval set to negative "
801 "value, using default instead (%d seconds)", br->name,
802 NF_ACTIVE_TIMEOUT_DEFAULT);
803 opts.active_timeout = -1;
806 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
807 if (opts.add_id_to_iface) {
808 if (opts.engine_id > 0x7f) {
809 VLOG_WARN("bridge %s: netflow port mangling may conflict "
810 "with another vswitch, choose an engine id less "
811 "than 128", br->name);
813 if (br->n_ports > 508) {
814 VLOG_WARN("bridge %s: netflow port mangling will conflict "
815 "with another port when more than 508 ports are "
820 opts.collectors.n = nf_cfg->n_targets;
821 opts.collectors.names = nf_cfg->targets;
822 if (ofproto_set_netflow(br->ofproto, &opts)) {
823 VLOG_ERR("bridge %s: problem setting netflow collectors",
827 ofproto_set_netflow(br->ofproto, NULL);
830 /* Set sFlow configuration on this bridge. */
831 if (br->cfg->sflow) {
832 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
833 struct ovsrec_controller **controllers;
834 struct ofproto_sflow_options oso;
835 size_t n_controllers;
837 memset(&oso, 0, sizeof oso);
839 oso.targets.n = sflow_cfg->n_targets;
840 oso.targets.names = sflow_cfg->targets;
842 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
843 if (sflow_cfg->sampling) {
844 oso.sampling_rate = *sflow_cfg->sampling;
847 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
848 if (sflow_cfg->polling) {
849 oso.polling_interval = *sflow_cfg->polling;
852 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
853 if (sflow_cfg->header) {
854 oso.header_len = *sflow_cfg->header;
857 oso.sub_id = sflow_bridge_number++;
858 oso.agent_device = sflow_cfg->agent;
860 oso.control_ip = NULL;
861 n_controllers = bridge_get_controllers(br, &controllers);
862 for (i = 0; i < n_controllers; i++) {
863 if (controllers[i]->local_ip) {
864 oso.control_ip = controllers[i]->local_ip;
868 ofproto_set_sflow(br->ofproto, &oso);
870 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
872 ofproto_set_sflow(br->ofproto, NULL);
875 /* Update the controller and related settings. It would be more
876 * straightforward to call this from bridge_reconfigure_one(), but we
877 * can't do it there for two reasons. First, and most importantly, at
878 * that point we don't know the dp_ifidx of any interfaces that have
879 * been added to the bridge (because we haven't actually added them to
880 * the datapath). Second, at that point we haven't set the datapath ID
881 * yet; when a controller is configured, resetting the datapath ID will
882 * immediately disconnect from the controller, so it's better to set
883 * the datapath ID before the controller. */
884 bridge_reconfigure_remotes(br, managers, n_managers);
886 LIST_FOR_EACH (br, node, &all_bridges) {
887 for (i = 0; i < br->n_ports; i++) {
888 struct port *port = br->ports[i];
891 port_update_vlan_compat(port);
892 port_update_bonding(port);
894 for (j = 0; j < port->n_ifaces; j++) {
895 iface_update_qos(port->ifaces[j], port->cfg->qos);
899 LIST_FOR_EACH (br, node, &all_bridges) {
900 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
903 LIST_FOR_EACH (br, node, &all_bridges) {
905 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
906 iface_update_cfm(iface);
914 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
915 const struct ovsdb_idl_column *column,
918 const struct ovsdb_datum *datum;
919 union ovsdb_atom atom;
922 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
923 atom.string = (char *) key;
924 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
925 return idx == UINT_MAX ? NULL : datum->values[idx].string;
929 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
931 return get_ovsrec_key_value(&br_cfg->header_,
932 &ovsrec_bridge_col_other_config, key);
936 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
937 struct iface **hw_addr_iface)
943 *hw_addr_iface = NULL;
945 /* Did the user request a particular MAC? */
946 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
947 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
948 if (eth_addr_is_multicast(ea)) {
949 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
950 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
951 } else if (eth_addr_is_zero(ea)) {
952 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
958 /* Otherwise choose the minimum non-local MAC address among all of the
960 memset(ea, 0xff, sizeof ea);
961 for (i = 0; i < br->n_ports; i++) {
962 struct port *port = br->ports[i];
963 uint8_t iface_ea[ETH_ADDR_LEN];
966 /* Mirror output ports don't participate. */
967 if (port->is_mirror_output_port) {
971 /* Choose the MAC address to represent the port. */
972 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
973 /* Find the interface with this Ethernet address (if any) so that
974 * we can provide the correct devname to the caller. */
976 for (j = 0; j < port->n_ifaces; j++) {
977 struct iface *candidate = port->ifaces[j];
978 uint8_t candidate_ea[ETH_ADDR_LEN];
979 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
980 && eth_addr_equals(iface_ea, candidate_ea)) {
985 /* Choose the interface whose MAC address will represent the port.
986 * The Linux kernel bonding code always chooses the MAC address of
987 * the first slave added to a bond, and the Fedora networking
988 * scripts always add slaves to a bond in alphabetical order, so
989 * for compatibility we choose the interface with the name that is
990 * first in alphabetical order. */
991 iface = port->ifaces[0];
992 for (j = 1; j < port->n_ifaces; j++) {
993 struct iface *candidate = port->ifaces[j];
994 if (strcmp(candidate->name, iface->name) < 0) {
999 /* The local port doesn't count (since we're trying to choose its
1000 * MAC address anyway). */
1001 if (iface->dp_ifidx == ODPP_LOCAL) {
1006 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1008 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1009 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1010 iface->name, strerror(error));
1015 /* Compare against our current choice. */
1016 if (!eth_addr_is_multicast(iface_ea) &&
1017 !eth_addr_is_local(iface_ea) &&
1018 !eth_addr_is_reserved(iface_ea) &&
1019 !eth_addr_is_zero(iface_ea) &&
1020 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
1022 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1023 *hw_addr_iface = iface;
1026 if (eth_addr_is_multicast(ea)) {
1027 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1028 *hw_addr_iface = NULL;
1029 VLOG_WARN("bridge %s: using default bridge Ethernet "
1030 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1032 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1033 br->name, ETH_ADDR_ARGS(ea));
1037 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1038 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1039 * an interface on 'br', then that interface must be passed in as
1040 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1041 * 'hw_addr_iface' must be passed in as a null pointer. */
1043 bridge_pick_datapath_id(struct bridge *br,
1044 const uint8_t bridge_ea[ETH_ADDR_LEN],
1045 struct iface *hw_addr_iface)
1048 * The procedure for choosing a bridge MAC address will, in the most
1049 * ordinary case, also choose a unique MAC that we can use as a datapath
1050 * ID. In some special cases, though, multiple bridges will end up with
1051 * the same MAC address. This is OK for the bridges, but it will confuse
1052 * the OpenFlow controller, because each datapath needs a unique datapath
1055 * Datapath IDs must be unique. It is also very desirable that they be
1056 * stable from one run to the next, so that policy set on a datapath
1059 const char *datapath_id;
1062 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1063 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1067 if (hw_addr_iface) {
1069 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1071 * A bridge whose MAC address is taken from a VLAN network device
1072 * (that is, a network device created with vconfig(8) or similar
1073 * tool) will have the same MAC address as a bridge on the VLAN
1074 * device's physical network device.
1076 * Handle this case by hashing the physical network device MAC
1077 * along with the VLAN identifier.
1079 uint8_t buf[ETH_ADDR_LEN + 2];
1080 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1081 buf[ETH_ADDR_LEN] = vlan >> 8;
1082 buf[ETH_ADDR_LEN + 1] = vlan;
1083 return dpid_from_hash(buf, sizeof buf);
1086 * Assume that this bridge's MAC address is unique, since it
1087 * doesn't fit any of the cases we handle specially.
1092 * A purely internal bridge, that is, one that has no non-virtual
1093 * network devices on it at all, is more difficult because it has no
1094 * natural unique identifier at all.
1096 * When the host is a XenServer, we handle this case by hashing the
1097 * host's UUID with the name of the bridge. Names of bridges are
1098 * persistent across XenServer reboots, although they can be reused if
1099 * an internal network is destroyed and then a new one is later
1100 * created, so this is fairly effective.
1102 * When the host is not a XenServer, we punt by using a random MAC
1103 * address on each run.
1105 const char *host_uuid = xenserver_get_host_uuid();
1107 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1108 dpid = dpid_from_hash(combined, strlen(combined));
1114 return eth_addr_to_uint64(bridge_ea);
1118 dpid_from_hash(const void *data, size_t n)
1120 uint8_t hash[SHA1_DIGEST_SIZE];
1122 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1123 sha1_bytes(data, n, hash);
1124 eth_addr_mark_random(hash);
1125 return eth_addr_to_uint64(hash);
1129 iface_refresh_status(struct iface *iface)
1133 enum netdev_flags flags;
1142 if (!netdev_get_status(iface->netdev, &sh)) {
1144 char **keys, **values;
1146 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1147 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1152 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1155 shash_destroy_free_data(&sh);
1157 error = netdev_get_flags(iface->netdev, &flags);
1159 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1162 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1165 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1167 ovsrec_interface_set_duplex(iface->cfg,
1168 netdev_features_is_full_duplex(current)
1170 /* warning: uint64_t -> int64_t conversion */
1171 bps = netdev_features_to_bps(current);
1172 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1175 ovsrec_interface_set_duplex(iface->cfg, NULL);
1176 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1180 ovsrec_interface_set_link_state(iface->cfg,
1181 netdev_get_carrier(iface->netdev)
1184 error = netdev_get_mtu(iface->netdev, &mtu);
1187 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1190 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1195 iface_refresh_cfm_stats(struct iface *iface)
1199 const struct ovsrec_monitor *mon;
1201 mon = iface->cfg->monitor;
1208 for (i = 0; i < mon->n_remote_mps; i++) {
1209 const struct ovsrec_maintenance_point *mp;
1210 const struct remote_mp *rmp;
1212 mp = mon->remote_mps[i];
1213 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1215 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1218 if (hmap_is_empty(&cfm->x_remote_mps)) {
1219 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1222 struct remote_mp *rmp;
1223 int64_t *x_remote_mps;
1225 length = hmap_count(&cfm->x_remote_mps);
1226 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1229 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1230 x_remote_mps[i++] = rmp->mpid;
1233 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1237 if (hmap_is_empty(&cfm->x_remote_maids)) {
1238 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1241 char **x_remote_maids;
1242 struct remote_maid *rmaid;
1244 length = hmap_count(&cfm->x_remote_maids);
1245 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1248 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1251 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1253 for (j = 0; j < CCM_MAID_LEN; j++) {
1254 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1259 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1261 for (i = 0; i < length; i++) {
1262 free(x_remote_maids[i]);
1264 free(x_remote_maids);
1267 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1271 iface_refresh_stats(struct iface *iface)
1277 static const struct iface_stat iface_stats[] = {
1278 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1279 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1280 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1281 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1282 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1283 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1284 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1285 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1286 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1287 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1288 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1289 { "collisions", offsetof(struct netdev_stats, collisions) },
1291 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1292 const struct iface_stat *s;
1294 char *keys[N_STATS];
1295 int64_t values[N_STATS];
1298 struct netdev_stats stats;
1300 /* Intentionally ignore return value, since errors will set 'stats' to
1301 * all-1s, and we will deal with that correctly below. */
1302 netdev_get_stats(iface->netdev, &stats);
1305 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1306 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1307 if (value != UINT64_MAX) {
1314 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1318 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1320 struct ovsdb_datum datum;
1324 get_system_stats(&stats);
1326 ovsdb_datum_from_shash(&datum, &stats);
1327 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1331 static inline const char *
1332 nx_role_to_str(enum nx_role role)
1337 case NX_ROLE_MASTER:
1342 return "*** INVALID ROLE ***";
1347 bridge_refresh_controller_status(const struct bridge *br)
1350 const struct ovsrec_controller *cfg;
1352 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1354 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1355 struct ofproto_controller_info *cinfo =
1356 shash_find_data(&info, cfg->target);
1359 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1360 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1361 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1362 (char **) cinfo->pairs.values,
1365 ovsrec_controller_set_is_connected(cfg, false);
1366 ovsrec_controller_set_role(cfg, NULL);
1367 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1371 ofproto_free_ofproto_controller_info(&info);
1377 const struct ovsrec_open_vswitch *cfg;
1379 bool datapath_destroyed;
1380 bool database_changed;
1383 /* Let each bridge do the work that it needs to do. */
1384 datapath_destroyed = false;
1385 LIST_FOR_EACH (br, node, &all_bridges) {
1386 int error = bridge_run_one(br);
1388 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1389 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1390 "forcing reconfiguration", br->name);
1391 datapath_destroyed = true;
1395 /* (Re)configure if necessary. */
1396 database_changed = ovsdb_idl_run(idl);
1397 cfg = ovsrec_open_vswitch_first(idl);
1399 /* Re-configure SSL. We do this on every trip through the main loop,
1400 * instead of just when the database changes, because the contents of the
1401 * key and certificate files can change without the database changing.
1403 * We do this before bridge_reconfigure() because that function might
1404 * initiate SSL connections and thus requires SSL to be configured. */
1405 if (cfg && cfg->ssl) {
1406 const struct ovsrec_ssl *ssl = cfg->ssl;
1408 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1409 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1412 if (database_changed || datapath_destroyed) {
1414 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1416 bridge_configure_once(cfg);
1417 bridge_reconfigure(cfg);
1419 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1420 ovsdb_idl_txn_commit(txn);
1421 ovsdb_idl_txn_destroy(txn); /* XXX */
1423 /* We still need to reconfigure to avoid dangling pointers to
1424 * now-destroyed ovsrec structures inside bridge data. */
1425 static const struct ovsrec_open_vswitch null_cfg;
1427 bridge_reconfigure(&null_cfg);
1431 /* Refresh system and interface stats if necessary. */
1432 if (time_msec() >= stats_timer) {
1434 struct ovsdb_idl_txn *txn;
1436 txn = ovsdb_idl_txn_create(idl);
1437 LIST_FOR_EACH (br, node, &all_bridges) {
1440 for (i = 0; i < br->n_ports; i++) {
1441 struct port *port = br->ports[i];
1444 for (j = 0; j < port->n_ifaces; j++) {
1445 struct iface *iface = port->ifaces[j];
1446 iface_refresh_stats(iface);
1447 iface_refresh_cfm_stats(iface);
1448 iface_refresh_status(iface);
1451 bridge_refresh_controller_status(br);
1453 refresh_system_stats(cfg);
1454 ovsdb_idl_txn_commit(txn);
1455 ovsdb_idl_txn_destroy(txn); /* XXX */
1458 stats_timer = time_msec() + STATS_INTERVAL;
1466 struct iface *iface;
1468 LIST_FOR_EACH (br, node, &all_bridges) {
1469 ofproto_wait(br->ofproto);
1470 if (ofproto_has_primary_controller(br->ofproto)) {
1474 mac_learning_wait(br->ml);
1477 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1479 cfm_wait(iface->cfm);
1483 ovsdb_idl_wait(idl);
1484 poll_timer_wait_until(stats_timer);
1487 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1488 * configuration changes. */
1490 bridge_flush(struct bridge *br)
1492 COVERAGE_INC(bridge_flush);
1494 mac_learning_flush(br->ml);
1497 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1498 * such interface. */
1499 static struct iface *
1500 bridge_get_local_iface(struct bridge *br)
1504 for (i = 0; i < br->n_ports; i++) {
1505 struct port *port = br->ports[i];
1506 for (j = 0; j < port->n_ifaces; j++) {
1507 struct iface *iface = port->ifaces[j];
1508 if (iface->dp_ifidx == ODPP_LOCAL) {
1517 /* Bridge unixctl user interface functions. */
1519 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1520 const char *args, void *aux OVS_UNUSED)
1522 struct ds ds = DS_EMPTY_INITIALIZER;
1523 const struct bridge *br;
1524 const struct mac_entry *e;
1526 br = bridge_lookup(args);
1528 unixctl_command_reply(conn, 501, "no such bridge");
1532 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1533 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1534 if (e->port < 0 || e->port >= br->n_ports) {
1537 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1538 br->ports[e->port]->ifaces[0]->dp_ifidx,
1539 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1541 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1545 /* Bridge reconfiguration functions. */
1546 static struct bridge *
1547 bridge_create(const struct ovsrec_bridge *br_cfg)
1552 assert(!bridge_lookup(br_cfg->name));
1553 br = xzalloc(sizeof *br);
1555 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1561 dpif_flow_flush(br->dpif);
1563 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1566 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1568 dpif_delete(br->dpif);
1569 dpif_close(br->dpif);
1574 br->name = xstrdup(br_cfg->name);
1576 br->ml = mac_learning_create();
1577 eth_addr_nicira_random(br->default_ea);
1579 hmap_init(&br->ifaces);
1581 shash_init(&br->port_by_name);
1582 shash_init(&br->iface_by_name);
1586 list_push_back(&all_bridges, &br->node);
1588 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1594 bridge_destroy(struct bridge *br)
1599 while (br->n_ports > 0) {
1600 port_destroy(br->ports[br->n_ports - 1]);
1602 list_remove(&br->node);
1603 error = dpif_delete(br->dpif);
1604 if (error && error != ENOENT) {
1605 VLOG_ERR("failed to delete %s: %s",
1606 dpif_name(br->dpif), strerror(error));
1608 dpif_close(br->dpif);
1609 ofproto_destroy(br->ofproto);
1610 mac_learning_destroy(br->ml);
1611 hmap_destroy(&br->ifaces);
1612 shash_destroy(&br->port_by_name);
1613 shash_destroy(&br->iface_by_name);
1620 static struct bridge *
1621 bridge_lookup(const char *name)
1625 LIST_FOR_EACH (br, node, &all_bridges) {
1626 if (!strcmp(br->name, name)) {
1633 /* Handle requests for a listing of all flows known by the OpenFlow
1634 * stack, including those normally hidden. */
1636 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1637 const char *args, void *aux OVS_UNUSED)
1642 br = bridge_lookup(args);
1644 unixctl_command_reply(conn, 501, "Unknown bridge");
1649 ofproto_get_all_flows(br->ofproto, &results);
1651 unixctl_command_reply(conn, 200, ds_cstr(&results));
1652 ds_destroy(&results);
1655 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1656 * connections and reconnect. If BRIDGE is not specified, then all bridges
1657 * drop their controller connections and reconnect. */
1659 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1660 const char *args, void *aux OVS_UNUSED)
1663 if (args[0] != '\0') {
1664 br = bridge_lookup(args);
1666 unixctl_command_reply(conn, 501, "Unknown bridge");
1669 ofproto_reconnect_controllers(br->ofproto);
1671 LIST_FOR_EACH (br, node, &all_bridges) {
1672 ofproto_reconnect_controllers(br->ofproto);
1675 unixctl_command_reply(conn, 200, NULL);
1679 bridge_run_one(struct bridge *br)
1682 struct iface *iface;
1684 error = ofproto_run1(br->ofproto);
1689 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1692 error = ofproto_run2(br->ofproto, br->flush);
1695 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1696 struct ofpbuf *packet;
1702 packet = cfm_run(iface->cfm);
1704 iface_send_packet(iface, packet);
1705 ofpbuf_uninit(packet);
1714 bridge_get_controllers(const struct bridge *br,
1715 struct ovsrec_controller ***controllersp)
1717 struct ovsrec_controller **controllers;
1718 size_t n_controllers;
1720 controllers = br->cfg->controller;
1721 n_controllers = br->cfg->n_controller;
1723 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1729 *controllersp = controllers;
1731 return n_controllers;
1735 bridge_reconfigure_one(struct bridge *br)
1737 struct shash old_ports, new_ports;
1738 struct svec snoops, old_snoops;
1739 struct shash_node *node;
1740 enum ofproto_fail_mode fail_mode;
1743 /* Collect old ports. */
1744 shash_init(&old_ports);
1745 for (i = 0; i < br->n_ports; i++) {
1746 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1749 /* Collect new ports. */
1750 shash_init(&new_ports);
1751 for (i = 0; i < br->cfg->n_ports; i++) {
1752 const char *name = br->cfg->ports[i]->name;
1753 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1754 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1759 /* If we have a controller, then we need a local port. Complain if the
1760 * user didn't specify one.
1762 * XXX perhaps we should synthesize a port ourselves in this case. */
1763 if (bridge_get_controllers(br, NULL)) {
1764 char local_name[IF_NAMESIZE];
1767 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1768 local_name, sizeof local_name);
1769 if (!error && !shash_find(&new_ports, local_name)) {
1770 VLOG_WARN("bridge %s: controller specified but no local port "
1771 "(port named %s) defined",
1772 br->name, local_name);
1776 /* Get rid of deleted ports.
1777 * Get rid of deleted interfaces on ports that still exist. */
1778 SHASH_FOR_EACH (node, &old_ports) {
1779 struct port *port = node->data;
1780 const struct ovsrec_port *port_cfg;
1782 port_cfg = shash_find_data(&new_ports, node->name);
1786 port_del_ifaces(port, port_cfg);
1790 /* Create new ports.
1791 * Add new interfaces to existing ports.
1792 * Reconfigure existing ports. */
1793 SHASH_FOR_EACH (node, &new_ports) {
1794 struct port *port = shash_find_data(&old_ports, node->name);
1796 port = port_create(br, node->name);
1799 port_reconfigure(port, node->data);
1800 if (!port->n_ifaces) {
1801 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1802 br->name, port->name);
1806 shash_destroy(&old_ports);
1807 shash_destroy(&new_ports);
1809 /* Set the fail-mode */
1810 fail_mode = !br->cfg->fail_mode
1811 || !strcmp(br->cfg->fail_mode, "standalone")
1812 ? OFPROTO_FAIL_STANDALONE
1813 : OFPROTO_FAIL_SECURE;
1814 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1815 && !ofproto_has_primary_controller(br->ofproto)) {
1816 ofproto_flush_flows(br->ofproto);
1818 ofproto_set_fail_mode(br->ofproto, fail_mode);
1820 /* Delete all flows if we're switching from connected to standalone or vice
1821 * versa. (XXX Should we delete all flows if we are switching from one
1822 * controller to another?) */
1824 /* Configure OpenFlow controller connection snooping. */
1826 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1827 ovs_rundir(), br->name));
1828 svec_init(&old_snoops);
1829 ofproto_get_snoops(br->ofproto, &old_snoops);
1830 if (!svec_equal(&snoops, &old_snoops)) {
1831 ofproto_set_snoops(br->ofproto, &snoops);
1833 svec_destroy(&snoops);
1834 svec_destroy(&old_snoops);
1836 mirror_reconfigure(br);
1839 /* Initializes 'oc' appropriately as a management service controller for
1842 * The caller must free oc->target when it is no longer needed. */
1844 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1845 struct ofproto_controller *oc)
1847 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1848 oc->max_backoff = 0;
1849 oc->probe_interval = 60;
1850 oc->band = OFPROTO_OUT_OF_BAND;
1851 oc->accept_re = NULL;
1852 oc->update_resolv_conf = false;
1854 oc->burst_limit = 0;
1857 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1859 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1860 struct ofproto_controller *oc)
1862 oc->target = c->target;
1863 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1864 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1865 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1866 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1867 oc->accept_re = c->discover_accept_regex;
1868 oc->update_resolv_conf = c->discover_update_resolv_conf;
1869 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1870 oc->burst_limit = (c->controller_burst_limit
1871 ? *c->controller_burst_limit : 0);
1874 /* Configures the IP stack for 'br''s local interface properly according to the
1875 * configuration in 'c'. */
1877 bridge_configure_local_iface_netdev(struct bridge *br,
1878 struct ovsrec_controller *c)
1880 struct netdev *netdev;
1881 struct in_addr mask, gateway;
1883 struct iface *local_iface;
1886 /* Controller discovery does its own TCP/IP configuration later. */
1887 if (strcmp(c->target, "discover")) {
1891 /* If there's no local interface or no IP address, give up. */
1892 local_iface = bridge_get_local_iface(br);
1893 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1897 /* Bring up the local interface. */
1898 netdev = local_iface->netdev;
1899 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1901 /* Configure the IP address and netmask. */
1902 if (!c->local_netmask
1903 || !inet_aton(c->local_netmask, &mask)
1905 mask.s_addr = guess_netmask(ip.s_addr);
1907 if (!netdev_set_in4(netdev, ip, mask)) {
1908 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1909 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1912 /* Configure the default gateway. */
1913 if (c->local_gateway
1914 && inet_aton(c->local_gateway, &gateway)
1915 && gateway.s_addr) {
1916 if (!netdev_add_router(netdev, gateway)) {
1917 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1918 br->name, IP_ARGS(&gateway.s_addr));
1924 bridge_reconfigure_remotes(struct bridge *br,
1925 const struct sockaddr_in *managers,
1928 const char *disable_ib_str, *queue_id_str;
1929 bool disable_in_band = false;
1932 struct ovsrec_controller **controllers;
1933 size_t n_controllers;
1936 struct ofproto_controller *ocs;
1940 /* Check if we should disable in-band control on this bridge. */
1941 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1942 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1943 disable_in_band = true;
1946 /* Set OpenFlow queue ID for in-band control. */
1947 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1948 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1949 ofproto_set_in_band_queue(br->ofproto, queue_id);
1951 if (disable_in_band) {
1952 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1954 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1956 had_primary = ofproto_has_primary_controller(br->ofproto);
1958 n_controllers = bridge_get_controllers(br, &controllers);
1960 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1963 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1964 for (i = 0; i < n_controllers; i++) {
1965 struct ovsrec_controller *c = controllers[i];
1967 if (!strncmp(c->target, "punix:", 6)
1968 || !strncmp(c->target, "unix:", 5)) {
1969 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1971 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1972 * domain sockets and overwriting arbitrary local files. */
1973 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1974 "\"%s\" due to possibility for remote exploit",
1975 dpif_name(br->dpif), c->target);
1979 bridge_configure_local_iface_netdev(br, c);
1980 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1981 if (disable_in_band) {
1982 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1987 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1988 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1991 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1992 ofproto_flush_flows(br->ofproto);
1995 /* If there are no controllers and the bridge is in standalone
1996 * mode, set up a flow that matches every packet and directs
1997 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1998 * switch is in secure mode and we won't pass any traffic until
1999 * a controller has been defined and it tells us to do so. */
2001 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2002 union ofp_action action;
2003 struct cls_rule rule;
2005 memset(&action, 0, sizeof action);
2006 action.type = htons(OFPAT_OUTPUT);
2007 action.output.len = htons(sizeof action);
2008 action.output.port = htons(OFPP_NORMAL);
2009 cls_rule_init_catchall(&rule, 0);
2010 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2015 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2020 for (i = 0; i < br->n_ports; i++) {
2021 struct port *port = br->ports[i];
2022 for (j = 0; j < port->n_ifaces; j++) {
2023 struct iface *iface = port->ifaces[j];
2024 shash_add_once(ifaces, iface->name, iface);
2026 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2027 shash_add_once(ifaces, port->name, NULL);
2032 /* For robustness, in case the administrator moves around datapath ports behind
2033 * our back, we re-check all the datapath port numbers here.
2035 * This function will set the 'dp_ifidx' members of interfaces that have
2036 * disappeared to -1, so only call this function from a context where those
2037 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2038 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2039 * datapath, which doesn't support UINT16_MAX+1 ports. */
2041 bridge_fetch_dp_ifaces(struct bridge *br)
2043 struct dpif_port_dump dump;
2044 struct odp_port odp_port;
2047 /* Reset all interface numbers. */
2048 for (i = 0; i < br->n_ports; i++) {
2049 struct port *port = br->ports[i];
2050 for (j = 0; j < port->n_ifaces; j++) {
2051 struct iface *iface = port->ifaces[j];
2052 iface->dp_ifidx = -1;
2055 hmap_clear(&br->ifaces);
2057 DPIF_PORT_FOR_EACH (&odp_port, &dump, br->dpif) {
2058 struct iface *iface = iface_lookup(br, odp_port.devname);
2060 if (iface->dp_ifidx >= 0) {
2061 VLOG_WARN("%s reported interface %s twice",
2062 dpif_name(br->dpif), odp_port.devname);
2063 } else if (iface_from_dp_ifidx(br, odp_port.port)) {
2064 VLOG_WARN("%s reported interface %"PRIu16" twice",
2065 dpif_name(br->dpif), odp_port.port);
2067 iface->dp_ifidx = odp_port.port;
2068 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2069 hash_int(iface->dp_ifidx, 0));
2072 iface_set_ofport(iface->cfg,
2073 (iface->dp_ifidx >= 0
2074 ? odp_port_to_ofp_port(iface->dp_ifidx)
2080 /* Bridge packet processing functions. */
2083 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2085 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2088 static struct bond_entry *
2089 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
2092 assert(port->bond_mode == BM_SLB);
2093 return &port->bond_hash[bond_hash(mac, vlan)];
2097 bond_choose_iface(const struct port *port)
2099 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2100 size_t i, best_down_slave = -1;
2101 long long next_delay_expiration = LLONG_MAX;
2103 for (i = 0; i < port->n_ifaces; i++) {
2104 struct iface *iface = port->ifaces[i];
2106 if (iface->enabled) {
2108 } else if (iface->delay_expires < next_delay_expiration) {
2109 best_down_slave = i;
2110 next_delay_expiration = iface->delay_expires;
2114 if (best_down_slave != -1) {
2115 struct iface *iface = port->ifaces[best_down_slave];
2117 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2118 "since no other interface is up", iface->name,
2119 iface->delay_expires - time_msec());
2120 bond_enable_slave(iface, true);
2123 return best_down_slave;
2127 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2128 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2130 struct iface *iface;
2132 assert(port->n_ifaces);
2133 if (port->n_ifaces == 1) {
2134 iface = port->ifaces[0];
2135 } else if (port->bond_mode == BM_AB) {
2136 if (port->active_iface < 0) {
2137 *tags |= port->no_ifaces_tag;
2140 iface = port->ifaces[port->active_iface];
2141 } else if (port->bond_mode == BM_SLB){
2142 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2143 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2144 || !port->ifaces[e->iface_idx]->enabled) {
2145 /* XXX select interface properly. The current interface selection
2146 * is only good for testing the rebalancing code. */
2147 e->iface_idx = bond_choose_iface(port);
2148 if (e->iface_idx < 0) {
2149 *tags |= port->no_ifaces_tag;
2152 e->iface_tag = tag_create_random();
2153 ((struct port *) port)->bond_compat_is_stale = true;
2155 *tags |= e->iface_tag;
2156 iface = port->ifaces[e->iface_idx];
2160 *dp_ifidx = iface->dp_ifidx;
2161 *tags |= iface->tag; /* Currently only used for bonding. */
2166 bond_link_status_update(struct iface *iface, bool carrier)
2168 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2169 struct port *port = iface->port;
2171 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2172 /* Nothing to do. */
2175 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2176 iface->name, carrier ? "up" : "down");
2177 if (carrier == iface->enabled) {
2178 iface->delay_expires = LLONG_MAX;
2179 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2180 iface->name, carrier ? "disabled" : "enabled");
2181 } else if (carrier && port->active_iface < 0) {
2182 bond_enable_slave(iface, true);
2183 if (port->updelay) {
2184 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2185 "other interface is up", iface->name, port->updelay);
2188 int delay = carrier ? port->updelay : port->downdelay;
2189 iface->delay_expires = time_msec() + delay;
2192 "interface %s: will be %s if it stays %s for %d ms",
2194 carrier ? "enabled" : "disabled",
2195 carrier ? "up" : "down",
2202 bond_choose_active_iface(struct port *port)
2204 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2206 port->active_iface = bond_choose_iface(port);
2207 port->active_iface_tag = tag_create_random();
2208 if (port->active_iface >= 0) {
2209 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2210 port->name, port->ifaces[port->active_iface]->name);
2212 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2218 bond_enable_slave(struct iface *iface, bool enable)
2220 struct port *port = iface->port;
2221 struct bridge *br = port->bridge;
2223 /* This acts as a recursion check. If the act of disabling a slave
2224 * causes a different slave to be enabled, the flag will allow us to
2225 * skip redundant work when we reenter this function. It must be
2226 * cleared on exit to keep things safe with multiple bonds. */
2227 static bool moving_active_iface = false;
2229 iface->delay_expires = LLONG_MAX;
2230 if (enable == iface->enabled) {
2234 iface->enabled = enable;
2235 if (!iface->enabled) {
2236 VLOG_WARN("interface %s: disabled", iface->name);
2237 ofproto_revalidate(br->ofproto, iface->tag);
2238 if (iface->port_ifidx == port->active_iface) {
2239 ofproto_revalidate(br->ofproto,
2240 port->active_iface_tag);
2242 /* Disabling a slave can lead to another slave being immediately
2243 * enabled if there will be no active slaves but one is waiting
2244 * on an updelay. In this case we do not need to run most of the
2245 * code for the newly enabled slave since there was no period
2246 * without an active slave and it is redundant with the disabling
2248 moving_active_iface = true;
2249 bond_choose_active_iface(port);
2251 bond_send_learning_packets(port);
2253 VLOG_WARN("interface %s: enabled", iface->name);
2254 if (port->active_iface < 0 && !moving_active_iface) {
2255 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2256 bond_choose_active_iface(port);
2257 bond_send_learning_packets(port);
2259 iface->tag = tag_create_random();
2262 moving_active_iface = false;
2263 port->bond_compat_is_stale = true;
2266 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2267 * bond interface. */
2269 bond_update_fake_iface_stats(struct port *port)
2271 struct netdev_stats bond_stats;
2272 struct netdev *bond_dev;
2275 memset(&bond_stats, 0, sizeof bond_stats);
2277 for (i = 0; i < port->n_ifaces; i++) {
2278 struct netdev_stats slave_stats;
2280 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2281 /* XXX: We swap the stats here because they are swapped back when
2282 * reported by the internal device. The reason for this is
2283 * internal devices normally represent packets going into the system
2284 * but when used as fake bond device they represent packets leaving
2285 * the system. We really should do this in the internal device
2286 * itself because changing it here reverses the counts from the
2287 * perspective of the switch. However, the internal device doesn't
2288 * know what type of device it represents so we have to do it here
2290 bond_stats.tx_packets += slave_stats.rx_packets;
2291 bond_stats.tx_bytes += slave_stats.rx_bytes;
2292 bond_stats.rx_packets += slave_stats.tx_packets;
2293 bond_stats.rx_bytes += slave_stats.tx_bytes;
2297 if (!netdev_open_default(port->name, &bond_dev)) {
2298 netdev_set_stats(bond_dev, &bond_stats);
2299 netdev_close(bond_dev);
2304 bond_run(struct bridge *br)
2308 for (i = 0; i < br->n_ports; i++) {
2309 struct port *port = br->ports[i];
2311 if (port->n_ifaces >= 2) {
2314 if (port->monitor) {
2315 assert(!port->miimon);
2317 /* Track carrier going up and down on interfaces. */
2318 while (!netdev_monitor_poll(port->monitor, &devname)) {
2319 struct iface *iface;
2321 iface = port_lookup_iface(port, devname);
2323 bool up = netdev_get_carrier(iface->netdev);
2325 bond_link_status_update(iface, up);
2326 port_update_bond_compat(port);
2331 assert(port->miimon);
2333 if (time_msec() >= port->bond_miimon_next_update) {
2334 for (j = 0; j < port->n_ifaces; j++) {
2335 struct iface *iface = port->ifaces[j];
2336 bool up = netdev_get_miimon(iface->netdev);
2338 bond_link_status_update(iface, up);
2339 port_update_bond_compat(port);
2341 port->bond_miimon_next_update = time_msec() +
2342 port->bond_miimon_interval;
2346 for (j = 0; j < port->n_ifaces; j++) {
2347 struct iface *iface = port->ifaces[j];
2348 if (time_msec() >= iface->delay_expires) {
2349 bond_enable_slave(iface, !iface->enabled);
2353 if (port->bond_fake_iface
2354 && time_msec() >= port->bond_next_fake_iface_update) {
2355 bond_update_fake_iface_stats(port);
2356 port->bond_next_fake_iface_update = time_msec() + 1000;
2360 if (port->bond_compat_is_stale) {
2361 port->bond_compat_is_stale = false;
2362 port_update_bond_compat(port);
2368 bond_wait(struct bridge *br)
2372 for (i = 0; i < br->n_ports; i++) {
2373 struct port *port = br->ports[i];
2374 if (port->n_ifaces < 2) {
2378 if (port->monitor) {
2379 netdev_monitor_poll_wait(port->monitor);
2383 poll_timer_wait_until(port->bond_miimon_next_update);
2386 for (j = 0; j < port->n_ifaces; j++) {
2387 struct iface *iface = port->ifaces[j];
2388 if (iface->delay_expires != LLONG_MAX) {
2389 poll_timer_wait_until(iface->delay_expires);
2392 if (port->bond_fake_iface) {
2393 poll_timer_wait_until(port->bond_next_fake_iface_update);
2399 set_dst(struct dst *dst, const struct flow *flow,
2400 const struct port *in_port, const struct port *out_port,
2403 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2404 : in_port->vlan >= 0 ? in_port->vlan
2405 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2406 : vlan_tci_to_vid(flow->vlan_tci));
2407 return choose_output_iface(out_port, flow->dl_src, dst->vlan,
2408 &dst->dp_ifidx, tags);
2412 swap_dst(struct dst *p, struct dst *q)
2414 struct dst tmp = *p;
2419 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2420 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2421 * that we push to the datapath. We could in fact fully sort the array by
2422 * vlan, but in most cases there are at most two different vlan tags so that's
2423 * possibly overkill.) */
2425 partition_dsts(struct dst_set *set, int vlan)
2427 struct dst *first = set->dsts;
2428 struct dst *last = set->dsts + set->n;
2430 while (first != last) {
2432 * - All dsts < first have vlan == 'vlan'.
2433 * - All dsts >= last have vlan != 'vlan'.
2434 * - first < last. */
2435 while (first->vlan == vlan) {
2436 if (++first == last) {
2441 /* Same invariants, plus one additional:
2442 * - first->vlan != vlan.
2444 while (last[-1].vlan != vlan) {
2445 if (--last == first) {
2450 /* Same invariants, plus one additional:
2451 * - last[-1].vlan == vlan.*/
2452 swap_dst(first++, --last);
2457 mirror_mask_ffs(mirror_mask_t mask)
2459 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2464 dst_set_init(struct dst_set *set)
2466 set->dsts = set->builtin;
2468 set->allocated = ARRAY_SIZE(set->builtin);
2472 dst_set_add(struct dst_set *set, const struct dst *dst)
2474 if (set->n >= set->allocated) {
2475 size_t new_allocated;
2476 struct dst *new_dsts;
2478 new_allocated = set->allocated * 2;
2479 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2480 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2484 set->dsts = new_dsts;
2485 set->allocated = new_allocated;
2487 set->dsts[set->n++] = *dst;
2491 dst_set_free(struct dst_set *set)
2493 if (set->dsts != set->builtin) {
2499 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2502 for (i = 0; i < set->n; i++) {
2503 if (set->dsts[i].vlan == test->vlan
2504 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2512 port_trunks_vlan(const struct port *port, uint16_t vlan)
2514 return (port->vlan < 0
2515 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2519 port_includes_vlan(const struct port *port, uint16_t vlan)
2521 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2525 port_is_floodable(const struct port *port)
2529 for (i = 0; i < port->n_ifaces; i++) {
2530 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2531 port->ifaces[i]->dp_ifidx)) {
2539 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2540 const struct port *in_port, const struct port *out_port,
2541 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2543 mirror_mask_t mirrors = in_port->src_mirrors;
2548 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2549 if (flow_vlan == 0) {
2550 flow_vlan = OFP_VLAN_NONE;
2553 if (out_port == FLOOD_PORT) {
2554 for (i = 0; i < br->n_ports; i++) {
2555 struct port *port = br->ports[i];
2557 && port_is_floodable(port)
2558 && port_includes_vlan(port, vlan)
2559 && !port->is_mirror_output_port
2560 && set_dst(&dst, flow, in_port, port, tags)) {
2561 mirrors |= port->dst_mirrors;
2562 dst_set_add(set, &dst);
2565 *nf_output_iface = NF_OUT_FLOOD;
2566 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2567 dst_set_add(set, &dst);
2568 *nf_output_iface = dst.dp_ifidx;
2569 mirrors |= out_port->dst_mirrors;
2573 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2574 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2576 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2577 && !dst_is_duplicate(set, &dst)) {
2578 dst_set_add(set, &dst);
2581 for (i = 0; i < br->n_ports; i++) {
2582 struct port *port = br->ports[i];
2583 if (port_includes_vlan(port, m->out_vlan)
2584 && set_dst(&dst, flow, in_port, port, tags))
2586 if (port->vlan < 0) {
2587 dst.vlan = m->out_vlan;
2589 if (dst_is_duplicate(set, &dst)) {
2593 /* Use the vlan tag on the original flow instead of
2594 * the one passed in the vlan parameter. This ensures
2595 * that we compare the vlan from before any implicit
2596 * tagging tags place. This is necessary because
2597 * dst->vlan is the final vlan, after removing implicit
2599 if (port == in_port && dst.vlan == flow_vlan) {
2600 /* Don't send out input port on same VLAN. */
2603 dst_set_add(set, &dst);
2608 mirrors &= mirrors - 1;
2611 partition_dsts(set, flow_vlan);
2614 static void OVS_UNUSED
2615 print_dsts(const struct dst_set *set)
2619 for (i = 0; i < set->n; i++) {
2620 const struct dst *dst = &set->dsts[i];
2622 printf(">p%"PRIu16, dst->dp_ifidx);
2623 if (dst->vlan != OFP_VLAN_NONE) {
2624 printf("v%"PRIu16, dst->vlan);
2630 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2631 const struct port *in_port, const struct port *out_port,
2632 tag_type *tags, struct ofpbuf *actions,
2633 uint16_t *nf_output_iface)
2640 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2643 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2644 if (cur_vlan == 0) {
2645 cur_vlan = OFP_VLAN_NONE;
2647 for (i = 0; i < set.n; i++) {
2648 const struct dst *dst = &set.dsts[i];
2649 if (dst->vlan != cur_vlan) {
2650 if (dst->vlan == OFP_VLAN_NONE) {
2651 nl_msg_put_flag(actions, ODPAT_STRIP_VLAN);
2654 tci = htons(dst->vlan & VLAN_VID_MASK);
2655 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2656 nl_msg_put_be16(actions, ODPAT_SET_DL_TCI, tci);
2658 cur_vlan = dst->vlan;
2660 nl_msg_put_u32(actions, ODPAT_OUTPUT, dst->dp_ifidx);
2665 /* Returns the effective vlan of a packet, taking into account both the
2666 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2667 * the packet is untagged and -1 indicates it has an invalid header and
2668 * should be dropped. */
2669 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2670 struct port *in_port, bool have_packet)
2672 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2673 if (in_port->vlan >= 0) {
2675 /* XXX support double tagging? */
2677 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2678 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2679 "packet received on port %s configured with "
2680 "implicit VLAN %"PRIu16,
2681 br->name, vlan, in_port->name, in_port->vlan);
2685 vlan = in_port->vlan;
2687 if (!port_includes_vlan(in_port, vlan)) {
2689 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2690 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2691 "packet received on port %s not configured for "
2693 br->name, vlan, in_port->name, vlan);
2702 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2703 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2704 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2706 is_gratuitous_arp(const struct flow *flow)
2708 return (flow->dl_type == htons(ETH_TYPE_ARP)
2709 && eth_addr_is_broadcast(flow->dl_dst)
2710 && (flow->nw_proto == ARP_OP_REPLY
2711 || (flow->nw_proto == ARP_OP_REQUEST
2712 && flow->nw_src == flow->nw_dst)));
2716 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2717 struct port *in_port)
2719 enum grat_arp_lock_type lock_type;
2722 /* We don't want to learn from gratuitous ARP packets that are reflected
2723 * back over bond slaves so we lock the learning table. */
2724 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2725 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2726 GRAT_ARP_LOCK_CHECK;
2728 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2731 /* The log messages here could actually be useful in debugging,
2732 * so keep the rate limit relatively high. */
2733 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2735 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2736 "on port %s in VLAN %d",
2737 br->name, ETH_ADDR_ARGS(flow->dl_src),
2738 in_port->name, vlan);
2739 ofproto_revalidate(br->ofproto, rev_tag);
2743 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2744 * dropped. Returns true if they may be forwarded, false if they should be
2747 * If 'have_packet' is true, it indicates that the caller is processing a
2748 * received packet. If 'have_packet' is false, then the caller is just
2749 * revalidating an existing flow because configuration has changed. Either
2750 * way, 'have_packet' only affects logging (there is no point in logging errors
2751 * during revalidation).
2753 * Sets '*in_portp' to the input port. This will be a null pointer if
2754 * flow->in_port does not designate a known input port (in which case
2755 * is_admissible() returns false).
2757 * When returning true, sets '*vlanp' to the effective VLAN of the input
2758 * packet, as returned by flow_get_vlan().
2760 * May also add tags to '*tags', although the current implementation only does
2761 * so in one special case.
2764 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2765 tag_type *tags, int *vlanp, struct port **in_portp)
2767 struct iface *in_iface;
2768 struct port *in_port;
2771 /* Find the interface and port structure for the received packet. */
2772 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2774 /* No interface? Something fishy... */
2776 /* Odd. A few possible reasons here:
2778 * - We deleted an interface but there are still a few packets
2779 * queued up from it.
2781 * - Someone externally added an interface (e.g. with "ovs-dpctl
2782 * add-if") that we don't know about.
2784 * - Packet arrived on the local port but the local port is not
2785 * one of our bridge ports.
2787 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2789 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2790 "interface %"PRIu16, br->name, flow->in_port);
2796 *in_portp = in_port = in_iface->port;
2797 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2802 /* Drop frames for reserved multicast addresses. */
2803 if (eth_addr_is_reserved(flow->dl_dst)) {
2807 /* Drop frames on ports reserved for mirroring. */
2808 if (in_port->is_mirror_output_port) {
2810 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2811 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2812 "%s, which is reserved exclusively for mirroring",
2813 br->name, in_port->name);
2818 /* Packets received on bonds need special attention to avoid duplicates. */
2819 if (in_port->n_ifaces > 1) {
2821 bool is_grat_arp_locked;
2823 if (eth_addr_is_multicast(flow->dl_dst)) {
2824 *tags |= in_port->active_iface_tag;
2825 if (in_port->active_iface != in_iface->port_ifidx) {
2826 /* Drop all multicast packets on inactive slaves. */
2831 /* Drop all packets for which we have learned a different input
2832 * port, because we probably sent the packet on one slave and got
2833 * it back on the other. Gratuitous ARP packets are an exception
2834 * to this rule: the host has moved to another switch. The exception
2835 * to the exception is if we locked the learning table to avoid
2836 * reflections on bond slaves. If this is the case, just drop the
2838 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2839 &is_grat_arp_locked);
2840 if (src_idx != -1 && src_idx != in_port->port_idx &&
2841 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2849 /* If the composed actions may be applied to any packet in the given 'flow',
2850 * returns true. Otherwise, the actions should only be applied to 'packet', or
2851 * not at all, if 'packet' was NULL. */
2853 process_flow(struct bridge *br, const struct flow *flow,
2854 const struct ofpbuf *packet, struct ofpbuf *actions,
2855 tag_type *tags, uint16_t *nf_output_iface)
2857 struct port *in_port;
2858 struct port *out_port;
2862 /* Check whether we should drop packets in this flow. */
2863 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2868 /* Learn source MAC (but don't try to learn from revalidation). */
2870 update_learning_table(br, flow, vlan, in_port);
2873 /* Determine output port. */
2874 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2876 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2877 out_port = br->ports[out_port_idx];
2878 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2879 /* If we are revalidating but don't have a learning entry then
2880 * eject the flow. Installing a flow that floods packets opens
2881 * up a window of time where we could learn from a packet reflected
2882 * on a bond and blackhole packets before the learning table is
2883 * updated to reflect the correct port. */
2886 out_port = FLOOD_PORT;
2889 /* Don't send packets out their input ports. */
2890 if (in_port == out_port) {
2896 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2904 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2905 struct ofpbuf *actions, tag_type *tags,
2906 uint16_t *nf_output_iface, void *br_)
2908 struct iface *iface;
2909 struct bridge *br = br_;
2911 COVERAGE_INC(bridge_process_flow);
2913 iface = iface_from_dp_ifidx(br, flow->in_port);
2915 if (cfm_should_process_flow(flow)) {
2916 if (packet && iface->cfm) {
2917 cfm_process_heartbeat(iface->cfm, packet);
2922 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2926 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2927 const struct nlattr *actions,
2929 unsigned long long int n_bytes, void *br_)
2931 struct bridge *br = br_;
2932 const struct nlattr *a;
2933 struct port *in_port;
2938 /* Feed information from the active flows back into the learning table to
2939 * ensure that table is always in sync with what is actually flowing
2940 * through the datapath.
2942 * We test that 'tags' is nonzero to ensure that only flows that include an
2943 * OFPP_NORMAL action are used for learning. This works because
2944 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2945 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2946 update_learning_table(br, flow, vlan, in_port);
2949 /* Account for bond slave utilization. */
2950 if (!br->has_bonded_ports) {
2953 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2954 if (nl_attr_type(a) == ODPAT_OUTPUT) {
2955 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2956 if (out_port && out_port->n_ifaces >= 2 &&
2957 out_port->bond_mode == BM_SLB) {
2958 uint16_t vlan = (flow->vlan_tci
2959 ? vlan_tci_to_vid(flow->vlan_tci)
2961 struct bond_entry *e = lookup_bond_entry(out_port,
2962 flow->dl_src, vlan);
2963 e->tx_bytes += n_bytes;
2970 bridge_account_checkpoint_ofhook_cb(void *br_)
2972 struct bridge *br = br_;
2976 if (!br->has_bonded_ports) {
2981 for (i = 0; i < br->n_ports; i++) {
2982 struct port *port = br->ports[i];
2983 if (port->n_ifaces > 1 && port->bond_mode == BM_SLB
2984 && now >= port->bond_next_rebalance) {
2985 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2986 bond_rebalance_port(port);
2991 static struct ofhooks bridge_ofhooks = {
2992 bridge_normal_ofhook_cb,
2993 bridge_account_flow_ofhook_cb,
2994 bridge_account_checkpoint_ofhook_cb,
2997 /* Bonding functions. */
2999 /* Statistics for a single interface on a bonded port, used for load-based
3000 * bond rebalancing. */
3001 struct slave_balance {
3002 struct iface *iface; /* The interface. */
3003 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3005 /* All the "bond_entry"s that are assigned to this interface, in order of
3006 * increasing tx_bytes. */
3007 struct bond_entry **hashes;
3012 bond_mode_to_string(enum bond_mode bm) {
3013 static char *bm_slb = "balance-slb";
3014 static char *bm_ab = "active-backup";
3017 case BM_SLB: return bm_slb;
3018 case BM_AB: return bm_ab;
3025 /* Sorts pointers to pointers to bond_entries in ascending order by the
3026 * interface to which they are assigned, and within a single interface in
3027 * ascending order of bytes transmitted. */
3029 compare_bond_entries(const void *a_, const void *b_)
3031 const struct bond_entry *const *ap = a_;
3032 const struct bond_entry *const *bp = b_;
3033 const struct bond_entry *a = *ap;
3034 const struct bond_entry *b = *bp;
3035 if (a->iface_idx != b->iface_idx) {
3036 return a->iface_idx > b->iface_idx ? 1 : -1;
3037 } else if (a->tx_bytes != b->tx_bytes) {
3038 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3044 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3045 * *descending* order by number of bytes transmitted. */
3047 compare_slave_balance(const void *a_, const void *b_)
3049 const struct slave_balance *a = a_;
3050 const struct slave_balance *b = b_;
3051 if (a->iface->enabled != b->iface->enabled) {
3052 return a->iface->enabled ? -1 : 1;
3053 } else if (a->tx_bytes != b->tx_bytes) {
3054 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3061 swap_bals(struct slave_balance *a, struct slave_balance *b)
3063 struct slave_balance tmp = *a;
3068 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3069 * given that 'p' (and only 'p') might be in the wrong location.
3071 * This function invalidates 'p', since it might now be in a different memory
3074 resort_bals(struct slave_balance *p,
3075 struct slave_balance bals[], size_t n_bals)
3078 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3079 swap_bals(p, p - 1);
3081 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3082 swap_bals(p, p + 1);
3088 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3090 if (VLOG_IS_DBG_ENABLED()) {
3091 struct ds ds = DS_EMPTY_INITIALIZER;
3092 const struct slave_balance *b;
3094 for (b = bals; b < bals + n_bals; b++) {
3098 ds_put_char(&ds, ',');
3100 ds_put_format(&ds, " %s %"PRIu64"kB",
3101 b->iface->name, b->tx_bytes / 1024);
3103 if (!b->iface->enabled) {
3104 ds_put_cstr(&ds, " (disabled)");
3106 if (b->n_hashes > 0) {
3107 ds_put_cstr(&ds, " (");
3108 for (i = 0; i < b->n_hashes; i++) {
3109 const struct bond_entry *e = b->hashes[i];
3111 ds_put_cstr(&ds, " + ");
3113 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3114 e - port->bond_hash, e->tx_bytes / 1024);
3116 ds_put_cstr(&ds, ")");
3119 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3124 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3126 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3129 struct bond_entry *hash = from->hashes[hash_idx];
3130 struct port *port = from->iface->port;
3131 uint64_t delta = hash->tx_bytes;
3133 assert(port->bond_mode == BM_SLB);
3135 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3136 "from %s to %s (now carrying %"PRIu64"kB and "
3137 "%"PRIu64"kB load, respectively)",
3138 port->name, delta / 1024, hash - port->bond_hash,
3139 from->iface->name, to->iface->name,
3140 (from->tx_bytes - delta) / 1024,
3141 (to->tx_bytes + delta) / 1024);
3143 /* Delete element from from->hashes.
3145 * We don't bother to add the element to to->hashes because not only would
3146 * it require more work, the only purpose it would be to allow that hash to
3147 * be migrated to another slave in this rebalancing run, and there is no
3148 * point in doing that. */
3149 if (hash_idx == 0) {
3152 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3153 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3157 /* Shift load away from 'from' to 'to'. */
3158 from->tx_bytes -= delta;
3159 to->tx_bytes += delta;
3161 /* Arrange for flows to be revalidated. */
3162 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3163 hash->iface_idx = to->iface->port_ifidx;
3164 hash->iface_tag = tag_create_random();
3168 bond_rebalance_port(struct port *port)
3170 struct slave_balance *bals;
3172 struct bond_entry *hashes[BOND_MASK + 1];
3173 struct slave_balance *b, *from, *to;
3174 struct bond_entry *e;
3177 assert(port->bond_mode == BM_SLB);
3179 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3180 * descending order of tx_bytes, so that bals[0] represents the most
3181 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3184 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3185 * array for each slave_balance structure, we sort our local array of
3186 * hashes in order by slave, so that all of the hashes for a given slave
3187 * become contiguous in memory, and then we point each 'hashes' members of
3188 * a slave_balance structure to the start of a contiguous group. */
3189 n_bals = port->n_ifaces;
3190 bals = xmalloc(n_bals * sizeof *bals);
3191 for (b = bals; b < &bals[n_bals]; b++) {
3192 b->iface = port->ifaces[b - bals];
3197 for (i = 0; i <= BOND_MASK; i++) {
3198 hashes[i] = &port->bond_hash[i];
3200 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3201 for (i = 0; i <= BOND_MASK; i++) {
3203 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3204 b = &bals[e->iface_idx];
3205 b->tx_bytes += e->tx_bytes;
3207 b->hashes = &hashes[i];
3212 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3213 log_bals(bals, n_bals, port);
3215 /* Discard slaves that aren't enabled (which were sorted to the back of the
3216 * array earlier). */
3217 while (!bals[n_bals - 1].iface->enabled) {
3224 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3225 to = &bals[n_bals - 1];
3226 for (from = bals; from < to; ) {
3227 uint64_t overload = from->tx_bytes - to->tx_bytes;
3228 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3229 /* The extra load on 'from' (and all less-loaded slaves), compared
3230 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3231 * it is less than ~1Mbps. No point in rebalancing. */
3233 } else if (from->n_hashes == 1) {
3234 /* 'from' only carries a single MAC hash, so we can't shift any
3235 * load away from it, even though we want to. */
3238 /* 'from' is carrying significantly more load than 'to', and that
3239 * load is split across at least two different hashes. Pick a hash
3240 * to migrate to 'to' (the least-loaded slave), given that doing so
3241 * must decrease the ratio of the load on the two slaves by at
3244 * The sort order we use means that we prefer to shift away the
3245 * smallest hashes instead of the biggest ones. There is little
3246 * reason behind this decision; we could use the opposite sort
3247 * order to shift away big hashes ahead of small ones. */
3250 for (i = 0; i < from->n_hashes; i++) {
3251 double old_ratio, new_ratio;
3252 uint64_t delta = from->hashes[i]->tx_bytes;
3254 if (delta == 0 || from->tx_bytes - delta == 0) {
3255 /* Pointless move. */
3259 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3261 if (to->tx_bytes == 0) {
3262 /* Nothing on the new slave, move it. */
3266 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3267 new_ratio = (double)(from->tx_bytes - delta) /
3268 (to->tx_bytes + delta);
3270 if (new_ratio == 0) {
3271 /* Should already be covered but check to prevent division
3276 if (new_ratio < 1) {
3277 new_ratio = 1 / new_ratio;
3280 if (old_ratio - new_ratio > 0.1) {
3281 /* Would decrease the ratio, move it. */
3285 if (i < from->n_hashes) {
3286 bond_shift_load(from, to, i);
3287 port->bond_compat_is_stale = true;
3289 /* If the result of the migration changed the relative order of
3290 * 'from' and 'to' swap them back to maintain invariants. */
3291 if (order_swapped) {
3292 swap_bals(from, to);
3295 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3296 * point to different slave_balance structures. It is only
3297 * valid to do these two operations in a row at all because we
3298 * know that 'from' will not move past 'to' and vice versa. */
3299 resort_bals(from, bals, n_bals);
3300 resort_bals(to, bals, n_bals);
3307 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3308 * historical data to decay to <1% in 7 rebalancing runs. */
3309 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3318 bond_send_learning_packets(struct port *port)
3320 struct bridge *br = port->bridge;
3321 struct mac_entry *e;
3322 struct ofpbuf packet;
3323 int error, n_packets, n_errors;
3325 if (!port->n_ifaces || port->active_iface < 0) {
3329 ofpbuf_init(&packet, 128);
3330 error = n_packets = n_errors = 0;
3331 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3332 union ofp_action actions[2], *a;
3338 if (e->port == port->port_idx
3339 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3343 /* Compose actions. */
3344 memset(actions, 0, sizeof actions);
3347 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3348 a->vlan_vid.len = htons(sizeof *a);
3349 a->vlan_vid.vlan_vid = htons(e->vlan);
3352 a->output.type = htons(OFPAT_OUTPUT);
3353 a->output.len = htons(sizeof *a);
3354 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3359 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3361 flow_extract(&packet, 0, ODPP_NONE, &flow);
3362 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3369 ofpbuf_uninit(&packet);
3372 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3373 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3374 "packets, last error was: %s",
3375 port->name, n_errors, n_packets, strerror(error));
3377 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3378 port->name, n_packets);
3382 /* Bonding unixctl user interface functions. */
3385 bond_unixctl_list(struct unixctl_conn *conn,
3386 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3388 struct ds ds = DS_EMPTY_INITIALIZER;
3389 const struct bridge *br;
3391 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3393 LIST_FOR_EACH (br, node, &all_bridges) {
3396 for (i = 0; i < br->n_ports; i++) {
3397 const struct port *port = br->ports[i];
3398 if (port->n_ifaces > 1) {
3401 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3402 bond_mode_to_string(port->bond_mode));
3403 for (j = 0; j < port->n_ifaces; j++) {
3404 const struct iface *iface = port->ifaces[j];
3406 ds_put_cstr(&ds, ", ");
3408 ds_put_cstr(&ds, iface->name);
3410 ds_put_char(&ds, '\n');
3414 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3418 static struct port *
3419 bond_find(const char *name)
3421 const struct bridge *br;
3423 LIST_FOR_EACH (br, node, &all_bridges) {
3426 for (i = 0; i < br->n_ports; i++) {
3427 struct port *port = br->ports[i];
3428 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3437 bond_unixctl_show(struct unixctl_conn *conn,
3438 const char *args, void *aux OVS_UNUSED)
3440 struct ds ds = DS_EMPTY_INITIALIZER;
3441 const struct port *port;
3444 port = bond_find(args);
3446 unixctl_command_reply(conn, 501, "no such bond");
3450 ds_put_format(&ds, "bond_mode: %s\n",
3451 bond_mode_to_string(port->bond_mode));
3452 ds_put_format(&ds, "bond-detect-mode: %s\n",
3453 port->miimon ? "miimon" : "carrier");
3456 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3457 port->bond_miimon_interval);
3460 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3461 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3463 if (port->bond_mode == BM_SLB) {
3464 ds_put_format(&ds, "next rebalance: %lld ms\n",
3465 port->bond_next_rebalance - time_msec());
3468 for (j = 0; j < port->n_ifaces; j++) {
3469 const struct iface *iface = port->ifaces[j];
3470 struct bond_entry *be;
3473 ds_put_format(&ds, "slave %s: %s\n",
3474 iface->name, iface->enabled ? "enabled" : "disabled");
3475 if (j == port->active_iface) {
3476 ds_put_cstr(&ds, "\tactive slave\n");
3478 if (iface->delay_expires != LLONG_MAX) {
3479 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3480 iface->enabled ? "downdelay" : "updelay",
3481 iface->delay_expires - time_msec());
3484 if (port->bond_mode != BM_SLB) {
3489 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3490 int hash = be - port->bond_hash;
3491 struct mac_entry *me;
3493 if (be->iface_idx != j) {
3497 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3498 hash, be->tx_bytes / 1024);
3501 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3504 if (bond_hash(me->mac, me->vlan) == hash
3505 && me->port != port->port_idx
3506 && choose_output_iface(port, me->mac, me->vlan,
3508 && dp_ifidx == iface->dp_ifidx)
3510 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3511 ETH_ADDR_ARGS(me->mac));
3516 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3521 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3522 void *aux OVS_UNUSED)
3524 char *args = (char *) args_;
3525 char *save_ptr = NULL;
3526 char *bond_s, *hash_s, *slave_s;
3528 struct iface *iface;
3529 struct bond_entry *entry;
3532 bond_s = strtok_r(args, " ", &save_ptr);
3533 hash_s = strtok_r(NULL, " ", &save_ptr);
3534 slave_s = strtok_r(NULL, " ", &save_ptr);
3536 unixctl_command_reply(conn, 501,
3537 "usage: bond/migrate BOND HASH SLAVE");
3541 port = bond_find(bond_s);
3543 unixctl_command_reply(conn, 501, "no such bond");
3547 if (port->bond_mode != BM_SLB) {
3548 unixctl_command_reply(conn, 501, "not an SLB bond");
3552 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3553 hash = atoi(hash_s) & BOND_MASK;
3555 unixctl_command_reply(conn, 501, "bad hash");
3559 iface = port_lookup_iface(port, slave_s);
3561 unixctl_command_reply(conn, 501, "no such slave");
3565 if (!iface->enabled) {
3566 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3570 entry = &port->bond_hash[hash];
3571 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3572 entry->iface_idx = iface->port_ifidx;
3573 entry->iface_tag = tag_create_random();
3574 port->bond_compat_is_stale = true;
3575 unixctl_command_reply(conn, 200, "migrated");
3579 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3580 void *aux OVS_UNUSED)
3582 char *args = (char *) args_;
3583 char *save_ptr = NULL;
3584 char *bond_s, *slave_s;
3586 struct iface *iface;
3588 bond_s = strtok_r(args, " ", &save_ptr);
3589 slave_s = strtok_r(NULL, " ", &save_ptr);
3591 unixctl_command_reply(conn, 501,
3592 "usage: bond/set-active-slave BOND SLAVE");
3596 port = bond_find(bond_s);
3598 unixctl_command_reply(conn, 501, "no such bond");
3602 iface = port_lookup_iface(port, slave_s);
3604 unixctl_command_reply(conn, 501, "no such slave");
3608 if (!iface->enabled) {
3609 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3613 if (port->active_iface != iface->port_ifidx) {
3614 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3615 port->active_iface = iface->port_ifidx;
3616 port->active_iface_tag = tag_create_random();
3617 VLOG_INFO("port %s: active interface is now %s",
3618 port->name, iface->name);
3619 bond_send_learning_packets(port);
3620 unixctl_command_reply(conn, 200, "done");
3622 unixctl_command_reply(conn, 200, "no change");
3627 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3629 char *args = (char *) args_;
3630 char *save_ptr = NULL;
3631 char *bond_s, *slave_s;
3633 struct iface *iface;
3635 bond_s = strtok_r(args, " ", &save_ptr);
3636 slave_s = strtok_r(NULL, " ", &save_ptr);
3638 unixctl_command_reply(conn, 501,
3639 "usage: bond/enable/disable-slave BOND SLAVE");
3643 port = bond_find(bond_s);
3645 unixctl_command_reply(conn, 501, "no such bond");
3649 iface = port_lookup_iface(port, slave_s);
3651 unixctl_command_reply(conn, 501, "no such slave");
3655 bond_enable_slave(iface, enable);
3656 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3660 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3661 void *aux OVS_UNUSED)
3663 enable_slave(conn, args, true);
3667 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3668 void *aux OVS_UNUSED)
3670 enable_slave(conn, args, false);
3674 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3675 void *aux OVS_UNUSED)
3677 char *args = (char *) args_;
3678 uint8_t mac[ETH_ADDR_LEN];
3682 char *mac_s, *vlan_s;
3683 char *save_ptr = NULL;
3685 mac_s = strtok_r(args, " ", &save_ptr);
3686 vlan_s = strtok_r(NULL, " ", &save_ptr);
3689 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3690 unixctl_command_reply(conn, 501, "invalid vlan");
3694 vlan = OFP_VLAN_NONE;
3697 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3698 == ETH_ADDR_SCAN_COUNT) {
3699 hash = bond_hash(mac, vlan);
3701 hash_cstr = xasprintf("%u", hash);
3702 unixctl_command_reply(conn, 200, hash_cstr);
3705 unixctl_command_reply(conn, 501, "invalid mac");
3712 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3713 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3714 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3715 unixctl_command_register("bond/set-active-slave",
3716 bond_unixctl_set_active_slave, NULL);
3717 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3719 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3721 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3724 /* Port functions. */
3726 static struct port *
3727 port_create(struct bridge *br, const char *name)
3731 port = xzalloc(sizeof *port);
3733 port->port_idx = br->n_ports;
3735 port->trunks = NULL;
3736 port->name = xstrdup(name);
3737 port->active_iface = -1;
3739 if (br->n_ports >= br->allocated_ports) {
3740 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3743 br->ports[br->n_ports++] = port;
3744 shash_add_assert(&br->port_by_name, port->name, port);
3746 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3753 get_port_other_config(const struct ovsrec_port *port, const char *key,
3754 const char *default_value)
3758 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3760 return value ? value : default_value;
3764 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3766 struct shash new_ifaces;
3769 /* Collect list of new interfaces. */
3770 shash_init(&new_ifaces);
3771 for (i = 0; i < cfg->n_interfaces; i++) {
3772 const char *name = cfg->interfaces[i]->name;
3773 shash_add_once(&new_ifaces, name, NULL);
3776 /* Get rid of deleted interfaces. */
3777 for (i = 0; i < port->n_ifaces; ) {
3778 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3779 iface_destroy(port->ifaces[i]);
3785 shash_destroy(&new_ifaces);
3789 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3791 const char *detect_mode;
3792 struct shash new_ifaces;
3793 long long int next_rebalance, miimon_next_update;
3794 unsigned long *trunks;
3800 /* Update settings. */
3801 port->updelay = cfg->bond_updelay;
3802 if (port->updelay < 0) {
3805 port->downdelay = cfg->bond_downdelay;
3806 if (port->downdelay < 0) {
3807 port->downdelay = 0;
3809 port->bond_rebalance_interval = atoi(
3810 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3811 if (port->bond_rebalance_interval < 1000) {
3812 port->bond_rebalance_interval = 1000;
3814 next_rebalance = time_msec() + port->bond_rebalance_interval;
3815 if (port->bond_next_rebalance > next_rebalance) {
3816 port->bond_next_rebalance = next_rebalance;
3819 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
3822 if (!strcmp(detect_mode, "carrier")) {
3823 port->miimon = false;
3824 } else if (!strcmp(detect_mode, "miimon")) {
3825 port->miimon = true;
3827 port->miimon = false;
3828 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
3829 "carrier", port->name, detect_mode);
3832 port->bond_miimon_interval = atoi(
3833 get_port_other_config(cfg, "bond-miimon-interval", "200"));
3834 if (port->bond_miimon_interval < 100) {
3835 port->bond_miimon_interval = 100;
3837 miimon_next_update = time_msec() + port->bond_miimon_interval;
3838 if (port->bond_miimon_next_update > miimon_next_update) {
3839 port->bond_miimon_next_update = miimon_next_update;
3842 if (!port->cfg->bond_mode ||
3843 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
3844 port->bond_mode = BM_SLB;
3845 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
3846 port->bond_mode = BM_AB;
3848 port->bond_mode = BM_SLB;
3849 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
3850 port->name, port->cfg->bond_mode,
3851 bond_mode_to_string(port->bond_mode));
3854 /* Add new interfaces and update 'cfg' member of existing ones. */
3855 shash_init(&new_ifaces);
3856 for (i = 0; i < cfg->n_interfaces; i++) {
3857 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3858 struct iface *iface;
3860 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3861 VLOG_WARN("port %s: %s specified twice as port interface",
3862 port->name, if_cfg->name);
3863 iface_set_ofport(if_cfg, -1);
3867 iface = iface_lookup(port->bridge, if_cfg->name);
3869 if (iface->port != port) {
3870 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3872 port->bridge->name, if_cfg->name, iface->port->name);
3875 iface->cfg = if_cfg;
3877 iface = iface_create(port, if_cfg);
3880 /* Determine interface type. The local port always has type
3881 * "internal". Other ports take their type from the database and
3882 * default to "system" if none is specified. */
3883 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3884 : if_cfg->type[0] ? if_cfg->type
3887 shash_destroy(&new_ifaces);
3892 if (port->n_ifaces < 2) {
3894 if (vlan >= 0 && vlan <= 4095) {
3895 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3900 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3901 * they even work as-is. But they have not been tested. */
3902 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3906 if (port->vlan != vlan) {
3908 bridge_flush(port->bridge);
3911 /* Get trunked VLANs. */
3913 if (vlan < 0 && cfg->n_trunks) {
3916 trunks = bitmap_allocate(4096);
3918 for (i = 0; i < cfg->n_trunks; i++) {
3919 int trunk = cfg->trunks[i];
3921 bitmap_set1(trunks, trunk);
3927 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3928 port->name, cfg->n_trunks);
3930 if (n_errors == cfg->n_trunks) {
3931 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3933 bitmap_free(trunks);
3936 } else if (vlan >= 0 && cfg->n_trunks) {
3937 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3941 ? port->trunks != NULL
3942 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3943 bridge_flush(port->bridge);
3945 bitmap_free(port->trunks);
3946 port->trunks = trunks;
3950 port_destroy(struct port *port)
3953 struct bridge *br = port->bridge;
3957 proc_net_compat_update_vlan(port->name, NULL, 0);
3958 proc_net_compat_update_bond(port->name, NULL);
3960 for (i = 0; i < MAX_MIRRORS; i++) {
3961 struct mirror *m = br->mirrors[i];
3962 if (m && m->out_port == port) {
3967 while (port->n_ifaces > 0) {
3968 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3971 shash_find_and_delete_assert(&br->port_by_name, port->name);
3973 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3974 del->port_idx = port->port_idx;
3976 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3978 netdev_monitor_destroy(port->monitor);
3980 bitmap_free(port->trunks);
3987 static struct port *
3988 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3990 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3991 return iface ? iface->port : NULL;
3994 static struct port *
3995 port_lookup(const struct bridge *br, const char *name)
3997 return shash_find_data(&br->port_by_name, name);
4000 static struct iface *
4001 port_lookup_iface(const struct port *port, const char *name)
4003 struct iface *iface = iface_lookup(port->bridge, name);
4004 return iface && iface->port == port ? iface : NULL;
4008 port_update_bonding(struct port *port)
4010 if (port->monitor) {
4011 netdev_monitor_destroy(port->monitor);
4012 port->monitor = NULL;
4014 if (port->n_ifaces < 2) {
4015 /* Not a bonded port. */
4016 if (port->bond_hash) {
4017 free(port->bond_hash);
4018 port->bond_hash = NULL;
4019 port->bond_compat_is_stale = true;
4022 port->bond_fake_iface = false;
4026 if (port->bond_mode == BM_SLB && !port->bond_hash) {
4027 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4028 for (i = 0; i <= BOND_MASK; i++) {
4029 struct bond_entry *e = &port->bond_hash[i];
4033 port->no_ifaces_tag = tag_create_random();
4034 bond_choose_active_iface(port);
4035 port->bond_next_rebalance
4036 = time_msec() + port->bond_rebalance_interval;
4038 if (port->cfg->bond_fake_iface) {
4039 port->bond_next_fake_iface_update = time_msec();
4041 } else if (port->bond_mode != BM_SLB) {
4042 free(port->bond_hash);
4043 port->bond_hash = NULL;
4045 port->bond_compat_is_stale = true;
4046 port->bond_fake_iface = port->cfg->bond_fake_iface;
4048 if (!port->miimon) {
4049 port->monitor = netdev_monitor_create();
4050 for (i = 0; i < port->n_ifaces; i++) {
4051 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4058 port_update_bond_compat(struct port *port)
4060 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
4061 struct compat_bond bond;
4064 if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) {
4065 proc_net_compat_update_bond(port->name, NULL);
4070 bond.updelay = port->updelay;
4071 bond.downdelay = port->downdelay;
4074 bond.hashes = compat_hashes;
4075 if (port->bond_hash) {
4076 const struct bond_entry *e;
4077 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4078 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
4079 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
4080 cbh->hash = e - port->bond_hash;
4081 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
4086 bond.n_slaves = port->n_ifaces;
4087 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
4088 for (i = 0; i < port->n_ifaces; i++) {
4089 struct iface *iface = port->ifaces[i];
4090 struct compat_bond_slave *slave = &bond.slaves[i];
4091 slave->name = iface->name;
4093 /* We need to make the same determination as the Linux bonding
4094 * code to determine whether a slave should be consider "up".
4095 * The Linux function bond_miimon_inspect() supports four
4096 * BOND_LINK_* states:
4098 * - BOND_LINK_UP: carrier detected, updelay has passed.
4099 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
4100 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
4101 * - BOND_LINK_BACK: carrier detected, updelay in progress.
4103 * The function bond_info_show_slave() only considers BOND_LINK_UP
4104 * to be "up" and anything else to be "down".
4106 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
4110 netdev_get_etheraddr(iface->netdev, slave->mac);
4113 if (port->bond_fake_iface) {
4114 struct netdev *bond_netdev;
4116 if (!netdev_open_default(port->name, &bond_netdev)) {
4118 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
4120 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
4122 netdev_close(bond_netdev);
4126 proc_net_compat_update_bond(port->name, &bond);
4131 port_update_vlan_compat(struct port *port)
4133 struct bridge *br = port->bridge;
4134 char *vlandev_name = NULL;
4136 if (port->vlan > 0) {
4137 /* Figure out the name that the VLAN device should actually have, if it
4138 * existed. This takes some work because the VLAN device would not
4139 * have port->name in its name; rather, it would have the trunk port's
4140 * name, and 'port' would be attached to a bridge that also had the
4141 * VLAN device one of its ports. So we need to find a trunk port that
4142 * includes port->vlan.
4144 * There might be more than one candidate. This doesn't happen on
4145 * XenServer, so if it happens we just pick the first choice in
4146 * alphabetical order instead of creating multiple VLAN devices. */
4148 for (i = 0; i < br->n_ports; i++) {
4149 struct port *p = br->ports[i];
4150 if (port_trunks_vlan(p, port->vlan)
4152 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
4154 uint8_t ea[ETH_ADDR_LEN];
4155 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
4156 if (!eth_addr_is_multicast(ea) &&
4157 !eth_addr_is_reserved(ea) &&
4158 !eth_addr_is_zero(ea)) {
4159 vlandev_name = p->name;
4164 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
4167 /* Interface functions. */
4170 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4173 union ofp_action action;
4175 memset(&action, 0, sizeof action);
4176 action.output.type = htons(OFPAT_OUTPUT);
4177 action.output.len = htons(sizeof action);
4178 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4180 flow_extract(packet, 0, ODPP_NONE, &flow);
4182 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4184 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4185 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4189 static struct iface *
4190 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4192 struct bridge *br = port->bridge;
4193 struct iface *iface;
4194 char *name = if_cfg->name;
4196 iface = xzalloc(sizeof *iface);
4198 iface->port_ifidx = port->n_ifaces;
4199 iface->name = xstrdup(name);
4200 iface->dp_ifidx = -1;
4201 iface->tag = tag_create_random();
4202 iface->delay_expires = LLONG_MAX;
4203 iface->netdev = NULL;
4204 iface->cfg = if_cfg;
4206 shash_add_assert(&br->iface_by_name, iface->name, iface);
4208 if (port->n_ifaces >= port->allocated_ifaces) {
4209 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4210 sizeof *port->ifaces);
4212 port->ifaces[port->n_ifaces++] = iface;
4213 if (port->n_ifaces > 1) {
4214 br->has_bonded_ports = true;
4217 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4225 iface_destroy(struct iface *iface)
4228 struct port *port = iface->port;
4229 struct bridge *br = port->bridge;
4230 bool del_active = port->active_iface == iface->port_ifidx;
4233 if (port->monitor) {
4234 netdev_monitor_remove(port->monitor, iface->netdev);
4237 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4239 if (iface->dp_ifidx >= 0) {
4240 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4243 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4244 del->port_ifidx = iface->port_ifidx;
4246 netdev_close(iface->netdev);
4249 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4250 bond_choose_active_iface(port);
4251 bond_send_learning_packets(port);
4254 cfm_destroy(iface->cfm);
4259 bridge_flush(port->bridge);
4263 static struct iface *
4264 iface_lookup(const struct bridge *br, const char *name)
4266 return shash_find_data(&br->iface_by_name, name);
4269 static struct iface *
4270 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4272 struct iface *iface;
4274 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4275 hash_int(dp_ifidx, 0), &br->ifaces) {
4276 if (iface->dp_ifidx == dp_ifidx) {
4283 /* Set Ethernet address of 'iface', if one is specified in the configuration
4286 iface_set_mac(struct iface *iface)
4288 uint8_t ea[ETH_ADDR_LEN];
4290 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4291 if (eth_addr_is_multicast(ea)) {
4292 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4294 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4295 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4296 iface->name, iface->name);
4298 int error = netdev_set_etheraddr(iface->netdev, ea);
4300 VLOG_ERR("interface %s: setting MAC failed (%s)",
4301 iface->name, strerror(error));
4307 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4309 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4312 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4316 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4318 * The value strings in '*shash' are taken directly from values[], not copied,
4319 * so the caller should not modify or free them. */
4321 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4322 struct shash *shash)
4327 for (i = 0; i < n; i++) {
4328 shash_add(shash, keys[i], values[i]);
4332 /* Creates 'keys' and 'values' arrays from 'shash'.
4334 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4335 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4336 * are populated with with strings taken directly from 'shash' and thus have
4337 * the same ownership of the key-value pairs in shash.
4340 shash_to_ovs_idl_map(struct shash *shash,
4341 char ***keys, char ***values, size_t *n)
4345 struct shash_node *sn;
4347 count = shash_count(shash);
4349 k = xmalloc(count * sizeof *k);
4350 v = xmalloc(count * sizeof *v);
4353 SHASH_FOR_EACH(sn, shash) {
4364 struct iface_delete_queues_cbdata {
4365 struct netdev *netdev;
4366 const struct ovsdb_datum *queues;
4370 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4372 union ovsdb_atom atom;
4374 atom.integer = target;
4375 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4379 iface_delete_queues(unsigned int queue_id,
4380 const struct shash *details OVS_UNUSED, void *cbdata_)
4382 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4384 if (!queue_ids_include(cbdata->queues, queue_id)) {
4385 netdev_delete_queue(cbdata->netdev, queue_id);
4390 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4392 if (!qos || qos->type[0] == '\0') {
4393 netdev_set_qos(iface->netdev, NULL, NULL);
4395 struct iface_delete_queues_cbdata cbdata;
4396 struct shash details;
4399 /* Configure top-level Qos for 'iface'. */
4400 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4401 qos->n_other_config, &details);
4402 netdev_set_qos(iface->netdev, qos->type, &details);
4403 shash_destroy(&details);
4405 /* Deconfigure queues that were deleted. */
4406 cbdata.netdev = iface->netdev;
4407 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4409 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4411 /* Configure queues for 'iface'. */
4412 for (i = 0; i < qos->n_queues; i++) {
4413 const struct ovsrec_queue *queue = qos->value_queues[i];
4414 unsigned int queue_id = qos->key_queues[i];
4416 shash_from_ovs_idl_map(queue->key_other_config,
4417 queue->value_other_config,
4418 queue->n_other_config, &details);
4419 netdev_set_queue(iface->netdev, queue_id, &details);
4420 shash_destroy(&details);
4426 iface_update_cfm(struct iface *iface)
4430 uint16_t *remote_mps;
4431 struct ovsrec_monitor *mon;
4432 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4434 mon = iface->cfg->monitor;
4440 if (netdev_get_etheraddr(iface->netdev, ea)) {
4441 VLOG_WARN("interface %s: Failed to get ethernet address. "
4442 "Skipping Monitor.", iface->name);
4446 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4447 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4452 iface->cfm = cfm_create();
4456 cfm->mpid = mon->mpid;
4457 cfm->interval = mon->interval ? *mon->interval : 1000;
4459 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4460 memcpy(cfm->maid, maid, sizeof cfm->maid);
4462 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4463 for(i = 0; i < mon->n_remote_mps; i++) {
4464 remote_mps[i] = mon->remote_mps[i]->mpid;
4466 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4469 if (!cfm_configure(iface->cfm)) {
4470 cfm_destroy(iface->cfm);
4475 /* Port mirroring. */
4477 static struct mirror *
4478 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4482 for (i = 0; i < MAX_MIRRORS; i++) {
4483 struct mirror *m = br->mirrors[i];
4484 if (m && uuid_equals(uuid, &m->uuid)) {
4492 mirror_reconfigure(struct bridge *br)
4494 unsigned long *rspan_vlans;
4497 /* Get rid of deleted mirrors. */
4498 for (i = 0; i < MAX_MIRRORS; i++) {
4499 struct mirror *m = br->mirrors[i];
4501 const struct ovsdb_datum *mc;
4502 union ovsdb_atom atom;
4504 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4505 atom.uuid = br->mirrors[i]->uuid;
4506 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4512 /* Add new mirrors and reconfigure existing ones. */
4513 for (i = 0; i < br->cfg->n_mirrors; i++) {
4514 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4515 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4517 mirror_reconfigure_one(m, cfg);
4519 mirror_create(br, cfg);
4523 /* Update port reserved status. */
4524 for (i = 0; i < br->n_ports; i++) {
4525 br->ports[i]->is_mirror_output_port = false;
4527 for (i = 0; i < MAX_MIRRORS; i++) {
4528 struct mirror *m = br->mirrors[i];
4529 if (m && m->out_port) {
4530 m->out_port->is_mirror_output_port = true;
4534 /* Update flooded vlans (for RSPAN). */
4536 if (br->cfg->n_flood_vlans) {
4537 rspan_vlans = bitmap_allocate(4096);
4539 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4540 int64_t vlan = br->cfg->flood_vlans[i];
4541 if (vlan >= 0 && vlan < 4096) {
4542 bitmap_set1(rspan_vlans, vlan);
4543 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4546 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4551 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4557 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4562 for (i = 0; ; i++) {
4563 if (i >= MAX_MIRRORS) {
4564 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4565 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4568 if (!br->mirrors[i]) {
4573 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4576 br->mirrors[i] = m = xzalloc(sizeof *m);
4579 m->name = xstrdup(cfg->name);
4580 shash_init(&m->src_ports);
4581 shash_init(&m->dst_ports);
4587 mirror_reconfigure_one(m, cfg);
4591 mirror_destroy(struct mirror *m)
4594 struct bridge *br = m->bridge;
4597 for (i = 0; i < br->n_ports; i++) {
4598 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4599 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4602 shash_destroy(&m->src_ports);
4603 shash_destroy(&m->dst_ports);
4606 m->bridge->mirrors[m->idx] = NULL;
4615 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4616 struct shash *names)
4620 for (i = 0; i < n_ports; i++) {
4621 const char *name = ports[i]->name;
4622 if (port_lookup(m->bridge, name)) {
4623 shash_add_once(names, name, NULL);
4625 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4626 "port %s", m->bridge->name, m->name, name);
4632 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4638 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4640 for (i = 0; i < cfg->n_select_vlan; i++) {
4641 int64_t vlan = cfg->select_vlan[i];
4642 if (vlan < 0 || vlan > 4095) {
4643 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4644 m->bridge->name, m->name, vlan);
4646 (*vlans)[n_vlans++] = vlan;
4653 vlan_is_mirrored(const struct mirror *m, int vlan)
4657 for (i = 0; i < m->n_vlans; i++) {
4658 if (m->vlans[i] == vlan) {
4666 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4670 for (i = 0; i < m->n_vlans; i++) {
4671 if (port_trunks_vlan(p, m->vlans[i])) {
4679 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4681 struct shash src_ports, dst_ports;
4682 mirror_mask_t mirror_bit;
4683 struct port *out_port;
4690 if (strcmp(cfg->name, m->name)) {
4692 m->name = xstrdup(cfg->name);
4695 /* Get output port. */
4696 if (cfg->output_port) {
4697 out_port = port_lookup(m->bridge, cfg->output_port->name);
4699 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4700 m->bridge->name, m->name);
4706 if (cfg->output_vlan) {
4707 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4708 "output vlan; ignoring output vlan",
4709 m->bridge->name, m->name);
4711 } else if (cfg->output_vlan) {
4713 out_vlan = *cfg->output_vlan;
4715 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4716 m->bridge->name, m->name);
4721 shash_init(&src_ports);
4722 shash_init(&dst_ports);
4723 if (cfg->select_all) {
4724 for (i = 0; i < m->bridge->n_ports; i++) {
4725 const char *name = m->bridge->ports[i]->name;
4726 shash_add_once(&src_ports, name, NULL);
4727 shash_add_once(&dst_ports, name, NULL);
4732 /* Get ports, and drop duplicates and ports that don't exist. */
4733 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4735 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4738 /* Get all the vlans, and drop duplicate and invalid vlans. */
4739 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4742 /* Update mirror data. */
4743 if (!shash_equal_keys(&m->src_ports, &src_ports)
4744 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4745 || m->n_vlans != n_vlans
4746 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4747 || m->out_port != out_port
4748 || m->out_vlan != out_vlan) {
4749 bridge_flush(m->bridge);
4751 shash_swap(&m->src_ports, &src_ports);
4752 shash_swap(&m->dst_ports, &dst_ports);
4755 m->n_vlans = n_vlans;
4756 m->out_port = out_port;
4757 m->out_vlan = out_vlan;
4760 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4761 for (i = 0; i < m->bridge->n_ports; i++) {
4762 struct port *port = m->bridge->ports[i];
4764 if (shash_find(&m->src_ports, port->name)
4767 ? port_trunks_any_mirrored_vlan(m, port)
4768 : vlan_is_mirrored(m, port->vlan)))) {
4769 port->src_mirrors |= mirror_bit;
4771 port->src_mirrors &= ~mirror_bit;
4774 if (shash_find(&m->dst_ports, port->name)) {
4775 port->dst_mirrors |= mirror_bit;
4777 port->dst_mirrors &= ~mirror_bit;
4782 shash_destroy(&src_ports);
4783 shash_destroy(&dst_ports);