1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
40 #include "dynamic-string.h"
46 #include "mac-learning.h"
50 #include "ofp-print.h"
52 #include "ofproto/netflow.h"
53 #include "ofproto/ofproto.h"
54 #include "ovsdb-data.h"
56 #include "poll-loop.h"
57 #include "proc-net-compat.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_reconfigure);
86 struct dst builtin[32];
91 static void dst_set_init(struct dst_set *);
92 static void dst_set_add(struct dst_set *, const struct dst *);
93 static void dst_set_free(struct dst_set *);
96 /* These members are always valid. */
97 struct port *port; /* Containing port. */
98 size_t port_ifidx; /* Index within containing port. */
99 char *name; /* Host network device name. */
100 tag_type tag; /* Tag associated with this interface. */
101 long long delay_expires; /* Time after which 'enabled' may change. */
103 /* These members are valid only after bridge_reconfigure() causes them to
105 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
106 int dp_ifidx; /* Index within kernel datapath. */
107 struct netdev *netdev; /* Network device. */
108 bool enabled; /* May be chosen for flows? */
109 const char *type; /* Usually same as cfg->type. */
110 struct cfm *cfm; /* Connectivity Fault Management */
111 const struct ovsrec_interface *cfg;
114 #define BOND_MASK 0xff
116 int iface_idx; /* Index of assigned iface, or -1 if none. */
117 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
118 tag_type iface_tag; /* Tag associated with iface_idx. */
121 #define MAX_MIRRORS 32
122 typedef uint32_t mirror_mask_t;
123 #define MIRROR_MASK_C(X) UINT32_C(X)
124 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
126 struct bridge *bridge;
129 struct uuid uuid; /* UUID of this "mirror" record in database. */
131 /* Selection criteria. */
132 struct shash src_ports; /* Name is port name; data is always NULL. */
133 struct shash dst_ports; /* Name is port name; data is always NULL. */
138 struct port *out_port;
142 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
144 struct bridge *bridge;
146 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
147 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
148 * NULL if all VLANs are trunked. */
149 const struct ovsrec_port *cfg;
152 /* An ordinary bridge port has 1 interface.
153 * A bridge port for bonding has at least 2 interfaces. */
154 struct iface **ifaces;
155 size_t n_ifaces, allocated_ifaces;
158 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
159 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
160 tag_type active_iface_tag; /* Tag for bcast flows. */
161 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
162 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
163 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
164 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
165 long long int bond_next_fake_iface_update; /* Time of next update. */
166 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
167 long long int bond_next_rebalance; /* Next rebalancing time. */
168 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
170 /* Port mirroring info. */
171 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
172 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
173 bool is_mirror_output_port; /* Does port mirroring send frames here? */
177 struct list node; /* Node in global list of bridges. */
178 char *name; /* User-specified arbitrary name. */
179 struct mac_learning *ml; /* MAC learning table. */
180 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
181 const struct ovsrec_bridge *cfg;
183 /* OpenFlow switch processing. */
184 struct ofproto *ofproto; /* OpenFlow switch. */
186 /* Kernel datapath information. */
187 struct dpif *dpif; /* Datapath. */
188 struct hmap ifaces; /* Contains "struct iface"s. */
192 size_t n_ports, allocated_ports;
193 struct shash iface_by_name; /* "struct iface"s indexed by name. */
194 struct shash port_by_name; /* "struct port"s indexed by name. */
197 bool has_bonded_ports;
202 /* Port mirroring. */
203 struct mirror *mirrors[MAX_MIRRORS];
206 /* List of all bridges. */
207 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
209 /* OVSDB IDL used to obtain configuration. */
210 static struct ovsdb_idl *idl;
212 /* Each time this timer expires, the bridge fetches systems and interface
213 * statistics and pushes them into the database. */
214 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
215 static long long int stats_timer = LLONG_MIN;
217 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
218 static void bridge_destroy(struct bridge *);
219 static struct bridge *bridge_lookup(const char *name);
220 static unixctl_cb_func bridge_unixctl_dump_flows;
221 static unixctl_cb_func bridge_unixctl_reconnect;
222 static int bridge_run_one(struct bridge *);
223 static size_t bridge_get_controllers(const struct bridge *br,
224 struct ovsrec_controller ***controllersp);
225 static void bridge_reconfigure_one(struct bridge *);
226 static void bridge_reconfigure_remotes(struct bridge *,
227 const struct sockaddr_in *managers,
229 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
230 static void bridge_fetch_dp_ifaces(struct bridge *);
231 static void bridge_flush(struct bridge *);
232 static void bridge_pick_local_hw_addr(struct bridge *,
233 uint8_t ea[ETH_ADDR_LEN],
234 struct iface **hw_addr_iface);
235 static uint64_t bridge_pick_datapath_id(struct bridge *,
236 const uint8_t bridge_ea[ETH_ADDR_LEN],
237 struct iface *hw_addr_iface);
238 static struct iface *bridge_get_local_iface(struct bridge *);
239 static uint64_t dpid_from_hash(const void *, size_t nbytes);
241 static unixctl_cb_func bridge_unixctl_fdb_show;
243 static void bond_init(void);
244 static void bond_run(struct bridge *);
245 static void bond_wait(struct bridge *);
246 static void bond_rebalance_port(struct port *);
247 static void bond_send_learning_packets(struct port *);
248 static void bond_enable_slave(struct iface *iface, bool enable);
250 static struct port *port_create(struct bridge *, const char *name);
251 static void port_reconfigure(struct port *, const struct ovsrec_port *);
252 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
253 static void port_destroy(struct port *);
254 static struct port *port_lookup(const struct bridge *, const char *name);
255 static struct iface *port_lookup_iface(const struct port *, const char *name);
256 static struct port *port_from_dp_ifidx(const struct bridge *,
258 static void port_update_bond_compat(struct port *);
259 static void port_update_vlan_compat(struct port *);
260 static void port_update_bonding(struct port *);
262 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
263 static void mirror_destroy(struct mirror *);
264 static void mirror_reconfigure(struct bridge *);
265 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
266 static bool vlan_is_mirrored(const struct mirror *, int vlan);
268 static struct iface *iface_create(struct port *port,
269 const struct ovsrec_interface *if_cfg);
270 static void iface_destroy(struct iface *);
271 static struct iface *iface_lookup(const struct bridge *, const char *name);
272 static struct iface *iface_from_dp_ifidx(const struct bridge *,
274 static void iface_set_mac(struct iface *);
275 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
276 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
277 static void iface_update_cfm(struct iface *);
278 static void iface_refresh_cfm_stats(struct iface *iface);
279 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
281 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
284 /* Hooks into ofproto processing. */
285 static struct ofhooks bridge_ofhooks;
287 /* Public functions. */
289 /* Initializes the bridge module, configuring it to obtain its configuration
290 * from an OVSDB server accessed over 'remote', which should be a string in a
291 * form acceptable to ovsdb_idl_create(). */
293 bridge_init(const char *remote)
295 /* Create connection to database. */
296 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
298 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
299 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
300 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
302 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
304 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
305 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
307 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
308 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
309 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
311 /* Register unixctl commands. */
312 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
313 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
315 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
323 struct bridge *br, *next_br;
325 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
328 ovsdb_idl_destroy(idl);
331 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
332 * but for which the ovs-vswitchd configuration 'cfg' is required. */
334 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
336 static bool already_configured_once;
337 struct svec bridge_names;
338 struct svec dpif_names, dpif_types;
341 /* Only do this once per ovs-vswitchd run. */
342 if (already_configured_once) {
345 already_configured_once = true;
347 stats_timer = time_msec() + STATS_INTERVAL;
349 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
350 svec_init(&bridge_names);
351 for (i = 0; i < cfg->n_bridges; i++) {
352 svec_add(&bridge_names, cfg->bridges[i]->name);
354 svec_sort(&bridge_names);
356 /* Iterate over all system dpifs and delete any of them that do not appear
358 svec_init(&dpif_names);
359 svec_init(&dpif_types);
360 dp_enumerate_types(&dpif_types);
361 for (i = 0; i < dpif_types.n; i++) {
366 dp_enumerate_names(dpif_types.names[i], &dpif_names);
368 /* For each dpif... */
369 for (j = 0; j < dpif_names.n; j++) {
370 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
372 struct svec all_names;
375 /* ...check whether any of its names is in 'bridge_names'. */
376 svec_init(&all_names);
377 dpif_get_all_names(dpif, &all_names);
378 for (k = 0; k < all_names.n; k++) {
379 if (svec_contains(&bridge_names, all_names.names[k])) {
384 /* No. Delete the dpif. */
388 svec_destroy(&all_names);
393 svec_destroy(&bridge_names);
394 svec_destroy(&dpif_names);
395 svec_destroy(&dpif_types);
398 /* Callback for iterate_and_prune_ifaces(). */
400 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
402 if (!iface->netdev) {
403 /* We already reported a related error, don't bother duplicating it. */
407 if (iface->dp_ifidx < 0) {
408 VLOG_ERR("%s interface not in %s, dropping",
409 iface->name, dpif_name(br->dpif));
413 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
414 iface->name, iface->dp_ifidx);
418 /* Callback for iterate_and_prune_ifaces(). */
420 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
421 void *aux OVS_UNUSED)
423 /* Set policing attributes. */
424 netdev_set_policing(iface->netdev,
425 iface->cfg->ingress_policing_rate,
426 iface->cfg->ingress_policing_burst);
428 /* Set MAC address of internal interfaces other than the local
430 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
431 iface_set_mac(iface);
437 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
438 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
439 * deletes from 'br' any ports that no longer have any interfaces. */
441 iterate_and_prune_ifaces(struct bridge *br,
442 bool (*cb)(struct bridge *, struct iface *,
448 for (i = 0; i < br->n_ports; ) {
449 struct port *port = br->ports[i];
450 for (j = 0; j < port->n_ifaces; ) {
451 struct iface *iface = port->ifaces[j];
452 if (cb(br, iface, aux)) {
455 iface_set_ofport(iface->cfg, -1);
456 iface_destroy(iface);
460 if (port->n_ifaces) {
463 VLOG_ERR("%s port has no interfaces, dropping", port->name);
469 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
470 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
471 * responsible for freeing '*managersp' (with free()).
473 * You may be asking yourself "why does ovs-vswitchd care?", because
474 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
475 * should not be and in fact is not directly involved in that. But
476 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
477 * it has to tell in-band control where the managers are to enable that.
478 * (Thus, only managers connected in-band are collected.)
481 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
482 struct sockaddr_in **managersp, size_t *n_managersp)
484 struct sockaddr_in *managers = NULL;
485 size_t n_managers = 0;
486 struct shash targets;
489 /* Collect all of the potential targets, as the union of the "managers"
490 * column and the "targets" columns of the rows pointed to by
491 * "manager_options", excluding any that are out-of-band. */
492 shash_init(&targets);
493 for (i = 0; i < ovs_cfg->n_managers; i++) {
494 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
496 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
497 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
499 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
500 shash_find_and_delete(&targets, m->target);
502 shash_add_once(&targets, m->target, NULL);
506 /* Now extract the targets' IP addresses. */
507 if (!shash_is_empty(&targets)) {
508 struct shash_node *node;
510 managers = xmalloc(shash_count(&targets) * sizeof *managers);
511 SHASH_FOR_EACH (node, &targets) {
512 const char *target = node->name;
513 struct sockaddr_in *sin = &managers[n_managers];
515 if ((!strncmp(target, "tcp:", 4)
516 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
517 (!strncmp(target, "ssl:", 4)
518 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
523 shash_destroy(&targets);
525 *managersp = managers;
526 *n_managersp = n_managers;
530 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
532 struct shash old_br, new_br;
533 struct shash_node *node;
534 struct bridge *br, *next;
535 struct sockaddr_in *managers;
538 int sflow_bridge_number;
540 COVERAGE_INC(bridge_reconfigure);
542 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
544 /* Collect old and new bridges. */
547 LIST_FOR_EACH (br, node, &all_bridges) {
548 shash_add(&old_br, br->name, br);
550 for (i = 0; i < ovs_cfg->n_bridges; i++) {
551 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
552 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
553 VLOG_WARN("more than one bridge named %s", br_cfg->name);
557 /* Get rid of deleted bridges and add new bridges. */
558 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
559 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
566 SHASH_FOR_EACH (node, &new_br) {
567 const char *br_name = node->name;
568 const struct ovsrec_bridge *br_cfg = node->data;
569 br = shash_find_data(&old_br, br_name);
571 /* If the bridge datapath type has changed, we need to tear it
572 * down and recreate. */
573 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
575 bridge_create(br_cfg);
578 bridge_create(br_cfg);
581 shash_destroy(&old_br);
582 shash_destroy(&new_br);
584 /* Reconfigure all bridges. */
585 LIST_FOR_EACH (br, node, &all_bridges) {
586 bridge_reconfigure_one(br);
589 /* Add and delete ports on all datapaths.
591 * The kernel will reject any attempt to add a given port to a datapath if
592 * that port already belongs to a different datapath, so we must do all
593 * port deletions before any port additions. */
594 LIST_FOR_EACH (br, node, &all_bridges) {
595 struct odp_port *dpif_ports;
597 struct shash want_ifaces;
599 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
600 bridge_get_all_ifaces(br, &want_ifaces);
601 for (i = 0; i < n_dpif_ports; i++) {
602 const struct odp_port *p = &dpif_ports[i];
603 if (!shash_find(&want_ifaces, p->devname)
604 && strcmp(p->devname, br->name)) {
605 int retval = dpif_port_del(br->dpif, p->port);
607 VLOG_ERR("failed to remove %s interface from %s: %s",
608 p->devname, dpif_name(br->dpif),
613 shash_destroy(&want_ifaces);
616 LIST_FOR_EACH (br, node, &all_bridges) {
617 struct odp_port *dpif_ports;
619 struct shash cur_ifaces, want_ifaces;
621 /* Get the set of interfaces currently in this datapath. */
622 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
623 shash_init(&cur_ifaces);
624 for (i = 0; i < n_dpif_ports; i++) {
625 const char *name = dpif_ports[i].devname;
626 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
629 /* Get the set of interfaces we want on this datapath. */
630 bridge_get_all_ifaces(br, &want_ifaces);
632 hmap_clear(&br->ifaces);
633 SHASH_FOR_EACH (node, &want_ifaces) {
634 const char *if_name = node->name;
635 struct iface *iface = node->data;
636 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
637 const char *type = iface ? iface->type : "internal";
640 /* If we have a port or a netdev already, and it's not the type we
641 * want, then delete the port (if any) and close the netdev (if
643 if ((dpif_port && strcmp(dpif_port->type, type))
644 || (iface && iface->netdev
645 && strcmp(type, netdev_get_type(iface->netdev)))) {
647 error = ofproto_port_del(br->ofproto, dpif_port->port);
654 netdev_close(iface->netdev);
655 iface->netdev = NULL;
659 /* If the port doesn't exist or we don't have the netdev open,
660 * we need to do more work. */
661 if (!dpif_port || (iface && !iface->netdev)) {
662 struct netdev_options options;
663 struct netdev *netdev;
666 /* First open the network device. */
667 options.name = if_name;
669 options.args = &args;
670 options.ethertype = NETDEV_ETH_TYPE_NONE;
674 shash_from_ovs_idl_map(iface->cfg->key_options,
675 iface->cfg->value_options,
676 iface->cfg->n_options, &args);
678 error = netdev_open(&options, &netdev);
679 shash_destroy(&args);
682 VLOG_WARN("could not open network device %s (%s)",
683 if_name, strerror(error));
687 /* Then add the port if we haven't already. */
689 error = dpif_port_add(br->dpif, netdev, NULL);
691 netdev_close(netdev);
692 if (error == EFBIG) {
693 VLOG_ERR("ran out of valid port numbers on %s",
694 dpif_name(br->dpif));
697 VLOG_ERR("failed to add %s interface to %s: %s",
698 if_name, dpif_name(br->dpif),
705 /* Update 'iface'. */
707 iface->netdev = netdev;
708 iface->enabled = netdev_get_carrier(iface->netdev);
710 } else if (iface && iface->netdev) {
714 shash_from_ovs_idl_map(iface->cfg->key_options,
715 iface->cfg->value_options,
716 iface->cfg->n_options, &args);
717 netdev_reconfigure(iface->netdev, &args);
718 shash_destroy(&args);
722 shash_destroy(&cur_ifaces);
723 shash_destroy(&want_ifaces);
725 sflow_bridge_number = 0;
726 LIST_FOR_EACH (br, node, &all_bridges) {
729 struct iface *local_iface;
730 struct iface *hw_addr_iface;
733 bridge_fetch_dp_ifaces(br);
735 iterate_and_prune_ifaces(br, check_iface, NULL);
737 /* Pick local port hardware address, datapath ID. */
738 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
739 local_iface = bridge_get_local_iface(br);
741 int error = netdev_set_etheraddr(local_iface->netdev, ea);
743 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
744 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
745 "Ethernet address: %s",
746 br->name, strerror(error));
750 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
751 ofproto_set_datapath_id(br->ofproto, dpid);
753 dpid_string = xasprintf("%016"PRIx64, dpid);
754 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
757 /* Set NetFlow configuration on this bridge. */
758 if (br->cfg->netflow) {
759 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
760 struct netflow_options opts;
762 memset(&opts, 0, sizeof opts);
764 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
765 if (nf_cfg->engine_type) {
766 opts.engine_type = *nf_cfg->engine_type;
768 if (nf_cfg->engine_id) {
769 opts.engine_id = *nf_cfg->engine_id;
772 opts.active_timeout = nf_cfg->active_timeout;
773 if (!opts.active_timeout) {
774 opts.active_timeout = -1;
775 } else if (opts.active_timeout < 0) {
776 VLOG_WARN("bridge %s: active timeout interval set to negative "
777 "value, using default instead (%d seconds)", br->name,
778 NF_ACTIVE_TIMEOUT_DEFAULT);
779 opts.active_timeout = -1;
782 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
783 if (opts.add_id_to_iface) {
784 if (opts.engine_id > 0x7f) {
785 VLOG_WARN("bridge %s: netflow port mangling may conflict "
786 "with another vswitch, choose an engine id less "
787 "than 128", br->name);
789 if (br->n_ports > 508) {
790 VLOG_WARN("bridge %s: netflow port mangling will conflict "
791 "with another port when more than 508 ports are "
796 opts.collectors.n = nf_cfg->n_targets;
797 opts.collectors.names = nf_cfg->targets;
798 if (ofproto_set_netflow(br->ofproto, &opts)) {
799 VLOG_ERR("bridge %s: problem setting netflow collectors",
803 ofproto_set_netflow(br->ofproto, NULL);
806 /* Set sFlow configuration on this bridge. */
807 if (br->cfg->sflow) {
808 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
809 struct ovsrec_controller **controllers;
810 struct ofproto_sflow_options oso;
811 size_t n_controllers;
813 memset(&oso, 0, sizeof oso);
815 oso.targets.n = sflow_cfg->n_targets;
816 oso.targets.names = sflow_cfg->targets;
818 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
819 if (sflow_cfg->sampling) {
820 oso.sampling_rate = *sflow_cfg->sampling;
823 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
824 if (sflow_cfg->polling) {
825 oso.polling_interval = *sflow_cfg->polling;
828 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
829 if (sflow_cfg->header) {
830 oso.header_len = *sflow_cfg->header;
833 oso.sub_id = sflow_bridge_number++;
834 oso.agent_device = sflow_cfg->agent;
836 oso.control_ip = NULL;
837 n_controllers = bridge_get_controllers(br, &controllers);
838 for (i = 0; i < n_controllers; i++) {
839 if (controllers[i]->local_ip) {
840 oso.control_ip = controllers[i]->local_ip;
844 ofproto_set_sflow(br->ofproto, &oso);
846 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
848 ofproto_set_sflow(br->ofproto, NULL);
851 /* Update the controller and related settings. It would be more
852 * straightforward to call this from bridge_reconfigure_one(), but we
853 * can't do it there for two reasons. First, and most importantly, at
854 * that point we don't know the dp_ifidx of any interfaces that have
855 * been added to the bridge (because we haven't actually added them to
856 * the datapath). Second, at that point we haven't set the datapath ID
857 * yet; when a controller is configured, resetting the datapath ID will
858 * immediately disconnect from the controller, so it's better to set
859 * the datapath ID before the controller. */
860 bridge_reconfigure_remotes(br, managers, n_managers);
862 LIST_FOR_EACH (br, node, &all_bridges) {
863 for (i = 0; i < br->n_ports; i++) {
864 struct port *port = br->ports[i];
867 port_update_vlan_compat(port);
868 port_update_bonding(port);
870 for (j = 0; j < port->n_ifaces; j++) {
871 iface_update_qos(port->ifaces[j], port->cfg->qos);
875 LIST_FOR_EACH (br, node, &all_bridges) {
876 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
879 LIST_FOR_EACH (br, node, &all_bridges) {
881 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
882 iface_update_cfm(iface);
890 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
891 const struct ovsdb_idl_column *column,
894 const struct ovsdb_datum *datum;
895 union ovsdb_atom atom;
898 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
899 atom.string = (char *) key;
900 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
901 return idx == UINT_MAX ? NULL : datum->values[idx].string;
905 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
907 return get_ovsrec_key_value(&br_cfg->header_,
908 &ovsrec_bridge_col_other_config, key);
912 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
913 struct iface **hw_addr_iface)
919 *hw_addr_iface = NULL;
921 /* Did the user request a particular MAC? */
922 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
923 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
924 if (eth_addr_is_multicast(ea)) {
925 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
926 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
927 } else if (eth_addr_is_zero(ea)) {
928 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
934 /* Otherwise choose the minimum non-local MAC address among all of the
936 memset(ea, 0xff, sizeof ea);
937 for (i = 0; i < br->n_ports; i++) {
938 struct port *port = br->ports[i];
939 uint8_t iface_ea[ETH_ADDR_LEN];
942 /* Mirror output ports don't participate. */
943 if (port->is_mirror_output_port) {
947 /* Choose the MAC address to represent the port. */
948 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
949 /* Find the interface with this Ethernet address (if any) so that
950 * we can provide the correct devname to the caller. */
952 for (j = 0; j < port->n_ifaces; j++) {
953 struct iface *candidate = port->ifaces[j];
954 uint8_t candidate_ea[ETH_ADDR_LEN];
955 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
956 && eth_addr_equals(iface_ea, candidate_ea)) {
961 /* Choose the interface whose MAC address will represent the port.
962 * The Linux kernel bonding code always chooses the MAC address of
963 * the first slave added to a bond, and the Fedora networking
964 * scripts always add slaves to a bond in alphabetical order, so
965 * for compatibility we choose the interface with the name that is
966 * first in alphabetical order. */
967 iface = port->ifaces[0];
968 for (j = 1; j < port->n_ifaces; j++) {
969 struct iface *candidate = port->ifaces[j];
970 if (strcmp(candidate->name, iface->name) < 0) {
975 /* The local port doesn't count (since we're trying to choose its
976 * MAC address anyway). */
977 if (iface->dp_ifidx == ODPP_LOCAL) {
982 error = netdev_get_etheraddr(iface->netdev, iface_ea);
984 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
985 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
986 iface->name, strerror(error));
991 /* Compare against our current choice. */
992 if (!eth_addr_is_multicast(iface_ea) &&
993 !eth_addr_is_local(iface_ea) &&
994 !eth_addr_is_reserved(iface_ea) &&
995 !eth_addr_is_zero(iface_ea) &&
996 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
998 memcpy(ea, iface_ea, ETH_ADDR_LEN);
999 *hw_addr_iface = iface;
1002 if (eth_addr_is_multicast(ea)) {
1003 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1004 *hw_addr_iface = NULL;
1005 VLOG_WARN("bridge %s: using default bridge Ethernet "
1006 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1008 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1009 br->name, ETH_ADDR_ARGS(ea));
1013 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1014 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1015 * an interface on 'br', then that interface must be passed in as
1016 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1017 * 'hw_addr_iface' must be passed in as a null pointer. */
1019 bridge_pick_datapath_id(struct bridge *br,
1020 const uint8_t bridge_ea[ETH_ADDR_LEN],
1021 struct iface *hw_addr_iface)
1024 * The procedure for choosing a bridge MAC address will, in the most
1025 * ordinary case, also choose a unique MAC that we can use as a datapath
1026 * ID. In some special cases, though, multiple bridges will end up with
1027 * the same MAC address. This is OK for the bridges, but it will confuse
1028 * the OpenFlow controller, because each datapath needs a unique datapath
1031 * Datapath IDs must be unique. It is also very desirable that they be
1032 * stable from one run to the next, so that policy set on a datapath
1035 const char *datapath_id;
1038 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1039 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1043 if (hw_addr_iface) {
1045 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1047 * A bridge whose MAC address is taken from a VLAN network device
1048 * (that is, a network device created with vconfig(8) or similar
1049 * tool) will have the same MAC address as a bridge on the VLAN
1050 * device's physical network device.
1052 * Handle this case by hashing the physical network device MAC
1053 * along with the VLAN identifier.
1055 uint8_t buf[ETH_ADDR_LEN + 2];
1056 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1057 buf[ETH_ADDR_LEN] = vlan >> 8;
1058 buf[ETH_ADDR_LEN + 1] = vlan;
1059 return dpid_from_hash(buf, sizeof buf);
1062 * Assume that this bridge's MAC address is unique, since it
1063 * doesn't fit any of the cases we handle specially.
1068 * A purely internal bridge, that is, one that has no non-virtual
1069 * network devices on it at all, is more difficult because it has no
1070 * natural unique identifier at all.
1072 * When the host is a XenServer, we handle this case by hashing the
1073 * host's UUID with the name of the bridge. Names of bridges are
1074 * persistent across XenServer reboots, although they can be reused if
1075 * an internal network is destroyed and then a new one is later
1076 * created, so this is fairly effective.
1078 * When the host is not a XenServer, we punt by using a random MAC
1079 * address on each run.
1081 const char *host_uuid = xenserver_get_host_uuid();
1083 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1084 dpid = dpid_from_hash(combined, strlen(combined));
1090 return eth_addr_to_uint64(bridge_ea);
1094 dpid_from_hash(const void *data, size_t n)
1096 uint8_t hash[SHA1_DIGEST_SIZE];
1098 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1099 sha1_bytes(data, n, hash);
1100 eth_addr_mark_random(hash);
1101 return eth_addr_to_uint64(hash);
1105 iface_refresh_tunnel_egress(struct iface *iface)
1107 const char *name = netdev_get_tnl_iface(iface->netdev);
1109 ovsrec_interface_set_tunnel_egress_iface(iface->cfg, name);
1113 iface_refresh_cfm_stats(struct iface *iface)
1117 const struct ovsrec_monitor *mon;
1119 mon = iface->cfg->monitor;
1126 for (i = 0; i < mon->n_remote_mps; i++) {
1127 const struct ovsrec_maintenance_point *mp;
1128 const struct remote_mp *rmp;
1130 mp = mon->remote_mps[i];
1131 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1133 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1136 if (hmap_is_empty(&cfm->x_remote_mps)) {
1137 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1140 struct remote_mp *rmp;
1141 int64_t *x_remote_mps;
1143 length = hmap_count(&cfm->x_remote_mps);
1144 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1147 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1148 x_remote_mps[i++] = rmp->mpid;
1151 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1155 if (hmap_is_empty(&cfm->x_remote_maids)) {
1156 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1159 char **x_remote_maids;
1160 struct remote_maid *rmaid;
1162 length = hmap_count(&cfm->x_remote_maids);
1163 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1166 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1169 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1171 for (j = 0; j < CCM_MAID_LEN; j++) {
1172 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1177 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1179 for (i = 0; i < length; i++) {
1180 free(x_remote_maids[i]);
1182 free(x_remote_maids);
1185 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1189 iface_refresh_stats(struct iface *iface)
1195 static const struct iface_stat iface_stats[] = {
1196 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1197 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1198 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1199 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1200 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1201 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1202 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1203 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1204 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1205 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1206 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1207 { "collisions", offsetof(struct netdev_stats, collisions) },
1209 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1210 const struct iface_stat *s;
1212 char *keys[N_STATS];
1213 int64_t values[N_STATS];
1216 struct netdev_stats stats;
1218 /* Intentionally ignore return value, since errors will set 'stats' to
1219 * all-1s, and we will deal with that correctly below. */
1220 netdev_get_stats(iface->netdev, &stats);
1223 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1224 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1225 if (value != UINT64_MAX) {
1232 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1236 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1238 struct ovsdb_datum datum;
1242 get_system_stats(&stats);
1244 ovsdb_datum_from_shash(&datum, &stats);
1245 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1252 const struct ovsrec_open_vswitch *cfg;
1254 bool datapath_destroyed;
1255 bool database_changed;
1258 /* Let each bridge do the work that it needs to do. */
1259 datapath_destroyed = false;
1260 LIST_FOR_EACH (br, node, &all_bridges) {
1261 int error = bridge_run_one(br);
1263 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1264 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1265 "forcing reconfiguration", br->name);
1266 datapath_destroyed = true;
1270 /* (Re)configure if necessary. */
1271 database_changed = ovsdb_idl_run(idl);
1272 cfg = ovsrec_open_vswitch_first(idl);
1273 if (database_changed || datapath_destroyed) {
1275 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1277 bridge_configure_once(cfg);
1278 bridge_reconfigure(cfg);
1280 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1281 ovsdb_idl_txn_commit(txn);
1282 ovsdb_idl_txn_destroy(txn); /* XXX */
1284 /* We still need to reconfigure to avoid dangling pointers to
1285 * now-destroyed ovsrec structures inside bridge data. */
1286 static const struct ovsrec_open_vswitch null_cfg;
1288 bridge_reconfigure(&null_cfg);
1293 /* Re-configure SSL. We do this on every trip through the main loop,
1294 * instead of just when the database changes, because the contents of the
1295 * key and certificate files can change without the database changing. */
1296 if (cfg && cfg->ssl) {
1297 const struct ovsrec_ssl *ssl = cfg->ssl;
1299 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1300 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1304 /* Refresh system and interface stats if necessary. */
1305 if (time_msec() >= stats_timer) {
1307 struct ovsdb_idl_txn *txn;
1309 txn = ovsdb_idl_txn_create(idl);
1310 LIST_FOR_EACH (br, node, &all_bridges) {
1313 for (i = 0; i < br->n_ports; i++) {
1314 struct port *port = br->ports[i];
1317 for (j = 0; j < port->n_ifaces; j++) {
1318 struct iface *iface = port->ifaces[j];
1319 iface_refresh_stats(iface);
1320 iface_refresh_cfm_stats(iface);
1321 iface_refresh_tunnel_egress(iface);
1325 refresh_system_stats(cfg);
1326 ovsdb_idl_txn_commit(txn);
1327 ovsdb_idl_txn_destroy(txn); /* XXX */
1330 stats_timer = time_msec() + STATS_INTERVAL;
1338 struct iface *iface;
1340 LIST_FOR_EACH (br, node, &all_bridges) {
1341 ofproto_wait(br->ofproto);
1342 if (ofproto_has_primary_controller(br->ofproto)) {
1346 mac_learning_wait(br->ml);
1349 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1351 cfm_wait(iface->cfm);
1355 ovsdb_idl_wait(idl);
1356 poll_timer_wait_until(stats_timer);
1359 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1360 * configuration changes. */
1362 bridge_flush(struct bridge *br)
1364 COVERAGE_INC(bridge_flush);
1366 mac_learning_flush(br->ml);
1369 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1370 * such interface. */
1371 static struct iface *
1372 bridge_get_local_iface(struct bridge *br)
1376 for (i = 0; i < br->n_ports; i++) {
1377 struct port *port = br->ports[i];
1378 for (j = 0; j < port->n_ifaces; j++) {
1379 struct iface *iface = port->ifaces[j];
1380 if (iface->dp_ifidx == ODPP_LOCAL) {
1389 /* Bridge unixctl user interface functions. */
1391 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1392 const char *args, void *aux OVS_UNUSED)
1394 struct ds ds = DS_EMPTY_INITIALIZER;
1395 const struct bridge *br;
1396 const struct mac_entry *e;
1398 br = bridge_lookup(args);
1400 unixctl_command_reply(conn, 501, "no such bridge");
1404 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1405 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1406 if (e->port < 0 || e->port >= br->n_ports) {
1409 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1410 br->ports[e->port]->ifaces[0]->dp_ifidx,
1411 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1413 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1417 /* Bridge reconfiguration functions. */
1418 static struct bridge *
1419 bridge_create(const struct ovsrec_bridge *br_cfg)
1424 assert(!bridge_lookup(br_cfg->name));
1425 br = xzalloc(sizeof *br);
1427 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1433 dpif_flow_flush(br->dpif);
1435 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1438 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1440 dpif_delete(br->dpif);
1441 dpif_close(br->dpif);
1446 br->name = xstrdup(br_cfg->name);
1448 br->ml = mac_learning_create();
1449 eth_addr_nicira_random(br->default_ea);
1451 hmap_init(&br->ifaces);
1453 shash_init(&br->port_by_name);
1454 shash_init(&br->iface_by_name);
1458 list_push_back(&all_bridges, &br->node);
1460 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1466 bridge_destroy(struct bridge *br)
1471 while (br->n_ports > 0) {
1472 port_destroy(br->ports[br->n_ports - 1]);
1474 list_remove(&br->node);
1475 error = dpif_delete(br->dpif);
1476 if (error && error != ENOENT) {
1477 VLOG_ERR("failed to delete %s: %s",
1478 dpif_name(br->dpif), strerror(error));
1480 dpif_close(br->dpif);
1481 ofproto_destroy(br->ofproto);
1482 mac_learning_destroy(br->ml);
1483 hmap_destroy(&br->ifaces);
1484 shash_destroy(&br->port_by_name);
1485 shash_destroy(&br->iface_by_name);
1492 static struct bridge *
1493 bridge_lookup(const char *name)
1497 LIST_FOR_EACH (br, node, &all_bridges) {
1498 if (!strcmp(br->name, name)) {
1505 /* Handle requests for a listing of all flows known by the OpenFlow
1506 * stack, including those normally hidden. */
1508 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1509 const char *args, void *aux OVS_UNUSED)
1514 br = bridge_lookup(args);
1516 unixctl_command_reply(conn, 501, "Unknown bridge");
1521 ofproto_get_all_flows(br->ofproto, &results);
1523 unixctl_command_reply(conn, 200, ds_cstr(&results));
1524 ds_destroy(&results);
1527 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1528 * connections and reconnect. If BRIDGE is not specified, then all bridges
1529 * drop their controller connections and reconnect. */
1531 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1532 const char *args, void *aux OVS_UNUSED)
1535 if (args[0] != '\0') {
1536 br = bridge_lookup(args);
1538 unixctl_command_reply(conn, 501, "Unknown bridge");
1541 ofproto_reconnect_controllers(br->ofproto);
1543 LIST_FOR_EACH (br, node, &all_bridges) {
1544 ofproto_reconnect_controllers(br->ofproto);
1547 unixctl_command_reply(conn, 200, NULL);
1551 bridge_run_one(struct bridge *br)
1554 struct iface *iface;
1556 error = ofproto_run1(br->ofproto);
1561 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1564 error = ofproto_run2(br->ofproto, br->flush);
1567 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1568 struct ofpbuf *packet;
1574 packet = cfm_run(iface->cfm);
1576 iface_send_packet(iface, packet);
1577 ofpbuf_uninit(packet);
1586 bridge_get_controllers(const struct bridge *br,
1587 struct ovsrec_controller ***controllersp)
1589 struct ovsrec_controller **controllers;
1590 size_t n_controllers;
1592 controllers = br->cfg->controller;
1593 n_controllers = br->cfg->n_controller;
1595 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1601 *controllersp = controllers;
1603 return n_controllers;
1607 bridge_reconfigure_one(struct bridge *br)
1609 struct shash old_ports, new_ports;
1610 struct svec snoops, old_snoops;
1611 struct shash_node *node;
1612 enum ofproto_fail_mode fail_mode;
1615 /* Collect old ports. */
1616 shash_init(&old_ports);
1617 for (i = 0; i < br->n_ports; i++) {
1618 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1621 /* Collect new ports. */
1622 shash_init(&new_ports);
1623 for (i = 0; i < br->cfg->n_ports; i++) {
1624 const char *name = br->cfg->ports[i]->name;
1625 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1626 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1631 /* If we have a controller, then we need a local port. Complain if the
1632 * user didn't specify one.
1634 * XXX perhaps we should synthesize a port ourselves in this case. */
1635 if (bridge_get_controllers(br, NULL)) {
1636 char local_name[IF_NAMESIZE];
1639 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1640 local_name, sizeof local_name);
1641 if (!error && !shash_find(&new_ports, local_name)) {
1642 VLOG_WARN("bridge %s: controller specified but no local port "
1643 "(port named %s) defined",
1644 br->name, local_name);
1648 /* Get rid of deleted ports.
1649 * Get rid of deleted interfaces on ports that still exist. */
1650 SHASH_FOR_EACH (node, &old_ports) {
1651 struct port *port = node->data;
1652 const struct ovsrec_port *port_cfg;
1654 port_cfg = shash_find_data(&new_ports, node->name);
1658 port_del_ifaces(port, port_cfg);
1662 /* Create new ports.
1663 * Add new interfaces to existing ports.
1664 * Reconfigure existing ports. */
1665 SHASH_FOR_EACH (node, &new_ports) {
1666 struct port *port = shash_find_data(&old_ports, node->name);
1668 port = port_create(br, node->name);
1671 port_reconfigure(port, node->data);
1672 if (!port->n_ifaces) {
1673 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1674 br->name, port->name);
1678 shash_destroy(&old_ports);
1679 shash_destroy(&new_ports);
1681 /* Set the fail-mode */
1682 fail_mode = !br->cfg->fail_mode
1683 || !strcmp(br->cfg->fail_mode, "standalone")
1684 ? OFPROTO_FAIL_STANDALONE
1685 : OFPROTO_FAIL_SECURE;
1686 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1687 && !ofproto_has_primary_controller(br->ofproto)) {
1688 ofproto_flush_flows(br->ofproto);
1690 ofproto_set_fail_mode(br->ofproto, fail_mode);
1692 /* Delete all flows if we're switching from connected to standalone or vice
1693 * versa. (XXX Should we delete all flows if we are switching from one
1694 * controller to another?) */
1696 /* Configure OpenFlow controller connection snooping. */
1698 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1699 ovs_rundir(), br->name));
1700 svec_init(&old_snoops);
1701 ofproto_get_snoops(br->ofproto, &old_snoops);
1702 if (!svec_equal(&snoops, &old_snoops)) {
1703 ofproto_set_snoops(br->ofproto, &snoops);
1705 svec_destroy(&snoops);
1706 svec_destroy(&old_snoops);
1708 mirror_reconfigure(br);
1711 /* Initializes 'oc' appropriately as a management service controller for
1714 * The caller must free oc->target when it is no longer needed. */
1716 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1717 struct ofproto_controller *oc)
1719 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1720 oc->max_backoff = 0;
1721 oc->probe_interval = 60;
1722 oc->band = OFPROTO_OUT_OF_BAND;
1723 oc->accept_re = NULL;
1724 oc->update_resolv_conf = false;
1726 oc->burst_limit = 0;
1729 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1731 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1732 struct ofproto_controller *oc)
1734 oc->target = c->target;
1735 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1736 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1737 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1738 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1739 oc->accept_re = c->discover_accept_regex;
1740 oc->update_resolv_conf = c->discover_update_resolv_conf;
1741 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1742 oc->burst_limit = (c->controller_burst_limit
1743 ? *c->controller_burst_limit : 0);
1746 /* Configures the IP stack for 'br''s local interface properly according to the
1747 * configuration in 'c'. */
1749 bridge_configure_local_iface_netdev(struct bridge *br,
1750 struct ovsrec_controller *c)
1752 struct netdev *netdev;
1753 struct in_addr mask, gateway;
1755 struct iface *local_iface;
1758 /* Controller discovery does its own TCP/IP configuration later. */
1759 if (strcmp(c->target, "discover")) {
1763 /* If there's no local interface or no IP address, give up. */
1764 local_iface = bridge_get_local_iface(br);
1765 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1769 /* Bring up the local interface. */
1770 netdev = local_iface->netdev;
1771 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1773 /* Configure the IP address and netmask. */
1774 if (!c->local_netmask
1775 || !inet_aton(c->local_netmask, &mask)
1777 mask.s_addr = guess_netmask(ip.s_addr);
1779 if (!netdev_set_in4(netdev, ip, mask)) {
1780 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1781 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1784 /* Configure the default gateway. */
1785 if (c->local_gateway
1786 && inet_aton(c->local_gateway, &gateway)
1787 && gateway.s_addr) {
1788 if (!netdev_add_router(netdev, gateway)) {
1789 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1790 br->name, IP_ARGS(&gateway.s_addr));
1796 bridge_reconfigure_remotes(struct bridge *br,
1797 const struct sockaddr_in *managers,
1800 const char *disable_ib_str, *queue_id_str;
1801 bool disable_in_band = false;
1804 struct ovsrec_controller **controllers;
1805 size_t n_controllers;
1808 struct ofproto_controller *ocs;
1812 /* Check if we should disable in-band control on this bridge. */
1813 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1814 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1815 disable_in_band = true;
1818 /* Set OpenFlow queue ID for in-band control. */
1819 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1820 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1821 ofproto_set_in_band_queue(br->ofproto, queue_id);
1823 if (disable_in_band) {
1824 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1826 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1828 had_primary = ofproto_has_primary_controller(br->ofproto);
1830 n_controllers = bridge_get_controllers(br, &controllers);
1832 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1835 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1836 for (i = 0; i < n_controllers; i++) {
1837 struct ovsrec_controller *c = controllers[i];
1839 if (!strncmp(c->target, "punix:", 6)
1840 || !strncmp(c->target, "unix:", 5)) {
1841 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1843 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1844 * domain sockets and overwriting arbitrary local files. */
1845 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1846 "\"%s\" due to possibility for remote exploit",
1847 dpif_name(br->dpif), c->target);
1851 bridge_configure_local_iface_netdev(br, c);
1852 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1853 if (disable_in_band) {
1854 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1859 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1860 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1863 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1864 ofproto_flush_flows(br->ofproto);
1867 /* If there are no controllers and the bridge is in standalone
1868 * mode, set up a flow that matches every packet and directs
1869 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1870 * switch is in secure mode and we won't pass any traffic until
1871 * a controller has been defined and it tells us to do so. */
1873 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1874 union ofp_action action;
1875 struct cls_rule rule;
1877 memset(&action, 0, sizeof action);
1878 action.type = htons(OFPAT_OUTPUT);
1879 action.output.len = htons(sizeof action);
1880 action.output.port = htons(OFPP_NORMAL);
1881 cls_rule_init_catchall(&rule, 0);
1882 ofproto_add_flow(br->ofproto, &rule, &action, 1);
1887 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1892 for (i = 0; i < br->n_ports; i++) {
1893 struct port *port = br->ports[i];
1894 for (j = 0; j < port->n_ifaces; j++) {
1895 struct iface *iface = port->ifaces[j];
1896 shash_add_once(ifaces, iface->name, iface);
1898 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1899 shash_add_once(ifaces, port->name, NULL);
1904 /* For robustness, in case the administrator moves around datapath ports behind
1905 * our back, we re-check all the datapath port numbers here.
1907 * This function will set the 'dp_ifidx' members of interfaces that have
1908 * disappeared to -1, so only call this function from a context where those
1909 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1910 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1911 * datapath, which doesn't support UINT16_MAX+1 ports. */
1913 bridge_fetch_dp_ifaces(struct bridge *br)
1915 struct odp_port *dpif_ports;
1916 size_t n_dpif_ports;
1919 /* Reset all interface numbers. */
1920 for (i = 0; i < br->n_ports; i++) {
1921 struct port *port = br->ports[i];
1922 for (j = 0; j < port->n_ifaces; j++) {
1923 struct iface *iface = port->ifaces[j];
1924 iface->dp_ifidx = -1;
1927 hmap_clear(&br->ifaces);
1929 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1930 for (i = 0; i < n_dpif_ports; i++) {
1931 struct odp_port *p = &dpif_ports[i];
1932 struct iface *iface = iface_lookup(br, p->devname);
1934 if (iface->dp_ifidx >= 0) {
1935 VLOG_WARN("%s reported interface %s twice",
1936 dpif_name(br->dpif), p->devname);
1937 } else if (iface_from_dp_ifidx(br, p->port)) {
1938 VLOG_WARN("%s reported interface %"PRIu16" twice",
1939 dpif_name(br->dpif), p->port);
1941 iface->dp_ifidx = p->port;
1942 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1943 hash_int(iface->dp_ifidx, 0));
1946 iface_set_ofport(iface->cfg,
1947 (iface->dp_ifidx >= 0
1948 ? odp_port_to_ofp_port(iface->dp_ifidx)
1955 /* Bridge packet processing functions. */
1958 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
1960 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
1963 static struct bond_entry *
1964 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
1967 return &port->bond_hash[bond_hash(mac, vlan)];
1971 bond_choose_iface(const struct port *port)
1973 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1974 size_t i, best_down_slave = -1;
1975 long long next_delay_expiration = LLONG_MAX;
1977 for (i = 0; i < port->n_ifaces; i++) {
1978 struct iface *iface = port->ifaces[i];
1980 if (iface->enabled) {
1982 } else if (iface->delay_expires < next_delay_expiration) {
1983 best_down_slave = i;
1984 next_delay_expiration = iface->delay_expires;
1988 if (best_down_slave != -1) {
1989 struct iface *iface = port->ifaces[best_down_slave];
1991 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1992 "since no other interface is up", iface->name,
1993 iface->delay_expires - time_msec());
1994 bond_enable_slave(iface, true);
1997 return best_down_slave;
2001 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2002 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2004 struct iface *iface;
2006 assert(port->n_ifaces);
2007 if (port->n_ifaces == 1) {
2008 iface = port->ifaces[0];
2010 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2011 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2012 || !port->ifaces[e->iface_idx]->enabled) {
2013 /* XXX select interface properly. The current interface selection
2014 * is only good for testing the rebalancing code. */
2015 e->iface_idx = bond_choose_iface(port);
2016 if (e->iface_idx < 0) {
2017 *tags |= port->no_ifaces_tag;
2020 e->iface_tag = tag_create_random();
2021 ((struct port *) port)->bond_compat_is_stale = true;
2023 *tags |= e->iface_tag;
2024 iface = port->ifaces[e->iface_idx];
2026 *dp_ifidx = iface->dp_ifidx;
2027 *tags |= iface->tag; /* Currently only used for bonding. */
2032 bond_link_status_update(struct iface *iface, bool carrier)
2034 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2035 struct port *port = iface->port;
2037 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2038 /* Nothing to do. */
2041 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
2042 iface->name, carrier ? "detected" : "dropped");
2043 if (carrier == iface->enabled) {
2044 iface->delay_expires = LLONG_MAX;
2045 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2046 iface->name, carrier ? "disabled" : "enabled");
2047 } else if (carrier && port->active_iface < 0) {
2048 bond_enable_slave(iface, true);
2049 if (port->updelay) {
2050 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2051 "other interface is up", iface->name, port->updelay);
2054 int delay = carrier ? port->updelay : port->downdelay;
2055 iface->delay_expires = time_msec() + delay;
2058 "interface %s: will be %s if it stays %s for %d ms",
2060 carrier ? "enabled" : "disabled",
2061 carrier ? "up" : "down",
2068 bond_choose_active_iface(struct port *port)
2070 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2072 port->active_iface = bond_choose_iface(port);
2073 port->active_iface_tag = tag_create_random();
2074 if (port->active_iface >= 0) {
2075 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2076 port->name, port->ifaces[port->active_iface]->name);
2078 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2084 bond_enable_slave(struct iface *iface, bool enable)
2086 struct port *port = iface->port;
2087 struct bridge *br = port->bridge;
2089 /* This acts as a recursion check. If the act of disabling a slave
2090 * causes a different slave to be enabled, the flag will allow us to
2091 * skip redundant work when we reenter this function. It must be
2092 * cleared on exit to keep things safe with multiple bonds. */
2093 static bool moving_active_iface = false;
2095 iface->delay_expires = LLONG_MAX;
2096 if (enable == iface->enabled) {
2100 iface->enabled = enable;
2101 if (!iface->enabled) {
2102 VLOG_WARN("interface %s: disabled", iface->name);
2103 ofproto_revalidate(br->ofproto, iface->tag);
2104 if (iface->port_ifidx == port->active_iface) {
2105 ofproto_revalidate(br->ofproto,
2106 port->active_iface_tag);
2108 /* Disabling a slave can lead to another slave being immediately
2109 * enabled if there will be no active slaves but one is waiting
2110 * on an updelay. In this case we do not need to run most of the
2111 * code for the newly enabled slave since there was no period
2112 * without an active slave and it is redundant with the disabling
2114 moving_active_iface = true;
2115 bond_choose_active_iface(port);
2117 bond_send_learning_packets(port);
2119 VLOG_WARN("interface %s: enabled", iface->name);
2120 if (port->active_iface < 0 && !moving_active_iface) {
2121 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2122 bond_choose_active_iface(port);
2123 bond_send_learning_packets(port);
2125 iface->tag = tag_create_random();
2128 moving_active_iface = false;
2129 port->bond_compat_is_stale = true;
2132 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2133 * bond interface. */
2135 bond_update_fake_iface_stats(struct port *port)
2137 struct netdev_stats bond_stats;
2138 struct netdev *bond_dev;
2141 memset(&bond_stats, 0, sizeof bond_stats);
2143 for (i = 0; i < port->n_ifaces; i++) {
2144 struct netdev_stats slave_stats;
2146 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2147 /* XXX: We swap the stats here because they are swapped back when
2148 * reported by the internal device. The reason for this is
2149 * internal devices normally represent packets going into the system
2150 * but when used as fake bond device they represent packets leaving
2151 * the system. We really should do this in the internal device
2152 * itself because changing it here reverses the counts from the
2153 * perspective of the switch. However, the internal device doesn't
2154 * know what type of device it represents so we have to do it here
2156 bond_stats.tx_packets += slave_stats.rx_packets;
2157 bond_stats.tx_bytes += slave_stats.rx_bytes;
2158 bond_stats.rx_packets += slave_stats.tx_packets;
2159 bond_stats.rx_bytes += slave_stats.tx_bytes;
2163 if (!netdev_open_default(port->name, &bond_dev)) {
2164 netdev_set_stats(bond_dev, &bond_stats);
2165 netdev_close(bond_dev);
2170 bond_run(struct bridge *br)
2174 for (i = 0; i < br->n_ports; i++) {
2175 struct port *port = br->ports[i];
2177 if (port->n_ifaces >= 2) {
2180 /* Track carrier going up and down on interfaces. */
2181 while (!netdev_monitor_poll(port->monitor, &devname)) {
2182 struct iface *iface;
2184 iface = port_lookup_iface(port, devname);
2186 bool carrier = netdev_get_carrier(iface->netdev);
2188 bond_link_status_update(iface, carrier);
2189 port_update_bond_compat(port);
2194 for (j = 0; j < port->n_ifaces; j++) {
2195 struct iface *iface = port->ifaces[j];
2196 if (time_msec() >= iface->delay_expires) {
2197 bond_enable_slave(iface, !iface->enabled);
2201 if (port->bond_fake_iface
2202 && time_msec() >= port->bond_next_fake_iface_update) {
2203 bond_update_fake_iface_stats(port);
2204 port->bond_next_fake_iface_update = time_msec() + 1000;
2208 if (port->bond_compat_is_stale) {
2209 port->bond_compat_is_stale = false;
2210 port_update_bond_compat(port);
2216 bond_wait(struct bridge *br)
2220 for (i = 0; i < br->n_ports; i++) {
2221 struct port *port = br->ports[i];
2222 if (port->n_ifaces < 2) {
2225 netdev_monitor_poll_wait(port->monitor);
2226 for (j = 0; j < port->n_ifaces; j++) {
2227 struct iface *iface = port->ifaces[j];
2228 if (iface->delay_expires != LLONG_MAX) {
2229 poll_timer_wait_until(iface->delay_expires);
2232 if (port->bond_fake_iface) {
2233 poll_timer_wait_until(port->bond_next_fake_iface_update);
2239 set_dst(struct dst *dst, const struct flow *flow,
2240 const struct port *in_port, const struct port *out_port,
2243 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2244 : in_port->vlan >= 0 ? in_port->vlan
2245 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2246 : vlan_tci_to_vid(flow->vlan_tci));
2247 return choose_output_iface(out_port, flow->dl_src, dst->vlan,
2248 &dst->dp_ifidx, tags);
2252 swap_dst(struct dst *p, struct dst *q)
2254 struct dst tmp = *p;
2259 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2260 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2261 * that we push to the datapath. We could in fact fully sort the array by
2262 * vlan, but in most cases there are at most two different vlan tags so that's
2263 * possibly overkill.) */
2265 partition_dsts(struct dst_set *set, int vlan)
2267 struct dst *first = set->dsts;
2268 struct dst *last = set->dsts + set->n;
2270 while (first != last) {
2272 * - All dsts < first have vlan == 'vlan'.
2273 * - All dsts >= last have vlan != 'vlan'.
2274 * - first < last. */
2275 while (first->vlan == vlan) {
2276 if (++first == last) {
2281 /* Same invariants, plus one additional:
2282 * - first->vlan != vlan.
2284 while (last[-1].vlan != vlan) {
2285 if (--last == first) {
2290 /* Same invariants, plus one additional:
2291 * - last[-1].vlan == vlan.*/
2292 swap_dst(first++, --last);
2297 mirror_mask_ffs(mirror_mask_t mask)
2299 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2304 dst_set_init(struct dst_set *set)
2306 set->dsts = set->builtin;
2308 set->allocated = ARRAY_SIZE(set->builtin);
2312 dst_set_add(struct dst_set *set, const struct dst *dst)
2314 if (set->n >= set->allocated) {
2315 size_t new_allocated;
2316 struct dst *new_dsts;
2318 new_allocated = set->allocated * 2;
2319 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2320 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2324 set->dsts = new_dsts;
2325 set->allocated = new_allocated;
2327 set->dsts[set->n++] = *dst;
2331 dst_set_free(struct dst_set *set)
2333 if (set->dsts != set->builtin) {
2339 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2342 for (i = 0; i < set->n; i++) {
2343 if (set->dsts[i].vlan == test->vlan
2344 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2352 port_trunks_vlan(const struct port *port, uint16_t vlan)
2354 return (port->vlan < 0
2355 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2359 port_includes_vlan(const struct port *port, uint16_t vlan)
2361 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2365 port_is_floodable(const struct port *port)
2369 for (i = 0; i < port->n_ifaces; i++) {
2370 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2371 port->ifaces[i]->dp_ifidx)) {
2379 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2380 const struct port *in_port, const struct port *out_port,
2381 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2383 mirror_mask_t mirrors = in_port->src_mirrors;
2388 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2389 if (flow_vlan == 0) {
2390 flow_vlan = OFP_VLAN_NONE;
2393 if (out_port == FLOOD_PORT) {
2394 for (i = 0; i < br->n_ports; i++) {
2395 struct port *port = br->ports[i];
2397 && port_is_floodable(port)
2398 && port_includes_vlan(port, vlan)
2399 && !port->is_mirror_output_port
2400 && set_dst(&dst, flow, in_port, port, tags)) {
2401 mirrors |= port->dst_mirrors;
2402 dst_set_add(set, &dst);
2405 *nf_output_iface = NF_OUT_FLOOD;
2406 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2407 dst_set_add(set, &dst);
2408 *nf_output_iface = dst.dp_ifidx;
2409 mirrors |= out_port->dst_mirrors;
2413 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2414 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2416 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2417 && !dst_is_duplicate(set, &dst)) {
2418 dst_set_add(set, &dst);
2421 for (i = 0; i < br->n_ports; i++) {
2422 struct port *port = br->ports[i];
2423 if (port_includes_vlan(port, m->out_vlan)
2424 && set_dst(&dst, flow, in_port, port, tags))
2426 if (port->vlan < 0) {
2427 dst.vlan = m->out_vlan;
2429 if (dst_is_duplicate(set, &dst)) {
2433 /* Use the vlan tag on the original flow instead of
2434 * the one passed in the vlan parameter. This ensures
2435 * that we compare the vlan from before any implicit
2436 * tagging tags place. This is necessary because
2437 * dst->vlan is the final vlan, after removing implicit
2439 if (port == in_port && dst.vlan == flow_vlan) {
2440 /* Don't send out input port on same VLAN. */
2443 dst_set_add(set, &dst);
2448 mirrors &= mirrors - 1;
2451 partition_dsts(set, flow_vlan);
2454 static void OVS_UNUSED
2455 print_dsts(const struct dst_set *set)
2459 for (i = 0; i < set->n; i++) {
2460 const struct dst *dst = &set->dsts[i];
2462 printf(">p%"PRIu16, dst->dp_ifidx);
2463 if (dst->vlan != OFP_VLAN_NONE) {
2464 printf("v%"PRIu16, dst->vlan);
2470 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2471 const struct port *in_port, const struct port *out_port,
2472 tag_type *tags, struct ofpbuf *actions,
2473 uint16_t *nf_output_iface)
2480 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2483 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2484 if (cur_vlan == 0) {
2485 cur_vlan = OFP_VLAN_NONE;
2487 for (i = 0; i < set.n; i++) {
2488 const struct dst *dst = &set.dsts[i];
2489 if (dst->vlan != cur_vlan) {
2490 if (dst->vlan == OFP_VLAN_NONE) {
2491 nl_msg_put_flag(actions, ODPAT_STRIP_VLAN);
2494 tci = htons(dst->vlan & VLAN_VID_MASK);
2495 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2496 nl_msg_put_be16(actions, ODPAT_SET_DL_TCI, tci);
2498 cur_vlan = dst->vlan;
2500 nl_msg_put_u32(actions, ODPAT_OUTPUT, dst->dp_ifidx);
2505 /* Returns the effective vlan of a packet, taking into account both the
2506 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2507 * the packet is untagged and -1 indicates it has an invalid header and
2508 * should be dropped. */
2509 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2510 struct port *in_port, bool have_packet)
2512 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2513 if (in_port->vlan >= 0) {
2515 /* XXX support double tagging? */
2517 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2518 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2519 "packet received on port %s configured with "
2520 "implicit VLAN %"PRIu16,
2521 br->name, vlan, in_port->name, in_port->vlan);
2525 vlan = in_port->vlan;
2527 if (!port_includes_vlan(in_port, vlan)) {
2529 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2530 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2531 "packet received on port %s not configured for "
2533 br->name, vlan, in_port->name, vlan);
2542 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2543 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2544 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2546 is_gratuitous_arp(const struct flow *flow)
2548 return (flow->dl_type == htons(ETH_TYPE_ARP)
2549 && eth_addr_is_broadcast(flow->dl_dst)
2550 && (flow->nw_proto == ARP_OP_REPLY
2551 || (flow->nw_proto == ARP_OP_REQUEST
2552 && flow->nw_src == flow->nw_dst)));
2556 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2557 struct port *in_port)
2559 enum grat_arp_lock_type lock_type;
2562 /* We don't want to learn from gratuitous ARP packets that are reflected
2563 * back over bond slaves so we lock the learning table. */
2564 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2565 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2566 GRAT_ARP_LOCK_CHECK;
2568 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2571 /* The log messages here could actually be useful in debugging,
2572 * so keep the rate limit relatively high. */
2573 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2575 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2576 "on port %s in VLAN %d",
2577 br->name, ETH_ADDR_ARGS(flow->dl_src),
2578 in_port->name, vlan);
2579 ofproto_revalidate(br->ofproto, rev_tag);
2583 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2584 * dropped. Returns true if they may be forwarded, false if they should be
2587 * If 'have_packet' is true, it indicates that the caller is processing a
2588 * received packet. If 'have_packet' is false, then the caller is just
2589 * revalidating an existing flow because configuration has changed. Either
2590 * way, 'have_packet' only affects logging (there is no point in logging errors
2591 * during revalidation).
2593 * Sets '*in_portp' to the input port. This will be a null pointer if
2594 * flow->in_port does not designate a known input port (in which case
2595 * is_admissible() returns false).
2597 * When returning true, sets '*vlanp' to the effective VLAN of the input
2598 * packet, as returned by flow_get_vlan().
2600 * May also add tags to '*tags', although the current implementation only does
2601 * so in one special case.
2604 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2605 tag_type *tags, int *vlanp, struct port **in_portp)
2607 struct iface *in_iface;
2608 struct port *in_port;
2611 /* Find the interface and port structure for the received packet. */
2612 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2614 /* No interface? Something fishy... */
2616 /* Odd. A few possible reasons here:
2618 * - We deleted an interface but there are still a few packets
2619 * queued up from it.
2621 * - Someone externally added an interface (e.g. with "ovs-dpctl
2622 * add-if") that we don't know about.
2624 * - Packet arrived on the local port but the local port is not
2625 * one of our bridge ports.
2627 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2629 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2630 "interface %"PRIu16, br->name, flow->in_port);
2636 *in_portp = in_port = in_iface->port;
2637 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2642 /* Drop frames for reserved multicast addresses. */
2643 if (eth_addr_is_reserved(flow->dl_dst)) {
2647 /* Drop frames on ports reserved for mirroring. */
2648 if (in_port->is_mirror_output_port) {
2650 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2651 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2652 "%s, which is reserved exclusively for mirroring",
2653 br->name, in_port->name);
2658 /* Packets received on bonds need special attention to avoid duplicates. */
2659 if (in_port->n_ifaces > 1) {
2661 bool is_grat_arp_locked;
2663 if (eth_addr_is_multicast(flow->dl_dst)) {
2664 *tags |= in_port->active_iface_tag;
2665 if (in_port->active_iface != in_iface->port_ifidx) {
2666 /* Drop all multicast packets on inactive slaves. */
2671 /* Drop all packets for which we have learned a different input
2672 * port, because we probably sent the packet on one slave and got
2673 * it back on the other. Gratuitous ARP packets are an exception
2674 * to this rule: the host has moved to another switch. The exception
2675 * to the exception is if we locked the learning table to avoid
2676 * reflections on bond slaves. If this is the case, just drop the
2678 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2679 &is_grat_arp_locked);
2680 if (src_idx != -1 && src_idx != in_port->port_idx &&
2681 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2689 /* If the composed actions may be applied to any packet in the given 'flow',
2690 * returns true. Otherwise, the actions should only be applied to 'packet', or
2691 * not at all, if 'packet' was NULL. */
2693 process_flow(struct bridge *br, const struct flow *flow,
2694 const struct ofpbuf *packet, struct ofpbuf *actions,
2695 tag_type *tags, uint16_t *nf_output_iface)
2697 struct port *in_port;
2698 struct port *out_port;
2702 /* Check whether we should drop packets in this flow. */
2703 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2708 /* Learn source MAC (but don't try to learn from revalidation). */
2710 update_learning_table(br, flow, vlan, in_port);
2713 /* Determine output port. */
2714 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2716 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2717 out_port = br->ports[out_port_idx];
2718 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2719 /* If we are revalidating but don't have a learning entry then
2720 * eject the flow. Installing a flow that floods packets opens
2721 * up a window of time where we could learn from a packet reflected
2722 * on a bond and blackhole packets before the learning table is
2723 * updated to reflect the correct port. */
2726 out_port = FLOOD_PORT;
2729 /* Don't send packets out their input ports. */
2730 if (in_port == out_port) {
2736 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2744 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2745 struct ofpbuf *actions, tag_type *tags,
2746 uint16_t *nf_output_iface, void *br_)
2748 struct iface *iface;
2749 struct bridge *br = br_;
2751 COVERAGE_INC(bridge_process_flow);
2753 iface = iface_from_dp_ifidx(br, flow->in_port);
2755 if (cfm_should_process_flow(flow)) {
2756 if (packet && iface->cfm) {
2757 cfm_process_heartbeat(iface->cfm, packet);
2762 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2766 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2767 const struct nlattr *actions,
2769 unsigned long long int n_bytes, void *br_)
2771 struct bridge *br = br_;
2772 const struct nlattr *a;
2773 struct port *in_port;
2778 /* Feed information from the active flows back into the learning table to
2779 * ensure that table is always in sync with what is actually flowing
2780 * through the datapath.
2782 * We test that 'tags' is nonzero to ensure that only flows that include an
2783 * OFPP_NORMAL action are used for learning. This works because
2784 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2785 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2786 update_learning_table(br, flow, vlan, in_port);
2789 /* Account for bond slave utilization. */
2790 if (!br->has_bonded_ports) {
2793 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2794 if (nl_attr_type(a) == ODPAT_OUTPUT) {
2795 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2796 if (out_port && out_port->n_ifaces >= 2) {
2797 uint16_t vlan = (flow->vlan_tci
2798 ? vlan_tci_to_vid(flow->vlan_tci)
2800 struct bond_entry *e = lookup_bond_entry(out_port,
2801 flow->dl_src, vlan);
2802 e->tx_bytes += n_bytes;
2809 bridge_account_checkpoint_ofhook_cb(void *br_)
2811 struct bridge *br = br_;
2815 if (!br->has_bonded_ports) {
2820 for (i = 0; i < br->n_ports; i++) {
2821 struct port *port = br->ports[i];
2822 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2823 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2824 bond_rebalance_port(port);
2829 static struct ofhooks bridge_ofhooks = {
2830 bridge_normal_ofhook_cb,
2831 bridge_account_flow_ofhook_cb,
2832 bridge_account_checkpoint_ofhook_cb,
2835 /* Bonding functions. */
2837 /* Statistics for a single interface on a bonded port, used for load-based
2838 * bond rebalancing. */
2839 struct slave_balance {
2840 struct iface *iface; /* The interface. */
2841 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2843 /* All the "bond_entry"s that are assigned to this interface, in order of
2844 * increasing tx_bytes. */
2845 struct bond_entry **hashes;
2849 /* Sorts pointers to pointers to bond_entries in ascending order by the
2850 * interface to which they are assigned, and within a single interface in
2851 * ascending order of bytes transmitted. */
2853 compare_bond_entries(const void *a_, const void *b_)
2855 const struct bond_entry *const *ap = a_;
2856 const struct bond_entry *const *bp = b_;
2857 const struct bond_entry *a = *ap;
2858 const struct bond_entry *b = *bp;
2859 if (a->iface_idx != b->iface_idx) {
2860 return a->iface_idx > b->iface_idx ? 1 : -1;
2861 } else if (a->tx_bytes != b->tx_bytes) {
2862 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2868 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2869 * *descending* order by number of bytes transmitted. */
2871 compare_slave_balance(const void *a_, const void *b_)
2873 const struct slave_balance *a = a_;
2874 const struct slave_balance *b = b_;
2875 if (a->iface->enabled != b->iface->enabled) {
2876 return a->iface->enabled ? -1 : 1;
2877 } else if (a->tx_bytes != b->tx_bytes) {
2878 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2885 swap_bals(struct slave_balance *a, struct slave_balance *b)
2887 struct slave_balance tmp = *a;
2892 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2893 * given that 'p' (and only 'p') might be in the wrong location.
2895 * This function invalidates 'p', since it might now be in a different memory
2898 resort_bals(struct slave_balance *p,
2899 struct slave_balance bals[], size_t n_bals)
2902 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2903 swap_bals(p, p - 1);
2905 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2906 swap_bals(p, p + 1);
2912 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2914 if (VLOG_IS_DBG_ENABLED()) {
2915 struct ds ds = DS_EMPTY_INITIALIZER;
2916 const struct slave_balance *b;
2918 for (b = bals; b < bals + n_bals; b++) {
2922 ds_put_char(&ds, ',');
2924 ds_put_format(&ds, " %s %"PRIu64"kB",
2925 b->iface->name, b->tx_bytes / 1024);
2927 if (!b->iface->enabled) {
2928 ds_put_cstr(&ds, " (disabled)");
2930 if (b->n_hashes > 0) {
2931 ds_put_cstr(&ds, " (");
2932 for (i = 0; i < b->n_hashes; i++) {
2933 const struct bond_entry *e = b->hashes[i];
2935 ds_put_cstr(&ds, " + ");
2937 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2938 e - port->bond_hash, e->tx_bytes / 1024);
2940 ds_put_cstr(&ds, ")");
2943 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2948 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2950 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2953 struct bond_entry *hash = from->hashes[hash_idx];
2954 struct port *port = from->iface->port;
2955 uint64_t delta = hash->tx_bytes;
2957 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2958 "from %s to %s (now carrying %"PRIu64"kB and "
2959 "%"PRIu64"kB load, respectively)",
2960 port->name, delta / 1024, hash - port->bond_hash,
2961 from->iface->name, to->iface->name,
2962 (from->tx_bytes - delta) / 1024,
2963 (to->tx_bytes + delta) / 1024);
2965 /* Delete element from from->hashes.
2967 * We don't bother to add the element to to->hashes because not only would
2968 * it require more work, the only purpose it would be to allow that hash to
2969 * be migrated to another slave in this rebalancing run, and there is no
2970 * point in doing that. */
2971 if (hash_idx == 0) {
2974 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2975 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2979 /* Shift load away from 'from' to 'to'. */
2980 from->tx_bytes -= delta;
2981 to->tx_bytes += delta;
2983 /* Arrange for flows to be revalidated. */
2984 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2985 hash->iface_idx = to->iface->port_ifidx;
2986 hash->iface_tag = tag_create_random();
2990 bond_rebalance_port(struct port *port)
2992 struct slave_balance *bals;
2994 struct bond_entry *hashes[BOND_MASK + 1];
2995 struct slave_balance *b, *from, *to;
2996 struct bond_entry *e;
2999 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3000 * descending order of tx_bytes, so that bals[0] represents the most
3001 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3004 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3005 * array for each slave_balance structure, we sort our local array of
3006 * hashes in order by slave, so that all of the hashes for a given slave
3007 * become contiguous in memory, and then we point each 'hashes' members of
3008 * a slave_balance structure to the start of a contiguous group. */
3009 n_bals = port->n_ifaces;
3010 bals = xmalloc(n_bals * sizeof *bals);
3011 for (b = bals; b < &bals[n_bals]; b++) {
3012 b->iface = port->ifaces[b - bals];
3017 for (i = 0; i <= BOND_MASK; i++) {
3018 hashes[i] = &port->bond_hash[i];
3020 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3021 for (i = 0; i <= BOND_MASK; i++) {
3023 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3024 b = &bals[e->iface_idx];
3025 b->tx_bytes += e->tx_bytes;
3027 b->hashes = &hashes[i];
3032 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3033 log_bals(bals, n_bals, port);
3035 /* Discard slaves that aren't enabled (which were sorted to the back of the
3036 * array earlier). */
3037 while (!bals[n_bals - 1].iface->enabled) {
3044 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3045 to = &bals[n_bals - 1];
3046 for (from = bals; from < to; ) {
3047 uint64_t overload = from->tx_bytes - to->tx_bytes;
3048 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3049 /* The extra load on 'from' (and all less-loaded slaves), compared
3050 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3051 * it is less than ~1Mbps. No point in rebalancing. */
3053 } else if (from->n_hashes == 1) {
3054 /* 'from' only carries a single MAC hash, so we can't shift any
3055 * load away from it, even though we want to. */
3058 /* 'from' is carrying significantly more load than 'to', and that
3059 * load is split across at least two different hashes. Pick a hash
3060 * to migrate to 'to' (the least-loaded slave), given that doing so
3061 * must decrease the ratio of the load on the two slaves by at
3064 * The sort order we use means that we prefer to shift away the
3065 * smallest hashes instead of the biggest ones. There is little
3066 * reason behind this decision; we could use the opposite sort
3067 * order to shift away big hashes ahead of small ones. */
3070 for (i = 0; i < from->n_hashes; i++) {
3071 double old_ratio, new_ratio;
3072 uint64_t delta = from->hashes[i]->tx_bytes;
3074 if (delta == 0 || from->tx_bytes - delta == 0) {
3075 /* Pointless move. */
3079 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3081 if (to->tx_bytes == 0) {
3082 /* Nothing on the new slave, move it. */
3086 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3087 new_ratio = (double)(from->tx_bytes - delta) /
3088 (to->tx_bytes + delta);
3090 if (new_ratio == 0) {
3091 /* Should already be covered but check to prevent division
3096 if (new_ratio < 1) {
3097 new_ratio = 1 / new_ratio;
3100 if (old_ratio - new_ratio > 0.1) {
3101 /* Would decrease the ratio, move it. */
3105 if (i < from->n_hashes) {
3106 bond_shift_load(from, to, i);
3107 port->bond_compat_is_stale = true;
3109 /* If the result of the migration changed the relative order of
3110 * 'from' and 'to' swap them back to maintain invariants. */
3111 if (order_swapped) {
3112 swap_bals(from, to);
3115 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3116 * point to different slave_balance structures. It is only
3117 * valid to do these two operations in a row at all because we
3118 * know that 'from' will not move past 'to' and vice versa. */
3119 resort_bals(from, bals, n_bals);
3120 resort_bals(to, bals, n_bals);
3127 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3128 * historical data to decay to <1% in 7 rebalancing runs. */
3129 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3138 bond_send_learning_packets(struct port *port)
3140 struct bridge *br = port->bridge;
3141 struct mac_entry *e;
3142 struct ofpbuf packet;
3143 int error, n_packets, n_errors;
3145 if (!port->n_ifaces || port->active_iface < 0) {
3149 ofpbuf_init(&packet, 128);
3150 error = n_packets = n_errors = 0;
3151 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3152 union ofp_action actions[2], *a;
3158 if (e->port == port->port_idx
3159 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3163 /* Compose actions. */
3164 memset(actions, 0, sizeof actions);
3167 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3168 a->vlan_vid.len = htons(sizeof *a);
3169 a->vlan_vid.vlan_vid = htons(e->vlan);
3172 a->output.type = htons(OFPAT_OUTPUT);
3173 a->output.len = htons(sizeof *a);
3174 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3179 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3181 flow_extract(&packet, 0, ODPP_NONE, &flow);
3182 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3189 ofpbuf_uninit(&packet);
3192 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3193 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3194 "packets, last error was: %s",
3195 port->name, n_errors, n_packets, strerror(error));
3197 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3198 port->name, n_packets);
3202 /* Bonding unixctl user interface functions. */
3205 bond_unixctl_list(struct unixctl_conn *conn,
3206 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3208 struct ds ds = DS_EMPTY_INITIALIZER;
3209 const struct bridge *br;
3211 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
3213 LIST_FOR_EACH (br, node, &all_bridges) {
3216 for (i = 0; i < br->n_ports; i++) {
3217 const struct port *port = br->ports[i];
3218 if (port->n_ifaces > 1) {
3221 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3222 for (j = 0; j < port->n_ifaces; j++) {
3223 const struct iface *iface = port->ifaces[j];
3225 ds_put_cstr(&ds, ", ");
3227 ds_put_cstr(&ds, iface->name);
3229 ds_put_char(&ds, '\n');
3233 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3237 static struct port *
3238 bond_find(const char *name)
3240 const struct bridge *br;
3242 LIST_FOR_EACH (br, node, &all_bridges) {
3245 for (i = 0; i < br->n_ports; i++) {
3246 struct port *port = br->ports[i];
3247 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3256 bond_unixctl_show(struct unixctl_conn *conn,
3257 const char *args, void *aux OVS_UNUSED)
3259 struct ds ds = DS_EMPTY_INITIALIZER;
3260 const struct port *port;
3263 port = bond_find(args);
3265 unixctl_command_reply(conn, 501, "no such bond");
3269 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3270 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3271 ds_put_format(&ds, "next rebalance: %lld ms\n",
3272 port->bond_next_rebalance - time_msec());
3273 for (j = 0; j < port->n_ifaces; j++) {
3274 const struct iface *iface = port->ifaces[j];
3275 struct bond_entry *be;
3278 ds_put_format(&ds, "slave %s: %s\n",
3279 iface->name, iface->enabled ? "enabled" : "disabled");
3280 if (j == port->active_iface) {
3281 ds_put_cstr(&ds, "\tactive slave\n");
3283 if (iface->delay_expires != LLONG_MAX) {
3284 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3285 iface->enabled ? "downdelay" : "updelay",
3286 iface->delay_expires - time_msec());
3290 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3291 int hash = be - port->bond_hash;
3292 struct mac_entry *me;
3294 if (be->iface_idx != j) {
3298 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3299 hash, be->tx_bytes / 1024);
3302 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3305 if (bond_hash(me->mac, me->vlan) == hash
3306 && me->port != port->port_idx
3307 && choose_output_iface(port, me->mac, me->vlan,
3309 && dp_ifidx == iface->dp_ifidx)
3311 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3312 ETH_ADDR_ARGS(me->mac));
3317 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3322 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3323 void *aux OVS_UNUSED)
3325 char *args = (char *) args_;
3326 char *save_ptr = NULL;
3327 char *bond_s, *hash_s, *slave_s;
3329 struct iface *iface;
3330 struct bond_entry *entry;
3333 bond_s = strtok_r(args, " ", &save_ptr);
3334 hash_s = strtok_r(NULL, " ", &save_ptr);
3335 slave_s = strtok_r(NULL, " ", &save_ptr);
3337 unixctl_command_reply(conn, 501,
3338 "usage: bond/migrate BOND HASH SLAVE");
3342 port = bond_find(bond_s);
3344 unixctl_command_reply(conn, 501, "no such bond");
3348 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3349 hash = atoi(hash_s) & BOND_MASK;
3351 unixctl_command_reply(conn, 501, "bad hash");
3355 iface = port_lookup_iface(port, slave_s);
3357 unixctl_command_reply(conn, 501, "no such slave");
3361 if (!iface->enabled) {
3362 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3366 entry = &port->bond_hash[hash];
3367 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3368 entry->iface_idx = iface->port_ifidx;
3369 entry->iface_tag = tag_create_random();
3370 port->bond_compat_is_stale = true;
3371 unixctl_command_reply(conn, 200, "migrated");
3375 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3376 void *aux OVS_UNUSED)
3378 char *args = (char *) args_;
3379 char *save_ptr = NULL;
3380 char *bond_s, *slave_s;
3382 struct iface *iface;
3384 bond_s = strtok_r(args, " ", &save_ptr);
3385 slave_s = strtok_r(NULL, " ", &save_ptr);
3387 unixctl_command_reply(conn, 501,
3388 "usage: bond/set-active-slave BOND SLAVE");
3392 port = bond_find(bond_s);
3394 unixctl_command_reply(conn, 501, "no such bond");
3398 iface = port_lookup_iface(port, slave_s);
3400 unixctl_command_reply(conn, 501, "no such slave");
3404 if (!iface->enabled) {
3405 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3409 if (port->active_iface != iface->port_ifidx) {
3410 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3411 port->active_iface = iface->port_ifidx;
3412 port->active_iface_tag = tag_create_random();
3413 VLOG_INFO("port %s: active interface is now %s",
3414 port->name, iface->name);
3415 bond_send_learning_packets(port);
3416 unixctl_command_reply(conn, 200, "done");
3418 unixctl_command_reply(conn, 200, "no change");
3423 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3425 char *args = (char *) args_;
3426 char *save_ptr = NULL;
3427 char *bond_s, *slave_s;
3429 struct iface *iface;
3431 bond_s = strtok_r(args, " ", &save_ptr);
3432 slave_s = strtok_r(NULL, " ", &save_ptr);
3434 unixctl_command_reply(conn, 501,
3435 "usage: bond/enable/disable-slave BOND SLAVE");
3439 port = bond_find(bond_s);
3441 unixctl_command_reply(conn, 501, "no such bond");
3445 iface = port_lookup_iface(port, slave_s);
3447 unixctl_command_reply(conn, 501, "no such slave");
3451 bond_enable_slave(iface, enable);
3452 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3456 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3457 void *aux OVS_UNUSED)
3459 enable_slave(conn, args, true);
3463 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3464 void *aux OVS_UNUSED)
3466 enable_slave(conn, args, false);
3470 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3471 void *aux OVS_UNUSED)
3473 char *args = (char *) args_;
3474 uint8_t mac[ETH_ADDR_LEN];
3478 char *mac_s, *vlan_s;
3479 char *save_ptr = NULL;
3481 mac_s = strtok_r(args, " ", &save_ptr);
3482 vlan_s = strtok_r(NULL, " ", &save_ptr);
3485 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3486 unixctl_command_reply(conn, 501, "invalid vlan");
3490 vlan = OFP_VLAN_NONE;
3493 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3494 == ETH_ADDR_SCAN_COUNT) {
3495 hash = bond_hash(mac, vlan);
3497 hash_cstr = xasprintf("%u", hash);
3498 unixctl_command_reply(conn, 200, hash_cstr);
3501 unixctl_command_reply(conn, 501, "invalid mac");
3508 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3509 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3510 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3511 unixctl_command_register("bond/set-active-slave",
3512 bond_unixctl_set_active_slave, NULL);
3513 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3515 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3517 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3520 /* Port functions. */
3522 static struct port *
3523 port_create(struct bridge *br, const char *name)
3527 port = xzalloc(sizeof *port);
3529 port->port_idx = br->n_ports;
3531 port->trunks = NULL;
3532 port->name = xstrdup(name);
3533 port->active_iface = -1;
3535 if (br->n_ports >= br->allocated_ports) {
3536 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3539 br->ports[br->n_ports++] = port;
3540 shash_add_assert(&br->port_by_name, port->name, port);
3542 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3549 get_port_other_config(const struct ovsrec_port *port, const char *key,
3550 const char *default_value)
3554 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3556 return value ? value : default_value;
3560 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3562 struct shash new_ifaces;
3565 /* Collect list of new interfaces. */
3566 shash_init(&new_ifaces);
3567 for (i = 0; i < cfg->n_interfaces; i++) {
3568 const char *name = cfg->interfaces[i]->name;
3569 shash_add_once(&new_ifaces, name, NULL);
3572 /* Get rid of deleted interfaces. */
3573 for (i = 0; i < port->n_ifaces; ) {
3574 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3575 iface_destroy(port->ifaces[i]);
3581 shash_destroy(&new_ifaces);
3585 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3587 struct shash new_ifaces;
3588 long long int next_rebalance;
3589 unsigned long *trunks;
3595 /* Update settings. */
3596 port->updelay = cfg->bond_updelay;
3597 if (port->updelay < 0) {
3600 port->downdelay = cfg->bond_downdelay;
3601 if (port->downdelay < 0) {
3602 port->downdelay = 0;
3604 port->bond_rebalance_interval = atoi(
3605 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3606 if (port->bond_rebalance_interval < 1000) {
3607 port->bond_rebalance_interval = 1000;
3609 next_rebalance = time_msec() + port->bond_rebalance_interval;
3610 if (port->bond_next_rebalance > next_rebalance) {
3611 port->bond_next_rebalance = next_rebalance;
3614 /* Add new interfaces and update 'cfg' member of existing ones. */
3615 shash_init(&new_ifaces);
3616 for (i = 0; i < cfg->n_interfaces; i++) {
3617 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3618 struct iface *iface;
3620 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3621 VLOG_WARN("port %s: %s specified twice as port interface",
3622 port->name, if_cfg->name);
3623 iface_set_ofport(if_cfg, -1);
3627 iface = iface_lookup(port->bridge, if_cfg->name);
3629 if (iface->port != port) {
3630 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3632 port->bridge->name, if_cfg->name, iface->port->name);
3635 iface->cfg = if_cfg;
3637 iface = iface_create(port, if_cfg);
3640 /* Determine interface type. The local port always has type
3641 * "internal". Other ports take their type from the database and
3642 * default to "system" if none is specified. */
3643 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3644 : if_cfg->type[0] ? if_cfg->type
3647 shash_destroy(&new_ifaces);
3652 if (port->n_ifaces < 2) {
3654 if (vlan >= 0 && vlan <= 4095) {
3655 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3660 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3661 * they even work as-is. But they have not been tested. */
3662 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3666 if (port->vlan != vlan) {
3668 bridge_flush(port->bridge);
3671 /* Get trunked VLANs. */
3673 if (vlan < 0 && cfg->n_trunks) {
3676 trunks = bitmap_allocate(4096);
3678 for (i = 0; i < cfg->n_trunks; i++) {
3679 int trunk = cfg->trunks[i];
3681 bitmap_set1(trunks, trunk);
3687 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3688 port->name, cfg->n_trunks);
3690 if (n_errors == cfg->n_trunks) {
3691 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3693 bitmap_free(trunks);
3696 } else if (vlan >= 0 && cfg->n_trunks) {
3697 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3701 ? port->trunks != NULL
3702 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3703 bridge_flush(port->bridge);
3705 bitmap_free(port->trunks);
3706 port->trunks = trunks;
3710 port_destroy(struct port *port)
3713 struct bridge *br = port->bridge;
3717 proc_net_compat_update_vlan(port->name, NULL, 0);
3718 proc_net_compat_update_bond(port->name, NULL);
3720 for (i = 0; i < MAX_MIRRORS; i++) {
3721 struct mirror *m = br->mirrors[i];
3722 if (m && m->out_port == port) {
3727 while (port->n_ifaces > 0) {
3728 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3731 shash_find_and_delete_assert(&br->port_by_name, port->name);
3733 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3734 del->port_idx = port->port_idx;
3736 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3738 netdev_monitor_destroy(port->monitor);
3740 bitmap_free(port->trunks);
3747 static struct port *
3748 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3750 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3751 return iface ? iface->port : NULL;
3754 static struct port *
3755 port_lookup(const struct bridge *br, const char *name)
3757 return shash_find_data(&br->port_by_name, name);
3760 static struct iface *
3761 port_lookup_iface(const struct port *port, const char *name)
3763 struct iface *iface = iface_lookup(port->bridge, name);
3764 return iface && iface->port == port ? iface : NULL;
3768 port_update_bonding(struct port *port)
3770 if (port->monitor) {
3771 netdev_monitor_destroy(port->monitor);
3772 port->monitor = NULL;
3774 if (port->n_ifaces < 2) {
3775 /* Not a bonded port. */
3776 if (port->bond_hash) {
3777 free(port->bond_hash);
3778 port->bond_hash = NULL;
3779 port->bond_compat_is_stale = true;
3780 port->bond_fake_iface = false;
3785 if (!port->bond_hash) {
3786 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3787 for (i = 0; i <= BOND_MASK; i++) {
3788 struct bond_entry *e = &port->bond_hash[i];
3792 port->no_ifaces_tag = tag_create_random();
3793 bond_choose_active_iface(port);
3794 port->bond_next_rebalance
3795 = time_msec() + port->bond_rebalance_interval;
3797 if (port->cfg->bond_fake_iface) {
3798 port->bond_next_fake_iface_update = time_msec();
3801 port->bond_compat_is_stale = true;
3802 port->bond_fake_iface = port->cfg->bond_fake_iface;
3804 port->monitor = netdev_monitor_create();
3805 for (i = 0; i < port->n_ifaces; i++) {
3806 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
3812 port_update_bond_compat(struct port *port)
3814 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3815 struct compat_bond bond;
3818 if (port->n_ifaces < 2) {
3819 proc_net_compat_update_bond(port->name, NULL);
3824 bond.updelay = port->updelay;
3825 bond.downdelay = port->downdelay;
3828 bond.hashes = compat_hashes;
3829 if (port->bond_hash) {
3830 const struct bond_entry *e;
3831 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3832 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3833 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3834 cbh->hash = e - port->bond_hash;
3835 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3840 bond.n_slaves = port->n_ifaces;
3841 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3842 for (i = 0; i < port->n_ifaces; i++) {
3843 struct iface *iface = port->ifaces[i];
3844 struct compat_bond_slave *slave = &bond.slaves[i];
3845 slave->name = iface->name;
3847 /* We need to make the same determination as the Linux bonding
3848 * code to determine whether a slave should be consider "up".
3849 * The Linux function bond_miimon_inspect() supports four
3850 * BOND_LINK_* states:
3852 * - BOND_LINK_UP: carrier detected, updelay has passed.
3853 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3854 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3855 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3857 * The function bond_info_show_slave() only considers BOND_LINK_UP
3858 * to be "up" and anything else to be "down".
3860 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3864 netdev_get_etheraddr(iface->netdev, slave->mac);
3867 if (port->bond_fake_iface) {
3868 struct netdev *bond_netdev;
3870 if (!netdev_open_default(port->name, &bond_netdev)) {
3872 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3874 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3876 netdev_close(bond_netdev);
3880 proc_net_compat_update_bond(port->name, &bond);
3885 port_update_vlan_compat(struct port *port)
3887 struct bridge *br = port->bridge;
3888 char *vlandev_name = NULL;
3890 if (port->vlan > 0) {
3891 /* Figure out the name that the VLAN device should actually have, if it
3892 * existed. This takes some work because the VLAN device would not
3893 * have port->name in its name; rather, it would have the trunk port's
3894 * name, and 'port' would be attached to a bridge that also had the
3895 * VLAN device one of its ports. So we need to find a trunk port that
3896 * includes port->vlan.
3898 * There might be more than one candidate. This doesn't happen on
3899 * XenServer, so if it happens we just pick the first choice in
3900 * alphabetical order instead of creating multiple VLAN devices. */
3902 for (i = 0; i < br->n_ports; i++) {
3903 struct port *p = br->ports[i];
3904 if (port_trunks_vlan(p, port->vlan)
3906 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3908 uint8_t ea[ETH_ADDR_LEN];
3909 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3910 if (!eth_addr_is_multicast(ea) &&
3911 !eth_addr_is_reserved(ea) &&
3912 !eth_addr_is_zero(ea)) {
3913 vlandev_name = p->name;
3918 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3921 /* Interface functions. */
3924 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
3927 union ofp_action action;
3929 memset(&action, 0, sizeof action);
3930 action.output.type = htons(OFPAT_OUTPUT);
3931 action.output.len = htons(sizeof action);
3932 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
3934 flow_extract(packet, 0, ODPP_NONE, &flow);
3936 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
3938 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3939 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
3943 static struct iface *
3944 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3946 struct bridge *br = port->bridge;
3947 struct iface *iface;
3948 char *name = if_cfg->name;
3950 iface = xzalloc(sizeof *iface);
3952 iface->port_ifidx = port->n_ifaces;
3953 iface->name = xstrdup(name);
3954 iface->dp_ifidx = -1;
3955 iface->tag = tag_create_random();
3956 iface->delay_expires = LLONG_MAX;
3957 iface->netdev = NULL;
3958 iface->cfg = if_cfg;
3960 shash_add_assert(&br->iface_by_name, iface->name, iface);
3962 if (port->n_ifaces >= port->allocated_ifaces) {
3963 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3964 sizeof *port->ifaces);
3966 port->ifaces[port->n_ifaces++] = iface;
3967 if (port->n_ifaces > 1) {
3968 br->has_bonded_ports = true;
3971 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3979 iface_destroy(struct iface *iface)
3982 struct port *port = iface->port;
3983 struct bridge *br = port->bridge;
3984 bool del_active = port->active_iface == iface->port_ifidx;
3987 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3989 if (iface->dp_ifidx >= 0) {
3990 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
3993 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3994 del->port_ifidx = iface->port_ifidx;
3996 netdev_close(iface->netdev);
3999 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4000 bond_choose_active_iface(port);
4001 bond_send_learning_packets(port);
4004 cfm_destroy(iface->cfm);
4009 bridge_flush(port->bridge);
4013 static struct iface *
4014 iface_lookup(const struct bridge *br, const char *name)
4016 return shash_find_data(&br->iface_by_name, name);
4019 static struct iface *
4020 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4022 struct iface *iface;
4024 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4025 hash_int(dp_ifidx, 0), &br->ifaces) {
4026 if (iface->dp_ifidx == dp_ifidx) {
4033 /* Set Ethernet address of 'iface', if one is specified in the configuration
4036 iface_set_mac(struct iface *iface)
4038 uint8_t ea[ETH_ADDR_LEN];
4040 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4041 if (eth_addr_is_multicast(ea)) {
4042 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4044 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4045 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4046 iface->name, iface->name);
4048 int error = netdev_set_etheraddr(iface->netdev, ea);
4050 VLOG_ERR("interface %s: setting MAC failed (%s)",
4051 iface->name, strerror(error));
4057 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4059 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4062 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4066 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4068 * The value strings in '*shash' are taken directly from values[], not copied,
4069 * so the caller should not modify or free them. */
4071 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4072 struct shash *shash)
4077 for (i = 0; i < n; i++) {
4078 shash_add(shash, keys[i], values[i]);
4082 struct iface_delete_queues_cbdata {
4083 struct netdev *netdev;
4084 const struct ovsdb_datum *queues;
4088 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4090 union ovsdb_atom atom;
4092 atom.integer = target;
4093 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4097 iface_delete_queues(unsigned int queue_id,
4098 const struct shash *details OVS_UNUSED, void *cbdata_)
4100 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4102 if (!queue_ids_include(cbdata->queues, queue_id)) {
4103 netdev_delete_queue(cbdata->netdev, queue_id);
4108 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4110 if (!qos || qos->type[0] == '\0') {
4111 netdev_set_qos(iface->netdev, NULL, NULL);
4113 struct iface_delete_queues_cbdata cbdata;
4114 struct shash details;
4117 /* Configure top-level Qos for 'iface'. */
4118 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4119 qos->n_other_config, &details);
4120 netdev_set_qos(iface->netdev, qos->type, &details);
4121 shash_destroy(&details);
4123 /* Deconfigure queues that were deleted. */
4124 cbdata.netdev = iface->netdev;
4125 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4127 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4129 /* Configure queues for 'iface'. */
4130 for (i = 0; i < qos->n_queues; i++) {
4131 const struct ovsrec_queue *queue = qos->value_queues[i];
4132 unsigned int queue_id = qos->key_queues[i];
4134 shash_from_ovs_idl_map(queue->key_other_config,
4135 queue->value_other_config,
4136 queue->n_other_config, &details);
4137 netdev_set_queue(iface->netdev, queue_id, &details);
4138 shash_destroy(&details);
4144 iface_update_cfm(struct iface *iface)
4148 uint16_t *remote_mps;
4149 struct ovsrec_monitor *mon;
4150 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4152 mon = iface->cfg->monitor;
4158 if (netdev_get_etheraddr(iface->netdev, ea)) {
4159 VLOG_WARN("interface %s: Failed to get ethernet address. "
4160 "Skipping Monitor.", iface->name);
4164 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4165 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4170 iface->cfm = cfm_create();
4174 cfm->mpid = mon->mpid;
4175 cfm->interval = mon->interval ? *mon->interval : 1000;
4177 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4178 memcpy(cfm->maid, maid, sizeof cfm->maid);
4180 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4181 for(i = 0; i < mon->n_remote_mps; i++) {
4182 remote_mps[i] = mon->remote_mps[i]->mpid;
4184 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4187 if (!cfm_configure(iface->cfm)) {
4188 cfm_destroy(iface->cfm);
4193 /* Port mirroring. */
4195 static struct mirror *
4196 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4200 for (i = 0; i < MAX_MIRRORS; i++) {
4201 struct mirror *m = br->mirrors[i];
4202 if (m && uuid_equals(uuid, &m->uuid)) {
4210 mirror_reconfigure(struct bridge *br)
4212 unsigned long *rspan_vlans;
4215 /* Get rid of deleted mirrors. */
4216 for (i = 0; i < MAX_MIRRORS; i++) {
4217 struct mirror *m = br->mirrors[i];
4219 const struct ovsdb_datum *mc;
4220 union ovsdb_atom atom;
4222 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4223 atom.uuid = br->mirrors[i]->uuid;
4224 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4230 /* Add new mirrors and reconfigure existing ones. */
4231 for (i = 0; i < br->cfg->n_mirrors; i++) {
4232 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4233 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4235 mirror_reconfigure_one(m, cfg);
4237 mirror_create(br, cfg);
4241 /* Update port reserved status. */
4242 for (i = 0; i < br->n_ports; i++) {
4243 br->ports[i]->is_mirror_output_port = false;
4245 for (i = 0; i < MAX_MIRRORS; i++) {
4246 struct mirror *m = br->mirrors[i];
4247 if (m && m->out_port) {
4248 m->out_port->is_mirror_output_port = true;
4252 /* Update flooded vlans (for RSPAN). */
4254 if (br->cfg->n_flood_vlans) {
4255 rspan_vlans = bitmap_allocate(4096);
4257 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4258 int64_t vlan = br->cfg->flood_vlans[i];
4259 if (vlan >= 0 && vlan < 4096) {
4260 bitmap_set1(rspan_vlans, vlan);
4261 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4264 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4269 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4275 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4280 for (i = 0; ; i++) {
4281 if (i >= MAX_MIRRORS) {
4282 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4283 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4286 if (!br->mirrors[i]) {
4291 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4294 br->mirrors[i] = m = xzalloc(sizeof *m);
4297 m->name = xstrdup(cfg->name);
4298 shash_init(&m->src_ports);
4299 shash_init(&m->dst_ports);
4305 mirror_reconfigure_one(m, cfg);
4309 mirror_destroy(struct mirror *m)
4312 struct bridge *br = m->bridge;
4315 for (i = 0; i < br->n_ports; i++) {
4316 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4317 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4320 shash_destroy(&m->src_ports);
4321 shash_destroy(&m->dst_ports);
4324 m->bridge->mirrors[m->idx] = NULL;
4333 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4334 struct shash *names)
4338 for (i = 0; i < n_ports; i++) {
4339 const char *name = ports[i]->name;
4340 if (port_lookup(m->bridge, name)) {
4341 shash_add_once(names, name, NULL);
4343 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4344 "port %s", m->bridge->name, m->name, name);
4350 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4356 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4358 for (i = 0; i < cfg->n_select_vlan; i++) {
4359 int64_t vlan = cfg->select_vlan[i];
4360 if (vlan < 0 || vlan > 4095) {
4361 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4362 m->bridge->name, m->name, vlan);
4364 (*vlans)[n_vlans++] = vlan;
4371 vlan_is_mirrored(const struct mirror *m, int vlan)
4375 for (i = 0; i < m->n_vlans; i++) {
4376 if (m->vlans[i] == vlan) {
4384 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4388 for (i = 0; i < m->n_vlans; i++) {
4389 if (port_trunks_vlan(p, m->vlans[i])) {
4397 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4399 struct shash src_ports, dst_ports;
4400 mirror_mask_t mirror_bit;
4401 struct port *out_port;
4408 if (strcmp(cfg->name, m->name)) {
4410 m->name = xstrdup(cfg->name);
4413 /* Get output port. */
4414 if (cfg->output_port) {
4415 out_port = port_lookup(m->bridge, cfg->output_port->name);
4417 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4418 m->bridge->name, m->name);
4424 if (cfg->output_vlan) {
4425 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4426 "output vlan; ignoring output vlan",
4427 m->bridge->name, m->name);
4429 } else if (cfg->output_vlan) {
4431 out_vlan = *cfg->output_vlan;
4433 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4434 m->bridge->name, m->name);
4439 shash_init(&src_ports);
4440 shash_init(&dst_ports);
4441 if (cfg->select_all) {
4442 for (i = 0; i < m->bridge->n_ports; i++) {
4443 const char *name = m->bridge->ports[i]->name;
4444 shash_add_once(&src_ports, name, NULL);
4445 shash_add_once(&dst_ports, name, NULL);
4450 /* Get ports, and drop duplicates and ports that don't exist. */
4451 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4453 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4456 /* Get all the vlans, and drop duplicate and invalid vlans. */
4457 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4460 /* Update mirror data. */
4461 if (!shash_equal_keys(&m->src_ports, &src_ports)
4462 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4463 || m->n_vlans != n_vlans
4464 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4465 || m->out_port != out_port
4466 || m->out_vlan != out_vlan) {
4467 bridge_flush(m->bridge);
4469 shash_swap(&m->src_ports, &src_ports);
4470 shash_swap(&m->dst_ports, &dst_ports);
4473 m->n_vlans = n_vlans;
4474 m->out_port = out_port;
4475 m->out_vlan = out_vlan;
4478 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4479 for (i = 0; i < m->bridge->n_ports; i++) {
4480 struct port *port = m->bridge->ports[i];
4482 if (shash_find(&m->src_ports, port->name)
4485 ? port_trunks_any_mirrored_vlan(m, port)
4486 : vlan_is_mirrored(m, port->vlan)))) {
4487 port->src_mirrors |= mirror_bit;
4489 port->src_mirrors &= ~mirror_bit;
4492 if (shash_find(&m->dst_ports, port->name)) {
4493 port->dst_mirrors |= mirror_bit;
4495 port->dst_mirrors &= ~mirror_bit;
4500 shash_destroy(&src_ports);
4501 shash_destroy(&dst_ports);