1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
40 #include "dynamic-string.h"
46 #include "mac-learning.h"
50 #include "ofp-print.h"
52 #include "ofproto/netflow.h"
53 #include "ofproto/ofproto.h"
54 #include "ovsdb-data.h"
56 #include "poll-loop.h"
57 #include "proc-net-compat.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_reconfigure);
86 struct dst builtin[32];
91 static void dst_set_init(struct dst_set *);
92 static void dst_set_add(struct dst_set *, const struct dst *);
93 static void dst_set_free(struct dst_set *);
96 /* These members are always valid. */
97 struct port *port; /* Containing port. */
98 size_t port_ifidx; /* Index within containing port. */
99 char *name; /* Host network device name. */
100 tag_type tag; /* Tag associated with this interface. */
101 long long delay_expires; /* Time after which 'enabled' may change. */
103 /* These members are valid only after bridge_reconfigure() causes them to
105 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
106 int dp_ifidx; /* Index within kernel datapath. */
107 struct netdev *netdev; /* Network device. */
108 bool enabled; /* May be chosen for flows? */
109 bool up; /* Is the interface up? */
110 const char *type; /* Usually same as cfg->type. */
111 struct cfm *cfm; /* Connectivity Fault Management */
112 const struct ovsrec_interface *cfg;
115 #define BOND_MASK 0xff
117 int iface_idx; /* Index of assigned iface, or -1 if none. */
118 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
119 tag_type iface_tag; /* Tag associated with iface_idx. */
123 BM_SLB, /* Source Load Balance (Default). */
124 BM_AB /* Active Backup. */
127 #define MAX_MIRRORS 32
128 typedef uint32_t mirror_mask_t;
129 #define MIRROR_MASK_C(X) UINT32_C(X)
130 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
132 struct bridge *bridge;
135 struct uuid uuid; /* UUID of this "mirror" record in database. */
137 /* Selection criteria. */
138 struct shash src_ports; /* Name is port name; data is always NULL. */
139 struct shash dst_ports; /* Name is port name; data is always NULL. */
144 struct port *out_port;
148 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
150 struct bridge *bridge;
152 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
153 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
154 * NULL if all VLANs are trunked. */
155 const struct ovsrec_port *cfg;
158 /* An ordinary bridge port has 1 interface.
159 * A bridge port for bonding has at least 2 interfaces. */
160 struct iface **ifaces;
161 size_t n_ifaces, allocated_ifaces;
164 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
165 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
166 tag_type active_iface_tag; /* Tag for bcast flows. */
167 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
168 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
169 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
170 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
171 bool miimon; /* Use miimon instead of carrier? */
172 long long int bond_miimon_interval; /* Miimon status refresh interval. */
173 long long int bond_miimon_next_update; /* Time of next miimon update. */
174 long long int bond_next_fake_iface_update; /* Time of next update. */
175 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
177 /* SLB specific bonding info. */
178 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
179 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
180 long long int bond_next_rebalance; /* Next rebalancing time. */
182 /* Port mirroring info. */
183 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
184 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
185 bool is_mirror_output_port; /* Does port mirroring send frames here? */
189 struct list node; /* Node in global list of bridges. */
190 char *name; /* User-specified arbitrary name. */
191 struct mac_learning *ml; /* MAC learning table. */
192 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
193 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
194 const struct ovsrec_bridge *cfg;
196 /* OpenFlow switch processing. */
197 struct ofproto *ofproto; /* OpenFlow switch. */
199 /* Kernel datapath information. */
200 struct dpif *dpif; /* Datapath. */
201 struct hmap ifaces; /* Contains "struct iface"s. */
205 size_t n_ports, allocated_ports;
206 struct shash iface_by_name; /* "struct iface"s indexed by name. */
207 struct shash port_by_name; /* "struct port"s indexed by name. */
210 bool has_bonded_ports;
215 /* Port mirroring. */
216 struct mirror *mirrors[MAX_MIRRORS];
219 /* List of all bridges. */
220 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
222 /* OVSDB IDL used to obtain configuration. */
223 static struct ovsdb_idl *idl;
225 /* Each time this timer expires, the bridge fetches systems and interface
226 * statistics and pushes them into the database. */
227 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
228 static long long int stats_timer = LLONG_MIN;
230 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
231 static void bridge_destroy(struct bridge *);
232 static struct bridge *bridge_lookup(const char *name);
233 static unixctl_cb_func bridge_unixctl_dump_flows;
234 static unixctl_cb_func bridge_unixctl_reconnect;
235 static int bridge_run_one(struct bridge *);
236 static size_t bridge_get_controllers(const struct bridge *br,
237 struct ovsrec_controller ***controllersp);
238 static void bridge_reconfigure_one(struct bridge *);
239 static void bridge_reconfigure_remotes(struct bridge *,
240 const struct sockaddr_in *managers,
242 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
243 static void bridge_fetch_dp_ifaces(struct bridge *);
244 static void bridge_flush(struct bridge *);
245 static void bridge_pick_local_hw_addr(struct bridge *,
246 uint8_t ea[ETH_ADDR_LEN],
247 struct iface **hw_addr_iface);
248 static uint64_t bridge_pick_datapath_id(struct bridge *,
249 const uint8_t bridge_ea[ETH_ADDR_LEN],
250 struct iface *hw_addr_iface);
251 static struct iface *bridge_get_local_iface(struct bridge *);
252 static uint64_t dpid_from_hash(const void *, size_t nbytes);
254 static unixctl_cb_func bridge_unixctl_fdb_show;
256 static void bond_init(void);
257 static void bond_run(struct bridge *);
258 static void bond_wait(struct bridge *);
259 static void bond_rebalance_port(struct port *);
260 static void bond_send_learning_packets(struct port *);
261 static void bond_enable_slave(struct iface *iface, bool enable);
263 static struct port *port_create(struct bridge *, const char *name);
264 static void port_reconfigure(struct port *, const struct ovsrec_port *);
265 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
266 static void port_destroy(struct port *);
267 static struct port *port_lookup(const struct bridge *, const char *name);
268 static struct iface *port_lookup_iface(const struct port *, const char *name);
269 static struct port *port_from_dp_ifidx(const struct bridge *,
271 static void port_update_bond_compat(struct port *);
272 static void port_update_vlan_compat(struct port *);
273 static void port_update_bonding(struct port *);
275 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
276 static void mirror_destroy(struct mirror *);
277 static void mirror_reconfigure(struct bridge *);
278 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
279 static bool vlan_is_mirrored(const struct mirror *, int vlan);
281 static struct iface *iface_create(struct port *port,
282 const struct ovsrec_interface *if_cfg);
283 static void iface_destroy(struct iface *);
284 static struct iface *iface_lookup(const struct bridge *, const char *name);
285 static struct iface *iface_from_dp_ifidx(const struct bridge *,
287 static void iface_set_mac(struct iface *);
288 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
289 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
290 static void iface_update_cfm(struct iface *);
291 static void iface_refresh_cfm_stats(struct iface *iface);
292 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
294 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
296 static void shash_to_ovs_idl_map(struct shash *,
297 char ***keys, char ***values, size_t *n);
300 /* Hooks into ofproto processing. */
301 static struct ofhooks bridge_ofhooks;
303 /* Public functions. */
305 /* Initializes the bridge module, configuring it to obtain its configuration
306 * from an OVSDB server accessed over 'remote', which should be a string in a
307 * form acceptable to ovsdb_idl_create(). */
309 bridge_init(const char *remote)
311 /* Create connection to database. */
312 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
314 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
315 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
316 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
318 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
320 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
321 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
323 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
324 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
325 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
327 /* Register unixctl commands. */
328 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
329 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
331 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
339 struct bridge *br, *next_br;
341 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
344 ovsdb_idl_destroy(idl);
347 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
348 * but for which the ovs-vswitchd configuration 'cfg' is required. */
350 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
352 static bool already_configured_once;
353 struct svec bridge_names;
354 struct svec dpif_names, dpif_types;
357 /* Only do this once per ovs-vswitchd run. */
358 if (already_configured_once) {
361 already_configured_once = true;
363 stats_timer = time_msec() + STATS_INTERVAL;
365 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
366 svec_init(&bridge_names);
367 for (i = 0; i < cfg->n_bridges; i++) {
368 svec_add(&bridge_names, cfg->bridges[i]->name);
370 svec_sort(&bridge_names);
372 /* Iterate over all system dpifs and delete any of them that do not appear
374 svec_init(&dpif_names);
375 svec_init(&dpif_types);
376 dp_enumerate_types(&dpif_types);
377 for (i = 0; i < dpif_types.n; i++) {
380 dp_enumerate_names(dpif_types.names[i], &dpif_names);
382 /* Delete each dpif whose name is not in 'bridge_names'. */
383 for (j = 0; j < dpif_names.n; j++) {
384 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
388 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
397 svec_destroy(&bridge_names);
398 svec_destroy(&dpif_names);
399 svec_destroy(&dpif_types);
402 /* Callback for iterate_and_prune_ifaces(). */
404 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
406 if (!iface->netdev) {
407 /* We already reported a related error, don't bother duplicating it. */
411 if (iface->dp_ifidx < 0) {
412 VLOG_ERR("%s interface not in %s, dropping",
413 iface->name, dpif_name(br->dpif));
417 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
418 iface->name, iface->dp_ifidx);
422 /* Callback for iterate_and_prune_ifaces(). */
424 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
425 void *aux OVS_UNUSED)
427 /* Set policing attributes. */
428 netdev_set_policing(iface->netdev,
429 iface->cfg->ingress_policing_rate,
430 iface->cfg->ingress_policing_burst);
432 /* Set MAC address of internal interfaces other than the local
434 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
435 iface_set_mac(iface);
441 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
442 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
443 * deletes from 'br' any ports that no longer have any interfaces. */
445 iterate_and_prune_ifaces(struct bridge *br,
446 bool (*cb)(struct bridge *, struct iface *,
452 for (i = 0; i < br->n_ports; ) {
453 struct port *port = br->ports[i];
454 for (j = 0; j < port->n_ifaces; ) {
455 struct iface *iface = port->ifaces[j];
456 if (cb(br, iface, aux)) {
459 iface_set_ofport(iface->cfg, -1);
460 iface_destroy(iface);
464 if (port->n_ifaces) {
467 VLOG_ERR("%s port has no interfaces, dropping", port->name);
473 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
474 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
475 * responsible for freeing '*managersp' (with free()).
477 * You may be asking yourself "why does ovs-vswitchd care?", because
478 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
479 * should not be and in fact is not directly involved in that. But
480 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
481 * it has to tell in-band control where the managers are to enable that.
482 * (Thus, only managers connected in-band are collected.)
485 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
486 struct sockaddr_in **managersp, size_t *n_managersp)
488 struct sockaddr_in *managers = NULL;
489 size_t n_managers = 0;
490 struct shash targets;
493 /* Collect all of the potential targets, as the union of the "managers"
494 * column and the "targets" columns of the rows pointed to by
495 * "manager_options", excluding any that are out-of-band. */
496 shash_init(&targets);
497 for (i = 0; i < ovs_cfg->n_managers; i++) {
498 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
500 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
501 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
503 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
504 shash_find_and_delete(&targets, m->target);
506 shash_add_once(&targets, m->target, NULL);
510 /* Now extract the targets' IP addresses. */
511 if (!shash_is_empty(&targets)) {
512 struct shash_node *node;
514 managers = xmalloc(shash_count(&targets) * sizeof *managers);
515 SHASH_FOR_EACH (node, &targets) {
516 const char *target = node->name;
517 struct sockaddr_in *sin = &managers[n_managers];
519 if ((!strncmp(target, "tcp:", 4)
520 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
521 (!strncmp(target, "ssl:", 4)
522 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
527 shash_destroy(&targets);
529 *managersp = managers;
530 *n_managersp = n_managers;
534 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
536 struct shash old_br, new_br;
537 struct shash_node *node;
538 struct bridge *br, *next;
539 struct sockaddr_in *managers;
542 int sflow_bridge_number;
544 COVERAGE_INC(bridge_reconfigure);
546 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
548 /* Collect old and new bridges. */
551 LIST_FOR_EACH (br, node, &all_bridges) {
552 shash_add(&old_br, br->name, br);
554 for (i = 0; i < ovs_cfg->n_bridges; i++) {
555 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
556 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
557 VLOG_WARN("more than one bridge named %s", br_cfg->name);
561 /* Get rid of deleted bridges and add new bridges. */
562 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
563 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
570 SHASH_FOR_EACH (node, &new_br) {
571 const char *br_name = node->name;
572 const struct ovsrec_bridge *br_cfg = node->data;
573 br = shash_find_data(&old_br, br_name);
575 /* If the bridge datapath type has changed, we need to tear it
576 * down and recreate. */
577 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
579 bridge_create(br_cfg);
582 bridge_create(br_cfg);
585 shash_destroy(&old_br);
586 shash_destroy(&new_br);
588 /* Reconfigure all bridges. */
589 LIST_FOR_EACH (br, node, &all_bridges) {
590 bridge_reconfigure_one(br);
593 /* Add and delete ports on all datapaths.
595 * The kernel will reject any attempt to add a given port to a datapath if
596 * that port already belongs to a different datapath, so we must do all
597 * port deletions before any port additions. */
598 LIST_FOR_EACH (br, node, &all_bridges) {
599 struct dpif_port_dump dump;
600 struct shash want_ifaces;
601 struct dpif_port dpif_port;
603 bridge_get_all_ifaces(br, &want_ifaces);
604 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
605 if (!shash_find(&want_ifaces, dpif_port.name)
606 && strcmp(dpif_port.name, br->name)) {
607 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
609 VLOG_ERR("failed to remove %s interface from %s: %s",
610 dpif_port.name, dpif_name(br->dpif),
615 shash_destroy(&want_ifaces);
617 LIST_FOR_EACH (br, node, &all_bridges) {
618 struct shash cur_ifaces, want_ifaces;
619 struct dpif_port_dump dump;
620 struct dpif_port dpif_port;
622 /* Get the set of interfaces currently in this datapath. */
623 shash_init(&cur_ifaces);
624 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
625 struct dpif_port *port_info = xmalloc(sizeof *port_info);
626 dpif_port_clone(port_info, &dpif_port);
627 shash_add(&cur_ifaces, dpif_port.name, port_info);
630 /* Get the set of interfaces we want on this datapath. */
631 bridge_get_all_ifaces(br, &want_ifaces);
633 hmap_clear(&br->ifaces);
634 SHASH_FOR_EACH (node, &want_ifaces) {
635 const char *if_name = node->name;
636 struct iface *iface = node->data;
637 struct dpif_port *dpif_port;
641 type = iface ? iface->type : "internal";
642 dpif_port = shash_find_data(&cur_ifaces, if_name);
644 /* If we have a port or a netdev already, and it's not the type we
645 * want, then delete the port (if any) and close the netdev (if
647 if ((dpif_port && strcmp(dpif_port->type, type))
648 || (iface && iface->netdev
649 && strcmp(type, netdev_get_type(iface->netdev)))) {
651 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
658 netdev_close(iface->netdev);
659 iface->netdev = NULL;
663 /* If the port doesn't exist or we don't have the netdev open,
664 * we need to do more work. */
665 if (!dpif_port || (iface && !iface->netdev)) {
666 struct netdev_options options;
667 struct netdev *netdev;
670 /* First open the network device. */
671 options.name = if_name;
673 options.args = &args;
674 options.ethertype = NETDEV_ETH_TYPE_NONE;
678 shash_from_ovs_idl_map(iface->cfg->key_options,
679 iface->cfg->value_options,
680 iface->cfg->n_options, &args);
682 error = netdev_open(&options, &netdev);
683 shash_destroy(&args);
686 VLOG_WARN("could not open network device %s (%s)",
687 if_name, strerror(error));
691 /* Then add the port if we haven't already. */
693 error = dpif_port_add(br->dpif, netdev, NULL);
695 netdev_close(netdev);
696 if (error == EFBIG) {
697 VLOG_ERR("ran out of valid port numbers on %s",
698 dpif_name(br->dpif));
701 VLOG_ERR("failed to add %s interface to %s: %s",
702 if_name, dpif_name(br->dpif),
709 /* Update 'iface'. */
711 iface->netdev = netdev;
712 iface->enabled = netdev_get_carrier(iface->netdev);
713 iface->up = iface->enabled;
715 } else if (iface && iface->netdev) {
719 shash_from_ovs_idl_map(iface->cfg->key_options,
720 iface->cfg->value_options,
721 iface->cfg->n_options, &args);
722 netdev_set_config(iface->netdev, &args);
723 shash_destroy(&args);
726 shash_destroy(&want_ifaces);
728 SHASH_FOR_EACH (node, &cur_ifaces) {
729 struct dpif_port *port_info = node->data;
730 dpif_port_destroy(port_info);
733 shash_destroy(&cur_ifaces);
735 sflow_bridge_number = 0;
736 LIST_FOR_EACH (br, node, &all_bridges) {
739 struct iface *local_iface;
740 struct iface *hw_addr_iface;
743 bridge_fetch_dp_ifaces(br);
745 iterate_and_prune_ifaces(br, check_iface, NULL);
747 /* Pick local port hardware address, datapath ID. */
748 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
749 local_iface = bridge_get_local_iface(br);
751 int error = netdev_set_etheraddr(local_iface->netdev, ea);
753 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
754 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
755 "Ethernet address: %s",
756 br->name, strerror(error));
759 memcpy(br->ea, ea, ETH_ADDR_LEN);
761 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
762 ofproto_set_datapath_id(br->ofproto, dpid);
764 dpid_string = xasprintf("%016"PRIx64, dpid);
765 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
768 /* Set NetFlow configuration on this bridge. */
769 if (br->cfg->netflow) {
770 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
771 struct netflow_options opts;
773 memset(&opts, 0, sizeof opts);
775 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
776 if (nf_cfg->engine_type) {
777 opts.engine_type = *nf_cfg->engine_type;
779 if (nf_cfg->engine_id) {
780 opts.engine_id = *nf_cfg->engine_id;
783 opts.active_timeout = nf_cfg->active_timeout;
784 if (!opts.active_timeout) {
785 opts.active_timeout = -1;
786 } else if (opts.active_timeout < 0) {
787 VLOG_WARN("bridge %s: active timeout interval set to negative "
788 "value, using default instead (%d seconds)", br->name,
789 NF_ACTIVE_TIMEOUT_DEFAULT);
790 opts.active_timeout = -1;
793 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
794 if (opts.add_id_to_iface) {
795 if (opts.engine_id > 0x7f) {
796 VLOG_WARN("bridge %s: netflow port mangling may conflict "
797 "with another vswitch, choose an engine id less "
798 "than 128", br->name);
800 if (br->n_ports > 508) {
801 VLOG_WARN("bridge %s: netflow port mangling will conflict "
802 "with another port when more than 508 ports are "
807 opts.collectors.n = nf_cfg->n_targets;
808 opts.collectors.names = nf_cfg->targets;
809 if (ofproto_set_netflow(br->ofproto, &opts)) {
810 VLOG_ERR("bridge %s: problem setting netflow collectors",
814 ofproto_set_netflow(br->ofproto, NULL);
817 /* Set sFlow configuration on this bridge. */
818 if (br->cfg->sflow) {
819 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
820 struct ovsrec_controller **controllers;
821 struct ofproto_sflow_options oso;
822 size_t n_controllers;
824 memset(&oso, 0, sizeof oso);
826 oso.targets.n = sflow_cfg->n_targets;
827 oso.targets.names = sflow_cfg->targets;
829 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
830 if (sflow_cfg->sampling) {
831 oso.sampling_rate = *sflow_cfg->sampling;
834 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
835 if (sflow_cfg->polling) {
836 oso.polling_interval = *sflow_cfg->polling;
839 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
840 if (sflow_cfg->header) {
841 oso.header_len = *sflow_cfg->header;
844 oso.sub_id = sflow_bridge_number++;
845 oso.agent_device = sflow_cfg->agent;
847 oso.control_ip = NULL;
848 n_controllers = bridge_get_controllers(br, &controllers);
849 for (i = 0; i < n_controllers; i++) {
850 if (controllers[i]->local_ip) {
851 oso.control_ip = controllers[i]->local_ip;
855 ofproto_set_sflow(br->ofproto, &oso);
857 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
859 ofproto_set_sflow(br->ofproto, NULL);
862 /* Update the controller and related settings. It would be more
863 * straightforward to call this from bridge_reconfigure_one(), but we
864 * can't do it there for two reasons. First, and most importantly, at
865 * that point we don't know the dp_ifidx of any interfaces that have
866 * been added to the bridge (because we haven't actually added them to
867 * the datapath). Second, at that point we haven't set the datapath ID
868 * yet; when a controller is configured, resetting the datapath ID will
869 * immediately disconnect from the controller, so it's better to set
870 * the datapath ID before the controller. */
871 bridge_reconfigure_remotes(br, managers, n_managers);
873 LIST_FOR_EACH (br, node, &all_bridges) {
874 for (i = 0; i < br->n_ports; i++) {
875 struct port *port = br->ports[i];
878 port_update_vlan_compat(port);
879 port_update_bonding(port);
881 for (j = 0; j < port->n_ifaces; j++) {
882 iface_update_qos(port->ifaces[j], port->cfg->qos);
886 LIST_FOR_EACH (br, node, &all_bridges) {
887 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
890 LIST_FOR_EACH (br, node, &all_bridges) {
892 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
893 iface_update_cfm(iface);
901 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
902 const struct ovsdb_idl_column *column,
905 const struct ovsdb_datum *datum;
906 union ovsdb_atom atom;
909 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
910 atom.string = (char *) key;
911 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
912 return idx == UINT_MAX ? NULL : datum->values[idx].string;
916 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
918 return get_ovsrec_key_value(&br_cfg->header_,
919 &ovsrec_bridge_col_other_config, key);
923 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
924 struct iface **hw_addr_iface)
930 *hw_addr_iface = NULL;
932 /* Did the user request a particular MAC? */
933 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
934 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
935 if (eth_addr_is_multicast(ea)) {
936 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
937 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
938 } else if (eth_addr_is_zero(ea)) {
939 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
945 /* Otherwise choose the minimum non-local MAC address among all of the
947 memset(ea, 0xff, sizeof ea);
948 for (i = 0; i < br->n_ports; i++) {
949 struct port *port = br->ports[i];
950 uint8_t iface_ea[ETH_ADDR_LEN];
953 /* Mirror output ports don't participate. */
954 if (port->is_mirror_output_port) {
958 /* Choose the MAC address to represent the port. */
959 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
960 /* Find the interface with this Ethernet address (if any) so that
961 * we can provide the correct devname to the caller. */
963 for (j = 0; j < port->n_ifaces; j++) {
964 struct iface *candidate = port->ifaces[j];
965 uint8_t candidate_ea[ETH_ADDR_LEN];
966 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
967 && eth_addr_equals(iface_ea, candidate_ea)) {
972 /* Choose the interface whose MAC address will represent the port.
973 * The Linux kernel bonding code always chooses the MAC address of
974 * the first slave added to a bond, and the Fedora networking
975 * scripts always add slaves to a bond in alphabetical order, so
976 * for compatibility we choose the interface with the name that is
977 * first in alphabetical order. */
978 iface = port->ifaces[0];
979 for (j = 1; j < port->n_ifaces; j++) {
980 struct iface *candidate = port->ifaces[j];
981 if (strcmp(candidate->name, iface->name) < 0) {
986 /* The local port doesn't count (since we're trying to choose its
987 * MAC address anyway). */
988 if (iface->dp_ifidx == ODPP_LOCAL) {
993 error = netdev_get_etheraddr(iface->netdev, iface_ea);
995 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
996 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
997 iface->name, strerror(error));
1002 /* Compare against our current choice. */
1003 if (!eth_addr_is_multicast(iface_ea) &&
1004 !eth_addr_is_local(iface_ea) &&
1005 !eth_addr_is_reserved(iface_ea) &&
1006 !eth_addr_is_zero(iface_ea) &&
1007 eth_addr_compare_3way(iface_ea, ea) < 0)
1009 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1010 *hw_addr_iface = iface;
1013 if (eth_addr_is_multicast(ea)) {
1014 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1015 *hw_addr_iface = NULL;
1016 VLOG_WARN("bridge %s: using default bridge Ethernet "
1017 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1019 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1020 br->name, ETH_ADDR_ARGS(ea));
1024 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1025 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1026 * an interface on 'br', then that interface must be passed in as
1027 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1028 * 'hw_addr_iface' must be passed in as a null pointer. */
1030 bridge_pick_datapath_id(struct bridge *br,
1031 const uint8_t bridge_ea[ETH_ADDR_LEN],
1032 struct iface *hw_addr_iface)
1035 * The procedure for choosing a bridge MAC address will, in the most
1036 * ordinary case, also choose a unique MAC that we can use as a datapath
1037 * ID. In some special cases, though, multiple bridges will end up with
1038 * the same MAC address. This is OK for the bridges, but it will confuse
1039 * the OpenFlow controller, because each datapath needs a unique datapath
1042 * Datapath IDs must be unique. It is also very desirable that they be
1043 * stable from one run to the next, so that policy set on a datapath
1046 const char *datapath_id;
1049 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1050 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1054 if (hw_addr_iface) {
1056 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1058 * A bridge whose MAC address is taken from a VLAN network device
1059 * (that is, a network device created with vconfig(8) or similar
1060 * tool) will have the same MAC address as a bridge on the VLAN
1061 * device's physical network device.
1063 * Handle this case by hashing the physical network device MAC
1064 * along with the VLAN identifier.
1066 uint8_t buf[ETH_ADDR_LEN + 2];
1067 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1068 buf[ETH_ADDR_LEN] = vlan >> 8;
1069 buf[ETH_ADDR_LEN + 1] = vlan;
1070 return dpid_from_hash(buf, sizeof buf);
1073 * Assume that this bridge's MAC address is unique, since it
1074 * doesn't fit any of the cases we handle specially.
1079 * A purely internal bridge, that is, one that has no non-virtual
1080 * network devices on it at all, is more difficult because it has no
1081 * natural unique identifier at all.
1083 * When the host is a XenServer, we handle this case by hashing the
1084 * host's UUID with the name of the bridge. Names of bridges are
1085 * persistent across XenServer reboots, although they can be reused if
1086 * an internal network is destroyed and then a new one is later
1087 * created, so this is fairly effective.
1089 * When the host is not a XenServer, we punt by using a random MAC
1090 * address on each run.
1092 const char *host_uuid = xenserver_get_host_uuid();
1094 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1095 dpid = dpid_from_hash(combined, strlen(combined));
1101 return eth_addr_to_uint64(bridge_ea);
1105 dpid_from_hash(const void *data, size_t n)
1107 uint8_t hash[SHA1_DIGEST_SIZE];
1109 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1110 sha1_bytes(data, n, hash);
1111 eth_addr_mark_random(hash);
1112 return eth_addr_to_uint64(hash);
1116 iface_refresh_status(struct iface *iface)
1120 enum netdev_flags flags;
1129 if (!netdev_get_status(iface->netdev, &sh)) {
1131 char **keys, **values;
1133 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1134 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1139 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1142 shash_destroy_free_data(&sh);
1144 error = netdev_get_flags(iface->netdev, &flags);
1146 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1149 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1152 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1154 ovsrec_interface_set_duplex(iface->cfg,
1155 netdev_features_is_full_duplex(current)
1157 /* warning: uint64_t -> int64_t conversion */
1158 bps = netdev_features_to_bps(current);
1159 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1162 ovsrec_interface_set_duplex(iface->cfg, NULL);
1163 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1167 ovsrec_interface_set_link_state(iface->cfg,
1168 netdev_get_carrier(iface->netdev)
1171 error = netdev_get_mtu(iface->netdev, &mtu);
1174 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1177 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1182 iface_refresh_cfm_stats(struct iface *iface)
1186 const struct ovsrec_monitor *mon;
1188 mon = iface->cfg->monitor;
1195 for (i = 0; i < mon->n_remote_mps; i++) {
1196 const struct ovsrec_maintenance_point *mp;
1197 const struct remote_mp *rmp;
1199 mp = mon->remote_mps[i];
1200 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1202 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1205 if (hmap_is_empty(&cfm->x_remote_mps)) {
1206 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1209 struct remote_mp *rmp;
1210 int64_t *x_remote_mps;
1212 length = hmap_count(&cfm->x_remote_mps);
1213 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1216 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1217 x_remote_mps[i++] = rmp->mpid;
1220 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1224 if (hmap_is_empty(&cfm->x_remote_maids)) {
1225 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1228 char **x_remote_maids;
1229 struct remote_maid *rmaid;
1231 length = hmap_count(&cfm->x_remote_maids);
1232 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1235 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1238 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1240 for (j = 0; j < CCM_MAID_LEN; j++) {
1241 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1246 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1248 for (i = 0; i < length; i++) {
1249 free(x_remote_maids[i]);
1251 free(x_remote_maids);
1254 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1258 iface_refresh_stats(struct iface *iface)
1264 static const struct iface_stat iface_stats[] = {
1265 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1266 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1267 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1268 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1269 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1270 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1271 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1272 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1273 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1274 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1275 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1276 { "collisions", offsetof(struct netdev_stats, collisions) },
1278 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1279 const struct iface_stat *s;
1281 char *keys[N_STATS];
1282 int64_t values[N_STATS];
1285 struct netdev_stats stats;
1287 /* Intentionally ignore return value, since errors will set 'stats' to
1288 * all-1s, and we will deal with that correctly below. */
1289 netdev_get_stats(iface->netdev, &stats);
1292 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1293 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1294 if (value != UINT64_MAX) {
1301 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1305 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1307 struct ovsdb_datum datum;
1311 get_system_stats(&stats);
1313 ovsdb_datum_from_shash(&datum, &stats);
1314 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1318 static inline const char *
1319 nx_role_to_str(enum nx_role role)
1324 case NX_ROLE_MASTER:
1329 return "*** INVALID ROLE ***";
1334 bridge_refresh_controller_status(const struct bridge *br)
1337 const struct ovsrec_controller *cfg;
1339 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1341 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1342 struct ofproto_controller_info *cinfo =
1343 shash_find_data(&info, cfg->target);
1346 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1347 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1348 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1349 (char **) cinfo->pairs.values,
1352 ovsrec_controller_set_is_connected(cfg, false);
1353 ovsrec_controller_set_role(cfg, NULL);
1354 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1358 ofproto_free_ofproto_controller_info(&info);
1364 const struct ovsrec_open_vswitch *cfg;
1366 bool datapath_destroyed;
1367 bool database_changed;
1370 /* Let each bridge do the work that it needs to do. */
1371 datapath_destroyed = false;
1372 LIST_FOR_EACH (br, node, &all_bridges) {
1373 int error = bridge_run_one(br);
1375 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1376 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1377 "forcing reconfiguration", br->name);
1378 datapath_destroyed = true;
1382 /* (Re)configure if necessary. */
1383 database_changed = ovsdb_idl_run(idl);
1384 cfg = ovsrec_open_vswitch_first(idl);
1386 /* Re-configure SSL. We do this on every trip through the main loop,
1387 * instead of just when the database changes, because the contents of the
1388 * key and certificate files can change without the database changing.
1390 * We do this before bridge_reconfigure() because that function might
1391 * initiate SSL connections and thus requires SSL to be configured. */
1392 if (cfg && cfg->ssl) {
1393 const struct ovsrec_ssl *ssl = cfg->ssl;
1395 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1396 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1399 if (database_changed || datapath_destroyed) {
1401 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1403 bridge_configure_once(cfg);
1404 bridge_reconfigure(cfg);
1406 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1407 ovsdb_idl_txn_commit(txn);
1408 ovsdb_idl_txn_destroy(txn); /* XXX */
1410 /* We still need to reconfigure to avoid dangling pointers to
1411 * now-destroyed ovsrec structures inside bridge data. */
1412 static const struct ovsrec_open_vswitch null_cfg;
1414 bridge_reconfigure(&null_cfg);
1418 /* Refresh system and interface stats if necessary. */
1419 if (time_msec() >= stats_timer) {
1421 struct ovsdb_idl_txn *txn;
1423 txn = ovsdb_idl_txn_create(idl);
1424 LIST_FOR_EACH (br, node, &all_bridges) {
1427 for (i = 0; i < br->n_ports; i++) {
1428 struct port *port = br->ports[i];
1431 for (j = 0; j < port->n_ifaces; j++) {
1432 struct iface *iface = port->ifaces[j];
1433 iface_refresh_stats(iface);
1434 iface_refresh_cfm_stats(iface);
1435 iface_refresh_status(iface);
1438 bridge_refresh_controller_status(br);
1440 refresh_system_stats(cfg);
1441 ovsdb_idl_txn_commit(txn);
1442 ovsdb_idl_txn_destroy(txn); /* XXX */
1445 stats_timer = time_msec() + STATS_INTERVAL;
1453 struct iface *iface;
1455 LIST_FOR_EACH (br, node, &all_bridges) {
1456 ofproto_wait(br->ofproto);
1457 if (ofproto_has_primary_controller(br->ofproto)) {
1461 mac_learning_wait(br->ml);
1464 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1466 cfm_wait(iface->cfm);
1470 ovsdb_idl_wait(idl);
1471 poll_timer_wait_until(stats_timer);
1474 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1475 * configuration changes. */
1477 bridge_flush(struct bridge *br)
1479 COVERAGE_INC(bridge_flush);
1481 mac_learning_flush(br->ml);
1484 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1485 * such interface. */
1486 static struct iface *
1487 bridge_get_local_iface(struct bridge *br)
1491 for (i = 0; i < br->n_ports; i++) {
1492 struct port *port = br->ports[i];
1493 for (j = 0; j < port->n_ifaces; j++) {
1494 struct iface *iface = port->ifaces[j];
1495 if (iface->dp_ifidx == ODPP_LOCAL) {
1504 /* Bridge unixctl user interface functions. */
1506 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1507 const char *args, void *aux OVS_UNUSED)
1509 struct ds ds = DS_EMPTY_INITIALIZER;
1510 const struct bridge *br;
1511 const struct mac_entry *e;
1513 br = bridge_lookup(args);
1515 unixctl_command_reply(conn, 501, "no such bridge");
1519 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1520 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1521 if (e->port < 0 || e->port >= br->n_ports) {
1524 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1525 br->ports[e->port]->ifaces[0]->dp_ifidx,
1526 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1528 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1532 /* Bridge reconfiguration functions. */
1533 static struct bridge *
1534 bridge_create(const struct ovsrec_bridge *br_cfg)
1539 assert(!bridge_lookup(br_cfg->name));
1540 br = xzalloc(sizeof *br);
1542 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1548 dpif_flow_flush(br->dpif);
1550 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1553 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1555 dpif_delete(br->dpif);
1556 dpif_close(br->dpif);
1561 br->name = xstrdup(br_cfg->name);
1563 br->ml = mac_learning_create();
1564 eth_addr_nicira_random(br->default_ea);
1566 hmap_init(&br->ifaces);
1568 shash_init(&br->port_by_name);
1569 shash_init(&br->iface_by_name);
1573 list_push_back(&all_bridges, &br->node);
1575 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1581 bridge_destroy(struct bridge *br)
1586 while (br->n_ports > 0) {
1587 port_destroy(br->ports[br->n_ports - 1]);
1589 list_remove(&br->node);
1590 error = dpif_delete(br->dpif);
1591 if (error && error != ENOENT) {
1592 VLOG_ERR("failed to delete %s: %s",
1593 dpif_name(br->dpif), strerror(error));
1595 dpif_close(br->dpif);
1596 ofproto_destroy(br->ofproto);
1597 mac_learning_destroy(br->ml);
1598 hmap_destroy(&br->ifaces);
1599 shash_destroy(&br->port_by_name);
1600 shash_destroy(&br->iface_by_name);
1607 static struct bridge *
1608 bridge_lookup(const char *name)
1612 LIST_FOR_EACH (br, node, &all_bridges) {
1613 if (!strcmp(br->name, name)) {
1620 /* Handle requests for a listing of all flows known by the OpenFlow
1621 * stack, including those normally hidden. */
1623 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1624 const char *args, void *aux OVS_UNUSED)
1629 br = bridge_lookup(args);
1631 unixctl_command_reply(conn, 501, "Unknown bridge");
1636 ofproto_get_all_flows(br->ofproto, &results);
1638 unixctl_command_reply(conn, 200, ds_cstr(&results));
1639 ds_destroy(&results);
1642 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1643 * connections and reconnect. If BRIDGE is not specified, then all bridges
1644 * drop their controller connections and reconnect. */
1646 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1647 const char *args, void *aux OVS_UNUSED)
1650 if (args[0] != '\0') {
1651 br = bridge_lookup(args);
1653 unixctl_command_reply(conn, 501, "Unknown bridge");
1656 ofproto_reconnect_controllers(br->ofproto);
1658 LIST_FOR_EACH (br, node, &all_bridges) {
1659 ofproto_reconnect_controllers(br->ofproto);
1662 unixctl_command_reply(conn, 200, NULL);
1666 bridge_run_one(struct bridge *br)
1669 struct iface *iface;
1671 error = ofproto_run1(br->ofproto);
1676 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1679 error = ofproto_run2(br->ofproto, br->flush);
1682 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1683 struct ofpbuf *packet;
1689 packet = cfm_run(iface->cfm);
1691 iface_send_packet(iface, packet);
1692 ofpbuf_uninit(packet);
1701 bridge_get_controllers(const struct bridge *br,
1702 struct ovsrec_controller ***controllersp)
1704 struct ovsrec_controller **controllers;
1705 size_t n_controllers;
1707 controllers = br->cfg->controller;
1708 n_controllers = br->cfg->n_controller;
1710 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1716 *controllersp = controllers;
1718 return n_controllers;
1722 bridge_reconfigure_one(struct bridge *br)
1724 struct shash old_ports, new_ports;
1725 struct svec snoops, old_snoops;
1726 struct shash_node *node;
1727 enum ofproto_fail_mode fail_mode;
1730 /* Collect old ports. */
1731 shash_init(&old_ports);
1732 for (i = 0; i < br->n_ports; i++) {
1733 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1736 /* Collect new ports. */
1737 shash_init(&new_ports);
1738 for (i = 0; i < br->cfg->n_ports; i++) {
1739 const char *name = br->cfg->ports[i]->name;
1740 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1741 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1746 /* If we have a controller, then we need a local port. Complain if the
1747 * user didn't specify one.
1749 * XXX perhaps we should synthesize a port ourselves in this case. */
1750 if (bridge_get_controllers(br, NULL)) {
1751 char local_name[IF_NAMESIZE];
1754 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1755 local_name, sizeof local_name);
1756 if (!error && !shash_find(&new_ports, local_name)) {
1757 VLOG_WARN("bridge %s: controller specified but no local port "
1758 "(port named %s) defined",
1759 br->name, local_name);
1763 /* Get rid of deleted ports.
1764 * Get rid of deleted interfaces on ports that still exist. */
1765 SHASH_FOR_EACH (node, &old_ports) {
1766 struct port *port = node->data;
1767 const struct ovsrec_port *port_cfg;
1769 port_cfg = shash_find_data(&new_ports, node->name);
1773 port_del_ifaces(port, port_cfg);
1777 /* Create new ports.
1778 * Add new interfaces to existing ports.
1779 * Reconfigure existing ports. */
1780 SHASH_FOR_EACH (node, &new_ports) {
1781 struct port *port = shash_find_data(&old_ports, node->name);
1783 port = port_create(br, node->name);
1786 port_reconfigure(port, node->data);
1787 if (!port->n_ifaces) {
1788 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1789 br->name, port->name);
1793 shash_destroy(&old_ports);
1794 shash_destroy(&new_ports);
1796 /* Set the fail-mode */
1797 fail_mode = !br->cfg->fail_mode
1798 || !strcmp(br->cfg->fail_mode, "standalone")
1799 ? OFPROTO_FAIL_STANDALONE
1800 : OFPROTO_FAIL_SECURE;
1801 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1802 && !ofproto_has_primary_controller(br->ofproto)) {
1803 ofproto_flush_flows(br->ofproto);
1805 ofproto_set_fail_mode(br->ofproto, fail_mode);
1807 /* Delete all flows if we're switching from connected to standalone or vice
1808 * versa. (XXX Should we delete all flows if we are switching from one
1809 * controller to another?) */
1811 /* Configure OpenFlow controller connection snooping. */
1813 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1814 ovs_rundir(), br->name));
1815 svec_init(&old_snoops);
1816 ofproto_get_snoops(br->ofproto, &old_snoops);
1817 if (!svec_equal(&snoops, &old_snoops)) {
1818 ofproto_set_snoops(br->ofproto, &snoops);
1820 svec_destroy(&snoops);
1821 svec_destroy(&old_snoops);
1823 mirror_reconfigure(br);
1826 /* Initializes 'oc' appropriately as a management service controller for
1829 * The caller must free oc->target when it is no longer needed. */
1831 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1832 struct ofproto_controller *oc)
1834 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1835 oc->max_backoff = 0;
1836 oc->probe_interval = 60;
1837 oc->band = OFPROTO_OUT_OF_BAND;
1838 oc->accept_re = NULL;
1839 oc->update_resolv_conf = false;
1841 oc->burst_limit = 0;
1844 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1846 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1847 struct ofproto_controller *oc)
1849 oc->target = c->target;
1850 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1851 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1852 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1853 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1854 oc->accept_re = c->discover_accept_regex;
1855 oc->update_resolv_conf = c->discover_update_resolv_conf;
1856 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1857 oc->burst_limit = (c->controller_burst_limit
1858 ? *c->controller_burst_limit : 0);
1861 /* Configures the IP stack for 'br''s local interface properly according to the
1862 * configuration in 'c'. */
1864 bridge_configure_local_iface_netdev(struct bridge *br,
1865 struct ovsrec_controller *c)
1867 struct netdev *netdev;
1868 struct in_addr mask, gateway;
1870 struct iface *local_iface;
1873 /* Controller discovery does its own TCP/IP configuration later. */
1874 if (strcmp(c->target, "discover")) {
1878 /* If there's no local interface or no IP address, give up. */
1879 local_iface = bridge_get_local_iface(br);
1880 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1884 /* Bring up the local interface. */
1885 netdev = local_iface->netdev;
1886 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1888 /* Configure the IP address and netmask. */
1889 if (!c->local_netmask
1890 || !inet_aton(c->local_netmask, &mask)
1892 mask.s_addr = guess_netmask(ip.s_addr);
1894 if (!netdev_set_in4(netdev, ip, mask)) {
1895 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1896 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1899 /* Configure the default gateway. */
1900 if (c->local_gateway
1901 && inet_aton(c->local_gateway, &gateway)
1902 && gateway.s_addr) {
1903 if (!netdev_add_router(netdev, gateway)) {
1904 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1905 br->name, IP_ARGS(&gateway.s_addr));
1911 bridge_reconfigure_remotes(struct bridge *br,
1912 const struct sockaddr_in *managers,
1915 const char *disable_ib_str, *queue_id_str;
1916 bool disable_in_band = false;
1919 struct ovsrec_controller **controllers;
1920 size_t n_controllers;
1923 struct ofproto_controller *ocs;
1927 /* Check if we should disable in-band control on this bridge. */
1928 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1929 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1930 disable_in_band = true;
1933 /* Set OpenFlow queue ID for in-band control. */
1934 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1935 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1936 ofproto_set_in_band_queue(br->ofproto, queue_id);
1938 if (disable_in_band) {
1939 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1941 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1943 had_primary = ofproto_has_primary_controller(br->ofproto);
1945 n_controllers = bridge_get_controllers(br, &controllers);
1947 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1950 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1951 for (i = 0; i < n_controllers; i++) {
1952 struct ovsrec_controller *c = controllers[i];
1954 if (!strncmp(c->target, "punix:", 6)
1955 || !strncmp(c->target, "unix:", 5)) {
1956 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1958 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1959 * domain sockets and overwriting arbitrary local files. */
1960 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1961 "\"%s\" due to possibility for remote exploit",
1962 dpif_name(br->dpif), c->target);
1966 bridge_configure_local_iface_netdev(br, c);
1967 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1968 if (disable_in_band) {
1969 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1974 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1975 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1978 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1979 ofproto_flush_flows(br->ofproto);
1982 /* If there are no controllers and the bridge is in standalone
1983 * mode, set up a flow that matches every packet and directs
1984 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1985 * switch is in secure mode and we won't pass any traffic until
1986 * a controller has been defined and it tells us to do so. */
1988 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1989 union ofp_action action;
1990 struct cls_rule rule;
1992 memset(&action, 0, sizeof action);
1993 action.type = htons(OFPAT_OUTPUT);
1994 action.output.len = htons(sizeof action);
1995 action.output.port = htons(OFPP_NORMAL);
1996 cls_rule_init_catchall(&rule, 0);
1997 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2002 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2007 for (i = 0; i < br->n_ports; i++) {
2008 struct port *port = br->ports[i];
2009 for (j = 0; j < port->n_ifaces; j++) {
2010 struct iface *iface = port->ifaces[j];
2011 shash_add_once(ifaces, iface->name, iface);
2013 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2014 shash_add_once(ifaces, port->name, NULL);
2019 /* For robustness, in case the administrator moves around datapath ports behind
2020 * our back, we re-check all the datapath port numbers here.
2022 * This function will set the 'dp_ifidx' members of interfaces that have
2023 * disappeared to -1, so only call this function from a context where those
2024 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2025 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2026 * datapath, which doesn't support UINT16_MAX+1 ports. */
2028 bridge_fetch_dp_ifaces(struct bridge *br)
2030 struct dpif_port_dump dump;
2031 struct dpif_port dpif_port;
2034 /* Reset all interface numbers. */
2035 for (i = 0; i < br->n_ports; i++) {
2036 struct port *port = br->ports[i];
2037 for (j = 0; j < port->n_ifaces; j++) {
2038 struct iface *iface = port->ifaces[j];
2039 iface->dp_ifidx = -1;
2042 hmap_clear(&br->ifaces);
2044 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2045 struct iface *iface = iface_lookup(br, dpif_port.name);
2047 if (iface->dp_ifidx >= 0) {
2048 VLOG_WARN("%s reported interface %s twice",
2049 dpif_name(br->dpif), dpif_port.name);
2050 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2051 VLOG_WARN("%s reported interface %"PRIu16" twice",
2052 dpif_name(br->dpif), dpif_port.port_no);
2054 iface->dp_ifidx = dpif_port.port_no;
2055 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2056 hash_int(iface->dp_ifidx, 0));
2059 iface_set_ofport(iface->cfg,
2060 (iface->dp_ifidx >= 0
2061 ? odp_port_to_ofp_port(iface->dp_ifidx)
2067 /* Bridge packet processing functions. */
2070 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2072 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2075 static struct bond_entry *
2076 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
2079 assert(port->bond_mode == BM_SLB);
2080 return &port->bond_hash[bond_hash(mac, vlan)];
2084 bond_choose_iface(const struct port *port)
2086 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2087 size_t i, best_down_slave = -1;
2088 long long next_delay_expiration = LLONG_MAX;
2090 for (i = 0; i < port->n_ifaces; i++) {
2091 struct iface *iface = port->ifaces[i];
2093 if (iface->enabled) {
2095 } else if (iface->delay_expires < next_delay_expiration) {
2096 best_down_slave = i;
2097 next_delay_expiration = iface->delay_expires;
2101 if (best_down_slave != -1) {
2102 struct iface *iface = port->ifaces[best_down_slave];
2104 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2105 "since no other interface is up", iface->name,
2106 iface->delay_expires - time_msec());
2107 bond_enable_slave(iface, true);
2110 return best_down_slave;
2114 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2115 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2117 struct iface *iface;
2119 assert(port->n_ifaces);
2120 if (port->n_ifaces == 1) {
2121 iface = port->ifaces[0];
2122 } else if (port->bond_mode == BM_AB) {
2123 if (port->active_iface < 0) {
2124 *tags |= port->no_ifaces_tag;
2127 iface = port->ifaces[port->active_iface];
2128 } else if (port->bond_mode == BM_SLB){
2129 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2130 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2131 || !port->ifaces[e->iface_idx]->enabled) {
2132 /* XXX select interface properly. The current interface selection
2133 * is only good for testing the rebalancing code. */
2134 e->iface_idx = bond_choose_iface(port);
2135 if (e->iface_idx < 0) {
2136 *tags |= port->no_ifaces_tag;
2139 e->iface_tag = tag_create_random();
2140 ((struct port *) port)->bond_compat_is_stale = true;
2142 *tags |= e->iface_tag;
2143 iface = port->ifaces[e->iface_idx];
2147 *dp_ifidx = iface->dp_ifidx;
2148 *tags |= iface->tag; /* Currently only used for bonding. */
2153 bond_link_status_update(struct iface *iface)
2155 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2156 struct port *port = iface->port;
2157 bool carrier = iface->up;
2159 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2160 /* Nothing to do. */
2163 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2164 iface->name, carrier ? "up" : "down");
2165 if (carrier == iface->enabled) {
2166 iface->delay_expires = LLONG_MAX;
2167 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2168 iface->name, carrier ? "disabled" : "enabled");
2169 } else if (carrier && port->active_iface < 0) {
2170 bond_enable_slave(iface, true);
2171 if (port->updelay) {
2172 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2173 "other interface is up", iface->name, port->updelay);
2176 int delay = carrier ? port->updelay : port->downdelay;
2177 iface->delay_expires = time_msec() + delay;
2180 "interface %s: will be %s if it stays %s for %d ms",
2182 carrier ? "enabled" : "disabled",
2183 carrier ? "up" : "down",
2190 bond_choose_active_iface(struct port *port)
2192 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2194 port->active_iface = bond_choose_iface(port);
2195 port->active_iface_tag = tag_create_random();
2196 if (port->active_iface >= 0) {
2197 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2198 port->name, port->ifaces[port->active_iface]->name);
2200 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2206 bond_enable_slave(struct iface *iface, bool enable)
2208 struct port *port = iface->port;
2209 struct bridge *br = port->bridge;
2211 /* This acts as a recursion check. If the act of disabling a slave
2212 * causes a different slave to be enabled, the flag will allow us to
2213 * skip redundant work when we reenter this function. It must be
2214 * cleared on exit to keep things safe with multiple bonds. */
2215 static bool moving_active_iface = false;
2217 iface->delay_expires = LLONG_MAX;
2218 if (enable == iface->enabled) {
2222 iface->enabled = enable;
2223 if (!iface->enabled) {
2224 VLOG_WARN("interface %s: disabled", iface->name);
2225 ofproto_revalidate(br->ofproto, iface->tag);
2226 if (iface->port_ifidx == port->active_iface) {
2227 ofproto_revalidate(br->ofproto,
2228 port->active_iface_tag);
2230 /* Disabling a slave can lead to another slave being immediately
2231 * enabled if there will be no active slaves but one is waiting
2232 * on an updelay. In this case we do not need to run most of the
2233 * code for the newly enabled slave since there was no period
2234 * without an active slave and it is redundant with the disabling
2236 moving_active_iface = true;
2237 bond_choose_active_iface(port);
2239 bond_send_learning_packets(port);
2241 VLOG_WARN("interface %s: enabled", iface->name);
2242 if (port->active_iface < 0 && !moving_active_iface) {
2243 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2244 bond_choose_active_iface(port);
2245 bond_send_learning_packets(port);
2247 iface->tag = tag_create_random();
2250 moving_active_iface = false;
2251 port->bond_compat_is_stale = true;
2254 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2255 * bond interface. */
2257 bond_update_fake_iface_stats(struct port *port)
2259 struct netdev_stats bond_stats;
2260 struct netdev *bond_dev;
2263 memset(&bond_stats, 0, sizeof bond_stats);
2265 for (i = 0; i < port->n_ifaces; i++) {
2266 struct netdev_stats slave_stats;
2268 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2269 /* XXX: We swap the stats here because they are swapped back when
2270 * reported by the internal device. The reason for this is
2271 * internal devices normally represent packets going into the system
2272 * but when used as fake bond device they represent packets leaving
2273 * the system. We really should do this in the internal device
2274 * itself because changing it here reverses the counts from the
2275 * perspective of the switch. However, the internal device doesn't
2276 * know what type of device it represents so we have to do it here
2278 bond_stats.tx_packets += slave_stats.rx_packets;
2279 bond_stats.tx_bytes += slave_stats.rx_bytes;
2280 bond_stats.rx_packets += slave_stats.tx_packets;
2281 bond_stats.rx_bytes += slave_stats.tx_bytes;
2285 if (!netdev_open_default(port->name, &bond_dev)) {
2286 netdev_set_stats(bond_dev, &bond_stats);
2287 netdev_close(bond_dev);
2292 bond_run(struct bridge *br)
2296 for (i = 0; i < br->n_ports; i++) {
2297 struct port *port = br->ports[i];
2299 if (port->n_ifaces >= 2) {
2302 if (port->monitor) {
2303 assert(!port->miimon);
2305 /* Track carrier going up and down on interfaces. */
2306 while (!netdev_monitor_poll(port->monitor, &devname)) {
2307 struct iface *iface;
2309 iface = port_lookup_iface(port, devname);
2311 bool up = netdev_get_carrier(iface->netdev);
2313 if (up != iface->up) {
2314 port->bond_compat_is_stale = true;
2321 assert(port->miimon);
2323 if (time_msec() >= port->bond_miimon_next_update) {
2324 for (j = 0; j < port->n_ifaces; j++) {
2325 struct iface *iface = port->ifaces[j];
2326 bool up = netdev_get_miimon(iface->netdev);
2328 if (up != iface->up) {
2329 port->bond_compat_is_stale = true;
2333 port->bond_miimon_next_update = time_msec() +
2334 port->bond_miimon_interval;
2338 for (j = 0; j < port->n_ifaces; j++) {
2339 bond_link_status_update(port->ifaces[j]);
2342 for (j = 0; j < port->n_ifaces; j++) {
2343 struct iface *iface = port->ifaces[j];
2344 if (time_msec() >= iface->delay_expires) {
2345 bond_enable_slave(iface, !iface->enabled);
2349 if (port->bond_fake_iface
2350 && time_msec() >= port->bond_next_fake_iface_update) {
2351 bond_update_fake_iface_stats(port);
2352 port->bond_next_fake_iface_update = time_msec() + 1000;
2356 if (port->bond_compat_is_stale) {
2357 port->bond_compat_is_stale = false;
2358 port_update_bond_compat(port);
2364 bond_wait(struct bridge *br)
2368 for (i = 0; i < br->n_ports; i++) {
2369 struct port *port = br->ports[i];
2370 if (port->n_ifaces < 2) {
2374 if (port->monitor) {
2375 netdev_monitor_poll_wait(port->monitor);
2379 poll_timer_wait_until(port->bond_miimon_next_update);
2382 for (j = 0; j < port->n_ifaces; j++) {
2383 struct iface *iface = port->ifaces[j];
2384 if (iface->delay_expires != LLONG_MAX) {
2385 poll_timer_wait_until(iface->delay_expires);
2388 if (port->bond_fake_iface) {
2389 poll_timer_wait_until(port->bond_next_fake_iface_update);
2395 set_dst(struct dst *dst, const struct flow *flow,
2396 const struct port *in_port, const struct port *out_port,
2399 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2400 : in_port->vlan >= 0 ? in_port->vlan
2401 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2402 : vlan_tci_to_vid(flow->vlan_tci));
2403 return choose_output_iface(out_port, flow->dl_src, dst->vlan,
2404 &dst->dp_ifidx, tags);
2408 swap_dst(struct dst *p, struct dst *q)
2410 struct dst tmp = *p;
2415 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2416 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2417 * that we push to the datapath. We could in fact fully sort the array by
2418 * vlan, but in most cases there are at most two different vlan tags so that's
2419 * possibly overkill.) */
2421 partition_dsts(struct dst_set *set, int vlan)
2423 struct dst *first = set->dsts;
2424 struct dst *last = set->dsts + set->n;
2426 while (first != last) {
2428 * - All dsts < first have vlan == 'vlan'.
2429 * - All dsts >= last have vlan != 'vlan'.
2430 * - first < last. */
2431 while (first->vlan == vlan) {
2432 if (++first == last) {
2437 /* Same invariants, plus one additional:
2438 * - first->vlan != vlan.
2440 while (last[-1].vlan != vlan) {
2441 if (--last == first) {
2446 /* Same invariants, plus one additional:
2447 * - last[-1].vlan == vlan.*/
2448 swap_dst(first++, --last);
2453 mirror_mask_ffs(mirror_mask_t mask)
2455 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2460 dst_set_init(struct dst_set *set)
2462 set->dsts = set->builtin;
2464 set->allocated = ARRAY_SIZE(set->builtin);
2468 dst_set_add(struct dst_set *set, const struct dst *dst)
2470 if (set->n >= set->allocated) {
2471 size_t new_allocated;
2472 struct dst *new_dsts;
2474 new_allocated = set->allocated * 2;
2475 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2476 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2480 set->dsts = new_dsts;
2481 set->allocated = new_allocated;
2483 set->dsts[set->n++] = *dst;
2487 dst_set_free(struct dst_set *set)
2489 if (set->dsts != set->builtin) {
2495 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2498 for (i = 0; i < set->n; i++) {
2499 if (set->dsts[i].vlan == test->vlan
2500 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2508 port_trunks_vlan(const struct port *port, uint16_t vlan)
2510 return (port->vlan < 0
2511 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2515 port_includes_vlan(const struct port *port, uint16_t vlan)
2517 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2521 port_is_floodable(const struct port *port)
2525 for (i = 0; i < port->n_ifaces; i++) {
2526 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2527 port->ifaces[i]->dp_ifidx)) {
2535 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2536 const struct port *in_port, const struct port *out_port,
2537 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2539 mirror_mask_t mirrors = in_port->src_mirrors;
2544 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2545 if (flow_vlan == 0) {
2546 flow_vlan = OFP_VLAN_NONE;
2549 if (out_port == FLOOD_PORT) {
2550 for (i = 0; i < br->n_ports; i++) {
2551 struct port *port = br->ports[i];
2553 && port_is_floodable(port)
2554 && port_includes_vlan(port, vlan)
2555 && !port->is_mirror_output_port
2556 && set_dst(&dst, flow, in_port, port, tags)) {
2557 mirrors |= port->dst_mirrors;
2558 dst_set_add(set, &dst);
2561 *nf_output_iface = NF_OUT_FLOOD;
2562 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2563 dst_set_add(set, &dst);
2564 *nf_output_iface = dst.dp_ifidx;
2565 mirrors |= out_port->dst_mirrors;
2569 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2570 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2572 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2573 && !dst_is_duplicate(set, &dst)) {
2574 dst_set_add(set, &dst);
2577 for (i = 0; i < br->n_ports; i++) {
2578 struct port *port = br->ports[i];
2579 if (port_includes_vlan(port, m->out_vlan)
2580 && set_dst(&dst, flow, in_port, port, tags))
2582 if (port->vlan < 0) {
2583 dst.vlan = m->out_vlan;
2585 if (dst_is_duplicate(set, &dst)) {
2589 /* Use the vlan tag on the original flow instead of
2590 * the one passed in the vlan parameter. This ensures
2591 * that we compare the vlan from before any implicit
2592 * tagging tags place. This is necessary because
2593 * dst->vlan is the final vlan, after removing implicit
2595 if (port == in_port && dst.vlan == flow_vlan) {
2596 /* Don't send out input port on same VLAN. */
2599 dst_set_add(set, &dst);
2604 mirrors &= mirrors - 1;
2607 partition_dsts(set, flow_vlan);
2610 static void OVS_UNUSED
2611 print_dsts(const struct dst_set *set)
2615 for (i = 0; i < set->n; i++) {
2616 const struct dst *dst = &set->dsts[i];
2618 printf(">p%"PRIu16, dst->dp_ifidx);
2619 if (dst->vlan != OFP_VLAN_NONE) {
2620 printf("v%"PRIu16, dst->vlan);
2626 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2627 const struct port *in_port, const struct port *out_port,
2628 tag_type *tags, struct ofpbuf *actions,
2629 uint16_t *nf_output_iface)
2636 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2639 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2640 if (cur_vlan == 0) {
2641 cur_vlan = OFP_VLAN_NONE;
2643 for (i = 0; i < set.n; i++) {
2644 const struct dst *dst = &set.dsts[i];
2645 if (dst->vlan != cur_vlan) {
2646 if (dst->vlan == OFP_VLAN_NONE) {
2647 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2650 tci = htons(dst->vlan & VLAN_VID_MASK);
2651 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2652 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2654 cur_vlan = dst->vlan;
2656 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2661 /* Returns the effective vlan of a packet, taking into account both the
2662 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2663 * the packet is untagged and -1 indicates it has an invalid header and
2664 * should be dropped. */
2665 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2666 struct port *in_port, bool have_packet)
2668 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2669 if (in_port->vlan >= 0) {
2671 /* XXX support double tagging? */
2673 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2674 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2675 "packet received on port %s configured with "
2676 "implicit VLAN %"PRIu16,
2677 br->name, vlan, in_port->name, in_port->vlan);
2681 vlan = in_port->vlan;
2683 if (!port_includes_vlan(in_port, vlan)) {
2685 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2686 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2687 "packet received on port %s not configured for "
2689 br->name, vlan, in_port->name, vlan);
2698 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2699 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2700 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2702 is_gratuitous_arp(const struct flow *flow)
2704 return (flow->dl_type == htons(ETH_TYPE_ARP)
2705 && eth_addr_is_broadcast(flow->dl_dst)
2706 && (flow->nw_proto == ARP_OP_REPLY
2707 || (flow->nw_proto == ARP_OP_REQUEST
2708 && flow->nw_src == flow->nw_dst)));
2712 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2713 struct port *in_port)
2715 enum grat_arp_lock_type lock_type;
2718 /* We don't want to learn from gratuitous ARP packets that are reflected
2719 * back over bond slaves so we lock the learning table. */
2720 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2721 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2722 GRAT_ARP_LOCK_CHECK;
2724 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2727 /* The log messages here could actually be useful in debugging,
2728 * so keep the rate limit relatively high. */
2729 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2731 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2732 "on port %s in VLAN %d",
2733 br->name, ETH_ADDR_ARGS(flow->dl_src),
2734 in_port->name, vlan);
2735 ofproto_revalidate(br->ofproto, rev_tag);
2739 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2740 * dropped. Returns true if they may be forwarded, false if they should be
2743 * If 'have_packet' is true, it indicates that the caller is processing a
2744 * received packet. If 'have_packet' is false, then the caller is just
2745 * revalidating an existing flow because configuration has changed. Either
2746 * way, 'have_packet' only affects logging (there is no point in logging errors
2747 * during revalidation).
2749 * Sets '*in_portp' to the input port. This will be a null pointer if
2750 * flow->in_port does not designate a known input port (in which case
2751 * is_admissible() returns false).
2753 * When returning true, sets '*vlanp' to the effective VLAN of the input
2754 * packet, as returned by flow_get_vlan().
2756 * May also add tags to '*tags', although the current implementation only does
2757 * so in one special case.
2760 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2761 tag_type *tags, int *vlanp, struct port **in_portp)
2763 struct iface *in_iface;
2764 struct port *in_port;
2767 /* Find the interface and port structure for the received packet. */
2768 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2770 /* No interface? Something fishy... */
2772 /* Odd. A few possible reasons here:
2774 * - We deleted an interface but there are still a few packets
2775 * queued up from it.
2777 * - Someone externally added an interface (e.g. with "ovs-dpctl
2778 * add-if") that we don't know about.
2780 * - Packet arrived on the local port but the local port is not
2781 * one of our bridge ports.
2783 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2785 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2786 "interface %"PRIu16, br->name, flow->in_port);
2792 *in_portp = in_port = in_iface->port;
2793 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2798 /* Drop frames for reserved multicast addresses. */
2799 if (eth_addr_is_reserved(flow->dl_dst)) {
2803 /* Drop frames on ports reserved for mirroring. */
2804 if (in_port->is_mirror_output_port) {
2806 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2807 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2808 "%s, which is reserved exclusively for mirroring",
2809 br->name, in_port->name);
2814 /* Packets received on bonds need special attention to avoid duplicates. */
2815 if (in_port->n_ifaces > 1) {
2817 bool is_grat_arp_locked;
2819 if (eth_addr_is_multicast(flow->dl_dst)) {
2820 *tags |= in_port->active_iface_tag;
2821 if (in_port->active_iface != in_iface->port_ifidx) {
2822 /* Drop all multicast packets on inactive slaves. */
2827 /* Drop all packets for which we have learned a different input
2828 * port, because we probably sent the packet on one slave and got
2829 * it back on the other. Gratuitous ARP packets are an exception
2830 * to this rule: the host has moved to another switch. The exception
2831 * to the exception is if we locked the learning table to avoid
2832 * reflections on bond slaves. If this is the case, just drop the
2834 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2835 &is_grat_arp_locked);
2836 if (src_idx != -1 && src_idx != in_port->port_idx &&
2837 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2845 /* If the composed actions may be applied to any packet in the given 'flow',
2846 * returns true. Otherwise, the actions should only be applied to 'packet', or
2847 * not at all, if 'packet' was NULL. */
2849 process_flow(struct bridge *br, const struct flow *flow,
2850 const struct ofpbuf *packet, struct ofpbuf *actions,
2851 tag_type *tags, uint16_t *nf_output_iface)
2853 struct port *in_port;
2854 struct port *out_port;
2858 /* Check whether we should drop packets in this flow. */
2859 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2864 /* Learn source MAC (but don't try to learn from revalidation). */
2866 update_learning_table(br, flow, vlan, in_port);
2869 /* Determine output port. */
2870 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2872 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2873 out_port = br->ports[out_port_idx];
2874 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2875 /* If we are revalidating but don't have a learning entry then
2876 * eject the flow. Installing a flow that floods packets opens
2877 * up a window of time where we could learn from a packet reflected
2878 * on a bond and blackhole packets before the learning table is
2879 * updated to reflect the correct port. */
2882 out_port = FLOOD_PORT;
2885 /* Don't send packets out their input ports. */
2886 if (in_port == out_port) {
2892 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2900 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2901 struct ofpbuf *actions, tag_type *tags,
2902 uint16_t *nf_output_iface, void *br_)
2904 struct iface *iface;
2905 struct bridge *br = br_;
2907 COVERAGE_INC(bridge_process_flow);
2909 iface = iface_from_dp_ifidx(br, flow->in_port);
2911 if (cfm_should_process_flow(flow)) {
2912 if (packet && iface->cfm) {
2913 cfm_process_heartbeat(iface->cfm, packet);
2918 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2922 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2923 const struct nlattr *actions,
2925 unsigned long long int n_bytes, void *br_)
2927 struct bridge *br = br_;
2928 const struct nlattr *a;
2929 struct port *in_port;
2934 /* Feed information from the active flows back into the learning table to
2935 * ensure that table is always in sync with what is actually flowing
2936 * through the datapath.
2938 * We test that 'tags' is nonzero to ensure that only flows that include an
2939 * OFPP_NORMAL action are used for learning. This works because
2940 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2941 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2942 update_learning_table(br, flow, vlan, in_port);
2945 /* Account for bond slave utilization. */
2946 if (!br->has_bonded_ports) {
2949 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2950 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
2951 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2952 if (out_port && out_port->n_ifaces >= 2 &&
2953 out_port->bond_mode == BM_SLB) {
2954 uint16_t vlan = (flow->vlan_tci
2955 ? vlan_tci_to_vid(flow->vlan_tci)
2957 struct bond_entry *e = lookup_bond_entry(out_port,
2958 flow->dl_src, vlan);
2959 e->tx_bytes += n_bytes;
2966 bridge_account_checkpoint_ofhook_cb(void *br_)
2968 struct bridge *br = br_;
2972 if (!br->has_bonded_ports) {
2977 for (i = 0; i < br->n_ports; i++) {
2978 struct port *port = br->ports[i];
2979 if (port->n_ifaces > 1 && port->bond_mode == BM_SLB
2980 && now >= port->bond_next_rebalance) {
2981 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2982 bond_rebalance_port(port);
2987 static struct ofhooks bridge_ofhooks = {
2988 bridge_normal_ofhook_cb,
2989 bridge_account_flow_ofhook_cb,
2990 bridge_account_checkpoint_ofhook_cb,
2993 /* Bonding functions. */
2995 /* Statistics for a single interface on a bonded port, used for load-based
2996 * bond rebalancing. */
2997 struct slave_balance {
2998 struct iface *iface; /* The interface. */
2999 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3001 /* All the "bond_entry"s that are assigned to this interface, in order of
3002 * increasing tx_bytes. */
3003 struct bond_entry **hashes;
3008 bond_mode_to_string(enum bond_mode bm) {
3009 static char *bm_slb = "balance-slb";
3010 static char *bm_ab = "active-backup";
3013 case BM_SLB: return bm_slb;
3014 case BM_AB: return bm_ab;
3021 /* Sorts pointers to pointers to bond_entries in ascending order by the
3022 * interface to which they are assigned, and within a single interface in
3023 * ascending order of bytes transmitted. */
3025 compare_bond_entries(const void *a_, const void *b_)
3027 const struct bond_entry *const *ap = a_;
3028 const struct bond_entry *const *bp = b_;
3029 const struct bond_entry *a = *ap;
3030 const struct bond_entry *b = *bp;
3031 if (a->iface_idx != b->iface_idx) {
3032 return a->iface_idx > b->iface_idx ? 1 : -1;
3033 } else if (a->tx_bytes != b->tx_bytes) {
3034 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3040 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3041 * *descending* order by number of bytes transmitted. */
3043 compare_slave_balance(const void *a_, const void *b_)
3045 const struct slave_balance *a = a_;
3046 const struct slave_balance *b = b_;
3047 if (a->iface->enabled != b->iface->enabled) {
3048 return a->iface->enabled ? -1 : 1;
3049 } else if (a->tx_bytes != b->tx_bytes) {
3050 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3057 swap_bals(struct slave_balance *a, struct slave_balance *b)
3059 struct slave_balance tmp = *a;
3064 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3065 * given that 'p' (and only 'p') might be in the wrong location.
3067 * This function invalidates 'p', since it might now be in a different memory
3070 resort_bals(struct slave_balance *p,
3071 struct slave_balance bals[], size_t n_bals)
3074 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3075 swap_bals(p, p - 1);
3077 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3078 swap_bals(p, p + 1);
3084 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3086 if (VLOG_IS_DBG_ENABLED()) {
3087 struct ds ds = DS_EMPTY_INITIALIZER;
3088 const struct slave_balance *b;
3090 for (b = bals; b < bals + n_bals; b++) {
3094 ds_put_char(&ds, ',');
3096 ds_put_format(&ds, " %s %"PRIu64"kB",
3097 b->iface->name, b->tx_bytes / 1024);
3099 if (!b->iface->enabled) {
3100 ds_put_cstr(&ds, " (disabled)");
3102 if (b->n_hashes > 0) {
3103 ds_put_cstr(&ds, " (");
3104 for (i = 0; i < b->n_hashes; i++) {
3105 const struct bond_entry *e = b->hashes[i];
3107 ds_put_cstr(&ds, " + ");
3109 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3110 e - port->bond_hash, e->tx_bytes / 1024);
3112 ds_put_cstr(&ds, ")");
3115 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3120 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3122 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3125 struct bond_entry *hash = from->hashes[hash_idx];
3126 struct port *port = from->iface->port;
3127 uint64_t delta = hash->tx_bytes;
3129 assert(port->bond_mode == BM_SLB);
3131 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3132 "from %s to %s (now carrying %"PRIu64"kB and "
3133 "%"PRIu64"kB load, respectively)",
3134 port->name, delta / 1024, hash - port->bond_hash,
3135 from->iface->name, to->iface->name,
3136 (from->tx_bytes - delta) / 1024,
3137 (to->tx_bytes + delta) / 1024);
3139 /* Delete element from from->hashes.
3141 * We don't bother to add the element to to->hashes because not only would
3142 * it require more work, the only purpose it would be to allow that hash to
3143 * be migrated to another slave in this rebalancing run, and there is no
3144 * point in doing that. */
3145 if (hash_idx == 0) {
3148 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3149 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3153 /* Shift load away from 'from' to 'to'. */
3154 from->tx_bytes -= delta;
3155 to->tx_bytes += delta;
3157 /* Arrange for flows to be revalidated. */
3158 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3159 hash->iface_idx = to->iface->port_ifidx;
3160 hash->iface_tag = tag_create_random();
3164 bond_rebalance_port(struct port *port)
3166 struct slave_balance *bals;
3168 struct bond_entry *hashes[BOND_MASK + 1];
3169 struct slave_balance *b, *from, *to;
3170 struct bond_entry *e;
3173 assert(port->bond_mode == BM_SLB);
3175 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3176 * descending order of tx_bytes, so that bals[0] represents the most
3177 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3180 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3181 * array for each slave_balance structure, we sort our local array of
3182 * hashes in order by slave, so that all of the hashes for a given slave
3183 * become contiguous in memory, and then we point each 'hashes' members of
3184 * a slave_balance structure to the start of a contiguous group. */
3185 n_bals = port->n_ifaces;
3186 bals = xmalloc(n_bals * sizeof *bals);
3187 for (b = bals; b < &bals[n_bals]; b++) {
3188 b->iface = port->ifaces[b - bals];
3193 for (i = 0; i <= BOND_MASK; i++) {
3194 hashes[i] = &port->bond_hash[i];
3196 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3197 for (i = 0; i <= BOND_MASK; i++) {
3199 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3200 b = &bals[e->iface_idx];
3201 b->tx_bytes += e->tx_bytes;
3203 b->hashes = &hashes[i];
3208 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3209 log_bals(bals, n_bals, port);
3211 /* Discard slaves that aren't enabled (which were sorted to the back of the
3212 * array earlier). */
3213 while (!bals[n_bals - 1].iface->enabled) {
3220 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3221 to = &bals[n_bals - 1];
3222 for (from = bals; from < to; ) {
3223 uint64_t overload = from->tx_bytes - to->tx_bytes;
3224 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3225 /* The extra load on 'from' (and all less-loaded slaves), compared
3226 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3227 * it is less than ~1Mbps. No point in rebalancing. */
3229 } else if (from->n_hashes == 1) {
3230 /* 'from' only carries a single MAC hash, so we can't shift any
3231 * load away from it, even though we want to. */
3234 /* 'from' is carrying significantly more load than 'to', and that
3235 * load is split across at least two different hashes. Pick a hash
3236 * to migrate to 'to' (the least-loaded slave), given that doing so
3237 * must decrease the ratio of the load on the two slaves by at
3240 * The sort order we use means that we prefer to shift away the
3241 * smallest hashes instead of the biggest ones. There is little
3242 * reason behind this decision; we could use the opposite sort
3243 * order to shift away big hashes ahead of small ones. */
3246 for (i = 0; i < from->n_hashes; i++) {
3247 double old_ratio, new_ratio;
3248 uint64_t delta = from->hashes[i]->tx_bytes;
3250 if (delta == 0 || from->tx_bytes - delta == 0) {
3251 /* Pointless move. */
3255 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3257 if (to->tx_bytes == 0) {
3258 /* Nothing on the new slave, move it. */
3262 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3263 new_ratio = (double)(from->tx_bytes - delta) /
3264 (to->tx_bytes + delta);
3266 if (new_ratio == 0) {
3267 /* Should already be covered but check to prevent division
3272 if (new_ratio < 1) {
3273 new_ratio = 1 / new_ratio;
3276 if (old_ratio - new_ratio > 0.1) {
3277 /* Would decrease the ratio, move it. */
3281 if (i < from->n_hashes) {
3282 bond_shift_load(from, to, i);
3283 port->bond_compat_is_stale = true;
3285 /* If the result of the migration changed the relative order of
3286 * 'from' and 'to' swap them back to maintain invariants. */
3287 if (order_swapped) {
3288 swap_bals(from, to);
3291 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3292 * point to different slave_balance structures. It is only
3293 * valid to do these two operations in a row at all because we
3294 * know that 'from' will not move past 'to' and vice versa. */
3295 resort_bals(from, bals, n_bals);
3296 resort_bals(to, bals, n_bals);
3303 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3304 * historical data to decay to <1% in 7 rebalancing runs. */
3305 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3314 bond_send_learning_packets(struct port *port)
3316 struct bridge *br = port->bridge;
3317 struct mac_entry *e;
3318 struct ofpbuf packet;
3319 int error, n_packets, n_errors;
3321 if (!port->n_ifaces || port->active_iface < 0) {
3325 ofpbuf_init(&packet, 128);
3326 error = n_packets = n_errors = 0;
3327 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3328 union ofp_action actions[2], *a;
3334 if (e->port == port->port_idx
3335 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3339 /* Compose actions. */
3340 memset(actions, 0, sizeof actions);
3343 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3344 a->vlan_vid.len = htons(sizeof *a);
3345 a->vlan_vid.vlan_vid = htons(e->vlan);
3348 a->output.type = htons(OFPAT_OUTPUT);
3349 a->output.len = htons(sizeof *a);
3350 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3355 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3357 flow_extract(&packet, 0, ODPP_NONE, &flow);
3358 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3365 ofpbuf_uninit(&packet);
3368 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3369 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3370 "packets, last error was: %s",
3371 port->name, n_errors, n_packets, strerror(error));
3373 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3374 port->name, n_packets);
3378 /* Bonding unixctl user interface functions. */
3381 bond_unixctl_list(struct unixctl_conn *conn,
3382 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3384 struct ds ds = DS_EMPTY_INITIALIZER;
3385 const struct bridge *br;
3387 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3389 LIST_FOR_EACH (br, node, &all_bridges) {
3392 for (i = 0; i < br->n_ports; i++) {
3393 const struct port *port = br->ports[i];
3394 if (port->n_ifaces > 1) {
3397 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3398 bond_mode_to_string(port->bond_mode));
3399 for (j = 0; j < port->n_ifaces; j++) {
3400 const struct iface *iface = port->ifaces[j];
3402 ds_put_cstr(&ds, ", ");
3404 ds_put_cstr(&ds, iface->name);
3406 ds_put_char(&ds, '\n');
3410 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3414 static struct port *
3415 bond_find(const char *name)
3417 const struct bridge *br;
3419 LIST_FOR_EACH (br, node, &all_bridges) {
3422 for (i = 0; i < br->n_ports; i++) {
3423 struct port *port = br->ports[i];
3424 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3433 bond_unixctl_show(struct unixctl_conn *conn,
3434 const char *args, void *aux OVS_UNUSED)
3436 struct ds ds = DS_EMPTY_INITIALIZER;
3437 const struct port *port;
3440 port = bond_find(args);
3442 unixctl_command_reply(conn, 501, "no such bond");
3446 ds_put_format(&ds, "bond_mode: %s\n",
3447 bond_mode_to_string(port->bond_mode));
3448 ds_put_format(&ds, "bond-detect-mode: %s\n",
3449 port->miimon ? "miimon" : "carrier");
3452 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3453 port->bond_miimon_interval);
3456 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3457 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3459 if (port->bond_mode == BM_SLB) {
3460 ds_put_format(&ds, "next rebalance: %lld ms\n",
3461 port->bond_next_rebalance - time_msec());
3464 for (j = 0; j < port->n_ifaces; j++) {
3465 const struct iface *iface = port->ifaces[j];
3466 struct bond_entry *be;
3469 ds_put_format(&ds, "slave %s: %s\n",
3470 iface->name, iface->enabled ? "enabled" : "disabled");
3471 if (j == port->active_iface) {
3472 ds_put_cstr(&ds, "\tactive slave\n");
3474 if (iface->delay_expires != LLONG_MAX) {
3475 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3476 iface->enabled ? "downdelay" : "updelay",
3477 iface->delay_expires - time_msec());
3480 if (port->bond_mode != BM_SLB) {
3485 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3486 int hash = be - port->bond_hash;
3487 struct mac_entry *me;
3489 if (be->iface_idx != j) {
3493 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3494 hash, be->tx_bytes / 1024);
3497 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3500 if (bond_hash(me->mac, me->vlan) == hash
3501 && me->port != port->port_idx
3502 && choose_output_iface(port, me->mac, me->vlan,
3504 && dp_ifidx == iface->dp_ifidx)
3506 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3507 ETH_ADDR_ARGS(me->mac));
3512 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3517 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3518 void *aux OVS_UNUSED)
3520 char *args = (char *) args_;
3521 char *save_ptr = NULL;
3522 char *bond_s, *hash_s, *slave_s;
3524 struct iface *iface;
3525 struct bond_entry *entry;
3528 bond_s = strtok_r(args, " ", &save_ptr);
3529 hash_s = strtok_r(NULL, " ", &save_ptr);
3530 slave_s = strtok_r(NULL, " ", &save_ptr);
3532 unixctl_command_reply(conn, 501,
3533 "usage: bond/migrate BOND HASH SLAVE");
3537 port = bond_find(bond_s);
3539 unixctl_command_reply(conn, 501, "no such bond");
3543 if (port->bond_mode != BM_SLB) {
3544 unixctl_command_reply(conn, 501, "not an SLB bond");
3548 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3549 hash = atoi(hash_s) & BOND_MASK;
3551 unixctl_command_reply(conn, 501, "bad hash");
3555 iface = port_lookup_iface(port, slave_s);
3557 unixctl_command_reply(conn, 501, "no such slave");
3561 if (!iface->enabled) {
3562 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3566 entry = &port->bond_hash[hash];
3567 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3568 entry->iface_idx = iface->port_ifidx;
3569 entry->iface_tag = tag_create_random();
3570 port->bond_compat_is_stale = true;
3571 unixctl_command_reply(conn, 200, "migrated");
3575 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3576 void *aux OVS_UNUSED)
3578 char *args = (char *) args_;
3579 char *save_ptr = NULL;
3580 char *bond_s, *slave_s;
3582 struct iface *iface;
3584 bond_s = strtok_r(args, " ", &save_ptr);
3585 slave_s = strtok_r(NULL, " ", &save_ptr);
3587 unixctl_command_reply(conn, 501,
3588 "usage: bond/set-active-slave BOND SLAVE");
3592 port = bond_find(bond_s);
3594 unixctl_command_reply(conn, 501, "no such bond");
3598 iface = port_lookup_iface(port, slave_s);
3600 unixctl_command_reply(conn, 501, "no such slave");
3604 if (!iface->enabled) {
3605 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3609 if (port->active_iface != iface->port_ifidx) {
3610 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3611 port->active_iface = iface->port_ifidx;
3612 port->active_iface_tag = tag_create_random();
3613 VLOG_INFO("port %s: active interface is now %s",
3614 port->name, iface->name);
3615 bond_send_learning_packets(port);
3616 unixctl_command_reply(conn, 200, "done");
3618 unixctl_command_reply(conn, 200, "no change");
3623 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3625 char *args = (char *) args_;
3626 char *save_ptr = NULL;
3627 char *bond_s, *slave_s;
3629 struct iface *iface;
3631 bond_s = strtok_r(args, " ", &save_ptr);
3632 slave_s = strtok_r(NULL, " ", &save_ptr);
3634 unixctl_command_reply(conn, 501,
3635 "usage: bond/enable/disable-slave BOND SLAVE");
3639 port = bond_find(bond_s);
3641 unixctl_command_reply(conn, 501, "no such bond");
3645 iface = port_lookup_iface(port, slave_s);
3647 unixctl_command_reply(conn, 501, "no such slave");
3651 bond_enable_slave(iface, enable);
3652 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3656 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3657 void *aux OVS_UNUSED)
3659 enable_slave(conn, args, true);
3663 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3664 void *aux OVS_UNUSED)
3666 enable_slave(conn, args, false);
3670 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3671 void *aux OVS_UNUSED)
3673 char *args = (char *) args_;
3674 uint8_t mac[ETH_ADDR_LEN];
3678 char *mac_s, *vlan_s;
3679 char *save_ptr = NULL;
3681 mac_s = strtok_r(args, " ", &save_ptr);
3682 vlan_s = strtok_r(NULL, " ", &save_ptr);
3685 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3686 unixctl_command_reply(conn, 501, "invalid vlan");
3690 vlan = OFP_VLAN_NONE;
3693 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3694 == ETH_ADDR_SCAN_COUNT) {
3695 hash = bond_hash(mac, vlan);
3697 hash_cstr = xasprintf("%u", hash);
3698 unixctl_command_reply(conn, 200, hash_cstr);
3701 unixctl_command_reply(conn, 501, "invalid mac");
3708 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3709 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3710 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3711 unixctl_command_register("bond/set-active-slave",
3712 bond_unixctl_set_active_slave, NULL);
3713 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3715 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3717 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3720 /* Port functions. */
3722 static struct port *
3723 port_create(struct bridge *br, const char *name)
3727 port = xzalloc(sizeof *port);
3729 port->port_idx = br->n_ports;
3731 port->trunks = NULL;
3732 port->name = xstrdup(name);
3733 port->active_iface = -1;
3735 if (br->n_ports >= br->allocated_ports) {
3736 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3739 br->ports[br->n_ports++] = port;
3740 shash_add_assert(&br->port_by_name, port->name, port);
3742 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3749 get_port_other_config(const struct ovsrec_port *port, const char *key,
3750 const char *default_value)
3754 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3756 return value ? value : default_value;
3760 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3762 struct shash new_ifaces;
3765 /* Collect list of new interfaces. */
3766 shash_init(&new_ifaces);
3767 for (i = 0; i < cfg->n_interfaces; i++) {
3768 const char *name = cfg->interfaces[i]->name;
3769 shash_add_once(&new_ifaces, name, NULL);
3772 /* Get rid of deleted interfaces. */
3773 for (i = 0; i < port->n_ifaces; ) {
3774 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3775 iface_destroy(port->ifaces[i]);
3781 shash_destroy(&new_ifaces);
3785 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3787 const char *detect_mode;
3788 struct shash new_ifaces;
3789 long long int next_rebalance, miimon_next_update;
3790 unsigned long *trunks;
3796 /* Update settings. */
3797 port->updelay = cfg->bond_updelay;
3798 if (port->updelay < 0) {
3801 port->downdelay = cfg->bond_downdelay;
3802 if (port->downdelay < 0) {
3803 port->downdelay = 0;
3805 port->bond_rebalance_interval = atoi(
3806 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3807 if (port->bond_rebalance_interval < 1000) {
3808 port->bond_rebalance_interval = 1000;
3810 next_rebalance = time_msec() + port->bond_rebalance_interval;
3811 if (port->bond_next_rebalance > next_rebalance) {
3812 port->bond_next_rebalance = next_rebalance;
3815 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
3818 if (!strcmp(detect_mode, "carrier")) {
3819 port->miimon = false;
3820 } else if (!strcmp(detect_mode, "miimon")) {
3821 port->miimon = true;
3823 port->miimon = false;
3824 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
3825 "carrier", port->name, detect_mode);
3828 port->bond_miimon_interval = atoi(
3829 get_port_other_config(cfg, "bond-miimon-interval", "200"));
3830 if (port->bond_miimon_interval < 100) {
3831 port->bond_miimon_interval = 100;
3833 miimon_next_update = time_msec() + port->bond_miimon_interval;
3834 if (port->bond_miimon_next_update > miimon_next_update) {
3835 port->bond_miimon_next_update = miimon_next_update;
3838 if (!port->cfg->bond_mode ||
3839 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
3840 port->bond_mode = BM_SLB;
3841 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
3842 port->bond_mode = BM_AB;
3844 port->bond_mode = BM_SLB;
3845 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
3846 port->name, port->cfg->bond_mode,
3847 bond_mode_to_string(port->bond_mode));
3850 /* Add new interfaces and update 'cfg' member of existing ones. */
3851 shash_init(&new_ifaces);
3852 for (i = 0; i < cfg->n_interfaces; i++) {
3853 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3854 struct iface *iface;
3856 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3857 VLOG_WARN("port %s: %s specified twice as port interface",
3858 port->name, if_cfg->name);
3859 iface_set_ofport(if_cfg, -1);
3863 iface = iface_lookup(port->bridge, if_cfg->name);
3865 if (iface->port != port) {
3866 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3868 port->bridge->name, if_cfg->name, iface->port->name);
3871 iface->cfg = if_cfg;
3873 iface = iface_create(port, if_cfg);
3876 /* Determine interface type. The local port always has type
3877 * "internal". Other ports take their type from the database and
3878 * default to "system" if none is specified. */
3879 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3880 : if_cfg->type[0] ? if_cfg->type
3883 shash_destroy(&new_ifaces);
3888 if (port->n_ifaces < 2) {
3890 if (vlan >= 0 && vlan <= 4095) {
3891 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3896 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3897 * they even work as-is. But they have not been tested. */
3898 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3902 if (port->vlan != vlan) {
3904 bridge_flush(port->bridge);
3907 /* Get trunked VLANs. */
3909 if (vlan < 0 && cfg->n_trunks) {
3912 trunks = bitmap_allocate(4096);
3914 for (i = 0; i < cfg->n_trunks; i++) {
3915 int trunk = cfg->trunks[i];
3917 bitmap_set1(trunks, trunk);
3923 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3924 port->name, cfg->n_trunks);
3926 if (n_errors == cfg->n_trunks) {
3927 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3929 bitmap_free(trunks);
3932 } else if (vlan >= 0 && cfg->n_trunks) {
3933 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3937 ? port->trunks != NULL
3938 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3939 bridge_flush(port->bridge);
3941 bitmap_free(port->trunks);
3942 port->trunks = trunks;
3946 port_destroy(struct port *port)
3949 struct bridge *br = port->bridge;
3953 proc_net_compat_update_vlan(port->name, NULL, 0);
3954 proc_net_compat_update_bond(port->name, NULL);
3956 for (i = 0; i < MAX_MIRRORS; i++) {
3957 struct mirror *m = br->mirrors[i];
3958 if (m && m->out_port == port) {
3963 while (port->n_ifaces > 0) {
3964 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3967 shash_find_and_delete_assert(&br->port_by_name, port->name);
3969 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3970 del->port_idx = port->port_idx;
3972 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3974 netdev_monitor_destroy(port->monitor);
3976 bitmap_free(port->trunks);
3983 static struct port *
3984 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3986 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3987 return iface ? iface->port : NULL;
3990 static struct port *
3991 port_lookup(const struct bridge *br, const char *name)
3993 return shash_find_data(&br->port_by_name, name);
3996 static struct iface *
3997 port_lookup_iface(const struct port *port, const char *name)
3999 struct iface *iface = iface_lookup(port->bridge, name);
4000 return iface && iface->port == port ? iface : NULL;
4004 port_update_bonding(struct port *port)
4006 if (port->monitor) {
4007 netdev_monitor_destroy(port->monitor);
4008 port->monitor = NULL;
4010 if (port->n_ifaces < 2) {
4011 /* Not a bonded port. */
4012 if (port->bond_hash) {
4013 free(port->bond_hash);
4014 port->bond_hash = NULL;
4015 port->bond_compat_is_stale = true;
4018 port->bond_fake_iface = false;
4022 if (port->bond_mode == BM_SLB && !port->bond_hash) {
4023 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4024 for (i = 0; i <= BOND_MASK; i++) {
4025 struct bond_entry *e = &port->bond_hash[i];
4029 port->no_ifaces_tag = tag_create_random();
4030 bond_choose_active_iface(port);
4031 port->bond_next_rebalance
4032 = time_msec() + port->bond_rebalance_interval;
4034 if (port->cfg->bond_fake_iface) {
4035 port->bond_next_fake_iface_update = time_msec();
4037 } else if (port->bond_mode != BM_SLB) {
4038 free(port->bond_hash);
4039 port->bond_hash = NULL;
4041 port->bond_compat_is_stale = true;
4042 port->bond_fake_iface = port->cfg->bond_fake_iface;
4044 if (!port->miimon) {
4045 port->monitor = netdev_monitor_create();
4046 for (i = 0; i < port->n_ifaces; i++) {
4047 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4054 port_update_bond_compat(struct port *port)
4056 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
4057 struct compat_bond bond;
4060 if (port->n_ifaces < 2 || port->bond_mode != BM_SLB) {
4061 proc_net_compat_update_bond(port->name, NULL);
4066 bond.updelay = port->updelay;
4067 bond.downdelay = port->downdelay;
4070 bond.hashes = compat_hashes;
4071 if (port->bond_hash) {
4072 const struct bond_entry *e;
4073 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4074 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
4075 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
4076 cbh->hash = e - port->bond_hash;
4077 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
4082 bond.n_slaves = port->n_ifaces;
4083 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
4084 for (i = 0; i < port->n_ifaces; i++) {
4085 struct iface *iface = port->ifaces[i];
4086 struct compat_bond_slave *slave = &bond.slaves[i];
4087 slave->name = iface->name;
4089 /* We need to make the same determination as the Linux bonding
4090 * code to determine whether a slave should be consider "up".
4091 * The Linux function bond_miimon_inspect() supports four
4092 * BOND_LINK_* states:
4094 * - BOND_LINK_UP: carrier detected, updelay has passed.
4095 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
4096 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
4097 * - BOND_LINK_BACK: carrier detected, updelay in progress.
4099 * The function bond_info_show_slave() only considers BOND_LINK_UP
4100 * to be "up" and anything else to be "down".
4102 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
4106 netdev_get_etheraddr(iface->netdev, slave->mac);
4109 if (port->bond_fake_iface) {
4110 struct netdev *bond_netdev;
4112 if (!netdev_open_default(port->name, &bond_netdev)) {
4114 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
4116 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
4118 netdev_close(bond_netdev);
4122 proc_net_compat_update_bond(port->name, &bond);
4127 port_update_vlan_compat(struct port *port)
4129 struct bridge *br = port->bridge;
4130 char *vlandev_name = NULL;
4132 if (port->vlan > 0) {
4133 /* Figure out the name that the VLAN device should actually have, if it
4134 * existed. This takes some work because the VLAN device would not
4135 * have port->name in its name; rather, it would have the trunk port's
4136 * name, and 'port' would be attached to a bridge that also had the
4137 * VLAN device one of its ports. So we need to find a trunk port that
4138 * includes port->vlan.
4140 * There might be more than one candidate. This doesn't happen on
4141 * XenServer, so if it happens we just pick the first choice in
4142 * alphabetical order instead of creating multiple VLAN devices. */
4144 for (i = 0; i < br->n_ports; i++) {
4145 struct port *p = br->ports[i];
4146 if (port_trunks_vlan(p, port->vlan)
4148 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
4150 uint8_t ea[ETH_ADDR_LEN];
4151 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
4152 if (!eth_addr_is_multicast(ea) &&
4153 !eth_addr_is_reserved(ea) &&
4154 !eth_addr_is_zero(ea)) {
4155 vlandev_name = p->name;
4160 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
4163 /* Interface functions. */
4166 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4169 union ofp_action action;
4171 memset(&action, 0, sizeof action);
4172 action.output.type = htons(OFPAT_OUTPUT);
4173 action.output.len = htons(sizeof action);
4174 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4176 flow_extract(packet, 0, ODPP_NONE, &flow);
4178 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4180 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4181 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4185 static struct iface *
4186 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4188 struct bridge *br = port->bridge;
4189 struct iface *iface;
4190 char *name = if_cfg->name;
4192 iface = xzalloc(sizeof *iface);
4194 iface->port_ifidx = port->n_ifaces;
4195 iface->name = xstrdup(name);
4196 iface->dp_ifidx = -1;
4197 iface->tag = tag_create_random();
4198 iface->delay_expires = LLONG_MAX;
4199 iface->netdev = NULL;
4200 iface->cfg = if_cfg;
4202 shash_add_assert(&br->iface_by_name, iface->name, iface);
4204 if (port->n_ifaces >= port->allocated_ifaces) {
4205 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4206 sizeof *port->ifaces);
4208 port->ifaces[port->n_ifaces++] = iface;
4209 if (port->n_ifaces > 1) {
4210 br->has_bonded_ports = true;
4213 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4221 iface_destroy(struct iface *iface)
4224 struct port *port = iface->port;
4225 struct bridge *br = port->bridge;
4226 bool del_active = port->active_iface == iface->port_ifidx;
4229 if (port->monitor) {
4230 netdev_monitor_remove(port->monitor, iface->netdev);
4233 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4235 if (iface->dp_ifidx >= 0) {
4236 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4239 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4240 del->port_ifidx = iface->port_ifidx;
4242 netdev_close(iface->netdev);
4245 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4246 bond_choose_active_iface(port);
4247 bond_send_learning_packets(port);
4250 cfm_destroy(iface->cfm);
4255 bridge_flush(port->bridge);
4259 static struct iface *
4260 iface_lookup(const struct bridge *br, const char *name)
4262 return shash_find_data(&br->iface_by_name, name);
4265 static struct iface *
4266 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4268 struct iface *iface;
4270 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4271 hash_int(dp_ifidx, 0), &br->ifaces) {
4272 if (iface->dp_ifidx == dp_ifidx) {
4279 /* Set Ethernet address of 'iface', if one is specified in the configuration
4282 iface_set_mac(struct iface *iface)
4284 uint8_t ea[ETH_ADDR_LEN];
4286 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4287 if (eth_addr_is_multicast(ea)) {
4288 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4290 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4291 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4292 iface->name, iface->name);
4294 int error = netdev_set_etheraddr(iface->netdev, ea);
4296 VLOG_ERR("interface %s: setting MAC failed (%s)",
4297 iface->name, strerror(error));
4303 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4305 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4308 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4312 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4314 * The value strings in '*shash' are taken directly from values[], not copied,
4315 * so the caller should not modify or free them. */
4317 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4318 struct shash *shash)
4323 for (i = 0; i < n; i++) {
4324 shash_add(shash, keys[i], values[i]);
4328 /* Creates 'keys' and 'values' arrays from 'shash'.
4330 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4331 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4332 * are populated with with strings taken directly from 'shash' and thus have
4333 * the same ownership of the key-value pairs in shash.
4336 shash_to_ovs_idl_map(struct shash *shash,
4337 char ***keys, char ***values, size_t *n)
4341 struct shash_node *sn;
4343 count = shash_count(shash);
4345 k = xmalloc(count * sizeof *k);
4346 v = xmalloc(count * sizeof *v);
4349 SHASH_FOR_EACH(sn, shash) {
4360 struct iface_delete_queues_cbdata {
4361 struct netdev *netdev;
4362 const struct ovsdb_datum *queues;
4366 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4368 union ovsdb_atom atom;
4370 atom.integer = target;
4371 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4375 iface_delete_queues(unsigned int queue_id,
4376 const struct shash *details OVS_UNUSED, void *cbdata_)
4378 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4380 if (!queue_ids_include(cbdata->queues, queue_id)) {
4381 netdev_delete_queue(cbdata->netdev, queue_id);
4386 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4388 if (!qos || qos->type[0] == '\0') {
4389 netdev_set_qos(iface->netdev, NULL, NULL);
4391 struct iface_delete_queues_cbdata cbdata;
4392 struct shash details;
4395 /* Configure top-level Qos for 'iface'. */
4396 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4397 qos->n_other_config, &details);
4398 netdev_set_qos(iface->netdev, qos->type, &details);
4399 shash_destroy(&details);
4401 /* Deconfigure queues that were deleted. */
4402 cbdata.netdev = iface->netdev;
4403 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4405 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4407 /* Configure queues for 'iface'. */
4408 for (i = 0; i < qos->n_queues; i++) {
4409 const struct ovsrec_queue *queue = qos->value_queues[i];
4410 unsigned int queue_id = qos->key_queues[i];
4412 shash_from_ovs_idl_map(queue->key_other_config,
4413 queue->value_other_config,
4414 queue->n_other_config, &details);
4415 netdev_set_queue(iface->netdev, queue_id, &details);
4416 shash_destroy(&details);
4422 iface_update_cfm(struct iface *iface)
4426 uint16_t *remote_mps;
4427 struct ovsrec_monitor *mon;
4428 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4430 mon = iface->cfg->monitor;
4436 if (netdev_get_etheraddr(iface->netdev, ea)) {
4437 VLOG_WARN("interface %s: Failed to get ethernet address. "
4438 "Skipping Monitor.", iface->name);
4442 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4443 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4448 iface->cfm = cfm_create();
4452 cfm->mpid = mon->mpid;
4453 cfm->interval = mon->interval ? *mon->interval : 1000;
4455 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4456 memcpy(cfm->maid, maid, sizeof cfm->maid);
4458 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4459 for(i = 0; i < mon->n_remote_mps; i++) {
4460 remote_mps[i] = mon->remote_mps[i]->mpid;
4462 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4465 if (!cfm_configure(iface->cfm)) {
4466 cfm_destroy(iface->cfm);
4471 /* Port mirroring. */
4473 static struct mirror *
4474 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4478 for (i = 0; i < MAX_MIRRORS; i++) {
4479 struct mirror *m = br->mirrors[i];
4480 if (m && uuid_equals(uuid, &m->uuid)) {
4488 mirror_reconfigure(struct bridge *br)
4490 unsigned long *rspan_vlans;
4493 /* Get rid of deleted mirrors. */
4494 for (i = 0; i < MAX_MIRRORS; i++) {
4495 struct mirror *m = br->mirrors[i];
4497 const struct ovsdb_datum *mc;
4498 union ovsdb_atom atom;
4500 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4501 atom.uuid = br->mirrors[i]->uuid;
4502 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4508 /* Add new mirrors and reconfigure existing ones. */
4509 for (i = 0; i < br->cfg->n_mirrors; i++) {
4510 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4511 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4513 mirror_reconfigure_one(m, cfg);
4515 mirror_create(br, cfg);
4519 /* Update port reserved status. */
4520 for (i = 0; i < br->n_ports; i++) {
4521 br->ports[i]->is_mirror_output_port = false;
4523 for (i = 0; i < MAX_MIRRORS; i++) {
4524 struct mirror *m = br->mirrors[i];
4525 if (m && m->out_port) {
4526 m->out_port->is_mirror_output_port = true;
4530 /* Update flooded vlans (for RSPAN). */
4532 if (br->cfg->n_flood_vlans) {
4533 rspan_vlans = bitmap_allocate(4096);
4535 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4536 int64_t vlan = br->cfg->flood_vlans[i];
4537 if (vlan >= 0 && vlan < 4096) {
4538 bitmap_set1(rspan_vlans, vlan);
4539 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4542 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4547 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4553 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4558 for (i = 0; ; i++) {
4559 if (i >= MAX_MIRRORS) {
4560 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4561 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4564 if (!br->mirrors[i]) {
4569 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4572 br->mirrors[i] = m = xzalloc(sizeof *m);
4575 m->name = xstrdup(cfg->name);
4576 shash_init(&m->src_ports);
4577 shash_init(&m->dst_ports);
4583 mirror_reconfigure_one(m, cfg);
4587 mirror_destroy(struct mirror *m)
4590 struct bridge *br = m->bridge;
4593 for (i = 0; i < br->n_ports; i++) {
4594 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4595 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4598 shash_destroy(&m->src_ports);
4599 shash_destroy(&m->dst_ports);
4602 m->bridge->mirrors[m->idx] = NULL;
4611 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4612 struct shash *names)
4616 for (i = 0; i < n_ports; i++) {
4617 const char *name = ports[i]->name;
4618 if (port_lookup(m->bridge, name)) {
4619 shash_add_once(names, name, NULL);
4621 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4622 "port %s", m->bridge->name, m->name, name);
4628 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4634 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4636 for (i = 0; i < cfg->n_select_vlan; i++) {
4637 int64_t vlan = cfg->select_vlan[i];
4638 if (vlan < 0 || vlan > 4095) {
4639 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4640 m->bridge->name, m->name, vlan);
4642 (*vlans)[n_vlans++] = vlan;
4649 vlan_is_mirrored(const struct mirror *m, int vlan)
4653 for (i = 0; i < m->n_vlans; i++) {
4654 if (m->vlans[i] == vlan) {
4662 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4666 for (i = 0; i < m->n_vlans; i++) {
4667 if (port_trunks_vlan(p, m->vlans[i])) {
4675 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4677 struct shash src_ports, dst_ports;
4678 mirror_mask_t mirror_bit;
4679 struct port *out_port;
4686 if (strcmp(cfg->name, m->name)) {
4688 m->name = xstrdup(cfg->name);
4691 /* Get output port. */
4692 if (cfg->output_port) {
4693 out_port = port_lookup(m->bridge, cfg->output_port->name);
4695 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4696 m->bridge->name, m->name);
4702 if (cfg->output_vlan) {
4703 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4704 "output vlan; ignoring output vlan",
4705 m->bridge->name, m->name);
4707 } else if (cfg->output_vlan) {
4709 out_vlan = *cfg->output_vlan;
4711 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4712 m->bridge->name, m->name);
4717 shash_init(&src_ports);
4718 shash_init(&dst_ports);
4719 if (cfg->select_all) {
4720 for (i = 0; i < m->bridge->n_ports; i++) {
4721 const char *name = m->bridge->ports[i]->name;
4722 shash_add_once(&src_ports, name, NULL);
4723 shash_add_once(&dst_ports, name, NULL);
4728 /* Get ports, and drop duplicates and ports that don't exist. */
4729 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4731 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4734 /* Get all the vlans, and drop duplicate and invalid vlans. */
4735 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4738 /* Update mirror data. */
4739 if (!shash_equal_keys(&m->src_ports, &src_ports)
4740 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4741 || m->n_vlans != n_vlans
4742 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4743 || m->out_port != out_port
4744 || m->out_vlan != out_vlan) {
4745 bridge_flush(m->bridge);
4747 shash_swap(&m->src_ports, &src_ports);
4748 shash_swap(&m->dst_ports, &dst_ports);
4751 m->n_vlans = n_vlans;
4752 m->out_port = out_port;
4753 m->out_vlan = out_vlan;
4756 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4757 for (i = 0; i < m->bridge->n_ports; i++) {
4758 struct port *port = m->bridge->ports[i];
4760 if (shash_find(&m->src_ports, port->name)
4763 ? port_trunks_any_mirrored_vlan(m, port)
4764 : vlan_is_mirrored(m, port->vlan)))) {
4765 port->src_mirrors |= mirror_bit;
4767 port->src_mirrors &= ~mirror_bit;
4770 if (shash_find(&m->dst_ports, port->name)) {
4771 port->dst_mirrors |= mirror_bit;
4773 port->dst_mirrors &= ~mirror_bit;
4778 shash_destroy(&src_ports);
4779 shash_destroy(&dst_ports);